TheoremBase

Proof of The Space of Piecewise Constant Paths in a Finite Set: Measurability of Evaluations, Restrictions, Left Limits, and Occupation Integrals

lemmalem:piecewise-constant-path-space-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: Proof of the measurability properties of the space of piecewise constant paths in a finite set.

Proof

Throughout, Path=Path(E,T)\mathsf{Path}=\mathsf{Path}(E,T), and for pPathp\in\mathsf{Path} we call the intervals [0,t1)[0,t_1), [tj,tj+1)[t_j,t_{j+1}) (1j<k1\le j<k) and [tk,T][t_k,T] of the definition (the single interval [0,T][0,T] when k=0k=0) the constancy intervals of pp; they partition [0,T][0,T], and each of them, except possibly the last, has its right endpoint strictly larger than its left endpoint and excluded, while the last one is [tk,T][t_k,T] or [0,T][0,T]. We use that the family of subsets of a set whose preimage under a given map lies in a given σ\sigma-algebra is itself a σ\sigma-algebra (preimages commute with complements and countable unions), so that a map into a generated σ\sigma-algebra is measurable as soon as the preimages of the generators are measurable. Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions is used for constants, indicators, linear combinations, products and pointwise limits of measurable real-valued maps.

Claim 1. Every subset SS of the finite set EE is the finite union of the singletons {y}\{y\}, ySy\in S, so the preimage of SS under pp(u)p\mapsto p(u) is the finite union of the generators {p:p(u)=y}\{p:p(u)=y\}, ySy\in S, a member of CT\mathcal{C}_T; thus the evaluation map is measurable. Since ϕ(p(u))=yEϕ(y)1{p(u)=y}\phi(p(u))=\sum_{y\in E}\phi(y)\mathbf{1}\{p(u)=y\}, the map pϕ(p(u))p\mapsto\phi(p(u)) is a linear combination of indicators of members of CT\mathcal{C}_T, hence measurable. Now let 0<TT0<T'\le T and pPathp\in\mathsf{Path} with times t1<<tkt_1<\dots<t_k; let kk' be the number of indices jj with tjTt_j\le T'. The restriction p[0,T]p|_{[0,T']} is constant on [0,t1)[0,t_1) if k1k'\ge1, on [tj,tj+1)[t_j,t_{j+1}) for j<kj<k', and on [tk,T][t_{k'},T'], the last interval being contained in [tk,tk+1)[t_{k'},t_{k'+1}) if k<kk'<k and in [tk,T][t_k,T] if k=kk'=k; if k=0k'=0 it is constant on [0,T][0,T'], which is contained in [0,t1)[0,t_1) (or in [0,T][0,T] when k=0k=0). Hence p[0,T]Path(E,T)p|_{[0,T']}\in\mathsf{Path}(E,T') with the times t1,,tkt_1,\dots,t_{k'}. The preimage under the restriction map of a generator {p:p(u)=y}\{p':p'(u)=y\} of CT\mathcal{C}_{T'} (u[0,T]u\in[0,T']) is the generator {p:p(u)=y}\{p:p(u)=y\} of CT\mathcal{C}_T, so the restriction map is measurable.

Claim 2. Let pPathp\in\mathsf{Path} with times t1<<tkt_1<\dots<t_k and let t(0,T]t\in(0,T]. If k=0k=0 or tt1t\le t_1, then pp is constant on [0,t)[0,t), and δ=t\delta=t together with the constant value works. Otherwise let jj be the largest index with tj<tt_j<t; then pp is constant on [tj,tj+1)[tj,t)[t_j,t_{j+1})\supseteq[t_j,t) if j<kj<k (as ttj+1t\le t_{j+1}), and on [tk,T][tk,t)[t_k,T]\supseteq[t_k,t) if j=kj=k; so δ=ttj\delta=t-t_j and the constant value work. If yy and yy' both have the property with δ\delta and δ\delta', then y=p(s)=yy=p(s)=y' for s=tmin(δ,δ)[0,t)s=t-\min(\delta,\delta')\in[0,t); so p(t)p(t-) is unique. For u[0,T)u\in[0,T), the constancy interval containing uu has right endpoint strictly larger than uu (if it is the last one, its right endpoint is T>uT>u and δ=Tu\delta=T-u works; otherwise it is of the form [a,b)[a,b) with u[a,b)u\in[a,b), so b>ub>u and it contains [u,u+δ)[u,u+\delta) for δ=bu\delta=b-u), which gives the right-continuity assertion. For measurability, fix t(0,T]t\in(0,T] and yEy\in E and let n0n_0 be a natural number with 1/n0t1/n_0\le t. We claim

{p:p(t)=y}=nn0 mn{p:p(t1/m)=y}.\{p:p(t-)=y\}=\bigcup_{n\ge n_0}\ \bigcap_{m\ge n}\{p:p(t-1/m)=y\}.

If p(t)=yp(t-)=y with the δ\delta above, then for every mmax(n0,1/δ)m\ge\max(n_0,1/\delta) the point t1/mt-1/m lies in [tδ,t)[t-\delta,t), so p(t1/m)=yp(t-1/m)=y, and pp belongs to the right side. Conversely, if p(t1/m)=yp(t-1/m)=y for all mnm\ge n, then, with δ\delta as in the existence part, p(t1/m)=p(t)p(t-1/m)=p(t-) for all mmax(n,n0,1/δ)m\ge\max(n,n_0,1/\delta), so p(t)=yp(t-)=y. The right side is a countable union of countable intersections of generators, hence a member of CT\mathcal{C}_T; as in claim 1 this gives the measurability of pp(t)p\mapsto p(t-).

Claim 3. Fix yEy\in E. For a natural number nn and i{1,,n}i\in\{1,\dots,n\} put In,i=((i1)T/n,iT/n]I_{n,i}=((i-1)T/n,\,iT/n], a member of B[0,T]\mathcal{B}_{[0,T]}, and define gn:[0,T][0,T]g_n:[0,T]\to[0,T] by gn(0)=0g_n(0)=0 and gn(u)=iT/ng_n(u)=iT/n for uIn,iu\in I_{n,i}; the sets {0}\{0\} and In,1,,In,nI_{n,1},\dots,I_{n,n} partition [0,T][0,T], and ugn(u)<u+T/nu\le g_n(u)<u+T/n for every uu. Then

1{p(gn(u))=y}=1{0}(u)1{p(0)=y}+i=1n1In,i(u)1{p(iT/n)=y}((u,p)[0,T]×Path),\mathbf{1}\{p(g_n(u))=y\}=\mathbf{1}_{\{0\}}(u)\,\mathbf{1}\{p(0)=y\}+\sum_{i=1}^{n}\mathbf{1}_{I_{n,i}}(u)\,\mathbf{1}\{p(iT/n)=y\}\qquad((u,p)\in[0,T]\times\mathsf{Path}),

where 1I\mathbf{1}_I denotes the indicator of a set II; each summand is the indicator of a measurable rectangle of B[0,T]CT\mathcal{B}_{[0,T]}\otimes\mathcal{C}_T, so the left side is B[0,T]CT\mathcal{B}_{[0,T]}\otimes\mathcal{C}_T-measurable as a function of (u,p)(u,p). For every (u,p)(u,p) the sequence 1{p(gn(u))=y}\mathbf{1}\{p(g_n(u))=y\} converges to 1{p(u)=y}\mathbf{1}\{p(u)=y\}: if u=Tu=T then gn(T)=Tg_n(T)=T for every nn; if u<Tu<T, claim 2 gives δ>0\delta>0 with p=p(u)p=p(u) on [u,u+δ)[0,T][u,u+\delta)\cap[0,T], and for n>T/δn>T/\delta we have gn(u)[u,u+δ)g_n(u)\in[u,u+\delta), so p(gn(u))=p(u)p(g_n(u))=p(u). Being a pointwise limit of measurable maps, (u,p)1{p(u)=y}(u,p)\mapsto\mathbf{1}\{p(u)=y\} is B[0,T]CT\mathcal{B}_{[0,T]}\otimes\mathcal{C}_T-measurable.

Now let ϕ:ER\phi:E\to\mathbb{R} and put M=maxyEϕ(y)M=\max_{y\in E}|\phi(y)|. The map (u,p)ϕ(p(u))=yEϕ(y)1{p(u)=y}(u,p)\mapsto\phi(p(u))=\sum_{y\in E}\phi(y)\mathbf{1}\{p(u)=y\} is B[0,T]CT\mathcal{B}_{[0,T]}\otimes\mathcal{C}_T-measurable and bounded by MM in absolute value. For fixed pp its section uϕ(p(u))u\mapsto\phi(p(u)) is B[0,T]\mathcal{B}_{[0,T]}-measurable by claim 3 of Sections of Product-Measurable Sets and Maps Are Measurable, and Insertion Maps into Products Are Measurable, and, being bounded, it is integrable over [0,T][0,T]: by claim 4 of Restricted Lebesgue Measure and Integral Toolkit on a Compact Interval (applied with the second function equal to the constant 11), a B[0,T]\mathcal{B}_{[0,T]}-measurable ff with [0,T]f2du<\int_{[0,T]}f^2\,du<\infty has integrable absolute value, and here f2M2f^2\le M^2 gives [0,T]f2duM2[0,T]1du=M2T\int_{[0,T]}f^2\,du\le M^2\int_{[0,T]}1\,du=M^2T by claim 1 of Linearity and Monotonicity of the Lebesgue Integral, the integral of the constant 11 being λ[0,T]([0,T])=T\lambda_{[0,T]}([0,T])=T by Simple Function and Its Integral and claim 1 of Restricted Lebesgue Measure and Integral Toolkit on a Compact Interval; so ff, and in particular uϕ(p(u))u\mapsto\phi(p(u)), is integrable. By claim 2 of Linearity and Monotonicity of the Lebesgue Integral, [0,T]ϕ(p(u))du[0,T]ϕ(p(u))du[0,T]Mdu=MT|\int_{[0,T]}\phi(p(u))\,du|\le\int_{[0,T]}|\phi(p(u))|\,du\le\int_{[0,T]}M\,du=MT, and, the indicators u1{p(u)=y}u\mapsto\mathbf{1}\{p(u)=y\} being integrable by the same argument,

[0,T]ϕ(p(u))du=yEϕ(y)[0,T]1{p(u)=y}du.\int_{[0,T]}\phi(p(u))\,du=\sum_{y\in E}\phi(y)\int_{[0,T]}\mathbf{1}\{p(u)=y\}\,du .

It remains to see that p[0,T]1{p(u)=y}dup\mapsto\int_{[0,T]}\mathbf{1}\{p(u)=y\}\,du is CT\mathcal{C}_T-measurable for each yy. For each nn and each pp, the map u1{p(gn(u))=y}u\mapsto\mathbf{1}\{p(g_n(u))=y\} is the nonnegative simple function 1{0}(u)1{p(0)=y}+i=1n1In,i(u)1{p(iT/n)=y}\mathbf{1}_{\{0\}}(u)\mathbf{1}\{p(0)=y\}+\sum_{i=1}^{n}\mathbf{1}_{I_{n,i}}(u)\mathbf{1}\{p(iT/n)=y\} on [0,T][0,T], whose integral is i=1n(T/n)1{p(iT/n)=y}\sum_{i=1}^{n}(T/n)\mathbf{1}\{p(iT/n)=y\} by additivity (claim 1 of Linearity and Monotonicity of the Lebesgue Integral) and Simple Function and Its Integral, the integral of c1Ic\,\mathbf{1}_I being cλ[0,T](I)c\,\lambda_{[0,T]}(I), with λ[0,T]({0})=0\lambda_{[0,T]}(\{0\})=0 and λ[0,T](In,i)=T/n\lambda_{[0,T]}(I_{n,i})=T/n by claim 1 of Restricted Lebesgue Measure and Integral Toolkit on a Compact Interval and Existence of Lebesgue Measure on the Real Line; this is a CT\mathcal{C}_T-measurable function of pp, being a linear combination of indicators of generators. For fixed pp, the maps u1{p(gn(u))=y}u\mapsto\mathbf{1}\{p(g_n(u))=y\} are B[0,T]\mathcal{B}_{[0,T]}-measurable, converge pointwise on [0,T][0,T] to u1{p(u)=y}u\mapsto\mathbf{1}\{p(u)=y\} as shown above, and are dominated by the constant 11, which is integrable over [0,T][0,T] since λ[0,T]([0,T])=T<\lambda_{[0,T]}([0,T])=T<\infty; by the dominated convergence theorem their integrals converge to [0,T]1{p(u)=y}du\int_{[0,T]}\mathbf{1}\{p(u)=y\}\,du. Hence p[0,T]1{p(u)=y}dup\mapsto\int_{[0,T]}\mathbf{1}\{p(u)=y\}\,du is a pointwise limit of CT\mathcal{C}_T-measurable maps, so it is CT\mathcal{C}_T-measurable, and the display shows that p[0,T]ϕ(p(u))dup\mapsto\int_{[0,T]}\phi(p(u))\,du is a linear combination of measurable maps. This completes the proof.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…