Oscillation Theory of Ordinary Linear Differential Equations

Oscillation Theory of Ordinary Linear Differential
Equations
JOHN H. BARRETT*
The University of Tennessee, Knoxville, Tennessee 37916
Preface
The purpose of this article [a compilation of lectures originally

presented at the Associated Western Universities Differential Equations
Symposium, Baulder, Colorado in the summer of 19671 is to give
motivating examples and ideas which ‘have influenced the author in his
studies of oscillation properties of solutions of linear ordinary differential
equations. This is a subject where the mathematical tools needed are
relatively elementary but where it is easy to state an unsolved problem.
For example, the familiar second-order equation y” +\ Q(X)JJ= 0 is still
a valid subject for research, although it has a voluminous literature.
As far as oscillation theory is concerned, most texts in Differential
Equations, both elementary and advanced, deal only with second-order
equations. A few deal with self-adjoint fourth-order equations and,
perhaps, those of arbitrary even order and systems of first or second-order
equations. Any discussion of oscillatory properties of third-order
equations or other nonself-adjoint equations is hard to find and that is the
lowest order where truly nonself-adjoint equations occur.
In this article an attempt is made to give a self-contained inductive
development from equations of one order to the next. Most of the
discussion will deal with equations of second, third, and fourth orders,
with linear systems of second-order equations and with generalizations
of those results to equations of higher orders. No attempt is made to
survey all of the oscillation theory of equations of orders higher than
four. Considerable attention is devoted to equations of order three
(Section II) and this is in line with the increased recent interest in these
equations, as the Bibliography will show.
Instead of the usual format where proofs follow the respective
* We regret to report that Dr. Barrett died on January 21, 1969, and therefore that
this paper is being published posthumously. We are grateful to Drs. John Bradley,
William J. Coles, and John W. Heidel for reading the proofs.
415
0 1969 by Academic Press, Inc.
6071314-1
416 JOHN H. BARRETT
theorems, motivating examples and developments of the ideas are given

first with statements of theorems following as summaries of what has
been established in the discussion.
I. Second-Order Equations
Much of the material in this introductory section is contained in

introductory texts on Ordinary Differential Equations. Only those
topics are included which are pertinent to the succeeding discussion of
equations of higher order. For further oscillation theory of second-order
equations see Chapters IV and XI of Hartman’s recent advanced text (47).
1.1. Basic Properties

The real linear second-order equation
Z,[y] - yv + A(x) y’ + B(x) y = 0;
(1.1) A and B E C(I), a<b<oo,
1 = [a, b),
is equivalent to a special case of the canonical self-adjoint form:
c4t) -qyl = (TY’)’ + qy = 0; r > 0, r & 9 E W),

where r = exp(JA) and q = rB. A function y is said to be admissiblefor
the operator L, on an interval I provided y and ry’ E C[I]. Note that y”
need not exist when Y is not differentiable. A solution of (E,) is an
admissible function for L, satisfying LJy] = 0 on I. Existence and
uniqueness theorems for (E,) may be obtained easily from the equivalent
vector-matrix form
(Vz)
Lemma 1.1. For any pair of numbers (cO, cl) and each c E I there
exists a unique solution of (I&) satisfying
Y(C) = co ! (r-y’)(c) = Cl .
There are several ways to transform (Z&2)back into the form (1.1).
Lemma 1.2. (a) If YE C’(I) then (I&) can be put into the form (1 .I).
OSCILLATION OF LINEAR DIFFERENTIAL EQUATIONS 417
(b) If r is not d#~entiable then the change of independent variable:
U-2) t = z (l/Y) 2 x = x(t), Y(t) = YWI

I a
yields the form (1 .l)-with respect to t-without a middle term,
ii+py=o, tat> = @?ew-
Note that Lemma 1.2(b) provides a method for removing the middle
term of (1.1) and differentiation of coefficients is not required, as it is in
the standard variation of parameters substitution
y = uv, v’ = (A/2)v.
Integration-by-parts of zZ,[y] (or zL,[y]), where z is an arbitrary

function with the exhibited derivatives, yields useful Lagrange Identities.
Lemma 1.3. (a) If y, z E C”(I) then
&[y] = {xy’ - yz’ + AYZ}’ + yZ2’[x],
where Z,+[z] = ( z’ - AZ)’ + Bz, an adjoint operator of 1, .

(b) If x and y are admissiblefunctions for L, then
GLYI = [GY’ - Y41’ + Y-u4

Note that I,+ = Za if A = 0, and in this special case I, and (1.1) are
self-adjoint. Since L, serves as its own adjoint operator, L, and the corre-
sponding equation (E,) are said to be self-adjoint.
1.2. Factoring and Disconjugacy

The Polya-Mammana (73, 79) factored form of L, is easy to establish
and is actually a variation of standard Wronskian properties.
Theorem 1.1. If L,[v] = 0 and v # 0 on I’ C I then
(4 VUYI = b4Y - YWI’

and
(1.4) UYI = (llV)PV2(Y14’l’

for each L,-admissible y on I’ and
(b) no nontrivial solution of (E,) has two zeros on I’.
418 JOHN H. BARRETT
Definition 1 .l. A second-order linear operator (e.g., LJ and the

corresponding homogeneous equation (e.g., E2) are said to be disconjugate
on an interval I provided that no nontrivial solution of the equation has
two zeros on I.
Let vr = r~r(x, a) be the solution of (E,) defined by the initial con-

ditions:
U-3) Yk4 = 0 and D,y(a) = (ry’)(u) = 1.
This solution is called the principal solution of (,?I&)at x = a and oscilla-

tion properties of (I&) can be given in terms of vr(x, a).
The Polya-Mammana factored form (I .4) may be used to prove the
following results which form a version of the Sturm comparison theorem.
Lemma 1.4. (a) If a 0 on
I0 = (a, b).
(b) If (E,) is disconjugate on an interval I then
c%‘) (ry’)’ + q1y = 0; q1 E C(I) and q1 < q on I
is disconjugate on I.
(c) If, in addition, r < r1 and r,&(I) then
(4”) (%Y’)’ + 91Y = 0
is disconjugate on I.
W. Leighton and 2. Nehari have isolated a crucial part of the standard

proof of the Sturm Separation Theorem as a Fundamental Lemma and
applied it to linear differential equations of orders greater than two (70).
Lemma 1.5. If u(x) and v(x) are dzflerentiable functions on [a, c],
a < c, u(a) = u(c) = 0 and v(x) # 0 on [a, c], then
(a) vu’ - uv’ = v2(u/v)’ and their Wronskian uv’ - vu’ has a zero
on (a, c) and
(b) there is a linear combination x = u - kv which has a double zero
on (a, c) (i.e., at x = f E (a, c) where z(f) = z’(f) = 0).
Note that if u and v are also solutions of (E,) and a < c < b then they
are linearly dependent, which contradicts the original assumptions of
Lemma 1.5, thus yielding the Sturm separation theorem.
Lemma 1.4(a) provides a method for proving oscillation theorems,

by use of nonlinear Riccati equations.
Lemma 1.6. Let y(x) be a solution of (E,) on I = [a, oo), such thet
y(u) = 0 undy’(a) > 0. Ify(x) > 0 on (a, co) then
(a) h = -ry’/y satisfies a Riccati equation
(1.5) h’ = q + (l/r)/22 on (a, co).
(b) If, in addition, y’(x) > 0 on [a, co) then Jz q is bounded on

[a, co). On the other hand, if jz q = co then y’(x) has a zero on (a, 00)
and y(x) is bounded on [a, co).
Once we have a zero of y’(x) we proceed to force a subsequent zero of
y(x). Hille (58) seemsto have been the first to note the following property,
which was of considerable use to Nehari (75) and the author (8) for
establishing necessary conditions for disconjugacy of (E,).
Lemma 1.7. Let y(x) be a solution of (E,) on I = [a, 00) such that
y(u) > 0, y’(u) = 0. If y(x) > 0 and q(x) 3 0, but f 0 for large x,
then &y(x) = (r(x) y’(x)) < 0 on (a, 00) and Jr (l/r) < co.
Theorem 1.2. If Jz (l/r) = CO, q 3 0 but $0 for large x, (E,) is

disconjugate on [a, co) and y(x) is any nontrivial solution of (E2) with
y(a) = 0, then y(x) y’(x) > 0 on (a, co).
It is often useful to know that disconjugacy allows the construction of
a nonzero solution on closed or open intervals (68). Recall that the
converse (Lemma 1.4) is also true.
The case of the finite closed interval is easy to prove. On the other
hand, if I = [a, b), a 0 on [a, x,> and (c~~)~+ (c~~)~= I,
where y%(x) = cl%(x) + c~%(x) and U, w is a given fundamental set

of solutions of (Q. There is a subsequence {n.J of {n], such that {c:j} and
{c?} both converge, and these limits define a positive solution on
10 = (a, b), which may or may not be zero at x = a. Similarly, a positive
solution may be found for open intervals I.
Theorem 1.3. If (E,) is disconjugute on an interval I then there exists

420 JOHN H. BARRETT
a positive solution of (E2) on I fog (a) I = [a, b], a < b < CC and (b)
I = (a, b), -CD < a 0 and J-z q = co then every

solution of (E2) h as inJiniteZy many zeros on [a, b).
This is a weak form of the Leighton-Wintner oscillation theorem. About

1949, both Leighton (67) and Wintner (108) eliminated the nonnegative
condition, q(x) 3 0. (See Theorem 1.2 below). However, the non-
negative coefficient case is of special interest and is more readily
generalized to certain equations of higher order.
Definition 1.2. A second-order operator L, , or the corresponding

equation (EJ, is said to be oscillatory {nonoscillatory) on an interval I
provided that every solution of (E2) h as in$nitely many (at most a Jinite
number of > zeros on I.
Although disconjugacy is an extreme case of nonoscillation they are

essentially equivalent for second-order equations.
Lemma 1.9. If L, , OY (E,), is nonoscillatory on [a, co) then there is

a number c E (a, co) such that L, , OY (E,), is disconjugate on [c, co).
Although it is well known that nonoscillation on [a, co) implies

disconjugacy for large x for second-order equations it was noted recently
by Nehari (77) that the analogous statement for equations of orders
greater than two is known to be true only for special cases. A useful
example and comparison equation is Euler’s Equation
(J-6) xy + ky = 0, K = constant,
which
(i) has solutions of the form xa where
01is real if K < *,
01is complex if K > 4,
(ii) is disconjugate if k < 4 and oscillatory if k > 4 on [I, co).

However, when (1.6) is put into the form (I$), with I = 1 and 4 = k/x2,
it does not satisfy the hypothesis of Theorem 1.1 on [l, 00) since ST q < a~
for all values of K, although we have oscillation for K > a. The following
is well-known.
Lemma 1.10. If Jz(l/r) < CL)and Jz 1q 1 < CO then equation (E,) is

nonoscillatory on [a, co).
Suppose that (E2) is nonoscillatory, i.e., there exists a solution U(X) and
a number b E (a, 00) such that U(X) > 0 on [b, co). Therefore, h = -YU’/U
satisfies the Riccati Equation
(1.5) h’ = q + k h2 on [b, cc).
Let Jam q = co and c 0 on [c, co) and, hence,
h(x) > g(x) = 1: (l/r) h2 > 0 on [c, 00).
Consequently,
g’ > (l19g2 and $(1/r) <g&

se
< co.
This proof for the following Leighton-Wintner Oscillation Theorem

was suggested by W. J. Coles.
Theorem 1.4. If sz l/r = COand J-z q = COthen the equation (E,) is

oscillatory on [a, co).
This theorem is the principal motivation for the discussion in the sub-
sequent chapters where higher-order analogs are established. Willet (107)
has recently pointed out that if sr (l/r) < co then the transformation
(1.6) t = [I; wy)]-1
yields the following corollary of Theorem 1.4.
Corollary 1.4.1. If Jc (l/r) < co but
(1.7) ,; P(S) [I; wy)]2 A = *
then (E2) is oscillatory on [a, CO).

422 JOHN H. BARRETT
In the case of (Ea), with Y = 1 and 4 3 0, Hille (58) achieved better

results than Lemma 1.8. Let sz (I/Y) = co, (2 3 0 but $0 for large x and
(Es) be disconjugate on [a, co) and let y(x) be a positive solution (EJ on
[a, CD). Then, by Theorem 1.2, h = --ry’/y < 0 on [a, co). Since h
satisfies the Riccati equation (1.5), it follows readily that
I &)I = 4.r) 3 j,” q and

s cl
z (l/Y) < -l/h(X).
Theorem 1.5. If Jz (l/r) = co, q(x) 3 0 but +O for large x on

[a, CO)then a necessarycondition for disconjugacy of (E,) on [a, CO)is
j: (I/Y) J^mq
x
< 1.
Hence, lim supsGr sz (1/Y) s: CJ> 1 is sufficient for oscillation of (E,).
1.4. The Priifey Transformation

The change of variables to polar coordinates of a nontrivial solution
Y(4 of Eq- (E2> in its phase plane
\ y(“) = P(X)sin e(x),

(1.9)
!&y(g = p(x) cos O(x); 4~ = YY',
is called a Priifer Transformation, after H. Priifer who introduced the

idea in 1926 (80). The approach given here is due to W. T. Reid (87) and
it is convenient for generalization to higher order self-adjoint systems.
Suppose that y(x) is a nontrivial solution of (E,) and let
(1.10) ~(4 = Z/Y"(X) + (D,Y(x))~ > 0.
Next normalize y and D,y by letting
(1.11) 44 = Y(4lf (4 and 44 = D,Y(~/P(~~-

By differentiating (1.10) and (1.11) we have
(1.12) P’/P = (1 /y - 4) SSl
and
(1.13) (Z,)’ = (&) 6;‘)(;l) ; b(x) = s,2(x)/Y(x) + q(x) ?(x).

Therefore, if e’(x) = b(x) then (1.9) is fulfilled and the Priifer differential
equations for (EJ are
(a) P’ = (lP)[(l/r - 4)~inWp,

(1.14) (b) 6’ = (l/y) cos2 0 + q sin2 f3
I = (1/2)(1/r + 4) + (1/2)(1/y - 4) cos B.
An alternate approach is to show that the nonlinear e-equation of

(1.14) has a unique solution for each given initial value of 8 at some a E I
and then to solve the other (linear) equation for p. Also, (1.14) may be
derived by differentiating (1.9).
Theorem 1.6. Each solution of equation (E2) may be expressed by

(I .9) whosepolar componentssatisfy ( 1.14).
Observe that if 19(x)is a solution of (1.14b) and B(b) = kr then
O’(b) = l/r(b) > 0.
Therefore, even though e(x) may not be monotone it is always increasing

at multiples of n, which has an important bearing on oscillation properties.
Theorem 1.7. In order for a nontrivial solution y(x) of (E,) to be

oscillatory on I = [a, 00) it is necessary and su@icient that fey any
e(x) E C’[a, CO),which satisfies (1.9),
A classical example of the usefulness of Theorem 1.7 is its application

to the Bessel differential equation. For simplicity let us take the order
zero and note that u = J,,(x) satisfies
x2d + xu’ + x224= 0 on (0, a).
Eliminating the middle term by
Y= 6 lo(x)
we have
y”+(l +$x”)y =0 on (0, a).
Now (1.14b) yields

0’ = 1 + (*x2) sin20
and, hence, t!?(x)+ co as x -+ co, so that Jo(x) is oscillatory on (0, 00).

424 JOHN H. BARRETT
The other Priifer equation (1.14b) also gives the useful information
that, for
P(X)= ~YYX>+ rr’cw~

p’ = [($x2) sin 201p
and, hence, for 0 < 1 < x < co,
Consequently y = l/x Jo(x) is bounded on [I, co) and
Jo(x) < M/z/X on [I, co), M < 00.
A boundedness theorem for equation (E,) may be found in the same

way from (1.14a). However, stronger results can be obtained by modifying
the transformation (I .9) into
(1.9’) y(x) = ~(4 sin 44, &y(x) = +> P(X) cosQ),
where w(x) is an arbitrary positive function of class C’(1). In this case

(see (4)) the Priifer equations are
fp’ = [(1/2)(w/r - q/w) sin 28 - (d/w) co9 t9] p,

(1.14’)
18’ = (1/2)(20/r + q/w) + (I /2)(zu/r - q/w) cos 20 + (w’/2w) sin 0.
The special case where W(X) = k > 0, a constant, yields immediately

a bound on solutions of (E,).
Theorem 1.8. If k is a positive number such that
.cnmI k/r - q/k / < cx)
then all solutions of (E,) are bounded on I = [a, CO). Furthermore, zy y(x)
is any nontrivial solution of (E,) and
B(x) = (1PI j-Z

nw + q/4
then there is a positive number A and a number a: such that
$+t [Y(X) - A sin[(@) + a)] = 0.

1.5. Quadratic Functionals

Let y(x) be any &-admissible function on [a, b], a 2 + d*

If, in addition,
UYI = 0, YW YW = 0, and YWYW = 09

then the quadratic functional
(1.16) qy; a, b] = J” [r(y’)2 - qy”] = 0.

a
Suppose that Eq. (Es) is’ d isconjugate on [a, co) and V(X) is a positive
solution of (Es) on [a, b], then the factored form (1.4) yields
(1.15’) YJ52[Yl = PY4YI4’) + ~~2KY/4’12 o* [a, bl.

By combining (1.15) and (1.17) and integrating over [a, b] we have
(1.17) I[y; a, b] z 1” [W2 - qy”] = 1; rv2[(J+J’12 + [J?y2+~l~.

a
If further conditions are imposed on q(x) then Theorem 1.2 gives the
sign of V’(X).
Theorem 1.9. If (E,) is disconjugute on [a, m), Jz (l/r) = 00 and

q(x) > 0 but gOfor large x then
(1.18) I@; a, 4 = I .b [W2 - qy21 > 0, u<b<co,

n.
for all nontrivial L,-admissible functions y(x) on [a, b] for which y(u) = 0.
W. T. Reid (84) established and utilized a more general criterion in
terms of focalpoints, i.e., x = a is said to be a focal point of x = b pro-
vided there is a nontrivial solutiony(x) of (E,) such that y(u) = 0 = y’(b).
Note that (1.17) holds for more general y(x).
Lemma 1.11. In order for (1.18) to hold for all nontrivial y(x) of
class D’ (i.e., piecewise continuous derivatiwe) satisfying y(u) = 0, it is
necessary and su#cient that if L2[y] = 0, y(b) = 1, y’(b) = 0 then
y(x) > 0 on [a, b], i.e., [a, b) contains no (left) focalpoint of x = b.
426 JOHN H. BARRETT
It is a simple matter to check that the positivity of the quadratic

functional (I. 18) is equivalent to Nehari’s criterion (75) which follows
in a slightly more general form given by the author (8).
Theorem 1.10. If A, denotes the least eigenvalue of
(yy’)’ + $y = 0, r(a) = y’(b) = 0, --co<a<b<co,
J” (I/r) = CO, q(x) 3 0 and q(x) $ 0 on any subinterval of [a, CO), then
(E,) is disconjugate on [a, 00) if, and only ;f, A, > 1 for all b E (a, co).
Various choices of functions of class D’ in Lemma 1.11 yield necessary
conditions
for disconjugacy, e.g., let 0 < 01< 1 < p < rx) and
(I/r)]6’2 for a < x < t < b,
wY)]“‘2 for a < t < x < b < co,
and the positivity criterion (1.18), together with Jz l/y = co, yields
Note that (1.8) of Theorem (1.5) is the special case when 131 = 0 and
/3 = 2. For other casesand for the discussion when Jf q = 00 see (8).
If we have the general casewhere we do not know the sign of Y’(X) we
have the same conclusion (1.18) when the admissible functions y(x) have
zeros at x = a and x = b. Recall, also, that (I. 17) holds for functions in
class D’ on [a, b], i.e., y E D’[a, b].
Lemma 1.12. If (I?,) is disconjugate on [a, b], a < b < co, then
(1.18) holds for each nontrivial function y(x) E D’[a, b] satisfying the zero
end conditions
y(a) = 0 = y(b).
Conversely, if there is a nontrivial solution U(X) of (E2) on [a, b] with

two zero, say,
U(Xl) = 0 = 24(x2), a SZ x1 < x2 < 6,
then
, elsewhere on [a, b]
E qa, b];
y(a) = 0 = y(b) and I[y; a, b] = 0. Hence, (1.18) is a sufficient con-

dition for disconjugacy.
The results of the preceding discussion are summarized in the concise
set of equivalent statements set down by Reid (83).
Theorem 1 .ll . The following statements are equivalent.

(i) Equation (E,) is disconjugate on [a, b].
(ii) If LJuJ = 0, uI(a) = 0 and u,‘(a) # 0 then u1 # 0 on (a, b].
(iii) If Ls[u,] = 0, z+(b) = 0 and u2’(b) # 0 then up g 0 on [a, b).
(iv) There is a nonzero solution u(x) of (E,) on [a, b].
(v) There is a continuousfunction u(x) such that
YU’ E C[a, b] and uL,[u] < 0 on [a, b].
(vi) I[y; a, b] > 0 for all nontrivial functions y(x) of class D’ on

[a, b] fog which y(a) = y(b) = 0.
More general admissible classesof functions have been used by Reid
(85-89).
Recently, Leighton (69) has used the positivity of the quadratic
functional to establish a very useful comparison theorem.
Theorem 1.12. If the coeficients of Eq. (E,) and another equation of

the same type,
F22) J%Yl = (WY’) + 4(X)Y = 0,

satisfy the inequality
(1.19) J; [(r - Y) zi2 + (q - q) u”] > Jb [ru’2 - qu2]

a
for someL,-admissible and E,-admissible function u(x) on [a, b] satisfying

u(a) = 0 = u(b), then any solution y(x) of (E.J such that y(a) = 0 has
a zero on (a, b).
42% JOHN H. BARRETT
A particularly useful special case is the following.
Corollary 1.12.1. If in Theorem 1.11, r(x) E f(x) on [a, b] then

(1.19) implies
b (q - q) 2 > 0.
s ,I
1.6. Asymptotic values at co

Hille (50) solved the integral equation
Y(x) = 1 - j= (t - x) q(t) Y(t) dt

+
to establish the following.
Theorem 1.13. If Y = 1 and 1: x / q(x)1 dx < GOand
(4 Yo(x>= 1, Y,(x) = 1 - sn (t - X) q(t) Elk&t) dt

.c
then
1 Y&x) - Yk&)l < (jn t / q(t)1 dt) k/k! ,

2
(b) There exists a solution Y(x) of the integral equation and, further-
more,
Y” + q(x) Y = 0 and lim Y(x) = 1.
x*m
Theorem 1.14. If Y = 1 and q(x) > 0 on [a, co) then in order for
(I?,) to have a solution approaching 1 as x < GOit is necessaryand sufficient
that Ja”xq(x) dx < co.
T. G. Hallam (44) has extended these theorems to equations of general
nth order.
1.7. Complex Equations with a Real Independent Variable

If complex coefficients and solutions are admitted in the second-order
equation (E,) some properties of real equations carry over but some do
not. For example, y = eis is a complex solution of y” + y = 0, which
has no zeros, although all real solutions are oscillatory. In general, if

U(X) and V(X) are linearly independent solutions of Eq. (E2) then
is a nonzero solution. Hence, the factoring (1.4) can be accomplished

for any operator L, if complex factors are allowed.
In (7) the author introduced a class of equations
(lo201 b’/!+)l + !+)Y = 0, dx) = q+) + $2@> f 0, qk E C[a, a>,
which for Q(X) real has the familiar solutions

z z
q cos
sin
I and
s q.
With this in mind, for any complex continuous function p(x) define
two complex functions
(1.21) s(x) = s[a, x; 41 and c(x) := c[a, x; q]
to form the unique solution pair of
(1.22) s‘ = qc, c’ = -4s
subject to the initial conditions
(1.123) s(a) = 0, c(a) = 1.
It is not difficult to verify the following identities, which shows that

(1.20), or (1.22) retains the boundedness property for complex q(x).
Lemma 1.13. 1 s I2 + 1 c I2 E 1.
The odd and even properties of the real case of (1.21) take on an
interesting form.
Lemma 1.14. If k is a complex number such that 1k 1 = 1 then
s[u, x; kq] = ks[u, x; q]

and
46 x; &II = 4% 3; 41 on [a, al).
430 JOHN H. BARRETT
By use of the complex polar form, i.e.,
~[a, x; q] = h(x) exp{k(x)}, q(x) = r(x) exp{iO(x)),
various oscillation and nonoscillation theorems can be established. In

particular, the system (1.22) becomes
(h’/Y)’ + ((1 + 6”) - (LX’/?,- 6)“) Yh = 0

(1.24)
h201’/r = 2 *‘bhh’, 6 = 0’/2v.
Jn
Theorem 1.15. If the real second-order
(y’/y)’ + (I + b”) Yy = 0
is nonoscillatory then s[a, x; q] has at mostajnite number of zeros on [a, 00).
Theorem 1.16. If r E C[a, GO),0, h and 01E C’[a, co) and

b(x) = B’(x)/~~(x) = constanton [a, co)
then s[a, x; q] has injinitely many zeros on [a, co) if and only if J-” 1P / = co.
Furthermore, zjcb # 0 then C[a, x; q] has no zeros on [a, CO).
The real Priifer transformation of Section 1.4 can be paralleled (7).
Theorem 1.17. If y( x ) is a nontrivial solution of (E,) with continuous

complex coeficients
Y(X) = YJX) + iY2(X), d-4 = 41w + cz2(~~)~ yk md qk E C[a, co),
such that y(a) = 0, then there exist a nonzero function p(x) E C’[a, co)
and a function Q(x) E C[a, co) such that
(1.25) \Y(4 = 44 4% xi !a
(Y(X)y’(x) = P(x)qu, x; Q].
Furthermore,
p’ = psc( 1/Y - q),
(I .26)
9 = (F!f)(I c l"/y + q I s I").
A slight modification of the first equation of (1.26) yields a boundedness

theorem.
Theorem 1.18. If k is a real positive constant and y(x) is a solution of

(E,) with complex coeficients then there is a positive constant M so that
A real matrix formulation of the complex scalar equation (E,) is noted

in Section V.
II. A. Third-Order Equations
Although differential equations of second and fourth orders have been

studied extensively, it is only recently that third-order equations have
been given serious study. Of course, there is the classical paper of
Birkhoff (16) of 1908 where he applied methods of projective geometry.
While BirkhofYs paper is a necessary reference fo any paper on third-
order equations, his results or methods are seldom quoted. In 1948
Sansone (91) gave a summary of results known to that date, as well as
a number of new results. The current interest in third-order equations
was kindled by Hanan’s 1961 paper (45), the 1960 paper of Azbelev and
Caljuk (3) and the papers of Gregus since 1955 (N-44). Other important
papers are those of Kondrat’ev (62,63), Svec (102) and Lazer (66)
2. I . Examples
If one solves and examines the oscillatory and nonoscillatory properties.
of solutions of
(a) y”’ + y = 0, (b) y”’ - y = 0,

(2.1)
(c) ym + y’ = 0, (d) y”’ - y’ = 0,
on half-line intervals [a, co), then he is armed with suitable motivating

examples for most of this chapter. For a long time it was thought that
the following was a typical property of all third-order equations.
Lemma 2.1. Every third-order equation with real constant coejicients,
(2.2) y”’ + czy” + q-y’ + coy = 0
has a nonxero solution on [a, 00).
60713 h-2
432 JOHN H. BARRETT
However, Sansone (91) and Gregus (30) have exhibited equations of

the form
o-2) [yr” + P(x)y]’ = yn’ + P(x)y’ + P’(x)y = 0; P(x) E qa, co),
for which every solution has infinitely many zeros on [a, CO). Lazer (66)
has recently used the following as motivating examples.
Lemma 2.2.
(a) If the constants c1 < 0 and q, > 0 then
yI” + c,y’ + coy = 0 0 on 1% a)
has a nontrivial solution with injinitely many zeros a., and only if,
Also, all solutions have infinitely many zeros, except for nonvanishing
solutions.
(b) If c1 < 0 and c,, < 0 then the above criterion is replaced by
I coI - -& / Cl 13’2> 0.
Also, there exist two independent solutions with infinitely many zeros.
(c) If c0 > 0 and c1 > 0 all solutions of (2.1) have injkitely many
zeros, except for constant multiples of one nonvanishing solution.
Hanan (45) has used the third-order Euler Equation
(2.3) yn’ + (a/xZ)y’ + (b/x”)y = 0; a, b = constants,
as his basic comparison equation.
Lemma 2.2. (a) If a > 1 then there is a nontrivial oscillatory solution

of (2.3) (i.e., one with infinitely many zeros) on [ 1, co).
(b) i’f a < 1 and a + b > 0 then there is a nontrivial oscillatory
solution of (2.3) ifand only if
afh-2(+q2>0.
(c) If a < 1 and a + b < 0 then there is a nontrivial oscillatory

solution of (2.3) if and only ;f
The author (13) used the following product principle of Appel (I) to
predict his general third-order Canonical Equation [see (Es) below].
This property is well known to those who work with Special Functions.
Lemma 2.3. If w and z are solutions of (E,) (ry’)’ + qy = 0 then

their product y = wz is a solution of the third-order equation
(2.4) WY’) + 2qyl)’ + 2q(ry’) = 0; Y > 0, Y & q E C(I).
Let v0 = vO(x, a) and vr = a,(~, a) be fundamental solutions of (E,) with

v,,(a) = 1, D,v,(a) = (rv,‘)(a) = 0 and vi(a) = 0, D,v,(a) = (r?+‘)(a) = 1,
then a fundamental set of solutions of (2.4) is
y = v12/2 : y(a) = D,y(a) = 0, D2y(u) = 1 QY = YY’,

y = vovl : y(a) = 0, Dly(u) = 1, Dzy(a) = 0 where
y = vo2: y(u) = 1, D,y(u) = 0, Dzy(u) = 0 1 t D,y = Q,y + 2qy)‘.
Note that Eq. (2.4) h as a nontrivial solution with three zeros if, and only
if, equation (IZ2) has a nontrivial solution with two zeros. Of course, two
arbitrary zeros can be specified for any linear third-order equation.
Definition 2.1. A third-order linear d#erential operator L[y] (and

its corresponding homogeneousequation L[y] = 0) is said to be disconjugate
on an interval I provided that no nontrivial solution has three zeros on I.
Hence, the third-order equation (2.4) is disconjugate if, and only if,
the second-order equation (E2) is disconjugate.
A well-known special case is that when r = 1 and q E C’, for which (2.4)
becomes
P-6) ye’ + 4py’ + 2q’y = 0,

whose solutions are linear combinations of products of two solutions of
y” + qy = 0. An interesting example of (2.6) is
(2.6’) y”’ + 4y’ = 0,
whose solutions are linear combinations of sin2 x, co9 x and sin x cos x.
434 JOHN H. BARRETT
Hanan (4.5) failed to take into account such examples which contradict
his Fundamental Lemma. Fortunately, such cases did not enter into his
subsequent applications. Note that y = sin2 x and y = sin x cos x are
independent nontrivial solutions with at least three zeros on [0, ~1.
In fact, y = sin2 x has infinitely many double zeros on [0, co). A cor-
rected version is given in Theorem 2.9.
2.2. A Canonical Form

The equation (2.4), which was generated by products of solutions of
(EQ), suggested to the author (13) the third-order canonicalform
(-4) MY1 = {rd%Y’) + %Yl>’ + 42hY’) = 0,

where ri > 0, ri and qi E C(I), I = [a, b), a < b < co. Let D,y = rly’,
D,y = r,L,[y] = r2[(r1y’)’ + qIy]; then (ES) becomes
UYI = (GY)’ + !&J&Y = 0.

It may be that y”’ + p(x)y = 0 is not a typical equation of third-order
but, instead, that either
(2.7) (a) (y” + PY)’ = 0 or (b) y”’ + py’ = 0
is more typical for predicting oscillation properties. It should be noted

that (ES) contains Shinn’s quasi-differential operator (99). A function y is
L,-admissible provided that y, Dry and D,y E c’(I), and a solution of
(EJ is an &-admissible function which satisfies (EJ.
The equation (E3) is equivalent to a special case of the .$rst-order
vector-matrix equation:
(V3) 01’ = B(x)ol; B(x) = (6ij(x>), 6ij E c(1)~

where
0
B(x) = -q1
t 0
in the sense that if y is a solution of (EJ then
Y
a=
0
Yl
‘Y2
is a solution of (I’,) and, conversely, if 01= (uJ is a (vector) solution of

(2.1) then y = a, is a solution of (Es).
Again, the standard uniqueness and existence properties follow easily
by use of the system (V,).
Lemma 2.4. Any classical third-order equation
(2.8) UYI = Y”’ + I%(4Yn + PdX>Y’ + POWY = 0; P, E w,
can be expressedin the canonical form ( E3) since it is equivalent to
(sy”)’ + SP,Y + SPOY= 0; s = exp P2 ,

(S 1
and
[SY” + (j SP,)Y]’ + [SP1- j SPo]Yl = 0.
Obviously, the canonical form (ES) is not unique.

Suppose that y is admissible for L, and z has the desired derivatives;
then the following adjoint operator and Lagrange Identity are easily
derived by integrations-by-parts.
Lemma 2.5. If y is L,-admissible and x is L,+-admissible, where the

tidj’oint operator is
(2.9) h+[Yl = {W,Y’) + %zYl>’+ 4&2Y’),

then y and z satisfy the Lagrange Identity
(2.10) z&[y] = {z&y - D,+zD,y + y&+.4’ - y-k’ [xl,
where Dl+y = r2 y’ and Dz+y = rJ(D,+y) + q2y].

The corresponding adjoint equation is
(%+I L,+[yl = p,+yy + qp,+y = 0.
Note that the adjoint operators L+ and Di+ are obtained from L and Di ,
respectively, by interchanging rl with r2 and q1 with q2 . If rl = y2 and
q1 = qz then L + = L and (ES) is said to be self-adjoint (sometimes called
anti-self-&joint for the odd order 3). Observe that Eq. (2.4) and, its
classical special case(2.6) are self-adjoint third-order equations. Also, the
436 JOHN H. BARRETT
equations (2.7) are adjoints of each other. Dejine certain fundamental

solutions of (E3) by the initial conditions:
ug = u&x, u) : y(a) = 0 = D,y(a), D&4 = 1,

(Fiyst principal solution)
P-11) u1 = 24,(x, a) : y(a) = 0, 4YW = 1, D&> = 0;
(Second principal solution)
u(J= u&Y, u) : y(a) = 1, D1y(c7)= 0 = D2y(a).
Define ui+ = ui+(x, a) to be the corresponding fundamental solutions

for the adjoint equation (E3+). Recall that these fundamental solutions of
the self-adjoint equation (2.4) are given by (2.5).
Lemma 2.6. (a) Any solution of (EJ with a zero at x = a has the
f orm
Y(X) = C1UI(X, a) + ce+, a).
(b) In order for there to exist a nontrivial solution of (EJ with a zero
at x = a and a double zero at x = b # a it is necessary and su#icient that
the “ Wronskian” of the principal solutions u2(x, a) and ul(x, a),
(2.12) u(x) = W[u, , uJ = uzD,u, ~ z+D,u, = Y&L+~ - uIuZ’),
vanish at x = b.
The Leighton-Nehari Fundamental Lemma 1.5 is very useful for
showing when (2.12) h as zeros. Apparently, Birkhoff (16) first noted that
the Wronskian of two solutions satisfies an adjoint equation.
Lemma 2.7. (a) If y an d x are solutions of (E3) then their Wronskian

w = yD,z - zD,y (note inclusion of rl) is a solution of the adjoint
equation (Es+).
(b) The special Wronskian (2.12) satisjies the initial conditions:
u(a) = 0 = u’(a), Dz+g(a) = 1 and, hence, is the first principal solution
of (Es+), i.e.,
(2.13) WC%9UJX, u) = u,+(x, a).
2.3. Self-adjoint Equations
We have already noted the self-adjoint third-order examples (2.4)

and (2.6) but under what conditions is the classical general equation
(2.8) equivalent to a self-adjoint equation? Recently Giuliano (29) gave

sufficient conditions, which were later simplified by the author (I.?).
Theorem 2.1. (a) The third-order operator
fJCY1
= Y”’ + P&b” + Pd4Y’ + POWY= 0; pi Ef?‘(f),
satisfies the identity,
with r = exp{( l/3) Jp2>.

(b) The equation (2.8), ZJy] = 0, is equiwalent to a self-adjoint
equation if z = r2 = exp{(2/3) Jp,) satisfies
(2.14) (22”- 2P9) + 4P@z= 0.
For p, = 0 the conditions (2.14) simplifr to the special case(2.6), i.e.,
(2.14’) PI’ = PlJ/2*

Since the canonical form (Ea) is not unique the self-adjointness
conditions ri = r2 and q1 = q2 can be improved slightly.
Theorem 2.2. The canonical third-order equation (E3) is equivalent

to a self-adjoint equation on an interval I if there exist constants k # 0 and
m such that
(2.15) r2(x) = kr,(x) and q2(x) = kq,(x) + m/r,(x) on I.
2.4. The Lagrange Bracket

The Lagrange Identity (2.10) contains the bilinear functional
(2; Y> = ~D,Y - D,+Gy + (D,+-~Y
which is sometimes known as the concomittant or the Lagrange Bracket

of the operator L, . Observe that if L3[y] = 0 and L,+[z] = 0 on an
interval I then
{z; y} = constant on 1,
and this constant is determined by initial values at any particular point

438 JOHN H. BARRETT
x = a. Also, if a is given then {z; y} is a third-order differential operator

in y.
Lemma 2.8.(a) {uzmt;UJ = 0 for i = 1, 2.

(b) If z # 0 on an interval I then
(2.16) {z; y} = Y$
(c) In the self-adioint case
(2.17) {Y;Y> = ~YD,Y - (DRY)“.
It is well-known that, for all x, t ~1,
(2.18) u&, t) ?I= uz+(t, x)
and recently, Dolan (24) has noted a more general identity.
Lemma 2.9. The fundamental solutions (2.11) of (EJ and its adjoint
(ES+) satisfy
(2.19) D,u,(x, t) = (- l)a+BD,&$, x) for (Y,/3 = 0, 1,2.
Note that (2.18) is the special case where a: = 0 and /3 = 2.
2.5. Factoring and Disconjugacy

If V+(X) is a nonzero solution of the adjoint equation (E+) on an interval
I, i.e.,
(9 L,‘-[v+] = 0 and vf # 0,
then the Lagrange Identity (2.10) yields
(2.20) UYI = (1iv’>@+; Y>’
for all &-admissible functions y(x) on I. Now the Lagrange Bracket

{v+; y} is a second-order differential operator (2,16) and is constant if
L,[y] = 0. Any solution z, of (ES) for which
(ii) tv+; v) = 0
at some point of I is by Lemma 2.8 a solution of the differential equation

(ii) on the whole interval I. If
(iii) L&J] = 0 and v#O
and (ii) is satisfied then the Polya-Mammana factoring (1.4) may be

applied to {VU+;y> = 0. Actually, such factoring was first done by
Frobenius (28) and later utilized by Polya (79) and Mammana (73).
Lemma 2.10. Under the conditions (i), (ii), and (iii) on an interval I,
the third-order canonical operator L, factors into
(2.21)
for all L,-admissible functions y(x) on I.
We have already noted that, if vi and ~1~are solutions of (E3), then their
Wronskian
w = W[o, ) W‘J = w,D,v, - o,D,v,
is a solution of the adjoint equation (E+). Furthermore, it turns out that
{w; Vi} = 0 for i=l,2.
Therefore, we can obtain the Polya-Mammana factored form (2.21) by

letting
v+ = W[w, , q] and v = vi for i = 1 or 2.
Note that (2.21) im pl ies that L, is disconjugate on I by Definition 2.1.

If we choose two principal solutions from (2.1 l),
q = Ul(X, 4 and v2 = u2@, a>
then ZI+ = W[U, , ui] = ua+ and it takes only a slight extension of the
argument for Lemma 2.10 to yield a criterion for disconjugacy.
Theorem 2.3. The operator L, [and the equation (ES)] is disconjugate

on the half-closed interval [a, b), a 0 and u,+(x, Q) > 0 on (a, b).

440 JOHN H. BARRETT
As a consequence of the symmetry property (2.18) Theorem 2.4

becomes
Corollary 2.3.1. The criterion (2.22) for disconjugacy on [a, b) or

(a, b] is equivalent to
(2.22’) u,(x, t> f 0 for x, t E (a, b), x # t.
Because of the dual role of L, and its adjoint, La+, we have
Corollary 2.3.2. An operator L, {equation (ES)} is disconjugate on

[a, b), a < b < co, if and onZy ;f its udjoint L,+{E,+} is disconjugute on
[a, b).
By combining two adjacent factors of L, in (2.21) with v+ = W, we
have another factored form which Mammana (73) noted for the classical
equation (2.8).
Lemma 2.11. If v is a nonzero solution of (EJ on I then
(2.18)’
where
and
l2CYl =
(Le,’ + ( DzV+$92~ jY’
for all L,-admissible functions on I.
2.6. First Conjugate Points

Suppose that there is a nontrivial solution y(x) of (ES) with three zeros
(counting any double zero as two zeros) x1 \< xp < x3, xi < x3 on
I = [a, co); then the disconjugacy criterion (2.19) asserts that either
u2(x, a) or z++(x, a) has a zero on (a, x3]. If uZ(x, u) has a zero at x = b
then y = ua(x, u) satisfies the two-point boundary conditions
(2.23) y(u) = Dly(u) = 0 = y(b).

On the other hand, if us+(x, u) has a zero at x = b then there is a non-

trivial solution y(x) of (I&) satisfying
(2.24) y(u) = 0 = y(6) = D,y(b).
Definition 2.2. Let ~~~(u){z~~(u)} be the minimum number b E (a, co)

such that the boundary conditions (2.23) ((2.24)) are sutisjied by a nontrivial
solution of (ES) and +(a) = co if the respective boundary conditions are
not sutisjied.
Note that for the examples (2.1),
(a): ~~(4 < co, h(u) = 00,
(b): Z&U) = co, Z+(U) < 00.
Definition 2.3. If (ES) has a nontrivial solution with three zeros on

[a, co) then thefirst conjugate point q,(u) of x = a is deJned to be
(2.25) ~(a) = inf{x, ; a~x,dx,~~x,,Y(xi)=O,Y~:O,~,rYl=0}.
If (E3) is disconjugute we write q,(a) = co.

The disconjugacy Theorem 2.3 may now be restated.
Theorem 2.4. ~~(a) = min{z,,(u), +(a)).

Azbelev and Caljuk (3) were the first to arrive at an equivalent result.
Note that in the self-&joint case (2.4),
d4 = %(4 = %&>’
For the example (2.1),
(c): .z12(U)= ZZl(U) = z&u) = a + 277,
(d): +(a) = zar(u) = Z+(U) = co.
Definition 2.2’. The number zS2(u) is the smallest b E (a, 00) such that
(2.26) Y(4 = DlYW = 0 = Y(b) = hY(4
is satisfied by a nontrivial solution of (E3). Otherwise, write z&u) = co.

Hanan’s special classesof classical third-order equations (2.8) may be
described simply as
c, : z&t) = cx) for all t f [a, co)
442 JOHN H. BARRETT
and
c,, : zzl(t) = co for all t E [a, 03).
Motivating examples in these classes are given by
yN’ + p(x) y = 0, P E w, a>
which belongs to Cr, if p < 0 and to C, if p > 0 on [a, 00). The author
(23) has extended Hanan’s classes to the canonical equation (Es) by
examining the behavior of the functional
@rYl = 2Y4Y - (DIY)2*
Theorem 2.5. If (r&J and (r,q, - r,q,) E C’[a, a) and
(2.27) (r,/r,)’ < owx, (f-2%- w,)’ < ow>
but +O on any subintermd of [a, co), then zzl(t) = a3 (z12(t) = m} for

all t E [a, co).
For the equation considered by Hanan,
(2.28) UYI = Yl” + PIWY + P,WY = 0; pi E cya, co),
the conditions (2.24) Theorem 2.6 reduces to:
but $0 on any subinterval.
2.7. DisconjugucY Numbers
While Hanan discussed the cases where either ,zr2 or zsr = cc;,
Azbelev and Caljuk (3) considered the more general case where both
xl2 and G might both be finite. As a motivating example, they con-
structed a third-order differential equation of the type
(2.29) (ry")' + qy' = 0, r > 0, q > 0, r & q E C[O, 24,
which has as a fundamental set of solutions
I, sin x + 0.001x3, 1 - cos x

and for which (in our notation)
a = 0 < +(O) = 5.7 < xzl(0) = 2Tr.
Azbelev and Caljuk defined and used the following for the classical
equation (2.3).
Definition 2.3. (a) An adjacent pair of zeros x1 , x2 of a nontrivial

solution of (ES) is said to be an (i, j)-adjacent pair provided the multiplicity
at x = x1 is at least i and at x2 , at leastj.
(b) For each pair (i, j) of positive integers
(2.30) rij(a) = sup{t: no (i, j)-adjucent pair of zerosin [a, t)}.
(c) The maximum interval of disconjugacy [a, r(a)) is defined by
(2.31) r(a) = sup{t: (ES) is disconjugateon [a, t)}.
Using these concepts, they proved that

(i) 44 = min[r&), rz,(41y
and purported to prove that
(ii) da) = maxi?,,(a),r21(41.
The first assertion is equivalent to Theorem 2.4 since
However, the assertion follows only for certain cases, as we will now see.
After these lectures were given, T. L. Sherman pointed out the
following example.
Example 2.1. The three linearly independent functions
24= (l/2) sin2x cosX, v = co9 x sin X, w=l
satisfy the third-order equation
(2.32) (Y”/q + C~,,/P,)Y’ = cl on [O, 001,
where
f)(i) &)
cfij = for i,j =0, 1,2,3.
I VW UU) I
444 JOHN H. BARRETT
Since a12 = 1 + (l/2) sin” x 3 1 > 0, then (2.32) is nonsingular. It is

easy to calculate that
Q(O) = Z12(0) = zzl(0) = Trj2 = Y12(0) = r,,(O) < Za2(0) = rz2(0) = 7r
and, since a01 = u2+(x, 0) = (l/S) sin2 (2x),
171+(0)= z:,(o) = 2&(O) = 42 = &2(O) = rm.
Therefore, the assertion (ii) does not hold for Eq. (2.32) but it does hold
for its adjoint equation.
With respect to the results of Azbelev and Caluk, a 2-2 adjacent
pair of zeros of a third-order equation (Es) does not imply such a pair
for the adjoint equation (Es+). Azbelev and Caljuk denoted
%4 = maxP12(4r&)1
and Dolan (24) introduced the number
(2.33) ZW = maxb12(4d41~
Let Q+, rt , z: , r+, Rf, and Z+ be the respective numbers for the
adjoint equation (Es+) corresponding to the previously defined numbers
for (Es). There are some immediate observations.
Theorem 2.6. (a) rij(a) = r;(a) for i + j = 3,

(b) ~,(a> = x+(a) = r(u) = r+(a),
and
(c) vi(t) is an increasing function of t.
Also, by definition
Since rii(a) = inf(,+(t), t > u} for i + J’ = 3 then for each E > 0

there exists a nontrivial solution y.(x) which has an (i, j)-adjacent pair of
zeros on [a, rii(a) + E) and which is positive between these zeros. Let
rij(~) < GO, E > 0 and 7, E [u, ~ij(a) + e) such that xij(Tc, Y,~(u) + 6).
Furthermore, let us normalize ye(x) so that
Now there exists a sequence {en} of positive numbers approaching zero

such that for i = 0, 1 and 2 the sequence (~2) and {T$} converge, say, to
ci and 7ii , respectively. Let
then {yJx)} converges uniformly to Y(X) on [a, r&a) + or]. Hence

Y(r,J = Y’(rij) = 0 and there exists 7tj = ~~~(a) E [a, rij(u)) such that
Y(T,~) = 0, i.e.,
(2.34) r&z) = z+(Tij).
Lemma 2.12. For i + j = 3, there exists Tij E [a, yii(u)) such that
r&Z) = Xij(Tij) and Tij(U) # Z+(t)fOr t > 7~ .
Of course, we know that if ~~(a) = rij(a) then ~<~(a) = a.

On the other hand, suppose that
We note an improved form of the Leighton-Nehari Fundamental

Lemma 1.5.
Lemma 2.13. If U(X) and v ( x ) are dz~erentiable functions on [a, b],

a < b < co, such that
V(X) # 0 on (a, b) and (@~)(a+) = 0 = (u/v)(b-)
then there is a number X and a number 4 E (a, b) such that u(x) + k(x)
has a double zero at x = .$.
Therefore, letting U(X) = uz(x, u) and v(x) = uz(x, x12) we see that the
latter has at least one zero c on (Q, ~3. Furthermore,
By an argument of the type used for Lemma 2.12, or by recalling that

solutions of (Ea) are continuous functions of initial conditions, we have:
Lemma 2.14. If X E (a, OI) and b E (a, X) then {uz(x, b - c))

approachesuz(x, b) uniformly on [a, x].
Hence, there is a positive number Esuch that u2(x, z&u) - 6) has a zero
446 JOHN H. BARRETT
o* (a, da> - E) and, of course, a double zero at al,(a) - E > a, i.e.,
a < ~I&) < %,(a>*
Putting this together with Lemma 2.12 we see that r12(u) = z12(-r12),
712 E (a, ~~~(a)), and u,(x, r12) > 0 on (712 , r12). By Lemma 2.14,
Y r 12 ) # 0 contradicts
4%dT12 the definition of rrz = rra(u). Conse-
quently,
and
for some positive constant R. Furthermore, u~+(x, or,) > 0 on (7r2 , rIz)
So that vr(Tr2) = r&U) = +(Tr2).
Theorem 2.7. If xzl(u) < zlz(u) < CO then r,,(a) E (.+(a), .+(a))
and there is a number T12(u) E (a, r,,(a)) such that ~~~(a), r12(u) form a
(2, 2)-adjacent pair of zeros for (E3).
For the other case,
(2.36) 71(a) = da) < %1(4 < a),
we recall that zij = a,: and rij = Y,: and, hence, Theorem 2.7 applies to
the adjoint equation (E3+). Consequently,
R+(a) = r&(a) = y&(a) < zq&7)

and, hence,
R(a) = YZl(4 < %W
Becauseof Example 2.1, it does not follow that R(a) = zzz(a) and we now
summarize the valid portions of the conclusions of Azbelev and Caljuk.
Theorem 2.8. Ifs, < R(u) < CO then there is a number T E (a, R(a))
suchthat
w = 44
and 7, R(u) form a 2-2 adjacent pair of zeros of
vu if 71(a)
= %1(4 (e.g., case2.35),
01
(Es+), ;f 44 = %2(4 (e.g., case 2.36).
We can draw a few more conclusions about the pair T, R(a). Returning
to the case (2.35), let R = R(a) and T be as given by Theorem 2.8,
c E (T, R) and define the solution
(2.37) W’(X) = %(X, c) ~,~,(X, T) - +I T) ~,u,(x, c)
of the adjoint equation (Es+). Then
W+(T) = W+(C) = W+(R) = 0
and, furthermore, these zeros are simple zeros. Finally, we note that
Corollary 2.8.1. If ~~(a) = zzl(u) < r12(u) = R(a) < CO and 7 is

the number guaranteed by Theorem 2.8 then T, R(a) form both a 2-l and
a l-2 adjacent pair of zeros of the adjoint equation (Es+) and, furthermore,
if c E (T, R(a)) then 7, c and R(a) are simple zeros of a solution of (Es+).
2.8. Subsequent Conjugate Points

For a third-order equation, having a nontrivial solution with v + 2
zeros, Hanan (45) defined the vth conjugate point.
Definition 2.4. If v is a positive integer and there is a nontrivial

solution y(x) of (E3) with v + 2 zeros, a = x1 < xB < --- < x,+~, then
If no solution has v + 2 zeros, let q”(u) = 00.
For v = 1 this definition differs from q,(a) in Definition 2.3 by requiring

that x = a is the first zero, i.e., y(a) = 0, but, it is readily seen that the
two definitions are equivalent. For v > 1 it is a different matter as noted
by Example 2.1, except when R(a) = cc for which Hanan established
the equivalence.
Suppose that n > 3 and (EJ has a nontrivial solution y(x) with n zeros,
a = x1 < x2 < x3 < *.* < x, < 00. For n = 3 we have already seen
the possible nontrivial solutions of (EJ which have three zeros on
607/3/4-3
448 JOHN H. BARRETT
[a, vr]. Note that in each such case one of the solutions has a double
zero, although, as in Example 2.1, another solution may have only
simple zeros.
For general Y, where ~(a) < co, Hanan established the existence of
a nontrivial solution of the classical equation (2.8), having v + 2 zeros on
[a, ~~(a)]. Suppose this is not so and let {x~~‘,} be a decreasing sequence of
numbers approaching Q(U) such that for each n, a$; is the (V + 2)th
zero on [a, co) of the normalized solution
(2.38) y(x) = c1u&, a> + q&, a), Cl2 + c22 = 1,
which have v + 2 zeros on [a, co). Therefore, there is a solution
Y(x) = C&x, a) + C,u,(x, 4

which is the uniform limit of a subsequence of solutions from the set
(2.42). Thus Y(X) has v + 2 zeros,
X~~X2~X~~~~~~X,+~E[a,oo)
and x,+~ = v”(u). Such a solution Y(x) is called an extremal solution.
Definition 2.5. If Q(U) < co then any nontrivial solution of (Es)

which has v + 2 zeros on [a, Q(U)] is called un extremul solution for y,(u).
Therefore the minimum in the definition of (u) can be replaced by the
minimum value of x,+~ . Note that if y(x) is an extremal solution for
s(u) then
Y(4 = 0 = YhW
Suppose that Y(x) is an extremal solution of (Es) for Q(U) < co and
that all n + 2 zeros xi of Y(X) on [a, ~~(a)] are simple, i.e., Y’(x,) # 0 and
a = x1 < 34 < *.- < x,+1 < x,+2 = 7&) = 7jJ

” *
Define a set of solutions of (I$) for E > 0 by
fY44 = 0 = YX?” - 61,

!Qy&) = qY(a) # 0.
Then for sufficiently small E > 0,
(2.39)
if
Consequently,
YE@>- w, uniformly on [a, r]J, as<-+0
and, since the IZ zeros of Y(x) on (a, 7”) are simple, there is a number
E > 0 such that ye(x) has la zeros on (a, 7” - l ). But this gives a total of
IZ + 2 zeros on [a, 7” - F], which contradicts the definition of 7” = ~“(a),
so one of the zeros of Y(x) is not simple. A similar argument can be
given if ~+(a, y”(a)) = u~+(~,(u), a) # 0. In this way Dolan (24) proved
a corrected version of Hanan’s Fundamental Lemma (45).
Theorem 2.9. If v”(u) < COand either
then for every extremal solution Y(x) of (ES) for q”(a) there is a number
T E [a, q”(a)] and a nonzero number k such that
Y(x) = f%(x, 4, x E [a, 00).
Note that for v = 1, T = a, or ~“(a). In fact, Hanan showed that a double

zero of an extremal solution occurs at x = a or 7”(u) if A(a) = 00.
His arguments also hold on [a, R(u)), which establishes the following
characterization.
Theorem 2.10.If ~“(a) < R(a) < co u&R(u) = ~~~(u){r~~(u)} then

u&, 4{u& S”(4)) is an extremul solution of (EJ for q”(a), is a simple
zero of u2(x, a), and T”(U) is a simple zero of u2(x, a){u,+(x, u)}. Furthermore,
(2.41) 7Yk4 = 7”‘W
We note that, in Example 2.1, Eq. (2.32) has the property that
R(a) = i7/2 = Q(O) = Q’(O) = Q+(o) < 72(O)= 57
and, hence, (2.41) does not always hold outside the interval [a, R(a)).
Another case where (2.41) obviously holds is that of self-adjoint
equations. Note that the hypothesis (2.40) eliminates self-adjoint
equations from consideration. But this caseis not difficult (24).
450 JOHN H. BARRETT
Theorem 2.11. If (EJ is self-adjoint on [a, CO) and q.(a) <c<j ,

then uz(x, u) is an extremul solution for Q(U) having double zeros at each of
the conjugate points ~~(a),..., qy( a ) an d zs
’ p oszt’ zve elsewhere on (a, q”(u)].
Dolan (24) also showed that {Q(U)) is a nondecreasing sequence of

numbers such that if ~y+z(u) < COthen either
(i> %+2(4 > %+1(U) 3 77”(4

or
A constructive characterization of all conjugate points y,(a) for v > 1

and for the general equation (I&) is still an open question.
2.9. Oscillation of (EJ and its Adjoint
Another basic question, which remains unanswered except for a few

special cases,is:
If (E3) is oscillatory on [a, a~) then is its udjoint equation (E3+)
also oscillatory ?
Hanan (45) and Svec (102) h ave answered this question in the affirmative
for (E3) in Hanan’s classesC, or C,r . But the answer is not known in the
general situation.
Dolan (24) h as given the following partial answers.
Lemma 2.15. If (EJ is oscillatory on [a, co) then
d4 < a3 for v = 1, 2, 3,....
Lemma 2.16. If q,(u) < co for each positive integer v then
Tj”(U)+ co, us v + co.
Theorem 2.12. If q”(u) ( co f or each positive integer v then either

(ES) or (E3+) is oscillatory on [a, 00).
Corollary 2.12.1. If (ES) zs

. nonoscillatory on [a, CQ) then either
(i) at most uJinite number of conjugate points Q(U) < co exist or

(ii) (ES+) is oscillatory.
Theorem 2.13. If (E3) is oscillatory and (ES+) is nonoscillatory on

[a, 00) then every solution of (E3) is oscilhztory on [a, a).
Sansone (92) has given examples of equations (ES) for which every
solution is oscillatory on [a, 00).
Corollary 2.13.1. If (ES) is oscillatory but has a nonoscillatory solution

on [a, 00) then (ES+) is oscillatory.
Note that in Hanan’s classes C, and C,, , either (ES) or (ES+) has a
nonoscillatory solution.
Theorem 2.14. If (ES) has an oscillatory nontrivial solution y(x) on

[a, co) such that y(a) = 0, (ES+) is nonoscillatory on [a, oo), and h is the
largest zero of uz+(x, a), then every solution y(x) of (ES) for which y(a) = 0
satisfies the second-order equation
(2.42)
c+yr’+&1 2
[D,+z + q1r24y = 0 on (4 m),
where x = us+(x, a).
Corollary 2.14.1. If (EJ is oscillatory on [a, co), r2q1E C’[a, 00) and
(r,q,)’ has constant sign on [a, co), then (ES+) is oscillatory.
Theorem 2.15. If (ES) and (ES+) are both nonoscillatory on [a, co)
then they are both disconjugatefor large x on [a, co).
I I. B. Nonnegative Coefficients
2.10. The Classical Equation

Here we will be concerned with
w f&J] = Y’n + P(x) y’ + Q(X)Y = 0; P + Q E CL a),

whose coefficients satisfy
v-4) P(4 3 0 and Q(x) 2 0 but P(x) + Q(x) + 0

on any subinterval of [u, co).
These are recent results of the author (14).

452 JOHN H. BARRETT
We recall from Section I that second-order equations with non-

negative coefficients have special properties, e.g., if P(x) 3 0 but +O
for large x, and the equation
w &[y] = y” + P(x) = 0; P E C[u, co)
is disconjugate on [a, co) then

(i) there is a positive solution v(x) of (8,) on [a, 61 for each b E (a, CO),
(ii) any nontrivial solution y(x) of (6,) for which y(a) = 0 satisfies
Y’(X) i; 0 on [a, a),

and
(iii) irn P<l/(x-a) for x E (a, co).

z
Hanan (4.5), Lazer (66), and Gregus (44,4.5) have recently studied
oscillatory properties for various cases of (GJ. Waltman (10.5) and
Heidel(49) have replaced the last term of (6s) with the possibly nonlinear
term Q(x)yy. Most of these discussions require one or both of the
additional assumptions:
WJ C,[y] = y” + P(x)y = 0 disconjugate on [a, co),
P3) PE C’[a, a) and 2Q-P’>O but $0

on any subinterval of [a, co).
Under the assumption (H,), the positive solution V(X), asserted in (i),
can be used to factor the first two terms of (E3) which then becomes
A [vz ($1’1’ + Qy = 0 on [a, b].
By means of the change of variables
z
(2.44) t=
s n v, l’(t) = Y(X)>
Eq. (EJ becomes
(2.45) (R(t)P) * + T(t)Y = 0; l?(t) = 73(x), T(t) = Q(x),

thus eliminating the middle term without changing signs of the coeffi-
cients. Therefore, as Hanan(45) noted, Eq. (2.45) is in class C, , i.e.,
Y,,(U) = co.
Also, Hanan showed that the hypothesis (Ha) restricts Eq. (Ea) to
class C, . Another way of showing this property will lead to some other
results. Rewrite Eq. (G3) as
(2.46) C,[yl = f + PY’ + (p’/Q + (Q - P’/2)y = 0
and note that the first three terms form a self-adjoint operator,
(2.47) c,*rJq = yw + Py’ + (p'/2)y = [y" + (W)yl' + (W)y'.
Therefore, we can express (Ga)in the form
(2.48) UYI = &*[Yl + M4Y = 0, m E 0, 00)s

which Gregus studied extensively (30-42). For any nontrivial solution
y(x) of (Q, yC,[y] = 0, which yields
(2.49) {y[y” + (P/2)yl - Y’W = -(Q - P’/~)Y’.
Hence, the bracketed quantity is decreasing so that again (6,) E Cr.

As we pointed out in Subsection 2.11, Hanan has shown that
Theorem 2.16. If R(a) = co and (ES) is nonoscillatory on [a, 00)

then (E,) is disconjugatefor large x.
If R(a) = co is replaced by (Hi) we have not been able to answer this

question but have been able to prove many of the other results of
Hanan, Lazer and Waltman for (&a) and show that neither (H,) nor
(H3) is necessary. We will be primarily concerned with necessary con-
ditions for disconjugacy of (Ga).
2.11. The Lagrange Identity and the Adjoint Equation

The Lagrange Identity for f,[y] is
(2.50) uC,[v] = (u; a}’ - “c,+[u],
where
(2.51) (21,w} = uvn - 24’21’

+ dBDzu,tclzy = l,[y] = y” + Py,
454 JOHN H. BARRETT
and the adjoint operator is
G+rYl = w?Y>’ - QY*
Note that if P E C’[a, co) then the adjoint operator can be written in the
more familiar forms
&‘[y] = y”’ + (Py)’ - Qy = yN’ + Py’ - (Q - P’)y.
The quantity yC,+[y] = 0 may be integrated by parts to obtain
Lemma 2.17. If y( x ) is a solution of the adjoint equation
(fh+) &‘[yl = [f + PYI’ - py = 0; P + QE c[a, a>,
then for x E [a, CO),
(2.52) Y%Y - ~‘“/21: = ,I QY” + j; PYY’.
If we can jind a solution y(x) of (G3+) whose coe$cients satisfy (H,), for
which
(2.53) Y(4 > 0, Y’(“) > 0 and t&y(x) < 0 on (a, 35
then (2.52) implies the inequalities
(2.54) ‘:Qyz < co, jx E’yy’ < co.

J n
2.12. Properties of Principal Solutions
Assume that equation (G3), whose coefficients satisfying (H,), is

disconjugate on [a, CO),i.e.,
(2.55) u&, a) > 0 and u2+(x, a) > 0 on [a, co).
Consider quotients of ui = ui(x, a) and ZQ+= ZQ+(X,a) and their

derivatives, using Qy = y” + Py in place of y” for ui+,
(2.56) A, = “&, , A, = u1’/u2” A, = qu; ,

(2.56+)
x0+ = u1+/u2+, /I,+ = tq’/lq’, A,+ = 9>,u,+p),u2+; qy = y” + Py.
The identity (2.13) applied to (&a) gives

+ = UlU2’ - U2UlJ’
112 u 2 = u 1+q - u2+lq,
u;’ = up; - up,” ) % ’ = u1+LD2u2+- u2+9p1+,
(2.57)
Il-)p2+ = ul’u[l - u2’zq , tD2u2 = upq” - q’u:“,
I
212 = up2u2+ - u,+‘tB2u,+.
The derivatives of the quotients (2.56) and (2.56’) are

(2.56’)
A,’ = -u2+ju22, A,’ = -LB>,u,‘~(u,‘)“, A,’ = -(PfJ&u,+ + Qu,+)~(u,“)~,
(2.56+‘) A;’ = -u2/(u2+)2, AZ’ = -%&/(u;‘)“, hi’ = Qu,‘/(Tp,+)*.
and the differences are
(2.56-)
A* - A, = U2f/U2U2’, A, - A, = tD2u,+lu2’u,” , A, - A, = ugu2uz” ,
(2.56*)
A,+ - Al+ = u2/u2i-u~~, Al+ - A2’ = up4p2u2+, AD-+- A2+ = u2’lu2+~2u2+.
Observe that if, in addition to (2.55),

(2.58) u21 > 0, 24;’ > 0, u; > 0 and B2u2+ > 0 on (a, co),
then
(2.59)
A,’ > A,* > A,+ > 0, x0” < 0, A,+’ <o and h2+’> 0 on (n? co).
Hence, there exists a positive constant h such that
h+(x) > h,+(x) > X > h,+(x) on (a, co)
or, in other words, the solution
(2.60) Y(x) = ul+(x, u) - Au,+@, a)
of the adjoint equation (&a+) satisfies the inequalities

(2.61) Y > 0, Y’ > 0 and t&Y = Y” + PY -c 0 on (n, cc).
Consequendy, (2.54) holds for y = Y.

456 JOHN H. BARRETT
We will now show that (2.58) follows from (2.55) and hypothesis (H,).
Similar results for special cases have been obtained by Waltman (10.5)
and Lazer (66).
For the special cases where P = 0 or Q = 0 it is a simple matter to
show that (2.55) also implies that
u,W(x,a) > 0 on [a, co).
But the general case is not so easy and will now be dealt with.
Suppose there exists a zero of ui(x, u) on (a, a) and let
(2.62) qpL1 >a) = 0, d&x, u) > 0 on [a, pJ.

By Lemma 2.24, ua’(x, u) has a zero on (pi , GO) and let
uz’(& , 4 = 0, u2’(x, a) > 0 on (4 5,).
Since UT = -Pu,' - Qua < 0, as long as ua’ > 0 and since P+ Q+ 0

on any subinterval, then u,“(x, u) has a second zero, i.e.,
However, if ua’(x, a) has a second zero ,$athen
Ue’(E2
,a> = 0, Q’(x,a) < 0 on (6 , L), 6 E(k, a),
and the first identity of (2.57) yields ~~‘(5~ , u) < 0 so that
4(x) - + m as x- t2 on (5; , &J.
But this contradicts
Lemma 2.18. Under hypotheses (H,), (2.57) and (2.62),

(a) ui(x, a) has a secondzero pa E (pr , co) and
(b) u2’(x, u) has only one zero 5, on (a, co), 5, E (pl , pJ and
u,‘(x, a) < 0 on (c-1, co).
The hypothesis (H,) implies that ui(x, a) > 0 immediately to the right
of x = pa . If there is a third zero pa ,
then the third identity of (2.57) implies that z& , a) > 0 and
AZ(x) --, + a3
But this contradicts
A,’ = -[P+,+ + Q~z+l/($)z< 0 on (p2, p3).

Furthermore, the Lagrange identity (2.51) yields
U2+fU2r= u3+u; + U,tD),U,f.
Lemma 2.19. (a) If u”(x, u) > 0 on (a, XI) then z+‘(x, a) > 0 and
$(X, a) > 0 on [a, co).
(b) If uz+(x, 4 > 0 on (a, co) then tD2uz+(x, a) > 1 > 0 on (a, co).
Next suppose that z+‘(x, a) > 0 on (a, a~) ; then z = z+’ is a positive
solution of a second-order equation with nonnegative coefficients,
zw+ [P + Quz/u2']z= 0 on (a, co),

and
z(a) = 0, Z’(U) = 1.
As we noted in Section 2.12 (i),
z’(x) = uI(x, a) > 0 on [a, co)
which proves the following.
Lemma 2.20. Ifu2'(x, a) > 0 and (H,) isassumedthen

z&Y,u) > 0 on [a, 00).
Conversely, suppose that ul’(x, a) has a zero on (a, co) and let
uz’ytl ) a) = 0, u$‘(x, u) > 0 on (a, tr).
From (2.55) and (2.57) it follows that u:‘(tr , a) < 0 and, hence,
X,‘(x) + -co, as x -+ t, on (a, h).
On the other hand, if u~(x, a) > 0 on (a, co) then
h,+(x) > X,+(x) = fi),u,+(x, u)/a),u,f(x, u) l=- -co on [a, cc).

458 JOHN H. BARRETT
Lemma 2.21. If, in addition to (H,) and (G3)disconjugate, it is assumed

that z/(x, a) > 0 and uz+(x, a) > 0 then
u,f’(x, u) > 0 on [a, co).
Lemma 2.22. Under hypotheses(H,), (2.57), and (2.62),
(4 4x, 4 > 0 on (h , a>,

(b) 4’@, a> < 0 on h, ~0).
We have already noted that EDaua+> 1 on [a, co). Therefore,
and Jr P = co so that the Leighton-Wintner oscillation theorem,

applied to C,[y] = y” + Py = 0, yields a contradiction.
Lemma 2.23. Let (GJ be disconjugate and its coeficients satisfy (H,).
In order for u~“(x, a) to have a zero on (a, CXI)it is necessaryand sufficient
that
Jn P = 00.
As a result of Lemmas 2.25, 2.26, and 2.27, the quotients X,(x), h,(x),
and As(x) are eventually decreasing. Therefore there is a number 01and
a number c E (pa , KI) such that
hi(X) < a for i= 1,2,3 and x E (c, CD).
Consequently, the solution z(x) = M&X, a) - ur(x, a) of (Ga) satisfies
z > 0, 2’ > 0 and z” > 0 on (c, CO).
But, as Lazer pointed out, these conditions provide w = z’ a positive

solution of
W” + [P@) + Q(x) +)/z’(x)Iw = 0 on (c, ~0)
which is disconjugate on (c, co); this contradicts Lemma 2.27 by the

Leighton-Wintner Oscillation Theorem.
Therefore, all of the inequalities of (2.58) follow from (2.57).
Theorem 2.17. If (G,) is disconjugate and its coeficients satisfy (H,)

then there is a solution
Y(x) = 2$+(x, a) - hu,+(x, a), h > 0,
of (ES+) such that Y >‘O, Y’ > 0, ‘B,Y < 0 on (a, ~0) and
(2.63)
I aaQYe < co, Jrn PYY’ < co.
a
2.13. Necessary Conditions for Disconjugacy

The results of the preceeding section are necessary conditions for
disconjugacy of (G3)under hypothesis (H,) on its coefficients. We continue
to make these assumptions and Theorem 2.17 has some immediate
corollaries.
Note that y = Y(x) is a positive solution of
and, consequently,
(2.64) C,[y] = y” + Py = 0 is disconjugateon [a, co).
Therefore, by Hille’s Theorem [i.e., (iii) of Subsection 2.101,
(2.65) mP < l/(x - a) on (a, co).

iz
It is easy to see that
Lemma 2.24. If Eq. (E,), under hypothesis (H,), is disconjugate on

[a, co) then
lkli ‘&Y(x) = 0, Jm(QY) = X

a
and
(2.66) %&Y(x) = Y”(x) + P(x) Y(x) = -sm QY on (a, co).

5
460 JOHN H. BARRETT
Furthermore, (2.63) implies that Jz Q < co and since Y(X) is increasing
Lemma 2.25. If Ep. (G3), under hypothesis (Hi), is disconjugate on

[a, co) then Jc Q < 00 and the second-order equation
(2.67) yn + [p(s) + j,Q] y = 0
is disconjugate on [a, co). Furthermore,
(2.67’) 1 P + (j,)” Q<l/(X-u) on (a,co).
Note that
(jp? = j; (t - x)Q(t) dt < co, x E (a, co)
if and only if
z
(2.68) Q(t) dt < 00.
s”
One of Hanan’s results follows immediately.
Lemma 2.26. If (8,) is d&conjugate, Q(x) 2 0, P(x) E C’[a, 00) and

2Q(x) - P’(x) > 0 on [a, co), then
s‘mt[2o(t) - P’(t)] dt < 00.

u,
Recently, Lazer (66) (in the course of proving his main oscillation
theorem) has shown, under hypotheses (Hi) and (Ha), that if (Ea) is
disconjugate then so is the second-order equation
(2.67*) Y” + [P(x) + m&(x)]y = 0
for each number m < l/2.

In a private communication, J. H. E. Cohn has provided examples
showing that oscillation of (2.67) or (2.67*) does not imply oscillation of
the other.
Now recall that Y’ is positive and decreasing and suppose that
(2.69) Y’(x) > K > 0, i.e., Y(x) > K(x - a) on (a, cc).
Then (2.63) implies
m
(2.70) m PQ(t) dt < co and tP(t) dt < co.
sa I a
But, according to a special case of the nth-order asymptotic results of

Hallam (M),
m
(2.70’) m t2 1Q(t)1 dt < co and t j P(t)1 dt < GO
In sa
implies disconjugacy of (Es) f or 1ar g e x and, hence, nonoscillation of
(‘3krntk o$er hand , if either
(2.71) ,m t”Q(t) dt = 00 or m tP(t) dt = 00,

sa sa
then Y’(x) decreases to zero as x -+ CC and integration of (2.66) yields
(2.72) Y’(4 = (jJz (QY) + /I (PY) on [a, 00).

5
Using, again, the fact that Y’(x) is decreasing,
(2.73) Y’(x) 3 H(x) Y(x); H(x) = (ly)2 Q + j-(0 P.

1: r
Note that (2.67’) of Lemma 2.25 implies that
4
H(x) < l/(x - a) on (a, co).
Also, (2.71) implies that
mH=co.
sa
462 JOHN H. BARRETT
The differential inequality (2.73) may be integrated to yield
(2.74) Y(x) 3 Y(b) exp (Jr H> , a < b < x < co,
which may be substituted into (2.63).
Theorem 2.18. If Eq. (G3), with coefficients satisfying (H,), is discon-

jugate then
(2.75)
/‘Q(t) exp (2 1’ Hi dt < co and jr P(t) H(t) exp (2 it H) dt < CO,

CL u (1 <I
where
H(x) = (jmj2Q + J‘= P = jz [(t - x)Q(t) + P(t)] dt.

z x z
2.14. Nonoscillation Theorems

Here we reprove an important nonoscillation theorem of Lazer.
For simplicity let us first consider the special caseof (Ga)with P(X) = 0,
(2.76)
T.lOIYl = Y”’ + Q(+ = 0, Q(x) 2 0, QEC[a,~)+O< jm
n
Q< m
Note that Eq. (2.76) E C, ; i.e.,
%2(x,b) > 0 for a<x<b<co.
In fact, (2.76) h as stronger properties, e.g.,
(2.77) uz’(x, b) < 0 for a,(x<b<co.
Suppose that (2.76) IS. oscillatory on [a, co) and y(x) is any nontrivial
solution so that for some c E [a, OO),y(a) = 0. Since (2.76) E C, and is
oscillatory then ua(x, c) is oscillatory. But y(x) and uZ(x, a) are both
solutions of the nonsingular second-order equation *
{$+;y} = 0 on (a, co),
where u2+ = ZQ+(X,a) > 0 on (a, co), and, hence, y(x) is oscillatory.
Since this conclusion does not depend on the particular equation (2.76)
we have a theorem first proved by Hanan.
Lemma 2.27. If (Q E C, and is oscillatory on [a, co) then any

solution which vanishes onceis also oscillatory.
Next suppose, not only that (2.76) is oscillatory, but that there is a
nonoscillatory solution U(X). Then by Lemma 2.30, U(X) # 0 on [a, co)
Let
u(x) > 0 on [a, co).
Since, u.Jx, a) is oscillatory let b be its first zero, i.e.,
uz(h 4 = 0, 4x, 4 > 0 on (a, V, b E (a, co).
Then by Lemma 2.13 there is a number h such that ua(x, u) - k(x)

has a double zero at some x = (y.E (a, b). Hence,
Au(x) - u&x, a) = Ku&, a); x > 0, k > 0.
Therefore, Au’(x) = ua’(x, CZ)+ kua’(x, CY.)

and from (2.77),
u’(u) = (k/h) u2’(u, a) < 0.
Lemma 2.28. If (G3)E C, and is oscillatory but has a nonoscillatory

solution u(x) on [a, 00) then U(X) # 0 on [a, 00). Furthermore, if (Q
satisjes the condition (2.77), then 1u(x)\’ < 0 on [a, co).
Suppose that y(z) is an oscillatory solution of (2.76); then z = (y/u)’
is an oscillatory solution of the second-order equation
(2.79) Z” + (3u’/u)z’ + (3u”/u)x = 0
and w = u3k is an oscillatory solution of
(2.80) w* + ($)[u”/u - ($)(u’/u)~]w = 0 on [a, co).
Since the nonoscillatory solution U(X) satisfies

a
U(X) > 0, u’(x) < 0 and u”(x) > 0,
it is not difficult to show that u(x) is bounded and
(2.81) g+$ u’(x) = li+i u”(x) = 0.
w/3/4-4
464 JOHN H. BARRETT
Furthermore, by multiplying u”’ + Pu = 0 by U(X) and integrating

we have
(2.82) { ud - 242)’ = -Qu” < 0
and by (2.81),
(2.83) C24~”- u’~/~}(x) = j,” Qu” > 0 on [a, co).
If, in addition, jr Q < co, then
(2.84) 1IdUn - u’VH-4 < ~“(-4 ,I Q
and comparison with (2.80) implies oscillation of
(2.85) y” + ($) (Jm Q) y = 0 on [a, 00).

z
Theorem 2.19. lf(2.85) is nonosciZZatory on [a, co) then y”’ + Qy = 0,
Q > 0, is nonoscillatory on [a, CO).
We will now extend the preceeding discussion to Lazer’s case of (G,),
whose coefficients satisfy hypotheses (Hi) and (Ha). Proceeding as
before, if y(x) is an oscillatory solution and U(X) is a nonzero solution of
(Ea) then eu = ~“‘“(y/u) ’ is an oscillatory solution of
(2.86) W” + {P + (3/2)[u”/u - ($)(u’/u)~]}w = 0.
Because of (H3), Eq. (g,) can be rewritten as
(63’) [Y” + U’P)yl + @WY + (8 - P’P1y = 0,
which is the form studied by Gregus (3442). Then for any solution
Y(X) of (h),
(2.87) (yQy - y’2/2}’ = -(Q - P’/2)y2 < 0; B,Y = yn + (W)y.
Note that (2.87) yields immediately that Eq. (G,) E C, ; i.e.,
u2(“Y t> > 0, a,(x<t<co.
For third-order equations in C, , Gregus (40,41) and Svec (102)

have constructed nonoscillatory solutions with properties needed here.
For each positive integer n > a let JJJX) be the solution of (Es) such that
which is normalized so that
Therefore, there exists a sequence {ni} of positive integers -+ co such that

{c;i} converges for j = 0, 1, 2, say,
{c;+ +c.. 3 , co2+ Cl2+ c22= 1.
Hence the sequence {m,(x)} converges to the nontrivial solution
Y(X) = c~z&, 0) + cluI(x, a> + c~~~(x,4 on [a, a)
and uniformly on closed finite subintervals [a, X], a < X < KI.
From (2.87) it follows that
Let a < X -=c 00. Then, for ni > X and x E [a, X],
from which we obtain
(Y9J - Y’2/2}(x) 3 jm (Q - P/2) Y2.

L
Now that the right-hand integral exists it is easy to complete the proof of
the following result due to Gregus (42).
Lemma 2.29. If the coeficients of (G,) satisfy (Ha) then there exists
a positive solution Y(x) on [a, co) such that
(2.88) {Y&Y - Y’2/2}(x) = Im (Q - P/2) 1-2 > 0 on [a, co).

2
466 JOHN H. BARRETT
If in addition Y’(x) < 0 and Jr Q < 00, then
(2.89) {Y‘B),Y - Y”/2)(x) < Y2(x) j=@ - P’/2) < y2(X) [p(x):2 + j’$?]
a z
and comparison with (2.86), where u = Y, yields that
(2.90) Y" + [p(x) + P/2) j: Q] Y = 0
is oscillatory on [a, Co).

Suppose that (Ga) is oscillatory but that (2.90) is nonoscillatory on
[a, 00); then
w I! 20 and y” + Py = 0 is nonoscillatory on [a, co).
Hanan (45) asserted that, under these assumptions, any nonoscillatory

solution is decreasing in absolute value. However, his proof appears to be
incorrect.
As we have already noted, (H4) implies that y” + Py = 0 is discon-
jugate for large x, i.e., there is a number c E [a, a) such that y” + Py = 0
is disconjugate on [c, co). Consequently, if d E (c, CD)there is a positive
solution V(X) of y” + Py = 0 on [c, d] and (Ea) can be rewritten in the
two-term form
62, [v”(y’/v)‘]’ + Qvy = 0 on [c, d].
From (2.19) we recall that
&‘(C, a) = -ul+(a, c)
and, since ur+(x, c) satisfies
&+, [(1/v>(v2y’>‘l’
- f&y = 0 on Cc,a>,
it is easy to show that
24,+(x, c) > 0 and U$‘(C, x) < 0 for c<x < co.
Now that property (2.77) is satisfied, we have Lemma 2.28 on [c, co).
Since Y(x) of Lemma 2.29 is a positive nonoscillatory solution then
Y’(x) < 0 on [c, co),

and (2.90) is oscillatory on [c, co), a contradiction, Note that Lazer’s

additional assumption that P( co) = 0 is unnecessary.
Theorem 2.20. If the coejicients of (Es) satisfy (Hi) and (H3),

Jr Q < GO, and the second-order equation
Y”+ PC4+ (W) ,I Q]y = 0

is nonoscillatory on [a, co), then so is the third-order equation (EJ.
2.15. A Third-order Priifer Transformation

Recalling the second-order Priifer Transformation (1.9) of Section I,
let y(x) be any nontrivial solution of the third-order equation (Es) and
define an amplitude function
(2.91) P(X) = dY2@) + (4Y)“@“) + P2Ym) > 0.
Next, normalize the phase components of y(x) by letting
(2.92) 44 = Y@-)/P(4~ 44 = 4Yc4/&)~ s2(4 = ~2YWlP w

Differentiating (2.91) we have
(2.93) P’ = [(l/r1 - 4J ssl + (1 iv2 - 4 v21f

and differentiating (2.92) yields the vector-matrix equation
s ’ 0 b,(x) 0 ’ s
(2.94) s1 = --b,(x) 0 b2W Sl ;
i!$2 i 0 --b,w 0 10 $2
bl = 1/~1-(1/~1-41)~2-(11~2-42)~~2,
b2 = cz2 + (l/r1 - 4J sf2 + (l/r2 - 4J s22-
This differential equation for the direction cosines may be rewritten

by use of the vector cross-product, i.e.,
(2.94’) u’=-pxu=ux/3
where u = (s, si , s2) , /3 = (b, , 0, b,). Note that the tangent vector u’
is orthogonal to the plane of /3 and u and the locus of U(X) is a path on the
468 JOHN H. BARRETT
unit sphere. For the case where b, and b, are linearly dependent the locus
of it is a circle. This interesting equation warrants further study and
should shed further light on third-oder oscillation properties.
By altering (2.92) slightly we let
(2.92’) y = ps, Dly = k,ps, , D,y = klkzpsz ; k, & k, = constants # 0
and obtain
(2.93’) P’ = RkJ~1 - Md ss1 + (k,lrz - &A vnl~~
which immediately yields bounds for solutions of (Ea) (13).
Theorem If there exist nonzero constants k, and k, such that

2.21.
Sa”I h/r, - dk, I < COand JQ”1k,/r, - qJk, 1 < COthen every solution
of (EJ is bounded on I = [a, CO).
Other boundedness theorems for third-order equations have been
reported by Dobrohotova (23) and Rab (81).
2.18. Gregus’ Asymptotic Theorems.

In his numerous papers (34-42) since 1955, Gregus has studied the
equation
(2.95) y”’ + 2Ay’ + (A’ + b)y = 0; A, A’ & b E C(I).
He observed that the classical equation
&[yl =y”I + pzy” +$J,y +Poy = 0; Pi fz CV),
can be transformed into the form (2.95) by means of the substitution
P = Y exp /(l/3) J-liPB(
and the subsequent equation (Ea) rewritten as (&a’), as in the preceding

section. In the form (Ga’), Eq. (2.95) becomes
(2.95’) [Y” + AY]’ + AY’ + by = 0; A & b E C(I),
and the coefficient A(x) need not be differentiable. Note that the first
three terms form a self-adjoint operator. Let zli(x, a) be the principal
solutions of
y” + (4qY = 0
defined by the initial conditions
y = v&, a), Y(U) = 1, y’(u) = 0
y = v,(x, a); Y(4 = 0, y’(u) = 1.
If we rewrite (2.95’) in the vector-matrix form
then a fundamental matrix solution of the homogeneous part is
VI212
Y = 2zot (:;zJf VIV,’ , where vi = q(x) = vi(x, a),
( 2(v,‘)2 2v,‘o,’ (VI’)2 1
and solutions of the complete nonhomogeneous equation are solutions

of the system of integral equations
(2.96)
-
i ’ G(x, t) b(t) r(t) 4
r
92YW = 2Y@;bow12 + 2YW ‘uow VI’W + h.YW[~I’(412
- z t) b(t)y(t)4
Ia K2(x,
where K(x, t) = [Q(X) vi(t) - q(x) v,(t)12/2 = q2(x, t)/2,
K&c, t) = I+, t) = v&, t) V1’(X,t),

and
K2@, t) = [v1’)1I(x,
t112.
Lemma 2.30. (a) If b(x) < 0 on I = [a, a) then u2(x, u) > 0 and
a)&(X, a) > 0 on [a, co).
470 JOHN H. BARRETT
(b) If, in addition, A(x) < 0 on [a, co) then

him 24,(x, a) = ;A~ u2’(.‘c, a) = co
and ‘Jl&x, a) is an increasing function on [a, CO).

Since ‘D>z~z 3 1 b ) r+, + 1 A / z+’ 3 0 on (a, co) then
Lemma 2.31. If, in addition to the hypothesesof Lemma 2.32, either

.n
0)
t2 1b(t)] dt = co
J ,I
01
(ii)
then
Also, the latter limit is assuredif

m
(iii) t2 / A(t)/ dt = co.

I (1
Note that the sufficient condition jz @[A’(t) - b(t)] dt = co of Gregus
(42) is implied by (i) or (ii) but Gregus’ conclusion that ua + 2Au, + CC
is slightly stronger.
Suppose with Gregus that
(2.97) 44 d 0 and b(x) > 0 but +O on any interval of [a, co).
Then, as in the preceeding Subsection, Eq. (2.95) E C, and
(Y" + 4)’ + 4’
is a disconjugate operator and, hence, there exists a solution Y(X) such

that
Y(x) > 0 and Y’(x) < 0 on [a, a3).
Also, the adjoint equation satisfies Lemma 2.32. The Lagrange identity
u,+Lo2Y - Zq’Y + Y92u2+ = Y(u)
then yields an improvement of Gregus’ theorem (42).
Theorem 2.22. Under hypothesis (2.97), there exists a solution Y(x) of

(2.95) such that
Y’(x) < 0 and Y”(x) > 0,
and limz+m Y’(x) = limr+m ‘&Y(x) = 0. If, in addition, either (i) or (ii)
of Lemma 2.33 is assumedthen
!& Y(x) = 0.
III. Fourth-Order Equations
3. I. Examples
Motivating examples, which illustrate the oscillatory behavior with
which we are concerned, are
(c) y’” + y’ = 0, (e) yiv + y” = 0,

(b) y’” + y = 0, (d) y’” - y’ = 0, (f) y’” - y” = 0,
The first two examples (3.la) and (3.lb) are special casesof
(4 (WYT - p(4r = 0,
(3.2)
(b) (+)Y”)” + PWY = 0; Y, p > 0, Y,p E C[a, a),
for which Nehari and Leighton (70) made a rather complete study.
Other authors (9-11,.54, 88-90) h ave extended certain parts of the
theory developed by Nehari and Leighton to
(3.3) [(l-w) + Qy’l’ + PY = 0.
Whyburn (106) and Kondrat’ev (62) h ave also contributed basic discus-
sions of Eqs. (3.2). Kamke (60) has listed other examples and Handelman
472 JOHN H. BARRETT
and Tu (46) studied a phsical application of (3.3), i.e., vibrations of

a beam with a compressive stress distribution.
Somewhat different examples are generated by recalling that the
product of two solutions of a second-order equation is a solution of a
third-order (self-adjoint) equation and by taking the product of three
solutions, which turns out to satisfy a fourth-order equation.
Lemma 3.1. Recall that LJy] = (ry’)’ + qy = 0 and let L,[vJ = 0

for i = 1, 2, 3. Then y = V~V~V~satisjies a fourth-order equation
(3.4) [~(~{~[(~Y’)’ + 3qyl)’ + 4PY’)l + 3qmy + 3qyl = 0.
An interesting special case of (3.4) is that when r = q = 1 and (3.4)

becomes
(3.4’) yiv + 1Oy” + 9y = 0
which has the fundamental set of solutions
y = sin3x, COG+
x, cos2x sin x, sin2x cosx.
An obvious way to generate fourth-order examples is to iterate

operators of lower order, e.g.,
(3.5) QYI = ~*~w2~2(w2~2rwl~l~Y~l~~~
~iCY1 = Y’ + PiYi q >o, wi , Pi E C(4

and
(3.6) UYI = ~2(whcYl);
where
h[Yl = (r,y’)’ + PiYl ri >o; w, t-i ; pi E C(I).
Note that for the operator of (3.5) every nontrivial solution of the
equation L[y] = 0 has at most three zeros in the interval I.
Definition 3.1. A fourth-order operator L[y] and the corresponding

equation L[y] = 0 is said to be disconjugate on an interval I if no nontrivial
solution hasfour zeros on I.
Later on we will show that any disconjugate fourth-order operator
can be factored into the form (3.5). An interesting special case of (3.6)
is the constant coefficient case,
L[y] = (II2 + m2)(D2 + n”)y = yiv + (m2 + n2)yn + m2n2y

[m, 12= arbitrary constants],
which discourages general separation theorem conjectures although such

theorems have been established for such special cases as (3.2) (63, 70).
Although the sum of two solutions of a single second-order equation
yields nothing of interest here, the sum (in fact, any linear combination)
of solutions of different second-order equations satisfies a fourth-order
equation.
Lemma 3.2. If (hi’)’ + Q pi = 0 for i= 1, 2 andQ, #Q, then

y = v1 + v2 satisfies
(3*7) IR
( ( + (‘l iQ2 )Y])‘/’
[(Ry’)‘+(g1~Q2jy]+(Q1;Qzjy=0.
The case Qr = -Qa # 0 and R = 1, where (3.7) becomes
!$ 1’ +Qly = 0,
has already been noted (9) ; and the constant coefficien teases, with
R = 1, QI = Kr , Qa = Ka [& = arbitrary constants], become
The product of solutions of different second-order equations is also

of interest. W. Hahn (43) h as made use of this idea in his study of
orthogonal functions.
Lemma 3.3. For solutions aI and v2 of Lemma 3.2, y = vlv2 satis$es
(3.8)
[ Q2 1, @[(RY’)’ + (Ql + !i?21~1)’ + (QI + Q2) Rq.‘]’ + (Q2 - QJY = 0.

474 JOHN H. BARRETT
Finally, if y(x) is a solution of L3[y] = h where h = 0, {h E C(I) and

h # 0} on 1 and L, is the canonical third-order operator, then y(x)
satisfies the fourth-order equation
&3(Y))= 0 $L3(Y,] = 01.
3.2. A Fourth-Order Canonical Form

The preceding suggest the following general fourth-order equation (12)
v41 UYI = C&Y)’ + a&y + ay = 0,
where D,y = r3L3[y], L, is the third-order canonical operator of

Subsection 2.2, which was defined in terms of the quasi-derivatives
D,y and D,y, and the coefficients satisfy
It is easy to see that (3.2), (3.3), (3.5), and (3.6)-(3.8) are special casesof
(E4). Shinn’s quasi-differential operator (98,99) is also equivalent to
(E4). Furthermore, the classical fourth-order equation
(3.9) 14[yl = Y’” + P3Y” + P,Y” + PlY’ + POY = 0
is equivalent to a special caseof (E4), since it is easily obtained from
(3-l 0) [Y”’expj P3 + (j Pl expj A) Y’] ’
+ (Pzexp
jP3- jAexpjP3)Yr’
+ (Poexp
jP3)Y=a
The corresponding phase-vector form of (E4) is
(3.11)
from which appropriate existence and uniqueness properties are easily

obtained. The preceding formulation is a special case of the nth-order
canonical form given by Zettl(110) w h o used it and the third-order case,

L, , as motivating examples.
The usual well-known methods provide the Lagrange Formula
(3.12) =qyl - YL+kl = {T Y>‘,
where the adjoint operators L4+, III+, D2+, D3f are obtained from L, ,
D, , D, , D, , respectively, by interchanging the coefficients:
rl with y3 and q1 with q3 .
Note that if LJy] = 0 and L3+[x] = 0 then {z; y) is a constant.
3.3. Self-Adjoint Equations

Equation (EJ with
Y1 = Y3 and 41 = 43
represents a large class of self-adjoint equations of order four; i.e.,
v4*1
LI*ru1 = MYdY1Y’) + PIYI)’ + W,Y’lI’ + W,[(~IY’)’ + P,Yl + P4Y = 0,
where Di = Di+ for i = 1, 2, 3.

Note that (3.2)-(3.4) and (3.7) are self-adjoint; (3.5) is self-adjoint if
X, = h, and wr = wa ; (3.6) is self-adjoint if /I, = il, ; and (3.10) is
equivalent to a self-adjoint equation if p, = 0 and p,’ = p, . Also, if
Y = exp(Jp,), and pi E IF-~)(I) for i = 1, 2, 3 then
(3.13) 9-~4lyl = (YY”)” + (ypz - yn)Yw + YPlY’ + YPOY,
and rlJy] is self-adjoint if r = exp(Jp,) satisfies the
(3.14) (T” - p,r)’ + p,r = 0.
Next we note that certain nonself-adjoint third-order equations can be

imbedded in self-adjoint fourth-order equations.
Lemma 3.4. If z(x) is a nonvanishing solution over an interval I of the

476 JOHN H. BARRETT
canonical third-order equation L&J] = 0, then ( l/z)(r1z2La[y])’ is a

self-adjoint fourth-order operator. In particular,
(3.15) (114(%~2k3rYI)’
= {r,PzY)’ - P*(W)’ + (y29*- ~&N DlYj’
+ @,Y + W24Y*
3.4. Factoring Fourth-Order Operators

Here we proceed as we did when factoring the third-order operator L, .
Suppose that the adjoint equation (Ed+) has a nonvanishing solution on an
interval 1, i.e.,
0) L,+[v+] = 0, v+ # 0 on I.
Then, by the Lagrange Formula (3.12),
-uYl = ; h-; Y>’
and {zJ+; JJ} is a third-order operator of the form L,[Y]; i.e.,
(3.16) {v’;y} = r3v+z )2
r,q,v+ + D,+v+
+I Y3V’+2 + + j +$I DlY
Next, assume there are two solutions wr and v2 of (EJ such that
(ii) {v+; Vi} = 0 for i-l,2 on I,

(iii) 7% # 0,
and
(iv> w2 = vlD,v2 - v2D,v, # 0 on I.
By Lemma 2.10 the third-order operator {u+; JJ} can be factored yielding
(3.17)
In addition, suppose that va is another solution of (Et) and consider the

Wronskian of the three solutions vi , u2 , ~a ,
(3.18) w3 = wh 1 212 , 2131 = %2 ,

where
Dp, Divz Div3
cqjk= Djvl Djv, Dp, .
DP~ DA D,v,
Successive differentiations of the determinant a,,rz yield
(3.19) Q+w, = 01013

> D,+w, = %23 9 DA = %23
and, finally, that ws satisfies the adjoint equation (E4+). Furthermore,

it is easy to show that
(3.20) {w3 ; Vi} = 0 for i= 1,2,3.
Consequently, ws may be substituted for v+ in (iii) and (3.17), the latter

becoming the Polya-Mammana factorization and a special case of (3.5).
Note that Lemma 3.4 is suggested by (3.17). By combining pairs of
factors in (3.17) we obtain a special case of (3.6), namely,
(3.21) MY1 = ~(r2~22~2rYl)~
where
+ (‘lQ2+ 2r2q1)
qyl = [Z)’ + [ ~l(~ZW2')'
2r2w22
w2
1Y
and &[y] is obtained from L&J] by replacing ri and q1 by rs and qa,
respectively. In the self-adjoint case, (3.21) becomes
(3.22) ’ h*rYl = J52(r2w2z~2[Yl) = L,+(J52rYl)-
Only the assumption (iv), w2 # 0, is needed for (3.21) and (3.22) to hold.
Since L,[vJ = 0 for i = 1 or 2, the Wronskian wa satisfies
(3.23) &[y] = (g)' + (,,,,,,, qy y = 0.

z a
The special case for Eq. (3.3) with Q = 0,
l2[Yl = (3)’ + (9) y = 0,
has been investigated .previously (9, 70).

478 JOHN H. BARRETT
Finally, under assumptions (iii) and (iv), i.e., disconjugacy,
(3.24) L*rYl = 4+hu4rYl)l

is another factoring of the self-adjoint operator, where
Z,[y] = Y1$(y/v~)’= YlV’iy’ - sy

[ 1 1.
The factored expressions (3.22) and (3.24) are equivalent to those
previously given by Mammana (73) for the special case (3.3).
3.5. First Conjugate Points

In the preceding section conditions (i)-(iv) were given which insure
that L,[y] can be factored into first-order real factors (3.17) over an
interval I and, hence, that (E4) is disconjugate on I. Conversely, if (E4)
has a nontrivial solution with four zeros (counting each multiple zero as
that many zeros) on I then either cot or ws or ws has a zero on I. This
result may be improved by choosing particular solutions of (E4).
Let ui(x, a) for i = 0, 1, 2, 3 be the fundamental set of solutions of
(IXJ) on I = [a, co) as defined by the initial conditions
(3.25) DjU<(U, U) = 6ij ; i,j =O, 1,2,3,
and ui+(x, a) be the corresponding solutions of the adjoint equation

(E4+). Because of these initial conditions at x = a, if or = U&X, a),
z~a= uz(x, a) and ~1s= ur(x, a) in (3.18), then
(3.26) w3 = w3p43 , u2 I %1(X, 4 = 213+(x, a).
Therefore, if a 0, u3+(x, u) > 0 and W,[u, , u,](x, a) > 0 on (a, b),
then (E4) is disconjugate on (a, b).

Suppose a = 0 on (a, 61,
which has u3 and ua as particular solutions. Again, because of the first and
last inequalities of (3.27), Eq. (3.28) is disconjugate, i.e., y(x) has at most
two zeros on (a, b), which determines y’(u) = 0 = Dir(u). But since
u3(x, a) > 0 and w2[u3 , u 2](x, a) > 0 on (a, b) then y(x) cannot have
two zeros on (a, b), so that y(a) = Dir(a) = Dsy(a) = 0. But this
implies the contradictory conclusion that y(x) = J&(X, u).
Theorem 3.1. If the inequalities (3.27) hold on the open interval

(a, b), a < b < co, then the Eq. (E4) is disconjugate on the half-closed
interval [a, b).
On the other hand, if (E4) h as a solution with four zeros on [a, b) then
either ua(x, a) or ua+(x, a) or W,[U, , u.J(x, u) has a zero in (a, b). If
ua(c, u) = 0, a < c < b, then +(x, u) satisfies the two-point boundary
conditions
(3.29) y(a) = D1y(u) = &y(a) = 0 = y(c).
Since us+ = W,[u, , u2 , UJ then ua+(c, a) = 0 yields a nontrivial

solution of (Ed) satisfying
(3.30) y(u) = 0 = y(c) = &y(c) = &y(c).
Similarly, W,[U, , ua](c, a) = 0 y’ieId s a solution of (E4) satisfying
(3.31) y(u) = Qy(u) = 0 = y(c) = 4y(c).
Definition 3.2. For i + j = 4 let +(a) be the smallest number c > a

such that (EJ has a solution satisfying
(3.32) y(u) = Dly(u) = *-- = Di+y(U) = 0 = y(C) = Dly(C) = '.' = Dj-a(C)
[i.e., y(x) has a zero of mu2tipZicity i at x = a and j at x = c = +(a)].

If no suchfinite number c exists let z+(u) = GO.
Note that c = xai(u), x&u), and z&a) satisfies (3.29), (3.30) and (3.31),
respectively.
Definition 3.3. If
(EJ h us a nontrivial solution with at leastfour zeros
on [a, co), the number
(3.33) ql(u) = inf{x, ; x1 < x2 < x3 < x4, y(xi) = 0,y f O,L4bl = 01
is called the first conjugate point of x = a with respect to (E4) in (a, 00). If
(E4) is disconjugate let ql(a) = 00.
f+/3/4-5
480 JOHN H. BARRETT
Coppel(22) h as g iven a more general definition of conjugate points and

has established some comparison theorems for “canonical” systems of
differential equations, but these will not be treated here.
Let c = min{zij(a); i + i = 4); then (E4) is disconjugate on [a, c) and
if c < a3 there is a nontrivial solution of (E4) having four zeros (counting
multiplicities) on [a, c].
Theorem 3.2. d4 = mink3(4, du), ~~~(41 d a.

For the self-adjoint equation (E4*), since z++(x, u) = u3(x, a), then
z13(u) = .+(a) and there is one less alternative.
Corollary 3.2.1. For the self-udjoint equation (Ed*),
It is now clear how the special cases (3.2), treated by Leighton

and Nehari (70), fit into the whole picture. Indeed, they are the
mutually exclusive extreme caseswhere for Eq. 3.2(a): Q(U) = 00 and
z&u) = ~~(a), and for Eq. 3.2(b): .z~~(u)= co and z13(u) = ~~(a) = xQl(u).
They also pointed out in certain cases in which the middle term of
Eq. (3.3) can be removed and we now give a simple explanation of this
procedure.
Suppose that the second-order equation
UYI = CRY’)’ + QY = 0
is disconjugate on [a, b) and a 0 and the change of variable

z
(3.35) t= s z 0 x = or(t)
a
yields the two-term form
(3.34’) @ii).. + 5Y = 0,
where X(t) = (Rz3)[cr(t)], T(t) = P[a(t)] and Y(t) = y[a(t)].

The author (10, II), W. T. Reid (88,89), D. Hinton (.52), and Howard
(54) established conditions for the existence of a finite z2a(a) for the
three-term equation (3.3). However, there has been very little on the
existence of a finite ~(a) beyond the simple case (3.28b) of Leighton
and Nehari. R. W. Hunt (56,57) h as extended much of this fourth-order
theory to two-term equations of higher orders. Most of the discussions
for even orders are direct generalizations of the vector-matrix formulation
which is given in the next section.
IV. Self-Adjoint Fourth-Order Equations
4.1. Vector-Matrix Formulations of Fourth-Order Scalar Equations

With the use of a standard formulation (28), as applied by Sternberg
(100) and the author (IO) to (3.3), the self-adjoint equation
(E4*) U’l = (QY)’ + &ty + w = 0

may be expressed in the vector-matrix system
a’= Aa+BS
(4.1) 1B’ = Cal - A+&
where A+ is the transpose of the matrix A and
O1= (D,y 1’
Y B=
i
--L&Y
D2y
1
7
By means of matrix integrating factors, the system (4.1) is simplified to
8’ = 4%
(4.3) 1,d’ = -Fp,
where
,t?= T-h E = T-lB(T+)-l T’=AT
(4.4) /‘j = T+& ’ F = T+CT ’ p” z -p/J’
Here, it is convenient to take the interval I = [a, CO)and
(4.5) T = (f)ge.
1 -‘l”)
u = (T-‘)t,
482 JOHN H. BARRETT
where u and z, are the fundamental set of solutions of the self-adjoint

second-order equation
(4.6) FIY’)’ + 4lY = 0

with g(a) = 1, D,u(a) = 0 and ~(a) = 0, &v(a) = 1. Therefore,
the coefficients of (4.3) become
Note that det(E,) = det(F,) = det(F,) = 0 and El , Fl , and F, are

symmetric and positive-semidefinite, which we write
4 >O, 4 30, F, > 0.
The special case of Eq. (3.3) w h erer, = 1,~~ = R,q, = O,q, =Q,
q3 = 0, q4 = P and a = 0 yields the matrix coefficients
has already been studied (Z0). R. W. Hunt (56) and R. L. Sternberg (100)
have investgated the higher-order case
(4.9) (Ryy~~) + Py = 0.
W. T. Reid (85-89) h as applied variational methods to the vector-

matrix system (4. l), including complex coefficients (see Subsection 4.4).
Note that E > 0 and if F < 0, then
(4.10) (P+/$ = B+(--F) P + PtE@ 3 0

and hence
(4.11) ,6+,g= a+8 = D,y D,y - yD,y
is nonincreasing on [a, co). Consequently, certain two-point boundary

problems are impossible in this case.
Theorem 4.1. If pa 3 0 but +O on any subinterval of [a, b] and

qz < 0 on [a, b] then no.nontriwial solution of (Ed*) satisfies any of the

following two-point boundary conditions:
(3.31) r(a) = 4y(a) = 0 = Y(b) = DlY(4, “clamped ends”;

(4.12) r(a) = au(a) = 0 = D,y(b) = D,Y(b), right “free end”;
(4.13) r(a) = DzY(@) = 0 = Y(b) = D,Y(@, “supported ends” [22];
(4.14) y(a) = Dzy(a) = 0 = Dly(b) = D&b).
Another way of stating the first conclusion of Theorem 4.1 is that

z&a) = co and one such case (3.2b) of this has been studied by Leighton
and Nehari (70), where they showed that the conjugate points of x = a
for (3.2b) are the zeros of ua(x, a).
In the following sections we will establish conditions on the coefficients
of (Ed) which insure that (3.31) is satisfied nontrivially, i.e., that
x22(4 < a*
Atkinson (2) has studied the boundary problem of (E4*) and (4.13) by
use of the vector-matrix system
and the author (9) has reported on the special case as (4.15) applies to the
two-term equation (3.2a), where he investigated relations between
solutions satisfying (3.31) and those satisfying (4.13). Coppel (21) has
considered more general boundary conditions.
4.2. Subwuronskians
Although the second-order equation (3.23) is sometimes useful (9, 70)
there are other relations involving 2 x 2 subwronskians (i.e., sub-
determinants of the Wronskian wa) whose coefficients are those of the
original differential equation.
For the fundamental solutions uZ(x, a) and ua(x, a) of the self-adjoint
equation (E4*) denote the various subwronskians by
Diuz Diua
for ;,j=O,1,2,3.
% = Diu, Dju3
Note that olij = -Q+ , aii = 0 and {~a ; ua} = 0 is equivalent to
(4.17) cl03 = %2*

484 JOHN H. BARRETT
Successive differentiations yield the following:
Lemma 4.1. (a) C& = c+Jra ;

PI 42 = 2%2/r, - 42ao1 ;
(4 "12' - %3/Yl - Pl~O2 ;
(4 43 = a23lr2 - %1%2 + q4no1 ;
(e> 0123
' = -42%3 + 44ao2 ;
and
(f) ~ol(4 = 0 = ao2(a) = c52W = ~13Wy
43(4 = l/r2(4 Nz3(a) = 1.
An equivalent result to Lemma 4.1 is that CJ= ao1is the unique solution
of the fifth-order self-adjoint equation and initial conditions
&[a] = (tD4u)' + q2Cd)p - q4Ca>p = 0,

(4.18)
( u(u) = YJ&u(u) = tD,u(u) = D,u(u) = 0, tDp(a) = 1,
where ‘J&O = go2, B2a = 201,~, ‘J&a = 01~~ , and ‘D40 = 01~~. Mammana
(73) noted the fifth-order equation of (4.18) for the special case (3.3).
Lemma 4.2. There exists a number 6 > 0 such that
in (a, a + a), i.e., a < zo2(a) < co.

A routine calculation of the Lagrange Formula for LJu] shows that it is
self-adjoint. Next we note a nonlinear identity among the subwronskians
aii , which is easily verified.
Lemma 4.3. 01~~01~~= 0~:~ + 01~~01~~.

The following Riccati-type equations are also useful.
Lemma 4.4. (a) If aI3 # 0 on an interval I then h = -cx~~/Lx~~

satisfies
(b) If 0123 # 0 on I then k = 0113/~23

satisjies
k' = ; + 92k2 - q4 (2," - 2q, ($j ,

(C) If aol # 0 on I then m = a13/0101satisjks
We now summarize the results obtained by the author (12) for the
special case (3.3), i.e., where
r1 = 1, r2 = R, q1 = 0, 42 ==Q, and 44 = p,
by use of Lemma 4.14.4.

Recall that the first zero on (a, co) of a01 is z2a(a) and designate the
first zeros on (a, CO) of the subwronskians 01~~and Celiaas pi(u) and cl(a),
respectively.
Theorem 4.2. For Eq. (3.3) in [a, co):

(a) Ifzzz(4 < 00 then a < k(u) < L&(U) < x22(4.
(b) If P < 0 and the second-order equation (Ry’)’ + Qy = 0 has
a nontrivial solution with three zeros, x1 < x2 < x3, then xS2(u) < xQ .
(c) If P < 0 and the second-order equation (Ry’)’ + Qy = 0 has
a nontrivial solution satisfying y(u) = y’(b) = 0, a 0, but / P 1 + Q + 0 for large x, and
J” (1 /R) = 00 then in order for zz2(u) < 00 it is necessary and su#icient
that pi(u) < co or .$,(a) < co.
4.3. Matrix Riccuti Equations

Here we apply the arguments, which established certain oscillation
results for scalar equations in Section I, to the matrix system
‘Y’ = EP,
(4.19)
t ir’ = -FY,
where E and F are the coefficient matrices (4.7) of the vector-matrix

equation (4.3). Note that det T(x) = 1 and
p = T+ -DD;u2 -D3u3)
2 2 D2”3
486 JOHN H. BARRETT
is a solution pair of (4.19) satisfying the initial conditions
(4.20) Y(u) = 0, &z) =q(: ,'i .
Howard (5.5) h as established a theorem involving the minimum eigen-

value of J” F. Reduced to the special case (4.3) his theorem becomes
Theorem H. If(i) min e.v.(J” F) + co as x + 00 and (ii) E(x) > e(x)l,
e >, 0 with J” e = co, then zz2(a) < co.
However, min e.v. E(x) = 0 and, hence, no such e(x) exists for (4.3).
Since det P(a) = 1 > 0, then Y-l(x) exists on [a, b) for some
b E (a, 00). Let K(x) = Y(x)Y-l(x), then
(4.21) K’ = E + KFK on [a, b); K(a) = 0.
Because of the fact that K(x) is symmetric at one point, x = a, it is

easy to show (9, 82) that K(x) is symmetric on [a, b).
By Lemma 4.2, det Y(x) > 0 ( a, a + 6) for some 6 > 0. Suppose that
det Y(X) > 0 on (a, co), i.e., that
(4.22) 2&z) = co.
Then H(x) = -Y(X) Y-‘(X) satisfies
(4.23) H’=F + HEH on (a, co),
and is symmetric, since H(x) = -K-l(x) near x = a.

Assume Howard’s property D (.55), i.e., that
(4.24) min e.v. (jz F) = ,in=fi[+ (SzF) 5 + ~0, as x + ~0.
Let b E (a, co); there exists a number c E (b, cc) such that
(4.25) H(x) > H(b) + jz F > I2 for a<b<c<x<a,

h
where the inequality is in the “positive-definite” sense, i.e.,
A >Bof+(A -B)(>O
for every nonzero vector 5 and where I, is the n x n identity. We now

parallel Coles’ proof of Theorem 1.4.
Let G(x) = 1, + jr (HEH); then H(x) > G(x) > 0 and G’(x) = HEH.
However, it does not follow that HEH > GEG. In fact, it does not
follow that H2 > G2, unless H and G commute (1.5). In spite of this
difficulty, we have -(G-l)’ = G-‘G’G-l = G-lHEHG-l and
(4.26) 0 < = (G-WEHG-‘) < G-‘(b) = I 2.

b
Taking the trace of (4.25), since E = &+G where
we have
(4.27) 2 >, jm tr(G-‘HEHG-l) > /m I/ EHG-l II2 3 jz 11G-‘HEHG-l jj.

b b b
Also, 11G-IHEHG-’ 113 11E/l/l/ H-IG /12, where the Euclidean norm
11A 11= (z afj)“’ for A = (~ij).
Dr. A. S. Householder suggested the following argument. Since

H-l < G-l on [c, co), there is a square root G112, i.e., G = (G1/2)2,
then G1/2H-1G1/2 < 12. Since G1/2H-1G1/2 and H-IG have the same
eigenvalues then 11H-lG 11< 2. Consequently, (4.27) implies that
(4.28)
If we violate (4.28) by assuming (4.24) then z2a(a) < 03.
Theorem 4.3. If u and v are solutions of (4.6) such that

z
-w2 -WV
(9 min e.v. -+a3 as x+-co
Si a -q4uv q2 - 4‘P21
and
m 22 + 02
(ii)
sa 7 = a,
then z22(a) < oofor equation (E4*).

488 JOHN H. BARRETT
Instead of violating (4.28) by condition (ii) we might allow the deter-

minant of J’zE to become unbounded.
Let D(x) = det(J,” E); then D’ = 0/r where
e(x) = j z ~[44 4t) - 44 WI2-df= z ___

v12(x,t, dt
a r(t) s a r(t)
and +uul(x,t) is the unique solution of
(4.30) (YIY’)’ + QIY = 0, r(t) = 0,
Theorem 4.4. The condition (ii) in Theorem 4.3 may be replaced by
(iii) det (1: E) = j: & (1: s’ ds) dt + co, as x+ co.
For the special case (3.3) ZI~(X,t) = x - t. Furthermore, if
A4 = -P(x) 3 0
and Q(x) < 0 then conditions for (i) are easily established.
Lemma 4.6. If Jz p = GO and p(x) > 0 on [a, a) then
min. e.v. j:p(t) i: i2) dt + 00, as x -+ co.
Corollary 4.4.1. For Eq. (3.3) ;f(i) P < 0 andQ > 0,
(ii) “;Pl=m
sn
st(s- t)2
and
(ii) -dsdt = co,

n y(s)
then .x,,(a) < co.
This is an improvement over Theorem 3.6 of (IO), since
(4.31)
(S
t $$
a
dtj2 < jt
a
9 ds - jt $,
a
where, instead of (iii), it was assumed that j” s2/r(s) ds = co. Dual

theorems are obtained by interchanging the roles of E and F.
Theorem 4.5. For Eq. (3.3), if(i) P < 0 andQ > 0,
(ii) sma (l/R) = co
and
(iii) either
I
oo or j: Ptt) j: P(s)(s - t)2 ds = co,
then 222(u) < co.

Because of an inequality like (4.31), the second alternative of (iii) is
satisfied if
j; P(t) [j: (t - s) P(s) ds]’ dt = co or mtzP(t) dt = 00,

sa
which were the best conditions given previously (10). Hunt’s oscillation
theorem (56) for
(4.32) [R(x)~(~)](~) + (-l)“+lP(x)y = 0, R(x) > 0, P(x) > 0
as given below can be established by paralleling the proof of Theorem 4.3.
Theorem 4.6. If J” P(x)(I~*P)~ dx = co, where PmP is an nth-iterated

integral of P(x), and j” (1 /R) = 00, then there exists a nontrivial solution of
(4.32) with a pair of n-fold zeros on [c, co) for each c E [a, CO).
Hunt (57) later proved a stronger result.
Theorem 4.7. Supposethat
s ‘*)[I”@-“/R(t))] P(x) dx = 00 and

i
m[(PnP)/R(x)] dx = co,
then there exists a set of 2n linearly independent solutions, with infinitely

many zeros on [a, co), of
[R(x) y(n)](n) + P(x) y = 0, n > 1,
where R(x) > 0, P(x) # 0 and R & P E C[a, a~).

490 JOHN H. BARRETT
It is not known whether the conclusion of Theorem 4.7 can be extended

to have all solutions oscillatory.
4.4. Quadratic Functionals and Wirtinger Inequalities

For the self-adjoint equation
(3.3) &Cyl = [(Ry”)’ + Qy’l’ + py = 0; R >O, R,Q&PEC[a, co),
it is a routine matter to calculate that
(4.32) MY; a, bl = jb(1[R(Y”)’ - Q(Y’)’ + b21
- / baYGLYI+ CY%Y- RY’Y%9
where g&y = (Ry”)’ + Qy’.

On the other hand if we emply a method of Coles (19) and integrate-
by-parts several times the right-hand side of
s b
a
Pw2 = -
we obtain on [c, b] C [a, co),
(4.33) I,[w; c, b] = -[[rB3y] w”/y - Ry”(~‘)~/y’]b,

b
- -Y2 ,‘-$,I’+ j~r[w~-~w~]2,
s c- Y [
where C,[y] = 0, y # 0 and y’ # 0 on [c, b] and w(x) is any C,-admis-
sible function on [c, b].
By making use of the derivatives and differences of the quotients
(4.34)
A0= u3(5ab2(x, a), A, = u3’/u2’, A, = qu2” ) h = UW’)’ + Qd

3 (Ru;)’ + Qu2’
the existence of an appropriate solution y(x) for (4.33) can be established

(11).
Lemma 4.7. If P(x) < 0 on [a, co), a o, Y’ > 0, (Ry”)’ + Qy’ < 0 on (a, b).
Furthermore, Ry” > 0 if Q > 0 on (a, b).

Using the solution guaranteed in Lemma 4.7 in (4.33) we have an analog
to the second-order focal point criterion in Theorem 1.9 (II).
Theorem 4.8.If P(x) < 0 on [a, co) then in order for pi(u) = co it
is necessary and suficient that for each b E (a, 00) and each nontrivial
admissiblefunction w(x) on [a, b] f or which w(a) = w’(a) = 0, the zero
of w’ being of order >i, it is true that
I&; a, b] = I” [Ye - Q(w’)~] > 0.

a
Recall that Theorem 4.2(d) sh ows that, under further conditions,

~~(a) = co if, and only if, x22(~) = 00. Various forms of I,[w; a, b] > 0
are also known as Wirtinger inequalities (19).
Corollary 4.8.1. If Q(x) > 0, P(x) < 0 and ]I (l/R) = co then

Eq. (3.3) has a nontrivial solution with a pair of double xeros, i.e.,
x~~(u) < co, if, and only zf,
w4 4 bl > 0, a<b<co,
for each admissiblefunction w(x) of Theorem 4.6.

W. T. Reid (83-89) proved Theorem 1.11 for vector-matrix systems,
which include (4. l), and his results will be stated here only in terms of
the special case (4.1).
Lemma 4.8. For Eq. (3.3), xzp(a) = 00 sf, and only if, for each
[c, 4 C [a, co>,
for all vector functions 7, .$ with T(X) absolutely continuous on [c, d]

and t(x) Lebesgue-measurableand essentially bounded on [c, d] with
492 JOHN H. BARRETT
q(c) = 0 = q(d), and equality only if Bt = 0 almost everywhere and

rl(x) = 0 on [c, d].
Hinton (52) recently made use of the following corollary of Reid’s

criterion.
Theorem 4.9. Equation (3.3) h as no nontrivial solution with a pair of

double zeros on [a, b], a 0
for every nontrivial function w E C”[a, b], where (Rw”)’ is piecewise

continuous on [a, b] satisfying
w(a) = W’(a) = 0 = w(b) = w’(b).
Following Leighton’s second-order results (69), Hinton established a

comparison theorem for fourth-order equations (-Ed).
Theorem 4.10. If (i) th ere is a nontrivial solution of (EJ with two

double zeros on [a, b], a < b < 00,
(ii) J?~ is the same operator as L, , except that r2 , q2 and q3 are replaced
by i, , & and & E C[a, b], respectively and
(iii) there is a function w(x) satisfying the conditions for w of
Theorem 4.9 such that
then Lb[ y] = 0 has a nontrivial solution with a pair of double zeros on [a, b].
Using the preceding comparison theorem, Hinton proved several
“oscillation” theorems for Eq. (3.3).
Theorem 4.11. If h(x) 3 0 and E C[a, a) such that

50
s xh(x) dx = co
a
and lim inf J(t) = - co as t + co, where
txh(x) dx , 2
x ds
/[Ja 1
then there is a number b E (a, 00) such that (3.3), f&y] = 0, has a non-
trivial solution with two double zeros on [a, b].
Corollary 49.1. If, as t + co,
lim inf t-4 t {R(x) - (t - x)” Q(x) + P(x)(t - ~)~/4} dx = -00,

sa
then the conclusion of Theorem 4.11 holds.
Corollary 4.9.2. If Jr [x/R(x)] dx = 00 and, as t + 00,
lim inf 1: ]P(x) [I: G do]’ - Q(X) [J: -&I21 dx/[j: & ds]I = -co,
then the conclusion of Theorem 4.11 holds.

In (53), Hinton has extended his results to
(Ry(yn) + (-Iy-lPy = 0.
Reid (88) has developed the quadratic functional criterion for complex
self-adjoint quasi-differential equations of order 2n,
(1.20) DWOy = 0,
where
Dck> = Dk, k = 0, l,..., n - 1;

Den) = p,, Dn + ip2n-l D”-l;
D(n+O = DDW-l> - ( -l)i[ip2n-2i+l - p2n-2jD”-i - $2n-2i-l D--l],
j = l,..., n - 1;
DW> = DD(2”-1, _ (-l)“[ip, - p, DO];
494 JOHN H. BARRETT
and
pj E wqu, b] for j = 0, I,..., 272.
An example of Reid’s results follows.
Theorem 4.12. Ifp,,-l(~) G 0, JT (I/&) = 00,
for all complex n-vectors r = (ri), and D(3”)y = 0 has no nontrivial

solution with a pair of n-fold zeros on [c, 00) for c su.ciently large, then
each of the integrals
is convergent.
s z p2,(t) t2n--2a-2 dt, N = 0, I,..., n - 1;
Since Reid (90) h as given an elegant summary of variational results for

self-adjoint systems, we will not repeat more of his many results here.
Also, we have concentrated on nonvariational methods in hopes that our
methods may also apply to nonself-adjoint equations.
V. Second-Order Matrix Equations and Related First-Order Systems
5.1. Examples
In Section 4.3 we studied a particular example (4.19) of the matrix

differential system
(5.1) Y’ = E(x)P, P’ = --F(x)E’,
where E and F are symmetric square matrices of continuous functions on

an interval 1. Another example is the matrix equation having the
coefficients of (4.15), i.e.,
1IT1 F= -44 -91

(5.2) E=b;l qlb i --41 l/r2 ! '
which was utilized by Atkinson (2) and Hartman and Wintner (48), as a
formulation of the special fourth-order scalar self-adjoint equation (3.3).
The second-order scalar complex differential equation of Subsection 1.7,
(5.3) (v')' + qy = 0; T = Y, + ir* # 0, q’ = q1 + ipz , Yi & pi E C(I),
is equivalent to the system (5.1) where
E = (l/l r I”) (-Z: Z;), F = (-E E).
For oscillation theory of (5.3) see (7).

The second-order matrix differential equation
(5.4) (R(x)Y')' + Q(x)Y = 0,

where R and Q are symmetric matrices of continuous functions on an
interval f and R > 0 (positive-definite) on 1, has been the subject of
numerous investigations (5, 6, 5.5, 87). Of course, if Y = RY’ then (5.4)
is equivalent to (5.1) with
(5.4') E = R-l, F = Q.
Note, also, that (5.3) is equivalent to (5.4) with
5.2. Basic Properties of Solutions

A matrix function Y(X) is said to be a solution of (5.1) if there exists a
corresponding matrix function Y(x) such that (Y, Y) is a solution pair
of the system (5.1).
Let Yr and Ya be two such matrix solutions of (5.1) on any interval I
and define
(5.5) W[Y, ) Y,] = Y,'Ps - E',+y, .
This matrix functional reduces to the usual Wronskian for the scalar
(n = 1) case and, in general, has analogous properties. One property
which does not hold in general is the identical vanishing of W[Y, Y]. In
fact the most that can be said is that W[Y, yl is skew-symmetric. In case
IV[Y, Y] = 0 then the column vectors of the matrix solution are said to
be conjoined (85). This terminology was introduced and the following
analogs of scalar properties were first proved by Reid and others who were
496 JOHN H. BARRETT
working in the Calculus of variations in the 1930’s (17, 74). The author
listed these properties later (5).
Theorem 5.1. Let U(x) be a matrix solution of (5.1).

(a) If A is a constant matrix then UA is a solution of (5.1) such that
W[UA, UA] = 0.
(b) If det U(X) # 0 on an interval I and a E I then
V(x) = U(x) JZ [ U-‘E( u-l)+], x E I,

a
is also a solution of (5.1) such that
wp, V] = 0 and W[U, V] = l,,, .
(c) If V(x) is a solution of (5.1) such that W[V, V] = 0 and

det( W[ U, V]) # 0 then every solution of (5.1) isgiven by
1’ = UA + T’B,
where A and B are arbitrary constant matrices.

Such a pair of solutions U, V can always be obtained by specifying the
initial conditions
(5.6) u(a) = I,, , O(a) = 0 and V(a) = 0, F(a) = I.
Recall the fact that if Y(X) is a matrix solution of (5.1) and y is

a constant vector then ,B = Yr is a vector solution of the related vector-
matrix equation
(5. I ‘) /I’ = Efl, p’ = -F/l.
Conversely any vector solution /3 of (5.1’) forms a column of some matrix

solution Y(X) of (5.1).
Corollary 5.1 .I. Under the hypotheses of Theorem 5.1(c), every

vector solution of (5.1’) isgiven by
p = uy + P-s,
where y and S are arbitrary constant vectors.

Corollary 5.1.2. Let V(x) be a (matrix) solution of(5.1) on an interval I

such that a E I, V(u) = 0 and det( V(u)) # 0, then in order for there to
exist a number b E I, b # a, and a nontrivial vector sol&ion ,Q of (5.1’)
such that
(5.7) B(a) = B(b) = 0
it is necessaryand suficient that det V(b) = 0.
Therefore, singularities of matrix solutions of (5.1) determine conjugate

points [i.e., (5.7) holds] of vector solutions of (5.1’).
5.3. Mutrix Trigonometric Functions
An interesting special case of (5.1) is that when E = F. For each

symmetric n x 1zmatrix Q(x) of continuous functions on [a, CO)define
(5.8) s = S[x, a; Ql, c = C[x, a; Q]

to be the solution pair of
S’ = QC, S(u) = 0,
(5.9) C’ = -QS, C(a) = I.
For the scalar (n = 1) case 5’ = sin(]z Q) and C = cos(Jz Q), hence, the
matrices (5.8) are said to be matrix sines and cosines. Note that if
det Q # 0 on an interval I then S and C are solutions of the second-order
matrix equation
(5.9‘) (Q-1Y')' + QY = 0 on I.
Furthermore, W[S, S] = W[C, C] = 0 and W[C, S] = 1, . The

matrix functions (5.5) were introduced by the author (6) and were
studied further by Reid (87) and Etgen (25-27). These and other
easily-proved results imply certain expected “trigonometric” identities.
Theorem 5.2.
498 JOHN H. BARRETT
Note that Theorem 5.2 is equivalent to saying that the 2n x 2n matrix
is orthogonal.
Etgen (25) has listed a number of other properties of S and C.
Theorem 5.3. (a) If b E (a, co) such that det(C[b, a; Q]) = 0 and
det(C[x, a; $31) > 0 012[a, 6) then det(S[x, a; Q] # 0 on (a, b).
(b) If det(C[x, a; Q]) # 0 on [b, c] C [a, 00) then det(S[x, a; Q]
has at most n zeros in [b, c].
(c) If det(S[x, a; Q]) # 0 on [b, c] C [a, a) then det(C[x, a; Q]) has
at most n zeros in [b, c].
He also gives an interesting simple example for which
det(C[x, a; 01) = 0 = det(S[x, a; Q]
for infinitely many values of x E [a, co); i.e.,
0 = (; =Y2)Y a = 0, s = (siy sin(;,2)x) )

cos Trx 0
c=
( 0 cos(n/2)x i *
However, it should be noted that if y is a constant vector such that, for

some x, Sy = Cy = 0 then y = 0. By means of a discussion similar
to that of Section 4.5, the author (6) first proved the following.
Theorem 5.4. (a>If Q(x) > 0 on [a, CO) and J” tr Q = GO then

WS[x, a, $21)hus at least one zero on (a, co).
(b) If Q(x) > 0 on [a, GO)and
s “trQ
a
<r/2&
then det(S[x, a; Q]) # 0 and det(C[x, a; Q]) # 0 on (a, co).

Reid (87) improved (b) by replacing rr/2 d/n by ~r/2 and showed the
following.
Theorem 5.5. If Q(x) > 0 on [a, m) then there exists a number

c E (a, 00) such that no nontrivial vector solution /3 of the system
(5.9’) B’ = Qb, 6’ = -QP
has two zeros on [c, 00) if, and only if, Jz tr Q = 00.
Later, Etgen [2.5] proved a slightly different version of Theorem 5.5
and, subsequently [26], established a “double angle” formula.
Theorem 5.6. If Y, Z is the solution pair of matrices of

Y’ = Q’Z + ZQ, T=-QY-YQ
and Y(u) = 0, Z(a) = I then Y and Z are symmetric and

Y = 2sct, z = cc+ - SS.
Finally, let us note [6] that S and C satisfy a Lipschitz condition

with respect to Q.
Theorem 5.7. If for i = 1 and 2, Qi(x) is a symmetric matrix of

continuousfunctions on [a, CO)with Si = S[a, X; QJ and Ci = C[a, X; QJ
then there exists a positive number M such that
5.4. A Matrix Prtifeer Transformation

In 1957 the first Priifer transformation for second-order matrix
differential equations (5.4) was published (6) and shortly thereafter
Reid (87) gave a more general development, which will be given here for
the real case.
Let Y(X) be a matrix solution of (5.1); i.e., there is a companion matrix
Y(x) such that
(5-l) Y’ = E(x) P, P’ = --F(x)Y,
where E and F are symmetric (square) matrices of continuous functions

on an interval I, and Y(a) = 0, J?(u) = 1, .
Suppose that there exist continuous matrix functions P(x) and Q(X) such
that
(5.10) Y(X) = S+[x; a, Q] P(X) and Y(x) = C’[X; a, Q] P(x).

500 JOHN H. BARRETT
Because of the identities in Theorem 5.2,
(5.1 I) P+P = E’+T’ + Ii+B > 0
and P(x) is nonsingular on I.

Differentiation of (5. IO) yields
\9P + SQP = EC+P,

(5.10’)
(C+P’ - S’QP = -FSP
and Theorem 5.2 allows us to solve for P’ and Q, so that
(P’ = [SEC+ - CFS+]P,

(5.12)
( Q = CECi + SFS+.
Conversely, suppose that P(x) satisfies (5.11) and
(5.13) p+pp’= YtEi’ - PfFy
and Q(x) satisfies
(5.14) PtQP = PtEP + Y’FY;
then it is a routine matter to check that (5.10) defines the solution of (5.1)
which also satisfies the initial conditions (5.1).
Thus Reid reduced the problem to finding a solution P(x) of (5.11)
and (5.13). Let
If(x) = Y+(x) Y(x) + P+(x) P(x) {>0 on [a, co)},
b E (a, 00) and k be a positive number such that
0 < k’M(x) < l,, on [a, b].
Then U(x) = 1, - Is~.M(x) satisfies
0 < U(x) < 1, on [u, 61.
Following Reid, let
V(x) = 1, - t cpyx), where (1 - x) 1’2 = 1 - f C,,Xk,

k=l k=l
and it follows that

(i) V2(,) = 1, - U(X),
(ii) V(x) > 0,
and
(iii) V(x) E C’[a, b].
If N(x) = (l/k) V(X) then N2(x) = M(i) on [a, 00). Furthermore,

if Nr2(x) = M(x) and N,(X) > 0 then N,(X) E N(X) > 0 & E C’[a, b],
and N,(X) E C’[a, co).
Lemma 5.1. There is a unique positive-de$nite matrix function PO(x)

such that
(5.15) P,2(x) = Y+Y + P+P, P&> > 0 and P, E qa, co).
It is easy to seethat all solutions of (5.11) are given by
(5.16) P = WPO)
where W is an orthogonal matrix, i.e., I%‘%’ = 1, , and condition (5.13)

is equivalent to
(5.17) W’ = W{P;l[Y+EP - Ei+FY - POP,‘] 5’) on [a, a~).
Now (5.17) is a linear equation and the solution IV, satisfying W(a) = 1,
is orthogonal on [a, co); and for
P = PI = W,P, ,
Q = Q, = P;-l[P+EP + Y+FY] P;l,
the solutions Y and Y are given by the Prtifer Transformation (5.10).

An alternate approach is to solve the functional equation
Q = CEC+ $ SFS+; C = C[x; a,81, S = S[x; a, Q]
for Q(X), which is possible because of the Lipschitz conditions of

Theorem 5.7, and then solve the other equation of (5.12) for P(x). The
extension to Reid’s more general case of complex Hermitian coefficients
in (5.1) may be obtained easily.
502 JOHN H. BARRETT
5.5 Atkinson’s Formulation

Another transformation of (5. I), which is similar to the Prefer Trans-
formation, has been introduced by Atkinson (2) and was later utilized by
Etgen (27) and Coppel(21).
Consider a matrix solution Y(X) of (5.1) such that
W[Y, k’] = 0 and det E;(x) + 0 on I
and, following Atkinson, let
(5.18) Z(x) = (P + iY)(P - iY)-1, i2 = -1,
Differentiation of (5.18) yields that Z(x) is a solution of the complex

matrix equation
(5.19) 2’ = 2iZG(x),
where G is the Hermitian matrix
(5.20) G = (Pi + W-l( step + YT’Y)(? - iY)-?
Furthermore,
Z(x) is a unitary matrix
and if wj(x), j = 1, 2 ,..., n, denote its eigenvalues then, for at least one j
andcEI,
Wj(C) = +I if and only if det Y(c) = 0
or
Wj(C) = -1 if and only if det P(c) = 0.
Using these ideas Atkinson established some separation theorems for

such numbers c. A detailed discussion is given in Chapter 10 of his book
(2) where the coefficient matrices E and F are Hermitian.
Etgen (27) pointed out that for the case where
Y(u) = 0 and P(a) = I
Atkinson’s transformation (5.18) is related to Etgen’s “double angle”

formulas of Theorem 5.6, i.e.,
(5.21) Z = (CC+ - SS) + 2iSC.

Coppel (22) utilized Atkinson’s formulation to establish comparison

theorem for more general conjugate points, which correspond to general
two-point homogeneous boundary conditions.
5.6. Oscillation Theorems

Most of the discussion of the 2 x 2 matrix equations of Section 4
holds also for the matrix equation (5.1) with E and F n x n symmetric
(or Hermitian) matrices. For example, the proof of Theorem 4.3 also
establishes a more general result.
Theorem 5.8. If in Ep. (5.1) over I = [a, CO)

(i) min e.v. (J,“F) + CO
and
(ii) max e.v. (Jz E) -+ CO as x -+ co, and E(x) > 0, then any
matrix solution Y(x) for which W[Y, yl = 0 has an oscillatory deter-
minant, i.e.,
det Y(x) has infnite~y many zeros as x + ~3.
Such a matrix system is said to be oscillatory. Howard (55) proved a

weaker form of Theorem 5.8 with (ii) replaced by
E(x)>e(x)l,, me= co.

i
Tomastik (ZO4) recently proved a more general theorem which holds

under the assumptions that
(iii) E > 0 and F > 0.
Using the integrated form of the Riccati equations (4.21) and (4.23),
H(x) = H(b) + jZF + jn HEH

b b
K(x) = K(b) + j= E + jz KFK

b b
and applying the Courant-Fischer min-max theorem to the eigenvalues

of H(x) and K(x) = -H-l(x), h e established the following for E > 0
and F > 0.
504 JOHN H. BARRETT
Theorem 5.9. If, as x -+ CO, the limits of I eigenvalues of jz E and

s eigenvalues of J-z F are positively injnite and r + s > n, then the system
(5.1) with nonnegative-dejinite coeficients (iii) is oscillatory on [a, CO).
Finally he gave an example to show that the inequality r + s > n cannot

be relaxed.
It would be interesting to know if the nonnegative conditions (iii)
could be relaxed, as they were in the special case, r = 1 and s = n, of
Theorem 5.8.
ACKNOWLEDGMENTS
The author is indebted to the Associated Western Universities and, in particular, to

the director of the Differential Equations Symposium, Professor R. W. McKelvey, and
his committee for the excellent arrangements for the lectures, the very able audience and
for extending the invitation to give lectures to one who welcomed the opportunity.
Thanks are also due to the author’s former students and colleagues at the Universities of
Utah and Tennessee for their patience and numerous suggestions for improvement
during earlier versions of these lectures.
BIBLIOGRAPHY
1. P. APPEL, Sur la transformation des equations differentielles lineares, Coopt. Rend.

(Paris) 91 (1880), 211-214.
2. F. V. ATKINSON, “Discrete and Continuous Boundary Problems.” Academic Press,
New York, 1964.
3. N. V. AZBELEV AND Z. B. CALJUK, On the question of distribution of zeros of solutions
of linear differential equations of the third order, Mat. Shorn. 51, 475-486 (1960).
[English Translation: A&IS Transl. 42, 233-245 (1964)].
4. J. H. BARRETT, Behavior of solutions of second-order self-adjoint differential
equations, Puoc. AMS 6, 247-251 (1955).
5. J. H. BARRETT, Matrix systems of second order differential equations, Portugal.
Math. 14, 79-89 (1955).
6. J. H. BARRETT, A Prufer transformation for matrix differential equations, PYOC.
Am. Math. Sot. 8, 510-518 (1957).
7. J. H. BARRETT, Second order complex
differential equations with a real independent
variable, Pact’& /. Math. 8, 187-200 (1958).
8. J. H. BARRETT, Disconjugacy of second order linear differential equations with
nonnegative coefficients, Proc. Am. Math. Sot. 10, 552-561 (1959).
9. J. H. BARRETT, Systems-disconjugacy of a fourth-order differential equation, PYOC.
Am. Math. Sot. 12, 205-213 (1961).
10. J. H. BARRETT, Fourth-order boundary value problems and comparison theorems,
Canadian /. Math. 13, 625-638 (1961).
II. J. H. BARRETT, Two-point boundary problems for linear self-adjoint differential

equations of the fourth order with middle term, Duke Math. 1. 29, 543-554 (1962).
12. J. H. BARRETT, Disconjugacy of generalized linear differential equations of third
and fourth order, (Abstract No. 594-36), Notices AMS 9, 469 (1962).
13. J. H. BARRETT, Canonical forms for third-order linear differential equations, Ann.
Mat. Pura Appl. 65, 253-274 (1964).
14. J. H. BARRETT, Third-order equations with nonnegative coefficients, J. Math. Anal.
Appl. 24, 212-224 (1968).
15. R. BELLMAN, “Introduction to Matrix Analysis.” McGraw-Hill, New York,
1960.
16. G. D. BIRKHOFF, On the solutions of ordinary linear homogeneous differential
equations of the third order, Ann. Math. 12, 103-127 (19l(tll).
17. G. A. BLISS, Lectures on the Calculus of Variations. The University of Chicago
Press, Chicago, 1946.
18. E. A. CODDINGTON AND N. LEVINSON, “Theory of Ordinary Differential Equations.”
McGraw-Hill, New York, 1955.
19. W. J. COLES, A general Wirtinger-type inequality, Duke Math. J. 27,133-138 (1960).
20. W. J. COLES, An oscillation criterion for second-order linear differential equations,
Proc. AMS (to appear).
21. W. A. COPPEL, Comparison theorems for canonical systems of differential equations,
/. Math. Anal. Appl. 12, 306315 (1965).
22. R. COURANT AND D. HILBERT, “Methods of Mathematical Physics,” Vol. I. Inter-
science, 1953, New York.
23. M. A. DOBROHOTOVA, On boundedness of solutions of linear differential equations
of third order (in Russian), Yaroslavl. Gosudarstvennyi Pedagogichoskii Institut
Uchenye Zapiski 34, 19-34 (1960).
24. J. M. DOLAN, Oscillatory Behavior of Solutions of Linear Ordinary Differential
Equations of Third Order, Unpublished doctoral dissertation, University of
Tennessee, 1967.
25. G. J. ETGEN, Oscillatory properties of certain nonlinear matrix differential systems
of second order, Trans. Am. Math. Sot. 122, 289-310 (1966).
26. G. J. ETGEN, “A note on trigonometric matrices,” Proc. Am. Math. Sot. 17, 1226-
1232 (1966).
27. G. J. ETGEN, “On the determinants of solutions of second order matrix differential
systems,” J. Math. Anal. Appl. 18, 585-598 (1967).
28. G. FROBENIUS, tiber adjugierte lineare Differentialausdrticke, Journalfiir Mathematik
85, 185-213 (1878).
29. LANDOLINO GIULIANO, “On linear self-adjoint equations of third order,” Boll.
Uniw. Mat. Ital. 12, 1618 (1957).
30. M. GREGUB, On certain new properties of solutions of the differential equation
y”’ + Qy’ + Q’y = 0 (in Czech.) Publ. Fat. Sci. Universite Masaryk, Bruo, Czech.
362, 237-251 (1955).
31. M. GREGU~, On a new boundary-value problem for differential equations of third
order (in Russian), Czech. Math. I. 7, 41-47 (1957).
32. M. GREG& On the linear differential equations of third order with constant
coefficients (in Slovak), Acta Fat. Rerum. Nutur. Uniw. Come&m. Math. 2, 61-65
(1957).
33. M. GREGUS, A remark about the dispersions and transformations of a differential
506 JOHN H. BARRETT
equation of third-order (in Czech.), Acta Fat. Rerum Natur. Univ. Comenian. Math.
4, 205-211 (1959).
34. M. GREG& Oxzillatorische Eigenschatten der Liisungen der linearen Differential-
gleichung dritter Ordnung y”’ + 2Ay’ + (A’ + b)y = 0, wo A = A(x) < 0 ist,
Czech. Mat. J. 9, 416-427 (1959).
35. M. GREGUS, tiber einige Eigenschaften der Losungen der Differentialgleichung
y”’ + 2Ay’ + (A’ + b)y = 0, A < 0, Czech. Mat. J. (86) 11, 106-l 15 (1961).
36. M. GREG&, “Oscillatory properties of the solutions of the third-order differential
equation of the type y’” + 2A(x)y’ + [A’(x) + b(x)]y = 0, Acta Fuc. Rerum natur.
Univ. Comenian. Math. 6, 275-300 (1961).
37. M. GREGUS, “Bemerkungen zu der unliisharen Randwert problemen dritter
Ordnung,” Acta Fat. Rerum Natur. Univ. Come&n. Math. 7, 639-647 (1963).
38. M. GREGUS, uber einige Eigenschatten der Lijsungen der Differentialgleichung
dritter Ordnung, Acta Fat. Rerum Natur. Univ. Cornet&n. Math. 7, 585-595 (1963).
39. M. GREG& iiber einige Radwertprobleme Dritter Ordnung, Czech. Math. J. 13,
551-560 (1963).
40. M. GREGUS, tfber die lineare homogene Differentialgleichung dritter Ordnung,
Wiss. Zeit. Martin-Luther Univ. Halle- Wittenberg X11/3, 265-286 (1963).
41. M. GREG& “Uber die asymptotischen Eigenschatten der Lijsungen der linearen
Differentialgleichung dritter Ordnung,” Ann. Mat. Pura Appl. 63, l-10 (1963).
42. M. GREGUS, tiber die Eigenschaften der Losungen einiger quasilinearer Gleichungen
3. Ordnung, Acta Fat. Rerum Natur. Univ. Corner&n. Math. 10, 1 l-22 (1965).
43. W. HAHN, tjber Orthogonalpolynome mit drei Parametern, Deutsche Math. 5,
273-278 (1940).
44. T. G. HALLAM, Asymptotic behavior of the solutions of an nth-order nonhomo-
geneous ordinary differential equation, Trans. Am. Math. Sot. 122, 177-194 (1966).
45. M. HANAN, Oscillation criteria for third-order linear differential equations, Pacific
J. Math. 11, 919-944 (1961).
46. G. HANDLEMAN AND YIH-0 Tu, Lateral vibrations of a beam under initial linear
axial stress, J. SIAM 9, 455-473 (1961).
47. P. HARTMAN, “Ordinary Differential Equations.” Wiley, New York, 1964.
48. P. HARTMAN AND A. WINTNER, On disconjugate differential systems, Canadian
J. Math. 8, 72-8 1 (I 956).
49. J. HEIDEL, Qualitative behavior of solutions of a third order nonlinear differential
equation, Pacific J. Math. 27, 507-526 (1968).
50. E. HILLE, Nonoscillation theorems, Trans. Am. Math. Sot. 64, 234-252 (1948).
51. D. B. HINTON, Disconjugate properties of a system of differential equations,
J. Differential Eqs. 2, 420-437 (1966).
52. I). B. HINTON, Clamped end boundary conditions for fourth-order self-adjoint
differential equations, Duke Math. /. 34, 131-138 (1967).
53. D. B. HINTON, A Criterion for n-n oscillations in differential equations of order 2n,
Proc. Am. Math. Sot. 19, 511-518 (1968).
54. H. C. HOWARD, Oscillation criteria for fourth-order linear differential equations,
Trans. Am. Math. Sot. 96, 296-311 (1960).
55. H. C. HOWARD, Oscillation criteria for matrix differential equations, Canadian
J. Math. 19, 184-199 (1967).
56. R. W. HUNT, The behavior of solutions of ordinary self-adjoint differential equations
of arbitrary even order, Pacific J. Math. 12, 945-961 (1962).
57. R. W. HUNT, Oscillation properties of even-order linear differential equations,

Trans. Am. Math. Sot. 115, 54-61 (1965).
58. E. L. Ince, “Ordinary Differential Equations,” Dover Publications, New York, 1944.
59. V. A. JAKUBOVIE, Oscillation properties of solutions of canonical equations, Mat. Sb.
56, 3-42 (1962); [English translation: Am. Math. Sot. AMS Transl. 42, 247-288
19641.
60. E. KAMKE, “Differentialgleichungen Lijsungsmethoden und Losungen, 1.” Becker
and Erler, Leipzig, 1943 and J. W. Edwards, Ann Arbor, 1945.
61. A. KNESER, Untersuchen iiber die reelen Nullstellen der Integrale linearer DifFeren-
tialgleichungen, Math. Ann. 42, 409-435 (1893).
62. V. A. KONDRAT’EV, On the oscillation of solutions of linear differential equations of
third and fourth order (in Russian), Trudy Mosk. Mat. Ohs%. 8, 259-282 (1959);
Dokl. Akad. Nauk. USSR 118, 22-24 (1958).
63. V. A. KONDRAT’EV, Oscillatory properties of solutions of the equation ~1%) +p(x)y = 0
(in Russian), Trudy Mask. Mat. ObE 10, 419-436 (1961) ; Dolk. Akad. Nauk USSR
120, 1180-1182 (1958).
64. A. LASOTA, Sur la distance entre les zeros de l’equation differentielle lineaire du
troisieme ordre, Ann. Polon. Math. 13, 129-132 (1963).
65. A. C. LAZER, The behavior of solutions of the differential equation y”’ + p(x)y’ +
q(x)y = 0, Pacific J. Math. 17, 435-466 (1966).
66. W. LEIGHTON, Principal quadratic functionals, Trans. Am. Math. Sot. 67, 253-274
(1949).
67. W. LEIGHTON, The detection of the oscillation of solutions of a second order linear
differential equation, Duke Math. J. 17, 57-62 (1950).
68. W. LEIGHTON, On self-adjoint differential equations of second-order, ].London Math.
Sot. 27, 37-47 (1952).
69. W. LEIGHTON, Comparison theorems for linear differential equations of second-order,
Proc. Am. Math. Sot. 13, 603-610 (1962).
70. W. LEIGHTON AND 2. NEHARI, On the oscillation of solutions of self-adjoint linear
differential equations of the fourth order, Trans. Am. Math. Sot. 89, 325-377 (1958).
(Contains an extensive bibliography).
71. A. Ju. LEVIN, Some questions on the oscillation of solutions of linear differential
equations (in Russian), Dokl. Akad. Nauk SSSR 148, 512-515 (1963); Soviet Math.
4, 121-124 (1963).
72. A. Ju. LEVIN, Distribution of the zeros of solutions of a linear differential equation,
(in Russian), Dokl. Akad. Nauk SSSR 156, 1281-1284 (1964); So&et Math. 5,
818-821 (1964).
73. G. MAMMANA, Decomposizione delle espressioni differenziali lineari omogenee in
prodotti di fattori simbolici e applicazione relativa allo studio delle equazioni
differenziali lineari, Math. Z. 33, 186-231 (1931).
74. M. MORSE, “The Calculus of Variations in the Large.” American Mathematical
Society Colloquium Publications, New York, 1934.
75. Z. NEHARI, Oscillation criteria for second-order linear differential equations, Trans.
Am. Math. Sot. 85, 428-445 (1958).
76. Z. NEHARI, On the zeros of solutions of nth order linear differential equations,
J. London Math. Sot. 39, 327-332 (1964).
77. Z. NEHARI, Nonoscillation criteria for nth order linear differential equations, Duke
Math. J, 32, 607-616 (1965).
508 JOHN H. BARRETT
78. Z. NEHARI, Disconjugate Linear Operators, Trans. Am. Moth. Sot. AMS 129,
50&516 (1967).
79. G. POLYA, On the mean-value theorem corresponding to a given linear homogeneous
differential equation, Trans. Am. Math. Sot. 24, 312-324 (1922).
80. H. PROOFER, Neue Herleitung der Sturm-Liouvilleschen Reihenentwicklung stetiger
Funktionen, Math. Ann. 95, 499-518 (1926).
8/. M. RAB, Asymptotische Eigenschatten der Liisungen Linearer Differentialgleichung
dritter Ordnung, Spisy Publ. Fat. Sci. Masaryk 374, 177-l 84 (1956/4).
82. W. T. REID, A matrix differential equation of the Riccati type, Am. /. Math. 68,
237-246 (1946). [See also Am. J. Math. 70, 460 (1948).]
83. W. T. REID, Oscillation criteria for linear differential equations with complex
coefficients, Pacific /. Math. 6, 733-751 (1956).
84. W. T. REID, A comparison theorem for self-adjoint differential equations of second
order, Ann. Moth. 65, 197-202 (1957).
85. W. T. REID, Oscillation criteria for linear differential systems with complex
coefficients, Pacific /. Math. 6, 733-751 (1956).
86. W. T. REID, Principal solutions of nonoscillatory self-adjoint linear differential
systems, Pacific J. Math. 8, 147-169 (1958).
87. W. T. REID, A Prefer transformation for differential systems, Pacific J. Moth. 8,
575-584 (1958).
88. W. T. REID, Oscillation criteria for self-adjoint differential systems, Trans. Am.
Math. Sot. 101, 91-106 (1961).
89. W. T. REID, Riccati matrix differential equations and nonoscillation criteria for
associated linear systems, Paci’c J. Math. 13, 664-685 (1963).
90. W. T. REID, Variational methods and boundary problems for ordinary linear differen-
tial systems, in “TheProceedings of the Japan-United States Seminar on Functional
and Differential Equations” (W. A. H arris and Y. Sibuya, Eds.), pp. 267-299.
Benjamin, New York, 1967.
9/. G. SANSONE, Studi sulle equazioni differenziali lineari omogenee di terzo
ordine nel campo real e, Reoista Mat. Fis. Tear. (Tucuman), 6, 195-253
(1948).
92. G. SANSONE, “Equazioni Differenziali nel Campo Reale,” Zanichelli, Bologna, 1956.
93. J. SCHRBDER, Randwertantgaben vieter Ordnung mit positiver Greenscher Funktion,
Math. Z. 90, 429-440 (1965).
94. J. SCHR~DER, Hinreichende Bedingungen bei Differentialgleichungen vierter
Ordnung, Math. Z. 92, 75-94 (1966).
95. B. SCHWARZ, Disconjugacy of Complex Differential Systems, Trans. Am. Math. Sot.
125, 482-496 (1966).
96. G. SEIFERT, A third order boundary value problem arising in aeroelastic wing theory,
@art. Appl. Math. 9, 210-218 (1951).
97. G. SEIFERT, A third order irregular boundary value problem and the associated
series, Pacific /. Math. 2, 395-406 (1952).
98. T. L. SHERMAN, Properties of solutions of nth order linear differential equations,
Pacific /. Math. 15, 1045-1060 (1965).
99. D. SHINN, Existence theorems for the quasi-differential equation of the nth order,
Compt. Rend. Acad. Sci. URSS 18, 515-518 (1938).
100. R. L. STERNBERG, Variational methods and nonoscillation theorems for systems of
differential equations, Duke Math. J. 19, 31 l-322 (1952).
101. M. ~EC, Sur une propritte des integrales de l’equation y’“’ + Q(x)y = 0, n = 3, 4,
Czech. Math. J. 7, 450-461 (1957).
102. M. SVEC, Some remarks on a third-order linear differential equation, (in Russian),
Czech. Math. 1. 15, 42-49 (1965).
103. M. SVEC, Einige asymptotische und oszillatorische Eigenschaften der Differential-
gleichung y”‘p+A(x)y’ + B(x)y = 0, Czech. Mat. J. 15, 378-393 (1965).
104. E. C. TOMASTIK, Singular quadratic functionals of n dependent variables, Trans.
Am. Math. Sot. 124, 60-76 (1966).
105. P. WALTMAN, Oscillation criteria for third order nonlinear differential equations,
Pacific J. Math. 18, 385-389 (1966).
106. W. M. WHYBURN, On self-adjoint ordinary differential equations of the fourth
order, Am. J. Math. 52, 171-196 (1930).
107. D. WILLET, On the Oscillatory Behavior of the Solutions of Second-Order Linear
Differential Equations, [AWU Differential Equations Symposium (1967)]
unpublished.
108. A. WINTNER, A criterion of oscillatory stability, Quart. Appl. Math. 7, 115-l 17
(1949).
109. M. ZEDEK, Cayley’s decomposition and Polya’s W-property of ordinary linear
differential equations, Israel j. Math. 3, 81-86 (1965).
110. A. J. ZETTL, Adjoint linear differential operators, PYOC. Am. Math. Sot. 16, 1239-
1241 (1965).
111. A. J. ZETTL (with E. A. CODDINGTON), Hermitian and anti-Hermitian properties of
Green’s matrices, Pacific J. Math. 18, 451454 (1966).
112. A. J. ZETTL, Some identities related to Polya’s property TV for linear differential
equations, Proc. Am. Math. Sot. 18, 992-994 (1967).

Oscillation Theory of Ordinary Linear Differential Equations

Uploaded by

Copyright:

Available Formats

Oscillation Theory of Ordinary Linear Differential Equations

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Oscillation Theory of Ordinary Linear Differential Equations

Uploaded by

Copyright:

Available Formats

Oscillation Theory of Ordinary Linear Differential

The purpose of this article [a compilation of lectures originally

theorems, motivating examples and developments of the ideas are given

Much of the material in this introductory section is contained in

1.1. Basic Properties

c4t) -qyl = (TY’)’ + qy = 0; r > 0, r & 9 E W),

(b) If r is not d#~entiable then the change of independent variable:

U-2) t = z (l/Y) 2 x = x(t), Y(t) = YWI

Integration-by-parts of zZ,[y] (or zL,[y]), where z is an arbitrary

Lemma 1.3. (a) If y, z E C”(I) then

&[y] = {xy’ - yz’ + AYZ}’ + yZ2’[x],

where Z,+[z] = ( z’ - AZ)’ + Bz, an adjoint operator of 1, .

GLYI = [GY’ - Y41’ + Y-u4

1.2. Factoring and Disconjugacy

Theorem 1.1. If L,[v] = 0 and v # 0 on I’ C I then

(4 VUYI = b4Y - YWI’

(1.4) UYI = (llV)PV2(Y14’l’

Definition 1 .l. A second-order linear operator (e.g., LJ and the

Let vr = r~r(x, a) be the solution of (E,) defined by the initial con-

U-3) Yk4 = 0 and D,y(a) = (ry’)(u) = 1.

This solution is called the principal solution of (,?I&)at x = a and oscilla-

c%‘) (ry’)’ + q1y = 0; q1 E C(I) and q1 < q on I

(4”) (%Y’)’ + 91Y = 0

W. Leighton and 2. Nehari have isolated a crucial part of the standard

Lemma 1.4(a) provides a method for proving oscillation theorems,

(1.5) h’ = q + (l/r)/22 on (a, co).

(b) If, in addition, y’(x) > 0 on [a, co) then Jz q is bounded on

Theorem 1.2. If Jz (l/r) = CO, q 3 0 but $0 for large x, (E,) is

Y&?J = 0, Y&) > 0 on [a, x,> and (c~~)~+ (c~~)~= I,

where y%(x) = cl%(x) + c~%(x) and U, w is a given fundamental set

Theorem 1.3. If (E,) is disconjugute on an interval I then there exists

By combining Lemma 1.6 with Theorem 1.2 we have:

Lemma 1.8. If Jz I/Y = co, q > 0 and J-z q = co then every

This is a weak form of the Leighton-Wintner oscillation theorem. About

Definition 1.2. A second-order operator L, , or the corresponding

Although disconjugacy is an extreme case of nonoscillation they are

Lemma 1.9. If L, , OY (E,), is nonoscillatory on [a, co) then there is

Although it is well known that nonoscillation on [a, co) implies

(ii) is disconjugate if k < 4 and oscillatory if k > 4 on [I, co).

Lemma 1.10. If Jz(l/r) < CL)and Jz 1q 1 < CO then equation (E,) is

(1.5) h’ = q + k h2 on [b, cc).

h(x) > g(x) = 1: (l/r) h2 > 0 on [c, 00).

g’ > (l19g2 and $(1/r) <g&

This proof for the following Leighton-Wintner Oscillation Theorem

Theorem 1.4. If sz l/r = COand J-z q = COthen the equation (E,) is

(1.6) t = [I; wy)]-1

yields the following corollary of Theorem 1.4.

Corollary 1.4.1. If Jc (l/r) < co but

(1.7) ,; P(S) [I; wy)]2 A = *

then (E2) is oscillatory on [a, CO).

In the case of (Ea), with Y = 1 and 4 3 0, Hille (58) achieved better

I &)I = 4.r) 3 j,” q and

Theorem 1.5. If Jz (l/r) = co, q(x) 3 0 but +O for large x on

Hence, lim supsGr sz (1/Y) s: CJ> 1 is sufficient for oscillation of (E,).

1.4. The Priifey Transformation

\ y(“) = P(X)sin e(x),

is called a Priifer Transformation, after H. Priifer who introduced the

(1.10) ~(4 = Z/Y"(X) + (D,Y(x))~ > 0.

Next normalize y and D,y by letting