TheoremBase

Proof of Cyclic Symmetry of Nested Conditional Expectations over a Common Marginal

lemmalem:nc-nested-expectation-cyclic-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 4,256 chars · 4 deps · depth 22 Reason: G4: proof of cyclic symmetry of nested expectations.

Induction on the depth: one layer is moved from one side of the trace to the other by the bimodule property of the conditional expectation and traciality.

Proof

Each result cited below is used for the data in its own statement. For ε∈{1,2}\varepsilon\in\{1,2\}, the maps πε\pi_{\varepsilon} and EεE_{\varepsilon} are those of Two Noncommutative Laws with a Common Marginal: Standing Notation for Their Amalgamated Free Product §embeddings. By Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §expectation, for S,T∈NS,T\in N and c∈Aεc\in A_{\varepsilon},

Eε(πε(S) c πε(T))=S Eε(c) Tandτμ(Eε(c))=τγε(c);E_{\varepsilon}\bigl(\pi_{\varepsilon}(S)\,c\,\pi_{\varepsilon}(T)\bigr)=S\,E_{\varepsilon}(c)\,T\qquad\text{and}\qquad\tau_{\mu}(E_{\varepsilon}(c))=\tau_{\gamma_{\varepsilon}}(c);

by Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §homomorphism, πε(I)=I\pi_{\varepsilon}(I)=I; by The Tracial Algebra of a Noncommutative Law: a Norm-Closed Unital *-Algebra with a Faithful Positive Trace, Determined by Vacuum Vectors, Closed under Square Roots §trace (for the law γε\gamma_{\varepsilon}), τγε(cc′)=τγε(c′c)\tau_{\gamma_{\varepsilon}}(cc')=\tau_{\gamma_{\varepsilon}}(c'c) for c,c′∈Aεc,c'\in A_{\varepsilon}; and all products below lie in the algebra indicated by The Tracial Algebra of a Noncommutative Law: a Norm-Closed Unital *-Algebra with a Faithful Positive Trace, Determined by Vacuum Vectors, Closed under Square Roots §star-algebra. Taking S=IS=I or T=IT=I in the first identity gives x Eε(c)=Eε(πε(x) c)x\,E_{\varepsilon}(c)=E_{\varepsilon}(\pi_{\varepsilon}(x)\,c) and Eε(c) z=Eε(c πε(z))E_{\varepsilon}(c)\,z=E_{\varepsilon}(c\,\pi_{\varepsilon}(z)) for x,z∈Nx,z\in N.

Step 1 (one layer). Let ε∈{1,2}\varepsilon\in\{1,2\}, u,v∈Aεu,v\in A_{\varepsilon} and x,z∈Nx,z\in N. Then

τμ(x Eε(u πε(z) v))=τμ(Eε(v πε(x) u) z).(1)\tau_{\mu}\bigl(x\,E_{\varepsilon}(u\,\pi_{\varepsilon}(z)\,v)\bigr)=\tau_{\mu}\bigl(E_{\varepsilon}(v\,\pi_{\varepsilon}(x)\,u)\,z\bigr).\tag{1}

Indeed, by the identities above the left side equals τμ(Eε(πε(x)uπε(z)v))=τγε(πε(x)uπε(z) v)\tau_{\mu}(E_{\varepsilon}(\pi_{\varepsilon}(x)u\pi_{\varepsilon}(z)v))=\tau_{\gamma_{\varepsilon}}(\pi_{\varepsilon}(x)u\pi_{\varepsilon}(z)\,v), which by the trace property (with c=πε(x)uπε(z)c=\pi_{\varepsilon}(x)u\pi_{\varepsilon}(z) and c′=vc'=v) equals τγε(vπε(x)u πε(z))=τμ(Eε(vπε(x)u πε(z)))=τμ(Eε(vπε(x)u) z)\tau_{\gamma_{\varepsilon}}(v\pi_{\varepsilon}(x)u\,\pi_{\varepsilon}(z))=\tau_{\mu}(E_{\varepsilon}(v\pi_{\varepsilon}(x)u\,\pi_{\varepsilon}(z)))=\tau_{\mu}(E_{\varepsilon}(v\pi_{\varepsilon}(x)u)\,z).

Step 2 (induction on kk). For k∈Nk\in\mathbb{N} let P(k)P(k) be the assertion: for all f1,…,fk∈{1,2}f_{1},\dots,f_{k}\in\{1,2\}, all uj,vj∈Afju_{j},v_{j}\in A_{f_{j}} (j∈[k]j\in[k]) and all x,y∈Nx,y\in N, the elements GkG_{k} and Gk′G_{k}' defined in the statement satisfy τμ(xGk)=τμ(Gk′y)\tau_{\mu}(xG_{k})=\tau_{\mu}(G_{k}'y). Let K={k∈N:P(k)}\mathcal{K}=\{k\in\mathbb{N}:P(k)\}; we show that 1∈K1\in\mathcal{K} and that k+1∈Kk+1\in\mathcal{K} whenever k∈Kk\in\mathcal{K}, so that K=N\mathcal{K}=\mathbb{N} by Principle of Induction for the Natural Numbers.

For k=1k=1: G1=Ef1(u1πf1(y)v1)G_{1}=E_{f_{1}}(u_{1}\pi_{f_{1}}(y)v_{1}) and G1′=Ef1(v1πf1(x)u1)G_{1}'=E_{f_{1}}(v_{1}\pi_{f_{1}}(x)u_{1}), and P(1)P(1) is (1) with ε=f1\varepsilon=f_{1}, u=u1u=u_{1}, v=v1v=v_{1}, z=yz=y.

Let k∈Kk\in\mathcal{K} and let data f1,…,fk+1f_{1},\dots,f_{k+1}, uj,vju_{j},v_{j}, x,yx,y be given, with G1,…,Gk+1G_{1},\dots,G_{k+1} and G1′,…,Gk+1′G_{1}',\dots,G_{k+1}' as in the statement. The first kk elements G1,…,GkG_{1},\dots,G_{k} are also the elements G1,…,GkG_{1},\dots,G_{k} of the truncated data (fj,uj,vj)j∈[k](f_{j},u_{j},v_{j})_{j\in[k]} with the same yy, since both are given by the same first term and the same recursion for the indices j≤kj\le k (induction on jj). Put

x′=Efk+1(vk+1 πfk+1(x) uk+1)∈N.x'=E_{f_{k+1}}\bigl(v_{k+1}\,\pi_{f_{k+1}}(x)\,u_{k+1}\bigr)\in N .

Since Gk+1=Efk+1(uk+1πfk+1(Gk)vk+1)G_{k+1}=E_{f_{k+1}}(u_{k+1}\pi_{f_{k+1}}(G_{k})v_{k+1}), (1) with ε=fk+1\varepsilon=f_{k+1} and z=Gkz=G_{k} gives τμ(xGk+1)=τμ(x′Gk)\tau_{\mu}(xG_{k+1})=\tau_{\mu}(x'G_{k}). By P(k)P(k) for the truncated data, with x′x' in place of xx and the same yy,

τμ(x′Gk)=τμ(Hk y),H1=Efk(vkπfk(x′)uk),Hj=Efk+1−j(vk+1−j πfk+1−j(Hj−1) uk+1−j) (2≤j≤k).\tau_{\mu}(x'G_{k})=\tau_{\mu}(H_{k}\,y),\qquad H_{1}=E_{f_{k}}\bigl(v_{k}\pi_{f_{k}}(x')u_{k}\bigr),\quad H_{j}=E_{f_{k+1-j}}\bigl(v_{k+1-j}\,\pi_{f_{k+1-j}}(H_{j-1})\,u_{k+1-j}\bigr)\ (2\le j\le k).

For the data of length k+1k+1 the statement defines G1′=x′G_{1}'=x' and Gj′=Efk+2−j(vk+2−jπfk+2−j(Gj−1′)uk+2−j)G_{j}'=E_{f_{k+2-j}}(v_{k+2-j}\pi_{f_{k+2-j}}(G_{j-1}')u_{k+2-j}) for 2≤j≤k+12\le j\le k+1. We claim Gj+1′=HjG_{j+1}'=H_{j} for every j∈[k]j\in[k], by induction on jj: for j=1j=1, G2′=Efk(vkπfk(G1′)uk)=Efk(vkπfk(x′)uk)=H1G_{2}'=E_{f_{k}}(v_{k}\pi_{f_{k}}(G_{1}')u_{k})=E_{f_{k}}(v_{k}\pi_{f_{k}}(x')u_{k})=H_{1}; and if j<kj<k and Gj+1′=HjG_{j+1}'=H_{j}, then, since k+2−(j+2)=k+1−(j+1)=k−jk+2-(j+2)=k+1-(j+1)=k-j,

Gj+2′=Efk−j(vk−j πfk−j(Gj+1′) uk−j)=Efk−j(vk−j πfk−j(Hj) uk−j)=Hj+1.G_{j+2}'=E_{f_{k-j}}\bigl(v_{k-j}\,\pi_{f_{k-j}}(G_{j+1}')\,u_{k-j}\bigr)=E_{f_{k-j}}\bigl(v_{k-j}\,\pi_{f_{k-j}}(H_{j})\,u_{k-j}\bigr)=H_{j+1}.

Hence Gk+1′=HkG_{k+1}'=H_{k}, and τμ(xGk+1)=τμ(x′Gk)=τμ(Gk+1′y)\tau_{\mu}(xG_{k+1})=\tau_{\mu}(x'G_{k})=\tau_{\mu}(G_{k+1}'y). This is P(k+1)P(k+1), so k+1∈Kk+1\in\mathcal{K}.

Therefore P(k)P(k) holds for every k∈Nk\in\mathbb{N}, which is the claim.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…