Pith. sign in

REVIEW 3 major objections 4 minor 5 references

On the dependence of the zero-free region of a partition function on the external field

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Linear multi-spin energy growth needs only log external field growth

desk verdict Solid new zero-free regime with a repairable gap in the branch-of-log step; worth refereeing. read the letter →

arxiv 2608.03687 v1 pith:2XUJE2HX submitted 2026-08-04 math-ph math.COmath.MPmath.PR

Alexander Barvinok This is my paper · ORCID
classification math-phmath.COmath.MPmath.PR MSC 82B2030C1568R0568W0560C05
keywords zero-freeregionpartitionfunctionexternalfieldmulti-spininteractionsp-biasedBooleancubeLipschitzfunctionsspinsystemsphasetransition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proves a zero-free region for a general class of partition functions on the p-biased Boolean cube: sums of 1-Lipschitz functions, each depending on at most r coordinates, with the total influence of every coordinate bounded by 1/(10√(rp)), have a nonzero expectation of exp(Σ λ_i φ_i) as long as rp≥12. In spin language, this means that when interactions involve r spins and the external field is not too weak, the strength of the external field needed to keep the partition function away from zero grows only logarithmically with the interaction energy, not linearly. A companion result covers the small-p regime, where a linear growth of the field is again required. The proof is an induction that compares the expectation before and after perturbing one interaction, controlled by a concentration lemma for complex Lipschitz functions on the biased cube. This yields, among other things, a quasi-polynomial algorithm for approximating such partition functions.

What carries the argument

The argument rests on Theorem 3.1, a concentration lemma for a single complex L-Lipschitz function f on the p-biased cube: if pn≥12 and L=1/(5√(pn)), then |E e^f| ≥ (1/2)E|e^f| (Part (1); Part (2) gives a factor 1/5 when p=e^{-6L}/n). This injects the 1/√(pn) scale into the induction. The induction (Claim 4.1) shows that replacing the last function φ_m by any 1-Lipschitz function within distance 1 that depends on the same coordinates changes the expectation by a factor e^α with |α| ≤ 2|λ_m|; the ratio bound is obtained by interpolating linearly between φ_m and the perturbed function, extending the expectation to the solid cube [0,1]^J to select a consistent branch of the logarithm, and apply

What would settle it

Directly test Theorem 3.1(1) by random or exhaustive search: for n with pn=12, set L=1/(5√(pn)), and sample complex 1-Lipschitz functions f (e.g., f(x)=L(g(x)+i h(x)) with g,h real 1-Lipschitz and Eg=Eh=0); compute the ratio |E e^f| / E|e^f|. A single sample with ratio below 1/2 would falsify the lemma and force a smaller constant in Theorem 1.2. Alternatively, enumerate small n to find a function achieving the infimum and check whether the infimum is indeed ≥1/2.

Watch

Extended reading notes

Core claim

Theorem 1.2 states: on the p-biased cube with p≤q and rp≥12, if φ_1,...,φ_m are complex 1-Lipschitz functions each depending on at most r coordinates, and the sum of |λ_i| over functions depending on coordinate j is at most 1/(10√(rp)) for every j, then E exp(Σ λ_i φ_i) ≠ 0. Reading λ_i as interaction strengths and p as a function of the external field α through p=e^{-βα}/(e^{βα}+e^{-βα}), this says the external field needs to grow only logarithmically with the energy of multi-spin interactions to keep the partition function zero-free. A second theorem covers the very small p regime (p=e^{-6L}/r), where the field must again grow linearly, matching cluster-expansion predictions.

Load-bearing premise

The whole argument rests on Theorem 3.1(1): a complex-valued function on the biased cube whose values never change by more than L=1/(5√(pn)) between neighboring points must have its expectation of e^f at least half as large in absolute value as the expectation of |e^f|; if that factor 1/2 or the 1/√(pn) scale fails, the per-coordinate bound 1/(10√(rp)) and the logarithmic-field conclusion would degrade.

Editorial extensions

If this is right

  • For a fixed inverse temperature β and interaction range r with rp≥12, multiplying the interaction strengths by a factor c only requires adding O(ln c) to the external field α to keep Z(H,β) ≠ 0 and the pressure analytic.
  • The zero-free region of the partition function as a function of the external field extends to a disc whose radius grows at least as p^{-1/2} as p decreases (for fixed r), up to constants.
  • The theorem yields a quasi-polynomial algorithm approximating the partition function within relative error ε, using O(m^k) expectations with k = O_δ(ln r + ln Σ|λ_i| - ln ε) (Section 1.5, Lemma 2.1).
  • In the hypergraph matching / polymer model, smaller selection probability p enlarges the zero-free disc for the penalty parameter λ, so stronger penalties can be charged while still approximating the matching statistics.
  • The dependence of the bound on r as r^{-1/2} is optimal up to a constant, per the author's earlier work [Ba26]; the constant 1/10 could presumably be improved but not the scaling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The crossover between the logarithmic regime (Theorem 1.2) and the linear regime (Theorem 1.3) is governed by the product rp: the paper's threshold rp≥12 suggests the transition occurs when the external field α is of order (1/β) ln(r/12); one could test numerically in a concrete model whether the zero-free region's shape changes sharply near that value.
  • The key lemma Theorem 3.1(1) has the flavor of an anti-concentration inequality; the same inductive scheme should transfer to other product spaces (e.g., continuous or discrete spins with more than two values), where a similar lower bound on |E e^f| would yield analogous logarithmic-field theorems.
  • The factor 1/2 in Theorem 3.1(1) is not optimized; if it could be improved toward 1, the per-coordinate bound 1/(10√(rp)) could be enlarged, potentially improving the constants in the algorithmic application.
  • The approximation algorithm derived from the zero-free region (Section 2) does not require real-rootedness or other algebraic structure, so the same strategy might apply to partition functions of hypergraph matchings, where real-rootedness fails; this could be checked by testing the algorithm's predicted O(m^k) complexity on random hypergraphs.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proves zero-freeness of the partition-function expectation E exp{Σ λ_i φ_i} on the p-biased Boolean cube under per-coordinate influence bounds. Theorem 1.2 gives the bound Σ_{i: j∈J_i}|λ_i| ≤ 1/(10√(rp)) when each φ_i is 1-Lipschitz, depends on at most r coordinates, and rp ≥ 12. Theorem 1.3 gives an analogous linear-in-L bound when p = e^{-6L}/r. The proofs are built on Theorem 3.1, a concentration-type lower bound for |E e^f| for complex Lipschitz f, proved via martingale and Laplace-transform estimates (Lemmas 3.2–3.5), and on an inductive claim (Claim 4.1) that modifying one interaction changes the log-partition by at most a constant multiple of |λ_m|. Section 2 sketches a quasi-polynomial approximation algorithm based on the zero-free regions.

Significance. If the result holds, it is significant: it extends Barvinok's earlier multi-spin zero-free work to include an external field, exhibits a regime where the external field needs only grow logarithmically in the interaction strength, and achieves a scaling in r that is optimal up to a constant. The local analytic estimates are clean, self-contained, and mostly machine-checkable. However, the induction in Section 4 contains a false geometric assertion about an affine extension, and the current proof of the key Lipschitz estimate for the log-partition is incomplete. The theorem may well be true and the gap appears repairable, but the manuscript as written does not establish the advertised result.

major comments (3)
  1. [§4.3, after Eq. (4.3.4)] The assertion that Ψ(t) is affine in each coordinate τ_j is false. For fixed other coordinates, each summand inside the expectation is of the form P(z) exp(A_z τ_j + B_z); a sum of such exponentials is not affine unless all A_z vanish. For example, with one inner coordinate, p=q=1/2, λ=1, φ(ξ)=ξ, Ψ(τ)=(e^τ+1)/2. Consequently, the argument of Ψ need not change by less than π along an edge, and the construction of ψ satisfying (4.3.2) is invalid as written. This is load-bearing: (4.3.2) supplies the Lipschitz bound for λ_m φ_τ + ψ used in (4.3.6) and hence in the final application of Theorem 3.1. The gap is likely repairable by defining log-ratio representatives on edges and checking consistency on 4-cycles using the smallness of the per-coordinate sums, but that argument is absent.
  2. [§4.2 and §4.3, applications of Theorem 3.1(1)] Theorem 3.1 Part (1) is applied to the cube {0,1}^J in the base case and in the induction step, but its hypothesis is p|J| ≥ 12. The theorem's assumption rp ≥ 12 only gives p|J| ≥ p·(something ≤ r), which can be much smaller than 12 when |J| is small. For instance, r=100, p=0.12, |J|=1 gives p|J|=0.12. Since J can be much smaller than r, the stated application is unjustified. A repair would embed J into r coordinates by adding dummy coordinates that the functions do not depend on; this is straightforward but not present.
  3. [§4.3, ratio bound for Ψ(x)/Ψ(y)] In the paragraph after Eq. (4.3.1), the paper claims that Ψ(x)/Ψ(y)=e^α with |α| ≤ 2Σ_{i∈I_j}|λ_i| = 2Σ_{1≤i≤m-1: j∈J_i}|λ_i|. The equality is false: functions i∉I can depend on the varying coordinate j∈\bar J and contribute a scalar factor e^{Σ λ_i(φ_i(x)-φ_i(y))} to the ratio. The final Lipschitz estimate can still be made to work using the larger bound 2Σ_{all i≤m-1: j∈J_i}|λ_i|, but the displayed equality and the subsequent conclusion are not justified as written.
minor comments (4)
  1. [§3.6, Part (2)] The displayed inequality E e^{2g} ≤ exp{e^{-5L}} is incorrect: since 2g is 2L-Lipschitz, Corollary 3.4(2) gives E e^{2g} ≤ exp{e^{-6L} e^{2L}} = exp{e^{-4L}}. The subsequent bound b ≤ √2 e^{-2} still holds with this correction because e^{-4L}/2 ≤ 1/2, so the error does not affect the final estimate.
  2. [§4.3 notation] The notation for J and its complement is very confusing, especially since overbars are lost in places. The same symbol J is used for the set of coordinates of φ_m, for the complement, and for indexing the solid cube. Please use distinct notation, e.g., J for the interaction set and K for its complement.
  3. [Claim 4.1, Eq. (4.1.2)] Eq. (4.1.2) writes E{λ_mφ_m + Σ...} where the exponential is missing; the text later uses exp. Similarly in Eq. (5.1.1). This is a typo that should be corrected.
  4. [Theorem 1.3] The condition p = e^{-6L}/r may violate the standing assumption 0 < p ≤ q unless L is sufficiently large (e.g., L ≥ (1/6)ln(2r)). The theorem statement should either impose this or note that the argument works for p>q as well.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proof is a self-contained derivation from Theorem 3.1, with no fitted parameters and no load-bearing self-citation.

full rationale

The paper's main theorems do not reduce to their inputs by construction. Theorems 1.2 and 1.3 are consequences of Theorem 3.1, which is proved in Section 3 from Lemmas 3.2-3.5 by elementary two-point estimates, a martingale induction, and Laplace-transform concentration bounds. The hypotheses of the theorems (the rp>=12 condition and the per-coordinate influence bounds) enter exactly where Theorem 3.1 is invoked, and the induction in Claim 4.1 proves the nonvanishing and ratio bounds rather than assuming them. The single self-citation [Ba26] is used only for methodological context and as an external optimality benchmark ('the dependence on r is optimal up to a constant [Ba26]'); it is not an input in the proof, and the paper explicitly states that the method of [Ba26] does not lead to the new results. The statistical-physics interpretation in Section 1.4 is a faithful translation of the theorem's quantitative bounds into the external-field parameterization p=e^{-beta alpha}/(e^{beta alpha}+e^{-beta alpha}); it is not a renaming of a fitted quantity or a prediction equivalent to an input. The skeptical remark about the affine-continuation argument in Section 4.3 and the displayed inequality slip in the proof of Theorem 3.1 Part (2) are internal correctness issues, not circularity: even if those steps are flawed, the paper's derivation chain does not become equivalent to its assumptions. No fitted parameters, no post-hoc exclusions, and no self-referential uniqueness claims are used. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new entities and no free parameters: every constant in the statements (10, 5, 6, 12, delta) arises from explicit bounds in the proofs, not from fitting or ad hoc tuning. The derivation is self-contained given standard complex analysis and probability tools, with [Ba26] cited only as an external benchmark.

assumptions (5)
  • standard math Borel-Carathéodory inequality and Cauchy bounds for holomorphic functions
    Used in Lemma 2.1 to bound the Taylor error of ln g via the sup of Re h on a larger disc; quoted from Lang [La99].
  • standard math Jensen's inequality, Markov inequality, Cauchy-Schwarz
    Used in Sections 3.5 and 3.6 to bound tail probabilities and to separate the real and imaginary parts of f in Theorem 3.1.
  • domain assumption Product measure on {0,1}^n with independent Bernoulli(p) coordinates, p ≤ q
    Standing assumption (1.1.1); the martingale concentration machinery (Lemmas 3.3 and 3.5) depends on coordinate independence and on the variance scale pq.
  • domain assumption Interactions are 1-Lipschitz in the Hamming metric with support of size at most r
    Defines the class of interactions; the paper controls the Lipschitz constant rather than the sup-norm of φ_i, which is the claimed improvement over cluster expansion.
  • domain assumption Optimality of the r^{-1/2} dependence up to a constant is taken from [Ba26]
    Section 1.2 states the dependence on r is optimal up to a constant citing the author's own prior work; this benchmark is not proven in the present paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the dependence of the zero-free region of a partition function on the external field." pith.science (2026). https://pith.science/paper/2XUJE2HX

@misc{pith2026260803687,
  author       = {Pith},
  title        = {Pith review of: On the dependence of the zero-free region of a partition function on the external field},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2XUJE2HX}},
  note         = {Machine review of arXiv:2608.03687}
}
abstract

Let $\{0, 1\}^n$ be the Boolean cube, endowed with the probability product measure, where ${\Bbb P}(1)=p$ and ${\Bbb P}(0)=q$ with $0 < p \leq q=1-p$. Let $\phi_i: \{0, 1\}^n \longrightarrow {\Bbb C}$ be $1$-Lipschitz functions in the Hamming metric, such that each $\phi_i$ depends on at most $r$ coordinates of $x \in \{0, 1\}^n$, where $rp \geq 12$. For $j=1, \ldots, n$, let $I_j $ be the set of indices $i$ such that $\phi_i$ depends on the $j$-th coordinate. We prove that $E\thinspace \exp\left\{ \sum_{i=1}^m \lambda_i \phi_i \right\} \ne 0$ provided $\lambda_i \in {\Bbb C}$ satisfy $\sum_{i \in I_j} |\lambda_i| \leq {1 \over 10 \sqrt{rp}}$ for all $j$. This translates into a regime for $\pm 1$ spin systems, where a linear increase in the energy of multi-spin interactions requires only a logarithmic increase of the external field to keep the partition function zero-free and the system away from the phase transition.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

5 extracted references · 4 canonical work pages

  1. [88]

    Jerrum and A

    [JS89] M. Jerrum and A. Sinclair,Approximating the permanent, SIAM Journal on Computing 18(1989), no. 6, 1149–1178. [La99] S. Lang,Complex Analysis, Graduate Texts in Mathematics,103, Springer-Verlag, New York. [LY52] T.D. Lee and C.N. Yang,Statistical theory of equations of state and phase transitions. II. Lattice gas and Ising model, Physical Review (2)...

    work page 1989

  2. [248]

    Mayer,The statistical mechanics of condensing systems

    [Ma37] J.E. Mayer,The statistical mechanics of condensing systems. I, The Journal of Chemical Physics5(1937), 67–73. [PR17] V. Patel and G. Regts,Deterministic polynomial-time approximation algorithms for partition functions and graph polynomials, SIAM Journal Computing46(2017), no. 6, 1893–1919. [SF71] M. Suzuki and M.E. Fisher,Zeros of the partition fun...

    work page 1937

  3. [2009]

    [L+19] J. Liu, A. Sinclair, and P. Srivastava,The Ising partition function: zeros and determin- istic approximation, Journal of Statistical Physics174(2019), no. 2, 287–315. [Mc98] C. McDiarmid,Concentration, Probabilistic Methods for Algorithmic Discrete Math- ematics, Algorithms and Combinatorics,16, Springer-Verlag, Berlin, 1998, pp. 195 –

    work page 2019

  4. [2016]

    Barvinok,Approximating real-rooted and stable polynomials, with combinatorial ap- plications, Online Journal of Analytic Combinatorics (2019), No

    [Ba19] A. Barvinok,Approximating real-rooted and stable polynomials, with combinatorial ap- plications, Online Journal of Analytic Combinatorics (2019), No. 14, Paper No. 8, 13 pp. [Ba26] A. Barvinok,On the zeros of partition functions with multi-spin interactions, online first, DOI 10.4171/AIHPD/228, Annales de l’Institut Henri Poincar´ e D. Combinatorics...

    doi:10.4171/aihpd/228 2019

  5. [2018]

    Galanis, L.A

    [G+22] A. Galanis, L.A. Goldberg and A. Herrera-Poyatos,The complexity of approximating the complex-valued Potts model, Computational Complexity31(2022), no. 1, Paper No. 2, 94 pp. [Go81] C.D. Godsil,Matchings and walks in graphs, Journal of Graph Theory5(1981), no. 3, 285–297. [HL72] O.J. Heilmann and E.H. Lieb,Theory of monomer-dimer systems, Communicat...

    work page 2022

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.