JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 17, 292-308 (1967)

Inequalities involving integrals of Functions

and Their Derivatives

DONALD C. BENSON

Department of Mathematics, University of California, Davis, California

Submitted by R. J. Dufin

1. INTRODUOTI~N

Inequalities of the sort to be discussed in this paper are considered in

Chapter 7 of Hardy, et al. [l]. Th ere these problems are considered prin-
cipally from the point of view of the calculus of variations, and many impor-
tant inequalities are obtained there in this way. In [I] there are a few examples
of the use of elementary identities to prove such inequalities. The purpose
of this paper is to uncover systematically elementary identities which yield
such inequalities. A few general forms for inequalities will be developed
which, by specialization, yield many of the inequalities in Chapter 7 of [l],
generalizations of these inequalities, and other inequalities, most of which are
believed to be new. It is hoped that it will be demonstrated that the discovery
of many such inequalities can be reduced to a more or less routine technique
and that the degree of difficulty is not much greater than the technique of
integration in elementary calculus. The method to be followed is roughly
this:

Method

(A) Use an “algebraic” inequality to find that a certain expression in-

volving the function and its derivatives is non-negative. For example, the
square of something real is non-negative.
(B) To both sides of this inequality add something which is “exact,”
i.e., a derivative, or (in the case of several independent variables) a divergence.
(C) Integrate. The integral of the exact expression yields boundary terms.
An inequality results.

This paper includes a geometrical application. An improvement of

Wirtinger’s inequality leads to an improvement of the isoperimetric inequa-
lity.
292
 INEQUALITIES 293

2. INEQUALITIES INVOLVING ut2

WIRTINGER'S INEQUALITY AND RELATED RESULTS

In the following it will always be assumed, unless there is an explicit

remark to the contrary, that u(x) belongs to the class C’ of continuously
differentiable functions. The domain of definition will be clear from the
context.
Let P(u, X) and G(u, X) be continuously differentiable for x in [a, b] and u
in the range of the function U(X). (These conditions are perhaps unnecessarily
restrictive, but we shall not be concerned here with the weakest possible
hypotheses.) It is further assumed P(u, x) > 0. We apply the method as
follows

(4 P(u’ - GJ’-‘)2 2 0

w P(u’ - G,J’-‘)2 + 22/G, + 2G, > $ (2G)

(C) j-” (Pu’” + P-lG,,2 + 2G,) dx > 2G(u(b), b) - 2G(u(a), a) (1)

a

Equality holds if and only if the differential equation u’ = G,F is satisfied.

Inequality (1) has many interesting applications by specializing the func-
tions P and G.

2.1 Generalization of Wirtinger’s Inequality

Let u be periodic of period L. Let m = inf U(X) and M = sup u(x), and let
x1 and x2 be such that 0 < x2 - xi <L and u(xr) = m, u(x2) = M. In (1)
put P(x) = 1 and

G(u, x) = f j ((M - u) (u - m))li2 du.

We have, using the positive sign,

s
; (u’~ + (M - u) (u - m)) dx 2 2 lM ((M - u) (u - m))1i2 du
m

= T(M - m)2 (2)

4 *
294 BENSON

Using the negative sign one obtains

21+L

121 (~‘2 + (M - u) (u - m)) dx 3 - 2 j 1 ((M - u) (u - rn))li2 du

- n(M - m)” (3)

4 *
Putting (2) and (3) together we have

j: (*Q- (24- VT) dx Z (yr (277-L) (4)

with equality only if u’a = (M - u) (u - m). We leave it to the reader to

show that this implies u(x) = A sin (x + CX)+ B for suitable choice of A, B
and 01,and hence that L = 2x.
If we specialize (4) by assuming that L = 2a we obtain

i:” (~‘2 - (u - v)‘) dx > 0. (5)

The further assumption s,“” u dx = 0 leads to

The inequality which results if we replace the right-hand side of (6) by zero
is known as Wirtinger’s inequality.

2.1.1 Problem
Use the methods of 2.1 to show the inequality:

II@ + 1)UP + 1)
I ;(zP + (A4- u)w (m - 24)“) dx 2 4(M - m>e#+l r(cr + B + 2) ,

where u is subject to the same conditions as in 2.1.

For what functions does equality occur ?

2.2 A Sharpening of the Isoperimetric Inequality

We use a standard proof [I, p. 1861 of the isoperimetric inequality, but in
place of Wirtinger’s inequality we use (6). The result which is obtained is
the following.
 INEQUALITIES 295

Consider a smooth simple closed curve C with perimeter L and area A.

Let P be the centroid and let & and 8, be a pair of parallel support lines of C
and &‘, parallel to and midway between /I and 4, . (See Fig. 1.) Let d be the
distance from P to la . Then we have

FIG. 1

L2/4~ > A + rrd2.

Thus the area becomes augmented by an amount not less than the area of a
circle of radius d if C is replaced by a circle of circumference L.

2.3 Example
We shall obtain a well-known inequality which is closely related to Wir-
tinger’s inequality by means of two quite different choices for G in (1).
Let u(x) be continuously diferentiable in [0,42] and supposeu(O) = 0. Then
n/2
I 0 @‘2 - u2) dx a O (7)

with equality only if u(x) = A sin x.

40911712-7
296 HENSOK

FIRST PROOF. Use (1) with P = 1 and G(u, x) = + j (K” - $)1/a &
where K = sup j u(x) / .

SECOND PROOF. Use (1) with P = 1 and G(u, x) = 3 u2 cot x. (This

function does not have a singularity at x = 0, because u(O) = 0 and u’(O)
exists.)
It is easy to see that the above result remains valid if the interval [0,42] is
replaced by [0, z-1and the hypothesis u(r) = 0 is added.

2.4 A Useful Special case of (1)

Applications to Eigenvalue Problems

In (1) put P 5 1 and G(u, x) .= 4 Sg(x), where g(x) is assumed continu-

ously differentiable, to obtain a useful generalization of (7):

sb

a W2 + (g’(x) + g(x)“) u”) dx 2 W2 g(b) - u(a)” g(a). (8)

Equality holds only if u’ = g(x) u. If one wishes to obtain an inequality for

s b (uf2 - hp(x) u”) dx,

a
P(X) 2 0,

it becomes necessary to solve the Riccati equation

g’(x) + g(x)” = - hP(X). (9)

One obtains such solutions most conveniently in the form v’(x)/v(x) where
v(x) is a solution of the second order equation

v” + Xp(x) v = 0. (10)

Let A, be the smallest eigenvalue of (lo), with the boundary condition

v(a) = v(b) = 0. Then v(x) f 0 for a < x < b and we putg(x) = v’(x)/v(x).
This function has singularities only at x = a and x = b because v(x)
vanishes only at the end points. But for any continuously differentiable func-
tion u(x) such that u(a) = u(b) = 0, the function u(x)“g(x) is continuously
differentiable and we see that (8) remains valid. We have, for any such u(x)
 INEQUALITIES 297

with equality only if u’ = ug(x), i.e., if u’/u = o’iv. Thus, equality in (11)
occurs only if u is a constant multiple of the eigenfunction belonging to A, .
Thus we have demonstrated Rayleigh’s variational characterization of A, .

2.4.1 Generalization of Weyl’s Inequality

In (8) we may replace g(x) by ,ug(x) where p is any constant. Consider the
discriminant of the resulting quadratic expression in p. We have

(Sbag’u2 dx + u(a)“g(a) - u(b)2g(b)j2 < 4 j” g2u2dx j” z12dx.

a a
(12)

If we put g(x) = x, a = 0 and let b tend to infinity, we obtain Weyl’s inequal-

ity

(j;Pdx)a <4 j;xQ”dx j;u’“dx,

provided that all the integrals exist. Here use is made of the fact that

pi u(b)2b = 0.

which can be shown from the existence of the integrals. (See [2, p. 3931.)

2.5 Probletns
1. Let p(x) be non-negative and continuously differentiable on [a, b]. Let
u and g satisfy the conditions set forth in 2.4. Show that we have

s (1 (P(X) u’~ + (P(X) g(x)” + (P(X) g(4)‘) u”) dx

> u(W2 ?(b) g(b) - 44” P(a) g(a), (13)

and that equality obtains if and only if the differential equation
u’(x) = g(x) u(x) is satisfied.
2. Show that if a < b and u E C’ then for any real A,

dx > + h(e-Abu(b)2- e-%(a)“).

Show that the coefficient of us cannot be replaced by a smaller number.

Solution. Use (13) with p(x) = e-A= and g(x) = h/2.

298 BENSON

3. Show that if u(O) = 0 and h > 0, b 3 0 then

j;exp
[-gy (u’~ - Xu2) dx > 0.

Show that the coefficient of u2 cannot be replaced by a smaller number.

Solution. Use (13) with g(x) = x-r and p(x) = exp (- hx2/2).
4. Show that if u(O) = 0 then

712
cos x(u’2 - 2u2) dx 2 0,
s0

with equality only if EC= A sin x.

Solution. Use (13) with p(x) = cos x and g(x) = cot x.
5. Show that if u(O) = 0, then

-I ((1 - x”) 24’2- 2u2) dx >, 0

i 0

with equality only if u = Ax.

Solution. Use (13) with p(x) = 1 - x2 and g(x) = l/x.
6. Show that if sup u(x) 2 1 and inf u(x) < 0, u(x) continuously dif-
ferentiable on (- co, co), then

co
u’2 + u2(1 - u)” dx > 5,
I -co
and equality holds only if u(x) = (1 + At+l, where A > 0.
Solution. Use (1) with P(x) = 1 and G(u, x) = (u2/2) - (g/3).
7. Show that if sup u(x) > n, inf u(x) < - W,then

sym
(u’2
+ cos4u) dx > 2a

with equality only if u = arc tan (x + A) for any constant A.

8. [ 1, Theorem 2621 If u(O) = u( 1) = 0, show that
1 u2 1

I x(1 -
0 x) I 0
d2 dx

with equality only if u = Ax(1 - x),

 INEQUALITIES 299

Solution. Use (8) with g(x) = (l/x) - (l/l - x). (The singularities are
harmless.)
9. Show that, if u(0) = 0, N > 0, then

N WV
S(0 up2-$)dx>- CW

with equality only if u(x) = Axliz.

Solution. Use (8) with g(x) = (2x)-l.

N
10. lim inf d2 -“)& 2 0, (u(0) = 0).
N-CO S(0 4x2

The constant $ cannot be replaced by a larger number. See [l, Theorem

2531.
11. Show that for any constant h, we have

with equality only if u = AeAs.

Solution. Use (8) with g(x) = h.

12. [l, Theorem 2631 Show that if lim,, u(x) = 0 then

with equality only if u = Ae-A2 for h > 0.

Solution. Consider the discriminant of the quadratic function of X in

the preceding problem.

3. INEQUALITIES INVOLVING HIGHER POWERS OF u’

A number of inequalities are based on the simple algebraic fact that the
polynomial qan(x) = x2* - 2nx + 2n - 1 is non-negative for all x for any
positive integer n. Further, q2Jx) has a double zero at x = 1 and no other
real zeros.
300 BENSON

Let us assume that U(X), P(u, x) and G( u, Y) are continuously differentiable,

just as in Section 2, and P(u, x) > 0. Because q&x) 3 0, we have

pU’Pn + (2fi - 1) P-1’(2n-1)G~‘t2n-1) + 2nG, > 2n(u’G, + G,)

= 2n & G(u(x), x).

Integrating both sides we have

sba(PUIBn+ (zn - 1)p-1’(2n-1)G~‘(zn-1)

+ ‘&t-3,) ,&
3 24W4), b)- G(44,4), (14)

where equality obtains only if the differential equation

*I = Gl’ (2dp-11(2n-1)
u

is satisfied. One sees that (14) is a generalization of (1). We shall study a few
special cases of (14).
Putting in (14), P = 1 and

G@, x) = “2”r-1 ,

where g(x) is continuously differentiable on [a, b], we obtain a generalization

of (8):

s b
a (u’~~ + (2n - 1) g(x)2n-2 (g(x)” + g’(x)) uzn) dx

> U(b)2”g(b)2”-1 - u(u)2n g(u)-1 (15)

with equality if and only if U’ = z&x).

If we put P = 1 and G(u, x) = Jf(~)~+l du we obtain

I1 (ZP + (2n - l)f(~)~“) dx > 2n Jl:yjf(~)2*1 du (16)

with equality if and only if U’ =f(u).

 INEQUALITIES 301

3.1 Problems

1. Generalization of [I, Theorem 2531. Let u be continuously differentia-

ble in the open interval (a, 6) where a = - cc or b = + co is allowed. Also
suppose

li:Pf !!@? > lim sup !QF.

@Z-l X-W+ #Z-l

Then

Show that equality is achieved only if

u = AXWn-l)/2n

and either 0 < a or 0 > b.

REMARK. The boundary conditions here are different from those in

[l, Theorem 2531 because with those narrower conditions an extremal
function does not exist.

Solution. Use (15) with g(x) = (1 - (1/2n)) x-l.

2. Generalization of (4). Let u be in C’ and periodic with period L.
Let m = inf u and M = sup u. Show that for any positive integer n we have

I uf2%+ (2n -
0
1) (u - m)n (M - u)” dx >, Tn(2n)!
(n!)” 24n-2
(M - m)2n

with equality if and only if u = A sin (X + B) + C for arbitrary choice of

A, B, C.

Solution. Use method of 2.1 but use (16) instead of (l), putting

f(u) = (u - rn)l12 (M - u)li2.

3. The following result, a generalization of (7), differs from [I, Theorem

2561, only by a trivial change in the variable of integration. The author of
the present paper felt particularly challenged to seek an elementary solution
because of the remark [l, p. 1831 concerning the possible difficulty of such a
proof.
Let u(O) = 0. Show that for any u E C’, and any positive integer IZ we have

s4(u
,I
‘2fi - (271- 1) u2”) dx > 0,
302 BENSOK

where

A, = ’ (1 - uzn)-l/zn du = 2; csc $ .
0

Equality obtains only if a constant multiple of u satisfies the differential

equation
fpn z 1 - uzn*

(Solutions of this equation are periodic and have period equal to 4A,.)
Solution. Let B = sup 1u(x) 1. Use (16) with

f(u) = (B2” - ZpyP.

We obtain immediately

j; (u'zn + p _ 1) (p - U2n>) dx a 2n jr (B2n - u2n)(2n-1)l(2n)du.

The result follows now from the observation that the right-hand side is
equal to (2n - 1) BanA, . This is seen by the following integration by parts:

I : (B27i _ u2n)l-W2n) du = j: (B2n - u2'9-1/2n (2~2 - 1) 242fldu

= _ (zn _ 1) j; (B2* - ~2n)l-(l/W du

+ (2n - 1) B2n jr (B2” - u2n)-1/2ndu,

hence

2,, 1 (p” - u291-WW du = (& - 1) B2n j; (Ban - zP)--lPn du

= (2n - 1) B2+‘A,.

Equality holds only if u’2n = B2n - uzn. Reflecting on the meaning of B,

we see that this equation is homogeneous! Thus u is a constant multiple
of a solution of the equation u12* = 1 - u2n, with the side conditions u(O) = 0
and sup 1 u 1 = 1. The solution, unique apart from sign, does indeed
satisfy sup (1 u(x) 1 : 0 < x < A,} = 1 as can be seen from the definition

A,,
=sl(1- $9~1Pfl
,,&. 0
 INEQUALITIES 303

4. Another generalization of (4). Let u, M, m, L satisfy conditions of

problem 2. Further put

A, = ’ (1 - zP)--1/2n du = & csc 2 .

s0

Then we have
L

S(ZP -
0
(272- 1) (u - V)p3 dx

>, 2-2n(2n - 1) (M - m)2n(4A, -L).

Equality holds only if a multiple of u is a solution of the differential equation

u’an=l-uan , and4A n =L *
Solution. Use method of 2.1. Use (16) with

M ; m)2n_ tu _ (M ; m)an)llzn
f(u) = ((-

together with identities used in the proof of Problem 3.

The problem is simplified if one observes that if u satisfies the inequality
then Bu + C satisfies the inequality for any constants B and C. Thus it
can be assumed with no loss of generality that M = + 1 and m = - 1.

3.2 An Inequality Similar to Weyl’s Inequality

Let us use (15) with g(x) = 01x,a = 0, n = 2. We have
b
~‘4 + 3u4(opx4+ ar3x2)dx > c?P~(b)~.
I 0

Now let

, B = 3 b u4x2 dx - b%~(b)~
s0
and

C= b up4dx.
s0

Thus we have 4(m) = Aor + Bar3 + C > 0 for all 01.By examining the critical
points of 4 we find a terrace point at 01= 0 and absolute minimum at
a0 = - (3B)/(4A). The condition $(LY~)> 0 yields:
304 BENSON

or

In order to have equality, we must have u’ -= uxu for some 01,which gives
u = A exp (~3/2), the same class of functions for which equality holds in
Weyl’s inequality.

4. INEQUALITIES INVOLVING d

Throughout this section it will be assumed that u belongs to C2, the class
of twice continuously differentiable functions. Let P(u’, II, x) and G(u’, u, X)
be continuously differentiable and P > 0.
Apply the method as follows:

6% P(u” - Gu,P-1)2 > 0

(B) Pu”~ + GuW-l + 2u’G, + 2G, > 2(u”G,~ + u’Gu + GE)

= 2 & G(u’(x), u(x), x)

CC) s b (Pu”” + G,f2P-l + 2u’G, + 2G,) dx

; 2G(u’(b), u(b),
6)- u(a),
a). 2G(u’(a), (17)

For equality we must have u” = G,*P-l.

4.1 Probkm
1. [l, Theorem 2601 Suppose

lj*T u(x) = pi u’(x) = 0.

Show

1iF-3&f - u’2 + u”) dx > 0,

s

and show that equality is achieved only if

u = Ae-tx sin (x sin y) (Y = + 4.

 INEQUALITIES 305

Solution. Use (17) with P E 1 and G = - + (u + u’)~. This is the second

proof given in [l] p. 189.
2. [l, Theorem 2671 If

gi u(x) = pi u’(x) = 0

show that we have

lim&f IN (UQ + 224’2+ 24”)dx 3 4 u(O)2

0

with equality only if u = ce+(x + 2).

Solution. Use (17) with P = 1 and G = - zL2 - uu’ - ua. We obtain

directly

lip &f + 2~4’~+ u”) dx > BUD + 2u(O) u’(0) + 2~‘(0)~

+

For equality we must have

u” = Gut = - zu’ - u and u’(0) + s u(O) = 0,

which implies u = Ce-5(x + 2).

3. Suppose u(a) u’(a) = u(b) u’(b) = 0. Show that

i ab(u”2+ u2 - 2~‘~) dx > 0,

and that equality is achieved only if u = A sin (x + B) and b - a is a mul-

tiple of 42.

Solution. Use (17) with G = - uu’.

REMARK. Certain inequalities due to Fan et aZ. [3] can be obtained by

combining the results of the preceding problem with well-known Wirtinger-
type inequalities (Sections 2.1 and 2.3). In particular
3a. If u(O) = u(z) = 0, then

?I
UC2
dx
> II
s I u2 dx,
0 0

with equality only if u = A sin x.

306 BENSON

3b. If
-71
u’(0) = u’(Tr) = 0, and udx = 0,
J0
then
m 6
u”= dx > 1.2dx,
i 0 I 0

with equality only if u = A cos x.

4. Suppose u(x) is positive twice continuously differentiable on the open
interval (a, b). It is allowed that a = - co or b = + co. Further suppose

lizrif 3 > lim sup 3

x-w

lirn_tsaup-fw3 < co
u(x)
Then

u- - -’ -“‘) dx > 0.
9 u2
.8-b-

Equality holds only if u = A(x - B)3/2 where either B < a or B 2 b.

Solution. Use (17) with P E 1 and G = uf3/9u.
5. Let G(u’, u, x) be continuously differentiable and u E P. Show

+ (2n - 1) PWVn-l) + 2nGUu’ + 2nG,) dx

I
2 24W’(b), u(b),b)- (-W(a),44,4 (18)
with equality only if u” = G:!‘2”-1’. Find a generalization of (18) for higher
derivatives.
6. Show
b
U “2n - (2~2- 1) A&u’ - A,u)~‘+~ dx
a
>, (u’(b) - A,u(~))~” - ($(a) - Anu(4)2”

where A, = (2n - 1)/2n.

 INEQUALITIES 307

7. (Cf. [l, Theorem 2641.) Suppose u’(x) >, 0,O < x < b. Let u’(O) = 0,
u(O) > 0, u(b) < 2n/(4n - 1) and u’(b) 3 1. Then

b 2n
I unZndx > 4n _
0
1

Equality is achieved only if

2n
and U(X) = ((2n -
‘=2n-1

Solution. Use problem 5 ‘with

G(u’, u, x) = (A) (u’)(~‘-/~~ -

Various other inequalities of this type may be obtained if it is observed

that the left-hand side is unchanged by the addition of a linear function.

5. INEQUALITIES FOR FUNCTIONS OF SEVERAL INDEPENDENT VARIABLES

Many of the inequalities of the preceding sections may be generalized to

the case of several independent variables.

5.1. Generalization of (1). Let D be a domain in E, with sufficient regu-

larity so that the divergence theorem can be applied. Let u(xr ,..., x,J = u(x)
be real valued continuously differentiable on ij, the closure of D. Let G(u, x)
be continuously differentiable into E, for x in D and u in the range of u(x).
Let P(u, x) be a real valued, positive, differentiable function. Let V denote
differentiation with respect to x; let aD be the boundary of D; and let n be
the unit outward normal of D.
We apply the method as follows:

(4 P~vu-P-q~2>0

P) PVU~+P-~IG,~~+~UV.G>,~V.(UG)

(c) 1 (PV~2+P-‘/G,12+2uV.G)dVB2~aDUG.ndA (19)

D

with equality only if Vu = P-G, .

This formula may be extended to surfaces as follows. Let D be a bounded
smooth orientable surface with a piecewise smooth boundary aD. Let
G(u, x) E C’ such that for any u, G(u, x) is a tangent vector field to D. Let
308 BENSOX

U(X) E C’ be real valued and defined on D. Let P(u, x) E C’ be real valued

and positive. Let n be a continuous unit normal on D, and t the unit tangent
to aD in the positive direction relative to 11.Then the surface divergence
theorem gives

sD
(PVu2 + P-l 1G, I2 -+ 2uV . G) dA > 2
i aD
un xG*tds (20)

with equality only if Vu = P-U;, . Here Vu and V . G denote the surface

gradient and surface divergence; Vu2 is Beltrami’s first differential parameter.
If we specialize (19) by putting

G(u, x) = $ u2g(x)
we obtain

f (I VU 12+ u2(v . g + 1g 12))dv 2 j aD

dg . n df

which generalizes an inequality of Beesack [4, p. 4981. For surfaces, of

course, we replace the right-hand side of (21) by

I aD
u2n x g . t ds.

REFERENCFS

1. G. H. HARDY, J. E. LITTLEWOOD, AND G. P~LYA. “Inequalities.” Cambridge

University Press, London, 1951.
2. H. WEYL. “The Theory of Groups and Quantum Mechanics.” Dover, New York
(English translation of second German edition, 1931).
3. K. FAN, OLGA TAUSSKY, AND J. TODD. Discrete analogs of the inequalities of
Wirtinger. Monatsh. Math. 59 (1955), 73-90.
4. P. R. BEESACK.Integral inequalities of the Wirtinger type. Duke /. iVZuth. 25 (1958),
477-498.