REVIEW 4 minor 13 references
The paper proves that a homogeneous posynomial that is zero-free on a product of open right half-planes has a concave degree-normalized root, extending Gårding's theorem to real exponents.
Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →
T0 review · deepseek-v4-flash
2026-08-02 07:41 UTC pith:NF42YIMH
load-bearing objection Gårding's theorem for posynomials is real: the paper closes the factor-of-two gap between sector stability and fractional log-concavity, and the proof is clean enough to send to referees.
G{aa}rding's Theorem for Posynomials
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is that zero-freeness in Γ1 (product of open right half-planes) is enough to make x ↦ p(x)^{1/d} concave on the positive orthant for any homogeneous posynomial of degree d. The proof establishes an angular-contraction estimate—that the argument of p along paths whose coordinate arguments lie in an interval of width φ < π is confined to [0, dφ]—and then uses the representation theorem for complete Bernstein functions to pass from that one-variable angular bound to concavity. The corollary sharpens the connection: Γα-stability implies α-fractional log-concavity, with the α dependence optimal.
What carries the argument
The angular-contraction lemma: for a Γ1-stable homogeneous posynomial p of degree d, if each coordinate of w has argument in [0, φ] with φ < π, then p(w) ≠ 0 and the continuous argument along the straight-line path satisfies 0 ≤ Arg(p(w)) ≤ dφ. This reduces the multivariate stability condition to a one-variable ray monotonicity statement for posynomials with real exponents, which is proved by a generalized Descartes rule; the one-variable statement then feeds a concavity criterion from the complete Bernstein function representation.
Load-bearing premise
The proof leans on the complete representation of a one-variable function that satisfies the angular condition 0 ≤ arg h(z) ≤ arg z; the concavity conclusion requires that the representing measure in that representation be positive, a fact the paper imports without proof.
What would settle it
Exhibit a homogeneous posynomial with positive real coefficients that is zero-free on (open right half-plane)^n but whose degree-normalized root p(x)^{1/d} is not concave on the positive orthant. Since the theorem is concise, the simplest check is numerical: sample two-variable posynomials with non-integer exponents, test zero-freeness on a grid of half-plane points, and test concavity by the midpoint inequality.
If this is right
- If p is Γα-stable, then x ↦ log p(x^α) is concave, i.e. α-fractional log-concavity, for all 0 < α ≤ 1.
- The α dependence is best possible: p(x,y)=x^β+y^β is Γα-stable exactly when αβ ≤ 1, and concavity of (x^{αβ}+y^{αβ})^{1/(αβ)} holds in the same range.
- For sector-stable generating polynomials of k-subset distributions, the k↔k−r down-up walk with r=⌈1/α⌉ mixes in time O_α(k^{1/α} log(1/μ(S0)) log(1/ε)) and has modified log-Sobolev constant Ω_α(k^{-1/α}).
- Domain sparsification reduces sampling from such distributions to sparse external fields supported on n^{1−α} poly(k) elements.
- For fixed-size matchings and nonsymmetric DPPs, the new certificate yields two-site walks with Ω(k^{-2}) log-Sobolev bounds and sparse domains of size n^{1/2} poly(k), improving the previous n^{3/4} poly(k).
Where Pith is reading between the lines
- The real-exponent formulation suggests that fractional log-concavity is not an artifact of integer exponents; the same sector-stability certificate might be used directly on generating functions with non-polynomial terms, e.g., partition functions with external fields, without first converting to polynomials.
- Because Corollary 2 is sharp, any sampling algorithm whose mixing guarantee is governed by fractional log-concavity cannot hope for a better exponent from this certificate alone; further gains must come from additional structure, such as negative dependence or spectral gaps.
- The angular-contraction technique may generalize to ratios of posynomials or to functions defined by integrals over the positive orthant, provided the one-dimensional ray argument can be extended to a suitable class of functions.
- An immediate testable extension is to non-homogeneous posynomials: if one normalizes by the total degree and checks Γ1-stability after homogenization, a similar concavity statement may hold; the current proof relies on homogeneity for the perspective step.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proves an analogue of Gårding's theorem for homogeneous posynomials with arbitrary nonnegative real exponents. The main result (Theorem 1) states that if such a posynomial is zero-free on a product of open right half-planes (Γ1-stable), then its degree-normalized root is concave on the positive orthant. Corollary 2 extends this to sector stability: Γα-stability implies α-fractional log-concavity. The proof proceeds through a real-exponent generalized Descartes rule (Lemma 3), a Sendov–Sendov ray-monotonicity theorem (Theorem 4), an angular-contraction lemma (Lemma 5), and a Pick-representation-based concavity lemma (Lemma 6). The applications section derives improved mixing-time and domain-sparsification bounds for fixed-size matchings and nonsymmetric determinantal point processes by removing a previously known factor-of-two loss.
Significance. If the result holds, it is a clean and useful extension of Gårding's theorem to posynomials, with a sharp dependence on the sector aperture α and concrete algorithmic consequences. The proof is essentially self-contained and, notably, has no fitted parameters or post hoc assumptions: the main theorem is derived from first principles with the only external input being the standard Pick/complete-Bernstein representation. The sharpness example shows the exponent range is optimal. The paper also transparently discloses the AI-assisted development of the proof. My reading of the central arguments is that they are correct: the generalized Descartes rule, ray monotonicity, angular contraction, and the perspective argument all check out. The one non-elementary step—the use of the complete-Bernstein characterization in Lemma 6—is correct but terse, and should be explained more explicitly for the intended CS readership.
minor comments (4)
- [§3, Lemma 6] The proof asserts, without derivation, that the hypotheses 0≤arg h(z)≤arg z imply that h is a complete Bernstein function, and then cites the Pick representation in [SSV12, Theorem 6.2]. This is the single most load-bearing step in the paper, and although the implication is correct (the sector condition is equivalent to h and -h/z being Pick functions, which is the standard characterization of complete Bernstein functions), the manuscript should either prove this equivalence or give a precise citation to the characterization. As written, a reader cannot tell whether the cited theorem is being used for the representation or for the classification.
- [§2, Theorem 4] In the definition of H_β, some coefficients a_j sin(λ_jθ−β) may vanish. The sign-change count should explicitly say that zero coefficients are discarded before applying Lemma 3. This is a minor amendment but avoids ambiguity.
- [§3, Lemma 5] The proof of Lemma 5 says that the lift obtained from Theorem 4 is the same branch as the one obtained by moving from 1 to e^{iφ}. This is true because f is zero-free on ℂ\ (−∞,0], which is simply connected, and both paths lie in that set. Stating this explicitly would improve clarity, since Theorem 4 is phrased along a ray of fixed angle, while the desired path is along the unit circle.
- [§1, Corollary 2] The sharpness example p(x,y)=x^β+y^β is stated without proof. A one-sentence explanation of why Γα-stability holds exactly when αβ≤1 (using the equation (z1/z2)^β=−1) would be useful.
Circularity Check
No significant circularity; the main theorem is derived from first principles with an external Bernstein-function representation theorem.
full rationale
The main derivation chain is Theorem 1 <= Lemma 5 <= Theorem 4 <= Lemma 3; all are proved inside the paper or from the classical (and cited) Pick/complete-Bernstein representation [SSV12, Theorem 6.2] in Lemma 6. None of these steps defines a quantity in terms of the claimed conclusion or fits a parameter and calls it a prediction. Theorem 4 is a strengthened version of Sendov-Sendov ray monotonicity and is proved, not assumed, using the generalized Descartes rule (Lemma 3), which is also proved. Lemma 6 is an external characterization theorem; the paper explicitly states the representation used and the positivity of the measure, so the concavity conclusion follows by integrating concave kernels against a positive measure. This is an input theorem, not a restatement of the target concavity. Corollaries 7 and 8 invoke the author's prior entropic-independence and domain-sparsification theorems as black-box application theorems; these citations are load-bearing only for the sampling consequences, not for the new Gårding-type theorem, and they state independent results rather than assuming the paper's conclusion. No fitted constants, parameter-renaming, or definitional equivalences occur. The only caveat is that the complete-Bernstein characterization is a nontrivial external input; relying on it is a correctness/verification concern, not circularity.
Axiom & Free-Parameter Ledger
axioms (5)
- domain assumption Real powers are defined using the principal logarithm on C\(-∞,0].
- standard math Generalized Descartes rule: a real-exponent exponential sum has at most as many positive zeros as its coefficient sign changes.
- standard math Complete Bernstein (Pick) representation theorem from [SSV12, Theorem 6.2].
- domain assumption Entropic-independence and domain-sparsification theorems from [Ana+21, Theorem 5] and [Ana+22].
- domain assumption Sector-stability is preserved by nonzero links and positive external fields.
Cite this review
Pith. "Pith review of G{\aa}rding's Theorem for Posynomials." pith.science (2026). https://pith.science/paper/NF42YIMH
@misc{pith2026260709168,
author = {Pith},
title = {Pith review of: G\aarding's Theorem for Posynomials},
year = {2026},
howpublished = {\url{https://pith.science/paper/NF42YIMH}},
note = {Machine review of arXiv:2607.09168}
}
read the original abstract
We extend G{\aa}rding's theorem to homogeneous posynomials: if a finite positive sum of monomials with arbitrary nonnegative real exponents is zero-free on a product of right half-planes, then its degree-normalized root is concave. Consequently, zero-freeness in a sector of aperture $\alpha\pi$ implies $\alpha$-fractional log-concavity. This sharpens generic mixing and domain-sparsification guarantees for fixed-size matchings and nonsymmetric determinantal point processes. The result was developed in an AI-assisted interaction initiated and checked by the author; Codex also assisted with assembling and typesetting the manuscript.
Reference graph
Works this paper leans on
-
[1]
An Inequality for Hyperbolic Polynomials , journal =
G. An Inequality for Hyperbolic Polynomials , journal =
-
[2]
Canadian Journal of Mathematics , volume =
Sendov, Blagovest and Sendov, Hristo , title =. Canadian Journal of Mathematics , volume =. 2021 , doi =
2021
-
[3]
Bernstein Functions: Theory and Applications , edition =
Schilling, Ren. Bernstein Functions: Theory and Applications , edition =. 2012 , doi =
2012
-
[4]
Negative Dependence and the Geometry of Polynomials , journal =
Borcea, Julius and Br. Negative Dependence and the Geometry of Polynomials , journal =. 2009 , doi =
2009
-
[5]
Proceedings of the 29th Conference on Learning Theory , series =
Anari, Nima and Oveis Gharan, Shayan and Rezaei, Alireza , title =. Proceedings of the 29th Conference on Learning Theory , series =
-
[6]
Lorentzian Polynomials , journal =
Br. Lorentzian Polynomials , journal =. 2020 , doi =
2020
-
[7]
Duke Mathematical Journal , volume =
Anari, Nima and Oveis Gharan, Shayan and Vinzant, Cynthia , title =. Duke Mathematical Journal , volume =. 2021 , doi =
2021
-
[8]
53rd Annual ACM SIGACT Symposium on Theory of Computing (STOC) , pages =
Alimohammadi, Yeganeh and Anari, Nima and Shiragur, Kirankumar and Vuong, Thuy-Duong , title =. 53rd Annual ACM SIGACT Symposium on Theory of Computing (STOC) , pages =. 2021 , doi =
2021
-
[9]
54th Annual ACM SIGACT Symposium on Theory of Computing (STOC) , pages =
Anari, Nima and Jain, Vishesh and Koehler, Frederic and Pham, Huy Tuan and Vuong, Thuy-Duong , title =. 54th Annual ACM SIGACT Symposium on Theory of Computing (STOC) , pages =. 2022 , doi =
2022
-
[10]
arXiv preprint arXiv:2106.04105 , year =
Anari, Nima and Jain, Vishesh and Koehler, Frederic and Pham, Huy Tuan and Vuong, Thuy-Duong , title =. arXiv preprint arXiv:2106.04105 , year =
-
[11]
and Vuong, Thuy-Duong , title =
Anari, Nima and Liu, Yang P. and Vuong, Thuy-Duong , title =. 63rd IEEE Annual Symposium on Foundations of Computer Science (FOCS) , pages =. 2022 , doi =
2022
-
[12]
Domain Sparsification of Discrete Distributions Using Entropic Independence , booktitle =
Anari, Nima and Derezi. Domain Sparsification of Discrete Distributions Using Entropic Independence , booktitle =. 2022 , doi =
2022
-
[13]
and Lieb, Elliott H
Heilmann, Ole J. and Lieb, Elliott H. , title =. Communications in Mathematical Physics , volume =. 1972 , doi =
1972
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.