{"id":"3c87eed5-a1d7-454e-988d-d36d8fc2d4d0","arxiv_id":"2412.16308","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The height of intersections of two hypersurfaces twisted by torsion points on a toric variety converges to an adelic sum of mixed integrals of roof and Ronkin functions.","lead":"This paper proves a formula for the typical height of the intersection of two polynomial equations on a toric variety, confirming a conjecture for the 2-dimensional case. It shows that after randomly twisting the equations by roots of unity, the height converges to a computable convex-geometric expression.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5.2's adelic-vanishing proof uses a false bound (5.5): max φ(ord χ^{m-m'}(ωℓ)) ≥ ord(ωℓ)^{c1} fails when the support of f spans a proper sublattice, leaving the convergence argument incomplete.","rationale":"The reader's verdict accepted the paper at moderate confidence and located the weakest assumption in the non-Archimedean extension of the auxiliary function in Section 3. My stress-test pass found a different, more concrete issue: Theorem 5.2, the adelic vanishing result that is indispensable for Theorem B, proves its first summand by using the false inequality (5.5). The counterexample with f = g = 1 + x1 + x2 in G_m^3 and a strict sequence whose order is dominated by the x3-coordinate shows that the maximum φ-value over characters from the support is controlled by a proper sublattice, not by the full order of the torsion point. The proof's use of dℓ = ord(ωℓ) is therefore not legitimate, and the convergence of the first summand is not established as written. The underlying statement of Theorem B is supported by the independent proof in [DHS24], so I do not recommend rejection; the appropriate action is a conditional acceptance requiring the adelic estimate to be repaired, for example by replacing dℓ with the order of the projection to the sublattice spanned by supp(f)−supp(f) and rederiving the corresponding bound. This is a genuine technical gap in a load-bearing step, not a disagreement with the consensus that the theorem is true.","tokens_in":41359,"tokens_out":32066,"duration_ms":298531,"concrete_test":"Work out Theorem 5.2's first-summand estimate for K = Q, T = G_m^3, f = g = 1 + x1 + x2, and the strict torsion sequence ωℓ = (ζ_{pℓ}, ζ_{pℓ}^2, ζ_{qℓ}) with pℓ, qℓ distinct primes and log qℓ ≥ pℓ. Compute the left side of (5.5): the relevant characters are e1, e2, e1−e2, so max φ ≈ pℓ, while dℓ = pℓ qℓ; verify that (5.5) fails for every c1 ≥ 1. Then check whether the proof's bound c2 log dℓ / dℓ^{c1} tends to 0; it does not when qℓ is chosen with log qℓ comparable to pℓ. Finally, verify that the actual sum over Aℓ = {pℓ} still tends to 0, so the defect is in the proof's overestimate rather than in the theorem's statement.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 5.2, which is essential for Theorem B, contains a concrete false inequality. In the first summand, the authors assert there is c1 ≥ 1 depending only on supp(f) such that (5.5) max_{m,m'} φ(ord(χ^{m-m'}(ωℓ))) ≥ dℓ^{c1}, where dℓ = ord(ωℓ) in T(K). This is not true when the differences m−m' for m,m' in supp(f) generate a proper sublattice of M. Example: K = Q, T = G_m^3, and f = g = 1 + x1 + x2, which is absolutely irreducible and non-binomial. Its support differences are e1, e2, e1−e2, all lying in the x1-x2 sublattice. Let pℓ and qℓ be distinct primes with qℓ >> pℓ, and take ωℓ = (ζ_{pℓ}, ζ_{pℓ}^2, ζ_{qℓ}). This sequence is strict: for every nonzero character (a,b,c), the value ζ_{pℓ}^{a+2b} ζ_{qℓ}^{c} is eventually different from 1, since for c=0 the exponent a+2b is fixed and nonzero, while for c≠0 the qℓ-part cannot cancel the pℓ-part for large pℓ. For this sequence, max_{m,m'} φ(ord(χ^{m-m'}(ωℓ))) = φ(pℓ) ≈ pℓ, but dℓ = pℓ qℓ, so (5.5) fails for any c1 > 0. The subsequent bound c2 log dℓ / dℓ^{c1} therefore does not follow, and the proof's control of the first summand is not established. The conclusion may still be true, but the written argument needs a different measure of strictness, e.g. the order of the projection of ωℓ to the subtorus generated by supp(f)−supp(f), rather than the full order in T(K).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem B: for two nonzero Laurent polynomials f and g over a number field K, a complete toric variety X compactifying a split torus T, and a quasi-strict sequence of torsion points (ωℓ) in T(K)^2, the height of the intersection cycle ZT(ωℓ,1^* f, ωℓ,2^* g) converges to an adelic sum of mixed integrals of roof functions and Legendre–Fenchel duals of Ronkin functions. The proof combines toric intersection theory, arithmetic Bézout, an extension of an auxiliary function to the analytic torus with logarithmic singularities (Section 3), local logarithmic equidistribution theorems (Section 4), and an adelic vanishing statement (Section 5). The paper also contains a detailed reduction appendix and applications to average heights over torsion sets.","tokens_in":41754,"tokens_out":7399,"duration_ms":71495,"significance":"If the proof is completed, the result establishes a nontrivial two-codimensional case of the authors' conjecture and gives an arithmetic analogue of the Bernstein–Kushnirenko–Khovanskii theorem for typical intersections. The paper is well structured and makes use of deep external results (Stoll, Tate–Voloch, Dimitrov–Habegger) in a modular way, with most reduction steps moved to the appendix. The main technical concern is a specific false inequality in Section 5 that is used to prove the adelic vanishing; this gap appears fixable, but it is load-bearing for the proof of Theorem B.","major_comments":[{"comment":"The inequality (5.5) is false when the differences m−m' for m,m' in supp(f) generate a proper sublattice of M. For example, take K=Q, T=G_m^3, and f=g=1+x1+x2, which is absolutely irreducible and non-binomial. Let pℓ and qℓ be distinct primes with qℓ >> pℓ, and set ωℓ=(ζ_{pℓ}, ζ_{pℓ}^{ℓ}, ζ_{qℓ}). This sequence is strict: for every nonzero character (a,b,c), the value ζ_{pℓ}^{a+bℓ} ζ_{qℓ}^{c} is eventually different from 1, since for c≠0 the qℓ-component cannot cancel the pℓ-component once qℓ>|c|, while for c=0 the relation a+bℓ=0 holds for at most one ℓ. However dℓ=ord(ωℓ)=pℓ qℓ, whereas every character χ^{m−m'} with m,m'∈supp(f) involves only x1 and x2, so max_{m,m'} φ(ord(χ^{m−m'}(ωℓ))) = φ(pℓ) ≈ pℓ. This contradicts (5.5) for any c1≥1. Consequently the bound c2 log dℓ / dℓ^{c1} used to prove that the first summand in Theorem 5.2 tends to zero does not follow. The proof needs to replace dℓ by a quantity that measures the order of the projection of ωℓ to the subtorus generated by supp(f)−supp(f), or equivalently the lcm of the finitely many orders ord(χ^{m−m'}(ωℓ)), rather than the full order in T(K).","section":"Section 5.B, Eq. (5.5)"}],"minor_comments":[{"comment":"In the sentence 'we need to show prove that there exists a finite subset S ⊂ M', the word 'prove' appears to be a typographical error and should be deleted.","section":"Introduction, p. 4"},{"comment":"In the phrase 'for all ℓ such that the strictness degree δ(ωℓ) si sufficiently large', 'si' should read 'is'.","section":"Section 4.A, proof of Theorem 4.1"},{"comment":"The final sentence 'the statement follows from Lemma 5.1' appears to cite the wrong lemma: Lemma 5.1 only establishes finiteness of S, whereas the displayed norm formula is proved in Lemma 5.4. The reference should be corrected.","section":"Section 5.A, proof of Proposition 5.3"},{"comment":"The notation 'MIZ2(0∆, ρ∨_{f,v}, ρ∨_{g,v})' is not defined in the paper; if it denotes the mixed integral for n=2 with the indicator of the standard simplex, this should be stated explicitly.","section":"Example 6.3"}],"recommendation":"major_revision","confidential_remarks":"The flaw in Section 5.B is real and load-bearing, but I do not think it is fatal: the convergence claim can likely be recovered by replacing dℓ with the lcm of the orders of the finitely many characters attached to supp(f)−supp(f), and the remainder of the proof structure appears sound. I recommend major revision rather than rejection, contingent on the authors supplying a corrected adelic vanishing argument."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main result here, the k=2 case of Conjecture A, was already proven in full generality by Destic–Hultberg–Szachniewicz, and the authors say so honestly. So the novelty is the method, not the statement. The method is worth taking seriously: reducing the height to a local error term, extending an auxiliary function with logarithmic singularities, and then using logarithmic equidistribution of torsion points is a genuinely different route from the globally valued fields approach. The local vanishing theorem (Theorem 4.4) and the explicit formula at good places (Proposition 5.3) are carefully done, and Appendix A collects useful reduction steps.\n\nBut there is a concrete problem in the proof of Theorem 5.2, the adelic vanishing. Inequality (5.5) asserts that max over support differences of phi(ord(chi^{m-m'}(omega_l))) grows at least like d_l^{c1} for some c1 depending only on supp(f). That is false when the differences m-m' generate a proper sublattice of M. The example from the stress test works: take T = G_m^3, f = g = 1 + x_1 + x_2, and omega_l = (zeta_{p_l}, zeta_{p_l}^2, zeta_{q_l}) with q_l >> p_l. The sequence is strict because the q_l-coordinate prevents any character from vanishing eventually, but every character coming from a difference in supp(f) has order dividing p_l, so the left side is phi(p_l) ~ p_l while d_l = p_l q_l. No positive c1 makes p_l >= (p_l q_l)^{c1}. The bound c2 log d_l / d_l^{c1} does not follow, and the first summand in the proof of Theorem 5.2 is not controlled.\n\nThis is load-bearing: Theorem 5.2 is essential for Theorem B. The conclusion may still be true, and the fix is likely to measure strictness through the order of the projection of omega_l to the subtorus generated by supp(f)-supp(f) rather than the full order d_l, but that is not what is written. The reader's take missed this. I would not accept the paper in its current form; it needs a corrected argument and a careful check of whether similar issues affect the second summand.\n\nWho gets value from this? Specialists in Arakelov geometry and toric varieties, especially those interested in quantitative versions of the limit or in the local equidistribution technique. It deserves a serious referee once the gap is fixed, and possibly also now as a way to pressure-test the method.","headline":"The proof of Theorem B has a real gap at inequality (5.5), but the strategy and local results are solid; the paper needs revision before acceptance.","tokens_in":42294,"tokens_out":2736,"would_cite":false,"duration_ms":26867,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14G40","11G50","14M25","52A39"],"pacs":[],"model":"deepseek-v4-flash","headline":"For two Laurent polynomials on a toric variety, the height of the intersection of their torsion-twisted hypersurfaces converges to an adelic sum of mixed integrals of roof functions and duals of Ronkin functions.","keywords":["toric variety","height of a variety","complete intersection","strict sequence of torsion points","Ronkin function","mixed integral","Arakelov geometry","logarithmic equidistribution"],"falsifier":"For a fixed number field and a pair of Laurent polynomials $f,g$, compute the Galois-orbit average of the local function $I_v$ along a strict sequence of torsion points at a non-Archimedean place where the reduction of $f$ or $g$ is reducible; if that average does not tend to 0, or the explicit norm identity in Lemma 5.4 fails, then Theorem 4.4 and hence Theorem B would be false.","tokens_in":41135,"feed_emoji":"🧮","tokens_out":9326,"duration_ms":78074,"temperature":0.7,"pith_summary":"The paper proves a limit formula for the height of the intersection of two hypersurfaces in a toric variety: twisting the defining Laurent polynomials by a quasi-strict sequence of torsion points, the height of the resulting 2-codimensional cycle converges to an adelic sum of mixed integrals of roof functions and convex duals of Ronkin functions. This is the k = 2 case of the authors' Conjecture A, which predicts that the typical arithmetic size of a complete intersection in a toric variety is determined entirely by convex-analytic data attached to the defining polynomials. The result matters because it extends the known height formulas for toric varieties and hypersurfaces to the first genuinely higher-codimensional situation, and it confirms a previously open conjecture for bivariate Fermat polynomials. The proof proceeds by decomposing the height into a well-understood hypersurface term plus local error terms, then showing each local error vanishes via logarithmic equidistribution of torsion points and a global adelic vanishing argument.","feed_headline":"Two twisted hypersurfaces: height has convex-analytic limit","feed_subtitle":"On toric varieties, the typical height of a 2-codimensional complete intersection equals an adelic sum of mixed integrals.","key_machinery":"The central object is the auxiliary function\n$$F_v(t) = \\int_{$X_v^{{\\mathrm{an}}$}} \\log |t^*g|_v \\, c_1(\\overline{D}_{0,v}) \\wedge \\dots \\wedge c_1(\\overline{D}_{n-2,v}) \\wedge \\delta_{$Z_v^{{\\mathrm{an}}$}},$$\nwhere $Z$ is the hypersurface defined by $f$. Theorem 3.4 shows that $F_v$ extends to the whole analytic torus with at most logarithmic singularities along the closed subset where $t^*g$ is proportional to $f$: a function with at most logarithmic singularities is continuous outside that subset and bounded below near it by a constant times the logarithm of the maximum of the defining equations. At Archimedean places this extension uses a continuity theorem for fiber integrals, while at non-Archimedean places it uses a formal-model description of the Monge\\,--Amp\\`ere measure as a weighted sum over irreducible components of the special fiber. Combined with the Archimedean logarithmic equidistribution theorem for torsion points and the non-Archimedean theorem on linear forms in roots of unity, this makes each local error term tend to zero; a separate adelic argument using the Poisson formula for the sparse resultant bounds the sum over all but finitely many places.","core_discovery":"The central claim (Theorem 6.2) is that for nonzero Laurent polynomials $f,g \\in K[M]$ and any quasi-strict sequence of torsion points $(\\omega_\\ell)_\\ell$ in $T(K)^2$, the height satisfies\n$$\\lim_{\\ell\\to\\infty} h_{D_0,\\dots,D_{n-2}}\\bigl(Z_T(\\omega_{\\ell,1}^* f,\\omega_{\\ell,2}^* g)\\bigr) = \\sum_{v\\in M} n_v \\, \\mathrm{MI}_M\\bigl(\\vartheta_{D_0,v},\\dots,\\vartheta_{D_{n-2},v},\\rho_{f,v}^\\vee,\\rho_{g,v}^\\vee\\bigr).$$\nThe left side is the height of the intersection cycle of two translated hypersurfaces inside a complete toric variety, and the right side is an adelic sum of mixed integrals of the $v$-adic roof functions of the metrized divisors and the convex duals of the $v$-adic Ronkin functions of $f$ and $g$. The theorem is proved by reducing to smooth projective toric data with very ample divisors and algebraic or smooth metrics, applying the arithmetic B\\'ezout theorem to separate a hypersurface height, and then proving that the remaining local integrals vanish both at each place, by logarithmic equidistribution, and away from a finite set of bad places, by an adelic argument.","pith_inferences":["The local extension theorem for the auxiliary function $F_v$ is the main technical bottleneck; the same strategy could in principle prove Conjecture A for $k>2$ once such an extension is available for $k$ polynomials.","Quantitative versions of the logarithmic equidistribution theorems used in the proof should yield explicit convergence rates for the height of twists of bounded torsion order, extending the special-case estimates the authors cite.","The limit formula suggests a convex-geometric picture of arithmetic complexity for complete intersections, analogous to the way sparse-resultant geometry computes typical degrees by mixed volumes; testing it on families with more than two polynomials or on non-split tori would be a natural next step."],"forward_implications":["For any two nonzero Laurent polynomials over a number field, the typical height of the intersection of their torsion-twisted hypersurfaces in a complete toric variety is explicitly computable from the Newton polytopes and the Ronkin data of $f$ and $g$.","The result settles the previously open case $k=2$ of Conjecture A and, for bivariate Fermat polynomials of arbitrary degrees, confirms the authors' earlier conjecture on limit heights.","Because the proof passes through reduction steps that allow arbitrary semipositive toric metrics to be approximated by smooth or algebraic ones, the formula holds for the full class of semipositive toric metrized divisors.","Averaging over strict sequences of finite sets of torsion points, the same limit describes the typical height and shows that almost all twists have height within any positive tolerance of the limit."],"supporting_citations":[{"why":"Supplies the height formula for toric varieties as adelic sums of mixed integrals of roof functions and the convex-analytic vocabulary used throughout.","marker":"[BPS14]"},{"why":"Gives the hypersurface height formula and the definition of v-adic Ronkin functions and their convex duals, which form the right-hand side of the limit formula.","marker":"[Gua18b]"},{"why":"Provides the Archimedean logarithmic equidistribution of Galois orbits of torsion points used to show each Archimedean local error term vanishes.","marker":"[DH24]"},{"why":"Provides the non-Archimedean theorem on linear forms in roots of unity used to show each p-adic local error term vanishes at the Gauss point.","marker":"[TV96]"},{"why":"Poisson formula for the sparse resultant, used to bound the auxiliary function and to prove the adelic vanishing outside the finite set of bad places.","marker":"[DS15]"},{"why":"Continuity of fiber integrals, used to prove the Archimedean part of the extension theorem for the auxiliary function.","marker":"[Sto67]"},{"why":"Non-Archimedean analytic geometry, used to define and study the extension of the auxiliary function and the peaked product.","marker":"[Ber90]"}],"fun_headline_variants":["Height of toric intersections: exact limit via adelic sums","Two twisted hypersurfaces: heights converge to mixed integrals","Toric heights: a limit formula for complete intersections","Adelic sums predict heights of toric intersections"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument stands on the claim that, at every place, the local integral function $F_v$ extends from torsion points to the whole analytic torus with only logarithmic singularities; if that extension broke down at even one place, the local error terms would not vanish and the limit formula would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Height of toric intersections: exact limit via adelic sums","Two twisted hypersurfaces: heights converge to mixed integrals","Toric heights: a limit formula for complete intersections","Adelic sums predict heights of toric intersections"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000713,"raw_usage":{"total_tokens":3213,"prompt_tokens":959,"completion_tokens":2254,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":575,"completion_tokens_details":{"reasoning_tokens":2189}},"tokens_in":575,"tokens_out":2254,"duration_ms":14087,"temperature":1.0,"reasoning_tokens":2189,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:43:07.468331+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed number field and a pair of Laurent polynomials $f,g$, compute the Galois-orbit average of the local function $I_v$ along a strict sequence of torsion points at a non-Archimedean place where the reduction of $f$ or $g$ is reducible; if that average does not tend to 0, or the explicit norm identity in Lemma 5.4 fails, then Theorem 4.4 and hence Theorem B would be false.","supporting_citations":[],"review_version":1}