Mathematica Bohemica

Rudolf Výborný
Some applications of Kurzweil-Henstock integration

Mathematica Bohemica, Vol. 118 (1993), No. 4, 425–441

Persistent URL: http://dml.cz/dmlcz/126154

Terms of use:
© Institute of Mathematics AS CR, 1993

Institute of Mathematics of the Czech Academy of Sciences provides access to digitized documents
strictly for personal use. Each copy of any part of this document must contain these Terms of use.

This document has been digitized, optimized for electronic delivery and
stamped with digital signature within the project DML-CZ: The Czech Digital
Mathematics Library http://dml.cz
118(1993) MATHEMATICA BOHEMICA No. 4, 425-441

SOME APPLICATIONS OF KURZWEIL-HENSTOCK INTEGRATION

RUDOLF VÝBORNÝ,1 Kennmore

(Received March 15, 1993)

Summary. Applications of ideas from Kurzweil-Henstock integration to elementary anal-

ysis on R, mean value theorems for vector valued functions, PHospital rule, theorems of
Taylor type and path independence of line integrals are discussed.
Keywords: Perron integral, KH-integral, fundamental theorem, PHospital, Taylor, differ-
entiation of series.
A MS classification: 26A39, 26A24

1. INTRODUCTION

One of the main features of the Perron integral is that it integrates every derivative
without any restriction. It seems that this property was not fully exploited e.g. in
teaching of analysis, because of the nonelementary character of Perron's definition.
In 1957 Kurzweil [K], in connection with research in differential equations, gave an
elementary definition equivalent to the Perron one, moreover the proof of the fun-
damental theorem became then extremely simple. For KurzweiVs own presentation
of the theory see [Kl]. Henstock later [H] independently rediscovered Kurzweil*s
approach and advanced it further [Hl-4]. It is the aim of this paper to give sev-
eral applications of these ideas in elementary analysis. Only the most rudimentary
knowledge of K-H integration is needed for understanding of this article.

1
This work was partly done while working at UBC and supported by a NSERC grant.

425
 2. NOTATION, BASIC FACTS

A partition of a compact interval [a, b] is a set of couples ((*, J*) such that the
points (* € [a,t], the intervals Ik are non-overlapping and

(1) l J / * = [o,6].
1

The set of couples (&, Ik) satisfying all the above properties except (1) is called a
subpartition. A partition is, of course, also a subpartition. A subpartition with the
additional property that £* € h is a P-subpartition. We shall be dealing only with
P-subpartitions and shall omit the qualifying letter P. Often it will be convenient to
have the intervals, /* =- [ufc,t>*], ordered hence for a partition II 5 {&, [ft*, v*]} we
have
II S a = Ui ^ & ^ Vi = U2 ^ 6 ^ ^2 < • • • < in ^ *n - &.

U6: [a, 6] .—• (0, oo) then a partition II for which

Si - 6(Si) < m ^ & ^ Vi < ^ + 6(Si)

for all j with 1 ^ t ^ n is called a 6-fine partition of [a, 6]. It is obvious what
we mean by a 6-fine subpartition. The set of alltf-finepartitions will be denoted by
9*(6). Hence, instead of saying that II istf-fine,we write II € &(6). For any positive
function 6 and any compact interval [a, 6] a 6-fine partition of [a, b] always exists. The
existence and the use of j-fine partitions has been traced by Mawhin [Ma] to Cousin
in the last century. We shall refer to the statement guaranteeing the existence of a 8-
fine partition for any positive function 6 as to Cousin's lemma. The contrast between
the early discovery of Cousin's lemma and rather late arrival of Kurzweil-Henstock
integral is a bit surprising. As Hadamard once said: "In mathematics simple ideas
come late."

3. REAL LINE APPLICATIONS

It has been shown, for instance in [VI], that £-fine partitions can be used advan-
tageously in teaching analysis on the real line.
As a a first example we prove the Bolzano-Cauchy convergence principle. The
usual argument can be used to show that a Cauchy sequence must be bounded.
Next we show that if n H-> xn is a Cauchy sequence and x is not a limit of n .-• xn
then there exists 6 > 0 such that the interval (x — 6, x + 6) contains only a finitely
many xn. Indeed, there exists a subsequence ib *-+ xnk and a positive £o such that

428
\x - Xnk\ ^ So for all k. Since n i-> xn is Cauchy there is an integer iV such that
\xn - x m | < eo/2 for n > N and m> N. Consequently, for some xnh with n* > N
and all n > N we have |x — xn\ ^ |a? — x n J — \xn% —xn\> eo/2 = 6. Assume now,
contrary to what we want to prove, that no a: € [—K, K] is a limit of our sequence
and that the interval [-K,K] contains all xn. By Cousin's lemma there exists a Wine
partition of [—K,K]. This partition has finitely many intervals and each of them
has in turn only finitely many xn. This contradicts the fact that [—K, K] contains
all xn.
Cousin's lemma has been shown to be equivalent to the least upper bound axiom
and has been used in the proofs of the following theorems: Bolzano-Weierstrass, in-
termediate value theorem, uniform continuity, uniform approximation of continuous
functions by piecewise linear functions, Weierstrass' theorems on boundedness and
extreme values of continuous functions and others. We shall not repeat these proofs
here but refer to articles by Botsko [Bo], [Bol], Bullen [B] and the author [VI]. Be-
sides all this there is a host of theorems asserting something about the increment of
a function from some information concerning the derivative. The simplest example
is: f > 0 =-> / increasing, a more sophisticated example is: / absolutely continuous,
g increasing and | / ' | ^ g a.e. implying |/(6) - f(a)\ ^ g(b) - g(a). We shall prove a
most general theorem in this direction but we need some definitions first. We shall
say that F: [a, 6] «—• C varies negligibly on a set M C [a, 6] if for every e > 0 there
exists a 6: [a, 6] H-> (0,OO) such that for every 6-fine subpartition II, with all & €E M
we have

(2) j2\F(v<)-FM\<e-
n.
Examples of F varying negligibly on M are:
(1) M countable and F continuous on Af;
(2) M of measure zero and F absolutely continuous;
(3) M = [a,fJ] C (a, 6) and F constant on M and continuous at or and /?.
We shall say that F has the strong Lusin property if it varies negligibly on every set
of measure zero. Instead of saying that F varies negligbly on M or that F has the
strong Lusin property we shall say that F € VNM or F € SL, respectively.

Theorem. Substitute for the Cauchy Mean Value Theorem. IfF: [a, 6] H->
R varies negligibly on M, g is increasing and

(3) \F'(x)\^g'(x)torxtM

then
(4) \F(b)-F(a)\<g(b)-g(a).

427
 Corollary. Inequality (3) implies (4) if
a. F has the strong Lusin property and M is of measure zero;
b. F is absolutely continuous and M is of measure zero;
c. M is countable and F is continuous on M.

P r o o f of t h e t h e o r e m . For every e > 0 and every £ £ M there is a

positive 6 such that

(5) \F(v)-F(u)\<([F'(0\ + e)(v-u)

and

(6) g'(0(v - ti) < g(v) - g(u) + e(t> - u),

for
(-*<tt<£0<( + «.
By (3), (5) and (6) we have

\F(v) - F(u)\ < g(v) - g(u) + 2e(v - u).

We define 5 on M by using the fact that F G VNM then we have (2). If now
II € &{$) we obtain

\F(b) - F(a)\ = £ |F(nO - F(u,)t ^ £ |F(*,) - F(«,)|

t=l C.€A#

+ £ w».) - *•(«.)..< £ !*(•») - *(«<)+2e<u« - u <)i+ e

tiiM (if*
<g(b)-g(a) + e[2(b-a) + l]

m
 4. A MEAN VALUE THEOREM

The Mean Value Theorem is, generally speaking, not valid for complex or vector-
valued functions. There are two types of valid generalizations. In the first the
equation f(b)-f(a) = / ' ( c ) ( 6 - a ) is replaced by an inequality, for a complex-valued
/ this inequality reads \f(b) - f(a)\ ^ \f'(c)\(b - a). This was already known in
1876 to Darboux and was extended to linear normed spaces by Aziz and Diaz. The
assertion in the second generalization states that the quotient [f(b) — f(a)]/(b — a)
lies in the convex hull of the range of the derivative. For linear normed spaces this
theorem is due to Wazewski but for complex valued function goes back as far as
Weierstrass. Both generalizations can be established by the same method used in
the proof of the Cauchy mean value theorem below. For sake of simplicity we restrict
our attention to a normed space X, denote the norm by | • | and extend the notion of
a vector valued F varying negligibly on M by interpreting | . | in (2) as the norm. If
A C X then cl_4 and co_4 denotes the topological closure and the convex hull of A,
respectively. The clco A stands for closed convex hull of A which is both the closure
of co _4 and the smallest closed convex set containing A.

The Cauchy Mean Value Theorem. If

(a) F:[a,b]~X;
(b) g is strictly increasing on [a, b];
(c) there is a set M such that F'(x) and g,(x) exist with §'(x) £ 0 for x £ M,
moreover [a,b]\M ^ 0;
(d) both F and geVNM;
(e) S> = {F*(z)tf(z);xe[a,b]\M}
then
F(b)-F(a) Є clcoS).
9(Ъ)-9(a)

R e m a r k . A statement similar to the Corollary holds, in particular, the theorem

remains valid if (d) is replaced by
(d ; ) both F and g are continuous on [a, b] and M is countabie.

P r o o f . Choose S € [a,&]\ M. Let e > 0. By using (d) we define 6 on M in

such a way that

J2\F(vi)-F(ui)\<ey(b)-g(a)]
n.
v) F'C^\ _

•<S)• Ek(«o - *(«oi < em -»(«))

n,
4m
for any 6-fine subpartition II, for which all £ € M. It follows from the definition of
derivative by simple calculation that for every £ g M there exists a positive 6 such
that

(8) F(v)-F(u)-^Q(g(v)-g(u)) < є(9(v) - »(«))•

for

Let II e &(6) and define

z7 - SГ Г^ (W
6)gK)-g(«.) , ^(5) v g(vi)-g(ui)
*- Ъ~ж *(&) g(b)-g(a) g'P) /-< g(b)-g(a) •

Clearly Zn G coO. We complete the proof by showing that Zn comes arbitrarily

close to (F(b) - F(a))/(g(b) - g(a)). Firstly we have

F(b)-F(a)= £ F(vi)-F(щ)+ £ F(vf) - F(щ)

CiЄAf UІM

and then by (8)

Y, (HVÍ) - HUÍ) - J§^^(VÍ) - </(«.))) | < e (g(b) - g(a))

This together with (7) gives

F(b)-F(a)
-Z„ <3є.
g(b)-g(a)
D
C o m m e n t s , c o u n t e r e x am p i e s . The theorem can be extended to locally
convex topological spaces but not much further. There is an example of a function
F and a linear topological space X such that F: [a, 6] .-• X, the derivative F' equals
zero everywhere on [a, 6] and F is not constant. See [R] or [Ya, p. 14].
Let

^{JШ^^У
It is a direct consequence of the above Cauchy Mean Value theorem that Q C clco 3)
and hence clcoO C clco 2). On the other hand 2) C clfl C clco.Q, consequently
clco 2) C clco £2. Hence we have:

430
 Corollary to CMVT. clcoS) = clcofl.
If X is finite dimensional then there is a linear variety of smallest possible dimen-
sion containing coS), call it H. Then (co©)°, the interior of coS) relative to /f, is
not empty and we show that

9(b) - ff(a)
Assume now, contrary to what we want to prove, that q is not in (coD)°. Then
there exists a supporting linear variety at q, i.e. there is a linear functional p and a
real a such that p(q) — a and p(x) ^ a for all iGcoS). However, there must be at
least one interval [u,v] such that

,F(v)-F(u).

otherwise H would not be minimal. Now we have2

p )~ (F(u)-F(á)\
a
«*-«. -
p{q) g(°)-g(
-wrw) \9(u)-9(a)J
9(v)~ g(u) (F(v)-F(u)\
P
g(b)-g(a) \g(v)-g(u)J
g(b)-g(v)jF(b)-F(v)\
g(b)-g(a)P\g(b)-g(v))<a'
a contradiction.
If X is not finite dimensional then q need not lie in coD, the closure is essential.
An example showing this is again in [Ya]. The CMVT can be further generalized,
any limiting value of (F(v) - F(u))/(g(v) - g(u)) can be used in the definition of ©
instead of F'(Z)/g'(£)y see e.g. [Mel]. Unfortunately, our method does not seems to
be easily adaptable to such an extension.

5. FUNDAMENTAL THEOREM

In Kurzweil-Henstock theory the formula

F' = F(b) - F(a)

/fa:
holds for a continuous F if
(i) the derivative F' exists except possibly a countable set M\
(ii) the derivative F' exists almost everywhere on [a, 6] and F G SL.
1
If u == a or b SB v the undefined terms should be omitted
431
There are other conditions under which the equation above holds but these two suffice
for our purposes. Of course, if (i) holds then so does (ii), we stated (i) because of its
frequent use.

5.1. Differentiation of series. The general theorem on term by term differen-

tiation of series is usually considered to be out of reach of an average undergraduate.
For instance, E. Landau—who certainly would not exaggerate difficulty—says in his
book [La] that the proof of the theorem on differentiation of series term by term is
one of the most difficult in the whole book. We are going to give a simple proof of
the following

Theorem. Let us assume that

(A) the condition (ii) is satisfied with F replaced by Fn} for every n 6 N;
(B) for some c £ [a, b]9 the sequence n »-> Fn(c) converges, say to F(c);
(C) the sequence n *-• Fn converges uniformly almost everywhere to g} say.
Then the sequence n •-* Fn converges uniformly on [a, 6] and the limit function F is
differentiate at every x for which lim Fn(x) exists and then F'(x) = g(x).
n—foo
Remark. The most important case is, of course, when lim F' exists every-
n—»-oo

where on (a, 6); the function F is also differentiate everywhere on (a, 6).
Proof. Firstly
Fn(«) = F(c) + jTfi.
By uniform convergence
F(x) = F(c) + jXg.
Moreover the convergence of Fn to F is clearly uniform. Assume now the existence
of
lim Fn(x).
n—>oo
For a positive e there is a natural n such that Fn(x) exists and

1 ft+h
гs+л
\Fn(x)-g(x)\<e and <є.
Ц (g(t)-Fn(t))dt
Since

F(x + h) - F(x) ._ 1 +h
Ѓ
h
-9І*)±ïJ (g(t)-Fn(t))dt
fi( łt R(<)
+ ' |- - F'(x) + F'(x) - g(x),

432
we obtain

F(x + h) - F(x) Fn(x + h)-Fn(x)

-g(x) <2є + -K(*)
By the choice of x and n there exists a positive S such that the last term is less than
e for 0 < \h\ < 6. Consequently

F(x + h)- F(x)

-g(x) <3e.

5.2. L'Hospital rule. This rule is not valid for complex valued functions. For a
counterexample define

f(x) = y/x, gi(x) = y/x + xsinx *, g^(x) — y/x+ xcosx - ì

and have

lim/(x) = 0,
arjO
lim(gi(x) + ig2(x)) = 0i
j?io

lim • , x # л«) __
i
*A0 0i(s) + *02(-c) l +i
lim „ {'(*) „ x
= 0,
*|o 0i'(x) + i$f2'(*)

Using the Fundamental Theorem, we prove below a theorem which states the ad­
ditional conditions under which ^Hospital's rule remains valid for complex valued
functions.

T h e o r e m . If /, g: (0, b) t-» C are continuous,

\imf(x) = \img(x) = 0,
ar|0 arj.0

the inequality g(x) ^ 0 hoick for all x € (0,6), the derivatives / ' and #' exist on [0,6]
except a countable set M,

(9) r J? 1^(01^
JV
^
hmsup , •;• . , — = A < oo,
*i0 |jf(*)l
433
and for every sequence xn —• 0 with xn £ M we have

(io) lim 4 r 4 = L e c,
0'(*n)

then
шm.L.
rjo flf(x)

R e m a r k . If Af is empty then (10) can be rephrased in the usual way as

\imf'(x)/g'(x) = L.

P r o o f . It follows from (10) and (9) that for every positive e there is a positive
6 such that for 0 < x < 6

\f'(x)-Lg'(x)\<e\g'(x)\, x$M

and
('\g'(t)Ut
Jo |y v n
— <K+1
W-)l
Consequently, for every x with 0 < x < 6, we have
|J?(Г(Q-if>(«))d.| cff 1^(01 di
'<*>-!
.?(» W«)l ^ lf(«)l < e ( K + 1)-
D
C o m m e n t s . The limit passage x [ 0 can be replaced by any of the following:
ic | o, x | a, a: -• o, x - > o o and a, —* —oo with obvious changes to the theorem.
The Fundamental Theorem was used in the proof of l'Hospital rule first probably by
Huntington [Hu] and advocated by Boas [Bs]. The assumption (9) was introduced
by G. Szabo [S] in case of real and absolutely continuous / and (/. A discussion for
one-sided derivatives is given in [VN]. As with many other theorems in mathematics,
1'HospitaPs rule is wrongly named—the theorem was discovered by Johann Bernoulli
who communicated it in a letter to marquis l'Hospital.

5.3. T h e Taylor Theorem. It is sometimes stated even in very good texts that
the Lagrange and Cauchy form of remainder are more general than the integral form.
This is not as much a statement about the nature of the theorem as it is about the
concept of the integral used in these formulae. Let us denote

T» = /(a) + f'(a)(b -«) + ... + r ( « ) ^ r ~

434
and

(11) Rn+l= f^+^lzJLdt.

Ja n!
The usual Taylor's formula with integral remainder

(12) f(b) = Tn+Rn+l

holds under the mere assumption that / ( n ) is continuous on [a, 6] and / ( n + 1 ) exists
on (a, b) except possibly a countable set. However, the integral (11) ought to be
interpreted as a Kurzweil integral. With our fairly general assumption on / ( n + 1 ) ,
the identity (12) is valid neither with Lebesgue nor with Riemann integral. Since the
derivative need not exist everywhere on (a,b) the Lagrange (or Cauchy) remainder is
not valid either. On the other hand it is possible to prove the mean value theorem of
integral calculus for the KH-integral and use it to deduce the Lagrange (or Cauchy)
remainder under the additional assumption that / ( n + 1 ) exits everywhere on (a, b). It
is therefore fair to say that the integral form of the remainder is more general than
either Cauchy's or Lagrange's. For proofs and details we refer to [T] or [Tl].
Our aim is to estimate the difference

<? - tm T *K») fna + b\ (*-«)"

Assuming continuity of / ( n + 2 ) , Poffald [Po] obtained the formula

for some f £ (a, 6). We give a different (simpler) proof and remove the continuity
assumption. We now have

Modified Taylor's formula. There exists a positive continuous function Kn,

n G N with

(14) fhK - " (6

ZfT+2

such that if / ( n + 1 ) is continuous on [a, 6] and /(*+2) exists on (a, 6) except possibly
a countable set then

(15) Sn+i= / Kn(t)fl«+*\t)dt.

Ja
435
 R e m a r k . Let H denote the Heaviside function, i.e. let H(t) = 0 for t < 0 and
H(t) = 1 for t > 0. Explicit formula can be given for Kny namely

(16) Kn(t) = j^jy \(b - <) n+1 + (b - a)nH (I£±-* - 1 ) ((» + 1)< -, „a - 6)].

Accepting (16) as the definition of Kn, it is easy to verify (14). Equation (16) shows
that Kn is continuous on [a, 6]. By using the formula [H(a — t)(t — a)]' = H(a — tf)
for t:/ a we see that

(17) Kn(t) = l[-(b-tr + H(^-t)(b-ar}.

Clearly,.#£ > 0 on (a, *$£) and KTn(a) = 0, consequently A'„ ^ 0 on [a,6]. If

f(n+2)(t) exists for every t 6 (a, 6) then as a derivative it has the intermediate value
property. Hence we can use the mean value theorem of integral calculus on the
integral in (15) and (14) leads directly to Poffald's result (13). The example of
f(t)-=z\tfl%\ with a = - 1 , 6.= 1, n ±= 1 shows that if /<n+2) does not exist at one
point then formula (13) may fail.
R e m a r k . For the proof we need integration by parts. It follows from the
Fundamental Theorem that the the usual formula

(18) / Fg = F(6)G(6) - F(a)G(a) - / fG

Ja Ja

holds if F and G are continuous on [a, 6], F' = / and G' = g on [a, 6] except a
countable set and one of the integrals in (18) exists.

Proof. We employ the definition of Tn and (12) and obtain

(•9) Snw = ^+1_ [/«> ( ^ ) -,<»>«] < t i ;

and with the use of the fundamental theorem, (11) and (17)

(20) 5n+1 = Rn+i -V^J^fW :- - jT>+l>JC

Since Kn(a) a Kn(b) -== 0, integrating the last integral in (20) by parts completes
the proof. D

436
 6. A LINE INTEGRAL

In this section the letters G and 5 will stand for an open set and an open star-
shaped set in R n , n ^ 2, respectively. We denote by d(M), A(M) and p(M) the
diameter, the two dimensional area and the perimeter3 of M, in that order. A
generic point in Rn will be denoted by (xlyx2>.. . , s n ) , hence for a mapping F:
G >-+ Rn we have F = (F\F2} ...Fn) with F4: G H R 1 . Partial derivatives will
be denoted by subscripts, consequently d/dxiF* = F\. The word path will be used
for a continuous map of bounded variation from an interval in R into Rn. We shall
allow a slight abuse of notation and use the same symbol for a closed path and
its geometrical image (correspondingly oriented). By a line integral f F(x)dx or
n
/ YL Fx(x) d-P* we understand the Kurzweil-Henstock limit of the Riemann sums
l

EE^CvfeWK)-^'^)).
k t.=i

It is a classical result that if F has continuous partial derivatives in S and

(21) fyx) = Ffa)

for all x E S then the line integral is independent of the path and there exists a
function U with Uti(x) = F*(x) for all x £ S. Moreover U is obtained by choosing an
arbitrary point x 0 in 5 and integrating F from XQ to a variable point x along any path
in 5. Our aim in this section is to reduce the assumption of continuous derivatives
to mere differentiabilty of F. Results of this nature can be also obtained by using
the work of Jarnik, Kurzweil, Schwabik, Mawhin and Pfeffer, however proving the
Stokes theorem in sufficient generality is a rather sophisticated matter (perhaps
not quite necessary for this purpose), whereas our approach is fairly simple and
straightforward. We shall need the following generalization of Cousin's lemma: If T
r
is a triangle and T*; k = 1, ..., r are nonoverlapping triangles with \jTk = T, points
l
yk belong to Tk and 6: T •-• (0, oo) then we say that the set {(yk, Tk); k = 1 , 2 , . . . , r}
is a 6-fine partition of T if d(Tk) < 6(yk). For a triangle T € Rn there always exists
a 6-fine partition consisting of triangles similar to T. An indirect proof can be given
which follows the usual pattern of the one-dimensional bisection argument except
that now T would be divided into four similar triangles formed by mid-points of
sides of T. We shall say that assumption & is satisfied in G if F is continuous in G
s
in the elementary geometrical sense

437
and there exists a countable set M such that F is differentiate and satisfies (21) in
G\M.

Lemma. Jf & is satisfied in G and the triangle T C G then

(22) / F(x)dx = 0.
JT

Proof. The elements of M can be enumerated and for wm G M and for

arbitrary e > 0 there exists a 6 > 0 such that

(23) \F(x)-F(wm)\<^

whenever \x — wm\ < 6. For y G S\M there is a S > 0 such that for i = 1 , 2 , . . . , n

(24) \F*(z) - F'(y) -J2F*j(y)(xi - y>)\ < i|» - x\

i=i

whenever \x — y| < 6. Let T C G and {j/*,T;b} a 6-fine partition of T with triangles

similar to T. If 2/* G M then t/j, = ^m for some m and we obtain from (23)

(25) / F(x) dx < ^p(Tk) < ±;p(T).

If t/jb £ M then because of (21)

(26) / J2 {**&)+E^(»)(^ - *0)d**' = °-

JT
* 1=1 i=l

This can be seen most easily by realizing that the integrand in (26) has a 'primitive'
U, where U(x) = £ F*(y)x* + i J Fj.(y)(a^ - y^)(x{ - t/'")- For yk <£ M we get
t=l i,j = l
from (24) and(26) that

m
(27)
\L * < €d(Tk)p(Tk).

The triangles Tk are similar to T, therefore there exists a constant C depending only
on T (and independent of k) such that d(Tk)p(Tk) < CA(Tk) for all k. Consequently
(27) becomes

(28) 1 / F(x)dx <CeA(Tk).

438
Obviously
F x dx
/ F(x)dx = J2 I () >
which finally by (25) and (28) gives

J F(x)dx<e(CA(T) + p(T)).

Theorem. Path independence. If 0 is satisfied in S then there is a function

U: 5 •-• R such that
n
(29) dU(x) = J2fi(x)dxi
t=i

for all x in S. If <p: [a, 6] »-* S is a path lying in S then

(30) f F(x)dx = U(<p(b))-U(<p(a)).

P r o o f . It suffices to prove (29), equation (30) then follows by an argument

similar to that one used in the proof of the Fundamental Theorem in K-H theory. To
see this choose e > 0. For every £ 6 [a, 6] there is a positive 6 such that if \z — £| < 6
then

\u(<p(z)) - u(<p(0) - it -'"(rtOX^to - *>'(0)| < -M*).- v(0)l-

1

Consequently, if£ — 5 < u . ^ £ ^ t ; < £ - M then

n
(31) W(<p(v)) - U(<p(u)) - VJ F'(rtO)(«>'(tO " ¥>'(«))! < e Var <p.
1

For II a 6-fine partition of [a, 6] we obtain with the help of (31)

\U(<p(b)) - U(<p(a)) - S E ^ ^ * ) ) ^ ^ * ) ~ *>>*))• < * V * r ^

n I

This establishes the implications (29)=>(30). We denote by l(x) the path whose
geometrical image joins the centre of the star-shaped region 5 with x and define

U(x)= [ F(z)dz
JiM
439
It follows from the Lemma that

U(x + h) - U(x) = ( F(z)dz,

Jtp

where %l>(t) = x + th, with 0 •$ t .^ 1, Routine continuity argument now gives (29).
•
G e n e r a l i z a t i o n s . If ^>i and <p2 are two paths homotopic with fixed ends in
G and assumption & is satisfied in G then

f F(z) áz = i F(z) áz

This can be proved in two different ways. Firstly the integral f F(x)dx is inde-
pendent of the path locally, i.e. in some neighborhood of every point in G. Using
this one can employ the usual homotopy argument (see e.g. [C]) to obtain the result
in the large. Alternatively, one can use Cousin's lemma (with squares rather than
triangles) for the homotopy square and proceed similarly as in the Lemma. This
would necessitate the use of a differentiable (compare [V2]) homotopy, the general
result must then be obtained by an approximation argument.

R e m a r k . The assumption Of can be weakened without essentially changing the

proofs. One needs to assume the differentiability on two-dimensional planes only,
the modified assumption reads as follows: We say that weak-0 is satisfied in G if F
is continuous in G and for every two-dimensional plane P there exists a countable
set Mp such that for every x G GC\(P\Mp) there exists a symmetric n by n matrix
[o,- j] with the following property: Given e > 0 and x G P fl G there is a positive 6
such that for every y and z in P with \x — y\ < 8 and \x — z\ < 8 we have

\í < e\y - z\ max(|:p - y|, \x - z\).

The lemma and the theorem on path independence remain valid if the assumption &
is replaced by weak-.^. This allows the set where F is not differentiable or equation
(21) is not satisfied to be uncountable. It is an interesting problem to determine how
big (say in measure theoretic terms) this set can be and still have the theorem on
path independence valid.

440
 References

[B] P.S. Bullen: Some applications of paгtitioning, Real Analysis Exch. 9 (1983-1984),
539-557.
[Bs] R.P. Boas: ĽHôpital rule without mean-value theorems, American Math. Monthly 76
(1969), 1051-1053.
[Bo] M. V. Botsko: The use of full covers in Real Analysis, AMM 96 (1989), 328-333.
[Bol] Michael W. Botsko: A Unifìed Treatment of Vaгious Theorems in Elementery Analysis.,
The American Mathematical Monthly 94 (1987), 450-453.
[C] John B. Conway: Functions of one comlex variable, Springeг, New York, Heidelbeгg,
Berlin, 1975.
[He] R.Henstock: Definitions of Riemann type of the variational integrals, Proc. London
Math. Soc. (3),11 (1961), 401-418.
[Hel] R. Henstock: Theory of Integration, Butterworths, London, 1963.
[He2] R. Henstock: Lineaг Analysis, Butterwoгths, London, 1967.
[HeЗ] R. Henstock: Lectures on the Theory of Integration, Woгld Scientifìc, Singapore, 1988.
[He4] R.Henstock: A Riemann Integral of Lebesgue Power, Canad. J. Math. 50(1968), 79-87.
[Hu] Huntigton E. U.: Simplified proof of ĽHôpitaľs theorem on ideterminate forms, Bulletin
of Amer. Math. Soc. 59(1923), 207.
[K] J. Kurzweil: Generalized Ordinary Differential Equations, Czechoslovak Math. Jour.
7(82) (1957), 418-446.
[Kl] J. Kurzweil: Nichtabsolut Konvergente Integrale, Teubner, Leipzig, 1980.
[La] E.G. Landau: Differential and Integral Calculus, Chelsea, New York, 1960.
[Ma] J. Mawhin: Introduction à ľAnalyse, Зrd ed., Cabay, Louvain-la-Neuve, 1983.
[Mcl] Robert M. McLeod: Mean Value Theorems foг Vector Valued Functions, Proc. of Edin.
Math. Soc. Ц (1964-5), 197-209.
[Po] E.I. Poffald: The Remainder in Tayloťs Formula, Ameгican Math. MontЫy. 97(1990),
205-213.
[R] 5. Rolewicz: Uwagi o funkcjach o pochodnej zero, Wiadomosci Matematyczne 3 (1959),
127-128.
[S] G. Szabó: A note on ĽHôpitaľs rule, Eiemente der Mathematik 44 (1989), 150-153.
[T] H.B. Thompson: Taylor's Theorem with the integral гemainder under very weak differ-
entiable assumptions, The Gazette of the Australian Math. Soc. 12 (1985), 1-6.
[Tl] H.B. Thompson: Taylor's Theorem using the Generalized Riemann integral, Ameгican
Mathematical Monthly 96(1989), 346-350.
[Ya] Sadayuki Yamamuro: Differential Calculus in Topological Lineaг Spaces., Springeг-Lec-
ture Notes in Mathematics, Berlin, Heidelberg, New Yoгk, 1974, pp. 14.
[Vl] R. Výborný: Remarks on elementary analysis, Normat 59 (1981), 72-74.
[V2] R. Výborný: On the use of a diffeгentiable homotopy in the pгoof of the Cauchy theorem,
American Math. Monthly 86(1979), 380-382.
[VN] R. Výborný. R. Nester: ĽHôpital rule, a counteгexample, Elemente der Matћematik 44
(1989), 116-121.

Authorfs address: 15 Rialanna St., Kenmoгe, Queensland, Australia 4069.

441