Enumerative Combinatorics: class

notes
Darij Grinberg
May 4, 2021 (unfinished!)
Status: Chapters 1 and 2 finished; Chapter 3 outlined.

Contents
1. Introduction 7
1.1. Domino tilings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.1. The problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2. The odd-by-odd case and the sum rule . . . . . . . . . . . . . 12
1.1.3. The symmetry and the bijection rule . . . . . . . . . . . . . . . 14
1.1.4. The m = 1 case . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.1.5. The m = 2 case and Fibonacci numbers . . . . . . . . . . . . . 19
1.1.6. Kasteleyn’s formula (teaser) . . . . . . . . . . . . . . . . . . . . 26
1.1.7. Axisymmetric domino tilings . . . . . . . . . . . . . . . . . . . 28
1.1.8. Tiling rectangles with k-bricks . . . . . . . . . . . . . . . . . . 31
1.2. Sums of powers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.2.1. The sum 1 + 2 + ... + n . . . . . . . . . . . . . . . . . . . . . . . 38
1.2.2. What is a sum, actually? . . . . . . . . . . . . . . . . . . . . . . 42
1.2.3. Rules for sums . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.2.4. While at that, what is a finite product? . . . . . . . . . . . . . 53
1.2.5. The sums 1k + 2k + ... + nk . . . . . . . . . . . . . . . . . . . . 54
1.3. Factorials and binomial coefficients . . . . . . . . . . . . . . . . . . . 58
1.3.1. Factorials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.3.2. Definition of binomial coefficients . . . . . . . . . . . . . . . . 59
1.3.3. Fundamental properties of the binomial coefficients . . . . . 61
1.3.4. Binomial coefficients count subsets . . . . . . . . . . . . . . . 69
1.3.5. Integrality and some arithmetic properties . . . . . . . . . . . 74
1.3.6. The binomial formula . . . . . . . . . . . . . . . . . . . . . . . 78
1.3.7. Other properties of binomial coefficients . . . . . . . . . . . . 86
1.4. Counting subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
1.4.1. All subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

1
Enumerative Combinatorics: class notes page 2

1.4.2. Lacunar subsets: the basics . . . . . . . . . . . . . . . . . . . . 94

1.4.3. Intermezzo: SageMath . . . . . . . . . . . . . . . . . . . . . . . 97
1.4.4. Counting lacunar subsets . . . . . . . . . . . . . . . . . . . . . 102
1.4.5. Counting k-element lacunar subsets . . . . . . . . . . . . . . . 107
1.4.6. Counting subsets with a odd and b even elements . . . . . . . 111
1.4.7. The addition formula for Fibonacci numbers . . . . . . . . . . 114
1.4.8. More subset counting . . . . . . . . . . . . . . . . . . . . . . . 119
1.4.9. Counting subsets containing a given subset . . . . . . . . . . 126
1.5. Counting tuples and maps . . . . . . . . . . . . . . . . . . . . . . . . . 128
1.5.1. Tuples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
1.5.2. Counting maps . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
1.5.3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
1.6. Interchange of summations . . . . . . . . . . . . . . . . . . . . . . . . 136
1.6.1. The finite Fubini principle . . . . . . . . . . . . . . . . . . . . . 136
1.6.2. The Fubini principle with a predicate . . . . . . . . . . . . . . 146
1.6.3. A cautionary tale about infinite sums . . . . . . . . . . . . . . 156
1.7. Counting permutations: an introduction . . . . . . . . . . . . . . . . 159
1.7.1. Permutations and derangements . . . . . . . . . . . . . . . . . 159
1.7.2. Only the size counts . . . . . . . . . . . . . . . . . . . . . . . . 161
1.7.3. Intermezzo: OEIS . . . . . . . . . . . . . . . . . . . . . . . . . . 166
1.7.4. The one-line notation . . . . . . . . . . . . . . . . . . . . . . . 168
1.7.5. Short-legged permutations . . . . . . . . . . . . . . . . . . . . 172
1.7.6. Long-legged permutations . . . . . . . . . . . . . . . . . . . . 176

2. Binomial coefficients 178

2.1. The alternating sum of a row of Pascal’s triangle . . . . . . . . . . . . 178
2.1.1. Telescoping sums . . . . . . . . . . . . . . . . . . . . . . . . . . 178
2.1.2. A war between the odd and the even . . . . . . . . . . . . . . 184
2.2. The trinomial revision formula . . . . . . . . . . . . . . . . . . . . . . 187
2.2.1. An algebraic proof . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.2.2. A double counting proof . . . . . . . . . . . . . . . . . . . . . 191
2.3. The hockey-stick identity revisited . . . . . . . . . . . . . . . . . . . . 198
2.4. Counting maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2.4.1. All maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2.4.2. Injective maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2.4.3. The pigeonhole principles . . . . . . . . . . . . . . . . . . . . . 210
2.4.4. Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
2.4.5. Surjective maps . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
2.5. 1m + 2m + ... + nm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
2.6. The Vandermonde convolution . . . . . . . . . . . . . . . . . . . . . . 226
2.6.1. The Vandermonde convolution theorem . . . . . . . . . . . . 226
2.6.2. The polynomial identity trick . . . . . . . . . . . . . . . . . . . 232
2.6.3. Salvaging the proofs of Theorem 2.6.1 . . . . . . . . . . . . . . 236
2.6.4. More consequences of the polynomial identity trick . . . . . . 238
Enumerative Combinatorics: class notes page 3

2.6.5. Mutating the Chu–Vandermonde identity . . . . . . . . . . . 241

2.7. Counting subsets again . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
2.8. Another use of polynomials . . . . . . . . . . . . . . . . . . . . . . . . 253
2.9. The Principle of Inclusion and Exclusion . . . . . . . . . . . . . . . . 257
2.9.1. The principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
2.9.2. The cancellation lemma . . . . . . . . . . . . . . . . . . . . . . 264
2.9.3. The proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
2.9.4. Application: Surjections . . . . . . . . . . . . . . . . . . . . . . 273
2.9.5. Application: Derangements . . . . . . . . . . . . . . . . . . . . 276
2.9.6. Application: Euler’s totient function . . . . . . . . . . . . . . . 282
2.9.7. Other cancellation-type lemmas . . . . . . . . . . . . . . . . . 290
2.10. Compositions and weak compositions . . . . . . . . . . . . . . . . . . 290
2.10.1. Compositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
2.10.2. Binary compositions . . . . . . . . . . . . . . . . . . . . . . . . 295
2.10.3. Weak compositions . . . . . . . . . . . . . . . . . . . . . . . . . 296
2.10.4. Other composition-like counting problems . . . . . . . . . . . 298
2.11. Multisubsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
2.11.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
2.11.2. Counting multisubsets of given size . . . . . . . . . . . . . . . 301
2.11.3. An application to lacunar subsets . . . . . . . . . . . . . . . . 302
2.12. Multinomial coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . 307
2.12.1. Definition and formulas . . . . . . . . . . . . . . . . . . . . . . 307
2.12.2. Counting maps that take values a given number of times . . 309
2.12.3. Counting anagrams . . . . . . . . . . . . . . . . . . . . . . . . 312
2.12.4. More formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . 315

3. The twelvefold way 317

3.1. What is the twelvefold way? . . . . . . . . . . . . . . . . . . . . . . . . 318
3.2. L→L. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
3.3. Equivalence relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
3.3.1. Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
3.3.2. Equivalence relations . . . . . . . . . . . . . . . . . . . . . . . . 326
3.3.3. Equivalence classes . . . . . . . . . . . . . . . . . . . . . . . . . 328
3.3.4. Defining unlabelled boxes and balls . . . . . . . . . . . . . . . 330
3.4. U→L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
3.5. L→U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
3.6. U → U and integer partitions . . . . . . . . . . . . . . . . . . . . . . . 341
3.7. Integer partitions (an introduction) . . . . . . . . . . . . . . . . . . . . 345
3.8. Odds and ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348

4. Permutations 349
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
4.2. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
4.3. Transpositions and cycles . . . . . . . . . . . . . . . . . . . . . . . . . 350
Enumerative Combinatorics: class notes page 4

4.4. Inversions and lengths . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

4.5. Descents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
4.6. Signs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

5. Lattice paths (brief introduction) 363

6. Generating functions (introduction) 367

7. Solutions and references to the exercises 369

7.1. Reference to solution to Exercise 1.3.1 . . . . . . . . . . . . . . . . . . 369
7.2. Solution to Exercise 1.3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 369
7.3. Reference to solution to Exercise 1.3.3 . . . . . . . . . . . . . . . . . . 369
7.4. Solution to Exercise 1.3.4 . . . . . . . . . . . . . . . . . . . . . . . . . . 370
7.5. Reference to solution to Exercise 1.3.5 . . . . . . . . . . . . . . . . . . 371
7.6. Solution to Exercise 1.3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . 371
7.7. Solution to Exercise 1.4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 372
7.8. Solution to Exercise 1.4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 374
7.9. Reference to solution to Exercise 1.4.6 . . . . . . . . . . . . . . . . . . 375
7.10. Reference to solution to Exercise 1.4.7 . . . . . . . . . . . . . . . . . . 375
7.11. Reference to solution to Exercise 1.4.8 . . . . . . . . . . . . . . . . . . 375
7.12. Reference to solution to Exercise 1.5.1 . . . . . . . . . . . . . . . . . . 375
7.13. Reference to solution to Exercise 1.5.2 . . . . . . . . . . . . . . . . . . 375
7.14. Reference to solution to Exercise 2.1.1 . . . . . . . . . . . . . . . . . . 375
7.15. Reference to solution to Exercise 2.2.1 . . . . . . . . . . . . . . . . . . 376
7.16. Solution to Exercise 2.2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 376
7.17. Reference to solution to Exercise 2.2.3 . . . . . . . . . . . . . . . . . . 378
7.18. Solution to Exercise 2.4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 378
7.19. Solution to Exercise 2.4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 380
7.20. Solution to Exercise 2.4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . 382
7.21. Solution to Exercise 2.4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . 383
7.22. Solution to Exercise 2.4.5 . . . . . . . . . . . . . . . . . . . . . . . . . . 389
7.23. Solution to Exercise 2.5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 390
7.24. Solution to Exercise 2.6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 392
7.25. Solution to Exercise 2.6.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 397
7.26. Solution to Exercise 2.6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . 397
7.27. Solution to Exercise 2.6.4 . . . . . . . . . . . . . . . . . . . . . . . . . . 398
7.28. Reference to solution to Exercise 2.6.5 . . . . . . . . . . . . . . . . . . 399
7.29. Reference to solution to Exercise 2.6.6 . . . . . . . . . . . . . . . . . . 399
7.30. Reference to solution to Exercise 2.6.7 . . . . . . . . . . . . . . . . . . 399
7.31. Reference to solution to Exercise 2.8.1 . . . . . . . . . . . . . . . . . . 399
7.32. Reference to solution to Exercise 2.9.1 . . . . . . . . . . . . . . . . . . 400
7.33. Solution to Exercise 2.9.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 400
7.34. Solution to Exercise 2.9.3 . . . . . . . . . . . . . . . . . . . . . . . . . . 405
7.35. Solution to Exercise 2.9.4 . . . . . . . . . . . . . . . . . . . . . . . . . . 410
Enumerative Combinatorics: class notes page 5

7.36. Reference to solution to Exercise 2.9.5 . . . . . . . . . . . . . . . . . . 417

7.37. Reference to solution to Exercise 2.9.6 . . . . . . . . . . . . . . . . . . 417
7.38. Reference to solution to Exercise 2.9.7 . . . . . . . . . . . . . . . . . . 417
7.39. Solution to Exercise 2.9.8 . . . . . . . . . . . . . . . . . . . . . . . . . . 418
7.40. Solution to Exercise 2.9.9 . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7.41. Solution to Exercise 2.9.10 . . . . . . . . . . . . . . . . . . . . . . . . . 420
7.42. Solution to Exercise 2.9.11 . . . . . . . . . . . . . . . . . . . . . . . . . 422
7.43. Reference to solution to Exercise 2.9.12 . . . . . . . . . . . . . . . . . 424
7.44. Solution to Exercise 2.10.1 . . . . . . . . . . . . . . . . . . . . . . . . . 424
7.45. Solution to Exercise 2.10.2 . . . . . . . . . . . . . . . . . . . . . . . . . 428
7.46. Solution to Exercise 2.10.3 . . . . . . . . . . . . . . . . . . . . . . . . . 432
7.47. Solution to Exercise 2.10.4 . . . . . . . . . . . . . . . . . . . . . . . . . 435
7.48. Solution to Exercise 2.10.5 . . . . . . . . . . . . . . . . . . . . . . . . . 436
7.49. Solution to Exercise 2.10.6 . . . . . . . . . . . . . . . . . . . . . . . . . 440
7.50. Solution to Exercise 2.10.7 . . . . . . . . . . . . . . . . . . . . . . . . . 447
7.51. Solution to Exercise 2.10.8 . . . . . . . . . . . . . . . . . . . . . . . . . 459
7.52. Solution to Exercise 2.10.9 . . . . . . . . . . . . . . . . . . . . . . . . . 461
7.53. Solution to Exercise 2.11.1 . . . . . . . . . . . . . . . . . . . . . . . . . 464
7.54. Solution to Exercise 2.11.2 . . . . . . . . . . . . . . . . . . . . . . . . . 465
7.55. Solution to Exercise 2.12.1 . . . . . . . . . . . . . . . . . . . . . . . . . 482
7.56. Solution to Exercise 2.12.1 . . . . . . . . . . . . . . . . . . . . . . . . . 485
7.57. Solution to Exercise 2.12.3 . . . . . . . . . . . . . . . . . . . . . . . . . 492
7.58. Solution to Exercise 2.12.4 . . . . . . . . . . . . . . . . . . . . . . . . . 493

This work is licensed under a Creative Commons “CC0

1.0 Universal” license.

Preface
This is the text accompanying my Math 222 (Enumerative Combinatorics) class at
Drexel University in Fall 2019. The website of this class can be found at
http://www.cip.ifi.lmu.de/~grinberg/t/19fco
and includes some extra materials (such as homeworks and solutions).
This document is a work in progress. It might become a textbook one day,
but for now only parts of it (currently Chapters 1 and 2) are at the level of detail
expected from a textbook and can be read on a standalone basis. The later chapters
resemble a construction zone.
Please report any errors you find to darijgrinberg@gmail.com .

What is this?
These notes cover the basics of enumerative combinatorics, with an emphasis on
counting, identities and bijections. We assume that you (the reader) are well famil-
Enumerative Combinatorics: class notes page 6

iar with the basics of rigorous mathematics (such as proof methods, the construc-
tions of integers and rationals, and basic properties of finite sets), as covered (for
example) in [LeLeMe16, Chapters 1–5], [Day16], [Hammac15] and [Newste19, Part
I and Appendices A–B]. We will not rely on any analysis, linear algebra or abstract
algebra except for the little that we introduce ourselves.
In terms of coverage, these notes do not set out to break any new ground.

• The first chapter (Chapter 1) is introductory and begins (in Section 1.1) with
a problem (that of counting domino tilings) that is not in itself particularly
important, but serves to motivate many basic ideas and notions (such as bi-
jective proofs, the sum rule and the product rule). We continue (in Section
1.2) with another elementary problem (viz., finding closed-form expressions
for sums of the form 1k + 2k + · · · + nk ), which we do not solve at this point (it
serves as a teaser for the next chapter) but which prompts us to introduce the
finite sum notation and get some passing acquaintance with certain combi-
natorial numbers that will later become important. The next section (Section
1.3) introduces factorials and binomial coefficients, and proves (mostly alge-
braically) their most basic properties, such as the recursion, the combinatorial
interpretation and the hockey-stick formula. In the sections that follow (Sec-
tions 1.4, 1.5, 1.6 and 1.7), we start counting for real: Various problems are
considered and solved, counting (certain kinds of) subsets, tuples, maps and
permutations. In the process we gradually introduce general strategies (such
as the difference rule, the isomorphism principle, or the interchange of sum-
mations) as they become useful. Among other things, we also provide brief
introductions to the use of modern electronic tools like the Online Encylope-
dia of Integer Sequences and the SageMath CAS.
• The second chapter (Chapter 2) focusses on binomial coefficients and related
concepts and problems. The first sections revisit some statements made in
the previous chapter, proving and re-proving them and demonstrating some
further techniques in the process. Section 2.4 answers one of the most ba-
sic counting problems, namely that of counting injective maps between two
finite sets; the analogous problem for surjective maps has no closed-form an-
swer, but two recurrences are derived. Section 2.5 then proves the formula for
1k + 2k + · · · + nk that was left unproved in Section 1.2. The next section (Sec-
tion 2.6) states, proves and applies the Vandermonde convolution theorem in
several ways, while also explaining the “polynomial identity trick” on which
all of the proofs rely and which is a versatile tool in the study of binomial
coefficients (despite not being in itself a result of combinatorics). The  next

n
sections continue with a re-proof of the combinatorial interpretation of
k
(Section 2.7), a first encounter with the method of generating functions (Sec-
tion 2.8), and the Principle of Inclusion and Exclusion (Section 2.9). Then, in
Section 2.10, we count several kinds and variants of compositions of an inte-
ger. In Section 2.11, we introduce multisubsets of a set, count them and show
Enumerative Combinatorics: class notes page 7

an application to an elementary (but far from simple) counting problem. In

Section 2.12, we introduce multinomial coefficients and establish their basic
properties.

• . . . (Further summaries will be written as the respective chapters will be fin-

ished.)

One particular goal of these notes is to develop at least the basics of the theory
rigorously – i.e., without handwaving, picture “proofs” and ambiguous terminol-
ogy like “ways to choose” or vaguely specified “arrangements”. This does not
mean that we shall avoid such semi-intuitive explanations whatsoever; but we aim
to ensure that they will never be used in a load-bearing capacity. Sometimes, a
statement will be first proved informally using such explanations, then re-proved
in a rigorous language.

These notes have their origin in the handwritten class notes I made for the Math 5705
(Enumerative Combinatorics) class at the UMN in Fall 20181 . However, they differ notice-
ably (both in coverage and in the order of the topics).

Class of 2019-09-23

1. Introduction
This is a text on enumerative combinatorics: the part of mathematics concerned with
the sizes of finite sets, particularly their computation and the proof of equalities
between them. More precisely, here are what I consider to be the three main threads
of enumerative combinatorics:

• Counting – i.e., finding formulas for the sizes of certain finite sets. For ex-
ample, we count the permutations of the set {1, 2, . . . , n}, or the k-subsets of
{1, 2, . . . , n} that contain no two consecutive elements. “Count” means find-
ing a formula that expresses the number of such permutations or k-subsets in
terms of n and k.
n
  polynomial identities (such as the binomial formula ( x + y) =
• Proving
n n k n−k
∑ x y or various deeper ones).
k =0 k

• Finding and studying interesting maps between finite sets. A basic exam-
ple of such a map is the “bit-set encoding”: the bijection from the set of all
subsets of {1, 2, . . . , n} (for a fixed positive integer n) to the set of all n-tuples
(i1 , i2 , . . . , in ) ∈ {0, 1}n (known as “length-n bitstrings”) which sends each

1 http://www.cip.ifi.lmu.de/~grinberg/t/18f
Enumerative Combinatorics: class notes page 8

(
1, if k ∈ S;
subset S of {1, 2, . . . , n} to the n-tuple (i1 , i2 , . . . , in ), where ik = .
0, if k ∈
/S
We will care particularly about bijections, since they directly help in counting,
but even non-bijective maps are fundamental to enumerative combinatorics.

You will see more examples of each of these three threads all over this text,
starting with this introductory chapter.
We will also occasionally see some connections to linear algebra, abstract algebra, num-
ber theory and graph theory. There are other threads in enumerative combinatorics that
we are not going to encounter (or only tangentially): applications (mostly), connections to
representation theory or geometry, asymptotics and many more. A one-semester course
needs to have its limits!
First, I will discuss some interesting (if you share my taste) questions, in no
particular order. Not all of them will be answered right away.

1.1. Domino tilings

1.1.1. The problem
Let n ∈ N and m ∈ N. Here and in the following, N means the set {0, 1, 2, . . .}.
Let Rn,m denote an n × m-rectangle, i.e., a rectangle with width n and height
m. (We imagine a specific such rectangle drawn somewhere in the plane.) For
example, R3,4 looks like this:2




y
4
x
 (1)



−−→ 3 ←−−

A domino shall mean a 1 × 2-rectangle or a 2 × 1-rectangle. More specifically:

A vertical domino shall mean a 1 × 2-rectangle; a horizontal domino shall mean a
2 × 1-rectangle. Here is how they look like:

; .
| {z }
| {z } horizontal domino
vertical domino

A domino tiling of Rn,m is a way to cover the rectangle Rn,m with non-overlapping
dominos.
2 We subdivide it with grid lines just to show its dimensions.
Enumerative Combinatorics: class notes page 9

For example, here are three domino tilings of the rectangle R3,4 (which rectangle
you have seen in (1)):

, , .

(Now, of course, we are no longer drawing the grid lines, but only the outlines of
the dominos.)
We can now state our first enumeration (i.e., counting) problem: How many
domino tilings does Rn,m have?
As we just saw, R3,4 has at least 3 domino tilings, but in fact you can find several
more. Counting them all is, at the very least, an unpleasant exercise in carefulness.
Let us try a simpler example:

Example 1.1.1. Here are all domino tilings of R3,2 :

, , .

If we are to solve the above problem in general, our first step should be making it
rigorous. We said that a domino tiling should be a way to cover the rectangle Rn,m
with non-overlapping dominos. What does “cover” mean, and what does “non-
overlapping” mean? Visually, it is pretty clear, but we do not have bulletproof
mathematical definitions yet. There are two ways to create such definitions:

• The geometric way: We really define Rn,m as a rectangle of width n and

height m in the Euclidean plane; for example, let us pick the rectangle with
vertices (0, 0), (n, 0), (n, m) and (0, m) (where we model the Euclidean plane
through a Cartesian coordinate system as usual). We say that a set of dominos
covers Rn,m if their union (as sets) is Rn,m . It is harder to define what it means
for a set of dominos to be non-overlapping; clearly, this is not quite the same as
them being disjoint as sets (because they are allowed to have edges or vertices
in common). There are several good ways to define non-overlappingness3 .
Unfortunately, once all these definitions are made, it is still far from clear how
to reason about them rigorously! For example, it may seem obvious, but why
exactly must all the dominos in a domino tiling of Rn,m be “snapped to the

3 For example, you can say that two dominos are non-overlapping if their intersection is either the
empty set or a point or a line segment. Then you can say that a set of dominos is non-overlapping
if any two distinct dominos in it are non-overlapping.
Enumerative Combinatorics: class notes page 10

grid” (i.e., why must their corners be grid points4 )? This is indeed true, but
proving this would take serious work. Thus, even our previous observation
that R3,2 has three domino tilings (shown visually in Example 1.1.1) would
become a nontrivial theorem. Thus, we leave this geometric model of domino
tilings aside, and instead define things in . . .

• The combinatorial way: We redefine Rn,m as the set [n] × [m], where we set
[k] = {1, 2, . . . , k} for each k ∈ N.
Its elements thus are the pairs (i, j) with i ∈ [n] and j ∈ [m]; we call these
pairs “squares”. Thus, Rn,m is a finite set of size5 | Rn,m | = nm.
A vertical domino shall mean a set of the form {(i, j) , (i, j + 1)} for some i, j ∈
Z.
A horizontal domino shall mean a set of the form {(i, j) , (i + 1, j)} for some
i, j ∈ Z.
A domino shall mean a set that is either a vertical domino or a horizontal
domino.
If S is a set of squares (for example, Rn,m ), then a domino tiling of S shall
mean a set {S1 , S2 , . . . , Sk } of disjoint dominos whose union is S (that is,
S1 ∪ S2 ∪ · · · ∪ Sk = S).
A few words are in order about what this has to do with our visual concept
of domino tilings.
First of all, why are we suddenly considering the finite set Rn,m to be a “rect-
angle”? Because we are no longer thinking in terms of all points in the plane,
but rather thinking in terms of grid squares (as in (1)). Thus, the rectangle
Rn,m is no longer an (infinite) set of points, but now becomes a (finite) set of
grid squares that lie in this rectangle. We label these grid squares by pairs
of integers (namely, we label each grid square by the pair (i, j) of Cartesian
coordinates of its northeastern corner6 ; thus, the southwesternmost square of

4A grid point means a point with integer coordinates.

5 The size of a finite set S is the number of elements of S. It is denoted by |S|. For example,
|{1, 4, 6}| = 3 and |{−1, 1, 6, 7}| = 4 and |∅| = 0.
6 Here is how R
3,4 looks like with each square labeled:

(1, 4) (2, 4) (3, 4)

(1, 3) (2, 3) (3, 3)

(1, 2) (2, 2) (3, 2)

(1, 1) (2, 1) (3, 1)

Enumerative Combinatorics: class notes page 11

Rn,m is labeled (1, 1), and the eastern neighbor of a square (i, j) is (i + 1, j),
whereas the northern neighbor of a square (i, j) is (i, j + 1)). In other words,
the square in column i (counted from the left) and row j (counted from the
bottom) is labelled by the pair (i, j).
Having thus redefined the rectangle Rn,m as a finite set of grid squares, we
then do the same for dominos and domino tilings. A domino, too, is not an
infinite set any more, but just a set of two adjacent grid squares. It is a vertical
domino if these grid squares differ in their y-coordinate (i.e., have the forms
(i, j) and (i, j + 1) for some i, j ∈ Z), and it is a horizontal domino if these grid
squares differ in their x-coordinate (i.e., have the forms (i, j) and (i + 1, j) for
some i, j ∈ Z). From this point of view, two dominos are non-overlapping
if they are literally disjoint (because they are sets of grid squares now, and
thus disjointness means that they have no grid squares in common; it does
not matter if they share an edge).
So we have obtained a new model for domino tilings, with simpler definitions
and with all sets involved being finite. This kind of model is called a discrete
model. It is much more manageable than the geometric one, and in partic-
ular, almost everything that is visually obvious is actually straightforward to
prove in this model (unlike in the geometric one). For example, it is easy
to rigorously reproduce our result from Example 1.1.1 saying that R3,2 has 3
domino tilings. In the discrete model, these domino tilings are7
 

 


 


 

 
{(1, 1) , (1, 2)}, {(2, 1) , (2, 2)}, {(3, 1) , (3, 2)} ,

 | {z } | {z } | {z }


 vertical domino vertical domino vertical domino 

covering the covering the covering the

 

 
leftmost column middle column rightmost column
 

 


 


 

 
{(1, 1) , (2, 1)}, {(1, 2) , (2, 2)}, {(3, 1) , (3, 2)} ,

 | {z } | {z } | {z }


 horizontal domino horizontal domino vertical domino 

 in the bottom row in the top row covering the 

 

rightmost column
 

 


 


 

{(1, 1) , (1, 2)}, {(2, 1) , (3, 1)}, {(2, 2) , (3, 2)} .

 | {z } | {z } | {z }
 vertical domino horizontal domino horizontal domino
 

 covering the
 in the bottom row in the top row 

leftmost column

7 listed here in the same order in which they appeared in Example 1.1.1
Enumerative Combinatorics: class notes page 12

You do need a bit of work to verify that no other domino tilings of R3,2 exist;
but it is very much doable. The surefire (but boring) way is to simply check
all possibilities by brute force: There are only 7 dominos that lie inside8 R3,2 ,
and clearly any domino tiling must consist of some of these 7 dominos; now,
you can check all the 27 possible subsets. (Of course, you can be a lot less
stupid.)

From now on, we shall always be using the discrete model when we study
domino tilings – i.e., we define Rn,m , dominos and domino tilings via the com-
binatorial way.
If n, m ∈ N, then let us define an integer dn,m by

dn,m = (# of domino tilings of Rn,m ) . (2)

Here and in the following, the symbol “#” always means “number” (or “the num-
ber”, depending on context).
Our problem thus asks us to compute dn,m . In Example 1.1.1, we have seen that
d3,2 = 3. For any fixed n and m, we can technically compute dn,m by brute force
(i.e., trying out all possible subsets of the set of dominoes lying inside Rn,m , and
counting the domino tilings among them). But this becomes forbiddingly slow
when n and m get even a little bit large (say, n = 8 and m = 8). We are looking for
something better: for an explicit formula for dn,m if possible, and otherwise at least
for faster algorithms that compute dn,m .

1.1.2. The odd-by-odd case and the sum rule

We begin with a particularly simple case:

Proposition 1.1.2. Assume that n and m are odd. Then, dn,m = 0.

Proof sketch. We have assumed that n and m are odd. Thus, the product nm is odd
as well. In other words, the size | Rn,m | is odd (since | Rn,m | = nm).
But each domino has even size (in fact, it has size 2).
If the set Rn,m had a domino tiling, then the size | Rn,m | of Rn,m would equal the
sum of the sizes of all the dominos in the tiling (because Rn,m is the union of the
dominos, and the dominos are disjoint). But the size | Rn,m | is odd, whereas the
sum of the sizes of all the dominos in the tiling is even (since each domino has
even size); thus the former cannot equal the latter. This shows that the set Rn,m has
no domino tilings. In other words, the # of domino tilings of Rn,m is 0. In other
words, dn,m = 0.
It is worth being a little bit more detailed once and look under the hood of this
proof. We have used the following basic fact:

8 To “lie inside” R3,2 means to be a subset of R3,2 here.

Enumerative Combinatorics: class notes page 13

Theorem 1.1.3 (The sum rule). If a finite set S is the union of k disjoint sets
S1 , S2 , . . . , Sk , then
| S | = | S1 | + | S2 | + · · · + | S k | .

Theorem 1.1.3 is known as the sum rule or the addition rule, and is so fundamental
for all of mathematics that you have probably not even noticed us tacitly using it in
the proof of Proposition 1.1.2 above. We shall not prove Theorem 1.1.3, since this
is a job for “axiomatic foundations of mathematics” courses and depends on the
“implementation details” of your mathematical “standard library” (such as: how
do you define the size of a finite set?).9
We can restate Theorem 1.1.3 as follows: If S1 , S2 , . . . , Sk are k disjoint finite sets,
then the set S1 ∪ S2 ∪ · · · ∪ Sk is finite and satisfies

| S1 ∪ S2 ∪ · · · ∪ S k | = | S1 | + | S2 | + · · · + | S k | . (3)

(Indeed, this follows from Theorem 1.1.3, applied to S = S1 ∪ S2 ∪ · · · ∪ Sk .)

Thus, if X and Y are two disjoint finite sets, then the set X ∪ Y is finite and
satisfies
| X ∪ Y | = | X | + |Y | . (4)
(Indeed, this follows from (3), applied to k = 2, S1 = X and S2 = Y.)
We have also used the visually obvious fact that | Rn,m | = nm (that is, Rn,m has
nm squares). Formally speaking, this is a consequence of another basic fact:

Theorem 1.1.4 (The product rule for two sets). Let X and Y be two finite sets.
Then, X × Y is a finite set with size

| X × Y | = | X | · |Y | . (5)

In other words, this theorem is saying that the number of pairs ( x, y) ∈ X × Y

(where X and Y are two given finite sets) is | X | · |Y |. This is intuitive, since such a
pair ( x, y) can be constructed by choosing an element x of X (there are | X | many
options for it) and choosing an element y of Y (there are |Y | many options for it);
since the two choices are completely independent, it is reasonable that they should
lead to | X | · |Y | many options for the whole pair ( x, y).

9 Forexample, if you define your sets and numbers in the old-fashioned Bourbakist way, then you
can find Theorem 1.1.3 with proof in [Bourba68, Chapter III, §3.3, Corollary]. If you are using
constructivist foundations, then Theorem 1.1.3 can be proven by induction on k (see [Loehr11,
proof of 1.2] for the details of this induction proof), relying on the fact that any two disjoint finite
sets A and B of | A ∪ B| = | A| + | B|. The latter fact (which is, of course, essentially equivalent to
the particular case of Theorem 1.1.3 for k = 2) can be proven directly by explicitly constructing
a bijection A ∪ B → [n + m] out of two bijections A → [n] and B → [m] (see [Loehr11, proof of
1.32] for the details of this construction). But we will not dwell on fundamental issues like this
here.
Enumerative Combinatorics: class notes page 14

Again, we will not prove Theorem 1.1.4, as it is sufficiently elementary.10 It

is called the product rule for two sets, as there is a product rule for k sets as well
(Theorem 1.5.3 below).
Let us now restate our above proof of Proposition 1.1.2 in a way that makes the
uses of Theorem 1.1.3 and of Theorem 1.1.4 in it explicit:
Proof of Proposition 1.1.2 (detailed version). For each k ∈ N, we have [k ] = {1, 2, . . . , k } and
thus |[k ]| = |{1, 2, . . . , k }| = k. Hence, |[n]| = n and |[m]| = m. But Rn,m = [n] × [m] and
thus

| Rn,m | = |[n] × [m]| = |[n]| · |[m]| (by (5), applied to X = [n] and Y = [m])
|{z} | {z }
=n =m
= nm.
We have assumed that n and m are odd. Thus, the product nm is odd as well. In other
words, the size | Rn,m | is odd (since | Rn,m | = nm).
Our next goal is to show that the set Rn,m has no domino tilings.
Indeed, let T be a domino tiling of Rn,m . We will derive a contradiction.
Write T in the form T = {S1 , S2 , . . . , Sk }, where S1 , S2 , . . . , Sk are distinct dominos11 .
Then, the sets S1 , S2 , . . . , Sk are dominos; thus, their sizes |S1 | , |S2 | , . . . , |Sk | are even (since
the size of each domino is even12 ). Hence, the sum |S1 | + |S2 | + · · · + |Sk | is even (being a
sum of even integers).
But the finite set Rn,m is a union of the k disjoint sets S1 , S2 , . . . , Sk (since {S1 , S2 , . . . , Sk } =
T is a domino tiling of Rn,m ). Hence, Theorem 1.1.3 (applied to S = Rn,m ) yields | Rn,m | =
|S1 | + |S2 | + · · · + |Sk |. Hence, | Rn,m | is even (since |S1 | + |S2 | + · · · + |Sk | is even). This
contradicts the fact that | Rn,m | is odd.
Now, forget that we fixed T. We thus have found a contradiction for each domino tiling
T of Rn,m . This shows that there exists no domino tiling of Rn,m . In other words, the # of
domino tilings of Rn,m is 0. In other words, dn,m = 0. This proves Proposition 1.1.2.

In the future, we will be using the sum rule (Theorem 1.1.3) many times, usually
without even mentioning it.

1.1.3. The symmetry and the bijection rule

Thus we have handled at least one case of our counting problem: the case when
n and m are odd. It remains to handle the case when at least one of n and m is
even. More precisely, it suffices to handle the case when n is even, because of the
following symmetry in the problem:
10 See [Loehr11, 1.5] for a proof (even of a more general statement).
11 Here, we are tacitly using the fact that T is finite. Why is T finite? Intuitively it is obvious. More
rigorously, you can argue this as follows: There are only finitely many dominos that are subsets
of Rn,m . The set T, being a domino tiling of Rn,m , must consist entirely of such dominos; thus, it
must be a subset of the (finite) set of these dominos. Hence, T is itself finite (since a subset of a
finite set is always finite).
We trust you to make such arguments whenever necessary; we will not dwell on them in the
future.
12 Indeed, the size of each domino is 2.
Enumerative Combinatorics: class notes page 15

Proposition 1.1.5. Let n, m ∈ N. Then, dn,m = dm,n .

Proof sketch. The idea is very simple: The rectangle Rm,n can be obtained by “flip-
ping” the rectangle Rn,m across the line with equation x = y (in Cartesian coordi-
nates). This “flip” operation turns domino tilings of Rm,n into domino tilings of
Rn,m and vice versa; here is an example:

flip
−→ .

Thus, the domino tilings of Rm,n are in 1-to-1 correspondence with the domino
tilings of Rn,m . This entails that the # of the former equals the # of the latter. Since
the # of the former is dm,n (by the definition of dm,n ), whereas the # of the latter is
dn,m (by the definition of dn,m ), we can rewrite this as follows: dm,n = dn,m . This
proves Proposition 1.1.5.
It is worth expanding this proof just to see what exactly we have done; again,
the underlying principle is very basic but worth stating at least once. The domino
tilings of Rm,n are not literally the same as the domino tilings of Rn,m (unless n = m
or one of n and m is 0); yet, we have argued that the former are in 1-to-1 correspon-
dence with the latter, and therefore equinumerous to the latter. Formally, a 1-to-1
correspondence between two sets of objects is given by a map from one set to the
other, but it cannot be just any map: It has to be a bijection (i.e., a bijective map)13 .
Thus, what we have used is the fact that if there is a bijection between two sets,
then these two sets have the same size. Let us state this a little bit more explicitly:

13 We refer to places like [LeLeMe16, §4.4–4.5], [Hammac15, §12.2] or [Day16, §3.F] for basic prop-
erties of bijections and bijectivity. Here come some brief reminders: A map f : X → Y between
two sets X and Y is said to be

• injective if it has the property that ( f ( x1 ) = f ( x2 )) =⇒ ( x1 = x2 ) for any two elements

x1 , x2 ∈ X (or, equivalently, if it maps any two distinct elements of X to two distinct
elements of Y);
• surjective if it has the property that for each y ∈ Y, there exists at least one x ∈ X satisfying
f ( x ) = y (in other words, every element of Y is a value of f );
• bijective if it is both injective and surjective.

It is easy to see that a map is bijective if and only if it is invertible (i.e., has an inverse).
The word “bijection” is a shorthand for “bijective map”. Likewise, the word “injection” is
a shorthand for “injective map”, whereas the word “surjection” is a shorthand for “surjective
map”.
Enumerative Combinatorics: class notes page 16

Theorem 1.1.6 (The bijection principle). If X and Y are two sets, and if f : X → Y
is a bijection (i.e., a bijective map), then

| X | = |Y | . (6)

We have not required X and Y to be finite in Theorem 1.1.6. The size of an infinite
set is a well-defined notion in mathematics (more commonly known as cardinality);
see, e.g., [LeLeMe16, §8.1] for an introduction to this or [Newste19, Chapter 9] for a
more in-depth treatment. We will only use Theorem 1.1.6 in the case when at least
one of the sets X and Y is known to be finite (in which case its size is a nonnegative
integer). In this case, Theorem 1.1.6 automatically yields that both sets X and Y
are finite, and have equal size.
Theorem 1.1.6 is known as the bijection principle or the bijection rule, and is suffi-
ciently basic that some authors consider it part of the definition of the cardinality
of a set; we are not going to reference it explicitly every time we use it. But just this
one time, let us do so, and while at that, also formalize the definition of the “flip”
operation that was used in our proof of Proposition 1.1.5:
Proof of Proposition 1.1.5 (detailed version). Define the map

F : Rn,m → Rm,n ,
(i, j) 7→ ( j, i ) .

(This notation is saying “the map F from Rn,m to Rm,n that sends each element (i, j) of
Rn,m to the element ( j, i ) of Rm,n ”. To see why this map is well-defined, just recall that
Rn,m = [n] × [m] and Rm,n = [m] × [n], which is why (i, j) ∈ Rn,m will always lead to
( j, i ) ∈ Rm,n .)
Visually speaking, this map F simply flips each square of the rectangle Rn,m across the
line with equation x = y. This yields a square of the rectangle Rm,n , of course. Thus, it is
clear that the map F is a bijection. This can be proved rigorously as follows: The map

G : Rm,n → Rn,m ,
(i, j) 7→ ( j, i )

is well-defined14 and inverse to F 15 . Thus, the map F is invertible, i.e., is a bijection.

Note that its inverse map G was defined in the same way as F, but simply with the roles of
n and m interchanged.
The map F is a bijection, but it is not the bijection that we are going to apply Theorem
1.1.6 to. (If we applied Theorem 1.1.6 to f = F, then we would conclude that | Rn,m | =
| Rm,n |, that is, nm = mn, which is reassuring but not what we are trying to prove.)
14 This is proved in the same way as we showed that F is well-defined.  
15 since F ◦ G = id (because each (i, j) ∈ Rm,n satisfies ( F ◦ G ) (i, j) = F  G (i, j) = F ( j, i ) =
 
| {z }
=( j,i )
(i, j) = id (i, j)) and G ◦ F = id (for similar reasons)
Enumerative Combinatorics: class notes page 17

The map F merely flips the squares of Rn,m across the x = y line; we want a map that
flips domino tilings. Of course, to flip a domino tiling, we have to flip each domino in it;
and to flip a domino, we have to flip each square in it. Thus, we define the following two
maps:

• Define the map

Fdom : {dominos inside Rn,m } → {dominos inside Rm,n } ,

D 7→ { F (d) | d ∈ D } .

(To spell this out: The map Fdom sends each domino D to the domino { F (d) | d ∈ D },
which is obtained from D by flipping each square d ∈ D. In other words, it flips each
domino by applying the flip map F to each square of the domino.)
It is easy to see that this map Fdom is a bijection. (Indeed, it has an inverse Gdom ,
which is defined in the same way but with the roles of n and m interchanged.)

• Define the map

Ftil : {domino tilings of Rn,m } → {domino tilings of Rm,n } ,

T 7→ { Fdom ( D ) | D ∈ T } .

(To spell this out: The map Ftil sends each domino tiling T to the domino tiling
{ Fdom ( D ) | D ∈ T }, which is obtained from T by flipping each domino D ∈ T. In
other words, it flips each domino tiling by applying the flip map Fdom to each domino
of the tiling.)
It is easy to see that this map Ftil is a bijection. (Indeed, it has an inverse Gtil , which
is defined in the same way but with the roles of n and m interchanged.)

To be fully rigorous, we would have to check that these two maps Fdom and Ftil are well-
defined (i.e., that flipping a domino inside Rn,m really results in a domino inside Rm,n , and
that flipping a domino tiling of Rn,m really results in a domino tiling of Rm,n ), but this is
intuitively clear, and the formal proof can easily be constructed by just “following your
nose”16 , which is why we omit it.
Anyway, we now have constructed a map
Ftil : {domino tilings of Rn,m } → {domino tilings of Rm,n } and showed that it is a bijection.
Hence, (6) (applied to X = {domino tilings of Rn,m } and Y = {domino tilings of Rm,n } and
f = Ftil ) shows that

|{domino tilings of Rn,m }| = |{domino tilings of Rm,n }| . (7)

But the definition of dn,m yields

dn,m = (# of domino tilings of Rn,m ) = |{domino tilings of Rn,m }| .

16 For example: If D is a horizontal domino {(i, j) , (i + 1, j)}, then flipping it yields the vertical
domino {( j, i ) , ( j, i + 1)}. Likewise, flipping a vertical domino yields a horizontal domino. Thus,
flipping a domino yields a domino. (Note that this would be false if our definition of dominos
didn’t have the symmetry built in!)
Enumerative Combinatorics: class notes page 18

The same reasoning shows that

dm,n = |{domino tilings of Rm,n }| .

In view of the latter two equalities, we can rewrite (7) as dn,m = dm,n . Thus Proposition
1.1.5 is proven.

For the sake of future use, we observe that Theorem 1.1.6 has a converse:

Theorem 1.1.7. If X and Y are two sets of the same size (that is, | X | = |Y |), then
there exists a bijection from X to Y.

Again, this holds for infinite sets just as it does for finite ones, but we will only
use it for finite sets.

1.1.4. The m = 1 case

Let us bite another piece off the problem:

Proposition 1.1.8. Assume that m = 1 and that n is even. Then, dn,m = 1.

Proof of Proposition 1.1.8 (sketched). We must show that there is exactly one domino
tiling of Rn,m . But this is visually obvious: The rectangle Rn,m has only one row,
and this row has an even number of squares, so we can cover it with horizontal
dominos in only one way. Here is how this domino tiling looks like:

··· .

This is not yet a rigorous proof, but it is fairly easy to turn it into one. Rigorously
speaking, the domino tiling we just described is17

{{(1, 1) , (2, 1)} , {(3, 1) , (4, 1)} , {(5, 1) , (6, 1)} , . . . , {(n − 1, 1) , (n, 1)}}
= {{(2k − 1, 1) , (2k, 1)} | k ∈ {1, 2, . . . , n/2}} . (8)

It is instantly clear that this is a domino tiling of Rn,m . In order to show that this is the only
domino tiling of Rn,m , we can let T be any domino tiling of Rn,m , and then argue as follows:

• The square (1, 1) must be contained in some domino A1 ∈ T (since T is a domino

tiling). This domino A1 must be either {(1, 1) , (2, 1)} or {(0, 1) , (1, 1)} or {(1, 1) , (1, 2)}
or {(1, 0) , (1, 1)} (since these are the only dominos that contain (1, 1)). But out of
these four dominos, only {(1, 1) , (2, 1)} is a subset of Rn,m (since m = 1). Hence, A1
must be the domino {(1, 1) , (2, 1)}. Thus, (2, 1) is also contained in A1 .
Now we know that our tiling T looks like this:

A1 A1 ? ? ? ? ? ? ? ? ? ? ? ? ? ··· ? ? ? ? ? ? ? ? ?

(where we have labeled the leftmost two squares with “A1 ” to signify that they are
contained in the domino A1 ).
17 Pay attention to where the set braces are! This is a set of sets of pairs of numbers.
Enumerative Combinatorics: class notes page 19

• The square (3, 1) must be contained in some domino A2 ∈ T. This domino A2 must
be either {(3, 1) , (4, 1)} or {(2, 1) , (3, 1)} or {(3, 1) , (3, 2)} or {(3, 0) , (3, 1)} (since
these are the only dominos that contain (3, 1)). But out of these four dominos, only
{(3, 1) , (4, 1)} and {(2, 1) , (3, 1)} are subsets of Rn,m (since m = 1). Hence, A2 must
be either {(3, 1) , (4, 1)} or {(2, 1) , (3, 1)}. But the dominos in a domino tiling must
be disjoint (by definition); hence, A2 cannot be {(2, 1) , (3, 1)} (because if A2 was
{(2, 1) , (3, 1)}, then it would fail to be disjoint from A1 = {(1, 1) , (2, 1)}). Thus, A2
must be {(3, 1) , (4, 1)}. Hence, (4, 1) is also contained in A2 .
Now we know that our tiling T looks like this:

A1 A1 A2 A2 ? ? ? ? ? ? ? ? ? ? ? ··· ? ? ? ? ? ? ? ? ?

• The square (5, 1) must be contained in some domino A3 ∈ T. This domino A3 must
be either {(5, 1) , (6, 1)} or {(4, 1) , (5, 1)} or {(5, 1) , (5, 2)} or {(5, 0) , (5, 1)} (since
these are the only dominos that contain (5, 1)). But out of these four dominos, only
{(5, 1) , (6, 1)} and {(4, 1) , (5, 1)} are subsets of Rn,m (since m = 1). Hence, A3 must
be either {(5, 1) , (6, 1)} or {(4, 1) , (5, 1)}. But the dominos in a domino tiling must
be disjoint (by definition); hence, A3 cannot be {(4, 1) , (5, 1)} (because if A3 was
{(4, 1) , (5, 1)}, then it would fail to be disjoint from A2 = {(3, 1) , (4, 1)}). Thus, A3
must be {(5, 1) , (6, 1)}. Hence, (6, 1) is also contained in A3 .
Now we know that our tiling T looks like this:

A1 A1 A2 A2 A3 A3 ? ? ? ? ? ? ? ? ? ··· ? ? ? ? ? ? ? ? ?

. . . and so on, proceeding further and further right until you hit the “eastern wall” of Rn,m
(that is, the square (n, 1)). Thus, the dominos appearing in T are uniquely determined:
They must be A1 = {(1, 1) , (2, 1)}, A2 = {(3, 1) , (4, 1)}, A3 = {(5, 1) , (6, 1)} and so on.
This is precisely the one tiling that we presented in (8). Thus, that one tiling is the only
domino tiling of Rn,m .
To be fully rigorous, this argument should be formalized as an induction proof (feel free
to do so!), but even if I wake you up at night, you will know how to construct this argument
if necessary, because the idea behind it is glaringly obvious (just walk the rectangle Rn,m
from its western wall to its eastern wall, and observe that at each step, there is only one
possible domino that fits in the rectangle without overlapping with the previous domino).

1.1.5. The m = 2 case and Fibonacci numbers

Between Proposition 1.1.2 and Proposition 1.1.8, we have fully covered the case
m = 1 of our problem. Let us now move on to the case m = 2. We compute
Enumerative Combinatorics: class notes page 20

dn,m = dn,2 for some small values of n simply by listing all domino tilings of Rn,m :

n dn,m domino tilings

0 d0,2 = 1

1 d1,2 = 1

2 d2,2 = 2 ,
.

3 d3,2 = 3 , ,

4 d4,2 = 5 , , ,

If the n = 0 case confuses you, keep in mind that the rectangle R0,2 is the empty
set (since R0,2 = [0] × [2] = ∅ × [2] = ∅) and thus has exactly one domino tiling
|{z}
=∅
– namely, the tiling that contains no dominos (i.e., the empty set).
Can you find d5,2 ?
Here is a quick way to the answer, at least as far as counting is concerned:

Proposition 1.1.9. For each integer n ≥ 2, we have dn,2 = dn−1,2 + dn−2,2 .

Class of 2019-09-25

Proof of Proposition 1.1.9 (sketched). Let n ≥ 2 be an integer. Consider the last18 col-
umn of Rn,2 (that is, the set {(n, 1) , (n, 2)}).
In any domino tiling T of Rn,2 , this last column is either covered by 1 vertical
domino, or covered by (parts of) 2 horizontal dominos.

18 i.e., easternmost
Enumerative Combinatorics: class notes page 21

In the former case, we shall call T a type-1 tiling; in the latter case, we shall call T
a type-2 tiling. Visually, these look as follows:

??????????????? ???????????????
,
??????????????? ???????????????
| {z } | {z }
type-1 tiling type-2 tiling

(where the question marks mean an unknown arrangement of dominos).

Let us now analyze type-1 tilings. A type-1 tiling consists of the single verti-
cal domino {(n, 1) , (n, 2)} that covers its last column, and a bunch of dominos
that cover all the remaining n − 1 columns. This latter bunch must thus be a
domino tiling of Rn−1,2 . Thus, a type-1 tiling consists of the single vertical domino
{(n, 1) , (n, 2)} and an arbitrary domino tiling of Rn−1,2 . (Visually, this means that
some domino
it looks as follows: .) Hence,19
tiling of Rn−1,2

(# of type-1 tilings) = (# of domino tilings of Rn−1,2 ) (9)

= dn−1,2 (10)
(since dn−1,2 was defined as the # of domino tilings of Rn−1,2 ).
Let us next analyze type-2 tilings. In a type-2 tiling, the last column is covered by
(parts of) 2 horizontal dominos. These 2 dominos must extend to the left (because
there is no space for them to extend to the right), and thus also cover the second-to-
last column. Explicitly speaking, these 2 dominos must be {(n − 1, 1) , (n, 1)} and
{(n − 1, 2) , (n, 2)}. All the other dominos in the tiling must then cover the remain-
ing n − 2 columns, i.e., must form a domino tiling of Rn−2,2 . Thus, a type-2 tiling
consists of the two horizontal dominos {(n − 1, 1) , (n, 1)} and {(n − 1, 2) , (n, 2)}
and an arbitrary domino tiling of Rn−2,2 . (Visually, this means that it looks as
some domino
follows: .) Hence,
tiling of Rn−2,2

(# of type-2 tilings) = (# of domino tilings of Rn−2,2 ) (11)

= dn−2,2 (12)
(since dn−2,2 was defined as the # of domino tilings of Rn−2,2 ).
Now, recall that each domino tiling of Rn,2 is either a type-1 tiling or a type-2
tiling (but cannot be both at the same time). Hence,
(# of domino tilings of Rn,2 )
= (# of type-1 tilings) + (# of type-2 tilings) (13)
= dn−1,2 + dn−2,2 (14)
19 When we say “type-1 tiling”, we mean “type-1 tiling of Rn,2 ”, of course. (The same will apply to
“type-2 tiling” later on.)
Enumerative Combinatorics: class notes page 22

(by adding the equalities (10) and (12) together). Now, the definition of dn,2 yields
dn,2 = (# of domino tilings of Rn,2 ) = dn−1,2 + dn−2,2
(by (14)). This proves Proposition 1.1.9.
Again, let us analyze what we have actually done in this proof:
1. The equality (9) follows from the bijection principle. Indeed, our argument for it
boils down to the (easily established) fact that there is a bijection
f : {domino tilings of Rn−1,2 } → {type-1 tilings}
(which takes any domino tiling of Rn−1,2 , and adds the vertical domino {(n, 1) , (n, 2)}
to it). Once you have convinced yourself of this fact, you can apply Theorem 1.1.6 to
X = {domino tilings of Rn−1,2 } and Y = {type-1 tilings}, and conclude that
|{domino tilings of Rn−1,2 }| = |{type-1 tilings}| .
In other words, (# of domino tilings of Rn−1,2 ) = (# of type-1 tilings). Thus, the
equality (9) is proven. The equality (11) is obtained similarly.
2. The equality (13) follows from the sum rule. Indeed, the sets {type-1 tilings} and
{type-2 tilings} are disjoint, and their union is the set {domino tilings of Rn,2 }. Hence,
Theorem 1.1.3 (applied to S = {domino tilings of Rn,2 }, k = 2, S1 = {type-1 tilings}
and S2 = {type-2 tilings}) yields
|{domino tilings of Rn,2 }| = |{type-1 tilings}| + |{type-2 tilings}| .
In other words,
(# of domino tilings of Rn,2 ) = (# of type-1 tilings) + (# of type-2 tilings) .
This proves (13).

Proposition 1.1.9 lets us compute the numbers dn,2 rather easily, if we compute
them in the appropriate order (i.e., start with d0,2 and d1,2 , then compute d2,2 , then
compute d3,2 , then compute d4,2 , and so on). For example, we get
d5,2 = d4,2 + d3,2 = 5 + 3 = 8;
|{z} |{z}
=5 =3
d6,2 = d5,2 + d4,2 = 8 + 5 = 13;
|{z} |{z}
=8 =5
d7,2 = d6,2 + d5,2 = 13 + 8 = 21;
|{z} |{z}
=13 =8
....
But what if we want to compute (say) d900,2 without having to first compute all the
previous numbers d0,2 , d1,2 , . . . , d899,2 ? Is there an explicit formula?
Before we answer this question, let us forget for a moment about domino tilings,
and define the following sequence of integers:
Enumerative Combinatorics: class notes page 23

Definition 1.1.10. The Fibonacci sequence is the sequence ( f 0 , f 1 , f 2 , . . .) of nonneg-

ative integers defined recursively by

f 0 = 0, f 1 = 1, and f n = f n−1 + f n−2 for all n ≥ 2.

This is a recursive definition – i.e., it tells us how to compute f n assuming that

the previous entries f 0 , f 1 , . . . , f n−1 of the sequence are already known. Thus, if we
want to compute f 5 using this definition, we have to compute f 0 , f 1 , f 2 , f 3 , f 4 first.
(Let us do this: The definition yields f 0 = 0 and f 1 = 1 immediately. Furthermore,
setting n = 2 in the equality f n = f n−1 + f n−2 , we obtain f 2 = f 1 + f 0 =
|{z} |{z}
=1 =0
1 + 0 = 1. Next, setting n = 3 in the equality f n = f n−1 + f n−2 , we obtain f 3 =
f 2 + f 1 = 1 + 1 = 2. Likewise, f 4 = f 3 + f 2 = 2 + 1 = 3. Likewise,
|{z} |{z} |{z} |{z}
=1 =1 =2 =1
f 5 = f 4 + f 3 = 5.)
|{z} |{z}
=3 =2
The entries f n of the Fibonacci sequence are called Fibonacci numbers. Here is a
table of the first 10 Fibonacci numbers:

n 0 1 2 3 4 5 6 7 8 9 ···
.
fn 0 1 1 2 3 5 8 13 21 34 ···

The Fibonacci sequence is famous – just look at its Wikipedia page! It also has
several books dedicated to it, such as Vorobiev’s [Vorobi02] (although, to be fully
honest, Vorobiev often uses it as a plug to pivot to other mathematics); there is also
a journal called The Fibonacci Quarterly (again, however, its actual scope is broader).
Now, we can reduce our problem of computing dn,2 to the problem of computing
Fibonacci numbers:

Proposition 1.1.11. We have dn,2 = f n+1 for each n ∈ N.

Proof of Proposition 1.1.11 (informal version). Here is the main idea of the proof: We
must show that the two sequences (d0,2 , d1,2 , d2,2 , d3,2 , . . .) and ( f 1 , f 2 , f 3 , f 4 , . . .) are
identical. Both of them have the property that their first entries are 1’s (that is,
d0,2 = 1 and f 1 = 1), their second entries are 1’s (that is, d1,2 = 1 and f 2 = 1), and
each of their further entries equals the sum of the preceding two entries (because
Proposition 1.1.9 shows that the dn,2 satisfy dn,2 = dn−1,2 + dn−2,2 , whereas the
definition of Fibonacci numbers shows that f n+1 = f n + f n−1 ). Thus, to put it in
practical terms: Both sequences start with the same two entries, and then are built
out of these two entries according to the same rule (namely, each further entry is
the sum of the preceding two entries). Hence, the two sequences must be the same.
This proves Proposition 1.1.11.
If you found this proof insufficiently rigorous, here is a formal version of this argument:
Enumerative Combinatorics: class notes page 24

Proof of Proposition 1.1.11 (formal version). We shall prove Proposition 1.1.11 by strong in-
duction on n.
A strong induction needs no induction base20 . Thus, we only do the induction step:
Induction step: Let m ∈ N. Assume (as the induction hypothesis) that Proposition 1.1.11
holds for each n < m. We must prove that Proposition 1.1.11 holds for n = m.
Our induction hypothesis says that Proposition 1.1.11 holds for each n < m. In other
words, we have
dn,2 = f n+1 for each n ∈ N satisfying n < m. (15)
Now, we must prove that Proposition 1.1.11 holds for n = m. In other words, we must
prove that dm,2 = f m+1 . If m = 0, then this is true (since d0,2 = 1 = f 1 = f 0+1 ). If m = 1,
then this is also true (since d1,2 = 1 = f 2 = f 1+1 ). Hence, it remains to prove this in the
case m ≥ 2. So let us WLOG21 assume that m ≥ 2. Then, m − 2 ∈ N. Hence, we can apply
(15) to n = m − 2 (since m − 2 < m), and obtain dm−2,2 = f (m−2)+1 = f m−1 . Furthermore,
m − 1 ∈ N (since m ≥ 2 ≥ 1). Thus, we can apply (15) to n = m − 1 (since m − 1 < m), and
obtain dm−1,2 = f (m−1)+1 = f m . However, Proposition 1.1.9 (applied to n = m) shows that

dm,2 = dm−1,2 + dm−2,2 = f m + f m−1 .

| {z } | {z }
= fm = f m −1

Comparing this with

f m+1 = f (m+1)−1 + f (m+1)−2 (by the definition of the Fibonacci sequence)

| {z } | {z }
= fm = f m −1

= f m + f m −1 ,

we obtain dm,2 = f m+1 . In other words, Proposition 1.1.11 holds for n = m. This completes
the induction step. Thus, Proposition 1.1.11 is proven.

So we have identified our numbers dn,2 as the famous Fibonacci numbers f n+1 .
Does this help us compute them directly? Yes, because there is a famous formula
for the Fibonacci numbers:

Theorem 1.1.12 (Binet’s formula). For each n ∈ N, we have

1
f n = √ ( ϕn − ψn ) ,
5
where
√ √
1+ 5 1− 5
ϕ= ≈ 1.618 . . . and ψ= ≈ −0.618 . . . .
2 2

20 See [Grinbe15, §2.8] for how strong induction works.

21 “WLOG” means “without loss of generality”. We can assume that m ≥ 2 without loss of general-
ity, since we have already proven our claim (that dm,2 = f m+2 ) in all other cases.
Enumerative Combinatorics: class notes page 25

A few words about this strange formula are in order. On its left hand side is
a nonnegative √ integer, f n . On its right side is an expression involving irrational
numbers like 5 as well as minus signs. How could an explicit formula for a
sequence of nonnegative integers require irrational numbers and subtraction?
Well, this is the price of asking for explicit formulas!
The numbers ϕ and ψ in Theorem 1.1.12 are known as the golden ratios (although
usually only ϕ is considered “the golden ratio”). They are the two roots of the
quadratic polynomial x2 − x − 1, so they satisfy ϕ2 = ϕ + 1 and ψ2 = ψ + 1. If this
looks like the Fibonacci recursion f n = f n−1 + f n−2 , don’t be surprised! This is the
reason why they show up in the explicit formula for the Fibonacci numbers.
How do you compute f 900 using Theorem 1.1.12? You may be tempted to just plug
n = 900 into the formula √ using your favorite computer algebra system; but there is a
subtlety involved: Since 5 is irrational, the computer may try to work with approximate
values, and then when you take n-th powers, the rounding errors will blow up. You will
probably not get an integer as a result, and even if you try to round it to the nearest
integer, you may well get the wrong √ value! Such is the price of blindly trusting floating-
point arithmetic. Fortunately, 5 is an algebraic number (i.e., a root of a polynomial with
rational coefficients), and this means that it is √possible to make exact computations with
it. You
√ just need to restrict yourself to the “ 5-rationals” (i.e., the numbers of the form
a + b 5 with a, b√∈ Q), and instead of approximating them with decimals, you just keep
√ a + b 5 form. It is easy to find rules for adding, subtracting, multiplying and
them in the
dividing 5-rationals by one another22 ; thus you don’t need approximate values. Using
the “exponentiation by squaring” trick, it is now easy to compute very high powers of ϕ
and ψ, and then Theorem 1.1.12 yields f n . For example, a computer will readily tell you
all the 188 digits of f 900 , the last six of which are 938800; you would probably have gotten
these wrong if you relied on approximate computation.
Theorem 1.1.12 also gives a very easy answer to the question (which we arguably haven’t
asked) how fast the Fibonacci numbers f n grow when n gets large. Indeed, |ψ| < 1, so that
1
ψn → 0 when n → ∞. Hence, for high enough n, we have f n ≈ √ ϕn . Of course, the
5
1 n
sequence of √ ϕ for n ∈ N is a geometric sequence with ratio ϕ ≈ 1.618 . . ., and grows
5
exponentially. Thus, we see that the Fibonacci numbers f n grow exponentially – slower than

22 To wit:
 √   √  √
a + b 5 + c + d 5 = ( a + c) + (b + d) 5;
 √   √  √
a + b 5 − c + d 5 = ( a − c) + (b − d) 5;
 √  √  √
a + b 5 c + d 5 = ( ac + 5bd) + ( ad + bc) 5;
√
 √  √  √
a+b 5 a + b 5 c − d 5 ( ac − 5bd) + (bc − ad) 5
√ =  √  √  = .
c+d 5 c+d 5 c−d 5 c2 − 5d2

Note in particular the last equation: This is why they taught you to rationalize denominators in
high school!
Enumerative Combinatorics: class notes page 26

the powers of 2 (since 1.618 . . . < 2), but still faster than any polynomial. The Fibonacci


number f n will have ≈ log10 ϕ · n ≈ 0.209 · n many digits.
Theorem 1.1.12 is easy to prove:
Proof of Theorem 1.1.12 (sketched). This can be proven by the same argument that
 for Proposition 1.1.11: We have to show thatthe sequences ( f 0 , f 1 , f 2 , . . .)
we used
1
1
1
and √ ϕ0 − ψ0 , √ ϕ1 − ψ1 , √ ϕ2 − ψ2 , . . . are identical. Both of them


5 5 5
have the property that their first entries are 0’s (this is easy to check), their second
entries are 1’s (this is easy to check), and each of their further entries equals the
sum of the preceding two entries23 . Thus, both sequences start with the same two
entries, and then are built out of these two entries according to the same rule.
Hence, the two sequences must be the same. This proves Theorem 1.1.12.
This proof should convince you that Theorem 1.1.12 holds, but does not explain
how you could have come up with Theorem 1.1.12. If time allows, we will later
explain this when we explore the concept of generating functions.

1.1.6. Kasteleyn’s formula (teaser)

Now we have computed dn,2 for each n ∈ N. What about dn,3 ?
Proposition 1.1.2 shows that dn,3 = 0 when n is odd. But computing dn,3 when n
is even is a lot harder. You might try to find a recursive formula such as Proposition
1.1.9, but this isn’t so easy any more. You can still try to separate the domino tilings
of Rn,3 into types according to how the last column looks like; however, there will be
three of these types now, and two of them will not fall into a 1-to-1 correspondence
with domino tilings of a smaller rectangle. For example, consider the following
domino tiling of R6,3 :

There is no tiling of R5,3 anywhere in it, nor of R4,3 , nor of R3,3 , nor of R2,3 , nor of
R1,3 . This thwarts our recursive approach.
Nevertheless, there is a nice recursion for dn,3 :
23 Indeed,
for the Fibonacci sequence ( f 0 , f 1 , f 2 , . . .), this is clear. For the second sequence, this boils
down to proving the identity

1 1   1  
√ ( ϕ n − ψ n ) = √ ϕ n −1 − ψ n −1 + √ ϕ n −2 − ψ n −2 ,
5 5 5
which however follows by subtracting the two easily verified identities

ϕ n = ϕ n −1 + ϕ n −2 and ψ n = ψ n −1 + ψ n −2
√
and dividing the result by 5.
Enumerative Combinatorics: class notes page 27

Proposition 1.1.13. We have dn,3 = 4dn−2,3 − dn−4,3 for each n ≥ 4.

As I said, there is no proof as easy as the one we gave for Proposition 1.1.9. We
will later learn the technique of generating functions, which can be used to give a rea-
sonably simple proof (see [Read80]), and probably a more complicated induction
could also do the trick.
What about dn,4 ? There is a recursion, too, according to [Read80, §2]:

dn,4 = dn−1,4 + 5dn−2,4 + dn−3,4 − dn−4,4 .

As you see, these are getting more complicated. In theory, recurrence relations like
these have explicit formulas like Binet’s formula for f n (Theorem 1.1.12). However,
these formulas become more and more complicated √as well, and in particular they
no longer involve “nice” irrational numbers like 5, but rather roots of higher-
degree polynomials, which at some point can no longer be expressed √ using rational
a
numbers, sums, differences, products, quotients and radicals (i.e., b terms)24 . This
looks like a dead end.
In the 20th Century, however, theoretical physicists studying thermodynamics
got interested in computing dn,m , as they considered a domino tiling to be a (rather
idealized) model for a liquid consisting of “dimers” (polymers that take up two
adjacent sites in a rectangular lattice; these are exactly our dominos). Even though
the 2-dimensionality of a rectangle makes it a somewhat unrealistic approximation
for real-world liquids, it found its use as a model for the adsorption of molecules
on a surface (see [Kastel61]). In 1961, Kasteleyn found the following surprising
formula for dn,m :

Theorem 1.1.14 (Kasteleyn’s formula). Assume that m is even and n ≥ 1. Then,

s
2 
m/2 n
kπ 2
 
jπ
dn,m = 2 mn/2
∏ ∏ cos m + 1 + cos n + 1 .
j =1 k =1

(Here, we are using the product sign ∏: That is, if a1 , a2 , . . . , a p are any numbers,
p
then ∏ ai means the product a1 a2 · · · a p . The presence of two product signs
i =1
directly following one another means that we are taking a product of products.)

Actually, “surprising” is an understatement; why cosines of angles would

√ appear
in a formula for the integer dn,m is even less transparent than why 5 should
appear in a formula for Fibonacci numbers!
24 The 2
√ of a quadratic polynomial x + ax + b can be expressed in this way: They are
roots
2
− a ± a − 4b
. The roots of a degree-3 or degree-4 polynomial can also be expressed in this
2
way, using complicated formulas due to Tartaglia, Ferrari, Cardano and Descartes. But the Abel–
Ruffini theorem says that for k ≥ 5, there is no formula that expresses the roots of a (general)
√
degree-k polynomial using only +, −, ·, / and .
Enumerative Combinatorics: class notes page 28

We won’t even get close to proving Theorem 1.1.14. A proof outline appears
in [Loehr11, Theorem 12.85], serving as a culmination of a graduate-level combi-
natorics textbook. A more self-contained exposition of the proof has been given
by Stucky in [Stucky15], but even that is only self-contained up to some advanced
linear algebra (Pfaffians and Kronecker products of matrices, as well as a good
understanding of eigenvectors are required).
For all its seeming extravagance, Theorem 1.1.14 actually provides a good way
jπ kπ
of computing dn,m . Indeed, cosines like cos and cos are algebraic num-
m+1 n+1
bers that lend themselves to exact computation (using cyclotomic polynomials – a
piece of abstract algebra we are also not coming close to), and the scary-looking
products don’t scare a good computer algebra system. For example, Kasteleyn’s
formula can be used to show that d8,8 = 12 988 816. This wouldn’t be so easy to
check by a brute force search for domino tilings!

Class of 2019-09-27

1.1.7. Axisymmetric domino tilings

Let us solve a few more counting exercises around domino tilings.

Exercise 1.1.1. Let n ∈ N. Say that a domino tiling T of Rn,2 is axisym-

metric if reflecting it across the vertical axis of symmetry of Rn,2 leaves it
unchanged (i.e., for each domino {(i, j) , (i0 , j0 )} ∈ T, the “mirror domino”
{(n + 1 − i, j) , (n + 1 − i0 , j0 )} also belongs to T).
For example, the tilings

and

are not axisymmetric (indeed, reflecting them across the vertical line transforms
them into one another, and they are not the same), but the tilings

and

are axisymmetric.
How many axisymmetric domino tilings does Rn,2 have?

Example 1.1.15. Let us list the axisymmetric domino tilings for Rn,2 when n is
Enumerative Combinatorics: class notes page 29

small:

their
n axisymmetric domino tilings
number

0 1

1 1

.
2 , 2

3 1

4 , , 3

Solution sketch to Exercise 1.1.1. Let us only show the main steps; a (more) detailed
solution to Exercise 1.1.1 can be found in [17f-hw1s, Exercise 5].
There are two cases to consider: the case when n is even, and the case when n is
odd.
Let us first consider the case when n is even. In this case, I claim that any
axisymmetric domino tiling of Rn,2 has one of the following two forms:

• Form 1: a domino tiling J of Rn/2,2 covering the left half of Rn,2 , and its mirror
image across the vertical axis covering the right half:

domino tiling J mirror image

.
of Rn/2,2 of J

(Note that the vertical axis cuts through the middle of this picture.)

• Form 2: a domino tiling J of Rn/2−1,2 covering the leftmost n/2 − 1 columns

of Rn,2 , and two horizontal dominos covering the (n/2)-th and (n/2 + 1)-th
columns25 , and the mirror image of J across the vertical axis covering the

25 We count columns from the left.

Enumerative Combinatorics: class notes page 30

rightmost n/2 − 1 columns of Rn,2 :

domino tiling J mirror image

.
of Rn/2−1,2 of J

(Again, the vertical axis cuts through the middle of this picture.)

If you agree with me that these are the only possible forms of an axisymmetric
domino tiling of Rn,2 , then it follows (by the same logic as in the proof of Proposi-
tion 1.1.9) that the # of axisymmetric domino tilings of Rn,2 is

(# of domino tilings of Rn/2,2 ) + (# of domino tilings of Rn/2−1,2 )

| {z } | {z }
=dn/2,2 = f n/2+1 =dn/2−1,2 = f n/2
(by Proposition 1.1.11, applied to n/2 (by Proposition 1.1.11, applied to n/2−1
instead of n) instead of n)
= f n/2+1 + f n/2 = f n/2+2
(by the recursive definition of the Fibonacci numbers). But why are the above-
mentioned two forms the only possible forms of an axisymmetric domino tiling of
Rn,2 ?
To prove this, we fix an axisymmetric domino tiling T of Rn,2 . We must show
that T has one of the above two forms. We can consider what dominos cover the
(n/2)-th column in T. If this column is covered by one vertical domino, then no
domino of T straddles the vertical axis, and thus the tiling T has Form 1 (indeed,
the part of T to the right of the vertical axis is a mirror image of the part to the
left, since T is axisymmetric). It remains to deal with the case when the (n/2)-th
column is covered by two horizontal dominos in T. The question is now where
these horizontal dominos fall. If both of them fall into columns n/2 − 1 and n/2,
then our tiling T has Form 1. If both of them fall into columns n/2 and n/2 + 1,
then our tiling T has Form 2. What about the remaining “rogue” possibility, that
one of them falls into columns n/2 − 1 and n/2, while the other falls into columns
n/2 and n/2 + 1 ? Here is a picture (which assumes that the former domino is in
the top row and the latter domino is in the bottom row; but the argument will be
the same in the opposite case):

some dominos some further

.
dominos

Note, however, that the part marked “some dominos” in this picture has an odd
number of squares (namely, 2 (n/2 − 1) − 1 many squares), so that it cannot ac-
tually be covered by dominos!26 Thus, we obtain a contradiction. Hence, this
“rogue” possibility is actually impossible. This shows that every axisymmetric
domino tiling T of Rn,2 either has Form 1 or has Form 2.
26 This is the same logic that we used to prove Proposition 1.1.2.
Enumerative Combinatorics: class notes page 31

Thus, we have proved that the # of axisymmetric domino tilings of Rn,2 is f n/2+2
in the case when n is even.
What about the case when n is odd? In this case, I claim that any axisymmetric
domino tiling of Rn,2 has the following form:

• Form 1: a domino tiling J of R(n−1)/2,2 covering the leftmost (n − 1) /2 columns

of Rn,2 , and one vertical domino covering the (n + 1) /2-th column, and the
mirror image of J across the vertical axis covering the rightmost (n − 1) /2
columns of Rn,2 :

domino tiling J mirror image

.
of R(n−1)/2,2 of J

(Again, the vertical axis cuts through the middle of this picture, which in this
case means cutting through the middle of the vertical domino.)

This time, there is no Form 2. Again, we need to prove that this is the only possible
form. This is even easier than in the previous case; the main idea is the following: If
an axisymmetric tiling T of Rn,2 contained any horizontal domino that intersects its
(n + 1) /2-th column, then it would also contain the mirror image of this domino
across the vertical axis (since T is axisymmetric), but then this domino and its
mirror image would be distinct but overlapping27 , which would contradict the
definition of a domino tiling. Thus, the (n + 1) /2-th column of any axisymmetric
tiling T of Rn,2 must be covered by a vertical domino. This shows that Form 1 is
the only possible form.
Thus, when n is odd, the # of axisymmetric domino tilings of Rn,2 equals the # of
domino tilings of R(n−1)/2,2 , which (by Proposition 1.1.11) is f (n−1)/2+1 = f (n+1)/2 .
Thus, the general formula for the # of axisymmetric domino tilings of Rn,2 is:
(
f n/2+2 , if n is even;
(# of axisymmetric domino tilings of Rn,2 ) = .
f (n+1)/2 , if n is odd

1.1.8. Tiling rectangles with k-bricks

Now, let us look at another related problem.
For the rest of Subsection 1.1.8, fix a positive integer k.
A k-brick shall mean a 1 × k-rectangle or a k × 1-rectangle. More specifically: A
vertical k-brick shall mean a 1 × k-rectangle (i.e., a set of the form

{(i, j) , (i, j + 1) , (i, j + 2) , . . . , (i, j + k − 1)}

27 They would overlap in the (n + 1) /2-th column.
Enumerative Combinatorics: class notes page 32

for some i, j ∈ Z); a horizontal k-brick shall mean a k × 1-rectangle (i.e., a set of the
form
{(i, j) , (i + 1, j) , (i + 2, j) , . . . , (i + k − 1, j)}
for some i, j ∈ Z). Here is how they look like:

; .
| {z }
horizontal k-brick
| {z }
vertical k-brick

If S is a set of squares, then a k-brick tiling of S means a way to cover the set
S with non-overlapping k-bricks (i.e., formally speaking: a set of disjoint k-bricks
whose union is S).
Thus, k-brick tilings are a generalization of domino tilings. More specifically: If
k = 2, then k-bricks are the same as dominos, and thus k-brick tilings are the same
as domino tilings.
When does a rectangle Rn,m have a k-brick tiling? The answer is surprisingly
simple:

Proposition 1.1.16. Let n, m ∈ N, and let k be a positive integer. Then, the

rectangle Rn,m has a k-brick tiling if and only if we have k | m or k | n.

In the case when k = 2, Proposition 1.1.16 says that the rectangle Rn,m has a
domino tiling if and only if we have 2 | m or 2 | n (that is, at least one of the
numbers m and n is even). This should not be surprising: One direction of this
equivalence (namely, the “=⇒” direction: i.e., the direction saying that if Rn,m has
a domino tiling, then at least one of m and n is even) follows from Proposition
1.1.2, whereas the other direction is easy. We could use a similar argument to
prove Proposition 1.1.16 whenever the number k is prime. Indeed, our proof of
Proposition 1.1.2 generalizes to show that if the rectangle Rn,m has a k-brick tiling,
then k | mn. When k is prime, the divisibility k | mn implies that k | m or k | n;
thus, the “=⇒” direction of Proposition 1.1.16 would follow immediately in this
case. But when k is not prime, we need a better argument.
Proof of Proposition 1.1.16. The claim we want to prove is an “if and only if” claim,
so it has two directions: the “⇐=” direction (also known as the “if” direction), and
the “=⇒” direction (also known as the “only if” direction). The former direction
claims that if we have k | m or k | n, then the rectangle Rn,m has a k-brick tiling. The
latter claims the converse of this statement.
We shall prove these two directions separately:
⇐=: Assume that k | m or k | n.
Enumerative Combinatorics: class notes page 33

If k | m, then Rn,m has a k-brick tiling consisting entirely of vertical k-bricks. It

looks as follows:

···

···

.. .. .. ..
.. .. . ..
.. .. ..

···

(with each column being covered by m/k many vertical k-bricks). Similarly, if k | n,
then Rn,m has a k-brick tiling consisting entirely of horizontal k-bricks. Thus, in
either case, Rn,m has a k-brick tiling. This proves the “⇐=” direction of Proposition
1.1.16.
=⇒: Assume that the rectangle Rn,m has a k-brick tiling. We must prove that we
have k | m or k | n.
Assume the contrary (for the sake of contradiction). Thus, k - m and k - n.
We shall use the following notations from elementary number theory: If a is an
integer and b is a positive integer, then

• we let a // b denote the quotient obtained when dividing a by b (in the sense
of division with remainder);

• we let a%b denote the remainder obtained when dividing a by b.

Both of these numbers a // b and a%b are integers, and they satisfy 0 ≤ a%b < b
and a = ( a // b) · b + ( a%b).
Let r = n%k and s = m%k. We have r = n%k 6= 0 (since k - n) and thus 0 < r < k
(since r is a remainder upon division by k). Likewise, 0 < s < k.
Now, we are going to color the squares of Rn,m with k colors. The k colors we
are going to use will be numbered 0, 1, . . . , k − 1. Each square (i, j) ∈ Rn,m will be
colored with the color (i + j − 2) %k. Here is how this coloring looks like (in the
Enumerative Combinatorics: class notes page 34

example where k = 4, n = 10 and m = 7):

2 3 0 1 2 3 0 1 2 3
1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1
3 0 1 2 3 0 1 2 3 0
2 3 0 1 2 3 0 1 2 3
1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1

(where we have written the color of each square as a number into this square). Note
that if k = 2, then there are two colors only, numbered 0 and 1; in this case, our
coloring is precisely the usual chessboard coloring (if we regard color 0 as black
and color 1 as white).
Let us make a few observation about how our coloring behaves:

• The southwesternmost square Rn,m is (1, 1), and thus has color (1 + 1 − 2) %k =
0%k = 0.

• As we move eastwards, the colors of the squares increase by 1 at each step,

until they reach k − 1, at which point they “fall back down” to 0 at the next
step.

• The same happens as we move northwards.

• Along each “northwest-to-southeast” diagonal (i.e., each line with slope −1),
the color stays constant.

Let us say that a finite set S of squares is balanced if it has equally many squares
of each color. In other words, a finite set S of squares is balanced if and only if for
any color h ∈ {0, 1, . . . , k − 1}, the # of all squares in S that have color h does not
depend on h.
Each horizontal k-brick has exactly 1 square of each color. Indeed, the colors of
the k squares in a horizontal k-brick look as follows:

u u+1 ··· k−1 0 1 ··· u−1

(where u is the color of the leftmost square of the k-brick). Thus, each horizontal k-
brick is balanced. Likewise, each vertical k-brick is balanced. Thus, we have shown
that each k-brick is balanced.
Enumerative Combinatorics: class notes page 35

But we have assumed that the rectangle Rn,m has a k-brick tiling. Hence, Rn,m
must, too, be balanced (by the sum rule)28 .
Let us now subdivide the rectangle Rn,m into several (disjoint) zones Zu,v by cut-
ting it with several lines29 . Namely, we cut it with horizontal lines every k squares
(counted from the bottom), and with vertical lines every k squares (counted from
the left). Thus, we obtain the following subdivision of Rn,m :

Z0,2 Z1,2 Z2,2 Z3,2

Z0,1 Z1,1 Z2,1 Z3,1

.

Z0,0 Z1,0 Z2,0 Z3,0

Formally speaking, our zones Zu,v are defined as follows:

Zu,v = {(i, j) ∈ Rn,m | (i − 1) // k = u and ( j − 1) // k = v}

for all u ∈ {0, 1, . . . , (n − 1) // k } and v ∈ {0, 1, . . . , (m − 1) // k } .
28 Here is the argument in details (albeit using the summation sign, which we won’t properly
introduce until Definition 1.2.2 below):
We have assumed that the rectangle Rn,m has a k-brick tiling. Let T be this k-brick tiling. Write
T in the form T = S1 , S2 , . . . , S j , where S1 , S2 , . . . , S j are distinct k-bricks. Hence, each square
in Rn,m belongs to exactly one of S1 , S2 , . . . , S j . Now, for each color h ∈ {0, 1, . . . , k − 1}, we have

(# of all squares in Rn,m that have color h)

= (# of all squares in Rn,m that belong to S1 and have color h)
+ (# of all squares in Rn,m that belong to S2 and have color h)
+···


+ # of all squares in Rn,m that belong to S j and have color h
!
by the sum rule, since each square in Rn,m
belongs to exactly one of S1 , S2 , . . . , S j
j
= ∑ (|# of all squares in Rn,m that {z
belong to Si and have color h)
i =1
}
=(# of all squares in Si that have color h)
(since Si ⊆ Rn,m )
j
= ∑ |(# of all squares in{zSi that have color h}) .
i =1
independent of h
(since Si is balanced)

Hence, for each h ∈ {0, 1, . . . , k − 1}, the # of all squares in Rn,m that have color h is independent
of h (since it is a sum of j numbers that are each independent of h). In other words, Rn,m is
balanced.
29 We will give a rigorous definition further below.
Enumerative Combinatorics: class notes page 36

All these zones Zu,v are rectangles. More precisely: Let us call a zone Zu,v (with
u ∈ {0, 1, . . . , (n − 1) // k } and v ∈ {0, 1, . . . , (m − 1) // k })

• generic if u < (n − 1) // k and v < (m − 1) // k;

• northern if u < (n − 1) // k and v = (m − 1) // k;

• eastern if u = (n − 1) // k and v < (m − 1) // k;

• northeastern if u = (n − 1) // k and v = (m − 1) // k.

Then,30

• each generic zone is a k × k-rectangle;

• each northern zone is a k × s-rectangle;

• each eastern zone is an r × k-rectangle;

• the northeastern zone is an r × s-rectangle.

Note that all these zones are nonempty, since r > 0 and s > 0 and k > 0.
Each northern zone is a k × s-rectangle, and thus can be tiled with horizontal
k-bricks; therefore it is balanced (since each k-brick is balanced). Likewise, each
eastern zone is balanced, and each generic zone is balanced. Thus, summarizing,
we have shown that each zone Z except for the northeastern zone is balanced. In
other words, if Z is a zone that is not the northeastern zone, then, for any color
h ∈ {0, 1, . . . , k − 1},

the # of all squares in Z that have color h is independent of h. (16)

But the zones are disjoint and their union is the whole rectangle Rn,m . Hence, it
follows that the northeastern zone is balanced as well31 .
But let us take a closer look at the northeastern zone. We denote this zone by Z.
This zone Z is an r × s-rectangle, and (just as for any zone) its southwestern corner

30 Recall that r = n%k and s = m%k.

31 Proof. For any color h ∈ {0, 1, . . . , k − 1}, we have

(# of all squares in Rn,m that have color h)

= ∑ (# of all squares in Z that have color h)
Z is a zone
= (# of all squares in the northeastern zone that have color h)
+ ∑ (# of all squares in Z that have color h) ,
Z is a zone;
Z is not the northeastern zone
Enumerative Combinatorics: class notes page 37

has color 0. Thus, the colors of its squares look as follows:

s−1 s ∗ ··· ∗ ∗

s−2 s−1 s ··· ∗ ∗

.. .. .. ... .. ..
. . . . . (17)

1 2 3 ··· r−1 r

0 1 2 ··· r−2 r−1

(where each asterisk ∗ stands for some entry we don’t need to know about).
We shall now show that this northeastern zone is not balanced. This will give us
a contradiction.
We WLOG assume that s ≤ r (since otherwise, the argument is similar32 ). Recall
that 0 < r < k and 0 < s < k. From 0 < s, we obtain s ≥ 1, so that s − 1 ≤ 0. Thus,
s −1 ≤ r − 1. Therefore, each row of Z contains at least one square with
0 ≤ |{z}
≤r
color s − 1. Since Z has s rows, we thus conclude that

Z contains at least s squares with color s − 1. (18)

(We can easily see that Z contains exactly s squares with color s − 1, but we won’t
need this.)
On the other hand, the bottommost row of Z only contains squares with colors
0, 1, . . . , r − 1. Hence, it does not contain a square with color r (since r > r − 1). But
each row of Z has width r < k, and thus contains at most one square with color
r (because if it contained two distinct squares with color r, then these two squares
would be a distance of ≥ k apart from one another, but the row only has width
r < k). So we know that the zone Z has s rows, one of which (the bottommost one)
so that

(# of all squares in the northeastern zone that have color h)

= (# of all squares in Rn,m that have color h)
| {z }
independent of h
(since Rn,m is balanced)

− ∑ (# of all squares in Z that have color h),

Z is a zone; | {z }
Z is not the northeastern zone independent of h
(by (16))

and thus the # of all squares in the northeastern zone that have color h is independent of h. In
other words, the northeastern zone is balanced.
32 More precisely, the same argument applies if we interchange the roles of rows and columns.
Enumerative Combinatorics: class notes page 38

contains no square with color r, while each of the other s − 1 rows contains at most
one square with color r. Hence, altogether,
Z contains at most s − 1 squares with color r. (19)
Comparing this with (18), we conclude that the # of squares with color r contained
in Z is different from the # of squares with color s − 1 contained in Z (since the for-
mer number is ≤ s − 1, while the latter number is ≥ s). Hence, Z is not balanced.
In other words, the northeastern zone is not balanced (since Z is the northeast-
ern zone). This contradicts the fact that the northeastern zone is balanced. This
contradiction shows that our assumption was false. Hence, the “=⇒” direction of
Proposition 1.1.16 is proven.
This was just a little taste of the theory of tilings of discrete (plane) shapes. See
the survey [ArdSta10] by Ardila and Stanley for an introduction to the varied ques-
tions and ideas of this theory. See also [18f-hw1s, Exercises 4 and 5] and [19f-hw1s,
Exercises 1, 2, 3] for further exercises on counting tilings, and [LeLeMe16, §5.1.5] for
a neat exercise in proving existence of tilings. Furthermore, the book [BenQui03]
by Benjamin and Quinn provides a lot of applications of tilings to enumerative
combinatorics, such as proofs of identities between Fibonacci numbers using their
domino-tiling interpretation (Proposition 1.1.11)33 . (See [BenQui04] for a “best-of”
of sorts.)

Class of 2019-09-30

1.2. Sums of powers

1.2.1. The sum 1 + 2 + ... + n
We now switch the subject and recall a famous result, known colloquially as the
“Little Gauss” formula due to the anecdote of Gauss inventing it in primary school:34

Theorem 1.2.1 (“Little Gauss” formula). Let n ∈ N. Then,

n ( n + 1)
1+2+···+n = .
2

Keep in mind that 0 ∈ N according to our conventions; thus, Theorem 1.2.1

applies to n = 0. And indeed, Theorem 1.2.1 holds for n = 0, since an empty sum
(i.e., a sum that consists of no addends) is defined to be 0, and thus we have
0 · (0 + 1)
1 + 2 + · · · + n = 1 + 2 + · · · + 0 = (empty sum) = 0 = .
2
33 To be more precise, Benjamin and Quinn work not with domino tilings of Rn,2 , but rather with
“square-and-domino tilings” of Rn,1 . These objects behave just as domino tilings of Rn,2 do
(there is an easy bijection between the latter and the former), but are easier to argue about.
34 In truth, this formula was known to the Ancient Greeks.
Enumerative Combinatorics: class notes page 39

First proof of Theorem 1.2.1. Induction on n. The details are completely straightfor-
ward and LTTR.
(The abbreviation “LTTR” stands for “left to the reader”. I will usually leave
arguments to the reader when they are straightforward or easy variations of argu-
ments shown before.)
Second proof of Theorem 1.2.1. We observe the following fact: If a1 , a2 , . . . , an are n
numbers (say, real numbers), and b1 , b2 , . . . , bn are n further numbers, then

( a1 + a2 + · · · + an ) + (b1 + b2 + · · · + bn )
= ( a1 + b1 ) + ( a2 + b2 ) + · · · + ( an + bn ) . (20)

(Indeed, both sides of this equality are just two ways to add all the 2n numbers
a1 , a2 , . . . , an , b1 , b2 , . . . , bn . This should convince you that they are the same, al-
though formally speaking, this is not a proof. See Subsection 1.2.2 below for some
references to a formal proof.)
Now,

2 · (1 + 2 + · · · + n )
= (1 + 2 + · · · + n ) + (1 + 2 + · · · + n )
| {z }
=n+(n−1)+···+1
(here, we have just reversed
the order of summation)
= (1 + 2 + · · · + n ) + ( n + ( n − 1) + · · · + 1)
= (1 + n) + (2 + (n − 1)) + · · · + (n + 1)
(by (20), applied to ai = i and bi = n + 1 − i )
= ( n + 1) + ( n + 1) + · · · + ( n + 1)
| {z }
n many addends

(since each of the numbers 1 + n, 2 + (n − 1) , . . . , n + 1 equals n + 1)

= n ( n + 1) .

n ( n + 1)
Dividing this equality by 2, we obtain 1 + 2 + · · · + n = . This proves
2
Theorem 1.2.1 again.
Third proof of Theorem 1.2.1 (sketched). Here is a picture proof (drawn for the case
n = 3):
∗ ∗ ∗ ∗
∗ ∗ ∗ ∗
.
∗ ∗ ∗ ∗
Enumerative Combinatorics: class notes page 40

This is a 4 × 3-rectangle35 (i.e., a rectangle of width 4 and height 3), subdivided into
two parts by a broken line which starts in the southwestern corner and winds its
way to the northeastern corner, making steps of length 1 eastwards and northwards
(by turns). The two parts (the one below and the one above the broken line) have
the same area, since they are symmetric to each other with respect to the center
of the square.36 Hence, the area of either part equals half the area of the whole
rectangle. Since the area of the whole rectangle is 3 · 4, we thus conclude that the
3·4
area of either part equals .
2
On the other hand, here is a different way to compute this area: Let us look at
the part below the broken line. This part has 0 squares in the 1-st column37 , 1
square in the 2-nd column, 2 squares in the 3-rd column, and 3 squares in the 4-th
column. Hence, in total, it has 0 + 1 + 2 + 3 many squares. In other words, its area
is 0 + 1 + 2 + 3.
Now we know that the area of the part of the rectangle below the broken line is
3·4
0 + 1 + 2 + 3, but we also know (from before) that it equals . Hence, 0 + 1 +
2
3·4 3·4
2+3 = . In view of 0 + 1 + 2 + 3 = 1 + 2 + 3, this rewrites as 1 + 2 + 3 = .
2 2
This is precisely the statement of Theorem 1.2.1 for n = 3.
The same argument (but using an (n + 1) × n-rectangle instead of a 4 × 3-rectangle)
proves Theorem 1.2.1 for arbitrary n.
How can we formalize this argument? Once again (as with domino tilings), it makes
sense to think of the (n + 1) × n-rectangle as the finite set [n + 1] × [n] (where, as before,
we set [k ] = {1, 2, . . . , k } for each k ∈ N) rather than a geometric shape in the real plane.38
The two parts into which this rectangle is divided by the broken line become the two sets

A := {(i, j) ∈ [n + 1] × [n] | i ≤ j} and

B := {(i, j) ∈ [n + 1] × [n] | i > j} .

(The letters A and B here stand for “above” and “below”.)

So let us redo this proof from scratch, using the combinatorial model (i.e., finite sets)
throughout it. We consider the finite set [n + 1] × [n], which has (n + 1) · n = n (n + 1)
many elements. Define two subsets

A := {(i, j) ∈ [n + 1] × [n] | i ≤ j} and

B := {(i, j) ∈ [n + 1] × [n] | i > j}

of [n + 1] × [n]. These two subsets A and B are disjoint (since no (i, j) ∈ [n + 1] × [n] can
satisfy i ≤ j and i > j at the same time), and their union is [n + 1] × [n] (since each (i, j) ∈
35 We have put an asterisk into each little square of the rectangle in order to make the squares easier
to discern.
36 Note that “area” is normally a geometric concept, but since both parts consist of full squares, we

can redefine it combinatorially as the number of squares in the respective part.

37 We count columns from the left.
38 Again, we let (i, j ) be the square in column i (counted from the left) and row j (counted from the

bottom).
Enumerative Combinatorics: class notes page 41

[n + 1] × [n] satisfies either i ≤ j or i > j). Hence, the sum rule shows that |[n + 1] × [n]| =
| A| + | B|, so that

| A| + | B| = |[n + 1] × [n]| = |[n + 1]| · |[n]| (by (5))

| {z } |{z}
= n +1 =n
= ( n + 1) · n = n ( n + 1) . (21)

On the other hand, the map

A → B,
(i, j) 7→ (n + 2 − i, n + 1 − j)

(which, visually, is just the reflection around the center of the (n + 1) × n-rectangle) is a
bijection39 . Hence, the bijection rule yields | A| = | B|. Thus, | A| + | B| = | B| + | B| = 2 · | B|.
|{z}
=| B|
Comparing this with (21), we obtain 2 · | B| = n (n + 1), so that

n ( n + 1)
| B| = . (22)
2
On the other hand, let us count the squares in B “by column”. Recall that, in our
combinatorial model, the x-coordinate i of a square (i, j) tells us which column it belongs to.
Thus, the squares c ∈ B in the k-th column (for any given k ∈ {1, 2, . . . , n + 1}) are precisely
the squares c ∈ B that have x-coordinate k. Hence, we should count the squares in B
according to their x-coordinate. Formally speaking, this means applying the sum rule to the
set B and its n + 1 disjoint subsets {c ∈ B | c has x-coordinate k } for k ∈ {1, 2, . . . , n + 1}.
These n + 1 subsets are disjoint (since any square c ∈ B has only one x-coordinate) and
their union is B (since each square c ∈ B has x-coordinate 1 or 2 or · · · or n + 1). Hence,
the sum rule yields

| B| = |{c ∈ B | c has x-coordinate 1}| + |{c ∈ B | c has x-coordinate 2}|

+ · · · + |{c ∈ B | c has x-coordinate n + 1}|
n +1
= ∑ |{c ∈ B | c has x-coordinate i }| . (23)
i =1

But the addends on the right hand side of this equality are easily computed:40 For each
39 Thisneeds to be proven, but the proof is really straightforward. First, you need to check that this
map is well-defined. This means showing that (n + 2 − i, n + 1 − j) ∈ B for each (i, j) ∈ A. Once
this is shown, you can show the bijectivity of this map by explicitly constructing an inverse;
namely, the inverse is the map

B → A,
(i, j) 7→ (n + 2 − i, n + 1 − j) .
(Don’t be surprised that it is given by the same formula: In order to undo a reflection around a
point, you have to reflect again around the same point.)
40 This is just the formalization of our (visually obvious) observation that the part below the broken

line (which we are now calling B) has 0 squares in the 1-st column, 1 square in the 2-nd column,
2 squares in the 3-rd column, etc..
Enumerative Combinatorics: class notes page 42

i ∈ {1, 2, . . . , n + 1}, we have

{c ∈ B | c has x-coordinate i }
= {c ∈ B | c = (i, j) for some j ∈ [n]}

since a square of [n + 1] × [n] has x-coordinate i


if and only if it has the form (i, j) for some j ∈ [n]

= {(i, j) | j ∈ [n] such that (i, j) ∈ B}
= {(i, j) | j ∈ [n] such that i > j}
 
since a square (i, j) ∈ [n + 1] × [n] satisfies (i, j) ∈ B
if and only if i > j (by the definition of B)
= {(i, 1) , (i, 2) , . . . , (i, i − 1)}

and thus

|{c ∈ B | c has x-coordinate i }| = |{(i, 1) , (i, 2) , . . . , (i, i − 1)}|

= i − 1. (24)

Hence, (23) becomes

n +1
| B| = ∑ |{ c∈B | c has x-coordinate i }|
i =1
| {z }
= i −1
(by (24))
n +1
= ∑ (i − 1) = 0 + 1 + 2 + · · · + n = 1 + 2 + · · · + n.
i =1

n ( n + 1)
Comparing this with (22), we obtain 1 + 2 + · · · + n = . This finally completes our
2
formalized picture proof of Theorem 1.2.1.
The moral of the story: A picture is worth a thousand words!

1.2.2. What is a sum, actually?

Throughout Subsection 1.2.1, we have been freely working with expressions like
1 + 2 + · · · + n and (more generally) a1 + a2 + · · · + an , where a1 , a2 , . . . , an are
n numbers. Looking back, you might wonder: Why are these expressions well-
defined? How is “the sum of n numbers” defined to begin with?
Let me explain what I mean by this. Let “number” mean “rational number”,
just to be specific here (although the same question and the same answer apply to
integers, real numbers and complex numbers). I assume you know what the sum
of two numbers is. Even when two numbers a and b are given without specifying
their order, their sum is well-defined, because the commutativity of addition (i.e., the
rule saying that a + b = b + a for any two numbers a and b) guarantees that the
two possible ways of adding them together yield the same result.
Enumerative Combinatorics: class notes page 43

Now, suppose you are given three numbers a, b and c. To add them all together
means to add two of them and then add the third one to the result. How many
ways are there to do this? There are 12, as you can easily check, namely

( a + b) + c, ( a + c) + b, (b + a) + c, (b + c) + a,
(c + a) + b, (c + b) + a, a + (b + c) , a + (c + b) ,
b + ( a + c) , b + (c + a) , c + ( a + b) , c + (b + a) .

If we want to make sense of a “sum of three numbers” without having to specify

the precise procedure of summation, we better hope that these 12 ways all lead to
the same result! And indeed, they do. This is not hard to derive from the above-
mentioned commutativity of addition, combined with the associativity of addition
(i.e., the rule saying that ( a + b) + c = a + (b + c) for any three numbers a, b and
c). For example,

a + (b + c) = ( a + b) +c (by associativity)
| {z }
=b+ a
(by commutativity)
= (b + a) + c = b + ( a + c) (by associativity)
| {z }
=c+ a
(by commutativity)
= b + (c + a) = (b + c) + a (by associativity)

and so on (in the sense that similar reasoning shows that all 12 ways give equal
results). Once we know that these 12 ways all lead to the same result, we can
call the result “the sum of the three numbers a, b and c” with a clear conscience,
and denote it by a + b + c without specifying the order in which the sum is taken
through well-placed parentheses.
Next, suppose you are given four numbers a, b, c and d. There are now 120 ways
of adding them together, including such ways as

(( a + b) + c) + d, ( a + b) + (c + d) , (b + d) + (c + a) , (b + (d + a)) + c

and many others. How can we tell that they all lead to the same result? They
do, and this can be proven as for three numbers (we don’t need any new rules;
commutativity and associativity suffice), but this is of course more laborious than
the case of three numbers.
Now, suppose you are given n numbers a1 , a2 , . . . , an , and you want to prove that
all ways of adding them together give the same result. For example, two of these
ways are

(· · · ((( a1 + a2 ) + a3 ) + a4 ) + · · · ) + an (“left-associative summation”)

and

a1 + (· · · + ( an−3 + ( an−2 + ( an−1 + an ))) · · · ) (“right-associative summation”);

Enumerative Combinatorics: class notes page 44

how do we know they are equivalent?

Let us close a ring around this problem. As long as we don’t know that all
ways to add n numbers give the same result, we cannot define “the sum” of n
numbers. But we can define the set of all possible sums of n numbers, i.e., the set
of all possible results that can be obtained by adding them together in all possible
orders. For example, for three numbers a, b, c, this set will be

{( a + b) + c, ( a + c) + b, . . . , c + (b + a)}

(containing all the 12 ways to add a, b, c together, listed above). This definition can
easily be done by recursion:

• There is only one way to add 0 numbers together. Namely, adding 0 numbers
always yields 0.

• There is only one way to add 1 number together. Namely, adding 1 number
a always yields a itself.

• If n > 1, then any way of adding n numbers together is given by splitting

them into two (disjoint nonempty) groups41 , then adding the numbers in
the first group together (in one of the many possible ways), then adding the
numbers in the second group together, and finally adding the two results
together.

Thus we can define the set of all possible sums of n numbers. Our goal is to prove
that there is only one such sum, i.e., that this set is a 1-element set. This is now a
rigorously stated claim42 , which we can try to prove! Thus, at the very least, we
have formalized our claim that we can add n numbers together in a well-defined
way. Proving it is another story, but it can be done with rather elementary tools.
See [Grinbe15, §2.14] for a very detailed proof43 (and [18s, lecture of 7th February
2018, pages 1–7] for a short version). Other proofs can be found in [Warner71,
Appendix A] and in [GalQua19, §3.3].
The upshot is that we have a well-defined notion of the sum of any finite family
of numbers, even if it is given without specifying an order. For example, “the sum
of all prime numbers smaller than 10” is well-defined (and equals 2 + 3 + 5 + 7 =
17). Likewise, if you are given a polygon, then “the sum of the sidelengths of the
polygon” is well-defined.
It is helpful to have a notation for these kinds of sums, so let us introduce it
(although we have already used it to some extent):

41 “Groups” in the sense of “batches”, not in the sense of group theory. For example, if our n
numbers are a, b, c, d, e, then the first group may be d, b and the second may be c, a, e.
42 known as the general commutativity theorem (“general” because it applies to any finite number of

addends rather than two)

43 More precisely, the proof is in [Grinbe15, §2.14.1–§2.14.6]. The remaining subsections of
[Grinbe15, §2.14] prove properties of sums.
Enumerative Combinatorics: class notes page 45

Definition 1.2.2. Let S be a finite set. For each s ∈ S, let as be a number (e.g., an
integer or a rational number or a real number or a complex number).
Then, ∑ as shall denote the sum of the numbers as for all s ∈ S. This notation
s∈S
is read as “the sum of as over all s ∈ S” or “the sum of as for s ranging over S” or
“the sum of as where s runs through S”. The “∑” sign in this notation is called
the summation sign.

Example 1.2.3. (a) We have ∑ s = 1 + 2 + 3 + 4 = 10.

s∈{1,2,3,4}
(b) We have ∑ s = 12 + 22 + 32
2 + 42 = 30.
s∈{1,2,3,4}
1 1 1 1 1 25
(c) We have ∑ = + + + = .
s∈{1,2,3,4} s 1 2 3 4 12
(d) We have ∑ s = 3 + 4 + 9 = 16.
s∈{3,4,9}

We need to note a few things about the summation sign:

• The letter “s” in the notation “ ∑ as ” is an instance of what is called a bound

s∈S
variable (or running index, or dummy variable): Its only purpose is to mark the
“moving part” of the sum. So it plays the same role as the letter “s” in the
set-comprehension notation “{s ∈ Z | s > 2}”, or the letter “s” in “the map
Z → Z, s 7→ s + 3”, or the letter “s” in the sentence “There exists no s ∈ Z
such that s2 = 2”. Thus, it is perfectly legitimate to replace it by any other
symbol (that is not used otherwise). For example, we can rewrite the sum
1 1 1 1 1
∑ as ∑ or as ∑ or as ∑ or as ∑ ;
s∈{1,2,3,4} s i ∈{1,2,3,4} i k∈{1,2,3,4} k S∈{1,2,3,4} S ♠∈{1,2,3,4} ♠
it will still be the same sum.

• In the notation “ ∑ as ”, the letter “s” is called the summation index; the set S
s∈S
is called the indexing set (or range) of the sum; and the numbers as are called
the addends of the sum. The whole expression ∑ as is called a finite sum.
s∈S

• Our definition of sums forces ∑ as to always be 0. This is called an empty

s ∈∅
sum. Thus, empty sums are 0. If this sounds like an arbitrary convention to
you, you should check that it is the only convention that makes (25) hold for
one-element sets S.

• Sums can have equal addends. For example,

∑ s2 = (−2)2 + (−1)2 + 02 + 12 + 22 = 4 + 1 + 0 + 1 + 4 = 10.

s∈{−2,−1,0,1,2}
Enumerative Combinatorics: class notes page 46

However,
∑ s = 02 + 12 + 22 = 0 + 1 + 4 = 5,
s∈{(−2)2 ,(−1)2 ,02 ,12 ,22 }
n o
2 2
because the set (−2) , (−1) , 02 , 12 , 22 is only a 3-element set (with ele-
ments 02 , 12 , 22 ).
Thus, if you want to write down a sum with some of its
entries equal, you still need to ensure that each of the addends gets a distinct
index in the indexing set. In general, the sum of 5 numbers is not the sum of
the elements of the set of these 5 numbers, because if some of these 5 numbers
are equal, then the set “forgets” that they appear more than once.
• The summation index does not always have to be a single letter. For instance,
if S is a set of pairs, then we can write ∑ a( x,y) (meaning the same as
( x,y)∈S
∑ as ). Here is an example of this notation:
s∈S
x 1 1 1 2 2 2 3 3 3
∑ y
= + + + + + + + + .
1 2 3 1 2 3 1 2 3
( x,y)∈{1,2,3}2

Here are some more examples, coming from more combinatorial questions:
Example 1.2.4. (a) Let P ( A) denote the powerset of a set A (that is, the set of all
subsets of A). Then, we can write the sum of the sizes of all subsets of {1, 2, 3}
as follows:

∑ | B| = |∅| + |{1}| + |{2}| + |{3}|

B∈P ({1,2,3})
+ |{1, 2}| + |{1, 3}| + |{2, 3}| + |{1, 2, 3}|
= 0 + 1 + 1 + 1 + 2 + 2 + 2 + 3 = 12.
(b) If A and B are two sets, then B A denotes the set of all maps from A to B.
For example, let A = {1, 2} and B = {1, 2}. Then, we can write the sum of the
sizes of the images of all maps from A to B as follows:

∑ | f ( A)|
f ∈B A
= |{1}| + |{1, 2}| + |{1, 2}| + |{2}|
| {z } | {z } | {z } | {z }
corresponding to the corresponding to the corresponding to the corresponding to the
! ! ! !
1 2 1 2 1 2 1 2
map map map map
1 1 1 2 2 1 2 2
= 1 + 2 + 2 + 1 = 6.
(Here, we have written out each map in two-line  notation:
 That is, the map
1 2
sending 1 and 2 to a1 and a2 has been written as .)
a1 a2
Enumerative Combinatorics: class notes page 47

There are several variants of the summation sign:

• The expression ∑ s2 stands for the sum of the squares of all odd
s∈{1,2,3,4,5,6};
s is odd
elements of {1, 2, 3, 4, 5, 6}. In general, if S is a set, and if A (s) is a logical
statement defined for each s ∈ S, then the notation ∑ as stands for the sum
s∈S;
A(s)
∑ as (that is, the sum of as not over all s ∈ S, but only over the ones
s∈{t∈S | A(t)}
which satisfy A (s)). For this sum to be well-defined, we do not need S to be
finite; we only need the subset {t ∈ S | A (t)} to be finite (i.e., we need there
to only be finitely many s ∈ S satisfying A (s)).
8
• The expression ∑ i2 stands for the sum ∑ i2 . In general, if p and q are
i =5 i ∈{5,6,7,8}
q
two integers, then the notation ∑ ai stands for the sum ∑ ai . Here
i= p i ∈{ p,p+1,...,q}
it should be kept in mind that the set { p, p + 1, . . . , q} is understood to be
empty if p > q. (Be warned that some authors have different conventions for
the latter case.)
q
In the notation ∑ ai , the integers p and q are called the bounds of the summa-
i= p
tion.
n
Note that the summation sign ∑ (for some n ∈ N) is thus equivalent to
i =1
∑ , hence also to ∑ .
i ∈{1,2,...,n} i ∈[n]

• The summation sign ∑ (where A is a given set) is shorthand for ∑ ,

B⊆ A B∈P ( A)
where P ( A) stands for the powerset of A (that is, the set of all subsets of A).
Thus, the result of Example 1.2.4 (a) could be rewritten as ∑ | B| = 12.
B⊆{1,2,3}

• The summation sign ∑ (where A and B are two sets) is shorthand for ∑ .
f :A→ B f ∈B A
Thus, the result of Example 1.2.4 (b) could be rewritten as ∑ | f ( A)| = 6.
f :A→ B

• Statements under the summation sign can be written out in words. For ex-
ample, “ ∑ ” means the same as “ ∑ ”.
B is a subset of A B⊆ A

These notations can be mixed and matched. For example, ∑ | f ( A)|

f :A→ B;
f is injective
means the sum of the sizes of the images of all injective maps from A to B.
Enumerative Combinatorics: class notes page 48

1.2.3. Rules for sums

Finite sums have lots of properties, such as the identity (20) which we have used
in the second proof of Theorem 1.2.1. See [Grinbe15, §1.4] for a long list of these
properties. Here I shall just list a few:44

• Splitting-off: Let S be a finite set. Let as be a number for each s ∈ S. Let

t ∈ S. Then,
∑ as = at + ∑ as . (25)
s∈S s∈S\{t}

In other words, we can always rewrite the sum ∑ as as its addend at plus
s∈S
the sum of all remaining addends. This is called splitting off (or extracting) an
addend from a sum. Particular cases of this identity are the identities
q q q q −1
∑ ai = a p + ∑ ai and ∑ ai = a q + ∑ ai
i= p i = p +1 i= p i= p

that hold whenever p ≤ q. (They do not hold for p > q, because we cannot
extract an addend from an empty sum.)

• Splitting: Let S be a finite set. Let X and Y be two subsets of S such that
X ∩ Y = ∅ and X ∪ Y = S. (Equivalently, X and Y are two subsets of S such
that each element of S lies in exactly one of X and Y.) Let as be a number for
each s ∈ S. Then,
∑ as = ∑ as + ∑ as . (26)
s∈S s∈ X s ∈Y
 
(The right hand side of this equality has to be understood as ∑ as +
  s ∈ X

∑ as . In general, finite sums reach past “·” signs but not past “+” signs;
s ∈Y  
thus, ∑ as b means ∑ ( as b), but ∑ as + b means ∑ as + b. This is consid-
s∈S s∈S s∈S s∈S
ered to be part of the PEMDAS convention, because finite sums are a form of
addition. When in doubt, always use parentheses.)
For example, we can apply (26) to S = {1, 2, 3, 4, 5} and X = {1, 3, 5} and
Y = {2, 4}. We then obtain

∑ as = ∑ as + ∑ as , or, equivalently,
s∈{1,2,3,4,5} s∈{1,3,5} s∈{2,4}
a1 + a2 + a3 + a4 + a5 = ( a1 + a3 + a5 ) + ( a2 + a4 ) .

44 See [Grinbe15, §1.4.2 and §2.14] for proofs of these properties.

Enumerative Combinatorics: class notes page 49

More generally, if S is a finite set of integers, then

∑ as = ∑ as + ∑ as (27)
s∈S s∈S; s∈S;
s is even s is odd

(by (26), applied to X = {s ∈ S | s is even} and Y = {s ∈ S | s is odd}).

More generally, let S be a finite set. For each s ∈ S, let A (s) be a logical
statement (which can be either true or false depending on s; for example,
A (s) could be “s is even” if S is a set of integers, or “s is empty” if S is a set
of sets), and let as be a number. Then,

∑ as = ∑ as + ∑ as . (28)
s∈S s∈S; s∈S;
A(s) is true A(s) is false

This follows by applying (26) to X = {s ∈ S | A (s) is true} and

Y = {s ∈ S | A (s) is false}. Of course, (27) is the particular case of (28) for
A (s) = (“s is even”).
• Summing equal values: Let S be a finite set. Let a be a number. Then,

∑ a = |S| · a. (29)
s∈S

(That is, summing n many copies of a number a results in the number na.)
Applying (29) to a = 1, we find

∑ 1 = |S| · 1 = |S| . (30)

s∈S

In other words, the size of a finite set is a particular case of a finite sum. This
trivial observation will prove rather useful to us below.
• Splitting an addend: Let S be a finite set. For every s ∈ S, let as and bs be
numbers. Then,
∑ ( a s + bs ) = ∑ a s + ∑ bs . (31)
s∈S s∈S s∈S
When S = {1, 2, . . . , n}, this becomes precisely the equality (20).
• Factoring out: Let S be a finite set. For every s ∈ S, let as be a number. Also,
let λ be a number. Then,
∑ λas = λ ∑ as . (32)
s∈S s∈S
This is a generalization of the distributive law λ ( a + b) = λa + λb.
• Zeroes sum to zero: Let S be a finite set. Then,

∑ 0 = 0. (33)
s∈S
Enumerative Combinatorics: class notes page 50

• Renaming the index: Let S be a finite set. Let as be a number for each s ∈ S.
Then,
∑ as = ∑ at .
s∈S t∈S
This is called renaming the summation index, and is justified by the fact that the
summation index is a dummy variable.

• Substituting the index I: Let S and T be two finite sets. Let f : S → T be a

bijection. Let at be a number for each t ∈ T. Then,

∑ at = ∑ a f (s) . (34)
t∈ T s∈S

The idea here is that the sum ∑ a f (s) contains the same addends as the sum
s∈S
∑ at .
t∈ T
When we apply (34), we say that we are substituting f (s) for t in the sum
∑ at .
t∈ T
Applying (34) to at = 1, we obtain ∑ 1 = ∑ 1, which rewrites as | T | = |S|
t∈ T s∈S
(because of (30)). This is precisely the bijection principle (Theorem 1.1.6).
Thus, (34) generalizes the bijection principle.
The equality (34) also shows that we can “turn a sum around”, in the sense
that we have
a 1 + a 2 + · · · + a n = a n + a n −1 + · · · + a 1 (35)
for any n ∈ N and any n numbers a1 , a2 , . . . , an . (This was used in the Second
proof of Theorem 1.2.1.) Indeed, the left hand side of (35) can be rewrit-
ten as ∑ at , whereas the right hand side can be rewritten as ∑ an+1−s =
t∈[n] s∈[n]
∑ a f (s) , where f : [n] → [n] is the map that sends each s ∈ [n] to n + 1 − s.
s∈[n]
Since this latter map f : [n] → [n] is a bijection, we can use (34) (applied to
S = [n] and T = [n]) to conclude that ∑ at = ∑ a f (s) ; but this is precisely
t∈[n] s∈[n]
the identity (35).

• Substituting the index II: Let S and T be two finite sets. Let f : S → T be a
bijection. Let as be a number for each s ∈ S. Then,

∑ a s = ∑ a f −1 ( t ) . (36)
s∈S t∈ T

This is, of course, just (34) but applied to T, S and f −1 instead of S, T and f .
We state it as a separate formula because we shall be using both versions.
Enumerative Combinatorics: class notes page 51

• Splitting a sum by a value of a function: Let S and W be two finite sets. Let
f : S → W be a map. Let as be a number for each s ∈ S. Then,

∑ as = ∑ ∑ as . (37)
s∈S w ∈W s∈S;
f (s)=w

(There are two summation signs on the right hand side, signifying that we
are taking a finite sum of finite sums.) This equality provides a way to split
a sum ∑ as into several batches by distributing all possible values of the
s∈S
summation index s into several “bins”45 , and then summing each batch of
addends corresponding to a bin together and then summing the “bin totals”.
For example, let us apply (37) to the 11-element set S = {−5, −4, . . . , 4, 5}
and the 6-element set W = {0, 1, . . . , 5} and the map f : S → W that sends
each s ∈ S to |s| ∈ W. We thus obtain

∑ as = ∑ ∑ as .
s∈{−5,−4,...,4,5} w∈{0,1,...,5} s∈{−5,−4,...,4,5};
|s|=w

Explicitly, this rewrites as

a −5 + a −4 + · · · + a 4 + a 5
= a 0 + ( a −1 + a 1 ) + ( a −2 + a 2 ) + ( a −3 + a 3 ) + ( a −4 + a 4 ) + ( a −5 + a 5 ) .
Here, the “bins” are {0}, {−1, 1}, {−2, 2}, {−3, 3}, {−4, 4} and {−5, 5}, and
the corresponding “bin totals” are a0 , a−1 + a1 , a−2 + a2 , a−3 + a3 , a−4 + a4
and a−5 + a5 .
• Splitting a sum into subsums: Let a finite set S be the union of k disjoint sets
S1 , S2 , . . . , Sk . Let as be a number for each s ∈ S. Then,
k
∑ as = ∑ ∑ as . (38)
s∈S w =1 s ∈ Sw

(The right hand side can be rewritten in the somewhat more familiar form
∑ as + ∑ as + · · · + ∑ as .)
s ∈ S1 s ∈ S2 s ∈ Sk

It is easy to see that (38) follows from (37) (by setting W = {1, 2, . . . , k},
and letting f : S → W be the map that sends each s ∈ S to the unique
w ∈ {1, 2, . . . , k } for which s ∈ Sw ).
Note that if we set as = 1 for each s ∈ S and recall the formula (30), then (38)
k
rewrites as |S| = ∑ |Sw | = |S1 | + |S2 | + · · · + |Sk |. Thus, we have recovered
w =1
the sum rule (Theorem 1.1.3) as a particular case of (38).
45 The “bins” are the elements of W here. Each value s ∈ S goes into bin f (s).
Enumerative Combinatorics: class notes page 52

There are further rules, such as Fubini’s principle for the interchange of summation
signs. We will not state them yet in order to keep this section reasonably short.
However, let us state an important application of summation signs in counting:

Theorem 1.2.5 (The sum rule, in summation-sign form). Let S and W be two
finite sets. Let f : S → W be a map. Then,

|S| = ∑ (# of s ∈ S satisfying f (s) = w) .

w ∈W

Theorem 1.2.5 is called the sum rule, since it is a more flexible version of Theorem
1.1.3. Indeed, both theorems tell us how to compute the size |S| of a set S that has
been split into several subsets46 ; but Theorem 1.1.3 requires the latter subsets to be
numbered by 1, 2, . . . , k, whereas Theorem 1.2.5 only needs them to be indexed by
elements w of W.

Example 1.2.6. Assume you have a finite set of socks, and each sock is either red
or blue or green. Then,

(# of socks) = (# of red socks) + (# of blue socks) + (# of green socks)

= ∑ (# of socks of color w) .
w∈{red,blue,green}

This is a consequence of Theorem 1.2.5, applied to S = {socks}, W =

{red, blue, green} and the map f : S → W defined by f (s) = (color of s).

Proof of Theorem 1.2.5. From (30), we obtain

|S| = ∑1= ∑ ∑ 1 (39)

s∈S w ∈W s∈S;
f (s)=w

(by (37), applied to as = 1).

46 In Theorem 1.1.3, the subsets are S1 , S2 , . . . , Sk ; in Theorem 1.2.5, the subsets are
{s ∈ S | f (s) = w} for various w ∈ W.
Enumerative Combinatorics: class notes page 53

Now, for each w ∈ W, we have

∑ 1= ∑ 1
s∈S; s∈{t∈S | f (t)=w}
f (s)=w
 

∑ ∑
 
since the sign “ ” is shorthand for “ ”
 
s∈S; s∈{t∈S | f (t)=w}
f (s)=w
= |{t ∈ S | f (t) = w}|
(by (30), applied to {t ∈ S | f (t) = w} instead of S)
= |{s ∈ S | f (s) = w}| (here, we have renamed the index t as s)
= (# of s ∈ S satisfying f (s) = w) . (40)

Hence, (39) becomes

|S| = ∑ ∑ 1 = ∑ (# of s ∈ S satisfying f (s) = w) .

w ∈W s∈S; w ∈W
f (s)=w
| {z }
=(# of s∈S satisfying f (s)=w)
(by (40))

This proves Theorem 1.2.5.

1.2.4. While at that, what is a finite product?

Similar to the notion of a finite sum is a notion of a finite product:

Definition 1.2.7. Let S be a finite set. For each s ∈ S, let as be a number (e.g., an
integer or a rational number or a real number or a complex number).
Then, ∏ as shall denote the product of the numbers as for all s ∈ S. This
s∈S
notation is read as “the product of as over all s ∈ S” or “the product of as for s
ranging over S” or “the product of as where s runs through S”. The “∏” sign in
this notation is called the product sign.

This notation is called a finite product. For example, ∏ s 3 = 43 · 53 · 73 · 93 .

s∈{4,5,7,9}
Most of what we said about finite sums applies (with the obvious changes) to
finite products. In particular, finite products are well-defined (in the sense that all
ways to multiply n given numbers together produce the same result) and satisfy
analogues of the properties of finite sums listed above (with the obvious changes
made: e.g., the equality (29) turns into ∏ a = a|S| ), except for (30). Notations like
s∈S
q
∑ as and ∑ as have their analogues for products, which look exactly the same
s∈S; i= p
A(s)
Enumerative Combinatorics: class notes page 54

q
(that is, ∏ as and ∏ as ) and are defined in the same way. We only need to be
s∈S; i= p
A(s)
careful in defining empty products correctly: While an empty sum was defined to
be 0, we must define an empty product (i.e., a product with no factors47 ) to be 1.
This is why, in particular, we have x0 = 1 for each x ∈ R. (Indeed, x m is defined as
· · · x} whenever m ∈ N and x ∈ R. Thus, x0 = |xx {z
|xx {z · · · x} = (empty product) = 1.)
m times 0 times

1.2.5. The sums 1k + 2k + ... + nk

Let us now go back to Theorem 1.2.1. That theorem gave an explicit formula for
the sum of the first n positive integers. A similar formula exists for the sum of the
squares of the first n positive integers:

Proposition 1.2.8. Let n ∈ N. Then,

n (n + 1) (2n + 1)
12 + 22 + · · · + n2 = .
6

Proof of Proposition 1.2.8 (sketched). Again, this can be shown by a straightforward

induction (LTTR).
Is there a picture proof for Proposition 1.2.8 as well, similar to the one we gave for
Theorem 1.2.1? Yes, but the picture will be 3-dimensional and thus much harder to reason
about.
Is there a proof of Proposition 1.2.8 that is analogous to our second proof of Theorem
1.2.1 above? Not to my knowledge. The numbers 12 + n2 , 22 + (n − 1)2 , . . . , n2 + 12 are
not equal, and I don’t know of a way to reorder the addends in the (expanded) sum
6 · 12 + 22 + · · · + n2 to obtain n (n + 1) (2n + 1).

Next, let us sum the cubes of the first n positive integers:

Proposition 1.2.9. Let n ∈ N. Then,

n2 ( n + 1)2
13 + 23 + · · · + n3 = .
4

Proof of Proposition 1.2.9 (sketched). Again, a straightforward induction (LTTR).

A picture proof appears even more hopeless for Proposition 1.2.9 than it was for Propo-
sition 1.2.8, as it would now require a 4-dimensional picture. But this is, to some ex-
tent, an illusion. When we formalized our picture proof of Theorem 1.2.1, we obtained

47 The analogue of “addend” in a finite product is “factor”.

Enumerative Combinatorics: class notes page 55

a combinatorial argument that did not rely on any pictures. Such arguments can be
made in any number of dimensions. And indeed, such proofs can be found for Propo-
sition 1.2.8 and for Proposition 1.2.9. For such proofs, see the math.stackexchange thread
https://math.stackexchange.com/questions/95047/ .

Now, a question suggests itself: Are Theorem 1.2.1, Proposition 1.2.8 and Propo-
sition 1.2.9 just the first three pieces in a sequence of formulas? In other words,
given a positive integer k, is there an explicit formula for 1k + 2k + · · · + nk ? Hav-
ing seen the answers for k = 1, k = 2 and k = 3, we would expect the right hand
side of such a formula to be a (k + 1)-st degree polynomial in n.
Let us first reword the question. Recall (from Subsection 1.2.2) that we are using
the shorthand notation
q
∑ ai for ∑ a i = a p + a p +1 + · · · + a q
i= p i ∈{ p,p+1,...,q}

whenever p and q are two integers and a p , a p+1 , . . . , aq are arbitrary numbers. Thus,
the sum that we are interested in, namely 1k + 2k + · · · + nk , can be rewritten as
n
∑ ik .
i =1
Let us list the answers to our question for small values of k:
n
n ( n + 1)
∑ i1 = 2
(by Theorem 1.2.1) ;
i =1
n
n (n + 1) (2n + 1)
∑ i2 = 6
(by Proposition 1.2.8) ;
i =1
n
n2 ( n + 1)2
∑i = 3
4
(by Proposition 1.2.9) ;
i =1
n 2−1


n ( 2n + 1 ) ( n + 1 ) 3n + 3n
∑ i4 = 30
;
i =1
n n2 (n + 1)2 2n + 2n2 − 1


∑i = 5
12
;
i =1
....
(We can prove each of these formulas by induction on n once we know how it looks
like.) For now, we don’t see much of a pattern. However, there is one – once we
rewrite the right hand sides in terms of binomial coefficients. We will give a more
general definition of binomial coefficients later, but for now let us just set
m ( m − 1) ( m − 2) · · · ( m − k + 1)
 
m
= (41)
k k ( k − 1) ( k − 2) · · · 1
for all m ∈ R and k ∈ N. (Here, on the right hand side, the numerator is a
product of k factors, the first of which is m and which decrease by 1 from each
Enumerative Combinatorics: class notes page 56

factor to the next; the denominator is a similar product, but starting at k instead
of m.) Furthermore, if k and m are two nonnegative integers, then let us use the
notation sur (k, i ) for the # of all surjective maps from [k ] to [i ]. (Recall that [k ]
stands for {1, 2, . . . , k }; in particular, [0] = ∅.) Now, we can answer our question
by the following formula:

Theorem 1.2.10. Let n ∈ N, and let k be a positive integer. Then,

n k 
n+1
∑ i = ∑ sur (k, i) · i + 1 .
k
i =1 i =0

We will prove this theorem in Section 2.5. For now, let us say a few things about
it.
First of all, why does Theorem 1.2.10 answer our question? Why does its right
hand side count as an explicit answer? Didn’t we just rewrite a finite sum as
another finite sum (a more complicated one to boot)? In a sense, yes; but if k is
fixed, then the sum on the right hand side has a fixed number of addends (namely,
k + 1), so it can indeed count as explicit, whereas the left hand side has n addends.
To illustrate this point, let us see how Proposition 1.2.8 can be straightforwardly
recovered from Theorem 1.2.10.
We will need to know the numbers sur (2, i ) for all i ∈ {0, 1, 2}, so let us agree
on a notation for maps between finite sets:

Definition 1.2.11. Let X be a finite set, and let Y be any set. Assume that X =
{ x1 , x2 , . . . , xk } for some distinct elements x1 , x2 , . . . , xk . Let y1 , y2 , . . . , yk be any
k elements of Y (not necessarily distinct). Then, the mapfrom X to Y that sends 
x1 x2 · · · x k
x1 , x2 , . . . , xk to y1 , y2 , . . . , yk , respectively, will be called .
y1 y2 · · · y k
This is called the two-line notation for maps.

3} and Y = Z,
For example, if X = {1, 2,  the map that sends each m ∈ X
 then
1 2 3 1 2 3
to m2 ∈ Z can be written as 2 2 2 = in two-line notation. It
1 2 3 1 4 9

2 1 3
can also be written as and in four other ways, since we have freedom
4 1 9
to decide in which order we list the elements of X.
The two-line notation makes it easy to see whether a map is injective or surjective:

Remark 1.2.12. Let f : X → Y be a map between two sets X and Y, where X is

finite. Assume that X = { x1 , x2 , . . . , xk } for some
 distinct elements x1 , x2 , . . . , xk .
x1 x2 · · · x k
Assume that f is written as in two-line notation. Then:
y1 y2 · · · y k
(a) The map f is injective if and only if y1 , y2 , . . . , yk are distinct.
(b) The map f is surjective if and only if Y = {y1 , y2 , . . . , yk }.
Enumerative Combinatorics: class notes page 57

With this in hand, we can easily compute sur (2, 0), sur (2, 1) and sur (2, 2):

• There exist no maps [2] → [0] (since [0] = ∅ is the empty set, so there is
nowhere to map 1 ∈ [2] to). Thus, a fortiori, there exist no surjective maps
[2] → [0]. Hence, sur (2, 0) = 0. (Similarly, sur (k, 0) = 0 for all k > 0.)
• There is only one map [2] → [1], namely  themap that sends both 1 and 2 to 1.
1 2
(In two-line notation, it is written as .) This map is surjective. Hence,
1 1
there is exactly 1 surjective map [2] → [1]. Hence, sur (2, 1) = 1. (Similarly,
sur (k, 1) = 1 for all k > 0.)

• There
 are
 four
 maps  [2] → [ 2]. In 
two-linenotation, they are written as
1 2 1 2 1 2 1 2
, , and . The second and the third
1 1 1 2 2 1 2 2
among them are surjective (since {1, 2} and {2, 1} both equal [2]), whereas
the first and the fourth are not. Hence, there are exactly 2 surjective maps
[2] → [2]. Hence, sur (2, 2) = 2.

Now, we can apply Theorem 1.2.10 to k = 2, and conclude that

n 2  
n+1
∑ i = ∑ sur (2, i) · i + 1
2
i =1 i =0
   
n+1 n+1
= sur (2, 0) · + sur (2, 1) ·
| {z } 0+1 | {z } 1+1
=0  | {z
 } =1  | {z }
n+1 n+1 n + 1 ( n + 1) n
= = = =
1 1 2 2·1
(by (41), applied (by (41), applied
to m=n+1 and k=1) to m=n+1 and k=2)
 
n+1
+ sur (2, 2) ·
| {z } 2+1
=2 | {z }
n + 1 ( n + 1) n ( n − 1)

= =
3 3·2·1
(by (41), applied
to m=n+1 and k =3)
n+1 ( n + 1) n ( n + 1) n ( n − 1)
= 0· +1· +2·
1 2·1 3·2·1
1 3 1 2 1
= n + n + n (after straightforward algebraic manipulations)
3 2 6
n (n + 1) (2n + 1)
=
6
for each n ∈ N. Thus, Proposition 1.2.8 is proven again. Similarly, we can recover
Theorem 1.2.1 and Proposition 1.2.9 and similar formulas for arbitrary values of k.
Enumerative Combinatorics: class notes page 58

Theorem 1.2.10 also confirms our suspicion that 1k + 2k + · · · + nk is given by

 (k + 1) polynomial in n. Indeed, it is easy to see that each of the terms
a degree-
n+1
on the right hand side of Theorem 1.2.10 is a degree-(i + 1) polynomial in
i+1
n (this follows easily from (41)).
Theorem 1.2.10 is likely an old result (18th Century?). Regrettably, I don’t know
its ultimate origins, since it has been overshadowed by an even more explicit for-
mula for 1k + 2k + · · · + nk , the Faulhaber formula [Knuth93].

1.3. Factorials and binomial coefficients

1.3.1. Factorials
We now switch from studying sums to studying products.
Definition 1.3.1. For any n ∈ N, we define a positive integer n! by

n! = 1 · 2 · · · · · n.
n
(Using the product sign as defined in Definition 1.2.7, this rewrites as n! = ∏ i.)
i =1
We shall refer to n! as “n factorial”.
We have agreed that empty products are defined to be 1. Thus, the above defini-
tion of 0! yields
0! = 1 · 2 · · · · · 0 = (empty product) = 1.
Definition 1.3.1 yields the following values for the first few factorials:
0! = 1,
1! = 1,
2! = 1 · 2 = 2,
3! = 1 · 2 · 3 = 6,
4! = 1 · 2 · 3 · 4 = 24,
5! = 1 · 2 · 3 · 4 · 5 = 120,
6! = 1 · 2 · 3 · 4 · 5 · 6 = 720,
7! = 1 · 2 · 3 · 4 · 5 · 6 · 7 = 5 040,
8! = 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 = 40 320.
Factorials can be computed recursively:
Proposition 1.3.2. If n is a positive integer, then n! = (n − 1)! · n.
Proof of Proposition 1.3.2. Let n be a positive integer. Then, n − 1 ∈ N, so that
Definition 1.3.1 yields (n − 1)! = 1 · 2 · · · · · (n − 1). But Definition 1.3.1 also yields
n! = 1 · 2 · · · · · n = (1 · 2 · · · · · (n − 1)) ·n = (n − 1)! · n.
| {z }
=(n−1)!
Enumerative Combinatorics: class notes page 59

This proves Proposition 1.3.2.

Exercise 1.3.1. Let n ∈ N. Prove that

(2n)!
(2n − 1) · (2n − 3) · · · · · 1 = .
2n n!
(The left hand side of this equality is understood to be the product of all odd
integers from 1 to 2n − 1.)

1.3.2. Definition of binomial coefficients

From now on, the word “number” (as in “Let n be any number”, without any
further specification) will mean an integer or a rational number or a real number
or a complex number, depending on the generality in which you want to work.
Let us now define binomial coefficients.
 
n
Definition 1.3.3. Let n and k be any two numbers. We define a number as
k
follows:

• If k ∈ N, then we set

n ( n − 1) ( n − 2) · · · ( n − k + 1)
 
n
= . (42)
k k!

/ N, then we set
• If k ∈  
n
= 0. (43)
k
 
n
We call a binomial coefficient, and we refer to it as “n choose k”.
k

This definition is standard across significant parts of the literature; in particu-

lar, it is followed in the book [GrKnPa94] (whose entire Chapter 5 is devoted to
binomial coefficients), in the book [Comtet74] (one of the classics on enumerative
combinatorics and binomial identities), and in [Grinbe15]. Some other authors use
other definitions. All definitions I am aware of are equivalent in the “core region”
of the binomial coefficients – that is, in the case when n ∈ N and k ∈ {0, 1, . . . , n}.
However, some definitions  yield values differing from ours when n < 0. Many
n
authors prefer to define only for n ∈ N, or only for k ∈ N, or even only in
k
  n ∈ N and k ∈ {0, 1, . . . , n}). Loehr, in [Loehr11],
the most restrictive case (when
n
seems to avoid defining for negative n at all, despite this case being rather
k
useful (as we will see soon).
Enumerative Combinatorics: class notes page 60

 
n
Some authors use notations like Ckn nC
ork or n Ck for .
k
We shalllater
 see why the words “binomial coefficient” and “choose” are appro-
n
priate for .
k
Definition 1.3.3 does not contradict our earlier definition (41) of binomial co-
efficients (but merely extends it to a more general setting). Indeed, the product
k (k − 1) (k − 2) · · · 1 in the denominator of (41) is precisely the k! in the denomina-
tor of (42) (since k! = 1  · 2
· · · · · k = k (k − 1) (k − 2) · · · 1).
n
Caution: The notation for a binomial coefficient risks getting confused for the two-
k
line notation for a map (if the domain of the map is a 1-element set) or for a column vector
with 2 entries. In practice, this kind of confusion rarely happens, but its possibility should
be kept in mind.

Example 1.3.4. Let us see some consequences of Definition 1.3.3.

(a) For any number n, we have
n ( n − 1) ( n − 2) · · · ( n − 0 + 1)
 
n
= (by (42), applied to k = 0)
0 0!
 
1 since n (n − 1) (n − 2) · · · (n − 0 + 1) = (empty product) = 1
=
1 and 0! = 1
= 1. (44)
(b) For any number n, we have
n ( n − 1) ( n − 2) · · · ( n − 1 + 1)
 
n
= (by (42), applied to k = 1)
1 1!
n
= (since n (n − 1) (n − 2) · · · (n − 1 + 1) = n and 1! = 1)
1
= n. (45)
(c) For any number n, we have
n ( n − 1) ( n − 2) · · · ( n − 2 + 1)
 
n
= (by (42), applied to k = 2)
2 2!
n ( n − 1)
= . (46)
2
(d) For any number n, we have
n ( n − 1) ( n − 2) n ( n − 1) ( n − 2)
 
n
= = (likewise) .
3 3! 6
(e) The equality (42) (applied to n = −1 and k = 5) yields
−1 (−1) (−1 − 1) (−1 − 2) · · · (−1 − 5 + 1)
 
(−1) (−2) (−3) (−4) (−5)
= =
5 5! 5!
(−1) (−2) (−3) (−4) (−5)
= = −1.
1·2·3·4·5
Enumerative Combinatorics: class notes page 61

(f) More generally, for any k ∈ N, we have

−1 (−1) (−1 − 1) (−1 − 2) · · · (−1 − k + 1)

 
=
k k!
(by (42), applied to n = −1)
(−1) (−2) · · · (−k) (−1) (−2) · · · (−k)
= =
k! 1·2·····k
k
= (−1) (47)
(
1, if k is even;
= .
−1, if k is odd
√
(g) The equality (42) (applied to n = 2 and k = 2) yields

√  √ √  √  √  √ √ 
2 2 2−1 2−2 ··· 2−2+1 2 2−1
= = .
2 2! 2
√
(h) The equality (43) (applied to n = 2 and k = 2) yields
  √
√2 = 0, / N.
since 2 ∈
2

Remark 1.3.5. In [19f-hw0s, Exercise 2], we have introduced the notation

nk (called a “falling factorial”) for the product n (n − 1) (n − 2) · · · (n − k + 1)
whenever n is a number and k is a nonnegative integer. Using this notation, we
can rewrite the equality (42) as

nk
 
n
= for all numbers n and all k ∈ N. (48)
k k!

Class of 2019-10-02

1.3.3. Fundamental properties of the binomial coefficients

The properties of binomial coefficients are one of the major topics in enumerative
combinatorics. We shall see many  of
 them in this text. Before we start, let us
n
tabulate the binomial coefficients for all n ∈ {−3, −2, −1, . . . , 6} and some of
k
the k ∈ {0, 1, 2, 3, 4, 5}. In the following table, each row corresponds to a value of
Enumerative Combinatorics: class notes page 62

n, while each southwest-northeast diagonal corresponds to a value of k:

k =0 k =1 k =2 k =3
. . . .
n = −3 → 1 −3 6 −10
n = −2 → 1 −2 3 −4
n = −1 → 1 −1 1 −1 1
n =0 → 1 0 0 0 0
n =1 → 1 1 0 0 0 0
n =2 → 1 2 1 0 0 0
n =3 → 1 3 3 1 0 0 0
n =4 → 1 4 6 4 1 0 0
n =5 → 1 5 10 10 5 1 0 0
n =6 → 1 6 15 20 15 1 6 0
 
n
Just by staring at this table, you will discover many properties of . For
k
example, all the zeroes in the right half of it suggest the following fact:
 
n
Proposition 1.3.6. Let n ∈ N and k ∈ R be such that k > n. Then, = 0.
k

/ N, then this follows immediately from (43). Thus,

Proof of Proposition 1.3.6. If k ∈
for the rest of this proof, we WLOG assume that k ∈ N. Hence, (42) yields

n ( n − 1) ( n − 2) · · · ( n − k + 1)
 
n
= . (49)
k k!

But n ∈ {0, 1, . . . , k − 1} (since n ∈ N and k ∈ N and k > n). Hence, one of the
factors of the product n (n − 1) (n − 2) · · · (n − k + 1) is n − n = 0. Therefore, the
product n (n − 1) (n − 2) · · · (n − k + 1) must be 0 (because if one ofthefactors of
n 0
a product is 0, then the whole product is 0). Thus, (49) rewrites as = . In
  k k!
n
other words, = 0. This proves Proposition 1.3.6.
k
This proof might remind you of the math joke “simplify ( x − a) ( x − b) · · · ( x − z)”. 48

 no longer holds if we drop the requirement n ∈ N.

Note that Proposition 1.3.6
1/2
For example, 1 > 1/2, yet = 1/2 6= 0. For another example, 1 > −1, yet
1
−1
 
= −1 6= 0.
1
48 Answer: 0, because the third factor from the right is x − x = 0.
Enumerative Combinatorics: class notes page 63

Another pattern that you might guess from the table above is that the binomial
coefficients in the “negative rows” (i.e., corresponding to n = −1, n = −2 and
further above) repeat the ones you find in the nonnegative rows up to sign. More
precisely:

Proposition 1.3.7 (Upper negation formula). Let n ∈ R and k ∈ Z. Then,

−n k n+k−1
   
= (−1) . (50)
k k

Proof of Proposition 1.3.7. We must be in one of the following two cases:

Case 1: We have k ∈ / N.
Case 2: We have k ∈ N.
Let us first
 consider Case / N. Thus, both binomial
1. In this case, we have k ∈
−n n+k−1
 
coefficients and equal 0 (by applying (43)). Hence, the equality
k k
(50) rewrites as 0 = (−1)k · 0, which is clearly true. Thus, we have proven (50) in
Case 1. In other words, Proposition 1.3.7 is proven in Case 1.
Let us now consider Case 2. In this case, we have k ∈ N. Thus, (42) (applied to
−n instead of n) yields
−n (−n) (−n − 1) (−n − 2) · · · (−n − k + 1)
 
= . (51)
k k!
Now, let us rewrite the numerator on the right hand side:

(−n) (−n − 1) (−n − 2) · · · (−n − k + 1)

= (−1)k · (n (n + 1) (n + 2) · · · (n + k − 1))
| {z }
=(n+k−1)(n+k−2)(n+k−3)···n
(here, we have turned the product around)

(here, we have factored out a − 1 from each of the k factors)

= (−1)k · ((n + k − 1) (n + k − 2) (n + k − 3) · · · n) .

Thus, (51) rewrites as

−n (−1)k · ((n + k − 1) (n + k − 2) (n + k − 3) · · · n)
 
= . (52)
k k!
On the other hand, (42) (applied to n + k − 1 instead of n) yields

n+k−1 ( n + k − 1) ( n + k − 2) ( n + k − 3) · · · ( n + k − 1 − k + 1)
 
=
k k!
( n + k − 1) ( n + k − 2) ( n + k − 3) · · · n
= .
k!
Enumerative Combinatorics: class notes page 64

Multiplying this equality by (−1)k , we find

k n+k−1 ( n + k − 1) ( n + k − 2) ( n + k − 3) · · · n
 
(−1) = (−1)k ·
k k!
(−1)k · ((n + k − 1) (n + k − 2) (n + k − 3) · · · n)
= .
k!
−n k n+k−1
   
Comparing this with (52), we obtain = (−1) . Thus, Proposi-
k k
tion 1.3.7 is proven in Case 2.
Proposition 1.3.7 and Proposition 1.3.6 show that muchofour above table of
n
binomial coefficients is redundant. In order to know all with n ∈ Z and
  k
n
k ∈ N, it suffices to tabulate the values for n ∈ N and k ∈ {0, 1, . . . , n}. Let us
k
do this here for n ∈ {0, 1, . . . , 8}:

k =0
.
k =1
n=0 → 1 .
k =2
n=1 → 1 1 .
k =3
n=2 → 1 2 1 .
k =4
n=3 → 1 3 3 1 .
k =5
n=4 → 1 4 6 4 1 .
k =6
n=5 → 1 5 10 10 5 1 .
k =7
n=6 → 1 6 15 20 15 6 1 .
n=7 → 1 7 21 35 35 21 7 1

n=8 → 1 8 28 56 70 56 28 8 1

This table is called Pascal’s triangle due to its triangular form.

One of the most fundamental properties of Pascal’s triangle is that each number
in it is the sum of the two numbers above it (i.e., the number above-left from it,
and the number above-right from it). For example, 56 = 21 + 35. This holds even
for the 1’s on the sides
 of the triangle, if you extend the triangle to include the
n
binomial coefficients with n < 0 or k < 0 or k < n. More generally, this holds
k  
n
for all binomial coefficients – not just for the ones that fit in Pascal’s triangle:
k
Enumerative Combinatorics: class notes page 65

Theorem 1.3.8 (Recurrence of the binomial coefficients). Let n ∈ R and k ∈ R.

Then,
n−1 n−1
     
n
= + .
k k−1 k

Proof of Theorem 1.3.8. We are in one of the following three cases:

Case 1: We have k ∈ {1, 2, 3, . . .}.
Case 2: We have k = 0.
Case 3: We have k ∈ / N.
(There are no other cases, because each k ∈ N fits either into Case 1 or into Case
2, whereas each k ∈ / N fits into Case 3.)
Let us first consider Case 3. In this case,  we  have k ∈/ N. Hence, k − 1 ∈ / N.
n−1
Thus, an application of (43) shows that = 0. Likewise, from k ∈ / N,
k−1
n−1
     
n n
we obtain = 0 and = 0. But we must prove the equality =
 k k k
n−1 n−1 n−1
   
+ . In view of the three equalities we just showed ( =0
k −1  k  k−1
n−1

n
and = 0 and = 0), this rewrites as 0 = 0 + 0, which of course is true.
k k
Thus, Theorem 1.3.8 is proven in Case 3.    
n n
Let us next consider Case 2. In this case, we have k = 0. Thus, = =1
k  0
n−1 n−1
 
(by (44)). Likewise, k −1 = −1 ∈
= 1. Also, |{z} / N, and thus =0
k k−1
=0
n−1
   
n
(by an application of (43)). But we must prove the equality = +
k k−1
n−1 n−1
  
n
. In view of the three equalities we just showed ( = 0 and =1
k k−1 k
n−1

and = 1), this rewrites as 1 = 0 + 1, which of course is true. Thus,
k
Theorem 1.3.8 is proven in Case 2.
Let us finally consider Case 1. In this case, we have k ∈ {1, 2, 3, . . .}. Hence,
Proposition 1.3.2 yields k! = (k − 1)! · k. Furthermore, both k and k − 1 belong to
n−1
   
n
N (since k ∈ {1, 2, 3, . . .}). Hence, all three binomial coefficients and
k k
n−1
 
and are defined using the formula (42) (applied to the appropriate num-
k−1
Enumerative Combinatorics: class notes page 66

bers). Explicitly:

n ( n − 1) ( n − 2) · · · ( n − k + 1)
 
n
=
k k!
n ( n − 1) ( n − 2) · · · ( n − k + 1)
= (53)
( k − 1) ! · k
(since k! = (k − 1)! · k)

and
n−1 (n − 1) (n − 2) (n − 3) · · · ((n − 1) − k + 1)
 
=
k k!
( n − 1) ( n − 2) ( n − 3) · · · ( n − k )
=
k!
((n − 1) (n − 2) (n − 3) · · · (n − k + 1)) · (n − k)
=
k!
((n − 1) (n − 2) (n − 3) · · · (n − k + 1)) · (n − k)
=
( k − 1) ! · k
(since k! = (k − 1)! · k)
( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1) n − k
= · (54)
( k − 1) ! k

and
n−1 (n − 1) (n − 2) (n − 3) · · · ((n − 1) − (k − 1) + 1)
 
=
k−1 ( k − 1) !
( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1)
= . (55)
( k − 1) !
Enumerative Combinatorics: class notes page 67

Adding the equalities (55) and (54) together, we obtain

n−1 n−1
   
+
k−1 k
( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1) ( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1) n − k
= + ·
( k − 1) ! ( k − 1) ! k
( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1) n−k
 
= 1+
( k − 1) ! k
| {z }
n
=
k
( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1) n
= ·
( k − 1) ! k
n · ((n − 1) (n − 2) (n − 3) · · · (n − k + 1))
=
( k − 1) ! · k
n ( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1)
=
( k − 1) ! · k
n ( n − 1) ( n − 2) ( n − 3) · · · ( n − k + 1)
= (since (k − 1)! · k = k!) ;
  k!
n
= (by (53)) .
k

Thus, Theorem 1.3.8 is proven in Case 1 as well.

We have thus proven Theorem 1.3.8 in all three Cases 1, 2 and 3.
The Wikipedia page for Pascal’s triangle includes many pictures of Pascal’s tri-
angle as well as numerous remarkable properties.

Theorem 1.3.9 (Factorial formula for the binomial coefficients). Let n ∈ N and
k ∈ N be such that k ≤ n. Then,
 
n n!
= .
k k! · (n − k )!

Proof of Theorem 1.3.9. Multiplying the equality (42) with k! · (n − k )!, we obtain

n ( n − 1) ( n − 2) · · · ( n − k + 1)
 
n
k! · (n − k)! · = k! · (n − k)! ·
k k!
= (n − k)! · (n (n − 1) (n − 2) · · · (n − k + 1))
| {z } | {z }
=1·2·····(n−k) =(n−k+1)·(n−k+2)·····n
(by the definition of (n−k)!) (here, we have turned the product around)

= (1 · 2 · · · · · (n − k)) · ((n − k + 1) · (n − k + 2) · · · · · n)
= 1 · 2 · · · · · n = n!
Enumerative Combinatorics: class notes page 68

   
n n
(since n! is defined to be 1 · 2 · · · · · n). Solving this for , we find =
k k
n!
. This proves Theorem 1.3.9.
k! · (n − k)!
Warning: Whenit  applies, Theorem 1.3.9 provides a simple and highly conve-
n
nient formula for . However, Theorem 1.3.9 only applies when n ∈ N and
k
−3
 
k ∈ N and k ≤ n. Thus, you cannot use Theorem 1.3.9 to compute or
    5
1/3 π
or √ . (There is a generalization of the factorial called the Gamma
4 2
function, which does not really alleviate this problem: it would make the right
hand side of Theorem 1.3.9 well-defined most of the time, but the equality would
still not be true in general.)  
n
Some authors use Theorem 1.3.9 as a definition of . But, as we just saw,
k
such a definition would only work in the case when n  ∈N  and k ∈ N and k ≤ n,
n
so it lacks the general applicability of our definition of .
k
Lemma
  1.3.10. Let n ∈ N and k ∈ R be such that k ∈ / {0, 1, . . . , n}. Then,
n
= 0.
k
 
n
Proof of Lemma 1.3.10. If k ∈ / N, then the claim (that = 0) follows immediately
k
from (43). Thus, for the rest of this proof, we WLOG assume that we do have
k ∈ N. Combining this with k ∈ / {0, 1, . . . , n}, we obtain k ∈ N \ {0, 1, . . . , n} =
{n +
 1, n + 2, n + 3, . . . } . Hence, k ≥ n + 1 > n. Thus, Proposition 1.3.6 yields
n
= 0. This proves Lemma 1.3.10.
k
Theorem 1.3.11 (Symmetry of the binomial coefficients). Let n ∈ N and k ∈ R.
Then,    
n n
= .
k n−k

Proof of Theorem 1.3.11. We are in one of the following two cases:

Case 1: We have k ∈ {0, 1, . . . , n}.
Case 2: We have k ∈ / {0, 1, . . . , n}.
Let us first consider Case 1. In this case, we have k ∈ {0, 1, . . . , n}, so that
n − k ∈ {0, 1, . . . , n} as well. Now, Theorem 1.3.9 yields
 
n n!
= .
k k! · (n − k )!
Enumerative Combinatorics: class notes page 69

Meanwhile, Theorem 1.3.9 (applied to n − k instead of k) yields

 
n n!
= (since n − k ∈ {0, 1, . . . , n})
n−k (n − k)! · (n − (n − k))!
n! n!
= =
(n − (n − k))! · (n − k)! k! · (n − k )!
   
n n
(since n − (n − k) = k). Comparing these two equalities, we obtain = .
k n−k
Thus, Theorem 1.3.11 is proven in Case 1.
Let us now consider Case 2. In this case, we have k ∈ / {0, 1, . . . , n}. Hence, we
have n − k ∈ 49 . Thus, Lemma 1.3.10 (applied to n − k instead
/
 { 0, 1,
 . . . , n } as well  
n n
of k) yields = 0. But Lemma 1.3.10 also yields = 0. Comparing these
n−k     k
n n
two equalities, we obtain = . Thus, Theorem 1.3.11 is proven in Case
k n−k
2.
We have now proven Theorem 1.3.11 in both possible cases.
Theorem 1.3.11 is known as the symmetry of binomial coefficients or the symmetry
of Pascal’s triangle.
−1
 
Warning: Theorem 1.3.11 does not hold for negative n. For example, =
1
−1 −1
   
−1 does not equal = = 0.
(−1) − 1 −2
 
n
Exercise 1.3.2. Let n ∈ N. Prove that = 1.
n

Class of 2019-10-04

1.3.4. Binomial coefficients count subsets

The following theorem is one of the central properties of binomial coefficients:

Theorem 1.3.12 (Combinatorial interpretation of the binomial coefficients). Let

n ∈ N and k ∈ R. Let S be an n-element set. Then,
 
n
= (# of k-element subsets of S) .
k

49 because
otherwise, we would have n − k ∈ {0, 1, . . . , n} and thus n − (n − k) ∈ {0, 1, . . . , n}, which
would contradict n − (n − k) = k ∈
/ {0, 1, . . . , n}
Enumerative Combinatorics: class notes page 70

Example 1.3.13. (a) Let n = 4 and k = 2 and S = {1, 2, 3, 4}. Then, the 2-element
subsets of S are
 {1, 2}, {1, 3}, {1, 4}, {2, 3}, {2, 4} and {3, 4}. The # of these
4
subsets is 6 = , which is exactly what Theorem 1.3.12 predicts.
2
(b) Now, let k = 5 instead (while n is still 4, and S is still {1, 2, 3, 4}). Then,
there are no 5-element subsets of S,  since
 S only has 4 elements. Thus, the #
4
of these 5-element subsets is 0 = , which is exactly what Theorem 1.3.12
5
predicts.
 
n
Theorem 1.3.12 is the reason why is called “n choose k”: It is the number of
k
ways to choose k distinct elements (without an order) from n given elements. The
k-element subsets of S are called the k-combinations  of S.
n
Warning: Theorem 1.3.12 says nothing about when n ∈/ N.
k
Before we prove Theorem 1.3.12, let us state a simple lemma:
 
0
Lemma 1.3.14. Let k ∈ R. Then, = [ k = 0].
k

Here, we are using the so-called Iverson bracket notation:

Definition 1.3.15. If A is any logical statement, then we define an integer [A] ∈

{0, 1} by (
1, if A is true;
[A] = .
0, if A is false
For example, [1 + 1 = 2] = 1 (since 1 + 1 = 2 is true), whereas [1 + 1 = 1] = 0
(since 1 + 1 = 1 is false).
If A is any logical statement, then the integer [A] is known as the truth value
of A.
 
0
Proof of Lemma 1.3.14. From (44), we obtain = 1. But [0 = 0] = 1 (since 0 = 0
0  
0
is true). Comparing these two equalities, we obtain = [0 = 0]. This shows
0
that Lemma 1.3.14 is true for k = 0. Hence, for the rest of this proof, we WLOG
assume that k 6= 0. Thus, the statement k = 0 is false. Hence, [k = 0] = 0. On the
other hand, from k 6= 0, weobtain
 k∈/ {0} = {0, 1, . . . , 0}. Therefore, Lemma 1.3.10
0
(applied to n = 0) yields = 0. Comparing this with [k = 0] = 0, we obtain
  k
0
= [k = 0]. This proves Lemma 1.3.14.
k
Proof of Theorem 1.3.12. Forget that we fixed n, k and S. (This means that instead of
considering n and k as fixed numbers and S as a fixed set, we imagine that Theorem
Enumerative Combinatorics: class notes page 71

1.3.12 begins with the words “For all n ∈ N and k ∈ R and any n-element set S”.
Thus we are proving the same theorem, but we have the freedom to change our
value of k in the proof.)
We proceed by induction on n:
Induction base: If S is a 0-element set, then S = ∅ and thus

(# of k-element subsets of S)
= (# of k-element subsets of ∅)
(  
1, if k = 0; since the empty set ∅ has only one subset,
=
0, if k 6= 0 namely the 0-element subset ∅
( !
1, if k = 0;
= [ k = 0] since the definition of [k = 0] yields [k = 0] =
0, if k 6= 0
 
0
= (by Lemma 1.3.14)
k

 each k ∈ R. In other words, for any k ∈ R and any 0-element set S, we have
for
0
= (# of k-element subsets of S). Thus, Theorem 1.3.12 is proven for n = 0.
k
This completes the induction base.
Induction step: Let m ∈ N. Assume (as the induction hypothesis) that Theorem
1.3.12 holds for n = m. (This means “for n = m and for all values of k ∈ R and all
n-element sets S”, because k and S are not fixed. This is important, since we will
later apply the induction hypothesis to two different values of k.)
Let k ∈ R. Let S be an (m + 1)-element set. We must prove that
 
m+1
= (# of k-element subsets of S) .
k

The set S is an (m + 1)-element set; thus, its size is |S| = m + 1 ≥ 1 > 0. Hence,
the set S is nonempty, i.e., there exists a t ∈ S. Fix such a t.
Now, we shall call a subset of S

• red if it contains t, and

• green if it does not contain t.

Thus, each subset of S is either red or green (but not both at the same time). Thus,
by the sum rule, we have

(# of k-element subsets of S)
= (# of k-element red subsets of S) + (# of k-element green subsets of S) .

We shall now compute the two addends on the right hand side.
Enumerative Combinatorics: class notes page 72

The green subsets of S are the subsets of S that don’t contain t. In other words,
the green subsets of S are exactly the subsets of S \ {t}. Hence,

(# of k-element green subsets of S)

= (# of k-element subsets of S \ {t}) . (56)

But S is an (m + 1)-element set, and thus S \ {t} is an m-element set (since t ∈ S).
Hence, our induction hypothesis shows that Theorem 1.3.12 can be applied to m
and S \ {t} instead of n and S. We thus obtain
 
m
= (# of k-element subsets of S \ {t}) .
k
Comparing this with (56), we find
 
m
(# of k-element green subsets of S) = . (57)
k
What about the red subsets?
Informally, a similar argument works: The red subsets of S are not exactly the
subsets of S \ {t}, but they “correspond to” the latter in a specific way. Namely,
since the red subsets of S are required to contain t, the only “information” that a
red subset of S really “carries” is what other elements (other than t) it contains. In
other words, each red subset R of S “corresponds to” the subset R \ {t} of S \ {t}.
Note that this correspondence changes the size of the subset: Namely, if R was a
k-element red subset of S, then R \ {t} will be a (k − 1)-element subset of S \ {t}.
Thus, the k-element red subsets of S “correspond to” the (k − 1)-element subsets
of S \ {t}.
Formally, this can be restated as follows: The map

f : {k-element red subsets of S} → {(k − 1) -element subsets of S \ {t}} ,

R 7→ R \ {t}

is well-defined and is a bijection. (Indeed, its inverse is

g : {(k − 1) -element subsets of S \ {t}} → {k-element red subsets of S} ,

P 7→ P ∪ {t} .

It is easy to verify that both the map f and its alleged inverse g are well-defined50 ,
and are indeed mutually inverse51 .) Therefore, the bijection principle (applied to
50 This means proving the following claims:

• If R is any k-element red subset of S, then R \ {t} is a (k − 1)-element subset of S \ {t}.

• If P is any (k − 1)-element subset of S \ {t}, then P ∪ {t} is a k-element red subset of S.

Both of these are simple exercises in set-theoretic basics.

51 This
means proving the following claims:
Enumerative Combinatorics: class notes page 73

the bijection f ) yields

|{k-element red subsets of S}|

= |{(k − 1) -element subsets of S \ {t}}| .

In other words,

(# of k-element red subsets of S)

= (# of (k − 1) -element subsets of S \ {t}) . (58)

But S \ {t} is an m-element set. Hence, our induction hypothesis shows that
Theorem 1.3.12 can be applied to m, k − 1 and S \ {t} instead of n, k and S. We
thus obtain
 
m
= (# of (k − 1) -element subsets of S \ {t}) .
k−1

Comparing this with (58), we find

 
m
(# of k-element red subsets of S) = . (59)
k−1

Now, we can finish our computation of (# of k-element subsets of S) that we

started above:

(# of k-element subsets of S)
= (# of k-element red subsets of S) + (# of k-element green subsets of S)
|  {z  } | {z  }
m m
= =
k−1 k
(by (59)) (by (57))
   
m m
= + . (60)
k−1 k

On the other hand, Theorem 1.3.8 (applied to n = m + 1) yields

( m + 1) − 1 ( m + 1) − 1
         
m+1 m m
= + = +
k k−1 k k−1 k

(since (m + 1) − 1 = m). Comparing this with (60), we find

 
m+1
= (# of k-element subsets of S) .
k
• If R is any k-element red subset of S, then ( R \ {t}) ∪ {t} = R.
• If P is any (k − 1)-element subset of S \ {t}, then ( P ∪ {t}) \ {t} = P.

Both of these are simple exercises in set-theoretic basics.

Enumerative Combinatorics: class notes page 74

Now, forget that we fixed k and S. We thus have shown that every k ∈ R and
every (m + 1)-element set S satisfy
 
m+1
= (# of k-element subsets of S) .
k
In other words, Theorem 1.3.12 holds for n = m + 1. This completes the induction
step, and thus the proof of Theorem 1.3.12.

Exercise 1.3.3. Prove the following rules for truth values:

(a) If A and B are two equivalent logical statements, then [A] = [B].
(b) If A is any logical statement, then [not A] = 1 − [A].
(c) If A and B are two logical statements, then [A ∧ B] = [A] [B].
(d) If A and B are two logical statements, then [A ∨ B] = [A] + [B] − [A] [B].
(e) If A, B and C are three logical statements, then

[A ∨ B ∨ C] = [A] + [B] + [C] − [A] [B] − [A] [C] − [B] [C] + [A] [B] [C] .

(f) If A1 , A2 , . . . , Ak are k logical statements (for some k ∈ N), then

[A1 ∧ A2 ∧ · · · ∧ Ak ] = [A1 ] · [A2 ] · · · · · [Ak ] .

(This allows for the possibility of k = 0, in which case the conjunction A1 ∧ A2 ∧

· · · ∧ Ak should be understood as a true statement.)

1.3.5. Integrality and some arithmetic properties

All the numbers we have seen in our tables of binomial coefficients are integers.
This is not a coincidence:

 1.3.16 (Integrality of the binomial coefficients). Let n ∈ Z and k ∈ Z.

Theorem
n
Then, ∈ Z.
k

In order to prove this theorem, it is perhaps easiest to start with the case when
n ∈ N, which we can declare a lemma:
 
n
Lemma 1.3.17. Let n ∈ N and k ∈ Z. Then, ∈ N.
k

Proof of Lemma 1.3.17. Let S = {1, 2, . . . , n}. Then, S is an n-element set. Hence,
Theorem 1.3.12 yields
 
n
= (# of k-element subsets of S) = |{all k-element subsets of S}| ∈ N
k
(since the size of any finite set is ∈ N). This proves Lemma 1.3.17.
Enumerative Combinatorics: class notes page 75

Alternatively, it is not hard to prove Lemma 1.3.17 by induction on n, using

Lemma 1.3.14 for the induction base and Theorem 1.3.8 for the induction step:
Exercise 1.3.4. Prove Lemma 1.3.17 without using Theorem 1.3.12.
We can now prove Theorem 1.3.16:

 Ifn ∈ N, then the claim of Theorem 1.3.16 is true, because

Proof of Theorem 1.3.16.
n
Lemma 1.3.17 yields ∈ N ⊆ Z in this case. Thus, for the rest of this proof, we
k
WLOG assume that n ∈ / N. Hence, n is a negative integer (since n ∈ Z but n ∈ / N).
Therefore, n ≤ −1, so that −n ≥ 1.
/ N, then the claim of Theorem 1.3.16 is true as well (since (43) yields
 If k ∈
n
= 0 ∈ Z in this case). Thus, for the rest of this proof, we WLOG assume that
k
k ∈ N. Hence, k ≥ 0.
Now, Proposition 1.3.7 (applied to −n instead of n) yields
− (−n) k −n + k − 1
   
= (−1) .
k k
In view of − (−n) = n, this rewrites as
k −n + k − 1
   
n
= (−1) .
k k
But −n + k − 1 is an integer satisfying |{z} k −1 ≥ 1 + 0 − 1 = 0. Hence,
−n + |{z}
≥1 ≥0
−n + k− 1 ∈ N. Therefore, Lemma 1.3.17 (applied to −n + k − 1 instead of n)
−n + k − 1

yields ∈ N ⊆ Z. Hence,
k
k −n + k − 1
   
n
= (−1) ∈Z
k k
| {z }
∈Z

(since multiplying an integer by (−1)k clearly results in an integer). Theorem 1.3.16

is thus proven.
 
n
Theorem 1.3.16 tells us that the binomial coefficients are integers, as long
k
as n and k are integers52 . Thus, it makes sense to ask arithmetical questions about
them, such as questions of divisibility and modular congruence53 . And there is a
lot to say about such questions:
 
52 Of n
course, it suffices to assume that n is an integer, since non-integer values of k lead to =0
k
anyway. But we won’t have much use for this generality here.
53 See, e.g., [Hammac15, §5.2], [LeLeMe16, §9.6] or [Grinbe15, §2.2] for the definition and basic

properties of modular congruence (“congruence modulo n”).

Enumerative Combinatorics: class notes page 76

 
p
Theorem 1.3.18. Let p be a prime. Let k ∈ {1, 2, . . . , p − 1}. Then, p | .
k

Theorem 1.3.18  that if p is a prime, then all entries in the p-th row of Pascal’s
 says  
n p
triangle (i.e., all with n = p) are divisible by p, except for the two entries
  k 0
p
and (which are equal to 1). This property actually characterizes primes: If
p
n > 1 is a composite integer (i.e., not a prime), then at least one entry in the n-th
  triangle (except for the entries equal to 1) is not divisible by n. For
row of Pascal’s
4
example, = 6 is not divisible by 4.
2
We shall not prove Theorem 1.3.18 here. See, e.g., [Vorobi02, §2.9] or [19s, The-
orem 2.17.19] for a proof. We shall likewise not prove the following two proposi-
tions, which provide two quick recursive algorithms for telling whether a binomial
coefficient is even or odd:

 Let 
Proposition 1.3.19. a ∈N and b ∈ N. Then:
2a a
(a) We have ≡ mod 2.
2b b  
2a + 1 a
(b) We have ≡ mod 2.
 2b  b
2a
(c) We have ≡ 0 mod 2.
2b + 1   
2a + 1 a
(d) We have ≡ mod 2.
2b + 1 b

 n Letn ∈N.Let a, b ∈ {0, 1, . . . , 2 − 1}. Then:

Proposition 1.3.20. n
2 +a a
(a) We have ≡ mod 2.
 n b  b 
2 +a a
(b) We have n
≡ mod 2.
2 +b b

Proposition 1.3.19 appears (with proof) in [18s-hw1s, Exercise 3]; Proposition

1.3.20 appears (with proof) in [18s-hw1s, Exercise 4]. Let us briefly mention a cu-
rious consequence of these propositions, though. Consider Pascal’s triangle again,
Enumerative Combinatorics: class notes page 77

but replace every odd entry with a “∗” and every even entry with an empty space:

k =0
.
k =1
n=0 → ∗ .
k =2
n=1 → ∗ ∗ .
k =3
n=2 → ∗ ∗ .
n=3 → ∗ ∗ ∗ ∗
n=4 → ∗ ∗
n=5 → ∗ ∗ ∗ ∗
n=6 → ∗ ∗ ∗ ∗
n=7 → ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
n=8 → ∗ ∗

If you are familiar with some fractals, you might recognize this picture as a (fi-
nite step towards) the Sierpinski triangle, with the “∗”s corresponding to the non-
removed points in the triangle. And this is indeed the case: For any r ∈ N, the first
2r rows of Pascal’s triangle (i.e., the rows from n = 0 to n = 2r − 1) form the image
obtained after r steps in the construction of the Sierpinski triangle. This can be
proved by induction on r, and the induction
  step requires understanding how the
n
with n ∈ 2r , 2r + 1, . . . , 2r+1 − 1 is con-

parity of the binomial coefficients
k  
n
nected with the parity of the binomial coefficients with n ∈ {0, 1, . . . , 2r − 1}.
k
Proposition 1.3.20 yields an easy answer to this question.
Here are two more results we shall not prove:

Theorem 1.3.21 (Lucas’s congruence). Let p be a prime. Let a, b ∈ Z. Let c, d ∈

{0, 1, . . . , p − 1}. Then,
    
pa + c a c
≡ mod p.
pb + d b d

Theorem 1.3.22 (Babbage’s congruence). Let p be a prime. Let a, b ∈ Z. Then,

   
pa a
≡ mod p2 .
pb b
Enumerative Combinatorics: class notes page 78

Elementary proofs of Theorem 1.3.21 and Theorem 1.3.22 can be found in [Grinbe17].
Note that Theorem 1.3.21 generalizes Proposition 1.3.19.
Remark 1.3.23. Lucas’s congruence has the following consequence: Let p be a
prime. Let a, b ∈ N. Write a and b in base p as follows:

a = a k p k + a k −1 p k −1 + · · · + a 0 p 0 and
b = bk pk + bk−1 pk−1 + · · · + b0 p0

with k ∈ N and ak , ak−1 , . . . , a0 , bk , bk−1 , . . . , b0 ∈ {0, 1, . . . , p − 1}. (Note that we

allow “leading zeroes” – i.e., any of ak and bk can be 0.) Then,
      
a ak a k −1 a
≡ · · · 0 mod p.
b bk bk − 1 b0

(This can be easily proven by induction on k, using Theorem 1.3.21 in the in-
a
duction step.) This allows for quick computation of remainders of modulo
b
prime numbers.
See [Mestro14] and [Granvi05] for overviews of more complicated divisibilities
and congruences for binomial coefficients.

1.3.6. The binomial formula

You have likely seen the following fact:
Theorem 1.3.24 (the binomial formula). Let x, y ∈ R. Let n ∈ N. Then,
n  
n k n−k
( x + y) = ∑
n
x y .
k =0
k

 
n
This is why the are called the “binomial coefficients” – they appear as
k
coefficients in the binomial formula.
Example 1.3.25. If n = 3, then the claim of Theorem 1.3.24 becomes
3 
3 k 3− k
( x + y) = ∑
3
x y
k =0
k
       
3 0 3−0 3 1 3−1 3 2 3−2 3 3 3−3
= x y + x y + x y + x y
0 |{z} |{z} 1 |{z} |{z} 2 |{z} 3 |{z}
|{z} =1 =y3 = x =y
|{z} = y2 |{z} |{z} =1
=1 =3 =3 =1
3 2 2 3
= y + 3xy + 3x y + x .
Enumerative Combinatorics: class notes page 79

Proof of Theorem
  1.3.24. Let me first “simplify” the  claim: I will rewrite the finite
n n k n−k n k n−k
sum ∑ x y as the infinite sum ∑ x y (that is, a sum over all
k =0 k k ∈Z k
integers k).
In order to do so, I need to justify two things:
 
n k n−k
• I need to explain why the infinite sum ∑ x y is well-defined. (We
k ∈Z k
have seen why finite sums are well-defined in Subsection 1.2.2, but infinite
sums are a different beast. You can easily write down nonsensical infinite
sums like ∑ 1 = 1 + 1 + 1 + · · · or ∑ k = 0 + 1 + 2 + 3 + · · · that will
k ∈N k ∈N
quickly lead you into contradiction-land. If we want to use an infinite sum,
we thus have to explain what it means.)
 
n n k n−k
• I need to explain why this infinite sum equals the finite sum ∑ x y .
k =0 k
 
n k n−k
Let me begin with the first justification: Why is the sum ∑ x y well-
k ∈Z k
defined?
If you think about this question very carefully, you  will
 discover that there is
n k n−k
another question underneath it: Why are the addends x y of this sum well-
k
defined? This is clear when k ∈ {0, 1, . . . , n}, because in this case both exponents k
and n − k are nonnegative integers. But if k < 0 or k > n, one of these exponents
will be negative, and this may cause trouble if x or y is 0 (because 0 j is undefined
when j is negative). We fix this problem by decree: We just decree that an ex-
pression of the form ab is always to be understood as 0 when a = 0, even if b is
undefined. Thus, 0 · 0−1 = 0, despite 0−1 being undefined.54  
n k n−k
Having convinced ourselves that the addends of the sum ∑ x y are
k ∈Z k
well-defined, we can now discuss the existence of the sum itself. Let us look at the
case n = 3. In this case, this sum has the form
y3 + 3xy2 + 3x2 y + |{z}
|· · · + 0{z+ 0 + 0} + |{z} x3 + 0| + 0 +{z0 + · · }· ,
|{z} |{z}
addends addend addend addend addend addends
for k<0 for k=0 for k=1 for k=2 for k=3 for k>3

54 Stillskeptical? Alright, there are a few more things to check. Conventions like this may lead
to errors if applied improperly; for example, after having made this particular convention, we
cannot also decree aa−1 to always equal 1 when a−1 is potentially undefined (since this would
lead to 0 · 0−1 = 1 6= 0). We need to check what exactly we are going to do with our convention.
This is easy: The only situation to which we will apply our ab = 0 convention is when a = 0 and
b = x k yn−k ,and it is 
easy to check that the only
 properties
 we are going to use from this notation
are that x · ax k yn−k = ax k+1 yn−k and y · ax k yn−k = ax k yn−k+1 and a1 b + a2 b = ( a1 + a2 ) b;
obviously, all three of these properties are true when a = 0 (resp. a1 = 0 or a2 = 0), even if the
other factors are undefined.
Enumerative Combinatorics: class notes page 80

 
n k n−k
because Lemma 1.3.10 renders all addends x y with k ∈
/ {0, 1, . . . , n} equal
k
to 0. This is an infinite sum, but only finitely many of its addends are nonzero.
Thus, it is clear how to give it a meaningful value: Just add the nonzero addends
together and drop the zero addends. The philosophy behind this is that zero ad-
dends are not supposed to contribute to sums (no matter how many of them are
present): for example, 0 + 0 + 0 + 0 + · · · = 0.
The same is true for any n ∈ N. Indeed, for any given n ∈ N, wecan  easily see
n k n−k
from Lemma 1.3.10 that the only nonzero addends in the sum ∑ x y are
k ∈Z k
the ones for which k ∈ {0, 1, . . . , n} (which is not saying that all of these addends
must be nonzero). Hence, this sum has only finitely many nonzero addends, and
thus is well-defined.  
n n k n−k
This also explains why this sum equals the finite sum ∑ x y : Indeed,
k =0 k
the only difference between the infinite and the finite sum is the presence of the
addends with k ∈ / {0, 1, . . . , n}; but these addends do not contribute anything,
because they all equal 0. Hence, the two sums are equal. In other words,
  n  
n k n−k n
∑ k x y = ∑ k x k yn−k . (61)
k ∈Z k =0

Thus, it remains to prove that

 
n k n−k
( x + y) = ∑
n
x y . (62)
k ∈Z
k

We will prove (62) by induction on n:

Induction base: If n = 0, then both sides of the equality (62) are equal to 1 (since
( x + y)n = ( x + y)0 = 1 and
  n  
n k n−k n
∑ k x y = ∑ k x k yn−k (by (61))
k ∈Z k =0
0  
0 k 0− k
= ∑ x y (since n = 0)
k =0
k
 
0
= x 0 y 0−0 = 1
0 |{z} |{z}
|{z} =1 =1
=1
(by (44))

in this case). Hence, (62) holds if n = 0. This completes the induction base.
Induction step: Fix m ∈ N. Assume that (62) holds for n = m. We must prove that
(62) holds for n = m + 1.
Enumerative Combinatorics: class notes page 81

We have assumed that (62) holds for n = m. In other words,

 
m k m−k
( x + y) = ∑
m
x y . (63)
k ∈Z
k

Now, for each k ∈ R, we have

( m + 1) − 1 ( m + 1) − 1
     
m+1
= +
k k−1 k
(by Theorem 1.3.8 (applied to n = m + 1))
   
m m
= + (64)
k−1 k
Enumerative Combinatorics: class notes page 82

(since (m + 1) − 1 = m). But

  !
m k m−k
( x + y ) m +1 = ( x + y ) ( x + y)m = ( x + y) ∑ k xy
k ∈Z
|  {z }
m k m−k
=∑ x y
k ∈Z k
(by (63))
   
m k m−k m k m−k
= x ∑ x y + y ∑ x y
k ∈Z
k k ∈Z
k
| {z } | {z }
m k m−k m k m−k
=∑ x x y =∑ y x y
k ∈Z k k ∈Z k
(by (32), or rather (by (32), or rather
its analogue for infinite sums) its analogue for infinite sums)

(by the distributive law)

   
m k m−k m k m−k
= ∑ x x y +∑ y x y
k ∈Z |
k k
 {z } k∈Z |  {z }
m k +1 m − k m k m − k +1
= x y = x y
k k
   
m k +1 m − k m k m − k +1
= ∑ x y + ∑ x y (65)
k ∈Z
k k ∈Z
k
   
m m k m − k +1
= ∑ (k−1)+1 m−(k−1)
x {z } y +∑ x y (66)
k ∈Z
k − 1 | | {z } k ∈Z
k | {z }
=xk = y m +1− k = y m +1− k
 
here, we substituted k − 1 for k in the first sum
(since the map Z → Z, k 7→ k − 1 is a bijection)
   
m m k m +1− k
= ∑ k m +1− k
x y + ∑ x y (67)
k ∈Z
k−1 k ∈Z
k
    
m m k m +1− k
= ∑ k m +1− k
x y + x y (68)
k ∈Z |
k−1 k

{z }
  
m m
= +  x k y m +1− k
k−1 k
(by (31), or rather its analogue for infinite sums)
     
m m m + 1 k m +1− k
= ∑ + k m +1− k
x y = ∑ x y .
k ∈Z |
k − 1 k k ∈Z
k
 {z  }
m+1
=
k
(by (64))

Thus, (62) holds for n = m + 1. This completes the induction step.

Enumerative Combinatorics: class notes page 83

This completes the induction

  proof of (62). In view of (61), this equality (62)
n n k n−k
rewrites as ( x + y)n = ∑ x y . Thus, Theorem 1.3.24 is proven.
k =0 k

It is instructive to try and prove Theorem 1.3.24 without taking the detour through
infinite sums that we did in our above proof. Such a proof would actually be
slightly shorter, but not by much: We would avoid having to explain why an in-
finite sum is well-defined (and what it means), but we would have to jump a few
extra hurdles in our computation. Namely, instead of (65), we would get the equal-
ity
m   m  
m k +1 m − k m k m − k +1
( x + y) = ∑
m
x y +∑ x y .
k =0
k k =0
k
Hence, instead of (67), we would get the equality
m +1   m  
m m k m +1− k
m
( x + y) = ∑ k−1
k m +1− k
x y +∑
k
x y
k =1 k =0

  1 and m + 1 because it originates from substituting

(where the first sum has bounds
m m k +1 m − k
k − 1 for k in the sum ∑ x y ). At this point, we would not be able to
k =0 k
combine the two sums on the right hand side using (31), since these two sums have
(slightly) different bounds of summation. We would have to fix this by “manually”
m +1
moving the bounds of summations so that both sums become ∑ sums:
k =0

m +1   m +1    
m m m
∑ k−1
k m +1− k
x y = ∑
k − 1
k m +1− k
x y −
0 − 1
x 0 y 0+1− k
k =1 k =0 | {z }
=0
(by (43))
 
here, we extended the sum by adding an
 extra addend for k = 0, and then promptly 
subtracted this new addend back
m +1  
m
= ∑ x k y m +1− k
k =0
k − 1
Enumerative Combinatorics: class notes page 84

and
m
m k m +1− k m +1 m k m +1− k
     
m
∑ k xy = ∑
k
x y −
m+1
x m+1 ym+1−(m+1)
k =0 k =0 | {z }
=0
(by Lemma 1.3.10)
 
here, we extended the sum by adding an
 extra addend for k = m + 1, and then promptly 
subtracted this new addend back
m +1  
m k m +1− k
= ∑ x y .
k =0
k

After these transformations, we could combine these two sums and proceed as in
the proof above. (A slightly different but similar proof of Theorem 1.3.24 can be
found in [Grinbe15, Exercise 3.6].)
So why did we take the trouble to turn our finite sum into an infinite sum, if
we could have just as easily gotten away without it? Because this trick is helpful
not merely in proving Theorem 1.3.24. Graham, Knuth and Patashnik ([GrKnPa94,
§5.1]) frequently apply this trick, in order to avoid having “to fuss over boundary
conditions” (i.e., over the bounds of the sums). Of course, this is not always possi-
ble; sometimes, when you extend a sum to a larger index set, the new addends you
get will not all be 0, and the new sum will either fail to be well-defined or at least
differ from the original sum. Thus, as in our proof of Theorem 1.3.24, the trick can
only be applied after checking that the newly added addends are 0. (You can often
do this checking in your head, but it needs to be done nevertheless.)
For the sake of future reference, let us summarize what we have learned about
infinite sums in the above proof of Theorem 1.3.24:

Definition 1.3.26. Let S be a set (not necessarily infinite). For each s ∈ S, let as
be a number.
Assume that only finitely many of these numbers are nonzero (i.e., only finitely
many s ∈ S satisfy as 6= 0). Then, the sum ∑ as is well-defined (even if S is
s∈S
infinite): Its value is just defined to be the finite sum ∑ as .
s∈S;
a s 6 =0

We refer to [Grinbe15, §2.14.15] for a rigorous treatment of such sums (called

“finitely supported sums”). In a nutshell, all their properties can be derived from
the analogous properties of finite sums simply by restricting oneself to certain finite
subsets of S.
The following corollary from Theorem 1.3.24 is perhaps the most famous:
Enumerative Combinatorics: class notes page 85

 
nn
Corollary 1.3.27. Let n ∈ N. Then, ∑ = 2n .
k =0 k
n
Proof
 ofCorollary 1.3.27. Theorem
    to x = 1 and y = 1) yields (1 + 1) =
1.3.24 (applied
n n n n n n
∑ 1 k
1 n−k
= ∑ . Hence, ∑ = (1 + 1)n = 2n . Corollary 1.3.27
k =0 k k =0 k k =0 k
|{z} |{z}
=1 =1
follows.
The following fact ([Grinbe15, Proposition 3.39 (c)]) can be regarded as a coun-
terpart to Corollary 1.3.27:

Proposition 1.3.28. Let n ∈ N. Then,

n  
n
∑ (−1) k = [n = 0] .
k
(69)
k =0

(See Definition 1.3.15 for the meaning of [n = 0] in this proposition.)

Proof of Proposition 1.3.28. If n 6= 0, then n is a positive integer (since n ∈ N) and
thus satisfies 0n = 0. On the other hand, if n = 0, then 0n = 00 = 1. Combining
these two observations, we obtain
(
n 1, if n = 0;
0 = . (70)
0, if n 6= 0
(
1, if n = 0;
On the other hand, the definition of [n = 0] yields [n = 0] = . Com-
0, if n 6= 0
paring this with (70), we find
0n = [ n = 0 ] . (71)
Theorem 1.3.24 (applied to x = −1 and y = 1) yields
n   n  
n k n
((−1) + 1) = ∑
n
(−1) 1|{z} = ∑ (−1)
k n−k
.
k =0 |
k k = 0
k
{z } =1
n
=(−1)k
k

Thus,  n
n  
n
∑ (−1)k k = (− 1 ) + 1  = 0n = [ n = 0 ] (by (71)) .
k =0
| {z }
=0
This proves Proposition 1.3.28.
Enumerative Combinatorics: class notes page 86

1.3.7. Other properties of binomial coefficients

Let us see more properties of binomial coefficients, mostly without proof. We will
prove some of these later (see also [Grinbe15, Chapter 3], or [GrKnPa94, Chapter
5], or essentially any other text on combinatorics). We begin with a summation
identity that is known as the hockey-stick identity:
Theorem 1.3.29 (“Hockey-stick identity”). Let n ∈ N and k ∈ N. Then,
         
0 1 2 n n+1
+ + +···+ = .
k k k k k+1

First proof of Theorem 1.3.29. Recall the notation nk introduced in [19f-hw0s, Exer-
cise 2] (and also in Remark 1.3.5). Now, [19f-hw0s, Exercise 2] says that
n
1
∑ i k = k + 1 ( n + 1 ) k +1 .
i =0

1
Multiplying this equality by yields
k!
1 n k 1 1 ( n + 1 ) k +1
k! i∑
k +1
i = · ( n + 1) = . (72)
=0
k! k + 1 k! · (k + 1)
But (48) (applied to n + 1 and k + 1 instead of n and k) yields

( n + 1 ) k +1 ( n + 1 ) k +1
 
n+1
= =
k+1 ( k + 1) ! k! · (k + 1)
(since Proposition 1.3.2 (applied to k + 1 instead of n) yields (k + 1)! = k! · (k + 1)).
Comparing this with (72), we obtain
1 n n n k
 
n+1 1 i
= ∑ ik = ∑ ik = ∑ .
k+1 k! i=0 i =0
k! i =0
k!
Comparing this with
n n
ik
         
0 1 2 n i
+ + +···+ =∑ = ∑ k! ,
k k k k i =0
k i =0
|{z}
ik
=
k!
(by (48), applied
to i instead of n)

we obtain          
0 1 2 n n+1
+ + +···+ = .
k k k k k+1
This proves Theorem 1.3.29.
Enumerative Combinatorics: class notes page 87

Alternatively, Theorem 1.3.29 can be proven by induction or combinatorially; we

will see those proofs in Section 2.3.
Here is an alternative way to express Theorem 1.3.29:

Corollary 1.3.30 (“Hockey-stick identity”, take 2). Let n ∈ N and k ∈ N. Then,

         
k k+1 k+2 n n+1
+ + +···+ = .
k k k k k+1
 
n m
(Here, the left-hand side means the sum ∑ . When n < k, this sum is
m=k k
empty and thus equals 0.)

Proof of Corollary 1.3.30. We have

 
m
=0 for every m ∈ N satisfying k > m (73)
k
(by Proposition 1.3.6, applied to m instead of n).  
0
Now, we would like to make the following argument: Splitting the sum +
      k
1 2 n
+ + ··· + into two parts (with the first part comprising its first k
k k k
addends, and the second part comprising everything else), we obtain
       
0 1 2 n
+ + +···+
k k k k
k−1
               
0 1 2 k k+1 k+2 n
= + + +···+ + + + +···+
k k k k k k k k
| {z }
=0
(since (73) shows that all
the addends in this sum are 0)
       
k k+1 k+2 n
= + + +···+ ,
k k k k
so that
       
k k+1 k+2 n
+ + +···+
k k k k
         
0 1 2 n n+1
= + + +···+ =
k k k k k+1
(by Theorem 1.3.29).    
0 1
Did you spot the (minor) hole in this argument? We split the sum + +
    k k
2 n
+···+ after its first k addends; but there is no guarantee that this sum
k k
Enumerative Combinatorics: class notes page 88

has k addends to begin with. Namely, if n < k − 1, then it has fewer than k addends.
In this case, our splitting argument does not work. This can be a source of real
errors55 . Fortunately, our proof can be easily salvaged. Namely, our argument
above worked fine in the case when n ≥ k − 1; thus, we only need to deal with the
case when n < k − 1 now. So letus assume
 that n < k − 1. Hence, n + 1 < k < k + 1,
n+1
so that k + 1 > n + 1 and thus = 0 (by Proposition 1.3.6, applied to n + 1
k+1
and k + 1 instead of n and k). Comparing this with
       
k k+1 k+2 n
+ + +···+ = (empty sum) (since n < k − 1 < k)
k k k k
= 0,
         
k k+1 k+2 n n+1
we obtain + + + ··· + = again. This com-
k k k k k+1
pletes our proof of Corollary 1.3.30.

Example 1.3.31. Applying Corollary 1.3.30 to k = 1, we obtain

           
1 2 3 n n+1 n+1
+ + +···+ = =
1 1 1 1 1+1 2

for each n ∈ N. In view of (45) and (46), this equality rewrites as

(n + 1) ((n + 1) − 1) ( n + 1) n n ( n + 1)
1+2+3+···+n = = = .
2 2 2
Thus, we have obtained a new proof of Theorem 1.2.1.

Let us illustrate Corollary 1.3.30 on a picture of Pascal’s triangle. On the fol-

lowing graphic (which shows the case n = 3 and k = 1), the addends on the left
hand side of Corollary 1.3.30 are the ones surrounded by the oblong green ellipse,
whereas the single binomial coefficient on the right hand side is in the red circle:

1 1

1 2 1

1 3 3 1

1 4 6 4 1
.
55 For example, 1 + 2 + · · · + k = (1 + 2) + (3 + 4 + · · · + k) does not hold for k = 1.
Enumerative Combinatorics: class notes page 89

This picture explains where the “hockey-stick identity” got its name.
Next, we are going to express the Fibonacci numbers of Definition 1.1.10 as sums
of binomial coefficients:

Proposition 1.3.32. Let n ∈ N. Then, the Fibonacci number f n+1 is

n 
n−k n−0 n−1 n−2 n−n
        
f n +1 = ∑ = + + +···+ .
k =0
k 0 1 2 n

We shall prove this later (in Subsection 1.4.5). For now, let us remark that roughly
half the addends on theright hand side of Proposition 1.3.32 are 0 (indeed, Propo-
n−k

sition 1.3.6 shows that = 0 for any k ∈ {0, 1, . . . , n} satisfying k > n/2)
k
and thus can be discarded; nevertheless it is easier to have the sum end at k = n
rather than figure out where exactly its nonzero addends stop.
Here are some more examples of binomial identities (i.e., identities involving
binomial coefficients):

Proposition 1.3.33. Let n ∈ N. Then,

−1/2 −1 n 2n
     
= .
n 4 n

Exercise 1.3.5. Prove Proposition 1.3.33.

The following fact is a neat way to restate Proposition 1.3.28:

Proposition 1.3.34. Let n be a positive integer. Then,

           
n n n n n n
+ + +··· = + + + · · · = 2n −1 .
0 2 4 1 3 5

(Here, both sums are infinite, but they are well-defined, since only finitely many
of their addends are nonzero.)

  of Proposition 1.3.34. For each k ∈ N satisfying k ∈

Proof / {0, 1, . . . , n},we have
n n
= 0 (by Lemma 1.3.10). Thus, all addends of the sum ∑ (−1)k with
k   k ∈N k
n
k ∈
/ {0, 1, . . . , n} are 0 (since they contain the factor, which is 0 when k ∈ /
  k
n
{0, 1, . . . , n}). Thus, this sum ∑ (−1)k has only finitely many nonzero ad-
k ∈N k
dends (namely, the ones for k ∈ {0, 1, . . . , n}), and is therefore well-defined. We
Enumerative Combinatorics: class notes page 90

can transform this sum as follows:

     
k n k n n
∑ (−1) k = ∑ (−1) k + ∑ (−1)k k
k ∈N k∈N; k∈N;
k≤n
| {z }
k>n
=0
(by Proposition 1.3.6)
since each k ∈ N satisfies either k ≤ n or k > n,
 

but not both at the same time

   
k n k n
= ∑ (−1) + ∑ (−1) 0 = ∑ (−1)
k

k∈N;
k k∈N; k∈N;
k
k≤n k≤n
|k>n {z }
  =0
n
= ∑ (−1)k k
k∈{0,1,...,n}
since the elements k ∈ N satisfying k ≤ n
 

are precisely the elements of {0, 1, . . . , n}

n  
k n
= ∑ (−1) = [ n = 0] (by Proposition 1.3.28)
k =0
k
=0

(since we don’t have n = 0 (because n is positive)). Hence,

     
k n n n
0 = ∑ (−1) = ∑ (−1) k
+ ∑ (−1) k

k ∈N
k k∈N;
| {z } k k∈N;
| {z } k
k is even =1 k is odd =−1
(since k is even) (since k is odd)
since each k ∈ N is either even or odd,
 

but not both at the same time

   
n n
= ∑ k
+ ∑ (−1)
k
k∈N; k∈N;
k is even k is odd
| {z   }
n
=− ∑
k∈N; k
k is odd
   
n n
= ∑ k
− ∑ k
k∈N; k∈N;
k is even k is odd
 |  {z }   |  {z
 } 
n n n n n n
= + + +··· = + + +···
0 2 4 1 3 5
             
n n n n n n
= + + +··· − + + +··· .
0 2 4 1 3 5
Enumerative Combinatorics: class notes page 91

In other words,
           
n n n n n n
+ + +··· = + + +··· . (74)
0 2 4 1 3 5
Now,
      
n n n
2· + + +···
0 2 4
             
n n n n n n
= + + +··· + + + +···
0 2 4 0 2 4
|    {z   }
n n n
= + + +···
1 3 5
(by (74))
             
n n n n n n
= + + +··· + + + +···
0 2 4 1 3 5
       
n n n n
= + + +··· = ∑
0 1 2 k ∈N
k
   
n n
= ∑ + ∑
k∈N;
k k∈N;
k
k≤n
| {z }
k>n
=0
(by Proposition 1.3.6)
since each k ∈ N satisfies either k ≤ n or k > n,
 

but not both at the same time

     
n n n
= ∑ + ∑ 0= ∑ = ∑
k∈N;
k k∈N; k∈N;
k k∈{0,1,...,n}
k
k≤n k≤n
|k>{z
n
}
=0
since the elements k ∈ N satisfying k ≤ n
 

are precisely the elements of {0, 1, . . . , n}

n  
n
= ∑ = 2n (by Corollary 1.3.27) .
k =0
k

Dividing this equality by 2, we obtain

     
n n n
+ + + · · · = 2n /2 = 2n−1 .
0 2 4
Combining this with (74), we obtain
           
n n n n n n
+ + +··· = + + + · · · = 2n −1 .
0 2 4 1 3 5
This proves Proposition 1.3.34.
Enumerative Combinatorics: class notes page 92

The next fact is highly useful (we will see some of its uses later on):
Proposition 1.3.35 (Trinomial revision formula). Let n, a, b ∈ R. Then,

n−b
     
n a n
= .
a b b a−b

Proposition 1.3.35 is the trinomial revision formula, and is not hard to prove.56 We
shall prove it in the next chapter.
Proposition 1.3.36 (Absorption formula I). Let n ∈ {1, 2, 3, . . .} and m ∈ R. Then,

m m−1
   
m
= .
n n n−1

Proposition 1.3.36 is the absorption formula, and is one of the major tools for trans-
forming binomial coefficients into more convenient forms. It is straightforward to
prove. The following proof of Proposition 1.3.36 is taken from [Grinbe15, proof of
Proposition 3.22]57 :
Proof of Proposition 1.3.36. We have n ∈ {1, 2, 3, . . .} ⊆ N. Thus, (42) (applied to m and n
instead of n and k) yields
m ( m − 1) ( m − 2) · · · ( m − n + 1)
 
m
=
n n!
m ( m − 1) ( m − 2) · · · ( m − n + 1)
= (75)
n · ( n − 1) !
(since Proposition 1.3.2 yields n! = (n − 1)! · n = n · (n − 1)!).
We have n − 1 ∈ N (since n ∈ {1, 2, 3, . . .}). Thus, (42) (applied to m − 1 and n − 1
instead of n and k) yields
m−1 (m − 1) ((m − 1) − 1) ((m − 1) − 2) · · · ((m − 1) − (n − 1) + 1)
 
=
n−1 ( n − 1) !
( m − 1) ( m − 2) ( m − 3) · · · ( m − n + 1)
=
( n − 1) !
(since (m − 1) − 1 = m − 2 and (m − 1) − 2 = m − 3 and (m − 1) − (n − 1) + 1 = m − n + 1).
m
Multiplying both sides of this equality by , we obtain
n
m m−1 m ( m − 1) ( m − 2) ( m − 3) · · · ( m − n + 1)
 
= ·
n n−1 n ( n − 1) !
m · ((m − 1) (m − 2) (m − 3) · · · (m − n + 1))
=
n · ( n − 1) !
m ( m − 1) ( m − 2) · · · ( m − n + 1)
=
n · ( n − 1) !
56 See [17f-hw1s, Exercise 2 (c)] for the most interesting case.
57 where it is stated only for m ∈ Q, but this makes no difference to the proof
Enumerative Combinatorics: class notes page 93

(since m · ((m − 1) (m − 2) (m
 −3) · · · (
m − n +1)) = m (m − 1) (m − 2) · · · (m − n + 1)). Com-
m m m−1
pared with (75), this yields = . This proves Proposition 1.3.36.
n n n−1
Theorem 1.3.37. Let n ∈ N, x ∈ R and y ∈ R. Then,
  n   
x+y x y
= ∑ .
n k =0
k n−k

Theorem 1.3.37 is the Chu–Vandermonde identity (aka the Vandermonde convolution

identity). Note its resemblance to Theorem 1.3.24; it is like a binomial formula for
binomial coefficients instead of powers. We will prove it later (in Section 2.6). For
now, see [Grinbe15, Theorem 3.29] for a proof (which is stated only for x, y ∈ Q,
but applies just as well to any numbers x and y).
The next two theorems appear in [18f-mt3s, Exercise 1]:
Theorem 1.3.38 (Cauchy’s binomial formula). Let n ∈ N and x, y, z ∈ R. Then,
n   n
n n!
∑ k ( x + kz ) k
( y − kz ) n−k
= ∑ k! (x + y)k zn−k .
k =0 k =0

Theorem 1.3.39 (Abel’s binomial formula). Let n ∈ N and x, y, z ∈ R. Then,

n  
n
∑ k |x (x +{zkz)k−}1 (y − kz)n−k = (x + y)n .
k =0
This should be read
as 1 if k=0

(More precisely, [18f-mt3s, Exercise 1] states these theorems in the particular

case when z = 1. The general case can easily be deduced from this particular case,
however.)
Note that both Theorem 1.3.38 and Theorem 1.3.39 generalize the binomial for-
mula; indeed, Theorem 1.3.24 can be recovered from either of them by setting z = 0.
 
n n
Exercise 1.3.6. Let n ∈ N. Prove that ∑ k = n · 2n −1 .
k =0 k

Class of 2019-10-07

1.4. Counting subsets

1.4.1. All subsets
Theorem 1.3.12 tells us how many k-element subsets a given n-element set has,
where n and k are fixed. What if we don’t fix k, and simply ask for the number of
all subsets of an n-element set? The following theorem gives the answer:
Enumerative Combinatorics: class notes page 94

Theorem 1.4.1. Let n ∈ N. Let S be an n-element set. Then,

(# of subsets of S) = 2n .

Exercise 1.4.1. Prove Theorem 1.4.1.

[Hint: Imitate the proof of Theorem 1.3.12.]

We will later give two other proofs of Theorem 1.4.1: an algebraic proof in Sub-
section 1.4.2 and a combinatorial proof in Subsection 1.5.3.58

1.4.2. Lacunar subsets: the basics

We have now seen the following examples of counting:

• An n-element set has 2n subsets. (This is Theorem 1.4.1.)

 
n
• An n-element set has many k-element subsets. (This is Theorem 1.3.12.)
k

Let us now ask subtler questions.

Definition 1.4.2. A set S of integers is said to be lacunar if it contains no two

consecutive integers (i.e., there is no integer i such that both i ∈ S and i + 1 ∈ S).

For example, the set {1, 5, 7} is lacunar, but {1, 5, 6} is not. Note that any 1-
element subset of Z is lacunar, and so is the empty set.
Some people say “sparse” instead of “lacunar”. I want to avoid this notation,
since the word “sparse” has a different meaning in computer science.
It is not hard to list the lacunar subsets of [5]: They are

∅, {1} , {2} , {3} , {4} , {5} , {1, 3} ,

{1, 4} , {1, 5} , {2, 4} , {2, 5} , {3, 5} , {1, 3, 5} .

We can now ask several natural questions:

Question 1.4.3. For given n, k ∈ N:

(a) How many lacunar subsets does [n] have?
(b) How many k-element lacunar subsets does [n] have?
(c) What is the largest size of a lacunar subset of [n] ?

Let us start with (c):

58 The impatient reader can find the latter proof in [19f-hw0s, Exercise 1 (a)]. (More precisely,
[19f-hw0s, Exercise 1 (a)] is the particular case of Theorem 1.4.1 when S = [n]. But it can easily
be adapted to the general case.)
Enumerative Combinatorics: class notes page 95

Definition 1.4.4. Let x ∈ R. Then:

• We let b x c denote the largest integer ≤ x. This integer b x c is called the floor
of x, or “rounding down x”.

• We let d x e denote the smallest integer ≥ x. This integer d x e is called the

ceiling of x, or “rounding up x”.

Example 1.4.5. We have

j√ k
b3c = 3; 2 = 1; bπ c = 3; b−π c = −4;
l√ m
d3e = 3; 2 = 2; dπ e = 4; d−π e = −3.

Let us note that each x ∈ R satisfies b x c ≤ x ≤ d x e.

It is intuitively clear that there is no way to pick more than dn/2e distinct num-
bers from the set {1, 2, . . . , n} without picking two consecutive integers. Let us state
and rigorously prove this:

Proposition 1.4.6. Let n ∈ N. Then, the largest size of a lacunar subset of [n] is
dn/2e.

Our proof of this will rely on the following basic fact:

Theorem 1.4.7. Let A be a finite set. Let B be a subset of A. Then:

(a) We have | A \ B| = | A| − | B|.
(b) We have | B| ≤ | A|.
(c) If | B| = | A|, then B = A.

This theorem is fundamental. Theorem 1.4.7 (a) is known as the difference rule,
and is part of the reason why A \ B is called the “set difference” of A and B 59 .
Parts (b) and (c) of Theorem 1.4.7 are (in a sense) the combinatorial manifestation
of the principle “the whole is greater than its part”. But before you declare them
completely obvious, keep in mind that Theorem 1.4.7 (c) is not true for infinite sets!
Indeed, if we set B = N and A = Z, then B ⊆ A holds, and so does | B| = | A|
(in the sense that the infinite sets B and A have the same cardinality), but it is not
true that B = A. Thus, it is worth proving the theorem. Fortunately, it is a simple
application of the sum rule:
Proof of Theorem 1.4.7. We know that B ⊆ A. Hence, the set A is the union of the two
disjoint sets B and A \ B. Hence, Theorem 1.1.3 (applied to S = A, k = 2, S1 = B and
59 although this terminology persists even in the case when B is not a subset of A, despite the fact
that | A \ B| is usually not | A| − | B| in this case
Enumerative Combinatorics: class notes page 96

S2 = A \ B) yields | A| = | B| + | A \ B|. Thus, | A \ B| = | A| − | B|. This proves Theorem

1.4.7 (a).
(b) We have | A| = | B| + | A \ B| ≥ | B|. In other words, | B| ≤ | A|. This proves Theorem
| {z }
≥0
1.4.7 (b).
(c) Assume that | B| = | A|. Hence, | B| = | A| = | B| + | A \ B|. Subtracting | B| from both
sides of this equation, we obtain 0 = | A \ B|, so that | A \ B| = 0. But this shows that A \ B
is the empty set. Hence, A ⊆ B, so that A = B (because B ⊆ A). In other words, B = A.
This proves Theorem 1.4.7 (c).

Proof of Proposition 1.4.6. The lacunar set60

{all odd elements of [n]} = {1, 3, 5, . . . , (n or n − 1)}
= {1 < 3 < 5 < · · · < (n or n − 1)}
(where the last element is n if n is odd, and n − 1 if n is even) is a subset of [n] and
has size dn/2e. Thus, there exists a lacunar subset of [n] having size dn/2e. Hence,
we only need to prove that no higher size is possible.
Let us prove this. Let L be a lacunar subset of [n]. We shall prove that | L| ≤
dn/2e.
Indeed, let L+ be the set {s + 1 | s ∈ L}. (In other words, L+ is the set L “shifted
by 1 to the right”, in the sense that L+ is obtained from L by adding 1 to each
element. For example, if L = {2, 4, 7, 10}, then L+ = {3, 5, 8, 11}.)
It is clear that | L| = | L+ |. (Formally speaking, this is a consequence of the
bijection principle, because the map L → L+ , s 7→ s + 1 is clearly a bijection.)
We have L ⊆ [n] = {1, 2, . . . , n} and therefore L+ ⊆ {2, 3, . . . , n + 1} ⊆ [n + 1].
Also, L ⊆ [n] ⊆ [n + 1]. Thus, we know that both L and L+ are subsets of [n + 1].
Hence, their union L ∪ L+ is a subset of [n + 1] as well. Therefore, Theorem 1.4.7
(b) (applied to L ∪ L+ and [n + 1] instead of B and A) yields
L ∪ L+ ≤ |[n + 1]| = n + 1 (since [n + 1] = {1, 2, . . . , n + 1}) .
On the other hand, recall that L is lacunar. Thus, the sets L and L+ are disjoint61 .
Hence, (4) (applied to X = L and Y = L+ ) yields
L ∪ L+ = | L| + L+ = | L| + | L| = 2 · | L| .
|{z}
=| L|
(since | L|=| L+ |)

60 We are using the notation { a1 < a2 < · · · < ak } (where a1 , a2 , . . . , ak are some integers) to mean
the set { a1 , a2 , . . . , ak } when we want to simultaneously assert that a1 < a2 < · · · < ak . For
example, {2 < 5 < 8} means the set {2, 5, 8}, whereas {4 < 2 < 7} is not well-defined.
61 Proof. Let t ∈ L ∩ L+ . Thus, t ∈ L ∩ L+ ⊆ L and t ∈ L ∩ L+ ⊆ L+ = { s + 1 | s ∈ L }. In other

words, there exists some s ∈ L such that t = s + 1. Consider this s. Now, the set L contains
the two consecutive integers s and s + 1 (since s ∈ L and s + 1 = t ∈ L). But L contains no
two consecutive integers (since L is lacunar). These two statements clearly contradict each other.
Thus we have obtained a contradiction.
Now, forget that we fixed t. We thus have found a contradiction for each t ∈ L ∩ L+ . Hence,
there exists no t ∈ L ∩ L+ . In other words, the sets L and L+ are disjoint.
Enumerative Combinatorics: class notes page 97

Hence,
2 · | L| = L ∪ L+ ≤ n + 1.
Thus,
| L| ≤ (n + 1) /2 = |{z} 1/2 < dn/2e + 1.
n/2 + |{z}
≤dn/2e <1

Hence,

| L| ≤ (dn/2e + 1) − 1 (because | L| and dn/2e + 1 are integers)

= dn/2e .

Now, forget that we fixed L. We thus have shown that any lacunar subset L of [n]
satisfies | L| ≤ dn/2e. In other words, any lacunar subset of [n] has size ≤ dn/2e.
Then, the largest size of a lacunar subset of [n] is dn/2e (because we already know
that there exists a lacunar subset of [n] having size dn/2e). This proves Proposition
1.4.6.
As another application of Theorem 1.4.7, let us prove Theorem 1.4.1 again:62

Second proof of Theorem 1.4.1. We have |S| = n, since S is an n-element set. If B is any
subset of S, then Theorem 1.4.7 (b) (applied to A = S) yields | B| ≤ |S| = n and therefore
| B| ∈ {0, 1, . . . , n}. Hence, the sum rule yields63

(# of subsets of S)
= ∑ (# of subsets B of S satisfying | B| = k)
k ∈{0,1,...,n}
| {z }
=(# of k-element
  subsets of S )
n
=
k
(by Theorem 1.3.12)
!
  n   n
n n
= ∑ k
=∑
k
since the symbol ∑ stands for ∑
k ∈{0,1,...,n} k =0 k =0 k∈{0,1,...,n}

= 2n (by Corollary 1.3.27) .

This proves Theorem 1.4.1.

1.4.3. Intermezzo: SageMath

Counting using SageMath. Our next goal is to address Question 1.4.3 (b). A good
way to start is by collecting some data. For example, counting the lacunar subsets of
[n] for the first (say) 10 positive integer values of n is a straightforward task best left
to a computer. The easiest way to do so is to use the SageMath computer algebra
62 The first proof was given in the solution to Exercise 1.4.1.
63 The sum rule that we are using here is Theorem 1.2.5. (Namely, we are applying Theorem 1.2.5
to {subsets of S} and {0, 1, . . . , n} instead of S and W, where the map f : {subsets of S} →
{0, 1, . . . , n} is defined by f ( B) = | B| for all B ∈ {subsets of S}.)
Enumerative Combinatorics: class notes page 98

system [sage], since it has almost all of the necessary functionality built in already.
You can either install SageMath on your computer (see here for Linux binaries, here
for Linux sources, and here for less up-to-date Windows installers), or run it on the
CoCalc cloud, or (if your computations are quick) use the SageMathCell browser
interface. The latter option is the easiest one, in my experience.
First, let us find all 3-element subsets of [5] = {1, 2, 3, 4, 5}. Typing in
Subsets ( {1 , 2 , 3 , 4 , 5 } , 3 )

returns:
Subsets of {1, 2, 3, 4, 5} of size 3
Okay, but how do we list them? For that, we use
list ( Subsets ( {1 , 2 , 3 , 4 , 5 } , 3 ) )

and obtain
[{1, 2, 3},
{1, 2, 4},
{1, 2, 5},
{1, 3, 4},
{1, 3, 5},
{1, 4, 5},
{2, 3, 4},
{2, 3, 5},
{2, 4, 5},
{3, 4, 5}]
These are our subsets indeed.64 Likewise, we can get SageMath to list all subsets
of [4] (of all sizes):
list ( Subsets ( {1 , 2 , 3 , 4 } ) )

This results in SageMath printing all 16 subsets of [4].

Now, let us see which of them are lacunar – more precisely, let us make SageMath
figure that out. For this, we need to define a function that checks whether a given
subset is lacunar. We call it is_lacunar, and define it as follows:65
def is_lacunar ( I ) :
# Check whether a set ‘I ‘ of integers is lacunar .
return all ( e + 1 not in I for e in I )
66 The all function here stands for the logical “for all” quantifier, so you should
read “all( e + 1 not in I for e in I )” as “for all e ∈ I, we have e + 1 ∈ / I”
(which is indeed a way to say that I is lacunar).
You can check that this function works as intended:
64 Squarebrackets [...] delimit lists in SageMath.
65 Thesecond line in this definition is unnecessary; it is just a comment explaining to us humans
what the function is supposed to do.
66 When pasting this code, don’t forget the indentation: The second and third lines need to be

indented, or else SageMath will not recognize that they are part of the definition of the function!
Enumerative Combinatorics: class notes page 99

is_lacunar ( {1 , 4 , 6 } )

returns True, whereas

is_lacunar ( {1 , 4 , 5 } )

returns False.
If you are using the SageMathCell browser interface, you need to put both the
definition of the function and these tests into the input box67 . So you have to write:
def is_lacunar ( I ) :
# Check whether a set ‘I ‘ of integers is lacunar .
return all ( e + 1 not in I for e in I )

is_lacunar ( {1 , 4 , 6 } )

Better yet, you can use the print instruction to do both of these tests at once:
def is_lacunar ( I ) :
# Check whether a set ‘I ‘ of integers is lacunar .
return all ( e + 1 not in I for e in I )

print ( is_lacunar ( {1 , 4 , 6 } ) )
print ( is_lacunar ( {1 , 4 , 5 } ) )

(Without the print instruction, SageMathCell would only give you the output of
the very last line.)
Now, let us make SageMath list the lacunar subsets of [4]. We shall ask it to do so
by walking through all the subsets of [4], and filtering them for lacunarity. Having
defined the is_lacunar function, this is easy:
[ A for A in Subsets ( {1 , 2 , 3 , 4 } ) if is_lacunar ( A ) ]

The meaning of this expression is very similar to the mathematical set-builder no-
tation { A | A ∈ P ({1, 2, 3, 4}) and A is lacunar}; the main difference is that we
are using square brackets rather than set-braces and therefore obtain a list rather
than a set. The result is what we would expect:
[{}, {1}, {2}, {3}, {4}, {1, 3}, {1, 4}, {2, 4}]
This is a list of all lacunar subsets of [4], each listed only once. Thus, the length of
the list is the number of said subsets. We can let SageMath compute this length as
follows:
len ( [ A for A in Subsets ( {1 , 2 , 3 , 4 } ) if is_lacunar ( A ) ] )

As expected, we get 8. In general, the SageMath function len can be applied both
to lists and to sets, yielding the length of the former and the size of the latter.
We can now put these to use and compute the number of lacunar subsets of [n]
for any n ∈ N that is not too high. For example, for n = 7, we just need to ask
SageMath the following:
67 and,
generally, you must keep the definition in the input box whenever you want to use it – as
SageMathCell won’t remember it for you
Enumerative Combinatorics: class notes page 100

len ( [ A for A in Subsets ( {1 , 2 , 3 , 4 , 5 , 6 , 7 } ) if is_lacunar ( A ) ] )

and we quickly obtain 34.

A bit of typing thus lets us build the following table of numbers of lacunar
subsets:
n 0 1 2 3 4 5 6 7 8 9 ···
# 1 2 3 5 8 13 21 34 55 89 ···
(where each entry in the bottom row is the # of lacunar subsets of [n] for the
appropriate value of n).
At this point, you may have a good guess of what the general answer is. We
will get to it, but beforehand, let me say a few words about how we could have
used SageMath better than we did above, because otherwise I would feel bad for
teaching you bad programming habits.

Working with general n. The way I showed above required me to type in a new
command for each new value of n. Why not have the computer calculate the whole
table at once? For this, we need to define a function which takes a number n as
input and returns the # of lacunar subsets of n. Let me call this function num_lacs,
and define it as follows:
def num_lacs ( n ) :
# Return the number of all lacunar subsets of ‘[ n ] ‘.
N = set ( range (1 , n + 1 ) ) # This is the set of ‘1 , 2 , ... , n ‘.
return len ( [ A for A in Subsets ( N ) if is_lacunar ( A ) ] )

Here is what this function does: It first constructs the set [n] (which it calls N). It
does this by first constructing the list (1, 2, . . . , n) (which is done via range(1, n+
1) because in general, range(a, b) returns the list ( a, a + 1, . . . , b − 1)), and then
making a set out of this list using the set function. Then, it returns the number of
all lacunar subsets of this set N = [n] (in the same way as we did above).
Having defined this function, we can make SageMath list its values for all n ∈
{0, 1, . . . , 9} as follows:68
for n in range (0 , 10 ) :
print ( " The number of lacunar subsets of [ " + str ( n ) + " ] is " + str (
num_lacs ( n ) ) )

This code runs through all elements n of range(0, 10), which means all elements
of {0, 1, . . . , 9}, and prints a certain piece of text for each of them, which always has
the form “The number of lacunar subsets of [”, followed by the value of n, followed
by the text “] is ”, followed by the # of lacunar subsets of [n]. Here is the output:
The number of lacunar subsets of [0] is 1
The number of lacunar subsets of [1] is 2
68 The following code fragment has only two lines. The linebreak between str( and num_lacs(n))
is a consequence of lack of space on this page.
Enumerative Combinatorics: class notes page 101

The number of lacunar subsets of [2] is 3

The number of lacunar subsets of [3] is 5
The number of lacunar subsets of [4] is 8
The number of lacunar subsets of [5] is 13
The number of lacunar subsets of [6] is 21
The number of lacunar subsets of [7] is 34
The number of lacunar subsets of [8] is 55
The number of lacunar subsets of [9] is 89
This is essentially our above table, in text form, generated all at once. (We could go
further and have SageMath generate the LaTeX sourcecode of the whole table, but
this would distract us here.)

Optimizing the code. Our code now looks like this:

def is_lacunar ( I ) :
# Check whether a set ‘I ‘ of integers is lacunar .
return all ( e + 1 not in I for e in I )

def num_lacs ( n ) :
# Return the number of all lacunar subsets of ‘[ n ] ‘.
N = set ( range (1 , n + 1 ) ) # This is the set of ‘1 , 2 , ... , n ‘.
return len ( [ A for A in Subsets ( N ) if is_lacunar ( A ) ] )

for n in range (0 , 10 ) :
print ( " The number of lacunar subsets of [ " + str ( n ) + " ] is " + str (
num_lacs ( n ) ) )

It is good enough for computing the required values for all n ∈ {0, 1, . . . , 18}, but
higher n’s will bring SageMath to its knees. (The SageMathCell browser interface
has a timeout – if your computation takes longer than 30 seconds69 , it will stop.
But even on your own machine, you don’t want to wait for days.)
It turns out that we can speed up our code a bit by optimizing the num_lacs
function. In its current form, this function first computes the list of all lacunar
subsets of [n] and then returns the length of this list. Thus, it walks through all
subsets of [n], and whenever it encounters a lacunar one, it “writes it down” into a
list (in memory). It would be faster to simply count these lacunar subsets, without
writing them down, since we only want to know their number. The easiest way
to do so is to simply start with the number 0 and then add 1 to it every time a
lacunar subset is encountered. This suggests the following modified version of the
num_lacs function:
def num_lacs ( n ) :
# Return the number of all lacunar subsets of ‘[ n ] ‘.
N = set ( range (1 , n + 1 ) ) # This is the set of ‘1 , 2 , ... , n ‘.
return sum ( 1 for A in Subsets ( N ) if is_lacunar ( A ) )

69 I believe it is 30 seconds.
Enumerative Combinatorics: class notes page 102

The line
sum ( 1 for A in Subsets ( N ) if is_lacunar ( A ) )

here corresponds to the mathematical expression ∑ 1, which is exactly the #

A⊆[n];
A is lacunar
of lacunar subsets of [n] (by the identity (30)).
This speeds up our code, but only by a little, because the true bulk of the time
is taken by walking through all the 2n subsets of [n]. Nevertheless, it is worth
remembering this tactic for reducing needless busywork. There are many situations
where it does make a major difference.

The manual. You have now seen a few pieces of SageMath’s functionality and
a few examples of how to use it. Of course, there is a lot more. Since this text
is not a SageMath reference book, you will need other sources to discover what
else SageMath can do for you. An overview is given in the text [SageBook]. In
order to systematically explore the abilities of SageMath, you will have to consult
the SageMath reference manual, which can currently be found at https://doc.
sagemath.org/html/en/reference/ . In particular, the Combinatorics page is of
most relevance to us.

1.4.4. Counting lacunar subsets

Let us now return to theorems and proofs. As our SageMath computations heavily
suggest, we should expect the # of lacunar subsets of [n] to be a Fibonacci number,
namely f n+2 . And this holds for all n ∈ N indeed. Moreover, it holds for n = −1
as well, if we extend our notation [k ] for {1, 2, . . . , k } to encompass the case when k
is negative in the following way:

Definition 1.4.8. Let k ∈ Z. Then, [k ] shall denote the set {1, 2, . . . , k }. This is
understood to be the empty set ∅ when k ≤ 0.

Let us state our claim about the # of lacunar subsets of [n] as a proposition:

Proposition 1.4.9. Let n ∈ {−1, 0, 1, . . .}. Then,

(# of lacunar subsets of [n]) = f n+2 .

There are several ways to prove Proposition 1.4.9.

First proof of Proposition 1.4.9 (sketched). The following proof is similar to that of The-
orem 1.3.12 above, so we will be brief and leave parts of it to the reader. A complete
Enumerative Combinatorics: class notes page 103

version of this proof can be found in [Grinbe15, solution to Exercise 4.3]70 or in

[17f-hw1s, Exercise 4 (c)]71 .
Forget that we fixed n. For each n ∈ {−1, 0, 1, . . .}, we define the integer `n by

`n = (# of lacunar subsets of [n]) .

Note that `0 = 1 (since the only lacunar subset of the empty set [0] is the empty set
itself) and `−1 = 1 (for similar reasons).
Now, we claim that

` n = ` n −1 + ` n −2 for each integer n ≥ 1. (76)

[Proof of (76): Let n ≥ 1 be an integer. We shall call a subset of [n]

• red if it contains n, and

• green if it does not contain n.

Thus, each subset of [n] is either red or green (but not both at the same time). Thus,
by the sum rule, we have

(# of lacunar subsets of [n])

= (# of lacunar red subsets of [n]) + (# of lacunar green subsets of [n]) .

We shall now compute the two addends on the right hand side.
First of all, we observe that the green subsets of [n] are precisely the subsets of
[n − 1]. Hence,

(# of lacunar green subsets of [n])

= (# of lacunar subsets of [n − 1]) = `n−1 (77)

(since `n−1 was defined to be the # of lacunar subsets of [n − 1]).

What about the red subsets?
A lacunar red subset of [n] must contain n (since it is red), and thus cannot
contain n − 1 (since it is lacunar, but n − 1 and n are two consecutive integers).
Hence, if R is a lacunar red subset of [n], then R \ {n} is not only a subset of [n − 1]
but actually a subset of [n − 2]. Hence, the map

f : {lacunar red subsets of [n]} → {lacunar subsets of [n − 2]} ,

R 7→ R \ {n}

70 Keep in mind that [Grinbe15, Exercise 4.3] is an equivalent restatement of our Proposition 1.4.9
(more precisely, our Proposition 1.4.9 is [Grinbe15, Exercise 4.3], applied to n + 2 instead of n).
71 Note that [17f-hw1s, Exercise 4 (c)] only covers the case when n ∈ N. But the remaining case

(viz., the case n = −1) is trivial.

Enumerative Combinatorics: class notes page 104

is well-defined. Moreover, it is easy to see that this map is a bijection (because if

P is a lacunar subset of [n − 2], then P ∪ {n} is a lacunar subset of [n]). Therefore,
the bijection principle (applied to the bijection f ) yields

|{lacunar red subsets of [n]}| = |{lacunar subsets of [n − 2]}| .

In other words,

(# of lacunar red subsets of [n])

= (# of lacunar subsets of [n − 2]) = `n−2 (78)

(since `n−2 was defined to be the # of lacunar subsets of [n − 2]).

Now,

(# of lacunar subsets of [n])

= (# of lacunar red subsets of [n]) + (# of lacunar green subsets of [n])
| {z } | {z }
=`n−2 =`n−1
= ` n −2 + ` n −1 = ` n −1 + ` n −2 .

This proves (76).]

The rest of the proof proceeds by the same argument that we used for Proposition
1.1.11: We have to show that

(# of lacunar subsets of [n]) = f n+2 for each n ∈ {−1, 0, 1, . . .} .

In other words, we have to show that

` n = f n +2 for each n ∈ {−1, 0, 1, . . .}

(since `n = (# of lacunar subsets of [n])). In other words, we have to show that

the sequences (`−1 , `0 , `1 , . . .) and ( f 1 , f 2 , f 3 , . . .) are identical. Both of them have
the property that their first entries are 1’s (since `−1 = 1 and f 1 = 1), their second
entries are 1’s (since `0 = 1 and f 2 = 1), and each of their further entries equals
the sum of the preceding two entries72 . Thus, both sequences start with the same
two entries, and then are built out of these two entries according to the same rule.
Hence, the two sequences must be the same. This proves Proposition 1.4.9.
The next proof of Proposition 1.4.9 we shall give is an example of a bijective proof :
a proof that uses the bijection principle to show that two numbers are equal. (Previ-
ous examples of such proofs are our proofs of Proposition 1.1.5 and of Proposition
1.1.9, as well as our third proof of Theorem 1.2.1.)

72 Indeed, for the shifted Fibonacci sequence ( f 1 , f 2 , f 3 , . . .), this is clear. For the sequence
(`−1 , `0 , `1 , . . .), this follows from (76).
Enumerative Combinatorics: class notes page 105

Second proof of Proposition 1.4.9 (sketched). Let us consider domino tilings of the rect-
angle Rn+1,2 . The # of such tilings is dn+1,2 (because we defined dn+1,2 to be this #).
Hence,

(# of domino tilings of Rn+1,2 )

= dn+1,2 = f (n+1)+1 (by Proposition 1.1.11, applied to n + 1 instead of n)
= f n +2 . (79)

If D = {(i, j) , (i + 1, j)} is a horizontal domino, then we say that D starts in

column i. In other words, a horizontal domino starts in its western column.
We want to define a map

h : {domino tilings of Rn+1,2 } → {lacunar subsets of [n]}

as follows: If T is any domino tiling of Rn+1,2 , then h ( T ) shall be the set of all
i ∈ [n + 1] such that at least one horizontal domino of T starts in column i.
Before we do anything with this map h, we need to check that it is well-defined.
First, let us illustrate it with an example: If n = 13 and if T is the domino tiling

1 2 3 4 5 6 7 8 9 10 11 12 13 14

(where the numbers on the top are just the numbers of the columns), then h ( T ) =
{2, 4, 7, 9, 13}, since the horizontal dominos of T start in columns 2, 4, 7, 9, 13.
The proof that h is well-defined will rest on the following two observations:

Observation 1: Let T be a domino tiling of Rn+1,2 . Then, no horizontal

domino of T starts in column n + 1.

[Proof of Observation 1: This is clear, since any horizontal domino that starts in
column n + 1 would continue into column n + 2 and thus fail to stay inside Rn+1,2 .]

Observation 2: Let T be a domino tiling of Rn+1,2 . Then, no two horizon-

tal dominos of T start in consecutive columns.

[Proof of Observation 2: This is easy to see by induction on n (as in the proof of

Proposition 1.1.9). An alternative proof proceeds as follows: Assume the contrary.
Thus, there are two horizontal dominos of T that start in consecutive columns. In
other words, there are two consecutive columns of Rn+1,2 , each of which has a
horizontal domino of T start in it. In other words, there exists an i ∈ [n] such that
each of the two consecutive columns i and i + 1 has a horizontal domino of T start
in it. Let us pick the smallest such i. Therefore, the column i − 1 has no horizontal
domino of T start in it (since otherwise, i − 1 would be a smaller candidate than i,
Enumerative Combinatorics: class notes page 106

because each of the two consecutive columns i − 1 and (i − 1) + 1 = i would have

a horizontal domino of T start in it).
We know that column i has a horizontal domino of T start in it, and that column
i + 1 has a horizontal domino of T start in it. Let these dominos be D and E (labeled
in such a way that D starts in column i, and E starts in column i + 1). Clearly,
the dominos D and E cannot lie in the same row, since otherwise they would
overlap. Thus, one of D and E lies in row 1, and the other lies in row 2 (where we
number rows from the bottom, so that row 1 is the bottom row of Rn+1,2 ). Let us
WLOG assume that D lies in row 1 and E lies in row 2 (since the opposite case is
analogous). Thus, D = {(i, 1) , (i + 1, 1)} and E = {(i + 1, 2) , (i + 2, 2)}. Here is
how D and E look like:
i i+1 i+2

E E

D D

(please pretend that the slightly oblong rectangles are actually squares).
Now, the square (i, 2) must belong to some domino F ∈ T. This domino F cannot
be vertical (since it would otherwise overlap with D), and thus is horizontal. More-
over, this horizontal domino F cannot start in column i (since it would otherwise
overlap with E), and thus must start in column i − 1. This contradicts the fact that
the column i − 1 has no horizontal domino of T start in it. This contradiction shows
that our assumption was false; hence, Observation 2 is proven.]
Now, we can confirm that the map h defined above is well-defined: If T is a
domino tiling of Rn+1,2 , then the set that we wanted to call h ( T ) is actually a
subset of [n] (since it is a subset of [n + 1], but Observation 1 shows that it does not
contain n + 1), and is lacunar (by Observation 2). Thus, h is well-defined.
Next, I claim that h is a bijection. Indeed, let me construct an inverse to h:
Consider the map

g : {lacunar subsets of [n]} → {domino tilings of Rn+1,2 }

that sends each lacunar subset S of [n] to the domino tiling of Rn+1,2 that is con-
structed by placing two horizontal dominos into columns i and i + 1 for each i ∈ S,
and filling the rest of Rn+1,2 with vertical dominos. It is easy to see that h ◦ g = id.
It is somewhat trickier to see that g ◦ h = id; this requires the following two obser-
vations:

Observation 3: Let T be a domino tiling of Rn+1,2 . Let i ∈ [n + 1] be such

that a horizontal domino of T starts in column i. Then, two horizon-
tal dominos of T start in column i, and these two dominos cover both
columns i and i + 1.

Observation 4: Let T be a domino tiling of Rn+1,2 . Let i ∈ [n + 1] be such

that no horizontal domino of T starts in column i and no horizontal
Enumerative Combinatorics: class notes page 107

domino of T starts in column i − 1 either. Then, column i is covered by

a vertical domino of T.

It is not hard to prove these two observations. In a nutshell: Observation 3

follows from Observation 2 (because a horizontal domino only covers one square
of column i, and the other square must be covered by another domino, which must
also be horizontal in order to avoid overlap). Observation 4 follows by noticing
that any horizontal domino that contains (i, 1) or (i, 2) must start either in column
i or in column i − 1.
Together, Observation 3 and Observation 4 show that g ◦ h = id (make sure you
understand how), and thus the maps g and h are mutually inverse (since h ◦ g = id
holds as well). Hence, the map h is invertible, i.e., a bijection. Thus, the bijection
principle yields

|{domino tilings of Rn+1,2 }| = |{lacunar subsets of [n]}| .

In other words,

(# of domino tilings of Rn+1,2 ) = (# of lacunar subsets of [n]) .

Comparing this with (79), we find

(# of lacunar subsets of [n]) = f n+2 .

This proves Proposition 1.4.9 again.

1.4.5. Counting k-element lacunar subsets

Proposition 1.4.9 has answered Question 1.4.3 (a). It remains to answer Question
1.4.3 (b). This is done by the following proposition:

Proposition 1.4.10. Let n ∈ Z and k ∈ N be such that k ≤ n + 1. Then,

n+1−k
 
(# of k-element lacunar subsets of [n]) = .
k

(To be pedantic, Proposition 1.4.10 only answers Question 1.4.3 (b) in the case
when k ≤ n + 1. But the remaining case is trivial: If k > n + 1, then k > n + 1 > n,
and thus the set [n] has no k-element subsets73 , let alone k-element lacunar subsets.)
Proposition 1.4.10 appears in [17f-hw2s, Exercise 3 (a)], with two proofs74 . One
way to prove it is by strong induction on n, similar to the first proof of Proposition
73 byTheorem 1.4.7 (b)
74 Tobe very pedantic: [17f-hw2s, Exercise 3 (a)] only states Proposition 1.4.10 in the case when
n ∈ N. But the remaining case is trivial (since k ≤ n + 1 leads to k = 0 when n is negative, and
thus we have to count 0-element subsets of an empty set, which is not a challenge at this point).
Enumerative Combinatorics: class notes page 108

1.4.9. (This is the second solution of [17f-hw2s, Exercise 3 (a)].) There is, however,
a more enlightening proof than that, which is bijective and explains the binomial
coefficient on the right hand side combinatorially. This latter proof will rely on the
following fact:75

Proposition 1.4.11. Let S be a finite set of integers. Then, there exists a unique
tuple (s1 , s2 , . . . , sk ) of integers satisfying {s1 , s2 , . . . , sk } = S and s1 < s2 < · · · <
sk .

Proposition 1.4.11 is just saying that any finite set of integers can be uniquely
listed in increasing order (with no repetitions). This is sufficiently self-evident that
I have never seen a combinatorics textbook that proves it. A detailed proof can be
found in [Grinbe15, proof of Theorem 2.46]76 , and a sketch at [19f-hw0s, Outline of
a proof of Proposition 1.3].
Let us furthermore make the following convention (which we have already made
in our proof of Proposition 1.4.6):

Definition 1.4.12. Let a1 , a2 , . . . , ak be k numbers (integers, rational numbers

or real numbers). The notation “{ a1 < a2 < · · · < ak }” shall denote the set
{ a1 , a2 , . . . , ak } and simultaneously assert that a1 < a2 < · · · < ak . In other
words, it means the set { a1 , a2 , . . . , ak } when a1 < a2 < · · · < ak , and otherwise
is meaningless. Thus, for example, {2 < 5 < 8} means the set {2, 5, 8}, whereas
{5 < 2 < 8} is meaningless.

Thus, Proposition 1.4.11 can be restated as follows: Each finite set of integers can
be uniquely expressed in the form {s1 < s2 < · · · < sk }. Of course, the k here is the
size of this set.
The following proposition is a variant of Proposition 1.4.11 for the case when |S|
is known:

Proposition 1.4.13. Let m ∈ N. Let S be an m-element set of integers. Then, there

exists a unique m-tuple (s1 , s2 , . . . , sm ) of integers satisfying {s1 , s2 , . . . , sm } = S
and s1 < s2 < · · · < sm .

The subtle difference between Proposition 1.4.11 and Proposition 1.4.13 is that
the size of the tuple in Proposition 1.4.13 is specified in advance (viz., it must be
m = |S|), whereas Proposition 1.4.13 is a statement about tuples of all possible
sizes.
The overly skeptical reader may want to solve the following exercise:

75 We use the word “tuple” as a synonym for “finite list”.

76 In [Grinbe15, §2.6], a tuple (s1 , s2 , . . . , sk ) of integers satisfying {s1 , s2 , . . . , sk } = S and s1 < s2 <
· · · < sk is called an increasing list of S. Thus, our Proposition 1.4.11 is simply saying that any
finite set S of integers has a unique increasing list.
Enumerative Combinatorics: class notes page 109

Exercise 1.4.2. Derive Proposition 1.4.13 from Proposition 1.4.11.

Now, we can give a bijective proof of Proposition 1.4.10:

Proof of Proposition 1.4.10 (sketched). We have n + 1 − k ∈ N (since k ≤ n + 1).
Hence, the set {0, 1, . . . , n − k } is an (n + 1 − k)-element set. Theorem 1.3.12 (ap-
plied to n + 1 − k and {0, 1, . . . , n − k } instead of n and S) thus shows that

n+1−k
 
(# of k-element subsets of {0, 1, . . . , n − k}) = . (80)
k

We shall put this equality to use by finding a bijection between

{k-element lacunar subsets of [n]} and {k-element subsets of {0, 1, . . . , n − k}}.
Proposition 1.4.13 (applied to m = k) shows that if S is any k-element set of
integers, then there exists a unique k-tuple (s1 , s2 , . . . , sk ) of integers satisfying
{s1 , s2 , . . . , sk } = S and s1 < s2 < · · · < sk . In other words, if S is any k-element set
of integers, then S can be uniquely expressed in the form {s1 < s2 < s3 < · · · < sk }.
Hence, if S is any k-element lacunar subset of [n], then S can be uniquely ex-
pressed in the form {s1 < s2 < s3 < · · · < sk }. Moreover, the integers s1 , s2 , . . . , sk
in this expression must satisfy not only si < si+1 for each i ∈ [k − 1], but also the
stronger inequality si + 1 < si+1 for each i ∈ [k − 1] (since otherwise, we would
have si + 1 = si+1 for some i ∈ [k − 1], and thus the subset S would contain the
two consecutive integers si and si+1 , which would contradict its lacunarity). Hence,
they must satisfy s1 − 1 < s2 − 2 < s3 − 3 < · · · < sk − k (because the inequality
si + 1 < si+1 can be equivalently rewritten as si − i < si+1 − (i + 1), and these in-
equalities can be spliced together to form the chain s1 − 1 < s2 − 2 < s3 − 3 < · · · <
sk − k). Furthermore, they must satisfy s1 − 1 ≥ 0 (since s1 ≥ 1) and sk − k ≤ n − k
(since sk ≤ n), and therefore the set {s1 − 1 < s2 − 2 < s3 − 3 < · · · < sk − k } is a
well-defined k-element subset of {0, 1, . . . , n − k }.
Thus, we can define a map

A : {k-element lacunar subsets of [n]} → {k-element subsets of {0, 1, . . . , n − k}} ,

{ s1 < s2 < s3 < · · · < s k } 7 → { s1 − 1 < s2 − 2 < s3 − 3 < · · · < s k − k } .

Conversely, it is easy to see that we can define a map

B : {k-element subsets of {0, 1, . . . , n − k }} → {k-element lacunar subsets of [n]} ,

{ t1 < t2 < t3 < · · · < t k } 7 → { t1 + 1 < t2 + 2 < t3 + 3 < · · · < t k + k } .

It is fairly obvious that the maps A and B are mutually inverse. Hence, the map
A is invertible, i.e., is a bijection. Thus, the bijection principle yields

|{k-element lacunar subsets of [n]}|

= |{k-element subsets of {0, 1, . . . , n − k}}| .
Enumerative Combinatorics: class notes page 110

In other words,
(# of k-element lacunar subsets of [n])
= (# of k-element subsets of {0, 1, . . . , n − k})
n+1−k
 
= (by (80)) .
k
This proves Proposition 1.4.10.
(The above proof of Proposition 1.4.10 can also be rewritten in terms of domino
tilings instead of lacunar subsets, since we know a bijection between the former
and the latter from our second proof of Proposition 1.4.9.)
By combining Proposition 1.4.9 and Proposition 1.4.10, we can now obtain a
bijective proof of Proposition 1.3.32:
Proof of Proposition 1.3.32. Let m = n − 1. Then, m ∈ {−1, 0, 1, . . .} (since n ∈ N =
{0, 1, 2, . . .}) and m + 1 = n. Hence, it is easy to see that |[m]| ≤ n 77 .
If B is any subset of [m], then Theorem 1.4.7 (b) (applied to A = [m]) yields
| B| ≤ |[m]| ≤ n and therefore | B| ∈ {0, 1, . . . , n}. Hence, the sum rule yields78
(# of lacunar subsets of [m])
= ∑ (# of lacunar subsets B of [m] satisfying | B| = k)
k∈{0,1,...,n}
| {z }
=(# of k-element lacunar subsets of [ m ])
m+1−k
 
=
k
(by Proposition 1.4.10, applied to m instead of n)
n  n 
m+1−k m+1−k n−k
   
= ∑ k
= ∑ k
= ∑ k
k∈{0,1,...,n} k =0 k =0

(since m + 1 = n). Comparing this with

(# of lacunar subsets of [m])
= f m +2 (by Proposition 1.4.9, applied to m instead of n)
 

= f n +1 since m + 2 = (m + 1) +1 = n + 1 ,
| {z }
=n
we obtain
n
n−k n−0 n−1 n−2 n−n
         
f n +1 = ∑ k
=
0
+
1
+
2
+···+
n
.
k =0
This proves Proposition 1.3.32.
77 Proof.If m ∈ N, then |[m]| = m = n − 1 ≤ n. Hence, if m ∈ N, then we are done. Thus, we WLOG
assume that m ∈/ N. Combining this with m ∈ {−1, 0, 1, . . .}, we obtain m ∈ {−1, 0, 1, . . .} \ N =
{−1}, so that m = −1. Thus, m + 1 = 0, so that n = m + 1 = 0. But m = −1 and thus [m] = ∅
and therefore |[m]| = |∅| = 0 = n (since n = 0). Hence, |[m]| ≤ n, qed.
78 Again, the sum rule that we are using here is Theorem 1.2.5.
Enumerative Combinatorics: class notes page 111

This is not the only known proof of Proposition 1.3.32. There is also a (more
straightforward) proof by strong induction on n using Theorem 1.3.8. (See [Vorobi02,
§1.15] for the details of this proof, or find it yourself.)

1.4.6. Counting subsets with a odd and b even elements

Question 1.4.3 is resolved. Here is another one, to illustrate a principle we shall
use more often later: Given n, a, b ∈ N, how many subsets of [n] have exactly a
even elements and exactly b odd elements? Here is the answer in the case when n
is even:

Proposition 1.4.14. Let n ∈ N be even. Let a ∈ N and b ∈ N. Then,

(# of subsets of [n] that contain a even elements and b odd elements)

  
n/2 n/2
= .
a b

(Here, “a even elements” means “exactly a even elements”, and “b odd elements”
means “exactly b odd elements”.)

The following proof is essentially copied from [18s-hw1s, Exercise 6 (a)]:

Proof of Proposition 1.4.14. Informally, our argument is as follows: Let E be the set
of all even elements of [n]; let O be the set of all odd elements of [n]. We want
to find the # of subsets of [n] that contain a even elements and b odd elements.
In order to construct such a subset, we can choose its a even elements and its b
odd elements separately. The a even elements must be chosen from the  set  E; thus,
| E|
choosing them means choosing an a-element subset of E. There are ways to
a
make this choice (by Theorem 1.3.12, applied to E, | E| and a instead of S, n and
k). The b odd elements must be chosen from the  setO; thus, choosing them means
|O|
choosing a b-element subset of O. There are ways to make this choice (by
b
Theorem 1.3.12, applied to O,  |O| and 
b instead of S, n and k). Hence, altogether,
| E| |O|
our choices can be made in many ways; thus,
a b

(# of subsets of [n] that contain a even elements and b odd elements)

  
| E| |O|
= . (81)
a b

But n is even, and thus

E = {2, 4, 6, . . . , n} and O = {1, 3, 5, . . . , n − 1} .

Enumerative Combinatorics: class notes page 112

Hence, each of the two sets E and O has exactly n/2 elements. In other words,
| E| = n/2 and |O| = n/2. Thus, the equality (81) rewrites as
(# of subsets of [n] that contain a even elements and b odd elements)
  
n/2 n/2
= .
a b
This proves Proposition 1.4.14.
Here is a more rigorous way to write up the same proof (without speaking of
“choices”, which arguably is not a well-defined mathematical notion). Let E be the
set of all even elements of [n]; let O be the set of all odd elements of [n]. Let N be
the set of all subsets of [n] that contain exactly a even elements and exactly b odd
elements. Thus,
(# of subsets of [n] that contain a even elements and b odd elements)
= |N | . (82)
  
n/2 n/2
Hence, we need to show that |N | = .
a b
For any set X and every k ∈ N, we let Pk ( X ) denote the set of all k-element
subsets of X. Now, the two maps
N → P a ( E) × Pb (O) ,
S 7→ (S ∩ E, S ∩ O)
and
P a ( E) × Pb (O) → N ,
(U, V ) 7→ U ∪ V
are well-defined and mutually inverse79 . Hence, these two maps are bijections.
79 Insteadof proving this in detail (which is straightforward), let me unfold the notations and
explain what the two maps do:
• The first map sends any S ∈ N (that is, any subset S of [n] that contains a even elements
and b odd elements) to the pair (S ∩ E, S ∩ O), which consists of
– the set S ∩ E (this is the set of all even elements of S; it is an a-element set, because
S contains exactly a even elements) and
– the set S ∩ O (this is the set of all odd elements of S; it is a b-element set).
In other words, the first map splits any S ∈ N into its set of even elements and its set of
odd elements.
• The second map sends any pair (U, V ) ∈ P a ( E) × Pb (O) to the union U ∪ V ∈ N . In
other words, the second map takes a pair (U, V ) consisting of an a-element set of even
numbers and a b-element set of odd numbers, and combines them to the set U ∪ V (which
belongs to N , because it has a even elements and b odd elements).
From this point of view, it should be clear that the two maps are well-defined and mutually
inverse.
Enumerative Combinatorics: class notes page 113

Thus, there exists a bijection N → P a ( E) × Pb (O). Therefore, the bijection princi-

ple shows that
|N | = |P a ( E) × Pb (O)| = |P a ( E)| · |Pb (O)| (83)
(by the product rule, i.e., by (5), applied to X = P a ( E) and Y = Pb (O)).
But every finite set S and every k ∈ N satisfy
Pk (S) = {k-element subsets of S} (by the definition of Pk (S))
and thus
|Pk (S)| = (# of k-element subsets of S)
 
|S|
= (84)
k
(by Theorem 1.3.12, applied to |S| instead of n).
Now, (83) becomes
  
| E| |O|
|N | = |P a ( E)| · |Pb (O)| = .
| {z } | {z } a b
| E| |O|
= =
a b
(by (84)) (by (84))

Now, recall that n is even. Thus, there are exactly n/2 even numbers in the set
[n] (namely, 2, 4, 6, . . . , n). In other words, the set of all even elements of [n] has size
n/2. In other words, the set E has size n/2 (since E is the set of all even elements
of [n]). In other words, | E| = n/2. Similarly, |O| = n/2. Now,
     
| E| |O| n/2 n/2
|N | = =
a b a b
(since | E| = n/2 and |O| = n/2). Hence, (82) becomes
(# of subsets of [n] that contain a even elements and b odd elements)
  
n/2 n/2
= |N | = .
a b
This proves Proposition 1.4.14.
The relation between the above informal proof and its formalized version is
worth pointing out explicitly: Informally, we argued that choosing a subset with a
even and b odd elements is tantamount to independently choosing a elements from
E and b elements from O. We formalized this argument as a bijection from N to
P a ( E) × Pb (O), and then used the product rule along with the bijection principle
to conclude that |N | = |P a ( E)| · |Pb (O)|. Thus, “independent choices” in infor-
mal arguments correspond to an application of the product rule and the bijection
principle when formalized.
The analogue of Proposition 1.4.14 in the case of odd n can be proven in the same
way:
Enumerative Combinatorics: class notes page 114

Proposition 1.4.15. Let n ∈ N be odd. Let a ∈ N and b ∈ N. Then,

(# of subsets of [n] that contain a even elements and b odd elements)

(n − 1) /2 (n + 1) /2
  
= .
a b

We leave the proof to the reader.

Class of 2019-10-09

1.4.7. The addition formula for Fibonacci numbers

Let us see some of these tactics being used again in proving an identity for Fi-
bonacci numbers:

Theorem 1.4.16. Let m, n ∈ N. Then, the Fibonacci sequence satisfies f m+n+1 =

f m f n + f m +1 f n +1 .

This is not a-priori a combinatorial result; it is just an algebraic identity for the
elements of a recursively defined sequence. Consequently, we should expect it to
have a purely algebraic proof. And this is indeed the case: Theorem 1.4.16 can eas-
ily be proven by induction on n. For such proofs, see [18s-mt1s, solution to Exercise
3 (e)] or (in a slightly more general setting) [Grinbe15, proof of Theorem 2.26 (a)]80 .
We shall, instead, derive Theorem 1.4.16 from the combinatorial interpretation of
the Fibonacci numbers.
To simplify this proof, we shall front-load a little argument into a separate propo-
sition (Proposition 1.4.18). We begin with a fairly standard piece of notation:

Definition 1.4.17. Let a and b be two integers. Then, the subset { a, a + 1, . . . , b}

of Z will be denoted by [ a, b]. (Recall that this subset is understood to be ∅ when
b < a.)
The subset [ a, b] is called the integer interval from a to b. (It is not an interval in
the sense of real analysis, since it only contains integers; but it is the combinato-
rial analogue of an interval.)

For example, using this notation, we have [k ] = [1, k] for each k ∈ Z.

Proposition 1.4.18. Let n ∈ {−1, 0, 1, . . .} and a ∈ Z. Then,

(# of lacunar subsets of [ a + 1, a + n]) = f n+2 .

80 To recover Theorem 1.4.16 from [Grinbe15, Theorem 2.26 (a)], set a = 1 and b = 1. Then,
the sequence ( x0 , x1 , x2 , . . .) in [Grinbe15, Theorem 2.26 (a)] becomes the Fibonacci sequence
( f 0 , f 1 , f 2 , . . . ).
Enumerative Combinatorics: class notes page 115

Proof of Proposition 1.4.18. Here is an outline: The integer interval [ a + 1, a + n] is

just the integer interval [n] shifted by a. Hence, the lacunar subsets of the former
interval are in 1-to-1 correspondence with the lacunar subsets of the latter interval.
Hence, the number of lacunar subsets of the former interval equals the number of
lacunar subsets of the latter interval. Thus,

(# of lacunar subsets of [ a + 1, a + n]) = (# of lacunar subsets of [n]) = f n+2

(by Proposition 1.4.9). This proves Proposition 1.4.18.

Formally speaking, the argument we have just made is a bijective proof, which can be
made rigorous as follows: For each subset S of [n], we let S + a denote the set {s + a | s ∈ S}.
This set S + a is clearly a subset of [ a + 1, a + n]. (Informally speaking, S + a is just the set
S shifted by a units on the number line.) Conversely, for each subset T of [ a + 1, a + n],
we let T − a denote the set {t − a | t ∈ T }. 81 This set T − a is clearly a subset of n .
[ ]
(Informally speaking, T − a is just the set T shifted by − a units on the number line.) Now,
it is easy to see that the maps

{lacunar subsets of [n]} → {lacunar subsets of [ a + 1, a + n]} ,

S 7→ S + a

and

{lacunar subsets of [ a + 1, a + n]} → {lacunar subsets of [n]} ,

T 7→ T − a

are well-defined82 and mutually inverse. Hence, they are bijections. Thus, the bijection
principle yields

|{lacunar subsets of [ a + 1, a + n]}| = |{lacunar subsets of [n]}| .

In other words,

(# of lacunar subsets of [ a + 1, a + n]) = (# of lacunar subsets of [n]) = f n+2

(by Proposition 1.4.9). This proves Proposition 1.4.18.

We now come to our combinatorial proof of Theorem 1.4.16:

Proof of Theorem 1.4.16 (sketched). If m = 0, then the claim of Theorem 1.4.16 boils
down to the equality f n+1 = f 0 f n + f 1 f n+1 , which is obvious (since f 0 = 0 and
f 1 = 1). Hence, for the rest of this proof, we WLOG assume that m 6= 0. For a
similar reason, we WLOG assume that n 6= 0. From m 6= 0, we obtain m ≥ 1
(since m ∈ N). Hence, m − 2 ≥ 1 − 2 = −1, so that m − 2 ∈ {−1, 0, 1, . . .}.
81 Minor annoying caveat: Some people use the minus sign − for set-theoretic difference (so they
write P − Q for P \ Q). This is not what we mean when we write T − a.
82 This means, in particular, that if a subset S of [ n ] is lacunar, then so is S + a. This is all straight-

forward.
Enumerative Combinatorics: class notes page 116

Similarly, n − 2 ∈ {−1, 0, 1, . . .}. Furthermore, |{z} n −1 ≥ 1 + 1 − 1 = 1, so

m + |{z}
≥1 ≥1
that m + n − 1 ∈ N ⊆ {−1, 0, 1, . . .}.
Now, let us count the lacunar subsets of [m + n − 1] in two different ways:
First way: Proposition 1.4.9 (applied to m + n − 1 instead of n) yields

(# of lacunar subsets of [m + n − 1]) = f (m+n−1)+2 = f m+n+1 . (85)

Second way: We shall call a subset of [m + n − 1]

• red if it contains m, and

• green if it does not contain m.

Thus, each subset of [m + n − 1] is either red or green (but not both at the same
time). Thus, by the sum rule, we have

(# of lacunar subsets of [m + n − 1])

= (# of lacunar red subsets of [m + n − 1])
+ (# of lacunar green subsets of [m + n − 1]) . (86)

We shall now compute the two addends on the right hand side.
A lacunar green subset S of [m + n − 1] does not contain m, and thus is the
union of its two subsets {s ∈ S | s < m} and {s ∈ S | s > m}. The former of
these two subsets is a lacunar subset of [m − 1], while the latter is a lacunar sub-
set of [m + 1, m + n − 1]. Hence, a lacunar green subset of [m + n − 1] is just the
union of a lacunar subset of [m − 1] with a lacunar subset of [m + 1, m + n − 1].
Conversely, any such union must be a lacunar green subset of [m + n − 1] (since an
element of [m − 1] and an element of [m + 1, m + n − 1] cannot be consecutive83 ).
Summarizing this discussion, we conclude that the map

{lacunar green subsets of [m + n − 1]}

→ {lacunar subsets of [m − 1]} × {lacunar subsets of [m + 1, m + n − 1]}

that sends any lacunar green subset S of [m + n − 1] to the pair

({s ∈ S | s < m} , {s ∈ S | s > m})

is well-defined and is a bijection (and the inverse of this map simply sends each
pair ( P, Q) to P ∪ Q). Thus, the bijection principle yields

|{lacunar green subsets of [m + n − 1]}|

= |{lacunar subsets of [m − 1]} × {lacunar subsets of [m + 1, m + n − 1]}|
= |{lacunar subsets of [m − 1]}| · |{lacunar subsets of [m + 1, m + n − 1]}|
83 because m separates them
Enumerative Combinatorics: class notes page 117

(by the product rule). In other words,

(# of lacunar green subsets of [m + n − 1])

= (# of lacunar subsets of [m − 1]) · (# of lacunar subsets of [m + 1, m + n − 1])
| {z } | {z }
= f (m−1)+2 = f (n−1)+2
(by Proposition 1.4.9, (by Proposition 1.4.18,
applied to m−1 instead of n) applied to m and n−1 instead of a and n)
= f (m−1)+2 · f (n−1)+2 = f m+1 f n+1 . (87)
| {z } | {z }
= f m +1 = f n +1

Now, let us take a look at lacunar red subsets.

A lacunar red subset S of [m + n − 1] contains m, and thus contains neither
m − 1 nor m + 1 (since it is lacunar); hence, it is the union of its two subsets
{s ∈ S | s < m − 1} and {s ∈ S | s > m + 1} and the one-element set {m}. The
former of these two subsets is a lacunar subset of [m − 2], while the latter is a lacu-
nar subset of [m + 2, m + n − 1]. Hence, a lacunar red subset of [m + n − 1] is just
the union of the following three sets: a lacunar subset of [m − 2], a lacunar subset of
[m + 2, m + n − 1], and the one-element set {m}. Conversely, any such union must
be a lacunar red subset of [m + n − 1] (since an element of [m − 2] and an element
of [m + 2, m + n − 1] cannot be consecutive, and furthermore no such element can
be consecutive to m). Summarizing this discussion, we conclude that the map

{lacunar red subsets of [m + n − 1]}

→ {lacunar subsets of [m − 2]} × {lacunar subsets of [m + 2, m + n − 1]}
that sends any lacunar red subset S of [m + n − 1] to the pair

({s ∈ S | s < m − 1} , {s ∈ S | s > m + 1})

is well-defined and is a bijection (and the inverse of this map simply sends each
pair ( P, Q) to P ∪ Q ∪ {m}). Thus, the bijection principle yields

|{lacunar red subsets of [m + n − 1]}|

= |{lacunar subsets of [m − 2]} × {lacunar subsets of [m + 2, m + n − 1]}|
= |{lacunar subsets of [m − 2]}| · |{lacunar subsets of [m + 2, m + n − 1]}|
(by the product rule). In other words,

(# of lacunar red subsets of [m + n − 1])

= (# of lacunar subsets of [m − 2]) · (# of lacunar subsets of [m + 2, m + n − 1])
| {z } | {z }
= f (m−2)+2 = f (n−2)+2
(by Proposition 1.4.9, (by Proposition 1.4.18,
applied to m−2 instead of n) applied to m+1 and n−2 instead of a and n)
= f (m−2)+2 · f (n−2)+2 = f m f n . (88)
| {z } | {z }
= fm = fn
Enumerative Combinatorics: class notes page 118

Hence, (86) becomes

(# of lacunar subsets of [m + n − 1])

= (# of lacunar red subsets of [m + n − 1])
| {z }
= fm fn
(by (88))
+ (# of lacunar green subsets of [m + n − 1])
| {z }
= f m +1 f n +1
(by (87))
= f m f n + f m +1 f n +1 . (89)

Comparing (85) with (89), we obtain f m+n+1 = f m f n + f m+1 f n+1 . Thus, Theorem
1.4.16 is proven.
The proof we just gave is an example of a proof by double counting: We have
counted the lacunar subsets of [m + n − 1] (that is, computed their number) in two
different ways. In the first way, we obtained (85); in the second, we found (89).
Comparing the results, we then obtained the equality f m+n+1 = f m f n + f m+1 f n+1 ,
which we had set out to prove.
Such proofs often look like magic, but they are not always hard to find: If you
want to prove an identity of the form A = B by double counting, all you have to do
is come up with a counting problem that can be solved in two ways, one of which
gives A as a result while the other gives B. The specifics of A and B often give
away how the problem could look like. For example, because the identity we were
proving was f m+n+1 = f m f n + f m+1 f n+1 , it stood to reason that the counting prob-
lem should be “how many lacunar subsets does [m + n − 1] have?”. This way, it is
immediately clear that one answer is f m+n+1 (by Proposition 1.4.9); it remained to
find a second way to solve the counting problem and get f m f n + f m+1 f n+1 instead.
The form of this expression suggests using the sum rule, so that the lacunar subsets
of [m + n − 1] had to somehow be subdivided into two categories (the “red” and
the “green” ones in the above proof) numbering f m f n and f m+1 f n+1 , respectively.
Furthermore, each of these categories should have a product-like structure (i.e., a
lacunar red subset should fall apart into two independent parts, and likewise for a
lacunar green subset), so that the products f m f n and f m+1 f n+1 could be explained
by the product rule. With these heuristics, it is not hard to come up with the exact
argument given above. This kind of reverse-engineering strategy does not always
work (it is an art, not a science, and it relies heavily on your preknowledge), but
when it does, the result is usually worth it.
Our proof of Proposition 1.3.32 was also a proof by double counting.
Benjamin and Quinn have devoted a whole book [BenQui03] to proofs by double
counting; it has many more examples. In particular, they give a proof of Theorem
1.4.16 (“Identity 3” in their book) which uses domino tilings, but essentially is the
same as our proof.84
84 Recall that domino tilings and lacunar subsets are more or less interchangeable, as our Second
Enumerative Combinatorics: class notes page 119

1.4.8. More subset counting

Let us explore some more classes of subsets, each time asking ourselves have many
there are. We begin with the following:

Exercise 1.4.3. A set S of integers is said to be self-counting if the size |S| is an

element of S. (For example, {1, 3, 5} is self-counting, since |{1, 3, 5}| = 3 ∈
{1, 3, 5}; but {1, 2, 5} is not self-counting.)
Let n be a positive integer.
(a) For each k ∈ [n], find the # of self-counting k-element subsets of [n].
(b) Find the # of all self-counting subsets of [n].

This exercise is essentially [17f-hw1s, Exercise 8].

Solution to Exercise 1.4.3 (sketched). (a) Fix k ∈ [n]. For any k-element subset S of [n], we
have the following chain of logical equivalences:

(S is self-counting)
⇐⇒ (the size |S| is an element of S) (by the definition of “self-counting”)
⇐⇒ (|S| ∈ S)
⇐⇒ (k ∈ S) (since |S| = k (because S is a k-element set))
⇐⇒ (S contains k) .
Thus, the self-counting k-element subsets of [n] are precisely the k-element subsets of [n]
that contain k. Hence,

(# of self-counting k-element subsets of [n])

= (# of k-element subsets of [n] that contain k) . (90)

But the maps

{k-element subsets of [n] that contain k} → {(k − 1) -element subsets of [n] \ {k}} ,
S 7→ S \ {k}

and

{(k − 1) -element subsets of [n] \ {k}} → {k-element subsets of [n] that contain k} ,
T 7→ T ∪ {k}

are well-defined85 and mutually inverse86 , and thus are bijections. Hence, the bijection
proof of Proposition 1.4.9 has shown.
85 This
means the following:
• If S is a k-element subset of [n] that contains k, then S \ {k} is a (k − 1)-element subset of
[ n ] \ { k }.
• If T is a (k − 1)-element subset of [n] \ {k}, then T ∪ {k} is a k-element subset of [n] that
contains k.
Checking this is straightforward; you can do it in your head, but don’t forget to do this!
86 For this, you need to show that
Enumerative Combinatorics: class notes page 120

principle yields

|{k-element subsets of [n] that contain k}|

= |{(k − 1) -element subsets of [n] \ {k}}|
= (# of (k − 1) -element subsets of [n] \ {k}) . (91)

But [n] is an n-element set; hence, [n] \ {k } is an (n − 1)-element set (since k ∈ [n]). There-
fore, n − 1 ∈ N, and furthermore, Theorem 1.3.12 (applied to n − 1, k − 1 and [n] \ {k }
instead of n, k and S) shows that
n−1
 
= (# of (k − 1) -element subsets of [n] \ {k}) .
k−1
Comparing this with (91), we obtain
n−1
 
|{k-element subsets of [n] that contain k}| = .
k−1
Now, (90) becomes

(# of self-counting k-element subsets of [n])

= (# of k-element subsets of [n] that contain k)
n−1
 
= |{k-element subsets of [n] that contain k}| = . (92)
k−1
This solves Exercise 1.4.3 (a).
(b) Let S be a self-counting subset of [n]. Then, |S| ∈ S (since S is self-counting), and
thus the set S has at least one element (namely, |S|). Thus, |S| ≥ 1. But Theorem 1.4.7 (b)
(applied to A = [n] and B = S) yields |S| ≤ |[n]| = n. Hence, |S| ∈ [n] (since |S| is an
integer satisfying |S| ≥ 1 and |S| ≤ n).
Now, forget that we fixed S. We thus have shown that each self-counting subset S of [n]
satisfies |S| ∈ [n]. Hence, by the sum rule, we have

(# of self-counting subsets of [n])

n−1
 
= ∑ (# of self-counting subsets S of [n] such that |S| = k) = ∑
k−1
k∈[n]
| {z } k∈[n]
=(# of self-counting
 k-element
 subsets of [n])
n−1
=
k−1
(by (92))
!
n n
n−1
 
=∑ since the summation sign ∑ means the same as ∑
k =1
k−1 k ∈[n] k =1
n −1 
n−1

= ∑ (here, we have substituted k for k − 1 in the sum) .
k =0
k

• If S is a k-element subset of [n] that contains k, then (S \ {k }) ∪ {k} = S.

• If T is a (k − 1)-element subset of [n] \ {k}, then ( T ∪ {k }) \ {k } = T.
This is again entirely straightforward.
Enumerative Combinatorics: class notes page 121

But n − 1 ∈ N (since n is apositive integer). Thus, Corollary 1.3.27 (applied to n − 1

n −1 n − 1
instead of n) yields ∑ = 2n−1 . Hence,
k =0 k
n −1 
n−1

(# of self-counting subsets of [n]) = ∑ = 2n −1 .
k =0
k
This solves Exercise 1.4.3 (b).

Our next exercise is a slight variation on this one, in which we count subsets of
[n] whose size is not just some element but in fact the smallest one:
Exercise 1.4.4. A set S of integers is said to be self-starting if the size |S| is
the smallest element of S. (For example, the set {3, 4, 6} is self-starting, since
|{3, 4, 6}| = 3 is its smallest element; but {1, 3, 6} and {1, 2, 6} are not self-
starting.)
Let n be a positive integer.
(a) For each k ∈ [n], find the # of self-starting k-element subsets of [n].
(b) Find the # of all self-starting subsets of [n].
This exercise is essentially [18s-hw1s, Exercise 7]. Note that self-starting sets are
called “maximal sets” in [Chu19].
Solution to Exercise 1.4.4 (sketched). Recall the notation from Definition 1.4.17; in particular,
if k ∈ Z, then [k + 1, n] will mean the integer interval {k + 1, k + 2, . . . , n}.
(a) Fix k ∈ [n]. For any k-element subset S of [n], we have the following chain of logical
equivalences:
(S is self-starting)
⇐⇒ (|S| is the smallest element of S) (by the definition of “self-starting”)
⇐⇒ (k is the smallest element of S)
(since |S| = k (because S is a k-element set))
⇐⇒ (k ∈ S, but all elements of S other than k are larger than k)
⇐⇒ (k ∈ S, but all elements of S other than k belong to [k + 1, n])
 
because an element of S is larger than k if and only if
it belongs to [k + 1, n] (since it belongs to [n] )
⇐⇒ (k ∈ S, but all elements of S \ {k} belong to [k + 1, n])
 
since the elements of S other than k
are precisely the elements of S \ {k }
⇐⇒ (k ∈ S and S \ {k} ⊆ [k + 1, n])
(since the statement “all elements of S \ {k } belong to [k + 1, n]” means the same as “S \
{k} ⊆ [k + 1, n]”). Thus, the self-starting k-element subsets of [n] are precisely the k-element
subsets of [n] that contain k and satisfy S \ {k } ⊆ [k + 1, n]. Hence, the maps
{self-starting k-element subsets of [n]} → {(k − 1) -element subsets of [k + 1, n]} ,
S 7→ S \ {k}
Enumerative Combinatorics: class notes page 122

and

{(k − 1) -element subsets of [k + 1, n]} → {self-starting k-element subsets of [n]} ,

T 7→ T ∪ {k}

are well-defined87 and mutually inverse, and thus are bijections. Hence, the bijection prin-
ciple yields

|{self-starting k-element subsets of [n]}|

= |{(k − 1) -element subsets of [k + 1, n]}|
= (# of (k − 1) -element subsets of [k + 1, n]) . (93)

But [k + 1, n] = {k + 1, k + 2, . . . , n} is an (n − k )-element set (since k ∈ [n] yields k ≤ n).

Therefore, n − k ∈ N, and furthermore, Theorem 1.3.12 (applied to n − k, k − 1 and [k + 1, n]
instead of n, k and S) shows that

n−k
 
= (# of (k − 1) -element subsets of [k + 1, n]) .
k−1

Comparing this with (93), we obtain

n−k
 
|{self-starting k-element subsets of [n]}| = .
k−1

In other words,

n−k
 
(# of self-starting k-element subsets of [n]) = . (94)
k−1

This solves Exercise 1.4.4 (a).

(b) Let S be a self-starting subset of [n]. Then, |S| is the smallest element of S (since S is
self-starting), and thus the set S has at least one element (namely, |S|). Thus, |S| ≥ 1. But
Theorem 1.4.7 (b) (applied to A = [n] and B = S) yields |S| ≤ |[n]| = n. Hence, |S| ∈ [n]
(since |S| is an integer satisfying |S| ≥ 1 and |S| ≤ n).
Now, forget that we fixed S. We thus have shown that each self-starting subset S of [n]

87 Asin the solution to Exercise 1.4.3, all these claims need to be thoroughly checked, but this can
be quickly done in your head.
Enumerative Combinatorics: class notes page 123

satisfies |S| ∈ [n]. Hence, by the sum rule, we have

(# of self-starting subsets of [n])

= ∑ (# of self-starting subsets S of [n] such that |S| = k)
k ∈[n]
| {z }
=(# of self-starting k-element subsets of [ n ])
n−k
 
=
k−1
(by (94))
n 
n−k ( n − 1) − ( k − 1) ( n − 1) − ( k − 1)
    
= ∑ = ∑ =∑
k ∈[n]
k−1 k ∈[n]
k−1 k =1
k−1
| {z }
( n − 1) − ( k − 1)
 
=
k−1
(since n−k =(n−1)−(k −1))
!
n
since the summation sign ∑ means the same as ∑
k ∈[n] k =1
n −1 
( n − 1) − k

= ∑ k
(95)
k =0

(here, we have substituted k for k − 1 in the sum).

But n − 1 ∈ N (since n is a positive integer). Thus, the first equality sign of Proposition
1.3.32 (applied to n − 1 instead of n) yields
n −1 
( n − 1) − k

f (n−1)+1 = ∑ k
.
k =0

Comparing this with (95), we obtain

(# of self-starting subsets of [n]) = f (n−1)+1 = f n . (96)

This solves Exercise 1.4.4 (b).

The answers we obtained for this exercise might make you wonder whether there is a di-
rect connection between self-starting subsets and lacunar subsets – after all, their numbers
are both given by Fibonacci numbers! More precisely, a comparison of (96) with Proposition
1.4.9 may suggest that there might be a bijection

{self-starting subsets of [n]} → {lacunar subsets of [n − 2]}

(since both of these sets have size f n ); furthermore, a comparison of (94) with Proposition
1.4.10 may suggest that there might be a bijection

{k-element self-starting subsets of [n]} → {(k − 1) -element lacunar subsets of [n − 2]}

for each k ∈ [n]. And indeed, such bijections exist, and both of them are given by the same
formula:

{ k < i 1 < i 2 < · · · < i k −1 } 7 → { i 1 − k < i 2 − ( k − 1 ) < i 3 − ( k − 2 ) < · · · < i k −1 − 2 } .

| {z }
={i j −(k− j+1) | j∈[k−1]}
Enumerative Combinatorics: class notes page 124

(We leave it to the reader to verify that this defines a valid bijection.)
We can mutate Exercise 1.4.4 into yet another similar-looking counting problem
by replacing “smallest element” by “largest element”:

Exercise 1.4.5. A set S of integers is said to be self-ending if the size |S| is

the largest element of S. (For example, the set {1, 2, 3} is self-ending, since
|{1, 2, 3}| = 3 is its largest element; but {1, 3, 6} and {1, 2, 6} are not self-ending.)
Let n be a positive integer.
(a) For each k ∈ [n], find the # of self-ending k-element subsets of [n].
(b) Find the # of all self-ending subsets of [n].

This exercise is essentially [18f-hw1s, Exercise 6 (a)]. It looks analogous to Exer-

cise 1.4.4, but is its answer also analogous to the answer of the latter exercise?
Solution to Exercise 1.4.5 (sketched). (a) Let k ∈ [n]. We claim the following:

Claim 1: The only self-ending k-element subset of [n] is [k ].

[Proof of Claim 1: The set [k ] = {1, 2, . . . , k } has size k, but its largest element is also
k. Thus, [k ] is self-ending. Consequently, [k ] is a self-ending k-element subset of [n]. It
remains to show that [k ] is the only self-ending k-element subset of [n].
Let S be a self-ending k-element subset of [n]. Thus, |S| = k (since S is a k-element set).
But |S| is the largest element of S (since S is self-ending). In other words, k is the largest
element of S (since |S| = k). Hence, all elements of S are ≤ k and thus belong to the set [k ].
In other words, S is a subset of [k ]. Therefore, Theorem 1.4.7 (c) (applied to A = [k ] and
B = S) shows that S = [k ] (since |S| = k = |[k ]|).
Now, forget that we fixed S. We thus have shown that each self-ending k-element subset
S of [n] satisfies S = [k ]. In other words, each self-ending k-element subset S of [n] must
equal [k ]. Hence, [k ] is the only self-ending k-element subset of [n] (since we already know
that [k ] is such a subset). This proves Claim 1.]
Claim 1 obviously yields that

(# of self-ending k-element subsets of [n]) = 1. (97)

This solves Exercise 1.4.5 (a).
(b) It is easy to see that each self-ending subset S of [n] satisfies |S| ∈ [n] 88 . Hence, by
the sum rule, we have

(# of self-ending subsets of [n])

= ∑ (# of self-ending subsets S of [n] such that |S| = k)
k ∈[n]
| {z }
=(# of self-ending k-element subsets of [n])
=1
(by (97))

= ∑ 1 = |[n]| · 1 = |[n]| = n.
k ∈[n]

This solves Exercise 1.4.5 (b).

88 Indeed,you can prove this in the same way as we showed the analogous claim about self-starting
subsets in our solution to Exercise 1.4.4 (b).
Enumerative Combinatorics: class notes page 125

The moral of this story is: similar-looking counting problems don’t always have
similar answers. Counting problems can vary wildly in difficulty and method of
solution. Eventually, we will see some where no good formula for the answer is
known!
Have you been experimenting in SageMath along with the above? In that case, you
have probably defined functions to check whether a set of integers is self-counting, resp.
self-starting, resp. self-ending. Here is one way to do so:
def is_self_counting ( S ) :
# Check whether a set ‘S ‘ of integers is self - counting .
return S . cardinality () in S
# Note : " S . cardinality () " is the size of ‘S ‘.

def is_self_starting ( S ) :
# Check whether a set ‘S ‘ of integers is self - starting .
c = S . cardinality ()
return c in S and all ( c < = i for i in S )
# This is checking that the size ‘c ‘ of ‘S ‘ is in ‘S ‘
# and is smaller or equal to all elements of ‘S ‘.

def is_self_ending ( S ) :
# Check whether a set ‘S ‘ of integers is self - ending .
c = S . cardinality ()
return c in S and all ( c > = i for i in S )

With these functions, you can ask SageMath to compute (for example) the # of self-starting
3-element subsets of [8] as follows:
sum ( 1 for S in Subsets (8 , 3 ) if is_self_starting ( S ) )

(Note that Subsets(8, 3) is a shorthand for Subsets({1, 2, 3, 4, 5, 6, 7, 8}, 3) that

SageMath understands.)

Exercise 1.4.6. A set S of integers is said to be O<E<O<E<... (this is an adjective)

if it can be written in the form S = {s1 , s2 , . . . , sk } where

• s1 < s2 < · · · < s k ;

• the integer si is even whenever i is even;

• the integer si is odd whenever i is odd.

(For example, {1, 4, 5, 8, 11} is an O<E<O<E<... set, while {2, 3} and {1, 4, 6}
are not. Note that k is allowed to be 0, whence ∅ is an O<E<O<E<... set.)
For each n ∈ N, we let a (n) denote the number of all O<E<O<E<... subsets
of [n], and let b (n) denote the number of all O<E<O<E<... subsets of [n] that
contain n.
(a) Show that a (n) = a (n − 1) + b (n) for each n > 0.
n
(b) Show that a (n) = 1 + ∑ b (k ) for each n ∈ N.
k =0
Enumerative Combinatorics: class notes page 126

(c) Show that b (n) = ∑ b (k ) + [n is odd] for each n ∈ N. (Here, we

k∈{0,1,...,n−1};
k≡n−1 mod 2
are using the Iverson bracket notation, defined in Definition 1.3.15.)
n −1
(d) Show that b (n) + b (n − 1) = 1 + ∑ b (k ) for each n > 0.
k =0
n −2
(e) Show that b (n) = 1 + ∑ b (k ) for each n > 0.
k =0
(f) Show that b (n) = a (n − 2) for each n ≥ 2.
(g) Show that a (n) = f n+2 for each n ∈ N.

Exercise 1.4.7. For each n ∈ N, we let c (n) denote the number of all subsets of
[n] that are simultaneously lacunar and O<E<O<E<....
Prove that c (n) = c (n − 2) + c (n − 3) for all n ≥ 3.

Remark 1.4.19. The sequence (c (0) , c (1) , c (2) , c (3) , . . .) from Exercise 1.4.7 is
the Padovan sequence (starting with 1, 2, 2, 3, 4, 5, 7, 9, 12, 16, 21, 28, 37, 49).

Exercise 1.4.8. A set S of integers is said to be 2-lacunar if every i ∈ S satisfies

i+1 ∈ / S and i + 2 ∈ / S. (That is, any two distinct elements of S are at least
a distance of 3 apart on the real axis.) For example, {1, 5, 8} is 2-lacunar, but
{1, 5, 7} and {1, 5, 6} are not.
For any n ∈ N, we let h (n) denote the number of all 2-lacunar subsets of [n].
(a) Prove that h (n) = h (n − 1) + h (n − 3
) for each n ≥ 3.
n + 2 − 2k
(b) Prove that h (n) = ∑ for each n ∈ N.
k∈N; k
2k≤n+2

1.4.9. Counting subsets containing a given subset

For a given k ∈ N, how many k-element subsets of a given set N contain a given
subset D as a subset? The following proposition gives an answer to this question:

Proposition 1.4.20. Let n ∈ N, d ∈ N and k ∈ R. Let N be an n-element set. Let

D be a d-element subset of N. Then,

n−d
 
(# of k-element subsets A of N satisfying D ⊆ A) = .
k−d

Example 1.4.21. Let N = {1, 2, 3, 4} and D = {1, 2}. Then, the subsets A of N
satisfying D ⊆ A are {1, 2}, {1, 2, 3}, {1, 2, 4} and {1, 2, 3, 4}. Thus, 1 of them has
2 elements; 2 have 3 elements; and 1 has 4 elements. This agrees with Proposition
1.4.20 (for n = 4 and d = 2).
Enumerative Combinatorics: class notes page 127

Proof of Proposition 1.4.20 (informal version). As usual, let me first show the informal
idea, and then its formalization.
Informally, how can we construct a k-element subset A of N satisfying D ⊆ A ?
We need to choose k elements of N to go into A; but we cannot choose them freely,
because we must ensure that D ⊆ A. Thus, the d elements of D must necessarily
go into A; the only freedom that we have is to decide which of the remaining n − d
elements of N will go into A. More precisely, we need to choose k − d many of
these n − d elements, because the total size of A has to be k (and d of its elements
have already been decided). Thus, our decision boils down to choosing a (k − d)-
element subset of the (n − d)-element set N \ D. According to Theorem  1.3.12
n−d

(applied to n − d, k − d and N \ D instead of n, k and S), there are many
k−d
such subsets. Thus, the total # of k-element subsets A of N satisfying D ⊆ A is
n−d
 
.
k−d
To formalize this argument, we have to replace our “boils down” handwaving with a
bijection. Here is how the result looks like:
Proof of Proposition 1.4.20 (formal version). We have | D | = d (since D is a d-element set) and
| N | = n (since N is an n-element set). Also, D is a subset of N. Thus, Theorem 1.4.7 (a)
(applied to A = N and B = D) yields | N \ D | = | N | − | D | = n − d. Hence, N \ D is an
|{z} |{z}
=n =d
(n − d)-element set. Thus, Theorem 1.3.12 (applied to n − d, k − d and N \ D instead of n,
k and S) yields
n−d
 
= (# of (k − d) -element subsets of N \ D ) . (98)
k−d
If A is a k-element subset of N satisfying D ⊆ A, then A \ D is a (k − d)-element subset
of N \ D 89 . Hence, the map
{k-element subsets A of N satisfying D ⊆ A} → {(k − d) -element subsets of N \ D } ,
A 7→ A \ D
is well-defined.
On the other hand, if B is a (k − d)-element subset of N \ D, then B ∪ D is a k-element
subset A of N that satisfies D ⊆ A 90 . Hence, the map
{(k − d) -element subsets of N \ D } → {k-element subsets A of N satisfying D ⊆ A} ,
B 7→ B ∪ D
89 Proof.Let A be a k-element subset of N satisfying D ⊆ A. Then, | A| = k (since A is a k-element
set). But D is a subset of A (since D ⊆ A). Hence, Theorem 1.4.7 (a) (applied to B = D) yields
| A \ D | = | A| − | D | = k − d. In other words, A \ D is a (k − d)-element set. Hence, A \ D is a
|{z} |{z}
=k =d
(k − d)-element subset of N \ D (because |{z}
A \ D ⊆ N \ D).
⊆N
90 Proof.Let B be a (k − d)-element subset of N \ D. Thus, B ⊆ N \ D and | B| = k − d (since B is a
( k − d )-element set). From B ⊆ N \ D, we conclude that the sets B and D are disjoint. Thus, (4)
Enumerative Combinatorics: class notes page 128

is well-defined.
It is easy to see that these two maps are mutually inverse91 . Thus, they are bijections.
Hence, the bijection principle yields

(# of k-element subsets A of N satisfying D ⊆ A)

n−d
 
= (# of (k − d) -element subsets of N \ D ) =
k−d

(by (98)). This proves Proposition 1.4.20.

1.5. Counting tuples and maps

1.5.1. Tuples
Before we continue with counting problems, let us recall the notion of the Cartesian
product of several sets:

Definition 1.5.1. Let A1 , A2 , . . . , An be any n sets. Then, the Cartesian prod-

uct A1 × A2 × · · · × An of these n sets is defined to be the set of all n-tuples
( a1 , a2 , . . . , an ) for which ai ∈ Ai for all i ∈ [n]. In other words,

A1 × A2 × · · · × A n
= {( a1 , a2 , . . . , an ) | ai ∈ Ai for all i ∈ [n]}
= {( a1 , a2 , . . . , an ) | a1 ∈ A1 and a2 ∈ A2 and · · · and an ∈ An } .

Applying Definition 1.5.1 to n = 2, we obtain the Cartesian product A1 × A2 of

two sets A1 and A2 ; it consists of pairs of an element of A1 and an element of A2 .
Applying Definition 1.5.1 to n = 1, we obtain the Cartesian product A1 of a single
set A1 ; this is the set of all 1-tuples ( a1 ) of a single element a1 ∈ A1 . The notation
hides a minor discrepancy: Strictly speaking, a 1-tuple ( a1 ) is not the same as its
underlying element a1 , and thus the Cartesian product of the single set A1 is not
the same as the set A1 itself, despite looking precisely the same in our notation.

(applied to X = B and Y = D) yields | B ∪ D | = | B| + | D | = (k − d) + d = k. Hence, B ∪ D is a

|{z} |{z}
=k−d =d
k-element set. Moreover, combining B ⊆ N \ D ⊆ N with D ⊆ N, we obtain B ∪ D ⊆ N. Hence,
B ∪ D is a k-element subset of N (since B ∪ D is a k-element set). Thus, B ∪ D is a k-element
subset A of N that satisfies D ⊆ A (since D ⊆ B ∪ D).
91 Indeed, this relies on the following two observations:

• If A is any k-element subset of N satisfying D ⊆ A, then ( A \ D ) ∪ D = A. (This follows

from D ⊆ A.)
• If B is any (k − d)-element subset of N \ D, then ( B ∪ D ) \ D = B. (This follows from
B ⊆ N \ D.)
Enumerative Combinatorics: class notes page 129

Fortunately, the sets are “as good as equal”: The map

(the original set A1 ) → (the Cartesian product of the single set A1 ) ,

a1 7 → ( a1 )

(sending each element a1 of A1 to the 1-tuple ( a1 )) is a bijection. Thus, the two sets
denoted A1 have the same size, at least, and it is rare that anything bad comes out
of equating them.
Applying Definition 1.5.1 to n = 0, we obtain the Cartesian product of 0 sets; this
is called the empty Cartesian product, and consists of all 0-tuples. There is only one
0-tuple, namely the empty list (); thus, this Cartesian product is the 1-element set
{()}.
The Cartesian product of sets is not quite associative! For example, if A, B
and C are three sets, then the three Cartesian products A × B × C, ( A × B) × C
and A × ( B × C ) are not literally the same. The first of them consists of triples92
( a, b, c); the second consists of nested pairs (( a, b) , c); the third consists of nested
pairs ( a, (b, c)). These are different things, and should not be equated93 . Again,
however, there are bijections between these products: The maps

A × B × C → ( A × B) × C,
( a, b, c) 7→ (( a, b) , c)

and

A × B × C → A × (B × C) ,
( a, b, c) 7→ ( a, (b, c))

are bijections. More generally:

Proposition 1.5.2. Let A1 , A2 , . . . , An be any n sets.

(a) If n > 0, then the map

A 1 × A 2 × · · · × A n → ( A 1 × A 2 × · · · × A n −1 ) × A n ,
( a1 , a2 , . . . , an ) 7→ (( a1 , a2 , . . . , an−1 ) , an )

is a bijection.
(b) If k ∈ {0, 1, . . . , n} is arbitrary, then the map

A 1 × A 2 × · · · × A n → ( A 1 × A 2 × · · · × A k ) × ( A k +1 × A k +2 × · · · × A n ) ,
( a1 , a2 , . . . , an ) 7→ (( a1 , a2 , . . . , ak ) , ( ak+1 , ak+2 , . . . , an ))

is a bijection.
92 Recall
that the word “triple” means a 3-tuple. Tuples are always ordered by definition.
93 Forexample, the first entry of the triple ( a, b, c) is a, while the first entry of the pair (( a, b) , c) is
the pair ( a, b).
Enumerative Combinatorics: class notes page 130

I consider Proposition 1.5.2 obvious (or, more precisely, to be part of an introduc-

tory course in axiomatic set theory that I am not giving here). From an algorithmic
viewpoint, the bijection in Proposition 1.5.2 (b) splits a list into two sublists (the
one formed by its first k entries, and the one formed by its remaining entries).
Using Proposition 1.5.2, we can generalize our product rule for two sets (Theo-
rem 1.1.4) to n sets:

Theorem 1.5.3 (The product rule for n sets). Let A1 , A2 , . . . , An be any n finite
sets. Then, A1 × A2 × · · · × An is a finite set with size

| A1 × A2 × · · · × A n | = | A1 | · | A2 | · · · · · | A n | .

The intuition behind Theorem 1.5.3 is the following: In order to choose an n-

tuple ( a1 , a2 , . . . , an ) ∈ A1 × A2 × · · · × An , it suffices to choose an element a1 of A1 ,
an element a2 of A2 , and so on (a total of n choices). These n choices are mutually
independent (e.g., the value of a1 we choose does not constrain us in our choice of
a2 ), and thus the total number of options should be the product of the total numbers
of options for each of the n choices (which numbers are | A1 | , | A2 | , . . . , | An |). In this
form, Theorem 1.5.3 is usually considered obvious and called the “rule of product”
or “multiplication principle”. We nevertheless give a rigorous proof in order to get
some feeling for Cartesian products:
Proof of Theorem 1.5.3. We will prove Theorem 1.5.3 by induction on n:
Induction base: We have

A1 × A2 × · · · × A0 = (empty Cartesian product) = {()}

(this is a one-element set) and therefore

| A1 × A2 × · · · × A0 | = |{()}| = 1.
Comparing this with

| A1 | · | A2 | · · · · · | A0 | = (empty product) = 1,
we obtain | A1 × A2 × · · · × A0 | = | A1 | · | A2 | · · · · · | A0 |. In other words, Theorem 1.5.3 is
true for n = 0. This completes the induction base.
Induction step: Let m ∈ N. Assume that Theorem 1.5.3 holds for n = m. We must prove
that Theorem 1.5.3 holds for n = m + 1.
Let A1 , A2 , . . . , Am+1 be any m + 1 finite sets. Recall that we assumed that Theorem 1.5.3
holds for n = m. Hence,

| A1 × A2 × · · · × A m | = | A1 | · | A2 | · · · · · | A m | .
But Proposition 1.5.2 (a) (applied to n = m + 1) yields that the map
 
A1 × A2 × · · · × Am+1 → A1 × A2 × · · · × A(m+1)−1 × Am+1 ,
  
( a1 , a2 , . . . , am+1 ) 7→ a1 , a2 , . . . , a(m+1)−1 , am+1
Enumerative Combinatorics: class notes page 131

is a bijection. Hence, the bijection principle shows that

| A 1 × A 2 × · · · × A m +1 |
 
= A1 × A2 × · · · × A(m+1)−1 × Am+1

= A1 × A2 × · · · × A(m+1)−1 · | Am+1 |
| {z }
=| A1 × A2 ×···× Am |
=| A1 |·| A2 |·····| Am |
 
by (5), applied to X = A1 × A2 × · · · × A(m+1)−1 and Y = Am+1
= (| A1 | · | A2 | · · · · · | Am |) · | Am+1 |
= | A 1 | · | A 2 | · · · · · | A m +1 | .

In particular, this shows that the set A1 × A2 × · · · × Am+1 is finite.

Forget that we fixed A1 , A2 , . . . , Am+1 . We thus have proven that if A1 , A2 , . . . , Am+1 are
any m + 1 finite sets, then A1 × A2 × · · · × Am+1 is a finite set with size

| A 1 × A 2 × · · · × A m +1 | = | A 1 | · | A 2 | · · · · · | A m +1 | .

In other words, Theorem 1.5.3 holds for n = m + 1. This completes the induction step, and
with it the proof of Theorem 1.5.3.

A particular case of a Cartesian product is obtained when all the “factors” are
equal:

Definition 1.5.4. Let A be a set, and let n ∈ N. Then, the n-th (Cartesian) power
An of A is defined to be the Cartesian product |A × A ×
{z· · · × A}.
n times

This Cartesian power An is often denoted by A×n as well.

For example, if A is any set, then A0 = {()}, whereas A1 is “more or less the
same as” A (as explained before, not exactly the same, but there is an obvious
bijection).

Corollary 1.5.5. Let A be a finite set. Let n ∈ N. Then, An is a finite set with size
| An | = | A|n .

Proof of Corollary 1.5.5. Definition 1.5.4 yields An = |A × A ×

{z· · · × A}. But Theorem 1.5.3
n times

{z· · · × A} is a finite set with size |A × A ×

(applied to Ai = A) shows that |A × A × {z· · · × A} =
n times n times
| A| · | A| · · · · · | A|. In view of |A × A × n · · · · · | A| = | A|n , we can
{z· · · × A} = A and ||A| · | A|{z
| {z } }
n times n times n times
rewrite this as follows: The set An is a finite set with size | An | = | A|n . This proves Corollary
1.5.5.
Enumerative Combinatorics: class notes page 132

1.5.2. Counting maps

Let us now use these results to count maps between two sets. First, we introduce a
notation for them:

Definition 1.5.6. Let A and B be two finite sets. Then, B A shall mean the set of
all maps from A to B.

This notation B A is somewhat counterintuitive (who in their right mind would

put the target B first?), but it is explained by the following property:

Theorem 1.5.7. Let A and B be two finite sets. Then, the set B A is finite, and its
size is
B A = | B || A| .

Example 1.5.8. Theorem 1.5.7 (applied to A = [2] and B = [3]) says that the set
[3][2] is finite, and that its size is [3][2] = |[3]||[2]| = 32 = 9. In other words, there
are exactly 9 maps from [2] to [3]. Here are these 9 maps (written in two-line
notation, as explained in Definition 1.2.11):
     
1 2 1 2 1 2
, ,
1 1 1 2 1 3
     
1 2 1 2 1 2
, ,
2 1 2 2 2 3
     
1 2 1 2 1 2
, , .
3 1 3 2 3 3

Proof of Theorem 1.5.7. Let ( a1 , a2 , . . . , ak ) be a list of all elements of A in some order

(with no repetitions). Thus, a1 , a2 , . . . , ak are distinct and satisfy A = { a1 , a2 , . . . , ak }.
Hence, | A| = k.
Recall that B A is the set of all maps from A to B. In other words, B A is the
set of all functions from A to B. (The nouns “map” and “function” are synonyms
in mathematics. I usually prefer “map” for its shortness, but here, I am using
“function” in order to separate the maps from A to B (which I call “functions”)
from maps between B A and Bk (which I shall consider soon, and will call “maps”).)
Consider the map

Φ : B A → Bk ,
f 7→ ( f ( a1 ) , f ( a2 ) , . . . , f ( ak )) .

This map Φ sends each function f : A → B to its list of values at the points94
a1 , a2 , . . . , ak . This list of values clearly determines f uniquely (since A = { a1 , a2 , . . . , ak },
94 I am just using “point” as just a suggestive way of saying “element of A” here.
Enumerative Combinatorics: class notes page 133

so that each value of f appears somewhere in the list95 ); thus, the map Φ is injective.
Furthermore, for each k-tuple (b1 , b2 , . . . , bk ) ∈ Bk , there exists a function f : A → B
such that ( f ( a1 ) , f ( a2 ) , . . . , f ( ak )) = (b1 , b2 , . . . , bk ) (indeed, this latter function f
is given by  
a1 a2 · · · a k
f =
b1 b2 · · · bk
in two-line notation). Hence, the map Φ is surjective. Thus, we know that the map
Φ is both injective and surjective. Therefore, Φ is bijective, i.e., is a bijection. Hence,
the bijection principle yields B A = Bk .
Corollary 1.5.5 (applied to B and k instead of A and n) shows that Bk is a finite
set with size Bk = | B|k . Hence, B A = Bk = | B|k = | B|| A| (since k = | A|). Thus,
in particular, B A is a finite set. This proves Theorem 1.5.7.
In one of the next chapters, we will count certain restricted types of maps (injec-
tive maps, surjective maps, etc.).

1.5.3. Applications
Theorem 1.5.7 and Theorem 1.5.3 can be used not only in the kind of problems
that are visibly concerned with subsets or tuples. In fact, often one can discover a
bijection between the objects one wants to count and some kind of maps or tuples.
The most famous example is a combinatorial proof of Theorem 1.4.1:
Third proof of Theorem 1.4.1. The set S is an n-element set. Let us denote its n ele-
ments by s1 , s2 , . . . , sn (in an arbitrary order, with no repetitions). Thus, s1 , s2 , . . . , sn
are n distinct elements of S and satisfy S = {s1 , s2 , . . . , sn }.
We will construct a bijection from the set {subsets of S} to {0, 1}n . This will then
let us apply Corollary 1.5.5 (and the bijection principle) and obtain |{subsets of S}| =
2n . So let us construct such a bijection:
We define the map A : {subsets of S} → {0, 1}n by setting96
A ( T ) = ([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]) for each subset T of S.
(This is well-defined, since each of the truth values [s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]
belongs to {0, 1}.) For example, if n = 5 and T = {s2 , s4 , s5 }, then A ( T ) =
(0, 1, 0, 1, 1). (Some call A ( T ) the indicator vector of T, since its i-th entry indicates
whether si belongs to T or not.)
In order to prove that A is a bijection, we shall next construct a map B in the
other direction, and then show that it is inverse to A.
We define the map B : {0, 1}n → {subsets of S} by setting
B ((i1 , i2 , . . . , in )) = {sk | k ∈ [n] and ik = 1}
for each (i1 , i2 , . . . , in ) ∈ {0, 1}n .
95 And we know where exactly it appears in the list, because a1 , a2 , . . . , ak are fixed.
96 We are using the Iverson bracket notation (defined in Definition 1.3.15).
Enumerative Combinatorics: class notes page 134

(Note that {sk | k ∈ [n] and ik = 1} is always a subset of S, since s1 , s2 , . . . , sn are

elements of S.)
We want to show that the maps A and B are mutually inverse. In order to do
this, we must show that A ◦ B = id and that B ◦ A = id.
Intuitively, this is fairly obvious: The map A encodes a subset of S as an n-tuple,
each of whose entries equals either 0 or 1 depending on whether the corresponding
element of S belongs to the subset or not.97 The map B, in turn, decodes such an
n-tuple back into a subset of S, by gathering the elements of S corresponding to
those positions at which the n-tuple has a 1. It is thus clear that the maps A and B
undo one another, i.e., they satisfy the two equalities A ◦ B = id and B ◦ A = id.
Here is a more formal way to prove these equalities:

• In order to show that A ◦ B = id, we fix some (i1 , i2 , . . . , in ) ∈ {0, 1}n . We shall show
that ( A ◦ B) ((i1 , i2 , . . . , in )) = (i1 , i2 , . . . , in ).
Set T = B ((i1 , i2 , . . . , in )). Thus, T = B ((i1 , i2 , . . . , in )) = {sk | k ∈ [n] and ik = 1}
(by the definition of B). Now,
 

( A ◦ B) ((i1 , i2 , . . . , in )) = A  B ((i1 , i2 , . . . , in )) = A ( T )

| {z }
=T
= ([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]) (99)
 
(by the definition of A). Now, we shall show that each j ∈ [n] satisfies s j ∈ T = i j .
Indeed, let j ∈ [n]. Thus, i j ∈ {0, 1} (since (i1 , i2 , . . . , in ) ∈ {0, 1}n ). Recall
 that

T = {sk | k ∈ [n] and ik = 1}. Hence,  if i j =
 1, then s j ∈ T and therefore s j ∈ T =
1 = i j . Therefore, we have proved s j ∈ T = i j in the case when i j = 1. On the
other hand, if i j 6= 1, then i j = 0 (since i j ∈ {0,1}) and s j ∈/ T 98 and therefore
 
s j ∈ T = 0 = i j . Thus, we have proved s j ∈ T = i j in the case when i j 6= 1. We
have now proved s j ∈ T = i j both in the case when i j = 1 and in  the case
 when
i j 6= 1. These two cases cover all possibilities; thus, we always have s j ∈ T = i j .
 
Forget that we fixed j. We thus have shown that s j ∈ T = i j for each j ∈ [n]. Hence,

([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]) = (i1 , i2 , . . . , in ) .

Hence, (99) becomes

( A ◦ B) ((i1 , i2 , . . . , in )) = ([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]) = (i1 , i2 , . . . , in )

= id ((i1 , i2 , . . . , in )) . (100)
97 The “corresponding element of S” to the k-th entry is understood to be sk . This correspon-
dence between the n positions in the n-tuple and the elements of S is fixed (since we have fixed
s1 , s2 , . . . , sn ).
98 Proof. Assume that i 6 = 1. We must prove that s ∈
j j / T.
Assume the contrary. Thus, s j ∈ T = {sk | k ∈ [n] and ik = 1}. In other words, there exists
some k ∈ [n] such that ik = 1 and s j = sk . Consider this k. From s j = sk , we obtain j = k (since
s1 , s2 , . . . , sn are distinct). Thus, i j = ik = 1. This contradicts i j 6= 1. This contradiction shows
that our assumption was false, qed.
Enumerative Combinatorics: class notes page 135

Now, forget that we fixed (i1 , i2 , . . . , in ). We thus have proven (100) for each (i1 , i2 , . . . , in ) ∈
{0, 1}n . Hence, A ◦ B = id.
• In order to prove that B ◦ A = id, we fix T ∈ {subsets of S}. We are going to show
that ( B ◦ A) ( T ) = T.
We know that T is a subset of S (since T ∈ {subsets of S}). In other words, T ⊆ S.
We have A ( T ) = ([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]) by the definition of A. Now,
 

( B ◦ A) ( T ) = B  A (T )
 

| {z }
=([s1 ∈ T ],[s2 ∈ T ],...,[sn ∈ T ])

= B (([s1 ∈ T ] , [s2 ∈ T ] , . . . , [sn ∈ T ]))

= {sk | k ∈ [n] and [sk ∈ T ] = 1} (101)

(by the definition of B).

From this, we can easily conclude that ( B ◦ A) ( T ) ⊆ T 99 and T ⊆ B ◦ A T
( )( )
100 . Combining these two relations, we obtain B ◦ A T = T. In other words,
( )( )
( B ◦ A) ( T ) = id ( T ) (since id ( T ) = T).
Now, forget that we fixed T. We thus have shown that ( B ◦ A) ( T ) = id ( T ) for each
T ∈ {subsets of S}. In other words, B ◦ A = id.

We now know that A ◦ B = id and B ◦ A = id. Combining these equalities,

we conclude that the maps A and B are mutually inverse. Hence, the map A is
invertible, i.e., bijective. In other words, A : {subsets of S} → {0, 1}n is a bijection.
Thus, Theorem 1.1.6 (applied to X = {subsets of S}, Y = {0, 1}n and f = A) yields

|{subsets of S}| = {0, 1}n = |{0, 1}|n

(by Corollary 1.5.5, applied to {0, 1} instead of A)
n
=2 (since |{0, 1}| = 2) .

Hence, 2n = |{subsets of S}| = (# of subsets of S). This proves Theorem 1.4.1 once
again.101
99 Proof. Let s ∈ ( B ◦ A) ( T ). Thus, s ∈ ( B ◦ A) ( T ) = {sk | k ∈ [n] and [sk ∈ T ] = 1}. In other
words, s = sk for some k ∈ [n] satisfying [sk ∈ T ] = 1. Consider this k. From [sk ∈ T ] = 1, we
conclude that sk ∈ T must be true (by the definition of the Iverson bracket notation). Hence,
s = sk ∈ T.
Forget that we fixed s. We thus have proved that s ∈ T for each s ∈ ( B ◦ A) ( T ). In other
words, ( B ◦ A) ( T ) ⊆ T.
100 Proof. Let s ∈ T. Then, s ∈ T ⊆ S = { s , s , . . . , s }. In other words, s = s for some j ∈ [ n ].
1 2  n j
Consider this j. Then, s j = s ∈ T. Hence, s j ∈ T = 1. Thus, j is a k ∈ [n] satisfying [sk ∈ T ] = 1
 
(because j ∈ [n] and s j ∈ T = 1). Hence, s j ∈ {sk | k ∈ [n] and [sk ∈ T ] = 1}. In view of
(101), this rewrites as s j ∈ ( B ◦ A) ( T ). Thus, s = s j ∈ ( B ◦ A) ( T ).
Forget that we fixed s. We thus have proved that s ∈ ( B ◦ A) ( T ) for each s ∈ T. In other
words, T ⊆ ( B ◦ A) ( T ).
101 This proof has been taken from [19f-hw0s, solution to Exercise 1 (a)] (but slightly adapted for the

fact that we are using the set S rather than [n] here).
Enumerative Combinatorics: class notes page 136

The key idea in the above proof of Theorem 1.4.1 was to encode any subset of
the n-element set S as an n-tuple (i1 , i2 , . . . , in ) ∈ {0, 1}n . Instead of using an n-
tuple, we could have just as well used a map from [n] to {0, 1}, since we learned to
translate between tuples and maps in the above proof of Theorem 1.5.7. The reader
can easily restate our above proof of Theorem 1.4.1 using maps instead of tuples.
Here are some other counting exercises that can easily be reduced to counting
tuples or maps:

Exercise 1.5.1. Let n ∈ N.

(a) Find the number of all triples ( A, B, C ) of subsets of [n] satisfying A ∪ B ∪
C = [n] and A ∩ B ∩ C = ∅.
(b) Find the number of all triples ( A, B, C ) of subsets of [n] satisfying B ∩ C =
C ∩ A = A ∩ B.
(c) Find the number of all triples ( A, B, C ) of subsets of [n] satisfying A ∩ B =
A ∩ C.
(d) Find the number of 4-tuples ( A, B, C, D ) of subsets of [n] satisfying A ∩ B =
C ∩ D.
(e) Find the number of 4-tuples ( A, B, C, D ) of subsets of [n] satisfying A ∩ B =
C ∪ D.

Exercise 1.5.2. A set S of integers is said to be shadowed if it has the following

property: Whenever an odd integer i belongs to S, the next integer i + 1 must
also belong to S. (For example, ∅, {2, 4} and {1, 2, 5, 6, 8} are shadowed, but
{1, 5, 6} is not, since 1 belongs to {1, 5, 6} but 2 does not.)
(a) Let n ∈ N be even. How many shadowed subsets of [n] exist?
(b) Let n ∈ N be odd. How many shadowed subsets of [n] exist?

Class of 2019-10-11

1.6. Interchange of summations

1.6.1. The finite Fubini principle
In Subsection 1.2.3, we have seen a few rules for manipulating finite sums. Let me
now show another: the (finite) Fubini principle. Let us first state it; we will see it in
action soon:

Theorem 1.6.1 ((Finite) Fubini principle). Let X and Y be two finite sets. Let
a( x,y) be a number for each pair ( x, y) ∈ X × Y. Then,

∑ ∑ a(x,y) = ∑ a( x,y) = ∑ ∑ a( x,y) . (102)

x ∈ X y ∈Y ( x,y)∈ X ×Y y ∈Y x ∈ X

Theorem 1.6.1 is called the finite Fubini principle or Fubini’s theorem for finite sums,
since it is a discrete analogue of Fubini’s principle for double integrals. A proof of
Enumerative Combinatorics: class notes page 137

Theorem 1.6.1 is outlined in [Grinbe15, §1.4.2] (and an alternative proof can easily
be obtained by induction on | X |); here I shall restrict myself to illustrating it on an
example:
Example 1.6.2. Let n ∈ N and m ∈ N. Let a( x,y) be a number for each pair
( x, y) ∈ [n] × [m]. Then, (102) (applied to X = [n] and Y = [m]) yields
n m m n
∑ ∑ a(x,y) = ∑ a( x,y) = ∑ ∑ a(x,y) . (103)
x =1 y =1 ( x,y)∈[n]×[m] y =1 x =1

This equality, when rewritten to avoid the use of summation signs, looks as
follows:
 
a(1,1) + a(1,2) + · · · + a(1,m)
 
+ a(2,1) + a(2,2) + · · · + a(2,m)
+···
 
+ a(n,1) + a(n,2) + · · · + a(n,m)
 
= a(1,1) + a(1,2) + · · · + a(n,m) this is the sum of all nm numbers a( x,y)
 
= a(1,1) + a(2,1) + · · · + a(n,1)
 
+ a(1,2) + a(2,2) + · · · + a(n,2)
+···
 
+ a(1,m) + a(2,m) + · · · + a(n,m) .

So this equality says that if we have a rectangular n × m-table of numbers:

a(1,1) a(1,2) ··· a(1,m)

a(2,1) a(2,2) ··· a(2,m)
.. .. ... .. ,
. . .
a(n,1) a(n,2) ··· a(n,m)

then the following three procedures all produce the same result:

(a) summing up the m numbers in each row of the table separately, and then
summing up these n “row tallies”;
(b) summing up all the nm numbers in the table;
(c) summing up the n numbers in each column of the table separately, and
then summing up these m “column tallies”.

This is, of course, exactly how you would expect sums to behave.
Enumerative Combinatorics: class notes page 138

I like to abbreviate the equality (102) as follows:

∑ ∑ = ∑ = ∑ ∑.
x ∈ X y ∈Y ( x,y)∈ X ×Y y ∈Y x ∈ X

This is an “equality between summation signs”; it should be understood as follows:

Every time you see an “ ∑ ∑ ” in an expression, you can replace it by a “ ∑ ”
x ∈ X y ∈Y ( x,y)∈ X ×Y
or by a “ ∑ ∑ ”, and similarly the other ways round. Thus, in particular, you can
y ∈Y x ∈ X
replace the sequence “ ∑ ∑ ” of two summation signs by the reverse sequence
x ∈ X y ∈Y
“ ∑ ∑ ” and vice versa.102 This is called interchange of summations or interchanging
y ∈Y x ∈ X
the order of summation.
As a first example of how (102) can be applied, let us solve a simple exercise:

Exercise 1.6.1. Let n ∈ N. Let S be an n-element set. Find the sum of the sizes
of all subsets of S; in other words, find the sum

∑ |T | .
T ⊆S

We shall solve this using the following simple fact:103

Proposition 1.6.3 (“Counting by roll call”). Let S be a finite set.

(a) If T is a subset of S, then

|T | = ∑ [s ∈ T ] .
s∈S

(b) For each s ∈ S, let A (s) be a logical statement (which can be either true
or false depending on s; for example, A (s) could be “s is even” if S is a set of
integers, or “s is empty” if S is a set of sets). Then,

(# of s ∈ S that satisfy A (s)) = ∑ [A (s)] .

s∈S

102 Ofcourse, you can do this only if X and Y are finite sets that have a well-defined meaning
independent of x and y. Thus, you cannot replace the sequence “ ∑ ∑ ” by “ ∑ ∑ ”: The
x ∈[n] y∈[ x ] y∈[ x ] x ∈[n]
latter sequence would be meaningless, since the “x” under the first summation sign now refers
to nothing. We will soon see how such double summations can be transformed.
103 For the meaning of the square brackets on the right hand side, see Definition 1.3.15.
Enumerative Combinatorics: class notes page 139

Example 1.6.4. (a) Applying Proposition 1.6.3 (a) to S = [5] and T = {1, 3, 5}, we
obtain
|{1, 3, 5}| = ∑ [s ∈ {1, 3, 5}] = 1 + 0 + 1 + 0 + 1.
s∈[5]

(b) Applying Proposition 1.6.3 (b) to S = [5] and A (s) = (“s is odd”), we
obtain

(# of s ∈ [5] that are odd) = ∑ [s is odd] = 1 + 0 + 1 + 0 + 1.

s∈[5]

Why am I calling this “counting by roll call”? Because Proposition 1.6.3 (a) for-
malizes a way to count the elements of the subset T by “calling the roll” of all
elements of S and adding a 1 every time an element of T is encountered. Proposi-
tion 1.6.3 (b) does the same thing, except we are no longer counting the elements
of T, but instead are counting the elements s of S that satisfy A (s). The two parts
of Proposition 1.6.3 are easily seen to be equivalent to each other, since any subset
T of S gives rise to a logical statement A (s) = (“s ∈ T”) for each s ∈ S, and con-
versely, any logical statement A (s) defined for each s ∈ S can be used to carve out
a subset {s ∈ S | A (s) holds} of S.
The formal proof is not much more complicated:
Proof of Proposition 1.6.3. (b) Let Q be the set of all s ∈ S that satisfy A (s). Then,

| Q| = (# of s ∈ S that satisfy A (s)) .

Furthermore, the elements s ∈ S for which A (s) is true are precisely the elements
of Q (because this is how Q is defined).
Now, apply the formula (28) to as = [A (s)]. Thus, we obtain

∑ [A (s)] = ∑ [A (s)] + ∑ [A (s)]

s∈S s∈S; s∈S;
| {z } | {z }
A(s) is true =1 A(s) is false =0
(since A(s) is true) (since A(s) is false)

= ∑ 1+ ∑ 0= ∑ 1= ∑1
s∈S; s∈S; s∈S; s∈ Q
A(s) is true A(s) is false A(s) is true
| {z }
=0
 
since the elements s ∈ S for which A (s) is true
are precisely the elements of Q
= | Q| (by (30), applied to Q instead of S)
= (# of s ∈ S that satisfy A (s)) .

This proves Proposition 1.6.3 (b).

Enumerative Combinatorics: class notes page 140

(a) Let T be a subset of S. Thus, S ∩ T = T. Applying Proposition 1.6.3 (b) to the

statements A (s) = (“s ∈ T”), we obtain
(# of s ∈ S that satisfy s ∈ T ) = ∑ [s ∈ T ] .
s∈S

Comparing this with

(# of s ∈ S that satisfy s ∈ T ) = {s ∈ S | s ∈ T } = | T | ,
| {z }
=S∩ T = T

we obtain | T | = ∑ [s ∈ T ]. This proves Proposition 1.6.3 (a).

s∈S

First solution to Exercise 1.6.1. Let P (S) denote the powerset of S. This is a finite
set, since S is finite.
We have |S| = n (since S is an n-element set). But

∑ |T | = ∑ ∑ [s ∈ T ] = ∑ ∑ [s ∈ T ] , (104)
T ⊆S T ⊆S s∈S s∈S T ⊆S
|{z}
= ∑ [s∈ T ]
s∈S
(by Proposition 1.6.3 (a))

where the last equality has been obtained by interchanging the two summation
signs (using the Fubini principle)104 . Hence,

∑ |T | = ∑∑ [s ∈ T ] = ∑ ∑ [x ∈ T] (105)
T ⊆S s∈S T ⊆S x ∈S T ⊆S

(here, we have renamed the summation index s as x in the outer sum).

Now, fix x ∈ S. We shall simplify the sum ∑ [ x ∈ T ]. Indeed, an application of
T ⊆S
Proposition 1.6.3 (b) yields

∑ [ x ∈ T ] = (# of subsets T of S that satisfy x ∈ T ) (106)

T ⊆S
104 In more detail: Theorem 1.6.1 (applied to X = P (S), Y = S and a( x,y) = [y ∈ x ]) yields

∑ ∑ [y ∈ x ] = ∑ [y ∈ x ] = ∑ ∑ [y ∈ x ] .
x ∈P (S) y∈S ( x,y)∈P (S)×S y∈S x ∈P (S)

Renaming the summation indices x and y as T and s (in order to match the notations in (104)),
we rewrite this as

∑ ∑ [s ∈ T ] = ∑ [s ∈ T ] = ∑ ∑ [s ∈ T ] .
T ∈P (S) s∈S ( T,s)∈P (S)×S s∈S T ∈P (S)

Since the summation sign ∑ is synonymous to ∑ , we can further rewrite this as follows:
T ∈P (S) T ⊆S

∑ ∑ [s ∈ T ] = ∑ [s ∈ T ] = ∑∑ [s ∈ T ] .
T ⊆S s∈S ( T,s)∈P (S)×S s∈S T ⊆S

This justifies the last equality sign in (104).

Enumerative Combinatorics: class notes page 141

105 . But there is a bijection

{subsets T of S that satisfy x ∈ T } → {subsets of S \ { x }} ,

R 7→ R \ { x }
106 . By the bijection principle, this entails that

|{subsets T of S that satisfy x ∈ T }| = |{subsets of S \ { x }}| .

In other words,

(# of subsets T of S that satisfy x ∈ T ) = (# of subsets of S \ { x }) .

Furthermore, S \ { x } is an (n − 1)-element set (since S is an n-element set, and

x ∈ S), and thus n − 1 = |S \ { x }| ∈ N; hence, Theorem 1.4.1 (applied to n − 1 and
S \ { x } instead of n and S) yields

(# of subsets of S \ { x }) = 2n−1 .

Thus, (106) becomes

∑ [ x ∈ T ] = (# of subsets T of S that satisfy x ∈ T )

T ⊆S
= (# of subsets of S \ { x }) = 2n−1 . (107)
105 In
more detail: We can apply Proposition 1.6.3 (b) to P (S) and (“x ∈ s”) instead of S and A (s).
This results in the equality

(# of s ∈ P (S) that satisfy x ∈ s) = ∑ [ x ∈ s] .

s∈P (S)

Renaming the “s” here as T, we can rewrite this as

(# of T ∈ P (S) that satisfy x ∈ T ) = ∑ [x ∈ T] .

T ∈P (S)

Since the summation sign ∑ is a synonym for ∑ , this further rewrites as

T ∈P (S) T ⊆S

(# of T ∈ P (S) that satisfy x ∈ T ) = ∑ [x ∈ T] .

T ⊆S

Hence,

∑ [ x ∈ T ] = (# of T ∈ P (S) that satisfy x ∈ T )

T ⊆S
= (# of subsets T of S that satisfy x ∈ T ) ,

because the T ∈ P (S) are precisely the subsets of S.

106 Indeed, this map is easily seen to be well-defined, and it has an inverse map which sends each
Q ∈ {subsets of S \ { x }} to Q ∪ { x } ∈ {subsets T of S that satisfy x ∈ T }. This is essentially the
same construction that we used to count red subsets in the proof of Theorem 1.3.12, except that
this time we are not fixing the size of our subsets but considering all subsets together.
Enumerative Combinatorics: class notes page 142

Now, forget that we fixed x. We thus have proved that (107) holds for each x ∈ S.
Hence, (105) becomes

∑ |T | = ∑ ∑ [x ∈ T] = ∑ 2n−1 = |{z}
| S | · 2n −1 = n · 2n −1 .
T ⊆S x ∈S T ⊆S s∈S
| {z } =n
=2n −1
(by (107))

This solves Exercise 1.6.1.

In the future, we will no longer explicitly cite the Fubini principle whenever we
interchange summation signs, let alone rename our summation indices to match
those in Theorem 1.6.1 and rewrite signs like ∑ in the standard form ∑ .
T ⊆S T ∈P (S)
We will also cut down on the details when invoking Proposition 1.6.3. Thus, the
solution we just gave will take the following short form:
First solution to Exercise 1.6.1 (short version). We have |S| = n (since S is an n-element
set). But

∑ |T | = ∑ ∑ [s ∈ T ] = ∑ ∑ [s ∈ T ]
T ⊆S T ⊆S s∈S s∈S T ⊆S
|{z}
= ∑ [s∈ T ] | {z }
s∈S =∑ ∑
(by Proposition 1.6.3 (a)) s∈S T ⊆S

= ∑ ∑ [x ∈ T] (108)
x ∈S T ⊆S

(here, we have renamed the summation index s as x in the outer sum).

Now, for each x ∈ S, we have

∑ [ x ∈ T ] = (# of subsets T of S that satisfy x ∈ T )

T ⊆S
(by an application of Proposition 1.6.3 (b))
= (# of subsets of S \ { x })
by the bijection principle, since there is a bijection
 

 from {subsets T of S that satisfy x ∈ T } 

 to {subsets of S \ { x }} 
that sends each R to R \ { x }
by Theorem 1.4.1, applied to n − 1
 
n −1
=2 . (109)
and S \ { x } instead of n and S
Hence, (108) becomes

∑ |T | = ∑ ∑ [x ∈ T] = ∑ 2n−1 = |{z}
| S | · 2n −1 = n · 2n −1 .
T ⊆S x ∈S T ⊆S s∈S
| {z } =n
=2n −1
(by (109))

This solves Exercise 1.6.1.

Enumerative Combinatorics: class notes page 143

Thus, interchanging summations (even if it requires strategically introducing ex-

tra summation signs) can lead to short and slick proofs.
Our solution to Exercise 1.6.1 resulted in the expression n · 2n−1 , which is rather
obviously the simplest possible answer (it cannot be simplified any further). Nev-
ertheless, it is worth seeing a different solution to Exercise 1.6.1, which results in
a different (more complicated) expression for the answer. This is not a bad thing,
because as a consequence we will immediately conclude the equality between the
two expressions (once again, a proof by double counting!).
Second solution to Exercise 1.6.1. Let P (S) denote the powerset of S. This is a finite
set, since S is finite. For each T ∈ P (S), we have | T | ∈ {0, 1, . . . , n} 107 . Hence,
we can define a map

f : P (S) → {0, 1, . . . , n} ,
T 7→ | T | .

Consider this map f . Now, (37) (applied to P (S), {0, 1, . . . , n} and |s| instead of S,
W and as ) yields
∑ |s| = ∑ ∑ |s| .
s∈P (S) w∈{0,1,...,n} s∈P (S);
f (s)=w

Renaming the summation index s as T everywhere in this equality, we can rewrite

this as
∑ |T | = ∑ ∑ |T | .
T ∈P (S) w∈{0,1,...,n} T ∈P (S);
f ( T )=w

In other words,
∑ |T | = ∑ ∑ |T |
T ⊆S w∈{0,1,...,n} T ⊆S;
f ( T )=w

(since “T ∈ P (S)” under a summation sign can always be abbreviated as “T ⊆ S”).

Renaming the summation index w as k on the right hand side, we can further
rewrite this as
∑ |T | = ∑ ∑ |T | . (110)
T ⊆S k∈{0,1,...,n} T ⊆S;
f ( T )=k

The left hand side of this equality is the sum we want to find; let us get a better
understanding of the right hand side.

107 Proof. Let T ∈ P (S). Thus, T is a subset of S (by the definition of P (S)). Hence, Theorem 1.4.7
(b) (applied to A = S and B = T) yields | T | ≤ |S| = n (since S is an n-element set). Hence,
| T | ∈ {0, 1, . . . , n} (since | T | is clearly a nonnegative integer).
Enumerative Combinatorics: class notes page 144

Let k ∈ {0, 1, . . . , n}. The definition of f shows that f ( T ) = | T | for each subset T
of S. Hence,

∑ |T | = ∑ |T | = ∑ k= ∑ k
T ⊆S; T ⊆S; T ⊆S; T ∈{k-element subsets of S}
|{z}
f ( T )=k | T |=k =k | T |=k
 
since the subsets T of S satisfying | T | = k
are precisely the k-element subsets of S
= |{k-element subsets of S}| ·k (by an application of (29))
| {z }
=(# of k-element
  subsets of S )
n
=
k
(by Theorem 1.3.12)
   
n n
= ·k = k . (111)
k k

Forget that we fixed k. We thus have proven (111) for each k ∈ {0, 1, . . . , n}. Hence,
(110) becomes
  n  
n n
∑ |T | = ∑ ∑ |T | = ∑ k k = ∑ k k .
T ⊆S k∈{0,1,...,n} T ⊆S; k∈{0,1,...,n} k =0
f ( T )=k
| {z }

n
=k
k
(by (111))

Thus we have solved Exercise 1.6.1 again.

The solution we just gave illustrates another method of dealing with sums:
namely, by splitting them into batches using (37). (We chose the batches in such a
way that all addends in any given batch are equal, which means that the sum of the
batch simplifies to a product: ∑ k = |{k-element subsets of S}| · k. This is not the
T ⊆S;
| T |=k
only strategy to split a sum, but it is one of the simplest strategies.) In the future,
we will not explicitly mention the formula (37) (let alone explain how exactly it is
being applied) when we perform such a splitting; we will simply say that we are
“splitting the sum ∑ | T | according to the value of | T |”. Thus, our Second solution
T ⊆S
to Exercise 1.6.1 takes the following short form:
Second solution to Exercise 1.6.1 (short version). If T is any subset of S, then | T | ∈
{0, 1, . . . , n} 108 . Hence, we can split the sum ∑ | T | according to the value of | T |
T ⊆S

108 becauseTheorem 1.4.7 (b) (applied to A = S and B = T) yields | T | ≤ |S| = n (since S is an

n-element set)
Enumerative Combinatorics: class notes page 145

as follows:
∑ |T | = ∑ ∑ |T | . (112)
T ⊆S k∈{0,1,...,n} T ⊆S;
| T |=k
But for each k ∈ {0, 1, . . . , n}, we have
   
n n
∑ |{z}| T | = ∑ k = (# of all T ⊆ S satisfying | T | = k) ·k =
k
·k = k
k
.
T ⊆S; T ⊆S;
| {z }
| T |=k =k | T |=k =(# of k-element
  subsets of S)
n
=
k
(by Theorem 1.3.12)

Hence, (112) becomes

  n  
n n
∑ |T | = ∑ ∑ |T | = ∑ k k = ∑ k k .
T ⊆S k∈{0,1,...,n} T ⊆S; k ∈{0,1,...,n} k =0
| T |=k
| {z }
n
=k
k
Thus we have solved Exercise 1.6.1 again.
Now, by comparing the results of the two solutions of Exercise 1.6.1, we get the
following identity for free:
Corollary 1.6.5. Let n ∈ N. Then,
n  
n
∑ k k = n · 2n −1 .
k =0

Proof of Corollary 1.6.5. Let S = [n]. Then, S is an n-element set.

In the first solution to Exercise 1.6.1, we have shown that

∑ | T | = n · 2n −1 .
T ⊆S

In the second solution to Exercise 1.6.1, we have shown that

n  
n
∑ |T | = ∑ k k .
T ⊆S k =0
 
n n
Comparing these two equalities, we obtain ∑ k = n · 2n−1 . This proves Corol-
k =0 k
lary 1.6.5.
There are easier ways to prove Corollary 1.6.5 as well; in particularly, it can be
proven purely algebraically using Proposition 1.3.36 and Corollary 1.3.27 (exer-
cise!).
Enumerative Combinatorics: class notes page 146

1.6.2. The Fubini principle with a predicate

Theorem 1.6.1 is a useful tool for interchanging summation signs that are indepen-
dent of one another – i.e., the indexing set of the inner summation sign must not
depend on the summation index of the other. Thus, for example, we can use The-
5 6 5 x
orem 1.6.1 to interchange the two summation signs in ∑ ∑ , but not in ∑ ∑ .
x =1 y =1 x =1 y =1
Nevertheless, in the latter situation, we would still like to bring the y “out”, i.e.,
5 x ? ?
to transform ∑ ∑ into ∑ ∑ for some values of the question marks. There is a
x =1 y =1 x =? y=?
Fubini-like rule for this, which we will see soon (Corollary 1.6.9). (The answer is:
5 x 5 5
∑ ∑ transforms into ∑ ∑ .)
x =1 y =1 y =1 x = y
First, let us state the most general version of this rule, which is a generalization
of Fubini’s principle to tables that have “gaps”:

Theorem 1.6.6 (Finite Fubini’s principle with a predicate). Let X and Y be two
finite sets. For each pair ( x, y) ∈ X × Y, let A ( x, y) be some statement (for
example, “x + y is even” or “x < y”, if X and Y are sets of numbers). For each
pair ( x, y) ∈ X × Y satisfying A ( x, y), let a( x,y) be a number. Then,

∑ ∑ a( x,y) = ∑ a( x,y) = ∑ ∑ a( x,y) . (113)

x ∈ X y∈Y; ( x,y)∈ X ×Y; y∈Y x ∈ X;
A( x,y) A( x,y) A( x,y)

(Here, the “A ( x, y)” under the summation signs means “A ( x, y) is true”.)

For example, if n, m ∈ N, and if a( x,y) is a number for each pair ( x, y) ∈

[n] × [m], then Theorem 1.6.6 (applied to X = [n] and Y = [m] and A ( x, y) =
(“x + y is even”)) shows that

∑ ∑ a( x,y) = ∑ a( x,y) = ∑ ∑ a( x,y) .

x ∈[n] y∈[m]; ( x,y)∈[n]×[m]; y∈[m] x ∈[n];
x +y is even x +y is even x +y is even

The idea behind Theorem 1.6.6 is the same as the one behind Theorem 1.6.1:
We have a table of numbers, and we can add up the numbers in this table in
three different ways (row by row, in a random order, or column by column). The
difference is that this time, the table can have empty cells (more precisely, only the
cells ( x, y) for which A ( x, y) is true have a number in them); of course, these empty
cells don’t contribute to the sums.109

109 We remark that Theorem 1.6.6 can easily be derived from Theorem 1.6.1. Indeed, if we define
the (hitherto undefined) number a( x,y) to be 0 whenever ( x, y) ∈ X × Y does not satisfy A ( x, y),
then we can apply Theorem 1.6.1, and the equality (102) we obtain quickly rewrites as (113).
Enumerative Combinatorics: class notes page 147

The formula (113) allows us to replace a sequence “ ∑ ∑ ” of summation

x ∈ X y∈Y;
A( x,y)
signs by “ ∑ ∑ ”. This is still considered to be an interchange of summations
y∈Y x ∈ X;
A( x,y)
– with the only caveat that the statement A ( x, y) has to stay put under the inner
sum, whatever this inner sum is.
By choosing appropriate statements A ( x, y) in Theorem 1.6.6, we can obtain
various templates for interchange of non-independent summation signs. Here is
one such template:

Corollary 1.6.7 (Triangular Fubini’s principle I). Let n ∈ N. For each pair ( x, y) ∈
[n] × [n] with x + y ≤ n, let a(x,y) be a number. Then,

n n− x n n−y
∑ ∑ a(x,y) = ∑ a( x,y) = ∑ ∑ a(x,y) .
x =1 y =1 ( x,y)∈[n]×[n]; y =1 x =1
x +y≤n

Example 1.6.8. If we set n = 4 in Corollary 1.6.7 and rewrite the result without
using summation signs, then we obtain
   
a(1,1) + a(1,2) + a(1,3) + a(2,1) + a(2,2) + a(3,1) + (empty sum)
 
= a(1,1) + a(1,2) + · · · + a(3,1) this is the sum of all the numbers a( x,y)
   
= a(1,1) + a(2,1) + a(3,1) + a(1,2) + a(2,2) + a(1,3) + (empty sum) .

In other words, if we are given a triangular table of numbers:

a(1,1) a(1,2) a(1,3)

a(2,1) a(2,2) ,
a(3,1)

then the following three procedures all produce the same result:

(a) summing up the numbers in each row of the table separately, and then
summing up these “row tallies”;

(b) summing up all the 6 numbers in the table;

(c) summing up the numbers in each column of the table separately, and then
summing up these “column tallies”.
Enumerative Combinatorics: class notes page 148

Proof of Corollary 1.6.7. Theorem 1.6.6 (applied to X = [n], Y = [n] and A ( x, y) = (“x + y ≤ n”))
yields
∑ ∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y) . (114)
x ∈[n] y∈[n]; ( x,y)∈[n]×[n]; y∈[n] x ∈[n];
x +y≤n x +y≤n x +y≤n

We shall now rewrite the two inner sums in this chain of equalities.
Let x ∈ [n]. Then, x ≥ 1 ≥ 0, so that n − x ≤ n. Thus, [n − x ] ⊆ [n], so that [n − x ] ∩ [n] =
[n − x ]. Hence, the elements y ∈ [n] that satisfy y ≤ n − x are precisely the elements of
[n − x ] (indeed, they are clearly the elements of [n − x ] ∩ [n]; but we just showed that
[n − x ] ∩ [n] = [n − x ]).
However, the statement “x + y ≤ n” for an integer y is equivalent to “y ≤ n − x”, and
thus can be replaced by the latter statement whenever it appears under a summation sign.
Thus,

∑ a(x,y) = ∑ a(x,y)
y∈[n]; y∈[n];
x +y≤n y≤n− x

since the elements y ∈ [n] that satisfy y ≤ n − x

 
= ∑ a(x,y)
are precisely the elements of [n − x ]
y∈[n− x ]
n− x
= ∑ a(x,y) . (115)
y =1

Forget that we fixed x. We thus have proved (115) for each x ∈ [n]. Likewise, we can
prove that
n−y
∑ a(x,y) = ∑ a(x,y) (116)
x ∈[n]; x =1
x +y≤n

for each y ∈ [n]. In light of the two equalities (115) and (116), we can rewrite the chain of
equalities (114) as follows:

n− x n−y
∑ ∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y) .
x ∈[n] y=1 ( x,y)∈[n]×[n]; y∈[n] x =1
x +y≤n

In other words,
n n− x n n−y
∑ ∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y)
x =1 y =1 ( x,y)∈[n]×[n]; y =1 x =1
x +y≤n
n
(since the summation sign ∑ is synonymous to ∑ , whereas the summation sign ∑ is
x ∈[n] x =1 y∈[n]
n
synonymous to ∑ ). This proves Corollary 1.6.7.
y =1

Let us apply Corollary 1.6.7 to solving an exercise (arguably, one that could be
more easily solved by other means, but this does not preclude it from being a useful
illustration):
Enumerative Combinatorics: class notes page 149

Exercise 1.6.2. Let n ∈ N. Prove that

n  
n+1
∑ k (n − k) = 3
.
k =1

Solution to Exercise 1.6.2. For each x ∈ [n], we have n − x ∈ N (since x ∈ [n] entails
x ≤ n and thus n − x ≥ 0), so that
n− x
∑ 1 = (n − x) · 1 = n − x. (117)
y =1

n
Renaming the index k as x in the sum ∑ k (n − k), we obtain
k =1

n n n n− x
∑ k (n − k) = ∑ x (|n {z
− x) = ∑ ∑1 x
k =1 x =1 x =1 y =1
}
n− x
=∑ 1
| {z }
y =1 n− x
= ∑ x ·1
(by (117)) y =1
(by (32))
n n− x n n− x
= x · 1 = ∑ ∑ x.
∑ ∑ |{z} (118)
x =1 y =1 = x x =1 y =1

But Corollary 1.6.7 (applied to a( x,y) = x) yields

n n− x n n−y
∑ ∑x= ∑ x= ∑ ∑ x. (119)
x =1 y =1 ( x,y)∈[n]×[n]; y =1 x =1
x +y≤n

Furthermore, each y ∈ [n] satisfies n − y ∈ N (since y ∈ [n] entails y ≤ n and thus

n − y ≥ 0), so that
n−y
(n − y) ((n − y) + 1)
∑ x = 1 + 2 + · · · + (n − y) = 2
x =1
(by Theorem 1.2.1, applied to n − y instead of n)
(n + 1 − y) (n − y) n+1−y
 
= = (120)
2 2

(since (46) (applied to n + 1 − y instead of n) yields

n+1−y (n + 1 − y) ((n + 1 − y) − 1) (n + 1 − y) (n − y)
 
= =
2 2 2
Enumerative Combinatorics: class notes page 150

). Hence, (118) becomes

n n n− x n n−y
∑ k (n − k) = ∑∑ x= ∑ ∑x (by (119))
k =1 x =1 y =1 y =1 x =1
 | {z } 
n+1−y
=
2
(by (120))
n n  
n+1−y
 
i
= ∑ 2
=∑
2
y =1 i =1
here, we have substituted i for n + 1 − y in the sum
 

(since the map [n] → [n] , y 7→ n − y + 1 is a bijection)

     
1 2 n
= + +···+
2 2 2
         
0 1 2 n 0
= + + +···+ −
2 2 2 2 2
| {z } |{z}
0 (0 − 1)
 
n+1
= =
2+1 2
(by Theorem 1.3.29, applied to k=2) (by (46))

0 (0 − 1)
     
n+1 n+1 n+1
= − = = .
2+1 | {z2 } 2+1 3
=0

This solves Exercise 1.6.2.

In the future, of course, we won’t be as detailed as in the solution we just gave;
instead of explicitly invoking Corollary 1.6.7, we will simply write the following:
n n− x n n−y
∑ ∑ x = ∑ ∑ x.
x =1 y =1 y =1 x =1
| {z }
n n−y
=∑ ∑
y =1 x =1

Here is another template for interchanging summation signs:

Corollary 1.6.9 (Triangular Fubini’s principle II). Let n ∈ N. For each pair
( x, y) ∈ [n] × [n] with x ≤ y, let a(x,y) be a number. Then,

n n n y
∑ ∑ a(x,y) = ∑ a( x,y) = ∑ ∑ a(x,y) .
x =1 y = x ( x,y)∈[n]×[n]; y =1 x =1
x ≤y
Enumerative Combinatorics: class notes page 151

Example 1.6.10. If we set n = 4 in Corollary 1.6.9 and rewrite the result without
using summation signs, then we obtain
 
a(1,1) + a(1,2) + a(1,3) + a(1,4)
 
+ a(2,2) + a(2,3) + a(2,4)
 
+ a(3,3) + a(3,4)
+ a(4,4)
 
= a(1,1) + a(1,2) + · · · + a(4,4) this is the sum of all the numbers a( x,y)
= a(1,1)
 
+ a(1,2) + a(2,2)
 
+ a(1,3) + a(2,3) + a(3,3)
 
+ a(1,4) + a(2,4) + a(3,4) + a(4,4) .

In other words, if we are given a triangular table of numbers:

a(1,1) a(1,2) a(1,3) a(1,4)

a(2,2) a(2,3) a(2,4)
,
a(3,3) a(3,4)
a(4,4)

then the following three procedures all produce the same result:

(a) summing up the numbers in each row of the table separately, and then
summing up these “row tallies”;

(b) summing up all the 10 numbers in the table;

(c) summing up the numbers in each column of the table separately, and then
summing up these “column tallies”.

Proof of Corollary 1.6.9. This is similar to the proof of Corollary 1.6.7; we just need to de-
fine the statements A ( x, y) differently. Theorem 1.6.6 (applied to X = [n], Y = [n] and
A ( x, y) = (“x ≤ y”)) yields
∑ ∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y) . (121)
x ∈[n] y∈[n]; ( x,y)∈[n]×[n]; y∈[n] x ∈[n];
x ≤y x ≤y x ≤y

We shall now rewrite the two inner sums in this chain of equalities.
Enumerative Combinatorics: class notes page 152

Let x ∈ [n]. Then, 1 ≤ x ≤ n. Hence, the elements y ∈ [n] that satisfy x ≤ y are precisely
the elements of { x, x + 1, . . . , n}. Thus,
n
∑ a(x,y) = ∑ a(x,y) = ∑ a(x,y) . (122)
y∈[n]; y∈{ x,x +1,...,n} y= x
x ≤y

Forget that we fixed x. We thus have proved (122) for each x ∈ [n].
Next, let y ∈ [n]. Then, 1 ≤ y ≤ n. Hence, the elements x ∈ [n] that satisfy x ≤ y are
precisely the elements of {1, 2, . . . , y}. Thus,
y
∑ a(x,y) = ∑ a(x,y) = ∑ a(x,y) . (123)
x ∈[n]; x ∈{1,2,...,y} x =1
x ≤y

Forget that we fixed y. We thus have proved (123) for each y ∈ [n].
In light of the two equalities (122) and (123), we can rewrite the chain of equalities (121)
as follows:
n y
∑ ∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y) .
x ∈[n] y= x ( x,y)∈[n]×[n]; y∈[n] x =1
x ≤y

In other words,
n n n y
∑∑ a(x,y) = ∑ a(x,y) = ∑ ∑ a(x,y)
x =1 y = x ( x,y)∈[n]×[n]; y =1 x =1
x ≤y
n
(since the summation sign ∑ is synonymous to ∑ , whereas the summation sign ∑ is
x ∈[n] x =1 y∈[n]
n
synonymous to ∑ ). This proves Corollary 1.6.9.
y =1

Let us illustrate Corollary 1.6.9 on two examples of its use. The first is an exercise
([19f-hw2s, Exercise 1]) about harmonic numbers:

Exercise 1.6.3. For each n ∈ N, we define the n-th harmonic number Hn by

n
1 1 1 1
Hn = + +···+ = ∑ .
1 2 n k =1
k

Prove that
H1 + H2 + · · · + Hn = (n + 1) ( Hn+1 − 1)
for each n ∈ N.

It is not hard to solve this exercise by induction on n (see [19f-hw2s, First solution
to Exercise 1]); but here is a different solution, using Corollary 1.6.9:
Enumerative Combinatorics: class notes page 153

Solution to Exercise 1.6.3. Each n ∈ N satisfies

n
1
Hn = ∑ k
k =1

(by the definition of Hn ). Renaming n as m in this statement, we obtain the follow-

ing: Each m ∈ N satisfies
m
1
Hm = ∑ . (124)
k =1
k
Now, let n ∈ N. Then,
n n m
1
H1 + H2 + · · · + Hn = ∑ Hm = ∑ ∑ .
|{z} k
(125)
m =1 m =1 k =1
m 1
=∑
k =1 k
(by (124))

1
Now, Corollary 1.6.9 (applied to a( x,y) = ) yields
x
n n n y
1 1 1
∑∑ x
= ∑ x
= ∑ ∑ x
.
x =1 y = x ( x,y)∈[n]×[n]; y =1 x =1
x ≤y

Renaming the summation indices x and y as k and m in this equality, we obtain

n n n m
1 1 1
∑ ∑ k
= ∑ k
= ∑ ∑ .
k
k =1 m = k (k,m)∈[n]×[n]; m =1 k =1
k≤m

Comparing this with (125), we obtain

n n n
1 1
H1 + H2 + · · · + Hn = ∑ ∑k = ∑ ( n − k + 1) · k .
k =1 m=k k =1
| {z }
1
=(n−k+1)·
k
(since this is a sum of n−k+1
many equal addends)
Enumerative Combinatorics: class notes page 154

Comparing this with

(n + 1) ( Hn+1 − 1)
= ( n + 1) Hn+1 − ( n + 1)
| {z } | {z }
n +1 1 n +1
=∑ =∑ 1
k =1 k k =1
(by (124), applied to m=n+1) n +1
(since ∑ 1 = (n + 1) · 1 = n + 1)
k =1
n +1 n +1 n +1   n +1
1 1 1
= ( n + 1) ∑
k k∑ ∑ ( n + 1) · k − 1 = ∑ ( n − k + 1) · k
− 1=
k =1 =1 k =1 | {z } k =1
1
=(n−k+1)·
k
n
1 1
= ( n − ( n + 1) + 1) · + ∑ ( n − k + 1) ·
| {z } n + 1 k =1 k
=0
(here, we have split off the addend for k = n + 1 from the sum)
n
1
= ∑ ( n − k + 1) · ,
k =1
k

we obtain H1 + H2 + · · · + Hn = (n + 1) ( Hn+1 − 1). This solves Exercise 1.6.3.

A more intricate application of Corollary 1.6.9 is the following identity for bino-
mial coefficients:

Exercise 1.6.4. Let n ∈ N. Then,

n
(−1)k−1 n
 
1 1 1
∑ k k
= + +···+ .
1 2 n
k =1

Exercise 1.6.4 is not too hard to solve by induction (see [Grinbe15, Exercise 3.19]
for such a solution); but for illustrative purposes, let me solve it using Corollary
1.6.9 here:
Solution to Exercise 1.6.4. Let k ∈ [n]. Then, 1 ≤ k ≤ n, so that k ≥ 1 and thus
k − 1 ∈ N. Note that k ≥ 1 also entails k ∈ {1, 2, 3, . . .}. Furthermore, n ≥ 1 (since
1 ≤ n) and thus n − 1 ∈ N. Hence, Theorem 1.3.29 (applied to n − 1 and k − 1
instead of n and k) yields

k−1 ( k − 1) + 1 ( k − 1) + 2 n−1
       
+ + +···+
k−1 k−1 k−1 k−1
( n − 1) + 1
   
n
= = (since (n − 1) + 1 = n and (k − 1) + 1 = k) .
( k − 1) + 1 k
Enumerative Combinatorics: class notes page 155

Hence,
k−1 ( k − 1) + 1 ( k − 1) + 2 n−1
         
n
= + + +···+
k k−1 k−1 k−1 k−1
n 
k−1 n−1 i−1
        
k k+1
= + + +···+ =∑ .
k−1 k−1 k−1 k−1 i =k
k − 1

(−1)k−1
Multiplying both sides of this equality by , we obtain
k

(−1)k−1 n (−1)k−1 n i − 1
   

k k
=
k ∑ k−1
i =k
n
(−1)k−1 i−1
 
=∑ (by (32))
i =k | {zk } k−1
(−1)k−1 i
= ·
i k
(since i 6=0 (because i ≥k≥1>0))
n
(−1)k−1 i i−1
 
=∑ ·
i =k
i k k−1
| {z  }
i
=
k
(since Proposition 1.3.36
(applied to iand
k instead
 of mand n)
i i i−1
yields = )
k k k−1
n
(−1)k−1 i
 
=∑ . (126)
i =k
i k

Now forget that we fixed k. We thus have proved (126) for each k ∈ [n]. Hence,
n
(−1)k−1 n n n
(−1)k−1 i
   
∑ k k
= ∑∑
i k
k =1 | {z } k =1 i = k
k −1  
n (−1) i
=∑
i =k i k
(by (126))
n i
(−1)k−1 i
 
=∑∑ . (127)
i =1 k =1
i k

Here, the last equality sign has been obtained by applying Corollary 1.6.9 to a( x,y) =
(−1) x−1 y
 
(and then renaming the summation indices x and y as k and i).
y x
Enumerative Combinatorics: class notes page 156

Now, fix i ∈ [n]. Thus, 1 ≤ i ≤ n, so that i ≥ 1 > 0 and therefore i 6= 0. Hence,

[i = 0] = 0. But (69) (applied to i instead of n) yields
i  
k i
∑ (−1) k = [i = 0] = 0.
k =0

Hence,
i     i  
i i k i
0 = ∑ (−1) k
= (−1) 0
+ ∑ (−1)
k =0
k | {z } 0 k =1
k
=1 |{z}
=1
(by (44))

(here, we have split off the addend for k = 0 from the sum)
i   i   i
i
= 1 + ∑ (−1) k
= 1 + ∑ − (−1) k −1

k =1
| {z } k k =1
k
k −1
=−(−1) | {z   }
i i
=− ∑ (−1)k−1
k =1 k
i  
i
= 1 − ∑ (−1)k−1 .
k =1
k
In other words,
i  
i
∑ (−1) k −1
k
= 1. (128)
k =1
Now, forget that we fixed i. We thus have proven (128) for each i ∈ [n]. Thus,
(127) becomes
n
(−1)k−1 n n i
(−1)k−1 i n i
     
1 k −1 i
∑ k k
=∑∑
i k
= ∑ ∑ · (−1)
i k
k =1 i =1 k=1 | {z } i =1 k =1
1
| {z   }
= ·(−1)k−1 1 i i
i = ∑ (−1)k−1
i k =1 k
(by (32))
n
1 i n n
 
k −1 i 1 1 1 1 1
= ∑ ∑ (−1) = ∑ ·1 = ∑ = + +···+ .
i
i =1 k =1
k i i 1 2 n
| {z } i =1 i =1
=1
(by (128))

This solves Exercise 1.6.4.

1.6.3. A cautionary tale about infinite sums

In Definition 1.3.26, I claimed that infinite sums (i.e., sums of the form ∑ a x where
x∈X
the indexing set X is infinite) are still well-defined as long as they only have finitely
Enumerative Combinatorics: class notes page 157

many nonzero addends; moreover, I claimed that they satisfy the same rules as
finite sums. This is true for all the rules listed above (including the Fubini principle
in all its forms) as long as we assume that each of the sums appearing in the rule
has only finitely many nonzero addends. Usually, it suffices to assume that some
of the sums have only finitely many nonzero addends, and then it follows that so
do the others.
Let us illustrate this on the example of (31). If S is an arbitrary set (not necessarily
finite), and if as and bs are two numbers for each s ∈ S, then the equality (31) holds
as long as we assume that the two sums ∑ as and ∑ bs have only finitely many
s∈S s∈S
nonzero addends (i.e., there are only finitely many s ∈ S satisfying as 6= 0, and
there are only finitely many s ∈ S satisfying bs 6= 0). We don’t need to assume
that the sum ∑ ( as + bs ) has only finitely many nonzero addends, because this
s∈S
follows “for free” from the analogous assumptions on the sums ∑ as and ∑ bs .
s∈S s∈S
However, it does not suffice to assume that the sum ∑ ( as + bs ) has only finitely
s∈S
many nonzero addends, if we don’t also make the analogous assumptions on ∑ as
s∈S
and ∑ bs . For example, if S is an infinite set, then the following transformation
s∈S
makes no sense:

∑ 0
|{z} = ∑ (1 + (−1)) = ∑ 1 + ∑ (−1) (by (31)) .
s∈S =1+(−1) s∈S s∈S s∈S

The formula (31) is being misapplied here, since the sums ∑ 1 and ∑ (−1) have
s∈S s∈S
infinitely many nonzero addends.
Something similar needs to be taken into account when extending Theorem 1.6.1
to infinite sums. It suffices to assume that the sum ∑ a( x,y) has only finitely
( x,y)∈ X ×Y
many nonzero addends (i.e., that only finitely many pairs ( x, y) ∈ X × Y satisfy
a( x,y) 6= 0). In other words, the following analogue of Theorem 1.6.1 holds for
infinite sums:
Theorem 1.6.11. Let X and Y be two sets. Let a( x,y) be a number for each pair
( x, y) ∈ X × Y. Assume that only finitely many pairs ( x, y) ∈ X × Y satisfy
a( x,y) 6= 0. Then, we have

∑ ∑ a(x,y) = ∑ a( x,y) = ∑ ∑ a( x,y) ,

x ∈ X y ∈Y ( x,y)∈ X ×Y y ∈Y x ∈ X

and in particular, each of the five sums in this equality has only finitely many
nonzero addends.
See [19s, proof of Proposition 7.2.11 (specifically, the proof of (227))] for a rigor-
ous proof of this theorem (but the idea is obvious: reduce it to the case of finite X
and Y by restricting the sums appropriately).
Enumerative Combinatorics: class notes page 158

Thus, in order to make Theorem 1.6.1 work for infinite sums, it suffices to as-
sume the middle sum ∑ a( x,y) to have only finitely many nonzero addends.
( x,y)∈ X ×Y
However, it is tempting to think that even if this middle sum has infinitely many
nonzero addends, we may still get the equality ∑ ∑ a( x,y) = ∑ ∑ a( x,y) as long
x ∈ X y ∈Y y ∈Y x ∈ X
as all the four sums appearing in this equality have only finitely many nonzero
addends. In other words, it is tempting to use the following “fact”:
Incorrect Fubini rule. Let X and Y be two sets. Let a( x,y) be a number
for each pair ( x, y) ∈ X × Y. Then,
∑ ∑ a(x,y) = ∑ ∑ a( x,y) , (129)
x ∈ X y ∈Y y ∈Y x ∈ X

as long as all four sums appearing in this equality have only finitely
many nonzero addends.
This, however, is false. Here is a counterexample:
Example 1.6.12 (“the serial debtor”). Let X = {1, 2, 3, . . .} and Y = {1, 2, 3, . . .}.
Define a number a( x,y) for each pair ( x, y) ∈ X × Y by setting
a( x,y) = [y = x ] − [y = x + 1] .
Here is a table showing these numbers:

a( x,y) y=1 y=2 y=3 y=4 y=5 ···

x=1 1 −1 ···
x=2 1 −1 ···
x=3 1 −1 ··· , (130)
x=4 1 −1 ···
x=5 1 ···
.. .. .. .. .. .. ...
. . . . . .

where empty slots are understood to contain the number 0. It is easy to see that
all four sums in (129) have only finitely many nonzero addends. Indeed, for each
x ∈ X, we have
∑ a(x,y) = 0 (131)
y ∈Y

(since the only nonzero addends of this sum are 1 and −1, which clearly sum up
to 0). Hence,
∑ ∑ a(x,y) = ∑ 0 = 0. (132)
x ∈ X y ∈Y x∈X
| {z }
=0
Enumerative Combinatorics: class notes page 159

On the other hand, for each y ∈ Y, we have

(
1, if y = 1;
∑ a(x,y) = 0, if y 6= 1
x∈X

(since the only nonzero addends of this sum are 1 and −1, except that there is
no −1 when y = 1). Hence,
(
1, if y = 1;
∑ ∑ a(x,y) = ∑ 0, if y 6= 1 = 1. (133)
y ∈Y x∈X y ∈Y
| {z }

1, if y = 1;
=
0, if y 6 = 1

Comparing this with (132), we see that (129) becomes 0 = 1, which is clearly
false. Hence, the Incorrect Fubini rule is false. But Theorem 1.6.11 does not
apply here, since infinitely many pairs ( x, y) ∈ X × Y satisfy a( x,y) 6= 0.

(Example 1.6.12 can be thought of as a mathematical model of check kiting, or

more generally of paying back a debt by opening a new line of credit. In this
metaphor, each row of the table (130) represents a line of credit, while each column
represents a day, with the entry 1 standing for incurring debt and the entry −1
standing for paying it back. The equality (131) says that each single debt gets
eventually paid back, but the equality (133) shows that the debtor has an extra
(“undeserved”) unit of money at his disposal. This paradox would not work if
there were only finitely many lines of credit; one day the debtor would have to pay
back his last debt.)
Note that we are only considering the simplest kind of infinite sums here: the
ones that have only finitely many nonzero addends. In analysis, more general
situations are studied in which an infinite sum can have a well-defined (finite)
value. In those situations, there are much subtler criteria for when summation signs
can be interchanged (such as absolute summability). But this is a combinatorics
course, so we will not need them.

1.7. Counting permutations: an introduction

We now come back to take another look at counting maps. Namely, we restrict
ourselves to a special class of maps: the permutations.

1.7.1. Permutations and derangements

There are two different things called “permutations” in mathematics. One is a
kind of maps (“active permutations”); the other is a kind of lists (“passive permu-
Enumerative Combinatorics: class notes page 160

tations”). We are always going to use the word “permutation” for the former, at
least when we are speaking of permutations of a set. Here is how it is defined:

Definition 1.7.1. A permutation of a set X means a bijection from X to X.

We shall use the two-line notation for maps (see Definition 1.2.11). Thus, for
example,  
0 1 5 7 9
1 7 5 0 9
denotes the map from {0, 1, 5, 7, 9} to {0, 1, 5, 7, 9} that sends 0, 1, 5, 7, 9 to 1, 7, 5, 0, 9,
respectively. This map is a permutation of the set {0, 1, 5, 7, 9}. On the other hand,
the map  
0 1 5 7 9
1 7 0 0 9
is not a permutation, since it fails to be injective (indeed, it sends both 5 and 7 to
0) and therefore also fails to be bijective. Also, the map
 
0 1 5 7 9
1 2 3 4 5

is not a permutation, since it is a map between two different sets (while a permu-
tation must be a map from a set to itself).
The most obvious counting question you can now ask has a simple answer:

Theorem 1.7.2. Let n ∈ N. Let X be an n-element set. Then,

(# of permutations of X ) = n!.

We shall prove this theorem later (in Subsection 2.4.4). (The idea of the proof is
simple, but formalizing it is tricky at this point.)

Example 1.7.3. The set [3] = {1, 2, 3} has 3! = 6 many permutations; they are
(written in two-line notation)
     
1 2 3 1 2 3 1 2 3
, , ,
1 2 3 1 3 2 2 1 3
     
1 2 3 1 2 3 1 2 3
, , .
2 3 1 3 1 2 3 2 1

Let us next focus on a specific kind of permutations.

Enumerative Combinatorics: class notes page 161

Definition 1.7.4. Let X be a set.

(a) If f : X → X is a map, then a fixed point of f means an element x ∈ X such
that f ( x ) = x.
(b) A derangement of X means a permutation of X that has no fixed points.

Example 1.7.5. (a) The fixed points of the map

 
0 1 5 7 9
1 7 5 0 9

are 5 and 9. Thus, this map is not a derangement of {0, 1, 5, 7, 9} (even though it
is a permutation of {0, 1, 5, 7, 9}).
(b) The map  
0 1 5 7 9
1 7 9 0 5
has no fixed points, and thus is a derangement of {0, 1, 5, 7, 9} (since it is a
permutation of {0, 1, 5, 7, 9}).

We can restate Definition 1.7.4 as follows: A derangement of a set X means a

permutation f of X such that

f ( x ) 6= x for all x ∈ X.

1.7.2. Only the size counts

How many derangements does an n-element set have?
A first simplification is to restrict our focus to a specific n-element set, namely
[n], because all n-element sets should be interchangeable as far as this question is
concerned. This relies on the following lemma:

Lemma 1.7.6. Let n ∈ N. Let X be any n-element set. Then,

(# of derangements of X ) = (# of derangements of [n]) .

This lemma might be so obvious to you that you would wonder what there is to
be proven about it; shouldn’t it be clear that the # of derangements of a set only
depends on the size of that set? The following proof, which we shall first give in
an informal version, just spells out this idea:
Proof of Lemma 1.7.6 (informal version). We have | X | = n (since X is an n-element
set) and |[n]| = n, so that | X | = n = |[n]|. Thus, the sets X and [n] have the same
size. Hence, Theorem 1.1.7 (applied to Y = [n]) shows that there is a bijection
φ : X → [n]. Fix such a φ.
We can regard this bijection φ as a way to label the elements of X by the numbers
1, 2, . . . , n (with each element a ∈ X getting the label φ ( a) ∈ [n]). For example, if X
Enumerative Combinatorics: class notes page 162

 
x y z
is a 3-element set { x, y, z}, and φ : X → [3] is the bijection110 , then we
1 2 3
can think of φ as labeling the elements x, y, z as 1, 2, 3, respectively.
Now, we claim that any derangement of X can be transformed into a derange-
ment of [n]. Indeed, if a derangement of X is written in two-line notation, then
we can simply replace each a ∈ X appearing in this two-line notation by its
“label” φ ( a) ∈ [n], and thus obtain a derangement of [n] written in two-line
notation.
 For example, if X is a 3-element set { x, y, z}, and φ is the bijection

x y z
: X → [3], then the derangement
1 2 3
 
x y z
of X
y z x
becomes the derangement
   
φ ( x ) φ (y) φ (z) 1 2 3
= of [3] .
φ (y) φ (z) φ ( x ) 2 3 1

This way, we have assigned a derangement of [n] to each derangement of X. This

assignment is “obviously”111 a bijection from the set {derangements of X } to the
set {derangements of [n]}. The bijection principle thus yields that
|{derangements of X }| = |{derangements of [n]}| ,
and this quickly yields Lemma 1.7.6.
This proof was somewhat informal: We have not shown that our assignment is
a well-defined map from {derangements of X } to {derangements of [n]} (and this
is not obvious, because a map usually can be written in two-line notation in several
ways); nor have we shown that it is a bijection. I shall thus rewrite this proof in a
more rigorous and formal way below.
First of all, let me recall some basic notations regarding maps. Recall that if
g : A → B and f : B → C are two maps (between some sets A, B and C), then the
composition f ◦ g of these two maps is defined to be the map
A → C,
a 7→ f ( g ( a)) .
(I pronounce “ f ◦ g” as “ f after g”, because this composition f ◦ g acts on an ele-
ment a ∈ A by first applying g and afterwards applying f to the result.)
The composition of maps is associative: i.e., if f , g and h are three maps for which
the compositions ( f ◦ g) ◦ h and f ◦ ( g ◦ h) make sense112 , then these two compo-
sitions ( f ◦ g) ◦ h and f ◦ ( g ◦ h) are equal, and thus both of them can be called
110 Inthis version of this proof, all maps will be written in two-line notation.
111 See below for a rigorous point of view.
112 This means that the target of h is the domain of g, and that the target of g is the domain of f .
Enumerative Combinatorics: class notes page 163

f ◦ g ◦ h. (See [Grinbe15, Proposition 2.82] for a proof of this fact, and [Grinbe15,
Theorem 2.86] for its analogue for more than three maps.)
It is easy to see that any composition of bijections is again a bijection. In par-
ticular, if f , g and h are three bijections for which the composition f ◦ g ◦ h makes
sense, then this composition f ◦ g ◦ h is a bijection.
Now, the bijection from {derangements of X } to {derangements of [n]} that we
have constructed in our above proof of Lemma 1.7.6 can be viewed in a new light:
This bijection sends any derangement ω of X to the composition

φ ◦ ω ◦ φ −1 : [ n ] → [ n ]

(where φ−1 : [n] → X is the inverse of the bijection φ : X → [n]). Indeed, if a

derangement ω of X is given by
 
a1 a2 · · · a n
ω= in two-line notation,
b1 b2 · · · bn

then
 
−1 φ ( a1 ) φ ( a2 ) · · · φ ( an )
φ◦ω◦φ = in two-line notation
φ (b1 ) φ (b2 ) · · · φ ( bn )

113 ,which is precisely what we obtain when we replace each a ∈ X appearing in

the two-line notation of ω by its “label” φ ( a) ∈ [n]. Thus, we don’t need to speak
about two-line notations at all when defining our bijection; we can simply define it
as the map that sends each derangement ω of X to the derangement φ ◦ ω ◦ φ−1 of
[n]. Thus, our above proof of Lemma 1.7.6 takes the following rigorous shape:
Proof of Lemma 1.7.6 (formal version). We have | X | = n (since X is an n-element set)
and |[n]| = n, so that | X | = n = |[n]|. Thus, the sets X and [n] have the same
size. Hence, Theorem 1.1.7 (applied to Y = [n]) shows that there is a bijection
φ : X → [n]. Fix such a φ.
Now, the map φ : X → [n] is a bijection; thus, it has an inverse map φ−1 : [n] →
X. Thus, for any map ω : X → X, we can form the composition

φ ◦ ω ◦ φ −1 : [ n ] → [ n ] .

113 because each i ∈ [n] satisfies

  
 
φ ◦ ω ◦ φ−1 (φ ( ai )) = φ ω φ−1 (φ ( ai )) = φ (ω ( ai )) = φ (bi )
  
| {z }
= ai
   
a1 a2 ··· an
since ω = shows that ω ( ai ) = bi
b1 b2 ··· bn
Enumerative Combinatorics: class notes page 164

Moreover, if ω is a derangement of X, then φ ◦ ω ◦ φ−1 is a derangement of [n] 114 .

Hence, we can define a map

A : {derangements of X } → {derangements of [n]} ,

ω 7 → φ ◦ ω ◦ φ −1 .

A similar argument shows that we can define a map

B : {derangements of [n]} → {derangements of X } ,

η 7→ φ−1 ◦ η ◦ φ.

It is easy to see that these two maps A and B are mutually inverse115 . Hence, the
map A is invertible, and thus a bijection. Therefore, the bijection principle yields

|{derangements of X }| = |{derangements of [n]}| .

In other words,

(# of derangements of X ) = (# of derangements of [n]) .

This proves Lemma 1.7.6.

114 Proof.
Let ω be a derangement of X. Thus, ω is a permutation of X that has no fixed points (by the
definition of a derangement). If the map φ ◦ ω ◦ φ−1 had a fixed point y ∈ [n], then φ−1 (y) ∈ X
would be a fixed point of ω (since it would satisfy
      
φ ω φ −1 ( y ) = φ ◦ ω ◦ φ −1 ( y ) = y since y is a fixed point of φ ◦ ω ◦ φ−1

and therefore ω φ−1 (y) = φ−1 (y)); but this would contradict the fact that ω has no fixed

points. Hence, the map φ ◦ ω ◦ φ−1 has no fixed points either. Furthermore, the map ω is a
permutation of X, that is, a bijection from X to X. Moreover, the maps φ and φ−1 are bijections.
Hence, the composition φ ◦ ω ◦ φ−1 : [n] → [n] is also a bijection (because it is a composition of
the three bijections φ, ω and φ−1 ). Thus, φ ◦ ω ◦ φ−1 is a bijection from [n] to [n]. In other words,
φ ◦ ω ◦ φ−1 is a permutation of [n]. Thus, φ ◦ ω ◦ φ−1 is a permutation of [n] that has no fixed
points (since we have previously shown that φ ◦ ω ◦ φ−1 has no fixed points). In other words,
φ ◦ ω ◦ φ−1 is a derangement of [n] (by the definition of a derangement). Qed.
115 Proof. We need to check that A ◦ B = id and B ◦ A = id. We shall only prove that A ◦ B = id, and

leave the analogous proof of B ◦ A = id to the reader.

Let η be a derangement of [n]. Then, B (η ) = φ−1 ◦ η ◦ φ (by the definition of B). Applying
the map A to both sides of this equality, we find
   
A ( B (η )) = A φ−1 ◦ η ◦ φ = φ ◦ φ−1 ◦ η ◦ φ ◦ φ−1 (by the definition of A)
= φ ◦ φ−1 ◦η ◦ φ ◦ φ−1 = id ◦η ◦ id = η.
| {z } | {z }
=id =id

Thus, ( A ◦ B) (η ) = A ( B (η )) = η = id (η ).
Forget that we fixed η. We thus have proved that ( A ◦ B) (η ) = id (η ) for each derangement
η of [n]. In other words, A ◦ B = id. As we said, the proof of B ◦ A = id is analogous. Thus, it
follows that the maps A and B are mutually inverse.
Enumerative Combinatorics: class notes page 165

Lemma 1.7.6 is not specific to derangements; it is merely an instance of a far-

ranging principle, which I will not state in its general form because that would
require a detour into category theory. Roughly speaking, the principle says the
following:

“Isomorphism principle” (informal): Assume that we are given a way

to assign a number f (S) to any finite set S, and this way does not
depend on what the elements of S are (i.e., it has a definition that works
the same no matter whether the elements of S are numbers or tuples or
maps or anything else). Let n ∈ N. Let X be any n-element set. Then,
f ( X ) = f ([n]).

Of course, this is not rigorous until we formalize what “does not depend on what
the elements of S are” means. This I am not going to do; but let me delineate it
through a few examples:
We can apply the “isomorphism principle” to f (S) = (# of derangements of S),
because the # of derangements of S is defined in a way that does not depend on
what the elements of S are. Thus, we obtain Lemma 1.7.6.
Likewise, we can apply the “isomorphism principle” to
f (S) = (# of permutations of S). Thus, we obtain the fact that if n ∈ N, and if X
is an n-element set, then

(# of permutations of X ) = (# of permutations of [n]) . (134)

Likewise, we can apply the “isomorphism principle” to f (S) = (# of subsets of S).

Thus, we obtain the fact that if n ∈ N, and if X is an n-element set, then

(# of subsets of X ) = (# of subsets of [n]) . (135)

Likewise, we can apply the “isomorphism principle” to

f (S) = (# of subsets of S having even size). Thus, we obtain the fact that if n ∈ N,
and if X is an n-element set, then

(# of subsets of X having even size) = (# of subsets of [n] having even size) .

The number f (S) in the “isomorphism principle” doesn’t have to be the # of

some objects. We can just as well apply the “isomorphism principle” to f (S) =
∑ | T | (that is, the sum of the sizes of all subsets of S). Thus, we obtain the fact
T ⊆S
that if n ∈ N, and if X is an n-element set, then

∑ |T | = ∑ |T | .
T⊆X T ⊆[n]

What we cannot do is apply the “isomorphism principle” to

f (S) = (# of lacunar subsets of S). Indeed, the word “lacunar” is only defined for
sets of integers, and whether a subset of S is lacunar depends on what its elements
Enumerative Combinatorics: class notes page 166

are. And unsurprisingly, it is not true that if n ∈ N, and if X is an n-element set of

integers, then

(# of lacunar subsets of X ) = (# of lacunar subsets of [n]) .

For example, if n = 3 and X = {1, 3, 5}, then all 8 subsets of X are lacunar, but
only 5 subsets of [n] = [3] are lacunar.
Likewise, we cannot apply the “isomorphism principle” to
f (S) = (# of self-counting subsets of S). Indeed, for example, if we relabel the
elements 1 and 2 of the set [2] as 8 and 9, then the self-counting subset [1] will no
longer be self-counting.
Since I am not proving the “isomorphism principle” in its general form, let me say a
few words on how to prove its specific instances. Essentially, every instance of the “iso-
morphism principle” can be proven by the same strategy that we used in our proof of
Lemma 1.7.6: Namely, we fix a bijection φ : X → [n], and we use it to construct a bijection
from {the objects counted by f (S)} to {the objects counted by f ([n])}, which proceeds by
replacing each element a ∈ X by its “label” φ ( a) ∈ [n]. When formalized, these proofs may
look rather different, but the underlying idea is always this one. In particular, the formal
proof of (134) resembles our formal proof of Lemma 1.7.6 very closely (but is shorter, be-
cause we don’t have to show that φ ◦ ω ◦ φ−1 and φ−1 ◦ η ◦ φ are derangements), whereas
the formal proof of (135) requires a different formula for the maps A and B 116 (but

otherwise, again, proceeds in the same way). I hope it is clear enough how these formal
proofs can be constructed more or less mechanically whenever the need arises.

1.7.3. Intermezzo: OEIS

Lemma 1.7.6 shows that, in order to count the derangements of any finite set X, it
suffices to count the derangements of [n] for any n ∈ N. Let us do this now.

Definition 1.7.7. For each n ∈ N, let

Dn = (# of derangements of [n]) .

116 namely,

A : {subsets of X } → {subsets of [n]} ,

U 7 → φ (U ) = { φ ( u ) | u ∈ U }

and

B : {subsets of [n]} → {subsets of X } ,

n o
V 7 → φ −1 ( V ) = φ −1 ( v ) | v ∈ V
Enumerative Combinatorics: class notes page 167

Example 1.7.8. Let us compute D3 . Indeed, we know from Example 1.7.3 that
the set [3] has 6 permutations (which are all listed in Example 1.7.3). Among
these 6 permutations, only 2 are derangements, namely
   
1 2 3 1 2 3
and
2 3 1 3 1 2

(written in two-line notation). Thus, the # of derangements of [3] is 2. In other

words, D3 = 2.

In a similar manner, we can count the derangements of any other given finite set.
In particular, we thus obtain

D0 = 1 (since id : ∅ → ∅ is a derangement) ;
D1 = 0 (since id : [1] → [1] is not a derangement) ;
  
1 2
D2 = 1 since the only derangement of [2] is ;
2 1
D3 = 2 (as we have just seen) ;
D4 = 9.

These days, when you have a sequence of integers you want to learn more about,
and you know a few of its values, you can look it up at the OEIS (the Online
Encyclopedia of Integer Sequences, available at https://oeis.org ). Let us try this
on the sequence ( D0 , D1 , D2 , D3 , . . .). We know its first five values D0 = 1, D1 =
0, D2 = 1, D3 = 2, D4 = 9, so let us enter “1,0,1,2,9” into the input field
of the OEIS. We press “Search” and are greeted with a list of several117 known
sequences that contain these five values somewhere in them (in five consecutive
positions). The first hit is a sequence labeled A000166, and named “Subfactorial
or rencontres numbers, or derangements: number of permutations of n elements
with no fixed points”. This is, of course, precisely our sequence (no wonder that
such a simple concept as derangements is well-known). The OEIS has collected a
lot of information on this sequence, including a long list of entries, combinatorial
interpretations, formulas (including many we don’t yet understand), references and
links. Here is only a small selection of the formulas that are claimed in the OEIS
for this sequence:

117 currently 45, but quite possibly more by the time you are reading this
Enumerative Combinatorics: class notes page 168

Theorem 1.7.9. (a) We have Dn = (n − 1) ( Dn−1 + Dn−2 ) for all n ≥ 2.

(b) We have Dn = nDn−1 + (−1)n for all integers n ≥ 1.
(c) We have
n  
n
n! = ∑ Dn − k for all n ∈ N.
k =0
k
(d) We have
n
n!
Dn = ∑ (−1)k k! for all n ∈ N.
k =0
 
1
(e) Set round ( x ) = x + for each x ∈ R. Then,
2
 
n!
Dn = round for all n ≥ 1,
e
∞ 1
where e = ∑ ≈ 2.718 . . . is the base of the natural logarithm.
k=0 k!

Some of these formulas will be proven in Subsection 2.9.5. (Theorem 1.7.9 (b)
follows from Theorem 1.7.9 (a) via [19f-hw1s, Exercise 5].) Note that parts (a)
and (b) of Theorem 1.7.9 are two recurrences for the sequence ( D0 , D1 , D2 , . . .) that
make its computation rather easy; both are due to Euler.
The OEIS can be used in many ways. It knows thousands of sequences that have ap-
peared in mathematics (not only combinatorics). You can search for them both by entries
(as we did above) and by keywords (such as “permutations” or “derangements”). It even
acts as a bare-bones social network: If two mathematicians encounter the same sequence in
their work, and the first one records it in the OEIS, the second can then find it and contact
the first. This has led to a number of collaborations between mathematicians working in
rather different disciplines; often, the appearance of the same integer sequence in different
places hints at a deeper connection between the theories.
It is known from basic set theory that there exists a bijection between N and N2 . Thus, a


“2-dimensional sequence” (i.e., a family ai,j (i,j)∈N2 of numbers whose entries are indexed
by two rather than one nonnegative integers) can be re-parametrized as a (regular, i.e.,
1-dimensional) sequence. Thus, such “2-dimensional sequences” and similar families of
numbers
  can also be found in the OEIS. For example, Pascal’s triangle (i.e., the family
n
) appears as sequence A007318.
k (n,k)∈N2 ; k≤n

1.7.4. The one-line notation

If σ is a permutation of the set [n] (for some n ∈ N), then we can write σ in two-line
notation as follows:  
1 2 ··· n
σ= .
σ (1) σ (2) · · · σ ( n )
Enumerative Combinatorics: class notes page 169

But you’ll quickly tire of this notation, since its whole top row is useless: It just
contains the numbers 1, 2, . . . , n in increasing order. Thus, we can omit this top row
and only write the bottom row down118 . The bottom row is usually written as an
n-tuple (so the elements σ (1) , σ (2) , . . . , σ (n) are separated by commas). This is
the so-called one-line notation of σ:

Definition 1.7.10. Let n ∈ N. Let σ be a permutation of [n]. Then, the one-line

notation of σ is defined to be the n-tuple (σ (1) , σ (2) , . . . , σ (n)).

Warning: This notation clashes with a different notation (the cycle notation),
which is used in many textbooks. We shall avoid this conflict by using a differ-
ent notation for cycles, but you should keep this in mind when consulting other
texts. Other authors often avoid this conflict by using square brackets instead of
parentheses in the one-line notation119 ; but we prefer not to do so, since we are
already using square brackets for myriad other things.

Example 1.7.11. The 6 permutations of the set [3] have been listed in Example
1.7.3. In one-line notation, they look as follows:

(1, 2, 3) , (1, 3, 2) , (2, 1, 3) , (2, 3, 1) , (3, 1, 2) , (3, 2, 1) .

Let us observe that the one-line notation (as defined in Definition 1.7.10) can be
used not only for permutations of [n], but for any maps from [n] to an arbitrary set
X.
Perhaps the most useful thing about one-line notations is that they turn permuta-
tions (a somewhat abstract object) into n-tuples of numbers (a much more tangible
object). Usually, it is easier to manipulate tuples (e.g., by inserting or removing
elements) than to manipulate permutations. Using the one-line notation, we can
substitute the former for the latter. This relies on the following simple fact:

Proposition 1.7.12. Let n ∈ N. Then:

(a) If σ is any permutation of [n], then the one-line notation of σ is an n-tuple
of distinct elements of [n] that contains all elements of [n].
(b) If (i1 , i2 , . . . , in ) is an n-tuple of distinct elements of [n] that contains all
elements of [n], then there exists a unique permutation σ of [n] such that the
one-line notation of σ is (i1 , i2 , . . . , in ).
(c) Let

Tn = {n-tuples of distinct elements of [n] that contain all elements of [n]} .

118 Of course, we can only do this if we agree that the top row is 1, 2, . . . , n rather than some other
ordering of the elements of [n].
119 i.e., they write the one-line notation of σ as [ σ (1) , σ (2) , . . . , σ ( n )]
Enumerative Combinatorics: class notes page 170

The map

{permutations of [n]} → Tn ,
σ 7→ (one-line notation of σ) = (σ (1) , σ (2) , . . . , σ (n))

is well-defined and is a bijection.

Proof of Proposition 1.7.12. This is a simple exercise in understanding what bijectivity means:
(a) Let σ be a permutation of [n]. Thus, σ is a bijection from [n] to [n] (by the definition
of a permutation). Hence, the map σ is bijective, i.e., both injective and surjective.
The one-line notation of σ is defined to be the n-tuple (σ (1) , σ (2) , . . . , σ (n)). This n-
tuple consists of n distinct entries120 , all of which are elements of [n]. Furthermore, this
n-tuple contains all elements of [n] 121 . Hence, the one-line notation of σ is an n-tuple of
distinct elements of [n] that contains all elements of [n]. This proves Proposition 1.7.12 (a).
(b) Let (i1 , i2 , . . . , in ) be an n-tuple of distinct elements of [n] that contains all elements of
[n]. Then, i1 , i2 , . . . , in are elements of [n]. Hence, we can define a map

f : [n] → [n] ,
j 7→ i j .

This map f is injective122 and surjective123 . Hence, f is bijective. In other words, f is a

bijection from [n] to [n]. In other words, f is a permutation of [n] (by the definition of a
permutation). Moreover, the one-line notation of f is

( f (1) , f (2) , . . . , f (n)) (by the definition of the one-line notation)

= ( i1 , i2 , . . . , i n ) since f ( j) = i j for each j ∈ [n] (by the definition of f ) .

Hence, there is at least one permutation σ of [n] such that the one-line notation of σ is
(i1 , i2 , . . . , in ) (namely, σ = f ). Moreover, it is easy to see that there is at most one such
permutation124 . Combining these two observations, we conclude that there is a unique

120 Proof. If i and j are two distinct elements of [n], then i 6= j and thus σ (i ) 6= σ ( j) (since σ is
injective). In other words, the n entries of the n-tuple (σ (1) , σ (2) , . . . , σ (n)) are distinct. Qed.
121 Proof. Let x ∈ [ n ]. We must prove that this n-tuple ( σ (1) , σ (2) , . . . , σ ( n )) contains x. In other

words, we must prove that there exists a y ∈ [n] such that σ (y) = x. But this follows from the
fact that σ is surjective. Qed.
122 Proof. The elements of the n-tuple (i , i , . . . , i ) are distinct (by the definition of this n-tuple).
1 2 n
Now, let a and b be two distinct elements of [n]. Thus, a 6= b, so that i a 6= ib (since the elements
of the n-tuple (i1 , i2 , . . . , in ) are distinct). The definition of f yields f ( a) = i a and f (b) = ib .
Hence, f ( a) = i a 6= ib = f (b).
Forget that we fixed a and b. We thus have shown that if a and b are two distinct elements of
[n], then f ( a) 6= f (b). In other words, the map f is injective.
123 Proof. Let x ∈ [ n ]. The n-tuple (i , i , . . . , i ) contains all elements of [ n ]. Thus, in particular, it
1 2 n
contains x. In other words, there exists some j ∈ [n] such that i j = x. Consider this j. The
definition of f now yields f ( j) = i j = x, so that x = f ( j). In other words, x is an image under
f.
Forget that we fixed x. We thus have shown that each x ∈ [n] is an image under f . In other
words, the map f is surjective.
124 Proof. Let α be any such permutation. We shall prove that α = f .
Enumerative Combinatorics: class notes page 171

permutation σ of [n] such that the one-line notation of σ is (i1 , i2 , . . . , in ). This proves
Proposition 1.7.12 (b).
(c) If σ is a permutation of [n], then the one-line notation of σ is an n-tuple of distinct
elements of [n] that contains all elements of [n] (by Proposition 1.7.12 (a)). In other words,
if σ is a permutation of [n], then the one-line notation of σ belongs to Tn (because of how
Tn was defined). Thus, the map

{permutations of [n]} → Tn ,
σ 7→ (one-line notation of σ) = (σ (1) , σ (2) , . . . , σ (n))

is well-defined. It remains to prove that this map is a bijection.

Let us denote this map by OLN (which is short for “one-line notation”). We thus must
prove that OLN is a bijection.
The map OLN is injective125 and surjective126 . Hence, OLN is bijective. In other words,

Indeed, α is a permutation σ of [n] such that the one-line notation of σ is (i1 , i2 , . . . , in ). In

other words, α is a permutation of [n], and its one-line notation is (i1 , i2 , . . . , in ). Hence,

(i1 , i2 , . . . , in ) = (the one-line notation of α) = (α (1) , α (2) , . . . , α (n)) ,

so that
(α (1) , α (2) , . . . , α (n)) = (i1 , i2 , . . . , in ) = ( f (1) , f (2) , . . . , f (n))
(since ( f (1) , f (2) , . . . , f (n)) = (i1 , i2 , . . . , in )). In other words, α ( x ) = f ( x ) for each x ∈ [n].
Since both α and f are maps from [n] to [n] (being permutations of [n]), we thus conclude that
α = f.
Now, forget that we fixed α. We thus have shown that if α is any permutation σ of [n] such that
the one-line notation of σ is (i1 , i2 , . . . , in ), then α = f . Hence, there is at most one permutation
σ of [n] such that the one-line notation of σ is (i1 , i2 , . . . , in ).
125 Proof. Let σ and τ be two permutations of [ n ] such that OLN ( σ ) = OLN ( τ ). Thus, the defini-

tion of OLN yields OLN (σ ) = (one-line notation of σ) = (σ (1) , σ (2) , . . . , σ (n)) and similarly
OLN (τ ) = (τ (1) , τ (2) , . . . , τ (n)). Hence,

(σ (1) , σ (2) , . . . , σ (n)) = OLN (σ) = OLN (τ ) = (τ (1) , τ (2) , . . . , τ (n)) .

In other words, σ ( x ) = τ ( x ) for each x ∈ [n]. Hence, σ = τ (since both σ and τ are maps from
[n] to [n]).
Forget that we fixed σ and τ. We thus have shown that if σ and τ are two permutations of [n]
such that OLN (σ ) = OLN (τ ), then σ = τ. In other words, the map OLN is injective.
126 Proof. Let x ∈ T . Thus, x ∈ T = { n-tuples of distinct elements of [ n ] that contain all elements of [ n ]};
n n
in other words, x is an n-tuple of distinct elements of [n] that contains all elements of [n]. Hence,
we can write x in the form x = (i1 , i2 , . . . , in ). Thus, (i1 , i2 , . . . , in ) = x is an n-tuple of distinct
elements of [n] that contains all elements of [n]. Therefore, Proposition 1.7.12 (b) shows that
there exists a unique permutation σ of [n] such that the one-line notation of σ is (i1 , i2 , . . . , in ).
Consider this σ. The definition of OLN yields

OLN (σ ) = (one-line notation of σ )

= ( i1 , i2 , . . . , i n ) (since the one-line notation of σ is (i1 , i2 , . . . , in ))
= x.

Hence, x is an image under the map OLN.

Now, forget that we fixed x. We thus have shown that every x ∈ Tn is an image under the
map OLN. In other words, the map OLN is surjective.
Enumerative Combinatorics: class notes page 172

OLN is a bijection. This completes the proof of Proposition 1.7.12 (c).

Class of 2019-10-16

1.7.5. Short-legged permutations

Derangements are just one special type of permutations. Many others can be de-
fined. For example, let me define what I call the short-legged permutations:

Definition 1.7.13. Let n ∈ N. A permutation σ of [n] is said to be short-legged if

each i ∈ [n] satisfies |σ (i ) − i | ≤ 1.

The word “short-legged” is my own creation (and refers to the idea that a short-
legged permutation of [n] “carries no i ∈ [n] any further than 1 step”, a way
of visualizing the inequality |σ (i ) − i | ≤ 1). But this type of permutations has
a long history, and appears naturally in the study of determinants (specifically,
of determinants of tridiagonal matrices – see, e.g., [Grinbe15, Second solution to
Exercise 6.27]).

Example 1.7.14. (a) Among the 6 permutations of [3], exactly 3 are short-legged,
namely
(1, 2, 3) , (1, 3, 2) , (2, 1, 3)
(in one-line notation). Indeed, for example, the permutation σ = (1, 3, 2) is short-
legged (since it has |σ (1) − 1| = |1 − 1| = 0 ≤ 1 and |σ (2) − 2| = |3 − 2| = 1 ≤ 1
and |σ (3) − 3| = |2 − 3| = 1 ≤ 1), whereas the permutation σ = (2, 3, 1) is not
short-legged (since it has |σ (3) − 3| = |1 − 3| = 2 > 1).
(b) Among the 24 permutations of [4], exactly 5 are short-legged, namely

(1, 2, 3, 4) , (1, 2, 4, 3) , (1, 3, 2, 4) , (2, 1, 3, 4) , (2, 1, 4, 3) .

We can now wonder: How many permutations of [n] are short-legged?

Proposition 1.7.15. Let n ∈ N. Then,

(# of short-legged permutations of [n]) = f n+1 .

That makes yet another question answered by the Fibonacci numbers! I will not
prove this proposition in full detail, but let me sketch the main ideas of two proofs:
First proof of Proposition 1.7.15 (sketched). Let wn denote the # of short-legged per-
mutations of [n]. Then, it suffices to show that wn = wn−1 + wn−2 for each n ≥ 2
(because we can then argue as in the proof of Proposition 1.1.11).
So let n ≥ 2. How do we prove wn = wn−1 + wn−2 ?
Enumerative Combinatorics: class notes page 173

If σ is any short-legged permutation of [n], then |σ (n) − n| ≤ 1 (by the definition

of “short-legged”), so that σ (n) ∈ {n − 1, n, n + 1} and therefore either σ (n) =
n − 1 or σ (n) = n (since σ (n) has to belong to [n]). We shall call σ red if σ (n) = n,
and green if σ (n) = n − 1. Thus, each short-legged permutation of [n] is either red
or green (but never both at the same time).
Our goal will be to show that

(# of red short-legged permutations of [n]) = wn−1

and
(# of green short-legged permutations of [n]) = wn−2 .
Proposition 1.7.12 (c) allows us to identify each permutation σ of [n] with its
one-line notation (σ (1) , σ (2) , . . . , σ (n)), which is an n-tuple of distinct elements
of [n]. Let us do so.127 Thus:

• A permutation of [n] is the same as an n-tuple of distinct elements of [n] that

contains all elements of [n].

• A short-legged permutation of [n] is the same as an n-tuple of distinct ele-

ments of [n] that contains all elements of [n] and whose i-th entry is either
i − 1 or i or i + 1 for each i ∈ [n].

• A red short-legged permutation of [n] is the same as a short-legged permuta-

tion whose last entry is n.

• A green short-legged permutation of [n] is the same as a short-legged permu-

tation whose last entry is n − 1.

We similarly identify permutations of [n − 1] with certain (n − 1)-tuples, and

permutations of [n − 2] with certain (n − 2)-tuples. However, the words “red”
and “green” will be used for permutations of [n] only (so “green short-legged
permutation” will mean “green short-legged permutation of [n]”, and likewise for
red).
So we know that a red short-legged permutation of [n] is the same as a short-
legged permutation of [n] whose last entry is n. If we remove said last entry, then
we obtain an (n − 1)-tuple of elements of [n − 1], which is easily seen to be a short-
legged permutation of [n − 1]. In other words, if σ = (σ (1) , σ (2) , . . . , σ (n)) is
a red short-legged permutation of [n], then (σ (1) , σ (2) , . . . , σ (n − 1)) is a short-
legged permutation of [n − 1]. Thus, the map

{red short-legged permutations} → {short-legged permutations of [n − 1]} ,

σ 7→ (σ (1) , σ (2) , . . . , σ (n − 1))
127 Actually,if you define an n-tuple of elements of a set X to be a map from [n] to X, then there
is nothing to identify here: In this case, a permutation of [n] is already identical to its one-line
notation. But I don’t want to make assumptions about your definition of tuples.
Enumerative Combinatorics: class notes page 174

is well-defined. It is easy to see that this map is a bijection128 . Thus, the bijection
principle yields

(# of red short-legged permutations)

= (# of short-legged permutations of [n − 1])
= w n −1

(since wn−1 was defined as the # of short-legged permutations of [n − 1]).

Next, we analyze the green short-legged permutations of [n]. Each green short-
legged permutation σ is an n-tuple whose last entry is n − 1. This entails that its
second-to-last entry must be n. Indeed, σ must contain n somewhere (since it is a
permutation of [n]), but the only positions in which a short-legged permutation can
contain n are its (n − 1)-st and n-th entries (because any other position of n would
violate the “short-legged” condition), and the n-th entry is already taken by n − 1.
Thus, each green short-legged permutation σ is an n-tuple that ends with the two
entries n and n − 1 (in this order). If we remove these two entries, then we obtain
an (n − 2)-tuple of elements of [n − 2], which is easily seen to be a short-legged
permutation of [n − 2]. In other words, if σ = (σ (1) , σ (2) , . . . , σ (n)) is a green
short-legged permutation of [n], then (σ (1) , σ (2) , . . . , σ (n − 2)) is a short-legged
permutation of [n − 2]. Thus, the map

{green short-legged permutations} → {short-legged permutations of [n − 2]} ,

σ 7→ (σ (1) , σ (2) , . . . , σ (n − 2))

is well-defined. It is easy to see that this map is a bijection129 . Thus, the bijection
principle yields

(# of green short-legged permutations)

= (# of short-legged permutations of [n − 2])
= w n −2

(since wn−2 was defined as the # of short-legged permutations of [n − 2]).

Now, recall that each short-legged permutation of [n] is either red or green (but
never both at the same time). Hence, the sum rule shows that

(# of short-legged permutations of [n])

= (# of red short-legged permutations) + (# of green short-legged permutations)
| {z } | {z }
= w n −1 = w n −2
= w n −1 + w n −2 .

128 Indeed, its inverse simply appends an n to the end of a short-legged permutation of [n − 1] (where
the latter is regarded as an (n − 1)-tuple).
129 Indeed, its inverse simply appends an n and then an n − 1 to the end of a short-legged permuta-

tion of [n − 2] (where the latter is regarded as an (n − 2)-tuple).

Enumerative Combinatorics: class notes page 175

Thus, the definition of wn yields

wn = (# of short-legged permutations of [n]) = wn−1 + wn−2 .

As we already know, this quickly completes the proof of Proposition 1.7.15.

Second proof of Proposition 1.7.15 (sketched). Here is the outline of a different proof.
Recall Definition 1.4.2. I claim that the map

F : {short-legged permutations of [n]} → {lacunar subsets of [n − 1]} ,

σ 7 → { i ∈ [ n − 1] | σ ( i ) = i + 1}

is well-defined and is a bijection. Once we have proven this, the bijection principle
will yield

(# of short-legged permutations of [n])

= (# of lacunar subsets of [n − 1])
= f (n−1)+2 (by Proposition 1.4.9, applied to n − 1 instead of n)
= f n +1 ,

and thus Proposition 1.7.15 will be proven.

Why is the map F well-defined and a bijection? In order to prove that F is well-
defined, we only need to check the following statement:

Statement 1: If σ is a short-legged permutation of [n], then

{i ∈ [n − 1] | σ (i ) = i + 1} is a lacunar subset of [n − 1].

I leave this to you, dear reader, with the following hint: If σ is a short-legged
permutation, and some a ∈ [n − 2] satisfies σ ( a) = a + 1 and σ ( a + 1) = a + 2, then
prove that σ ( a + 2) = a + 3 (because the values a + 1 and a + 2 are already taken)
and σ ( a + 3) = a + 4 (because the values a + 2 and a + 3 are already taken) and
σ ( a + 4) = a + 5 (likewise) and so on... but this eventually leads to σ (n) = n + 1,
which is absurd.
Statement 1 shows that F is well-defined. In order to show that F is bijective, we
need to prove the following two facts:

Statement 2: If σ and τ are two short-legged permutations of [n] satisfy-

ing F (σ) = F (τ ), then σ = τ.

Statement 3: If J is a lacunar subset of [n − 1], then there exists a short-

legged permutation σ of [n] satisfying F (σ) = J.

To prove Statement 2, you need to find a way to reconstruct a short-legged per-

mutation σ from the lacunar subset F (σ). Here is such a way: The permutation
Enumerative Combinatorics: class notes page 176


 j + 1, if j ∈ F (σ) ;

σ will send each j ∈ [n] to j − 1, if j − 1 ∈ F (σ ) ; . Of course, this needs to be

j, otherwise

proven – again, LTTR. Same applies for Statement 3.
Alternatively, this proof can be restated in terms of domino tilings of Rn,2 . Indeed, we
already know from the Second proof of Proposition 1.4.9 that there is a bijection
h : {domino tilings of Rn+1,2 } → {lacunar subsets of [n]} .
Applying this to n − 1 instead of n, we obtain a bijection
h0 : {domino tilings of Rn,2 } → {lacunar subsets of [n − 1]} .
Composing the inverse of this bijection h0 with the bijection F above, we obtain a bijection
{short-legged permutations of [n]} → {domino tilings of Rn,2 } .
We leave it to the reader to describe this bijection more directly and prove its bijectivity
without the detour through lacunar subsets.

1.7.6. Long-legged permutations

If short-legged permutations were so easy to count, what about the “opposite”
notion?

Definition 1.7.16. Let n ∈ N. A permutation σ of [n] is said to be long-legged if

each i ∈ [n] satisfies |σ (i ) − i | > 1.

How many long-legged permutations does [n] have?

As in Subsection 1.4.3, we can compute these #s using SageMath for small n. As
in Subsection 1.7.3, we can then enter the results into the OEIS and hope that the
sequence will be recognized. Here is the SageMath code:130
def is_longlegged ( s ) :
# Check whether a permutation ‘s ‘ is long - legged .
n = len ( s )
# This finds the ‘n ‘ such that
# ‘s ‘ is a permutation of ‘[ n ] ‘.
return all ( abs ( s ( i ) - i ) > 1 for i in range (1 , n + 1 ) )

def num _of_long legged ( n ) :

# Return the number of long - legged permutations
# of ‘[ n ] ‘.
return sum ( 1 for s in Permutations ( n ) if is_longlegged ( s ) )

for n in range ( 10 ) :
print ( " The number of long - legged permutations of [ " + str ( n ) + " ] is
" + str ( num_ of_longl egged ( n ) ) )

130 Thiscomputes the #s for n ∈ {0, 1, . . . , 9}. You can try to go further, but keep in mind that the #
of permutations of [n] is n!, which grows a lot faster than 2n (the # of subsets of [n]).
Enumerative Combinatorics: class notes page 177

Its output is:

The number of long-legged permutations of [0] is 1
The number of long-legged permutations of [1] is 0
The number of long-legged permutations of [2] is 0
The number of long-legged permutations of [3] is 0
The number of long-legged permutations of [4] is 1
The number of long-legged permutations of [5] is 4
The number of long-legged permutations of [6] is 29
The number of long-legged permutations of [7] is 206
The number of long-legged permutations of [8] is 1708
The number of long-legged permutations of [9] is 15702
So we enter these values 1, 0, 0, 0, 1, 4, 29, 206, 1708, 15702 into OEIS and instantly
obtain the result: OEIS sequence A001883, described as “Number of permutations
s of {1,2,...,n} such that |s(i)-i|>1 for each i=1,2,...,n”, which clears any doubts as to
whether it is the right sequence.
In the “Formula” section of the OEIS page, we spot the following recursion:
a(n) = n*a(n-1) + 4*a(n-2) - 3*(n-3)*a(n-3) + (n-4)*a(n-4)
+ 2*(n-5)*a(n-5) - (n-7)*a(n-6) - a(n-7).
In other words:

Proposition 1.7.17. For each n ∈ N, let

an = (# of long-legged permutations of [n]) .

Then, for each integer n ≥ 7, we have

an = nan−1 + 4an−2 − 3 (n − 3) an−3 + (n − 4) an−4

+ 2 ( n − 5 ) a n −5 − ( n − 7 ) a n −6 − a n −7 .

This recurrence is credited to Vaclav Kotesovec (2017), and probably has been
proved by reducing it to another recurrence (by J. Riordan, in an unpublished Bell
Labs memorandum from 1963), which involves the Lucas sequence (a variant of
the Fibonacci sequence: same recursion, but different starting values). The latter
recurrence has been proved (in said memorandum, which can be downloaded from
the OEIS page) using generating functions and an enumerative technique known as
rook theory (see [BCHR11] for an introduction). The argument would be both too
advanced and too intricate to present at this point.
Thus ends our introductory tour of enumerative combinatorics. We shall next
focus on one of the most basic topics: the binomial coefficients.
Enumerative Combinatorics: class notes page 178

2. Binomial coefficients
We have already started exploring the binomial coefficients in Section 1.3. In this
chapter, we shall explore them further, with a focus on proving identities involving
them. We will build up a toolbox of techniques for this, some combinatorial and
some algebraic.

2.1. The alternating sum of a row of Pascal’s triangle

We have previously proved Proposition 1.3.28 using the binomial formula. Let us
now give two more proofs for it, as they involve techniques worth seeing.

2.1.1. Telescoping sums

For the second proof, we will use the telescope principle – or, more precisely, the
telescoping sum principle (as there is an analogous version for products):

Theorem 2.1.1 (Telescoping sum principle). Let u and v be two integers with
u ≤ v + 1. Let au , au+1 , . . . , av+1 be any numbers. Then,
v
∑


a j +1 − a j = a v +1 − a u .
j=u

First proof of Theorem 2.1.1 (informal version). If u = v + 1, then

v v
∑ ∑



a j +1 − a j = a j+1 − a j = (empty sum) = 0
j=u j = v +1
= au − a u = a v +1 − a u .
|{z}
= a v +1
(since u=v+1)

Hence, Theorem 2.1.1 is proved in the case when u = v + 1. Thus, for the rest of
this proof, we WLOG assume that u 6= v + 1. Hence, u < v + 1 (since u ≤ v + 1),
so that u + 1 ≤ v + 1 (since u and v + 1 are integers). In other words, u ≤ v.
Now, here is an informal version of how the proof proceeds from here: We have
v
∑ | a j+1{z− a j }


j=u
=− a j + a j+1
v
∑


= − a j + a j +1
j=u
= (− au + au+1 ) + (− au+1 + au+2 ) + (− au+2 + au+3 ) + · · · + (− av + av+1 ) .
Enumerative Combinatorics: class notes page 179

The right hand side of this equality is a sum of 2 · (v − u + 1) addends (once ex-
panded out), but all but 2 of them cancel out (indeed, the second addend in each
parenthesis cancels the first addend in the next parenthesis). The 2 addends left
over are − au and av+1 . Thus, the sum “contracts” like a telescope (thus the name
“telescoping sum principle”):

(− au + au+1 ) + (− au+1 + au+2 ) + (− au+2 + au+3 ) + · · · + (− av + av+1 )

= − a u + a v +1 = a v +1 − a u .
Combining, we obtain
v
∑


a j +1 − a j
j=u
= (− au + au+1 ) + (− au+1 + au+2 ) + (− au+2 + au+3 ) + · · · + (− av + av+1 )
= a v +1 − a u .
This proves Theorem 2.1.1.
You might not like how we waved our hands talking of cancelling addends in the proof
above. But formalizing the proof is a simple matter of reframing it as an induction argu-
ment:

First proof of Theorem 2.1.1 (formal version). From u ≤ v + 1, we obtain v + 1 − u ≥ 0 and

thus v + 1 − u ∈ N. Hence, we can prove Theorem 2.1.1 by induction on v + 1 − u:
Induction base: It is easy to see that Theorem 2.1.1 holds if v + 1 − u = 0 131 . This

completes the induction base.

Induction step: Let m ∈ N. Assume (as the induction hypothesis) that Theorem 2.1.1
holds if v + 1 − u = m. We must prove that Theorem 2.1.1 holds if v + 1 − u = m + 1.
We have assumed that Theorem 2.1.1 holds if v + 1 − u = m. In other words, the follow-
ing claim holds:

Claim 1: Let u and v be two integers with u ≤ v + 1 and v + 1 − u = m. Let

au , au+1 , . . . , av+1 be any numbers. Then,
v
∑


a j +1 − a j = a v +1 − a u .
j=u

We must prove that Theorem 2.1.1 holds if v + 1 − u = m + 1. In other words, we must

prove the following claim:
131 Proof.
Let u, v and au , au+1 , . . . , av+1 be as in Theorem 2.1.1, and assume that v + 1 − u = 0. Thus,
u = v + 1 (since v + 1 − u = 0). Therefore,
v v
∑ ∑



a j +1 − a j = a j+1 − a j = (empty sum) = 0 = au − a u = a v +1 − a u .
|{z}
j=u j = v +1 = a v +1
(since u=v+1)

This is precisely the claim of Theorem 2.1.1. Hence, we have showed that Theorem 2.1.1 holds if
v + 1 − u = 0.
Enumerative Combinatorics: class notes page 180

Claim 2: Let u and v be two integers with u ≤ v + 1 and v + 1 − u = m + 1. Let

au , au+1 , . . . , av+1 be any numbers. Then,
v
∑


a j +1 − a j = a v +1 − a u .
j=u

[Proof of Claim 2: We have v − u = (v + 1 − u) −1 = (m + 1) − 1 = m ≥ 0 (since m ∈ N)

| {z }
= m +1
v
and thus v ≥ u. Hence, we can split off the addend for j = v from the sum ∑ a j+1 − a j .


j=u
We thus obtain
v v −1
∑ a j +1 − a j = ( a v +1 − a v ) + ∑ a j +1 − a j .



(136)
j=u j=u

But u ≤ (v − 1) + 1 (since (v − 1) + 1 = v ≥ u) and (v − 1) + 1 − u = v − u = m. Hence,

we can apply Claim 1 to v − 1 instead of v. We thus obtain
v −1
∑


a j+1 − a j = a(v−1)+1 − au = av − au .
j=u | {z }
= av

Thus, (136) becomes

v v −1
∑ a j +1 − a j = ( a v +1 − a v ) + ∑ a j +1 − a j = ( a v +1 − a v ) + ( a v − a u ) = a v +1 − a u .



j=u j=u
| {z }
= av − au

This proves Claim 2.]

Now, we have proved Claim 2. In other words, we have proved that Theorem 2.1.1 holds
if v + 1 − u = m + 1. This completes the induction step. Thus, Theorem 2.1.1 is proved by
induction.

There is also a different proof of Theorem 2.1.1, relying on basic manipulations of sums:

Second proof of Theorem 2.1.1. We have

v v v v +1 v
∑ ∑ a j +1 − ∑ a j = ∑ ∑ aj


a j +1 − a j = aj − (137)
j=u j=u j=u j = u +1 j=u

(here, we have substituted j for j + 1 in the first sum).

Now, it is tempting to argue that
v +1 v v v
∑ a j = a v +1 + ∑ aj and ∑ a j = au + ∑ aj
j = u +1 j = u +1 j=u j = u +1

(and then subtract these two equalities). Unfortunately, these two equalities are false in the
border case u = v + 1, because you cannot split off a (non-existing) addend from an empty
sum. Of course, we could get around this issue by treating this border case separately
Enumerative Combinatorics: class notes page 181

(it is trivial). But it is a bit easier to make a similar argument that avoids this nuisance
v +1 v
altogether: Instead of making the sums ∑ a j and ∑ a j equal by splitting off an addend
j = u +1 j=u
from each, we make them equal by adding an extra addend to each. This can be done as
follows:
v +1
We have u ≤ v + 1. Thus, we can split off the addend for j = v + 1 from the sum ∑ a j ,
j=u
and obtain
v +1 v
∑ a j = a v +1 + ∑ a j .
j=u j=u

Thus,
v v +1
∑ a j = ∑ a j − a v +1 .
j=u j=u

v +1
But we can also split off the addend for j = u from the sum ∑ a j , and obtain
j=u

v +1 v +1
∑ a j = au + ∑ aj.
j=u j = u +1

Thus,
v +1 v +1
∑ aj = ∑ a j − au .
j = u +1 j=u

Hence, (137) becomes

v v +1 v
∑ ∑ ∑ aj


a j +1 − a j = aj −
j=u j = u +1 j=u
| {z } | {z }
v +1 v +1
= ∑ a j − au = ∑ a j − a v +1
j=u j=u
! !
v +1 v +1
= ∑ a j − au − ∑ a j − a v +1 = a v +1 − a u .
j=u j=u

This proves Theorem 2.1.1 again.

Remark 2.1.2. Equivalent versions of Theorem 2.1.1 abound all across the literature. For
example, [19f-hw0s, Proposition 2.2] says that if m ∈ N and if a0 , a1 , . . . , am are any m + 1
numbers, then
m
∑ ( a i − a i −1 ) = a m − a 0 .
i =1
m
This follows from Theorem 2.1.1, because substituting j + 1 for i in the sum ∑ ( ai − ai−1 )
i =1
Enumerative Combinatorics: class notes page 182

yields
m m −1
∑ ( a i − a i −1 ) = ∑


a j+1 − a j = a(m−1)+1 − a0
i =1 j =0 | {z }
= am
(by Theorem 2.1.1, applied to u = 0 and v = m − 1)
= a m − a0 .

Also, [Grinbe15, (16)] says that if u and v are two integers such that u − 1 ≤ v, and as is
a number for each s ∈ {u − 1, u, . . . , v}, then
v
∑ ( a s − a s −1 ) = a v − a u −1 .
s=u

This is precisely Theorem 2.1.1, applied to a j−1 instead of a j .

Let us note that Theorem 2.1.1 is a discrete version of the famous “Second Fundamental
Theorem of Calculus”, which says that any two real numbers u and v with u ≤ v and any
differentiable function F : [u, v]R → R satisfy132
Z v
F 0 ( x ) dx = F (v) − F (u) .
u

Indeed, the summation sign ∑ is a discrete analogue of the integral sign , whereas the
R

differences a j+1 − a j of consecutive entries of the sequence au , au+1 , . . . , av+1 are a discrete
analogue of (the values of) the derivative of a function.

We can now re-prove Proposition 1.3.28:

Second proof of Proposition 1.3.28. If n = 0, then Proposition 1.3.28 boils down to the
identity 1 = 1, which is true. Thus, for the rest of this proof, we WLOG assume
that n 6= 0. Hence, [n = 0] = 0 and n ≥ 1 (since n ∈ N). From n ≥ 1, we obtain
n−1 ∈  N. Hence, Proposition 1.3.6 (applied to n − 1 and n instead of n and k)
n−1

yields = 0 (since n − 1 < n). Now, for each k ∈ N, we have
n
 
k n
(−1)
k
|{z }
n−1 n−1
 
= +
k−1 k
(by Theorem 1.3.8)
n−1 n−1 n−1 k n−1
       
k k
= (−1) + = (−1) + (−1)
k−1 k | {z } k−1 k
=−(−1)k−1
n−1 n−1 n−1 k −1 n − 1
       
k −1
= − (−1) + (−1)k = (−1) k
− (−1) .
k−1 k k k−1
132 Here, [u, v]R denotes the real interval from u to v.
Enumerative Combinatorics: class notes page 183

Hence,
n  
n
∑ k
(−1)
k
k =0 | {z } 
n−1 n−1
  
k k −1
=(−1) −(−1)
k k−1
n 
k n−1 k −1 n − 1
   
= ∑ (−1) − (−1)
k =0
k k−1
n −1
j +1 n − 1 j n−1
    
= ∑ (−1) − (−1)
j=−1
j + 1 j
(here, we have substituted j + 1 for k in the sum)
n−1 n−1
   
(n−1)+1 −1
= (−1) − (−1)
( n − 1) + 1 −1
|  {z  } | {z }
n−1 =0
= =0 (by (43), applied to n−1 and −1
n instead of n and k)

j n−1
  
by Theorem 2.1.1, applied to u = −1, v = n − 1 and a j = (−1)
j
= 0 − 0 = 0 = [ n = 0] (since [n = 0] = 0) .

This proves Proposition 1.3.28 again.

When applying Theorem 2.1.1, one usually does not directly mention Theorem
v
2.1.1; instead, one says that the sum ∑ a j+1 − a j “telescopes to” av+1 − au . Thus,


j=u
for example, inthe proof
 of Proposition 1.3.28 we just gave, the sum 
n −1
j +1 n − 1 j n−1 n−1
   
(n−1)+1
∑ (−1) − (−1) telescoped to (−1) −
j=−1 j+1 j ( n − 1) + 1
−1 n − 1
 
(−1) .
−1
Exercise 2.1.1. (a) Let n ∈ R. Let m ∈ N. Prove that
m
m n−1
   
k n
∑ (−1) k = (−1) m
.
k =0

(b) Explain why Proposition 1.3.28 is a particular case of Exercise 2.1.1 (a).

Let us illustrate the telescoping sum principle on another application: a new

proof of the hockey-stick identity (Theorem 1.3.29):
Enumerative Combinatorics: class notes page 184

Second proof of Theorem 1.3.29. For each j ∈ Z, we have

( j + 1) − 1 ( j + 1) − 1
     
j+1
= +
k+1 ( k + 1) − 1 k+1
 
by Theorem 1.3.8, applied to j + 1 and k + 1
instead of n and k
   
j j
= + (since ( j + 1) − 1 = j and (k + 1) − 1 = k)
k k+1
and therefore      
j j+1 j
= − . (138)
k k+1 k+1
But n ∈ N, thus 0 ≤ n ≤ n + 1. Also, k ∈ N, thus k + 1 > 0 and therefore
k + 1 6= 0. Hence,  [k + 1 = 0] = 0. Now, Lemma 1.3.14 (applied to k + 1 instead of
0
k) yields = [k + 1 = 0] = 0.
k+1
But
       
0 1 2 n
+ + +···+
k k k k
n   n        
j j+1 j n+1 0
=∑ =∑ − = −
j =0
k j =0
k+1 k+1 k+1 k+1
|{z} | {z }
=0
   
j+1 j
= −
k+1 k+1
(by (138))
  
j
by Theorem 2.1.1, applied to u = 0 and v = n and a j =
k+1
 
n+1
= .
k+1
This proves Theorem 1.3.29 again.

2.1.2. A war between the odd and the even

Proposition  1.3.28
 does not seem vulnerable to bijective proof strategies: The sum
n n
k
∑ (−1) contains both positive and negative numbers, and of course negative
k =0 k
numbers cannot be interpreted as sizes of sets. Nevertheless, it turns out that there
is a bijective proof of Proposition 1.3.28, and a rather slick and simple one:
Third proof of Proposition 1.3.28 (sketched). As in the Second proof of Proposition 1.3.28
above, we see that Proposition 1.3.28 holds for n = 0. Thus, we WLOG assume that
n 6= 0. Hence, [n = 0] = 0. We thus need to show that
n  
k n
∑ (−1) k = 0.
k =0
Enumerative Combinatorics: class notes page 185

But the left hand side of this equality is

n    
k n k n
∑ (−1) k = ∑ (−1) k
k =0 k∈{0,1,...,n}
   
n n
= ∑ (−1) k
k
+ ∑ (−1) k
k
k∈{0,1,...,n}; k∈{0,1,...,n};
| {z } | {z }
k is even
=1 k is odd
=−1
(since k is even) (since k is odd)
 
since each k ∈ {0, 1, . . . , n} is either even or odd,
but not both at the same time
   
n n
= ∑ k
+ ∑ (−1)
k
k∈{0,1,...,n}; k∈{0,1,...,n};
k is even k is odd
| {z  }
n
=− ∑
k∈{0,1,...,n}; k
k is odd
   
n n
= ∑ k
− ∑ k
.
k∈{0,1,...,n}; k∈{0,1,...,n};
k is even k is odd

Thus, we only need to prove that

   
n n
∑ k
= ∑ k
. (139)
k∈{0,1,...,n}; k∈{0,1,...,n};
k even k odd

We can prove this bijectively:

The left-hand side of (139) is the # of subsets of [n] of even size133 . The right-
hand side of (139) is the # of subsets of [n] of odd size134 . Hence, the equality (139)
133 Proof.We have |[n]| = n. If B is any subset of [n], then Theorem 1.4.7 (b) (applied to A = [n])
yields | B| ≤ |[n]| = n and therefore | B| ∈ {0, 1, . . . , n}. Furthermore, if B is any subset of [n] of
even size, then | B| is an even element of {0, 1, . . . , n} (because we just saw that | B| ∈ {0, 1, . . . , n},
but we also know now that | B| is even). Hence, the sum rule yields

(# of subsets of [n] of even size)

= ∑ (# of subsets B of [n] satisfying | B| = k)
k ∈{0,1,...,n};
| {z }
k is even =(# of k-element
 subsets of [ n ])
n
=
k
(by Theorem 1.3.12,
applied to S=[n])
 
n
= ∑ k
= (the left-hand side of (139)) .
k ∈{0,1,...,n};
k is even

134 for an analogous reason

Enumerative Combinatorics: class notes page 186

is equivalent to

(# of subsets of [n] of even size) = (# of subsets of [n] of odd size) .

This will follow immediately from the bijection principle, once we can find a bijec-
tion
{subsets of [n] of even size} → {subsets of [n] of odd size} .
Here is such a bijection:

{subsets of [n] of even size} → {subsets of [n] of odd size} ,

(
A ∪ {1} , if 1 ∈
/ A;
A 7→ .
A \ {1} , if 1 ∈ A

We need to actually convince ourselves that this map is well-defined135 and actually
a bijection136 . Thus, (139) is proved. As we have said, this entails Proposition
1.3.28.
The bijection we have used in this proof can be rewritten in an even simpler way
using a basic set-theoretical operation (almost as basic as union and intersection of
sets), called the symmetric difference:

Definition 2.1.3. Let X and Y be two sets. Then, the symmetric difference of X and
Y is the set

{all elements that lie in exactly one of the two sets X and Y }
= ( X \ Y ) ∪ (Y \ X ) = ( X ∪ Y ) \ ( X ∩ Y ) .
(
135 This A ∪ {1} , if 1 ∈
/ A;
means proving that if A is any subset of [n] of even size, then is a
A \ {1} , if 1 ∈ A
subset of [n] of odd size. This relies on the following facts:
• We have 1 ∈ [n] (since n 6= 0).
• If 1 ∈
/ A, then | A ∪ {1}| = | A| + 1 is odd (since | A| is even).
• If 1 ∈ A, then | A \ {1}| = | A| − 1 is odd (since | A| is even).

136 Toprove this, we need to find an inverse to it. This is easy: The inverse is given by the exact same
formula! That is, the inverse is the map

{subsets of [n] of odd size} → {subsets of [n] of even size} ,

(
A ∪ {1} , if 1 ∈
/ A;
A 7→ .
A \ {1} , if 1 ∈ A

In order to see that this is really an inverse to our map, we need to show the following:
• If 1 ∈
/ A, then 1 ∈ A ∪ {1} and ( A ∪ {1}) \ {1} = A.
• If 1 ∈ A, then 1 ∈
/ A \ {1} and ( A \ {1}) ∪ {1} = A.
Of course, both of these claims are completely obvious.
Enumerative Combinatorics: class notes page 187

This set is denoted by X 4 Y.

The Venn diagram for X 4 Y is

(where the two circles represent X and Y).

Example 2.1.4. We have {1, 2, 3, 4} 4 {3, 4, 5, 6} = {1, 2, 5, 6}.

The notion of symmetric difference has the following properties (which are all
straightforward to check using Venn diagrams or elementwise verification):

Proposition 2.1.5. (a) We have X 4 X = ∅ for any set X.

(b) We have X 4 ∅ = X for any set X.
(c) We have X 4 Y = Y 4 X for any two sets X and Y.
(d) We have ( X 4 Y ) 4 Z = X 4 (Y 4 Z ) for any three sets X, Y and Z.
(e) We have X ∩ (Y 4 Z ) = ( X ∩ Y ) 4 ( X ∩ Z ) for any three sets X, Y and Z.
(f) If X and Y are two finite sets, then | X 4 Y | = | X | + |Y | − 2 | X ∩ Y | ≡
| X | + |Y | mod 2.

Using symmetric differences, we can now rewrite the bijection that we used in
the above Third proof of Proposition 1.3.28 as

{subsets of [n] of even size} → {subsets of [n] of odd size} ,

A 7 → A 4 {1} .

This makes it even easier to check that this map is well-defined and a bijection.

2.2. The trinomial revision formula

2.2.1. An algebraic proof
Recall the trinomial revision formula (Proposition 1.3.35), which states that

n−b
     
n a n
= for all n, a, b ∈ R.
a b b a−b

We have previously stated this without proof; let us prove it now:

Proof of Proposition 1.3.35. We must be in one of the following four cases:
Case 1: We have b ∈ / N.
Case 2: We have b ∈ N and a ∈ / N.
Enumerative Combinatorics: class notes page 188

Case 3: We have b ∈ N and a ∈ N but a < b.

Case 4: We have b ∈ N and a ∈ N and a ≥ b.
 we have b ∈
Let us first consider Case 1. In this case, / N. Hence, (43) (applied
a
to a and b instead of n and k) yields = 0. Also, (43) (applied to k = b) yields
  b
n
= 0. Now, comparing
b

n−b
     
n a n
=0 with = 0,
a b b a−b
|{z} | {z }
=0 =0

n−b
     
n a n
we obtain = . Thus, Proposition 1.3.35 is proven in Case 1.
a b b a−b
Let us now consider Case  2. In this case, we have b ∈ N and a ∈ / N. Hence, (43)
n
(applied to k = a) yields = 0. Also, a − b ∈/ N 137 . Hence, (43) (applied to
a
n−b
 
n − b and a − b instead of n and k) yields = 0. Now, comparing
a−b

n−b
     
n a n
=0 with = 0,
a b b a−b
| {z } | {z }
=0 =0

n−b
     
n a n
we obtain = . Thus, Proposition 1.3.35 is proven in Case 2.
a b b a−b
Let us now consider Case 3. In this case, we have b ∈ N and a ∈ N but a < b.
From a < b, we obtain a − b < 0, sothat a −b ∈/ N. Hence, (43) (applied to n − b
n−b
and a − b instead of n and k) yields = 0. Also, b > a (since a < b). Hence,
a−b  
a
Proposition 1.3.6 (applied to a and b instead of n and k) yields = 0. Now,
b
comparing
n−b
     
n a n
=0 with = 0,
a b b a−b
|{z} | {z }
=0 =0
n−b
     
n a n
we obtain = . Thus, Proposition 1.3.35 is proven in Case 3.
a b b a−b
Let us finally consider Case 4. In this case, we have b ∈ N and a ∈ N and a ≥ b.
From this, we conclude that a − b ∈ N and b ≤ a. Hence, Theorem 1.3.9 (applied
137 Proof. Assume the contrary. Thus, a − b ∈ N. Hence, a = a − b + |{z}
b ∈ N (since the sum of two
| {z }
∈N ∈N
elements of N is always an element of N), which contradicts a ∈
/ N. Hence, we have obtained a
contradiction. This shows that our assumption was false, qed.
Enumerative Combinatorics: class notes page 189

to a and b instead of n and k) yields

 
a a!
= . (140)
b b! · ( a − b)!
Also, (42) (applied to k = a) yields
n ( n − 1) ( n − 2) · · · ( n − a + 1)
 
n
= .
a a!
Multiplying this equality by (140), we obtain
n ( n − 1) ( n − 2) · · · ( n − a + 1)
  
n a a!
= ·
a b a! b! ( a − b)!
n ( n − 1) ( n − 2) · · · ( n − a + 1)
= . (141)
b! · ( a − b)!
On the other hand, (42) (applied to k = b) yields
n ( n − 1) ( n − 2) · · · ( n − b + 1)
 
n
= .
b b!
Furthermore, (42) (applied to n − b and a − b instead of n and k) yields
n−b (n − b) (n − b − 1) (n − b − 2) · · · ((n − b) − ( a − b) + 1)
 
=
a−b ( a − b)!
( n − b ) ( n − b − 1) ( n − b − 2) · · · ( n − a + 1)
= .
( a − b)!
Multiplying these two equalities, we find
n−b
  
n
b a−b
n ( n − 1) ( n − 2) · · · ( n − b + 1) ( n − b ) ( n − b − 1) ( n − b − 2) · · · ( n − a + 1)
= ·
b! ( a − b)!
(n (n − 1) (n − 2) · · · (n − b + 1)) · ((n − b) (n − b − 1) (n − b − 2) · · · (n − a + 1))
=
b! · ( a − b)!
n ( n − 1) ( n − 2) · · · ( n − a + 1)
=
b! · ( a − b)!
(because
(n (n − 1) (n − 2) · · · (n − b + 1)) · ((n − b) (n − b − 1) (n − b − 2) · · · (n − a + 1))
= n ( n − 1) ( n − 2) · · · ( n − a + 1)
since 0 ≤ b ≤ a, and thus the
 
 product n (n − 1) (n − 2) · · · (n − a + 1) can be split into 
 
 (n (n − 1) (n − 2) · · · (n − b + 1)) 
· ((n − b) (n − b − 1) (n − b − 2) · · · (n − a + 1))
Enumerative Combinatorics: class notes page 190

). Comparing this with (141), we obtain

n−b
     
n a n
= .
a b b a−b

Thus, Proposition 1.3.35 is proven in Case 4.

We have now proved Proposition 1.3.35 in each of the four Cases 1, 2, 3 and 4.
Hence, Proposition 1.3.35 always holds.
It is worth taking a second look at the structure of our above proof of Proposition
1.3.35: This proof was subdivided into four cases. In the first three of these cases,
the proposition boiled down to 0 = 0 (but for different reasons in each case). Only
in the fourth case did we end up proving anything interesting. This suggests that
the generality in which we stated Proposition 1.3.35 (allowing all of n, a, b to be ar-
bitrary reals) was perhaps more of a handicap than a helpful feature. Nevertheless,
this generality pays off: We only need to prove the proposition once, and then we
get to use it as often as we want, so a few extra cases are worth the trouble.
As an application of Proposition 1.3.35, let us prove a slight generalization of the
absorption formula (Proposition 1.3.36):

Corollary 2.2.1 (Absorption formula II). Let n ∈ R \ {0} and m ∈ R. Then,

m m−1
   
m
= .
n n n−1

Proof of Corollary 2.2.1. Proposition 1.3.35 (applied to m, n and 1 instead of n, a and

b) yields
m−1
     
m n m
= . (142)
n 1 1 n−1
   
n m
But we have = n (by (45)) and = m (likewise). Hence, the formula (142)
1 1
rewrites as
m−1
   
m
n=m .
n n−1
Dividing both sides of this equality by n, we find

m−1
 
m
m m−1
   
m n−1
= = .
n n n n−1

This proves Corollary 2.2.1.

Enumerative Combinatorics: class notes page 191

Exercise 2.2.1. Let n ∈ Q, a ∈ N and b ∈ N.

(a) Prove that every integer j ≥ a satisfies

n−j n−a n−a−b

       
n j n
= .
j a b a b j−a

n−j
   
n n j
(b) Compute the sum ∑ for every integer n ≥ a. (The result
j= a j a b
should contain no summation signs.)

Class of 2019-10-18

2.2.2. A double counting proof

Let us give a different proof of Proposition 1.3.35 – or, more precisely, a partial
proof, since it will only work in the case when n ∈ N. (In Subsection 2.6.4, we
will learn a trick that will allow us to extend this proof to a complete proof of
Proposition 1.3.35;
 essentially,we
will learn
 a way to (rigorously) argue that if an
n−b
 
n a n
identity such as = holds for all n ∈ N, then it must also
a b b a−b
hold for all n ∈ R, and therefore it suffices to prove it for n ∈ N only. This is called
the “polynomial
  identity trick”,
 and,  in a nutshell, the thing that makes it work is
n−b

n a n
that both and are polynomials in n. But for now, let us just
a b b a−b
prove Proposition 1.3.35 in the case when n ∈ N.)
Let us first present our partial proof in an informal way:
Second proof of Proposition 1.3.35 for n ∈ N (informal version). Assume that n ∈ N.
Fix a set N of n people.
We consider pairs ( A, B), where A is a committee consisting of a people from N
(that is, an a-element subset of N) and where B is a subcommittee consisting of b
people from this committee A (that is, a b-element subset of A). 138 Formally

speaking, these are simply pairs ( A, B) of sets satisfying B ⊆ A ⊆ N and | A| = a

138 Here is a symbolic picture of this situation:

B A

.
Enumerative Combinatorics: class notes page 192

and | B| = b. How many such pairs ( A, B) are there? We shall refer to such pairs as
CS pairs139 , and we shall count their # in two ways140 :
First way: In order to construct a CS pair ( A, B), we
 first choose the committee A
n
and then choose its subcommittee B. We have many options for A (because
a
A hastobe an a-element subset of the n-element set N), and after choosing A, we
a
have many options for B (because B has to be a b-element subset of A). Thus,
b    
n a
the total # of CS pairs is a sum of many copies of . In other words, this #
   a b
n a
is .
a b
Second way: In order to construct a CS pair ( A, B), wefirst  choose the subcom-
n
mittee B and then choose the committee A. We have many options for B
b
(because B has to be a b-element subset of the n-element set N). After choosing B,
how many options do we have for A ? The committee A has to be an a-element
subset of N, but it is also required to contain the (already chosen) subcommittee
B as a subset, so that b of its a elements are already decided. We only need to
choose the remaining a − b elements of A. These elements have to come from the
(n − b)-element set N \ B (because the elements of B have already been chosen to
lie in A). Thus, we are choosing an ( a − b)-element subset of the (n − b)-element
n−b

set N \ B. Thus, we have many options for choosing A. Hence, the total
a − b
n−b
   
n
# of CS pairs is a sum of many copies of . In other words, this # is
b a−b
n−b
  
n
.
b a−b
Now, we have computed
   the # of CS pairs in two different ways.  The first way
n−b
 
n a n
gave us the result , while the second way gave us the result .
a b b a−b
n−b
     
n a n
Comparing these results, we find = . Thus, Proposition
a b b a−b
1.3.35 is proven (in the case when n ∈ N).
How can we formalize this argument? It is clear how to formalize the notion of
a CS pair (just replace “n people” by n mathematical objects, such as the numbers
1, 2, . . . , n). But how do we formalize our two ways of counting the # of CS pairs?
Let us look at the first way. On its surface, it may seem like we have applied
n a
the product rule (Theorem 1.1.4) in order to obtain (similarly to how we
a b
applied it back in the proof of Proposition 1.4.14). But this appearance is mislead-
139 short for “committee-subcommittee pairs”
140 We will be using Theorem 1.3.12 in both of these ways.
Enumerative Combinatorics: class notes page 193

  
n a
ing. If we wanted to apply Theorem 1.1.4, we would have to interpret
a b
as | X | · |Y | for two sets X and Y. Thus, we would need two sets X and Y such
that the CS pairs are elements of X × Y (or at least are in obvious bijection to
the latter141 ). The obvious choice would be X = { a-element subsets of N } and
Y = {b-element subsets of A}, but this does not work: The set Y needs to be de-
fined independently of A; but here it clearly is not. This is no surprise: If we choose
a CS pair ( A, B) by first choosing A, then the options available for B will depend
on which A we have chosen (because we must have B ⊆ A). We are inluck  because
a
the number of these options does not depend on A (but always is ); but the
b
set of these options does depend on A. Theorem 1.1.4 is not general enough to
accommodate this situation.
There are two ways to resolve this problem. One is to generalize Theorem 1.1.4
to cover dependent choices. This results in a “dependent product rule”, which
(informally) says the following:

“Dependent product rule” for two sets: Consider a situation in which

you have to make two choices (sequentially). Assume that you have a1
options available in choice 1, and then (after making choice 1) you have
a2 options available in choice 2 (no matter which option you chose in
choice 1). Then, the total # of ways to make both choices142 is a1 a2 .

To formalize this, we can let X be the set of all available options in choice 1, and
let Y be the set of all options that may be available in choice 2 (including the ones
whose availability depends on what you chose in choice 1), and let S be the set of
all ways to make both choices. Then, our “dependent product rule” turns into the
following rigorous fact:

Theorem 2.2.2. Let a1 and a2 be two numbers. Let X and Y be finite sets. Let S
be a subset of X × Y. Assume that | X | = a1 . Assume further that for each x ∈ X,
there are precisely a2 many elements y ∈ Y such that ( x, y) ∈ S (even though the
set of these y may depend on x). Then, |S| = a1 a2 .

We leave the proof of this theorem to the reader, as we will not use it:

Exercise 2.2.2. Prove Theorem 2.2.2.

More generally, Theorem 1.5.3 can be generalized to cover several dependent

choices:

141 We are using some basic combinatorial slang here: We say that objects of some type are in bi-
jection with objects of another type if there exists a bijection from {objects of the first type} to
{objects of the second type}.
142 i.e., of pairs ( x, y ), where x is an option you can choose in choice 1, and y is an option you can

choose in choice 2 after choosing x in choice 1

Enumerative Combinatorics: class notes page 194

“Dependent product rule” for n sets: Consider a situation in which

you have to make n choices (sequentially). Assume that you have a1
options available in choice 1, and then (after making choice 1) you have
a2 options available in choice 2 (no matter which option you chose in
choice 1), and then (after both choices 1 and 2) you have a3 options
available in choice 3 (no matter which options you chose in choices 1
and 2), and so on. Then, the total # of ways to make all n choices is
a1 a2 · · · a n .

Formal versions of this rule can be found in [Loehr11, §1.8] and in [Newste19,
Theorem 7.2.19].
However, all of this is entirely optional, because there is a simpler way to for-
malize our proof of Proposition 1.3.35. Instead of adapting the product rule, we
can simply apply the sum rule (more precisely, Theorem 1.2.5) and observe that all
addends of the resulting sum are equal (which means that the sum simplifies to a
product). This leads to the following proof:
Second proof of Proposition 1.3.35 for n ∈ N (formal version). Let N = [n]. Thus, N is
an n-element set.
A CS pair shall mean a pair ( A, B) of sets satisfying B ⊆ A ⊆ N and | A| = a and
| B| = b. Hence, if ( A, B) is a CS pair, then A is an a-element subset of N (since
A ⊆ N and | A| = a), and B is a b-element subset of N (since B ⊆ N and | B| = b).
We shall compute the number |{CS pairs}| in two ways.
First way: Let f : {CS pairs} → { a-element subsets of N } be the map that sends
each CS pair ( A, B) to its first entry A. (This is well-defined, because if ( A, B) is a
CS pair, then A is an a-element subset of N.)
Theorem 1.2.5 (applied to S = {CS pairs} and W = { a-element subsets of N })
yields

|{CS pairs}|
= ∑ (# of s ∈ {CS pairs} satisfying f (s) = w)
w∈{ a-element subsets of N }

= ∑ (# of s ∈ {CS pairs} satisfying f (s) = C )

C ∈{ a-element subsets of N }
| {z }
=(# of CS pairs s satisfying f (s)=C )
=(# of CS pairs whose first entry is C )
(because if s is a CS pair, then f (s) is the first entry of s
(by the definition of f ))

(here, we have renamed the index w as C )

= ∑ (# of CS pairs whose first entry is C ) . (143)
C ∈{ a-element subsets of N }

Now, fix C ∈ { a-element subsets of N }. Thus, C is an a-element subset of N.

Thus, |C | = a, so that a = |C | ∈ N. Now, what are the CS pairs whose first entry is
C ? They are precisely the CS pairs of the form (C, B) for some b-element subset B
Enumerative Combinatorics: class notes page 195

of C. Thus, the map

{CS pairs whose first entry is C } → {b-element subsets of C } ,

(C, B) 7→ B

is well-defined. In the other direction, there is a map

{b-element subsets of C } → {CS pairs whose first entry is C } ,

B 7→ (C, B) .

Clearly, these two maps are mutually inverse. Thus, they are bijections. Hence, the
bijection principle yields

(# of CS pairs whose first entry is C )

 
a
= (# of b-element subsets of C ) = (144)
b

(by Theorem 1.3.12, applied to a, C and b instead of n, S and k).

Now, forget that we fixed C. We thus have shown that (144) holds for every
C ∈ { a-element subsets of N }. Thus, (143) becomes

|{CS pairs}|
= ∑ (# of CS pairs whose first entry is C ) (145)
C ∈{ a-element subsets of N }
| {z 
 }
a
=
b
(by (144))
   
a a
= ∑ b
= |{ a-element subsets of N }| ·
b
C ∈{ a-element subsets of N }
| {z }
=(# of a-element
 subsets of N )
n
=
a
(by Theorem 1.3.12, applied to S= N and k= a)
  
n a
= . (146)
a b

Second way: Forget about the map f defined previously.

Instead, let f : {CS pairs} → {b-element subsets of N } be the map that sends
each CS pair ( A, B) to its second entry B. (This is well-defined, because if ( A, B) is
a CS pair, then B is a b-element subset of N.)
Theorem 1.2.5 (applied to S = {CS pairs} and W = {b-element subsets of N })
Enumerative Combinatorics: class notes page 196

yields

|{CS pairs}|
= ∑ (# of s ∈ {CS pairs} satisfying f (s) = w)
w∈{b-element subsets of N }

= ∑ (# of s ∈ {CS pairs} satisfying f (s) = D )

D ∈{b-element subsets of N }
| {z }
=(# of CS pairs s satisfying f (s)= D )
=(# of CS pairs whose second entry is D )
(because if s is a CS pair, then f (s) is the second entry of s
(by the definition of f ))

(here, we have renamed the index w as D )

= ∑ (# of CS pairs whose second entry is D ) . (147)
D ∈{b-element subsets of N }

Now, fix D ∈ {b-element subsets of N }. Thus, D is a b-element subset of N.

Hence, | D | = b, so that b = | D | ∈ N. Now, what are the CS pairs whose second
entry is D ? They are precisely the CS pairs of the form ( A, D ) for some a-element
subset A of N satisfying D ⊆ A. Thus, the map

{CS pairs whose second entry is D } → { a-element subsets A of N satisfying D ⊆ A} ,

( A, D ) 7→ A

is well-defined. In the other direction, there is a map

{ a-element subsets A of N satisfying D ⊆ A} → {CS pairs whose second entry is D } ,

A 7→ ( A, D ) .

Clearly, these two maps are mutually inverse. Thus, they are bijections. Hence, the
bijection principle yields

(# of CS pairs whose second entry is D )

= (# of a-element subsets A of N satisfying D ⊆ A)
n−b
 
= (148)
a−b

(by Proposition 1.4.20, applied to d = b and k = a).

Now, forget that we fixed D. We thus have shown that (148) holds for every
Enumerative Combinatorics: class notes page 197

D ∈ {b-element subsets of N }. Thus, (147) becomes

|{CS pairs}|
= ∑ (# of CS pairs whose second entry is D )
D ∈{b-element subsets of N }
|  {z  }
n−b
=
a−b
(by (148))
n−b n−b
   
= ∑ a−b
= |{b-element subsets of N }| ·
a−b
D ∈{b-element subsets of N }
| {z }
=(# of b-element
 subsets of N )
n
=
b
(by Theorem 1.3.12, applied to S= N and k=b)
n−b
  
n
= . (149)
b a−b
Now, comparing (146) with (149), we obtain
n−b
     
n a n
= .
a b b a−b
This proves Proposition 1.3.35 in the case when n ∈ N.

Remark 2.2.3. Something looks fishy. In the second proof of Proposition 1.3.35
we gave above, we have assumed that n ∈ N, but we have not assumed that
a ∈N. But then, how did we manage to
  obtain
 a combinatorial interpretation for
a a
such as (144), which shows that is an integer? Doesn’t this contradict
b   b
a
the fact that is usually not an integer when a ∈ / N ? Where is our mistake?
b
It turns out that there is nothing wrong with our proof. The crux of the
matter is the little sentence “Now, fix C ∈ { a-element subsets of N }”. This
sentence provides the context in which the equality (144) was stated. In
other words, we have only claimed that  the equality (144) holds for each
a
C ∈ { a-element subsets of N }. Thus, will be an integer for each C ∈
b
 
a
{ a-element subsets of N }. Even though does not depend on C, we can-
b
not just throw the “for each C ∈ { a-element subsets of N }” part of this state-
ment away, because there is no a-priori guarantee that there exists a C ∈
{ a-element subsets of N }. And indeed, such a C can exist only when   a ∈ N
a
(and, even stronger, a ∈ {0, 1, . . . , n}); but if it does not exist, then will not
b
always be an integer. So the equality (144) leads to a contradiction if we take it
out of its context, but inside its context it is perfectly correct.
Enumerative Combinatorics: class notes page 198

The sentence “Now, forget that we fixed C” that appeared later on in the proof
marks the end of this context. After this sentence, C is no longer a fixed a-element
subset of N, and the equality (146) can be applied to any a-element subset of N in
the role of C. (Thus, in particular, (146) can be applied to manipulate the addends
in the sum appearing in (145), because this sum ranges over all a-element subsets
of N. Of course, when a ∈ / N, then this sum is an empty sum.)

Exercise 2.2.3. Let k be a positive integer.

(a) How many k-digit numbers are there? (A “k-digit number” means a non-
negative integer that has k digits without leading zeroes. For example, 3902 is
a 4-digit number, not a 5-digit number. Note that 0 counts as a 0-digit number,
not as a 1-digit number.)
(b) How many k-digit numbers are there that have no two equal digits?
(c) How many k-digit numbers have an even sum of digits?
(d) How many k-digit numbers are palindromes? (A “palindrome” is a number
such that reading its digits from right to left yields the same number. For exam-
ple, 5 and 1331 and 49094 are palindromes. Your answer may well depend on
the parity of k.)

2.3. The hockey-stick identity revisited

Next, we return to the hockey-stick identity (Theorem 1.3.29), which we have al-
ready proven in two ways (once in Subsection 1.3.7, and once in Subsection 2.1.1).
Let us give two more proofs for it:
 
n+1
Third proof of Theorem 1.3.29 (sketched). Let us try to rewrite as a sum of
k+1
binomial  coefficients
 with a k at the bottom (that is, of binomial coefficients of
?
the form for some values of ?). The first step towards this goal is to rewrite
  k
n+1
as follows:
k+1
     
n+1 n n
= + (by Theorem 1.3.8) .
k+1 k k+1
 
n
This is a good start, since it gives us a addend, which is of the desired form.
  k
n
But the other addend, , is not. So let us rewrite it in the same way, viz., as
k+1
Enumerative Combinatorics: class notes page 199

follows:
n−1 n−1 n−1
           
n+1 n n
= + = + + .
k+1 k k+1 k k k+1
 | {z }
n−1 n−1

= +
k k+1
(by Theorem 1.3.8)

This looks even better,since we

 now have two addends of the desired form; but we
n−1
still have an addend, , which is not of this form. So let us rewrite it in the
k+1
same way, and keep doing this over and over:
     
n+1 n n
= + (by Theorem 1.3.8)
k+1 k k+1
 | {z }
n−1 n−1

= +
k k+1
(by Theorem 1.3.8)
n−1 n−1
    
n
= + +
k k k+1
 | {z  }
n−2 n−2

= +
k k+1
(by Theorem 1.3.8)
n−1 n−2 n−2
      

n
= + + +
k k k k+1
 | {z  }
n−3 n−3

= +
k k+1
(by Theorem 1.3.8)
n−1 n−2 n−3 n−3
         
n
= + + + +
k k k k k+1
= ········· (keep doing the same transformation) .

We could go on like this forever, but let us stop at the point when the  single 
“bad”
0
addend (i.e., the binomial coefficient that has a k + 1 at the bottom) is . We
k+1
Enumerative Combinatorics: class notes page 200

thus have found

 
n+1
k+1
n−1 n−2
         
n 0 0
= + + +···+ + (150)
k k k k k+1
| {z }
=0
(by Lemma 1.3.14,
since k+1>0)
n−1 n−2
       
n 0
= + + +···+
k k k k
     
0 1 n
= + +···+ .
k k k

Thus, Theorem 1.3.29 holds.

This is not yet a formal proof, because “keep doing the same transformation” is
not a well-defined term in rigorous mathematics. (A computation cannot have an
indefinite length!) But it is easy to formalize this argument: We are really proving
the equality

n−1 n−i+1 n−i+1

         
n+1 n
= + +···+ +
k+1 k k k k+1
| {z }
a sum of i binomial coefficients
with n,n−1,...,n−i +1 at the top and (always) k at the bottom
 
n+1
for each i ∈ N. (This equality is our expression for after i transformations
k+1
of the form described above.) Of course, this equality is proved by induction on i.
Once it is proved, we can apply it to i = n + 1, and obtain (150), after which we
proceed as before. The details are LTTR.
Fourth proof of Theorem 1.3.29 (sketched). Let us count the (k + 1)-element subsets of
[n + 1] according to their largest element.
First of all, why are their largest elements well-defined? Each (k + 1)-element
subset of [n + 1] is nonempty (since its size is k + 1 > k ≥ 0), and thus has a unique
largest element. This largest element obviously belongs to [n + 1]. (Even better,
this largest element belongs to {k + 1, k + 2, . . . , n + 1}, but we don’t need this.)
Theorem 1.3.12 (applied to n + 1, k + 1 and [n + 1] instead of n, k and S) yields
 
n+1
= (# of (k + 1) -element subsets of [n + 1])
k+1
= ∑ (# of (k + 1) -element subsets of [n + 1] whose largest element is j)
j∈[n+1]
(151)
Enumerative Combinatorics: class notes page 201

as a consequence of the sum rule143 (since each (k + 1)-element subset of [n + 1]

has a unique largest element, which belongs to [n + 1]).
Now, fix j ∈ [n + 1]. How many (k + 1)-element subsets of [n + 1] are there
whose largest element is j ?
The answer is:

(# of (k + 1) -element subsets of [n + 1] whose largest element is j)

j−1
 
= . (152)
k

[Proof of (152): Informally, this can be justified as follows: If we want to choose a

(k + 1)-element subset of [n + 1] whose largest element is j, then we already know
that j has to be an element of this subset, and we only need to choose its remaining k
elements. These remaining k elements have to belong to {1, 2, . . . , j − 1} (in order to
ensure that j is the largest element of the resulting subset). Thus, we are really just
choosing a k-element subset  of { 1, 2, . . . , j − 1}. Since {1, 2, . . . , j − 1} is a ( j − 1)-
j−1
element set, we thus have many choices (by Theorem 1.3.12). This proves
k
(152).
Here is a rigorous version of this argument: If S is a (k + 1)-element subset of
[n + 1] whose largest element is j, then S \ { j} is a k-element subset of {1, 2, . . . , j − 1}.
Hence, there is a map

{(k + 1) -element subsets of [n + 1] whose largest element is j}

→ {k-element subsets of {1, 2, . . . , j − 1}}

which sends each S to S \ { j}. This map is furthermore a bijection144 . Thus, the
bijection principle yields

(# of (k + 1) -element subsets of [n + 1] whose largest element is j)

= (# of k-element subsets of {1, 2, . . . , j − 1})
j−1
 
=
k

(by Theorem 1.3.12, applied to j − 1 and {1, 2, . . . , j − 1} instead of n and S). This
proves (152) formally.]

143 To be more precise: of Theorem 1.2.5, applied to S = {(k + 1) -element subsets of [n + 1]}, W =
[n + 1] and f being the map that sends each (k + 1)-element subset of [n + 1] to its largest
element
144 Its inverse map sends each T to T ∪ { j }. (We leave all the necessary verifications to the reader,

who by now should find them completely routine.)

Enumerative Combinatorics: class notes page 202

Either way, we have thus shown that (152) holds. Hence, (151) becomes
 
n+1
k+1
= ∑ (# of (k + 1) -element subsets of [n + 1] whose largest element is j)
j∈[n+1]
|  {z  }
| {z } j−1
n +1 =
=∑ k
j =1 (by (152))
n +1  n  
j−1

i
= ∑ =∑ (here, we have substituted i for j − 1 in the sum)
j =1
k i =0
k
     
0 1 n
= + +···+ .
k k k

This proves Theorem 1.3.29 again.

Remark 2.3.1. In the above proof of Theorem 1.3.29, we have observed that the
largest element of a (k + 1)-element subset of [n + 1] not only belongs to [n + 1],
but actually belongs to the smaller set {k + 1, k + 2, . . . , n + 1} as well. But we
have not used this observation. What if we had used it, thus obtaining a sum
over all j ∈ {k + 1, k + 2, . . . , n + 1} instead of a sum over all j ∈ [n + 1] ?
We would have obtained Corollary 1.3.30 instead of Theorem 1.3.29.

2.4. Counting maps

2.4.1. All maps
How many maps are there from an m-element set to an n-element set? We already
know the answer (Theorem 1.5.7); let us just restate it a bit:

Theorem 2.4.1. Let m, n ∈ N. Let A be an m-element set. Let B be an n-element

set. Then,
(# of maps from A to B) = nm .

Proof of Theorem 2.4.1. We have B A = {maps A → B} (by the definition of B A ).

Also, | A| = m (since A is an m-element set) and | B| = n (likewise). Now,

(# of maps from A to B) = |{maps A → B}|

 
A A
= B since {maps A → B} = B

= | B || A| (by Theorem 1.5.7)

= nm (since | B| = n and | A| = m) .

This proves Theorem 2.4.1.

Enumerative Combinatorics: class notes page 203

Now, what if we want to count not all maps, but only some maps?
Recall that if f : A → B is a map between any two sets, then f ( A) denotes
the image of this map (i.e., the subset { f ( a) | a ∈ A} of B). More generally, if
f : A → B is a map and S is any subset of A, then f (S) denotes the subset
{ f ( a) | a ∈ S} of B.

Exercise 2.4.1. Let A and B be two sets. Let C be a finite subset of B. Assume
that A is finite, too.
(a) Prove that

(# of maps f : A → B satisfying f ( A) ⊆ C ) = |C || A| .

(b) Prove that

(# of maps f : A → B satisfying f ( A) = C ) = (# of surjective maps from A to C ) .

2.4.2. Injective maps

In Remark 1.3.5 (and in [19f-hw0s, Exercise 2]), we have introduced the notation
nk (called a “falling factorial”). Since we are now going to use this notation more
substantially, let us state its definition in detail:

Definition 2.4.2. Let n ∈ R and k ∈ N. Then, the falling factorial nk is the number
defined by

n k = n ( n − 1) ( n − 2) · · · ( n − k + 1) (153)
k −1
= ∏ (n − i ) . (154)
i =0

Falling factorials are also called lower factorials or descending factorials. Let us state
a few of their basic properties:

Proposition 2.4.3. Let n ∈ R.

(a) We have n0 = 1.
(b) We have n1 = n.  
n
(c) We have nk = k! · for each k ∈ N.
k
(d) If n ∈ N, then nn = n!.
(e) If n ∈ N, and if k ∈ N satisfies k > n, then nk = 0.
(f) We have nk · (n − k ) = nk+1 for each k ∈ N.

By now, the proof of this is a near-trivial exercise:

Enumerative Combinatorics: class notes page 204

Proof of Proposition 2.4.3. (a) Applying (153) to k = 0, we obtain

n0 = n (n − 1) (n − 2) · · · (n − 0 + 1) = (empty product) = 1.
| {z }
a product with 0 factors

This proves Proposition 2.4.3 (a).

(b) Applying (153) to k = 1, we obtain

n1 = n (n − 1) (n − 2) · · · (n − 1 + 1) = n. (155)
| {z }
a product with 1 factor

This proves Proposition 2.4.3 (b).

(c) Let k ∈ N. Then, multiplying both sides of the equality (42) by k!, we obtain

n ( n − 1) ( n − 2) · · · ( n − k + 1)
 
n
k! · = k! · = n ( n − 1) ( n − 2) · · · ( n − k + 1) .
k k!
 
k n
Comparing this with (153), we obtain n = k! · . This proves Proposition 2.4.3 (c).
k
(d) Let n ∈ N. Then, (153) (applied to k = n) yields

nn = n (n − 1) (n − 2) · · · (n − n + 1) = n (n − 1) (n − 2) · · · 1 = 1 · 2 · · · · · n = n!

(since n! is defined to be 1 · 2 · · · · · n). This proves Proposition 2.4.3 (d).

(e) Let n ∈ N, and let k ∈ N satisfy k > n. Then, Proposition 2.4.3 (c) yields
 
k n
n = k! · = 0.
k
| {z }
=0
(by Proposition 1.3.6)

This proves Proposition 2.4.3 (e).

(f) Let k ∈ N. Multiplying both sides of the equality (153) by n − k, we obtain

nk · (n − k ) = (n (n − 1) (n − 2) · · · (n − k + 1)) · (n − k ) = n (n − 1) (n − 2) · · · (n − k ) .

On the other hand, (153) (applied to k + 1 instead of k) yields

n k +1 = n ( n − 1 ) ( n − 2 ) · · · ( n − ( k + 1 ) + 1 ) = n ( n − 1 ) ( n − 2 ) · · · ( n − k ) .

Comparing these two equalities, we obtain nk · (n − k ) = nk+1 . This proves Proposition

2.4.3 (f).

We are now ready to count the injective maps between two finite sets:

Theorem 2.4.4. Let m, n ∈ N. Let A be an m-element set. Let B be an n-element

set. Then,
(# of injective maps from A to B) = nm . (156)
Enumerative Combinatorics: class notes page 205

Remark 2.4.5. (a) If m = 0, then the right-hand side of (156) is n0 = 1 (by

Proposition 2.4.3 (a)). And indeed, there is exactly one injective map from A = ∅
to B (namely, the “empty map”, which sends nothing anywhere).
(b) If m > n, then the right-hand side of (156) is nm = 0 (by Proposition 2.4.3
(e)). Thus, Theorem 2.4.4 yields that there are no injective maps from A to B
in this case. This is intuitively clear (as B has “not enough elements” to find
a distinct image for each a ∈ A); we are later going to state this as a theorem,
which we shall call the “Pigeonhole Principle for Injections”.

Let us first prove Theorem 2.4.4 in an informal way, using the “dependent prod-
uct rule” for n sets that we have stated in Subsection 2.2.2:
Proof of Theorem 2.4.4 (informal version). Let a1 , a2 , . . . , am be the m elements of A
(listed without repetitions). Then, in order to construct a map f from A to B,
we only have to choose the m values f ( a1 ) , f ( a2 ) , . . . , f ( am ). Moreover, to ensure
that this map f is injective, we need to choose these m values to be distinct. In
other words, we need to choose these m values to satisfy

f ( a1 ) ∈ B;
f ( a2 ) ∈ B \ { f ( a1 )} ;
f ( a3 ) ∈ B \ { f ( a1 ) , f ( a2 )} ;
f ( a4 ) ∈ B \ { f ( a1 ) , f ( a2 ) , f ( a3 )} ;
..
.
f ( am ) ∈ B \ { f ( a1 ) , f ( a2 ) , . . . , f ( am−1 )} .

We can construct such an injective map f by the following m-step procedure:

• first choose the value f ( a1 ) to be any element of B (there are n options for it,
since B is an n-element set);

• then choose the value f ( a2 ) to be any element of B \ { f ( a1 )} (there are n − 1

options for it145 );

• then choose the value f ( a3 ) to be any element of B \ { f ( a1 ) , f ( a2 )} (there

are n − 2 options for it146 );

145 since B \ { f ( a1 )} is an (n − 1)-element set

|{z} | {z }
an n-element set a 1-element subset of B
146 since B \ { f ( a1 ) , f ( a2 )} is an (n − 2)-element set
|{z} | {z }
an n-element set a 2-element subset of B
(since f ( a1 ) and f ( a2 ) are distinct
(because we chose f ( a2 ) out of B\{ f ( a1 )}))
Enumerative Combinatorics: class notes page 206

• then choose the value f ( a4 ) to be any element of B \ { f ( a1 ) , f ( a2 ) , f ( a3 )}

(there are n − 3 options for it147 );

• and so on;

• at last, choose the value f ( am ) to be any element of B \ { f ( a1 ) , f ( a2 ) , . . . , f ( am−1 )}

(there are n − (m − 1) options for it).

Thus, by the dependent product rule, the total # of ways to make all m choices is

n (n − 1) (n − 2) · · · (n − (m − 1)) = n (n − 1) (n − 2) · · · (n − m + 1) = nm

(since nm is defined to be n (n − 1) (n − 2) · · · (n − m + 1)). Thus, the # of injective

maps from A to B is nm . This proves Theorem 2.4.4.
How can we formalize this argument? The words “and so on” suggest that we
should frame it as an induction over m, and the use of the dependent product
rule hints at using Theorem 1.2.5 (which, as we recall from our Second proof of
Proposition 1.3.35, can be used as a stand-in replacement for the dependent product
rule, at least for its two-set version; but as we will see, it works for the general case
just as well). This leads us to the following proof:
Proof of Theorem 2.4.4 (formal version). Forget that we fixed n, m, A and B. If A and
B are any two sets, then we let Inj ( A, B) denote the set of all injective maps from
A to B.
Now, we shall prove Theorem 2.4.4 by induction on m:
Induction base: Theorem 2.4.4 holds when m = 0 148 . This completes the

induction base.
Induction step: Let k ∈ N. Assume (as the induction hypothesis) that Theorem
2.4.4 holds for m = k. We must now prove that Theorem 2.4.4 holds for m = k + 1.

147 since B \ { f ( a1 ) , f ( a2 ) , f ( a3 )} is an (n − 3)-element set

|{z} | {z }
an n-element set a 3-element subset of B
(since f ( a1 ), f ( a2 ) and f ( a3 ) are distinct
(since we chose f ( a2 ) out of B\{ f ( a1 )},
and since we chose f ( a3 ) out of B\{ f ( a1 ), f ( a2 )}))
148 Proof.Let m, n, A, B be as in Theorem 2.4.4, and assume that m = 0. We must prove the equality
(156).
The set A is an m-element set; thus, | A| = m = 0. Hence, A = ∅. Thus, there is exactly
one map from A to B (namely, the “empty map”); this map is clearly injective. Hence, there is
exactly one injective map from A to B. In other words, (# of injective maps from A to B) = 1.
Comparing this with

n m = n0 (since m = 0)
=1 (by Proposition 2.4.3 (a)) ,

we obtain (# of injective maps from A to B) = nm . This proves the equality (156) under the
assumption that m = 0. Hence, Theorem 2.4.4 holds when m = 0.
Enumerative Combinatorics: class notes page 207

Let n ∈ N. Let A be a (k + 1)-element set, and let B be an n-element set. We

must show that
(# of injective maps from A to B) = nk+1 . (157)
The set A is a (k + 1)-element set; thus, | A| = k + 1 > k ≥ 0. Hence, A is
nonempty. Choose any a ∈ A. (Such an a exists, since A is nonempty.)
The set A \ { a} has size

| A \ { a}| = | A| −1 (since a ∈ A)
|{z}
= k +1
= (k + 1) − 1 = k.

Hence, A \ { a} is a k-element set. Thus, we can apply Theorem 2.4.4 to A \ { a} and

k instead of A and m (since our induction hypothesis says that Theorem 2.4.4 holds
for m = k). As a result, we obtain

(# of injective maps from A \ { a} to B) = nk .

But Inj ( A \ { a} , B) is the set of all injective maps from A \ { a} to B (since this is
how Inj ( A \ { a} , B) is defined). Thus,

|Inj ( A \ { a} , B)| = (# of injective maps from A \ { a} to B)

= nk . (158)

Now, if h ∈ Inj ( A, B), then h is an injective map from A to B, and therefore the
restriction h | A\{a} of h to A \ { a} is an injective map from A \ { a} to B (since a
restriction of an injective map to a subset is still injective). Thus, the map

R : Inj ( A, B) → Inj ( A \ { a} , B) ,
h 7→ h | A\{a}

is well-defined. (All this map does is to restrict an injective map h ∈ Inj ( A, B) to

the subset A \ { a}. In other words, it “throws away” the  value of the map at a.)
1 2 3
[Example: If A = {1, 2, 3} and B is any set and h = (a map from A
x y z
 
1 3
to B in two-line notation) and a = 2, then R (h) = h |{1,3} = .]
x z
We now ask ourselves: For a given injective map g ∈ Inj ( A \ { a} , B), how many
injective maps h ∈ Inj ( A, B) are there that satisfy R (h) = g ? In other words,
how many ways are there to extend a given injective map g ∈ Inj ( A \ { a} , B) to an
injective map h ∈ Inj ( A, B) ? (The word “extends” here means that h | A\{a} should
be g, that is, R (h) = g.)
The answer to this question turns out to be nice – in particular, it does not depend
on g:
Enumerative Combinatorics: class notes page 208

Observation 1: Let g ∈ Inj ( A \ { a} , B). Then,

(# of maps h ∈ Inj ( A, B) satisfying R (h) = g) = n − k.

[Proof of Observation 1: Let us first argue informally: The map g : A \ { a} →

B is injective (since g ∈ Inj ( A \ { a} , B)); thus, it has k different values (since
| A \ { a}| = k). In other words, | g ( A \ { a})| = k. Hence, | B \ g ( A \ { a})| = n − k
(since g ( A \ { a}) is a subset of the n-element set B).
To construct an h ∈ Inj ( A, B) satisfying R (h) = g, we only need to choose a
value h ( a) (since the condition R (h) = g already forces all the other values of h
to equal the corresponding values of g). This value h ( a) must be distinct from all
values of g (in order for h to be injective); in other words, it has to belong to B \
g ( A \ { a}). Thus, there are n − k many options for h ( a) (since | B \ g ( A \ { a})| =
n − k). Hence, there are n − k many maps h ∈ Inj ( A, B) satisfying R (h) = g. This
proves Observation 1.
Here is a formal version of this argument:
The map g : A \ { a} → B is injective (since g ∈ Inj ( A \ { a} , B)). But if X and Y are
two finite sets, and if z : X → Y is an injective map, then |z ( X )| = | X | 149 . Applying
this to X = A \ { a}, Y = B and z = g, we obtain | g ( A \ { a})| = | A \ { a}| = k. But recall
that | B| = n (since B is an n-element set). Also, g ( A \ { a}) is clearly a subset of B. Hence,
Theorem 1.4.7 (a) (applied to B and g ( A \ { a}) instead of A and B) yields

| B \ g ( A \ { a})| = | B| − | g ( A \ { a})| = n − k.
|{z} | {z }
=n =k

If h ∈ Inj ( A, B) satisfies R (h) = g, then h ( a) ∈ B \ g ( A \ { a}) 150 . Hence, the map

{h ∈ Inj ( A, B) | R (h) = g} → B \ g ( A \ { a}) ,

h 7→ h ( a) (159)
149 Proof. This is a classical fact, with a simple proof: Let X and Y be two finite sets, and let z :
X → Y be an injective map. Let x1 , x2 , . . . , x p be all elements of X (listed without repetitions);
then, the elements x1 , x2 , . . . , x p are distinct, and | X | = p. But the map z is
injective, and thus
sends distinct elements to distinct elements. Hence, z ( x1 ) , z ( x2 ) , . .
. , z x p are distinct (since
x1 , x2 , . . . , x p are distinct).
 Therefore, the
set z ( x1 ) , z ( x2 ) , .. . , z x p has p distinct elements;
in other words, z ( x1 ) , z ( x2 ) , . . . , z x p = p. But X = x1 , x2 , . . . , x p (since x1 , x2 , . . . , x p
are all elements of X) and thus



z ( X ) = z x1 , x2 , . . . , x p = z ( x1 ) , z ( x2 ) , . . . , z x p ,


so that |z ( X )| = z ( x1 ) , z ( x2 ) , . . . , z x p = p = | X |, qed.
150 Proof. Let h ∈ Inj ( A, B ) satisfy R ( h ) = g. Thus, g = R ( h ) = h |
A\{ a} (by the definition of
R), and furthermore, the map h : A → B is injective (since h ∈ Inj ( A, B)). Now, if we had
h ( a) ∈ g ( A \ { a}), then there would existsome u  ∈ A \ { a} such that h ( a) = g (u). This u ∈ A
would then satisfy h ( a) = g (u) = h | A\{ a} (u) = h (u) and thus h (u) = g (u) = h ( a),
|{z}
=h| A\{ a}
which would lead to u = a (since h is injective); but this would contradict u ∈ A \ { a}. Thus, we
cannot have h ( a) ∈ g ( A \ { a}). Hence, h ( a) ∈
/ g ( A \ { a}). Combining this with h ( a) ∈ B, we
obtain h ( a) ∈ B \ g ( A \ { a}), qed.
Enumerative Combinatorics: class notes page 209

is well-defined. It is easy to see that this map is a bijection151 . Thus, the bijection principle
yields
|{h ∈ Inj ( A, B) | R (h) = g}| = | B \ g ( A \ { a})| = n − k.
Thus,

(# of maps h ∈ Inj ( A, B) satisfying R (h) = g)

= |{h ∈ Inj ( A, B) | R (h) = g}| = n − k.

Thus, Observation 1 is rigorously proven.]

Now, Theorem 1.2.5 (applied to S = Inj ( A, B), W = Inj ( A \ { a} , B) and f = R)
yields

|Inj ( A, B)| = ∑ (# of s ∈ Inj ( A, B) satisfying R (s) = w)

w∈Inj( A\{ a},B)

= ∑ (# of h ∈ Inj ( A, B) satisfying R (h) = g)

g∈Inj( A\{ a},B)
| {z }
=(# of maps h∈Inj( A,B) satisfying R(h)= g)
=n−k
(by Observation 1)

(here, we have renamed the indices w and s as g and h)

= ∑ (n − k) = |Inj ( A \ { a} , B)| · (n − k)
g∈Inj( A\{ a},B)
| {z }
=nk
(by (158))

= n k · ( n − k ) = n k +1 (by Proposition 2.4.3 (f)) .

But recall that Inj ( A, B) is the set of all injective maps from A to B (since this is
how Inj ( A, B) is defined). Thus,

|Inj ( A, B)| = (# of injective maps from A to B) ,

so that
(# of injective maps from A to B) = |Inj ( A, B)| = nk+1 .

151 Indeed, we can construct an inverse ( to it: For each b ∈ B \ g ( A \ { a}), we let hb be the map from
b, if c = a;
A to B that sends each c ∈ A to . It is easy to see that this map hb is injective.
g (c) , if c 6= a
(In fact, its values at the elements of A \ { a} are precisely the values of g, and thus are distinct
because g is injective; but its value at a is b, which is distinct from its values at the elements
of A \ { a} because b ∈ / g ( A \ { a}).) Hence, for each b ∈ B \ g ( A \ { a}), we have defined an
injective map hb from A to B which furthermore satisfies R (hb ) = g (since hb (c) = g (c) for all
c ∈ A \ { a}). Thus, we obtain a map

B \ g ( A \ { a}) → {h ∈ Inj ( A, B) | R (h) = g} ,

b 7→ hb .

It is now completely straightforward to check that this map is inverse to the map (159).
Enumerative Combinatorics: class notes page 210

Now, forget that we fixed n, A and B. We thus have shown that if n ∈ N, if A is

a (k + 1)-element set, and if B is an n-element set, then

(# of injective maps from A to B) = nk+1 .

In other words, Theorem 2.4.4 holds for m = k + 1. This completes the induction
step. Thus, Theorem 2.4.4 is proven by induction.

2.4.3. The pigeonhole principles

We shall now state two fundamental facts about maps between finite sets – the
so-called Pigeonhole Principles. These facts are intuitively fairly self-evident, but we
will nevertheless give rigorous proofs due to their importance.
The first of these facts is the Pigeonhole Principle for Injections:

Theorem 2.4.6 (Pigeonhole Principle for Injections). Let A and B be two finite
sets. Let f : A → B be an injective map. Then:
(a) We have | A| ≤ | B|.
(b) If | A| = | B|, then f is bijective.

Proof of Theorem 2.4.6. Let a1 , a2 , . . . , ak be the elements of A (listed without repeti-

tions). Thus, the elements a1 , a2 , . . . , ak are distinct and we have | A| = k. But the
map f is injective, and thus sends distinct elements to distinct elements. Hence,
f ( a1 ) , f ( a2 ) , . . . , f ( ak ) are distinct (since a1 , a2 , . . . , ak are distinct). Hence, the set
{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )} has k distinct elements. In other words,
|{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| = k = | A| . (160)
But { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} is clearly a subset of B. Therefore, Theorem 1.4.7 (b)
(applied to B and { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} instead of A and B) yields
|{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| ≤ | B| .
Now, (160) shows that | A| = |{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| ≤ | B|. This proves Theo-
rem 2.4.6 (a).
(b) Assume that | A| = | B|. Then, (160) becomes |{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| =
| A| = | B|. Hence, Theorem 1.4.7 (c) (applied to B and { f ( a1 ) , f ( a2 ) , . . . , f ( ak )}
instead of A and B) yields { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} = B. But this means that f
is surjective152 . Thus, f is bijective (since f is both injective and surjective). This
proves Theorem 2.4.6 (b).
The following similar-looking theorem is known as the Pigeonhole Principle for
Surjections:

152 Proof.Let b ∈ B. Then, b ∈ B = { f ( a1 ) , f ( a2 ) , . . . , f ( ak )}. In other words, b = f ( ai ) for some

i ∈ [k]. Consider this i. Thus, b is a value of f (namely, the value at ai ).
Forget that we fixed b. We thus have showed that each b ∈ B is a value of f . In other words,
f is surjective.
Enumerative Combinatorics: class notes page 211

Theorem 2.4.7 (Pigeonhole Principle for Surjections). Let A and B be two finite
sets. Let f : A → B be a surjective map. Then:
(a) We have | A| ≥ | B|.
(b) If | A| = | B|, then f is bijective.
Proof of Theorem 2.4.7. Let a1 , a2 , . . . , ak be the elements of A (listed without repeti-
tions). Thus, | A| = k and A = { a1 , a2 , . . . , ak }. The k elements f ( a1 ) , f ( a2 ) , . . . , f ( ak )
may or may not be distinct; thus, the set { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} has at most k
elements153 . In other words,
|{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| ≤ k = | A| . (161)
But f is surjective; hence, { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} = B 154 .
Thus, (161) rewrites
as | B| ≤ | A|. In other words, | A| ≥ | B|. This proves Theorem 2.4.7 (a).
(b) Assume that | A| = | B|. We must prove that f is bijective. First, we shall show
that f is injective.
Indeed, let x and y be two distinct elements of A. We shall show that f ( x ) 6=
f ( y ).
Indeed, assume the contrary. Thus, f ( x ) = f (y). But x ∈ A = { a1 , a2 , . . . , ak };
thus, x can be written in the form x = ai for some i ∈ [k]. Likewise, y can be written
in the form y = a j for some j ∈ [k]. Consider these i and j. If we had i = j, then we
would have ai = a j and thus x = ai = a j = y, which would contradict the fact that
x and y are distinct. Hence, we must have i 6= j.

But recall that f ( x ) = f (y). This rewrites as f ( ai ) = f a j (since x = ai and
y = a j ). Since i 6= j, this equality reveals that at least two of the k elements


f ( a1 ) , f ( a2 ) , . . . , f ( ak ) are equal (namely, f ( ai ) and f a j ). Therefore, there are
at most k − 1 distinct elements among these k elements f ( a1 ) , f ( a2 ) , . . . , f ( ak ). In
other words,
|{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| ≤ k − 1.
155 In view of { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} = B, this rewrites as | B| ≤ k − 1. Hence,
| B| ≤ k − 1 < k = | A| = | B|. This is, of course, absurd. This contradiction shows
that our assumption was false.
153 Here, we are tacitly using the following fact: If u1 , u2 , . . . , uk are any k objects (distinct or not),
then the set {u1 , u2 , . . . , uk } has at most k elements. Shouldn’t this, too, be proved?
(Yes, but I am leaving this to the texts on foundations. That said, you can easily construct a
proof by induction on k, using ( the even more fundamental fact that every finite set A and every
| A| , if s ∈ A;
object s satisfy | A ∪ {s}| = .)
| A| + 1, if s ∈ /A
154 Proof. Let b ∈ B. Then, there exists some a ∈ A such that b = f ( a ) (since f is surjective). Consider

 a∈
this a. We have A = { a1 , a2 , . . . , ak }. In other words, a = ai for some i ∈ [k]. Consider this i.

a  = f ( ai ) ∈ { f ( a1 ) , f ( a2 ) , . . . , f ( ak )}.
Now, b = f |{z}
= ai
Forget that we fixed b. We thus have shown that b ∈ { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} for
each b ∈ B. In other words, B ⊆ { f ( a1 ) , f ( a2 ) , . . . , f ( ak )}. Combining this with
{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )} ⊆ B (which is obvious), we obtain { f ( a1 ) , f ( a2 ) , . . . , f ( ak )} = B.
155 If you insist on proving this more formally than this, here is the argument that we tacitly used
Enumerative Combinatorics: class notes page 212

Hence, we have shown that f ( x ) 6= f (y).

Now, forget that we fixed x and y. We thus have proved that if x and y are two
distinct elements of A, then f ( x ) 6= f (y). In other words, the map f sends distinct
elements to distinct elements. In other words, f is injective.
Hence, f is bijective (since f is both injective and surjective). This proves Theo-
rem 2.4.7 (b).

Remark 2.4.8. The parts (b) of both Pigeonhole Principles become false if the
sets A and B are not finite. For example:

• The map N → N, i 7→ i + 1 is injective, but not bijective. Thus, Theorem

2.4.6 (b) fails if A and B are infinite.
(
0, if i = 0;
• The map N → N, i 7→ is surjective, but not bijective.
i − 1, if i > 0
Thus, Theorem 2.4.7 (b) fails if A and B are infinite.

2.4.4. Permutations
We can now prove Theorem 1.7.2:
Proof of Theorem 1.7.2. The set X is finite (since it is an n-element set) and satisfies
| X | = | X |. Hence, Theorem 2.4.6 (b) shows that if f : X → X is an injective map,
then f is bijective. In other words, every injective map from X to X is bijective.
The converse also holds (by definition). Thus, the injective maps from X to X are

here: Let G be the set






f ( a 1 ) , f ( a 2 ) , . . . , f a j −1 , f a j +1 , f a j +2 , . . . , f ( a k ) = f a p | p ∈ [ k ] \ { j } ,

which collects all of the k elements f ( a1 ) , f ( a2 ) , . . . , f ( ak ) except
for the
j-th one.

This set
G clearly has size | G | ≤ k − 1 (because f ( a1 ) , f ( a2 ) , . . . , f a j−1 , f a j+1 , f a j+2 , . . . , f ( ak )
are k − 1 elements).
Since i
6= j, the
element f ( ai ) is one of the k − 1 elements
f ( a1 ) , f ( a2 ) , . . . , f a j−1 , f a j+1 , f a j+2 , . . . , f ( ak ) that constitute the set G (because i 6= j
entails i ∈ [k] \ { j}), and thus belongs to G. Therefore, { f ( ai )} ⊆ G, so that G ∪ { f ( ai )} = G.
Hence,




G = G ∪ { f ( ai )} = G ∪ f a j since f ( ai ) = f a j





= f ( a 1 ) , f ( a 2 ) , . . . , f a j −1 , f a j +1 , f a j +2 , . . . , f ( a k ) ∪ f a j





since G = f ( a1 ) , f ( a2 ) , . . . , f a j−1 , f a j+1 , f a j+2 , . . . , f ( ak )





= f ( a 1 ) , f ( a 2 ) , . . . , f a j −1 , f a j +1 , f a j +2 , . . . , f ( a k ) , f a j
= { f ( a1 ) , f ( a2 ) , . . . , f ( ak )}





(because the k elements f ( a1 ) , f ( a2 ) , . . . , f a j−1 , f a j+1 , f a j+2 , . . . , f ( ak ) , f a j are ex-
actly the k elements f ( a1 ) , f ( a2 ) , . . . , f ( ak ), up to order). Thus, the inequality | G | ≤ k − 1,
which we have previously proved, rewrites as |{ f ( a1 ) , f ( a2 ) , . . . , f ( ak )}| ≤ k − 1.
Enumerative Combinatorics: class notes page 213

precisely the bijective maps from X to X. Hence,

{injective maps from X to X } = {bijective maps from X to X }

= {bijections from X to X } = {permutations of X }

(because the permutations of X are defined to be the bijections from X to X). Hence,

(# of injective maps from X to X ) = (# of permutations of X ) .

But Theorem 2.4.4 (applied to m = n, A = X and B = X) shows that

(# of injective maps from X to X ) = nn = n!

(by Proposition 2.4.3 (d)). Comparing these two equalities, we obtain

(# of permutations of X ) = n!.

This proves Theorem 1.7.2.

2.4.5. Surjective maps

After having counted injective maps, let us count surjective maps. The answer will
not be as simple as the one we obtained for injective maps in Theorem 2.4.4. First,
we introduce a notation (which we already have used in Theorem 1.2.10):

Definition 2.4.9. Let m ∈ N and n ∈ N. Then, sur (m, n) shall mean the # of
surjective maps from [m] to [n].

Example 2.4.10. We have sur (3, 2) = 6, because there are exactly 6 surjective
maps from [3] to [2]; namely, these maps are
     
1 2 3 1 2 3 1 2 3
, , ,
1 1 2 1 2 1 1 2 2
     
1 2 3 1 2 3 1 2 3
, ,
2 1 1 2 1 2 2 2 1

(written in two-line notation).

A priori, these numbers sur (m, n) only count surjective maps between the spe-
cific finite sets [m] and [n]; but it is easy to see that they also count surjective maps
between any two finite sets:

Proposition 2.4.11. Let m, n ∈ N. Let A be an m-element set. Let B be an

n-element set. Then,

(# of surjective maps from A to B) = sur (m, n) .

Enumerative Combinatorics: class notes page 214

Proof of Proposition 2.4.11 (sketched). This is a relabeling argument, similar to our

proof of Lemma 1.7.6: Relabel the m elements of A as 1, 2, . . . , m, and relabel the n
elements of B as 1, 2, . . . , n. The details are LTTR.156
Recall that “surjection” is a shorthand for “surjective map”. Thus, for any m, n ∈
N, we have

(# of surjections from [m] to [n])

= (# of surjective maps from [m] to [n]) = sur (m, n) (162)

(since sur (m, n) is defined as the # of surjective maps from [m] to [n]).
Let us next determine some values of sur (m, n) (with heavy use of Iverson brack-
ets):

Proposition 2.4.12. (a) We have sur (m, 0) = [m = 0] for all m ∈ N.

(b) We have sur (m, 1) = [m 6= 0] = 1 − [m = 0] for all m ∈ N.
(c) We have sur (m, 2) = 2m − 2 + [m = 0] for all m ∈ N.
(d) We have sur (0, k ) = [k = 0] for all k ∈ N.
(e) We have sur (1, k ) = [k = 1] for all k ∈ N.
(f) We have sur (m, n) = 0 for all m, n ∈ N satisfying m < n.

Proving this proposition is an easy exercise in understanding what surjectivity

of maps means (as well as applying Theorem 2.4.1 and Theorem 2.4.7 (a)):
Proof of Proposition 2.4.12. The set [0] is empty. Thus, there is exactly one map from [0] to
[0] (namely, the “empty map”, which sends nothing anywhere). This map is surjective157 .
156 Forthe formal proof, you have to construct a bijection from the set {surjective maps from A to B}
to the set {surjective maps from [m] to [n]}. Here is how to do this:
We have | A| = m (since A is an m-element set) and |[m]| = m, so that | A| = m = |[m]|. Thus,
the sets A and [m] have the same size. Hence, Theorem 1.1.7 (applied to X = A and Y = [m])
shows that there is a bijection α : A → [m]. Likewise, there is a bijection β : B → [n]. Consider
these α and β. Now, it is easy to see that the maps

{surjective maps from A to B} → {surjective maps from [m] to [n]} ,

f 7 → β ◦ f ◦ α −1

and

{surjective maps from [m] to [n]} → {surjective maps from A to B} ,

f 7 → β −1 ◦ f ◦ α

are well-defined and mutually inverse; thus, they are bijections. Hence, the bijection principle
yields

(# of surjective maps from A to B)

= (# of surjective maps from [m] to [n]) = sur (m, n)
(since sur (m, n) was defined to be the # of surjective maps from [m] to [n]).
157 Indeed, a map f : A → B (where A and B are two sets) is surjective if and only if each element of
B lies in its image. If the set B is empty, then this is vacuously true.
Enumerative Combinatorics: class notes page 215

Hence, there is exactly one surjective map from [0] to [0]. In other words, sur (0, 0) = 1
(since sur (0, 0) is defined as the # of surjective maps from [0] to [0]). Comparing sur (0, 0) =
1 with [0 = 0] = 1, we obtain sur (0, 0) = [0 = 0].
(a) Let m ∈ N. Recall that sur (0, 0) = [0 = 0]. In other words, Proposition 2.4.12
(a) holds for m = 0. Thus, for the rest of this proof, we WLOG assume that m 6= 0.
Hence, m > 0 (since m ∈ N). Thus, 1 ∈ [m]. Hence, there exists no map from [m] to
[0] (because such a map would have to send 1 to some element of [0], but there is no
element of [0] to send it to). Thus, a fortiori, there exists no surjective map from [m] to
[0]. In other words, sur (m, 0) = 0 (since sur (m, 0) is defined as the # of surjective maps
from [m] to [0]). Comparing this with [m = 0] = 0 (which follows from m 6= 0), we obtain
sur (m, 0) = [m = 0]. Thus, Proposition 2.4.12 (a) is proved.
(f) Let m, n ∈ N be such that m < n. Theorem 2.4.7 (a) (applied to A = [m] and
B = [n]) shows that if f : [m] → [n] is a surjective map, then |[m]| ≥ |[n]|; but this
contradicts |[m]| = m < n = |[n]|. Thus, we obtain a contradiction for each surjective map
f : [m] → [n]. Hence, there exists no such map. In other words, there is no surjective
map from [m] to [n]. In other words, sur (m, n) = 0 (since sur (m, n) is defined as the # of
surjective maps from [m] to [n]). This proves Proposition 2.4.12 (f).
(d) Let k ∈ N. Recall that sur (0, 0) = [0 = 0]. In other words, Proposition 2.4.12 (d)
holds for k = 0. Hence, for the rest of this proof, we WLOG assume that k 6= 0. Hence,
k > 0 (since k ∈ N), so that 0 < k. Thus, Proposition 2.4.12 (f) (applied to m = 0 and
n = k) yields sur (0, k ) = 0. Comparing this with [k = 0] = 0 (which follows from k 6= 0),
we obtain sur (0, k ) = [k = 0]. This proves Proposition 2.4.12 (d).
(b) Let m ∈ N. We need to prove the chain of equalities sur (m, 1) = [m 6= 0] = 1 −
[m = 0]. The second of these equalities is easily checked158 . Thus, it remains to prove the
first equality, i.e., the equality sur (m, 1) = [m 6= 0].
Proposition 2.4.12 (d) (applied to k = 1) yields sur (0, 1) = [1 = 0] = 0 (since 1 6= 0).
Comparing this with [0 6= 0] = 0 (since 0 = 0), we obtain sur (0, 1) = [0 6= 0]. In other
words, the equality sur (m, 1) = [m 6= 0] (which we intend to prove) holds for m = 0. Thus,
for the rest of this proof, we WLOG assume that m 6= 0. Hence, m > 0 (since m ∈ N),
so that 1 ∈ [m]. Now, the set [1] has only one element, namely 1. Thus, there is exactly
one map from [m] to [1]: namely, the map f that sends each element of [m] to 1 (because
the only possible value for this map is 1). This map is surjective, because the only element
of [1] (namely, 1) is a value of this map (indeed, it is the value of this map at 1 ∈ [m]).
Thus, there is exactly one surjective map from [m] to [1]. In other words, sur (m, 1) = 1
(since sur (m, 1) is defined as the # of surjective maps from [m] to [1]). Comparing this with
[m 6= 0] = 1 (which follows from m 6= 0), we obtain sur (m, 1) = [m 6= 0]. This completes
our proof of Proposition 2.4.12 (b).
(e) Let k ∈ N. Proposition 2.4.12 (b) (applied to m = 1) yields sur (1, 1) = [1 6= 0] =
1 − [1 = 0]. Hence, sur (1, 1) = [1 6= 0] = 1 (since 1 6= 0). Compared with [1 = 1] = 1, this
yields sur (1, 1) = [1 = 1]. In other words, Proposition 2.4.12 (e) holds for k = 1. Hence,
for the rest of this proof, we WLOG assume that k 6= 1.
Proposition 2.4.12 (a) (applied to m = 1) yields sur (1, 0) = [1 = 0] = 0 (since 1 6= 0).
Compared with [0 = 1] = 0, this yields sur (1, 0) = [0 = 1]. In other words, Proposition

158 Indeed,it is a particular case of the equality [not A] = 1 − [A], which holds for any logical
statement A. Applying the latter equality to A = (“m = 0”), we obtain [not (m = 0)] = 1 −
[m = 0]. In other words, [m 6= 0] = 1 − [m = 0].
Enumerative Combinatorics: class notes page 216

2.4.12 (e) holds for k = 0. Hence, for the rest of this proof, we WLOG assume that k 6= 0.
We now have k 6= 0 and k 6= 1. Hence, k ≥ 2 (since k ∈ N). Thus, k ≥ 2 > 1, so that
1 < k. Hence, Proposition 2.4.12 (f) (applied to m = 1 and n = k) yields sur (1, k ) = 0.
Comparing this with [k = 1] = 0 (which is because k 6= 1), we obtain sur (1, k ) = [k = 1].
This proves Proposition 2.4.12 (e).
(c) Let m ∈ N. Proposition 2.4.12 (d) (applied to k = 2) yields sur (0, 2) = [2 = 0] = 0.
Comparing this with |{z} 20 −2 + [0 = 0] = 1 − 2 + 1 = 0, we obtain sur (0, 2) = 20 − 2 +
| {z }
=1 =1
[0 = 0]. Thus, Proposition 2.4.12 (c) holds for m = 0. Hence, for the rest of this proof, we
WLOG assume that m 6= 0. Hence, 1 ∈ [m]. Now, Theorem 2.4.1 (applied to n = 2, A = [m]
and B = [2]) yields
(# of maps from [m] to [2]) = 2m .
In other words, there are exactly 2m maps from [m] to [2]. Which of these 2m maps are
surjective? Obviously, in order for a map f : [m] → [2] to be surjective, it suffices that both
1 and 2 appear as its values; in other words, it must avoid sending all elements of [m] to
1, and it must avoid sending all elements of [m] to 2. Thus, among the 2m maps from [m]
to [2], only two fail to be surjective: namely, the map α that sends all elements of [m] to 1,
and the map β that sends all elements of [m] to 2. Moreover, these two maps α and β are
distinct (because α (1) = 1 6= 2 = β (1)). Hence, exactly 2 out of the 2m maps from [m] to
[2] fail to be surjective. Thus,
(# of surjective maps from [m] to [2]) = 2m − 2.
This rewrites as sur (m, 2) = 2m − 2 (since sur (m, 2) is defined as the # of surjective maps
from [m] to [2]). Compared with 2m − 2 + [m = 0] = 2m − 2, this yields sur (m, 2) =
| {z }
=0
(since m6=0)
2m − 2 + [m = 0]. Thus, Proposition 2.4.12 (c) is proved.

Proposition 2.4.12 gives us many values of sur (m, n), but far from all of them –
and certainly not the most interesting ones. For example, what is sur (30, 20) ?
It thus is reasonable to look for a recursive formula for sur (m, n). We shall show
two different approaches to this question, giving two different recursive formulas.
1st approach: Fix m ∈ N and a positive integer n > 0. Thus, n ∈ [n].
Given a surjective map f : [m] → [n], we let J f be the set of all i ∈ [m] such that
f (i ) = n. 159 Clearly, this set J f is a subset of [m] and satisfies J f 6= ∅ (since f
is surjective, so there exists at least one i ∈ [m] such that f (i ) = n). Thus, the sum
rule yields
(# of all surjective maps f : [m] → [n])
∑


= # of all surjective maps f : [m] → [n] satisfying J f = J . (163)
J ⊆[m];
J 6 =∅
159 For
 = 3, and if f : [m] → [n] is the surjective map written in two-line
example,if m = 4 and n
1 2 3 4
notation as , then J f = {1, 3}, because 1 and 3 are precisely the i ∈ [4] such that
3 1 3 2
f (i ) = 3.
Enumerative Combinatorics: class notes page 217

Now, fix a subset J of [m] that satisfies J 6= ∅. What is the # of all surjective maps
f : [m] → [n] satisfying J f = J ? Such a map f must satisfy


J = J f = {i ∈ [ m ] | f (i ) = n } by the definition of J f .

In other words, such a map f must send all elements of J, but no other elements, to
n. In other words, it must send all elements of J to n, and send all other elements
of [m] to numbers different from n. The latter numbers are, of course, the elements
of [n] \ {n} = [n − 1]. Thus, a surjective map f : [m] → [n] satisfying J f = J
must send all elements of J to n, and send all elements of [m] \ J to elements of
[n − 1]. Hence, if f is such a map, then f is uniquely determined by its values at
the elements of [m] \ J, and said values must belong to [n − 1]. In other words,
if f is such a map, then f is uniquely determined by its restriction f |[m]\ J , and
furthermore this restriction must “actually” be a map from [m] \ J to [n − 1], in
the sense that all its values belong to [n − 1]. Moreover, in order for the map f to
be surjective, all of the numbers 1, 2, . . . , n must be values of f . Since J 6= ∅, we
know that n will always be a value of f (indeed, f sends any element of J to n,
and there is at least one element of J, since J 6= ∅); we thus only need to ensure
that 1, 2, . . . , n − 1 are values of f as well. Obviously, these values must be taken at
elements of [m] \ J (because f sends all elements of J to n). In other words, if we
restrict f to the set [m] \ J, then the resulting map f |[m]\ J : [m] \ J → [n] needs to
take all the numbers 1, 2, . . . , n − 1 as values; i.e., if we consider f |[m]\ J as a map
from [m] \ J to [n − 1], then f |[m]\ J should be surjective. Thus, in order to construct
a surjective map f : [m] → [n] satisfying J f = J, we need to construct a surjective
map from [m] \ J to [n − 1]. Hence,


# of all surjective maps f : [m] → [n] satisfying J f = J
= (# of all surjective maps from [m] \ J to [n − 1]) . (164)

(Formally speaking, this can be proved using the bijection principle160 .)

On the other hand, J ⊆ [m], and thus an application of Theorem 1.4.7 (a) yields
|[m] \ J | = |[m]| − | J | = m − | J |. Thus, [m] \ J is an (m − | J |)-element set. Also,
| {z }
=m
[n − 1] is an (n − 1)-element set (since n is a positive integer, so that n − 1 ∈ N).
Hence, Proposition 2.4.11 (applied to [m] \ J, [n − 1], |[m] \ J |, and |[n − 1]| instead
of A, B, m and n) yields

(# of all surjective maps [m] \ J → [n − 1]) = sur (m − | J | , n − 1) .

160 More precisely, we need to apply the bijection principle to the bijection
n o
surjective maps f : [m] → [n] satisfying J f = J → {surjective maps from [m] \ J to [n − 1]}

that sends each f to its restriction f |[m]\ J (considered as a map from [m] \ J to [n − 1]).
Here, of course, we are using the fact that if f : [m] → [n] satisfies J f = J, then the restriction
f |[m]\ J can be considered as a map from [m] \ J to [n − 1], since it does not take the value n.
Enumerative Combinatorics: class notes page 218

Thus, (164) becomes

# of all surjective maps f : [m] → [n] satisfying J f = J
= (# of all surjective maps from [m] \ J to [n − 1])
= sur (m − | J | , n − 1) . (165)

Now, forget that we fixed J. We thus have proved (165) for each subset J of [m]
satisfying J 6= ∅. Hence, (163) becomes

(# of all surjective maps f : [m] → [n])

∑


= # of all surjective maps f : [m] → [n] satisfying J f = J
J ⊆[m]; | {z }
J 6 =∅ =sur(m−| J |,n−1)
(by (165))
 

= ∑ sur (m − | J | , n − 1) = ∑ ∑ sur m − | J | , n − 1
 
J ⊆[m]; j∈{1,2,...,m} J ⊆[m];
|{z}
J 6 =∅ | {z } J 6=∅; =j
m | J |= j
=∑
j =1

here, we have split the sum according to the value of | J | ,

 

 because each subset J of [m] satisfying J 6= ∅ (that is, 

 each nonempty subset J of [m] ) satisfies | J | ∈ {1, 2, . . . , m} 
(since J ⊆ [m] entails | J | ≤ m, whereas J 6= ∅ entails | J | ≥ 1)
m
= ∑ ∑ sur (m − j, n − 1)
j =1 J ⊆[m];
J 6=∅;
| J |= j
| {z }
=(# of J ⊆[m] such that J 6=∅ and | J |= j)·sur(m− j,n−1)
m
= ∑ (# of J ⊆ [m] such that J 6= ∅ and | J | = j) · sur (m − j, n − 1) . (166)
j =1

We shall next simplify the # on the right hand side of this equation. Indeed, let
j ∈ [m]. Thus, j ≥ 1 > 0. Hence, each subset J of [m] satisfying | J | = j must
automatically be nonempty (since | J | = j > 0); in other words, it must satisfy
J 6= ∅. Thus, the requirement J 6= ∅ in “# of J ⊆ [m] such that J 6= ∅ and | J | = j”
is redundant. Hence,

(# of J ⊆ [m] such that J 6= ∅ and | J | = j)

= (# of J ⊆ [m] such that | J | = j)
 
m
= (# of j-element subsets of [m]) = (167)
j

(by Theorem 1.3.12, applied to m, j and [m] instead of n, k and S).

Enumerative Combinatorics: class notes page 219

Forget that we fixed j. We thus have proved (167) for each j ∈ [m]. Hence, (166)
becomes

(# of all surjective maps f : [m] → [n])

m
= ∑ |(# of J ⊆ [m] such that J 6= ∅ and | J | = j) · sur (m − j, n − 1)
j =1 {z  }
m
=
j
(by (167))
m   m −1  
m m
= ∑ j
· sur (m − j, n − 1) = ∑ m−j
· sur (m − j, n − 1)
j =1 j =0
| {z } 
m
=
m−j
(by Theorem 1.3.11,
applied to m and j
instead of n and k)
m −1  
m
= ∑ j
· sur ( j, n − 1)
j =0

(here, we have substituted m − j for j in the sum). Hence,

sur (m, n) = (# of all surjective maps f : [m] → [n])

m   m −1  
m m
=∑ · sur (m − j, n − 1) = ∑ · sur ( j, n − 1) .
j =1
j j =0
j

Forget that we fixed m and n. Thus, we have proved the following recursive
formula for the numbers sur (m, n):

Proposition 2.4.13. Let m ∈ N, and let n be a positive integer. Then,

 m m −1  
m m
sur (m, n) = ∑ · sur (m − j, n − 1) = ∑ · sur ( j, n − 1) .
j =1
j j =0
j

Using this proposition and Proposition 2.4.12 (a), it is easy to compute sur (m, n)
recursively.

Class of 2019-10-23

2nd approach: Here is another way to get a recursive formula for sur (m, n).
Fix two positive integers m and n. Let us classify the surjections [m] → [n]
according to the image of m.
A surjection f : [m] → [n] will be called
Enumerative Combinatorics: class notes page 220

• red if f (m) = f (i ) for some i ∈ [m − 1];

• green if it is not red (i.e., if f (m) 6= f (i ) for all i ∈ [m − 1]).

[Examples: If m =4 and n = 3, then the surjection f 1 : [m] → [n] written in
1 2 3 4
two-line notation as is red (since f 1 (m) = f 1 (1) with 1 ∈ [m − 1]),
2 1 3 2  
1 2 3 4
while the surjection f 2 : [m] → [n] written in two-line notation as
2 1 2 3
is green (since f 2 (m) = 3 6= f 2 (i ) for all i ∈ [m − 1]).]
If f : [m] → [n] is a red surjection, then its restriction f |[m−1] : [m − 1] → [n] is
“still” a surjection161 . Conversely, if f : [m] → [n] is any map whose restriction
f |[m−1] : [m − 1] → [n] is a surjection, then f must be a red surjection162 .
Thus, we get the following algorithm for constructing a red surjection f : [m] →
[ n ]:

• first, we choose f (m) (there are n choices for this, since we want f (m) ∈ [n]);

• then, we choose f (1) , f (2) , . . . , f (m − 1); in other words, we choose the re-
striction f |[m−1] (there are sur (m − 1, n) choices for this, since f |[m−1] has to
be a surjection from [m − 1] to [n]).

Hence, the dependent product rule yields

(# of red surjections f : [m] → [n]) = n · sur (m − 1, n) . (168)

161 Formal proof. Let f : [m] → [n] be a red surjection. Then, f is surjective; thus, f ([m]) =
[n]. Also, f (m) = f (i ) for some i ∈ [m − 1] (since f is red). Consider this i. From
i ∈ [m − 1] = {1, 2, . . . , m − 1}, we obtain f (i ) ∈ { f (1) , f (2) , . . . , f (m − 1)}. Hence, f (m) =
f (i ) ∈ { f (1) , f (2) , . . . , f (m − 1)}, so that { f (m)} ⊆ { f (1) , f (2) , . . . , f (m − 1)}.
Now, the image of the restriction f |[m−1] is the set
 

 


 


 

     
f |[m−1] ([m − 1]) = f | [ m −1] ( j ) | j ∈ [m − 1] = { f ( j) | j ∈ [m − 1]}

 | {z } 


 



 = f ( j ) 


(since f |[m−1] is a restriction of f )

= { f (1) , f (2) , . . . , f (m − 1)} = { f (1) , f (2) , . . . , f (m − 1)} ∪ { f (m)}

(since { f (m)} ⊆ { f (1) , f (2) , . . . , f (m − 1)})
= { f (1) , f (2) , . . . , f (m − 1) , f (m)} = { f (1) , f (2) , . . . , f (m)} = f ([m]) = [n]

(since f is surjective). Thus, the map f |[m−1] : [m − 1] → [n] is surjective. In other words,
f |[m−1] : [m − 1] → [n] is a surjection.
162 Check this! (The idea is: Since f |
[m−1] : [ m − 1] → [ n ] is surjective, we know that each element
of [n] is a value of f at one of the numbers 1, 2, . . . , m − 1. Hence, f is surjective. Moreover,
f (m) ∈ [n] must also be a value of f |[m−1] (since f |[m−1] is surjective); but this is saying that
f (m) = f (i ) for some i ∈ [m − 1]. This shows that the surjection f is red.)
Enumerative Combinatorics: class notes page 221

(You might want to see how this argument can be formalized without speaking
of “choices”. See Exercise 2.4.2 further below for the answer.)
On the other hand, if f : [m] → [n] is a green surjection, then the restriction
f |[m−1] has image [n] \ { f (m)} 163 , and thus can be viewed as a surjection

[m − 1] → [n] \ { f (m)}. Conversely, if f : [m] → [n] is any map whose restriction

f |[m−1] is a surjection from [m − 1] to [n] \ { f (m)}, then f must be a green surjec-
tion164 . Hence, we get the following algorithm for constructing a green surjection
f : [ m ] → [ n ]:

• first, we choose f (m) (there are n choices);

• then, we choose f (1) , f (2) , . . . , f (m − 1); in other words, we choose the re-
striction f |[m−1] (there are sur (m − 1, n − 1) choices for this, because f |[m−1]
needs to be a surjection [m − 1] →  [n] \ { f (m)}, and Proposition
 2.4.11 shows

that the # of such surjections is sur |[m − 1]|, |[n] \ { f (m)}| = sur (m − 1, n − 1)).
 
| {z } | {z }
= m −1 = n −1

Hence,

(# of green surjections f : [m] → [n]) = n · sur (m − 1, n − 1) . (169)

(Again, see Exercise 2.4.2 further below for a rigorous proof of this equality.)
Hence, (162) becomes

sur (m, n)
= (# of surjections f : [m] → [n])
= (# of red surjections f : [m] → [n]) + (# of green surjections f : [m] → [n])
| {z } | {z }
=n·sur(m−1,n) =n·sur(m−1,n−1)
(by (168)) (by (169))
 
since each surjection f : [m] → [n] is either red or green
(but not both at the same time)
= n · sur (m − 1, n) + n · sur (m − 1, n − 1)
= n · (sur (m − 1, n) + sur (m − 1, n − 1)) .

Forget that we fixed m and n. Thus, we have proved the following fact:

163 The proof proceeds roughly as follows: Let f : [m] → [n] be a green surjection. Then, the image
of f contains all elements of [n] (since f is a surjective), but the value n is taken by f only once,
namely at m (because f is green). Thus, if we remove m from the domain of f , then f will no
longer take the value n. In other words, the restriction f |[m−1] no longer takes the value n.
Other than that, of course, this restriction takes all the values that f takes; thus, it takes all the
elements of [n] as values except for f (m). In other words, it has image [n] \ { f (m)}.
164 Check this!
Enumerative Combinatorics: class notes page 222

Proposition 2.4.14. Let m and n be positive integers. Then,

sur (m, n) = n · (sur (m − 1, n) + sur (m − 1, n − 1)) .

Exercise 2.4.2. Give rigorous proofs of the equalities (168) and (169) above.

Proposition 2.4.14 is an even better recursion than Proposition 2.4.13; it is almost

as simple as Theorem 1.3.8. Let us draw a conclusion from it:

Corollary 2.4.15. (a) We have sur (n, n) = n! for all n ∈ N.

(b) The integer sur (m, n) is a multiple of n! for all n ∈ N and m ∈ N.

First proof of Corollary 2.4.15 (rough idea). Both parts can easily be shown by induc-
tion using Proposition 2.4.14. (Use induction on n for part (a), and induction on m
for part (b).)

Exercise 2.4.3. Complete this proof of Corollary 2.4.15.

Second proof of Corollary 2.4.15 (rough idea). (a) Let n ∈ N. The surjections [n] →
[n] are bijections (by Theorem 2.4.7 (b)). Thus, they are precisely the bijections
[n] → [n]. In other words, they are precisely the permutations of [n]. Hence, their
number is n! (according to Theorem 1.7.2). Thus, sur (n, n) = n!. This proves
Corollary 2.4.15 (a).
(b) Here is just the idea behind the proof (as we are, so far, missing the language
for formalizing it): Each surjection f : [m] → [n] provides a way of “grouping” the
elements of [m] into n nonempty (disjoint) subsets – namely, the subsets

{ i ∈ [ m ] | f ( i ) = 1}, { i ∈ [ m ] | f ( i ) = 2}, ..., {i ∈ [ m ] | f (i ) = n } .

| {z } | {z } | {z }
the elements the elements the elements
sent to 1 by f sent to 2 by f sent to n by f

(These n subsets are nonempty, since f is surjective.) For example, the surjection
f : [6] → [3] that is given in two-line notation as
 
1 2 3 4 5 6
2 1 2 3 3 2

groups the elements of [6] into the 3 subsets

{2} , {1, 3, 6} , {4, 5} .

|{z} | {z } | {z }
the elements the elements the elements
sent to 1 by f sent to 2 by f sent to 3 by f

Conversely, each such “grouping” comes from a unique surjection f : [m] → [n]
(namely, the surjection which sends the elements of the 1-st subset to 1, the ele-
ments of the 2-nd subset to 2, and so on). Hence, the # of such “groupings” is
Enumerative Combinatorics: class notes page 223

sur (m, n). However, if we forget the order of the subsets in each grouping (i.e., we
forget which subset is the 1-st subset, which subset is the 2-nd subset, and so on,
and only remember the set of these subsets), then the # of all such “groupings”
becomes sur (m, n) /n!, because each unordered “grouping” can be ordered in pre-
cisely n! many different ways. Thus, sur (m, n) /n! ∈ N. This proves Corollary
2.4.15 (b).
For the details of this argument, see [18s-hw3s, §0.3 (on “set partitions”)] (specif-
ically, [18s-hw3s, Definition 0.13]).

Remark   Let m ∈ N and n ∈ N. The number sur (m, n) /n! is often

2.4.16.
m
denoted , and is called a Stirling number of the second kind. This number is
n
an integer, because of Corollary 2.4.15 (b).

(Of course, the expression “sur (m, n) /n!” is to be interpreted as (sur (m, n)) /n!.)
Here is the most explicit formula known for sur (m, n):

Theorem 2.4.17. Let m ∈ N and n ∈ N. Then,

n  
n m
sur (m, n) = ∑ (−1) n −i
i
i .
i =0

The most conceptual proof of Theorem 2.4.17 uses the Principle of Inclusion and
Exclusion, which we have not learned yet; this proof will appear in Subsection
2.9.4165 . But there is also an alternative proof, which requires nothing that we have
not already seen:

Exercise 2.4.4. Prove Theorem 2.4.17 by strong induction on m, using Proposition

2.4.13.

Applying Theorem 2.4.17 to n = 3, we find that each m ∈ N satisfies

3  
3 m
sur (m, 3) = ∑ (−1) 3− i
i
0m +3 · |{z}
i = − |{z} 1m −3 · 2m + 3m
i =0 =[m=0] =1
m m
= 3 − 3 · 2 + 3 − [ m = 0] .

Likewise, we can recover parts (a), (b) and (c) of Proposition 2.4.12 by applying
Theorem 2.4.17 to n = 0, n = 1 and n = 2, respectively.
The following exercise gives some variations on the formula in Theorem 2.4.17:

165 This proof also appears in [18s-hw3s, Exercise 2 (b)].

Enumerative Combinatorics: class notes page 224

Exercise 2.4.5. Let m be a positive integer. Let n ∈ N.

(a) Prove that
n
n − i n − 1 m −1
 
sur (m, n) = n ∑ (−1) i .
i =0
i − 1
(b) Prove that
n
im
 
m
= ∑ (−1)n−i .
n i =0
i! ( n − i ) !

2.5. 1m + 2m + ... + nm
We have still not proved Theorem 1.2.10. But we are getting close. The crucial tool
will be the following fact:

Theorem 2.5.1. Let k ∈ N and m ∈ N. Then,

m  
k
k = ∑ sur (m, i ) ·
m
.
i =0
i

Proof of Theorem 2.5.1. We shall prove the theorem by double counting. Let us com-
pute the # of all maps f : [m] → [k ]. We shall compute this # in two ways:
1st way: We have

(# of maps f : [m] → [k]) = (# of maps from [m] to [k])

= km (170)

(by Theorem 2.4.1, applied to k, [m] and [k ] instead of n, A and B).

2nd way: We can construct a map f : [m] → [k ] by the following method:

• First, we choose the number | f ([m])| (this is the size of the image of f , i.e.,
the # of distinct values of f ). This is an integer in {0, 1, . . . , m} 166 . Let us
call this integer i.

  we choose the set f ([m]) (this is the set of all values of f ). There are
• Then,
k
many choices for this (since f ([m]) must be an i-element subset of [k ]).
i
• Finally, we choose the map f itself. The image of this map f must be the
already chosen i-element set f ([m]). Thus, what we are choosing is essentially
a surjection from the m-element set [m] to the already chosen i-element set
f ([m]). Hence, there are sur (m, i ) many choices here (by Proposition 2.4.11).

166 since the set f ([m]) = f ({1, 2, . . . , m}) = { f (1) , f (2) , . . . , f (m)} clearly has at most m elements
Enumerative Combinatorics: class notes page 225

Thus, by the dependent product rule, we obtain

m  
k
(# of maps f : [m] → [k]) = ∑ · sur (m, i ) . (171)
i =0
i

(See Exercise 2.5.1 below for a rigorous presentation of this proof.)

Comparing (170) with (171), we obtain
m   m  
k k
k =∑
m
· sur (m, i ) = ∑ sur (m, i ) · .
i =0
i i =0
i

This proves Theorem 2.5.1.

Exercise 2.5.1. Prove (171) rigorously.

Theorem 2.5.2. Let n ∈ N and m ∈ N. Then,

n m  
n+1
∑ k = ∑ sur (m, i) · i + 1 .
m

k =0 i =0

Proof of Theorem 2.5.2. We have

n n m   m n  
k k
∑ k m
|{z} = ∑ ∑ sur (m, i ) ·
i
= ∑ ∑ sur (m, i ) ·
i
k =0   k =0 i =0 i =0 k =0
m k | {z } | {z   }
= ∑ sur(m,i )· m n
i =0 i =∑ ∑ n k
(by Theorem 2.5.1) i =0 k =0 =sur(m,i )· ∑
(here, we interchanged k =0 i
the summation signs,
by the Fubini principle)
m n
  m  
k n+1
= ∑ sur (m, i ) · ∑ i = ∑ sur (m, i ) · .
i =0 k =0 i =0
i+1
  | {z }  
0 1 n
= + +···+
i i  i
n+1
=
i+1
(by Theorem 1.3.29,
applied to k=i)

This proves Theorem 2.5.2.

Proof of Theorem 1.2.10. Forget that we fixed k. Instead, let m be a positive integer.
Theorem 2.5.2 yields
n m  
n+1
∑ k = ∑ sur (m, i) · i + 1 .
m

k =0 i =0
Enumerative Combinatorics: class notes page 226

Comparing this with

n n
∑ km = 0m
|{z} +1m + 2m + · · · + n m = 1m + 2m + · · · + n m = ∑ im ,
k =0 =0 i =1
(since m>0)

we obtain
n m  
n+1
∑ i = ∑ sur (m, i) · i + 1 .
m
i =1 i =0
Now, forget that we fixed m. We thus have proved that
n m  
n+1
∑ i = ∑ sur (m, i) · i + 1
m
for each positive integer m.
i =1 i =0

Renaming m as k in this statement, we conclude that

n k  
n+1
∑ i = ∑ sur (k, i) · i + 1
k
for each positive integer k.
i =1 i =0

This proves Theorem 1.2.10.

An alternative, combinatorial proof of Theorem 1.2.10 (by double counting) can
be found in [Galvin17, Proposition 23.2] (but, in a sense, is merely a translation of
our above proof into the language of counting).

Class of 2019-10-25

2.6. The Vandermonde convolution

2.6.1. The Vandermonde convolution theorem
The next theorem is one of the most fundamental identities for binomial coeffi-
cients:

Theorem 2.6.1 (The Vandermonde convolution, or the Chu–Vandermonde iden-

tity). Let n ∈ N and x, y ∈ R. Then,
  n   
x+y x y
= ∑ (172)
n k =0
k n−k
  
x y
=∑ . (173)
k
k n−k

Here, the summation sign “∑” on the right hand side of (173) means a sum
k
over all k ∈ Z. (We are thus implicitly claiming that this sum over all k ∈ Z is
well-defined, i.e., that it has only finitely many nonzero addends.)
Enumerative Combinatorics: class notes page 227

Note that (172) is precisely Theorem 1.3.37.

Remark 2.6.2. Before proving Theorem 2.6.1, let us explain why the right hand
sides of (172) and (173) are equal. Indeed, fix n ∈ N andx, y ∈ R. Now, consider
x
any k ∈ Z. If k < 0, then we have k ∈ / N and thus = 0 (by (43)), and
   k
x y
thus the product is 0. But if k > n, then we have n − k ∈/ N and
  k n−k   
y x y
thus = 0 (again by (43)), and thus the product is again
n−k    k n−k
x y
0. Thus, the product is 0 whenever the integer k satisfies either
k n − k   
x y
k < 0 or k > n. In other words, = 0 whenever k ∈ / {0, 1, . . . , n}
k n−k
(because an integer k satisfies either k < 0 or k > n if and only if it satisfies
k ∈/ {0, 1, . . . , n}). Therefore, even though the sums on the right hand sides of
(172) and (173) differ in some addends, these addends in which they differ are
all 0. Thus, these sums are equal.
0
We shall use the symbol “=” (an equality sign with a 0 on top of it) to describe
situations like this. More specifically: If A and B are two sums that only differ in
0 0
addends that are 0, then we will write A = B. (Of course, A = B implies A = B,
which is why we are using the equality sign.) Thus, the argument we have just
made shows that
n      
x y x y
∑ k n−k = ∑ k n−k .
0

k =0 k
Likewise,
n n
∑k= ∑k
0
for each n ∈ N.
k =0 k =1
Likewise, each n ∈ N satisfies
n   n +1      
n 0 n 0 n 0 n
∑ k =∑ k = ∑ k =∑ k
k =0 k =0 k ∈N k ∈Z
 
n
(since = 0 whenever k ∈ / {0, 1, . . . , n}). Likewise, the equality that we spent
k
the main part of our proof of Corollary 1.3.30 arguing can be written as
               
0 1 2 n 0 k k+1 k+2 n
+ + +···+ = + + +···+ .
k k k k k k k k

We will give two proofs of Theorem 2.6.1 and a reference to a third one; but none
of our two proofs will be complete right away. Instead, our first proof will only
Enumerative Combinatorics: class notes page 228

prove Theorem 2.6.1 in the particular case when x ∈ N, and our second proof will
only prove Theorem 2.6.1 in the (even more particular) case when both x and y
belong to N. Later – in Subsection 2.6.3 – we will see how to regain the generality
of Theorem 2.6.1 and transform these proofs into complete proofs of Theorem 2.6.1.
First proof of Theorem 1.3.37 for x ∈ N (informal version). Let u ∈ R and v ∈ N. Then,
Theorem 1.3.8 yields
u−1 u−1
     
u
= + .
n n−1 n
Let us now apply Theorem 1.3.8 again to each addend on the right hand side of
this equation:
u−1 u−1
     
u
= +
n n−1 n
 | {z  }  | {z  }
u−2 u−2 u−2 u−2
 
= + = +
n−2 n−1 n−1 n
(by Theorem 1.3.8) (by Theorem 1.3.8)
u−2 u−2 u−2
     
= +2 +
n−2 n−1 n
 | {z  }  | {z  }  | {z  }
u−3 u−3 u−3 u−3 u−3 u−3
  
= + = + = +
n−3 n−2 n−2 n−1 n−1 n
(by Theorem 1.3.8) (by Theorem 1.3.8) (by Theorem 1.3.8)
u−3 u−3 u−3 u−3
       
= +3 +3 +
n−3 n−2 n−1 n
u−4 u−4 u−4 u−4 u−4
         
= +4 +6 +4 +
n−4 n−3 n−2 n−1 n
(by applying Theorem 1.3.8 to each addend again) .
We can keep applying Theorem 1.3.8 like this indefinitely. After v steps, we
obtain an equality of the form
u−v u−v u−v u−v
         
u
= αv,v + αv,v−1 + · · · + αv,1 + αv,0
n n−v n − ( v − 1) n−1 n
v
u−v
 
= ∑ αv,k , (174)
k =0
n−k

where αv,0 , αv,1 , . . . , αv,v are certain numbers. Due to how we are combining terms,
we see that these numbers satisfy

αv,0 = 1, αv,v = 1 and αv,i = αv−1,i + αv−1,i−1 for each i ∈ [v − 1]

u − ( v − 1) u − ( v − 1) u−v
     
(because both and spawn a term when
n−i n − ( i − 1) n−i
rewritten using Theorem 1.3.8). This is a recurrence that allows us to compute
Enumerative Combinatorics: class notes page 229

the αv,i for all v and 

i; but
 more importantly, this is the same recurrence that the
v
binomial coefficients satisfy:
i

v−1 v−1
         
v v v
= 1, = 1, and = + for each i ∈ [v − 1]
0 v i i i−1

(by Theorem 1.3.8). Hence, a straightforward induction on v shows that

 
v
αv,k = for each k ∈ {0, 1, . . . , v} .
k

Thus, (174) rewrites as

v  
u−v
  
u v
= ∑ .
n k =0
k n − k

Now, forget that we fixed u and v. We thus have proved the identity
v  
u−v
  
u v
= ∑ (175)
n k =0
k n−k

for any u ∈ R and v ∈ N.

Now, let x ∈ N and y ∈ R. Applying (175) to u = x + y and v = x, we get167
x   x  
( x + y) − x
   
x+y x x y
= ∑ = ∑
n k =0
k n−k k n−k
   k=0   
x y x
= ∑
0
since = 0 whenever k > x
k ∈N
k n−k k
n       
x y y
= ∑
0
since = 0 whenever k > n .
k =0
k n − k n − k

Thus, (172) is proved. Hence, (173) follows as well168 . This proves Theorem 2.6.1
for x ∈ N.
The above proof was informal, but it is easy to formalize:

Exercise 2.6.1. Transform the above informal proof of Theorem 2.6.1 into a rig-
orous proof. (Of course, this proof will still require x ∈ N.)

Now, let us come to our second proof of Theorem 2.6.1. As I said, it is even less
complete than the first one: it only covers the case when x, y ∈ N. However, it is
neater:
167 Here, 0
we are using the “=” notation that we introduced in Remark 2.6.2.
168 because we have seen in Remark 2.6.2 that the right hand sides of (172) and of (173) are equal
Enumerative Combinatorics: class notes page 230

Second proof of Theorem 2.6.1 for x, y ∈ N (sketched). Let x, y ∈ N. We shall prove

(172) by double counting. How many ways are there to choose an n-element subset
of the set {1, 2, . . . , x } ∪ {−1, −2, . . . , −y} ? Let us answer this question in two
ways:  
x+y
1st way: The answer is (by Theorem 1.3.12, since {1, 2, . . . , x } ∪ {−1, −2, . . . , −y}
n
is an ( x + y)-element set).
2nd way: We choose an n-element subset of the set {1, 2, . . . , x } ∪ {−1, −2, . . . , −y}
as follows:
• First, we decide how many positive elements our subset will have. Let’s say
it will have k positive elements (with k ∈ {0, 1, . . . , n}).
 
x
• Then, we choose these k positive elements. (There are choices for them,
k
since they must belong to the x-element set {1, 2, . . . , x }.)
• Then, we choose
 the remaining n − k negative elements of our subset. (There
y
are choices for them, since they must belong to the y-element set
n−k
{−1, −2, . . . , −y}).
  
n x y
Thus, the answer is ∑ . (Formally speaking, this follows by applying
k =0 k n−k
the sum rule and the product rule.)  
x+y
Now, we can compare the two answers we have found. We obtain =
   n
n x y
∑ . This proves (172) when x, y ∈ N. Hence, (173) follows as well169
k =0 k n − k
(for x, y ∈ N). Therefore, Theorem 2.6.1 is proved for x, y ∈ N.
(See [19s, proof of Lemma 2.17.15] for a more detailed version of this proof.)
A third proof of  Theorem  be obtained by induction on n, using the
2.6.1 can
y y−1
 
y
absorption identity = . See [Grinbe15, first proof of Theorem 3.29]
n n n−1
for this proof. Unlike our first two proofs above, it proves Theorem 2.6.1 in full
generality (without any extra assumptions on x and y); but on the flipside, it is
somewhat laborious and unmemorable.
Before I explain how the above two proofs above can be extended to the general
case (x, y ∈ R), let me draw a couple conclusions from the theorem:
Corollary 2.6.3. Let x ∈ R and y ∈ N. Then,
y     
x y x+y
∑ k k = y .
k =0

169 because we have seen in Remark 2.6.2 that the right hand sides of (172) and of (173) are equal
Enumerative Combinatorics: class notes page 231

Proof of Corollary 2.6.3. We have

y     y     
x y x y x+y
∑ k k
= ∑
k y−k
=
y
k =0 k =0
|{z} 
y
=
y−k
(by Theorem 1.3.11,
applied to n=y)

(by (172), applied to n = y). This proves Corollary 2.6.3.

Exercise 2.6.2. We have so far not proved Theorem 2.6.1 in full generality. How-
ever, our first proof of Theorem 2.6.1 above proves the particular case of Theorem
2.6.1 for x ∈ N.
Give an alternative proof of Corollary 2.6.3 which relies only on this particular
case.

Corollary 2.6.4. Let n ∈ N. Then,

n  2  
n 2n
∑ k = n .
k =0

Proof of Corollary 2.6.4. We have

n  2 n     
n n n n+n
∑ k =∑ k k = n
k =0 k =0
(by Corollary 2.6.3 (applied to x = n and y = n))
 
2n
= .
n
This proves Corollary 2.6.4.
 2
n
n
Remark 2.6.5. Corollary 2.6.4 gives an explicit formula for ∑ , while
k =0 k
 1
n n
Corollary 1.3.27 gives an explicit formula for ∑ . Is there a formula for
k =0 k
 3
n n
∑ as well? No explicit formula is known, but some recursions can be
k =0 k
 3
n n
found. For example, if we set an = ∑ for each n ∈ N, then
k =0 k
 
2
(n + 1) an+1 = 7n + 7n + 2 an + 8n2 an−1
2
for each n ≥ 1.
Enumerative Combinatorics: class notes page 232

(See [Stanle01, Exercise 6.54 b] for this formula, which was found by Franel in
1894. Recursions
r like this, although increasingly more complicated, exist for all
n n
∑ . Note that the sequence ( a0 , a1 , a2 , . . .) is OEIS sequence A000172.)
k =0 k
The situation
 r is somewhat nicer for alternating sums – i.e., sums of the forms
n
k n
∑ (−1) . Indeed, there are simple formulas for these sums for all r ∈
k =0 k
{1, 2, 3}. We have already seen the formula for r = 1 (this is Proposition 1.3.28),
and will soon see a formula for r = 2 (in Exercise 2.8.1 (c)). The formula for
r = 3 is a special case of Dixon’s identity ([Ward91, (∗)]), which we may see later
too.

Exercise 2.6.3. Explain why Exercise 2.1.1 (a) is also a particular case of Theorem
1.3.37.
[Hint: Apply Theorem 1.3.37 to y = −1.]

2.6.2. The polynomial identity trick

We shall next discuss a “simple” trick that will help us complete the first two proofs
of Theorem 2.6.1 (i.e., to prove the theorem in full generality, not just for x, y ∈ N).
Essentially, this trick allows us to argue that if a certain kind of identity involving
a variable x holds for all x ∈ N, then it must also hold for all x ∈ R. This is the
“polynomial identity trick”.
Let me begin with a reminder on polynomials. I will not formally define poly-
nomials here (I might get to it later, in the chapter on generating functions), but
merely recall the main features of this concept. Polynomials are not functions, but
you can nevertheless substitute numbers into them. Informally, a polynomial (with
real coefficients, in 1 variable X) is a “formal expression” of the form

αX a + βX b + γX c + · · · + ωX z with α, β, γ, . . . , ω ∈ R and a, b, c, . . . , z ∈ N

– that is, a finite sum170 of terms of the form

(a real number) · X (a nonnegative integer) .

For example,

3 1
2X 3 + 7X 5 , X + 6X 0 , 7X 5 , (−2) X 0 , 0
2
are polynomials (where 0 is understood to be an empty sum). Two such expressions
are understood to be equal to each other if they can be transformed into one another
by the following rules:

170 possibly an empty sum

Enumerative Combinatorics: class notes page 233

• The terms can be swapped at will (i.e., we can replace ϕX n + ψX m by ψX m +

ϕX n ).
• Terms involving the same power of X can be combined (i.e., we can replace
ϕX n + ψX n by ( ϕ + ψ) X n ). This is called “combining like terms”. The oppo-
site operation (i.e., splitting a term) can also be done.
• Terms of the form 0X m can be removed or introduced at will.
We write X 0 as 1 (so the term αX 0 is written as α), and we write X 1 as X.
If P is a polynomial, then we can write P as a finite sum a0 X 0 + a1 X 1 + · · · + an X n
for some n ∈ N and some real numbers a0 , a1 , . . . , an . (Indeed, we can transform
P into this form by swapping the terms until they are in the order of increasing
exponents171 , then combining like terms until no two terms with the same expo-
nent remain, and finally introducing 0X m terms wherever necessary.) Moreover,
the representation of P as such a sum is unique up to “trailing zeroes”: We can
always make n larger by introducing 0X m terms (for example, 2X 0 + 7X 1 can be
rewritten as 2X 0 + 7X 1 + 0X 2 or as 2X 0 + 7X 1 + 0X 2 + 0X 3 ), but other than that the
representation of P is unique.
We can also write a polynomial P as an infinite sum a0 X 0 + a1 X 1 + a2 X 2 + · · · ,
where a0 , a1 , a2 , . . . are real numbers such that only finitely many of them are
nonzero. (Indeed, we obtain this representation by writing P as a finite sum
a0 X 0 + a1 X 1 + · · · + an X n and then setting an+1 = an+2 = an+3 = · · · = 0.)
If P is written in such a form a0 X 0 + a1 X 1 + a2 X 2 + · · · , then ai is called the i-th
coefficient of P (or the coefficient of X i in P) whenever i ∈ N.
The zero polynomial is the polynomial 0 (that is, the polynomial whose all coeffi-
cients are 0). A polynomial is said to be nonzero if it is not the zero polynomial.
Addition of polynomials is defined by just throwing their terms together: e.g.,
   
αX + βX + γX + δX = αX a + βX b + γX c + δX d .
a b c d

Subtraction is defined like addition, except that the terms of the subtrahend get
their signs flipped:
   
αX a + βX b − γX c + δX d = αX a + βX b + (−γ) X c + (−δ) X d .

Multiplication is defined by the distributivity law and the rule (αX a ) βX b =

(αβ) X a+b . For example,

  
αX a + βX b γX c + δX d
      
a c a d b c b d
= (αX ) (γX ) + (αX ) δX + βX (γX ) + βX δX
| {z } | {z } | {z } | {z }
=αγXa+c
=αδX a+d = βγX b+c = βδX b+d

= αγX a+c + αδX a+d + βγX b+c + βδX b+d .

171 The “exponent” of a term βX b is understood to be the integer b.
Enumerative Combinatorics: class notes page 234

We cannot always divide a polynomial by another polynomial; but we can divide a

P
polynomial by a nonzero real number. Namely, is defined to be r −1 · P whenever
r
r is a nonzero real number and P is a polynomial. Furthermore, if n ∈ N and if P
is a polynomial, then Pn (that is, the n-th power of P) is defined to be the product
· · · P}.
|PP{z
n times
The degree of a nonzero polynomial P is the largest i ∈ N such that the i-th
coefficient of P is nonzero. For example, the degree of 2X 2 + 7X 5 − X is 5. (We say
that the degree of the zero polynomial 0 is −∞.)
Substituting a number (or square matrix, or another polynomial) x into a polyno-
mial P = αX a + βX b + γX c + · · · yields αx a + βx b + γx c + · · · . This result is called
P ( x ).
The discussion of polynomials we just had is not fully rigorous and self-contained.
We could make it rigorous by formally defining the notion of “formal expression”,
but we would have to verify that it behaves as nicely as we would hope for, meaning
that the following facts are true (among others):

• The representation of a given polynomial P as a finite sum a0 X 0 + a1 X 1 +

· · · + an X n is unique up to “trailing zeroes”.
• The representation of a given polynomial P as an infinite sum a0 X 0 + a1 X 1 +
a2 X 2 + · · · is unique. (This is necessary to ensure that the i-th coefficient of P
is well-defined.)

• Our definitions of addition, subtraction and multiplication of polynomials are

well-defined. (This means, for example, that the product PQ does not depend
on how P and Q are expressed. For example, if we “combine like terms” or
remove a 0X m term from P, then the product PQ also stays the same.)

• The result of substituting a number x into a polynomial P does not depend

on how the polynomial is expressed.

All of these facts are true and can be proven, but it requires some work. Thus,
it is more common to define polynomials in a different way, namely as infinite
sequences ( a0 , a1 , a2 , . . .) of real numbers having only finitely many nonzero en-
tries. (Such a sequence ( a0 , a1 , a2 , . . .) corresponds to the polynomial a0 X 0 + a1 X 1 +
a2 X 2 + · · · .) This definition of polynomials can be found in [Grinbe15, §1.5] or
[Loehr11, Sections 7.1–7.3] or [19s, §7.4] or most good algebra textbooks; we will
also outline it in the generating functions chapter of this course.
We shall silently identify each number u with the polynomial uX 0 (which is
called a constant polynomial, and has degree 0 when u 6= 0 and degree −∞ when
u = 0).
Let us now discuss roots of polynomials:
Enumerative Combinatorics: class notes page 235

Definition 2.6.6. A number x (say, a rational or real or complex number) is a root

of a polynomial P if and only if P ( x ) = 0.

A crucial property of polynomials is that they cannot have too many roots (unless
they are the zero polynomial):

Theorem 2.6.7. Let n ∈ N. Then, a nonzero polynomial of degree ≤ n has ≤ n

roots.
Here, “polynomial” means “polynomial with real coefficients in 1 variable X”, and
“root” means “real root”. However, the same theorem can be stated using rational
or complex numbers instead (and the same proofs apply).

Example 2.6.8. (a) The polynomial X 2 − 2X − 3 has degree 2; thus, Theorem 2.6.7
(applied to n = 2) shows that it has ≤ 2 roots. And indeed, it has 2 roots: 3 and
−1.
(b) The polynomial X 2 − 2X + 1 has degree 2; thus, Theorem 2.6.7 (applied to
n = 2) shows that it has ≤ 2 roots. And indeed, it has 1 root: 1. (We could say
that 1 is a “double root” of this polynomial, since X 2 − 2X + 1 = ( X − 1)2 ; but
this requires the somewhat subtle concept of “double roots” and, more generally,
“multiple roots”, which I don’t want to introduce here.)

Theorem 2.6.7 is often called the “easy half of the Fundamental Theorem of Al-
gebra”. We will not prove it here, but proofs of this theorem are not hard to find.
I believe any good textbook on abstract algebra contains a proof of Theorem 2.6.7
somewhere. (In particular, proofs appear in [Goodma15, Corollary 1.8.24], [Joyce17,
Theorem 1.58], [Walker87, Corollary 4.5.10], [CoLiOs15, Chapter 1, §5, Corollary 3],
[19s, Theorem 7.6.11], [Elman19, Corollary 33.7] and [Knapp16, Corollary 1.14].)
The following corollary of Theorem 2.6.7 is almost trivial:

Corollary 2.6.9. If a polynomial P has infinitely many roots, then P is the zero
polynomial.

(Again, “polynomial” means “polynomial with real coefficients in 1 variable X”,

and “root” means “real root” here.)
Proof of Corollary 2.6.9. Let P be a polynomial that has infinitely many roots. We
must prove that P is the zero polynomial. Indeed, assume the contrary. Thus, P
is nonzero. Let n be the degree of P. Thus, P is a nonzero polynomial of degree
≤ n. Hence, Theorem 2.6.7 shows that P has ≤ n roots. But P has > n roots (since
P has infinitely many roots). The previous two sentences contradict each other.
This contradiction shows that our assumption was false. Hence, Corollary 2.6.9 is
proven.
A corollary of this corollary will prove the most useful to us:
Enumerative Combinatorics: class notes page 236

Corollary 2.6.10. Let P and Q be polynomials (with real coefficients in 1 variable

X). Assume that
P (x) = Q (x) for all x ∈ N. (176)
Then, P = Q.

Proof of Corollary 2.6.10. Consider the polynomial P − Q. For each x ∈ N, we have

( P − Q) ( x ) = P ( x ) − Q ( x ) = 0 (by (176)) .
In other words, each x ∈ N is a root of P − Q. Hence, P − Q has infinitely many
roots (since there are infinitely many x ∈ N). Thus, Corollary 2.6.9 (applied to
P − Q instead of P) shows that P − Q is the zero polynomial. In other words,
P − Q = 0. Hence, P = Q. This proves Corollary 2.6.10.

2.6.3. Salvaging the proofs of Theorem 2.6.1

How does Corollary 2.6.10 help us prove Theorem 2.6.1? What do the binomial
coefficients appearing in Theorem 2.6.1 have to do with polynomials?
To understand this, let us extend our definition of binomial coefficients some-
what:
 
n
Definition 2.6.11. In Definition 1.3.3, we defined the binomial coefficient
k
whenever n and k are numbers. Let us now extend this definition to the case
when n is a polynomial (although k should still be a number).

Since polynomials are not usually denoted byn,let me restate this definition: If
P
P is any polynomial and k is any number, then is a polynomial defined by
k
P ( P − 1) ( P − 2) · · · ( P − k + 1)
 
P
= if k ∈ N
k k!
and  
P
=0 / N.
if k ∈
k
Note the following fact:

Proposition 2.6.12. Let P be a polynomial,

  and let n and k betwo numbers.
 Then,
P P (n)
substituting n into the polynomial yields the number .
k k

This proposition is intuitively clear (it is just saying that it does notmatter

P
whether we substitute n for X before computing the binomial coefficient or
k
after it). For a formal proof, see [19s, Corollary 7.6.16 (c)].
Enumerative Combinatorics: class notes page 237

Let us now come back to our first proof of Theorem 2.6.1. In that proof, we only
showed that Theorem 2.6.1 holds for all x ∈ N. We shall now extend it to all real
values of x:
Completion of the first proof of Theorem 2.6.1. Fix y ∈ R and n ∈ N. We have already
proved that Theorem 2.6.1 holds for x ∈ N. Thus, the equality
  n   
x+y x y
= ∑ (177)
n k =0
k n−k
holds for all x ∈ N. We now want to prove this equality for all x ∈ R.
Define two polynomials P and Q (in 1 variable X, with real coefficients) by
  n   
X+y X y
P= and Q= ∑ .
n k =0
k n − k
These are well-defined polynomials, and can be written explicitly as follows:
( X + y ) ( X + y − 1) ( X + y − 2) · · · ( X + y − n + 1)
 
X+y
P= =
n n!
and
n n
X ( X − 1) ( X − 2) · · · ( X − k + 1)
    
X y y
Q= ∑ = ∑ .
k =0
k n−k k =0
k! n−k
Now, for each x ∈ R, we have
    
x+y X+y
P (x) = by Proposition 2.6.12, since P =
n n
and
n   
x y
Q (x) = ∑ (likewise) .
k =0
k n−k
In view of these two equalities, we can rewrite (177) as
P (x) = Q (x) .
Hence, we know that P ( x ) = Q ( x ) holds for all x ∈ N (because we know that (177)
holds for all x ∈ N). Therefore, Corollary 2.6.10 yields P = Q. Thus, P ( x ) = Q ( x )
for all x ∈ R. In other words,
  n   
x+y x y
= ∑
n k =0
k n−k
    
x+y n x y
for all x ∈ R (since P ( x ) = and Q ( x ) = ∑ for each x ∈ R).
n k =0 k n−k
Thus, we have proved the equality (172) for all x ∈ R. Hence, (173) holds for all
x ∈ R as well172 . Therefore, Theorem 2.6.1 is proved (this time for all values of x
and y).
172 because we have seen in Remark 2.6.2 that the right hand sides of (172) and of (173) are equal
Enumerative Combinatorics: class notes page 238

Our use of Corollary 2.6.10 in the proof we just gave is an instance of what I call
the “polynomial identity trick”: We wanted to prove the equality P ( x ) = Q ( x ) for
all x ∈ R, where P and Q were two given polynomials. We did this as follows:
We first proved it for all x ∈ N; then, we used this to conclude that P = Q (by
Corollary 2.6.10); finally we substituted x in P = Q to obtain P ( x ) = Q ( x ) for all
x ∈ R. What made this argument possible is that the equality that we were aiming
to show (in our case, (177)) was an equality between two polynomials in x (that
is, we could rewrite it as P ( x ) = Q ( x ) for two polynomials P and Q). Corollary
2.6.10 allows us to use this kind of argument in proving any such equality. This
is called the “polynomial identity trick” or (in [GrKnPa94, §5.1]) the “polynomial
argument”.
We can also salvage our second proof of Theorem 2.6.1 in a similar way. Here we
need to apply the polynomial identity trick twice:

• Step 1: Fix y ∈ N and n ∈ N. Use the same argument as before to prove that
(177) holds for all x ∈ R. Thus, (177) is proved for all x ∈ R and y ∈ N.
• Step 2: Fix x ∈ R and n ∈ N. Use an analogous argument (using y instead of
x) to prove that (177) holds for all y ∈ R.

See [19s, §2.17.3] for the details of this argument173 , and [Grinbe15, Second proof
of Theorem 3.30] for a different variant of this argument (using polynomials in 2
variables).

2.6.4. More consequences of the polynomial identity trick

The polynomial identity trick can be applied to several other identities. For exam-
ple:

• Proposition 1.3.35 (the trinomial revision formula) says that

n−b
     
n a n
= (178)
a b b a−b
for all n, a, b ∈ R. Our second proof of this proposition (in Subsection 2.2.2)
only showed that it holds for all n ∈ N. But using the polynomial identity
trick, we can now complete it to obtain a proof of Proposition 1.3.35 for all
n ∈ R: Namely, we fix a, b ∈ R, and we rewrite the equality (178) as

P (n) = Q (n) ,

where the polynomials P and Q are defined by

X−b
     
X a X
P= and Q= .
a b b a−b
173 I use Q instead of R in [19s, §2.17.3], but otherwise the argument proceeds exactly in the same
way.
Enumerative Combinatorics: class notes page 239

Our second proof of Proposition 1.3.35 thus shows that P (n) = Q (n) for
all n ∈ N. Hence, using the polynomial identity trick, we conclude that
P (n) = Q (n) for all n ∈ R. This completes the second proof of Proposition
1.3.35.

• Proposition 1.3.28 says that

n  
n
∑ (−1) k = [n = 0]
k
(179)
k =0

holds for all n ∈ N. Can we use the polynomial identity trick to generalize
this to all n ∈ R ? No, because n appears as a summation bound on the left
hand side, and it would make no sense to replace it by a non-integer number
√
2
(because what would “ ∑ ” even mean?).
k =0
We could try to sneak past this issue by rewriting the left hand side somewhat.
For any n ∈ N, we can rewrite the identity (179) as
 
k n
∑ (−1) k = [n = 0] (180)
k ∈N
   
n n 0
k k n
(because ∑ (−1) = ∑ (−1) ). But even in this form, we cannot
k =0 k k ∈N k
generalize it to n ∈ R using the polynomial identity trick, because the right
hand side of (180) cannot be written as a polynomial in n. In other words,
there exists no polynomial Q such that

Q ( n ) = [ n = 0] for all n ∈ N

(in fact, such a polynomial Q would have infinitely many roots, contradicting
Corollary 2.6.9).
Note that we cannot generalize (180) to all n ∈ R in any way (by the polyno-
mial identity trick or otherwise), because if we try to substitute n = −1 into
(180), then the left hand side will become

k −1
 
∑ (−1) k = ∑ (− 1)k (−1)k = ∑ 1,
k ∈N | {z } k∈N k ∈N
| {z }
k+k
=(−1) =1
=(−1)k (since k+k=2k is even)
(by (47))

which is an ill-defined infinite sum (with infinitely many nonzero

 addends).
n
More generally, for any n ∈ R \ N, the infinite sum ∑ (−1)k will have
k ∈N k
infinitely many nonzero addends (unlike for n ∈ N).
Enumerative Combinatorics: class notes page 240

• Theorem 2.5.1 claims that

m  
k
k = ∑ sur (m, i ) ·
m
(181)
i =0
i
for all k ∈ N and m ∈ N. Can we use the polynomial identity trick to
generalize this to all k ∈ R ? What about generalizing it to all m ∈ R ?
The answer to the second question is negative, because km is clearly not a
polynomial in m (that is, there exists no polynomial P such that P (m) = km
for all m ∈ N, where k is fixed). Thus, m has to remain a nonnegative integer.
But the answer to the first question is positive. Indeed, if we fix m ∈ N, then
both sides of (181) are polynomials in k; that is, we can rewrite the equality
(181) as
m  
X
P (k) = Q (k) , where P = X and Q = ∑ sur (m, i ) ·
m
.
i =0
i
From Theorem 2.5.1, we know that this equality holds for all k ∈ N. Thus,
the polynomial identity trick shows that it holds for all k ∈ R. Thus, we have
proved that
m  
k
k = ∑ sur (m, i ) ·
m
(182)
i =0
i
for all k ∈ R and m ∈ N. This is a significant generalization of Theorem 2.5.1.
• Exercise 2.1.1 (a) claims the equality
m
m n−1
   
k n
∑ (−1) k = (−1) m
k =0

for all n ∈ R and m ∈ N. We  cannot generalize this equality to m ∈ R

n − 1
(because, for example, (−1)m is not a polynomial in m). But the
m
polynomial identity trick shows that if we had only proved it for n ∈ N, we
could automatically conclude that it holds for all n ∈ R (since both sides are
polynomials in n). (But we have no need for this, since we have proved it for
all n ∈ R already.)
• We cannot generalize Theorem 1.3.29 to all n ∈ R, because n appears as a
summation bound on the left hand side. The polynomial identity trick does
not help here, since the left hand side is not a polynomial in n. (The right hand
side is, but this alone is not sufficient.) We also cannot generalize Theorem
1.3.29 to all k ∈ R,because  it would be false for k = −1 and n = 0 (and,
n+1
of course, because is not a polynomial in k). (It would be true for
k+1
all k 6= −1, but only because it would boil down to 0 = 0 in the stupidest
possible way when k + 1 ∈ / N.) For similar reasons, we cannot generalize
Corollary 1.3.30.
Enumerative Combinatorics: class notes page 241

Class of 2019-10-28

2.6.5. Mutating the Chu–Vandermonde identity

We shall next see some “mutated” versions of the Chu–Vandermonde identity
(Theorem 2.6.1). Here, “mutated” means that these versions are obtained from
the Chu–Vandermonde identity by transforming the binomial coefficients via the
upper negation and symmetry identities174 . One example of such a “mutated”
version is Corollary 2.6.3, since that corollary was proved by applying the Chu–
Vandermonde identity and rewriting one of the binomial coefficients using sym-
metry. Here is another (which we shall prove soon):

Proposition 2.6.13. Let n, x, y ∈ N. Then,

n  
n−k
  
n+1 k
= ∑ .
x+y+1 k =0
x y

This proposition is often called the “upside-down Vandermonde convolution”,

since its right hand side is the right hand side of (172) with all binomial coefficients
turned “upside down”. It behaves rather differently in some aspects, however. In
particular, (172) holds for all x, y ∈ R, whereas Proposition 2.6.13 requires x, y ∈ N
(and indeed, would be false for n = 3, x = −1 and y = 2). The polynomial identity
trick cannot be used to generalize Proposition 2.6.13, since its right hand side is
neither a polynomial in n nor a polynomial in x nor a polynomial in y.
We shall give two proofs of Proposition 2.6.13: one algebraic and one by dou-
ble counting. The algebraic proof will explain why we call it a “mutated” Chu–
Vandermonde identity; the double-counting proof will shine some light on its com-
binatorial meaning. Which is the “right” proof? You decide.
First proof of Proposition 2.6.13. This proof is a close relative of the proof given in
[Grinbe15, proof of Proposition 3.32 (f)] (but is more direct).
We shall first show an auxiliary claim:

Claim 1: Let k ∈ {0, 1, . . . , n} be such that k < x or n − k < y. Then,

n−k
  
k
= 0.
x y

n−k
    
k k
[Proof of Claim 1: Assume the contrary. Thus, 6= 0. Hence, 6= 0 and
x y x
n−k
 
6= 0. Note that k ∈ {0, 1, . . . , n}, so that both k and n − k are nonnegative integers.
y
Thus, k ∈ N and n − k ∈ N.
174 i.e., Proposition 1.3.7 and Theorem 1.3.11
Enumerative Combinatorics: class notes page 242

 
k
From k ∈ N and 6= 0, we obtain k ≥ x, because otherwise, Proposition 1.3.6
x  
k
(applied to k and x instead of n and k) would yield = 0.
x
n−k
 
From n − k ∈ N and 6= 0, we obtain n − k ≥ y, because otherwise, Proposition
y
n−k
 
1.3.6 (applied to n − k and y instead of n and k) would yield = 0.
y
We have thus found k ≥ x and n − k ≥ y. This contradicts the assumption that k < x
or n − k < y. This contradiction shows that our assumption was wrong. Hence, Claim 1 is
proven.]
We are in one of the following two cases:
Case 1: We have n < x + y.
Case 2: We have n ≥ x + y.
Let us first consider Case 1. In this case, we have n < x + y. Therefore, each
k ∈ {0, 1, . . . , n} satisfies either k < x or n − k < y 175 . Hence, each k ∈ {0, 1, . . . , n}
satisfies
n−k
  
k
=0 (183)
x y
(by Claim 1).
Now, n < x + y, thus n + 1 < x + y + 1. Thus, Proposition
 1.3.6 (applied to n + 1
n+1
and x + y + 1 instead of n and k) yields = 0. Comparing this with
x+y+1
n   n
n−k

k
∑ x y
= ∑ 0 = 0,
k =0 | {z } k =0
=0
(by (183))

we obtain
n
n−k
    
n+1 k
= ∑ .
x+y+1 k =0
x y
Thus, Proposition 2.6.13 is proven in Case 1.
Let us now consider Case 2. In this case, we have n ≥ x + y. Thus, n − y ≥ x.
Thus, x ≤ n − y, so that 0 ≤ x ≤ n − y ≤ n (since y ∈ N). Therefore, the integer
interval [ x, n − y] = { x, x + 1, . . . , n − y} is a subset of [0, n] = {0, 1, . . . , n}. Thus,
the elements k ∈ {0, 1, . . . , n} that satisfy k ≥ x and k ≤ n − y are precisely the
elements of { x, x + 1, . . . , n − y}.
Now, each k ∈ {0, 1, . . . , n} satisfies

either (k ≥ x and n − k ≥ y) or (k < x or n − k < y)

175 Proof.
Assume the contrary. Thus, neither k < x nor n − k < y holds. Hence, we have k ≥ x
and n − k ≥ y. Adding these two inequalities together, we obtain k + (n − k ) ≥ x + y. This
contradicts k + (n − k ) = n < x + y. This contradiction shows that our assumption was wrong.
Qed.
Enumerative Combinatorics: class notes page 243

(butnot
both of
 these statements simultaneously). Thus, we can split the sum
n k n−k
∑ as follows:176
k =0 x y
n
n−k
  
k
∑ x y
k =0
n−k n−k
     
k k
= ∑ x y
+ ∑ x y
k∈{0,1,...,n}; k∈{0,1,...,n}; | {z }
k≥ x and n−k≥y k< x or n−k<y =0
| {z } (by Claim 1)
= ∑
k∈{0,1,...,n};
k≥ x and k≤n−y
(since the inequality n−k≥y
is equivalent to k≤n−y)
n−k
  
k
= ∑ x y
+ ∑ 0
k∈{0,1,...,n}; k∈{0,1,...,n};
k≥ x and k≤n−y k< x or n−k<y
n−k
  
k
= ∑ x y
k∈{0,1,...,n};
k≥ x and k≤n−y
n−y 
n−k
 
k
= ∑ x y
(184)
k= x

(because the elements k ∈ {0, 1, . . . , n} that satisfy k ≥ x and k ≤ n − y are precisely

the elements of { x, x + 1, . . . , n − y}).
Now, fix k ∈ { x, x + 1, . . . , n − y}. Thus, x ≤ k ≤ n − y, so that 0 ≤ x ≤ k ≤
n − y ≤ n. Thus, k ∈ N (since 0 ≤ k) and n − k ∈ N (since k ≤ n). Theorem 1.3.11
(applied to k and x instead of n and k) yields

− (−k)
     
k k
= = (since k = − (−k))
x k−x k−x
k− x −k + k − x − 1
 
= (−1)
k−x
 
by Proposition 1.3.7, applied to − k and k − x
instead of n and k
−x − 1
 
= (−1)k−x (185)
k−x

(since −k + k − x − 1 = − x − 1). Furthermore, recall that n − k ∈ N; hence,

n
176 Here we are using the splitting rule (28), along with the fact that the “ ∑ ” sign is a shorthand for
k =0
“ ∑ ”.
k ∈{0,1,...,n}
Enumerative Combinatorics: class notes page 244

Theorem 1.3.11 (applied to n − k and y instead of n and k) yields

n−k n−k − (k − n)
     
= = (since n − k = − (k − n))
y n−k−y n−k−y
n−k−y k − n + n − k − y − 1
 
= (−1)
n−k−y
 
by Proposition 1.3.7, applied to k − n and n − k − y
instead of n and k
−y − 1
 
= (−1)n−k−y (186)
n−y−k

(since k − n + n − k − y − 1 = −y − 1 and n − k − y = n − y − k). Multiplying the

equalities (185) and (186), we obtain

n−k k− x − x − 1 −y − 1
      
k n−k−y
= (−1) (−1)
x y k−x n−y−k
−x − 1 −y − 1
  
k− x n−k−y
= (−1) (−1)
| {z } k−x n−y−k
=(−1)(k− x)+(n−k−y) =(−1)n− x−y
(since (k− x )+(n−k−y)=n− x −y)
−x − 1 −y − 1
  
n− x −y
= (−1) . (187)
k−x n−y−k

Now, forget that we fixed k. We thus have proved the equality (187) for each
Enumerative Combinatorics: class notes page 245

k ∈ { x, x + 1, . . . , n − y}. Now, (184) becomes

n  
n−k

k
∑ x y
k =0
n−y n−y
n−k n− x −y − x − 1 −y − 1
     
k
= ∑ = ∑ (−1)
k= x
x y k= x
k−x n−y−k
| {z  }
−x − 1 −y − 1
 
=(−1)n− x−y
k−x n−y−k
(by (187))
n−y− x
−x − 1 −y − 1
   
= ∑ (−1) n− x −y
(k + x ) − x n − y − (k + x )
k =0 |  {z  } | {z
| {z } }
n− x −y −x − 1 −y − 1
= ∑ = =
k =0 k n−x−y−k
(since n−y− x =n− x −y) (since (k+ x )− x =k) (since n−y−(k+ x )=n− x −y−k)

(here, we have substituted k + x for k in the sum)

n− x −y
n− x −y − x − 1 −y − 1
  
= ∑ (−1)
k =0
k n−x−y−k
n− x −y 
−x − 1 −y − 1
 
= (−1) n− x −y
∑ k n−x−y−k
. (188)
k =0

Let us now look at the sum on the right hand side of this equality more carefully.
It is a sum of products of binomial coefficients, and this time the running index k
of the sum appears only in the bottom arguments of these coefficients. This looks
similar to the right hand side of (172), and indeed we can easily recognize the right
hand side of (172) in our sum, with n − x − y, − x − 1 and −y − 1 substituted for n,
x and y.
Let us see how this works in detail: We have n − x − y ≥ 0 (since n ≥ x + y) and
thus n − x − y ∈ N. Thus, (172) (applied to n − x − y, − x − 1 and −y − 1 instead
of n, x and y) yields
n− x −y 
(− x − 1) + (−y − 1) −x − 1 −y − 1
   

n−x−y
= ∑ k n−x−y−k
.
k =0

In view of (− x − 1) + (−y − 1) = − ( x + y + 2), this rewrites as

n− x −y 
− ( x + y + 2) −x − 1 −y − 1
   

n−x−y
= ∑ k n−x−y−k
. (189)
k =0
Enumerative Combinatorics: class notes page 246

Thus, (188) becomes

n  
n−k

k
∑ x y
k =0
n− x −y 
−x − 1 −y − 1
 
= (−1) n− x −y
∑ k n−x−y−k
k =0
| {z }
− ( x + y + 2)
 
=
n−x−y
(by (189))
− ( x + y + 2)
 
n− x −y
= (−1)
n−x−y
| {z }
− x−y−1

n− x −y
x + y + 2 + n
=(−1)
n−x−y
(by Proposition 1.3.7,
applied to x +y+2 and n− x −y instead of n and k)
x+y+2+n−x−y−1
 
n− x −y n− x −y
= (−1) (−1)
| {z } n−x−y
=(−1)(n− x−y)+(n− x−y) =1
|  {z  }
(since (n− x −y)+(n− x −y)=2(n− x −y) n + 1
is even)
=
n−x−y
(since x +y+2+n− x −y−1=n+1)
   
n+1 n+1
= =
n−x−y ( n + 1) − ( n − x − y )
(by Theorem 1.3.11, applied to n + 1 and n − x − y instead of n and k)
 
n+1
= (since (n + 1) − (n − x − y) = x + y + 1) .
x+y+1

Thus, Proposition 2.6.13 is proven in Case 2.

We have now proved Proposition 2.6.13 in both Cases 1 and 2. Hence, Proposition
2.6.13 always holds.
Second proof of Proposition 2.6.13 (sketched). We shall prove Proposition 2.6.13 by dou-
ble counting. How many ( x + y + 1)-element subsets does the set [n + 1] have? Let
us answer this question in two ways:
1st way: Theorem 1.3.12 (applied to n + 1, x + y + 1 and [n + 1] instead of n, k
and S) shows that

(# of ( x + y + 1) -element subsets of [n + 1])

 
n+1
= . (190)
x+y+1

2nd way: Here is a weird procedure to construct an ( x + y + 1)-element subset U

of [n + 1]:
Enumerative Combinatorics: class notes page 247

• First, we choose the ( x + 1)-st smallest element of the subset U. Let us call it
m. Thus, m ∈ [n + 1].

• Then, we choose the x smallest elements of U. These x smallest elements

must belong to the (m − 1)-element set {1, 2, . . . , m − 1} (since they must be
smaller
 than  the ( x + 1)-st smallest element of U, which is m); thus, there
m−1
are choices for them (by Theorem 1.3.12, applied to m − 1, x and
x
{1, 2, . . . , m − 1} instead of n, k and S).
• Finally, we choose the remaining y elements of U. These y elements must
belong to the (n − m + 1)-element set {m + 1, m + 2, . . . , n + 1} (since they
must be larger  ( x + 1)-st smallest element of U, which is m); thus,
than the
n−m+1

there are choices for them (by Theorem 1.3.12, applied to n −
y
m + 1, y and {m + 1, m + 2, . . . , n + 1} instead of n, k and S).

m−1 n−m+1
  
The total # of ways to make these decisions is clearly ∑ .
m∈[n+1] x y
Thus,

(# of ( x + y + 1) -element subsets of [n + 1])

m−1 n−m+1
  
= ∑ . (191)
m∈[n+1]
x y

Now, comparing (190) with (191), we obtain

m−1 n−m+1
     
n+1
= ∑
x+y+1 m∈[n+1]
x y
| {z }
− −
| {z } 
n +1 n ( m 1 )
= ∑ =
m =1 y
(since n−m+1=n−(m−1))
n +1  n
m−1 n − ( m − 1) n−k
    
k
= ∑ x y
= ∑
x y
m =1 k =0

(here, we have substituted k for m − 1 in the sum). Thus, Proposition 2.6.13 is

proved again.

Exercise 2.6.4. Explain why Theorem 1.3.29 is a particular case of Proposition

2.6.13.

There are several other “mutated versions” of the Chu–Vandermonde identity.

Two of them are given in the following exercises:
Enumerative Combinatorics: class notes page 248

Exercise 2.6.5. Let x, y ∈ R and n ∈ N. Show that

n
x−y k+y−1
    
x
= ∑ (−1) k
.
n k =0
n − k k

Exercise 2.6.6. Let x ∈ N, y ∈ R and n ∈ N be such that x ≤ n. Prove that

n
y−x−1
    
k− x k y
= ∑ (−1) .
n−x k =0
x n − k

Here is another application of the Chu–Vandermonde identity (joint with the

trinomial revision formula):
Exercise 2.6.7. Let a, b ∈ N and x ∈ R. Prove that
   a+b    
x x i a x
a b
= ∑ a a+b−i i
.
i=a

2.7. Counting subsets again

One of the most important facts we have seen so far was Theorem 1.3.12, which
said that if S is an n-element set and k ∈ R, then
 
n
= (# of k-element subsets of S) .
k
We have proved this fact by induction on n. We shall now re-prove it using the sum
rule applied “backwards”. The underlying idea of the argument can be summa-
rized by the saying “to count sheep, first count the legs and then divide by 4”. In
our case, the sheep will be the k-element subsets of S, and their legs will be injective
k-tuples of elements of S (i.e., k-tuples (s1 , s2 , . . . , sk ) ∈ Sk such that s1 , s2 , . . . , sk are
distinct). More precisely, an injective k-tuple (s1 , s2 , . . . , sk ) ∈ Sk will be considered
a “leg” of a sheep W if and only if W = {s1 , s2 , . . . , sk }. What is so useful about this
definition of “legs” is that
• the legs are easier to count than the sheep, and
• each sheep has the same number of legs (although not 4).
This trick for counting is sometimes called the “multijection principle”177 , but it
requires no new theorems; the sum rule is all we need.
Here are the details. First, let us define the injective k-tuples – the “legs” of our
“sheep”:
177 or “strategic overcounting” (but there is yet another trick that shares this name)
Enumerative Combinatorics: class notes page 249

Definition 2.7.1. Let S be a set. Let k ∈ N.

(a) A k-tuple (s1 , s2 , . . . , sk ) ∈ Sk is said to be injective if s1 , s2 , . . . , sk are distinct.
k be the set of all injective k-tuples ( s , s , . . . , s ) ∈ Sk . This is a
(b) Let Sinj 1 2 k
subset of Sk .

For example, the 3-tuple (3, 2, 5) is injective (since 3, 2, 5 are distinct), whereas
the 3-tuple (4, 2, 4) is not (since 4 = 4).
In probability theory, injective k-tuples are called “ordered k-samples without
replacement” (while arbitrary k-tuples are called “ordered k-samples with replace-
ment”).
Warning: Some authors also refer to injective k-tuples (s1 , s2 , . . . , sk ) ∈ Sk as k-
permutations of S. But this is not the meaning that the word “permutation” has in
this text!
We promised that the injective k-tuples (the “legs”) would be easy to count; let
us do this:

Proposition 2.7.2. Let S be a finite set. Let k ∈ N. Then, Sinj

k = |S|k .

Here, we are using the falling factorial notation nk (as defined in Definition 2.4.2).
It is easy to prove Proposition 2.7.2 using the “dependent product rule”178 . How-
ever, we have already made more or less the same argument when we proved The-
orem 2.4.4, so instead of doing it again, I find it easier to piggyback on Theorem
2.4.4:
Proof of Proposition 2.7.2. Note that S is a |S|-element set, whereas [k] is a k-element
set.
The injective k-tuples (s1 , s2 , . . . , sk ) ∈ Sk are in one-to-one correspondence with
the injective maps from [k] to S. More precisely: If f is an injective map from [k ]
to S, then the k-tuple ( f (1) , f (2) , . . . , f (k )) ∈ Sk is injective (since the injectivity
of f forces the values f (1) , f (2) , . . . , f (k ) to be distinct) and thus belongs to Sinjk .

Thus, we can define a map

k
α : {injective maps from [k ] to S} → Sinj ,
f 7→ ( f (1) , f (2) , . . . , f (k )) .

178 Just k can be constructed by first choosing its

argue that an injective k-tuple (s1 , s2 , . . . , sk ) ∈ Sinj
first entry s1 (there are |S| many choices), then choosing its second entry s2 (there are |S| − 1
many choices, since s2 has to be distinct from s1 ), then choosing its third entry s3 (there are
|S| − 2 many choices, since s3 has to be distinct from the two distinct elements s1 and s2 ), and
k must be | S | · (| S | − 1) ·
so on, and therefore the total # of injective k-tuples (s1 , s2 , . . . , sk ) ∈ Sinj
(|S| − 2) · · · · · (|S| − k + 1) = |S|k .
Enumerative Combinatorics: class notes page 250

This map α simply sends every injective map f : [k ] → S to its list of values. It is
easy to see that this map α is a bijection179 . Thus, the bijection principle yields

k
|{injective maps from [k] to S}| = Sinj .

Hence,
k
Sinj = |{injective maps from [k] to S}|
= (# of injective maps from [k] to S) = |S|k

(by Theorem 2.4.4, applied to m = k, n = |S|, A = [k ] and B = S). This proves

Proposition 2.7.2.
We shall now apply this to obtain a second proof of Theorem 1.3.12, as promised:
Second proof of Theorem 1.3.12 (sketched). We must prove the equality
 
n
= (# of k-element subsets of S) .
k

If k ∈/ N, then both sides of this equality are 0 (because S has no k-element subsets
when k ∈ / N), and thus the equality is clearly true. Hence, for the rest of this proof,
we WLOG assume that k ∈ N.
We have |S| = n (since S is an n-element set).
For each injective k-tuple s = (s1 , s2 , . . . , sk ) ∈ Sinj k , we let set s denote the set

{s1 , s2 , . . . , sk } of all entries of s. This set set s is a k-element set (because s1 , s2 , . . . , sk

are distinct180 ) and a subset of S (because s1 , s2 , . . . , sk are elements of S 181 ), and
thus is a k-element subset of S.
In our informal discussion above, set s was playing the role of the “sheep” whose
“leg” is the k-tuple s. Now, how many “legs” does a “sheep” have? Let us find out.
Let us fix a k-element subset W of S. Thus, |W | = k (since W is a k-element set).
The following observation is easy:

179 Indeed, α has an inverse, which can be constructed as follows: For each injective k-tuple
k , we let f : [ k ] → S be the map that sends 1, 2, . . . , k to s , s , . . . , s , re-
s = (s1 , s2 , . . . , sk ) ∈ Sinj s 1 2 k
spectively. This map f s is injective (since its values s1 , s2 , . . . , sk are distinct (because (s1 , s2 , . . . , sk )
is an injective k-tuple)), and thus belongs to {injective maps from [k ] to S}. Hence, we can de-
fine a map
k
β : Sinj → {injective maps from [k] to S} ,
s 7→ f s .

It is straightforward to see that the maps α and β are mutually inverse. Hence, the map α is
invertible, i.e., bijective, i.e., a bijection.
180 since ( s , s , . . . , s ) is an injective k-tuple
1 2 k
181 since ( s , s , . . . , s ) ∈ Sk ⊆ Sk
1 2 k inj
Enumerative Combinatorics: class notes page 251

k satisfying set s = W are

Observation 1: The injective k-tuples s ∈ Sinj
k .
precisely the injective k-tuples in Winj

[Proof of Observation 1: We must prove the following two claims:

k satisfying set s = W is an injective
Claim 1.1: Every injective k-tuple s ∈ Sinj
k .
k-tuple in Winj

k is an injective k-tuple s ∈ Sk satisfying

Claim 1.2: Every injective k-tuple in Winj inj
set s = W.

[Proof of Claim 1.1: Let s ∈ Sinj k be an injective k-tuple satisfying set s = W. We must prove

that s is an injective k-tuple in Winj k .

Write the k-tuple s as s = (s1 , s2 , . . . , sk ). Then, set s = {s1 , s2 , . . . , sk } (by the definition
of set s), so that {s1 , s2 , . . . , sk } = set s = W. Hence, the elements s1 , s2 , . . . , sk belong to W
(because they clearly belong to {s1 , s2 , . . . , sk }). Therefore, (s1 , s2 , . . . , sk ) ∈ W k . In other
words, s ∈ W k (since s = (s1 , s2 , . . . , sk )). Since the k-tuple s is injective, this shows that
s ∈ Winjk . Hence, s is an injective k-tuple in W k . This completes the proof of Claim 1.1.]
inj
[Proof of Claim 1.2: Let w be an injective k-tuple in Winj k . We must prove that w is an

injective k-tuple s ∈ Sinj k satisfying set s = W. In other words, we must prove that w is an

injective k-tuple that belongs to Sinj k and satisfies set w = W.

We have w ∈ Winj k ⊆ W k ⊆ Sk (since W ⊆ S). Thus, w is an injective k-tuple in Sk ; in

other words, w ∈ Sinj k . Now, write the k-tuple w in the form w = w , w , . . . , w . Then,
( 1 2 k)
k k
w1 , w2 , . . . , wk ∈ W (since (w1 , w2 , . . . , wk ) = w ∈ Winj ⊆ W ), so that {w1 , w2 , . . . , wk } is
a subset of W. Moreover, the k-tuple (w1 , w2 , . . . , wk ) = w is injective; in other words,
w1 , w2 , . . . , wk are distinct. Thus, {w1 , w2 , . . . , wk } is a k-element set. In other words,
|{w1 , w2 , . . . , wk }| = k. Hence, |{w1 , w2 , . . . , wk }| = k = |W | (since |W | = k). Hence, The-
orem 1.4.7 (c) (applied to A = W and B = {w1 , w2 , . . . , wk }) yields {w1 , w2 , . . . , wk } = W.
Now, from w = (w1 , w2 , . . . , wk ), we obtain set w = {w1 , w2 , . . . , wk } (by the definition of
set w), so that set w = {w1 , w2 , . . . , wk } = W. Thus, altogether, we have shown that w is an
injective k-tuple that belongs to Sinj k and satisfies set w = W. This proves Claim 1.2.]

Combining Claim 1.1 with Claim 1.2, we obtain precisely the claim of Observation 1.
Hence, Observation 1 is proven.]
Now, Observation 1 shows that
 
k
# of injective k-tuples s ∈ Sinj satisfying set s = W
 
k
= # of injective k-tuples in Winj
 
k k
= Winj since all elements of Winj are injective k-tuples

= |W | k (by Proposition 2.7.2, applied to W instead of S)

= kk (since |W | = k)
= k! (192)
Enumerative Combinatorics: class notes page 252

(by Proposition 2.4.3 (d), applied to k instead of n).

Now, forget that we fixed W. We thus have proved (192) for each k-element
subset W of S.
Now, recall that for each injective k-tuple s ∈ Sinjk , we have defined a k-element

subset set s of S. In other words, there is a map

n o
k
injective k-tuples s ∈ Sinj → {k-element subsets of S} ,
s 7→ set s.

Hence, the sum rule (Theorem 1.2.5) shows that

 
k
# of injective k-tuples s ∈ Sinj
 
= ∑ # of injective k-tuples s ∈ S k
inj satisfying set s = W
W is a k-element | {z }
subset of S =k!
(by (192))

= ∑ k! = (# of k-element subsets of S) · k!.

W is a k-element
subset of S

Comparing this equality with

 
k
# of injective k-tuples s ∈ Sinj
 
k k
= Sinj since all elements of Sinj are injective k-tuples

= |S|k (by Proposition 2.7.2)

= nk (since |S| = n)
 
n
= k! · (by Proposition 2.4.3 (c)) ,
k

we obtain  
n
k! · = (# of k-element subsets of S) · k!.
k
We can divide both sides of this equality by k! (since k! = 1 · 2 · · · · · k is clearly a
nonzero integer), and thus obtain
 
n
= (# of k-element subsets of S) .
k

Hence, Theorem 1.3.12 is proved again.

Class of 2019-10-30
Enumerative Combinatorics: class notes page 253

2.8. Another use of polynomials

We have already seen how binomial identities (i.e., identities that involve binomial
coefficients) can be proved using the “polynomial identity trick”. This is not the
only application of polynomials to proving binomial identities. We shall now see
another application (which is actually a foretaste of a comprehensive method – the
method of generating functions), in which polynomials are not merely evaluated at
numbers, but also compared coefficientwise. This application will be (yet) another
proof of the Chu–Vandermonde identity (specifically, (172)). As in our second
proof of Theorem 2.6.1, we will only prove it in the case when x, y ∈ N, but as
we already know, the polynomial identity trick allows us to extend such a proof
to the general case more or less automatically, so we don’t lose any generality by
restricting ourselves to such a case.
Before we come to this new proof of the Chu–Vandermonde identity, let us state
a simple lemma:

Lemma 2.8.1. Let p ∈ N. Then,

 
p
(1 + X ) = ∑
p
Xm
m ∈N
m

(an equality between polynomials in 1 variable X with real coefficients). Here,

the sum on the right hand side is infinite, but it is well-defined because only
finitely many of its addends are nonzero.

Proof of Lemma 2.8.1 (sketched). We have stated Theorem 1.3.24 only for the case
when x and y are real numbers; however, it also holds when x and y are poly-
nomials (and the same proof that we gave applies in this case). Thus, we can apply
Theorem 1.3.24 to n = p, x = X and y = 1. We thus obtain182
p   p    
p p p
( X + 1) = ∑
p
X 1|{z} = ∑
k p−k
X = ∑
k 0
Xk.
k =0
k k =0
k k ∈N
k
=1

0
  the “=” sign is owed to the fact that all integers k satisfying
Here,   k > p satisfy
p p
= 0 (by Proposition 1.3.6, applied to n = p) and thus X k = 0. This also
k   k
p
shows that the infinite sum ∑ X k has only finitely many nonzero addends,
k ∈N k
and thus is well-defined. Now,
   
p p
( X + 1) = ∑
p
X = ∑
k
Xm
k ∈N
k m ∈N
m

182 See 0
Remark 2.6.2 for the meaning of the “=” sign.
Enumerative Combinatorics: class notes page 254

(here, we have renamed the summation index k as m). In view of X + 1 = 1 + X,

we can rewrite this as  
p
(1 + X ) = ∑
p
Xm.
m ∈N
m
This proves Lemma 2.8.1.
Fourth proof of Theorem 2.6.1 for x, y ∈ N (sketched). Assume that x, y ∈ N. We shall
only prove the equality (172) (since we have already seen how to derive (173) from
it).
Rename x and y as a and b. Thus, a, b ∈ N, and we must prove the equality
  n   
a+b a b
= ∑ . (193)
n k =0
k n − k

Let us work with polynomials in one variable X, with real coefficients. Lemma
2.8.1 (applied to p = a) yields
   
a a
(1 + X ) = ∑
a
X = ∑
m
Xi
m ∈N
m i ∈N
i

(here, we have renamed the summation index m as i). Similarly,

 
b
(1 + X ) = ∑
b
X j.
j ∈N
j
Enumerative Combinatorics: class notes page 255

Multiplying these two equalities, we obtain

(1 + X ) a · (1 + X ) b
  !   !
a b
= ∑ Xi · ∑ Xj
i ∈N
i j ∈N
j
       
a b a b
= ∑ X ∑
i
X = ∑ ∑
j
Xi Xj
i ∈N
i j ∈N
j i ∈N j ∈N |
i j
| {z }   {z  }
   
a i b j a b i+ j
= X
= ∑ X X i j
j ∈N i j
     
a b a b
= ∑ ∑ X = ∑
i+ j
X i+ j
i ∈N j ∈N
i j (i,j)∈N2
i j
by the first equality sign in Theorem 1.6.11,
 
  
a b
since only finitely many (i, j) ∈ N satisfy 2 X i+ j 6= 0
 
i j
  
a b
= ∑ ∑ i j X i+ j
|{z}
m ∈N (i,j)∈N ;2
=X m
i + j=m (since i + j=m)

(here, we have used the analogue of (37) for infinite sums)

  
a b
= ∑ ∑ i j Xm
m ∈N (i,j)∈N ;
2
i + j=m
|

{z 
}
  

 a b 
= ∑ Xm
(i,j)∈N2 ; i j 


i + j=m
 
  
a b 
∑ ∑  Xm.

= 
m ∈N
 i j 
(i,j)∈N2 ;
i + j=m

Comparing this with

(1 + X ) a · (1 + X ) b
= (1 + X ) a + b
!
since the product rule for exponents, u a · ub = u a+b ,
holds for polynomials just as it holds for numbers
 
a+b
= ∑ Xm (by Lemma 2.8.1, applied to p = a + b) ,
m ∈N
m
Enumerative Combinatorics: class notes page 256

we obtain  
    
a b  a + b
∑ ∑ X = ∑
m
Xm.

 (194)
m ∈N
 i j  m ∈N
m
(i,j)∈N2 ;
i + j=m

We can thus define a polynomial P by

 
    
a b  a+b
P= ∑  ∑ X = ∑
m
Xm.

 
m ∈N
i j 
m ∈N
m
2 (i,j)∈N ;
i + j=m
 
  
 a b  X m , we obtain
Now, from P = ∑  ∑
m∈N (i,j)∈N2 ; i j 

i + j=m
  
a b
(the n-th coefficient of P) = ∑ i j
. (195)
(i,j)∈N2 ;
i + j=n
 
a+b
Meanwhile, from P = ∑ X m , we obtain
m ∈N m
 
a+b
(the n-th coefficient of P) = .
n
Comparing this with (195), we obtain
       
a+b a b a b
= ∑ = ∑
n i j k n−k
2 (i,j)∈N ;
k∈{0,1,...,n}
i + j=n

(here, we have substituted (k, n − k ) for (i, j) in the sum, since the map
n o
{0, 1, . . . , n} → (i, j) ∈ N | i + j = n ,
2

k 7→ (k, n − k )

is a bijection183 ). This is precisely the equality (193) (since the summation sign
n
∑ is a synonym for ∑ ). Thus, (193) holds. As we have said, this proves
k =0 k∈{0,1,...,n}
Theorem 2.6.1 in the case when x, y ∈ N.
183 In informal terms, this is just saying that each pair (i, j) ∈ N2 of nonnegative integers satisfying
i + j = n has the form (k, n − k) for a unique k ∈ {0, 1, . . . , n} (namely, for k = i), and conversely,
any pair of the latter form is a pair (i, j) ∈ N2 of nonnegative integers satisfying i + j = n.
Enumerative Combinatorics: class notes page 257

The proof we just gave used polynomials, but not via the polynomial identity
trick. Instead, we obtained an equality between two polynomials (viz., (194)), and
then compared coefficients (i.e., argued that two equal polynomials have equal
n-th coefficients). This relies on the fact that any polynomial has a well-defined
n-th coefficient for each n ∈ N, which does not depend on how the polynomial
is expressed. This is a versatile tactic, and we will later apply it to more general
objects than polynomials. But even when restricted to polynomials, it is rather
useful. Instead of proving Theorem 2.6.1 again and again, we can apply it to obtain
a new equality:

Exercise 2.8.1. (a) Let n ∈ N and m ∈ N. Prove that

  
m     m/2 n
n n  (−1) , if m is even;
∑ (−1)k k m − k =  m/2 .
k =0 0, if m is odd

(b) Prove that the claim of Exercise 2.8.1 (a) holds, more generally, for all n ∈ R
(rather than only for n ∈ N).
(c) Conclude that every n ∈ N satisfies
  
n  2  n/2 n
n  (−1) , if n is even;
∑ (−1)k k =  n/2 .
k =0 0, if n is odd

[Hint: For part (a), compare coefficients in the equality (1 − X )n · (1 + X )n =

n
1 − X 2 (expanded, once again, using the binomial formula). For part (b),
apply the polynomial identity trick.]

2.9. The Principle of Inclusion and Exclusion

We shall now come to an important tool for counting: the Principle of Inclusion and
Exclusion. We shall first present this principle in four forms (Theorem 2.9.1, The-
orem 2.9.6, Theorem 2.9.7 and Theorem 2.9.9); then we will prove it in Subsection
2.9.3 (after proving a simple but important identity in Subsection 2.9.2, which will
help in our proofs), and show some applications in the remaining subsections.

2.9.1. The principles

We begin with some simple facts about unions and intersections of finite sets.
If A1 and A2 are two disjoint finite sets, then | A1 ∪ A2 | = | A1 | + | A2 |. (This is just
the equality (4), with X and Y renamed as A1 and A2 .) Is there a similar formula
for | A1 ∪ A2 | when A1 and A2 are not disjoint? Yes, if | A1 ∩ A2 | is known. Namely,

| A1 ∪ A2 | = | A1 | + | A2 | − | A1 ∩ A2 | (196)
Enumerative Combinatorics: class notes page 258

for any two finite sets A1 and A2 . This is fairly easy to check. A similar but more
complicated formula holds for 3 sets: namely,
| A1 ∪ A2 ∪ A3 | = | A1 | + | A2 | + | A3 | − | A1 ∩ A2 | − | A1 ∩ A3 | − | A2 ∩ A3 |
+ | A1 ∩ A2 ∩ A3 | (197)
for any three finite sets A1 , A2 and A3 . A further formula of the same kind holds
for 4 sets: namely,
| A1 ∪ A2 ∪ A3 ∪ A4 |
= | A1 | + | A2 | + | A3 | + | A4 |
− | A1 ∩ A2 | − | A1 ∩ A3 | − | A1 ∩ A4 | − | A2 ∩ A3 | − | A2 ∩ A4 | − | A3 ∩ A4 |
+ | A1 ∩ A2 ∩ A3 | + | A1 ∩ A2 ∩ A4 | + | A1 ∩ A3 ∩ A4 | + | A2 ∩ A3 ∩ A4 |
− | A1 ∩ A2 ∩ A3 ∩ A4 | (198)
for any four finite sets A1 , A2 , A3 and A4 . These three formulas (along with the
trivial formula | A1 | = | A1 | for a single set A1 ) all are particular cases of one general
fact, which expresses the size of the union of n finite sets through the sizes of their
intersections (more precisely, of the intersection of each k of these sets):
Theorem 2.9.1 (Principle of Inclusion and Exclusion (union form)). Let n ∈ N.
Let A1 , A2 , . . . , An be n finite sets. Then,

| A1 ∪ A2 ∪ · · · ∪ A n |
n
= ∑ (−1)m−1 ∑ A i1 ∩ A i2 ∩ · · · ∩ A i m . (199)
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im

The nested sum on the right hand side of (199) is somewhat daunting, so let us
write it out: It is
| A1 | + | A2 | + · · · + | A n |
| {z }
the sizes of all Ai
(summed with + signs)
− | A 1 ∩ A 2 | − | A 1 ∩ A 3 | − · · · − | A n −1 ∩ A n |
| {z }
the sizes of all intersections Ai ∩ A j (with i < j)
(summed with − signs)
+ | A 1 ∩ A 2 ∩ A 3 | + | A 1 ∩ A 2 ∩ A 4 | + · · · + | A n −2 ∩ A n −1 ∩ A n |
| {z }
the sizes of all intersections Ai ∩ A j ∩ Ak (with i < j<k)
(summed with + signs)
±···
+ (−1)n−1 | A1 ∩ A2 ∩ · · · ∩ An | .
| {z }
the size of the intersection A1 ∩ A2 ∩···∩ An
(summed with a + or − sign depending on the parity of n)
Enumerative Combinatorics: class notes page 259

Thus, it is a sum that contains the sizes of all possible intersections of some of the
sets A1 , A2 , . . . , An , each with a + or − sign depending on how many sets are being
intersected. Clearly, the right hand sides of (196), (197) and (198) are the particular
cases of this sum obtained for n = 2, n = 3 and n = 4, respectively.
Theorem 2.9.1 is known as the Principle of Inclusion and Exclusion (often abbre-
viated PIE) or the Sylvester sieve formula. We will not prove it right away, since
this will become easier once we have restated it in a more abstract form. This ab-
stract form (which also makes it more convenient to apply) relies on the following
notation from set theory:
Definition 2.9.2. Let I be a nonempty set. For each i ∈ I, let Ai be a set. Then,
T
Ai denotes the set
i∈ I
{ x | x ∈ Ai for each i ∈ I } .
Ai is called the intersection of all Ai with i ∈ I.
T
This set
i∈ I

It is easy to see that if I = {i1 , i2 , . . . , im } is a finite set, and if Ai is a set for each
i ∈ I, then \
A i = A i1 ∩ A i2 ∩ · · · ∩ A i m . (200)
i∈ I
Example 2.9.3. Let I = [5]. For each i ∈ I, let Ai = [i − 8, i ] =
{i − 8, i − 7, . . . , i − 1, i } (where we are using the notation from Definition 1.4.17).
Then,
\
Ai = A1 ∩ A2 ∩ A3 ∩ A4 ∩ A5
|{z} |{z} |{z} |{z} |{z}
i∈ I =[2−8,2] =[3−8,3] =[5−8,5]
=[1−8,1] =[4−8,4]
=[−7,1] =[−6,2] =[−5,3] =[−4,4] =[−3,5]
= [−7, 1] ∩ [−6, 2] ∩ [−5, 3] ∩ [−4, 4] ∩ [−3, 5]
= [−3, 1] = {−3, −2, −1, 0, 1} .

T
Note the similarity between the notation Ai (for the intersection of a family of
i∈ I
sets) and the notation ∑ ai (for the sum of a finite family of numbers). However, in
i∈ I
Ai , the set I must be nonempty (unlike in ∑ ai ) but can be infinite
T
the notation
i∈ I i∈ I
(unlike in ∑ ai , which is usually undefined when I is infinite).
i∈ I
We can now rewrite the right hand side of (199) as follows:
Proposition 2.9.4. Let n ∈ N. Let A1 , A2 , . . . , An be n finite sets. Then,
n
∑ (−1)m−1 ∑ ∑ (−1)| I |−1
\
A i1 ∩ A i2 ∩ · · · ∩ A i m = Ai .
m =1 m
(i1 ,i2 ,...,im )∈[n] ; I ⊆[n]; i∈ I
i1 <i2 <···<im I 6 =∅
Enumerative Combinatorics: class notes page 260

Example 2.9.5. Set n = 3 in Proposition 2.9.4. Then, the claim of Proposition

2.9.4 becomes

(−1)1−1 (| A1 | + | A2 | + | A3 |)
+ (−1)2−1 (| A1 ∩ A2 | + | A1 ∩ A3 | + | A2 ∩ A3 |)
+ (−1)3−1 | A1 ∩ A2 ∩ A3 |

= (−1)|{1}|−1 Ai + (−1)|{2}|−1 Ai + (−1)|{3}|−1

\ \ \
Ai
i ∈{1} i ∈{2} i ∈{3}

+ (−1)|{1,2}|−1 Ai + (−1)|{1,3}|−1
\ \
Ai
i ∈{1,2} i ∈{1,3}

+ (−1)|{2,3}|−1 Ai + (−1)|{1,2,3}|−1
\ \
Ai
i ∈{2,3} i ∈{1,2,3}

(since the subsets I of [n] that satisfy I 6= ∅ are

{1} , {2} , {3} , {1, 2} , {1, 3} , {2, 3} , {1, 2, 3}). You can easily confirm this
by checking that the addends on the left hand side (after the parentheses are
expanded) are precisely the addends on the right hand side.

Proof of Proposition 2.9.4. We first claim the following:

Claim 1: Let m ∈ {1, 2, . . . , n}. Then,

∑ ∑
\
A i1 ∩ A i2 ∩ · · · ∩ A i m = Ai .
(i1 ,i2 ,...,im )∈[n]m ; I ⊆[n]; i ∈ I
i1 <i2 <···<im I 6=∅;
| I |=m

[Proof of Claim 1: We have m > 0 (since m ∈ {1, 2, . . . , n}). Now, if I is a subset

of [n] that satisfies | I | = m, then it satisfies I 6= ∅ (since | I | = m > 0). Hence, the
condition “I 6= ∅” under the summation sign ∑ is redundant (i.e., the sum does
I ⊆[n];
I 6=∅;
| I |=m
not change if we remove it). Thus,

∑ ∑
\ \
Ai = Ai
I ⊆[n]; i ∈ I I ⊆[n]; i ∈ I
I 6=∅; | I |=m
| I |=m

∑
\
= Ai . (201)
I ∈{m-element subsets of [n]} i ∈ I
Enumerative Combinatorics: class notes page 261

For any m-tuple (i1 , i2 , . . . , im ) ∈ [n]m satisfying i1 < i2 < · · · < im , the set
{i1 , i2 , . . . , im } is an m-element subset of [n] 184 . Thus, the map

(i1 , i2 , . . . , im ) ∈ [n]m | i1 < i2 < · · · < im → {m-element subsets of [n]} ,



( i1 , i2 , . . . , i m ) 7 → { i1 , i2 , . . . , i m }

is well-defined. Moreover, this map is bijective185 . Hence, we can substitute

{i1 , i2 , . . . , im } for I in the sum ∑
T
Ai . We thus obtain
I ∈{m-element subsets of [n]} i ∈ I

∑ ∑
\ \
Ai = Ai
m
I ∈{m-element subsets of [n]} i ∈ I (i1 ,i2 ,...,im )∈[n] ; i ∈{i1 ,i2 ,...,im }
i1 <i2 <···<im | {z }
= Ai1 ∩ Ai2 ∩···∩ Aim
(by (200))

= ∑ m
A i1 ∩ A i2 ∩ · · · ∩ A i m .
(i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im

Hence, (201) becomes

∑ ∑ ∑
\ \
Ai = Ai = A i1 ∩ A i2 ∩ · · · ∩ A i m .
I ⊆[n]; i ∈ I I ∈{m-element subsets of [n]} i ∈ I (i1 ,i2 ,...,im )∈[n]m ;
I 6=∅; i1 <i2 <···<im
| I |=m

This proves Claim 1.]

Now, for any subset I of [n] satisfying I 6= ∅, we have | I | ∈ {1, 2, . . . , n} 186 .

184 Proof. Let (i1 , i2 , . . . , im ) ∈ [n]m be an m-tuple satisfying i1 < i2 < · · · < im . Thus, i1 , i2 , . . . , im ∈
[n]; hence, {i1 , i2 , . . . , im } is a subset of [n]. Moreover, the numbers i1 , i2 , . . . , im are distinct (since
i1 < i2 < · · · < im ), and thus the set {i1 , i2 , . . . , im } has m elements. Hence, {i1 , i2 , . . . , im } is an
m-element set. Therefore, {i1 , i2 , . . . , im } is an m-element subset of [n].
185 Indeed, this can easily be derived from Proposition 1.4.13.
186 Proof. Let I be a subset of [ n ] satisfying I 6 = ∅. Then, Theorem 1.4.7 (b) (applied to A = [ n ] and

B = I) yields | I | ≤ |[n]| = n. Also, | I | > 0 (since I 6= ∅) and thus | I | ≥ 1. Combining this with
| I | ≤ n, we obtain | I | ∈ {1, 2, . . . , n}, qed.
Enumerative Combinatorics: class notes page 262

Hence, an application of (37) yields

∑ (−1)| I |−1 ∑ ∑ (−1)| I |−1

\ \
Ai = Ai
I ⊆[n]; i∈ I m∈{1,2,...,n} I ⊆[n]; i∈ I
| {z }
I 6 =∅ I 6=∅; =(−1)m−1
| I |=m (since | I |=m)

∑ ∑ (−1)m−1
\
= Ai
m∈{1,2,...,n} I ⊆[n]; i∈ I
I 6=∅;
| I |=m
| {z }
=(−1)m−1 ∑
T
Ai
I ⊆[n]; i∈ I
I 6=∅;
| I |=m

∑ (−1)m−1 ∑
\
= Ai
m∈{1,2,...,n} I ⊆[n]; i ∈ I
I 6=∅;
| I |=m
| {z }
= ∑ | Ai1 ∩ Ai2 ∩···∩ Aim |
(i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im
(by Claim 1)

= ∑ (−1)m−1 ∑ A i1 ∩ A i2 ∩ · · · ∩ A i m
m∈{1,2,...,n} (i1 ,i2 ,...,im )∈[n]m ;
| {z } i1 <i2 <···<im
n
= ∑
m =1
n
= ∑ (−1)m−1 ∑ A i1 ∩ A i2 ∩ · · · ∩ A i m .
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im

This proves Proposition 2.9.4.

We can now restate Theorem 2.9.6 as follows:

Theorem 2.9.6 (Principle of Inclusion and Exclusion (union form)). Let n ∈ N.

Let A1 , A2 , . . . , An be n finite sets. Then,

∑ (−1)| I |−1
\
| A1 ∪ A2 ∪ · · · ∪ A n | = Ai . (202)
I ⊆[n]; i∈ I
I 6 =∅

We will soon prove Theorem 2.9.6 (and thus conclude Theorem 2.9.1 from it).
First, let us state yet another equivalent version of this principle:
Enumerative Combinatorics: class notes page 263

Theorem 2.9.7 (Principle of Inclusion and Exclusion (complement form)). Let

n ∈ N. Let U be a finite set. Let A1 , A2 , . . . , An be n subsets of U. Then,

∑ (−1)| I |
\
|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = Ai . (203)
I ⊆[n] i∈ I
T
Here, the “empty” intersection Ai is understood to mean the set U.
i ∈∅

Example 2.9.8. Let U be a finite set, and let A1 and A2 be two subsets of U. Then,
Theorem 2.9.7 (applied to n = 2) says that

∑ (−1)| I |
\
|U \ ( A1 ∪ A2 )| = Ai
I ⊆[2] i∈ I

= (−1)|∅| + (−1)|{1}|
\ \
Ai Ai
i ∈∅
| {z } | {z }
=1 =−1 i ∈{1}
| {z } | {z }
=U = A1
(by definition)

+ (−1)|{2}| Ai + (−1)|{1,2}|
\ \
Ai
| {z } | {z }
=−1 i ∈{2} =1 i ∈{1,2}
| {z } | {z }
= A2 = A1 ∩ A2
(since the subsets of [2] are ∅, {1} , {2} , {1, 2})
= |U | + (−1) | A1 | + (−1) | A2 | + | A1 ∩ A2 |
= |U | − | A 1 | − | A 2 | + | A 1 ∩ A 2 | .

T
A remark is in order here. In Definition 2.9.2, we have defined Ai for any
i∈ I
nonempty set I. This definition cannot be extended to the case when I = ∅,
T
because it would entail that Ai should be the set of all objects whatsoever, but
i ∈∅
there is no such set187 . However, in order for the right hand side of (203) to make
Ai for all I ⊆ [n], including I = ∅. Thus, we
T
sense, we need a well-defined set
T i∈ I T
need to define Ai somehow. It turns out that defining Ai to be U is the right
i ∈∅ i ∈∅
thing to do (in this particular situation). It is justified here, because all of the sets
involved are subsets of U, so “all objects” can be interpreted as “all elements of U”
187 See [LeLeMe16, §8.3] for an explanation.
Enumerative Combinatorics: class notes page 264

(and then the set of all objects really becomes U). This is, by the way, the reason
why we chose the letter U in Theorem 2.9.7; it stands for “universe”.
Finally, let us restate Theorem 2.9.7 in an equivalent form that is slightly easier to
T
actually use, because it shuns the notation Ai (and thus avoids having to define
i∈ I
it separately for I = ∅):

Theorem 2.9.9 (Principle of Inclusion and Exclusion (complement form, simpli-

fied)). Let n ∈ N. Let U be a finite set. Let A1 , A2 , . . . , An be n subsets of U.
Then,

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| .

I ⊆[n]

Note that the addend corresponding to I = ∅ on the right hand side of this
equality is simply |U | 188 .

2.9.2. The cancellation lemma

We have now stated the Principle of Inclusion and Exclusion in four forms (Theo-
rem 2.9.1, Theorem 2.9.6, Theorem 2.9.7 and Theorem 2.9.9). Soon we shall prove
them. First, we state a crucial theorem that will serve as an auxiliary tool in the
proofs:

Proposition 2.9.10. Let S be a finite set. Then,

∑ (−1)| I | = [S = ∅] .
I ⊆S

Once again, we are using the Iverson bracket notation here (see Definition 1.3.15).

Example 2.9.11. The subsets of {1, 2} are ∅, {1}, {2} and {1, 2}. Thus, applying
Proposition 2.9.10 to S = {1, 2}, we find

(−1)|∅| + (−1)|{1}| + (−1)|{2}| + (−1)|{1,2}| = [{1, 2} = ∅] .

| {z } | {z } | {z } | {z }
=1 =−1 =−1 =1
188 Proof. This addend is

(−1)|∅| {s ∈ U | s ∈ Ai for all i ∈ ∅} = { s ∈ U } = |U | .

| {z } | {z } | {z }
=1 ={s∈U } =U
(because each s∈U satisfies (s∈ Ai for all i ∈∅)
(indeed, this is vacuously true, since there exists no i ∈∅))
Enumerative Combinatorics: class notes page 265

Indeed, both sides of this equality are 0 (the left hand side because the addends
cancel; the right hand side because {1, 2} 6= ∅).

First proof of Proposition 2.9.10. This will be similar to the proof of Theorem 1.4.1
that we gave in Subsection 1.4.2.
We have S = ∅ if and only if |S| = 0. In other words, “S = ∅” and “|S| = 0”
are two equivalent logical statements. Hence, Exercise 1.3.3 (a) (applied to A =
(“S = ∅”) and B = (“ |S| = 0”)) yields [S = ∅] = [|S| = 0].
Let n = |S|. Hence, S is an n-element set. If I is any subset of S, then Theorem
1.4.7 (b) (applied to A = S and B = I) yields | I | ≤ |S| = n and therefore | I | ∈
{0, 1, . . . , n}. Hence, an application of (37) yields

∑ (−1)| I |
I ⊆S

= ∑ ∑ (−1)| I | = ∑ ∑ (−1)k
k∈{0,1,...,n} I ⊆S; k∈{0,1,...,n} I ⊆S;
| {z }
| I |=k =(−1)k | I |=k
(since | I |=k) | {z }
=(# of subsets I of S satisfying | I |=k)·(−1)k

= ∑ (# of subsets I of S satisfying | I | = k) · (−1)k

k∈{0,1,...,n}
| {z }
| {z } =( # of k-element
  subsets of S )
n n
=∑ =
k =0 k
(by Theorem 1.3.12)
n   n  
n k n
= ∑ (−1) = ∑ (−1)
k
= [ n = 0] (by Proposition 1.3.28)
k =0
k k =0
k
= [|S| = 0] (since n = |S|)
= [ S = ∅] (since [S = ∅] = [|S| = 0]) .

This proves Proposition 2.9.10.

Second proof of Proposition 2.9.10. If S = ∅, then Proposition 2.9.10 holds189 . Hence,
for the rest of this proof, we WLOG assume that we don’t have S = ∅. Thus, S 6= ∅.
189 Proof.
Assume that S = ∅. Thus, there exists exactly one subset of S, namely the empty set ∅.
Hence, the sum ∑ (−1)| I | has exactly one addend, namely the addend for I = ∅. Thus, this
I ⊆S
sum simplifies as follows:

∑ (−1)| I | = (−1)|∅| = 1 (since |∅| = 0 is even) .

I ⊆S

Comparing this with

[S = ∅] = 1 (since S = ∅) ,
we obtain ∑ (−1)| I | = [S = ∅]. Hence, Proposition 2.9.10 is proven under the assumption that
I ⊆S
S = ∅.
Enumerative Combinatorics: class notes page 266

Hence, there exists some g ∈ S. Consider this g. (We can have many choices for g,
but we just pick one.)
Each subset I of S must satisfy either g ∈ I or g ∈
/ I (but not both at the same
|I|
time). Hence, we can split the sum ∑ (−1) as follows:
I ⊆S

∑ (−1)| I | = ∑ (−1)| I | + ∑ (−1)| I | . (204)

I ⊆S I ⊆S; I ⊆S;
g∈ I ∈I
g/

We shall now “set the two sums on the right hand side up to fight each other”.
Each subset J of S satisfies J ∪ { g} ⊆ S (because g ∈ S) and g ∈ J ∪ { g} (obvi-
ously). Thus, the map190

{I ⊆ S | g ∈
/ I} → {I ⊆ S | g ∈ I} ,
J 7→ J ∪ { g}

is well-defined. The map

{I ⊆ S | g ∈ I} → {I ⊆ S | g ∈
/ I} ,
K 7→ K \ { g}

is also well-defined. These two maps are mutually inverse191 , and thus are bijec-
tions. Hence, in particular, the map

{I ⊆ S | g ∈
/ I} → {I ⊆ S | g ∈ I} ,
J 7→ J ∪ { g}

is a bijection. Thus, we can substitute J ∪ { g} for I in the sum ∑ (−1)| I | . We thus

I ⊆S;
g∈ I
obtain

∑ (−1)| I | = ∑ (−1)| J ∪{ g}| = ∑ (−1)| J |+1 = − ∑ (−1)| J |

I ⊆S; J ⊆S; J ⊆S; J ⊆S;
| {z } | {z }
g∈ I g/∈J =(−1)| J |+1 ∈ J =−(−1)| J |
g/ ∈J
g/
(since | J ∪{ g}|=| J |+1
(because g/ ∈ J))

=− ∑ (−1)| I | (205)
I ⊆S;
∈I
g/

190 The notation “{ I ⊆ S | g ∈/ I }” means “the set of all subsets I of S satisfying g ∈/ I”. Similarly,
the notation “{ I ⊆ S | g ∈ I }” means “the set of all subsets I of S satisfying g ∈ I”.
191 because of the following two (easily proven) facts:

• Every subset J of S satisfying g ∈

/ J must satisfy ( J ∪ { g}) \ { g} = J.
• Every subset K of S satisfying g ∈ K must satisfy (K \ { g}) ∪ { g} = K.
Enumerative Combinatorics: class notes page 267

(here, we have renamed the summation index J as I).

Now, (204) becomes

∑ (−1)| I | = ∑ (−1)| I | + ∑ (−1)| I | = − ∑ (−1)| I | + ∑ (−1)| I | = 0.

I ⊆S I ⊆S; I ⊆S; I ⊆S; I ⊆S;
g∈ I ∈I
g/ ∈I
g/ ∈I
g/
| {z }
=− ∑ (−1)| I |
I ⊆S;
∈I
g/
(by (205))

Comparing this with

[ S = ∅] = 0 (since S 6= ∅) ,

we obtain ∑ (−1)| I | = [S = ∅]. This proves Proposition 2.9.10.

I ⊆S

We note that there is a generalization of Proposition 2.9.10, which we will not

use anytime soon but which is nevertheless rather useful:

Proposition 2.9.12. Let S be a finite set. Let T be a subset of S. Then,

∑ (−1)| I | = (−1)|T | [S = T ] .
I ⊆S;
T⊆ I

Exercise 2.9.1. Prove Proposition 2.9.12.

Clearly, Proposition 2.9.10 is the particular case of Proposition 2.9.12 when T = ∅

(since every subset I of S satisfies ∅ ⊆ I).

2.9.3. The proofs

Class of 2019-11-01

We are now ready to prove the Principle of Inclusion and Exclusion in all its
forms. We begin with Theorem 2.9.9:
Proof of Theorem 2.9.9. We shall use the Iverson bracket notation (see Definition
1.3.15).
Here is the plan: We will approach the sum

∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| (206)

I ⊆[n]

by looking at each particular s ∈ U and computing how often s is “counted” in

this sum. In other words, instead of computing (206) directly, we shall compute the
Enumerative Combinatorics: class notes page 268

sum ∑ (−1)| I | [s ∈ Ai for all i ∈ I ] for a single s ∈ U first. Then, we will sum it
I ⊆[n]
over all s ∈ U, and obtain (206), because of Proposition 1.6.3 (b).
Here are the details: Fix s ∈ U. Define a subset S of [n] by

S = {i ∈ [ n ] | s ∈ Ai } .

Thus, S is the set of all i ∈ [n] such that s ∈ Ai . Hence, S = ∅ holds if and only if
there exists no i ∈ [n] such that s ∈ Ai . Thus, we have the following chain of logical
equivalences:

(S = ∅) ⇐⇒ (there exists no i ∈ [n] such that s ∈ Ai )

⇐⇒ (we have s ∈/ Ai for all i ∈ [n])
⇐⇒ (s ∈
/ A1 ∪ A2 ∪ · · · ∪ A n )
⇐⇒ (s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )) (since s ∈ U ) .

Thus, “S = ∅” and “s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )” are two equivalent logical state-

ments. Hence, Exercise 1.3.3 (a) (applied to A = (“S = ∅”) and
B = (“s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An ) ”)) yields

[S = ∅] = [s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )] . (207)

On the other hand, let I be any subset of [n]. Then, we have the following chain
of logical equivalences:

(s ∈ Ai for each i ∈ I )
⇐⇒ (i ∈ S for each i ∈ I )
 
because for any given i ∈ I, the statement “s ∈ Ai ”
is equivalent to “i ∈ S” (by the definition of S)
⇐⇒ ( I ⊆ S) .

Thus, “s ∈ Ai for each i ∈ I” and “I ⊆ S” are two equivalent logical statements.

Consequently, Exercise 1.3.3 (a) (applied to A = (“s ∈ Ai for each i ∈ I”) and B =
(“I ⊆ S”)) yields
[s ∈ Ai for each i ∈ I ] = [ I ⊆ S] . (208)
Forget that we fixed I. We thus have proved (208) for each subset I of [n].
Enumerative Combinatorics: class notes page 269

Now,

∑ (−1)| I | [s ∈ Ai for each i ∈ I ]

I ⊆[n]
| {z }
=[ I ⊆S]
(by (208))

= ∑ (−1)| I | [ I ⊆ S]
I ⊆[n]

= ∑ (−1)| I | [ I ⊆ S] + ∑ (−1)| I | [ I ⊆ S]
I ⊆[n]; I ⊆[n];
| {z } | {z }
I ⊆S =1 not I ⊆S
=0
| {z } (since I ⊆S) (since we don’t have I ⊆S)
=∑
I ⊆S
(since S⊆[n],
and thus any subset I of S
is a subset of [n])
(since each subset I of [n] satisfies either I ⊆ S or not I ⊆ S)
= ∑ (−1)| I | + ∑ (−1)| I | 0 = ∑ (−1)| I | = [S = ∅] (by Proposition 2.9.10)
I ⊆S I ⊆[n]; I ⊆S
not I ⊆S
| {z }
=0
= [s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )] (209)
(by (207)).
Forget that we fixed s. We thus have proved (209) for each s ∈ U. Now, if I is
any subset of [n], then
|{s ∈ U | s ∈ Ai for all i ∈ I }|
= (# of s ∈ U that satisfy (s ∈ Ai for all i ∈ I ))
= ∑ [s ∈ Ai for all i ∈ I ] (210)
s ∈U

(by Proposition 1.6.3 (b), applied to S = U and A (s) = (“s ∈ Ai for all i ∈ I”)).
Hence,

∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }|

I ⊆[n]
| {z }
= ∑ [s∈ Ai for all i ∈ I ]
s ∈U
(by (210))

= ∑ (−1)| I | ∑ [s ∈ Ai for all i ∈ I ] = ∑ ∑ (−1)| I | [s ∈ Ai for all i ∈ I ]

I ⊆[n] s ∈U I ⊆[n] s∈U
| {z }
=∑ ∑
s∈U I ⊆[n]

= ∑ ∑ (−1)| I | [s ∈ Ai for all i ∈ I ] = ∑ [s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )] .

s∈U I ⊆[n] s ∈U
| {z }
=[s∈U \( A1 ∪ A2 ∪···∪ An )]
(by (209))
Enumerative Combinatorics: class notes page 270

On the other hand, Proposition 1.6.3 (a) (applied to U and U \ ( A1 ∪ A2 ∪ · · · ∪ An )

instead of S and T) yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = ∑ [s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ An )] .
s ∈U

Comparing these two equalities, we obtain

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| .

I ⊆[n]

This proves Theorem 2.9.9.

Let us next derive Theorem 2.9.7 from Theorem 2.9.9:
Proof of Theorem 2.9.7. In order to derive this from Theorem 2.9.9, we need to show
the following fact:

Statement 1: Let I be a subset of [n]. Then,

\
Ai = {s ∈ U | s ∈ Ai for all i ∈ I } .
i∈ I

[Proof of Statement 1: We are in one of the following two cases:

Case 1: We have I = ∅.
Case 2: We have I 6= ∅.
T T
Let us first consider Case 1. In this case, we have I = ∅. Hence, Ai = Ai =
T i∈ I i ∈∅
U (since we defined Ai to be U in Theorem 2.9.7). Comparing this with
i ∈∅

{s ∈ U | s ∈ Ai for all i ∈ I }
= {s ∈ U | s ∈ Ai for all i ∈ ∅} (since I = ∅)
 
since each s ∈ U satisfies (s ∈ Ai for all i ∈ ∅)
= {s ∈ U }
(indeed, this is vacuously true, since there exists no i ∈ ∅)
= U,
Ai = {s ∈ U | s ∈ Ai for all i ∈ I }. Thus, Statement 1 is proved in
T
we obtain
i∈ I
Case 1.
Let us next consider Case 2. In this case, we have I 6= ∅. Thus, the set I is
T
nonempty. Hence, the definition of Ai yields
i∈ I
\
Ai = { x | x ∈ Ai for each i ∈ I } .
i∈ I

But I is nonempty; thus, there exists some j ∈ I. Consider this j. From j ∈ I, we

Ai ⊆ A j (since the intersection of a family of sets is clearly contained in
T
obtain
i∈ I
Enumerative Combinatorics: class notes page 271

each of these sets). But A j is a subset of U (since A1 , A2 , . . . , An are subsets of U).

 
Hence, A j ⊆ U, so that Ai ⊆ A j ⊆ U and thus U ∩
T T T
Ai = Ai . Hence,
i∈ I i∈ I i∈ I
!
\ \
Ai = U ∩ Ai
i∈ I i∈ I
| {z }
={ x | x ∈ Ai for each i ∈ I }
= U ∩ { x | x ∈ Ai for each i ∈ I } = { x ∈ U | x ∈ Ai for each i ∈ I }
= { x ∈ U | x ∈ Ai for all i ∈ I } = {s ∈ U | s ∈ Ai for all i ∈ I }

(here, we have renamed the index x as s). Thus, Statement 1 is proved in Case 2.
We have now proved Statement 1 in both Cases 1 and 2. Hence, Statement 1 is
always proved.]
Now, Theorem 2.9.9 yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| .

I ⊆[n]

Comparing this with

∑ (−1)| I | ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| ,

\
Ai =
I ⊆[n] i∈ I I ⊆[n]
| {z }
={s∈U | s∈ Ai for all i ∈ I }
(by Statement 1)

we obtain

∑ (−1)| I |
\
|U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = Ai .
I ⊆[n] i∈ I

This proves Theorem 2.9.7.

We can now derive Theorem 2.9.6 from this:
Proof of Theorem 2.9.6. Let U be the set A1 ∪ A2 ∪ · · · ∪ An . Then, A1 , A2 , . . . , An are
subsets of U. The set U = A1 ∪ A2 ∪ · · · ∪ An is finite (since A1 , A2 , . . . , An are
finite). From U = A1 ∪ A2 ∪ · · · ∪ An , we obtain U \ ( A1 ∪ A2 ∪ · · · ∪ An ) = ∅ and
thus |U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = |∅| = 0.
T
Let us agree to understand the “empty” intersection Ai to mean the set U.
i ∈∅
Enumerative Combinatorics: class notes page 272

Then, Theorem 2.9.7 yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )|

∑ (−1)| I |
\
= Ai
I ⊆[n] i∈ I

= (−1)|∅| ∑ (−1)| I |
\ \
Ai + Ai
i ∈∅ I ⊆[n]; i∈ I
| {z } | {z }
=1 | {z } I 6=∅ =−(−1)
| I |−1
(since |∅|=0
=U
is even)
(here, we have split off the addend for I = ∅ from the sum)
  \
= U
|{z} + ∑ − (−1)| I |−1 Ai
= A1 ∪ A2 ∪···∪ An I ⊆[n]; i∈ I
I 6 =∅
| {z }
| I |−1 T
=− ∑ (−1) Ai
I ⊆[n]; i∈ I
I 6 =∅

∑ (−1)| I |−1
\
= | A1 ∪ A2 ∪ · · · ∪ A n | − Ai .
I ⊆[n]; i∈ I
I 6 =∅

Comparing this with |U \ ( A1 ∪ A2 ∪ · · · ∪ An )| = 0, we obtain

∑ (−1)| I |−1
\
| A1 ∪ A2 ∪ · · · ∪ A n | − Ai = 0.
I ⊆[n]; i∈ I
I 6 =∅

In other words,

∑ (−1)| I |−1
\
| A1 ∪ A2 ∪ · · · ∪ A n | = Ai .
I ⊆[n]; i∈ I
I 6 =∅

This proves Theorem 2.9.6.

Proof of Theorem 2.9.1. Theorem 2.9.6 yields

∑ (−1)| I |−1
\
| A1 ∪ A2 ∪ · · · ∪ A n | = Ai
I ⊆[n]; i∈ I
I 6 =∅
n
= ∑ (−1)m−1 ∑ A i1 ∩ A i2 ∩ · · · ∩ A i m
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im

(by Proposition 2.9.4). This proves Theorem 2.9.1.

Enumerative Combinatorics: class notes page 273

We thus have proved all three forms of the Principle of Inclusion and Exclusion.
We notice that Theorem 2.9.6 also appears in [Smid09] (with a proof by induction),
in [Galvin17, Theorem 16.1] (with two proofs) and in [Grinbe15, Theorem 3.42] (in a
slightly more general form). Likewise, Theorem 2.9.7 also appears in [White10], in
[Galvin17, (12)] and in [Grinbe15, Theorem 3.43] (again, in a slightly more general
form). Several generalizations of the Principle can be found in [Grinbe15, Theorems
3.44, 3.45 and 3.46].
The following exercise provides an analogue of Theorem 2.9.1 with the roles of
∪ and ∩ swapped:

Exercise 2.9.2. Let n be a positive integer. Let A1 , A2 , . . . , An be n finite sets.

Prove that
n
| A1 ∩ A2 ∩ · · · ∩ A n | = ∑ (−1)m−1 ∑ m
A i1 ∪ A i2 ∪ · · · ∪ A i m .
m =1 (i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im

We shall now explore some applications of the Principle of Inclusion and Exclu-
sion.

2.9.4. Application: Surjections

Let us recall the numbers sur (m, n) we introduced in Definition 2.4.9. We have
proved two recursive formulas for them in Subsection 2.4.5; but we left a more
explicit formula (Theorem 2.4.17) without an appropriately elucidating proof192 . It
is now time to give such a proof using Theorem 2.9.9:
Proof of Theorem 2.4.17. Recall that sur (m, n) is the # of surjective maps from [m] to
[n] (by Definition 2.4.9). In order to compute this using Theorem 2.9.9, we want to
find a finite set (“universe”) U and n subsets A1 , A2 , . . . , An of U such that

U \ ( A1 ∪ A2 ∪ · · · ∪ An ) = {surjective maps from [m] to [n]} . (211)

We find them as follows: We define

U = [n][m] = {maps from [m] to [n]} . (212)

This is a finite set. Thus, the elements f ∈ U are the maps f : [m] → [n]. For each
i ∈ {1, 2, . . . , n}, we define a subset Ai of U by

Ai = {maps f : [m] → [n] | i ∈

/ f ([m])} . (213)

(This is the set of all maps f : [m] → [n] which never take i as a value.) Then,
A1 , A2 , . . . , An are n subsets of U. Let us check that they satisfy (211).
192 unless you count the inductive proof in Exercise 2.4.4 as a such
Enumerative Combinatorics: class notes page 274

Indeed, we have

{surjective maps from [m] to [n]}

= {maps f : [m] → [n] | f is surjective}
= {maps f : [m] → [n] | every i ∈ [n] is taken as a value by f } (214)

(because a map f : [m] → [n] is surjective if and only if every i ∈ [n] is taken as a
value by f ). But from (213), we obtain

A1 ∪ A2 ∪ · · · ∪ An = {maps f : [m] → [n] | at least one i ∈ [n] satisfies i ∈

/ f ([m])}

and thus

U \ ( A1 ∪ A2 ∪ · · · ∪ A n )
= U
|{z}
={maps from [m] to [n]}
={maps f :[m]→[n]}
\ {maps f : [m] → [n] | at least one i ∈ [n] satisfies i ∈
/ f ([m])}
= {maps f : [m] → [n]}
\ {maps f : [m] → [n] | at least one i ∈ [n] satisfies i ∈
/ f ([m])}
= {maps f : [m] → [n] | every i ∈ [n] satisfies i ∈ f ([m])}
 
since the negation of the statement
 “at least one i ∈ [n] satisfies i ∈
/ f ([m]) ” 
is “every i ∈ [n] satisfies i ∈ f ([m]) ”
= {maps f : [m] → [n] | every i ∈ [n] is taken as a value by f }
 
since the statement “every i ∈ [n] satisfies i ∈ f ([m]) ”
means the same as “every i ∈ [n] is taken as a value by f ”
= {surjective maps from [m] to [n]} (215)

(by (214)). Thus, (211) is indeed satisfied.

Now, (215) yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )|
= |{surjective maps from [m] to [n]}|
= (# of surjective maps from [m] to [n]) = sur (m, n) (216)

(since sur (m, n) was defined as the # of surjective maps from [m] to [n]).
In order to apply Theorem 2.9.9 successfully, we want to understand the sets
{s ∈ U | s ∈ Ai for all i ∈ I } (or at least their sizes). But this is easy: If I is a
subset of [n], then

{s ∈ U | s ∈ Ai for all i ∈ I }
= {maps f : [m] → [n] | f ([m]) ⊆ [n] \ I } . (217)
Enumerative Combinatorics: class notes page 275

[Proof of (217): Let I be a subset of [n]. We have

{s ∈ U | s ∈ Ai for all i ∈ I }
= {s ∈ U | s ∈ Ai for each i ∈ I }
= { f ∈ U | f ∈ Ai for each i ∈ I } (here, we have renamed the index s as f )
= {maps f : [m] → [n] | f ∈ Ai for each i ∈ I } (218)
(since the elements f ∈ U are the maps f : [m] → [n]) .

But if f : [m] → [n] is any map, then we have the following chain of logical equivalences:

( f ∈ Ai for each i ∈ I )
⇐⇒ (i ∈ / f ([m]) for each i ∈ I )
(because an i ∈ I satisfies f ∈ Ai if and only if i ∈
/ f ([m]) (by (213)))
⇐⇒ (none of the elements of I is contained in f ([m]))
⇐⇒ (the sets f ([m]) and I are disjoint)
⇐⇒ ( f ([m]) is disjoint from I )
⇐⇒ ( f ([m]) ⊆ [n] \ I )

(because f ([m]) is a subset of [n], and thus is disjoint from I if and only if f ([m]) ⊆ [n] \ I).
In light of this equivalence, we can rewrite (218) as

{s ∈ U | s ∈ Ai for all i ∈ I } = {maps f : [m] → [n] | f ([m]) ⊆ [n] \ I } .

This proves (217).]

Now, if I is a subset of [n], then

|{s ∈ U | s ∈ Ai for all i ∈ I }|

= |{maps f : [m] → [n] | f ([m]) ⊆ [n] \ I }| (by (217))
= (# of maps f : [m] → [n] satisfying f ([m]) ⊆ [n] \ I )
= |[n] \ I ||[m]| (by Exercise 2.4.1 (a), applied to A = [m] , B = [n] and C = I )
 
since |[n] \ I | = |[n]| − | I |
= (|[n]| − | I |)|[m]|
(by Theorem 1.4.7 (a), since I is a subset of [n] )
= (n − | I |)m (since |[n]| = n and |[m]| = m) . (219)

Now, let us note that each subset I of [n] satisfies | I | ∈ {0, 1, . . . , n} (since Theo-
Enumerative Combinatorics: class notes page 276

rem 1.4.7 (b) yields | I | ≤ |[n]| = n). Theorem 2.9.9 yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )|
= ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }|
I ⊆[n]
| {z }
=(n−| I |)m
(by (219))
n
= ∑ (−1)| I | (n − | I |)m = ∑ ∑ (−1)| I | (n − | I |)m
I ⊆[n] k =0 I ⊆[n];
| {z }
| I |=k =(−1)k (n−k)m
(since | I |=k)
 
by an application of (37), since
each subset I of [n] satisfies | I | ∈ {0, 1, . . . , n}
n
= ∑ ∑ (−1)k (n − k)m
k =0 I ⊆[n];
| I |=k
| {z }
=(# of subsets I of [n] satisfying | I |=k)·(−1)k (n−k)m
n
= ∑ |(# of subsets I of [n] satisfying | I | = k) · (−1)k (n − k)m
k =0
{z   }
n
=(# of k-element subsets of [n])=
k
(by Theorem 1.3.12, applied to S=[n])
n   n  
n n
= ∑ · (−1)k (n − k)m = ∑ (−1) k
(n − k )m
k =0
k k =0
| {z } k
=(−1)n−(n−k) | {z } 
(since k=n−(n−k)) n
=
n−k
(by Theorem 1.3.11)
n   n  
n n −i n m
= ∑ (−1) n−(n−k)
n−k
(n − k) = ∑ (−1)
m
i
i
k =0 i =0

(here, we have substituted i for n − k in the sum). Comparing this with (216), we
obtain
n  
n −i n m
sur (m, n) = ∑ (−1) i .
i =0
i
This proves Theorem 2.4.17.

2.9.5. Application: Derangements

In Definition 1.7.4, we defined the notion of a derangement; in Definition 1.7.7, we
introduced the notation Dn for the # of derangements of [n]. We now intend to use
the Principle of Inclusion and Exclusion to obtain a formula for Dn – namely, the
formula from Theorem 1.7.9 (d).
Enumerative Combinatorics: class notes page 277

We shall use the following notation (which is fairly widespread in abstract alge-
bra):
Definition 2.9.13. Let X be any set. Then, the set of all permutations of X is
denoted by SX .
Thus, if X is a finite set, then
SX = {permutations of X }
and therefore
|SX | = (# of permutations of X ) = | X |! (220)
(by Theorem 1.7.2, applied to n = | X |), since X is an | X |-element set.
In order to count the derangements of [n] for some n ∈ N, we can try to apply
Theorem 2.9.9 to
n o
U = S[ n ] and A i = σ ∈ S[ n ] | σ ( i ) = i .

This would shift our problem to the computation of |{s ∈ U | s ∈ Ai for all i ∈ I }|
for subsets I of [n]. The following proposition will help us do so:
Proposition 2.9.14. Let X be a set. Let I be a subset of X. Then, there is a bijection

from {σ ∈ SX | σ (i ) = i for each i ∈ I } to SX \ I .

Example 2.9.15. For this example, set X = [4] and I = {1, 3}. Then, Proposition
2.9.14 claims that there is a bijection
n o
from σ ∈ S[4] | σ (i ) = i for each i ∈ {1, 3} to S{2,4}

(since X \ I = [4] \ {1, 3} = {2, 4}). Let us check this: The permutations σ ∈ S[4]
satisfying (σ (i ) = i for each i ∈ {1, 3}) are
   
1 2 3 4 1 2 3 4
and
1 2 3 4 1 4 3 2

(in two-line notation). Meanwhile, the permutations of {2, 4} are

   
2 4 2 4
and
2 4 4 2

(in two-line notation). Thus,

n there is a bijection from the former o two permu-
tations (i.e., from the set σ ∈ S[4] | σ (i ) = i for each i ∈ {1, 3} ) to the latter
two permutations (i.e., to the set S{2,4} ): Namely, the bijection is “throw away
the values at 1 and at 3” or, more formally, “restrict the permutation to the sub-
set {2, 4}”.
Enumerative Combinatorics: class notes page 278

The following proof of Proposition 2.9.14 generalizes the bijection we have found
in this example:
Proof of Proposition 2.9.14 (sketched). Here is the rough idea of the proof: If a per-
mutation σ ∈ SX satisfies (σ (i ) = i for each i ∈ I ), then it has to map each element
of I to itself, and therefore must permute the remaining elements of X among
themselves (i.e., it must send any element of X \ I to an element of X \ I), since
it would otherwise fail to be injective. Thus, a permutation σ ∈ SX satisfying
(σ (i ) = i for each i ∈ I ) is “nothing but” a permutation of the set X \ I (since the
requirement (σ (i ) = i for each i ∈ I ) uniquely determines its values on the set I).
This is not rigorous, because strictly speaking a permutation of X cannot be a per-
mutation of X \ I (after all, the former has domain X while the latter only has
domain X \ I). Here is a rigorous version of the argument we just made:
To each permutation σ ∈ SX satisfying (σ (i ) = i for each i ∈ I ), we can assign a
permutation e σ of X \ I by letting

σ ( p) = σ ( p)
e for each p ∈ X \ I.
193 This defines a map

A : {σ ∈ SX | σ (i ) = i for each i ∈ I } → SX \ I ,
σ 7→ e
σ.

Conversely, to each permutation τ of X \ I, we can assign a permutation τb ∈ SX

satisfying (τb (i ) = i for each i ∈ I ) by setting
(
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I

194 This defines a map

B : SX \ I → {σ ∈ SX | σ (i ) = i for each i ∈ I } ,
τ 7→ τb.

The maps A and B are well-defined and mutually inverse195 . Hence, they are
bijections. Thus, there is a bijection from the set {σ ∈ SX | σ (i ) = i for each i ∈ I }
to the set SX \ I (namely, A). This proves Proposition 2.9.14.

193 Why σ well-defined, and why is it really a permutation of X \ I ? Try to answer these
is this map e
questions (and similar questions that we leave unanswered in this proof) on your own, or look
up the answers in the solution to Exercise 2.9.3 below.
194 Once again: This should be proved! (Proving this is part of Exercise 2.9.3 below.)
195 Once again: This should be proved! (Proving this is part of Exercise 2.9.3 below.)
Enumerative Combinatorics: class notes page 279

Exercise 2.9.3. Fill in the missing details in the above (sketched) proof of Propo-
sition 2.9.14.

Corollary 2.9.16. Let n ∈ N. Let X be an n-element set. Let I be a subset of X.

Then,
|{σ ∈ SX | σ (i ) = i for each i ∈ I }| = (n − | I |)!.

Proof of Corollary 2.9.16. The set X is finite (since it is an n-element set). Hence, its
subset X \ I is finite as well. Moreover, I is a subset of X; thus, Theorem 1.4.7 (a)
(applied to A = X and B = I) yields

|X \ I | = |X| − |I| = n − |I| .

|{z}
=n
(since X is
an n-element set)

Proposition 2.9.14 shows that there is a bijection

from {σ ∈ SX | σ (i ) = i for each i ∈ I } to SX \ I .

Hence, the bijection principle yields

|{σ ∈ SX | σ (i ) = i for each i ∈ I }| = SX \ I = | X \ I |!

(by (220), applied to X \ I instead of X). In view of | X \ I | = n − | I |, this rewrites

as
|{σ ∈ SX | σ (i ) = i for each i ∈ I }| = (n − | I |)!
This proves Corollary 2.9.16.
We can now prove Theorem 1.7.9 (d):
Proof of Theorem 1.7.9 (d). Let n ∈ N. Set U = S[n] . For each i ∈ [n], we define a set
n o
A i = σ ∈ S[ n ] | σ ( i ) = i . (221)

This set Ai is a subset of S[n] = U. Thus, we have found n subsets A1 , A2 , . . . , An of

U. We claim that

U \ ( A1 ∪ A2 ∪ · · · ∪ An ) = {derangements of [n]} . (222)

[Proof of (222): Let σ be a permutation of [n]. Hence, σ ∈ S[n] (by the definition of
S[n] ). Hence, for any i ∈ [n], we have the equivalence (σ ∈ Ai ) ⇐⇒ (σ (i ) = i ) (by
(221)). Thus, we have the following equivalence:

(there exists no i ∈ [n] such that σ ∈ Ai )

⇐⇒ (there exists no i ∈ [n] such that σ (i ) = i ) . (223)
Enumerative Combinatorics: class notes page 280

But a fixed point of σ is defined to be an element i ∈ [n] such that σ (i ) = i.

Hence, we have the following chain of equivalences:

(σ has no fixed points)

⇐⇒ (there exists no i ∈ [n] such that σ (i ) = i )
⇐⇒ (there exists no i ∈ [n] such that σ ∈ Ai ) (by (223))
⇐⇒ (σ ∈ / Ai for all i ∈ [n])
⇐⇒ (σ ∈ / A1 ∪ A2 ∪ · · · ∪ A n ) . (224)

Forget that we fixed σ. We thus have proved the equivalence (224) for each
permutation σ of [n].
Recall that a derangement of [n] is defined as a permutation of [n] that has no
fixed points. Hence,

{derangements of [n]}
= {permutations of [n] that have no fixed points}
= {σ is a permutation of [n] | σ has no fixed points}
= {σ is a permutation of [n] | σ ∈ / A1 ∪ A2 ∪ · · · ∪ A n }
(by the equivalence (224))
n o
= σ ∈ S[ n ] | σ ∈/ A1 ∪ A2 ∪ · · · ∪ A n
 
since the permutations of [n] are the elements of S[n]
= S[ n ] \ ( A1 ∪ A2 ∪ · · · ∪ A n ) = U \ ( A1 ∪ A2 ∪ · · · ∪ A n ) .
|{z}
=U

This proves (222).]

Now, (222) yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )|
= |{derangements of [n]}|
= (# of derangements of [n]) = Dn (225)

(since Dn was defined as the # of derangements of [n]).

In order to apply Theorem 2.9.9 successfully, we want to understand the sets
{s ∈ U | s ∈ Ai for all i ∈ I } (or at least their sizes). But this is easy: If I is a
subset of [n], then

{s ∈ U | s ∈ Ai for all i ∈ I }
n o
= σ ∈ S[n] | σ (i ) = i for each i ∈ I . (226)
Enumerative Combinatorics: class notes page 281

[Proof of (226): Let I be a subset of [n]. We have

{s ∈ U | s ∈ Ai for all i ∈ I }
= {s ∈ U | s ∈ Ai for each i ∈ I }
= {σ ∈ U | σ ∈ Ai for each i ∈ I }
(here, we have renamed the index s as σ)
n o  
= σ ∈ S[n] | σ ∈ Ai for each i ∈ I since U = S[n]
n o
= σ ∈ S[n] | σ (i ) = i for each i ∈ I

(because for each σ ∈ S[n] and each i ∈ I, we have the equivalence (σ ∈ Ai ) ⇐⇒ (σ (i ) = i )

196 ).
This proves (226).]
Now, if I is a subset of [n], then

|{s ∈ U | s ∈ Ai for all i ∈ I }|

n o
= σ ∈ S[n] | σ (i ) = i for each i ∈ I (by (226))
= (n − | I |)! (227)

(by Corollary 2.9.16, applied to X = [n]).

Now, let us note that each subset I of [n] satisfies | I | ∈ {0, 1, . . . , n} (since Theo-

196 by (221)
Enumerative Combinatorics: class notes page 282

rem 1.4.7 (b) yields | I | ≤ |[n]| = n). Theorem 2.9.9 yields

|U \ ( A1 ∪ A2 ∪ · · · ∪ An )|
= ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }|
I ⊆[n]
| {z }
=(n−| I |)!
(by (227))
n
= ∑ (−1)| I | (n − | I |)! = ∑ ∑ (−1)| I | (n − | I |)!
I ⊆[n] k =0 I ⊆[n];
| {z }
| I |=k =(−1)k (n−k)!
(since | I |=k)
 
by an application of (37), since
each subset I of [n] satisfies | I | ∈ {0, 1, . . . , n}
n
= ∑ ∑ (−1)k (n − k)!
k =0 I ⊆[n];
| I |=k
| {z }
=(# of subsets I of [n] satisfying | I |=k)·(−1)k (n−k)!
n
= ∑ |(# of subsets I of [n] satisfying | I | = k) · (−1)k (n − k)!
k =0
{z   }
n
=(# of k-element subsets of [n])=
k
(by Theorem 1.3.12, applied to S=[n])
n   n
n n!
= ∑ k
· (−1) (n − k)! = ∑
k
k! · (n − k )!
· (−1)k (n − k)!
k =0 | {z } k =0 | {z }
n! n!
= =(−1)k
k! · (n − k )! k!
(by Theorem 1.3.9)
n
n!
= ∑ (−1)k k! .
k =0

Comparing this with (225), we obtain

n
n!
Dn = ∑ (−1)k k! .
k =0

This proves Theorem 1.7.9 (d).

Exercise 2.9.4. Prove parts (a), (b) and (c) of Theorem 1.7.9.

2.9.6. Application: Euler’s totient function

We recall two standard notions from elementary number theory (see, e.g., [19s, §2.9
and §2.10]):
Enumerative Combinatorics: class notes page 283

• The greatest common divisor of two positive integers a and b is defined to be the
greatest positive integer that divides both a and b. It is denoted by gcd ( a, b).
For example, gcd (4, 6) = 2 and gcd (3, 7) = 1 and gcd (10, 25) = 5.
• Two positive integers a and b are said to be coprime (or relatively prime) if
gcd ( a, b) = 1.
• A positive integer a is said to be coprime to a positive integer b if gcd ( a, b) = 1
(that is, if a and b are coprime).
(These definitions can be extended to arbitrary integers, if we set gcd (0, 0) = 0
(despite 0 not literally being the greatest common divisor of 0 and 0). But we will
only need them for positive integers.)
We now define a function that has many uses in number theory (see, e.g., [19s,
§2.14]):
Definition 2.9.17. The function φ : {1, 2, 3, . . .} → N (called Euler’s totient function
or Euler’s φ-function) is defined by

φ (u) = (# of all m ∈ [u] that are coprime to u) .

See [Tou17] for a history of this function (and an attempt at explaining its name).
Example 2.9.18. (a) The definition of φ yields

φ (2) = (# of all m ∈ [2] that are coprime to 2) . (228)

Among the elements 1, 2 of [2], only 1 is coprime to 2 (since gcd (1, 2) = 1 and
gcd (2, 2) = 2). Thus, the # of all m ∈ [2] that are coprime to 2 is 1. Hence, (228)
becomes
φ (2) = 1.
(b) The definition of φ yields

φ (4) = (# of all m ∈ [4] that are coprime to 4) . (229)

Among the elements 1, 2, 3, 4 of [4], only 1 and 3 are coprime to 4 (since

gcd (1, 4) = 1, gcd (2, 4) = 2, gcd (3, 4) = 1 and gcd (4, 4) = 4). Thus, the #
of all m ∈ [4] that are coprime to 4 is 2. Hence, (229) becomes

φ (4) = 2.

(c) Likewise,
φ (6) = 2,
since the only elements among 1, 2, 3, 4, 5, 6 that are coprime to 6 are 1 and 5.
(d) Likewise,
φ (7) = 6,
Enumerative Combinatorics: class notes page 284

since the only elements among 1, 2, 3, 4, 5, 6, 7 that are coprime to 7 are

1, 2, 3, 4, 5, 6.
(e) Likewise,
φ (12) = 4,
since the only elements among 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 that are coprime to
12 are 1, 5, 7, 11.

Here is a table of the first 15 values of φ:

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 ···
.
φ (n) 1 1 2 2 4 2 6 4 6 4 10 4 12 6 8 ···

The function φ can, of course, be seen as a sequence of integers (namely, the

sequence (φ (1) , φ (2) , φ (3) , . . .)). As such, it appears in the OEIS as OEIS sequence
A000010.
Here are two neat (but inconsequential, at least for us) exercises on the function
φ (which can be solved essentially combinatorially):

Exercise 2.9.5. Let n ∈ N satisfy n > 2. Prove that φ (n) is even.

Exercise 2.9.6. Let n ∈ N satisfy n > 1. Prove that

∑ i = nφ (n) /2.
i ∈{1,2,...,n};
i is coprime to n

We shall apply the Principle of Inclusion and Exclusion to proving the following
“explicit” formula for φ (u):

Theorem 2.9.19. Let u be a positive integer. Let p1 , p2 , . . . , pn be the distinct

primes that divide u. Then,
n  
1
φ (u) = u · ∏ 1 − .
i =1
pi

Example 2.9.20. Let u = 18. What does Theorem 2.9.19 say in this case? The
distinct primes that divide u are 2 and 3 (since u = 18 = 2 · 32 ). Hence, we can
apply Theorem 2.9.19 to n = 2, p1 = 2 and p2 = 3. We thus obtain
2     
1 1 1
φ (u) = u · ∏ 1− = u· 1− 1− .
i =1
pi 2 3
Enumerative Combinatorics: class notes page 285

In view of u = 18, this rewrites as

  
1 1
φ (18) = 18 · 1 − 1− = 6.
2 3

This is easy to verify by actually counting all m ∈ [18] that are coprime to 18.

There are many ways to prove Theorem 2.9.19; for example, one can be found
in [19s, §2.16.3]. We shall give a combinatorial proof that uses the Principle of
Inclusion and Exclusion as well as a few elementary lemmas. The first lemma is a
pair of simple identities:

Lemma 2.9.21. Let n ∈ N. Let a1 , a2 , . . . , an be any n numbers. Then:

(a) We have
∑ ∏ a i = (1 + a1 ) (1 + a2 ) · · · (1 + a n ) .
I ⊆[n] i ∈ I

(b) We have

∑ (−1)| I | ∏ ai = (1 − a1 ) (1 − a2 ) · · · (1 − an ) .
I ⊆[n] i∈ I

Example 2.9.22. If n = 3, then Lemma 2.9.21 (a) says that

1 + a1 + a2 + a3 + a1 a2 + a1 a3 + a2 a3 + a1 a2 a3 = (1 + a1 ) (1 + a2 ) (1 + a3 ) .

This is no surprise: The left hand side is what you obtain if you expand the right
hand side.

Exercise 2.9.7. Prove Lemma 2.9.21.

The remaining lemmas that we will need come from elementary number theory.
The first is a simple counting exercise:197

Lemma 2.9.23. Let u and v be two positive integers such that v divides u. Then,

(# of m ∈ [u] such that v divides m) = u/v.

Proof of Lemma 2.9.23 (sketched). The positive integers u and v have the property that
v divides u. Hence, u/v is a positive integer. Moreover, if m ∈ [u], then m ≤ u.

197 Inthe following, we will (mostly) avoid using the symbol “|” for the word “divides” (in the
notation “a | b”). Instead, we will just write out the word “divides”. We will do this in order to
avoid confusing expressions like “{m ∈ [u] | v | m}”, because the symbol “|” is already used in
set-builder notation (e.g., in “{m ∈ [u] | v divides m}”).
Enumerative Combinatorics: class notes page 286

Hence, if m ∈ [u] has the property that v divides m, then m/v is a positive integer
satisfying m/v ≤ u/v (since m ≤ u), and thus belongs to [u/v]. Hence, the map
A : {m ∈ [u] | v divides m} → [u/v] ,
m 7→ m/v
is well-defined. Consider this map A.
On the other hand, if j ∈ [u/v], then jv is an element of [u] (because j ∈ [u/v]
yields 0 < j ≤ u/v and thus 0 < jv ≤ u) and has the property that v divides jv.
Hence, the map
B : [u/v] → {m ∈ [u] | v divides m} ,
j 7→ jv
is well-defined. Consider this map B.
The maps A and B are mutually inverse (since A divides the input by v, whereas
B multiplies the input by v), and hence are bijections. Thus, the bijection principle
yields
|{m ∈ [u] | v divides m}| = |[u/v]| = u/v
(since u/v is a positive integer). Hence,
(# of m ∈ [u] such that v divides m) = |{m ∈ [u] | v divides m}| = u/v.
This proves Lemma 2.9.23.
The next lemma is a particular case of [19s, Exercise 2.10.3]:

Lemma 2.9.24. Let c ∈ Z. Let b1 , b2 , . . . , bk be k positive integers that are mutually

coprime. (“Mutually coprime” means that bi is coprime to b j whenever i 6= j).
Assume that bi | c for each i ∈ {1, 2, . . . , k }. Then, b1 b2 · · · bk | c.

Using this lemma, we can easily see the following fact:

Lemma 2.9.25. Let c ∈ Z. Let b1 , b2 , . . . , bk be k distinct primes. Assume that

bi | c for each i ∈ {1, 2, . . . , k }. Then, b1 b2 · · · bk | c.

Exercise 2.9.8. Derive Lemma 2.9.25 from Lemma 2.9.24.

From Lemma 2.9.25, in turn, we can easily derive the following lemma:

Lemma 2.9.26. Let I be a finite set. For each i ∈ I, let pi be a prime. Assume that
all these primes pi (for different i ∈ I) are distinct. Let m be an integer. Then, we
have the equivalence
!
p j divides m for each j ∈ I ⇐⇒ ∏ pi divides m .


(230)
i∈ I
Enumerative Combinatorics: class notes page 287

Exercise 2.9.9. Prove Lemma 2.9.26.

The next lemma stems easily from the definition of a greatest common divisor:

Lemma 2.9.27. Let u be a positive integer. Let p1 , p2 , . . . , pn be the distinct primes

that divide u. Let m be a positive integer. Then, we have the equivalence

(m is coprime to u) ⇐⇒ (( pi does not divide m) for each i ∈ [n]) .

Exercise 2.9.10. Prove Lemma 2.9.27.

[Hint: A well-known fact says that if n > 1 is an integer, then there exists at
least one prime p such that p | n.]

We can now prove Theorem 2.9.19 using Theorem 2.9.9:

Proof of Theorem 2.9.19. Define a finite set U of positive integers by

U = [u] .

For each i ∈ [n], define a subset Ai of U by

Ai = {m ∈ U | pi divides m} . (231)

Thus, we have found n subsets A1 , A2 , . . . , An of U.

For any m ∈ U, we have the logical equivalence

(m ∈ Ai ) ⇐⇒ ( pi divides m) (232)

(by (231)), and thus we also have the logical equivalence

(m ∈
/ Ai ) ⇐⇒ ( pi does not divide m) . (233)

We claim that

U \ ( A1 ∪ A2 ∪ · · · ∪ An ) = {m ∈ [u] | m is coprime to u} . (234)

[Proof of (234): For any m ∈ U, we have the logical equivalence

(m is coprime to u) ⇐⇒ (( pi does not divide m) for each i ∈ [n])

(by Lemma 2.9.27). Thus,

{m ∈ U | m is coprime to u}
= {m ∈ U | ( pi does not divide m) for each i ∈ [n]} . (235)
Enumerative Combinatorics: class notes page 288

But the definition of set difference yields

U \ ( A1 ∪ A2 ∪ · · · ∪ A n )
= {m ∈ U | m ∈ / A1 ∪ A2 ∪ · · · ∪ A n }
 

 


 


 

 
= m∈U | m∈ / Ai for each i ∈ [n]

 | {z } 



 This is equivalent to 



 “pi does not divide m” 

(by (233))
= {m ∈ U | ( pi does not divide m) for each i ∈ [n]}
= {m ∈ U | m is coprime to u} (by (235))
= {m ∈ [u] | m is coprime to u} (since U = [u]) .
This proves (234).]
Now, the definition of φ yields
φ (u) = (# of all m ∈ [u] that are coprime to u)
= |{m ∈ [u] | m is coprime to u}|
= |U \ ( A1 ∪ A2 ∪ · · · ∪ An )| (236)
(since (234) yields {m ∈ [u] | m is coprime to u} = U \ ( A1 ∪ A2 ∪ · · · ∪ An )).
In order to apply Theorem 2.9.9 successfully, we want to understand the sets
{s ∈ U | s ∈ Ai for all i ∈ I } (or at least their sizes). But this is easy: If I is a
subset of [n], then
{s ∈ U | s ∈ Ai for all i ∈ I }
( )
= m ∈ [u] | ∏ pi divides m . (237)
i∈ I

[Proof of (237): Let I be a subset of [n]. We have

{s ∈ U | s ∈ Ai for all i ∈ I }
= {s ∈ U | s ∈ Ai for each i ∈ I }
= {m ∈ U | m ∈ Ai for each i ∈ I }
(here, we have renamed the index s as m)
= {m ∈ U | pi divides m for each i ∈ I }
(due to the equivalence (232), which holds for every m ∈ U )
= {m ∈ [u] | pi divides m for each i ∈ I } (since U = [u])

= m ∈ [u] | p j divides m for each j ∈ I
(here, we have renamed the index i as j)
( )
= m ∈ [u] | ∏ pi divides m (by the equivalence (230)) .
i∈ I
Enumerative Combinatorics: class notes page 289

This proves (237).]

As a consequence of (237), we can now easily see the following: If I is a subset
of [n], then
1
|{s ∈ U | s ∈ Ai for all i ∈ I }| = u · ∏ . (238)
i∈ I
p i

[Proof of (238): Let I be a subset of [n]. All the primes pi (for different i ∈ I) are
distinct (since p1 , p2 , . . . , pn are the distinct primes that divide u). Furthermore, p j
divides u for each j ∈ I (for the same reason). But Lemma 2.9.26 (applied to m = u)
shows that we have the equivalence
!
p j divides u for each j ∈ I ⇐⇒ ∏ pi divides u .


i∈ I

Hence, ∏ pi divides u (since p j divides u for each j ∈ I). Thus, Lemma 2.9.23
i∈ I
(applied to v = ∏ pi ) yields that
i∈ I
! !
# of m ∈ [u] such that ∏ pi divides m = u/ ∏ pi .
i∈ I i∈ I

But (237) yields

|{s ∈ U | s ∈ Ai for all i ∈ I }|

( )
= m ∈ [u] | ∏ pi divides m
i∈ I
! !
= # of m ∈ [u] such that ∏ pi divides m = u/ ∏ pi
i∈ I i∈ I
1 1
= u· = u·∏ .
∏ pi p
i∈ I i
i∈ I
| {z }
1
=∏
i∈ I pi

This proves (238).]

Enumerative Combinatorics: class notes page 290

Now, everything is in place. The equality (236) becomes

φ (u) = |U \ ( A1 ∪ A2 ∪ · · · ∪ An )|
= ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }| (by Theorem 2.9.9)
I ⊆[n]
| {z }
1
=u· ∏
i∈ I pi
(by (238))
1 1
= ∑ (−1)| I | u · ∏
pi
= u· ∑ (−1)| I | ∏
pi
I ⊆[n] i∈ I I ⊆[n] i∈ I
| {z }
1 1 1
! ! !
= 1− 1− ··· 1−
p1 p2 pn
1
(by Lemma 2.9.21 (b), applied to ai =
)
pi
     n  
1 1 1 1
= u· 1− 1− ··· 1− = u·∏ 1− .
p1 p2 pn i = 1
pi
| {z }
1
!
n
= ∏ 1−
i =1 pi
This proves Theorem 2.9.19.

2.9.7. Other cancellation-type lemmas

Propositions 2.9.10 and 2.9.12 belong to a genre of results saying that certain sums
simplify to 0 in most cases. The following exercises give other specimens of this
kind of result:
Exercise 2.9.11. Let S and T be two finite sets. Prove that
(
2|S| , if S ⊆ T;
∑ (−1) = 0, otherwise .
| I \T|
I ⊆S

Exercise 2.9.12. Let S be a finite set. Let X and Y be two distinct subsets of S.
Prove that
∑ (−1)|X∩ I |+|Y∩ I | = 0.
I ⊆S

Class of 2019-11-04

2.10. Compositions and weak compositions

We already know how to count n-tuples of elements of a set A: They are the
elements of An , so their number is | An | = | A|n (by Corollary 1.5.5). We shall now
Enumerative Combinatorics: class notes page 291

turn to some subtler tuple-counting problems. Specifically, we will count tuples of

integers with a given sum.

2.10.1. Compositions
How many ways are there to write 5 as a sum of 3 positive integers, if the order
matters? In other words, how many 3-tuples ( x1 , x2 , x3 ) of positive integers satisfy
x1 + x2 + x3 = 5 ? The answer is that there are 6 such 3-tuples (i.e., there are 6 ways
to write 5 as a sum of 3 positive integers), namely

(1, 1, 3) , (1, 3, 1) , (3, 1, 1) , (1, 2, 2) , (2, 1, 2) , (2, 2, 1) .

Indeed,

5 = 1 + 1 + 3 = 1 + 3 + 1 = 3 + 1 + 1 = 1 + 2 + 2 = 2 + 1 + 2 = 2 + 2 + 1.

What if we replace 5 and 3 by two arbitrary nonnegative integers n and k ?

Theorem 2.10.1. Let P = {1, 2, 3, . . .} be the set of all positive integers. Let n ∈ N
and k ∈ N. Then,
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n
n−1
 
= (239)
n−k

 n − 1 , if n > 0;


= k−1 . (240)

 [ k = 0] , if n = 0

Before we prove Theorem 2.10.1, let us introduce some terminology for the tuples
counted in it:

Definition 2.10.2. (a) A composition shall mean a tuple (i.e., a finite list) of positive
integers.
(b) If k ∈ N, then a composition into k parts shall mean a k-tuple of positive
integers.
(c) If n ∈ N, then a composition of n shall mean a tuple of positive integers
whose sum is n.
(d) If n ∈ N and k ∈ N, then a composition of n into k parts shall mean a k-tuple
of positive integers whose sum is n.

Example 2.10.3. (a) The compositions of 5 into 3 parts are

(1, 1, 3) , (1, 3, 1) , (3, 1, 1) , (1, 2, 2) , (2, 1, 2) , (2, 2, 1) .

Enumerative Combinatorics: class notes page 292

This is just a restatement of what we said before Theorem 2.10.1.

(b) The compositions of 3 are

(3) , (1, 2) , (2, 1) , (1, 1, 1) .

The first of these is a composition into 1 part; the last is a composition into 3
parts; the other two are compositions into 2 parts.
(c) The only composition of 0 is the 0-tuple (); it is a composition into 0 parts.

If n ∈ N and k ∈ N, then the compositions of n into k parts are the k-tuples of

positive integers whose sum is n. In other words, they are the k-tuples ( x1 , x2 , . . . , xk ) ∈
Pk satisfying x1 + x2 + · · · + xk = n (where we are using the notation P = {1, 2, 3, . . .}).
Hence, Theorem 2.10.1 can be restated as follows:
(# of compositions of n into k parts)
n−1
 
=
n−k

 n−1

 , if n > 0;
= k−1

 [ k = 0] , if n = 0

for any n ∈ N and k ∈ N.

We shall now outline a proof of Theorem 2.10.1, referring to [19f-hw0s, solution
to Exercise 1 (b)] (where a similar argument is made in a slightly different proof)
and to Exercise 2.10.1 for some missing details. A different proof will be given in
Exercise 2.10.2 further below.
Proof of Theorem 2.10.1 (sketched). It is easy to see that Theorem 2.10.1 holds when
n = 0 (see Exercise 2.10.1 (b) below for the proof). Thus, for the rest of this proof,
we WLOG assume that n 6= 0. Hence, n ≥ 1 (since n ∈ N). Thus, n − 1 ∈ N.
It is easy to see that Theorem 2.10.1 holds when k = 0 (see Exercise 2.10.1 (c)
below for the proof). Thus, for the rest of this proof, we WLOG assume that k 6= 0.
Hence, k ≥ 1 (since k ∈ N). Thus, k − 1 ∈ N.
A k-tuple of positive integers whose sum is n is the same as a composition of n
into k parts198 . In other words, a k-tuple ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 +
· · · + xk = n is the same as a composition of n into k parts199 . Therefore,
 
k
# of ( x1 , x2 , . . . , xk ) ∈ P satisfying x1 + x2 + · · · + xk = n
= (# of compositions of n into k parts) . (241)
Now, we define a map
C : {compositions of n into k parts} → {(k − 1) -element subsets of [n − 1]}
198 since this is how a composition of n into k parts was defined in Definition 2.10.2 (d)
199 since P is the set of all positive integers
Enumerative Combinatorics: class notes page 293

by setting

C (( a1 , a2 , . . . , ak )) = { a1 + a2 + · · · + ai | i ∈ [k − 1]}
= { a 1 , a 1 + a 2 , a 1 + a 2 + a 3 , . . . , a 1 + a 2 + · · · + a k −1 }
for all compositions ( a1 , a2 , . . . , ak ) of n into k parts.

It is easy to see that this map C is well-defined200 .

(For example, if n = 10 and k = 3, then applying the map C to the composition
(3, 5, 2) yields C ((3, 5, 2)) = {3, 3 + 5} = {3, 8}.)
Next, we define a map

D : {(k − 1) -element subsets of [n − 1]} → {compositions of n into k parts}

as follows: Let S be a (k − 1)-element subset of [n − 1]. Thus, Proposition 1.4.13 (ap-

plied to m = k − 1) shows that there exists a unique (k − 1)-tuple (s1 , s2 , . . . , sk−1 )
of integers satisfying {s1 , s2 , . . . , sk−1 } = S and s1 < s2 < · · · < sk−1 . Consider
this (k − 1)-tuple, and extend it further to a (k + 1)-tuple (s0 , s1 , . . . , sk ) by setting
s0 = 0 and sk = n. Then, it is easy to see that s0 < s1 < · · · < sk 201 . Now, set

D ( S ) = ( s 1 − s 0 , s 2 − s 1 , . . . , s k − s k −1 ) .

This defines D. It is easy to see that this map D is well-defined202 .

Note that the map D depends on n. For example, if n = 6 and k = 4, then
D ({2, 3, 5}) = (2, 1, 2, 1), but if n = 7 and k = 4, then D ({2, 3, 5}) = (2, 1, 2, 2).
Now, it is easy to check that the maps C and D are mutually inverse203 . Thus,
the map

C : {compositions of n into k parts} → {(k − 1) -element subsets of [n − 1]}

200 Proof.We need to show that { a1 + a2 + · · · + ai | i ∈ [k − 1]} is actually a (k − 1)-element subset
of [n − 1] whenever ( a1 , a2 , . . . , ak ) is a composition of n into k parts.
So let ( a1 , a2 , . . . , ak ) be a composition of n into k parts. We must show that
{ a1 + a2 + · · · + ai | i ∈ [k − 1]} is a (k − 1)-element subset of [n − 1]. In [19f-hw0s,
solution to Exercise 1 (b), proof of the well-definedness of C], it is shown that
{ a1 + a2 + · · · + ai | i ∈ [k − 1]} is a subset of [n − 1]. It thus remains to only show that this
subset is a (k − 1)-element subset.
The k-tuple ( a1 , a2 , . . . , ak ) is a composition; thus, its entries a1 , a2 , . . . , ak are positive integers.
Hence, we have the chain of inequalities

a 1 < a 1 + a 2 < a 1 + a 2 + a 3 < · · · < a 1 + a 2 + · · · + a k −1 .

Thus, the k − 1 numbers a1 , a1 + a2 , a1 + a2 + a3 , . . . , a1 + a2 + · · · + ak−1 are distinct. In

other words, the k − 1 numbers a1 + a2 + · · · + ai for i ∈ [k − 1] are distinct. Hence, the set
{ a1 + a2 + · · · + ai | i ∈ [k − 1]} is a (k − 1)-element set. Since this set is a subset of [n − 1], we
thus conclude that it is a (k − 1)-element subset of [n − 1]. This completes our proof.
201 Indeed, this can be proved just as s < s < · · · < s
0 1 k +1 is proved in [19f-hw0s, solution to Exercise
1 (b), construction of the map D] (with the caveat that what we call k here corresponds to the
k + 1 in [19f-hw0s, solution to Exercise 1 (b), construction of the map D]).
202 This can be done as in [19f-hw0s, solution to Exercise 1 (b), proof of the well-definedness of D].
203 This can be done as in [19f-hw0s, solution to Exercise 1 (b)].
Enumerative Combinatorics: class notes page 294

is invertible, i.e., is a bijection. Hence, the bijection principle yields

|{compositions of n into k parts}|

= |{(k − 1) -element subsets of [n − 1]}|
= (# of (k − 1) -element subsets of [n − 1])
n−1
 
= (242)
k−1

(by Theorem 1.3.12, applied to [n − 1], n − 1 and k − 1 instead of S, n and k).

Now, (241) becomes
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n
= (# of compositions of n into k parts)
= |{compositions of n into k parts}|
n−1
 
= (by (242))
k−1
n−1
   
by Theorem 1.3.11 (applied to n − 1 and k − 1
=
( n − 1) − ( k − 1) instead of n and k), since n − 1 ∈ N
n−1
 
= (since (n − 1) − (k − 1) = n − k) .
n−k

This proves (239). Comparing this with


 n − 1 , if n > 0;


n−1
 
k−1 = (since n > 0)

 [ k = 0] , k−1
if n = 0
n−1
 
= ,
n−k

we obtain (240). Thus, Theorem 2.10.1 is proved.

Exercise 2.10.1. The purpose of this exercise is to supply some missing details
for the above proof of Theorem 2.10.1.
(a) Prove that any n ∈ N and k ∈ N satisfy

n−1

 
n−1
  , if n > 0;
= k−1 .
n−k 
 [ k = 0] , if n = 0

(b) Prove that Theorem 2.10.1 holds for n = 0.

(c) Prove that Theorem 2.10.1 holds for k = 0.
Enumerative Combinatorics: class notes page 295

Exercise 2.10.2. Prove Theorem 2.10.1 by induction on k.

2.10.2. Binary compositions

Let us now count the ways to write a given n ∈ N not as a sum of k positive
integers, but as a sum of k elements of the set {0, 1}:

Theorem 2.10.4. Let n ∈ Z and k ∈ N. Then,

 
# of ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 + · · · + xk = n
 
k
= .
n

Proof of Theorem 2.10.4 (sketched). Here is the idea of the proof:

If ( x1 , x2 , . . . , xk ) ∈ {0, 1}k is a k-tuple, then each index i ∈ [k ] satisfies either
xi = 0 or xi = 1; furthermore, the sum x1 + x2 + · · · + xk of all entries of this k-
tuple is precisely the number of indices i ∈ [k ] satisfying xi = 1 (because each such
index contributes a 1 to this sum, whereas all remaining indices contribute 0’s).
Thus, in order to construct a k-tuple ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 +
· · · + xk = n, we just need to choose which n indices i ∈ [k] will satisfy xi = 1
(because then,  the remaining k − n indices i ∈ [k ] will necessarily satisfy xi = 0).
k
There are many options for this, since this boils down to choosing an n-
n  
k
element subset of [k ] (and Theorem 1.3.12 says that there are such subsets).
n
A formal version of this argument would proceed using the bijection principle. Namely,
it would argue that the map
n o
( x1 , x2 , . . . , xk ) ∈ {0, 1}k | x1 + x2 + · · · + xk = n → {n-element subsets of [k]} ,
( x1 , x2 , . . . , x k ) 7 → { i ∈ [ k ] | x i = 1}

is a bijection; therefore, the bijection principle yields

 
# of ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 + · · · + xk = n
= (# of n-element subsets of [k])
 
k
= (by Theorem 1.3.12, applied to k and n instead of n and k) .
n
Enumerative Combinatorics: class notes page 296

Exercise 2.10.3. Prove Theorem 2.10.4 by induction on k.

2.10.3. Weak compositions

Let us next count the ways to write a given n ∈ N as a sum of k nonnegative
integers:

Theorem 2.10.5. Let n ∈ N and k ∈ N. Then,

 
# of ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n
n+k−1
 
= (243)
n

 n + k − 1 , if k > 0;


= k−1 . (244)

 [ n = 0] , if k = 0

Remark 2.10.6. Tuples of nonnegative integers are called weak compositions. Thus,
Theorem 2.10.5 can be considered as a formula for counting (certain) weak com-
positions.

Example 2.10.7. For n = 2 and k = 3, the k-tuples ( x1 , x2 , . . . , xk ) ∈ Nk satisfying

x1 + x2 + · · · + xk = n are

(0, 0, 2) , (0, 2, 0) , (2, 0, 0) , (0, 1, 1) , (1, 0, 1) , (1, 1, 0) ,

because

2 = 0 + 0 + 2 = 0 + 2 + 0 = 2 + 0 + 0 = 0 + 1 + 1 = 1 + 0 + 1 = 1 + 1 + 0.
2+3−1
   
4
Thus, their # is 6. This is exactly what (243) yields, since = = 6.
2 2

Proof of Theorem 2.10.5. Let P denote the set {1, 2, 3, . . .} of all positive integers.
If x is a nonnegative integer, then x + 1 is a positive integer. In other words,
if x ∈ N, then x + 1 ∈ P. Hence, if a k-tuple ( x1 , x2 , . . . , xk ) belongs to Nk and
satisfies x1 + x2 + · · · + xk = n, then the k-tuple ( x1 + 1, x2 + 1, . . . , xk + 1) belongs
to Pk and satisfies

( x1 + 1) + ( x2 + 1) + · · · + ( xk + 1) = ( x1 + x2 + · · · + xk ) + 1| + 1 +{z· · · + 1} = n + k.
| {z }
=n k times
| {z }
=k
Enumerative Combinatorics: class notes page 297

Hence, the map

n o
k
( x1 , x2 , . . . , x k ) ∈ N | x1 + x2 + · · · + x k = n
n o
→ ( x 1 , x 2 , . . . , x k ) ∈ Pk | x 1 + x 2 + · · · + x k = n + k ,
( x1 , x2 , . . . , xk ) 7→ ( x1 + 1, x2 + 1, . . . , xk + 1)
is well-defined. It is easy to see that this map is a bijection. Thus, the bijection
principle yields
n o
k
( x1 , x2 , . . . , x k ) ∈ N | x1 + x2 + · · · + x k = n
n o
= ( x 1 , x 2 , . . . , x k ) ∈ Pk | x 1 + x 2 + · · · + x k = n + k .

In other words,
 
k
# of ( x1 , x2 , . . . , xk ) ∈ N satisfying x1 + x2 + · · · + xk = n
 
= # of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n + k
(n + k) − 1
 
= (by (239), applied to n + k instead of n)
(n + k) − k
n+k−1
 
= (since (n + k) − 1 = n + k − 1 and (n + k) − k = n) .
n
This proves (243).
Continuing this computation, we obtain
 
# of ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n

n+k−1

 
n+k−1
  , if k > 0;
= = k−1
n 
 [ n = 0] , if k = 0

(by Theorem 1.3.11 and an easy distinction of cases204 ). This proves (244). Thus,
Theorem 2.10.5 is proven.

Exercise 2.10.4. Let n ∈ N and k ∈ N. Prove that


n+k−1

 
n+k−1
  , if k > 0;
= k−1 .
n 
 [ n = 0] , if k = 0

The equality (243) also appears in [Grinbe15, Exercise 3.15] (with a proof by
induction).
204 See Exercise 2.10.4 below for a detailed justification of the last equality sign.
Enumerative Combinatorics: class notes page 298

2.10.4. Other composition-like counting problems

Theorem 2.10.1, Theorem 2.10.5 and Theorem 2.10.4 are perhaps the three most
important results about counting tuples of integers with a given sum. But there are
many others. First of all, we can count compositions of n without specifying the
number of parts:

Theorem 2.10.8. Let n ∈ N. Then,

(
2n −1 , if n > 0
(# of compositions of n) = .
1, if n = 0

For the proof of Theorem 2.10.8, see [19f-hw0s, Exercise 1 (b)].

We may also want to count compositions of n whose entries belong to {1, 2} –
that is, ways to write n as a sum of 1’s and 2’s. The answer turns out to be simple
again:

Exercise 2.10.5. Let n ∈ N.

A {1, 2}-composition of n shall mean a composition ( x1 , x2 , . . . , xk ) of n such
that x1 , x2 , . . . , xk ∈ {1, 2}.
For example, the {1, 2}-compositions of 5 are

(1, 1, 1, 1, 1) , (1, 1, 1, 2) , (1, 1, 2, 1) , (1, 2, 1, 1) ,

(2, 1, 1, 1) , (2, 2, 1) , (2, 1, 2) , (1, 2, 2) .

(a) Prove that

(# of {1, 2} -compositions of n) = f n+1
(where ( f 0 , f 1 , f 2 , . . .) denotes the Fibonacci sequence, as defined in Definition
1.1.10).
(b) Let k ∈ N. A {1, 2}-composition of n into k parts shall mean a composition
( x1 , x2 , . . . , xk ) of n into k parts such that x1 , x2 , . . . , xk ∈ {1, 2}.
Prove that
 
k
(# of {1, 2} -compositions of n into k parts) = .
n−k

Also, we can try to generalize Theorem 2.10.4 by replacing {0, 1} by {0, 1, . . . , p − 1}

for a given p ∈ N. Thus, we look for the # of ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k
satisfying x1 + x2 + · · · +
 xk = n. The answer to this counting problem, however,
k
will not be as simple as :
n
Enumerative Combinatorics: class notes page 299

Exercise 2.10.6. Let p ∈ N. Let n ∈ Z and k ∈ N. Prove that

 
# of ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k satisfying x1 + x2 + · · · + xk = n
k
n − pj + k − 1
  
j k
= ∑ (−1) .
j =0
j n − pj

[Hint: Use the Principle of Inclusion and Exclusion, keeping in mind that
{ 1, . . . , p − 1} = N \ {i ∈ N | i ≥ p}.]
0,

The next exercise is a simple-looking identity for binomial coefficients, which can
be nicely solved using weak compositions despite not explicitly involving them:

Exercise 2.10.7. Let n ∈ N and k ∈ N. Prove that

k 
n+i−1 n+k−1
   
n
∑ i k − 2i
=
k
.
i =0

[Hint: Any nonnegative integer u can be written as u = 2q + r for a unique

q ∈ N and a unique r ∈ {0, 1}. What does this mean for weak compositions of k
into n parts?]

Exercise 2.10.8. Generalize Exercise 2.10.7 to all n ∈ R.

Rather than count the compositions of an integer, we can also count them “with
weight” – i.e., sum certain values over them. Perhaps the simplest such value is the
length. Recall that the length of a tuple ( x1 , x2 , . . . , xk ) is defined to be the integer k.
In other words, the length of a tuple α is the k ∈ N such that α is a k-tuple. Thus,
each composition has a well-defined length (since it is a tuple).

Exercise 2.10.9. Let n be a positive integer. Prove that the sum of the lengths of
all compositions of n is (n + 1) 2n−2 .

2.11. Multisubsets
2.11.1. Definitions
The notion of a set is sufficiently versatile that it is often used as a foundation to
build all of mathematics upon. But as it is, it has its limitations; in particular, a set
cannot contain multiple copies of the same element. (More precisely, a set S can
either contain or not contain an element x; it makes no sense to ask “how often” it
contains x.) Often, one wants to use a kind of collection that can contain elements
with multiplicities205 . The notion of a multiset addresses this need (at least if the

205 For example, the roots of a polynomial naturally come with multiplicities.
Enumerative Combinatorics: class notes page 300

multiplicities of the elements are nonnegative integers206 ). Rather than define this
notion, we shall however introduce a slight modification thereof, namely the notion
of a multisubset of a given set T:

Definition 2.11.1. Let T be a set.

A multisubset of T is formally defined as a map f : T → N such that only
finitely many t ∈ T satisfy f (t) 6= 0.
Informally, we regard such a map f : T → N as a way to encode a “set with
multiplicities” – namely, the “set” in which each t ∈ T appears f (t) many times.
Accordingly, we will use the notation { a1 , a2 , . . . , ak }multi for this multisubset f ,
where ( a1 , a2 , . . . , ak ) is a list of elements of T such that each t ∈ T appears in
this list exactly f (t) times.

For example, if T = [8], then the map from [8] to N given in two-line notation as
 
1 2 3 4 5 6 7 8
1 0 0 2 1 0 3 0

is a multisubset of [8], namely the multisubset {1, 4, 4, 5, 7, 7, 7}multi (that is, the
multisubset of [8] in which 1 appears 1 time, 4 appears 2 times, 5 appears 1 time,
and 7 appears 3 times). Of course, it can be written in many other ways, too
(for example, as {7, 5, 4, 7, 4, 1, 7}multi ). Note that {1, 1}multi 6= {1}multi , despite
{1, 1} = {1}.
The requirement that “only finitely many t ∈ T satisfy f (t) 6= 0” in Definition
2.11.1 prevents us from having infinite multisubsets. Of course, if the set T is finite,
then this requirement is automatically satisfied.
We note that many authors use the notation {{ a1 , a2 , . . . , ak }} instead of our
notation { a1 , a2 , . . . , ak }multi for a multisubset. However, the former notation can
be misunderstood to mean a set consisting of the set { a1 , a2 , . . . , ak }; thus we are
avoiding it.
As we said, instead of defining multisubsets of a set T, we could have defined
multisets in general.207 We are preferring multisubsets of T here, as they are better
adapted to counting purposes.

Definition 2.11.2. Let T be a set. If f : T → N is a multisubset of T, then the size

of this multisubset f is defined to be the nonnegative integer ∑ f (t).
t∈ T

206 There are also variants that allow infinite multiplicities.

207 The easiest way to define them is as follows: A multiset is a pair (S, f ), where S is a finite set
and f : S → {1, 2, 3, . . .} is a map. The elements of S are considered to be the elements of this
multiset, and the values f (s) are understood to be their multiplicities (i.e., any element s ∈ S
is understood to appear f (s) times in the multiset). This time, the map f must have target
{1, 2, 3, . . .} rather than N (that is, it cannot have 0 in its image), since otherwise our multiset
could have “ghost” elements s that are considered to be its elements but appear with multiplicity
0.
Enumerative Combinatorics: class notes page 301

For example, the above-mentioned multisubset {1, 4, 4, 5, 7, 7, 7}multi of [8] has

size 1 + 2 + 1 + 3 = 7. Note that the sum ∑ f (t) in Definition 2.11.2 is well-
t∈ T
defined, since we required in Definition 2.11.1 that only finitely many t ∈ T satisfy
f (t) 6= 0.
If a1 , a2 , . . . , ak are k objects, then the set { a1 , a2 , . . . , ak } doesn’t always have size
k; its size will be smaller than k if some of a1 , a2 , . . . , ak are equal. But the multiset
{ a1 , a2 , . . . , ak }multi always has size k, as the following simple exercise shows:

Exercise 2.11.1. Let T be a set. Let k ∈ N. Let a1 , a2 , . . . , ak be any k elements of

T (not necessarily distinct). Prove that the multisubset { a1 , a2 , . . . , ak }multi of T
has size k.

2.11.2. Counting multisubsets of given size

A finite nonempty set T has infinitely many multisubsets, since any element can
appear any number of times. Thus, counting all multisubsets of a given finite set T
is not a promising combinatorial question. However, we get a finite number if we
restrict ourselves to counting multisubsets with a given size:

Corollary 2.11.3. Let n ∈ N and k ∈ N. Let T be a k-element set. Then,

n+k−1
 
(# of multisubsets of T having size n) = .
n

Example 2.11.4. The multisubsets of [3] having size 2 are

{1, 1}multi , {1, 2}multi , {1, 3}multi ,

{2, 2}multi , {2, 3}multi , {3, 3}multi .

In terms of their definition, these multisubsets are the maps from [3] to N with
the two-line notations
     
1 2 3 1 2 3 1 2 3
, , ,
2 0 0 1 1 0 1 0 1
     
1 2 3 1 2 3 1 2 3
, , ,
0 2 0 0 1 1 0 0 2

respectively. Either way, we see

 that there
 are 6 of them, and this is what Corol-
2+3−1
lary 2.11.3 predicts (because = 6).
2

Proof of Corollary 2.11.3. Let t1 , t2 , . . . , tk be the k elements of T, listed in some order

(with no repetitions). Recall that a multisubset of T is the same as a map f : T → N
such that only finitely many t ∈ T satisfy f (t) 6= 0 (according to Definition 2.11.1).
Enumerative Combinatorics: class notes page 302

In other words, a multisubset of T is the same as a map f : T → N (because

the property that “only finitely many t ∈ T satisfy f (t) 6= 0” is automatically
satisfied208 ). Moreover, if f : T → N is a multisubset of T, then
(the size of f ) = ∑ f (t) (by Definition 2.11.2)
t∈ T
= f ( t1 ) + f ( t2 ) + · · · + f ( t k )
(since t1 , t2 , . . . , tk are the k elements of T). Hence, the multisubsets of T having
size n are precisely the maps f : T → N satisfying f (t1 ) + f (t2 ) + · · · + f (tk ) = n.
Clearly, such maps f : T → N are uniquely determined by their lists of values
( f (t1 ) , f (t2 ) , . . . , f (tk )) ∈ Nk (since t1 , t2 , . . . , tk are the k elements of T). Thus,
the map
n o
{multisubsets of T having size n} → ( x1 , x2 , . . . , xk ) ∈ Nk | x1 + x2 + · · · + xk = n ,
f 7→ ( f (t1 ) , f (t2 ) , . . . , f (tk ))
is a bijection. Hence, the bijection principle yields
(# of multisubsets of T having size n)
 
k
= # of ( x1 , x2 , . . . , xk ) ∈ N satisfying x1 + x2 + · · · + xk = n
n+k−1
 
= (by (243)) .
n
This proves Corollary 2.11.3.

2.11.3. An application to lacunar subsets

We shall now outline an application of multisubsets. First, however, let us state
two propositions, which are fairly obvious intuitively (even though proving them
formally is somewhat time-consuming):
Proposition 2.11.5. Let T be a set of integers. Let S be a multisubset of
T. Then, there exists a unique tuple (s1 , s2 , . . . , sk ) of integers in T satisfying
{s1 , s2 , . . . , sk }multi = S and s1 ≤ s2 ≤ · · · ≤ sk .

Proposition 2.11.6. Let T be a set of integers. Let m ∈ N. Let S be a multisubset

of T having size m. Then, there exists a unique m-tuple (s1 , s2 , . . . , sm ) of integers
in T satisfying {s1 , s2 , . . . , sm }multi = S and s1 ≤ s2 ≤ · · · ≤ sm .

Proposition 2.11.5 and Proposition 2.11.6 are multiset analogues of Proposition

1.4.11 and Proposition 1.4.13, respectively. The latter two propositions say that a
finite set of integers has a unique strictly increasing list of elements; the former two
say that a finite multiset of integers has a unique weakly increasing list of elements.

208 since T is finite

Enumerative Combinatorics: class notes page 303

Exercise 2.11.2. Prove Proposition 2.11.5 and Proposition 2.11.6.

As an application of multisubsets, we shall sketch a solution to the following
exercise (whose statement involves no multisubsets!):
Exercise 2.11.3. Let m ∈ N and a, b ∈ {0, 1, . . . , m}. Prove that

(# of lacunar subsets of [2m] with exactly a even and b odd elements)

m−a m−b
   
= · .
b a

Exercise 2.11.3 originates in [MusPro07, Theorem 3] (although the authors of this

paper forget to assume that a, b ∈ {0, 1, . . . , m}). One might try to solve it using
the dependent product rule, but this does not work: If we (say) first choose the
b odd elements of our lacunar subset, then the # of ways to subsequently choose
the remaining a even elements will depend on our chosen b odd elements. The
exercise is too subtle for this method. I have outlined two solutions of Exercise
2.11.3 in [18s-hw2s, Exercise 3]. A simpler solution has been found by Musiker
and Propp (in [MusPro07]) using multisubsets. Before I sketch this solution, let me
define a multiset analogue of the notion of the union of two disjoint sets:
Definition 2.11.7. Let A and B be two disjoint sets. Let X be a multisubset of A.
Let Y be a multisubset of B.
Then, X ∪ Y shall denote the multisubset of A ∪ B defined by

X ∪ Y = { x1 , x2 , . . . , xk , y1 , y2 , . . . , y` }multi , (245)

where we have written X in the form X = { x1 , x2 , . . . , xk }multi and written Y in

the form Y = {y1 , y2 , . . . , y` }multi .
Here is a different (but equivalent) way to define this multisubset X ∪ Y: Recall
that a multisubset of A is the same as a map f : A → N. Thus, the multisubset X
of A is a map f : A → N. Likewise, the multisubset Y of B is a map g : B → N.
Now, define a map h : A ∪ B → N by setting
(
f (u) , if u ∈ A;
h (u) = for each u ∈ A ∪ B. (246)
g (u) , if u ∈ B

(This is well-defined, since no u ∈ A ∪ B can satisfy both u ∈ A and u ∈ B at the

same time.) This map h : A ∪ B → N is a multisubset of A ∪ B; we define X ∪ Y
to be this multisubset.
The multisubset X ∪ Y of A ∪ B is called the multiset union of X and Y.
For example, the multiset union of the multisubset {1, 2, 2, 4}multi of {1, 2, 3, 4, 5}
with the multisubset {6, 6, 8, 9, 9}multi of {6, 7, 8, 9, 10} is the multisubset
{1, 2, 2, 4}multi ∪ {6, 6, 8, 9, 9}multi = {1, 2, 2, 4, 6, 6, 8, 9, 9}multi
Enumerative Combinatorics: class notes page 304

of {1, 2, . . . , 10}.

Remark 2.11.8. Definition 2.11.7 can be extended to the case when A and B are not
disjoint. However, there are two non-equivalent ways to do so. In other words, there are
two different reasonable notions of X ∪ Y when A and B are not disjoint.
One of these two notions can be defined by the equality (245). Equivalently, it can be
defined by regarding X and Y as maps f : A → N and g : B → N (as in Definition
2.11.7), and defining X ∪ Y as the map h : A ∪ B → N given by

 f (u) ,
 if u ∈ A \ B;
h (u) = g (u) , if u ∈ B \ A; for each u ∈ A ∪ B.

f (u) + g (u) , if u ∈ A ∩ B


For example, this definition entails that the multiset union of the multisubset
{1, 2, 2, 3}multi of [3] with the multisubset {2, 3, 3, 4}multi of [4] is the multisubset

{1, 2, 2, 3}multi ∪ {2, 3, 3, 4}multi = {1, 2, 2, 3, 2, 3, 3, 4}multi = {1, 2, 2, 2, 3, 3, 3, 4}multi

of [4].
The second notion of X ∪ Y can be defined by regarding X and Y as maps f : A → N
and g : B → N (as in Definition 2.11.7), and defining X ∪ Y as the map h0 : A ∪ B → N
given by

 f (u) ,
 if u ∈ A \ B;
0
h (u) = g (u) , if u ∈ B \ A; for each u ∈ A ∪ B

max { f (u) , g (u)} , if u ∈ A ∩ B


(where max S denotes the largest element of any given set S ⊆ N). For example, this
definition entails that the multiset union of the multisubset {1, 2, 2, 3}multi of [3] with the
multisubset {2, 3, 3, 4}multi of [4] is the multisubset

{1, 2, 2, 3}multi ∪ {2, 3, 3, 4}multi = {1, 2, 2, 3, 3, 4}multi

of [4].
These two notions of X ∪ Y coincide when X and Y have no elements in common;
otherwise, the first notion of X ∪ Y has larger size than the second. Note that the second
notion (unlike the first) does not satisfy (245) in general.

It is easy to check that if A and B are two disjoint sets, then any multisubset of
A ∪ B can be uniquely split up as a multiset union of a multisubset of A with a
multisubset of B, in the following sense:

Lemma 2.11.9. Let A and B be two disjoint sets. Let Z be a multisubset of

A ∪ B. Then, there is a unique way to write Z in the form X ∪ Y, where X is a
multisubset of A and where Y is a multisubset of B. (In other words, there exists
a unique pair ( X, Y ) such that X is a multisubset of A and Y is a multisubset of
B and Z = X ∪ Y.)
Enumerative Combinatorics: class notes page 305

Proof of Lemma 2.11.9 (sketched). Let us view the multisubset Z of A ∪ B as a map

f : A ∪ B → N. Then, its restriction f | A : A → N is a multisubset of A, and its
restriction f | B : B → N is a multisubset of B. If we denote these two restrictions
by X and Y, we then have Z = X ∪ Y. It is easy to see that this is the only way
to write Z in the form X ∪ Y, where X is a multisubset of A and where Y is a
multisubset of B. Thus, Lemma 2.11.9 follows.
The following is also rather obvious:

Lemma 2.11.10. Let A and B be two disjoint sets. Let X be a multisubset of A.

Let Y be a multisubset of B. Then, the size of X ∪ Y is the sum of the sizes of X
and Y.

We can now solve Exercise 2.11.3:

Solution to Exercise 2.11.3 (sketched). Let g = m − a − b + 1.
The set [2g] is the union of its two disjoint g-element subsets

E := {even elements of [2g]} = {2, 4, 6, . . . , 2g} and

O := {odd elements of [2g]} = {1, 3, 5, . . . , 2g − 1} .

Let S be a lacunar subset of [2m] with exactly a even and b odd elements. Thus,
|S| = a + b, so that we can write S uniquely in the form S = {s1 < s2 < · · · < s a+b }
(by Proposition 1.4.13). Let us consider these s1 , s2 , . . . , s a+b . Then, since S is lacu-
nar, we have

s1 − 0 ≤ s2 − 2 ≤ s3 − 4 ≤ · · · ≤ s a + b − 2 ( a + b − 1) (247)
209 . Moreover, all the a + b entries si − 2 (i − 1) (for i ∈ [ a + b]) of this chain of

209 Proof. We must show that si − 2 (i − 1) ≤ si+1 − 2i for each i ∈ [ a + b − 1].

So let i ∈ [ a + b − 1]. Then, si < si+1 (since s1 < s2 < · · · < s a+b ), so that si ≤ si+1 − 1 (since si
and si+1 are integers). But recall that the set S is lacunar, and thus contains no two consecutive
integers. If we had si = si+1 − 1, then si and si+1 would be two consecutive integers in S, which
would contradict the previous sentence. Hence, we cannot have si = si+1 − 1. Thus, we have
si 6= si+1 − 1. Combined with si ≤ si+1 − 1, this yields si < si+1 − 1. Since si and si+1 − 1 are
integers, this entails si ≤ (si+1 − 1) − 1 = si+1 − 2. Subtracting 2 (i − 1) from both sides of this
inequality, we obtain si − 2 (i − 1) ≤ si+1 − 2 − 2 (i − 1) = si+1 − 2i.
Now, forget that we fixed i. We thus have showed that si − 2 (i − 1) ≤ si+1 − 2i for each
i ∈ [ a + b − 1]. In other words,

s1 − 0 ≤ s2 − 2 ≤ s3 − 4 ≤ · · · ≤ s a + b − 2 ( a + b − 1) .
Enumerative Combinatorics: class notes page 306

inequalities belong to the set [2g] 210 . Hence, we can define the multisubset211

MS := {s1 − 0 ≤ s2 − 2 ≤ s3 − 4 ≤ · · · ≤ s a+b − 2 ( a + b − 1)}multi

of [2g]. This multisubset MS has exactly a even and b odd elements212 (since the
integers s1 − 0, s2 − 2, s3 − 4, . . . , s a+b − 2 ( a + b − 1) have the same parities213 as the
integers s1 , s2 , s3 , . . . , s a+b , respectively214 ). In other words, the multisubset MS has
exactly a elements from E and exactly b elements from O.
Recall that MS is a multisubset of [2g] = E ∪ O. Hence, we can (using Lemma
2.11.9) uniquely split MS up as a multiset union MS,even ∪ MS,odd , where

MS,even is a multisubset of E

and
MS,odd is a multisubset of O.
Moreover, the multisubset MS,even here has size a (since MS has exactly a elements
from E), and the multisubset MS,odd here has size b (since MS has exactly b elements
from O).
Forget that we fixed S. Thus, for each lacunar subset S of [2m] with exactly a
even and b odd elements, we have constructed a size-a multisubset MS,even of E
and a size-b multisubset MS,odd of O. Hence, we can define a map

{lacunar subsets of [2m] with exactly a even and b odd elements}

→ {size-a multisubsets of E} × {size-b multisubsets of O} ,
S 7→ ( MS,even , MS,odd ) .

210 Proof. We must show that si − 2 (i − 1) ∈ [2g] for each i ∈ [ a + b]. So let us fix i ∈ [ a + b].
Then, 1 ≤ i ≤ a + b. Hence, s1 − 0 ≤ si − 2 (i − 1) ≤ s a+b − 2 ( a + b − 1) (by (247)). Hence,
si − 2 (i − 1) ≥ s1 − 0 = s1 ≥ 1 (since s1 ∈ {s1 < s2 < · · · < s a+b } = S ⊆ [2m]). Also, s a+b ≤ 2m
(since s a+b ∈ {s1 < s2 < · · · < s a+b } = S ⊆ [2m]), and si − 2 (i − 1) ≤ s a+b −2 ( a + b − 1) ≤
|{z}
≤2m
2m − 2 ( a + b − 1) = 2 (m − a − b + 1) = 2g. Combining si − 2 (i − 1) ≥ 1 with si − 2 (i − 1) ≤
| {z }
=g
2g, we obtain si − 2 (i − 1) ∈ {1, 2, . . . , 2g} = [2g], qed.
211 We are using an analogue of the notation from Definition 1.4.12 for multisets here: If T is a set of

integers, and if a1 , a2 , . . . , ak are k numbers from T, then the notation “{ a1 ≤ a2 ≤ · · · ≤ ak }multi ”

shall denote the multisubset { a1 , a2 , . . . , ak }multi of T and simultaneously assert that a1 ≤ a2 ≤
· · · ≤ ak .
212 counted with multiplicity
213 The parity of an integer k is defined to be 0 if k is even and 1 if k is odd. Thus, two integers k and

` have the same parity if and only if they are either both even or both odd.
214 because the former integers differ from the latter integers by a multiple of 2 (to wit: s − 2 (i − 1)
i
differs from si by 2 (i − 1)).
Enumerative Combinatorics: class notes page 307

This map is easily seen to be a bijection215 . Hence, the bijection principle yields

(# of lacunar subsets of [2m] with exactly a even and b odd elements)

= |{size-a multisubsets of E} × {size-b multisubsets of O}|
= (# of size-a multisubsets of E) · (# of size-b multisubsets of O)
| {z } | {z }
=(# of multisubsets of E having size a ) =( # of multisubsets of O having size b )
a+g−1 b+g−1
   
= =
a b
(by Corollary 2.11.3, (by Corollary 2.11.3,
applied to a, g and E instead of n, k and T) applied to b, g and O instead of n, k and T)

(by the product rule)

a+g−1 b+g−1
   
= ·
a b
|  {z } |  {z }
m−b m−a
= =
a b
(since a+ g−1=m−b (since b+ g−1=m− a
(because g=m− a−b+1)) (because g=m− a−b+1))
m−b m−a m−a m−b
       
= · = · .
a b b a

This solves Exercise 2.11.3.

Note that, as a consequence of Exercise 2.11.3, we can obtain the following for-
mula for Fibonacci numbers:
m m 
m−a m−b
  
f 2m+2 = ∑ ∑ · for all m ∈ N.
a =0 b =0 b a

(See [18s-mt1s, solution to Exercise 3 (f)] for more details on how this is proved.)

2.12. Multinomial coefficients

2.12.1. Definition and formulas
Let us next introduce another family of combinatorially meaningful numbers.

215 The inverse map sends any pair ( X, Y ) ∈ {size-a multisubsets of E} × {size-b multisubsets of O}
to the lacunar subset S of [2m] constructed as follows: Write the multiset union X ∪ Y in the
form X ∪ Y = {t1 ≤ t2 ≤ · · · ≤ t a+b }multi (Proposition 2.11.6 guarantees that this can be done in
a unique way), and define S by

S = {t1 + 0 < t2 + 2 < t3 + 4 < · · · < t a+b + 2 ( a + b − 1)} .

Enumerative Combinatorics: class notes page 308

Definition 2.12.1. Let n ∈ N and n1 , n2 , . . . ,nk ∈ N be such  that n1 + n2 + · · · +

n
nk = n. Then, we define a rational number by
n1 , n2 , . . . , n k
 
n n!
= .
n1 , n2 , . . . , n k n1 !n2 ! · · · nk !
 
n
This number is called a multinomial coefficient.
n1 , n2 , . . . , n k
 
n
Remark 2.12.2. (a) We are not defining for negative or non-
n1 , n2 , . . . , n k
 
n
integer n. (This is in contrast to the notion of binomial coefficients .)
  k
n
(b) The notation for multinomial coefficients may get mis-
n1 , n2 , . . . , n k
taken for a binomial coefficient in the case when k = 1. However, in this trivial
case, both notations lead to the same value (namely, 1) 216 , and thus the ambi-
guity is harmless.

Example 2.12.3. We have 7 ∈ N and 2, 3, 2 ∈ N and 2 + 3 + 2 = 7. Thus,

Definition 2.12.1 yields
 
7 7! 5040
= = = 210.
2, 3, 2 2! · 3! · 2! 2·6·2

Here are some algebraic properties of multinomial coefficients:

Proposition 2.12.4. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 + · · · +

nk = n. Then:

216 Inmore details: Assume that n ∈ N and n1 , n2 , . . . , nk ∈ N satisfy n1 + n2 + · · · + nk = n

and k = 1. Then, n =n1 + n2 +· · ·  + nk = n1 (since k = 1), so that n1 = n. Thus, the
n n n
binomial coefficient satisfies = = 1 (by Exercise 1.3.2). But the definition of
n1  n1  n  
n n n! n!
the multinomial coefficient yields = = (since
 n1 , n2 , . . . , nk n1 , n2 , . . . , n k n1 !n2 ! · · · nk ! n1 !
n n!
k = 1), so that = = 1 (since n = n1 ). Hence, both the binomial coefficient
  n ,
1 2n , . . . , n k n 1!  
n n
and the multinomial coefficient equal 1 in this case.
n1 n1 , n2 , . . . , n k
Enumerative Combinatorics: class notes page 309

(a) We have
k
n − n 1 − n 2 − · · · − n i −1
   
n
=∏
n1 , n2 , . . . , n k i =1
ni
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −1
     
n
= ···
n1 n2 n3 nk
| {z }
=1
k −1 
n − n 1 − n 2 − · · · − n i −1

= ∏ ni
.
i =1
 
n
(b) We have ∈ N.
n1 , n2 , . . . , n k

Exercise 2.12.1. Prove Proposition 2.12.4.

2.12.2. Counting maps that take values a given number of times

Next, we shall give two combinatorial interpretations of multinomial coefficients:

Proposition 2.12.5. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 + · · · +

nk = n. Then, the # of maps f : [n] → [k] satisfying

(# of a ∈ [n] such that f ( a) = i ) = ni for each i ∈ [k ]

 
n
is .
n1 , n2 , . . . , n k

Example 2.12.6. We have 7 ∈ N and 2, 3, 2 ∈ N and 2 + 3 + 2 = 7. Hence,

Proposition 2.12.5 (applied to n = 7 and k = 3 and (n1 , n2 , n3 ) = (2, 3, 2)) says
that the # of maps f : [7] → [3] satisfying

(# of a ∈ [n] such that f ( a) = 1) = 2 and

(# of a ∈ [n] such that f ( a) = 2) = 3 and
(# of a ∈ [n] such that f ( a) = 3) = 2
 
7
is = 210. In other words, it says that the # of maps f : [7] → [3] that
2, 3, 2
take the value 1 exactly twice, take the value
 2 exactly thrice, and 
take the value
1 2 3 4 5 6 7
3 exactly twice is 210. One such map is (in two-line
1 3 2 2 1 2 3
notation).

Class of 2019-11-06
Enumerative Combinatorics: class notes page 310

Let us first give an informal way of proving Proposition 2.12.5:

Proof of Proposition 2.12.5 (informal version, sketched). A relevant map shall mean a map
f : [n] → [k ] satisfying

(# of a ∈ [n] such that f ( a) = i ) = ni

for each i ∈ [k ] . (248)
 
n
Thus, we need to prove that the # of all relevant maps is .
n1 , n2 , . . . , n k
A map f : [n] → [k ] (not necessarily relevant) is uniquely determined if you
know which elements of [n] it sends to 1, which elements of [n] it sends to 2, which
elements of [n] it sends to 3, and so on. In other words, a map f : [n] → [k ] is
uniquely determined if you know the k sets

{ a ∈ [n] | f ( a ) = 1} ,
{ a ∈ [n] | f ( a ) = 2} ,
{ a ∈ [n] | f ( a ) = 3} ,
...,
{ a ∈ [n] | f ( a) = k} .

These k sets should be k disjoint subsets of [n] 217 , and their union must be

[n] 218 . If we want the map f to be relevant, then these k sets should have sizes
n1 , n2 , n3 , . . . , nk , respectively219 .
Thus, the following algorithm can be used to construct any relevant map f :
[ n ] → [ k ]:

• First, choose the set { a ∈ [n] | f ( a) = 1}. This must be a subset of [n] having
size n1 . In other words,  must be an n1 -element subset of the n-element
 this
n
set [n]. Thus, we have choices for it.
n1
• Then, choose the set { a ∈ [n] | f ( a) = 2}. This must be a subset of [n] having
size n2 and disjoint from the already chosen set { a ∈ [n] | f ( a) = 1}. In
other words, this must be an n2 -element  subset of the (n − n1 )-element set
n − n1
[n] \ { a ∈ [n] | f ( a) = 1}. Thus, we have choices for it.
n2
• Then, choose the set { a ∈ [n] | f ( a) = 3}. This must be a subset of [n] hav-
ing size n3 and disjoint from the two already chosen sets { a ∈ [n] | f ( a) = 1}
and { a ∈ [n] | f ( a) = 2}. In other words, this must be an n3 -element subset
217 Indeed, they must be disjoint, since any element a in the intersection of two of them would have
two different values of f ( a) at the same time.
218 since each a ∈ [ n ] should have some value f ( a ) ∈ [ k ] assigned to it, and thus lie in one of these k

sets
219 because the equality (248) says that |{ a ∈ [ n ] | f ( a ) = i }| = n for each i ∈ [ k ]
i
Enumerative Combinatorics: class notes page 311

of the (n − n1 −n2 )-element

set [n] \ { a ∈ [n] | f ( a) = 1} \ { a ∈ [n] | f ( a) = 2}.
n − n1 − n2
Thus, we have choices.
n3

 until the last set { a ∈ [n] | f ( a) = k} has been chosen (for which
• And so on,
n − n 1 − n 2 − · · · − n k −1
we have choices).
nk
The dependent product rule thus shows that the total # of relevant maps is
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −1
     
n
···
n1 n2 n3 nk
k 
n − n 1 − n 2 − · · · − n i −1

=∏
i =1
ni
 
n
= (by Proposition 2.12.4 (a)) .
n1 , n2 , . . . , n k
Thus, Proposition 2.12.5 is proved.

Exercise 2.12.2. Give a formal proof of Proposition 2.12.5 (using induction in-
stead of the dependent product rule).

If A is any set and n ∈ N, then the maps from [n] to A are in bijection with the
n-tuples of elements of A (as we have seen, e.g., in the proof of Theorem 1.5.7).
Thus, any properties of the former can be restated in terms of the latter (and vice
versa). For example, we can restate Proposition 2.12.5 in terms of n-tuples:

Proposition 2.12.7. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 + · · · +

nk = n. Then, the # of n-tuples (u1 , u2 , . . . , un ) ∈ [k ]n satisfying

(# of a ∈ [n] such that u a = i ) = ni for each i ∈ [k ] (249)

 
n
is .
n1 , n2 , . . . , n k

Note that the n-tuples (u1 , u2 , . . . , un ) ∈ [k ]n satisfying (249) in Proposition 2.12.7

are simply the n-tuples of elements of [k ] that contain the entry 1 exactly n1 many
times, contain the entry 2 exactly n2 many times, contain the entry 3 exactly n3
many times, and so on.
Proof of Proposition 2.12.7 (sketched). We introduce some shorthand terminology: A
relevant n-tuple shall mean an n-tuple (u1 , u2 , . . . , un ) ∈ [k ]n satisfying

(# of a ∈ [n] such that u a = i ) = ni for each i ∈ [k ] .

 
n
Thus, we need to prove that the # of all relevant n-tuples is .
n1 , n2 , . . . , n k
Enumerative Combinatorics: class notes page 312

A relevant map shall mean a map f : [n] → [k ] satisfying

(# of a ∈ [n] such that f ( a) = i ) = ni

for each i ∈ [k ] .
 
n
Thus, Proposition 2.12.5 shows that the # of all relevant maps is .
n1 , n2 , . . . , n k
Recall that [k ][n] = {maps from [n] to [k ]}. Consider the map

Φ : {relevant maps} → {relevant n-tuples} ,

f 7→ ( f (1) , f (2) , . . . , f (n)) .

This map Φ sends each relevant map f : [n] → [k ] to its list of values at the points220
1, 2, . . . , n. It is clear that this map Φ is a bijection221 . Thus, the bijection principle
yields
|{relevant maps}| = |{relevant n-tuples}| .
In other words,

(# of all relevant maps) = (# of all relevant n-tuples) . (250)

 
n
But we know that the # of all relevant maps is . Thus, (250) shows
 n1 , n2 , . . . , nk
n
that the # of all relevant n-tuples is . This proves Proposition 2.12.7.
n1 , n2 , . . . , n k

2.12.3. Counting anagrams

For a second combinatorial interpretation of multinomial coefficients, we need the
notion of an “anagram”:

Definition 2.12.8. Let n ∈ N. Let α be an n-tuple (of any objects). An anagram of

α shall mean an n-tuple that can be obtained from α by permuting its entries.
In other words (more formally): Write α as α = (α1 , α2 , . . ., αn ). Then, an
anagram of α is an n-tuple of the form ασ(1) , ασ(2) , . . . , ασ(n) , where σ is a
permutation of [n].

“Anagrams” are often called “permutations”, but the latter terminology risks
confusion with the notion of permutations defined in Definition 1.7.1.

220 I am just using “point” as just a suggestive way of saying “element of [n]” here.
221 Indeed, the definition of a “relevant map” requires the map to take the value i exactly ni many
times (for each i ∈ [k]), whereas the definition of a “relevant n-tuple” requires the n-tuple to
contain the entry i exactly ni many times (for each i ∈ [k]). It is clear that if we encode a map by
its list of values, then the former requirement translates into the latter requirement.
Enumerative Combinatorics: class notes page 313

Example 2.12.9. (a) The anagrams of the 3-tuple (1, 4, 6) are (1, 4, 6), (1, 6, 4),
(4, 1, 6), (4, 6, 1), (6, 1, 4) and (6, 4, 1).
(b) The anagrams of the 3-tuple (1, 4, 1) are (1, 4, 1), (4, 1, 1) and (1, 1, 4).
(There are 6 ways to permute the 3 entries of (1, 4, 1), but they only produce
3 different results.)
(c) The only anagram of the 3-tuple (2, 2, 2) is (2, 2, 2) itself. (All 6 ways of
permuting the 3 entries of (2, 2, 2) yield the same result.)

We can now interpret multinomial coefficients as counting anagrams:

Proposition 2.12.10. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 +

· · · + nk = n.
Let α be the n-tuple
 
1, 1, . . . , 1, 2, 2, . . . , 2, . . . , k, k, . . . , k  .
| {z } | {z } | {z }
n1 times n2 times nk times

Then, the # of distinct anagrams ofα (that is, n-tuples obtained from α by per-
n
muting its entries) is .
n1 , n2 , . . . , n k

Example 2.12.11. How many anagrams does the word “anagram” have? (Here,
we regard a word as a tuple of letters; it doesn’t have to make any sense in
English. For example, “nagrama” is an anagram of “anagram”. And yes, a word
counts as its own anagram.)
Equivalently, how many anagrams does the word “aaagmnr” have? (This is an
equivalent question, because the words “anagram” and “aaagmnr” are anagrams
of one another, and thus have the same set of anagrams.)
Equivalently, how many anagrams does the 7-tuple “(1, 1, 1, 2, 3, 4, 5)” have?
(This is an equivalent question, because “(1, 1, 1, 2, 3, 4, 5)” is obtained from
“aaagmnr” by replacing each letter by a number222 .)
Proposition 2.12.10 (applied to n = 7, k = 5 and (n1 , n2 , . . . , nk ) = (3, 1, 1, 1, 1))
says that the answer to the last question is
 
7 7! 7!
= = = 7 · 6 · 5 · 4 = 840.
3, 1, 1, 1, 1 3! · 1! · 1! · 1! · 1! 3!

Thus, the word “anagram” has 840 anagrams.

Let us prove Proposition 2.12.10 informally:

222 Namely, we replaced the letters “a”, “g”, “m”, “n” and “r” by the numbers 1, 2, 3, 4 and 5, respec-
tively. (It doesn’t matter which letter we choose to replace by which number; the replacement
merely has to guarantee that equal letters become equal numbers, while distinct letters become
distinct numbers.)
Enumerative Combinatorics: class notes page 314

Proof of Proposition 2.12.10 (informal version). If we permute the entries of an n-tuple,

then the order of its entries will change (usually), but some things won’t change:
The number of times 1 appears as an entry won’t change; the number of times 2
appears as an entry won’t change; the number of times 3 appears as an entry won’t
change; and so on. Thus, if β is any anagram of α, and if i ∈ [k ] is arbitrary, then

(# of times i appears in β) = (# of times i appears in α) (251)

= ni

(since i appears ni times in the n-tuple α). Hence, if β is an anagram of α, then

(# of times i appears in β) = ni for each i ∈ [k ] . (252)

Conversely, if β ∈ [k ]n is an n-tuple that satisfies (252), then β is an anagram

of α. Indeed, we can sort the entries of β in increasing order, thus obtaining

an n-tuple of the form  1, 1, . . . , 1 , 2, 2, . . . , 2 , . . . , k, k, . . . , k ; but the equalities

| {z } | {z } | {z }
some # of 1’s some # of 2’s some # of k’s
(252) show that the # of 1’s in this n-tuple will be n1 , and the # of 2’s will be
n
2 , and the # of 3’s will be n3 , and
 so on, so that this n-tuple will be precisely
1, 1, . . . , 1, 2, 2, . . . , 2, . . . , k, k, . . . , k  = α. This shows that α can be obtained by
| {z } | {z } | {z }
n1 times n2 times nk times
sorting β, and thus β is an anagram of α.
Combining our observations, we conclude that the anagrams of α are precisely
the n-tuples β ∈ [k ]n that satisfy (252). Renaming β as (u1 , u2 , . . . , un ), we can
rewrite this as follows: The anagrams of α are precisely the n-tuples (u1 , u2 , . . . , un ) ∈
[k]n that satisfy

(# of times i appears in (u1 , u2 , . . . , un )) = ni for each i ∈ [k ] .

In other words, the anagrams of α are precisely the n-tuples (u1 , u2 , . . . , un ) ∈ [k ]n

satisfying

(# of a ∈ [n] such that u a = i ) = ni for each i ∈ [k ] .

 
n
But we know (from Proposition 2.12.7) that the # of such n-tuples is .
 n 1 , n 2 , .. . , n k
n
Hence, we conclude that the # of distinct anagrams of α is . This
n1 , n2 , . . . , n k
proves Proposition 2.12.10.
To formalize the above proof of Proposition 2.12.10, we need to fill two gaps. The
first is a rigorous justification of the equality (251) (assuming that β is an anagram
of α). This can easily be done on a more general level:
Enumerative Combinatorics: class notes page 315

Exercise 2.12.3. Let X be any set. Let n ∈ N. Let α ∈ X n be an n-tuple. Let β be

an anagram of α. Let i ∈ X. Prove that

(# of times i appears in β) = (# of times i appears in α) .

Applying Exercise 2.12.3 to X = [k], we immediately obtain (251).

But there is a second gap in the above informal proof of Proposition 2.12.10 (at
least from a formal point of view); indeed, we used the notion of “sorting” an
n-tuple (when we said that “we can sort the entries of β in increasing order”).
Behind this notion stands the fundamental but nontrivial fact that any n-tuple of
integers (or, more generally, of real numbers) can be sorted into weakly increasing
order by permuting its entries. This fact is not hard to prove (see, e.g., [Grinbe15,
Proposition 6.40 (a)] and its proof in [Grinbe15, solution to Exercise 6.13]).
There is also a second way of closing this second gap. Indeed, we can avoid the
notion of “sorting” altogether, and instead show the following fact (a converse to
Exercise 2.12.3):

Exercise 2.12.4. Let X be any set. Let n ∈ N. Let α ∈ X n and β ∈ X n be two

n-tuples. Assume that

(# of times i appears in β) = (# of times i appears in α) (253)

for each i ∈ X. Prove that β is an anagram of α.

Example 2.12.12. Let X = Z and n = 5 and α = (1, 4, 1, 2, 0) and β = (2, 0, 1, 1, 4).

Then, each integer appears as often in the 5-tuple α as it appears in the 5-tuple
β. In other words, (253) holds for each i ∈ X. Hence, Exercise 2.12.4 shows that
β is an anagram of α. And indeed, if we write α as(α1 , α2 , α3 , α4 , α5 ), then we can

easily find a permutation σ of [5] such that β = α σ (1) , α σ (2) , α σ (3) , α σ (4) , α σ (5) .
 
1 2 3 4 5
(For example, the permutation given in two-line notation as
4 5 1 3 2
will do.)

Once Exercise 2.12.4 is solved, we no longer have need for sorting. Instead, we
can apply Exercise 2.12.4 to X = [k], and thus conclude that if β ∈ [k]n is an n-tuple
that satisfies (252), then β is an anagram of α. So the second gap in the above proof
of Proposition 2.12.10 is filled again.
Note that Exercise 2.12.4 is a weaker version of [19s, Lemma 2.13.20].

2.12.4. More formulas

We shall now state a few more basic properties of multinomial coefficients.
First, we observe that the multinomial coefficients generalize the binomial coef-
ficients that you see in Pascal’s triangle:
Enumerative Combinatorics: class notes page 316

Proposition 2.12.13. Let n ∈ N and k ∈ {0, 1, . . . , n}. Then,

   
n n
= .
k k, n − k
| {z } | {z }
binomial coefficient multinomial coefficient

Proof of Proposition 2.12.13. From k ∈ {0, 1, . . . , n}, we obtain k ≤ n and thus n − k ∈

N. Also, k ∈ {0, 1, . . . , n} ⊆ N.  So we have k, n − k ∈ N and k + (n − k) = n.
n
Thus, the multinomial coefficient is well-defined. Moreover, its defini-
k, n − k
tion yields  
n n! n!
= = .
k, n − k k! (n − k )! k! · (n − k )!
Comparing this with
 
n n!
= (by Theorem 1.3.9) ,
k k! · (n − k )!
   
n n
we obtain = . This proves Proposition 2.12.13.
k k, n − k
Let us prove some further properties of multinomial coefficients.
Proposition 2.12.14. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 +
· · · + nk = n. Let σ be a permutation of [k]. Then,
   
n n
= .
n1 , n2 , . . . , n k n σ (1) , n σ (2) , . . . , n σ ( k )

Proof of Proposition 2.12.14. The map σ is a permutation of [k ], that is, a bijection

from [k] to [k ]. Hence, the product nσ(1) !nσ(2) ! · · · nσ(k) ! has the same factors as
the product n1 !n2 ! · · · nk ! (just in a different order). Thus, nσ(1) !nσ(2) ! · · · nσ(k) ! =
n1 !n2 ! · · · nk !. For similar reasons, we have nσ(1) + nσ(2) + · · · + nσ(k) = n1 + n2 +
· · · + nk . Thus, nσ(1) + nσ(2) + · · · + nσ(k) = n1 + n2 + · · · + nk = n, so that the
 
n
multinomial coefficient is well-defined.
n σ (1) , n σ (2) , . . . , n σ ( k )
 
n
But the definition of the multinomial coefficient yields
n σ (1) , n σ (2) , . . . , n σ ( k )
 
n n! n!
= =
n σ (1) , n σ (2) , . . . , n σ ( k ) nσ(1) !nσ(2) ! · · · nσ(k) ! n1 !n2 ! · · · nk !
 
since nσ(1) !nσ(2) ! · · · nσ(k) ! = n1 !n2 ! · · · nk !
 
n
=
n1 , n2 , . . . , n k
Enumerative Combinatorics: class notes page 317

 
n n!
(since the multinomial coefficient is defined to be ).
n1 , n2 , . . . , n k n1 !n2 ! · · · nk !
This proves Proposition 2.12.14.

Theorem 2.12.15 (Recurrence of the multinomial coefficients). Let n ∈ N and

n1 , n2 , . . . , nk ∈ N be such that n1 + n2 + · · · + nk = n > 0. Then,
k
n−1
   
n
=∑ .
n1 , n2 , . . . , n k n , . . . , ni−1 , ni − 1, ni+1 , . . . , nk
i =1 | 1 {z }
This should be interpreted as 0 if ni =0

Proof. LTTR.
Just as we tabulated the binomial  coefficients  in Pascal’s triangle, we can tab-
n
ulate the multinomial coefficients for a given value of k in a k-
n1 , n2 , . . . , n k
dimensional pyramid-like table, called “Pascal’s pyramid” (see [Uecker17, p. 20,
Figure 10] fora picture). Theorem 2.12.15 shows that each entry in this table (ex-
0
cept for the = 1 at its apex) is the sum of its k upper neighbors
0, 0, . . . , 0
(generalizing the analogous property of Pascal’s triangle).
The following theorem generalizes the binomial formula (because of Proposition
2.12.13):

Theorem 2.12.16 (the multinomial formula). Let x1 , x2 , . . . , xk be k numbers. Let

n ∈ N. Then,
 
n
∑
n
n
( x1 + x2 + · · · + x k ) = x1n1 x2n2 · · · xk k .
n ,
1 2n , . . . , n k
k (n1 ,n2 ,...,nk )∈N ;
n1 +n2 +···+nk =n

Proof. LTTR. There are many ways to prove this: Either do it combinatorially using
Proposition 2.12.5 (see [Galvin17, Theorem 11.7] for this proof); or use induction
on n; or use induction on k (see [Grinbe15, Exercise 6.2] for this proof223 ).

3. The twelvefold way

Let us look back at the counting problems we have so far been solving. We have
counted subsets, tuples, maps, integers, tilings and more complicated combinations
 
223 Keep n
in mind that what we call is denoted by m (n1 , n2 , . . . , nk ) in [Grinbe15,
n1 , n2 , . . . , n k
Exercise 6.2].
Enumerative Combinatorics: class notes page 318

of these (such as tuples of subsets). While these problems were often interconnected
(in particular, many of them were solved using the same methods, and bijections let
us translate between differently looking objects), the problems themselves followed
no system other than “what else can we count using the methods seen so far?”.
Historically, this has led to combinatorics being regarded more as a hodgepodge of
problems and tricks rather than as a systematic theory.
To some extent, this is still the case, and it is not just an artifact of viewpoint;
combinatorics really is a broad field with much less of a clear direction than many
other mathematical subjects.224 Nevertheless, there is more structure in combina-
torics than meets the eye, and there have been several successful attempts to classify
and organize the most common enumerative problems in a coherent theory. One
of the most successful attempts of this kind is the twelvefold way, introduced in
the 1960s by Gian-Carlo Rota under the name of “theory of distribution and occu-
pancy” (or, less grandiloquently, “placing balls in boxes”) and since popularized as
the “twelvefold way”.

3.1. What is the twelvefold way?

The twelvefold way is a collection of 12 standard counting problems (usually ar-
ranged as a table with 4 rows and 3 columns) that frequently appear in combina-
torics (in the sense that many other counting problems can be reduced to them).
Let us first introduce it informally, and then define it formally and study it in detail.

Convention 3.1.1. For the rest of Chapter 3, we fix two finite sets A and X. We
shall refer to the elements of A as balls, and we shall refer to the elements of X
as boxes. (The letters A and X have been chosen because “a” appears in “ball”
while “x” appears in “box”. I cannot use “B” for both sets...)

A placement shall mean a way to distribute all the balls a ∈ A into the boxes
x ∈ X. Rigorously, we can think of a placement as a map from A to X. More
precisely, this is what we will call an “L → L placement”; other kinds of placements
will be defined below.
How many L → L placements are there? Of course, there are | X || A| many (by
Theorem 2.4.1). But let us look at them more closely.

Example 3.1.2. Let X = [2] and A = [3], so that | X | = 2 and | A| = 3. Thus, we

are trying to place 3 balls (called 1, 2, 3) into 2 boxes (called 1, 2).

224 This is not entirely a bad thing:

It means that an average result in combinatorics relies less on prior
knowledge and mastery of complex theory than in most other pure-mathematical disciplines.
Already early on in your journey, you have more directions to “branch out” into, and more
results you can achieve without years of training but merely with a good dose of creativity and
perseverance. But on the flip side, it means that said results will be less likely to be foundational
to others’ work after you. (And this does not spell significance for texts like this...)
Enumerative Combinatorics: class notes page 319

We will draw such placements as follows: We will always draw the boxes in
increasing order:

| {z } | {z }
box 1 box 2

We draw each ball as a pair of parentheses with a number inside – so, for exam-
ple, the ball 3 will be drawn as “(3)”.
So the L → L placements of the balls 1, 2, 3 ∈ A in the boxes 1, 2 ∈ X are

(1)(2)(3) , (1)(2) (3) ,

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

(1)(3) (2) , (2)(3) (1) ,

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

(3) (1)(2) , (2) (1)(3) ,

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

(1) (2)(3) , (1)(2)(3) .

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

The order of balls within a single box does not matter:

(3) (1)(2) = (3) (2)(1) .

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

As we said, counting these L → L placements was easy. But we can consider the
following variations of the problem:
• What if we require our maps f : A → X to be injective (i.e., each box contains
≤ 1 ball) or surjective (i.e., each box contains ≥ 1 ball)?
• What if the balls are unlabelled (i.e., indistinguishable)?
For example, the two L → L placements

(1)(2) (3) and (2)(3) (1)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2
Enumerative Combinatorics: class notes page 320

were counted as different, because the former has ball 1 in box 1 whereas
the latter has ball 1 in box 2. But if the balls are indistinguishable, then this
difference disappears, and these placements become identical, thus reducing
the count.
Intuitively, it may be clear what “indistinguishable” means, but it is a bit
tricky to make this mathematically rigorous. After all, if we define place-
ments as maps, then we cannot just pretend that two different maps are equal
because we “chose to think of the balls as indistinguishable”. (There is no
notion of a “set with indistinguishable elements” in mathematics.225 )
We will make this rigorous using the notion of equivalence classes (which we
will introduce in Section 3.3). These are sets of mutually equivalent objects
(in our case, maps f : A → X), where the word “equivalent” is understood
with respect to a specific relation. Equivalent objects will contribute to the
same equivalence class; thus, if we want to “count objects up to equivalence”,
all we will have to do is counting their equivalence classes. This is a valid
and (once you have gotten used to it) fairly convenient rigorous alternative to
“pretending” that equivalent objects are “the same”.
Here is how this general idea will be applied to our question about indistin-
guishable balls: We will say that two maps f , g : A → X are “ball-equivalent”
if there is a permutation τ of the balls (i.e., formally: a permutation τ of the
set A) that transforms one into the other (i.e., that satisfies f = g ◦ τ). Then,
“U → L placements” (i.e., placements of unlabelled balls in labelled boxes)
will be defined as the equivalence classes of ball-equivalence.
For example, the two L → L placements

(1)(2) (3) and (2)(3) (1)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are ball-equivalent, and therefore contribute to only 1 ball-equivalence class –

i.e., they count as one and the same U → L placement.
We shall draw U → L placements in the same way as we draw L → L place-
ments, but representing balls by “•” symbols (rather than by numbers in
parentheses). Thus, the U → L placement that we have just discussed be-
comes
•• • .
| {z } | {z }
box 1 box 2

This shall visualize the fact that balls are indistinguishable.

225 Actually,
there is: the notion of a multiset, as defined in Section 2.11. However, it would only help
us if we defined maps between multisets (as well as maps between multisets and sets), which
we are not going to do (as it comes with its own difficulties).
Enumerative Combinatorics: class notes page 321

• What if the boxes are unlabelled (i.e., indistinguishable)?

This can be made rigorous in a similar way, but using permutations of boxes
instead of permutations of balls. This gives “L → U placements” (i.e., place-
ments of labelled balls into unlabelled boxes). For example, the two L → L
placements

(1)(3) (2) and (2) (1)(3)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are “box-equivalent” (i.e., can be transformed into one another by permuting

the boxes) and thus contribute to only 1 box-equivalence class – i.e., they
count as one and the same L → U placement.
We shall draw L → U placements in the same way as we draw L → L place-
ments, but without marking the boxes as “box 1”, “box 2” etc.. Thus, the
L → U placement that we have just discussed becomes

(1)(3) (2) .

• What if both the boxes and the balls are indistinguishable? This kind of place-
ments will be called “U → U placements”. We shall draw them in the same
way as we draw L → L placements, but representing balls by “•” symbols
and without marking the boxes.

We can combine these variations, by

• requiring f either to be arbitrary (i.e., not requiring anything about f ), or to

be injective, or to be surjective;

• counting L → L placements or U → L placements or L → U placements or

L → L placements.

In total, we thus get 3 · 4 = 12 many different counting problems. We list them in

a table:
arbitrary injective surjective

L→L | X || A|
U→L
L→U
U→U
Here:
Enumerative Combinatorics: class notes page 322

• The columns specify whether we are requiring f to be arbitrary or injective

or surjective.
• The rows specify what kind of placements we are counting:
– “L → L” means “balls are labelled, boxes are labelled”, so we are just
counting maps f : A → X.
– “U → L” means “balls are unlabelled, boxes are labelled”.
– “L → U” means “balls are labelled, boxes are unlabelled”.
– “U → U” means “balls are unlabelled, boxes are unlabelled”.

Example 3.1.3. Let X = [2] and A = [3]. Then, let us count how many placements
of each kind we have:
arbitrary injective surjective
L→L 8 0 6
U→L 4 0 2
L→U 4 0 3
U→U 2 0 1

In fact:
• The U → L placements are

••• , •• • ,
| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

• •• , ••• .
| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

• The L → U placements are

(1)(2)(3) , (1)(2) (3) ,

(1)(3) (2) , (2)(3) (1) .

• The U → U placements are

••• , •• • .
Enumerative Combinatorics: class notes page 323

In general, not every of the 12 questions in the table above has a closed-form
answer. But each has, at least, a good recursive answer, and there is much to be
learned from attempting to answer all the questions. We shall thus strive to fill in
the whole “twelvefold way” table, row by row.

3.2. L → L
We begin with the first row of the twelvefold way: the counting of L → L place-
ments. Let us repeat how we defined them above:

Definition 3.2.1. An L → L placement from A to X is just a map from A to X.

The value of this map at some a ∈ A is called the box in which ball a is placed.

We can now count the L → L placements:

Proposition 3.2.2. We have

(# of L → L placements A → X )
= (# of maps from A to X ) = | X || A| .

Proof. The first equality sign follows from Definition 3.2.1. The second equality
sign follows from Theorem 2.4.1 (applied to m = | A|, n = | X | and B = X).

Proposition 3.2.3. We have

(# of injective L → L placements A → X )
= (# of injective maps from A to X ) = | X || A| .

Proof. The first equality sign follows from Definition 3.2.1. The second equality
sign follows from Theorem 2.4.4 (applied to m = | A|, n = | X | and B = X).

Proposition 3.2.4. We have

(# of surjective L → L placements A → X )
= (# of surjective maps from A to X ) = sur (| A| , | X |) .

Proof. The first equality sign follows from Definition 3.2.1. The second equality
sign follows from Theorem 2.4.11 (applied to m = | A|, n = | X | and B = X).

Example 3.2.5. Here are some typical counting problems that boil down to count-
ing L → L placements:
Enumerative Combinatorics: class notes page 324

• How many ways are there to assign grades (from a finite set X) to students
(from a finite set A) ? These assignments are arbitrary L → L placements.

• How many ways are there to assign IP addresses to a given set of com-
puters? These assignments are injective L → L placements, since IP
addresses should be distinct and come from a well-specified set (say,
{32-bit numbers} for IPv4 addresses).
• How many 8-digit telephone numbers are there with no 2 equal digits?
Of course, these telephone numbers are precisely the injective 8-tuples
(i1 , i2 , . . . , i8 ) ∈ {0, 1, . . . , 9}8inj (see Definition 2.7.1 for the meaning of these
words), and thus Proposition 2.7.2 shows that there are 108 many of them.
Alternatively, let us restate this counting problem in terms of L → L place-
ments: The 8-digit telephone numbers with no 2 equal digits correspond to
injective L → L placements with A = [8] and X = {0, 1, . . . , 9}; for exam-
ple, the telephone number 20354986 corresponds to the L → L placement

(2) (1) (3) (5) (4) (8) (7) (6) .

|{z} |{z} |{z} |{z} |{z} |{z} |{z} |{z} |{z} |{z}
box 0 box 1 box 2 box 3 box 4 box 5 box 6 box 7 box 8 box 9

Thus, the total # of such telephone numbers is 108 = 10 · 9 · · · · · 3 = 10!/2!.

3.3. Equivalence relations

Now, we take a break from counting questions to build some theory that will help
us rigorously formulate the other three rows of the twelvefold way (i.e., the notions
of U → L placements, L → U placements, and U → U placements). These notions
depend on having “unlabelled balls” and/or “unlabelled boxes”. What does it
mean for balls, or boxes, to be unlabelled?
Rigorously, it means that we are counting not the maps f : A → X themselves,
but rather their equivalence classes with respect to some relation.
We are going to explain what these are. First, let us introduce (or recall, de-
pending on your background) the notions of relations and, in particular, equivalence
relations.

Class of 2019-11-08

3.3.1. Relations
I will follow [19s, Chapter 3] in this subsection.

Unpolished stuff starts here

Enumerative Combinatorics: class notes page 325

Definition 3.3.1. Let S be a set. A (binary) relation on S is formally defined as

a subset of S × S. Informally, it is a statement x R y defined for every pair of
elements ( x, y) ∈ S × S. For each pair ( x, y) ∈ S × S, this statement x R y is
either true or false.

The informal and the formal definitions are equivalent, since:

• Given a statement x R y defined for every pair ( x, y) ∈ S × S, we can encode

it as a subset of S × S, namely as the subset

{( x, y) ∈ S × S | x R y} .

• Conversely, every subset T of S × S can be decoded into a statement (namely,

“( x, y) ∈ T”).

In the following, we will say “relation” for “binary relation”.

Example 3.3.2. Let S = Z.

(a) The relation = is a binary relation on S. As a subset of S × S, this relation
is

{( a, b) ∈ S × S | a = b}
= {(c, c) | c ∈ S} = {. . . , (−2, −2) , (−1, −1) , (0, 0) , (1, 1) , (2, 2) , . . .} .

(b) The relation < is a binary relation on S. As a subset of S × S, this relation

is
{( a, b) ∈ S × S | a < b} .
(c) The relation ≤ is a binary relation on S.
(d) The relation 6= is a binary relation on S.
(e) Fix n ∈ Z. Define a binary relation ≡ on S = Z by
n
 
a ≡ b ⇐⇒ ( a ≡ b mod n) ⇐⇒ (n | a − b)
n
⇐⇒ ( a = b + kn for some k ∈ Z) .

This relation ≡ is called congruence modulo n. Note that the relation ≡ is =.

n 0
When n is positive, we have a ≡ b mod n if and only if a%n = b%n.
For n = 1, we have

a−b
   
a ≡ b ⇐⇒ (1 | a − b) ⇐⇒ ∈ Z ⇐⇒ ( a − b ∈ Z) ⇐⇒ (true) .
1 1

Thus, the relation ≡ always holds (i.e., we have a ≡ b for all ( a, b) ∈ S × S).
1 1
Thus, it equals the whole set S × S (when considered as a subset of S × S).
Enumerative Combinatorics: class notes page 326

(f) Define a binary relation N on S by

( a N b) ⇐⇒ (false) .

That is, there is no pair ( a, b) ∈ S × S such that a N b. As a subset of S × S, this

relation is ∅.
(g) Define a binary relation A on S by

( a A b) ⇐⇒ (true) .

That is, every pair ( a, b) ∈ S × S satisfies a A b. As a subset of S × S, this relation

is S × S. This is the above relation ≡, but the way we just defined it, it makes
1
sense for every set S (not just for Z).
(h) The relation | is a binary relation on S.
(i) The relation “is coprime to” is a binary relation on S.

There are other examples on other sets. For example, “parallel” is a relation
between lines in plane geometry.

3.3.2. Equivalence relations

Definition 3.3.3. Let R be a binary relation on a set S.

(a) We say that R is reflexive if every a ∈ S satisfies a R a.
(b) We say that R is symmetric if every a, b ∈ S satisfying a R b satisfy b R a.
(c) We say that R is transitive if every a, b, c ∈ S satisfying a R b and b R c
satisfy a R c.

Example 3.3.4. Let S be the set Z.

(a) The relation = is reflexive, symmetric and transitive.
(b) The relation < is not reflexive, not symmetric but transitive.
(c) The relation ≤ is reflexive, not symmetric but transitive.
(d) The relation 6= is not reflexive, but symmetric. It is not transitive (e.g.,
2 6= 3 and 3 6= 2, but 2 = 2).
(e) For every n ∈ Z, the congruence relation ≡ is reflexive (since n | a − a),
n
symmetric (since n | a − b implies n | b − a) and transitive (since n | a − b and
n | b − c implies n | a − c).
(f) The relation N is symmetric (vacuously) and transitive (vacuously), but not
reflexive.
(g) The relation A is reflexive, symmetric and transitive.
(h) The divisibility relation | is reflexive but not symmetric. It is transitive (if
a | b and b | c, then a | c).
(i) The coprimality relation is symmetric but neither reflexive nor transitive.
Enumerative Combinatorics: class notes page 327

Definition 3.3.5. An equivalence relation on a set S means a relation on S that

is reflexive, symmetric and transitive.

Example 3.3.6. Let S be any set. The relation = on S is an equivalence relation.

Example 3.3.7. Let n ∈ Z. Then, the congruence relation ≡ on Z is an equiva-

n
lence relation.

Example 3.3.8. The relation “parallel” on {lines in the plane} is an equivalence

relation.

Example 3.3.9. The relation “similar” on {triangles in the plane} is an equiva-

lence relation.

Example 3.3.10. The relation “directly similar” (= similar with the same orienta-
tion) on {triangles in the plane} is an equivalence relation.

Example 3.3.11. The relation “indirectly similar” (= similar with opposite ori-
entation) on {triangles in the plane} is not an equivalence relation, because it is
not reflexive.

Example 3.3.12. Let S and T be two sets. Let f : S → T be a map. Define a

relation ≡ on S by
f
 
a ≡ b ⇐⇒ ( f ( a) = f (b)) .
f

This relation ≡ is an equivalence relation.

f

Example 3.3.13. Let S be {all points on the landmass of the Earth}, and define a
relation ∼ by

( a ∼ b) ⇐⇒ (there is a land route from a to b) .

Then, ∼ is an equivalence relation.

Example 3.3.14. Let

n o
S = Z × (Z \ {0}) = ( a1 , a2 ) ∈ Z | a1 ∈ Z and a2 ∈ Z \ {0} .
2

Define a relation ∼ on S by
∗
 
( a1 , a2 ) ∼ (b1 , b2 ) ⇐⇒ ( a1 b2 = a2 b1 ) .
∗

This relation ∼ is an equivalence relation.

∗
Enumerative Combinatorics: class notes page 328

3.3.3. Equivalence classes

Definition 3.3.15. Let ∼ be an equivalence relation on a set S.

(a) For each a ∈ S, we define a subset [ a]∼ of S by

[ a]∼ = {b ∈ S | b ∼ a} .

This subset [ a]∼ is called the equivalence class of a (for the relation ∼), or the
∼-equivalence class of a.
(b) The equivalence classes of ∼ are defined to be the sets [ a]∼ for a ∈ S. They
are also known as the ∼-equivalence classes.

Example 3.3.16. Consider the relation ≡ on Z. We have

3
 
[5] ≡ = b∈Z | b≡5 = {b ∈ Z | b ≡ 5 mod 3}
3 3
= {5 + 3k | k ∈ Z} = {. . . , −4, −1, 2, 5, 8, 11, 14, . . .}

and
 
[3] ≡ = b∈Z | b≡3 = {b ∈ Z | b ≡ 3 mod 3}
3 3
= {3 + 3k | k ∈ Z} = {. . . , −6, −3, 0, 3, 6, 9, . . .}

and
 
[2] ≡ = b∈Z | b≡2 = {b ∈ Z | b ≡ 2 mod 3}
3 3
= {2 + 3k | k ∈ Z} = {. . . , −4, −1, 2, 5, 8, 11, 14, . . .} .

Note that [2]≡ = [5]≡ .

3 3

Let us state some basic properties of equivalence classes. For their proofs, see [19s,
§3.3.2] or any sufficiently thorough introduction to proofs.

Proposition 3.3.17. Let ∼ be an equivalence relation on a set S. Let a ∈ S. Then,

[ a]∼ = {b ∈ S | a ∼ b} .

Proposition 3.3.18. Let ∼ be an equivalence relation on a set S. Let a ∈ S. Then,

a ∈ [ a]∼ .
Enumerative Combinatorics: class notes page 329

Proposition 3.3.19. Let ∼ be an equivalence relation on a set S. Let x, y ∈ S.

(a) If x ∼ y, then [ x ]∼ = [y]∼ .
(b) If not x ∼ y, then [ x ]∼ and [y]∼ are disjoint.
(c) We have x ∼ y if and only if x ∈ [y]∼ .
(d) We have x ∼ y if and only if y ∈ [ x ]∼ .
(e) We have x ∼ y if and only if [ x ]∼ = [y]∼ .

As a consequence, any two ∼-equivalence classes are either identical or disjoint.

Example 3.3.20. Consider the set

n o
S = Z × (Z \ {0}) = ( a1 , a2 ) ∈ Z2 | a1 ∈ Z and a2 ∈ Z \ {0} .

Consider the relation ∼ defined on the set S by

∗
 
( a1 , a2 ) ∼ (b1 , b2 ) ⇐⇒ ( a1 b2 = a2 b1 ) .
∗

Then, ∼ is an equivalence relation, and its equivalence classes are the rational
∗
numbers. This is how the rational numbers are rigorously defined! So the re-
lation ∼ tells us which pairs of integers give the same rational number. The
∗
a
equivalence class [( a1 , a2 )]∼ is denoted by 1 .
∗ a2

Example 3.3.21. Let S be {all points on the landmass of the Earth}, and consider
the relation ∼ defined by

( a ∼ b) ⇐⇒ (there is a land route from a to b) .

Then, the ∼-equivalence classes are the continents and islands.

Example 3.3.22. Let A be a set, and k ∈ N. The relation ∼ on Ak is defined as

perm
follows:
 
p ∼ q ⇐⇒ (p is an anagram of q)
perm

⇐⇒ (p can be obtained from q by permuting the entries) .

For example, (3, 8, 8, 2) ∼ (8, 3, 2, 8).

perm
The equivalence classes of ∼ are called unordered k-tuples of elements of
perm
A. They are in bijection with size-k multisubsets of A.
Enumerative Combinatorics: class notes page 330

Example 3.3.23. Let n ∈ Z. The equivalence classes of the relation ≡ are called
n
“integers modulo n”, and their set is denoted by Z/nZ or Zn or Z/n or Zn or
Cn (depending on author and context). In particular, the sets Z/12Z and Z/24Z
have well-known meanings: they stand for the hours on a clock (a 12-hour clock
or a 24-hour clock, respectively). For example, when you say “3 hours after 11 is
2”, you are really saying that 11 + 3 ≡ 2 mod 12 or, equivalently, [11]≡ + [3]≡ =
12 12
[2]≡ for an appropriate definition of addition of equivalence classes.
12

Next, we will define box-equivalence and ball-equivalence on placements of balls

into boxes. Then, by taking equivalence classes, we will be able to make sense of
“unlabelled balls” and “unlabelled boxes”.
Class of 2019-11-11

3.3.4. Defining unlabelled boxes and balls

What does it mean for balls, or boxes, to be unlabelled (in the twelvefold way)?
Rigorously, it means that we are counting
not the maps f : A → X,
but their equivalence classes with respect to certain equivalence relations.
What relations?
Definition 3.3.24. Let f , g : A → X be two maps. Then, we say that

box
• f is box-equivalent to g (written f ∼ g) if and only if there is a permuta-
tion σ of X such that f = σ ◦ g (in other words, f can be obtained from g
by permuting boxes).
ball
• f is ball-equivalent to g (written f ∼ g) if and only if there is a permuta-
tion τ of A such that f = g ◦ τ (in other words, f can be obtained from g
by permuting balls).
box
• f is box-ball-equivalent to g (written f ∼ g) if and only if there is a
ball
permutation σ of X and a permutation τ of A such that f = σ ◦ g ◦ τ.

Example 3.3.25. Let A = [3] and X = [2].

(a) The two L → L placements

(1) (2)(3) and (2)(3) (1)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are box-equivalent to one another (since they can be obtained from one another
by swapping the two boxes), but they are not ball-equivalent to one another
Enumerative Combinatorics: class notes page 331

(since a permutation of balls cannot change the fact that the first placement has
1 ball in the first box, whereas the second placement has 2).
(b) The two L → L placements

(1) (2)(3) and (2) (1)(3)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are ball-equivalent to one another (since they can be obtained from one another
by swapping the balls 1 and 2), but they are not box-equivalent to one another
(since a permutation of boxes cannot change the fact that balls 2 and 3 are to-
gether in a single box in the first placement, but not in the second).
(c) The two L → L placements

(2)(3) (1) and (2) (1)(3)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are box-ball-equivalent to one another (since they can be obtained from one an-
other by swapping the two boxes and swapping balls 1 and 2). But they are
neither box-equivalent nor ball-equivalent.
(d) The two L → L placements

(1)(2)(3) and (2)(3) (1)

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

are not box-ball-equivalent to one another (since no permutation of boxes or

of balls can change the fact that the first placement has an empty box, but the
second does not).

box ball box

Proposition 3.3.26. All of ∼ , ∼ and ∼ are equivalence relations on the set
ball
{maps A → X }.

Proof. Easy. Hint:

• If σ and τ are two permutations of a set S, then σ ◦ τ is a permutation of S.

• If σ is a permutation of a set S, then σ−1 is a permutation of S.

Definition 3.3.27. Now, we define:

ball
• U → L placements to be the ∼ -equivalence classes.
Enumerative Combinatorics: class notes page 332

box
• L → U placements to be the ∼ -equivalence classes.
box
• U → U placements to be the ∼ -equivalence classes.
ball

Example 3.3.28. Consider again our running example, with X = [2] and A = [3].
Let g be the L → L placement

(1) (2)(3) .
| {z } | {z }
box 1 box 2

box
(a) The ∼ -equivalence class of g is
 

 

 
(1) (2)(3) , (2)(3) (1) .
 
 | {z } | {z }
 | {z } | {z } 

box 1 box 2 box 1 box 2

ball
(b) The ∼ -equivalence class of g is




(1) (2)(3) , (2) (1)(3) ,

 | {z } | {z } | {z } | {z }

box 1 box 2 box 1 box 2




(3) (1)(2) .
| {z } | {z } 


box 1 box 2

box
(c) The ∼ -equivalence class of g is
ball




(1)(2) (3) , (1)(3) (2) ,

 | {z } | {z } | {z } | {z }

box 1 box 2 box 1 box 2

(2)(3) (1) , (3) (1)(2) ,

| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2




(2) (1)(3) , (1) (2)(3) .
| {z } | {z } | {z } | {z } 


box 1 box 2 box 1 box 2
Enumerative Combinatorics: class notes page 333

Recall the following crucial general fact about equivalence classes:

Proposition 3.3.29. Let ∼ be an equivalence relation on a set S. Let x ∈ S and

y ∈ S. Then, x ∼ y if and only if the ∼-equivalence classes of x and y are
identical (that is, [ x ]∼ = [y]∼ ).

Thus, “counting elements of S up to equivalence” means counting ∼-equivalence

classes. Thus,
ball
• by counting ∼ -equivalence classes, we will be counting placements of balls
in boxes where the balls are unlabelled (but the boxes are labelled);
box box
• similarly for ∼ and ∼ .
ball

3.4. U → L
ball
Recall: The U → L placements are the ∼ -equivalence classes.
Enumerative Combinatorics: class notes page 334

Example 3.4.1. For X = [2] and A = [3], there are 4 different U → L placements:
 

 

 
(1)(2)(3) ,

 | {z } | {z } 

 
box 1 box 2





(1)(2) (3) , (1)(3) (2) ,

 | {z } | {z } | {z } | {z }

box 1 box 2 box 1 box 2




(2)(3) (1) ,
| {z } | {z } 


box 1 box 2





(1) (2)(3) , (2) (1)(3) ,

 | {z } | {z } | {z } | {z }

box 1 box 2 box 1 box 2




(3) (1)(2) ,
| {z } | {z } 


box 1 box 2

 

 

 
(1)(2)(3) .

 | {z } | {z } 

 
box 1 box 2

Notation: When visualizing a U → L placement, we will just draw the balls as

circles, without putting any numbers in them. So the four U → L placements for
X = [2] and A = [3] are (shown here in the same order as in the example above)

••• , •• • ,
| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2

• •• , ••• .
| {z } | {z } | {z } | {z }
box 1 box 2 box 1 box 2
Enumerative Combinatorics: class notes page 335

Now, we are ready to count the U → L placements in the general setting:

Proposition 3.4.2. We have

(# of U → L placements A → X )
   
|X|
= # of x1 , x2 , . . . , x|X | ∈ N satisfying x1 + x2 + · · · + x|X | = | A|
| A| + | X | − 1
 
= .
| A|

Proof. The second equality sign follows from (243).

It remains to prove the first equality sign. First, we WLOG assume that X = [| X |]
(so the boxes are labelled 1, 2, . . . , | X |). Then, consider the bijection
n  o
{U → L placements A → X } → x 1 , x 2 , . . . , x | X | ∈ N| X | | x 1 + x 2 + · · · + x | X | = | A |

that sends each U → L placement a1 balls a2 balls · · · a|X | balls to the

 
ball
| X |-tuple a1 , a2 , . . . , a|X | . (More rigorously: The ∼ -equivalence class of a map
f : A → X is sent to the | X |-tuple
 
f −1 (1) , f −1 (2) , . . . , f −1 (| X |) ,

where
f −1 ( x ) : = { a ∈ A | f ( a ) = x } .
To see that this is a bijection, we need to show that if f , g : A → X are two maps
such that
f −1 ( x ) = g −1 ( x ) for all x ∈ X,
ball
then f ∼ g.)
Thus, the bijection principle yields

(# of U → L placements A → X )
   
|X|
= # of x1 , x2 , . . . , x|X | ∈ N satisfying x1 + x2 + · · · + x|X | = | A| .

This is exactly the first equality sign of Proposition 3.4.2.

Recall: An L → L placement is surjective if and only if each box has at least one
ball in it.
ball
We define surjectivity of U → L placements in the same way. Thus, a ∼ -
equivalence class is a surjective U → L placement if and only if all its elements
are surjective maps.
Enumerative Combinatorics: class notes page 336

Proposition 3.4.3. We have

(# of surjective U → L placements A → X )
   
|X|
= # of x1 , x2 , . . . , x|X | ∈ P satisfying x1 + x2 + · · · + x|X | = | A|
| A| − 1
 
= .
| A| − | X |

Proof. The first equality sign is proved similarly to the first equality sign in Propo-
sition 3.4.2.
The second equality sign follows from (239).
Recall: An L → L placement is injective if and only if each box has at most one
ball in it.
ball
We define injectivity of U → L placements in the same way. Thus, a ∼ -
equivalence class is an injective U → L placement if and only if all its elements
are injective maps.

Proposition 3.4.4. We have

(# of injective U → L placements A → X )
   
|X|
= # of x1 , x2 , . . . , x|X | ∈ {0, 1} satisfying x1 + x2 + · · · + x|X | = | A|
 
|X|
= .
| A|

Proof. The first equality sign is proved similarly to the first equality sign in Propo-
sition 3.4.2.
The second equality sign follows from Theorem 2.10.4.
Thus, our “twelvefold way” table now looks as follows:

arbitrary injective surjective

L→L | X || A| | X || A| sur (| X | , | A|)

| A| + | X | − 1 | A| − 1
     
|X| .
U→L
| A| | A| | A| − | X |
L→U
U→U
Enumerative Combinatorics: class notes page 337

3.5. L → U
box
Recall: The L → U placements are the ∼ -equivalence classes. They are placements
of labelled balls in unlabelled boxes.

Example 3.5.1. For X = [2] and A = [3], there are 4 different L → U placements:
 

 

 
(1)(2)(3) , (1)(2)(3) ,
 | {z } | {z } 
 | {z } | {z }
 

box 1 box 2 box 1 box 2

 

 

 
(1)(2) (3) , (3) (1)(2) ,

 | {z } | {z } | {z } | {z } 

 
box 1 box 2 box 1 box 2

 

 

 
(1)(3) (2) , (2) (1)(3) ,

 | {z } | {z } | {z } | {z } 

 
box 1 box 2 box 1 box 2

 

 

 
(2)(3) (1) , (1) (2)(3) .

 | {z } | {z } | {z } | {z } 

 
box 1 box 2 box 1 box 2

Proposition 3.5.2. We have

(# of injective L → U placements A → X ) = [| A| ≤ | X |] .

Proof. If | A| > | X |, then the Pigeonhole Principle shows that no such placements
exist, and thus (# of injective L → U placements) = 0.
Now assume that | A| ≤ | X |. WLOG assume that X = [| X |], so that the boxes are
labelled 1, 2, . . . , | X |. Thus, injective L → U placements do exist: for example, we
can place each ball i into box i (thus leaving the last | X | − | A| many boxes empty).
Furthermore, any two injective L → U placements are identical. (Indeed, any such
placement consists of one box with ball 1, one box with ball 2, one box with ball
3, and so on, totalling to | A| many boxes filled with 1 ball each, and furthermore
| X | − | A| many empty boxes. It is clear that this uniquely describes it as an L → U
placement, because any two L → L placements of this form can be turned into one
Enumerative Combinatorics: class notes page 338

another by a permutation of the boxes. For example,

(2) (1) (3)

| {z } | {z } | {z } | {z } | {z }
box 1 box 2 box 3 box 4 box 5

box
∼ (1) (2) (3) .
| {z } | {z } | {z } | {z } | {z }
box 1 box 2 box 3 box 4 box 5

Thus, any two injective L → U placements are identical.) Hence, the total # of
injective L → U placements is 1. But this is precisely [| A| ≤ | X |], since | A| ≤
| X |.
 
n
Recall: If n ∈ N and k ∈ N, then := sur (n, k ) /k! is called a Stirling
k
number of the second kind.
Proposition 3.5.3. We have
 
| A| sur (| A| , | X |)
(# of surjective L → U placements A → X ) = = .
|X| | X |!

Proof. The crux of the proof is the following claim:

box
Claim 1: Each ∼ -equivalence class of surjective maps (i.e., each surjec-
tive L → U placement) contains exactly | X |! many maps A → X.
box
[Proof of Claim 1: Consider the ∼ -equivalence class of some surjective map g :
A → X. The elements of this class are all the maps of the form σ ◦ g with σ a
permutation of X. There are | X |! many permutations of X, and they all lead to
distinct maps σ ◦ g (because if you know σ ◦ g and you have an x ∈ X, then you
can tell what σ ( x ) is by looking at a ball placed in box x by g, and checking which
box contains this ball in σ ◦ g). Thus, there are exactly | X |! many distinct maps
σ ◦ g in the class of g. This proves Claim 1.]
Now, Proposition 2.4.11 yields
sur (| A| , | X |)
= (# of surjections A → X )
!
here, we have split the sum
= ∑ |C |
|{z} box
according to the ∼ -equivalence class
box
C is a ∼ -equivalence =| X |!
class of surjections (by Claim 1)
 
∑
box
= | X |! = # of ∼ -equivalence classes of surjections · | X |!.
box
C is a ∼ -equivalence
class of surjections
Enumerative Combinatorics: class notes page 339

Thus, dividing by | X |!, we obtain

 
box
# of ∼ -equivalence classes of surjections
 
| A|
= sur (| A| , | X |) / | X |! =
|X|
 
| A|
(since was defined to be sur (| A| , | X |) / | X |!). This proves Proposition 3.5.3
|X|
box
(since the ∼ -equivalence classes of surjections are precisely the surjective L → U
placements A → X).

Class of 2019-11-13

Proposition 3.5.4. We have

     
| A| | A| | A|
(# of L → U placements A → X ) = + +···+ .
0 1 |X|

Proof. We can prove a better claim: For each k ∈ {0, 1, . . . , | X |}, we have
 
| A|
(# of L → U placements with exactly k nonempty boxes) = .
k
(This is not hard to see, because in an L → U placement, we can WLOG assume
that all empty boxes are at the end, and thus we can simply ignore the empty
boxes.)
Adding these equalities up for all k ∈ {0, 1, . . . , | X |}, we obtain precisely the
claim of Proposition 3.5.4. (Details LTTR.)
Thus, our table now looks as follows:

arbitrary injective surjective

L→L | X || A| | X || A| sur (| X | , | A|)

| A| + | X | − 1 | A| − 1
     
|X|
U→L
| A| | A| | A| − | X | .
       
| A| | A| | A| | A|
L→U + +···+ [| A| ≤ | X |]
0 1 |X| |X|
U→U

Let us say a few words about an equivalent version of surjective L → U place-

ments: the set partitions.
Enumerative Combinatorics: class notes page 340

Definition 3.5.5. Let S be a set.

(a) A set partition of S is a set F of disjoint nonempty subsets of S such that
the union of these subsets is S.
In other words, a set partition of S is a set {S1 , S2 , . . . , Sk } of nonempty subsets
of S such that each element of S lies in exactly one Si . (Here, we are assuming
that S is finite.)
(b) If F is a set partition of S, then the elements of F are called the parts (or
blocks) of F . Keep in mind that they are subsets of S.
(c) If a set partition F of S has k parts, then we say that F is a set partition of
S into k parts.

Example 3.5.6. Here are all set partitions of the set [3] = {1, 2, 3}:

{{1, 2, 3}} , {{1, 2} , {3}} , {{1, 3} , {2}} , {{2, 3} , {1}} ,

{{1} , {2} , {3}} .

And here are the same set partitions, drawn as pictures (each part of the set
partition corresponds to a blob):

3 3 3
3 3

1 2 1 2 1 2
1 2 1 2

Proposition 3.5.7. Let A be an n-element set. Let k ∈ N. Then,

 
n
(# of set partitions of A into k parts) = .
k

Proof. Let X = [k ]. Then, there is a bijection

{set partitions of A into k parts} → {surjective L → U placements A → X } ,
{S1 , S2 , . . . , Sk } 7→ (all elements in Si go into box i ) .
(This is well-defined, because the resulting L → U placement does not depend on
the order in which we have listed the blocks of our set partition; any two orders
box
lead to ∼ -equivalent maps.) Now, it remains to recall the bijection principle and
apply Proposition 3.5.3.
 
m
Recall that we have shown a bunch of properties of sur (m, n). Since =
  n
m
sur (m, n) /n!, we can translate them into properties of :
n
Enumerative Combinatorics: class notes page 341

 Let n ∈ N and k ∈ N.
Proposition 3.5.8.
n
(a) We have = [ n = 0].
0 
0
(b) We have = [ k = 0].
 k
n
(c) We have = 0 if k > n.
k  
n−1 n−1
  
n
(d) We have = +k if n > 0 and k > 0.
k k − 1 k
n −1 n
    
n j
(e) We have = ∑ /k if k > 0.
k j =0 j k−1
   
n 1 k k −i k n
(f) We have = ∑ (−1) i .
k k! i=0 i

Proof. (a) Follows from Proposition 2.4.12 (a).

(b) Follows from Proposition 2.4.12 (d).
(c) Follows from Proposition 2.4.12 (f).
(d) Follows from Proposition 2.4.14.
(e) Follows from Proposition 2.4.13.
(f) Follows from Theorem 2.4.17.

Remark 3.5.9. Let n ∈ N. The n-th Bell number B (n) is defined as the # of all
set partitions of [n]. Thus, Proposition 3.5.3 yields
     
n n n
B (n) = + +···+ .
0 1 n

For example, B (3) = 5.

There is no explicit formula for B (n), but there is a recurrence relation:
n  
n
B ( n + 1) = ∑ B (i ) .
i =0
i

See [Guicha20, Theorem 1.4.3] for a proof of this recurrence relation, and
[18s-hw3s, §0.3.2] for more about the Bell numbers.

3.6. U → U and integer partitions

It remains to count U → U placements.
A U → U placement looks like this:

••• • •• •• ,
Enumerative Combinatorics: class notes page 342

with the understanding that the boxes, too, are interchangeable. Thus, we can
order the boxes by decreasing number of balls:

••• •• •• • .

You can encode this U → U placement by a sequence of numbers, which say how
many balls lie in each box:
(3, 2, 2, 1, 0, 0, 0) .
The decreasing order makes this sequence unique.
Let us introduce a name for such sequences, more precisely, for such sequences
that don’t contain zeroes:

Definition 3.6.1. A partition of an integer n is a weakly decreasing list

( a1 , a2 , . . . , ak ) of positive integers whose sum is n (that is, a1 ≥ a2 ≥ · · · ≥ ak > 0
and a1 + a2 + · · · + ak = n).
Instead of “partition”, we can also say “integer partition”.
The positive integers a1 , a2 , . . . , ak are called the parts of the partition
( a1 , a2 , . . . , a k ).
If a partition of n has k parts, then we say that it is a partition of n into k parts.

Example 3.6.2. The partitions of 5 are

(5) , (4, 1) , (3, 2) , (3, 1, 1) ,

(2, 2, 1) , (2, 1, 1, 1) , (1, 1, 1, 1, 1) .

The partition (2, 2, 1) is a partition of 5 into 3 parts.

Remark 3.6.3. A partition of n is the same as a weakly decreasing composition

of n.

Definition 3.6.4. Let n ∈ Z and k ∈ N. Then, we set

pk (n) = (# of partitions of n into k parts) .

Example 3.6.5. We have

p0 (5) = 0, p1 (5) = 1, p2 (5) = 2, p3 (5) = 2,

p4 (5) = 1, p5 (5) = 1, pk (5) = 0 for k > 5.

Proposition 3.6.6. Let n ∈ Z and k ∈ N.

(a) We have pk (n) = 0 when n < 0.
(b) We have pk (n) = 0 when k > n.
Enumerative Combinatorics: class notes page 343

(c) We have p0 (n) = [n = 0].

(d) We have p1 (n) = [n > 0].
(e) We have pk (n) = pk (n − k) + pk−1 (n − 1) when k ≥ 1.
(f) We have p2 (n) = bn/2c if n ∈ N.

Proof. (a) A sum of positive integers is never negative. Thus, there exist no parti-
tions of n when n < 0.
(b) If ( a1 , a2 , . . . , ak ) is a partition of n into k parts, then

n = a1 + a2 + · · · + ak ≥ 1| + 1 +{z· · · + 1} = k.
|{z} |{z} |{z}
≥1 ≥1 ≥1 k times

Thus, no such partition exists if k > n.

(c) A partition of n into 0 parts is a 0-tuple of positive integers whose sum is n.
But the sum of a 0-tuple is always 0. Hence, such a partition exists only for n = 0,
and is unique. Thus, p0 (n) = 1 if n = 0 and is 0 otherwise.
(d) If n > 0, then there is only one partition of n into 1 part, namely the 1-tuple
(n). If n = 0, then there is no partition of n into 1 part.
(e) Assume that k ≥ 1. Let us call a partition of n

• red if 1 is a part of it;

• green if 1 is not a part of it.

Any red partition of n must end with a 1 (since it contains a 1 but is weakly
decreasing). Thus, the map

{red partitions of n into k parts} → {partitions of n − 1 into k − 1 parts} ,

( a 1 , a 2 , . . . , a k −1 , 1 ) 7 → ( a 1 , a 2 , . . . , a k −1 )
is well-defined. It is easy to see that this map is a bijection. Hence, the bijection
principle yields

(# of red partitions of n into k parts)

= (# of partitions of n − 1 into k − 1 parts) = pk−1 (n − 1) .
All parts of a green partition of n are > 1 (since they are positive integers and
6= 1). Thus, the map
{green partitions of n into k parts} → {partitions of n − k into k parts}
( a1 , a2 , . . . , ak ) 7→ ( a1 − 1, a2 − 1, . . . , ak − 1)
is well-defined. It is easy to see that this map is a bijection. Hence, the bijection
principle yields

(# of green partitions of n into k parts)

= (# of partitions of n − k into k parts) = pk (n − k) .
Enumerative Combinatorics: class notes page 344

Adding these two equalities together, we get the claim of (e).

(f) Let n ∈ N. Then, the partitions of n into 2 parts are

(n − 1, 1) , (n − 2, 2) , (n − 3, 3) , ..., (dn/2e , bn/2c)

(where d x e denotes the ceiling of the real number x). So there are bn/2c many of
them.

Proposition 3.6.7. We have

(# of surjective U → U placements A → X ) = p|X | (| A|) .

Proof. Proof idea: Encode a surjective U → U placement  A → X as a partition of

| A| into | X | parts: namely the partition a1 , a2 , . . . , a|X | , where

ai = (# of balls in the box with the i-th largest # of balls) .

This is a bijection.

Proposition 3.6.8. We have

(# of injective U → U placements A → X ) = [| A| ≤ | X |] .

Proof. This follows from Proposition 3.5.2.

Proposition 3.6.9. We have

(# of U → U placements A → X ) = p0 (| A|) + p1 (| A|) + · · · + p|X | (| A|) .

Proof. Similar to the proof of Proposition 3.5.4.

Thus, we have obtained a full table of answers to the “twelvefold way” counting
problems:

arbitrary injective surjective

L→L | X || A| | X || A| sur (| X | , | A|)

| A| + | X | − 1 | A| − 1
     
|X|
U→L
| A| | A| | A| − | X | .
       
| A| | A| | A| | A|
L→U + +···+ [| A| ≤ | X |]
0 1 |X| |X|
U→U p0 (| A|) + p1 (| A|) + · · · + p|X | (| A|) [| A| ≤ | X |] p|X | (| A|)
Enumerative Combinatorics: class notes page 345

By the way, what happens if we add an extra column for “bijective”? Here is this
column:
bijective
L→L [| A| = | X |] · | X |!
U→L [| A| = | X |] .
L→U [| A| = | X |]
U→U [| A| = | X |]
(See [17f-hw1s, Exercise 7] for the proofs.) So that’s not a very interesting column.

Class of 2019-11-15

3.7. Integer partitions (an introduction)

Recall that we have set

pk (n) := (# of partitions of n into k parts)

for all n ∈ Z and k ∈ N.

Proposition 3.6.6 (e) says that

p k ( n ) = p k ( n − k ) + p k −1 ( n − k ) for all n ∈ Z and k ≥ 1.

Let us also set

p (n) := (# of partitions of n)
for all n ∈ Z. Then,

p ( n ) = p0 ( n ) + p1 ( n ) + · · · + p n ( n ) for each n ∈ Z

(by Proposition 3.6.6 (b)).

Let us now count partitions with some more special properties.

Definition 3.7.1. Let n ∈ Z.

(a) Let

podd (n) = (# of partitions of n into odd parts)

= (# of partitions ( a1 , a2 , . . . , ak ) of n such that all ai are odd) .

(b) Let

pdist (n) = (# of partitions of n into distinct parts)

= (# of partitions ( a1 , a2 , . . . , ak ) of n such that a1 > a2 > · · · > ak ) .
Enumerative Combinatorics: class notes page 346

Example 3.7.2. (a) We have

podd (7) = |{(7) , (5, 1, 1) , (3, 3, 1) , (3, 1, 1, 1, 1) , (1, 1, 1, 1, 1, 1, 1)}| = 5.

(b) We have

pdist (7) = |{(7) , (6, 1) , (5, 2) , (4, 3) , (4, 2, 1)}| = 5.

The following theorem is due to Euler:

Theorem 3.7.3. Let n ∈ Z. Then, podd (n) = pdist (n).

Proof. Here is a rough outline. See [Andrew16] or [Galvin17, Proposition 18.1] for
a more detailed version (albeit with a slightly different version of the bijection).
We construct a map

A : {partitions of n into odd parts} → {partitions of n into distinct parts}

which transforms a partition as follows: Repeatedly merge two equal parts until no
more equal parts can be found. “Merging two equal parts” means replacing two
equal parts a, a by the single part 2a, and (if necessarily) rearranging the resulting
tuple back into weakly decreasing order.
(Examples: Let us compute A (5, 5, 3, 1, 1, 1):



5, 5, 3, 1, 1, 1 → 10, 3, 1, 1, 1 → (10, 3, 2, 1)

(where we underline equal entries that are about to get merged). Thus, A (5, 5, 3, 1, 1, 1) =
(10, 3, 2, 1).
Let us compute A (5, 3, 1, 1, 1, 1):




5, 3, 1, 1, 1, 1 → 5, 3, 2, 1, 1 → 5, 3, 2, 2 → (5, 4, 3) .

Thus, A (5, 3, 1, 1, 1, 1) = (5, 4, 3).)

Why is this map A well-defined? This is not obvious.
Our definition of A was non-deterministic: It tells us to merge equal parts; but it
does not tell us which equal parts to choose first (and there can be several choices).
Thus, we have to prove that the result of our many merges does not depend on the
order in which we do the merges.
One way to prove this is using something called the diamond lemma. Another
way is by writing the parts of the partitions in binary (this is what Andrews does
in [Andrew16]; he doesn’t even talk about merging).
The inverse of A transforms a partition by repeatedly splitting even parts into
two equal pieces.
(The map A is called the Glaisher bijection; there are several other bijections that
work.)
Enumerative Combinatorics: class notes page 347

Proposition 3.7.4. Let n ∈ Z and k > 0. Then,

pk (n) = (# of partitions of n whose largest part is k) .

Example 3.7.5. We have p3 (5) = (# of partitions of 5 whose largest part is 3)

(indeed, both numbers are 2). Indeed, the partitions of 5 into 3 parts are (3, 1, 1)
and (2, 2, 1).

Proof. Picture proof: e.g., let n = 14 and k = 4. Start with the partition λ =
(5, 4, 4, 1) of n into k parts. Draw a table of k left-aligned rows, where the length of
each row equals the corresponding part of λ:

5→
4→
.
4→
1→

Now, flip the table across the main diagonal (i.e., the diagonal going from the
top-left to the bottom-right), so that the rows become columns and vice versa:
5 4 4 1
↓ ↓ ↓ ↓

The lengths of the rows of the resulting table again form a partition of n. The
largest part of this new partition is k (because our original table had k rows, so the
flipped table has k columns). In our example, this new partition is (4, 3, 3, 3, 1).
Thus, we obtain a map
{partitions of n into k parts} → {partitions of n whose largest part is k} ,
which transforms a partition by flipping its table.
This map is a bijection; indeed, its inverse map is defined in the same way. This
bijection is called conjugation of partitions. The bijection principle now yields the
proposition.
(Note: The table that we constructed above is called the Young diagram of λ, or
the Ferrers diagram of λ.)
Enumerative Combinatorics: class notes page 348

Definition 3.7.6. For any k ∈ Z, define wk ∈ N by

(3k − 1) k
wk = .
2
This is called a pentagonal number.

Theorem 3.7.7 (Euler’s recursion for the partition numbers). For each n > 0, we
have

p (n)
= p (n − 1) + p (n − 2) − p (n − 5) − p (n − 7) + p (n − 12) + p (n − 15) ± · · ·
= ∑ (−1)k−1 p (n − wk ) .
k∈Z;
k 6 =0

We might get to prove this later on, using the technique of generating functions.
There is, of course, much more to say about partitions. See [Wilf09] or [AndEri04]
for two introductions.

3.8. Odds and ends

Here are some random counting exercises.

Exercise 3.8.1. Given n persons (n > 0) and k tasks (k > 0).

(a) What is the # of ways to assign a task to each person such that each task
has at least 1 person working on it?
(b) What if we additionally want to choose a leader for each task (among the
people assigned to this task)?
(c) What if, instead, we want to choose a vertical hierarchy (between all people
working on the task) for each task? (A “vertical hierarchy” means a ranking of
all people working on the task, with no ties.)
Example: Assume n = 8 and k = 3. Let our 8 people be 1, 2, 3, 4, 5, 6, 7, 8, and
let our 3 tasks be A, B, C.
(a) One option is
task people working on it
A 1, 2, 5
B 3
C 4, 6, 7, 8
Enumerative Combinatorics: class notes page 349

(b) One option is

task people working on it
A 1, 2, 5 with leader 2
B 3 with leader 3
C 4, 6, 7, 8 with leader 7
(c) One option is
task people working on it
A 1>5>2
B 3
C 7>8>4>6
(where the “>” signs stand for “is ranked above”).

Proof. (Solution sketch.)

(a) sur (n, k ).
Proof. Choosing such an arrangement is tantamount to choosing a surjection
{people} → {tasks}.
(b) nk · kn−k .
Proof. First, choose a leader for each task. There are nk options for this (since the
leaders have to be distinct). Then, every of the remaining n − k people joins one of
the k leaders. n−k options for this.
 There  are k
n−1
(c) n! · .
k−1
Proof. First, order all the n people in some way. There are n! options for this.
Then, splitthis ordering into k nonempty chunks (corresponding to the k tasks).
n−1

There are options for this, since we need to put k − 1 separators into the
k−1
n − 1 positions between two consecutive people in our ordering.

Class of 2019-11-18

4. Permutations
4.1. Introduction
We will now talk about permutations in more detail. For deeper treatments, see
[Bona12], [Sagan01] and [Stanle11, Chapter 1]. See also [Grinbe15, Chapter 5] for a
detailed exposition of the basics.
Recall: A permutation of a set X is a bijection from X to X.
Enumerative Combinatorics: class notes page 350

4.2. Definitions
Definition 4.2.1. Let n ∈ N. Let Sn be the set of all permutations of [n]. This
set Sn is called the n-th symmetric group. It is closed under composition (i.e.,
for any α ∈ Sn and β ∈ Sn , we have α ◦ β ∈ Sn ) and under inverses (i.e., for any
σ ∈ Sn , we have σ−1 ∈ Sn ), and contains the identity map id[n] .

Definition 4.2.2. Let n ∈ N and σ ∈ Sn . We introduce two notations for σ:

(a) The one-line notation of σ is the n-tuple (σ (1) , σ (2) , . . . , σ (n)). (Conven-
tionally, authors use square brackets for it, but we use parentheses.)
(b) The cycle digraph of σ is defined (informally) as follows:
For each i ∈ [n], draw a point (“node”) labelled i.
For each i ∈ [n], draw an arrow (“arc”) from the node labelled i to the node
labelled σ (i ).
The result is called the cycle digraph of σ.

Example 4.2.3. Let σ : [10] → [10] be the map that sends the elements
1, 2, 3, 4, 5, 6, 7, 8, 9, 10 to 5, 4, 3, 2, 6, 10, 1, 9, 8, 7, respectively. Then, σ is a bijection,
hence a permutation of [10]. The one-line notation of σ is (5, 4, 3, 2, 6, 10, 1, 9, 8, 7).
The cycle digraph of σ is

5 9 8

1 6

7 10 2 4
.

4.3. Transpositions and cycles

Definition 4.3.1. (a) Let i and j be two distinct elements of a set X.
Then, the transposition ti,j is the permutation of X that sends i to j, sends j to
i, and leaves all other elements in their places.
If X = [n] for some n ∈ N, and if i < j, then the one-line notation of ti,j is
 

 1, 2, . . . , i − 1 , j, i + 1, i + 2, . . . , j − 1, i, j + 1, j + 2, . . . , n  .
 
| {z } | {z } | {z }
integers from 1 to i −1 integers from i +1 to j−1 integers from j+1 to n
Enumerative Combinatorics: class notes page 351

(b) Let n ∈ N and i ∈ [n − 1]. Then, the simple transposition si is defined

by si = ti,i+1 ∈ Sn . So a simple transposition is a transposition that swaps two
consecutive integers.

Convention 4.3.2. If α and β are two permutations of a set X, then we write αβ

for α ◦ β.
Also, if α is any permutation of X, then we set αi := |α ◦ α ◦{z· · · ◦ α} for each
i times
i ∈ N. If i = 0, this is understood to be idX .

Proposition 4.3.3. Let n ∈ N.

(a) We have s2i = id for all i ∈ [n − 1]. (Recall: s2i = si ◦ si .)
(b) We have si s j = s j si for all i, j ∈ [n − 1] with |i − j| > 1.
(c) We have si si+1 si = si+1 si si+1 = ti,i+2 for all i ∈ [n − 2]. (This is known as
“the braid relation for permutations”.)

Proof. Straightforward verification that both sides send each k ∈ [n] to the same
value. (The “hardest” part is part (c), which is proved in [Grinbe15, solution to
Exercise 5.1 (a)].)

Definition 4.3.4. Let n ∈ N. Let w0 be the permutation in Sn that sends each

i ∈ [n] to n + 1 − i.
In other words, it “reflects” all numbers from 1 to n across the middle of [n].
It is the unique strictly decreasing permutation of [n].
 
1 2 3 4 5
Example 4.3.5. If n = 5, then w0 = in two-line notation, and
5 4 3 2 1
w0 = (5, 4, 3, 2, 1) in one-line notation.

Definition 4.3.6. Let X be a set. Let i1 , i2 , . . . , ik be k distinct elements of X. Then,

cyci1 ,i2 ,...,ik

means the permutation of X that sends i1 7→ i2 , i2 7→ i3 , i3 7→ i4 , . . ., ik−1 7→ ik ,

ik 7→ i1 and leaves all other elements of X unchanged. This is called a k-cycle.

Remark 4.3.7. People often write (i1 , i2 , . . . , ik ) for cyci1 ,i2 ,...,ik . (But we won’t.)

Example 4.3.8. Let X = [8]. Then, the permutation cyc2,6,5 of [8] has two-line
 
1 2 3 4 5 6 7 8
notation .
1 6 3 4 2 5 7 8
Enumerative Combinatorics: class notes page 352

Remark 4.3.9. A permutation α is called an involution if α2 = id. For example,

all transpositions ti,j are involutions. Also, both id and w0 are involutions. But
k-cycles with k > 2 are not involutions.

Proposition 4.3.10. Let n ∈ N.

(a) For any k distinct elements i1 , i2 , . . . , ik of [n], we have

cyci1 ,i2 ,...,ik = ti1 ,i2 ti2 ,i3 · · · tik−1 ,ik .

| {z }
k−1 transpositions

(b) For any i ∈ [n] and k ∈ N such that i + k − 1 ≤ n, we have

cyci,i+1,...,i+k−1 = si si+1 · · · si+k−2 .

(c) For any i ∈ [n], we have cyci = id.

(d) For any distinct i, j ∈ [n], we have cyci,j = ti,j .
(e) For any k distinct elements i1 , i2 , . . . , ik of [n], we have

cyci1 ,i2 ,...,ik = cycik ,i1 ,i2 ,...,ik−1 .

(f) For any k distinct elements i1 , i2 , . . . , ik of [n] and any σ ∈ Sn , then

σ cyci1 ,i2 ,...,ik σ−1 = cycσ(i1 ),σ(i2 ),...,σ(ik ) .

(g) If i and j are two integers satisfying 1 ≤ i < j ≤ n, then

ti,j = si si+1 · · · s j−1 · · · si+1 si

 
this is a product starting at si , then walking up to s j−1
and back down to si
= s j −1 s j −2 · · · s i · · · s j −2 s j −1
!
this is a product starting at s j−1 , then walking down to si
.
and back up to s j−1

(h) We have

w0 = cyc1,2,...,n cyc1,2,...,n−1 · · · cyc1

= ( s 1 s 2 · · · s n −1 ) ( s 1 s 2 · · · s n −2 ) · · · ( s 1 s 2 ) s 1
= cyc1 cyc2,1 cyc3,2,1 · · · cycn,n−1,...,1
= s 1 ( s 2 s 1 ) ( s 3 s 2 s 1 ) · · · ( s n −1 s n −2 · · · s 1 ) .

Proof. All of these are doable with some bookkeeping and induction.
(a) This is [Grinbe15, Exercise 5.16].
Enumerative Combinatorics: class notes page 353

(b) This follows from part (a).

(c), (d) and (e) are obvious consequences of the definitions.
(f) This is [Grinbe15, Exercise 5.17 (a)].
(g) This can be proved by induction on j − i or otherwise. Perhaps the quickest
way, however, is using part (a): To show that ti,j = si si+1 · · · s j−1 · · · si+1 si , we notice
that part (a) yields
si si+1 · · · s j−1 = ti,i+1 ti+1,i+2 · · · t j−1,j = cyci,i+1,...,j (by part (a))
and
s j−2 s j−3 · · · si+1 si = t j−1,j−2 t j−2,j−3 · · · ti+2,i+1 ti+1,i = cyc j−1,j−2,...,i+1,i (by part (a)) .
Now,
s i s i +1 · · · s j −1 · · · s i +1 s i



= s i s i +1 · · · s j −1 s j −2 s j −3 · · · s i +1 s i
| {z }| {z }
=cyci,i+1,...,j =cyc j−1,j−2,...,i+1,i

= cyci,i+1,...,j cyc j−1,j−2,...,i+1,i = ti,j (this is easy to check) .

This proves the first equality sign of part (g); the second is proved similarly.
(h) TODO. For now, see [18s].

4.4. Inversions and lengths

Definition 4.4.1. Let n ∈ N and σ ∈ Sn .
(a) An inversion of σ is a pair (i, j) of elements of [n] such that i < j and
σ ( i ) > σ ( j ).
(b) The length (or Coxeter length) of σ is the # of inversions of σ. It is called
` ( σ ).
(In LaTeX, you can obtain the symbol “`” by typing “\ell”. If you just type “l”,
then you get the less remarkable letter “l”.)
Example 4.4.2. Let π = (3, 1, 4, 2) ∈ S4 (in one-line notation).
The inversions of π are
 

(1, 4) since 1 < 4 and π (1) > π (4) and

| {z } | {z }
=3 =2
 

(3, 4) since 3 < 4 and π (3) > π (4) and

| {z } | {z }
=4 =2
 

(1, 2) since 1 < 2 and π (1) > π (2) .

| {z } | {z }
=3 =1
Enumerative Combinatorics: class notes page 354

So the length of π is 3.
 
n
Remark 4.4.3. If σ ∈ Sn , then 0 ≤ ` (σ) ≤ .
2
The only σ ∈ Sn with ` (σ) = 0 isid.
n
The only σ ∈ Sn with ` (σ) = is w0 .
2
Inbetween, there are many permutations with a given ` (σ).

Class of 2019-11-20

Remark 4.4.4. If n ∈ N, then the permutations of [n] can be represented as the

vertices of an (n − 1)-dimensional polyhedron in n-dimensional space. Namely,
each permutation σ of [n] gives rise to the point (σ (1) , σ (2) , . . . , σ (n)) ∈ Rn ,
and the convex hull of all these points is the polyhedron. This is called the
permutahedron.

Proposition 4.4.5. For every σ ∈ Sn , we have ` σ−1 = ` (σ ).

Proof. Recall: An inversion of σ is a pair (i, j) ∈ [n] × [n] such that i < j and
σ (i ) > σ ( j). An inversion of σ−1 is a pair (u, v) ∈ [n] × [n] such that u < v and
σ −1 ( u ) > σ −1 ( v ).
The map
n o
{inversions of σ} → inversions of σ−1 ,
(i, j) 7→ (σ ( j) , σ (i ))

is well-defined and bijective (its inverse map sends each (u, v) ∈ inversions of σ−1


to σ−1 (v) , σ−1 (u) ). So the bijection principle yields ` (σ) = ` σ−1 .

(For details, see [Grinbe15, Exercise 5.2 (f)].)

Recall from last time

• the transpositions ti,j (swapping i with j while leaving all other numbers un-
changed), and

• the simple transpositions si = ti,i+1 .

Proposition 4.4.6. Let n ∈ N, σ ∈ Sn and k ∈ [n − 1].

(a) We have
(
` (σ) + 1, if σ (k) < σ (k + 1) ;
` (σ ◦ sk ) =
` (σ) − 1, if σ (k) > σ (k + 1) .
Enumerative Combinatorics: class notes page 355

(b) We have
(
` (σ) + 1, if σ−1 (k) < σ−1 (k + 1) ;
` (sk ◦ σ ) =
` (σ) − 1, if σ−1 (k) > σ−1 (k + 1) .

[Note: σ−1 (i ) is the position in which i appears in the one-line notation of σ.

For example, if σ = (5, 1, 2, 3, 6, 4) in one-line notation, then σ−1 (6) = 5.]
Proof. (b) How do the inversions of sk ◦ σ differ from the inversions of σ ?
I claim that they are the same, except that (when we go from σ to sk ◦ σ)
• if σ−1 (k ) < σ−1 (k + 1), then we gain a new inversion σ−1 (k ) , σ−1 (k + 1) ;

• if σ−1 (k ) > σ−1 (k + 1), then we lose an existing inversion σ−1 (k + 1) , σ−1 (k ) .

For example, let

n=7 and σ = (6, 2, 4, 1, 7, 3, 5) and k = 4.
Thus, σ−1 (k) < σ−1 (k + 1) and sk ◦ σ = (6, 2, 5, 1, 7, 3, 4). So we gain a new inver-
sion (3, 7) since 5 > 4, but all other inversions remain the same.
For another example, let
n=7 and σ = (3, 1, 4, 2, 7, 6, 5) and k = 2.
Thus, σ −1 ( k )> σ −1 ( k + 1 )
and sk ◦ σ = (2, 1, 4, 3, 7, 6, 5). So we lose the inversion
(1, 4) since 3 > 2, but all other inversions remain the same.
Thus, when going from σ to sk ◦ σ, the # of inversions increases by 1 if σ−1 (k ) <
σ 1 (k + 1) and decreases by 1 if σ−1 (k ) > σ−1 (k + 1). This is precisely the claim
−

of (b).
(a) Apply part (b) to σ−1 instead of σ. We get

  ` σ−1
+ 1, if σ−1
−1 (k ) < σ−1
−1 (k + 1) ;
` s k ◦ σ −1 =
` σ−1
− 1, if σ−1
−1 (k ) > σ−1
−1 (k + 1)
(
` (σ) + 1, if σ (k) < σ (k + 1) ;
=
` (σ) − 1, if σ (k) > σ (k + 1)

−1
(since ` σ−1 = ` (σ ) (by previous Proposition) and σ−1


= σ). On the other
hand, the previous Proposition yields
 
 
 
 
 −1
 −1 −1
       
−1
` s k ◦ σ −1 = `  s k ◦ σ −1 = ` ◦ ( s )
  
  σ k 
  | {z } | {z } 
=sk
 | {z } 
 = σ−1 −1 ◦(s )−1  =σ (since sk ◦sk =id )
( ) k
(since (α◦ β)−1 = β−1 ◦α−1 )
= ` (σ ◦ sk ) .
Enumerative Combinatorics: class notes page 356

Comparing these equalities, we get

(
` (σ) + 1, if σ (k) < σ (k + 1) ;
` (σ ◦ sk ) =
` (σ) − 1, if σ (k) > σ (k + 1) .

(See [Grinbe15, Exercise 5.2 (a)] for a similar proof in more detail.)

Remark 4.4.7. Let n ∈ N and σ ∈ Sn . Let i, j ∈ [n] be such that i < j and
σ (i ) > σ ( j). Is

` σ ◦ ti,j < ` (σ) ?
It’s not obvious, but the answer is “yes”. See [Grinbe15, Exercise 5.20].

Recall: A simple transposition in Sn means one of the transpositions s1 , s2 , . . . , sn−1 .

We shall occasionally abbreviate “simple transposition” as “simple”.

Theorem 4.4.8. Let n ∈ N and σ ∈ Sn .

(a) We can write σ as a composition of ` (σ) simples.
(b) The number ` (σ ) is the smallest p ∈ N such that we can write σ as a
composition of p simples.
[Keep in mind: The composition of 0 simples is id.]

Example 4.4.9. Let σ ∈ S4 be the permutation (4, 1, 3, 2) (in one-line notation).

How can we represent σ as a composition of simples? There are several ways to
do so, for example

σ = s2 ◦ s3 ◦ s2 ◦ s1 = s3 ◦ s2 ◦ s3 ◦ s1 = s3 ◦ s2 ◦ s1 ◦ s3
| {z } | {z }
= s3 ◦ s2 ◦ s3 = s1 ◦ s3
= s2 ◦ s1 ◦ s1 ◦ s3 ◦ s2 ◦ s1 = · · · .

The shortest of these representations involve 4 simples, as the Theorem above

predicts (since ` (σ) = 4).

Proof. (Proof of Theorem.)

(a) Induction on ` (σ ).
Induction base: If ` (σ) = 0, then σ = id, so we can write σ as a composition of 0
simples.
Induction step: Fix h ∈ N. Assume (as the IH) that Theorem (a) holds for ` (σ) =
h.
Now, let σ ∈ Sn be such that ` (σ ) = h + 1.
Hence, ` (σ) = h + 1 > 0, so σ 6= id.
Therefore, there exists some k ∈ [n − 1] such that σ (k ) > σ (k + 1) (because
otherwise, we would have σ (1) ≤ σ (2) ≤ · · · ≤ σ (n), and this would imply
` (σ) = 0, whence σ = id). Fix such a k.
Enumerative Combinatorics: class notes page 357

Part (a) of the previous proposition yields

(
` (σ) + 1, if σ (k) < σ (k + 1) ;
` (σ ◦ sk ) = = ` ( σ ) −1
` (σ) − 1, if σ (k) > σ (k + 1) | {z }
= h +1
= (h + 1) − 1 = h.

Thus, the IH (applied to σ ◦ sk instead of σ) shows that we can write σ ◦ sk as a

composition of ` (σ ◦ sk ) = h simples:

σ ◦ s k = s i1 ◦ s i2 ◦ · · · ◦ s i h for some i1 , i2 , . . . , ih ∈ [n − 1] .

Composing both sides of this equality with (sk )−1 , we obtain

σ = s i1 ◦ s i2 ◦ · · · ◦ s i h ◦ ( s k ) −1


| {z }
=sk


= s i1 ◦ s i2 ◦ · · · ◦ s i h ◦ s k = s i1 ◦ s i2 ◦ · · · ◦ s i h ◦ s k .

This shows that we can write σ as a composition of h + 1 = ` (σ ) simples. Thus,

Theorem (a) holds for ` (σ) = h + 1. This completes the induction step, and so
Theorem (a) is proved by induction.
[The idea behind this proof is known as “bubblesort”.]
(b) Omitted.
(For details, see [Grinbe15, Exercise 5.2 (g)].)

Corollary 4.4.10. Let n ∈ N.

(a) We have ` (σ ◦ τ ) ≡ ` (σ ) + ` (τ ) mod 2 for all σ ∈ Sn and τ ∈ Sn . (In other
words, if ` (σ ◦ τ ) is even, then ` (σ) + ` (τ ) is even, and the same for “odd”.)
(b) We have ` (σ ◦ τ ) ≤ ` (σ) + ` (τ ).
(c) If σ = sk1 ◦ sk2 ◦ · · · ◦ skq , then q ≥ ` (σ) and q ≡ ` (σ) mod 2.

Example 4.4.11. Let n = 4 and σ = (3, 2, 1, 4) and τ = (3, 1, 4, 2) in one-line

notation. Then, ` (σ) = 3 and ` (τ ) = 3. Now, σ ◦ τ = (1, 3, 4, 2) has ` (σ ◦ τ ) = 2.
Corollary (a) says ` (σ ◦ τ ) ≡ ` (σ) + ` (τ ) mod 2. In other words, 2 ≡ 3 +
3 mod 2.
Corollary (b) says ` (σ ◦ τ ) ≤ ` (σ) + ` (τ ). In other words, 2 ≤ 3 + 3.

Proof. (Proof of Corollary) See [Grinbe15, Exercises 5.2 and 5.3]. (Or do it yourself,
e.g., by induction on ` (σ ).)

Proposition 4.4.12. Let n ∈ N.

(a) We have ` (sk )
= 1 for any k ∈ [n − 1].
(b) We have ` ti,j = 2 |i − j| − 1 for any distinct i, j ∈ [n].
 
(c) We have ` cyci,i+1,...,i+k−1 = k − 1 for all i, k.
Enumerative Combinatorics: class notes page 358

 
(d) We have ` cyci1 ,i2 ,...,ik ≥ k − 1 for all distinct i1 , i2 , . . . , ik ∈ [n].
 
n
(e) We have ` (id) = 0 and ` (w0 ) = . (Recall: w0 is the “reflection across
2
the middle of [n]”.)
Proof. (a) follows from (b).
(b) is [Grinbe15, Exercise 5.10 (a)], but also easy to check.
(c) and (d) are parts of [Grinbe15, Exercise 5.17].
(e) is trivial.
Class of 2019-11-25
Remark 4.4.13. For a given k and n, how many σ ∈ Sn have length k ?

• The # of σ ∈ Sn having ` (σ) = 0 is 1 (namely, just σ = id).

• The # of σ ∈ Sn having ` (σ) = 1 is n − 1 (namely, just σ = sk with k ∈

[n − 1]).
• The # of σ ∈ Sn having ` (σ) = 2 is (n − 1) (n + 1) /2. (Exercise.)

What about the general case?

There is no explicit formula, but there is a generating function:

Proposition 4.4.14. Let n ∈ N. Then,

∑ x `(w)
w ∈ Sn
n −1  
= ∏ 1 + x + x2 + · · · + xi
i =1
     
= (1 + x ) · 1 + x + x 2 · 1 + x + x 2 + x 3 · · · · · 1 + x + x 2 + · · · + x n −1 .

Example 4.4.15. Applying this to n = 3, we obtain

 
∑ x = (1 + x ) · 1 + x + x .
`(w) 2
w ∈ S3

Let us check this:

∑ x `(w) = x `(1,2,3) + x `(1,3,2) + x `(2,1,3) + x `(2,3,1) + x `(3,1,2) + x `(3,2,1)
w ∈ S3
(where we are writing each permutation in one-line notation)
= x + x1 + x1 + x2 + x2 + x3 = x0 + 2x1 + 2x2 + x3
0
 
= (1 + x ) · 1 + x + x 2 .
Enumerative Combinatorics: class notes page 359

Proof. The proposition is [Grinbe15, Corollary 5.53], and a proof appears in [Grinbe15,
solution to Exercise 5.18].

4.5. Descents
Definition 4.5.1. Let n ∈ N and σ ∈ Sn .
A descent of σ means a k ∈ [n − 1] such that σ (k ) > σ (k + 1).
The descent set of σ, denoted Des σ, is the set of all descents of σ.

Example 4.5.2. Let π ∈ S4 be the permutation with one-line notation (3, 1, 4, 2).
Then, 1 is a descent of π (since π (1) = 3 > 1 = π (2)), and so is 3, but not 2.
Thus, Des π = {1, 3}.

Exercise 4.5.1. Fix n ≥ 4.

(a) How many σ ∈ Sn have 0 descents?
(b) How many σ ∈ Sn have 1 descent?
(c) How many σ ∈ Sn have n − 1 descents?
(d) How many σ ∈ Sn satisfy 1 ∈ Des σ (that is, σ (1) > σ (2)) ?
(e) How many σ ∈ Sn satisfy 1, 2 ∈ Des σ (that is, σ (1) > σ (2) > σ (3)) ?
(f) How many σ ∈ Sn satisfy 1, 3 ∈ Des σ (that is, σ (1) > σ (2) and σ (3) >
σ (4)) ?

Proof. (Solution sketch.) (a) The answer is 1 (namely, σ = id).

n!
(d) The answer is .
2
First proof: The map

{σ ∈ Sn | σ (1) > σ (2)} → {σ ∈ Sn | σ (1) < σ (2)} ,

σ 7 → σ ◦ s1

is bijective. So each of the 2 sets is half as large as Sn (because each σ ∈ Sn satisfies

either σ (1) > σ (2) or σ (1) < σ (2)). But |Sn | = n!.
Second proof: To construct a σ ∈ Sn satisfying σ (1) > σ (2), we can proceed as
follows:
 
n
• Choose the set {σ (1) , σ (2)}. There are options.
2
• Thus, σ (1) and σ (2) are already uniquely determined, because they have to
satisfy σ (1) > σ (2).

• Choose σ (3) , σ (4) , . . . , σ (n). There are (n − 2)! options.

 
n n! n!
=⇒ The total # is · ( n − 2) ! = = .
2 2! 2
Enumerative Combinatorics: class notes page 360

n!
(e) The answer is .
3!
For the detailed proof, see [18s-mt1s, solution to Exercise 2 (b)].
n! n!
(f) The answer is = .
2! · 2! 4
For the detailed proof, see [18s-mt1s, solution to Exercise 2 (a)].
More generally, you can ask how many permutations σ ∈ Sn have a given bunch
of numbers in their descent set. A similar question is answered in [18f-hw4s, Exer-
cise 4 (a)].
(b) First of all, fix i ∈ [n − 1]. How many σ ∈ Sn have Des σ = {i } ?
In other words, how many σ ∈ Sn satisfy

σ (1) < σ (2) < · · · < σ ( i ) > σ ( i + 1) < σ ( i + 2) < · · · < σ ( n ) .

We have

(# of σ ∈ Sn such that σ (1) < σ (2) < · · · < σ (i ) and σ (i + 1) < σ (i + 2) < · · · < σ (n))
 
n
=
i

(because in order to construct such a σ, it suffices to choose the i-element subset

{σ (1) , σ (2) , . . . , σ (i )} of [n]). Thus,

(# of σ ∈ Sn such that σ (1) < σ (2) < · · · < σ (i ) > σ (i + 1) < σ (i + 2) < · · · < σ (n))
= (# of σ ∈ Sn such that σ (1) < σ (2) < · · · < σ (i ) and σ (i + 1) < σ (i + 2) < · · · < σ (n))
| {z  }
n
=
i
− (# of σ ∈ Sn such that σ (1) < σ (2) < · · · < σ (i ) < σ (i + 1) < σ (i + 2) < · · · < σ (n))
| {z }
=1
 
n
= − 1.
i

Now, forget that we fixed i. Summing this over all i ∈ [n − 1], we get

(# of σ ∈ Sn that have exactly 1 descent)

n−1    n −1  
n n
= ∑ −1 = ∑ − ( n − 1 ) = 2n − 2 − ( n − 1 ) = 2n − ( n + 1 ) .
i =1
i i
|i=1 {z }
=2n −2

(c) Only 1 permutation σ ∈ Sn has n − 1 descents, namely w0 .

4.6. Signs
Enumerative Combinatorics: class notes page 361

Definition 4.6.1. Let n ∈ N. The sign of a permutation σ ∈ Sn is (−1)`(σ) .

It is called (−1)σ or sgn (σ) or sign (σ) or ε (σ).

Proposition 4.6.2. Let n ∈ N.

(a) We have (−1)id = 1.
(b) We have (−1)ti,j = −1.
cyc
(c) We have (−1) i1 ,i2 ,...,ik = (−1)k−1 for any k ≥ 1 and any distinct
i1 , i2 , . . . , i k ∈ [ n ].
(d) We have (−1)σ◦τ = (−1)σ · (−1)τ for any σ, τ ∈ Sn . (In the lingo of abstract
algebra, this is saying “The sign is a group homomorphism from Sn to {1, −1}”.)
(e) We have (−1)σ1 ◦σ2 ◦···◦σp = (−1)σ1 (−1)σ2 · · · (−1)σp for any σ1 , σ2 , . . . , σp ∈
Sn .
−1
(f) We have (−1)σ = (−1)σ for any σ ∈ Sn . (The LHS has to be read as
−1
(−1)(σ ) .)
(g) We have

σ (i ) − σ ( j )
(−1)σ = ∏ i−j
for each σ ∈ Sn .
1≤ i < j ≤ n

(h) If x1 , x2 , . . . , xn are any numbers, and σ ∈ Sn , then

 
∏ ∏ xi − x j .
σ

x σ (i ) − x σ( j) = (−1)
1≤ i < j ≤ n 1≤ i < j ≤ n

Proof. Most of this follows easily from what we have proved above, but here are
references to complete proofs:
(a) This is [Grinbe15, Proposition 5.15 (a)].
(b) This is [Grinbe15, Exercise 5.10 (b)].
(c) This is [Grinbe15, Exercise 5.17 (d)].
(d) This is [Grinbe15, Proposition 5.15 (c)].
(e) This is [Grinbe15, Proposition 5.28].
(f) This is [Grinbe15, Proposition 5.15 (d)].
(g) This is [Grinbe15, Exercise 5.13 (c)].
(h) This is [Grinbe15, Exercise 5.13 (a)].

Definition 4.6.3. Let n ∈ N. A permutation σ ∈ Sn is said to be

• even if (−1)σ = 1 (that is, if ` (σ) is even);

• odd if (−1)σ = −1 (that is, if ` (σ) is odd).

Enumerative Combinatorics: class notes page 362

Corollary 4.6.4. Let n ≥ 2. Then,

(# of even σ ∈ Sn ) = (# of odd σ ∈ Sn ) = n!/2.

Proof. The map

{even σ ∈ Sn } → {odd σ ∈ Sn } ,
σ 7 → σ ◦ s1
is a bijection. (See [Grinbe15, Exercise 5.4] for details.)
Example 4.6.5. The 15-game
1 2 3 4 1 2 3 4
5 6 7 8 5 6 7 8
−→
9 10 11 12 9 10 11 12
13 15 14 13 14 15
(via swaps of the empty cell with the neighboring square (“slides”)) is unsolv-
able.
Why?
Proof sketch.
(a) Let us first consider the 3 × 3-analogue:
1 2 3 1 2 3
4 5 6 −→ 4 5 6 .
8 7 7 8
This is also impossible.
Proof: For each position P, let σP be the permutation of [8] whose one-line
notation is what you get if you read P row by row from left to right.
a b c
P= d e 7→ σP = ( a, b, c, d, e, f , g, h) (in one-line notation).
f g h
Now, a slide changes σP either not at all or by multiplying it with a 3-cycle
| {z }
cyc p,q,r
for some p,q,r
(that is, σP becomes σP ◦ cyc p,q,r ). Thus, the sign of σP never changes (since 3-
cycles have sign 1). Thus, the 3 × 3-game is unsolvable (since the initial position
and the target position have different signs of σP ).
(b) Now to the 4 × 4-version.
The sign of σP is no longer invariant. Instead, every vertical slide flips the sign
of σP . Therefore,
(−1)σP · (−1)which row has an empty square
is invariant. This again proves that the game is unsolvable, because the initial
and target positions have different values of this invariant.
Enumerative Combinatorics: class notes page 363

Remark 4.6.6. For any n ∈ N, the set of all even permutations σ ∈ Sn is called
the alternating group An .

2019-12-04 notes

5. Lattice paths (brief introduction)

We shall say a few words about lattice paths. See [18f, Chapter 6] for more, and see
[Kratte17] for much more.

Definition 5.0.1. The integer lattice (or, for short, lattice) is the set Z2 = Z × Z.
Its elements are called points; indeed, any element ( a, b) ∈ Z2 can be identi-
fied with the point with coordinates a and b on the plane.
Points can be added and subtracted entrywise: e.g.,

( a, b) − (c, d) = ( a − c, b − d) .

If ( a, b) ∈ Z2 and (c, d) ∈ Z2 are two points, then a lattice path (for short, LP)
from ( a, b) to (c, d) means

• (informally) a path from ( a, b) to (c, d) that uses only 2 kinds of steps:

– “up-steps” (U) going ( p, q) 7→ ( p, q + 1);
– “right-steps” (R) going ( p, q) 7→ ( p + 1, q).

• (rigorously) a tuple (v0 , v1 , . . . , vn ) of points vi ∈ Z2 such that

v0 = ( a, b) and vn = (c, d)
 

 

and vi − vi−1 ∈ (0, 1) , (1, 0) for each i ∈ [n] .
 | {z }
 | {z } 

up-step right-step

Example 5.0.2. The following picture shows an LP from (0, 0) to (5, 3):
Enumerative Combinatorics: class notes page 364

Formally speaking, this LP is the 9-tuple

((0, 0) , (1, 0) , (1, 1) , (2, 1) , (3, 1) , (4, 1) , (4, 2) , (4, 3) , (5, 3)) .

Its “step sequence” (i.e., the sequence of the directions of its steps) is
RURRRUUR (meaning that its first step is a right-step, its second step is an
up-step, its third step is a right-step, and so on).
Note that any LP is uniquely determined by its starting point and its “step se-
quence”.

Proposition 5.0.3. Let ( a, b) ∈ Z2 and (c, d) ∈ Z2 be two points. Then,


 c + d − a − b , if c + d ≥ a + b;


(# of LPs from ( a, b) to (c, d)) = c−a .

0, if c + d < a + b

Proof. In each LP, the x-coordinates of the points weakly increase at each step, and
so do the y-coordinates. Thus, LPs from ( a, b) to (c, d) can only exist when c ≥ a
and d ≥ b.
Furthermore, each step of a LP increases
(x-coordinate) + (y-coordinate)
by exactly 1 (in the sense that if ( xi , yi ) and ( xi+1 , yi+1 ) are two consecutive points
on an LP, then xi+1 + yi+1 = ( xi + yi ) + 1). Hence, each LP (v0 , v1 , . . . , vn ) from
( a, b) to (c, d) must have n = c + d − a − b. Thus, if c + d < a + b, then the # of LPs
from ( a, b) to (c, d) is 0. Otherwise, the bijection
{LPs from ( a, b) to (c, d)} → {(c − a) -element subsets of [c + d − a − b]} ,
(v0 , v1 , . . . , vn ) 7→ {i ∈ [n] | vi − vi−1 = (1, 0)}
c+d−a−b
 
shows that the # of LPs is .
c−a
Definition 5.0.4. Let v = (v0 , v1 , . . . , vn ) be a LP from ( a, b) to (c, d). Let p ∈ Z2 .
We say that p ∈ v (in words: p lies on v) if p ∈ {v0 , v1 , . . . , vn }.

Exercise 5.0.1. Find the # of LPs v from (0, 0) to (6, 6) such that (2, 2) ∈ v.
Proof. (Solution sketch.) Each such v consists of a LP from (0, 0) to (2, 2) and a LP
from (2, 2) to (6, 6). Thus, the product rule yields
(# of LPs v from (0, 0) to (6, 6) such that (2, 2) ∈ v)
= (# of LPs from (0, 0) to (2, 2)) · (# of LPs from (2, 2) to (6, 6)) .
| {z } | {z }
2+2−0−0 6+6−2−2
   
= =
2−0 6−2
Enumerative Combinatorics: class notes page 365

We shall now introduce a special class of LPs:

Definition 5.0.5. A LP v is said to be Catalan if x ≥ y for each ( x, y) ∈ v.
(Visually, this means that v never strays above the diagonal x = y.)
(In [18f], I say “legal” instead of “Catalan”.)
Example 5.0.6. The following picture shows a Catalan LP from (0, 0) to (4, 4):

(Here, the region shaded cyan is the set of all ( x, y) ∈ R2 satisfying x ≥ y; this is
the region in which a Catalan LP must stay.)
The following LP from (0, 0) to (4, 4) is not Catalan:

Definition 5.0.7. If n, m ∈ Z, then we set

Ln,m = (# of Catalan LPs from (0, 0) to (n, m)) .

Proposition 5.0.8. (a) We have Ln,m = Ln−1,m + Ln,m−1 for any n ∈ Z and m ∈ Z
satisfying n ≥ m and (n, m) 6= (0, 0).
(b) If n ∈ N and m ∈ N satisfy n < m, then Ln,m = 0.
(c) If n ∈ N and m ∈ N satisfy n ≥ m − 1, then
   
n+m n+m
Ln,m = − .
m m−1
Enumerative Combinatorics: class notes page 366

(d) If n ∈ N and m ∈ N satisfy n ≥ m − 1, then

n+1−m n+m
 
Ln,m = .
n+1 m

(e) For any n ∈ N, we have

 
1 2n
Ln,n = .
n+1 n

Proof. Main idea: Recall the notion of “upsided tuples” from MT3 exercise 4. There
is a bijection
{Catalan LPs from (0, 0) to (n, m)}
→ {upsided (n + m) -tuples (i1 , i2 , . . . , in+m ) with i1 + i2 + · · · + in+m = n} ,
( v0 , v1 , . . . , v n + m ) 7 → ( i1 , i2 , . . . , i n + m ) , where ik = [vk − vk−1 = (1, 0)] .
Thus, the claims of (b) and (c) follow from MT3 exercise 4. The claim of (d) follows
by simple algebra from (c). The claim of (e) follows by applying (d) to m = n. The
claim of (a) follows by looking at the last step of a Catalan LP.
   
1 2n 2n
Definition 5.0.9. For any n ∈ N, the number Ln,n = = −
  n+1 n n
2n
is called the n-th Catalan number and is denoted by Cn .
n−1
A whole book ([Stanle15]) has been written about Catalan numbers. A few other
places where they appear are:
• Let us say that a permutation σ ∈ Sn is 123-avoiding if there exist no i < j < k
in [n] such that σ (i ) < σ ( j) < σ (k ). Then, the # of 123-avoiding permutations
σ ∈ Sn is Cn .
• Consider all possible ways to fully parenthesize a given expression a1 + a2 +
· · · + an . For example, for n = 4, these are
( a1 + a2 ) + ( a3 + a4 ) , ( a1 + ( a2 + a3 )) + a4 ,
a1 + (( a2 + a3 ) + a4 ) , a1 + ( a2 + ( a3 + a4 )) ,
(( a1 + a2 ) + a3 ) + a4 .
For any n ∈ N, there are Cn−1 ways to do this.
• Fix an n ≥ 3, and a convex n-gon Gn . How many ways are there to triangulate
Gn (i.e., to subdivide Gn into triangles whose vertices are vertices of Gn ) ?
The answer is Cn−2 .
There are several variations on Catalan numbers and Catalan LPs, such as r-
Catalan numbers.
Enumerative Combinatorics: class notes page 367

6. Generating functions (introduction)

We have already seen a few generating functions. Here are two more examples.
Basic idea: Any sequence ( a0 , a1 , a2 , . . .) of numbers gives rise to a “power se-
ries” a0 + a1 x + a2 x2 + a3 x3 + · · · , called its “generating function”. What exactly
it is is a question that would take us a while (see [Loehr11, Chapter 7 (in the 1st
edition)] or [Niven69] or [19s, Chapter 7]), but let us just calculate with these gen-
erating functions as if they were polynomials in x. (Of course, they are not literally
polynomials in x, since they can have infinitely many nonzero coefficients.)
Example 1. Recall the Fibonacci sequence ( f 0 , f 1 , f 2 , . . .) with

f 0 = 0, f 1 = 1, f n = f n −1 + f n −2 .

Consider its gf (= generating function)

F ( x ) = f 0 + f 1 x + f 2 x2 + · · ·
= 0 + 1x + 1x2 + 2x3 + 3x4 + 5x5 + 8x6 + · · · .

Then,

F ( x ) = f 0 + f 1 x + f 2 x2 + f 3 x3 + f 4 x4 + · · ·
2
{z1x} + |( f 1 + f 0 ) x + ( f 2 + f 1{z
= 0| + ) x3 + ( f 3 + f 2 ) x4 + · · ·
}
=x
=( f 1 x2 + f 2 x3 + f 3 x4 +··· )
+( f 0 x2 + f 1 x3 + f 2 x4 +··· )
   
2 3 4 2 3 4
= x + f1 x + f2 x + f3 x + · · · + f0 x + f1 x + f2 x + · · ·
| {z }
= f 0 x + f 1 x2 + f 2 x3 + f 3 x4 +···
(since f 0 =0)
   
2 3 4 2 3 4
= x + f0 x + f1 x + f2 x + f3 x + · · · + f0 x + f1 x + f2 x + · · ·
| {z } | {z }
= xF ( x ) = x2 F ( x )
= x + xF ( x ) + x2 F ( x ) .

Solving this equation for F ( x ), we get

x x
F (x) = 2
=
1−x−x (1 − φx ) (1 − ψx )
√ √
1+ 5 1− 5
where φ = and ψ = are the “golden ratios”. Applying partial
2 2
fraction decomposition to the RHS, we obtain

x 1 1 1 1
F (x) = =√ · −√ · .
(1 − φx ) (1 − ψx ) 5 1 − φx 5 1 − ψx
Enumerative Combinatorics: class notes page 368

1
Now, what are the coefficients of for α ∈ C ?
1 − αx
Well:
1
= 1 + x + x2 + x3 + · · · ,
1−x
because
 
(1 − x ) 1 + x + x 2 + x 3 + · · ·
   
= 1 + x + x2 + x3 + · · · − x 1 + x + x2 + x3 + · · ·
   
2 3 2 3
= 1 + x + x + x + · · · − x + x + x + · · · = 1.
If we substitute αx for x here, we obtain
1
= 1 + αx + (αx )2 + (αx )3 + · · ·
1 − αx
= 1 + αx + α2 x2 + α3 x3 + · · · .
Thus, our formula for F ( x ) becomes
1 1 1 1
F (x) = √ · −√ ·
5 1 − φx 5 1 − ψx
1   1  
= √ · 1 + φx + φ2 x2 + φ3 x3 + · · · − √ · 1 + ψx + ψ2 x2 + ψ3 x3 + · · ·
5 5
1−1 φ−ψ 2 2
φ −ψ 2 φ −ψ 3 3 3
= √ + √ x+ √ x + √ x +··· .
5 5 5 5
Now, comparing coefficients before x n , we get
φn − ψn
fn = √ for each n ∈ N.
5
This is exactly Binet’s formula. Unlike the first time we saw it, we now have a
motivated “proof” of it.
However, of course, this is only a proof if we can explain
• what a power series is;
• what x is;
• why we can divide by power series like 1 − x − x2 ;
• why we can substitute αx for x into a power series;
• why we can expand infinite sums;
• ...
This is done nicely in [Loehr11, Chapter 7 (in the 1st edition)] and [19s, Chapter
7] and [21s] and in most detailed textbooks on abstract algebra.
[18f, Chapter 8] gives a quick overview. So does [Niven69]. Many applications
are found in [Wilf09].
Enumerative Combinatorics: class notes page 369

7. Solutions and references to the exercises

This chapter contains solutions to the exercises not solved in the text, or references
to places where these solutions can be found.

7.1. Reference to solution to Exercise 1.3.1

Exercise 1.3.1 is [17f-hw2s, Exercise 1 (a)].

7.2. Solution to Exercise 1.3.2

Solution to Exercise 1.3.2. Theorem 1.3.11 (applied to k = n) yields
     
n n n
= = (since n − n = 0)
n n−n 0
=1 (by (44)) .
This solves Exercise 1.3.2.

7.3. Reference to solution to Exercise 1.3.3

Parts (a), (b), (c), (d) and (e) of Exercise 1.3.3 appear (with solution) in [17f-hw1s,
Exercise 1] and in [Grinbe15, Exercise 3.12]. Thus, it remains to solve part (f):
Solution to Exercise 1.3.3 (f). (f) We shall show that

[A1 ∧ A2 ∧ · · · ∧ An ] = [A1 ] · [A2 ] · · · · · [An ] (254)

for each n ∈ {0, 1, . . . , k}.

[Proof of (254): We shall prove (254) by induction on n:
Induction base: The conjunction A1 ∧ A2 ∧ · · · ∧ A0 of no statements is a true state-
ment (by definition), and thus its truth value is 1. In other words, [A1 ∧ A2 ∧ · · · ∧ A0 ] =
1. Comparing this with [A1 ] · [A2 ] · · · · · [A0 ] = (empty product) = 1, we obtain
[A1 ∧ A2 ∧ · · · ∧ A0 ] = [A1 ] · [A2 ] · · · · · [A0 ]. In other words, (254) holds for n = 0.
This completes the induction base.
Induction step: Let m ∈ {0, 1, . . . , k} be positive. Assume that (254) holds for
n = m − 1. We must prove that (254) holds for n = m as well.
We have assumed that (254) holds for n = m − 1. In other words, we have

[A1 ∧ A2 ∧ · · · ∧ Am−1 ] = [A1 ] · [A2 ] · · · · · [Am−1 ] .

But the statement A1 ∧ A2 ∧ · · · ∧ Am is equivalent to (A1 ∧ A2 ∧ · · · ∧ Am−1 ) ∧
Am 226 . Hence, Exercise 1.3.3 (a) (applied to A = A ∧ A ∧ · · · ∧ A and
1 2 m
226 Makesure you understand why this is true even for m = 1 ! (In this case, it comes helpful that
the conjunction A1 ∧ A2 ∧ · · · ∧ A0 of no statements was defined to be a true statement, so that
(A1 ∧ A2 ∧ · · · ∧ A0 ) ∧ A1 is equivalent to A1 .)
Enumerative Combinatorics: class notes page 370

B = (A1 ∧ A2 ∧ · · · ∧ Am−1 ) ∧ Am ) yields

[A1 ∧ A2 ∧ · · · ∧ Am ]
= [(A1 ∧ A2 ∧ · · · ∧ Am−1 ) ∧ Am ]
= [A1 ∧ A2 ∧ · · · ∧ Am−1 ] [Am ]
| {z }
=[A1 ]·[A2 ]·····[Am−1 ]
(by Exercise 1.3.3 (c), applied to A = A1 ∧ A2 ∧ · · · ∧ Am−1 and B = Am )
= ([A1 ] · [A2 ] · · · · · [Am−1 ]) · [Am ]
= [A1 ] · [A2 ] · · · · · [Am ] .
In other words, (254) holds for n = m. This completes the induction step. Thus,
the induction proof of (254) is complete.]
Now, (254) (applied to n = k) yields [A1 ∧ A2 ∧ · · · ∧ Ak ] = [A1 ] · [A2 ] · · · · · [Ak ].
This solves Exercise 1.3.3 (f).

7.4. Solution to Exercise 1.3.4

In order to solve Exercise 1.3.4, we must prove Lemma 1.3.17 without using Theo-
rem 1.3.12. Here is such a proof:
Second proof of Lemma 1.3.17. Forget that we fixed n and k. We shall prove Lemma
1.3.17 by induction on n:
Induction base: For each k ∈ Z, we have
 
0
= [ k = 0] (by Lemma 1.3.14)
k
∈ {0, 1} (since [A] ∈ {0, 1} for any statement A)
⊆ N.
 
0
Thus, we have ∈ N for each k ∈ Z. In other words, Lemma 1.3.17 holds for
k
n = 0. Thus, the induction base is complete.
Induction step: Let m be a positive integer. Assume that Lemma 1.3.17 holds for
n = m − 1. We must prove that Lemma 1.3.17 holds for n = m.
Let k ∈ Z. Recall that Lemma 1.3.17 holds for n = m − 1(by ourassumption).
m−1
Thus, Lemma 1.3.17 (applied to m − 1 instead of n) yields ∈ N. Also,
k 
m−1

Lemma 1.3.17 (applied to m − 1 and k − 1 instead of n and k) yields ∈N
k−1
(because Lemma 1.3.17 holds for n = m − 1). Now, Theorem 1.3.8 (applied to
n = m) yields
m−1 m−1
     
m
= + ∈N
k k−1 k
| {z } | {z }
∈N ∈N
Enumerative Combinatorics: class notes page 371

(since the sum of two elements of N is always an element

  of N).
m
Forget that we fixed k. We thus have proven that ∈ N for each k ∈ Z. In
k
other words, Lemma 1.3.17 holds for n = m. This completes the induction step.
Thus, the inductive proof of Lemma 1.3.17 is finished.

7.5. Reference to solution to Exercise 1.3.5

Proposition 1.3.33 is proved in [17f-hw2s, Exercise 1 (b)], in [Grinbe15, Exercise 3.2
(b)] and in [18f-hw3s, Exercise 3 (a)].

7.6. Solution to Exercise 1.3.6

First solution to Exercise 1.3.6. If n = 0, then Exercise 1.3.6 holds227 . Hence, for the
rest of this solution, we can WLOG assume that n 6= 0. Assume this. Combining
n 6= 0 with n ∈ N, we obtain n ∈ {1, 2, 3, . . .}, so that n − 1 ∈ N.
We have
n     n    
n n n here, we have split off the
∑k k =0 0 +∑k k addend for k = 0 from the sum
k =0 | {z } k=1 | {z } 
=0 n n−1

=
k k−1
(by Proposition 1.3.36,
applied to n and k
instead of m and n)
n n
n n−1 n−1
   
= ∑ k· = ∑n
k k−1
k=1 |{z} k =1
k−1
=n
n −1
n−1
   
here, we have substituted k
= ∑n
k =0
k for k − 1 in the sum
n −1
n−1
 
=n ∑ k (by an application of (32))
k =0
| {z }
=2n −1
(by Corollary 1.3.27,
applied to n−1 instead of n)

= n · 2n −1 .

This solves Exercise 1.3.6.

     
n n 0 0 0
227 Proof. Assume that n = 0. Then, ∑ k = ∑ k = 0 = 0. Comparing this with
 k = 0 k k = 0 k 0
n n
n−1 = 0, we obtain = n · 2n−1 . Thus, we have solved Exercise 1.3.6 under the
n · 2 ∑ k
|{z} k =0 k
=0
assumption that n = 0.
Enumerative Combinatorics: class notes page 372

Second solution to Exercise 1.3.6. The following solution uses the same “Little Gauss”
trick that you have seen in the second proof of Theorem 1.2.1, except that we are
now using the summation sign. We have
n   n   n   n   n  
n n n n n
2· ∑ k = ∑k +∑k = ∑k + ∑ (n − k)
k =0
k k =0
k k =0
k k =0
k k =0
n−k
| {z  }
n
=
k
(by Theorem 1.3.11)

(here, we have substituted n − k for k in the second sum)

n   n   n     
n n n n
= ∑k + ∑ (n − k) = ∑ k + (n − k)
k =0
k k =0
k k =0 |
k k
{z   }
n
=(k+(n−k))
k
(by an application of (31))
n   n   n  
n n n
= ∑ (k + (n − k)) = ∑n =n ∑
k =0
| {z } k k =0
k k =0
k
=n | {z }
=2n
(by Corollary 1.3.27)

(by an application of (32))

= n · 2n .
 
n n
Dividing both sides of this equality by 2, we obtain ∑ k = n · 2| n{z
/2} = n · 2n−1 .
k =0 k
=2n −1
This solves Exercise 1.3.6 again.

7.7. Solution to Exercise 1.4.1

In order to solve Exercise 1.4.1, we must prove Theorem 1.4.1.
First proof of Theorem 1.4.1. Forget that we fixed n and S.
We proceed by induction on n:
Induction base: If S is a 0-element set, then S = ∅ and thus

(# of subsets of S) = (# of subsets of ∅)
 
since the empty set ∅ has only one subset,
=1
namely the empty subset ∅
 
= 20 since 20 = 1 .

In other words, for any 0-element set S, we have (# of subsets of S) = 20 . Thus,

Theorem 1.4.1 is proven for n = 0. This completes the induction base.
Enumerative Combinatorics: class notes page 373

Induction step: Let m ∈ N. Assume (as the induction hypothesis) that Theorem
1.4.1 holds for n = m.
Let S be an (m + 1)-element set. We must prove that

(# of subsets of S) = 2m+1 .

The set S is an (m + 1)-element set; thus, its size is |S| = m + 1 ≥ 1 > 0. Hence,
the set S is nonempty, i.e., there exists a t ∈ S. Fix such a t.
Now, we shall call a subset of S

• red if it contains t, and

• green if it does not contain t.

Thus, each subset of S is either red or green (but not both at the same time). Thus,
by the sum rule, we have

(# of subsets of S) = (# of red subsets of S) + (# of green subsets of S) .

We shall now compute the two addends on the right hand side.
The green subsets of S are the subsets of S that don’t contain t. In other words,
the green subsets of S are exactly the subsets of S \ {t}. Hence,

(# of green subsets of S) = (# of subsets of S \ {t}) . (255)

But S is an (m + 1)-element set, and thus S \ {t} is an m-element set (since t ∈ S).
Hence, our induction hypothesis shows that Theorem 1.4.1 can be applied to m and
S \ {t} instead of n and S. We thus obtain

(# of subsets of S \ {t}) = 2m .

Hence, (255) becomes

(# of green subsets of S) = (# of subsets of S \ {t}) = 2m . (256)

What about the red subsets?

Informally, a similar argument works: The red subsets of S are not exactly the
subsets of S \ {t}, but they “correspond to” the latter in a specific way. Namely,
since the red subsets of S are required to contain t, the only “information” that a
red subset of S really “carries” is what other elements (other than t) it contains. In
other words, each red subset R of S “corresponds to” the subset R \ {t} of S \ {t}.
Formally, this can be restated as follows: The map

f : {red subsets of S} → {subsets of S \ {t}} ,

R 7→ R \ {t}
Enumerative Combinatorics: class notes page 374

is well-defined and is a bijection.228 Therefore, the bijection principle (applied to

the bijection f ) yields

|{red subsets of S}| = |{subsets of S \ {t}}| .

In other words,

(# of red subsets of S) = (# of subsets of S \ {t}) . (257)

But S \ {t} is an m-element set. Hence, our induction hypothesis shows that
Theorem 1.4.1 can be applied to m and S \ {t} instead of n and S. We thus obtain

(# of subsets of S \ {t}) = 2m .

Hence, (257) becomes

(# of red subsets of S) = (# of subsets of S \ {t}) = 2m . (258)

Now, we can finish our computation of (# of subsets of S) that we started above:

(# of subsets of S) = (# of red subsets of S) + (# of green subsets of S)

| {z } | {z }
=2m =2m
(by (258)) (by (256))

= 2m + 2m = 2 · 2m = 2m +1 .

Now, forget that we fixed S. We thus have shown that every (m + 1)-element set
S satisfies
(# of subsets of S) = 2m+1 .
In other words, Theorem 1.4.1 holds for n = m + 1. This completes the induction
step, and thus the proof of Theorem 1.4.1.

7.8. Solution to Exercise 1.4.2

In order to solve Exercise 1.4.2, we need to prove Proposition 1.4.13.
Proof of Proposition 1.4.13. The set S has size |S| = m (since it is an m-element set).
Thus, it is finite. Hence, Proposition 1.4.11 yields that there exists a unique tuple
(s1 , s2 , . . . , sk ) of integers satisfying {s1 , s2 , . . . , sk } =
S and s1 < s2 < · ·
· < sk .
Consider this tuple, and denote it by t1 , t2 , . . . , t p . Thus, t1 , t2 , . . . , t p is the
unique tuple of integers satisfying t1 , t2 , . . . , t p = S and t1 < t2 < · · · < t p . From
t1 < t2 <  · · · < t p , we conclude that the p numbers t1 , t2 , . . . , t p are distinct. Hence,
 set t1 , t2 , . . . , t p has size p. In other words, t1 , t2 , . . . , t p = p. In view of
the
t1 , t2 , . . . , t p = S, this rewrites as |S| = p. Comparing this with |S| = m, we
obtain p = m.
228 This
can be showed in the same way as we proved the corresponding claim in our proof of
Theorem 1.3.12.
Enumerative Combinatorics: class notes page 375


Recall that the tuple t1 , t2 , . . . , t p is a p-tuple of integers satisfying t1 , t2 , . . . , t p =
S and t1 < t2 < · · · < t p . In view of p = m, we can rewrite this as follows: The
tuple (t1 , t2 , . . . , tm ) is an m-tuple of integers satisfying {t1 , t2 , . . . , tm } = S and
t1 < t2 < · · · < tm . Hence, there exists at least one m-tuple (s1 , s2 , . . . , sm ) of in-
tegers satisfying {s1 , s2 , . . . , sm } = S and s1 < s2 < · · · < sm (namely, the m-tuple
(t1 , t2 , . . . , tm )). Furthermore, there exists at most one such m-tuple (s1 , s2 , . . . , sm )
229 . Combining the claims of the preceding two sentences, we conclude that there

exists a unique m-tuple (s1 , s2 , . . . , sm ) of integers satisfying {s1 , s2 , . . . , sm } = S and

s1 < s2 < · · · < sm . This proves Proposition 1.4.13.

7.9. Reference to solution to Exercise 1.4.6

Exercise 1.4.6 is [17f-hw1s, Exercise 5].

7.10. Reference to solution to Exercise 1.4.7

Exercise 1.4.7 is [17f-hw1s, Exercise 6].

7.11. Reference to solution to Exercise 1.4.8

Exercise 1.4.8 is [17f-hw3s, Exercise 3].

7.12. Reference to solution to Exercise 1.5.1

Parts (a), (b) and (c) of Exercise 1.5.1 are [17f-hw3s, Exercise 1].
Part (d) of Exercise 1.5.1 is [19f-hw2s, Exercise 2].
Part (e) of Exercise 1.5.1 is [19f-mt1s, Exercise 2].

7.13. Reference to solution to Exercise 1.5.2

Exercise 1.5.2 is [17f-hw3s, Exercise 4].

7.14. Reference to solution to Exercise 2.1.1

Exercise 2.1.1 (a) is essentially [18f-hw2s, Exercise 4]. (More precisely, [18f-hw2s,
Exercise 4] is the particular case of Exercise 2.1.1 (a) where n is assumed to be a
positive integer. However, the first two solutions to [18f-hw2s, Exercise 4] apply

229 Proof.


Recall that t1 , t2 , . . . , t p is the unique tuple (s1 , s2 , . . . , sk ) of integers satisfying
{s1 , s2 , . . . , sk } = S and s1 < s2 < · · · < sk . Hence, if (s1 , s2 , . . . , sk ) is any tuple of integers sat-

isfying {s1 , s2 , . . . , sk } = S and s1 < s2 < · · · < sk , then (s1 , s2 , . . . , sk ) must equal t1 , t2 , . . . , t p .
Applying this to k = m, we conclude that if (s1 , s2 , . . . , sm ) is any m-tuple of integers satisfying

{s1 , s2 , . . . , sm } = S and s1 < s2 < · · · < sm , then (s1 , s2 , . . . , sm ) must equal t1 , t2 , . . . , t p . Thus,
there exists at most one such m-tuple (s1 , s2 , . . . , sm ).
Enumerative Combinatorics: class notes page 376

just as well to the general case. Only the third solution to [18f-hw2s, Exercise 4]
really requires n to be a positive integer.)
Let us solve Exercise 2.1.1 (b) now. In order to do so, we need to prove Proposi-
tion 1.3.28 again:
Alternative proof of Proposition 1.3.28 using Exercise 2.1.1 (a). If n = 0, then
n   0    
n k 0 0
∑ (−1) k = ∑ (−1) k = (−
k
1) 0
| {z } 0
=1
k =0 k =0
=1 |{z}
=1
(by (44))
= [ n = 0] (since [n = 0] = 1 (because n = 0)) .

Thus, Proposition 1.3.28 is proved in the case when n = 0. Hence, for the rest of this
proof, we WLOG assume that n 6= 0. Thus, [n = 0] = 0. Also, n is a positive integer
(since n ∈ N and n 6= 0), and thus n − 1 ∈ N. Furthermore, n > n− 1. Hence,
n−1

Proposition 1.3.6 (applied to n − 1 and n instead of n and k) yields = 0.
n
Now, Exercise 2.1.1 (a) (applied to m = n) yields
n
n n−1
   
k n
∑ (−1) k = (−1) n
= 0 = [ n = 0]
k =0 | {z }
=0

(since [n = 0] = 0). This proves Proposition 1.3.28.

Thus, Exercise 2.1.1 (b) is solved.

7.15. Reference to solution to Exercise 2.2.1

Exercise 2.2.1 is [17f-hw2s, Exercise 2].

7.16. Solution to Exercise 2.2.2

In order to solve Exercise 2.2.2, we need to prove Theorem 2.2.2. Let us now do
this:
Proof of Theorem 2.2.2. The set X × Y is finite (since X and Y are finite). Thus, the
set S is finite (since S is a subset of X × Y).
The elements of S are pairs ( x, y) ∈ X × Y with x ∈ X and y ∈ Y (since S is a
subset of X × Y). Let f : S → X be the map that sends each pair ( x, y) ∈ S to its
first entry x ∈ X. Then, Theorem 1.2.5 (applied to W = X) yields

|S| = ∑ (# of s ∈ S satisfying f (s) = w) . (259)

w∈ X
Enumerative Combinatorics: class notes page 377

Now, fix w ∈ X. We have assumed that for each x ∈ X, there are precisely a2
many elements y ∈ Y such that ( x, y) ∈ S. Applying this to x = w, we conclude
that there are precisely a2 many elements y ∈ Y such that (w, y) ∈ S. In other
words,
(# of elements y ∈ Y such that (w, y) ∈ S) = a2 . (260)
We shall next prove that
(# of elements y ∈ Y such that (w, y) ∈ S) = (# of s ∈ S satisfying f (s) = w) .
Indeed, this will follow from the bijection principle, once we have set up the ap-
propriate bijections. Let us do this now:
If y is an element of Y satisfying (w, y) ∈ S, then the pair (w, y) is an s ∈ S
satisfying f (s) = w (since the definition of f yields f ((w, y)) = w). Hence, the
map
α : {y ∈ Y | (w, y) ∈ S} → {s ∈ S | f (s) = w} ,
y 7→ (w, y)
is well-defined.
On the other hand, recall again that the elements of S are pairs ( x, y) ∈ X × Y
with x ∈ X and y ∈ Y. In other words, they are pairs (u, v) ∈ X × Y with u ∈ X
and v ∈ Y. If (u, v) is an element s of S satisfying f (s) = w, then v is an element
y ∈ Y such that (w, y) ∈ S 230 . Hence, the map
β : {s ∈ S | f (s) = w} → {y ∈ Y | (w, y) ∈ S} ,
(u, v) 7→ v
is well-defined.
It is straightforward to see that our maps α and β are mutually inverse231 . Thus,
α and β are bijections. Hence, the bijection principle yields
|{y ∈ Y | (w, y) ∈ S}| = |{s ∈ S | f (s) = w}| .
230 Proof. Let (u, v) be an element s of S satisfying f (s) = w. Thus, (u, v) ∈ S and f ((u, v)) = w. But
the definition of f yields f ((u, v)) = u, so that u = f ((u, v)) = w. Hence, (u, v) = (w, v), so
that (w, v) = (u, v) ∈ S. In other words, v is an element y ∈ Y such that (w, y) ∈ S.
231 Proof. We need to show that α ◦ β = id and β ◦ α = id. The second of these two equalities is easy

to check; we thus only need to prove the first one.

Let s be an element of S satisfying f (s) = w. Write s in the form s = (u, v) for some u ∈ U
and v ∈ V. (We can do this, since every element of S has this form.) Now, f (s) = w rewrites
as f ((u, v)) = w (since s = (u, v)), so that w = f ((u, v)) = u (by the definition of f ). Hence,
(w, v) = (u, v) = s. But the definition of β yields β ((u, v)) = v. This rewrites as β (s) = v (since
s = (u, v)). Applying the map α to both sides of this equality, we obtain
α ( β (s)) = α (v) = (w, v) (by the definition of α)
= s.
Thus, (α ◦ β) (s) = α ( β (s)) = s = id (s).
Forget that we fixed s. We thus have showed that (α ◦ β) (s) = id (s) for each s ∈ S satisfying
f (s) = w. In other words, α ◦ β = id. This completes our proof that α and β are mutually
inverse (since β ◦ α = id).
Enumerative Combinatorics: class notes page 378

In other words,
(# of elements y ∈ Y such that (w, y) ∈ S) = (# of s ∈ S satisfying f (s) = w) .
Comparing this with (260), we find
(# of s ∈ S satisfying f (s) = w) = a2 . (261)
Now, forget that we fixed w. We thus have proved (261) for each w ∈ X. Thus,
(259) becomes
|S| = ∑ (# of s ∈ S satisfying f (s) = w) = ∑ a2 = | X | · a2 = a1 a2 .
w∈ X
| {z } w∈ X |{z}
= a2 = a1
(by (261))

This proves Theorem 2.2.2.

7.17. Reference to solution to Exercise 2.2.3

Exercise 2.2.3 is [17f-hw1s, Exercise 3].

7.18. Solution to Exercise 2.4.1

Solution to Exercise 2.4.1. (a) Informally speaking, the claim of Exercise 2.4.1 (a) is
obvious: The maps f : A → B satisfying f ( A) ⊆ C are “the same as” the maps
from A to C, since their values belong to C; thus, the # of the former maps equals
the # of the latter maps, which is |C || A| (by Theorem 2.4.1, applied to C instead of
B).
But from a rigorous point of view, this argument is not precise; the maps f :
A → B satisfying f ( A) ⊆ C are not literally maps from A to C, even though their
values belong to C (unless, of course, C = B). In fact, a map (as defined nowadays)
“knows” what its domain and its codomain are; thus, two maps with different
codomains cannot be equal, even if they have the same domain and the same value
on each element of this domain.
Thus, we need to slightly refine our argument. If f : A → B is a map satisfying
f ( A) ⊆ C, then we can define a map

f |C : A → C,
a 7→ f ( a)

(because each a ∈ A satisfies f ( a) ∈ f ( A) ⊆ C). This map f |C may be called

the corestriction of f to C. (Unlike a restriction, it has the same domain as A but a
different codomain. Informally, you can think of f |C as being the map f “wearing
a tighter cloak”.) Thus, the map
{maps f : A → B satisfying f ( A) ⊆ C } → {maps from A to C } ,
f 7 → f |C (262)
Enumerative Combinatorics: class notes page 379

is well-defined. Conversely, if g is any map from A to C, then we can define a map

g | B : A → B,
a 7→ g ( a)

(because each a ∈ A satisfies g ( a
) ∈ C ⊆ B), and this map g | B is a map f : A → B

satisfying f ( A) ⊆ C (since g | B ( A) = g ( A) ⊆ C). Thus, the map
{maps from A to C } → {maps f : A → B satisfying f ( A) ⊆ C } ,
g 7→ g | B (263)
is well-defined. Note that neither of the two maps (262) and (263) changes the
values of the map they are being applied to; all they change is the codomain. Thus,
it is clear that the two maps (262) and (263) are mutually inverse; hence, they are
bijections. Thus, the bijection principle yields
|{maps f : A → B satisfying f ( A) ⊆ C }| = |{maps from A to C }| .
In other words,

(# of maps f : A → B satisfying f ( A) ⊆ C ) = (# of maps from A to C ) = |C || A|

(by Theorem 2.4.1, applied to C, | A| and |C | instead of B, m and n). This solves
Exercise 2.4.1 (a).
(b) In the above solution to Exercise 2.4.1 (a), we have defined a map f |C : A →
C for each map f : A → B satisfying f ( A) ⊆ C. Thus, in particular, the map
f |C : A → C is defined for each map f : A → B satisfying f ( A) = C (because
f ( A) = C implies f ( A) ⊆ C). Moreover, it is easy to see that if f : A → B is a map
satisfying f ( A) = C, then the map f |C : A → C is surjective (because its image is
f |C ( A) = f ( A) = C). Hence, the map

{maps f : A → B satisfying f ( A) = C } → {surjective maps from A to C } ,

f 7 → f |C (264)
is well-defined. Conversely, if g is any surjective map from A to C, then the map
g | B can be defined as in our above solution to Exercise 2.4.1 (a); this latter map g | B
is a map f : A → B satisfying f ( A) = C (since g | B ( A) = g ( A) = C (because g

is surjective)). Thus, the map

{surjective maps from A to C } → {maps f : A → B satisfying f ( A) = C } ,
g 7→ g | B (265)
is well-defined. It is easy to see that the two maps (264) and (265) are mutually
inverse232 ; hence, they are bijections. Thus, the bijection principle yields
|{maps f : A → B satisfying f ( A) = C }| = |{surjective maps from A to C }| .
232 Indeed, this can be proven just as in our above solution to Exercise 2.4.1 (a).
Enumerative Combinatorics: class notes page 380

In other words,

(# of maps f : A → B satisfying f ( A) = C ) = (# of surjective maps from A to C ) .

This solves Exercise 2.4.1 (b).

7.19. Solution to Exercise 2.4.2

Solution to Exercise 2.4.2. We need to prove the equalities (168) and (169) rigorously.
The equality (168) can be proved either using the sum rule or using the product
rule; the equality (169) can only be proved using the sum rule233 . Thus, in order to
be maximally consistent, we shall prove both equalities using the sum rule.
[Proof of (168): If f : [m] → [n] is a red surjection, then f (m) is some element of
[n] (since m ∈ {1, 2, . . . , m} = [m]). Thus, the sum rule yields

(# of red surjections f : [m] → [n])

= ∑ (# of red surjections f : [m] → [n] such that f (m) = i ) . (266)
i ∈[n]

But let i ∈ [n]. Then, the map

{red surjections f : [m] → [n] such that f (m) = i } → {surjections from [m − 1] to [n]} ,
f 7 → f | [ m −1]

is well-defined (because if f : [m] → [n] is a red surjection, then its restriction

f |[m−1] : [m − 1] → [n] is “still” a surjection), and is a bijection234 . Thus, the
bijection principle yields

(# of red surjections f : [m] → [n] such that f (m) = i )

= (# of surjections from [m − 1] to [n])
= sur (m − 1, n) (267)

(by (162), applied to m − 1 instead of m).

Now, forget that we fixed i. Hence, we have proved (267) for each i ∈ [n]. Thus,

233 since the options available for f (1) , f (2) , . . . , f (m − 1) in our proof of (169) depend on the choice
of f (m)
234 Indeed, its inverse is the map

{surjections from [m − 1] to [n]} → {red surjections f : [m] → [n] such that f (m) = i } ,
g 7→ (the extension of g to [m] that sends m to i ) .

It is straightforward to see that this is well-defined.

Enumerative Combinatorics: class notes page 381

(266) becomes

(# of red surjections f : [m] → [n])

= ∑ (# of red surjections f : [m] → [n] such that f (m) = i )
i ∈[n]
| {z }
=sur(m−1,n)
(by (267))

= ∑ sur (m − 1, n) = |[n]| · sur (m − 1, n) = n · sur (m − 1, n) .

i ∈[n]
|{z}
=n

This formally proves (168).]

[Proof of (169): If f : [m] → [n] is a green surjection, then f (m) is some element
of [n]. Thus, the sum rule yields

(# of green surjections f : [m] → [n])

= ∑ (# of green surjections f : [m] → [n] such that f (m) = i ) . (268)
i ∈[n]

But let i ∈ [n]. Thus, [n] \ {i } is an (n − 1)-element set (since [n] is an n-element
set). Also, [m − 1] is an (m − 1)-element set. Now, the map235

{green surjections f : [m] → [n] such that f (m) = i }

→ {surjections from [m − 1] to [n] \ {i }} ,
[n]\{i }
f 7 → f | [ m −1]

is well-defined (because if f : [m] → [n] is a green surjection, then its restriction

f |[m−1] has image [n] \ {i }, and thus can be regarded as a surjective map from
[m − 1] to [n] \ {i }), and is a bijection236 . Thus, the bijection principle yields

(# of green surjections f : [m] → [n] such that f (m) = i )

= (# of surjections from [m − 1] to [n] \ {i })
= (# of surjective maps from [m − 1] to [n] \ {i })
= sur (m − 1, n − 1) (269)

(by Proposition 2.4.11, applied to m − 1, n − 1, [m − 1] and [n] \ {i }).

235 Here, [n]\{i }

f |[m−1] means the restriction f |[m−1] , regarded as a map from [m − 1] to [n] \ {i }. (This is
well-defined, because if f : [m] → [n] is a green surjection such that f (m) = i, then the image of
f |[m−1] is [n] \ {i }.)
236 Indeed, its inverse is the map

{surjections from [m − 1] to [n] \ {i }} → {green surjections f : [m] → [n] such that f (m) = i } ,
g 7→ (the extension of g to [m] that sends m to i ) .

It is straightforward to see that this is well-defined.

Enumerative Combinatorics: class notes page 382

Now, forget that we fixed i. Hence, we have proved (269) for each i ∈ [n]. Thus,
(268) becomes
(# of green surjections f : [m] → [n])
= ∑ (# of green surjections f : [m] → [n] such that f (m) = i )
i ∈[n]
| {z }
=sur(m−1,n−1)
(by (269))

= ∑ sur (m − 1, n − 1) = |[n]| · sur (m − 1, n − 1) = n · sur (m − 1, n − 1) .

i ∈[n]
|{z}
=n
This formally proves (169).]
Thus, Exercise 2.4.2 is solved.

7.20. Solution to Exercise 2.4.3

In order to solve Exercise 2.4.3, we need to prove Corollary 2.4.15.
Proof of Corollary 2.4.15. (a) We shall prove Corollary 2.4.15 (a) by induction on n:
Induction base: Proposition 2.4.12 (a) (applied to m = 0) yields sur (0, 0) =
[0 = 0] = 1 = 0! (since 0! = 1). In other words, Corollary 2.4.15 (a) holds for
n = 0. This completes the induction base.
Induction step: Let k be a positive integer. Assume that Corollary 2.4.15 (a) holds
for n = k − 1. We must prove that Corollary 2.4.15 (a) holds for n = k.
We have assumed that Corollary 2.4.15 (a) holds for n = k − 1. In other words,
sur (k − 1, k − 1) = (k − 1)!.
We have k − 1 ∈ N (since k is a positive integer) and k − 1 < k. Hence, Proposi-
tion 2.4.12 (f) (applied to m = k − 1 and n = k) yields sur (k − 1, k ) = 0.
Now, Proposition 2.4.14 (applied to m = k and n = k) yields
 

sur (k, k ) = k · sur (k − 1, k ) + sur (k − 1, k − 1) = k · (k − 1)! = (k − 1)! · k.

 
| {z } | {z }
=0 =(k−1)!

Comparing this with

k! = (k − 1)! · k (by Proposition 1.3.2, applied to n = k) ,
we obtain sur (k, k ) = k!. In other words, Corollary 2.4.15 (a) holds for n = k. This
completes the induction step. Thus, Corollary 2.4.15 (a) is proven by induction.
(b) We shall prove Corollary 2.4.15 (b) by induction on m:
Induction base: It is easy to see that the integer sur (0, n) is a multiple of n! for all
n ∈ N 237 . In other words, Corollary 2.4.15 (b) holds for m = 0. This completes
237 Proof.Let n ∈ N. We need to prove that sur (0, n) is a multiple of n!. We have sur (0, n) = [n = 0]
(by Proposition 2.4.12 (d), applied to k = n). Hence, if n 6= 0, then sur (0, n) = [n = 0] = 0 (since
n 6= 0) is clearly a multiple of n!. Thus, our proof is complete if n 6= 0. Hence, for the rest of this
proof, we WLOG assume that n = 0. Thus, n! = 0! = 1, so that every integer is a multiple of n!.
Hence, in particular, sur (0, n) is a multiple of 0!. Qed.
Enumerative Combinatorics: class notes page 383

the induction base.

Induction step: Let k be a positive integer. Assume that Corollary 2.4.15 (b) holds
for m = k − 1. We must prove that Corollary 2.4.15 (b) holds for m = k. In other
words, we must prove that the integer sur (k, n) is a multiple of n! for all n ∈ N.
So let n ∈ N. We must prove that the integer sur (k, n) is a multiple of n!. If
n = 0, then this is obvious238 . Thus, for the rest of this proof, we WLOG assume
that n 6= 0. Thus, n is a positive integer. Proposition 2.4.14 (applied to m = k) thus
yields
sur (k, n) = n · (sur (k − 1, n) + sur (k − 1, n − 1)) .
Also, Proposition 1.3.2 yields n! = (n − 1)! · n.
But we have assumed that Corollary 2.4.15 (b) holds for m = k − 1. Thus, Corol-
lary 2.4.15 (b) (applied to k − 1 instead of m) yields that the integer sur (k − 1, n) is
a multiple of n!. Hence, sur (k − 1, n) = n! · a for some integer a. Consider this a.
Likewise, Corollary 2.4.15 (b) (applied to k − 1 and n − 1 instead of m and n)
yields that the integer sur (k − 1, n − 1) is a multiple of (n − 1)!. Hence, sur (k − 1, n − 1) =
(n − 1)! · b for some integer b. Consider this b.
Now,
 

sur (k, n) = n · sur (k − 1, n) + sur (k − 1, n − 1) = n · (n! · a + (n − 1)! · b)

 
| {z } | {z }
=n!· a =(n−1)!·b
= n · n! · a + n · (n − 1)! ·b = n · n! · a + n! · b = n! · (na + b) .
| {z }
=(n−1)!·n=n!

This shows that the integer sur (k, n) is a multiple of n!.

Forget that we fixed n. We thus have shown that the integer sur (k, n) is a multi-
ple of n! for all n ∈ N. In other words, Corollary 2.4.15 (b) holds for m = k. This
completes the induction step. Thus, Corollary 2.4.15 (b) is proven by induction.
Thus, Exercise 2.4.3 is solved.

7.21. Solution to Exercise 2.4.4

In order to solve Exercise 2.4.4, we need to prove Theorem 2.4.17. Before we do so,
let us state a simple corollary of the binomial formula:

Corollary 7.21.1. Let x ∈ R and n ∈ N. Then,

n −1  
n j
n n
( x + 1) − x = ∑ j
x.
j =0

238 Proof.Assume that n = 0. Thus, n! = 0! = 1. Hence, every integer is a multiple of n!. Thus, in
particular, sur (k, n) is a multiple of n!.
Enumerative Combinatorics: class notes page 384

Proof of Corollary 7.21.1. Theorem 1.3.24 (applied to y = 1) yields

n   n   n  
n k n−k n k n j
( x + 1) = ∑
n
x 1|{z} = ∑ x =∑ x
k =0
k k =0
k j =0
j
=1
(here, we have renamed the summation index k as j)
n −1    
n j n
= ∑ x + xn
j =0
j n
| {z }
=1
(by Exercise 1.3.2)

(here, we have split off the addend for j = n from the sum)
n −1  
n j
= ∑ x + xn .
j =0
j

n −1
 
n n j
Subtracting xn from this equality, we find ( x + 1) − xn = ∑ x . This proves
j =0 j
Corollary 7.21.1.
Proof of Theorem 2.4.17. We shall prove Theorem 2.4.17 by strong induction over m:
Induction step: Let q ∈ N. Assume that Theorem 2.4.17 holds for all m < q. We
must now prove that Theorem 2.4.17 holds for m = q.
We have assumed that Theorem 2.4.17 holds for all m < q. In other words, we
have
n  
n −i n m
sur (m, n) = ∑ (−1) i (270)
i =0
i
for every m ∈ N and n ∈ N satisfying m < q.
Now, let n ∈ N. We are going to prove that
n  
n q
sur (q, n) = ∑ (−1) n −i
i
i . (271)
i =0

[Proof of (271): If q = 0, then (271) is easy to check239 . Hence, for the rest of this
proof, we WLOG assume that q 6= 0. Hence, q ∈ {1, 2, 3, . . .} (since q ∈ N), so that
q > 0. Since q 6= 0, we have [q = 0] = 0.

239 Proof. Assume that q = 0. Thus, sur (q, n) = sur (0, n) = [n = 0] (by Proposition 2.4.12 (d),
Enumerative Combinatorics: class notes page 385

If n = 0, then (271) is also easy to check240 . Hence, for the rest of this proof, we
WLOG assume that n 6= 0. Hence, n ∈ {1, 2, 3, . . .} (since n ∈ N), so that n > 0.
Hence, n − 1 ∈ N. Therefore, for every m ∈ N satisfying m < q, we have
n −1
n−1 m
 
sur (m, n − 1) = ∑ (−1) (n−1)−i
i
i (272)
i =0

(by (270) (applied to n − 1 instead of n)).

Applying Proposition 2.4.13 to m = q, we find
q
  q −1  
q q
sur (q, n) = ∑ · sur (q − j, n − 1) = ∑ · sur ( j, n − 1) .
j =1
j j =0
j

applied to k = n). Comparing this with

n   n  
n q n −i n
∑ (−1) n −i
i
i = ∑ (−1)
i
i0
|{z} (since q = 0)
i =0 i =0 =1
n   n  
n n
= ∑ (−1)n−i = ∑ (−1)n−k
i =0
i k =0
k
| {z }
 
n
=
n−k
(by Theorem 1.3.11)

(here, we have renamed the summation index i as k)

n   n    
n n here, we have substituted k
= ∑ (−1)n−k = ∑ (−1)k
k =0
n−k k =0
k for n − k in the sum
= [ n = 0] (by Proposition 1.3.28) ,
 
n
n −i n q
we obtain sur (q, n) = ∑ (−1) i . Thus, we have proven (271) under the assumption that
i =0 i
q = 0.
240 Proof. Assume that n = 0. Thus, sur ( q, n ) = sur ( q, 0) = [ q = 0] (by Proposition 2.4.12 (a), applied

to m = q). Thus, sur (q, n) = [q = 0] = 0. Comparing this with

n   0  
n q 0− i 0 q
∑ (−1) n −i
i
i = ∑ (−1)
i
i (since n = 0)
i =0 i =0
 
0 q
= (−1)0−0 0 = 0q = 0 (since q > 0) ,
| {z } 0
=1 |{z}
=1
 
n
n −i n q
we obtain sur (q, n) = ∑ (−1) i . Thus, we have proven (271) under the assumption that
i =0 i
n = 0.
Enumerative Combinatorics: class notes page 386

Hence,
q −1  
q
sur (q, n) = ∑ j
· sur ( j, n − 1)
j =0
| {z  } 
n −1 n−1 j
= ∑ (−1)(n−1)−i i
i =0 i
(by (272) (applied to m= j)
(since j≤q−1<q and j∈N))
q −1  q −1 n −1  
q n −1 (n−1)−i n − 1 j (n−1)−i n − 1 j
    
q
= ∑ j
· ∑ (−1)
i
i = ∑ ∑
j
(−1)
i
i
j =0 i =0 j =0 i =0
| {z }
n −1 q −1
=∑ ∑
i =0 j =0

n −1 q −1 
n−1 j
  
q
= ∑∑ j
(−1)(n−1)−i
i
i
i =0 j =0
n −1  q −1  
(n−1)−i n − 1

q j
= ∑ (−1) ∑ i. (273)
i =0
i j =0
j

Recall that n − 1 ∈ N.Therefore, Proposition 1.3.6 (applied to n − 1 and n instead

n−1

of n and k) shows that = 0 (since n > n − 1).
n
Now, every g ∈ Z satisfies
n   n    
n −i g q n −i g q n −0 g
∑ (−1) i
i = ∑ (−1)
i
i + (−1)
0
0q
|{z}
i =0 i =1 =0
(since q>0)
(here, we have split off the addend for i = 0 from the sum)
n    
n −i g q n −0 g
= ∑ (−1) i + (−1) 0
i =1
i 0
| {z }
=0
n  
n −i g q
= ∑ (−1) i . (274)
i =1
i
Enumerative Combinatorics: class notes page 387

Applying this to g = n, we obtain

n   n  
n −i n q n
∑ (−1) i
i = ∑ (−1)n−i
i
iq
i =0 i =1 |{z }
n−1 n−1
 
= +
i−1 i
(by Theorem 1.3.8, applied to k=i)
n
n−1 n−1

  
= ∑ (−1) n −i
+ iq
i =1
i − 1 i
n n
n −i n − 1 q n −i n − 1 q
   
= ∑ (−1) i + ∑ (−1) i . (275)
i =1
i−1 i =1
i
But
n
n − 1 q n −1 n−1
   
∑ (−1) n −i
i−1
i = ∑ (−1) n−(i +1)
( i + 1) − 1
( i + 1) q
i =1 i =0
| {z }
=(−1)n−i−1
|  {z  }
n−1
=
i
(here, we have substituted i + 1 for i in the sum)
n −1
n − i −1 n − 1
 
= ∑ (−1) ( i + 1) q . (276)
i =0
i
Also, (274) (applied to g = n − 1) yields
n n
n −i n − 1 q n −i n − 1 q
   
∑ (−1) i
i = ∑ (−1)
i
i ,
i =0 i =1

and thus we have

n n
n −i n − 1 q n −i n − 1 q
   
∑ (−1) i
i = ∑ (−1)
i
i
i =1 i =0
n −1
n−1 q n−n n − 1
   
= ∑ (−1) n −i
i + (−1) nq
i =0
| {z } i n
n − i −1 | {z }
=−(−1)
=0
(here, we have split off the addend for i = n from the sum)
n −1    n − 1
= ∑ − (−1) n − i −1
iq + (−1)n−n 0nq
i =0
i | {z }
=0
n −1    n − 1
= ∑ − (−1)n−i−1
i
iq
i =0
n −1
n − i −1 n − 1 q
 
= − ∑ (−1) i . (277)
i =0
i
Enumerative Combinatorics: class notes page 388

Now, (275) becomes

n  
n −i n q
∑ (−1) i
i
i =0
n n
n −i n − 1 q n −i n − 1 q
   
= ∑ (−1) i + ∑ (−1) i
i =1
i−1 i =1
i
| {z  } | {z  }
n−1 n−1 q

n −1 n −1
= ∑ (−1)n−i−1 ( i +1) q =− ∑ (−1)n−i−1 i
i =0 i i =0 i
(by (276)) (by (277))
n −1 n −1  !
n−1 n − i −1 n − 1 q
  
= ∑ (−1) n − i −1
(i + 1) + − ∑ (−1)
q
i
i =0
i i =0
i
n −1 n −1
n − i −1 n − 1 n − i −1 n − 1 q
   
= ∑ (−1) (i + 1) − ∑ (−1)
q
i
i =0
i i =0
i
n −1
n − i −1 n − 1
 
= ∑ (−1) ( i + 1) q − i q


i =0
| {z } i | {z
  }
(n−1)−i
=(−1) q −1 q j
=∑ i
j =0 j
(by Corollary 7.21.1, applied to q
and i instead of n and x)
n −1  q −1  
n−1

q
= ∑ (−1) (n−1)−i
i ∑ j ij.
i =0 j =0

Comparing this with (273), we obtain

n  
n q
sur (q, n) = ∑ (−1) n −i
i
i .
i =0

This proves (271).]  

n q n
n −i
Now, forget that we fixed n. We thus have proven that sur (q, n) = ∑ (−1) i
i =0 i
for every n ∈ N. In other words, Theorem 2.4.17 holds for m = q. This completes
the induction step. Thus, Theorem 2.4.17 is proved.
Thus, Exercise 2.4.4 is solved.241

241 This solution has been taken from [17f-hw2s, Exercise 4].
Enumerative Combinatorics: class notes page 389

7.22. Solution to Exercise 2.4.5

Solution to Exercise 2.4.5. (a) Theorem 2.4.17 yields

sur (m, n)
n     n  
n m n −0 n n m
= ∑ (−1) n −i
i
i = (−1)
0
0 m
|{z} + ∑ (−1) n −i
i
i
i =0 =0 i =1
(since m is positive)

(here, we have split off the addend for i = 0 from the sum)
  n   n  
n −0 n n −i n m n
= (−1) 0 + ∑ (−1) i = ∑ (−1) n −i
im
0 i i |{z}
| {z } i =1 i =1 | {z }  =iim−1
=0 (since m is positive)
n n−1

=
i i−1
(by Proposition 1.3.36,
applied to n and i
instead of m and n)
n n
n n − 1 m −1 n − 1 m −1
   
= ∑ (−1) n −i
ii = ∑ (−1) n
n −i
i
i =1
i i−1 i = 1
i−1
|  {z  }
n − 1 m −1
=n i
i−1
n
n − i n − 1 m −1
 
= n ∑ (−1) i .
i =1
i−1

Comparing this with

n
n − 1 m −1
 
n ∑ (−1) n −i
i−1
i
i =0
| {z }
n − 1 m −1 n n − 1 m −1
 
n −0 n −i
=(−1) 0 + ∑ (−1) i
0−1 i =1 i−1
(here, we have split off the addend for i =0 from the sum)
 
 
n
n−1 n − 1 m −1 
    
= n (−1)n−0 0m−1 + ∑ (−1)n−i i
 
0−1 i−1

i =1
 
 | {z } 
=0
∈N)
(by (43), since 0−1/
 
n
n − 1 m −1 
 
= n (−1)n−0 0 · 0m−1 + ∑ (−1)n−i i
| {z } i =1 i−1
=0
n
n − i n − 1 m −1
 
= n ∑ (−1) i ,
i =1
i − 1
Enumerative Combinatorics: class notes page 390

we obtain
n
n − 1 m −1
 
sur (m, n) = n ∑ (−1) n −i
i .
i =0
i−1
This solves Exercise 2.4.5 (a).  
m
(b) The definition of the Stirling number (in Remark 2.4.16) yields
n

1 n
   
m 1 n
= sur (m, n) /n! = sur (m, n) = · ∑ (−1) n −i
im
n n! | {z  } n! i=0 i
n m
| {z }
n
n −i n!
= ∑ (−1) i
i =0 i =
(by Theorem 2.4.17) i! · (n − i )!
(by Theorem 1.3.9,
applied to k=i)

1 n n! 1 n
1
= · ∑ (−1)n−i im = · n! ∑ (−1)n−i im
n! i=0 i! · (n − i )! n!
| {z } i=0 i! ( n − i ) !
=1
n n
1 im
= ∑ (−1)n−i i! (n − i )!
im = ∑ (−1)n−i
i! (n − i )!
.
i =0 i =0

This solves Exercise 2.4.5 (b). 242

7.23. Solution to Exercise 2.5.1

Solution to Exercise 2.5.1. Fix k ∈ N and m ∈ N. We must prove the equality (171)
rigorously.
Our informal proof of this equality involved a sequence of three choices, each
depending on the previous one. The easiest way to formalize such an argument
is by reframing it as two applications of the sum rule (one corresponding to the
choice of i = | f ([m])|, and one corresponding to the choice of f ([m])). Thus, we
obtain the following proof:
[Proof of (171): If f : [m] → [k ] is any map, then | f ([m])| ∈ {0, 1, . . . , m} 243 .
Hence, the sum rule yields

(# of maps f : [m] → [k])

= ∑ (# of maps f : [m] → [k] satisfying | f ([m])| = i ) . (278)
i ∈{0,1,...,m}

242 This solution to Exercise 2.4.5 has been taken from [17f-hw3s, Exercise 2].
243 Proof. Let f : [m] → [k ]. Then, [m] = {1, 2, . . . , m} and thus f ([m]) = f ({1, 2, . . . , m}) =
{ f (1) , f (2) , . . . , f (m)}. But the set { f (1) , f (2) , . . . , f (m)} has at most m elements (since
f (1) , f (2) , . . . , f (m) are m elements). In other words, |{ f (1) , f (2) , . . . , f (m)}| ≤ m. In view of
f ([m]) = { f (1) , f (2) , . . . , f (m)}, this rewrites as | f ([m])| ≤ m. Hence, | f ([m])| ∈ {0, 1, . . . , m}
(since | f ([m])| is clearly a nonnegative integer).
Enumerative Combinatorics: class notes page 391

Now, fix i ∈ {0, 1, . . . , m}. If f : [m] → [k ] is any map satisfying | f ([m])| = i, then
f ([m]) is an i-element subset of [k ] (since f ([m]) ⊆ [k ] and | f ([m])| = i). Thus, the
sum rule yields

(# of maps f : [m] → [k] satisfying | f ([m])| = i )

= ∑ (# of maps f : [m] → [k] satisfying | f ([m])| = i and f ([m]) = I ) .
I is an i-element
subset of [k]
(279)

Now, fix an i-element subset I of [k ]. Then, | I | = i. Hence, each map f : [m] → [k ]

satisfying f ([m]) = I automatically satisfies | f ([m])| = i (since f ([m]) = I leads to
| f ([m])| = | I | = i). Hence, the condition “| f ([m])| = i” in
“(# of maps f : [m] → [k ] satisfying | f ([m])| = i and f ([m]) = I )” is redundant. Thus,

(# of maps f : [m] → [k] satisfying | f ([m])| = i and f ([m]) = I )

= (# of maps f : [m] → [k] satisfying f ([m]) = I )
= (# of surjective maps from [m] to I )
(by Exercise 2.4.1 (b) (applied to A = [m] , B = [k] and C = I))
= sur (m, i ) (280)

(by Proposition 2.4.11, applied to n = i, A = [m] and B = I).

Now, forget that we fixed I. We thus have proved (280) for each i-element subset
I of [k ]. Thus, (279) becomes

(# of maps f : [m] → [k] satisfying | f ([m])| = i )

= ∑ (# of maps f : [m] → [k] satisfying | f ([m])| = i and f ([m]) = I )
I is an i-element
| {z }
subset of [k] =sur(m,i )
(by (280))

= ∑ sur (m, i ) = (# of i-element subsets of [k ]) · sur (m, i )

I is an i-element
| {z 
 }
subset of [k] k
=
i
(by Theorem 1.3.12,
applied to k, i and [k] instead of n, k and S)
 
k
= · sur (m, i ) . (281)
i

Now, forget that we fixed i. We thus have proved (281) for each i ∈ {0, 1, . . . , m}.
Enumerative Combinatorics: class notes page 392

Thus, (278) becomes

(# of maps f : [m] → [k])

= ∑ (# of maps f : [m] → [k] satisfying | f ([m])| = i )
i ∈{0,1,...,m}
|  {z }
k
= ·sur(m,i )
i
(by (281))
  m  
k k
= ∑ i
· sur (m, i ) = ∑
i
· sur (m, i ) .
i ∈{0,1,...,m} i =0
| {z }
m
=∑
i =0

This proves (171).]

Thus, Exercise 2.5.1 is solved.

7.24. Solution to Exercise 2.6.1

In order to solve Exercise 2.6.1, we need to formalize our proof of Theorem 1.3.37.
Here is one way to do so:
First proof of Theorem 1.3.37 for x ∈ N (formal version). Let us first show the follow-
ing:

Claim 1: We have
v  
u−v
  
u v
= ∑
n k =0
k n−k
for any u ∈ R and v ∈ N.

[Proof of Claim 1: We shall prove

 Claim 1 by induction on v.
u−0
  
u 0 0
Induction base: We have = ∑ for any u ∈ R 244 . In other
n k =0 k n − k
words, Claim 1 holds for v = 0. This completes the induction base.
Induction step: Let w ∈ N. Assume that Claim 1 holds for v = w. We must prove
that Claim 1 holds for v = w + 1.

244 Proof. Let u ∈ R. Then,

0
u−0 u−0 u−0
         
0 0 u
∑ k n−k = 0 n−0 = n−0 = n .
k =0 |{z}
=1

u−0
    
u 0 0
In other words, = ∑ , qed.
n k =0 k n−k
Enumerative Combinatorics: class notes page 393

Let u ∈ R. We have assumed that Claim 1 holds for v = w. Thus, we have

w  
u−w
   
u w
= ∑
n k =0
k n−k
| {z  }
u−w−1 u−w−1
 
= +
n−k−1 n−k
(by Theorem 1.3.8, applied to u−w and n−k instead of n and k)
w   
u−w−1 u−w−1
  
w
= ∑ +
k =0
k n−k−1 n−k
w
u−w−1 u−w−1
     
w w
= ∑ +
k =0
k n−k−1 k n−k
w w
u−w−1 u−w−1
     
w w
= ∑ +∑ . (282)
k =0
k n − k − 1 k =0
k n − k
 
w
On the other hand, (43) (applied to w and −1 instead of n and k) yields =0
  − 1
w
(since −1 ∈/ N). Furthermore, w + 1 > w and thus = 0 (by Proposition
w+1
1.3.6, applied to w and w + 1 instead of n and k).
But each k ∈ N satisfies
( w + 1) − 1 ( w + 1) − 1
     
w+1
= +
k k−1 k
(by Theorem 1.3.8, applied to w + 1 instead of n)
   
w w
= + (283)
k−1 k

(since (w + 1) − 1 = w). Hence,

w +1
u − ( w + 1)
   
w+1
∑ k n−k
k =0
 | {z }  |  {z }
w w u−w−1
= + =
k−1 k n−k
(by (283)) (since u−(w+1)=u−w−1)
w+1 
u−w−1

   
w w
= ∑ +
k =0
k−1 k n−k
w +1
u−w−1 u−w−1
     
w w
= ∑ +
k =0
k−1 n−k k n−k
w +1   w+1  
u−w−1 u−w−1
 
w w
= ∑ + ∑ . (284)
k =0
k−1 n−k k =0
k n−k
Enumerative Combinatorics: class notes page 394

We intend to show that the right hand sides of the equalities (282) and (284) are
equal. To that aim, we transform the sums on the right hand side of the latter. Let
us begin with the first sum:
w +1 
u−w−1
 
w
∑ k−1 n−k
k =0
 w +1 
u−w−1 u−w−1
    
w w
= + ∑
0−1 n−0 k =1
k−1 n−k
| {z }
w
= =0
−1
(here, we have split off the addend for k = 0 from the sum)
w +1  w 
u−w−1 u−w−1
    
w w
= ∑ = ∑
k =1
k − 1 n − k k =0 |
( k + 1) − 1 n − ( k + 1)
{z  } |  {z }
w u−w−1
= =
k n−k−1
(since (k+1)−1=k) (since n−(k+1)=n−k−1)
(here, we have substituted k + 1 for k in the sum)
w
u−w−1
  
w
= ∑ . (285)
k =0
k n−k−1

Let us next transform the second sum:

w+1  
u−w−1

w
∑ k n−k
k =0
w  
u−w−1 u−w−1
   
w w
= +∑
w+1 n − ( w + 1) k =0
k n−k
| {z }
=0
(here, we have split off the addend for k = w + 1 from the sum)
w
u−w−1
  
w
= ∑ . (286)
k =0
k n−k
Enumerative Combinatorics: class notes page 395

Thus, (284) becomes

w +1 
u − ( w + 1)
 
w+1
∑ k n−k
k =0
w +1  w+1  
u−w−1 u−w−1
  
w w
= ∑ + ∑
k =0
k−1 n−k k =0
k n−k
| {z } | {z }
u−w−1 u−w−1
    
w w w w
=∑ =∑
k =0 k n−k−1 k =0 k n−k
(by (285)) (by (286))
w w  
u−w−1 u−w−1
   
w w
= ∑ +∑ .
k =0
k n − k − 1 k =0
k n − k

Comparing this with (282), we obtain

  w +1 
u − ( w + 1)
 
u w+1
= ∑ .
n k =0
k n − k

Now, forget that we fixed u. We thus have proved that

  w +1 
u − ( w + 1)
 
u w+1
= ∑ for any u ∈ R.
n k =0
k n−k

In other words, Claim 1 holds for v = w + 1. This completes the induction step.
Thus, Claim 1 is proven by induction.]
Now, let x ∈ N and y ∈ R. Applying Claim 1 to u = x + y and v = x, we get
x   x  
( x + y) − x
   
x+y x x y
= ∑ = ∑ (287)
n k =0
k n−k k =0
k n−k

(since ( x + y) − x = y).
Now, recall that x ∈ N. Hence, every k ∈ N satisfying k > x must satisfy
 
x
=0 (288)
k
  
x y
(by Proposition 1.3.6, applied to x instead of n) and thus = 0. Thus,
k n−k
| {z }
   =0
x y
in the infinite sum ∑ , every addend with k > x is 0. Hence, this
k ∈N k n−k
infinite sum has only finitely many nonzero addends (namely, the addends with
Enumerative Combinatorics: class notes page 396

k ≤ x). Thus, it is well-defined. Moreover,

         
x y x y x y
∑ k n−k = ∑ k n−k
+ ∑
k n−k
k ∈N k∈N; k∈N; | {z }
k≤ x k> x
=0
(by (288))
|{z}
= ∑
k∈{0,1,...,x }
x
=∑
k =0
x      x   
x y y x y
= ∑ + ∑ 0 = ∑ .
k =0
k n−k k∈N;
n−k k =0
k n−k
|k> x {z }
=0

Comparing this with (287), we obtain

    
x+y x y
= ∑ . (289)
n k ∈N
k n−k

On the other hand, n ∈ N. If k is an integer satisfying k > n, then n − k < 0 and

/ N and therefore
thus n − k ∈  
y
=0 (290)
n−k
(by (43), applied to y and n − k instead of n and k). Thus,
        
x y x y x y
∑ k n−k = ∑ k n−k
+ ∑
k n−k
k ∈N k∈N; k∈N;
k≤n
| {z }
k>n
=0
(by (290))
|{z}
= ∑
k ∈{0,1,...,n}
n
=∑
k =0
n
     n   
x y x x y
= ∑ + ∑ 0= ∑ .
k =0
k n − k k∈N;
k k =0
k n − k
|k>n {z }
=0

Hence, (289) becomes

     n   
x+y x y x y
= ∑ = ∑ .
n k ∈N
k n−k k =0
k n−k

Thus, (172) is proved. Hence, (173) follows as well245 . This proves Theorem 2.6.1
for x ∈ N.
Thus, Exercise 2.6.1 is solved.
245 because we have seen in Remark 2.6.2 that the right hand sides of (172) and of (173) are equal
Enumerative Combinatorics: class notes page 397

7.25. Solution to Exercise 2.6.2

In order to solve Exercise 2.6.2, we need to give an alternative proof of Corollary
2.6.3:
Alternative proof of Corollary 2.6.3. We have y ∈ N. Hence, we can apply (172) to y,
x and y instead of x, y and n (since we have proved (172) for x ∈ N already). We
thus obtain
  y    y   
y+x y x y x
= ∑ = ∑
y k =0
k y − k k =0
y−k y − (y − k)
| {z  }
x
=
k
(since y−(y−k)=k)
(here, we have substituted y − k for k in the sum)
y    y   
y x x y
= ∑ = ∑ .
k =0 |
y−k k k = 0
k y−k
  {z }
x y
=
k y−k

But
y     y   
x y x y
∑ k k
= ∑
k y−k
.
k =0 k =0
|{z} 
y
=
y−k
(by Theorem 1.3.11,
applied to n=y)

Comparing these two equalities, we find

y       
x y y+x x+y
∑ k k = y = y (since y + x = x + y) .
k =0

This proves Corollary 2.6.3 again.

7.26. Solution to Exercise 2.6.3

In order to solve Exercise 2.6.3, we need to solve Exercise 2.1.1 (a) again using
Theorem 1.3.37:
Alternative solution to Exercise 2.1.1 (a) using Theorem 1.3.37. The equality (172) (ap-
plied to n, −1 and m instead of x, y and n) yields
m  
−1
  
n + (−1) n
= ∑ . (291)
m k =0
k m − k
Enumerative Combinatorics: class notes page 398

But every k ∈ {0, 1, . . . , m} satisfies m − k ∈ {0, 1, . . . , m} ⊆ N and thus

−1
 
= (−1)m−k (292)
m−k
(by (47), applied to m − k instead of k). Hence, (291) becomes
m   m  
−1
  
n + (−1) n n
= ∑ = ∑ (−1)m−k
m k =0
k m − k k
| {z } k=0
| {z }
m+k
=(−1)
=(−1)m−k (since m−k≡m+k mod 2)
(by (292))
m   m  
n n
= ∑ (−1) m+k
= ∑ (−1)m (−1)k
k =0
k | {z } k =0 |
k
=(−1)m (−1)k
{z   }
n
=(−1)m (−1)k
k
m   m  
k n k n
= ∑ (−1) (−1)
m
= (−1) ∑ (−1)
m
.
k =0
k k =0
k

In view of n + (−1) = n − 1, this rewrites as

m
n−1
   
k n
= (−1) ∑ (−1)
m
.
m k =0
k

Multiplying both sides of this equality by (−1)m , we find

m m
m n−1
     
k n k n
} k∑
(−1) = (−1) (−1) m m
(−1) = ∑ (−1) .
m | {z =0
k k =0
k
=(−1)m+m =1
(since m+m=2m is even)

Thus, Exercise 2.1.1 (a) is solved again.

7.27. Solution to Exercise 2.6.4

In order to solve Exercise 2.6.4, we must give an alternative proof of Theorem 1.3.29
using Proposition 2.6.13. Let us do this:
Alternative proof of Theorem 1.3.29. Forget that we fixed k. For each x ∈ N, we have
n   
n−k
  
n+1 k
= ∑ (by Proposition 2.6.13, applied to y = 0)
x+1 k =0
x 0
| {z }
=1
(by (44),
applied to n−k
instead of n)
n          
k 0 1 2 n
= ∑ = + + +···+
k =0
x x x x x
Enumerative Combinatorics: class notes page 399

and thus          
0 1 2 n n+1
+ + +···+ = .
x x x x x+1
Renaming x as k in this statement, we obtain the following: For each k ∈ N, we
have          
0 1 2 n n+1
+ + +···+ = .
k k k k k+1
Thus, Theorem 1.3.29 is proved again.

7.28. Reference to solution to Exercise 2.6.5

Exercise 2.6.5 appears (with solution) in [Grinbe15, Proposition 3.32 (d)]. (To be
more precise, [Grinbe15, Proposition 3.32 (d)] is only the particular case of Exer-
cise 2.6.5 when x, y ∈ Z; but the general case can be proved by the exact same
argument.)

7.29. Reference to solution to Exercise 2.6.6

Exercise 2.6.6 appears (with solution) in [Grinbe15, Proposition 3.32 (e)]. (To be
more precise, [Grinbe15, Proposition 3.32 (e)] is only the particular case of Exercise
2.6.6 when y ∈ Z; but the general case can be proved by the exact same argument.)

7.30. Reference to solution to Exercise 2.6.7

Exercise 2.6.7 appears (with solution) in [Grinbe15, Proposition 3.37] (with x re-
named as m). (To be more precise, [Grinbe15, Proposition 3.37] is only the partic-
ular case of Exercise 2.6.7 when x ∈ Q; but the general case can be proved by the
exact same argument.)

7.31. Reference to solution to Exercise 2.8.1

Exercise 2.8.1 appears (with solution) in [Grinbe15, §7.30]. More precisely:

• Exercise 2.8.1 (a) is [Grinbe15, Lemma 7.49] (with n and x renamed as m and
n).

• Exercise 2.8.1 (b) follows from [Grinbe15, Exercise 3.22]. (Namely, rename
n as m in [Grinbe15, Exercise 3.22], and substitute n for X in the resulting
equality between polynomials; then, you obtain precisely the claim of Exercise
2.8.1 (b).)

• Exercise 2.8.1 (c) is [Grinbe15, Corollary 7.53].

Enumerative Combinatorics: class notes page 400

7.32. Reference to solution to Exercise 2.9.1

Proposition 2.9.12 appears (with proof) in [18f-hw3s, Proposition 6.4] (with S and
T renamed as G and S).

7.33. Solution to Exercise 2.9.2

Before we solve Exercise 2.9.2, let us recall two fundamental facts from elementary
set theory (often called de Morgan’s laws):

Proposition 7.33.1. Let U, X and Y be three sets. Then:

(a) We have U \ ( X ∪ Y ) = (U \ X ) ∩ (U \ Y ).
(b) We have U \ ( X ∩ Y ) = (U \ X ) ∪ (U \ Y ).

We will not prove Proposition 7.33.1, as it is the domain of introductory textbooks

on proofs and sets; suffice to say that both parts of the proposition can be proved
in straightforward ways using basic logic and the fact that two sets P and Q satisfy
P = Q if and only if ( P ⊆ Q and Q ⊆ P).
We will use the following generalization of Proposition 7.33.1 to multiple sets:

Proposition 7.33.2. Let k be a positive integer. Let U be a set. Let P1 , P2 , . . . , Pk

be any k sets. Then:
(a) We have

U \ ( P1 ∪ P2 ∪ · · · ∪ Pk ) = (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ Pk ) .

(b) We have

U \ ( P1 ∩ P2 ∩ · · · ∩ Pk ) = (U \ P1 ) ∪ (U \ P2 ) ∪ · · · ∪ (U \ Pk ) .

Proof of Proposition 7.33.2. (a) We shall prove Proposition 7.33.2 (a) by induction on
k:
Induction base: It is easy to see that Proposition 7.33.2 (a) holds for k = 1 246 .
This completes the induction base.
Induction step: Let m be a positive integer. Assume that Proposition 7.33.2 (a)
holds for k = m. We must prove that Proposition 7.33.2 (a) holds for k = m + 1.
Let U be a set. Let P1 , P2 , . . . , Pm+1 be any m + 1 sets. We have assumed that
Proposition 7.33.2 (a) holds for k = m. Thus, Proposition 7.33.2 (a) (applied to
246 Proof. If U is a set, and if P1 , P2 , . . . , P1 are any 1 sets, then

U \ ( P1 ∪ P2 ∪ · · · ∪ P1 ) = U \ P1 = (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ P1 )
| {z }
= P1

(since (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ P1 ) = U \ P1 ). In other words, Proposition 7.33.2 (a) holds

for k = 1.
Enumerative Combinatorics: class notes page 401

k = m) yields

U \ ( P1 ∪ P2 ∪ · · · ∪ Pm ) = (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ Pm ) .

Now,

U \ ( P1 ∪ P2 ∪ · · · ∪ Pm+1 )
| {z }
=( P1 ∪ P2 ∪···∪ Pm )∪ Pm+1
= U \ (( P1 ∪ P2 ∪ · · · ∪ Pm ) ∪ Pm+1 )
= (U \ ( P1 ∪ P2 ∪ · · · ∪ Pm )) ∩ (U \ Pm+1 )
| {z }
=(U \ P1 )∩(U \ P2 )∩···∩(U \ Pm )
(by Proposition 7.33.1 (a), applied to X = P1 ∪ P2 ∪ · · · ∪ Pm and Y = Pm+1 )
= ((U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ Pm )) ∩ (U \ Pm+1 )
= (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ Pm+1 ) .

Now, forget that we fixed U and P1 , P2 , . . . , Pm+1 . We thus have shown that if U
is a set, and if P1 , P2 , . . . , Pm+1 are any m + 1 sets, then

U \ ( P1 ∪ P2 ∪ · · · ∪ Pm+1 ) = (U \ P1 ) ∩ (U \ P2 ) ∩ · · · ∩ (U \ Pm+1 ) .

In other words, Proposition 7.33.2 (a) holds for k = m + 1. This completes the
induction step. Hence, Proposition 7.33.2 (a) is proved by induction.
(b) In order to obtain a proof of Proposition 7.33.2 (b), it suffices to make some
simple modifications to our above proof of Proposition 7.33.2 (a) (namely: replace
all “∪” signs by “∩” signs and vice versa, and replace the reference to Proposition
7.33.1 (a) by a reference to Proposition 7.33.1 (b)).
Now, let us solve Exercise 2.9.2:
Solution to Exercise 2.9.2. Let U be the set A1 ∪ A2 ∪ · · · ∪ An . Thus, U is the union
of the n finite sets A1 , A2 , . . . , An , and thus itself a finite set. Moreover, all the n sets
A1 , A2 , . . . , An clearly are subsets of their union A1 ∪ A2 ∪ · · · ∪ An . In other words,
all the n sets A1 , A2 , . . . , An are subsets of U (since U = A1 ∪ A2 ∪ · · · ∪ An ).
For each i ∈ {1, 2, . . . , n}, we define a set Bi by Bi = U \ Ai . Thus, for each
i ∈ {1, 2, . . . , n}, the set Bi is a subset of the set U (since Bi = U \ Ai ), and thus is
finite (since U is finite). Hence, Theorem 2.9.1 (applied to Bi instead of Ai ) yields

| B1 ∪ B2 ∪ · · · ∪ Bn |
n
= ∑ (−1)m−1 ∑ Bi1 ∩ Bi2 ∩ · · · ∩ Bim . (293)
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im
Enumerative Combinatorics: class notes page 402

On the other hand, Theorem 2.9.1 (applied to U instead of Ai ) yields

n
| ∪U∪
U {z· · · ∪ U} = ∑ (−1)m−1 ∑ m
| ∩U∩
U {z· · · ∩ U}
n times m =1 (i1 ,i2 ,...,im )∈[n] ; m times
i1 <i2 <···<im | {z }
=U
(since m≥1)
n
= ∑ (−1)m−1 ∑ |U | .
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im

In view of
| ∪U∪
U {z· · · ∪ U} = U (since n is positive) ,
n times
this rewrites as
n
|U | = ∑ (−1)m−1 ∑ m
|U | . (294)
m =1 (i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im

However, we have the following two statements:

Statement 1: We have

| B1 ∪ B2 ∪ · · · ∪ Bn | = |U | − | A1 ∩ A2 ∩ · · · ∩ An | .

Statement 2: Let m ∈ {1, 2, . . . , n}. Let (i1 , i2 , . . . , im ) ∈ [n]m . Then,

Bi1 ∩ Bi2 ∩ · · · ∩ Bim = |U | − Ai1 ∪ Ai2 ∪ · · · ∪ Aim .

[Proof of Statement 1: Recall that all the n sets A1 , A2 , . . . , An are subsets of U.

Thus, in particular, A1 is a subset of U. In other words, A1 ⊆ U. Hence, A1 ∩
A2 ∩ · · · ∩ An is a subset of U as well (since A1 ∩ A2 ∩ · · · ∩ An ⊆ A1 ⊆ U). Hence,
Theorem 1.4.7 (a) (applied to U and A1 ∩ A2 ∩ · · · ∩ An instead of A and B) yields

|U \ ( A1 ∩ A2 ∩ · · · ∩ An )| = |U | − | A1 ∩ A2 ∩ · · · ∩ An | . (295)

But each i ∈ {1, 2, . . . , n} satisfies Bi = U \ Ai (by the definition of Bi ). Hence,

( B1 , B2 , . . . , Bn ) = (U \ A1 , U \ A2 , . . . , U \ An ) .
Thus,
B1 ∪ B2 ∪ · · · ∪ Bn = (U \ A1 ) ∪ (U \ A2 ) ∪ · · · ∪ (U \ An ) .
But Proposition 7.33.2 (b) (applied to k = n and Pj = A j ) yields

U \ ( A 1 ∩ A 2 ∩ · · · ∩ A n ) = (U \ A 1 ) ∪ (U \ A 2 ) ∪ · · · ∪ (U \ A n ) .
Enumerative Combinatorics: class notes page 403

Comparing these two equalities, we obtain

U \ ( A1 ∩ A2 ∩ · · · ∩ An ) = B1 ∪ B2 ∪ · · · ∪ Bn .

In light of this equality, we can rewrite (295) as

| B1 ∪ B2 ∪ · · · ∪ Bn | = |U | − | A1 ∩ A2 ∩ · · · ∩ An | .

This proves Statement 1.]

[Proof of Statement 2: We have (i1 , i2 , . . . , im ) ∈ [n]m . Thus, i1 , i2 , . . . , im are ele-
ments of [n] = {1, 2, . . . , n}.
Recall that all the n sets A1 , A2 , . . . , An are subsets of U. Hence, in particu-
lar, the sets Ai1 , Ai2 , . . . , Aim are subsets of U (since i1 , i2 , . . . , im are elements of
{1, 2, . . . , n}). Thus, their union Ai1 ∪ Ai2 ∪ · · · ∪ Aim is a subset of U as well (since
a union of subsets of U is always a subset of U). Hence, Theorem 1.4.7 (a) (applied
to U and Ai1 ∪ Ai2 ∪ · · · ∪ Aim instead of A and B) yields


U \ A i1 ∪ A i2 ∪ · · · ∪ A i m = |U | − A i1 ∪ A i2 ∪ · · · ∪ A i m . (296)

But each j ∈ {1, 2, . . . , m} satisfies Bi j = U \ Ai j (by the definition of Bi j ). Hence,

Bi1 , Bi2 , . . . , Bim = U \ Ai1 , U \ Ai2 , . . . , U \ Aim .

Thus,


Bi1 ∩ Bi2 ∩ · · · ∩ Bim = U \ Ai1 ∩ U \ Ai2 ∩ · · · ∩ (U \ Aim ) .
But Proposition 7.33.2 (a) (applied to k = m and Pj = Ai j ) yields




U \ A i1 ∪ A i2 ∪ · · · ∪ A i m = U \ A i1 ∩ U \ A i2 ∩ · · · ∩ (U \ A i m ) .

Comparing these two equalities, we obtain

U \ Ai1 ∪ Ai2 ∪ · · · ∪ Aim = Bi1 ∩ Bi2 ∩ · · · ∩ Bim .

In light of this equality, we can rewrite (296) as

Bi1 ∩ Bi2 ∩ · · · ∩ Bim = |U | − Ai1 ∪ Ai2 ∪ · · · ∪ Aim .

This proves Statement 2.]

Enumerative Combinatorics: class notes page 404

Now, (293) becomes

| B1 ∪ B2 ∪ · · · ∪ Bn |
n
= ∑ (−1)m−1 ∑ m
Bi1 ∩ Bi2 ∩ · · · ∩ Bim
m =1 (i1 ,i2 ,...,im )∈[n] ;
| {z }
i1 <i2 <···<im =|U |−| Ai1 ∪ Ai2 ∪···∪ Aim |
(by Statement 2)
n
∑ (−1)m−1 ∑


= |U | − A i1 ∪ A i2 ∪ · · · ∪ A i m
m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im
| {z }
= ∑ |U |− ∑ | Ai1 ∪ Ai2 ∪···∪ Aim |
(i1 ,i2 ,...,im )∈[n]m ; (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im i1 <i2 <···<im
 
n
∑ (−1)m−1  ∑ ∑
 
=  |U | − A i1 ∪ A i2 ∪ · · · ∪ A i m 

m =1 (i1 ,i2 ,...,im )∈[n]m ; (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im i1 <i2 <···<im
| {z }
=(−1)m−1 ∑ |U |−(−1) m −1
∑ | Ai1 ∪ Ai2 ∪···∪ Aim |
(i1 ,i2 ,...,im )∈[n]m ; (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im i1 <i2 <···<im
 
n
∑ (−1)m−1 ∑ |U | − (−1)m−1 ∑
 
=  A i1 ∪ A i2 ∪ · · · ∪ A i m 

m =1 (i1 ,i2 ,...,im )∈[n]m ; (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im i1 <i2 <···<im
n n
= ∑ (−1) m −1
∑ |U | − ∑ (−1)m−1 ∑ A i1 ∪ A i2 ∪ · · · ∪ A i m
m =1 (i1 ,i2 ,...,im )∈[n]m ; m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im i1 <i2 <···<im
| {z }
=|U |
(by (294))
n
= |U | − ∑ (−1)m−1 ∑ m
A i1 ∪ A i2 ∪ · · · ∪ A i m . (297)
m =1 (i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im

Now, Statement 1 yields

| B1 ∪ B2 ∪ · · · ∪ Bn | = |U | − | A1 ∩ A2 ∩ · · · ∩ An | .
Enumerative Combinatorics: class notes page 405

Solving this equality for | A1 ∩ A2 ∩ · · · ∩ An |, we find

| A1 ∩ A2 ∩ · · · ∩ A n |
= |U | − | B ∪ B2 ∪ · · · ∪ Bn |
| 1 {z }
n
m −1
=|U |− ∑ (−1) ∑
m
| Ai1 ∪ Ai2 ∪···∪ Aim |
m =1 (i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im
(by (297))
 
n
∑ (−1)m−1 ∑
 
= |U | − 
 |U | − A i1 ∪ A i2 ∪ · · · ∪ A i m 

m =1 (i1 ,i2 ,...,im )∈[n]m ;
i1 <i2 <···<im
n
= ∑ (−1)m−1 ∑ m
A i1 ∪ A i2 ∪ · · · ∪ A i m .
m =1 (i1 ,i2 ,...,im )∈[n] ;
i1 <i2 <···<im

This solves Exercise 2.9.2.

7.34. Solution to Exercise 2.9.3

Solution to Exercise 2.9.3. Here are the statements that were left unproved in our
(sketched) proof of Proposition 2.9.14:247

Statement 1: Let σ ∈ SX be a permutation satisfying (σ (i ) = i for each i ∈ I ).

Then, we can define a map e σ : X \ I → X \ I by setting

σ ( p) = σ ( p)
e for each p ∈ X \ I.

Statement 2: Let σ ∈ SX be a permutation satisfying (σ (i ) = i for each i ∈ I ).

Then, the map e σ : X \ I → X \ I defined in Statement 1 is a permutation
of X \ I (and thus belongs to SX \ I ).

Statement 3: Let τ be a permutation of X \ I. Then, we can define a map

τb : X → X by setting
(
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I

Statement 4: Let τ be a permutation of X \ I. Then, the map τb : X → X

defined in Statement 3 is a permutation in SX satisfying (τb (i ) = i for each i ∈ I )
(and thus belongs to {σ ∈ SX | σ (i ) = i for each i ∈ I }).
247 Statement2 shows that the map A is well-defined. Statement 4 shows that the map B is well-
defined. Statement 5 and Statement 6 (when combined) show that the maps A and B are mutu-
ally inverse.
Enumerative Combinatorics: class notes page 406

Statement 5: The maps A and B satisfy A ◦ B = id.

Statement 6: The maps A and B satisfy B ◦ A = id.

Let us now prove these statements:

[Proof of Statement 1: We have σ ( p) ∈ X \ I for each p ∈ X \ I 248 . Hence, we
can define a map e σ : X \ I → X \ I by setting

σ ( p) = σ ( p)
e for each p ∈ X \ I.

This proves Statement 1.]

[Proof of Statement 2: The map σ is a permutation of X (since σ ∈ SX ), thus a
bijection from X to X. Hence, σ is bijective, thus injective and surjective.
We must prove that e σ is a permutation of X \ I. In other words, we must prove
that eσ is a bijection from X \ I to X \ I.
It is easy to see that the map e σ is injective249 and surjective250 . Hence, e σ is
bijective. In other words, e σ is a bijection from X to X. Hence, e σ is a permutation of
X \ I. In other words, e σ belongs to SX \ I . This proves Statement 2.]
(
τ ( p) , if p ∈
/ I;
[Proof of Statement 3: Let p ∈ X. We claim that τ ∈ X.
p, if p ∈ I

248 Proof. Let p ∈ X \ I. We must prove that σ ( p) ∈ X \ I.

Assume the contrary. Thus, σ ( p) ∈ / X \ I. Combining σ ( p) ∈ X with σ ( p) ∈ / X \ I, we obtain
σ ( p) ∈ X \ ( X \ I ) ⊆ I.
Now, recall that (σ (i ) = i for each i ∈ I ). Applying this to i = σ ( p), we obtain σ (σ ( p)) =
σ ( p ).
But the map σ is a permutation of X (since σ ∈ SX ), thus a bijection from X to X. Hence, σ is
bijective, thus injective. In other words, if u and v are two elements of X such that σ (u) = σ (v),
then u = v. Applying this to u = σ ( p) and v = p, we find σ ( p) = p (since σ (σ ( p)) = σ ( p)).
Hence, σ ( p) = p ∈ X \ I, so that σ ( p) ∈ / I. This contradicts σ ( p) ∈ I. This contradiction shows
that our assumption was wrong. Hence, σ ( p) ∈ X \ I is proved.
249 Proof. Let u and v be two elements of X \ I such that e σ (u) = e
σ (v). We shall show that u = v.
The definition of e σ yields eσ (u) = σ (u) and e σ (v) = σ (v). Hence, σ (u) = e σ (u) = e σ (v) =
σ (v). Since σ is injective, we can thus conclude that u = v.
Now, forget that we fixed u and v. We thus have proved that if u and v are two elements of
X \ I such that e σ (u) = e σ (v), then u = v. In other words, the map e σ is injective.
250 Proof. Let p ∈ X \ I. We shall prove that there exists some x ∈ X \ I such that p = e σ ( x ).
Recall that the map σ is injective. Thus, there exists some q ∈ X such that p = σ (q). Consider
this q. Thus, σ (q) = p ∈ / I (since p ∈ X \ I).
If we had q ∈ I, then we would have σ (q) = q (by the assumption “σ (i ) = i for each i ∈ I”,
applied to i = q), which would entail σ (q) = q ∈ I; but this would contradict σ (q) ∈ / I. This,
we cannot have q ∈ I. Hence, we have q ∈ / I. Combining q ∈ X with q ∈ / I, we obtain q ∈ X \ I.
Hence, e σ (q) is well-defined. Moreover, the definition of e σ yields eσ (q) = σ (q) = p. In other
words, p = e σ (q). Hence, there exists some x ∈ X \ I such that p = e σ ( x ) (namely, x = q).
Forget that we fixed p. We thus have proved that for each p ∈ X \ I, there exists some x ∈ X \ I
such that p = e σ ( x ). In other words, eσ is surjective.
Enumerative Combinatorics: class notes page 407

Indeed, if p ∈ I, then this follows from

(
τ ( p) , if p ∈
/ I;
=p (since p ∈ I )
p, if p ∈ I
∈ X.

Hence, for the rest of this proof, we WLOG assume that p ∈ / I. Hence, p ∈ X \ I
(since p ∈ X and p ∈ / I), so that τ ( p) is well-defined (since τ is a permutation of
X \ I). Moreover,
(
τ ( p) , if p ∈
/ I;
= τ ( p) ∈ X \ I (since τ is a permutation of X \ I )
p, if p ∈ I
⊆ X.
(
τ ( p) , if p ∈
/ I;
Thus, ∈ X is proved.
p, if p ∈ I
(
τ ( p) , if p ∈
/ I;
Now, forget that we fixed p. We thus have shown that ∈ X for
p, if p ∈ I
each p ∈ X. Hence, we can define a map τb : X → X by setting
(
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I

This proves Statement 3.]

[Proof of Statement 4: We shall first prove that τb is a permutation of X.
There are two ways to prove this. One way to do so is by showing that τb is both
surjective and injective. This is straightforward but boring. A slicker approach is
by constructing its inverse map. Let us do the latter.
The map τ is a permutation of X \ I, thus a bijection from X \ I to X \ I. Hence,
its inverse τ −1 is well-defined, and is also a bijection from X \ I to X \ I. In other
words, τ −1 is also a permutation of X \ I. Thus, Statement 4 (applied to τ −1 instead
of τ) shows that we can define a map τd −1 : X → X by setting

(
−1 ( p ) =
τ −1 ( p) , if p ∈
/ I;
τd for each p ∈ X.
p, if p ∈ I

Consider this map τd −1 . Note that τ ◦ τ −1 = id and τ −1 ◦ τ = id. We shall now

prove that τb ◦ τd−1 = id and τ d−1 ◦ τ

b = id.  
[Proof of τb ◦ τ −1 = id: Let q ∈ X. We shall show that τb ◦ τd
d −1 ( q ) = q.

Indeed, we are in one of the following two cases:

Case 1: We have q ∈ I.
Enumerative Combinatorics: class notes page 408

Case 2: We have q ∈ / I.
Let us first consider ( Case 1. In this case, we have q ∈ I. Now, the definition of
−1 yields τ −1 ( q ) =
τ −1 (q) , if q ∈ / I;
τd d = q (since q ∈ I). Applying the map τb to
q, if q ∈ I
both sides of this equality, we find
 
(
 −1 τ (q) , if q ∈ / I;
(q) = τb (q) = (by the definition of τb)

τb τd
| {z } q, if q ∈ I
=q
=q (since q ∈ I ) .
     
Hence, τb ◦ τ d− 1 − 1
(q) = τb τ (q) = q. Thus, τb ◦ τ
d d− 1 (q) = q is proven in Case
1.
Let us next consider Case 2. In this case, we have q ∈ / I. Combining q ∈ X
with q ∈ / I, we find q ∈ X \ I. Hence, τ −1 (q) is well-defined (since τ −1 is a map
from from X \ I to X \ I) and satisfies τ −1 (q) ∈ X \ I (for the same reason). From
τ −1 (q) ∈ X \ I, we obtain τ −1 (q) ∈ X and(τ −1 (q) ∈ / I.
− 1
τ (q) , if q ∈ / I;
The definition of τd−1 yields τd−1 ( q ) = = τ −1 (q) (since q ∈
/
q, if q ∈ I
I). Applying the map τb to both sides of this equality, we find
 
 
− 1 −1
τ τ (q) = τ τ (q)
d 
b b
| {z }
=q
(
τ τ −1 ( q ) , if τ −1 (q) ∈


/ I;
= (by the definition of τb)
τ −1 ( q ) , if τ −1 (q) ∈ I
   
= τ τ −1 ( q ) since τ −1 (q) ∈
/I
= q.
     
Hence, τb ◦ τd −1 ( q ) = τ
b τd−1 ( q ) = q. Thus, τ −1 ( q ) = q is proven in Case
b ◦ τd
2.  
We have now proved τb ◦ τd −1 ( q ) = q in each of the two Cases 1 and 2. Thus,
   
τ◦τ
b d− 1 (q) = q always holds. Hence, τ ◦ τb d− 1 (q) = q = id (q).
 
Now, forget that we fixed q. We thus know that τb ◦ τd −1 ( q ) = id ( q ) for each

q ∈ X. In other words, τb ◦ τd −1 = id. The same argument (but with the roles of τ

and τ −1 interchanged251 ) shows that τd −1 ◦ τb = id. Combining these two equalities,

we conclude that the maps τb and τ −1 are mutually inverse. Hence, the map τb is
d

251 This −1 also get interchanged.

means that the roles of τb and τd
Enumerative Combinatorics: class notes page 409

invertible. In other words, τb is a bijection from X to X. In other words, τb is a

permutation of X. In other words, τb ∈ SX .
For each i ∈ I, we have
(
τ (i ) , if i ∈
/ I;
τb (i ) = (by the definition of τb)
i, if i ∈ I
=i (since i ∈ I ) .

Thus, we have shown that (τb (i ) = i for each i ∈ I ). Hence, τb is an element of SX

that satisfies (τb (i ) = i for each i ∈ I ).
In other words, τb belongs to {σ ∈ SX | σ (i ) = i for each i ∈ I }. This proves
Statement 4.]
[Proof of Statement 5: Let τ ∈ SX \ I . We shall show that ( A ◦ B) (τ ) = τ.
Indeed, let p ∈ X \ I. Thus, p ∈ X and p ∈ / I. Now, B (τ ) = τb (by the definition
of B). Also,

( A ◦ B) (τ ) = A ( B (τ )) = B]
(τ ) (by the definition of A) .

Applying both sides of this equality to p, we obtain

   
(( A ◦ B) (τ )) ( p) = B] (τ ) ( p) = ( B (τ )) ( p) by the definition of B](τ )
| {z }
=τb
(
τ ( p) , if p ∈/ I;
= τb ( p) = (by the definition of τb)
p, if p ∈ I
= τ ( p) (since p ∈
/ I) .

Forget that we fixed p. We thus have shown that (( A ◦ B) (τ )) ( p) = τ ( p) for

each p ∈ X \ I. In other words, ( A ◦ B) (τ ) = τ = id (τ ).
Forget that we fixed τ. We thus have shown that ( A ◦ B) (τ ) = id (τ ) for each
τ ∈ SX \ I . In other words, A ◦ B = id. This proves Statement 5.]
[Proof of Statement 6: Let γ ∈ {σ ∈ SX | σ (i ) = i for each i ∈ I }. We shall show
that ( B ◦ A) (γ) = γ.
We have γ ∈ {σ ∈ SX | σ (i ) = i for each i ∈ I }. In other words, γ is a σ ∈
SX such that (σ (i ) = i for each i ∈ I ). In other words, γ is an element of SX and
satisfies
(γ (i ) = i for each i ∈ I ) . (298)
Thus, γ is a permutation of X (since γ ∈ SX ).
Now, let p ∈ X. We shall show that (( B ◦ A) (γ)) ( p) = γ ( p).
We have A (γ) = γe (by the definition of A). Furthermore,

( B ◦ A) (γ) = B ( A (γ)) = \
A ( γ ).
Enumerative Combinatorics: class notes page 410

Applying both sides of this equality to p, we obtain

 
(( B ◦ A) (γ)) ( p) = A (γ) ( p)
\
(
( A (γ)) ( p) , if p ∈
/ I;
= (299)
p, if p ∈ I

(by the definition of \A (γ)).

Now, we are in one of the following two cases:
Case 1: We have p ∈ I.
Case 2: We have p ∈ / I.
Let us first consider Case 1. In this case, we have p ∈ I. Thus, γ ( p) = p (by
(298), applied to i = p). But (299) becomes
(
( A (γ)) ( p) , if p ∈
/ I;
(( B ◦ A) (γ)) ( p) = =p (since p ∈ I )
p, if p ∈ I
= γ ( p) (since γ ( p) = p) .

Hence, (( B ◦ A) (γ)) ( p) = γ ( p) is proved in Case 1.

Let us now consider Case 2. In this case, we have p ∈ / I. Thus, p ∈ X \ I (since
p ∈ X and p ∈ / I). Now, (299) becomes
(
( A (γ)) ( p) , if p ∈
/ I;
(( B ◦ A) (γ)) ( p) = = ( A (γ)) ( p) (since p ∈
/ I)
p, if p ∈ I | {z }
=γ
e
=γ
e ( p) = γ ( p) (by the definition of γ
e) .

Hence, (( B ◦ A) (γ)) ( p) = γ ( p) is proved in Case 2.

We have now proved (( B ◦ A) (γ)) ( p) = γ ( p) in each of the two Cases 1 and 2.
Thus, (( B ◦ A) (γ)) ( p) = γ ( p) always holds.
Forget that we fixed p. We thus have proved that (( B ◦ A) (γ)) ( p) = γ ( p) for
each p ∈ X. In other words, ( B ◦ A) (γ) = γ = id (γ).
Forget that we fixed γ. We thus have proved that ( B ◦ A) (γ) = id (γ) for each
γ ∈ {σ ∈ SX | σ (i ) = i for each i ∈ I }. In other words, B ◦ A = id. This proves
Statement 6.]
We have now proved all six Statements 1, 2, 3, 4, 5 and 6. This fills all the gaps
in our proof of Proposition 2.9.14. Thus, Exercise 2.9.3 is solved.

7.35. Solution to Exercise 2.9.4

In order to solve Exercise 2.9.4, we have to prove parts (a), (b) and (c) of Theorem
1.7.9. In doing so, we can use part (d) of this theorem, since it has already been
proven.
Enumerative Combinatorics: class notes page 411

Proof of Theorem 1.7.9 (b). Let n be an integer such that n ≥ 1. Thus, n is a positive
integer, so that n ∈ N and n − 1 ∈ N. Hence, Theorem 1.7.9 (d) (applied to n − 1
instead of n) yields
n −1
( n − 1) !
Dn−1 = ∑ (−1)k . (300)
k =0
k!
Also, Theorem 1.7.9 (d) yields
n n −1
n! n! n!
Dn = ∑ (−1)k
k!
= (−1)n
n!
+ ∑ (−1)k
k!
k =0 |{z} k =0 |{z}
=1 ( n − 1) ! · n
=
k!
(since Proposition 1.3.2
yields n!=(n−1)!·n)

(here, we have split off the addend for k = n from the sum)
n −1 n −1
k ( n − 1) ! · n ( n − 1) !
= (−1)n + ∑ (− 1 ) = (− 1 ) n
+ ∑ n (−1)k
k =0 | {z k! k!
|k=0
}
( n − 1) !
{z }
=n(−1) k n −1 ( n − 1) !
k! =n ∑ (−1)k
k =0 k!
n −1
( n − 1) !
= (−1)n + n ∑ (−1)k k!
= (−1)n + nDn−1 = nDn−1 + (−1)n .
|k=0 {z }
= Dn −1
(by (300))

This proves Theorem 1.7.9 (b).

Now that we have proved Theorem 1.7.9 (b), we shall derive Theorem 1.7.9 (a)
from it. In doing so, we shall use the following simple lemma ([19f-hw1s, Exercise
5]):

Lemma 7.35.1. Let (u0 , u1 , u2 , . . .) be a sequence of real numbers such that every
integer n ≥ 1 satisfies
un = nun−1 + (−1)n . (301)
Then, un = (n − 1) (un−1 + un−2 ) for each integer n ≥ 2.

Proof of Lemma 7.35.1. Let n ≥ 2 be an integer. Thus, n − 1 ≥ 2 − 1 = 1. Hence,

(301) (applied to n − 1 instead of n) yields

un−1 = (n − 1) u(n−1)−1 + (−1)n−1 = (n − 1) un−2 + − (−1)n

| {z } | {z }
= u n −2 =−(−1)n
n
= (n − 1) un−2 − (−1) .
Enumerative Combinatorics: class notes page 412

Thus,
un−1 + (−1)n = (n − 1) un−2 . (302)
But n ≥ 2 ≥ 1. Hence, (301) yields

un = n
|{z} un−1 + (−1)n = ((n − 1) + 1) un−1 + (−1)n
| {z }
=(n−1)+1 =(n−1)un−1 +un−1
= (n − 1) un−1 + un−1 + (−1)n = (n − 1) un−1 + ( n − 1 ) u n −2
| {z }
=(n−1)un−2
(by (302))
= ( n − 1 ) ( u n −1 + u n −2 ) .

This proves Lemma 7.35.1.

Proof of Theorem 1.7.9 (a). Theorem 1.7.9 (b) shows that every integer n ≥ 1 satisfies
Dn = nDn−1 + (−1)n . Hence, Lemma 7.35.1 (applied to u j = D j ) shows that
Dn = (n − 1) ( Dn−1 + Dn−2 ) for each integer n ≥ 2. This proves Theorem 1.7.9
(a).
It remains to prove Theorem 1.7.9 (c). There are several ways to do this. In partic-
ular, there is a rather simple double-counting argument, which uses the following
concept:

Definition 7.35.2. Let X be a set. Let f : X → X be any map. Then, Fix f shall
denote the set of all fixed points of f .

We will need a fact similar to Proposition 2.9.14:

Proposition 7.35.3. Let X be a set. Let I be a subset of X. Then, there is a bijection

from {σ ∈ SX | Fix σ = I } to {derangements of X \ I } .

Proof of Proposition 7.35.3. Let us first give a sketch of the proof (similarly to the
proof of Proposition 2.9.14 we sketched), and then fill in the details (similarly to
the solution of Exercise 2.9.3):
To each permutation σ ∈ SX satisfying Fix σ = I, we can assign a derangement e σ
of X \ I by letting
σ ( p) = σ ( p)
e for each p ∈ X \ I.
252 This defines a map

A : {σ ∈ SX | Fix σ = I } → {derangements of X \ I } ,
σ 7→ e
σ.
252 Why σ well-defined, and why is it really a derangement of X \ I ? This will follow
is this map e
from Statement 1’ and Statement 2’ further below.
Enumerative Combinatorics: class notes page 413

Conversely, to each derangement τ of X \ I, we can assign a permutation τb ∈ SX

satisfying Fix τb = I by setting
(
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I

253 This defines a map

B : {derangements of X \ I } → {σ ∈ SX | Fix σ = I } ,
τ 7→ τb.

The maps A and B are well-defined and mutually inverse254 . Hence, they are bi-
jections. Thus, there is a bijection from the set {σ ∈ SX | Fix σ = I } to the set
{derangements of X \ I } (namely, A). This proves Proposition 7.35.3, once the de-
tails are filled in.
Let us now fill in the details. Here are the statements that were left unproved in
our above proof:255

Statement 1’: Let σ ∈ SX be a permutation satisfying Fix σ = I. Then, we

can define a map e σ : X \ I → X \ I by setting

σ ( p) = σ ( p)
e for each p ∈ X \ I.

Statement 2’: Let σ ∈ SX be a permutation satisfying Fix σ = I. Then, the

map eσ : X \ I → X \ I defined in Statement 1’ is a derangement of X \ I
(and thus belongs to {derangements of X \ I }).

Statement 3’: Let τ be a derangement of X \ I. Then, we can define a

map τb : X → X by setting
(
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I

Statement 4’: Let τ be a derangement of X \ I. Then, the map τb : X → X

defined in Statement 3’ is a permutation in SX satisfying Fix τb = I (and
thus belongs to {σ ∈ SX | Fix σ = I }).

Statement 5’: The maps A and B satisfy A ◦ B = id.

Statement 6’: The maps A and B satisfy B ◦ A = id.

253 Again, this will be proved below.
254 Again, this will be proved below.
255 Statement 2’ shows that the map A is well-defined. Statement 4’ shows that the map B is well-

defined. Statement 5’ and Statement 6’ (when combined) show that the maps A and B are
mutually inverse.
Enumerative Combinatorics: class notes page 414

Let us now prove these statements:

[Proof of Statement 1’: Let i ∈ I. Then, i ∈ I = Fix σ (since Fix σ = I). This means
that i is a fixed point of σ (since Fix σ was defined as the set of all fixed points of
σ). In other words, σ (i ) = i (by the definition of “fixed point”).
Forget that we fixed i. We thus have shown that σ (i ) = i for each i ∈ I. Thus,
Statement 1 from our above solution to Exercise 2.9.3 shows that we can define a
map e σ : X \ I → X \ I by setting
σ ( p) = σ ( p)
e for each p ∈ X \ I.
This proves Statement 1’.]
[Proof of Statement 2’: The map σ is a permutation of X (since σ ∈ SX ), thus
a bijection from X to X. Moreover, we have σ (i ) = i for each i ∈ I. (Indeed,
this can be proven just as in our proof of Statement 1’ above.) Thus, Statement
2 from our above solution to Exercise 2.9.3 shows that e σ is a permutation of X \
I. Moreover, this permutation e 256
σ has no fixed points ; in other words, e σ is a
derangement of X \ I (by the definition of “derangement”). In other words, e σ
belongs to {derangements of X \ I }. This proves Statement 2’.]
[Proof of Statement 3’: The map τ is a derangement of X \ I, and thus is a permu-
tation of X \ I (since any derangement is a permutation). Hence, Statement 3 from
our above solution to Exercise 2.9.3 shows that we can define a map τb : X → X by
setting (
τ ( p) , if p ∈
/ I;
τb ( p) = for each p ∈ X.
p, if p ∈ I
This proves Statement 3’.]
[Proof of Statement 4’: The map τ is a derangement of X \ I, and thus is a permu-
tation of X \ I (since any derangement is a permutation). Hence, Statement 4 from
our above solution to Exercise 2.9.3 shows that the map τb : X → X is a permutation
in SX satisfying (τb (i ) = i for each i ∈ I ). Now, we shall show that Fix τb = I.
Indeed, I ⊆ Fix τb 257 and Fix τb ⊆ I 258 . Combining these two facts, we obtain
Fix τb = I. Thus, we have showed that τb is an element of SX that satisfies Fix τb = I.
256 Proof. Let z be a fixed point of e σ. Thus, z is an element of X \ I satisfying e σ (z) = z (by the
definition of “fixed point”). But the definition of e σ yields eσ (z) = σ (z), so that σ (z) = eσ (z) = z.
In other words, z is a fixed point of σ. In other words, z ∈ Fix σ (by the definition of Fix σ).
Hence, z ∈ Fix σ = I. But recall that z ∈ X \ I, so that z ∈ X and z ∈ / I. Now, z ∈ I contradicts
z∈/ I.
Forget that we fixed z. We thus have obtained a contradiction for each fixed point z of e σ.
Hence, there exist no fixed points of e σ. In other words,
( σ
e has no fixed points.
257 Proof. Let i ∈ I. The definition of τ τ (i ) , if i ∈
/ I;
b yields τb (i ) = = i (since i ∈ I). In other
i, if i ∈ I
words, i is a fixed point of τb. In other words, i ∈ Fix τb (by the definition of Fix τb).
Forget that we fixed i. We thus have showed that i ∈ Fix τb for each i ∈ I. In other words,
I ⊆ Fix τb.
258 Proof. Let p ∈ Fix τ
b. Thus, p is a fixed point of τb (by the definition of Fix τb). In other words, p is
an element of X satisfying τb ( p) = p.
We want to prove that p ∈ I. Indeed, assume the contrary. Thus, p ∈ / I. Combining p ∈ X
Enumerative Combinatorics: class notes page 415

In other words, τb belongs to {σ ∈ SX | Fix σ = I }. This proves Statement 4’.]

[Proof of Statement 5’: The proof of Statement 5’ is entirely analogous to the proof
of Statement 5 in our above solution to Exercise 2.9.3 (since any derangement of
X \ I is a permutation of X \ I).]
[Proof of Statement 6’: The proof of Statement 6’ is entirely analogous to the proof
of Statement 6 in our above solution to Exercise 2.9.3.]
We have now proved all six Statements 1’, 2’, 3’, 4’, 5’ and 6’. This fills all the
gaps in our above argument. Hence, Proposition 7.35.3 is completely proved.
We can draw the following conclusion from Proposition 7.35.3 (analogous to
Corollary 2.9.16):

Corollary 7.35.4. Let n ∈ N. Let X be an n-element set. Let I be a subset of X.

Then,
|{σ ∈ SX | Fix σ = I }| = Dn−| I | .

Proof of Corollary 7.35.4. The set X \ I is finite and satisfies | X \ I | = n − | I |. (This

can be proved in the same way as in our above proof of Corollary 2.9.16.)
Proposition 7.35.3 shows that there is a bijection
from {σ ∈ SX | Fix σ = I } to {derangements of X \ I } .
Hence, the bijection principle yields
|{σ ∈ SX | Fix σ = I }| = |{derangements of X \ I }|
= (# of derangements of X \ I ) .
But X \ I is an (n − | I |)-element set (since | X \ I | = n − | I |), and we have n − | I | =
| X \ I | ∈ N. Hence, Lemma 1.7.6 (applied to X \ I and n − | I | instead of X and n)
shows that
(# of derangements of X \ I ) = (# of derangements of [n − | I |]) = Dn−| I |
(since Dn−| I | is defined as the # of derangements of [n − | I |]). Thus,

|{σ ∈ SX | Fix σ = I }| = (# of derangements of X \ I ) = Dn−| I | .

This proves Corollary 7.35.4.
with p ∈
/ I, we obtain p ∈ X \ I. Now, the definition of τb yields
(
τ ( p) , if p ∈
/ I;
τb ( p) = = τ ( p) (since p ∈
/ I) ,
p, if p ∈ I

so that τ ( p) = τb ( p) = p. In other words, p is a fixed point of τ. But the permutation τ

is a derangement, and thus has no fixed points (by the definition of “derangement”). Hence,
p cannot be a fixed point of τ. This contradicts the fact that p is a fixed point of τ. This
contradiction shows that our assumption was false. Hence, p ∈ I is proven.
Now, forget that we fixed p. We thus have proved that p ∈ I for each p ∈ Fix τb. In other
words, Fix τb ⊆ I.
Enumerative Combinatorics: class notes page 416

We can now prove Theorem 1.7.9 (c) at last:

Proof of Theorem 1.7.9 (c). Let n ∈ N. Then, Fix σ is a subset of [n] for each permu-
tation σ ∈ S[n] (by the definition of Fix σ). Hence, the sum rule (Theorem 1.2.5)
yields  
S[ n ] = ∑ # of σ ∈ S[n] satisfying Fix σ = I .
I is a subset of [n]

Comparing this with

S[n] = |[n]| ! (by (220), applied to X = [n])

|{z}
=n
= n!,

we obtain
 
n! = ∑ # of σ ∈ S [n] satisfying Fix σ = I
I is a subset of [n] | {z }
=|{σ∈S[n] | Fix σ= I }|
= Dn−| I |
(by Corollary 7.35.4, applied to X =[n])

= ∑ Dn−| I | . (303)
I is a subset of [n]

Now, let us note that each subset I of [n] satisfies | I | ∈ {0, 1, . . . , n} (since Theo-
rem 1.4.7 (b) yields | I | ≤ |[n]| = n). Hence, we can split the sum ∑ Dn−| I |
I is a subset of [n]
according to the value of | I |, obtaining the following:

∑ Dn−| I | = ∑ ∑ Dn−| I |
I is a subset of [n] k∈{0,1,...,n} I is a subset of [n]; | {z }
| I |=k = Dn − k
(since | I |=k)

= ∑ ∑ Dn − k
k∈{0,1,...,n} I is a subset of [n];
| {z } | I |=k
n
=∑
| {z }
k =0 =(# of subsets I of [n] satisfying | I |=k) Dn−k
n
= ∑ (|# of subsets I of [n] satisfying | I | = k) Dn−k
k =0
{z }
=(# of k-element
 subsets of [ n ])
n
=
k
(by Theorem 1.3.12, applied to S=[n])
n  
n
= ∑ Dn − k .
k =0
k
Enumerative Combinatorics: class notes page 417

Thus, (303) becomes

n 
n
n! = ∑ Dn−| I | = ∑
k
Dn − k .
I is a subset of [n] k =0

This proves Theorem 1.7.9 (c).

We have now proved parts (a), (b) and (c) of Theorem 1.7.9. This solves Exercise
2.9.4.

7.36. Reference to solution to Exercise 2.9.5

Exercise 2.9.5 is [19s, Exercise 2.14.4].

7.37. Reference to solution to Exercise 2.9.6

Exercise 2.9.6 is [19s, Exercise 2.14.5].

7.38. Reference to solution to Exercise 2.9.7

Exercise 2.9.7 asks us to prove Lemma 2.9.21.
Lemma 2.9.21 appears (with a sketch of a proof) in [Galvin17, Lemma 16.2].
Alternatively, it can be proved by induction on n, similarly to how Theorem 1.4.1
was proved in our solution of Exercise 1.4.1. (Actually, if we apply Lemma 2.9.21 (a)
to ai = 1, then we obtain (# of subsets of [n]) = 2n , which is exactly what Theorem
1.4.1 says for S = [n].)
Alternatively, Lemma 2.9.21 can be easily obtained from the following identity:

Lemma 7.38.1. Let n ∈ N. Let a1 , a2 , . . . , an be n numbers. Let b1 , b2 , . . . , bn be n

further numbers. Then,
! 
n
∏ ( a i + bi ) = ∑ ∏ a i  ∏ bi  .
i =1 I ⊆[n] i∈ I i ∈[n]\ I

Lemma 7.38.1 appears (with detailed proof) in [Grinbe15, Exercise 6.1 (a)]. Now,
Lemma 2.9.21 can easily be derived from it:
Proof of Lemma 2.9.21. (a) Lemma 7.38.1 (applied to bi = 1) yields
! 
n
∏ ( a i + 1) = ∑ ∏ a i  ∏ 1 = ∑ ∏ ai .
i =1 I ⊆[n] i∈ I i ∈[n]\ I I ⊆[n] i ∈ I
| {z }
=1
Enumerative Combinatorics: class notes page 418

Hence,
n n
∑ ∏ ai = + 1) = ∏ (1 + a i ) = (1 + a1 ) (1 + a2 ) · · · (1 + a n ) .
∏ (|ai {z
I ⊆[n] i ∈ I i =1
} i =1
=1+ a i

This proves Lemma 2.9.21 (a).

(b) Lemma 2.9.21 (a) (applied to − ai instead of ai ) yields

∑ ∏ (−ai ) = (1 + (−a1 )) (1 + (−a2 )) · · · (1 + (−an ))

I ⊆[n] i ∈ I
= (1 − a1 ) (1 − a2 ) · · · (1 − a n )

(since 1 + (− ai ) = 1 − ai for each i ∈ [n]). Comparing this with

∑ ∏ (−ai ) = ∑ (−1)| I | ∏ ai ,
I ⊆[n] i ∈ I I ⊆[n] i∈ I
| {z }
=(−1)| I | ∏ ai
i∈ I

we obtain
∑ (−1)| I | ∏ ai = (1 − a1 ) (1 − a2 ) · · · (1 − an ) .
I ⊆[n] i∈ I

This proves Lemma 2.9.21 (b).

7.39. Solution to Exercise 2.9.8

In order to solve Exercise 2.9.8, we need to prove Lemma 2.9.25. First, we will show
an auxiliary fact:

Lemma 7.39.1. Any two distinct primes are coprime.

Proof of Lemma 7.39.1. Let p and q be two distinct primes. We must prove that p
and q are coprime. In other words, we must prove that gcd ( p, q) = 1 (since this is
what it means for p and q to be coprime).
Assume the contrary. Thus, gcd ( p, q) 6= 1.
The definition of gcd ( p, q) shows that gcd ( p, q) is the greatest positive integer
that divides both p and q. Thus, gcd ( p, q) is a positive integer and divides both p
and q. In particular, gcd ( p, q) is a positive divisor of p (since gcd ( p, q) is positive
and divides p).
But p is prime. In other words, p is an integer greater than 1 and having the prop-
erty that the only positive divisors of p are 1 and p. In particular, the only positive
divisors of p are 1 and p. Hence, gcd ( p, q) must be either 1 or p (since gcd ( p, q)
is a positive divisor of p). Since gcd ( p, q) cannot be 1 (because gcd ( p, q) 6= 1), we
thus conclude that gcd ( p, q) = p.
Similarly, we obtain gcd ( p, q) = q (since gcd ( p, q) divides not only p but also q).
Enumerative Combinatorics: class notes page 419

Comparing gcd ( p, q) = p with gcd ( p, q) = q, we obtain p = q. This contradicts

the fact that p and q are distinct. This contradiction shows that our assumption
was false. Hence, gcd ( p, q) = 1 is proved. In other words, p and q are coprime.
This proves Lemma 7.39.1.
We can now prove Lemma 2.9.25:
Proof of Lemma 2.9.25. The k numbers b1 , b2 , . . . , bk are primes, and thus are positive
integers.
Lemma 7.39.1 shows that any two distinct primes are coprime. Hence, any k
distinct primes are mutually coprime. Thus, the k positive integers b1 , b2 , . . . , bk are
mutually coprime (since they are k distinct primes). Thus, Lemma 2.9.24 shows
that b1 b2 · · · bk | c. This proves Lemma 2.9.25.
Thus, Exercise 2.9.8 is solved.

7.40. Solution to Exercise 2.9.9

In order to solve Exercise 2.9.9, we must prove Lemma 2.9.26.
Proof of Lemma 2.9.26. We must prove the equivalence (230). We shall do this by
separately proving the “=⇒” and “⇐=” implications of this equivalence:
Proof of the “=⇒” implication of (230): Assume that p j divides m for each j ∈ I.
We must show that ∏ pi divides m.
i∈ I
Let h1 , h2 , . . . , hk be all elements of I, listed in some order (with no repetitions).
Thus, I = { h1 , h2 , . . . , hk } and ∏ pi = ph1 ph2 · · · phk .
i∈ I
The primes ph1 , ph2 , . . . , phk are distinct259 .
Let i ∈ {1, 2, . . . , k }. Hence, hi ∈ {h1 , h2 , . . . , hk } = I. But p j divides m for each
j ∈ I (by our assumption). Applying this to j = hi , we obtain that phi divides m
(since hi ∈ I). In other words, phi | m.
Forget that we fixed i. We thus have showed that phi | m for each i ∈ {1, 2, . . . , k}.
Hence, Lemma 2.9.25 (applied to bi = phi ) shows that ph1 ph2 · · · phk | m (since
ph1 , ph2 , . . . , phk are k distinct primes). In view of ∏ pi = ph1 ph2 · · · phk , this rewrites
i∈ I
as ∏ pi | m. In other words, ∏ pi divides m.
i∈ I i∈ I
We thus have shown that if p j divides m for each j ∈ I, then ∏ pi divides m. In
i∈ I
other words, we have proved the “=⇒” implication of (230).

259 Proof.Let u and v be two elements of [k] such that u 6= v. We shall show that phu 6= phv .
The elements h1 , h2 , . . . , hk are distinct (since they are all elements of I, listed with no repeti-
tions). Hence, from u 6= v, we obtain hu 6= hv . Therefore, phu 6= phv , since all the primes pi (for
different i ∈ I) are distinct.
Forget that we fixed u and v. We thus have shown that if u and v are two elements of [k] such
that u 6= v, then phu 6= phv . In other words, the primes ph1 , ph2 , . . . , phk are distinct.
Enumerative Combinatorics: class notes page 420

Proof of the “⇐=” implication of (230): Assume that ∏ pi divides m. We must

i∈ I
show that p j divides m for each j ∈ I.
Fix j ∈ I. Then, we can split off the factor for i = j from the product ∏ pi . We
i∈ I
thus obtain
∏ pi = p j · ∏ pi .
i∈ I i ∈ I \{ j}
| {z }
an integer

Hence, p j | ∏ pi . But ∏ pi | m (since ∏ pi divides m). Combining the preceding

i∈ I i∈ I i∈ I
two sentences, we conclude that p j | ∏ pi | m, thus p j | m 260 . In other words, p j
i∈ I
divides m.
Forget that we fixed j. We thus have proved that p j divides m for each j ∈ I.
We thus have shown that if ∏ pi divides m, then p j divides m for each j ∈ I. In
i∈ I
other words, we have proved the “⇐=” implication of (230).
We now have proved both “=⇒” and “⇐=” implications of (230). Thus, (230)
holds, and therefore Lemma 2.9.26 is proved.
This solves Exercise 2.9.9.

7.41. Solution to Exercise 2.9.10

In order to solve Exercise 2.9.10, we must prove Lemma 2.9.27. First, we recall a
very elementary fact ([19s, Proposition 2.13.8]):

Proposition 7.41.1. Let n > 1 be an integer. Then, there exists at least one prime
p such that p | n.

Proof of Lemma 2.9.27. We must prove the equivalence

(m is coprime to u)
⇐⇒ (( pi does not divide m) for each i ∈ [n]) . (304)

We shall do this by separately proving the “=⇒” and “⇐=” implications of this
equivalence:
Proof of the “=⇒” implication of (304): Assume that m is coprime to u. We must
show that ( pi does not divide m) for each i ∈ [n].
Fix i ∈ [n]. We shall show that pi does not divide m.
Indeed, assume the contrary. Thus, pi divides m. But pi is a prime that divides u
(since all of p1 , p2 , . . . , pn are primes that divide u); thus, pi divides u. Hence, the
positive integer pi divides both m and u. Since pi is a prime, we have pi > 1.
260 Here we are using the transitivity of divisibility (i.e., the fact that if a, b and c are three integers
satisfying a | b and b | c, then a | c). This is [19s, Proposition 2.2.4 (b)] (and is easy to prove).
Enumerative Combinatorics: class notes page 421

We have assumed that m is coprime to u. In other words, gcd (m, u) = 1 (by the
definition of “coprime”). But gcd (m, u) is defined to be the greatest positive integer
that divides both m and u. Therefore, if r is any positive integer that divides both
m and u, then r ≤ gcd (m, u). Applying this to r = pi , we obtain pi ≤ gcd (m, u)
(since pi is a positive integer that divides both m and u). Thus, pi ≤ gcd (m, u) = 1.
This contradicts pi > 1.
This contradiction shows that our assumption was wrong. Hence, pi does not
divide m.
Forget that we fixed i. We thus have proved that ( pi does not divide m) for each
i ∈ [ n ].
We thus have shown that if m is coprime to u, then ( pi does not divide m) for
each i ∈ [n]. In other words, we have proved the “=⇒” implication of (304).
Proof of the “⇐=” implication of (304): Assume that ( pi does not divide m) for each
i ∈ [n]. We must show that m is coprime to u. In other words, we must show that
gcd (m, u) = 1 (because this is what it means for m to be coprime to u).
Indeed, assume the contrary. Thus, gcd (m, u) 6= 1. But gcd (m, u) is the great-
est positive integer that divides both m and u (by the definition of gcd (m, u)).
Hence, gcd (m, u) is a positive integer. Therefore, from gcd (m, u) 6= 1, we obtain
gcd (m, u) > 1. Hence, Proposition 7.41.1 (applied to gcd (m, u) instead of n) yields
that there exists at least one prime p such that p | gcd (m, u). Consider this p.
Recall that gcd (m, u) divides both m and u. In other words, gcd (m, u) | m and
gcd (m, u) | u.
We have p | gcd (m, u) and gcd (m, u) | m. Combining these two divisibilities, we
find p | gcd (m, u) | m, thus p | m 261 . Likewise, combining p | gcd (m, u) with
gcd (m, u) | u, we find p | gcd (m, u) | u, so that p | u 262 . Thus, p is a prime that
divides u (since p is a prime). Hence, p must be one of the n primes p1 , p2 , . . . , pn
(since these n primes p1 , p2 , . . . , pn are precisely the distinct primes that divide u).
In other words, p = p j for some j ∈ [n]. Consider this j.
Recall that we assumed that ( pi does not divide m) for each i ∈ [n]. Applying
this to i = j, we conclude that p j does not divide m. In other words, p does not
divide m (since p = p j ). But this contradicts p | m. This contradiction shows that
our assumption was false. Hence, we have proved that gcd (m, u) = 1. In other
words, m is coprime to u.
We thus have shown that if ( pi does not divide m) for each i ∈ [n], then m is
coprime to u. In other words, we have proved the “⇐=” implication of (304).
We now have proved both “=⇒” and “⇐=” implications of (304). Thus, (304)
holds, and therefore Lemma 2.9.27 is proved.
This solves Exercise 2.9.10.

261 Here we are using the transitivity of divisibility (i.e., the fact that if a, b and c are three integers
satisfying a | b and b | c, then a | c). This is [19s, Proposition 2.2.4 (b)] (and is easy to prove).
262 again, by the transitivity of divisibility
Enumerative Combinatorics: class notes page 422

7.42. Solution to Exercise 2.9.11

Solution to Exercise 2.9.11. We proceed similarly to the second proof of Proposition
2.9.10.
We are in one of the following two cases:
Case 1: We have S ⊆ T.
Case 2: We don’t have S ⊆ T.
Let us first consider Case 1. In this case, we have S ⊆ T. Hence, every subset I
of S satisfies
(−1)| I \T | = 1. (305)
[Proof of (305): Let I be a subset of S. Thus, I ⊆ S ⊆ T, so that I \ T = ∅. Thus,
| I \ T | = |∅| = 0. Hence, (−1)| I \T | = (−1)0 = 1. This proves (305).]
On the other hand, S is a |S|-element set. Hence, Theorem 1.4.1 (applied to
n = |S|) yields (# of subsets of S) = 2|S| .
Now,

∑ (− 1)| I \T | = ∑ 1 = (# of subsets I of S) · 1
I ⊆S I ⊆S
| {z }
=1
(by (305))

= (# of subsets I of S) = (# of subsets of S) = 2|S| .

Comparing this with

(
2|S| , if S ⊆ T;
= 2| S | (since S ⊆ T ) ,
0, otherwise

we obtain (
2|S| , if S ⊆ T;
∑ (−1) | I \T|
=
0, otherwise
.
I ⊆S

Thus, Exercise 2.9.11 is solved in Case 1.

Let us now consider Case 2. In this case, we don’t have S ⊆ T. Hence, there
exists some g ∈ S such that g ∈ / T. Consider this g. (We can have many choices
for g, but we just pick one.) Note that the sets { g} and T are disjoint (since g ∈ / T).
Hence, { g} \ T = { g}.
Each subset I of S must satisfy either g ∈ I or g ∈ / I (but not both at the same
| I \T|
time). Hence, we can split the sum ∑ (−1) as follows:
I ⊆S

∑ (−1)| I \T| = ∑ (−1)| I \T | + ∑ (−1)| I \T | . (306)

I ⊆S I ⊆S; I ⊆S;
g∈ I ∈I
g/
Enumerative Combinatorics: class notes page 423

The map263

{I ⊆ S | g ∈
/ I} → {I ⊆ S | g ∈ I} ,
J 7→ J ∪ { g}

is a bijection. (This can be proved in the exact same way as it was proved during
our second proof of Proposition 2.9.10.)
Moreover, every subset J of S satisfying g ∈
/ J must satisfy

(−1)|( J ∪{ g})\T | = − (−1)| J \T | . (307)

[Proof of (307): Let J be a subset of S satisfying g ∈ / J. Then, we cannot have

g ∈ J \ T (since g ∈ J \ T would imply g ∈ J \ T ⊆ J, which would contradict
g ∈/ J). Thus, we must have g ∈ / J \ T. But any three sets X, Y and Z satisfy
( X ∪ Y ) \ Z = ( X \ Z ) ∪ (Y \ Z ). Applying this to X = J, Y = { g} and Z = T, we
find
( J ∪ { g}) \ T = ( J \ T ) ∪ ({ g} \ T ) = ( J \ T ) ∪ { g} .
| {z }
={ g}

Hence, |( J ∪ { g}) \ T | = |( J \ T ) ∪ { g}| = | J \ T | + 1 (since g ∈ / J \ T). Thus,

|( J ∪{ g})\ T | | J \ T |+1 | J \T|
(−1) = (−1) = − (−1) . This proves (307).]
Now, we can substitute J ∪ { g} for I in the sum ∑ (−1)| I \T | (since the map
I ⊆S;
g∈ I
{I ⊆ S | g ∈ / I } → { I ⊆ S | g ∈ I } , J 7→ J ∪ { g} is a bijection). We thus obtain
 
∑ (− 1 ) | I \T|
= ∑| (− 1 ) |( J ∪{ g})\ T |
= ∑ − (− 1 ) | J \T|
= − ∑ (−1)| J \T |
I ⊆S; J ⊆S; J ⊆S; J ⊆S;
{z }
g∈ I ∈J
g/ =−(−1)| J \T | ∈J
g/ ∈J
g/
(by (307))

=− ∑ (−1)| I \T | (308)
I ⊆S;
∈I
g/

(here, we have renamed the summation index J as I).

Now, (306) becomes

∑ (−1)| I \T| = ∑ (−1)| I \T | + ∑ (−1)| I \T | = − ∑ (−1)| I \T | + ∑ (−1)| I \T | = 0.

I ⊆S I ⊆S; I ⊆S; I ⊆S; I ⊆S;
g∈ I ∈I
g/ ∈I
g/ ∈I
g/
| {z }
| I \T|
=− ∑ (−1)
I ⊆S;
∈I
g/
(by (308))

263 Thenotation “{ I ⊆ S | g ∈/ I }” means “the set of all subsets I of S satisfying g ∈/ I”. Similarly,
the notation “{ I ⊆ S | g ∈ I }” means “the set of all subsets I of S satisfying g ∈ I”.
Enumerative Combinatorics: class notes page 424

Comparing this with

(
2|S| , if S ⊆ T;
=0 (since we don’t have S ⊆ T ) ,
0, otherwise

we obtain (
2|S| , if S ⊆ T;
∑ (−1)| I \T | =
0, otherwise
.
I ⊆S

Thus, Exercise 2.9.11 is solved in Case 2.

Thus, Exercise 2.9.11 is solved (since we have solved it in both Cases 1 and 2).

7.43. Reference to solution to Exercise 2.9.12

Exercise 2.9.12 is [19f-mt2s, Exercise 2].

7.44. Solution to Exercise 2.10.1

Solution to Exercise 2.10.1. (a) Let n ∈ N and k ∈ N. We must prove that

n−1

 
n−1
  , if n > 0;
= k−1 . (309)
n−k 
 [ k = 0] , if n = 0

We are in one of the following two cases:

Case 1: We have n = 0.
Case 2: We have n 6= 0.
Let us first consider Case 1. In this case, we have n = 0. We are in one of the
following two subcases:
Subcase 1.1: We have k = 0.
Subcase 1.2: We have k 6= 0.
Let us first consider Subcase
  1.1. In  this
 subcase, we have k = 0. From n = 0 and
n−1 0−1 −1
 
k = 0, we obtain = = = 1 (by (44)). Comparing this with
n−k 0−0 0

 n − 1 , if n > 0;


k−1 = [ k = 0] (since n = 0)

 [ k = 0] , if n = 0
=1 (since k = 0) ,

 n−1 ,


n−1
 
if n > 0;
we find = k−1 . Thus, (309) is proved in Subcase 1.1.
n−k 
 [ k = 0] , if n = 0
Enumerative Combinatorics: class notes page 425

Let us next consider Subcase 1.2. In this subcase, we have k 6= 0. Hence, k > 0
(since k ∈ N), so that −k < 0 and therefore −k ∈ / N. Thus, |{z}n −k = 0 − k =
=0
− / N.
k ∈  Therefore, (43) (applied to n − 1 and n − k instead of n and k) yields
n−1
= 0. Comparing this with
n−k

 n − 1 , if n > 0;


k−1 = [ k = 0] (since n = 0)

 [ k = 0] , if n = 0
=0 (since k 6= 0) ,

n−1


n−1
   , if n > 0;
we find = k−1 . Thus, (309) is proved in Subcase 1.2.
n−k 
 [ k = 0] , if n = 0
We have now proved (309) in each of the two Subcases 1.1 and 1.2. Hence, (309)
is proved in Case 1 (since Subcases 1.1 and 1.2 cover Case 1).
Let us next consider Case 2. In this case, we have n 6= 0. Hence, n is a positive
integer (since n ∈ N). Thus, n ≥ 1, so that n − 1 ∈ N. Therefore, Theorem 1.3.11
(applied to n − 1 and n − k instead of n and k) yields

n−1 n−1 n−1

     
= = (since (n − 1) − (n − k) = k − 1) .
n−k ( n − 1) − ( n − k ) k−1

Comparing this with


 n − 1 , if n > 0;


n−1
 
k−1 = (since n > 0 (because n ≥ 1)) ,

 [ k = 0] , k−1
if n = 0

n−1


n−1
   , if n > 0;
we obtain = k−1 . Thus, (309) is proved in Case 2.
n−k 
 [ k = 0] , if n = 0
We have now proved (309) in each of the two Cases 1 and 2. Hence, (309) always
holds.
Thus, we have proved (309). This solves Exercise 2.10.1 (a).
(b) We must prove that Theorem 2.10.1 holds for n = 0.
So let n ∈ N and k ∈ N, and assume that n = 0. We must prove the claim of
Theorem 2.10.1.
We shall first prove (239). We are in one of the following two cases:
Case 1: We have k = 0.
Case 2: We have k 6= 0.
Enumerative Combinatorics: class notes page 426

Let us first consider Case 1. In this case, we have k = 0. Hence,

 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n
 

= # of ( x1 , x2 , . . . , x0 ) ∈ P0 satisfying x1 + x2 + · · · + x0 = |{z}
n 
 
| {z }
=(empty sum)=0 =0
 
= # of ( x1 , x2 , . . . , x0 ) ∈ P0 satisfying 0 = 0
 
= # of ( x1 , x2 , . . . , x0 ) ∈ P 0
 
since every ( x1 , x2 , . . . , x0 ) ∈ P0 satisfies 0 = 0
=1

(because there exists exactly one 0-tuple ( x1 , x2 , . . . , x0 ) ∈ P0 : namely, the empty

list ()). Comparing this with

n−1 0−1
   
= (since n = 0 and k = 0)
n−k 0−0
0−1
 
= =1 (by (44)) ,
0

we obtain
   n − 1
k
# of ( x1 , x2 , . . . , xk ) ∈ P satisfying x1 + x2 + · · · + xk = n = .
n−k

Hence, (239) is proved in Case 1.

Let us next consider Case 2. In this case, we have k 6= 0. Hence, k > 0 (since
k ∈ N), so that n − |{z}
k < n = 0 and therefore n − k ∈/ N. Thus, (43) (applied to
>0
n−1
 
n − 1 and n − k instead of n and k) yields = 0.
n−k
We have k > 0. Thus, x1 + x2 + · · · + xk 6= n for each ( x1 , x2 , . . . , xk ) ∈ Pk 264 . In
other words, there exists no ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n.
Thus,
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n = 0.

Let ( x1 , x2 , . . . , xk ) ∈ Pk . Thus, x1 , x2 , . . . , xk are elements of P; in other words, x1 , x2 , . . . xk

264 Proof.

are positive integers (since P is the set of all positive integers). Thus, x1 + x2 + · · · + xk is a sum
of k positive integers. Thus, x1 + x2 + · · · + xk is a positive integer itself (since any sum of k
positive integers is a positive integer (because k 6= 0)). Hence, x1 + x2 + · · · + xk > 0, so that
x1 + x2 + · · · + xk 6= 0. In other words, x1 + x2 + · · · + xk 6= n (since n = 0). Qed.
Enumerative Combinatorics: class notes page 427

n−1
 
Comparing this with = 0, we obtain
n−k
n−1
   
k
# of ( x1 , x2 , . . . , xk ) ∈ P satisfying x1 + x2 + · · · + xk = n = .
n−k
Hence, (239) is proved in Case 2.
Hence, we have proved (239) in each of the two Cases 1 and 2. Thus, (239) always
holds.
Now, (239) yields
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n

n−1

 
n−1
  , if n > 0;
= = k−1 (by Exercise 2.10.1 (a)) .
n−k 
 [ k = 0] , if n = 0

Thus, we have proved Theorem 2.10.1 under the assumption that n = 0. So we

know that Theorem 2.10.1 holds for n = 0. This completes the solution to Exercise
2.10.1 (b).
(c) We must prove that Theorem 2.10.1 holds for k = 0.
So let n ∈ N and k ∈ N, and assume that k = 0. We must prove the claim of
Theorem 2.10.1.
If n = 0, then this follows from Exercise 2.10.1 (b) (which was solved above).
Hence, for the rest of this proof, we WLOG assume that n 6= 0. Hence, n ≥ 1 (since
n ∈ N), so that n − 1 ≥ 0 and thus n − 1 ∈ N. Moreover, n − |{z} k = n > n − 1.
=0
Hence, Proposition 1.3.6 (applied to n − 1 and n − k instead of n and k) yields
n−1

= 0.
n−k
Also, we have x1 + x2 + · · · + xk 6= n for each ( x1 , x2 , . . . , xk ) ∈ Pk 265 . In other
words, there exists no ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n. Thus,
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n = 0.

n−1
 
Comparing this with = 0, we obtain
n−k
 
# of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n

n−1

 
n−1
  , if n > 0;
= = k−1 (by Exercise 2.10.1 (a)) .
n−k 
 [ k = 0] , if n = 0

Let ( x1 , x2 , . . . , xk ) ∈ Pk . From k = 0, we obtain x1 + x2 + · · · + xk = x1 + x2 + · · · + x0 =

265 Proof.

( empty sum) = 0 6= n (since n 6= 0). Qed.

Enumerative Combinatorics: class notes page 428

Thus, we have proved Theorem 2.10.1 under the assumption that k = 0. So we

know that Theorem 2.10.1 holds for k = 0. This completes the solution to Exercise
2.10.1 (c).

7.45. Solution to Exercise 2.10.2

In order to solve Exercise 2.10.2, we need to give an alternative proof of Theorem
2.10.1.
Second proof of Theorem 2.10.1. Forget that we fixed n and k.
First, we shall prove (239) by induction on k.
Induction base: We know (from Exercise 2.10.1 (c)) that Theorem 2.10.1 holds for
k = 0. Hence, in particular, (239) holds for k = 0. This completes the induction
base.
Induction step: Let q be a positive integer. Assume (as the induction hypothesis)
that (239) holds for k = q − 1. We must show that (239) holds for k = q.
We have assumed that (239) holds for k = q − 1. In other words, we have
 
# of x1 , x2 , . . . , xq−1 ∈ Pq−1 satisfying x1 + x2 + · · · + xq−1 = n

n−1
 
= (310)
n − ( q − 1)

for each n ∈ N.
Now, let n ∈ N. We shall compute the # of x1 , x2 , . . . , xq ∈ Pq satisfying

x1 + x2 + · · · + xq = n.
First, we notice that if x1 , x2 , . . . , xq ∈ Pq satisfies x1 + x2 + · · · + xq = n, then

xq ∈ [n] 266 . Hence, the sum rule (Theorem 1.2.5) yields

# of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n

= ∑ # of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n and xq = p .



p∈[n]

Now, let p ∈ [n]. Then, p is a positive integer, so that p ∈ P. Moreover, from

p ∈ [n] = {1, 2, . . . , n}, we obtain p ≤ n, so that n − p ∈ N. Hence, (310) (applied

Let x1 , x2 , . . . , xq ∈ Pq be such that x1 + x2 +
· · · + xq = n. We must prove that xq ∈ [n].

266 Proof.

We have x1 , x2 , . . . , xq ∈ P (since x1 , x2 , . . . , xq ∈ Pq ). In other words, x1 , x2 , . . . , xq are

positive integers (since P is the set of all positive integers). Thus, in particular, xq is a positive
integer. Moreover, x1 + x2 + · · · + xq−1 is a sum of positive integers (since x1 , x2 , . . . , xq are
positive integers), and thus nonnegative. Hence, x1 + x2 + · · · + xq−1 ≥ 0. Now,


n = x 1 + x 2 + · · · + x q = x 1 + x 2 + · · · + x q −1 + x q ≥ x q .
| {z }
≥0

In other words, xq ≤ n. Since xq is a positive integer, we thus obtain xq ∈ {1, 2, . . . , n} = [n].

Qed.
Enumerative Combinatorics: class notes page 429

to n − p instead of n) yields
 
q −1
# of x1 , x2 , . . . , xq−1 ∈ P


satisfying x1 + x2 + · · · + xq−1 = n − p
(n − p) − 1
 
= . (311)
( n − p ) − ( q − 1)

If x1 , x2 , . . . , xq ∈ Pq satisfies x1 + x2 + · · · + xq = n and xq = p, then x1 + x2 +

· · · + xq−1 = n − p 267 . Hence, we can define a map

x1 , x2 , . . . , xq ∈ Pq | x1 + x2 + · · · + xq = n and xq = p


n o
→ x 1 , x 2 , . . . , x q −1 ∈ Pq −1 | x 1 + x 2 + · · · + x q −1 = n − p

that sends each q-tuple x1 , x2 , . . . , xq to x1 , x2 , . . . , xq−1 . Conversely, we can
define a map
n o
q −1
x 1 , x 2 , . . . , x q −1 ∈ P


| x 1 + x 2 + · · · + x q −1 = n − p
→ x1 , x2 , . . . , xq ∈ Pq | x1 + x2 + · · · + xq = n and xq = p


that sends each (q − 1)-tuple x1 , x2 , . . . , xq−1 to x1 , x2 , . . . , xq−1 , p (because if
x1 , x2 , . . . , xq−1 ∈ Pq−1 satisfies x1 + x2 + · · · + xq−1 = n − p, then the entries of




the q-tuple x1 , x2 , . . . , xq−1 , p sum up to x1 + x2 + · · · + xq−1 + p = (n − p) + p =
| {z }
=n− p
n). These two maps are mutually inverse, and thus are bijections. Hence, the
bijection principle yields

x1 , x2 , . . . , xq ∈ Pq | x1 + x2 + · · · + xq = n and xq = p


n o
x 1 , x 2 , . . . , x q −1 ∈ Pq −1 | x 1 + x 2 + · · · + x q −1 = n − p


=
 
= # of x1 , x2 , . . . , xq−1 ∈ Pq−1 satisfying x1 + x2 + · · · + xq−1 = n − p

(n − p) − 1
 
= (by (311)) . (312)
( n − p ) − ( q − 1)

Forget that we fixed p. We thus have proved (312) for each p ∈ [n]. Now, we can

267 because



n = x 1 + x 2 + · · · + x q = x 1 + x 2 + · · · + x q −1 + x q = x 1 + x 2 + · · · + x q −1 + p
|{z}
=p
Enumerative Combinatorics: class notes page 430

continue our computation from before:

# of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n

= ∑ # of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n and xq = p



p∈[n]
| {z }
=|{( x1 ,x2 ,...,xq )∈Pq | x1 + x2 +···+ xq =n and xq = p}|
(n − p) − 1
 
=
( n − p ) − ( q − 1)
(by (312))
n
(n − p) − 1 (n − p) − 1
   
= ∑ ( n − p ) − ( q − 1)
= ∑ ( n − p ) − ( q − 1)
p∈[n] p =1
|{z}
n
=∑
p =1

n −1 
j−1

= ∑ j − ( q − 1)
(313)
j =0

(here, we have substituted j for n − p in the sum).

But let j ∈ Z. Then, Theorem 1.3.8 (applied to j and j − (q − 1) instead of n and
k) yields

j−1 j−1 j−1 j−1

         
j
= + = +
j − ( q − 1) j − ( q − 1) − 1 j − ( q − 1) j−q j − ( q − 1)

j−1
 
(since j − (q − 1) − 1 = j − q). Solving this equation for , we obtain
j − ( q − 1)

j−1 j−1
     
j
= −
j − ( q − 1) j − ( q − 1) j−q
| {z }
( j + 1) − 1
 
=
(( j + 1) − 1) − (q − 1)
(since j=( j+1)−1)
( j + 1) − 1 j−1
   
= −
(( j + 1) − 1) − (q − 1) j−q
( j + 1) − 1 j−1
   
= − (314)
( j + 1) − q j−q

(since (( j + 1) − 1) − (q − 1) = ( j + 1) − q).
Forget that we fixed j. We thus have proved (314) for each j ∈ Z.
Enumerative Combinatorics: class notes page 431

Now, (313) becomes

# of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n

n −1
j−1
 
= ∑
j =0
j − ( q − 1)
 | {z  }
( j + 1) − 1 j−1

= −
( j + 1) − q j−q
(by (314))
n−1 
( j + 1) − 1 j−1 ((n − 1) + 1) − 1 0−1
      
= ∑ − = −
j =0
( j + 1) − q j−q ((n − 1) + 1) − q 0−q
j−1
  
by Theorem 2.1.1, applied to u = 0 and v = n − 1 and a j =
j−q
n−1 0−1
   
= − (since (n − 1) + 1 = n and 0 − q = −q) .
n−q −q
But q is positive, and thus −q is negative.  −q ∈
 Hence, / N. Hence, (43) (applied
0−1
to 0 − 1 and −q instead of n and k) yields = 0. Thus, we obtain
−q

# of x1 , x2 , . . . , xq ∈ Pq satisfying x1 + x2 + · · · + xq = n

n−1 0−1 n−1

     
= − = .
n−q −q n−q
| {z }
=0

Now, forget that we fixed n. We thus have proved that

n−1
 
q
# of x1 , x2 , . . . , xq ∈ P satisfying x1 + x2 + · · · + xq = n =



n−q

for each n ∈ N. In other words, (239) holds for k = q. This completes the induction
step.
Thus, we have proved (239) by induction. Now, for each n ∈ N and k ∈ N, we
have
 
k
# of ( x1 , x2 , . . . , xk ) ∈ P satisfying x1 + x2 + · · · + xk = n
n−1
 
= (by (239))
n−k

 n − 1 , if n > 0;


= k−1 (by Exercise 2.10.1 (a)) .

 [ k = 0] , if n = 0

This proves Theorem 2.10.1.

Enumerative Combinatorics: class notes page 432

7.46. Solution to Exercise 2.10.3

In order to solve Exercise 2.10.3, we need to give an alternative proof of Theorem
2.10.4.
Second proof of Theorem 2.10.4. Forget that we fixed n and k.
We shall prove Theorem 2.10.4 by induction on k.
Induction base: It is easy to see that Theorem 2.10.4 holds for k = 0 268 . This
completes the induction base.

268 Proof. Assume that k = 0. Let n ∈ Z be arbitrary. Now,

 
# of ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 + · · · + xk = n
 

= # of ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 satisfying x1 + x2 + · · · + x0 = n

 
| {z }
=(empty sum)=0
(since k = 0)
 
= # of ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 satisfying 0 = n .

But the right hand side is easy to compute:

 
• If n = 0, then # of ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 satisfying 0 = n = 1 (because there is ex-
actly one 0-tuple ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 (namely, the empty list ()), and clearly this
0-tuple satisfies 0 = n (since n = 0)).
 
• If n 6= 0, then # of ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 satisfying 0 = n = 0 (because there is no
0-tuple ( x1 , x2 , . . . , x0 ) ∈ {0, 1}0 satisfying 0 = n (since 0 6= n)).

Combining these two observations, we conclude that

(

0
 1, if n = 0;
# of ( x1 , x2 , . . . , x0 ) ∈ {0, 1} satisfying 0 = n = .
0, if n 6= 0

Comparing this with

   
k 0
= (since k = 0)
n n
= [ n = 0] (by Lemma 1.3.14, applied to n instead of k)
( (
1, if n = 0; 1, if n = 0;
= = ,
0, otherwise 0, if n 6= 0

we obtain
 

k k

# of ( x1 , x2 , . . . , xk ) ∈ {0, 1} satisfying x1 + x2 + · · · + xk = n = .
n

In other words, Theorem 2.10.4 holds for our k and n. We have thus shown that Theorem 2.10.4
holds under the assumption that k = 0. Hence, Theorem 2.10.4 holds for k = 0.
Enumerative Combinatorics: class notes page 433

Induction step: Let q be a positive integer. Assume (as the induction hypothesis)
that Theorem 2.10.4 holds for k = q − 1. We must show that Theorem 2.10.4 holds
for k = q.
We have assumed that Theorem 2.10.4 holds for k = q − 1. In other words, we
have
 
# of x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 satisfying x1 + x2 + · · · + xq−1 = n

q−1
 
= (315)
n

for each n ∈ Z.
Now, let n ∈ Z. We shall compute the # of x1 , x2 , . . . , xq ∈ {0, 1}q satisfying

x1 + x2 + · · · + xq = n.
Clearly, if x1 , x2 , . . . , xq ∈ {0, 1}q satisfies x1 + x2 + · · · + xq = n, then xq ∈

{0, 1}. Hence, the sum rule (Theorem 1.2.5) yields

# of x1 , x2 , . . . , xq ∈ {0, 1}q satisfying x1 + x2 + · · · + xq = n

= ∑ # of x1 , x2 , . . . , xq ∈ {0, 1}q satisfying x1 + x2 + · · · + xq = n and xq = p .

p∈{0,1}

Now, let p ∈ {0, 1}. Then, (315) (applied to n − p instead of n) yields

 
# of x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 satisfying x1 + x2 + · · · + xq−1 = n − p

q−1
 
= . (316)
n−p

If x1 , x2 , . . . , xq ∈ {0, 1}q satisfies x1 + x2 + · · · + xq = n and xq = p, then

x1 + x2 + · · · + xq−1 = n − p 269 . Hence, we can define a map

x1 , x2 , . . . , xq ∈ {0, 1}q | x1 + x2 + · · · + xq = n and xq = p



n o

q −1
→ x1 , x2 , . . . , xq−1 ∈ {0, 1} | x 1 + x 2 + · · · + x q −1 = n − p



that sends each q-tuple x1 , x2 , . . . , xq to x1 , x2 , . . . , xq−1 . Conversely, we can
define a map
n o
x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 | x1 + x2 + · · · + xq−1 = n − p

→ x1 , x2 , . . . , xq ∈ {0, 1}q | x1 + x2 + · · · + xq = n and xq = p



that sends each (q − 1)-tuple x1 , x2 , . . . , xq−1 to x1 , x2 , . . . , xq−1 , p (because if
x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 satisfies x1 + x2 + · · · + xq−1 = n − p, then the entries

269 because



n = x 1 + x 2 + · · · + x q = x 1 + x 2 + · · · + x q −1 + x q = x 1 + x 2 + · · · + x q −1 + p
|{z}
=p
Enumerative Combinatorics: class notes page 434

of the q-tuple x1 , x2 , . . . , xq−1 , p sum up to x1 + x2 + · · · + xq−1 + p = (n − p) +
| {z }
=n− p
p = n). These two maps are mutually inverse, and thus are bijections. Hence, the
bijection principle yields

x1 , x2 , . . . , xq ∈ {0, 1}q | x1 + x2 + · · · + xq = n and xq = p



n o
x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 | x1 + x2 + · · · + xq−1 = n − p


=
 
= # of x1 , x2 , . . . , xq−1 ∈ {0, 1}q−1 satisfying x1 + x2 + · · · + xq−1 = n − p

q−1
 
= (by (316)) . (317)
n−p

Forget that we fixed p. We thus have proved (317) for each p ∈ {0, 1}. Now, we
can continue our computation from before:

# of x1 , x2 , . . . , xq ∈ {0, 1}q satisfying x1 + x2 + · · · + xq = n

= ∑ # of x1 , x2 , . . . , xq ∈ {0, 1}q satisfying x1 + x2 + · · · + xq = n and xq = p

p∈{0,1}
| {z }
=|{( x1 ,x2 ,...,xq )∈{0,1}q | x1 + x2 +···+ xq =n and xq = p}|
q−1
 
=
n−p
(by (317))
q−1 q−1 q−1 q−1 q−1
     
   
= ∑ = + = +
p∈{0,1}
n−p n−0 n−1 n−1 n−0
q−1 q−1
   
= + (since n − 0 = n) .
n−1 n

Comparing this with

q−1 q−1
       
q by Theorem 1.3.8, applied to q and n
= + ,
n n−1 n instead of n and k

we obtain
 

q
q
# of x1 , x2 , . . . , xq ∈ {0, 1} satisfying x1 + x2 + · · · + xq = n = .
n

Forget that we fixed n. We thus have proved

 

q
q
# of x1 , x2 , . . . , xq ∈ {0, 1} satisfying x1 + x2 + · · · + xq = n =
n

for each n ∈ Z. In other words, Theorem 2.10.4 holds for k = q. This completes the
induction step. Thus, Theorem 2.10.4 is proved by induction.
Enumerative Combinatorics: class notes page 435

7.47. Solution to Exercise 2.10.4

Solution to Exercise 2.10.4. Let n ∈ N and k ∈ N. We are in one of the following
two cases:
Case 1: We have k = 0.
Case 2: We have k 6= 0.
Let us first consider Case 1. In this case, we have k = 0. We are in one of the
following two subcases:
Subcase 1.1: We have n = 0.
Subcase 1.2: We have n 6= 0.
Let us first consider Subcase
 1.1. In this subcase, we have n = 0. From n = 0 and
n+k−1 0+0−1
 
k = 0, we obtain = = 1 (by (44)). Comparing this with
n 0

 n + k − 1 , if k > 0;


k−1 = [ n = 0] (since k = 0)

 [ n = 0] , if k = 0
=1 (since n = 0) ,

 n + k − 1 , if k > 0;


n+k−1
 
we find = k−1 . Thus, Exercise 2.10.4 is proved
n 
 [ n = 0] , if k = 0
in Subcase 1.1.
Let us next consider Subcase 1.2. In this subcase, we have n 6= 0. Hence, n ≥ 1
(since n ∈ N), so that n − 1 ≥ 0 and thus n − 1  ∈ N. Hence, Proposition 1.3.6
n−1

(applied to n − 1 and n instead of n and k) yields = 0 (since n > n − 1).
 n 
n+k−1 n−1
 
k −1 = n − 1, so that
Now, n + |{z} = = 0. Comparing this
n n
=0
with

 n+k−1 ,


if k > 0;
k−1 = [ n = 0] (since k = 0)

 [ n = 0] , if k = 0
=0 (since n 6= 0) ,

 n + k − 1 , if k > 0;


n+k−1
 
we find = k−1 . Thus, Exercise 2.10.4 is proved
n 
 [ n = 0] , if k = 0
in Subcase 1.2.
We have now proved Exercise 2.10.4 in each of the two Subcases 1.1 and 1.2.
Hence, Exercise 2.10.4 is proved in Case 1 (since Subcases 1.1 and 1.2 cover Case 1).
Enumerative Combinatorics: class notes page 436

Let us next consider Case 2. In this case, we have k 6= 0. Hence, k ≥ 1 (since k ∈

N), so that n + |{z}
k −1 ≥ n + 1 − 1 = n ≥ 0 (since n ∈ N). Hence, n + k − 1 ∈ N.
≥1
Therefore, Theorem 1.3.11 (applied to n + k − 1 and n instead of n and k) yields
n+k−1 n+k−1 n+k−1
     
= = (since (n + k − 1) − n = k − 1) .
n ( n + k − 1) − n k−1
Comparing this with

 n + k − 1 , if k > 0;


n+k−1
 
k−1 = (since k > 0 (because k ≥ 1)) ,

 [ n = 0] , k−1
if k = 0

n+k−1

 
n+k−1
  , if k > 0;
we obtain = k−1 . Thus, Exercise 2.10.4 is proved
n 
 [ n = 0] , if k = 0
in Case 2.
We have now proved Exercise 2.10.4 in each of the two Cases 1 and 2. Hence,
Exercise 2.10.4 always holds.

7.48. Solution to Exercise 2.10.5

Our solution to Exercise 2.10.5 will use the following simple lemma:

Lemma 7.48.1. Let n ∈ N. Let ( x1 , x2 , . . . , xk ) be a composition of n. Then, k ≤ n.

Proof of Lemma 7.48.1. We know that ( x1 , x2 , . . . , xk ) is a composition of n. In other

words, ( x1 , x2 , . . . , xk ) is a tuple of positive integers whose sum is n (by Definition
2.10.2 (c)). Hence, x1 , x2 , . . . , xk are positive integers whose sum is n. In other
words, x1 , x2 , . . . , xk are positive integers and satisfy x1 + x2 + · · · + xk = n.
For each i ∈ {1, 2, . . . , k}, the number xi is a positive integer (since x1 , x2 , . . . , xk
are positive integers) and thus satisfies xi ≥ 1 (since any positive integer is ≥ 1).
Summing up these inequalities for all i ∈ {1, 2, . . . , k }, we obtain x1 + x2 + · · · +
xk ≥ 1| + 1 +{z· · · + 1} = k · 1 = k. In view of x1 + x2 + · · · + xk = n, this rewrites as
k times
n ≥ k. In other words, k ≤ n. This proves Lemma 7.48.1.
Solution to Exercise 2.10.5. There are many ways to solve Exercise 2.10.5 (a). One
way is to use strong induction on n. Another is to construct a bijection
{{1, 2} -compositions of n} → {lacunar subsets of [n − 1]}
(defined in the same way as the bijection C in our first proof of Theorem 2.10.1),
and then use the bijection principle (and Proposition 1.4.9). Yet another is to find a
bijection
{{1, 2} -compositions of n} → {domino tilings of Rn,2 } ,
Enumerative Combinatorics: class notes page 437

and then use the bijection principle (and Proposition 1.1.11). All these approaches
can be easily adapted to yield Exercise 2.10.5 (b) as well.
We shall, however, use the conceptually simplest approach: We will first solve
Exercise 2.10.5 (b) by deriving it from Theorem 2.10.4, and then solve Exercise 2.10.5
(a) by summing the result over all k. So let us start with the solution to Exercise
2.10.5 (b):
(b) If (u1 , u2 , . . . , uk ) is a {1, 2}-composition of n into k parts, then
(u1 − 1, u2 − 1, . . . , uk − 1) is a k-tuple ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 +
· · · + xk = n − k 270 . Hence, we can define a map

A : {{1, 2} -compositions of n into k parts}

n o
k
→ ( x1 , x2 , . . . , xk ) ∈ {0, 1} | x1 + x2 + · · · + xk = n − k ,
(u1 , u2 , . . . , uk ) 7→ (u1 − 1, u2 − 1, . . . , uk − 1) .

Let us now construct

n a map in the opposite direction. o
If ( p1 , p2 , . . . , pk ) ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1}k | x1 + x2 + · · · + xk = n − k , then
( p1 + 1, p2 + 1, . . . , pk + 1) is a {1, 2}-composition of n into k parts271 . Hence, we
270 Proof. Let (u1 , u2 , . . . , uk ) be a {1, 2}-composition of n into k parts. We must show that
(u1 − 1, u2 − 1, . . . , uk − 1) is a k-tuple ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 + · · · + xk =
n − k.
We know that (u1 , u2 , . . . , uk ) is a {1, 2}-composition of n into k parts. In other words,
(u1 , u2 , . . . , uk ) is a composition ( x1 , x2 , . . . , xk ) of n into k parts such that x1 , x2 , . . . , xk ∈ {1, 2}
(by the definition of a “{1, 2}-composition of n into k parts”). In other words, (u1 , u2 , . . . , uk ) is
a composition of n into k parts and satisfies u1 , u2 , . . . , uk ∈ {1, 2}. From u1 , u2 , . . . , uk ∈ {1, 2},
we obtain u1 − 1, u2 − 1, . . . , uk − 1 ∈ {0, 1} (since p − 1 ∈ {0, 1} for each p ∈ {1, 2}). In other
words, (u1 − 1, u2 − 1, . . . , uk − 1) ∈ {0, 1}k . Moreover, (u1 , u2 , . . . , uk ) is a composition of n into
k parts; in other words, (u1 , u2 , . . . , uk ) is a k-tuple of positive integers whose sum is n (by Def-
inition 2.10.2 (d)). In other words, (u1 , u2 , . . . , uk ) is a k-tuple of positive integers and satisfies
u1 + u2 + · · · + uk = n. Hence,

(u1 − 1) + (u2 − 1) + · · · + (uk − 1) = (u1 + u2 + · · · + uk ) − (1 + 1 + · · · + 1) = n − k.

| {z } | {z }
=n k times
| {z }
=k

Thus, we have shown that (u1 − 1, u2 − 1, . . . , uk − 1) ∈ {0, 1}k and (u1 − 1) + (u2 − 1) + · · · +
(uk − 1) = n − k. In other words, (u1 − 1, u2 − 1, . . . , uk − 1) is a k-tuple ( x1 , x2 , . . . , xk ) ∈ {0, 1}k
satisfying x1 + x2 + · · · + xk = n
n − k. Qed. o
271 Proof. Let ( p1 , p2 , . . . , pk ) ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1}k | x1 + x2 + · · · + xk = n − k . Thus,
k
( p1 , p2 , . . . , pk ) is a k-tuple ( x1 , x2 , . . . , xk ) ∈ {0, 1} satisfying x1 + x2 + · · · + xk = n − k. In
other words, ( p1 , p2 , . . . , pk ) ∈ {0, 1}k is a k-tuple satisfying p1 + p2 + · · · + pk = n − k. We must
prove that ( p1 + 1, p2 + 1, . . . , pk + 1) is a {1, 2}-composition of n into k parts.
We have ( p1 , p2 , . . . , pk ) ∈ {0, 1}k . In other words, p1 , p2 , . . . , pk ∈ {0, 1}. Hence,
p1 + 1, p2 + 1, . . . , pk + 1 ∈ {1, 2} (since q + 1 ∈ {1, 2} for each q ∈ {0, 1}). Hence,
( p1 + 1, p2 + 1, . . . , pk + 1) ∈ {1, 2}k ⊆ Pk (since {1, 2} ⊆ P). In other words,
p
( 1 + 1, p 2 + 1, . . . , p k + 1 ) is a k-tuple of positive integers (since P is the set of all positive inte-
Enumerative Combinatorics: class notes page 438

can define a map

n o
k
B : ( x1 , x2 , . . . , xk ) ∈ {0, 1} | x1 + x2 + · · · + xk = n − k
→ {{1, 2} -compositions of n into k parts} ,
( p1 , p2 , . . . , pk ) 7→ ( p1 + 1, p2 + 1, . . . , pk + 1) .

Clearly, the maps A and B are mutually inverse (since the map A subtracts 1
from each entry of the k-tuple it is applied to, whereas the map B adds 1 to each
entry of the k-tuple it is applied to). Thus, the map A is invertible, i.e., is a bijection.
Hence, the bijection principle yields

|{{1, 2} -compositions of n into k parts}|

n o
= ( x1 , x2 , . . . , xk ) ∈ {0, 1}k | x1 + x2 + · · · + xk = n − k
 
= # of ( x1 , x2 , . . . , xk ) ∈ {0, 1}k satisfying x1 + x2 + · · · + xk = n − k
 
k
= (by Theorem 2.10.4, applied to n − k instead of n) .
n−k

Hence,

(# of {1, 2} -compositions of n into k parts)

= |{{1, 2} -compositions of n into k parts}|
 
k
= .
n−k

This solves Exercise 2.10.5 (b).

(a) Recall that the length of
a list is the number of entries of this list. Thus, the
length of a list a1 , a2 , . . . , a p is the integer p.
Let k ∈ N. Comparing the definitions of “composition of n” and “composition
of n into k parts”, we see that a composition of n into k parts is the same thing as a
composition of n that is a k-tuple. In other words, a composition of n into k parts
is the same thing as a composition of n whose length is k.

gers). Moreover, the sum of the entries of the k-tuple ( p1 + 1, p2 + 1, . . . , pk + 1) is

( p1 + 1) + ( p2 + 1) + · · · + ( pk + 1) = ( p1 + p2 + · · · + pk ) + (1 + 1 + · · · + 1) = (n − k) + k = n.
| {z } | {z }
=n−k k times
| {z }
=k

Hence, ( p1 + 1, p2 + 1, . . . , pk + 1) is a k-tuple of positive integers whose sum is n. In

other words, ( p1 + 1, p2 + 1, . . . , pk + 1) is a composition of n into k parts. There-
fore, ( p1 + 1, p2 + 1, . . . , pk + 1) is a composition ( x1 , x2 , . . . , xk ) of n into k parts such
that x1 , x2 , . . . , xk ∈ {1, 2} (since p1 + 1, p2 + 1, . . . , pk + 1 ∈ {1, 2}). In other words,
( p1 + 1, p2 + 1, . . . , pk + 1) is a {1, 2}-composition of n into k parts (by the definition of a “{1, 2}-
composition of n into k parts”). Qed.
Enumerative Combinatorics: class notes page 439

Thus, comparing the definitions of “{1, 2}-composition of n” and “{1, 2}-composition

of n into k parts”, we conclude that a {1, 2}-composition of n into k parts is the same
thing as a {1, 2}-composition of n whose length is k. Thus,

(# of {1, 2} -compositions of n into k parts)

= (# of {1, 2} -compositions of n whose length is k) .

Hence,

(# of {1, 2} -compositions of n whose length is k)

= (# of {1, 2} -compositions of n into k parts)
 
k
= (318)
n−k

(by Exercise 2.10.5 (b)).

Now, forget that we fixed k. We thus have proved (318) for each k ∈ N.
If s is a {1, 2}-composition of n, then the length of s is an element of {0, 1, . . . , n}
272 . Hence, the sum rule (Theorem 1.2.5) yields

(# of {1, 2} -compositions of n)
= ∑ |(# of {1, 2} -compositions of n whose length is k )
k∈{0,1,...,n}  {z  }
| {z } k
n =
=∑ n−k
k =0 (by (318))
n n
n−k
   
k
= ∑ n−k
= ∑
n − (n − k)
k =0 k =0 |
 {z  }
n−k
=
k
(since n−(n−k)=k)
(here, we have substituted n − k for k in the sum)
n
n−k
 
= ∑ .
k =0
k

272 Proof. Let s be a {1, 2}-composition of n. We must prove that the length of s is an element of
{0, 1, . . . , n}.
We know that s is a {1, 2}-composition of n. In other words, s is a composition ( x1 , x2 , . . . , xk )
of n such that x1 , x2 , . . . , xk ∈ {1, 2} (by the definition of a “{1, 2}-composition of n”). Hence,
s has the form s = ( x1 , x2 , . . . , xk ), where ( x1 , x2 , . . . , xk ) is a composition of n such that
x1 , x2 , . . . , xk ∈ {1, 2}. Consider this ( x1 , x2 , . . . , xk ). From s = ( x1 , x2 , . . . , xk ), we conclude
that the length of s is k.
Lemma 7.48.1 yields k ≤ n. Hence, k ∈ {0, 1, . . . , n} (since k ∈ N). In other words, k is an
element of {0, 1, . . . , n}. In other words, the length of s is an element of {0, 1, . . . , n} (since the
length of s is k). Qed.
Enumerative Combinatorics: class notes page 440

But Proposition 1.3.32 yields

n
n−k
 
f n +1 = ∑ k
.
k =0

Comparing these two equalities, we obtain

(# of {1, 2} -compositions of n) = f n+1 .

This solves Exercise 2.10.5 (a).

7.49. Solution to Exercise 2.10.6

For our solution to Exercise 2.10.6, we will need a minor generalization of (243):

Lemma 7.49.1. Let n ∈ Z and k ∈ N. Then,

n+k−1
   
k
# of ( x1 , x2 , . . . , xk ) ∈ N satisfying x1 + x2 + · · · + xk = n = .
n

Proof of Lemma 7.49.1. If n ∈ N, then this follows immediately from (243). Hence,
for the rest of this proof, we WLOG assume  that n ∈ / N. Hence, (43) (applied to
n+k−1

n + k − 1 and n instead of n and k) yields = 0.
n
But there exists no ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n 273 .
Hence,
 
# of ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n = 0.

n+k−1
 
Comparing this with = 0, we obtain
n

n+k−1
   
k
# of ( x1 , x2 , . . . , xk ) ∈ N satisfying x1 + x2 + · · · + xk = n = .
n

Thus, Lemma 7.49.1 is proved.

273 Proof. Let ( x1 , x2 , . . . , xk ) ∈ Nk . Hence, the k elements x1 , x2 , . . . , xk are k elements of N. Thus,

the sum x1 + x2 + · · · + xk of these k elements must also be an element of N. In other words,
x1 + x2 + · · · + xk ∈ N. If we had x1 + x2 + · · · + xk = n, then this would entail that n = x1 +
x2 + · · · + xk ∈ N, which would contradict n ∈ / N. Hence, we cannot have x1 + x2 + · · · + xk = n.
In other words, we have x1 + x2 + · · · + xk 6= n.
Forget that we fixed ( x1 , x2 , . . . , xk ). We thus have showed that each ( x1 , x2 , . . . , xk ) ∈ Nk
satisfies x1 + x2 + · · · + xk 6= n. In other words, there exists no ( x1 , x2 , . . . , xk ) ∈ Nk satisfying
x1 + x2 + · · · + xk = n.
Enumerative Combinatorics: class notes page 441

Solution to Exercise 2.10.6. Set

n o
k
U = ( x1 , x2 , . . . , x k ) ∈ N | x1 + x2 + · · · + x k = n . (319)

Thus,
n o
k
|U | = ( x1 , x2 , . . . , x k ) ∈ N | x1 + x2 + · · · + x k = n
 
= # of ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n
n+k−1
 
= (by Lemma 7.49.1) .
n

This shows that U is a finite set. Also, clearly, U ⊆ Nk (by the definition of U).
For each i ∈ {1, 2, . . . , k }, we define a subset Ai of U by

Ai = {( x1 , x2 , . . . , xk ) ∈ U | xi ≥ p} . (320)

Then, A1 , A2 , . . . , Ak are k subsets of U. Moreover, for any (y1 , y2 , . . . , yk ) ∈ U and

any i ∈ {1, 2, . . . , k }, we have the logical equivalence

((y1 , y2 , . . . , yk ) ∈ Ai ) ⇐⇒ (yi ≥ p) (321)

(by (320)).
Now, we claim the following:

Statement 1: We have

U \ ( A1 ∪ A2 ∪ · · · ∪ A k )
n o
k
= ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1} | x1 + x2 + · · · + xk = n .

[Proof of Statement 1: It is easy to see that

U \ ( A1 ∪ A2 ∪ · · · ∪ A k )
n o
⊆ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n . (322)

274

274 Proof of (322): Let t ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ). Thus, t ∈ U and t ∈ / A1 ∪ A2 ∪ · · · ∪ Ak . From

n o
t ∈ U = ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n ,

we conclude that t can be written in the form t = ( x1 , x2 , . . . , xk ) for some ( x1 , x2 , . . . , xk ) ∈ Nk

satisfying x1 + x2 + · · · + xk = n. Consider this ( x1 , x2 , . . . , xk ), and denote it by (y1 , y2 , . . . , yk ).
Thus, (y1 , y2 , . . . , yk ) ∈ Nk satisfies y1 + y2 + · · · + yk = n and t = (y1 , y2 , . . . , yk ).
Let i ∈ {1, 2, . . . , k }. If we had t ∈ Ai , then we would have t ∈ Ai ⊆ A1 ∪ A2 ∪ · · · ∪ Ak , which
would contradict t ∈ / A1 ∪ A2 ∪ · · · ∪ Ak . Hence, we do not have t ∈ Ai . In other words, we do
Enumerative Combinatorics: class notes page 442

It is also easy to see that

n o
k
( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1} | x1 + x2 + · · · + xk = n
⊆ U \ ( A1 ∪ A2 ∪ · · · ∪ A k ) . (323)
275

not have (y1 , y2 , . . . , yk ) ∈ Ai (since t = (y1 , y2 , . . . , yk )). Because of the equivalence (321), we can
thus conclude that we do not have yi ≥ p. In other words, we have yi < p. But yi ∈ N (since
(y1 , y2 , . . . , yk ) ∈ Nk ). Hence, yi ∈ {0, 1, . . . , p − 1} (since yi ∈ N and yi < p).
Forget that we fixed i. We thus have proved that yi ∈ {0, 1, . . . , p − 1} for each i ∈ {1, 2, . . . , k}.
Therefore, (y1 , y2 , . . . , yk ) ∈ {0, 1, . . . , p − 1}k . Thus, (y1 , y2 , . . . , yk ) is an ( x1 , x2 , . . . , xk ) ∈
{0, 1, . . . , p − 1}k satisfying x1 + x2 + · · · + xk = n (since y1 + y2 + · · · + yk = n). In other words,
n o
(y1 , y2 , . . . , yk ) ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n .

In view of t = (y1 , y2 , . . . , yk ), this rewrites as

n o
t ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n .

Forget that we fixed t. We thus have showed that

n o
t ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n

for each t ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ). In other words,

n o
U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ) ⊆ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n .

This proves (322). n o

275 Proof of (323): Let s ∈ ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n . Thus, s can
be written in the form s = ( x1 , x2 , . . . , xk ) for some ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k satisfying
x1 + x2 + · · · + xk = n. Consider this ( x1 , x2 , . . . , xk ), and denote it by (y1 , y2 , . . . , yk ). Thus,
(y1 , y2 , . . . , yk ) ∈ {0, 1, . . . , p − 1}k satisfies y1 + y2 + · · · + yk = n and s = (y1 , y2 , . . . , yk ).
We have (y1 , y2 , . . . , yk ) ∈ {0, 1, . . . , p − 1}k ⊆ Nk (since {0, 1, . . . , p − 1} ⊆ N). Hence,
(y1 , y2 , . . . , yk ) is an ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n (since y1 + y2 +
· · · + yk = n). In other words,
n o
( y 1 , y 2 , . . . , y k ) ∈ ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n .

In view of (319), this rewrites as (y1 , y2 , . . . , yk ) ∈ U. Hence, s = (y1 , y2 , . . . , yk ) ∈ U.

Next, we shall prove that s ∈ / A1 ∪ A2 ∪ · · · ∪ Ak . Indeed, assume the contrary. Thus, s ∈
A1 ∪ A2 ∪ · · · ∪ Ak . In other words, s ∈ Ai for some i ∈ {1, 2, . . . , k}. Consider this i. We have
(y1 , y2 , . . . , yk ) = s ∈ Ai . Because of the equivalence (321), this entails that yi ≥ p. But from
(y1 , y2 , . . . , yk ) ∈ {0, 1, . . . , p − 1}k , we obtain yi ∈ {0, 1, . . . , p − 1}, so that yi ≤ p − 1 < p. This
contradicts yi > p. This contradiction shows that our assumption was wrong. Hence, we have
proved that s ∈ / A1 ∪ A2 ∪ · · · ∪ A k .
Combining s ∈ U with s ∈ / A1 ∪ A2 ∪ · · · ∪ Ak , we obtain s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ).
Forget that we fixed s. We thus have proved that s ∈ U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ) for each s ∈
Enumerative Combinatorics: class notes page 443

Combining (322) with (323), we obtain

U \ ( A1 ∪ A2 ∪ · · · ∪ A k )
n o
k
= ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1} | x1 + x2 + · · · + xk = n .

This proves Statement 1.]

Next, we claim the following:

Statement 2: Let I be a subset of [k ]. Then,

n − p |I| + k − 1
 
|{s ∈ U | s ∈ Ai for all i ∈ I }| = .
n − p |I|

[Proof of Statement 2: We shall use the Iverson bracket notation (as defined in
Definition 1.3.15). Corollary 1.6.3 (a) (applied to S = [k ] and T = I) yields

|I| = ∑ [s ∈ I ] = ∑ [i ∈ I ] (324)
s∈[k] i ∈[k]

(here, we have renamed the summation index s as i).

Define two sets

P = {s ∈ U | s ∈ Ai for all i ∈ I } and

n o
Q = ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n − p | I | .

We shall find a bijection from P to Q.

The definition of P shows that P ⊆ U ⊆ Nk , so that every element of P is a
k-tuple of nonnegative integers. The same holds for Q (since the definition of Q
shows that Q ⊆ Nk ).
For each i ∈ {1, 2, . . . , k }, we define an integer qi by qi = p [i ∈ I ]. Thus, we have
defined k integers q1 , q2 , . . . , qk . Their sum is

q1 + q2 + · · · + q k
k
= ∑ qi = ∑ p [i ∈ I ] = p ∑ [i ∈ I ]
i =1 i ∈[k] i ∈[k]
|{z}
|{z} = p [i ∈ I ] | {z }
= ∑ (by the definition of qi ) =| I |
i ∈{1,2,...,k }
(by (324))
= ∑
i ∈[k ]

= p |I| .
n o
( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n . In other words,
n o
( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n ⊆ U \ ( A1 ∪ A2 ∪ · · · ∪ Ak ) .

This proves (323).

Enumerative Combinatorics: class notes page 444

If (y1 , y2 , . . . , yk ) ∈ P, then (y1 − q1 , y2 − q2 , . . . , yk − qk ) ∈ Q 276 . Hence, we can

define a map

α : P → Q,
( y1 , y2 , . . . , y k ) 7 → ( y1 − q1 , y2 − q2 , . . . , y k − q k ) .

If (z1 , z2 , . . . , zk ) ∈ Q, then (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ P 277 . Hence, we can

276 Proof. Let (y1 , y2 , . . . , yk ) ∈ P. Thus,

n o
( y 1 , y 2 , . . . , y k ) ∈ P ⊆ U = ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n .

In other words, (y1 , y2 , . . . , yk ) is an ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n. In

other words, (y1 , y2 , . . . , yk ) is an element of Nk and satisfies y1 + y2 + · · · + yk = n.
Recall that (y1 , y2 , . . . , yk ) ∈ P = {s ∈ U | s ∈ Ai for all i ∈ I }. In other words, (y1 , y2 , . . . , yk )
is an s ∈ U satisfying s ∈ Ai for all i ∈ I. In other words, (y1 , y2 , . . . , yk ) is an element of U and
satisfies
( y1 , y2 , . . . , y k ) ∈ A i for all i ∈ I. (325)
Now, let i ∈ {1, 2, . . . , k }. Then, yi ∈ N (since (y1 , y2 , . . . , yk ) is an element of Nk ). We shall
now show that yi − qi ∈ N. Indeed, this is obvious in the case when qi = 0 (because in this
case, we have yi − qi = yi ∈ N). Hence, for the rest of this proof of yi − qi ∈ N, we WLOG
|{z}
=0
assume that qi 6= 0. Hence, we have i ∈ I (since otherwise, we would have [i ∈ I ] = 0 and thus
qi = p [i ∈ I ] = 0, which would contradict qi 6= 0). Hence, (325) yields (y1 , y2 , . . . , yk ) ∈ Ai .
| {z }
=0
Because of the equivalence (321), this entails that yi ≥ p. Thus, yi − p ≥ 0, so that yi − p ∈ N.
Now, qi = p [i ∈ I ] = p. Hence, yi − qi = yi − p ∈ N. Thus, yi − qi ∈ N is proved.
| {z } |{z}
=1 =p
(since i ∈ I)
Now, forget that we fixed i. We thus have showed that yi − qi ∈ N for each i ∈ {1, 2, . . . , k }.
In other words, (y1 − q1 , y2 − q2 , . . . , yk − qk ) ∈ Nk . Moreover,

( y1 − q1 ) + ( y2 − q2 ) + · · · + ( y k − q k ) = ( y1 + y2 + · · · + y k ) − ( q1 + q2 + · · · + q k ) = n − p | I | .
| {z } | {z }
=n = p| I |

Hence, (y1 − q1 , y2 − q2 , . . . , yk − qk ) is an ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk =

n − p | I | (since (y1 − q1 , y2 − q2 , . . . , yk − qk ) ∈ Nk ). In other words,
n o
( y 1 − q 1 , y 2 − q 2 , . . . , y k − q k ) ∈ ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n − p | I | .

In
n other words, (y1 − q1 , y2 − q2 , . . . , yk − qo
k) ∈ Q (since Q =
( x1 , x2 , . . . , xk ) ∈ N | x1 + x2 + · · · + xk = n − p | I | ). Qed.
k

277 Proof. Let (z1 , z2 , . . . , zk ) ∈ Q. Thus,

n o
( z 1 , z 2 , . . . , z k ) ∈ Q = ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n − p | I | .

In other words, (z1 , z2 , . . . , zk ) is an ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n − p | I |.

In other words, (z1 , z2 , . . . , zk ) is an element of Nk and satisfies z1 + z2 + · · · + zk = n − p | I |.
Enumerative Combinatorics: class notes page 445

define a map

β : Q → P,
( z1 , z2 , . . . , z k ) 7 → ( z1 + q1 , z2 + q2 , . . . , z k + q k ) .

The maps α and β are mutually inverse278 , and thus are bijective. Hence, in

For each i ∈ {1, 2, . . . , k}, we have zi ∈ N (since (z1 , z2 , . . . , zk ) is an element of Nk ) and thus

zi + qi ∈ N.
|{z} |{z}
∈N = p[i ∈ I ]∈N
(since p∈N and [i ∈ I ]∈N)

In other words, (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ Nk . Moreover,

( z1 + q1 ) + ( z2 + q2 ) + · · · + ( z k + q k )
= (z1 + z2 + · · · + zk ) + (q1 + q2 + · · · + qk ) = (n − p | I |) + p | I | = n.
| {z } | {z }
=n− p| I | = p| I |

Hence, (z1 + q1 , z2 + q2 , . . . , zk + qk ) is an ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n

(since (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ Nk ). In other words,
n o
( z 1 + q 1 , z 2 + q 2 , . . . , z k + q k ) ∈ ( x 1 , x 2 , . . . , x k ) ∈ Nk | x 1 + x 2 + · · · + x k = n .

In view of (319), this rewrites as (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ U.

Now, let i ∈ I. We shall show that (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ Ai .
Indeed, i ∈ I, so that [i ∈ I ] = 1. Hence, the definition of qi yields qi = p [i ∈ I ] = p.
| {z }
=1
Also, zi ∈ N (since (z1 , z2 , . . . , zk ) is an element of Nk ) and thus zi ≥ 0. Thus, zi + qi ≥
|{z} |{z}
≥0 =p
0 + p = p. Hence, (z1 + q1 , z2 + q2 , . . . , zk + qk ) is an ( x1 , x2 , . . . , xk ) ∈ U satisfying xi ≥ p (since
(z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ U). In other words,

(z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ {( x1 , x2 , . . . , xk ) ∈ U | xi ≥ p} .

In view of (320), this rewrites as (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ Ai .

Now, forget that we fixed i. We thus have showed that (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ Ai for
all i ∈ I. Hence, (z1 + q1 , z2 + q2 , . . . , zk + qk ) is an s ∈ U that satisfies (s ∈ Ai for all i ∈ I ) (since
(z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ U). In other words,

(z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ {s ∈ U | s ∈ Ai for all i ∈ I } .

In view of P = {s ∈ U | s ∈ Ai for all i ∈ I }, this rewrites as (z1 + q1 , z2 + q2 , . . . , zk + qk ) ∈ P.

Qed.
278 Indeed, the map α subtracts q , q , . . . , q from the entries of the k-tuple that it is applied to,
1 2 k
whereas the map β adds q1 , q2 , . . . , qk to the entries of the k-tuple that it is applied to. Thus, the
maps α and β clearly undo each other. Hence, they are mutually inverse.
Enumerative Combinatorics: class notes page 446

particular, α is a bijection. Thus, the bijection principle yields

n o
k
| P | = | Q | = ( x1 , x2 , . . . , x k ) ∈ N | x1 + x2 + · · · + x k = n − p | I |
 n o
since Q = ( x1 , x2 , . . . , xk ) ∈ Nk | x1 + x2 + · · · + xk = n − p | I |
 
= # of ( x1 , x2 , . . . , xk ) ∈ Nk satisfying x1 + x2 + · · · + xk = n − p | I |
n − p |I| + k − 1
 
= (by Lemma 7.49.1, applied to n − p | I | instead of n) .
n − p |I|

In view of P = {s ∈ U | s ∈ Ai for all i ∈ I }, this rewrites as

n − p |I| + k − 1
 
|{s ∈ U | s ∈ Ai for all i ∈ I }| = .
n − p |I|

This proves Statement 2.]

Now, let us note that each subset I of [k ] satisfies | I | ∈ {0, 1, . . . , k } (since Theo-
rem 1.4.7 (b) yields | I | ≤ |[k]| = k).
But Statement 1 yields

U \ ( A1 ∪ A2 ∪ · · · ∪ A k )
n o
k
= ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1} | x1 + x2 + · · · + xk = n .

Hence,

|U \ ( A1 ∪ A2 ∪ · · · ∪ Ak )|
n o
= ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k | x1 + x2 + · · · + xk = n
 
= # of ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k satisfying x1 + x2 + · · · + xk = n .
Enumerative Combinatorics: class notes page 447

Thus,
 
# of ( x1 , x2 , . . . , xk ) ∈ {0, 1, . . . , p − 1}k satisfying x1 + x2 + · · · + xk = n
= |U \ ( A1 ∪ A2 ∪ · · · ∪ Ak )|
= ∑ (−1)| I | |{s ∈ U | s ∈ Ai for all i ∈ I }|
I ⊆[k]
| {z }
n − p |I| + k − 1
 
=
n − p |I|
(by Statement 2)

(by Theorem 2.9.9 (applied to k instead of n))

k
|I| n − p | I | + k − 1 |I| n − p | I | + k − 1
   
= ∑ (−1) = ∑ ∑ (−1)
I ⊆[k]
n − p |I| j=0 I ⊆[k]; |
n − p |I|
{z }
| I |= j
− −

n pj + k 1
=(−1) j
n − pj
(since | I |= j)
 
by an application of (37), since
each subset I of [k ] satisfies | I | ∈ {0, 1, . . . , k }
k
j n − pj + k − 1
 
= ∑ ∑ (−1) n − pj
j =0 I ⊆[k];
| I |= j
| {z }
n − pj + k − 1
 
j
=(# of subsets j of [k] satisfying | I |= j)·(−1)
n − pj
k
j n − pj + k − 1
 
= ∑ (# of subsets j of [k] satisfying | I | = j) · (−1)
j =0
| {z   } n − pj
k
=(# of j-element subsets of [k])=
j
(by Theorem 1.3.12, applied to [k], k and j
instead of S, n and k)
k k
j n − pj + k − 1 n − pj + k − 1
      
k k
=∑ · (−1) = ∑ (−1) j
j
.
j =0
j n − pj j =0
n − pj

This solves Exercise 2.10.6.

7.50. Solution to Exercise 2.10.7

Our solution to Exercise 2.10.7 is taken from [Grinbe16b]. We first recall some
notations and basic facts from number theory:
Enumerative Combinatorics: class notes page 448

Definition 7.50.1. If a is an integer and b is a positive integer, then

• we let a // b denote the quotient obtained when dividing a by b (in the sense
of division with remainder);

• we let a%b denote the remainder obtained when dividing a by b.

Proposition 7.50.2. Let b be a positive integer.

(a) We have a = ( a // b) · b + ( a%b) for every a ∈ Z.
(b) We have a%b ∈ {0, 1, . . . , b − 1} for every a ∈ Z.
(c) If a ∈ N, then a // b ∈ N.
(d) Any q ∈ Z and r ∈ {0, 1, . . . , b − 1} satisfy (qb + r ) // b = q.
(e) Any q ∈ Z and r ∈ {0, 1, . . . , b − 1} satisfy (qb + r ) %b = r.

Proof of Proposition 7.50.2. (a) Proposition 7.50.2 (a) is [19s, Corollary 2.6.9 (d)] (with
u and n renamed as a and b).
(b) Proposition 7.50.2 (b) is the first claim of [19s, Corollary 2.6.9 (a)] (with u and
n renamed as a and b).
(c) Let a ∈ N. Then, Proposition 7.50.2 (b) yields a%b ∈ {0, 1, . . . , b − 1}, so that
a%b ≤ b − 1 < b. But Proposition 7.50.2 (a) yields

a = ( a // b) · b + ( a%b) < ( a // b) · b + b = (( a // b) + 1) b.
| {z }
<b

Hence,
(( a // b) + 1) b > a ≥ 0 (since a ∈ N) .
We can divide both sides of this inequality by b (since b is positive), and thus obtain
( a // b) + 1 > 0. In other words, a // b > −1. Since a // b ∈ Z, we thus conclude
that a // b ≥ (−1) + 1 = 0. In other words, a // b ∈ N. This proves Proposition
7.50.2 (c).
(d) Proposition 7.50.2 (d) is [19s, Exercise 2.6.4 (c)] (with n renamed as b).
(e) Proposition 7.50.2 (e) is [19s, Exercise 2.6.4 (d)] (with n renamed as b).
Solution to Exercise 2.10.7. This will be a proof by double counting. We shall count
the n-tuples ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k in two different
ways.
The first way is easy: The equality (243) (applied to k and n instead of n and k)
shows that
k+n−1
 
n
(# of ( x1 , x2 , . . . , xn ) ∈ N satisfying x1 + x2 + · · · + xn = k) =
k
n+k−1
 
= (326)
k
Enumerative Combinatorics: class notes page 449

(since k + n = n + k).
The second way is trickier. Let me first outline the idea, before showing the
argument in full detail.

Outline: It is well-known (a particular case of division with remainder) that any

nonnegative integer x can be written as x = 2q + r for a unique q ∈ N and a
unique r ∈ {0, 1}. Thus, each n-tuple ( x1 , x2 , . . . , xn ) ∈ Nn can be written uniquely
in the form (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) for some q1 , q2 , . . . , qn ∈ N and some
r1 , r2 , . . . , rn ∈ {0, 1}. In other  words, for each n-tuple  ( x 1 , x 2 , . . . , x n ) ∈ Nn ,
q1 q2 · · · q n
there exists a unique 2 × n-table with q1 , q2 , . . . , qn ∈ N and
r1 r2 · · · r n
r1 , r2 , . . . , rn ∈ {0, 1} satisfying ( x1 , x2 , . . . , xn ) = (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ).
Moreover, an n-tuple ( x1 , x2 , . . . , xn )∈ Nn satisfies x1  + x2 + · · · + xn = k if and
q1 q2 · · · q n
only if the corresponding 2 × n-table satisfies 2 (q1 + q2 + · · · + qn ) +
r1 r2 · · · r n
(r1 + r2 + · · · + rn ) = k. Thus, there is a bijection
  
q1 q2 · · · q n
2 × n-tables with q1 , q2 , . . . , qn ∈ N and r1 , r2 , . . . , rn ∈ {0, 1}
r1 r2 · · · r n

satisfying 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k

→ {n-tuples ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k}
that sends
 
q1 q2 · · · qn
each 2 × n-table
r1 r2 · · · rn
to the n-tuple (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) .

Hence, the bijection principle yields

(# of n-tuples ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k)
= (# of nice tables) , (327)
 
q1 q2 · · · q n
where a “nice table” shall mean a 2 × n-table with q1 , q2 , . . . , qn ∈
r1 r2 · · · r n
N and r1 , r2 , . . . , rn ∈ {0, 1} satisfying 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) =
k. It thus remains to count the nice tables.  But we can do  this using the sum
q1 q2 · · · q n
rule: Define the top-sum of a nice table to be the sum q1 +
r1 r2 · · · r n
q2 + · · · + qn . The top-sum of a nice table is always a nonnegative integer ≤ k
(actually, ≤ bk/2c, but we will not use this stronger statement), i.e., an element of
{0, 1, . . . , k}. Moreover, for each i ∈ {0, 1, . . . , k}, the product rule yields
n+i−1
  
n
(# of nice tables with top-sum i ) = ,
i k − 2i
Enumerative Combinatorics: class notes page 450

 
q1 q2 · · · qn
because we can construct a nice table with top-sum i by the
r1 r2 · · · rn
following algorithm:

• First, choose the n entries q1 , q2 , . . . , qn of the top row of the table. These n
entries must belong to N and sum to q1 + q2 + · · · + qn = i; in other words, we
are choosing an n-tuple (q1 , q2 , . . . , qn ) ∈ Nn satisfying q1 + q2 + · · · + qn = i.
Thus, (243) (applied to i 
and n
 instead of  n and k) shows that the # of options
i+n−1 n+i−1

at this step is = .
i i
• Then, choose the n entries r1 , r2 , . . . , rn of the bottom row of the table. These n
entries must belong to {0, 1} and sum to r1 + r2 + · · · + rn = k − 2i (since we
want to have 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k, but we have q1 +
q2 + · · · + qn = i); in other words, we are choosing an n-tuple (r1 , r2 , . . . , rn ) ∈
{0, 1}n satisfying r1 + r2 + · · · + rn = k − 2i. Thus, Theorem 2.10.4 (applied
 k − 2i and n instead of n and k) shows that the # of options at this step is
to
n
.
k − 2i
Combining these observations using the sum rule, we obtain (# of nice tables) =
n+i−1
  
k n
∑ . In view of (327), this rewrites as
i =0 i k − 2i

(# of n-tuples ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k)
k 
n+i−1
 
n
=∑ .
i =0
i k − 2i

Comparing this with (326), we obtain the claim of the exercise.

Detailed argument: Here comes the detailed (and formalized) version of the argu-
ment we just outlined.  
279 q1 q2 · · · q n
We define a nice table to mean a 2 × n-table with q1 , q2 , . . . , qn ∈
r1 r2 · · · r n
N and r1 , r2 , . . . , rn ∈ {0, 1} satisfying 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) =
k.  
q1 q2 · · · q n
The top-sum of a nice table is defined to be the sum q1 +
r1 r2 · · · r n
q2 + · · · + q n .
We now state our first observation:

Statement 1: We have

(# of ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k)
= (# of nice tables) .
279 i.e., a rectangular table with 2 rows and n columns
Enumerative Combinatorics: class notes page 451

 
q1 q2 · · · q n
[Proof of Statement 1: If is a nice table, then
r1 r2 · · · r n
(2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) is an n-tuple ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 +
x2 + · · · + xn = k 280 . Hence, we can define a map

A : {nice tables} → {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = k } ,
 
q1 q2 · · · q n
7→ (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) .
r1 r2 · · · r n

Consider this map A.

Now, if ( x1 , x2 , . . . , xn ) ∈ Nn is an n-tuple satisfying x1 + x2 + · · · + xn = k, then
 
x1 // 2 x2 // 2 · · · xn // 2
x1 %2 x2 %2 · · · xn %2

 
280 Proof. q1 q2 · · · q n
Let be a nice table. According to the definition of a “nice ta-
r1 r2 · · · r n
ble”, this means that q1 , q2 , . . . , qn ∈ N and r1 , r2 , . . . , rn ∈ {0, 1} and 2 (q1 + q2 + · · · + qn ) +
(r1 + r2 + · · · + rn ) = k. Hence, for each i ∈ [n], we have qi ∈ N (since q1 , q2 , . . . , qn ∈ N)
and ri ∈ {0, 1} (since r1 , r2 , . . . , rn ∈ {0, 1}) and thus, 2 qi + ri ∈ N. In other
|{z} |{z}
∈N ∈{0,1}⊆N
words, the n numbers 2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn all belong to N. In other words,
(2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) ∈ Nn .
Hence, (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) is an n-tuple ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 +
· · · + xn = k (since

(2q1 + r1 ) + (2q2 + r2 ) + · · · + (2qn + rn ) = 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k

). Qed.
Enumerative Combinatorics: class notes page 452

is a nice table281 . Hence, we can define a map

B : {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = k } → {nice tables} ,
 
x1 // 2 x2 // 2 · · · xn // 2
( x1 , x2 , . . . , x n ) 7 → .
x1 %2 x2 %2 · · · xn %2
281 Proof. Let ( x1 , x2 , . . . , xn ) ∈ Nn be an n-tuple satisfying x1 + x2 + · · · + xn = k. We have
x1 , x2 , . . . , xn ∈ N (since ( x1 , x2 , . . . , xn ) ∈ Nn ). Thus, for each i ∈ [n], we have xi ∈ N and there-
fore xi // 2 ∈ N (by Proposition 7.50.2 (c), applied to a = xi and b = 2). In other words, the n
numbers x1 // 2, x2 // 2, . . . , xn // 2 all belong to N. In other words, x1 // 2, x2 // 2, . . . , xn // 2 ∈ N.
Furthermore, for each i ∈ [n], we have

xi %2 ∈ {0, 1, . . . , 2 − 1} (by Proposition 7.50.2 (b), applied to a = xi and b = 2)

= {0, 1} .

In other words, the n numbers x1 %2, x2 %2, . . . , xn %2 all belong to {0, 1}. In other words,
x1 %2, x2 %2, . . . , xn %2 ∈ {0, 1}.
Finally, for each i ∈ [n], we have xi = ( xi // 2) · 2 + ( xi %2) (by Proposition 7.50.2 (a), applied
to a = xi and b = 2). In other words, we have the n equalities

x1 = ( x1 // 2) · 2 + ( x1 %2) ,
x2 = ( x2 // 2) · 2 + ( x2 %2) ,
..
.
xn = ( xn // 2) · 2 + ( xn %2) .

Adding these n equalities together, we find

x1 + x2 + · · · + x n
= (( x1 // 2) · 2 + ( x1 %2)) + (( x2 // 2) · 2 + ( x2 %2)) + · · · + (( xn // 2) · 2 + ( xn %2))
= (( x1 // 2) + ( x2 // 2) + · · · + ( xn // 2)) · 2 + (( x1 %2) + ( x2 %2) + · · · + ( xn %2))
= 2 (( x1 // 2) + ( x2 // 2) + · · · + ( xn // 2)) + (( x1 %2) + ( x2 %2) + · · · + ( xn %2)) .

Comparing this with x1 + x2 + · · · + xn = k, we obtain

2 (( x1 // 2) + ( x2 // 2) + · · · + ( xn // 2)) + (( x1 %2) + ( x2 %2) + · · · + ( xn %2)) = k.

Thus, we have proved that x1 // 2, x2 // 2, . . . , xn // 2 ∈ N and x1 %2, x2 %2, . . . , xn %2 ∈ {0, 1}

and

2 (( x1 // 2) + ( x2 // 2) + · · · + ( xn // 2)) + (( x1 %2) + ( x2 %2) + · · · + ( xn %2)) = k.

In other words,  
x1 // 2 x2 // 2 · · · xn // 2
x1 %2 x2 %2 · · · xn %2
 
q1 q2 · · · q n
is a 2 × n-table with q1 , q2 , . . . , qn ∈ N and r1 , r2 , . . . , rn ∈ {0, 1} satisfying
r1 r2 · · · r n
2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k. In other words,
 
x1 // 2 x2 // 2 · · · xn // 2
x1 %2 x2 %2 · · · xn %2

is a nice table (because this is how a “nice table” was defined). Qed.
Enumerative Combinatorics: class notes page 453

Consider this map B.

It is easy to see that A ◦ B = id 282 and B ◦ A = id 283 . Combining these two
equalities, we conclude that the maps A and B are mutually inverse. Hence, the

282 Proof. Let x ∈ {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = k }. Thus, x can be written in the form

x = ( x1 , x2 , . . . , xn ) for some ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x 1 + x2 + · · · + xn = k. Consider this

x1 // 2 x2 // 2 · · · xn // 2
( x1 , x2 , . . . , xn ). The definition of B yields B ( x1 , x2 , . . . , xn ) = .
x1 %2 x2 %2 · · · xn %2
Applying the map B to both sides of the equality x = ( x1 , x2 , . . . , xn ), we obtain
 
x1 // 2 x2 // 2 · · · xn // 2
B ( x ) = B ( x1 , x2 , . . . , x n ) = .
x1 %2 x2 %2 · · · xn %2

Applying the map A to both sides of this equality, we find

 
x1 // 2 x2 // 2 · · · xn // 2
A ( B (x)) = A
x1 %2 x2 %2 · · · xn %2
= (2 ( x1 // 2) + ( x1 %2) , 2 ( x2 // 2) + ( x2 %2) , . . . , 2 ( xn // 2) + ( xn %2)) (328)

(by the definition of the map A).

On the other hand, for each i ∈ [n], we have

xi = ( xi // 2) · 2 + ( xi %2) (by Proposition 7.50.2 (a), applied to a = xi and b = 2)

= 2 ( xi // 2) + ( xi %2) .

In other words, we have the n equalities

x1 = 2 ( x1 // 2) + ( x1 %2) ,
x2 = 2 ( x2 // 2) + ( x2 %2) ,
..
.
xn = 2 ( xn // 2) + ( xn %2) .

Hence,

( x1 , x2 , . . . , xn ) = (2 ( x1 // 2) + ( x1 %2) , 2 ( x2 // 2) + ( x2 %2) , . . . , 2 ( xn // 2) + ( xn %2)) .

Comparing this with (328), we obtain

A ( B (x)) = ( x1 , x2 , . . . , xn ) = x = id (x) .

Hence, ( A ◦ B) (x) = A ( B (x)) = id (x).

Forget that we fixed x. We thus have showed that ( A ◦ B) (x) = id (x) for each x ∈
{( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = k}. In other words, A ◦ B = id.
283 Proof. Let T ∈ {nice tables}. We shall show that ( B ◦ A ) ( T ) = T.

We have  T ∈ {nice tables}; in other words, T is a nice table. In other words, T is a
q1 q2 · · · q n
2 × n-table with q1 , q2 , . . . , qn ∈ N and r1 , r2 , . . . , rn ∈ {0, 1} satisfying
r1 r2 · · · r n
2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k (since this is how a “nice table” is defined). Con-
sider these q1 , q2 , . . . , qn and r1 , r2 , . . . , rn . Thus,
 
q1 q2 · · · q n
T= . (329)
r1 r2 · · · r n
Enumerative Combinatorics: class notes page 454

map B is invertible, i.e., is a bijection. Thus, the bijection principle yields

|{( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = k}|
= |{nice tables}| .

In other words,

(# of ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k)
= (# of nice tables) .

This proves Statement 1.]

The next observation is simpler:

Statement 2: The top-sum of any nice table belongs to {0, 1, . . . , k }.

Applying the map A to both sides of this equality, we obtain
 
q1 q2 · · · q n
A (T) = A = (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) (330)
r1 r2 · · · r n

(by the definition of A). Define an integer xi by xi = 2qi + ri for each i ∈ [n]. Thus,

( x1 , x2 , . . . , xn ) = (2q1 + r1 , 2q2 + r2 , . . . , 2qn + rn ) .

Comparing this with (330), we obtain A (T) = ( x1 , x2 , . . . , xn ). Applying the map B to both sides
of this equality, we find
 
x1 // 2 x2 // 2 · · · xn // 2
B ( A (T)) = B ( x1 , x2 , . . . , xn ) = (331)
x1 %2 x2 %2 · · · xn %2

(by the definition of B).

Now, let i ∈ [n]. The definition of xi yields xi = 2qi + ri = qi · 2 + ri . From q1 , q2 , . . . , qn ∈ N,
we obtain qi ∈ N ⊆ Z. Also, from r1 , r2 , . . . , rn ∈ {0, 1}, we obtain ri ∈ {0, 1} = {0, 1, . . . , 2 − 1}.
Hence, Proposition 7.50.2 (d) (applied to b = 2 and q = qi and r = ri ) yields (qi · 2 + ri ) // 2 = qi .
In view of xi = qi · 2 + ri , this rewrites as xi // 2 = qi . Furthermore, Proposition 7.50.2 (e) (applied
to b = 2 and q = qi and r = ri ) yields (qi · 2 + ri ) %2 = ri . In view of xi = qi · 2 + ri , this rewrites
as xi %2 = ri .
Now, forget that we fixed i. We thus have proved that

xi // 2 = qi and xi %2 = ri

for each i ∈ [n]. In other words,

   
x1 // 2 x2 // 2 ··· xn // 2 q1 q2 ··· qn
= .
x1 %2 x2 %2 ··· xn %2 r1 r2 ··· rn

Hence, (331) becomes

   
x1 // 2 x2 // 2 ··· xn // 2 q1 q2 ··· qn
B ( A (T)) = = =T
x1 %2 x2 %2 ··· xn %2 r1 r2 ··· rn

(by (329)). Thus, ( B ◦ A) (T) = B ( A (T)) = T = id (T).

Forget that we fixed T. We thus have proved that ( B ◦ A) (T) = id (T) for each T ∈
{ nice tables}. In other words, B ◦ A = id.
Enumerative Combinatorics: class notes page 455

[Proof of Statement 2: Let T be a nice table. We must show that the top-sum of T
belongs to {0, 1, . . . , k }.
We have assumed that T is a nice table.  According to  the definition of a “nice
q1 q2 · · · q n
table”, this means that T is a 2 × n-table with q1 , q2 , . . . , qn ∈ N
r1 r2 · · · r n
and r1 , r2 , . . . , rn ∈ {0, 1} satisfying 2 (q1 + q2 + · · · + qn ) + ( r1 + r2 + · · · + r n )  =
q1 q2 · · · q n
k. Consider these q1 , q2 , . . . , qn and r1 , r2 , . . . , rn . Thus, T = .
r1 r2 · · · r n
Hence, the top-sum of T is q1 + q2 + · · · + qn (by the definition of “top-sum”).
We have r1 , r2 , . . . , rn ∈ {0, 1}. Thus, r1 , r2 , . . . , rn are nonnegative integers. Hence,
their sum r1 + r2 + · · · + rn is a nonnegative integer. Thus, r1 + r2 + · · · + rn ≥ 0.
Moreover, from q1 , q2 , . . . , qn ∈ N, we obtain q1 + q2 + · · · + qn ∈ N. Thus,
q1 + q2 + · · · + qn ≥ 0, so that
2 ( q1 + q2 + · · · + q n ) ≥ q1 + q2 + · · · + q n .
But from 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k, we obtain
k = 2 ( q1 + q2 + · · · + q n ) + (r1 + r2 + · · · + r n )
| {z }
≥0
≥ 2 ( q1 + q2 + · · · + q n ) ≥ q1 + q2 + · · · + q n .
In other words, q1 + q2 + · · · + qn ≤ k. Combining this with q1 + q2 + · · · + qn ∈ N,
we obtain q1 + q2 + · · · + qn ∈ {0, 1, . . . , k }. In other words, q1 + q2 + · · · + qn
belongs to {0, 1, . . . , k }. In other words, the top-sum of T belongs to {0, 1, . . . , k }
(since the top-sum of T is q1 + q2 + · · · + qn ). This completes the proof of Statement
2.]
Finally, let us count the nice tables with a given top-sum i ∈ {0, 1, . . . , k }:
Statement 3: Let i ∈ {0, 1, . . . , k }. Then,
n+i−1
  
n
(# of nice tables having top-sum i ) = .
i k − 2i

[Proof of Statement 3: Define a set

Q = {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = i } .
Thus,
| Q| = |{( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = i }|
= (# of ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = i )
i+n−1
 
=
i
(by (243) (applied to i and n instead of n and k))
n+i−1
 
= (since i + n = n + i ) . (332)
i
Enumerative Combinatorics: class notes page 456

Define a set

R = ( x1 , x2 , . . . , xn ) ∈ {0, 1}n | x1 + x2 + · · · + xn = k − 2i .


Thus,

| R| = ( x1 , x2 , . . . , xn ) ∈ {0, 1}n | x1 + x2 + · · · + xn = k − 2i


= # of ( x1 , x2 , . . . , xn ) ∈ {0, 1}n satisfying x1 + x2 + · · · + xn = k − 2i

 
n
= (333)
k − 2i
(by Theorem 2.10.4 (applied to k − 2i and n instead of n and k)) .
 
q1 q2 · · · q n
Now, if is a nice table having top-sum i, then
r1 r2 · · · r n

((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × R
Enumerative Combinatorics: class notes page 457

284 . Thus, we can define a map

C : {nice tables having top-sum i } → Q × R,

 
q1 q2 · · · q n
7→ ((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) .
r1 r2 · · · r n

Consider this map C.

On the other hand, any element of Q × R has the form ((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn ))
for some (q1 , q2 , . . . , qn ) ∈ Nn and some (r1 , r2 , . . . , rn ) ∈ {0, 1}n (since any element
of Q has the form (q1 , q2 , . . . , qn ) for some (q1 , q2 , . . . , qn ) ∈ Nn , whereas any ele-
ment of R has the form (r1 , r2 , . . . , rn ) for some (r1 , r2 , . . . , rn ) ∈ {0, 1}n ). For any
((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × R, we have
 
q1 q2 · · · q n
∈ {nice tables having top-sum i }
r1 r2 · · · r n

 
284 Proof. q1 q2 · · · q n
Let be a nice table having top-sum i. We must prove that
r1 r2 · · · r n
((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × 
R.
q1 q2 · · · q n
We know that is a nice table. According to the definition of a “nice
r1 r2 · · · r n
table”, this means that q1 , q2 , . . . , qn ∈ N and r1 , r2 , . . . , rn ∈ {0, 1} and 2 (q1 + q2 + · · · + qn ) +
(r1 + r2 + · · · + rn ) = k.  
q1 q2 · · · q n
Also, we know that the nice table has top-sum i. Hence, the top-sum of
r1 r2 · · · r n
 
q1 q2 · · · q n
the nice table is i. Thus,
r1 r2 · · · r n
  
q1 q2 ··· qn
i= the top-sum of the nice table = q1 + q2 + · · · + q n
r1 r2 ··· rn

(by the definition of “top-sum”). Thus, q1 + q2 + · · · + qn = i. Moreover, from

2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k, we obtain

r1 + r2 + · · · + rn = k − 2 (q1 + q2 + · · · + qn ) = k − 2i.
| {z }
=i

Now, (q1 , q2 , . . . , qn ) is an element of Nn (since q1 , q2 , . . . , qn ∈ N) and satisfies q1 + q2 + · · · +

qn = i. In other words, (q1 , q2 , . . . , qn ) is an ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = i.
In other words, (q1 , q2 , . . . , qn ) ∈ {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = i }. This rewrites
as (q1 , q2 , . . . , qn ) ∈ Q (since Q = {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = i }).
Also, (r1 , r2 , . . . , rn ) is an element of {0, 1}n (since r1 , r2 , . . . , rn ∈ {0, 1}) and satis-
fies r1 + r2 + · · · + rn = k − 2i. In other words, (r1 , r2 , . . . , rn ) is an ( x1 , x2 , . . . , xn ) ∈
{0, 1}n satisfying x1 + x2 + · · · + xn = k − 2i. In other words, (r1 , r2 , . . . , rn ) ∈
( x1 , x2 , . . ., xn ) ∈ {0, 1}n | x1 + x2 + · · · + xn = k − 2i . This rewrites as (r1 , r2 , . . . , rn ) ∈ R
(since R = ( x1 , x2 , . . . , xn ) ∈ {0, 1}n | x1 + x2 + · · · + xn = k − 2i ).
Combining (q1 , q2 , . . . , qn ) ∈ Q with (r1 , r2 , . . . , rn ) ∈ R, we obtain
((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × R. Qed.
Enumerative Combinatorics: class notes page 458

285 . Hence, we can define a map

D : Q × R → {nice tables having top-sum i } ,

 
q1 q2 · · · q n
((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) 7→ .
r1 r2 · · · r n

Consider this map D.

The maps C and D are mutually inverse286 . Hence, the map C is invertible, i.e.,
is a bijection. Thus, the bijection principle yields

|{nice tables having top-sum i }|

= | Q × R| = | Q| · | R| (by Theorem 1.1.4, applied to X = Q and Y = R) .

 
285 Proof. q1 q2 ··· qn
Let ((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × R. We must prove that ∈
r1 r2 ··· rn
{nice tables having top-sum i }.
We have ((q1 , q2 , . . . , qn ) , (r1 , r2 , . . . , rn )) ∈ Q × R. In other words, (q1 , q2 , . . . , qn ) ∈ Q and
(r1 , r2 , . . . , rn ) ∈ R.
We have (q1 , q2 , . . . , qn ) ∈ Q = {( x1 , x2 , . . . , xn ) ∈ Nn | x1 + x2 + · · · + xn = i }. In other
words, (q1 , q2 , . . . , qn ) is an ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = i. In other
words, (q1 , q2 , . . . , qn ) ∈ Nn and q1 + q2 + · · · + qn = i.
We have (r1 , r2 , . . . , rn ) ∈ R = ( x1 , x2 , . . . , xn ) ∈ {0, 1}n | x1 + x2 + · · · + xn = k − 2i . In
other words, (r1 , r2 , . . . , rn ) is an ( x1 , x2 , . . . , xn ) ∈ {0, 1}n satisfying x1 + x2 + · · · + xn = k − 2i.
In other words, (r1 , r2 , . . . , rn ) ∈ {0, 1}n and r1 + r2 + · · · + rn = k − 2i.
Now,
2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = 2i + (k − 2i ) = k.
| {z } | {z }
=i =k−2i

Now, we have q1 , q2 , . . . , qn ∈ N (since (q1 , q2 , . . . , qn ) ∈ Nn ) and r1 , r2 , . . . , rn ∈ {0, 1} (since

(r1 , r2 , . . . , rn ) ∈ {0, 1}n ).  
q1 q2 · · · q n
Thus, we know that the 2 × n-table satisfies q1 , q2 , . . . , qn ∈ N and
r1 r2 · · · r n
r1 , r2 , . . . , rn ∈ {0, 1 } and 2 (q1 + q2 + · · · + qn ) + (r1 + r2 + · · · + rn ) = k. In other words,
q1 q2 · · · q n
is a nice table (because this is how a “nice table” was defined). Fur-
r1 r2 · · · r n
 
q1 q2 · · · q n
thermore, the top-sum of this nice table is
r1 r2 · · · r n

q1 + q2 + · · · + q n (by the definition of the “top-sum”)

= i.
 
q1 q2 · · · q n
Hence, is a nice table having top-sum i. In other words,
r1 r2 · · · r n
 
q1 q2 · · · q n
∈ {nice tables having top-sum i }. Qed.
r1 r2 · · · r n
286 Proof. This follows immediately from the definitions of C and D.
Enumerative Combinatorics: class notes page 459

Hence,

(# of nice tables having top-sum i )

= |{nice tables having top-sum i }|
n+i−1
  
n
= | Q| · | R| = .
|{z} |{z} i k − 2i
n+i−1
   
n
= =
i k − 2i
(by (332)) (by (333))

This proves Statement 3.]

Now, we have almost solved the exercise. Statement 2 shows that the top-sum of
any nice table belongs to {0, 1, . . . , k}. Hence, the sum rule (specifically, Theorem
1.2.5) yields

(# of nice tables) = ∑ (# of nice tables having top-sum i )

i ∈{0,1,...,k}
| {z }
n+i−1
 
n
=
i k − 2i
(by Statement 3)
k
n+i−1 n+i−1
     
n n
= ∑ i k − 2i
= ∑ i k − 2i
.
i ∈{0,1,...,k} i =0
| {z }
k
=∑
i =0

But Statement 1 yields

(# of ( x1 , x2 , . . . , xn ) ∈ Nn satisfying x1 + x2 + · · · + xn = k)
k 
n+i−1
 
n
= (# of nice tables) = ∑ .
i =0
i k − 2i

Comparing this with (326), we find

k
n+i−1 n+k−1
    
n
∑ i k − 2i
=
k
.
i =0

This solves Exercise 2.10.7.

7.51. Solution to Exercise 2.10.8

Exercise 2.10.8 wants us to generalize Exercise 2.10.7 to all n ∈ R. Here is this
generalization:
Enumerative Combinatorics: class notes page 460

Proposition 7.51.1. Let n ∈ R and k ∈ N. Prove that

k
n+i−1 n+k−1
    
n
∑ i k − 2i
=
k
.
i =0

Proof of Proposition 7.51.1. For each x ∈ N, we have

k
x+i−1 x+k−1
    
x
∑ i k − 2i
=
k
(334)
i =0

(by Exercise 2.10.7 (applied to x instead of n)).

Define two polynomials P and Q (in 1 variable X, with real coefficients) by

k
X+i−1 X+k−1
    
X
P= ∑ i k − 2i
and Q=
k
.
i =0

These are well-defined polynomials287 .

Now, for each x ∈ R, we have

x+k−1 X+k−1
    
Q (x) = by Proposition 2.6.12, since Q =
k k

and
k
x+i−1
  
x
P (x) = ∑ i k − 2i
(likewise) .
i =0
In view of these two equalities, we can rewrite (334) as

P (x) = Q (x) .

287 They can be written explicitly as follows:

k
X+i−1
  
X
P= ∑ i k − 2i
i =0
k
( X + i − 1) ( X + i − 2) · · · ( X + 1) X
= ∑ i!
i =0

 X ( X − 1) ( X − 2) · · · ( X − (k − 2i ) + 1) , if k − 2i ∈ N;

· (k − 2i )!
/N

0, if k − 2i ∈

and
X+k−1 ( X + k − 1) ( X + k − 2) · · · ( X + 1) X
 
Q= = .
k k!
Enumerative Combinatorics: class notes page 461

Hence, we know that P ( x ) = Q ( x ) holds for all x ∈ N (because we know that (334)
holds for all x ∈ N). Therefore, Corollary 2.6.10 yields P = Q. Thus, P ( x ) = Q ( x )
for all x ∈ R. In other words,
k
x+i−1 x+k−1
    
x
∑ i k − 2i
=
k
i =0

x+k−1 x+i−1
    
k x
for all x ∈ R (since Q ( x ) = and P ( x ) = ∑ for
k i =0 i k − 2i
each x ∈ R). Applying this to x = n, we obtain

k
n+i−1 n+k−1
    
n
∑ i k − 2i
=
k
.
i =0

This proves Proposition 7.51.1.

Thus, Exercise 2.10.8 is solved.

7.52. Solution to Exercise 2.10.9

Exercise 2.10.9 is [18s-mt2s, Exercise 6], and two solutions can be found on the
course website (1 and 2). Let us give another, based on the following lemma:

Lemma 7.52.1. Let n be a positive integer. Then,

n 
n−1

∑ k k − 1 = ( n + 1 ) 2n −2 .
k =0

Proof of Lemma 7.52.1. We have n − 1 ∈ N (since n is a positive integer).

Enumerative Combinatorics: class notes page 462

n−1
 
n
We can split off the addend for k = 0 from the sum ∑ k . This yields
k =0 k−1
n n  n 
n−1 n−1 n−1 n−1
     
∑ k k−1 =0
0−1
+∑k
k − 1
= ∑k
k−1
k =0 | {z } k=1 k =1
=0
n −1
n−1
 
= ∑ ( k + 1)
k =0
( k + 1) − 1
|  {z  }
n−1
=
k
(here, we have substituted k + 1 for k in the sum)
n −1 n −1  
n−1 n−1 n−1
    
= ∑ ( k + 1) = ∑ k +
k =0
k k = 0
k k
| {z  }
n−1 n−1
 
=k +
k k
n −1 n −1 
n−1 n−1
  
= ∑k k + ∑ k
k =0 k =0
| {z } | {z }
=(n−1)·2(n−1)−1 =2n −1
(by Exercise 1.3.6, (by Corollary 1.3.27,
applied to n−1 instead of n) applied to n−1 instead of n)

= (n − 1) · 2| (n− 1)−1 n −1
{z } + 2|{z} = (n − 1) · 2
n −2
+ 2 · 2n −2
=2n −2 =2 ·2n −2
n −2
= ((n − 1) + 2) 2 = ( n + 1 ) 2n −2 .
| {z }
= n +1

This proves Lemma 7.52.1.

Solution to Exercise 2.10.9. Let P = {1, 2, 3, . . .} be the set of all positive integers.
If α is any composition of n, then (the length of α) ∈ {0, 1, . . . , n} 288 .
Let k ∈ N. Then, a k-tuple of positive integers is the same as an element of Pk
(since P is the set of all positive integers).
Recall that a composition of n is the same as a tuple of positive integers whose

288 Proof.Let α be a composition of n.

Write α in the form α = ( x1 , x2 , . . . , xk ). Then, α is a k-tuple; thus, the length of α is k (by
the definition of “length”). But α is a composition of n; in other words, ( x1 , x2 , . . . , xk ) is a
composition of n (since α = ( x1 , x2 , . . . , xk )). Hence, Proposition 7.48.1 yields k ≤ n. Thus, k
is an element of {0, 1, . . . , n} (since k ∈ N). In other words, the length of α is an element of
{0, 1, . . . , n} (since the length of α is k). In other words, (the length of α) ∈ {0, 1, . . . , n}. Qed.
Enumerative Combinatorics: class notes page 463

sum is n (by the definition of a “composition of n”). Hence,

{α is a composition of n | (the length of α) = k}

= {α is a tuple of positive integers whose sum is n | (the length of α) = k}
= {α is a tuple of positive integers whose sum is n | α is a k-tuple}
 
because a tuple α satisfies (the length of α) = k if and only if
it is a k-tuple
= {α is a k-tuple of positive integers whose sum is n}
= {( x1 , x2 , . . . , xk ) is a k-tuple of positive integers whose sum is n}
(here, we have renamed the index α as ( x1 , x2 , . . . , xk ))
= {( x1 , x2 , . . . , xk ) is a k-tuple of positive integers | x1 + x2 + · · · + xk = n}
n o
= ( x 1 , x 2 , . . . , x k ) ∈ Pk | x 1 + x 2 + · · · + x k = n
 
k
since a k-tuple of positive integers is the same as an element of P .

Therefore,

|{α is a composition of n | (the length of α) = k}|

n o
= ( x 1 , x 2 , . . . , x k ) ∈ Pk | x 1 + x 2 + · · · + x k = n
 
= # of ( x1 , x2 , . . . , xk ) ∈ Pk satisfying x1 + x2 + · · · + xk = n

 n − 1 , if n > 0;


= k−1 (by (240))

 [ k = 0] , if n = 0
n−1
 
= (since n > 0) . (335)
k−1

Forget that we fixed k. We thus have proved (335) for each k ∈ N.

Enumerative Combinatorics: class notes page 464

Now,

(the sum of the lengths of all compositions of n)

= ∑ (the length of α)
α is a composition of n

= ∑ ∑ (the length of α)
k ∈{0,1,...,n} α is a composition of n;
| {z }
(the length of α)=k =k
 
by (37), because if α is any composition of n,
then (the length of α) ∈ {0, 1, . . . , n}
= ∑ ∑ k
k∈{0,1,...,n} α is a composition of n;
(the length of α)=k
| {z }
=|{α is a composition of n | (the length of α)=k}|·k
= ∑ |{α is a composition of n | (the length of α) = k}| ·k
k∈{0,1,...,n}
|  {z  }
| {z } n−1
n =
=∑ k−1
k =0 (by (335))
n n
n−1 n−1
   
= ∑ k−1
·k = ∑ k
k−1
= ( n + 1 ) 2n −2 (by Lemma 7.52.1) .
k =0 k =0

This solves Exercise 2.10.9.

7.53. Solution to Exercise 2.11.1

Before we solve Exercise 2.11.1, let us recall the definition of the multisubset
{ a1 , a2 , . . . , ak }multi :

Definition 7.53.1. Let T be a set. Let k ∈ N. Let a1 , a2 , . . . , ak be any k elements

of T (not necessarily distinct). Then, { a1 , a2 , . . . , ak }multi is the multisubset of T
defined as the map

T → N,
t 7→ (# of i ∈ [k ] satisfying ai = t) .

Indeed, Definition 7.53.1 is simply a more explicit restatement of the last sentence
of Definition 2.11.1.289

289 Infact, in the last sentence of Definition 2.11.1, we said that a multisubset of T given as a map
f : T → N shall be denoted by { a1 , a2 , . . . , ak }multi if ( a1 , a2 , . . . , ak ) is a list of elements of T such
that each t ∈ T appears in this list exactly f (t) times. In other words, { a1 , a2 , . . . , ak }multi is the
Enumerative Combinatorics: class notes page 465

Solution to Exercise 2.11.1. Definition 7.53.1 shows that { a1 , a2 , . . . , ak }multi is the mul-
tisubset of T defined as the map
T → N,
t 7→ (# of i ∈ [k ] satisfying ai = t) .
Let us denote this map by f . Thus, { a1 , a2 , . . . , ak }multi = f . Hence,
(the size of { a1 , a2 , . . . , ak }multi ) = (the size of f )
=∑ f (t) (by Definition 2.11.2)
t∈ T
|{z}
=(# of i ∈[k] satisfying ai =t)
(by the definition of f )

= ∑ (# of i ∈ [k] satisfying ai = t) .
t∈ T

Comparing this with

k = |[k]| (since [k] = {1, 2, . . . , k} is a k-element set)
= ∑ (# of i ∈ [k] satisfying ai = t)
t∈ T
(by the sum rule (Theorem 1.2.5), since ai ∈ T for each i ∈ [k]) ,
we obtain (the size of { a1 , a2 , . . . , ak }multi ) = k. In other words, the multisubset
{ a1 , a2 , . . . , ak }multi of T has size k. This solves Exercise 2.11.1.

7.54. Solution to Exercise 2.11.2

Exercise 2.11.2 asks us to prove Proposition 2.11.5 and Proposition 2.11.6. We shall
prove Proposition 2.11.5 first. This proof will imitate the proof of Proposition 1.4.11
that is given in [Grinbe15, proof of Theorem 2.46]. First, however, we need to build
up some basics of the theory of multisubsets. Since these fundamental construc-
tions are hard to find in the mathematical literature, I will do them in detail; very
little (if anything) in them is non-obvious.
We recall that multisubsets of a set T have been defined as (certain) maps from
T to N. Even if we can think of them as analogues of sets, they are thus (formally
speaking) maps. Hence, we will tend to use letters like f , g, h (rather than S or A)
for them in this section.
multisubset of T defined as the map

T → N,
t 7→ (# of times t appears in the list ( a1 , a2 , . . . , ak )) .

But this is equivalent to the definition of { a1 , a2 , . . . , ak }multi in Definition 7.53.1, because each
t ∈ T satisfies

(# of times t appears in the list ( a1 , a2 , . . . , ak )) = (# of i ∈ [k] satisfying ai = t) .

Enumerative Combinatorics: class notes page 466

Definition 7.54.1. Let T be a set. Let f be any multisubset of T. Then, we let | f |

denote the size of f . Thus,

| f | = (the size of f ) = ∑ f (t) (336)

t∈ T

(by the definition of the size of a multisubset).

Let us next define a concept of inclusion for multisubsets of T:

Definition 7.54.2. Let T be a set. Let f and g be two multisubsets of T. Then, we

shall say that f ⊆ g if and only if each t ∈ T satisfies f (t) ≤ g (t).

Thus, we have defined a relation “⊆” between multisubsets of T. It is an ana-

logue of the relation “⊆” between sets, but it takes the multiplicities into account:
Two multisubsets f and g of T satisfy f ⊆ g if and only if each element of T appears
at most as often in f as it appears in g. For example, we have {1, 3, 3, 5, 6}multi ⊆
{1, 1, 3, 3, 3, 5, 6, 7}multi , but we don’t have {1, 3, 3, 5}multi ⊆ {1, 1, 3, 5, 5}multi .

Definition 7.54.3. Let T be a set. Let f and g be two multisubsets of T. Then, the
map

T → N,
t 7→ f (t) + g (t)

is a multisubset of T (by Proposition 7.54.4 (a) below). This multisubset of T will

be denoted by f ∪ g.

The multisubset f ∪ g is an analogue of the union of two sets, but it takes the
multiplicities into account: If an element of T appears a times in f and appears b
times in g, then it appears a + b times in f ∪ g. For example, {1, 3, 4, 4, 5}multi ∪
{2, 3, 4}multi = {1, 2, 3, 3, 4, 4, 4, 5}multi .
Let us prove a few properties of this “multisubset union” f ∪ g:

Proposition 7.54.4. Let T be a set. Let f and g be two multisubsets of T. Then:

(a) The map

T → N,
t 7→ f (t) + g (t)

is a multisubset of T.
(b) We have | f ∪ g| = | f | + | g|.
(c) We have f ⊆ f ∪ g and g ⊆ f ∪ g.
(d) We have f ∪ g = g ∪ f .
Enumerative Combinatorics: class notes page 467

Proof of Proposition 7.54.4. We know that f and g are multisubsets of T; thus, f and
g are maps from T to N.
We know that f is a multisubset of T. According to Definition 2.11.1, this means
that only finitely many t ∈ T satisfy f (t) 6= 0. Hence, we know that only finitely
many t ∈ T satisfy f (t) 6= 0. In other words, the set {t ∈ T | f (t) 6= 0} is finite.
Similarly, the set {t ∈ T | g (t) 6= 0} is finite.
(a) We know that f and g are maps from T to N. Hence, each t ∈ T satisfies
f (t) ∈ N and g (t) ∈ N and therefore f (t) + g (t) ∈ N. Thus, the map

T → N,
t 7→ f (t) + g (t)

is well-defined. Moreover, it is easy to see that this map is a multisubset of T 290 .

This proves Proposition 7.54.4 (a).

290 Proof. We need to show that the map

T → N,
t 7→ f (t) + g (t)

is a multisubset of T. According to Definition 2.11.1, this means showing that only finitely many
t ∈ T satisfy f (t) + g (t) 6= 0. Thus, we need to show that only finitely many t ∈ T satisfy
f (t) + g (t) 6= 0. In other words, we need to show that the set {t ∈ T | f (t) + g (t) 6= 0} is
finite.
Let s ∈ {t ∈ T | f (t) + g (t) 6= 0}. We shall show that s ∈ {t ∈ T | f (t) 6= 0} ∪
{ t ∈ T | g ( t ) 6 = 0}.
Indeed, assume the contrary. Hence, s ∈ / {t ∈ T | f (t) 6= 0} ∪ {t ∈ T | g (t) 6= 0}. In other
words, we have both s ∈ / {t ∈ T | f (t) 6= 0} and s ∈
/ { t ∈ T | g ( t ) 6 = 0}.
But s ∈ {t ∈ T | f (t) + g (t) 6= 0}. Thus, s is a t ∈ T satisfying f (t) + g (t) 6= 0. In other
words, s ∈ T and f (s) + g (s) 6= 0.
If we had f (s) 6= 0, then s would be an element t ∈ T satisfying f (t) 6= 0, and therefore
we would have s ∈ {t ∈ T | f (t) 6= 0}; but this would contradict s ∈ / { t ∈ T | f ( t ) 6 = 0}.
Hence, we cannot have f (s) 6= 0. Thus, we must have f (s) = 0. Similarly, g (s) = 0. Hence,
f (s) + g (s) = 0, which contradicts f (s) + g (s) 6= 0.
|{z} |{z}
=0 =0
This contradiction shows that our assumption was false. Hence, s ∈ {t ∈ T | f (t) 6= 0} ∪
{t ∈ T | g (t) 6= 0} is proved.
Now, forget that we fixed s. We thus have proved that s ∈ {t ∈ T | f (t) 6= 0} ∪
{t ∈ T | g (t) 6= 0} for each s ∈ {t ∈ T | f (t) + g (t) 6= 0}. In other words,

{t ∈ T | f (t) + g (t) 6= 0} is a subset of {t ∈ T | f (t) 6= 0} ∪ {t ∈ T | g (t) 6= 0} .

But the set {t ∈ T | f (t) 6= 0} ∪ {t ∈ T | g (t) 6= 0} is finite (since it is the union of

the two finite sets {t ∈ T | f (t) 6= 0} and {t ∈ T | g (t) 6= 0}). Hence, its subset
{t ∈ T | f (t) + g (t) 6= 0} must be finite as well (since a subset of a finite set is always finite).
In other words, only finitely many t ∈ T satisfy f (t) + g (t) 6= 0. In other words, the map

T → N,
t 7→ f (t) + g (t)

is a multisubset of T. Qed.
Enumerative Combinatorics: class notes page 468

(b) The equality (336) yields | f | = ∑ f (t). Also, applying (336) to g instead of
t∈ T
f , we find | g| = ∑ g (t). Adding these two equalities together, we obtain
t∈ T

| f | + | g| = ∑ f (t) + ∑ g (t) = ∑ ( f (t) + g (t)) .

t∈ T t∈ T t∈ T

Comparing this with

| f ∪ g| = ∑ ( f ∪ g) (t) (by (336), applied to f ∪ g instead of f )

t∈ T
| {z }
= f (t)+ g(t)
(by the definition of f ∪ g)

= ∑ ( f (t) + g (t)) ,
t∈ T

we obtain | f ∪ g| = | f | + | g|. This proves Proposition 7.54.4 (b).

(c) Each t ∈ T satisfies
( f ∪ g) (t) = f (t) + g (t) (by the definition of f ∪ g)
|{z}
≥0
(since g(t)∈N
(because g is a map from T to N))
≥ f (t) .
In other words, each t ∈ T satisfies f (t) ≤ ( f ∪ g) (t). In other words, f ⊆ f ∪ g
(according to Definition 7.54.2). An analogous argument shows that g ⊆ f ∪ g.
Hence, Proposition 7.54.4 (c) is proved.
(d) For each t ∈ T, we have
( f ∪ g) (t) = f (t) + g (t) (337)
(by the definition of f ∪ g) and
( g ∪ f ) (t) = g (t) + f (t) (338)
(by the definition of g ∪ f ). Hence, for each t ∈ T, we have
( f ∪ g) (t) = f (t) + g (t) (by (337))
= g (t) + f (t) = ( g ∪ f ) (t) (by (338)) .
In other words, f ∪ g = g ∪ f (since both f ∪ g and g ∪ f are maps from T to N).
This proves Proposition 7.54.4 (d).

Proposition 7.54.5. Let T be a set. Let k ∈ N and ` ∈ N. Let a1 , a2 , . . . , ak be any

k elements of T (not necessarily distinct). Let b1 , b2 , . . . , b` be any ` elements of T
(not necessarily distinct). Then,

{ a1 , a2 , . . . , ak }multi ∪ {b1 , b2 , . . . , b` }multi = { a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi .

Enumerative Combinatorics: class notes page 469

Proof of Proposition 7.54.5. Definition 7.53.1 says that { a1 , a2 , . . . , ak }multi is the mul-
tisubset of T defined as the map

T → N,
t 7→ (# of i ∈ [k ] satisfying ai = t) .

Hence, for each t ∈ T, we have291

{ a1 , a2 , . . . , ak }multi (t)
= (# of i ∈ [k] satisfying ai = t)
= (# of times t appears in the k-tuple ( a1 , a2 , . . . , ak )) . (339)

Likewise, for each t ∈ T, we have

{b1 , b2 , . . . , b` }multi (t)

= (# of times t appears in the `-tuple (b1 , b2 , . . . , b` )) . (340)

Likewise, for each t ∈ T, we have

{ a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi (t)
= (# of times t appears in the (k + `) -tuple ( a1 , a2 , . . . , ak , b1 , b2 , . . . , b` )) .

Hence, for each t ∈ T, we have

{ a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi (t)
= (# of times t appears in the (k + `) -tuple ( a1 , a2 , . . . , ak , b1 , b2 , . . . , b` ))
= (# of times t appears in the k-tuple ( a1 , a2 , . . . , ak ))
| {z }
={ a1 ,a2 ,...,ak }multi (t)
(by (339))
+ (# of times t appears in the `-tuple (b1 , b2 , . . . , b` ))
| {z }
={b1 ,b2 ,...,b` }multi (t)
(by (340))
= { a1 , a2 , . . . , ak }multi (t) + {b1 , b2 , . . . , b` }multi (t) . (341)

On the other hand, for each t ∈ T, we have

({ a1 , a2 , . . . , ak }multi ∪ {b1 , b2 , . . . , b` }multi ) (t)

= { a1 , a2 , . . . , ak }multi (t) + {b1 , b2 , . . . , b` }multi (t)
(by the definition of the map { a1 , a2 , . . . , ak }multi ∪ {b1 , b2 , . . . , b` }multi )
= { a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi (t)
291 Recall that any multisubset of T is a map from T to N. Hence, for example, the multisubset
{ a1 , a2 , . . . , ak }multi of T is a map from T to N, and thus can be evaluated at t ∈ T. Thus,
expressions like “{ a1 , a2 , . . . , ak }multi (t)” make sense.
Enumerative Combinatorics: class notes page 470

(by (341)). In other words,

{ a1 , a2 , . . . , ak }multi ∪ {b1 , b2 , . . . , b` }multi = { a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi
(because both { a1 , a2 , . . . , ak }multi ∪ {b1 , b2 , . . . , b` }multi and { a1 , a2 , . . . , ak , b1 , b2 , . . . , b` }multi
are maps from T to N). This proves Proposition 7.54.5.
Having defined the “union” f ∪ g of two multisubsets f and g, let us next define
their “difference” g \ f . This is an analogue of the set difference of two sets, but
we shall only define it in the simplest case – namely, when f ⊆ g. (This is the only
case in which we will need it.)
Definition 7.54.6. Let T be a set. Let f and g be two multisubsets of T such that
f ⊆ g. Then, the map

T → N,
t 7→ g (t) − f (t)

is a multisubset of T (by Proposition 7.54.7 (a) further below). This multisubset

of T is denoted by g \ f .

Proposition 7.54.7. Let T be a set. Let f and g be two multisubsets of T such that
f ⊆ g.
(a) The map

T → N,
t 7→ g (t) − f (t)

is a multisubset of T.
(b) We have ( g \ f ) ∪ f = g.
(c) We have g \ f ⊆ g.
Proof of Proposition 7.54.7. (a) We have f ⊆ g. In other words, each t ∈ T satisfies
f (t) ≤ g (t) (by Definition 7.54.2). Hence, each t ∈ T satisfies g (t) − f (t) ∈ N
(since the previous sentence says that f (t) ≤ g (t), hence g (t) ≥ f (t) and therefore
g (t) − f (t) ∈ N 292 ). Thus, the map
T → N,
t 7→ g (t) − f (t)
is well-defined. Moreover, it is easy to see that this map is a multisubset of T 293 .

292 sinceg (t) − f (t) is an integer

293 The proof of this fact is analogous to our proof (during the proof of Proposition 7.54.4 (a) above)
of the fact that
T → N,
t 7→ f (t) + g (t)
is a multisubset of T.
Enumerative Combinatorics: class notes page 471

This proves Proposition 7.54.7 (a).

(b) For each t ∈ T, we have

( g \ f ) (t) = g (t) − f (t) (342)

(by the definition of g \ f ). Hence, for each t ∈ T, we have

(( g \ f ) ∪ f ) (t) = ( g \ f ) (t) + f (t) (by the definition of ( g \ f ) ∪ f )

= g (t) (by (342)) .

In other words, ( g \ f ) ∪ f = g (since both ( g \ f ) ∪ f and g are maps from T to N).

This proves Proposition 7.54.7 (b).
(c) Proposition 7.54.4 (c) (applied to g \ f and f instead of f and g) yields g \ f ⊆
( g \ f ) ∪ f and f ⊆ ( g \ f ) ∪ f . Thus, in particular, g \ f ⊆ ( g \ f ) ∪ f = g (by
Proposition 7.54.7 (b)). This proves Proposition 7.54.7 (c).
Proposition 7.54.7 (b) is a multiset analogue of the standard fact that if A and B
are two sets satisfying A ⊆ B, then ( B \ A) ∪ A = B. A similar fact is that if A and
B are two disjoint sets, then ( B ∪ A) \ A = B. This, too, has a multiset analogue
(but without any disjointness requirement):

Proposition 7.54.8. Let T be a set. Let f and g be two multisubsets of T. Then,

( g ∪ f ) \ f = g.

Proof of Proposition 7.54.8. Proposition 7.54.4 (c) (applied to g and f instead of f and
g) yields g ⊆ g ∪ f and f ⊆ g ∪ f . Hence, ( g ∪ f ) \ f is well-defined. Moreover, for
each t ∈ T, we have
( g ∪ f ) (t) = g (t) + f (t) (343)
(by the definition of g ∪ f ). Hence, for each t ∈ T, we have

(( g ∪ f ) \ f ) (t) = ( g ∪ f ) (t) − f (t) (by the definition of ( g ∪ f ) \ f )

= g (t) (by (343)) .

In other words, ( g ∪ f ) \ f = g (since both ( g ∪ f ) \ f and g are maps from T to N).

This proves Proposition 7.54.8.
Next, we define the support of a multisubset f of T:

Definition 7.54.9. Let T be a set. Let f be a multisubset of T. Then, we define

Supp f to be the set {t ∈ T | f (t) 6= 0}. We call this set Supp f the support of f .

Thus, the support of a multisubset f of T is the set of all t ∈ T that occur

in f (with a positive multiplicity). Thus, intuitively speaking, the support of a
multisubset f is obtained by turning f into a set (by forgetting the multiplicities).
From this viewpoint, the following fact should be evident:
Enumerative Combinatorics: class notes page 472

Proposition 7.54.10. Let T be a set. Let k ∈ N. Let a1 , a2 , . . . , ak be any k elements

of T (not necessarily distinct). Then,

Supp ({ a1 , a2 , . . . , ak }multi ) = { a1 , a2 , . . . , ak } .

Proof of Proposition 7.54.10. The definition of Supp ({ a1 , a2 , . . . , ak }multi ) yields

Supp ({ a1 , a2 , . . . , ak }multi )
= {t ∈ T | { a1 , a2 , . . . , ak }multi (t) 6= 0} . (344)
Now, let u ∈ Supp ({ a1 , a2 , . . . , ak }multi ). We shall show that u ∈ { a1 , a2 , . . . , ak }.
We have u ∈ Supp ({ a1 , a2 , . . . , ak }multi ) = {t ∈ T | { a1 , a2 , . . . , ak }multi (t) 6= 0}
(by (344)). In other words, u is a t ∈ T satisfying { a1 , a2 , . . . , ak }multi (t) 6= 0. In
other words, u ∈ T and { a1 , a2 , . . . , ak }multi (u) 6= 0.
But recall how the map { a1 , a2 , . . . , ak }multi was defined in Definition 7.53.1. This
definition shows that
{ a1 , a2 , . . . , ak }multi (u) = (# of i ∈ [k] satisfying ai = u) .
Hence, (# of i ∈ [k ] satisfying ai = u) = { a1 , a2 , . . . , ak }multi (u) 6= 0. In other words,
there exists at least one i ∈ [k ] satisfying ai = u. Consider this i. Hence, u = ai ∈
{ a1 , a2 , . . . , a k }.
Forget that we fixed u. We thus have proved that u ∈ { a1 , a2 , . . . , ak } for each
u ∈ Supp ({ a1 , a2 , . . . , ak }multi ). In other words,
Supp ({ a1 , a2 , . . . , ak }multi ) ⊆ { a1 , a2 , . . . , ak } . (345)
Next, let v ∈ { a1 , a2 , . . . , ak }. We shall show that v ∈ Supp ({ a1 , a2 , . . . , ak }multi ).
We have v ∈ { a1 , a2 , . . . , ak }. In other words, v = a j for some j ∈ [k]. Consider
this j. We have v = a j ∈ T (since a1 , a2 , . . . , ak are elements of T). Also, v = a j , so
that a j = v. Hence, there exists at least one i ∈ [k] satisfying ai = v (namely, i = j).
Recall how the map { a1 , a2 , . . . , ak }multi was defined in Definition 7.53.1. This
definition shows that
{ a1 , a2 , . . . , ak }multi (v) = (# of i ∈ [k] satisfying ai = v) ≥ 1
(since there exists at least one i ∈ [k ] satisfying ai = v). Thus, { a1 , a2 , . . . , ak }multi (v) ≥
1 > 0, so that { a1 , a2 , . . . , ak }multi (v) 6= 0.
Now, we know that v ∈ T and { a1 , a2 , . . . , ak }multi (v) 6= 0. In other words, v is a
t ∈ T satisfying { a1 , a2 , . . . , ak }multi (t) 6= 0. In other words,
v ∈ {t ∈ T | { a1 , a2 , . . . , ak }multi (t) 6= 0}. In view of (344), this rewrites as v ∈
Supp ({ a1 , a2 , . . . , ak }multi ).
Forget that we fixed v. We thus have proved that v ∈ Supp ({ a1 , a2 , . . . , ak }multi )
for each v ∈ { a1 , a2 , . . . , ak }. In other words,
{ a1 , a2 , . . . , ak } ⊆ Supp ({ a1 , a2 , . . . , ak }multi ) .
Combining this with (345), we obtain Supp ({ a1 , a2 , . . . , ak }multi ) = { a1 , a2 , . . . , ak }.
This proves Proposition 7.54.10.
Enumerative Combinatorics: class notes page 473

Here are three more simple properties of supports:

Proposition 7.54.11. Let T be a set. Let f be a multisubset of T. Let u ∈ Supp f .

Then:
(a) We have {u}multi ⊆ f .
(b) We have | f \ {u}multi | = | f | − 1.

Proof of Proposition 7.54.11. We know that f is a multisubset of T, thus a map from

T to N.
(a) The definition of Supp f yields Supp f = {t ∈ T | f (t) 6= 0} = {s ∈ T | f (s) 6= 0}
(here, we have renamed the index t as s).
Definition 7.53.1 (applied to 1 and the 1-tuple (u) instead of k and the k-tuple
( a1 , a2 , . . . , ak )) tells us that {u}multi is defined as the map

T → N,
t 7→ (# of i ∈ [1] satisfying u = t) .

Hence,
{u}multi (t) = (# of i ∈ [1] satisfying u = t) (346)
for each t ∈ T.
Let t ∈ T. We shall show that {u}multi (t) ≤ f (t).
Indeed, assume the contrary. Thus, {u}multi (t) > f (t). But f (t) ∈ N (since
f is a map from T to N), hence f (t) ≥ 0. Thus, {u}multi (t) > f (t) ≥ 0. In
view of (346), this rewrites as (# of i ∈ [1] satisfying u = t) > 0. In other words,
there exists at least one i ∈ [1] satisfying u = t. Thus, of course, we have u = t
(since the statement “u = t” does not depend on i). Hence, t = u ∈ Supp f =
{s ∈ T | f (s) 6= 0}. In other words, t is an s ∈ T such that f (s) 6= 0. In other
words, t ∈ T and f (t) 6= 0. From f (t) 6= 0, we obtain f (t) ≥ 1 (since f (t) ∈ N).
Hence, {u}multi (t) > f (t) ≥ 1.
But (346) becomes

{u}multi (t) = (# of i ∈ [1] satisfying u = t)

= (# of i ∈ [1]) (since we know that u = t)
= |[1]| = 1.

This contradicts {u}multi (t) > 1. This contradiction shows that our assumption
was wrong. Hence, {u}multi (t) ≤ f (t) is proven.
Forget that we fixed t. We thus have showed that each t ∈ T satisfies {u}multi (t) ≤
f (t). This means that {u}multi ⊆ f (according to Definition 7.54.2). Thus, we have
shown that {u}multi ⊆ f . This proves Proposition 7.54.11 (a).
(b) Exercise 2.11.1 (applied to 1 and the 1-tuple (u) instead of k and the k-tuple
( a1 , a2 , . . . , ak )) yields that the multisubset {u}multi of T has size 1. In other words,
|{u}multi | = 1 (since the size of {u}multi is denoted by |{u}multi |).
Enumerative Combinatorics: class notes page 474

We have {u}multi ⊆ f (by Proposition 7.54.11 (a)). Thus, Proposition 7.54.7 (b)
(applied to {u}multi and f instead of f and g) yields ( f \ {u}multi ) ∪ {u}multi = f .
Hence, f = ( f \ {u}multi ) ∪ {u}multi . Thus,

| f | = |( f \ {u}multi ) ∪ {u}multi | = | f \ {u}multi | + |{u}multi |

| {z }
=1
 
by Proposition 7.54.4 (b),
applied to f \ {u}multi and {u}multi instead of f and g
= | f \ {u}multi | + 1.

In other words, | f \ {u}multi | = | f | − 1. This proves Proposition 7.54.11 (b).

Proposition 7.54.12. Let T be a set. Let f be a multisubset of T.

(a) The set Supp f is finite.
(b) We have | f | = ∑ f (t).
t∈Supp f
(c) Assume that | f | > 0. Then, Supp f is a finite nonempty subset of T.

Proof of Proposition 7.54.12. We know that f is a multisubset of T. According to

Definition 2.11.1, this means that only finitely many t ∈ T satisfy f (t) 6= 0. Hence,
we know that only finitely many t ∈ T satisfy f (t) 6= 0. In other words, the set
{t ∈ T | f (t) 6= 0} is finite. Since Supp f = {t ∈ T | f (t) 6= 0} (by the definition
of Supp f ), we can rewrite this as follows: The set Supp f is finite. This proves
Proposition 7.54.12 (a).
(b) From (336), we obtain

|f| = ∑ f (t) = ∑ f (t) + ∑ f (t)

t∈ T t∈ T; t∈ T;
|{z}
f (t)=0 =0 f (t)6=0
(since each t ∈ T satisfies either f (t) = 0 or f (t) 6= 0 (but not both))
= ∑ 0+ ∑ f (t) = ∑ f (t) .
t∈ T; t∈ T; t∈ T;
f (t)=0 f (t)6=0 f (t)6=0
| {z }
=0

Comparing this with

∑ f (t) = ∑ f (t) ,
t∈Supp f t∈ T;
| {z } f (t)6=0
= ∑
t∈ T;
f (t)6=0
(since Supp f ={t∈ T | f (t)6=0})

we obtain | f | = ∑ f (t). Thus, Proposition 7.54.12 (b) is proved.

t∈Supp f
Enumerative Combinatorics: class notes page 475

(c) Proposition 7.54.12 (b) yields | f | = ∑ f (t), thus ∑ f (t) = | f | > 0.

t∈Supp f t∈Supp f
If we had Supp f = ∅, then we would have

∑ f (t) = ∑ f (t) = (empty sum) = 0,

t∈Supp f t ∈∅

which would contradict ∑ f (t) > 0. Thus, we cannot have Supp f = ∅.

t∈Supp f
Hence, Supp f 6= ∅. In other words, Supp f is nonempty. Moreover, Supp f is finite
(by Proposition 7.54.12 (a)). Finally, Supp f = {t ∈ T | f (t) 6= 0} ⊆ T. Hence,
Supp f is a subset of T. Altogether, this proves Proposition 7.54.12 (c).

Proposition 7.54.13. Let T be a set. Let f and g be two multisubsets of T such

that f ⊆ g. Then, Supp f ⊆ Supp g.

Proof of Proposition 7.54.13. Let u ∈ Supp f . We shall show that u ∈ Supp g.

The definition of Supp f yields Supp f = {t ∈ T | f (t) 6= 0}. The definition of
Supp g yields Supp g = {t ∈ T | g (t) 6= 0}.
Now, u ∈ Supp f = {t ∈ T | f (t) 6= 0}; in other words, u is a t ∈ T satisfying
f (t) 6= 0. In other words, u ∈ T and f (u) 6= 0. But f is a multisubset of T, thus a
map from T to N. Hence, f (u) ∈ N. Thus, from f (u) 6= 0, we obtain f (u) > 0.
But f ⊆ g. In other words, each t ∈ T satisfies f (t) ≤ g (t) (by Definition
7.54.2). Applying this to t = u, we obtain f (u) ≤ g (u). Hence, g (u) ≥ f (u) > 0,
so that g (u) 6= 0. Hence, u is a t ∈ T satisfying g (t) 6= 0 (since u ∈ T). In
other words, u ∈ {t ∈ T | g (t) 6= 0}. This rewrites as u ∈ Supp g (since Supp g =
{t ∈ T | g (t) 6= 0}).
Forget that we fixed u. We thus have proved that u ∈ Supp g for each u ∈ Supp f .
In other words, Supp f ⊆ Supp g. This proves Proposition 7.54.13.
Next, let us introduce a word for the kind of tuple whose existence (and unique-
ness) is claimed in Proposition 2.11.5:

Definition 7.54.14. Let T be a set of integers. Let f be a multisubset of T. An

increasing list of f shall mean a tuple (s1 , s2 , . . . , sk ) of elements of T such that
f = {s1 , s2 , . . . , sk }multi and s1 ≤ s2 ≤ · · · ≤ sk .

Our goal is thus to prove the following:

Proposition 7.54.15. Let T be a set of integers. Let f be a multisubset of T. Then,

f has exactly one increasing list.

We shall prove this by induction on the size of this multisubset; the induction
base will rely on the following simple fact (an analogue of [Grinbe15, Proposition
2.48]):
Enumerative Combinatorics: class notes page 476

Proposition 7.54.16. Let T be a set of integers. Let f be a multisubset of T such

that | f | = 0. Then:
(a) We have f = {}multi .
(b) The multisubset f has exactly one increasing list: namely, the empty list ().

Of course, the multisubset {}multi is a multiset analogue of the empty set ∅.

Proof of Proposition 7.54.16. This is an essentially trivial argument that takes long to
explain due to the notational awkwardness of working with the empty multisubset
and the empty list. So here we go:
From (336), we obtain | f | = ∑ f (t). Hence, ∑ f (t) = | f | = 0.
t∈ T t∈ T
But f is a multisubset of T, thus a map from T to N. Hence, each t ∈ T satisfies
f (t) ∈ N. Hence, all addends of the sum ∑ f (t) are nonnegative integers. Thus,
t∈ T
∑ f (t) is a sum of nonnegative integers. But this sum equals 0 (since ∑ f (t) = 0).
t∈ T t∈ T
If a sum of nonnegative integers equals 0, then all its addends must be 0 (since
otherwise, the sum would be positive rather than 0). We can apply this to the sum
∑ f (t) (because ∑ f (t) is a sum of nonnegative integers and equals 0). Thus, we
t∈ T t∈ T
conclude that all addends of this sum must be 0. In other words,

f (t) = 0 for each t ∈ T. (347)

Let us denote the 0-tuple () by ( a1 , a2 , . . . , a0 ). Thus, ( a1 , a2 , . . . , a0 ) = (), so that

{ a1 , a2 , . . . , a0 }multi = {}multi .
Definition 7.53.1 (applied to 0 and the 0-tuple ( a1 , a2 , . . . , a0 ) instead of k and the
k-tuple ( a1 , a2 , . . . , ak )) tells us that { a1 , a2 , . . . , a0 }multi is defined as the map

T → N,
t 7→ (# of i ∈ [0] satisfying ai = t) .
294 Hence, for each t ∈ T, we have

{ a1 , a2 , . . . , a0 }multi (t) = (# of i ∈ [0] satisfying ai = t) = 0 (348)

(since there exist no i ∈ [0] satisfying ai = t (because [0] is an empty set)).

Now, for each t ∈ T, we have

f (t) = 0 (by (347))

= { a1 , a2 , . . . , a0 }multi (t) (by (348))
| {z }
={}multi
= {}multi (t) .

294 Being able to write down this map is the reason why we denoted the 0-tuple () by ( a1 , a2 , . . . , a0 ).
Enumerative Combinatorics: class notes page 477

In other words, f = {}multi (since both f and {}multi are maps from T to N). This
proves Proposition 7.54.16 (a).
(b) The empty list () satisfies f = {}multi (as we have just proved). Thus, the
empty list () is a tuple (s1 , s2 , . . . , sk ) of elements of T such that f = {s1 , s2 , . . . , sk }multi
and s1 ≤ s2 ≤ · · · ≤ sk (indeed, the chain of inequalities s1 ≤ s2 ≤ · · · ≤ sk is vac-
uously true for the empty list (), because it contains no inequality signs). In other
words, the empty list () is an increasing list of f (by the definition of an increasing
list). It remains to show that it is the only increasing list of f .
Let (s1 , s2 , . . . , sk ) be any increasing list of f . Thus, (s1 , s2 , . . . , sk ) is a tuple of
elements of T such that f = {s1 , s2 , . . . , sk }multi and s1 ≤ s2 ≤ · · · ≤ sk (by the
definition of an increasing list). But Exercise 2.11.1 (applied to ai = si ) yields that
the multisubset {s1 , s2 , . . . , sk }multi of T has size k. In other words, the multisubset f
of T has size k (since f = {s1 , s2 , . . . , sk }multi ). In other words, | f | = k (since the size
of f is denoted by | f |). Thus, k = | f | = 0. Hence, (s1 , s2 , . . . , sk ) = (s1 , s2 , . . . , s0 ) =
().
Now, forget that we fixed (s1 , s2 , . . . , sk ). We thus have shown that if (s1 , s2 , . . . , sk )
is any increasing list of f , then (s1 , s2 , . . . , sk ) = (). In other words, any increasing
list of f is (). Therefore, the multisubset f has exactly one increasing list: namely,
the empty list () (since we already know that () is an increasing list of f ). This
proves Proposition 7.54.16 (b).
The induction step of our proof of Proposition 7.54.15 will rely on the following
fact (an analogue of [Grinbe15, Proposition 2.49]):

Proposition 7.54.17. Let T be a set of integers. Let f be a multisubset of T such

that | f | > 0. Let m be the largest element of Supp f . Let (s1 , s2 , . . . , sk ) be an
increasing list of f . Then:
(a) We have k ≥ 1 and sk = m.
(b) We have {m}multi ⊆ f .
(c) The list (s1 , s2 , . . . , sk−1 ) is an increasing list of the multisubset f \ {m}multi
of T.

Proof of Proposition 7.54.17. We know that (s1 , s2 , . . . , sk ) is an increasing list of f . In

other words, (s1 , s2 , . . . , sk ) is a list of elements of T such that f = {s1 , s2 , . . . , sk }multi
and s1 ≤ s2 ≤ · · · ≤ sk (by the definition of an “increasing list”).
Exercise 2.11.1 (applied to ai = si ) yields that the multisubset {s1 , s2 , . . . , sk }multi
of T has size k. In other words, the multisubset f of T has size k (since f =
{s1 , s2 , . . . , sk }multi ). In other words, | f | = k (since the size of f is denoted by | f |).
Hence, k = | f | > 0. Thus, k ≥ 1 (since k is an integer). Therefore, sk is well-defined.
Clearly, k ∈ [k ] (since k ≥ 1).
Proposition 7.54.10 (applied to ai = si ) yields
Enumerative Combinatorics: class notes page 478

Supp ({s1 , s2 , . . . , sk }multi ) = {s1 , s2 , . . . , sk }. Thus,

Supp f = Supp ({s1 , s2 , . . . , sk }multi )
|{z}
={s1 ,s2 ,...,sk }multi
= { s1 , s2 , . . . , s k } . (349)
We have s1 ≤ s2 ≤ · · · ≤ sk . Hence, the largest element of {s1 , s2 , . . . , sk } is sk . In
view of (349), this rewrites as follows: The largest element of Supp f is sk . In other
words, m is sk (since m is the largest element of Supp f ). In other words, m = sk .
Hence, sk = m. This completes the proof of Proposition 7.54.17 (a).
(b) We have m = sk ∈ {s1 , s2 , . . . , sk } = Supp f (by (349)). Hence, Proposition
7.54.11 (a) (applied to u = m) yields {m}multi ⊆ f . This proves Proposition 7.54.17
(b).
(c) From s1 ≤ s2 ≤ · · · ≤ sk , we obtain s1 ≤ s2 ≤ · · · ≤ sk−1 .
Furthermore, k − 1 ∈ N (since k ≥ 1). Hence, Proposition 7.54.5 (applied to k − 1
and 1 and the (k − 1)-tuple (s1 , s2 , . . . , sk−1 ) and the 1-tuple (sk ) instead of k and `
and the k-tuple ( a1 , a2 , . . . , ak ) and the `-tuple (b1 , b2 , . . . , b` )) yields
{s1 , s2 , . . . , sk−1 }multi ∪ {sk }multi = {s1 , s2 , . . . , sk−1 , sk }multi = {s1 , s2 , . . . , sk }multi .
Now, Proposition 7.54.8 (applied to {s1 , s2 , . . . , sk−1 }multi and {sk }multi instead of
g and f ) yields
({s1 , s2 , . . . , sk−1 }multi ∪ {sk }multi ) \ {sk }multi = {s1 , s2 , . . . , sk−1 }multi .
Hence,
 
 
{s1 , s2 , . . . , sk−1 }multi = ({s1 , s2 , . . . , sk−1 }multi ∪ {sk }multi ) \ sk
| {z } |{z} 
={s1 ,s2 ,...,sk }multi = f =m multi
(since f ={s1 ,s2 ,...,sk }multi )

= f \ {m}multi .
Thus, f \ {m}multi = {s1 , s2 , . . . , sk−1 }multi .
Now, we know that (s1 , s2 , . . . , sk−1 ) is a list of elements of T such that f \
{m}multi = {s1 , s2 , . . . , sk−1 }multi and s1 ≤ s2 ≤ · · · ≤ sk−1 . In other words,
(s1 , s2 , . . . , sk−1 ) is an increasing list of f \ {m}multi (by the definition of an “in-
creasing list”). This proves Proposition 7.54.17 (c).
We are now ready to prove Proposition 7.54.15:
Proof of Proposition 7.54.15. We shall prove Proposition 7.54.15 by induction on | f |:
Induction base: Proposition 7.54.15 holds under the condition that | f | = 0 295 .
This completes the induction base.
295 Proof.
Let f be as in Proposition 7.54.15, and assume that | f | = 0. We must show that the claim
of Proposition 7.54.15 holds.
Indeed, Proposition 7.54.16 (b) shows that the multisubset f has exactly one increasing list.
Thus, the claim of Proposition 7.54.15 holds. This completes our proof.
Enumerative Combinatorics: class notes page 479

Induction step: Let g ∈ N. Assume that Proposition 7.54.15 holds under the
condition that | f | = g. We shall now show that Proposition 7.54.15 holds under the
condition that | f | = g + 1.
We have assumed that Proposition 7.54.15 holds under the condition that | f | = g.
In other words,
 
if f is a multisubset of T satisfying | f | = g,
. (350)
then f has exactly one increasing list

Now, let f be a multisubset of T satisfying | f | = g + 1. We want to prove that f

has exactly one increasing list.
We have | f | = g + 1 > g ≥ 0. Hence, Proposition 7.54.12 (c) shows that Supp f
is a finite nonempty subset of T. Hence, Supp f is a finite nonempty set of integers
(since T is a set of integers). Thus, Supp f has a largest element (since any finite
nonempty set of integers has a largest element). Let m be this largest element.
Thus, m is the largest element of Supp f . Hence, in particular, m is an element
of Supp f , so that we have m ∈ Supp f . Hence, Proposition 7.54.11 (a) (applied
to u = m) yields {m}multi ⊆ f . Also, Proposition 7.54.11 (b) (applied to u = m)
yields | f \ {m}multi | = | f | − 1 = g (since | f | = g + 1). Hence, (350) (applied to
f \ {m}multi instead
of f ) shows that f \ {m}multi has exactly one increasing list.

Let t1 , t2 , . . . , t j be this list. We extend this list to a ( j + 1)-tuple t1 , t2 , . . . , t j+1
by setting t j+1 = m. Thus, t j+1 = m ∈ Supp f ⊆ T (since Supp f is a subset of T).
Recall that {m}multi ⊆ f . Hence, Proposition 7.54.7 (c) (applied to {m}multi and f
instead of f and g) yields f \ {m}multi ⊆ f . Therefore, Proposition 7.54.13 (applied
to f \ {m}multi and f instead of
f and g) yields Supp ( f \ {m}multi ) ⊆ Supp f .
We have defined t1 , t2 , . . . , t j as an increasing list of the multisubset f \ {m}multi .


In other words, t1 , t2 , . . . , t j is a tuple of elements of T such that f \ {m}multi =

t1 , t2 , . . . , t j multi and t1 ≤ t2 ≤ · · · ≤ t j (by the definition of an “increasing list”).


Thus, t1 , t2 , . . . , t j are elements of T (since t1 , t2 , . . . , t j is a tuple of elements of T).


Proposition 7.54.10 (applied to j and the j-tuple t1 , t2 , . . . , t j instead of k and
the k-tuple ( a1 , a2 , . . . , ak )) yields
  
Supp t1 , t2 , . . . , t j multi = t1 , t2 , . . . , t j .

In view of f \ {m}multi = t1 , t2 , . . . , t j , this rewrites as
multi

Supp ( f \ {m}multi ) = t1 , t2 , . . . , t j .

Hence, 
t1 , t2 , . . . , t j = Supp ( f \ {m}multi ) ⊆ Supp f .
We claim that
t 1 ≤ t 2 ≤ · · · ≤ t j +1 . (351)
[Proof of (351): If j + 1 ≤ 1, then the chain of inequalities (351) is vacuously true
(since it contains no inequality signs). Thus, for the rest of this proof of (351), we
Enumerative Combinatorics: class notes page 480

WLOG assume that we don’t have j + 1 ≤ 1. Hence, j + 1 > 1, so that j > 0 and
thus j ≥ 1 (since j is an integer). Hence, t j is well-defined. We have j ≥ 1 and thus

j ∈ [ j]. Hence, t j ∈ t1 , t2 , . . . , t j ⊆ Supp f .
But m is the largest element of Supp f . Hence, m ≥ x for each x ∈ Supp f .
Applying this to x = t j , we obtain m ≥ t j (since t j ∈ Supp f ). Thus, t j ≤ m = t j+1
(since t j+1 = m). Combining the chain of inequalities t1 ≤ t2 ≤ · · · ≤ t j with the
single inequality t j ≤ t j+1 , we obtain the longer chain of inequalities t1 ≤ t2 ≤
· · · ≤ t j ≤ t j+1 . In other words, t1 ≤ t2 ≤ · · · ≤ t j+1 . This proves (351).]
Next, we shall prove that

f = t1 , t2 , . . . , t j+1 multi . (352)

[Proof of (352): We have {m}multi ⊆ f and thus ( f \ {m}multi ) ∪ {m}multi = f (by

Proposition 7.54.7 (b), applied to {m}multi and f instead of f and g). Thus,
 
 
     
f =  f \ {m}multi  ∪ |{z}
  m = t1 , t2 , . . . , t j multi ∪ t j+1 multi
| {z }  
= t j +1
={t1 ,t2 ,...,t j }multi multi

= t1 , t2 , . . . , t j , t j+1 multi
 
by Proposition 7.54.5, applied to the numbers j and 1 and



 the j-tuple t1 , t2 , . . . , t j and the 1-tuple t j+1 instead of the numbers k 
 
and ` and the k-tuple ( a1 , a2 , . . . , ak ) and the `-tuple (b1 , b2 , . . . , b` )

= t1 , t2 , . . . , t j+1 multi .

This proves (352).]

Clearly, t1 , t2 , . . . , t j+1 are elements of the set T (since t1 , t2 , . . . , t j are elements of


T, and since t j+1 ∈ T). Hence, t1 , t2 , . . . , t j+1 is a tuple of elements of T.


Thus, t1 , t2 , . . . , t j+1 is a tuple of elements of T such that f = t1 , t2 , . . . , t j+1 multi


(by (352)) and t1 ≤ t2 ≤ · · · ≤ t j+1 (by (351)). In other words, t1 , t2 , . . . , t j+1 is an
increasing list of f (by the definition of an “increasing list”).
Hence, the multisubset
f has at least one increasing list (namely, t1 , t2 , . . . , t j+1 ).


We shall next show that t1 , t2 , . . . , t j+1 is the only increasing list of f . Indeed, let
(s1 , s2 , . . . , sk ) be any increasing list of f . Then, Proposition 7.54.17 (a) shows that
k ≥ 1 and sk = m. Also, Proposition 7.54.17 (c) shows that the list (s1 , s2 , . . . , sk−1 )
is an increasing list of the multisubset f \ {m}multi of T.
But recall that f \ {m}multi has exactly one increasing list. Thus, in particular,
f \ {m}multi has at most one increasing list. In other words, any two increasing
lists
of f \ {m}multi are equal. Hence, the lists (s1 , s2 , . . . , sk−1 ) and t1 , t2 , . . . , t j must be
equal (since both of these lists
are increasing lists of f \ {m}multi ). In other words,
(s1 , s2 , . . . , sk−1 ) = t1 , t2 , . . . , t j . In other words, k − 1 = j and

(si = ti for each i ∈ {1, 2, . . . , k − 1}) . (353)

Enumerative Combinatorics: class notes page 481

From k − 1 = j, we obtain k = j + 1. Hence, tk = t j+1 = m. Next, we claim that

si = ti for each i ∈ {1, 2, . . . , k } . (354)
[Proof of (354): Let i ∈ {1, 2, . . . , k }. We must prove that si = ti . If i ∈ {1, 2, . . . , k − 1},
then this follows from (353). Hence, for the rest of this proof, we WLOG assume
that we don’t have i ∈ {1, 2, . . . , k − 1}. Hence, i ∈ / {1, 2, . . . , k − 1}. Combining
i ∈ {1, 2, . . . , k } with i ∈
/ {1, 2, . . . , k − 1}, we obtain
i ∈ {1, 2, . . . , k} \ {1, 2, . . . , k − 1} = {k } .
In other words, i = k. Hence, si = sk = m = tk (since tk = m). In view of k = i, this
rewrites as si = ti . This proves (354).]

From (354), we obtain (s1 , s2 , . . . , sk ) = (t1 , t2 , . . . , tk ) = t1 , t2 , . . . , t j+1 (since
k = j + 1).
Now, forget that we fixed (s1 , s2 , . . . , sk ). We thus have proven that
if (s1 , s2 , . . . , sk )
is any increasing list of f , then (s1 , s2 , . . . , sk ) = t1 , t2 , . . . , t j+1 . In other words,


any increasing list of f equals t1 , t2 , . . . , t j+1 . Thus, the multisubset f has at most
one increasing list. Since we also know that the multisubset f has at least one
increasing list, we thus conclude that f has exactly one increasing list.
Now, forget that we fixed f . We thus have shown that
 
if f is a multisubset of T satisfying | f | = g + 1,
.
then f has exactly one increasing list
In other words, Proposition 7.54.15 holds under the condition that | f | = g + 1. This
completes the induction step. Hence, Proposition 7.54.15 is proven by induction.

It is now easy to prove the two propositions that Exercise 2.11.2 wants us to
prove:
Proof of Proposition 2.11.5. Proposition 7.54.15 (applied to f = S) shows that S has
exactly one increasing list. In other words, there exists exactly one increasing list of
S. In other words, there exists a unique increasing list of S. In other words, there ex-
ists a unique tuple (s1 , s2 , . . . , sk ) of elements of T such that S = {s1 , s2 , . . . , sk }multi
and s1 ≤ s2 ≤ · · · ≤ sk (because an increasing list of S is defined to be a tuple
(s1 , s2 , . . . , sk ) of elements of T such that S = {s1 , s2 , . . . , sk }multi and s1 ≤ s2 ≤
· · · ≤ sk ). In other words, there exists a unique tuple (s1 , s2 , . . . , sk ) of integers in T
satisfying S = {s1 , s2 , . . . , sk }multi and s1 ≤ s2 ≤ · · · ≤ sk (because elements of T are
the same as integers in T). In other words, there exists a unique tuple (s1 , s2 , . . . , sk )
of integers in T satisfying {s1 , s2 , . . . , sk }multi = S and s1 ≤ s2 ≤ · · · ≤ sk (because
“S = {s1 , s2 , . . . , sk }multi ” is equivalent to “{s1 , s2 , . . . , sk }multi = S”). This proves
Proposition 2.11.5.
Now, Proposition 2.11.6 can be derived from Proposition 2.11.5 in the same way
as Proposition 1.4.13 was derived from Proposition 1.4.11 (in the solution to Exer-
cise 1.4.2):
Enumerative Combinatorics: class notes page 482

Proof of Proposition 2.11.6. Proposition 2.11.5 yields that there exists a unique tuple
(s1 , s2 , . . . , sk ) of integers in T satisfying {s1 , s2 , . . . , sk }multi

= S and s1 ≤ s2 ≤

··· ≤
sk . Consider this tuple, and denote it by t1 , t2 , . . . , t p . Thus, t1 , t2 , . . . , t p is the
unique tuple of integers in T satisfying t1 , t2 , . . . , t p multi = S and t1 ≤ t2 ≤ · · · ≤
tp.
 Exercise 2.11.1 (applied to p and ti instead of k and ai ) yields that the multisubset
t1 , t2 , . . . , t p multi of T has size p. In other words, the multisubset S of T has size

p (since t1 , t2 , . . . , t p multi = S). This rewrites as |S| = p (since the size of S is
denoted by |S|). But we know that S has size m; in other words, we have |S| = m
(since the size of S is denoted by |S|). Comparing this with |S| = p, we obtain
p = m.

 Recall that the tuple t1 , t2 , . . . , t p is a p-tuple of integers in T satisfying
t1 , t2 , . . . , t p multi = S and t1 ≤ t2 ≤ · · · ≤ t p . In view of p = m, we can rewrite
this as follows: The tuple (t1 , t2 , . . . , tm ) is an m-tuple of integers in T satisfying
{t1 , t2 , . . . , tm }multi = S and t1 ≤ t2 ≤ · · · ≤ tm . Hence, there exists at least one
m-tuple (s1 , s2 , . . . , sm ) of integers in T satisfying {s1 , s2 , . . . , sm }multi = S and s1 ≤
s2 ≤ · · · ≤ sm (namely, the m-tuple (t1 , t2 , . . . , tm )). Furthermore, there exists at
most one such m-tuple (s1 , s2 , . . . , sm ) 296 . Combining the claims of the preceding
two sentences, we conclude that there exists a unique m-tuple (s1 , s2 , . . . , sm ) of
integers in T satisfying {s1 , s2 , . . . , sm }multi = S and s1 ≤ s2 ≤ · · · ≤ sm . This proves
Proposition 2.11.6.
Thus, Proposition 2.11.5 and Proposition 2.11.6 are proven, so that Exercise 2.11.2
is solved.

7.55. Solution to Exercise 2.12.1

In order to solve Exercise 2.12.1, we need to prove Proposition 2.12.4.
Proof of Proposition 2.12.4. By assumption, we have n1 + n2 + · · · + nk = n, so that
n = n1 + n2 + · · · + n k .
For each i ∈ [k + 1], we define a nonnegative integer si ∈ N by si = ni + ni+1 +
· · · + nk . Thus, in particular, s1 = n1 + n2 + · · · + nk = n and sk+1 = nk+1 + nk+2 +
· · · + nk = (empty sum) = 0.
For each i ∈ [k + 1], we have

n
|{z} − n 1 − n 2 − · · · − n i −1 = ( n 1 + n 2 + · · · + n k ) − n 1 − n 2 − · · · − n i −1
=n1 +n2 +···+nk
= n i + n i +1 + · · · + n k = s i (355)
296 Proof.


Recall that t1 , t2 , . . . , t p is the unique tuple (s1 , s2 , . . . , sk ) of integers in T satisfying
{s1 , s2 , . . . , sk }multi = S and s1 ≤ s2 ≤ · · · ≤ sk . Hence, if (s1 , s2 , . . . , sk ) is any tuple of integers
in T satisfying
{s1 , s2 , . . . , sk }multi = S and s1 ≤ s2 ≤ · · · ≤ sk , then (s1 , s2 , . . . , sk ) must equal
t1 , t2 , . . . , t p . Applying this to k = m, we conclude that if (s1 , s2 , . . . , sm ) is any m-tuple of
integers in T satisfying
{s1 , s2 , . . . , sm }multi = S and s1 ≤ s2 ≤ · · · ≤ sm , then (s1 , s2 , . . . , sm ) must
equal t1 , t2 , . . . , t p . Thus, there exists at most one such m-tuple (s1 , s2 , . . . , sm ).
Enumerative Combinatorics: class notes page 483

(since si is defined by si = ni + ni+1 + · · · + nk ).

(a) It is easy to see that Proposition 2.12.4 (a) holds if k = 0 297 . Hence, for the
rest of this proof, we WLOG assume that k 6= 0. Hence, k ≥ 1.
We shall now show two simple observations:
n − n 1 − n 2 − · · · − n i −1
 
si !
Observation 1: We have = for each
ni n i ! · s i +1 !
i ∈ [ k ].
[Proof of Observation 1: Let i ∈ [k]. Then, the definition of si+1 yields si+1 =
ni+1 + ni+2 + · · · + nk . But the definition of si yields
s i = n i + n i +1 + · · · + n k = n i + ( n i +1 + n i +2 + · · · + n k ) = n i + s i +1 .
| {z }
= s i +1

Hence, si − ni = si+1 ≥ 0 (since si+1 ∈ N), so that ni ≤ si . Thus, Proposition 1.3.9

(applied to si and ni instead of n and k) yields
 
si si ! si !
= = (since si − ni = si+1 ) .
ni ni ! · ( si − ni ) ! n i ! · s i +1 !
297 Proof. Assume that k = 0. Then,
n = n1 + n2 + · · · + nk = (empty sum) (since k = 0)
= 0.
 
n
Thus, the definition of yields
n1 , n2 , . . . , n k
 
n n! 1
= =
n1 , n2 , . . . , n k n1 !n2 ! · · · nk ! n1 !n2 ! · · · n0 !
 
since n ! = 0! = 1 and k = 0
|{z}
=0
1
= (since n1 !n2 ! · · · n0 ! = (empty product) = 1)
1
= 1.
Moreover, the products
k 
n − n 1 − n 2 − · · · − n i −1

∏ ni
,
i =1
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −1
     
n
··· ,
n1 n2 n3 nk
| {z }
=1
k −1 
n − n 1 − n 2 − · · · − n i −1

∏ ni
i =1

are all empty products (since k = 0) and thus equal 1 as well. Hence, the claim of Proposition
2.12.4 (a) boils down to 1 = 1 = 1 = 1, which is obvious. Thus, Proposition 2.12.4 (a) is proved
under the assumption that k = 0.
Enumerative Combinatorics: class notes page 484

But (355) yields n − n1 − n2 − · · · − ni−1 = si . Thus,

n − n 1 − n 2 − · · · − n i −1
   
s si !
= i = .
ni ni n i ! · s i +1 !
This proves Observation 1.]

n − n 1 − n 2 − · · · − n k −1
 
Observation 2: We have = 1.
nk

[Proof of Observation 2: We have k ∈ [k ] (since k ≥ 1). From n = n1 + n2 + · · · + nk ,

we obtain

n
|{z} − n 1 − n 2 − · · · − n k −1 = ( n 1 + n 2 + · · · + n k ) − n 1 − n 2 − · · · − n k −1 = n k .
=n1 +n2 +···+nk

n − n 1 − n 2 − · · · − n k −1
   
nk
Hence, = = 1 (by Exercise 1.3.2, applied to nk
nk nk
instead of n). This proves Observation 2.]
Now,
k 
n − n 1 − n 2 − · · · − n i −1

∏ n
i =1 | {zi }
si !
=
n i ! · s i +1 !
(by Observation 1)
k
k ∏ si !
si !
=∏ = i =1
n ! · s i +1 !
  
k k
i =1 i
∏ n i ! · ∏ s i +1 !
i =1 i =1
s1 ! · (s2 !s3 ! · · · sk !)
=
(n1 !n2 ! · · · nk !) · (s2 !s3 ! · · · sk !) · sk+1 !
 
k
 since ∏ si ! = s1 !s2 ! · · · sk ! = s1 ! · (s2 !s3 ! · · · sk !) 
 i =1 
 
 k 

 and ∏ ni ! = n1 !n2 ! · · · nk ! 

 i =1 
 k 
and ∏ si+1 ! = s2 !s3 ! · · · sk+1 ! = (s2 !s3 ! · · · sk !) · sk+1 !
 
i =1
 
s1 ! n!
= = since s1 = n and sk+1 ! = 0! = 1
(n1 !n2 ! · · · nk !) · sk+1 ! (n1 !n2 ! · · · nk !) · 1 |{z}
=0
n!
= . (356)
n1 !n2 ! · · · nk !
Enumerative Combinatorics: class notes page 485

 
n
But the definition of yields
n1 , n2 , . . . , n k
 
n n!
=
n1 , n2 , . . . , n k n1 !n2 ! · · · nk !
k 
n − n 1 − n 2 − · · · − n i −1

=∏ (by (356))
i =1
ni
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −1
     
n
= ···
n1 n2 n3 nk
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −2
     
n
= ···
n1 n2 n3 n k −1
n − n 1 − n 2 − · · · − n k −1
 
·
nk
| {z }
=1
(by Observation 2)

(here, we have split off the last factor from our product)
n − n1 n − n1 − n2 n − n 1 − n 2 − · · · − n k −2
     
n
= ···
n1 n2 n3 n k −1
k −1 
n − n 1 − n 2 − · · · − n i −1

=∏ .
i =1
ni

Thus, Proposition 2.12.4 (a) is proved.

(b) Proposition 2.12.4 (a) yields
k k
n − n 1 − n 2 − · · · − n i −1
     
n si
=∏ =∏ ∈N
n1 , n2 , . . . , n k ni n
i =1 | {z  } i =1 | {zi }
si ∈N
= (by Lemma 1.3.17,
ni applied to si and ni
(since (355) yields n−n1 −n2 −···−ni−1 =si ) instead of n and k)

(since any product of elements of N must itself belong to N). This proves Proposi-
tion 2.12.4 (b). 298

7.56. Solution to Exercise 2.12.1

In order to solve Exercise 2.12.1, we first introduce a convenient notation:

Definition 7.56.1. Let A be a set. Let k ∈ N. Let n1 , n2 , . . . , nk ∈ N. A map

f : A → [k ] is said to be (n1 , n2 , . . . , nk )-fine if it satisfies

(# of a ∈ A such that f ( a) = i ) = ni for each i ∈ [k ] . (357)

298 This proof of Proposition 2.12.4 (b) is essentially taken from [Grinbe15, solution to Exercise 3.1].
Enumerative Combinatorics: class notes page 486

Example 7.56.2. The map f : {2, 5, 6, 9, 10} → [2] that sends 2, 6, 9 to 1 and sends
5, 10 to 2 is (3, 2)-fine, since it satisfies

(# of a ∈ {2, 5, 6, 9, 10} such that f ( a) = 1) = |{2, 6, 9}| = 3 and

(# of a ∈ {2, 5, 6, 9, 10} such that f ( a) = 2) = |{5, 10}| = 2.

We can now (slightly) generalize Proposition 2.12.5 as follows:

Proposition 7.56.3. Let n ∈ N and n1 , n2 , . . . , nk ∈ N be such that n1 + n2 + · · · +

nk = n. Let A  be a set such that  | A| = n. Then, the # of (n1 , n2 , . . . , nk )-fine maps
n
f : A → [k ] is .
n1 , n2 , . . . , n k

Proposition 2.12.5 is the particular case of Proposition 7.56.3 when A = [n].

We shall now prove Proposition 7.56.3.
Proof of Proposition 7.56.3. We shall prove Proposition 7.56.3 by induction on k.
Induction base: It is easy to see that Proposition 7.56.3 holds when k = 0 299 .
This completes the induction base.
Induction step: Let m be a positive integer. Assume that Proposition 7.56.3 holds
when k = m − 1. We must prove that Proposition 7.56.3 holds when k = m.

299 Proof.
Consider the situation of Proposition 7.56.3, and assume that k = 0. We must show that
Proposition 7.56.3 holds.
Definition 2.12.1 yields
 
n n! 0!
= = (since k = 0 and n = 0)
n1 , n2 , . . . , n k n1 !n2 ! · · · nk ! n1 !n2 ! · · · n0 !
0!
= (since n1 !n2 ! · · · n0 ! = (empty product) = 1)
1
= 0! = 1.

We have

| A | = n = n1 + n2 + · · · + n k = n1 + n2 + · · · + n0 (since k = 0)
= (empty sum) = 0.

Hence, A = ∅. Also, k = 0 and thus [k] = [0] = ∅. Thus, there exists no i ∈ [k].
Both sets A and [k] are empty (since A = ∅ and [k ] = ∅). Hence, there exists exactly
one map f : A → [k] (namely, the map that sends nothing to nothing). This map f satisfies
(357) (since (357) is a vacuously true statement (because there exists no i ∈ [k])), and thus is
(n1 , n2 , . . . , nk )-fine (by the definition of “(n1 , n2 , . . . , nk )-fine”). Thus, there exists exactly one
(n1 , n2 , . . . , nk )-fine map f : A → [k] (namely, the map  we just found).  In other words, the # of
n
(n1 , n2 , . . . , nk )-fine maps f : A → [k] is 1. In view of = 1, we can rewrite this as
n1 , n
2 , . . . , nk 
n
follows: The # of (n1 , n2 , . . . , nk )-fine maps f : A → [k] is . Hence, Proposition
n1 , n2 , . . . , n k
7.56.3 is proved (under the assumption that k = 0). Thus, Proposition 7.56.3 holds when k = 0.
Enumerative Combinatorics: class notes page 487

Note that m ∈ [m] (since m is positive).

We have assumed that Proposition 7.56.3 holds when k = m − 1. In other words,
the following statement holds:
Statement 1: Let n ∈ N and n1 , n2 , . . . , nm−1 ∈ N be such that n1 +
n2 + · · · + nm−1 = n. Let A be a set such that | A| = n. Then, the  # of
n
(n1 , n2 , . . . , nm−1 )-fine maps f : A → [m − 1] is .
n 1 , n 2 , . . . , n m −1
We must prove that Proposition 7.56.3 holds when k = m. In other words, we
must prove the following statement:
Statement 2: Let n ∈ N and n1 , n2 , . . . , nm ∈ N be such that n1 +
n2 + · · · + nm = n. Let A be a set suchthat | A| = n.Then, the # of
n
(n1 , n2 , . . . , nm )-fine maps f : A → [m] is .
n1 , n2 , . . . , n m
[Proof of Statement 2: The following proof is rather similar to the 1st approach to
counting sur (m, n) in Subsection 2.4.5.
From n1 + n2 + · · · + nm = n, we obtain
n = n 1 + n 2 + · · · + n m = ( n 1 + n 2 + · · · + n m −1 ) + n m .
Thus, n − nm = n1 + n2 + · · · + nm−1 ∈ N (since n1 , n2 , . . . , nm−1 ∈ N). Hence,
n − nm ≥ 0, so that nm ≤ n. Thus, Theorem 1.3.9 (applied to k = nm ) yields
 
n n!
= . (358)
nm nm ! · (n − nm )!
But A is an n-element set (since | A| = n). Hence, Theorem 1.3.12 (applied to
k = nm and S = A) yields
 
n
= (# of nm -element subsets of A)
nm
= (# of subsets J of A satisfying | J | = nm ) .
Comparing this with (358), we obtain
n!
(# of subsets J of A satisfying | J | = nm ) = . (359)
nm ! · (n − nm )!
Given a map f : A → [m], we let J f be the set of all a ∈ A such that f ( a) = m. 300
Clearly, this set J f is a subset of A. Moreover, if f : A → [m] is an (n1 , n2 , . . . , nm )-
fine map, then this subset J f of A satisfies J f = nm 301 . Hence, the sum rule

300 For example,if A = {2, 5, 6, 9, 13 } and m = 2, and if f : A → [m] is the map written in two-line
2 5 6 9 13
notation as , then J f = {2, 6, 9}, because 2, 6, 9 are precisely the a ∈ A such
2 1 2 2 1
that f ( a) = 2.
301 Proof. Let f : A → [ m ] be an ( n , n , . . . , n )-fine map. Thus, we have
1 2 m

(# of a ∈ A such that f ( a) = i ) = ni for each i ∈ [m]

Enumerative Combinatorics: class notes page 488

yields

(# of all (n1 , n2 , . . . , nm ) -fine maps f : A → [m])

∑


= # of all (n1 , n2 , . . . , nm ) -fine maps f : A → [m] satisfying J f = J .
J ⊆ A;
| J |=nm
(360)

Now, fix a subset J of A that satisfies | J | = nm . What is the # of all (n1 , n2 , . . . , nm )-

fine maps f : A → [m] satisfying J f = J ? Such a map f must satisfy


J = J f = { a ∈ A | f ( a) = m} by the definition of J f .

In other words, such a map f must send all elements of J, but no other elements,
to m. In other words, it must send all elements of J to m, and send all other
elements of A to numbers different from m. The latter numbers are, of course, the
elements of [m] \ {m} = [m − 1]. Thus, an (n1 , n2 , . . . , nm )-fine map f : A → [m]
satisfying J f = J must send all elements of J to m, and send all elements of A \ J
to elements of [m − 1]. Hence, if f is such a map, then f is uniquely determined
by its values at the elements of A \ J, and said values must belong to [m − 1]. In
other words, if f is such a map, then f is uniquely determined by its restriction
f | A\ J , and furthermore this restriction must “actually” be a map from A \ J to
[m − 1], in the sense that all its values belong to [m − 1]. Moreover, in order for
the map f to be (n1 , n2 , . . . , nm )-fine, its restriction f | A\ J (regarded as a map from
A \ J to [m − 1]) must be (n1 , n2 , . . . , nm−1 )-fine302 . This is actually a necessary
(by the definition of “(n1 , n2 , . . . , nm )-fine”). We can apply this to i = m (since m ∈ [m]), and
thus obtain
(# of a ∈ A such that f ( a) = m) = nm .
But J f was defined as the set of all a ∈ A such that f ( a) = m. Hence,

J f = (# of a ∈ A such that f ( a) = m) = nm .

Qed.
302 Proof. Let f : A → [m] be a map satisfying J f = J. We must prove that the map f is
(n1 , n2 , . . . , nm )-fine if and only if its restriction f | A\ J (regarded as a map from A \ J to [m − 1])
is (n1 , n2 , . . . , nm−1 )-fine.
For each i ∈ [m], we let Ai be the logical statement “(# of a ∈ A such that f ( a) = i ) = ni ”.
Recall that J f = J, so that
 
J = J f = { a ∈ A | f ( a) = m} by the definition of J f .

Hence, | J | = |{ a ∈ A | f ( a) = m}| = (# of a ∈ A such that f ( a) = m). Therefore,

(# of a ∈ A such that f ( a) = m) = | J | = nm .

In other words, the statement Am holds (since Am was defined to be the logical statement
“(# of a ∈ A such that f ( a) = m) = nm ”). In logical notation, this rewrites as follows: Am ⇐⇒
(true).
Enumerative Combinatorics: class notes page 489

and sufficient condition; i.e., if f : A → [m] is a map satisfying J f = J, and if its

restriction f | A\ J (regarded as a map from A \ J to [m − 1]) is (n1 , n2 , . . . , nm−1 )-fine,

Now, let i ∈ [m − 1]. Then, i ≤ m − 1 < m, so that i 6= m. But J = { a ∈ A | f ( a) = m}.

Hence,
each a ∈ J satisfies f ( a) = m. (361)
Hence, if there was an element a ∈ A satisfying both f ( a) = i and a ∈ J at the same time, then
this element would satisfy f ( a) = m (by (361)), which would contradict f ( a) = i 6= m. Thus,
there is no such element a. In other words, there exists no a ∈ A such that f ( a) = i and a ∈ J.
In other words,
(# of a ∈ A such that f ( a) = i and a ∈ J ) = 0.
But each a ∈ A satisfies either a ∈ J or a ∈
/ J (but not both at the same time). Hence, the sum
rule yields

(# of a ∈ A such that f ( a) = i )
= (# of a ∈ A such that f ( a) = i and a ∈ J ) + (# of a ∈ A such that f ( a) = i and a ∈
/ J)
| {z }
=0
= (# of a ∈ A such that f ( a) = i and a ∈
/ J)
= (# of a ∈ A such that a ∈
/ J and f ( a) = i )
= (# of a ∈ A \ J such that f ( a) = i )

(since the a ∈ A satisfying a ∈

/ J are precisely the elements of A \ J).
Now, forget that we fixed i. We thus have proved that

(# of a ∈ A such that f ( a) = i )
= (# of a ∈ A \ J such that f ( a) = i ) (362)

for each i ∈ [m − 1].

We have the following chain of logical equivalences:
 
f | A\ J (regarded as a map from A \ J to [m − 1] ) is (n1 , n2 , . . . , nm−1 ) -fine
  
    
⇐⇒ # of a ∈ A \ J such that f | A\ J ( a) = i  = ni for each i ∈ [m − 1]
  
 | {z }  
= f ( a)
(by the definition of “ (n1 , n2 , . . . , nm−1 ) -fine”)
 
 
 
⇐⇒ (# of a ∈ A \ J such that f ( a) = i ) = ni for each i ∈ [m − 1]


| {z } 
=(# of a∈ A such that f ( a)=i )
(by (362))
 
 
 
⇐⇒ 
 (# of a ∈ A such that f ( a) = i ) = ni for each i ∈ [m − 1]

 | {z } 
⇐⇒ (Ai holds)
(since Ai is the logical statement “(# of a∈ A such that f ( a)=i )=ni ”)

⇐⇒ (Ai holds for each i ∈ [m − 1])

⇐⇒ (A1 ∧ A2 ∧ · · · ∧ Am−1 ) . (363)
Enumerative Combinatorics: class notes page 490

then the map f itself must be (n1 , n2 , . . . , nm )-fine303 . Thus, in order to construct
an (n1 , n2 , . . . , nm )-fine map f : A → [m] satisfying J f = J, we need to construct an
(n1 , n2 , . . . , nm−1 )-fine map f : A \ J → [m − 1]. Hence,


# of all (n1 , n2 , . . . , nm ) -fine maps f : A → [m] satisfying J f = J
= (# of all (n1 , n2 , . . . , nm−1 ) -fine maps f : A \ J → [m − 1]) . (364)

(Formally speaking, this can be proved using the bijection principle304 .)

On the other hand, J ⊆ A, and thus an application of Theorem 1.4.7 (a) yields
| A \ J | = | A| − | J | = n − nm ∈ N. Also, recall that n − nm = n1 + n2 + · · · +
|{z} |{z}
=n =nm
nm−1 , so that n1 + n2 + · · · + nm−1 = n − nm . Hence, Statement 1 (applied to n − nm
and A \ J instead of n and A) yields that the # of (n1 , n2 , . . . , nm−1 )-fine maps

On the other hand, we have the following chain of logical equivalences:

( f is (n1 , n2 , . . . , nm ) -fine)
 
 
 
⇐⇒ 
 (# of a ∈ A such that f ( a) = i ) = ni for each i ∈ [m]

 | {z } 
⇐⇒ (Ai holds)
(since Ai is the logical statement “(# of a∈ A such that f ( a)=i )=ni ”)

(by the definition of “ (n1 , n2 , . . . , nm ) -fine”)

⇐⇒ (Ai holds for each i ∈ [m])
⇐⇒ (A1 ∧ A2 ∧ · · · ∧ Am )
 

⇐⇒ (A1 ∧ A2 ∧ · · · ∧ Am−1 ) ∧ Am 
 
|{z}
⇐⇒(true)
⇐⇒ ((A1 ∧ A2 ∧ · · · ∧ Am−1 ) ∧ (true))
⇐⇒ (A1 ∧ A2 ∧ · · · ∧ Am−1 )
 
⇐⇒ f | A\ J (regarded as a map from A \ J to [m − 1] ) is (n1 , n2 , . . . , nm−1 ) -fine

(by (363)). In other words, the map f is (n1 , n2 , . . . , nm )-fine if and only if its restriction f | A\ J
(regarded as a map from A \ J to [m − 1]) is (n1 , n2 , . . . , nm−1 )-fine. Qed.
303 This was proven in the previous footnote.
304 More precisely, we need to apply the bijection principle to the bijection

n o
(n1 , n2 , . . . , nm ) -fine maps f : A → [m] satisfying J f = J
→ {(n1 , n2 , . . . , nm−1 ) -fine maps f : A \ J → [m − 1]}

that sends each f to its restriction f | A\ J (considered as a map from A \ J to [m − 1]).

Here, of course, we are using the fact that if an (n1 , n2 , . . . , nm )-fine map f : A → [m] satisfies
J f = J, then the restriction f | A\ J can be considered as a map from A \ J to [m − 1], since it does
not take the value m.
Enumerative Combinatorics: class notes page 491

n − nm
 
f : A \ J → [m − 1] is . In other words,
n 1 , n 2 , . . . , n m −1
(# of all (n1 , n2 , . . . , nm−1 ) -fine maps f : A \ J → [m − 1])
n − nm (n − nm )!
 
= =
n 1 , n 2 , . . . , n m −1 n1 !n2 ! · · · nm−1 !
n − nm
 
(by the definition of ). Thus, (364) becomes
n 1 , n 2 , . . . , n m −1


# of all (n1 , n2 , . . . , nm ) -fine maps f : A → [m] satisfying J f = J
= (# of all (n1 , n2 , . . . , nm−1 ) -fine maps f : A \ J → [m − 1])
(n − nm )!
= . (365)
n1 !n2 ! · · · nm−1 !
Now, forget that we fixed J. We thus have proved (365) for each subset J of A
satisfying | J | = nm . Hence, (360) becomes
(# of all (n1 , n2 , . . . , nm ) -fine maps f : A → [m])
∑


= # of all ( n 1 , n 2 , . . . , n m ) -fine maps f : A → [ m ] satisfying J f = J
J ⊆ A; | {z }
| J |=nm (n − nm )!
=
n1 !n2 ! · · · nm−1 !
(by (365))
(n − nm )! (n − nm )!
= ∑ n1 !n2 ! · · · nm−1 !
= (# of subsets J of A satisfying | J | = nm )
} n1 !n2 ! · · · nm−1 !
J ⊆ A;
| {z
| J |=nm n!
=
nm ! · (n − nm )!
(by (359))
n! (n − nm )! n! n!
= · = =
nm ! · (n − nm )! n1 !n2 ! · · · nm−1 ! nm ! · (n1 !n2 ! · · · nm−1 !) n1 !n2 ! · · · nm !
(since nm ! · (n1 !n2 ! · · · nm−1 !) = (n1 !n2 ! · · · nm−1 !) · nm ! = n1 !n2 ! · · · nm !)
 
n
=
n1 , n2 , . . . , n m
 
n n!
(since was defined to be ). In other words, the # of
n1 , n2 , . . . , n m n1 !n2 ! · · · nm ! 
n
(n1 , n2 , . . . , nm )-fine maps f : A → [m] is . This proves Statement
n1 , n2 , . . . , n m
2.]
Now, Statement 2 is proven. In other words, Proposition 7.56.3 holds when
k = m. This completes the induction step. Thus, Proposition 7.56.3 is proved.
Proof of Proposition 2.12.5. We have |[n]| = n. Hence, Proposition 7.56.3  (applied to
n
A = [n]) yields that the # of (n1 , n2 , . . . , nk )-fine maps f : [n] → [k ] is .
n1 , n2 , . . . , n k
Enumerative Combinatorics: class notes page 492

In other words, the # of maps f : [n] → [k ] satisfying

(# of a ∈ [n] such that f ( a) = i ) = ni for each i ∈ [k ] (366)

 
n
is (because the (n1 , n2 , . . . , nk )-fine maps f : [n] → [k ] are precisely
n1 , n2 , . . . , n k
the maps f : [n] → [k] satisfying (366)305 ). This proves Proposition 2.12.5.
Thus, Exercise 2.12.2 is solved.

7.57. Solution to Exercise 2.12.3

Solution to Exercise 2.12.3. Write the n-tuple α ∈ X n in the form α = (α1 , α2 , . . . , αn ).
Thus,
(# of times i appears in α) = (# of a ∈ [n] such that α a = i ) . (367)
We have assumed that β is an anagram of the n-tuple α. Thus, β is an n-tuple that
can be obtained from α by permuting its entries
 (by the definition ofan “anagram”).
In other words, β is an n-tuple of the form ασ(1) , ασ(2) , . . . , ασ(n) , where σ is a
of [n]. Consider this σ.
permutation  
Thus, β = ασ(1) , ασ(2) , . . . , ασ(n) . Thus,
 
(# of times i appears in β) = # of b ∈ [n] such that ασ(b) = i . (368)

But σ is a permutation of [n]. In other words, σ is a bijection from [n] to [n]

(by the definition of a permutation). Hence, its inverse map σ−1 : [n] → [n] is
well-defined.
If a ∈ [n] satisfies α a = i, then the element σ−1 ( a) is a b ∈ [n] such that ασ(b) = i
306 . Hence, the map

n o
f : { a ∈ [ n ] | α a = i } → b ∈ [ n ] | ασ(b) = i ,
a 7 → σ −1 ( a )

is well-defined. Consider this map f .

305 bythe definition of “(n1 , n2 , . . . , nk )-fine”

306 Proof.Let a ∈ [n] satisfy α a = i. We must prove that the element σ−1 ( a) is a b ∈ [n] such that
ασ(b) = i.
Clearly, σ−1 ( a) ∈ [n]. Also, σ σ−1 ( a) = a, so that ασ(σ−1 (a)) = α a = i. Thus, σ−1 ( a) is a

b ∈ [n] such that ασ(b) = i (since σ−1 ( a) ∈ [n] and ασ(σ−1 (a)) = i). Qed.
Enumerative Combinatorics: class notes page 493

If b ∈ [n] satisfies ασ(b) = i, then the element σ (b) is an a ∈ [n] such that α a = i
307 .Hence, the map
n o
g : b ∈ [ n ] | ασ(b) = i → { a ∈ [ n ] | α a = i } ,
b 7→ σ (b)
is well-defined. Consider this map g.
It is easy to see that f ◦ g = id 308 . Similarly, g ◦ f = id. Combining f ◦ g = id
with g ◦ f = id, we conclude thatnthe maps f and g are o mutually inverse. Hence, the
map f : { a ∈ [n] | α a = i } → b ∈ [n] | ασ(b) = i is invertible, i.e., is bijective.
Thus, the bijection principle yields
n o
|{ a ∈ [n] | α a = i }| = b ∈ [n] | ασ(b) = i .

Now, (367) yields

(# of times i appears in α) = (# of a ∈ [n] such that α a = i )
n o
= |{ a ∈ [n] | α a = i }| = b ∈ [n] | ασ(b) = i
 
= # of b ∈ [n] such that ασ(b) = i
= (# of times i appears in β) (by (368)) .
In other words, (# of times i appears in β) = (# of times i appears in α). This solves
Exercise 2.12.3.

7.58. Solution to Exercise 2.12.4

Instead of solving Exercise 2.12.4 directly, we shall first prove a more general state-
ment (stated in the more flexible language of maps instead of tuples):
Theorem 7.58.1. Let A and B be two finite sets. Let X be any set. Let f : A → X
and g : B → X be two maps. Assume that

(# of a ∈ A such that f ( a) = i )
= (# of b ∈ B such that g (b) = i ) (369)
307 Proof.Let b ∈ [n] satisfy ασ(b) = i. We must prove that the element σ (b) is an a ∈ [n] such that
α a = i. In other
n words, we must prove
o that σ (b) ∈ [n] and ασ(b) = i. But this is obvious. Qed.
308 Proof. Let c ∈ b ∈ [n] | ασ(b) = i . We shall show that ( f ◦ g) (c) = id (c).
). But ( f ◦ g) (c) = f ( g (c)) = σ−1 ( g (c)) (by the
Indeed, the definition of g yields g (c)= σ (c

definition of f ). Thus, ( f ◦ g) (c) = σ−1  g (c)  = σ−1 (σ (c)) = c = id (c).

 
|{z}
=σ(c)

n Forget that we fixed

o c. We thus have shown that ( f ◦ g) (c) = id (c) for each c ∈
b ∈ [n] | ασ(b) = i . In other words, f ◦ g = id.
Enumerative Combinatorics: class notes page 494

for each i ∈ X. Then, there exists a bijection σ : A → B such that f = g ◦ σ.

Proof of Theorem 7.58.1. The set A is finite; i.e., there are only finitely many a ∈ A.
Likewise, there are only finitely many b ∈ B.
Let Y be the subset { f ( a) | a ∈ A} of X. Then, Y is finite (since there are only
finitely many a ∈ A).
Let Z be the subset { g (b) | b ∈ B} of X. Then, Z is finite (since there are only
finitely many b ∈ B).
Let W be the union Y ∪ Z. This set W is a finite set (since it is the union of the
two finite sets Y and Z) and is a subset of X (since Y and Z are subsets of X).
It is easy to see that

f (c) ∈ W for each c ∈ A (370)

309 . Likewise,
g (d) ∈ W for each d ∈ B. (371)
Let i ∈ W. We shall construct a bijection from the set { a ∈ A | f ( a) = i } to the
set {b ∈ B | g (b) = i }.
Indeed, i ∈ W ⊆ X. Also, we have

|{ a ∈ A | f ( a) = i }| = (# of a ∈ A such that f ( a) = i )
= (# of b ∈ B such that g (b) = i ) (by (369))
= |{b ∈ B | g (b) = i }| .

In other words, { a ∈ A | f ( a) = i } and {b ∈ B | g (b) = i } are two sets of the

same size. Thus, Theorem 1.1.7 (applied to { a ∈ A | f ( a) = i } and {b ∈ B | g (b) = i }
instead of X and Y) yields that there exists a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i }. Choose such a bijection, and denote it by hi .
Forget that we fixed i. Thus, for each i ∈ W, we have found

a bijection hi from { a ∈ A | f ( a) = i } to {b ∈ B | g (b) = i } .

310

We shall now “glue” these bijections hi together to a single bijection u : A → B

which will satisfy f = g ◦ u. Here is how this is done in detail:

309 Proof. Let c ∈ A. We have Y = { f ( a) | a ∈ A} (by the definition of Y). But from c ∈ A, we obtain
f (c) ∈ { f ( a) | a ∈ A} = Y ⊆ Y ∪ Z = W (since W was defined to be Y ∪ Z). This proves (370).
310 If you are aware of the Axiom of Choice, you might wonder whether we have used it here. The

answer is “no, not really”. We have made a choice (namely, choosing a bijection hi ) for each
i ∈ W. This amounts to finitely many choices (since W is a finite set). But you can make
finitely many choices without invoking the Axiom of Choice (indeed, this can easily be proved
by induction on the number of choices). Thus, we are not using the Axiom of Choice here.
Enumerative Combinatorics: class notes page 495

For each c ∈ A, the element h f (c) (c) ∈ B is well-defined311 . Hence, we can define
a map
u : A → B,
c 7 → h f (c) ( c ) .
  −1
For each d ∈ B, the element h g(d) (d) ∈ A is well-defined312 . Hence, we can
define a map
v : B → A,
  −1
d 7 → h g(d) (d) .

We have u ◦ v = id 313 and v ◦ u = id 314 . Thus, the maps u and v are

mutually inverse. Hence, the map u : A → B is invertible, i.e., bijective. In other

311 Proof. Let c ∈ A. Let i = f (c). Then, c is an a ∈ A satisfying f ( a) = i (since c ∈ A and f (c) = i).
In other words, c ∈ { a ∈ A | f ( a) = i }.
But i = f (c) ∈ W (by (370)). Hence, hi is a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i } (by the definition of hi ). Therefore, hi (c) is a well-defined element of
{b ∈ B | g (b) = i } (since c ∈ { a ∈ A | f ( a) = i }). Thus, hi (c) ∈ {b ∈ B | g (b) = i } ⊆ B.
In view of i = f (c), this rewrites as h f (c) (c) ∈ B. Thus, the element h f (c) (c) ∈ B is well-defined.
Qed.
312 Proof. Let d ∈ B. Let i = g ( d ). Then, d is a b ∈ B satisfying g ( b ) = i (since d ∈ B and g ( d ) = i).

In other words, d ∈ {b ∈ B | g (b) = i }.

But i = g (d) ∈ W (by (371)). Hence, hi is a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i } (by the definition of hi ). Thus, the inverse map (hi )−1 is a bijection from
{b ∈ B | g (b) = i } to { a ∈ A | f ( a) = i }. Therefore, (hi )−1 (d) is a well-defined element of
{ a ∈ A | f ( a) = i } (since d ∈ {b ∈ B | g (b) = i }). Thus, (hi )−1 (d) ∈ { a ∈ A | f ( a) = i } ⊆
  −1   −1
A. In view of i = g (d), this rewrites as h g(d) (d) ∈ A. Thus, the element h g(d) (d) ∈ A
is well-defined. Qed.
313 Proof. Let d ∈ B. Let i = g ( d ). Then, d is a b ∈ B satisfying g ( b ) = i (since d ∈ B and g ( d ) = i).

In other words, d ∈ {b ∈ B | g (b) = i }.

But i = g (d) ∈ W (by (371)). Hence, hi is a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i } (by the definition of hi ). Thus, the inverse map (hi )−1 is a bijection from
{b ∈ B | g (b) = i } to { a ∈ A | f ( a) = i }. Therefore, (hi )−1 (d) is a well-defined element of
{ a ∈ A | f ( a) = i } (since d ∈ {b ∈ B | g (b) = i }). Thus, (hi )−1 (d) ∈ { a ∈ A | f ( a) = i }.
  −1
But the definition of v yields v (d) = h g(d) (d) = (hi )−1 (d) (since g (d) = i). Hence,
hi (v (d)) = d.
Also, v (d) = (hi )−1 (d) ∈ { a ∈ A | f ( a) = i }. In other words, v (d) is an a ∈ A satisfying
f ( a) = i. In other words, v (d) ∈ A and f (v (d)) = i. Now,
(u ◦ v) (d) = u (v (d)) = h f (v(d)) (v (d)) (by the definition of u)
= hi (v (d)) (since f (v (d)) = i )
= d = id (d) .
Forget that we fixed d. We thus have proved that (u ◦ v) (d) = id (d) for each d ∈ B. In other
words, u ◦ v = id.
314 Proof. Let c ∈ A. Let i = f ( c ). Then, c is an a ∈ A satisfying f ( a ) = i (since c ∈ A and f ( c ) = i).
Enumerative Combinatorics: class notes page 496

words, u : A → B is a bijection. Moreover, we have f = g ◦ u 315 . Hence, there

exists a bijection σ : A → B such that f = g ◦ σ (namely, σ = u). This proves
Theorem 7.58.1.
Exercise 2.12.4 now easily follows:
Solution to Exercise 2.12.4. Write the n-tuple α ∈ X n in the form α = (α1 , α2 , . . . , αn ).
Thus, for each i ∈ X, we have

(# of times i appears in α) = (# of a ∈ [n] such that α a = i ) . (372)

Write the n-tuple β ∈ X n in the form β = ( β 1 , β 2 , . . . , β n ). Thus, for each i ∈ X,

we have

(# of times i appears in β) = (# of b ∈ [n] such that β b = i ) . (373)

Define a map

f : [n] → X,
a 7→ β a .

In other words, c ∈ { a ∈ A | f ( a) = i }.
But i = f (c) ∈ W (by (370)). Hence, hi is a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i } (by the definition of hi ). Therefore, hi (c) is a well-defined element of
{b ∈ B | g (b) = i } (since c ∈ { a ∈ A | f ( a) = i }). Thus, hi (c) ∈ {b ∈ B | g (b) = i } ⊆ B.
But the definition of u yields u (c) = h f (c) (c) = hi (c) (since f (c) = i).
Also, u (c) = hi (c) ∈ {b ∈ B | g (b) = i }. In other words, u (c) is a b ∈ B satisfying g (b) = i.
In other words, u (c) ∈ B and g (u (c)) = i. Now,
  −1
(v ◦ u) (c) = v (u (c)) = h g(u(c)) (u (c)) (by the definition of v)
= (hi )−1 (u (c)) (since g (u (c)) = i )
=c (since u (c) = hi (c))
= id (c) .

Forget that we fixed c. We thus have proved that (v ◦ u) (c) = id (c) for each c ∈ A. In other
words, v ◦ u = id.
315 Proof. Let c ∈ A. Let i = f ( c ). Then, c is an a ∈ A satisfying f ( a ) = i (since c ∈ A and f ( c ) = i).

In other words, c ∈ { a ∈ A | f ( a) = i }.
But i = f (c) ∈ W (by (370)). Hence, hi is a bijection from { a ∈ A | f ( a) = i } to
{b ∈ B | g (b) = i } (by the definition of hi ). Therefore, hi (c) is a well-defined element of
{b ∈ B | g (b) = i } (since c ∈ { a ∈ A | f ( a) = i }). Thus, hi (c) ∈ {b ∈ B | g (b) = i }.
But the definition of u yields u (c) = h f (c) (c) = hi (c) (since f (c) = i). Hence, u (c) = hi (c) ∈
{b ∈ B | g (b) = i }. In other words, u (c) is a b ∈ B satisfying g (b) = i. In other words,
u (c) ∈ B and g (u (c)) = i. Comparing f (c) = i with ( g ◦ u) (c) = g (u (c)) = i, we obtain
f ( c ) = ( g ◦ u ) ( c ).
Forget that we fixed c. We thus have proved that f (c) = ( g ◦ u) (c) for each c ∈ A. In other
words, f = g ◦ u.
Enumerative Combinatorics: class notes page 497

Define a map

g : [n] → X,
b 7→ αb .

For each i ∈ X, we have

 
 
 
# of a ∈ [n] such that f ( a) = i
 | {z } 
 
=βa
(by the definition of f )
= (# of a ∈ [n] such that β a = i )
= (# of b ∈ [n] such that β b = i ) (here, we have renamed the index a as b)
= (# of times i appears in β) (374)

(by (373)). Likewise, for each i ∈ X, we have

(# of b ∈ [n] such that g (b) = i )

= (# of times i appears in α) . (375)

Now, for each i ∈ X, we have

(# of a ∈ [n] such that f ( a) = i )

= (# of times i appears in β) (by (374))
= (# of times i appears in α) (by (253))
= (# of b ∈ [n] such that g (b) = i ) (by (375)) .

Hence, Theorem 7.58.1 (applied to A = [n] and B = [n]) yields that there exists a
bijection σ : [n] → [n] such that f = g ◦ σ. Consider this σ.
The map σ is a bijection from [n] to [n]. In other words, σ is a permutation
of [n]. Moreover, we have β i = ασ(i) for each i ∈ [n] 316 . In other words,
 
( β 1 , β 2 , . . . , β n ) = α σ (1) , α σ (2) , . . . , α σ ( n ) .
 
Hence, β = ( β 1 , β 2 , . . . , β n ) = ασ(1) , ασ(2) , . . . , ασ(n) . Thus, the n-tuple β is ob-
tained from the n-tuple (α1 , α2 , . . . , αn ) by permuting its entries (namely, using the
permutation σ). In other words, β is an anagram of (α1 , α2 , . . . , αn ) (by the definition
of an anagram). In other words, β is an anagram of α (since α = (α1 , α2 , . . . , αn )).
This solves Exercise 2.12.4.

316 Proof. Let i ∈ [n]. The definition of f yields f (i ) = β i . Hence, β i = f (i ) = ( g ◦ σ ) (i ) =

|{z}
= g◦σ
g (σ (i )) = ασ(i) (by the definition of g). Qed.
Enumerative Combinatorics: class notes page 498

References
[17f-hw1s] Darij Grinberg, UMN Fall 2017 Math 4707 & Math 4990 homework
set #1 with solutions, http://www.cip.ifi.lmu.de/~grinberg/t/17f/
hw1s.pdf

[17f-hw2s] Darij Grinberg, UMN Fall 2017 Math 4707 & Math 4990 homework
set #2 with solutions, http://www.cip.ifi.lmu.de/~grinberg/t/17f/
hw2s.pdf

[17f-hw3s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #3 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw3os.pdf

[17f-hw4s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #4 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw4os.pdf

[17f-hw5s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #5 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw5os.pdf

[17f-hw6s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #6 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw6os.pdf

[17f-hw7s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #7 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw7os.pdf

[17f-hw8s] Darij Grinberg, UMN Fall 2017 Math 4990 homework set #8 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/17f/hw8os.pdf

[18f] Darij Grinberg, UMN Fall 2018 Math 5705: “black”board notes, https:
//www.cip.ifi.lmu.de/~grinberg/t/18f/

[18f-hw1s] Darij Grinberg, UMN Fall 2018 Math 5705 homework set #1 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/hw1s.pdf

[18f-hw2s] Darij Grinberg, UMN Fall 2018 Math 5705 homework set #2 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/hw2s.pdf

[18f-hw3s] Darij Grinberg, UMN Fall 2018 Math 5705 homework set #3 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/hw3s.pdf

[18f-hw4s] Darij Grinberg, UMN Fall 2018 Math 5705 homework set #4 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/hw4s.pdf

[18f-mt1s] Darij Grinberg, UMN Fall 2018 Math 5705 midterm #1 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/mt1s.pdf

[18f-mt2s] Darij Grinberg, UMN Fall 2018 Math 5705 midterm #2 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/mt2s.pdf
Enumerative Combinatorics: class notes page 499

[18f-mt3s] Darij Grinberg, UMN Fall 2018 Math 5705 midterm #3 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18f/mt3s.pdf
[18s] Darij Grinberg, UMN Spring 2018 Math 4707: “black”board notes, https:
//www.cip.ifi.lmu.de/~grinberg/t/18s/
[18s-hw1s] Darij Grinberg, UMN Spring 2018 Math 4707 homework set #1 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/18s/hw1s.pdf
[18s-hw2s] Darij Grinberg, UMN Spring 2018 Math 4707 homework set #2 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/18s/hw2s.pdf
[18s-hw3s] Darij Grinberg, UMN Spring 2018 Math 4707 homework set #3 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/18s/hw3s.pdf
[18s-hw4s] Darij Grinberg, UMN Spring 2018 Math 4707 homework set #4 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/18s/hw4s.pdf
[18s-hw5s] Darij Grinberg, UMN Spring 2018 Math 4707 homework set #5 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/18s/hw5s.pdf
[18s-mt1s] Darij Grinberg, UMN Spring 2018 Math 4707 midterm #1 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18s/mt1s.pdf
[18s-mt2s] Darij Grinberg, UMN Spring 2018 Math 4707 midterm #2 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/18s/mt2s.pdf
[19f-hw0s] Darij Grinberg, Drexel Fall 2019 Math 222 homework set #0 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/hw0s.pdf
[19f-hw1s] Darij Grinberg, Drexel Fall 2019 Math 222 homework set #1 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/hw1s.pdf
[19f-hw2s] Darij Grinberg, Drexel Fall 2019 Math 222 homework set #2 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/hw2s.pdf
[19f-hw3s] Darij Grinberg, Drexel Fall 2019 Math 222 homework set #3 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/hw3s.pdf
[19f-hw4s] Darij Grinberg, Drexel Fall 2019 Math 222 homework set #4 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/hw4s.pdf
[19f-mt1s] Darij Grinberg, Drexel Fall 2019 Math 222 midterm #1 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/mt1s.pdf
[19f-mt2s] Darij Grinberg, Drexel Fall 2019 Math 222 midterm #2 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/mt2s.pdf
[19f-mt3s] Darij Grinberg, Drexel Fall 2019 Math 222 midterm #3 with solutions,
http://www.cip.ifi.lmu.de/~grinberg/t/19fco/mt3s.pdf
Enumerative Combinatorics: class notes page 500

[19s] Darij Grinberg, Introduction to Modern Algebra (UMN Spring 2019 Math
4281 notes), 29 June 2019.
http://www.cip.ifi.lmu.de/~grinberg/t/19s/notes.pdf

[19s-hw0s] Darij Grinberg, UMN Spring 2019 Math 4281 homework set #0 with solu-
tions, http://www.cip.ifi.lmu.de/~grinberg/t/19s/hw0s.pdf

[21s] Darij Grinberg, Algebraic Combinatorics (Drexel Spring 2021 Math 701
lecture notes), 4 May 2021.
https://www.cip.ifi.lmu.de/~grinberg/t/21s/lecs.pdf

[Aigner07] Martin Aigner, A Course in Enumeration, Graduate Texts in Mathemat-

ics #238, Springer 2007.

[AigZie14] Martin Aigner, Günter M. Ziegler, Proofs from the Book, 6th edition,
Springer 2018.

[AndEri04] George E. Andrews, Kimmo Eriksson, Integer Partitions, Cambridge

University Press 2004.

[AndFen04] Titu Andreescu, Zuming Feng, A Path to Combinatorics for Undergradu-

ates: Counting Strategies, Springer 2004.

[Andrew16] George E. Andrews, Euler’s Partition Identity – Finite Version, 2016.

http://www.personal.psu.edu/gea1/pdf/317.pdf

[ArdSta10] Federico Ardila, Richard P. Stanley, Tilings, The Mathematical In-

telligencer 32 (2010), pp. 32–43. A preprint can be found at
arXiv:math/0501170v3 .

[BCHR11] Fred Butler, Mahir Can, Jim Haglund, Jeffrey B. Remmel, Rook Theory
Notes, 3 January 2011.
http://www.math.ucsd.edu/~remmel/files/Book.pdf

[BenDre07] Arthur T. Benjamin and Gregory P. Dresden, A Combinatorial Proof of

Vandermonde’s Determinant, The American Mathematical Monthly, Vol.
114, No. 4 (Apr., 2007), pp. 338–341.
(Also available at http://scholarship.claremont.edu/hmc_fac_pub/
524/ .)

[BenQui03] Arthur T. Benjamin and Jennifer J. Quinn, Proofs that Really Count: The
Art of Combinatorial Proof, The Mathematical Association of America,
2003.

[BenQui04] Arthur T. Benjamin and Jennifer J. Quinn, Proofs that Really Count: The
Magic of Fibonacci Numbers and More, Mathematical Adventures for Stu-
dents and Amateurs, (David F. Hayes and Tatiana Shubin, editors),
Spectrum Series of MAA, pp. 83–98, 2004.
Enumerative Combinatorics: class notes page 501

[BenQui08] Arthur T. Benjamin and Jennifer J. Quinn, An Alternate Approach to Al-

ternating Sums: A Method to DIE for, The College Mathematics Journal,
Volume 39, Number 3, May 2008, pp. 191-202(12).

[Bona12] Miklos Bóna, Combinatorics of Permutations, 2nd edition, Tay-

lor&Francis 2012.
https://doi.org/10.1201/b12210

[Bourba68] Nicolas Bourbaki, Theory of Sets, Springer 1968.

[Chu19] Hùng Viê.t Chu, The Fibonacci Sequence and Schreier-Zeckendorf Sets,
Journal of Integer Sequences, 22 (2019), Article 19.6.5.
https://cs.uwaterloo.ca/journals/JIS/VOL22/Chu2/chu9.html

[CoLiOs15] David A. Cox, John Little, Donal O’Shea, Ideals, Varieties, and Algo-
rithms: An Introduction to Computational Algebraic Geometry and Commu-
tative Algebra, 4th edition, Springer 2015.

[Comtet74] Louis Comtet, Advanced Combinatorics: The Art of Finite and Infinite Ex-
pansions, D. Reidel Publishing Company, 1974.

[Day16] Martin V. Day, An Introduction to Proofs and the Mathematical Vernacular,

7 December 2016.
https://web.archive.org/web/20180712152432/https://www.math.
vt.edu/people/day/ProofsBook/IPaMV.pdf .

[Elman19] Richard Elman, Lectures on Abstract Algebra, 20 September 2019.

https://www.math.ucla.edu/~rse/algebra_book.pdf

[GalQua19] Jean Gallier and Jocelyn Quaintance, Algebra, Topology, Differential Cal-
culus, and Optimization Theory For Computer Science and Engineering, 2
August 2019.
https://www.cis.upenn.edu/~jean/gbooks/geomath.html

[Galvin17] David Galvin, Basic discrete mathematics, 13 December 2017.

http://www-users.math.umn.edu/~dgrinber/comb/
60610lectures2017-Galvin.pdf
(The URL might change, and the text may get updated. In
order to reliably obtain the version of 13 December 2017,
you can use the archive.org Wayback Machine: https:
//web.archive.org/web/20180205122609/http://www-users.math.
umn.edu/~dgrinber/comb/60610lectures2017-Galvin.pdf .)

[Goodma15] Frederick M. Goodman, Algebra: Abstract and Concrete, edition 2.6, 1

May 2015.
http://homepage.math.uiowa.edu/~goodman/algebrabook.dir/
book.2.6.pdf .
Enumerative Combinatorics: class notes page 502

[Granvi05] Andrew Granville, Binomial coefficients modulo prime powers, preprint.

https://web.archive.org/web/20181024055320/http://ebooks.
bharathuniv.ac.in/gdlc1/gdlc1/EngineeringMergedLibraryv3.0/
AndrewGranville/BinomialCoefficientsModuloPrimePowers(5579)
/BinomialCoefficientsModuloPrimePowers-AndrewGranville.pdf

[Grinbe15] Darij Grinberg, Notes on the combinatorial fundamentals of algebra, 10 Jan-

uary 2019.
http://www.cip.ifi.lmu.de/~grinberg/primes2015/sols.pdf
The numbering of theorems and formulas in this link might shift
when the project gets updated; for a “frozen” version whose num-
bering is guaranteed to match that in the citations above, see https:
//github.com/darijgr/detnotes/releases/tag/2019-01-10 .

[Grinbe16a] Darij Grinberg, Notes on linear algebra, version of 13 December 2016.

https://github.com/darijgr/lina

[Grinbe16b] Darij Grinberg, 4th QEDMO, Problem 13 with solution, version of 28

May 2016.
http://www.cip.ifi.lmu.de/~grinberg/QEDMO4P13.pdf

[Grinbe17] Darij Grinberg, The Lucas and Babbage congruences, 10 January 2019.
https://www.cip.ifi.lmu.de/~grinberg/lucascong.pdf

[GrKnPa94] Ronald L. Graham, Donald E. Knuth, Oren Patashnik, Concrete Mathe-

matics, Second Edition, Addison-Wesley 1994.
See https://www-cs-faculty.stanford.edu/~knuth/gkp.html for er-
rata.

[Guicha20] David Guichard, An Introduction to Combinatorics and Graph Theory, 30

January 2020.
https://www.whitman.edu/mathematics/cgt_online/book/

[Hammac15] Richard Hammack, Book of Proof, 3rd edition.

https://www.people.vcu.edu/~rhammack/BookOfProof/

[Joyce17] David Joyce, Introduction to Modern Algebra, 5 December 2017.

https://mathcs.clarku.edu/~djoyce/ma225/algebra.pdf

[Joyner08] W. D. Joyner, Mathematics of the Rubik’s cube, 19 August 2008.

https://web.archive.org/web/20160304122348/http://www.
permutationpuzzles.org/rubik/webnotes/ (link to the PDF at
the bottom).

[Kastel61] P. W. Kasteleyn, The statistics of dimers on a lattice: I. The number of dimer

arrangements on a quadratic lattice, Physica 27 (1961), pp. 1209–1225.
https://doi.org/10.1016/0031-8914(61)90063-5
Enumerative Combinatorics: class notes page 503

[Knapp16] Anthony W. Knapp, Basic Algebra, digital second edition 2016.

http://www.math.stonybrook.edu/~aknapp/download.html

[Knuth93] Donald E. Knuth, Johann Faulhaber and sums of powers, Math. Comp. 61
(1993), no. 203, 277–294.
https://www.ams.org/journals/mcom/1993-61-203/
S0025-5718-1993-1197512-7/

[Kratte17] Christian Krattenthaler, Lattice Path Enumeration, arXiv:1503.05930v3,

published in: Handbook of Enumerative Combinatorics, M. Bóna
(ed.), Discrete Math. and Its Appl., CRC Press, Boca Raton-London-
New York, 2015, pp. 589–678.
https://arxiv.org/abs/1503.05930v3

[LeLeMe16] Eric Lehman, F. Thomson Leighton, Albert R. Meyer, Mathematics for

Computer Science, revised Tuesday 6th June 2018,
https://courses.csail.mit.edu/6.042/spring18/mcs.pdf .

[LLPT95] D. Laksov, A. Lascoux, P. Pragacz, and A. Thorup, The LLPT Notes,

edited by A. Thorup, 1995,
http://www.math.ku.dk/~thorup/notes/sympol.pdf .

[Loehr11] Nicholas A. Loehr, Bijective Combinatorics, Chapman & Hall/CRC 2011.

[Mestro14] Romeo Meštrović, Lucas’ theorem: its generalizations, extensions and ap-
plications (1878–2014), arXiv:1409.3820v1.

[MusPro07] Gregg Musiker, James Propp, Combinatorial Interpretations for Rank-Two

Cluster Algebras of Affine Type, The Electronic Journal of Combinatorics
14 (2007), #R15.
http://www.combinatorics.org/ojs/index.php/eljc/article/
view/v14i1r15

[Newste19] Clive Newstead, An Infinite Descent into Pure Mathematics, version 0.4,
1 January 2020.
https://infinitedescent.xyz

[Niven69] Ivan Niven, Formal Power Series, The American Mathematical Monthly
76, No. 8 (Oct., 1969), pp. 871–889.
https://www.maa.org/programs/maa-awards/writing-awards/
formal-power-series

[Read80] Ronald C. Read, A Note on Tiling Rectangles with Dominoes, Fibonacci

Quarterly 18 (1980), pp. 24–27.
https://www.fq.math.ca/Scanned/18-1/read.pdf
Enumerative Combinatorics: class notes page 504

[Sagan01] Bruce Sagan, The Symmetric Group, Graduate Texts in Mathematics 203,
2nd edition 2001.
https://doi.org/10.1007/978-1-4757-6804-6
See https://users.math.msu.edu/users/bsagan/Books/Sym/errata.
pdf for errata.

[Sagan19] Bruce Sagan, Combinatorics: The Art of Counting, preliminary version, 7

January 2020.
https://users.math.msu.edu/users/bsagan/Books/Aoc/aoc.pdf

[sage] The Sage Developers, SageMath, the Sage Mathematics Software System
(Version 9.0), 2019, https://www.sagemath.org .

[SageBook] Paul Zimmermann et al, Computational Mathematics with SageMath, De-

cember 2018.
http://sagebook.gforge.inria.fr/english.html

[Smid09] Vita Smid, Inclusion-Exclusion Principle: Proof by Mathematical Induction,

2 December 2009.
https://faculty.math.illinois.edu/~nirobles/files453/iep_
proof.pdf

[Stanle15] Richard P. Stanley, Catalan Numbers, 1st edition 2015.

See http://math.mit.edu/~rstan/catalan/ for errata.

[Stanle11] Richard P. Stanley, Enumerative Combinatorics, volume 1, Second edition,

version of 15 July 2011. Available at http://math.mit.edu/~rstan/ec/
.
See http://math.mit.edu/~rstan/ec/ for errata.

[Stanle01] Richard P. Stanley, Enumerative Combinatorics, volume 2, First edition,

Cambridge University Press 2001.
See http://math.mit.edu/~rstan/ec/ for errata.

[Strick13] Neil Strickland, MAS201 Linear Mathematics for Applications, lecture

notes, 28 September 2013.
http://neil-strickland.staff.shef.ac.uk/courses/MAS201/

[Stucky15] Eric Stucky, An Exposition of Kasteleyn’s Solution of the Dimer Model,

senior thesis at Harvey Mudd College, 2015.
https://scholarship.claremont.edu/hmc_theses/89/

[Tou17] Erik R. Tou, Math Origins: The Totient Function, Convergence (Septem-
ber 2017), DOI:10.4169/convergence20170923.
https://www.maa.org/press/periodicals/convergence/
math-origins-the-totient-function
Enumerative Combinatorics: class notes page 505

[Uecker17] Torsten Ueckerdt, Lecture Notes Combinatorics (2017), 30 May 2017.

http://www.math.kit.edu/iag6/lehre/combinatorics2017s/media/
script.pdf
See http://www.cip.ifi.lmu.de/~grinberg/algebra/
ueckerdt-script2017-errata.pdf for an inofficial list of errata.

[Vorobi02] Nicolai N. Vorobiev, Fibonacci Numbers, Translated from the Russian by

Mircea Martin, Springer 2002 (translation of the 6th Russian edition).

[Walker87] Elbert A. Walker, Introduction to Abstract Algebra, Random House-

/Birkhauser, New York, 1987.

[Ward91] James Ward, 100 years of Dixon’s identity, Irish Mathematical Society
Bulletin 27 (1991), pp. 46–54.
https://www.maths.tcd.ie/pub/ims/bull27/bull27_46-54.pdf

[Warner71] Seth Warner, Classical Modern Algebra, Prentice-Hall 1971.

[White10] Dennis White, Math 4707: Inclusion-Exclusion and Derangements, 18

October 2010.
http://www-users.math.umn.edu/~reiner/Classes/Derangements.
pdf

[Wilf04] Herbert S. Wilf, generatingfunctionology, 2nd edition 2004.

https://www.math.upenn.edu/~wilf/DownldGF.html

[Wilf09] Herbert S. Wilf, Lectures on Integer Partitions, 2009.

https://www.math.upenn.edu/~wilf/PIMS/PIMSLectures.pdf