Measure Theory (VIII): The Lebesgue Measure on Rn
24 Oct 2018
One of the most important applications of the Carathéodory construction of measures is to extend the notion of “length” or “volume” in Rn to the Lebesgue measure, yielding a more powerful integration theory. In this post, we cover the basic properties of the Lebesgue measure.
The Lebesgue measure
Recall that for a=(a1,…,an), b=(b1,…,bn), with −∞≤ak≤bk≤∞, the half-open interval in Rn is the set
[a,b)=[a1,b1)×⋯×[an,bn)⊆Rn.
We write Hn for the collection of all half-open intervals in Rn. We have shown that Hn generates the Borel σ-algebra, ie. σ(Hn)=Bn.
Define λ0:Hn→[0,∞] by
λ0[a,b)=k=1∏n(bk−ak).
Then the Carathéodory construction gives the Lebesgue outer measure λ∗ on Rn, which is given by
λ∗(A)=inf{k=1∑∞λ0(Ik):Ik∈Hn,A⊆k=1⋃∞Ik}.
The λ∗-measurable sets, called Lebesgue measurable sets, form a σ-algebra Σ, and the corresponding measure λ=λ∗∣Σ is called the Lebesgue measure on Rn. As we have shown in the previous post, (Rn,Σ,λ) is a complete measure space.
We now consider some elementary properties of the Lebesgue measure. For any A⊆Rn, x∈Rn, and c∈R\{0}, define
A+xcA={ω+x:ω∈A},={cω:ω∈A}.
Geometrically, these operations are translation by x and scaling by c, respectively. We first show that the Lebesgue measure is translation-invariant and homogeneous under scaling.
Proposition: Let A⊆Rn, x∈Rn, and c>0. Then
λ∗(A+x)λ∗(cA)=λ∗(A),=cnλ∗(A).
Proof: First note that for all I∈Hn, we have
λ0(I+x)λ0(cI)=λ0(I),=cnλ0(I).
Let (Ik) be a countable covering of A with half-open intervals. Then (Ik+x) and (cIk) are coverings of A+x and cA, respectively. Hence
λ∗(A+x)λ∗(cA)≤k∑λ0(Ik+x)=k∑λ0(Ik),≤k∑λ0(cIk)=cnk∑λ0(Ik).
Taking infimum over all (Ik) gives
λ∗(A+x)λ∗(cA)≤λ∗(A),≤cnλ∗(A).
Now the same argument with −x replacing x and c1 replacing c gives
λ∗(A+x)λ∗(cA)≥λ∗((A+x)−x)=λ∗(A),≥cnλ∗(c1(cA))=λ∗(A),
and we are done. □
Corollary: Let A⊆Rn, x∈Rn, and c>0. Then A is Lebesgue measurable if and only if A+x is Lebesgue measurable, if and only if cA is Lebesgue measurable.
Proof: If A is Lebesgue measurable, then for any S⊆Rn, we have
λ∗(S)=λ∗(S−x)=λ∗((S−x)∩A)+λ∗((S−x)∩Ac)=λ∗(S∩(A+x))+λ∗(S∩(A+x)c),
so A+x is Lebesgue measurable. Conversely, if A+x is Lebesgue measurable, then replacing x by −x gives (A+x)−x=A is measurable.
Similarly,
λ∗(S)=cnλ∗(c1S)=cn(λ∗((c1S)∩A)+λ∗((c1S)∩Ac))=λ∗(S∩(cA))+λ∗(S∩(cA)c),
so cA is Lebesgue measurable. Conversely, if cA is Lebesgue measurable, then replacing c by c1 gives c1(cA)=A is measurable. □
Corollary: Let f:Rn→[−∞,∞], x∈Rn, and c>0. Define g(ω)=f(ω−x) and h(ω)=f(cω). Then:
- f is Lebesgue measurable if and only if g is Lebesgue measurable, if and only if h is Lebesgue measurable;
- f is Lebesgue integrable if and only if g is Lebesgue integrable, if and only if h is Lebesgue integrable, in which case we have
∫gdλ=∫fdλ,∫hdλ=cn∫fdλ.
Proof: (1): This follows from g−1(a,∞]=f−1(a,∞]+x and h−1(a,∞]=cf−1(a,∞].
(2): Note that g±(ω)=f±(ω−x) and h±(ω)=f±(cω). Hence it suffices to prove the identity for f≥0.
For any sequence of step functions 0≤φ1≤φ≤⋯ with (φk)→f, let ψk(ω)=φk(ω−x). Then 0≤ψ1≤ψ2≤⋯ is a sequence of step functions with (ψk)→g. Now
∫ψkdλ=j∑ajλ(ψk−1(aj))=j∑ajλ(φk−1(aj)+x)=j∑ajλ(φk−1(aj))=∫φkdλ.
Taking k→∞ gives ∫gdλ=∫fdλ.
Similarly, let ηk(ω)=φk(cω), so 0≤η1≤η2≤⋯ is a sequence of step functions with (ηk)→h. Now
∫ηkdλ=j∑ajλ(ηk−1(aj))=j∑ajλ(cφk−1(aj))=cnj∑ajλ(φk−1(aj))=cn∫φkdλ.
Taking k→∞ gives ∫hdλ=cn∫fdλ. □
Note that we have omitted the case of c<0: the main problem here is that I∈Hn does not imply that −I=(−1)I∈Hn. To deal with this case, we will need stronger theorems about the structure of Lebesgue measurable sets, which we will cover in the next post.
Half-open intervals
Due to the indirect nature of the Carathéodory construction, it is not obvious that the Lebesgue measure λ on half-open intervals corresponds to their volume λ0; indeed, it is not obvious that half-open intervals are (λ∗-)measurable at all! The proof of these two statements are technical, and will take up the rest of the post.
Lemma: Let I,I1,I2,…,In∈Hn be half-open intervals.
-
If I is the disjoint union of I1,…,Im, then
λ0(I)=k=1∑mλ0(Ik).
-
If I is contained in the (not necessarily disjoint) union of I1,…,Im, then
λ0(I)≤k=1∑mλ0(Ik).
Proof: (1): Consider the special case of a “product partition” of the form
[a,b)=i1=0⋃m1−1⋯in=0⋃mn−1[xi11,xi1+11)×⋯×[xinn,xin+1n),
where ak=x0k<x1k<⋯<xmkk=bk. Writing the half-open intervals in the partition as Si1…in, we have
i1,…,in∑λ0(Si1…in)=i1,…,in∑(xi1+11−xi11)⋯(xin+1n−xin1)=(i1∑(xi1+11−xi11))⋯(in∑(xin+1n−xin1))=(b1−a1)⋯(bn−an)=λ0[a,b),
so the statement holds in this special case.
For a general partition of I=[a,b), we “extend the sides” of each half-open interval to get a sub-partition of I, as follows. For each k, list all lower and upper endpoints of the k-th coordinates of every Ij, and arrange them in increasing order as ak=x0k<x1k<⋯<xmkk=bk.
Now consider the product partition of I associated to these xik; by the special case, we have
i1,…,in∑λ0(Si1…in)=λ0(I).
On the other hand, note that for each Ij, the Si1…in which intersect Ij also form a product partition of Ij, and each Si1…in is contained in exactly one Ij. Hence
λ0(I)=i1,…,in∑λ0(Si1…in)=j∑λ0(Ij),
and we are done.
(2): Without loss of generality, we may replace Ij by I∩Ij. Then the same operation as above gives a product partition of I. However, now each Si1…in is contained in at least one Ij, so
λ0(I)=i1,…,in∑λ0(Si1…in)≤j∑λ0(Ij),
as desired. □
Proposition: For any half-open interval I∈Hn, we have λ∗(I)=λ0(I).
Proof: We first consider the case where I is finite, ie. bounded. Since {I} is a covering of I by half-open intervals, we have λ∗(I)≤λ0(I); in particular, λ∗(I)<∞. Hence, we only need to consider countable coverings (Ik) of I by half-open intervals where ∑kλ0(Ik)<∞; this implies that each Ik is finite.
Fix ε>0. By shrinking I slightly, we can get I′∈Hn such that I⊇I′, the closure of I′, and
λ0(I′)≥λ0(I)−ε.
Similarly, by enlarging Ik slightly, we can get Ik′∈Hn such that Ik⊆Ik′∘, the interior of Ik′, and
λ0(Ik′)≤λ0(Ik)+2kε.
Now (Ik′) is an open covering of the compact set I′, so it has a finite subcovering, say Ik1′∪⋯∪Ikm′. By the previous lemma, we have
λ0(I)≤λ0(I′)+ε≤i=1∑mλ0(Iki′)+ε≤k=1∑∞λ0(Ik)+2ε,
and taking ε→0+ gives ∑kλ0(Ik)≥λ0(I). Now taking infimum over all (Ik) gives λ∗(I)≥λ0(I), as desired.
Now suppose that I=[a,b) is not finite, say bk=∞ (the argument for ak=−∞ is analogous). Then
S=[a1,b1)×⋯×[ak,ak+N)×⋯×[an,bn)
is contained in I, so
λ∗(I)≥λ∗(S)=Ni=k∏(bi−ai).
Taking N→∞ gives λ∗(I)=∞=λ0(I), and we are done. □
Proposition: Every half-open interval I∈Hn is Lebesgue measurable.
Proof: We need to show that
λ∗(S)≥λ∗(S∩I)+λ∗(S∩Ic)
for any S⊆Rn. If λ∗(S)=∞ there is nothing to prove, so assume that λ∗(S)<∞.
Fix ε>0, and take a countable cover (Rk) of S by half-open intervals, such that
k=1∑∞λ0(Rk)≤λ∗(S)+ε.
Note that each Rk can be written as the disjoint union of finitely many half-open intervals,
Rk=R~k∪i=1⋃NSki,
where R~k=Rk∩I, by “extending the sides” into a product partition. By the lemma, we have
λ0(Rk)=λ0(R~k)+i=1∑Nλ0(Ski),
so
λ∗(S)+ε≥k=1∑∞λ0(Rk)=k=1∑∞λ0(R~k)+k=1∑∞i=1∑Nλ0(Ski)≥λ∗(S∩I)+λ∗(S∩Ic),
since (R~k) cover S∩I and (Ski) cover S∩Ic. Taking ε→0+ gives
λ∗(S)≥λ∗(S∩I)+λ∗(S∩Ic),
as desired. □
Corollary: Every Borel subset of Rn is Lebesgue measurable.
Proof: Since the collection Σ of all Lebesgue measurable sets is a σ-algebra, Hn⊆Σ implies Bn=σ(Hn)⊆Σ. □
Comments (0)
Be the first to comment!