8.10: Integration in Generalized Measure Spaces
- Page ID
- 32381
( \newcommand{\kernel}{\mathrm{null}\,}\)
Let (S,M,s) be a generalized measure space. By Note 1 in §3, a map f is s-measurable iff it is vs-measurable. This naturally leads us to the following definition.
Definition
A map f:S→E is s-integrable on a set A iff it is vs-integrable on A. (Recall that vs, the total variation of s, is a measure.)
Note 1. Here the range spaces of f and s are assumed complete and such that f(x)sA is defined for x∈S and A∈M. Thus if s is vector valued, f must be scalar valued, and vice versa. Later, if a factor p occurs, it must be such that pf(x)sA is defined, i.e., at least two of p,f(x), and sA are scalars.
Note 2. If s is a measure (≥0), then vs=s+=s (Corollary 3 in Chapter 7, §11); so our present definition agrees with the previous ones (as in Theorem 1 of §7).
Lemma 8.10.1
If m′ and m′′ are measures, with m′≥m′′ on M, then
∫A|f|dm′≥∫A|f|dm′′
for all A∈M and any f:S→E.
- Proof
-
First, take any elementary and nonnegative map g≥|f|,
g=∑iCAiai on A.
Then (§4)
∫Agdm′=∑aim′Ai≥∑aim′′Ai=∫Agdm′′.
Hence by Definition 1 in §5,
∫A|f|dm′=infg≥|f|∫Agdm′≥infg≥|f|∫Agdm′′=∫A|f|dm′′,
as claimed.◻
Lemma 8.10.2
(i) If s:M→En(Cn) with s=(s1,…,sn), and if f is s-integrable on A∈M, then f is sk-integrable on A for k=1,2,…,n.
(ii) If s is a signed measure and f is s-integrable on A, then f is integrable on A with respect to both s+ and s− (with s+ and s− as in formula (3) in Chapter 7, §11).
Note 3. The converse statements hold if f is M-measurable on A.
- Proof
-
(i) If s=(s1,…,sn), then (Problem 4 of Chapter 7, §11)
vs≥vsk,k=1,…,n.
Hence by Definition 1 and Lemma 1, the s-integrability of f implies
∞>∫A|f|dvs≥∫A|f|dvsk.
Also, f is vs-measurable, i.e., M-measurable on A−Q, with
0=vsQ≥vskQ≥0.
Thus f is sk -integrable on A,k=1,…,n, as claimed.
(ii) If s=s+−s−, then by Theorem 4 in Chapter 7, §11, and Corollary 3 there, s+ and s− are measures (≥0) and vs=s++s−, so that both
vs≥s+=vs+ and vs≥s−=vs−.
Thus the desired result follows exactly as in part (i) of the proof.◻
We leave Note 3 as an exercise.
Definition
If f is s-integrable on A∈M, we set
(i) in the case s:M→E∗,
∫Afds=∫Afds+−∫Afds−,
with s+ and s− as in formula (3) of Chapter 7, §11;
(ii) in the case s:M→En(Cn),
∫Afds=n∑k=1→ek∫Afdsk,
with →ek as in Theorem 2 of Chapter 3, §§1-3;
(iii) if s:M→C,
∫Afds=∫Afdsre+i⋅∫Afdsim.
(See also Problems 2 and 3.)
Note 4. If s is a measure, then
s=s+=sre=s1
and
0=s−=sim=s2;
so Definition 2 agrees with our previous definitions. Similarly for s:M→ En(Cn).
Below, s,t, and u are generalized measures on M as in Definition 2, while f,g:S→E are functions, with E a complete normed space, as in Note 1.
Theorem 8.10.1
The linearity, additivity, and σ-additivity properties (as in §7, Theorems 2 and 3) also apply to integrals
∫Afds,
with s as in Definition 2.
- Proof
-
(i) Linearity: Let f,g:S→E be s-integrable on A∈M. Let p,q be suitable constants (see Note 1).
If s is a signed measure, then by Lemma 2(ii) and Definitions 1 and 2, f is integrable with respect to vs,s+, and s−. As these are measures, Theorem 2 in §7 shows that pf+qg is integrable with respect to vs,s+, and s−, and by Definition 2,
∫A(pf+qg)ds=∫A(pf+qg)ds+−∫A(pf+qg)ds−=p∫Afds++q∫Agds+−p∫Afds−−q∫Agds−=p∫Afds+q∫Agds.
Thus linearity holds for signed measures. Via components, it now follows for s:M→En(Cn) as well. Verify!
(ii) Additivity and σ-additivity follow in a similar manner.◻
Corollary 8.10.1
Assume f is s-integrable on A, with s as in Definition 2.
(i) If f is constant (f=c) on A, we have
∫Afds=c⋅sA.
(ii) If
f=∑iaiCAi
for an M-partition {Ai} of A, then
∫Afds=∑iaisAi and ∫A|f|ds=∑i|ai|sAi
(both series absolutely convergent).
(iii) |f|<∞ a.e. on A.
(iv) ∫A|f|dvs=0 iff f=0 a.e. on A.
(v) The set A (f≠0) is (vs) σ-finite (Definition 4 in Chapter 7, §5).
(vi) ∫Afds=∫A−Qfds if vsQ=0 or f=0 on Q (Q∈M).
(vii) f is s-integrable on any M-set B⊆A.
- Proof
-
(i) If s=s+−s− is a signed measure, we have by Definition 2 that
∫Afds=∫Afds+−∫Afds−=c(s+A−s−A)=c⋅sA,
as required.
For s:M→En(Cn), the result now follows via components. (Verify!)
(ii) As f=ai on Ai, clause (i) yields
∫Aifds=aisAi,i=1,2,….
Hence by σ-additivity,
∫Afds=∑i∫Aifds=∑iaisAi,
as claimed.
Clauses (iii), (iv), and (v) follow by Corollary 1 in §5 and Theorem 1(b)(h) there, as vs is a measure; (vi) is proved as §5, Corollary 2. We leave (vii) as an exercise.◻
Theorem 8.10.2 (dominated convergence)
If
f=limi→∞fi (pointwise)
on A−Q(vsQ=0) and if each fi is s-integrable on A, so is f, and
∫Afds=limi→∞∫Afids,
all provided that
(∀i)|fi|≤g
for some map g with ∫Agdvs<∞.
- Proof
-
If s is a measure, this follows by Theorem 5 in §6. Thus as vs is a measure, f is vs-integrable (hence s-integrable) on A, as asserted.
Next, if s=s+−s− is a signed measure, Lemma 2 shows that f and the fi are s+ and s−-integrable as well, with
∫A|fi|ds+≤∫A|fi|dvs≤∫Agdvs<∞;
similarly for
∫A|fi|ds−.
As s+ and s− are measures, Theorem 5 of §6 yields
∫Afds=∫Afds+−∫Afds−=lim(∫Afids+−∫Afids−)=lim∫Afids.
Thus all is proved for signed measures.
In the case s:M→En(Cn), the result now easily follows by Definition 2(ii)(iii) via components.quad◻
Theorem 8.10.3 (uniform convergence)
If fi→f (uniformly) on A−Q (vsA<∞,vsQ=0), and if each fi is s-integrable on A, so is f, and
∫Afds=limi→∞∫Afids.
- Proof
-
Argue as in Theorem 2, replacing §6, Theorem 5, by §7, Lemma 1.◻
Our next theorem shows that integrals behave linearly with respect to measures.
Theorem 8.10.4
Let t,u:M→E∗(En,Cn), with vt<∞ on M, and let
s=pt+qu
for finite constants p and q. Then the following statements are true.
(a) If t and u are generalized measures, so is s.
(b) If, further, f is M-measurable on a set A and is both t-and u-integrable on A, it is also s-integrable on A, and
∫Afds=p∫Afdt+q∫Afdu.
- Proof
-
We consider only assertion (b) for s=t+u; the rest is easy.
First, let f be M-elementary on A. By Corollary 1(ii), we set
∫Afdt=∑iaitAi and ∫Afdu=∑iaiuAi.
Also, by integrability,
∞>∫A|f|dvt=∑|ai|vtAi and ∞>∫A|f|dvu=∑i|ai|vuAi.
Now, by Problem 4 in Chapter 7, §11,
vs=vt+u≤vt+vu;
so
∫A|f|dvs=∑i|ai|vsAi≤∑i|ai|(vtAi+vuAi)=∫A|f|dvt+∫A|f|dvu<∞.
As f is also M-measurable (even elementary), it is s-integrable on A (by Definition 1), and
∫Afds=∑iaisAi=∑iai(tAi+uAi)=∫Afdt+∫Afdu,
as claimed.
Next, suppose f is M-measurable on A and vuA<∞. By assumption, vtA<∞, too; so
vsA≤vtA+vuA<∞.
Now, by Theorem 3 in §1,
f=limi→∞fi (uniformly)
for some M-elementary maps fi on A. By Lemma 2 in §7, for large i, the fi are integrable with respect to both vt and vu on A. By what was shown above, they are also s-integrable, with
∫Afids=∫Afidt+∫Afidu.
With i→∞, Theorem 3 yields the result.
Finally, let vuA=∞. By Corollary 1(v), we may assume (as in Lemma 3 of §7) that Ai↗A, with vuAi<∞, and vtAi<∞ (since vt<∞, by assumption). Set
fi=fCAi→f (pointwise)
on A, with |fi|≤|f|. (Why?)
As fi=f on Ai and fi=0 on A−Ai, all fi are both t- and u-integrable on A (for f is). Since vtAi<∞ and vuAi<∞, the fi are also s-integrable (as shown above), with
∫Afids=∫Aifids=∫Aifidt+∫Aifidu=∫Afidt+∫Afidu.
With i→∞, Theorem 2 now yields the result.
To complete the proof of (b), it suffices to consider, along similar lines, the case s=pt (or s=qu). We leave this to the reader.
For (a), see Chapter 7, §11.◻
Theorem 8.10.5
If f is s-integrable on A, so is |f|, and
|∫Afds|≤∫A|f|dvs.
- Proof
-
By Definition 1, and Theorem 1 of §1, f and |f| are M-measurable on A−Q,vsQ=0, and
∫A|f|dvs<∞;
so |f| is s-integrable on A.
The desired inequality is immediate by Corollary 1(ii) if f is elementary.
Next, exactly as in Theorem 4, one obtains it for the case vsA<∞, and then for vsA=∞. We omit the details.◻
Definition
We write
"ds=gdt in A"
or
"s=∫gdt in A"
iff g is t-integrable on A, and
sX=∫Xgdt
for A⊇X,X∈M.
We then call s the indefinite integral of g in A. (∫Xgdt may be interpreted as in Problems 2-4 below.)
Lemma 8.10.3
If A∈M and
ds=gdt in A,
then
dvs=|g|dvt in A.
- Proof
-
By assumption, g and |g| are vt-integrable on X, and
sX=∫Xgdt
for A⊇X,X∈M. We must show that
vsX=∫X|g|dvt
for such X.
This is easy if g=c (constant) on X. For by definition,
vsX=supP∑i|sXi|,
over all M-partitions P={Xi} of X. As
sXi=∫Xigdt=c⋅tXi,
we have
vsX=supP∑i|c||tXi|=|c|supP∑i|tXi|=|c|vtX;
so
vsX=∫X|g|dvt.
Thus all is proved for constant g.
Hence by σ-additivity, the lemma holds for M-elementary maps g. (Why?)
In the general case, g is t-integrable on X, hence M-measurable and finite on X−Q,vtQ=0. By Corollary 1(iii), we may assume g finite and measurable on X; so
g=limk→∞gk (uniformly)
on X for some M-elementary maps gk, all integrable on X, with respect to vt (and t).
Let
sk=∫gkdt
in X. By what we just proved for elementary and integrable maps,
vskX=∫X|gk|dvt,k=1,2,….
Now, if vtX<∞, Theorem 3 yields
∫X|g|dvt=limk→∞∫X|gk|dvt=limk→∞vskX=vsX
(see Problem 6). Thus all is proved if vtX<∞.
If, however, vtX=∞, argue as in Theorem 4 (the last step), using the left continuity of vs and of
∫|g|dvt.
Verify!◻
Theorem 8.10.6 (change of measure)
If f is s-integrable on A∈M, with
ds=gdt in A,
then (subject to Note 1) fg is t-integrable on A and
∫Afds=∫Afgdt.
(Note the formal substitution of "gdt" for "ds.")
- Proof
-
The proof is easy if f is constant or elementary on A (use Corollary 1(ii)). We leave this case to the reader, and next we assume g is bounded and vtA<∞.
By s-integrability, f is M-measurable and finite on A−Q, with
0=vsQ=∫Q|g|dvt
by Lemma 3. Hence 0=g=fg on Q−Z,vtZ=0. Therefore,
∫Qfgdt=0=∫Qfds
for vsQ=0. Thus we may neglect Q and assume that f is finite and M-measurable on A.
As ds=gdt, Definition 3 and Lemma 3 yield
vsA=∫A|g|dvt<∞.
Also (Theorem 3 in Chapter 8, §1),
f=limk→∞fk(uniformly)
for elementary maps fk, all vs-integrable on A (Lemma 2 in §7). As g is bounded, we get on A
fg=limk→∞fkg(uniformly).
Moreover, as the theorem holds for elementary and integrable maps, fkg is t-integrable on A, and
∫Afkds=∫Afkgdt,k=1,2,….
Since vsA<∞ and vtA<∞, Theorem 3 shows that fg is t-integrable on A, and
∫Afds=limk→∞∫Afkds=limk→∞∫Afkgdt=∫Afgdt.
Thus all is proved if vtA<∞ and g is bounded on A.
In the general case, we again drop a null set to make f and g finite and M-measurable on A. By Corollary 1(v), we may again assume Ai↗A, with vtAi<∞ (∀i).
Now for i=1,2,… set
gi={g on Ai(|g|≤i),0 elsewhere.
Then each gi is bounded,
gi→g (pointwise),
and
|gi|≤|g|
on A. We also set fi=fCAi; so fi→f (pointwise) and |fi|≤|f| on A. Then
∫Afids=∫Aifids=∫Aifigidt=∫Afigidt.
(Why?) Since |figi|≤|fg| and figi→fg, the result follows by Theorem 2.◻