JOURNAL

OF ALGEBRA

149, 438453

(1992)

Derived

Functors of /-adic Completion and Local Homology

J. P. C.
GREENLEES
Unil;ersi/y

AND

J. P.

MAY
Chicago, Illinois 60637

Department

q/ Mathematics, Communicated

of Chicago,

by Richard

G. Swan

Received

June 1. 1990

In recent topological work [2], we were forced to consider the left derived functors of the I-adic completion functor, where I is a finitely generated ideal in a commutative ring A. While our concern in [2] was with a particular class of rings, namely the Burnside rings A(G) of compact Lie groups G, much of the foundational work we needed was not restricted to this special case. The essential point is that the modules we consider in [2] need not be finitely generated and, unless G is finite, say, the ring ,4(G) is not Noetherian. There seems to be remarkably little information in the literature about the behavior of I-adic completion in this generality. We presume that interesting non-Noetherian commutative rings and interesting non-finitely generated modules arise in subjects other than topology. We have therefore chosen to present our algebraic work separately, in the hope that it may be of value to mathematicians working in other fields. One consequence of our study, explained in Section 1, is that I-adic completion is exact on a much larger class of modules than might be expected from the key role played by the Artin-Rees lemma and that the deviations from exactness can be computed in terms of torsion products. However, the most interesting consequence, discussed in Section 2, is that the left derived functors of I-adic completion usually can be computed in terms of certain local homology groups, which are defined in a fashion dual to the definition of the classical local cohomology groups of Grothendieck. These new local homology groups may well be relevant to algebraists and algebraic geometers. In particular, we obtain a universal coefficients theorem for calculating these groups from local cohomology in Section 3; the classical local duality spectral sequence is a very special case. The brief Section 4 gives an analysis of the behavior of composites of left derived functors of I-adic completion. The still briefer Section 5 describes 438
0021-8693/92 $5.00
Copyright 0 1992 by Academic Press, Inc All rights ol reproduction m any form reserved

DERIVED FUNCTORSOF COMPLETION

439

the right derived functors of I-adic completion, which are much less interesting (and irrelevant to our topological applications). We restrict ourselves to the main points here, and the arguments are quite elementary. Commutative ring theorists will see that we have left many very natural questions unaswered. In particular, we have left sheaf theoretic generalizations to the reader.

0. PRELIMINARIES

To establish notations and context, we recall briefly the definitions of left derived functors and of some basic constructions that we shall use. Let A&! be the category of modules over a commutative ring A. A a-functor 9 is a sequence {Oi 1 ia 0} of covariant functors Di: A& -+ A& together with natural connecting homomorphisms ai: .;(M) -+ D,+ ,(M) for short exact sequences

such that the following are zero sequences (all composites are zero): . . + DJM) + D,(M) -+ Di(M)
-+ . . . + D,(M) + D,(M) -+ D,+ ,(M)

+ 0;

9 is exact if these sequences are exact. 9 is effaceable if, for each M, there is an epimorphism N -+ M such that DiN-+ DiM is zero for i > 0. This obviously holds if Di F = 0 for i > 0 when F is free. Let l? A& -+ A,,& be an additive functor. Its left derived functors are given by an exact and effaceable a-functor Pr= {LiT) together with a natural transformation E: LOT+ r, which is an isomorphism on free modules. The functor LJ is right exact and its left derived functors for i > 0 are the same as those of r. For any &functor 9, a natural tranformation J,: D, + L,T extends uniquely to a map {fi}: 9 -+ yr of %functors. Moreover, {f,} is an isomorphism if and only if 9 is exact and effaceable andf, is an isomorphism on free modules. For an A-module M, 2IM can be constructed by taking the homology of the complex obtained by applying r to a free resolution of M. Details may be found in [l, V, Sects. 2-3 1. Define the cone, or caliber, Ck of a chain map k: X+ Y by (Ck)i= Yi@XiPl, with differential di(y,x)=(di(y)+ki-,(x), -dip,(x)). Define the suspension CX and desuspension C- X by (ZX), = Xi-, and (c-lx)i=xi+*, with the differential -d. We have a short exact sequence 0 + Y -+ Ck + CX -+ 0, and the connecting homomorphism of the derived

440

GREENLEES

AND

MAY

long exact sequence in homology is k,. It is convenient to define the fiber of k to be Fk=CC(-k). Given a sequence of chain maps f: x + x+ , r b 0, define a map 1: @r + @Xr by I(X) =x -,f(x) for x E X. Define the homotopy colimit, or telescope, of the sequence {fr} to be Cr and denote it Tel(X). Then Hi(Tel(X)) = Colim H,(X). The composite of the projection from Cr to its first variable and the canonical map @X + Colim X is a homology isomorphism [: Tel(X) + Colim X. We shall need an observation about the behavior of telescopes with respect to tensor products. Given two sequences f r: X -+X+ and g: Y-+ Y+l, we obtain a sequencefrOg: Xr@ Y-x+@ Y+.
LEMMA 0.1. There is a natural homology isomorphism

4: Tel(X@

Y) + Tel(X)@Tel(

I).

ProoJ: Using an ordered pair notation for elements of the relevant calibers, we specify 5 by the explicit formula WOY, x0.Y) = (0, x)0 (y, 0) + (- 1 )dbyf(X), O)@ (0, y)

+ (x, 0) 0 (y, 0).

A tedious computation homology isomorphism Tel(X@ Y)

shows that 4 commutes with differentials. because the diagram b Tel(X)@Tel( Y)

It is a

Ii Colim(X@
r

Y) -+ Colim(X@ r. A

Il@i Y) Y) 2 Colim(X)@Colim(
r s

commutes. Here the bottom left arrow is the diagonal cofmality isomorphism. Dually to the telescope, given chain maps f: x -+ x- for r b 1, define a map rr: xX+ x X by rr(x) = (xr-fr+l(x+ )). Define the homotopy limit, or microscope, of the sequence {fr> to be Fx and denote it Mic(X). Then there are short exact sequences
(0.2)

0 -+ Lim Hi+ l(Xr) -+ H,(Mic(X))

+ Lim H,(F)

-+ 0.

Observe that a degreewise short exact sequence o+{cr}+{Y}+{z}~o

DERIVED FUNCTORSOFCOMPLETION

441

of systems of chain complexes gives rise to a short exact sequence 0 -+ Mic(X) + Mic( Y) -+ Mic(Z) -+ 0

and thus to a long exact sequence of homology groups. Here Lim denotes the first right derived functor of the inverse limit functor. We shall be concerned only with inverse sequences, for which the higher right derived functors of Lim vanish. Thus a short exact sequence of inverse sequences gives a six term exact sequence of Lims and Lims. We say that an inverse sequence {M} is pro-zero if, for each r, there exists s > r such that M --f M is zero; of course, if {M) is pro-zero, then LimM=O and LimM=O.

1. THE LEFT DERIVED FUNCTORS OF I-ADIC COMPLETION Let I be an ideal in our commutative ring A. For an A-module IV, define M; = Lim M/IM. Let L: denote the ith left derived functor of I-adic completion. We begin by obtaining a construction of these functors that leads to a description of the Lf(M) in terms of torsion products. Let x be a free resolution of A/I and construct chain maps f: x-+ XP1 over the quotient maps A/I + A/I~ . 1.1. The functors Lj(A4) are computable as the homology groups of the complexes Mic(X@ M). Therefore, by (0.2), there are short
PROPOSITION

exact sequences (the rightmost term in the zerolh being M, )

0 -+ Lim Torf+ ,(A/Z, M) + Lf + Lim Tor:(A/Z,

M) + 0.

Proof: The H,(Mic(X@ M)) clearly give an exact &functor. If M is free, the evident natural map E: H,(Mic(X@ M)) + Lim(A/I@ M) = M, is an isomorphism and H, (Mic(X@ M)) = 0 for i > 0.

We need some restrictive hypotheses to proceed further. In the rest of the paper, all ideals are assumed to be finitely generated.
DEFINITION 1.2. Let a E A. For an A-module M, let T(a; M) denote the kernel of a: M + A4 and observe that T(a; M) c r(ar ; M) for r > 1. Say that M has bounded a-torsion if this increasing sequence stabilizes, for example if A is Noetherian and A4 is finitely generated.

Remarks 1.3. (i) Observe that a: M+ M restricts to a map T(a+ ; M) -+ r(a; M) for each r. It is easily checked that M has bounded a-torsion if and only if the inverse sequence {T(a; M)} is pro-zero. Thus Lim T(a; M) = 0 and Lim f(ar; M) = 0 if M has bounded a-torsion.

442

GREENLEESAND

MAY

(ii) If NcM, then r(crr; N)=f(cr; M)nN, so that N has bounded cc-torsion if M does. If each of a set M, of A-modules has bounded a-torsion with a common bound r, then the sum and product of the M, have bounded cr-torsion. In particular, if A itself has bounded z-torsion, then so does every submodule of any free A-module.
EXAMPLES 1.4. If A is the quotient of the polynomial ring generated by (~1, 1r 3 1) by the ideal generated by {cY~, 1Y> 1}, then A has y, unbounded a-torsion. As pointed out by Swan, if k is a field and if CI,/I, and x,, s 2 1, are indeterminates, then the sub k-algebra A of k(a, fi, x,~) which is generated by tl, /I, the x,, and the elements y,>, = CL~X,//Y s 3 r 2 1 is r for an example of an integral domain in which A/(B) has unbounded u-torsion for every r. PROPOSITION 1.5. Let I= (LX)and assumethat A has bounded cc-torsion. If Lim IJar; M) = 0 and Lim ZJcC; M) = 0, for example if M has bounded cc-torsion, then LA(M) z M; and L:(M) = 0 for i > 0. Moreover, the following conclusionshold for any A-module M.

(i)

There is a short exact sequence 0 + Lim Tor;(A/I, M) + L;(M) -+ M; M). -+ 0.

(ii) (iii) Proof.

L:(M) L:(M)

E Lim Tor$A/Z,
for i>2.

Tensoring M with the diagram O-+f(cc; A)+ A +(a)+0

0 + (u)+

A + A/(a) 4 0

and inspecting, we see that Tor,(A/Z, M) z r(a; M)/~(cx; A) M. There results an exact sequence

of inverse systems. If Lim r(ccr; M) = 0 and Lim T(a; M) = 0, the six term Lim-Lim exact sequence and our hypothesis on A imply that Lim Tor,(A/Z, M) = 0 and Lim Tor,(A/I, M) = 0. In view of Proposition 1.1, it remains to show that Lim Tor,(A/I, M)=O and LimTor,(A/I,M)=O for all M when i>2. If O+N-+F+M+O is

DERIVED

FUNCTORS

OF COMPLETION

443

exact, where F is free, then N has bounded cr-torsion. The conclusion follows inductively from the connecting isomorphisms Tori+ ,(A/Z, M) E Tor,(A/I, N),
i> 1.

To generalize to arbitrary finitely generated ideals, we need to understand the behavior of composites of completions. We begin with the following observation (in which J need not be finitely generated).
LEMMA

1.6. Let I = (J, ~1) and suppose that Lim Lim T(cc; M/JM) s r = 0.

Then M, Proof

is isomorphic

to (M;

),^ .

For each r and s, we have the two short exact sequences

0 + T(cc; M/JM) + M/JM + cr(M/JM) + 0

O-+

cr(M/JM)

+M,:M+M,(s,S)M+O

For each fixed s, the Lim-Lim 0 -+ Lim T(cr; M/SM)

exact sequence gives exact sequences + Lim T(cc; M/SM)

+ 0

+ MJ -+ Lim tl( M/JM)

0 -+ Lim a(M/JM)

25 I + Mj + Lim M/(cr, J) M + 0

This diagram implies the short exact sequence 0 -+ aM; + Lim aM/JM -+ Lim T(cc; M/JM)
+ 0 (*I

and the map of short exact sequences O+

CYM,^ +M; I/ -+ M; + M;/aM; -+ 0 (**I

I 0 + Lim crM/JM

I + Lim M/(a, J) M ---f 0.

As s varies, the sequences above all give exact sequences of inverse systems. By hypothesis, the Lim-Lim exact sequence, and the fact that Lim Lim is always zero for bi-countably indexed systems (e.g., by a spectral sequence in Roos [7]), the exact sequences (*) give rise to isomorphisms Lim crMj + Lim Lim crM/SM

444 and

GREENLEES

AND

MAY

Lim aM,^ + Lim Lim aM/JM. Now application the commutative of the Lim-Lim exact sequence to the diagram (**) gives diagram with exact rows -+ 0

0 -+ Lim cxM; + A4; + (MS ),^ -+ LimcrM;

0 -+ Lim Lim $M/JM

Iz II I
-+ M;

--f M,* + Lim Lim c?M/JM is an isomorphism.

1%

-+ 0

By the live lemma, (MS ),^ + Ml

We need a conveniently verifiable criterion for checking that the hypothesis of the previous lemma holds. The following observation gives us one (and, here again, J need not be finitely generated). It also gives a means of verifying the hypothesis of Proposition 1.5 for modules of the form M; that does not require boundedness of their a-torsion.
LEMMA 1.7. Let I = (J, a). Multiplication M/J+ M + M/JM induce a map T(M+; M/J+ by a and the quotient map

M) + r(cC; M/JM). is pro-zero, for example if = 0, r

If the resulting inverse system { r(~; MJJM)} each MIJM has bounded cc-torsion, then

Lim Lim T(cc, M/JM)

1 r

Lim T(a; A4 ; ) = 0,
r Proof:

and

Lim f(cr; M,^ ) = 0.

The left exactness of Lim implies that, for any cc,J, and M, T(a; M; ) = r( a; Lim M/FM) z Lim T(a; M/FM).

Write Y,, s = ~(cc; M/SM). This is a bi-indexed system, and the diagonal system Y, r is cofinal in it. We have an isomorphism Lim Lim Y, J g Lim Y,, ~. r ., r. J By a spectral sequence of Roos [7], we also have a short exact sequence O+LimLim r and similarly s Y,,-+Lim r. s Y,,r+LimLim , s Yr,s-+O,

with the roles of r and s reversed. The result follows.

DERIVEDFUNCTORSOF

COMPLETION

445

DEFINITION 1.8. Let a= {ul, .... a,} be a sequence of elements of A. Write Z(0) = 0 and Z(i) = (a,, .... a;). Say that a is a pro-regular sequence for A4 if the inverse sequence f (a:; M/1( i - 1) M) is pro-zero for 1 < i < n. Say that the ring A is good if every a is a pro-regular sequence for A. Clearly A is good if A/J has bounded a-torsion for every finitely generated ideal J (including 0) and every element a. THEOREM 1.9. Let I= (a,, .... a,,) and write J= (a,, .... z,_ L) and a = a,. Assume that A has bounded ai-torsion for each i and that a is a pro-regular sequencefor A. If a is a pro-regular sequencefor an A-module M, then L;(M) z A4; and L:(M) = 0 for i > 0. Moreover, the following conclusions hold,for any A-module M.

(i) (ii)

LA(M) g Li(L$M)). For 1 Gidn-

1, there is a short exact sequence

-+ L;(M) -+ L;(L;,(M)). ,(M)) -+ 0.

0 -+ L;((LJ(M))

(iii) (iv)

L:(M)

2 LI;(Li-

L:(M)=0

for i>n+

1.

Prooj Proposition 1.5 handles the case n = 1. Assume inductively that the conclusion holds for J. By [ 1, XVII, Sect. 71, there is a pair of cornposite functor spectral sequences, (Ei, ,} and { EL, u}, which both converge to the same hyperhomology groups 2*. They have E*-terms

E;, 4= LpW;o (j 1) (Ml = &(L;U,^

where X is a free resolution of M, and
E;, y = L;@;(M)).

)I>

It is clear that a is a pro-regular sequence for any free A-module. Proposition 1.5 and the previous two lemmas give that
Ei,,=O

Thus

for q> 1

and

q, 0 = Hp(u-;

LA ) = ffp(X,^ 1.

Therefore Y,, = Lj,M. For the first statement of the theorem, the induction hypothesis, Proposition 1.5, and the previous two lemmas give that
E&=0

for q>O,

Ei,,=L;(M;

)=0 for p>O,

and

E;,,=M;.

It follows that Y* = 0 for p > 0 and that Y0 = M, . For the second statement, the induction hypothesis implies that E; 4 =0 for p > 1 and for q > n - 1. Thus E* = E, and (i) through (iv) follow.

446

GREENLEESAND MAY

Note that (i) holds even though Ml need not be isomorphic to (MJ^ )% in general. The point is that these two functors agree on free modules and so have the same derived functors. Theorems 3.3 and 3.4 below imply better vanishing results than (iv) for Noetherian rings and Burnside rings. For a good ring A, we conclude from the first statement that I-adic completion is an exact functor when restricted to those A-modules A4 for which a is a pro-regular sequence. It is obvious that Noetherian rings are good, and so are all Burnside rings A(G) [2]. Some bad rings are exhibited in Examples 1.4.

2. LOCAL HOMOLOGY AND DERIVED FUNCTORS

We begin by recalling Grothendiecks cohomology groups [4; 5, Sect. 21. definition and calculation of local

DEFINITIONS 2.1. For CIE A, let K.(R) be the chain complex LX:A + A, where the two copies of A are in degrees 1 and 0, respectively. For a sequence a = {cil, .... a,}, let K.(a) = K.(cc,) @ . . . @K.(a,). The identity map in degree 0 and multiplication by CI in degree 1 give a chain map zqcc+ ) + K.(d), and thus, by tensoring, a chain map K.(a+) + K.(a). Let M be an A-module and define the local cohomology groups of M at the ideal Z= (cz,, .... LY,)to be

H?(M)

= Z-Z*(Colim Hom(K.(a),

M)).

For a space X, a closed subspace Y, and a sheaf 9 of Abelian groups over X, let Z,(X; 9) be the group of sections of F with support in Y. The functor Tr(X, ?) on sheaves is left exact, and its right derived functors are denoted H*y(X, 9).
THEOREM 2.2. Let X= Spec(A) H:(M) and Y = V(Z). Then z Ht(X; ATi),

where fi is the associated sheaf of M.

Zf A is Illoetherian, then
M).

H:(M)

z Colim Ext*(A/Z,

This identifies local cohomology groups as right derived functors. We shall define certain local homology groups and verify that they agree with the left derived functors L:(M) under mild hypotheses. We begin with a reformulation of the definition of local cohomology.

DERIVEDFUNCTORSOFCOMPLETION

447

Remarks 2.3. For tx E A, let K(N) be the cochain complex ~1: A -+ A, where the two copies of A are in degrees 0 and 1, respectively. For a sequence a = { ct,, .... an}, let K(a) = K(cc,)@ ... 0 K(cc,). The identity map in degree 0 and multiplication by GI in degree 1 give a cochain map K(a) -+ K(a+ ), and thus, by tensoring, a cochain map K(a) -+ K(a + ). These cochain complexes and cochain maps are obtained by applying Hom(?, A) to the chain complexes and chain maps in Definitions 2.1, and we have an isomorphism of direct systems Hom(K.(a), M) z K(a)
@ M.

Define K(a) = Colim K(a), and observe that K(a) complex A -+ A[l/a]. We have an evident isomorphism
H:(M) 2 H*(K(a) @ M).

is just the cochain

The homology isomorphism Tel K(a) -+ K(az ) gives a projective approximation of the flat cochain complex K(a). By the Kiinneth spectral sequence, this approximation induces an isomorphism H:(M) z H*(Tel
K(a) 0 M).

This suggests the following definition,

DEFINITION

which seems to be new. groups of A4 at I by

M)).

2.4.

Define the local homology H,(M) = H,(Hom(Tel

K(a),

A formal duality argument shows that Hom(Te1 K(a), M) z Mic Hom(K(a), M), these isomorphisms

and clearly Hom(K(a), M) g K.(a) 0 M. Putting together, we obtain the alternative description
H:(M) z H, (Mic(K.(a) @ M)).

The resemblance to the description L:(M) in Proposition no surprise. E H,(Mic(X@M))

1.1 is obvious, and the following result should now come as

THEOREM 2.5. Let Z= (a,, .... a,). Assume that A has bounded a,-torsion for each i and that a is a pro-regular sequence for A. Then

Hpf)

z L,(M).

448

GREENLEESAND MAY

Since the K.(a) are free chain complexes, the H:(M) certainly give an exact Sfunctor. We need only construct a natural mapf,: Hi(M) + L;(M) and show that f0 is an isomorphism and Hi(M) = 0 for i> 0 when M is free. We proceed in three steps, first handling the case n = 1, next constructing a spectral sequence that will allow induction, and then completing the proof.
LEMMA 2.6. Let I= (a), where A has bounded cc-torsion. Then H/,(M) z L:(M).

Proof: The free complex K.(cx) over A/(a) is not a resolution, but it gives the first two terms of a free resolution X. We thus obtain a map of inverse systems K.(cc) + X and thus a map of microscopes. The homology of K.(cc) @ M is M/crM in degree zero and r(cC; M) in degree one. If M is free, the system T(a; M) is pro-zero and thus H;(M) = M; and H!(M) = 0 for i > 0 by the short exact sequence for the computation of the homology of microscopes. LEMMA 2.7. Let I= J+ K. Then there is a spectral sequence {E} conuerges to H:(M) and has Ei. y = H$H,(M)). which

Prooj Let a and p be sequences of generators for J and K. By Lemma 0.1 and the evident adjunction, we have a homology isomorphism

<*: Hom(Te1 K(a),

Hom(Te1 K(p),

M)) + Hom(Te1 K(a, /Y), M).

A standard argument with double complexes yields the conclusion.

Proof of Theorem 2.5. Let J= Z(n - 1) and a = CC,.Lemma 2.6 gives the result for (c() and we may assume it for J. By the previous result, the induction hypothesis, and Theorem 1.9(i), we have H;(M) z Hpz;(M)) z LG(Li(M)) z L&w).

If M is free, Theorem 1.9 gives that Hi(M) z L:(M) is zero for q > 0 and is M,^ for q = 0, and Proposition 1.5 gives that Hg(MJ^ ) z Lg(M; ) is zero for p>O. Thus, when M is free, Ez,,= 0 unless p = q = 0 and therefore H:(M) = 0 for n > 0. This completes the proof.
3. A UNIVERSAL COEFFICIENTS SPECTRAL SEQUENCE

We can use the relationship between local homology and local cohomology to obtain a duality, or universal coefficients, spectral sequence. It is the most useful tool for explicit calculation of local homology groups.

DERIVED FUNCTORSOFCOMPLETION PROPOSITION

449

3.1.

There is a fourth quadrant spectral sequence {E,;d,: EW+E;+W-r+l)

which converges to H,(M)

in total (homological) degree -(p + q) and has E;,y = ExV(HJA),

M).

Proof Replace A4 in Hom(Tei K(a), M) by an injective resolution Y of M. To keep track of the grading, think of Tel K(a) as a complex graded in non-positive degrees, so that Hom(Te1 K(a), Y) is a (cohomological) bicomplex. Filtering so as to take the homology of Y first we obtain Hom(Te1 K(a), M) on the E,-level and H!+(M) on the E,-level, with no further differentials and with trivial extensions. Filtering so as to take the homology of Tel K(a) first, we obtain the spectral sequence we want. This spectral sequence looks a little strange. If H;(A) = 0 for k > n, then the non-zero terms of EFy lie on the Oth through ( -n)th rows of the fourth quadrant, while the non-zero terms of EC4 lie on the Oth through nth diagonals in the seventh octant; that is, E;, y = 0 if q < -n or q > 0 and EP;y = 0 if either -(p + q) < 0 or - (p + q) > n. The differentials wipe out all but finitely many of the non-zero terms present in E,. The following immediate observation is quite useful.
COROLLARY

3.2. Zf Hi(A ) = 0 for i > k, then H:(M) = 0 for i > k.

This gains force from the following 3.6.51 or [6, 2.71).

THEOREM

theorem of Grothendieck

(see [3,

3.3. rf A is Noetherian, then Hi(A) = 0 for i > dim A.

Even though Burnside rings need not be Noetherian, conclusion holds for them [2]; recall that they have dimension
THEOREM

the same one.

Hj(A)=O

3.4. If A is the Burnside ring of a compact Lie group, then for i> 1.

COROLLARY 3.5. Suppose that A is a Noetherian ring of dimension one or a Burnside ring. Then there is an exact sequence

0 + Exti(H:(A),

M) + Hi(M) -+ Hom(Hy(A),

M) + Exti(H:(A),

M) + 0

and an isomorphism H:(M) g Hom(H:(A),

481/149/2-12

M).

450

GREENLEESAND

MAY

The spectral sequence of Proposition 3.1 generalizes Grothendiecks local duality spectral sequence. To see this, we consider modules of the form M= Hom(N, Q), where Q is injective. Here our &-term and abutment take the following alternative forms.
LEMMA 3.6. For modules L and N and injective modules Q, there is a natural isomorphism

ExtP(L, Hom(N,
Proof

Q)) g ExtP(N, Hom(L,

Q)).

There is an evident natural isomorphism Hom(L, Hom(N, Q)) E Hom(N, Hom(L, Q)), Q) is an injective

If X is a projective resolution resolution of Hom(L, Q).

LEMMA

of L, then Hom(X,

3.7. For modules N and injective modules Q, there is a natural isomorphism

Hom(H;(N),
Proof:

Q) g H!(Hom(N,

Q)).

Apply homology

to the evident isomorphisms Q)).

Hom(Te1 K(a) 0 N, Q) g Hom(Te1 K(a), Hom(N,

After the second degree is raised by n so as to put the non-zero terms in the first quadrant, the spectral sequence of Proposition 3.1 takes the same form as the local duality spectral sequence.
PROPOSITION 3.8. Write DN= Hom(N, Q), where Q is injective, and assumethat H:(A) = 0 for q > n. There is a spectral sequence

{E,;d,: which converges to DH:(N)

EFY

+EP+r,q-r+I

in total degree n - q-p

and has

E; y = ExtP( N, DHl- Y(A)).

Here A is any commutative ring, I is any finitely generated ideal, N is any A-module, and Q is any injective A-module. In the special case when A is a complete local ring of dimension n, I is its maximal ideal, N is finitely generated, and Q is a dualizing module, this is precisely [S, Theorem 6.83.

DERIVED FUNCTORSOFCOMPLETION

451

4. COMPOSITES OF DERIVED FUNCTORS Let E: LAM+ M,^ be the natural epimorphism. We also have a natural and y is an isomorphism if A4 = N,^ . Since the zeroth left map y: M-+M,^, derived functor of the identity functor is the identity functor, there results a natural map I]: M-+ LhA4 such that ~oyl= y. In our topological work in [2], the map YZ appears naturally and plays a far more central role than the more intuitive map y. In fact, we were led there to say that A4 is Z-complete if q: M+ LiM is an isomorphism. With this sense of the term Z-complete, the following result shows that M; and all of the L:M are Z-complete; it also shows that Li N = 0 for p 3 1 when N is Z-complete.
THEOREM

4.1.

Assume the hypotheses of Theorem 2.5, so that H#f) 2 L,(M). is an

or L:M for some q 3 0. Then I: N -+ LhN Let N be either M; isomorphism and LL N = 0 for p 3 1.

Proof We agree to write L, for Li throughout the proof. It suffices to prove that r~: N + L,N is an isomorphism for the specified N and that L,L,M=O forpa 1 and any M. We can let Z= J= K in Lemma 2.7, using the same list of generators twice, and so obtain a spectral sequence {E} converging from L,L,M to L,M. In total degree zero, the spectral sequence collapses to an isomorphism L,Mz L,L,M. Writing down an explicit construction of 1 and using the proof of Theorem 2.5, we easily check that the isomorphism is in fact given by 9. Since Lo&: L,L,M -+ L,M,* is an epimorphism, it follows by a little diagram chase that 1: M; -+ L,M,^ is an epimorphism, and q is certainly a monomorphism since E0r] is the isomorphism y. We will prove at the end that r~: L,M + L,L,M is an isomorphism for q > 0. Suppose next that F is a free module. Then the E,-term of the spectral sequence above is zero unless q=O, when it is L,LOF= L,F,, while the limit term is zero except in degree 0. Thus L,L,F= 0 for p > 1. Given a general module M, construct a short exact sequence

where F is free. We first show that LI L,M = 0. Since L, F = 0, we have an exact sequence O-+L,M-+L,R-tL,F+L,M-kO.

452

GREENLEESAND MAY L,M

Let K be the kernel of L,F-+ short exact sequences

O+L,M+L,R+K+O L, + L, L, is an isomorphism

and break this sequence into the two and

O+K+L,F+L,M+O.

The first gives an epimorphism L,L,R + L, K, and the fact that q: implies that r]: K -+ L, K is an epimorphism. Using the second and the fact that L, L,F= 0, we obtain a commutative diagram with exact rows
O+ K + L,F + L,M -+O

Chasing the diagram, we see that n: K -+ L,K is an isomorphism, hence that L,K -+ L, L,F is a monomorphism, hence that L, L,M = 0. Since L, + , L,M z L, K z L, L, K, it follows inductively that L, L,M = 0 for all p > 1. Finally, q: L,M + L, L,M is an isomorphism for q = 1 since q: L,R + L,L,R and ?z: K + L,K are isomorphisms; it is an isomorphism for q 3 2 since, inductively, r~: L, ~, R -+ LOL, ~, R is an isomorphism and L y~, R is isomorphic to L, M.

5. THE RIGHT DERIVED FUNCTORS OF I-ADIC COMPLETION

Let I= (a,, .... LX,) and let Rj be the ith right derived functor of I-adic completion, These functors are much less interesting than the functors Lf. The main reason is the following observation, which surely must be known. For an A-module M, define T(Z, M), the annihilator of Z in M, to be {m 1Z.m=O}cM. Write Z(Z)=ZJZ,A).
LEMMA 5.1. For an injectioe A-module N, IN = T(f(Z), if A is an integral domain, then ZN = N. N); in particular,

Proof Clearly Z(ZJZ), N) = { n 1a. I= 0 implies a. n = 0} contains IN. The injectivity of N implies the reverse inclusion. To see this, note that T(Z)=T(cr,)n ... nf(cr,) is the kernel of the map A+(a,)@ ... @(a,) with coordinates m,. Thus we have inclusions

A/r(z) -, (aI) 0 ... @(a,)+A@

... @A.

We may identify T(T(Z), N) with Hom(A/Z(Z), N). By extending maps over (a,)@ ... @(a,) and then over A@ ... @A, we see that
T(I(Z), N)=xf(T(a,), N)=zaiN=ZN.

DERIVED FUNCTORS OF COMPLETION

453

Now assume that A has bounded cc,-torsion for all i. Using that f(Z) = f(q) n . . . nr(a,) and that Z is generated by the monomials of degree r in the tl,, we see that A has bounded Z-torsion. That is, there exists r such that T(r) = f(Z) for all sat-. We conclude from the lemma that N; = N/Z-(r(r), N) for injective A-modules N. For an arbitrary A-module M, the right derived A-modules RIM are computed by applying I-adic completion to an injective resolution of A4 and then taking homology. In particular, if A is an integral domain, then N, = 0 for any injective module N and we conclude that R>M = 0 for any A-module A4 and all i 2 0. Note that the functor RM= M/ZJZ-(r), M) of A4 preserves monomorphisms and epimorphisms but fails to be half exact in general. For a short exact sequence, O+M-+M-+M-+O, the middle homology group measuring the deviation from exactness is

(m 1al = 0 implies am E M }/M + ZJ r(T), M). Of course, when the functor R is exact, Ry = R and R, = 0 for i > 0.

ACKNOWLEDGMENTS It is a pleasure to thank Dick Swan for his very careful reading of an earlier draft, which uncovered several errors. We are also grateful to Gennady Lyubeznik and Bill Dwyer for helpful conversations.

REFERENCES 1. H. CARTAN AND S. EILENBERG,Homological Algebra, Princeton Univ. Press, Princeton, NJ, 1956. 2. J. P. C. GREENLEES AND J. P. MAY, Completions of G-spectra at ideals of the Burnside ring, Proceedings of the Adams Memorial Symposium, Lecture Notes in Mathematics, SpringerVerlag, New York/Berlin, to appear in 1992. 3. A. GROTHENDIECK,Sur quelques points dalgbbre homologique, T6hoku Math. J. 9 (1957), 119-221. 4. A. GROTHENDIECK,EGA III. etude cohomologique des faisceaux cohkrents, Publ. Math. IHES 11 (1961), 17 (1963). 5. A. GROTHENDIECK, Local Cohomology (notes by R. Hartshorne), Lecture Notes in Mathematics, Vol. 41, Springer-Verlag, New York/Berlin, 1967. 6. R. HARTSHORNE,Algebraic Geometry, Springer-Verlag, New York/Berlin, 1977. 7. J. E. RWS, Sur les foncteurs dtrivbs de lim. Applications. C. R. Acnd. Sci. Paris 252 (l961),
3702-3704.