PACIFIC JOURNAL OF MATHEMATICS

Vol. 75, No. 1, 1978

UNIQUENESS OF LINEAR BOUNDARY VALUE

PROBLEMS FOR DIFFERENTIAL SYSTEMS

H. GlNGOLD

A unified approach to "Uniqueness of Boundary Value

Problems for Linear and Nonlinear Differential Systems" is
expounded. New results are obtained and old one's become
transparent.

1. Introduction. In this paper we expound a unified ap-

proach for determining whether a linear boundary value problem of a
non-linear differential system has a unique solution. The advantages of
our method are:
(i) It provides a unified approach to linear boundary value
problems of a general type.
(ii) It is applicable to linear and non-linear differential systems.
(iii) It provides a tool for investigation of non-linear boundary
value problems for non-linear differential systems.
(iv) It makes many old results transparent.
An application of our standpoint to two-point boundary value
problems of second order differential systems and equations was demon-
strated in Gingold [16].
We will not treat in this paper the uniqueness problem of an n-th
order differential equation that includes the De la Vallee-Poussin bound-
ary value problem though our method aplies to that problem as well.
We observe from Gingold [16], that many boundary value problems
for n-th order differential equations exhibit "some sort of
singularity." Besides this, the importance of the uniqueness of a De la
Vallee-Poussin problem (e.g. see Bessmertnykh [6] for an existence
theorem, Levin [34-38] for uniqueness) justifies a special
discussion. This will be carried out in Gingold and Gustafson [17].
In §2 we introduce assumptions and notations.
In §3 we write a necessary and sufficient condition for the boundary
value problems of linear differential systems to possess a unique
solution. We also point out how the domain of uniqueness of the
boundary value problem may be extended.
In §4 we modify the results of §3 to nonlinear differential systems.
In §5 we illuminate Kim's result [28], and show how nonuniqueness
on the boundary 3D of a simply connected domain D, is related to
uniqueness of the boundary value problem in D.
In §6 we discuss sufficient conditions for the uniqueness of a special
107
108 H. GINGOLD

form of boundary conditions for "triangular" nonlinear systems. In §7

we make use of quadratic forms to obtain other sufficient conditions for
uniqueness of special boundary value problems.
In §8 we characterize "best possible constants" for uniqueness of
boundary value problems, and in §9 we estimate them.
In §10 we conclude with some historical remarks.

2. Preliminaries. Let

(2.1) y'

be a nonlinear differential system where y (t) is an n-column vector

y (0 =
y»(0

and /(ί,y) an n column vector function defined in a domain to be

introduced below.

NOTATION 2.1. Let D be either the interval [α,b], or an open

simply connected domain in C with boundary 3D which is a piecewise-
smooth closed Jordan path. We also denote by D the closure of D.

ASSUMPTION 2.1. The vector function /(ί, y) is analytic in D x Cn,

for D CC.

ASSUMPTION 2.2. The vector function f(t>y) is continuous in

n
( D U 3 D ) x C , for D C C .

ASSUMPTION 2.3. The vector function f(t,y) is a mapping from

[α, b] x Rn into Rn or from [α, b] x C" into Cn, such that;
(i) for every fixed y,f(t,y) is piecewise continuous in [α, b].
(ii) for every fixed t, f(ty y) and its Jacobian, fy (t, y) are continuous
in R" or C .

NOTATION 2.2. B.V.P. will be the abbreviation of Boundary Value

Problem.

NOTATION 2.3. A solution of (2.1) will be an analytic n-column

vector function y (t) on D, continuously differentiable in D U 3D which
satisfies (2.1) subject to assumptions 2.1, 2.2, or a piecewise smooth
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 109

n-column vector function y(ί) on [a, b] which satisfies (2.1) subject to

assumption 2.3.

NOTATION 2.4. Let y (t), y(t) be two solutions of (2.1), we denote

by JB(ί, y, y) the following matrix

(2.2) B(t,yJ)= f f,(/,(λy(O + (l-λ)^(O)rfλ

Jo

NOTATION 2.5. The nonhomogeneous linear system (2.3) will be

(2.3) y' = f(t,9) = A(t)y+g(t)

where A(t) is an n x n matrix and g(t) is an n-column vector.

NOTATION 2.6. The homogeneous linear system (2.2) will be

(2.4) y'=ϊby) = A(t)y

where A(t) is an n x n matrix.

NOTATION 2.7. We denote by E the n x n identity matrix.

NOTATION 2.8. Whenever

f'2 encountered, for f t two points
b 2

belonging to a domain £>, then; if D =[a,b] the integration path is taken

along a real segment.
If D is a domain in the complex plane,
f'2 will be assumed to be
Jtl
taken along a simple Jordan arc imbedded in D and which connects tx
and t2.
In order to bring out that the integration in the complex plane is
taken along a specific simple Jordan arc Γ, we will use the notation
JtιΓ

NOTATION 2.9. We denote by C(ί, t0) the solution of the matrix

integral equation

(2.5) C(t, to) = E+ίt B (5, y, y)C(s, to)ds

J

where ί, t0 belong to [a, b] or to D U

Similarly we denote by Q(ί, t0) the solution of the integral equation
110 H. GINGOLD

(2.6) Q(t, to) = E+ί' A (s)Q(s, to)ds.

Jto

We recall that C(t, ί0), Q(t,t0) are given by the infinite series

C(t,to) = E+ [ B(s,y,y)ds + P B(s,y,y) Γ B(si,y,y)dsi<fc

Jίo Jto Jto
(2 7)
r -- r - r-
+ B (s, y , y ) β(su y,y)\ B ( s 2 , y , y ) d s 2 d S ι d s + •••
J to J to J to

Q(t,to) = E+ f A(s)ds+ !' A(s) \S A(s1)ds1ds

J to J to J to

(2.8)
+ ί A ( s ) f S A ( s , ) ί " A ( s 2 ) d s 2 d s ι d s + •••
J to J to J to

We define the operator T", n = 0,1, as follows

T° = E, T'= [ A(s)ds,
Jto

T" = f' A(s) fS A(st), , f " " A(«„_,) ^ - , ^ - 2 , , dsids, n g 1.

J to J to Jto

We observe that

(2.9) Σ

NOTATION 2.10. Let y{t) be a solution of (2.1).

Let U G D, ί = 1, , m (or ( £ D U ΛD), be m points, not necessar-
ily distinct. Let Pn i = l, ,m, be m n X n matrices. Then the
boundary conditions will be given by the equation

([Link]) Σ WO=£

where C is a given n-column vector belonging to Rn or Cn.

Throughout the paper, it will be assumed that (ΣΓ=i Pi) is invertible.

NOTATION 2.11. We denote by rj the initial value taken by the

solution y(t) at the initial point t0.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 111

3. A uniqueness criteria. We are ready now to formulate a

necessary and sufficient condition for linear differential systems to
possess a unique solution to B.V.P. (2.1), (2.10).

PROPOSITION 3.1. Given B.V.P. (2.1), (2.10) let Assumptions 2.1,

2.2 or Assumption 2.3 be satisfied then:
(i) The B.V.P. (2.1), (2.10) has a unique solution iff

(3.1) d(ίo, *i, t2, , tm)έ 0, where

(3.2) d(t0, tu t29 , O = det ( J i> + 5 P, P A (

\ί = l i=ί Jto

(ii) Let L and R be n x n invertible matrices. Then B. V.P. (2.3),

(2.10), has a unique solution iff

(3.3) dct\ L(Σ P^R+Σ U>tΓ A(s)Q(s,to)Rds]j£O

L \i = l / ι= l Jto J

(iii) CΛoo5e in (3.3) L = (ΣΓ-iPj)"1 and R = E.

We denote by

(3.4) P, =

Then B.V.P. (2.1), (2.10) has a unique solution iff

(3.5) det [E + Σ ft I* A (s)O(s, to)dsj ϊ 0.

Proof. With the Notation 2.11, let y (t) and y (/) be two solutions to
B.V.P. (2.3), (2.10). Then by

(3.6) y(ti) = Q(th to)η = η + (J* A(s)Q(s, to)ds) ή,

(3.7) p^) = Qith to)η = η

Utilizing (3.6), (3.7) one obtains from (2.10) by subtraction

(3.8) J P,[y(0- yit)] = [ 2 Pi + S P( ί" A(s)Q(s, η - η) = 0.

i=l Li = l ί= l Jίo
112 H. GINGOLD

Formula (3.8) immediately yields the results (i) to (iii) of our

proposition. An inspection of formula (3.2) shows that if A = 0 on D
then (3.2) is satisfied. This is a first justification to the assumption that
(ΣΓ-i-P.) is invertible. If we let tx = t for i = 1,2, , m, then

det ΓJ Pt + Σ Pt Γ A(s)Q(s,to)ds\
Li = l 1 =1 Jto J
(3.9)
= [det ( g P.) j [det (E + £ A(5)0(5, fo)<fa) ].

From formula (3.9) we get

(3.10) det ( B + Γ A (5)0(5, ίo)ds) 7^ 0 for ί G D,

therefore if we would like our uniqueness criteria to include initial value

problems then assumption on the invertability of (ΣΓ=iP/) is
necessary. It is also easily observed from formula (3.2) that if A is
"small" then by (2.6), Q(tu t0) is also small. Therefore if A (t) is "small"
enough then

Σ f l Γ A(s)Q(sίto)ds)
i=l Jto I

is small enough and this implies that the "perturbation" of the invertible
matrix (ΣΠiPi) will be still invertible to guarantee a unique solution to
B.V.P. (2.1), (2.10). The uniqueness criteria for the Nicoletti problem
which appears in the literature under the term nonoscillation is a
particular case of our formulation.

PROPOSITION 3.2. Assume the systems (2.3) and (2.4) to satisfy

Assumptions 2.1, 2.2, or 2.3.

Then B.V.P. (2.3), (2.10) has a unique solution for every vector C
and every m points U ED, i = 1,2, , m iff the only solution to the
B.V.P. ΣfLi PSiti) = 0 of the system (2.4), for every m points ί, G D, is the
zero solution.

Proof. The difference of two solutions of (2.3) is a solution of the

corresponding homogeneous system.

PROPOSITION 3.3. Let (2.3), (2.4) satisfy Assumptions 2.1, 2.2,

or 2.3.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 113

Denote by PNh i = 1,2, , n the matrices which have all of their

entries zero, except the entry in the (i, /) place is 1.

0. T
0
''0
(3.11) —>
\o
0 1 '
i 0

Then (2.4) is nonoscillatory on D iff for every n points t, ED, i =

l,2, ,n

(3.12) det \E + 2 ^ f'' A (s)G(s, ίo)dsl ^ 0.

L i= l Jίo J

Proof. This follows from the observation

(3.12)

and Proposition 3.1.

We observe from Proposition 3.1 part (ii) that the B.V.P. (2.10), and
the B.V.P. with

(3.13)

are equivalent from the point of view of uniqueness.

We make another observation.

PROPOSITION 3.4. Assume 2.1, 2.2 or 2.3 to hold in D and in a

larger domain D with D CD and D/D/0 for the system
(2.3). Assume B.V.P. (2.3), (2.10) to possess a unique solution in
D. Then there exists a domain D, D CD CD such that D/D^0, and
is such that B.V.P. (2.3), (2.10) has a unique solution in D.
 114 H. GINGOLD

Proof. Since d(t0, tu t2, , tm) is a continuous function in all of its

variables for U E D and for ί, E D, / = 0,1, , n, it follows by Proposi-
tion 3.1 that d(to,tut2, -m,tm)/0 in a domain D with the desired
properties.

4. Uniqueness and non-linear systems. We point out in

this section the modified results of §3 which carry over to nonlinear
systems.

PROPOSITION 4.1. Lety(t), y(t), be two solutions of the system (2.1)

which satisfies Assumptions 2.1, 2.2 or Assumption 2.3.

Then their difference z (t),

(4.1) i(O = y(O-y(O,

satisfies the linear differential system

(4.2) z' = B(t,y,f)z,

and the integral equation

(4.3) z(0 = f(ίo)+ P B(s,y,f)2(s)ds.

Jto

Proof Since y(ί), y ( 0 are solutions of (2.1) we obtain by subtrac-

tion

(4.4) y'(O-yV) = B(t,yJ)(y(t)-f(t))

using Notation 2.4. Formula (4.1) yields immediately (4.2) and (4.3).

PROPOSITION 4.2. With the assumptions of Proposition 4.1 we have

(i) The B.V.P. (2.1), (2.10) has a unique solution if

(4.5) d(ίo,ίi, ,ίm,y,y)^O

where

ΣPi + ΣPi B(s,yJ)C(s,ίo)ds\.

ι =l i=l Jίo J

(ii) Let L and R b e n X n invertible matrices, then B.V.P. (2.1),

(2.10) has a unique solution if
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 115

L ( Σ Pi) R + Σ LPi I B (s, y, y4)C(s, ro)*<fe J ^ 0

(iii) Choose in (4.7) L = (ΣjΊi P,)"1 and JR = E using the notation of

(3.4), we have that B.V.P. (2.1), (2.10) has a unique solution if

[ E + ΣPil f
B
J

& y» y )<?(*. ίo)rfs J ϊ o.

i= l Jίo
Proof. By use of Proposition 4.1 we deduce that if y(ί), y(ί), are
two solutions which satisfy (2.1), (2.10), then their difference satisfies the
linear differential system (4.2). We deduce further that

(4.9) ( Σ P, + Σ Pi I" B(s, y, f )C(s, to)ds) (η-η) = 0.

Formula (4.9) implies that if (4.5) holds then rj - rj = 0. Since our

system is locally Lipschitzian, rj = rj implies y (t) = y (t) for every t G D
and parts (i) to (iii) follow.
We turn now to an analog of Proposition 3.4.

PROPOSITION 4.3. Let Assumptions 2.1, 2.2 or 2.3 hold on a domain

D and D._
Let D CD and D/D/0.
Let d(to,tu- -,tm,y,y)?έO for any two solutions of (2.1), for all
U E D, i = 0,1, , m, (including y{t) = y(t)).

Assume B.V.P. (2.1), (2.10) to possess a unique solution in D for

ί E D , i = D,l, ,m.
Assume that from the family of pairs of distinct solutions of B.V P.
(2.1), (2.10) in D it is always possible to choose a uniformly convergent
subsequence of pairs of solutions. Then B.V.P. (2.1), (2.10) has a unique
solution for all U E D, i = 0,1, , m where D is a domain such that
DCDCD and

Proof Assume on the contrary that there does not exist a D with
the required properties.

Let Dr be a sequence of connected domains such that Π "=1 Dr = D,

DC, , CD. , CD3CD2CD1 and Dr/D^0. Then there exist m
tuples (ίlr, , tmr), r = 1,2, , with at least one coordinate in Dr/D, and
there exist pairs of distinct solutions (yr(t),yr(t)) in Dr such that

(4.10) d(tln ,tmnynyr) =0

116 H. GINGOLD

and such that

(4.11) limy Γ (ί) = y(O and lim yΓ(ί) = y(t).

Choose now in (4.10) a subsequence of r = 1,2, namely rk —»oo

with fe -» oo such that ί/rk —» U and ί, E D. This yields a contradiction to
our assumption.

5. Nonuniqueness.

PROPOSITION 5.1. Let Assumptions 2.1, 2.2 hold for the system
(2.3).

Assume B.V.P. (2.3), (2.10) has a unique solution for all U E D,

Ϊ = 1, , m, and all C E C", but does not have a unique solution for all
( 6 D U dD, i = 1, , m, and all C E C .
Then there exist two distinct solutions to B.V.P. (2.3), (2.10), with
ti E dD, i = 1, ,m.

Proof. This theorem is a corollary of the maximum

principle. Consider d(t0, tu t2, , ί m ) as a function of m variables in the
closed domain (D U <9D)m and also as a function of m variables in the
open domain Dm.
Consider now a sequence of simply connected open domains Dn
r = l,2,

D 1 CD 2 C£>3, ,CDΓCD.

Denote their boundaries by ΘDr and let dDt Π ^D 7 = 0 for iV / and

(9DΓ Π 5D = 0 . Let also UΓ-i D Γ = D.
By Proposition 3.1, d(to,tl9 ,tm)έθ for t> E Dn r = l,2, .
Therefore by the maximum principle

(5.1) min \d(t0, tu , ί M )| = | d(ί 0 , ίi π ^ , tmr)\ tx EDrU dD,

where tιr E ^D Γ , i = 1,2, , m.

This follows from the fact that our function d(tOy tu , tm) is analytic
m
in (DΓ U <9DΓ) .
Let (ί!, , tm) be an m-tuple in (D U 3D)m such that
d(ίo, ίi, * * *, tm) = 0 and such that at least one of the coordinates tx belongs
to dD.
Choose a sequence (ί l r , , ί mr ) of m -tuples which converges to
(r b , tm) for r -» + oo, such that tir E Dr U dDr.
By (5.1) we have
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 117

(5.2) Id ( t l n , tmr)I i= I d ( t l n , i m r )I tir G D r U 3 D n

tir G 3Dr.

Choose a subsequence from r = 1,2, such that (ίi rk , , ί mrk ) tends to

some point (iu , ί m ) for /c ->°o, £ . E ^D.
Since
(5.3) lir

we obtain d(tu , ί m ) = 0 and the result follows.

We thus obtain Kim's result [28].

PROPOSITION 5.2. Let (2.4) satisfy assumptions of Proposition 5.1.

Let (2.4) 6e oscillatory on D U 3D but non-oscillatory on D. Then
(2.4) is oscillatory on 3D.

Proof. Choose in Proposition 5.1 the matrices PNι given by (3.11),

m = n, and observe by Proposition 3.3 the equivalence of nonoscillation
and the uniqueness of B.V.P. (2.4), (2.10).

6. A sufficient condition for uniqueness. In this section

we show that a special form of the matrices in (2.10) combined with a
proper form of the vector function f(t, y) leads to uniqueness of B.V.P.
(2.1), (2.10).

PROPOSITION 6.1. Let Assumptions 2.1, 2.2 or 2.3 be satisfied.

Let f(t, y) be such that

(6.1) fi{Uy) = f{t,yU"^yi\ i = 1,2, •• ,m,

namely that f(t, y) depends on yu , y, but not on y, + 1, , yn.

Let the matrices Pi which appear in (2.10) be such that;
Pi are lower triangular matrices having all elements of the diagonals of
Pi zero except the element in the (i, i) place.

In each F, the first i - 1 rows have zero elements. Also, take n = m

in (2.10).
0 , ,0

(6.1) o, ,o, ,0

',- P'u, 0, ,0

, p'k = 0 for k ^j,

118 H. GINGOLD

Let y(t), y(ί) be two solutions of B.V.P. (2.1), (2.10). Then

y(
Proof. Because of the structure of /(ί, y) it is easily verified that
B(s, y, y) is a lower triangular matrix and therefore C(s, t0) is also lower
triangular. Inspection of the formula (4.6) reveals that condition (4.5)
will be satisfied if the elements on the diagonal of the lower triangular
matrix

(6.2) ( 2 P, + Σ Pi Π B(s, y, y )C(s, to)ds), are nonzero.

\i = l i=l Jto I

But this is true since the elements of the diagonal of the above
matrix have the form

p«exρ y bu(s9y,y)ds)

where bu(s9y9y) are the corresponding elements on the diagonal of

B(s, y,y).
A more direct proof of this theorem proceeds as follows.
By the structure of the Ph i = 1,2, , n one obtains that yι(t) and
yj(r) satisfy a first order differential equation and yi(ίi) = yi(ίi) which
implies y^t) = yλ(t) on D.
Assume by induction that y}(t) = y; (ί) on D for j <n and proceed to
prove y/+1(ί) = y/+1(ί) on D.
This is true since yj+i(t) and y;+i(0 are solutions of the same first
order differential equation yf = fj+1(t, yι(t), ,yy(O>y) a n d satisfy the
same initial condition at ί/+1.

7. A positivity criterion for uniqueness. Let us borrow

a definition which is a variant of a common concept in the theory of
quadratic forms.

DEFINITION 7.1. An n x n real matrix A will be called positive

definite (positive semi-definite) if the quadratic form x'Ax is positive
definite (positive semi-definite). By this definition A need not be a
symmetric matrix.
We denote by x, a real n column vector, and by xf its transpose.

DEFINITION 7.2. An nx n complex matrix A will be called posi-

tive definite (positive semi-definite) if for every n -column vector x, the
form JC'AJC is "positive definite" ("positive semi-definite"), JC is the
conjugate of x.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 119

PROPOSITION 7.1. Let A(t) be a real nXn matrix function

piecewise-continuous on [a,b].
Let g(t) be an n-column vector function piecewise-continuous on
[a,b].
LetA(t), L, R, Ph i = 1,2, , m be n x n real matrices such that;
L(ΣT^iPi)R is positive definite,
LPι(Urj=,1 A (tj))R are positive semi-definite matrices for i = 1,2, , m,
n = l,2, .
Then B.V.P. (2.3), (2.10) has a unique solution. (We consider only
real solutions).

Proof We use Proposition 3.1, formula (3.2). Multiply the matrix

in (3.3) on the right by x and on the left by JC'.
Then (3.2) holds iff for any nonzero vector x the quadratic form

A s
(7.1) JC'L (Σ P) RX + Σ x'LPi Γ ( )O(s, to)Rxds
\i = l ! i= l Jto

is positive.
Since by assumption xιL{ΣT=ιPi)Rx is positive it suffices to show
that for every ί = 1,2, ,m

(7.2) Vt = xιLPx P A (s)Q(s, to)Rxdsy

Jto

is nonnegative. By use of the infinite series (2.9) we get

(7.3) Vt= x<P, Γ A (s)Σ TnRxds.

Jto 0

Our result will follow if Vin given by

(7.4) Vιn=xtLPιί' A(s)TnRxds, i = 1, ,m, n = 0,1, ,

Jto

is nonnegative.
We choose t0 = a and proceed to prove that Vin given by

(7.5) Vin = Γ Γ Γ\ , p " xXF,Λ(5) (Π A(fi)) Λ x ^ Ίdζ1ds9

Jto Jto Jto Jto \/ = l /

is nonnegative. (The formula (7.5) is valid for n > 1 its modification for
n = 0,1, is obvious). But this is true since ί, g t0 and the integrand is a
120 H. GINGOLD

nonnegative function, and the result follows. A similar result holds for
complex-valued systems.
We state the result as a proposition without proof.

PROPOSITION 7.2. Assume the n x n matrix function A (t) and the

n-column vector function g(t) to be piecewise-continuous on [a,b].
Let A(f), L, R, Pt9 i = 1,2, , m be n x n matrices such that:
L(ΣΊL1Pi)R is a positive definite matrix,
LPi (ΣJΪ; A (tj))R are positive semi-definite matrices for i = 1,2, , m,
r = l,2, .
Then B.V.P. (2.3), (2.10) has a unique solution.
We turn to a variant of Proposition 7.1 and 7.2.

PROPOSITION 7.3. Assume the nx n function matrix A (t) and the n

column vector function g(t) to be piecewise-continuous on [a,b].
Let A(t), L, R, Ph i = 1,2, , ra, be n x n matrices such that;
L(ΣilιPi)R is a positive definite matrix,
L&?-ΐPi)<p*i.ιA{ζi))R, forv = l,2,-;m-l,r = l92,-' are posi-
tive semi-definite matrices.
Then B.V.P. (2.3), (2.10) has a unique solution if U ^ ί 2 ,'' s = tm.

Proof The proof follows by rearranging the terms that appear in

the matrix considered in (3.3).
Since

(7.6) Σ Pι IU A (s)Q(s, to)ds = Σ ( Σ P ) \V+l A (s)Q(s, to)ds.

i= l Jto v=0 \i = v+\ I Jtv

We obtain by Proposition 3.1,

(7.7) det [ Σ Pi + Σ ( Σ P) Γ+ί A (5)0(5, /o)^l * 0

Li = 1 ^=0 \i = v+l / Jtv J

implies that B.V.P. (2.3), (2.10) has a unique solution.

We choose in (7.7) t0 = tx and repeat the argument of Proposition 7.1
to obtain the desired result.
Our theorem requires that a product of matrices be positive
definite. We may expect in some circumstances to use the well known
theorems where the signs of minors of a symmetric matrix are used to
determine whether it is positive definite. With this in mind, we mention
the following lemma.

LEMMA 7.1. Let A and B be two n x n totally positive matrices (See

Karlin [25] Ch. 1), then AB is a totally positive matrix.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 121

Proof. See Karlin [25] Ch. 1, 2.

We thus have obtained the following proposition.

PROPOSITION7.4. Let A (t) be an n x n matrix function and g(t) an

n vector -function both piecewise continuous on [a, b]. Let the matrices
L, R, Ph i = 1,2, , ra, ΣΠiP. , A(t) be totally positive.
Let L(ET=iPi)R, LPtmz[A(ζ}))R be symmetric matrices. Then
B.V.P. (2.3), (2.10) has a unique solution.

Proof. By lemma 7.1 all the principal minors of the symmetric

matrices that are involved in V, and Vin (7.3), (7.5), are
nonnegative. This implies by Gantmacher [15] Ch. X, that V) are
nonnegative definite.
Similarly, L(ΣΓ=\Pi)R is symmetric and has all its principal minors
positive. Thus, all conditions of Proposition 7.1 are fulfilled; and the
result follows.

AN EXAMPLE. Let Ph i = 1,2, , m, be totally positive symmetric

matrices which commute with the totally positive symmetric matrix
A. Let z{t) be a nonnegative, piecewise continuous function on [α, b].
Let the eigenvalues of (ΣΓ-i-R) be positive. Then the B.V.P.

y' = a(t)Ay,

has a unique solution. This is true by choosing in Proposition 7.4

L = R = E.

8. Characterization of best constants. We are going to

determine the existence of best constants p such that

(8.1) Γ||A(ί)||A<p,
Ja
will imply that B.V.P. (2.3), (2.10) has a unique solution. First, consider
a couple of lemmas borrowed from Schwarz [53].

LEMMA 8.1. With every simply connected domain D which has at

least two boundary points, there exists a family of nonintersecting, smooth,
closed Jordan arcs dDn depending continuously on the parameter r,
0 < r < l , such that the interiors of dDr are bounded simply-connected
domains Dr.

Moreover;
122 H. GINGOLD

(8.2) A, CD,, for r,<r 2

(8.3) dDn(ΛdDn = 0, for r^ r2

(8.4) lim dD, = 3D

r-*l

(8.5) U dD, = U D, = DU 3D.

0<r<l 0<r<l

Proof. The lemma is a corollary of Riemann's mapping

theorem. If ψ(z) is the univalent function in \z \ < 1 which maps the
unit circle onto D, then dDr are the level curves corresponding to | z | = r,
and D r are the images of \z \ < r under φ(z).

8.2. Let A(z) be an analytic matrix function in D, continu-

LEMMA
ous in D U dD. Define I{r) for 0 ^ r < 1 to be

(8.6) J(r)=f ||A(2)|| I dz I,

JdDr

and set

(8.7) ί(l) = limί(r),

whenever the limit exists.

(If 1(1)<oo, A(z) is said to belong to Hx). Then \\A{z)\\ is a

continuous subharmonic function in D ; and I(r) is a nondecreasing
function of r. Moreover, if 1(1) < °°, then there exists a constant c such
that

(8.8) Γ
J io Γ

where t0 = φ(0), t = ^>(Reiθ)E D, and Γί? the path of integration, is the

image of the segment s = re'ι\ 0 ^ r ^ jR.

/. This lemma is a restatement of Lemmas 2 and 3 in Schwarz

[53].

DEFINITION 8.1. Let || || be any norm defined in the set of n x k

matrices, and such that
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 123

(8.9) \\AB\\ίa\\A\\\\Bl

where A, B are n x k and k x v matrices respectively. The number a

may depend on A and B but ά given by

(8.10) ά=supα<oo

is independent of A and B. Such a norm will be referred to as an

algebra norm.

PROPOSITION 8.1. Let || || be an algebra norm.

(i) Let (2.1), (2.3) satisfy Assumptions 2.3.
Then there exists a best possible constant ph 0 < pi ^ <*> such that

(8.11) ίb\\B(s,yJ)\\ds<Pl
Ja

implies B.V.P. (2.1), (2.10) has a unique solution.

(8.12) PJ = inf Γ ||A(s)||ds = supr

Ja

where the inf is taken over all matrices A (t) piecewise continuous on [α, b]
and satisfying

det ( Σ P, + J P, Γ A(s)Q(s, to)ds) = 0

\i = l i=l Jίo /
(8.13)
ί, £ D, i = 0,1, , m.

77te sup is taken over all r such that

(8.14) Γ||A(s)||ώ<r
Ja

implies B.V.P. (2.3), (2.10) has a unique solution.

REMARK. If A(t) is a matrix function on [α, b]9 and y(t), y(t), of

(8.11), are real valued solutions of (2.1), (2.10), we will obtain some pi
which will be denoted by ρIR.
If A (t) is a complex valued matrix function on [a, b], and y (/), y (/),
of (8.11) are complex valued vector solutions of (2.1), (2.10), we will
obtain some pi that will be denoted by ρic.
Clearly, ρΪC ^ pm
124 H. GINGOLD

(ii) Let (2.1) and (2.3) satisfy Assumption 2.1.

Then there exists a best possible constant p D , 0 < p D ^ o o such that for
fixed t0E D and every t E D connected to t0 by a Jordan path Γ CD, the
inequality

(8.15) Γ \\B(s,yJ)\\\ds\<pD
JtoΓ

implies B.V.P. (2.1), (2.10) has a unique solution.

(8.16) pD = i n f Γ
Jto

where the inf is taken over all A (t) which satisfy Assumptions 2.1, and are
such that

(8.17) det(2fl + Σ ^ Γ A(s)Q(s,to)ds) = O,

\ί = l 1= 1 Jtor I

/<ρr 4; E D, i = 0,1, , m and for every set of paths of integration, Γ, in D,

such that (8.17) is true.
The sup is taken over all r such that

(8.18) f ||A(s)|||<fe|<r,
J tor

for t0 fixed in D, and t E D, (8.18) implies the B.V.P. (2.3), (2.10) has a
unique solution. Γ is some path of integration imbedded in D which
connects t and t0.
(iii) Let (2.1) and (2.3) satisfy Assumptions 2.1, 2.2.
Then there exists a best possible constant pdD, 0 < pdD ^ & such that

(8.19) f ||B(s,y,y)|||</s|<p,D
JdD

i m p l i e s B . V . P . ( 2 . 1 ) , ( 2 . 1 0 ) possesses a u n i q u e s o l u t i o n for ( 6 D U 3 D ,
i = 0,1, , m.

(8.20) PdD =infί ||A(s)|||Λ| =

JdD

where the inf is taken over all A (t) such that

 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 125

det ( Σ P, + 5 Pf Π A(s)Q(s, to)ds) = 0,

\j = l i =l Jto J
(8.21)
t,ED, ι=0,l, ,m.

The sup is ίαfcen over all r such that

(8.22) ί ||A(s)|||ώ|<r
JdD

implies B.V.P. (2.3), (2.10) has a unique solution.

Proof. In view of Proposition 4.1 it is enough to consider linear

homogeneous differential systems.
We observe that a best possible constant may be + °o in the case
where P4 = 0, i = 2,3, , m, since then our problem is equivalent to an
initial value problem.
From the equation (2.6) one obtains after multiplying it on the left by
A(t) and integrating:

(8.23) Γ A(s)Q(s,to)ds = P A(s)ds + Γ A(s) f$ A(ζ)Q(ζ,to)dζds.

Jto Jto Jto Jto

By a norm estimation

(8.24) I f A(s)Q(Mo)&l
(I J II

By Gronwall's inequality (see Hille [23], Ch. 1)

Γ A(s)Q(s,to)ds\g{' ||A(s)|||Λ|
J to I! J to

(8.25)

+ £ αjAWϋ ( £ llA(OIΠ^l) [exp ([ aζlA(ζ)\\\dζ\)] \ds\

Applying integration by parts to the right hand side of (8.25) one obtains

g £ || A (5)|| [exp
126 H. GINGOLD

(8.26)

Set

(8.27) max£ \\A(s)\\\ds\ = r.

Combining (8.26), formula (3.5), and demanding

one obtains that

(8.29) r<ά-l

is sufficient to guarantee (3.5).

(i) The inequality (8.14) guarantees the existence of a pi < 0 such
that (8.11) implies uniqueness for B.V.P. (2.1), (2.10). (Choose for
instance t0 = a). The proof for the inf sup characterization of the best
possible pi follows from its definition. Consider the norm || ||1>2 to be
that induced on a matrix A by two vector norms || \\u \\ \\2 by defining

(8.30) Milk

In the case m - n, PNi = Ph \\ || = \\ | | u defined above, we see that pm

coincides with

(8.31)

where the inf is taken over y ( ί ) ^ 0 and y,(ίi) = 0.

The result (8.31) is due to Friedland [14]. By his definition of the ρh
it is not clear that p 7 > 0 and that there exists a best possible p / c . Our
approach yields this information immediately.
(ii) It follows from (8.29) that if for every t E D, there exists a
Jordan path Γ C D such that ί \\B(s,y,y)||\ds\<r holds, then B.V.P.
(2.1), (2.10) has a unique solution. By definition, pD is the best possible.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 127

(ii) By Proposition 5.1 we are guaranteed that B.V.P. (2.3), (2.10)

has a unique solution for ί, G D U 3D iff B.V.P. (2.3), (2.10) with U E 3Dr,
i=0,l, ,m, for every r, 0 < r ^ l , has a unique solution. Recall
Lemmas 8.1, 8.2. By the inequality (8.29) we are guaranteed that there
exists βdD such that (8.19) implies B.V.P. (2.1), (2.10) has a unique
solution, for tx E 3D, i = 0,1, , m. By Lemma 8.2 this implies that
also

f \\A(s)\\\ds\<βdD for 0<r=Sl,

J dDr

which implies that B.V.P. has a unique solution for tx E D U 3D, i =

0, l, ,m. This argument shows that pdD given by (8.20) is the best
possible.

PROPOSITION 8.2. With the assumptions of Proposition 8.1, with

m =2,

(8.32) 2pίC^p,D.

Proof. Because of the Riemann mapping theorem and Lemma 8.1

it is enough to consider the case D = { ί | | f | < l } . Let expiθi, exρ/02,
0 < 0j ^ 02 < 2ττ, be two points belonging to j t \ = 1.
Then,

2pIC>ί ||A(ί)|||Λ|= Γ'||A(exp/β)||dβ

J|f| = l Jo

(8.33) = Γ2" \\A(e»)\\dθ =2 Γ \\A(expiθ)\\dθ

Jθi J θi

θo

for some 0O, Θί<θo<2π + θu

By (8.11), B.V.P. (2.10) for the system

(8.34)

must have a unique solution on each of the intervals [θu 0O], [0o, 0i + 2ττ].
(Remark, the best constants pIR or pIC do not depend on the endpoints a,
b, of an interval.)
Since 02 must be either in [θl9 0O] or in [0O, 0i + 2ττ] our result follows.
The relationship among ρh p D , pdD seem to be of a very interesting
128 H. GINGOLD

nature. However a general theory linking their magnitudes seems to be

lacking.
For the Nicoletti problem there are results indicating that ρD = 2pIR
for certain norms.
Compare, for instance, Lasota and Olech [32] and Nehari [45] (who
found ρIR = τr/2, for the norm induced on A(t) by the Euclidean vector
norm) with Schwartz [55] and Nehari [47], who found ρdD = π.

9. Estimation of best constants.

PROPOSITION 9.1. Let assumptions of Proposition 8.1 hold.

(i) If || || is an algebra norm, then

(9.1)

(9.2)

(9.3)

(ii) Let L be a matrix in a block partitioned form

L= = 1,2, ••-,*.
(9.4)

(9.5) L — (Li, L2, * , Lk),

then if || || is an algebra norm such that

(9.6) S max HI, |

(9.7)

(9.8)
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 129

Proof, (i) We demand that (8.28) will hold, so as to guarantee

(3.5). A fortiori, if

Γ A(s)Q(s,to)ds
Jto

(9.10) ^ ( I \\Pt l) max

then (8.28) is satisfied.

If D is the interval [α, b], choose t0 such that

(9.11) \ ίb\\A(s)\\ds= ί" \\A(s)\\ds = Γ\\A(s)ds\\^ ί" \\A(s)\\\ds\.

" J a J to J a J to

Since (8.26) implies

(9.12)

we obtain (9.1) and (9.3).

In order to prove (9.2) we choose on 3D a point tQ such that

(9.13) if
^ JdD

and (9.2) follows.

(ii) We proceed as in part (i) and exploit the norm property to
obtain

U A(s)Q(s,to)dsl
|| i = l Jto ||
(9.14)
g \\(PU A, , Pm)\\ max I f' A (s)Q(s, to)ds I.

We use (9.12) and repeat the same arguments as in part (i) to obtain our
result.
In particular, choose m = n and PNi = P, given by (3.11). If we also
take our norm to be the maximal row norm, then the condition of our
proposition part (ii) holds with a = 1. Then \\(Ph , Pm)\\ = 1 and we
obtain, Schwartz [53],
130 H. GINGOLD

(9.15) pm^p[C^ 2 In 2.

It was shown by Friedland [14] that actually ρIC = 2 In 2.

By Proposition 3.1 it follows that any hypothesis on A(t) which
makes the eigenvalues of

(9.16) ΣPiί' A(s)Q(s,to)ds

i =l Jto

lie in the interior of the unit disk serves our purposes. Let us introduce
a few notations for the next proposition.

NOTATION 9.1. Given a matrix A = (α iy ), i = 1, , n, / = 1, , k,

we denote by \A | the matrix \A \ = (| ai} |).

9.2. Given the matrices A and M of the same dimen-

NOTATION
sions, \A\^M means that for every i and /, | αί; | ^ mf/ .

NOTATION 9.3. We denote by | λ | (A) the maximum of the absolute

values of the eigenvalues of A.

PROPOSITION 9.2. Let the assumptions of Proposition 8.1

hold. Then :
(i) If

(9.17) \B(t,yJ)\^M for te[ayb]

and

(9.18) |λ I ( ( I \P, |) {[exp(|(6 - a)M] - E}) < 1

or if

(9.19)

and

(9.20) min|λ I {(M), ( ( g \P, | ) M ( £ - Af)-)} < 1

ί/ien B.V.P. (2.1), (2.10) has a unique solution.

 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 131

In particular, if ΣΓ=i|P; | = E, condition (9.18) is replaced by

(9.21) | λ | ( M ) < 2 ( f c - α ) - 1 In 2

and condition (9.20) is replaced by

(9.22) | |

(ii) Let D be a simply connected domain in the complex plane such

that for any m points in D, ί, E D, there are m simple Jordan arcs Γ,,
i = 1, , m, connecting those points to a point t0 in D such that the
length of each Jordan arc does not exceed /. Then if

(9.23) \B(thyJ)\^M for ί G D

and

(9.24)

or

(9.25) f" \B(s,y,f)\ds<M,

JtoΓ.

and

(9.26) min |λ I ί(M), ( ( | ) |P, | ) M(E - My1)} < 1

then B.V.P. (2.1), (2.10) has a unique solution.

Proof. The proof is based on majorizing the infinite series

(2.7). By (2.7) it follows that condition (9.17) implies

ΣP.r B < * Λ Λ C ( * « * s(ΣIΛI) (Σ flί^fMI')

«=i ' Jto \i = i I \k=\ K! /
(9 27)
= ( Σ \P> l) {[expθ(* - β)M)] - E)

Conditions (9.19) and (9.20) imply by the use of (2.7), that

132 H. GINGOLD

.l B(s,yJ)C(s,to)ds
J \i = l / \fc=l /
(9.28)

We recall from Gantmacher [15] Vol. 2, Ch. 2, that \A \ ^ M implies

(9.29)

Since for a function / and an n x n matrix A, we have λ(/(A)) =

/(λ (A)), it follows that (9.21) and (9.22) hold. A similar discussion leads
to the results (9.24), (9.26).

10. Some historical r e m a r k s . In 1897 Nicoletti [49], consi-

dered the existence and uniqueness of the boundary value problem (2.1),
(2.10), with PNi = Ph m = n given by (3.11). His main result may be
reformulated as follows:
Given the differential system (2.1) subject to the Assumption 2.3
there exists an interval [a, c], a<c^b, such that for every n-tuple
(tu t2, -,tn), a ^ti^c, i = 1, , rc, there exists a unique continuously
differentiable solution which satisfies (2.1), (2.10).
Since then, this B.V.P. has been known as the Nicoletti
problem. Various generalizations of this B.V.P. have been considered.
For the existence problem, and for the existence and uniqueness for
general B.V.P. the reader may consult Antosiewicz [1,2], Conti [8,10],
Grandolfi [19], Lasota and Opial [30,31], Lasota and Olech [32], and
their references.
Assuming that a solution to the Nicoletti problem exists, the
question of its uniqueness gave rise to a wide literature which uses the
key word nonoscillation. If the matrix in (2.4) is the companion matrix
of an nth order differential equation and the Nicoletti problem has a
unique solution for every ff E D, ί = 1, , n, and every C in (2.10), then
the corresponding homogeneous nth order linear equation is said to be
disfocal in D.
For sufficient conditions for a system (2.4) to be nonoscillatory on
[α, b]9 the reader is referred to Lasota and Olech [32], Nehari [45,46],
and Friedland [13,14]. For sufficient conditions for (2.4) to be nonoscil-
latory in a complex domain D when A{t) is analytic in D the reader is
referred to Nehari [47], Schwartz [52-55], London and Schwarz [39], and
their references. It is possible to distinguish three directions in the
investigation of nonoscillation of (2.4).
(I) Formulation of necessary and sufficient conditions for (2.4) to
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 133

be nonoscillatory in D. For example, Proposition 3.1 satisfies such a

criterion. In general the necessary and sufficient conditions are formu-
lated on derived functions of the coefficient matrix A(t) rather then
directly on the matrix A(t). This is many steps before the criteria are
applicable and practical.
(II) Sufficient conditions on functionals defined on A (ί), to guaran-
tee the nonoscillation of (2.4) for all systems (2.4) were formulated. The
main theme which appears in them is as follows.
Γb
Let || || be an algebra norm. If A (s)ds is less then p, where p is
Ja
some positive constant, then (2.4) is nonoscillatory. The subject of
finding the appropriate p when || || is a norm induced onA(ί) by vector
norms took two courses;
(i) By direct estimation. The main tune in this trend is as
follows. If y(t) is a nonzero vector solution of (2.4) which satisfies
y,(t) = 0, i = 1, , n, then by direct estimation one shows that

lly(θf Λ - p
Γb
for some p > 0 . By using an ad-absurdum argument, ||A(s)||ds ^ p
Ja
is found to guarantee the nonoscillation of (2.4). Lasota and Olech [32],
used this geometrical argument to find for || ||2 (the norm induced on A
by the vector Euclidean norm) that p = π/2 is the best constant. This
result was reestablished later by Nehari [45]. Ad-absurdum arguments
to derive (8.1) as a sufficient condition for nonoscillation in case D is the
interval [α, b] or D a domain in the complex plane, were used by Nehari
[47], Kim [28], Schwarz [52-55].
(ii) By direct use of calculus of variations for real systems. See
Friedland [13,14].
(III) Simplification of the problem. Given a specific norm || ||,
induced on A by a vector norm it may happen that the best constant p in
(8.1) which guarantees nonoscillation of (2.4) does not depend on the
dimension n of the system (2.4). Indeed, this was hinted by Lasota and
Olech's work [32], verified by calculus of variations by Friedland [14] for
various norms, and established by Nehari [48], for || ||p, 1 ^ p, where || ||p
are the norms induced on A by the Holder vector norms. By Friedland
[14] it is not true that the best constant p is independent of n for every
norm.
Another simplification in the search for best constants p in (8.1) to
imply nonoscillation is due to Nehari [46], who showed that it is enough
to consider a "degenerate Nicoletti" B.V.P., namely that ί, E[α, 6],
 134 H. GINGOLD

i = 1, , n, may be chosen such that U coincides either with a or with

b. (A similar phenomenon occurs in the De la Vallee-Poussin problem
whose uniqueness is investigated under the name of disconjugacy of nth
order linear homogeneous differential equations. See Levin [35]. A
third type of simplification for nonoscillation problems in a closed
domain D in the complex plane was supplied by Kim [28], who made it
possible to substitute the question of nonoscillation of a system (2.4) in D,
by a question of nonoscillation of the system (2.4) on a closed Jordan arc
imbedded in D.
(IV) Conditions on the form of the matrix A(t) or on special
relations among its elements to imply nonoscillation of (2.4). Not much
in this direction seems to exist in the literature.
(V) Necessary conditions on A (t) if A (t) is analytic in D and (2.4)
is nonoscillatory. In this direction consult Lavie [33].

Acknowledgement. I wish to thank Professor G. Gustafson

whose valuable suggestions helped to bring this paper to its final form.

REFERENCES

1. H. A. Antosiewicz, Boundary value problems for nonlinear ordinary differential equations, Pacific
J. Math., 17 (1966), 191-197.
2. , Linear problems for nonlinear ordinary differential equations, Proc. U.S.-Japan Sem. Diff.
and Funct. Eqns., W. A. Benjamin, New York, 1967, 1-11.
3. P. R. Beesack, Integral inequalities of the Wirtinger type, Duke Math. J., 25 (1958), 477-498.
4. , On the Green's function of an N-point boundary value problem, Pacific J. Math., 12
(1962), 801-812.
5. R. Bellman, Introduction to Matrix Analysis, McGraw-Hill, 1970.
6. G. A. Bessmertnykh, Existence and uniqueness of solutions of a multipoint De la Vallee-Poussin
boundary-value problem for nonlinear differential equations, Translated from DifferentiaΓnye Urav-
neniya, Vol. 6, No. 2, (1970), 298-310.
7. E. A. Coddington and N. Levinson, Theory of Ordinary Differential Equations, McGraw-Hill,
New York, 1955.
8. R. Conti, Recent trends in the theory of boundary value problems for ordinary differential
equations, Bull. Un. Mat. Ital., 22 (1967), 135-178.
9. , Equazioni differenziali ordinarie quasilineari con condizioni lineari, Ann. Math. Pura.
Appl., 57 (1962), 48-62.
10. , Problems lineaires pour les equations differentielles ordinaires, Math. Nachr., 23 (1961),
161-178.
11. M. M. Day, Normed Linear Spaces, Springer-Verlag, 1962.
12. N. Dunford and J. T. Schwartz, Linear Operators, Part I, Interscience, New York, 1967.
13. S. Friedland, Non-oscillation and integral inequalities, Bull. Amer. Math. Soα, 80 (1974),
715-717.
14. , Non-oscillation, disconjugacy and integral inequalities, Mem. Amer. Math. Soc, 7
(1976), 1-78.
15. F. R. Gantmacher, The theory of matrices, Chelsea, New York, 1966.
 UNIQUENESS OF LINEAR BOUNDARY VALUE PROBLEMS 135

16. H. Gingold, Uniqueness of nonlinear boundary value problems of second order differential
systems and equations, (to appear).
17. H. Gingold and G. Gustafson, On the uniqueness of the De Vallee-Poussin problem, (to appear).
18. W. A. Goppel, Disconjugacy, Lecture notes in Mathematics, 220, Springer-Verlag, 1971.
19. Maria Grandolfi, Problemi ai limiti per le equazioni differentielles multivoche, Atti. Accad. Naz.
Lincei Rend, Cl Sci. Fix. Mat. Natur., 42 (1967), 355-360.
20. P. Hartman and A. Wintner, On disconjugate differential systems, Canad. J. Math., 8 (1956),
72-81.
21. P. Hartman, Ordinary Differential Equations, Wiley, New York, 1964.
22. E. Hille, Analytic Function Theory, Ginn and Company, 1962.
23. , Lectures on Differential Equations, Addison-Wesley publishing Co., 1969.
24. D. B. Hinton, Disconjugate properties of a system of differential equations, J. Differential
Equations, 2 (1966), 420-437.
25. S. Karlin, Total Positivity, Vol. I, Stanford University Press, California, 1968.
26. , Total positivity, interpolation by Splines and Green's functions of differential operators, J.
Approximation Theory, 4 (1971), 91-112.
27. , Some extremal problems for eigenvalues of certain matrix and integral operators,
Advances in Math., 9 (1972), 93-136.
28. W. J. Kim, Disconjugacy and disfocality of differential systems, J. Math. Anal. Appl., 26 (1969),
9-19.
29. H. B. Keller, Numerical methods for two-point boundary value problems, Blaisdell, Waltham,
Massachusetts, 1968.
30. A. Lasota and Z. Opial, An application of the Kakutani -Ky fan theorem in the theory of ordinary
differential equations, Bull. Acad. Polon. Sci. Ser. Sci. Math. Astronom. Phys., 13 (1965), 781-786.
31. , On the existence of solutions of linear problems for ordinary differential equations, Ibid.,
14 (1966), 371-376.
32. A. Lasota and C. Olech, An optimal solution of NicolettVs boundary value problem, Ann. Polon.
Math., 18 (1966), 131-139.
33. M. Lavie, Some function-theoretic aspects of disconjugacy of linear differential systems, Trans.
Amer. Math. Soc., 143 (1969), 153-171.
34. A. Ju. Levin, On certain estimates for differentiate functions, Dokl. Akad. Nauk. SSSR, 138
(1961), 37-38.
35. , Some questions on the oscillation of solutions of linear differential equations, Dokl. Akad.
Nauk SSSR, 148 (1963), 512-515.
36. , Distribution of the zeros of solutions of a linear differential equation, Dokl. Akad. Nauk
SSSR, 156 (1964), 1281-1284.
37. , Fredholm equation with a smooth kernel and boundary value problems for linear
differential equations, Dokl. Akad. Nauk SSSR, 159 (1964), 13-16.
38. A. Ju. Levin, The non-oscillation of solutions of the differential equation x(n) 4- pi(t)x(n~1) 4- +
pn(t)x = 0", Uspeh. Mat. Nauk, 24 (1969), 43-96.
39. D. London and B. Schwarz, Disconjugacy of complex difierential systems and equation, Trans.
Amer. Math. Soc, 135 (1969), 487-505.
40. J. L. Massera and J. J. Schaffer, Linear Differential Equations and Function Spaces, Academic
Press, New York, 1966.
41. A. McNabb and Alan Shumitzky, Factorization and imbedding for general linear boundary value
problems, J. Differential Equations, 14 (1973), 518-546.
42. Z. Nehari, The Schwarzian derivative and Schlicht functions, Bull. Amer. Math. Soc, 55 (1949),
545-551.
43. , Some inequalities in the Theory of functions, Trans. Amer. Math. Soc, 75 (1953),
256-286.
136 H. GINGOLD

44. , Some eigenvalue estimates, J. Analyse Math., 7 (1959), 79-88.

45. , Oscillation theorems for systems of linear equations, Trans. Amer. Math. Soc, 139
(1969), 339-347.
46. , Non-oscillation and disconjugacy of systems of linear differential equations, J. Math.
Anal. Appl., 42 (1973), 237-254.
47. , Oscillatory properties of complex differential systems, J. Analyse Math., 26 (1973),
413-429.
48. , A minimal-length problem in R", J. London Math. Soc. 8 (1974), 22-28.
49. O. Nicoletti, Sulle condizionia inυiziali che determinano gli integrali delle equazioni ordinaire,
Atti. Accad. Sci. Torino, 33 (1897-98), 746-759.
50. W. T. Reid, Principal solutions of non-oscillatory self-adjoint linear differential equations, Pacific
J. Math., 8 (1958), 147-169.
51. , Oscillation criteria for self-adjoint differential systems, Trans. Amer. Math. Soc, 101
(1961), 91-106.
52. B. Schwarz, Disconjugacy of complex differential systems, Trans. Amer. Math. Soc, 125 (1966),
482-496.
53. , Norm conditions for disconjugacy of complex differential systems, J. Math. Anal. Appl.,
28 (1969), 553-568.
54# y Mappings of domains by components of solutions of differential systems, J. Differential

Equations, 10 (1971), 314-323.

55. , Curves on the unit sphere and disconjugacy of differential systems, J. Math. Anal. Appl.,
39 (1972), 75-86.
56. R. L. Sternberg, Variational methods and non-oscillation theorems for systems of differential
equations, Duke Math. J., 19 (1952), 311-322.
57. D. V. V. Wend, On the zeros of solutions of some linear complex differential equations, Pacific J.
Math., 10 (1960), 713-722.

Received May 20, 1976 and in revised form April 6, 1977.

UNIVERSITY OF UTAH
SALT LAKE CITY, UT 84112