CORRECTIONS

AN INTRODUCTION TO NUMERICAL ANALYSIS, 2ND ED.

Page

Line

Corrected version of text

5*

(1.1.7)

n+1
1
(x
2

+
8

xn+1
(1 + x )n+1
6)2

@2f ( ; )
+ 2(x
@x2

2)

@2f ( ; )
@x@y

-3, -4
+(y

10

13

13

Table 1.1

16

14, 15

28*

(1.4.20)

f (xT ; yT )

32

: : : has a mean of

32

(1.5.10)

32*

6)(y

(1.5.12)

kvk

2)2

@2f ( ; )
@y 2

0; kvk = 0 if and only if: : :

For the rst entry for PRIME 850, the value should be 1:19E
:
= 2 22 = 2:38 10 7 chopped arithmetic
:
= 2 23 = 1:19 10 7 rounded arithmetic
:::

E
p

: : : xT x =

:::
"
m
P

j=1

j)

=2; : : :

x2j

#1=2

33

(1.5.15)

: : : (1 +

36*

K(x) = Supremum

44

56

1726

64

20

67

table heading

: : : g(x) dx = : : :
This paragraph should not be indented because it is not part
of the denition preceding it.
f (xn )
xn =
f 0( n)
Replace xn xn 1 with xn+1 xn .

71*

:::

75

16

98*

1. jf (z)j

x=x
= Supremum : : :
y=y
y

f (xn )
f 0 (xn )

n
10

10

1:001

7.

Page

Line

100*

(2.9.22)

113

: : : Davidon: : :

123

Replace x
^n

123

14

g(x)

126

Prob. 47, Line 2

132*

(3.1.4)

Corrected version of text

+

:002j 6 ( 1)j 1
(j 1)!(7 j)!

with x
^n .

a = (x

a)h(x)

: : : y = :5 + h tan
Y
:::

:::
2

(x + y 2 )

0 j<i n

140*

Table 3.1, col. 3

Add entry f [x4 ; x5 ] at the bottom of the column.

148*

1
1
(r + 1)h r!hr

150*
163

2
4

: : : ( :000005) = 1:466288
have been used in a: : :

165

Replace

165

(3.7.8)

:::

166

(3.7.10)

:::

xi

xi

with

Max
1

t xi

Max
1

t xi

Maxf

f1

:::
:00000012

i.

f (4) (t)
f (4) (t)

176*

k 3; : : :

183

: : : the most widely used: : :

191

Prob. 32(a), Line 4

192*

Prob. 38, Line 5

: : : x0 x x3 :
Z xn
2
[^
s00 (x)] dx
x0

207

: : : Given f 2 C[a; b]; : : :

q
: : : '2 (x) = 12 52 (3x2 1)

210*

211

(4.4.14)

Replace m = m = 0 with m = n = 0.

216

10, 11

Replace orthogonal with orthonormal.

216

Change bj ) to bj .

221

-3

Change

226

11

227*

: : : theory of Fourier series: : :

p
:
: : : 2 1 = 0:414. Then

227*

n
X

to

j=0

<

n
X
0
j=0

2n+5
2
2
2n + 5 1

Page

Line

Corrected version of text

228*

:::

232*

(4.7.31)

2n+3

2n + 3
=

n
P
0

:::

cn;k Tk (x)

k=0

241

Prob. 11(a), Line 2

258*

q1 (x) = :955 + :414x

n=2
P

f (4) ( j )

j=1

258

(5.1.18)

261*

272*

(5.3.12)

:
Replace = with =.
p
: : : f (x) = x3 x : : :
+

n
P

~ j (x)
f 0 (xj )h

j=1

313*

(5.6.28)

315
320
338*

2
3
12

the cases w(x) : : :

Change h = :0022 to h = :022
= f (x; Y (x; )) : : :

338
341

10
17

Replace Z(x) with Z(x; )

Change given to give.

348

Change at least to about.

348

: : : leads to (6.2.24).

350

14

Further assume that Y0 = y~0 . Then

352

: : : h = :01 case: : :

352

(6.2.34)

360

jB(x; h)j

361

16

362

6, 7

366

370*

(6.5.18)

370

12

372

17

372

table heading

+ h[

:::

1 (j

1) + : : :

chp ;

yh (xn )j

x0

:::

: : : each step has an error

To analyze the convergence of (6.4.2), we use Theorem 6.6.
From (6.4.1), (6.3.4), (6.3.5), we easily obtain that
Change decrease to oscillate.
ih 2
i
: : : h12 Y (3) 1
: : : Assuming e0 =
=

1
3

[yh (xn )

0h

+ :::

:::

Change last column heading to

1
3

[yh (x)

y2h (x)].

Page

Line

Corrected version of text

377

Step 20

Append: Also, x1 := x0 + h.

379*

: : : Max jY (t)

384*

: : : and Yj0 = Y 0 (xj )

388

with xn

399*

: : : (6.8.24) is simply

403

: : : solution (6.8.22).

408*

: : : of h are

412

(6.9.9)

419

10

with r = 2 yields

421*

:::

423

: : : +fy4 f 2

425

(6.10.27)

:::

426
426
428

13
13
5

433

15

+ h n (Y )
Remove the equation label (6.10.30)
large nor too small.: : :
p
P
=
j Vj

un (t)j

xn t x

p+1

xn

xn+1 : : :

1 < h < 0.

rp

+h

1 f (x; y)

+ :::

1=2

F (x; z; h; f )j

:::

j=1

y 0 (0) = y 0 (1) = 0;

436*

437

jy(x)j <

438

(6.11.18)

438

10

: : : values are obtained from those in (6.11.16): : :

452

Prob. 14, Line 2

yn+1 = 12 (yn + yn

469*

x = (1; 2; 3)

470*

(x; u(j) ) =

490

493

k(I

494*

=A

499*

Prob. 13(b), Line 2

: : : then Y (x; s ) will satisfy

'(sm )
: : : sm
:::
'0 (sm )

1 (u

(1)

1)

+ :::

; u(j) ) +

n (u

(n)

; u(j) )

By examining in detail the struture of D and N , based on

their origin in the Jordan canonical form of A, we have
DN = N D. Then
1

A)
1

(B

: : : x; y 2 Rn

m+1

k kAk

A)B

Page
501*

Line
Prob. 21

511

517

519*

Corrected version of text

Prove the following: for x 2 Cn
3
2
u1j
6 .. 7
6 . 7
7
6
6 ujj 7
7
6
:::6
7
6 0 7
6 . 7
4 .. 5
0
of A by an appropriate: : :
aij
bij =
;
j = 1; : : : ; n
si
(k)

519

(8.2.12)

:::

aik

(k)

si

Insert the following sentence into Line 2, following

(k)
rows i and k: Also interchange the values of si
(k)
and sk , and denote the resulting new values by
(k+1)
sj
; j = 1; : : : ; n, most of which have not changed
[see step 10 of the algorithm Factor, given below].

520

521

10

17. det := ann det; ier := 0 and exit the algorithm.

522

13

Gauss-Jordan method: : :

525

11

: : : (8.3.9)(8.3.12): : :

525

: : : the rst row of L times: : :

525

(8.3.14)

: : : j = 1; : : : ; i

530

13

:::

kek
kxk

530

kIk = kAA

537

: : : bound (8.4.23) is: : :

537

Change k^
xk to k^
xk1 .

538*

544*

kRk
kAk kCk
2
=

6
4

:::

k:::

kCk
1
4
1
2
1
4

1
2

0
1
2

1
4
1
2
1
4

:::
3
7
5

Page

Line

548*

12

Corrected version of text

iP1
aij
i =
aii : : :
j=1

552

553*

10

: : : rather than the much smaller: : :

561*

10

: : : the bilinear interpolation: : :

564

Replace that with than.

567

14

: : : and van der Vorst

574

575*

Prob. 4

580*

Prob. 29, Line 3

: : : van der Vorst (1986): : :

R1
: : : 0 cos( st)x(t) dt : : :

581*

Prob. 31, Line 3

x(k+1) = b + Ax(k) ;

582*

Prob. 34(a), Line 1

: : : in (8.8.1) to the

590

1821

593
600

1
15

600

(9.1.41)

partial dierential equations.

: : : kAi k + kCi k <

1
;:::
kBi 1 k
k

Let S be a connected union of m of the circles, all of which

are disjoint from the remaining n m circles. Each path
f i( ) j 0
1g which begins at a center aii
i
within S must remain within it. To see this, rst note from
above that i must remain in the union of all Zk (1). If i
does not remain in S, then it must be in one of the remaining
n m circles for some values of . But this will contradict
the continuity of i ( ) as S is not connected to the remaining
n m circles. Since the number of eigenvalues counted as
roots of det(A
I) is constant, the above argument shows
that the number of such eigenvalues within each connected
component S must remain constant for 0
1. This
proves the second result.
kEk2 = : : :
are the normalized eigenvectors: : :
n
P
= uk + "ak uk +
j=1
j6=k

603

Let

603

z (m) =

be a: : :
w(m)

:::

Page

Line

Corrected version of text

603

10

z (1) =

w(1)

= :::

603*

:::

603

603

A2 z (0)
kAz (0) k1

:::

z (2) =

w(2)

= :::

2
(m)
1

Am z (0)

604*

610*

611*

11

w=

616

T is related to A by

621

(1; 1; 0; 1; 1; 0; 1)

627

(9.5.16)

627

10

: : : as m ! 1

629

11

630

(9.6.11)

632

632

632

10

632

11

632

635

11

637

15

= 0r
0r
v

kAm 1 z (0) k1 k

^
1; w

:::

T T

From (9.5.13),
(9.2.2)(9.2.3); and for simplicity in analysis and
implementation, we replace m by kw(m+1) k1 .

: : : U w(1) = e
E z^ +

z^(m)
kwk
^ 2

p
Using kz (m) k2
nkz (m) k1
residual satises
p
n
+
kwk
^ 2
p
n
+
kwk
^ 2
= A^
z

z^ = : : :

: : : For any x 2 Rn and any: : :

Replace (8.7.20) with (9.7.20)

n, the

Page
639

Line
9

639

Table 9.3

640

Figure 9.2

Change label on horizontal scale from x to t.

643

11

mn2 +

644

11

644

9, 8

649*

655

-4

664*

15

665*

26

669

Prob. 39

Corrected version of text

: : : From (9.7.19), if: : :
Replace the table headings as follows:
xi ! ti , yi ! bi

n3
3
T T
: : : [x A Ax

0 for all x] : : :

above the diagonal. Thus R has the form of the

matrix F of
p (9.7.5). We will then have R = F
with i =
i . Letting B = AU in (9.7.42), we
have the desired SVD:
uk ( ) = uk (0) + : : :
+ AT A)

AT = A+

NETLIB@ORNL.GOV on INTERNET
The Mathworks, Inc.
3 Apple Hill Drive
Natick, MA 01760-2098
URL: www.mathworks.com
E-mail: info@mathworks.com
xi+m x
x xi
(m)
(m 1)
(m
Bi (x) =
B
(x) +
B
xi+m xi+1 i+1
xi+m 1 xi i

1)

(x)