17.

Elliptic Integrals
L. M. MILNE-THOMEION

Contents
Mathematical hpertiea . . . . . . . . . . . . . . . . . . . 17.1. Definition of Elliptic Integrals . . . . . . . . . . . . 17.2. Canonical Forms . . . . . . . . . . . . . . . . . . 17.3. Complete Elliptic Integrals of the First and Second Kinds . 17.4. Incomplete Elliptic Integrals of the First and Second Kinds 17.5. Landens Transformation . . . . . . . . . . . . . . . 17.6. The Process of the Arithmetic-Geometric Mean . . . . . 17.7. Elliptic Integrals of the Third Kind . . . . . . . . . .
page
. .

.
. .

. . .

589 589 589 590 592 597 598 599 600 600 606 608

Numerical Methods . . . . . . . . . . . . . . . : . . . . . . 17.8. Use and Extension of the Tables . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . Table 17.1. Complete Elliptic Integrals of the First and Second Kinds and the Nome q With Argument the Parameter m . . . . . . . . .

K(m), KW, 15D; dm), a(m), m=0(.01) 1

15D; Wm),E(4, 9D

Table 17.2. Complete Elliptic Integrals of the Fixst and Second Kinds and the Nome pRith Argument the Modular Angle a . . . . . .

610

K(4, K(a), d4, q1(a), Eb), EM, 15D a= Oo(l0)9Oo

Table 17.3. Parameter m With Argument K(m)/K(m) . . . . . . . 6 12

K(m)/K(m).3(.02)3, 1OD = Table 17.4. Auxiliary.Functionsfor Computation of the Nome Q and the Parameter m . . . . . . . . . . . . . . . . . . . . . . . . .

612

Q(m) (m) =qi

15D
T

L(m)=--K(m)+K*)
ml=0(.01). 15

In () $ I

lOD

Table 17.5. Elliptic Integral of the First Kind F(q\a) . . . . . . . ~ ~ = 0 ~ ( 2 ~ 5(100)850,q=0(50)900, 8D )90~, Table 17.6. Elliptic Integral of the Second Kind E(q\a) a=0(20)900,5(100)850,q=0(50)900, 8D
. . . . . .

613 616

University of Arimna. (Prepared under contract with the National Bureau of Bt.ndsrde.) 587

588

ELLIFTIC INTEQRALS

Table 17.7. Jacobian Zeta Function Z(q\a) . . . . . . . . . . . . Values of K(a)Z(q\a) a=0(20)900,5(100)850,(0=0~(5~)90~, 6D Table 17.8. Heumans Lambda Function &(q\a) . . . . . . . . . .

Psge

619

622

a=0(20)900,5(100)850,q=0(50)900 Table 17.9. Elliptic Integral of the Third Kind n(n;q\a) . . . . . . n=O(.1)1, q, a=0(150)900, 5D
625

The author acknowledges with thanks the assistance of Ruth Zucker in the computation of the examplea, Ruth E. Capuano for Table 17.3, David S. Liepman for Table 17.4, and Andrew Schopf for Table 17.9.

589

1 7 .Elliptic Integrals
Mathematical Properties
17.1. Delinition of Elliptic Integrals 17.1.5
(2-8)bJ,-a+

If R(z, y) is a rational function of z and y, where $ is equal to a cubic or quartic polynomial in z the integral ,
17.1.1

4 bi (3-28)J,-z +b,( 1-8)Js-l

(a=& 2, 3, . .
.)

++ 1-28)J,-8brJ.+i=y(Z-C)-' ba(
By means of these reduction formulae and certain transformations (see Examples 1 and 2) every elliptic integral can be brought to depend on the integral of a rational function and on three canonical forms for elliptic integrals.
17.2. Canonical Forms 1.. 721
Dcdinitions

SR(z,W

is called an elliptic integml. The elliptic integral just defined can not, in general, be expressed in terms of elementary functions. Exceptions to this are (i) when R(z, y) contains no odd powers of y. (ii) when the polynomial $ has a repeated factor. We therefore exclude these c-. By substituting for $ and denoting by p,(z) a polynomial in z we get

m=sina a; m is the parameter, C is the modular angle Y

17.2.2 17.2.3 1.. 724

x=sin 9=sn u

m cp=cn u

(1-m sina cp))=dnu=A(cp), the delta amplitude where Rl(z) and R2(z) are rational functions of 2. Hence, by expreasing %(z) as the sum of a polynomial and partial fractions
JNZ,

1.. 725

cp=arcsin (en u)=am u, the amplitude

Elliptic Inof the &t Kind

Y)dz=~l(Z)dz+2sl,J3-Ldz
+B,B,J[ (Z-C)'Yl-'dz

1.. 726

~(cp\a) =F(qlm)

=S
0

(1 -sins a sin2 e)-W

Let 1.. 712

Reduction Formulae

Jo

$=d+alz"+4+a~+a4 (I.ol+la11 #O) =~O(Z-C)' +b! (z-c)'+ bz (x-c)'+~~(z-c) +b4

1.. 713

=l
Elliptic In-1

dw=u
of the Second Kind

I =J = [v(z-C) 'I-'& , t*y-'d;G, ,

By integrating the derivatives of yz' and y(z-c)-' we get the reduction formulae 17.1.4
(8+2)d,+8+

(lbl+l~ll#O)

17.2.8

E(p\a)=E(ulm)=

(I -t') -+( 1-mt2)*dt

3)I,+2+a2(8+ 1)I,+i +4%(2?+l)I,+aaJ,-l=ZLy @=O, 1,2, . . .)

4 ai

17.2.10 17.2.11

=pnswdW =rn,u+m$~cn%dw 0

'See (17.71 Z4.72.

590
17.2.12 E(p\a)=u-m

ELLIPTIC INTEGRALS

LU

sn2wdw

In computation the modulus is of minimal importance, since it is the parameter and its complement which arise naturally. The parameter and the modular angle will be employed in this chapter to the exclusion of the modulus.
The Characteristic n

(For theta functions, see chapter 16.)

17.2.14
Elliptic Integral of the Third Kind

n(n;p\a)

=so

(1 --n sin2 e)-1 [1 -sin2 a sin2 e

1 - 1 ~

If z=sn (ulm), 17.2.15

The elliptic integral of the third kind depends on three variables namely (i) the parameter, (ii) the amplitude, (iii) the characteristic n. When real, the characteristic may be any number in the interval ( - a , a ) .The properties of the integral depend upon the location of the characteristic in this interval, see 17.7. 17.3. Complete Elliptic Integrals of the First and Second Kinds Referred to the canonical forms of 17.2, the elliptic integrals are said to be complete when the amplitude is +r and so z=1. These complete integrals are designated aa follows 17.3.1

17.2.16

n(n; ul m)= (l--nt2)-[(1-P) I

(1 -mt2)]-1/2dt

=L(l--nsn2 (wlm))-ldw
The Amplitude
(p

17.2.17

p=am u=arcsin (an u)=arcsin

can be calculated from Tables 17.5 and 4.14.

The Parameter m

[K(m)]=K= [(1 -t2) (1-mt2)1-1/2dt

Dependence on the parameter m is denoted by a vertical stroke preceding the parameter, e.g., F(vlm). Together with the parameter we define the complementary parameter ml by 17.2.18

s.

m+m1= 1

When the parameter is real, it can always be arrsnged, see 17.4, that O<m<l.
The Modular Angle a

Dependence on the modular angle a, defined in terms of the parameter by 17.2.1, is denoted by a backward stroke preceding the modular angle, thus E(v\a). The complementary d & r angle is ~12-a or 90-a according to the unit and thus ml=sin2 (90-a)=cos2 a.

We also define 17.3.5

gf=K(ml)=K(l-m)=J (l-ml sin2 e)-lw 0

17.3.6 17.3.7 ~=~(m~)=E(i-m)=

r/2

The Modulum k

K=F(hlml)=F(br\br-a)

In terms of Jacobian elliptic functions (chapter 16), the modulus k and the complementary modulus are defined by 17.2.19 k=ns (K+iK), k=dn K. They are related to the parameter by kz=m, k= m,. Dependence on the modulus is denoted by a comma preceding it, thus n(n; u, k).

s,(1-ml

sin2 O)12dB

17.3.8 E=E[K(mi)]=E(mi)=E(~\~-a)

K and iK are the real and irnagbq puarter-periods of the corresponding Jacobian elliptic functions (see chapter 16).

ELLIPTIC INTEGRALS

591

Relation to the Hypergeometric Function (see chapter 15)

17.3.9 17.3.10 17.3.11

K=4*FO,4;1; m) E=+rF(-+, 1; m) +;
Infinite Series

17.3.23
8-1

17.3.24 am u = v + c 2* sin 280where v=xu/(2K)

OD

8-1

8(1+?)

17.3.25 17.3.12 17.3.26 17.3.27 17.3.28

Mo

l K h (E-K) =O
mn -l
m+l

lim [K-4 In (16/m1)]=0

m 4
mrtl

lim m-,(K--E)=lh m-l(E-mnr,K)=r/4

m 4

lim q/m=lim pl/m1=l/16

Alternative Evaluations of K and E (see also 17.5) Legendres Relation

17.3.13

EK+ EK-KK=
Auxiliary Function

17.3.29 K(m)=2[1+m:2]-K([(1-

1 + rn:2)l*)*

17.3.14

Lm= * ()K
r

In --K(m) 16
m1

17.3.15 m= 1 -16 exp [- r(K(m)+L(m))/K(m)] 17.3.16 m=16 exp [-r(K(m) +L(ml))/K(m)] The function L(m) is tabulated in Table 17.4.
q-Series

17.3.31

K(a)=2F(wcttm (s~c/ a)\a)

17.3.32 &(a) =2E(anttan (secl a)\a)-1 +COS a

The Nome p and the Complementary Nome p , 17.3.17 17.3.18 17.3.19 17.3.20
1 1 loglo- log,, -= ! 2 P 1
(T log,,

p=p(m)=exp [-rK/K]

pl=p(ml) =exp [-rK/K]

1 1 In - In -4 .Q P I

~o=1.3862944 ai= .11197 23 a= .07252 96

bo=.5 b1=.12134 78 b2=.02887 29

e)l= 1.86152 28349 to lOD

(lo=

17.3.21

1.38629 4361 12 a, = .09666 344259 %= .03590 092383 %= .03742 563713 U4= .01451 196212

bo= .5 bl = .12498 593597 ba=.06880 248576 b,= .03328 355346 b4=.00441 787012

a The approximations 17.3.33-17.3.36 are from C. H u t tinge, Jr., Approximations for Digital Computers, Princeton Univ. Prees. Princeton, N. J. (with parmission).

*sec pogc 11.

592

ELUPTIC INTEGRALS

17.3.35 E(m)=[l +a1m1+a&l+[b1m,+

bm:] In (llml)
(r(m)[<4X10-'

+4m)
al= A6301 51 %=.lo778 12
bl=.24527 27 b2=.04124 96

17.3.36 E(m)=[l+alml+. . . +a4m:]+[blml+. . . +brm:lln a1=.44325 141463 ~1x.06260 601220 %=.04757 383546 ad= .01736 506451
( I l m i )

+t(m)

I4m) 1<2x

lo-'

b1=.24998 b,=.09200 ba=.04069 b4=.00526

368310 180037 697526 449639

17.4. Incomplete Ellipfic Integrals of the First and Second Kinds

Extension of the Tablea
Negative Amplitude

FIQUBE 17.1. cbmplete dt?ipt!&~?Mk@'d the $I'd of kind.

17.4.1 17.4.2 17.4.3 17.4.4 17.4.5

F(--(Plm) = -F(-(Plm)

E(-vlm) = -E(elm)
Amplitude of Any Magnitude

F(s*f(plm)=2~KfF((pplm)

E(u+~K) =E(u)+~E

E(u+~=') =E(u)+2i(K'-E')

17.4.6 E(u+2mK+2niKf) =E(u)+2&+2ni(K'-E') 17.4.7 E(K-u)=E-E(u)+~

I m a g i n a r y

u cd u

Amplitude

If tm B=&h
17.4.8

(p

F(iup\a)=iF(e\**-a)

17.4.9 E(+\a) = -iE(e\3*-a)+iF(e\3*-a) +i e(1--ooe2 a sin*0)' tan

Jumbi'm Imaginary Transformadon

17.4.10 E(iulm)=i[2~+ dn(ulmi)w(Ulmi) -E(ulml)

FIQUBB 17.2. Complete 4iptk integral o the f

8 d

compkp Amplitude

kind.

17.4.11

F(cp+ym) =F(Xlm) +iF(&i)

where cota A is the positive root of the equation $-[cota (p+m s h h v ~ c a ~ - m 1 1 5 - m 1 cotz~=o and m tanz p= tanz(pcot%- 1. 17.4.12
E((p+$\a) =E(X\a) -.iE(p\90"-a) bi +2 +iF(p\900-a)+2

Parameter Greater Than Unity

17.4.15 F((pplm)=m-tF(@lm-l), sin

17.4.16 E(ulm) =m+E(um+(m-i) (m- 1)u by which a parameter greater than unity can be replaced by a parameter less than unity.
Negative Parameter

where

b3

17.4.17

bl=sinza sin X cos X sina~((l-sin' a sinaX)'

bx=(l-sina a sinz X ) ( l - W S a a sina p)) sin
b3=wsZp+sina a sinz X sin' p
Amplitude Near t */2 (see also 17.5) o
p COS p

F($01-m) = (1 +m) -+K(m( 1+m) -1)

-(l+m)-+F
17.4.18

If cos a tan

(p

tan #=1

E(uI -m) = (1 +m)+ E(u(1 +m ) + l m ( m + I)-1)

-m(l+m)-tsn(.u(l+m)+lm(l

17.4.13 F((p\a) +F(#\a) =F(r/2\a) =K 17.4.14

E((P\a)+E(#\a) =E(r/Z\a) +sinaa SiW Sin#

Values when (p is near to r/2and m is near to unity can be calculated by these formulae.

whereby computations can be made for negative parameters, and therefore for pure imaginary modulus.
F((P\UI

e=m+ sin

(p

--(pi

m(l+m)-l)

+m)-')

cd(u(1 +m)'lm(l+m!-l)

FIGURE 17.3. Incomplete diptic @8t kin&.

F(o\a),

itga of ne tl

the

FIGWE 17.4. Incomplete diptic ktegral #rst kind.

P((p\a),
u conatmnt

o the f

o conetent

594

ELLIPTIC INTEGRALS

FIGURE 17.7. Incomplete elliptic btegral o the f SeCond kind.

E(v\a),
a constant

FIGURE 17.5. (p-90' F(9'a), - a constant. K

FIGURE 17.6. InCOmPkte elliptic ~?'&h?@d the o f s e e d kind. E(v\ U), (p constant
s a i l caoem p ca

17.4.22 17.4.23 11.4.24 17.4.25 11.4.26

F(iP\goO)=i arctan (&h (p>

E((P\O) = (P

E(i(p\O) =i(P E((p\90)=sin (p E(i(p\90) sinh (p =i

17.4.19 11.4.20

F(P\O)=cp
F i p O =i(p (( \ )

Jacmbi*o Zeta Function

Heuman's Lambda Function

17.4.!27 Z(v\a) =E(v\a)-E(a)F(cp\a)/K(a) 17.4.28 17.4.29 17.4.30 17.4.31 Z(ulm)=Z(u) =E@) --uE(m)/K(m) Z(-u)= -Z(u)
Z U+2K)=Z(U) (

17.4.39

ao(v\a)= F(~\90"-a) +2- K(~)z(~\w-~) Kt(a) 1

17.4.40
=- K(a)E((p\W-a)

n-

Z(K-U)= -Z(K+u)

-[ma) -E(ol)lF(v\90O--a)

596
ELLIPTIC INTEGRALS

2
n

6 + uu 'z1x "a

h
h
"

"

$2 +I+ tlt
%"a
W

Y a0 1

-Y

+ %E:
V

e .

+ E:

c t
I
V

uu
" U U

8 ." c,
r n
u

1 3

5 ;
+

ELLIPTIC INTEQRALS

597
cos (P
2 - 1 4 2-1+43
0

Some Important Special Gues

-(P

15O

17.4.53
2 5

2'+ 1

45O

17.4.54

1-2' -

l+z'

45O

43+1--2
+1+2

15O

1 -

45O

43-

l+z &+1-z

75O

17.4.56

45O

1-43-2 1+2r3-2

75O

Reduction of Sdt/@ where P=P(t) is a cubic polynomial with three real factors P= (t--81) (t-a) (t--8a) where 81>82>a. write 17.4.61

Reduction of

J&/@

when P=P(t)=P+alt*

+ad+% is a cubic polynomial with only one reel

roott=/3. We form the first and second derive tivea P'(t), P"(t) with reape& to t and then write 17.4.70 X*=[P'@)]'n,
1 m=&* a=--- 1 P"@) 2 8 [P'@)]"*

ml=cos' a -R--& =
17.4.62

17.4.71

8 -I 1- %

17.4.72 17.4.63 17.4.73 = d t 17.4.64 17.4.74 17.4.65 "dt

cu
17.5. Landen's Transformation

17.4.66 = d t 17.4.67

M a n Tdormation

Let an, a,,.+lbe two modular a n g h such that 17.5.1

(l+&u*~)(~+cos an)=2

+a (+ a< . l , d

C P
17.4.68

and let pn, pM1 two corresponding amplitudee be such that 17.5.2

tan (

~ n + l - d = ~aa

tan pa

(pn+1>Vn)

17.4.69

C P

8 The emphssis here is on the modular angle since thin is an argument of the Tablea. All formulae concerning Ianden's tramformation mny ab0 be expressed in term of the modulus k=m)=sin a and ita complement k'=m!
==Coo a.

598

ELLIPTIC INTEGRAL8

Thus the step from n to n+ 1 decreases the modular angle but increases the amplitude. By iterating the process we can descend from a given modular angle to one whose magnitude is negligible, when 17.4.19 becomes applicable. With %=awe have

With * = a we have 17.5.13

F((p\a)=2(1+sin a)-'F((q\al)
n-1

17.5.14 F((p\a)=2* II (l+sin a,)-lF((pn\%)

a-0

17.5.15

F((p\a)= II (1 +COS aJF((p,,\a,,)

8-1
m

17.5.16 F((p\a)=[csc a II sin as]J tan In

8-1

(i?r+3@)

17.5.4 F((p\a)=2-" 17.5.5 17.5.6 17.5.7 17.5.8 17.5.9

E((p\al=~('p\a)
1 +$ sin al sin a,+

S=l

6 (l+sina,)~((p,,\a,)
m
8-1

17.5.17

@=lim (pn
A +

Neighborhood of a Right Angle (see 8180 17.4.13)

F((p\a)=@ II (l+sin a,)

1 Pn @=lim - F(cp,\a,,)=lim n+m 2" n+- 2"

0

When both (p and a are near to a right angle, interpolation in the table F((p\a) is difficult. Either Landen's transformation can then be used with advantage to increase the modular angle and decrease the amplitude or vice-versa. 17.6. The Pracess of the Arithmetic-Geometric
Mean

K=F(b\a) =

a-1

II (1 +sin a,)

F((p\a)=2r-'K@

[I-z1 sin*a (1+z1 sin al

. . .)]+sin a

[i
1

Starting with a given number triple (a, co) bo, we proceed to determine number triples (al, bl, cJ, a ~ bz, CZ),. . (UN, b ~ cN) according to ( , , the following scheme of arithmetic and geometric means

(sin al)1/2sin (pl

+-1 (sin al sin a2)1/2 e+ . . . sin 22

17.5.10

17.6.1
a0

3(a0+bo) 'z 3(% 4- bl) a

a = 1
aN=3(cGN-l+bN-l)

a,
bi= ( d o ) ' bz=(aibi)'

sin al+- sin al sin a2 22

1 +g sin al sin a2sin as+
Amending Landen Traneformation

bN=(aN-lbN-l)'

...

eo c1=3(ao-bo) c2=4(a1-h)

Let am, a,,+lbe two modular angles such that 17.5.11

(l+sh a,,)(l+m~a,,+1)=2
(P,,+~

(a,,++l>a,)

CN=3(aN-,-bN--I).

and let (pn, such that 17.5.12

be two corresponding amplitudes

(Pn+l<(pJ

sin (2qm+l--(p,) =sin amsin p ,

Thus the step from n to n+l increases the modular angle but decreases the amplitude. By iterating the p r o m we can ascend from a given modular angle to one whose difTerence from a right angle is BO small that 17.4.21 becomes applicable.

We stop at the Nth step when aN= i.e., when bN, cN=O to the degree of accuracy to which the numbers are required. To determine the -complete elliptic integrals K(a), E(a) we start with 17.6.2 whenco 17.6.3
m a )=& i

%=I, b o = ca, co=sin a ~

r

ELLIPTIC INTEGRALS

599

17.7.4 To determine K'(a), E'(a) we start with 17.6.5 whence 17.6.6 17.6.7
d=1,bi=sin a, C;=COS
a

17.7.5

K'(u)=2a;v

In the above we can also use Neville's theta functions 16.36. 17.7.6
lI(n\a) =K(u) +&K(a)Z(r\a)

CMC (ii) Hyperbolic G e I >1 e

To calculate F(q\a), E(v\a) start from 17.5.2 which corresponds to the descending Landen transformation and determine +e,, (p1, . . . , vN successively from the relation 17.6.8 tan
(vn+t-vn)=(bn/an)

The case n>l can be reduced to the case u by writing ONsn < <i z 17.7.7 N=n-' sinZa, p,=[(n-l)(l-n-' 17.7.8 sin2 a)]'

tan vnpn,
*

&=v

Then to the prescribed accuracy 17.6.9 17.6.10

-n(n; v\4 = --n(N;

1

v\4 +F(v\a)
-p1 tan v)-'I

F(v\a)=csy/(2NaN)

4 g [(A(v)+m tan v) -

a v \ 4 =E(v\a)- (-/rnF(v\a)
*

where A(p) is the delta amplitude, 17.2.4. 17.7.9

=cI sin e+cZsin pi+ . . . +cNsin (pN

17.7. Elliptic Integrals of the Third Kind

Cam? (iii) Circular CMCsin' <P I <

n(n\a) =K(a> --n(N\a)

Ole<

17.7.1

r=arcain [(l-n)/Cos' a]'

0

. 4

n(n;v\a) =J'(1 -n

sin* e)-l(i -sin2 a sin*e)-+te

B= )?rF(e\\90-a)/K(a)
q=!z(d
17.7.10

17.7.2

II ( ; 4 ?r\a) I (n\a) n =I
Case (i)Hyperbolic CMC O<n<

sin'

v= 4 rF(v\a)/K(a) & = [n(1 -n) -'(n -sinz a)-']* ,

17.7.11 17.7.12 X=arctan (tanh B tan v) +2 f: (-1)'-'8-'p(l-p)-' 2m sinh 2sB sin
8-1

-n(n; v\a>=&(A-4/4

17.7.13

17.7.14

II(n\a)=K(a) 41&[1 -A~(e\a)] +

where A, is Heurnan's Lambda function, 17.4.39.

600

ELLIPTIC INTEQRALs

Special Casea

17.7.18 17.7.19

w; cp\4

n=O =F(cp\a)

n=Q, a = O

No; cp\O)=v
17.7.20 n(n; cp\O)=(l-n)-+
a=O

arctan [(l-n)' tan

cp],

= (n- 1)-i arctanh [(n- 1)' tan cp],

nl < n>l

FIGURE 17.11. EUiptic inhgrd of the third kind

Wn; cp\a).
Case (iv) Circular CPW n<O

=tancp

n=l

The case O n < can be reduced to the case sin' N a< <l by writing 17.7.15

a= u/2 17.7.21 n(n; cp\r/2) = (1 -n)-'[ln (tan q+sec q) -3ni In (l+n+sin cp)(l-n# sin cp)-']

n#1

17.7.22

n=fsin a

N=(sin' a-n) (1 -n)-I

pz=[-n(l -n)-'(sin' 17.7.16 [(l-n)(l-n-I a-n)]'

(17sin a){2n(fsin a; q\a)-F(q\a)} =arctan [(lT-sin a) tan q/A(cp)] 17.7.23 n=lfcos a 2 cos d ( 1 f c o s a; q\a)=fi In[(l+tan q *A(cp))(l-tan P - A ( v ) ) - ~ I + ~ [(A(v) +cos a -tan cp)(A(cp)-cos a tan q)-']
I n

sin' a)]'n(n; q\a) sin*a)]'n(N; q\a)

Sinz aF(q\a) +mCtan

=[(l-iV)(I-N-'

T- (1 Fcos a)F(cp\a)

+Pi'

[hSin ~v/Nv)I

17.7.17
II(n\a) = (-n cos*a) (1 -n)-'(sin* a-n)-' II(N\a)

17.7.24 II(sin' a; c \ ) ' p a=

n=sina a aE(p\a) -(tanz a sin 2q)/(2A(cp))

+sin' a(sht a-n)-'K(a)

17.7.25 n=l II(1;cp\a)=F(q\a)-se.czaE(cp\a)+sec2 a tan cpA(cp)

17.8. U e and Extension of the Tables s

Example 1. Reduce to canonical form Jy-I&,

(10+8X)2-4(3+X)(9+10h)=0;

i.e., if A=-1 7

2 1 or 3 3

where V=-3Z'+3428--119Z'+ 1722-90

and then
Q1+3

Qz=g

(z-l)',

Q1-3

Qz=z (2-2)'

By inspection or by solving an equation of the fourth degree we find that

~'=QIQ~

Solving for Q1 and Qz we get Q1= (2- 1)'+2 (2-2)*, Q =2(2- 1y-3 (2-2)' z

Where Q1=32P-IOZ+g, Q2=-Z'+&-IO

First Method

The substitution t = (2- I)/@-% then gives

Sy-'dz=*S[(t'+2)(2tD-3)1-Uf

fect ~square if ~the discriminant ) Z + ~ + ~ O Xb 8 pWQ -XQ~=( +X)~P-(~O+~X

ELLIPTIC INTEGRALS

601

If the quartic y2=0 has four real roots in 2 (or in the case of a cubic all three roots are real), we must so combine the factors that no root of Ql=O lies between the roots of Ql=O and no root of Ql=O lies between the roots of Ql=O. Provided this condition is observed the method just described will always lead to real values of A. These values may, however, be irrational.
Sccond Method

The discriminant of Q2w- Q1=2w- (2w+ 1)2+ 1 is 4 W = ( 2 ~ +1)2-4~=4++1 so that 1 W=Ad+Bw+C where A=l, B=O, C=4 and if we write zZ= W/w and 22=(B--zZ)4AC=

(zZ)2- 1 = (z2--1)(zZ+ I),

write

Jy-dz=*S[(zZ-l)(zZi.1)]- dz

and let the discriminant of Q2t*-Ql be 4Ta=(8t2+10)*--4(t2+3)(10t4+9) =4(3t2+2)(2tP- 1)

Then

~y-dz=*~T-Ldt=f~[(3t2+2)(2ts-l)]-tdl

The first method of Example 1 fails with the above values of Ql and Q2 since the root of Ql=O lies between the roots of Qa=O, and we get imaginary values of A. The method succeeds, however, if we take Q1=z, Q2=(2-1)(2-2), for then the roots of Ql=O do not lie between those of Q2=O. Example 3. Find K(80/81).
h t Method

This method will succeed if, as here, T2 as a function of t2 has real factors. If the coefficients of the given quartic are rational numbers, the factors of T2 will likewise be rational.
Third Method

Use 17.3.29 with m=80/81, m1=l/81, m:/=1/9. = Since [(l -mi/) (1 + m:/)-I2 .64, K(80/81) = 1.8 K(.64) =3.59154 500 to 8D, taking K(.64) from Table 17.1.
Sccond Method

write

and let the discriminant of Q2w-Q1 be 4 W=4(3~+2) 2 ~ - 1 )=4(Ad+Bw+ C) ( Then if zZ= W/w and 22=(B-zZ)2-4AC=(21-l)3+48

Table 17.4 giving L(m) is useful for computing K(m) when m is near unity or K(m) when m is near zero.

K(80/81)= K(80/81) (16X81)--L(80/81). h

?r

By interpolation in Tablea 17.1 and 17.4, since Sol81 =.98765 43210,

K(80/81) =1.57567 8423 L(80/81)=.00311 16543 K(80/81)=r-(1.575678423) (7.16703 7877) -.00311 16543 =3.59154 5000 to 9D.
Third Method

However, in this case the factors of 2 are complex and the method fails. Of the second and third methods one will always succeed where the other fails, and if the coefficients of the given quartic are rational numbers, the factors of Ta or 22, the case may be, will be rational. Example 2. Reduce to canonical form y-& where y2=z(z- 1) (2-2). We use the third method of Example 1 taking Qi=(~-l), Q~=s(z-~) and writing
a s

The polynomial approximation 17.3.34 gives to 8D K(80/81)=3.59154 501

Fourth Method, Arithketic-Geometric Mern

Here sin2 a=80/81 and we start with

%=l,bo=,,

c o = W = . 9 9 3 8 0 79900

giving

602

ELLIPTIC INTEORALS

1. oomo 00000 2 .44444 .43738 .43736 .43736

.66566 66665 .33333 33333

44444 79636 95008 95003 .43033 .43733 .43736 .43736

I I

. l l l l l 11111

14829 10380 94999 95003

I I

0'

.99380 .44444 . 11111 .00705 oooO2

79900 44444 11111 64808 84628

1 Thus K(80/81)=3 ~~;'=3.59154 5001.

Example 4. Find E(80/81).

Firat Method

The computation could also be made using common logarithms with the aid of 17.3.20. The point of this procedure is that it enables us to calculate ql without the loss of significant figurea which would result from direct interpolation in Table 17.1. By this means In (l/pl) can be found without loss of accuracy. Example 6. Find m to 10D when K'/K=.25 and when K'/K= 3.5. From 17.3.15 with K'/K=.25 we can write the iteration formula

m'"+"= 1 -1 6 ~ " [-WL(~'"')/K'(~'"))]. exp

Then by iteration using Tables 17.1 and 17.4

Urn 17.3.30 which gives, with m=80/81

E(80/81)= ;

E(.64)-g K(.64) 1
0 1 2 3 1. 39994 42026 .99994 42041 .99994 42041

=1.01910 6047 taking E(.64)and K(.64) from Table 17.1.

Sccond Method

Polynomial approximation, 17.3.36 gives E(80/81)=1.01910 6060. The lest two figures must be dropped tokeep within the limit of accuracy of the method.
Third Method

Thus m=.99994 42041. From 17.3.16 with K'/K=3.5 we can write the iteration formula,

Arithmetic-gwmetric mean, 17.6. The numbera were calculated in Example 3, fourth method, and we have K(80/81)-E(80/81) 1 =- [4+24+2*4+ . . . +2%] K(80/81) 2 1 =%[1.43249 712981 =.71624 85649. Using the value of K(80/81) found in Example 3, fourth method, we have E(80/81)=1.01910 6048 to 9D. Example 5. Find p when m=.9995. Here ml=.0005 and so from Table 11.4
0 1 2 3
I

0 .(3)26841 25043 3)26837 66 :!3)26837 66

Thus m=.00026 83765. The above methods in conjunction with the auxiliary Table 17.4 of L(m)enable us to extend Table 17.3 for K'/K>3, and for K'lK<.3. Example 7. Calculate to 5D the Jacobian elliptic fmction sn (.753421.7) using Table 17.5. Here m=sina a=.7, a=56.789089'. Thus, an (.753421.7)=sin (p where from
(p

&(m) =.06251 563013

~1=m,&(m)=.0000312578 15. From 17.3.19

i determined s

(i)=#/h

(i)=r'/10.37324 1132

F(~\56.789089')=.75342. Inspection of Table 17.5 shows that up lies between 40' and 45'. We have from the table of F(v\a)

=.95144 84701 q=.38618 125.

ELLIPTIC INTEGRALS

603

56' 35' 40' 45'

58'
.63945 .74138 .84788 .95974

1
A

Second Method, Numerical Quadrature

6' 0
.64085 . 74358 .85122 .96465

.63803
. 73914 .84450 .95479

Simpson'sformula with 11 ordinates and interval .1 gives .141117. Example 9. Evaluate

50'

J [(tz-2)(t2-4)]-'dt.
~

From this we form the table of F((p\56.789089')

9

First Method, Reduction to Standard Form and Bivariate Interpolation

F
.63859 .74003 .84584 .95674

--

a,

35' 40' 45'

Here we can use 17.4.48 noting that az-4, b2=2, and that

10144 10581 11090

437 509

72

50'

so'(t2-2) (t2--4)]-'dt= [
1 =- [1.854075-.535623]= .659226

A rough estimate now shows that q lioa between 40' and 41'. We therefore form the following table of F((p\56.789089') by direct interpolation in the foregoing table
9

where Thus

2 . sin (p1=-1~111 2

2 2 e=-,. a=--. 4 sin2 4

a=45', a=90,~ = 3 0 ' .

Seeond Method, Numerical Integration

40.0' 40.5' 41.0'

.74003 .75040 .76082

whence by linear inverse interpolation (p=40.5'+.5' .75.342- .75040 =40.64490 [ .76082- .750401

If we wish to use numerical integration we must observe that the integrand has a singularity at t=2 where it behaves like [8(t-2)]-'. We remove the singularity at t=2, by writing

and so sin (p=.65137=sn (.753421.7). This method of bivariate interpolation is given merely as an illustration. Other more direct methods such as that of the arithmetic-geometric mean described in 17.6 and illustrated for the Jacobian functions in chapter 16 are less laborious. Example 8. Evaluate

L4

[(tz-2) (t2- 4) ]%!t=lf(

t)dt +J4t -2)]-Ut [8(

Z

where
f (t)=[(t*- 2) (t'- 4) I-' -[8(t -2)I-'.

If we define f (2)=O,

can be calculated by numerical quadrature. Also

F h t Method, Bivariate Interpolation

From 17.4.50 we have

I4
1+

[8(t -2)] -Mt = [

$(t

&

s'

and thus we calculate the integral as

[(2t2+l)(t2-2)]-"2dt=F(~\a)-F($3\U,

where

14f

-2)' I 1 =

(t)dt= 1-.340773= .659227.

1 Jz 4 sinza=-' cos ql=-, cos e=5 3 2

Example 10. Evaluate

Thus a=26.56505 12') (pl=61.87449 43') ~ = 4 5 ' , F(*\a)=1.115921 and F(cpl\a)=.800380 and therefore the integral is equal to .141114.

2 - 7 ~ + 6 = ( ~ - 1 ) ( ~ - 2 ) ( ~ + 3 )and we use 17.4.65 with &=2, &=1, p*=-3,

604
Thus a=63.434949', c0=3O0 and

ELLIPTIC INTEGRALS

m=sin2 u=4/5, X4%/2,cos' ~ = 3 / 4 .

the integral=F(9O0\30')

=z (1.071797)(1.001292) 2
=1.68575 to 5D.

~=2(5)-+F(30'\63.434949') =2(5)-+(.543604)=.486214from Table 17.5.

The above integral is of the Weierstraas type and in fact 17= @(h; -24) (see chapter 18). 28, Example 11. Evaluate

Second Method, Aacendiqg T d o n u a t i o n

We use 17.5.11 to give

I+COS

CYn+l=2/(1+&

C, Y )

s,""(24-12t+2t2-t3)-1'P d t .
We have 24 -12t+2t2-18= -(t-2) (P+ 12)= -P(t). There is only one real zero and we therefore use 17.4.74 with P(t)=tS-2t2+12t-24, @=2 so that P'(2) = 16, P"(2) =8, X=2 and therefore
1 2 3
.33333 333 ,02943 725 .O0021 673 .94280 904 .99956 663 .99999 998

m=sin2

CY-ZI

a=30.

sin (2a -90') =sin 30, sin (2cp,-~1)=sb a sin a, 1 sin ( 2 e - e ) =sin sin e, sin (2cp4-~)=sin assin e,
From 17.5.16 2 2

n=60' e=57.367805' *=57.348426' v4=57.348425'=@.

Therefore the given integrrtl'is

F(90a\300)=E 1.99956 663 1.94280 904

where 1.99999 998 In tan(,,+;@) =1.37288 050 In tan 73.674213'
= 1.37288 050(1.22789 30)

a=3'
cos e=2j

*=70.52877 93' e=60

F(9Oo\3O0)= 1.68575 to 5D. Example 13. Find the value of F(89.5'\89.5').

First Method

and the integral=i[1.510344- 1.2125971=.148874. Example 12. Use Landen's transformation to evaluate

First Method, Demcending Tramformation

L'' -:
(1

This is a case where interpolation in Table 17.5 is not possible. We use 17.4.13 which gives
F(89.5'\89.5')=F(90'\89.5') where -F(#\89.5')

sin28)-1'2 dB to 5D.

We me 17.5.1 to give
1 +sin al=

1 +cos 30'-

-1.071797

5) cot *=sin (3')cot (.5O)=cos ( ' $=45.00109 084' and P'(*\89.5') = .881390 from Table 17.5.

~~l)]"~=.997419 F(90'\89.5') =K(sh' 89.5') =K(.99992 38476) =6.12777 88 2 1+sin a2= =1.001292; COS ~~2=.999999 1 +cos CY1 Thus F(89.5'\89.5') = 5.246389.
COS C Y ~ = [ ( ~ - S ~ I aJ(l+sh

1+sin CY3'

1 +cos a 2

=1.000000

Seeond Method

Thus from 17.5.7,

Landen's ascending transformation, 17.5.11, gives

ELLIPTIC INTEGRALS
COS

605
a, 17.7.3

al=(l-sin 89.5')/(l+sin 89.5') sin q = [(1 --COS al)(1 +COS al)]*= .99999 99997
a2=0

This is case (i) of integrals of the third kind,

Onsn < <i 2

COS

sin q = l .
17.5.12 'then gives sin (2fi-89.5')=sh 89.5' sin 89.5' = .99992 38476 2fi-89.5'=89.2929049', fic89.39645 245'

e = a r ~ ~ i(n/sin2a)*=3Oo, n B= ~F(30'\30'>/K(30') = .49332 60 ~=+rF(45'\30')/K(30') = .74951 51, 61= (1 6/45)*

and so from 17.7.3

sin @(pn-fi)=sin sin (2cpa-(pn)=sin

F(89.5'\89.5') =

a 1
(pn,

sin

fi,

(pn=89.39645602'

(Pa=yL1=@.

II(3;45'\30) =

Thus 17.5.16 gives

>' ($9996 19231

I n

(tan 89.69822 801')=5.24640.

Using the qeeries, 16.27, for the 9 functions we get

11 (f;45'\30') =(16/45)* -.02995 89 +(1.86096 21)(.74951 51)) =.813845. Table 17.9 gives .81385 with 4 point Lagrangian interpolation. Example 16. Evaluate the complete elliptic integral 11 (&\30) to 6D.

Example 14. Evaluate

From 17.4.51 the given integral

q=.O1797 24.

where
sin a=#) a=36.869900
Sin

From 17.7.6 we have

&,

fi=48.18968'

II (&\30') =K(3Oo)+(16/45)1'2K(a)z(e\300)
where e=arcain(n/sinaa)*=300. Thus using Table 17.7 II (&\30')=1.743055.

By bivariate interpolation in Table 17.6 we find that the given integral

=[ ;

.80904-.41192]=.0O496.

Table 17.9 gives 1.74302 with 5 point Lagrangian interpolation. Example 17. Evaluate
11 (g; 45'\30')

Simpson's rule-with3 ordinates gives

6 [.005O4+.01975+.OO5]=.00496.
Example 15. Evaluate
II (A; 45'\30) =

to 6D.

(l-&~in~O)-~(l--+ sin00)-*d8to 6D.

This is case (iii) of integrals of the third kind, sin0anl < <, n=#,(p=45', a=30

606

and 80

c=armin [(l-n)/cosa a]$=45O @=4~F(45~\60~)/K(30~) =.79317 74 ~=~~F(45"\30~)/r((30~) 51 = .74951 &= (40/9)* q=.O1797 24 from 17.7.11

1 (#; 45'\3O0)= (40/9)1'2(h-4~) 1 ~ 2 . 1 0 8 1 8 .55248 32-4(.03854 26) 51 (.74951 51) 1 =.921129.

Table 17.9 gives .92113 with 4 point Lagrangian interpolation. Example 18. Evaluate the complete elliptic intqpl II (#\30)to 5D. F'rom 17.7.14 we have

ELLIPTIC INTEGRALS

Here n =5 (p=45O,a=30 and since the characterz istic is greater than unity we use 17.7.7 N=n-I sin2 a=.2, pl=(1/5)* II (i); 45O\3Oo) -II( 2; 45O\3Oo)+F(45'\3O0)

=- .83612+.8O437

=1.13214. Numerical quadrature gives the same result. Example 20. Evaluate

II (#\30) =K(30)

+E 8

[l -A0(e\3Oo)

J
Here the characteristic is negative aHd we therefore use 17.7.15 with n=-1

where c=arcsin [(l-n)/co~~a]~/~=45~. using Thus Table 17.8 II (#\30)=2.80099. Table 17.9 gives 2.80126 by 6 point Lagrangian interpolation. The discrepancy results from interpolation with respect to n for p=90 in Table 17.9. Example 19. Evaluate
II

4'

1 sinaa=4

(i); 450\30)
=I" (1 i-)

sins e)-1(1-+

sin2 e)-112 d~

Using Tables 4.14, 17.5, and 17.9 we get

to 5D.

II (-*; 45O\3Oo)= .76987

References
TCSt8

[17.1] A. Csyley, An elementary treatise on elliptic functions (Dover Publications, Inc., New York, N.Y., 1956). (17.21 A. Erdblyi et al., Higher transcendental functions, vol. 2, oh. 23 (McGraw-Hill Book Co., Inc., New York, N.Y., 1953). (17.31 L. V. King, On the direct numerical calculation of elliptic functions and integrals (Cambridge Univ. Press, Cambridge, England, 1924). l17.41 E. H. Neville, Jacobian elliptic functions, 2d ed. (Oxford Univ. Press, London, England, 1951). (17.5) F. Oberhettinger and W. Magnus, Anwendung der elliptisohen Funktionen in Physik und Technik (Springer-Verlsg, Berlin, Germany, 1949).

[17.6] F. Tricomi, Elliptische Funktionen (Akademiache Verlagsgesellschaft, Leipzig, Germany, 1948). [17.7] E. T. Whittaker and G. N. Watson, A course of modern analysis, chs. 20, 21, 22, 4th ed. (Cambridge Univ. Press, Cambridge, England, 1952).
Tables

117.8) P. F. Byrd and M. D. Friedman, Handbook of elliptic integrals for engineers and physicists (Springer-Verlag, Berlin, Germany, 1954). [17.9] C. Heuman, Tables of complete elliptic integrals, J. Math. Phys. 20, 127-206 (1941). (17.101 J. Houel, Recueil de formules et de tables numbriques (Gauthier-Villars, Paris, France, 1901).

ELLIPTIC INTEGRALS

607

[17.11] E. Jahnke and F. Em&, Tables of functions, 4th ed. (Dover Publications, Inc., New York, N.Y., 1945). [17.12] L. M. Milne-Thomson, Jacobian elliptic function tables (Dover Publications, Inc., New York, N.Y., 1956). (17.131 L. M. Milne-Thomeon, Ten-figure table of the complete elliptic integrals K, K,E, E and a 1 9 Roc. h n d o n Math. table Of 8 md m S o c . 33 (1931). 2,

(17.141 L. M. Milne-Thornson, The Zeta function of Jacobi, Roc. Roy. S o c .Edinburgh 52 (1931). [17.15] L. M. Milne-Thomson, Die elliptischen Funktionen von Jacobi (Julius Springer, Berlin, Germany, 1931). [17.16] K. Pearson, Tabla of the complete and incomplete elliptic integrals (Cambridge Univ. Press, Cambridge, England, 1934). [17.17] G. W. and R. M. Spenceley, Smithsonian elliptic function tablea, Smithsonian Miscellaneous Collection, vol. 109 (Washington, D.C., 1947).