Analysis 101:

Funct.<.ol'U> 06 a Single VIVUabte

 ABSTRACr

These notes are a chapter in Real Analysis, While prinarily standard,

tbe reader will find a discussion of certain topics that are ordinarily not

covered in the standard accounts.

My thanks to Judith Clare for a superb job of difficult technical typing.

 FUtcTIOOS Cf ASING£ VAAIAil.E

§0. RADON MEASURES

§ 1. VARIATION OF A FUNCTION
§2. LIMIT AND OSCILLATION
§ 3. FACTS AND EXAMPLES
§4. PROPERTIES
§5. REGULATED FUNCTIONS
§6. POSITIVE AND NEGATIVE
§7. CONTINUITY
§8. ABSOLUTE CONTINUITY I
§9. DINI DERIVATIVES
§10. DIFFERENTIATION
§11. ESTIMATE OF THE IMAGE
§12. ABSOLUTE CONTINUITY I I
§13. MU!. TIPLICITIES
§14. LOWER SEMICONTINUITY
§15. FUNCTIONAL ANALYSI S
§16. DUALITY
§17. INTEGRAL MEANS
§18. ESSENTIAL VARIATION
§19.. BVC
§20. ABSOLUTE CONTINUITY III

ADPENDIX
Ref: Advanc.ed Ano.f.y1,,U, 011 .tlte Real U11e, R. Kannan aro carole King Krueger,

Springer-Verlag, 1996
 1.

§0. RAVON MEASURES

I.et X be a locally compact Hausdorff space.

1: NOI'ATION C (X) is the set of real valued continuous functions on X

and BC(X) is the set of bounded real valued continuous functions on X.

2: DEFINITION Given f E C (X) , its supp::>rt., denoted spt (f) , is the smallest

closed subset of X outside of which f vanishes, i.e. , the closure of {x: f (x) f 0 },

and f is said to be COII\.'OClCtly supported provided spt (f) is compact.

3: NOTATION C (X) is the subset of C (X) vvh.ose elements are compactly

c
supp::>rted.

4: DEFINITION A function f E C(X) is said to vanish at infinity if

v e > 0, the set

{x: jf (x) I ::> d

is compact.

5: NOI'ATION c0 (X) is the subset of C (X) whose elements vanish at infinity.

7: LEMMA c (X) is the closure of Cc (X) in the unifonn metric:

0

d(f,g) = I If - gl l 00
•

8: DEFINITION A linear functional I :Cc (X) + R is positive if

f ~ 0 => I(f) ~ O.
 2.

9: LEMMA If I is a positive linear functional on C (X) , then for each

c

compact set K c X there is a constant S< :;: 0 such that

for all f E C (X) such that spt (f) c K.

c

10: DEFINITION A Radon measure on X is a Borel measure µ that is finite

on compact sets, outer regular on Borel sets, and inner regular on open sets.

11: EXAMPLE Take X = rf .,. ,_ then the restriction of Lebesgue measure A.

to the Borel sets in X is a Radon measure.

Every Radon measure µ on X gives rise to a positive linear functional on

C (X) , viz. the assignment

c
f + fx f dµ.

And all such arise in this fashion:

12: RIESZ REPRESENTATION THEOREM If I is a rx>sitive linear functional

on Cc (X), then there exists a unique Radon measureµ on X such that

I(f) = fx f dµ
for all f E C (X) •
c

13: EXAMPLE Take X = R and define I by the rule

I(f) = !R f dx (Riemann integral).

Then the Radon measure in this setup per the RRr is the restriction of Lebesgue

measure A. on the line to the Borel sets.

 3.

14 : RAPPEL Cc (X) is a complete top::>logical vector space when equipped

with the inductive top:>logy, i.e., the toI:X>logy of nnifonn convergence on canpact
sets.

15: DEFINITION A distribution of order 0 is a continuous linear functional

T:C (X)
c
+ R.

16: LEMMA A linear functional T:C (X) + R is a distribution of order O

c

iff for each canpa.ct set K c X there is a constant ~ > O such that

for all f E C (X) such that spt(f) c K.

c

Therefore a I:X>Sitive linear fnnction.. I:C (X) + R is a distribution of order O,

c
hence is continuous in the inductive top::>logy.

D:mote the set of distributions of order Oby the symbol v0 •

17: LEMMA v0 is a vector lattice.

If T E V0 , then its Jordan decomp:>sition is given by

T = T
+ .... T
-I

\Nhere

T+(f) = sup T(g)

O<g<f
- .,...
(f) = inf T(g}.
0:5_g:5.f
 4.

Here T+, T E v0 are pJsitive linear functionals and

T=T
+ - T-.

Therefore

T
+ ~-c->1.1
+

(Radon),

T <-> µ

so V f E C (X),
c
+ -
T(f) = fx f dµ - f x f dµ

and
+ -
ITI (f) = Jx f d (µ + µ )•

18: N.B. Both µ+ and µ.,..might have infinite measure, thus in general
their difference is not defined.

19: REMARK As we have seen, the pJsitive linear functionals on Cc (X)

can be identified with the Radon measures. Bearing in mind that c0 (X) is the

uniform closure of Cc(X), the positive linear functionals on c0 (X) can be identified

with the finite Radon measures.

**********

I.et X be a compa.ct Hausdorff spa.ce.

20: N.B. It is clear that in this situation C (X) == C(X).

c

Equip C(X) with the uniform norm:

II f 11 co. = sup If i .
x
 5.

Tnen the pair (C (X) , 11 · 11 00) is a Banach space. Let C (X) * be the dual space of

C (X) , i.e. , the linear space of all continuous linear functionals J\ on C (X) --

then the prescription

11 J\ I I* = inf {M 2: 0 : IA( f) I ~ M I If 11 00 ( f E c (X) ) }

is a nonn on C (X) * under which the pair (C (X) *, 11 · 11 *) is a Banach space.

21: N.B. V f E C(X), VJ\ E C(X)*,

22 : RAPPEL A signed Radon measure is a signed Borel measure µ whose

positive variationµ+ is Ra.don and whose negative variationµ- is Radon.

[Note: As usual, µ = µ+..,... µ"""" is the Jordan decomposition of µand its total

variation, denoted Iµ I , is by definition IµI = µ+ + µ- In addition, µ is finite

if lµI is finite, i.e., if lµI ex> < + oo.J

23: RIESZ REPRESENTATION THEOREM Given a A E C (X) *, there exists a

unique finite signed Radon measure u such that V f E C(X),

A(f) = f x f dµ.

And

II AII * = Iµ I ex> •

24 : NOI'ATION M(X) is the set of finite signed Radon measures on X.

25: LEMMA M(X) is a vector space of R.

26: NOTATION Given µ E M(X) , put

I Iit! IM(X) = Iµ I (X) •

 6.

27: LEMMA 11 ·I IM(X) is a nonn on M(X) under which the pair (M(X), 11 ·I J M(X))

is a Banach space.

28: THEOREM Define an arrow

J\ : M(X) -+ c (X) *
by the rule

J\(µ) (f) = fx f dµ.

Then J\ is an isometric isomorphism.

[E.g.:

IA(µ) (f) I= lfx f dµI

Sr Ix I£ I d IµI < 11 £ 1100 IµI ex>

= I If I Ioo I IµI IMex) ·

Therefore

l\(µ) E C(X)*]

**********

If Xis not compact, then the story for c 0 (x) is the same as that for C(X)

when X is compact. Without stopping to spell it all out, once again the bounded
linear functionals are in a one-to-one correspondence with the finite signed Radon
measures and

I I1\ I I* = Iµ I ex> •
 1.

§1. VARIATION OF A FUNCTIOM

Let [a,b] c R be a closed interval (a < b, - oo < a < b < + oo).

1: DEFINITION A partition of [a,b] is a finite set P = {x , ... ,xn} c [a,b],

0
where

2: NarATION The set of all partitions of [a,b] is denoted by P[a,b].

3; EXAMPLE
{a,b} E P[a,b].

Let (X,d) be a metric space and let f:[a,b]-+ X be a function.

4: DEFINITION Given a partition P E P[a,b]/ put

b n
V (f ;P) = L
. 1
d(f(x.),f(x.
l l -1
)),
a 1=

the variation of f in P.

5: NarATION Put

b
Tf [a,b} = sup V (f ;P),
PEP[a,b] a

the total variation off in [a,b].

6: N.B. Here, (X,d) is implicit •.••

One can then develop the basics at this level of generality but we shall
 2.

instead specialize irrmediately and take

X = R, d(x,y) = jx - YI,
thus now f: [a,b] + R. later on, we shall deal with the situation when the da.uain
[a,.b] is replaced by the open interval ]a,b[ (or in principle, by any nonempty
open set st c R (recall that such an st can be written as an at most countable
union of pairwise disjoint open intervals), e.g. st = R). As for the range, we
shall stick with R for the time being but will eventually consider matters when

R is replaced by RM (M = 1,2, ••• ) (curve theory).

 1.

§2. LIMIT ANV OSCILLATION

Let f: [a,b] + R.

1: DEFINITION Given a closErl subinterval I = [x,y] c [a,b], put

v(f;I) = lf(y) - f(x) I,

the variation of f in I.

2: DEFINITION Given a J?Clrtition P E P[a,b], put

b n
V (f;P) = k 1£(x.)
1
- f(x. 1 ) I
a i=l i-

n
= l:. v (f; I . ) (I.= [x. ,x.]),
. 1 1 1 l.-1 1
1=

the variation of f in P.

3: NOI1ATION Put

b
Tf[a,b] = sup V (f;P),
pEP[a,b] a

the total variation of f in [a,b].

4: DEFINITION A function f: [a, b] + R is of bormded. variation in [a, b]

provided

Tf [a,b] < + 00
•

5: NOTATION BV[a,b] is the set of functions of bounded variation in [a,b].

 2.

6: EXAMPLE Take [a, b] = [ 0, l] and def.ine f: [ 0, l] -+ R by the rule

0 if x is irrational
f (x) =
1 if x is rational.

Then f ~ BV[O,l].

7: NarATION Given P E P[a,b], put

I IPI I= rnax(x. J_
- x.1-1 ) (i = l, ... ,n).

8: THEOREM Let f E BV [a, b] . Assume: f is continuous -- then

b
Tf [a,b] = l.im V (f ;P) •
jjPll -+Oa

[Note: The continuity assumption is essential. E.g. , take [a, b] = [..... 1, + l]

and consider f (O) = 1, f (x) = 0 (x ~ 0) .]

Let f:[a,b] -+ R.

9: DEFINITION Given a closed subinterval I = [x,y] c [a,b], denote by M

and m the supremum and inf imum of f in I and put

osc(f;I) = M -·m,

the oscillation of f in I.

[Note: Since the diameter of f (I) is the supremum of the distances between

pairs of points of f(I), it follows that

M - m = diam f (I)

or still,

osc(f;I) =diam f (I).

 3.

And, of course,

v(f;I) ~diam f (I).]

Let
n
v(f;[a,b]) = sup E osc(f;I.).
PEP[a,b] i=l 1.

10: THEOREM

Tf[a,b] = v(f; [a,b]).

PROOF It is obvious that

Tf[a,b] ~ v(f; [a,b]).

To go the other way, fix s > 0. Choose a partition P of [a,b] such that if
!J..
l
= osc(f;I.),
l
then
n
CT= L. /J..
i=l l

-1
is greater than v(f; [a,b]) - s or s according to whether v(f; [a,b])< + 00
or

v(f; [a,b]) = + 00
• To deal with the first p:::>ssibility, note that in each interval

I. = [x.i - ,x.] there are two p:::>ints ~! , l;'.' with

1. 1 l l 1.

If ( ~ '.' > - f ( ~ ! > I > 11. - ~ •

i i 1. n

The p:::>ints ~ ! , ~'.' divide I.1. into one or two or three subintervals. Call
l l

the partition of [a, b] thereby determined -- then the sum (i) L: If (y . ) ~ (y . , 1 > I ( [v. l' YJ· ]
J J-· -] ....

..,.. -s
n
. Therefore

m
L 1£(yJ.) - f (yJ.-l) I
j~l
 4.

n
= L (i)Llf(yj) - f(yj-1) I
i=l

n
> L (LL - --)
s
1 n
i=l

n s n
= L: /1. - - E 1
1
i=l n i=l

=0 - c

> v(f; [a,b]) ~c -s,

from which

Tf[a,b] ~ v(f; [a,b]).

 1.

§3. FACTS ANV EXAMPLES

1: FACT Suppose that f E BV[a,b] -- then f is rounded. on [a,b].

[Given a ~ x ::: b, write

If (x) I= lf(x) - f (a) + f(a) I

< lf(x) - f(a)I + lf(a) I
< If (x) ~ f (a) I+ If (b) -- f (x) I+ If (a) I
~ Tf[a,b] + If (a) I < + oo.}

2: FACr A function f: [a,b] + R constant iff Tf [a,b] = O.

[A constant function certainly has the stated property. Conversely, if f is

not constant on [a,b], then the claim is that Tf[a,b] ':f 0. Thus choose

x1 ':f x 2 E [a,b] such that f(x1 ) ':f f(x2 ), say x 1 < x 2 -- then

=>

3: FACT If f: [a,b] + R is increasing, then f E BV[a,b] and

Tf[a,b] = f(b) - f(a).

[If P = {x0 , ••• ,xn} is a partition of [a,b], then

b n
V (f; P) = i:l jf (xi) - f (xi-1) I
a

n
= l: (f (x. ) -- f (x. ) f (b) - f (a) • ]
i=l 1 1-1
 2.

4: FACT If f: [a,b] -+ Rsatisfies a Lipschitz condition, then f E BV[a,b].

['Ib say that f satisfies a Lipschitz condition means that there exists a

constant K > 0 such that for all x,y E [a,b],

If (x) - f (y) I S K Ix - YI·]

5: FACT If f: [a,b] -+ R is differentiable on [a,b] and if its derivative

f': [a,b] -+ R is bounded on Ia,b], then f E BV[a,b].

[The mean value theorem implies that f satisfies a Lipschitz condition on

[a,b] . ]

[Note: Tnerefore r:olynomials on [a 1.b] are in BV [a,. b] . ]

6: FACT If f: [a,b] -+ R has finitely many relative maxima and minima, say

at the pJints

a< ~l < , .• < ~n < b,

then

< + oo,

so f E BV[a,b].

7: EXAMPLE Take f (x) = sin x (0 S x ~ 2TI) ~ then Tf[0,2TI] = 4.

Neither continuity and/or boundedness on [a,b] suffices to force bounded

variation.

8: EXA.1\1.PLE Take Ia,b] -- IO ,.l] and let

x sin(l/x) (O <x < 1)

f (;x:) ;::::;

0 (x ;::::; 0) •.
 3.

Then f (x) is continuous and rounded but f tf_ BV [O, l] •

[Note: On the other hand,

2
x sin(l/x) (0 < x :s. 1)
f (x) =
0 x = 0

is continuous and of rounded variation in [O,l].]

The composition of two functions of bounded variation need not be of bounded

variation.

9: EXAMPLE Work on [0,1] and take f(x) = h",

2 2
x sin (1/x) (0 <x ~ 1)

g(x) =
0 (x = 0).

Then f: [O,l] -+ R, g: [O,l] -+ [O,l] are of rounded variation but f 0

g: [O,l] -+ R is
not of bounded variation.

10: FA.CI' Suppose that f: [a,b] -+ [a,b] ...--,... then the composition

fog E BV[a,b] for all g:[a,b] -+ [a,b] of rounded variation iff f satisfies a
Lipschitz condition.
[In one direction, suppose that

lf(x) - f(y) I~ Klx ~ yj (x,y E [a,b]).

Let P E P[a,b]:

b n
v (f o g;P) = E I (f 0 g) (xi) - (f 0 g) (xi-l) I
a i=l
 4.

n
~ L K!g(x.) - g(x ._ )
. 1 l. 1 1
I
i=

b
~ KV (g:P) ~ KTg[a,bJ < + oo.]
a
 1.

§4. PROPERTIES

1: TBEOHEM If f ,g E BV[a,b], then f + g E BV[a,b] and

2: T'"rlEOREM If f E BV[a,b] and c E R, then cf E BV[a,b] and

3: SCHOLIUM BV [a, b] is a linear space.

4: THEOREM If f,g E BV[a,,b], then fg E BV[a,b] and

Tf [a, b] ::: ( sup Ig I)Tf [a, b] + ( sup If I)T [~ .b] •

g [a,b] [a,b] g

5: SCHOLIUM BV [a ,b] is an algebra.

6: THEOREI'1 Let f E BV[a,b] and let a < c < b -,- then

f E BV[a,c]

f E BV[c,b]

and

7: CRITERION Suppose given a function f: [a, b] -+ R with the property that

[a,b] can be divided into a finite number of subintervals on each of which f is

rocmotonic .,..- then f E BV [a, b] •

8: EXAMPLE A function of rounded variation need not be monotonic in any

subinterval of its domain.

 2.

[Take [a,b] = [O,l] and let r ,r2 , •.• be an ordering of the rational numbers
1
in ]O,l[. Fix 0 < c < 1 and define f:[O,l]-+ R by

f (x) =
0 otherwise.

Then f is nowhere nonotonic but it is of bounded variation in [O,l]:

2c .]
- c

f E BV[a,b] => 1£1 E BV[a,b].

Therefore BV[a,b] is closed under the formation of the combinations

} Cf + If I>

1
2 Cf - I£ I) .
 1.

§5. REGULATEV FUNCTIONS

Given a function f: [a,b] -+ R and a point c E ] a,b [,

f (c+) = limit from the right= lim r(x)

xk

f (c~) = limit from the left = lim f (x).

xtc

[Note: Define f (a+) and f (b-) in the obvious way.]

1: DEFINITION f is said to be regulated if

• f (c+) exists for all a ~ c < b.

• f (c-) exists for all a < c ~ b.

2: NarATION REG [a,b] is the set of regulated functions in [a,b] .

3: THEOREM. REG[a,b] is a linear space.

[Sums and scalar multiples of regulated functions are regulated.]

4: N.B. Continuous functions f: [a,b] -+ R are regulated, i.e.,

C[a,b] c REG[a 1b].

5: THEOREM Let f E REG [a ,b] ~""- then the discontinuity set of f is at

:rrost countable.

6: DEFINITION A function f: [a, b] -+ R is

right continuous if for all a ~ c < b,

f (c) = f(c+).

7~ DEFINITION Let f E REG[a,b] -. then the right continuous modification

 2.

f r of f is defined by

f (x) = f (x+) (a < x < b).

r

8: LEMMA Up to an a-e. rrost countable set, f = f.

r

[The set of points at which f is not right continuous is a subset of the

set of points at which f is not continuous.]

9: LEMMA f is right continuous.

r
[For

f (c+) = l.im f (x) = l.im f(x) = f(c+) = f (c).]

r x+c r x+c r

10: DEFINITION Let f: [a,b] -+ R.

• If f(x) = xI(x), where I= [a,b], or ]a,b[, or [a,b[, or ]a,b], then

f is said to be a single step function.

• If f is a finite linear combination of single step functions, then f is

said to be a step function.

11: LEMMA A function f: [a,b] -+ R is a step function iff there are points

a=x 0 <x1 <···<xn =b

such that f is constant on each open interval ]x. ,x. [ (i = 1, ... ,n).
l -1 l

12: THEOREM Let f:[a,b] -+ R ~- then f is regulated iff f is a unifonn

limit of a sequence of step functions.

13 : N. B. Regulated frmctions are rounded.

[Take an f E REG[a,b] and choose a step function g such that 11£ - gj 100 $ 1,
 3.

hence V x E [a~b],

If (x) I < 11 f - g I Ico + I Ig I Ico ~ 1 + 11 g I Ico· J

14: THEOREM I.et f E BV[a,b] ~ then f is regulated.

PROOF Suppose that a < c ~band f (c-) does not exist - then there is a

positive number s and a sequence of real numbers ck increasing to c such that

for all k,

It therefore follows that for all n,

n
+co ~ Tf [a,b] ~ L If (ck) ~ f (ck+l) I > ns,
k=l

an impossibility. In the same vein, f (c+) must exist for all a :S c < b.

15: SCHOLIUM

BV[a,b] c REG[a,b]~

In particular: The discontinuity set of an f E BV [a ,,b] is at n:ost countable.

16: THEOREM REG[a,b] is a Banach space in the unifonn nonn and BV[a,b]

is a dense linear subspace of REG[a,b], thus

BV[a,b] = REG[a,b]
 1.

§6. POSITIVE ANV NEGATIVE

1: NorATION Given a real number x, put

+ 1
x =max(x,O) = 2 <lxl +x)

x .... =max(..- x,O) - 1 ( Ix I - x).

2

Given a function f:fa,b] -+ R, let

n
+ +
Tf fa,b] sup l: (f (xi) - f (x.1.-1 ))
PEP[a,b]

n
-
T;[a,b] == sup l: (f (xi) - f (x.J..-1 )) ,
PEP[a,b] i=l
the

:positive
total variation
negative

off in [a,b].

Obviously

so T;[a,b], T~[a,b], Tf[a,b] are all finite if f E BV[a,b].

2: N .B. Abbreviate
n
L: {f{x.) - f(x. )) to l:,
i=l l. J..-1
 2.

n
L (f(x.) - f(x. ))- to L-.
i=l }_ i -1

Tnen

=>
2L+ = L + f(b) - f(a)~ 2L- = r - f(b) + f(a).

3: THEOREM If f E BV[a,b], then

T;[a,b] - T~[a,b] = f(b) - f(a).

Replace 11 b" by "x:i and assume that f E BV[a,b].

+
• Tf[a,x] = 2-1 (Tf[a,x] + f(x) - f(a))

=>
1 + -1
2 (Tf [a,x] + f(x)) = Tf[a,x] + 2 f(a)

-1
• T~[a,x] = 2 (Tf[a,b] - f(x) + f(a))

=>

1 - -1
2 (Tf[a,x] - f(x)) = Tf[a,x] - 2 f(a).

4: LEMMA The functions

,, Tf [a,a] = 0
1
x +
2 (Tf[a,x] - f(x))
 3.

are increasing.
PRCX)F Let a :: x < y :;: b.

1 1
• 2 (Tf[a,y] + f(y)) - 2 (Tf[a,x] + f(x))

-- 21 (Tf(a,y] - Tf[a,x] + f(y) - f(x))

~ ~ (Tf [x,y] - lf(y) - f (x) I> > 0.

1 1
• 2 (Tf[a,y] - f(y)) - 2 (Tf[a,x] - f(x))

= 21 (Tf[a,y] - Tf[a,x] - f(y) + f(x))

~ ~ (Tf[x,y] - lf(y) - f(x) I> ~ 0.

5: DEFINITION The representation

1 1
f(x) = 2 (Tf[a,x] + f(x)) - 2 (Tf[a,x] - f(x))

is the Jordan deca.tlJ;.Osition of f.

§_:_ REMARK To arrive at a representation of f as the difference of two

strictly increasing functions, write

1 1
f(x) = c2 (Tf[a,x] + f(x)) + x - c2 (Tf[a,x] - f(x)) + x).

7: THEOREM Suppose that f E BV [a ,b] - then f is Borel measurable.

[For this is the case of an increasing function.]
 1.

§7. CONTINUITY

1: THEOREM Let f E BV[a,b]. Supp:Jse that f is continuous at c E [a,bJ --

then Tf[a,-] is continuous at c E [a,b].

PROOF The function x -+ Tf [a,x] is increasing, hence both one sided limits

exist at all points c E [a,b], the claim being that

lim Tf[a,x] = Tf[a,c].

x -+ c

'Ib this end, it will be shown that the right hand limit of Tf [a,x] as x -+ c is

equal to Tf[a,c], where a_:: c < b, the discussion for the left hand limit being

analogous. So let.s-> O and choose a7 Osuch that

o<x - c < a => lf <x> - f Cc> I < f.

Partition [c ,b] by the scha:ne
n
Tf[c~b] < r jf(x.) ~ f(x1.~1 ) [ + £ (x0 c, x = b).
i=l i 2 n

If x
1
..... c < c, then

n
Tf [c,b] - ~ <-· If (x1 ) - f (c) I + E [f (xi) ·- f (xi·.-l> I
i=2

=>

On the other hand, if x1 - c ~ o, add a point x to the partition subject to

 2.

x - c < 0 thus
1

n
Tf [c,b] - ~ < If (x1 ) - f (x0 ) I+ l: If (xi) - f (xi-l) I
i=2

n
+ l: If (xi) - f (xi-1) I
i=2

n
< If (x ) - f (x) I + -€ +
1 2
E
. 2
If (x. ) - f (x. 1 )
1 i-
I·
i=

Since
{x,x1 , ••• ,xn}

is a partition of [x,b], it follows that

=>

Finally

< s

if x - c < o. Therefore

so Tf [a,x] is right continuous at c.

2: SCHOLilM If f E BV[a,b] n C[a,b]. 1 then

Tf [a,,,_] E C [a 11 bJ •
 3.

3: REMARK It is also true that

Tf+ [a,-] E C[a,b]

Ti[a,-] E C[a,b].

Proof:
+ -1 .
Tf[a,x] = 2 (Tf[a,x] + f(x) - f(a))

-
Tf[a,x] = 2-1 (Tf[a,x] - f(x) + f(a)).

4: 'YtlEOREM f E BV[a,b] is continuous, then f can be written as the

difference of two increasing continuous fnnctions.

[In view of what has been said al:::cve, this is obvious.]

5: LEMMA let f E BV[a,b]. Assmne: Tf [a,-] is continuous at c E [a,b] . ,__

then f is continuous at c E [a,b].
PROOF For

c < x => jf (x) - f (c) I ~ Tf[c,x] = Tf[a,x] - Tf[a,c]

x < c => jf(c) .... f(x) I. ~-Tf[x,c] = Tf[a,c] - Tf[a,x].

6: RAPPEL let f: [a,b] + R be increasing and let xl'x ,.. • be an entnneration

2
of the interior points of discontinuity of f --- then the saltus function sf: [a,b] + R
attache::i to f is define::i by

and if a < x ~ b, by
 4.

(f (a+) - f(a)) + L: (f (~+) - f(~-))

~<'.x

+ (f (x) - f (x-)) •

7: FACT The difference f - sf is an increasing continuous function.

Assume again that f E BV[a,bl and put

V(x) = Tf [a,x] , F (x) = V(x) - f (x) (a ~ x ~ b) •

8: N.B. V(x) and F(x) are increasing functions of x.

Let

be the set comprised of the discontinuity points of V.

9: REMARK The discontinuity set of V coincides with the discontinuity

set of f and the discontinuity set of F is contained in the discontinuity set of f.

Introduce

8v(x) = (V(a+) - V(a)) + L: (V(~+) - V(~-) )

~<x

+ (V (x) - V (x-))

and

= (F(a+) - F(a)) + L: (F(xk+) - F(xk-))

~<x

+ (F (x) - F (x-)) ,

where a < x < b and take

 5.

10: LEMi:V:lA sv is the saltus function of V and sF is the saltus function

of F.

[Per V, this is true by its very construction. A.s for F, if ~ is not a

discontinuity point, then

thus such a term does not participate.]

11: DEFINITION The saltus function sf: [a,b] -+ R attached to f is the

difference

Spelled out,

and

= (f(a+) - f(a)) + L (f(~+) - f(~-))

~<x

+ (f(x) - f(x-))

subject to a < x S b.

12 : SCHOLIUM The functions

- V(x) - sV(x)
x -+ '
F(x) - sF(x)
I
are increasing and continuous. Therefore

= (V(x) - sv(x)) - (F(x) - sF(x))

is a continuous function of bounded variation.

 1.

§8. ABSOLUTE CONTINUITY 1

1: DEFINITION A function f: [a,b] + R is absolutely continuous \:/ s > O,

3 o> 0 such that

n
Z: lf(1\:_) - f{ak) I < s
k=l

whenever
<a <b <b
- n n

for which
n
L (1\:_ - ~) < o.
k=l

2: NOI'ATION" AC[a,b] is the set of absolutely continuous functions in [a,b].

3: THEOREM An absolutely continuous function is unifonnly continuous.

4: THEOREM

f E AC[a,b] => Jf! E AC{a,b].

5: THEOREM If f,g E AC[a,b], then so do their smn, difference, and product.

6: THEOREM

AC[a,b] c BV[a,b].

7: SCHOLIUM f E C[a,b] but f ¢ BV[a,b], then f ¢ AC[a,b].

8: CRITERION If f is continuous in [a, b] and if f" exists and is bounded

in ] a,b [, then f is absolutely continuous in [a,b] •

[Define M > 0 by jf~ (x) I < M for all x in ]a,b[. Take s > 0 and consider
 2.

n
l: If <~> - f cak> I ,
k=l
where

n
s
l: (~ - ~) < M.
k=l

ONing to the Mean Value Theorem, 3 ~ E ]ak 1 1\:_[ such that

f (~) -- f (~)
= f I (X. ) •
~ -~ K

Therefore

n
= l:
k=l

n
= L If'(~) I I~ --·~I
k=l

E
<MM=s.]

9: EXAJYIPLE It can happen that a continuous function with an unbounded

derivative is absolutely continuous.

[Consider f (x) = rx (O ~ x ~ 1) - then f E AC[O,lJ but

 3.

1
f' (x) = - - (0 < x < 1).]
2£"

10: EXAMPLE Consider

2
x sin (l/x) (O < x ~ 1)
f (x) =
0 (x = 0).

Then f E BV[O,l]. But more is true, viz. f E AC[O,l]. In fact, in ]0,1[,

f' (x) = 2x sin(l/x) - cos(l/x)

=>

If' (x) I ~ 2 Ix I Isin (l/x) I + !cos (l/x) I

< 3.

11: THEORE.M Iet f E BV[a,b] -.· then f E AC[a,b] iff Tf [a,-] E AC[a,b].

PROOF Suppose first that f is absolutely continuous. Given s > 0, introduce

the pairs

subject to

n
L (bk - ck) < a,
k=l

thus
n
L: I£ (bk> - f <~ > I < s.
k=l

For each k, let

p :a.
k K
= X.
KO
< X.
Kl
< • • • < Xi.rn
n ...k
= bk
 4.

be a partition of [a.K:,1\-_] - then

n n
l: . l: (~. - ~ .) = E (~ .... ~} < 0
k=l i=l l i-1

=>

n ~
E E If(~. } - f (~ ~ )I < s.
k=l i=l l. i-1

Vart now the Pk through p( [~,bk] ) and take the supremum, hence

or still,

So Tf [a,-] E 'AC. [a,b] • In the other direction, simply note that

J f (1\-_) - f (~) J ~ Tf [a,bk] - Tf [a,~].

Recall that the Jordan decomposition of f ~the representation

f(x) = 21 (Tf[a,x] + f(x))

1
2 (Tf[a,x] ~ f(x)).

12: SCHOLIUM If f E 'AC.[a.,b], then f can be represented as the difference

of bNO increasing absolutely continuous functions.

Here is a useful teclm.icality.

13: LEMMA Suppose that f: [a,b] + R is absolutely continuous 'I."!"""" then

v s > 0, 3 o> 0 such that for an arbitrary finite or countable system of pairwise
 5.

disjoint open intervals { (~ ,1\:)} with

I (1\: ..,.. ·~) < o,

k

the inequality

I osc(f; [~,~]) < s

k

obtains.

14: DEFINITION A function f: [a,b] + R is said to have property (N) if

f sends sets of Lebesgue measure 0 to sets of Lebesgue measure 0:

E c [a,b] & :\(E) = 0 => :\(£ (E)) = 0.

15: THEORE!v'.l If f: [a,b] + R is absolutely continuous, then f has property (N).

PROOF Suppose that :\(E) = 0 and assume that a rJ_ E, b ¢E (this omission has

no bearing on the final outcome). Notationally s, o, and { (~,11c)} are per #13,

thus

'lb fix the data and thereby pin matters down, start by putting

hence

Since :\(E) = 0, there exists an open set S c. [a,b] such that

E c S, :\ (S) < 6.

Decompose S into its connected components ] ~,bk [, so

 6.

Next

f (E) c f (S) = l: f (] ~ ,~ [)
k

or still

f.* (f (E)) ~ l: f.* (f ( [~ 1 ~])).

k

But

Therefore

f.*(f(E)) < l: (Mx: - ~) < s.

k

Since s is arbitrary, it follows that

A(f(E)) = 0.

16: THEOREM If f: [a,b] -+ R is continuous, then f has property (N) iff

for every Lebesgue measurable set E c:. [a,b] 1 f(E) is Lebesgue measurable.
PROOF Assuming that f has prqperty {N), take an E and -write
00

E = ( U K .) u S (Kl c K
2
c ••• ) ,
j=l J

where each K. is compact and S has Lebesgue measure O. Since f is continuous,

J

f (K,) is compact, hence

J 00

U f (K.)
j=l J

Lebesgue measurable. But f has property (N), hence f (S) has Lebesgue measure O.

Therefore
00

f (E) = ( U f(K.)) u f(S)

j=l J
 7.

is Lebesgue measurable. In the other direction, supp:>se that f does not p:>ssess
property (N), thus that there exists a set E c [a,b] of Lebesgue measure O such
that f (E) is not a set of Lebesgue measure 0.

e If f (E) Lebesgue measurable, then it contains a nonmeasurable subset.

• If f(E) is not I.ebesgue measurable, then it contains (is ••. ) a non-

measurable set.
So there exists a nonmeasurable set A c f (E) • Put S = f-l (A) n E: S is

Lebesgue measurable (being a subset of E, a set of Lebesgue measure O), yet f(S) =A

is not Lebesgue measurable.

17 : SCHOLIUM An absolute!y continuous function sends Lebesgue measurable

sets to Lebesgue measurable sets.

18: REMA.RI< I.et E c [a,b] be Lebesgue measurable - . - then its image f (E)
under a continuous function f:[a,b] + R need not be Lebesgue measurable.

19: RAPPEL If E c R is a set of Lebesgue measure 0, then its complement

Ec is a dense subset of R.

[In fact, Ec n I t- ¢ for every open interval I.]

2O: LEMMA Sup:i;:ose that f, g: [a ,b] + R are continuous and f =g alrrost

everywhere -- then f = g.
[The set
E {x E [a,b] 1f (x) I g(x)}

is a set of Lebesgue measure 0. ]

21 :z APPLICATION Two absolutely continuous functions which are equal aLn:ost

everywhere are equal.
 l~

§9. VIN! VER!VATIVES

'l. DEFINITION Let f ~ [a,b] + R.

• Given x E [a,b[,

.
1lJl1 f (.x+h) - f (x)
::::; sup h
hi-0

is the ~pper riqht derivativ~ of f at x and

(D f) (x) = lim inf f (x+h) f (x) '!'"

+ htO
h

is the '].~er riqht derivative of f at x.

• Given x E ]a,b],

(D-f) (x) = lim sup f (x+h) - f (x)

htO h

is the upper left derivative of f at x and

(D_f) (x) = lim inf f (x+h) - f (x)

htO h

is the lOiNer left derivative of f at x.

2: N. B. Collectively, these are the Dini derivatives.

3: EXAMPLE Suppose that a < b and c < d. I.et

. 1
sm- cos -1 (x > 0)
x x
f (x) = 0 (x = 0)

=[ sin ~ T a{ + cos -1
x
-12 (x < 0) •
 2.

Then

(D-f) (0) = d > c = (D f) (0).

~

If (D+f) (x) = (D+f) (x), then the comron value is called the right derivative

of f at x, denoted (Drf) (x), and f is said to be !ight differentiable at x if

this common value is finite.

If (D-f) (x) = (D_f) (x), then the C0Im1.0n value is called the left derivative

off at x, denoted (D,e_f) (x), and f is said to be left differentiable at x if this

co:rnrron value is finite.

4: EXAMPLE Take f(x) = lxl -.-- then

=> (D f) (O) = 1
r

and

(D_f) (0) = - 1

If (Drf) (x) and (D,e_f) (x) exist and are equal, then their common value is

denoted by f' (x) and is called the derivative of f at x, f being differentiable at

x if f' (x) is finite.
 3.

[So the relations

are tantamount to the differentiability of f at x.]

5: EXAMPLE Take f (x) = l (x "f O) / f (0) = 0 - then

(D f) (0) = + oo
r

Therefore f' (O) =+ oo but f is not differentiable at O.

There is much that can be said about Dini derivatives but we shall limit

ourselves to a few p:>ints that are relevant for the sequel.

.§.:_ THEOREM Let f: [a,b] -+ R -.. then for any real number r, each of the

following sets is at rrost cotmtablel

{x: (D+f) (x) > r and (D-f) (x) < r},

-
{x: (D_ f) (x) > r and (D+f) (x) < r},

+
{x: (D f) (x) < r and (D__f) (x) > r},
-
{x: (D'""'f) (x) < r and (D+f) (x) > r}.
-.-

7l APPLICATION" Let f: Ia,b] -+ R - then up to an at rrost c0tmtable set,

(D+f) (x) > (D_f) (x)

~

8l THEOREM Let f: [a 1 bJ -+ R be a Lebesgue measurable function - then its

 4.

Dini derivatives are Lebesgue measurable functions.

To fix the ideas, let us consider a special case. So suppose that f: [a,b] + R

is a Lebesgue measurable function and E c [a,b[ is a Lebesgue measurable subset

of [a,b]. Assume: Dr f exists on E -· then Dr f is a Lebesgue measurable function

on E.

To establish this, extend the definition of f to R by setting f = 0 in

R - [a, b] • Define a sequence g1 , g 2 , • • • of Lebesgue measurable functions via the

prescription

1
g (x) = n(f(x + ) - f(x)).
n n

Let D be the subset of R comprised of those x such that lim g (x) exists in
e n + oo
n

[ - oo, + oo] - then D is a Lebesgue measurable set and

e

lim gn : De + [""" 00 I + 00]

n + oo

is a Lebesgue measurable function. Take now an x E E and write

(D f) (x) = lim f (x+h) - f (x)

r h+O h

+ 1 ) · - f (x)
f (x
= lim --·--=-~---- = lim gn (x) •
n-+oo n-+oo
n

Consequently E c D and
e

D f =
r
lim gn
n + oo

in E, hence Drf is a Lebesgue measurable function on E.

9: N.B. ·Analogous considerations apply to D,e.f and f".

 1.

§10. VIFFERENT1ATTON

We shall first review some fundamental points.

1: FA.Cr I.et f: [a,b] + R be an increasing function - then f is differ--

entiable in ]a,b[ - E, where E is a set of I.ebesgue measure 0 contained in ]a,b[.
[Note: Bear in mind that ndifferentiablen means that at x E ]a,b[ - E, f~ (x)

exists and is finite. MJreover f 1 (x) = + oo is p::>ssible only on a set of I.e..besgue

measure 0. ]

2: N.B.

f': [a,b] - E + R>O

is a Lebesgue measurable function.

3: REMARK If E c ]a,b[ a set of I.ebes::JUe measure 0, then it can be

shoVJn that there exists a continuous increasing function f which is not differentiable
at any point of E.

4: RAPPEL If ¢ is a Lebesgue measurable function and if l/J = ¢ alm::>st

everywhere, then l/J is a I.ebesgue measurable function.

5: FACT I.et f: [a ,b] + R be an increasing function -.~- then f 1

is integrable

on [a,b] and

~ft
a
< f(b) -
-
f (a).

[Note: This estimate can be sharpE!necl to

rb f i~ < f (b--) .... f (a+) • ]

a ""'
 2.

6: EXAr"'.ll?LE One can construct a function f ~ [a,b] -+ R that continuous

and strictly increasing in [a,b] such that f' = 0 alm::>st everywhere, hence

0 = ~a f' < f (b) - f (a) •

1
7: FACT Given an f EL [a,b], put

F (x) = ~ f (a ~ x :s_ b) •

Then F E AC[a,b] and F' =f alrrost everywhere.

8: FA.Cr Sup:p:>se that f: [a,b] -+ R is absolutely continuous --- then

f (x) = f(a) + f:a f' (a < x :_ b).

9: FUBINI 'S LEMMA I.et {f } (n

n
= 1, 2 , ••• ) be a sequence of increasing

functions in [a,b]. Assume that the series

00

2: fn (x)
n=l

converges pointwise in [a,b] to a function F -- then F is differentiable aJrrost

everywhere in [a ,b] and

co
p~· (x) L: f~ (x)
n=l

off of a set of Lebesgue measure 0.

PRCXJF Without loss of generality, take fk(a) =0 for all k and observing

that Fis increasing, let Ebe the set of points x E ]a,b[such that the derivatives

F' (x), fi (x), f,2 {x), ••• all exist and are finite .,..~ then [a,b] ·,"!. E has Lebesgue

measure 0. I.et
n
Fn(x) ~ E fk(x).
~l
 3.

Supp:Jse that x EE and h is chosen small enough to ensure that x + h E [a,b] - then

00
F (x + h) ·.- F (x)
h' = l:

F(x + h) - F(x) n fk(x + h) - fk(x)

> 2:
h
k=l

=>
n
F' (x) > l: ' (x) == F' (x) •
- k=l n

The fk_ are nonnegative and the sequence

{ F~ (x) } (n == 1, 2 , ••• )

is bounded above by F 1 (x) , hence is convergent. It remains to establish that

lim F' = pt..

n -+ oo
n

alnost everywhere in [a,b]. Since

lim Fn(b) = F(b),

n -+ oo

there exists a subsequence {Fn. (b) } such that

J

F(a) - Fn. (a) =0 < F(b) - F

~- n. (b) < 2-j.
J J

But F - F is an increasing function, thus

n.
J
-j
0 ~ F(x) - F (x) < 2
n.
J

for all x E [a,b]and so the series

00

L: (F' - F' )
n.
J
 4.

is a pointwise convergent series of increasing functions. Reasoning as above,

we conclude that the series
00

2: (F' - F' )
n.
j=l J

is convergent al.rrost everywhere in [a,b] and from this it follows that

F' (x) - F • (x) -+ 0

n

as n -+ oo for almost all x E [a,b].

10: APPLICATION Supp::>se that ff [a,b] -+ R is increasing and let sf 1 [a,b] -+ R

be the saltus function attached to f - then sf =0 aJ..nost everywhere.

[In general, sf is not continuous. Still, a continuous singular filllction is

a continuous function '\vhose derivative exists and is zero abrost everywhere. 'lb

illustrate, write

where by construction rf is increasing and continuous. And alnost everywhere

Introduce F by the rule

F(x)

and set

f
cs = rf .... F.

f'cs = r~
f
- F' = f' - f' = 0.
Therefore f cs is a continuous singular increasing function and
 5.

The fact that an f E BV[a,b] can be represented as the difference of b.vo

increasing functions implies that f is differentiable aLrrost everywhere.

[Note: Therefore a continuous nowhere differentiable function is not of

bounded variation.]

11: THEOREM Supp:::>se that f E BV [a ,b] ~ then for a.J..rrost all x E [a, b] ,

PROJF Given n E N, choose a partition P E P[a,b] such that

n

In the segment ~-l ~ x ~ xk of Pn' let

or

where the constants are chosen so that fn(a) = 0 and the values of fn at~ agree -

then

so

-n
< 2

On the other hand, the function

 6.

is increasing, hence

< 2
-n
=>

00 00
--n
E (Tf[a,x] - fn(x)) < E 2 < + oo.
n~- n=l

The series
00

E (Tf[a,x] - fn(x))
n=l

is therefore J?Ointwise convergent, thus by Fubini's lemma, the derived series

converges aJ.rrost everywhere, thus

aJ.rrost everywhere. But

f' (x) = ± f' (x) •

n

Since T:f [a,x] > 0 (Tf[a,x] being increasing), the upshot is that

If' (x) I = T£ [a,x]

aJ.rrost everywhere.

12: APPLICATION
1
f E BV[a,b] => f' E L [a,b].

[For
 7.

13: THEOREM Given an f E L1 [a,.b] , put

F(x) = _ta L

Then

PROOF Given a PE P[a,b],

= ~ l!"k f I '.:f~ !fl < + 00

k=l ~--1

=>

'Ib reverse this, recall that FE AC[a,b]; that F' = f alm:>st everywhere, and that

aln10st everywhere. Therefore

!If 11 L1 = I!a IF' I

= /!a T' [ a - ]
F '

14: LEMMA Suppose that f:[a,b]-+ R is increasing -- then f E AC[a,b] iff

I:a f' =f (b) - f (a).

 8.

PROOF If f E AC[a,b], then

f (x) = f (a) + ;X-a £' (a ~ x ~ b)

=>

f(b) - f(a) = Jba f'.

Conversely, write

f (x) = a £' + f cs (x) + sf(x).

fx

Then

f (x) = f (a) + fX-a f' + g(x),

where

f cs (x) + sf(x) = f(a) + g(x),

=>

f cs (a) + sf (a) = f (a) + g(a)

=>

rf (a) - F(a) + sf (a) = f (a) + g(a)

=>

(f - sf) (a) + sf(a) = f (a) + g(a)

=>

f (a) = f (a) + g(a)

=>

g (a) = 0.

In addition, the assumption that

!b f' = f(b) - f (a)

a
 9.

implies that

g{b) =f {b) - f (a) - f>a ft

= 0.

Since g is increasing, it follows that g{x) = 0 for all x E [a,b], hence

f(x) = f(a) +I!"a f~~

15: THEOREM Supµ::>se that f E BV[a,b] - then f E AC[a,b]

[a,bJ = f>a IP 1.
PRCX>F On the one J1an.d,

f E AC[a,.b] => f~ E L [a,b]

1

On the other hand, assume the stated relation. Since for alm:::>st all x in [a ,b] ,

we have

or still,

T f[a,b] - T f[a,a] = fab T:f [a,-] .

But Tf [a,-] is increasing, thus in view of the lerrrna, Tf [a,-] absolutely

continuous, which in tum implies that f absolutely continuous.

 1.

§11. ESTIMATE OF THE IMAGE

1: RAPPEL

A. = Lebesgue measure

A.* = outer Lebesgue measure't

2: LEMMA Let f: [a,b] + R. Suppose that E c. [a,b] is a subset in which

f 1
exists, subject to If' I ~ K --- then

A.*(f(E)) < KA.*(E).

Tne proof will be carried out in seven steps.

Step 1: Given x E E, f"' (x) exists and

If' (x) I = lim f(y)_: ~(x) I < K.

I
y + x y

So, V x EE, 3 o> 0:

lf(y) - f(x) I ~ Kly ~xi (y E ]x - o, x + o[ n [a,b]).

If now for n = 1,2, .•. ,

En == {x I
E E: f (y) - f (x) I I
~ K y - x I (y 1 x + n1 [ )'} ,
E ] x - n,

then each x E E belongs to E (n > > 0) , hence

n
00

E c;. U E •
n=l n

On the other hand, V n, En c. E and {En} is increasing. Therefore

E = U E = lim E •
n=l n n + oo n
 2.

Step 2: Consequently

lim A*(En ) = A*(E).

n -+ co

But
co co

f {E) = f( U En) = U f(En) = lim f (En)

n=l n=l n -+ co

=>

lim A*(f(En)) = A*(f(E)).

n -+ co

Step 3: Let e: > 0 be given and let In, k (k = 1, 2,, •.• ) be a sequence of

open intervals such that

00
1
A(In k) <- E
n' n , c u I
k=l n 1 k'

and
co
l: A(In k) < A*(En ) + e:.
1 -

Step 4:
00

E
n
= k=l
U (E
n
n I
n,
k)

and
co

f (E ) = U f (E nI k).
n k=l n n,

Step 5:

=>

A* (f (E n I k)) < K/din k) •

n n,. - /
 3.

Step 6:

00

).* (f (E ) )
n = A* ( U f (E n In;k))
k=l n
00

< L: )i.*(f(E n rn,k))

n

00

< E K)i.(I k) < K ( A* (E ) + s) •

k=l n, - n

Step 7 :_

)i.*(f(E)) = lim A*(f(En))

n + oo

:::_ K( lim A* (En) + c)

n + oo

= K(A* (E) + e)

=>

A*(f{E)) :::_ KA*(E) (s ~ 0),

the assertion of the lerrma.

3: THEOREM I.et f: [a, b] -+ R be Lebesgue measurable. SupJ?Ose that E c [a ,b]

is a Lebesgue measurable subset in which f is differentiable --- then

A* (f (E) ) s f E If 1 (x) I •

PROOF Note that fl :E -+ R is a Lebesgue measurable function. This said, to

begin with, assume that in E, If' J < M (a J?Ositive integer). I.et

K1k = {x E Et k """
1
< If' (x)
~
I < -~}
2n n ' 2

where
 4.

Then for each n,

A*(f(E)) = A*(f(u ~))

k

Therefore

A* (f (E) ) < lim O: ~ - l A. (,..n) + _.!_ E A. (,..n) )

-n+ook 2n K 2nk K

'Ib treat the case of an unbounded f' , let

~ = {x E E : k - 1 :: If ' (x) I < k} (k = 1, 2 , ••• ) •

Then
A.*(f(E)) = A.*(f(U 1\:_))
k

~ A.* (~ f (1\))

~ E A.*(f(l\:_))
-k
 5.

[Note: In point of fact, f (E) is Lebesgue measurable, so

A*(f(E)) = A(f(E)).]

4: N.B. It follows that

A*(f(E)) =0
if f i = o.
[It can be shown conversely that

A*(f(E)) = 0

implies that f , . = 0 alrrost every.Nhere in E. ]

5: SCHOLIUM Suppose that f has a finite derivative on a set E ...... then

A* (f (E)) = 0 iff P = 0 alnost everywhere on E.

 1.

§12. ABSOLUTE CONTINUITY 11

1: THEOREM If f: [a,b] -+ R is absolutely continuous and f' (x) =0

alrrost everyw-here, then f is a constant function.
[Let
E = {x e [a, b] : f ' (x) = O}
and let
E' = [a,b] - E.
The assumption that f E AC[a,b] implies that f has property (N) which in turn
implies that f sends Lebesgue measurable sets to Lebesgue measurable sets. In
particular: f (E) , f (E' ) are Lebesgue measurable and

A(f[a,b]) ...< A(f(E)) + A.(f(E~)) •

So first
A.(f(E)} < 0 A.(E) =0 ("K" = 0).
And second, E' is a set of Lebesgue measure O, hence the same is true of f(E•).
All told then
A.(f[a,b]) = 0.

O..Ving now to the continuity of f, the image f ( [a,b]} is a point or a closed interval.
But the latter is a non-sequitur, thus f ( [a,b]) is a singleton.]

2: MAIN THEOREM. Let fl [a,b] -+ R -. then f is absolutely continuous

the following four conditions are satisfiedi
(1) f is continuous.
(2} f ,. exists almost everywhere.

(3) £' E L1 [a,b].

(4) f has property (N).
 2.

PRCXJF An absolutely continuous function has these pro:perties. Conversely,

assume tbat f satisfies the stated conditions~ ONing to (3), given s > O, there

exists o> 0 such that

E c [a, b] & :\ (E) < o = > f E lf • I < s,

Fix

-< •••
- - -
<a
- n
<b
n
<b

with
n
z:: <~ - C\:) < a~
k=l

'Ihen

n
E f If" I < £.
k=l [C\:,bk]

I.et

~ = {x E [C\:,~] tf' (x) exists}.

Thanks to (2), [C\:,~] - ~ is a set of Iebesgue measure O, hence thanks to (4),

f ( [~ ,~] - ~) is a set of Iebesgue measure 0. Therefore

n
:s. k=l
E fA
~x
If' I

n
= ~ I
k=l [~,~]
1£' I

< E:.
 3.

3: SCHOLIU1v1 f E BV[a,b] is continuous and possesses property (N), then

f E AC[a,b].

[One has only to note that if f is of bounded variation, then f' exists alm:>st

1
everywhere and P E L [a,b].]

4: LEMMA If f: [a,b] -+ R has a finite derivative at every :i;oint x E [a,b],

then f has property (N) •

PROOF Suppose that A(E) = 0 (E c [a 1b]). For each positive integer n, let

En = {x E E: If t (x) I: n}.

Then A(E ) = 0 and

n

A*(f(E )) < nA*(E )

n - n
nA(E ) = 0
n
=>

A(f(E )) = 0.
n
Since
00

E = U E
n
n=l
and
00 00

f (E) =f( U En) = U f(En),

n=l n=l

the conclusion is that

00

A*(f(E)) $ E A*(f(En}}
n=l

= E A. (f (En)}
n=l

= 0.
 4.

I.e.: A(f(E)) = 0.

5: EXAMPLE One can construct a continuous function f: [a ,b] + R with a

finite derivative alrrost everywhere 'Which fails to have pror:erty (N) •

6: THEOREM I.et f: [a,b] -+ R. Assrnne: f' (x) exists and is finite for all

x E [a,b] and that f' is integrable there -- then f is absolutely continuous.

PROOF Condition (1) of the Main Theorem is satisfied ( "differentiabilityn =>

ncontinuity"), conditions (2) and (3) are given, and (4) is satisfied in view of

the previous lemna.

Tne composition of two absolutely continuous functions need not be absolutely

continuous. However:

7: FAcr Suppose that f: [a,b] -+ [c,d] and g: [c,d] + R are absolutely

continuous -- then g Q f E AC[a,b] iff (g• 0 f)f' is integrable.

[Note: Interpret g' (f (x) ) f' (x) to be zero vvh.enever f' (x) = 0. ]
 1.

§13. MULTIPLICITIES

Let f: [a ,b] -+ R be a continuous function. Put

m = min f, M max f.
[a,b] [a,b]

1: NDrATION ~fine a function N(fi-) :]- 00 , + 00 [ -+ R by stipulating that

N(f;y) is the number of times that f assumes the value y in [a,b], i.e., the m.nnber

of solutions of the equation

f (x) =y (a ~ x ~ b).
[Note: N (f ;y) is either 0, or a :i;:x:>sitive integer, or + oo.]

2: DEFINITION N (f ;-) the multiplicity function attached to f.

THEOREM N(f;-) is a Borel measurable function and

+oo
f -oo N (f ,-) = Tf [a,b] •

PRCX)F Subdivide [a,b] into 2n eefUal parts 11 let

Ini = [a, a+ (b - a)/2n], i = 1,

and let

Ini = ]a+ (i . . . 1) (b - a)/2n, a+ i(b - a)/2n], i = 2,3, ••• ,2n.

Then f maps each to a segment (closed or not), viz. the segment from m. to M.,
J.. .1.

where
m. = inf f, M. sup f .
.1. I . .1. I .
ni Ill.

The characteristic function xni. of the set f (I . ) is zero for y > M.

m i & y < m. ,
i

one form.
l.
< y < M.,
.1.
while it may be zero or one at the two end:i;:x:>ints. Therefore
 2.

Xni. is Borel measurable, thus so is the function

2n
Xn(y) = L Xni(y) (- oo < y < + oo).
i=l

And

(M. - m.)
1 1

2n
= L osc(f;I .) .
. ni
1=1

I·:breover

Xn ~ O, Xn ~ Xn+l'

which implies that

X lirn Xn
n + oo

is Borel measurable. Pass then to the limit:

+oo X = -l'Jm
f-oo f-oo
+oo Xn = Tf [a,b ] ,
n + oo

f being continuous. Matters thereby reduce to establishing that

X = N(f;-).

First
V n, X < N(f;-) => X < N(f;-).
n - -

Iet now q be a natural number not greater than N(f;y), giving rise to q distinct
 3.

roots

of the equation

f (x) =y (a < x < b) •

Ui;x:>n choosing n > > 0:

b - a .
< mm(x.+l - x.),
2n J.. J..

it follows that all q roots will fall into distinct intervals I . , hence
ni
~ ~ q => x~ q.

If N(f;y) =+ oo, q can be chosen arbitrarily large, thus x{y) = + oo. On the other

hand, if N(f;y) is finite, take q = N(f;y) to get

X(y) ~ N ( f i y) = > X 2:. N ( f ;-) •

4: SCHOLIUM A continuous function f: [a,b] -+ R is of rounded variation iff

its multiplicity function N(f;-) is integrable.

5: N.B. If f E BV[a,b] n C[a,b], then

{y:N(f;y) = + oo}
is a set of Lebesgue measure 0.

[In fact, N (f ; - ) is integrable, thus is finite alrrost everywhere.]

Maintain the assumption that f: [a,b] -+ R is continuous.

6: NarATION Given J = [c,d] c. [a,b], write

+ l if f (c) < y < f (d}

ct> (f ;J ,y) = - 1 if f (c) > y > f (d)
0 otherwise,

where - oo < y < + oo.

 4.

7: LEMMA If

c=y
O <y1 <··· <ym =d

is a partition of J = [c,d] into the m intervals Jj = [y}·~l' yj] and f (yj) "I y
for j = 0, 1, ••• ,m, then

m
¢(f7J,y) = E ¢(f;Jj 1 y) ~

8: NarATION Given a finite system S of nonoverlapping intervals J = [c,d]

in [a,b], put

cN(f;y) = sup E l¢(f;J,y) I.

S JES

9: DEFINITION" cN(f;y) is the corrected multiplicity function attached to f.

Obviously

O ~ cl\J(f;-) ::: + oo.

10: T.HEOREl:".l Vy, - oo < y < + oo,

0 ~ cN(f;y) ~ N(f;y)

and

cN(f;y) = N(f;y)

for all but countably many y.

Therefore

Tf [a,b] = ~= N(f;-) = .(: cN(f;-).

 1.

§14. LOWER SEM1CONT1NU1TY

1: EXAMPLE (Fatou's Iarma) SuppJse given a measure space (X,µ) and a

sequence {f } of nonnegative integrable functions such that f + f al.nost every-

n n
where -- then

Jx f dµ < lim inf fx fn dµ.

n + co

2: 'I'HEORE.1\1 Suppose that f : [a,b]

n
-+ R (n = 1,2, ••• ) is a sequence of

functions that converges pJintwise to f:[a,b]-+ R -- then

Tf[a,b] ~ lim inf Tf [a,b].

n -+ co n

PROOF Given c: > 0, there exists a partition P = {x , ••• ,xm} of [a,b] such
0
that

b m
V (f ;P) = E If (x . ) - f (x. 1 > I
a j=l J J-

-1
> Tf [a,b] - 2 £

Since f (x . ) -+ f (x . } at each of the

n J J

m + 1 pJints x , •.. ,xm' there is an nc: such that

0
-1 -1
lf(x.)
J
- f (x.)
n J I< 4 m £

for all n > n and j = O, ..• ,m, hence if n > n,

·- € - £

lf(xj) - f(xj-l) I
= If (x. ) - f (x. ) + f (x . ) - f (x. ) - f (x. > + f (x. > I
J n J n J n J·-1 J·-1 n J-1
 2.

< If (x. ) - f (x. ) I + j'f (x . ) - f (x. ) J

J n J J-1 n J-1

+ If (x . ) - f (x. 1 ) I
n J n J-

=>

m -1 -1 m
l: If (xJ.) - f (xJ. _1 ) I < 4 s + 4 s + L: If (x.) - f (x. ) I
j=l j=l n J n J-1

or still,
m -1
L: lf(xJ.) - f(xJ..1 > I - 2 s
j=l

m
< L: If (x.) - f (x. 1 > t
j=l n J n J-

~ Tf [a,b].
n

Case 1: Tf [a,b] < + oo - - then

m
L: lf(xJ.) - f(xJ.-l) I
j=l

-1 -1
> Tf [a,b] - 2 s ~ 2 E

= Tf [a,b] - s
=>

[a,b] - s < Tf [a,b] (n > n )

- E:
n
=>

Tf [a,b] - s ~ lim inf Tf [a,b]

n -+ oo n

=> (s + 0)
 3.

Tf[a,b] ~ lim inf Tf [a,b].

n -+ oo n

case 2: Tf [a,b] =+ oo - - then

rn
L:

-1 -1
> e:: - 2 E:

==>

-1 -1
s - 2 s < Tf [a,b] (n > n )
n - e

=>

+ 00
= Tf[a,b] = lim inf Tf [a,b].
n -+ 00 n

3: RELvJARK One cannot in general replace pointwise convergence by convergence

alnost everywhere, i.e., it can happen that under such circumstances

lim inf Tf [a,b] < Tf[a,b].

n -+ oo n

4: EXAMPLE 'Work On [ 0, 2 TI] and take

so f (x) = O -- then f -+ f unifonuly,

n

Tf [0,21T] = 0, Tf [0,21T] = 4.
n

5: EXAJ:VIl?LE work on [0,21T] and take

f (x) =
1
sin (n2x),
n n

so f (x) == 0 -- then f n -+ f unifonuly,

 4.

Tf (0,2TI] = 0, Tf [0,2TI] =+ oo.

n

6: THEOREM I.et f: [a ,b] -+ R be a continuous function -- then cN (f ;-)

is lower semicontinuous in ] - 00 , + 00 [ , i.e. , V y ,

0

cN ( f; y ) < Lim inf cN (f; y) •

0
y-+ Yo

7: THEOREM Suppose that f : [a,b] -+ R is a sequence of continuous functions

n

that converges pointwise to f:[a,b] -+ R -- then Vy,

cN(f;y) ~ lirn inf cN(fn;y).

n -+ co

8: REMARK These statements ensure that cN is lower semicontinuous w. r. t.

to f and w. r. t. y separately. IV.bre is true= cN is lower semicontinuous w. r. t.

the pair (f,y), i.e., if fn -+ f, y-+ y , then

0

9: N.B. In the foregoing, one cannot in general replace cN by N.

 1.

§15. FUNCTIONAL ANALYSIS

1: THEOREM BV [a,b] is a Banach space under the nonn

[Note: Tf [a,b] is not a nonn since a constant function f has zero total

variation, hence the introduction of If (a) I• Recall, however, that

and

As a preliminary to the proof, consider a Cauchy sequence {fk} in BV [a ,b] •

Given c > 0, there exists C E N such that

£

for all k,l > C . T'nerefore

- E

thus the sequence {fk} converges uniforroly to a bounded function f: [a ,b] -+ R, the

claim being that f E BV[a,b].

This said, take a partition P E p[a,b] and note that

for all k,l > C • From here, send l to + co to get

- E:
n
2: I (fk - f) (xi) - (f k - f) (xi-l) I ~ c::
i=l
 2.

for all k -> CE , hence

Tf -f [a,b] < E
k

for all k -> C E • And

Therefore

for all k -> CE . I'-breover

<+ oo.

So f E BV[a,b] and fk +fin BV[a,b].

2: REMARK BV [a ,b] , equipped with the nonn [ I• [ IBV' is not separable.

[Take [a,b] = [O,l] and for f E BV[O,l], r > 0, let

S ( f, r) = { g E BV [ 0 , 1] : I Ig - f I IBV < r} ·

Call >Ct (O < t < 1) the characteristic function of {t} -- then for t
1
~ t 2,

[ [~ - ~ 11 BV = (~ - ~ ) (a) + TX. _ X. [ 0 r l}
1 2 1 2 , 'tl , 't2

= 0 + T [O ,l]
~1- ~2

= 4.
But this implies that

S(Xt ,1) n S(~ ,1) = ~-

1 2
 3.

In fact

l lh - 4 I IBV < l
1

I lh - 4 I IBV < l
2

=>

1141- .xt2 I IBv = 1141 - h + h - ~2 I IBv

:'.: 11~ - h I IBv + 114 - h I IBv

1 2

< 1 + 1 = 2.
Accordingly there exists a continmnn_of disjoint spheres S Cxt'l) c. s (0,3), hence an

arbitrary sphere S(f,r) contains a continuum of disjoint spheres S(r'Xt/3 + f,r/3) .]

3: THEOREM BV[a,b] .is ·.a ·complete metric space under the distance function

The issue is completeness and for this, it suffices to establish that the

balls ~ .of radius.:M centered at 0 are compact, the claim being that every sequence

{fn} c ~has a subsequence converging to a limit in ~·

4: N.B. Spelled out, BM is the set of functions f E BV[a,b] satisfying

the condition

5: HELLY' S SELECTION THEOREM Let F be an infinite family of functions

in BV[a,b]. Assume that there exists a point x E [a,b] and a constant K > 0
0
 4.

such that v f E F,

Then there exists a sequence {f }

n
c F and a function g E BV[a,b] such that

f + g (n + oo)
n

pointwise in [a,b].

6: LEMMA V f E ~v1'

1
If (a) I ~ M (1 + b - a) •

PROOF Write

f (a) = f (a) - f(x) + f(x)

=>

If (a) I s If (a) .... f (x) I + If (x) I

=>

=>
-
< M(b - a) + M

In the HST, take F = {fn}, x 0 = a, and

K = M (1 +b l ) + M.
- a

Then there exists a subsequence {fl'\:} and a function g E BV[a,b] such that

fl'\: + g {k + oo)

pointwise in [a,b] .
 5.

7: LEMMA V nk' V x E [a,b],

If~ (a)l + Tf [a,b] < + oo.

~

The f are therefore bounded, hence by dominated convergence,

11<
f -+ g (k -+ oo)
~

Consider now the numbers

Tf [a , b] (k = 1, 2 , • • • ) •
11<
They constitute a bounded set, hence there exists a subsequence {Tf [a,b]} (not
nk
relabeled) which converges to a limit T. Since f tends to g pointwise, on the
nk
basis of lower sernicontinuity, it follows that

T [a,b] ~ lim Tf [a,b] ,

g k-+oo nk

which implies that

Adjusting g at a if necessary, matters can be arranged so as to ensure that

T [a,b] = T.
g

Consequently

d_~ 7 (fn. ,g) = ! 1 If - gl + jTf [a,b] - T [a,b] ~·

-Ev K 0 nk ~ g

+ (k -+ oo) -} (k -+ oo)

0 1-r-T I
 6.

I.e.:

lirn ~v(f ,g) = o.

k + 00 I1K:_

The final detail is the verification that g E BM. To this end, fix E: > O --

then for k > > O,

< E: + M.

8: LEMMA In the ~V metric, BV [a ,b] is separable.

9: LEMMA Va E R,V f,g E BV[a,b],

~v(af,ag) = lal ~(f,g).

1
10 : 'YtlEOREM Let a E L [a ,b] -- then the assignment

f + f°a fa = Aa (f)
is a continuous linear functional on BV[a,b] when equipped with the ~V metric.

PROOF To establish the continuity, take an f E BV[a,l)] and suppose that {fn}

is a sequence in BV[a,b] such that

the objective being to show that if E: > 0 be given, then

IAa (fn ) - A (f)

a I< E:

provided n > > 0.

So fix a constant C > 0: v n,

J~ Ifn - f I + ITf [a ,b] - Tf [a, b] I ~ C.

n
 7.

For each n choose a point xn such that

1f
n (xn > - f exn > s c
1

and note that for all x E [a,b] ,

- Ifn (x) - f n (xn ) I s Tf [a,bJ

n

and

Tf [a,b] ~ Tf [a,b] + C
n

=>

Ifn (x) - f (x) I

< (i ) I + lf<x> - fCin >I + Ifnn

Ifn ex> - f nn ci l - £Cxn >I

s Tf [a,bJ + Tf[a,b] + lfn(xn) - f(xn) I

n

:: K.
On general grounds (absolute continuity of the integral), givens> 0 there
exists 8 > 0 such that

if A(E) < 8. Take nCJii/ N > > 0:

A(~) < o (~ = {x; la (x) I > N}).

 8.

Then

IAa (fn ) - A (f)

a I

= IJoa £na - Joa f a.1

= JE If - fl la.I+ f lfn- fl lo.I

N n ~

~ !~Kial + f ~ lfn - fl !al

< s/2 + f If - f I Ia I.
~ n

And

x E ~ => Ia (x) I < N

=> f c lfn - fl lal ~ Nf

0
lfn - fl
EN EN
-< Nf°a Ifn - f I < s/2 (n > > O).

Therefore in the end

for all n sufficiently large.

11: N.B.

iff a.1 = a 2 almost everywhere.

[Suppose that A =A • Define ft E BV[a,b] by the prescription

al a.2
 9.

1 (O ~ x ~ t)

0 (t < x ~ 1).

Then

=>

=>

almost everywhere.]
 1.

§16. VUALITY

In the abstract theory, take X = [a,b] -- then there is an isometric iso-

rrorphism

J\.: M( [a,b]) -+ C [a,b]*,

viz. the rule that sends a finite signed measure µto the bounded linear functional

f -+ J [a,b]f dµ.

On the other hand, it is a point of some im;portance that there is another descrip-

tion of C[a,b]* which does not involve any measure theory at all.

1: RAPPEL If f is continuous on [a,b] and if g E BV [a,b], then the

Stieltjes integral

~a f (x) dg(x)

exists.

2: NorATIOO C [a,b] is the set of continuous functions on [a,b] equipped

with the suprernum nonn:

llflLxi= sup lfl,

[a,b]
and C[a,b]* is its dual.

--3: LEMl.\1A I.et g E BV[a,b] ,_.,_ then the assigrunent

f -+ / : f (x) dg(x)
a

defines a bounded linear functional A E C[a,b]*.

g
{Note:
 2.

hence

4: RIESZ REPRESENTATION THEOREJ'Yl If A is a bounded linear functional

on C [a,b], then there exists a g E BV [a,b] such that

A(f) = /: f(x) dg(x) (=A (f))

a g

for all f E C [a,b]. And:

I ]A 11 = T g (a,b] •

PROOF Extend A to [a,b] .:JC [a,b] without increasing its nonn (Hahn-Banach).

Given x E [a,b], let

- 1 (a :: t ::_ x)

O (x < t < b)

and put

Claim: g E BV[a,b] and in fact

Thus take a partition P E P[a,b] and let

s.
1
= sgn(g(x.)
1
- g(x.
1-1
)) (i = l, ••• ,n).

Then
n n
L lg(x .) - g(x ._ ) I= L ~. (g(x.) - g(x. ))
i=l 1 1 1 i=l 1 1 J..- 1

n
= E (A(ux.> - A(u ))
i=l 1 xi-1
 3.

n
=A( E (u - u ))
i=l xi xi-1

n
< I IA 11 11 E (u - u ) 11
i=l xi xi-1

< I IA 11 ·
Therefore

Tg[a,b] :5. I IA! I < + 00 = g E BV[a,b].

Suppose next that f E C[a,b] and let

X. = a + i (b-a) (i' 0 , ... ,n ) .

J.. n

Define
n
fn(x) = . E1 f(x.) (u (x) - u
.1. x.1 x.1-1
(x)).
1=

Then

11 f - fn 1100 = sup If - fn I
[a,b]

< max
- .
sup{ jf(x) - f(x.)
1
I :x.J..-1 -< x <- x..l. }.
l<i<n

Invoking unifonn continuity:, it follOiilTs that

11 f - fn I loo + 0 (n + + oo) '

i.e.,

fn + f => A(f) = lim A(fn)

n + co

n
= lim E f(x.) (A(u ) - A(u
. 1 .1. x..1. x.J..-1
))
n + oo i=

n
= lim E f(x.) (g(x.) - g(x. 1 >>
. 1 .1. .1. J..-
n + oo i=
 4.

= ~a f(x) dq(x)
-
=Ag (f).

From the above,

and

I IA 11 ~ '1~[a,b] ·

So

I IA 11 = T [a,b],
g
as contended.

The "g" that figures in this theorem is definitely not unique. To remedy

this, proceed as follows.

5: DEFINITION g E BV[a,b] is nonnalize~ if g(a) =0 and g(x+) = g(x)

when a < x < b.

[Note: Since g (a) = 0,

I lg I IBv = Tg[a,b].

Observe too that by definition, the right continuous modification g of g in ]a1b [

r
is given by the formula

g (x)
r
= g(x+),

so the assumption is that g = g, i.e., in ]a,b[, g is right continuous.]

r

6: NDrATION Write NBV[a,b] for the linear subspace of BV[a,b] whose

elements are normalized.

7: THEOREM The arrow

NBV[a,b] + C[a,b]*

that sends g to Ag is an isometric isonorphism~

 5.

Here is a sketch of the proof.

Define an equivalence relation in BV[a,b] by writing g ,. ., g

1 2

2: Note that

g ~ 0 => 0 = /:!a dg(x) = g(b) - g(a)

=> g(a) g(b).

Step 3: Establish that

g ~ 0

=>

g(a) = g(c+) = g(c-) = g(b)

if a < c < b.
[Suppose that

a ~ c < b, 0 < h < b - c

and define

1 (a < x < c)
x_-_c
__
f (x) = 1
h
(c < x < c + h)

0 (c +h ~ x :: b) •

Then
g ~ 0

=>

0 = !~ f (x) dg(x) = g(c) - g(a) + !~+h f (x) dg(x).

 6.

Integrate
rh
c
f (x) dg (x)

by parts to get

- g(c) + k~h g(x) dx

==> (h -+ 0)

0 == g(c) - g(a) - g(c) - g(c+)

==>

g(a) == g(c+).
Analogously
a< c ~ b => g(b) = g(c-).]

Step 4: Establish that g E BV [a,b] and if

g(a) = g(c+) = g(c-) = g(b)

when a < c < b, then g ~ 0.

[In fact, g(x) = g(a) at x = a, x = b, and at all interior fX)ints of [a,b]

at which g is continuous, thus V f E C[a,b],

fb f (x) dg(x) = ~ f(x) dh(x) = O,

a a

where h (x) = g (a) .]

Step 5: Every .equivalence class contains at nnst one normalized function.

g (a)
1
= 0, g (a)
2
= O, so

(gl - g2) (a) = O => (gl - g2) (b) =0

 7.

Moreover
g(c+) = g(a) = 0

=> gl (c+) - g2 (c+) = 0

On the other hand,

g1 E NBV [a,b] => g1 (c+) = g1 (c)

g E NEV [a,b] => g (c+) = g (c)

2 2 2

Step 6: Every equivalence class contains at least one normalized function.

[Given g E BV [a,b], define g* E BV [a,b] as follows:

g*(a) = 0, g*(b) = g(b) - g(a)

g*(x) = g(x+) - g(a) (a< x < b).

Then g* E NBV[a,b] and g* . . . g. The verification that g* E NBV[a,b] is irmnediate.

There remains the claim that g* - g ,. ., 0.

• (g* - g) (a)= g*(a) - g(a) =- g(a).

• (g* - g) (b) = g*(b) - g(b) =g(b) - g(a) - g(b) =~ g(a).

When a < x < b,
g*(x) = gr(x) - g(a).

And for c E 1a ,b [ ,
- lim gr(x) = lim g(x)
x+c x+c

lim g (x) = lim g(x).

XtC r Xtc
 8.

• (g* - g) (c+)

= g* (c+) - g (c+)

= gr(c+) - g(a) - g(c+)

= lirn g (x) - g(a) - g(c+)

x+c r

= lirn g(x) - g(a) - g(c+)

Xi-C

= g(c+) - g(a) - g(c+)

- g(a).

• {g* - g) (c-)

= g*(c-) - g(c-)

= lirn g (x) - g(a) - g(c-)

xtc r

= lirn g(x) - g(a) - g{c-)

xtc

= g(c-) - g(a) - g(c-)

=- g(a).
Therefore

g* - g ~ 0 => g* ~ g.]

Step 7:

Tg *[a,b] -< Tg[a,b].

[I.et P E P [a,b]:
 9.

Given e: > 0, ch<x>se FOints y , ••• ,y at which g is continuous with y. ro close

1 n-1 i

to x. (on the right) that

l.

Taking y = a, yn = b, there follOVJs

9

n
E lg* (xi) - g* (xi-1) I
n
= L lg (x.+) - g(a) - g(x. l+) + g (a)
l. l.-
I
i=l
n
< E jg(x.+) - g (yi) -~
- l.

n
+ z:: lg<xi-1+) - g(yi-1) I
i=l

n
+ E lg (yi) - g (yi-1) I
i=l
n
~ L Ig (yi) - g (yi-1) I+ E:

=>

T *[a,b] < T [a,b] + e:

g - g

=> (e: + 0)

T *[a,b] < T [a,b].

g - g

Consider nov:1 the arrow

NBV[a,b] + C[a,b]*

that sends g to A . 'lb see that it is surjective, let A E C[a,b]* and choose
g

a g E BV[a,b] such that

A = /\..
g
 10.

The equivalence class to which g belongs contains a unique normalized element g*,
so g* ~ g

=>

A*=.A.
g g
=.A..

Finally, as regards the nonns,

1111.11 = 1111.g II = 1111.g *II

Meanwhile

Tg*[a,b] = I lg* I IBv =>1 IAI I = I lg* I IBv·

 1.

§17. INTEGRAL MEANS

To simplify the notation, work in [O,l] (the generalization to [a,b] being

straightforward) .

1: NorATICT\I I= [0,1], 0 < 0 < 1, Io= [0,1 - o] (=> 1 - 0 > O)'

o < h < o C=> i -- h > i - o).

2: DEFINITION I.et f E BV [0, 1] and suppose that f is continuous -- then

its integral m=an is the function f- on [O,l - o] defined by the prescription

f-cx) = k~ f (x + t)dt (O < x < 1 .... o) •

3: LEMMA f- E C[Io] and

f- -r f (h -r 0)

unifonnly in I .
0

4: LEMMA The derivative of f- exists in ] 0, 1 - o [ and is given there

by the fonnula

cl1) I (X) = f (X + h) - f(X) •

h

[Note: Therefore /1 has a continuous first d.erivative in the interior of I • ]

0

5: LEMMA

t1- E AC[O, 1 - o].

PROOF let

M = sup If I·
[O ,l]
 2.

Then for fixed h,

I cl1i I (x) I= If (x + ~) ~ f I
(x) (0 < x < 1 - 0)

2M
~ 11"•

Choose a < b such that

0 < a < b < 1 - 8.

Then

=>

lfhCb) - :r11ca) I~ 2M(~ - a) (O < a < b < 1 - 8)

or still, by continuity,

j:r11(b) - l1(a) I~ 2M(~ - a) (0 ~a< b ~ 1 - 8).

And this implies that fh is absolutely continuous.

[In the usual notation,

6: LEMMA let

Then
(O < h < 8) •
 3.

PROOF Take a finite system of intervals [a. ,b.] (1 < i ::: n) without coIIlITOn
l l

interior points in [a,b] -- then

n
L:
. l
If (b.J. + t) - f (a.J. + t) I ~ Tf [a,b + o]
1=

=>
n
L: 1t1cb.)
J..
- t1ca.)
J..
I
i=l

= Tf [a,b + o]
=>

(O < h < o) •

7: THEOREM I.et

Then

(O < h -+ 0) •

PROOF

(O < h < 8)

=>

lim sup T h[a,b] ~ Tf[a,b + o].

h-+ 0 f

Since

it follows that

lim sup T _h [a,b] ~ Tf [a,b].

h -+ 0 f- ~
 4.

By hyp::>thesis , [a ,b] c. I and in I ,

0 0

:ri + f (h + 0)

lUlifonnly, hence pointwise. Therefore

lim inf T_h[a,b] ~ Tf[a,b].

h + 0 f-

8: SCHOLIUM CMing to the absolute continuity of fh in I , for any

0
[a,b] c I , we have
0

= f ~ \f (x + ~) - f (x) I dx
and

!~ \f(x + ~) - f (x) I dx + Tf[a,b] (0 < h -r 0) •

 1.

§18. ESSENTIAL VARIATION

1 1
1: DEFINITIOO BVL ].a,b[ is the subset of L ]a,b[ consisting of those

f whose distributional derivative Df is represented by a finite signed Radon

Ireasure in ] a,b [ of finite total variation, i.e., if

00

f]a,b[ f¢' =- f]a,b[ ¢ d.Df (V ¢ E C ] a,b [)

c

for so.rre finite signed Radon measure Df with

IDfiJ a,b [ < + 00

•

1
[Note: TwO L -functions which are equal almost everywhere define the same

distribution (and so have the sane distributional derivative) • ]

2: N.B. A smoothing argument shc:Ms that the integration by parts

1
fonnula is still true for all ¢ E C ] a ,b [.
c

Of course it may happen that Df is a flmction, say Df = gdx, hence

1
V ¢ E C ]a,b[,
c

f]a,b[ f¢' = - f]a,b[ ¢ gdx.

3: EXAMPLE Work in ] 0, 2 [ and let

x (0 < x < 1)
-
f (x) =
1 (1 < x < 2) •

Put

1 (O < x < 1)
g(x) =
0 (1 < x < 2) •
 2.

1
Then Df = gdx.. In fact, V cp E C ] 0, 2 [,
c

f .2 f cf> I dx = ! 1 Xcf> I dx + !2 cp I dx
0 0 1

=-f ~ cp dx + cp (1) - cp (1)

1 2
=- f 0 cp dx = - f 0 <P gdx.

~: EXAMPLE I.etµ be a finite signed Radon nEasure in ] a,b [. Put f (x) =

µ (]a ,x [) -- then the distributional derivative of f is µ.

1
[\/ ¢ E C ]a,b[,
c

f]a,b[ f(x)¢' (x) dx = f]a,b[ f]a,x[ <P' (x) dµ(y)dx

= f]a,b[ f]y,b[ <P' (x) dxdµ(y)

= - f]a,b[ cp(y) dµ(y).]

5: NOTATION let f:]a,b[ -r R -- then the total variation Tf]a,b[ of fin

] a ,b [ is the supremum of the total variations of f in the closed subintervals of

] a,b [.

6: FACT If f: [a,b] -r R, then

+ jf(a+) ..... f(a) I+ jf(h-) - f (b) I·

7: N. B. Therefore

whenever f is continuous.
 3.

8: DEFINITION A function f: ] a ,b [ + R is of bounded variation in ]a,b [

provided

9: NarATION BV]a,b [ is the set of functions of bo1.ll1ded variation in ]a,b [.

10~ N.B. Elements of BV]a,b [ are bounded, hence are integrable:

1
BV]a,b [ c L ]a,b [.

Moreover, v f E BV] a ,b [,

f (a+)
exist.
f (b-)

11: EXAMPLE Take ]a,b[ = ]O,l[ -- then

1
f (x) =
-x

is increasing and of bounded variation in every closed subinterval of ] 0, l [, yet

f ¢ BV]O,l [.

The initial step in the theoretical development is to characterize the

1
elements of BVL ] a ,b [.

12: FACT I.et µbe a finite signed Radon measure in ]a,b[ - then for

any open set S c ]a ,b I ,

1
13: DEFINITION Given f E L ]a,b[, let
 4.

1
V (f; ] a, b [) = sup u a, b [ f <P ' : <P E C ] a, b [ ,
c l I<P l Ioo -< 1 }.
1

1
14: THEOREM I.et f E L Ja,b [ -- then f E BVL1 ] a,b [ iff
V(f; ]a,b [) < + oo.
And when this is so,

V(f; ]a,b [) = IDf IIa,b [.

1
PROOF Suppose first that f E BVL ]a,b [ -- then

V (f; ]a,b [)

1
=sup {- f]a,b[ <P dDf:¢ E Cc]a,b[, I l<P l 100 _s l}

= sup { - f ]a,b [ <P dDf: <P E Cc]a,b [, I l<P 11 00 ~ l}

= I- Df I] a ,b [

= jDf jJa,b[ < + ~.

Conversely assume that

V(f;]a,b[) < + oo.

Then

!J]a,b[ f¢' I~ V(f;]a,b[) I !<Pl loo·

Since C~] a ,b [ is dense in c 0 ] a ,b [, the linear functional

1
J\:Cc Ja,b[ + R

defined by the rule

<P + 1Ja,b[ f<P'

can be extended uniquely to a continuous linear functional
 5.

where

I !All*_:: V(f;]a,b[).

Thanks to the "Co" version of the RRT, there exists a finite signed Ra.don measure

µin ]a,b[ such that

I IA 11* = Iµ I <Ja,b D
and

A(¢) =J]a,b[ <Pdµ (v¢ EC ]a,b[).

0

Definition:

Df = µ

=>

IDf I] a,b [ = Iµ I (] a,b [)

= I IA! I*
< V(f;]a,bl) < + oo.

15: LEMMA The map

f -+ V (f; ] a ,b [)

is la.ver semicontinuous in the Lioc]a,b[ to}?Ology.

16: APPLICATION The map

f -+ IDf I] a ,b [

is lo.r.rer sernicontinuous in the Lioc] a,b [ to}?Ology.

17: SUBLEMMA Any element of BV] a,b [ can be represented as the difference

of two bounded increasing functions.

18: LEMMA 'v' f E BV]a,b[,

V(f;]a,b[) ~ Tf]a,b[ ( < + oo).

 6.

PROOF Construct a sequence Xn of step functions such that

Xn -+ f (n -+ oo)

in Lioc]a,b [ and v n,

Thanks nCNJ to lower semicontinuity,

V (f; ]a,b [) < lim inf V(Xn; ]a,b [)

n -+ oo

19: 92IDLIUM
1
BV]a,b[ cBVL ]a,b[.

[Note: If f: [a,b] -+ R is in BV [a,b], then its restriction to ]a,b [ is in

1
BV]a,b[, hence is in BVL ]a,b[.]

1
20 : DEFINITIQ.\J I.et f E L ] a ,b [ -- then the essential variation of f,

denoted e - Tf]a,b [, is the set

inf {T ] a ,b [ : g = f a.1.most every.where} •

g
1
[Note: If f ,f E L ]a,b[ and if f = f alroost everywhere, then
1 2 1 2

e - Tf ]a,b[ = e - Tf ]a,b[.]
1 2

1
21: LEMMA I.et f E L ] a ,b [ -- then

e - Tf]a,b[ = V(f;]a,b[).

Consequently
1
22: THEOREM I.et f E L ]a,b[ - then
 7.

1
e - Tf]a,b [ < + oo <=> f E BVL ]a,b [.

And then

ID£ IJa,b [ == e - Tf ]a,b [.

1
23: LEMMA I.et f E BVL ]a,b [. Assl..nte: Df == O -- then f is (equivalent

to) a unique constant.

1
Assuming still that f E BVL ]a,b [, let µ == Df and put w(x) µ(]a,x[) - then

ow-== µ, thus D(f-w) 2: O, so there exists a unique constant C such that

f =C+w

almost everywhere.

24: LEMMA

PROOF Take points

in ]a,b[ -- then

n
l: I(C+w) (xi) - (C+w) (xi-l) I < lu I (]a,b[)
i=l

==>

TC+w]a,b[ ~ V(f;]a,b[)

== e - Tf]a,b[.

1 1
25: DEFINITION Given f E BVL ]a,b[, a function g E L ]a,b[ such that

g = f almost everywhere is admissible

Tg]a,b[ =e - Tf]a,b[.
 8.

[Note: Since

this says that f is equivalent to g, where g E BV]a,b [.]

So, in this tenninology, C+w is admissible, i.e. ,

fl(x) :: C + Df]a,x [

is admissible, the same being the case of

~ (x) C + Df]a,x].

26: LEMMA

fl is left continuous

~ is right continuous.

27 : RErJIARK

- l _!
f (x) - f-(y) = Df [y,x[
(a < y < x < b).

fr(x) - ~(y) = Df]y,x]

1
28: THEOREM A function g E L Ja,b[ is admissible iff

g E {e~ + (1 - e )~:o ~ e ~ l}.

29: N.B. Denote by ATf the atans of the theory, i.e., the x E ] a,b [

such that Df ({x}) ':/ 0 -- then I = ~ in ]a,b[ - AT£ and every admissible g is

continuous in ]a,b[ - ATf.

1
30: LEMMA Suppose that g E L Ja ,b [ is admissible .....- then g is differentiable

almost everywhere and its derivative g' is the density of Df w.r.t. Lebesgue

measure.
 9.

There is a characterization of the essential variation which is purely

internal.

31: NorATIONGiven an f E L 1 Ja,b[, let Cap (f) stand for its set of points
of approximate continuity.
[Recall that Cap (f) is a subset of ]a,b[ of full measure.]

32: LEMMA

n
e - Tf]a,b[ =sup L !f(xi) - f(xi-l) I,
i=l
where the supremum is taken over all finite collections of points xi E Cap(f)
subject to
 1.

§19. BVC

1: NOI1ATION Given a subset M c ] a ,b [ of Lebesgue measure O, denote by

~M] a,b [ the set of all sequences

P:x0 < x 1 < • • • < xn '

where

x < b
n

and

x.
l
E ]a,b[ - M (i = 0,1, ••• ,n).

[Note: The possibility that M = ~ is not excluded"]

2: NOI1ATIQ.1\J Given a function f: ] a ,b [ -+ R, let fM be the resriction of

f to ]a,b[ - M.

3: NorATION Given an ele:rrEnt p E pM] a' b [ ' put

b n
~ (fM;P) = i:l j~(xi) - fM(xi-1) j.

4: NorATION Given a function f: ] a ,b [ -+ R, put

b
Tf ] a,b [ = sup V (fM;P).
M PEP a,b[ a M]

5: DEFINITIOO Tf ]a,b[ is the total variation of fM in ]a,b[ - M.

M

1
6: DEFINITION A function f E L ]a,b[ is said to be of bounded variation
 2.

in the sense of Cesari if there exists a subset M c ]a ,b [ of Iebesgue measure O

such that

T~ ]a,b [ < + oo.

7: NDrATION BVC]a,b [ is the set of functions which are of bounded

variation in the sense of Cesari.

8: EXAMPLE
BV]a,b [ c BVC]a,b [ (M = fO).

9: THEOREM
1
BVC]a,b [ = BVL ]a,b [.

Proceed via a couple of lemmas.

1
10: LEMMA Suppose that f E BVL Ja,b [ -- then f E BVC]a,b [.

PROOF The assumption that

1
f E BVL ]a,b[ => e - Tf]a,b[ < + 00 •

So there exists a g: g = f almost everywhere and

T ] a,b [ <
g
+ 00 •

Take noo for M the set of x such that g (x) :I f (x) , the complement ] a ,b [ - M being

the set of x where g (x) = f (x) • Consider a typical sum

which is equal to
n
E !g (xi) - g (xi-1) !
i=l
 3.

which is less than or equal to

T ] a, b [ <
g
+ co.

Therefore f E BVC]a,b[.

11: SUBLEMMA If Tf ]a,b [ < + co, then there exists a g: ]a,b [ + R such that
M

T ]a,b [ = Tf ]a,b [.
g M

12: 1
LEMMA Suppose that f E BVC]a,b [ -- then f E BVL ]a,b [.

PROOF The assumption that f E BVC]a,b [ produces an "M" and from the preceding

consideration,

gM = fM => g IJa,b[ - M = f IJa,b[ - M,

hence g = f almost everywhere. But

T~]a,b[ < + co=> Tg]a,b[ < + co

1
=> g E BV]a,b [ => g E BVL ]a,b [.

Since g = f al.Irost everywhere, they have the same distributional derivative, thus

1
f E BVL ]a,b[.

Let Mbe the set of all subsets of ] a ,b [ of Lebesgue measure 0.

1
13: NorATION Given an f E BVL Ja ,b [, put

~(f) = inf Tf ]a,b[.

MEM M

14: THEOREM
 4.

PROOF To begin with,

1
f E BVL ]a,b[ => e -Tf]a,b[ < + oo.

On the other hand, f E BVC]a,b [, so there exists M E M:

Tf ]a,b [ < + oo => cp(f) < + oo.

M

[Denote by Mf the subset of Mconsisting of those M such that Tf ]a,b [ < + oo.
M

Assign to each M E Mf a flmction g: ]a,b [ + R such that gM = fM and

T ]a,b[ = Tf ]a,b[.
g M

Therefore

{Tf ]a,b[:M E Mf}

M

c {T }a,bI:g = f almost everywhere}

g
=>

cp(f) = inf Tf ]a,b[

MEMf M

[Denote by ~ the subset of M consisting of those M that arise from the

elements Tg]a,b[ in the set defining e - Tfla,b[ (i.e., per the requirement that

g = f al.most everywhere.) -- then

 5.

hence

~(f) = inf Tf ]a,b[

MEM M

~ inf Tf la,b [
ME~ M

::: inf {Tg] a,b [ :g =f almost everywhere}

=e - Tf] a,b [.]

1
15: THEOREM I.et f E BVL Ja ,b [ - then there exists a g E BV] a ,b [ which

is equal to f almost everywhere and has the property that

~(£) = Tg ]a,b[.
PRCDF Take g admissible:
 1.

§20~ ABSOLUTE CONTINUITY III

1: DEFINITION A function f: ]a,b [ -+ R is said to be absolutely continuous

in ]a,bl if for ever:y e: > 0 there exists cS > 0 such that for any collection of

non overlapping closed intervals

then

n n
L (1\: - ~) < cS => L 1£ (~) - f ('1<:) I < e:.
k=l k=l

2: NorATION AC]a,b [ is the set of absolutely continuous functions in ]a,b [.

3: N.B. An absolutely continuous function f: ]a,b [ -+ R is unifonnly con-

tinuous.

4: RAPPEL A unifonnly continuous function f: ] a,b [ -+ R can be extended

uniquelll to [a,b] in such a way that the extended function remains unifonnly

continuous.

5: LEl.\1MA If f E AC]a,b [, then its extension to [a,b] belongs to AC [a,b].

6: THEOREM Let f:] a ,b [ -+ R -- then f is absolutely continuous iff the

following four conditions are satisfied.

(1) f is continuous.

(2) f' exists a.lrrost everywhere.

(3) f' E LP]a,b[ for some 1 ~ p < + oo.

(4) V x,x E ] a,b[,

0

f(x) = f(x ) + ?:_ ft dL 1 •

0 XO
 2.

1
Here (and infra),, L is Lebesgue measure on ] a,b [.

7: N.B. For the record,

p 1
L ]a, b[c L ]a,b[ (l _s p < + oo).

8: DEFINITION I.et 1 ~ p < + oo -- then a function f E L~oc] a ,b .[ adrnits a

weak derivative in LPia,b[ if there exists a function - : E rfJ a,b [ such that

df 1 l
J]a,b[¢ dx dl = - J]a,b[ ¢' fdL •

9: CRITERIOO If f E L~oc]a,b[ and if V ¢ E C~]a,b[,

1
J]a,b[ ¢f dl = O,

then f = O almost everywhere.

10: SCHOLIUM A weak derivative of f in LP]a,b[, if it exists at all,

is unique up to a set of Lebesgue measure 0 • For suppose you have two weak

derivatives u,v in rPJa,b[, thus V ¢ E C~J.a,b[,

J ]a,b[ '¥~u dL J]a,b[ 'I'~· fdL

1 1
· = -

J ]a,b[ 'I'~v dL = - J Ia,b[ 'I'~· fdL

1 1

=>
1
J]a,b[ ¢(u - v) dL =0
00 • •
and so u = v a.lnost everywhere, ¢ E Cc]a,b[ being arbitrary.

11: N.B .. If f,g E L; ]a,b[ are equal a.J..rrost everywhere, then they have
-- .{..0C
the nsame weak derivative.
11

df 1 1
f Ja,b [ ¢ax dL = - J]a,b [ m1 fdL
'+'

=- J]a,b[ cp I gdll

= J]a,b ['+'ax
dg ,+. dll

so
df - dg
dx - dx

almost everywhere. ]

12: LEMMA I.et f, g E L;, ] a ,b [ and suppose that each of them a.dnlits a
- -{,0C

weak derivative -- then f + g admits a weak derivative and

d
ax (f + )
g
= ax
df + dg
ax·
00
PROOF V ¢ E C ]a,b[,
c

- df 1 dg 1
- 11a,b[ <P dx dl + 1 Ja,b[ <Pax dL

=- J]a,b[ <P' fdL 1 - J]a,b[ <P' gdL 1

= - f]a,b[ ~· (f + gtdll.

13: LEMMA If 1fJ E C~]a,b[ and if f admits a weak derivative : , then

 4.

1J;f admits a weak derivative and

! (1J;f) = 1JJ'f + 1J; : •

00
PROOF V <P E C ] a ,b [,
c

f] a,b [ <P' (1JJf) dl 1 = f] a,b [ (f (1};¢) ' - f (1/J • <P)) dl 1

00
14: SUBLEMMA Given <P E C ] a,b [, let
c
1
<I> (x) = f] a,x [ <P dl

and suppose that

1
f]a,b[ <P dl = O.
00
Then <I> E C ]a,b[.
c

15: LEMMA let f E L; ]a,b[ and assume that f has weak derivative 0 --
- ..t..OC

then f coincides alrrost everywhere in ] a,b [ with a constant flmction.

PROOF Fix 1/Jo E C~]a,b[: J]a,b[ 1/Jo dL

1
=1, and given any <PE C~]a,b[, put

1
I(cp) = f]a,b[ <P dl -- then

I(¢ - I(¢)1/Jo) = I(¢) - I(¢)I(1/Jo) = 0,

hence

Since f has weak derivative 0,

f ] a,b [ w ddxf dll = O,

T
 5.

=>

1
O = 11a,b[ \!'' ·fdL

1
= 11a,b [ (¢ - I(¢)1/J0 )fdl

1 1 1
= 11a,b [ ¢ fdL - (J]a,b[ ¢ dL ) (J]a,b[ f1jJ0dL )

1
= 1J a,b [ ¢(f - c0 )dl ,

where

Therefore f - c0 = 0 alrrost everywhere or still, f = c0 almost everywhere.

16: NDrATION Let 1 ~ p <+ oo -- then w11 P]a,b [ is the set of all functions

f E rY]a,b [ which possess a weak derivative : in rJ']a,b [.

1
17: N.B. w1 11 Ja,b [ is contained in BVL Ia,b [.

[Take an f E w1' 1 Ja,b [ and consider

Df (E) = f df dLl (E E BO]a,b [),

E dx
i.e.'

00
Then V ¢ E C ]a,b [,
c
 6.

so by definition, f E Bvr.ha,b[. ]

[Note: The contaimnent is strict.]

18: THEOREM Let 1 ~ p < + oo - - then a flIDction f:] a ,b [ + R belongs to

w1'P]a,b[ iff it admits an absolutely continuous representative f:]a,b[ + R such

that f and its ordinary derivative f' belong to LP]a,b[.

00
19: LEMMA If f E AC]a,b[, then V ¢ E C ]a,b[,
c

there being no boundary term in the (implicit) integration by parts since <P has

canpact support in ] a ,b [.

20: SCHOLIUM If f is absolutely continuous, then its ordinary derivative

f • is a weak derivative.

One direction of the theorem is imnediate. For suppose that f: ] a ,b [ + R

admits an absolutely continuous representative f :]a,b[ + R such that f and f'

are in LP]a,b[ -- then the claim is that f E w1 1 P]a,b[. Of course, f E LP]a,b[.

As for the existence of the weak derivative : , note that v <P E C~] a,b [,
- 1 - 1
f]a,b[ <P f'dl =- f]a,b[ ¢' fdL
or still,
- 1 1
J]a,b[ <P f 'dL =- f ]a,b[ ¢' fdL ,

since f =f alnost everywhere. Therefore f' a weak derivative of f in LP]a,b[.

Turning to the converse, let f E w1 1 P]a,b[, fix a point x E ]a,b[, and put
0
 7.

(x E ] a,b [).

Then f E AC]a,b[ and alnost everywhere,

i.e., abnost everywhere,

f' - df
dx
=0 1

or still, almost everywhere,

d -
dx (f - f) = 0,

which implies that there exists a constant C such that f - f C abnost everywhere,
thus f has an absolutely continuous representative f such that it and its ordinary
derivative belong to LP]a,b[.

21: REMARK Matters simplify slightly when p = l:f E w1 11]a,b[ iff f

admits an absolutely continuous representative f.