A SURVEY IN MEAN VALUE THEOREMS

by

David A. Neuser

A thesis submitted in partial fulfillment

of the requirements for the degree
of
MASTER OF SCIENCE
in
Mathematics

Approved:

UTAH STATE UNIVERSITY

Logan, Utah
1970
 ii

ACKNOWLEDGMENT

I sincerely thank Professor Stanley G. Wayment for his

assistance, encouragement, and confidence.

David A. Neuser
 iii

To "Twig"
 iv

TABLE
OF CONTENTS

Page
ACKNOWLEDGEMENT ii
ABSTRACT V

Chapter
I. INTRODUCTION
. 1
II . MEAN
VALUE
THEOREMS
FORDERIVATIVES 5
III. MEAN
VALUE
THEOREMS
FORINTEGRA
LS 20
IV. OTHER
MEAN
VALUE
THEOREMS 28

BIBLIOGRAPHY 42
VITA . 44
 V

ABSTRACT
A Survey in Mean Value Theorems
by
David A. Neuser, Master of Science
Utah State University, 1970
Major Professor: Dr. Stanley G. Wayment
Department: Mathematics
A variety of new mean value theorems are presented along with
interesting proofs and generalizations of the standard theorems.
Three proofs are given for the ordinary Mean Value Theorem
for derivatives, the third of which is interesting in that it is
independant of Rolle's Theorem. The Second Mean Value Theorem for
derivatives is generalized, with the use of determinants, to three
functions and also generalized in terms of n-th order derivatives.
Observing that under certain conditions the tangent line to the
curve of a differentiable function passes through the initial point,
we find a new type of mean value theorem for derivatives. This
theorem is extended to two functions and later in the paper an integral
analog is given together with integral mean value theorems.
Many new mean value theorems are presented in their respective
settings including theorems for the total variation of a function, the
arc length of the graph of a function, and for vector-valued functions.
A mean value theorem in the complex plane is given in which the dif­
ference quotient is equal to a linear combination of the values of the
 vi

derivative. Using a regular derivative, the ordinary Mean Value

Theorem for derivatives is extended into R n ,n>l.

(so pages)
 CHAPTER I
INTRODUCTION

Early in the study of calculus we learn of the ordinary

mean value theorems for derivatives and integrals, and of their
generalizations. However, there are many other less well known
theorems which can be classified under the general heading of
mean value theorems. In this paper we will present many of these
theorems together with interesting proofs and generalizations of
the standard theorems.
In Chapter II we restrict ourselves to mean value theorems
for derivatives of real-valued functions. We present three proofs
of the ordinary Mean Value Theorem: the first being the standard
proof using Rolle 1 s Theorem, the second using an area function,
determinants, and Rolle 1 s Theorem, and the third using a sliding
interval technique by R. J. Easton and S. G. Wayment [2] which is,
interestingly enough, independent of Rolle 1 s Theorem. This same
technique is then used to prove a theorem by Darboux [ll] showing
that the derivative possesses the intermediate value property. We
will at times refer to this property of the derivative as the
Darboux property.
The Second Mean Value Theorem for derivatives, often referred
to as the Generalized Mean Value Theorem, is presented in a linear
form, thus avoiding any assumptions about non-vanishing terms [l].
Devinatz [l] then generalizes the theorem to three functions with
 2

the use of determinants. Another generalization is given by D. H.

Trahan [14] in terms of n-th order derivatives.
An new mean value theorem for derivatives is given by T. M.
Flett [4] in which he observes that for a differentiable function f
defined on [ a,b], there is a point c in (a,b) at which (c - a)f'(c) =

f(c) - f(a), provided f'(a) = f'(b). D. H. Trahan [13] generalizes

Flett's results by relaxing the condition that f'(a) = f'(b), and then
follows this with an extension of the theorem to two functions.
In Chapter III we consider integral mean value theorems for real­
valued functions. The ordinary Integral Mean Value Theorem is proved
under a stronger hypothesis than is necessary. As a result, the
theorem has an immediate generalization in which the requirement of
continuity is relaxed. K. S. Miller [7] has extended the Integral
Mean Value Theorem in what he refers to as the ''First Mean Value
Theorem" and the "Second Mean Value Theorem." We follow these theorems
with two more general theorems, the proofs of which are relatively
less involved.
Recently S. G. Wayment [16] submitted an integral analog to
Flett's Theorem in which he observed that for a continuous function
f defined on [a,b], there is a point w in (a,b) at which (w - a)f(w) =

J;f(x)dx, provided f(a) = f(b). We include his proof along with an­
other proof for comparison. The additional proof is interesting in
that it uses Flett's Theorem. Wayment's proof could be generalized if
the sum of a continuous function and a function with the intermediate
value property was in turn a function possessing the intermediate
value property. However, we provide a counterexample, the construction
of which was suggested by J. W. Cannon.
 3
In Chapter IV we look at a variety of different mean value
theorems in their respective settings. If f is an absolutely con­
tinuous function defined on [a,b] which is differentiable on (a,b),
then S. G. Wayment has shown [17], with the use of the Lebesgue
integral, that there exists mean value theorems for the total varia­
tion of f over [a,] and the arc length of the graph of f on [a,].
The proofs would also follow using the Riemann integral if, under the
given hypotheses, the absolute value of the derivative was necessarily
Riemann integrable. However, E. W. Hobson [5] provides us with a
counterexample. The theorems are then generalized by relaxing the
condition that f be absolutely continuous.
For a real-valued function f, the total variation of f and the
arc length of the graph of f are quite different. However, for vector­
valued functions it is customary to define the arc length of the
graph of the function to be the total variation [12:. If
we think of the real-valued function f rather as the vector-valued
function given by f:x+(x,f(x)), we find that its total variation and
the arc length of its graph are equivalent. With the above definition
in mind, similar proofs would follow if the norm of the derivative
had the intermediate value property. However, a counterexample is
given. We also provide a counterexample showing that the origin al
theorems cannot be extended to Rn ,n>2.
If we consider the differentiable vector-valued function v(t) =

(t,f(t)) defined on the unit interval [a,b], we find an interesting

vector analog to the ordinary Mean Value Theorem for derivatives in
that there is a point p in (a,b) at which v 1 (p) has the same direction
and magnitude as the vector (b - a,f(b) - f(a)). Under certain
 4

conditions we find similar results with the more general function v(t) =

(x(t),y(t)). However, we find it easy to construct a differentiable

3
vector-valued function v from [a,b] into R whose derivative is dif­
ferent in magnitude and direction from [v(b) - v(a)]/(b - a) at each
point in (a,b) .
For a holomorphic complex-valued function f defined on a connected
open subset G of the complex plane where z EG, z2EG, and A is an arc
1
in G connecting z1 and z2, there may not exist a point ZEA at which
(z - z )f 1 (z) = f(z ) - f(z1). However, McLeod [6] gives us
2 1 2
a mean value theorem in which the difference quotient is equal to a
linear combination of the values of the derivatives.
The main problem in establishing a mean value theorem for functions
defined on R n ,n>l, is that there are many ways of defining the deriva­
tive. Using a regular derivative [9], R. J. Easton and S. G. Wayment
[3] give a mean value theorem with the additional hypothesis that the
function be absolutely continuous with respect to Lebesgue measure.
For notational purposes, the proof is given only for the case n = 2
and uses a sliding 1 1 interval 11 technique similar to that used in the
third proof of the ordinary Mean Value Theorem for derivatives. We
remark that L. Misik [8] gives the same theorem without the added
condition of absolute continuity.
 5

CHAPTER I I
MEAN VALUE THEOREMS FOR DERIVATIVES

For completeness we begin with some elementary results.

Let D be the set of all real-valued functions f which are
defined and continuous on [a,b] and such that f' exists at each point
in (a,b).

Theorem 2. l (Rolle' s Theorem)

If fED and f(a) = f(b), then there exists a point CE(a,b) such
that f'(c) = 0.
Proof. Since fED, there exists points m and M in [a,] such that
f(m)�f(x).2._f(M) for each XE[a,b]. If f(m) = f(M), then f is constant
on [a,b] implying that f'(x) = 0 for each XE(a,b). If f(m) f f(M),
assume f(a)<f(M) and hence ME(a,b). Since fED, f'(M) exists and

f�(M) =lim (1/h) [f(m + h) - f(M)] = lim (l/h) [f(M + h) - f(M)] =

h-�-<l h+O
f�(M).

But f'(M)<O
- and f'(M)>O
- - so it follows that f'(M) = 0.
+
The case f(a)>f(m) is handled similarly.
The following theorem is a statement of the ordinary Mean Value
Theorem for derivatives. We follow it with three proofs, the first
two of which depend on Rolle's Theorem.

Theorem 2.2 (Mean Value Theorem)

If fED, then there exists a point CE(a,b) such that (b - a)f'(c) =

f(b) - f(a).
 6

Proof 1. Let g(x) be the equation of the straight line joining

the points (a,f(a)) and (b,f(b)), that is

g(x) = [f(b) - f(a)J [(x - a)/(b - a)]+ f(a).

Define d(x) = f(x) - g(x). Since fED and gED it follows that
dED. Now d(a) = 0 = d(b) so applying Rolle's Theorem, there exists a
point cE(a,b) such that d' (c) = 0. Thus
d' (c) = f' (c) - g' (c) = f' (c) -[f(b) - f(a)]/(b - a) = 0
and the result follows.
We note in the above proof that d(x) attains either a relative
maximum or a minimum value at the point x = c. Let A, X, and Bbe the
point's (a,f(a)),(_x,f(x)), and (b,f(b)),respectively. Let ABdenote
the line segment joining A and Band Lx denote the line segment joining
X and (x,g(x)). Since the acute angle formed by LX and ABis constant
for each XE(a,b), it follows that the perpendicular distance h(x) from
X to ABis in direct proportion to d(x). Thus by maximizing (or
minimizing) d(x), we maximize (or minimize) h(x) which results in maxi­
mizing (or minimizing) the area of the triangle having verticies at A,
X, and B. This gives us a slightly similar but interesting way of
proving the Mean Value Theorem.
Proof 2. Using the notation in the above paragraph, we find that
the area T(x) of the triangle having verticies at A, X, and Bis given
by T(x) = (l/2)luXvl where u and v are the vectors (x - a, f(x) - f(a),0)
and (b - a, f(b) - f(a), 0), respectively. Now

i j k
uXv = x - a f(x) f(a) 0 = k[(a - b)f(x) - (x - b)f(a) + (x - a)f(b)]
b - a f(b) - f(a) 0
 7

where k is the unit vector (0,0,l). Since T(x).:::_0, we now define a

somewhat similar but less restrictive area function A(x) as

A(x) = ( 1 /2) I k I [(a - b) f(x) - (x - b)f(a) + ( x - a)f(b)].

If we let

f(x) X 1
F(x) = f (a) a 1
f(b) b 1

then A(x) = (l/2)F(x). Since AED and A(a) = 0 = A(b), it follows by

Rolle's Theorem that there exists a point CE(a,b) such that A'(c) = 0.
Thus

f'(c) O
A'(c) = (l/2) f(a) al = (l/2) {(a - b)f'(c) - [f(a) - f(b)]} = 0
f( b) b

and the result follows.

In most standard calculus text, the proof of the ordinary Mean
Value Theorem is dependent on Rolle's Theorem. The third proof of
Theorem 2.2 is dependent on the nested interval theorem rather than
on Rolle's Theorem [2]. We first consider the following lemmas.

Lemma 2.1
Let a,b,c, and d be real numbers with d>O and b>O. If a/b�c/d,
then a/b�(a + c)/(b + d)�c/d.
Proof. Multiplying both sides of our given equality by bd>-0, we
obtain ad<bc. We now add cd to both sides and factor to obtain
 8

(a + c)d.::.(b + d)c and hence (a + c)/(b + d).::. c/d. Similarly, if we

add ab to both sides of the inequality ad.::.bc and factor we obtain
a(b + d).::.b(a + c) which in turn yields a/b.::.(a + c)/(b +d). Therefore,
a/b.::.(a + c)/(b + d).::.c/d.

Lemma 2.2
If f' exists at each point of the open set G and XEG, then f'(x)
exists if and only if

L = lim f(b)-f(a) exists

a+x- b -a
b+x+
Proof. If L exists and b = x, then L = fJx). Similarly, if a_
x, then L = f+(x) and hence f'(x) exists.
If f'(x) exists, then f'(x) = f+(x) = f Jx). Now for a fixed a
and b, we can assume without loss of generality that

f(x)-f(a) < f(b) - f(x).

x -a - b -a

By lemma 2.1 we have that

f(x) - f(a) f(b) -f(a) < f(b) - f(x)

x-a ..'.:. b-a - b-x

and hence

f�(x) -f(a) < L< lim f(b) - f(x)

lim f(x)x-a = f+(X).
= - - b-+x+ b - x
a-+x

Therefore L = f'(x) and L exists.

We now establish some notation for use in the following lemma and
in the third proof of Theorem 2.2. If r = [a,b], then let F(I ) = f(b)
1 1
f(a) and /1 = b - a. Observe that for a.::. x .::_b, if I'= [a,x] and I''=
1
J

[x, b J, then F (I 1) = F(I') + F(I'') and I I1 J = II 'J + JI'' I . A1so, 1et h = jI1I-
 9

Lemma 2.3
If f 1 exists at each point of the open set G and I1 = [a,b]CG,
then there exists a point cd1, such that f1(c) = F(I1)/ r1 I I·
Proof. Divide I1 into tv.10 equal intervals, namely I11 =[a,a + h/2]
and I12 =[a+ h/2,b]. Since F(I1) = F(I11) + F(I12), we can assume
without loss of generality that F(I11)2_(1/2)F(I1)2_F(I12). Define the
auxiliary function

g(t) = f(a + h/2+ t) - f(a+ t)

which is continuous on [O, h/2]. Since g(O) = F(I11)�(l/2)F(I1)2_

F(r12) = g(h/2), it follows from the intermediate value property for

a continuous function that there exists a point t 0s[O,h/2] at which
g(t 0) ( l/2)F(I1). This value of t0 determines the interval I2 =
=

[a+ t0, a+ h/2+ t0 ]CI 1 where F(I2) = ( l/2)F(I1) and I I2 I = (1/2) I I1 / ·

Thus

We proceed inductively to obtain a nested sequence of closed intervals

{I ,} such that

F(I i + 1 ) = F(I;)
I I =llil·
2
and Ii + 1 2

If we let () I. = {c}, then it follows from lemma 2.2 that

i
= 1 1

f1(c) = lim F(Ii) = F( 1)

l

i� ITT ITT
 10

We now turn to the third proof of Theorem 2.2.

Proof 3. In view of lemma 2.3, it will suffice to show that there
exists a closed interval I 0 C(a,b) with the property that F(I 0 );j1 0 j =

F(I 1 )/II 1 I·

Jivide into two equal intervals I . If F(Ill)<(l/2)

ll and 1 1 2
r1
F(I1)<F(I12), then the procedure in the proof of lemma 2.3 yields an
interval 12 = I0(':(a,b) with the desired property. If F(I12)<(1/2)
F(I 1 )<F(Ill), define the auxiliary function g as

g(t) = f(b - t) - f(b - h/2 - t) where tE[O,h/2].

Then following the procedure described in the proof of lemma 2.3, we

again obtain an interval I0c(a,b) with the desired property. If
F(Ill) = (l/2)F(I1) = F(r 1 2), let Ill = I2 and divide 12 into two equal
intervals 121 and I22" If F(I21)<(1/2)F(I2) or F(I21)>( 1 /2)F(I2),
then the procedure in lemma 2.3 gives an interval I3 = I 0 C(a,b) with
the desired property. If F(I21) = (l/2)F(I2) = F(I22), choose 122 =
I 0c(a,b). Therefore, from the above arguments and lemma 2.3, there
exists a point cEI0C(a,b) such that (b - a)f'(c) = f(b) - f(a).
A function f defined on [a,b] is said to have the intermediate
value property provided the close d interval from f(x) to f(y) is con­
tained in the image of the closed interval from x to y for each x and
y in [a,b]. In the following theorem, Darboux [1 1 ] has shown that if
f 1 exists on [a,b], then f' has the intermediate value property. The
proof will use a sliding interval technique [2] similar to that used
in the third proof of Theorem 2.2.
 11

Theorem 2.3 (Theorem of Darboux)

If f 1 exists on [a,b], then f 1 has the intermediate value property.
Proof. Let [x,y] be an arbitrary closed subinterval of [a,b].
Since the result is obvious when f 1 (x) = f 1 (y), we will assume without
loss of generality that f 1 (x)<f 1 (y). For any positive s <[f 1 (y) -
f 1 (x)]/2, there exists an h such that O<h<y - x and

/u(x + h) - f(x)]/h - f 1 (x)/< E and l[f(y) - f(y - h)]/h - f 1 (y)/< E.

If we define F(t) = [f(x + h + t) - f(x + t)]/h, we see that F is con­

tinuous on [0,y - x - h] and thus has the intermediate value property.
By the way s was chosen, F(O)<f 1 (x) + s<f 1 (y) - s<F(y - x - h) and
hence [f 1 (x) + s,f 1 (y) - s]CF([O,y - x - h]). Now for each ts[O,y - x - h]
we can apply the Mean Value Theorem to f on [x + t,x + h + t] and obtain
a t'E(x + t,x + h + t) such that f'(t 1 ) = F(t) and hence F([O,y - x - h])
Cf 1 ([x,y]). Since E can be made arbitrarily small, it follows that
(f 1 (x),f 1 (y))Cf 1 ([x,y]) and thus [f 1 (x),f 1 (y)]Cf 1 ([x,y]).
Theorem 2.2 is a special case of a more general theorem. In con­
trast to the standard form of the theorem, the Second Mean Value Theorem
(or Generalized Mean Value Theorem) is given here without any assump­
tions about non-vanishing terms [l].

Theorem 2.4 (Second Mean Value Theorem)

If fsD and gsD, then there exists a point cs(a,b) such that

g 1 (c) [f(b) - f(a)J = f 1 (c) [g(b) - g(a)J.

 12

Proof. Define

F(x)= [g(b) - g( a)J [f(a) - f(x)J + [g(x) - g(a)] [f(b) - f(a )J.

Since fED and gED, it follows that FED also. Now F(a)= 0 = F(b) so
applying Rolle 1 s Theorem, there exists a point CE(a,b) such that
F 1 (c)=O. Thus

F 1 (c)=g 1 (c) [f(b) - f(a)J - f 1 (c) [g(b) - g(a)J = 0

and the result follows.

We remark that Theorem 2.2 follows as a corollary if g is the
identity function defined on [a,b].
Theorem 2.4 has the following generalization with the use of
determinants [l].

Theorem 2.5
If fED, gED, and hED, then there exists a point CE(a,b) such that
9 1 (c)[f(b)h(a) - f(a)h(b)]= f 1 (c)[g(b)h(a) - g(a)h(b)] + h 1 (c)
[f(b)g(a) - f(a)g(b)].
Proof. Define

f(x) g(x) h(x)

F(x) = f(a) g(a) h(a) where XE [a,b].
f ( b) g(b) h(b)

Since fED, gED, and hED, it follows that FED, Now F(a)= 0 = F(b) so
applying Rolle 1 s Theorem, there exists a point CE(a,b) such that
F 1 (c)= O. Thus
 13
f 1 ( c) g 1 ( c) h 1 ( c)
F 1 (c) = f ( a) g ( a) h ( a ) = 0
f( b) g ( b) h(b)

and the result follows

We note that Theorems 2 .4 and 2 .5 are extensions of the Mean Value
Theorem in terms of the number of functions involved. We now turn our
attention to extending the Second Mean Value Theorem in terms of n - th
order derivatives [14]. It will be convenient in the following theorem
to let the symbol n! ! represent 1!2!3! • • • n!.

Theorem 2 .6
If f and g are continuous on [a,a+ nh], h>0, the n-th deriva­
tives of f and g exist on ( a,a+ nh), and ck =
( -1/[�J , then there
exists a point cE ( a,a+ nh) such that f (n)( c) [c0 g( a)+ • • · + cng ( a+ nh)] =
g (n)(c) [c0 f ( a)+ · • •+ cnf ( a+ nh)].
Proof. Define
X X
2
• X
n -1 g ( x) f ( x)
1 a a an - 1 g ( a) f(a)
. . .
2

¢ ( x) = 1 a+ h (a+ h)2 ( a+ h)n -1 g(a+ h) f ( a+ h)

. . . .
1 a+ nh ( a+ nh) 2 . . . ( a+ nh r -, g ( a+ nh ) f ( a+ n h )

and note that¢ (a+ ih) = 0 for each i = 0,1,2 , • • •,n. Since¢ is
continuous on each [a+ (j - l)h,a+ jh], j = 1,2,3, • • · ,n, and dif­
ferentiable on each (a+ ( j - l)h,a+ jh), it follows by Rolle 1 s
Theorem that in each ( a+ ( j - 1)h,a+ jh) there exists a point b.
such that¢ 1 ( bj) = 0. Now¢ ( 2 ) exists on each [bk,bk k = 1,2 ,
+ 1 J,
• • •,n - 1, and hence¢ 1 is contin uous on each ( bk,bk+ 1). Again
 14

applying Rolle 1 s Theorem we find that in each (b k ,b k +1 ) there exists

a point p k such that ¢ (2 )(pk) =
0. Continuing in this way we arrive
at a unique point cE(a,a+nh) such that n) = 0. That is,
¢( (c)

0 0 0 ·0 g(n)(c) f(n)(c)
·a n-1
2
l a a g(a) f(a)
¢(n)(c) l a+ h .
(a+h)
2
• ·(a+h) n-1 g(a+h) f(a+h)
. .... ..... ..... ... ... = 0.
=

. .
l a+ h a+ hl· .
n ( n • (a+nh) n-1 g(a+nh) f(a+nh)

The minor of f(n)(c) can be calculated by expandin g down its last

column, for the subsequent minors of g(a+ k h),0�k�n, are Vandermonde
determinants. The minor of f(n)(c) is

[hn(n - l)/ 2(n - l)!!J[c0 g(a)+ • • • cn g(a+hn)J.

Since the minor of g(n)(c) is similar, it follows that

equals zero.
We note that Theorem 2.4 follows as an immediate corollary if n =l
and h = b - a.

Corollary 2.1
If f is continuous on [a,a+nh], h>0, the n -th derivatives of
f exists on (a,a+ nh), and c k (- l) k [�), then there exists a point
=

n
cs(a,a+nh) such that h f(n)(c) =
co f(a+nh)+ • • • +cn f(a).
Proof. I n Theorem 2.6, if we let g(x) =
xn/ n!, then c 0 g(a)+
n n n
• +cn g(a+nh ) = [co a +c1 (a+h) + • • • +cn (a+nh) ]/ n !.
 15

The corollary follows from the last equation and the equations below:

1 co + c, + c2 + . . . +
n = 0
C

2 c, + 2c2+ .. . + nc = 0
n
3 c, + 22c2 + .. . 2
+ n e = 0
n
. . .. . . . .. . .
l .. . n - 1e
n c, + 2n- c2 + + n
n
= 0
(n + 1) c, + 2n c 2 + .. . + n
n
C
n
= (- 1) n !
n

The above equations can be generated in the following manner.

n
The binomial expansion of (1 X ) gives

which when evaluated at x = l yields equation 1. By taking the

derivative of i) and multiplying through by x we obtain

which when evaluated at x = 1 yields equation 2. By taking the deriva­

tive of ii) and multiplying through by x we obtain

which when evaluated at x = 1 yields equation 3. Continuing in this

way we obtain the above n + 1 equations.
T. M. Fleet [4] first observed that for a differentiable function
f on [a,b] where f 1 (a) = f 1 (b), that at some point in the interval, the
tangent to the curve at that point passes through the initial point
(a,f(a)). Thus we have the following new type of mean value theorem
for derivatives.
 16

Theorem 2.7 (Flett 1 s (Theorem)

If f 1 exists on [a,b] and f 1 (a) f 1 (b), then there exists a
point cs(a,b) such that (c - a)f 1 (c) f(c) - f(a).
=

Proof. Consider the function g defined by

=

g(X) ( X) - f(a) - f 1 (a), a<X<b

c X - a
0,x = a
=

We note that g is continuous on [a,b] and differentiable on (a,b], and

g 1 (x) - - f(x) - f(a) + f 1 (x) for a<x<b.

2
(x - a) x - a

It will therefore be sufficient to prove that there exists some point

cc(a,b) such that g 1 (c) 0.
If g(b) 0, the result follows immediately by Rolle 1 s Theorem.
=

Suppose then that g(b)>0, so that g 1 (b) g(b)/(b - a)<0. Thus

=

there exists a point x1c(a,b) such that g(x1)>g(b). Since g is con­

= -

tinuous on [a,x1J and g(a)<g(b)<g(x1), there exists a point x2c(a,x1)

such that g(x2) = g(b). Now applying Rolle 1 s Theorem to the function
g on [x2,b], there exists a point cc(x2,b) such that g 1 (c) = 0. A
similar argument applies if g(b)<0 and the proof is completed.
We note that in the cases g(b)>0 and g(b)<0, the existance of
the point c is also guaranteed by Darboux 1 s Theorem. For if we assume
g(b)>0, then there exists a point x0 c(a,b) such that g 1 (x0 )>0, for if
not then g 1 (x)<0 for each xc(a,b). But since g(a) 0, this would
imply that g(b)<0, a contradiction. Now g 1 (x0 )>0>g 1 (b), so by Darboux 1 s
=

Theorem there exists a point cc(x0 ,b) such that 9 1 (c) 0. A similar
argument would apply if g(b)<0.
=
 17

D. H. Trahan [13] has generalized Flett 1 s result in the following

two theorems. We first consider the following lemma.

Lemma 2.4
If f is continuous on [a,b] and f 1 exists on (a,b] , and f'(b)•
[f(b) - f(a)]_:::_0, then there exists a point cs(a,b] such that f 1 (c) = 0.
Proof. If f(a) = f(b), the result follows from Rolle 1 s Theorem.
If f 1 (b) = 0, then choose c = b. If f 1 (b)[f(b) - f(a)] <0, then f
assumes a maximum or minimum value at some point cs(a,b) and f 1 (c) = 0.

Theorem 2.8
If f 1 exists on [a,b] and

then there exists a point cs(a,b] such that (c - a)f 1 (c) = f(c) - f(a).
Proof. Define the auxiliary function

C
(x) - f(a), a<x.:::_b
x _ a
h(x) =
f (a)
1
, x = a

We note that h is continuous on [a,b] and differentiable on (a, b],and

h'(x) (x - a)f (x) - [ (x) - f(a)J

1
=
�
(x - a)

) - f(a)
L J
b - a)f'(b) - [f�b) - f(a)Jl
Since h'(b)[h(b) -h(a)J " 1T �
(b - a) b - a

l f";'(b) _ f(b) - f(a)I f;'(a)-f(b) - f(afl<O,

_ f'(a-;i
�
= -
(b - a) L b - a :Jl: b - a :J-
it follows from lemma 2.4 that there exists a point cs(a,b] such that
h 1 (c) = 0 and hence (c - a)f 1 (c) = f(c) - f(a).
 18

We observe that Flett's Theorem follows as a corollary to

Theorem 2.8. For if (b - a)f'(b) = f(b) - f(a) we consider the function
h as defined in the above proof and note that since f'(a) = f'(b), then
h(a) = h(b). Applying Rolle's Theorem to h on [a,b] we find a point
cs(a,b) such that h'(c) = O and the result follows. On the other hand,
if (b - a)f'(b)tf(b) - f(a), then either i) (b - a)f'(b)<f(b) - f(a) or
ii) (b - a)f'(b)>f(b) - f(a), and we note that cfb. If i) holds , then

I+' (b)
L
_ f(b) - f(a)J
b - a
·E - f(a) >O
(a) _ f(b)b - a J -

and the result follows from Theorem 2.8. A similar argument applies if
ii) holds.
The following corollary is a stronger statement of Flett's
Theorem in that the condition f'(a) = f'(b) is relaxed. The proof
follows directly from Theorem 2.8 and will be deleted.

Corollary 2.2
If f' exists on [a,b] and f'(a) and f'(b) are both less than or
both greater than [f(b) - f(a)]/(b - a), then there exists a point
cs(a,b) such that (c - a)f'(c) = f(c) - f(a).

Theorem 2.9
If f' and g' exist on [a,b], g'(a)to, g(x)tg(a) for all xs(a,b],
and

�:f:l - ml= :f:8§(b) - g(a)J f'(b) - [f(b) _ f(a)J g'(b)J"-o,

then there exists a point cs(a,b] such that [g(c) - g(a)J f'(c) =
[f(c) - f(a)] g'(c).
 19

Proof. Define the auxiliary function

-_ f(a) , a<X<b
g(a)
h (x) =

We note that h is continuous on [a,b] and h' exists on (a,b], and

h, (x) =
[g(x) - g(a)] f'(x) - [f(x) - f(a)J g 1 (x)
2
[g(x) - g(a)J

Since h'(b)[h(b) - h(a)]_::0, it follows from lemma 2 .4 that there exists

a point CE(a,b] such that h'(c) = 0 and hence

[g(c) - g(a)J f'(c) = [f(c) - f(a)] g'(c).

We note that Theorem 2 .8 follows as an immediate corollary if
g(x) = x for each XE[a,b].

Corollary 2 .3
If f' and g' exist on [a,b], g'(a)tO,g(x)tg(a) for each XE(a,b],
g'(b)[g(b) - g(a)J 0, and f'(a)/g'(a) = f'(b)/g'(b), then there exists
a point cs(a,b) such that [g(c) - g(a)J f'(c) = [f(c) - f(a)J g'(c).
Proof. If

f(b) - f(a) _ f'(b),

g(b) - g(a) - g'(b)

define h as in the proof of Theorem 2 .9. Since f'(a)/g'(a) = f'(b)/g'(b),

it follows that h(a) = h(b). Thus by Rolle's Theorem, there exists a
point CE(a,b) such that h'(c) = 0 and the conclusion follows. If the
above equality does not hold, the conclusion follows from Theorem 2 .9
and the fact that g'(b)[g(b) - g(a)]>O.
 20

CHAPTER III
MEAN VALUE THEOREMS FOR INTEGRALS

The integral mean value theorems of introductory calculus are

usually proved under stronger hypotheses than are required. Continuity
is often demanded when only a consequence of continuity is necessary.

Theorem 3.1 (Integral Mean Value Theorem)

If f is continuous on [a,b], then there exists a point ws(a,b) such
·b
that (b - a)f(w) = Jaf(x)dx.
Proof. Consider the inequality

inf f(x).2.n � a f�f(x)dx.:5..sup f(x).

If equality l1olds on either side of this inequality, then f(x) = inf

f(x) a.e. or f(x) = sup f(x) a.e. and thus there exists at least one
point ws(a,b) with the desired property. If strict inequality holds
on both sides of this inequality, then by the intermediate value
property of a continuous function there exists a point ws(a,b) such
that (b - a)f(w) = fabf(x)dx.
We note that continuity is required only to insure that f be
integrable and have the intermediate value property. The function f
2
defined by f(x) = x sin(l/x )for xfo and f(O) = 0 does not have a con­
tinuous derivative on [- l,l] but the derivative does have the inter­
mediate value property and is Riemann integrable. Hence the conditions
of the ordinary Integral Mean Value Theorem are not met. We also note
that there is a function F whose derivative exists at each point in
 21

[a,b] but F' is not Riemann integrable (see example 4.1). In view of
these results, the Integral Mean Value Theorem has the following im­
mediate generalization. The proof is similar to the proof of Theorem
].1 and will be omitted.

Theorem 3.2
If f has the intermediate value property and is integrable
(Riemann or Lebesgue), then there exists a point ws(a,b) such that
(b - a)f(w) = f�f(x)dx.
It is interesting to note that there exists a Lebesgue integrable
function which possesses the intermediate value property and is dis­
continuous at each point. Such a function is given in example 3.1.
We next present a generalization of a theorem due to K. S. Miller
[7] and note that Theorem 3.1 follows as an immediate corollary if
g(x) = l for each xs[a,b].

Theorem 3.3
If f is continuous on [a,b] and g is Riemann integrable on [a,b]
and either g(x)�O or g(x)-2_0 for all xs[a,b], then there exists a
point ws(a,b) such that f(w)f�g(x)dx = f�f(x)g(x)dx.
Proof. We will prove only the case g(x)�O. Let M = sup f(x)
and m = inf f(x), then

mfabg(x)dx.::_ fabf(x)g(x)dx.2_M fabg(x)dx.

If equality holds on either side of the above inequaltiy, then

f(x) = m a.e. or f(x) = M a.e. which implies that there exists at
least one point ws(a,b) such that f(w)f�g(x)dx = f�f(x)g(x)dx. If
 22

strict inequality holds on both sides of the a b ove inequality, then it

follows by the intermediate value property of a continuous function
that there exists a point WE(a, b ) such that f(w)fabg(x)dx = b
faf(x)g(x)dx.
K. S. Miller [7] states the Second Mean Value Theorem with a
stronger hypothesis than is necessary. We therefore follow his result
with a more general theorem.

Theorem 3.4 (Second Mean Value Theorem)

If f is a continuous, monotone increasing function defined on
[a, b ], and g(x)�O is integrable on [a, b ], then there exists a point
WE[a, b ] such that

J�f(x)g(x)dx = f(a)J�g(x)dx + f( b )f�g(x)dx.

Proof. By Theorem 3.4, there exists a point pE(a, b) such that

f�f(x)g(x)dx = f(p)f�g(x)dx. Define ¢(x) = f(x) - f(a) and note that
¢ is a continuous, nonnegative, monotone increasing function on [a, b ].
Now

b
¢(p)fag(x)dx = f(p)fab g(x)dx - f(a)f bg(x)dx,
a
so

f(p)fabg(x)dx =
b b
¢(p)fag(x)dx + f(a)fag(x)dx.

b
If we define G(x) = fX g(x)dx, then G is a continuous, nonnegative,
monotone decreasing function on [a, b]. Since G( b) = 0,

¢(p)f�g(x)dx = ¢(p)G(a) = ¢(p)G(a) + [¢( b) - ¢(p)]G( b).

But ¢(p)�O, ¢( b) - ¢(p)�O, so there exists a num ber a with

G( b).:_ a .:_G(a) such that
 23

¢(p)fabg(x)dx = a {¢(p) + [¢(b) - ¢(p)]} = a¢(b).

By the intermediate value property of a continuous function, it follows

that there exists some point WE[a,b] such that G(w) = a and thus

b
¢(p)fag(x)dx = G(w)¢(b) = [f(b) - f(a)Jfwb g(x)dx.

Therefore,

b b
faf(x)g(x)dx = f(p)fag(x)

= ¢(p)f�g(x)dx + f (a)J�g(x)dx

= [f(b)Jeg(x)dx - f(a)Jeg(w)dx] + [f(a)f:g(x)dx + f(a)feg(x)dx]

= f(a)f:g(x)dx + f(b)feg(x)dx

which completes the proof.

By relaxing the condition that f be a monotone increasing function
on [a,b] we obtain the following more general theorem, the proof of
which is similar to the proof of Theorem 3.4 and thus will be deleted.

Theorem 3.5
If f is continuous on [a,b] and f(a).::_f(x).::_f(b) for each XE[a,b],
and if g(x)�O is integrable on [a,b], then there exists a point
WE[a,b] such that

f�f (x)g(x)dx = f(a)J:g(x)dx + f(b)feg(x)dx.

We now turn our attention to an integral analog to Flett's result

(Theorem 2.5). S. G. Wayment [16] first posed the following theorem
and we include his proof (Proof 2) here along with an additional proof
for comparison.
 24

Theorem 3.6
If f(t) is a continuous function defined on [a,b] and f(a) = f(b),
then there exists a point ws(a,b) such that (w - a)f(w) = f:f(t)dt.
Proof 1. Let F(x) = J;f(t)dt. Since f is continuous, we have that
F is differentiable and F'(x) = f(x) for all xs[a,b]. Now F'(a) =

f(a) = f(b) = F'(b), so applying Flett's result, there exists a point

ws(a,b) such that (w - a)F'(w) = F(w) - F(a). This reduces to
(w - a)f(w) = J:f(t)dt, the desired result.
We remark that Proof 1 uses the full power of continuity to
insure that F'(x) = f(x) for all xs[a,b].
Proof 2. Since f is continuous, there exist points t1 and t2 in
[a,b] such that f(t1).:::_f(x) and f(t2)�f(x) for all xs[a,b]. Now
(t 1 - a)f(t1)2:_ J�1f(t)dt and (t2 - a)f(t2)� J�2f(t)dt. Define
F(x) = J;f(t)dt - (x - a)f(x). Since f is continuous, we have that F
is continuous also. Now F(t 1 ).2_0 and F(t2)�0 so by the intermediate
value property of a continuous function, there exists a point ws(a,b)
such that F(w) = 0 and hence (w - a)f(w) = f:f(t)dt.
Proof 2 can be generalized if F has the intermediate value
property whenever f does. This, in turn, would be implied if the sum
of a continuous function and a function with the intermediate value
property were necessarily a function with the intermediate value
property. The following example shows that such is not the case. The
techniques used in the construction of this example are somewhat
standard in topology and were suggested by J. W. Cannon.
Example 3. 1 . We first generate a sequence {Ci} of Cantor sets in
the following way. Construct c 1 on [0,1] and let L1 be the largest
 25

number for which there exists at least one open interval (a1,b1) in
�c with length L1. Construct c on the closed middle third of
1 2
(a1,b1). Let L2 be the largest number for which there exists at least
one open interval (a2,b2) in �(c 1uc 2 ) with length L2_2.L1. Construct c3
on the closed middle third of (a ,b2). In general, construct en on the
2
closed middle third of the open interval in�(; 9
1 Ci) with length Ln,

where Ln is the largest number for which there exists at least one open
interval in �(i Y 1 Ci) with length Ln-2.Ln _ 1.
For each i, let gi be a continuous, monotone increasing function
which maps Ci onto [0,l]. The construction of such functions can be
found in [10]. Define.
(
\ g (x) if xsCi
g(x) -) i
(o if xs[O,l] - i V l C i.
We note that g has the intermediate value property, for given any
arbitrary open interval (a,b) contained in [0,l], there is an i such
that cic(a,b) and g(Ci) = [O,l].
For a real valued function p(x) defined on [O,l], let IIP/1 = sup
lp(x)I. We next define a sequence {fi } of continuous functions on
[O,l] such that 'f IJf-l/< and hence f = 'f f.1 is continuous on
00

i=l l i=l
[O,l]. The {fi} will be constructed in such a way that (f + g)(x)tl/2
for any value of x in [0,l]. Since g is continuous on c1, there is a
finite collection u1 of open intervals whose closures are pairwise
disjoint which covers c 1 and such that if x and y are in o 1isu1 for
some i, then Jg(x) - g(y)l<s1<l/3. Let v 1 be the collection of those
open intervals in u1 which contain values of x satisfying g(x) = 1/2.
 26

Def i ne f 1 (x)=El for the values of x covered by v 1 , and let f 1 (x)=0

for the values of x wh i ch are covered by u 1 but not covered by v 1 .
Extend f 1 to be cont i nuous on [0,1 ] w i th functional values between
0 and E1. If XEC 1 and i s covered by v 1 , then g(x)>l/2 - E l and hence
(f 1 + g)(x)>l/ 2 . Thus f 1 +g i s cont i nuous on c 1 and (f 1 +g)(x)tl/2
for any xEC1 .Let o 1 be the distance from 1 / 2 to the image of c1
under f1 +g, that is, o1 = p(l/2 ,[f1 +g](c1)). We note that
o1..::_E 1 and choose Ez such that 0<E 2 <o1 /3. Let h2 =f 1 +g on c2 . Since
h2 i s continuous on c2 , there is a fin i te collect i on u 2 of open i nter­
vals whose closures are pa i rw i se disjoint which covers c2 and such that
if x and y are i n o 2 i EU 2 for some i, then lh2 (x) - h2 (y) l<E 2 . Let v 2
be the collection of those open i ntervals in u 2 wh i ch conta i n values
of x satisfying h2 (x) =1/ 2 . Define f2 (x)=Ez for the values of x
covered by v 2 , and let f2 (x)=0 for the values of x which are covered
by u 2 but not covered by v 2 . Extend f2 to be cont i nuous on [0,1 ] with
functional values between 0 and E 2 . If XEC2 and i s covered by v 2 ,
then h2 (x)>l/2 - Ez and hence (f2 +h2 )(x)>l/2 . Thus f2 +h2 i s
continuous on c2 and (f2 +h2 )(x)tl/2 for any XEC2 . Let o2 = p(l/2 ,
[f2 +h2 J(c2 )). We note that o2..::_E 2 and choose E3 such that 0<E3<o2;3.
Let h3=f2 + h2 on c3. We proceed i nductively to define {E;}, {o i }'
{f.},
1
and {h.}
1
such that 6.1 = p(l/2 ,[f.1 +h.](C.)),
1 1
not i ng that
00
o-<E., and 0<E.1 + 1 <o./3. If f= I: f., then s i nce
1
1- 1
i = 1 1
p(l/ 2 ,[f. +h.](C.))= p(l/2 ,[g + ! f.](C.)) = oi and
J J J i=l 1 J
00 i00
r W-11< r o.( 1 /3 )=o./ 2 , it follows by the triangle inequal i ty
i=j+l l i=lJ J
00
that (f+g)(x)tl/2 for any x in C .• If xs -U C . , then
J 1. = 1 1
f(x)2- ! llf- II< ! i
1 ;3 = 1;2 .
i =l l i =l
 27

We remark that it is possible to construct a function f which has

the intermediate value property, is Riemann integrable, and such that
J:f(x) - (x - a)f(x) does not have the intermediate value property.
 28

CHAPTER IV
OTHER MEAN VALUE THEOREMS

For a real-valued fu n ctio n f defin ed on [a, b ], let Labf represent

the arc length of the graph of f o n [a, b ] an d let Vabf represe n t the
total variatio n of f over [a, b ]. I n tuitively, L�f is the total dis­
tan ce a particle would travel alo n g the graph of f from point (a,f(a))
to point ( b ,f( b)), whereas V�f is the total distan ce a particle's
projected image onto the y-axis v�ould travel as the particle itself
moves alo n g the graph of f from poin t (a,f(a)) to poin t ( b ,f( b )).
We n ote that the above relationship betwee n Vab f a n d L ab f is the
exceptio n rather than the rule. For if f:[a, b ]+R n is a rectifia b le
curve, i.e., co n ti n uous a n d of b ounded variation, it is customary to
defin e the arc le n gth of the graph of f to be V�f [1 2 ]. Thus for the
real-valued fu n ctio n f defined on [a, b ], we fin d that V�f = L�f if we
consider f as the mapping from [a, b ] in to R 2 given by f:x+(x,f(x)).
We now tur n our atte n tio n to esta b lishing mea n value theorems
for V�f a n d L�f [15,17].
Let D b e the set of all real -valued fun ctions f which are co n ­
tin uous o n [a, b ] and such that f' exists o n (a, b ), a n d let A b e the set
of all real-valued fu n ctio n s f which are a b solutely co n tin uous on
[a, b ].

Theorem 4.1
If fsAOD, then there exists a poin t ps(a, b) such that V�f =

( b - a)jf'(p)j.
 29

Proof. Since f is a bsolutely continuous, it follows that

f�lf' I = V�f where the integral is the Lebesgue integral. Now

( b - a)infjf'(t)i< <( b - a)supjf'(t)j.

J�lf'l-
ts(a, b ) - ts(a, b )

If equality holds on either side of this inequality, then

if'(t)i = supjf'(t)i a.e. or If'(t)I = infjf 1 (t) I a.e.

ts(a, b ) ts(a, b )

b
Thus Vaf = ( b - a) if'(t)i a.e. If strict inequality holds on both
sides of this inequality, it follows from Dar boux 1 s Theorem (Theorem
2 .3) that there exists a point ps(a, b ) such that V�f = ( b - a)lf'(p)I.

Theorem 4. 2
If fsAOD, then there exists a point ps(a, b ) such that Labf =

(b - a)�l+ lf'(p)l2 ,
.,
2
Proof. We can say that L�f = f� � 1 + If 1
1 provided this
Le besgue integral exists [1 2 ]. Since

it follows that

< J b 11 + /f 1
Vabf - 2
1 < ( b - a)+ v b f.
a'-l 1_ a

Thus for a continuous function we can conclude that V�f exists (that
is, f is of bounded variation) if and only if L�f exists. Since
frAf\D, it follows that L�f exists. By the Dar boux property, f 1 has
2 2
the intermediate value property and hence if' I, jf 1 1 , 1 + jf 1 1 , and
2
�l+ If' 1 ' do also. We note that in general the sum of a continuous
 30

function and a function with the Darboux property need not be a

function with the Darboux property (see example 3.1 ). Now

< (b - a) sup :JI l + jf'(t)I2'.

(b - a)inf --...tI l + jf'(t)I 2' � fab--.JI l + If'I 2'-
ts(a,b) ts(a,o)

If equality holds on either side of this inequality, then either

1
2 2
�l + jf' (t) 1 = inf �l + If'(t) 1 , a.e.
tda, b)

or
2
�1 + if'(t) ' 1
2

= sup �l + If'(t) 1 a.e.

tda,b)

Thus we have that L�f = (b - a)�l + jf'(t)i 2 a.e. If strict in­

equality holds on both sides, then by the Oarboux property for
�l + jf' ', there exists a point pE:(a,b) such that
1
2

J� �l + If I I , - a)�l + I f 1 (p)I 2 ,
2
= (b

Thus this integral exists and we have that L� f = (b - a)�l + jf'(p)j 2•

We remark that the proofs of Theorems 4.1 and 4. 2 would follow
using the Riemann integral provided jf' I is Riemann integrable. Since
jf' I is bounded and if jf' I is Riemann integrable, then jf' I is Lebesgue
integrable. However, example 4.1 provides a function fsA"D such that
If' I is not Riemann integrable [5].
Example 4.1. Let G be a perfect non-dense set of points in the
interval (a,b) and such that its measure is greater than zero. Let
(a, S) be an arbitrary open interval in the complement of G and define
the following function on (a, S):
 31

¢(x,a) = (x - a) 2 sin -­
x - a

Thus¢ 1 (x,a) = 2 (x - a)sin x � a - cos

We note that¢'(x,a) = 0 at an infinite num b er of points in (a, s).

Let a+ A= max {xlx<(l/2)(a + S) and¢'(x,a) = O}. Define the
following function on each component (a, S) of the complement of G:

(x,a) if xda, S) and a < x < a+ A

F(x) = e(a+ )..,a) if a+ )..< x <B - A
¢(x,S) if S - A�X < S

and F(x) = 0 for each XEG. The function F is continuous and has a
bounded derivative on the interval (a, b). If XEG, we note that
F'(x) 0 and that in any a-neigh borhood of x there is an interval in
the complement of G in which there are an infinite num ber of points at
=

which F' is greater than l. Therefore, F' is discontinuous at each

point of G. Since G has positive measure and since F'(x) = IF'(x)I = 0
at each XEG, it follows that IF' I is not Riemann integra ble.
Theorems 4.1 and 4. 2 can b e made somewhat stronger if we relax
the requirement that f be a bsolutely continuous.

Theorem 4.3
If fED and f is of bounded variation on [a, b ], then there exists
a point pE(a, b) such that V�f = ( b - a)/f'(p)/.

Proof. Assume that there does not exist a point pE(a, b) such
that V bf ( b - a)/f 1 (p) I, Since f' has the Darboux property on (a, b)
and there is not point ps(a,b) at which lf 1 (p)/ V�f/( b - a), it
=

follows that either i) /f'(x)/>V�f/( b - a) for each xda,b ), or ii)

=

/f'(x) l<V�f/(b - a) for each XE(a, b).

 32

Suppose i) holds. Let a = x 0 <x1< • • • <xn = b be an arbitrary

partition of [a,b], and let pi be the point in (xi _ 1,xi) guaranteed
by the Mean Value Theorem such that f(xi) - f(x _ 1) = f 1 (p i)6xi = where 6xi
i
xi - xi _ 1. Then taking the suprema over all possible partitions of
[a,b], we find that
n
V�f = sup I /f(xi) - f(xi _ 1) I
.;
I
-- ,
. i

n
= sup I
/f' (pi)/6xi
i =l

which is clearly a contradiction.

Suppose ii) holds. This implies that /f'/ is bounded and hence
f is absolutely continuous. By Theorem 4.1, there exists a point
pE(a,b) such that V�f = (b - a)/f 1 (p) I which contradicts our original
assumption.

Theorem 4.4
If fED and f is of bounded variation on [a,b], then there exists
a point pE(a,b) such that L�f = (b - a)�l + /f'(p)/2 .
Proof. As in the proof of Theorem 4.2, we note that for a
continuous function f, V�f exists if and only if L�f exists. Since
fED and is of bounded variation,we have that L�f exists. Since f 1 has
2
the Darboux property, it follows that�l + /f 1 1 ' does also. If we
now assume that there does not exist a point PE(a,b) at which L�f =
2
(b - a)�l + /f 1 (p)/ , the proof will follow by similar arguments as
used in the proof of Theorem 4.3.
 33

I n view of the discussio n prior to Theorem4. 1, it would seem

clear that similar results to Theorems
4.1 through
4.4could be
o b tai n ed if f is a fun ction o n [a,b] i n to R n such that f is absolutely
conti n uous o n [a,b], f 1 exists on (a,b), an d lif
1
I I has the in termediate
value property. However, the followin g example shows that llf 1 II n eed
n ot have the in termediate value property.
Example4. 2 . For a <O an d b>O, defin e

(x2 sin (l/x),x2 cos(l/x),x) for XE[a,O) U(O,b]

F (x) =

Then

( 2 xsi n (l/x) - cos(l/x),2 xcos(l/x) + sin (l/x),l) for

F (x)=
1
XE[a,O)U(O, b ]
(0,0,l) for x = 0

We o bserve that IIF 1 (0) II= 1, whereas IIF 1 (x) 11 = �4x2 + 2�� for
XE[a,O) U(O,b].
A vector-valued fun ctio n f is said to be absolutely co n ti n uous
if an d o n ly if each of its compon e n ts is a bsolutely co n ti n uous [12 ].
In the fu n ctio n F defin ed in example4. 2 , if we let fi represe n t the
i - th component of F, the n lfi(x) l2_3 a n d he n ce f. is absolutely
I

l -b�
co n tin uous. Thus a llF II= Ja4x + 2
1 2
Vab F. Now�4x2 + 2""7is a
co n ti n uous function on [a, b ] an d thus b y Theorem 3.1, there exists a
poi n t ps(a, b ) at which J��4x2 + 2 � p2 +
= (b - a)4 2 So we have
that V�f = (b - a)�4p2 + 2' = (b - a) IIF 1 (p) II even though /iF 1 11 does
n ot have the intermediate value property o n [a,b].
O n e might ask at this point if f bein g a differe n tiable and
absolutely co n ti n uous fu n ction from [a, b ] into R n is sufficie n t for
 34

there to exist a point pE (a,b) such that V�f = (b - a) IJf 1 (p) 11-

Unfortunately, the answer to this question is no, as the following

example shows.
Example 4.3. Define
(Ex sin(l/x),E x cos(l/x),x), if XE[­ l , 0)
2 2

gE(x) = [(x sin(l/�),x os(l/x,x), if x E(0,l]

2 2
�
(0,0,0), lf X - 0

then

( 2E xsin(l/x) - cos(l/x), 2E xcos(l/x) + sin(l/x),l), if XE[- 1,0)

g (x) = ( 2xsin(l/x) - cos(l/x), 2xcos(l/x) + sin(l/x),l), if xE (0,l]
1

(0,0,l), if X = 0

and

�4x E + E + l if X [- 1,0)
2 2 2

l if X = 0

t:
Let XE [- l,0), then for any 6>0 there exists an E >0 such that

l�jJg�(x)Jr_l + 6. Let F(t) = (xJ/[t - (- l)], then Fis a

1gE
continuous function of t on [0,l]. Now

and

(l) = ( l/ 2) V � gE(x) = ( l/ 2)
F l J� l I I g�
(x) 11 + (l/ 2) f � I I g�(x) 11
2_(l/ 2) J� 1
l + (l / 2 ) J� IT

= l + R
2
Thus for sufficiently small 6>0, there exists a point tE (0,l) such
 35

l + o<F( t)< 1 + 12 < 12 and consequently there cannot exist a

t ha t
2
point ps(- l, t ) at which

V t lg E( X) = ( t + l) Ilg EI ( X) I I .

The ordinary Mean Value Theorem for deriva tives of a real-valued

function f from [a,b] into R guarantees t he existence of a point in
(a,b) a t which t he deriva t ive has t he same slope as t he line joining
t he points (a,f(a)) and (b,f(b)). If we consider f as a vec t or­
valued func tion from [a,b] into R2 , we get t he following interest ing
results.

Theorem 4.5
If v is a differen tiable vec t or-valued func t ion defined on [a,b],
given by v( t ) = ( t ,f( t )) where f is a real-valued func t ion from [a,b]
in t o R, then t here exis ts a poin t ps(a,b) such t ha t v 1 (p) has t he same
direc t ion as t he vector (b - a,f(b) - f(a)).
Proof. Since v is differentiable on [a,b], it follows t hat f is
different iable and t ha t v 1 ( t) = (l,f 1 ( t )). By Theorem 2.2, t here
exist s a point ps(a,b) such t hat f 1 (p) = [f(b) - f(a)]/(b - a). Thus

v 1 (p) = (b - a,f(b) - f(a))·[l/(b - a)].

Since 1/(b - a) is posit ive, it follows t ha t v 1 (p) has t he same direc­

tion as the vector (b - a,f(b) - f(a)).

Corollary 4.1
If b - a = l, t hen v'(p) will also have t he same magni t ude as the
vector (b - a,f(b) - f(a)).
 36

We might ask if similar results can be obtained if v is a more

2
general vector-valued function from [a,b] into R . The answer is yes
if we place a few more restrictions on v.

Theorem 4.6
If v is a differentiable vector-valued function from [a,b] into R 2
given by v(t)= (x(t),y(t)) such that v'(t) is nowhere zero on [a,b]
and v(a)tv(b), then there exists a point pE(a,b) such that v'(p) is
parallel to the vector u= (x(b) - x(a),y(b) - y(a)).
Proof. Since v is differentiable on [a,b], we have that v'(t)=
(x'(t),y'(t)). Also, since v(a)tv(b), it follows that ut(O,O) and
hence either x(b) - x(a)to or y(b) - y(a)to.
Suppose x(b) - x(a)to. Define the auxillary function

(b) - y(a)[x(t) - x(a)J

¢(t)= y(t) - y(a) - yx(b) - x(a)

then¢(a)= 0= ¢(b). Also,¢ is continuous and differentiable on

[a,b] since v is. Thus

¢ 1 (t)= y' (t) - �f�j = �f!j · x'(t)

exists for all tE[a,b]. By Rolle's Theorem, there exists a point
pE(a,b) such that¢ 1 (p)= 0 and hence

y' b) - y(a) • x'(p).

(p)= yx�b) - x(a)
X ( p)
I

Therefore v'(p)= u • -x-(b�)--�x(-a�) which implies that v'(p) and u are

parallel vectors.
Suppose y(b) - y(a)fO. The proof is similar to the proof in the
case above if we define our auxillary function to be
 37

8(t) = x(t) - x(a) x(b) - x(a) [y(t) - y(a)J.

(b) - y(a)

Corollary 4.2
If y'(p) a n d y(b) - y(a), or x' (p) and x(b) - x(a) have the
same sign, then v'(p) has the same direction as u.

Corollary 4.3
If y'(p)= y(b) - y(a) or x '(p) = x(b) - x(a), then v'(p) has
the same magn itude as u.
If we try to ge n eralize the above results to a differentiable
vector-valued function v from [a,b] in to R n , n>2, we fin d that there
need n ot exist a poin t pE(a,b) at which v'(p) has the same direction
or magn itude as the vector [v(b) - v(a)]/(b - a). The following is
an example of such a fu n ction .
Example 4.4. Define v(t)= (cos t, sin t, t) on [0,2IT]. Now
[v(b) - v(a)]/(b - a)= (0,0,1) whereas v'(t)= (- sin t, cos t,l).
In order for these two vectors to be parallel, on e must be a multiple
of the other for some tE[0,2IT] which is clearly impossible.
Sin ce the mag n itude of v' (t)= /2 for all tE[0,2IT] an d the
mag n itude of [v(b) - v(a)]/(b - a) = l, it fol lows that there does not
exist a poin t pE[0,2IT] at which v' (p) and [v(b) - v(a)]/(b - a) have the
same magnitude.
If we consider f as a holomorphic complex-valued function defin ed
on a subset of the complex plane C, we find that a mean value theorem
in the form of the ordin ary Mean Value Theorem (Theorem 2.2) may n ot
always be possible. However, McLeod [6] has shown that for a holo­
morphic function f defined on a con n ected open set GCC where z a n d
1
 38

z2 are poin ts in G such that the segment joining them is also in G,

that

for some p1 and p2 on the segment joining z1 and z2, and some non ­
negative real numbers Al and A2 such that Al + A2 = l. This seems
i ntuitively clear since as z2 gets ''close" to z1,

[f(z2) - f(z1)]/(z2 - z1), f'(p ), f'(p ), and hence

1 2
Alf'(pl) + A2f'(p2) get "close" to f'(zl ).

We note that the linear expression A1f'(p1) + A2f'(p2) cannot

in general be replaced by a vlaue f'(p). For example, if we define
f(z) = e2 an d choose z2 = z1 + 2rri, we fin d that f(z2) - f(z1) = 0
p
whereas (z2 - z1) f' (p) = 2rrie to for any p. However, if we let
Al = 1 /2 = A2 and choose p1 = z1 an d p2 = z1 + rri, we see that McLeod's
result holds.
The main difficulty in establishing a mean value theorem in Rn,n >l,
is that there are many different ways of defining a derivative in
this setting. Before we present a mean value theorem in Rn we first
make the following definitions relative to Rn [9]. The set ICRn is a
closed interval if I {(x1,x2,· • ·,xn ) la.1-<x-<b.1 ,i = 1,2, • • · .J n}, and
=

the set JC Rn is a closed cube if J is a closed interval having equal,

non -zero sides. The diameter of I is given by d(I) = sup{p(x,y)lxsI,ysI}
where p(x,y) represents the distan ce between points x and y, and its
n
Lebesgue measure is given by µ(I)= · IT l (b.1 - a.).
1
A sequence {I.}
1
of
1 =
closed intervals is said to converge to xsRn, den oted I.1 +x, if xsI.1 for
 39

each i, a n d l. im d(I.)
1
= 0. A sequence {I.}
1
is said to be a regular
1-)-00
seque n ce if I.1 +x, and if there is a consta n t a>O, called the parameter
of regularity, such that for each i, there is a cube J1. .:)I. for which
1

µ(Ii )
--> a

µ(Ji) -

In tuitively, a regular seque n ce is one i n which the intervals do not

become too "thi n ." If we let T be a fi n itely additive set function
defined o n at least the closed intervals, the n the upper a n d lower
regular derivatives of T are give n by
'* _T(I ) '* l. T(I.)
f ( X) = sup � im (Ii.) a n d T (x) = i nf 1m
i-)-00
1
µ(I.1 ) '
I 1. +x 1 -)-00 µ 1 I.1 +x

respectively, where the sup and inf are taken over regular seque n ces
'*
which con verge to x. The regular derivative, de n oted T (x), is said
I* I*
to exist at the poi n t x if f (x)
(x).= l
'*
L. Misik has show n [8] that if T exists at each point of a n
i nterval i n Rn , the n a mea n value theorem holds. However, we will give
here a simpler a nd more direct proof of his result, as give n by R. J.
Easto n a n d S. G. Wayme n t [3], using the additional hypothesis that T
be absolutely conti n uous with respect toµ.
For notational purposes, we will restrict the proof of the
following theorem to R2 In this setting, a closed interval becomes
a closed recta n gle and its Lebesgue measure becomes its area. For this
reason, we will use a more suggestive notatio n by letting R = [a,b;c,d] =

{(x,y)lxE[a,b] and yE[c,d]} represent an arbitrary i n terval in R2 and

 40

A(R) represent its area. Also, let h = b - a and k = d - c. We define

the auxiliary function gR (x,y) = T(R), where (x,y) is the midpoin t of R.
We note that since T is absolutely con tinuous with respect toµ, that
gR is a continuous function of x and y. In general, we would define
gI on the closed intervals in Rn to be a function of n-variables. The
following proof will use a sliding interval technique similar to that
used in the third proof of Theorem 2.2.

Theorem 4.7
'*
If T (x) exists at each point w of a closed rectangle R and T is
absolutely continuous with respect toµ, then there exists a poin t
'*
pER such that T (p) = T(R)/A(R).
Proof. Let R1 = R = [a,b;c,d] and divide R1 into four similar
rectangles, namely R11 = [a,a+ h/2;c,c+ k/2], R12 = [a,a+ h/2;d - k/2,d],
R13 = [b - h/2,b;d - k/2,d], and R14 = [b - h/2,b;c,c+ k/2]. Since

T(Rl) = I T(Rli),
i = 1

we can assume without loss of generality that T(R11)�(1/4)T(R1)�T(R13).

The other cases follow in a similar manner. Let a = k/h and define the
auxiliary function

f(t) = gR (a+ t + h/4,c+ at+ k/4)

n
which is continuous on [0,h/2]. T(R11)�(1/4)T(R1)�T(R13) =
Since f(O) =

f(h/2), it follows from the intermediate value property for a continuous

function that there exists a point t0E[0,h/2] such that f(t 0) =
This value of t0 determines a rectangle
 41

where T(R2) = (l/4)T(R1) and A(R2) = (l/4)A(R1).

Thus

T(R )2
1 =-- T(R )
--
A(Rl) A(R2)

We proceed inductively to obtain a nested sequence of closed

rectangles {Ri}, each of which is geometrically similar to R1 and
having sides parallel to R1, such that
A(Ri)
and A(R.1 + 1) =
4

By the nested interval theorem there exists exactly one point

psn Ri, and
i = 1

'* T ( R.) T(Ri)

T (p) = 1 ,m 1
. =--
i-+m A(R.)1
A(R1)

We note that by changing the selection process slightly for R3 and R4,
we can insure psint(R1).
 42

BIBLIOGRAPHY

1. Devinatz, A., Advanced Calculus, Holt, Rinehart andWinston, Inc.,

New York, 1968, pp. 151-152.

2. Easton, R. J. and S. G.Wayment, "The Sliding Interval Technique,"

The American Mathematical Monthly, Vol. 75, October (1968), pp.
886-888.

3. �------��-----' "A Mean Value Theorem." The

American Mathematical Monthly, Vol. 77, February (1970), pP:-170-
172.

4. Flett, T. M., 11 A Mean Value Theorem," Mathematical Gazette, Vol. 42,

1958, pp. 38-39.

5. Hobson, E.W., The Theory of Functions of a Real Variable, Vol. I,

Dover Publications, Inc., New York, 1927, pp. 490-491.

6. McLeod, R. M., "Mean Value Theorems for Vector Valued Functions,"

Proceedings of the Edinburg Mathematics Society, (2), Vol. 14,
1964-65, pp. 197-209.

7. Miller, K. S., Advanced Real Calculus, Harper and Brothers, New

York, 1957, pp. 75-77.

8. Misik, L., "Der Mittelwertsatz fLlr Additive Intervallfunktionen,"

Fundamenta Mathematica, Vol. 45, 1957, pp. 64-70.

9. Munroe, M. E., Introduction to Measure and Integration, Addison­

Wesley Publishing Co., Inc., Reading, Massachusetts, 1953. 212 p.

10. Royden, H. L., Real Analysis, The Macmillan Co., New York, 1968,
pp. 47-48.

11. Taylor, A. E., Advanced Calculus, Ginn and Co., Boston, 1955, p.
107.

12. -=--,--,---=-=--=--,--' General Theory of Functions and Integration,

Blaisdell Publishing Co., New York, 1965, pp. 417-418.

13. Trahan, D. H., 11 A New Type of Mean Value Theorem," 1�athematics

Magazine, Vol. 39, November 1966, pp. 264-268.

14. �-�- ��' "An N - th Order Second Mean Va1ue Theorem," The
American [�athematical Monthly, Vol. 72, March 1965, pp. 300-3m.
 43
15. Valentine, J. E. and S. G. Wayment, 110n a Mean Value Theorem for
a Metric Derivative, 11 Rev. Roum. Math. Pures et Appligues,
January 1970.
16. Wayment, S. G., 11 An Integral Mean Value Theorem, 11 Mathematical
Gazette, (to appear).
17. ______, 11 Arc Length and the Mean Value Theorem, 11 (submitted).
 44

VITA
David A. Neuser
Candidate for the Degree of
Master of Science
Thesis: A Survey in Mean Value Theorems
Major Field: Mathematics
Biographical Information:
Personal Data: Born May 20, 1944, in Eau Claire, Wisconsin.
Education: Graduate of Memorial High School, Eau Claire,
Wisconsin in 1962. Bachelor of Science degree in Mathematics
from Wisconsin State University, Eau Claire, Wisconsin, June,
1966.