JOURN.

XL OF DIFFERENTIAL EQUATIONS 16, 257-269 (1974)

Boundary Value Problems for Nonlinear Second-Order

Vector Differential Equations

Imtitut Matlz~matique, Univenite’ Catholique de Louvaiz,

E-1348 Louoain-la-Neuve, Belgium

Received August 16, 1972

1. INTRODUCTION

A number of papers have been devoted to the study of different types of

boundary value problems for a vector second-order differential equation

xv = f(t, x, x’), (1-l)

where f : [a, b] x Rn x Rn ---f R" and X’ = ax,/&. For boundary conditions

of the form
x(u) = x, , x(b) = Xb ) (1.2)

with X, , xb E Rn (Picard BVP), references can be found in Hartman’s book [l];

see also recent papers by K.A. Heimes [Z], and J.H. George and W.C.
Sutton [3]. For periodic boundary conditions

x(u) = x(b), x’(u) = x’(b) 0.3)

(Poincare’ BVP) references can be found in the papers of H.W. Knobloch
[4] and K. Schmitt [5].
The purpose of this work is to show how the author’s coincidence degree
[6], an extension of Leray-Schauder theory, can be used to give a simple and
unified proof of somewhat more general versions of the results of P. Hartman
[I, 71, H. W. Knobloch [4], and K. Schmitt [5j. Applications to special
vector second-order differential equations are given and a detailed study of
the scalar case wiI1 be published elsewhere.
Let us also note that a recent paper of A. Lasota and J. A. Yorke [IO] uses
Leray-Schauder theory to generalize an existence theorem of P. Hartman for
the Picard BVP distinct of the one extended here.

* Work realized in part when the author was visiting Professor in the “Istituto
Matematico U. Dini” of the University of Firenze (Italy), under sponsorship of the
C.N.R. which is here acknowledged.
257
Copyright Q 1974 by Academic Press, Inc.
All rights of reproduction in any form reserved.
258 J. MAWHIN

Moreover, since the time where this paper was submitted for publication,
a number of related results have appeared in the literature. First, Corollary
6.2 of the present paper, in the periodic case, has been proved independantly
by Bebernes and Schmitt [I 1, 121, together with other results, via one theorem
of Krasnosel’skii [13]. A direct proof of this corollary, by an argument
similar to the one used here, can also be found in the joint book [14] of Rouche
and the author (Chapter XI, Section 5). To end with the periodic case, let us
also note that Bebernes [15] has recently shown how to use Leray-Schauder
continuation theorem to prove results in the line of this paper and Knobloch’s
one quoted above.
Concerning other boundary value problems for (l.l), the author [16] has
used coincidence degree theory to prove that the conclusions of the Theorem
6.1 of this paper still hold for the boundary conditions

x(a) = x, , x’(6) = 0, (1.4)

and
x’(a) = x’(b) = 0, (1.5)

with the same assumptions than in the periodic case for (1.5) and with (6.1)
replaced by
T/‘&J < 0

for (1.4). Moreover, the same technique is applied in [16] to mixed boundary
conditions of the form

AX(U) - A’x’(a) = Bx(b) + B’x(6) = 0, (1.6)

where 8, A’, B, B' are diagonal (n x n)-matrices with non-negative elements

such that A + A’ and B + B' are positive definite. It is proved that Corollary
6.2 of this paper still holds for the boundary conditions (1.6), which implies,
as a special case, some results of Corduneanu [17].

2. A BASIC EXISTENCE THEOREM FOR NONLINEAR OPERATOR EQUATIONS

X and 2 being locally convex separated topological vector spaces, let

L : Dom L C X+ 2 be a (not necessarily continuous) Fredholm mapping
of index zero and N : Dom N C X + 2 a (not necessarily linear) continuous
mapping and let us choose an orientation upon Ker L and Coker L. Then, if
Q C X is a finitely bounded open set with closure cl Q and boundary bdry Q
such that:
 BOUNDARY VALUE PROBLEMS 2.59

(i) clJ2CDomiV and cIQnDomL # O;

(ii) either N is completely continuous on cl Q and L has a continuous
right inverse on Im L or N(cl Q) is bounded and L has a compact right inverse
on Im L;

(iii) 0 6 (L - N)(bdry D n DomL);

it is possible to define for the couple (L, N) an integer, the A?+-coincidence

degree in Q d[(L, N), J2] such that, if X = Z and L = I, the identity,

d[(L N), Ql = &s[~ - N, % 01

(the Leray-Schauder degree at zero of I - N in Sz) [6]. This coicidence degree

conserves most of the basic properties of Leray-Schauder degree and furnishes
a natural proof of the following existence theorem (see [6j):

PROPOSITION 2.1. If assumptions (i) and (ii) above are satisjied and, y being
aJixed element of Im L, if the following conditions are verified:

(a) Lx f ANx + y for every x E bdry Q n Dom L and every X E]O,I[;

(b) L-ly E D or QA?(c + Ky) # 0 for every c E bdry( -Ky + sz> n Ker L
depending on whether Ker L = (0) OY Ker L # {O], with Q any continuous
projector on 2 such that Z = Im Q @ Im L and K a right inverse of L mapping
Im L into a topological complement of Ker L in X,
(c) if Ker L + (O}, the Brouwer degree

ddn,, (-KY + Q) n KerL, 01

is not zero, where n, : (-KY + cl !2) n Ker L -+ Ker L is the mapping

defined by
no(c) = --JQN(c + Ky)

with J : Im Q -+ Ker L any isomorphism, then, for every h E [0, l], the equation

Lx =XNx+y (2.1)

has at least one solution in cl Sz and hence the same is true for the equation

Lx=Nx+y. Q.2)
Note that this result is an extension to the case where KerL + {O} of a
well-known existence theorem for the case where L-l exists.
260 J. MAWHIN

3. BASIC ASSUMPTIONS AND REDUCTION TO EQUATION (2.2)

Let us consider the second-order vector differential equation

xv = f(t, x, x’), (3.1)

where f : [a, b] x R” x Rn --z R” is continuous and a < b. If x, , x, are
fixed elements of Rn, a solution of the Picard boundary value problem (PdBVP,
for short) for equation (3.1) will be a mapping Y : [a, b] + Rn of class C2
which satisfies (3.1) and the boundary conditions

x(u) = x, ) x(b) = X~ . (3.2)

A solution of the PoincarC boundary value problem or periodic boundary

value problem (PBVP, for short) for equation (3.1) will be a mapping x:
[a, b] -+ R” of class Cz which satisfies (3.1) and the boundary conditions

x(a) = x(b), x’(u) = x’(b). (3.3)

Now, j . 1 being the Euclidean norm in R”, let B = Cl([a, b], Rn) be the
Banach space of mappings x : [a, b] + R” of class Cl with the norm

and let & be the subspace of B defined by

B == (x E B : x(a) = x(b), x’(a) := x’(b)}

with the induced norm. It is very easy to verify the following

PROPOSITION 3.1. The PdBVP for equation (3.1) can be written in the form
(2.2) ;f we take

X=DomN=B,Z=B x R” x’R”,DomL =(xEB:xisofclassC2),

L : x H [x”, x(a), x(b)], N : x I-+ [f(-, x(.), x’(.)), 0, 01, y = (0, x, , xb).

Moreover, KerL = (O), ImL = 2, L-l is compact and N is continuous and

takes bounded sets into bounded sets.

PROPOSITION 3.2. The PBVP fog equation (3.1) can be zvritten in the
form (2.2) if we take

X=DomN=~,Z=B,DomL==(x~B:xisofclassC~},
L : x m++
x”, N : x wf(-, x(.), x’(e)), y = 0.
 BOUNDARY VALUE PROBLEMS 261

Ker L == (x E & : x is a constant mapping]

= [x&x(t) =(h-a)-‘~‘BS(s)~,vtEju,b]~,
LX

ImL = (a E B : /’ m(s) & = 01

I a
and heme a projector ,O in B such that B -= Im Q @ Im L is given by

Q: N t-+ (b - a)-’ r*” x(s) ds,

-a
which yields ImQ == Ker L. Lastly, L has a compact right inverse and N is
continuous and takes bounded sets i?tto bounded sets.

Note that the compactness conclusions in the two propositions above are
obtained as usual by an easy application of Arzela-Ascoli theorem.

4. THE CONCEPT OF BOUNDING FUNCTION RELATIVE TO EQUATION (3.1)

Let us introduce the following

DEFINITION 4.1. A bounding function relative to equation (3.1) is a function

V : Rn + R of class C” which satisfies the following conditions:

(a) the set

@ = {x E R’&: V(x) < O>
is bounded;
(b) if zl : Rn ---f fit and W : Rn --t B(Rn, R”) denote, respectively,
the gradient vector and the Hessian matrix functions of V, then ( , ) being
the scalar product in R”,

< qx)Y, 30 + W(4, f (4 x, A> > 0 (4.1)

for every (t, ~,y, ,\) E [a, b] x R* x Rn ~10, 1[ such that

V(x) = 0 and <44, Y> = 0.

The name “bounding function” follows from the fact that the existence of
such a function implies the existence of an “a priori bozcnd” for solutions of
the PdBVP and the PBVP for every equation of the family
x” = Af(t, N, x’) (4.2)
with h E IO, I [ as follows from
262 J. MAWHIN

PROPOSITION 4.1. Let us suppose that there exists a bounding function relative
to equation (3.1). Then, for every h E IO, I[, each possible solution x of the
PBVP for (4.2) is such that

T+(t)] < 0 V t E [a, b], (4.3)

or
V[x(7)] > 0 for some 7 E [a, 61, (4.4)

ad if%, , x, E RR satisfJ1 the relations

w%) -=c 0, %%> < 0, (4.5)

the same is true for the corresponding PdBVP.

Proof. Let h ~10, I[ and, x being a possible solution of the PdBVP or

PBVP for equation (4.2), let us define the function z, : [a, b] + R by

w(t)= q401, t E [a, b].

Then 21is of class Cs in [a, e] and

w’(t) = ($x(t)], x’(t)} (4.6)

w”(t) = (W[x(t)] x’(t), x’(t)>+ W[x(t)l, fP, x(t),W)l). (4.7)
If w(t) has no negative absolute maximum in [a, b], there will exist a
7 E [a, b] such that V(T) > 0 and the Lemma is proved. If w(t) has a negative
absolute maximum which is attained at s E ]a, b[, then

4s) < 0, o’(s) = 0, v”(s) d 0,

and it follows immediately from (4.6), (4.7), and the definition of a bounding
function that v(s) < 0, and hence

Y[x(t)] = w(t) < w(s) < 0 ‘v’t E [a, b].

If the negative absolute maximum of z’ is attained at a or b, then it follows

immediately from (4.5) that V[x(t)] < 0 V t E [a, b] in the PdBVP case. For
the PBVP, we then have necessarily

w(t) < w(a) = w(6) < 0 V t E [a, 61,

which immediately implies that

0 < (u[x(b)], x’(b)) = (u[x(a)], x’(a)) < 0. (4.8)

But, then, v(a) = w(b) < 0 because, if v(a) = v(b) = 0, it follows from (4.6),
 BOUNDARY VALUE PROBLEMS 263

(4.7), (4.Q and the fact that V is a bounding function that v(t) > ~(a) in
some neighborhood of a and v(t) > u(b) in some neighborhood of b, a
contradiction. Thus Proposition 4.1 is completely proved.
h particular bounding function has been used by P. Hartman [7] in his
existence theorem for the PdBVP relative to (3.1) in which he introduces the
assumption:
There exists R > 0 such that

for every (t, X, y) E [u, b] x RTL x RTL such that j x ! = R and (x, yj = 0.
In this case, the function B : x w 1x !2 - Rz is a bounding function relative
to (3.1) because (4.9) clearly implies that

for every (t, s, y, A) ~]a, b[ x Rn x Rn x }O, l[ such that [ N / = R and

(x, y) = 0.
Let us also note that, in their existence theorems, H.W. Knobloch [4]
and J.H. George and W.G. Sutton [3] h ave used auxiliary functions allowed
to depend also on t or on t and x’ but, on the other side, submitted to more
restrictive supplementary conditions than our bounding functions.

5. THE CONCEPT OF NAGUMO EQUATION WITH RESPECT TO A BOUNDING

FUNCTION AND A BOUNDARY VALUE PROBLEM

Let us introduce the following

DEFINITION 5.1. If I’ is a bounding function relative to (3.1), this equa-

tion will be called a Nagurno equation with respect to V and a given boundmy
value problenl if, for each A E IO, l[, every possible solution x of this boundary
value problem for equation (4.2) which satisfies the relation

q@)]< 0, v t f [a,q,
is such that
1x’(t)] < s, v t E [a, b]

with S > 0 independant of A.

The name Nagumo equation follows from the fact that in the scalar case
and for F’(x) = 1x Ia - @, a sufficient condition for (3.1) to be a Nagumo
equation was first introduced by M. Nagumo [S]. It has been extended to the
vector case by P. Hartman [7] in the following way

505,W2-5
264 J. MAWHIN

Nagumo-Hartman condition. If the following conditions hold:

(i) there exists a (Nagumo) function v : R, ---f RR, such that

w sds
s s=co
and
Im “Y Y>l G dl Y I>
in
E(R) = W, x,Y>: t E[a,bl, 1x1 <R, YEW
with R > 0;
(ii) whenn > 1,

1f(t, x, $1 d 24&f@, X, y)) + 1y I”] + K

in E(R), where (II 3 0 and K > 0 are constants, then every solution E of equation
(3.1) which satisjies
I @)I d R, vtE[a,4,
is such that
j x’(t)1 < s, V t E[a,4,
where S depends only on v(s), 01,R, K and b - a.

For a proof, see for example [I], pp. 428-431.

An easy consequence is the following

PROPOSITION 5.1. If equation (3.1) satisjes the Nagumo-Hartman condition,

then, for every bounding function V relative to (3.1) such that

the closed ball of center xero and radius R, (3.1) is a Nagumo equation with respect
to V and the PdBVP or the PBVP.

Proof. The proof follows at once from the fact that if the right-hand side
of (3.1) verifies a Nagumo-Hartman condition, the same is true, with the same
functions and constants, for the right-hand side of equation (4.2) for every
A E]O, l[.
But (3.1) can be a Nagumo equation without verifying the Nagumo-
Hartman condition as follows from

PROPOSITION 5.2. If g : Rn 3 R is class Cl and la : [a, b] x R* + Rn is

continuous, then the equation
X” = gradg(.*‘) + h(t, x) (5-l)
is a Nagumo equation with respect to any bounding function V and the PBVP.
 BOUNDARY VALUE PROBLEMS 265

Boof. If X ~10, I[ an d 1‘f x is any possible solution of the PBVP for equation

X” = X[gradg(x’j + Iz(t, xj],

then we have

(b - a)-1 s” <x”(t), x”(t)) dt = (b - a)-’ Jab (gradg[x’(t)], x”(t)> dt

a

+ (b - a)-’ j-” <h[t> x(t)], x”(t)> dt,

n

and hence, using the periodicity of x’ and Schwarz inequality,

[(b - a)-‘s,b [ x”(t)/” dt]l” < H

with

and R > 0 such that the set CDcorresponding to v is contained in the closed
ball of center zero and radius R. From (5.2) and a well known inequality we
obtain then

1x’(t)\ < (b - a)H, v t E [a, b]

and Proposition 5.2 is proved.

6.-LTHE BASIC EXISTENCE THEOREM FOR PdBVP AND PBVP

AND APPLICATIONS

We can now use the results and concepts of the preceding sections to prove
the following basic

THEOREM 6.1. If there exists a boundiy function 1;’ relative to (3.1) for
which (3.1) is a Nagumo equatiou for the PdBVP (resp. PBVP) and if

V{(b - a)-l[(t - a) xb + (b - t) x,]} < 0 ‘d t E [a, b] (6.1)

(resp. if
dJgrad v, 0, 0] =# O} (6.2)
then, for every h E [O, 11, the PdBVP (resp. PBVP) fur equatiofz (4.2) has at least
one solution x such that

V[x(t)] < 0 V t E [a, b-j,

and the same is true for equation (3.1).

266 J. MAWHIN

Proof. Let us introduce the open bounded set

J-2 = (x E x : r+(t)] < 0, Vt E [a, 61, s;pb, [ x’(t)] < S>

where X = B or P depending on whether we consider the PdBVP or the

PBVP, and S is given by definition 5.1. Using Propositions 3.1, 3.2, 4.1, and
Definition 5.1, we see that assumption (a) of Proposition 2.1 is satisfied both
for the PdBVP and the PBVP for the set Q defined above. Now, (6.1) implies
that condition (b) of Proposition 2.1 is satisfied (we are in the case where
Ker L = (0)) and hence Theorem 6.1 is proved for the PdBVP. In the periodic
case, we have

n,,(c) = QN(c) = (6 - a)-’ labf(t, c, 0) dt, c E Rn

(we have used the natural isomorphism between Ker L and Rn) and it follows
easily from (4.1) with y = 0 that

(grad W, (6 - 4-l jabf(t. c, 0,) dt) > 0, (6.3)

for every c E R” such that V(c) = 0 and hence for every

cEbdry@ = KerLn bdry9.

Thus, condition (b) of Proposition 2.1 is satisfied in the PBVP case and
moreover, using (6.3), the PoincarC-Bohl theorem of Brouwer degree theory
and condition (6.2) we find

d&o ,Q n KerL, 0] = d,[grad V, @, 0] # 0.

Thus, condition (c) of Proposition 2.1 is verified and Theorem 6.1 is proved
for the PBVP.

COROLLARY 6.1. Theorem 6.1 still holds if the strict iltequality in conditions
(4.1) and (6.1) is replaced by a nonstrict one.
Proof. The proof follows by exactly the same approximation and com-
pactness argument as used in [l], p. 433, last paragraph.

COROLLARY 6.2. If thepe exists R > 0 such that

IYl"+<%f(t,x,Y)~Bo (6.4)
for every (t, x, y) E [a, b] x Rn x Rn verify&g [ x ( = R and <x, y} = 0,
and if (3.1) is a Nagumo equation with respect to V(x) = ( x jB - R2 and the
 BOUNDARY VALUE PROBLEMS 267

PdBVP with 1X, (, 1.z’~j < R (resp. the PBVP), then the PdBVP (resp. PBVPj
for (3.1) has at least one solution 31:such that

I x(t)1B R v t E[a,b].
Proof. It suffices to take the bounding function V(X) = / x js - R’ and
to use Corollary 6.1.
When the condition that (3.1) to be a Nagumo equation is replaced by
the Nagumo-Hartman condition, Corollary 4.2 was first proved in the
PdBVP case, in a different way, by P. Hartman [7] and, in the PBVP case,
under different types of more restrictive conditions, by H. W. Knobloch [4]
and K. Schmitt [5].

COROLLARY 6.3. Let g and h be as in Proposition 5.2 and let k : R” + R”

be a globally Lipschitzian mappi?tg. If grad g(y) (res?. k( y)) has either the same
or the opposite direction of y and if there exists R > 0 such that

(x, h(t, x)i 3 0 (6.5)

for every (t, x) E [a, b] x R* such that / x / = R, then the PBVP for equation

xn = gradg(x’) + h(t, X) (6.6)

(resp. X” = k(x’) + h(t, x)) (6.7)

ha-s at least one solution x such that / x(t)1 < R for every I E [a, b].

Proof. For equations (6.6) and (6.7), condition (6.4) of Corollary 6.2
follows from (6.5) because

(grad g( y), xj = (k(y), x) = 0

if (2, y) = 0. The global Lipschitz condition imposed on k implies that (6.7)

satisfies a Nagumo-Hartman condition (see [l], p. 433); on the other hand,
it follows from Proposition 5.2 that (6.6) is a Nagumo equation with respect
to the PBVP and the bounding function If(x) = 1x I2 - R”. The result then
follows from Corollary 6.2.
In the case of equation (6.7) with h(t, X) = Ax, A a positive definite
matrix, CorolIary 6.3 has been proved in a different way by K. Schmitt [5]
under the supplementary condition

where q is the Lipschitz constant of the mapping k, and TVthe least eigenvalue
of matrix A.
268 J. MAWHIN

COROLLARY 6.4. For each i = 1,. . ., n, let ai(t, x) and Pi(t, x) be continuous
fumtims on [a, b] x Rn into R having the properties that there exist constants
L$ > 0 and Ai > 0 such that ai(t, x) 3 Si and ! ,&(t, x)1 < Ai (i = l,..., n)
for t?uery (t, X) E [a, b] x R”. Then the PBVPfor the system

x; = xic& x) + p (t, m) (i = l,..., n) (6.8)

has at least one solution.

Proof. The system (6.8) clearly satisfies a Nagumo-Hartman condition.

On the other hand, if

S = min(S, ,..., S,),

then for every R > A/S, the function

t’(x) = x2 - R‘J

is a bounding function for (6.8) because

gl [x:Q, x) + x&3&, x)] 3 6 1s lB - A 1x I = 6 1x j (I x 1 - A/S).

The result then follows from Corollary 6.2.

This corollary has been proved in a different way by K. Schmitt [5]. For a
more general existence theorem concerning the PBVP for a more general
system of type (6.8), p roved directly from the argument used in Proposition
2.1, see [9].

REFERENCES

1. P. HARTWIN, “Ordinary Differential Equations,” Wiley, New York, 1964.

2. K. A. HEIMES, Boundary value problems for ordinary nonlinear second-order
systems, J. Dzz. Eq. 2 (1966), 449-463.
3. J. H. GEONX AND W. G. SUTTON, Application of Liapunov theory to boundary
value problems, PIWC. Amer. Muth. Sot. 25 (1970), 666-671.
4. H. W. KNOBLOCH, On the existence of periodic solutions for second order vector
differential equations, I. 0;s. Eq. 9 (1971), 67-8.5.
5. I(. Scmwrr, Periodic solutions of systems of second-order differential equations,
J. Dif. Eq. II (1972), 180-192.
6. J. MAWHIN, Equivalence theorems for nonlinear operator equations and coin-
cidence degree theory for some mappings in locally convex topological rector
spaces, J. D$?. Eq. 12 (1972), 610-636.
7. P. HARTMAN, On boundary value problems for systems of ordinary, nonlinear,
second-order diflerential equations, Trans. i2*~zer. Muth. Sot. 96 (1960), 493-509.
 ROUNDARY VALUE PROBLEMS 269

8. PvI. NAGIJMO, Ueber die DiRerentialgleichung y” = f(x, y, y’), PTOC. P!zys.-LMClth.

sot. Japan, Ser. 3 19 (1937), 861-866.
9. J. MA\~HIN, Existence of periodic solutions for higher-order differential systems
that are not of class D, J. Dig. Eq. 8 (1970), 523-530.
10. A. LASOTA AND J. A. YORKE, Existence of solutions of two-point boundary value
problems for nonlinear systems, J. DifJ. Eq. 11 (1972), 509-518.
il. J. W. BEBERNES AND K. SCHMITT, Periodic boundary value problems for systems
of second order differential equations, J. Di#. Eq. 13 (1973), 32-47.
12. J. W. BEBERNES AND K. SCHMITT, An existence theorem for periodic boundary
value problems for systems of second-order differential equations, r2rciz. Math.
(Brno), to appear.
13. M. A. KRASNOSEL’SKII, “Translation aIong Trajectories of Differential Equations,”
Transl. of R~u.wian Mmograplzs Ser., Vol. 19, Amer. Math. Sot., Providence,
Rhode Island, 1968.
14. N. ROUCHE AND J. M.XWHIN, “Equations diffkentielles ordinaires. Tome 2:
StabilitC et Solutions Ptriodiques,” Masson, Paris, 1973.
15. J. W. BEBERNES, -4 simple alternative problem for finding periodic solutions of
second-order ordinary differential systems, PYOC.Amel-. Mnth. Sot., to appear.
16. J. MAWHIN, DegrC de coyncidence et probl&mes aux limites pour des equations
diffkrentielles ordinaires et fonctionnelles, S&zin. Math. Appl. M&z. Univ. Louvni~r,
Rapport no 64, Vander, Louvain, 1973.
17. C. CORDUNEANU, Sopra i problemi ai limiti per alcuni sistemi di equazioni
dieerenziali non lineari, Rexd. Act. Sci. Fiz. Mut. ,vapoli (4), 25 (1958), 98-106.