You are on page 1of 14

JOURNAL

OF

MATHEMATICAL

ANALYSIS

AND

APPLICATIONS

2, 145-158

(1961)

Global

Behavior

of Solutions of Ordinary Equations


FRED BRAUER of British Levinson

Differential

Department

of Mathematics, Submitted

University by Norman

Columbia,

Vancouver

SECTION

Lyapunovs second method is a useful approach to the study of stability of solutions of ordinary differential equations. Roughly, the method involves the use of a real nonnegative function V(t, x), where t is the independent variable and x is the dependent variable (possibly a vector). If the derivative of V along any solution curve x(t) of the differential equation is negative, then some conclusions may be drawn about the size of solutions for large t. Our approach, following Conti [2], is to regard V(t, x(t)) as a measure of the size of the solution x(t), and to study it as a function of t. We bound the derivative of I above, not necessarily by zero, but by a suitable function of t and V. Then we obtain information about the size of I/ in terms of the size of solutions of a first order differential equation. Many properties of solutions of this first order equation are reflected by properties of solutions of the original equation. By this approach we are able to deduce Lyapunovs classical results on stability [33, and also to generalize various results of Winter [s-12] on global existence, limits of solutions, the possibility of assigning initial conditions at infinity, and the smoothness of the flow defined by a differential equation. Our results also cover a type of limiting behaviour mentioned only briefly in Wintners work, where the function V tends to a limit but the solution itself does not. Such behaviour may be exhibited, for example by a periodic solution. The solution may be regarded as spiralling about a surface on which V is constant in this case. The author is indebted to Professor Shlomo Sternberg for many valuable suggestions, including Theorem 5 in its entirety, as well as several other ideas. 145

146

BKAUER

SECTION

II

We are interested in the behaviour of the solutions x(t) of

X = ,(t, x),
satisfying the initial condition

(1)

x(O) = x0,

(4

where x, x0, and f are real n-dimensional vectors, and 0 < t < co. The norm of a vector x will be denoted by 1x1and will mean the sum of the absolute values of the components of X. In applications, the Euclidean norm, denoted by j/xl/ will a1 be used. Derivatives with respect to t so will be denoted by . It will always be assumed that f is continuous for 0 < t < 00, 1x1< CQ. Let V(t, x) be a function defined for 0 < t < 03 and vectors X, with nonnegative real values, which is continuous in (t, x), and has one-sided partial derivatives with respect to t and the components of X. It will be assumed that for each fixed t, the inverse image under V in R of a lim V(t, X) = 00. It compact set is compact, or equivalently, that bl--+oo will also be assumed that V is positive definite, that is, given any E > 0 there exists Q > 0 depending only on E, such that 1x1< E whenever V(t, x) < S. This assumption is actually not needed until Theorem 4 except for Corollary 2 of Theorem 2, but it will be made throughout to shorten the statements of the results. V/t will denote a partial derivative of I with respect to t, V, some gradient vector of V with respect to X, and , the usual scalar product of vectors. Any condition which involves V, or V, will be understood to be required for all one-sided derivatives. \fre define

v*(t, x) = V&, x) + V&P x) .f(h 4.


All our results will depend on bounds on V*(t, x). Except in Theorem 3, only a one-sided bound will be required. These bounds will be of the form

or V(k x) 2 qt, V(k q), (4) for 0 < t < 00, 1x1< 00, where ~(t, Y) and 13(t,r) are continuous functions on 0 < t < 00, Y > 0, and ~((t, 0) = 0(t, 0) = 0. An important tool in our results is the following lemma, due to Conti [2].

ORDINARY DIFFERENTIAL

EQUATIONS

147

LEMMA 1: Let x(t) be a solution of (l), and let m(t) = V(t, x(t)). Y* = m(0) = v(0, x(O)), and let r(t) be the maximum s&&on of

Let

r = o.qt, Y) satisfying the initial conditiovt Y(0) = 0.

(5)

(6)

Theuz, if (3) is satisfied, the solution x(t) can 6~ continued to the right as a function of t as far as r(t) exists, atid m(t) 6 r(t) fol all such 1. Conti assumes that V has continuous partial derivatives with respect to t and the components of x. Because of (3) and the fact that x(t) satisfies (l), this implies that m(t) is differentiable, and m(t) = V*(t, x(t)) < co(t, m(t)). We have assumed only the existence of one-sided derivatives, principally to permit us to consider applications with V(t, x) = 1x1. However, the argument used for the proof of Theorem 1 of [l] shows that even with these weaker hypotheses, we still have

E?:;p [m(t + h) - m(t)l/h < u(t,m(t)),


and this suffices to continue the proof of the lemma as in [2]. It is easy to prove an analogous result if the condition (3) is replaced by (4). In this case, lip;f [m(t + 32)- m(t)]/h > Ott, m(t)),

and an analogous argument yields this result.


LEMMA 2: Let x(t) be a soltitiort of (1) and Eetm(t) = V(t, x(t)). r,, = m(0) and let p(t) be the minimum solution of

Let

Y = e(t,r),

(7)

which satisfies (6); Then, if (4) is satisfied, m(t) >, p(t) as far to the right as both p(t) and x(t) exist. The main result on global existence of solutions depends on Lemma 1 and the following result of Wintner 171,
THEOREM 1: lf x(t) is a solution of (1) which is defined for some tinterval bowzded oe the right by t,,, then either /x(t)/ + 00 as t - to, or x(t) cas be continued beyond to. Combination of Lemma 1 and Theorem 1 yields this global existence theorem, due to Conti [2].

148

I:KAUER

THEOREM 2: If (3) zs satisfied and aW so.&ions of (5) can he continued for all t, then all solutions of (1) can be continued for all t.

The proof is immediate. For any solution x(t) of (1) and any finite t,, lim Ix(t)] = 00 would imply lim m(t) =T cu. This contradicts Lemma 1 t-t, t-d, and the assumption that the solutions of (5) can be continued beyond t,. Now Theorem 1 shows that x(t) can be continued beyond t,.
COROLLARE' 1: If alL solutims solutions of (1) are hounded. COROLLARY 2: If all solutions of (I) tend to zero. of (5) aye Bounded for all t, then all

of (5) tend to zero, then all solutions

PROOF: If x(t) is any solution of (1) and m(t), r(t) are as in Lemma I,

Given any E > 0, we can find 6 > 0 so that lx(t)/ < E 44 < m whenever m(t) > 6. By making t large enough, we can make m(t) < S, and hence Ix(t)1 < E. Thus x(t) tends to zero. This proof requires V to be positive definite. We can also obtain a converse result, based on Lemma 2. lirn 1x1 = 00, if (4) is satisfied, and qt, n)+ co tions of (7) tend to 00, then for any sol&ion x(t) of (l), Ix(t)1 COROLLARY 3: If PROOF:

if

all solu-

W.

Either lx(t) 1 + m for some finite t or x(t) can be continued for all t. In this case, Lemma 2 implies lim V(t, x(t)) = 00, whence t+m lim lx(t)] = bc1. t--tm
SECTION III

In this section, we study the behaviour of the solutions of (1) for large t. Our first result requires a two-sided condition on V*(t, x), but, as will be seen, it is possible to weaken this to an upper bound if this upper bound is nonpositive.
Sq5pose that the functions u), 8 are nzonotone MOWTHEOREM 3: decreasing in r for each fixed t, that (3) and (4) hold, artd that ali solutions of (5) a& (7) tend to finite limits as t - 00. Then for every solution x(t) of (I), V(t, x(t)) tends to a finite limit as t - cu. If rO = V(0, x(O)), y(t) is the maximtim sol&ion of (5) and p(t) is the minimum solution of (7) satisfying (6), then

lim t+m

p(t) <

lim t-03

V(t, x(t)) <

lim t-m

r(t).

(8)

ORDINARY DIFFERENTIAL PROOF:

EQUATIONS

149

If t>a>O,

qt, - Jf(a, j T/*($<jm,s, ds, w) +4) x(s)) qs, = ds %+))]


a a

using (3) and the fact that x(t) is a solution of (1). Since V(s, x(s)) < Y(S), 01 is monotone nondecreasing in I, and r(t) satisfies (5),

v(t, x(t)) - v(a, a(u)) < 1 a+, r(s)) ds = I(i) - r(u). a


A similar argument using (4) and (7) instead of (3) and (5) yields

Since y(t) and p(t) tend to finite limits as t -3 bo, both lo - r(a)] and ip(t) - p(a)/ can be made less than an arbitrary E > 0 by making a sufficiently large, the choice of a depending only on E and V(0, x(O)). Then - E < V(t, x(t)) - V(a, x(a)) < E for t > a, and V(& x(t)) tends to a limit as t + 00. The inequality (8) follows immediately from Lemmas 1 and 2. In this case, If o(t, Y) < 0, the condition (4) is unnecessary. V*(t, x(t)) < 0, which implies that V(t, x(t)) is decreasing. Since V(t, x(t)) is bounded below by zero, it must tend to a limit. We may also remark that the condition V*(t, x(t)) > 0 can play the role of (4). This corresponds to t3(t, I) = 0, and for this choice of 9, the solutions of (7) tend to limits. It might be natural to conjecture that under the hypotheses of Theorem 3, not only does V(t, x(t)) tend to a limit, but the vector x(t) itself tends to a limit vector. This stronger statement, however, is false unless some additional hypotheses are made. Consider the two-dimensional system

whose general solution xl(t) = a cos t + b sin t, x2(t) = - a sin t + b cos t


(10)

is periodic and thus does not tend to a limit. We use V(t, x) = x * x, so that V/t = 0, and VZ = (2x,, 2x,). The hypotheses of Theorem 3 are satisfied with o(t, 1) = 0, but no solution vector tends to a limit.

To prove a result to the effect that x(t) tends to a limit, we must make some further assumption on F(t, x). This further assumption is strong enough to permit us to dispense with the lower bound (4) for v*p, x).
THEOREM 4: Suppose thut for t > a > 0 atid atiy continuous rector x(t),

Stifipose also that (3) holds for some cc) which is monotolze nolzdecveasing in I for each fixed t, and that all solutions of (5) tend to finite limits as t + 00. Then al,? solutions of (1) tend to finite limit vectors as t ---f CO.

PROOF: If t > a, and x(t) is any solution of (l), t t

J(t, x(t) - x(a)) = v (4 if(s, x(s)) ds] < 1 J*(s, x(s)) ds t


< I a co [s, V(s, x(s))] ds <

t
w(s, Y(S)) ds = r(t) - r(a),

using (3), (II), and the fact that o is monotone in Y. Here, r(t) is the maximum solution of (5) with r(O) = r/(0, x(0)). Given any E > 0, we can find 6 > 0 so that (x(t)/ < E whenever V(C, x(t)) < S. Then we can choose a large enough, depending only on E and V(0, x(O)), that for t > ti., 0 < V(t, x(t) - x(a)) < 6. But then Ix(t) - x(a)1 < F, and thus x(t) tends to a finite limit vector as t - co. SECTION IV A problem which has been considered by Wintner [lo] is the possibility of assigning initial conditions at infinity. Possibly it would be appropriate to describe these as final conditions. To prove a resuIt in this direction, we need some sort of uniformity for all solutions of (1) which pass through a given point in x-space. We will assume that if x is any point of 67, there is a solution r(t) of (5), possibly depending on z, such that any solution y(t) of (1) p assing through z (for any value of t) obeys

ORDINARY DIFFERENTIAL

EQUATIONS

151

This condition is satisfied, for example, if V(t, X) is independent of t or nonincreasing in t for each X, and co(t, Y) > 0. In this case, r(t) may be taken as the maximum solution of (5) with r(O) = V(0, z). If y(t) is a solution of (1) with y(tr) = x for some t,, 0 < t, < 00, let r,(t) be the maximum solution of (5) with r,(t,) = V(t,, z) < V(0, z). It follows easily from ~(t, Y) 3 0 that pi <r(O) < r(C,), and vi(t) < r(t) for 0 < t < co. Lemma 1 implies V(t, y(t)) < rr(t) < y(t) for t, < t < 03. Since r(t) and l(t, y(t)) are nondecreasing functions of t and V(tl, ~(2,)) < Y(O), (12) must actually hold for 0 < t < W. The statement and proof of the following result are due to S. Sternberg.
THEOREM 5: Suppose that the hypotheses of Theorem 4 are satisfied, and that for any fioint z of R there is a solution r(t) of (5), such that alzy solution y(t) of (1) p assing through z for any value of t satisfies (12). Then the differential equation (1) with the initial condition at i#utity

lim x(t) = x, f-boo


has a solution PROOF: x(t) on 0 < t < oo for atiy vector x,.

(13)

Let x*(t) be a solution of (1) with

h&(n) = xco,

n = 1,2,. , . .

(14)

By Theorem 4, x&) tends to a finite limit vector X,,(W) as t - 00. The assumption on the uniformity of solutions of (1) makes it possible to use r(t) in the proof of Theorem 4 depending only on X, and not on n. Then the convergence of xn(t) to X,(W) as t is uniform in n. Next, P [t, f(t, x,(t))1 is bounded uniformly in n on every bounded t-interval, which implies that If(t, x%(t))\ is also bounded uniformly in n on every bounded t-interval. Since f(t, x,(t)) = x%(t), {x&>> is equicontinuous on every bounded t-interval, and contains a subsequence (x,,k(t)> which converges uniformly on every bounded &interval. Since every x,(t) tends to a limit x,( CQ),this subsequence can be chosen so that the corresponding subsequence of limits xmk( CW) also converges. Then the resulting subsequence {xnk(t)} converges uniformly on 0 6 t < 03 because the convergence of xnk(t) to xnR(oo) as t --j COis uniform in n. Now let x(t) be the limit function of such a subsequence. Then x(t) is a solution of (l), and by Theorem 4, x(t) tends to a limit x( CG) as t --t CXJ. Also,

and these two facts together with (14) imply that x( CW)must be x,. Thus we have constructed a solution x(t) of (1) satisfying (13).

115.2

We now have a mapping 7 of K onto itself defined by the differential equation (I). If x0 is any point of R, let x(t) be a solution of (1) which satisfies (2). Then if X, is defined by (13), we let TX, = x,. Also, if x, is any point of RI, we can let x(t) be a solution of (1) which satisfies (Is), and then define T-l x, = x(0). We can do the same for (5). If Y,,3 0, let r(t) be a soIution of (5) which satisfies (Ij), and let Y, = lim y(t), Then we can define the mapping S of Y > 0 onto itself by Sr,, = Y,. Also, if Y, 3 0, we let r(t) be a solution of (5) which tends to Y, as t --, m, and S-l Y, = r(0). As we have made no assumptions to assure the uniqueness of solutions of (l), the mappings T and T-l are not necessariIy well-defined. If such an assumption is made, we can obtain some information about the mapping T.

THEOREM Suppose that the ky#otheses of Theorem 5 are satisfied, 6:


that V(t, x) is small uniformly

in t if and only if 1x1 is small, thut

V&, x - Y) + u4 x - Y) * [f(k 4 - f(h 341< 4,

w> x -

Y)l

(15)

do, where the only solution o# (5) satisfying r(O) = r(0) = 0 on any bounded t-interval is the trivial solution, and that S is a homeomorphism of I 3 0 onto itself. Then T is well-defined and is a homeomorphism of R onto itseZf.

f010 <t < m, 1x1 03,jyl < <

PROOF: The hypothesis involving (15) implies that there is a unique solution of (1) on 0 < t < 00 which satisfies (2) (cf. [l]), and thus that T is well-defined and one to one. Let x(t) and y(1) be two solutions of (1) Th en if r(t) is the maximum solution and let m(t) = V [t, x(C) - y(t)]. of (5) with r(O) = m(O), it follows, as shown in [l], that m(t) < r(t) for 0 < t < 00. Both m(t) and r(t) tend to limits m(m), ~(00) respectively as t 3 00, and m(m) < y(m). The continuity of S implies that Y(W), and hence m(m), can be made arbitrarily small by making m(0) == Y(O) small enough. Since V(t, X) is small uniformly in t if and only if 1x1 is small by small, this means that IX(W) - y(m) j can be made arbitrarily making Ix(O) - y(0) 1 small enough, or, in other words, that T is continuous. The continuity of T-l follows from the continuity of S-l in the same way. Thus T is a homeomorphism.
SECTION

VI

The scope of our results can be indicated by various well-known results which axe special cases. A first example is the classical work of

ORDINARY DIFFERENTIAL

EQUATIONS

153

Lyapunov [3]. A Lyapunov function is a function I/-@,x) defined for 0 < 2 < 00, 1x1< B with the positivity and differentiability properties that we have been assuming throughout, except that as V(t, x) is only defined for uniformly bounded 1 we can dispense with the condition xl, lim V(t, x) = co. It is convenient to assume f(t, 0) = 0, so that x = 0 I++~ is a solution of (I). This solution is said to be stable if given any E > 0, any solution x(t) of (1) with sufficiently small ix(O)1 obeys lx(t) 1< F for 0 < t < co. The solution x = 0 is said to be asymptotically stable if every solution x(t) with sufficiently small lx(O)1 tends to zero as t --f a3.
THEOREM 7: If V is a Lyapunov fu+zction and V*(t, x) < 0, then the solution x = 0 of (1) 2sstable. If V tends to xero with 1x1 wiformly ix t and V*(t, x) is negative definite, then the solution x = 0 is asymptotically stable. PROOF: If V*(t, x) < 0, we can take cu(t, Y) = 0, and the solutions of (5) are constants. It is easy to see that the proof of Theorem 2 is valid even if (3) only holds in a bounded region, provided the solutions of (1) under consideration do not leave this region. By Theorem 2, V(t, x(t)) < V(0, x(0)) fort > 0 if x(t) is any solution of (1) with /x(O) j < L3. Thus, with a suitable bound on 1x(0)1, V(0, x(0)) < 6, and this implies lx(t)l < E for t 2 0, proving the stability of x = 0. The assumptions that V(t, X) tends to zero with 1x1 uniformly in t and that V*(t, x) is negative definite imply an inequality of the form V*(t, X) < - cV(t, x) for sufficiently small 1 with c some positive constant. Then (3) is xl, satisfied with w(t, .y) = - CY,and by the second corollary to Theorem 2, all solutions of (1) tend to zero if Ix(O)1 is sufficiently small. Thus x = 0 is asymptotically stable. In this connection we mention also the work of Yoshizawa [13], which includes many results on boundedness, stability, and asymptotic stability of a more precise nature than these results of Lyapunov. Some of Yoshizawas results can be obtained easily from ours, using either ~(t, Y) = 0 or ~(t, Y) = - CI (c > 0). Others appear to be too sharp to be proved by this approach. Another type of problem to which our results may be applied is the question of stability of linear systems or small perturbations of linear systems. Consider the system

(l(i) where A(t) is a real, continuous n x n matrix. Let A(t) denote the transpose of A(t) and let A(t) be the largest eigenvalue of +(A(t) + A*(t)), necessarily real since *(A(t) + AT(t)) 1s sy mmetric. The following result . is essentially due to Wazewski [7: (cf. [6], pp. 591, 602).

x = A(d)&

THE~KEM 8:
and asym#totically

Ilhe soLutio+z x y= o oi (16) is stable ii St A(t) dt < ~3,


stable

if jy A(t)dt = ~ 00.

PROOF: We take L-(2,x) =I x * x. It is cIear that I (t, X) is small if and only if lxi is small, and that V(d, X) has all the necessary continuity and differentiability properties. Also, V,(t, X) +- I,(t, X) . A (t)?~= 2x - A(t) .2: = x * (A(t) -f- A T(t )) X. This quadratic form is no greater than &I($)x * x, and thus we can take o)(t, Y) = U(t)y. The solution of (5) satisfying (6) is y(t) = y. eSin(s)dr, and the result follows from the first two corollaries of Theorem 2. It is easy to prove (cf. [6], p. 588) by means of the variation of constants formula that the stability of solutions of (16) is equivalent to the stability of solutions of the linear nonhomogeneous system x zzzA(t)x + h(t). For nonlinear systems of the form x := Ax -/- g(t, x), (17)

where A is a constant matrix, the classical result is due to Perron 151. Here it is convenient to use the Euclidean norm / 1x1 for vectors X. j THEOREM 9: I/' the eigenvalties of A all have negative real part, anLZi/, given. E > 0 there exists iT > 0, indefiendent of t, such that

then the zero solzttion

of (17) is

asymfxfotically

stable.

PROOF: We can give a direct proof only under the more restrictive assumption that the eigenvalues of +(A + AT) are all negative. However, the general case can be reduced to this. Let - A(1 > 0) be the largest eigenvalues of $(A + AT). We again use V(t, X) = xv x, and then (3) is satisfied with co(t, r) = -- 2(1 - E)Y for l]xjl < 6, or V(t, X) < d2. As in the proof of Theorem 7, we note that the proof of Theorem 2 is valid here even though (3) holds only in a bounded region. If we choose E < A, the solution of (5) satisfying (6) is the monotone decreasing function r(t) = Yge-2(n-))t. Thus, if x(t) is a solution of (1) with ]lx(O)(I < 6, or V(0, x(O))< d2, V(t, x(t)) < V/(0, x(0)) e-2z(a-E)t,or ilX(t)jl < ]Ix(O)]j B-++. This proves the desired asymptotic stability. We reduce the general case to the one in which the eigenvalues of $(A + A*) are negative by a device used by Marcus r4J. The transformation x = cy, (18)

OKDINARY DIFFERENTIAL

EQUATIONS

155

where C is a real nonsingular

constant matrix,

changes (17) to (19)

y = c-1 ACy + h(2, y),

where h(t, y) = C-l g(t, Cy) obeys the same smallness condition as g(t, x), Thus we may assume that such has already been made to bring A into triangular form. Thus, if the elements of A are aii, we may assume aii = 0 for j < i. We now make another transformation (18) using a diagonal matrix C whose diagonal elements are 1,~, v2,. . . , r- l. The result is a system (19) with a matrix D = C--l AC such that Rl Ai, *(D +
D?)ij

;=i i<j
i>j

=
I

&vj - t aii, 1 i-jaji 2rl


)

where 3Li are the eigenvalues of A, so that Rl & < 0 (i = 1,. . . , YZ). Since the eigenvalues of a matrix depend continuously on the entries, the largest eigenvalue of *(II + 0) can be made arbitrarily close to the real part of the eigenvalue of A with largest real part by choosing q sufficiently close to zero. Then @I + Dr) has negative eigenvalues, and we can apply our earlier argument to prove the asymptotic stability of the zero solution of (19), which is obviously equivalent to the asymptotic stability of the zero solution of (17). We can also use this approach to treat an equation of the form
x = A (t)x + g(t, x), W)
A(t)

in which the linear term depends on t. As before, we let largest eigenvalue of $(A (t) + A (t) ) ,
t+co E > 0 there exists 6 > 0, independent 1w Then the zero solution
of

denote the

THEOREM 10: Sz@ose that

lim l/f j,!, A(s) ds = c < 0, and that given

of

t such that llxll < 6.


stable.

x)11 < fllXllJ

(20) is asymPtotically

PROOF: We use V(t, x) = x - x, and then we see, following the methods used in Theorems 8 and 9, that (3) is satisfied with co(t, Y) = 2@(t) + E). !a The solution of (5) satisfying (6) is r(t) = y0 e2[Jf) A(s) + et] . If E is chosen less than 1~1, -+ 0 as t + 00. Then if x(t) is any solution of (20) with r(t) llmll < 4 IIx(t)jl -+o bY th e second corollary to Theorem 2, and this proves the asymptotic stability of the zero solution.

The origin of all these results is the work of Wintner i8-12,. \\:intners results can be divided into two classes, as indicated hv the following two theorems.
THEOIZEM 11: I,/ lf(h x)1 < qh IX!), (21)

and all solutions oj (5) ca?t he continued for all t, then all solutio~zs of (I)
can be continued /or all t. lf al2 solutions of (5) are hourtded, then alL solutions

of (1) are bounded.


THEOREM 12: Su@ose ilk 4 j G WMx), (22)

where A(t), $( Y) aYe continuous and nonnegative on 0 < t < 03, 0 < Y < 00 reqbectiaely, and

Then all solutions

of (1) can be contiwed


m

for all t. If, in addition,

A(t) dt < c-,

P4)

for

then every soZutio?z of (1) tends to a finite limit vector as t -+ co. Also, the differentiaL equation (1) with the initial conditiott (13) any vector x,, at infivtity has a so&ion on 0 < t < 00. If these conditions aye supplemented by

and

(26)
T of Section-r I/ is a homeomorphism. To prove these results, we use V(t, x) = 1x1. Then V(t, X) has the necessary differentiability properties. In fact, it was to allow this choice of I/ that we assumed only the existence of one-sided derivatives of V, rather than differentiability. Obviously V(t, X) is small if and only if 1x1 is small, uniformly in t. Then V,(t, x) has all components either 1 or - 1, then the ma@ing

ORDIN.4RY

DIFFERENTIAL

EQUATIONS

157

V,(t, x) = 0, and V*(t, x) is bounded above by I#, x)1 and below by - If(t, x) /. The condition (21) plays the role of (3) in Theorem 11, which thus reduces to Theorem 2. In Theorem 12, (22) plays the role of (3), so that o(t, r) = a(t)+(~). The properties of solutions of (5) needed for Theorems 4, 5, 6 can be verified for this chaise of co@, using (23), (24), r) and Wintners proofs include these verifications. The additional condiThe tions (25), (26) correspond to (15) and also assure uniqueness. condition (11) introduced in Theorem 4 is just the well-known fact that the norm of an integral is no greater than the integral of the norm, with this choice of I. As we have remarked above Theorem 5, the uniformity condition required there is satisfied if V is independent of t and (o(t, Y) is nonnegative, which is the case here. Then Theorem 12 follows from Theorems 4, 5, 6. In fact, we can give the following unwieldy generalization of Theorem I?!, in the style of Theorem 11.
THEOREM 13. S@pose f satisfies (21) for some o which is monotone nondecreasing in Y for each t. If all solutions of (5) tend to finite limits as t + co, the same is true of all solutions of (I). In addition, the differential equation (1) with the initial condition. (13) at infinity has a solution. Finally, suppo.se that

if@>4 - f(t) Y)l G 4,

Ix - YIL

for 0 < t < W, 1x1< m, IyI < 00, where the only solution of (5) satisfying Y(0) = r(0) = 0 on any bounded t-interval is the trivial solution, and that the mafiping S defined on 7 Z,S0 by the solutions of (5) is a homeomorphism. Thelz the mapping T of R onto itself defined by the so&ions of (1) is also homeomor$hism.
REFERENCES
1. BRAUER, F. AND STERNBERG, S. Local uniqueness, existence in the large, and the convergence of successive approximations, Am. J. MatA. 80, 421-430 (1958) ; Ibid. 81, 797 (1959). 2. CONTI, R. Sulla prolungabilita deile soluzioni di un sistema di equazioni differenziali ordinarie, BOX Unione Mat. Ital. 11, MO-514 (1956). 3. LIAPOUNOFF, A. M. Problbme genera1 de la stabilite du mouvement, Princeton, 1947. 4. MARCUS, M. Some results on the asymptotic behaviour of linear systems, Can. /. Math. 7, 531-538 (1955). 5. PERRON, 0. uber Stabilitat und asymptotisches Verhalten der Integrale von Differentialgleichungssystemen, Math. 2. 29, 129-160 (1929). 6. SANSONE,G., AND CONTI, R. Equazioni differenziali non lineari, Rome, 1956. 7. WAZEWSKI, T. Sur la limitation des integrales des systemes dequations differentielles lineaires ordinaires, Studia Math. 10, 48-59 (1948).

Am. J. Mafh. 68, 125-132 (1946). 8. WINTNER, A. Xsymptotic equilibria, 9. WINTNER, A. The infinities in the non-local existence problem of ordinary differential equations, ilm. J. JIuth. 68, 173-178 (1946). 10. WINTNER, A. An abelian lemma concerning asymptotic equilibria, Am. ,J. Math. 68, 451-454 (1946). 11. WINTNER, A. Asymptotic integration constants, /Im. J. Math. 68, 553-959 (1946). 12. WINTNER, A. Ordinary differential equations and Laplace transforms (appendix), Am. J, Math. 59, 265-294 (1957). 13. YOSHIZAWA, T. Liapunovs function and boundedness of solutions, Funkcial. Ekvac. 2, 95-142 (1959).

You might also like