{"id":"76bf3c87-f431-4fe2-bfcf-3f34efbad774","arxiv_id":"2607.19207","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A quasi-polynomial is eventually nondecreasing exactly when certain moment inequalities on its h-vector hold, and the number of such nonnegative h-vectors of fixed degree and period is a quasi-polynomial in the volume.","lead":"This paper gives complete conditions that tell when a quasi-polynomial—a function built from a few repeating polynomial formulas—eventually stops decreasing and only grows. It also counts these functions with nonnegative h-vectors, showing the count itself follows a repeating polynomial pattern.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5.4 is false as stated: CM alone does not determine v without fixing support; the theorem is likely salvageable but the proof needs revision.","rationale":"Theorem 3.1 and the non-polyhedrality proof appear internally sound; the reader's weakest-assumption pick, Lemma 4.4, actually holds. However, the paper contains a false supporting lemma (Lemma 5.4) that is used in the proof of Theorem 5.3, and a related flaw in Lemma 8.7. Since the vertex classification underpins the growth-rate results in Section 7, the proof as written has a genuine gap. The gap is localized: all applications of Lemma 5.4 occur with a fixed zero pattern, and an easy repair is to restate the lemma with the support included. Therefore I do not reject, but I would make acceptance conditional on correcting these statements and re-verifying Theorem 5.3's proof.","tokens_in":23968,"tokens_out":43737,"duration_ms":412872,"concrete_test":"Use a polyhedral library (e.g., SageMath) to enumerate all vertices of S1 for small parameters, say (d,p) = (3,3), (4,3), and (2,4). Check whether the vertex list exactly matches the three conditions in Theorem 5.3. If every vertex satisfies the conditions and every vector satisfying the conditions is a vertex, the central vertex classification survives despite the flawed Lemma 5.4. Independently, evaluate the two explicit 15-dimensional vectors above to confirm they are distinct elements of S1 with equal CM vectors, which settles the false statement.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The reader's flagged worry about Lemma 4.4 does not land: the total-sum condition exactly supplies the boundary value h_{p-1,d} = -α_{p-1,d}, so the cumulative-sum construction is coherent. The real soft spot is Lemma 5.4. As stated it claims that if v ∈ S1 has at most two nonzero entries in each mod class, then the vector CM(v) uniquely determines v. This is false. For d = 4, p = 3, take v = (0.5,0,0.5,0,0 | 0,1,0,0,0 | 0,1,0,0,0) and w = (0,1,0,0,0 | 0,1,0,0,0 | 0,1,0,0,0). Both lie in S1, both have at most two nonzero entries per mod class, and CM(v) = CM(w) = (1,1,1), but v ≠ w. The proof implicitly assumes the support pair (α,ω) is fixed. Every actual use in Theorem 5.3 and Proposition 8.3 does fix a zero pattern, so the main classification is probably recoverable, but the written proof of Theorem 5.3 relies on an overstrong false statement. A related overstrong claim appears in Lemma 8.7: for CM = (1.8,1.2) the prescribed rounding order produces a vector violating (5.1); this only occurs when max(h) < 1, so Lemma 8.1 already covers those cases, but as a standalone lemma it is not correct.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies quasi-polynomials of degree d and period dividing p, encoded by h-vectors in R^{p(d+1)}, and characterizes those that are eventually nondecreasing. Theorem 3.1 reduces the infinite family of inequalities L_h(n+1) >= L_h(n) to finitely many sign conditions on certain moment differences A_r^(ell)(h) and B^(ell)(h). The closure of the eventually nondecreasing cone is shown to be a polyhedral cone (Cor. 3.5), while C_N is polyhedral exactly for d=1,2 (Thm. 4.6). The paper then restricts to nonnegative h-vectors, studies the polytope S_1, and classifies its vertices (Thm. 5.3). Section 6 gives sufficient criteria for a vertex to lie in C_END. Section 7 proves that the number of nonnegative integer h-vectors of volume V that are eventually nondecreasing is a quasipolynomial in V of degree pd (Thm. 7.2), with explicit generating functions for small parameters.","tokens_in":24332,"tokens_out":7207,"duration_ms":71752,"significance":"If the results hold, the paper gives a clean finite characterization of eventual nondecreasingness for quasi-polynomials, a structural dichotomy for the cones C_N, a full vertex description of the nonnegative part of the closure, and a quasipolynomial growth result. The paper is largely self-contained, provides detailed proofs, and includes several worked examples and explicit SageMath-generated generating functions. These are substantial contributions to the study of quasi-polynomial h-vectors.","major_comments":[{"comment":"Lemma 5.4 is false as stated. For d=4, p=3, take v=(1/2,0,1/2,0,0 | 0,1,0,0,0 | 0,1,0,0,0) and w=(0,1,0,0,0 | 0,1,0,0,0 | 0,1,0,0,0). Both lie in S_1 (nonnegative, volume equalities, first-level inequalities) and both have at most two nonzero entries in each mod class, and CM(v)=CM(w)=(1,1,1), yet v≠w. The proof of Lemma 5.4 solves (5.3) only after fixing the support pair (α,ω); CM alone does not determine that support. This false lemma is used in the reverse direction of Theorem 5.3 to conclude w=v from CM(w)=CM(v), and in Proposition 8.3 to construct v_i with the same zero pattern as h. In those particular uses the zero pattern is fixed by the facet equations, so the intended argument is likely salvageable by restating Lemma 5.4 with an explicit support hypothesis, but the current statement and proof are overstrong and the written proof of Theorem 5.3 relies on them.","section":"§5, Lemma 5.4"},{"comment":"Lemma 8.7 is false as stated. For p=2, CM=(1.8,1.2) satisfies (5.1), since -1+1.8=0.8 ≤ 1.2 ≤ 1.8. The lemma prescribes s=p-1=1 and, after deleting integer coordinates, the ordered list (s_1)=(0). Thus g^(1) rounds down coordinate 0 and up coordinate 1, giving (1,2), which violates (5.1) because 2≤1 is false. Rounding down coordinate 1 instead yields (2,1), which satisfies (5.1). Hence the columns g^(i) in Proposition 8.3 are not always in CEND, so the decomposition argument in Proposition 8.3 has a real gap. For the vertex classification this case may be excluded by Lemma 8.1 when max(h)<1, but the stated lemma and proposition are not correct and need an added hypothesis or a refined case division.","section":"§8, Lemma 8.7 and Prop. 8.3"}],"minor_comments":[{"comment":"The chamber decomposition does not handle the case a_r=d (or b=d) where A^(a_r+1)_r or B^(b+1) is not defined. If all A^(ℓ)_r vanish for ℓ=0,...,d, the intended cell should be included as an all-vanishing relative-open cone. This is a small but necessary clarification for the union decomposition used later in Section 7.","section":"§3, Cor. 3.2"},{"comment":"In the proof, the function gnd(p,d;V) is written instead of gnd(d,p;V). This is a typo, but it should be corrected.","section":"§7, Thm. 7.2 proof"},{"comment":"In the d=3, p=2 rational generating function, the expression contains an extra parenthesis: '... + 3t + 1))/' appears to have one unmatched ')'.","section":"§7, Example 7.4"},{"comment":"The proof would be clearer if it explicitly states that the support (α,ω) is assumed fixed when solving the two equations; as written, the uniqueness assertion is the source of the error described in the major comments.","section":"§5, Lemma 5.4 proof"}],"recommendation":"major_revision","confidential_remarks":"The paper is ambitious and the main classification theorems appear likely to be correct, but the current written proof of Theorem 5.3 depends on two false lemmas (Lemma 5.4 and Lemma 8.7). Both are local and likely repairable: Lemma 5.4 should be restated with an explicit support condition, and Lemma 8.7/Proposition 8.3 need a hypothesis that excludes the rounding-order failure (e.g., using Lemma 8.1 when max(h)<1). I recommend major revision rather than rejection because the central framework is sound and the errors are fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main takeaway: the paper is worth taking seriously. It gives a finite moment-inequality criterion for a quasipolynomial to be eventually nondecreasing, proves a sharp polyhedrality dichotomy for the cones C_N, classifies the vertices of the nonnegative slice, and shows the enumeration is a quasipolynomial. The h-vector encoding is natural and the proofs are mostly detailed and honest.\n\nThe genuine soft spot is Lemma 5.4, which is false as stated. For d=4,p=3, v=(0.5,0,0.5,0,0 | 0,1,0,0,0 | 0,1,0,0,0) and w=(0,1,0,0,0 | 0,1,0,0,0 | 0,1,0,0,0) both lie in S1, both have at most two nonzero entries per mod class, and CM(v)=CM(w)=(1,1,1), yet v≠w. The proof implicitly assumes the support pair (α,ω) is known. That is not a fatal problem: every actual use in Theorem 5.3 and Proposition 8.3 fixes the zero pattern, so the main classification is recoverable by rephrasing the lemma. But the written theorem relies on an overstrong false statement and needs revision.\n\nLemma 8.7 has a milder version of the same problem: the prescribed rounding order is not always valid. For CM=(1.8,1.2) with p=2 it produces (1,2), violating (5.1). Again, the cases where this happens have max(h)<1 and are already covered by Lemma 8.1, so the gap is repairable rather than structural. The worry about Lemma 4.4 that got flagged does not land: the total-sum condition exactly supplies the boundary value h_{p-1,d}, so the cumulative-sum construction is coherent.\n\nMinor presentation issues: Corollary 3.2 doesn't explicitly handle the edge case a_r=d where all A^(ℓ) vanish, and the proof of Theorem 7.2 swaps d and p in one expression. These are typos.\n\nOverall: the central characterization and the main structural results are sound and new. The paper deserves a serious referee; I'd recommend accept after the lemmas are fixed. I'd bring it to a reading group for the h-vector crowd.","headline":"Strong, mostly correct paper with a false-but-repairable Lemma 5.4; the main results survive with revisions.","tokens_in":24812,"tokens_out":4869,"would_cite":true,"duration_ms":47731,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper characterizes eventually nondecreasing quasi-polynomials by a finite list of moment-difference sign checks on their h-vectors, and describes the polyhedral geometry of this class.","keywords":["eventually nondecreasing","quasi-polynomial","h-vector","moment difference","polyhedral cone","center of mass","Ehrhart theory","growth rate"],"falsifier":"Take the explicit h-vector from Example 4.3 and evaluate L_h(n) for n around 18 in exact arithmetic: the theorem predicts a single descent at n=18 and no others; any additional descent would disprove the realizability construction. Equivalently, search over all h-vectors with d=2, p=3 and compare the membership tests for C_0, C_1,... computed by brute force with the polyhedral description of Proposition 4.2; a mismatch would falsify the finite description.","tokens_in":23857,"feed_emoji":"📉","tokens_out":6178,"duration_ms":51777,"temperature":0.7,"pith_summary":"The paper asks which quasi-polynomials—functions that are polynomials on each residue class modulo p—eventually stop decreasing. Working with the standard h-vector encoding of a quasi-polynomial, the authors prove a complete characterization: a quasi-polynomial is eventually nondecreasing if and only if, for each pair of adjacent residue classes, the first moment difference between their h-vectors that does not vanish has the correct sign. This reduces an infinite system of inequalities to finitely many checks. The authors then show the closure of the set of all such quasi-polynomials is a polyhedral cone, classify the rays of its nonnegative part, and prove that the count of nonnegative integer h-vectors of a given volume is itself a quasi-polynomial of degree pd.","feed_headline":"Moment gaps decide if a quasi-polynomial ever stops decreasing","feed_subtitle":"One finite check per residue class replaces an infinite list of inequalities with a sign test.","key_machinery":"The central object is the h-vector h=(h_0,…,h_{p(d+1)}) of a quasi-polynomial, defined by the generating function Σ L_h(n) z^n = h(z)/(1−z^p)^{d+1}. The key identity expands L_h(tp+r) in the binomial basis binom{t+d−j}{d}; the coefficients are moments Σ_j (−j)^ℓ h_{r,j} (with (−j−1)^ℓ for the last residue). The difference L_h(n+1)−L_h(n) therefore has each constituent's leading coefficient proportional to the first nonvanishing moment difference A_r^(ℓ)(h) or B^(ℓ)(h) between adjacent residue classes. This identity converts the infinite family of inequalities L_h(n+1)≥L_h(n) into finitely many sign checks, and it is the reason the closure CEND is polyhedral.","core_discovery":"The central result (Theorem 3.1) states that an h-vector h belongs to the eventually nondecreasing cone C_END exactly when, for each residue r between 0 and p−2, either all moment differences A_r^(ℓ)(h)=Σ_j(−j)^ℓ h_{r+1,j} − Σ_j(−j)^ℓ h_{r,j} vanish, or the first nonvanishing one is positive, and similarly for the wrap-around quantity B^(ℓ)(h) involving residues p−1 and 0. These moment differences control the leading coefficients of the constituents of L_h(n+1)−L_h(n), so the condition says exactly that each constituent is eventually nonnegative. A direct corollary is that the closure CEND is the polyhedral cone defined by the volume equalities (the total sums of adjacent residue classes agr","pith_inferences":["The moment-difference condition is a kind of stochastic monotonicity test: if one treats each residue class of the h-vector as a probability distribution (after normalizing volume), the condition says the first moment that distinguishes two adjacent classes must be ordered in a fixed direction. This suggests the result may transfer to problems about comparing distributions (e.g., stochastic domina","The conservation of volume across residue classes is a strong constraint; I suspect it corresponds to a known invariant in Ehrhart theory, and the first-level inequalities might encode a mean-width condition for the associated lattice polytope, which could give a geometric reinterpretation of eventual monotonicity.","The non-polyhedrality for d≥3 implies that any algorithm deciding C_N membership for a fixed N must inspect infinitely many inequalities, which may explain why no simple local monotonicity criterion was known before; a testable consequence is that the cone C_N has infinitely many facets, which one could verify by computing the polar cone for small d=3, p=2.","A concrete extension would be to compute the moment differences for Ehrhart series of rational polytopes and check whether the sign pattern correlates with known monotonicity/unimodality results for Ehrhart h*-vectors; if so, Theorem 3.1 would give a new unified proof of such results."],"forward_implications":["For fixed d and p, checking whether a quasi-polynomial is eventually nondecreasing requires only computing the first d+1 moment differences for each residue—a finite linear algebra calculation.","Optimization over the eventually nondecreasing cone (e.g., extremal h-vectors) becomes a linear program, since the closure is polyhedral with explicit inequalities.","For degree 1 and 2 the cone C_N (nondecreasing from step N) is polyhedral, so eventual nondecreasing can be certified by checking one period's worth of values; for degree ≥3 it cannot, as single decreases may be placed arbitrarily far out.","The nonnegative part of the closure has vertices with an explicit combinatorial form—each residue class is either a unit mass or two masses straddling an integer center-of-mass—so all ray generators are known.","The number of nonnegative integer h-vectors of volume V in the eventually nondecreasing cone grows as a quasi-polynomial of degree pd, meaning the enumeration has a closed quasi-polynomial form."],"fun_headline_variants":["Finite moment checks decide if a quasi-polynomial eventually stops dipping","Moment signs tell when quasi-polynomials stop decreasing forever","Residue-class moment test for eventually nondecreasing quasi-polynomials","A sign test per residue class ends the infinite decrease check","When do quasi-polynomials cease to fall? Moment gaps give the answer"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The characterization assumes the h-vector encoding of quasi-polynomials is complete and faithful—that every degree-d, period-p quasi-polynomial corresponds to exactly one such vector—and the non-polyhedrality claim additionally relies on the realizability lemma that any zero-sum family of difference polynomials can be realized by an h-vector.","fun_headline_variants_meta":{"raw":{"variants":["Finite moment checks decide if a quasi-polynomial eventually stops dipping","Moment signs tell when quasi-polynomials stop decreasing forever","Residue-class moment test for eventually nondecreasing quasi-polynomials","A sign test per residue class ends the infinite decrease check","When do quasi-polynomials cease to fall? Moment gaps give the answer"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00013,"raw_usage":{"total_tokens":958,"prompt_tokens":733,"completion_tokens":225,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":133}},"tokens_in":477,"tokens_out":225,"duration_ms":3213,"temperature":1.0,"reasoning_tokens":133,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T13:05:35.741711+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the explicit h-vector from Example 4.3 and evaluate L_h(n) for n around 18 in exact arithmetic: the theorem predicts a single descent at n=18 and no others; any additional descent would disprove the realizability construction. Equivalently, search over all h-vectors with d=2, p=3 and compare the membership tests for C_0, C_1,... computed by brute force with the polyhedral description of Proposition 4.2; a mismatch would falsify the finite description.","supporting_citations":[],"review_version":1}