{"id":"a66a0305-9f9f-4960-8c0f-c0e40233038a","arxiv_id":"2608.07173","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The higher theta series on Hermitian shtukas are shown to be modular, independent of the chosen Lagrangian, with a stronger supermodularity result for general linear groups.","lead":"This paper proves a conjecture about special counting functions built from curves and vector bundles: the functions do not depend on certain choices of extra subbundles. It also proves a stronger version for general linear groups, and establishes a new structural link between virtual cycle classes and categorical traces.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of Theorem 9.2.3 depends on unproved formal properties of the motivic derived Fourier transform deferred to [Zho]; without them, the core Fourier-duality comparison in Lemma 9.1.4 is unsupported.","rationale":"The reader's weakest_assumption identified the restriction to the trivial-similitude fiber L=O_X as the primary concern. I agree that this is a genuine gap in the written proof: the statement of Theorem 1.1.1 covers all L, while the proof in Section 9 only treats L=O_X and asserts the general case follows 'straightforwardly.' However, a structurally more fundamental concern is the paper's reliance on the unproved motivic derived Fourier transform formalism of Theorem 5.3.1, whose proofs are entirely deferred to the forthcoming work [Zho]. This dependency is load-bearing because the key comparison in Lemma 9.1.4 uses Proposition 6.4.1 and Proposition 8.4.5, both built on Theorem 5.3.1. Even the paper's own Lemma 5.4.3 invokes the Plancherel formula from that theorem. Thus, without [Zho], the central argument does not stand even in the trivial fiber case. I do not see an internal contradiction, and the missing pieces are plausibly fillable, so the appropriate verdict remains CONDITIONAL rather than REJECT. My agreement with the reader is partial because the reader lists [Zho] only as a secondary fragility, whereas I regard it as the principal load-bearing gap.","tokens_in":76707,"tokens_out":5582,"duration_ms":54342,"concrete_test":"Obtain [Zho] (or an independent proof of Theorem 5.3.1) and check that the seven properties hold exactly as stated, especially the Plancherel formula (5.3.9) and the Gysin/forget-supports compatibility (5.3.12). Then re-run the proof of Lemma 5.4.3 and Proposition 8.4.5: if any shift or Tate twist differs, the normalized Gaussian identity (8.4.22) and the comparison (9.1.17) must be re-evaluated. A stronger test is to provide a fully written proof of Theorem 5.3.1(5) for derived vector bundles over derived Artin stacks and verify the cancellation in Lemma 5.4.3. If the Plancherel isomorphism holds as stated, the conditional concern is resolved; if not, Theorem 9.2.3 is in doubt.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 5.3 states Theorem 5.3.1 with the words: \"The proofs of all statements in this subsection are deferred to the forthcoming work of Tong Zhou [Zho].\" These formal properties are not optional commentary: Lemma 9.1.4, the Fourier-duality comparison at the heart of the proof of Theorem 9.1.1, uses Proposition 6.4.1 and Proposition 8.4.5, both of which invoke Theorem 5.3.1. Even the in-paper Lemma 5.4.3 (invertibility of the relative Gauss cohomology) cites the Plancherel isomorphism (5.3.9) from Theorem 5.3.1. Consequently, without [Zho], the central equality (9.1.17) — hence Theorem 9.2.3 — is unsupported. This is a dependency on an unavailable proof, not an internal contradiction; however, it means the paper does not establish the Modularity Conjecture on its own. The restriction to the trivial-similitude fiber L=O_X in Section 9 is a further limitation, but the Fourier-technique dependence is the more load-bearing gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript claims a proof of the Modularity Conjecture for higher theta series on moduli stacks of Hermitian shtukas, i.e., independence of the series eZ^{n,r}_m(G,E) of the choice of Lagrangian subbundle E⊂G, for all coranks m≤n and all r. The strategy is: (i) prove the Trace Conjecture for Hitchin stacks in the low-corank range m≤n/3 in §§3–4, linking derived fundamental classes of special cycles to categorical traces; (ii) develop a motivic derived Fourier transform in §§5–7 and use a Gaussian identity in §8 to compare the theta series for two transverse Lagrangians, yielding modularity in low corank (Theorem 9.1.1); (iii) extend to all coranks by an embedding trick and cancellation at a rank-2n point (Theorem 9.2.3). Part 4 establishes a stronger supermodularity for split covers/general linear groups. As submitted, the proof is written only for the trivial-similitude fiber L=O_X, the formal properties of the motivic Fourier transform are deferred to the forthcoming paper [Zho], and the split-case Trace Conjecture is asserted as a specialization of Theorem 4.2.1.","tokens_in":76931,"tokens_out":15347,"duration_ms":138222,"significance":"If the deferred foundations hold, this is a substantial result: it realizes the function-field analogue of Kudla's modularity conjecture at the integral-Chow level, where number-field analogues are largely conjectural, and it supplies the input used in [FHM25] for higher theta lifting. The paper contains genuinely new, checkable ingredients: the low-corank dimension bound (Theorem 3.1.1) with its stratification of framed shtukas (Corollary 3.4.3), the trace formula for non-proper correspondences (Proposition 7.1.1), the explicit scalar bookkeeping in Lemmas 9.1.4–9.1.5, and the supermodularity phenomenon. The Fourier-Gaussian comparison in §8.4 is concrete enough to be falsifiable. However, as submitted the main theorem is contingent on the unavailable [Zho] and on an unwritten extension from L=O_X to general L, so the significance is conditional on those gaps being filled.","major_comments":[{"comment":"Theorem 1.1.1 and Conjecture 9.0.1 are stated for arbitrary similitude fiber L, but the proof in Section 9 is carried out only for L=O_X. The restriction is explicit: 'To keep the notation manageable, we write out the proof below only in the case of the trivial-similitude fiber L=O_X. The same ideas apply straightforwardly to the general case.' This is a load-bearing gap, not a notational simplification: for general L the Hermitian structure is G ≅ σ^*G^∨⊗ν^*L and the exact sequences (8.1.1) acquire extra tensor factors ν^*L; the normalization in (9.0.3) contains q^{n(deg E − deg L − deg ω_X)/2}; and the determinant/character computation (9.1.28)–(9.1.29) together with the Riemann–Roch identities (9.1.26)–(9.1.27) are written only for L=O_X. The embedding trick of §9.2 is likewise formulated only for trivial-similitude G. As submitted, Theorem 1.1.1 is therefore not proved at its stated strength; the extension must either be written out or the theorem restricted to the case actually proved.","section":"§1.1 / Conjecture 9.0.1 / §9 (p. 52)"},{"comment":"Theorem 5.3.1 states the full set of formal properties of the motivic derived Fourier transform—base change (5.3.1)–(5.3.2), involutivity (5.3.3), linear-map functoriality (5.3.4), Plancherel (5.3.9), and the Gysin/forget-supports compatibility (5.3.12)—and defers all proofs to the forthcoming work [Zho]. These properties are used at load-bearing points: Lemma 5.4.3 uses the Plancherel isomorphism (5.3.9) to prove invertibility of the relative Gauss cohomology; the Fourier transform of cohomological correspondences in §6.4, formulas (6.4.4) and (6.4.6), is asserted by verbatim carry-over of [FYZ23, §7]; and Lemma 9.1.4, through Proposition 8.4.5 and equation (9.1.17), uses Proposition 6.4.1 and Proposition 5.3.2 to obtain the central Fourier-duality comparison. Consequently Theorem 9.1.1, hence Theorem 9.2.3, is not established within the manuscript: the central equality (9.1.17) is unsupported unless the contents of [Zho] are available. The paper needs to supply proofs, or at least complete and verifiable statements, of the properties actually invoked, or to state the main theorem as conditional on [Zho].","section":"§5.3, Theorem 5.3.1; used in Lemma 9.1.4, Eq. (9.1.17)"},{"comment":"The split-case Trace Conjecture is the input for the low-corank supermodularity theorem, yet its proof consists of one paragraph asserting that the proof of Theorem 4.2.1 'goes through with no essential changes.' This is not visibly a specialization: Theorem 12.2.1 has separate bounds m_1≤n/3 and m_2≤n/3, the legs are signed sequences µ∈{±1}^r on the two components X^{(1)}⊔X^{(2)}, and the degree formula d_µ=r_+(n−m_1)+r_−(n−m_2) differs from d(h_r)=r(n−m). In the connected proof of Proposition 4.3.1, the choice of x_0 with ν^{-1}(x_0) disjoint from the exceptional leg points is used to kill boundary classes; no analogue is supplied for the split two-component situation. Since the supermodularity claim in the abstract and Theorem 10.1.1 depend on this theorem, the reduction needs to be written out or the theorem proved directly.","section":"§12.2, Theorem 12.2.1; used in Theorem 12.2.2"}],"minor_comments":[{"comment":"The relation between the leg count r of Sht^r_{U(n)} in Part 3 and the signed sequence µ∈{±1}^r with ∑µ_i=0 in Part 4 (Sht^µ_n) is only explained in §12.1; a forward reference at the first use of Sht^µ_n in §10 would avoid confusion.","section":"§2.1 / §10.1 / §12.1"},{"comment":"The displayed headings 'F ramed Hitchin stacks' and 'F ramed shtukas' have a missing accented character, and in Proposition 3.3.1 the labels in 'type0,+,−,±' are rendered without separators; these should be cleaned up.","section":"§3.3–§3.4"},{"comment":"The symbol TrSht is used both for the categorical trace of a cohomological correspondence and for its Frobenius evaluation (a locally constant function or Chow class), e.g., in the paragraphs around (9.1.21); the two uses should be distinguished following the conventions of [FK24, §6.4.3], since one is an object and the other is its evaluation.","section":"§9.1.4, Lemma 9.1.5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is the third in a self-authored series and proves a conjecture formulated in [FYZ25], which is normal for such a series, but two dependencies deserve editorial scrutiny: the core formal properties of the motivic derived Fourier transform are deferred to the forthcoming [Zho], and the proof of the main theorem is written only for L=O_X with an asserted 'straightforward' extension. The review could not access [Zho]; if the journal requires self-contained proofs of load-bearing formalism, the paper as submitted should be returned for revision. The abbreviated proof of the split-case Trace Conjecture (Theorem 12.2.1) also merits a request for the actual argument. The AI-usage disclosure in §1.6 is transparent; the GPT-assisted proof of Theorem 3.1.1 is accompanied by the authors' own write-up and does not affect the mathematical assessment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, the main result is real and significant: it proves the Modularity Conjecture for higher theta series on unitary shtukas, and it adds supermodularity for the general linear case. The low-corank Trace Conjecture, the dimension bounds that feed it, and the embedding trick are genuine advances. The overall strategy is coherent: start with the r=0 Poisson argument, promote it through motivic Fourier duality, then extract Chow-valued modularity via the sheaf-cycle correspondence. I believe the authors know what they are doing, and the paper is not a performance; it is dense, honest, serious mathematics.\n\nSecond, the stress-test note is right. Section 5.3 states the formal properties of the motivic derived Fourier transform and explicitly defers all proofs to Tong Zhou's forthcoming work. These properties are not optional commentary: Lemma 9.1.4, the Fourier-duality comparison at the heart of the proof of Theorem 9.1.1, invokes Proposition 6.4.1 and Proposition 8.4.5, both of which use Theorem 5.3.1. Even Lemma 5.4.3 uses the Plancherel isomorphism from that theorem. So without [Zho], the central equality (9.1.17) is unsupported. This is a dependency on an unavailable proof, not an internal contradiction, but it means the paper as submitted does not establish the Modularity Conjecture on its own.\n\nThere are two smaller soft spots. Section 9 proves the main theorem only in the trivial-similitude fiber L=O_X and says the general case follows straightforwardly. That may be true, but the theorem is stated for all L, so the reader cannot verify the full claim from the text. And the split-case Trace Conjecture, Theorem 12.2.1, is asserted to go through with no essential changes; that is plausible but again not shown.\n\nThe citation pattern is fine: relying on [FYZ23] and [FK24] is legitimate because those foundations are published. The problem is specifically [Zho], which is not available to the reader.\n\nWho is this for? Specialists in the Kudla program over function fields, derived algebraic geometry, and the geometry of shtukas. It absolutely deserves a serious referee. My recommendation: send it to peer review, but the referee should have access to [Zho] or the authors should be asked to include the deferred proofs. If the editors require self-containedness, this is a major revision; if not, it is a strong paper whose verification is conditional on a forthcoming reference.","headline":"A serious proof of a real conjecture, but the submitted text leans on a forthcoming Fourier-transform paper for a load-bearing step and only writes the main proof in the trivial-similitude case.","tokens_in":77462,"tokens_out":1738,"would_cite":true,"duration_ms":18929,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F27","14C15","14D23","14F42"],"pacs":[],"model":"deepseek-v4-flash","headline":"The Modularity Conjecture for higher theta series on Hermitian shtukas is proved: the Fourier sum over special cycles depends only on the skew-Hermitian bundle, not on the Lagrangian subbundle.","keywords":["higher theta series","Modularity Conjecture","Hermitian shtukas","special cycles","Trace Conjecture","motivic derived Fourier transform","sheaf-cycle correspondence","supermodularity"],"falsifier":"A decisive check would fix $k=\\mathbb F_q$, a nontrivial quadratic cover $X'\\to X$, rank $n=3$, corank $m=1$, and $r=1$, choose a trivial-similitude skew-Hermitian bundle $G$ with two transverse Lagrangians $E_1,E_2$, and compare the two classes in $\\mathrm{CH}^{2}(\\mathrm{Sht}^1_{U(3)})$; because the proof identifies the classes through Frobenius traces of the Gauss cohomology, a single discrepancy between the point counts of the two special-cycle stacks over $\\mathbb F_{q^k}$ would disprove Theorem 9.2.3.","tokens_in":76428,"feed_emoji":"🧮","tokens_out":16002,"duration_ms":137411,"temperature":0.7,"pith_summary":"This paper proves the Modularity Conjecture for higher $\\theta$ series on moduli stacks of Hermitian shtukas (bundles with prescribed modifications at moving points) over function fields. The higher $\\theta$ series $\\tilde Z^{n,r}_m(G,E)$ is a Fourier series with coefficients in the Chow group of the shtuka stack, defined from a skew-Hermitian bundle $G$ and a Lagrangian subbundle $E\\subset G$; the conjecture asserts that this series is independent of $E$. The proof establishes the low-corank Trace Conjecture, which identifies the virtual fundamental class of each special cycle with the shtuka-twisted categorical trace of a cohomological correspondence, and then lifts the result to all coranks by an embedding trick. For general linear groups the paper proves a stronger statement, supermodularity: every parabolic-refined higher $\\theta$ series is essentially the special-cycle class of the bundle alone, sheared by an explicit power of $q$. This gives the function-field analogue, in integral Chow-valued form, of the strongest tier of the arithmetic $\\theta$ series program.","feed_headline":"Higher theta series depend only on the skew-Hermitian bundle","feed_subtitle":"Modularity now holds in integral Chow groups, and general linear theta series shed all parabolic data.","key_machinery":"The argument is carried by the motivic derived Fourier transform acting on cohomological correspondences between derived vector bundles $U,V,W$ attached to a transverse pair of Lagrangians and to the torsion sheaf $Q_1$ they generate. The key identity is the varying-base Gaussian correspondence: the normalized Fourier transform of the $\\beta$-Gaussian correspondence equals the $-\\beta$-Gaussian correspondence tensored with the relative Gauss cohomology $G_Q$, whose Frobenius trace is the explicit scalar $q^{d/2}\\eta_{F'/F}(D_Q)^n$. Once the identity is pushed to the shtuka fixed-point stack and traced, the sheaf-cycle correspondence converts it into the Chow-valued modularity equality. The low-corank Trace Conjecture supplies the other half of the mechanism: it identifies the virtual fundamental class of each special cycle with the shtuka-twisted trace, so the Fourier coefficients of the $\\theta$ series are recognized as traces of the same correspondences.","core_discovery":"Theorem 1.1.1 states that the Modularity Conjecture of [FYZ25] holds. Concretely, for every trivial-similitude skew-Hermitian bundle $G$ of rank $2m$ and any two Lagrangian subbundles $E_1,E_2\\subset G$, one has $\\tilde Z^{n,r}_m(G,E_1)=\\tilde Z^{n,r}_m(G,E_2)$ in $\\mathrm{CH}^{r(n-m)}(\\mathrm{Sht}^r_{U(n)})$, so the Fourier sum over special cycles descends from the pair $(G,E)$ to the isomorphism class of $G$ alone. The proof rests on Theorem 4.1.3, the Trace Conjecture in low corank $m\\le n/3$: the identity $\\mathrm{Tr}_{\\mathrm{Sht}}(c_{\\mathcal M})=[\\mathrm{Sht}^r_{\\mathcal M}]$ realizes derived fundamental classes of special cycles as categorical traces of cohomological correspondences. An embedding-and-cancellation argument then reduces every rank to the low-corank range. For split double covers, the same machinery yields supermodularity (Theorem 10.1.1): $\\tilde Z^{\\psi,\\mu}_{m_1,m_2}(E_1,G)=\\sum_d q^{-m_2 d+m_2 n(g-1)}[Z^{\\mu}_{G,0}]_d$ in $\\mathrm{CH}^{\\frac r2(2n-m)}(\\mathrm{Sht}^\\mu_n)$, expressing all parabolic refinements through the special-cycle class of $G$.","pith_inferences":["A natural extension, not pursued in the paper, is to use the same low-corank trace plus embedding-and-cancellation scheme for orthogonal and symplectic dual pairs; the authors state the decisive steps are insensitive to the group-theoretic setup.","The low-corank dimension bound on the injective core is a purely geometric statement that can be tested independently of the trace formalism; its function-field proof suggests an analogous expected-dimension prediction for unitary Shimura varieties when the corank is at most one third of the target rank.","Since the proof identifies the two theta series by pushing forward cohomological correspondences, a plausible sharpening would be a canonical isomorphism of the correspondences themselves before taking traces; the paper's results supply the isomorphism at the level of traced classes, leaving the full derived-category refinement open."],"forward_implications":["The Modularity Conjecture holds: for any trivial-similitude skew-Hermitian bundle $G$ of rank $2m$ and any Lagrangians $E_1,E_2\\subset G$, $\\tilde Z^{n,r}_m(G,E_1)=\\tilde Z^{n,r}_m(G,E_2)$ in $\\mathrm{CH}^{r(n-m)}(\\mathrm{Sht}^r_{U(n)})$.","The Trace Conjecture for Hitchin stacks is true for $m\\le n/3$, so in this range virtual fundamental classes of special cycles are realized as shtuka-twisted categorical traces without restricting the shtuka legs; this is what makes the proof integral rather than generic-fiber only.","For general linear groups, supermodularity (Theorem 10.1.1) holds: every parabolic-refined higher theta series $\\tilde Z^{\\psi,\\mu}_{m_1,m_2}(E_1,G)$ equals the sheared special-cycle class $[Z^{\\mu}_{G,0}]^{\\langle}$, independent of the parabolic and of the additive character.","The higher theta lifting and higher arithmetic inner product formula that were conditional on the Modularity Conjecture become unconditional.","Because modularity is now proved in integral Chow groups, applications to arithmetic intersection theory over function fields and to local special-cycle questions, including a higher arithmetic fundamental lemma, come within reach."],"supporting_citations":[{"why":"Constructs the higher theta series, formulates the Modularity Conjecture (Conjecture 4.15), and supplies the explicit virtual fundamental classes of special cycles that the present proof re-expresses as traces.","marker":"[FYZ25]"},{"why":"Introduced the sheaf-cycle correspondence and proved generic-fiber modularity for unitary groups; the present Fourier-duality argument refines and replaces the level-structure method used there.","marker":"[FYZ23]"},{"why":"Developed the motivic sheaf-cycle correspondence and the homogeneous motivic Fourier transform; the paper's refined inhomogeneous transform and trace machinery build on this framework.","marker":"[FK24]"},{"why":"Formulated the Trace Conjecture for Hitchin stacks (Conjecture 9.4.2); Theorem 4.1.3 of this paper proves the low-corank case.","marker":"[FH25]"},{"why":"Gives the construction and dimension/level-structure properties of Hermitian shtukas and special cycles used throughout the proof, including the smoothness of the leg maps and the quasi-finiteness of the injective-locus maps.","marker":"[FYZ24]"},{"why":"Forthcoming work supplying the formal properties of the motivic derived Fourier transform (Theorem 5.3.1) on which the core Fourier-duality comparison in Lemma 9.1.4 rests.","marker":"[Zho]"},{"why":"Provides the relative fundamental class and Gysin formalism on derived Artin stacks, including the excess intersection formula, used to define the cohomological correspondences and their traces.","marker":"[Kha19]"}],"fun_headline_variants":["Higher theta series become pure bundle invariants","Modularity conjecture proven for higher theta series","General linear theta series shed all parabolic data","Supermodularity: theta series forget Lagrangians"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The full-strength theorem assumes that the formal properties of the motivic derived Fourier transform stated in Theorem 5.3.1, whose proofs are deferred to a forthcoming paper, hold as stated, and that the argument written for the trivial-similitude fiber $L=\\mathcal O_X$ extends to arbitrary line bundles $L$ with only routine modifications.","fun_headline_variants_meta":{"raw":{"variants":["Higher theta series become pure bundle invariants","Modularity conjecture proven for higher theta series","General linear theta series shed all parabolic data","Supermodularity: theta series forget Lagrangians"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000225,"raw_usage":{"total_tokens":1458,"prompt_tokens":930,"completion_tokens":528,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":469}},"tokens_in":546,"tokens_out":528,"duration_ms":5222,"temperature":1.0,"reasoning_tokens":469,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T13:35:32.962923+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive check would fix $k=\\mathbb F_q$, a nontrivial quadratic cover $X'\\to X$, rank $n=3$, corank $m=1$, and $r=1$, choose a trivial-similitude skew-Hermitian bundle $G$ with two transverse Lagrangians $E_1,E_2$, and compare the two classes in $\\mathrm{CH}^{2}(\\mathrm{Sht}^1_{U(3)})$; because the proof identifies the classes through Frobenius traces of the Gauss cohomology, a single discrepancy between the point counts of the two special-cycle stacks over $\\mathbb F_{q^k}$ would disprove Theorem 9.2.3.","supporting_citations":[],"review_version":1}