Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-adaptedPipeline-generatedprecheck passverified 2026-10-22 (gpt-6.1-sol)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Finite relative-entropy chain rule and event grouping

Statement

Let I,J be finite nonempty sets, let P,Q be probability vectors on I×J (a probability vector p on a finite nonempty set H is a function p:H→[0,1] with ∑z∈Hp(z)=1), and assume Q(x,y)>0 for every (x,y). For probability vectors p,q on a finite set with q strictly positive, define D(p∥q)=∑z:p(z)>0p(z)ln⁡p(z)q(z), with zero-mass terms omitted. For C⊆I×J, write P(C)=∑z∈CP(z) and Q(C)=∑z∈CQ(z). Define the first-coordinate marginals by PX(x)=∑y∈JP(x,y) and QX(x)=∑y∈JQ(x,y). For x with PX(x)>0, define PY∣x(y)=P(x,y)/PX(x); define QY∣x(y)=Q(x,y)/QX(x) for every x, since strict positivity of Q gives QX(x)>0. These are probability vectors on J; terms weighted by PX(x)=0 are omitted. Then D(P∥Q)=D(PX∥QX)+∑x:PX(x)>0PX(x)D(PY∣x∥QY∣x). For every subset C⊆I×J, grouping the atoms into C and its complement cannot increase relative entropy: D(P∥Q)≥d(P(C)∥Q(C)),d(a∥b)=aln⁡ab+(1−a)ln⁡1−a1−b, with zero-weight terms omitted. For the empty or full event, the binary divergence is 0, and the grouping bound reduces to D(P∥Q)≥0, with equality exactly when P=Q. For a nonempty proper event C, equality holds exactly when, on each of C and Cc having positive P-mass, the conditional probability vectors of P and Q agree.

Facts & Assumptions

Given: Finite nonempty sets I,J, probability vectors P,Q on I×J, strict positivity of Q, and the natural logarithm convention.

[F1]

Nonnegativity and equality for finite classical relative entropy: for finite probability vectors with positive reference vector, relative entropy as defined above is nonnegative and vanishes exactly when the probability vectors coincide; its proof derives the scalar inequality −ln⁡u≥1−u from the logarithm derivative.

Proof

1.1givenF2algebra

Write px=PX(x) and qx=QX(x). Strict positivity of Q gives qx>0 and QY∣x(y)>0 for each x,y. If P(x,y)>0, then px>0 and ln⁡P(x,y)Q(x,y)=ln⁡pxqx+ln⁡PY∣x(y)QY∣x(y) by [F2]. Multiply by P(x,y) and sum over positive joint atoms. For each x with px>0, summing the first term over y gives pxln⁡(px/qx), and summing the second gives pxD(PY∣x∥QY∣x). Atoms with px=0 contribute no terms. This is exactly the claimed chain rule.

2.1givenF1F2algebracases∎

If C is empty or all of I×J, then P(C)=Q(C) is respectively 0 or 1, so the binary divergence is 0 under the stated zero-weight convention. The grouping bound is then D(P∥Q)≥0, with equality exactly when P=Q by [F1]. For a nonempty proper C, strict positivity gives 0<Q(C)<1. Put a=P(C) and b=Q(C). When a>0, the conditional laws on C are PC(z)=P(z)/a and QC(z)=Q(z)/b; the latter is strictly positive on C. The logarithmic identity ln⁡(P(z)/Q(z))=ln⁡(a/b)+ln⁡(PC(z)/QC(z)) for each z∈C with P(z)>0 yields ∑z∈C:P(z)>0P(z)ln⁡P(z)Q(z)=aln⁡ab+aD(PC∥QC). If a=0, both the sum and the weighted conditional term are zero, so the same identity holds with that term omitted. Applying the identical calculation to Cc and adding gives D(P∥Q)=d(a∥b)+aD(PC∥QC)+(1−a)D(PCc∥QCc), with zero-weight conditional terms omitted. Both weighted conditional divergences are nonnegative by [F1], which proves the grouping inequality, including a=0 or 1. Equality holds precisely when every conditional divergence with positive weight vanishes; by the equality clause of [F1], this is equivalent to agreement of the conditional probability vectors on each positive-P-mass block. Zero-P-mass blocks impose no conditional-law condition.

Depends on

Dependency tree · 2 levels

Sources