This action might not be possible to undo. Are you sure you want to continue?
and Nonlinear Equations
C. T. Kelley
North Carolina State University
Society for Industrial and Applied Mathematics
Philadelphia 1995
Untitled1 9/20/2004, 2:59 PM 3
To Polly H. Thomas, 19061994, devoted mother and grandmother
1
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
2
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Contents
Preface xi
How to Get the Software xiii
CHAPTER 1. Basic Concepts and Stationary Iterative Methods 3
1.1 Review and notation . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 The Banach Lemma and approximate inverses . . . . . . . . . . 5
1.3 The spectral radius . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Matrix splittings and classical stationary iterative methods . . 7
1.5 Exercises on stationary iterative methods . . . . . . . . . . . . 10
CHAPTER 2. Conjugate Gradient Iteration 11
2.1 Krylov methods and the minimization property . . . . . . . . . 11
2.2 Consequences of the minimization property . . . . . . . . . . . 13
2.3 Termination of the iteration . . . . . . . . . . . . . . . . . . . . 15
2.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Preconditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 CGNR and CGNE . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.7 Examples for preconditioned conjugate iteration . . . . . . . . 26
2.8 Exercises on conjugate gradient . . . . . . . . . . . . . . . . . . 30
CHAPTER 3. GMRES Iteration 33
3.1 The minimization property and its consequences . . . . . . . . 33
3.2 Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Preconditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 GMRES implementation: Basic ideas . . . . . . . . . . . . . . . 37
3.5 Implementation: Givens rotations . . . . . . . . . . . . . . . . . 43
3.6 Other methods for nonsymmetric systems . . . . . . . . . . . . 46
3.6.1 BiCG. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6.2 CGS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6.3 BiCGSTAB. . . . . . . . . . . . . . . . . . . . . . . . . 50
vii
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
viii CONTENTS
3.6.4 TFQMR. . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.7 Examples for GMRES iteration . . . . . . . . . . . . . . . . . . 54
3.8 Examples for CGNR, BiCGSTAB, and TFQMR iteration . . . 55
3.9 Exercises on GMRES . . . . . . . . . . . . . . . . . . . . . . . . 60
CHAPTER 4. Basic Concepts and FixedPoint Iteration 65
4.1 Types of convergence . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Fixedpoint iteration . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 The standard assumptions . . . . . . . . . . . . . . . . . . . . . 68
CHAPTER 5. Newton’s Method 71
5.1 Local convergence of Newton’s method . . . . . . . . . . . . . . 71
5.2 Termination of the iteration . . . . . . . . . . . . . . . . . . . . 72
5.3 Implementation of Newton’s method . . . . . . . . . . . . . . . 73
5.4 Errors in the function and derivative . . . . . . . . . . . . . . . 75
5.4.1 The chord method. . . . . . . . . . . . . . . . . . . . . . 76
5.4.2 Approximate inversion of F
. . . . . . . . . . . . . . . . 77
5.4.3 The Shamanskii method. . . . . . . . . . . . . . . . . . 78
5.4.4 Diﬀerence approximation to F
. . . . . . . . . . . . . . . 79
5.4.5 The secant method. . . . . . . . . . . . . . . . . . . . . 82
5.5 The Kantorovich Theorem . . . . . . . . . . . . . . . . . . . . . 83
5.6 Examples for Newton’s method . . . . . . . . . . . . . . . . . . 86
5.7 Exercises on Newton’s method . . . . . . . . . . . . . . . . . . 91
CHAPTER 6. Inexact Newton Methods 95
6.1 The basic estimates . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.1 Direct analysis. . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.2 Weighted norm analysis. . . . . . . . . . . . . . . . . . . 97
6.1.3 Errors in the function. . . . . . . . . . . . . . . . . . . . 100
6.2 Newtoniterative methods . . . . . . . . . . . . . . . . . . . . . 100
6.2.1 Newton GMRES. . . . . . . . . . . . . . . . . . . . . . . 101
6.2.2 Other Newtoniterative methods. . . . . . . . . . . . . . 104
6.3 NewtonGMRES implementation . . . . . . . . . . . . . . . . . 104
6.4 Examples for NewtonGMRES . . . . . . . . . . . . . . . . . . 106
6.4.1 Chandrasekhar Hequation. . . . . . . . . . . . . . . . . 107
6.4.2 Convectiondiﬀusion equation. . . . . . . . . . . . . . . 108
6.5 Exercises on inexact Newton methods . . . . . . . . . . . . . . 110
CHAPTER 7. Broyden’s method 113
7.1 The Dennis–Mor´e condition . . . . . . . . . . . . . . . . . . . . 114
7.2 Convergence analysis . . . . . . . . . . . . . . . . . . . . . . . . 116
7.2.1 Linear problems. . . . . . . . . . . . . . . . . . . . . . . 118
7.2.2 Nonlinear problems. . . . . . . . . . . . . . . . . . . . . 120
7.3 Implementation of Broyden’s method . . . . . . . . . . . . . . . 123
7.4 Examples for Broyden’s method . . . . . . . . . . . . . . . . . . 127
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONTENTS ix
7.4.1 Linear problems. . . . . . . . . . . . . . . . . . . . . . . 127
7.4.2 Nonlinear problems. . . . . . . . . . . . . . . . . . . . . 128
7.5 Exercises on Broyden’s method . . . . . . . . . . . . . . . . . . 132
CHAPTER 8. Global Convergence 135
8.1 Single equations . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.2 Analysis of the Armijo rule . . . . . . . . . . . . . . . . . . . . 138
8.3 Implementation of the Armijo rule . . . . . . . . . . . . . . . . 141
8.3.1 Polynomial line searches. . . . . . . . . . . . . . . . . . 142
8.3.2 Broyden’s method. . . . . . . . . . . . . . . . . . . . . . 144
8.4 Examples for Newton–Armijo . . . . . . . . . . . . . . . . . . . 146
8.4.1 Inverse tangent function. . . . . . . . . . . . . . . . . . 146
8.4.2 Convectiondiﬀusion equation. . . . . . . . . . . . . . . 146
8.4.3 Broyden–Armijo. . . . . . . . . . . . . . . . . . . . . . . 148
8.5 Exercises on global convergence . . . . . . . . . . . . . . . . . . 151
Bibliography 153
Index 163
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
x CONTENTS
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Preface
This book on iterative methods for linear and nonlinear equations can be used
as a tutorial and a reference by anyone who needs to solve nonlinear systems
of equations or large linear systems. It may also be used as a textbook for
introductory courses in nonlinear equations or iterative methods or as source
material for an introductory course in numerical analysis at the graduate level.
We assume that the reader is familiar with elementary numerical analysis,
linear algebra, and the central ideas of direct methods for the numerical
solution of dense linear systems as described in standard texts such as [7],
[105], or [184].
Our approach is to focus on a small number of methods and treat them
in depth. Though this book is written in a ﬁnitedimensional setting, we
have selected for coverage mostly algorithms and methods of analysis which
extend directly to the inﬁnitedimensional case and whose convergence can be
thoroughly analyzed. For example, the matrixfree formulation and analysis for
GMRES and conjugate gradient is almost unchanged in an inﬁnitedimensional
setting. The analysis of Broyden’s method presented in Chapter 7 and
the implementations presented in Chapters 7 and 8 are diﬀerent from the
classical ones and also extend directly to an inﬁnitedimensional setting. The
computational examples and exercises focus on discretizations of inﬁnite
dimensional problems such as integral and diﬀerential equations.
We present a limited number of computational examples. These examples
are intended to provide results that can be used to validate the reader’s own
implementations and to give a sense of how the algorithms perform. The
examples are not designed to give a complete picture of performance or to be
a suite of test problems.
The computational examples in this book were done with MATLAB
(version 4.0a on various SUN SPARCstations and version 4.1 on an Apple
Macintosh Powerbook 180) and the MATLAB environment is an excellent one
for getting experience with the algorithms, for doing the exercises, and for
smalltomedium scale production work.
1
MATLAB codes for many of the
algorithms are available by anonymous ftp. A good introduction to the latest
1
MATLAB is a registered trademark of The MathWorks, Inc.
xi
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
xii PREFACE
version (version 4.2) of MATLAB is the MATLAB Primer [178]; [43] is also
a useful resource. If the reader has no access to MATLAB or will be solving
very large problems, the general algorithmic descriptions or even the MATLAB
codes can easily be translated to another language.
Parts of this book are based upon work supported by the National
Science Foundation and the Air Force Oﬃce of Scientiﬁc Research over
several years, most recently under National Science Foundation Grant Nos.
DMS9024622 and DMS9321938. Any opinions, ﬁndings, and conclusions or
recommendations expressed in this material are those of the author and do not
necessarily reﬂect the views of the National Science Foundation or of the Air
Force Oﬃce of Scientiﬁc Research.
Many of my students and colleagues discussed various aspects of this
project with me and provided important corrections, ideas, suggestions, and
pointers to the literature. I am especially indebted to Jim Banoczi, Jeﬀ Butera,
Steve Campbell, Tony Choi, Moody Chu, Howard Elman, Jim Epperson,
Andreas Griewank, Laura Helfrich, Ilse Ipsen, Lea Jenkins, Vickie Kearn,
Belinda King, Debbie Lockhart, Carl Meyer, Casey Miller, Ekkehard Sachs,
Jeﬀ Scroggs, Joseph Skudlarek, Mike Tocci, Gordon Wade, Homer Walker,
Steve Wright, Zhaqing Xue, Yue Zhang, and an anonymous reviewer for their
contributions and encouragement.
Most importantly, I thank ChungWei Ng and my parents for over one
hundred and ten years of patience and support.
C. T. Kelley
Raleigh, North Carolina
January, 1998
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
How to get the software
A collection of MATLAB codes has been written to accompany this book. The
MATLAB codes can be obtained by anonymous ftp from the MathWorks server
ftp.mathworks.com in the directory pub/books/kelley, from the MathWorks
World Wide Web site,
http://www.mathworks.com
or from SIAM’s World Wide Web site
http://www.siam.org/books/kelley/kelley.html
One can obtain MATLAB from
The MathWorks, Inc.
24 Prime Park Way
Natick, MA 01760,
Phone: (508) 6531415
Fax: (508) 6532997
Email: info@mathworks.com
WWW: http://www.mathworks.com
xiii
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 1
Basic Concepts and Stationary Iterative Methods
1.1. Review and notation
We begin by setting notation and reviewing some ideas from numerical linear
algebra that we expect the reader to be familiar with. An excellent reference
for the basic ideas of numerical linear algebra and direct methods for linear
equations is [184].
We will write linear equations as
Ax = b, (1.1)
where A is a nonsingular N N matrix, b ∈ R
N
is given, and
x
∗
= A
−1
b ∈ R
N
is to be found.
Throughout this chapter x will denote a potential solution and ¦x
k
¦
k≥0
the
sequence of iterates. We will denote the ith component of a vector x by (x)
i
(note the parentheses) and the ith component of x
k
by (x
k
)
i
. We will rarely
need to refer to individual components of vectors.
In this chapter   will denote a norm on R
N
as well as the induced matrix
norm.
Definition 1.1.1. Let   be a norm on R
N
. The induced matrix norm
of an N N matrix A is deﬁned by
A = max
x=1
Ax.
Induced norms have the important property that
Ax ≤ Ax.
Recall that the condition number of A relative to the norm   is
κ(A) = AA
−1
,
where κ(A) is understood to be inﬁnite if A is singular. If   is the l
p
norm
x
p
=
_
_
N
j=1
[(x)
i
[
p
_
_
1/p
3
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
4 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
we will write the condition number as κ
p
.
Most iterative methods terminate when the residual
r = b −Ax
is suﬃciently small. One termination criterion is
r
k

r
0

< τ, (1.2)
which can be related to the error
e = x −x
∗
in terms of the condition number.
Lemma 1.1.1. Let b, x, x
0
∈ R
N
. Let A be nonsingular and let x
∗
= A
−1
b.
e
e
0

≤ κ(A)
r
r
0

. (1.3)
Proof. Since
r = b −Ax = −Ae
we have
e = A
−1
Ae ≤ A
−1
Ae = A
−1
r
and
r
0
 = Ae
0
 ≤ Ae
0
.
Hence
e
e
0

≤
A
−1
r
A
−1
r
0

= κ(A)
r
r
0

,
as asserted.
The termination criterion (1.2) depends on the initial iterate and may result
in unnecessary work when the initial iterate is good and a poor result when the
initial iterate is far from the solution. For this reason we prefer to terminate
the iteration when
r
k

b
< τ. (1.4)
The two conditions (1.2) and (1.4) are the same when x
0
= 0, which is a
common choice, particularly when the linear iteration is being used as part of
a nonlinear solver.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BASIC CONCEPTS 5
1.2. The Banach Lemma and approximate inverses
The most straightforward approach to an iterative solution of a linear system
is to rewrite (1.1) as a linear ﬁxedpoint iteration. One way to do this is to
write Ax = b as
x = (I −A)x +b, (1.5)
and to deﬁne the Richardson iteration
x
k+1
= (I −A)x
k
+b. (1.6)
We will discuss more general methods in which ¦x
k
¦ is given by
x
k+1
= Mx
k
+c. (1.7)
In (1.7) M is an NN matrix called the iteration matrix. Iterative methods of
this form are called stationary iterative methods because the transition from x
k
to x
k+1
does not depend on the history of the iteration. The Krylov methods
discussed in Chapters 2 and 3 are not stationary iterative methods.
All our results are based on the following lemma.
Lemma 1.2.1. If M is an N N matrix with M < 1 then I − M is
nonsingular and
(I −M)
−1
 ≤
1
1 −M
. (1.8)
Proof. We will show that I − M is nonsingular and that (1.8) holds by
showing that the series
∞
l=0
M
l
= (I −M)
−1
.
The partial sums
S
k
=
k
l=0
M
l
form a Cauchy sequence in R
N×N
. To see this note that for all m > k
S
k
−S
m
 ≤
m
l=k+1
M
l
.
Now, M
l
 ≤ M
l
because   is a matrix norm that is induced by a vector
norm. Hence
S
k
−S
m
 ≤
m
l=k+1
M
l
= M
k+1
_
1 −M
m−k
1 −M
_
→ 0
as m, k → ∞. Hence the sequence S
k
converges, say to S. Since MS
k
+ I =
S
k+1
, we must have MS +I = S and hence (I −M)S = I. This proves that
I −M is nonsingular and that S = (I −M)
−1
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
6 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Noting that
(I −M)
−1
 ≤
∞
l=0
M
l
= (1 −M)
−1
.
proves (1.8) and completes the proof.
The following corollary is a direct consequence of Lemma 1.2.1.
Corollary 1.2.1. If M < 1 then the iteration (1.7) converges to
x = (I −M)
−1
c for all initial iterates x
0
.
A consequence of Corollary 1.2.1 is that Richardson iteration (1.6) will
converge if I − A < 1. It is sometimes possible to precondition a linear
equation by multiplying both sides of (1.1) by a matrix B
BAx = Bb
so that convergence of iterative methods is improved. In the context of
Richardson iteration, the matrices B that allow us to apply the Banach lemma
and its corollary are called approximate inverses.
Definition 1.2.1. B is an approximate inverse of A if I −BA < 1.
The following theorem is often referred to as the Banach Lemma.
Theorem 1.2.1. If A and B are NN matrices and B is an approximate
inverse of A, then A and B are both nonsingular and
A
−1
 ≤
B
1 −I −BA
, B
−1
 ≤
A
1 −I −BA
, (1.9)
and
A
−1
−B ≤
BI −BA
1 −I −BA
, A−B
−1
 ≤
AI −BA
1 −I −BA
. (1.10)
Proof. Let M = I −BA. By Lemma 1.2.1 I −M = I −(I −BA) = BA is
nonsingular. Hence both A and B are nonsingular. By (1.8)
A
−1
B
−1
 = (I −M)
−1
 ≤
1
1 −M
=
1
1 −I −BA
. (1.11)
Since A
−1
= (I −M)
−1
B, inequality (1.11) implies the ﬁrst part of (1.9). The
second part follows in a similar way from B
−1
= A(I −M)
−1
.
To complete the proof note that
A
−1
−B = (I −BA)A
−1
, A−B
−1
= B
−1
(I −BA),
and use (1.9).
Richardson iteration, preconditioned with approximate inversion, has the
form
x
k+1
= (I −BA)x
k
+Bb. (1.12)
If the norm of I − BA is small, then not only will the iteration converge
rapidly, but, as Lemma 1.1.1 indicates, termination decisions based on the
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BASIC CONCEPTS 7
preconditioned residual Bb − BAx will better reﬂect the actual error. This
method is a very eﬀective technique for solving diﬀerential equations, integral
equations, and related problems [15], [6], [100], [117], [111]. Multigrid methods
[19], [99], [126], can also be interpreted in this light. We mention one other
approach, polynomial preconditioning, which tries to approximate A
−1
by a
polynomial in A [123], [179], [169].
1.3. The spectral radius
The analysis in ¸ 1.2 related convergence of the iteration (1.7) to the norm of
the matrix M. However the norm of M could be small in some norms and
quite large in others. Hence the performance of the iteration is not completely
described by M. The concept of spectral radius allows us to make a complete
description.
We let σ(A) denote the set of eigenvalues of A.
Definition 1.3.1. The spectral radius of an N N matrix A is
ρ(A) = max
λ∈σ(A)
[λ[ = lim
n→∞
A
n

1/n
. (1.13)
The term on the righthand side of the second equality in (1.13) is the limit
used by the radical test for convergence of the series
A
n
.
The spectral radius of M is independent of any particular matrix norm of
M. It is clear, in fact, that
ρ(A) ≤ A (1.14)
for any induced matrix norm. The inequality (1.14) has a partial converse that
allows us to completely describe the performance of iteration (1.7) in terms of
spectral radius. We state that converse as a theorem and refer to [105] for a
proof.
Theorem 1.3.1. Let A be an N N matrix. Then for any > 0 there is
a norm   on R
N
such that
ρ(A) > A −.
A consequence of Theorem 1.3.1, Lemma 1.2.1, and Exercise 1.5.1 is a
characterization of convergent stationary iterative methods. The proof is left
as an exercise.
Theorem 1.3.2. Let M be an NN matrix. The iteration (1.7) converges
for all c ∈ R
N
if and only if ρ(M) < 1.
1.4. Matrix splittings and classical stationary iterative methods
There are ways to convert Ax = b to a linear ﬁxedpoint iteration that are
diﬀerent from (1.5). Methods such as Jacobi, Gauss–Seidel, and sucessive
overrelaxation (SOR) iteration are based on splittings of A of the form
A = A
1
+A
2
,
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
8 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
where A
1
is a nonsingular matrix constructed so that equations with A
1
as
coeﬃcient matrix are easy to solve. Then Ax = b is converted to the ﬁxed
point problem
x = A
−1
1
(b −A
2
x).
The analysis of the method is based on an estimation of the spectral radius of
the iteration matrix M = −A
−1
1
A
2
.
For a detailed description of the classical stationary iterative methods the
reader may consult [89], [105], [144], [193], or [200]. These methods are usually
less eﬃcient than the Krylov methods discussed in Chapters 2 and 3 or the more
modern stationary methods based on multigrid ideas. However the classical
methods have a role as preconditioners. The limited description in this section
is intended as a review that will set some notation to be used later.
As a ﬁrst example we consider the Jacobi iteration that uses the splitting
A
1
= D, A
2
= L +U,
where D is the diagonal of A and L and U are the (strict) lower and upper
triangular parts. This leads to the iteration matrix
M
JAC
= −D
−1
(L +U).
Letting (x
k
)
i
denote the ith component of the kth iterate we can express Jacobi
iteration concretely as
(x
k+1
)
i
= a
−1
ii
_
_
b
i
−
j=i
a
ij
(x
k
)
j
_
_
. (1.15)
Note that A
1
is diagonal and hence trivial to invert.
We present only one convergence result for the classical stationary iterative
methods.
Theorem 1.4.1. Let A be an N N matrix and assume that for all
1 ≤ i ≤ N
0 <
j=i
[a
ij
[ < [a
ii
[. (1.16)
Then A is nonsingular and the Jacobi iteration (1.15) converges to x
∗
= A
−1
b
for all b.
Proof. Note that the ith row sum of M = M
JAC
satisﬁes
N
j=1
[m
ij
[ =
j=i
[a
ij
[
[a
ii
[
< 1.
Hence M
JAC

∞
< 1 and the iteration converges to the unique solution of
x = Mx + D
−1
b. Also I − M = D
−1
A is nonsingular and therefore A is
nonsingular.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BASIC CONCEPTS 9
Gauss–Seidel iteration overwrites the approximate solution with the new
value as soon as it is computed. This results in the iteration
(x
k+1
)
i
= a
−1
ii
_
_
b
i
−
j<i
a
ij
(x
k+1
)
j
−
j>i
a
ij
(x
k
)
j
_
_
,
the splitting
A
1
= D +L, A
2
= U,
and iteration matrix
M
GS
= −(D +L)
−1
U.
Note that A
1
is lower triangular, and hence A
−1
1
y is easy to compute for vectors
y. Note also that, unlike Jacobi iteration, the iteration depends on the ordering
of the unknowns. Backward Gauss–Seidel begins the update of x with the Nth
coordinate rather than the ﬁrst, resulting in the splitting
A
1
= D +U, A
2
= L,
and iteration matrix
M
BGS
= −(D +U)
−1
L.
A symmetric Gauss–Seidel iteration is a forward Gauss–Seidel iteration
followed by a backward Gauss–Seidel iteration. This leads to the iteration
matrix
M
SGS
= M
BGS
M
GS
= (D +U)
−1
L(D +L)
−1
U.
If A is symmetric then U = L
T
. In that event
M
SGS
= (D +U)
−1
L(D +L)
−1
U = (D +L
T
)
−1
L(D +L)
−1
L
T
.
From the point of view of preconditioning, one wants to write the stationary
method as a preconditioned Richardson iteration. That means that one wants
to ﬁnd B such that M = I − BA and then use B as an approximate inverse.
For the Jacobi iteration,
B
JAC
= D
−1
. (1.17)
For symmetric Gauss–Seidel
B
SGS
= (D +L
T
)
−1
D(D +L)
−1
. (1.18)
The successive overrelaxation iteration modiﬁes Gauss–Seidel by adding a
relaxation parameter ω to construct an iteration with iteration matrix
M
SOR
= (D +ωL)
−1
((1 −ω)D −ωU).
The performance can be dramatically improved with a good choice of ω but
still is not competitive with Krylov methods. A further disadvantage is that
the choice of ω is often diﬃcult to make. References [200], [89], [193], [8], and
the papers cited therein provide additional reading on this topic.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
10 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
1.5. Exercises on stationary iterative methods
1.5.1. Show that if ρ(M) ≥ 1 then there are x
0
and c such that the iteration
(1.7) fails to converge.
1.5.2. Prove Theorem 1.3.2.
1.5.3. Verify equality (1.18).
1.5.4. Show that if A is symmetric and positive deﬁnite (that is A
T
= A and
x
T
Ax > 0 for all x ,= 0) that B
SGS
is also symmetric and positive
deﬁnite.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 2
Conjugate Gradient Iteration
2.1. Krylov methods and the minimization property
In the following two chapters we describe some of the Krylov space methods
for linear equations. Unlike the stationary iterative methods, Krylov methods
do not have an iteration matrix. The two such methods that we’ll discuss in
depth, conjugate gradient and GMRES, minimize, at the kth iteration, some
measure of error over the aﬃne space
x
0
+/
k
,
where x
0
is the initial iterate and the kth Krylov subspace /
k
is
/
k
= span(r
0
, Ar
0
, . . . , A
k−1
r
0
)
for k ≥ 1.
The residual is
r = b −Ax.
So ¦r
k
¦
k≥0
will denote the sequence of residuals
r
k
= b −Ax
k
.
As in Chapter 1, we assume that A is a nonsingular N N matrix and let
x
∗
= A
−1
b.
There are other Krylov methods that are not as well understood as CG or
GMRES. Brief descriptions of several of these methods and their properties
are in ¸ 3.6, [12], and [78].
The conjugate gradient (CG) iteration was invented in the 1950s [103] as a
direct method. It has come into wide use over the last 15 years as an iterative
method and has generally superseded the Jacobi–Gauss–Seidel–SOR family of
methods.
CG is intended to solve symmetric positive deﬁnite (spd) systems. Recall
that A is symmetric if A = A
T
and positive deﬁnite if
x
T
Ax > 0 for all x ,= 0.
11
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
12 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
In this section we assume that A is spd. Since A is spd we may deﬁne a norm
(you should check that this is a norm) by
x
A
=
√
x
T
Ax. (2.1)
 
A
is called the Anorm. The development in these notes is diﬀerent from
the classical work and more like the analysis for GMRES and CGNR in [134].
In this section, and in the section on GMRES that follows, we begin with a
description of what the algorithm does and the consequences of the minimiza
tion property of the iterates. After that we describe termination criterion,
performance, preconditioning, and at the very end, the implementation.
The kth iterate x
k
of CG minimizes
φ(x) =
1
2
x
T
Ax −x
T
b (2.2)
over x
0
+/
k
.
Note that if φ(˜ x) is the minimal value (in R
N
) then
∇φ(˜ x) = A˜ x −b = 0
and hence ˜ x = x
∗
.
Minimizing φ over any subset of R
N
is the same as minimizing x −x
∗

A
over that subset. We state this as a lemma.
Lemma 2.1.1. Let o ⊂ R
N
. If x
k
minimizes φ over o then x
k
also
minimizes x
∗
−x
A
= r
A
−1 over o.
Proof. Note that
x −x
∗

2
A
= (x −x
∗
)
T
A(x −x
∗
) = x
T
Ax −x
T
Ax
∗
−(x
∗
)
T
Ax + (x
∗
)
T
Ax
∗
.
Since A is symmetric and Ax
∗
= b
−x
T
Ax
∗
−(x
∗
)
T
Ax = −2x
T
Ax
∗
= −2x
T
b.
Therefore
x −x
∗

2
A
= 2φ(x) + (x
∗
)
T
Ax
∗
.
Since (x
∗
)
T
Ax
∗
is independent of x, minimizing φ is equivalent to minimizing
x −x
∗

2
A
and hence to minimizing x −x
∗

A
.
If e = x −x
∗
then
e
2
A
= e
T
Ae = (A(x −x
∗
))
T
A
−1
(A(x −x
∗
)) = b −Ax
2
A
−1
and hence the Anorm of the error is also the A
−1
norm of the residual.
We will use this lemma in the particular case that o = x
0
+/
k
for some k.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 13
2.2. Consequences of the minimization property
Lemma 2.1.1 implies that since x
k
minimizes φ over x
0
+/
k
x
∗
−x
k

A
≤ x
∗
−w
A
(2.3)
for all w ∈ x
0
+/
k
. Since any w ∈ x
0
+/
k
can be written as
w =
k−1
j=0
γ
j
A
j
r
0
+x
0
for some coeﬃcients ¦γ
j
¦, we can express x
∗
−w as
x
∗
−w = x
∗
−x
0
−
k−1
j=0
γ
j
A
j
r
0
.
Since Ax
∗
= b we have
r
0
= b −Ax
0
= A(x
∗
−x
0
)
and therefore
x
∗
−w = x
∗
−x
0
−
k−1
j=0
γ
j
A
j+1
(x
∗
−x
0
) = p(A)(x
∗
−x
0
),
where the polynomial
p(z) = 1 −
k−1
j=0
γ
j
z
j+1
has degree k and satisﬁes p(0) = 1. Hence
x
∗
−x
k

A
= min
p∈P
k
,p(0)=1
p(A)(x
∗
−x
0
)
A
. (2.4)
In (2.4) P
k
denotes the set of polynomials of degree k.
The spectral theorem for spd matrices asserts that
A = UΛU
T
,
where U is an orthogonal matrix whose columns are the eigenvectors of A and
Λ is a diagonal matrix with the positive eigenvalues of A on the diagonal. Since
UU
T
= U
T
U = I by orthogonality of U, we have
A
j
= UΛ
j
U
T
.
Hence
p(A) = Up(Λ)U
T
.
Deﬁne A
1/2
= UΛ
1/2
U
T
and note that
x
2
A
= x
T
Ax = A
1/2
x
2
2
. (2.5)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
14 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Hence, for any x ∈ R
N
and
p(A)x
A
= A
1/2
p(A)x
2
≤ p(A)
2
A
1/2
x
2
≤ p(A)
2
x
A
.
This, together with (2.4) implies that
x
k
−x
∗

A
≤ x
0
−x
∗

A
min
p∈P
k
,p(0)=1
max
z∈σ(A)
[p(z)[. (2.6)
Here σ(A) is the set of all eigenvalues of A.
The following corollary is an important consequence of (2.6).
Corollary 2.2.1. Let A be spd and let ¦x
k
¦ be the CG iterates. Let k be
given and let ¦¯ p
k
¦ be any kth degree polynomial such that ¯ p
k
(0) = 1. Then
x
k
−x
∗

A
x
0
−x
∗

A
≤ max
z∈σ(A)
[ ¯ p
k
(z)[. (2.7)
We will refer to the polynomial ¯ p
k
as a residual polynomial [185].
Definition 2.2.1. The set of kth degree residual polynomials is
T
k
= ¦p [ p is a polynomial of degree k and p(0) = 1.¦ (2.8)
In speciﬁc contexts we try to construct sequences of residual polynomials,
based on information on σ(A), that make either the middle or the right term
in (2.7) easy to evaluate. This leads to an upper estimate for the number of
CG iterations required to reduce the Anorm of the error to a given tolerance.
One simple application of (2.7) is to show how the CG algorithm can be
viewed as a direct method.
Theorem 2.2.1. Let A be spd. Then the CG algorithm will ﬁnd the
solution within N iterations.
Proof. Let ¦λ
i
¦
N
i=1
be the eigenvalues of A. As a test polynomial, let
¯ p(z) =
N
i=1
(λ
i
−z)/λ
i
.
¯ p ∈ T
N
because ¯ p has degree N and ¯ p(0) = 1. Hence, by (2.7) and the fact
that ¯ p vanishes on σ(A),
x
N
−x
∗

A
≤ x
0
−x
∗

A
max
z∈σ(A)
[ ¯ p(z)[ = 0.
Note that our test polynomial had the eigenvalues of A as its roots. In
that way we showed (in the absence of all roundoﬀ error!) that CG terminated
in ﬁnitely many iterations with the exact solution. This is not as good as it
sounds, since in most applications the number of unknowns N is very large,
and one cannot aﬀord to perform N iterations. It is best to regard CG as an
iterative method. When doing that we seek to terminate the iteration when
some speciﬁed error tolerance is reached.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 15
In the two examples that follow we look at some other easy consequences
of (2.7).
Theorem 2.2.2. Let A be spd with eigenvectors ¦u
i
¦
N
i=1
. Let b be a linear
combination of k of the eigenvectors of A
b =
k
l=1
γ
l
u
i
l
.
Then the CG iteration for Ax = b with x
0
= 0 will terminate in at most k
iterations.
Proof. Let ¦λ
i
l
¦ be the eigenvalues of A associated with the eigenvectors
¦u
i
l
¦
k
l=1
. By the spectral theorem
x
∗
=
k
l=1
(γ
l
/λ
i
l
)u
i
l
.
We use the residual polynomial,
¯ p(z) =
k
l=1
(λ
i
l
−z)/λ
i
l
.
One can easily verify that ¯ p ∈ T
k
. Moreover, ¯ p(λ
i
l
) = 0 for 1 ≤ l ≤ k and
hence
¯ p(A)x
∗
=
k
l=1
¯ p(λ
i
l
)γ
l
/λ
i
l
u
i
l
= 0.
So, we have by (2.4) and the fact that x
0
= 0 that
x
k
−x
∗

A
≤ ¯ p(A)x
∗

A
= 0.
This completes the proof.
If the spectrum of A has fewer than N points, we can use a similar technique
to prove the following theorem.
Theorem 2.2.3. Let A be spd. Assume that there are exactly k ≤ N
distinct eigenvalues of A. Then the CG iteration terminates in at most k
iterations.
2.3. Termination of the iteration
In practice we do not run the CG iteration until an exact solution is found, but
rather terminate once some criterion has been satisﬁed. One typical criterion is
small (say ≤ η) relative residuals. This means that we terminate the iteration
after
b −Ax
k

2
≤ ηb
2
. (2.9)
The error estimates that come from the minimization property, however, are
based on (2.7) and therefore estimate the reduction in the relative Anorm of
the error.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
16 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Our next task is to relate the relative residual in the Euclidean norm to
the relative error in the Anorm. We will do this in the next two lemmas and
then illustrate the point with an example.
Lemma 2.3.1. Let A be spd with eigenvalues λ
1
≥ λ
2
≥ . . . λ
N
. Then for
all z ∈ R
N
,
A
1/2
z
2
= z
A
(2.10)
and
λ
1/2
N
z
A
≤ Az
2
≤ λ
1/2
1
z
A
. (2.11)
Proof. Clearly
z
2
A
= z
T
Az = (A
1/2
z)
T
(A
1/2
z) = A
1/2
z
2
2
which proves (2.10).
Let u
i
be a unit eigenvector corresponding to λ
i
. We may write A = UΛU
T
as
Az =
N
i=1
λ
i
(u
T
i
z)u
i
.
Hence
λ
N
A
1/2
z
2
2
= λ
N
N
i=1
λ
i
(u
T
i
z)
2
≤ Az
2
2
=
N
i=1
λ
2
i
(u
T
i
z)
2
≤ λ
1
N
i=1
λ
i
(u
T
i
z)
2
= λ
1
A
1/2
z
2
2
.
Taking square roots and using (2.10) complete the proof.
Lemma 2.3.2.
b
2
r
0

2
b −Ax
k

2
b
2
=
b −Ax
k

2
b −Ax
0

2
≤
_
κ
2
(A)
x
k
−x
∗

A
x
∗
−x
0

A
(2.12)
and
b −Ax
k

2
b
2
≤
_
κ
2
(A)r
0

2
b
2
x
k
−x
∗

A
x
∗
−x
0

A
. (2.13)
Proof. The equality on the left of (2.12) is clear and (2.13) follows directly
from (2.12). To obtain the inequality on the right of (2.12), ﬁrst recall that if
A = UΛU
T
is the spectral decomposition of A and we order the eigenvalues
such that λ
1
≥ λ
2
≥ . . . λ
N
> 0, then A
2
= λ
1
and A
−1

2
= 1/λ
N
. So
κ
2
(A) = λ
1
/λ
N
.
Therefore, using (2.10) and (2.11) twice,
b −Ax
k

2
b −Ax
0

2
=
A(x
∗
−x
k
)
2
A(x
∗
−x
0
)
2
≤
¸
λ
1
λ
N
x
∗
−x
k

A
x
∗
−x
0

A
as asserted.
So, to predict the performance of the CG iteration based on termination on
small relative residuals, we must not only use (2.7) to predict when the relative
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 17
Anorm error is small, but also use Lemma 2.3.2 to relate small Anorm errors
to small relative residuals.
We consider a very simple example. Assume that x
0
= 0 and that the
eigenvalues of A are contained in the interval (9, 11). If we let ¯ p
k
(z) =
(10 −z)
k
/10
k
, then ¯ p
k
∈ T
k
. This means that we may apply (2.7) to get
x
k
−x
∗

A
≤ x
∗

A
max
9≤z≤11
[ ¯ p
k
(z)[.
It is easy to see that
max
9≤z≤11
[ ¯ p
k
(z)[ = 10
−k
.
Hence, after k iterations
x
k
−x
∗

A
≤ x
∗

A
10
−k
. (2.14)
So, the size of the Anorm of the error will be reduced by a factor of 10
−3
when
10
−k
≤ 10
−3
,
that is, when
k ≥ 3.
To use Lemma 2.3.2 we simply note that κ
2
(A) ≤ 11/9. Hence, after k
iterations we have
Ax
k
−b
2
b
2
≤
√
11 10
−k
/3.
So, the size of the relative residual will be reduced by a factor of 10
−3
when
10
−k
≤ 3 10
−3
/
√
11,
that is, when
k ≥ 4.
One can obtain a more precise estimate by using a polynomial other than
p
k
in the upper estimate for the righthand side of (2.7). Note that it is always
the case that the spectrum of a spd matrix is contained in the interval [λ
N
, λ
1
]
and that κ
2
(A) = λ
1
/λ
N
. A result from [48] (see also [45]) that is, in one
sense, the sharpest possible, is
x
k
−x
∗

A
≤ 2x
0
−x
∗

A
_
_
κ
2
(A) −1
_
κ
2
(A) + 1
_
k
. (2.15)
In the case of the above example, we can estimate κ
2
(A) by κ
2
(A) ≤ 11/9.
Hence, since (
√
x −1)/(
√
x + 1) is an increasing function of x on the interval
(1, ∞).
_
κ
2
(A) −1
_
κ
2
(A) + 1
≤
√
11 −3
√
11 + 3
≈ .05.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
18 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Therefore (2.15) would predict a reduction in the size of the Anorm error by
a factor of 10
−3
when
2 .05
k
< 10
−3
or when
k > −log
10
(2000)/ log
10
(.05) ≈ 3.3/1.3 ≈ 2.6,
which also predicts termination within three iterations.
We may have more precise information than a single interval containing
σ(A). When we do, the estimate in (2.15) can be very pessimistic. If the
eigenvalues cluster in a small number of intervals, the condition number can
be quite large, but CG can perform very well. We will illustrate this with an
example. Exercise 2.8.5 also covers this point.
Assume that x
0
= 0 and the eigenvalues of A lie in the two intervals (1, 1.5)
and (399, 400). Based on this information the best estimate of the condition
number of A is κ
2
(A) ≤ 400, which, when inserted into (2.15) gives
x
k
−x
∗

A
x
∗

A
≤ 2 (19/21)
k
≈ 2 (.91)
k
.
This would indicate fairly slow convergence. However, if we use as a residual
polynomial ¯ p
3k
∈ T
3k
¯ p
3k
(z) =
(1.25 −z)
k
(400 −z)
2k
(1.25)
k
400
2k
.
It is easy to see that
max
z∈σ(A)
[ ¯ p
3k
(z)[ ≤ (.25/1.25)
k
= (.2)
k
,
which is a sharper estimate on convergence. In fact, (2.15) would predict that
x
k
−x
∗

A
≤ 10
−3
x
∗

A
,
when 2 (.91)
k
< 10
−3
or when
k > −log
10
(2000)/ log
10
(.91) ≈ 3.3/.04 = 82.5.
The estimate based on the clustering gives convergence in 3k iterations when
(.2)
k
≤ 10
−3
or when
k > −3/ log
10
(.2) = 4.3.
Hence (2.15) predicts 83 iterations and the clustering analysis 15 (the smallest
integer multiple of 3 larger than 3 4.3 = 12.9).
From the results above one can see that if the condition number of A is near
one, the CG iteration will converge very rapidly. Even if the condition number
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 19
is large, the iteration will perform well if the eigenvalues are clustered in a few
small intervals. The transformation of the problem into one with eigenvalues
clustered near one (i.e., easier to solve) is called preconditioning. We used
this term before in the context of Richardson iteration and accomplished the
goal by multiplying A by an approximate inverse. In the context of CG, such
a simple approach can destroy the symmetry of the coeﬃcient matrix and a
more subtle implementation is required. We discuss this in ¸ 2.5.
2.4. Implementation
The implementation of CG depends on the amazing fact that once x
k
has been
determined, either x
k
= x
∗
or a search direction p
k+1
,= 0 can be found very
cheaply so that x
k+1
= x
k
+ α
k+1
p
k+1
for some scalar α
k+1
. Once p
k+1
has
been found, α
k+1
is easy to compute from the minimization property of the
iteration. In fact
dφ(x
k
+αp
k+1
)
dα
= 0 (2.16)
for the correct choice of α = α
k+1
. Equation (2.16) can be written as
p
T
k+1
Ax
k
+αp
T
k+1
Ap
k+1
−p
T
k+1
b = 0
leading to
α
k+1
=
p
T
k+1
(b −Ax
k
)
p
T
k+1
Ap
k+1
=
p
T
k+1
r
k
p
T
k+1
Ap
k+1
. (2.17)
If x
k
= x
k+1
then the above analysis implies that α = 0. We show that
this only happens if x
k
is the solution.
Lemma 2.4.1. Let A be spd and let ¦x
k
¦ be the conjugate gradient iterates.
Then
r
T
k
r
l
= 0 for all 0 ≤ l < k. (2.18)
Proof. Since x
k
minimizes φ on x
0
+/
k
, we have, for any ξ ∈ /
k
dφ(x
k
+tξ)
dt
= ∇φ(x
k
+tξ)
T
ξ = 0
at t = 0. Recalling that
∇φ(x) = Ax −b = −r
we have
∇φ(x
k
)
T
ξ = −r
T
k
ξ = 0 for all ξ ∈ /
k
. (2.19)
Since r
l
∈ /
k
for all l < k (see Exercise 2.8.1), this proves (2.18).
Now, if x
k
= x
k+1
, then r
k
= r
k+1
. Lemma 2.4.1 then implies that
r
k

2
2
= r
T
k
r
k
= r
T
k
r
k+1
= 0 and hence x
k
= x
∗
.
The next lemma characterizes the search direction and, as a side eﬀect,
proves that (if we deﬁne p
0
= 0) p
T
l
r
k
= 0 for all 0 ≤ l < k ≤ n, unless the
iteration terminates prematurely.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
20 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Lemma 2.4.2. Let A be spd and let ¦x
k
¦ be the conjugate gradient iterates.
If x
k
,= x
∗
then x
k+1
= x
k
+ α
k+1
p
k+1
and p
k+1
is determined up to a scalar
multiple by the conditions
p
k+1
∈ /
k+1
, p
T
k+1
Aξ = 0 for all ξ ∈ /
k
. (2.20)
Proof. Since /
k
⊂ /
k+1
,
∇φ(x
k+1
)
T
ξ = (Ax
k
+α
k+1
Ap
k+1
−b)
T
ξ = 0 (2.21)
for all ξ ∈ /
k
. (2.19) and (2.21) then imply that for all ξ ∈ /
k
,
α
k+1
p
T
k+1
Aξ = −(Ax
k
−b)
T
ξ = −∇φ(x
k
)
T
ξ = 0. (2.22)
This uniquely speciﬁes the direction of p
k+1
as (2.22) implies that p
k+1
∈ /
k+1
is Aorthogonal (i.e., in the scalar product (x, y) = x
T
Ay) to /
k
, a subspace
of dimension one less than /
k+1
.
The condition p
T
k+1
Aξ = 0 is called Aconjugacy of p
k+1
to /
k
. Now, any
p
k+1
satisfying (2.20) can, up to a scalar multiple, be expressed as
p
k+1
= r
k
+w
k
with w
k
∈ /
k
. While one might think that w
k
would be hard to compute, it
is, in fact, trivial. We have the following theorem.
Theorem 2.4.1. Let A be spd and assume that r
k
,= 0. Deﬁne p
0
= 0.
Then
p
k+1
= r
k
+β
k+1
p
k
for some β
k+1
and k ≥ 0. (2.23)
Proof. By Lemma 2.4.2 and the fact that /
k
= span(r
0
, . . . , r
k−1
), we need
only verify that a β
k+1
can be found so that if p
k+1
is given by (2.23) then
p
T
k+1
Ar
l
= 0
for all 0 ≤ l ≤ k −1.
Let p
k+1
be given by (2.23). Then for any l ≤ k
p
T
k+1
Ar
l
= r
T
k
Ar
l
+β
k+1
p
T
k
Ar
l
.
If l ≤ k −2, then r
l
∈ /
l+1
⊂ /
k−1
. Lemma 2.4.2 then implies that
p
T
k+1
Ar
l
= 0 for 0 ≤ l ≤ k −2.
It only remains to solve for β
k+1
so that p
T
k+1
Ar
k−1
= 0. Trivially
β
k+1
= −r
T
k
Ar
k−1
/p
T
k
Ar
k−1
(2.24)
provided p
T
k
Ar
k−1
,= 0. Since
r
k
= r
k−1
−α
k
Ap
k
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 21
we have
r
T
k
r
k−1
= r
k−1

2
2
−α
k
p
T
k
Ar
k−1
.
Since r
T
k
r
k−1
= 0 by Lemma 2.4.1 we have
p
T
k
Ar
k−1
= r
k−1

2
2
/α
k
,= 0. (2.25)
This completes the proof.
The common implementation of conjugate gradient uses a diﬀerent form
for α
k
and β
k
than given in (2.17) and (2.24).
Lemma 2.4.3. Let A be spd and assume that r
k
,= 0. Then
α
k+1
=
r
k

2
2
p
T
k+1
Ap
k+1
(2.26)
and
β
k+1
=
r
k

2
2
r
k−1

2
2
. (2.27)
Proof. Note that for k ≥ 0
p
T
k+1
r
k+1
= r
T
k
r
k+1
+β
k+1
p
T
k
r
k+1
= 0 (2.28)
by Lemma 2.4.2. An immediate consequence of (2.28) is that p
T
k
r
k
= 0 and
hence
p
T
k+1
r
k
= (r
k
+β
k+1
p
k
)
T
r
k
= r
k

2
2
. (2.29)
Taking scalar products of both sides of
r
k+1
= r
k
−α
k+1
Ap
k+1
with p
k+1
and using (2.29) gives
0 = p
T
k+1
r
k
−α
k+1
p
T
k+1
Ap
k+1
= r
T
k

2
2
−α
k+1
p
T
k+1
Ap
k+1
,
which is equivalent to (2.26).
To get (2.27) note that p
T
k+1
Ap
k
= 0 and hence (2.23) implies that
β
k+1
=
−r
T
k
Ap
k
p
T
k
Ap
k
. (2.30)
Also note that
p
T
k
Ap
k
= p
T
k
A(r
k−1
+β
k
p
k−1
)
= p
T
k
Ar
k−1
+β
k
p
T
k
Ap
k−1
= p
T
k
Ar
k−1
.
(2.31)
Now combine (2.30), (2.31), and (2.25) to get
β
k+1
=
−r
T
k
Ap
k
α
k
r
k−1

2
2
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
22 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Now take scalar products of both sides of
r
k
= r
k−1
−α
k
Ap
k
with r
k
and use Lemma 2.4.1 to get
r
k

2
2
= −α
k
r
T
k
Ap
k
.
Hence (2.27) holds.
The usual implementation reﬂects all of the above results. The goal is to
ﬁnd, for a given , a vector x so that b−Ax
2
≤ b
2
. The input is the initial
iterate x, which is overwritten with the solution, the right hand side b, and a
routine which computes the action of A on a vector. We limit the number of
iterations to kmax and return the solution, which overwrites the initial iterate
x and the residual norm.
Algorithm 2.4.1. cg(x, b, A, , kmax)
1. r = b −Ax, ρ
0
= r
2
2
, k = 1.
2. Do While
√
ρ
k−1
> b
2
and k < kmax
(a) if k = 1 then p = r
else
β = ρ
k−1
/ρ
k−2
and p = r +βp
(b) w = Ap
(c) α = ρ
k−1
/p
T
w
(d) x = x +αp
(e) r = r −αw
(f) ρ
k
= r
2
2
(g) k = k + 1
Note that the matrix A itself need not be formed or stored, only a routine
for matrixvector products is required. Krylov space methods are often called
matrixfree for that reason.
Now, consider the costs. We need store only the four vectors x, w, p, and r.
Each iteration requires a single matrixvector product (to compute w = Ap),
two scalar products (one for p
T
w and one to compute ρ
k
= r
2
2
), and three
operations of the form ax +y, where x and y are vectors and a is a scalar.
It is remarkable that the iteration can progress without storing a basis for
the entire Krylov subspace. As we will see in the section on GMRES, this is
not the case in general. The spd structure buys quite a lot.
2.5. Preconditioning
To reduce the condition number, and hence improve the performance of the
iteration, one might try to replace Ax = b by another spd system with the
same solution. If M is a spd matrix that is close to A
−1
, then the eigenvalues
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 23
of MA will be clustered near one. However MA is unlikely to be spd, and
hence CG cannot be applied to the system MAx = Mb.
In theory one avoids this diﬃculty by expressing the preconditioned
problem in terms of B, where B is spd, A = B
2
, and by using a twosided
preconditioner, S ≈ B
−1
(so M = S
2
). Then the matrix SAS is spd and its
eigenvalues are clustered near one. Moreover the preconditioned system
SASy = Sb
has y
∗
= S
−1
x
∗
as a solution, where Ax
∗
= b. Hence x
∗
can be recovered
from y
∗
by multiplication by S. One might think, therefore, that computing
S (or a subroutine for its action on a vector) would be necessary and that a
matrixvector multiply by SAS would incur a cost of one multiplication by A
and two by S. Fortunately, this is not the case.
If y
k
, ˆ r
k
, ˆ p
k
are the iterate, residual, and search direction for CG applied
to SAS and we let
x
k
= Sˆ y
k
, r
k
= S
−1
ˆ r
k
, p
k
= Sˆ p
k
, and z
k
= Sˆ r
k
,
then one can perform the iteration directly in terms of x
k
, A, and M. The
reader should verify that the following algorithm does exactly that. The input
is the same as that for Algorithm cg and the routine to compute the action of
the preconditioner on a vector. Aside from the preconditioner, the arguments
to pcg are the same as those to Algorithm cg.
Algorithm 2.5.1. pcg(x, b, A, M, , kmax)
1. r = b −Ax, ρ
0
= r
2
2
, k = 1
2. Do While
√
ρ
k−1
> b
2
and k < kmax
(a) z = Mr
(b) τ
k−1
= z
T
r
(c) if k = 1 then β = 0 and p = z
else
β = τ
k−1
/τ
k−2
, p = z +βp
(d) w = Ap
(e) α = τ
k−1
/p
T
w
(f) x = x +αp
(g) r = r −αw
(h) ρ
k
= r
T
r
(i) k = k + 1
Note that the cost is identical to CG with the addition of
• the application of the preconditioner M in step 2a and
• the additional inner product required to compute τ
k
in step 2b.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
24 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Of these costs, the application of the preconditioner is usually the larger. In
the remainder of this section we brieﬂy mention some classes of preconditioners.
A more complete and detailed discussion of preconditioners is in [8] and a
concise survey with many pointers to the literature is in [12].
Some eﬀective preconditioners are based on deep insight into the structure
of the problem. See [124] for an example in the context of partial diﬀerential
equations, where it is shown that certain discretized secondorder elliptic
problems on simple geometries can be very well preconditioned with fast
Poisson solvers [99], [188], and [187]. Similar performance can be obtained from
multigrid [99], domain decomposition, [38], [39], [40], and alternating direction
preconditioners [8], [149], [193], [194]. We use a Poisson solver preconditioner
in the examples in ¸ 2.7 and ¸ 3.7 as well as for nonlinear problems in ¸ 6.4.2
and ¸ 8.4.2.
One commonly used and easily implemented preconditioner is Jacobi
preconditioning, where M is the inverse of the diagonal part of A. One can also
use other preconditioners based on the classical stationary iterative methods,
such as the symmetric Gauss–Seidel preconditioner (1.18). For applications to
partial diﬀerential equations, these preconditioners may be somewhat useful,
but should not be expected to have dramatic eﬀects.
Another approach is to apply a sparse Cholesky factorization to the
matrix A (thereby giving up a fully matrixfree formulation) and discarding
small elements of the factors and/or allowing only a ﬁxed amount of storage
for the factors. Such preconditioners are called incomplete factorization
preconditioners. So if A = LL
T
+ E, where E is small, the preconditioner
is (LL
T
)
−1
and its action on a vector is done by two sparse triangular solves.
We refer the reader to [8], [127], and [44] for more detail.
One could also attempt to estimate the spectrum of A, ﬁnd a polynomial
p such that 1 −zp(z) is small on the approximate spectrum, and use p(A) as a
preconditioner. This is called polynomial preconditioning. The preconditioned
system is
p(A)Ax = p(A)b
and we would expect the spectrum of p(A)A to be more clustered near z = 1
than that of A. If an interval containing the spectrum can be found, the
residual polynomial q(z) = 1 − zp(z) of smallest L
∞
norm on that interval
can be expressed in terms of Chebyshev [161] polynomials. Alternatively
q can be selected to solve a least squares minimization problem [5], [163].
The preconditioning p can be directly recovered from q and convergence rate
estimates made. This technique is used to prove the estimate (2.15), for
example. The cost of such a preconditioner, if a polynomial of degree K is
used, is K matrixvector products for each application of the preconditioner
[5]. The performance gains can be very signiﬁcant and the implementation is
matrixfree.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 25
2.6. CGNR and CGNE
If A is nonsingular and nonsymmetric, one might consider solving Ax = b by
applying CG to the normal equations
A
T
Ax = A
T
b. (2.32)
This approach [103] is called CGNR [71], [78], [134]. The reason for this name
is that the minimization property of CG as applied to (2.32) asserts that
x
∗
−x
2
A
T
A
= (x
∗
−x)
T
A
T
A(x
∗
−x)
= (Ax
∗
−Ax)
T
(Ax
∗
−Ax) = (b −Ax)
T
(b −Ax) = r
2
is minimized over x
0
+/
k
at each iterate. Hence the name Conjugate Gradient
on the Normal equations to minimize the Residual.
Alternatively, one could solve
AA
T
y = b (2.33)
and then set x = A
T
y to solve Ax = b. This approach [46] is now called CGNE
[78], [134]. The reason for this name is that the minimization property of CG
as applied to (2.33) asserts that if y
∗
is the solution to (2.33) then
y
∗
−y
2
AA
T
= (y
∗
−y)
T
(AA
T
)(y
∗
−y) = (A
T
y
∗
−A
T
y)
T
(A
T
y
∗
−A
T
y)
= x
∗
−x
2
2
is minimized over y
0
+/
k
at each iterate. Conjugate Gradient on the Normal
equations to minimize the Error.
The advantages of this approach are that all the theory for CG carries over
and the simple implementation for both CG and PCG can be used. There
are three disadvantages that may or may not be serious. The ﬁrst is that the
condition number of the coeﬃcient matrix A
T
A is the square of that of A.
The second is that two matrixvector products are needed for each CG iterate
since w = A
T
Ap = A
T
(Ap) in CGNR and w = AA
T
p = A(A
T
p) in CGNE.
The third, more important, disadvantage is that one must compute the action
of A
T
on a vector as part of the matrixvector product involving A
T
A. As we
will see in the chapter on nonlinear problems, there are situations where this
is not possible.
The analysis with residual polynomials is similar to that for CG. We
consider the case for CGNR, the analysis for CGNE is essentially the same.
As above, when we consider the A
T
A norm of the error, we have
x
∗
−x
2
A
T
A
= (x
∗
−x)
T
A
T
A(x
∗
−x) = A(x
∗
−x)
2
2
= r
2
2
.
Hence, for any residual polynomial ¯ p
k
∈ T
k
,
r
k

2
≤ ¯ p
k
(A
T
A)r
0

2
≤ r
0

2
max
z∈σ(A
T
A)
[ ¯ p
k
(z)[. (2.34)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
26 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
There are two major diﬀerences between (2.34) and (2.7). The estimate is
in terms of the l
2
norm of the residual, which corresponds exactly to the
termination criterion, hence we need not prove a result like Lemma 2.3.2. Most
signiﬁcantly, the residual polynomial is to be maximized over the eigenvalues
of A
T
A, which is the set of the squares of the singular values of A. Hence the
performance of CGNR and CGNE is determined by the distribution of singular
values.
2.7. Examples for preconditioned conjugate iteration
In the collection of MATLAB codes we provide a code for preconditioned
conjugate gradient iteration. The inputs, described in the comment lines,
are the initial iterate, x
0
, the right hand side vector b, MATLAB functions for
the matrixvector product and (optionally) the preconditioner, and iteration
parameters to specify the maximum number of iterations and the termination
criterion. On return the code supplies the approximate solution x and the
history of the iteration as the vector of residual norms.
We consider the discretization of the partial diﬀerential equation
−∇
˙
(a(x, y)∇u) = f(x, y) (2.35)
on 0 < x, y < 1 subject to homogeneous Dirichlet boundary conditions
u(x, 0) = u(x, 1) = u(0, y) = u(1, y) = 0, 0 < x, y < 1.
One can verify [105] that the diﬀerential operator is positive deﬁnite in the
Hilbert space sense and that the ﬁvepoint discretization described below is
positive deﬁnite if a > 0 for all 0 ≤ x, y ≤ 1 (Exercise 2.8.10).
We discretize with a ﬁvepoint centered diﬀerence scheme with n
2
points
and mesh width h = 1/(n + 1). The unknowns are
u
ij
≈ u(x
i
, x
j
)
where x
i
= ih for 1 ≤ i ≤ n. We set
u
0j
= u
(n+1)j
= u
i0
= u
i(n+1)
= 0,
to reﬂect the boundary conditions, and deﬁne
α
ij
= −a(x
i
, x
j
)h
−2
/2.
We express the discrete matrixvector product as
(Au)
ij
= (α
ij
+α
(i+1)j
)(u
(i+1)j
−u
ij
)
−(α
(i−1)j
+α
ij
)(u
ij
−u
(i−1)j
) + (α
i(j+1)
+α
ij
)(u
i(j+1)
−u
ij
)
−(α
ij
+α
i(j−1)
)(u
ij
−u
i(j−1)
)
(2.36)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 27
for 1 ≤ i, j ≤ n.
For the MATLAB implementation we convert freely from the representa
tion of u as a twodimensional array (with the boundary conditions added),
which is useful for computing the action of A on u and applying fast solvers,
and the representation as a onedimensional array, which is what pcgsol ex
pects to see. See the routine fish2d in collection of MATLAB codes for an
example of how to do this in MATLAB.
For the computations reported in this section we took a(x, y) = cos(x) and
took the right hand side so that the exact solution was the discretization of
10xy(1 −x)(1 −y) exp(x
4.5
).
The initial iterate was u
0
= 0.
In the results reported here we took n = 31 resulting in a system with
N = n
2
= 961 unknowns. We expect secondorder accuracy from the method
and accordingly we set termination parameter = h
2
= 1/1024. We allowed
up to 100 CG iterations. The initial iterate was the zero vector. We will report
our results graphically, plotting r
k

2
/b
2
on a semilog scale.
In Fig. 2.1 the solid line is a plot of r
k
/b and the dashed line a plot of
u
∗
−u
k

A
/u
∗
−u
0

A
. Note that the reduction in r is not monotone. This is
consistent with the theory, which predicts decrease in e
A
but not necessarily
in r as the iteration progresses. Note that the unpreconditioned iteration is
slowly convergent. This can be explained by the fact that the eigenvalues are
not clustered and
κ(A) = O(1/h
2
) = O(n
2
) = O(N)
and hence (2.15) indicates that convergence will be slow. The reader is asked
to quantify this in terms of execution times in Exercise 2.8.9. This example
illustrates the importance of a good preconditioner. Even the unpreconditioned
iteration, however, is more eﬃcient that the classical stationary iterative
methods.
For a preconditioner we use a Poisson solver. By this we mean an operator
G such that v = Gw is the solution of the discrete form of
−v
xx
−v
yy
= w,
subject to homogeneous Dirichlet boundary conditions. The eﬀectiveness of
such a preconditioner has been analyzed in [124] and some of the many ways
to implement the solver eﬃciently are discussed in [99], [188], [186], and [187].
The properties of CG on the preconditioned problem in the continuous
case have been analyzed in [48]. For many types of domains and boundary
conditions, Poisson solvers can be designed to take advantage of vector and/or
parallel architectures or, in the case of the MATLAB environment used in
this book, designed to take advantage of fast MATLAB builtin functions.
Because of this their execution time is less than a simple count of ﬂoating
point operations would indicate. The fast Poisson solver in the collection of
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
28 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
0 10 20 30 40 50 60
10
4
10
3
10
2
10
1
10
0
10
1
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
a
n
d
A

n
o
r
m
o
f
E
r
r
o
r
Fig. 2.1. CG for 2D elliptic equation.
0 10 20 30 40 50 60
10
4
10
3
10
2
10
1
10
0
10
1
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 2.2. PCG for 2D elliptic equation.
codes fish2d is based on the MATLAB fast Fourier transform, the builtin
function fft.
In Fig. 2.2 the solid line is the graph of r
k

2
/b
2
for the preconditioned
iteration and the dashed line for the unpreconditioned. The preconditioned
iteration required 5 iterations for convergence and the unpreconditioned
iteration 52. Not only does the preconditioned iteration converge more
rapidly, but the number of iterations required to reduce the relative residual
by a given amount is independent of the mesh spacing [124]. We caution
the reader that the preconditioned iteration is not as much faster than the
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 29
unpreconditioned one as the iteration count would suggest. The MATLAB
flops command indicates that the unpreconditioned iteration required roughly
1.2 million ﬂoatingpoint operations while the preconditioned iteration required
.87 million ﬂoatingpoint operations. Hence, the cost of the preconditioner is
considerable. In the MATLAB environment we used, the execution time of
the preconditioned iteration was about 60% of that of the unpreconditioned.
As we remarked above, this speed is a result of the eﬃciency of the MATLAB
fast Fourier transform. In Exercise 2.8.11 you are asked to compare execution
times for your own environment.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
30 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
2.8. Exercises on conjugate gradient
2.8.1. Let ¦x
k
¦ be the conjugate gradient iterates. Prove that r
l
∈ /
k
for all
l < k.
2.8.2. Let A be spd. Show that there is a spd B such that B
2
= A. Is B
unique?
2.8.3. Let Λ be a diagonal matrix with Λ
ii
= λ
i
and let p be a polynomial.
Prove that p(Λ) = max
i
[p(λ
i
)[ where   is any induced matrix norm.
2.8.4. Prove Theorem 2.2.3.
2.8.5. Assume that A is spd and that
σ(A) ⊂ (1, 1.1) ∪ (2, 2.2).
Give upper estimates based on (2.6) for the number of CG iterations
required to reduce the A norm of the error by a factor of 10
−3
and for
the number of CG iterations required to reduce the residual by a factor
of 10
−3
.
2.8.6. For the matrix A in problem 5, assume that the cost of a matrix vector
multiply is 4N ﬂoatingpoint multiplies. Estimate the number of ﬂoating
point operations reduce the A norm of the error by a factor of 10
−3
using
CG iteration.
2.8.7. Let A be a nonsingular matrix with all singular values in the interval
(1, 2). Estimate the number of CGNR/CGNE iterations required to
reduce the relative residual by a factor of 10
−4
.
2.8.8. Show that if A has constant diagonal then PCG with Jacobi precondi
tioning produces the same iterates as CG with no preconditioning.
2.8.9. Assume that A is N N, nonsingular, and spd. If κ(A) = O(N), give
a rough estimate of the number of CG iterates required to reduce the
relative residual to O(1/N).
2.8.10. Prove that the linear transformation given by (2.36) is symmetric and
positive deﬁnite on R
n
2
if a(x, y) > 0 for all 0 ≤ x, y ≤ 1.
2.8.11. Duplicate the results in ¸ 2.7 for example, in MATLAB by writing the
matrixvector product routines and using the MATLAB codes pcgsol
and fish2d. What happens as N is increased? How are the performance
and accuracy aﬀected by changes in a(x, y)? Try a(x, y) =
√
.1 +x and
examine the accuracy of the result. Explain your ﬁndings. Compare
the execution times on your computing environment (using the cputime
command in MATLAB, for instance).
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
CONJUGATE GRADIENT ITERATION 31
2.8.12. Use the Jacobi and symmetric Gauss–Seidel iterations from Chapter 1
to solve the elliptic boundary value problem in ¸ 2.7. How does the
performance compare to CG and PCG?
2.8.13. Implement Jacobi (1.17) and symmetric Gauss–Seidel (1.18) precondi
tioners for the elliptic boundary value problem in ¸ 2.7. Compare the
performance with respect to both computer time and number of itera
tions to preconditioning with the Poisson solver.
2.8.14. Modify pcgsol so that φ(x) is computed and stored at each iterate
and returned on output. Plot φ(x
n
) as a function of n for each of the
examples.
2.8.15. Apply CG and PCG to solve the ﬁvepoint discretization of
−u
xx
(x, y) −u
yy
(x, y) +e
x+y
u(x, y) = 1, 0 < x, y, < 1,
subject to the inhomogeneous Dirichlet boundary conditions
u(x, 0) = u(x, 1) = u(1, y) = 0, u(0, y) = 1, 0 < x, y < 1.
Experiment with diﬀerent mesh sizes and preconditioners (Fast Poisson
solver, Jacobi, and symmetric Gauss–Seidel).
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
32 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 3
GMRES Iteration
3.1. The minimization property and its consequences
The GMRES (Generalized Minimum RESidual) was proposed in 1986 in [167]
as a Krylov subspace method for nonsymmetric systems. Unlike CGNR,
GMRES does not require computation of the action of A
T
on a vector. This is
a signiﬁcant advantage in many cases. The use of residual polynomials is made
more complicated because we cannot use the spectral theorem to decompose
A. Moreover, one must store a basis for /
k
, and therefore storage requirements
increase as the iteration progresses.
The kth (k ≥ 1) iteration of GMRES is the solution to the least squares
problem
minimize
x∈x
0
+K
k
b −Ax
2
. (3.1)
The beginning of this section is much like the analysis for CG. Note that
if x ∈ x
0
+/
k
then
x = x
0
+
k−1
j=0
γ
j
A
j
r
0
and so
b −Ax = b −Ax
0
−
k−1
j=0
γ
j
A
j+1
r
0
= r
0
−
k
j=1
γ
j−1
A
j
r
0
.
Hence if x ∈ x
0
+/
k
then r = ¯ p(A)r
0
where ¯ p ∈ T
k
is a residual polynomial.
We have just proved the following result.
Theorem 3.1.1. Let A be nonsingular and let x
k
be the kth GMRES
iteration. Then for all ¯ p
k
∈ T
k
r
k

2
= min
p∈P
k
¯ p(A)r
0

2
≤ ¯ p
k
(A)r
0

2
. (3.2)
From this we have the following corollary.
Corollary 3.1.1. Let A be nonsingular and let x
k
be the kth GMRES
iteration. Then for all ¯ p
k
∈ T
k
r
k

2
r
0

2
≤ ¯ p
k
(A)
2
. (3.3)
33
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
34 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
We can apply the corollary to prove ﬁnite termination of the GMRES
iteration.
Theorem 3.1.2. Let A be nonsingular. Then the GMRES algorithm will
ﬁnd the solution within N iterations.
Proof. The characteristic polynomial of A is p(z) = det(A − zI). p has
degree N, p(0) = det(A) ,= 0 since A is nonsingular, and so
¯ p
N
(z) = p(z)/p(0) ∈ T
N
is a residual polynomial. It is well known [141] that p(A) = ¯ p
N
(A) = 0. By
(3.3), r
N
= b −Ax
N
= 0 and hence x
N
is the solution.
In Chapter 2 we applied the spectral theorem to obtain more precise infor
mation on convergence rates. This is not an option for general nonsymmetric
matrices. However, if A is diagonalizable we may use (3.2) to get information
from clustering of the spectrum just like we did for CG. We pay a price in that
we must use complex arithmetic for the only time in this book. Recall that
A is diagonalizable if there is a nonsingular (possibly complex!) matrix V such
that
A = V ΛV
−1
.
Here Λ is a (possibly complex!) diagonal matrix with the eigenvalues of A on
the diagonal. If A is a diagonalizable matrix and p is a polynomial then
p(A) = V p(Λ)V
−1
A is normal if the diagonalizing transformation V is orthogonal. In that case
the columns of V are the eigenvectors of A and V
−1
= V
H
. Here V
H
is the
complex conjugate transpose of V . In the remainder of this section we must
use complex arithmetic to analyze the convergence. Hence we will switch to
complex matrices and vectors. Recall that the scalar product in C
N
, the space
of complex Nvectors, is x
H
y. In particular, we will use the l
2
norm in C
N
.
Our use of complex arithmetic will be implicit for the most part and is needed
only so that we may admit the possibility of complex eigenvalues of A.
We can use the structure of a diagonalizable matrix to prove the following
result.
Theorem 3.1.3. Let A = V ΛV
−1
be a nonsingular diagonalizable matrix.
Let x
k
be the kth GMRES iterate. Then for all ¯ p
k
∈ T
k
r
k

2
r
0

2
≤ κ
2
(V ) max
z∈σ(A)
[ ¯ p
k
(z)[. (3.4)
Proof. Let ¯ p
k
∈ T
k
. We can easily estimate ¯ p
k
(A)
2
by
¯ p
k
(A)
2
≤ V 
2
V
−1

2
¯ p
k
(Λ)
2
≤ κ
2
(V ) max
z∈σ(A)
[ ¯ p
k
(z)[,
as asserted.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 35
It is not clear how one should estimate the condition number of the
diagonalizing transformation if it exists. If A is normal, of course, κ
2
(V ) = 1.
As we did for CG, we look at some easy consequences of (3.3) and (3.4).
Theorem 3.1.4. Let A be a nonsingular diagonalizable matrix. Assume
that A has only k distinct eigenvalues. Then GMRES will terminate in at most
k iterations.
Theorem 3.1.5. Let A be a nonsingular normal matrix. Let b be a linear
combination of k of the eigenvectors of A
b =
k
l=1
γ
l
u
i
l
.
Then the GMRES iteration, with x
0
= 0, for Ax = b will terminate in at most
k iterations.
3.2. Termination
As is the case with CG, GMRES is best thought of as an iterative method.
The convergence rate estimates for the diagonalizable case will involve κ
2
(V ),
but will otherwise resemble those for CG. If A is not diagonalizable, rate
estimates have been derived in [139], [134], [192], [33], and [34]. As the set of
nondiagonalizable matrices has measure zero in the space of N N matrices,
the chances are very high that a computed matrix will be diagonalizable. This
is particularly so for the ﬁnite diﬀerence Jacobian matrices we consider in
Chapters 6 and 8. Hence we conﬁne our attention to diagonalizable matrices.
As was the case with CG, we terminate the iteration when
r
k

2
≤ ηb
2
(3.5)
for the purposes of this example. We can use (3.3) and (3.4) directly to estimate
the ﬁrst k such that (3.5) holds without requiring a lemma like Lemma 2.3.2.
Again we look at examples. Assume that A = V ΛV
−1
is diagonalizable,
that the eigenvalues of A lie in the interval (9, 11), and that κ
2
(V ) = 100.
We assume that x
0
= 0 and hence r
0
= b. Using the residual polynomial
¯ p
k
(z) = (10 −z)
k
/10
k
we ﬁnd
r
k

2
r
0

2
≤ (100)10
−k
= 10
2−k
.
Hence (3.5) holds when 10
2−k
< η or when
k > 2 + log
10
(η).
Assume that I − A
2
≤ ρ < 1. Let ¯ p
k
(z) = (1 − z)
k
. It is a direct
consequence of (3.2) that
r
k

2
≤ ρ
k
r
0

2
. (3.6)
The estimate (3.6) illustrates the potential beneﬁts of a good approximate
inverse preconditioner.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
36 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
The convergence estimates for GMRES in the nonnormal case are much
less satisfying that those for CG, CGNR, CGNE, or GMRES in the normal
case. This is a very active area of research and we refer to [134], [33], [120],
[34], and [36] for discussions of and pointers to additional references to several
questions related to nonnormal matrices.
3.3. Preconditioning
Preconditioning for GMRES and other methods for nonsymmetric problems is
diﬀerent from that for CG. There is no concern that the preconditioned system
be spd and hence (3.6) essentially tells the whole story. However there are two
diﬀerent ways to view preconditioning. If one can ﬁnd M such that
I −MA
2
= ρ < 1,
then applying GMRES to MAx = Mb allows one to apply (3.6) to the
preconditioned system. Preconditioning done in this way is called left
preconditioning. If r = MAx − Mb is the residual for the preconditioned
system, we have, if the product MA can be formed without error,
e
k

2
e
0

2
≤ κ
2
(MA)
r
k

2
r
0

2
,
by Lemma 1.1.1. Hence, if MA has a smaller condition number than A, we
might expect the relative residual of the preconditioned system to be a better
indicator of the relative error than the relative residual of the original system.
If
I −AM
2
= ρ < 1,
one can solve the system AMy = b with GMRES and then set x = My. This is
called right preconditioning. The residual of the preconditioned problem is the
same as that of the unpreconditioned problem. Hence, the value of the relative
residuals as estimators of the relative error is unchanged. Right preconditioning
has been used as the basis for a method that changes the preconditioner as the
iteration progresses [166].
One important aspect of implementation is that, unlike PCG, one can
apply the algorithm directly to the system MAx = Mb (or AMy = b). Hence,
one can write a single matrixvector product routine for MA (or AM) that
includes both the application of A to a vector and that of the preconditioner.
Most of the preconditioning ideas mentioned in ¸ 2.5 are useful for GMRES
as well. In the examples in ¸ 3.7 we use the Poisson solver preconditioner for
nonsymmetric partial diﬀerential equations. Multigrid [99] and alternating
direction [8], [182] methods have similar performance and may be more
generally applicable. Incomplete factorization (LU in this case) preconditioners
can be used [165] as can polynomial preconditioners. Some hybrid algorithms
use the GMRES/Arnoldi process itself to construct polynomial preconditioners
for GMRES or for Richardson iteration [135], [72], [164], [183]. Again we
mention [8] and [12] as a good general references for preconditioning.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 37
3.4. GMRES implementation: Basic ideas
Recall that the least squares problem deﬁning the kth GMRES iterate is
minimize
x∈x
0
+K
k
b −Ax
2
.
Suppose one had an orthogonal projector V
k
onto /
k
. Then any z ∈ /
k
can
be written as
z =
k
l=1
y
l
v
k
l
,
where v
k
l
is the lth column of V
k
. Hence we can convert (3.1) to a least squares
problem in R
k
for the coeﬃcient vector y of z = x −x
0
. Since
x −x
0
= V
k
y
for some y ∈ R
k
, we must have x
k
= x
0
+V
k
y where y minimizes
b −A(x
0
+V
k
y)
2
= r
0
−AV
k
y
2
.
Hence, our least squares problem in R
k
is
minimize
y∈R
kr
0
−AV
k
y
2
. (3.7)
This is a standard linear least squares problem that could be solved by a QR
factorization, say. The problem with such a direct approach is that the matrix
vector product of A with the basis matrix V
k
must be taken at each iteration.
If one uses Gram–Schmidt orthogonalization, however, one can represent
(3.7) very eﬃciently and the resulting least squares problem requires no extra
multiplications of A with vectors. The Gram–Schmidt procedure for formation
of an orthonormal basis for /
k
is called the Arnoldi [4] process. The data are
vectors x
0
and b, a map that computes the action of A on a vector, and a
dimension k. The algorithm computes an orthonormal basis for /
k
and stores
it in the columns of V .
Algorithm 3.4.1. arnoldi(x
0
, b, A, k, V )
1. Deﬁne r
0
= b −Ax
0
and v
1
= r
0
/r
0

2
.
2. For i = 1, . . . , k −1
v
i+1
=
Av
i
−
i
j=1
((Av
i
)
T
v
j
)v
j
Av
i
−
i
j=1
((Av
i
)
T
v
j
)v
j

2
If there is never a division by zero in step 2 of Algorithm arnoldi, then
the columns of the matrix V
k
are an orthonormal basis for /
k
. A division by
zero is referred to as breakdown and happens only if the solution to Ax = b is
in x
0
+/
k−1
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
38 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Lemma 3.4.1. Let A be nonsingular, let the vectors v
j
be generated by
Algorithm arnoldi, and let i be the smallest integer for which
Av
i
−
i
j=1
((Av
i
)
T
v
j
)v
j
= 0.
Then x = A
−1
b ∈ x
0
+/
i
.
Proof. By hypothesis Av
i
∈ /
i
and hence A/
i
⊂ /
i
. Since the columns of
V
i
are an orthonormal basis for /
i
, we have
AV
i
= V
i
H,
where H is an i i matrix. H is nonsingular since A is. Setting β = r
0

2
and e
1
= (1, 0, . . . , 0)
T
∈ R
i
we have
r
i

2
= b −Ax
i

2
= r
0
−A(x
i
−x
0
)
2
.
Since x
i
− x
0
∈ /
i
there is y ∈ R
i
such that x
i
− x
0
= V
i
y. Since r
0
= βV
i
e
1
and V
i
is an orthogonal matrix
r
i

2
= V
i
(βe
1
−Hy)
2
= βe
1
−Hy
R
i+1,
where  
R
k+1 denotes the Euclidean norm in R
k+1
.
Setting y = βH
−1
e
1
proves that r
i
= 0 by the minimization property.
The upper Hessenberg structure can be exploited to make the solution of
the least squares problems very eﬃcient [167].
If the Arnoldi process does not break down, we can use it to implement
GMRES in an eﬃcient way. Set h
ji
= (Av
j
)
T
v
i
. By the Gram–Schmidt
construction, the k+1k matrix H
k
whose entries are h
ij
is upper Hessenberg,
i.e., h
ij
= 0 if i > j +1. The Arnoldi process (unless it terminates prematurely
with a solution) produces matrices ¦V
k
¦ with orthonormal columns such that
AV
k
= V
k+1
H
k
. (3.8)
Hence, for some y
k
∈ R
k
,
r
k
= b −Ax
k
= r
0
−A(x
k
−x
0
) = V
k+1
(βe
1
−H
k
y
k
).
Hence x
k
= x
0
+V
k
y
k
, where y
k
minimizes βe
1
−H
k
y
2
over R
k
. Note that
when y
k
has been computed, the norm of r
k
can be found without explicitly
forming x
k
and computing r
k
= b −Ax
k
. We have, using the orthogonality of
V
k+1
,
r
k

2
= V
k+1
(βe
1
−H
k
y
k
)
2
= βe
1
−H
k
y
k

R
k+1. (3.9)
The goal of the iteration is to ﬁnd, for a given , a vector x so that
b −Ax
2
≤ b
2
.
The input is the initial iterate, x, the righthand side b, and a map that
computes the action of A on a vector. We limit the number of iterations
to kmax and return the solution, which overwrites the initial iterate x and the
residual norm.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 39
Algorithm 3.4.2. gmresa(x, b, A, , kmax, ρ)
1. r = b −Ax, v
1
= r/r
2
, ρ = r
2
, β = ρ, k = 0
2. While ρ > b
2
and k < kmax do
(a) k = k + 1
(b) for j = 1, . . . , k
h
jk
= (Av
k
)
T
v
j
(c) v
k+1
= Av
k
−
k
j=1
h
jk
v
j
(d) h
k+1,k
= v
k+1

2
(e) v
k+1
= v
k+1
/v
k+1

2
(f) e
1
= (1, 0, . . . , 0)
T
∈ R
k+1
Minimize βe
1
−H
k
y
k

R
k+1 over R
k
to obtain y
k
.
(g) ρ = βe
1
−H
k
y
k

R
k+1.
3. x
k
= x
0
+V
k
y
k
.
Note that x
k
is only computed upon termination and is not needed within
the iteration. It is an important property of GMRES that the basis for the
Krylov space must be stored as the iteration progress. This means that in
order to perform k GMRES iterations one must store k vectors of length N.
For very large problems this becomes prohibitive and the iteration is restarted
when the available room for basis vectors is exhausted. One way to implement
this is to set kmax to the maximum number m of vectors that one can store,
call GMRES and explicitly test the residual b−Ax
k
if k = m upon termination.
If the norm of the residual is larger than , call GMRES again with x
0
= x
k
,
the result from the previous call. This restarted version of the algorithm is
called GMRES(m) in [167]. There is no general convergence theorem for the
restarted algorithm and restarting will slow the convergence down. However,
when it works it can signiﬁcantly reduce the storage costs of the iteration. We
discuss implementation of GMRES(m) later in this section.
Algorithm gmresa can be implemented very straightforwardly in MAT
LAB. Step 2f can be done with a single MATLAB command, the backward
division operator, at a cost of O(k
3
) ﬂoatingpoint operations. There are more
eﬃcient ways to solve the least squares problem in step 2f, [167], [197], and we
use the method of [167] in the collection of MATLAB codes. The savings are
slight if k is small relative to N, which is often the case for large problems, and
the simple oneline MATLAB approach can be eﬃcient for such problems.
A more serious problem with the implementation proposed in Algo
rithm gmresa is that the vectors v
j
may become nonorthogonal as a result of
cancellation errors. If this happens, (3.9), which depends on this orthogonality,
will not hold and the residual and approximate solution could be inaccurate. A
partial remedy is to replace the classical Gram–Schmidt orthogonalization in
Algorithm gmresa with modiﬁed Gram–Schmidt orthogonalization. We replace
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
40 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
the loop in step 2c of Algorithm gmresa with
v
k+1
= Av
k
for j = 1, . . . k
v
k+1
= v
k+1
−(v
T
k+1
v
j
)v
j
.
While modiﬁed Gram–Schmidt and classical Gram–Schmidt are equivalent in
inﬁnite precision, the modiﬁed form is much more likely in practice to maintain
orthogonality of the basis.
We illustrate this point with a simple example from [128], doing the
computations in MATLAB. Let δ = 10
−7
and deﬁne
A =
_
_
_
1 1 1
δ δ 0
δ 0 δ
_
_
_.
We orthogonalize the columns of A with classical Gram–Schmidt to obtain
V =
_
_
_
1.0000e + 00 1.0436e −07 9.9715e −08
1.0000e −07 1.0456e −14 −9.9905e −01
1.0000e −07 −1.0000e + 00 4.3568e −02
_
_
_.
The columns of V
U
are not orthogonal at all. In fact v
T
2
v
3
≈ −.004. For
modiﬁed Gram–Schmidt
V =
_
_
_
1.0000e + 00 1.0436e −07 1.0436e −07
1.0000e −07 1.0456e −14 −1.0000e + 00
1.0000e −07 −1.0000e + 00 4.3565e −16
_
_
_.
Here [v
T
i
v
j
−δ
ij
[ ≤ 10
−8
for all i, j.
The versions we implement in the collection of MATLAB codes use modi
ﬁed Gram–Schmidt. The outline of our implementation is Algorithm gmresb.
This implementation solves the upper Hessenberg least squares problem using
the MATLAB backward division operator, and is not particularly eﬃcient. We
present a better implementation in Algorithm gmres. However, this version is
very simple and illustrates some important ideas. First, we see that x
k
need
only be computed after termination as the least squares residual ρ can be used
to approximate the norm of the residual (they are identical in exact arithmetic).
Second, there is an opportunity to compensate for a loss of orthogonality in
the basis vectors for the Krylov space. One can take a second pass through the
modiﬁed Gram–Schmidt process and restore lost orthogonality [147], [160].
Algorithm 3.4.3. gmresb(x, b, A, , kmax, ρ)
1. r = b −Ax, v
1
= r/r
2
, ρ = r
2
, β = ρ, k = 0
2. While ρ > b
2
and k < kmax do
(a) k = k + 1
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 41
(b) v
k+1
= Av
k
for j = 1, . . . k
i. h
jk
= v
T
k+1
v
j
ii. v
k+1
= v
k+1
−h
jk
v
j
(c) h
k+1,k
= v
k+1

2
(d) v
k+1
= v
k+1
/v
k+1

2
(e) e
1
= (1, 0, . . . , 0)
T
∈ R
k+1
Minimize βe
1
−H
k
y
k

R
k+1 to obtain y
k
∈ R
k
.
(f) ρ = βe
1
−H
k
y
k

R
k+1.
3. x
k
= x
0
+V
k
y
k
.
Even if modiﬁed Gram–Schmidt orthogonalization is used, one can still
lose orthogonality in the columns of V . One can test for loss of orthogonality
[22], [147], and reorthogonalize if needed or use a more stable means to create
the matrix V [195]. These more complex implementations are necessary if A is
ill conditioned or many iterations will be taken. For example, one can augment
the modiﬁed Gram–Schmidt process
• v
k+1
= Av
k
for j = 1, . . . k
h
jk
= v
T
k+1
v
j
v
k+1
= v
k+1
−h
jk
v
j
• h
k+1,k
= v
k+1

2
• v
k+1
= v
k+1
/v
k+1

2
with a second pass (reorthogonalization). One can reorthogonalize in every
iteration or only if a test [147] detects a loss of orthogonality. There is nothing
to be gained by reorthogonalizing more than once [147].
The modiﬁed Gram–Schmidt process with reorthogonalization looks like
• v
k+1
= Av
k
for j = 1, . . . , k
h
jk
= v
T
k+1
v
j
v
k+1
= v
k+1
−h
jk
v
j
• h
k+1,k
= v
k+1

2
• If loss of orthogonality is detected
For j = 1, . . . , k
h
tmp
= v
T
k+1
v
j
h
jk
= h
jk
+h
tmp
v
k+1
= v
k+1
−h
tmp
v
j
• h
k+1,k
= v
k+1

2
• v
k+1
= v
k+1
/v
k+1

2
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
42 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
One approach to reorthogonalization is to reorthogonalize in every step.
This doubles the cost of the computation of V and is usually unnecessary.
More eﬃcient and equally eﬀective approaches are based on other ideas. A
variation on a method from [147] is used in [22]. Reorthogonalization is done
after the Gram–Schmidt loop and before v
k+1
is normalized if
Av
k

2
+δv
k+1

2
= Av
k

2
(3.10)
to working precision. The idea is that if the new vector is very small relative
to Av
k
then information may have been lost and a second pass through
the modiﬁed Gram–Schmidt process is needed. We employ this test in the
MATLAB code gmres with δ = 10
−3
.
To illustrate the eﬀects of loss of orthogonality and those of reorthogonal
ization we apply GMRES to the diagonal system Ax = b where b = (1, 1, 1)
T
,
x
0
= (0, 0, 0)
T
, and
A =
_
_
_
.001 0 0
0 .0011 0
0 0 10
4
_
_
_. (3.11)
While in inﬁnite precision arithmetic only three iterations are needed to solve
the system exactly, we ﬁnd in the MATLAB environment that a solution to full
precision requires more than three iterations unless reorthogonalization is ap
plied after every iteration. In Table 3.1 we tabulate relative residuals as a func
tion of the iteration counter for classical Gram–Schmidt without reorthogonal
ization (CGM), modiﬁed Gram–Schmidt without reorthogonalization (MGM),
reorthogonalization based on the test (3.10) (MGMPO), and reorthogonaliza
tion in every iteration (MGMFO). While classical Gram–Schmidt fails, the
reorthogonalization strategy based on (3.10) is almost as eﬀective as the much
more expensive approach of reorthogonalizing in every step. The method based
on (3.10) is the default in the MATLAB code gmresa.
The kth GMRES iteration requires a matrixvector product, k scalar
products, and the solution of the Hessenberg least squares problem in step 2e.
The k scalar products require O(kN) ﬂoatingpoint operations and the cost
of a solution of the Hessenberg least squares problem, by QR factorization or
the MATLAB backward division operator, say, in step 2e of gmresb is O(k
3
)
ﬂoatingpoint operations. Hence the total cost of the m GMRES iterations is
m matrixvector products and O(m
4
+m
2
N) ﬂoatingpoint operations. When
k is not too large and the cost of matrixvector products is high, a brute
force solution to the least squares problem using the MATLAB backward
division operator is not terribly ineﬃcient. We provide an implementation
of Algorithm gmresb in the collection of MATLAB codes. This is an appealing
algorithm, especially when implemented in an environment like MATLAB,
because of its simplicity. For large k, however, the bruteforce method can be
very costly.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 43
Table 3.1
Eﬀects of reorthogonalization.
k CGM MGM MGMPO MGMFO
0 1.00e+00 1.00e+00 1.00e+00 1.00e+00
1 8.16e01 8.16e01 8.16e01 8.16e01
2 3.88e02 3.88e02 3.88e02 3.88e02
3 6.69e05 6.42e08 6.42e08 6.34e34
4 4.74e05 3.70e08 5.04e24
5 3.87e05 3.04e18
6 3.35e05
7 3.00e05
8 2.74e05
9 2.53e05
10 2.37e05
3.5. Implementation: Givens rotations
If k is large, implementations using Givens rotations [167], [22], Householder
reﬂections [195], or a shifted Arnoldi process [197] are much more eﬃcient
than the bruteforce approach in Algorithm gmresb. The implementation in
Algorithm gmres and in the MATLAB code collection is from [167]. This
implementation maintains the QR factorization of H
k
in a clever way so that
the cost for a single GMRES iteration is O(Nk) ﬂoatingpoint operations. The
O(k
2
) cost of the triangular solve and the O(kN) cost of the construction of
x
k
are incurred after termination.
A 2 2 Givens rotation is a matrix of the form
G =
_
c −s
s c
_
, (3.12)
where c = cos(θ), s = sin(θ) for θ ∈ [−π, π]. The orthogonal matrix G rotates
the vector (c, −s), which makes an angle of −θ with the xaxis through an
angle θ so that it overlaps the xaxis.
G
_
c
−s
_
=
_
1
0
_
.
An N N Givens rotation replaces a 2 2 block on the diagonal of the
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
44 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
N N identity matrix with a 2 2 Givens rotation.
G =
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
1 0 . . . 0
0
.
.
.
.
.
.
.
.
. c −s
.
.
. s c 0
.
.
.
0 1
.
.
.
.
.
.
.
.
. 0
0 . . . 0 1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. (3.13)
Our notation is that G
j
(c, s) is an N N givens rotation of the form (3.13)
with a 2 2 Givens rotation in rows and columns j and j + 1.
Givens rotations are used to annihilate single nonzero elements of matrices
in reduction to triangular form [89]. They are of particular value in reducing
Hessenberg matrices to triangular form and thereby solving Hessenberg least
squares problems such as the ones that arise in GMRES. This reduction can be
accomplished in O(N) ﬂoatingpoint operations and hence is far more eﬃcient
than a solution by a singular value decomposition or a reduction based on
Householder transformations. This method is also used in the QR algorithm
for computing eigenvalues [89], [184].
Let H be an N M (N ≥ M) upper Hessenberg matrix with rank M.
We reduce H to triangular form by ﬁrst multiplying the matrix by a Givens
rotation that annihilates h
21
(and, of course, changes h
11
and the subsequent
columns). We deﬁne G
1
= G
1
(c
1
, s
1
) by
c
1
= h
11
/
_
h
2
11
+h
2
21
and s
1
= −h
21
/
_
h
2
11
+h
2
21
. (3.14)
If we replace H by G
1
H, then the ﬁrst column of H now has only a single
nonzero element h
11
. Similarly, we can now apply G
2
(c
2
, s
2
) to H, where
c
2
= h
22
/
_
h
2
22
+h
2
32
and s
2
= −h
32
/
_
h
2
22
+h
2
32
. (3.15)
and annihilate h
32
. Note that G
2
does not aﬀect the ﬁrst column of H.
Continuing in this way and setting
Q = G
N
. . . G
1
we see that QH = R is upper triangular.
A straightforward application of these ideas to Algorithm gmres would
solve the least squares problem by computing the product of k Givens rotations
Q, setting g = βQe
1
, and noting that
βe
1
−H
k
y
k

R
k+1 = Q(βe
1
−H
k
y
k
)
R
k+1 = g −R
k
y
k

R
k+1,
where R
k
is the k + 1 k triangular factor of the QR factorization of H
k
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 45
In the context of GMRES iteration, however, we can incrementally perform
the QR factorization of H as the GMRES iteration progresses [167]. To see
this, note that if R
k
= Q
k
H
k
and, after orthogonalization, we add the new
column h
k+2
to H
k
, we can update both Q
k
and R
k
by ﬁrst multiplying h
k+2
by Q
k
(that is applying the ﬁrst k Givens rotations to h
k+2
), then computing
the Givens rotation G
k+1
that annihilates the (k + 2)nd element of Q
k
h
k+2
,
and ﬁnally setting Q
k+1
= G
k+1
Q
k
and forming R
k+1
by augmenting R
k
with
G
k+1
Q
k
h
k+2
.
The MATLAB implementation of Algorithm gmres stores Q
k
by storing
the sequences ¦c
j
¦ and ¦s
j
¦ and then computing the action of Q
k
on a vector
x ∈ R
k+1
by applying G
j
(c
j
, s
j
) in turn to obtain
Q
k
x = G
k
(c
k
, s
k
) . . . G
1
(c
1
, s
1
)x.
We overwrite the upper triangular part of H
k
with the triangular part of the
QR factorization of H
k
in the MATLAB code. The MATLAB implementation
of Algorithm gmres uses (3.10) to test for loss of orthogonality.
Algorithm 3.5.1. gmres(x, b, A, , kmax, ρ)
1. r = b −Ax, v
1
= r/r
2
, ρ = r
2
, β = ρ,
k = 0; g = ρ(1, 0, . . . , 0)
T
∈ R
kmax+1
2. While ρ > b
2
and k < kmax do
(a) k = k + 1
(b) v
k+1
= Av
k
for j = 1, . . . k
i. h
jk
= v
T
k+1
v
j
ii. v
k+1
= v
k+1
−h
jk
v
j
(c) h
k+1,k
= v
k+1

2
(d) Test for loss of orthogonality and reorthogonalize if necessary.
(e) v
k+1
= v
k+1
/v
k+1

2
(f) i. If k > 1 apply Q
k−1
to the kth column of H.
ii. ν =
_
h
2
k,k
+h
2
k+1,k
.
iii. c
k
= h
k,k
/ν, s
k
= −h
k+1,k
/ν
h
k,k
= c
k
h
k,k
−s
k
h
k+1,k
, h
k+1,k
= 0
iv. g = G
k
(c
k
, s
k
)g.
(g) ρ = [(g)
k+1
[.
3. Set r
i,j
= h
i,j
for 1 ≤ i, j ≤ k.
Set (w)
i
= (g)
i
for 1 ≤ i ≤ k.
Solve the upper triangular system Ry
k
= w.
4. x
k
= x
0
+V
k
y
k
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
46 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
We close with an example of an implementation of GMRES(m) . This
implementation does not test for success and may, therefore, fail to terminate.
You are asked to repair this in exercise 7. Aside from the parameter m, the
arguments to Algorithm gmresm are the same as those for Algorithm gmres.
Algorithm 3.5.2. gmresm(x, b, A, , kmax, m, ρ)
1. gmres(x, b, A, , m, ρ)
2. k = m
3. While ρ > b
2
and k < kmax do
(a) gmres(x, b, A, , m, ρ)
(b) k = k +m
The storage costs of m iterations of gmres or of gmresm are the m + 2
vectors b, x, and ¦v
k
¦
m
k=1
.
3.6. Other methods for nonsymmetric systems
The methods for nonsymmetric linear systems that receive most attention in
this book, GMRES, CGNR, and CGNE, share the properties that they are easy
to implement, can be analyzed by a common residual polynomial approach, and
only terminate if an acceptable approximate solution has been found. CGNR
and CGNE have the disadvantages that a transposevector product must be
computed for each iteration and that the coeﬃcient matrix A
T
A or AA
T
has
condition number the square of that of the original matrix. In ¸ 3.8 we give
an example of how this squaring of the condition number can lead to failure
of the iteration. GMRES needs only matrixvector products and uses A alone,
but, as we have seen, a basis for /
k
must be stored to compute x
k
. Hence,
the storage requirements increase as the iteration progresses. For large and
illconditioned problems, it may be impossible to store enough basis vectors
and the iteration may have to be restarted. Restarting can seriously degrade
performance.
An ideal method would, like CG, only need matrixvector products, be
based on some kind of minimization principle or conjugacy, have modest
storage requirements that do not depend on the number of iterations needed
for convergence, and converge in N iterations for all nonsingular A. However,
[74], methods based on shortterm recurrences such as CG that also satisfy
minimization or conjugacy conditions cannot be constructed for general
matrices. The methods we describe in this section fall short of the ideal, but
can still be quite useful. We discuss only a small subset of these methods and
refer the reader to [12] and [78] for pointers to more of the literature on this
subject. All the methods we present in this section require two matrixvector
products for each iteration.
Consistently with our implementation of GMRES, we take the view that
preconditioners will be applied externally to the iteration. However, as with
CG, these methods can also be implemented in a manner that incorporates
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 47
the preconditioner and uses the residual for the original system to control
termination.
3.6.1. BiCG. The earliest such method BiCG (Biconjugate gradient)
[122], [76], does not enforce a minimization principle; instead, the kth residual
must satisfy the biorthogonality condition
r
T
k
w = 0 for all w ∈ /
k
, (3.16)
where
/
k
= span(ˆ r
0
, A
T
ˆ r
0
, . . . , (A
T
)
k−1
ˆ r
0
)
is the Krylov space for A
T
and the vector ˆ r
0
. ˆ r
0
is a usersupplied vector and is
often set to r
0
. The algorithm gets its name because it produces sequences of
residuals ¦r
k
¦, ¦ˆ r
k
¦ and search directions ¦p
k
¦, ¦ˆ p
k
¦ such that biorthogonality
holds, i. e. ˆ r
T
k
r
l
= 0 if k ,= l and the search directions ¦p
k
¦ and ¦ˆ p
k
¦ satisfy
the biconjugacy property
ˆ p
T
k
Ap
l
= 0 if k ,= l.
In the symmetric positive deﬁnite case (3.16) is equivalent to the minimization
principle (2.2) for CG [89].
Using the notation of Chapter 2 and [191] we give an implementation of Bi
CG making the choice ˆ r
0
= r
0
. This algorithmic description is explained and
motivated in more detail in [122], [76], [78], and [191]. We do not recommend
use of BiCG and present this algorithm only as a basis for discussion of some
of the properties of this class of algorithms.
Algorithm 3.6.1. bicg(x, b, A, , kmax)
1. r = b −Ax, ˆ r = r, ρ
0
= 1, ˆ p = p = 0, k = 0
2. Do While r
2
> b
2
and k < kmax
(a) k = k + 1
(b) ρ
k
= ˆ r
T
r, β = ρ
k
/ρ
k−1
(c) p = r +βp, ˆ p = ˆ r +βˆ p
(d) v = Ap
(e) α = ρ
k
/(ˆ p
T
v)
(f) x = x +αp
(g) r = r −αv; ˆ r = ˆ r −αA
T
ˆ p
Note that if A = A
T
is spd, BiCG produces the same iterations as CG
(but computes everything except x twice). If, however, A is not spd, there is
no guarantee that ρ
k
in step 2b or ˆ p
T
Ap, the denominator in step 2e, will not
vanish. If either ρ
k−1
= 0 or ˆ p
T
Ap = 0 we say that a breakdown has taken
place. Even if these quantities are nonzero but very small, the algorithm can
become unstable or produce inaccurate results. While there are approaches
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
48 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
[80], [78], [148], [79], [81], to avoid some breakdowns in many variations of
BiCG, there are no methods that both limit storage and completely eliminate
the possibility of breakdown [74]. All of the methods we present in this section
can break down and should be implemented with that possibility in mind.
Once breakdown has occurred, one can restart the BiCG iteration or pass
the computation to another algorithm, such as GMRES. The possibility of
breakdown is small and certainly should not eliminate the algorithms discussed
below from consideration if there is not enough storage for GMRES to perform
well.
Breakdowns aside, there are other problems with BiCG. A transpose
vector product is needed, which at the least will require additional program
ming and may not be possible at all. The performance of the algorithm can
be erratic or even unstable with residuals increasing by several orders of mag
nitude from one iteration to the next. Moreover, the eﬀort in computing ˆ r
at each iteration is wasted in that ˆ r makes no contribution to x. However,
BiCG sometimes performs extremely well and the remaining algorithms in
this section represent attempts to capture this good performance and damp
the erratic behavior when it occurs.
We can compare the bestcase performance of BiCG with that of GMRES.
Note that there is ¯ p
k
∈ T
k
such that both
r
k
= ¯ p
k
(A)r
0
and ˆ r
k
= ¯ p
k
(A
T
)ˆ r
0
. (3.17)
Hence, by the minimization property for GMRES
r
GMRES
k

2
≤ r
Bi−CG
k

2
,
reminding us that GMRES, if suﬃcient storage is available, will always
reduce the residual more rapidly than BiCG (in terms of iterations, but not
necessarily in terms of computational work). One should also keep in mind that
a single BiCG iteration requires two matrixvector products and a GMRES
iterate only one, but that the cost of the GMRES iteration increases (in terms
of ﬂoatingpoint operations) as the iteration progresses.
3.6.2. CGS. A remedy for one of the problems with BiCG is the Conjugate
Gradient Squared (CGS) algorithm [180]. The algorithm takes advantage of
the fact that (3.17) implies that the scalar product ˆ r
T
r in BiCG (step 2b of
Algorithm bicg) can be represented without using A
T
as
r
T
k
ˆ r
k
= (¯ p
k
(A)r
0
)
T
(¯ p
k
(A
T
)ˆ r
0
) = (¯ p
k
(A)
2
r
0
)
T
ˆ r
0
.
The other references to A
T
can be eliminated in a similar fashion and an
iteration satisfying
r
k
= ¯ p
k
(A)
2
r
0
(3.18)
is produced, where ¯ p
k
is the same polynomial produced by BiCG and used in
(3.17). This explains the name, Conjugate Gradient Squared.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 49
The work used in BiCG to compute ˆ r is now used to update x. CGS
replaces the transposevector product with an additional matrixvector product
and applies the square of the BiCG polynomial to r
0
to produce r
k
. This may,
of course, change the convergence properties for the worse and either improve
good convergence or magnify erratic behavior [134], [191]. CGS has the same
potential for breakdown as BiCG. Since ¯ p
2
k
∈ T
2k
we have
r
GMRES
2k

2
≤ r
CGS
k

2
. (3.19)
We present an algorithmic description of CGS that we will refer to in our
discussion of TFQMR in ¸ 3.6.4. In our description of CGS we will index the
vectors and scalars that would be overwritten in an implementation so that we
can describe some of the ideas in TFQMR later on. This description is taken
from [77].
Algorithm 3.6.2. cgs(x, b, A, , kmax)
1. x
0
= x; p
0
= u
0
= r
0
= b −Ax
2. v
0
= Ap
0
; ˆ r
0
= r
0
3. ρ
0
= ˆ r
T
0
r
0
; k = 0
4. Do While r
2
> b
2
and k < kmax
(a) k = k + 1
(b) σ
k−1
= ˆ r
T
0
v
k−1
(c) α
k−1
= ρ
k−1
/σ
k−1
(d) q
k
= u
k−1
−α
k−1
v
k−1
(e) x
k
= x
k−1
+α
k−1
(u
k−1
+q
k
)
r
k
= r
k−1
−α
k−1
A(u
k−1
+q
k
)
(f) ρ
k
= ˆ r
T
0
r
k
; β
k
= ρ
k
/ρ
k−1
u
k
= r
k
+β
k
q
k
p
k
= u
k
+β
k
(q
k
+β
k
p
k−1
)
v
k
= Ap
k
Breakdowns take place when either ρ
k−1
or σ
k−1
vanish. If the algorithm
does not break down, then α
k−1
,= 0 for all k. One can show [180], [77], that
if ¯ p
k
is the residual polynomial from (3.17) and (3.18) then
¯ p
k
(z) = ¯ p
k−1
(z) −α
k−1
z¯ q
k−1
(z) (3.20)
where the auxiliary polynomials ¯ q
k
∈ T
k
are given by ¯ q
0
= 1 and for k ≥ 1 by
¯ q
k
(z) = ¯ p
k
(z) +β
k
¯ q
k−1
(z). (3.21)
We provide no MATLAB implementation of BiCG or CGS. BiCGSTAB
is the most eﬀective representative of this class of algorithms.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
50 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
3.6.3. BiCGSTAB. The Biconjugate gradient stabilized method (Bi
CGSTAB) [191] attempts to smooth the convergence of CGS by replacing
(3.18) with
r
k
= q
k
(A)p
k
(A)r
0
(3.22)
where
q
k
(z) =
k
i=1
(1 −ω
i
z).
The constant ω
i
is selected to minimize r
i
= q
i
(A)p
i
(A)r
0
as a function of ω
i
.
The examples in [191] indicate the eﬀectiveness of this approach, which can
be thought of [12] as a blend of BiCG and GMRES(1). The performance in
cases where the spectrum has a signiﬁcant imaginary part can be improved by
constructing q
k
to have complex conjugate pairs of roots, [98].
There is no convergence theory for BiCGSTAB and no estimate of the
residuals that is better than that for CGS, (3.19). In our description of the
implementation, which is described and motivated fully in [191], we use ˆ r
0
= r
0
and follow the notation of [191].
Algorithm 3.6.3. bicgstab(x, b, A, , kmax)
1. r = b −Ax, ˆ r
0
= ˆ r = r, ρ
0
= α = ω = 1, v = p = 0, k = 0, ρ
1
= ˆ r
T
0
r
2. Do While r
2
> b
2
and k < kmax
(a) k = k + 1
(b) β = (ρ
k
/ρ
k−1
)(α/ω)
(c) p = r +β(p −ωv)
(d) v = Ap
(e) α = ρ
k
/(ˆ r
T
0
v)
(f) s = r −αv, t = As
(g) ω = t
T
s/t
2
2
, ρ
k+1
= −ωˆ r
T
0
t
(h) x = x +αp +ωs
(i) r = s −ωt
Note that the iteration can break down in steps 2b and 2e. We provide an
implementation of Algorithm bicgstab in the collection of MATLAB codes.
The cost in storage and in ﬂoatingpoint operations per iteration re
mains bounded for the entire iteration. One must store seven vectors
(x, b, r, p, v, ˆ r
0
, t), letting s overwrite r when needed. A single iteration re
quires four scalar products. In a situation where many GMRES iterations are
needed and matrixvector product is fast, BiCGSTAB can have a much lower
average cost per iterate than GMRES. The reason for this is that the cost of
orthogonalization in GMRES can be much more than that of a matrixvector
product if the dimension of the Krylov space is large. We will present such a
case in ¸ 3.8.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 51
3.6.4. TFQMR. Now we consider the quasiminimal residual (QMR)
family of algorithms [80], [81], [77], [37], [202]. We will focus on the transpose
free algorithm TFQMR proposed in [77], to illustrate the quasiminimization
idea. All algorithms in this family minimize the norm of an easily computable
quantity q = f − Hz, a quasiresidual over z ∈ R
N
. The quasiresidual is
related to the true residual by a fullrank linear transformation r = Lq and
reduction of the residual can be approximately measured by reduction in q.
The speciﬁc formulation of q and L depends on the algorithm.
Returning to Algorithm cgs we deﬁne sequences
y
m
=
_
¸
_
¸
_
u
k−1
if m = 2k −1 is odd,
q
k
if m = 2k is even,
and
w
m
=
_
¸
_
¸
_
(¯ p
k
(A))
2
r
0
if m = 2k −1 is odd,
¯ p
k
(A)¯ p
k−1
(A)r
0
if m = 2k is even.
Here the sequences ¦¯ p
k
¦ and ¦¯ q
k
¦ are given by (3.20) and (3.21). Hence, the
CGS residual satisﬁes
r
CGS
k
= w
2k+1
.
We assume that CGS does not break down and, therefore, α
k
,= 0 for all
k ≥ 0. If we let ¸r be the nearest integer less than or equal to a real r we
have
Ay
m
= (w
m
−w
m+1
)/α
(m−1)/2
, (3.23)
where the denominator is nonzero by assumption. We express (3.23) in matrix
form as
AY
m
= W
m+1
B
m
, (3.24)
where Y
m
is the N m matrix with columns ¦y
j
¦
m
j=1
, W
m+1
the N (m+1)
matrix with columns ¦w
j
¦
m+1
j=1
, and B
m
is the (m+ 1) m matrix given by
B
m
=
_
_
_
_
_
_
_
_
_
_
1 0 . . . 0
−1 1
.
.
.
.
.
.
0
.
.
.
.
.
. 0
.
.
.
.
.
. −1 1
0 . . . 0 −1
_
_
_
_
_
_
_
_
_
_
diag(α
0
, α
0
, . . . , α
(m−1)/2
)
−1
.
Our assumptions that no breakdown takes place imply that /
m
is the span
of ¦y
j
¦
m
j=1
and hence
x
m
= x
0
+Y
m
z (3.25)
for some z ∈ R
m
. Therefore,
r
m
= r
0
−AY
m
z = W
m+1
(e
1
−B
m
z), (3.26)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
52 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
where, as in our discussion of GMRES, e
1
= (1, 0, . . . , 0)
T
∈ R
m
. In a sense,
the conditioning of the matrix W
m+1
can be improved by scaling. Let
Ω = diag(ω
1
, ω
2
, . . . , ω
m
) (3.27)
and rewrite (3.26) as
r
m
= r
0
−AY
m
z = W
m+1
Ω
−1
m+1
(f
m+1
−H
m
z) (3.28)
where f
m+1
= Ω
m+1
e
1
= ω
m+1
e
1
and H
m
= Ω
m+1
B
m
is bidiagonal and hence
upper Hessenberg. The diagonal entries in the matrix Ω
m+1
are called weights.
The weights make it possible to derive straightforward estimates of the residual
in terms of easily computed terms in the iteration.
If W
m+1
Ω
−1
m+1
were an orthogonal matrix, we could minimize r
m
by solving
the least squares problem
minimize
z∈R
mf
m+1
−H
m
z
2
. (3.29)
The quasiminimization idea is to solve (3.29) (thereby quasiminimizing the
residual) to obtain z
m
and then set
x
m
= x
0
+Y
m
z
m
. (3.30)
Note that if k corresponds to the iteration counter for CGS, each k produces
two approximate solutions (corresponding to m = 2k −1 and m = 2k).
The solution to the least squares problem (3.29) can be done by using
Givens rotations to update the factorization of the upper Hessenberg matrix
H
m
[77] and this is reﬂected in the implementation description below. As
one can see from that description, approximate residuals are not computed in
Algorithm tfqmr. Instead we have the quasiresidual norm
τ
m
= f
m+1
−H
m
z
m

2
.
Now if we pick the weights ω
i
= w
i

2
then the columns of W
m+1
Ω
−1
m+1
have
norm one. Hence
r
m

2
= W
m+1
Ω
−1
m+1
(f
m+1
−H
m
z)
2
≤ τ
m
√
m+ 1. (3.31)
We can, therefore, base our termination criterion on τ
m
, and we do this in
Algorithm tfqmr.
One can [80], [78], also use the quasiminimization condition to estimate
the TFQMR residual in terms of residual polynomials.
Theorem 3.6.1. Let A be an N N matrix and let x
0
, b ∈ R
N
be given.
Assume that the TFQMR iteration does not break down or terminate with
the exact solution for k iterations and let 1 ≤ m ≤ 2k. Let r
GMRES
m
be the
residual for the mth GMRES iteration. Let ξ
m
be the smallest singular value
of W
m+1
Ω
−1
m+1
. Then if ξ
m
> 0
τ
m
≤ r
GMRES
m

2
/ξ
m
. (3.32)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 53
Proof. We may write
r
GMRES
m
= r
0
+Y
m
z
GMRES
m
and obtain
r
GMRES
m

2
= W
m+1
Ω
−1
m+1
(f
m+1
−H
m
z
GMRES
m
)
2
≥ ξ
m
f
m+1
−H
m
z
GMRES
m

2
.
Hence, by the quasiminimization property
r
GMRES
m

2
≥ ξ
m
τ
m
as desired.
As a corollary we have a ﬁnite termination result and a convergence
estimate for diagonalizable matrices similar to Theorem 3.1.3.
Corollary 3.6.1. Let A be an N N matrix and let x
0
, b ∈ R
N
be given.
Then within (N +1)/2 iterations the TFQMR iteration will either break down
or terminate with the solution.
We apply Theorem 3.1.3 to (3.32) and use (3.31) to obtain the following
result.
Theorem 3.6.2. Let A = V ΛV
−1
be a nonsingular diagonalizable matrix.
Assume that TFQMR does not break down or terminate with the solution for k
iterations. For 1 ≤ m ≤ 2k let ξ
m
be the smallest singular value of W
m+1
Ω
−1
m+1
and let x
m
be given by (3.30). Then, if ξ
m
> 0,
r
m

2
r
0

2
≤
√
m+ 1ξ
−1
m
κ(V ) max
z∈σ(A)
[φ(z)[
for all φ ∈ T
m
.
The implementation below follows [77] with ˆ r
0
= r
0
used throughout.
Algorithm 3.6.4. tfqmr(x, b, A, , kmax)
1. k = 0, w
1
= y
1
= r
0
= b −Ax, u
1
= v = Ay
1
, d = 0
ρ
0
= r
T
0
r
0
, τ = r
2
, θ = 0, η = 0
2. Do While k < kmax
(a) k = k + 1
(b) σ
k−1
= r
T
0
v, α = ρ
k−1
/σ
k−1
, y
2
= y
1
−α
k−1
v, u
2
= Ay
2
(c) For j = 1, 2 (m = 2k −2 +j)
i. w = w −α
k−1
u
j
ii. d = y
j
+ (θ
2
η/α
k−1
)d
iii. θ = w
2
/τ, c = 1/
√
1 +θ
2
iv. τ = τθc, η = c
2
α
k−1
v. x = x +ηd
vi. If τ
√
m+ 1 ≤ b
2
terminate successfully
(d) ρ
k
= r
T
0
w, β = ρ
k
/ρ
k−1
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
54 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
(e) y
1
= w +βy
2
, u
1
= Ay
1
(f) v = u
1
+β(u
2
+βv)
Note that y
2
and u
2
= Ay
2
need only be computed if the loop in step 2c
does not terminate when j = 1. We take advantage of this in the MATLAB
code tfqmr to potentially reduce the matrixvector product cost by one.
3.7. Examples for GMRES iteration
These examples use the code gmres from the collection of MATLAB codes. The
inputs, described in the comment lines, are the initial iterate x
0
, the righthand
side vector b, a MATLAB function for the matrixvector product, and iteration
parameters to specify the maximum number of iterations and the termination
criterion. We do not pass the preconditioner to the GMRES code, rather
we pass the preconditioned problem. In all the GMRES examples, we limit
the number of GMRES iterations to 60. For methods like GMRES(m), Bi
CGSTAB, CGNR, and TFQMR, whose storage requirements do not increase
with the number of iterations, we allow for more iterations.
We consider the discretization of the partial diﬀerential equation
(Lu)(x, y) = −(u
xx
(x, y) +u
yy
(x, y)) +a
1
(x, y)u
x
(x, y)
+a
2
(x, y)u
y
(x, y) +a
3
(x, y)u(x, y) = f(x, y)
(3.33)
on 0 < x, y < 1 subject to homogeneous Dirichlet boundary conditions
u(x, 0) = u(x, 1) = u(0, y) = u(1, y) = 0, 0 < x, y < 1.
For general coeﬃcients ¦a
j
¦ the operator L is not selfadjoint and its discretiza
tion is not symmetric. As in ¸ 2.7 we discretize with a ﬁvepoint centered dif
ference scheme with n
2
points and mesh width h = 1/(n + 1). The unknowns
are
u
ij
≈ u(x
i
, x
j
)
where x
i
= ih for 1 ≤ i ≤ n. We compute Lu in a matrixfree manner
as we did in ¸ 2.7. We let n = 31 to create a system with 961 unknowns.
As before, we expect secondorder accuracy and terminate the iteration when
r
k

2
≤ h
2
b
2
≈ 9.8 10
−4
b
2
.
As a preconditioner we use the fast Poisson solver fish2d described in ¸ 2.7.
The motivation for this choice is similar to that for the conjugate gradient
computations and has been partially explained theoretically in [33], [34], [121],
and [139]. If we let Gu denote the action of the Poisson solver on u, the
preconditioned system is GLu = Gf.
For the computations reported in this section we took
a
1
(x, y) = 1, a
2
(x, y) = 20y, and a
3
(x, y) = 1.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 55
u
0
= 0 was the initial iterate. As in ¸ 2.7 we took the right hand side so that
the exact solution was the discretization of
10xy(1 −x)(1 −y) exp(x
4.5
).
In Figs. 3.1 and 3.2 we plot iteration histories corresponding to precondi
tioned or unpreconditioned GMRES. For the restarted iteration, we restarted
the algorithm after three iterations, GMRES(3) in the terminology of ¸ 3.4.
Note that the plots are semilog plots of r
2
/b
2
and that the righthand
sides in the preconditioned and unpreconditioned cases are not the same. In
both ﬁgures the solid line is the history of the preconditioned iteration and the
dashed line that of the unpreconditioned. Note the importance of precondi
tioning. The MATLAB flops command indicates that the unpreconditioned
iteration required 7.6 million ﬂoating point operations and converged in 56
iterations. The preconditioned iteration required 1.3 million ﬂoating point op
erations and terminated after 8 iterations. Restarting after 3 iterations is more
costly in terms of the iteration count than not restarting. However, even in
the unpreconditioned case, the restarted iteration terminated successfully. In
this case the MATLAB flops command indicates that the unpreconditioned
iteration terminated after 223 iterations and 9.7 million ﬂoating point opera
tions and the preconditioned iteration required 13 iterations and 2.6 million
ﬂoating point operations.
The execution times in our environment indicate that the preconditioned
iterations are even faster than the diﬀerence in operation counts indicates.
The reason for this is that the FFT routine is a MATLAB intrinsic and is not
interpreted MATLAB code. In other computing environments preconditioners
which vectorize or parallelize particularly well might perform in the same way.
3.8. Examples for CGNR, BiCGSTAB, and TFQMR iteration
When a transpose is available, CGNR (or CGNE) is an alternative to GMRES
that does not require storage of the iteration history. For elliptic problems like
(3.33), the transpose (adjoint) of L is given by
L
∗
u = −u
xx
−u
yy
−a
1
u
x
−a
2
u
y
+a
3
u
as one can derive by integration by parts. To apply CGNR to the precondi
tioned problem, we require the transpose of GL, which is given by
(GL)
∗
u = L
∗
G
∗
u = L
∗
Gu.
The cost of the application of the transpose of L or GL is the same as that
of the application of L or GL itself. Hence, in the case where the cost of the
matrixvector product dominates the computation, a single CGNR iteration
costs roughly the same as two GMRES iterations.
However, CGNR has the eﬀect of squaring the condition number, this eﬀect
has serious consequences as Fig. 3.3 shows. The dashed line corresponds to
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
56 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
0 10 20 30 40 50 60
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 3.1. GMRES for 2D elliptic equation.
0 50 100 150 200 250
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 3.2. GMRES(3) for 2D elliptic equation.
CGNR on the unpreconditioned problem, i.e., CG applied to L
∗
Lu = L
∗
f.
The matrix corresponding to the discretization of L
∗
L has a large condition
number. The solid line corresponds to the CGNR applied to the preconditioned
problem, i.e., CG applied to L
∗
G
2
Lu = L
∗
G
2
f. We limited the number of
iterations to 310 = 10m and the unpreconditioned formulation of CGNR had
made very little progress after 310 iterations. This is a result of squaring
the large condition number. The preconditioned formulation did quite well,
converging in eight iterations. The MATLAB flops command indicates that
the unpreconditioned iteration required 13.7 million ﬂoatingpoint operations
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 57
and the preconditioned iteration 2.4 million. However, as you are asked to
investigate in Exercise 3.9.8, the behavior of even the preconditioned iteration
can also be poor if the coeﬃcients of the ﬁrst derivative terms are too large.
CGNE has similar performance; see Exercise 3.9.9.
0 50 100 150 200 250 300 350
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 3.3. CGNR for 2D elliptic equation.
The examples for BiCGSTAB use the code bicgstab from the collection
of MATLAB codes. This code has the same input/output arguments as gmres.
We applied BiCGSTAB to both the preconditioned and unpreconditioned
forms of (3.33) with the same data and termination criterion as for the GMRES
example.
We plot the iteration histories in Fig. 3.4. As in the previous examples, the
solid line corresponds to the preconditioned iteration and the dashed line to
the unpreconditioned iteration. Neither convergence history is monotonic, with
the unpreconditioned iteration being very irregular, but not varying by many
orders of magnitude as a CGS or BiCG iteration might. The preconditioned
iteration terminated in six iterations (12 matrixvector products) and required
roughly 1.7 million ﬂoatingpoint operations. The unpreconditioned iteration
took 40 iterations and 2.2 million ﬂoatingpoint operations. We see a
considerable improvement in cost over CGNR and, since our unpreconditioned
matrixvector product is so inexpensive, a better cost/iterate than GMRES in
the unpreconditioned case.
We repeat the experiment with TFQMR as our linear solver. We use
the code tfqmr that is included in our collection. In Fig. 3.5 we plot the
convergence histories in terms of the approximate residual
√
2k + 1τ
2k
given
by (3.31). We plot only full iterations (m = 2k) except, possibly, for the
ﬁnal iterate. The approximate residual is given in terms of the quasiresidual
norm τ
m
rather than the true residual, and the graphs are monotonic. Our
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
58 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
10
0
10
1
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 3.4. BiCGSTAB for 2D elliptic equation.
0 10 20 30 40 50 60 70
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
A
p
p
r
o
x
i
m
a
t
e
R
e
s
i
d
u
a
l
Fig. 3.5. TFQMR for 2D elliptic equation.
termination criterion is based on the quasiresidual and (3.31) as described in
¸ 3.6.4, stopping the iteration when τ
m
√
m+ 1/b
2
≤ h
−2
. This led to actual
relative residuals of 7 10
−5
for the unpreconditioned iteration and 2 10
−4
for the preconditioned iteration, indicating that it is reasonable to use the
quasiresidual estimate (3.31) to make termination decisions.
The unpreconditioned iteration required roughly 4.1 million ﬂoatingpoint
operations and 67 iterations for termination. The preconditioned iteration
was much more eﬃcient, taking 1.9 million ﬂoatingpoint operations and 7
iterations. The plateaus in the graph of the convergence history for the
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 59
unpreconditioned iteration are related (see [196]) to spikes in the corresponding
graph for a CGS iteration.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
60 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
3.9. Exercises on GMRES
3.9.1. Give an example of a matrix that is not diagonalizable.
3.9.2. Prove Theorem 3.1.4 and Theorem 3.1.5.
3.9.3. Prove that the Arnoldi process (unless it terminates prematurely with a
solution) produces a basis for /
k+1
that satisﬁes (3.8).
3.9.4. Duplicate Table 3.1 and add two more columns corresponding to classical
Gram–Schmidt orthogonalization with reorthogonalization at every step
and determine when the test in (3.10) detects loss of orthogonality. How
do your conclusions compare with those in [160]?
3.9.5. Let A = J
λ
be an elementary Jordan block of size N corresponding to
an eigenvalue λ ,= 0. This means that
J
λ
=
_
_
_
_
_
_
_
_
_
_
λ 1 0
0 λ 1 0
.
.
.
.
.
.
.
.
.
.
.
.
0 λ 1 0
0 λ 1
0 λ
_
_
_
_
_
_
_
_
_
_
.
Give an example of x
0
, b ∈ R
N
, and λ ∈ R such that the GMRES
residuals satisfy r
k

2
= r
0

2
for all k < N.
3.9.6. Consider the centered diﬀerence discretization of
−u
+u
+u = 1, u(0) = u(1) = 0.
Solve this problem with GMRES (without preconditioning) and then
apply as a preconditioner the map M deﬁned by
−(Mf)
= f, (Mf)(0) = (Mf)(1) = 0.
That is, precondition with a solver for the highorder term in the
diﬀerential equation using the correct boundary conditions. Try this
for meshes with 50, 100, 200, . . . points. How does the performance of
the iteration depend on the mesh? Try other preconditioners such as
Gauss–Seidel and Jacobi. How do other methods such as CGNR, CGNE,
BiCGSTAB, and TFQMR perform?
3.9.7. Modify the MATLAB GMRES code to allow for restarts. One way to do
this is to call the gmres code with another code which has an outer loop
to control the restarting as in Algorithm gmresm. See [167] for another
example. How would you test for failure in a restarted algorithm? What
kinds of failures can occur? Repeat Exercise 6 with a limit of three
GMRES iterations before a restart.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GMRES ITERATION 61
3.9.8. Duplicate the results in ¸ 3.7 and ¸ 3.8. Compare the execution times
in your computing environment. Experiment with diﬀerent values of m.
Why might GMRES(m) be faster than GMRES? Vary the dimension and
the coeﬃcient a
y
. Try a
y
= 5y. Why is the performance of the methods
sensitive to a
y
and a
x
?
3.9.9. Apply CGNE to the PDE in ¸ 3.7. How does the performance diﬀer from
CGNR and GMRES?
3.9.10. Consider preconditioning from the right. For the PDE in ¸ 3.7, apply
GMRES, CGNR, BiCGSTAB, TFQMR, and/or CGNE, to the equation
LGw = f and then recover the solution from u = Gw. Compare timings
and solution accuracy with those obtained by left preconditioning.
3.9.11. Are the solutions to (3.33) in ¸ 3.7 and ¸ 3.8 equally accurate? Compare
the ﬁnal results with the known solution.
3.9.12. Implement BiCG and CGS and try them on some of the examples in
this chapter.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
62 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 4
Basic Concepts and FixedPoint Iteration
This part of the book is about numerical methods for the solution of systems
of nonlinear equations. Throughout this part, we will let   denote a norm
on R
N
as well as the induced matrix norm.
We begin by setting the notation. We seek to solve
F(x) = 0. (4.1)
Here F : R
N
→ R
N
. We denote the ith component of F by f
i
. If the
components of F are diﬀerentiable at x ∈ R
N
we deﬁne the Jacobian matrix
F
(x) by
F
(x)
ij
=
∂f
i
∂x
j
(x).
The Jacobian matrix is the vector analog of the derivative. We may express
the fundamental theorem of calculus as follows.
Theorem 4.0.1. Let F be diﬀerentiable in an open set Ω ⊂ R
N
and let
x
∗
∈ Ω. Then for all x ∈ Ω suﬃciently near x
∗
F(x) −F(x
∗
) =
_
1
0
F
(x
∗
+t(x −x
∗
))(x −x
∗
) dt.
4.1. Types of convergence
Iterative methods can be classiﬁed by the rate of convergence.
Definition 4.1.1. Let ¦x
n
¦ ⊂ R
N
and x
∗
∈ R
N
. Then
• x
n
→ x
∗
qquadratically if x
n
→ x
∗
and there is K > 0 such that
x
n+1
−x
∗
 ≤ Kx
n
−x
∗

2
.
• x
n
→ x
∗
qsuperlinearly with qorder α > 1 if x
n
→ x
∗
and there is
K > 0 such that
x
n+1
−x
∗
 ≤ Kx
n
−x
∗

α
.
• x
n
→ x
∗
qsuperlinearly if
lim
n→∞
x
n+1
−x
∗

x
n
−x
∗

= 0.
65
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
66 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
• x
n
→ x
∗
qlinearly with qfactor σ ∈ (0, 1) if
x
n+1
−x
∗
 ≤ σx
n
−x
∗

for n suﬃciently large.
Definition 4.1.2. An iterative method for computing x
∗
is said to be
locally (qquadratically, qsuperlinearly, qlinearly, etc.) convergent if the
iterates converge to x
∗
(qquadratically, qsuperlinearly, qlinearly, etc.) given
that the initial data for the iteration is suﬃciently good.
Note that a qsuperlinearly convergent sequence is also qlinearly conver
gent with qfactor σ for any σ > 0. A qquadratically convergent sequence is
qsuperlinearly convergent with qorder 2.
In general, a qsuperlinearly convergent method is preferable to a qlinearly
convergent one if the cost of a single iterate is the same for both methods. We
will see that often a method that is more slowly convergent, in terms of the
convergence types deﬁned above, can have a cost/iterate so low that the slow
iteration is more eﬃcient.
Sometimes errors are introduced into the iteration that are independent of
the errors in the approximate solution. An example of this would be a problem
in which the nonlinear function F is the output of another algorithm that
provides error control, such as adjustment of the mesh size in an approximation
of a diﬀerential equation. One might only compute F to low accuracy in the
initial phases of the iteration to compute a solution and tighten the accuracy
as the iteration progresses. The rtype convergence classiﬁcation enables us to
describe that idea.
Definition 4.1.3. Let ¦x
n
¦ ⊂ R
N
and x
∗
∈ R
N
. Then ¦x
n
¦ converges
to x
∗
r(quadratically, superlinearly, linearly) if there is a sequence ¦ξ
n
¦ ⊂ R
converging q(quadratically, superlinearly, linearly) to zero such that
x
n
−x
∗
 ≤ ξ
n
.
We say that ¦x
n
¦ converges rsuperlinearly with rorder α > 1 if ξ
n
→ 0
qsuperlinearly with qorder α.
4.2. Fixedpoint iteration
Many nonlinear equations are naturally formulated as ﬁxedpoint problems
x = K(x) (4.2)
where K, the ﬁxedpoint map, may be nonlinear. A solution x
∗
of (4.2) is
called a ﬁxed point of the map K. Such problems are nonlinear analogs of the
linear ﬁxedpoint problems considered in Chapter 1. In this section we analyze
convergence of ﬁxedpoint iteration, which is given by
x
n+1
= K(x
n
). (4.3)
This iterative method is also called nonlinear Richardson iteration, Picard
iteration, or the method of successive substitution.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
FIXEDPOINT ITERATION 67
Before discussing convergence of ﬁxedpoint iteration we make two deﬁni
tions.
Definition 4.2.1. Let Ω ⊂ R
N
and let G : Ω → R
M
. G is Lipschitz
continuous on Ω with Lipschitz constant γ if
G(x) −G(y) ≤ γx −y
for all x, y ∈ Ω.
Definition 4.2.2. Let Ω ⊂ R
N
. K : Ω → R
N
is a contraction mapping
on Ω if K is Lipschitz continuous on Ω with Lipschitz constant γ < 1.
The standard result for ﬁxedpoint iteration is the Contraction Mapping
Theorem [9]. Compare the proof to that of Lemma 1.2.1 and Corollary 1.2.1.
Theorem 4.2.1. Let Ω be a closed subset of R
N
and let K be a contraction
mapping on Ω with Lipschitz constant γ < 1 such that K(x) ∈ Ω for all x ∈ Ω.
Then there is a unique ﬁxed point of K, x
∗
∈ Ω, and the iteration deﬁned by
(4.3) converges qlinearly to x
∗
with qfactor γ for all initial iterates x
0
∈ Ω.
Proof. Let x
0
∈ Ω. Note that ¦x
n
¦ ⊂ Ω because x
0
∈ Ω and K(x) ∈ Ω
whenever x ∈ Ω. The sequence ¦x
n
¦ remains bounded since for all i ≥ 1
x
i+1
−x
i
 = K(x
i
) −K(x
i−1
) ≤ γx
i
−x
i−1
 . . . ≤ γ
i
x
1
−x
0
,
and therefore
x
n
−x
0
 = 
n−1
i=0
x
i+1
−x
i

≤
n−1
i=0
x
i+1
−x
i
 ≤ x
1
−x
0

n−1
i=0
γ
i
≤ x
1
−x
0
/(1 −γ).
Now, for all n, k ≥ 0,
x
n+k
−x
n
 = K(x
n+k−1
) −K(x
n−1
)
≤ γx
n+k−1
−x
n−1

≤ γK(x
n+k−2
) −K(x
n−2
)
≤ γ
2
x
n+k−2
−x
n−2
 ≤ . . . ≤ γ
n
x
k
−x
0

≤ γ
n
x
1
−x
0
/(1 −γ).
Hence
lim
n,k→∞
x
n+k
−x
n
 = 0
and therefore the sequence ¦x
n
¦ is a Cauchy sequence and has a limit x
∗
[162].
If K has two ﬁxed points x
∗
and y
∗
in Ω, then
x
∗
−y
∗
 = K(x
∗
) −K(y
∗
) ≤ γx
∗
−y
∗
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
68 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Since γ < 1 this can only hold if x
∗
−y
∗
 = 0, i. e. x
∗
= y
∗
. Hence the ﬁxed
point is unique.
Finally we note that
x
n+1
−x
∗
 = K(x
n
) −K(x
∗
) ≤ γx
n
−x
∗
,
which completes the proof.
One simple application of this theorem is to numerical integration of
ordinary diﬀerential equations by implicit methods. For example, consider
the initial value problem
y
= f(y), y(t
0
) = y
0
and its solution by the backward Euler method with time step h,
y
n+1
−y
n
= hf(y
n+1
). (4.4)
In (4.4) y
k
denotes the approximation of y(t
0
+ kh). Advancing in time from
y
n
to y
n+1
requires solution of the ﬁxedpoint problem
y = K(y) = y
n
+hf(y).
For simplicity we assume that f is bounded and Lipschitz continuous, say
[f(y)[ ≤ M and [f(x) −f(y)[ ≤ M[x −y[
for all x, y.
If hM < 1, we may apply the contraction mapping theorem with Ω = R
since for all x, y
K(x) −K(y) = h[f(x) −f(y)[ ≤ hM[x −y[.
From this we conclude that for h suﬃciently small we can integrate forward
in time and that the time step h can be chosen independently of y. This time
step may well be too small to be practical, and the methods based on Newton’s
method which we present in the subsequent chapters are used much more often
[83]. This type of analysis was used by Picard to prove existence of solutions
of initial value problems [152]. Nonlinear variations of the classical stationary
iterative methods such as Gauss–Seidel have also been used; see [145] for a
more complete discussion of these algorithms.
4.3. The standard assumptions
We will make the standard assumptions on F.
Assumption 4.3.1.
1. Equation 4.1 has a solution x
∗
.
2. F
: Ω → R
N×N
is Lipschitz continuous with Lipschitz constant γ.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
FIXEDPOINT ITERATION 69
3. F
(x
∗
) is nonsingular.
These assumptions can be weakened [108] without sacriﬁcing convergence
of the methods we consider here. However the classical result on quadratic
convergence of Newton’s method requires them and we make them throughout.
Exercise 5.7.1 illustrates one way to weaken the standard assumptions.
Throughout this part, we will always denote a root of F by x
∗
. We let
B(r) denote the ball of radius r about x
∗
B(r) = ¦x[ e < r¦,
where
e = x −x
∗
.
The notation introduced above will be used consistently. If x
n
is the nth iterate
of a sequence, e
n
= x
n
−x
∗
is the error in that iterate.
Lemma 4.3.1 is an important consequence of the standard assumptions.
Lemma 4.3.1. Assume that the standard assumptions hold. Then there is
δ > 0 so that for all x ∈ B(δ)
F
(x) ≤ 2F
(x
∗
), (4.5)
F
(x)
−1
 ≤ 2F
(x
∗
)
−1
, (4.6)
and
F
(x
∗
)
−1

−1
e/2 ≤ F(x) ≤ 2F
(x
∗
)e. (4.7)
Proof. Let δ be small enough so that B(δ) ⊂ Ω. For all x ∈ B(δ) we have
F
(x) ≤ F
(x
∗
) +γe.
Hence (4.5) holds if γδ < F
(x
∗
).
The next result (4.6) is a direct consequence of the Banach Lemma if
I −F
(x
∗
)
−1
F
(x) < 1/2.
This will follow from
δ <
F
(x
∗
)
−1

−1
2γ
since then
I −F
(x
∗
)
−1
F
(x) = F
(x
∗
)
−1
(F
(x
∗
) −F
(x))
≤ γF
(x
∗
)
−1
e ≤ γδF
(x
∗
)
−1
 < 1/2.
(4.8)
To prove the ﬁnal inequality (4.7), we note that if x ∈ B(δ) then x
∗
+te ∈
B(δ) for all 0 ≤ t ≤ 1. We use (4.5) and Theorem 4.0.1 to obtain
F(x) ≤
_
1
0
F
(x
∗
+te)e dt ≤ 2F
(x
∗
)e
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
70 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
which is the right inequality in (4.7).
To prove the left inequality in (4.7) note that
F
(x
∗
)
−1
F(x) = F
(x
∗
)
−1
_
1
0
F
(x
∗
+te)e dt
= e −
_
1
0
(I −F
(x
∗
)
−1
F
(x
∗
+te))e dt,
and hence, by (4.8)
F
(x
∗
)
−1
F(x) ≥ e(1 −
_
1
0
I −F
(x
∗
)
−1
F
(x
∗
+te)dt) ≥ e/2.
Therefore
e/2 ≤ F
(x
∗
)
−1
F(x) ≤ F
(x
∗
)
−1
F(x),
which completes the proof.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 5
Newton’s Method
5.1. Local convergence of Newton’s method
We will describe iterative methods for nonlinear equations in terms of the
transition from a current iterate x
c
to a new iterate x
+
. In this language,
Newton’s method is
x
+
= x
c
−F
(x
c
)
−1
F(x
c
). (5.1)
We may also view x
+
as the root of the twoterm Taylor expansion or linear
model of F about x
c
M
c
(x) = F(x
c
) +F
(x
c
)(x −x
c
).
In the context of single systems this method appeared in the 17th century
[140], [156], [13], [137].
The convergence result on Newton’s method follows from Lemma 4.3.1.
Theorem 5.1.1. Let the standard assumptions hold. Then there are K > 0
and δ > 0 such that if x
c
∈ B(δ) the Newton iterate from x
c
given by (5.1)
satisﬁes
e
+
 ≤ Ke
c

2
. (5.2)
Proof. Let δ be small enough so that the conclusions of Lemma 4.3.1 hold.
By Theorem 4.0.1
e
+
= e
c
−F
(x
c
)
−1
F(x
c
) = F
(x
c
)
−1
_
1
0
(F
(x
c
) −F
(x
∗
+te
c
))e
c
dt.
By Lemma 4.3.1 and the Lipschitz continuity of F
e
+
 ≤ (2F
(x
∗
)
−1
)γe
c

2
/2.
This completes the proof of (5.2) with K = γF
(x
∗
)
−1
.
The proof of convergence of the complete Newton iteration will be complete
if we reduce δ if needed so that Kδ < 1.
Theorem 5.1.2. Let the standard assumptions hold. Then there is δ such
that if x
0
∈ B(δ) the Newton iteration
x
n+1
= x
n
−F
(x
n
)
−1
F(x
n
)
converges qquadratically to x
∗
.
71
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
72 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Proof. Let δ be small enough so that the conclusions of Theorem 5.1.1
hold. Reduce δ if needed so that Kδ = η < 1. Then if n ≥ 0 and x
n
∈ B(δ),
Theorem 5.1.1 implies that
e
n+1
 ≤ Ke
n

2
≤ ηe
n
 < e
n
 (5.3)
and hence x
n+1
∈ B(ηδ) ⊂ B(δ). Therefore if x
n
∈ B(δ) we may continue the
iteration. Since x
0
∈ B(δ) by assumption, the entire sequence ¦x
n
¦ ⊂ B(δ).
(5.3) then implies that x
n
→ x
∗
qquadratically.
The assumption that the initial iterate be “suﬃciently near” the solution
(x
0
∈ B(δ)) may seem artiﬁcial at ﬁrst look. There are, however, many
situations in which the initial iterate is very near the root. Two examples are
implicit integration of ordinary diﬀerential equations and diﬀerential algebraic
equations, [105], [83], [16], where the initial iterate is derived from the solution
at the previous time step, and solution of discretizations of partial diﬀerential
equations, where the initial iterate is an interpolation of a solution from a
coarser computational mesh [99], [126]. Moreover, when the initial iterate is
far from the root and methods such as those discussed in Chapter 8 are needed,
local convergence results like those in this and the following two chapters
describe the terminal phase of the iteration.
5.2. Termination of the iteration
We can base our termination decision on Lemma 4.3.1. If x ∈ B(δ), where δ
is small enough so that the conclusions of Lemma 4.3.1 hold, then if F
(x
∗
) is
well conditioned, we may terminate the iteration when the relative nonlinear
residual F(x)/F(x
0
) is small. We have, as a direct consequence of applying
Lemma 4.3.1 twice, a nonlinear form of Lemma 1.1.1.
Lemma 5.2.1. Assume that the standard assumptions hold. Let δ > 0 be
small enough so that the conclusions of Lemma 4.3.1 hold for x ∈ B(δ). Then
for all x ∈ B(δ)
e
4e
0
κ(F
(x
∗
))
≤
F(x)
F(x
0
)
≤
4κ(F
(x
∗
))e
e
0

,
where κ(F
(x
∗
)) = F
(x
∗
)F
(x
∗
)
−1
 is the condition number of F
(x
∗
)
relative to the norm  .
From Lemma 5.2.1 we conclude that if F
(x
∗
) is well conditioned, the size
of the relative nonlinear residual is a good indicator of the size of the error.
However, if there is error in the evaluation of F or the initial iterate is near a
root, a termination decision based on the relative residual may be made too
late in the iteration or, as we will see in the next section, the iteration may
not terminate at all. We raised this issue in the context of linear equations
in Chapter 1 when we compared (1.2) and (1.4). In the nonlinear case, there
is no “righthand side” to use as a scaling factor and we must balance the
relative and absolute size of the nonlinear residuals in some other way. In all
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 73
of our MATLAB codes and algorithms for nonlinear equations our termination
criterion is to stop the iteration if
F(x) ≤ τ
r
F(x
0
) +τ
a
, (5.4)
where the relative error tolerance τ
r
and absolute error tolerance τ
a
are input
to the algorithm. Combinations of relative and absolute error tolerances are
commonly used in numerical methods for ordinary diﬀerential equations or
diﬀerential algebraic equations [16], [21], [23], [151].
Another way to decide whether to terminate is to look at the Newton step
s = −F
(x
c
)
−1
F(x
c
) = x
+
−x
c
,
and terminate the iteration when s is suﬃciently small. This criterion is
based on Theorem 5.1.1, which implies that
e
c
 = s +O(e
c

2
).
Hence, near the solution s and e
c
are essentially the same size.
For methods other than Newton’s method, the relation between the step
and the error is less clear. If the iteration is qlinearly convergent, say, then
e
+
 ≤ σe
c

implies that
(1 −σ)e
c
 ≤ s ≤ (1 +σ)e
c
.
Hence the step is a reliable indicator of the error provided σ is not too near
1. The diﬀerential equations codes discussed in [16], [21], [23], and [151], make
use of this in the special case of the chord method.
However, in order to estimate e
c
 this way, one must do most of the work
needed to compute x
+
, whereas by terminating on small relative residuals or
on a condition like (5.4) the decision to stop the iteration can be made before
computation and factorization of F
(x
c
). If computation and factorization of
the Jacobian are inexpensive, then termination on small steps becomes more
attractive.
5.3. Implementation of Newton’s method
In this section we assume that direct methods will be used to solve the linear
equation for the Newton step. Our examples and discussions of operation
counts make the implicit assumption that the Jacobian is dense. However,
the issues for sparse Jacobians when direct methods are used to solve linear
systems are very similar. In Chapter 6 we consider algorithms in which iterative
methods are used to solve the equation for the Newton step.
In order to compute the Newton iterate x
+
from a current point x
c
one
must ﬁrst evaluate F(x
c
) and decide whether to terminate the iteration. If
one decides to continue, the Jacobian F
(x
c
) must be computed and factored.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
74 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Then the step is computed as the solution of F
(x
c
)s = −F(x
c
) and the iterate
is updated x
+
= x
c
+s. Of these steps, the evaluation and factorization of F
are the most costly. Factorization of F
in the dense case costs O(N
3
) ﬂoating
point operations. Evaluation of F
by ﬁnite diﬀerences should be expected to
cost N times the cost of an evaluation of F because each column in F
requires
an evaluation of F to form the diﬀerence approximation. Hence the cost of a
Newton step may be roughly estimated as N + 1 evaluations of F and O(N
3
)
ﬂoatingpoint operations. In many cases F
can be computed more eﬃciently,
accurately, and directly than with diﬀerences and the analysis above for the
cost of a Newton iterate is very pessimistic. See Exercise 5.7.21 for an example.
For general nonlinearities or general purpose codes such as the MATLAB code
nsol from the collection, there may be little alternative to diﬀerence Jacobians.
In the future, automatic diﬀerentiation (see [94] for a collection of articles on
this topic) may provide such an alternative.
For deﬁniteness in the description of Algorithm newton we use an LU
factorization of F
. Any other appropriate factorization such as QR or
Cholesky could be used as well. The inputs of the algorithm are the initial
iterate x, the nonlinear map F, and a vector of termination tolerances
τ = (τ
r
, τ
a
) ∈ R
2
.
Algorithm 5.3.1. newton(x, F, τ)
1. r
0
= F(x)
2. Do while F(x) > τ
r
r
0
+τ
a
(a) Compute F
(x)
(b) Factor F
(x) = LU.
(c) Solve LUs = −F(x)
(d) x = x +s
(e) Evaluate F(x).
One approach to reducing the cost of items 2a and item 2b in the Newton
iteration is to move them outside of the main loop. This means that the linear
approximation of F(x) = 0 that is solved at each iteration has derivative
determined by the initial iterate.
x
+
= x
c
−F
(x
0
)
−1
F(x
c
).
This method is called the chord method. The inputs are the same as for
Newton’s method.
Algorithm 5.3.2. chord(x, F, τ)
1. r
0
= F(x)
2. Compute F
(x)
3. Factor F
(x) = LU.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 75
4. Do while F(x) > τ
r
r
0
+τ
a
(a) Solve LUs = −F(x)
(b) x = x +s
(c) Evaluate F(x).
The only diﬀerence in implementation from Newton’s method is that the
computation and factorization of the Jacobian are done before the iteration
is begun. The diﬀerence in the iteration itself is that an approximation to
F
(x
c
) is used. Similar diﬀerences arise if F
(x
c
) is numerically approximated
by diﬀerences. We continue in the next section with an analysis of the eﬀects
of errors in the function and Jacobian in a general context and then consider
the chord method and diﬀerence approximation to F
as applications.
5.4. Errors in the function and derivative
Suppose that F and F
are computed inaccurately so that F + and F
+ ∆
are used instead of F and F
in the iteration. If ∆ is suﬃciently small, the
resulting iteration can return a result that is an O() accurate approximation
to x
∗
. This is diﬀerent from convergence and was called “local improvement”
in [65]. These issues are discussed in [201] as well. If, for example, is entirely
due to ﬂoatingpoint roundoﬀ, there is no reason to expect that F(x
n
) will
ever be smaller than in general. We will refer to this phase of the iteration
in which the nonlinear residual is no longer being reduced as stagnation of the
iteration.
Theorem 5.4.1. Let the standard assumptions hold. Then there are
¯
K > 0, δ > 0, and δ
1
> 0 such that if x
c
∈ B(δ) and ∆(x
c
) < δ
1
then
x
+
= x
c
−(F
(x
c
) + ∆(x
c
))
−1
(F(x
c
) +(x
c
))
is deﬁned (i.e., F
(x
c
) + ∆(x
c
) is nonsingular) and satisﬁes
e
+
 ≤
¯
K(e
c

2
+∆(x
c
)e
c
 +(x
c
)). (5.5)
Proof. Let δ be small enough so that the conclusions of Lemma 4.3.1 and
Theorem 5.1.1 hold. Let
x
N
+
= x
c
−F
(x
c
)
−1
F(x
c
)
and note that
x
+
= x
N
+
+ (F
(x
c
)
−1
−(F
(x
c
) + ∆(x
c
))
−1
)F(x
c
) −(F
(x
c
) + ∆(x
c
))
−1
(x
c
).
Lemma 4.3.1 and Theorem 5.1.1 imply
e
+
 ≤ Ke
c

2
+ 2F
(x
c
)
−1
−(F
(x
c
) + ∆(x
c
))
−1
)F
(x
∗
)e
c

+(F
(x
c
) + ∆(x
c
))
−1
(x
c
).
(5.6)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
76 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
If
∆(x
c
) ≤ F
(x
∗
)
−1

−1
/4
then Lemma 4.3.1 implies that
∆(x
c
) ≤ F
(x
c
)
−1

−1
/2
and the Banach Lemma implies that F
(x
c
) + ∆(x
c
) is nonsingular and
(F
(x
c
) + ∆(x
c
))
−1
 ≤ 2F
(x
c
)
−1
 ≤ 4F
(x
∗
)
−1
.
Hence
F
(x
c
)
−1
−(F
(x
c
) + ∆(x
c
))
−1
 ≤ 8F
(x
∗
)
−1

2
∆(x
c
).
We use these estimates and (5.6) to obtain
e
+
 ≤ Ke
c

2
+ 16F
(x
∗
)
−1

2
F
(x
∗
)∆(x
c
)e
c
 + 4F
(x
∗
)
−1
(x
c
).
Setting
¯
K = K + 16F
(x
∗
)
−1

2
F
(x
∗
) + 4F
(x
∗
)
−1

completes the proof.
The remainder of this section is devoted to applications of Theorem 5.4.1.
5.4.1. The chord method. Recall that the chord method is given by
x
+
= x
c
−F
(x
0
)
−1
F(x
c
).
In the language of Theorem 5.4.1
(x
c
) = 0, ∆(x
c
) = F
(x
0
) −F
(x
c
).
Hence, if x
c
, x
0
∈ B(δ) ⊂ Ω
∆(x
c
) ≤ γx
0
−x
c
 ≤ γ(e
0
 +e
c
). (5.7)
We apply Theorem 5.4.1 to obtain the following result.
Theorem 5.4.2. Let the standard assumptions hold. Then there are
K
C
> 0 and δ > 0 such that if x
0
∈ B(δ) the chord iterates converge qlinearly
to x
∗
and
e
n+1
 ≤ K
C
e
0
e
n
. (5.8)
Proof. Let δ be small enough so that B(δ) ⊂ Ω and the conclusions of
Theorem 5.4.1 hold. Assume that x
n
∈ B(δ). Combining (5.7) and (5.5)
implies
e
n+1
 ≤
¯
K(e
n
(1 +γ) +γe
0
)e
n
 ≤
¯
K(1 + 2γ)δe
n
.
Hence if δ is small enough so that
¯
K(1 + 2γ)δ = η < 1
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 77
then the chord iterates converge qlinearly to x
∗
. Qlinear convergence implies
that e
n
 < e
0
 and hence (5.8) holds with K
C
=
¯
K(1 + 2γ).
Another variation of the chord method is
x
+
= x
c
−A
−1
F(x
c
),
where A ≈ F
(x
∗
). Methods of this type may be viewed as preconditioned
nonlinear Richardson iteration. Since
∆(x
c
) = A−F
(x
c
) ≤ A−F
(x
∗
) +F
(x
∗
) −F
(x
c
),
if x
c
∈ B(δ) ⊂ Ω then
∆(x
c
) ≤ A−F
(x
∗
) +γe
c
 ≤ A−F
(x
∗
) +γδ.
Theorem 5.4.3. Let the standard assumptions hold. Then there are
K
A
> 0, δ > 0, and δ
1
> 0, such that if x
0
∈ B(δ) and A−F
(x
∗
) < δ
1
then
the iteration
x
n+1
= x
n
−A
−1
F(x
n
)
converges qlinearly to x
∗
and
e
n+1
 ≤ K
A
(e
0
 +A−F
(x
∗
))e
n
. (5.9)
5.4.2. Approximate inversion of F
. Another way to implement chord
type methods is to provide an approximate inverse of F
. Here we replace
F
(x)
−1
by B(x), where the action of B on a vector is less expensive to compute
than a solution using the LU factorization. Rather than express the iteration
in terms of B
−1
(x) −F
(x) and using Theorem 5.4.3 one can proceed directly
from the deﬁnition of approximate inverse.
Note that if B is constant (independent of x) then the iteration
x
+
= x
c
−BF(x
c
)
can be viewed as a preconditioned nonlinear Richardson iteration.
We have the following result.
Theorem 5.4.4. Let the standard assumptions hold. Then there are
K
B
> 0, δ > 0, and δ
1
> 0, such that if x
0
∈ B(δ) and the matrixvalued
function B(x) satisﬁes
I −B(x)F
(x
∗
) = ρ(x) < δ
1
(5.10)
for all x ∈ B(δ) then the iteration
x
n+1
= x
n
−B(x
n
)F(x
n
)
converges qlinearly to x
∗
and
e
n+1
 ≤ K
B
(ρ(x
n
) +e
n
)e
n
. (5.11)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
78 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Proof. On one hand, this theorem could be regarded as a corollary of the
Banach Lemma and Theorem 5.4.1. We give a direct proof.
First, by (5.10) we have
B(x) = B(x)F
(x
∗
)F
(x
∗
)
−1
 ≤ B(x)F
(x
∗
)F
(x
∗
)
−1

≤ M
B
= (1 +δ
1
)F
(x
∗
)
−1
.
(5.12)
Using (5.12) and
e
+
= e
c
−B(x
c
)F(x
c
) =
_
1
0
(I −B(x
c
)F
(x
∗
+te
c
))e
c
dt
= (I −B(x
c
)F
(x
∗
))e
c
+B(x
c
)
_
1
0
(F
(x
∗
) −F
(x
∗
+te
c
))e
c
dt
we have
e
+
 ≤ ρ(x
c
)e
c
 +M
B
γe
c

2
/2.
This completes the proof with K
B
= 1 +M
B
γ/2.
5.4.3. The Shamanskii method. Alternation of a Newton step with a
sequence of chord steps leads to a class of highorder methods, that is, methods
that converge qsuperlinearly with qorder larger than 2. We follow [18] and
name this method for Shamanskii [174], who considered the ﬁnite diﬀerence
Jacobian formulation of the method from the point of view of eﬃciency. The
method itself was analyzed in [190]. Other highorder methods, especially for
problems in one dimension, are described in [190] and [146].
We can describe the transition from x
c
to x
+
by
y
1
= x
c
−F
(x
c
)
−1
F(x
c
),
y
j+1
= y
j
−F
(x
c
)
−1
F(y
j
) for 1 ≤ j ≤ m−1,
x
+
= y
m
.
(5.13)
Note that m = 1 is Newton’s method and m = ∞ is the chord method with
¦y
j
¦ playing the role of the chord iterates.
Algorithmically a second loop surrounds the factor/solve step. The inputs
for the Shamanskii method are the same as for Newton’s method or the chord
method except for the addition of the parameter m. Note that we can overwrite
x
c
with the sequence of y
j
’s. Note that we apply the termination test after
computation of each y
j
.
Algorithm 5.4.1. sham(x, F, τ, m)
1. r
0
= F(x)
2. Do while F(x) > τ
r
r
0
+τ
a
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 79
(a) Compute F
(x)
(b) Factor F
(x) = LU.
(c) for j = 1, . . . m
i. Solve LUs = −F(x)
ii. x = x +s
iii. Evaluate F(x).
iv. If F(x) ≤ τ
r
r
0
+τ
a
exit.
The convergence result is a simple application of Theorem 5.4.2.
Theorem 5.4.5. Let the standard assumptions hold and let m ≥ 1 be
given. Then there are K
S
> 0 and δ > 0 such that if x
0
∈ B(δ) the Shamanskii
iterates converge qsuperlinearly to x
∗
with qorder m+ 1 and
e
n+1
 ≤ K
S
e
n

m+1
. (5.14)
Proof. Let δ be small enough so that B(δ) ⊂ Ω and that the conclusions
of Theorem 5.4.2 hold. Then if x
n
∈ B(δ) all the intermediate iterates ¦y
j
¦
are in B(δ) by Theorem 5.4.2. In fact, if we set y
1
= x
n
, (5.8) implies that for
1 ≤ j ≤ m
y
j
−x
∗
 ≤ K
C
x
n
−x
∗
y
j−1
−x
∗
 = K
C
e
n
y
j−1
−x
∗
 ≤ . . . ≤ K
j
C
e
n

j+1
.
Hence x
n+1
∈ B(δ). Setting j = m in the above inequality completes the
proof.
The advantage of the Shamanskii method over Newton’s method is
that high qorders can be obtained with far fewer Jacobian evaluations or
factorizations. Optimal choices of m as a function of N can be derived
under assumptions consistent with dense matrix algebra [18] by balancing the
O(N
3
) cost of a matrix factorization with the cost of function and Jacobian
evaluation. The analysis in [18] and the expectation that, if the initial iterate is
suﬃciently accurate, only a small number of iterations will be taken, indicate
that the chord method is usually the best option for very large problems.
Algorithm nsol is based on this idea and using an idea from [114] computes
and factors the Jacobian only until an estimate of the qfactor for the linear
convergence rate is suﬃciently low.
5.4.4. Diﬀerence approximation to F
. Assume that we compute
F(x) + (x) instead of F(x) and attempt to approximate the action of F
on a vector by a forward diﬀerence. We would do this, for example, when
building an approximate Jacobian. Here the jth column of F
(x) is F
(x)e
j
where e
j
is the unit vector with jth component 1 and other components 0.
A forward diﬀerence approximation to F
(x)w would be
F(x +hw) +(x +hw) −F(x) −(x)
h
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
80 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Assume that (x) ≤ ¯ . Then
F
(x)w −
F(x +hw) +(x +hw) −F(x) −(x)
h
= O(h + ¯ /h).
The quantity inside the Oterm is minimized when h =
√
¯ . This means, for
instance, that very small diﬀerence steps h can lead to inaccurate results. In
the special case where (x) is a result of ﬂoatingpoint roundoﬀ in full precision
(¯ ≈ 10
−15
), the analysis here indicates that h ≈ 10
−7
is a reasonable choice.
The choice h ≈ 10
−15
in this case can lead to disaster (try it!). Here we have
made the implicit assumption that x and w are of roughly the same size. If
this is not the case, h should be scaled to reﬂect that. The choice
h = ¯
1/2
x/w
reﬂects all the discussion above.
A more important consequence of this analysis is that the choice of step
size in a forward diﬀerence approximation to the action of the Jacobian on
a vector must take into account the error in the evaluation of F. This can
become very important if part of F is computed from measured data.
If F(x) has already been computed, the cost of a forward diﬀerence
approximation of F
(x)w is an additional evaluation of F (at the point x+hw).
Hence the evaluation of a full ﬁnite diﬀerence Jacobian would cost N function
evaluations, one for each column of the Jacobian. Exploitation of special
structure could reduce this cost.
The forward diﬀerence approximation to the directional derivative is not
a linear operator in w. The reason for this is that the derivative has been
approximated, and so linearity has been lost. Because of this we must
carefully specify what we mean by a diﬀerence approximation to the Jacobian.
Deﬁnition 5.4.1 below speciﬁes the usual choice.
Definition 5.4.1. Let F be deﬁned in a neighborhood of x ∈ R
N
.
(∇
h
F)(x) is the N N matrix whose jth column is given by
(∇
h
F)(x)
j
=
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
F(x +hxe
j
) −F(x)
hx
x ,= 0
F(he
j
) −F(x)
h
x = 0
We make a similar deﬁnition of the diﬀerence approximation of the
directional derivative.
Definition 5.4.2. Let F be deﬁned in a neighborhood of x ∈ R
N
and let
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 81
w ∈ R
N
. We have
D
h
F(x : w) =
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
0, w = 0,
w
F(x +hxw/w) −F(x)
hx
, w, x ,= 0,
w
F(hw/w) −F(x)
h
, x = 0, w ,= 0.
(5.15)
The standard assumptions and the Banach Lemma imply the following
result.
Lemma 5.4.1. Let the standard assumptions hold. Then there are δ, ¯ ,
¯
h >
0 such that if x ∈ B(δ), h ≤
¯
h, and (x) ≤ ¯ for all x ∈ B(δ), then
∇
h
(F +)(x) is nonsingular for all x ∈ B(δ) and there is M
F
> 0 such that
F
(x) −∇
h
(F +)(x) ≤ M
F
(h + ¯ /h).
If we assume that the forward diﬀerence step has been selected so that
h = O(
√
¯ ), then ∆(x) = O(
√
¯ ) by Lemma 5.4.1. Theorem 5.4.1 implies the
following result.
Theorem 5.4.6. Let the standard assumptions hold. Then there are
positive δ, ¯ , and K
D
such that if x
c
∈ B(δ), (x) ≤ ¯ for all x ∈ B(δ),
and there is M
−
> 0 such that
h
c
> M
−
√
¯
then
x
+
= x
c
−∇
hc
(F +)(x
c
)
−1
(F(x
c
) +(x
c
))
satisﬁes
e
+
 ≤ K
D
(¯ + (e
c
 +h
c
)e
c
).
Note that Theorem 5.4.6 does not imply that an iteration will converge to
x
∗
. Even if is entirely due to ﬂoatingpoint roundoﬀ, there is no guarantee
that (x
n
) → 0 as x
n
→ x
∗
. Hence, one should expect the sequence ¦e
n
¦
to stagnate and cease to decrease once e
n
 ≈ ¯ . Therefore in the early
phases of the iteration, while e
n
 >>
√
¯ , the progress of the iteration will
be indistinguishable from Newton’s method as implemented with an exact
Jacobian. When e
n
 ≤
√
¯ , Theorem 5.4.6 says that e
n+1
 = O(¯ ). Hence
one will see quadratic convergence until the error in F admits no further
reduction.
A diﬀerence approximation to a full Jacobian does not use any special
structure of the Jacobian and will probably be less eﬃcient than a hand coded
analytic Jacobian. If information on the sparsity pattern of the Jacobian is
available it is sometimes possible [47], [42], to evaluate ∇
h
F with far fewer than
N evaluations of F. If the sparsity pattern is such that F
can be reduced to
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
82 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
a block triangular form, the entire solution process can take advantage of this
structure [59].
For problems in which the Lipschitz constant of F
is large, the error in
the diﬀerence approximation will be large, too. Moreover, D
h
F(x : w) is not,
in general, a linear function of w and it is usually the case that
(∇
h
F(x))w ,= D
h
F(x : w).
If F
(x
∗
) is ill conditioned or the Lipschitz constant of F
is large, the
diﬀerence between them may be signiﬁcant and a diﬀerence approximation to
a directional derivative, which uses the size of w, is likely to be more accurate
than ∇
h
F(x)w.
While we used full matrix approximations to numerical Jacobians in all
the examples reported here, they should be used with caution.
5.4.5. The secant method. The results in this section are for single
equations f(x) = 0, where f is a realvalued function of one real variable.
We assume for the sake of simplicity that there are no errors in the evaluation
of f, i.e., (x) = 0. The standard assumptions, then, imply that f
(x
∗
) ,= 0.
There is no reason to use a diﬀerence approximation to f
for equations in a
single variable because one can use previously computed data to approximate
f
(x
n
) by
a
n
=
f(x
n
) −f(x
n−1
)
x
n
−x
n−1
.
The resulting method is called the secant method and is due to Newton
[140], [137]. We will prove that the secant method is locally qsuperlinearly
convergent if the standard assumptions hold.
In order to start, one needs two approximations to x
∗
, x
0
and x
−1
. The
local convergence theory can be easily described by Theorem 5.4.1. Let x
c
be
the current iterate and x
−
the previous iterate. We have (x
c
) = 0 and
∆(x
c
) =
f(x
c
) −f(x
−
)
x
c
−x
−
−f
(x
c
). (5.16)
We have the following theorem.
Theorem 5.4.7. Let the standard assumptions hold with N = 1. Then
there is δ > 0 such that if x
0
, x
−1
∈ B(δ) and x
0
,= x
−1
then the secant iterates
converge qsuperlinearly to x
∗
.
Proof. Let δ be small enough so that B(δ) ⊂ Ω and the conclusions of
Theorem 5.4.1 hold. Assume that x
−1
, x
0
, . . . , x
n
∈ B(δ). If x
n
= x
∗
for some
ﬁnite n, we are done. Otherwise x
n
,= x
n−1
. If we set s = x
n
−x
n−1
we have
by (5.16) and the fundamental theorem of calculus
∆(x
n
) =
_
1
0
(f
(x
n−1
+ts) −f
(x
n
)) dt
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 83
and hence
[∆(x
n
)[ ≤ γ[x
n−1
−x
n
[/2 ≤
γ([e
n
[ +[e
n−1
[)
2
.
Applying Theorem 5.4.1 implies
[e
n+1
[ ≤
¯
K((1 +γ/2)[e
n
[
2
+γ[e
n
[[e
n−1
[/2). (5.17)
If we reduce δ so that
¯
K(1 +γ)δ = η < 1
then x
n
→ x
∗
qlinearly. Therefore
[e
n+1
[
[e
n
[
≤
¯
K((1 +γ/2)[e
n
[ +γ[e
n−1
[/2) → 0
as n → ∞. This completes the proof.
The secant method and the classical bisection (see Exercise 5.7.8) method
are the basis for the popular method of Brent [17] for the solution of single
equations.
5.5. The Kantorovich Theorem
In this section we present the result of Kantorovich [107], [106], which states
that if the standard assumptions “almost” hold at a point x
0
, then there
is a root and the Newton iteration converges rquadratically to that root.
This result is of use, for example, in proving that discretizations of nonlinear
diﬀerential and integral equations have solutions that are near to that of the
continuous problem.
We require the following assumption.
Assumption 5.5.1. There are β, η, ¯ r, and γ with βηγ ≤ 1/2 and x
0
∈ R
N
such that
1. F is diﬀerentiable at x
0
,
F
(x
0
)
−1
 ≤ β, and F
(x
0
)
−1
F(x
0
) ≤ η.
2. F
is Lipschitz continuous with Lipschitz constant γ in a ball of radius
¯ r ≥ r
−
about x
0
where
r
−
=
1 −
√
1 −2βηγ
βγ
.
We will let B
0
be the closed ball of radius r
−
about x
0
B
0
= ¦x[ x −x
0
 ≤ r
−
¦.
We do not prove the result in its full generality and refer to [106], [145],
[57], and [58] for a complete proof and for discussions of related results.
However, we give a simpliﬁed version of a Kantorovichlike result for the chord
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
84 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
method to show how existence of a solution can be derived from estimates
on the function and Jacobian. Unlike the chord method results in [106],
[145], and [57], Theorem 5.5.1 assumes a bit more than Assumption 5.5.1 but
has a simple proof that uses only the contraction mapping theorem and the
fundamental theorem of calculus and obtains qlinear convergence instead of
rlinear convergence.
Theorem 5.5.1. Let Assumption 5.5.1 hold and let
βγη < 1/2. (5.18)
Then there is a unique root x
∗
of F in B
0
, the chord iteration with x
0
as the
initial iterate converges to x
∗
qlinearly with qfactor βγr
−
, and x
n
∈ B
0
for
all n. Moreover x
∗
is the unique root of F in the ball of radius
min
_
¯ r,
1 +
√
1 −2βηγ
βγ
_
about x
0
.
Proof. We will show that the map
φ(x) = x −F
(x
0
)
−1
F(x)
is a contraction on the set
o = ¦x[ x −x
0
 ≤ r
−
¦.
By the fundamental theorem of calculus and Assumption 5.5.1 we have for
all x ∈ o
F
(x
0
)
−1
F(x) = F
(x
0
)
−1
F(x
0
)
+F
(x
0
)
−1
_
1
0
F
(x
0
+t(x −x
0
))(x −x
0
) dt
= F
(x
0
)
−1
F(x
0
) +x −x
0
+F
(x
0
)
−1
_
1
0
(F
(x
0
+t(x −x
0
)) −F
(x
0
))(x −x
0
) dt
(5.19)
and
φ(x) −x
0
= −F
(x
0
)
−1
F(x
0
)
−F
(x
0
)
−1
_
1
0
(F
(x
0
+t(x −x
0
)) −F
(x
0
))(x −x
0
) dt.
Hence,
φ(x) −x
0
 ≤ η +βγr
2
−
/2 = r
−
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 85
and φ maps o into itself.
To show that φ is a contraction on o and to prove the qlinear convergence
we will show that
φ(x) −φ(y) ≤ βγr
−
x −y
for all x, y ∈ o. If we do this, the contraction mapping theorem (Theo
rem 4.2.1) will imply that there is a unique ﬁxed point x
∗
of φ in o and
that x
n
→ x
∗
qlinearly with qfactor βγr
−
. Clearly x
∗
is a root of F because
it is a ﬁxed point of φ. We know that βγr
−
< 1 by our assumption that
βηγ < 1/2.
Note that for all x ∈ B
0
φ
(x) = I −F
(x
0
)
−1
F
(x) = F
(x
0
)
−1
(F
(x
0
) −F
(x))
and hence
φ
(x) ≤ βγx −x
0
 ≤ βγr
−
< 1.
Therefore, for all x, y ∈ B
0
,
φ(x) −φ(y) =
_
_
_
_
_
1
0
φ
(y +t(x −y))(x −y) dt
_
_
_
_
≤ βγr
−
x −y.
This proves the convergence result and the uniqueness of x
∗
in B
0
.
To prove the remainder of the uniqueness assertion let
r
+
=
1 +
√
1 −2βηγ
βγ
and note that r
+
> r
−
because βηγ < 1/2. If ¯ r = r
−
, then we are done by the
uniqueness of x
∗
in B
0
. Hence we may assume that ¯ r > r
−
. We must show
that if x is such that
r
−
< x −x
0
 < min(¯ r, r
+
)
then F(x) ,= 0. Letting r = x −x
0
 we have by (5.19)
F
(x
0
)
−1
F(x) ≥ r −η −βγr
2
/2 > 0
because r
−
< r < r
+
. This completes the proof.
The Kantorovich theorem is more precise than Theorem 5.5.1 and has a
very diﬀerent proof. We state the theorem in the standard way [145], [106],
[63], using the rquadratic estimates from [106].
Theorem 5.5.2. Let Assumption 5.5.1 hold. Then there is a unique root
x
∗
of F in B
0
, the Newton iteration with x
0
as the initial iterate converges to
x
∗
, and x
n
∈ B
0
for all n. x
∗
is the unique root of F in the ball of radius
min
_
¯ r,
1 +
√
1 −2βηγ
βγ
_
about x
0
and the errors satisfy
e
n
 ≤
(2βηγ)
2
n
2
n
βγ
. (5.20)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
86 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
If βηγ < 1/2 then (5.20) implies that the convergence is rquadratic (see
Exercise 5.7.16).
5.6. Examples for Newton’s method
In the collection of MATLAB codes we provide an implementation nsol
of Newton’s method, the chord method, the Shamanskii method, or a
combination of all of them based on the reduction in the nonlinear residual.
This latter option decides if the Jacobian should be recomputed based on the
ratio of successive residuals. If
F(x
+
)/F(x
c
)
is below a given threshold, the Jacobian is not recomputed. If the ratio is too
large, however, we compute and factor F
(x
+
) for use in the subsequent chord
steps. In addition to the inputs of Algorithm sham, a threshold for the ratio
of successive residuals is also input. The Jacobian is recomputed and factored
if either the ratio of successive residuals exceeds the threshold 0 < ρ < 1 or
the number of iterations without an update exceeds m. The output of the
MATLAB implementation of nsol is the solution, the history of the iteration
stored as the vector of l
∞
norms of the nonlinear residuals, and an error ﬂag.
nsol uses diffjac to approximate the Jacobian by ∇
h
F with h = 10
−7
.
While our examples have dense Jacobians, the ideas in the implementation
of nsol also apply to situations in which the Jacobian is sparse and direct
methods for solving the equation for the Newton step are necessary. One would
still want to minimize the number of Jacobian computations (by the methods
of [47] or [42], say) and sparse matrix factorizations. In Exercise 5.7.25, which
is not easy, the reader is asked to create a sparse form of nsol.
Algorithm 5.6.1. nsol(x, F, τ, m, ρ)
1. r
c
= r
0
= F(x)
2. Do while F(x) > τ
r
r
0
+τ
a
(a) Compute ∇
h
F(x) ≈ F
(x)
(b) Factor ∇
h
F(x) = LU.
(c) i
s
= 0, σ = 0
(d) Do While i
s
< m and σ ≤ ρ
i. i
s
= i
s
+ 1
ii. Solve LUs = −F(x)
iii. x = x +s
iv. Evaluate F(x)
v. r
+
= F(x), σ = r
+
/r
c
, r
c
= r
+
vi. If F(x) ≤ τ
r
r
0
+τ
a
exit.
In the MATLAB code the iteration is terminated with an error condition
if σ ≥ 1 at any point. This indicates that the local convergence analysis in this
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 87
section is not applicable. In Chapter 8 we discuss how to recover from this.
nsol is based on the assumption that a diﬀerence Jacobian will be used. In
Exercise 5.7.21 you are asked to modify the code to accept an analytic Jacobian.
Many times the Jacobian can be computed eﬃciently by reusing results that
are already available from the function evaluation. The parameters m and ρ
are assigned the default values of 1000 and .5. With these default parameters
the decision to update the Jacobian is made based entirely on the reduction in
the norm of the nonlinear residual. nsol allows the user to specify a maximum
number of iterations; the default is 40.
The Chandrasekhar Hequation. The Chandrasekhar Hequation, [41],
[30],
F(H)(µ) = H(µ) −
_
1 −
c
2
_
1
0
µH(ν) dν
µ +ν
_
−1
= 0, (5.21)
is used to solve exit distribution problems in radiative transfer.
We will discretize the equation with the composite midpoint rule. Here we
approximate integrals on [0, 1] by
_
1
0
f(µ) dµ ≈
1
N
N
j=1
f(µ
j
),
where µ
i
= (i −1/2)/N for 1 ≤ i ≤ N. The resulting discrete problem is
F(x)
i
= x
i
−
_
_
1 −
c
2N
N
j=1
µ
i
x
j
µ
i
+µ
j
_
_
−1
. (5.22)
It is known [132] that both (5.21) and the discrete analog (5.22) have two
solutions for c ∈ (0, 1). Only one of these solutions has physical meaning,
however. Newton’s method, with the 0 function or the constant function with
value 1 as initial iterate, will ﬁnd the physically meaningful solution [110].
The standard assumptions hold near either solution [110] for c ∈ [0, 1). The
discrete problem has its own physical meaning [41] and we will attempt to
solve it to high accuracy. Because H has a singularity at µ = 0, the solution of
the discrete problem is not even a ﬁrstorder accurate approximation to that
of the continuous problem.
In the computations reported here we set N = 100 and c = .9. We used the
function identically one as initial iterate. We computed all Jacobians with the
MATLAB code diffjac. We set the termination parameters τ
r
= τ
a
= 10
−6
.
We begin with a comparison of Newton’s method and the chord method.
We present some iteration statistics in both tabular and graphical form. In
Table 5.1 we tabulate the iteration counter n, F(x
n
)
∞
/F(x
0
)
∞
, and the
ratio
R
n
= F(x
n
)
∞
/F(x
n−1
)
∞
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
88 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Table 5.1
Comparison of Newton and chord iteration.
n F(x
n
)
∞
/F(x
0
)
∞
R
n
F(x
n
)
∞
/F(x
0
)
∞
R
n
0 1.000e+00 1.000e+00
1 1.480e01 1.480e01 1.480e01 1.480e01
2 2.698e03 1.823e02 3.074e02 2.077e01
3 7.729e07 2.865e04 6.511e03 2.118e01
4 1.388e03 2.132e01
5 2.965e04 2.136e01
6 6.334e05 2.136e01
7 1.353e05 2.136e01
8 2.891e06 2.136e01
for n ≥ 1 for both Newton’s method and the chord method. In Fig. 5.1
the solid curve is a plot of F(x
n
)
∞
/F(x
0
)
∞
for Newton’s method and
the dashed curve a plot of F(x
n
)
∞
/F(x
0
)
∞
for the chord method. The
concave iteration track is the signature of superlinear convergence, the linear
track indicates linear convergence. See Exercise 5.7.15 for a more careful
examination of this concavity. The linear convergence of the chord method
can also be seen in the convergence of the ratios R
n
to a constant value.
For this example the MATLAB flops command returns an fairly accurate
indicator of the cost. The Newton iteration required over 8.8 million ﬂoating
point operations while the chord iteration required under 3.3 million. In
Exercise 5.7.18 you are asked to obtain timings in your environment and will
see how the ﬂoatingpoint operation counts are related to the actual timings.
The good performance of the chord method is no accident. If we think
of the chord method as a preconditioned nonlinear Richardson iteration with
the initial Jacobian playing the role of preconditioner, then the chord method
should be much more eﬃcient than Newton’s method if the equations for the
steps can be solved cheaply. In the case considered here, the cost of a single
evaluation and factorization of the Jacobian is far more than the cost of the
entire remainder of the chord iteration.
The ideas in Chapters 6 and 7 show how the low iteration cost of the
chord method can be combined with the small number of iterations of a
superlinearly convergent nonlinear iterative method. In the context of this
particular example, the cost of the Jacobian computation can be reduced by
analytic evaluation and reuse of data that have already been computed for the
evaluation of F. The reader is encouraged to do this in Exercise 5.7.21.
In the remaining examples, we plot, but do not tabulate, the iteration
statistics. In Fig. 5.2 we compare the Shamanskii method with m = 2 with
Newton’s method. The Shamanskii computation required under 6 million
ﬂoatingpoint operations, as compared with over 8.8 for Newton’s method.
However, for this problem, the simple chord method is the most eﬃcient.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 89
0 1 2 3 4 5 6 7 8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 5.1. Newton and chord methods for Hequation, c = .9.
0 0.5 1 1.5 2 2.5 3 3.5 4
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 5.2. Newton and Shamanskii method for Hequation, c = .9.
The hybrid approach in nsol with the default parameters would be the chord
method for this problem.
We now reconsider the Hequation with c = .9999. For this problem the
Jacobian at the solution is nearly singular [110]. Aside from c, all iteration
parameters are the same as in the previous example. While Newton’s method
converges in 7 iterations and requires 21 million ﬂoatingpoint operations, the
chord method requires 188 iterations and 10.6 million ﬂoatingpoint operations.
The qfactor for the chord method in this computation was over .96, indicating
very slow convergence. The hybrid method in nsol with the default parameters
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
90 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
computes the Jacobian four times during the course of the iteration, converges
in 14 iterations, and requires 12 million ﬂoatingpoint operations. In Fig. 5.3
we compare the Newton iteration with the hybrid.
0 2 4 6 8 10 12 14
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 5.3. Newton and hybrid methods for Hequation c = .9999.
One should approach plots of convergence histories with some caution. The
theory for the chord method asserts that the error norms converge qlinearly to
zero. This implies only that the nonlinear residual norms ¦F(x
n
)¦ converge
rlinearly to zero. While the plots indicate qlinear convergence of the nonlinear
residuals, the theory does not, in general, support this. In practice, however,
plots like those in this section are common.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 91
5.7. Exercises on Newton’s method
In some of the exercises here, and in the rest of this part of the book, you will
be asked to plot or tabulate iteration statistics. When asked to do this, for
each iteration tabulate the iteration counter, the norm of F, and the ratio
of F(x
n
)/F(x
n−1
) for n ≥ 1. A better alternative in the MATLAB
environment is to use the semilogy command to plot the norms. When
one does this, one can visually inspect the plot to determine superlinear
convergence (concavity) without explicitly computing the ratios. Use the l
∞
norm.
5.7.1. A function G is said to be H¨older continuous with exponent α in Ω if
G(x) −G(y) ≤ Kx −y
α
for all x, y ∈ Ω. Show that if the Lipschitz
continuity condition on F
in the standard assumptions is replaced by
H¨older continuity with exponent α > 0 that the Newton iterates converge
with qorder 1 +α.
5.7.2. Can the performance of the Newton iteration be improved by a linear
change of variables? That is, for nonsingular N N matrices A and
B, can the Newton iterates for F(x) = 0 and AF(Bx) = 0 show any
performance diﬀerence when started at the same initial iterate? What
about the chord method?
5.7.3. Let ∇
h
F be given by Deﬁnition 5.4.1. Given the standard assumptions,
prove that the iteration given by
x
n+1
= x
n
−(∇
hn
F(x
n
))
−1
F(x
n
)
is locally qsuperlinearly convergent if h
n
→ 0. When is it locally q
quadratically convergent?
5.7.4. Suppose F
(x
n
) is replaced by ∇
hn
F(x
n
) in the Shamanskii method.
Discuss how h
n
must be decreased as the iteration progresses to preserve
the qorder of m+ 1 [174].
5.7.5. In what way is the convergence analysis of the secant method changed if
there are errors in the evaluation of f?
5.7.6. Prove that the secant method converges rsuperlinearly with rorder
(
√
5 + 1)/2. This is easy.
5.7.7. Show that the secant method converges qsuperlinearly with qorder
(
√
5 + 1)/2.
5.7.8. The bisection method produces a sequence of intervals (x
k
l
, x
k
r
) that
contain a root of a function of one variable. Given (x
k
l
, x
k
r
) with
f(x
k
l
)f(x
k
r
) < 0 we replace one of the endpoints with y = (x
k
l
+ x
k
r
)/2
to force f(x
k+1
l
)f(x
k+1
r
) < 0. If f(y) = 0, we terminate the iteration.
Show that the sequence x
k
= (x
k
l
+ x
k
r
)/2 is rlinearly convergent if f is
continuous and there exists (x
0
l
, x
0
r
) such that f(x
0
l
)f(x
0
r
) < 0.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
92 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
5.7.9. Write a program that solves single nonlinear equations with Newton’s
method, the chord method, and the secant method. For the secant
method, use x
−1
= .99x
0
. Apply your program to the following
function/initial iterate combinations, tabulate or plot iteration statistics,
and explain your results.
1. f(x) = cos(x) −x, x
0
= .5
2. f(x) = tan
−1
(x), x
0
= 1
3. f(x) = sin(x), x
0
= 3
4. f(x) = x
2
, x
0
= .5
5. f(x) = x
2
+ 1, x
0
= 10
5.7.10. For scalar functions of one variable a quadratic model of f about a point
x
c
is
m
c
(x) = f(x
c
) +f
(x
c
)(x −x
c
) +f
(x
c
)(x −x
c
)
2
/2.
Consider the iterative method that deﬁnes x
+
as a root of m
c
. What
problems might arise in such a method? Resolve these problems and
implement your algorithm. Test it on the problems in Exercise 5.7.9. In
what cases is this algorithm superior to Newton’s method? For hints and
generalizations of this idea to functions of several variables, see [170]. Can
the ideas in [170] be applied to some of the highorder methods discussed
in [190] for equations in one unknown?
5.7.11. Show that if f is a real function of one real variable, f
is Lipschitz
continuous, and f(x
∗
) = f
(x
∗
) = 0 but f
(x
∗
) ,= 0 then the iteration
x
n+1
= x
n
−2f(x
n
)/f
(x
n
)
converges locally qquadratically to x
∗
provided x
0
is suﬃciently near to
x
∗
, but not equal to x
∗
[171].
5.7.12. Show that if f is a real function of one real variable, the standard
assumptions hold, f
(x
∗
) = 0 and f
(x) is Lipschitz continuous, then
the Newton iteration converges cubically (qsuperlinear with qorder 3).
5.7.13. Show that if f is a real function of one real variable, the standard
assumptions hold, x
0
is suﬃciently near x
∗
, and f(x
0
)f
(x
0
) > 0, then
the Newton iteration converges monotonically to x
∗
. What if f
(x
∗
) = 0?
How can you extend this result if f
(x
∗
) = f
(x
∗
) = . . . f
(k)
(x
∗
) = 0 ,=
f
(k+1)
(x
∗
)? See [131].
5.7.14. How would you estimate the qorder of a qsuperlinearly convergent
sequence? Rigorously justify your answer.
5.7.15. If x
n
→ x
∗
qsuperlinearly with qorder α > 1, show that log(e
n
) is a
concave function of n in the sense that
log(e
n+1
) −log(e
n
)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
NEWTON’S METHOD 93
is a decreasing function of n for n suﬃciently large. Is this still the case
if the convergence is qsuperlinear but there is no qorder? What if the
convergence is qlinear?
5.7.16. Show that the sequence on the right side of (5.20) is rquadratically (but
not qquadratically) convergent to 0 if βηγ < 1/2.
5.7.17. Typically (see [89], [184]) the computed LU factorization of F
(x) is the
exact LU factorization of a nearby matrix.
LU = F
(x) +E.
Use the theory in [89], [184] or another book on numerical linear algebra
and Theorem 5.4.1 to describe how this factorization error aﬀects the
convergence of the Newton iteration. How does the analysis for the QR
factorization diﬀer?
5.7.18. Duplicate the results on the Hequation in ¸ 5.6. Try diﬀerent values of
N and c. How does the performance of the methods diﬀer as N and c
change? Compare your results to the tabulated values of the H function
in [41] or [14]. Compare execution times in your environment. Tabulate
or plot iteration statistics. Use the MATLAB cputime command to see
how long the entire iteration took for each N, c combination and the
MATLAB flops command to get an idea for the number of ﬂoating
point operations required by the various methods.
5.7.19. Solve the Hequation with c = 1 and N = 100. Explain your results
(see [50] and [157]). Does the trick in Exercise 5.7.11 improve the
convergence? Try one of the approaches in [51], [49], or [118] for
acceleration of the convergence. You might also look at [90], [170], [75],
[35] and [155] for more discussion of this. Try the chord method and
compare the results to part 4 of Exercise 5.7.9 (see [52]).
5.7.20. Solve the Hequation with N = 100, c = .9 and c = .9999 with ﬁxedpoint
iteration. Compare timings and ﬂop counts with Newton’s method, the
chord method, and the algorithm in nsol.
5.7.21. Compute by hand the Jacobian of F, the discretized nonlinearity from
the H equation in (5.22). Prove that F
can be computed at a cost of
less than twice that of an evaluation of F. Modify nsol to accept this
analytic Jacobian rather than evaluate a diﬀerence Jacobian. How do
the execution times and ﬂop counts change?
5.7.22. Modify dirder by setting the forward diﬀerence step h = 10
−2
. How
does this change aﬀect the convergence rate observations that you have
made?
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
94 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
5.7.23. Add noise to your function evaluation with a random number generator,
such as the the MATLAB rand function. If the size of the noise is
O(10
−4
) and you make the appropriate adjustments to dirder, how are
the convergence rates diﬀerent? What is an appropriate termination
criterion? At what point does the iteration stagnate?
5.7.24. Assume that the standard assumptions hold, that the cost of a function
evaluation is O(N
2
) ﬂoatingpoint operations, the cost of a Jacobian is
O(N) function evaluations, and that x
0
is near enough to x
∗
so that the
Newton iteration converges qquadratically to x
∗
. Show that the number
n of Newton iterations needed to obtain e
n
 ≤ e
0
 is O(log()) and
that the number of ﬂoatingpoint operations required is O(N
3
log()).
What about the chord method?
5.7.25. Develop a sparse version of nsol for problems with tridiagonal Jacobian.
In MATLAB, for example, you could modify diffjac to use the
techniques of [47] and nsol to use the sparse matrix factorization in
MATLAB. Apply your code to the central diﬀerence discretization of the
twopoint boundary value problem
−u
= sin(u) +f(x), u(0) = u(1) = 0.
Here, the solution is u
∗
(x) = x(1 −x), and
f(x) = 2 −sin(x(1 −x)).
Let u
0
= 0 be the initial iterate.
5.7.26. Suppose you try to ﬁnd an eigenvectoreigenvalue pair (φ, λ) of an NN
matrix A by solving the system of N + 1 nonlinear equations
F(φ, λ) =
_
Aφ −λφ
φ
T
φ −1
_
=
_
0
0
_
(5.23)
for the vector x = (φ
T
, λ)
T
∈ R
N+1
. What is the Jacobian of this
system? If x = (φ, λ) ∈ R
N+1
is an eigenvectoreigenvalue pair, when is
F
(x) nonsingular? Relate the application of Newton’s method to (5.23)
to the inverse power method. See [150] for more development of this
idea.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 6
Inexact Newton Methods
Theorem 5.4.1 describes how errors in the derivative/function aﬀect the
progress in the Newton iteration. Another way to look at this is to ask how
an approximate solution of the linear equation for the Newton step aﬀects the
iteration. This was the view taken in [55] where inexact Newton methods in
which the step satisﬁes
F
(x
c
)s +F(x
c
) ≤ η
c
F(x
c
) (6.1)
are considered. Any approximate step is accepted provided that the relative
residual of the linear equation is small. This is quite useful because conditions
like (6.1) are precisely the small linear residual termination conditions for
iterative solution of the linear system for the Newton step. Such methods are
not new. See [145] and [175], for example, for discussion of these ideas in the
context of the classical stationary iterative methods. In most of this chapter
we will focus on the approximate solution of the equation for the Newton step
by GMRES, but the other Krylov subspace methods discussed in Chapters 2
and 3 and elsewhere can also be used. We follow [69] and refer to the term η
c
on the right hand side of (6.1) as the forcing term.
6.1. The basic estimates
We will discuss speciﬁc implementation issues in ¸ 6.2. Before that we will
give the basic result from [55] to show how a step that satisﬁes (6.1) aﬀects
the convergence. We will present this result in two parts. In ¸ 6.1.1 we give a
straightforward analysis that uses the techniques of Chapter 5. In ¸ 6.1.2 we
show how the requirements of the simple analysis can be changed to admit a
more aggressive (i.e., larger) choice of the parameter η.
6.1.1. Direct analysis. The proof and application of Theorem 6.1.1 should
be compared to that of Theorem 5.1.1 and the other results in Chapter 5.
Theorem 6.1.1. Let the standard assumptions hold. Then there are δ and
K
I
such that if x
c
∈ B(δ), s satisﬁes (6.1), and x
+
= x
c
+s then
e
+
 ≤ K
I
(e
c
 +η
c
)e
c
. (6.2)
95
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
96 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Proof. Let δ be small enough so that the conclusions of Lemma 4.3.1 and
Theorem 5.1.1 hold. To prove the ﬁrst assertion (6.2) note that if
2
r = −F
(x
c
)s −F(x
c
)
is the linear residual then
s +F
(x
c
)
−1
F(x
c
) = −F
(x
c
)
−1
r
and
e
+
= e
c
+s = e
c
−F
(x
c
)
−1
F(x
c
) −F
(x
c
)
−1
r. (6.3)
Now, (6.1), (4.7), and (4.6) imply that
s +F
(x
c
)
−1
F(x
c
) ≤ F
(x
c
)
−1
η
c
F(x
c
)
≤ 4κ(F
(x
∗
))η
c
e
c
.
Hence, using (6.3) and Theorem 5.1.1
e
+
 ≤ e
c
−F
(x
c
)
−1
F(x
c
) + 4κ(F
(x
∗
))η
c
e
c

≤ Ke
c

2
+ 4κ(F
(x
∗
))η
c
e
c
,
where K is the constant from (5.2). If we set
K
I
= K + 4κ(F
(x
∗
)),
the proof is complete.
We summarize the implications of Theorem 6.1.1 for iterative methods in
the next result, which is a direct consequence of (6.2) and will suﬃce to explain
most computational observations.
Theorem 6.1.2. Let the standard assumptions hold. Then there are δ and
¯ η such that if x
0
∈ B(δ), ¦η
n
¦ ⊂ [0, ¯ η], then the inexact Newton iteration
x
n+1
= x
n
+s
n
,
where
F
(x
n
)s
n
+F(x
n
) ≤ η
n
F(x
n
)
converges qlinearly to x
∗
. Moreover
• if η
n
→ 0 the convergence is qsuperlinear, and
• if η
n
≤ K
η
F(x
n
)
p
for some K
η
> 0 the convergence is qsuperlinear
with qorder 1 +p.
2
Our deﬁnition of r is consistent with the idea that the residual in a linear equation is b −Ax. In
[55] r = F
(xc)s +F(xc).
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 97
Proof. Let δ be small enough so that (6.2) holds for x
c
∈ B(δ). Reduce δ
and ¯ η if needed so that
K
I
(δ + ¯ η) < 1,
where K
I
is from (6.2). Then if n ≥ 0 and x
n
∈ B(δ) we have
e
n+1
 ≤ K
I
(e
n
 +η
n
)e
n
 ≤ K
I
(δ + ¯ η)e
n
 < e
n
.
This proves qlinear convergence with a qfactor of K
I
(δ + ¯ η).
If η
n
→ 0 then qsuperlinear convergence follows from the deﬁnition. If
η
n
≤ K
η
F(x
n
)
p
then we may use (4.7) and (6.2) to conclude
e
n+1
 ≤ K
I
(e
n

1−p
+K
η
2
p
F
(x
∗
)
p
)e
n

1+p
which completes the proof.
6.1.2. Weighted norm analysis. Since K
I
= O(κ(F
(x
∗
))), one might
conclude from Theorem 6.1.2 that if F
(x
∗
) is ill conditioned very small forcing
terms must be used. This is not the case and the purpose of this subsection
is to describe more accurately the necessary restrictions on the forcing terms.
The results here diﬀer from those in ¸ 6.1.1 in that no restriction is put on
the sequence ¦η
n
¦ ⊂ [0, 1) other than requiring that 1 not be an accumulation
point.
Theorem 6.1.3. Let the standard assumptions hold. Then there is δ such
that if x
c
∈ B(δ), s satisﬁes (6.1), x
+
= x
c
+s, and η
c
≤ η < ¯ η < 1, then
F
(x
∗
)e
+
 ≤ ¯ ηF
(x
∗
)e
c
. (6.4)
Proof. To prove (6.4) note that Theorem 4.0.1 implies that
F(x
c
) ≤ F
(x
∗
)e
c
 +
γe
c

2
2
.
Since
e
c
 = F
(x
∗
)
−1
F
(x
∗
)e
c
 ≤ F
(x
∗
)
−1
F
(x
∗
)e
c

we have, with
M
0
=
γF
(x
∗
)
−1

2
F(x
c
) ≤ (1 +M
0
δ)F
(x
∗
)e
c
. (6.5)
Now,
F
(x
∗
)e
+
= F
(x
∗
)(e
c
+s)
= F
(x
∗
)(e
c
−F
(x
c
)
−1
F(x
c
) −F
(x
c
)
−1
r).
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
98 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
By Theorem 5.1.1
F
(x
∗
)(e
c
−F
(x
c
)
−1
F(x
c
)) ≤ KF
(x
∗
)e
c

2
.
Hence,
F
(x
∗
)e
+
 ≤ F
(x
∗
)F
(x
c
)
−1
r +KF
(x
∗
)e
c

2
. (6.6)
Since
F
(x
∗
)F
(x
c
)
−1
r ≤ r +(F
(x
∗
) −F
(x
c
))F
(x
c
)
−1
r
≤ (1 + 2γF
(x
∗
)
−1
e
c
)r,
we may set
M
1
= 2γF
(x
∗
)
−1
, and M
2
= KF
(x
∗
)
and obtain, using (4.7) and (6.5),
F
(x
∗
)e
+
 ≤ (1 +M
1
δ)r +M
2
δe
c

≤ (1 +M
1
δ)(1 +M
0
δ)η
c
F
(x
∗
)e
c

+M
2
δF
(x
∗
)
−1
F
(x
∗
)e
c

≤ ((1 +M
1
δ)(1 +M
0
δ)η +M
2
δF
(x
∗
)
−1
)F
(x
∗
)e
c
.
(6.7)
Now let δ be small enough so that
(1 +M
1
δ)(1 +M
0
δ)η +M
2
δF
(x
∗
)
−1
 ≤ ¯ η
and the proof is complete.
Note that the distance δ from the initial iterate to the solution may have
to be smaller for (6.4) to hold than for (6.2). However (6.4) is remarkable in its
assertion that any method that produces a step with a linear residual less than
that of the zero step will reduce the norm of the error if the error is measured
with the weighted norm
 
∗
= F
(x
∗
) .
In fact, Theorem 6.1.3 asserts qlinear convergence in the weighted norm if the
initial iterate is suﬃciently near x
∗
. This is made precise in Theorem 6.1.4.
The application of Theorem 6.1.3 described in Theorem 6.1.4 diﬀers from
Theorem 6.1.2 in that we do not demand that ¦η
n
¦ be bounded away from 1
by a suﬃciently large amount, just that 1 not be an accumulation point. The
importance of this theorem for implementation is that choices of the sequence
of forcing terms ¦η
n
¦ (such as η
n
= .5 for all n) that try to minimize the
number of inner iterations are completely justiﬁed by this result if the initial
iterate is suﬃciently near the solution. We make such a choice in one of the
examples considered in ¸ 6.4 and compare it to a modiﬁcation of a choice from
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 99
[69] that decreases η
n
rapidly with a view toward minimizing the number of
outer iterations.
Theorem 6.1.4. Let the standard assumptions hold. Then there is δ such
that if x
0
∈ B(δ), ¦η
n
¦ ⊂ [0, η] with η < ¯ η < 1, then the inexact Newton
iteration
x
n+1
= x
n
+s
n
,
where
F
(x
n
)s
n
+F(x
n
) ≤ η
n
F(x
n
)
converges qlinearly with respect to  
∗
to x
∗
. Moreover
• if η
n
→ 0 the convergence is qsuperlinear, and
• if η
n
≤ K
η
F(x
n
)
p
for some K
η
> 0 the convergence is qsuperlinear
with qorder 1 +p.
Proof. The proof uses both (6.4) and (6.2). Our ﬁrst task is to relate the
norms   and  
∗
so that we can estimate δ. Let δ
0
be such that (6.4) holds
for e
c
 < δ
0
.
For all x ∈ R
N
,
x
∗
≤ F
(x
∗
)x ≤ κ(F
(x
∗
))x
∗
, (6.8)
Note that e < δ
0
if
e
∗
< δ
∗
= F
(x
∗
)δ
0
Set δ = F
(x
∗
)
−1
δ
∗
. Then e
0

∗
< δ
∗
if e
0
 < δ.
Our proof does not rely on the (possibly false) assertions that e
n
 < δ for
all n or that x
n
→ x
∗
qlinearly with respect to the unweighted norm. Rather
we note that (6.4) implies that if e
n

∗
< δ
∗
then
e
n+1

∗
≤ ¯ ηe
n

∗
< δ
∗
(6.9)
and hence x
n
→ x
∗
qlinearly with respect to the weighted norm. This proves
the ﬁrst assertion.
To prove the assertions on qsuperlinear convergence, note that since
x
n
→ x
∗
, eventually e
n
 will be small enough so that the conclusions of
Theorem 6.1.1 hold. The superlinear convergence results then follow exactly
as in the proof of Theorem 6.1.2.
We close this section with some remarks on the two types of results in
Theorems 6.1.2 and 6.1.4. The most signiﬁcant diﬀerence is the norm in which
qlinear convergence takes place. qlinear convergence with respect to the
weighted norm  
∗
is equivalent to qlinear convergence of the sequence of
nonlinear residuals ¦F(x
n
)¦. We state this result as Proposition 6.1.1 and
leave the proof to the reader in Exercise 6.5.2. The implications for the choice of
the forcing terms ¦η
n
¦ are also diﬀerent. While Theorem 6.1.2 might lead one
to believe that a small value of η
n
is necessary for convergence, Theorem 6.1.4
shows that the sequence of forcing terms need only be kept bounded away from
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
100 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
1. A constant sequence such as η
n
= .5 may well suﬃce for linear convergence,
but at a price of a greater number of outer iterations. We will discuss other
factors in the choice of forcing terms in ¸ 6.3.
Proposition 6.1.1. Let the standard assumptions hold and let x
n
→ x
∗
.
Then F(x
n
) converges qlinearly to 0 if and only if e
n

∗
does.
6.1.3. Errors in the function. In this section, we state two theorems on
inexact local improvement. We leave the proofs, which are direct analogs to
the proofs of Theorems 6.1.1 and 6.1.3, to the reader as exercises. Note that
errors in the derivative, such as those arising from a diﬀerence approximation
of the action of F
(x
c
) on a vector, can be regarded as part of the inexact
solution of the equation for the Newton step. See [55] or Proposition 6.2.1 and
its proof for an illustration of this point.
Theorem 6.1.5. Let the standard assumptions hold. Then there are δ and
K
I
such that if x
c
∈ B(δ), s satisﬁes
F
(x
c
)s +F(x
c
) +(x
c
) ≤ η
c
F(x
c
) +(x
c
) (6.10)
and x
+
= x
c
+s, then
e
+
 ≤ K
I
((e
c
 +η
c
)e
c
 +(x
c
)). (6.11)
Theorem 6.1.6. Let the standard assumptions hold. Then there is δ such
that if x
c
∈ B(δ), s satisﬁes (6.10), x
+
= x
c
+ s, and η
c
≤ η < ¯ η < 1, then
there is K
E
such that
F
(x
∗
)e
+
 ≤ ¯ ηF
(x
∗
)e
c
 +K
E
(x
c
). (6.12)
6.2. Newtoniterative methods
A Newtoniterative method realizes (6.1) with an iterative method for the linear
system for the Newton step, terminating when the relative linear residual is
smaller than η
c
(i.e, when (6.1) holds). The name of the method indicates
which linear iterative method is used, for example, NewtonSOR, NewtonCG,
NewtonGMRES, would use SOR, CG, or GMRES to solve the linear system.
This naming convention is taken from [175] and [145]. These methods have also
been called truncated Newton methods [56], [136] in the context of optimization.
Typically the nonlinear iteration that generates the sequence ¦x
n
¦ is called
the outer iteration and the linear iteration that generates the approximations
to the steps is called the inner iteration.
In this section we use the l
2
norm to measure nonlinear residuals. The
reason for this is that the Krylov methods use the scalar product and the
estimates in Chapters 2 and 3 are in terms of the Euclidean norm. In the
examples discussed in ¸ 6.4 we will scale the l
2
norm by a factor of 1/N so
that the results for the diﬀerential and integral equations will be independent
of the computational mesh.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 101
6.2.1. Newton GMRES. We provide a detailed analysis of the Newton
GMRES iteration. We begin by discussing the eﬀects of a forward diﬀerence
approximation to the action of the Jacobian on a vector.
If the linear iterative method is any one of the Krylov subspace methods
discussed in Chapters 2 and 3 then each inner iteration requires at least one
evaluation of the action of F
(x
c
) on a vector. In the case of CGNR and CGNE,
an evaluation of the action of F
(x
c
)
T
on a vector is also required. In many
implementations [20], [24], the action of F
(x
c
) on a vector w is approximated
by a forward diﬀerence, (5.15), D
h
F(x : w) for some h. It is important to note
that this is entirely diﬀerent from forming the ﬁnite diﬀerence Jacobian ∇
h
F(x)
and applying that matrix to w. In fact, as pointed out in [20], application
of GMRES to the linear equation for the Newton step with matrixvector
products approximated by ﬁnite diﬀerences is the same as the application of
GMRES to the matrix G
h
F(x) whose last k −1 columns are the vectors
v
k
= D
h
F(x : v
k−1
)
The sequence ¦v
k
¦ is formed in the course of the forwarddiﬀerence GMRES
iteration.
To illustrate this point, we give the forwarddiﬀerence GMRES algorithm
for computation of the Newton step, s = −F
(x)
−1
F(x). Note that matrix
vector products are replaced by forward diﬀerence approximations to F
(x), a
sequence of approximate steps ¦s
k
¦ is produced, that b = −F(x), and that the
initial iterate for the linear problem is the zero vector. We give a MATLAB
implementation of this algorithm in the collection of MATLAB codes.
Algorithm 6.2.1. fdgmres(s, x, F, h, η, kmax, ρ)
1. s = 0, r = −F(x), v
1
= r/r
2
, ρ = r
2
, β = ρ, k = 0
2. While ρ > ηF(x)
2
and k < kmax do
(a) k = k + 1
(b) v
k+1
= D
h
F(x : v
k
)
for j = 1, . . . k
i. h
jk
= v
T
k+1
v
j
ii. v
k+1
= v
k+1
−h
jk
v
j
(c) h
k+1,k
= v
k+1

2
(d) v
k+1
= v
k+1
/v
k+1

2
(e) e
1
= (1, 0, . . . , 0)
T
∈ R
k+1
Minimize βe
1
−H
k
y
k

R
k+1 to obtain y
k
∈ R
k
.
(f) ρ = βe
1
−H
k
y
k

R
k+1.
3. s = V
k
y
k
.
In our MATLAB implementation we solve the least squares problem in
step 2e by the Givens rotation approach used in Algorithm gmres.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
102 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
As should be clear from inspection of Algorithm gmres, the diﬀerence
between ∇
h
F and G
h
F is that the columns are computed by taking directional
derivatives based on two diﬀerent orthonormal bases: ∇
h
F using the basis of
unit vectors in coordinate directions and G
h
F the basis for the Krylov space /
k
constructed by algorithm fdgmres. The action of G
h
F on /
⊥
k
, the orthogonal
complement of /
k
, is not used by the algorithm and, for the purposes of
analysis, could be speciﬁed by the action of ∇
h
F on any basis for /
⊥
k
. Hence
G
h
F is also ﬁrst order accurate in h. Assuming that there is no error in the
evaluation of F we may summarize the observations above in the following
proposition, which is a special case of the more general results in [20].
Proposition 6.2.1. Let the standard assumptions hold. Let η ∈ (0, 1).
Then there are C
G
,
¯
h, and δ such that if x ∈ B(δ), h ≤
¯
h, and Algo
rithm fdgmres terminates with k < kmax then the computed step s satisﬁes
F
(x)s +F(x)
2
< (η +C
G
h)F(x)
2
. (6.13)
Proof. First let δ be small enough so that B(δ) ⊂ Ω and the conclusions
to Lemma 4.3.1 hold. Let ¦u
j
¦ be any orthonormal basis for R
N
such that
u
j
= v
j
for 1 ≤ j ≤ k, where ¦v
j
¦
k
j=1
is the orthonormal basis for /
k
generated
by Algorithm fdgmres.
Consider the linear map deﬁned by its action on ¦u
j
¦
Bu
j
= D
h
F(x : u
j
).
Note that
D
h
F(x : u
j
) =
_
1
0
F
(x +thx
2
u
j
)u
j
dt
= F
(x)u
j
+
_
1
0
(F
(x +thx
2
u
j
) −F
(x))u
j
dt.
Since F
is Lipschitz continuous and ¦u
j
¦ is an orthonormal basis for R
N
we
have, with ¯ γ = γ(x
∗

2
+δ),
B −F
(x)
2
≤ hx
2
γ/2 ≤ h¯ γ/2 (6.14)
Since the linear iteration (fdgmres) terminates in k < kmax iterations, we
have, since B and G
h
F agree on /
k
,
Bs +F(x)
2
≤ ηF(x)
2
and therefore
F
(x)s +F(x)
2
≤ ηF(x)
2
+h¯ γs
2
/2. (6.15)
Assume that
¯
h¯ γ ≤ F
(x
∗
)
−1

−1
2
/2
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 103
Then Lemma 4.3.1 and (6.15) imply
F
(x
∗
)
−1

−1
2
s
2
/2 ≤ F
(x)
−1

−1
2
s
2
≤ F
(x)s
2
≤ (1 +η)F(x)
2
+
¯
h¯ γs
2
/2.
Therefore,
s
2
≤ 4(1 +η)F
(x
∗
)
−1

2
F(x)
2
. (6.16)
Combining (6.16) with (6.15) completes the proof with C
G
= 4¯ γ(1 +
η)F
(x
∗
)
−1

2
.
Proposition 6.2.1 implies that a ﬁnite diﬀerence implementation will not
aﬀect the performance of NewtonGMRES if the steps in the forward diﬀerence
approximation of the derivatives are suﬃciently small. In an implementation,
therefore, we must specify not only the sequence ¦η
n
¦, but also the steps ¦h
n
¦
used to compute forward diﬀerences. Note also that Proposition 6.2.1 applies
to a restarted GMRES because upon successful termination (6.15) will hold.
We summarize our results so far in the analogs of Theorem 6.1.4 and
Theorem 6.1.2. The proofs are immediate consequences of Theorems 6.1.2,
6.1.4, and Proposition 6.2.1 and are left as (easy) exercises.
Theorem 6.2.1. Assume that the assumptions of Proposition 6.2.1 hold.
Then there are δ, ¯ σ such that if x
0
∈ B(δ) and the sequences ¦η
n
¦ and ¦h
n
¦
satisfy
σ
n
= η
n
+C
G
h
n
≤ ¯ σ
then the forward diﬀerence NewtonGMRES iteration
x
n+1
= x
n
+s
n
where s
n
is computed by Algorithm fdgmres with arguments
(s
n
, x
n
, F, h
n
, η
n
, kmax, ρ)
converges qlinearly and s
n
satisﬁes
F
(x
n
)s
n
+F(x
n
)
2
< σ
n
F(x
n
)
2
. (6.17)
Moreover,
• if σ
n
→ 0 the convergence is qsuperlinear, and
• if σ
n
≤ K
η
F(x
n
)
p
2
for some K
η
> 0 the convergence is qsuperlinear
with qorder 1 +p.
Theorem 6.2.2. Assume that the assumptions of Proposition 6.2.1 hold.
Then there is δ such that if x
0
∈ B(δ) and the sequences ¦η
n
¦ and ¦h
n
¦ satisfy
σ
n
= η
n
+C
G
h
n
∈ [0, ¯ σ]
with 0 < ¯ σ < 1 then the forward diﬀerence NewtonGMRES iteration
x
n+1
= x
n
+s
n
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
104 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
where s
n
is computed by Algorithm fdgmres with arguments
(s
n
, x
n
, F, h
n
, η
n
, kmax, ρ)
converges qlinearly with respect to  
∗
and s
n
satisﬁes (6.17). Moreover
• if σ
n
→ 0 the convergence is qsuperlinear, and
• if σ
n
≤ K
η
F(x
n
)
p
2
for some K
η
> 0 the convergence is qsuperlinear
with qorder 1 +p.
In the case where h
n
is approximately the square root of machine roundoﬀ,
σ
n
≈ η
n
if η
n
is not too small and Theorem 6.2.2 states that the observed
behavior of the forward diﬀerence implementation will be the same as that of
an implementation with exact derivatives.
6.2.2. Other Newtoniterative methods. Other iterative methods may
be used as the solver for the linear equation for the Newton step. Newton
multigrid [99] is one approach using a stationary iterative method. See also
[6], [111]. If F
is symmetric and positive deﬁnite, as it is in the context of
unconstrained minimization, [63], NewtonCG is a possible approach.
When storage is restricted or the problem is very large, GMRES may not
be practical. GMRES(m), for a small m, may not converge rapidly enough
to be useful. CGNR and CGNE oﬀer alternatives, but require evaluation of
transposevector products. BiCGSTAB and TFQMR should be considered in
all cases where there is not enough storage for GMRES(m) to perform well.
If a transpose is needed, as it will be if CGNR or CGNE is used as the
iterative method, a forward diﬀerence formulation is not an option because
approximation of the transposevector product by diﬀerences is not possible.
Computation of ∇
h
F(x) and its transpose analytically is one option as is
diﬀerentiation of F automatically or by hand. The reader may explore this
further in Exercise 6.5.13.
The reader may also be interested in experimenting with the other Krylov
subspace methods described in ¸ 3.6, [78], and [12].
6.3. NewtonGMRES implementation
We provide a MATLAB code nsolgm in the collection that implements
a forwarddiﬀerence NewtonGMRES algorithm. This algorithm requires
diﬀerent inputs than nsol. As input we must give the initial iterate, the
function, the vector of termination tolerances as we did for nsol. In addition,
we must provide a method for forming the sequence of forcing terms ¦η
n
¦.
We adjust η as the iteration progresses with a variation of a choice from
[69]. This issue is independent of the particular linear solver and our discussion
is in the general inexact Newton method setting. Setting η to a constant for
the entire iteration is often a reasonable strategy as we see in ¸ 6.4, but the
choice of that constant depends on the problem. If a constant η is too small,
much eﬀort can be wasted in the initial stages of the iteration. The choice in
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 105
[69] is motivated by a desire to avoid such over solving. Over solving means
that the linear equation for the Newton step is solved to a precision far beyond
what is needed to correct the nonlinear iteration. As a measure of the degree
to which the nonlinear iteration approximates the solution we begin with
η
A
n
= γF(x
n
)
2
/F(x
n−1
)
2
,
where γ ∈ (0, 1] is a parameter. If η
A
n
is uniformly bounded away from 1,
then setting η
n
= η
A
n
for n > 0 would guarantee qquadratic convergence by
Theorem 6.1.1. In this way, the most information possible would be extracted
from the inner iteration. In order to specify the choice at n = 0 and bound
the sequence away from 1 we set
η
B
n
=
_
¸
_
¸
_
η
max
, n = 0,
min(η
max
, η
A
n
), n > 0.
(6.18)
In (6.18) the parameter η
max
is an upper limit on the sequence ¦η
n
¦. In [69]
the choices γ = .9 and η
max
= .9999 are used.
It may happen that η
B
n
is small for one or more iterations while x
n
is still
far from the solution. A method of safeguarding was suggested in [69] to avoid
volatile decreases in η
n
. The idea is that if η
n−1
is suﬃciently large we do not
let η
n
decrease by much more than a factor of η
n−1
.
η
C
n
=
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
η
max
, n = 0,
min(η
max
, η
A
n
), n > 0, γη
2
n−1
≤ .1,
min(η
max
, max(η
A
n
, γη
2
n−1
)), n > 0, γη
2
n−1
> .1.
(6.19)
The constant .1 is somewhat arbitrary. This safeguarding does improve the
performance of the iteration.
There is a chance that the ﬁnal iterate will reduce F far beyond the
desired level and that the cost of the solution of the linear equation for the last
step will be more accurate than is really needed. This oversolving on the ﬁnal
step can be controlled comparing the norm of the current nonlinear residual
F(x
n
) to the nonlinear residual norm at which the iteration would terminate
τ
t
= τ
a
+τ
r
F(x
0
)
and bounding η
n
from below by a constant multiple of τ
t
/F(x
n
). The
algorithm nsolgm does this and uses
η
n
= min(η
max
, max(η
C
n
, .5τ
t
/F(x
n
))). (6.20)
Exercise 6.5.9 asks the reader to modify nsolgm so that other choices of
the sequence ¦η
n
¦ can be used.
We use dirder to approximate directional derivatives and use the default
value of h = 10
−7
. In all the examples in this book we use the value γ = .9 as
recommended in [69].
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
106 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Algorithm 6.3.1. nsolgm(x, F, τ, η)
1. r
c
= r
0
= F(x)
2
/
√
N
2. Do while F(x)
2
/
√
N > τ
r
r
0
+τ
a
(a) Select η.
(b) fdgmres(s, x, F, η)
(c) x = x +s
(d) Evaluate F(x)
(e) r
+
= F(x)
2
/
√
N, σ = r
+
/r
c
, r
c
= r
+
(f) If F(x)
2
≤ τ
r
r
0
+τ
a
exit.
Note that the cost of an outer iterate is one evaluation of F to compute
the value at the current iterate and other evaluations of F to compute the
forward diﬀerences for each inner iterate. Hence if a large value of η can be
used, the cost of the entire iteration can be greatly reduced. If a maximum
of m GMRES iterations is needed (or GMRES(m) is used as a linear solver)
the storage requirements of Algorithm nsolgm are the m+ 5 vectors x, F(x),
x +hv, F(x +hv), s, and the Krylov basis ¦v
k
¦
m
k=1
.
Since GMRES forms the inner iterates and makes termination decisions
based on scalar products and l
2
norms, we also terminate the outer iteration on
small l
2
norms of the nonlinear residuals. However, because of our applications
to diﬀerential and integral equations, we scale the l
2
norm of the nonlinear
residual by a factor of 1/
√
N so that constant functions will have norms that
are independent of the computational mesh.
For large and poorly conditioned problems, GMRES will have to be
restarted. We illustrate the consequences of this in ¸ 6.4 and ask the reader to
make the necessary modiﬁcations to nsolgm in Exercise 6.5.9.
The preconditioners we use in ¸ 6.4 are independent of the outer iteration.
Since the Newton steps for MF(x) = 0 are the same as those for F(x) = 0, it
is equivalent to precondition the nonlinear equation before calling nsolgm.
6.4. Examples for NewtonGMRES
In this section we consider two examples. We revisit the Hequation and solve
a preconditioned nonlinear convectiondiﬀusion equation.
In all the ﬁgures we plot the relative nonlinear residual F(x
n
)
2
/F(x
0
)
2
against the number of function evaluations required by all inner and outer
iterations to compute x
n
. Counts of function evaluations corresponding to
outer iterations are indicated by circles. From these plots one can compare
not only the number of outer iterations, but also the total cost. This enables
us to directly compare the costs of diﬀerent strategies for selection of η. Note
that if only a single inner iteration is needed, the total cost of the outer iteration
will be two function evaluations since F(x
c
) will be known. One new function
evaluation will be needed to approximate the action of F
(x
c
) on a vector
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 107
and then F(x
+
) will be evaluated to test for termination. We also count
the evaluation of F(x
0
) as an additional cost in the evaluation of x
1
, which
therefore has a minimum cost of three function evaluations. For each example
we compare a constant value of η
n
with the choice given in (6.20) and used as
the default in nsolgm.
6.4.1. Chandrasekhar Hequation. We solve the Hequation on a 100
point mesh with c = .9 using two schemes for selection of the parameter η.
The initial iterate is the function identically one. We set the parameters in
(6.20) to γ = .9 and η
max
= .25.
We used τ
r
= τ
a
= 10
−6
in this example.
In Fig. 6.1 we plot the progress of the iteration using η = .1 with the solid
line and using the sequence given by (6.20) with the dashed line. We set the
parameters in (6.20) to γ = .9 and η
max
= .25. This choice of η
max
was the
one that did best overall in our experiments on this problem.
We see that the constant η iteration terminates after 12 function evalua
tions, 4 outer iterations, and roughly 275 thousand ﬂoatingpoint operations.
This is a slightly higher overall cost than the other approach in which ¦η
n
¦ is
given by (6.20), which terminated after 10 function evaluations, 3 outer itera
tions, and 230 thousand ﬂoatingpoint operations. The chord method with the
diﬀerent (but very similar for this problem) l
∞
termination condition for the
same problem, reported in ¸ 5.6 incurred a cost of 3.3 million ﬂoatingpoint
operations, slower by a factor of over 1000. This cost is entirely a result of the
computation and factorization of a single Jacobian.
0 2 4 6 8 10 12
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 6.1. NewtonGMRES for the Hequation, c = .9.
In Fig. 6.2 the results change for the more illconditioned problem with
c = .9999. Here the iteration with η
n
= .1 performed slightly better and
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
108 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
0 5 10 15 20 25
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 6.2. NewtonGMRES for the Hequation, c = .9999
0 2 4 6 8 10 12 14 16 18
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 6.3. NewtonGMRES for the PDE, C = 20.
terminated after 22 function evaluations and 7 outer iterations, and roughly
513 thousand ﬂoatingpoint operations. The iteration with the decreasing ¦η
n
¦
given by (6.20) required 23 function evaluations, 7 outer iterations, and 537
thousand ﬂoatingpoint operations.
6.4.2. Convectiondiﬀusion equation. We consider the partial diﬀeren
tial equation
−∇
2
u +Cu(u
x
+u
y
) = f (6.21)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 109
with homogeneous Dirichlet boundary conditions on the unit square (0, 1)
(0, 1). f has been constructed so that the exact solution was the discretization
of
10xy(1 −x)(1 −y) exp(x
4.5
).
We set C = 20 and u
0
= 0. As in ¸ 3.7 we discretized (6.21) on a 31 31 grid
using centered diﬀerences. To be consistent with the secondorder accuracy of
the diﬀerence scheme we used τ
r
= τ
a
= h
2
, where h = 1/32.
We compare the same possibilities for ¦η
n
¦ as in the previous example and
report the results in Fig. 6.3. In (6.20) we set γ = .9 and η
max
= .5. The
computation with constant η terminated after 19 function evaluations and 4
outer iterates at a cost of roughly 3 million ﬂoatingpoint operations. The
iteration with ¦η
n
¦ given by (6.20) required 16 function evaluations, 4 outer
iterations, and 2.6 million ﬂoatingpoint operations. In the computation we
preconditioned (6.21) with the fast Poisson solver fish2d.
In this case, the choice of ¦η
n
¦ given by (6.20) reduced the number of inner
iterations in the early stages of the iteration and avoided oversolving. In cases
where the initial iterate is very good and only a single Newton iterate might be
needed, however, a large choice of η
max
or constant η may be the more eﬃcient
choice. Examples of such cases include (1) implicit time integration of nonlinear
parabolic partial diﬀerential equations or ordinary diﬀerential equations in
which the initial iterate is either the converged solution from the previous
time step or the output of a predictor and (2) solution of problems in which
the initial iterate is an interpolant of the solution from a coarser mesh.
Preconditioning is crucial for good performance. When preconditioning
was not done, the iteration for C = 20 required more than 80 function
evaluations to converge. In Exercise 6.5.9 the reader is asked to explore this.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
110 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
6.5. Exercises on inexact Newton methods
6.5.1. Verify (6.8).
6.5.2. Prove Proposition 6.1.1 by showing that if the standard assumptions
hold, 0 < < 1, and x is suﬃciently near x
∗
then
F(x)/(1 +) ≤ F
(x
∗
)e ≤ (1 +)F(x).
6.5.3. Verify (6.14).
6.5.4. Prove Theorems 6.2.2 and 6.2.1.
6.5.5. Prove Theorems 6.1.5 and 6.1.6.
6.5.6. Can anything like Proposition 6.2.1 be proved for a ﬁnitediﬀerence Bi
CGSTAB or TFQMR linear solver? If not, what are some possible
implications?
6.5.7. Give an example in which F(x
n
) → 0 qlinearly but x
n
does not converge
to x
∗
qlinearly.
6.5.8. In the DASPK code for integration of diﬀerential algebraic equations,
[23], preconditioned NewtonGMRES is implemented in such a way
that the data needed for the preconditioner is only computed when
needed, which may be less often that with each outer iteration. Discuss
this strategy and when it might be useful. Is there a relation to the
Shamanskii method?
6.5.9. Duplicate the results in ¸ 6.4. For the Hequation experiment with
various values of c, such as c = .5, .99999, .999999, and for the convection
diﬀusion equation various values of C such as C = 1, 10, 40 and how
preconditioning aﬀects the iteration. Experiment with the sequence of
forcing terms ¦η
n
¦. How do the choices η
n
= 1/(n + 1), η
n
= 10
−4
[32], η
n
= 2
1−n
[24], η
n
= min(F(x
n
)
2
, (n + 2)
−1
) [56], η
n
= .05, and
η
n
= .75 aﬀect the results? Present the results of your experiments as
graphical iteration histories.
6.5.10. For the Hequation example in ¸ 6.4, vary the parameter c and see how
the performance of NewtonGMRES with the various choices of ¦η
n
¦ is
aﬀected. What happens when c = 1? See [119] for an explanation of
this.
6.5.11. Are the converged solutions for the preconditioned and unpreconditioned
convectiondiﬀusion example in ¸ 6.4 equally accurate?
6.5.12. For the convectiondiﬀusion example in ¸ 6.4, how is the performance
aﬀected if GMRES(m) is used instead of GMRES?
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INEXACT NEWTON METHODS 111
6.5.13. Apply NewtonCGNR, NewtonBiCGSTAB, or NewtonTFQMR to the
two examples in ¸ 6.4. How will you deal with the need for a transpose in
the case of CGNR? How does the performance (in terms of time, ﬂoating
point operations, function evaluations) compare to NewtonGMRES?
Experiment with the forcing terms. The MATLAB codes fdkrylov,
which allows you to choose from a variety of forward diﬀerence Krylov
methods, and nsola, which we describe more fully in Chapter 8, might
be of use in this exercise.
6.5.14. If you have done Exercise 5.7.25, apply nsolgm to the same twopoint
boundary value problem and compare the performance.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
112 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 7
Broyden’s method
QuasiNewton methods maintain approximations of the solution x
∗
and the
Jacobian at the solution F
(x
∗
) as the iteration progresses. If x
c
and B
c
are
the current approximate solution and Jacobian, then
x
+
= x
c
−B
−1
c
F(x
c
). (7.1)
After the computation of x
+
, B
c
is updated to form B
+
. The construction of
B
+
determines the quasiNewton method.
The advantages of such methods, as we shall see in ¸ 7.3, is that solution of
equations with the quasiNewton approximation to the Jacobian is often much
cheaper than using F
(x
n
) as the coeﬃcient matrix. In fact, if the Jacobian is
dense, the cost is little more than that of the chord method. The method we
present in this chapter, Broyden’s method, is locally superlinearly convergent,
and hence is a very powerful alternative to Newton’s method or the chord
method.
In comparison with Newtoniterative methods, quasiNewton methods
require only one function evaluation for each nonlinear iterate; there is no cost
in function evaluations associated with an inner iteration. Hence, if a good
preconditioner (initial approximation to F
(x
∗
)) can be found, these methods
could have an advantage in terms of function evaluation cost over Newton
iterative methods.
Broyden’s method [26] computes B
+
by
B
+
= B
c
+
(y −B
c
s)s
T
s
T
s
= B
c
+
F(x
+
)s
T
s
T
s
. (7.2)
In (7.2) y = F(x
+
) −F(x
c
) and s = x
+
−x
c
.
Broyden’s method is an example of a secant update. This means that the
updated approximation to F
(x
∗
) satisﬁes the secant equation
B
+
s = y. (7.3)
In one dimension, (7.3) uniquely speciﬁes the classical secant method. For
equations in several variables (7.3) alone is a system of N equations in N
2
113
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
114 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
unknowns. The role of the secant equation is very deep. See [62] and [64] for
discussion of this topic. In this book we focus on Broyden’s method and our
methods of analysis are not as general as those in [62] and [64]. In this chapter
we will assume that the function values are accurate and refer the reader to [66],
[65], and [114] for discussions of quasiNewton methods with inaccurate data.
Locally superlinearly convergent secant methods can be designed to maintain
the sparsity pattern, symmetry, or positive deﬁniteness, of the approximate
Jacobian, and we refer the reader to [28], [63], [62], and [64] for discussion and
analysis of several such methods. More subtle structural properties can also
be preserved by secant methods. Some examples can be found in [101], [96],
[95], [113], and [116].
Broyden’s method is also applicable to linear equations [29], [82], [84],
[104], [143], [130], Ax = b, where B ≈ A is updated. One can also apply the
method to linear least squares problems [84], [104]. The analysis is somewhat
clearer in the linear case and the plan of this chapter is to present results for
the linear case for both the theory and the examples. One interesting result,
which we will not discuss further, is that for linear problems Broyden’s method
converges in 2N iterations [29], [82], [84], [143].
We will express our results in terms of the error in the Jacobian
E = B −F
(x
∗
) (7.4)
and the step s = x
+
−x
c
. While we could use E = B−F
(x), (7.4) is consistent
with our use of e = x − x
∗
. When indexing of the steps is necessary, we will
write
s
n
= x
n+1
−x
n
.
In this chapter we show that if the data x
0
and B
0
are suﬃciently good,
the Broyden iteration will converge qsuperlinearly to the root.
7.1. The Dennis–Mor´e condition
In this section we consider general iterative methods of the form
x
n+1
= x
n
−B
−1
n
F(x
n
) (7.5)
where B
n
= F
(x
∗
)+E
n
≈ F
(x
∗
) is generated by some method (not necessarily
Broyden’s method).
Veriﬁcation of qsuperlinear convergence cannot be done by the direct
approaches used in Chapters 5 and 6. We must relate the superlinear
convergence condition that η
n
→ 0 in Theorem 6.1.2 to a more subtle, but
easier to verify condition in order to analyze Broyden’s method. This technique
is based on veriﬁcation of the Dennis–Mor´e condition [61], [60] on the sequence
of steps ¦s
n
¦ and errors in the Jacobian ¦E
n
¦
lim
n→∞
E
n
s
n

s
n

= 0. (7.6)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 115
The main result in this section, Theorem 7.1.1 and its corollary for linear
problems, are used to prove superlinear convergence for Broyden’s method in
this chapter and many other quasiNewton methods. See [61], [28], [62], and
[64] for several more examples. Our formulation of the Dennis–Mor´e result is
a bit less general than that in [60] or [63].
Theorem 7.1.1. Let the standard assumptions hold, let ¦B
n
¦ be a sequence
of nonsingular N N matrices, let x
0
∈ R
N
be given and let ¦x
n
¦
∞
n=1
be given
by (7.5). Assume that x
n
,= x
∗
for any n. Then x
n
→ x
∗
qsuperlinearly if
and only if x
n
→ x
∗
and the Dennis–Mor´e condition (7.6) holds.
Proof. Since
−F(x
n
) = B
n
s
n
= F
(x
∗
)s
n
+E
n
s
n
we have
E
n
s
n
= −F
(x
∗
)s
n
−F(x
n
) = −F
(x
∗
)e
n+1
+F
(x
∗
)e
n
−F(x
n
). (7.7)
We use the fundamental theorem of calculus and the standard assumptions to
obtain
F
(x
∗
)e
n
−F(x
n
) =
_
1
0
(F
(x
∗
) −F
(x
∗
+te
n
))e
n
dt
and hence
F
(x
∗
)e
n
−F(x
n
) ≤ γe
n

2
/2.
Therefore, by (7.7)
E
n
s
n
 ≤ F
(x
∗
)e
n+1
 +γe
n

2
/2. (7.8)
Now, if x
n
→ x
∗
qsuperlinearly, then for n suﬃciently large
s
n
/2 ≤ e
n
 ≤ 2s
n
. (7.9)
The assumption that x
n
,= x
∗
for any n implies that the sequence
ν
n
= e
n+1
/e
n
 (7.10)
is deﬁned and and ν
n
→ 0 by superlinear convergence of ¦x
n
¦. By (7.9) we
have
e
n+1
 = ν
n
e
n
 ≤ 2ν
n
s
n
,
and hence (7.8) implies that
E
n
s
n
 ≤ (2F
(x
∗
)ν
n
+γe
n
)s
n

which implies the DennisMor´e condition (7.6).
Conversely, assume that x
n
→ x
∗
and that (7.6) holds. Let
µ
n
=
E
n
s
n

s
n

.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
116 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
We have, since
E
n
s
n
= (B
n
−F
(x
∗
))s
n
= −F(x
n
) −F
(x
∗
)s
n
, (7.11)
s
n
F
(x
∗
)
−1

−1
≤ F
(x
∗
)s
n
 ≤ E
n
s
n
 +F(x
n
)
= µ
n
s
n
 +F(x
n
).
(7.12)
Since µ
n
→ 0 by assumption,
µ
n
≤ F
(x
∗
)
−1

−1
/2, (7.13)
for n suﬃciently large. We assume now that n is large enough so that (7.13)
holds. Hence,
s
n
 ≤ 2F
(x
∗
)
−1
F(x
n
). (7.14)
Moreover, using the standard assumptions and (7.11) again,
F(x
n
) +F
(x
n
)s
n
 ≤ F(x
n
) +F
(x
∗
)s
n
 +(F
(x
∗
) −F
(x
n
))s
n

≤ (µ
n
+γe
n
)s
n
.
(7.15)
Since x
n
→ x
∗
by assumption, η
n
→ 0 where
η
n
= 2F
(x
∗
)
−1
(µ
n
+γe
n
).
Combining (7.14) and (7.15) implies that
F(x
n
) +F
(x
n
)s
n
 ≤ η
n
F(x
n
), (7.16)
which is the inexact Newton condition. Since η
n
→ 0, the proof is complete
by Theorem 6.1.2.
7.2. Convergence analysis
Our convergence analysis is based on an approach to inﬁnitedimensional
problems from [97]. That paper was the basis for [168], [104], and [115], and
we employ notation from all four of these references. We will use the l
2
norm
throughout our discussion. We begin with a lemma from [104].
Lemma 7.2.1. Let 0 <
ˆ
θ < 1 and let
¦θ
n
¦
∞
n=0
⊂ (
ˆ
θ, 2 −
ˆ
θ).
Let ¦
n
¦
∞
n=0
⊂ R
N
be such that
n

n

2
< ∞,
and let ¦η
n
¦
∞
n=0
be a set of vectors in R
N
such that η
n

2
is either 1 or 0 for
all n. Let ψ
0
∈ R
N
be given. If ¦ψ
n
¦
∞
n=1
is given by
ψ
n+1
= ψ
n
−θ
n
(η
T
n
ψ
n
)η
n
+
n
(7.17)
then
lim
n→∞
η
T
n
ψ
n
= 0. (7.18)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 117
Proof. We ﬁrst consider the case in which
n
= 0 for all n. In that case, we
can use the fact that
θ
n
(2 −θ
n
) >
ˆ
θ
2
> 0
to show that the sequence ¦ψ
n
¦ is bounded in l
2
norm by ψ
0

2
and, in fact,
ψ
n+1

2
2
≤ ψ
n

2
2
−θ
n
(2 −θ
n
)(η
T
n
ψ
n
)
2
≤ ψ
n

2
2
−
ˆ
θ
2
(η
T
n
ψ
n
)
2
≤ ψ
n

2
2
.
Therefore, for any M > 0,
M
n=0
(η
T
n
ψ
n
)
2
≤
ψ
0

2
2
−ψ
M+1

2
2
ˆ
θ
2
≤
ψ
0

2
ˆ
θ
2
.
We let M → ∞ to obtain
∞
n=0
(η
T
n
ψ
n
)
2
< ∞. (7.19)
Convergence of the series in (7.19) implies convergence to zero of the terms in
the series, which is (7.18).
To prove the result for
n
,= 0 we use the inequality
_
a
2
−b
2
≤ a −
b
2
2a
, (7.20)
which is valid for a > 0 and [b[ ≤ a. This inequality is used often in the analysis
of quasiNewton methods [63]. From (7.20) we conclude that if ψ
n
,= 0 then
ψ
n
−θ
n
(η
T
n
ψ
n
)η
n

2
≤
_
ψ
n

2
2
−θ
n
(2 −θ
n
)(η
T
n
ψ
n
)
2
≤ ψ
n

2
−
θ
n
(2 −θ
n
)(η
T
n
ψ
n
)
2
2ψ
n

2
.
Hence if ψ
n
,= 0
ψ
n+1

2
≤ ψ
n

2
−
θ
n
(2 −θ
n
)(η
T
n
ψ
n
)
2
2ψ
n

2
+
n

2
. (7.21)
Hence
(η
T
n
ψ
n
)
2
≤
2ψ
n

2
θ
n
(2 −θ
n
)
(ψ
n

2
−ψ
n+1

2
+
n

2
), (7.22)
which holds even if ψ
n
= 0.
From (7.21) we conclude that
ψ
n+1

2
≤ µ,
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
118 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
where
µ =
∞
i=0

i

2
+ψ
0

2
.
Hence
M
n=0
(η
T
n
ψ
n
)
2
≤
2µ
ˆ
θ
2
M
n=0
(ψ
n

2
−ψ
n+1

2
+
n

2
)
=
2µ
ˆ
θ
2
_
ψ
0

2
−ψ
M+1

2
+
M
n=0

n

2
_
≤
2µ
2
ˆ
θ
2
.
Hence (7.19) holds and the proof is complete.
7.2.1. Linear problems. We may use Lemma 7.2.1 to prove a result from
[130] on the convergence of Broyden’s method for linear problems. The role of
the parameter
ˆ
θ and the sequence ¦θ
n
¦ in Lemma 7.2.1 is nicely illustrated by
this application of the lemma.
In this section F(x) = Ax−b and B
n
= A+E
n
. The standard assumptions
in this case reduce to nonsingularity of A. Our ﬁrst task is to show that
the errors in the Broyden approximations to A do not grow. This property,
called bounded deterioration [28], is an important part of the convergence
analysis. The reader should compare the statement of the result with that
in the nonlinear case.
In the linear case, the Broyden update has a very simple form. We will
consider a modiﬁed update
B
+
= B
c
+θ
c
(y −B
c
s)s
T
s
T
s
= B
c
+θ
c
(Ax
+
−b)s
T
s
T
s
. (7.23)
If x
c
and B
c
are the current approximations to x
∗
= A
−1
b and A and
s = x
+
−x
c
then
y −B
c
s = (Ax
+
−Ax
c
) −B
c
(x
+
−x
c
) = −E
c
s
and therefore
B
+
= B
c
+θ
c
(y −B
c
s)s
T
s
2
2
= B
c
−θ
c
(E
c
s)s
T
s
2
2
. (7.24)
Lemma 7.2.2. Let A be nonsingular, θ
c
∈ [0, 2], and x
c
∈ R
N
be given. Let
B
c
be nonsingular and let B
+
be formed by the Broyden update (7.23). Then
E
+

2
≤ E
c

2
. (7.25)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 119
Proof. By subtracting A from both sides of (7.24) we have
E
+
= E
c
−θ
c
(E
c
s)s
T
s
2
2
= E
c
(I −θ
c
P
s
),
(7.26)
where P
s
is the orthogonal projector
P
s
=
ss
T
s
2
2
. (7.27)
This completes the proof as
E
+

2
≤ E
c

2
I −θ
c
P
s

2
and I −θ
c
P
s

2
≤ 1 by orthogonality of P
s
and the fact that 0 ≤ θ
c
≤ 2.
Now, θ
c
can always be selected to make B
+
nonsingular. In [130] one
suggestion was
θ
c
=
_
_
_
1, [γ
c
[ ≥ σ,
1 −sign(γ
c
)σ
1 −γ
c
, otherwise,
where
γ
c
=
(B
−1
c
y)
T
s
s
2
2
=
(B
−1
c
As)
T
s
s
2
2
and σ ∈ (0, 1) is ﬁxed. However the results in [130] assume only that the
sequence ¦θ
n
¦ satisﬁes the hypotheses of Lemma 7.2.1 for some
ˆ
θ ∈ (0, 1) and
that θ
c
is always chosen so that B
+
is nonsingular.
For linear problems one can show that the Dennis–Mor´e condition implies
convergence and hence the assumption that convergence takes place is not
needed. Speciﬁcally
Proposition 7.2.1. Let A be nonsingular. Assume that ¦θ
n
¦ satisﬁes
the hypotheses of Lemma 7.2.1 for some
ˆ
θ ∈ (0, 1). If ¦θ
n
¦ is such that
the matrices B
n
obtained by (7.23) are nonsingular and ¦x
n
¦ is given by the
modiﬁed Broyden iteration
x
n+1
= x
n
−B
−1
n
(Ax
n
−b) (7.28)
then the Dennis–Mor´e condition implies convergence of ¦x
n
¦ to x
∗
= A
−1
b.
We leave the proof as Exercise 7.5.1.
The superlinear convergence result is remarkable in that the initial iterate
need not be near the root.
Theorem 7.2.1. Let A be nonsingular. Assume that ¦θ
n
¦ satisﬁes the
hypotheses of Lemma 7.2.1 for some
ˆ
θ ∈ (0, 1). If ¦θ
n
¦ is such the matrices B
n
obtained by (7.23) are nonsingular then the modiﬁed Broyden iteration (7.28)
converges qsuperlinearly to x
∗
= A
−1
b for every initial iterate x
0
.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
120 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Proof. Our approach is to show that for every φ ∈ R
N
that
lim
n→∞
φ
T
_
E
n
s
n
s
n

2
_
= 0. (7.29)
Assuming that (7.29) holds, setting φ = e
j
, the unit vector in the jth
coordinate direction, shows that the j component of
E
n
s
n
s
n

2
has limit zero.
As j was arbitrary, we have
lim
n→∞
E
n
s
n
s
n

2
= 0
which is the Dennis–Mor´e condition. This will imply qsuperlinear convergence
by Theorem 7.1.1 and Proposition 7.2.1.
Let φ ∈ R
N
be given. Let
P
n
=
s
n
s
T
n
s
n

2
2
.
Since for all v ∈ R
N
and all n
v
T
(E
T
n
φ) = φ
T
(E
n
v)
and, by (7.26)
E
T
n+1
φ = (I −θ
n
P
n
)E
T
n
φ,
we may invoke Lemma 7.2.1 with
η
n
= s
n
/s
n

2
, ψ
n
= E
T
n
φ, and
n
= 0
to conclude that
η
T
n
ψ
n
=
(E
T
n
φ)
T
s
n
s
n

2
= φ
T
E
n
s
n
s
n

2
→ 0. (7.30)
This completes the proof.
The result here is not a local convergence result. It is global in the sense
that qsuperlinear convergence takes place independently of the initial choices
of x and B. It is known [29], [82], [84], [143], that the Broyden iteration will
terminate in 2N steps in the linear case.
7.2.2. Nonlinear problems. Our analysis of the nonlinear case is diﬀerent
from the classical work [28], [63]. As indicated in the preface, we provide a
proof based on ideas from [97], [115], and [104], that does not use the Frobenius
norm and extends to various inﬁnitedimensional settings.
For nonlinear problems our analysis is local and we set θ
n
= 1. However
we must show that the sequence ¦B
n
¦ of Broyden updates exists and remains
nonsingular. We make a deﬁnition that allows us to compactly express this
fact.
Definition 7.2.1. We say that the Broyden sequence (¦x
n
¦, ¦B
n
¦) for the
data (F, x
0
, B
0
) exists if F is deﬁned at x
n
for all n, B
n
is nonsingular for all
n, and x
n+1
and B
n+1
can be computed from x
n
and B
n
using (7.1) and (7.2).
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 121
We will show that if the standard assumptions hold and x
0
and B
0
are
suﬃciently good approximations to x
∗
and F
(x
∗
) then the Broyden sequence
exists for the data (F, x
0
, B
0
) and that the Broyden iterates converge q
superlinearly to x
∗
. The development of the proof is complicated. We ﬁrst
prove a bounded deterioration result like Lemma 7.2.2. However in the
nonlinear case, the errors in the Jacobian approximation can grow as the
iteration progresses. We then show that if the initial data is suﬃciently good,
this growth can be limited and therefore the Broyden sequence exists and
the Broyden iterates converges to x
∗
at least qlinearly. Finally we verify the
Dennis–Mor´e condition (7.6), which, together with the qlinear convergence
proved earlier completes the proof of local qsuperlinear convergence.
Bounded deterioration. Our ﬁrst task is to show that bounded deteriora
tion holds. Unlike the linear case, the sequence ¦E
n
¦ need not be monotone
nonincreasing. However, the possible increase can be bounded in terms of the
errors in the current and new iterates.
Theorem 7.2.2. Let the standard assumptions hold. Let x
c
∈ Ω and a
nonsingular matrix B
c
be given. Assume that
x
+
= x
c
−B
−1
c
F(x
c
) = x
c
+s ∈ Ω
and B
+
is given by (7.2). Then
E
+

2
≤ E
c
 +γ(e
c

2
+e
+

2
)/2 (7.31)
Proof. Note that (7.7) implies that
E
c
s = −F(x
+
) + (F(x
+
) −F(x
c
) −F
(x
∗
)s)
= −F(x
+
) +
_
1
0
(F
(x
c
+ts) −F
(x
∗
))s dt.
(7.32)
We begin by writing (7.32) as
F(x
+
) = −E
c
s +
_
1
0
(F
(x
c
+ts) −F
(x
∗
))s dt
and then using (7.2) to obtain
E
+
= E
c
(I −P
s
) +
(∆
c
s)s
T
s
2
2
where P
s
is is given by (7.27) and
∆
c
=
_
1
0
(F
(x
c
+ts) −F
(x
∗
)) dt.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
122 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Hence, by the standard assumptions
∆
c

2
≤ γ(e
c

2
+e
+

2
)/2.
Just as in the proof of Lemma 7.2.2 for the linear case
E
+

2
≤ E
c

2
+∆
c

2
and the proof is complete.
Local qlinear convergence. Theorem 7.2.3 states that Broyden’s method
is locally qlinearly convergent. The proof is more complex than similar results
in Chapters 5 and 6. We must explore the relation between the size of E
0

needed to enforce qlinear convergence and the qfactor.
Theorem 7.2.3. Let the standard assumptions hold. Let r ∈ (0, 1) be
given. Then there are δ and δ
B
such that if x
0
∈ B(δ) and E
0

2
< δ
B
the
Broyden sequence for the data (F, x
0
, B
0
) exists and x
n
→ x
∗
qlinearly with
qfactor at most r.
Proof. Let δ and δ
1
be small enough so that the conclusions of Theo
rem 5.4.3 hold. Then, reducing δ and δ
1
further if needed, the qfactor is at
most
K
A
(δ +δ
1
) ≤ r,
where K
A
is from the statement of Theorem 5.4.3.
We will prove the result by choosing δ
B
and reducing δ if necessary so that
E
0

2
≤ δ
B
will imply that E
n

2
≤ δ
1
for all n, which will prove the result.
To do this reduce δ if needed so that
δ
2
=
γ(1 +r)δ
2(1 −r)
< δ
1
(7.33)
and set
δ
B
= δ
1
−δ
2
. (7.34)
We show that E
n

2
≤ δ
1
by induction. Since E
0

2
< δ
B
< δ
1
we may
begin the induction. Now assume that E
k

2
< δ
1
for all 0 ≤ k ≤ n. By
Theorem 7.2.2,
E
n+1

2
≤ E
n

2
+γ(e
n+1

2
+e
n

2
)/2.
Since E
n

2
< δ
1
, e
n+1

2
≤ re
n

2
and therefore
E
n+1

2
≤ E
n

2
+γ(1 +r)e
n

2
/2 ≤ E
n

2
+γ(1 +r)r
n
δ/2.
We can estimate E
n

2
, E
n−1

2
, . . . in the same way to obtain
E
n+1

2
≤ E
0

2
+
(1 +r)γδ
2
n
j=0
r
j
≤ δ
B
+
(1 +r)γδ
2(1 −r)
≤ δ
1
,
which completes the proof.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 123
Veriﬁcation of the Dennis–Mor´e condition. The ﬁnal task is to verify
the Dennis–Mor´e condition.
Theorem 7.2.4. Let the standard assumptions hold. Then there are δ and
δ
B
such that if x
0
∈ B(δ) and E
0

2
< δ
B
the Broyden sequence for the data
(F, x
0
, B
0
) exists and x
n
→ x
∗
qsuperlinearly.
Proof. Let δ and δ
B
be such that the conclusions of Theorem 7.2.3 hold.
By Theorem 7.1.1 we need only show that Dennis–Mor´e condition (7.6) holds
to complete the proof.
As in the proof of Theorem 7.2.1 let
P
n
=
s
n
s
T
n
s
n

2
2
.
Set
∆
n
=
_
1
0
(F
(x
n
+ts
n
) −F
(x
∗
)) dt.
Let φ ∈ R
N
be arbitrary. Note that
E
T
n+1
φ = (I −P
n
)E
T
n
φ +P
n
∆
T
n
φ.
We wish to apply Lemma 7.2.1 with
ψ
n
= E
T
n
φ, η
n
= s
n
/s
n

2
, and
n
= P
n
∆
T
n
φ.
The hypothesis in the lemma that
n

n

2
< ∞
holds since Theorem 7.2.3 and the standard assumptions imply
∆
n

2
≤ γ(1 +r)r
n
δ/2.
Hence Lemma 7.2.1 implies that
η
T
n
ψ
n
=
(E
T
n
φ)
T
s
n
s
n

2
= φ
T
E
n
s
n
s
n

2
→ 0.
This, as in the proof of Theorem 7.2.1, implies the Dennis–Mor´e condition and
completes the proof.
7.3. Implementation of Broyden’s method
The implementation of Broyden’s method that we advocate here is related
to one that has been used in computational ﬂuid mechanics [73]. Some
such methods are called limited memory formulations in the optimization
literature [138], [142], [31]. In these methods, after the storage available for
the iteration is exhausted, the oldest of the stored vectors is replaced by the
most recent. Another approach, restarting, clears the storage and starts over.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
124 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
This latter approach is similar to the restarted GMRES iteration. Our basic
implementation is a nonlinear version of the implementation for linear problems
in [67]. We use a restarted approach, though a limited memory approach could
be used as well.
We begin by showing how the initial approximation to F
(x
∗
), B
0
, may be
regarded as a preconditioner and incorporated into the deﬁntion of F just as
preconditioners were in the implementation of NewtonGMRES in ¸ 6.4.
Lemma 7.3.1. Assume that the Broyden sequence (¦x
n
¦, ¦B
n
¦) for the data
(F, x
0
, B
0
) exists. Then the Broyden sequence (¦y
n
¦, ¦C
n
¦) exists for the data
(B
−1
0
F, x
0
, I) and
x
n
= y
n
, B
n
= B
0
C
n
(7.35)
for all n.
Proof. We proceed by induction. Equation (7.35) holds by deﬁnition when
n = 0. If (7.35) holds for a given n, then
y
n+1
= y
n
−C
−1
n
B
−1
0
F(y
n
)
= x
n
−(B
0
C
n
)
−1
F(x
n
) = x
n
−B
−1
n
F(x
n
)
= x
n+1
.
(7.36)
Now s
n
= x
n+1
−x
n
= y
n+1
−y
n
by (7.36). Hence
B
n+1
= B
n
+
F(x
n+1
)s
T
n
s
n

2
2
= B
0
C
n
+B
0
B
−1
0
F(y
n+1
)s
T
n
s
n

2
2
= B
0
C
n+1
.
This completes the proof.
The consequence of this proof for implementation is that we can assume
that B
0
= I. If a better choice for B
0
exists, we can incorporate that into the
function evaluation. Our plan will be to store B
−1
n
in a compact and easy to
apply form. The basis for this is the following simple formula [68], [176], [177],
[11].
Proposition 7.3.1. Let B be a nonsingular N N matrix and let
u, v ∈ R
N
. Then B + uv
T
is invertible if and only if 1 + v
T
B
−1
u ,= 0. In
this case,
(B +uv
T
)
−1
=
_
I −
(B
−1
u)v
T
1 +v
T
B
−1
u
_
B
−1
. (7.37)
The expression (7.37) is called the Sherman–Morrison formula. We leave
the proof to Exercise 7.5.2.
In the context of a sequence of Broyden updates ¦B
n
¦ we have for n ≥ 0,
B
n+1
= B
n
+u
n
v
T
n
,
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 125
where
u
n
= F(x
n+1
)/s
n

2
and v
n
= s
n
/s
n

2
.
Setting
w
n
= (B
−1
n
u
n
)/(1 +v
T
n
B
−1
n
u
n
)
we see that, if B
0
= I,
B
−1
n
= (I −w
n−1
v
T
n−1
)(I −w
n−2
v
T
n−2
) . . . (I −w
0
v
T
0
)
=
n−1
j=0
(I −w
j
v
T
j
).
(7.38)
Since the empty matrix product is the identity, (7.38) is valid for n ≥ 0.
Hence the action of B
−1
n
on F(x
n
) (i.e., the computation of the Broyden
step) can be computed from the 2n vectors ¦w
j
, v
j
¦
n−1
j=0
at a cost of O(Nn)
ﬂoatingpoint operations. Moreover, the Broyden step for the following
iteration is
s
n
= −B
−1
n
F(x
n
) = −
n−1
j=0
(I −w
j
v
T
j
)F(x
n
). (7.39)
Since the product
n−2
j=0
(I −w
j
v
T
j
)F(x
n
)
must also be computed as part of the computation of w
n−1
we can combine
the computation of w
n−1
and s
n
as follows:
w =
n−2
j=0
(I −w
j
v
T
j
)F(x
n
)
w
n−1
= C
w
w where C
w
= (s
n−1

2
+v
T
n−1
w)
−1
s
n
= −(I −w
n−1
v
T
n−1
)w.
(7.40)
One may carry this one important step further [67] and eliminate the need
to store the sequence ¦w
n
¦. Note that (7.40) implies that for n ≥ 1
s
n
= −w +C
w
w(v
T
n−1
w) = w(−1 +C
w
(C
−1
w
−s
n−1

2
))
= −C
w
ws
n−1

2
= −s
n−1

2
w
n−1
.
Hence, for n ≥ 0,
w
n
= −s
n+1
/s
n

2
. (7.41)
Therefore, one need only store the steps ¦s
n
¦ and their norms to construct
the sequence ¦w
n
¦. In fact, we can write (7.38) as
B
−1
n
=
n−1
j=0
_
I +
s
j+1
s
T
j
s
j

2
2
_
. (7.42)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
126 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
We cannot use (7.42) and (7.39) directly to compute s
n+1
because s
n+1
appears
on both sides of the equation
s
n+1
= −
_
I +
s
n+1
s
T
n
s
n

2
2
_
n−1
j=0
_
I +
s
j+1
s
T
j
s
j

2
2
_
F(x
n+1
)
= −
_
I +
s
n+1
s
T
n
s
n

2
2
_
B
−1
n
F(x
n+1
).
(7.43)
Instead, we solve (7.43) for s
n+1
to obtain
s
n+1
= −
B
−1
n
F(x
n+1
)
1 +s
T
n
B
−1
n
F(x
n+1
)/s
n

2
2
. (7.44)
By Proposition 7.3.1, the denominator in (7.44)
1 +s
T
n
B
−1
n
F(x
n+1
)/s
n

2
2
= 1 +v
T
n
B
−1
n
u
n
is nonzero unless B
n+1
is singular.
The storage requirement, of n +O(1) vectors for the nth iterate, is of the
same order as GMRES, making Broyden’s method competitive in storage with
GMRES as a linear solver [67]. Our implementation is similar to GMRES(m),
restarting with B
0
when storage is exhausted.
We can use (7.42) and (7.44) to describe an algorithm. The termination
criteria are the same as for Algorithm nsol in ¸ 5.6. We use a restarting
approach in the algorithm, adding to the input list a limit nmax on the number
of Broyden iterations taken before a restart and a limit maxit on the number
of nonlinear iterations. Note that B
0
= I is implicit in the algorithm. Our
looping convention is that the loop in step 2(e)ii is skipped if n = 0; this is
consistent with setting the empty matrix product to the identity. Our notation
and algorithmic framework are based on [67].
Algorithm 7.3.1. brsol(x, F, τ, maxit, nmax)
1. r
0
= F(x)
2
, n = −1,
s
0
= −F(x), itc = 0.
2. Do while itc < maxit
(a) n = n + 1; itc = itc + 1
(b) x = x +s
n
(c) Evaluate F(x)
(d) If F(x)
2
≤ τ
r
r
0
+τ
a
exit.
(e) if n < nmax then
i. z = −F(x)
ii. for j = 0, n −1
z = z +s
j+1
s
T
j
z/s
j

2
2
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 127
iii. s
n+1
= z/(1 −s
T
n
z/s
n

2
2
)
(f) if n = nmax then
n = −1; s = −F(x);
If n < nmax, then the nth iteration of brsol, which computes x
n
and
s
n+1
, requires O(nN) ﬂoatingpoint operations and storage of n + 3 vectors,
the steps ¦s
j
¦
n
j=0
, x, z, and F(x), where z and F(x) may occupy the same
storage location. This requirement of one more vector than the algorithm
in [67] needed is a result of the nonlinearity. If n = nmax, the iteration
will restart with x
nmax
, not compute s
nmax+1
, and therefore need storage for
nmax+2 vectors. Our implementation of brsol in the collection of MATLAB
codes stores z and F separately in the interest of clarity and therefore requires
nmax + 3 vectors of storage.
The Sherman–Morrison approach in Algorithm brsol is more eﬃcient, in
terms of both time and storage, than the dense matrix approach proposed
in [85]. However the dense matrix approach makes it possible to detect
ill conditioning in the approximate Jacobians. If the data is suﬃciently
good, bounded deterioration implies that the Broyden matrices will be well
conditioned and superlinear convergence implies that only a few iterates will
be needed. In view of all this, we feel that the approach of Algorithm brsol
is the best approach, especially for large problems.
A MATLAB implementation of brsol is provided in the collection of
MATLAB codes.
7.4. Examples for Broyden’s method
In all the examples in this section we use the l
2
norm multiplied by 1/
√
N.
We use the MATLAB code brsol.
7.4.1. Linear problems. The theory in this chapter indicates that Broy
den’s method can be viewed as an alternative to GMRES as an iterative method
for nonsymmetric linear equations. This point has been explored in some detail
in [67], where more elaborate numerical experiments that those given here are
presented, and in several papers on inﬁnitedimensional problems as well [91],
[104], [120]. As an example we consider the linear PDE (3.33) from ¸ 3.7. We
use the same mesh, righthand side, coeﬃcients, and solution as the example
in ¸ 3.7.
We compare Broyden’s method without restarts (solid line) to Broyden’s
method with restarts every three iterates (dashed line). For linear problems,
we allow for increases in the nonlinear residual. We set τ
a
= τ
r
= h
2
where
h = 1/32. In Fig. 7.1 we present results for (3.33) preconditioned with the
Poisson solver fish2d.
Broyden’s method terminated after 9 iterations at a cost of roughly 1.5
million ﬂoating point operations, a slightly higher cost than the GMRES
solution. The restarted iteration required 24 iterations and 3.5 million ﬂoating
point operations. One can also see highly irregular performance from the
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
128 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
restarted method.
In the linear case [67] one can store only the residuals and recover the
terminal iterate upon exit. Storage of the residual, the vector z, and the
sequence of steps saves a vector over the nonlinear case.
0 5 10 15 20 25
10
3
10
2
10
1
10
0
10
1
Iterations
R
e
l
a
t
i
v
e
R
e
s
i
d
u
a
l
Fig. 7.1. Broyden’s method for the linear PDE.
7.4.2. Nonlinear problems. We consider the Hequation from Chapters 5
and 6 and a nonlinear elliptic partial diﬀerential equation. Broyden’s method
and NewtonGMRES both require more storage as iterations progress, Newton
GMRES as the inner iteration progresses and Broyden’s method as the outer
iteration progresses. The cost in function evaluations for Broyden’s method is
one for each nonlinear iteration and for NewtonGMRES one for each outer
iteration and one for each inner iteration. Hence, if function evaluations are
very expensive, the cost of a solution by Broyden’s method could be much less
than that of one by NewtonGMRES. If storage is at a premium, however,
and the number of nonlinear iterations is large, Broyden’s method could be at
a disadvantage as it might need to be restarted many times. The examples
in this section, as well as those in ¸ 8.4 illustrate the diﬀerences in the two
methods. Our general recommendation is to try both.
Chandrasekhar Hequation. As before we solve the Hequation on a 100
point mesh with c = .9 (Fig. 7.2) and c = .9999 (Fig. 7.3). We compare
Broyden’s method without restarts (solid line) to Broyden’s method with
restarts every three iterates (dashed line). For c = .9 both methods required six
iterations for convergence. The implementation without restarts required 153
thousand ﬂoatingpoint operations and the restarted method 150, a substantial
improvement over the NewtonGMRES implementation.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 129
For c = .9999, the restarted iteration has much more trouble. The un
restarted iteration converges in 10 iterations at a cost of roughly 249 thousand
ﬂoatingpoint operations. The restarted iteration requires 18 iterations and
407 ﬂoatingpoint operations. Even so, both implementations performed bet
ter than NewtonGMRES.
0 1 2 3 4 5 6
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 7.2. Broyden’s method for the Hequation, c = .9.
0 2 4 6 8 10 12 14 16 18
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 7.3. Broyden’s method for the Hequation, c = .9999.
One should take some care in drawing general conclusions. While Broyden’s
method on this example performed better than NewtonGMRES, the storage
in NewtonGMRES is used in the inner iteration, while that in Broyden in the
nonlinear iteration. If many nonlinear iterations are needed, it may be the case
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
130 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
that Broyden’s method may need restarts while the GMRES iteration for the
linear problem for the Newton step may not.
Convectiondiﬀusion equation. We consider the nonlinear convection
diﬀusion equation from ¸ 6.4.2. We use the same mesh width h = 1/32 on
a uniform square grid. We set C = 20, u
0
= 0, and τ
r
= τ
a
= h
2
. We
preconditioned with the Poisson solver fish2d.
We allowed for an increase in the residual, which happened on the second
iterate. This is not supported by the local convergence theory for nonlinear
problems; however it can be quite successful in practice. One can do this in
brsol by setting the third entry in the parameter list to 1, as one would if the
problem were truly linear. In ¸ 8.4.3 we will return to this example.
In Fig. 7.4 we compare Broyden’s method without restarts (solid line) to
Broyden’s method with restarts every 8 iterates (dashed line). The unrestarted
Broyden iteration converges in 12 iterations at a cost of roughly 2.2 million
ﬂoatingpoint operations. This is a modest improvement over the Newton
GMRES cost (reported in ¸ 6.4.2) of 2.6 million ﬂoatingpoint operations.
However, the residual norms do not monotonically decrease in the Broyden
iteration (we ﬁx this in ¸ 8.3.2) and more storage was used.
0 2 4 6 8 10 12 14 16
10
4
10
3
10
2
10
1
10
0
10
1
Iterations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 7.4. Broyden’s method for nonlinear PDE.
At most ﬁve inner GMRES iterations were required. Therefore the GMRES
inner iteration for the NewtonGMRES approach needed storage for at most
10 vectors (the Krylov basis, s, x
c
, F(x
c
), F(x
c
+ hv)) to accommodate the
NewtonGMRES iteration. The Broyden iteration when restarted every n
iterates, requires storage of n+2 vectors ¦s
j
¦
n−1
j=0
, x, and z (when stored in the
same place as F(x)). Hence restarting the Broyden iteration every 8 iterations
most closely corresponds to the NewtonGMRES requirements. This approach
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 131
took 15 iterations to terminate at a cost of 2.4 million ﬂoatingpoint operations,
but the convergence was extremely irregular after the restart.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
132 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
7.5. Exercises on Broyden’s method
7.5.1. Prove Proposition 7.2.1.
7.5.2. Prove Proposition 7.3.1.
7.5.3. Extend Proposition 7.3.1 by showing that if A is a nonsingular N N
matrix, U is N M, V is M N, then A + UV is nonsingular if and
only if I + V A
−1
U is a nonsingular M M matrix. If this is the case,
then
(A+UV )
−1
= A
−1
−A
−1
U(I +V A
−1
U)
−1
V A
−1
.
This is the Sherman–Morrison–Woodbury formula [68], [199]. See [102]
for further generalizations.
7.5.4. Duplicate the results reported in ¸ 7.4. Vary the parameters in the
equations and the number of vectors stored by Broyden’s method and
report on their eﬀects on performance. What happens in the diﬀerential
equation examples if preconditioning is omitted?
7.5.5. Solve the Hequation with Broyden’s method and c = 1. Set τ
a
= τ
r
=
10
−8
. What qfactor for linear convergence do you see? Can you account
for this by applying the secant method to the equation x
2
= 0? See [53]
for a complete explanation.
7.5.6. Try to solve the nonlinear convectiondiﬀusion equation in ¸ 6.4 with
Broyden’s method using various values for the parameter C. How does
the performance of Broyden’s method diﬀer from NewtonGMRES?
7.5.7. Modify nsol to use Broyden’s method instead of the chord method after
the initial Jacobian evaluation. How does this compare with nsol on the
examples in Chapter 5.
7.5.8. Compare the performance of Broyden’s method and NewtonGMRES on
the Modiﬁed Bratu Problem
−∇
2
u +du
x
+e
u
= 0
on (0, 1) (0, 1) with homogeneous Dirichlet boundary conditions.
Precondition with the Poisson solver fish2d. Experiment with various
mesh widths, initial iterates, and values for the convection coeﬃcient d.
7.5.9. The bad Broyden method [26], so named because of its inferior perfor
mance in practice [63], updates an approximation to the inverse of the
Jacobian at the root so that B
−1
≈ F
(x
∗
)
−1
satisﬁes the inverse secant
equation
B
−1
+
y = s (7.45)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BROYDEN’S METHOD 133
with the rankone update
B
−1
+
= B
−1
c
+
(s −B
−1
c
y)y
T
y
2
2
.
Show that the bad Broyden method is locally superlinearly convergent
if the standard assumptions hold. Try to implement the bad Broyden
method in a storageeﬃcient manner using the ideas from [67] and ¸ 7.3.
Does anything go wrong? See [67] for more about this.
7.5.10. The Schubert or sparse Broyden algorithm [172], [27] is a quasiNewton
update for sparse matrices that enforces both the secant equation and the
sparsity pattern. For problems with tridiagonal Jacobian, for example,
the update is
(B
+
)
ij
= (B
c
)
ij
+
(y −B
c
s)
i
s
j
i+1
k=i−1
s
2
k
for 1 ≤ i, j ≤ N and [i − j[ ≤ 1. Note that only the subdiagonal, su
perdiagonal, and main diagonal are updated. Read about the algorithm
in [172] and prove local superlinear convergence under the standard as
sumptions and the additional assumption that the sparsity pattern of B
0
is the same as that of F
(x
∗
). How would the Schubert algorithm be
aﬀected by preconditioning? Compare your analysis with those in [158],
[125], and [189].
7.5.11. Implement the bad Broyden method, apply it to the examples in ¸ 7.4,
and compare the performance to that of Broyden’s method. Discuss the
diﬀerences in implementation from Broyden’s method.
7.5.12. Implement the Schubert algorithm on an unpreconditioned discretized
partial diﬀerential equation (such as the Bratu problem from Exer
cise 7.5.8) and compare it to NewtonGMRES and Broyden’s method.
Does the relative performance of the methods change as h is decreased?
This is interesting even in one space dimension where all matrices are
tridiagonal. References [101], [93], [92], and [112], are relevant to this
exercise.
7.5.13. Use your result from Exercise 5.7.14 in Chapter 5 to numerically estimate
the qorder of Broyden’s method for some of the examples in this
section (both linear and nonlinear). What can you conclude from your
observations?
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
134 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Chapter 8
Global Convergence
By a globally convergent algorithm we mean an algorithm with the property
that for any initial iterate the iteration either converges to a root of F or fails
to do so in one of a small number of ways. Of the many such algorithms we
will focus on the class of line search methods and the Armijo rule [3], [88],
implemented inexactly [24], [25], [70].
Other methods, such as trust region [153], [154], [181], [129], [63], [24],
[25], [173], [70], and continuation/homotopy methods [1], [2], [109], [159],
[198], can be used to accomplish the same objective. We select the line search
paradigm because of its simplicity, and because it is trivial to add a line search
to an existing locally convergent implementation of Newton’s method. The
implementation and analysis of the line search method we develop in this
chapter do not depend on whether iterative or direct methods are used to
compute the Newton step or upon how or if the Jacobian is computed and
stored. We begin with single equations, where the algorithms can be motivated
and intuition developed with a very simple example.
8.1. Single equations
If we apply Newton’s method to ﬁnd the root x
∗
= 0 of the function
f(x) = arctan(x) with initial iterate x
0
= 10 we ﬁnd that the initial iterate
is too far from the root for the local convergence theory in Chapter 5 to be
applicable. The reason for this is that f
(x) = (1 + x
2
)
−1
is small for large
x; f(x) = arctan(x) ≈ ±π/2, and hence the magnitude of the Newton step
is much larger than that of the iterate itself. We can see this eﬀect in the
sequence of iterates:
10, −138, 2.9 10
4
, −1.5 10
9
, 9.9 10
17
,
a failure of Newton’s method that results from an inaccurate initial iterate.
If we look more closely, we see that the Newton step s = −f(x
c
)/f
(x
c
) is
pointing toward the root in the sense that sx
c
< 0, but the length of s is too
large. This observation motivates the simple ﬁx of reducing the size of the step
until the size of the nonlinear residual is decreased. A prototype algorithm is
given below.
135
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
136 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Algorithm 8.1.1. lines1(x, f, τ)
1. r
0
= [f(x)[
2. Do while [f(x)[ > τ
r
r
0
+τ
a
(a) If f
(x) = 0 terminate with failure.
(b) s = −f(x)/f
(x) (search direction)
(c) x
t
= x +s (trial point)
(d) If [f(x
t
)[ < [f(x)[ then x = x
t
(accept the step)
else
s = s/2 goto 2c (reject the step)
When we apply Algorithm lines1 to our simple example, we obtain the
sequence
10, −8.5, 4.9, −3.8, 1.4, −1.3, 1.2, −.99, .56, −0.1, 9 10
−4
, −6 10
−10
.
The quadratic convergence behavior of Newton’s method is only apparent
very late in the iteration. Iterates 1, 2, 3, and 4 required 3, 3, 2, and 2
steplength reductions. After the fourth iterate, decrease in the size of the
nonlinear residual was obtained with a full Newton step. This is typical of the
performance of the algorithms we discuss in this chapter.
We plot the progress of the iteration in Fig. 8.1. We plot
[arctan(x)/arctan(x
0
)[
as a function of the number of function evaluations with the solid line. The
outer iterations are identiﬁed with circles as in ¸ 6.4. One can see that most of
the work is in the initial phase of the iteration. In the terminal phase, where
the local theory is valid, the minimum number of two function evaluations per
iterate is required. However, even in this terminal phase rapid convergence
takes place only in the ﬁnal three outer iterations.
Note the diﬀerences between Algorithm lines1 and Algorithm nsol. One
does not set x
+
= x +s without testing the step to see if the absolute value of
the nonlinear residual has been decreased. We call this method a line search
because we search along the line segment
(x
c
, x
c
−f(x
c
)/f
(x
c
))
to ﬁnd a decrease in [f(x
c
)[. As we move from the right endpoint to the left,
some authors [63] refer to this as backtracking.
Algorithm lines1 almost always works well. In theory, however, there is
the possibility of oscillation about a solution without convergence [63]. To
remedy this and to make analysis possible we replace the test for simple
decrease with one for suﬃcient decrease. The algorithm is to compute a search
direction d which for us will be the Newton direction
d = −f(x
c
)/f
(x
c
)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 137
0 5 10 15 20 25 30 35
10
10
10
9
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.1. Newton–Armijo iteration for inverse tangent.
and then test steps of the form s = λd, with λ = 2
−j
for some j ≥ 0, until
f(x +s) satisﬁes
[f(x
c
+λd)[ < (1 −αλ)[f(x
c
)[. (8.1)
The condition in (8.1) is called suﬃcient decrease of [f[. The parameter
α ∈ (0, 1) is a small, but positive, number intended to make (8.1) as easy
as possible to satisfy. We follow the recent optimization literature [63] and
set α = 10
−4
. Once suﬃcient decrease has been obtained we accept the step
s = λd. This strategy, from [3] (see also [88]) is called the Armijo rule. Note
that we must now distinguish between the step and the Newton direction,
something we did not have to do in the earlier chapters.
Algorithm 8.1.2. nsola1(x, f, τ)
1. r
0
= [f(x)[
2. Do while [f(x)[ > τ
r
r
0
+τ
a
(a) If f
(x) = 0 terminate with failure.
(b) d = −f(x)/f
(x) (search direction)
(c) λ = 1
i. x
t
= x +λd (trial point)
ii. If [f(x
t
)[ < (1 −αλ)[f(x)[ then x = x
t
(accept the step)
else
λ = λ/2 goto 2(c)i (reject the step)
This is essentially the algorithm implemented in the MATLAB code nsola.
We will analyze Algorithm nsola in ¸ 8.2. We remark here that there is
nothing critical about the reduction factor of 2 in the line search. A factor of
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
138 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
10 could well be better in situations in which small values of λ are needed for
several consecutive steps (such as our example with the arctan function). In
that event a reduction of a factor of 8 would require three passes through the
loop in nsola1 but only a single reduction by a factor of 10. This could be
quite important if the search direction were determined by an inexact Newton
method or an approximation to Newton’s method such as the chord method
or Broyden’s method.
On the other hand, reducing λ by too much can be costly as well. Taking
full Newton steps ensures fast local convergence. Taking as large a fraction as
possible helps move the iteration into the terminal phase in which full steps
may be taken and fast convergence expected. We will return to the issue of
reduction in λ in ¸ 8.3.
8.2. Analysis of the Armijo rule
In this section we extend the onedimensional algorithm in two ways before
proceeding with the analysis. First, we accept any direction that satisﬁes the
inexact Newton condition (6.1). We write this for the Armijo sequence as
F
(x
n
)d
n
+F(x
n
) ≤ η
n
F(x
n
). (8.2)
Our second extension is to allow more ﬂexibility in the reduction of λ. We
allow for any choice that produces a reduction that satisﬁes
σ
0
λ
old
≤ λ
new
≤ σ
1
λ
old
,
where 0 < σ
0
< σ
1
< 1. One such method, developed in ¸ 8.3, is to minimize
an interpolating polynomial based on the previous trial steps. The danger is
that the minimum may be too near zero to be of much use and, in fact, the
iteration may stagnate as a result. The parameter σ
0
safeguards against that.
Safeguarding is an important aspect of many globally convergent methods and
has been used for many years in the context of the polynomial interpolation
algorithms we discuss in ¸ 8.3.1 [54], [63], [86], [133], [87]. Typical values of σ
0
and σ
1
are .1 and .5. Our algorithm nsola incorporates these ideas.
While (8.2) is motivated by the Newtoniterative paradigm, the analysis
here applies equally to methods that solve the equations for the Newton step
directly (so η
n
= 0), approximate the Newton step by a quasiNewton method,
or even use the chord method provided that the resulting direction d
n
satisﬁes
(8.2). In Exercise 8.5.9 you are asked to implement the Armijo rule in the
context of Algorithm nsol.
Algorithm 8.2.1. nsola(x, F, τ, η).
1. r
0
= F(x)
2. Do while F(x) > τ
r
r
0
+τ
a
(a) Find d such that F
(x)d +F(x) ≤ ηF(x)
If no such d can be found, terminate with failure.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 139
(b) λ = 1
i. x
t
= x +λd
ii. If F(x
t
) < (1 −αλ)F(x) then x = x
t
else
Choose σ ∈ [σ
0
, σ
1
]
λ = σλ
goto 2(b)i
Note that step 2a must allow for the possibility that F
is illconditioned
and that no direction can be found that satisﬁes (8.2). If, for example, step
2a were implemented with a direct factorization method, illconditioning of F
might be detected as part of the factorization.
Let ¦x
n
¦ be the iteration produced by Algorithm nsola with initial iterate
x
0
. The algorithm can fail in some obvious ways:
1. F
(x
n
) is singular for some n. The inner iteration could terminate with
a failure.
2. x
n
→ ¯ x, a local minimum of F which is not a root.
3. x
n
 → ∞.
Clearly if F has no roots, the algorithm must fail. We will see that if F has no
roots then either F
(x
n
) has a limit point which is singular or ¦x
n
¦ becomes
unbounded.
The convergence theorem is the remarkable statement that if ¦x
n
¦ and
¦F
(x
n
)
−1
¦ are bounded (thereby eliminating premature failure and local
minima of F that are not roots) and F
is Lipschitz continuous in a
neighborhood of the entire sequence ¦x
n
¦, then the iteration converges to a
root of F at which the standard assumptions hold, full steps are taken in the
terminal phase, and the convergence is qquadratic.
We begin with a formal statement of our assumption that F
is uniformly
Lipschitz continuous and bounded away from zero on the sequence ¦x
n
¦.
Assumption 8.2.1. There are r, γ, m
f
> 0 such that F is deﬁned, F
is
Lipschitz continuous with Lipschitz constant γ, and F
(x)
−1
 ≤ m
f
on the
set
Ω(¦x
n
¦, r) =
∞
_
n=0
¦x[ x −x
n
 ≤ r¦.
Assumption 8.2.1 implies that the line search phase (step 2b in Algorithm
nsola) will terminate in a number of cycles that is independent of n. We show
this, as do [25] and [70], by giving a uniform lower bound on λ
n
(assuming
that F(x
0
) ,= 0). Note how the safeguard bound σ
0
enters this result. Note
also the very modest conditions on η
n
.
Lemma 8.2.1. Let x
0
∈ R
N
and α ∈ (0, 1) be given. Assume that ¦x
n
¦ is
given by Algorithm nsola with
¦η
n
¦ ⊂ (0, ¯ η] ⊂ (0, 1 −α) (8.3)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
140 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
and that Assumption 8.2.1 holds. Then either F(x
0
) = 0 or
λ
n
≥
¯
λ = σ
0
min
_
r
m
f
F(x
0
)
,
2(1 −α − ¯ η)
m
2
f
F(x
0
)(1 + ¯ η)
2
γ
_
. (8.4)
Proof. Let n ≥ 0. By the fundamental theorem of calculus and (8.2) we
have, for any λ ∈ [0, 1]
F(x
n
+λd
n
) = F(x
n
) +λF
(x
n
)d
n
+
_
1
0
(F
(x
n
+tλd
n
) −F
(x
n
))λd
n
dt
= (1 −λ)F(x
n
) +λξ
n
+
_
1
0
(F
(x
n
+tλd
n
) −F
(x
n
))λd
n
dt,
where, by the bound η
n
≤ ¯ η,
ξ
n
 ≤ η
n
F(x
n
) ≤ ¯ ηF(x
n
).
The step acceptance condition in nsola implies that ¦F(x
n
)¦ is a decreasing
sequence and therefore
d
n
 = F
(x
n
)
−1
(ξ
n
−F(x
n
)) ≤ m
f
(1 + ¯ η)F(x
n
).
Therefore, by Assumption 8.2.1 F
is Lipschitz continuous on the line segment
[x
n
, x
n
+λd
n
] whenever
λ ≤
¯
λ
1
= r/(m
f
F(x
0
)).
Lipschitz continuity of F
implies that if λ ≤
¯
λ
1
then
F(x
n
+λd
n
) ≤ (1 −λ)F(x
n
) +λ¯ ηF(x
n
) +
γ
2
λ
2
d
n

2
≤ (1 −λ)F(x
n
)[ +λ¯ ηF(x
n
) +
m
2
f
γ(1 + ¯ η)
2
λ
2
F(x
n
)
2
2
≤ (1 −λ + ¯ ηλ)F(x
n
) +λF(x
n
)
m
2
f
γλ(1 + ¯ η)
2
F(x
0
)
2
< (1 −αλ)F(x
n
),
which will be implied by
λ ≤
¯
λ
2
= min
_
¯
λ
1
,
2(1 −α − ¯ η)
(1 + ¯ η)
2
m
2
f
γF(x
0
)
_
.
This shows that λ can be no smaller than σ
0
¯
λ
2
, which completes the proof.
Now we can show directly that the sequence of Armijo iterates either is
unbounded or converges to a solution. Theorem 8.2.1 says even more. The
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 141
sequence converges to a root at which the standard assumptions hold, so in the
terminal phase of the iteration the step lengths are all 1 and the convergence
is qquadratic.
Theorem 8.2.1. Let x
0
∈ R
N
and α ∈ (0, 1) be given. Assume that ¦x
n
¦
is given by Algorithm nsola, ¦η
n
¦ satisﬁes (8.3), ¦x
n
¦ is bounded, and that
Assumption 8.2.1 holds. Then ¦x
n
¦ converges to a root x
∗
of F at which the
standard assumptions hold, full steps are taken for n suﬃciently large, and the
convergence behavior in the ﬁnal phase of the iteration is that given by the local
theory for inexact Newton methods (Theorem 6.1.2).
Proof. If F(x
n
) = 0 for some n, then we are done because Assump
tion 8.2.1 implies that the standard assumptions hold at x
∗
= x
n
. Otherwise
Lemma 8.2.1 implies that F(x
n
) converges qlinearly to zero with qfactor at
most (1 −α
¯
λ).
The boundedness of the sequence ¦x
n
¦ implies that a subsequence ¦x
n
k
¦
converges, say to x
∗
. Since F is continuous, F(x
∗
) = 0. Eventually
[x
n
k
− x
∗
[ < r, where r is the radius in Assumption 8.2.1 and therefore the
standard assumptions hold at x
∗
.
Since the standard assumptions hold at x
∗
, there is δ such that if x ∈ B(δ),
the Newton iteration with x
0
= x will remain in B(δ) and converge q
quadratically to x
∗
. Hence as soon as x
n
k
∈ B(δ), the entire sequence, not
just the subsequence, will remain in B(δ) and converge to x
∗
. Moreover,
Theorem 6.1.2 applies and hence full steps can be taken.
At this stage we have shown that if the Armijo iteration fails to converge
to a root either the continuity properties of F or nonsingularity of F
break
down as the iteration progresses (Assumption 8.2.1 is violated) or the iteration
becomes unbounded. This is all that one could ask for in terms of robustness
of such an algorithm.
Thus far we have used for d the inexact Newton direction. It could well be
advantageous to use a chord or Broyden direction, for example. All that one
needs to make the theorems and proofs in ¸ 8.2 hold is (8.2), which is easy to
verify, in principle, via a forward diﬀerence approximation to F
(x
n
)d.
8.3. Implementation of the Armijo rule
The MATLAB code nsola is a modiﬁcation of nsolgm which provides for a
choice of several Krylov methods for computing an inexact Newton direction
and globalizes Newton’s method with the Armijo rule. We use the l
2
norm in
that code and in the numerical results reported in ¸ 8.4. If GMRES is used
as the linear solver, then the storage requirements are roughly the same as for
Algorithm nsolgm if one reuses the storage location for the ﬁnite diﬀerence
directional derivative to test for suﬃcient decrease.
In Exercise 8.5.9 you are asked to do this with nsol and, with each
iteration, numerically determine if the chord direction satisﬁes the inexact
Newton condition. In this case the storage requirements are roughly the same
as for Algorithm nsol. The iteration can be very costly if the Jacobian must be
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
142 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
evaluated and factored many times during the phase of the iteration in which
the approximate solution is far from the root.
We consider two topics in this section. The ﬁrst, polynomial interpolation
as a way to choose λ, is applicable to all methods. The second, how Broyden’s
method must be implemented, is again based on [67] and is more specialized.
Our Broyden–Armijo code does not explicitly check the inexact Newton
condition and thereby saves an evaluation of F. Evaluation of F
(x
c
)d
c
if the
full step is not accepted (here d
c
is the Broyden direction) would not only verify
the inexact Newton condition but provide enough information for a twopoint
parabolic line search. The reader is asked to pursue this in Exercise 8.5.10.
8.3.1. Polynomial line searches. In this section we consider an elabora
tion of the simple line search scheme of reduction of the steplength by a ﬁxed
factor. The motivation for this is that some problems respond well to one or
two reductions in the steplength by modest amounts (such as 1/2) and others
require many such reductions, but might respond well to a more aggressive
steplength reduction (by factors of 1/10, say). If one can model the decrease
in F
2
as the steplength is reduced, one might expect to be able to better
estimate the appropriate reduction factor. In practice such methods usually
perform better than constant reduction factors.
If we have rejected k steps we have in hand the values
F(x
n
)
2
, F(x
n
+λ
1
d
n
)
2
, . . . F(x
n
+λ
k−1
d
n
)
2
.
We can use this iteration history to model the scalar function
f(λ) = F(x
n
+λd
n
)
2
2
with a polynomial and use the minimum of that polynomial as the next
steplength. We consider two ways of doing this that use second degree
polynomials which we compute using previously computed information.
After λ
c
has been rejected and a model polynomial computed, we compute
the minimum λ
t
of that polynomial analytically and set
λ
+
=
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
σ
0
λ
c
if λ
t
< σ
0
λ
c
,
σ
1
λ
c
if λ
t
> σ
1
λ
c
,
λ
t
otherwise.
(8.5)
Twopoint parabolic model. In this approach we use values of f and f
at λ = 0 and the value of f at the current value of λ to construct a 2nd degree
interpolating polynomial for f.
f(0) = F(x
n
)
2
2
is known. Clearly f
(0) can be computed as
f
(0) = 2(F
(x
n
)
T
d
n
)
T
F(x
n
) = 2F(x
n
)
T
(F
(x
n
)d
n
). (8.6)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 143
Use of (8.6) requires evaluation of F
(x
n
)d
n
, which can be obtained by
examination of the ﬁnal residual from GMRES or by expending an additional
function evaluation to compute F
(x
n
)d
n
with a forward diﬀerence. In any
case, our approximation to f
(0) should be negative. If it is not, it may
be necessary to compute a new search direction. This is a possibility with
directions other than inexact Newton directions, such as the Broyden direction.
The polynomial p(λ) such that p, p
agree with f, f
at 0 and p(λ
c
) = f(λ
c
)
is
p(λ) = f(0) +f
(0)λ +cλ
2
,
where
c =
f(λ
c
) −f(0) −f
(0)λ
c
λ
2
c
.
Our approximation of f
(0) < 0, so if f(λ
c
) > f(0), then c > 0 and p(λ) has a
minimum at
λ
t
= −f
(0)/(2c) > 0.
We then compute λ
+
with (8.5).
Threepoint parabolic model. An alternative to the twopoint model that
avoids the need to approximate f
(0) is a threepoint model, which uses f(0)
and the two most recently rejected steps to create the parabola. The MATLAB
code parab3p implements the threepoint parabolic line search and is called
by the two codes nsola and brsola.
In this approach one evaluates f(0) and f(1) as before. If the full step is
rejected, we set λ = σ
1
and try again. After the second step is rejected, we
have the values
f(0), f(λ
c
), and f(λ
−
),
where λ
c
and λ
−
are the most recently rejected values of λ. The polynomial
that interpolates f at 0, λ
c
, λ
−
is
p(λ) = f(0) +
λ
λ
c
−λ
−
_
(λ −λ
−
)(f(λ
c
) −f(0))
λ
c
+
(λ
c
−λ)(f(λ
−
) −f(0))
λ
−
_
.
We must consider two situations. If
p
(0) =
2
λ
c
λ
−
(λ
c
−λ
−
)
(λ
−
(f(λ
c
) −f(0)) −λ
c
(f(λ
−
) −f(0)))
is positive, then we set λ
t
to the minimum of p
λ
t
= −p
(0)/p
(0)
and apply the safeguarding step (8.5) to compute λ
+
. If p
(0) ≤ 0 one could
either set λ
+
to be the minimum of p on the interval [σ
0
λ, σ
1
λ] or reject the
parabolic model and simply set λ
+
to σ
0
λ
c
or σ
1
λ
c
. We take the latter approach
and set λ
+
= σ
1
λ
c
. In the MATLAB code nsola from the collection, we
implement this method with σ
0
= .1, σ
1
= .5.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
144 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Interpolation with a cubic polynomial has also been suggested [87], [86],
[63], [133]. Clearly, the polynomial modeling approach is heuristic. The
safeguarding and Theorem 8.2.1 provide insurance that progress will be made
in reducing F.
8.3.2. Broyden’s method. In Chapter 7 we implemented Broyden’s
method at a storage cost of one vector for each iteration. The relation
y − B
c
s = F(x
+
) was not critical to this and we may also incorporate a line
search at a cost of a bit of complexity in the implementation. As in ¸ 7.3, our
approach is a nonlinear version of the method in [67].
The diﬀerence from the development in ¸ 7.3 is that the simple relation
between the sequence ¦w
n
¦ and ¦v
n
¦ is changed. If a line search is used, then
s
n
= −λ
n
B
−1
n
F(x
n
)
and hence
y
n
−B
n
s
n
= F(x
n+1
) −(1 −λ
n
)F(x
n
). (8.7)
If, using the notation in ¸ 7.3, we set
u
n
=
y
n
−B
n
s
n
s
n

2
, v
n
=
s
n
s
n

2
, and w
n
= (B
−1
n
u
n
)/(1 +v
T
n
B
−1
n
u
n
),
we can use (8.7) and (7.38) to obtain
w
n
= λ
n
_
−d
n+1
s
n

2
+ (λ
−1
n
−1)
s
n
s
n

2
_
= λ
n
_
−s
n+1
/λ
n+1
s
n

2
+ (λ
−1
n
−1)
s
n
s
n

2
_
,
(8.8)
where
d
n+1
= −B
−1
n+1
F(x
n+1
)
is the search direction used to compute x
n+1
. Note that (8.8) reduces to (7.41)
when λ
n
= 1 and λ
n+1
= 1 (so d
n+1
= s
n+1
).
We can use the ﬁrst equality in (8.8) and the relation
d
n+1
= −
_
I −
w
n
s
T
n
s
n

2
_
B
−1
n
F(x
n+1
)
to solve for d
n+1
and obtain an analog of (7.44)
d
n+1
= −
s
n

2
2
B
−1
n
F(x
n+1
) −(1 −λ
n
)s
T
n
B
−1
n
F(x
n+1
)s
n
s
n

2
2
+λ
n
s
T
n
B
−1
n
F(x
n+1
)
(8.9)
We then use the second equality in (8.8) to construct B
−1
c
F(x
+
) as we did in
Algorithm brsol. Veriﬁcation of (8.8) and (8.9) is left to Exercise 8.5.5.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 145
Algorthm brsola uses these ideas. We do not include details of the line
search or provide for restarts or a limited memory implementation. The
MATLAB implementation, provided in the collection of MATLAB codes, uses
a threepoint parabolic line search and performs a restart when storage is
exhausted like Algorithm brsol does.
It is possible that the Broyden direction fails to satisfy (8.2). In fact, there
is no guarantee that the Broyden direction is a direction of decrease for F
2
.
The MATLAB code returns with an error message if more than a ﬁxed number
of reductions in the step length are needed. Another approach would be to
compute F
(x
c
)d numerically if a full step is rejected and then test (8.2) before
beginning the line search. As a side eﬀect, an approximation of f
(0) can be
obtained at no additional cost and a parabolic line search based on f(0), f
(0),
and f(1) can be used. In Exercise 8.5.10 the reader is asked to fully develop
and implement these ideas.
Algorithm 8.3.1. brsola(x, F, τ, maxit, nmax)
1. Initialize:
F
c
= F(x) r
0
= F(x)
2
, n = −1,
d = −F(x), compute λ
0
, s
0
= λ
0
d
2. Do while n ≤ maxit
(a) n = n + 1
(b) x = x +s
n
(c) Evaluate F(x)
(d) If F(x)
2
≤ τ
r
r
0
+τ
a
exit.
(e) i. z = −F(x)
ii. for j = 0, n −1
a = −λ
j
/λ
j+1
, b = 1 −λ
j
z = z + (as
j+1
+bs
j
)s
T
j
z/s
j

2
2
(f) d = (s
n

2
2
z + (1 −λ
n
)s
T
n
zs
n
)/(s
n

2
2
−λ
n
s
T
n
z)
(g) Compute λ
n+1
with a line search.
(h) Set s
n+1
= λ
n+1
d, store s
n+1

2
and λ
n+1
.
Since d
n+1
and s
n+1
can occupy the same location, the storage require
ments of Algorithm brsola would appear to be essentially the same as those
of brsol. However, computation of the step length requires storage of both the
current point x and the trial point x + λs before a step can be accepted. So,
the storage of the globalized methods exceeds that of the local method by one
vector. The MATLAB implementation of brsola in the collection illustrates
this point.
8.4. Examples for Newton–Armijo
The MATLAB code nsola is a modiﬁcation of nsolgm that incorporates the
threepoint parabolic line search from ¸ 8.3.1 and also changes η using (6.20)
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
146 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
once the iteration is near the root. We compare this strategy, using GMRES
as the linear solver, with the constant η method as we did in ¸ 6.4 with γ = .9.
As before, in all the ﬁgures we plot F(x
n
)
2
/F(x
0
)
2
against the number
of function evaluations required by all line searches and inner and outer
iterations to compute x
n
. Counts of function evaluations corresponding to
outer iterations are indicated by circles. We base our absolute error criterion
on the norm  
2
/
√
N as we did in ¸ 6.4.2 and 7.4.
In ¸ 8.4.3 we compare the Broyden–Armijo method with NewtonGMRES
for those problems for which both methods were successful. We caution the
reader now, and will repeat the caution later, that if the initial iterate is far
from the solution, an inexact Newton method such as NewtonGMRES can
succeed in many case in which a quasiNewton method can fail because the
quasiNewton direction may not be an inexact Newton direction. However,
when both methods succeed, the quasiNewton method, which requires a single
function evaluation for each outer iteration when full steps can be taken, may
well be most eﬃcient.
8.4.1. Inverse tangent function. Since we have easy access to analytic
derivatives in this example, we can use the twopoint parabolic line search.
We compare the twopoint parabolic line search with the constant reduction
search (σ
0
= σ
1
= .5) for the arctan function. In Fig. 8.2 we plot the iteration
history corresponding to the parabolic line search with the solid line and that
for the constant reduction with the dashed line. We use x
0
= 10 as the initial
iterate with τ
r
= τ
a
= 10
−8
. The parabolic line search required 7 outer iterates
and 21 function evaluations in contrast with the constant reduction searches 11
outer iterates and 33 function evaluations. In the ﬁrst outer iteration, both line
searches take three steplength reductions. However, the parabolic line search
takes only one reduction in the next three iterations and none thereafter. The
constant reduction line search took three reductions in the ﬁrst two outer
iterations and two each in following two.
8.4.2. Convectiondiﬀusion equation. We consider a much more diﬃ
cult problem. We take (6.21) from ¸ 6.4.2,
−∇
2
u +Cu(u
x
+u
y
) = f
with the same right hand side f, initial iterate u
0
= 0, 31 31 mesh and
homogeneous Dirichlet boundary conditions on the unit square (0, 1) (0, 1)
that we considered before. Here, however, we set C = 100. This makes a
globalization strategy critical (Try it without one!). We set τ
r
= τ
a
= h
2
/10.
This is a tighter tolerance that in ¸ 6.4.2 and, because of the large value of C
and resulting illconditioning, is needed to obtain an accurate solution.
In Fig. 8.3 we show the progress of the iteration for the unpreconditioned
equation. For this problem we plot the progress of the iteration using η = .25
with the solid line and using the sequence given by (6.20) with the dashed
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 147
0 5 10 15 20 25 30 35
10
12
10
10
10
8
10
6
10
4
10
2
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.2. Newton–Armijo for the arctan function.
line. We set the parameters in (6.20) to γ = .9 and η
max
= .25. This problem
is very poorly conditioned and much tighter control on the inner iteration is
needed than for the other problems. The two approaches to selection of the
forcing term performed very similarly. The iterations required 75.3 (constant
η) and 75.4 (varying η) million ﬂoating point operations, 759 (constant η) and
744 (varying η) function evaluations, and 25 (constant η) and 22 (varying η)
outer iterations.
0 100 200 300 400 500 600 700 800
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.3. Newton–Armijo for the PDE, C = 100.
We consider the preconditioned problem in Fig. 8.4. In the computation we
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
148 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
preconditioned (6.21) with the fast Poisson solver fish2d. In Fig. 8.4 we show
the progress of the iteration for the preconditioned equation. For this problem
we plot the progress of the iteration using η = .25 with the solid line and using
the sequence given by (6.20) with the dashed line. We set the parameters in
(6.20) to γ = .9 and η
max
= .99.
The constant η iteration terminated after 79 function evaluations, 9 outer
iterations, and roughly 13.5 million ﬂoatingpoint operations. The line search
in this case reduced the step length three times on the ﬁrst iteration and twice
on the second and third.
The iteration in which ¦η
n
¦ is given by (6.20) terminated after 70 function
evaluations, 9 outer iterations, and 12.3 million ﬂoatingpoint operations. The
line search in the nonconstant η case reduced the step length three times on
the ﬁrst iteration and once on the second. The most eﬃcient iteration, with
the forcing term given by (6.20), required at most 16 inner iterations while the
constant η approach needed at most 10.
0 10 20 30 40 50 60 70 80
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.4. Newton–Armijo for the PDE, C = 100.
8.4.3. Broyden–Armijo. We found that the Broyden–Armijo line search
failed on all the unpreconditioned convection diﬀusion equation examples. In
these cases the Jacobian is not well approximated by a low rank perturbation
of the identity [112] and the ability of the inexact Newton iteration to ﬁnd a
good search direction was the important factor.
We begin by revisiting the example in ¸ 6.4.2 with C = 20. As reported
in ¸ 6.4.2, the NewtonGMRES iteration always took full steps, terminating
in 4 outer iterations, 16 function evaluations, and 2.6 million ﬂoatingpoint
operations. When compared to the Broyden’s method results in ¸ 6.4.2,
when increases in the residual were allowed, the Broyden–Armijo costs reﬂect
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 149
an improvement in eﬃciency. The Broyden iteration in ¸ 6.4.2 took 12
iterations and 2.8 million ﬂoatingpoint operations while the Broyden–Armijo
took 9 iterations and required 2 million ﬂoatingpoint operations. We set
τ
a
= τ
r
= h
2
, η
max
= .5, and γ = .9 as we did in ¸ 6.4.2.
In our implementation of brsola the threepoint parabolic line search was
used. The steplength was reduced on the second iteration (twice) and the third
(once). Figure 8.5 compares the Broyden–Armijo iteration (solid line) to the
GMRES iteration (dashed line).
0 2 4 6 8 10 12 14 16
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.5. NewtonGMRES and Broyden–Armijo for the PDE, C = 20.
For the more diﬃcult problem with C = 100, the performance of the
two methods is more similar. In Fig. 8.6 we compare our implementation
of Newton–GMRES–Armijo (dashed line) to Broyden–Armijo (solid line). We
set τ
a
= τ
r
= h
2
/10, η
max
= .99, and γ = .9. The Broyden–Armijo iteration
required 34 nonlinear iterations, roughly 16 million ﬂoatingpoint operations,
and 85 function evaluations. In terms of storage, the Broyden–Armijo iteration
required storage for 37 vectors while the NewtonGMRES iteration needed at
most 16 inner iterations, needing to store 21 vectors, and therefore was much
more eﬃcient in terms of storage.
Restarting the Broyden iteration every 19 iterations would equalize the
storage requirements with Newton–GMRES–Armijo. Broyden’s method suf
fers under these conditions as one can see from the comparison of Broyden–
Armijo (solid line) and Newton–GMRES–Armijo (dashed line) in Fig. 8.7. The
restarted Broyden–Armijo iteration took 19.5 million ﬂoatingpoint operations,
42 outer iterates, and 123 function evaluations.
From these examples we see that Broyden–Armijo can be very eﬃcient pro
vided only a small number of iterations are needed. The storage requirements
increase as the nonlinear iteration progresses and this fact puts Broyden’s
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
150 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
0 10 20 30 40 50 60 70 80 90
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.6. NewtonGMRES and Broyden–Armijo for the PDE, C = 100.
0 20 40 60 80 100 120 140
10
5
10
4
10
3
10
2
10
1
10
0
Function Evaluations
R
e
l
a
t
i
v
e
N
o
n
l
i
n
e
a
r
R
e
s
i
d
u
a
l
Fig. 8.7. NewtonGMRES and restarted Broyden–Armijo for the PDE, C = 100.
method at a disadvantage to NewtonGMRES when the number of nonlinear
iterations is large.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
GLOBAL CONVERGENCE 151
8.5. Exercises on global convergence
8.5.1. How does the method proposed in [10] diﬀer from the one implemented in
nsola? What could be advantages and disadvantages of that approach?
8.5.2. In what ways are the results in [25] and [70] more general than those in
this section? What ideas do these papers share with the analysis in this
section and with [3] and [88]?
8.5.3. Implement the Newton–Armijo method for single equations. Apply your
code to the following functions with x
0
= 10. Explain your results.
1. f(x) = arctan(x) (i.e., Duplicate the results in ¸ 8.1.)
2. f(x) = arctan(x
2
)
3. f(x) = .9 + arctan(x)
4. f(x) = x(1 + sin(1/x))
5. f(x) = e
x
6. f(x) = 2 + sin(x)
7. f(x) = 1 +x
2
8.5.4. A numerical analyst buys a German sports car for $50,000. He puts
$10,000 down and takes a 7 year installment loan to pay the balance. If
the monthly payments are $713.40, what is the interest rate? Assume
monthly compounding.
8.5.5. Verify (8.8) and (8.9).
8.5.6. Show that if F
is Lipschitz continuous and the iteration ¦x
n
¦ produced
by Algorithm nsola converges to x
∗
with F(x
∗
) ,= 0, then F
(x
∗
) is
singular.
8.5.7. Use nsola to duplicate the results in ¸ 8.4.2. Vary the convection
coeﬃcient in the convectiondiﬀusion equation and the mesh size and
report the results.
8.5.8. Experiment with other linear solvers such as GMRES(m), BiCGSTAB,
and TFQMR. This is interesting in the locally convergent case as well.
You might use the MATLAB code nsola to do this.
8.5.9. Modify nsol, the hybrid Newton algorithm from Chapter 5, to use the
Armijo rule. Try to do it in such a way that the chord direction is used
whenever possible.
8.5.10. Modify brsola to test the Broyden direction for the descent property
and use a twopoint parabolic line search. What could you do if the
Broyden direction is not a descent direction? Apply your code to the
examples in ¸ 8.4.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
152 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
8.5.11. Modify nsola to use a cubic polynomial and constant reduction line
searches instead of the quadratic polynomial line search. Compare the
results with the examples in ¸ 8.4.
8.5.12. Does the secant method for equations in one variable always give a
direction that satisﬁes (8.2) with η
n
bounded away from 1? If not, when
does it? How would you implement a secantArmijo method in such a
way that the convergence theorem 8.2.1 is applicable?
8.5.13. Under what conditions will the iteration given by nsola converge to a
root x
∗
that is independent of the initial iterate?
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Bibliography
[1] E. L. Allgower and K. Georg, Simplicial and continuation methods for
approximating ﬁxed points and solutions to systems of equations, SIAM Rev., 22
(1980), pp. 28–85.
[2] , Numerical Continuation Methods : An Introduction, SpringerVerlag, New
York, 1990.
[3] L. Armijo, Minimization of functions having Lipschitzcontinuous ﬁrst partial
derivatives, Paciﬁc J. Math., 16 (1966), pp. 1–3.
[4] W. E. Arnoldi, The principle of minimized iterations in the solution of the
matrix eigenvalue problem, Quart. Appl. Math., 9 (1951), pp. 17–29.
[5] S. F. Ashby, T. A. Manteuffel, and J. S. Otto, A comparison of
adaptive Chebyshev and least squares polynomial preconditioning for Hermetian
positive deﬁnite linear systems, SIAM J. Sci. Statist. Comput., 13 (1992), pp. 1–
29.
[6] K. E. Atkinson, Iterative variants of the Nystr¨ om method for the numerical
solution of integral equations, Numer. Math., 22 (1973), pp. 17–31.
[7] , An Introduction to Numerical Analysis, 2nd. ed., John Wiley, New York,
1989.
[8] O. Axelsson, Iterative Solution Methods, Cambridge University Press, Cam
bridge, 1994.
[9] S. Banach, Sur les op´erations dans les ensembles abstraits et leur applications
aux ´equations int´egrales, Fund. Math, 3 (1922), pp. 133–181.
[10] R. E. Bank and D. J. Rose, Global approximate Newton methods, Numer.
Math., 37 (1981), pp. 279–295.
[11] M. S. Barlett, An inverse matrix adjustment arising in discriminant analysis,
Ann. Math. Statist., 22 (1951), pp. 107–111.
[12] R. Barrett, M. Berry, T. F. Chan, J. Demmel, J. Donato, J. Don
garra, V. Eijkhout, R. Pozo, C. Romine, and H. van der Vorst, Templates
for the Solution of Linear Systems: Building Blocks for Iterative Methods, Society
for Industrial and Applied Mathematics, Philadelphia, PA, 1993.
[13] N. Bi´ cani´ c and K. H. Johnson, Who was ‘Raphson’?, Internat. J. Numer.
Methods. Engrg., 14 (1979), pp. 148–152.
[14] P. B. Bosma and W. A. DeRooij, Eﬃcient methods to calculate Chan
drasekhar’s Hfunctions, Astron. Astrophys., 126 (1983), p. 283.
[15] H. Brakhage,
¨
Uber die numerische Behandlung von Integralgleichungen nach
der Quadraturformelmethode, Numer. Math., 2 (1960), pp. 183–196.
[16] K. E. Brenan, S. L. Campbell, and L. R. Petzold, Numerical Solution
153
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
154 BIBLIOGRAPHY
of Initial Value Problems in DiﬀerentialAlgebraic Equations, no. 14 in Classics in
Applied Mathematics, SIAM, Philadelphia, 1996.
[17] R. Brent, Algorithms for Minimization Without Deriviatives, PrenticeHall,
Englewood Cliﬀs, NJ, 1973.
[18] , Some eﬃcient algorithms for solving systems of nonlinear equations, SIAM
J. Numer. Anal., 10 (1973), pp. 327–344.
[19] W. Briggs, A Multigrid Tutorial, Society for Industrial and Applied Mathemat
ics, Philadelphia, PA, 1987.
[20] P. N. Brown, A local convergence theory for combined inexact–Newton/ ﬁnite–
diﬀerence projection methods, SIAM J. Numer. Anal., 24 (1987), pp. 407–434.
[21] P. N. Brown, G. D. Byrne, and A. C. Hindmarsh, VODE: A variable
coeﬃcient ode solver, SIAM J. Sci. Statist. Comput., 10 (1989), pp. 1038–1051.
[22] P. N. Brown and A. C. Hindmarsh, Reduced storage matrix methods in stiﬀ
ODE systems, J. Appl. Math. Comput., 31 (1989), pp. 40–91.
[23] P. N. Brown, A. C. Hindmarsh, and L. R. Petzold, Using Krylov methods
in the solution of largescale diﬀerentialalgebraic systems, SIAM J. Sci. Comput., 15
(1994), pp. 1467–1488.
[24] P. N. Brown and Y. Saad, Hybrid Krylov methods for nonlinear systems of
equations, SIAM J. Sci. Statist. Comput., 11 (1990), pp. 450–481.
[25] , Convergence theory of nonlinear NewtonKrylov algorithms, SIAM J.
Optim., 4 (1994), pp. 297–330.
[26] C. G. Broyden, A class of methods for solving nonlinear simultaneous equa
tions, Math. Comput., 19 (1965), pp. 577–593.
[27] , The convergence of an algorithm for solving sparse nonlinear systems,
Math. Comput., 25 (1971), pp. 285–294.
[28] C. G. Broyden, J. E. Dennis, and J. J. Mor´ e, On the local and superlinear
convergence of quasiNewton methods, J. Inst. Maths. Applic., 12 (1973), pp. 223–
246.
[29] W. Burmeister, Zur Konvergenz einiger verfahren der konjugierten Richtun
gen, in Proceedings of Internationaler Kongreß ¨ uber Anwendung der Mathematik in
dem Ingenieurwissenschaften, Weimar, 1975.
[30] I. W. Busbridge, The Mathematics of Radiative Transfer, Cambridge Tracts,
No. 50, Cambridge Univ. Press, Cambridge, 1960.
[31] R. H. Byrd, J. Nocedal, and R. B. Schnabel, Representation of quasi
Newton matrices and their use in limited memory methods, Math. Programming, 63
(1994), pp. 129–156.
[32] X.C. Cai, W. D. Gropp, D. E. Keyes, and M. D. Tidriri, NewtonKrylov
Schwarz methods in CFD, in Proceedings of the International Workshop on the
NavierStokes Equations, R. Rannacher, ed., Notes in Numerical Fluid Mechanics,
Braunschweig, 1994, Vieweg Verlag.
[33] S. L. Campbell, I. C. F. Ipsen, C. T. Kelley, and C. D. Meyer, GMRES
and the minimal polynomial, Tech. Report CRSCTR9410, North Carolina State
University, Center for Research in Scientiﬁc Computation, July 1994. BIT, to appear.
[34] S. L. Campbell, I. C. F. Ipsen, C. T. Kelley, C. D. Meyer, and Z. Q.
Xue, Convergence estimates for solution of integral equations with GMRES, Tech.
Report CRSCTR9513, North Carolina State University, Center for Research in
Scientiﬁc Computation, March 1995. Journal of Integral Equations and Applications,
to appear.
[35] R. Cavanaugh, Diﬀerence Equations and Iterative Processes, PhD thesis,
University of Maryland, 1970.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BIBLIOGRAPHY 155
[36] F. ChaitinChatelin, Is nonnormality a serious diﬃculty ?, Tech. Report
TR/PA/94/18, CERFACS, December 1994.
[37] T. Chan, E. Gallopoulos, V. Simoncini, T. Szeto, and C. Tong, A quasi
minimal residual variant of the BiCGSTAB algorithm for nonsymmetric systems,
SIAM J. Sci. Comput., 15 (1994), p. 338.
[38] T. Chan, R. Glowinski, J. P´ eriaux, and O. Widlund, eds., Domain
Decomposition Methods,Proceedings of the Second International Symposium on
Domain Decomposition Methods, Los Angeles, CA, January 14–16, 1988; Society
for Industrial and Applied Mathematics, Philadelphia, PA, 1989.
[39] , eds., Domain Decomposition Methods, Proceedings of the Third Interna
tional Symposium on Domain Decomposition Methods, Houston, TX, 1989; Society
for Industrial and Applied Mathematics, Philadelphia, PA, 1990.
[40] , eds., Domain Decomposition Methods, Proceedings of the Fourth Interna
tional Symposium on Domain Decomposition Methods, Moscow, USSR, 1990; Soci
ety for Industrial and Applied Mathematics, Philadelphia, PA, 1991.,
[41] S. Chandrasekhar, Radiative Transfer, Dover, New York, 1960.
[42] T. F. Coleman and J. J. Mor´ e, Estimation of sparse Jacobian matrices and
graph coloring problems, SIAM J. Numer. Anal., 20 (1983), pp. 187–209.
[43] T. F. Coleman and C. VanLoan, Handbook for Matrix Computations,
Frontiers in Appl. Math., No. 4, Society for Industrial and Applied Mathematics,
Philadelphia, PA, 1988.
[44] P. Concus, G. H. Golub, and G. Meurant, Block preconditioning for the
conjugate gradient method, SIAM J. Sci. Statist. Comput., 6 (1985), pp. 220–252.
[45] P. Concus, G. H. Golub, and D. P. O’Leary, A generalized conjugate
gradient method for the numerical solution of elliptic partial diﬀerential equations,
in Sparse Matrix Computations, J. R. Bunch and D. J. Rose, eds., Academic Press,
1976, pp. 309–332.
[46] E. J. Craig, The Nstep iteration procedures, J. Math. Phys., 34 (1955), pp. 64–
73.
[47] A. R. Curtis, M. J. D. Powell, and J. K. Reid, On the estimation of sparse
Jacobian matrices, J. Inst. Math. Appl., 13 (1974), pp. 117–119.
[48] J. W. Daniel, The conjugate gradient method for linear and nonlinear operator
equations, SIAM J. Numer. Anal., 4 (1967), pp. 10–26.
[49] D. W. Decker, H. B. Keller, and C. T. Kelley, Convergence rates for
Newton’s method at singular points, SIAM J. Numer. Anal., 20 (1983), pp. 296–314.
[50] D. W. Decker and C. T. Kelley, Newton’s method at singular points I, SIAM
J. Numer. Anal., 17 (1980), pp. 66–70.
[51] , Convergence acceleration for Newton’s method at singular points, SIAM J.
Numer. Anal., 19 (1982), pp. 219–229.
[52] , Sublinear convergence of the chord method at singular points, Numer.
Math., 42 (1983), pp. 147–154.
[53] , Broyden’s method for a class of problems having singular Jacobian at the
root, SIAM J. Numer. Anal., 22 (1985), pp. 566–574.
[54] T. J. Dekker, Finding a zero by means of successive linear interpolation, in
Constructive Aspects of the Fundamental Theorem of Algebra, P. Henrici, ed., 1969,
pp. 37–48.
[55] R. Dembo, S. Eisenstat, and T. Steihaug, Inexact Newton methods, SIAM
J. Numer. Anal., 19 (1982), pp. 400–408.
[56] R. Dembo and T. Steihaug, Truncated Newton algorithms for largescale
optimization, Math. Programming, 26 (1983), pp. 190–212.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
156 BIBLIOGRAPHY
[57] J. E. Dennis, On the Kantorovich hypothesis for Newton’s method, SIAM J.
Numer. Anal., 6 (1969), pp. 493–507.
[58] , Toward a uniﬁed convergence theory for Newtonlike methods, in Nonlinear
Functional Analysis and Applications, L. B. Rall, ed., Academic Press, New York,
1971, pp. 425–472.
[59] J. E. Dennis, J. M. Martinez, and X. Zhang, Triangular decomposition
methods for solving reducible nonlinear systems of equations, SIAM J. Optim., 4
(1994), pp. 358–382.
[60] J. E. Dennis and J. J. Mor´ e, A characterization of superlinear convergence
and its application to quasiNewton methods, Math. Comput., 28 (1974), pp. 549–560.
[61] , QuasiNewton methods, methods, motivation and theory, SIAM Rev., 19
(1977), pp. 46–89.
[62] J. E. Dennis and R. B. Schnabel, Least change secant updates for quasi
Newton methods, SIAM Rev., 21 (1979), pp. 443–459.
[63] , Numerical Methods for Unconstrained Optimization and Nonlinear Equa
tions, no. 16 in Classics in Applied Mathematics, SIAM, Philadelphia, 1996.
[64] J. E. Dennis and H. F. Walker, Convergence theorems for least change secant
update methods, SIAM J. Numer. Anal., 18 (1981), pp. 949–987.
[65] , Inaccuracy in quasiNewton methods: Local improvement theorems, in
Mathematical Programming Study 22: Mathematical programming at Oberwolfach
II, North–Holland, Amsterdam, 1984, pp. 70–85.
[66] , Leastchange sparse secant updates with inaccurate secant conditions,
SIAM J. Numer. Anal., 22 (1985), pp. 760–778.
[67] P. Deuflhard, R. W. Freund, and A. Walter, Fast secant methods for
the iterative solution of large nonsymmetric linear systems, Impact of Computing in
Science and Engineering, 2 (1990), pp. 244–276.
[68] W. J. Duncan, Some devices for the solution of large sets of simultaneous
linear equations (with an appendix on the reciprocation of partitioned matrices), The
London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science,
Seventh Series, 35 (1944), pp. 660–670.
[69] S. C. Eisenstat and H. F. Walker, Choosing the forcing terms in an inexact
Newton method, SIAM J. Sci. Comput., 17 (1996), pp. 16–32.
[70] , Globally convergent inexact Newton methods, SIAM J. Optim., 4 (1994),
pp. 393–422.
[71] H. C. Elman, Iterative Methods for Large, Sparse, Nonsymmetric Systems of
Linear Equations, PhD thesis, Yale University, New Haven, CT, 1982.
[72] H. C. Elman, Y. Saad, and P. E. Saylor, A hybrid ChebyshevKrylov
subspace algorithm for solving nonsymmetric systems of linear equations, SIAM J.
Sci. Statist. Comput., 7 (1986), pp. 840–855.
[73] M. Engelman, G. Strang, and K. J. Bathe, The application of quasi
Newton methods in ﬂuid mechanics, Internat. J. Numer. Methods Engrg., 17 (1981),
pp. 707–718.
[74] V. Faber and T. A. Manteuffel, Necessary and suﬃcient conditions for the
existence of a conjugate gradient method, SIAM J. Numer. Anal., 21 (1984), pp. 352–
362.
[75] D. Feng, P. D. Frank, and R. B. Schnabel, Local convergence analysis of
tensor methods for nonlinear equations, Math. Programming, 62 (1993), pp. 427–459.
[76] R. Fletcher, Conjugate gradient methods for indeﬁnite systems, in Numerical
Analysis Dundee 1975, G. Watson, ed., SpringerVerlag, Berlin, New York, 1976,
pp. 73–89.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BIBLIOGRAPHY 157
[77] R. W. Freund, A transposefree quasiminimal residual algorithm for non
Hermitian linear systems, SIAM J. Sci. Comput., 14 (1993), pp. 470–482.
[78] R. W. Freund, G. H. Golub, and N. M. Nachtigal, Iterative solution of
linear systems, Acta Numerica, 1 (1991), pp. 57–100.
[79] R. W. Freund, M. H. Gutknecht, and N. M. Nachtigal, An implemen
tation of the lookahead Lanczos algorithm for nonHermitian matrices, SIAM J. Sci.
Comput., 14 (1993), pp. 137–158.
[80] R. W. Freund and N. M. Nachtigal, QMR: a quasiminimal residual
algorithm for nonhermitian linear systems, Numer. Math., 60 (1991), pp. 315–339.
[81] , An implementation of the QMR method based on coupled twoterm
recurrences, SIAM J. Sci. Comput., 15 (1994), pp. 313–337.
[82] D. M. Gay, Some convergence properties of Broyden’s method, SIAM J. Numer.
Anal., 16 (1979), pp. 623–630.
[83] C. W. Gear, Numerical Initial Value Problems in Ordinary Diﬀerential Equa
tions, PrenticeHall, Englewood Cliﬀs, NJ, 1971.
[84] R. R. Gerber and F. T. Luk, A generalized Broyden’s method for solving
simultaneous linear equations, SIAM J. Numer. Anal., 18 (1981), pp. 882–890.
[85] P. E. Gill and W. Murray, QuasiNewton methods for unconstrained
optimization, J. I. M. A., 9 (1972), pp. 91–108.
[86] , Safeguarded steplength algorithms for optimization using descent methods,
Tech. Report NAC 37, National Physical Laboratory Report, Teddington, England,
1974.
[87] P. E. Gill, W. Murray, and M. H. Wright, Practical Optimization,
Academic Press, London, 1981.
[88] A. A. Goldstein, Constructive Real Analysis, Harper and Row, New York,
1967.
[89] G. H. Golub and C. G. VanLoan, Matrix Computations, Johns Hopkins
University Press, Baltimore, 1983.
[90] A. Griewank, Analysis and modiﬁcation of Newton’s method at singularities,
PhD thesis, Australian National University, 1981.
[91] , Rates of convergence for secant methods on nonlinear problems in Hilbert
space, in Numerical Analysis, Proceedings Guanajuato , Mexico 1984, Lecture Notes
in Math., No, 1230, J. P. Hennart, ed., SpringerVerlag, Heidelberg, 1986, pp. 138–
157.
[92] , The solution of boundary value problems by Broyden based secant methods,
in Computational Techniques and Applications: CTAC 85, Proceedings of CTAC,
Melbourne, August 1985, J. Noye and R. May, eds., North Holland, Amsterdam,
1986, pp. 309–321.
[93] , On the iterative solution of diﬀerential and integral equations using secant
updating techniques, in The State of the Art in Numerical Analysis, A. Iserles and
M. J. D. Powell, eds., Clarendon Press, Oxford, 1987, pp. 299–324.
[94] A. Griewank and G. F. Corliss, eds., Automatic Diﬀerentiation of Algo
rithms: Theory, Implementation, and Application, Society for Industrial and Applied
Mathematics, Philadelphia, PA, 1991.
[95] A. Griewank and P. L. Toint, Local convergence analysis for partitioned
quasinewton updates, Numer. Math., 39 (1982), pp. 429–448.
[96] , Partitioned variable metric methods for large sparse optimization problems,
Numer. Math., 39 (1982), pp. 119–137.
[97] W. A. Gruver and E. Sachs, Algorithmic Methods In Optimal Control,
Pitman, London, 1980.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
158 BIBLIOGRAPHY
[98] M. H. Gutknecht, Variants of BICGSTAB for matrices with complex spectrum,
SIAM J. Sci. Comput., 14 (1993), pp. 1020–1033.
[99] W. Hackbusch, MultiGrid Methods and Applications, vol. 4 of Springer Series
in Computational Mathematics, SpringerVerlag, New York, 1985.
[100] , Multigrid methods of the second kind, in Multigrid Methods for Integral
and Diﬀerential Equations, Oxford University Press, Oxford, 1985.
[101] W. E. Hart and S. O. W. Soul, QuasiNewton methods for discretized
nonlinear boundary problems, Journal of the Institute of Applied Mathematics, 11
(1973), pp. 351–359.
[102] H. V. Henderson and S. R. Searle, On deriving the inverse of a sum of
matrices, SIAM Rev., 23 (1981), pp. 53–60.
[103] M. R. Hestenes and E. Steifel, Methods of conjugate gradient for solving
linear systems, J. of Res. Nat. Bureau Standards, 49 (1952), pp. 409–436.
[104] D. M. Hwang and C. T. Kelley, Convergence of Broyden’s method in Banach
spaces, SIAM J. Optim., 2 (1992), pp. 505–532.
[105] E. Isaacson and H. B. Keller, Analysis of numerical methods, John Wiley,
New York, 1966.
[106] L. Kantorovich and G. Akilov, Functional Analysis, 2nd ed., Pergamon
Press, New York, 1982.
[107] L. V. Kantorovich, Functional analysis and applied mathematics, Uspehi Mat.
Nauk., 3 (1948), pp. 89–185. translation by C. Benster as Nat. Bur. Standards Report
1509. Washington, D. C., 1952.
[108] H. B. Keller, Newton’s method under mild diﬀerentiability conditions, J.
Comput. Sys. Sci, 4 (1970), pp. 15–28.
[109] , Lectures on Numerical Methods in Bifurcation Theory, Tata Institute of
Fundamental Research, Lectures on Mathematics and Physics, SpringerVerlag, New
York, 1987.
[110] C. T. Kelley, Solution of the Chandrasekhar Hequation by Newton’s method,
J. Math. Phys., 21 (1980), pp. 1625–1628.
[111] , A fast multilevel algorithm for integral equations, SIAM J. Numer. Anal.,
32 (1995), pp. 501–513.
[112] C. T. Kelley and E. W. Sachs, A quasiNewton method for elliptic boundary
value problems, SIAM J. Numer. Anal., 24 (1987), pp. 516–531.
[113] , A pointwise quasiNewton method for unconstrained optimal control
problems, Numer. Math., 55 (1989), pp. 159–176.
[114] , Fast algorithms for compact ﬁxed point problems with inexact function
evaluations, SIAM J. Sci. Statist. Comput., 12 (1991), pp. 725–742.
[115] , A new proof of superlinear convergence for Broyden’s method in Hilbert
space, SIAM J. Optim., 1 (1991), pp. 146–150.
[116] , Pointwise Broyden methods, SIAM J. Optim., 3 (1993), pp. 423–441.
[117] , Multilevel algorithms for constrained compact ﬁxed point problems, SIAM
J. Sci. Comput., 15 (1994), pp. 645–667.
[118] C. T. Kelley and R. Suresh, A new acceleration method for Newton’s method
at singular points, SIAM J. Numer. Anal., 20 (1983), pp. 1001–1009.
[119] C. T. Kelley and Z. Q. Xue, Inexact Newton methods for singular problems,
Optimization Methods and Software, 2 (1993), pp. 249–267.
[120] , GMRES and integral operators, SIAM J. Sci. Comput., 17 (1996), pp. 217–
226.
[121] T. Kerkhoven and Y. Saad, On acceleration methods for coupled nonlinear
elliptic systems, Numerische Mathematik, 60 (1992), pp. 525–548.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BIBLIOGRAPHY 159
[122] C. Lanczos, Solution of linear equations by minimized iterations, J. Res. Natl.
Bur. Stand., 49 (1952), pp. 33–53.
[123] T. A. Manteuffel, Adaptive procedure for estimating parameters for the
nonsymmetric Tchebychev iteration, Numer. Math., 31 (1978), pp. 183–208.
[124] T. A. Manteuffel and S. Parter, Preconditioning and boundary conditions,
SIAM J. Numer. Anal., 27 (1990), pp. 656–694.
[125] E. S. Marwil, Convergence results for Schubert’s method for solving sparse
nonlinear equations, SIAM J. Numer. Anal., (1979), pp. 588–604.
[126] S. McCormick, Multilevel Adaptive Methods for Partial Diﬀerential Equations,
Society for Industrial and Applied Mathematics, Philadelphia, PA, 1989.
[127] J. A. Meijerink and H. A. van der Vorst, An iterative solution method
for linear systems of which the coeﬃcient matrix is a symmetric Mmatrix, Math.
Comput., 31 (1977), pp. 148–162.
[128] C. D. Meyer, Matrix Analysis and Applied Linear Algebra, forthcoming.
[129] J. J. Mor´ e, Recent developments in algorithms and software for trust region
methods, in Mathematical Programming: The State of the Art, A. Bachem,
M. Gr¨ oschel, and B. Korte, eds., SpringerVerlag, Berlin, 1983, pp. 258–287.
[130] J. J. Mor´ e and J. A. Trangenstein, On the global convergence of Broyden’s
method, Math. Comput., 30 (1976), pp. 523–540.
[131] T. E. Mott, Newton’s method and multiple roots, Amer. Math. Monthly, 64
(1957), pp. 635–638.
[132] T. W. Mullikin, Some probability distributions for neutron transport in a half
space, J. Appl. Probab., 5 (1968), pp. 357–374.
[133] W. Murray and M. L. Overton, Steplength algorithms for minimizing a
class of nondiﬀerentiable functions, Computing, 23 (1979), pp. 309–331.
[134] N. M. Nachtigal, S. C. Reddy, and L. N. Trefethen, How fast are
nonsymmetric matrix iterations?, SIAM J. Matrix Anal. Appl., 13 (1992), pp. 778–
795.
[135] N. M. Nachtigal, L. Reichel, and L. N. Trefethen, A hybrid gmres
algorithm for nonsymmetric linear systems, SIAM J. Matrix Anal. Appl., 13 (1992).
[136] S. G. Nash, Preconditioning of truncated Newton methods, SIAM J. Sci. Statist.
Comput., 6 (1985), pp. 599–616.
[137] , Who was Raphson? (an answer). Electronic Posting to NADigest,
v92n23, 1992.
[138] J. L. Nazareth, Conjugate gradient methods less dependent on conjugacy,
SIAM Rev., 28 (1986), pp. 501–512.
[139] O. Nevanlinna, Convergence of Iterations for Linear Equations, Birkh¨ auser,
Basel, 1993.
[140] I. Newton, The Mathematical Papers of Isaac Newton (7 volumes), D. T.
Whiteside, ed., Cambridge University Press, Cambridge, 1967–1976.
[141] B. Noble, Applied Linear Algebra, Prentice Hall, Englewood Cliﬀs, NJ, 1969.
[142] J. Nocedal, Theory of algorithms for unconstrained optimization, Acta Numer
ica, 1 (1991), pp. 199–242.
[143] D. P. O’Leary, Why Broyden’s nonsymmetric method terminates on linear
equations, SIAM J. Optim., 4 (1995), pp. 231–235.
[144] J. M. Ortega, Numerical Analysis A Second Course, Classics in Appl. Math.,
No. 3, Society for Industrial and Applied Mathematics, Philadelphia, PA, 1990.
[145] J. M. Ortega and W. C. Rheinboldt, Iterative Solution of Nonlinear
Equations in Several Variables, Academic Press, New York, 1970.
[146] A. M. Ostrowski, Solution of Equations and Systems of Equations, Academic
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
160 BIBLIOGRAPHY
Press, New York, 1960.
[147] B. N. Parlett, The Symmetric Eigenvalue Problem, Prentice Hall, Englewood
Cliﬀs, NJ, 1980.
[148] B. N. Parlett, D. R. Taylor, and Z. A. Liu, A lookahead Lanczos
algorithm for unsymmetric matrices, Math. Comput., 44 (1985), pp. 105–124.
[149] D. W. Peaceman and H. H. Rachford, The numerical solution of parabolic
and elliptic diﬀerential equations, J. Soc. Indus. Appl. Math., 11 (1955), pp. 28–41.
[150] G. Peters and J. Wilkinson, Inverse iteration, illconditioned equations and
Newton’s method, SIAM Rev., 29 (1979), pp. 339–360.
[151] L. R. Petzold, A description of DASSL: a diﬀerential/algebraic system solver,
in Scientiﬁc Computing, R. S. Stepleman et al., eds., North Holland, Amsterdam,
1983, pp. 65–68.
[152] E. Picard, M´emoire sur la th´eorie des ´equations aux d´eriv´ees partielles et la
m´ethode des approximations successives, J. de Math. ser 4, 6 (1890), pp. 145–210.
[153] M. J. D. Powell, A hybrid method for nonlinear equations, in Numerical
Methods for Nonlinear Algebraic Equations, Gordon and Breach, New York, 1970,
pp. 87–114.
[154] , Convergence properties of a class of minimization algorithms, in Nonlinear
Programming 2, O. L. Mangasarian, R. R. Meyer, and S. M. Robinson, eds.,
Academic Press, New York, 1975, pp. 1–27.
[155] L. B. Rall, Convergence of the Newton process to multiple solutions, Numer.
Math., 9 (1961), pp. 23–37.
[156] J. Raphson, Analysis aequationum universalis seu ad aequationes algebraicas
resolvendas methodus generalis, et expedita, ex nova inﬁnitarum serierum doctrina,
deducta ac demonstrata. Original in British Library, London, 1690.
[157] G. W. Reddien, On Newton’s method for singular problems, SIAM J. Numer.
Anal., 15 (1978), pp. 993–986.
[158] J. K. Reid, Least squares solution of sparse systems of nonlinear equations by a
modiﬁed Marquardt algorithm, in Proc. NATO Conf. at Cambridge, July 1972, North
Holand, Amsterdam, 1973, pp. 437–445.
[159] W. C. Rheinboldt, Numerical Analysis of Parametrized Nonlinear Equations,
John Wiley, New York, 1986.
[160] J. R. Rice, Experiments on GramSchmidt orthogonalization, Math. Comput.,
20 (1966), pp. 325–328.
[161] T. J. Rivlin, The Chebyshev Polynomials, John Wiley, New York, 1974.
[162] H. L. Royden, Real Analysis, 2nd ed., Macmillan, New York, 1968.
[163] Y. Saad, Practical use of polynomial preconditionings for the conjugate gradient
method, SIAM J. Sci. Statist. Comput., 6 (1985), pp. 865–881.
[164] , Least squares polynomials in the complex plane and their use for solving
nonsymmetric linear systems, SIAM J. Numer. Anal., 24 (1987), pp. 155–169.
[165] , ILUT: A dual threshold incomplete LU factorization, Tech. Report 9238,
Computer Science Department, University of Minnesota, 1992.
[166] , A ﬂexible innerouter preconditioned GMRES algorithm, SIAM J. Sci.
Comput., (1993), pp. 461–469.
[167] Y. Saad and M. Schultz, GMRES a generalized minimal residual algorithm
for solving nonsymmetric linear systems, SIAM J. Sci. Statist. Comput., 7 (1986),
pp. 856–869.
[168] E. Sachs, Broyden’s method in Hilbert space, Math. Programming, 35 (1986),
pp. 71–82.
[169] P. E. Saylor and D. C. Smolarski, Implementation of an adaptive algorithm
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
BIBLIOGRAPHY 161
for Richardson’s method, Linear Algebrra Appl., 154/156 (1991), pp. 615–646.
[170] R. B. Schnabel and P. D. Frank, Tensor methods for nonlinear equations,
SIAM J. Numer. Anal., 21 (1984), pp. 815–843.
[171] E. Schr¨ oder,
¨
Uber unendlich viele Algorithmen zur Auﬂosung der Gleichungen,
Math. Ann., 2 (1870), pp. 317–365.
[172] L. K. Schubert, Modiﬁcation of a quasiNewton method for nonlinear equations
with sparse Jacobian, Math. Comput., 24 (1970), pp. 27–30.
[173] G. A. Schultz, R. B. Schnabel, and R. H. Byrd, A family of trustregion
based algorithms for unconstrained minimization with strong global convergence
properties, SIAM J. Numer. Anal., 22 (1985), pp. 47–67.
[174] V. E. Shamanskii, A modiﬁcation of Newton’s method, Ukran. Mat. Zh., 19
(1967), pp. 133–138. (In Russian.)
[175] A. H. Sherman, On Newtoniterative methods for the solution of systems of
nonlinear equations, SIAM J. Numer. Anal., 14 (1978), pp. 755–774.
[176] J. Sherman and W. J. Morrison, Adjustment of an inverse matrix corre
sponding to changes in the elements of a given column or a given row of the original
matrix (abstract), Ann. Math. Statist., 20 (1949), p. 621.
[177] , Adjustment of an inverse matrix corresponding to a change in one element
of a given matrix, Ann. Math. Statist., 21 (1950), pp. 124–127.
[178] K. Sigmon, MATLAB Primer, Fourth Edition, CRC Press, Boca Raton, FL,
1994.
[179] D. C. Smolarski and P. E. Saylor, An optimum iterative method for solving
any linear system with a square matrix, BIT, 28 (1988), pp. 163–178.
[180] P. Sonneveld, CGS, a fast Lanczostype solver for nonsymmetric linear
systems, SIAM J. Sci. Statist. Comput., 10 (1989), pp. 36–52.
[181] D. C. Sorensen, Newton’s method with a model trust region modiﬁcation, SIAM
J. Numer. Anal., 19 (1982), pp. 409–426.
[182] G. Starke, Alternating direction preconditioning for nonsymmetric systems of
linear equations, SIAM J. Sci. Comput., 15 (1994), pp. 369–385.
[183] G. Starke and R. S. Varga, A hybrid ArnoldiFaber iterative method for
nonsymmetric systems of linear equations, Numer. Math., 64 (1993), pp. 213–239.
[184] G. W. Stewart, Introduction to matrix computations, Academic Press, New
York, 1973.
[185] E. L. Stiefel, Kernel polynomials in linear algebra and their numerical
applications, U.S. National Bureau of Standards, Applied Mathematics Series, 49
(1958), pp. 1–22.
[186] P. N. Swarztrauber, The methods of cyclic reduction, Fourier analysis and
the FACR algorithm for the discrete solution of Poisson’s equation on a rectangle,
SIAM Rev., 19 (1977), pp. 490–501.
[187] , Approximate cyclic reduction for solving Poisson’s equation, SIAM J. Sci.
Statist. Comput., 8 (1987), pp. 199–209.
[188] P. N. Swarztrauber and R. A. Sweet, Algorithm 541: Eﬃcient FORTRAN
subprograms for the solution of elliptic partial diﬀerential equations, ACM Trans.
Math. Software, 5 (1979), pp. 352–364.
[189] P. L. Toint, On sparse and symmetric matrix updating subject to a linear
equation, Math. Comput., 31 (1977), pp. 954–961.
[190] J. F. Traub, Iterative Methods for the Solution of Equations, Prentice Hall,
Englewood Cliﬀs, NJ, 1964.
[191] H. A. van der Vorst, BiCGSTAB: A fast and smoothly converging variant to
BiCG for the solution of nonsymmetric systems, SIAM J. Sci. Statist. Comput., 13
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
162 BIBLIOGRAPHY
(1992), pp. 631–644.
[192] H. A. van der Vorst and C. Vuik, The superlinear convergence behaviour
of GMRES, Journal Comput. Appl. Math., 48 (1993), pp. 327–341.
[193] R. S. Varga, Matrix Iterative Analysis, Prentice Hall, Englewood Cliﬀs, NJ,
1962.
[194] E. L. Wachspress, Iterative Solution of Elliptic Systems and Applications to
the Neutron Diﬀusion Equations of Reactor Physics, Prentice Hall, Englewood Cliﬀs,
NJ, 1966.
[195] H. F. Walker, Implementation of the GMRES method using Householder
transformations, SIAM J. Sci. Statist. Comput., 9 (1989), pp. 815–825.
[196] , Residual smoothing and peak/plateau behavior in Krylov subspace methods,
Applied Numer. Math., 19 (1995), pp. 279–286.
[197] H. F. Walker and L. Zhou, A simpler GMRES, J. Numer. Lin. Alg. Appl.,
6 (1994), pp. 571–581.
[198] L. T. Watson, S. C. Billups, and A. P. Morgan, Algorithm 652:
HOMPACK: A suite of codes for globally convergent homotopy algorithms, ACM
Trans. Math. Software, 13 (1987), pp. 281–310.
[199] M. A. Woodbury, Inverting modiﬁed matrices, Memorandum Report 42,
Statistical Research Group, Princeton NJ, 1950.
[200] D. M. Young, Iterative Solution of Large Linear Systems, Academic Press, New
York, 1971.
[201] T. J. Ypma, The eﬀect of rounding errors on Newtonlike methods, IMA J.
Numer. Anal., 3 (1983), pp. 109–118.
[202] L. Zhou and H. F. Walker, Residual smoothing techniques for iterative
methods, SIAM J. Sci. Comput., 15 (1994), pp. 297–312.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Index
Aconjugacy, 20
Anorm, 12
Approximate inverse, 6, 77
Armijo rule, 137
Arnoldi process, 37
Backtracking, 136
Bad Broyden method, 132
Banach Lemma, 6
BiCG, 47
algorithm, 47
BiCGSTAB, 50
algorithm, 50
ﬁnite diﬀerence, 110
Biconjugacy, 47
BiConjugate Gradient, 47
algorithm, 47
Biorthogonality, 47
Bounded deterioration, 118, 121
Breakdown, 47
Broyden sequence, 120
Broyden’s method, 113
convergence
linear equations, 118
nonlinear equations, 120
implementation, 123
limited memory, 123
restarting, 123
sparse, 133
Brsol
algorithm, 126
Brsola
algorithm, 145
Cauchy sequence, 67
CGNE, 25
CGNR, 25
CGS, 48
algorithm, 49
Chandrasekhar Hequation, 87
solution
Broyden’s method, 128
Newton’s method, 87
NewtonGMRES, 107
Characteristic polynomial, 34
Chord method, 74
algorithm, 74
convergence, 76
Condition number, 3
Conjugate Gradient
algorithm, 22
ﬁnite termination, 14
minimization property, 12
use of residual polynomials, 14
Conjugate Gradient Squared, 48
algorithm, 49
Conjugate transpose, 34
Contraction mapping, 67
Contraction Mapping Theorem, 67
ConvectionDiﬀusion Equation
Broyden’s method, 130
Newton–Armijo, 146
NewtonGMRES, 108
solution
Broyden’s method, 130
Convergence theorem
Broyden’s method
linear equations, 119
163
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
164 INDEX
nonlinear equations, 123
chord method, 76, 77
contraction mapping, 67
ﬁnite diﬀerence Newton, 81
inexact Newton’s method, 96,
99
Newton’s method, 71
Newton–Armijo, 141
NewtonGMRES, 103
secant method, 82
Shamanskii method, 79
DASPK, 110
Dennis–Mor´e Condition, 114
superlinear convergence, 115
Diagonalizable matrix, 34
Diagonalizing transformation, 34
Diﬀerence approximation, 79
choice of h, 80
directional derivative, 80
Jacobian, 80
nonlinearity, 80
Elementary Jordan block, 60
Fixed point, 66
Fixedpoint iteration, 66
Forcing term, 95
choice of, 104
Gauss–Seidel
algorithm, 9
Givens rotation, 43
Globally convergent algorithm, 135
GMRES
algorithm, 40, 45
diagonalizable matrices, 34
ﬁnite diﬀerences, 101
ﬁnite termination, 34
GMRES(m), 39, 46
algorithm, 46
loss of orthogonality, 41
minimization property, 33
restarts, 39, 46
use of residual polynomials, 33
Gram–Schmidt process, 37
loss of orthogonality, 40, 41
modiﬁed, 40
Hequation, 87
solution
Broyden’s method, 128
Newton’s method, 87
NewtonGMRES, 107
H¨older Continuity, 91
High order methods, 78
Hybrid algorithms, 36
Induced matrix norm
deﬁnition, 3
Inexact Newton method, 95
Inner iteration, 100
Inverse power method, 94
Inverse secant equation, 132
Inverse Tangent
Newton–Armijo, 146
Iteration matrix, 5
Iteration statistics, 91
Jacobi
algorithm, 8
convergence result, 8
preconditioning, 9
Jacobi preconditioning, 24
Jordan block, 60
Kantorovich Theorem, 83
Krylov subspace, 11
Left preconditioning, 36
Limited memory, 123
Line Search
parabolic
threepoint, 143
twopoint, 142
polynomial, 142
Line search, 135, 136
Linear PDE
BiCGSTAB, 57
Broyden’s method, 127
CGNR, 55
GMRES, 54
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
INDEX 165
solution
Broyden’s method, 127
TFQMR, 58
Lipschitz continuous
deﬁnition, 67
Local improvement, 75
inexact, 100
Matrix norm, 3
Matrix splitting, 7
Modiﬁed Bratu Problem, 132
Newton direction, 136
Newton’s method, 71
algorithm, 74
convergence rate, 71
ﬁnite diﬀerences, 81
stagnation, 81
inexact, 95
monotone convergence, 92
termination, 72
NewtonGMRES, 101
convergence rates, 103
ﬁnite diﬀerences, 101
safeguarding, 105
Newtoniterative method, 100
Normal matrix, 34
Nsol
algorithm, 86
Nsola
algorithm, 138
Nsolgm
algorithm, 105
Outer iteration, 100
Over solving, 104
Picard iteration, 66
Poisson solver, 24, 27
preconditioner
for CG, 24, 27
for GMRES, 36, 54
Polynomial preconditioning, 24, 36
Richardson iteration, 7
Positive deﬁnite matrix, 11
Preconditioned CG
algorithm, 23
Preconditioning
conjugate gradient, 19
GMRES, 36
incomplete
factorization, 24, 36
Jacobi, 24
Poisson solver, 27, 36, 54
polynomial, 24, 36
Richardson iteration, 6
Quasiminimization, 52
QuasiNewton methods, 113
Quasiresidual, 51
Quasiresidual norm, 52
Relaxation parameter, 9
Reorthogonalization, 41
Residual
linear, 11
as error estimate, 4
nonlinear
as error estimate, 72
relative nonlinear, 72
Residual polynomials
application to CG, 14
application to CGNR, 25
application to GMRES, 33
application to TFQMR, 52
deﬁnition, 14
Richardson iteration, 5
nonlinear, 66
Right preconditioning, 36
Schubert algorithm, 133
Search direction, 136
Secant equation, 113
Secant method, 82
convergence, 82
qorder, 91
rorder, 91
Secant update, 113
Shamanskii method, 78
algorithm, 78
convergence, 79
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
166 INDEX
ﬁnite diﬀerences, 91
Sherman–Morrison
formula, 124, 132
Sherman–Morrison–Woodbury for
mula, 132
Simple decrease, 136
SOR
algorithm, 9
SPD matrix, 12
Spectral radius, 7
Stagnation, 94
Newton’s method, 75
Stationary iterative methods, 5
as preconditioners, 24
Successive substitution, 66
Suﬃcient decrease, 137
Symmetric Gauss–Seidel
algorithm, 9
preconditioning, 9
Symmetric matrix, 11
Symmetric positive deﬁnite matrix,
12
Termination
CG, 16
GMRES, 35, 38
linear equations, 4
nonlinear equations, 72
TFQMR, 52
TFQMR, 51
algorithm, 53
ﬁnite diﬀerence, 110
weights, 52
Weighted norm, 98
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
To Polly H. Thomas, 19061994, devoted mother and grandmother
1
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
2
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
. . . . . . . . . 10 CHAPTER 2. . . . . . .1 BiCG. . . . . .5 Implementation: Givens rotations . 2. . . . . . . . . . . . 5 1. . . 2. GMRES Iteration 3.3 BiCGSTAB. . . . . . . 2. . . . . . . . . . . . .3 Preconditioning . . . . . . . . . .5 Preconditioning . .4 GMRES implementation: Basic ideas . .6. . . 7 1. . . . . . . . . . . . . . . . . . .2 Consequences of the minimization property . . . . . . . . Conjugate Gradient Iteration 2. . . . . . . . . . . . . 7 1. . . . . . . . . . . . 3. . . . . . . .3 Termination of the iteration . . This electronic version is for personal use and may not be duplicated or distributed. . . . . . . . . . . . . 3. . .1 Krylov methods and the minimization property .4 Matrix splittings and classical stationary iterative methods . . . . . . . 2. . . . . . . .7 Examples for preconditioned conjugate iteration 2. . . . . .ecsecurehost. . . . . . .6.2 Termination . . . . . . . . . . . . . . . . . . . . . .html. . . 3 1. . . Basic Concepts and Stationary Iterative Methods 3 1.3 The spectral radius . . . . . . . . . . . . . . .1 Review and notation . . .6. . . .1 The minimization property and its consequences 3. . . . . .8 Exercises on conjugate gradient . 3. .2 CGS. . . . . . . . . . . . . . .6 CGNR and CGNE . . . . . . . . . . . . . . . . . . . . . . . . .2 The Banach Lemma and approximate inverses . . . .5 Exercises on stationary iterative methods . . . . . . . . . . . . . . . . . .4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . vii Buy this book from SIAM at http://www. . . . 2. . . . . . . . . . Contents Preface xi How to Get the Software xiii CHAPTER 1. . . . . . . . . . . CHAPTER 3. . . . . . . . 3. . . . . . . . . . . 11 11 13 15 19 22 25 26 30 33 33 35 36 37 43 46 47 48 50 . . . . . . . . . .6 Other methods for nonsymmetric systems . . 3. . . . . . . . . 3. . . . . 3.com/SIAM/FR16. . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . .Copyright ©1995 by the Society for Industrial and Applied Mathematics. . . . . . . . . . . .
. 4. . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2.4. . . . . . . . . . . . . . 6. . .2 Weighted norm analysis. 6.html. .3 The Shamanskii method. . . . . . .4 Diﬀerence approximation to F . . . . . . . . 5.4 Errors in the function and derivative . . . Exercises on GMRES . . . . . CHAPTER 6.8 3. . . . . . . . . . . . .3 The standard assumptions . . . . 6. . . Inexact Newton Methods 6.2 Other Newtoniterative methods. .1. . . . .ecsecurehost. . . . . . . . . . . . . . . . . .1 Newton GMRES. . .3 NewtonGMRES implementation . . . . . .1 Local convergence of Newton’s method 5. . 7. . . . . . . . . . .Copyright ©1995 by the Society for Industrial and Applied Mathematics. . . . CHAPTER 5. . . 5. . . . 5. . . . . . . Broyden’s method 7. . . . . . . . . . . . . . 6. . . . . . . .9 . . . . . . . . . .1 Chandrasekhar Hequation. . . . . . . . . . . .2 Fixedpoint iteration . . . . . . . 5. . . . . . . . . . .1 The basic estimates . . . . . . 6. . . . . . . .2 Termination of the iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 3. . . . . . .4 Examples for NewtonGMRES . . . . . . . . . . . . . . . . . . . . . . . 5. . . .4. . . . . This electronic version is for personal use and may not be duplicated or distributed. . . . 5. .1 The chord method. . . . . . . . . . . . . . . . . . . . Buy this book from SIAM at http://www. . . . . . . .4. . . . . . . . . . . . . . . . . . . . and TFQMR iteration . . . . . . . . . . . . . . . . . .4 TFQMR. . . . .3 Implementation of Newton’s method . .2. . . 51 54 55 60 65 65 66 68 71 71 72 73 75 76 77 78 79 82 83 86 91 95 95 95 97 100 100 101 104 104 106 107 108 110 113 114 116 118 120 123 127 CHAPTER 4. . 7. . . . . . . . . 4. . . . 6.2 Newtoniterative methods . . .7 Exercises on Newton’s method . . . . . . . . . . .4 Examples for Broyden’s method . 5. . . . . . . . . . . . . . 5. . . . . . . . . . . .2. . . . . . . . . . . . . . . . .6 Examples for Newton’s method . . . . . .5 The secant method.com/SIAM/FR16. . . 5. . . . Basic Concepts and FixedPoint Iteration 4. . . . . . . . . . . 6. .1 Linear problems. . . . . . . . . . . . . . . . . . . . . . . . . .2 Convergence analysis . . . . . . . . . . . .3 Implementation of Broyden’s method 7. . . . . . 6. . . . . . . . . . . . . . . . . .1. . . . . . . . . .2. . viii 3. . .4. . . . . . . . .2 Nonlinear problems. . . . . . . . . . . . . . . . . . BiCGSTAB. . . . . . . . . . .4. . . .6. . . . . . . . . . . . . 6. . . .1 The Dennis–Mor´ condition . . .1. . . .3 Errors in the function. . . . .4. . . Newton’s Method 5. . . . . . . 5. . . . 6. . .1 Types of convergence . . CHAPTER 7. 7. . . . . . . . . . . . . . . . . . CONTENTS 3. e 7. . . . . .2 Approximate inversion of F . . . .2 Convectiondiﬀusion equation. . Examples for GMRES iteration . . Examples for CGNR.1 Direct analysis. . . . . . . . . . . . . . . . . . . .5 The Kantorovich Theorem . . . . . . . . . . . . . . . .5 Exercises on inexact Newton methods . . . 6. . . . . .
8. . . . . . . . . .4. . . . . . . . . . . . . . . . . . .3. .1 Inverse tangent function. .2 Broyden’s method. . . . . . . 135 135 138 141 142 144 146 146 146 148 151 153 163 CHAPTER 8. This electronic version is for personal use and may not be duplicated or distributed. . . . . . . Global Convergence 8. . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . .3. . . . . 132 . . . . . . . . .2 Convectiondiﬀusion equation. . .2 Nonlinear problems. . . . . . . . . . . . . .4 Examples for Newton–Armijo . . . 8. . . . . . .1 Linear problems. . . .Copyright ©1995 by the Society for Industrial and Applied Mathematics. . . . . . . . . 8. . . .5 7. . . . . . . . 127 7. . .4. . . . 8. . 8. . . . .3 Implementation of the Armijo rule . . . . . . .4. . . . . . .1 Polynomial line searches. . . . . 8. . . .ecsecurehost. . . . . . . 8. .4. . . . . . . . 8. . . .html. . . . . . CONTENTS ix 7. . . . . . . . . . . .1 Single equations . . Bibliography Index Buy this book from SIAM at http://www. . . . . . . . . . . . . 8. . . . . . .3 Broyden–Armijo. . . . 128 Exercises on Broyden’s method . . . . . . . . . . . . . . . .5 Exercises on global convergence . . . . .2 Analysis of the Armijo rule . . . . .com/SIAM/FR16. . . . . . . . . .
This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost.com/SIAM/FR16. .html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. x CONTENTS Buy this book from SIAM at http://www.
Inc. and the central ideas of direct methods for the numerical solution of dense linear systems as described in standard texts such as [7]. .0a on various SUN SPARCstations and version 4.com/SIAM/FR16. the matrixfree formulation and analysis for GMRES and conjugate gradient is almost unchanged in an inﬁnitedimensional setting. We assume that the reader is familiar with elementary numerical analysis. xi Buy this book from SIAM at http://www. Preface This book on iterative methods for linear and nonlinear equations can be used as a tutorial and a reference by anyone who needs to solve nonlinear systems of equations or large linear systems. We present a limited number of computational examples. for doing the exercises. The analysis of Broyden’s method presented in Chapter 7 and the implementations presented in Chapters 7 and 8 are diﬀerent from the classical ones and also extend directly to an inﬁnitedimensional setting. [105]. A good introduction to the latest 1 MATLAB is a registered trademark of The MathWorks.ecsecurehost.1 on an Apple Macintosh Powerbook 180) and the MATLAB environment is an excellent one for getting experience with the algorithms. and for smalltomedium scale production work. The computational examples in this book were done with MATLAB (version 4. For example. This electronic version is for personal use and may not be duplicated or distributed. These examples are intended to provide results that can be used to validate the reader’s own implementations and to give a sense of how the algorithms perform.html. Our approach is to focus on a small number of methods and treat them in depth. Though this book is written in a ﬁnitedimensional setting. or [184]. linear algebra. The examples are not designed to give a complete picture of performance or to be a suite of test problems. It may also be used as a textbook for introductory courses in nonlinear equations or iterative methods or as source material for an introductory course in numerical analysis at the graduate level. The computational examples and exercises focus on discretizations of inﬁnitedimensional problems such as integral and diﬀerential equations. we have selected for coverage mostly algorithms and methods of analysis which extend directly to the inﬁnitedimensional case and whose convergence can be thoroughly analyzed.Copyright ©1995 by the Society for Industrial and Applied Mathematics.1 MATLAB codes for many of the algorithms are available by anonymous ftp.
Belinda King. If the reader has no access to MATLAB or will be solving very large problems.html. most recently under National Science Foundation Grant Nos. Ilse Ipsen. Parts of this book are based upon work supported by the National Science Foundation and the Air Force Oﬃce of Scientiﬁc Research over several years. Lea Jenkins. Steve Wright. Carl Meyer. Gordon Wade. . Steve Campbell. Vickie Kearn.ecsecurehost. Many of my students and colleagues discussed various aspects of this project with me and provided important corrections. Zhaqing Xue.2) of MATLAB is the MATLAB Primer [178]. Moody Chu. Debbie Lockhart. Tony Choi. Joseph Skudlarek. Jeﬀ Scroggs. I thank ChungWei Ng and my parents for over one hundred and ten years of patience and support. Jim Epperson. North Carolina January. 1998 Buy this book from SIAM at http://www. C. [43] is also a useful resource. Homer Walker. ﬁndings. Yue Zhang. Mike Tocci. Laura Helfrich. T.com/SIAM/FR16. Ekkehard Sachs. Andreas Griewank. Any opinions. Most importantly. and conclusions or recommendations expressed in this material are those of the author and do not necessarily reﬂect the views of the National Science Foundation or of the Air Force Oﬃce of Scientiﬁc Research. Jeﬀ Butera. Howard Elman.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Kelley Raleigh. and an anonymous reviewer for their contributions and encouragement. suggestions. Casey Miller. ideas. xii PREFACE version (version 4. and pointers to the literature. the general algorithmic descriptions or even the MATLAB codes can easily be translated to another language. DMS9024622 and DMS9321938. I am especially indebted to Jim Banoczi. This electronic version is for personal use and may not be duplicated or distributed.
html.mathworks. 24 Prime Park Way Natick. from the MathWorks World Wide Web site. Phone: (508) 6531415 Fax: (508) 6532997 Email: info@mathworks.html One can obtain MATLAB from The MathWorks.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The MATLAB codes can be obtained by anonymous ftp from the MathWorks server ftp.com WWW: http://www.com or from SIAM’s World Wide Web site http://www. .mathworks.mathworks.com xiii Buy this book from SIAM at http://www.siam.org/books/kelley/kelley.ecsecurehost. http://www. How to get the software A collection of MATLAB codes has been written to accompany this book. This electronic version is for personal use and may not be duplicated or distributed.com in the directory pub/books/kelley. MA 01760.com/SIAM/FR16. Inc.
Recall that the condition number of A relative to the norm κ(A) = A A−1 .1.com/SIAM/FR16. If · · is is the lp norm x p = N j=1 1/p (x)i  p 3 Buy this book from SIAM at http://www. x =1 where A is a nonsingular N × N matrix.ecsecurehost. In this chapter · will denote a norm on RN as well as the induced matrix norm. Chapter 1 Basic Concepts and Stationary Iterative Methods 1.1) Ax = b. Throughout this chapter x will denote a potential solution and {xk }k≥0 the sequence of iterates. We will rarely need to refer to individual components of vectors.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The induced matrix norm of an N × N matrix A is deﬁned by A = max Ax . This electronic version is for personal use and may not be duplicated or distributed. and Induced norms have the important property that Ax ≤ A x . b ∈ RN is given. Review and notation We begin by setting notation and reviewing some ideas from numerical linear algebra that we expect the reader to be familiar with. Let · be a norm on RN .1.1. We will denote the ith component of a vector x by (x)i (note the parentheses) and the ith component of xk by (xk )i . where κ(A) is understood to be inﬁnite if A is singular. We will write linear equations as (1.html. . x∗ = A−1 b ∈ RN is to be found. Definition 1. An excellent reference for the basic ideas of numerical linear algebra and direct methods for linear equations is [184].
x. The termination criterion (1. .2) which can be related to the error e = x − x∗ in terms of the condition number.Copyright ©1995 by the Society for Industrial and Applied Mathematics.1. Most iterative methods terminate when the residual r = b − Ax is suﬃciently small. Let A be nonsingular and let x∗ = A−1 b. For this reason we prefer to terminate the iteration when rk (1. 4 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS we will write the condition number as κp . One termination criterion is (1. r0 as asserted. Since r = b − Ax = −Ae we have e = A−1 Ae ≤ A−1 and r0 = Ae0 ≤ A e0 .ecsecurehost.3) Proof. particularly when the linear iteration is being used as part of a nonlinear solver.html.2) depends on the initial iterate and may result in unnecessary work when the initial iterate is good and a poor result when the initial iterate is far from the solution. x0 ∈ RN . Let b. (1.com/SIAM/FR16.4) are the same when x0 = 0. e0 r0 rk < τ.2) and (1. Lemma 1. b The two conditions (1.4) < τ. e0 A −1 r0 r0 Ae = A−1 r e r ≤ κ(A) . Buy this book from SIAM at http://www.1. This electronic version is for personal use and may not be duplicated or distributed. which is a common choice. Hence e A−1 r r ≤ = κ(A) .
If M is an N × N matrix with M < 1 then I − M is nonsingular and 1 (I − M )−1 ≤ (1.6) xk+1 = (I − A)xk + b.8) . we must have M S + I = S and hence (I − M )S = I. .5) x = (I − A)x + b.html. say to S.2.8) holds by showing that the series ∞ M l = (I − M )−1 . This electronic version is for personal use and may not be duplicated or distributed.1.7) M is an N ×N matrix called the iteration matrix. The Banach Lemma and approximate inverses The most straightforward approach to an iterative solution of a linear system is to rewrite (1.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Iterative methods of this form are called stationary iterative methods because the transition from xk to xk+1 does not depend on the history of the iteration. In (1. Hence l Ml .ecsecurehost. k l=0 Ml To see this note that for all m > k m l=k+1 Sk − Sm ≤ Now. The Krylov methods discussed in Chapters 2 and 3 are not stationary iterative methods. One way to do this is to write Ax = b as (1.com/SIAM/FR16. Buy this book from SIAM at http://www. M l ≤ M norm. l=0 The partial sums Sk = form a Cauchy sequence in RN ×N . because m l=k+1 · is a matrix norm that is induced by a vector 1 − M m−k 1− M Sk − Sm ≤ M l = M k+1 →0 as m. and to deﬁne the Richardson iteration (1. We will show that I − M is nonsingular and that (1.1) as a linear ﬁxedpoint iteration. This proves that I − M is nonsingular and that S = (I − M )−1 .7) xk+1 = M xk + c. k → ∞. 1− M Proof. Since M Sk + I = Sk+1 . All our results are based on the following lemma. BASIC CONCEPTS 5 1.2. Lemma 1. We will discuss more general methods in which {xk } is given by (1. Hence the sequence Sk converges.
1.7) converges to x = (I − M )−1 c for all initial iterates x0 .9). By Lemma 1. inequality (1. then not only will the iteration converge rapidly.1 I − M = I − (I − BA) = BA is nonsingular.2.1 indicates.11) implies the ﬁrst part of (1. The following corollary is a direct consequence of Lemma 1.8) and completes the proof. The following theorem is often referred to as the Banach Lemma.9) and (1. Hence both A and B are nonsingular. Corollary 1.1. It is sometimes possible to precondition a linear equation by multiplying both sides of (1. 1 − I − BA A I − BA . (1.2.2. proves (1.1.Copyright ©1995 by the Society for Industrial and Applied Mathematics. termination decisions based on the Buy this book from SIAM at http://www. has the form xk+1 = (I − BA)xk + Bb. the matrices B that allow us to apply the Banach lemma and its corollary are called approximate inverses. 1 − I − BA A − B −1 ≤ Proof.1. If M < 1 then the iteration (1. Let M = I − BA. . 1 − I − BA Since A−1 = (I − M )−1 B.10) A−1 − B ≤ A−1 ≤ B .9). To complete the proof note that A−1 − B = (I − BA)A−1 . 1 − I − BA B I − BA . A − B −1 = B −1 (I − BA). In the context of Richardson iteration. Definition 1. The second part follows in a similar way from B −1 = A(I − M )−1 . Richardson iteration. then A and B are both nonsingular and (1.12) If the norm of I − BA is small.6) will converge if I − A < 1. and use (1.html.ecsecurehost. By (1.1) by a matrix B BAx = Bb so that convergence of iterative methods is improved.2. but.8) (1. Theorem 1. B is an approximate inverse of A if I − BA < 1.2. If A and B are N × N matrices and B is an approximate inverse of A. This electronic version is for personal use and may not be duplicated or distributed. as Lemma 1. A consequence of Corollary 1. 6 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Noting that (I − M )−1 ≤ ∞ l=0 M l = (1 − M )−1 .1.2. preconditioned with approximate inversion. 1 − I − BA B −1 ≤ A .1 is that Richardson iteration (1.11) A−1 B −1 = (I − M )−1 ≤ 1 1− M = 1 .com/SIAM/FR16.
Matrix splittings and classical stationary iterative methods There are ways to convert Ax = b to a linear ﬁxedpoint iteration that are diﬀerent from (1.1. and Exercise 1.5. [99]. . [126].14) has a partial converse that allows us to completely describe the performance of iteration (1.1. can also be interpreted in this light.1 is a characterization of convergent stationary iterative methods. We mention one other approach. Methods such as Jacobi. [6]. The spectral radius of M is independent of any particular matrix norm of M . and related problems [15]. The spectral radius The analysis in § 1.3. Theorem 1. This electronic version is for personal use and may not be duplicated or distributed.7) to the norm of the matrix M . Hence the performance of the iteration is not completely described by M .13) ρ(A) = max λ = lim An λ∈σ(A) n→∞ 1/n .ecsecurehost. Let M be an N ×N matrix. [117]. [179].3. The proof is left as an exercise. [169]. Lemma 1. that (1. The term on the righthand side of the second equality in (1. Then for any > 0 there is a norm · on RN such that ρ(A) > A − . Let A be an N × N matrix. Theorem 1. in fact.1.html.7) converges for all c ∈ RN if and only if ρ(M ) < 1. This method is a very eﬀective technique for solving diﬀerential equations. A consequence of Theorem 1.1. However the norm of M could be small in some norms and quite large in others.3. We state that converse as a theorem and refer to [105] for a proof.4.7) in terms of spectral radius. integral equations. The iteration (1. 1.14) ρ(A) ≤ A for any induced matrix norm. which tries to approximate A−1 by a polynomial in A [123].2.13) is the limit used by the radical test for convergence of the series An . [111].com/SIAM/FR16.3. Multigrid methods [19]. BASIC CONCEPTS 7 preconditioned residual Bb − BAx will better reﬂect the actual error. Buy this book from SIAM at http://www. and sucessive overrelaxation (SOR) iteration are based on splittings of A of the form A = A1 + A2 . We let σ(A) denote the set of eigenvalues of A.5). Definition 1. The spectral radius of an N × N matrix A is (1. polynomial preconditioning.2 related convergence of the iteration (1. Gauss–Seidel.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The concept of spectral radius allows us to make a complete description. It is clear.3. 1. [100]. The inequality (1.2.
1 The analysis of the method is based on an estimation of the spectral radius of the iteration matrix M = −A−1 A2 . Theorem 1. [105]. We present only one convergence result for the classical stationary iterative methods. 0< j=i Then A is nonsingular and the Jacobi iteration (1. Proof. As a ﬁrst example we consider the Jacobi iteration that uses the splitting A1 = D. [193]. This leads to the iteration matrix MJAC = −D−1 (L + U ).1. These methods are usually less eﬃcient than the Krylov methods discussed in Chapters 2 and 3 or the more modern stationary methods based on multigrid ideas.html.15) converges to x∗ = A−1 b for all b. Let A be an N × N matrix and assume that for all 1≤i≤N (1. Hence MJAC ∞ < 1 and the iteration converges to the unique solution of x = M x + D−1 b. where D is the diagonal of A and L and U are the (strict) lower and upper triangular parts. [144].Copyright ©1995 by the Society for Industrial and Applied Mathematics. Letting (xk )i denote the ith component of the kth iterate we can express Jacobi iteration concretely as (1. Then Ax = b is converted to the ﬁxedpoint problem x = A−1 (b − A2 x).15) (xk+1 )i = a−1 bi ii − j=i aij (xk )j . The limited description in this section is intended as a review that will set some notation to be used later.ecsecurehost.com/SIAM/FR16. Note that A1 is diagonal and hence trivial to invert. A2 = L + U. . This electronic version is for personal use and may not be duplicated or distributed. 8 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS where A1 is a nonsingular matrix constructed so that equations with A1 as coeﬃcient matrix are easy to solve.4. 1 For a detailed description of the classical stationary iterative methods the reader may consult [89].16) aij  < aii . However the classical methods have a role as preconditioners. Also I − M = D−1 A is nonsingular and therefore A is nonsingular. or [200]. Buy this book from SIAM at http://www. Note that the ith row sum of M = MJAC satisﬁes N j=1 mij  = j=i aij  aii  < 1.
A2 = L. A1 = D + L. From the point of view of preconditioning.ecsecurehost. Note also that. the iteration depends on the ordering of the unknowns.html. and hence A−1 y is easy to compute for vectors 1 y.Copyright ©1995 by the Society for Industrial and Applied Mathematics. BJAC = D−1 . MGS = −(D + L)−1 U. For the Jacobi iteration. The performance can be dramatically improved with a good choice of ω but still is not competitive with Krylov methods. . A further disadvantage is that the choice of ω is often diﬃcult to make. Backward Gauss–Seidel begins the update of x with the N th coordinate rather than the ﬁrst. BASIC CONCEPTS 9 Gauss–Seidel iteration overwrites the approximate solution with the new value as soon as it is computed. and the papers cited therein provide additional reading on this topic. unlike Jacobi iteration. resulting in the splitting A1 = D + U. References [200]. This results in the iteration (xk+1 )i = a−1 bi − ii the splitting and iteration matrix aij (xk+1 )j − j<i j>i aij (xk )j . [89].17) For symmetric Gauss–Seidel (1. If A is symmetric then U = LT . This leads to the iteration matrix MSGS = MBGS MGS = (D + U )−1 L(D + L)−1 U. [193]. That means that one wants to ﬁnd B such that M = I − BA and then use B as an approximate inverse. This electronic version is for personal use and may not be duplicated or distributed. Note that A1 is lower triangular. [8]. In that event MSGS = (D + U )−1 L(D + L)−1 U = (D + LT )−1 L(D + L)−1 LT . (1. one wants to write the stationary method as a preconditioned Richardson iteration. A2 = U. The successive overrelaxation iteration modiﬁes Gauss–Seidel by adding a relaxation parameter ω to construct an iteration with iteration matrix MSOR = (D + ωL)−1 ((1 − ω)D − ωU ).18) BSGS = (D + LT )−1 D(D + L)−1 . Buy this book from SIAM at http://www.com/SIAM/FR16. and iteration matrix MBGS = −(D + U )−1 L. A symmetric Gauss–Seidel iteration is a forward Gauss–Seidel iteration followed by a backward Gauss–Seidel iteration.
Buy this book from SIAM at http://www.Copyright ©1995 by the Society for Industrial and Applied Mathematics.5.3.4.2.3. 1.2.5. 1.5.com/SIAM/FR16. Show that if A is symmetric and positive deﬁnite (that is AT = A and xT Ax > 0 for all x = 0) that BSGS is also symmetric and positive deﬁnite.html. . This electronic version is for personal use and may not be duplicated or distributed. 1.ecsecurehost.7) fails to converge. 10 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 1.5. Show that if ρ(M ) ≥ 1 then there are x0 and c such that the iteration (1. Verify equality (1. Prove Theorem 1.1. Exercises on stationary iterative methods 1.18).5.
conjugate gradient and GMRES. The residual is r = b − Ax. Ar0 . So {rk }k≥0 will denote the sequence of residuals rk = b − Axk . we assume that A is a nonsingular N × N matrix and let x∗ = A−1 b. The conjugate gradient (CG) iteration was invented in the 1950s [103] as a direct method. This electronic version is for personal use and may not be duplicated or distributed. Ak−1 r0 ) for k ≥ 1. There are other Krylov methods that are not as well understood as CG or GMRES. Chapter 2 Conjugate Gradient Iteration 2. some measure of error over the aﬃne space x0 + Kk .Copyright ©1995 by the Society for Industrial and Applied Mathematics. The two such methods that we’ll discuss in depth. .ecsecurehost.html.com/SIAM/FR16. and [78]. at the kth iteration. Krylov methods and the minimization property In the following two chapters we describe some of the Krylov space methods for linear equations.6. . where x0 is the initial iterate and the kth Krylov subspace Kk is Kk = span(r0 .1. minimize. Krylov methods do not have an iteration matrix. Recall that A is symmetric if A = AT and positive deﬁnite if xT Ax > 0 for all x = 0. Brief descriptions of several of these methods and their properties are in § 3. [12]. CG is intended to solve symmetric positive deﬁnite (spd) systems. It has come into wide use over the last 15 years as an iterative method and has generally superseded the Jacobi–Gauss–Seidel–SOR family of methods. Unlike the stationary iterative methods. . . As in Chapter 1. 11 Buy this book from SIAM at http://www. .
The development in these notes is diﬀerent from the classical work and more like the analysis for GMRES and CGNR in [134]. We state this as a lemma. Since A is spd we may deﬁne a norm (you should check that this is a norm) by (2. Since A is symmetric and Ax∗ = b −xT Ax∗ − (x∗ )T Ax = −2xT Ax∗ = −2xT b. Therefore x − x∗ 2 A = 2φ(x) + (x∗ )T Ax∗ . and in the section on GMRES that follows. Let S ⊂ RN . In this section.1. Lemma 2. Since (x∗ )T Ax∗ is independent of x. Buy this book from SIAM at http://www. We will use this lemma in the particular case that S = x0 + Kk for some k.1. After that we describe termination criterion. Proof.html. 12 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS In this section we assume that A is spd. ˜ Minimizing φ over any subset of RN is the same as minimizing x − x∗ A over that subset. · A is called the Anorm. we begin with a description of what the algorithm does and the consequences of the minimization property of the iterates. and at the very end.com/SIAM/FR16. the implementation. . Note that if φ(˜) is the minimal value (in RN ) then x ∇φ(˜) = A˜ − b = 0 x x and hence x = x∗ . performance. This electronic version is for personal use and may not be duplicated or distributed. preconditioning.2) 1 φ(x) = xT Ax − xT b 2 over x0 + Kk .ecsecurehost.1) x A = √ xT Ax. Note that x − x∗ 2 A = (x − x∗ )T A(x − x∗ ) = xT Ax − xT Ax∗ − (x∗ )T Ax + (x∗ )T Ax∗ . minimizing φ is equivalent to minimizing x − x∗ 2 and hence to minimizing x − x∗ A .Copyright ©1995 by the Society for Industrial and Applied Mathematics. If xk minimizes φ over S then xk also minimizes x∗ − x A = r A−1 over S. The kth iterate xk of CG minimizes (2. A If e = x − x∗ then e 2 A = eT Ae = (A(x − x∗ ))T A−1 (A(x − x∗ )) = b − Ax 2 A−1 and hence the Anorm of the error is also the A−1 norm of the residual.
1. Since U U T = U T U = I by orthogonality of U . .ecsecurehost.p(0)=1 min p(A)(x∗ − x0 ) A. 2 Buy this book from SIAM at http://www. where U is an orthogonal matrix whose columns are the eigenvectors of A and Λ is a diagonal matrix with the positive eigenvalues of A on the diagonal.Copyright ©1995 by the Society for Industrial and Applied Mathematics. In (2.4) Pk denotes the set of polynomials of degree k. Hence p(A) = U p(Λ)U T . γj z j+1 has degree k and satisﬁes p(0) = 1. Since any w ∈ x0 + Kk can be written as w= k−1 j=0 γj Aj r0 + x0 for some coeﬃcients {γj }. Hence (2.5) x = xT Ax = A1/2 x 2 .3) x∗ − xk A ≤ x∗ − w A for all w ∈ x0 + Kk .2. we have Aj = U Λj U T . CONJUGATE GRADIENT ITERATION 13 2. we can express x∗ − w as x∗ − w = x∗ − x0 − Since Ax∗ = b we have r0 = b − Ax0 = A(x∗ − x0 ) and therefore x∗ − w = x∗ − x0 − where the polynomial p(z) = 1 − k−1 j=0 k−1 j=0 k−1 j=0 γj Aj r0 .html.1 implies that since xk minimizes φ over x0 + Kk (2. 2 A Deﬁne A1/2 = U Λ1/2 U T and note that (2.4) x∗ − xk A = p∈Pk . The spectral theorem for spd matrices asserts that A = U ΛU T . This electronic version is for personal use and may not be duplicated or distributed.com/SIAM/FR16. Consequences of the minimization property Lemma 2. γj Aj+1 (x∗ − x0 ) = p(A)(x∗ − x0 ).
7) and the fact ¯ that p vanishes on σ(A). Let {λi }N be the eigenvalues of A.7) is to show how the CG algorithm can be viewed as a direct method.com/SIAM/FR16.7) xk − x∗ x0 − x∗ A A ≤ max ¯k (z).1. let i=1 p(z) = ¯ N i=1 (λi − z)/λi . based on information on σ(A). since in most applications the number of unknowns N is very large.6) xk − x∗ A ≤ x0 − x∗ A p∈Pk . In that way we showed (in the absence of all roundoﬀ error!) that CG terminated in ﬁnitely many iterations with the exact solution.6). Let A be spd and let {xk } be the CG iterates.2. This electronic version is for personal use and may not be duplicated or distributed. Let A be spd. As a test polynomial.2. This leads to an upper estimate for the number of CG iterations required to reduce the Anorm of the error to a given tolerance. and one cannot aﬀord to perform N iterations. Hence.2. Then p ¯ (2. Here σ(A) is the set of all eigenvalues of A. for any x ∈ RN and p(A)x A = A1/2 p(A)x 2 ≤ p(A) 2 A1/2 x 2 ≤ p(A) 2 x A. together with (2.8) Pk = {p  p is a polynomial of degree k and p(0) = 1. Corollary 2.4) implies that (2. Theorem 2. p Note that our test polynomial had the eigenvalues of A as its roots. One simple application of (2. p z∈σ(A) We will refer to the polynomial pk as a residual polynomial [185]. This is not as good as it sounds. Proof.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The following corollary is an important consequence of (2. Let k be given and let {¯k } be any kth degree polynomial such that pk (0) = 1. 14 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Hence. This. It is best to regard CG as an iterative method.1.7) easy to evaluate. ¯ Definition 2. Then the CG algorithm will ﬁnd the solution within N iterations. When doing that we seek to terminate the iteration when some speciﬁed error tolerance is reached.ecsecurehost.} In speciﬁc contexts we try to construct sequences of residual polynomials.p(0)=1 z∈σ(A) min max p(z). that make either the middle or the right term in (2. The set of kth degree residual polynomials is (2. by (2. .1. Buy this book from SIAM at http://www. ¯ xN − x∗ A ≤ x0 − x∗ A z∈σ(A) max ¯(z) = 0.html. ¯ ¯ p ∈ PN because p has degree N and p(0) = 1.
Assume that there are exactly k ≤ N distinct eigenvalues of A.2. Let b be a linear i=1 combination of k of the eigenvectors of A b= k l=1 γl uil . we can use a similar technique to prove the following theorem. Moreover. Let {λil } be the eigenvalues of A associated with the eigenvectors {uil }k .4) and the fact that x0 = 0 that xk − x∗ A ≤ p(A)x∗ ¯ A = 0. CONJUGATE GRADIENT ITERATION 15 In the two examples that follow we look at some other easy consequences of (2. ¯ One can easily verify that p ∈ Pk . 2. are based on (2. This completes the proof. This electronic version is for personal use and may not be duplicated or distributed. Then the CG iteration terminates in at most k iterations. One typical criterion is small (say ≤ η) relative residuals.7) and therefore estimate the reduction in the relative Anorm of the error.2. Let A be spd. This means that we terminate the iteration after (2. Termination of the iteration In practice we do not run the CG iteration until an exact solution is found.2. however. p(z) = ¯ k l=1 (λil − z)/λil . By the spectral theorem l=1 x∗ = k l=1 (γl /λil )uil .3.3.7). Theorem 2. p(λil ) = 0 for 1 ≤ l ≤ k and ¯ hence p(A)x∗ = ¯ k l=1 p(λil )γl /λil uil = 0.ecsecurehost. Buy this book from SIAM at http://www. we have by (2. . If the spectrum of A has fewer than N points.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The error estimates that come from the minimization property. but rather terminate once some criterion has been satisﬁed.9) b − Axk 2 ≤ η b 2 .html. Let A be spd with eigenvectors {ui }N . Theorem 2.com/SIAM/FR16. ¯ So. We use the residual polynomial. Proof. Then the CG iteration for Ax = b with x0 = 0 will terminate in at most k iterations.
Let A be spd with eigenvalues λ1 ≥ λ2 ≥ . 2 Taking square roots and using (2. .13) follows directly from (2. to predict the performance of the CG iteration based on termination on small relative residuals.12) and (2.11) Proof.7) to predict when the relative Buy this book from SIAM at http://www.html.ecsecurehost. Then for all z ∈ RN . We may write A = U ΛU T as Az = Hence λN A1/2 z 2 2 N i=1 λi (uT z)ui . Let ui be a unit eigenvector corresponding to λi . 16 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Our next task is to relate the relative residual in the Euclidean norm to the relative error in the Anorm. ﬁrst recall that if A = U ΛU T is the spectral decomposition of A and we order the eigenvalues such that λ1 ≥ λ2 ≥ .Copyright ©1995 by the Society for Industrial and Applied Mathematics.12).11) twice. λN . . Proof. So. . using (2.2. A1/2 z 2 = z A (2.13) b r0 2 2 b − Axk b 2 b − Axk b 2 2 = b − Axk b − Ax0 2 2 ≤ κ2 (A) xk − x∗ x∗ − x0 xk − x∗ x∗ − x0 A A A A 2 ≤ κ2 (A) r0 b 2 2 .3.10) and (2. .3.com/SIAM/FR16. Therefore. Clearly z 2 A λN 1/2 z A ≤ Az 2 ≤ λ1 1/2 z A.10) and (2.10) complete the proof. we must not only use (2. Lemma 2. To obtain the inequality on the right of (2. The equality on the left of (2. So κ2 (A) = λ1 /λN .12) is clear and (2. We will do this in the next two lemmas and then illustrate the point with an example. (2. i = λN ≤ Az ≤ λ1 N T 2 i=1 λi (ui z) 2 2 = N 2 T 2 i=1 λi (ui z) N T 2 i=1 λi (ui z) = λ1 A1/2 z 2 . This electronic version is for personal use and may not be duplicated or distributed. then A 2 = λ1 and A−1 2 = 1/λN . Lemma 2.12). λN > 0.10). b − Axk b − Ax0 2 2 = A(x∗ − xk ) A(x∗ − x0 ) 2 2 ≤ λ1 x∗ − xk λN x∗ − x0 A A as asserted. = z T Az = (A1/2 z)T (A1/2 z) = A1/2 z 2 2 which proves (2. .1.
p ≤ x∗ A 10 −k .14) xk − x∗ A A ≤ x∗ A 9≤z≤11 max ¯k (z). Hence.15) xk − x ∗ A ≤ 2 x0 − x ∗ A κ2 (A) − 1 κ2 (A) + 1 k .html. after k iterations we have Axk − b 2 √ ≤ 11 × 10−k /3. Assume that x0 = 0 and that the eigenvalues of A are contained in the interval (9. 11). when k ≥ 4. p 9≤z≤11 max ¯k (z) = 10−k . So. ≤√ κ2 (A) + 1 11 + 3 Buy this book from SIAM at http://www. λ1 ] and that κ2 (A) = λ1 /λN .2 to relate small Anorm errors to small relative residuals. b 2 So. but also use Lemma 2. We consider a very simple example. we can estimate κ2 (A) by κ2 (A) ≤ 11/9.7) to get ¯ xk − x∗ It is easy to see that Hence. A result from [48] (see also [45]) that is. In the case of the above example. then pk ∈ Pk . that is. the size of the Anorm of the error will be reduced by a factor of 10−3 when 10−k ≤ 10−3 . This means that we may apply (2. that is. the size of the relative residual will be reduced by a factor of 10−3 when √ 10−k ≤ 3 × 10−3 / 11.2 we simply note that κ2 (A) ≤ 11/9. √ √ Hence. is (2. . √ κ2 (A) − 1 11 − 3 ≈ . when k ≥ 3. in one sense. after k iterations (2. To use Lemma 2. This electronic version is for personal use and may not be duplicated or distributed. One can obtain a more precise estimate by using a polynomial other than pk in the upper estimate for the righthand side of (2.7).ecsecurehost. ∞). the sharpest possible.05. since ( x − 1)/( x + 1) is an increasing function of x on the interval (1.3. CONJUGATE GRADIENT ITERATION 17 Anorm error is small.com/SIAM/FR16. If we let pk (z) = ¯ (10 − z)k /10k .Copyright ©1995 by the Society for Industrial and Applied Mathematics.3. Note that it is always the case that the spectrum of a spd matrix is contained in the interval [λN .
5.ecsecurehost.15) would predict that xk − x∗ when 2 × (. The estimate based on the clustering gives convergence in 3k iterations when (. (1.3/.com/SIAM/FR16. .3.04 = 82. In fact. Based on this information the best estimate of the condition number of A is κ2 (A) ≤ 400. 1. A ≤ 10−3 x∗ A.25 − z)k (400 − z)2k .2)k ≤ 10−3 or when k > −3/ log10 (. (2. p which is a sharper estimate on convergence. This would indicate fairly slow convergence. However. Exercise 2.91)k < 10−3 or when k > − log10 (2000)/ log10 (.5 also covers this point. if we use as a residual polynomial p3k ∈ P3k ¯ p3k (z) = ¯ It is easy to see that z∈σ(A) (1.91) ≈ 3.8.25)k = (.15) can be very pessimistic. but CG can perform very well.3 ≈ 2.25)k × 4002k max ¯3k (z) ≤ (.91)k .3/1. the CG iteration will converge very rapidly.05) ≈ 3. the estimate in (2.Copyright ©1995 by the Society for Industrial and Applied Mathematics.05k < 10−3 or when k > − log10 (2000)/ log10 (. This electronic version is for personal use and may not be duplicated or distributed. Assume that x0 = 0 and the eigenvalues of A lie in the two intervals (1.25/1.15) predicts 83 iterations and the clustering analysis 15 (the smallest integer multiple of 3 larger than 3 × 4. when inserted into (2. the condition number can be quite large. which also predicts termination within three iterations.html. 400).6.3 = 12. 18 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Therefore (2.2) = 4. From the results above one can see that if the condition number of A is near one. If the eigenvalues cluster in a small number of intervals. which.15) would predict a reduction in the size of the Anorm error by a factor of 10−3 when 2 × .15) gives xk − x∗ x∗ A A ≤ 2 × (19/21)k ≈ 2 × (. Hence (2. When we do.5) and (399.9). We may have more precise information than a single interval containing σ(A).2)k . Even if the condition number Buy this book from SIAM at http://www. We will illustrate this with an example.
18) Proof. Lemma 2.e. . In fact dφ(xk + αpk+1 ) (2. Now. This electronic version is for personal use and may not be duplicated or distributed.1.1). In the context of CG. Buy this book from SIAM at http://www. Let A be spd and let {xk } be the conjugate gradient iterates.com/SIAM/FR16. Implementation The implementation of CG depends on the amazing fact that once xk has been determined.Copyright ©1995 by the Society for Industrial and Applied Mathematics.17) αk+1 = pT (b − Axk ) p T rk k+1 = T k+1 . Then T rk rl = 0 for all 0 ≤ l < k. Lemma 2. the iteration will perform well if the eigenvalues are clustered in a few small intervals. then rk = rk+1 . pT Apk+1 pk+1 Apk+1 k+1 If xk = xk+1 then the above analysis implies that α = 0. 2. for any ξ ∈ Kk dφ(xk + tξ) = ∇φ(xk + tξ)T ξ = 0 dt at t = 0. Since xk minimizes φ on x0 + Kk . We discuss this in § 2.18). (2.html. CONJUGATE GRADIENT ITERATION 19 is large.4. We used this term before in the context of Richardson iteration and accomplished the goal by multiplying A by an approximate inverse. Recalling that ∇φ(x) = Ax − b = −r we have (2.16) can be written as pT Axk + αpT Apk+1 − pT b = 0 k+1 k+1 k+1 leading to (2.1 then implies that T T rk 2 = rk rk = rk rk+1 = 0 and hence xk = x∗ .4. Equation (2. proves that (if we deﬁne p0 = 0) pT rk = 0 for all 0 ≤ l < k ≤ n.16) =0 dα for the correct choice of α = αk+1 . easier to solve) is called preconditioning.ecsecurehost. αk+1 is easy to compute from the minimization property of the iteration. We show that this only happens if xk is the solution.8. if xk = xk+1 . Once pk+1 has been found.4. either xk = x∗ or a search direction pk+1 = 0 can be found very cheaply so that xk+1 = xk + αk+1 pk+1 for some scalar αk+1 . 2 The next lemma characterizes the search direction and. we have..5. unless the l iteration terminates prematurely. as a side eﬀect. this proves (2.19) T ∇φ(xk )T ξ = −rk ξ = 0 for all ξ ∈ Kk . Since rl ∈ Kk for all l < k (see Exercise 2. The transformation of the problem into one with eigenvalues clustered near one (i. such a simple approach can destroy the symmetry of the coeﬃcient matrix and a more subtle implementation is required.
Let A be spd and let {xk } be the conjugate gradient iterates. . it is. (2. . k+1 Proof. If xk = x∗ then xk+1 = xk + αk+1 pk+1 and pk+1 is determined up to a scalar multiple by the conditions (2. Let pk+1 be given by (2. Lemma 2. (2.ecsecurehost.. Now. Since k rk = rk−1 − αk Apk Buy this book from SIAM at http://www. While one might think that wk would be hard to compute. Trivially k+1 (2. a subspace of dimension one less than Kk+1 . pT Aξ = 0 for all ξ ∈ Kk .23) Proof. in the scalar product (x. This electronic version is for personal use and may not be duplicated or distributed. We have the following theorem.4.23) then pT Arl = 0 k+1 for all 0 ≤ l ≤ k − 1. rk−1 ).21) then imply that for all ξ ∈ Kk . Then pk+1 = rk + βk+1 pk for some βk+1 and k ≥ 0. 20 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Lemma 2. Deﬁne p0 = 0. k+1 This uniquely speciﬁes the direction of pk+1 as (2.2 and the fact that Kk = span(r0 .23).2.4. Since Kk ⊂ Kk+1 .html. By Lemma 2.24) T βk+1 = −rk Ark−1 /pT Ark−1 k provided pT Ark−1 = 0. any k+1 pk+1 satisfying (2. Let A be spd and assume that rk = 0. then rl ∈ Kl+1 ⊂ Kk−1 . . (2. we need only verify that a βk+1 can be found so that if pk+1 is given by (2.e.Copyright ©1995 by the Society for Industrial and Applied Mathematics.22) αk+1 pT Aξ = −(Axk − b)T ξ = −∇φ(xk )T ξ = 0.21) ∇φ(xk+1 )T ξ = (Axk + αk+1 Apk+1 − b)T ξ = 0 for all ξ ∈ Kk . k+1 It only remains to solve for βk+1 so that pT Ark−1 = 0.com/SIAM/FR16.1. Theorem 2. up to a scalar multiple. The condition pT Aξ = 0 is called Aconjugacy of pk+1 to Kk .19) and (2. Then for any l ≤ k T pT Arl = rk Arl + βk+1 pT Arl . .4. . in fact.20) can.4. k+1 k If l ≤ k − 2. be expressed as pk+1 = rk + wk with wk ∈ Kk .22) implies that pk+1 ∈ Kk+1 is Aorthogonal (i. y) = xT Ay) to Kk .20) pk+1 ∈ Kk+1 . trivial.2 then implies that pT Arl = 0 for 0 ≤ l ≤ k − 2. (2.
4.27) Proof. Then (2.2.com/SIAM/FR16. k+1 which is equivalent to (2.30) Also note that (2. This electronic version is for personal use and may not be duplicated or distributed. Let A be spd and assume that rk = 0. pT Apk k Now combine (2.25) pT Ark−1 = rk−1 2 /αk = 0. The common implementation of conjugate gradient uses a diﬀerent form for αk and βk than given in (2.26). Note that for k ≥ 0 (2.4. k 2 This completes the proof.23) implies that k+1 (2. To get (2.25) to get βk+1 = T −rk Apk αk .28) T pT rk+1 = rk rk+1 + βk+1 pT rk+1 = 0 k+1 k αk+1 = rk 2 2 pT Apk+1 k+1 rk rk−1 2 2 2.27) note that pT Apk = 0 and hence (2. k T Since rk rk−1 = 0 by Lemma 2. (2.html.29) gives T 0 = pT rk − αk+1 pT Apk+1 = rk k+1 k+1 2 2 − αk+1 pT Apk+1 . k k k βk+1 = T −rk Apk . (2.3.28) is that pT rk = 0 and k hence pT rk = (rk + βk+1 pk )T rk = rk 2 . . An immediate consequence of (2.1 we have (2. 2 βk+1 = by Lemma 2.30).Copyright ©1995 by the Society for Industrial and Applied Mathematics.31).4.29) k+1 2 Taking scalar products of both sides of rk+1 = rk − αk+1 Apk+1 with pk+1 and using (2.17) and (2.26) and (2. and (2.24). Lemma 2.31) pT Apk = pT A(rk−1 + βk pk−1 ) k k = pT Ark−1 + βk pT Apk−1 = pT Ark−1 .ecsecurehost. CONJUGATE GRADIENT ITERATION 21 2 2 we have T rk rk−1 = rk−1 − αk pT Ark−1 . rk−1 2 2 Buy this book from SIAM at http://www.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
22
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Now take scalar products of both sides of rk = rk−1 − αk Apk with rk and use Lemma 2.4.1 to get rk
2 2 T = −αk rk Apk .
Hence (2.27) holds. The usual implementation reﬂects all of the above results. The goal is to ﬁnd, for a given , a vector x so that b−Ax 2 ≤ b 2 . The input is the initial iterate x, which is overwritten with the solution, the right hand side b, and a routine which computes the action of A on a vector. We limit the number of iterations to kmax and return the solution, which overwrites the initial iterate x and the residual norm. Algorithm 2.4.1. cg(x, b, A, , kmax) 1. r = b − Ax, ρ0 = r 2 , k = 1. 2 √ 2. Do While ρk−1 > b 2 and k < kmax (a) if k = 1 then p = r else β = ρk−1 /ρk−2 and p = r + βp (b) w = Ap (c) α = ρk−1 /pT w (d) x = x + αp (e) r = r − αw (f) ρk = r
2 2
(g) k = k + 1 Note that the matrix A itself need not be formed or stored, only a routine for matrixvector products is required. Krylov space methods are often called matrixfree for that reason. Now, consider the costs. We need store only the four vectors x, w, p, and r. Each iteration requires a single matrixvector product (to compute w = Ap), two scalar products (one for pT w and one to compute ρk = r 2 ), and three 2 operations of the form ax + y, where x and y are vectors and a is a scalar. It is remarkable that the iteration can progress without storing a basis for the entire Krylov subspace. As we will see in the section on GMRES, this is not the case in general. The spd structure buys quite a lot. 2.5. Preconditioning To reduce the condition number, and hence improve the performance of the iteration, one might try to replace Ax = b by another spd system with the same solution. If M is a spd matrix that is close to A−1 , then the eigenvalues
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
CONJUGATE GRADIENT ITERATION
23
of M A will be clustered near one. However M A is unlikely to be spd, and hence CG cannot be applied to the system M Ax = M b. In theory one avoids this diﬃculty by expressing the preconditioned problem in terms of B, where B is spd, A = B 2 , and by using a twosided preconditioner, S ≈ B −1 (so M = S 2 ). Then the matrix SAS is spd and its eigenvalues are clustered near one. Moreover the preconditioned system SASy = Sb has y ∗ = S −1 x∗ as a solution, where Ax∗ = b. Hence x∗ can be recovered from y ∗ by multiplication by S. One might think, therefore, that computing S (or a subroutine for its action on a vector) would be necessary and that a matrixvector multiply by SAS would incur a cost of one multiplication by A and two by S. Fortunately, this is not the case. If y k , rk , pk are the iterate, residual, and search direction for CG applied ˆ ˆ to SAS and we let ˆ ˆ ˆ ˆ xk = S y k , rk = S −1 rk , pk = S pk , and zk = S rk , then one can perform the iteration directly in terms of xk , A, and M . The reader should verify that the following algorithm does exactly that. The input is the same as that for Algorithm cg and the routine to compute the action of the preconditioner on a vector. Aside from the preconditioner, the arguments to pcg are the same as those to Algorithm cg. Algorithm 2.5.1. pcg(x, b, A, M, , kmax) 1. r = b − Ax, ρ0 = r 2 , k = 1 2 √ 2. Do While ρk−1 > b 2 and k < kmax (a) z = M r (b) τk−1 = z T r (c) if k = 1 then β = 0 and p = z else β = τk−1 /τk−2 , p = z + βp (d) w = Ap (e) α = τk−1 /pT w (f) x = x + αp (g) r = r − αw (h) ρk = rT r (i) k = k + 1 Note that the cost is identical to CG with the addition of • the application of the preconditioner M in step 2a and • the additional inner product required to compute τk in step 2b.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
24
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Of these costs, the application of the preconditioner is usually the larger. In the remainder of this section we brieﬂy mention some classes of preconditioners. A more complete and detailed discussion of preconditioners is in [8] and a concise survey with many pointers to the literature is in [12]. Some eﬀective preconditioners are based on deep insight into the structure of the problem. See [124] for an example in the context of partial diﬀerential equations, where it is shown that certain discretized secondorder elliptic problems on simple geometries can be very well preconditioned with fast Poisson solvers [99], [188], and [187]. Similar performance can be obtained from multigrid [99], domain decomposition, [38], [39], [40], and alternating direction preconditioners [8], [149], [193], [194]. We use a Poisson solver preconditioner in the examples in § 2.7 and § 3.7 as well as for nonlinear problems in § 6.4.2 and § 8.4.2. One commonly used and easily implemented preconditioner is Jacobi preconditioning, where M is the inverse of the diagonal part of A. One can also use other preconditioners based on the classical stationary iterative methods, such as the symmetric Gauss–Seidel preconditioner (1.18). For applications to partial diﬀerential equations, these preconditioners may be somewhat useful, but should not be expected to have dramatic eﬀects. Another approach is to apply a sparse Cholesky factorization to the matrix A (thereby giving up a fully matrixfree formulation) and discarding small elements of the factors and/or allowing only a ﬁxed amount of storage for the factors. Such preconditioners are called incomplete factorization preconditioners. So if A = LLT + E, where E is small, the preconditioner is (LLT )−1 and its action on a vector is done by two sparse triangular solves. We refer the reader to [8], [127], and [44] for more detail. One could also attempt to estimate the spectrum of A, ﬁnd a polynomial p such that 1 − zp(z) is small on the approximate spectrum, and use p(A) as a preconditioner. This is called polynomial preconditioning. The preconditioned system is p(A)Ax = p(A)b and we would expect the spectrum of p(A)A to be more clustered near z = 1 than that of A. If an interval containing the spectrum can be found, the residual polynomial q(z) = 1 − zp(z) of smallest L∞ norm on that interval can be expressed in terms of Chebyshev [161] polynomials. Alternatively q can be selected to solve a least squares minimization problem [5], [163]. The preconditioning p can be directly recovered from q and convergence rate estimates made. This technique is used to prove the estimate (2.15), for example. The cost of such a preconditioner, if a polynomial of degree K is used, is K matrixvector products for each application of the preconditioner [5]. The performance gains can be very signiﬁcant and the implementation is matrixfree.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
As above.34) rk 2 ≤ pk (AT A)r0 ¯ 2 ≤ r0 2 z∈σ(AT A) max ¯k (z). 2 Hence. for any residual polynomial pk ∈ Pk . more important. [134].33) then y∗ − y 2 AAT = (y ∗ − y)T (AAT )(y ∗ − y) = (AT y ∗ − AT y)T (AT y ∗ − AT y) = x∗ − x 2 2 is minimized over y0 + Kk at each iterate. The ﬁrst is that the condition number of the coeﬃcient matrix AT A is the square of that of A. there are situations where this is not possible. CGNR and CGNE If A is nonsingular and nonsymmetric.ecsecurehost. one might consider solving Ax = b by applying CG to the normal equations (2. The third. The analysis with residual polynomials is similar to that for CG. [134].33) AAT y = b and then set x = AT y to solve Ax = b. when we consider the AT A norm of the error.html. The reason for this name is that the minimization property of CG as applied to (2. Hence the name Conjugate Gradient on the Normal equations to minimize the Residual. The second is that two matrixvector products are needed for each CG iterate since w = AT Ap = AT (Ap) in CGNR and w = AAT p = A(AT p) in CGNE. As we will see in the chapter on nonlinear problems. [78]. This approach [103] is called CGNR [71]. one could solve (2. CONJUGATE GRADIENT ITERATION 25 2. p Buy this book from SIAM at http://www.32) asserts that x∗ − x 2 AT A = (x∗ − x)T AT A(x∗ − x) = (Ax∗ − Ax)T (Ax∗ − Ax) = (b − Ax)T (b − Ax) = r 2 is minimized over x0 + Kk at each iterate. There are three disadvantages that may or may not be serious.6. This electronic version is for personal use and may not be duplicated or distributed.com/SIAM/FR16. Alternatively. Conjugate Gradient on the Normal equations to minimize the Error. ¯ (2.32) AT Ax = AT b. the analysis for CGNE is essentially the same.Copyright ©1995 by the Society for Industrial and Applied Mathematics. we have x∗ − x 2 AT A = (x∗ − x)T AT A(x∗ − x) = A(x∗ − x) 2 2 = r 2.33) asserts that if y ∗ is the solution to (2. . This approach [46] is now called CGNE [78]. We consider the case for CGNR. disadvantage is that one must compute the action of AT on a vector as part of the matrixvector product involving AT A. The advantages of this approach are that all the theory for CG carries over and the simple implementation for both CG and PCG can be used. The reason for this name is that the minimization property of CG as applied to (2.
which corresponds exactly to the termination criterion. y < 1.html. We discretize with a ﬁvepoint centered diﬀerence scheme with n2 points and mesh width h = 1/(n + 1). the residual polynomial is to be maximized over the eigenvalues of AT A. Most signiﬁcantly.ecsecurehost.7.8. to reﬂect the boundary conditions. The inputs.36) = (αij + α(i+1)j )(u(i+1)j − uij ) −(α(i−1)j + αij )(uij − u(i−1)j ) + (αi(j+1) + αij )(ui(j+1) − uij ) −(αij + αi(j−1) )(uij − ui(j−1) ) Buy this book from SIAM at http://www.10). We consider the discretization of the partial diﬀerential equation (2. 1) = u(0. The unknowns are uij ≈ u(xi . described in the comment lines.3. y ≤ 1 (Exercise 2. . the right hand side vector b. MATLAB functions for the matrixvector product and (optionally) the preconditioner. hence we need not prove a result like Lemma 2. y) = 0.com/SIAM/FR16. y) on 0 < x. and deﬁne αij = −a(xi . On return the code supplies the approximate solution x and the history of the iteration as the vector of residual norms. We express the discrete matrixvector product as (Au)ij (2. 0) = u(x. We set u0j = u(n+1)j = ui0 = ui(n+1) = 0. are the initial iterate.35) ˙ −∇(a(x. This electronic version is for personal use and may not be duplicated or distributed. 26 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS There are two major diﬀerences between (2. and iteration parameters to specify the maximum number of iterations and the termination criterion. y < 1 subject to homogeneous Dirichlet boundary conditions u(x. xj ) where xi = ih for 1 ≤ i ≤ n.Copyright ©1995 by the Society for Industrial and Applied Mathematics. which is the set of the squares of the singular values of A.7). y)∇u) = f (x. One can verify [105] that the diﬀerential operator is positive deﬁnite in the Hilbert space sense and that the ﬁvepoint discretization described below is positive deﬁnite if a > 0 for all 0 ≤ x.2. The estimate is in terms of the l2 norm of the residual.34) and (2. Hence the performance of CGNR and CGNE is determined by the distribution of singular values. Examples for preconditioned conjugate iteration In the collection of MATLAB codes we provide a code for preconditioned conjugate gradient iteration. 2. x0 . 0 < x. xj )h−2 /2. y) = u(1.
CONJUGATE GRADIENT ITERATION 27 for 1 ≤ i. The eﬀectiveness of such a preconditioner has been analyzed in [124] and some of the many ways to implement the solver eﬃciently are discussed in [99].8. [188]. however. The reader is asked to quantify this in terms of execution times in Exercise 2. which is useful for computing the action of A on u and applying fast solvers.ecsecurehost. In the results reported here we took n = 31 resulting in a system with N = n2 = 961 unknowns. For a preconditioner we use a Poisson solver. By this we mean an operator G such that v = Gw is the solution of the discrete form of −vxx − vyy = w. For the computations reported in this section we took a(x. Note that the unpreconditioned iteration is slowly convergent. subject to homogeneous Dirichlet boundary conditions. The properties of CG on the preconditioned problem in the continuous case have been analyzed in [48]. and the representation as a onedimensional array. is more eﬃcient that the classical stationary iterative methods. See the routine fish2d in collection of MATLAB codes for an example of how to do this in MATLAB.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The fast Poisson solver in the collection of Buy this book from SIAM at http://www. j ≤ n. For many types of domains and boundary conditions. The initial iterate was u0 = 0. y) = cos(x) and took the right hand side so that the exact solution was the discretization of 10xy(1 − x)(1 − y) exp(x4. This can be explained by the fact that the eigenvalues are not clustered and κ(A) = O(1/h2 ) = O(n2 ) = O(N ) and hence (2. which is what pcgsol expects to see. The initial iterate was the zero vector. Because of this their execution time is less than a simple count of ﬂoatingpoint operations would indicate. in the case of the MATLAB environment used in this book. In Fig. 2. We allowed up to 100 CG iterations.5 ). Note that the reduction in r is not monotone.1 the solid line is a plot of rk / b and the dashed line a plot of u∗ −uk A / u∗ −u0 A . We expect secondorder accuracy from the method and accordingly we set termination parameter = h2 = 1/1024. [186]. plotting rk 2 / b 2 on a semilog scale. Poisson solvers can be designed to take advantage of vector and/or parallel architectures or. .com/SIAM/FR16. designed to take advantage of fast MATLAB builtin functions. This example illustrates the importance of a good preconditioner. This is consistent with the theory. We will report our results graphically.15) indicates that convergence will be slow. For the MATLAB implementation we convert freely from the representation of u as a twodimensional array (with the boundary conditions added).html. which predicts decrease in e A but not necessarily in r as the iteration progresses. Even the unpreconditioned iteration. and [187].9. This electronic version is for personal use and may not be duplicated or distributed.
28 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 1 10 Relative Residual and Anorm of Error 10 0 10 1 10 2 10 3 10 4 0 10 20 30 Iterations 40 50 60 Fig.2. but the number of iterations required to reduce the relative residual by a given amount is independent of the mesh spacing [124]. In Fig. 2. CG for 2D elliptic equation.com/SIAM/FR16. 10 1 10 0 Relative Residual 10 1 10 2 10 3 10 4 0 10 20 30 Iterations 40 50 60 Fig. the builtin function fft. . PCG for 2D elliptic equation.html. 2. The preconditioned iteration required 5 iterations for convergence and the unpreconditioned iteration 52.2 the solid line is the graph of rk 2 / b 2 for the preconditioned iteration and the dashed line for the unpreconditioned.ecsecurehost. This electronic version is for personal use and may not be duplicated or distributed. codes fish2d is based on the MATLAB fast Fourier transform. Not only does the preconditioned iteration converge more rapidly. We caution the reader that the preconditioned iteration is not as much faster than the Buy this book from SIAM at http://www.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 2.1.
the cost of the preconditioner is considerable. Hence. this speed is a result of the eﬃciency of the MATLAB fast Fourier transform. . the execution time of the preconditioned iteration was about 60% of that of the unpreconditioned.ecsecurehost.8.87 million ﬂoatingpoint operations.2 million ﬂoatingpoint operations while the preconditioned iteration required .11 you are asked to compare execution times for your own environment. In Exercise 2. As we remarked above.com/SIAM/FR16. CONJUGATE GRADIENT ITERATION 29 unpreconditioned one as the iteration count would suggest. In the MATLAB environment we used. The MATLAB flops command indicates that the unpreconditioned iteration required roughly 1. Buy this book from SIAM at http://www.html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Duplicate the results in § 2. in MATLAB by writing the matrixvector product routines and using the MATLAB codes pcgsol and fish2d. y) > 0 for all 0 ≤ x. y ≤ 1.8. What happens as N is increased? How are the performance √ and accuracy aﬀected by changes in a(x.ecsecurehost. y)? Try a(x.8.5.7. 30 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 2.3.9.2.8. 2.8. assume that the cost of a matrix vector multiply is 4N ﬂoatingpoint multiplies.2.8.36) is symmetric and 2 positive deﬁnite on Rn if a(x. give a rough estimate of the number of CG iterates required to reduce the relative residual to O(1/N ).4. Exercises on conjugate gradient 2. Let Λ be a diagonal matrix with Λii = λi and let p be a polynomial. y) = . Let A be a nonsingular matrix with all singular values in the interval (1.8. Prove that rl ∈ Kk for all l < k. . Let {xk } be the conjugate gradient iterates.8. 2. 2. 2. 1. 2). 2. 2. Show that if A has constant diagonal then PCG with Jacobi preconditioning produces the same iterates as CG with no preconditioning. Compare the execution times on your computing environment (using the cputime command in MATLAB. 2.8. Estimate the number of CGNR/CGNE iterations required to reduce the relative residual by a factor of 10−4 . Estimate the number of ﬂoatingpoint operations reduce the A norm of the error by a factor of 10−3 using CG iteration. for instance). Assume that A is N × N . 2. If κ(A) = O(N ). 2. Explain your ﬁndings.com/SIAM/FR16. Give upper estimates based on (2.8. Buy this book from SIAM at http://www.6) for the number of CG iterations required to reduce the A norm of the error by a factor of 10−3 and for the number of CG iterations required to reduce the residual by a factor of 10−3 .1) ∪ (2. This electronic version is for personal use and may not be duplicated or distributed.8. Show that there is a spd B such that B 2 = A.7 for example.11. Let A be spd.1 + x and examine the accuracy of the result. Prove Theorem 2.8.6. Assume that A is spd and that σ(A) ⊂ (1. Prove that p(Λ) = maxi p(λi ) where · is any induced matrix norm.8. Is B unique? 2.html. nonsingular. 2.1.10.2). and spd.3. For the matrix A in problem 5. Prove that the linear transformation given by (2.8.Copyright ©1995 by the Society for Industrial and Applied Mathematics.
14. Plot φ(xn ) as a function of n for each of the examples. Apply CG and PCG to solve the ﬁvepoint discretization of −uxx (x. subject to the inhomogeneous Dirichlet boundary conditions u(x. Jacobi.8.com/SIAM/FR16. y.Copyright ©1995 by the Society for Industrial and Applied Mathematics.12. and symmetric Gauss–Seidel). Compare the performance with respect to both computer time and number of iterations to preconditioning with the Poisson solver. y) + ex+y u(x. Modify pcgsol so that φ(x) is computed and stored at each iterate and returned on output.8. 2. y) = 1. y) = 1. 2. y) − uyy (x.7.15. This electronic version is for personal use and may not be duplicated or distributed. How does the performance compare to CG and PCG? 2.18) preconditioners for the elliptic boundary value problem in § 2. y < 1.html. CONJUGATE GRADIENT ITERATION 31 2.13.7. Experiment with diﬀerent mesh sizes and preconditioners (Fast Poisson solver. 0 < x. Implement Jacobi (1. .8. 1) = u(1. 0 < x.ecsecurehost. y) = 0. Use the Jacobi and symmetric Gauss–Seidel iterations from Chapter 1 to solve the elliptic boundary value problem in § 2. u(0. Buy this book from SIAM at http://www. < 1.17) and symmetric Gauss–Seidel (1. 0) = u(x.8.
32 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Buy this book from SIAM at http://www.html.com/SIAM/FR16. This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost. .Copyright ©1995 by the Society for Industrial and Applied Mathematics.
¯ From this we have the following corollary. Theorem 3. Let A be nonsingular and let xk be the kth GMRES iteration. one must store a basis for Kk .1. Then for all pk ∈ Pk ¯ (3. Corollary 3. Unlike CGNR.1.3) rk r0 2 2 ≤ pk (A) 2 . This is a signiﬁcant advantage in many cases.1.1.com/SIAM/FR16. Chapter 3 GMRES Iteration 3. We have just proved the following result. and therefore storage requirements increase as the iteration progresses. GMRES does not require computation of the action of AT on a vector. The use of residual polynomials is made more complicated because we cannot use the spectral theorem to decompose A.1. (3.1) The beginning of this section is much like the analysis for CG. This electronic version is for personal use and may not be duplicated or distributed. ¯ ¯ Hence if x ∈ x0 + Kk then r = p(A)r0 where p ∈ Pk is a residual polynomial. Note that if x ∈ x0 + Kk then x = x0 + and so b − Ax = b − Ax0 − k−1 j=0 k−1 j=0 k j=1 γj Aj r0 γj A j+1 r0 = r0 − γj−1 Aj r0 . ¯ 33 Buy this book from SIAM at http://www. Then for all pk ∈ Pk ¯ (3. Moreover. The minimization property and its consequences The GMRES (Generalized Minimum RESidual) was proposed in 1986 in [167] as a Krylov subspace method for nonsymmetric systems. Let A be nonsingular and let xk be the kth GMRES iteration.2) rk 2 = min p(A)r0 ¯ p∈Pk 2 ≤ pk (A)r0 2 .ecsecurehost. The kth (k ≥ 1) iteration of GMRES is the solution to the least squares problem minimizex∈x0 +Kk b − Ax 2 .html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. .
Here V H is the complex conjugate transpose of V .3. Our use of complex arithmetic will be implicit for the most part and is needed only so that we may admit the possibility of complex eigenvalues of A. 2 by ≤ V 2 V −1 2 pk (Λ) ¯ 2 ≤ κ2 (V ) max ¯k (z). We pay a price in that we must use complex arithmetic for the only time in this book. In that case the columns of V are the eigenvectors of A and V −1 = V H . By ¯ (3. Let A = V ΛV −1 be a nonsingular diagonalizable matrix. Proof.1.com/SIAM/FR16. We can use the structure of a diagonalizable matrix to prove the following result. We can easily estimate pk (A) ¯ pk (A) ¯ as asserted. The characteristic polynomial of A is p(z) = det(A − zI). Here Λ is a (possibly complex!) diagonal matrix with the eigenvalues of A on the diagonal. If A is a diagonalizable matrix and p is a polynomial then p(A) = V p(Λ)V −1 A is normal if the diagonalizing transformation V is orthogonal. In Chapter 2 we applied the spectral theorem to obtain more precise information on convergence rates. This is not an option for general nonsymmetric matrices. However.ecsecurehost. 34 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS We can apply the corollary to prove ﬁnite termination of the GMRES iteration. Recall that the scalar product in C N . Then for all pk ∈ Pk ¯ (3. Let pk ∈ Pk . p(0) = det(A) = 0 since A is nonsingular. the space of complex N vectors. if A is diagonalizable we may use (3. Let A be nonsingular. Let xk be the kth GMRES iterate. Theorem 3. Theorem 3. p has degree N . we will use the l2 norm in C N . p z∈σ(A) 2 ¯ Proof. Recall that A is diagonalizable if there is a nonsingular (possibly complex!) matrix V such that A = V ΛV −1 .2. Then the GMRES algorithm will ﬁnd the solution within N iterations. p z∈σ(A) Buy this book from SIAM at http://www.4) rk r0 2 2 ≤ κ2 (V ) max ¯k (z). It is well known [141] that p(A) = pN (A) = 0. rN = b − AxN = 0 and hence xN is the solution. Hence we will switch to complex matrices and vectors. and so pN (z) = p(z)/p(0) ∈ PN ¯ is a residual polynomial.3). This electronic version is for personal use and may not be duplicated or distributed.Copyright ©1995 by the Society for Industrial and Applied Mathematics.html. In particular.1. In the remainder of this section we must use complex arithmetic to analyze the convergence. is xH y.2) to get information from clustering of the spectrum just like we did for CG. .
Assume that A has only k distinct eigenvalues. The estimate (3.1. GMRES is best thought of as an iterative method.5) holds without requiring a lemma like Lemma 2.4) directly to estimate the ﬁrst k such that (3. As the set of nondiagonalizable matrices has measure zero in the space of N × N matrices.5) holds when 102−k < η or when k > 2 + log10 (η). Termination As is the case with CG. Let A be a nonsingular normal matrix. we terminate the iteration when (3. We can use (3. with x0 = 0.html.3) and (3. but will otherwise resemble those for CG.5. Then the GMRES iteration. Let b be a linear combination of k of the eigenvectors of A b= k l=1 γl uil . We assume that x0 = 0 and hence r0 = b. Let pk (z) = (1 − z)k . The convergence rate estimates for the diagonalizable case will involve κ2 (V ). [33]. If A is not diagonalizable. 11).4).3) and (3. If A is normal.1. κ2 (V ) = 1.2) that (3.4. we look at some easy consequences of (3. Buy this book from SIAM at http://www. Theorem 3.ecsecurehost. for Ax = b will terminate in at most k iterations. Assume that I − A consequence of (3. Using the residual polynomial pk (z) = (10 − z)k /10k we ﬁnd ¯ rk r0 2 2 ≤ (100)10−k = 102−k .5) rk 2 ≤η b 2 for the purposes of this example. and that κ2 (V ) = 100. [134].6) illustrates the potential beneﬁts of a good approximate inverse preconditioner. Assume that A = V ΛV −1 is diagonalizable. It is a direct ¯ rk 2 ≤ ρk r0 2 .2. As we did for CG. This electronic version is for personal use and may not be duplicated or distributed. rate estimates have been derived in [139]. Let A be a nonsingular diagonalizable matrix. Hence (3.6) 2 ≤ ρ < 1. Hence we conﬁne our attention to diagonalizable matrices.com/SIAM/FR16.Copyright ©1995 by the Society for Industrial and Applied Mathematics. GMRES ITERATION 35 It is not clear how one should estimate the condition number of the diagonalizing transformation if it exists. 3.3. Then GMRES will terminate in at most k iterations. [192]. This is particularly so for the ﬁnite diﬀerence Jacobian matrices we consider in Chapters 6 and 8. and [34]. Theorem 3. the chances are very high that a computed matrix will be diagonalizable. that the eigenvalues of A lie in the interval (9.2. . Again we look at examples. of course. As was the case with CG.
and [36] for discussions of and pointers to additional references to several questions related to nonnormal matrices. if the product M A can be formed without error. [33]. [164]. If I − AM 2 = ρ < 1. The residual of the preconditioned problem is the same as that of the unpreconditioned problem.7 we use the Poisson solver preconditioner for nonsymmetric partial diﬀerential equations. This is a very active area of research and we refer to [134].6) to the preconditioned system. If one can ﬁnd M such that I − MA 2 = ρ < 1. Preconditioning Preconditioning for GMRES and other methods for nonsymmetric problems is diﬀerent from that for CG. Right preconditioning has been used as the basis for a method that changes the preconditioner as the iteration progresses [166]. Preconditioning done in this way is called left preconditioning. [34].1. CGNR.com/SIAM/FR16. Multigrid [99] and alternating direction [8]. [120]. Hence. one can apply the algorithm directly to the system M Ax = M b (or AM y = b). we might expect the relative residual of the preconditioned system to be a better indicator of the relative error than the relative residual of the original system. then applying GMRES to M Ax = M b allows one to apply (3. unlike PCG.1.html. the value of the relative residuals as estimators of the relative error is unchanged. Buy this book from SIAM at http://www. Most of the preconditioning ideas mentioned in § 2. One important aspect of implementation is that. we have. This electronic version is for personal use and may not be duplicated or distributed. [182] methods have similar performance and may be more generally applicable.Copyright ©1995 by the Society for Industrial and Applied Mathematics. . This is called right preconditioning. by Lemma 1. 36 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS The convergence estimates for GMRES in the nonnormal case are much less satisfying that those for CG.ecsecurehost. [72]. one can write a single matrixvector product routine for M A (or AM ) that includes both the application of A to a vector and that of the preconditioner. [183].3. if M A has a smaller condition number than A. There is no concern that the preconditioned system be spd and hence (3. Incomplete factorization (LU in this case) preconditioners can be used [165] as can polynomial preconditioners. one can solve the system AM y = b with GMRES and then set x = M y. Hence. or GMRES in the normal case. In the examples in § 3. Some hybrid algorithms use the GMRES/Arnoldi process itself to construct polynomial preconditioners for GMRES or for Richardson iteration [135]. However there are two diﬀerent ways to view preconditioning. 3. Again we mention [8] and [12] as a good general references for preconditioning. CGNE. Hence. If r = M Ax − M b is the residual for the preconditioned system. ek e0 2 2 ≤ κ2 (M A) rk r0 2 2 .5 are useful for GMRES as well.6) essentially tells the whole story.
.7) very eﬃciently and the resulting least squares problem requires no extra multiplications of A with vectors. a map that computes the action of A on a vector. Suppose one had an orthogonal projector Vk onto Kk . our least squares problem in Rk is (3. however. A division by zero is referred to as breakdown and happens only if the solution to Ax = b is in x0 + Kk−1 . Then any z ∈ Kk can be written as z= vlk k l=1 yl vlk . Hence. . GMRES implementation: Basic ideas Recall that the least squares problem deﬁning the kth GMRES iterate is minimizex∈x0 +Kk b − Ax 2 . For i = 1. . k. one can represent (3. The Gram–Schmidt procedure for formation of an orthonormal basis for Kk is called the Arnoldi [4] process. The algorithm computes an orthonormal basis for Kk and stores it in the columns of V .7) minimizey∈Rk r0 − AVk y 2 .html.com/SIAM/FR16. arnoldi(x0 . .1. If one uses Gram–Schmidt orthogonalization. GMRES ITERATION 37 3. .Copyright ©1995 by the Society for Industrial and Applied Mathematics.1) to a least squares where problem in Rk for the coeﬃcient vector y of z = x − x0 . Hence we can convert (3. Algorithm 3. The problem with such a direct approach is that the matrix vector product of A with the basis matrix Vk must be taken at each iteration. 2. Since x − x0 = Vk y for some y ∈ Rk . we must have xk = x0 + Vk y where y minimizes b − A(x0 + Vk y) 2 = r0 − AVk y 2 . Buy this book from SIAM at http://www.ecsecurehost. A. say. k − 1 vi+1 = Avi − Avi − i T j=1 ((Avi ) vj )vj i T j=1 ((Avi ) vj )vj 2 If there is never a division by zero in step 2 of Algorithm arnoldi. V ) 1. The data are vectors x0 and b.4. This electronic version is for personal use and may not be duplicated or distributed. and a dimension k. then the columns of the matrix Vk are an orthonormal basis for Kk . b.4. This is a standard linear least squares problem that could be solved by a QR factorization. Deﬁne r0 = b − Ax0 and v1 = r0 / r0 2 . is the lth column of Vk .
. . we can use it to implement GMRES in an eﬃcient way. 0)T ∈ Ri we have ri 2 2 = b − Axi 2 = r0 − A(xi − x0 ) 2 . .. where y k minimizes βe1 − Hk y 2 over Rk . The upper Hessenberg structure can be exploited to make the solution of the least squares problems very eﬃcient [167]. using the orthogonality of Vk+1 . If the Arnoldi process does not break down. the norm of rk can be found without explicitly forming xk and computing rk = b − Axk . . which overwrites the initial iterate x and the residual norm. a vector x so that b − Ax 2 AVk = Vk+1 Hk .Copyright ©1995 by the Society for Industrial and Applied Mathematics. We limit the number of iterations to kmax and return the solution. i. let the vectors vj be generated by Algorithm arnoldi. Setting y = βH −1 e1 proves that ri = 0 by the minimization property. 0. Since xi − x0 ∈ Ki there is y ∈ Ri such that xi − x0 = Vi y. and a map that computes the action of A on a vector. and let i be the smallest integer for which Avi − i j=1 ((Avi )T vj )vj = 0.html. Then x = A−1 b ∈ x0 + Ki . The input is the initial iterate.ecsecurehost. the righthand side b. Buy this book from SIAM at http://www. where · Rk+1 denotes the Euclidean norm in Rk+1 . This electronic version is for personal use and may not be duplicated or distributed. for a given . Note that when y k has been computed.com/SIAM/FR16. x. Since r0 = βVi e1 and Vi is an orthogonal matrix ri 2 = Vi (βe1 − Hy) 2 = βe1 − Hy Ri+1 . Since the columns of Vi are an orthonormal basis for Ki . Proof. for some y k ∈ Rk . rk = b − Axk = r0 − A(xk − x0 ) = Vk+1 (βe1 − Hk y k ). Hence xk = x0 + Vk y k . .8) Hence. ≤ b 2. hij = 0 if i > j + 1. Let A be nonsingular. Setting β = r0 and e1 = (1.4. the k +1×k matrix Hk whose entries are hij is upper Hessenberg.1. rk 2 = Vk+1 (βe1 − Hk y k ) 2 = βe1 − Hk y k Rk+1 . By hypothesis Avi ∈ Ki and hence AKi ⊂ Ki .e. We have. By the Gram–Schmidt construction. The Arnoldi process (unless it terminates prematurely with a solution) produces matrices {Vk } with orthonormal columns such that (3. we have AVi = Vi H. H is nonsingular since A is. 38 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Lemma 3.9) The goal of the iteration is to ﬁnd. (3. where H is an i × i matrix. Set hji = (Avj )T vi .
While ρ > b 2 and k < kmax do (a) k = k + 1 (b) for j = 1. We discuss implementation of GMRES(m) later in this section. (g) ρ = βe1 − Hk y k 3. . and we use the method of [167] in the collection of MATLAB codes. 0. which is often the case for large problems. For very large problems this becomes prohibitive and the iteration is restarted when the available room for basis vectors is exhausted. We replace Rk+1 . 0)T (f) e1 = ∈ Rk+1 Minimize βe1 − Hk y k Rk+1 over Rk to obtain y k . kmax. k hjk = (Avk )T vj (c) vk+1 = Avk − (d) hk+1. will not hold and the residual and approximate solution could be inaccurate.com/SIAM/FR16. A more serious problem with the implementation proposed in Algorithm gmresa is that the vectors vj may become nonorthogonal as a result of cancellation errors. β = ρ. However. and the simple oneline MATLAB approach can be eﬃcient for such problems. It is an important property of GMRES that the basis for the Krylov space must be stored as the iteration progress. . There is no general convergence theorem for the restarted algorithm and restarting will slow the convergence down.9). xk = x0 + Vk y k . r = b − Ax. k = 0 2. .ecsecurehost. gmresa(x. ρ) 1. This means that in order to perform k GMRES iterations one must store k vectors of length N . This electronic version is for personal use and may not be duplicated or distributed. (3. .html. call GMRES and explicitly test the residual b−Axk if k = m upon termination. Algorithm gmresa can be implemented very straightforwardly in MATLAB.4. GMRES ITERATION 39 Algorithm 3. call GMRES again with x0 = xk . A. [197]. Note that xk is only computed upon termination and is not needed within the iteration. Step 2f can be done with a single MATLAB command. If this happens. the backward division operator. .Copyright ©1995 by the Society for Industrial and Applied Mathematics. A partial remedy is to replace the classical Gram–Schmidt orthogonalization in Algorithm gmresa with modiﬁed Gram–Schmidt orthogonalization. .2. . the result from the previous call. b. Buy this book from SIAM at http://www. . The savings are slight if k is small relative to N . [167]. If the norm of the residual is larger than . which depends on this orthogonality. at a cost of O(k 3 ) ﬂoatingpoint operations. when it works it can signiﬁcantly reduce the storage costs of the iteration. One way to implement this is to set kmax to the maximum number m of vectors that one can store. .k = vk+1 k j=1 hjk vj 2 2 (e) vk+1 = vk+1 / vk+1 (1. This restarted version of the algorithm is called GMRES(m) in [167]. There are more eﬃcient ways to solve the least squares problem in step 2f. ρ = r 2 . . v1 = r/ r 2 .
δ 0 δ We orthogonalize the columns of A with classical Gram–Schmidt to obtain 1. .9715e − 08 −9. .0000e − 07 −1. k T vk+1 = vk+1 − (vk+1 vj )vj . While modiﬁed Gram–Schmidt and classical Gram–Schmidt are equivalent in inﬁnite precision. In fact v2 v3 ≈ −. [160]. and is not particularly eﬃcient.0436e − 07 1. Let δ = 10−7 and deﬁne 1 1 1 A = δ δ 0 . While ρ > b 2 and k < kmax do (a) k = k + 1 Buy this book from SIAM at http://www.Copyright ©1995 by the Society for Industrial and Applied Mathematics. V = 1. Second. One can take a second pass through the modiﬁed Gram–Schmidt process and restore lost orthogonality [147]. 40 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS the loop in step 2c of Algorithm gmresa with vk+1 = Avk for j = 1. . ρ) 1.4. For modiﬁed Gram–Schmidt 1. The versions we implement in the collection of MATLAB codes use modiﬁed Gram–Schmidt. The outline of our implementation is Algorithm gmresb. gmresb(x. We illustrate this point with a simple example from [128].0000e + 00 . j. v1 = r/ r 2 . k = 0 2. A. this version is very simple and illustrates some important ideas. there is an opportunity to compensate for a loss of orthogonality in the basis vectors for the Krylov space.0436e − 07 −1. β = ρ.com/SIAM/FR16.3568e − 02 T The columns of VU are not orthogonal at all. . We present a better implementation in Algorithm gmres.0000e + 00 1.html.9905e − 01 . First. b.0000e − 07 1.0456e − 14 1.3565e − 16 T Here vi vj − δij  ≤ 10−8 for all i. This electronic version is for personal use and may not be duplicated or distributed. V = 1. This implementation solves the upper Hessenberg least squares problem using the MATLAB backward division operator. we see that xk need only be computed after termination as the least squares residual ρ can be used to approximate the norm of the residual (they are identical in exact arithmetic).0000e − 07 −1.0436e − 07 9.0456e − 14 1.ecsecurehost. . r = b − Ax.004. ρ = r 2 .0000e + 00 4. the modiﬁed form is much more likely in practice to maintain orthogonality of the basis.3.0000e − 07 1.0000e + 00 1. However.0000e + 00 4. kmax. Algorithm 3. doing the computations in MATLAB.
. One can test for loss of orthogonality [22]. . .ecsecurehost. .Copyright ©1995 by the Society for Industrial and Applied Mathematics. one can still lose orthogonality in the columns of V . .com/SIAM/FR16.k = vk+1 2 2 • vk+1 = vk+1 / vk+1 Buy this book from SIAM at http://www. . . Even if modiﬁed Gram–Schmidt orthogonalization is used. .html. . For example. GMRES ITERATION 41 (b) vk+1 = Avk for j = 1. One can reorthogonalize in every iteration or only if a test [147] detects a loss of orthogonality. xk = x0 + Vk y k . k T hjk = vk+1 vj vk+1 = vk+1 − hjk vj • hk+1. [147]. k T i. . . and reorthogonalize if needed or use a more stable means to create the matrix V [195]. hjk = vk+1 vj ii. These more complex implementations are necessary if A is ill conditioned or many iterations will be taken. 0. 0)T (e) e1 = ∈ Rk+1 Minimize βe1 − Hk y k Rk+1 to obtain y k ∈ Rk .k = vk+1 2 Rk+1 . . . . . There is nothing to be gained by reorthogonalizing more than once [147]. • vk+1 = vk+1 / vk+1 2 with a second pass (reorthogonalization). The modiﬁed Gram–Schmidt process with reorthogonalization looks like • vk+1 = Avk for j = 1.k = vk+1 2 2 (d) vk+1 = vk+1 / vk+1 (1. .k = vk+1 2 • If loss of orthogonality is detected For j = 1. vk+1 = vk+1 − hjk vj (c) hk+1. k T hjk = vk+1 vj vk+1 = vk+1 − hjk vj • hk+1. one can augment the modiﬁed Gram–Schmidt process • vk+1 = Avk for j = 1. (f) ρ = βe1 − Hk y k 3. . . This electronic version is for personal use and may not be duplicated or distributed. k T htmp = vk+1 vj hjk = hjk + htmp vk+1 = vk+1 − htmp vj • hk+1. .
Copyright ©1995 by the Society for Industrial and Applied Mathematics. When k is not too large and the cost of matrixvector products is high.0011 0 . the bruteforce method can be very costly.1 we tabulate relative residuals as a function of the iteration counter for classical Gram–Schmidt without reorthogonalization (CGM). especially when implemented in an environment like MATLAB. The k scalar products require O(kN ) ﬂoatingpoint operations and the cost of a solution of the Hessenberg least squares problem. and (3. The idea is that if the new vector is very small relative to Avk then information may have been lost and a second pass through the modiﬁed Gram–Schmidt process is needed. Hence the total cost of the m GMRES iterations is m matrixvector products and O(m4 + m2 N ) ﬂoatingpoint operations. 42 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS One approach to reorthogonalization is to reorthogonalize in every step.10) is the default in the MATLAB code gmresa. the reorthogonalization strategy based on (3. however. This doubles the cost of the computation of V and is usually unnecessary. reorthogonalization based on the test (3. by QR factorization or the MATLAB backward division operator. More eﬃcient and equally eﬀective approaches are based on other ideas. x0 = (0. A= 0 0 0 104 While in inﬁnite precision arithmetic only three iterations are needed to solve the system exactly. While classical Gram–Schmidt fails. To illustrate the eﬀects of loss of orthogonality and those of reorthogonalization we apply GMRES to the diagonal system Ax = b where b = (1. In Table 3. We provide an implementation of Algorithm gmresb in the collection of MATLAB codes. For large k. This is an appealing algorithm. 1. This electronic version is for personal use and may not be duplicated or distributed.10) (MGMPO). Reorthogonalization is done after the Gram–Schmidt loop and before vk+1 is normalized if (3.11) . a bruteforce solution to the least squares problem using the MATLAB backward division operator is not terribly ineﬃcient.001 0 0 . modiﬁed Gram–Schmidt without reorthogonalization (MGM). in step 2e of gmresb is O(k3 ) ﬂoatingpoint operations. We employ this test in the MATLAB code gmres with δ = 10−3 . Buy this book from SIAM at http://www. because of its simplicity.html.10) is almost as eﬀective as the much more expensive approach of reorthogonalizing in every step. 0)T . 1)T . and the solution of the Hessenberg least squares problem in step 2e. A variation on a method from [147] is used in [22].com/SIAM/FR16. k scalar products.ecsecurehost. we ﬁnd in the MATLAB environment that a solution to full precision requires more than three iterations unless reorthogonalization is applied after every iteration. The method based on (3. . say.10) Avk 2 + δ vk+1 2 = Avk 2 to working precision. and reorthogonalization in every iteration (MGMFO). 0. The kth GMRES iteration requires a matrixvector product.
88e02 6.69e05 4. Implementation: Givens rotations If k is large.00e+00 8.16e01 3. . [22].Copyright ©1995 by the Society for Industrial and Applied Mathematics. s = sin(θ) for θ ∈ [−π. which makes an angle of −θ with the xaxis through an angle θ so that it overlaps the xaxis. GMRES ITERATION Table 3.html.1 Eﬀects of reorthogonalization.12) G= .88e02 6. −s).74e05 3. G c −s = 1 0 . where c = cos(θ).00e+00 8.34e34 3.04e18 MGMPO 1.37e05 MGM 1. 43 k 0 1 2 3 4 5 6 7 8 9 10 CGM 1.04e24 MGMFO 1.00e+00 8.42e08 5. An N × N Givens rotation replaces a 2 × 2 block on the diagonal of the Buy this book from SIAM at http://www.35e05 3. Householder reﬂections [195].88e02 6. The orthogonal matrix G rotates the vector (c. This electronic version is for personal use and may not be duplicated or distributed.87e05 3.88e02 6.00e05 2.74e05 2. The implementation in Algorithm gmres and in the MATLAB code collection is from [167].16e01 3.53e05 2.5.00e+00 8.16e01 3.16e01 3. π]. This implementation maintains the QR factorization of Hk in a clever way so that the cost for a single GMRES iteration is O(N k) ﬂoatingpoint operations.com/SIAM/FR16.70e08 3.ecsecurehost. implementations using Givens rotations [167]. The O(k 2 ) cost of the triangular solve and the O(kN ) cost of the construction of xk are incurred after termination. A 2 × 2 Givens rotation is a matrix of the form c −s s c (3.42e08 3. or a shifted Arnoldi process [197] are much more eﬃcient than the bruteforce approach in Algorithm gmresb.
s c 1 0 . s1 ) by (3. s2 ) to H. where (3. changes h11 and the subsequent columns). . 0 . 44 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS N × N identity matrix with a 2 × 2 Givens rotation.14) c1 = h11 / h2 + h2 and s1 = −h21 / h2 + h2 . we can now apply G2 (c2 .. This reduction can be accomplished in O(N ) ﬂoatingpoint operations and hence is far more eﬃcient than a solution by a singular value decomposition or a reduction based on Householder transformations. of course. Let H be an N × M (N ≥ M ) upper Hessenberg matrix with rank M . then the ﬁrst column of H now has only a single nonzero element h11 .. .13) 0 .15) c2 = h22 / h2 + h2 and s2 = −h32 / h2 + h2 . [184]. . We deﬁne G1 = G1 (c1 . .. Similarly.. 22 32 22 32 and annihilate h32 . They are of particular value in reducing Hessenberg matrices to triangular form and thereby solving Hessenberg least squares problems such as the ones that arise in GMRES. . G= . . A straightforward application of these ideas to Algorithm gmres would solve the least squares problem by computing the product of k Givens rotations Q. This electronic version is for personal use and may not be duplicated or distributed. 0 . . . setting g = βQe1 . 1 . 0 1 0 0 . .Copyright ©1995 by the Society for Industrial and Applied Mathematics. . c −s . Our notation is that Gj (c. where Rk is the k + 1 × k triangular factor of the QR factorization of Hk . and noting that βe1 − Hk y k Rk+1 = Q(βe1 − Hk y k ) Rk+1 = g − Rk y k Rk+1 .. Continuing in this way and setting Q = GN . G1 we see that QH = R is upper triangular. Givens rotations are used to annihilate single nonzero elements of matrices in reduction to triangular form [89]. This method is also used in the QR algorithm for computing eigenvalues [89]. 0 . (3.html. We reduce H to triangular form by ﬁrst multiplying the matrix by a Givens rotation that annihilates h21 (and. .. . .. ..13) with a 2 × 2 Givens rotation in rows and columns j and j + 1. . 11 21 11 21 If we replace H by G1 H. Buy this book from SIAM at http://www. s) is an N × N givens rotation of the form (3.com/SIAM/FR16..ecsecurehost.. Note that G2 does not aﬀect the ﬁrst column of H.
(e) vk+1 = vk+1 / vk+1 (f) ii. hk+1.10) to test for loss of orthogonality. kmax.ecsecurehost. k T i.k . Buy this book from SIAM at http://www. ck = hk. then computing the Givens rotation Gk+1 that annihilates the (k + 2)nd element of Qk hk+2 . A.k = 0 iv.j = hi. sk )g. GMRES ITERATION 45 In the context of GMRES iteration. we can update both Qk and Rk by ﬁrst multiplying hk+2 by Qk (that is applying the ﬁrst k Givens rotations to hk+2 ).j for 1 ≤ i. To see this. . b. r = b − Ax. . Algorithm 3. vk+1 = vk+1 − hjk vj (c) hk+1. g = ρ(1.k − sk hk+1. . v1 = r/ r 2 . We overwrite the upper triangular part of Hk with the triangular part of the QR factorization of Hk in the MATLAB code. . iii. sk ) . If k > 1 apply Qk−1 to the kth column of H.Copyright ©1995 by the Society for Industrial and Applied Mathematics.k = vk+1 2 (d) Test for loss of orthogonality and reorthogonalize if necessary. we can incrementally perform the QR factorization of H as the GMRES iteration progresses [167]. xk = x0 + Vk y k . 4. . G1 (c1 . and ﬁnally setting Qk+1 = Gk+1 Qk and forming Rk+1 by augmenting Rk with Gk+1 Qk hk+2 . . β = ρ. hjk = vk+1 vj ii. ρ) 1. note that if Rk = Qk Hk and. j ≤ k. This electronic version is for personal use and may not be duplicated or distributed. 0)T ∈ Rkmax+1 2. Set ri. h2 + h2 k. sk = −hk+1. ρ = r 2 . Solve the upper triangular system Ry k = w. The MATLAB implementation of Algorithm gmres uses (3. (g) ρ = (g)k+1 . k = 0. .1. The MATLAB implementation of Algorithm gmres stores Qk by storing the sequences {cj } and {sj } and then computing the action of Qk on a vector x ∈ Rk+1 by applying Gj (cj . While ρ > b 2 and k < kmax do (a) k = k + 1 (b) vk+1 = Avk for j = 1. . 0. . sj ) in turn to obtain Qk x = Gk (ck . s1 )x.html.k /ν hk. we add the new column hk+2 to Hk . .com/SIAM/FR16. after orthogonalization.k /ν.k = ck hk.5.k . however. g = Gk (ck . ν = 2 i. . gmres(x. 3. Set (w)i = (g)i for 1 ≤ i ≤ k.k k+1.
x. and {vk }m . A. Other methods for nonsymmetric systems The methods for nonsymmetric linear systems that receive most attention in this book.Copyright ©1995 by the Society for Industrial and Applied Mathematics. However. the arguments to Algorithm gmresm are the same as those for Algorithm gmres. CGNR and CGNE have the disadvantages that a transposevector product must be computed for each iteration and that the coeﬃcient matrix AT A or AAT has condition number the square of that of the original matrix. . m. we take the view that preconditioners will be applied externally to the iteration. can be analyzed by a common residual polynomial approach. Restarting can seriously degrade performance. While ρ > b 2 and k < kmax do (a) gmres(x. kmax. 46 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS We close with an example of an implementation of GMRES(m) . In § 3.8 we give an example of how this squaring of the condition number can lead to failure of the iteration.com/SIAM/FR16. . b. have modest storage requirements that do not depend on the number of iterations needed for convergence. This implementation does not test for success and may. methods based on shortterm recurrences such as CG that also satisfy minimization or conjugacy conditions cannot be constructed for general matrices. ρ) 2.2. This electronic version is for personal use and may not be duplicated or distributed. An ideal method would. . Consistently with our implementation of GMRES. m. Algorithm 3. the storage requirements increase as the iteration progresses. gmresm(x.5. as with CG. therefore. these methods can also be implemented in a manner that incorporates Buy this book from SIAM at http://www.html. [74]. . CGNR. a basis for Kk must be stored to compute xk . fail to terminate. GMRES. All the methods we present in this section require two matrixvector products for each iteration. Hence. and CGNE. but. ρ) (b) k = k + m The storage costs of m iterations of gmres or of gmresm are the m + 2 vectors b. like CG. Aside from the parameter m. and only terminate if an acceptable approximate solution has been found. We discuss only a small subset of these methods and refer the reader to [12] and [78] for pointers to more of the literature on this subject. be based on some kind of minimization principle or conjugacy. k = m 3. b.6. but can still be quite useful. A. gmres(x.ecsecurehost. as we have seen. k=1 3. ρ) 1. GMRES needs only matrixvector products and uses A alone. it may be impossible to store enough basis vectors and the iteration may have to be restarted. The methods we describe in this section fall short of the ideal. For large and illconditioned problems. However. share the properties that they are easy to implement. m. and converge in N iterations for all nonsingular A. b. You are asked to repair this in exercise 7. only need matrixvector products. A.
instead. While there are approaches Buy this book from SIAM at http://www. there is no guarantee that ρk in step 2b or pT Ap. [78]. . Kk = span(ˆ0 . GMRES ITERATION 47 the preconditioner and uses the residual for the original system to control termination. . ρ0 = 1. If either ρk−1 = 0 or p ˆ place. The algorithm gets its name because it produces sequences of residuals {rk }. We do not recommend use of BiCG and present this algorithm only as a basis for discussion of some of the properties of this class of algorithms. [76]. Do While r 2 > b 2 and k < kmax (a) k = k + 1 (b) ρk = rT r. does not enforce a minimization principle.com/SIAM/FR16. {ˆk } and search directions {pk }.1. p = r + β p ˆ ˆ ˆ (d) v = Ap (e) α = ρk /(ˆT v) p (f) x = x + αp (g) r = r − αv. This algorithmic description is explained and ˆ motivated in more detail in [122]. 3. the algorithm can become unstable or produce inaccurate results. p = p = 0. r = r. e. however. A. BiCG produces the same iterations as CG (but computes everything except x twice).16) is equivalent to the minimization principle (2. the denominator in step 2e. (AT )k−1 r0 ) r ˆ ˆ is the Krylov space for AT and the vector r0 . b. BiCG. A is not spd. rk rl = 0 if k = l and the search directions {pk } and {ˆk } satisfy ˆT p the biconjugacy property pT Apl = 0 if k = l. . will not ˆ T Ap = 0 we say that a breakdown has taken vanish. and [191].6. . AT r0 . Algorithm 3. Using the notation of Chapter 2 and [191] we give an implementation of BiCG making the choice r0 = r0 . {ˆk } such that biorthogonality r p holds.16) where T rk w = 0 for all w ∈ Kk . i. r0 is a usersupplied vector and is ˆ ˆ often set to r0 . The earliest such method BiCG (Biconjugate gradient) [122]. kmax) ˆ 1.ecsecurehost. [76]. r = b − Ax. r = r − αAT p ˆ ˆ ˆ Note that if A = AT is spd. the kth residual must satisfy the biorthogonality condition (3. β = ρk /ρk−1 ˆ (c) p = r + βp.2) for CG [89]. . If. k = 0 ˆ 2. bicg(x.1.html.6. Even if these quantities are nonzero but very small. This electronic version is for personal use and may not be duplicated or distributed. . ˆk In the symmetric positive deﬁnite case (3.Copyright ©1995 by the Society for Industrial and Applied Mathematics.
if suﬃcient storage is available. one can restart the BiCG iteration or pass the computation to another algorithm. but that the cost of the GMRES iteration increases (in terms of ﬂoatingpoint operations) as the iteration progresses. CGS. ˆ BiCG sometimes performs extremely well and the remaining algorithms in this section represent attempts to capture this good performance and damp the erratic behavior when it occurs.2. We can compare the bestcase performance of BiCG with that of GMRES. there are no methods that both limit storage and completely eliminate the possibility of breakdown [74]. [148]. The algorithm takes advantage of the fact that (3. All of the methods we present in this section can break down and should be implemented with that possibility in mind. the eﬀort in computing r ˆ at each iteration is wasted in that r makes no contribution to x. 48 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS [80]. The performance of the algorithm can be erratic or even unstable with residuals increasing by several orders of magnitude from one iteration to the next. [79]. The possibility of breakdown is small and certainly should not eliminate the algorithms discussed below from consideration if there is not enough storage for GMRES to perform well.17) rk = pk (A)r0 and rk = pk (AT )ˆ0 .html. ˆ p p r p ˆ The other references to AT can be eliminated in a similar fashion and an iteration satisfying rk = pk (A)2 r0 (3. such as GMRES.18) ¯ is produced. One should also keep in mind that a single BiCG iteration requires two matrixvector products and a GMRES iterate only one. [81]. 3. ¯ ˆ ¯ r Hence. Conjugate Gradient Squared.17). Moreover. [78]. This electronic version is for personal use and may not be duplicated or distributed. but not necessarily in terms of computational work).ecsecurehost. there are other problems with BiCG. which at the least will require additional programming and may not be possible at all. This explains the name.17) implies that the scalar product rT r in BiCG (step 2b of ˆ Algorithm bicg) can be represented without using AT as T rk rk = (¯k (A)r0 )T (¯k (AT )ˆ0 ) = (¯k (A)2 r0 )T r0 . However. by the minimization property for GMRES GM rk RES 2 Bi−CG ≤ rk 2. to avoid some breakdowns in many variations of BiCG. Breakdowns aside.com/SIAM/FR16. Once breakdown has occurred. will always reduce the residual more rapidly than BiCG (in terms of iterations. A transposevector product is needed.Copyright ©1995 by the Society for Industrial and Applied Mathematics. . reminding us that GMRES. where pk is the same polynomial produced by BiCG and used in ¯ (3.6. A remedy for one of the problems with BiCG is the Conjugate Gradient Squared (CGS) algorithm [180]. Buy this book from SIAM at http://www. Note that there is pk ∈ Pk such that both ¯ (3.
CGS ˆ replaces the transposevector product with an additional matrixvector product and applies the square of the BiCG polynomial to r0 to produce rk .6. of course. .17) and (3.com/SIAM/FR16.21) ¯ ¯ qk (z) = pk (z) + βk qk−1 (z). b. v0 = Ap0 . cgs(x. BiCGSTAB is the most eﬀective representative of this class of algorithms. βk = ρk /ρk−1 ˆT uk = rk + βk qk pk = uk + βk (qk + βk pk−1 ) vk = Apk Breakdowns take place when either ρk−1 or σk−1 vanish. GMRES ITERATION 49 The work used in BiCG to compute r is now used to update x.18) then ¯ (3.6.html. ¯ We provide no MATLAB implementation of BiCG or CGS. that if pk is the residual polynomial from (3. Do While r 2 > b 2 and k < kmax (a) k = k + 1 ˆT (b) σk−1 = r0 vk−1 (c) αk−1 = ρk−1 /σk−1 (d) qk = uk−1 − αk−1 vk−1 (e) xk = xk−1 + αk−1 (uk−1 + qk ) rk = rk−1 − αk−1 A(uk−1 + qk ) (f) ρk = r0 rk . In our description of CGS we will index the vectors and scalars that would be overwritten in an implementation so that we can describe some of the ideas in TFQMR later on. kmax) 1.19) GM r2k RES 2 CGS ≤ rk 2. Since p2 ∈ P2k we have ¯k (3. This may. then αk−1 = 0 for all k.4. This description is taken from [77]. p0 = u0 = r0 = b − Ax 2. Algorithm 3. k = 0 ˆT 4. One can show [180].Copyright ©1995 by the Society for Industrial and Applied Mathematics. We present an algorithmic description of CGS that we will refer to in our discussion of TFQMR in § 3. If the algorithm does not break down. A. change the convergence properties for the worse and either improve good convergence or magnify erratic behavior [134]. [77].ecsecurehost. This electronic version is for personal use and may not be duplicated or distributed. x0 = x. r0 = r0 ˆ 3. ρ0 = r0 r0 . [191].20) ¯ ¯ pk (z) = pk−1 (z) − αk−1 z qk−1 (z) ¯ ¯ where the auxiliary polynomials qk ∈ Pk are given by q0 = 1 and for k ≥ 1 by ¯ (3. CGS has the same potential for breakdown as BiCG. Buy this book from SIAM at http://www.2. .
ecsecurehost. BiCGSTAB. The cost in storage and in ﬂoatingpoint operations per iteration remains bounded for the entire iteration. p.Copyright ©1995 by the Society for Industrial and Applied Mathematics. kmax) ˆ ˆT 1. r0 = r = r. . The examples in [191] indicate the eﬀectiveness of this approach. There is no convergence theory for BiCGSTAB and no estimate of the residuals that is better than that for CGS.22) where qk (z) = k i=1 (1 − ωi z).19). [98].3. we use r0 = r0 ˆ and follow the notation of [191]. This electronic version is for personal use and may not be duplicated or distributed. We provide an implementation of Algorithm bicgstab in the collection of MATLAB codes.18) with rk = qk (A)pk (A)r0 (3. r. b. The reason for this is that the cost of orthogonalization in GMRES can be much more than that of a matrixvector product if the dimension of the Krylov space is large.html. Do While r 2 > b 2 and k < kmax (a) k = k + 1 (b) β = (ρk /ρk−1 )(α/ω) (c) p = r + β(p − ωv) (d) v = Ap rT (e) α = ρk /(ˆ0 v) (f) s = r − αv. letting s overwrite r when needed. r0 . 50 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 3.8. v = p = 0. Algorithm 3. One must store seven vectors (x. A single iteration reˆ quires four scalar products. r = b − Ax.3. ρ1 = r0 r ˆ 2. t = As rT (g) ω = tT s/ t 2 . b. which is described and motivated fully in [191].com/SIAM/FR16. t). Buy this book from SIAM at http://www. The performance in cases where the spectrum has a signiﬁcant imaginary part can be improved by constructing qk to have complex conjugate pairs of roots. which can be thought of [12] as a blend of BiCG and GMRES(1). The Biconjugate gradient stabilized method (BiCGSTAB) [191] attempts to smooth the convergence of CGS by replacing (3. BiCGSTAB can have a much lower average cost per iterate than GMRES.6. In a situation where many GMRES iterations are needed and matrixvector product is fast. ρ0 = α = ω = 1. In our description of the implementation. v. We will present such a case in § 3. . k = 0. ρk+1 = −ωˆ0 t 2 (h) x = x + αp + ωs (i) r = s − ωt Note that the iteration can break down in steps 2b and 2e. (3. A. The constant ωi is selected to minimize ri = qi (A)pi (A)r0 as a function of ωi .6. bicgstab(x.
. The quasiresidual is related to the true residual by a fullrank linear transformation r = Lq and reduction of the residual can be approximately measured by reduction in q. . Wm+1 the N × (m + 1) j=1 matrix with columns {wj }m+1 . .23) in matrix form as AYm = Wm+1 Bm . . . if m = 2k − 1 is odd. Therefore. We express (3. a quasiresidual over z ∈ RN . Now we consider the quasiminimal residual (QMR) family of algorithms [80]. α0 . −1 .. Returning to Algorithm cgs we deﬁne sequences ym = and wm = uk−1 q k if m = 2k − 1 is odd.25) for some z ∈ Rm . We will focus on the transposefree algorithm TFQMR proposed in [77]. TFQMR.. [37]. if m = 2k is even. [77]. . . therefore.26) rm = r0 − AYm z = Wm+1 (e1 − Bm z)..21).ecsecurehost. . [202]. αk = 0 for all k ≥ 0.. the p CGS residual satisﬁes CGS rk = w2k+1 .4. . 0 −1 −1 0 diag(α0 . .Copyright ©1995 by the Society for Industrial and Applied Mathematics.com/SIAM/FR16. . (3. . 0 . The speciﬁc formulation of q and L depends on the algorithm. All algorithms in this family minimize the norm of an easily computable quantity q = f − Hz.20) and (3. to illustrate the quasiminimization idea.6. 1 0 1 . and Bm is the (m + 1) × m matrix given by j=1 −1 Bm = 0 . . We assume that CGS does not break down and. If we let r be the nearest integer less than or equal to a real r we have Aym = (wm − wm+1 )/α (m−1)/2 .html. .24) where Ym is the N × m matrix with columns {yj }m . . α (m−1)/2 ) . (3. This electronic version is for personal use and may not be duplicated or distributed. . Buy this book from SIAM at http://www. 1 Our assumptions that no breakdown takes place imply that Km is the span of {yj }m and hence j=1 xm = x0 + Ym z (3. Hence. (3. . . GMRES ITERATION 51 3. [81]. .23) where the denominator is nonzero by assumption. (¯k (A))2 r0 p p (A)¯ ¯k pk−1 (A)r0 q Here the sequences {¯k } and {¯k } are given by (3. 0 . . if m = 2k is even.
28) rm = r0 − AYm z = Wm+1 Ω−1 (fm+1 − Hm z) m+1 Ω = diag(ω1 . This electronic version is for personal use and may not be duplicated or distributed. The solution to the least squares problem (3. Note that if k corresponds to the iteration counter for CGS. therefore. In a sense. base our termination criterion on τm . 0. approximate residuals are not computed in Algorithm tfqmr.ecsecurehost. Buy this book from SIAM at http://www. . . e1 = (1. Assume that the TFQMR iteration does not break down or terminate with GM the exact solution for k iterations and let 1 ≤ m ≤ 2k.Copyright ©1995 by the Society for Industrial and Applied Mathematics. and we do this in Algorithm tfqmr. . as in our discussion of GMRES.html.6.29) minimizez∈Rm fm+1 − Hm z 2 . Then if ξm > 0 m+1 (3. we could minimize rm by solving m+1 the least squares problem (3.32) GM τm ≤ rm RES 2 /ξm . The diagonal entries in the matrix Ωm+1 are called weights. . . Hence (3.26) as (3. One can [80]. Now if we pick the weights ωi = wi norm one. The weights make it possible to derive straightforward estimates of the residual in terms of easily computed terms in the iteration.31) rm 2 2 then the columns of Wm+1 Ω−1 have m+1 2 = Wm+1 Ω−1 (fm+1 − Hm z) m+1 √ ≤ τm m + 1. Let rm RES be the residual for the mth GMRES iteration. [78]. b ∈ RN be given. If Wm+1 Ω−1 were an orthogonal matrix.30) xm = x0 + Ym zm . Theorem 3. .com/SIAM/FR16. We can. the conditioning of the matrix Wm+1 can be improved by scaling. each k produces two approximate solutions (corresponding to m = 2k − 1 and m = 2k). Instead we have the quasiresidual norm τm = fm+1 − Hm zm 2 . 0)T ∈ Rm . . 52 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS where.27) and rewrite (3.29) (thereby quasiminimizing the residual) to obtain zm and then set (3. . . Let A be an N × N matrix and let x0 . also use the quasiminimization condition to estimate the TFQMR residual in terms of residual polynomials.29) can be done by using Givens rotations to update the factorization of the upper Hessenberg matrix Hm [77] and this is reﬂected in the implementation description below.1. As one can see from that description. Let ξm be the smallest singular value of Wm+1 Ω−1 . ωm ) where fm+1 = Ωm+1 e1 = ωm+1 e1 and Hm = Ωm+1 Bm is bidiagonal and hence upper Hessenberg. Let (3. The quasiminimization idea is to solve (3. ω2 .
u2 = Ay2 (c) For i.1. iii. β = ρk /ρk−1 Buy this book from SIAM at http://www.6. Do While k < kmax (a) k = k + 1 T (b) σk−1 = r0 v. Let A be an N × N matrix and let x0 .6. by the quasiminimization property GM rm RES 2 ≥ ξm τm as desired. Assume that TFQMR does not break down or terminate with the solution for k iterations. ˆ Algorithm 3.4. Hence. d = 0 T ρ0 = r0 r0 . iv. Let A = V ΛV −1 be a nonsingular diagonalizable matrix.1. u1 = v = Ay1 .html. Corollary 3. We may write GM GM rm RES = r0 + Ym zm RES and obtain GM rm RES 2 GM = Wm+1 Ω−1 (fm+1 −Hm zm RES ) m+1 2 GM ≥ ξm fm+1 −Hm zm RES 2. b ∈ RN be given. A. 2 (m = 2k − 2 + j) w = w − αk−1 uj d = yj + (θ2 η/αk−1 )d √ θ = w 2 /τ .3 to (3. tfqmr(x.32) and use (3. η = c2 αk−1 x = x + ηd √ If τ m + 1 ≤ b 2 terminate successfully T (d) ρk = r0 w. For 1 ≤ m ≤ 2k let ξm be the smallest singular value of Wm+1 Ω−1 m+1 and let xm be given by (3.2. rm r0 2 2 ≤ √ −1 m + 1ξm κ(V ) max φ(z) z∈σ(A) for all φ ∈ Pm . if ξm > 0. k = 0. kmax) 1. j = 1. τ = r 2 .Copyright ©1995 by the Society for Industrial and Applied Mathematics. b.1. α = ρk−1 /σk−1 . vi. c = 1/ 1 + θ2 τ = τ θc.31) to obtain the following result.com/SIAM/FR16. y2 = y1 − αk−1 v. ii. . This electronic version is for personal use and may not be duplicated or distributed. We apply Theorem 3. Then within (N + 1)/2 iterations the TFQMR iteration will either break down or terminate with the solution. v. The implementation below follows [77] with r0 = r0 used throughout. η = 0 2. θ = 0.ecsecurehost.3. Theorem 3. Then. w1 = y1 = r0 = b − Ax. As a corollary we have a ﬁnite termination result and a convergence estimate for diagonalizable matrices similar to Theorem 3. GMRES ITERATION 53 Proof.30). .6.
y) = 1. y) = 0. a MATLAB function for the matrixvector product. and TFQMR. y)u(x. [34]. described in the comment lines. whose storage requirements do not increase with the number of iterations. the preconditioned system is GLu = Gf .com/SIAM/FR16. For general coeﬃcients {aj } the operator L is not selfadjoint and its discretization is not symmetric. 54 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS (e) y1 = w + βy2 . Examples for GMRES iteration These examples use the code gmres from the collection of MATLAB codes.7. As before. y) on 0 < x. In all the GMRES examples. y)uy (x. the righthand side vector b.8 × 10−4 b 2 . As in § 2. we limit the number of GMRES iterations to 60. We let n = 31 to create a system with 961 unknowns. The inputs. y) = 20y. are the initial iterate x0 . y) + a3 (x.33) (Lu)(x. and a3 (x. y < 1. We compute Lu in a matrixfree manner as we did in § 2. [121].7 we discretize with a ﬁvepoint centered difference scheme with n2 points and mesh width h = 1/(n + 1). If we let Gu denote the action of the Poisson solver on u. and iteration parameters to specify the maximum number of iterations and the termination criterion. 0) = u(x. we expect secondorder accuracy and terminate the iteration when rk 2 ≤ h2 b 2 ≈ 9. We take advantage of this in the MATLAB code tfqmr to potentially reduce the matrixvector product cost by one.7. y) = u(1. y) = −(uxx (x.html. 1) = u(0. y) + uyy (x. y) = 1. We consider the discretization of the partial diﬀerential equation (3.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The motivation for this choice is similar to that for the conjugate gradient computations and has been partially explained theoretically in [33]. 3. BiCGSTAB.ecsecurehost. and [139]. For the computations reported in this section we took a1 (x. As a preconditioner we use the fast Poisson solver fish2d described in § 2. This electronic version is for personal use and may not be duplicated or distributed. y)ux (x. a2 (x. The unknowns are uij ≈ u(xi . CGNR. xj ) where xi = ih for 1 ≤ i ≤ n. y)) + a1 (x. For methods like GMRES(m). y < 1 subject to homogeneous Dirichlet boundary conditions u(x. .7. y) = f (x. 0 < x. We do not pass the preconditioner to the GMRES code. u1 = Ay1 (f) v = u1 + β(u2 + βv) Note that y2 and u2 = Ay2 need only be computed if the loop in step 2c does not terminate when j = 1. Buy this book from SIAM at http://www. we allow for more iterations. rather we pass the preconditioned problem. y) +a2 (x.
For the restarted iteration. BiCGSTAB. This electronic version is for personal use and may not be duplicated or distributed. As in § 2.7 we took the right hand side so that the exact solution was the discretization of 10xy(1 − x)(1 − y) exp(x4. Hence. However.com/SIAM/FR16. Restarting after 3 iterations is more costly in terms of the iteration count than not restarting.html. a single CGNR iteration costs roughly the same as two GMRES iterations. CGNR has the eﬀect of squaring the condition number. GMRES(3) in the terminology of § 3. CGNR (or CGNE) is an alternative to GMRES that does not require storage of the iteration history. 3.7 million ﬂoating point operations and the preconditioned iteration required 13 iterations and 2. in the case where the cost of the matrixvector product dominates the computation. the restarted iteration terminated successfully. GMRES ITERATION 55 u0 = 0 was the initial iterate.3 shows. which is given by (GL)∗ u = L∗ G∗ u = L∗ Gu. In this case the MATLAB flops command indicates that the unpreconditioned iteration terminated after 223 iterations and 9.3 million ﬂoating point operations and terminated after 8 iterations.1 and 3. Note the importance of preconditioning. The cost of the application of the transpose of L or GL is the same as that of the application of L or GL itself.8.33). However. we require the transpose of GL. In both ﬁgures the solid line is the history of the preconditioned iteration and the dashed line that of the unpreconditioned.6 million ﬂoating point operations and converged in 56 iterations. The MATLAB flops command indicates that the unpreconditioned iteration required 7.5 ).2 we plot iteration histories corresponding to preconditioned or unpreconditioned GMRES. The dashed line corresponds to Buy this book from SIAM at http://www. even in the unpreconditioned case.Copyright ©1995 by the Society for Industrial and Applied Mathematics.4. we restarted the algorithm after three iterations. Examples for CGNR. and TFQMR iteration When a transpose is available. In other computing environments preconditioners which vectorize or parallelize particularly well might perform in the same way. . the transpose (adjoint) of L is given by L∗ u = −uxx − uyy − a1 ux − a2 uy + a3 u as one can derive by integration by parts. The preconditioned iteration required 1. Note that the plots are semilog plots of r 2 / b 2 and that the righthand sides in the preconditioned and unpreconditioned cases are not the same. 3. In Figs. The execution times in our environment indicate that the preconditioned iterations are even faster than the diﬀerence in operation counts indicates.ecsecurehost. 3. The reason for this is that the FFT routine is a MATLAB intrinsic and is not interpreted MATLAB code.6 million ﬂoating point operations. To apply CGNR to the preconditioned problem. this eﬀect has serious consequences as Fig. For elliptic problems like (3.
2.e.ecsecurehost. GMRES for 2D elliptic equation... .Copyright ©1995 by the Society for Industrial and Applied Mathematics. This is a result of squaring the large condition number. The matrix corresponding to the discretization of L∗ L has a large condition number. 56 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 0 10 10 Relative Residual 1 10 2 10 3 10 4 0 10 20 30 Iterations 40 50 60 Fig.e. The MATLAB flops command indicates that the unpreconditioned iteration required 13. This electronic version is for personal use and may not be duplicated or distributed. GMRES(3) for 2D elliptic equation. 3. i.com/SIAM/FR16. converging in eight iterations. i.7 million ﬂoatingpoint operations Buy this book from SIAM at http://www. 3. We limited the number of iterations to 310 = 10m and the unpreconditioned formulation of CGNR had made very little progress after 310 iterations.1. 10 0 10 Relative Residual 1 10 2 10 3 10 4 0 50 100 Iterations 150 200 250 Fig. CGNR on the unpreconditioned problem. CG applied to L∗ Lu = L∗ f .html. CG applied to L∗ G2 Lu = L∗ G2 f . The preconditioned formulation did quite well. The solid line corresponds to the CGNR applied to the preconditioned problem.
Our Buy this book from SIAM at http://www. with the unpreconditioned iteration being very irregular. and the graphs are monotonic. Neither convergence history is monotonic. GMRES ITERATION 57 and the preconditioned iteration 2. We see a considerable improvement in cost over CGNR and. a better cost/iterate than GMRES in the unpreconditioned case. The preconditioned iteration terminated in six iterations (12 matrixvector products) and required roughly 1.4.4 million. In Fig. since our unpreconditioned matrixvector product is so inexpensive.√3. However. for the ﬁnal iterate.31).Copyright ©1995 by the Society for Industrial and Applied Mathematics.com/SIAM/FR16. the behavior of even the preconditioned iteration can also be poor if the coeﬃcients of the ﬁrst derivative terms are too large.7 million ﬂoatingpoint operations.html. possibly. The approximate residual is given in terms of the quasiresidual norm τm rather than the true residual. This electronic version is for personal use and may not be duplicated or distributed. This code has the same input/output arguments as gmres.5 we plot the convergence histories in terms of the approximate residual 2k + 1τ2k given by (3. The examples for BiCGSTAB use the code bicgstab from the collection of MATLAB codes. 10 0 10 Relative Residual 1 10 2 10 3 10 4 0 50 100 150 200 Iterations 250 300 350 Fig. The unpreconditioned iteration took 40 iterations and 2.8. We plot only full iterations (m = 2k) except. As in the previous examples.ecsecurehost. 3. the solid line corresponds to the preconditioned iteration and the dashed line to the unpreconditioned iteration. CGNE has similar performance. as you are asked to investigate in Exercise 3. 3. . We repeat the experiment with TFQMR as our linear solver. see Exercise 3.9. but not varying by many orders of magnitude as a CGS or BiCG iteration might.9. CGNR for 2D elliptic equation. We use the code tfqmr that is included in our collection.9. We applied BiCGSTAB to both the preconditioned and unpreconditioned forms of (3.3. We plot the iteration histories in Fig.2 million ﬂoatingpoint operations.33) with the same data and termination criterion as for the GMRES example.
4.1 million ﬂoatingpoint operations and 67 iterations for termination. 3. BiCGSTAB for 2D elliptic equation.31) to make termination decisions. TFQMR for 2D elliptic equation.html.com/SIAM/FR16. This led to actual relative residuals of 7 × 10−5 for the unpreconditioned iteration and 2 × 10−4 for the preconditioned iteration.4. termination criterion is based on the quasiresidual and (3. 10 0 10 Relative Approximate Residual 1 10 2 10 3 10 4 10 5 0 10 20 30 40 Iterations 50 60 70 Fig. stopping the iteration when τm m + 1/ b 2 ≤ h−2 . 58 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 1 10 10 0 Relative Residual 10 1 10 2 10 3 10 4 0 5 10 15 20 Iterations 25 30 35 40 Fig. .9 million ﬂoatingpoint operations and 7 iterations.ecsecurehost. The preconditioned iteration was much more eﬃcient. 3.Copyright ©1995 by the Society for Industrial and Applied Mathematics. indicating that it is reasonable to use the quasiresidual estimate (3. This electronic version is for personal use and may not be duplicated or distributed.31) as described in √ § 3.6. taking 1. The unpreconditioned iteration required roughly 4.5. The plateaus in the graph of the convergence history for the Buy this book from SIAM at http://www.
. GMRES ITERATION 59 unpreconditioned iteration are related (see [196]) to spikes in the corresponding graph for a CGS iteration. This electronic version is for personal use and may not be duplicated or distributed.com/SIAM/FR16. Buy this book from SIAM at http://www.Copyright ©1995 by the Society for Industrial and Applied Mathematics.html.ecsecurehost.
Exercises on GMRES 3.5. Consider the centered diﬀerence discretization of −u + u + u = 1.1. This electronic version is for personal use and may not be duplicated or distributed. 200. That is. and TFQMR perform? 3.9. and λ ∈ R such that the GMRES residuals satisfy rk 2 = r0 2 for all k < N . . .3.4.9. Prove that the Arnoldi process (unless it terminates prematurely with a solution) produces a basis for Kk+1 that satisﬁes (3. . Modify the MATLAB GMRES code to allow for restarts. b ∈ RN .5. Let A = Jλ be an elementary Jordan block of size N corresponding to an eigenvalue λ = 0. 0 0 . . 3..9. 0 1 . 0 1 Give an example of x0 . How do your conclusions compare with those in [160]? 3.8). Try this for meshes with 50.9. One way to do this is to call the gmres code with another code which has an outer loop to control the restarting as in Algorithm gmresm. . How would you test for failure in a restarted algorithm? What kinds of failures can occur? Repeat Exercise 6 with a limit of three GMRES iterations before a restart. . . How does the performance of the iteration depend on the mesh? Try other preconditioners such as Gauss–Seidel and Jacobi.10) detects loss of orthogonality. points.7. 3.1 and add two more columns corresponding to classical Gram–Schmidt orthogonalization with reorthogonalization at every step and determine when the test in (3.9. 3. This means that Jλ = λ 0 1 λ .Copyright ©1995 by the Society for Industrial and Applied Mathematics. 3. How do other methods such as CGNR.html.9. λ 0 .9.. Give an example of a matrix that is not diagonalizable. .. BiCGSTAB.ecsecurehost. Prove Theorem 3. Duplicate Table 3.2.4 and Theorem 3.1.9. Solve this problem with GMRES (without preconditioning) and then apply as a preconditioner the map M deﬁned by −(M f ) = f.1.com/SIAM/FR16. u(0) = u(1) = 0. 60 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 3.. precondition with a solver for the highorder term in the diﬀerential equation using the correct boundary conditions. CGNE. (M f )(0) = (M f )(1) = 0. 1 λ 0 λ .6. Buy this book from SIAM at http://www. See [167] for another example. 100.
TFQMR.8. Apply CGNE to the PDE in § 3. Compare the execution times in your computing environment.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Are the solutions to (3.9. Why might GMRES(m) be faster than GMRES? Vary the dimension and the coeﬃcient ay .ecsecurehost. Buy this book from SIAM at http://www.8 equally accurate? Compare the ﬁnal results with the known solution.7.7. Experiment with diﬀerent values of m. Compare timings and solution accuracy with those obtained by left preconditioning. to the equation LGw = f and then recover the solution from u = Gw. Duplicate the results in § 3. and/or CGNE. How does the performance diﬀer from CGNR and GMRES? 3.html. CGNR.9.9.12. apply GMRES. Implement BiCG and CGS and try them on some of the examples in this chapter. Consider preconditioning from the right.com/SIAM/FR16.8.11. GMRES ITERATION 61 3. 3. 3.33) in § 3.9. Try ay = 5y. BiCGSTAB.7 and § 3. For the PDE in § 3. This electronic version is for personal use and may not be duplicated or distributed.10. Why is the performance of the methods sensitive to ay and ax ? 3.9. .9.7 and § 3.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
62
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
Chapter 4
Basic Concepts and FixedPoint Iteration
This part of the book is about numerical methods for the solution of systems of nonlinear equations. Throughout this part, we will let · denote a norm on RN as well as the induced matrix norm. We begin by setting the notation. We seek to solve (4.1) F (x) = 0. Here F : RN → RN . We denote the ith component of F by fi . If the components of F are diﬀerentiable at x ∈ RN we deﬁne the Jacobian matrix F (x) by ∂fi F (x)ij = (x). ∂xj The Jacobian matrix is the vector analog of the derivative. We may express the fundamental theorem of calculus as follows. Theorem 4.0.1. Let F be diﬀerentiable in an open set Ω ⊂ RN and let ∗ ∈ Ω. Then for all x ∈ Ω suﬃciently near x∗ x F (x) − F (x∗ ) = 4.1. Types of convergence Iterative methods can be classiﬁed by the rate of convergence. Definition 4.1.1. Let {xn } ⊂ RN and x∗ ∈ RN . Then • xn → x∗ qquadratically if xn → x∗ and there is K > 0 such that xn+1 − x∗ ≤ K xn − x∗ 2 . • xn → x∗ qsuperlinearly with qorder α > 1 if xn → x∗ and there is K > 0 such that xn+1 − x∗ ≤ K xn − x∗ α . • xn → x∗ qsuperlinearly if
n→∞ 1 0
F (x∗ + t(x − x∗ ))(x − x∗ ) dt.
lim
xn+1 − x∗ = 0. xn − x∗
65
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
66
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
• xn → x∗ qlinearly with qfactor σ ∈ (0, 1) if xn+1 − x∗ ≤ σ xn − x∗ for n suﬃciently large. Definition 4.1.2. An iterative method for computing x∗ is said to be locally (qquadratically, qsuperlinearly, qlinearly, etc.) convergent if the iterates converge to x∗ (qquadratically, qsuperlinearly, qlinearly, etc.) given that the initial data for the iteration is suﬃciently good. Note that a qsuperlinearly convergent sequence is also qlinearly convergent with qfactor σ for any σ > 0. A qquadratically convergent sequence is qsuperlinearly convergent with qorder 2. In general, a qsuperlinearly convergent method is preferable to a qlinearly convergent one if the cost of a single iterate is the same for both methods. We will see that often a method that is more slowly convergent, in terms of the convergence types deﬁned above, can have a cost/iterate so low that the slow iteration is more eﬃcient. Sometimes errors are introduced into the iteration that are independent of the errors in the approximate solution. An example of this would be a problem in which the nonlinear function F is the output of another algorithm that provides error control, such as adjustment of the mesh size in an approximation of a diﬀerential equation. One might only compute F to low accuracy in the initial phases of the iteration to compute a solution and tighten the accuracy as the iteration progresses. The rtype convergence classiﬁcation enables us to describe that idea. Definition 4.1.3. Let {xn } ⊂ RN and x∗ ∈ RN . Then {xn } converges to x∗ r(quadratically, superlinearly, linearly) if there is a sequence {ξn } ⊂ R converging q(quadratically, superlinearly, linearly) to zero such that xn − x∗ ≤ ξn . We say that {xn } converges rsuperlinearly with rorder α > 1 if ξn → 0 qsuperlinearly with qorder α. 4.2. Fixedpoint iteration Many nonlinear equations are naturally formulated as ﬁxedpoint problems (4.2) x = K(x) where K, the ﬁxedpoint map, may be nonlinear. A solution x∗ of (4.2) is called a ﬁxed point of the map K. Such problems are nonlinear analogs of the linear ﬁxedpoint problems considered in Chapter 1. In this section we analyze convergence of ﬁxedpoint iteration, which is given by (4.3) xn+1 = K(xn ).
This iterative method is also called nonlinear Richardson iteration, Picard iteration, or the method of successive substitution.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
and therefore xn − x0 = ≤ n−1 i=0 xi+1 n−1 i=0 − xi n−1 i i=0 γ xi+1 − xi ≤ x1 − x0 ≤ x1 − x0 /(1 − γ). Let Ω ⊂ RN and let G : Ω → RM . Let x0 ∈ Ω. This electronic version is for personal use and may not be duplicated or distributed. .2. ≤ γ i x1 − x0 . ≤ γ n xk − x0 ≤ γ n x1 − x0 /(1 − γ).2. .2.com/SIAM/FR16. Let Ω be a closed subset of RN and let K be a contraction mapping on Ω with Lipschitz constant γ < 1 such that K(x) ∈ Ω for all x ∈ Ω. . then x∗ − y ∗ = K(x∗ ) − K(y ∗ ) ≤ γ x∗ − y ∗ .1. If K has two ﬁxed points x∗ and y ∗ in Ω. Then there is a unique ﬁxed point of K. Definition 4. .2. FIXEDPOINT ITERATION 67 Before discussing convergence of ﬁxedpoint iteration we make two deﬁnitions. and the iteration deﬁned by (4.ecsecurehost.k→∞ and therefore the sequence {xn } is a Cauchy sequence and has a limit x∗ [162].Copyright ©1995 by the Society for Industrial and Applied Mathematics. . Now.html. Note that {xn } ⊂ Ω because x0 ∈ Ω and K(x) ∈ Ω whenever x ∈ Ω. The sequence {xn } remains bounded since for all i ≥ 1 xi+1 − xi = K(xi ) − K(xi−1 ) ≤ γ xi − xi−1 . k ≥ 0.1 and Corollary 1. The standard result for ﬁxedpoint iteration is the Contraction Mapping Theorem [9].1. Hence lim xn+k − xn = 0 n. Theorem 4. x∗ ∈ Ω. Let Ω ⊂ RN . Compare the proof to that of Lemma 1.2. G is Lipschitz continuous on Ω with Lipschitz constant γ if G(x) − G(y) ≤ γ x − y for all x. xn+k − xn = K(xn+k−1 ) − K(xn−1 ) ≤ γ xn+k−1 − xn−1 ≤ γ K(xn+k−2 ) − K(xn−2 ) ≤ γ 2 xn+k−2 − xn−2 ≤ . y ∈ Ω. Buy this book from SIAM at http://www. K : Ω → RN is a contraction mapping on Ω if K is Lipschitz continuous on Ω with Lipschitz constant γ < 1.2.3) converges qlinearly to x∗ with qfactor γ for all initial iterates x0 ∈ Ω. Definition 4. Proof.1. for all n.
we may apply the contraction mapping theorem with Ω = R since for all x. (4. Advancing in time from y n to y n+1 requires solution of the ﬁxedpoint problem y = K(y) = y n + hf (y). say f (y) ≤ M and f (x) − f (y) ≤ M x − y for all x. y(t0 ) = y0 and its solution by the backward Euler method with time step h.3.1 has a solution x∗ . From this we conclude that for h suﬃciently small we can integrate forward in time and that the time step h can be chosen independently of y. For simplicity we assume that f is bounded and Lipschitz continuous. Equation 4.Copyright ©1995 by the Society for Industrial and Applied Mathematics. x∗ = y ∗ . Hence the ﬁxed point is unique. Buy this book from SIAM at http://www. y K(x) − K(y) = hf (x) − f (y) ≤ hM x − y.1. 1. F : Ω → RN ×N is Lipschitz continuous with Lipschitz constant γ. see [145] for a more complete discussion of these algorithms.3. which completes the proof. consider the initial value problem y = f (y).4) y n+1 − y n = hf (y n+1 ). .html. This time step may well be too small to be practical. Nonlinear variations of the classical stationary iterative methods such as Gauss–Seidel have also been used. 4. Finally we note that xn+1 − x∗ = K(xn ) − K(x∗ ) ≤ γ xn − x∗ . 2.com/SIAM/FR16. 68 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Since γ < 1 this can only hold if x∗ − y ∗ = 0. For example. The standard assumptions We will make the standard assumptions on F . In (4. If hM < 1.4) y k denotes the approximation of y(t0 + kh). Assumption 4. One simple application of this theorem is to numerical integration of ordinary diﬀerential equations by implicit methods. e. and the methods based on Newton’s method which we present in the subsequent chapters are used much more often [83].ecsecurehost. i. This type of analysis was used by Picard to prove existence of solutions of initial value problems [152]. y. This electronic version is for personal use and may not be duplicated or distributed.
Hence (4.5) holds if γδ < F (x∗ ) . For all x ∈ B(δ) we have F (x) ≤ F (x∗ ) + γ e . Throughout this part.1 is an important consequence of the standard assumptions. The notation introduced above will be used consistently.Copyright ©1995 by the Society for Industrial and Applied Mathematics.5) and Theorem 4. −1 e /2 ≤ F (x) ≤ 2 F (x∗ ) e . These assumptions can be weakened [108] without sacriﬁcing convergence of the methods we consider here.7. This electronic version is for personal use and may not be duplicated or distributed. This will follow from δ< since then (4. FIXEDPOINT ITERATION 69 3. We use (4.7).1 illustrates one way to weaken the standard assumptions. The next result (4. Let δ be small enough so that B(δ) ⊂ Ω. where e = x − x∗ . . Lemma 4.7) F (x∗ )−1 F (x) ≤ 2 F (x∗ ) . we note that if x ∈ B(δ) then x∗ + te ∈ B(δ) for all 0 ≤ t ≤ 1. Lemma 4. If xn is the nth iterate of a sequence. Exercise 5.8) I − F (x∗ )−1 F (x) = F (x∗ )−1 (F (x∗ ) − F (x)) ≤ γ F (x∗ )−1 e ≤ γδ F (x∗ )−1 < 1/2.3. we will always denote a root of F by x∗ .1 to obtain F (x) ≤ 1 0 F (x∗ + te) e dt ≤ 2 F (x∗ ) e Buy this book from SIAM at http://www. F (x)−1 ≤ 2 F (x∗ )−1 .1. Then there is δ > 0 so that for all x ∈ B(δ) (4.6) and (4. F (x∗ )−1 2γ −1 To prove the ﬁnal inequality (4.ecsecurehost. F (x∗ ) is nonsingular.html.3. Proof.5) (4. en = xn − x∗ is the error in that iterate.6) is a direct consequence of the Banach Lemma if I − F (x∗ )−1 F (x) < 1/2. Assume that the standard assumptions hold.0. However the classical result on quadratic convergence of Newton’s method requires them and we make them throughout. We let B(r) denote the ball of radius r about x∗ B(r) = {x  e < r}.com/SIAM/FR16.
70 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS which is the right inequality in (4.html.com/SIAM/FR16. e /2 ≤ F (x∗ )−1 F (x) ≤ F (x∗ )−1 F (x) . To prove the left inequality in (4.ecsecurehost.Copyright ©1995 by the Society for Industrial and Applied Mathematics.7). .8) F (x∗ )−1 F (x) ≥ e (1 − Therefore 1 0 1 0 1 0 F (x∗ + te)e dt (I − F (x∗ )−1 F (x∗ + te))e dt. by (4.7) note that F (x∗ )−1 F (x) = F (x∗ )−1 =e− and hence. This electronic version is for personal use and may not be duplicated or distributed. Buy this book from SIAM at http://www. which completes the proof. I − F (x∗ )−1 F (x∗ + te)dt ) ≥ e /2.
1) We may also view x+ as the root of the twoterm Taylor expansion or linear model of F about xc Mc (x) = F (xc ) + F (xc )(x − xc ). [137].1. [156].3. Then there are K > 0 and δ > 0 such that if xc ∈ B(δ) the Newton iterate from xc given by (5. The proof of convergence of the complete Newton iteration will be complete if we reduce δ if needed so that Kδ < 1.com/SIAM/FR16. Theorem 5.2) Proof.0.2. Local convergence of Newton’s method We will describe iterative methods for nonlinear equations in terms of the transition from a current iterate xc to a new iterate x+ . By Lemma 4. Let δ be small enough so that the conclusions of Lemma 4.1) satisﬁes e+ ≤ K ec 2 .1.1 e+ = ec − F (xc )−1 F (xc ) = F (xc )−1 1 0 (F (xc ) − F (x∗ + tec ))ec dt.ecsecurehost.3.1.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Chapter 5 Newton’s Method 5.1. 71 Buy this book from SIAM at http://www. Theorem 5. In the context of single systems this method appeared in the 17th century [140].1 and the Lipschitz continuity of F e+ ≤ (2 F (x∗ )−1 )γ ec 2 /2. This completes the proof of (5. The convergence result on Newton’s method follows from Lemma 4. [13].1. . (5. Newton’s method is x+ = xc − F (xc )−1 F (xc ). Let the standard assumptions hold. Then there is δ such that if x0 ∈ B(δ) the Newton iteration xn+1 = xn − F (xn )−1 F (xn ) converges qquadratically to x∗ .1 hold. This electronic version is for personal use and may not be duplicated or distributed. In this language.3. Let the standard assumptions hold.html. (5.2) with K = γ F (x∗ )−1 . By Theorem 4.
[105].1. Moreover. There are.1 twice. This electronic version is for personal use and may not be duplicated or distributed. if there is error in the evaluation of F or the initial iterate is near a root. Since x0 ∈ B(δ) by assumption. In all Buy this book from SIAM at http://www. a termination decision based on the relative residual may be made too late in the iteration or. Two examples are implicit integration of ordinary diﬀerential equations and diﬀerential algebraic equations. where the initial iterate is derived from the solution at the previous time step.1 implies that (5. local convergence results like those in this and the following two chapters describe the terminal phase of the iteration. [126].1 we conclude that if F (x∗ ) is well conditioned. Therefore if xn ∈ B(δ) we may continue the iteration.2.1. Reduce δ if needed so that Kδ = η < 1.1. the entire sequence {xn } ⊂ B(δ).3.1 hold. as we will see in the next section. the size of the relative nonlinear residual is a good indicator of the size of the error.4). where the initial iterate is an interpolation of a solution from a coarser computational mesh [99]. We have.2) and (1.1. as a direct consequence of applying Lemma 4.3) then implies that xn → x∗ qquadratically. [16].1 hold.3.Copyright ©1995 by the Society for Industrial and Applied Mathematics. there is no “righthand side” to use as a scaling factor and we must balance the relative and absolute size of the nonlinear residuals in some other way.1 hold for x ∈ B(δ).ecsecurehost. many situations in which the initial iterate is very near the root. We raised this issue in the context of linear equations in Chapter 1 when we compared (1. Termination of the iteration We can base our termination decision on Lemma 4. where δ is small enough so that the conclusions of Lemma 4. however. From Lemma 5. 5.1. and solution of discretizations of partial diﬀerential equations. If x ∈ B(δ). Theorem 5. then if F (x∗ ) is well conditioned.3.2. a nonlinear form of Lemma 1.1. Then for all x ∈ B(δ) e F (x) 4κ(F (x∗ )) e ≤ ≤ . Let δ > 0 be small enough so that the conclusions of Lemma 4. Let δ be small enough so that the conclusions of Theorem 5.3) en+1 ≤ K en 2 ≤ η en < en and hence xn+1 ∈ B(ηδ) ⊂ B(δ).com/SIAM/FR16. . 72 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Proof. ∗ )) 4 e0 κ(F (x F (x0 ) e0 where κ(F (x∗ )) = F (x∗ ) F (x∗ )−1 is the condition number of F (x∗ ) relative to the norm · .2. Assume that the standard assumptions hold. In the nonlinear case. we may terminate the iteration when the relative nonlinear residual F (x) / F (x0 ) is small. Lemma 5. Then if n ≥ 0 and xn ∈ B(δ). [83]. However. the iteration may not terminate at all.html. when the initial iterate is far from the root and methods such as those discussed in Chapter 8 are needed. (5.3. The assumption that the initial iterate be “suﬃciently near” the solution (x0 ∈ B(δ)) may seem artiﬁcial at ﬁrst look.
If the iteration is qlinearly convergent. where the relative error tolerance τr and absolute error tolerance τa are input to the algorithm. the Jacobian F (xc ) must be computed and factored. If one decides to continue.1. then termination on small steps becomes more attractive.3. If computation and factorization of the Jacobian are inexpensive.Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed. whereas by terminating on small relative residuals or on a condition like (5. [23]. 5. . the issues for sparse Jacobians when direct methods are used to solve linear systems are very similar. This criterion is based on Theorem 5. make use of this in the special case of the chord method. Our examples and discussions of operation counts make the implicit assumption that the Jacobian is dense. The diﬀerential equations codes discussed in [16]. which implies that ec = s + O( ec 2 ). Hence the step is a reliable indicator of the error provided σ is not too near 1. [21]. the relation between the step and the error is less clear. [23]. and [151]. Buy this book from SIAM at http://www.html. then e+ ≤ σ ec implies that (1 − σ) ec ≤ s ≤ (1 + σ) ec . in order to estimate ec this way. However. For methods other than Newton’s method.1.ecsecurehost. one must do most of the work needed to compute x+ . However. and terminate the iteration when s is suﬃciently small. Combinations of relative and absolute error tolerances are commonly used in numerical methods for ordinary diﬀerential equations or diﬀerential algebraic equations [16]. Another way to decide whether to terminate is to look at the Newton step s = −F (xc )−1 F (xc ) = x+ − xc .4) the decision to stop the iteration can be made before computation and factorization of F (xc ). Hence. Implementation of Newton’s method In this section we assume that direct methods will be used to solve the linear equation for the Newton step. NEWTON’S METHOD 73 of our MATLAB codes and algorithms for nonlinear equations our termination criterion is to stop the iteration if (5. [151].4) F (x) ≤ τr F (x0 ) + τa .com/SIAM/FR16. say. near the solution s and ec are essentially the same size. In order to compute the Newton iterate x+ from a current point xc one must ﬁrst evaluate F (xc ) and decide whether to terminate the iteration. [21]. In Chapter 6 we consider algorithms in which iterative methods are used to solve the equation for the Newton step.
there may be little alternative to diﬀerence Jacobians. One approach to reducing the cost of items 2a and item 2b in the Newton iteration is to move them outside of the main loop. r0 = F (x) 2. r0 = F (x) 2. F. . τ ) 1.ecsecurehost. automatic diﬀerentiation (see [94] for a collection of articles on this topic) may provide such an alternative. This method is called the chord method. Buy this book from SIAM at http://www. and a vector of termination tolerances τ = (τr . and directly than with diﬀerences and the analysis above for the cost of a Newton iterate is very pessimistic.7. This electronic version is for personal use and may not be duplicated or distributed. τa ) ∈ R2 . The inputs are the same as for Newton’s method. F.3. Any other appropriate factorization such as QR or Cholesky could be used as well. Of these steps.com/SIAM/FR16. See Exercise 5. newton(x.2. In many cases F can be computed more eﬃciently. x+ = xc − F (x0 )−1 F (xc ). This means that the linear approximation of F (x) = 0 that is solved at each iteration has derivative determined by the initial iterate.html. Evaluation of F by ﬁnite diﬀerences should be expected to cost N times the cost of an evaluation of F because each column in F requires an evaluation of F to form the diﬀerence approximation. The inputs of the algorithm are the initial iterate x. In the future. τ ) 1. chord(x. 74 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Then the step is computed as the solution of F (xc )s = −F (xc ) and the iterate is updated x+ = xc + s.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Algorithm 5.21 for an example. the nonlinear map F . Algorithm 5.1. Do while F (x) > τr r0 + τa (a) Compute F (x) (b) Factor F (x) = LU . Hence the cost of a Newton step may be roughly estimated as N + 1 evaluations of F and O(N 3 ) ﬂoatingpoint operations. Factorization of F in the dense case costs O(N 3 ) ﬂoatingpoint operations. Compute F (x) 3. For general nonlinearities or general purpose codes such as the MATLAB code nsol from the collection.3. For deﬁniteness in the description of Algorithm newton we use an LU factorization of F . (c) Solve LU s = −F (x) (d) x = x + s (e) Evaluate F (x). the evaluation and factorization of F are the most costly. Factor F (x) = LU . accurately.
4. the resulting iteration can return a result that is an O( ) accurate approximation to x∗ . Similar diﬀerences arise if F (xc ) is numerically approximated by diﬀerences.e. and δ1 > 0 such that if xc ∈ B(δ) and ∆(xc ) < δ1 then x+ = xc − (F (xc ) + ∆(xc ))−1 (F (xc ) + (xc )) is deﬁned (i. We will refer to this phase of the iteration in which the nonlinear residual is no longer being reduced as stagnation of the iteration.. δ > 0.4. Let xN = xc − F (xc )−1 F (xc ) + and note that x+ = xN + (F (xc )−1 − (F (xc ) + ∆(xc ))−1 )F (xc ) − (F (xc ) + ∆(xc ))−1 (xc ).1 imply (5. + Lemma 4.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Let δ be small enough so that the conclusions of Lemma 4.5) ¯ e+ ≤ K( ec 2 + ∆(xc ) ec + (xc ) ).1 and Theorem 5.ecsecurehost.1 hold. Let the standard assumptions hold. If. + (F (xc ) + ∆(xc ))−1 Buy this book from SIAM at http://www. This electronic version is for personal use and may not be duplicated or distributed.3.3. These issues are discussed in [201] as well. If ∆ is suﬃciently small.1.1. Do while F (x) > τr r0 + τa (a) Solve LU s = −F (x) (b) x = x + s (c) Evaluate F (x). Proof. We continue in the next section with an analysis of the eﬀects of errors in the function and Jacobian in a general context and then consider the chord method and diﬀerence approximation to F as applications. The diﬀerence in the iteration itself is that an approximation to F (xc ) is used. F (xc ) + ∆(xc ) is nonsingular) and satisﬁes (5.com/SIAM/FR16. there is no reason to expect that F (xn ) will ever be smaller than in general.1 and Theorem 5. Theorem 5. The only diﬀerence in implementation from Newton’s method is that the computation and factorization of the Jacobian are done before the iteration is begun. 5. Errors in the function and derivative Suppose that F and F are computed inaccurately so that F + and F + ∆ are used instead of F and F in the iteration. .1. Then there are ¯ K > 0. is entirely due to ﬂoatingpoint roundoﬀ.html. This is diﬀerent from convergence and was called “local improvement” in [65]. NEWTON’S METHOD 75 4.6) e+ ≤ K ec 2 + 2 F (xc )−1 − (F (xc ) + ∆(xc ))−1 ) F (x∗ ) ec (xc ) . for example.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
76 If
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
∆(xc ) ≤ F (x∗ )−1 ∆(xc ) ≤ F (xc )−1
−1
/4
then Lemma 4.3.1 implies that
−1
/2
and the Banach Lemma implies that F (xc ) + ∆(xc ) is nonsingular and (F (xc ) + ∆(xc ))−1 ≤ 2 F (xc )−1 ≤ 4 F (x∗ )−1 . Hence F (xc )−1 − (F (xc ) + ∆(xc ))−1 ≤ 8 F (x∗ )−1 We use these estimates and (5.6) to obtain e+ ≤ K ec Setting
2 2
∆(xc ) .
+ 16 F (x∗ )−1
2
F (x∗ ) ∆(xc ) ec + 4 F (x∗ )−1
2
(xc ) .
¯ K = K + 16 F (x∗ )−1
F (x∗ ) + 4 F (x∗ )−1
completes the proof. The remainder of this section is devoted to applications of Theorem 5.4.1. 5.4.1. The chord method. Recall that the chord method is given by
x+ = xc − F (x0 )−1 F (xc ). In the language of Theorem 5.4.1 (xc ) = 0, ∆(xc ) = F (x0 ) − F (xc ). Hence, if xc , x0 ∈ B(δ) ⊂ Ω (5.7) ∆(xc ) ≤ γ x0 − xc ≤ γ( e0 + ec ).
We apply Theorem 5.4.1 to obtain the following result. Theorem 5.4.2. Let the standard assumptions hold. Then there are KC > 0 and δ > 0 such that if x0 ∈ B(δ) the chord iterates converge qlinearly to x∗ and (5.8) en+1 ≤ KC e0 en . Proof. Let δ be small enough so that B(δ) ⊂ Ω and the conclusions of Theorem 5.4.1 hold. Assume that xn ∈ B(δ). Combining (5.7) and (5.5) implies ¯ ¯ en+1 ≤ K( en (1 + γ) + γ e0 ) en ≤ K(1 + 2γ)δ en . Hence if δ is small enough so that ¯ K(1 + 2γ)δ = η < 1
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
NEWTON’S METHOD
77
then the chord iterates converge qlinearly to x∗ . Qlinear convergence implies ¯ that en < e0 and hence (5.8) holds with KC = K(1 + 2γ). Another variation of the chord method is x+ = xc − A−1 F (xc ), where A ≈ F (x∗ ). Methods of this type may be viewed as preconditioned nonlinear Richardson iteration. Since ∆(xc ) = A − F (xc ) ≤ A − F (x∗ ) + F (x∗ ) − F (xc ) , if xc ∈ B(δ) ⊂ Ω then ∆(xc ) ≤ A − F (x∗ ) + γ ec ≤ A − F (x∗ ) + γδ. Theorem 5.4.3. Let the standard assumptions hold. Then there are KA > 0, δ > 0, and δ1 > 0, such that if x0 ∈ B(δ) and A − F (x∗ ) < δ1 then the iteration xn+1 = xn − A−1 F (xn ) converges qlinearly to x∗ and (5.9) en+1 ≤ KA ( e0 + A − F (x∗ ) ) en .
5.4.2. Approximate inversion of F . Another way to implement chordtype methods is to provide an approximate inverse of F . Here we replace F (x)−1 by B(x), where the action of B on a vector is less expensive to compute than a solution using the LU factorization. Rather than express the iteration in terms of B −1 (x) − F (x) and using Theorem 5.4.3 one can proceed directly from the deﬁnition of approximate inverse. Note that if B is constant (independent of x) then the iteration x+ = xc − BF (xc ) can be viewed as a preconditioned nonlinear Richardson iteration. We have the following result. Theorem 5.4.4. Let the standard assumptions hold. Then there are KB > 0, δ > 0, and δ1 > 0, such that if x0 ∈ B(δ) and the matrixvalued function B(x) satisﬁes (5.10) I − B(x)F (x∗ ) = ρ(x) < δ1
for all x ∈ B(δ) then the iteration xn+1 = xn − B(xn )F (xn ) converges qlinearly to x∗ and (5.11) en+1 ≤ KB (ρ(xn ) + en ) en .
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
78
ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS
Proof. On one hand, this theorem could be regarded as a corollary of the Banach Lemma and Theorem 5.4.1. We give a direct proof. First, by (5.10) we have (5.12) B(x) = B(x)F (x∗ )F (x∗ )−1 ≤ B(x)F (x∗ ) F (x∗ )−1 ≤ MB = (1 + δ1 ) F (x∗ )−1 .
Using (5.12) and e+ = ec − B(xc )F (xc ) =
1 0
(I − B(xc )F (x∗ + tec ))ec dt
1 0
= (I − B(xc )F (x∗ ))ec + B(xc ) we have
(F (x∗ ) − F (x∗ + tec ))ec dt
e+ ≤ ρ(xc ) ec + MB γ ec 2 /2.
This completes the proof with KB = 1 + MB γ/2. 5.4.3. The Shamanskii method. Alternation of a Newton step with a sequence of chord steps leads to a class of highorder methods, that is, methods that converge qsuperlinearly with qorder larger than 2. We follow [18] and name this method for Shamanskii [174], who considered the ﬁnite diﬀerence Jacobian formulation of the method from the point of view of eﬃciency. The method itself was analyzed in [190]. Other highorder methods, especially for problems in one dimension, are described in [190] and [146]. We can describe the transition from xc to x+ by y1 (5.13) = xc − F (xc )−1 F (xc ),
yj+1 = yj − F (xc )−1 F (yj ) for 1 ≤ j ≤ m − 1, x+ = ym .
Note that m = 1 is Newton’s method and m = ∞ is the chord method with {yj } playing the role of the chord iterates. Algorithmically a second loop surrounds the factor/solve step. The inputs for the Shamanskii method are the same as for Newton’s method or the chord method except for the addition of the parameter m. Note that we can overwrite xc with the sequence of yj ’s. Note that we apply the termination test after computation of each yj . Algorithm 5.4.1. sham(x, F, τ, m) 1. r0 = F (x) 2. Do while F (x) > τr r0 + τa
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Optimal choices of m as a function of N can be derived under assumptions consistent with dense matrix algebra [18] by balancing the O(N 3 ) cost of a matrix factorization with the cost of function and Jacobian evaluation. if the initial iterate is suﬃciently accurate. when building an approximate Jacobian. m i. Let the standard assumptions hold and let m ≥ 1 be given. for example. Diﬀerence approximation to F .com/SIAM/FR16. if we set y1 = xn . (5. Assume that we compute F (x) + (x) instead of F (x) and attempt to approximate the action of F on a vector by a forward diﬀerence. Evaluate F (x).4. This electronic version is for personal use and may not be duplicated or distributed. . Algorithm nsol is based on this idea and using an idea from [114] computes and factors the Jacobian only until an estimate of the qfactor for the linear convergence rate is suﬃciently low. .Copyright ©1995 by the Society for Industrial and Applied Mathematics. The analysis in [18] and the expectation that. Then there are KS > 0 and δ > 0 such that if x0 ∈ B(δ) the Shamanskii iterates converge qsuperlinearly to x∗ with qorder m + 1 and (5.4.4.2. 5. A forward diﬀerence approximation to F (x)w would be F (x + hw) + (x + hw) − F (x) − (x) . Then if xn ∈ B(δ) all the intermediate iterates {yj } are in B(δ) by Theorem 5. Theorem 5. indicate that the chord method is usually the best option for very large problems. The advantage of the Shamanskii method over Newton’s method is that high qorders can be obtained with far fewer Jacobian evaluations or factorizations. We would do this.14) en+1 ≤ KS en m+1 . ≤ KC en j+1 .4. iv. (c) for j = 1.ecsecurehost. Here the jth column of F (x) is F (x)ej where ej is the unit vector with jth component 1 and other components 0. .html.4.4.5. Let δ be small enough so that B(δ) ⊂ Ω and that the conclusions of Theorem 5. Proof.2. . If F (x) ≤ τr r0 + τa exit.2 hold. . The convergence result is a simple application of Theorem 5. x = x + s iii. h Buy this book from SIAM at http://www. Hence xn+1 ∈ B(δ). In fact. Setting j = m in the above inequality completes the proof. NEWTON’S METHOD 79 (a) Compute F (x) (b) Factor F (x) = LU . only a small number of iterations will be taken. Solve LU s = −F (x) ii.8) implies that for 1≤j≤m yj −x∗ ≤ KC xn −x∗ yj−1 −x∗ = KC en j yj−1 −x∗ ≤ . .
Let F be deﬁned in a neighborhood of x ∈ RN and let Buy this book from SIAM at http://www. the cost of a forward diﬀerence approximation of F (x)w is an additional evaluation of F (at the point x+hw). The forward diﬀerence approximation to the directional derivative is not a linear operator in w. h F (x)w − √ The quantity inside the Oterm is minimized when h = ¯. and so linearity has been lost. (∇h F )(x) is the N × N matrix whose jth column is given by F (x + h x ej ) − F (x) h x F (he ) − F (x) j x=0 x=0 (∇h F )(x)j = h We make a similar deﬁnition of the diﬀerence approximation of the directional derivative. .4. If this is not the case. If F (x) has already been computed. Definition 5. The choice h ≈ 10−15 in this case can lead to disaster (try it!).4. A more important consequence of this analysis is that the choice of step size in a forward diﬀerence approximation to the action of the Jacobian on a vector must take into account the error in the evaluation of F .html. Definition 5. h should be scaled to reﬂect that.1 below speciﬁes the usual choice. In the special case where (x) is a result of ﬂoatingpoint roundoﬀ in full precision (¯ ≈ 10−15 ). The reason for this is that the derivative has been approximated. one for each column of the Jacobian. Let F be deﬁned in a neighborhood of x ∈ RN . Because of this we must carefully specify what we mean by a diﬀerence approximation to the Jacobian. This means.ecsecurehost.1. Then F (x + hw) + (x + hw) − F (x) − (x) = O(h + ¯/h). This can become very important if part of F is computed from measured data. Deﬁnition 5.2. that very small diﬀerence steps h can lead to inaccurate results. Exploitation of special structure could reduce this cost. the analysis here indicates that h ≈ 10−7 is a reasonable choice. for instance. Here we have made the implicit assumption that x and w are of roughly the same size.4.com/SIAM/FR16. This electronic version is for personal use and may not be duplicated or distributed. Hence the evaluation of a full ﬁnite diﬀerence Jacobian would cost N function evaluations. The choice h = ¯1/2 x / w reﬂects all the discussion above.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 80 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Assume that (x) ≤ ¯.
6 says that en+1 = O(¯). h x = 0.15)Dh F (x : w) = 0. If we assume that the forward diﬀerence step has been selected so that √ √ h = O( ¯). w = 0. ¯. then ∇h (F + )(x) is nonsingular for all x ∈ B(δ) and there is MF > 0 such that F (x) − ∇h (F + )(x) ≤ MF (h + ¯/h). Then there are δ. w = 0.4. and there is M− > 0 such that √ hc > M− ¯ then satisﬁes e+ ≤ KD (¯ + ( ec + hc ) ec ).4.ecsecurehost. there is no guarantee that (xn ) → 0 as xn → x∗ . [42]. Hence. and KD such that if xc ∈ B(δ). Therefore in the early phases of the iteration. F (x + h x w/ w ) − F (x) . w. h > ¯ 0 such that if x ∈ B(δ).1 implies the following result. and (x) ≤ ¯ for all x ∈ B(δ).Copyright ©1995 by the Society for Industrial and Applied Mathematics.com/SIAM/FR16. (x) ≤ ¯ for all x ∈ B(δ). If information on the sparsity pattern of the Jacobian is available it is sometimes possible [47]. one should expect the sequence { en } to stagnate and cease to decrease once√ en ≈ ¯.1. while en >> ¯.1.html. to evaluate ∇h F with far fewer than N evaluations of F . When en ≤ ¯.6. This electronic version is for personal use and may not be duplicated or distributed. x+ = xc − ∇hc (F + )(xc )−1 (F (xc ) + (xc )) Buy this book from SIAM at http://www. NEWTON’S METHOD 81 w ∈ RN .6 does not imply that an iteration will converge to Even if is entirely due to ﬂoatingpoint roundoﬀ. x = 0.4. the progress of the iteration will be indistinguishable from Newton’s method as implemented with an exact √ Jacobian. Let the standard assumptions hold.4. ¯ Lemma 5. Note that Theorem 5. . Theorem 5. w w The standard assumptions and the Banach Lemma imply the following result. A diﬀerence approximation to a full Jacobian does not use any special structure of the Jacobian and will probably be less eﬃcient than a hand coded analytic Jacobian. h ≤ h.4. Hence one will see quadratic convergence until the error in F admits no further reduction. h x F (hw/ w ) − F (x) . We have (5. Let the standard assumptions hold. Theorem 5. Then there are positive δ.4. then ∆(x) = O( ¯) by Lemma 5. ¯. Theorem 5. If the sparsity pattern is such that F can be reduced to x∗ .
. Moreover. . the error in the diﬀerence approximation will be large. they should be used with caution. We will prove that the secant method is locally qsuperlinearly convergent if the standard assumptions hold.4. the diﬀerence between them may be signiﬁcant and a diﬀerence approximation to a directional derivative. This electronic version is for personal use and may not be duplicated or distributed. . an = xn − xn−1 The resulting method is called the secant method and is due to Newton [140].4. 5. . i. (x) = 0. a linear function of w and it is usually the case that (∇h F (x))w = Dh F (x : w).4. xc − x− We have the following theorem. There is no reason to use a diﬀerence approximation to f for equations in a single variable because one can use previously computed data to approximate f (xn ) by f (xn ) − f (xn−1 ) . x−1 ∈ B(δ) and x0 = x−1 then the secant iterates converge qsuperlinearly to x∗ . the entire solution process can take advantage of this structure [59]. The standard assumptions. If we set s = xn − xn−1 we have by (5. one needs two approximations to x∗ . which uses the size of w. Assume that x−1 . [137]. The secant method. . While we used full matrix approximations to numerical Jacobians in all the examples reported here. We have (xc ) = 0 and (5. Then there is δ > 0 such that if x0 . Let xc be the current iterate and x− the previous iterate. in general.16) and the fundamental theorem of calculus ∆(xn ) = 1 0 (f (xn−1 + ts) − f (xn )) dt Buy this book from SIAM at http://www. We assume for the sake of simplicity that there are no errors in the evaluation of f . x0 and x−1 . Let δ be small enough so that B(δ) ⊂ Ω and the conclusions of Theorem 5. Let the standard assumptions hold with N = 1. . The results in this section are for single equations f (x) = 0.Copyright ©1995 by the Society for Industrial and Applied Mathematics.5.4. In order to start.e. Theorem 5. where f is a realvalued function of one real variable. then. Otherwise xn = xn−1 . If xn = x∗ for some ﬁnite n. 82 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS a block triangular form. is likely to be more accurate than ∇h F (x)w.com/SIAM/FR16. imply that f (x∗ ) = 0. x0 . Dh F (x : w) is not.7.ecsecurehost.16) ∆(xc ) = f (xc ) − f (x− ) − f (xc ). For problems in which the Lipschitz constant of F is large. The local convergence theory can be easily described by Theorem 5. too. Proof.html. we are done. xn ∈ B(δ). If F (x∗ ) is ill conditioned or the Lipschitz constant of F is large.1 hold.1.
r− = βγ We will let B0 be the closed ball of radius r− about x0 B0 = {x  x − x0 ≤ r− }. Assumption 5.com/SIAM/FR16.5. This result is of use. then there is a root and the Newton iteration converges rquadratically to that root. [57]. 5.4. This completes the proof.5. ¯ K(1 + γ)δ = η < 1 If we reduce δ so that then xn → x∗ qlinearly. F (x0 )−1 ≤ β. in proving that discretizations of nonlinear diﬀerential and integral equations have solutions that are near to that of the continuous problem.ecsecurehost.Copyright ©1995 by the Society for Industrial and Applied Mathematics. and [58] for a complete proof and for discussions of related results. we give a simpliﬁed version of a Kantorovichlike result for the chord Buy this book from SIAM at http://www. F is Lipschitz continuous with Lipschitz constant γ in a ball of radius r ≥ r− about x0 where ¯ √ 1 − 1 − 2βηγ . r. and γ with βηγ ≤ 1/2 and x0 ∈ RN ¯ such that 1. F is diﬀerentiable at x0 . There are β. This electronic version is for personal use and may not be duplicated or distributed. However. Therefore en+1  ¯ ≤ K((1 + γ/2)en  + γen−1 /2) → 0 en  as n → ∞. η. . The Kantorovich Theorem In this section we present the result of Kantorovich [107].7. 2 and hence ∆(xn ) ≤ γxn−1 − xn /2 ≤ Applying Theorem 5. We do not prove the result in its full generality and refer to [106].1. for example. [145]. [106]. and F (x0 )−1 F (x0 ) ≤ η.17) ¯ en+1  ≤ K((1 + γ/2)en 2 + γen en−1 /2). We require the following assumption. 2.1 implies (5. which states that if the standard assumptions “almost” hold at a point x0 .html. NEWTON’S METHOD 83 γ(en  + en−1 ) .8) method are the basis for the popular method of Brent [17] for the solution of single equations. The secant method and the classical bisection (see Exercise 5.
1 we have for all x ∈ S F (x0 )−1 F (x) = F (x0 )−1 F (x0 ) +F (x0 )−1 (5. Theorem 5. Let Assumption 5. Moreover x∗ is the unique root of F in the ball of radius √ 1 + 1 − 2βηγ ¯ min r. Theorem 5. We will show that the map φ(x) = x − F (x0 )−1 F (x) is a contraction on the set S = {x  x − x0 ≤ r− }. This electronic version is for personal use and may not be duplicated or distributed.5.5. and xn ∈ B0 for all n. Buy this book from SIAM at http://www.1 but has a simple proof that uses only the contraction mapping theorem and the fundamental theorem of calculus and obtains qlinear convergence instead of rlinear convergence. Then there is a unique root x∗ of F in B0 .5.1.19) 1 0 F (x0 + t(x − x0 ))(x − x0 ) dt = F (x0 )−1 F (x0 ) + x − x0 +F (x0 )−1 1 0 (F (x0 + t(x − x0 )) − F (x0 ))(x − x0 ) dt and φ(x) − x0 = −F (x0 )−1 F (x0 ) −F (x0 )−1 Hence. βγ about x0 .18) βγη < 1/2. . Proof.1 hold and let (5.1 assumes a bit more than Assumption 5.com/SIAM/FR16. the chord iteration with x0 as the initial iterate converges to x∗ qlinearly with qfactor βγr− . Unlike the chord method results in [106]. and [57].html. By the fundamental theorem of calculus and Assumption 5. 2 φ(x) − x0 ≤ η + βγr− /2 = r− .5.Copyright ©1995 by the Society for Industrial and Applied Mathematics.5. 1 0 (F (x0 + t(x − x0 )) − F (x0 ))(x − x0 ) dt. 84 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS method to show how existence of a solution can be derived from estimates on the function and Jacobian.ecsecurehost. [145].
Theorem 5. Clearly x∗ is a root of F because it is a ﬁxed point of φ. We state the theorem in the standard way [145]. Then there is a unique root x∗ of F in B0 . This completes the proof.2. y ∈ B0 .19) F (x0 )−1 F (x) ≥ r − η − βγr2 /2 > 0 because r− < r < r+ . and xn ∈ B0 for all n. This electronic version is for personal use and may not be duplicated or distributed. the contraction mapping theorem (Theorem 4. We know that βγr− < 1 by our assumption that βηγ < 1/2. ≤ 2n βγ n Buy this book from SIAM at http://www. φ(x) − φ(y) = φ (y + t(x − y))(x − y) dt ≤ βγr− x − y . NEWTON’S METHOD 85 and φ maps S into itself.html.2. Hence we may assume that r > r− . for all x. If r = r− . Note that for all x ∈ B0 φ (x) = I − F (x0 )−1 F (x) = F (x0 )−1 (F (x0 ) − F (x)) and hence φ (x) ≤ βγ x − x0 ≤ βγr− < 1. 1 0 Therefore. To prove the remainder of the uniqueness assertion let √ 1 + 1 − 2βηγ r+ = βγ ¯ and note that r+ > r− because βηγ < 1/2.1 and has a very diﬀerent proof. x∗ is the unique root of F in the ball of radius √ 1 + 1 − 2βηγ ¯ min r.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The Kantorovich theorem is more precise than Theorem 5. Let Assumption 5. [63]. r+ ) r then F (x) = 0.1) will imply that there is a unique ﬁxed point x∗ of φ in S and that xn → x∗ qlinearly with qfactor βγr− . This proves the convergence result and the uniqueness of x∗ in B0 .1 hold. using the rquadratic estimates from [106]. [106].5. βγ about x0 and the errors satisfy (5.5. We must show ¯ that if x is such that r− < x − x0 < min(¯. then we are done by the uniqueness of x∗ in B0 .20) en (2βηγ)2 . the Newton iteration with x0 as the initial iterate converges to x∗ .com/SIAM/FR16. . y ∈ S. If we do this.5. To show that φ is a contraction on S and to prove the qlinear convergence we will show that φ(x) − φ(y) ≤ βγr− x − y for all x. Letting r = x − x0 we have by (5.ecsecurehost.
16). This electronic version is for personal use and may not be duplicated or distributed. however.7. the chord method. 86 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS If βηγ < 1/2 then (5. r+ = F (x) .com/SIAM/FR16.html. Algorithm 5.1.25. In addition to the inputs of Algorithm sham. Evaluate F (x) v. 5. m. One would still want to minimize the number of Jacobian computations (by the methods of [47] or [42]. rc = r+ vi. This indicates that the local convergence analysis in this Buy this book from SIAM at http://www. σ = r+ /rc . the Shamanskii method. x = x + s iv. and an error ﬂag. or a combination of all of them based on the reduction in the nonlinear residual. The output of the MATLAB implementation of nsol is the solution.7. the ideas in the implementation of nsol also apply to situations in which the Jacobian is sparse and direct methods for solving the equation for the Newton step are necessary. nsol uses diffjac to approximate the Jacobian by ∇h F with h = 10−7 .6. F. we compute and factor F (x+ ) for use in the subsequent chord steps. the Jacobian is not recomputed.6. Solve LU s = −F (x) iii. Do while F (x) > τr r0 + τa (a) Compute ∇h F (x) ≈ F (x) (b) Factor ∇h F (x) = LU . Examples for Newton’s method In the collection of MATLAB codes we provide an implementation nsol of Newton’s method. This latter option decides if the Jacobian should be recomputed based on the ratio of successive residuals. (c) is = 0.ecsecurehost. ρ) 1. which is not easy. If F (x) ≤ τr r0 + τa exit. If F (x+ ) / F (xc ) is below a given threshold. While our examples have dense Jacobians. say) and sparse matrix factorizations. The Jacobian is recomputed and factored if either the ratio of successive residuals exceeds the threshold 0 < ρ < 1 or the number of iterations without an update exceeds m. In Exercise 5. rc = r0 = F (x) 2. τ. .20) implies that the convergence is rquadratic (see Exercise 5. In the MATLAB code the iteration is terminated with an error condition if σ ≥ 1 at any point. the history of the iteration stored as the vector of l∞ norms of the nonlinear residuals. If the ratio is too large. nsol(x.Copyright ©1995 by the Society for Industrial and Applied Mathematics. σ = 0 (d) Do While is < m and σ ≤ ρ i. the reader is asked to create a sparse form of nsol. is = is + 1 ii. a threshold for the ratio of successive residuals is also input.
The standard assumptions hold near either solution [110] for c ∈ [0. c 2 1 µH(ν) dν 0 −1 µ+ν = 0. Here we approximate integrals on [0. Newton’s method. We will discretize the equation with the composite midpoint rule.9.5. with the 0 function or the constant function with value 1 as initial iterate. (5. The resulting discrete problem is (5. Because H has a singularity at µ = 0.22) have two solutions for c ∈ (0. The Chandrasekhar Hequation. 1). In Exercise 5. With these default parameters the decision to update the Jacobian is made based entirely on the reduction in the norm of the nonlinear residual.7. where µi = (i − 1/2)/N for 1 ≤ i ≤ N . F (xn ) ∞ / F (x0 ) ∞ .21) F (H)(µ) = H(µ) − 1 − The Chandrasekhar Hequation. NEWTON’S METHOD 87 section is not applicable.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The parameters m and ρ are assigned the default values of 1000 and . is used to solve exit distribution problems in radiative transfer.html. nsol is based on the assumption that a diﬀerence Jacobian will be used. [41]. We used the function identically one as initial iterate. however.com/SIAM/FR16. In Chapter 8 we discuss how to recover from this. In the computations reported here we set N = 100 and c = .ecsecurehost. and the ratio Rn = F (xn ) ∞ / F (xn−1 ) ∞ Buy this book from SIAM at http://www. We set the termination parameters τr = τa = 10−6 .21) and the discrete analog (5.1 we tabulate the iteration counter n. This electronic version is for personal use and may not be duplicated or distributed. In Table 5. Only one of these solutions has physical meaning. . We begin with a comparison of Newton’s method and the chord method. The discrete problem has its own physical meaning [41] and we will attempt to solve it to high accuracy. [30]. the default is 40. We present some iteration statistics in both tabular and graphical form. will ﬁnd the physically meaningful solution [110]. Many times the Jacobian can be computed eﬃciently by reusing results that are already available from the function evaluation. It is known [132] that both (5.21 you are asked to modify the code to accept an analytic Jacobian. We computed all Jacobians with the MATLAB code diffjac. the solution of the discrete problem is not even a ﬁrstorder accurate approximation to that of the continuous problem. 1] by 1 0 f (µ) dµ ≈ 1 N N j=1 f (µj ). 1).22) F (x)i = xi − 1 − c 2N µi xj µi + µj j=1 N −1 . nsol allows the user to specify a maximum number of iterations.
In the context of this particular example. the cost of a single evaluation and factorization of the Jacobian is far more than the cost of the entire remainder of the chord iteration.965e04 6.698e03 7.com/SIAM/FR16. The good performance of the chord method is no accident. See Exercise 5. In Fig. The Shamanskii computation required under 6 million ﬂoatingpoint operations. In Exercise 5. then the chord method should be much more eﬃcient than Newton’s method if the equations for the steps can be solved cheaply. In Fig.136e01 2. The linear convergence of the chord method can also be seen in the convergence of the ratios Rn to a constant value.353e05 2. The ideas in Chapters 6 and 7 show how the low iteration cost of the chord method can be combined with the small number of iterations of a superlinearly convergent nonlinear iterative method.511e03 1.3 million. In the case considered here. the cost of the Jacobian computation can be reduced by analytic evaluation and reuse of data that have already been computed for the evaluation of F .000e+00 1. 5.15 for a more careful examination of this concavity.136e01 for n ≥ 1 for both Newton’s method and the chord method.823e02 2. for this problem.18 you are asked to obtain timings in your environment and will see how the ﬂoatingpoint operation counts are related to the actual timings.077e01 2. In the remaining examples.480e01 2. 88 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Table 5. The concave iteration track is the signature of superlinear convergence.118e01 2.1 the solid curve is a plot of F (xn ) ∞ / F (x0 ) ∞ for Newton’s method and the dashed curve a plot of F (xn ) ∞ / F (x0 ) ∞ for the chord method. .8 for Newton’s method.891e06 ∞ Rn 1.html. The Newton iteration required over 8.1 Comparison of Newton and chord iteration. the simple chord method is the most eﬃcient.21.480e01 1.136e01 2.2 we compare the Shamanskii method with m = 2 with Newton’s method. The reader is encouraged to do this in Exercise 5.865e04 F (xn ) ∞ / F (x0 ) 1.136e01 2.388e03 2. 5. the linear track indicates linear convergence.480e01 3. For this example the MATLAB flops command returns an fairly accurate indicator of the cost.ecsecurehost.729e07 ∞ Rn 1.480e01 2. If we think of the chord method as a preconditioned nonlinear Richardson iteration with the initial Jacobian playing the role of preconditioner.7. but do not tabulate. This electronic version is for personal use and may not be duplicated or distributed.000e+00 1. as compared with over 8.132e01 2. n 0 1 2 3 4 5 6 7 8 F (xn ) ∞ / F (x0 ) 1. However.074e02 6. we plot.7.7. Buy this book from SIAM at http://www. the iteration statistics.8 million ﬂoatingpoint operations while the chord iteration required under 3.Copyright ©1995 by the Society for Industrial and Applied Mathematics.334e05 1.
This electronic version is for personal use and may not be duplicated or distributed.1. While Newton’s method converges in 7 iterations and requires 21 million ﬂoatingpoint operations. Newton and chord methods for Hequation. c = . NEWTON’S METHOD 0 89 10 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 10 6 10 7 0 1 2 3 4 Iterations 5 6 7 8 Fig.9999. 10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 10 6 10 7 0 0.Copyright ©1995 by the Society for Industrial and Applied Mathematics. all iteration parameters are the same as in the previous example.html. Aside from c.96.6 million ﬂoatingpoint operations.9.2. The qfactor for the chord method in this computation was over . For this problem the Jacobian at the solution is nearly singular [110].ecsecurehost.com/SIAM/FR16. 5. .9. indicating very slow convergence.5 1 1. 5.5 2 Iterations 2. We now reconsider the Hequation with c = . the chord method requires 188 iterations and 10.5 4 Fig. The hybrid method in nsol with the default parameters Buy this book from SIAM at http://www.5 3 3. Newton and Shamanskii method for Hequation. The hybrid approach in nsol with the default parameters would be the chord method for this problem. c = .
however. 10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 10 6 0 2 4 6 8 Iterations 10 12 14 Fig.9999.3 we compare the Newton iteration with the hybrid. Buy this book from SIAM at http://www. This electronic version is for personal use and may not be duplicated or distributed. 90 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS computes the Jacobian four times during the course of the iteration. plots like those in this section are common. Newton and hybrid methods for Hequation c = . In Fig.ecsecurehost. 5. support this.3. converges in 14 iterations. . 5. in general. and requires 12 million ﬂoatingpoint operations. While the plots indicate qlinear convergence of the nonlinear residuals. One should approach plots of convergence histories with some caution.Copyright ©1995 by the Society for Industrial and Applied Mathematics. This implies only that the nonlinear residual norms { F (xn ) } converge rlinearly to zero. The theory for the chord method asserts that the error norms converge qlinearly to zero.html.com/SIAM/FR16. In practice. the theory does not.
4. When is it locally qquadratically convergent? 5. NEWTON’S METHOD 91 5. 5. 5. A better alternative in the MATLAB environment is to use the semilogy command to plot the norms.1.6. .4.7. Can the performance of the Newton iteration be improved by a linear change of variables? That is. we terminate the iteration.3. can the Newton iterates for F (x) = 0 and AF (Bx) = 0 show any performance diﬀerence when started at the same initial iterate? What about the chord method? 5. Show that if the Lipschitz G(x) − G(y) ≤ K x − y continuity condition on F in the standard assumptions is replaced by H¨lder continuity with exponent α > 0 that the Newton iterates converge o with qorder 1 + α. for nonsingular N × N matrices A and B. xk ) that r l contain a root of a function of one variable. Prove that the secant method converges rsuperlinearly with rorder √ ( 5 + 1)/2.7. the norm of F . Exercises on Newton’s method In some of the exercises here. x0 ) such that f (x0 )f (x0 ) < 0. y ∈ Ω.2.7. 5. When one does this. 5. A function G is said to be H¨lder continuous with exponent α in Ω if o α for all x. This electronic version is for personal use and may not be duplicated or distributed. If f (y) = 0. and the ratio of F (xn ) / F (xn−1 ) for n ≥ 1.7.7.8.7. Given the standard assumptions. Given (xk . one can visually inspect the plot to determine superlinear convergence (concavity) without explicitly computing the ratios.html. r l Show that the sequence xk = (xk + xk )/2 is rlinearly convergent if f is r l continuous and there exists (x0 . Suppose F (xn ) is replaced by ∇hn F (xn ) in the Shamanskii method. In what way is the convergence analysis of the secant method changed if there are errors in the evaluation of f ? 5. xk ) with r l f (xk )f (xk ) < 0 we replace one of the endpoints with y = (xk + xk )/2 r r l l to force f (xk+1 )f (xk+1 ) < 0. This is easy. you will be asked to plot or tabulate iteration statistics.Copyright ©1995 by the Society for Industrial and Applied Mathematics.7. Discuss how hn must be decreased as the iteration progresses to preserve the qorder of m + 1 [174]. Show that the secant method converges qsuperlinearly with qorder √ ( 5 + 1)/2.7. and in the rest of this part of the book.com/SIAM/FR16.ecsecurehost.1.5. prove that the iteration given by xn+1 = xn − (∇hn F (xn ))−1 F (xn ) is locally qsuperlinearly convergent if hn → 0. for each iteration tabulate the iteration counter. When asked to do this.7. Use the l∞ norm. 5.7. r r l l Buy this book from SIAM at http://www. Let ∇h F be given by Deﬁnition 5. The bisection method produces a sequence of intervals (xk .
15. and explain your results. x0 = 10 5.9. but not equal to x∗ [171]. Consider the iterative method that deﬁnes x+ as a root of mc . tabulate or plot iteration statistics.10.12. x0 = 1 3.html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. What problems might arise in such a method? Resolve these problems and implement your algorithm. show that log( en ) is a concave function of n in the sense that log( en+1 ) − log( en ) Buy this book from SIAM at http://www.7. . x0 is suﬃciently near x∗ .7. 5. For scalar functions of one variable a quadratic model of f about a point xc is mc (x) = f (xc ) + f (xc )(x − xc ) + f (xc )(x − xc )2 /2.7.99x0 . If xn → x∗ qsuperlinearly with qorder α > 1.7.7. and f (x0 )f (x0 ) > 0. use x−1 = .13. 1. How would you estimate the qorder of a qsuperlinearly convergent sequence? Rigorously justify your answer. Test it on the problems in Exercise 5. In what cases is this algorithm superior to Newton’s method? For hints and generalizations of this idea to functions of several variables. then the Newton iteration converges monotonically to x∗ . 5. Show that if f is a real function of one real variable. and f (x∗ ) = f (x∗ ) = 0 but f (x∗ ) = 0 then the iteration xn+1 = xn − 2f (xn )/f (xn ) converges locally qquadratically to x∗ provided x0 is suﬃciently near to x∗ . the standard assumptions hold.7. see [170]. . f is Lipschitz continuous. and the secant method. f (x∗ ) = 0 and f (x) is Lipschitz continuous.7. then the Newton iteration converges cubically (qsuperlinear with qorder 3). x0 = 3 4.ecsecurehost. the chord method. f (x) = x2 + 1.5 2. 5. f (k) (x∗ ) = 0 = f (k+1) (x∗ )? See [131].11. f (x) = x2 . Can the ideas in [170] be applied to some of the highorder methods discussed in [190] for equations in one unknown? 5. What if f (x∗ ) = 0? How can you extend this result if f (x∗ ) = f (x∗ ) = .14. f (x) = tan−1 (x). Show that if f is a real function of one real variable. Show that if f is a real function of one real variable. x0 = . f (x) = sin(x). the standard assumptions hold. This electronic version is for personal use and may not be duplicated or distributed.5 5. .9.7. Write a program that solves single nonlinear equations with Newton’s method. f (x) = cos(x) − x. x0 = . 5. For the secant method.com/SIAM/FR16. 92 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 5. Apply your program to the following function/initial iterate combinations.
6.7.22. Typically (see [89].7. [184]) the computed LU factorization of F (x) is the exact LU factorization of a nearby matrix.7. Does the trick in Exercise 5. Tabulate or plot iteration statistics. c combination and the MATLAB flops command to get an idea for the number of ﬂoatingpoint operations required by the various methods.22). Explain your results (see [50] and [157]). LU = F (x) + E. How do the execution times and ﬂop counts change? 5. . Duplicate the results on the Hequation in § 5.9 and c = . Use the theory in [89].19. 5. How does the performance of the methods diﬀer as N and c change? Compare your results to the tabulated values of the H function in [41] or [14].20) is rquadratically (but not qquadratically) convergent to 0 if βηγ < 1/2. Modify nsol to accept this analytic Jacobian rather than evaluate a diﬀerence Jacobian. Compute by hand the Jacobian of F .9999 with ﬁxedpoint iteration. Try diﬀerent values of N and c. [170].11 improve the convergence? Try one of the approaches in [51]. How does this change aﬀect the convergence rate observations that you have made? Buy this book from SIAM at http://www. You might also look at [90].7. c = . the discretized nonlinearity from the H equation in (5. Use the MATLAB cputime command to see how long the entire iteration took for each N. Is this still the case if the convergence is qsuperlinear but there is no qorder? What if the convergence is qlinear? 5. or [118] for acceleration of the convergence. Compare timings and ﬂop counts with Newton’s method.1 to describe how this factorization error aﬀects the convergence of the Newton iteration.Copyright ©1995 by the Society for Industrial and Applied Mathematics.17. Modify dirder by setting the forward diﬀerence step h = 10−2 .7. Compare execution times in your environment.20. Solve the Hequation with N = 100. How does the analysis for the QR factorization diﬀer? 5. and the algorithm in nsol.9 (see [52]). 5. Prove that F can be computed at a cost of less than twice that of an evaluation of F . [49].21. This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost. Try the chord method and compare the results to part 4 of Exercise 5. 5.16. Solve the Hequation with c = 1 and N = 100. [75]. [35] and [155] for more discussion of this.com/SIAM/FR16.7.7.html. [184] or another book on numerical linear algebra and Theorem 5. NEWTON’S METHOD 93 is a decreasing function of n for n suﬃciently large.7. the chord method.4.18. Show that the sequence on the right side of (5.7. 5.
λ) of an N ×N matrix A by solving the system of N + 1 nonlinear equations (5. What is the Jacobian of this system? If x = (φ. such as the the MATLAB rand function. for example. λ) = Aφ − λφ φT φ − 1 = 0 0 for the vector x = (φT . how are the convergence rates diﬀerent? What is an appropriate termination criterion? At what point does the iteration stagnate? 5. λ)T ∈ RN +1 . This electronic version is for personal use and may not be duplicated or distributed. you could modify diffjac to use the techniques of [47] and nsol to use the sparse matrix factorization in MATLAB.7. λ) ∈ RN +1 is an eigenvectoreigenvalue pair. when is F (x) nonsingular? Relate the application of Newton’s method to (5. Let u0 = 0 be the initial iterate. Show that the number n of Newton iterations needed to obtain en ≤ e0 is O(log( )) and that the number of ﬂoatingpoint operations required is O(N 3 log( )).23) to the inverse power method. Suppose you try to ﬁnd an eigenvectoreigenvalue pair (φ.7. Apply your code to the central diﬀerence discretization of the twopoint boundary value problem −u = sin(u) + f (x). and f (x) = 2 − sin(x(1 − x)).html. Buy this book from SIAM at http://www.25. 94 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 5. the solution is u∗ (x) = x(1 − x). Develop a sparse version of nsol for problems with tridiagonal Jacobian. Here.26. What about the chord method? 5.ecsecurehost. In MATLAB. .7. If the size of the noise is O(10−4 ) and you make the appropriate adjustments to dirder.com/SIAM/FR16. u(0) = u(1) = 0. the cost of a Jacobian is O(N ) function evaluations. 5. and that x0 is near enough to x∗ so that the Newton iteration converges qquadratically to x∗ . that the cost of a function evaluation is O(N 2 ) ﬂoatingpoint operations.7. Add noise to your function evaluation with a random number generator.24.23. See [150] for more development of this idea.23) F (φ. Assume that the standard assumptions hold.Copyright ©1995 by the Society for Industrial and Applied Mathematics.
Let the standard assumptions hold.4. We will present this result in two parts.2) e+ ≤ KI ( ec + ηc ) ec .1 we give a straightforward analysis that uses the techniques of Chapter 5. See [145] and [175]. 6. but the other Krylov subspace methods discussed in Chapters 2 and 3 and elsewhere can also be used. This electronic version is for personal use and may not be duplicated or distributed. Before that we will give the basic result from [55] to show how a step that satisﬁes (6. for discussion of these ideas in the context of the classical stationary iterative methods.1) as the forcing term. This was the view taken in [55] where inexact Newton methods in which the step satisﬁes (6.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Chapter 6 Inexact Newton Methods Theorem 5. The proof and application of Theorem 6.1.1. Such methods are not new.1.1 should be compared to that of Theorem 5. s satisﬁes (6. Theorem 6.2.. and x+ = xc + s then (6. This is quite useful because conditions like (6. larger) choice of the parameter η. Another way to look at this is to ask how an approximate solution of the linear equation for the Newton step aﬀects the iteration. In § 6.e.1 describes how errors in the derivative/function aﬀect the progress in the Newton iteration. In § 6. In most of this chapter we will focus on the approximate solution of the equation for the Newton step by GMRES. Then there are δ and KI such that if xc ∈ B(δ).1. .1.ecsecurehost. 6. The basic estimates We will discuss speciﬁc implementation issues in § 6.1.1) F (xc )s + F (xc ) ≤ ηc F (xc ) are considered. for example. Any approximate step is accepted provided that the relative residual of the linear equation is small.com/SIAM/FR16. We follow [69] and refer to the term ηc on the right hand side of (6.1). 95 Buy this book from SIAM at http://www.2 we show how the requirements of the simple analysis can be changed to admit a more aggressive (i.1) are precisely the small linear residual termination conditions for iterative solution of the linear system for the Newton step.1.1 and the other results in Chapter 5.html.1) aﬀects the convergence.1.1. Direct analysis.
1.3.1 e+ ≤ ec − F (xc )−1 F (xc ) + 4κ(F (x∗ ))ηc ec ≤ K ec 2 + 4κ(F (x∗ ))ηc ec .1. and • if ηn ≤ Kη F (xn ) with qorder 1 + p.ecsecurehost. the proof is complete.1).2. {ηn } ⊂ [0. η ].3) and Theorem 5. p for some Kη > 0 the convergence is qsuperlinear 2 Our deﬁnition of r is consistent with the idea that the residual in a linear equation is b − Ax. Let the standard assumptions hold. Theorem 6. then the inexact Newton iteration ¯ ¯ xn+1 = xn + sn . (6.2) and will suﬃce to explain most computational observations. 96 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Proof. where K is the constant from (5. We summarize the implications of Theorem 6. This electronic version is for personal use and may not be duplicated or distributed. Hence.html. which is a direct consequence of (6. Buy this book from SIAM at http://www. . and (4.2).1 and Theorem 5. Let δ be small enough so that the conclusions of Lemma 4.6) imply that s + F (xc )−1 F (xc ) ≤ F (xc )−1 ηc F (xc ) ≤ 4κ(F (x∗ ))ηc ec . Moreover • if ηn → 0 the convergence is qsuperlinear. If we set KI = K + 4κ(F (x∗ )). using (6.1.Copyright ©1995 by the Society for Industrial and Applied Mathematics.com/SIAM/FR16.2) note that if2 r = −F (xc )s − F (xc ) is the linear residual then s + F (xc )−1 F (xc ) = −F (xc )−1 r and (6. Now.1 hold. (4. To prove the ﬁrst assertion (6.1 for iterative methods in the next result. where F (xn )sn + F (xn ) ≤ ηn F (xn ) converges qlinearly to x∗ .1. In [55] r = F (xc )s + F (xc ).7).3) e+ = ec + s = ec − F (xc )−1 F (xc ) − F (xc )−1 r. Then there are δ and η such that if x0 ∈ B(δ).
Then there is δ such that if xc ∈ B(δ).4) F (x∗ )e+ ≤ η F (x∗ )ec . then ¯ (6.1.1.ecsecurehost. M0 = (6. This electronic version is for personal use and may not be duplicated or distributed.Copyright ©1995 by the Society for Industrial and Applied Mathematics.com/SIAM/FR16.1.1 implies that F (xc ) ≤ F (x∗ )ec + Since we have. F (x∗ )e+ = F (x∗ )(ec + s) = F (x∗ )(ec − F (xc )−1 F (xc ) − F (xc )−1 r).2) to conclude en+1 ≤ KI ( en which completes the proof.1 in that no restriction is put on the sequence {ηn } ⊂ [0. Since KI = O(κ(F (x∗ ))).5) Now.3. The results here diﬀer from those in § 6.1). ¯ This proves qlinear convergence with a qfactor of KI (δ + η ). ¯ 1−p p + Kη 2p F (x∗ ) p ) en 1+p Proof.7) and (6. ¯ If ηn → 0 then qsuperlinear convergence follows from the deﬁnition.2) holds for xc ∈ B(δ). Buy this book from SIAM at http://www.4) note that Theorem 4. s satisﬁes (6. Theorem 6.2. 1) other than requiring that 1 not be an accumulation point. with γ ec 2 2 . Let δ be small enough so that (6. . F (x∗ )ec ec = F (x∗ )−1 F (x∗ )ec ≤ F (x∗ )−1 γ F (x∗ )−1 2 F (xc ) ≤ (1 + M0 δ) F (x∗ )ec . If ηn ≤ Kη F (xn ) then we may use (4. 6. This is not the case and the purpose of this subsection is to describe more accurately the necessary restrictions on the forcing terms. To prove (6. Then if n ≥ 0 and xn ∈ B(δ) we have en+1 ≤ KI ( en + ηn ) en ≤ KI (δ + η ) en < en .1. x+ = xc + s.html. Weighted norm analysis. Reduce δ and η if needed so that ¯ KI (δ + η ) < 1.2). and ηc ≤ η < η < 1.0. ¯ where KI is from (6.2 that if F (x∗ ) is ill conditioned very small forcing terms must be used. one might conclude from Theorem 6. Let the standard assumptions hold. INEXACT NEWTON METHODS 97 Proof.
Hence.5 for all n) that try to minimize the number of inner iterations are completely justiﬁed by this result if the initial iterate is suﬃciently near the solution.7) ≤ (1 + M1 δ) r + M2 δ ec ≤ (1 + M1 δ)(1 + M0 δ)ηc F (x∗ )ec +M2 δ F (x∗ )−1 F (x∗ )ec ≤ ((1 + M1 δ)(1 + M0 δ)η + M2 δ F (x∗ )−1 ) F (x∗ )ec .4) is remarkable in its assertion that any method that produces a step with a linear residual less than that of the zero step will reduce the norm of the error if the error is measured with the weighted norm · ∗ = F (x∗ ) · .html.3 described in Theorem 6. F (x∗ )e+ ≤ F (x∗ )F (xc )−1 r + K F (x∗ ) ec 2 .6) Since F (x∗ )F (xc )−1 r ≤ r + (F (x∗ ) − F (xc ))F (xc )−1 r ≤ (1 + 2γ F (x∗ )−1 we may set ec ) r .5).4) to hold than for (6. just that 1 not be an accumulation point.2 in that we do not demand that {ηn } be bounded away from 1 by a suﬃciently large amount.4 diﬀers from Theorem 6.1. using (4. This electronic version is for personal use and may not be duplicated or distributed.3 asserts qlinear convergence in the weighted norm if the initial iterate is suﬃciently near x∗ .1 F (x∗ )(ec − F (xc )−1 F (xc )) ≤ K F (x∗ ) ec 2 .7) and (6. The application of Theorem 6. 98 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS By Theorem 5.1. M1 = 2γ F (x∗ )−1 .2).1. Theorem 6.com/SIAM/FR16.4.Copyright ©1995 by the Society for Industrial and Applied Mathematics. This is made precise in Theorem 6.4 and compare it to a modiﬁcation of a choice from Buy this book from SIAM at http://www. .1. Now let δ be small enough so that ¯ (1 + M1 δ)(1 + M0 δ)η + M2 δ F (x∗ )−1 ≤ η and the proof is complete. Note that the distance δ from the initial iterate to the solution may have to be smaller for (6. In fact. and M2 = K F (x∗ ) and obtain. F (x∗ )e+ (6. (6. The importance of this theorem for implementation is that choices of the sequence of forcing terms {ηn } (such as ηn = .1. However (6.ecsecurehost.1. We make such a choice in one of the examples considered in § 6.
e < δ∗ = F (x∗ ) δ0 ∗ Set δ = F (x∗ ) −1 δ∗ . Our proof does not rely on the (possibly false) assertions that en < δ for all n or that xn → x∗ qlinearly with respect to the unweighted norm. qlinear convergence with respect to the weighted norm · ∗ is equivalent to qlinear convergence of the sequence of nonlinear residuals { F (xn ) }.1.9) en+1 ∗ ≤ η en ¯ ∗ < δ∗ and hence xn → x∗ qlinearly with respect to the weighted norm.1.4) holds for ec < δ0 . While Theorem 6.4.Copyright ©1995 by the Society for Industrial and Applied Mathematics.4.4) and (6. For all x ∈ RN . (6. Theorem 6. {ηn } ⊂ [0. To prove the assertions on qsuperlinear convergence.html. The proof uses both (6. Let δ0 be such that (6.1.1.com/SIAM/FR16. Rather we note that (6.1 and leave the proof to the reader in Exercise 6.1. The most signiﬁcant diﬀerence is the norm in which qlinear convergence takes place.1 hold. The implications for the choice of the forcing terms {ηn } are also diﬀerent.1. .5. η] with η < η < 1. Moreover • if ηn → 0 the convergence is qsuperlinear.4) implies that if en ∗ < δ∗ then (6.2 and 6.2 might lead one to believe that a small value of ηn is necessary for convergence. The superlinear convergence results then follow exactly as in the proof of Theorem 6. We close this section with some remarks on the two types of results in Theorems 6.8) Note that e < δ0 if x ∗ ≤ F (x∗ ) x ≤ κ(F (x∗ )) x ∗ . Proof. and • if ηn ≤ Kη F (xn ) p for some Kη > 0 the convergence is qsuperlinear with qorder 1 + p.1. Theorem 6.2.2).1. then the inexact Newton ¯ iteration xn+1 = xn + sn . This proves the ﬁrst assertion. This electronic version is for personal use and may not be duplicated or distributed. Let the standard assumptions hold. eventually en will be small enough so that the conclusions of Theorem 6.2. Then e0 ∗ < δ∗ if e0 < δ. Then there is δ such that if x0 ∈ B(δ).ecsecurehost. Our ﬁrst task is to relate the norms · and · ∗ so that we can estimate δ. INEXACT NEWTON METHODS 99 [69] that decreases ηn rapidly with a view toward minimizing the number of outer iterations. note that since xn → x∗ .4 shows that the sequence of forcing terms need only be kept bounded away from Buy this book from SIAM at http://www. We state this result as Proposition 6. where F (xn )sn + F (xn ) ≤ ηn F (xn ) converges qlinearly with respect to · ∗ to x∗ .
1.html. such as those arising from a diﬀerence approximation of the action of F (xc ) on a vector.5. CG. Then F (xn ) converges qlinearly to 0 if and only if en ∗ does. but at a price of a greater number of outer iterations. x+ = xc + s. We will discuss other factors in the choice of forcing terms in § 6. This electronic version is for personal use and may not be duplicated or distributed. Let the standard assumptions hold. Then there are δ and KI such that if xc ∈ B(δ). would use SOR. Errors in the function.e.11) e+ ≤ KI (( ec + ηc ) ec + (xc ) ).1) holds). These methods have also been called truncated Newton methods [56].10). In the examples discussed in § 6. terminating when the relative linear residual is smaller than ηc (i. ¯ 6.ecsecurehost.1 and 6.1. This naming convention is taken from [175] and [145].5 may well suﬃce for linear convergence. Then there is δ such that if xc ∈ B(δ). we state two theorems on inexact local improvement. s satisﬁes (6.1.1 and its proof for an illustration of this point. NewtonGMRES. Buy this book from SIAM at http://www. [136] in the context of optimization.3.4 we will scale the l2 norm by a factor of 1/N so that the results for the diﬀerential and integral equations will be independent of the computational mesh. or GMRES to solve the linear system. In this section we use the l2 norm to measure nonlinear residuals.12) F (x∗ )e+ ≤ η F (x∗ )ec + KE (xc ) . Theorem 6. Let the standard assumptions hold. Proposition 6.Copyright ©1995 by the Society for Industrial and Applied Mathematics.3.1. We leave the proofs.1) with an iterative method for the linear system for the Newton step.10) F (xc )s + F (xc ) + (xc ) ≤ ηc F (xc ) + (xc ) and x+ = xc + s.2. Theorem 6. The name of the method indicates which linear iterative method is used.com/SIAM/FR16. See [55] or Proposition 6.1. A constant sequence such as ηn = . Note that errors in the derivative. then (6. NewtonCG. s satisﬁes (6. Let the standard assumptions hold and let xn → x∗ . . The reason for this is that the Krylov methods use the scalar product and the estimates in Chapters 2 and 3 are in terms of the Euclidean norm. for example.3.1. which are direct analogs to the proofs of Theorems 6.2. and ηc ≤ η < η < 1. Newtoniterative methods A Newtoniterative method realizes (6. to the reader as exercises. NewtonSOR. In this section.6. 100 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 1. 6. when (6. can be regarded as part of the inexact solution of the equation for the Newton step. Typically the nonlinear iteration that generates the sequence {xn } is called the outer iteration and the linear iteration that generates the approximations to the steps is called the inner iteration.1. then ¯ there is KE such that (6.
and that the initial iterate for the linear problem is the zero vector. While ρ > η F (x) (a) k = k + 1 (b) vk+1 = Dh F (x : vk ) for j = 1. Newton GMRES. We begin by discussing the eﬀects of a forward diﬀerence approximation to the action of the Jacobian on a vector. . kmax. Rk+1 . In the case of CGNR and CGNE. s = Vk y k . We give a MATLAB implementation of this algorithm in the collection of MATLAB codes. β = ρ. v1 = r/ r 2 . To illustrate this point. η. ρ) 1. fdgmres(s. It is important to note that this is entirely diﬀerent from forming the ﬁnite diﬀerence Jacobian ∇h F (x) and applying that matrix to w. In many implementations [20]. k = 0 2. h. hjk = vk+1 vj ii. an evaluation of the action of F (xc )T on a vector is also required. If the linear iterative method is any one of the Krylov subspace methods discussed in Chapters 2 and 3 then each inner iteration requires at least one evaluation of the action of F (xc ) on a vector. .Copyright ©1995 by the Society for Industrial and Applied Mathematics. k T i. . . a sequence of approximate steps {sk } is produced.1. (f) ρ = βe1 − Hk y k 3.15). In fact. ρ = r 2 . This electronic version is for personal use and may not be duplicated or distributed. that b = −F (x).html. (5. as pointed out in [20]. INEXACT NEWTON METHODS 101 6. . .ecsecurehost. 0.2. [24].com/SIAM/FR16. Buy this book from SIAM at http://www. s = 0. the action of F (xc ) on a vector w is approximated by a forward diﬀerence. x. 0)T (e) e1 = ∈ Rk+1 Minimize βe1 − Hk y k Rk+1 to obtain y k ∈ Rk . . Algorithm 6. we give the forwarddiﬀerence GMRES algorithm for computation of the Newton step. F. In our MATLAB implementation we solve the least squares problem in step 2e by the Givens rotation approach used in Algorithm gmres. r = −F (x).2.k = vk+1 2 2 2 and k < kmax do (d) vk+1 = vk+1 / vk+1 (1. . application of GMRES to the linear equation for the Newton step with matrixvector products approximated by ﬁnite diﬀerences is the same as the application of GMRES to the matrix Gh F (x) whose last k − 1 columns are the vectors vk = Dh F (x : vk−1 ) The sequence {vk } is formed in the course of the forwarddiﬀerence GMRES iteration.1. Dh F (x : w) for some h. Note that matrixvector products are replaced by forward diﬀerence approximations to F (x). We provide a detailed analysis of the NewtonGMRES iteration. s = −F (x)−1 F (x). vk+1 = vk+1 − hjk vj (c) hk+1.
Let {uj } be any orthonormal basis for RN such that uj = vj for 1 ≤ j ≤ k. Note that Dh F (x : uj ) = 1 0 F (x + th x 2 uj )uj dt 1 0 (F = F (x)uj + (x + th x 2 uj ) − F (x))uj dt. Bs + F (x) and therefore (6.1.com/SIAM/FR16. h ≤ h. h. Proposition 6.14) B − F (x) 2 ≤ h x 2 γ/2 ≤ h¯ /2 γ Since the linear iteration (fdgmres) terminates in k < kmax iterations.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Let η ∈ (0. ¯ ¯ Then there are CG . Since F is Lipschitz continuous and {uj } is an orthonormal basis for RN we have. Assuming that there is no error in the evaluation of F we may summarize the observations above in the following proposition. . Let the standard assumptions hold. and Algorithm fdgmres terminates with k < kmax then the computed step s satisﬁes (6. The action of Gh F on Kk . where {vj }k is the orthonormal basis for Kk generated j=1 by Algorithm fdgmres. with γ = γ( x∗ 2 + δ).2. Hence Gh F is also ﬁrst order accurate in h. Proof. which is a special case of the more general results in [20].1 hold.13) F (x)s + F (x) 2 < (η + CG h) F (x) 2 .3. the orthogonal complement of Kk . ¯ (6. could be speciﬁed by the action of ∇h F on any basis for Kk . 1). since B and Gh F agree on Kk . This electronic version is for personal use and may not be duplicated or distributed. we have.html.ecsecurehost. γ ¯γ h¯ ≤ F (x∗ )−1 −1 2 /2 Buy this book from SIAM at http://www. for the purposes of ⊥ analysis. the diﬀerence between ∇h F and Gh F is that the columns are computed by taking directional derivatives based on two diﬀerent orthonormal bases: ∇h F using the basis of unit vectors in coordinate directions and Gh F the basis for the Krylov space Kk ⊥ constructed by algorithm fdgmres. First let δ be small enough so that B(δ) ⊂ Ω and the conclusions to Lemma 4.15) Assume that F (x)s + F (x) 2 2 ≤ η F (x) 2 ≤ η F (x) 2 + h¯ s 2 /2. 102 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS As should be clear from inspection of Algorithm gmres. Consider the linear map deﬁned by its action on {uj } Buj = Dh F (x : uj ). is not used by the algorithm and. and δ such that if x ∈ B(δ).
1 and (6. Then there is δ such that if x0 ∈ B(δ) and the sequences {ηn } and {hn } satisfy σn = ηn + CG hn ∈ [0. . γ Combining (6.1. η) F (x 2 Proposition 6.2. kmax. and • if σn ≤ Kη F (xn ) p for some Kη > 0 the convergence is qsuperlinear 2 with qorder 1 + p.ecsecurehost.15) will hold. and Proposition 6. σ such that if x0 ∈ B(δ) and the sequences {ηn } and {hn } ¯ satisfy σn = ηn + CG hn ≤ σ ¯ then the forward diﬀerence NewtonGMRES iteration xn+1 = xn + sn where sn is computed by Algorithm fdgmres with arguments (sn .4 and Theorem 6. but also the steps {hn } used to compute forward diﬀerences.3.1 applies to a restarted GMRES because upon successful termination (6.2. Assume that the assumptions of Proposition 6. The proofs are immediate consequences of Theorems 6.2.16) with (6. Then there are δ. Therefore. Theorem 6. Moreover.2.1. This electronic version is for personal use and may not be duplicated or distributed. ηn .html.1 implies that a ﬁnite diﬀerence implementation will not aﬀect the performance of NewtonGMRES if the steps in the forward diﬀerence approximation of the derivatives are suﬃciently small.com/SIAM/FR16. hn . We summarize our results so far in the analogs of Theorem 6.2.1.17) F (xn )sn + F (xn ) 2 < σn F (xn ) 2 . 6.2. F. Theorem 6.1. σ ] ¯ with 0 < σ < 1 then the forward diﬀerence NewtonGMRES iteration ¯ xn+1 = xn + sn Buy this book from SIAM at http://www. • if σn → 0 the convergence is qsuperlinear.2.2.Copyright ©1995 by the Society for Industrial and Applied Mathematics. therefore. we must specify not only the sequence {ηn }.1 and are left as (easy) exercises. INEXACT NEWTON METHODS 103 Then Lemma 4. xn . Assume that the assumptions of Proposition 6. Note also that Proposition 6. ρ) converges qlinearly and sn satisﬁes (6.15) imply F (x∗ )−1 −1 2 s 2 /2 ≤ F (x)−1 ≤ F (x)s 2 −1 2 s 2 2 ≤ (1 + η) F (x) 2 ¯γ + h¯ s 2 /2. (6.2.15) completes the proof with CG = 4¯ (1 + ∗ )−1 .1 hold.16) s 2 ≤ 4(1 + η) F (x∗ )−1 F (x) 2 . In an implementation.4.1.2.1 hold.
We adjust η as the iteration progresses with a variation of a choice from [69]. but require evaluation of transposevector products.6. The reader may also be interested in experimenting with the other Krylov subspace methods described in § 3. Other Newtoniterative methods. Setting η to a constant for the entire iteration is often a reasonable strategy as we see in § 6. the vector of termination tolerances as we did for nsol. but the choice of that constant depends on the problem. As input we must give the initial iterate. The choice in Buy this book from SIAM at http://www. ηn . a forward diﬀerence formulation is not an option because approximation of the transposevector product by diﬀerences is not possible. If F is symmetric and positive deﬁnite. 104 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS where sn is computed by Algorithm fdgmres with arguments (sn . CGNR and CGNE oﬀer alternatives. as it is in the context of unconstrained minimization.ecsecurehost.2. 6.2. GMRES(m). See also [6]. for a small m. The reader may explore this further in Exercise 6.com/SIAM/FR16. In the case where hn is approximately the square root of machine roundoﬀ. we must provide a method for forming the sequence of forcing terms {ηn }. GMRES may not be practical.17). Moreover • if σn → 0 the convergence is qsuperlinear. may not converge rapidly enough to be useful. NewtonGMRES implementation We provide a MATLAB code nsolgm in the collection that implements a forwarddiﬀerence NewtonGMRES algorithm. Newtonmultigrid [99] is one approach using a stationary iterative method. Computation of ∇h F (x) and its transpose analytically is one option as is diﬀerentiation of F automatically or by hand. If a constant η is too small. [111]. When storage is restricted or the problem is very large.2 states that the observed behavior of the forward diﬀerence implementation will be the same as that of an implementation with exact derivatives. This issue is independent of the particular linear solver and our discussion is in the general inexact Newton method setting. This electronic version is for personal use and may not be duplicated or distributed. and [12]. ρ) converges qlinearly with respect to · ∗ and sn satisﬁes (6. If a transpose is needed. much eﬀort can be wasted in the initial stages of the iteration.html. Other iterative methods may be used as the solver for the linear equation for the Newton step. σn ≈ ηn if ηn is not too small and Theorem 6. hn .5. F. [63]. .Copyright ©1995 by the Society for Industrial and Applied Mathematics.4. BiCGSTAB and TFQMR should be considered in all cases where there is not enough storage for GMRES(m) to perform well.13. In addition. and • if σn ≤ Kη F (xn ) p for some Kη > 0 the convergence is qsuperlinear 2 with qorder 1 + p. xn .3. the function. This algorithm requires diﬀerent inputs than nsol. kmax. NewtonCG is a possible approach. [78]. 6.2. as it will be if CGNR or CGNE is used as the iterative method.
18) B ηn = ηmax . The idea is that if ηn−1 is suﬃciently large we do not let ηn decrease by much more than a factor of ηn−1 . A for n > 0 would guarantee qquadratic convergence by then setting ηn = ηn Theorem 6. We use dirder to approximate directional derivatives and use the default value of h = 10−7 . . 2 A 2 min(ηmax . γηn−1 )). the most information possible would be extracted from the inner iteration.18) the parameter ηmax is an upper limit on the sequence {ηn }. max(ηn . The constant . ηn ). Exercise 6.1 is somewhat arbitrary. In this way. The algorithm nsolgm does this and uses (6. B It may happen that ηn is small for one or more iterations while xn is still far from the solution. n = 0. A where γ ∈ (0. As a measure of the degree to which the nonlinear iteration approximates the solution we begin with A ηn = γ F (xn ) 2 / F (xn−1 ) 2 .19) C ηn = A min(ηmax . If ηn is uniformly bounded away from 1.html. There is a chance that the ﬁnal iterate will reduce F far beyond the desired level and that the cost of the solution of the linear equation for the last step will be more accurate than is really needed. min(η A max .Copyright ©1995 by the Society for Industrial and Applied Mathematics. In [69] the choices γ = . A method of safeguarding was suggested in [69] to avoid volatile decreases in ηn . 2 n > 0. n = 0.5τt / F (xn ) )). . In order to specify the choice at n = 0 and bound the sequence away from 1 we set (6. This electronic version is for personal use and may not be duplicated or distributed. This safeguarding does improve the performance of the iteration. γηn−1 > .9 as recommended in [69]. n > 0. (6.1. n > 0.1. γηn−1 ≤ .com/SIAM/FR16.9 and ηmax = .20) C ηn = min(ηmax . This oversolving on the ﬁnal step can be controlled comparing the norm of the current nonlinear residual F (xn ) to the nonlinear residual norm at which the iteration would terminate τt = τa + τr F (x0 ) and bounding ηn from below by a constant multiple of τt / F (xn ) . Buy this book from SIAM at http://www. Over solving means that the linear equation for the Newton step is solved to a precision far beyond what is needed to correct the nonlinear iteration. In (6.9999 are used.9 asks the reader to modify nsolgm so that other choices of the sequence {ηn } can be used. In all the examples in this book we use the value γ = . 1] is a parameter. max(ηn .1. INEXACT NEWTON METHODS 105 [69] is motivated by a desire to avoid such over solving.ecsecurehost.1.5. ηn ). ηmax .
we also terminate the outer iteration on small l2 norms of the nonlinear residuals. and the Krylov basis {vk }m . Note that if only a single inner iteration is needed. In all the ﬁgures we plot the relative nonlinear residual F (xn ) 2 / F (x0 ) 2 against the number of function evaluations required by all inner and outer iterations to compute xn . η) (c) x = x + s (d) Evaluate F (x) (f) If F (x) 2 √ (e) r+ = F (x) 2 / N . However. GMRES will have to be restarted. but also the total cost. (b) fdgmres(s. F. The preconditioners we use in § 6. We illustrate the consequences of this in § 6. Hence if a large value of η can be used. Note that the cost of an outer iterate is one evaluation of F to compute the value at the current iterate and other evaluations of F to compute the forward diﬀerences for each inner iterate.4 and ask the reader to make the necessary modiﬁcations to nsolgm in Exercise 6. because of our applications to diﬀerential and integral equations.1. F (x + hv). Examples for NewtonGMRES In this section we consider two examples. Since the Newton steps for M F (x) = 0 are the same as those for F (x) = 0. we scale the l2 norm of the nonlinear √ residual by a factor of 1/ N so that constant functions will have norms that are independent of the computational mesh. 106 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Algorithm 6. This electronic version is for personal use and may not be duplicated or distributed. If a maximum of m GMRES iterations is needed (or GMRES(m) is used as a linear solver) the storage requirements of Algorithm nsolgm are the m + 5 vectors x.ecsecurehost. From these plots one can compare not only the number of outer iterations. the total cost of the outer iteration will be two function evaluations since F (xc ) will be known. s. . it is equivalent to precondition the nonlinear equation before calling nsolgm. σ = r+ /rc .Copyright ©1995 by the Society for Industrial and Applied Mathematics. 6. This enables us to directly compare the costs of diﬀerent strategies for selection of η.5. We revisit the Hequation and solve a preconditioned nonlinear convectiondiﬀusion equation.html. x. the cost of the entire iteration can be greatly reduced.4. For large and poorly conditioned problems. F (x).com/SIAM/FR16. One new function evaluation will be needed to approximate the action of F (xc ) on a vector Buy this book from SIAM at http://www.9.4 are independent of the outer iteration. nsolgm(x. rc = r+ ≤ τr r0 + τa exit.3. η) √ 1. x + hv. Counts of function evaluations corresponding to outer iterations are indicated by circles. τ. k=1 Since GMRES forms the inner iterates and makes termination decisions based on scalar products and l2 norms. rc = r0 = F (x) 2 / N √ 2. F. Do while F (x) 2 / N > τr r0 + τa (a) Select η.
Here the iteration with ηn = . We see that the constant η iteration terminates after 12 function evaluations. NewtonGMRES for the Hequation.Copyright ©1995 by the Society for Industrial and Applied Mathematics.2 the results change for the more illconditioned problem with c = . We set the parameters in (6. INEXACT NEWTON METHODS 107 and then F (x+ ) will be evaluated to test for termination. 6. The chord method with the diﬀerent (but very similar for this problem) l∞ termination condition for the same problem. The initial iterate is the function identically one. This electronic version is for personal use and may not be duplicated or distributed. and 230 thousand ﬂoatingpoint operations. 3 outer iterations.1 with the solid line and using the sequence given by (6.4.20).25.1. Chandrasekhar Hequation. slower by a factor of over 1000. This cost is entirely a result of the computation and factorization of a single Jacobian.20) with the dashed line. which therefore has a minimum cost of three function evaluations. 10 10 10 10 10 10 10 10 10 0 1 2 Relative Nonlinear Residual 3 4 5 6 7 8 0 2 4 6 8 Function Evaluations 10 12 Fig.9999. This is a slightly higher overall cost than the other approach in which {ηn } is given by (6.9.com/SIAM/FR16.9 and ηmax = . We set the parameters in (6. This choice of ηmax was the one that did best overall in our experiments on this problem.9 and ηmax = .3 million ﬂoatingpoint operations.1 we plot the progress of the iteration using η = . reported in § 5.25. In Fig.1 performed slightly better and Buy this book from SIAM at http://www. For each example we compare a constant value of ηn with the choice given in (6. 6. and roughly 275 thousand ﬂoatingpoint operations. 4 outer iterations. In Fig. .1.20) to γ = .20) and used as the default in nsolgm.html. 6. We solve the Hequation on a 100point mesh with c = . We used τr = τa = 10−6 in this example. 6. c = .ecsecurehost.6 incurred a cost of 3.9 using two schemes for selection of the parameter η. We also count the evaluation of F (x0 ) as an additional cost in the evaluation of x1 .20) to γ = . which terminated after 10 function evaluations.
21) We consider the partial diﬀeren − ∇2 u + Cu(ux + uy ) = f Buy this book from SIAM at http://www.9999 10 0 Relative Nonlinear Residual 10 1 10 2 10 3 10 4 0 2 4 6 8 10 12 Function Evaluations 14 16 18 Fig.3. and 537 thousand ﬂoatingpoint operations. C = 20. This electronic version is for personal use and may not be duplicated or distributed.2. 108 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 0 10 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 10 6 10 7 0 5 10 15 Function Evaluations 20 25 Fig. terminated after 22 function evaluations and 7 outer iterations. The iteration with the decreasing {ηn } given by (6. 6.4.com/SIAM/FR16.2.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 6.20) required 23 function evaluations. NewtonGMRES for the PDE. c = . Convectiondiﬀusion equation. 7 outer iterations. NewtonGMRES for the Hequation.html. 6. tial equation (6. .ecsecurehost. and roughly 513 thousand ﬂoatingpoint operations.
20) required 16 function evaluations.9 and ηmax = .7 we discretized (6. a large choice of ηmax or constant η may be the more eﬃcient choice. 1) × (0. In cases where the initial iterate is very good and only a single Newton iterate might be needed. and 2.com/SIAM/FR16.html. 6. 4 outer iterations. In the computation we preconditioned (6. however. . where h = 1/32. We compare the same possibilities for {ηn } as in the previous example and report the results in Fig. Preconditioning is crucial for good performance. To be consistent with the secondorder accuracy of the diﬀerence scheme we used τr = τa = h2 . The iteration with {ηn } given by (6.20) reduced the number of inner iterations in the early stages of the iteration and avoided oversolving. Buy this book from SIAM at http://www.21) on a 31 × 31 grid using centered diﬀerences. The computation with constant η terminated after 19 function evaluations and 4 outer iterates at a cost of roughly 3 million ﬂoatingpoint operations. INEXACT NEWTON METHODS 109 with homogeneous Dirichlet boundary conditions on the unit square (0.5.Copyright ©1995 by the Society for Industrial and Applied Mathematics.20) we set γ = . As in § 3. When preconditioning was not done. We set C = 20 and u0 = 0. f has been constructed so that the exact solution was the discretization of 10xy(1 − x)(1 − y) exp(x4. the iteration for C = 20 required more than 80 function evaluations to converge. In Exercise 6. Examples of such cases include (1) implicit time integration of nonlinear parabolic partial diﬀerential equations or ordinary diﬀerential equations in which the initial iterate is either the converged solution from the previous time step or the output of a predictor and (2) solution of problems in which the initial iterate is an interpolant of the solution from a coarser mesh.6 million ﬂoatingpoint operations.5. the choice of {ηn } given by (6. In (6. 1).9 the reader is asked to explore this.21) with the fast Poisson solver fish2d. In this case.ecsecurehost. This electronic version is for personal use and may not be duplicated or distributed.3.5 ).
5. What happens when c = 1? See [119] for an explanation of this.2.12. how is the performance aﬀected if GMRES(m) is used instead of GMRES? Buy this book from SIAM at http://www.99999. For the Hequation example in § 6.Copyright ©1995 by the Society for Industrial and Applied Mathematics. and x is suﬃciently near x∗ then F (x) /(1 + ) ≤ F (x∗ )e ≤ (1 + ) F (x) . 6.4. Can anything like Proposition 6.14). 6. and for the convectiondiﬀusion equation various values of C such as C = 1.4 equally accurate? 6.4.5.7. Give an example in which F (xn ) → 0 qlinearly but xn does not converge to x∗ qlinearly. such as c = .html.5 and 6. ηn = 10−4 [32].3. Prove Proposition 6.5. what are some possible implications? 6. Is there a relation to the Shamanskii method? 6. Experiment with the sequence of forcing terms {ηn }.11. 6. For the convectiondiﬀusion example in § 6. ηn = min( F (xn ) 2 . 110 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 6.05. 6. Verify (6.5.6. In the DASPK code for integration of diﬀerential algebraic equations.1. which may be less often that with each outer iteration.2.10.1 by showing that if the standard assumptions hold. (n + 2)−1 ) [56].4.8). 0 < < 1. .5.2.4. . This electronic version is for personal use and may not be duplicated or distributed.2. Prove Theorems 6.5. 6. Exercises on inexact Newton methods 6. How do the choices ηn = 1/(n + 1).1.5.1. Are the converged solutions for the preconditioned and unpreconditioned convectiondiﬀusion example in § 6.9.com/SIAM/FR16. [23]. ηn = 21−n [24]. 6. Prove Theorems 6. Duplicate the results in § 6. For the Hequation experiment with various values of c. Discuss this strategy and when it might be useful.5.5. ηn = .1 be proved for a ﬁnitediﬀerence BiCGSTAB or TFQMR linear solver? If not.5.5.1.5.6. 40 and how preconditioning aﬀects the iteration.75 aﬀect the results? Present the results of your experiments as graphical iteration histories.999999. 6. and ηn = .5. preconditioned NewtonGMRES is implemented in such a way that the data needed for the preconditioner is only computed when needed.2 and 6. vary the parameter c and see how the performance of NewtonGMRES with the various choices of {ηn } is aﬀected. Verify (6.5.8.ecsecurehost.5. 10. 6.1. .
Copyright ©1995 by the Society for Industrial and Applied Mathematics.13.5. Buy this book from SIAM at http://www. . and nsola. or NewtonTFQMR to the two examples in § 6.7.14.com/SIAM/FR16.ecsecurehost. which we describe more fully in Chapter 8. If you have done Exercise 5. ﬂoatingpoint operations. function evaluations) compare to NewtonGMRES? Experiment with the forcing terms. INEXACT NEWTON METHODS 111 6. This electronic version is for personal use and may not be duplicated or distributed. might be of use in this exercise. How will you deal with the need for a transpose in the case of CGNR? How does the performance (in terms of time. 6. NewtonBiCGSTAB. which allows you to choose from a variety of forward diﬀerence Krylov methods.5.4.html. apply nsolgm to the same twopoint boundary value problem and compare the performance. Apply NewtonCGNR. The MATLAB codes fdkrylov.25.
112 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Buy this book from SIAM at http://www.com/SIAM/FR16. .ecsecurehost.html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
as we shall see in § 7. Chapter 7 Broyden’s method QuasiNewton methods maintain approximations of the solution x∗ and the Jacobian at the solution F (x∗ ) as the iteration progresses. sT s sT s In (7. The construction of B+ determines the quasiNewton method. This means that the updated approximation to F (x∗ ) satisﬁes the secant equation (7. If xc and Bc are the current approximate solution and Jacobian. .3) alone is a system of N equations in N 2 113 Buy this book from SIAM at http://www. Broyden’s method is an example of a secant update.2) y = F (x+ ) − F (xc ) and s = x+ − xc .2) B+ = Bc + (y − Bc s)sT F (x+ )sT = Bc + .com/SIAM/FR16.1) −1 x+ = xc − Bc F (xc ). The advantages of such methods. In fact. After the computation of x+ . Hence.3) B+ s = y. For equations in several variables (7. is locally superlinearly convergent. these methods could have an advantage in terms of function evaluation cost over Newtoniterative methods. quasiNewton methods require only one function evaluation for each nonlinear iterate. In one dimension.Copyright ©1995 by the Society for Industrial and Applied Mathematics. if a good preconditioner (initial approximation to F (x∗ )) can be found. This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost. (7. Bc is updated to form B+ . Broyden’s method. if the Jacobian is dense. Broyden’s method [26] computes B+ by (7. and hence is a very powerful alternative to Newton’s method or the chord method.3. the cost is little more than that of the chord method. The method we present in this chapter. there is no cost in function evaluations associated with an inner iteration. In comparison with Newtoniterative methods. is that solution of equations with the quasiNewton approximation to the Jacobian is often much cheaper than using F (xn ) as the coeﬃcient matrix. then (7.3) uniquely speciﬁes the classical secant method.html.
the Broyden iteration will converge qsuperlinearly to the root. of the approximate Jacobian. Locally superlinearly convergent secant methods can be designed to maintain the sparsity pattern. [84]. (7. [82]. . [84]. which we will not discuss further. [96]. 7. 114 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS unknowns. We will express our results in terms of the error in the Jacobian (7. One interesting result. In this book we focus on Broyden’s method and our methods of analysis are not as general as those in [62] and [64].1. and [116]. but easier to verify condition in order to analyze Broyden’s method. See [62] and [64] for discussion of this topic.6) n→∞ lim En sn = 0. When indexing of the steps is necessary. Veriﬁcation of qsuperlinear convergence cannot be done by the direct approaches used in Chapters 5 and 6. Broyden’s method is also applicable to linear equations [29]. The role of the secant equation is very deep.4) is consistent with our use of e = x − x∗ . In this chapter we will assume that the function values are accurate and refer the reader to [66]. or positive deﬁniteness.5) −1 xn+1 = xn − Bn F (xn ) where Bn = F (x∗ )+En ≈ F (x∗ ) is generated by some method (not necessarily Broyden’s method). and [64] for discussion and analysis of several such methods. [60] on the sequence e of steps {sn } and errors in the Jacobian {En } (7. and we refer the reader to [28]. symmetry. This electronic version is for personal use and may not be duplicated or distributed. and [114] for discussions of quasiNewton methods with inaccurate data.Copyright ©1995 by the Society for Industrial and Applied Mathematics. The analysis is somewhat clearer in the linear case and the plan of this chapter is to present results for the linear case for both the theory and the examples.4) E = B − F (x∗ ) and the step s = x+ −xc . [82]. [130]. where B ≈ A is updated. [104]. Some examples can be found in [101].com/SIAM/FR16. [63]. sn Buy this book from SIAM at http://www. Ax = b.2 to a more subtle. [143]. While we could use E = B −F (x).html. This technique is based on veriﬁcation of the Dennis–Mor´ condition [61]. More subtle structural properties can also be preserved by secant methods. In this chapter we show that if the data x0 and B0 are suﬃciently good. The Dennis–Mor´ condition e In this section we consider general iterative methods of the form (7. is that for linear problems Broyden’s method converges in 2N iterations [29]. We must relate the superlinear convergence condition that ηn → 0 in Theorem 6. [65]. [104]. [113]. we will write sn = xn+1 − xn . [143]. [95].ecsecurehost. [62]. One can also apply the method to linear least squares problems [84].1.
. then for n suﬃciently large (7.10) νn = en+1 / en is deﬁned and and νn → 0 by superlinear convergence of {xn }.7) En sn = −F (x∗ )sn − F (xn ) = −F (x∗ )en+1 + F (x∗ )en − F (xn ). Now. [28].9) we have en+1 = νn en ≤ 2νn sn . Let the standard assumptions hold.1. Our formulation of the Dennis–Mor´ result is e a bit less general than that in [60] or [63].Copyright ©1995 by the Society for Industrial and Applied Mathematics.6) holds. Assume that xn = x∗ for any n. assume that xn → x∗ and that (7. Theorem 7. by (7. are used to prove superlinear convergence for Broyden’s method in this chapter and many other quasiNewton methods. By (7. sn Buy this book from SIAM at http://www. Then xn → x∗ qsuperlinearly if and only if xn → x∗ and the Dennis–Mor´ condition (7. Theorem 7.8) En sn ≤ F (x∗ )en+1 + γ en 2 /2.1.7) (7. This electronic version is for personal use and may not be duplicated or distributed. 1 0 (F (x∗ ) − F (x∗ + ten ))en dt F (x∗ )en − F (xn ) ≤ γ en 2 /2.8) implies that En sn ≤ (2 F (x∗ ) νn + γ en ) sn which implies the DennisMor´ condition (7.html. and hence (7.5).6) holds.1 and its corollary for linear problems.com/SIAM/FR16. and [64] for several more examples. The assumption that xn = x∗ for any n implies that the sequence (7.ecsecurehost.6). if xn → x∗ qsuperlinearly. See [61]. We use the fundamental theorem of calculus and the standard assumptions to obtain F (x∗ )en − F (xn ) = and hence Therefore. e Proof. [62]. Since −F (xn ) = Bn sn = F (x∗ )sn + En sn we have (7. e Conversely. Let µn = En sn .1. let {Bn } be a sequence of nonsingular N × N matrices. BROYDEN’S METHOD 115 The main result in this section. let x0 ∈ RN be given and let {xn }∞ be given n=1 by (7.9) sn /2 ≤ en ≤ 2 sn .
11) (7. using the standard assumptions and (7.1.16) F (xn ) + F (xn )sn ≤ ηn F (xn ) .2. .13) µn ≤ F (x∗ )−1 −1 /2.ecsecurehost. and we employ notation from all four of these references.11) again.html.15) ≤ F (xn ) + F (x∗ )sn + (F (x∗ ) − F (xn ))sn ≤ (µn + γ en ) sn . Convergence analysis Our convergence analysis is based on an approach to inﬁnitedimensional problems from [97]. This electronic version is for personal use and may not be duplicated or distributed. n=0 Let { n }∞ ⊂ RN be such that n=0 n n 2 Since xn → x∗ by assumption.12) sn En sn = (Bn − F (x∗ ))sn = −F (xn ) − F (x∗ )sn . Since µn → 0 by assumption. [104]. 116 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS We have. Hence. for n suﬃciently large.2. We will use the l2 norm throughout our discussion.18) T ψn+1 = ψn − θn (ηn ψn )ηn + n is either 1 or 0 for lim η T ψn n→∞ n = 0. Let ψ0 ∈ RN be given. since (7. Combining (7. We begin with a lemma from [104]. F (x∗ )−1 −1 ≤ F (x∗ )sn ≤ En sn + F (xn ) = µn sn + F (xn ) .14) and (7. which is the inexact Newton condition. Buy this book from SIAM at http://www. sn ≤ 2 F (x∗ )−1 F (xn ) . If {ψn }∞ is given by n=1 (7. (7. (7. 2 and let {ηn }∞ be a set of vectors in RN such that ηn n=0 all n. the proof is complete by Theorem 6. 7. and [115].17) then (7. Let 0 < θ < 1 and let ˆ ˆ {θn }∞ ⊂ (θ. That paper was the basis for [168].1.15) implies that (7. ˆ Lemma 7. 2 − θ).com/SIAM/FR16. ηn = 2 F (x∗ )−1 (µn + γ en ).13) holds. Since ηn → 0.Copyright ©1995 by the Society for Industrial and Applied Mathematics. We assume now that n is large enough so that (7.14) Moreover. F (xn ) + F (xn )sn (7. ηn → 0 where < ∞.2.
we can use the fact that ˆ θn (2 − θn ) > θ2 > 0 to show that the sequence {ψn } is bounded in l2 norm by ψ0 ψn+1 2 2 2 and. This electronic version is for personal use and may not be duplicated or distributed.Copyright ©1995 by the Society for Industrial and Applied Mathematics.21) we conclude that ψn+1 2 ≤ µ. From (7. BROYDEN’S METHOD 117 Proof. This inequality is used often in the analysis of quasiNewton methods [63]. Buy this book from SIAM at http://www. 2 ψn 2 ( ψn θn (2 − θn ) 2 − ψn+1 2 + n 2 ). 2a which is valid for a > 0 and b ≤ a. Convergence of the series in (7.html. ˆ θ2 We let M → ∞ to obtain (7. To prove the result for n = 0 we use the inequality (7.19) implies convergence to zero of the terms in the series. which holds even if ψn = 0. in fact.20) a2 − b2 ≤ a − b2 .22) T (ηn ψn )2 ≤ − ψn+1 2 ≤ ψn 2 − T θn (2 − θn )(ηn ψn )2 + 2 ψn 2 n 2. for any M > 0. M n=0 T (ηn ψn )2 ≤ ψ0 2 2 − ψM +1 ˆ θ2 2 2 ≤ ψ0 2 .com/SIAM/FR16.ecsecurehost. From (7. which is (7. 2 ψn 2 ≤ ψn Hence if ψn = 0 (7.18). ≤ ψn ≤ ψn 2 2 2 2 T − θn (2 − θn )(ηn ψn )2 ˆ T − θ2 (ηn ψn )2 ≤ ψn 2 . In that case.20) we conclude that if ψn = 0 then T ψn − θn (ηn ψn )ηn 2 ≤ ψn 2 2 2 T − θn (2 − θn )(ηn ψn )2 T θn (2 − θn )(ηn ψn )2 . .21) Hence (7.19) ∞ n=0 T (ηn ψn )2 < ∞. We ﬁrst consider the case in which n = 0 for all n. 2 Therefore.
In the linear case. sT s sT s If xc and Bc are the current approximations to x∗ = A−1 b and A and s = x+ − xc then y − Bc s = (Ax+ − Axc ) − Bc (x+ − xc ) = −Ec s and therefore (7. Let Bc be nonsingular and let B+ be formed by the Broyden update (7. 2]. ˆ θ2 Hence (7. The reader should compare the statement of the result with that in the nonlinear case. The role of ˆ the parameter θ and the sequence {θn } in Lemma 7. the Broyden update has a very simple form. We will consider a modiﬁed update (7.1 is nicely illustrated by this application of the lemma. is an important part of the convergence analysis. Then (7. and xc ∈ RN be given.2.1 to prove a result from [130] on the convergence of Broyden’s method for linear problems. 7. In this section F (x) = Ax−b and Bn = A+En . s 2 s 2 2 2 Lemma 7.2. ≤ 2µ ˆ θ2 2µ ˆ θ2 M n=0 ( ψn 2 − ψn+1 2 + M n 2) = ψ0 2 − ψM +1 2+ n 2 n=0 ≤ 2µ2 . The standard assumptions in this case reduce to nonsingularity of A.23) B+ = Bc + θc (y − Bc s)sT (Ax+ − b)sT = Bc + θc . .2. Let A be nonsingular.ecsecurehost. We may use Lemma 7. Buy this book from SIAM at http://www.2.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Linear problems. 118 where ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS ∞ i 2 i=0 µ= Hence M T 2 n=0 (ηn ψn ) + ψ0 2 . This property. θc ∈ [0.2.23).25) E+ 2 ≤ Ec 2 .19) holds and the proof is complete.24) B+ = Bc + θc (y − Bc s)sT (Ec s)sT = Bc − θc .1. called bounded deterioration [28]. Our ﬁrst task is to show that the errors in the Broyden approximations to A do not grow.html. This electronic version is for personal use and may not be duplicated or distributed.com/SIAM/FR16.
1) and that θc is always chosen so that B+ is nonsingular.1 for some θ ∈ (0. This electronic version is for personal use and may not be duplicated or distributed. By subtracting A from both sides of (7.Copyright ©1995 by the Society for Industrial and Applied Mathematics.2.2. Assume that {θn } satisﬁes the ˆ hypotheses of Lemma 7. In [130] one suggestion was γc  ≥ σ. 1.27) This completes the proof as E+ 2 Ps = ssT .1.2.2.26) = Ec (I − θc Ps ). 1). Speciﬁcally Proposition 7. Now. For linear problems one can show that the Dennis–Mor´ condition implies e convergence and hence the assumption that convergence takes place is not needed. However the results in [130] assume only that the ˆ sequence {θn } satisﬁes the hypotheses of Lemma 7. 1 − γc where γc = −1 −1 (Bc y)T s (Bc As)T s = 2 s 2 s 2 2 and σ ∈ (0. Buy this book from SIAM at http://www. where Ps is the orthogonal projector (7.1 for some θ ∈ (0. θc can always be selected to make B+ nonsingular.1. Let A be nonsingular. Let A be nonsingular.24) we have E+ = E c − θ c (Ec s)sT s 2 2 (7. If {θn } is such that the matrices Bn obtained by (7. 1 − sign(γc )σ θc = . s 2 2 ≤ Ec 2 I − θ c Ps 2 and I − θc Ps 2 ≤ 1 by orthogonality of Ps and the fact that 0 ≤ θc ≤ 2.2. e We leave the proof as Exercise 7. 1) is ﬁxed.28) −1 xn+1 = xn − Bn (Axn − b) then the Dennis–Mor´ condition implies convergence of {xn } to x∗ = A−1 b.5. . otherwise. BROYDEN’S METHOD 119 Proof.23) are nonsingular and {xn } is given by the modiﬁed Broyden iteration (7. Assume that {θn } satisﬁes ˆ the hypotheses of Lemma 7.28) converges qsuperlinearly to x∗ = A−1 b for every initial iterate x0 . If {θn } is such the matrices Bn obtained by (7.23) are nonsingular then the modiﬁed Broyden iteration (7.1.1 for some θ ∈ (0.html. Theorem 7.com/SIAM/FR16. 1).ecsecurehost. The superlinear convergence result is remarkable in that the initial iterate need not be near the root.
we have n→∞ lim En sn =0 sn 2 which is the Dennis–Mor´ condition. Definition 7. It is known [29].2).2. .html.1) and (7.2. [63].30) T ηn ψ n = T En sn (En φ)T sn = φT → 0.1 and Proposition 7. and xn+1 and Bn+1 can be computed from xn and Bn using (7. [143]. The result here is not a local convergence result.29) holds. we provide a proof based on ideas from [97].2. setting φ = ej . 120 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Proof. x0 . sn 2 2 and. ψn = En φ. Let φ ∈ RN be given. Assuming that (7.2.1. that the Broyden iteration will terminate in 2N steps in the linear case. sn 2 As j was arbitrary. However we must show that the sequence {Bn } of Broyden updates exists and remains nonsingular.com/SIAM/FR16. and we may invoke Lemma 7. This electronic version is for personal use and may not be duplicated or distributed. Our analysis of the nonlinear case is diﬀerent from the classical work [28]. This will imply qsuperlinear convergence e by Theorem 7. shows that the j component of has limit zero. by (7. We say that the Broyden sequence ({xn }. Buy this book from SIAM at http://www. the unit vector in the jth En sn coordinate direction. Bn is nonsingular for all n.Copyright ©1995 by the Society for Industrial and Applied Mathematics.2. [115].26) T T En+1 φ = (I − θn Pn )En φ. We make a deﬁnition that allows us to compactly express this fact. 7. that does not use the Frobenius norm and extends to various inﬁnitedimensional settings. {Bn }) for the data (F. Let Pn = Since for all v ∈ RN and all n T v T (En φ) = φT (En v) sn sT n . and [104].1. T ηn = sn / sn 2 . It is global in the sense that qsuperlinear convergence takes place independently of the initial choices of x and B. [82]. Our approach is to show that for every φ ∈ RN that (7.ecsecurehost.1 with n =0 to conclude that (7. For nonlinear problems our analysis is local and we set θn = 1. [84]. As indicated in the preface. B0 ) exists if F is deﬁned at xn for all n. Nonlinear problems.1. sn 2 sn 2 This completes the proof.29) n→∞ lim φT En sn sn 2 = 0.
Then (7. together with the qlinear convergence e proved earlier completes the proof of local qsuperlinear convergence.2.html.2.com/SIAM/FR16. the errors in the Jacobian approximation can grow as the iteration progresses.2) to obtain E+ = Ec (I − Ps ) + where Ps is is given by (7.32) as F (x+ ) = −Ec s + and then using (7. x0 .2. this growth can be limited and therefore the Broyden sequence exists and the Broyden iterates converges to x∗ at least qlinearly. the sequence { En } need not be monotone nonincreasing. Assume that −1 x+ = xc − Bc F (xc ) = xc + s ∈ Ω and B+ is given by (7. . Let xc ∈ Ω and a nonsingular matrix Bc be given. Finally we verify the Dennis–Mor´ condition (7. This electronic version is for personal use and may not be duplicated or distributed. Note that (7. (F (xc + ts) − F (x∗ ))s dt (∆c s)sT s 2 2 (F (xc + ts) − F (x∗ )) dt. Unlike the linear case. However.ecsecurehost. Let the standard assumptions hold. However in the nonlinear case. which.32) = −F (x+ ) + We begin by writing (7.31) E+ 2 ≤ Ec + γ( ec 2 + e+ 2 )/2 Proof. B0 ) and that the Broyden iterates converge qsuperlinearly to x∗ . Our ﬁrst task is to show that bounded deterioration holds.27) and ∆c = 1 0 1 0 1 0 (F (xc + ts) − F (x∗ ))s dt. Bounded deterioration. We ﬁrst prove a bounded deterioration result like Lemma 7. The development of the proof is complicated. BROYDEN’S METHOD 121 We will show that if the standard assumptions hold and x0 and B0 are suﬃciently good approximations to x∗ and F (x∗ ) then the Broyden sequence exists for the data (F.2. the possible increase can be bounded in terms of the errors in the current and new iterates.7) implies that Ec s = −F (x+ ) + (F (x+ ) − F (xc ) − F (x∗ )s) (7.6).2). We then show that if the initial data is suﬃciently good.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Theorem 7. Buy this book from SIAM at http://www.
2. in the same way to obtain En+1 2 ≤ E0 2+ rj ≤ δB + which completes the proof. This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost. . + γ(1 + r)rn δ/2. .3 states that Broyden’s method is locally qlinearly convergent. Since E0 2 < δB < δ1 we may begin the induction. 2(1 − r) Buy this book from SIAM at http://www. en+1 2 2 and therefore 2 En+1 ≤ En 2 + γ(1 + r) en 2 /2 ≤ En (1 + r)γδ 2 We can estimate En 2 .2. We will prove the result by choosing δB and reducing δ if necessary so that E0 2 ≤ δB will imply that En 2 ≤ δ1 for all n.3 hold. Proof. Theorem 7. . 2 2 2 We show that En 2 ≤ δ1 by induction.html. Let the standard assumptions hold. n j=0 < δ1 . Local qlinear convergence. which will prove the result. Let r ∈ (0.34) δB = δ1 − δ 2 . 1) be given. En−1 2 . x0 . Then. reducing δ and δ1 further if needed. + ∆c Just as in the proof of Lemma 7. . Now assume that Ek 2 < δ1 for all 0 ≤ k ≤ n. where KA is from the statement of Theorem 5. B0 ) exists and xn → x∗ qlinearly with qfactor at most r.2. by the standard assumptions ∆c 2 ≤ γ( ec ≤ Ec 2 + e+ 2 )/2.2 for the linear case E+ and the proof is complete.2.com/SIAM/FR16. To do this reduce δ if needed so that γ(1 + r)δ < δ1 (7. Let δ and δ1 be small enough so that the conclusions of Theorem 5. En+1 Since En 2 2 2 ≤ En ≤ r en 2 + γ( en+1 2 + en 2 )/2.4. The proof is more complex than similar results in Chapters 5 and 6.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 122 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Hence. We must explore the relation between the size of E0 needed to enforce qlinear convergence and the qfactor.3. Theorem 7.2.3. (1 + r)γδ ≤ δ1 .33) δ2 = 2(1 − r) and set (7. the qfactor is at most KA (δ + δ1 ) ≤ r. By Theorem 7. Then there are δ and δB such that if x0 ∈ B(δ) and E0 2 < δB the Broyden sequence for the data (F.4.
Let φ ∈ RN be arbitrary. [31].4. sn 2 sn 2 This. [142]. By Theorem 7.2. Hence Lemma 7. The ﬁnal task is to verify e the Dennis–Mor´ condition. Let δ and δB be such that the conclusions of Theorem 7. clears the storage and starts over. Buy this book from SIAM at http://www.3 and the standard assumptions imply ∆n 2 ≤ γ(1 + r)rn δ/2.2. Then there are δ and δB such that if x0 ∈ B(δ) and E0 2 < δB the Broyden sequence for the data (F. as in the proof of Theorem 7. ηn = sn / sn 2 . and n = Pn ∆T φ. Another approach. Proof. after the storage available for the iteration is exhausted. BROYDEN’S METHOD 123 Veriﬁcation of the Dennis–Mor´ condition. the oldest of the stored vectors is replaced by the most recent.2.Copyright ©1995 by the Society for Industrial and Applied Mathematics. B0 ) exists and xn → x∗ qsuperlinearly.3. Implementation of Broyden’s method The implementation of Broyden’s method that we advocate here is related to one that has been used in computational ﬂuid mechanics [73].2. 7. implies the Dennis–Mor´ condition and e completes the proof. As in the proof of Theorem 7.6) holds e to complete the proof. x0 .1 implies that T ηn ψn = T En sn (En φ)T sn = φT → 0. Let the standard assumptions hold.ecsecurehost. Note that T T En+1 φ = (I − Pn )En φ + Pn ∆T φ.2. n We wish to apply Lemma 7. e Theorem 7. In these methods.3 hold.2.1. restarting. .com/SIAM/FR16. Some such methods are called limited memory formulations in the optimization literature [138]. This electronic version is for personal use and may not be duplicated or distributed. n The hypothesis in the lemma that n n 2 <∞ holds since Theorem 7. sn 2 2 (F (xn + tsn ) − F (x∗ )) dt.1 let Pn = Set ∆n = 1 0 sn sT n .1 we need only show that Dennis–Mor´ condition (7.1.1 with T ψn = En φ.html.2.
Proposition 7. Then the Broyden sequence ({yn }. [177]. B0 .35) for all n. x0 . we can incorporate that into the −1 function evaluation. We begin by showing how the initial approximation to F (x∗ ). Now sn = xn+1 − xn = yn+1 − yn by (7. Bn = B0 Cn (7.1.1. We leave the proof to Exercise 7. We use a restarted approach. This electronic version is for personal use and may not be duplicated or distributed. Our basic implementation is a nonlinear version of the implementation for linear problems in [67].ecsecurehost.4. Lemma 7. (B −1 u)v T (7. B0 ) exists. Our plan will be to store Bn in a compact and easy to apply form.com/SIAM/FR16. x0 . Then B + uv T is invertible if and only if 1 + v T B −1 u = 0. Equation (7.35) holds for a given n.3. (B + uv T )−1 = I − 1 + v T B −1 u The expression (7. [11].37) B −1 . In the context of a sequence of Broyden updates {Bn } we have for n ≥ 0.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Let B be a nonsingular N × N matrix and let u. If (7. The basis for this is the following simple formula [68].3. Hence Bn+1 = Bn + F (xn+1 )sT n sn 2 2 −1 B0 F (yn+1 )sT n = B0 Cn+1 . sn 2 2 = B0 Cn + B0 This completes the proof. If a better choice for B0 exists. Buy this book from SIAM at http://www. The consequence of this proof for implementation is that we can assume that B0 = I. Assume that the Broyden sequence ({xn }.5.37) is called the Sherman–Morrison formula. may be regarded as a preconditioner and incorporated into the deﬁntion of F just as preconditioners were in the implementation of NewtonGMRES in § 6. We proceed by induction.36). v ∈ RN . {Bn }) for the data (F. then −1 −1 yn+1 = yn − Cn B0 F (yn ) (7.36) −1 = xn − (B0 Cn )−1 F (xn ) = xn − Bn F (xn ) = xn+1 . In this case. T Bn+1 = Bn + un vn . .35) holds by deﬁnition when n = 0. though a limited memory approach could be used as well. I) and xn = yn . {Cn }) exists for the data −1 (B0 F.2.html. [176]. Proof. 124 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS This latter approach is similar to the restarted GMRES iteration.
−1 = (I − w T T T Bn n−1 vn−1 )(I − wn−2 vn−2 ) .e.40) implies that for n ≥ 1 T −1 sn = −w + Cw w(vn−1 w) = w(−1 + Cw (Cw − sn−1 2 )) = −Cw w sn−1 Hence. Moreover.39) (I − wj vj )F (xn ). (I − w0 v0 ) 2 and vn = sn / sn 2 . −1 Hence the action of Bn on F (xn ) (i. the Broyden step for the following iteration is n−1 −1 T (7.38) = n−1 j=0 T (I − wj vj ). Since the empty matrix product is the identity. This electronic version is for personal use and may not be duplicated or distributed. if B0 = I. Therefore. −1 T −1 wn = (Bn un )/(1 + vn Bn un ) (7. one need only store the steps {sn } and their norms to construct the sequence {wn }. vj }n−1 at a cost of O(N n) j=0 ﬂoatingpoint operations. sn = −Bn F (xn ) = − j=0 Since the product n−2 j=0 T (I − wj vj )F (xn ) must also be computed as part of the computation of wn−1 we can combine the computation of wn−1 and sn as follows: w (7. we can write (7. the computation of the Broyden step) can be computed from the 2n vectors {wj .html.Copyright ©1995 by the Society for Industrial and Applied Mathematics. . . One may carry this one important step further [67] and eliminate the need to store the sequence {wn }. .41) 2 = − sn−1 2 wn−1 .42) −1 Bn = n−1 j=0 I+ sj+1 sT j sj 2 2 . wn = −sn+1 / sn 2 .ecsecurehost. Note that (7.38) as (7. (7.38) is valid for n ≥ 0.40) = n−2 j=0 T (I − wj vj )F (xn ) 2 T + vn−1 w)−1 wn−1 = Cw w where Cw = ( sn−1 sn T = −(I − wn−1 vn−1 )w.. BROYDEN’S METHOD 125 where un = F (xn+1 )/ sn Setting we see that. In fact. for n ≥ 0. (7.com/SIAM/FR16. Buy this book from SIAM at http://www.
Our implementation is similar to GMRES(m). r0 = F (x) 2 .43) for sn+1 to obtain (7. is of the same order as GMRES. restarting with B0 when storage is exhausted.44) −1 1 + sT Bn F (xn+1 )/ sn n 2 2 T −1 = 1 + vn Bn un is nonzero unless Bn+1 is singular. of n + O(1) vectors for the nth iterate. Algorithm 7. the denominator in (7. for j = 0. τ. F. adding to the input list a limit nmax on the number of Broyden iterations taken before a restart and a limit maxit on the number of nonlinear iterations.43) =− I+ sn+1 sT n sn 2 2 sn+1 sT n sn 2 2 n−1 j=0 I+ sj+1 sT j 2 sj 2 F (xn+1 ) −1 Bn F (xn+1 ). . n = −1.ecsecurehost.42) and (7. this is consistent with setting the empty matrix product to the identity.42) and (7.1. The storage requirement. we solve (7. z = −F (x) ii.39) directly to compute sn+1 because sn+1 appears on both sides of the equation sn+1 = − I + (7. brsol(x. Instead. Our notation and algorithmic framework are based on [67].44) to describe an algorithm. Note that B0 = I is implicit in the algorithm.3. nmax) 1.com/SIAM/FR16.6. By Proposition 7. itc = 0.html. This electronic version is for personal use and may not be duplicated or distributed.1. making Broyden’s method competitive in storage with GMRES as a linear solver [67]. We can use (7. 2. maxit. We use a restarting approach in the algorithm. (e) if n < nmax then i. itc = itc + 1 (b) x = x + sn (c) Evaluate F (x) (d) If F (x) 2 ≤ τr r0 + τa exit.3. 126 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS We cannot use (7. s0 = −F (x). Do while itc < maxit (a) n = n + 1.44) sn+1 = − −1 Bn F (xn+1 ) −1 1 + sT Bn F (xn+1 )/ sn n 2 2 .Copyright ©1995 by the Society for Industrial and Applied Mathematics. The termination criteria are the same as for Algorithm nsol in § 5. n − 1 z = z + sj+1 sT z/ sj j 2 2 Buy this book from SIAM at http://www. Our looping convention is that the loop in step 2(e)ii is skipped if n = 0.
and solution as the example in § 3. and in several papers on inﬁnitedimensional problems as well [91]. in terms of both time and storage. not compute snmax+1 . z. than the dense matrix approach proposed in [85]. For linear problems. If the data is suﬃciently good.33) preconditioned with the Poisson solver fish2d.1. We compare Broyden’s method without restarts (solid line) to Broyden’s method with restarts every three iterates (dashed line).4. and F (x).7. bounded deterioration implies that the Broyden matrices will be well conditioned and superlinear convergence implies that only a few iterates will be needed. . If n < nmax. which computes xn and sn+1 . Examples for Broyden’s method Buy this book from SIAM at http://www. The theory in this chapter indicates that Broyden’s method can be viewed as an alternative to GMRES as an iterative method for nonsymmetric linear equations. 7. x.com/SIAM/FR16. The restarted iteration required 24 iterations and 3. where z and F (x) may occupy the same j=0 storage location. This point has been explored in some detail in [67]. [104]. The Sherman–Morrison approach in Algorithm brsol is more eﬃcient.33) from § 3. where more elaborate numerical experiments that those given here are presented. a slightly higher cost than the GMRES solution. sn+1 = z/(1 − sT z/ sn 2 ) n 2 (f) if n = nmax then n = −1. 7. We use the MATLAB code brsol. If n = nmax. Linear problems. coeﬃcients. Broyden’s method terminated after 9 iterations at a cost of roughly 1. This electronic version is for personal use and may not be duplicated or distributed. √ In all the examples in this section we use the l2 norm multiplied by 1/ N . we allow for increases in the nonlinear residual. then the nth iteration of brsol. BROYDEN’S METHOD 127 iii. In Fig. especially for large problems. Our implementation of brsol in the collection of MATLAB codes stores z and F separately in the interest of clarity and therefore requires nmax + 3 vectors of storage.Copyright ©1995 by the Society for Industrial and Applied Mathematics. requires O(nN ) ﬂoatingpoint operations and storage of n + 3 vectors. This requirement of one more vector than the algorithm in [67] needed is a result of the nonlinearity. the iteration will restart with xnmax .ecsecurehost. However the dense matrix approach makes it possible to detect ill conditioning in the approximate Jacobians.4. One can also see highly irregular performance from the 7.html.5 million ﬂoating point operations.1 we present results for (3. and therefore need storage for nmax + 2 vectors. We set τa = τr = h2 where h = 1/32.5 million ﬂoating point operations.7. righthand side. [120]. As an example we consider the linear PDE (3. the steps {sj }n . we feel that the approach of Algorithm brsol is the best approach. s = −F (x). A MATLAB implementation of brsol is provided in the collection of MATLAB codes. In view of all this. We use the same mesh.
7. Broyden’s method for the linear PDE.4 illustrate the diﬀerences in the two methods. The cost in function evaluations for Broyden’s method is one for each nonlinear iteration and for NewtonGMRES one for each outer iteration and one for each inner iteration. The examples in this section. and the number of nonlinear iterations is large. Broyden’s method could be at a disadvantage as it might need to be restarted many times. In the linear case [67] one can store only the residuals and recover the terminal iterate upon exit. if function evaluations are very expensive. however. We compare Broyden’s method without restarts (solid line) to Broyden’s method with restarts every three iterates (dashed line).com/SIAM/FR16. Buy this book from SIAM at http://www.3).4. Nonlinear problems. Hence. Chandrasekhar Hequation. as well as those in § 8.9999 (Fig.9 both methods required six iterations for convergence.ecsecurehost. and the sequence of steps saves a vector over the nonlinear case. As before we solve the Hequation on a 100point mesh with c = . Broyden’s method and NewtonGMRES both require more storage as iterations progress. 7. This electronic version is for personal use and may not be duplicated or distributed. Our general recommendation is to try both. 10 1 10 Relative Residual 0 10 1 10 2 10 3 0 5 10 Iterations 15 20 25 Fig.2. We consider the Hequation from Chapters 5 and 6 and a nonlinear elliptic partial diﬀerential equation. NewtonGMRES as the inner iteration progresses and Broyden’s method as the outer iteration progresses.1. the vector z. If storage is at a premium. 7.html.Copyright ©1995 by the Society for Industrial and Applied Mathematics.2) and c = . the cost of a solution by Broyden’s method could be much less than that of one by NewtonGMRES. 128 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS restarted method. For c = . The implementation without restarts required 153 thousand ﬂoatingpoint operations and the restarted method 150.9 (Fig. . 7. Storage of the residual. a substantial improvement over the NewtonGMRES implementation.
One should take some care in drawing general conclusions. Broyden’s method for the Hequation. the storage in NewtonGMRES is used in the inner iteration. If many nonlinear iterations are needed.html.ecsecurehost. it may be the case Buy this book from SIAM at http://www.9. Even so.Copyright ©1995 by the Society for Industrial and Applied Mathematics.9999. 7. 10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 10 6 10 7 0 2 4 6 8 10 Iterations 12 14 16 18 Fig. While Broyden’s method on this example performed better than NewtonGMRES.3. The unrestarted iteration converges in 10 iterations at a cost of roughly 249 thousand ﬂoatingpoint operations. The restarted iteration requires 18 iterations and 407 ﬂoatingpoint operations. while that in Broyden in the nonlinear iteration. 7.com/SIAM/FR16. 10 10 10 10 10 10 10 10 10 0 1 2 Relative Nonlinear Residual 3 4 5 6 7 8 0 1 2 3 Iterations 4 5 6 Fig. This electronic version is for personal use and may not be duplicated or distributed. .9999. both implementations performed better than NewtonGMRES. c = . c = . the restarted iteration has much more trouble. BROYDEN’S METHOD 129 For c = . Broyden’s method for the Hequation.2.
F (xc + hv)) to accommodate the NewtonGMRES iteration. however it can be quite successful in practice. . as one would if the problem were truly linear. Therefore the GMRES inner iteration for the NewtonGMRES approach needed storage for at most 10 vectors (the Krylov basis. The unrestarted Broyden iteration converges in 12 iterations at a cost of roughly 2. 10 1 10 Relative Nonlinear Residual 0 10 1 10 2 10 3 10 4 0 2 4 6 8 Iterations 10 12 14 16 Fig.4. We consider the nonlinear convectiondiﬀusion equation from § 6. The Broyden iteration when restarted every n iterates.ecsecurehost. u0 = 0. We set C = 20.html. Hence restarting the Broyden iteration every 8 iterations most closely corresponds to the NewtonGMRES requirements. Broyden’s method for nonlinear PDE. This is not supported by the local convergence theory for nonlinear problems.4.2) of 2. x. This is a modest improvement over the NewtonGMRES cost (reported in § 6.2 million ﬂoatingpoint operations. 7. At most ﬁve inner GMRES iterations were required.Copyright ©1995 by the Society for Industrial and Applied Mathematics. One can do this in brsol by setting the third entry in the parameter list to 1. This approach Buy this book from SIAM at http://www. 7. and τr = τa = h2 . This electronic version is for personal use and may not be duplicated or distributed. s. F (xc ).4. the residual norms do not monotonically decrease in the Broyden iteration (we ﬁx this in § 8.6 million ﬂoatingpoint operations.2. We use the same mesh width h = 1/32 on a uniform square grid.com/SIAM/FR16. However.2) and more storage was used.4. xc .4 we compare Broyden’s method without restarts (solid line) to Broyden’s method with restarts every 8 iterates (dashed line). 130 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS that Broyden’s method may need restarts while the GMRES iteration for the linear problem for the Newton step may not. requires storage of n + 2 vectors {sj }n−1 . In Fig.3. which happened on the second iterate. We preconditioned with the Poisson solver fish2d.3 we will return to this example. and z (when stored in the j=0 same place as F (x)). In § 8. Convectiondiﬀusion equation. We allowed for an increase in the residual.
BROYDEN’S METHOD 131 took 15 iterations to terminate at a cost of 2. but the convergence was extremely irregular after the restart.ecsecurehost.html.com/SIAM/FR16.4 million ﬂoatingpoint operations.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Buy this book from SIAM at http://www. . This electronic version is for personal use and may not be duplicated or distributed.
so named because of its inferior performance in practice [63]. 7. U is N × M .5.5.5.8.5. The bad Broyden method [26].5. V is M × N . initial iterates. 7. Precondition with the Poisson solver fish2d. and values for the convection coeﬃcient d. Duplicate the results reported in § 7.Copyright ©1995 by the Society for Industrial and Applied Mathematics.5. 132 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 7. 7. Exercises on Broyden’s method 7. Experiment with various mesh widths.1 by showing that if A is a nonsingular N × N matrix.9.7.3. Try to solve the nonlinear convectiondiﬀusion equation in § 6.5.3.5. If this is the case. Extend Proposition 7. 7.1.2. Set τa = τr = 10−8 . Solve the Hequation with Broyden’s method and c = 1. [199].4.3.5. Vary the parameters in the equations and the number of vectors stored by Broyden’s method and report on their eﬀects on performance. What happens in the diﬀerential equation examples if preconditioning is omitted? 7. 1) with homogeneous Dirichlet boundary conditions.5.6. How does the performance of Broyden’s method diﬀer from NewtonGMRES? 7. Prove Proposition 7. 1) × (0.html.5.2.ecsecurehost. This electronic version is for personal use and may not be duplicated or distributed. See [102] for further generalizations. updates an approximation to the inverse of the Jacobian at the root so that B −1 ≈ F (x∗ )−1 satisﬁes the inverse secant equation −1 B+ y = s (7.1. How does this compare with nsol on the examples in Chapter 5. This is the Sherman–Morrison–Woodbury formula [68]. 7. Prove Proposition 7. Modify nsol to use Broyden’s method instead of the chord method after the initial Jacobian evaluation. What qfactor for linear convergence do you see? Can you account for this by applying the secant method to the equation x2 = 0? See [53] for a complete explanation. then (A + U V )−1 = A−1 − A−1 U (I + V A−1 U )−1 V A−1 . 7.45) Buy this book from SIAM at http://www.4. . Compare the performance of Broyden’s method and NewtonGMRES on the Modiﬁed Bratu Problem −∇2 u + dux + eu = 0 on (0.4 with Broyden’s method using various values for the parameter C.com/SIAM/FR16.1. then A + U V is nonsingular if and only if I + V A−1 U is a nonsingular M × M matrix.
3. Discuss the diﬀerences in implementation from Broyden’s method. and [112].11.12. How would the Schubert algorithm be aﬀected by preconditioning? Compare your analysis with those in [158].com/SIAM/FR16. and main diagonal are updated.5. References [101].10. Does anything go wrong? See [67] for more about this. Use your result from Exercise 5. 7. apply it to the examples in § 7. and compare the performance to that of Broyden’s method.ecsecurehost.5. This electronic version is for personal use and may not be duplicated or distributed. are relevant to this exercise. and [189]. [125]. What can you conclude from your observations? Buy this book from SIAM at http://www.5. y 2 2 Show that the bad Broyden method is locally superlinearly convergent if the standard assumptions hold. For problems with tridiagonal Jacobian. [92].5. j ≤ N and i − j ≤ 1. Does the relative performance of the methods change as h is decreased? This is interesting even in one space dimension where all matrices are tridiagonal. for example. Read about the algorithm in [172] and prove local superlinear convergence under the standard assumptions and the additional assumption that the sparsity pattern of B0 is the same as that of F (x∗ ). Note that only the subdiagonal. BROYDEN’S METHOD 133 with the rankone update −1 −1 B+ = Bc + −1 (s − Bc y)y T .14 in Chapter 5 to numerically estimate the qorder of Broyden’s method for some of the examples in this section (both linear and nonlinear). Try to implement the bad Broyden method in a storageeﬃcient manner using the ideas from [67] and § 7.8) and compare it to NewtonGMRES and Broyden’s method. Implement the Schubert algorithm on an unpreconditioned discretized partial diﬀerential equation (such as the Bratu problem from Exercise 7.5. [93].4.7. 7. [27] is a quasiNewton update for sparse matrices that enforces both the secant equation and the sparsity pattern. Implement the bad Broyden method. 7. The Schubert or sparse Broyden algorithm [172]. superdiagonal. 7.Copyright ©1995 by the Society for Industrial and Applied Mathematics. .13. the update is (y − Bc s)i sj (B+ )ij = (Bc )ij + i+1 2 k=i−1 sk for 1 ≤ i.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. 134 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Buy this book from SIAM at http://www.com/SIAM/FR16.ecsecurehost.html. . This electronic version is for personal use and may not be duplicated or distributed.
but the length of s is too large. a failure of Newton’s method that results from an inaccurate initial iterate. [173]. can be used to accomplish the same objective. [154]. 9. [25]. We begin with single equations. The implementation and analysis of the line search method we develop in this chapter do not depend on whether iterative or direct methods are used to compute the Newton step or upon how or if the Jacobian is computed and stored. If we look more closely. we see that the Newton step s = −f (xc )/f (xc ) is pointing toward the root in the sense that sxc < 0. [24]. −138. [88]. We can see this eﬀect in the sequence of iterates: 10.9 × 104 . and because it is trivial to add a line search to an existing locally convergent implementation of Newton’s method. [159].ecsecurehost. [2]. Of the many such algorithms we will focus on the class of line search methods and the Armijo rule [3]. [70]. f (x) = arctan(x) ≈ ±π/2. where the algorithms can be motivated and intuition developed with a very simple example. implemented inexactly [24]. This observation motivates the simple ﬁx of reducing the size of the step until the size of the nonlinear residual is decreased. −1.html.5 × 109 . [63]. The reason for this is that f (x) = (1 + x2 )−1 is small for large x. [181]. [25]. and hence the magnitude of the Newton step is much larger than that of the iterate itself. [198].Copyright ©1995 by the Society for Industrial and Applied Mathematics.1.com/SIAM/FR16. We select the line search paradigm because of its simplicity. A prototype algorithm is given below. [109]. [70]. 2. This electronic version is for personal use and may not be duplicated or distributed. 8. Other methods. and continuation/homotopy methods [1]. Single equations If we apply Newton’s method to ﬁnd the root x∗ = 0 of the function f (x) = arctan(x) with initial iterate x0 = 10 we ﬁnd that the initial iterate is too far from the root for the local convergence theory in Chapter 5 to be applicable. such as trust region [153]. 135 Buy this book from SIAM at http://www. Chapter 8 Global Convergence By a globally convergent algorithm we mean an algorithm with the property that for any initial iterate the iteration either converges to a root of F or fails to do so in one of a small number of ways.9 × 1017 . [129]. .
Do while f (x) > τr r0 + τa (a) If f (x) = 0 terminate with failure. −6 × 10−10 . One can see that most of the work is in the initial phase of the iteration. However. 3. decrease in the size of the nonlinear residual was obtained with a full Newton step. This electronic version is for personal use and may not be duplicated or distributed. some authors [63] refer to this as backtracking.Copyright ©1995 by the Society for Industrial and Applied Mathematics. xc − f (xc )/f (xc )) to ﬁnd a decrease in f (xc ). −1. the minimum number of two function evaluations per iterate is required. Algorithm lines1 almost always works well. . . 2. and 4 required 3. we obtain the sequence 10.8.4. Iterates 1. 136 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Algorithm 8.html. The quadratic convergence behavior of Newton’s method is only apparent very late in the iteration. After the fourth iterate. This is typical of the performance of the algorithms we discuss in this chapter.3.99. The outer iterations are identiﬁed with circles as in § 6. however. 3. We plot arctan(x)/arctan(x0 ) as a function of the number of function evaluations with the solid line.4. where the local theory is valid. −8. 2. −0.1. In theory.1. In the terminal phase. (b) s = −f (x)/f (x) (search direction) (c) xt = x + s (trial point) (d) If f (xt ) < f (x) then x = xt (accept the step) else s = s/2 goto 2c (reject the step) When we apply Algorithm lines1 to our simple example.ecsecurehost. One does not set x+ = x + s without testing the step to see if the absolute value of the nonlinear residual has been decreased. τ ) 1. −3. 1. As we move from the right endpoint to the left. even in this terminal phase rapid convergence takes place only in the ﬁnal three outer iterations. We call this method a line search because we search along the line segment (xc . f. and 2 steplength reductions. 9 × 10−4 .com/SIAM/FR16.1.5. 4.9. 8.56. r0 = f (x) 2. Note the diﬀerences between Algorithm lines1 and Algorithm nsol. −. We plot the progress of the iteration in Fig. lines1(x. there is the possibility of oscillation about a solution without convergence [63].1. To remedy this and to make analysis possible we replace the test for simple decrease with one for suﬃcient decrease.2. 1. The algorithm is to compute a search direction d which for us will be the Newton direction d = −f (xc )/f (xc ) Buy this book from SIAM at http://www.
from [3] (see also [88]) is called the Armijo rule. 8. (b) d = −f (x)/f (x) (search direction) (c) λ = 1 i. Algorithm 8.com/SIAM/FR16. This electronic version is for personal use and may not be duplicated or distributed.1) f (xc + λd) < (1 − αλ)f (xc ). xt = x + λd (trial point) ii. The condition in (8. We follow the recent optimization literature [63] and set α = 10−4 .2. 1) is a small. nsola1(x. r0 = f (x) 2. with λ = 2−j for some j ≥ 0. Newton–Armijo iteration for inverse tangent. number intended to make (8. . something we did not have to do in the earlier chapters.html. Once suﬃcient decrease has been obtained we accept the step s = λd.2. We remark here that there is nothing critical about the reduction factor of 2 in the line search. GLOBAL CONVERGENCE 0 137 10 10 10 Relative Nonlinear Residual 10 10 10 10 10 10 10 10 1 2 3 4 5 6 7 8 9 10 0 5 10 15 20 Function Evaluations 25 30 35 Fig.Copyright ©1995 by the Society for Industrial and Applied Mathematics. A factor of Buy this book from SIAM at http://www. but positive. and then test steps of the form s = λd. f. If f (xt ) < (1 − αλ)f (x) then x = xt (accept the step) else λ = λ/2 goto 2(c)i (reject the step) This is essentially the algorithm implemented in the MATLAB code nsola. Do while f (x) > τr r0 + τa (a) If f (x) = 0 terminate with failure. We will analyze Algorithm nsola in § 8. Note that we must now distinguish between the step and the Newton direction. until f (x + s) satisﬁes (8. The parameter α ∈ (0. τ ) 1.1.1) as easy as possible to satisfy.1.ecsecurehost.1) is called suﬃcient decrease of f . This strategy.
3.3. This could be quite important if the search direction were determined by an inexact Newton method or an approximation to Newton’s method such as the chord method or Broyden’s method. 1.2) is motivated by the Newtoniterative paradigm. τ. Analysis of the Armijo rule In this section we extend the onedimensional algorithm in two ways before proceeding with the analysis. We allow for any choice that produces a reduction that satisﬁes σ0 λold ≤ λnew ≤ σ1 λold . Do while F (x) > τr r0 + τa (a) Find d such that F (x)d + F (x) ≤ η F (x) If no such d can be found.1.1 and . Buy this book from SIAM at http://www. where 0 < σ0 < σ1 < 1. Typical values of σ0 and σ1 are . the iteration may stagnate as a result. . we accept any direction that satisﬁes the inexact Newton condition (6.2.5. In Exercise 8. Taking full Newton steps ensures fast local convergence. Algorithm 8. the analysis here applies equally to methods that solve the equations for the Newton step directly (so ηn = 0).1). Our algorithm nsola incorporates these ideas. terminate with failure. in fact. 8. is to minimize an interpolating polynomial based on the previous trial steps. This electronic version is for personal use and may not be duplicated or distributed. reducing λ by too much can be costly as well.html. 138 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 10 could well be better in situations in which small values of λ are needed for several consecutive steps (such as our example with the arctan function).Copyright ©1995 by the Society for Industrial and Applied Mathematics.1 [54].2). [133]. In that event a reduction of a factor of 8 would require three passes through the loop in nsola1 but only a single reduction by a factor of 10. The parameter σ0 safeguards against that. Taking as large a fraction as possible helps move the iteration into the terminal phase in which full steps may be taken and fast convergence expected. [87]. First. Our second extension is to allow more ﬂexibility in the reduction of λ. F. On the other hand.2) F (xn )dn + F (xn ) ≤ ηn F (xn ) . We will return to the issue of reduction in λ in § 8.3. approximate the Newton step by a quasiNewton method. r0 = F (x) 2.2.9 you are asked to implement the Armijo rule in the context of Algorithm nsol. or even use the chord method provided that the resulting direction dn satisﬁes (8.5. [86]. nsola(x. [63]. η).ecsecurehost.com/SIAM/FR16. Safeguarding is an important aspect of many globally convergent methods and has been used for many years in the context of the polynomial interpolation algorithms we discuss in § 8. developed in § 8. The danger is that the minimum may be too near zero to be of much use and. While (8. One such method. We write this for the Armijo sequence as (8.
xn → ∞. Note how the safeguard bound σ0 enters this result. as do [25] and [70]. If F (xt ) < (1 − αλ) F (x) then x = xt else Choose σ ∈ [σ0 . Let x0 ∈ RN and α ∈ (0.ecsecurehost. The algorithm can fail in some obvious ways: 1.2. 1 − α) Buy this book from SIAM at http://www. for example. If. n=0 Assumption 8.2. Let {xn } be the iteration produced by Algorithm nsola with initial iterate x0 . step 2a were implemented with a direct factorization method. r) = ∞ {x  x − xn ≤ r}. γ. by giving a uniform lower bound on λn (assuming that F (x0 ) = 0). ¯ 3. The inner iteration could terminate with a failure.Copyright ©1995 by the Society for Industrial and Applied Mathematics. full steps are taken in the terminal phase. Assume that {xn } is given by Algorithm nsola with (8.1. 2. We begin with a formal statement of our assumption that F is uniformly Lipschitz continuous and bounded away from zero on the sequence {xn }. . Note also the very modest conditions on ηn . Assumption 8.2. 1) be given.2).com/SIAM/FR16. There are r. xt = x + λd ii. xn → x. F is Lipschitz continuous with Lipschitz constant γ. a local minimum of F which is not a root. the algorithm must fail. mf > 0 such that F is deﬁned. Lemma 8. We will see that if F has no roots then either F (xn ) has a limit point which is singular or {xn } becomes unbounded. σ1 ] λ = σλ goto 2(b)i Note that step 2a must allow for the possibility that F is illconditioned and that no direction can be found that satisﬁes (8. and the convergence is qquadratic.1. then the iteration converges to a root of F at which the standard assumptions hold. η ] ⊂ (0. We show this. This electronic version is for personal use and may not be duplicated or distributed. The convergence theorem is the remarkable statement that if {xn } and { F (xn )−1 } are bounded (thereby eliminating premature failure and local minima of F that are not roots) and F is Lipschitz continuous in a neighborhood of the entire sequence {xn }. and F (x)−1 ≤ mf on the set Ω({xn }.3) ¯ {ηn } ⊂ (0. F (xn ) is singular for some n. GLOBAL CONVERGENCE 139 (b) λ = 1 i.html.1 implies that the line search phase (step 2b in Algorithm nsola) will terminate in a number of cycles that is independent of n. illconditioning of F might be detected as part of the factorization. Clearly if F has no roots.
by the bound ηn ≤ η .2) we have.com/SIAM/FR16. for any λ ∈ [0.Copyright ©1995 by the Society for Industrial and Applied Mathematics.2. by Assumption 8. ¯ Lipschitz continuity of F implies that if λ ≤ λ1 then F (xn + λdn ) ≤ (1 − λ) F (xn ) + λ¯ F (xn ) + η ≤ (1 − λ) F (xn ) + λ¯ F (xn ) + η γ 2 λ dn 2 2 1 0 (F (xn + tλdn ) − F (xn ))λdn dt (F (xn + tλdn ) − F (xn ))λdn dt. Proof. Then either F (x0 ) = 0 or (8. 140 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS and that Assumption 8. Now we can show directly that the sequence of Armijo iterates either is unbounded or converges to a solution. This electronic version is for personal use and may not be duplicated or distributed.2. Let n ≥ 0. ¯ The step acceptance condition in nsola implies that { F (xn ) } is a decreasing sequence and therefore dn = F (xn )−1 (ξn − F (xn )) ≤ mf (1 + η ) F (xn ) . which will be implied by ¯ ¯ λ ≤ λ2 = min λ1 .ecsecurehost. Theorem 8. mf F (x0 ) m2 F (x0 ) (1 + η )2 γ ¯ f . 2(1 − α − η ) ¯ (1 + η )2 m2 γ F (x0 ) ¯ f . 1] F (xn + λdn ) = F (xn ) + λF (xn )dn + = (1 − λ)F (xn ) + λξn + where. ¯ ξn ≤ ηn F (xn ) ≤ η F (xn ) .1 F is Lipschitz continuous on the line segment [xn . The Buy this book from SIAM at http://www.1 holds. .4) ¯ λn ≥ λ = σ0 min r 2(1 − α − η ) ¯ . ¯ Therefore.html. which completes the proof. xn + λdn ] whenever ¯ λ ≤ λ1 = r/(mf F (x0 ) ). By the fundamental theorem of calculus and (8.2. 1 0 ¯ m2 γ(1 + η )2 λ2 F (xn ) f 2 ¯ m2 γλ(1 + η )2 F (x0 ) f 2 2 ≤ (1 − λ + η λ) F (xn ) + λ F (xn ) ¯ < (1 − αλ) F (xn ) .1 says even more. ¯ This shows that λ can be no smaller than σ0 λ2 .
and the convergence behavior in the ﬁnal phase of the iteration is that given by the local theory for inexact Newton methods (Theorem 6. Since F is continuous. F (x∗ ) = 0.2.5.2).com/SIAM/FR16.1 is violated) or the iteration becomes unbounded. Theorem 8. full steps are taken for n suﬃciently large. The boundedness of the sequence {xn } implies that a subsequence {xnk } converges. {xn } is bounded. say to x∗ . then we are done because Assumption 8. for example.2.1 implies that F (xn ) converges qlinearly to zero with qfactor at ¯ most (1 − αλ). We use the l2 norm in that code and in the numerical results reported in § 8.9 you are asked to do this with nsol and. there is δ such that if x ∈ B(δ). In Exercise 8. then the storage requirements are roughly the same as for Algorithm nsolgm if one reuses the storage location for the ﬁnite diﬀerence directional derivative to test for suﬃcient decrease.1. This electronic version is for personal use and may not be duplicated or distributed. {ηn } satisﬁes (8. in principle. Since the standard assumptions hold at x∗ . numerically determine if the chord direction satisﬁes the inexact Newton condition. and that Assumption 8. 8.1. Hence as soon as xnk ∈ B(δ). Thus far we have used for d the inexact Newton direction. Then {xn } converges to a root x∗ of F at which the standard assumptions hold.2 hold is (8. Let x0 ∈ RN and α ∈ (0. Proof. Eventually xnk − x∗  < r. so in the terminal phase of the iteration the step lengths are all 1 and the convergence is qquadratic.3.2. Assume that {xn } is given by Algorithm nsola. which is easy to verify.1.1 implies that the standard assumptions hold at x∗ = xn .ecsecurehost. will remain in B(δ) and converge to x∗ . the entire sequence. with each iteration.2). It could well be advantageous to use a chord or Broyden direction.html.1 holds. If F (xn ) = 0 for some n. the Newton iteration with x0 = x will remain in B(δ) and converge qquadratically to x∗ . At this stage we have shown that if the Armijo iteration fails to converge to a root either the continuity properties of F or nonsingularity of F break down as the iteration progresses (Assumption 8. All that one needs to make the theorems and proofs in § 8.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Otherwise Lemma 8.2 applies and hence full steps can be taken. Implementation of the Armijo rule The MATLAB code nsola is a modiﬁcation of nsolgm which provides for a choice of several Krylov methods for computing an inexact Newton direction and globalizes Newton’s method with the Armijo rule.3).2. The iteration can be very costly if the Jacobian must be Buy this book from SIAM at http://www.1 and therefore the standard assumptions hold at x∗ . via a forward diﬀerence approximation to F (xn )d.2. GLOBAL CONVERGENCE 141 sequence converges to a root at which the standard assumptions hold. 1) be given. Theorem 6. .4. Moreover. where r is the radius in Assumption 8. This is all that one could ask for in terms of robustness of such an algorithm. If GMRES is used as the linear solver. In this case the storage requirements are roughly the same as for Algorithm nsol.2. not just the subsequence.
. In practice such methods usually perform better than constant reduction factors. The second.Copyright ©1995 by the Society for Industrial and Applied Mathematics. In this section we consider an elaboration of the simple line search scheme of reduction of the steplength by a ﬁxed factor. F (xn + λ1 dn ) 2 .3. In this approach we use values of f and f at λ = 0 and the value of f at the current value of λ to construct a 2nd degree interpolating polynomial for f .6) f (0) = 2(F (xn )T dn )T F (xn ) = 2F (xn )T (F (xn )dn ). but might respond well to a more aggressive steplength reduction (by factors of 1/10. we compute the minimum λt of that polynomial analytically and set σ0 λc if λt < σ0 λc . how Broyden’s method must be implemented. 142 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS evaluated and factored many times during the phase of the iteration in which the approximate solution is far from the root. . If we have rejected k steps we have in hand the values F (xn ) 2 . is again based on [67] and is more specialized. If one can model the decrease in F 2 as the steplength is reduced. This electronic version is for personal use and may not be duplicated or distributed. F (xn + λk−1 dn ) 2 . . if λt > σ1 λc . (8.10. 8. Evaluation of F (xc )dc if the full step is not accepted (here dc is the Broyden direction) would not only verify the inexact Newton condition but provide enough information for a twopoint parabolic line search. otherwise. Our Broyden–Armijo code does not explicitly check the inexact Newton condition and thereby saves an evaluation of F . We consider two topics in this section.com/SIAM/FR16. After λc has been rejected and a model polynomial computed. Clearly f (0) can be computed as 2 (8. Buy this book from SIAM at http://www. say).html.ecsecurehost. f (0) = F (xn ) 2 is known.5) λ+ = σ 1 λc λt Twopoint parabolic model. polynomial interpolation as a way to choose λ. The motivation for this is that some problems respond well to one or two reductions in the steplength by modest amounts (such as 1/2) and others require many such reductions. The ﬁrst. is applicable to all methods. We can use this iteration history to model the scalar function f (λ) = F (xn + λdn ) 2 2 with a polynomial and use the minimum of that polynomial as the next steplength. one might expect to be able to better estimate the appropriate reduction factor.1.5. The reader is asked to pursue this in Exercise 8. . Polynomial line searches. We consider two ways of doing this that use second degree polynomials which we compute using previously computed information.
In any case. f at 0 and p(λc ) = f (λc ) is p(λ) = f (0) + f (0)λ + cλ2 . If p (0) = 2 (λ− (f (λc ) − f (0)) − λc (f (λ− ) − f (0))) λc λ− (λc − λ− ) is positive. Buy this book from SIAM at http://www.5). GLOBAL CONVERGENCE 143 Use of (8.6) requires evaluation of F (xn )dn . If it is not.Copyright ©1995 by the Society for Industrial and Applied Mathematics. σ1 = . λ− is p(λ) = f (0) + λ λc − λ− (λ − λ− )(f (λc ) − f (0)) (λc − λ)(f (λ− ) − f (0)) + . it may be necessary to compute a new search direction. If p (0) ≤ 0 one could either set λ+ to be the minimum of p on the interval [σ0 λ. then we set λt to the minimum of p λt = −p (0)/p (0) and apply the safeguarding step (8. we implement this method with σ0 = . We then compute λ+ with (8. We take the latter approach and set λ+ = σ1 λc . Threepoint parabolic model. If the full step is rejected. This is a possibility with directions other than inexact Newton directions. . p agree with f. where λc and λ− are the most recently rejected values of λ. The MATLAB code parab3p implements the threepoint parabolic line search and is called by the two codes nsola and brsola. which uses f (0) and the two most recently rejected steps to create the parabola. and f (λ− ). λ2 c Our approximation of f (0) < 0. which can be obtained by examination of the ﬁnal residual from GMRES or by expending an additional function evaluation to compute F (xn )dn with a forward diﬀerence. we set λ = σ1 and try again. we have the values f (0). The polynomial that interpolates f at 0. our approximation to f (0) should be negative.1. The polynomial p(λ) such that p. where c= f (λc ) − f (0) − f (0)λc .5. This electronic version is for personal use and may not be duplicated or distributed. then c > 0 and p(λ) has a minimum at λt = −f (0)/(2c) > 0.5) to compute λ+ . In this approach one evaluates f (0) and f (1) as before.com/SIAM/FR16.html. λc λ− We must consider two situations. λc . After the second step is rejected. so if f (λc ) > f (0). σ1 λ] or reject the parabolic model and simply set λ+ to σ0 λc or σ1 λc . such as the Broyden direction. In the MATLAB code nsola from the collection.ecsecurehost. f (λc ). An alternative to the twopoint model that avoids the need to approximate f (0) is a threepoint model.
ecsecurehost. sn 2 sn 2 If.5. In Chapter 7 we implemented Broyden’s method at a storage cost of one vector for each iteration.38) to obtain wn = λn (8.8) and (8.2. yn − Bn sn sn −1 T −1 .7) yn − Bn sn = F (xn+1 ) − (1 − λn )F (xn ). the polynomial modeling approach is heuristic.8) to construct Bc F (x+ ) as we did in Algorithm brsol.8) reduces to (7.41) when λn = 1 and λn+1 = 1 (so dn+1 = sn+1 ). The relation y − Bc s = F (x+ ) was not critical to this and we may also incorporate a line search at a cost of a bit of complexity in the implementation. This electronic version is for personal use and may not be duplicated or distributed.8) and the relation dn+1 = − I − wn sT n sn 2 −1 Bn F (xn+1 ) to solve for dn+1 and obtain an analog of (7. vn = .3 is that the simple relation between the sequence {wn } and {vn } is changed. and wn = (Bn un )/(1 + vn Bn un ).7) and (7. [63]. we set un = we can use (8.2. If a line search is used.Copyright ©1995 by the Society for Industrial and Applied Mathematics.html.8) = λn where −dn+1 sn + (λ−1 − 1) n sn 2 sn 2 −sn+1 /λn+1 sn + (λ−1 − 1) n sn 2 sn −1 dn+1 = −Bn+1 F (xn+1 ) 2 . Clearly. our approach is a nonlinear version of the method in [67]. We can use the ﬁrst equality in (8. Broyden’s method.3. 144 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS Interpolation with a cubic polynomial has also been suggested [87]. is the search direction used to compute xn+1 . then −1 sn = −λn Bn F (xn ) and hence (8.5.9) dn+1 = − −1 −1 sn 2 Bn F (xn+1 ) − (1 − λn )sT Bn F (xn+1 )sn n 2 −1 sn 2 + λn sT Bn F (xn+1 ) n 2 −1 We then use the second equality in (8.9) is left to Exercise 8. The diﬀerence from the development in § 7. Buy this book from SIAM at http://www. As in § 7. [133]. 8. The safeguarding and Theorem 8. using the notation in § 7. .com/SIAM/FR16. [86].44) (8. Veriﬁcation of (8.3.3.1 provide insurance that progress will be made in reducing F . Note that (8.
20) Buy this book from SIAM at http://www. compute λ0 . f (0). τ. The MATLAB implementation.html. brsola(x.5.1. n − 1 a = −λj /λj+1 .10 the reader is asked to fully develop and implement these ideas. It is possible that the Broyden direction fails to satisfy (8. The MATLAB code returns with an error message if more than a ﬁxed number of reductions in the step length are needed. n = −1. So. Initialize: Fc = F (x) r0 = F (x) 2 . an approximation of f (0) can be obtained at no additional cost and a parabolic line search based on f (0). We do not include details of the line search or provide for restarts or a limited memory implementation. Do while n ≤ maxit (a) n = n + 1 (b) x = x + sn (c) Evaluate F (x) (d) If F (x) (e) 2 ≤ τr r0 + τa exit. i. provided in the collection of MATLAB codes. (h) Set sn+1 = λn+1 d. z = −F (x) ii. Since dn+1 and sn+1 can occupy the same location. and f (1) can be used. there is no guarantee that the Broyden direction is a direction of decrease for F 2 . b = 1 − λj z = z + (asj+1 + bsj )sT z/ sj j 2 2 2 2 (f) d = ( sn 2 z + (1 − λn )sT zsn )/( sn n 2 (g) Compute λn+1 with a line search. In fact.2). . maxit.com/SIAM/FR16. s0 = λ0 d 2. the storage of the globalized methods exceeds that of the local method by one vector. As a side eﬀect. uses a threepoint parabolic line search and performs a restart when storage is exhausted like Algorithm brsol does.4. the storage requirements of Algorithm brsola would appear to be essentially the same as those of brsol. This electronic version is for personal use and may not be duplicated or distributed.2) before beginning the line search. The MATLAB implementation of brsola in the collection illustrates this point. However. computation of the step length requires storage of both the current point x and the trial point x + λs before a step can be accepted. d = −F (x). for j = 0.Copyright ©1995 by the Society for Industrial and Applied Mathematics. In Exercise 8. store sn+1 2 − λn sT z) n and λn+1 . Algorithm 8.1 and also changes η using (6. nmax) 1. 8. GLOBAL CONVERGENCE 145 Algorthm brsola uses these ideas. F.ecsecurehost.3. Another approach would be to compute F (xc )d numerically if a full step is rejected and then test (8. Examples for Newton–Armijo The MATLAB code nsola is a modiﬁcation of nsolgm that incorporates the threepoint parabolic line search from § 8.3.
1. We base our absolute error criterion on the norm · 2 / N as we did in § 6. For this problem we plot the progress of the iteration using η = . This is a tighter tolerance that in § 6.25 with the solid line and using the sequence given by (6.2. and will repeat the caution later. In Fig.21) from § 6. which requires a single function evaluation for each outer iteration when full steps can be taken. both line searches take three steplength reductions.4. using GMRES as the linear solver. Convectiondiﬀusion equation. 8. Counts of function evaluations corresponding to outer iterations are√ indicated by circles. initial iterate u0 = 0. However. the parabolic line search takes only one reduction in the next three iterations and none thereafter. We caution the reader now. the quasiNewton method. We consider a much more diﬃ −∇2 u + Cu(ux + uy ) = f with the same right hand side f . 146 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS once the iteration is near the root.9. As before.4. In the ﬁrst outer iteration. However. We compare this strategy. 1) × (0. We take (6. 8. 1) that we considered before. The parabolic line search required 7 outer iterates and 21 function evaluations in contrast with the constant reduction searches 11 outer iterates and 33 function evaluations. we can use the twopoint parabolic line search. when both methods succeed. The constant reduction line search took three reductions in the ﬁrst two outer iterations and two each in following two.4 with γ = .Copyright ©1995 by the Society for Industrial and Applied Mathematics.4.20) with the dashed Buy this book from SIAM at http://www.3 we show the progress of the iteration for the unpreconditioned equation.4. In Fig. We compare the twopoint parabolic line search with the constant reduction search (σ0 = σ1 = .5) for the arctan function. cult problem. . We set τr = τa = h2 /10.2.com/SIAM/FR16.4.4. This makes a globalization strategy critical (Try it without one!). however. with the constant η method as we did in § 6. In § 8. Inverse tangent function. that if the initial iterate is far from the solution. Since we have easy access to analytic derivatives in this example.ecsecurehost.html. 31 × 31 mesh and homogeneous Dirichlet boundary conditions on the unit square (0. because of the large value of C and resulting illconditioning. 8.4.2 and 7. in all the ﬁgures we plot F (xn ) 2 / F (x0 ) 2 against the number of function evaluations required by all line searches and inner and outer iterations to compute xn . we set C = 100. This electronic version is for personal use and may not be duplicated or distributed. Here.3 we compare the Broyden–Armijo method with NewtonGMRES for those problems for which both methods were successful. an inexact Newton method such as NewtonGMRES can succeed in many case in which a quasiNewton method can fail because the quasiNewton direction may not be an inexact Newton direction.2 we plot the iteration history corresponding to the parabolic line search with the solid line and that for the constant reduction with the dashed line. is needed to obtain an accurate solution.2 and. 8. may well be most eﬃcient. We use x0 = 10 as the initial iterate with τr = τa = 10−8 .
This electronic version is for personal use and may not be duplicated or distributed. This problem is very poorly conditioned and much tighter control on the inner iteration is needed than for the other problems. C = 100. We set the parameters in (6.9 and ηmax = . 759 (constant η) and 744 (varying η) function evaluations. 8. line.html.20) to γ = . Newton–Armijo for the PDE. We consider the preconditioned problem in Fig.4.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 8.4 (varying η) million ﬂoating point operations.25. In the computation we Buy this book from SIAM at http://www. . The two approaches to selection of the forcing term performed very similarly.3 (constant η) and 75.3. Newton–Armijo for the arctan function. 8.com/SIAM/FR16.2. GLOBAL CONVERGENCE 0 147 10 10 Relative Nonlinear Residual 2 10 4 10 6 10 8 10 10 10 12 0 5 10 15 20 Function Evaluations 25 30 35 Fig. The iterations required 75. and 25 (constant η) and 22 (varying η) outer iterations.ecsecurehost. 10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 0 100 200 300 400 500 Function Evaluations 600 700 800 Fig.
25 with the solid line and using the sequence given by (6.21) with the fast Poisson solver fish2d.Copyright ©1995 by the Society for Industrial and Applied Mathematics. the NewtonGMRES iteration always took full steps. The line search in this case reduced the step length three times on the ﬁrst iteration and twice on the second and third. 16 function evaluations. 9 outer iterations.4 we show the progress of the iteration for the preconditioned equation.20) to γ = .4. 8. This electronic version is for personal use and may not be duplicated or distributed. We set the parameters in (6.3.4. with the forcing term given by (6. C = 100. 8.20) terminated after 70 function evaluations. and 2. The line search in the nonconstant η case reduced the step length three times on the ﬁrst iteration and once on the second.20). When compared to the Broyden’s method results in § 6.99.4.9 and ηmax = . In Fig.20) with the dashed line. As reported in § 6. The constant η iteration terminated after 79 function evaluations. 8. .2 with C = 20.4.3 million ﬂoatingpoint operations.html.4. In these cases the Jacobian is not well approximated by a low rank perturbation of the identity [112] and the ability of the inexact Newton iteration to ﬁnd a good search direction was the important factor.ecsecurehost. the Broyden–Armijo costs reﬂect Buy this book from SIAM at http://www. 148 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS preconditioned (6. We begin by revisiting the example in § 6. 10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 0 10 20 30 40 50 Function Evaluations 60 70 80 Fig. The most eﬃcient iteration. and roughly 13.com/SIAM/FR16. terminating in 4 outer iterations. For this problem we plot the progress of the iteration using η = .2.6 million ﬂoatingpoint operations. Broyden–Armijo. We found that the Broyden–Armijo line search failed on all the unpreconditioned convection diﬀusion equation examples.2. 9 outer iterations. Newton–Armijo for the PDE.5 million ﬂoatingpoint operations. and 12. required at most 16 inner iterations while the constant η approach needed at most 10. The iteration in which {ηn } is given by (6. when increases in the residual were allowed.
42 outer iterates.4. We set τa = τr = h2 . C = 20. The storage requirements increase as the nonlinear iteration progresses and this fact puts Broyden’s Buy this book from SIAM at http://www. From these examples we see that Broyden–Armijo can be very eﬃcient provided only a small number of iterations are needed. ηmax = .2. and γ = . roughly 16 million ﬂoatingpoint operations. The Broyden–Armijo iteration required 34 nonlinear iterations.99.5 million ﬂoatingpoint operations.2 took 12 iterations and 2. GLOBAL CONVERGENCE 149 an improvement in eﬃciency. In Fig. the Broyden–Armijo iteration required storage for 37 vectors while the NewtonGMRES iteration needed at most 16 inner iterations. and 85 function evaluations.com/SIAM/FR16.7.5.Copyright ©1995 by the Society for Industrial and Applied Mathematics.6 we compare our implementation of Newton–GMRES–Armijo (dashed line) to Broyden–Armijo (solid line). 8.4. This electronic version is for personal use and may not be duplicated or distributed. the performance of the two methods is more similar. and therefore was much more eﬃcient in terms of storage. The Broyden iteration in § 6. 8. Figure 8.8 million ﬂoatingpoint operations while the Broyden–Armijo took 9 iterations and required 2 million ﬂoatingpoint operations. . Restarting the Broyden iteration every 19 iterations would equalize the storage requirements with Newton–GMRES–Armijo. Broyden’s method suffers under these conditions as one can see from the comparison of Broyden– Armijo (solid line) and Newton–GMRES–Armijo (dashed line) in Fig. needing to store 21 vectors. 10 0 Relative Nonlinear Residual 10 1 10 2 10 3 0 2 4 6 8 10 Function Evaluations 12 14 16 Fig.5 compares the Broyden–Armijo iteration (solid line) to the GMRES iteration (dashed line). The restarted Broyden–Armijo iteration took 19.9. and γ = . For the more diﬃcult problem with C = 100. In terms of storage.ecsecurehost.9 as we did in § 6. 8. and 123 function evaluations. We set τa = τr = h2 /10.5. ηmax = . The steplength was reduced on the second iteration (twice) and the third (once). NewtonGMRES and Broyden–Armijo for the PDE. In our implementation of brsola the threepoint parabolic line search was used.html.
10 0 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 0 20 40 60 80 Function Evaluations 100 120 140 Fig. NewtonGMRES and restarted Broyden–Armijo for the PDE.6. method at a disadvantage to NewtonGMRES when the number of nonlinear iterations is large.7.Copyright ©1995 by the Society for Industrial and Applied Mathematics.ecsecurehost. C = 100.html.com/SIAM/FR16. . Buy this book from SIAM at http://www. C = 100. 150 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 0 10 10 Relative Nonlinear Residual 1 10 2 10 3 10 4 10 5 0 10 20 30 40 50 60 Function Evaluations 70 80 90 Fig. NewtonGMRES and Broyden–Armijo for the PDE. 8. This electronic version is for personal use and may not be duplicated or distributed. 8.
Duplicate the results in § 8. the hybrid Newton algorithm from Chapter 5. .000 down and takes a 7 year installment loan to pay the balance. You might use the MATLAB code nsola to do this.5. f (x) = x(1 + sin(1/x)) 5. Modify brsola to test the Broyden direction for the descent property and use a twopoint parabolic line search.9 + arctan(x) 4. Try to do it in such a way that the chord direction is used whenever possible. Buy this book from SIAM at http://www. then F (x∗ ) is singular.5. to use the Armijo rule.5.e.5. How does the method proposed in [10] diﬀer from the one implemented in nsola? What could be advantages and disadvantages of that approach? 8. Vary the convection coeﬃcient in the convectiondiﬀusion equation and the mesh size and report the results. 8. He puts $10. f (x) = ex 6.9. Apply your code to the following functions with x0 = 10.2. and TFQMR. 8. f (x) = .7.8.4. GLOBAL CONVERGENCE 151 8. 8.5.000. Use nsola to duplicate the results in § 8. This is interesting in the locally convergent case as well.40. f (x) = 2 + sin(x) 7. 8.1. What could you do if the Broyden direction is not a descent direction? Apply your code to the examples in § 8.Copyright ©1995 by the Society for Industrial and Applied Mathematics.4.8) and (8. Show that if F is Lipschitz continuous and the iteration {xn } produced by Algorithm nsola converges to x∗ with F (x∗ ) = 0. Experiment with other linear solvers such as GMRES(m).ecsecurehost. Implement the Newton–Armijo method for single equations.9). f (x) = arctan(x2 ) 3. f (x) = arctan(x) (i. Modify nsol.6. In what ways are the results in [25] and [70] more general than those in this section? What ideas do these papers share with the analysis in this section and with [3] and [88]? 8. 1.5. what is the interest rate? Assume monthly compounding.com/SIAM/FR16.5. 8. Exercises on global convergence 8.4.3. f (x) = 1 + x2 8.5. A numerical analyst buys a German sports car for $50.5.) 2.1.10.5. Explain your results.2.. 8.5.5. If the monthly payments are $713.html. Verify (8. BiCGSTAB. This electronic version is for personal use and may not be duplicated or distributed.
2.2) with ηn bounded away from 1? If not.ecsecurehost.5. Does the secant method for equations in one variable always give a direction that satisﬁes (8. . Under what conditions will the iteration given by nsola converge to a root x∗ that is independent of the initial iterate? Buy this book from SIAM at http://www. This electronic version is for personal use and may not be duplicated or distributed.12.13.Copyright ©1995 by the Society for Industrial and Applied Mathematics.11.5.html. Modify nsola to use a cubic polynomial and constant reduction line searches instead of the quadratic polynomial line search. 8.1 is applicable? 8.com/SIAM/FR16.4.5. Compare the results with the examples in § 8. when does it? How would you implement a secantArmijo method in such a way that the convergence theorem 8. 152 ITERATIVE METHODS FOR LINEAR AND NONLINEAR EQUATIONS 8.
C. ´ ´ [13] N. 133–181. Who was ‘Raphson’?. Comput. J. ed. Bosma and W. The principle of minimized iterations in the solution of the matrix eigenvalue problem. Numerical Solution 153 Buy this book from SIAM at http://www. Armijo. Romine. 28–85.. Bibliography [1] E.. Rose. Arnoldi. Astrophys. Templates for the Solution of Linear Systems: Building Blocks for Iterative Methods.. 9 (1951). [3] L. 1990. Iterative Solution Methods. 183–196. J.. Math. 13 (1992). Manteuffel. Philadelphia. Numer. 22 (1973). J. V. Donato. [8] O. pp. Sur les op´rations dans les ensembles abstraits et leur applications e aux ´quations int´grales. Bicanic and K. F. Brenan. pp. Math. Brakhage. Cambridge. E.com/SIAM/FR16. An inverse matrix adjustment arising in discriminant analysis. 2 (1960). SIAM J. p.. E. [14] P.html. R. 2nd.ecsecurehost. This electronic version is for personal use and may not be duplicated or distributed. 283. SIAM Rev. Pozo. R. Statist.. Johnson. Eﬃcient methods to calculate Chandrasekhar’s Hfunctions. [16] K. Engrg. Eijkhout. 1989. 37 (1981). pp. Simplicial and continuation methods for approximating ﬁxed points and solutions to systems of equations. PA. F. van der Vorst. T. 1993. [5] S. Math. Banach. 17–31. New York. 16 (1966). A. DeRooij. Demmel. and J. pp. pp. 1994. Allgower and K. John Wiley. 279–295. B. Quart. Ann. Global approximate Newton methods. H. 22 (1951). Cambridge University Press. M. pp. S. 126 (1983). Appl. 22 (1980). [11] M. Barlett. J. Ashby. Society for Industrial and Applied Mathematics.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Math. [12] R. 17–29. e e [10] R.. Paciﬁc J. E. Math. [9] S. Georg. Astron. Numer. Numer. A. T. Uber die numerische Behandlung von Integralgleichungen nach der Quadraturformelmethode. An Introduction to Numerical Analysis. and H. Numer. [6] K. Math. New York. S. pp. Minimization of functions having Lipschitzcontinuous ﬁrst partial derivatives. . Barrett. 148–152. Petzold. Math. 1– 29. [4] W. 3 (1922). pp. pp.. 1–3. Fund. Numerical Continuation Methods : An Introduction.. Atkinson. [7] . Otto. 107–111. Iterative variants of the Nystr¨m method for the numerical o solution of integral equations. Campbell.. S. [2] . ¨ [15] H. Axelsson. Sci. Methods. and L. Chan. J. A comparison of adaptive Chebyshev and least squares polynomial preconditioning for Hermetian positive deﬁnite linear systems. 14 (1979). Dongarra. Statist. SpringerVerlag. pp. Berry. L. Bank and D. Internat.. E. L.
154 BIBLIOGRAPHY of Initial Value Problems in DiﬀerentialAlgebraic Equations. J. . Englewood Cliﬀs. pp. University of Maryland. Using Krylov methods in the solution of largescale diﬀerentialalgebraic systems. I. PhD thesis. Optim. Vieweg Verlag. C. Burmeister.. D.. W. 40–91. NewtonKrylovSchwarz methods in CFD. Cai. in Proceedings of Internationaler Kongreß uber Anwendung der Mathematik in ¨ dem Ingenieurwissenschaften. Press. J. [19] W. Ipsen. Campbell. D. Brown. Broyden. Brown and A. I. This electronic version is for personal use and may not be duplicated or distributed. Report CRSCTR9513. ed. A local convergence theory for combined inexact–Newton/ ﬁnite– diﬀerence projection methods. Convergence theory of nonlinear NewtonKrylov algorithms. R. to appear. 297–330. and J. 24 (1987). Comput. pp. The convergence of an algorithm for solving sparse nonlinear systems. and L. 577–593. pp. Meyer. 1038–1051. The Mathematics of Radiative Transfer. GMRES and the minimal polynomial. 129–156.. J. and Z. R. SIAM J. [35] R. pp. Center for Research in Scientiﬁc Computation. 12 (1973). Hindmarsh. Ipsen.. Byrd. Notes in Numerical Fluid Mechanics. Byrne. Maths. [33] S. W. in Proceedings of the International Workshop on the NavierStokes Equations. [31] R. On the local and superlinear convergence of quasiNewton methods. [26] C. N.com/SIAM/FR16. J. and R. Tech. C. Center for Research in Scientiﬁc Computation. Kelley. 19 (1965). T. Inst. 1973. A Multigrid Tutorial. C.html. SIAM J. July 1994. No. 1960. Buy this book from SIAM at http://www. 63 (1994). pp. Petzold. and C. N. Convergence estimates for solution of integral equations with GMRES. Zur Konvergenz einiger verfahren der konjugierten Richtungen. B. E. to appear. G. Applic.. SIAM J. E.. 25 (1971). pp. Schnabel. [22] P. H. N. Journal of Integral Equations and Applications. [32] X. C. [18] . 1975. Statist. 407–434. Tech. 1994. Programming. G. Math. Busbridge. pp. Some eﬃcient algorithms for solving systems of nonlinear equations. Society for Industrial and Applied Mathematics. [21] P. 1467–1488. North Carolina State University. F. 31 (1989). [23] P. J. Saad.C. [30] I. Comput.. T. 1987. BIT. Sci. 1996. D. Cavanaugh. Anal. Sci. NJ. Philadelphia. 14 in Classics in Applied Mathematics. Math. D. Cambridge Univ. A class of methods for solving nonlinear simultaneous equations. pp. D. 285–294. 327–344. 450–481. Algorithms for Minimization Without Deriviatives. Appl. C. Brown and Y. Xue. F. and M. SIAM J. [27] . C. ´ [28] C.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Representation of quasiNewton matrices and their use in limited memory methods. PA. [34] S. Hindmarsh. Dennis. 10 (1989). Q. Diﬀerence Equations and Iterative Processes. Numer. Briggs. Kelley. Reduced storage matrix methods in stiﬀ ODE systems. Sci. SIAM. Brown. Comput. 223– 246. Brent. PrenticeHall. Nocedal. and A. North Carolina State University. Campbell. 50. 10 (1973). Broyden.. L. Rannacher. N. L. C. C. Statist. Gropp. Anal. pp. Comput. Meyer. D. SIAM J. Philadelphia. Keyes. Brown. pp.ecsecurehost. [17] R. Comput. pp. VODE: A variable coeﬃcient ode solver. Hybrid Krylov methods for nonlinear systems of equations. Weimar. N. A. Report CRSCTR9410. Cambridge. Cambridge Tracts. [29] W.. no. [20] P. 11 (1990)... [25] . More. Numer. March 1995. 4 (1994). [24] P. Braunschweig. 1970. G. SIAM J. Math. 15 (1994). Math. Comput. Hindmarsh. Tidriri.
Decker and C. pp. Statist. 219–229. [49] D. and C. Steihaug. pp. H.. 1976. Comput. p. 13 (1974). 66–70. Numer. Proceedings of the Fourth International Symposium on Domain Decomposition Methods. A generalized conjugate gradient method for the numerical solution of elliptic partial diﬀerential equations.ecsecurehost. B. 22 (1985). Numer. Comput. ´ [42] T. [54] T. Reid. Buy this book from SIAM at http://www. Simoncini. Moscow. Philadelphia. G. E. Math. Domain Decomposition Methods. [56] R.. SIAM J. ChaitinChatelin. and G. 42 (1983). 187–209. D. pp. Numer. Domain Decomposition Methods. 296–314. Anal. The Nstep iteration procedures. New York. . Society for Industrial and Applied Mathematics. eds. SIAM J. eds. Craig. 4 (1967). Tong. Chan. Daniel.. 26 (1983). H. Math. Periaux. SIAM J. O’Leary. 19 (1982). PA. 10–26. No. 147–154. Appl. J. P. Finding a zero by means of successive linear interpolation. R.. Golub. S. Programming. 1969. 220–252. 1990. Los Angeles. 1988. Numer. [39] . 190–212. 1990. G. and C. 1989. [46] E. [52] . Dembo and T. Chan. Estimation of sparse Jacobian matrices and graph coloring problems. F. SIAM J. Philadelphia. Academic Press... Widlund. P. 15 (1994). [55] R. Phys. Kelley. Radiative Transfer. 37–48. TX. J. Society for Industrial and Applied Mathematics. Curtis. Anal.. J. Chandrasekhar. eds. Math. in Constructive Aspects of the Fundamental Theorem of Algebra. pp. Anal. January 14–16. Domain Decomposition Methods. December 1994. [45] P. Anal. Kelley. Sci. T.. Dembo. pp. J. 1989. [53] . and D. On the estimation of sparse Jacobian matrices.html. eds. Broyden’s method for a class of problems having singular Jacobian at the root. Anal. [41] S. Proceedings of the Third International Symposium on Domain Decomposition Methods. Coleman and J. Powell. Steihaug. 117–119. [40] . J. and O.. Math. The conjugate gradient method for linear and nonlinear operator equations. Inst. USSR. Convergence rates for Newton’s method at singular points.. [47] A. SIAM J. A quasiminimal residual variant of the BiCGSTAB algorithm for nonsymmetric systems. 1960. [37] T. Rose. SIAM J. J.. Block preconditioning for the conjugate gradient method. pp. [48] J. This electronic version is for personal use and may not be duplicated or distributed. 309–332. 1988. SIAM J. pp.Copyright ©1995 by the Society for Industrial and Applied Mathematics. pp. PA. pp. Glowinski. More. Anal. W. 6 (1985). [51] . and T. Tech. SIAM J. Golub. 338. 1991. Society for Industrial and Applied Mathematics. H. BIBLIOGRAPHY 155 [36] F. R. Sci. pp. Society for Industrial and Applied Mathematics. 19 (1982). Dekker. PA. V.. Handbook for Matrix Computations. T. 400–408. Dover. Szeto. 64– 73. Sublinear convergence of the chord method at singular points.com/SIAM/FR16..Proceedings of the Second International Symposium on Domain Decomposition Methods. and J. PA.. 566–574. Truncated Newton algorithms for largescale optimization. Houston.. M. T. 17 (1980). Numer. Frontiers in Appl. J. ´ [38] T. 34 (1955). [44] P. [43] T. VanLoan. Convergence acceleration for Newton’s method at singular points. ed. Bunch and D. SIAM J. Is nonnormality a serious diﬃculty ?.. Report TR/PA/94/18. F. 4. Anal. 20 (1983). J. Decker. W. Concus. pp. [50] D. 20 (1983). Eisenstat. Math. W. Philadelphia. Coleman and C. Meurant. Concus. K. pp. in Sparse Matrix Computations. J. Numer. CERFACS. Inexact Newton methods. Keller. CA.. Numer. pp. pp.. Newton’s method at singular points I.. R. Henrici. Philadelphia. Gallopoulos. Numer.
[71] H. 1984. 443–459. J. Anal. Y. 352– 362. [73] M. [72] H. pp. Duncan. A hybrid ChebyshevKrylov subspace algorithm for solving nonsymmetric systems of linear equations. 4 (1994).. C.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 1982. in Mathematical Programming Study 22: Mathematical programming at Oberwolfach II. CT. Walker. J. F. Impact of Computing in Science and Engineering. Numer. Sparse. B. [76] R. Least change secant updates for quasiNewton methods. Elman. J. 707–718. 1976. [67] P. G. New York. Optim. pp. 7 (1986). Frank. in Numerical Analysis Dundee 1975. 19 (1977). 1996. SIAM J. 21 (1979). Seventh Series. 18 (1981). A characterization of superlinear convergence and its application to quasiNewton methods. 62 (1993). [59] J. Schnabel. ´ [60] J.. 425–472. 73–89. The application of quasiNewton methods in ﬂuid mechanics. Saad. 760–778.. Watson. W.. E. [63] . Sci. L. Programming. Toward a uniﬁed convergence theory for Newtonlike methods. methods. G. and R. M. pp. A. [61] . 70–85. Numer.. Leastchange sparse secant updates with inaccurate secant conditions. SIAM J. 1971. [70] . pp. pp. E. SIAM. Saylor. and Dublin Philosophical Magazine and Journal of Science. Anal. Philadelphia. 4 (1994).. Dennis and J. Iterative Methods for Large. Nonsymmetric Systems of Linear Equations. 949–987. Necessary and suﬃcient conditions for the existence of a conjugate gradient method. E. Sci. This electronic version is for personal use and may not be duplicated or distributed. Deuflhard. 16–32. Local convergence analysis of tensor methods for nonlinear equations. Edinburgh. Walter. Statist. Methods Engrg. SIAM J. 46–89. pp. Fletcher. New York. pp. pp. C. 28 (1974). 840–855. 16 in Classics in Applied Mathematics. Amsterdam. Math. B. Fast secant methods for the iterative solution of large nonsymmetric linear systems. Choosing the forcing terms in an inexact Newton method. pp. On the Kantorovich hypothesis for Newton’s method. Manteuffel. 17 (1996). Rall. 21 (1984).. 156 BIBLIOGRAPHY [57] J. SIAM Rev. 22 (1985). Feng. SIAM J. 2 (1990). Anal. Schnabel. Globally convergent inexact Newton methods. [68] W. Some devices for the solution of large sets of simultaneous linear equations (with an appendix on the reciprocation of partitioned matrices). Comput. and X. Optim. Freund. Faber and T. Berlin. J. Bathe. pp. Zhang. SIAM J. ed. Convergence theorems for least change secant update methods. SpringerVerlag. Triangular decomposition methods for solving reducible nonlinear systems of equations. Dennis. pp.. E. Eisenstat and H. motivation and theory. pp. D. SIAM J. and A. pp.. 17 (1981). pp. 427–459. More. Internat. 6 (1969). QuasiNewton methods. C. Comput. Numer.. Dennis and H. Yale University. [58] . [75] D.html. .. SIAM J. E. Buy this book from SIAM at http://www. B. Numerical Methods for Unconstrained Optimization and Nonlinear Equations.ecsecurehost. 493–507. Engelman. Academic Press. and P. SIAM J. no. Walker. F. Math... Anal. [74] V.. pp. PhD thesis.com/SIAM/FR16. [69] S. and K. [64] J. 244–276. 35 (1944). ed. P. North–Holland. 393–422. Dennis and R. [65] . Numer. The London. 549–560. 660–670. Strang. pp. pp. Martinez. pp. J. Elman. Inaccuracy in quasiNewton methods: Local improvement theorems. 358–382. Conjugate gradient methods for indeﬁnite systems. Comput. in Nonlinear Functional Analysis and Applications. R. SIAM Rev. Dennis. E. New Haven. Numer. [62] J. [66] .
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
BIBLIOGRAPHY
157
[77] R. W. Freund, A transposefree quasiminimal residual algorithm for nonHermitian linear systems, SIAM J. Sci. Comput., 14 (1993), pp. 470–482. [78] R. W. Freund, G. H. Golub, and N. M. Nachtigal, Iterative solution of linear systems, Acta Numerica, 1 (1991), pp. 57–100. [79] R. W. Freund, M. H. Gutknecht, and N. M. Nachtigal, An implementation of the lookahead Lanczos algorithm for nonHermitian matrices, SIAM J. Sci. Comput., 14 (1993), pp. 137–158. [80] R. W. Freund and N. M. Nachtigal, QMR: a quasiminimal residual algorithm for nonhermitian linear systems, Numer. Math., 60 (1991), pp. 315–339. [81] , An implementation of the QMR method based on coupled twoterm recurrences, SIAM J. Sci. Comput., 15 (1994), pp. 313–337. [82] D. M. Gay, Some convergence properties of Broyden’s method, SIAM J. Numer. Anal., 16 (1979), pp. 623–630. [83] C. W. Gear, Numerical Initial Value Problems in Ordinary Diﬀerential Equations, PrenticeHall, Englewood Cliﬀs, NJ, 1971. [84] R. R. Gerber and F. T. Luk, A generalized Broyden’s method for solving simultaneous linear equations, SIAM J. Numer. Anal., 18 (1981), pp. 882–890. [85] P. E. Gill and W. Murray, QuasiNewton methods for unconstrained optimization, J. I. M. A., 9 (1972), pp. 91–108. [86] , Safeguarded steplength algorithms for optimization using descent methods, Tech. Report NAC 37, National Physical Laboratory Report, Teddington, England, 1974. [87] P. E. Gill, W. Murray, and M. H. Wright, Practical Optimization, Academic Press, London, 1981. [88] A. A. Goldstein, Constructive Real Analysis, Harper and Row, New York, 1967. [89] G. H. Golub and C. G. VanLoan, Matrix Computations, Johns Hopkins University Press, Baltimore, 1983. [90] A. Griewank, Analysis and modiﬁcation of Newton’s method at singularities, PhD thesis, Australian National University, 1981. [91] , Rates of convergence for secant methods on nonlinear problems in Hilbert space, in Numerical Analysis, Proceedings Guanajuato , Mexico 1984, Lecture Notes in Math., No, 1230, J. P. Hennart, ed., SpringerVerlag, Heidelberg, 1986, pp. 138– 157. [92] , The solution of boundary value problems by Broyden based secant methods, in Computational Techniques and Applications: CTAC 85, Proceedings of CTAC, Melbourne, August 1985, J. Noye and R. May, eds., North Holland, Amsterdam, 1986, pp. 309–321. [93] , On the iterative solution of diﬀerential and integral equations using secant updating techniques, in The State of the Art in Numerical Analysis, A. Iserles and M. J. D. Powell, eds., Clarendon Press, Oxford, 1987, pp. 299–324. [94] A. Griewank and G. F. Corliss, eds., Automatic Diﬀerentiation of Algorithms: Theory, Implementation, and Application, Society for Industrial and Applied Mathematics, Philadelphia, PA, 1991. [95] A. Griewank and P. L. Toint, Local convergence analysis for partitioned quasinewton updates, Numer. Math., 39 (1982), pp. 429–448. [96] , Partitioned variable metric methods for large sparse optimization problems, Numer. Math., 39 (1982), pp. 119–137. [97] W. A. Gruver and E. Sachs, Algorithmic Methods In Optimal Control, Pitman, London, 1980.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
158
BIBLIOGRAPHY
[98] M. H. Gutknecht, Variants of BICGSTAB for matrices with complex spectrum, SIAM J. Sci. Comput., 14 (1993), pp. 1020–1033. [99] W. Hackbusch, MultiGrid Methods and Applications, vol. 4 of Springer Series in Computational Mathematics, SpringerVerlag, New York, 1985. [100] , Multigrid methods of the second kind, in Multigrid Methods for Integral and Diﬀerential Equations, Oxford University Press, Oxford, 1985. [101] W. E. Hart and S. O. W. Soul, QuasiNewton methods for discretized nonlinear boundary problems, Journal of the Institute of Applied Mathematics, 11 (1973), pp. 351–359. [102] H. V. Henderson and S. R. Searle, On deriving the inverse of a sum of matrices, SIAM Rev., 23 (1981), pp. 53–60. [103] M. R. Hestenes and E. Steifel, Methods of conjugate gradient for solving linear systems, J. of Res. Nat. Bureau Standards, 49 (1952), pp. 409–436. [104] D. M. Hwang and C. T. Kelley, Convergence of Broyden’s method in Banach spaces, SIAM J. Optim., 2 (1992), pp. 505–532. [105] E. Isaacson and H. B. Keller, Analysis of numerical methods, John Wiley, New York, 1966. [106] L. Kantorovich and G. Akilov, Functional Analysis, 2nd ed., Pergamon Press, New York, 1982. [107] L. V. Kantorovich, Functional analysis and applied mathematics, Uspehi Mat. Nauk., 3 (1948), pp. 89–185. translation by C. Benster as Nat. Bur. Standards Report 1509. Washington, D. C., 1952. [108] H. B. Keller, Newton’s method under mild diﬀerentiability conditions, J. Comput. Sys. Sci, 4 (1970), pp. 15–28. [109] , Lectures on Numerical Methods in Bifurcation Theory, Tata Institute of Fundamental Research, Lectures on Mathematics and Physics, SpringerVerlag, New York, 1987. [110] C. T. Kelley, Solution of the Chandrasekhar Hequation by Newton’s method, J. Math. Phys., 21 (1980), pp. 1625–1628. [111] , A fast multilevel algorithm for integral equations, SIAM J. Numer. Anal., 32 (1995), pp. 501–513. [112] C. T. Kelley and E. W. Sachs, A quasiNewton method for elliptic boundary value problems, SIAM J. Numer. Anal., 24 (1987), pp. 516–531. [113] , A pointwise quasiNewton method for unconstrained optimal control problems, Numer. Math., 55 (1989), pp. 159–176. [114] , Fast algorithms for compact ﬁxed point problems with inexact function evaluations, SIAM J. Sci. Statist. Comput., 12 (1991), pp. 725–742. [115] , A new proof of superlinear convergence for Broyden’s method in Hilbert space, SIAM J. Optim., 1 (1991), pp. 146–150. [116] , Pointwise Broyden methods, SIAM J. Optim., 3 (1993), pp. 423–441. [117] , Multilevel algorithms for constrained compact ﬁxed point problems, SIAM J. Sci. Comput., 15 (1994), pp. 645–667. [118] C. T. Kelley and R. Suresh, A new acceleration method for Newton’s method at singular points, SIAM J. Numer. Anal., 20 (1983), pp. 1001–1009. [119] C. T. Kelley and Z. Q. Xue, Inexact Newton methods for singular problems, Optimization Methods and Software, 2 (1993), pp. 249–267. [120] , GMRES and integral operators, SIAM J. Sci. Comput., 17 (1996), pp. 217– 226. [121] T. Kerkhoven and Y. Saad, On acceleration methods for coupled nonlinear elliptic systems, Numerische Mathematik, 60 (1992), pp. 525–548.
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
Copyright ©1995 by the Society for Industrial and Applied Mathematics. This electronic version is for personal use and may not be duplicated or distributed.
BIBLIOGRAPHY
159
[122] C. Lanczos, Solution of linear equations by minimized iterations, J. Res. Natl. Bur. Stand., 49 (1952), pp. 33–53. [123] T. A. Manteuffel, Adaptive procedure for estimating parameters for the nonsymmetric Tchebychev iteration, Numer. Math., 31 (1978), pp. 183–208. [124] T. A. Manteuffel and S. Parter, Preconditioning and boundary conditions, SIAM J. Numer. Anal., 27 (1990), pp. 656–694. [125] E. S. Marwil, Convergence results for Schubert’s method for solving sparse nonlinear equations, SIAM J. Numer. Anal., (1979), pp. 588–604. [126] S. McCormick, Multilevel Adaptive Methods for Partial Diﬀerential Equations, Society for Industrial and Applied Mathematics, Philadelphia, PA, 1989. [127] J. A. Meijerink and H. A. van der Vorst, An iterative solution method for linear systems of which the coeﬃcient matrix is a symmetric Mmatrix, Math. Comput., 31 (1977), pp. 148–162. [128] C. D. Meyer, Matrix Analysis and Applied Linear Algebra, forthcoming. ´ [129] J. J. More, Recent developments in algorithms and software for trust region methods, in Mathematical Programming: The State of the Art, A. Bachem, M. Gr¨schel, and B. Korte, eds., SpringerVerlag, Berlin, 1983, pp. 258–287. o ´ [130] J. J. More and J. A. Trangenstein, On the global convergence of Broyden’s method, Math. Comput., 30 (1976), pp. 523–540. [131] T. E. Mott, Newton’s method and multiple roots, Amer. Math. Monthly, 64 (1957), pp. 635–638. [132] T. W. Mullikin, Some probability distributions for neutron transport in a half space, J. Appl. Probab., 5 (1968), pp. 357–374. [133] W. Murray and M. L. Overton, Steplength algorithms for minimizing a class of nondiﬀerentiable functions, Computing, 23 (1979), pp. 309–331. [134] N. M. Nachtigal, S. C. Reddy, and L. N. Trefethen, How fast are nonsymmetric matrix iterations?, SIAM J. Matrix Anal. Appl., 13 (1992), pp. 778– 795. [135] N. M. Nachtigal, L. Reichel, and L. N. Trefethen, A hybrid gmres algorithm for nonsymmetric linear systems, SIAM J. Matrix Anal. Appl., 13 (1992). [136] S. G. Nash, Preconditioning of truncated Newton methods, SIAM J. Sci. Statist. Comput., 6 (1985), pp. 599–616. [137] , Who was Raphson? (an answer). Electronic Posting to NADigest, v92n23, 1992. [138] J. L. Nazareth, Conjugate gradient methods less dependent on conjugacy, SIAM Rev., 28 (1986), pp. 501–512. [139] O. Nevanlinna, Convergence of Iterations for Linear Equations, Birkh¨user, a Basel, 1993. [140] I. Newton, The Mathematical Papers of Isaac Newton (7 volumes), D. T. Whiteside, ed., Cambridge University Press, Cambridge, 1967–1976. [141] B. Noble, Applied Linear Algebra, Prentice Hall, Englewood Cliﬀs, NJ, 1969. [142] J. Nocedal, Theory of algorithms for unconstrained optimization, Acta Numerica, 1 (1991), pp. 199–242. [143] D. P. O’Leary, Why Broyden’s nonsymmetric method terminates on linear equations, SIAM J. Optim., 4 (1995), pp. 231–235. [144] J. M. Ortega, Numerical Analysis A Second Course, Classics in Appl. Math., No. 3, Society for Industrial and Applied Mathematics, Philadelphia, PA, 1990. [145] J. M. Ortega and W. C. Rheinboldt, Iterative Solution of Nonlinear Equations in Several Variables, Academic Press, New York, 1970. [146] A. M. Ostrowski, Solution of Equations and Systems of Equations, Academic
Buy this book from SIAM at http://www.ecsecurehost.com/SIAM/FR16.html.
. Amsterdam. in Nonlinear Programming 2. New York. pp. Math. D. 20 (1966).. [168] E. M. University of Minnesota. 7 (1986). New York. The Chebyshev Polynomials. pp. K. A description of DASSL: a diﬀerential/algebraic system solver. Sci. [155] L. SIAM J. ex nova inﬁnitarum serierum doctrina. [159] W. Reid. Math. July 1972. [160] J. C.. North Holand. Sachs. Comput. SIAM J. Saad. Petzold. Sci. SIAM Rev. J. pp. and S. pp. Soc.. Meyer. eds.. 145–210. New York. 160 BIBLIOGRAPHY Press. Comput. de Math. 1970. 24 (1987). Math. [157] G. Academic Press. W. A ﬂexible innerouter preconditioned GMRES algorithm. Report 9238. S. illconditioned equations and Newton’s method. 1980. eds. (1993). Programming. [153] M. 44 (1985). 1975. pp. Macmillan. at Cambridge. Parlett. Robinson. 11 (1955). D. Rachford. pp. Numer.Copyright ©1995 by the Society for Industrial and Applied Mathematics. R. Stepleman et al. Saylor and D. pp. John Wiley. in Proc. New York. ILUT: A dual threshold incomplete LU factorization.com/SIAM/FR16. The Symmetric Eigenvalue Problem. pp. Broyden’s method in Hilbert space. in Numerical Methods for Nonlinear Algebraic Equations. . NJ. R. Numer. L. Rice. [154] . Anal. [148] B. Real Analysis. Appl. [152] E. Statist. Statist. Numer. North Holland. Experiments on GramSchmidt orthogonalization. 1992. 29 (1979). Numerical Analysis of Parametrized Nonlinear Equations. This electronic version is for personal use and may not be duplicated or distributed.. Comput. Taylor. New York. deducta ac demonstrata. Prentice Hall. 65–68. Inverse iteration. Comput. Rall. [164] . M´moire sur la th´orie des ´quations aux d´riv´es partielles et la e e e e e e m´thode des approximations successives. 1968. Liu. Peaceman and H. Amsterdam.. Least squares solution of sparse systems of nonlinear equations by a modiﬁed Marquardt algorithm.. A lookahead Lanczos algorithm for unsymmetric matrices. Original in British Library. [151] L. O. and Z. [162] H.. The numerical solution of parabolic and elliptic diﬀerential equations. 15 (1978). Picard. Indus. Least squares polynomials in the complex plane and their use for solving nonsymmetric linear systems.. Powell. Math. Reddien. 1–27. C. 23–37. NATO Conf. pp. 1983. 1690. H. Analysis aequationum universalis seu ad aequationes algebraicas resolvendas methodus generalis. John Wiley. R. Saad and M. On Newton’s method for singular problems. B. Math. Royden. 865–881. Implementation of an adaptive algorithm Buy this book from SIAM at http://www. Schultz. pp. 1973. N. Comput. pp. 1974. A. [161] T. GMRES a generalized minimal residual algorithm for solving nonsymmetric linear systems. Convergence properties of a class of minimization algorithms. 325–328. Parlett. R. A hybrid method for nonlinear equations. Englewood Cliﬀs. Smolarski. J. J. Gordon and Breach. 105–124. Computer Science Department. London. New York. [165] . [167] Y. Mangasarian. ser 4. 461–469. SIAM J. J. [147] B. Raphson. Wilkinson. E. Rheinboldt. 993–986. 1960. R. SIAM J. et expedita. 339–360. L. 437–445. Practical use of polynomial preconditionings for the conjugate gradient method. 6 (1890). [149] D. [166] . pp.. in Scientiﬁc Computing. [156] J. Anal. 87–114... pp. Peters and J. 6 (1985). Convergence of the Newton process to multiple solutions. pp. pp. 9 (1961). 155–169.html. 28–41. 71–82. N. 856–869. W.ecsecurehost. Tech. Sci. Rivlin. [169] P. 2nd ed. [163] Y. SIAM J. R. 35 (1986). pp. [158] J. [150] G. 1986..
1973. pp. Sci. Anal. Comput. 24 (1970). . E. [181] D. E. 409–426. Math. Algorithm 541: Eﬃcient FORTRAN subprograms for the solution of elliptic partial diﬀerential equations. 317–365. 14 (1978). 31 (1977). Math. Sigmon. Approximate cyclic reduction for solving Poisson’s equation. Schultz. 369–385. [180] P. S. Uber unendlich viele Algorithmen zur Auﬂosung der Gleichungen. Software. Comput. SIAM J. Schroder. 36–52.. On sparse and symmetric matrix updating subject to a linear equation. Numer. Schnabel. [182] G. 199–209. 64 (1993). B. [178] K. pp. 8 (1987). MATLAB Primer. N. Zh. Introduction to matrix computations. N. [191] H. Comput. [187] . Starke and R. Mat. A modiﬁcation of Newton’s method. FL. 21 (1984). pp.ecsecurehost. 5 (1979). 815–843. L.. A. National Bureau of Standards. Statist. D. pp. W...com/SIAM/FR16. 19 (1982). SIAM J. [190] J. 10 (1989).. 15 (1994). 490–501. Comput. Saylor. B. [189] P. [174] V. [170] R. Statist. 352–364. ACM Trans. Numer. Englewood Cliﬀs. [185] E. H. Statist. CRC Press. Sherman and W.. Numer. Fourier analysis and the FACR algorithm for the discrete solution of Poisson’s equation on a rectangle. Comput. Math. F. NJ. Applied Mathematics Series. 27–30. Frank. BiCGSTAB: A fast and smoothly converging variant to BiCG for the solution of nonsymmetric systems. pp. 213–239. Anal. [177] . 47–67. pp. Schubert. Ann. 28 (1988). 154/156 (1991). [186] P. CGS. 2 (1870). pp. Schnabel and P. Stiefel. Statist. 163–178. C.. Iterative Methods for the Solution of Equations. U. 19 (1977). K. pp. [176] J. BIBLIOGRAPHY 161 for Richardson’s method. H. Adjustment of an inverse matrix corresponding to changes in the elements of a given column or a given row of the original matrix (abstract). pp. Numer. Sherman. The methods of cyclic reduction. BIT. SIAM J. Sci.. This electronic version is for personal use and may not be duplicated or distributed.. pp. L. 49 (1958). pp. 124–127. 133–138. Sweet. Ann. Comput. Sorensen.html. Linear Algebrra Appl. Academic Press. van der Vorst. Tensor methods for nonlinear equations. Varga. [173] G. Modiﬁcation of a quasiNewton method for nonlinear equations with sparse Jacobian.. Fourth Edition..S. Ukran. Numer. SIAM J. New York. Prentice Hall. Statist. p.. Boca Raton. Math.) [175] A. 22 (1985). 755–774. (In Russian. 954–961. 621. 615–646. SIAM Rev. 21 (1950). A. Byrd. pp. SIAM J.. Stewart. Math. 19 (1967). A family of trustregionbased algorithms for unconstrained minimization with strong global convergence properties. [172] L. An optimum iterative method for solving any linear system with a square matrix. Math. Swarztrauber and R. Ann. pp. 1–22. Traub. Starke. J. R. ¨ ¨ [171] E. Math. pp. pp. A. Sonneveld. Anal. [179] D.. [184] G. Swarztrauber. Newton’s method with a model trust region modiﬁcation. Toint. Shamanskii. SIAM J. a fast Lanczostype solver for nonsymmetric linear systems. 1994. Morrison. [183] G. Sci. Alternating direction preconditioning for nonsymmetric systems of linear equations. 1964. and R. A hybrid ArnoldiFaber iterative method for nonsymmetric systems of linear equations. 20 (1949).Copyright ©1995 by the Society for Industrial and Applied Mathematics. [188] P.. C. Kernel polynomials in linear algebra and their numerical applications. SIAM J. pp. Anal. SIAM J. 13 Buy this book from SIAM at http://www. Sci.. pp.. pp. Adjustment of an inverse matrix corresponding to a change in one element of a given matrix. On Newtoniterative methods for the solution of systems of nonlinear equations. Smolarski and P.
Young. pp. 162 BIBLIOGRAPHY (1992). Academic Press. F. 9 (1989). [193] R. pp. C. S. SIAM J. P. L. Walker. Princeton NJ. S. 19 (1995). Implementation of the GMRES method using Householder transformations. pp. Lin. A simpler GMRES.Copyright ©1995 by the Society for Industrial and Applied Mathematics. Comput. Matrix Iterative Analysis. The eﬀect of rounding errors on Newtonlike methods. . 281–310. Zhou and H. 3 (1983). Comput. [192] H. Englewood Cliﬀs. F. NJ. Prentice Hall. This electronic version is for personal use and may not be duplicated or distributed. 1950. [199] M.. Journal Comput. Algorithm 652: HOMPACK: A suite of codes for globally convergent homotopy algorithms. [195] H.. J. pp. 571–581. Iterative Solution of Large Linear Systems. Woodbury. Walker and L. Morgan. 109–118. 15 (1994). 6 (1994).. 48 (1993). Math. Wachspress. IMA J. Watson.html. 297–312. Memorandum Report 42. [197] H. Billups. Buy this book from SIAM at http://www. Numer. Math. 815–825. Varga. Prentice Hall. A. [201] T.com/SIAM/FR16. and A. Ypma. Residual smoothing and peak/plateau behavior in Krylov subspace methods. A. J. pp. Statistical Research Group. [200] D. Walker. Appl. Appl. [196] . 279–286. Statist. SIAM J.. F. Englewood Cliﬀs. Sci. M. 1966. pp. Inverting modiﬁed matrices. Zhou. ACM Trans. Residual smoothing techniques for iterative methods. T.. 1971. pp. [202] L. Vuik. Alg. Software. NJ. Anal. 13 (1987). Applied Numer. pp. Sci. Iterative Solution of Elliptic Systems and Applications to the Neutron Diﬀusion Equations of Reactor Physics. 327–341. Numer. [194] E.. 1962. The superlinear convergence behaviour of GMRES. [198] L.ecsecurehost. van der Vorst and C. 631–644. New York. Math.
. 136 Bad Broyden method. 87 solution Broyden’s method. 3 Conjugate Gradient algorithm. 22 ﬁnite termination. 121 Breakdown. 77 Armijo rule. 123 restarting. 132 Banach Lemma. 145 163 Cauchy sequence. 6.ecsecurehost. 47 algorithm. 34 Contraction mapping. 110 Biconjugacy. 130 Convergence theorem Broyden’s method linear equations. 25 CGNR. 14 minimization property. 12 use of residual polynomials. 126 Brsola algorithm. 49 Chandrasekhar Hequation. 120 Broyden’s method.com/SIAM/FR16. 47 Biorthogonality.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 67 CGNE. 47 BiCGSTAB. 119 Buy this book from SIAM at http://www. 108 solution Broyden’s method. 74 algorithm. 50 algorithm. 48 algorithm. 67 ConvectionDiﬀusion Equation Broyden’s method. Index Aconjugacy. 50 ﬁnite diﬀerence. 133 Brsol algorithm. 12 Approximate inverse. 76 Condition number. 48 algorithm. 37 Backtracking. 47 Broyden sequence. 130 Newton–Armijo. 14 Conjugate Gradient Squared. 47 BiConjugate Gradient. 137 Arnoldi process. 123 limited memory. 6 BiCG. 113 convergence linear equations. 34 Chord method. 107 Characteristic polynomial. 120 implementation. 74 convergence. 47 algorithm. 67 Contraction Mapping Theorem.html. 128 Newton’s method. 118 nonlinear equations. 47 Bounded deterioration. 123 sparse. 49 Conjugate transpose. 118. 20 Anorm. This electronic version is for personal use and may not be duplicated or distributed. 87 NewtonGMRES. 25 CGS. 146 NewtonGMRES.
40 Hequation. This electronic version is for personal use and may not be duplicated or distributed. 115 Diagonalizable matrix. 8 convergence result. 79 DASPK. . 164 nonlinear equations. 80 nonlinearity. 141 NewtonGMRES. 78 Hybrid algorithms. 142 Line search. 77 contraction mapping. 46 loss of orthogonality. 91 o High order methods. 46 use of residual polynomials. 107 H¨lder Continuity. 143 twopoint. 95 Inner iteration. 8 preconditioning. 66 Forcing term. 80 directional derivative. 55 GMRES. 37 INDEX loss of orthogonality. 80 Elementary Jordan block. 54 Buy this book from SIAM at http://www. 36 Induced matrix norm deﬁnition. 146 Iteration matrix. 11 Left preconditioning. 87 solution Broyden’s method. 41 modiﬁed. 24 Jordan block. 136 Linear PDE BiCGSTAB. 9 Givens rotation. 95 choice of.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 41 minimization property. 9 Jacobi preconditioning. 80 Jacobian. 45 diagonalizable matrices. 76. 67 ﬁnite diﬀerence Newton. 104 Gauss–Seidel algorithm. 43 Globally convergent algorithm. 57 Broyden’s method. 127 CGNR. 110 Dennis–Mor´ Condition. 83 Krylov subspace. 3 Inexact Newton method. 66 Fixedpoint iteration. 33 Gram–Schmidt process. 39. 34 GMRES(m). 40. 103 secant method. 123 Line Search parabolic threepoint. 94 Inverse secant equation. 135 GMRES algorithm.ecsecurehost. 71 Newton–Armijo. 33 restarts. 34 Diagonalizing transformation. 114 e superlinear convergence. 99 Newton’s method. 87 NewtonGMRES. 79 choice of h. 81 inexact Newton’s method. 60 Fixed point.com/SIAM/FR16. 96. 60 Kantorovich Theorem. 135. 123 chord method. 36 Limited memory. 39. 34 ﬁnite diﬀerences. 82 Shamanskii method. 40. 34 Diﬀerence approximation. 5 Iteration statistics. 100 Inverse power method. 132 Inverse Tangent Newton–Armijo. 91 Jacobi algorithm. 101 ﬁnite termination. 46 algorithm. 142 polynomial.html. 128 Newton’s method.
11 as error estimate. 36. 36 Schubert algorithm. 27 for GMRES. 86 Nsola algorithm. 6 Quasiminimization. 51 Quasiresidual norm. 58 Lipschitz continuous deﬁnition. 91 rorder. 34 Nsol algorithm. 24. 11 Preconditioned CG Buy this book from SIAM at http://www. 78 convergence. 33 application to TFQMR. 71 ﬁnite diﬀerences. 24. 27 preconditioner for CG. 23 Preconditioning conjugate gradient. 25 application to GMRES. 72 NewtonGMRES. 71 algorithm. 52 QuasiNewton methods. 52 Relaxation parameter. 5 nonlinear. .com/SIAM/FR16. 24. 4 nonlinear as error estimate. This electronic version is for personal use and may not be duplicated or distributed. 27. 52 deﬁnition. 36. 100 Normal matrix. 36 Richardson iteration.html. 24. 92 termination. 113 Quasiresidual. 81 inexact. 113 Secant method. 9 Reorthogonalization. 103 ﬁnite diﬀerences. 14 application to CGNR. 81 stagnation. 74 convergence rate. 100 Matrix norm. 136 Secant equation. 24.ecsecurehost. 105 Newtoniterative method. INDEX 165 algorithm. 78 algorithm. 91 Secant update. 19 GMRES.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 72 relative nonlinear. 36 Richardson iteration. 138 Nsolgm algorithm. 101 safeguarding. 132 Newton direction. 7 Modiﬁed Bratu Problem. 127 TFQMR. 82 qorder. 54 Polynomial preconditioning. 113 Shamanskii method. 66 Poisson solver. 75 inexact. 24 Poisson solver. 3 Matrix splitting. 100 Over solving. 54 polynomial. 101 convergence rates. 133 Search direction. 7 Positive deﬁnite matrix. 136 Newton’s method. 14 Richardson iteration. 36 Jacobi. 105 Outer iteration. 66 Right preconditioning. 82 convergence. 67 Local improvement. 95 monotone convergence. 41 Residual linear. 79 solution Broyden’s method. 36 incomplete factorization. 72 Residual polynomials application to CG. 104 Picard iteration.
52 TFQMR. 16 GMRES.com/SIAM/FR16. 136 SOR algorithm. 9 SPD matrix. 124. 110 weights. 94 Newton’s method. 75 Stationary iterative methods.Copyright ©1995 by the Society for Industrial and Applied Mathematics. 66 Suﬃcient decrease.html. 7 Stagnation. 9 Symmetric matrix. 5 as preconditioners. 9 preconditioning. 51 algorithm. 12 Termination CG. . 11 Symmetric positive deﬁnite matrix. 137 Symmetric Gauss–Seidel algorithm. 132 Simple decrease. 38 linear equations. This electronic version is for personal use and may not be duplicated or distributed.ecsecurehost. 35. 166 ﬁnite diﬀerences. 12 Spectral radius. 52 Weighted norm. 98 INDEX Buy this book from SIAM at http://www. 53 ﬁnite diﬀerence. 24 Successive substitution. 132 Sherman–Morrison–Woodbury formula. 72 TFQMR. 91 Sherman–Morrison formula. 4 nonlinear equations.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.