TheoremBase

Proof of Inner and Outer Regularity of a Finite Borel Measure on a Metric Space, and Lipschitz Approximation of Indicators

lemmalem:regularity-finite-borel-measure-metric-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 13,472 chars · 28 deps · depth 13 Reason: Phase B2b: proof by the good-sets argument, with continuity from above derived from continuity from below on complements, and the Lipschitz function built by truncating a multiple of the distance to the inner closed set.

The sets approximable from inside by closed sets and from outside by open sets form a sigma-algebra containing the closed sets, whose neighbourhoods shrink to them by continuity from above; the regularity statements follow, and the Lipschitz function is built by truncating a multiple of the distance to the inner closed set.

Proof

Throughout, each result cited is universally quantified over the data appearing in its own statement and is applied to the data named here. Write C\mathcal{C} for the family of subsets of XX closed in (X,Td)(X,\mathcal{T}_{d}) and U\mathcal{U} for the family of subsets open in (X,d)(X,d); by claim 1 of Borel Measurability and Bounded Integration on a Metric Space both are contained in B(X)\mathcal{B}(X), and B(X)\mathcal{B}(X) is the σ\sigma-algebra generated by C\mathcal{C}. Since μ(X)<\mu(X)<\infty, claim 2 of Basic Properties of a Measure makes μ(A)\mu(A) a real number with 0μ(A)μ(X)0\le\mu(A)\le\mu(X) for every AB(X)A\in\mathcal{B}(X), and for AAA\subseteq A' in B(X)\mathcal{B}(X) claim 3 of that lemma gives μ(AA)=μ(A)μ(A)\mu(A'\setminus A)=\mu(A')-\mu(A). Let ι\iota be the canonical map from N\mathbb{N} to R\mathbb{R}; by claim 3 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field each ι(m)\iota(m) is positive, as is its inverse ι(m)1\iota(m)^{-1}. Open balls Bd(x,r)B_{d}(x,r) are those of Open Ball in a Metric Space.

Step 1 (Continuity from above). Let (Am)mN(A_{m})_{m\in\mathbb{N}} be a sequence in B(X)\mathcal{B}(X) with Am+1AmA_{m+1}\subseteq A_{m} for every mm, and let A=mNAmA=\bigcap_{m\in\mathbb{N}}A_{m}. Writing Bm=XAmB_{m}=X\setminus A_{m}, one has A=XmNBmA=X\setminus\bigcup_{m\in\mathbb{N}}B_{m}, so AB(X)A\in\mathcal{B}(X) because a σ\sigma-algebra is closed under complements and countable unions. The sets BmB_{m} increase, mBm=XA\bigcup_{m}B_{m}=X\setminus A, and every μ(Bm)\mu(B_{m}) is real and at most μ(X)\mu(X), so claim 5 of Basic Properties of a Measure applies and (μ(Bm))mN(\mu(B_{m}))_{m\in\mathbb{N}} converges to μ(XA)\mu(X\setminus A). By claim 3 of Basic Properties of a Measure, μ(Am)=μ(X)μ(Bm)\mu(A_{m})=\mu(X)-\mu(B_{m}) for every mm and μ(A)=μ(X)μ(XA)\mu(A)=\mu(X)-\mu(X\setminus A); hence, by claim 3 of Arithmetic of Limits of Real Sequences applied to the constant sequence with value μ(X)\mu(X) and to (μ(Bm))m(\mu(B_{m}))_{m}, the sequence (μ(Am))mN(\mu(A_{m}))_{m\in\mathbb{N}} converges to μ(A)\mu(A).

Step 2 (A closed set contains its closure points). Let FCF\in\mathcal{C} and let xx belong to the closure clX(F)\operatorname{cl}_{X}(F). If xFx\notin F then xXFx\in X\setminus F, which is open because FF is closed, so there is a positive real ε\varepsilon with Bd(x,ε)XFB_{d}(x,\varepsilon)\subseteq X\setminus F, that is, Bd(x,ε)F=B_{d}(x,\varepsilon)\cap F=\varnothing; this contradicts claim 2 of Characterization of the Closure in a Metric Space by Open Balls. Hence xFx\in F.

Step 3 (Shrinking neighbourhoods of a nonempty closed set). Let FCF\in\mathcal{C} be nonempty, let distd(,F)\operatorname{dist}_{d}(\cdot,F) be the distance to FF, and for mNm\in\mathbb{N} put

Vm={xX:distd(x,F)<ι(m)1}.V_{m}=\{x\in X:\operatorname{dist}_{d}(x,F)<\iota(m)^{-1}\}.

Each VmV_{m} is open: for xVmx\in V_{m} the real number r=ι(m)1distd(x,F)r=\iota(m)^{-1}-\operatorname{dist}_{d}(x,F) is positive, and every yBd(x,r)y\in B_{d}(x,r) satisfies distd(y,F)distd(x,F)+d(x,y)<ι(m)1\operatorname{dist}_{d}(y,F)\le\operatorname{dist}_{d}(x,F)+d(x,y)<\iota(m)^{-1} by claim 4 of The Distance to a Set is Nonexpansive together with claim 3 of Properties of the Absolute Value in an Ordered Field, so Bd(x,r)VmB_{d}(x,r)\subseteq V_{m}. Each VmV_{m} is Borel, being open. Moreover FVmF\subseteq V_{m}, since distd(x,F)=0\operatorname{dist}_{d}(x,F)=0 for xFx\in F by Vanishing of the Distance to a Set Characterizes the Closure; the sets decrease, since ι(m)ι(m+1)\iota(m)\le\iota(m+1) by claim 6 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field, and multiplying that inequality by ι(m)1ι(m+1)1\iota(m)^{-1}\iota(m+1)^{-1}, which is nonnegative by claims 7 and 5 of Elementary Order Arithmetic in an Ordered Field, gives ι(m+1)1ι(m)1\iota(m+1)^{-1}\le\iota(m)^{-1} by claim 5 of Elementary Arithmetic in an Ordered Field; and mNVm=F\bigcap_{m\in\mathbb{N}}V_{m}=F, because a point xx of the intersection has distd(x,F)<ι(m)1\operatorname{dist}_{d}(x,F)<\iota(m)^{-1} for every mm, hence distd(x,F)=0\operatorname{dist}_{d}(x,F)=0 by claim 3 of The Archimedean Property of the Real Numbers and the nonnegativity of the distance, hence xclX(F)x\in\operatorname{cl}_{X}(F) by Vanishing of the Distance to a Set Characterizes the Closure and so xFx\in F by Step 2. By Step 1, (μ(Vm))mN(\mu(V_{m}))_{m\in\mathbb{N}} converges to μ(F)\mu(F).

Step 4 (Closed sets are approximable). Let A\mathcal{A} be the family of those BB(X)B\in\mathcal{B}(X) such that for every positive real ε\varepsilon there are FCF\in\mathcal{C} and UUU\in\mathcal{U} with FBUF\subseteq B\subseteq U and μ(UF)ε\mu(U\setminus F)\le\varepsilon. Then CA\mathcal{C}\subseteq\mathcal{A}. Indeed, let FCF\in\mathcal{C} and let ε\varepsilon be a positive real number. If F=F=\varnothing, take FF itself and U=U=\varnothing, which is both closed and open by claim 1 of Complements, Unions and Intersections of Closed Sets in a Topological Space and Metric Open Sets Form a Topology, so that μ(UF)=μ()=0ε\mu(U\setminus F)=\mu(\varnothing)=0\le\varepsilon. If FF\ne\varnothing, let VmV_{m} be as in Step 3; since (μ(Vm))m(\mu(V_{m}))_{m} converges to μ(F)\mu(F) there is mm with μ(Vm)μ(F)ε\mu(V_{m})-\mu(F)\le\varepsilon, and FFVmF\subseteq F\subseteq V_{m} with μ(VmF)=μ(Vm)μ(F)ε\mu(V_{m}\setminus F)=\mu(V_{m})-\mu(F)\le\varepsilon.

Step 5 (The approximable sets form a σ\sigma-algebra); claim 1. We show A\mathcal{A} is a σ\sigma-algebra on XX. It contains XX, which is closed by claim 1 of Complements, Unions and Intersections of Closed Sets in a Topological Space, by Step 4.

Complements. Let BAB\in\mathcal{A} and let ε\varepsilon be a positive real number; choose FBUF\subseteq B\subseteq U as in the definition of A\mathcal{A}. Then XUXBXFX\setminus U\subseteq X\setminus B\subseteq X\setminus F, the set XUX\setminus U is closed and XFX\setminus F is open by Closed Subset of a Topological Space, and (XF)(XU)=UF(X\setminus F)\setminus(X\setminus U)=U\setminus F, so μ((XF)(XU))ε\mu\bigl((X\setminus F)\setminus(X\setminus U)\bigr)\le\varepsilon. Hence XBAX\setminus B\in\mathcal{A}.

Countable unions. Let (Bm)mN(B_{m})_{m\in\mathbb{N}} be a sequence in A\mathcal{A} and B=mBmB=\bigcup_{m}B_{m}. Fix a positive real ε\varepsilon first; then, for each mm, choose FmCF_{m}\in\mathcal{C} and UmUU_{m}\in\mathcal{U} with FmBmUmF_{m}\subseteq B_{m}\subseteq U_{m} and μ(UmFm)ε(12)m+1\mu(U_{m}\setminus F_{m})\le\varepsilon\,(\tfrac{1}{2})^{m+1}. Put U=mUmU=\bigcup_{m}U_{m}, which is open by Metric Open Sets Form a Topology, and G=mFmB(X)G=\bigcup_{m}F_{m}\in\mathcal{B}(X). Then FmBUF_{m}\subseteq B\subseteq U for every mm, so GBUG\subseteq B\subseteq U. Since UGm(UmFm)U\setminus G\subseteq\bigcup_{m}(U_{m}\setminus F_{m}), claims 2 and 4 of Basic Properties of a Measure and claims 3 and 4 of Series of Nonnegative Real Numbers, Comparison, and the Geometric Series give

μ(UG)m=1ε(12)m+1=ε2m=1(12)m=ε2.\mu(U\setminus G)\le\sum_{m=1}^{\infty}\varepsilon\,(\tfrac{1}{2})^{m+1}=\tfrac{\varepsilon}{2}\sum_{m=1}^{\infty}(\tfrac{1}{2})^{m}=\tfrac{\varepsilon}{2}.

For NNN\in\mathbb{N} let HN=m=1NFmH_{N}=\bigcup_{m=1}^{N}F_{m}, closed by claim 2 of Complements, Unions and Intersections of Closed Sets in a Topological Space. The HNH_{N} increase with union GG, so (μ(HN))N(\mu(H_{N}))_{N} converges to μ(G)\mu(G) by claim 5 of Basic Properties of a Measure, and there is NN with μ(G)μ(HN)ε2\mu(G)-\mu(H_{N})\le\tfrac{\varepsilon}{2}, that is μ(GHN)ε2\mu(G\setminus H_{N})\le\tfrac{\varepsilon}{2}. Now HNBUH_{N}\subseteq B\subseteq U and UHN=(UG)(GHN)U\setminus H_{N}=(U\setminus G)\cup(G\setminus H_{N}), so μ(UHN)ε\mu(U\setminus H_{N})\le\varepsilon by claim 4 of Basic Properties of a Measure. Hence BAB\in\mathcal{A}.

Thus A\mathcal{A} is a σ\sigma-algebra on XX containing C\mathcal{C}, so B(X)A\mathcal{B}(X)\subseteq\mathcal{A} by claim 2 of Intersections of Sigma-Algebras and Minimality of the Generated Sigma-Algebra; as AB(X)\mathcal{A}\subseteq\mathcal{B}(X) by definition, A=B(X)\mathcal{A}=\mathcal{B}(X). This is claim 1.

Step 6 (Claims 2 and 3). Let BB(X)B\in\mathcal{B}(X) and let S={μ(F):FB, FC}S=\{\mu(F):F\subseteq B,\ F\in\mathcal{C}\}. It is nonempty, since C\varnothing\in\mathcal{C} and B\varnothing\subseteq B, and μ(B)\mu(B) is an upper bound for it by claim 2 of Basic Properties of a Measure; so SS is bounded above. Let bb be an upper bound for SS and suppose, for a contradiction, that b<μ(B)b<\mu(B). Then ε=μ(B)b\varepsilon=\mu(B)-b is positive, and claim 1, applied with the positive real number ε2\tfrac{\varepsilon}{2}, gives FCF\in\mathcal{C} and UUU\in\mathcal{U} with FBUF\subseteq B\subseteq U and μ(UF)ε2\mu(U\setminus F)\le\tfrac{\varepsilon}{2}. Since BFUFB\setminus F\subseteq U\setminus F, claim 2 of Basic Properties of a Measure gives μ(B)μ(F)=μ(BF)ε2\mu(B)-\mu(F)=\mu(B\setminus F)\le\tfrac{\varepsilon}{2}, so μ(F)μ(B)ε2=b+ε2\mu(F)\ge\mu(B)-\tfrac{\varepsilon}{2}=b+\tfrac{\varepsilon}{2}; since 0<ε20<\tfrac{\varepsilon}{2} by claim 8 of Elementary Order Arithmetic in an Ordered Field, claim 1 of that lemma gives b<b+ε2b<b+\tfrac{\varepsilon}{2} and claim 2 then gives b<μ(F)b<\mu(F). As μ(F)S\mu(F)\in S, this contradicts the assumption that bb is an upper bound for SS. Hence μ(B)b\mu(B)\le b for every upper bound bb of SS, and μ(B)\mu(B) is the least upper bound of SS. This is claim 2.

For claim 3 let S={μ(U):BU, UU}S'=\{\mu(U):B\subseteq U,\ U\in\mathcal{U}\}. It is nonempty, since XX is open and BXB\subseteq X, is bounded below by 00, and μ(B)\mu(B) is a lower bound for it by claim 2 of Basic Properties of a Measure. Let bb' be a lower bound for SS' and suppose μ(B)<b\mu(B)<b'. Then ε=bμ(B)\varepsilon'=b'-\mu(B) is positive, and claim 1, applied with ε2\tfrac{\varepsilon'}{2}, gives FCF\in\mathcal{C} and UUU\in\mathcal{U} with FBUF\subseteq B\subseteq U and μ(UF)ε2\mu(U\setminus F)\le\tfrac{\varepsilon'}{2}. Since UBUFU\setminus B\subseteq U\setminus F, claim 2 of Basic Properties of a Measure gives μ(U)μ(B)=μ(UB)ε2\mu(U)-\mu(B)=\mu(U\setminus B)\le\tfrac{\varepsilon'}{2}, so μ(U)μ(B)+ε2=bε2\mu(U)\le\mu(B)+\tfrac{\varepsilon'}{2}=b'-\tfrac{\varepsilon'}{2}; since 0<ε20<\tfrac{\varepsilon'}{2} by claim 8 of Elementary Order Arithmetic in an Ordered Field, claims 1 and 2 of that lemma give μ(U)<b\mu(U)<b', contradicting that bb' is a lower bound for SS'. Hence bμ(B)b'\le\mu(B) for every lower bound bb' of SS', and μ(B)\mu(B) is the greatest lower bound of SS'.

Step 7 (Claim 4). Let BB(X)B\in\mathcal{B}(X) and let ε\varepsilon be a positive real number. By claim 1, applied with the positive real number ε2\tfrac{\varepsilon}{2}, there are FCF\in\mathcal{C} and UUU\in\mathcal{U} with FBUF\subseteq B\subseteq U and μ(UF)ε2\mu(U\setminus F)\le\tfrac{\varepsilon}{2}; in particular μ(BF)ε2\mu(B\setminus F)\le\tfrac{\varepsilon}{2}, since BFUFB\setminus F\subseteq U\setminus F.

If F=F=\varnothing, take hh to be the constant function with value 00, which is Lipschitz with constant 00 and takes values in {0}\{0\}, and take N=BN=B; then μ(N)=μ(BF)ε\mu(N)=\mu(B\setminus F)\le\varepsilon and h(x)=0=1B(x)h(x)=0=\mathbf{1}_{B}(x) for every xXNx\in X\setminus N.

Suppose FF\ne\varnothing and let VmV_{m} be as in Step 3. Since (μ(Vm))m(\mu(V_{m}))_{m} converges to μ(F)\mu(F), choose kNk\in\mathbb{N} with μ(Vk)μ(F)ε2\mu(V_{k})-\mu(F)\le\tfrac{\varepsilon}{2}, that is μ(VkF)ε2\mu(V_{k}\setminus F)\le\tfrac{\varepsilon}{2}; the order of choice is ε\varepsilon, then FF and UU, then kk. Define

h(x)=max(0, 1ι(k)distd(x,F))(xX),h(x)=\max\bigl(0,\ 1-\iota(k)\operatorname{dist}_{d}(x,F)\bigr)\qquad(x\in X),

the maximum of two real numbers. Then 0h(x)0\le h(x), and h(x)1h(x)\le1 because 1ι(k)distd(x,F)11-\iota(k)\operatorname{dist}_{d}(x,F)\le1, the distance being nonnegative and ι(k)\iota(k) positive; so hh takes values in [0,1][0,1], the lower bound by claim 1 of Elementary Properties of the Maximum of Two Elements and the upper bound by claim 3 of that lemma, applied with the upper bound 11 for both 00 and 1ι(k)distd(x,F)1-\iota(k)\operatorname{dist}_{d}(x,F).

The map hh is Lipschitz with constant ι(k)\iota(k). To see this, note first that max(0,a)max(0,b)ab|\max(0,a)-\max(0,b)|\le|a-b| for all real a,ba,b: by symmetry we may assume max(0,b)max(0,a)\max(0,b)\le\max(0,a); if max(0,a)=0\max(0,a)=0 then max(0,b)=0\max(0,b)=0 by claim 1 of Elementary Properties of the Maximum of Two Elements and the difference is 00; otherwise max(0,a)=a\max(0,a)=a and bmax(0,b)b\le\max(0,b), so max(0,a)max(0,b)abab\max(0,a)-\max(0,b)\le a-b\le|a-b| by claim 3 of Properties of the Absolute Value in an Ordered Field. Applying this with a=1ι(k)distd(x,F)a=1-\iota(k)\operatorname{dist}_{d}(x,F) and b=1ι(k)distd(y,F)b=1-\iota(k)\operatorname{dist}_{d}(y,F) and using claim 4 of The Distance to a Set is Nonexpansive together with claim 4 of Properties of the Absolute Value in an Ordered Field,

h(x)h(y)ι(k)distd(x,F)distd(y,F)ι(k)d(x,y).|h(x)-h(y)|\le\iota(k)\,\bigl|\operatorname{dist}_{d}(x,F)-\operatorname{dist}_{d}(y,F)\bigr|\le\iota(k)\,d(x,y).

Put N=(BF)(VkF)N=(B\setminus F)\cup(V_{k}\setminus F), a member of B(X)\mathcal{B}(X) with μ(N)ε\mu(N)\le\varepsilon by claim 4 of Basic Properties of a Measure. Let xXNx\in X\setminus N. If xFx\in F then distd(x,F)=0\operatorname{dist}_{d}(x,F)=0 by Vanishing of the Distance to a Set Characterizes the Closure, so h(x)=max(0,1)=1h(x)=\max(0,1)=1, and 1B(x)=1\mathbf{1}_{B}(x)=1 because FBF\subseteq B. If xFx\notin F then xVkx\notin V_{k}, since xVkFx\notin V_{k}\setminus F, so ι(k)1distd(x,F)\iota(k)^{-1}\le\operatorname{dist}_{d}(x,F), and multiplying by the nonnegative number ι(k)\iota(k) gives 1ι(k)distd(x,F)1\le\iota(k)\operatorname{dist}_{d}(x,F) by claim 5 of Elementary Arithmetic in an Ordered Field, giving h(x)=max(0,c)h(x)=\max(0,c) with c0c\le0, that is h(x)=0h(x)=0; and xBx\notin B, since xBFx\notin B\setminus F and xFx\notin F, so 1B(x)=0\mathbf{1}_{B}(x)=0. In both cases h(x)=1B(x)h(x)=\mathbf{1}_{B}(x), which proves claim 4.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…