Regulated functions and the regulated integral

Jordan Bell
April 3, 2014

1 Regulated functions and step functions

Let I=[a,b] and let X be a normed space. A function f:I→X is said to be regulated if for all t∈[a,b) the limit lims→t+⁡f⁢(s) exists and for all t∈(a,b] the limit lims→t-⁡f⁢(s) exists. We denote these limits respectively by f⁢(t+) and f⁢(t-). We define R⁢(I,X) to be the set of regulated functions I→X. It is apparent that R⁢(I,X) is a vector space. One checks that a regulated function is bounded, and that R⁢(I,X) is a normed space with the norm ∥f∥∞=supt∈[a,b]⁡∥f⁢(t)∥.

Theorem 1.

If I is a compact interval in R and X is a normed algebra, then R⁢(I,X) is a normed algebra.

Proof.

If f,g∈R⁢(I,X), then f⁢g∈R⁢(I,X) because the limit of a product is equal to a product of limits. For t∈I we have

∥(f⁢g)⁢(t)∥=∥f⁢(t)⁢g⁢(t)∥≤∥f⁢(t)∥⁢∥g⁢(t)∥≤∥f∥∞⁢∥g∥∞,

so ∥f⁢g∥∞≤∥f∥∞⁢∥g∥∞. ∎

A function f:I→X, where I=[a,b], is said to be a step function if there are a=s0<s1<⋯<sk=b for which f is constant on each open interval (si-1,si). We denote the set of step functions I→X by S⁢(I,X). It is apparent that S⁢(I,X) is contained in R⁢(I,X) and is a vector subspace, and the following theorem states that if X is a Banach space then S⁢(I,X) is dense in R⁢(I,X).11 1 Jean Dieudonné, Foundations of Modern Analysis, enlarged and corrected printing, p. 145, Theorem 7.6.1; Rodney Coleman, Calculus on Normed Vector Spaces, p. 70, Proposition 3.3; cf. Robert G. Bartle, A Modern Theory of Integration, p. 49, Theorem 3.17.

Theorem 2.

Let I be a compact interval in R, let X be a Banach space, and let f∈XI. f∈R⁢(I,X) if and only if for all ϵ>0 there is some g∈S⁢(I,X) such that ∥f-g∥∞<ϵ.

We prove in the following theorem that the set of regulated functions from a compact interval to a Banach space is itself a Banach space.

Theorem 3.

If I is a compact interval in R and X is a Banach space, then R⁢(I,X) is a Banach space.

Proof.

Let fn∈R⁢(I,X) be a Cauchy sequence. For each t∈I we have

∥fn⁢(t)-fm⁢(t)∥≤∥fn-fm∥∞,

hence fn⁢(t) is a Cauchy sequence in X. As X is a Banach space, this Cauchy sequence converges to some limit, and we define f⁢(t) to be this limit. Thus f∈XI and ∥f-fn∥∞→0. We have to prove that f∈R⁢(I,X). Let ϵ>0. There is some N for which n≥N implies that ∥f-fn∥∞<ϵ; in particular, ∥f-fN∥∞<ϵ. By Theorem 2, there is some gN∈S⁢(I,X) with ∥fN-gN∥∞<ϵ. Then,

∥f-gN∥∞≤∥f-fN∥∞+∥fN-gN∥∞<2⁢ϵ,

and by Theorem 2 this implies that f∈R⁢(I,X). ∎

The following lemma shows that the set of points of discontinuity of a regulated function taking values in a Banach space is countable.

Lemma 4.

If I is a compact interval in R, X is a Banach space, and f∈R⁢(I,X), then

{t∈I:f is discontinuous at t}

is countable.

Proof.

For each n let gn∈S⁢(I,X) satisfy ∥f-gn∥≤1n, and let

Dn={t∈I:gn is discontinuous at t}.

gn is a step function so Dn is finite, and hence D=⋃n=1∞Dn is countable. It need not be true that f is discontinuous at each point in D, but we shall prove that if t∈I∖D then f is continuous at t, which will prove the claim.

Suppose that t∈I∖D, let ϵ>0, and take N>1ϵ. As t∉DN, the step function gN is continuous at t, and hence there is some δ>0 for which |s-t|<δ implies that ∥gN⁢(s)-gN⁢(t)∥<ϵ. If |s-t|<δ, then

∥f⁢(s)-f⁢(t)∥ ≤ ∥f⁢(s)-gN⁢(s)∥+∥gN⁢(s)-gN⁢(t)∥+∥gN⁢(t)-f⁢(t)∥
≤ 2⁢∥f-gN∥∞+∥gN⁢(s)-gN⁢(t)∥
< 2N+ϵ
< 3⁢ϵ,

showing that f is continuous at t. ∎

2 Integrals of step functions

Let I=[a,b] and let X be a normed space. If f∈S⁢(I,X) then there is a subdivision a=s0<s1<⋯<sk=b of [a,b] and there are ci∈X such that f takes the value ci on the open interval (si-1,si). Suppose that there is a subdivision a=t0<t1<⋯<tl=b of [a,b] and di∈X such that f takes the value di on the open interval (ti-1,ti). One checks that

∑i=1k(si-si-1)⁢ci=∑i=1l(ti-ti-1)⁢di.

We define the integral of f to be the above element of X, and denote this element of X by ∫If=∫abf.

Lemma 5.

If I is a compact interval in R and X is a normed space, then ∫I:S⁢(I,X)→X is linear.

Lemma 6.

If I=[a,b] and X is a normed space, then ∫I:S⁢(I,X)→X is a bounded linear map with operator norm b-a.

Proof.

If f∈S⁢(I,X), let a=s0<s1<⋯<sk=b be a subdivision of [a,b] and let ci∈X such that f takes the value ci on the open interval (si-1,si). Then,

∥∫If∥≤∑i=1k(si-si-1)⁢∥ci∥≤∑i=1k(si-si-1)⁢∥f∥∞=(b-a)⁢∥f∥∞.

This shows that ∥∫I∥≤b-a, and if f is constant, say f⁢(t)=c∈X for all t∈I, then ∫If=(b-a)⁢c and ∥∫If∥=(b-a)⁢∥c∥=(b-a)⁢∥f∥∞, showing that ∥∫I∥=b-a. ∎

Lemma 7.

If a≤b≤c, if X is a normed space, and if g∈S⁢([a,c],X), then

∫acg=∫abg+∫bcg.

3 The regulated integral

Let I be a compact interval in ℝ and let X be a Banach space. Theorem 2 shows that S⁢(I,X) is a dense subspace of R⁢(I,X), and therefore if T0∈ℬ⁢(S⁢(I,X),X) then there is one and only one T∈ℬ⁢(R⁢(I,X),X) whose restriction to S⁢(I,X) is equal to T0, and this operator satisfies ∥T∥=∥T0∥. Lemma 6 shows that ∫I:S⁢(I,X)→X is a bounded linear operator, thus there is one and only one bounded linear operator R⁢(I,X)→X whose restriction to S⁢(I,X) is equal to ∫I, and we denote this operator R⁢(I,X)→X also by ∫I. With I=[a,b], we have ∥∫I∥=b-a. We call ∫I:R⁢(I,X)→X the regulated integral.

Lemma 8.

If a≤b≤c, if X is a Banach space, and if f∈R⁢([a,c],X), then

∫acf=∫abf+∫bcf.
Proof.

Let I1=[a,b], I2=[b,c], I=[a,c], and let f1 and f2 be the restriction of f to I1 and I2 respectively. From the definition of regulated functions, f1∈R⁢(I1,X) and f2∈R⁢(I2,X). By Theorem 2, for any ϵ>0 there is some g∈S⁢(I,X) satisfying ∥f-g∥∞<ϵ. Taking g1 and g2 to be the restriction of g to I1 and I2, we check that g1∈S⁢(I1,X) and g2∈S⁢(I2,X). Then by Lemma 7,

∥∫If-∫I1f1-∫I2f2∥∞ ≤ ∥∫If-∫Ig∥∞+∥∫Ig-∫I1g1-∫I2g2∥∞
+∥∫I1g1+∫I2g2-∫I1f1-∫I2f2∥∞
= ∥∫I(f-g)∥∞+0
+∥∫I1(g1-f1)∥∞+∥∫I2(g2-f2)∥∞
≤ (c-a)⁢∥f-g∥∞+(b-a)⁢∥g1-f2∥∞
+(c-b)⁢∥g2-f2∥∞.

But ∥g1-f1∥∞≤∥g-f∥∞ and ∥g2-f2∥∞≤∥g-f∥∞, hence we obtain

∥∫If-∫I1f1-∫I2f2∥∞<(c-a)⁢ϵ+(b-a)⁢ϵ+(c-b)⁢ϵ=2⁢(c-a)⁢ϵ.

Since ϵ>0 was arbitrary, we get

∥∫If-∫I1f1-∫I2f2∥∞=0,

so

∫If=∫I1f1+∫I2f2,

proving the claim. ∎

We prove that applying a bounded linear map and taking the regulated integral commute.22 2 Jean-Paul Penot, Calculus Without Derivatives, p. 124, Proposition 2.18.

Lemma 9.

Suppose that I is a compact interval in R and that X and Y are Banach spaces. If f∈R⁢(I,X) and T∈B⁢(X,Y), then T∘f∈R⁢(I,Y) and

∫IT∘f=T⁢∫If.
Proof.

Because T is continuous we have T∘f∈R⁢(I,Y). For ϵ>0, there is some g∈S⁢(I,X) satisfying ∥f-g∥∞<ϵ. Write I=[a,b]. Because g is a step function, there is a subdivision a=s0<s1<⋯<sk=b of I and there are ci∈X such that g takes the value ci on the open interval (si-1,si). Furthermore, T∘g takes the value T⁢ci on the open interval (si-1,si) so T∘g∈S⁢(I,Y), and then because T is linear,

∫IT∘g=∑i=1k(si-si-1)⁢T⁢ci=T⁢∑i=1k(si-si-1)⁢ci=T⁢∫Ig.

Using this,

∥∫IT∘f-T⁢∫If∥ ≤ ∥∫IT∘f-∫IT∘g∥+∥∫IT∘g-T⁢∫Ig∥
+∥T⁢∫Ig-T⁢∫If∥
= ∥∫IT∘(f-g)∥+∥T⁢∫I(f-g)∥
≤ (b-a)⁢∥T∘(f-g)∥∞+∥T∥⁢∥∫I(f-g)∥
≤ (b-a)⁢∥T∥⁢∥f-g∥∞+∥T∥⁢(b-a)⁢∥f-g∥∞
< 2⁢(b-a)⁢∥T∥⁢ϵ.

As ϵ>0 is arbitrary, this means that

∥∫IT∘f-T⁢∫If∥=0,

and so

∫IT∘f=T⁢∫If.

∎

4 Left and right derivatives

Suppose that I is an open interval in ℝ, X is a normed space, f∈XI, and t∈I. We say that f is right-differentiable at t if f⁢(t+h)-f⁢(t)h has a limit as h→0+, and that f is left-differentiable at t if f⁢(t+h)-f⁢(t)h has a limit as h→0-. We call these limits respectively the right derivative of f at t and the left derivative of f at t, denoted respectively by f+′⁢(t) and f-′⁢(t). For f to be differentiable at t means that f+′⁢(t) and f-′⁢(t) exist and are equal.

The following is the mean value theorem for functions taking values in a Banach space.33 3 Henri Cartan, Differential Calculus, p. 39, Theorem 3.1.3.

Theorem 10 (Mean value theorem).

Suppose that I=[a,b], that X is a Banach space, and that f:I→X and g:I→R are continuous functions. If there is a countable set D⊂I such that t∈I∖D implies that f+′⁢(t) and g+′⁢(t) exist and satisfy ∥f+′⁢(t)∥≤g+′⁢(t), then

∥f⁢(b)-f⁢(a)∥≤g⁢(b)-g⁢(a).
Corollary 11.

Suppose that I=[a,b], that X is a Banach space, and that f:I→X is continuous. If there is a countable set D⊂I such that t∈I∖D implies that f+′⁢(t)=0, then f is constant on I.

5 Primitives

Let I=[a,b], let X be a normed space, and let f,g∈XI. We say that g is a primitive of f if g is continuous and if there is a countable set D⊂I such that t∈I∖D implies that g is differentiable at t and g′⁢(t)=f⁢(t).

Lemma 12.

Suppose that I is a compact interval in R, that X is a Banach space, and that f:I→X is a function. If g1,g2:I→X are primitives of f, then g1-g2 is constant on I.

Proof.

For i=1,2, as gi is a primitive of f there is a countable set Di⊂I such that t∈I∖Di implies that gi is differentiable at t and gi′⁢(t)=f⁢(t). Let D=D1∪D2, which is a countable set. Both g1 and g2 are continuous so g=g1-g2:I→X is continuous, and if t∈I∖D then g is differentiable at t and g′⁢(t)=g1′⁢(t)-g2′⁢(t)=f⁢(t)-f⁢(t)=0. Then Corollary 11 shows that g is constant on I, i.e., that g1-g2 is constant on I.∎

We now give a construction of primitives of regulated functions.44 4 Jean-Paul Penot, Calculus Without Derivatives, p. 124, Theorem 2.19.

Theorem 13.

If I=[a,b], X is a Banach space, and f∈R⁢(I,X), then the map g:I→X defined by g⁢(t)=∫atf is a primitive of f on I.

Proof.

For t∈[a,b) and ϵ>0, because f is regulated there is some 0<δ<b-t such that 0<r≤δ implies that ∥f⁢(t+r)-f⁢(t+)∥≤ϵ. For 0<r≤δ and for any 0<η<r, using Lemma 8 we have

∥∫at+rf-∫atf-∫tt+rf⁢(t+)∥ = ∥∫tt+rf-∫tt+rf⁢(t+)∥
= ∥∫tt+η(f-f⁢(t+))+∫t+ηt+r(f-f⁢(t+))∥
≤ η⁢supt≤s≤t+η⁡∥f⁢(s)-f⁢(t+)∥
+(r-η)⁢supt+η≤s≤t+r⁡∥f⁢(s)-f⁢(t+)∥
≤ 2⁢∥f∥∞⁢η+(r-η)⁢ϵ.

This is true for all 0<η<r, so

∥∫at+rf-∫atf-∫tt+rf⁢(t+)∥≤r⁢ϵ,

i.e.

∥g⁢(t+r)-g⁢(t)r-f⁢(t+)∥≤ϵ.

This shows that

g+′⁢(t)=f⁢(t+).

Similarly,

g-′⁢(t)=f⁢(t-).

Because f is regulated, Lemma 4 shows that there is a countable set D⊂I such that t∈I∖D implies that f is continuous at t. Therefore, if t∈I∖D then f⁢(t+)=f⁢(t-)=f⁢(t), so g+′⁢(t)=g-′⁢(t), which means that if t∈I∖D then g is differentiable at t, with g′⁢(t)=f⁢(t). To prove that g is a primitive of f on I it suffices now to show that g is continuous. For ϵ>0 and t∈I, let δ=ϵ∥f∥∞, and then for |s-t|<δ we have by Lemma 8 that

∥g⁢(s)-g⁢(t)∥=∥∫asf-∫atf∥=∥∫stf∥≤|t-s|⁢∥f∥∞⁢<δ∥⁢f∥∞=ϵ,

showing that g is continuous at t, completing the proof. ∎

Suppose that X is a Banach space and that f:[a,b]→X is a primitive of a regulated function h:[a,b]→X. Because h is regulated, by Theorem 13 the function g:[a,b]→X defined by g⁢(t)=∫atf is a primitive of f on [a,b]. Then applying Lemma 12, there is some c∈X such that f⁢(t)-g⁢(t)=c for all t∈[a,b]. But f⁢(a)-g⁢(a)=f⁢(a), so c=f⁢(a). Hence, for all t∈[a,b],

f⁢(t)=f⁢(a)+∫ath.

But

∫ath=∫aa+η1h+∫a+η1t-η2h+∫t-η2th=∫aa+η1h+∫a+η1t-η2f′+∫t-η2th

and

∥∫aa+η1h∥≤η1⁢∥h∥∞,∥∫t-η2th∥≤η2⁢∥h∥∞,

hence as η1→0+ and η2→0+,

∫a+η1t-η2f′→∫ath,

and so it makes sense to write

∫atf′=∫ath,

and thus for all t∈[a,b],

f⁢(t)=f⁢(a)+∫atf′.