{"id":"9ceae891-060f-4d85-b116-ab0833c3af16","arxiv_id":"2412.04976","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"New power-saving bounds are proven for all generalized Kloosterman sums on GL_n, using explicit parametrizations and the Weil bound.","lead":"This paper proves new upper bounds on generalized Kloosterman sums for the matrix group GL_n, exponential sums used throughout analytic number theory. The bounds beat the previous trivial estimates for every admissible permutation, and improve a known result for the longest permutation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The power-saving exponent 1/(4l(w)) rests entirely on the unproved combinatorial inequality (8); the B_k construction and m=0 replacement rules are too under-specified to verify.","rationale":"The reader's weakest_assumption pinpoints the same combinatorial claim in §4.2, and I agree that it is the most load-bearing part of the proof. The claimed power saving is precisely the content of (8): the size of the Kloosterman set is |Cw(m)| ≈ p^{Σ(j-i+1)m_{i,j}}, and Lemma 4.2 gives a bound of |Cw(m)|^{1+ε} p^{1/4 Σ b*}. Inequality (8) is what turns the b* sum into a negative multiple of the exponent of |Cw(m)|, yielding the 1/(4l(w)) saving. If the combinatorial inequality is false or merely unproven, the central theorem reduces to a trivial or much weaker bound. The proof in the paper is a single paragraph with no induction details; the B_k construction is described in prose and the m=0 replacement rules are recursive and context-dependent. This is a genuine proof gap, not a stylistic issue. I also considered the character-dependency passage at the end of §4.2, where the proof says the loss is at most max_j max(|ψ_j|^{-1/2}, |ψ'_j|^{-1/2})^{l(w)/2}, while Theorem 1 states C^{l(w)/2} with C = max_j min(|ψ_j|^{-1/2}, p^{r_j/2}); this discrepancy is secondary because it affects only the implied constant, not the power-saving exponent, and it may be repairable. The recommended verdict remains CONDITIONAL: the paper presents a credible strategy and substantial technical work, but the central combinatorial lemma must be either fully proved or machine-checked before the main theorem can be accepted. No change to the reader's verdict is needed.","tokens_in":29810,"tokens_out":10789,"duration_ms":219273,"concrete_test":"Implement the §4.2 construction literally (fixing the ambiguities by the most natural reading) for all admissible Weyl elements of GL_N with 2≤N≤7. For each w, enumerate all exponent vectors m with entries in {0,1,2} and all zero patterns, construct the B_k sets, and verify inequalities (6), (7), and (8) exactly as stated. If any violation or non-termination occurs, the exponent 1/(4l(w)) in Theorem 1 is not established; if the check passes for all small cases, it would at least show the combinatorial claim is plausible, though the missing induction proof would still need to be supplied.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1's saving of 1/(4l(w)) follows from Lemma 4.2 together with inequality (8): l(w)·Σ b*_{i,j} ≤ -Σ (j-i+1)m_{i,j}. Without (8), the final exponent in Theorem 1 is unsupported. The proof given in §4.2 is a short heuristic paragraph. The sets B_k are defined by a recursive 'furthest to the right' procedure whose output for a general multi-block Weyl element is not uniquely pinned down; the multiplicities m_k are said to yield 'at most l(w)' occurrences of each b_{i,j}, but the promised induction over blocks is not carried out. The m_{i,j}=0 replacement rules on p.34 have three cases, each with a 'continue as if b_{i,j} was added' instruction, and it is not demonstrated that the procedure terminates or preserves the inequalities (6)–(7). Because these inequalities are exactly what converts the Weil-bound savings into a power saving, a failure or even a hidden ambiguity in this combinatorial step would break the central claim. This is an internal gap, not a disagreement with consensus, and no formal verification or exhaustive check is provided.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an explicit parametrization of local Kloosterman sums for GL_{N+1} for all admissible Weyl elements, building on the Dąbrowski–Reeder stratification. Two-block Weyl elements are treated by an explicit Bruhat decomposition encoded in path diagrams, and general Weyl elements are handled by an induction that splits off the top block. The resulting exponential sums are bounded by repeated applications of the Weil bound together with a combinatorial inequality (8), yielding Theorem 1 (power saving 1/(4l(w))) and Theorem 2 (a variant with a milder character dependency), including a version for the congruence subgroup Γ0(q).","tokens_in":29908,"tokens_out":14327,"duration_ms":151238,"significance":"If the combinatorial inequality (8) is fully established, Theorem 1 supplies the first non-trivial bounds for many admissible Weyl elements and improves the long-element bound of Blomer–Man by a factor of 4. The parametrization itself (Theorem 3 and Corollary 2) is a concrete and potentially reusable tool. The paper is largely self-contained relative to the cited stratification and Weil-bound benchmarks, and the main derivation is not circular. However, the advertised power saving rests on an unproved and under-specified combinatorial claim, and the uniform character dependence in Theorem 1 also needs qualification. No machine-checked proofs or exhaustive verifications are provided.","major_comments":[{"comment":"Inequality (8) is the load-bearing step that converts the Weil-bound savings into the exponent 1/(4l(w)) in Theorem 1. The proof given for the first inequality in (8) is the assertion that each b_{i,j} appears at most l(w) times, followed by 'The maximal sum over these multiplicities can be computed inductively over the number of blocks and is exactly l(w)'. The induction is not carried out, and the sets B_k are only described informally. Since any weakening or hidden multiplicity in this step would change the final exponent, a complete definition of B_k for arbitrary multi-block diagrams and a full proof of (8) are required.","section":"§4.2, Eq. (8)"},{"comment":"The three replacement rules on p. 34 for handling m_{i,j}=0 are recursive and contain instructions such as 'continue as if b_{i,j} was added'. It is not demonstrated that this procedure terminates for all diagrams, nor is it proved that inequalities (6) and (7) survive the replacement. The final claim that the new sets B_k still satisfy both inequalities in (8) is asserted rather than proved. This is an essential part of the argument because zero exponents occur generically when the exponent vector r has zero coordinates.","section":"§4.2, m_{i,j}=0 replacement rules"},{"comment":"The statement 'improves the trivial bound by a power saving of 1/(4l(w))' is not uniform in the characters. For a character with |ψ_j|^{-1/2}_p = p^{r_j/2} (for example a character of conductor p^{r_j}), one has C = p^{r_j/2}; then C^{l(w)/2} already contributes p^{l(w)r_j/4}. When only one r_j is nonzero and l(w)>4, the right-hand side of Theorem 1 exceeds the trivial size p^{Σ r_k}. The authors should either state the character regime in which the displayed bound actually beats the Dąbrowski–Reeder trivial bound or explicitly compare the full expression (including C) with the trivial bound.","section":"Theorem 1 and Remark 2"}],"minor_comments":[{"comment":"In Lemma 4.1, the sentence 'It can maybe be removed' about the factor ∏ e_i is not a mathematical statement; because the factor is later absorbed into ε, this is harmless, but the remark should be phrased more precisely or deleted.","section":"Remark 15"},{"comment":"The text says 'Theorem 3 claims C = 1/l(w)', but Theorem 3 is the parametrization; the relevant claim is inequality (8) in Section 4.2, not Theorem 3. This misreference should be corrected.","section":"§4.2, paragraph after Eq. (5)"},{"comment":"The expression '1/2(N−1)' is ambiguous; it should read '1/(2(N−1))' or be parenthesized accordingly.","section":"Remark 16"},{"comment":"In the displayed formula for L_{2,5}, the exponent '−m22,−m2,3−m2,4' appears to be a typo for '−m_{2,2}−m_{2,3}−m_{2,4}'.","section":"Section 3.2, example after Theorem 4"},{"comment":"The symbol C_w(m) is defined once just before Theorem 3 and again in Theorem 5, with slightly different conventions for the range of c_{i,j}. The two definitions should be explicitly reconciled.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The paper is a serious contribution and the central strategy is credible, but the incomplete proof of inequality (8) is a genuine obstruction: the advertised power saving is not established without it. I would not reject on the current evidence, since the gap may be fillable, but a revision must contain a complete, termination-guaranteed construction of the sets B_k and a full verification of (8). The character-dependency issue in Theorem 1 should also be clarified. If (8) turns out to fail, the main theorem may reduce to a bound weaker than the trivial one in the relevant ranges."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe headline is simple: this paper claims the first power-saving bound for every admissible Weyl element of GL_n, and the claim is plausible. The parametrization via Dabrowski-Reeder stratification is explicit, the Weil-bound application is structured, and the main theorem's exponent 1/(4l(w)) is derived, not fitted. Prior work covered only special elements (long, w*, GL3, GL4, Sp4, orthogonal), so the breadth here is genuinely new. The induction over blocks in Section 3 is the right idea, and Theorem 4 gives real explicit Bruhat-decomposition formulas that look like they hold.\n\nWhere it is soft, in proportion: the proof of Lemma 4.1 is only sketched and carries a product-of-e_i factor that the author says 'can maybe be removed' but then absorbs into epsilon. That is a minor gap, since the absorption is legitimate even without removal. The bigger soft spot is Section 4.2. Inequality (8) is the load-bearing combinatorial step: it converts the Weil-bound savings into the final power saving. The proof there is a heuristic paragraph. The sets B_k are defined by a recursive 'furthest to the right' procedure with three replacement rules for m_{i,j}=0, and the termination/uniqueness of that procedure is not demonstrated. The claimed bound 'each b_{i,j} appears at most l(w) times' is asserted with a sketch, not a proof. The m=0 replacement rules on p.34 are under-specified; it is not clear that they preserve inequalities (6) and (7). If (8) fails, Theorem 1's exponent breaks. Nothing in my reading suggests it actually fails—the example and the one-block computation are consistent—but the proof as written is not checkable without extra work.\n\nThe reader's stress-test note flagged exactly this, and I agree with the diagnosis. I do not agree with any suggestion that the whole paper is hollow: the parametrization theorem and the two-block theorem are substantial, and the combinatorial step is isolated. I also note the paper is formally self-contained against external benchmarks: no circularity, no fitted exponents, no invented entities. The citation pattern is appropriate, including use of [BM24] and [DR98].\n\nMy bottom line: this deserves a serious referee and likely a conditional acceptance after the combinatorial section is expanded. The paper is for analytic number theorists working on relative trace formulae and density theorems. I would bring it to a reading group, and I would cite it if the combinatorial lemma survives scrutiny. Recommend sending it to peer review; do not desk reject.\n\nBest,\n[You]","headline":"A serious, credible preprint that plausibly proves power-saving bounds for all admissible Weyl elements on GL_n, but the pivotal combinatorial inequality (8) is under-verified and needs a real referee.","tokens_in":30567,"tokens_out":659,"would_cite":true,"duration_ms":9914,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11L05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that every admissible Weyl-element Kloosterman sum on GL_n admits a power saving of size 1/(4l(w)) over the trivial bound.","keywords":["Kloosterman sums","GL_n","Weyl elements","exponential sums","Weil bound","Bruhat decomposition","p-adic groups","relative trace formula"],"falsifier":"Take the five-block Weyl element $w_2$ in Section 4.2, enumerate all admissible $m$ for a small prime $p$, compute the sets $B_k$ explicitly, and check inequality (8): any tuple with $l(w)\\sum b^*_{i,j} > -\\sum (j-i+1)m_{i,j}$ would destroy the $p^{1/(4l(w))}$ saving and falsify Theorem 1.","tokens_in":29484,"feed_emoji":"📐","tokens_out":9814,"duration_ms":94153,"temperature":0.7,"pith_summary":"The paper establishes a power-saving upper bound for the local Kloosterman sums of GL_{N+1} attached to any admissible Weyl element, not just the special cases treated before. For a Weyl element w of length l(w), the sum is shown to be at most $C^{l(w)/2}(\\prod_{k=1}^N p^{r_k})^{1-1/(4l(w))+\\varepsilon}$, where the $r_k$ encode the diagonal modulus and $C$ measures the conductors of the characters. This improves the trivial bound by a power saving of $1/(4l(w))$; for all but three Weyl elements it is the first non-trivial bound, and for the long Weyl element it quadruples the previous saving. The same bound holds for the congruence subgroup $\\Gamma_0(q)$ with $q$ a power of $p$. The reason to care is that these sums appear on the geometric side of relative trace formulae and in Fourier coefficients of automorphic forms, where bounds control the size of the terms.","feed_headline":"Power saving for all Weyl-element Kloosterman sums on GL(n)","feed_subtitle":"New parametrisation plus repeated Weil bounds beats the trivial bound by p^{1/(4L)} for every admissible Weyl element of length L.","key_machinery":"The load-bearing objects are, first, the explicit parametrization of the Kloosterman sum by the sets $C_w(m)$ of exponent coordinates and, second, a directed diagram whose vertices are the roots in $R(w^{-1})$ and whose edges carry the coordinates; the diagram is augmented with dotted edges for the second character and is used both to write the exponential sum (Corollary 2) and to define the grouping of summation variables. The grouping into two families of non-adjacent variables is what allows several Weil bounds to be applied at once, and an induction over the number of blocks in the Weyl element reduces every admissible $w$ to two-block elements.","core_discovery":"The central claim, Theorem 1, is that for an admissible block-diagonal Weyl element $w$ in $\\mathrm{GL}_{N+1}$ and a modulus with exponent vector $r$, the Kloosterman sum satisfies $\\mathrm{Kl}_p(\\psi,\\psi',n)\\ll_\\varepsilon C^{l(w)/2}(\\prod_{k=1}^N p^{r_k})^{1-1/(4l(w))+\\varepsilon}$ with $C=\\max_j \\min(|\\psi_j|_p^{-1/2}, p^{r_j/2})$, and the identical bound holds for $\\Gamma_0(q)$ with $q$ a power of $p$. The proof is constructive: the sum is parametrized explicitly as an exponential sum over a product of congruence classes (Theorem 3), using the Bruhat decomposition of the group elements $b_\\alpha(a)$. A diagram attached to the Weyl element encodes the $p$-adic powers in the summands, and its vertices are two-coloured so that disjoint groups of variables can be summed one after another; each single-variable sum is a $\\mathrm{GL}_2$ Kloosterman sum to which the Weil bound applies. A lemma controlling the average dependence on neighbouring variables gives the stated saving.","pith_inferences":["The genuine bottleneck is the combinatorial inequality (8); if it fails, the clean exponent $1/(4l(w))$ would need to be replaced by a smaller saving, but the parametrization and two-colouring scheme would still give some power saving.","The paper's example with all $m_{i,j}$ equal suggests that the method cannot reach savings better than $1/(2(N-1))$ for the long element; a plausible next step is to compute the exact saving for small $N$ to see whether $1/O(N)$ is the right order.","The diagram language is group-independent in structure, so a parallel treatment for other reductive groups with the same kind of stratification and a Weil bound would be a natural test of the method's scope."],"forward_implications":["For every admissible Weyl element other than the identity, $w^*$, and the long element, Theorem 1 is the first non-trivial bound; applications to trace formulae no longer have to single out these cases.","The bound applies uniformly to $\\mathrm{GL}_{N+1}(\\mathbb{Z}_p)$ and to $\\Gamma_0(q)$, $q$ a power of $p$, so the same saving is available in congruence-subgroup settings used in density theorems.","Theorem 2 gives an alternative with a much lighter dependence on the characters, replacing $C^{l(w)/2}$ by $C$ at the cost of a saving of $1/(2Nl(w))$ instead of $1/(4l(w))$.","Since the proof uses only the stratification of the Kloosterman set and the Weil bound, the same statements hold over any non-archimedean local field, not only $\\mathbb{Q}_p$."],"supporting_citations":[{"why":"Supplies the stratification of the Kloosterman set and the exact size of the quotient used to set up the parametrisation.","marker":"[DR98]"},{"why":"Introduces the parametrisation-plus-Weil-bound strategy and the lemma that the paper generalises; the long-element saving here is four times that of this paper.","marker":"[BM24]"},{"why":"Provides the $\\mathrm{GL}_2$ Weil bound applied to each separated variable in the exponential sum.","marker":"[Wei48]"},{"why":"Establishes which Weyl elements are admissible and identifies the special element with hyper-Kloosterman sums, fixing the class of $w$ considered.","marker":"[Fri87]"},{"why":"Defines the generalized Kloosterman sums in the Chevalley group setting, giving the object studied.","marker":"[Dą93]"}],"fun_headline_variants":["Explicit parametrisation beats trivial bound for all GL(n) Kloosterman sums","Weil bound gives power saving for all GL(n) Kloosterman sums","Parametrising Kloosterman sums with Bruhat decomposition beats trivial bound","All Weyl-element Kloosterman sums on GL(n) get power saving","New proof beats Dabrowski-Reeder bound for all GL(n) sums"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The bound collapses if the combinatorial grouping of exponent coordinates in Section 4.2 fails to satisfy inequality (8), a step that is asserted with only a sketch and no worked induction.","fun_headline_variants_meta":{"raw":{"variants":["Explicit parametrisation beats trivial bound for all GL(n) Kloosterman sums","Weil bound gives power saving for all GL(n) Kloosterman sums","Parametrising Kloosterman sums with Bruhat decomposition beats trivial bound","All Weyl-element Kloosterman sums on GL(n) get power saving","New proof beats Dabrowski-Reeder bound for all GL(n) sums"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000994,"raw_usage":{"total_tokens":4170,"prompt_tokens":861,"completion_tokens":3309,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":3207}},"tokens_in":477,"tokens_out":3309,"duration_ms":25519,"temperature":1.0,"reasoning_tokens":3207,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T21:01:39.502302+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the five-block Weyl element $w_2$ in Section 4.2, enumerate all admissible $m$ for a small prime $p$, compute the sets $B_k$ explicitly, and check inequality (8): any tuple with $l(w)\\sum b^*_{i,j} > -\\sum (j-i+1)m_{i,j}$ would destroy the $p^{1/(4l(w))}$ saving and falsify Theorem 1.","supporting_citations":[],"review_version":1}