AN E X A C T A N A L Y T I C A L SOLUTION OF

KEPLER'S EQUATION

C. E. S I E W E R T
Dept. of Nuclear Engineering, North Carolina State University, Raleigh, N.C., U.S.A.
and
E. E. B U R N I S T O N
Dept. of Mathematics, North Carolina State University, Raleigh, N.C., U.S.A.

(Received 8 February, 1972)

Abstract. Complex-variable analysis is used to develop an exact solution to Kepler's equation, for
both elliptic and hyperbolic orbits. The method is based on basic properties of canonical solutions to
appropriately posed Riemann problems, and the final results are expressed in terms of elementary
quadratures.

1. Introduction

Kepler's equation is, of course, basic to studies of celestial mechanics and consequently
has been of interest for several centuries. For elliptic orbits, the equation is normally
written as
e sinE = E - M , (1)

whereas for hyperbolic orbits the form is

e sinh F = F + N . (2)

In Equation (1), we presume that the eccentricity ee(0,1) and the mean anomaly
M e [0,2rc] are given and thus consider the equation transcendental in E, the eccentric
anomaly. In Equation (2), we view e > 1 and the hyperbolic mean anomaly N > 0 as
given and thus seek the hyperbolic eccentric anomaly F.
To solve Equation (1), we first let

E = M + (e/z), co = (l/e), and = (M/e) (3)

and subsequently consider the equivalent problem of seeking the zeros of

A (z) = 1 + {z - coz sin-1 (l/z) (4)
in an appropriately cut plane. We note that each branch of

(5)
with log z denoting the principal branch of the log-function,

f(z)= zZ 1, and f(oo)=i, (6)

Celestial Mechanics 6 (1972) 294-304. All Rights Reserved

Copyright 9 1972 by D. Reidel Publishing Company, Dordrecht-Holland
 AN EXACT ANALYTICAL SOLUTION OF KEPLER'S EQUATION 295

is analytic in the plane cut from - 1 to 1 along the real axis. It therefore follows that
any zero zk~r ( - 1, 1) of

Ak(z)=l+~z--o)zlkrc+(--1)krC )k i ( - - 1 log I f ( z ) + lZl] ,

k = 0, • 1 , _ 2, ..., (7)

will yield, by way of Equations (3), a solution of Kepler's equation, in the elliptic form.
We note that only the branch corresponding to k = 1 in Equation (5) is such that
sin- 1 ( _ z ) = - sin- 1 z.
In a similar manner, it follows that any zero zk~ of

k=0,__l,
[ !1]
i (-- l)k log f ( z ) + ,

+ 2, . . . , (8)
can be used with
e 1 N
F =- N + i-, co = - , and ~= i (9)
z e e

to yield a solution to Kepler's equation, in the hyperbolic form. Here we use the
circumflex superscript to denote quantities basic to the solution of Equation (2).

2. Basic Analysis
Our previously reported procedure (Siewert and Burniston, 1972; Burniston and
Siewert, 1972) for solving a class of transcendental equations is based on the fact that
if an appropriate Riemann problem can be formulated, then the solution(s) of the
considered transcendental equation can be expressed in terms of a canonical solution
of that Riemann problem.
From Equation (7), we find the boundary values of Ak(z), as z approaches the
branch cut [ - 1, 1] from above ( + ) and below ( - ) , to be

A~ ( t ) = 1 -+- t [ - ~ - co~zA(k)] + (-- 1)kr 7C ~[tl T i ( - 1)kcotC (t), (10)

where we have let

A ( k ) = k + ( - 1) ~ (11)
and

F 1]
C ( t ) = In f ( t ) + ~ . (12)

We now introduce
a~ (z) = A~ (z) A~ ( - z) (13)

and consider the Riemann problem defined by the boundary condition

9 ~- (t)= Gk(t) ~- (t), te(O, 1), (14)

296 C.E. SIEWERT AND E. E. BURNISTON

where
Gk (t) = O; (t) = exp [-2i arg O~- (t)]. (15)

Here we seek a function ~k (Z)which is analytic in the plane cut from 0 to 1 along
the real axis and nonvanishing in the finite plane. Since Gk (t) is continuous (Muskhelish-
vili, 1953) and nonvanishing for re(0,1), the desired canonical solution can be written
as 1

e (z) = (1 - z) exp I ! f argf2~ (t) t -dt z 1 ' (16)

0

with arg O~-(0)=0. In addition the index Nk is such that 2riNk is the change in the
argument of Gk (t) as t varies from 0 to 1.
We now restrict our efforts to seeking real solutions of Equation (1), for ee(0,1)
and Me[0,2n], and thus in view of Equations (3) can consider in Equation (7) only
those values of k for which the resulting Ak (z) will have real zeros. It is interesting
to observe that the relevant values of k depend on the eccentricity and the mean
anomaly. The argument principle (Ahlfors, 1953) can be used to determine, for a fixed
ee(0,1) and M e [0,2n], the number of zeros of Ok(Z) in the cut plane. In general, this
number can be expressed as 2 (Nk + 1). We find, upon considering all k, that Nk can
be - 1 , 0, 1, or 2. The case Nk = - 1 is not interesting since the corresponding Ok (Z)
does not have any zeros in the cut plane. For Nk =0, we find that Ok (Z) has two real
zeros in the cut plane, and thus we must consider all such possibilities; however, since
we are seeking only the real solutions of Equation (1), we can dismiss all cases for
which Nk = 1. We conclude that we must also include all possibilities for which Nk = 2.
For the allowed ee(0,1) and M e [0,2n], we can summarize our conclusions regarding
the appropriate values of k and Nk in terms of the three open regions

Rl:e< M,
2
n 3n
R2:e> --M and e> M
2 2'
3n
R3:e<M.
2
such that
{e, M}eR1 =~ k = 1 and ~ 1 "- 2 ,

{e, M}eR2 =~ k = 0 and N 0 =0,

{e, M} e R 3 => k = 3 and N3 = 2 .

Of course, Equation (1) can be solved immediately for any of the special values of
(e, M).
Observing that the considered Ok (Z) is such that
Ok (Z) = Ok (-- Z) and Ok (Z) = Ok (5), (17)
 AN EXACT ANALYTICAL SOLUTION OF KEPLER'S EQUATION 297

where the bar denotes the complex conjugate, we conclude that fak(Z ) cP[I(--Z) is
also a solution to the Riemann problem defined by Equation (14). It therefore follows
that
E2k(z) cI'[ ~ (-- z) = cI)k (z) Pk (z), (18)
where Pk (Z) is a polynomial in z. Since ~bk(z) is nonvanishing in the finite plane, we
deduce that
Ilk+ 1
P~(z)= B~ H [zL- zq, (19)
~=1
with
Bk = ~ -- oo~A ( k ) , (20)
and we can therefore write Equation (18) as
Nk+ I
a~ (z) = ~ (z) ~ ( - ~) B~ 1Yl [ z L - z ~ ] . (21)
~=1

If we now consider the hyperbolic form of Kepler's equation, we find analogously

to the foregoing, that
O, (z) = ~, (z) 21 (- z) (22)
can be factored in the manner
N1+1
o, (~)= ~ (z) ~, (- z) ~ I1 [z~,- z~], (23)
~=1
where

~1 ( z ) - (1 - z) -~1 exp I ! f arg ~ - (t) t dt Z1 (24)

0

with arg ~ (0)=0. In regard to Equation (2), we note that only the case k = 1 in
Equation (8) need be considered, since we seek only the real solution of Kepler's
A

equation; however, we find again that the index N1 depends on the parameters, e and
N. In terms of the open regions defined by
7C
~1: e cosh N > -
2'

/)2:e cosh N < ~,

we find that
{e, N} e k l =,. N1 = O,
{e, N} eR2 =,. fr = 2.
We prefer to consider separately those special values of e and N for which
71;
{e, N} e/~s ~ e cosh N = - (25)
2'
298 C . E . SIEWERT A N D E. E. B U R N I S T O N

since here the boundary values ~0~ (t) have zeros on the cut [-1,1l. It follows that,
for {e, N} e/~s, the argument of the coefficient

(~1 (t) = ~ [ ( t ) (26)

~O~-(t) '

in the Riemann problem defined by

+
~1 ( t ) = Gx (t) ~ - (t), t ~ (0, 1), (27)

is discontinuous at
2e
t= to - . (28)

It thus follows immediately that Equation (2) admits the solutions

7r,
F=-N+_i-2, {e, N } e/~s, (29)

but, of course, we seek a real solution. A canonical solution of equation (27) can be
written as

~1 (z)-- (t o -- z) -1 exp arg ~0~-(t) t-- z '

dr] {e, N} e/~s, (30)
0

with arg .O[ (0)=0. It is important to note here, for {e, N}e/~s, that arg .O~-(t) has
a discontinuity of rc at t = to, so that arg ~ - (1) - 0 . Equation (30) can now be used to
factor ~1 (z) in the manner
.01 (Z)= ~I (Z) (~i (-- Z) ~2 [t 2 _ Z2] [Z21 - z2], {e, N } e/~s. (31)

3. Explicit Solutions

Having established the required formalism, we are now able to solve Equation (1)
almost immediately. Since our explicit solution is most concise for {e, M}eR2, we
consider that case first. Noting for this case that k = 0 and No=0, we can solve
Equation (21) to obtain
Zo21 = z 2 + 12o (z) EBb@o (z) ~o ( - z)] - I , {e, M } ~ R 2 , (32)

which can be evaluated at any convenient z to yield __+Zol, respectively the real zeros
of Ao (z) and Ao ( - z). If we now let z = iy, y real, then Equations (7) and (13) can be
used to yield

e2Qk(iY)= ~ + 1 + ~]]2 + y2 [ M -
[ e + ( - - 1 ) k y In IX/y1 rcA (k)-] 2 .

(33)
 AN EXACTANALYTICALSOLUTIONOF KEPLER'SEQUATION 299

We also wish to define

1
11
Ek (iy) = exp [ - -~z t argO~- (t) t2 + y2 de] (34)
0
with
2 (--1)k+ ltC (t) [e + (--1) k rc Itl
a r g O : (t) = tan -1
[e + ( _ 1)kT~ ]2 tEc2(t)
_ ~ltl - t ~ [M - =A (k)] ~

(35)
and arg O~ (0)=0. Equation (32) can now be used with Equations (3) to establish the
desired solution of Equation (1)
E = M - e (M - re) [e200 (iy) E~ (iy) - y2 (M - 7r)2] - 1/2 , {e, M } e R 2 .
(36)
Different choices of y in Equation (36) can, of course, alter the computational merits
of that result; the choice y = 0 yields
1

E=M-(M- r e ) e x p [ ! f a r g o o (t) ~ ] , {e, M} e R2, (37)

0
or, if we let y tend to infinity in Equation (36), we obtain the equivalent solution

E = M - e (M - ~z) [(e + 1) z (M- ~)~ -2 j,b t a r g o o+ (t) dt 1 - 1/2 , {e, M} e R 2.

0
(38)
We must now consider the remaining cases {e, M}eR1 or R3. For both of these
cases the appropriate Nk is 2, and thus we choose to evaluate Equation (21) at three
distinct points, say z = iy, y=o~, fl and ),, to obtain
Fk (i~) = [z~1 + ~2] [zkZ2+ ~2] [z23 + ~2], (39a)
e~ (~fl)= [zL +/32] [z~ +/32] [z~ + fl~], (39b)
and
Fk (i),)= [z~1 + ),2] [Z#z + ),2] [zkZ3 + ),2], (39c)
for k = 1 or 3. Here
e20k (iy) [1 + y212 E~ (iy)
Fk (iy) = [M -- zr,A (k)] 2 " (40)

Equations (39) can be reduced by elimination to yield a cubic equation in z2~. The
300 C . E . SIEWERT A N D E. E. B U R N I S T O N

resulting cubic equation can, naturally, be solved analytically; some care must be
exercised, however, in selecting the correct value of Zk~ to be used with Equations (3)
to yield the real solution of Equation (1). We find that the desired solution of Equation
(1) can be expressed as

E=M+e(2- k) [Slk (~, fl, 7) + S2k (a, fl, 7) --]A2g (~, fl, 7)]-1/2
{e, M} eRk, k = 1 or 3, (41)
where
Sjk (~, fl, Y) = [Dk (~, fl, Y) -- (-- 1) j [ D2 (~, ]3, Y) + Qk3 (a, fl, 7)]1/2] 1/3
(42)
with
Dk fl, 7) = [Alk fl, 7) A2k fl, 7) -- 3Aok fl, 7)] --
- (43)
and
Qk (o~,fl, 7)= ~Alk (o~,fl, 7 ) - [-}Azk (~, fi, 7)] z. (44)
In addition,
Aok (a, fl, 7) = a2f1272 + fl272 (f12 _ 72) TFk (ia) +
+ 72a2 (72 _ a2) TF, (ifl) + a2f12 (~2 _ f12) TFk (i7), (45a)
AI, (a, fl, 7) = a2fl 2 + fl272 + 72a2 + (f14 _ 74) TFk (ia) +
+ (74 - a 4) TF, (ifl) + (a 4 - f14) TFk (i7), (45b)
and
A2k (c~, fl, 7) = a2 + f12 + 72 + (f12 _ 72) TF k (ia) +
+ (72 _ a2) TFk (ifl) + (~2 _ flz) TFk (i7), (45c)
where
T = [(~2 _ fie) (fiE _ 72)(72 _ a2)]-i, (46)

The choice of c~, fl, and '2 in Equation (41) can, of course, alter the accuracy of a
calculational scheme based on that result; a convenient choice is a = 0 , fl= 1, and
7 = 2, for which Equations (45) reduce to

Aok (0, 1, 2) = -- Fk (0), (47a)

A lk (0, 1, 2) = 4 -- -}Fk (0) + 4F k (i) 1(2
12Fk 0 , (47b)
and
Azk (0, 1, 2) = 5 -- --}Fk (0) + -}Fa (i) l~Fk (2i) . (47C)
Equations (47) can be used to write an analytical solution of Equation (1) as

E = M + e(2 - k) [Slk (0, 1, 2) + Sza (0, 1, 2) -- 89 (0, 1, 2)] -1/2 ,

{e, M} ~ Rk, k = 1 or 3. (48)

In order to establish, for {e, M } e R 1 or Ra, a solution to Equation (1) equivalent in

form to Equation (38), we now replace c~, fl, and 7 in Equation (41) by at, fit, and 7t
and subsequently observe the limit as t tends to infinity to obtain

E = M + e (2 - k) [Slk + S2k -- ]Azk]- 1/2, {e, M} E Rk, k=lor3

(49)
 A N EXACT A N A L Y T I C A L SOLUTION OF K E P L E R ' S E Q U A T I O N 301

where
Sjk = [ D k - (-- 1) j [D 2 + Q311/211/3 (50)
D k = -~ [ A ~ k A z k - 3Aok] -- [89 3, (51)
and
Qk = -}A lk - [-}3Azk] 2. (52)
To complete the solution given by Equation (49)7 we list the explicit results
Aok = ~313k + 4IlkIak + 215g + J (k) [ M - rcA (k)] -2 , (53a)
Alk = 2I~k + 213k + [-- 2 ( e - 1) 2 Ilk --[-- } ( e - 1)3 [M - rcA (k)] -z
(53b)
and
Azk = 2 I l k - (e- 1) 2 [ M - rca (k)] -2 (53c)
where
J (k)=- 2 ( e - 1) 2 [112 + Iak] + ( e - 1) [211k + ~o] " 36 (54)
and 1

I~k = -if
rc
t ~ arg Q~- (t) dr. 2
~+1
. (55)
0

Equations (36; 37; 38) and (41; 48; 49) are our final solutions of the elliptic form of
Kepler's equation. Although we have restricted our attention to those values of e and
M of physical interest, we wish to emphasize that our method can be used for all
values, real or complex. In the same vein, although we have sought only the real
solutions of Equation (1), the method clearly is not restricted to real solutions; in fact,
even for {e, M } e R 1 or R3 the complex solutions of Equation (1) deriving from the
branches corresponding to k - 1 and k = 3 are immediately available from two of the
complex solutions of Equations (39).
We now wish to solve Equation (2), the form of Kepler's equation appropriate to
hyperbolic orbits. Except for those special values of e and N for which {e, N}e/~s,
the solutions required here follow almost immediately from the previous consider-
ations. First of all, for {e, N} e/~1, Equation (23) for z = iy, y real, yields
F = - N + e N [e2~1 (iy) E f (iy) + N2y2] - 1/2 {e, N} e/~l, (56)
after Equations (9) are invoked. Here

e E ~ l ( i y ) = [e - Y In [~/~2 + 1 + ~112 - - N 2 y 2 (57)

1

JE'I (iy) = exp Ill7c- +

t arg ~1 (t) t2 ~ y2 , at 1 (58)
0
and
2tC (t) e it[
2
arg ~ - (t) = tan -1 2 (59)
7r Itl + t2 [N z
Ie - c (t)]
302 C.E. SIEWERTAND E. E. BURNLSTON

with arg ~0~ (0) =0. We can now set y = 0 , or let y tend to infinity, in Equation (56) to
obtain two equivalent solutions of Equation (2):

F=-N+Nexp[!f argO~(t)~l, {e, N} ~.[{1 (60)

0
and
1
F=_N+eNI(e - 1)2+N2 -2; t arg .0~- (t) dt 1-1/2, {e, N } ~/~1 (61)
7C0
If we now consider {e, N} e/~2, then, in the established way, we can evaluate Equation
(23) at three distinct points, say z = ia, z= ifl, and z=i2, with c~, fl, and ]/real, and
solve the resulting equations to find
F = - N + e[Sll (a, fl, ]/) + $21 (a, fl, ]/)- ~d21 (a, fl, ]/)-]-1/2,
{e, N} e[{2 , (62)
where
/3
9

(63)
b~(~, fl, r) ~[dll (a, fl, ]/)d21 (a, fl, ]/)- 3dol (a, fl, V)] -
- [3~&, (~, fl, r)]~ (64)
and
Q1 (0~, fl, ]/)--- ~ d l l (0~, fl, ] / ) - []d21 (0~, fl, ]/)]2. (65)
In addition,
d01 (0~, fl, ]/) - - - - ~2fl2]/2 __ fl2]/2 (f12 __ ]/2) T p I (ioc) - -

- ~,~ (,~ - ~,~) vf, (~fl) - =~fl~ (o, ~ - fl:) r& (g~,) (66a)
fl** (~,, fl, v) = ~,~fl: + fl:v~ + ~,:~,: + (f14 _ r TPl (i~) +
+ (]/4 _ e4) TF1 (ifl) + (a4 _ f14) T/~I (i7), (66b)
and
d21 (0~, fl, ]/) -- -- (X2 -- f12 __ ]/})2 __ (f12 __ ]/2) T/~I (i~) --
_ (]/2 _ a2) TF1 (ifl) - (a2 _ f12) T/~I (i]/), (66c)
where
e2
F1 (iy)=- N2 ~21(iy) [1 + y212 ~12 (iy). (67)

Again, the choice a =0, fl = 1, and ]/= 2 yields

F = - N + e [$11 (0, 1, 2) + $21 (0, 1, 2 ) - 89 (0, 1, 2 ) ] - * / 2 ,
{e, N} ~/~2, (68)
with A

2o1(0,1,2) = r , (o), (69a)

flll (0, 1, 2) = 4 - 88 (0) + 4ff 1 (i) 1 F1 (2i)
12 (69b)
 AN EXACT ANALYTICAL SOLUTION OF KEPLER'S EQUATION 303

and
A2, (0, 1, 2) = - 5 + -}Fi (0) - } f l (i) + 1-~-f~ (2i). (69c)

If we now let c~, fl, and ? tend to infinity, as previously, in Equation (62), we find

F =- N + e [,.~11 q- S21 ld21 ]- 1/2, {e, N} ~/~2 ], (70)

where
Nj~ - [D~ - ( - I) j 1-132-4- 0 3 ] I / 2 ] I/3 , (71)
131 = -~ [ d l i d 2 i - 3Aol] - [3-I-d21] 3 , (72)
and
Q1 -" 11 -- 21 9 (73)
In addition,

d01 --- 4;~3

3Xll N 4[~/3 1 -- 215 1 + Y (1) N - 2 (74a)
A l l = 2/'21 + 2131 - [ - 2 (e - 1)21fl 1 + 89(e - 1)l N - 2 (74b)
and
221 = - 21,1 - ( e - 1) 2 N -2 , (74c)
where
] ( 1 ) = - 2 ( e - 1) 2 [121 + 1'31] + (e - 1) [2111 -It- 2--~] 1
36
(75)
and
1

2
Li - 1 t" arg ~ l (t) dt - (76)
7~ a+l
0

Finally we need consider the special case {e, N}eRs. If we solve Equation (31), with
z - i y , for zl 1 and use that result with Equations (9), we find
F = - N + eN[e2Di (iy) EZl(iy ) + N2y2] -1/2 , {e, N}e[~s. (77)
We can now set y = 0 , or let y tend to infinity in Equation (77) to obtain

F = - N + N exp arg {)~ (t) , {e, N} e k e , (78)

o

or

F_ N+eN[e l 2+N22f ]1/2

1

- - t arg D~ (t) dt , {e, N) e~s (79)

7g
0

Though Equations (77), (78), and (79) are the same as those derived for {e, N}e/~l,
we wish to reiterate that for {e, N } e R s the resulting arg D~-(t) has a discontinuity
of ~ at t=to, so that arg D~- (1)=0, {e, N}e/?l or Rs. Equations (56; 60; 61), (62;
68; 70), and (77; 78; 79) are our final solutions of the hyperbolic form of Kepler's
equation. Although we have restricted our attention to those values of e and N of
304 C.E. SIEWERT AND E. E. BURNISTON

physical interest and have sought only real solutions of Equation (2), the method is
not limited to either real parameters or real solutions.
A Gaussian quadrature integration procedure has been used to evaluate numerically
all of our explicit solutions, for numerous cases, and accuracy to within ten significant
figures was achieved, quite straightforwardly.

Acknowledgements

The authors wish to express their gratitude to Y. Ishiguro and J. T. Kriese for numer-
ically evaluating our solutions to obtain known results, and to J. M. A. Danby,
T. W. Mullikin and H. Pollard for the interest shown in this work.

References
Ahlfors, L. V.: 1953, Complex Analysis, McGraw Hill, New York.
Burniston, E. E. and Siewert, C. E.: 1972, Proe. Camb. Phil. Soc. (in press).
Muskhelishvili, N. I.: 1953, Singular Integral Equations, Noordhoff, Groningen, The Netherlands.
Siewert, C. E. and Burniston, E. E.: 1972, Astrophys. Y. 173, 405.