{"id":"b2ddb0bd-999c-445b-be2e-ce0d6937316c","arxiv_id":"2608.00470","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A new arithmetic D-module realizes reductive-group hypergeometric exponential sums as Frobenius traces and yields an explicit q'^(d/2) bound with a Weyl-chamber volume constant.","lead":"This paper builds a p-adic differential equation object whose Frobenius traces are exponential sums attached to Laurent polynomials on reductive algebraic groups, then uses it to prove an explicit size bound for those sums. The result gives a p-adic counterpart to the authors' earlier l-adic construction and extends the p-adic hypergeometric method from tori to reductive groups.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 1.11 proves a degree-independent q^{d/2} bound; for k'/k of degree m>1 this must be reconciled with Theorem 0.2's q'^{d/2}=q^{md/2}.","rationale":"The central claim is the explicit q'^{d/2} bound in Theorem 0.2. The only place the extension degree m enters the final inequality is Prop 1.10 and the weight estimate in §1.11. As typeset, those two steps use incompatible normalizations: Prop 1.10 introduces a factor q^{-1}, and the weight bound also uses q, so the resulting estimate for the exponential sum is q^{d/2}, not q'^{d/2}. This is not a cosmetic issue: for m>1 it would assert a bound independent of the number q^m of terms, which is stronger than the known Weil/Deligne bounds for Kloosterman-type sums on G_m (d=1). The geometric Assumptions 2.1/4.3 are stated hypotheses, and the import from [FL] concerns the explicit constant; both matter, but the normalization mismatch directly breaks the derivation of the main quantitative claim as written. The reader's rationale flags the same q-versus-q' issue, although the reader's formal weakest_assumption is the compactification hypothesis, hence partial agreement. The issue is plausibly fixable by consistently using q' in the trace formula and in the weight bound for extensions, so the verdict remains CONDITIONAL rather than moving to REJECT.","tokens_in":43928,"tokens_out":35397,"duration_ms":321685,"concrete_test":"Recompute the trace formula for the Dwork module L_π at a k'-point y of degree m: evaluate Tr(F^m,i_y^+L_π) directly from [M, Theorem 4.4] and the Frobenius structure in §1.2. If the result is -q^{-m}θ_m(y) rather than -q^{-1}θ_m(y), then Prop 1.10 must involve q'^{-1}, and the inequality in §1.11 must use q^{m(d-2)/2}; rerunning the argument with these corrections should yield exactly q'^{d/2} d!∫.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Prop 1.10 states Tr(F^m,i_A^+Hyp_{π,!}) = (-1)^{d+n} q^{-1} S_A for k'/k of degree m. Section 1.11 then bounds |Tr(F^m,i_A^+Hyp_{π,!})| ≤ q^{(d-2)/2} rank, using that H^{-n}(i_A^*Hyp_{π,!}(n)[n]) has weight ≤ d-2, and concludes |S_A| ≤ q^{d/2} d!∫. This bound is independent of m, while Theorem 0.2 claims q'^{d/2}=q^{md/2}. For m>1 the written proof would establish a much stronger estimate; e.g. for d=1 it gives O(q^{1/2}) for exponential sums over G(k') with |k'|=q^m, contradicting the expected Deligne/Weil size q^{m/2}. The inconsistency must be resolved either by replacing the q^{-1} in Prop 1.10 with q'^{-1}=q^{-m}, or by using the degree-aware weight bound |α| ≤ q^{mw/2} for the m-th Frobenius iterate, or both. As written, the chain Prop 1.10 → §1.11 does not literally establish Theorem 0.2 for non-trivial extensions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an arithmetic-D-module approach to exponential sums of the form S_A = Σ_{g∈G(k')} ψ(Tr_{k'/k}(Σ_j Tr(A_j ρ_j(g)))), where G is a split reductive group, ρ_j are representations, and A = (A_j) is a tuple of endomorphisms. The authors define arithmetic hypergeometric D-modules Hyp_{π,+} and Hyp_{π,!} on V = ∏_j End(V_j), identify them as Fourier transforms of ι_+O^†_G and ι_!O^†_G, prove overholonomicity under a quasi-finiteness assumption, and show that over the nondegenerate locus V^gen they are overconvergent F-isocrystals whose rank is bounded by d! times an explicit Weyl-chamber integral. From this they derive Theorem 0.2, an explicit Deligne-style bound |S_A| ≤ q'^{d/2} d! ∫_{Δ_∞∩C} ∏_{α∈R_+} (λ,H_α)^2/(ρ,H_α)^2 dλ. The proof follows the Berthelot–Abe–Caro formalism and relies on the existence of a good integral equivariant compactification (Assumptions 2.1 and 4.3).","tokens_in":44161,"tokens_out":10826,"duration_ms":100809,"significance":"If correct, the paper provides a systematic p-adic framework for hypergeometric exponential sums on reductive groups and gives a parameter-free, explicit upper bound with a geometric constant. The construction is substantial: it defines Frobenius-structured arithmetic D-modules, proves their overconvergence away from the degenerate locus, and reduces the exponential-sum estimate to a rank computation plus Abe–Caro weights. The main derivation is detailed and does not use fitted parameters or normalization tricks. The principal weakness is a normalization mismatch between the trace formula and the weight estimate when the extension degree m = [k':k] exceeds 1; as written, the proof does not literally establish the stated q'^{d/2} bound. This issue is local and apparently correctable, so the underlying approach remains credible.","major_comments":[{"comment":"The chain from Proposition 1.10 to Theorem 0.2 does not prove the stated bound for nontrivial extensions. Proposition 1.10 asserts Tr(F^m,i_A^+Hyp_{π,!}) = (-1)^{d+n} q^{-1} S_A, while Theorem 0.2 claims |S_A| ≤ q'^{d/2} d!∫ = q^{md/2} d!∫ for m=[k':k]. Independently, §1.11 bounds |Tr(F^m,i_A^+Hyp_{π,!})| by q^{(d-2)/2} rank, using that H^{-n}(i_A^+Hyp_{π,!}) has weight ≤ d-2. Since weights are measured relative to the q-Frobenius, the m-th iterate has absolute size q^{m(d-2)/2}, not q^{(d-2)/2}. For m>1 the written proof would yield |S_A| ≤ q^{d/2} d!∫, which is stronger than the expected Deligne/Weil size; e.g. for d=1 it gives O(q^{1/2}) for sums over G(k') with |k'|=q^m, contradicting the expected O(q^{m/2}). The normalization must be repaired: either Proposition 1.10 should have q'^{-1}=q^{-m} and §1.11 should use q^{m(d-2)/2}, or an equivalent degree-aware argument must be supplied. As written, the proof of Theorem 0.2 is incomplete for m>1.","section":"§1.9–1.11, Proposition 1.10 and proof of Theorem 0.2"}],"minor_comments":[{"comment":"There are several typos, e.g. “exponential sum associated this Laurent polynomial” in the abstract and “degernate locus” in the introduction; a copyedit pass is needed.","section":"Abstract and Introduction"},{"comment":"The displayed trace formula omits the inner trace in Σ_j A_j ρ_j(g); the proof in equation (1.10.3) correctly has Tr(A_j ρ_j(g)). The statement should be made consistent with the proof.","section":"Proposition 1.10"},{"comment":"The notation for the volume integrand is inconsistent: Theorem 0.2 writes (λ,H_α)^2/(ρ,H_α)^2 while Proposition 4.22 writes (λ,α)^2/(ρ,α)^2. The co-root/root convention should be fixed throughout.","section":"Theorem 0.2 vs. Proposition 4.22"},{"comment":"“Techm¨uller” is a misspelling of Teichmüller; it occurs in at least two places.","section":"§1.9 and §1.11"},{"comment":"The statement says F_π(N)^{(m)}|_{V^gen} is coherent as an O_{\\hat V}-module, while the proof concludes O_{V_k}-coherence on the special fiber; please clarify the completed/module statement.","section":"Corollary 4.20"}],"recommendation":"major_revision","confidential_remarks":"The central construction appears sound and the extension-degree mismatch in §1.9–1.11 is local and fixable: the authors need to correct the trace-formula normalization to q'^{-1} and the weight bound to q^{m(d-2)/2}, or otherwise reconcile the two. I would be willing to accept after verifying that repair. The dependency on the companion paper [FL] for the volume identity in Proposition 4.22 should be checked by the editor for availability and non-circularity."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth reading: Lei, Xuanyou, and Chenhan construct arithmetic hypergeometric D-modules Hyp_{π,!} and Hyp_{π,+} on V_k whose Frobenius traces match reductive-group hypergeometric sums, and they prove overconvergence on the nondegenerate locus. That is a real step beyond the torus case in [FLWZ] and the l-adic companion [FL]. The rank bound is an explicit Weyl-chamber integral, and the proof is detailed and innocent of fitted parameters.\n\nThe soft spot is real and likely fixable. Proposition 1.10 states that the trace of F^m on i_A^+Hyp_{π,!} is q^{-1} times the exponential sum over G(k') with [k':k]=m. Section 1.11 then bounds that trace by q^{(d-2)/2} times the rank, independent of m, and concludes |S_A| ≤ q^{d/2} rank·integral. For m>1 that is far stronger than the q'^{d/2} bound in Theorem 0.2; for d=1 it would give O(q^{1/2}) for a sum over a group with q^m points, contradicting the expected Weil size q^{m/2}. The fix is almost certainly to replace q^{-1} by q'^{-1}=q^{-m} in Prop 1.10, or to make the weight bound degree-aware, (q^m)^{(d-2)/2}. As written, the chain Prop 1.10 → §1.11 does not establish Theorem 0.2 for nontrivial extensions.\n\nSecond, the rank bound imports [FL, Theorem 3.10] from the companion preprint. That is a parameter-free geometric computation, not a circular use of the target bound, but it is unpublished and should be checked.\n\nThird, the theorem is conditional on Assumptions 2.1/4.3, the existence of a good equivariant compactification over R. The authors note this holds for almost all primes in a Dedekind setup, but as a blanket hypothesis it is real.\n\nAll told, this is not a desk-reject. The construction is novel, the proof is earnest, and the flaw is identifiable and likely repairable. I would send it to a serious referee, expecting a request for revision on the normalization point. I would not cite it in its current form.","headline":"Worth a serious referee: the paper builds new arithmetic hypergeometric D-modules with the right Frobenius traces, but a q vs q' normalization slip leaves the proof of Theorem 0.2 incomplete for nontrivial field extensions.","tokens_in":44743,"tokens_out":6283,"would_cite":false,"duration_ms":53619,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F10","14F30","14M27","11L07"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs an arithmetic hypergeometric D-module whose Frobenius trace is a hypergeometric exponential sum on a reductive group, and proves an explicit exponential-sum bound with a Weyl-chamber constant.","keywords":["arithmetic D-module","hypergeometric exponential sum","reductive group","overconvergent F-isocrystal","Fourier transformation","spherical variety","Weyl chamber","Newton polytope"],"falsifier":"Enumerate all $A=(A_1,\\dots,A_N)\\in\\prod_j \\operatorname{End}(V_j)(\\mathbb{F}_q)$ for a small reductive group such as $G=\\mathrm{GL}_2$ with the standard representation and $q\\in\\{2,3,5,7\\}$, identify the nondegenerate ones via the critical-point criterion (0.1.2), and compare $|\\sum_{g\\in G(\\mathbb{F}_q)} \\psi(\\operatorname{Tr}(A\\rho(g)))|$ with $q^{d/2}d!\\int_{\\Delta_\\infty\\cap C}\\prod_{\\alpha\\in R^+}\\frac{\\lambda(H_\\alpha)^2}{\\rho(H_\\alpha)^2}\\,d\\lambda$; any violation would disprove Theorem 0.2.","tokens_in":43685,"feed_emoji":"","tokens_out":7008,"duration_ms":60238,"temperature":0.7,"pith_summary":"This paper studies exponential sums of the form sum over g in G(k') of psi(sum_j Tr(A_j rho_j(g))), where G is a reductive group over a finite field and the rho_j are representations. The authors construct an arithmetic hypergeometric D-module whose Frobenius trace at A is exactly this sum, so p-adic cohomology can bound it. Their main theorem states that for every nondegenerate A the absolute value is at most q'^(d/2) d! times an explicit integral over the dominant Weyl chamber, a bound of the same shape as classical weight bounds but with a concrete constant. The construction works through Fourier transform, invariant differential operators, and a formal model that turns the module into an overconvergent F-isocrystal on the nondegenerate locus. A reader should care because it gives a uniform, explicit estimate for a broad family of exponential sums attached to reductive groups.","feed_headline":"New D-module gives explicit bounds for group exponential sums","feed_subtitle":"The Frobenius trace of the new module is the character sum; weight theory turns it into an explicit bound.","key_machinery":"The load-bearing object is the arithmetic hypergeometric D-module $\\operatorname{Hyp}_{\\pi,!}=\\pi_{2,!}f^+L_\\pi[d+n-1]$, where $L_\\pi$ is the Dwork overconvergent $F$-isocrystal on $\\mathbb{A}^1$, $f(g,A)=\\sum_j \\operatorname{Tr}(A_j\\rho_j(g))$, and $\\pi_2$ is the projection $G\\times V\\to V$. Its Frobenius trace at $A$ computes the exponential sum. The proof replaces $\\operatorname{Hyp}_{\\pi,!}$ by a direct summand, the modified hypergeometric D-module $F_\\pi(N)$, whose integral models at finite level $m$ are described by invariant differential operators; the associated graded modules have support controlled by the critical locus of $f_{\\tau,A}$ on $G\\times G$-orbits. The rank estimate comes from the degree of the closure of $G$ in projective space, computed as the displayed Weyl-chamber integral.","core_discovery":"The paper's central discovery is a p-adic object that packages the entire family of hypergeometric exponential sums: an overholonomic arithmetic D-module $\\operatorname{Hyp}_{\\pi,!}$ on the space $V=\\prod_j \\operatorname{End}(V_j)$, with a Frobenius structure, for which $\\operatorname{Tr}(F^m, i_A^+ \\operatorname{Hyp}_{\\pi,!})$ equals $(-1)^{d+n}q^{-1}$ times the sum over $G(k')$. This module is the Fourier transform of the direct image of the structure sheaf from $G$, and a modified version $F_\\pi(N)$ admits an integral model whose associated graded module is supported in the critical locus of the Laurent polynomial. Over the nondegenerate open set $V^{\\mathrm{gen}}$, $\\operatorname{Hyp}_{\\pi,!}$ is an overconvergent $F$-isocrystal of rank at most $d!$ times the Weyl-chamber integral; the p-adic weight theorem then bounds Frobenius traces and yields the main estimate.","pith_inferences":["If the main theorem is correct, the same explicit bound should hold for any quasi-finite representation family once an integral equivariant compactification exists; the geometric Assumptions 2.1 and 4.3 are likely an artifact of the method rather than a genuine restriction on the phenomenon.","For the torus $G=\\mathbb{G}_m^n$, this construction should reproduce the known p-adic GKZ hypergeometric bounds, giving a concrete consistency check of both the rank formula and the Weyl-chamber integral.","The explicit constant $d!\\int_{\\Delta_\\infty\\cap C}\\prod_{\\alpha}\\frac{\\lambda(H_\\alpha)^2}{\\rho(H_\\alpha)^2}\\,d\\lambda$ may be non-optimal; small-case computations could reveal whether the factorial factor can be replaced by a smaller group-theoretic invariant.","A direct computational search over small finite fields for nondegenerate $A$ could test the inequality numerically and calibrate the constant before deeper arithmetic applications."],"forward_implications":["For any nondegenerate $A$, the exponential sum is bounded by $q'^{d/2}\\, d! \\int_{\\Delta_\\infty\\cap C}\\prod_{\\alpha\\in R^+}\\frac{\\lambda(H_\\alpha)^2}{\\rho(H_\\alpha)^2}\\,d\\lambda$.","The full family of hypergeometric sums is controlled by one overholonomic D-module, with cohomology concentrated in degree zero over $V^{\\mathrm{gen}}$.","Over $V^{\\mathrm{gen}}$, $\\operatorname{Hyp}_{\\pi,!}$ becomes an overconvergent $F$-isocrystal whose rank is at most the explicit Weyl-chamber integral times $d!$.","The Fourier-transform identity $\\operatorname{Hyp}_{\\pi,!}\\simeq \\mathcal{F}_\\pi(\\iota_{k!}\\mathcal{O}_G^\\dagger)$ ties the construction to Fourier analysis on the affine space of endomorphisms.","The p-adic weight argument supplies the mixedness needed to convert the rank estimate into an exponential-sum bound.","The construction gives a blueprint for bounding other character sums on reductive groups by similar arithmetic D-modules."],"supporting_citations":[{"why":"Defines the Laurent-polynomial setup and the hypergeometric exponential sums on reductive groups that this paper studies.","marker":"[FL]"},{"why":"Establishes the torus case by p-adic methods; the present paper extends that approach to general reductive groups.","marker":"[FLWZ]"},{"why":"Supplies the six-functor formalism for overholonomic arithmetic D-modules and the p-adic weight theorem used to pass from ranks to character-sum bounds.","marker":"[AC]"},{"why":"Provides the Fourier transform of arithmetic D-modules used to identify $\\operatorname{Hyp}_{\\pi,!}$ with the Fourier transform of $\\iota_{k!}\\mathcal{O}_G^\\dagger$.","marker":"[NH2]"},{"why":"Constructs the overconvergent power series whose special values give the additive character $\\psi$, making the Frobenius trace computable.","marker":"[M]"},{"why":"Defines overholonomicity and the independence of the compactification for overconvergent arithmetic D-modules.","marker":"[Ca2]"},{"why":"Supplies toroidal equivariant compactifications of reductive group embeddings used in Assumptions 2.1 and 4.3.","marker":"[BK]"},{"why":"Sets up arithmetic differential operators of finite level and the sheaves $D^\\dagger$, which form the foundation of the construction.","marker":"[B3]"}],"fun_headline_variants":["New p-adic D-module bounds hypergeometric exponential sums","Frobenius trace equals group sum, weight theory bounds it","Overholonomic D-module yields explicit bounds for character sums","Arithmetic D-module: from Frobenius trace to sum estimates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The bound is conditional on the existence of an integral equivariant compactification $\\tilde{Y}\\to Y$ over $R$ satisfying Assumptions 2.1 and 4.3—in particular an open piece isomorphic to $(U^+\\times U^-)\\times S$ with $S$ a toric scheme over the maximal torus—and if such a model does not exist for a representation family, the formal model, overconvergence, and rank bound do not follow.","fun_headline_variants_meta":{"raw":{"variants":["New p-adic D-module bounds hypergeometric exponential sums","Frobenius trace equals group sum, weight theory bounds it","Overholonomic D-module yields explicit bounds for character sums","Arithmetic D-module: from Frobenius trace to sum estimates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000545,"raw_usage":{"total_tokens":2580,"prompt_tokens":891,"completion_tokens":1689,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":1614}},"tokens_in":507,"tokens_out":1689,"duration_ms":13498,"temperature":1.0,"reasoning_tokens":1614,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:20:34.176435+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate all $A=(A_1,\\dots,A_N)\\in\\prod_j \\operatorname{End}(V_j)(\\mathbb{F}_q)$ for a small reductive group such as $G=\\mathrm{GL}_2$ with the standard representation and $q\\in\\{2,3,5,7\\}$, identify the nondegenerate ones via the critical-point criterion (0.1.2), and compare $|\\sum_{g\\in G(\\mathbb{F}_q)} \\psi(\\operatorname{Tr}(A\\rho(g)))|$ with $q^{d/2}d!\\int_{\\Delta_\\infty\\cap C}\\prod_{\\alpha\\in R^+}\\frac{\\lambda(H_\\alpha)^2}{\\rho(H_\\alpha)^2}\\,d\\lambda$; any violation would disprove Theorem 0.2.","supporting_citations":[],"review_version":2}