Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-generatedPipeline-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.

Numerical FK cylinder limits and pressure comparison

Statement

Let q≥1 be a finite real number. On the nearest-neighbor square lattice let Λn={−n,…,n}2, let En consist of edges with both endpoints in Λn, and let Bn={x∈Λn:∥x∥∞=n}. For p∈(0,1) and a partition ξ of Bn, give A⊆En probability proportional to p∣A∣(1−p)∣En∖A∣qkξ(A), where kξ counts components after the partition classes are identified. Write ϕn,p,qb for the free partition when b=0 and the one-class wired partition when b=1.

For every event C determined by finitely many edge coordinates, the numerical limits πp,qb(C):=lim⁡n→∞ϕn,p,qb(C),b∈{0,1}, exist. They are consistent, nonnegative normalized finite-coordinate probability families, invariant under translations and quarter-turns. On increasing cylinder events they are increasing in p, with π0≤π1. Every cylinder-probability limit of arbitrary boundary partitions lies between these two families on increasing cylinder events. There is an at most countable set Dq⊂(0,1) such that πp,q0(C)=πp,q1(C) for every cylinder event C whenever p∉Dq.

Let Dr be the event that actual open edges inside Λr join 0 to Br, and define the real numbers Θb(p,q)=inf⁡r≥1πp,qb(Dr). Then Θ1(p,q)≤Θ0(p′,q) for 0<p<p′<1, and lim⁡n→∞ϕn,p,q1(0↔Bn)=Θ1(p,q). Connections here use actual edges, not artificial boundary-identification paths. These are statements about real limits and finite probability tables; existence of a countably additive probability on the infinite configuration space is not asserted.

Facts & Assumptions

Given: Finite real q≥1, p∈(0,1), the finite box laws in the Statement, and cylinder events depending on a fixed finite set of edges.

[F1]

The positive finite FK weights and arbitrary boundary partitions are defined in Finite FK random-cluster weights for real q and boundary partitions.

[F2]

Finite FK laws have spatial Markov conditioning, positive association and monotonicity in parameter and boundary wiring. Their common-uniform heat-bath coupling preserves coordinatewise order and converges to the corresponding finite laws Finite FK association, spatial Markov property, and boundary monotonicity.

Proof

technique · finite boundary comparison, positive-activity tiling and one-variable convexity

Given: [F1]–[F2] and the data in the Statement.

1.1F1F2algebra

If n<m, condition the law on Em on edges outside En. Its induced partition on Bn refines the wired partition and is coarser than the free partition. By [F2], free laws increase under restriction as volume grows and wired laws decrease. Every all-open event on a fixed finite set therefore has a bounded monotone sequence of probabilities. An atom prescribing open edges O and closed edges H has probability ∑T⊆H(−1)∣T∣ϕ(all edges of O∪T open). Finite inclusion-exclusion gives the limit of every atom and hence every cylinder event. Limits of finite sums preserve positivity, normalization and all restriction-consistency identities. No measure extension is used.

1.2F1algebra

Write h=log⁡(p/(1−p)) and v=eh. For QN={0,…,N−1}2 define the activity sum and its volume normalization YNb(h)=∑A⊆E(QN)eh∣A∣qkb(A),FNb(h)=N−2log⁡YNb(h). The finite law is unchanged by replacing the original weights by these activity weights. Since 1≤qk(A)≤qN2 and ∣E(QN)∣=2N(N−1), one has 0≤FN0(h)≤log⁡q+2log⁡(1+eh). For fixed L, write N=kL+s, 0≤s<L, tile the kL-square by k2 disjoint L-squares, and leave the remaining vertices as singletons. Restricting the sum to configurations closing every interpiece edge gives YN0(h)≥(YL0(h))k2qN2−k2L2. Closed edges have activity factor 1 and component factors multiply over the pieces. Hence lim inf⁡NFN0(h)≥FL0(h) for each L. The finite bound above and FN0≤sup⁡LFL0 prove that F(h)=lim⁡NFN0(h)=sup⁡LFL0(h) exists and is finite.

2.1F2step 1.1algebra

The same comparison holds for any two nested translated boxes. For a fixed z∈Z2, put r=∥z∥∞; then Λn−r⊆Λn+z⊆Λn+r. For every fixed increasing cylinder event, the translated free probability is sandwiched between those from the two unshifted boxes, with reverse order for wired probabilities. Their limits coincide by step 1.1. Inclusion-exclusion gives equality for all cylinders. Quarter-turn invariance holds already in each centered finite box. Thus each edge orientation and position has the same limiting open probability ub(p,q). Parameter monotonicity and π0≤π1 follow from [F2]. For an arbitrary-boundary cylinder limit ν, condition on the exterior of a fixed box Λr, use ϕr0(C)≤νj(C)≤ϕr1(C) for increasing C supported inside it, pass to the given numerical limit, then let r→∞. This proves the asserted sandwich.

2.2F1step 1.2algebra

Wiring boundary vertices lowers component counts by an integer between 0 and ∣∂QN∣−1, so q−(∣∂QN∣−1)YN0(h)≤YNξ(h)≤YN0(h). As ∣∂QN∣≤4N, every boundary partition has the same pressure limit F(h). Differentiating the finite sums gives (FNb)′(h)=N−2EQN,h,qb∣A∣,(FNb)′′(h)=N−2Var⁡QN,h,qb(∣A∣)≥0. Thus these functions and their finite pointwise limit F are convex.

3.1step 2.2algebra

Ordered secant slopes of a finite convex function have finite one-sided limits F−′(h)≤F+′(h), bounded locally by secants to fixed points on either side. If h<k then F+′(h)≤F−′(k). Therefore the nonempty intervals (F−′(h),F+′(h)) at nondifferentiability points are pairwise disjoint. Assign each its first rational in a fixed enumeration: this is an injection into a countable set. Let Dq be the image of these at most countably many h under h↦eh/(1+eh). At a differentiability point, for each t>0 convexity gives FNb(h)−FNb(h−t)t≤(FNb)′(h)≤FNb(h+t)−FNb(h)t. Pass to lower and upper numerical limits as N→∞, then let t↓0; the common pressure limit proves (FNb)′(h)→F′(h) for both boundaries.

4.1F2step 1.1step 2.1step 3.1algebra

Finite free edge probabilities in QN are at most the infinite-volume numerical value u0, and finite wired ones are at least u1, by nested-box comparison and step 2.1. Thus (FN0)′(h)≤(2−2/N)u0≤(2−2/N)u1≤(FN1)′(h). At p∉Dq both outside quantities tend to F′(h), so u0=u1. For an arbitrary cylinder event C supported on H, use the common-uniform ordered finite heat-bath chains from [F2]. At every time the difference of their C probabilities in absolute value is at most their probability of differing on H, which by order and a union bound is at most the sum over e∈H of their mean edge differences. Let time tend to infinity, then the finite box tend to the lattice. This gives ∣πp,q1(C)−πp,q0(C)∣≤∣H∣(u1−u0)=0. This proves equality of the numerical families, without constructing an infinite coupling or law.

5.1F2step 1.1step 4.1algebra∎

If p<p′, select one t∈(p,p′)∖Dq. Such a point exists because a countable set cannot fill a real interval; it requires no sequence of arbitrary choices. For every increasing cylinder event C, πp,q1(C)≤πt,q1(C)=πt,q0(C)≤πp′,q0(C). Apply this to Dr and take the infimum over r to obtain Θ1(p,q)≤Θ0(p′,q). Finally let an=ϕn,p,q1(Dn). For m≥r, truncation of a path at its first smaller-boundary hit and [F2] give am≤ϕm,p,q1(Dr)≤ar. In particular an decreases to a real L. Letting m→∞ gives L≤πp,q1(Dr)≤ar. Taking the infimum over r proves L=Θ1(p,q).

Remarks

The proof constructs finite-coordinate numerical families. Turning them into countably additive Borel laws requires the separately stated measure-extension hypotheses. No equality of free and wired laws is asserted at an exceptional parameter.

Depends on

Dependency tree · 2 levels

Sources