Introduction to Numerical Analysis for
Engineers
 Fundamentals of Digital Computing
Digital Computer Models
Convergence, accuracy and stability
Number representation
Arithmetic operations
Recursion algorithms
Error propagation  numerical stability
Error estimation
Error cancellation
Condition numbers
 Error Analysis
13.002
Numerical Methods for Engineers
Lecture 2
Floating Number Representation
Examples
m
b
e
Mantissa
Base
Exponent
Decimal
Binary
Convention
Decimal
Binary
Max mantissa
Min mantissa
Max exponent
General
13.002
Min exponent
Numerical Methods for Engineers
Lecture 2
Error Analysis
Number Representation
Addition and Subtraction
Absolute Error
Shift mantissa of largest number
Relative Error
Result has exponent of largest number
Absolute Error
Relative Error
Unbounded
Multiplication and Division
Relative Error
13.002
Numerical Methods for Engineers
Bounded
Lecture 2
Error Propagation
Spherical Bessel Functions
Differential Equation
Solutions
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Spherical Bessel Functions
Forward Recurrence
Forward Recurrence
Unstable
Backward Recurrence
Millers algorithm
Stable
N ~ x+20
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Differential Equation
Eulers Method
Example
Discretization
Finite Difference (forward)
Recurrence
euler.m
Central Finite Difference
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Absolute Errors
'y~f (x)'x
Function of one variable
'x = x - x
General Error Propagation Formula
x
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Example
Error Propagation Formula
Multiplication
=>
=>
=>
=>
Relative Errors Add for Multiplication
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Expectation of Errors
Addition
Standard Error
Truncation
Error Expectation
Rounding
Standard Error better measure
of expected errors
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Error Cancellation
Function of one variable
Max. error
Error cancellation
Stand. error
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Condition Number
y = f(x)
x = x(1 + D)
y = y(1 + E)
Problem Condition Number
Problem ill-conditioned
Error cancellation example
Well-conditioned problem
13.002
Numerical Methods for Engineers
Lecture 2
Error Propagation
Condition Number
Problem Condition Number
K A is algorith condition number, which
may be much larger than the K P due
to limited number representation.
Solution
 Higher precision
 Rewrite algorithm
4 Significant Digits
Well-conditioned Algorithm
Algorithm Condition Number
13.002
Numerical Methods for Engineers
Lecture 2