Beruflich Dokumente
Kultur Dokumente
1994
A criterion for the linear stability of relative equilibria of the Newtonian n-body
problem is found in the case when n - 1 of the masses are small. Several stable
periodic orbits of the problem are presented as examples.
KEY WORDS: Celestial mechanics; relative equilibria; stability.
1. I N T R O D U C T I O N
In 1772, Lagrange discovered his remarkable equilateral periodic solutions
of the planar three-body problem (4). For any choice of the three masses,
there is a periodic solution for which the configuration of the bodies is
always an equilateral triangle which rotates rigidly about its center of mass.
Later, Routh showed that these periodic orbits are linearly stable if one
mass is much larger than the other two (9).
In 1859, James Clerk Maxwell published his study of the rings of
Saturn (5, 6). As a first approximation, he treats the ring as a rigidly rotating regular polygon of n equal masses. He carries out a complete analysis
of the linear stability of this periodic orbit and finds that such a ring is
linearly stable provided the central mass is sufficiently large compared
to the masses in the ring. This is indeed the case, but only under the
additional assumption that n I> 7. A minor error in computing one of the
characteristic exponents caused Maxwell to miss this necessary condition.
Both the equilaterial triangle and the regular n-gon with a central
mass are examples of relative equilibria, that is, configurations which
38
Moeckel
2. RELATIVE EQUILIBRIA
Consider the Newtonian (n + 1)-body problem in the plane. Let the
mass and position of the ith body be denoted m~ and qiER 2, i = O . . . n .
39
= n,(q, p) = M - ~
p = --Hq(q, p)=VU(q)
where q = ( q o ..... qn) e R 2"+~, p = ( p o ..... Pn)~R2~+2,
Hamiltonian function,
H(q, p)= 8 9
H(q,p)
is the
U(q)
U(q) = E mirn/
i < j rr
Here rr = l q ~ - qj[ is the Euclidean distance between two of the bodies.
A relative equilibrium is a configuration x which becomes an equilibrium of Newton's equations in a uniformly rotating coordinate system.
Let R(t) denote the linear operator on R ~+2 which rotates each pair of
coordinates counterclockwise by angle t. Setting x = R(t)q and y = R(t) p,
one finds that x and y satisfy Hamilton's equations,
= M - l y -- Kx
~=VU(x)-Ky
(1)
= JVH(z)
(2)
,[o o']
and I denotes an identity matrix.
A restpoint of Eqs. (1) satisfies y = MKx and
VU(x) + M x = 0
(3)
40
~oeekel
(4)
i~l
Hall showed that for such a limiting configuration, 20 = (0, 0) and ~ lies
on the unit circle for 1 ~<i ~<n. It is possible that several of the small masses
coalesce in the limit but it is assumed here that ~ ~-2j for i # j.
To prove Hall's results in this special case, consider the ith pair of
components of Eq. (3) divided by ms:
j,,i
(5)
If i = 0, then using mj--- e/aj together with the fact that the limiting positions
are distinct gives
Xo = O ( t ) ,
20 = 0
12,1 = 1
I] = 0
(6)
Linear
41
for 1 <~i<<.n. Hall observed that this equation can be viewed as the
equation for critical points of the function:
v(o)= 5".
+ - 5".
i<jre(O) 2i<j
(7)
where the ranges of the summation indices do not include 0 and r~(O)=
2(1-cos(01-0j)). Thus the relative equilibria of the (1 + n)-body problem
are exactly the critical points of V(O).
The Morse index of a critical point of V(O) is determined by the
Hessian quadratic forrn Voo. Because of the symmetry Of the problem with
respect to simultaneous rotation of all masses, this quadratic form always
has nullity at least 1. It is appropriate to call a relative equilibrium of the
(1 + n)-body problem nondegenerate if ~i # ~y for i # j and if the nullity of
V00 is equal to 1. In particular, 2 is a nondegenerate local minimum of V(O)
provided that Voo is positive semidefinite with nullity 1. It follows from the
implicit function theorem that any nondegenerate critical point, 2, is a
relative equilibrium of the (1 + n)-body problem, that is, a family of relative
equilibria, x *, converging to ~ exists.
3. LINEAR STABILITY
A relative equilibrium of the (n + 1)-body problem determines a
periodic orbit for which the configuration rotates rigidly. The main goal of
this paper is to study the linear stability of the periodic orbits associated
to relative equilibria in a family x" converging to a relative equilibrium,
~, of the (l+n)-body problem. Equivalently, one can study the linear
stability of the associated restpoints of Eq. (1).
Using the notation of Eq. (2), the linearized equations at a restpoint
z = (x, y) are
9 =Aw
(8)
where
42
Moeckel
(0, Mx)
(Kx, 0)
(0, KMx)
Then using the homogeneity of the Newtonian potential and the fact that
x is a relative equilibrium, one finds that W~ is an invariant subspace for A.
The eigenvalues of A I ,,1 are 0, 0, + i and there is a nontrivial Jordan block
associated to the repeated 0. Thus the restpoint is not linearly stable in the
conventional sense. However, this instability arises from the fact that the
given periodic orbit is part of a family of rigidly rotating periodic orbits
with different rotation frequencies; the angular positions of nearby
solutions in this family drift away from each other and this is reflected in
the nontrivial Jordan block.
A similar drift occurs in the four-dimensional subspace, W2, spanned
by
(r 0)
(0, Me)
(r/, 0)
(0, Mr/)
where ~ = (I, 0, I, 0,...) e C 2n+2 and !/= (0, 1, 0, 1,...) e C z~+2. This subspace
is also invariant and the eigenvalues of A [ ,,2 are + i, + i with a nontrivial
Jordan block. Clearly this is associated to a drift in the center of mass.
It is traditional in celestial mechanics to view the drifts in these two
subspaces as harmless. Indeed, they can be eliminated by fixing the
momentum, angular momentum and center of mass and passing to a
quotient manifold under the action of the rotational symmetry group. Thus
it is reasonable to formulate a definition of linear stability based o n the
behavior of A in a complementary subspace. To define such a subspace, it
is necessary to introduce the skew inner product of two complex vectors,
0, W ~ C4n+4:
~2(v, w) = vrJw
A matrix of the form A =JS with S r = S is called Hamiltonian. The
Hamiltonian property is equivalent to
Aw)= --~(Av, w)
(10)
43
The proof is carried out in the rest of this section. The first step is to
establish the behavior of the eigenvalues of A ] ,, as 8 --* 0. Throughout the
proof, 8 is suppressed in the notation.
Let v E W denote an eigenvector of A with eigenvalue 2 = ~ + ifl.
Writing v = (w, w') with w, w ' ~ C 2"+2 and using (9) gives
w' = M ( K + 2 I ) w
(11)
Bw = 0
where
B = M - tD VU(x) - ( K + 21) 2
This reduces the problem from 4n + 4 to 2n + 2 dimensions. In what
follows, the quantity 12(v, ~), where ~ denotes the complex conjugate, is
significant. A short computation shows that
f2(v, ~) = 2 i f l w r M ~ - 2wrKM~:
(12)
dt-814nWn=O
44
Moeekd
and let p = (Pi ..... p,) and z = (zt ..... x,). Then w=R(p, ~), where R is an
orthogonal 2nx2n matrix. Setting C f R - ~ B R , one finds, after some
computation, that
C=[
(3 - ~2)i+ 0(8)
- 2 M + O(s)
(13)
where/~ = diag(#~ ..... #,). Moreover, writing Eq. (12) in these variables and
dividing by ~ gives
1
-~2(v,O)=2ifl(pTl~#+Zrl~{)-2zr#:+2prlJs
8
(14)
where, the O(e) arises from the terms involving We. The normalization
condition adopted above can be written
prlx:+zrUg= 1
(15)
(16)
Elimination of z from these two equations gives 2(1 +22)p = 0(8), while
elimination of p gives 22(1-I-,~2)'c=O(8). It follows from these together
with (15) that
,~2(1 + ,~2) = 0 ( 8 )
45
(17)
for any 2 ~ C. Using this, one can prove the following well-known fact [7]:
if v and w are generalized eigenvectors of A corresponding to eigenvalues
2 and # and if 2 +/z # 0, then f2(v, w) = 0. The following lemma uses these
properties of Hamiltonian matrices to derive a simple stability criterion.
Lemma 1. Suppose A is a Hamiltonian matrix such that every eigenvector v ~ C ~ of A satisfies O(v, ~ ) ~ 0 (where ~ denotes the complex
conjugate of v). Then
46
Moeckel
+2iprlz~+ 0(8)
Moreover, Eq. (15) shows that 5prize= 1 +O(8). It follows that for 8
sufficiently small, t2(v, g) # 0 as required.
If 2=O(q/~), then 2 = q / ~ , where ( 2 = ~ + o ( I ) and r is a nonzero
eigenvalue of -3/~-1V~. Since $ is a local minimum, all of the nonzero
eigenvalues of the matrix -31z-lVoo are negative. Hence ~ - i v + o(1),
where ? is real and bounded away from zero. The imaginary part of ,t is
therefore of the form
(18)
Now Eqs, (16) show that p = O(V~ ). Setting p = ~/~ (7 in the equation
o(,/7)
18ate,
= -
i vq
+ o(vq)
Since p = O(v/~), zr/zf ffi 1 + O(e) and so t2(v, ~ ) # 0 for 8 sufficiently small.
This completes the proof of the theorem.
4. EXAMPLES
This section contains several applications of the theorem.
47
89 "/3 - 89
J
which has eigenvalues 0 and #1/a2(89+ ,,//3)> 0 as required.
Vo,= ~ sin(O,--Oj) [1
j= t
j#i
r#
where r~2 = 2(1 - c o s ( 0 , - 0j)). Differentiating again gives the entries of the
Hessian Voo
Vo,oj=
cos(0~- 0j)
2r3.
2r 3
Vo~o~=-- ~ Vo~oj
j=l
j#i
- cos(0,-- 0j)
for
i# j
48
Moeckel
This is a circnlant matrix, that is, Vo,oj-- Vo,+~o~+kfor all/, j, and k with
indices interpreted modulo n. It is possible simply to write down the eigenvalues of such a matrix. The eigenvectors take the form u~ffi (p, p2,..., pn)
where p -- e (2~u/n) is an nth root of unity. Since the nth coordinate of ut is
1, the corresponding eigenvalue is just the nth coordinate of Voou~. Since
the eigenvalues are real, one finds
n-I
ln-ll_c/cjl
3 ~zl l _ c j l
~.~= ~ c j ( 1 - c j t ) + ~ j ~
r3 + ~ L
~
j~t
jffil
r~
(19)
Hence A~>0 for 1=2 ..... n/2 and it remains only to consider ;h (Maxwell
did not recognize that the case 1= 1 is special). The formula for r] gives
~l(3+c,)(1-cj)
2r]
~[l=jffil
n "~'3+c,
--]ffij=l 4rj
n
2
1 n-t
Using the formula Q=2 sin(nj/n), one finds that the second sum can be
computed explicitly. Its value is 88cot(n/2n). Hence ,11 > 0 if and only if
ln-tl
zc
An = n j ~ , ;>~nn c~ ~n + ~
and A~< 0 ff the inequality is reversed. The crucial quantity An is just an
average reciprocal distance between the points of a regular n-gon.
49
For small values of n one can evaluate both sides of the inequality
with a computer. In this way one sees that the inequality is true for n - - 7
but the reverse inequality holds for 2 ~<n ~< 6. It is now shown that the
inequality is true for n >/8. In fact the stronger inequality
1
A~>G+~
nA~-- Y~ f(j)
j~l
@
Q
0
"
o
0
"
9
9
0
9
"
O
O
86516/1-4
50
Meeekel
fT
- l f ( x ) dx < 89
f ( 2 ) + ... + f ( n - 2 ) + 8 9
An>A~ =lln
ll=~+cos(x/n) ~-~
~ csc
1 7t
One can verify that A~- is monotonically increasing and also that A~->
(I/2x) + 89 which completes the proof.
ACKNOWLEDGMENTS
This research was supported by the NSF and the Sloan Foundation.
REFERENCES
1. Andoyer, M. H. (1906). Sur les solutiones p~riodiques voisines des position d'b:luilibre
relatif dans le probleme d~ n corps. Bull. Astron. 23, 129-146.
2. Brumberg,V. A. (1957). Permanentconfigurationsin the problem of four bodies and their
stability. Soviet Astron. 1(1), 57-79.
3. Hall, G. R. Central configurationsin the planar 1 + n body problem. Boston University,
preprint.
4. Lagsange,J. L. (1873). Essai sur le probl6me des trois corps. In Ouvres, Voi.6, GauthierVlllars, Paris.
5. Maxwell,J. C. (1890). Stabilityof the motion of Saturn's rings. In Niven, W. D. (ed.), The
Scientific PaperJ of James Clerk Maxwell, Cambridge University Press, Cambridge.
51
6. Maxwell, J. C. (1983). Stability of the motion of Saturn's rings. In Brush, S., Everitt,
C. W. F., and Garber, E. (eds.), Maxwell on Saturn's Rings, MIT Press, Cambridge, MA.
7. Meyer, K., and Hall, G. R. (1992). Introduction to Hamiltonian Dynamical Systems and the
N-Body Problem, Fol. 90, Applied Mathematical Sciences, Springer, New York.
8. Pedersen, P. (1952). Stabilitatsuntersuchung im restringierten Vierk6rperproblem. Dan.
Mat. Fys. Medd. 26, 16.
9. Routh, E. J. 0875). On Lapace's three particles with a supplement on the stability of their
motion. Proc. Lond. Math. Soc. 6, 86-97.
I0. Sim6, C. (1978). Relative equuilibria of the four body problem. Cel. Mech. 18:165-184.