Downloaded 11/25/14 to 169.233.7.23. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.

php

SIAM J. MATH. ANAL.

Vol. 12, No. 2, March 1981

1981 Society for Industrial and Applied Mathematics

0036-1410/81/1202-0001 $01.00/0

A UNIQUENESS THEOREM FOR

ORDINARY DIFFERENTIAL EQUATIONS*
M. J. NORRIS,"

AND

R. D. DRIVERS

Abstract. The uniqueness theorem of this paper answers an open question for a system of differential
equations arising in a certain n-body problem of classical electrodynamics. The essence of the result can be
illustrated using the scalar prototype equation x gl(x)+ g2(t + x) with x(0)=0. The solution of the latter
will be unique provided gl and g2 are continuous positive functions of bounded variation.

The theorem proved in this paper presents a criterion weaker than a Lipschitz
condition which assures uniqueness of solutions of a system of ordinary differential
equations. It was designed to resolve an open question in classical electrodyamics
described at the end of the paper.
Before stating the theorem, let us illustrate it with two scalar examples typifying the
problems we had in mind. These examples are easily treated with the theorem which
follows. We are unaware of any previous uniqueness theorem which would handle them
or the electrodynamics problem of Example 3.
Example 1. If gl and g2 are continuous positive functions of bounded variation on
an open interval containing 0, then the equation
with x(0)

x=g(x)+g2(t+x)

has a unique solution on some open interval containing 0.

Example 2. The equation

x= (t + x5/3) /3

for t_-> 0 with x(0)

has a unique solution.

The theorem itself treats a system of n ordinary differential equations

x=f(t,x)

(1)
with initial conditions

X(to)=Xo.
Let S be a subset (not necessarily open) of R n/l, and let f" S--> R n. Then, given
(to, x0) S, a solution of (1) and (2) is defined as any differentiable function x on an
interval J such that (t, x(t)) S and x= f(t, x(t)) for J, while to J and X(to) Xo. (If J
contains either of its endpoints, x(t) is a one-sided derivative there).
(2)

R" is I111- Ei=I Ii].

The norm used in this paper for a vector
THEOREM. Let f" S--> R be continuous and satisfy the following condition. Each
point in S has an open neighborhood U, a constant K > O, an integer m >-_ O, and functions
m such that
1,.
h and gi for

(3)

Ill(t, )-f(t, n)ll<-_gll-nll+g

]g(h(t, ))-g(h(t,

* Received by the editors May 30, 1980.

t Applied Mathematics Department 5640, Sandia National Laboratories, Albuquerque, New Mexico
87115. The work of this author was supported in part by the U.S. Department of Energy under contract
DE-AC04-76DP00789.
$ Department of Mathematics, University of Rhode Island, Kingston, Rhode Island 02881. The work of
this author was supported in part by the U.S. Air Force Office of Scientific Research under contract
F49620-79-C-0129.
141

142

M. J. NORRIS AND R. D. DRIVER

Downloaded 11/25/14 to 169.233.7.23. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

on U (q $, where

hi" U--> R is continuously differentiable with

(4)

ohm(t, fi(t,j)#O onUOS,

Ot

i=1

and each gi R R is continuous and of bounded variation on bounded subintervals.

Then (1) and (2) with any point (to, Xo)S S have at most one solution on any interval J.
Remarks. The theorem of course does not guarantee the existence of a solution on
a nontrivial interval J. Existence would follow, for example, if S were open.
To treat Example 1, define hi(t, )= s: and hz(t, )= t+. For Example 2, let
h(t, s) + :5/3 and g(:)= sa/3.
Proof of the theorem. Suppose there were two different solutions, x and y, on some
interval J [to, b) where b > to. (The case J (b, to] is handled similarly.) Let
a

=-- inf {t (to, b)x(t) y (t)}.

Then x(a)= y(a).

For the point (a, x(a)) S let U, K, m, hi and gi be as described in the hypotheses of
the theorem. Without loss of generality, assume that for each ] the expression in (4) is
positive at (a, x(a)), Then, reducing U if necessary, the continuity of the derivatives of
hi assures that there exist positive constants p and M such that, for ] 1,..., m,

l:O h

ohi(t,
"+
fi(t, j) >- P
E
Ot
i=1

(5)

on U f3 S

and

[hi(t )-hi(t w)l<=Mll#-w[I

(6)

on U.

Choose a bounded interval [at,/3i] which contains hi(UfqS), reducing U if

necessary. Then gi is the difference of two continuous nondecreasing functions on
[ai,/3i], and each of the latter can be extended to a continuous nondecreasing function
on R by defining it to be constant on (-c, ai] and constant on [fli, ). Without loss of
generality, we shall assume that each gi is itself nondecreasing on R and that

gi(hi(a, x(a))) O.

(7)
Define

z(t)

Ia

IIx(s)-y(s)ll ds fora<-t<b.

Then z(a)=0, z(a)=0, z and z are continuous, z(t)>-O and

Ilx(t)-y(t)ll<-z(t) on

[a,b).
Choose c (a, b] sufficiently small so that (s, x(s)) and (s, y(s)) remain in U for
a _<-s < c. Then, from (6),

hi(s, x(s))-Mz(s) <- hi(s, y(s)) <_- hi(s x(s)) + Mz(s),

and, from (5),
d

d-- hi(s x(s))

for a <_-s <c and/= 1,.

, m.

>-_ p

UNIQUENESS THEOREM FOR ORDINARY DIFFERENTIAL EQUATIONS

Downloaded 11/25/14 to 169.233.7.23. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Thus for a =< < c, using (3) and the monotonicity of each

z(t)<-K

g/.

143

gives

IIx(s)-y(s)]l+ 2 Igi(hi(s,x(s)))-gi(hi(s, y(s)))l ds

/=1

<=K(t-a)z(t)+Kp

/.=l

fa

[gi(h/.(sx(s))+Mz(s))
d

-g/.(h/.(s, x(s))-mz(s))]-ssh/(s, x(s)) ds

K(t-a)z(t)+-p

K
p

/=l

g/(u) du
/=1

h(t,x(t))-Mz(t)

[g/(h/(s, x(s))+ Mz(s))+ g/(h/(s, x(s))-Mz(s))]Mz(s) ds.

Choose 61 > 0 such that for each/"

[g/(u)l <

when lu

6m---

h/(a, x(a))l < 61.

Then choose 8 e (0, 1/6K) such that a + 6 <= c and, for each/,

Now for
proof.

[hi(t, x(t))-h/(a,x(a))l+Mz(t)<3a
a < < a + 6 one finds z(t) <= 5z(t)/6.

when a -<t <a +6.

This contradiction completes the

The motivation for this paper was the following problem from classical electrodynamics.
Example 3. Consider n electrically charged point particles moving along the x-axis
at distinct positions, Xl(t),x2(t)," ,Xn(t). Assume that the motion of particle ]
depends only on the electromagnetic fields produced by the other n 1 particles, with
these fields traveling to particle ] at the speed of light, c.
The required fields are calculated in terms of the trajectories of the other particles
from the retarded Li6nard-Wiechert potentials; they are substituted into the Lorentz
force law for particle/. Introducing vi x/c for the velocity of particle as a multiple of
c, one obtains a system of delay differential equations with state-dependent delays:

(8)

v__v)3/2=E_Ki/

(1

where each Ki/ is a constant,

(9)

oi/ + V
oi

ri/

IA

[x/(0)- xi(0)], and where ri/ > 0 satisfies

v/-vi(t-ri/) for

tri/ ---sgn

ri/

cri/

vi(t- ri/)

].

In these equations, v/ and ri/ without an argument stand for v/(t) and ri/(t).
In order to solve the system of n 2 equations represented by (8) and (9) when -> 0,
one should know not only

(10)

v/(O) and ri/(O)

for all f and all

s f,

1,..., n.
but also the values of vi(t) for _-< 0,
Now, consideration of the problem in three-dimensional motion has led to the
conclusion that accelerations should not be assumed continuous, but only integrable
[2]. Thus it seems reasonable even in the case of one-dimensional motion to assume

144

M. J, NORRIS AND R. D. DRIVER

Downloaded 11/25/14 to 169.233.7.23. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

that the given past history of

(11)

vi,

say

vi(t) gi(t) for t_0,

1...., n,

is merely absolutely continuous, not, in general, locally Lipschitzian.

Substituting (11) into the right-hand sides of (8) and (9), one gets a system of
ordinary differential equations which satisfies the uniqueness criterion of the present
paper. Thus, a unique solution exists at least as long as each t-rii(t)<=O and each
Ivy(t)[ < 1. (Further extension of the solution would use a "method-of-steps" argument
which is not relevant to this paper.)
The above uniqueness problem was solved earlier for the case of two particles in
one-dimensional motion 1]. But the method used did not seem to extend to the n-body

problem.
REFERENCES

[1] R. D. DRIVER

AND M. J. NORRIS, Note on uniqueness ]or a one-dimensional two-body problem of

classical electrodynamics, Ann. Physics, 42 (1967), pp. 347-351.
[2] R. D. DRIVER, Topologies for equations of neutral type and classical electrodynamics, in Differencialnye
Uravnenija Otklonjajuimsja Argumentom, Naukova Dumka Kiev, 1977, pp. 113-127 (in
Russian); transl, as Tech. Rept. 60, Dept. of Mathematics, University of Rhode Island, Kingston,
RI, 1975.