TheoremBase

Proof of Entropy on the Configuration Space: Tensorization, Subadditivity over the Particles, and the One-Particle Marginal

lemmalem:entropy-tensor-marginal-euclidean-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 7,762 chars · 15 deps · depth 35 Reason: Phase N1a proof.

Tensor powers satisfy rho(x)(n+1)rho^{(x)(n+1)} = rho(x)nrho^{(x)n} boxtimes rho by uniqueness, and the block maps of Rq(n+1)R^{q(n+1)} factor through the two projections, so tensorization and subadditivity follow by induction on the number of particles from the product and two-marginal subadditivity results, and the one-particle marginal bound follows from convexity applied to the average of the particle laws.

Proof

Each result cited is universally quantified over the data in its own statement. For n∈Nn\in\mathbb{N} write pk(n)\mathfrak{p}^{(n)}_{k}, k∈[n]k\in[n], for the block maps of Rqn\mathbb{R}^{qn} (Particle Blocks of the Configuration Space: Block Maps, Configurations, Product Maps and Diagonal Points §blocks with N=nN=n), and ρ⊗n\rho^{\otimes n} for the tensor power in Rqn\mathbb{R}^{qn} (The Tensor Power of a Probability Measure on Euclidean Space §tensor with N=nN=n). A natural number used as a real number is its image under the canonical map, as in Properties of the Canonical Map from the Natural Numbers to an Ordered Field.

Step 1 (One particle). For n=1n=1 we have q⋅1=qq\cdot1=q and b(1,i)=(1−1)q+i=ib(1,i)=(1-1)q+i=i for i∈[q]i\in[q], so p1(1)(x)=(x1,…,xq)=x\mathfrak{p}^{(1)}_{1}(x)=(x_{1},\dots,x_{q})=x: the map p1(1)\mathfrak{p}^{(1)}_{1} is the identity of Rq\mathbb{R}^{q}. Hence (p1(1))#P=P(\mathfrak{p}^{(1)}_{1})_{\#}P=P for every P∈P(Rq)P\in\mathcal{P}(\mathbb{R}^{q}), and for ρ∈P(Rq)\rho\in\mathcal{P}(\mathbb{R}^{q}) and B∈B(Rq)B\in\mathcal{B}(\mathbb{R}^{q}), by The Tensor Power of a Probability Measure on Euclidean Space §tensor and Finite Product Notation, ρ⊗1(B)=ρ⊗1((p1(1))−1(B))=∏k=11ρ(B)=ρ(B)\rho^{\otimes1}(B)=\rho^{\otimes1}\bigl((\mathfrak{p}^{(1)}_{1})^{-1}(B)\bigr)=\prod_{k=1}^{1}\rho(B)=\rho(B); so ρ⊗1=ρ\rho^{\otimes1}=\rho.

Step 2 (Adding a particle). Let n∈Nn\in\mathbb{N}, and write ι=ιqn,q\iota=\iota^{qn,q}, pr1=pr1qn,q\mathrm{pr}_{1}=\mathrm{pr}^{qn,q}_{1} and pr2=pr2qn,q\mathrm{pr}_{2}=\mathrm{pr}^{qn,q}_{2}, on Rq(n+1)=Rqn+q\mathbb{R}^{q(n+1)}=\mathbb{R}^{qn+q}. Every z∈Rq(n+1)z\in\mathbb{R}^{q(n+1)} equals ι(pr1(z),pr2(z))\iota(\mathrm{pr}_{1}(z),\mathrm{pr}_{2}(z)) by Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets §projections, so Particle Blocks: Linearity, Splitting of Inner Products, Product Maps and Diagonal Shifts §concatenation gives

pk(n+1)=pk(n)∘pr1(k∈[n]),pn+1(n+1)=pr2.(1)\mathfrak{p}^{(n+1)}_{k}=\mathfrak{p}^{(n)}_{k}\circ\mathrm{pr}_{1}\quad(k\in[n]),\qquad\mathfrak{p}^{(n+1)}_{n+1}=\mathrm{pr}_{2}.\qquad(1)

Consequently, for P∈P(Rq(n+1))P\in\mathcal{P}(\mathbb{R}^{q(n+1)}) and Q1=(pr1)#PQ_{1}=(\mathrm{pr}_{1})_{\#}P, the definition of the push-forward in Probability Measures on Euclidean Space and Random Vectors: Standing Notation §pushforward gives, for B∈B(Rq)B\in\mathcal{B}(\mathbb{R}^{q}) and k∈[n]k\in[n], P((pk(n+1))−1(B))=P(pr1−1((pk(n))−1(B)))=Q1((pk(n))−1(B))P\bigl((\mathfrak{p}^{(n+1)}_{k})^{-1}(B)\bigr)=P\bigl(\mathrm{pr}_{1}^{-1}((\mathfrak{p}^{(n)}_{k})^{-1}(B))\bigr)=Q_{1}\bigl((\mathfrak{p}^{(n)}_{k})^{-1}(B)\bigr); that is,

(pk(n+1))#P=(pk(n))#Q1(k∈[n]),(pn+1(n+1))#P=(pr2)#P.(2)(\mathfrak{p}^{(n+1)}_{k})_{\#}P=(\mathfrak{p}^{(n)}_{k})_{\#}Q_{1}\quad(k\in[n]),\qquad(\mathfrak{p}^{(n+1)}_{n+1})_{\#}P=(\mathrm{pr}_{2})_{\#}P.\qquad(2)

Now let ρ∈P(Rq)\rho\in\mathcal{P}(\mathbb{R}^{q}); we show ρ⊗(n+1)=ρ⊗n⊠ρ\rho^{\otimes(n+1)}=\rho^{\otimes n}\boxtimes\rho. Let B1,…,Bn+1∈B(Rq)B_{1},\dots,B_{n+1}\in\mathcal{B}(\mathbb{R}^{q}) and A=⋂k=1n(pk(n))−1(Bk)A=\bigcap_{k=1}^{n}(\mathfrak{p}^{(n)}_{k})^{-1}(B_{k}), which belongs to B(Rqn)\mathcal{B}(\mathbb{R}^{qn}) since each pk(n)\mathfrak{p}^{(n)}_{k} is Borel (Particle Blocks: Linearity, Splitting of Inner Products, Product Maps and Diagonal Shifts §linear). Every k∈[n+1]k\in[n+1] lies in [n][n] or equals n+1n+1 (claim 5 of Properties of the Order on the Natural Numbers), so by (1) a point zz lies in ⋂k=1n+1(pk(n+1))−1(Bk)\bigcap_{k=1}^{n+1}(\mathfrak{p}^{(n+1)}_{k})^{-1}(B_{k}) exactly when pr1(z)∈A\mathrm{pr}_{1}(z)\in A and pr2(z)∈Bn+1\mathrm{pr}_{2}(z)\in B_{n+1}. As z=ι(pr1(z),pr2(z))z=\iota(\mathrm{pr}_{1}(z),\mathrm{pr}_{2}(z)) and pr1(ι(u,v))=u\mathrm{pr}_{1}(\iota(u,v))=u, pr2(ι(u,v))=v\mathrm{pr}_{2}(\iota(u,v))=v, this holds exactly when z∈ι(A×Bn+1)z\in\iota(A\times B_{n+1}). By Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets §product, The Tensor Power of a Probability Measure on Euclidean Space §tensor for ρ⊗n\rho^{\otimes n}, and the recursion of Finite Product Notation (n+1≥2n+1\ge2),

(ρ⊗n⊠ρ)(⋂k=1n+1(pk(n+1))−1(Bk))=ρ⊗n(A) ρ(Bn+1)=(∏k=1nρ(Bk))ρ(Bn+1)=∏k=1n+1ρ(Bk).(\rho^{\otimes n}\boxtimes\rho)\Bigl(\bigcap_{k=1}^{n+1}(\mathfrak{p}^{(n+1)}_{k})^{-1}(B_{k})\Bigr)=\rho^{\otimes n}(A)\,\rho(B_{n+1})=\Bigl(\prod_{k=1}^{n}\rho(B_{k})\Bigr)\rho(B_{n+1})=\prod_{k=1}^{n+1}\rho(B_{k}).

Since ρ⊗n⊠ρ∈P(Rqn+q)=P(Rq(n+1))\rho^{\otimes n}\boxtimes\rho\in\mathcal{P}(\mathbb{R}^{qn+q})=\mathcal{P}(\mathbb{R}^{q(n+1)}), the uniqueness in Existence and Uniqueness of Tensor Powers, and the Average of the Block Marginals §tensor (with N=n+1N=n+1) gives ρ⊗(n+1)=ρ⊗n⊠ρ\rho^{\otimes(n+1)}=\rho^{\otimes n}\boxtimes\rho.

Step 3 (Claim 1). Fix ρ∈P2Ent(Rq)\rho\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) and let TT be the set of n∈Nn\in\mathbb{N} with ρ⊗n∈P2Ent(Rqn)\rho^{\otimes n}\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{qn}) and Ent(ρ⊗n)=n Ent(ρ)\mathrm{Ent}(\rho^{\otimes n})=n\,\mathrm{Ent}(\rho). By Step 1, 1∈T1\in T. If n∈Tn\in T, then by Step 2 and Entropy of Products, Subadditivity over Two Marginals, and Convexity of the Entropy §product (with dimensions qnqn and qq), ρ⊗(n+1)=ρ⊗n⊠ρ∈P2Ent(Rq(n+1))\rho^{\otimes(n+1)}=\rho^{\otimes n}\boxtimes\rho\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q(n+1)}) and Ent(ρ⊗(n+1))=n Ent(ρ)+Ent(ρ)=(n+1) Ent(ρ)\mathrm{Ent}(\rho^{\otimes(n+1)})=n\,\mathrm{Ent}(\rho)+\mathrm{Ent}(\rho)=(n+1)\,\mathrm{Ent}(\rho), using claim 1 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field; so n+1∈Tn+1\in T. As n+1n+1 is the successor of nn (claim 1 of Arithmetic of Addition on the Natural Numbers), Principle of Induction for the Natural Numbers gives T=NT=\mathbb{N}, and in particular N∈TN\in T.

Step 4 (Claim 2). Let UU be the set of n∈Nn\in\mathbb{N} such that for every P∈P2Ent(Rqn)P\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{qn}) each (pk(n))#P(\mathfrak{p}^{(n)}_{k})_{\#}P, k∈[n]k\in[n], belongs to P2Ent(Rq)\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) and ∑k=1nEnt((pk(n))#P)≤Ent(P)\sum_{k=1}^{n}\mathrm{Ent}\bigl((\mathfrak{p}^{(n)}_{k})_{\#}P\bigr)\le\mathrm{Ent}(P). By Step 1 and ∑k=11ak=a1\sum_{k=1}^{1}a_{k}=a_{1} (claim 1 of Properties of Finite Sums), 1∈U1\in U. Let n∈Un\in U and P∈P2Ent(Rq(n+1))P\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q(n+1)}), and put Q1=(pr1)#PQ_{1}=(\mathrm{pr}_{1})_{\#}P, Q2=(pr2)#PQ_{2}=(\mathrm{pr}_{2})_{\#}P as in Step 2. By Entropy of Products, Subadditivity over Two Marginals, and Convexity of the Entropy §subadditive (with dimensions qnqn and qq), Q1∈P2Ent(Rqn)Q_{1}\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{qn}), Q2∈P2Ent(Rq)Q_{2}\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) and Ent(Q1)+Ent(Q2)≤Ent(P)\mathrm{Ent}(Q_{1})+\mathrm{Ent}(Q_{2})\le\mathrm{Ent}(P). By (2), (pn+1(n+1))#P=Q2(\mathfrak{p}^{(n+1)}_{n+1})_{\#}P=Q_{2} and, for k∈[n]k\in[n], (pk(n+1))#P=(pk(n))#Q1(\mathfrak{p}^{(n+1)}_{k})_{\#}P=(\mathfrak{p}^{(n)}_{k})_{\#}Q_{1}, which belongs to P2Ent(Rq)\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) since n∈Un\in U. Every k∈[n+1]k\in[n+1] lies in [n][n] or equals n+1n+1, so all particle laws of PP belong to P2Ent(Rq)\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}), and by the recursion in claim 1 of Properties of Finite Sums and n∈Un\in U,

∑k=1n+1Ent((pk(n+1))#P)=∑k=1nEnt((pk(n))#Q1)+Ent(Q2)≤Ent(Q1)+Ent(Q2)≤Ent(P).\sum_{k=1}^{n+1}\mathrm{Ent}\bigl((\mathfrak{p}^{(n+1)}_{k})_{\#}P\bigr)=\sum_{k=1}^{n}\mathrm{Ent}\bigl((\mathfrak{p}^{(n)}_{k})_{\#}Q_{1}\bigr)+\mathrm{Ent}(Q_{2})\le\mathrm{Ent}(Q_{1})+\mathrm{Ent}(Q_{2})\le\mathrm{Ent}(P).

Hence n+1∈Un+1\in U, and Principle of Induction for the Natural Numbers gives U=NU=\mathbb{N}; in particular N∈UN\in U.

Step 5 (Claim 3). Let P∈P2Ent(RqN)P\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{qN}) and μk=(pk)#P\mu_{k}=(\mathfrak{p}_{k})_{\#}P for k∈[N]k\in[N]; by Step 4, μk∈P2Ent(Rq)\mu_{k}\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) and ∑k=1NEnt(μk)≤Ent(P)\sum_{k=1}^{N}\mathrm{Ent}(\mu_{k})\le\mathrm{Ent}(P). Put tk=N−1t_{k}=N^{-1}, which is positive by claim 3 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field. By The One-Particle Marginal of a Probability Measure on the Configuration Space §marginal and claim 3 of Properties of Finite Sums, P[1](B)=N−1∑k=1Nμk(B)=∑k=1Ntkμk(B)P^{[1]}(B)=N^{-1}\sum_{k=1}^{N}\mu_{k}(B)=\sum_{k=1}^{N}t_{k}\mu_{k}(B) for B∈B(Rq)B\in\mathcal{B}(\mathbb{R}^{q}); for B=RqB=\mathbb{R}^{q}, where μk(Rq)=P(RqN)=1\mu_{k}(\mathbb{R}^{q})=P(\mathbb{R}^{qN})=1, this reads ∑k=1Ntk=P[1](Rq)=1\sum_{k=1}^{N}t_{k}=P^{[1]}(\mathbb{R}^{q})=1, P[1]P^{[1]} being a probability measure. Hence Entropy of Products, Subadditivity over Two Marginals, and Convexity of the Entropy §convex (with n=Nn=N) applies and gives P[1]∈P2Ent(Rq)P^{[1]}\in\mathcal{P}_{2}^{\mathrm{Ent}}(\mathbb{R}^{q}) and, with claim 3 of Properties of Finite Sums and claim 5 of Elementary Arithmetic in an Ordered Field (0≤N−10\le N^{-1}),

Ent(P[1])≤∑k=1NN−1 Ent(μk)=N−1∑k=1NEnt(μk)≤N−1 Ent(P).\mathrm{Ent}(P^{[1]})\le\sum_{k=1}^{N}N^{-1}\,\mathrm{Ent}(\mu_{k})=N^{-1}\sum_{k=1}^{N}\mathrm{Ent}(\mu_{k})\le N^{-1}\,\mathrm{Ent}(P).

Multiplying by N≥0N\ge0 (claim 5 of Elementary Arithmetic in an Ordered Field) gives N Ent(P[1])≤Ent(P)N\,\mathrm{Ent}(P^{[1]})\le\mathrm{Ent}(P).

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…