{"id":"7892cb41-443f-4a26-85de-a9f271861d80","arxiv_id":"2509.05526","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"For GL_n x GL_{n-1}-Shtukas over F_q, the sigma-isotypic Rankin-Selberg cycle self-intersection equals an r-th derivative of the normalized Rankin-Selberg L-function, provided one uses the author's constructed isotypic subspace.","lead":"This paper proves that certain 'fake' special cycle classes from earlier work coincide with genuine geometric special cycles on Shtukas, via categorical trace. It applies this to express self-intersection numbers of Rankin-Selberg cycles as higher derivatives of Rankin-Selberg L-functions over function fields.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.7 replaces the spectral-action isotypic subspace (1.21) by the ad hoc injection (1.22) without proving equality, so the computed self-intersection is not established to be that of the true isotypic part.","rationale":"The reader's weakest_assumption identifies precisely the same primary load-bearing gap: the unproved equality between the spectral-action isotypic subspace (1.21) and the constructed image of (1.22). I agree that this is the most important issue for the central claim, since Theorem 1.7 is explicitly formulated only after replacing (1.21) by (1.22). The paper's own §1.1.7 concedes the equality is not shown, so the main theorem does not yet confirm Conjecture 1.6 as originally stated. The reader also flagged Lemma 5.16's dependence on [LW25, Conjecture 4.45] and Assumption 4.46; I regard this as a second, compounding dependency because it affects the proof even of the modified statement. However, the paper is transparent about its conditionality, and no internal inconsistency or circular reasoning is evident. The verdict CONDITIONAL therefore remains appropriate; my read does not change it.","tokens_in":960,"tokens_out":881,"duration_ms":55125,"concrete_test":"Verify the equality (1.21)=(1.22) by checking that H^*ξ_{σ,ϵ} is a map of modules for the excursion algebra Γ(Loc^{arith}_{GL_n×GL_{n-1}},O): for every excursion operator f, show f∘H^*ξ_{σ,ϵ} = H^*ξ_{σ,ϵ}∘(f·id). Both sides are perfect complexes, and by [GR25] the spectral-action isotypic part has the same rank as H^r(C^I,(σ_n⊠σ_{n-1})^ϵ); equivariance plus injectivity would then force equality of images. If this commutativity cannot be derived from [GR25], Theorem 1.7 must be restated as a theorem about the image of (1.22) only. Independently, re-derive the commutator relation in Lemma 5.16 from [LW25, Corollary 6.12] without invoking Conjecture 4.45, or supply a proof of that conjecture for the specific Rankin–Selberg correspondences used.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 1.7, which states that Conjectures 1.5 and 1.6 hold after replacing the spectral-action isotypic subspace (1.21) with the image of H^*ξ_{σ,ϵ} (1.22). The paper explicitly does not prove that these subspaces coincide; §1.1.7 defers this to [GR25]. Consequently, the self-intersection number in (1.17) is not shown to be the self-intersection of the σ-isotypic part defined by the excursion-operator action. If the images differ, the theorem proves a statement about an auxiliary subspace, not the conjecture as formulated. This gap is load-bearing because the advertised application—higher Gross–Zagier for the isotypic part—requires equality. A second unproved input compounds the issue: Lemma 5.16, used in Proposition 5.14 and hence in Theorem 6.12 and the pairing computation (6.32), invokes [LW25, Conjecture 4.45] and Assumption 4.46 without proof. If that conjecture fails, the formula for the restricted intersection pairing (6.20) and the final intersection number would be unsupported even for the modified subspace. The main theorem is thus conditional twice over: once for identification with (1.21), and once for the commutator relation underlying the diagonal-cycle computation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a categorical-trace interpretation of special cycle classes on Shtukas and uses it to relate the isotypic part of Rankin–Selberg cycles on GL_n × GL_{n-1}-Shtukas to derivatives of Rankin–Selberg L-functions. The main theorem (Theorem 1.7) states that, after replacing the spectral-action isotypic subspace (1.21) by the image of the constructed map H^*ξ_{σ,ϵ} (1.22), and assuming σ_n, σ_{n-1} are geometrically irreducible and p > n, Conjectures 1.5 and 1.6 hold. The proof combines: (i) a new compatibility theorem for geometric trace with push-forward along maps admitting contracting boundaries (Theorem 2.27); (ii) the identification of special cycle classes with geometric traces of cohomological correspondences (Theorems 3.2, 3.3); (iii) a comparison of geometric trace with categorical trace (Theorem 4.9); and (iv) the fake-cycle computations of [LW25]. The paper explicitly states in §1.1.7 that equality of (1.22) with (1.21) is not shown and is deferred to [GR25], and Lemma 5.16 relies on [LW25, Conjecture 4.45] and Assumption 4.46.","tokens_in":67119,"tokens_out":3543,"duration_ms":39342,"significance":"If the main theorem holds as stated for the true spectral-action isotypic part, the paper would establish a function-field higher Gross–Zagier formula for GL_n × GL_{n-1}, a significant step beyond the rank-one cases treated by Yun–Zhang. The technical framework is valuable: the paper gives a detailed theory of cohomological correspondences with kernels, a new compatibility result for geometric trace and contracting boundaries, and a careful reduction of the intersection number to the fake-cycle identities of [LW25]. The author is admirably explicit about the conditionalities of the argument, flagging the unproved equality (1.21)=(1.22) and the dependence on [LW25, Conjecture 4.45]. These strengths are real: the proof is not circular, the main formula is not fitted to the answer, and the comparison of fake and geometric cycles is a substantive theorem. However, the two advertised gaps are load-bearing, and the paper as it stands proves a conditional statement about an auxiliary subspace rather than the original conjecture.","major_comments":[{"comment":"Theorem 1.7 replaces the spectral-action isotypic subspace (1.21) by the image of the map H^*ξ_{σ,ϵ} in (1.22), but the paper does not prove that these two subspaces coincide. The author states in §1.1.7 that this equality is not shown and is deferred to [GR25]. This is load-bearing: the advertised application—a higher Gross–Zagier formula for the σ-isotypic part defined by the excursion-operator action—requires equality of (1.22) with (1.21). If the images differ, the self-intersection number computed in (1.17) is that of an auxiliary subspace, not the true isotypic part. The theorem should either include a proof of the equality, or be reformulated explicitly as a result about the image of (1.22), with the relationship to Conjectures 1.5 and 1.6 clearly marked as conditional on [GR25].","section":"§1.1.7, Eq. (1.22)"},{"comment":"Lemma 5.16 is used in Proposition 5.14, which in turn feeds into Theorem 6.12 and the pairing formula (6.20)/(6.32). Its proof invokes [LW25, Conjecture 4.45] and Assumption 4.46, and then asserts that the condition g(C)≠1 'can be easily removed' because one leg is fixed. No proof of this removal is supplied. Since the automorphic commutator relation is not established in this paper, the restricted intersection pairing (6.20) and the final intersection number in Theorem 1.7 are conditional on an unproved conjecture. This is a second load-bearing dependency. The author should either prove the needed case of [LW25, Conjecture 4.45], or clearly state in Theorem 1.7 that the result is conditional on it.","section":"§5.5, Lemma 5.16"},{"comment":"The conversion from the intersection pairing ⟨−,−⟩ to the bilinear form ω_{M^{⊗r}} is a crucial step in the final computation. Equation (6.32) is obtained by combining Proposition 6.11 and Theorem 6.12. Both of these rely on the previously noted unproved identification (1.21)=(1.22) and on the commutator relation of Lemma 5.16. Consequently, the displayed equality (6.32) is not an unconditional theorem. The proof of Theorem 1.7 should make explicit which parts of the argument are proved in the paper and which are imported as conjectural inputs from [LW25] and [GR25].","section":"§6.4, Eq. (6.32)"}],"minor_comments":[{"comment":"The exposition is generally careful, but the paper is long and several notational conventions (e.g., the switch between H^*ξ_{σ,ϵ}, ξ^e_{σ,I}, and ξ^{d}_{σ,I}) could be unified. A table of the main maps and their definitions would help readers.","section":"General"},{"comment":"The phrase 'the integration 1 of f' appears to contain a typo; it likely should read 'the integration of f'.","section":"§1.1.2"},{"comment":"In the proof of Lemma 2.35, the sentence 'the quotient stack X•/TX is representable by a scheme, which is clearly normal and of finite type' appears to conflate two assertions; it should be split and the finite-type claim justified before it is used.","section":"§2.3.1"},{"comment":"The statement 'One can easily remove the condition g(C)≠1' in Lemma 5.16 is unsupported and should be either proved or deleted; as written it is a dangling claim that undermines the reader's confidence in that lemma.","section":"§5.5.3"},{"comment":"The bibliography is appropriate, but several results from [LW25] and [GR25] are imported without precise theorem numbers in the text (e.g., the exact statement of [LW25, Conjecture 4.45]). Adding specific references would improve verifiability.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is substantial and the main idea—identifying fake cycle classes with restrictions of geometric special cycles via categorical trace—is sound and interesting. However, the two explicitly advertised gaps (the equality of (1.22) with (1.21) and the dependence on [LW25, Conjecture 4.45]) are not cosmetic; they directly affect the truth of the main theorem as a statement about the σ-isotypic part. I would recommend major revision rather than rejection, because the gaps are clearly identified and the author gives a credible route to filling them, but the current manuscript does not yet establish the advertised theorem unconditionally."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a conditional result, not the full conjecture. Theorem 1.7 as stated replaces the spectral-action isotypic subspace (1.21) with an injected subspace (1.22), and the equality is explicitly deferred to [GR25]. The paper also relies, in Lemma 5.16, on [LW25, Conjecture 4.45]. Those are the two load-bearing gaps; neither is hidden. If you want to know whether the advertised higher Gross–Zagier is proven: no, not yet. If you want a careful formal framework and a clear conditional statement: yes.\n\nWhat is genuinely useful: the categorical trace interpretation of special cycle classes (Sections 4–5), the identification of fake cycles with geometric traces after isotypic restriction (Proposition 5.8), and the push-forward compatibility for contracting boundaries (Theorem 2.27). This last piece looks like a real technical contribution, not a formality. The paper is carefully organized, and the limitations are stated plainly.\n\nSoft spots: The first gap is structural. Conjectures 1.5 and 1.6 are about the σ-isotypic part defined by spectral action, while Theorem 1.7 proves a statement about the image of H*ξ_{σ,ε}. The author says in §1.1.7 that equality should follow from [GR25], but does not prove it. If the images differ, the computed self-intersection is not for the isotypic part named in the conjecture. The second gap is technical: Lemma 5.16 is used to prove Proposition 5.14, which drives Theorem 6.12 and equation (6.20). Its proof invokes [LW25, Conjecture 4.45] and Assumption 4.46, and the sentence “one can easily remove” the g(C)=1 condition is not a proof. These are real dependencies, not cosmetic.\n\nProportion: the reader's soundness 4 and CONDITIONAL verdict are about right. I do not see circularity or fitted parameters. The reliance on a companion preprint is legitimate, but it makes the current theorem conditional on that conjecture. The math is dense but written carefully.\n\nBottom line: send to referees. The topic is important, the conditional results are clearly valuable, and the gaps can in principle be fixed or explicitly certified. A referee should focus on the (1.21)=(1.22) identification and on the status of [LW25, Conjecture 4.45]. I would cite this paper once those dependencies are resolved; even now, the categorical trace framework is worth quoting.","headline":"Conditional higher Gross–Zagier for GL_n x GL_{n-1} Shtukas, honestly stated but with two load-bearing gaps: the (1.21)/(1.22) subspace identification and the borrowed [LW25, Conjecture 4.45].","tokens_in":67549,"tokens_out":3288,"would_cite":true,"duration_ms":31248,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F67","11G40","14D24"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that, with geometrically irreducible local systems and p > n, the summed self-intersection of the σ-isotypic part of Rankin–Selberg cycles on GL_n × GL_{n−1}-shtukas equals q^{dim Bun_{GL_{n−1}}}(ln q)^{−r−2} times the r-th","keywords":["special cycles on shtukas","categorical trace","Rankin–Selberg L-functions","higher Gross–Zagier formula","geometric Langlands","cohomological correspondences","intersection numbers","isotypic parts"],"falsifier":"Take the smallest nontrivial case — n=2, r=1, C an elliptic curve over F_q, σ_1 the trivial local system, and σ_2 a rank-2 geometrically irreducible local system with explicitly computable L-function — and compute both sides of (1.17): the self-intersection number from the construction (1.22) on the left and the derivative/residue formula on the right; a mismatch would refute Theorem 1.7, and checking whether image(1.22) is invariant under the excursion operators in this example would directly test the deferred identification with the spectral-action subspace.","tokens_in":66608,"feed_emoji":"🧮","tokens_out":14050,"duration_ms":122721,"temperature":0.7,"pith_summary":"Special cycle classes on shtukas — moduli spaces of G-bundles over a curve with Frobenius and point modifications — can be represented as traces of correspondences, both geometrically (Borel–Moore classes of a fixed-point stack) and categorically (traces of Hecke endofunctors). The paper proves that the 'fake' special cycle classes previously attached to a Hecke eigensheaf L_σ through period integrals are in fact the genuine special cycle classes restricted to the σ-isotypic part of shtuka cohomology. As an application, for GL_n × GL_{n−1} with geometrically irreducible local systems σ_n, σ_{n−1} and p > n, the summed self-intersection number of the σ-isotypic part of the Rankin–Selberg cycles equals an explicit formula in the r-th derivative of the normalized Rankin–Selberg L-function divided by residues of the two adjoint L-functions. This is a direct higher-dimensional generalization of the GL_2 higher L-derivative formula, and it confirms Conjectures 1.5 and 1.6 once the σ-isotypic part is taken to be the image of the constructed map (1.22), an identification deferred to the geometric Langlands theorem [GR25].","feed_headline":"Shtuka cycle self-pairings equal higher L-derivative values","feed_subtitle":"GL_2 higher L-derivative formula now extends to GL_n × GL_{n-1}: cycle self-pairings encode Rankin–Selberg derivatives.","key_machinery":"The argument turns on the categorical trace formalism: for a dualizable category C and an endofunctor F, the trace tr(F, C) is an object that can represent a space of interest. Here, shtuka cohomology is exhibited as the categorical trace of the Hecke operator (Frob×id)_! ∘ T_{V^I} acting on ShvNilp(Bun_G) ⊗ QLisse(C^I), via the LT_Serre isomorphism of [AGK+22a]. To connect this to geometry, the paper develops cohomological correspondences with kernels and their geometric trace on the shtuka fixed-point stack, proving a new push-forward compatibility (Theorem 2.27) for maps with a contracting boundary — needed because the diagonal Bun_H → Bun_G is not proper, and a Drinfeld-style compactific","core_discovery":"The central claim is Theorem 1.7: take σ_n, σ_{n−1} geometrically irreducible Weil local systems of ranks n and n−1 over a smooth projective curve C over F_q, set σ = (σ_n, σ_{n−1}), and assume p > n. After replacing the spectral-action σ-isotypic subspace (1.21) of the compact-support cohomology of GL_n × GL_{n−1}-shtukas by the image of the explicitly constructed injection H^r(C^I, (σ_n ⊗ σ_{n−1})^ϵ) → H^{2(n−1)r}_c(Sht_{GL_n×GL_{n−1},(Stdn⊠Std_{n−1})^ϵ}) (equation 1.22), the paper proves that Conjecture 1.5 and Conjecture 1.6 hold. In particular, the summed self-intersection number of the σ-isotypic parts of the Rankin–Selberg cycles equals q^{dim Bun_{GL_{n−1}}} (ln q)^{−r−2} (d/ds)^r|_{","pith_inferences":["If, as expected, [GR25] indeed proves image(1.22) = subspace(1.21), then Theorem 1.7 upgrades to a proof of the original Conjecture 1.6 without the substitution; until then, the numerical content of the formula depends on a non-canonical choice of what counts as the σ-isotypic part.","The same trace-theoretic mechanism — cohomological correspondences with kernels plus a contracting-boundary push-forward theorem — should produce higher-derivative formulas for other spherical pairs H ⊂ G satisfying Assumption 2.36, e.g., symmetric varieties, yielding analogous L-function identities for other Langlands products.","The non-degeneracy result (Corollary 6.13) suggests that isotypic parts of shtuka cohomology are finite-dimensional and self-dual in a robust way; this could be useful in defining intersection-theoretic invariants for non-compact cycles more broadly.","Testing whether the image of (1.22) is invariant under the full excursion-operator algebra would give a concrete, low-complexity check of the deferred identification with the spectral-action subspace, independent of the full geometric Langlands theorem."],"forward_implications":["For geometrically irreducible σ_n, σ_{n−1} with p > n, the r-th central derivative of the normalized Rankin–Selberg L-function is read off from the self-intersection of Rankin–Selberg cycles on GL_n × GL_{n−1}-shtukas, giving higher L-derivative formulas for all n rather than just GL_2.","The intersection pairing on the σ-isotypic part of compact-support cohomology of shtukas is non-degenerate (Conjecture 1.5), so a canonical self-intersection number exists even though the special cycles themselves are not compact.","The 'fake' special cycle classes of [LW25] are demystified: they are the isotypic parts of the actual special cycle classes, confirming the expectation in that earlier work.","The categorical-trace framework yields a uniform treatment of two kinds of special cycles — minuscule homogeneous ones (Rankin–Selberg) and diagonal ones (intersection pairings) — through a single commutative diagram (1.25).","Theorem 1.7 reduces Conjecture 1.6 to the equality between the image of (1.22) and the spectral-action subspace (1.21), making the remaining gap a precise, checkable statement in geometric Langlands."],"supporting_citations":[{"why":"Supplies the categorical-trace interpretation of shtuka cohomology and the LT_Serre isomorphism that Theorem 4.9 relies on.","marker":"[AGK+22a]"},{"why":"Provides the category ShvNilp(Bun_G), excursion operators, the spectral action, and QLisse(C^I), the framework in which Conjectures 1.2, 1.5, and 1.6 are formulated.","marker":"[AGK+22c]"},{"why":"Geometric Langlands in positive characteristic; load-bearing for Assumption 1.1, for compactness of Hecke eigensheaves (Proposition 6.1), and for the unproved identification of (1.22) with (1.21).","marker":"[GR25]"},{"why":"Defines fake special cycle classes, the Clifford algebra action and Kolyvagin system, and supplies Theorem 6.5 and the commutator relations used in Section 6; also the source of Lemma 5.16's unproved Conjecture 4.45 and Assumption 4.46.","marker":"[LW25]"},{"why":"The GL_2 formula for higher L-derivatives that Theorem 1.7 directly generalizes, and the model for self-intersections of isotypic parts of special cycles on shtukas.","marker":"[YZ17]"},{"why":"Constructs Hecke eigensheaves for geometrically irreducible GL_n-local systems, used to define L_σ and the isotypic-part map ξ_{σ,ϵ}.","marker":"[FGV02]"},{"why":"Proves the finiteness / closedness of special cycles for shtukas, used to define the pushforward π_{Sht,I,!} of the Rankin–Selberg cycles.","marker":"[Yun22]"}],"fun_headline_variants":["Shtuka cycle self-pairings capture L-derivative values","Self-pairings on Shtuka cycles equal L-function derivatives","Special cycle self-intersections match L-derivative values","GL_n Shtuka cycles encode Rankin-Selberg derivatives","Higher L-derivatives from Shtuka special cycle pairings"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The main formula is proven for the image of the explicitly constructed map (1.22), but the paper does not prove that this image equals the σ-isotypic subspace (1.21) cut out by the spectral action, deferring that identification to the geometric Langlands theorem [GR25]; a key algebraic step (Lemma 5.16) also relies on an unproved conjecture from [LW25].","fun_headline_variants_meta":{"raw":{"variants":["Shtuka cycle self-pairings capture L-derivative values","Self-pairings on Shtuka cycles equal L-function derivatives","Special cycle self-intersections match L-derivative values","GL_n Shtuka cycles encode Rankin-Selberg derivatives","Higher L-derivatives from Shtuka special cycle pairings"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000958,"raw_usage":{"total_tokens":3911,"prompt_tokens":728,"completion_tokens":3183,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":3094}},"tokens_in":472,"tokens_out":3183,"duration_ms":21186,"temperature":1.0,"reasoning_tokens":3094,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:23:39.188386+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the smallest nontrivial case — n=2, r=1, C an elliptic curve over F_q, σ_1 the trivial local system, and σ_2 a rank-2 geometrically irreducible local system with explicitly computable L-function — and compute both sides of (1.17): the self-intersection number from the construction (1.22) on the left and the derivative/residue formula on the right; a mismatch would refute Theorem 1.7, and checking whether image(1.22) is invariant under the excursion operators in this example would directly test the deferred identification with the spectral-action subspace.","supporting_citations":[],"review_version":1}