3.

Elementary Analytical Methods

MILTONABRAMOWITZ

Contents
PBge
Elementary Analytical Methods . . . . . . . . . . . . . . . . . 10
3.1. Binomial Theorem and Binomial Coefficients; Arithmetic and
Geometric Progressions; Arithmetic. Geometric. Harmonic
and Generalized Means . . . . . . . . . . . . . . . 10
3.2. Inequalities . . . . . . . . . . . . . . . . . . . . . . 10
3.3. Rules for Differentiation and Integration . . . . . . . . . 11
3.4. Limits. Maxima and Minima . . . . . . . . . . . . . . 13
3.5. Absolute and Relative Errors . . . . . . . . . . . . . . 14
3.6. Infinite Series . . . . . . . . . . . . . . . . . . . . . 14
3.7. Complex Numbers and Functions . . . . . . . . . . . . 16
3.8. Algebraic Equations . . . . . . . . . . . . . . . . . . 17
3.9. Successive Approximation Methods . . . . . . . . . . . 18
3.10. Theorems on Continued Fractions . . . . . . . . . . . . 19
Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . 19
3.11. Use and Extension of the Tables . . . . . . . . . . . . 19
3.12. Computing Techniques . . . . . . . . . . . . . . . . . 19
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Table 3.1. Powers and Roots . . . . . . . . . . . . . . . . . . 24
nk.k=1(1)10. 24. 1/2. 1/3. 1/4. 1/5
n=2(1)999. Exact or 10s

The author acknowledges the assistance of Peter J . O’Haraand Kermit C. Nelson in

the preparation and checking of the table of powers and roots.

1 National Bureau of Standards. (Deceased.)

9
 3. Elementary Analytical Methods

3.1. Binomial Theorem and Binomial Coeffi- 3.1.9

cients; Arithmetic and Geometric Progres-
Sum of Arithmetic Progreslrion to n Terms
sions; Arithmetic, Geometric, Harmonic and
Generalized Means aS-(a+d)+(a+2d)+ - . +(a+(n--l)d) *

Binomial Theorem =na+-1 n(n- l)d=2

n (a+I),
3.1.1 2
last term in series=Z=a+(n-1)d
Sum of Geometric Progreeeion to n Terms
+C)a"-ag3+. . . +b" 3.1.10
(na positive integer)
Binomial Coefficients (seechapter 24) lim s,=a~(l-r) (-l<r<l)
3.1.2 n-m

n(n-1). . . (n-k+l)-
- n! Arithmetic Mean of n Quantitiea A
k! (n- k)
!k!
3.1.11 A= al+az+ +an
k-n-1 n
3*1*3 (IE)=(nnk)=(-l)'( k ) Geometric Mean of n Quantitiea G
n+l 3.1.12 G= (al@ . . . an)'"' (ak>0,k=1,2,.. .,n)
3.1.4 (k )=(IE)+(knl)
Harmonic Mean of n Quantities H
3.1.13
3.1.5
C)=C)=l 1 1 1 1
-=-(-+-+
H n a l @ .. . +')an (at>O, k=1,2, . . ., n)
3.1.6 . . +c)=2"
l+c)+C) +n.
Generalized Mean

3.1.7 l-c)+c)-.
. . +(-l)"C)=O 3.1.14
3.1.15 M(t)=O(t<O, some a k zero)
Table of Binomial Coefficienta
(3 -
3.1.16 limM(t)=max.
t+
(al,%, . . .,a.)=rnc~x.a
3.1.8
3.1.17 limM(t)=min. (al,@, . . .,a,)=min. a
t+- m
11 12
-- 3.1.18 lim M(t)= G
1 4
3.1.19 M(l)=A
3.1.20 M (- 1)=H
3.2. Inequalities
Relation Between Arithmetic, Geometric, Harmonic
and Generalized Means
3.2.1
1
12 1
A 2 Q>H, equality if nnd only if a1=a2=. . . =a,,
For a more extensive table see chapter 24. 3.2.2 min. a<M(t)<mux. a
*See page II.
10
 ELEMENTARY ANALYTICAL METHODS 11
3.2.3 min. a<G<max. a Minkowski’sInequality for Sums

equality holds if all a k are equal, or O<t If p>l and U.k, br>O for all k,
and an a k is zero 3.2.12
3.2.4 M(t)<M(s) s if< t unless all a k are equal,
orO
<s and an a k is zero.

Triangle Inequalities equality holds if and only if bk=cuk (c=con-

stan t>O) .
3.2.5 la*I-lazl I la1+azlI la1I + la21
Minkowski’sInequality for Integrals

If P>l,
Chebyshev’sInequality 3.2.13
If a l 2 ~ 2 2 a 3 2. . . >a,
bl>b,>b32 . . . Ib,

3.2.7
k
k = l ukbk>($ uk) (& bk)

Holder’sInequality for Sums equality holds if and only if g(x)=cj(x) (c=con-

stant>O).
1 1
If -+-=l,p>l, q>l
P q
3.3. Rules for Differentiation and Integration
Derivatives

d du
3.3.1 - (cu)=c 9- c constant
equality holds if and only if lbkl=cluklp-l (c=con- dx dx
stant>O). If p=q=2 we get
d du dv
3.3.2 - (u+v)=-+-
Cauchy’sInequality dx dx dx
3.2.9
d dv du
3.3.3 - (uv)=u -+v -
dx dx dx
c constant).

Holder’sInequality for Integrals

1 1
If -+-=l,p>l, q>l d du dv
P P 3.3.5 - u(v)=- -
dx dv dx
3.2.10

J’ I-mdX) ldxI[ J’If(4I ’dx] [J’Id4PdX] 3.3.6 d (uo)=uo g+ln u 2)

equality holds if and enly if Ig(x) I =clf(r) 1p-l
Leibniz’sTheorem for DilTerentiation of an Integral
(c=constant>O).
If p=q=2 we get 3.3.7

Schware’sInequality
3.2.11
12 ELEMENTARY ANALYTICAL METHODS,

Leibniz's Theorem for Differentiation of a Product The following- formulas are useful for evaluating
s(a:$y'c)n
-
3.3.8 where P(x) is a polynomial and
d"
dz" '+c) 2
(uv)=g n dn-2ud2u
+(2) d T 2 @ n>l is aninteger.
3.3.16
n dn-'u d'v dnv
+ . . . +(J&z=+.
. . + u -dx"
dx 2ax+ b
arctan
(4ac- b2)t
3.3.9 !%@
dy dx
(b2-4~<0)

3.3.17

3.3.10
-2
3.3.18 =- (b2-4~=0)
2ax+ b
3.3.19
Ja2$:+c=2a 1 In
Integration by Parts

3.3.12 3.3.20
dx -=- 1 In c+dx
- (ad2bc)
J(a+bx) (c+dx) ad- bc la+bd
3.3.13 Juvdx=(s1..> v-J(sudx) $dx 3.3.21
dx 1 bx
a
Integrals of Rational Algebraic Functiona
3.3.22
(Integration constants are omitted)
dx
3.3.23
(ax+b)"+'
(n#-1)
3.3.14 S(ax+b)"dx= a(n+l) dx arctan-+
X X
a 2az(9+a9
dx -X
3.3.15

Integrals of Irrational Algebraic Functions

dx -
3.3.26
s[
(a+ bt) (c+ dx)]'/'-(- bd)'l2
arctan

3.3.27

3.3.28 =- 2 In I[bd(n+bx)]'l2+ b(c+dx)1/21 (bd>O)

(bd) 'la

3.3.29

3.3.30
 ELEMENTARY ANALY'MCAL METHODS 13

ax 3.3.44
-(dG:)2 S[(a+bx)(c+dx)]ll2
3.3.32

- ax
2b S[(a+bx) (c+d~)]'/~
3.3.33
dx
3.3.47 J (2a2E 2)'
x-a
=arcsin -
a

s (aZ"+ bx+c)'I2 3.3.443

=,-'I2 bx+c)1/2+2ax+bl (a>O)
In !2a112(az2+

3.3.34
S @ax- x2)+dx=-
2
a2 arcsin x-a
(2ax-x2)++3 -
a

3.3.49

3.3.35

3.3.36
(aO
<, b2>4ac, 12ax+ bl< (b2--4aC)'/?
3.3.37 3.3.50

J (ax2- bz+c)1/2dx=-2ax+
4a
b (ax2+bx+c)'I2 -
- 1
1
[b(~Z"+d)]++~(b~-d)+
2[b(bc--ad)]+ In [b(cx2+d)]+- ~(bc-d)'
ax (bc>d)
(az2+bx +c)
3.3.38
ax at 3.4. Limits, Maxima and Minima
1/2 where t=l/x
s x ( a 9 + bx+c) lI2---S(a+bt+ct )
Indeterminate Forme (L'Hospital's Rule)
3.3.39 3.4.1 Let f(x) and g(x) be differentiable on an
XdX interval a<x<b for which g'(x) #O.
s (az" +bx+c) If
limf(x)=O and lim g(x)=O
z4- 24-

or if
lim f(2)= OJ and lim g(x) = OJ
24- z4-
and if
ljm f ! = l then lim fM=,.
z4- g'(4 24- g(4
Both b and I may be finite or infinite.
14 ELEMENTARY ANALYTJCAL METHODB

Maxima and Minima Approximate Values

3.4.2 (1) Functions of One Variable If lel<l, ltl<1, b
a<,
The function y=f(x) has a maximum at x = q
if f’(q)=O and f”(q)<O, and a minimum at 3.5.8 (a+b)L =:d+k&’b
s=q if j’(q,)=O and f”(xJ>O. Points 4 for 3.5.9 (1 +4(1 +d = 1 +€+T
which f’(q)=O are called stationary points.
3.4.3 (2) Functions of Two Variables 3.5.10 --
l+E-l+E-B
The functionj(x,y) has a maximum or minimum 1+11
for those values of (q,, yo)for which 3.6. Infinite Series
Taylor’sFormula for a Single Variable
3.6.1

(a) f(x,y) has a maximum hn- 1

+ . . . +-(n-l)! j‘”-”(x)+R,

(b) j(x,y) has a minimum

if ->0 and ->0 at (xo,yo).
ax2 by2
3.5. Absolute and Relative Errors
(I) If q is an approximation to the true value
of x, then
3.5.1 (a) the absold error of q is Ax=xo-x,
x--% is the correction to x.
Ax Ax
3.5.2 (b) the relative error of x,, is ax=- =-
2 xo
3.5.3 (c) the percentage error is 100 times the
relative error.
3.5.4 (2) The absolute error of the s u m or
difference of several numbers is’at most equal to
the sum of the absolute errors of the individual
numbers. 3.6.6

3.5.5 (3) If f(zl, q, . . ., x,) is a function of

xl, x2, . . ., xn and the absolute error in xi 3.6.7
(i=l,2, . . . n) is Axi, then the absolute error
infis

3.5.6 (4) The relative error of the product or

quotient of several factors is at most equal to the where g(x) is any function indefinitely differenti-
sum of t.herelative errors of the individual factors. able.
Binomial Series
3.5.7 3.6.8
‘Y
(5) If y=f(x), the relative error 8y=-=- f’(‘) Ax (--1<x<1)
Y Ax) k -0
 ELEMENTARY ANALYTICAL METHODS 15
3.6.9 3.6.14
a(a-1) a(a-I)(a-2)
(1 +x)==I+ax+- x’+ 2+ . . ., 1 2 14 35
2! 3! (1+x)-+=1-- X f - 2-- 23- 2 4
3 9 81 243
3.6.10
(1+x)-1=1-x+x2-2+x4-- ... (-l<x<l)
3.6.11
Asymptotic Expansions
x 2 2 5 3 5x4 7x5 212’
(1 +x)t= 1 +---+---+---
2 8 16 128 256 1024
+.. .
m

(-l<x<l)
3.6.15 A series C
k=O
akx-k s said to be an asymp-
3.6.12 totic expansion of a functir nf(x) if
x 3x2 5 2 35x4 63x5
(1+x)-t=1--+---+---
2 8 16 128 256 n-1
f(x)-C
k=O
akx-k=o\c-.) a8 x+-

for everyn=l,2, . . .. We I Tite

3.6.13
1 1 5 10
(1+x)+=1+-5-- 2+- 2-- 5 4
3 9 81 243
+-22 x5-- 154 The series itself may be ei her convergent or
xs+. . . (-l<x<l)
729 6561 divergent.

Operations With Series

Let sl=l+alx+a&+a31+a4z”+ . . .
sz=l+blz+b22+b329+b~~+. . .
s,=1+c~x+C~22+C~l+C~x~+
. . .

Operation c2 c3

3.6.16 a: -a1
3.6.17 3 4 - 2a2
1 1
3.6.18 --a:
8

3.6.19

3.6.20 ~(n-l)clul+na2
1 *

3.6.21
3.6.22

3.6.23

3.6.24

*See page 11.

16 ELEMENTARY ANALYTICAL METHODS

Revenion of Serieo Polar Form

3.6.25 Given 3.7.2 z=refe=r(cose+isin e)
y=a;~.+b2?+cza+d;c'+ezs+fx~+~'+ . . .
3.7.3 Modulus: lzl = (2+y2)r=r
then
3.7.4 Argument: arg z=arctan (&)=e (other
x=Ay+Byg+Cy8+Dy'+E$+F~+C4y7+ notations for arg z are am z and ph z).
where
3.7.5 Real Part: X = ~ Z = T COS e
aA= 1
3.7.6 Imaginary Part: y=Yz=r sin t9
a*B=-b
a6C=2b2--ac Complex Conjugate of I

a7D=5abc-a2d-5ba 3.7.7 z=x-iy

aOE=6a2bd+3a28+ 14b4-aae-2 lab2c
3.7.8 14=1~1
a"F= 7aabe+7aacd+84ab3c-a4f -
-28a2b8-42 b6- 28a2bgd 3.7.9 arg z=-arg 2

alacf =8a'bf +8a%e+4aW +12Oa2bad Multiplication and Division

+ 180a2b28+132b*-a6g-36a3b2e
-72aabcd-12a38- 330ab'c If zl =z1+iyl, z2=z2+iy2, then
Kummer's Transformation of Series 3-7-10 +i(~ly2 ~ l y d
Z!Z~=XIZ~-YIV~
m
3.6.26 L e t F ak=sbe agiven convergentseries and 3.7.11 IZlzZl = 1211 1221
-0
gock=c be a given convergent series with known
0)
3.7.12 arg (z1zz)=arg Z l + q zz
sum c such that lim %=AZO.
k-m ck
Then
s=xc+& (1-A 2) Uk. 3.7.14

Euler'o Tramformation of Serieo 3.7.15

3.6.27 If & (-l)kak=%-@+az-
0)

. . . is a con-
Powers
vergent seriea with sum s then 3.7.16 zn=rnein0

-(-l)*Akao k 3.7.17 =rnCOS ne+irn sin ne

2k+l ' Ak%ao-x
(-I)* (m) ak-m
k
m-0 (n=O,fl,f2,. . .)
3.7.18 9=2?-yZ+i(2xy)
Euler-Maclaurin Summation Formula
3.6.28 3i7.19 23=3?-32$+i(3*-$)
3.7.20 z'=a? -6 w +y'+ i(42Y-44)

3.7.21 9 ~ 2 -1 0 2 9+52yli-i (5x'y- lox2$ $1

3.7.22

3.7. Complex Numbers and F U ~ C ~ ~ O M

Cartesian Form

3.7.1 z=x+ig/ (n=1:2,. . .)

 ELEMENTARY ANALYTICAL METHODS 17
If zn=un+ivn, then ~~+1=u,,+~+iv~+l
where 3.8. Algebraic Equations

3.7.23 ~ n + l =X U ~ - Y V ~ ; Vn+1= zt)n+yUn Solution of Quadratic Equations

9 2 " and Yzn are called harmonic polynomials.
3.8.1 Given a$+bz+c=O,
3.7.24

3.7.25 z1+ a= -b/a, z1z2=cia

If p>O, two real roots,
ROO- p=O, two equal roots,
- ,O<p pair of complex conjugate roots.
3.7.26 z*=Jz=r*e*'@=r*COS @+idsin l e Solution of Cubic Equations
If -r<e_<r this is the principal root. The 3.8.2 Given 9+a&+alz+a,,=0, let
other root has the opposite sign. The principal
root is given by 1 1
p=s al-g 4;r=%1 (a16-3h)-- 1
27 4.
3.7.27 z*=[l(r +x)]t f i[4 (r-x) ]*=u f iv where
2uv=y and where the ambiguous sign is taken to If $+r2>0, one real root and a pair of complex
be the same as the sign of y. conjugate roots,
3.7.28 zl/n=rl"Ve~n, (principal root if -w<e 5~). $+r?=O, all roots real and at least two are
Other roots are rl/nef(@+zr*'/n
(k= 1,2,3, . . ., n- 1). equal,
$+rP<O, all roots real (irreducible case).
Inequalities
Let
3-1-29 ~lzll-lz21~I lZ1fz2l512*1+1221 SI=[?-+ ($+r")'I', 82=[r- (a"+r">+l*
then
Complex Functions, Cauchy-Riemann Equation0 21=(81+82)-~ a2

f(d=f(x +iy) =N x , Y) +Wx, Y) whereu(x,y), w (x,y)

are real, is analytic at those points z=x+iy' at
which
3.7.30
au- dv bu- bv
-
ax-&' &-Xi
If Z= re *e,
If zl, a,23 are the roots of the cubic equation
b lau
bu l -
-=- du
3.7.31
br T be' re=-br 21+ 2 2 +a=-a2
Laplace's Equation z1z2+z1z3+z2z3=al
ZlZ2Z3= -UC,
The functions u(z, y) and v(z, y) are called
harmonic functions and satisfy Laplace's equation: Solution of Quartic Equations

Cartesian Coordinates 3.8.3 Given z'+a3~+a&+~z+a,,=0, find the

real root ul of the cubic equation
u3-&u2+ (a1~-4a,,)u- (a: +& - 4 a ~ )= 0
and determine the four roots of the quartic aa
Polar Coordinates solutions of the two quadratic equations
18 ELEMENTARY ANALYTICAL METHODS

If all roots of the cubic equation are real, use Method of Iteration (Successive Substitution)
the value of it1 which gives real coefficients in tlw 3.9.4 The iteration scheme xk+l =F(%k) will
quadratic equation and select signs so that if converge to a zero of z=F(x) if
(1) JF’(x)Jsq<l for asz_<b,

Newton’aMethod of Successive Approximations

3.9.5
Newton’sRule
If is an approximation to the solution
E=[ of f(x)=O then the sequence
3.9. Successive Approximath Metha&
General Comments

3.9.1 Let z=zl be an approximation to z=€ will converge quadratically to x=[: (if instead of
where f(€)=O and both xl and [are in the interval the condition (2) above),
a s x s b . We define
(1) Monotonic convergence, f(zo)fp’(x0)>0
x.+l=z.+cJ(x.) (n=1, 2, . . .I. and j’(x),f”(x) do not change sign in the
interval (20, E), or
Then, if f’(x)>O and the constants c,, are
negative and bounded, the sequence x. converges (2) Oscillatory convergence, f(aa)f”(zO)
<O
monotonically to the mot f. and f’(z),f”(z) do not change sign in the
I€ c,,=c=constant<O and f’(z)>O, then the interval (xo, q),xo 5t 5q.
process converges but not necessarily monotoni-
Newton’sMethod Applied to Real nth Roota
cally.
Degree of Convergence of an Approxirnntion Procem 3.9.6 Given x”=N, if Zk is an approximation
z= N“”then the sequence
3.9.2 Let xl,x2, x3, . . . be an infinite sequence
of approximations to a number 4. Then, if
IG+~-€I<A~~~-€I~, (n=1, 2, *)
will converge quadratically to 2.
where A and k are independent of n, the sequence
is said to have convergence of at most the kth
degree (or order or index) to E. If k=1 and
Al< the convergence is linear; if k=2 the con-
vergence is quadratic.
Regula Falsi (Fabe Position)
Aitken’s@-Process for Acceleration of sequences
3.9.3 Given y=f(x) to fina E such that f(€)=O,
choose ro and zl such that j ’r0)and f(xl) have 3.9.7 If Xk, zk+l,xp+z are three successive iterates
opposite signs and compute in a. sequence converging with an error which is
approximately in geometric progression, then
x,=x,-- (zl-zo) fl=f1aa-foZ1.
(f-f
1 0) f1-fo
Then continue with xz and either of zo or x1 for A2xk=xk-2xk+ 1 +x k +2
which f(zo)or f(q) is of opposite sign to f(z2).
Regula falsi is equivalent to inverse linear inter- is an improved estimate of x. In fact, if xk=x+*
polation. O(Ak) then Z=z+O(Ac), Ihl<l.
*See pr*e 11.
 I1 *
3.10. Theorems on Continued Fractions (4) AnBn-i-An-i13n=(-l)n-' LI ~ f i
P-1
Definitions (5) For every n>O,
3.10.1
(1) Let J=b0++,
a1
jn=
cia1
bo+cxT x+
Ci&CGL
+
=
C&Qs
* * *
Cn-lCnan.
cnbn

b2+%
(6) l+bl+btba+ .. - +b2b3 - - . bn
ba+ ' * '
=--1 b2 -
b3
. . .
bn
1 - bz+l- b3+1- -bn+l
- a1 az %
---
-bo+bl+ bz+ b3+ * a * 1 1
-+-+ . . . +-=-
1 1 21: ...
1
%-1
u1 uz Un ~ 1 -~ 1 + ~ 2 - -21,- 1 +Un
If the number of ternis is finite, .f is cidled (I 1 G xz X0
teriiiinating continued fraction. If the iiuriiber a ~ - = ~ +-aoalaa - + (-1)" aoala2. . . a,
* a

of ternis is infinite,j is ctilled m infinite continued

fraction and the terminating fraction =- 1 aoa: - alx ~
an-15
%+ al-x+ %-x+ ' ' ' +an-%
-?=bo+-
A aa
a1 - 5
j"-Bn h+ b2+ ' ' ' ba
is called the nth convergent of j.
lim An
(2) If n+m - exists, the infinite continued frac-
Bn
tionf is said to be convergent. If a r = l and the
br are integers there is always convergence.
Theorems
(1) If at and br w e positive then jZn<f2n+Z,

fin-1 >jin+l*

A.
(2) If jn=X
An= bnAn- 1+an& -2
Bn= b&-1 +anBn-z
where A-l=l, &=bo, B-l=O, Bo=l.
An-1 4 - 2 FIGURE
3.1. y=xn.
(3) [:;]=[Bn-, Bn-J [:] fn-0,
1 1
-9-1
5 2
1, 2, 5.

3.11. Use and Extension of the Tables Linear interpolation in Table 3.1 gives
(919.826)'"m5.507144.
ExamPlel* Compute X1* and x4' for X=29 By Newton's method for fourth rootg with
using Table 3.1. N=9 19.826,
x 1 L XO.Xl0

= (1.45071 4598. 101s)(4.207072333. 1014)

I4 [(5.507144)a
919'826 +a(5.507144)]=5.50714 3845

=6.10326 1248 lon *

a?'= (z")'/x
~ ( 1 . 2 5 1 8 49008*1W)*/29
=5.40388 2547. low
Example 2. Compute a?1'
(9.19826)"'= (919.826/100)1'4=
(919.826)1'4/108
Repetition yields the same result. Thus,
~'"=5.507143845/101=1.74151 1796,
~ - ~ " = ~ * / ~ = . 1 805683.
933
3.12. Computing Techniques
for x=9.19826.
Example 3. Solve the quadratic equation
22- 18.224-.056 given the coefficients aa 18.2f .1,
 20 ELEMENTARY ANALYTICAL METHODS

.056 f .001. From 3.8.1 the solution is Example 5. Solve the cubic equation 2- 18.12
x=+(lS.2f[ (18.~)2-4(.058)]+) -34.8 = 0.
To use Newton’s method we first form the
=$(l 8 . 2 f [331.016]*)= $(18.2f 18.1939) table of f(z)=2- 18.12-34.8
=18.g&9, .003
2 f(4
The smaller root may be obtained more accurately 4 -43.2
from 5 - .3
* .05fi/18.=9= .0031f .OOO1. 6 72.6
Example 4. Compute (-3 + .0076i)l. 7 181.5
From 3.7.26, (-3+.0076i)*=u+iv where We obtain by linear inverse interpolation:
y T--2 3 0- (- .3)
u=--,
2v v+) 9 T = (x2+yZ)t =5.004.
x0=5+72.6- (- .3)
Thus Using Newton’smethod, f’(x)=3x2- 18.1 we get
T= [(-a)*+ (.0076)‘]4= (9.OOOO5776)i=3.OOOOO9627
21 =s-f (XO>/f’(20)
v=r.OOOOO ””-
2 1
‘,1.73205 2196
z 5.004-
9936) =5.00526,
57.020048
u=Y= .0076 -.00219 392926 Repetition yields x1=5.00526 5097. Dividing
2v 2(1.73205 2196)- f(x) by 2-5.00526 5097 gives xa+5.00526 50972
We note that the principal square root has been 4-6.95267 869 the zeros of which are -2.50263 2549
computed. f 33036 800i.

Example 6. Solve the quartic equation I

2‘-2.37752 4922~‘+6.07350 5?41x2
- 11.17938 023~+9.052655259=0. We seek that value of p1 for which y(pl)=O.
Resolution Into Quadratic Factors Inverse interpolation in y(pl) gives y(pl)=O for
(z~+plz+pl)(z~+pzZ+pz) pI ~ 2 . 0 0 3 . Then,
by Inverse Interpolation
~~ ~~ ~

Starting with t,he triiil vdue ql= 1 we compute

QI Qz PI P2 Y(Q1)
successively -- --
2.003 4. 520 -2.550 . 172 .011

Inverse interpolation between pl=2.2 Ltnd pl=

1 5.383 2.003 gives ql=2.0041, nnd thus,
‘032
-2.023

Inverse interpolution gives q,=2.00420 2152, nnd we get finally,

91 Q2 PI P2 !/(PI)

2.00420 2152 4. 51683 7410 -2. 55283 358 . 17530 8659 A. 00000 0011

*See page 11.

 ELEMENTARY ANALYTICAL METHODS 21
Double Precision Multiplication and Division on a Method @)-If N and d axe numbers each not
Desk Calculator more than 19 digits let N=Nl+NolOg,d=dl+
Example 7. Multiply M=20243 97459 71664 32102 dolo9where No and do contain 10 digits and Nl
by m=69732 82428 43662 95023 on a 10XlOX20 and d, not more than 9 digits. Then
desk calculating machine.
Let M0=20243 97459, &f1=71664 32102, mo=
69732 82428, m1=43662 95023. Then Mm= Here
MomlOzo + (Moml+MI%)10"+Mm.
(1) Multiply M1m1=31290 75681 96300 28346 N=14116 69523 40138 1761,
and record the digits 96300 28346 appearing in d=20243 97459 71664 3210
positions 1 to 10 of the product dial. No=14116 69523, &=20243 97459,
(2) Transfer the digits 31290 75681 from posi- d1=71664 3210
tions 11 to 20 of the product dial to positions 1 to
10 of the product dial. (1) N&=10116 63378 42188 8830 (productdid).
(3) Multiply cumulatively Ml%+Moml +31290 (2) (N&)&,=49973 55504 (quotient dial).
75681=58812 67160 12663 25894 and record the (3) N- (N&/&=14116 69522 90164 62106
digits 12663 25894 in positions 1 to 10. (product dial).
(4) Transfer the digits 58812 67160 from posi- (4) [N- (NdI)/do]/dJ09=.69732 82428=ht 10
tions 11 to 20 to positions 1 to 10. digits of quotient in quotient dial. Remainder
(5) Multiply cumulatively M0m+58812 67160 =~=0883911654, in positions 1 to 10 of product
=14116 69523 40138 17612. The results as ob- dial.
tained are shown below, (5) r/(&1O9)=.436629502.10-10=next9digitsof
9630028346 quotient. N/d=.69732 82428 43662 9502. This
12663 25894 method may be modified t~ give the quotient of
14116695234013817612 20 digit numbers. Method (1) may be extended
14116695234013817612 12663258949630028346 to quotients of numbers containing more than 20
If the product Mm is wanted to 20 digits, only digits by employing higher order interpolation.
the result obtained in step 5 need be recorded. Example 9. Sum the seriea S=l-+++-t
Further, if the allowable error in the 20th place is + . . . to 5D using the Euler transform.
a unit, the operation Mlml may be omitted. The s u m of the h t 8 terms is .634524 to 6D.
When either of the factors M or m contains less If u,,=l/n we get
than 20 digits it is convenient to position the
numbers as if they both had 20 digits. This n Un Aun A%,, Aaun A%,,
multiplication process may be extended to any 9 .llllll
higher accuracy desired. -11111
10 .100000 2020
Example 8. Divide N=14116 69523 40138 17612 -9091 -505
by d=20243 97459 71664 32102. 11 .090909 1515 156
Method (1 )--linear interpolation. -7576 -349
12 .083333 1166
N/20243 97459-10"= .69732 82430'90519 39054 -6410
N/20243 97460.1010=-69732 82427 46057 26941 13 .076923
Dif€erence=3 14462 12113.
From 3.6.27 we then obtain
Difference X .71664 32102=24685 644028.10-* -111111 (- .011111) .002020
(note this is an 11x10 multiplication). S=.634524+-- 22 +28
2
Quotient= (- .000505) .000156
(69732 82430 90519 39054-246856 44028).
- 24 +,s
= ,69732 82428 43662 95028 = .634524+.055556+.002778+.000253
There is an error of 3 uriita in the 20th place due +.oooO32+ .000005
=.693148
to neglect of the contribution from second m e r -
enwe. (S=h2=.6931472 to 7D).
22 ELEMENTARY ANALYTICAL METHOD8

Example 10. Evaluate the integral lm

sy dx
r
-- to 4D using the Euler transform.
-2

+-720
1
jr- . . .
where j(k)=(k+lO)-'. Thus,

Evaluating t,he integrals in the last sum by k-'=1.54976 7731+.1

numerical integration we get
-.005+.00016 6667-.OOOOO 0333
k =1.64493 4085,

1.85194 aa compared with x= 1.64493 4067.

f

Example 12. Compute

.43379
x 2 42 92
.25661 arctan x-- ---
-1+ 34- 54- 74- * * '

. 18260 A A2 As A4

. 14180
-2587
. 11593 799
-1788 -321
.09805 478 153
-1310 -168
.08495 310
-1000
.07495
The sum to k=3 is 1.49216. Applying the
Kl=l
0
1
1
0 1.1 /=I *:

Euler transform to the remainder we obtain

1 1 1 L:l=I
.2
1
0
1 lIJ=l 3::4
-
2
(.14180)--zy (-.02587)+~ (.00799)
1 1
A*
.6 .2 1 /=I 3.032 14=.197396

= .07862
--
24
(- .00321) +5 (.00153)

=.07090+.00647+.00100+.00020
+.OW05
[ ]=I
A,
B,
3.04 1
3 032 .6
*

15.36 3.04 .36

1 7 I=/
.16 15.36 BS
21.440 I ~ = . I w ~ M
108.6144 B4

We obtain the value of the integral aa 1.57018 as Nota that in carrying out the recurrence method
for computing continued fractions the numerators
compared with 1.57080.
Example 11. Sum the series zl
f
OD

k-'=x using
the Euler-Maclaurin summation formula.
From 3.6.28 we have for n= = ,
A,, and the denominators B,, must be used as
originally computed. The numerators and de-
nominators obtained by reducing A,,/B. to lower
terms must not be uaed.
 ELEMENTARY ANALYTICAL METHODS
References
Testa
[3.1] R. A. Buckingham, Numerical methods (Pitman
Publishing Corp., New York, N.Y., 1957).
[3.2] T. Fort, Finite differences (Clarendon Press, Oxford,
England, 1948).
[3.3] L. Fox, The use and construction of mathematical
tables, Mathematical Tables, vol. 1, National
Physical Laboratory (Her Majesty’s Stationery
Office, London, England, 1956).
[3.4] G. H. Hardy, A course of pure mathematics, 9th
ed. (Cambridge Univ. Press, Cambridge, England,
and The Macmillan Co., New York, N.Y., 1947).
[3.5] D. R. Hartree, Numerical analysis (Clarendon
Press, Oxford, England, 1952).
13.61 F. B. Hildebrand, Introduction to numerical analysis
(McGraw-Hill Book Co., Inc., New York, N.Y.,
1956).
[3.7] A. S. Householder, Principles of numerical analysis
(McGraw-Hill Book Co., Inc., New York, N.Y.,
1953).
[3.8] L. V. Kantorowitach and V. I. Krylow, Naherungs-
methoden der Hoheren Analysis (VEB Deutscher
Verlag der Wissenschaften, Berlin, Germany,
1956; translated from Russian, Moscow, U.S.S.R.,
1952).
[3.9] K. Knopp, Theory and application of infinite series
(Blackie and Son, Ltd., London, England, 1951).
[3.10] Z. Kopal, Numerical analysis (John Wiley & Sons,
Inc., New York, N.Y., 1955).
[3.11J G. Kowalewski, Interpolation und geniiherte Quad-
ratur (B. G. Teubner, Leipaig, Germany, 1932).
[3.12] K. S. Kunz, Numerical analysis (McGraw-Hill
Book Co.,Inc., New York, N.Y., 1957).
[3.13] C. Lancaos, Applied analysis (Prentice-Hall, Inc.,
Englewood Cliffs, N.J., 1956).
[3.14] I. M. Longman, Note on a method for computing
infinite integrals of oscillatory functions, Proc.
Cambridge Philos. SOC.62, 764 (1958).
[3.15] S. E. Mikeladae, Numerical method8 of mathe-
matical analysis (Russian) (Cos. Izdat. Tehn.-
Teor. Lit., Moscow, U.S.S.R., 1953).
[3.16] W. E. Milne, Numerical calculus (Princeton Univ.
Press, Princeton, N.J., 1949).
[3.17] L. M. Milne-Thomson, The calculus of finite difler-
ences (Macmillan and Co., Ltd., London, England,
1951).
23
[3.18] H. Mineur, Techniques de calcul numhrique
(Librairie Polytechnique Ch. Beranger, Paris,
France, 1952).
[3.19] National Physical Laboratory, Modern computing
methods, Notea on Applied Science No. 16 (Her
Majesty’s Stationery Office, London, England,
1957).
13.201 J. B. Rosser, Transformations to speed the con+
vergence of series, J. Research NBS 46, 58-84
(1951).
[3.21] J. B. Scarborough, Numerical mathematical anal-
ysis, 3d ed. (The Johns Hopkins Press, Baltimore,
Md.; Oxford Univ. Press, London, England,
1955).
[3.22] J. F. Steffensen, Interpolation (Chelsea Publishing
Co., New York, N.Y., 1950).
[3.23] H. 6. Wall, Analytic theory of continued fraction8
(D. Van Nostrand Co., Inc., New York, N.Y.,
1948).
t3.241 E. T. Whittaker and G. Robinson, The calculus of
observations, 4th ed. (Blackie and Son, Ltd.,
London, England, 1944).
[3.25] R. Zurmiihl, Praktische Mathematik (Springer-
Verlag, Berlin, Germany, 1953).
Mathematical Tables and Collectiono of Formulam
[3.28] E. P. Adam, Smithsonian mathematical f o r m u
and tables of elliptic functions, 3d reprint (The
Smithaonian Institution, Washington, D.C., 1957).
[3.27] L. J. Comrie, Barlow’stables of squares, cubes,
square roota, cube roota and reciprocals of all
integers up to 12,600 (Chemical Publishing Co.,
Inc., New York, N.Y., 1954).
[3.28] H. B. Dwight, Tables of integrals and other mathe-
matical data, 3d ed. (The Macmillan CQ., New
York, N.Y., 1957).
(3.291 Gt. Britain H.M. Nautical Almanac Office, .Inter-
polation and allied tables (Her Majesty’s Sta-
tionery Office, London, England, 1950).
(3.301 B. 0. Peirce, A short table of integrals, 4th ed.
(Ginn and Co., Boston, Mess., 1950).
[3.3 11 G. Schulz, Formelsarnmlung xur praktischen Mathe-
matik (de Gruyter and Co., Berlin, Germany,
1945).