{"id":"5283d95c-fe79-43a3-93ce-9e7463caac92","arxiv_id":"1909.02697","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A global proof of the arithmetic fundamental lemma for unitary groups over Q_p (odd p >= n) via an SL_2-equivariant relative trace formula and derived CM cycles, which also yields the Jacquet-Rallis fundamental lemma for residue field size q >= n.","lead":"Zhang proves the arithmetic fundamental lemma for unitary groups over Q_p when p >= n, a core local identity in the arithmetic Gan-Gross-Prasad program, and also proves the Jacquet-Rallis fundamental lemma when the residue field has size q >= n. The proof is global: it uses an SL_2-symmetry from the Weil representation in a linearized relative trace formula, modular generating functions of special divisors, and a new derived CM cycle.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The adaptation of Theorem 8.6 from [6] to the non-maximal level structure at primes dividing d is justified in a single paragraph, yet this modularity result is load-bearing for the global comparison and ultimately for the AFL.","rationale":"The paper's central claim is a proof of the arithmetic fundamental lemma for unitary groups over Q_p when p >= n, on strongly regular semisimple elements, plus the Jacquet-Rallis fundamental lemma for q >= n. The visible structure is coherent: the FL is proved in Theorem 13.9, the arithmetic comparison is developed in Sections 9-14, and the final Section 15 is not in the review copy. The most load-bearing assumption is the modularity of the arithmetic generating series, Theorem 8.6, because it is the bridge that identifies the global arithmetic intersection generating series with a holomorphic modular form, allowing the local AFL to be extracted via Fourier coefficients and Lemma 13.6. The theorem is imported from [6] and adapted to the level structure at primes dividing d in one paragraph; the stated localization argument does not by itself establish compatibility of the theta lift construction with the non-maximal level and the Eisenstein condition in Definition 6.1. This is not an objection to [6] itself, but to the unsupported adaptation. The reader's weakest_assumption identified the same point, and my concern does not move the verdict: the paper should remain conditional until Theorem 8.6's level-d variant is either written out in detail, verified in a concrete example, or replaced by a fully proven statement. No ad hominem is intended; the critique targets an import step, not the author. Other possible concerns, such as the srs restriction or the use of partial Gaussian test functions, are acknowledged in the paper and appear less likely to invalidate the stated theorem.","tokens_in":78157,"tokens_out":9234,"duration_ms":94392,"concrete_test":"For a small explicit case, e.g., n=2 or 3 with one prime ell dividing d and a non-maximal K_G,ell, compute the first positive Fourier coefficient of \\hat Z^B(\\tau,\\varphi) by the regularized theta lift of [6] and compare it with the KR divisor Z(1,\\varphi) with automorphic Green function, working in \\hat{Ch}^1(M) modulo cycles supported at ell. Agreement for several primes and levels would support Theorem 8.6; disagreement would isolate a failure of the adaptation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 8.6 states that the generating series of KR special divisors with automorphic Green functions is a holomorphic modular form of weight n with coefficients in the arithmetic Chow group of M = M_{K_{\\widetilde G}}(\\widetilde G). Its proof consists of one paragraph: the assertion is said to follow from [6], with the explanation that the arithmetic Chow group omits a finite set S of bad places and therefore the computations of divisors of regularized theta lifts and Borcherds products over Spec O_E[1/d] still apply. This is a genuine reduction step, but not a proof of compatibility with the level structure introduced in Definition 6.1: at primes dividing d, the compact open K_{\\widetilde G}=K_{Z_Q}\\times K_G is not the maximal level used in [6], and the moduli functor imposes an Eisenstein condition and a non-maximal level structure on the rational Tate module. The subsequent argument depends on Theorem 8.6 in an essential way: (9.7) places Int(\\tau,\\Phi) in the same finite-dimensional space A^{hol}(\\Gamma(N),n)_Q\\otimes R_{S,Q} as the analytic object 2\\partial J^{hol}, and Lemma 13.6 then upgrades equality of almost all Fourier coefficients into equality of modular forms. If Theorem 8.6 fails for the level-d components, the q-expansion of Int(\\tau,\\Phi) could differ in the coefficients whose indices are divisible only by primes in S, and the comparison with the analytic side would not follow. The paper provides no independent check of this adaptation, and no machine-checked or numerical verification is supplied for any instance of Theorem 8.6 at non-maximal level.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves the arithmetic fundamental lemma (AFL) conjecture for unitary groups in n variables over Q_p when p is odd and p >= n, on the strongly regular semisimple locus, and also proves the Jacquet--Rallis fundamental lemma (FL) for unramified quadratic extensions with residue field size q >= n. The method is global: the author fixes an irreducible conjugate self-reciprocal polynomial alpha, constructs a partially linearized relative trace formula with an SL_2-symmetry supplied by the Weil representation, and compares its first derivative against a generating series of arithmetic intersections of Kudla--Rapoport special divisors with a newly introduced derived CM cycle. The comparison is performed in a finite-dimensional space of holomorphic modular forms; local FL and AFL identities are extracted by controlling Fourier coefficients away from a finite set of bad places, using local constancy of orbital integrals and of intersection numbers. The main theorems are Theorem 13.9 (FL), Theorem 14.6 (global comparison), and Theorem 15.1 (AFL in the stated range).","tokens_in":78342,"tokens_out":10484,"duration_ms":115641,"significance":"If the proof is complete, the paper is a major advance: it establishes a central conjecture in the arithmetic Gan--Gross--Prasad program and provides a new global induction mechanism that also yields the Jacquet--Rallis FL. The paper contains substantial original constructions, including the semi-Lie algebra version of the relative trace formula, partial Gaussian test functions, the derived fat big CM cycle, and a local constancy theorem for Rapoport--Zink intersection numbers via the Vollaard--Wedhorn stratification. The author is admirably explicit about the strongly regular semisimple restriction and about the dependence of the proof on imported modularity results. The main unresolved point is whether the imported modularity statements, especially Theorem 8.6, are valid with the non-maximal level structure used in Definition 6.1; this is load-bearing and needs to be addressed before the central claim can be regarded as fully established.","major_comments":[{"comment":"Theorem 8.6 is load-bearing for the global comparison: it is what places the arithmetic intersection generating series Int(tau, Phi) in the same finite-dimensional space A^hol(Gamma(N), n)_Q tensor R_{S,Q} as the analytic object 2 dJ^hol, after which Lemma 13.6 upgrades equality of almost all Fourier coefficients into equality of modular forms. Its proof, however, is only a one-paragraph appeal to [6] that does not address the level structure introduced in Definition 6.1. In the present paper the compact open subgroup K_{\\tilde G}=K_{Z_Q} times K_G is allowed to be non-maximal at all primes v dividing d, and the moduli functor imposes an Eisenstein condition and a non-maximal rational Tate-module level structure; the computations in [6] are made for a maximal-level moduli problem over the full ring of integers. Passing to O_E[1/d] removes the fibers over the finite set S, but it does not automatically transfer the divisor and Borcherds-product computations to the level-d components. The coefficients of q^xi whose prime factors lie in S are precisely the ones needed in the 'almost all coefficients' argument of Lemma 13.6, so a failure of Theorem 8.6 at the level-d components would break Theorem 14.6 and hence Theorem 15.1. The authors should either prove Theorem 8.6 for the level structure of Definition 6.1 or cite a statement that literally covers it.","section":"§8.4, Theorem 8.6"},{"comment":"Proposition 2.7 is used in the induction in Theorem 13.9: part (i) is the equivalence between the group FL for S_n and the semi-Lie algebra FL for S_{n-1} times V'_{n-1}, and part (ii) derives the semi-Lie version from the group version. The proof of Proposition 2.7 is explicitly omitted, with the text saying only that the proof of the analogous Proposition 4.12 will be given and that 'the proof also works here.' This is not a routine reversal: Proposition 4.12 concerns derivatives of orbital integrals and intersection numbers in the AFL setting, whereas Proposition 2.7 concerns plain orbital integrals of characteristic functions in the FL setting. The reduction requires the same orbit maps r_xi, r^natural_xi and the Cayley-transform lemmas of §4.3 to be adapted to the FL case, including the transfer-factor bookkeeping of Lemma 4.11. Since the FL induction depends on this statement, it should be proved in the text or supplied with a precise reference that covers the semi-Lie algebra FL case.","section":"§2.4, Proposition 2.7"},{"comment":"The archimedean correction input is also imported: Theorem 8.4 is stated to follow from Ehlen--Sankaran [8] by an embedding trick, but the generating function Z_{v0,corr} in (8.14) uses Schwartz functions of the form phi = 1_{Lambda^d} tensor phi_d with arbitrary level at primes dividing d, and Proposition 14.5 then uses this to cancel the non-holomorphic terms in 2 dJ(h, Phi') + Int^{K-B}(h, Phi). The one-sentence proof of Theorem 8.4 does not verify that the level structure of the unitary Shimura variety or the non-maximal components at primes dividing d are compatible with the arguments in [8]. As in the case of Theorem 8.6, this is a load-bearing step for the holomorphic projection in Proposition 14.5. The authors should at least state the precise form of the Ehlen--Sankaran theorem they need and indicate where in [8] the non-maximal-level unitary case is covered.","section":"§8.3 and §14.2"}],"minor_comments":[{"comment":"The displayed definition of phi^flat_{h_f} contains corrupted tokens ('/d47/d47') that make the Iwasawa decomposition and the formula illegible; this rendering should be cleaned.","section":"§1.2"},{"comment":"There is a parenthesis typo in the second summand: Orb((gamma, 0-, Phi', s) should read Orb((gamma, 0-), Phi', s).","section":"§12.6"},{"comment":"The proof of Proposition 2.6 refers to [46, Lem. 2.5.5] for the Lie algebra version and says the argument is the same for the semi-Lie version; a short explanation of how the linear V-component is handled would improve readability.","section":"§2.3"},{"comment":"The terms 'naive fat big CM cycle' and 'derived CM cycle' are introduced rapidly; the reader would benefit from a short summary of the relationship to Bruinier--Kudla--Yang and Howard and of why the derived structure repairs the expected dimension, since this construction is advertised as the main novel geometric input.","section":"§7.4--7.6"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, this is the real result: Zhang proves the Jacquet–Rallis fundamental lemma for q ≥ n and the arithmetic fundamental lemma for unitary groups over Q_p for odd p ≥ n, at least for strongly regular semisimple elements. The main theorem is the long-awaited removal of the minuscule and n ≤ p restrictions. Second, the proof is a serious global argument, not a routine extension of his earlier AFL papers. The semi-Lie algebra RTF with the SL_2-action from the Weil representation, the derived CM cycle, and the Fourier-coefficient density lemma (Lemma 13.6) are genuinely new tools, and they fit together coherently.\n\nWhat is good: the visible argument is detailed and rigorous. Theorem 13.9 (FL) is proved by a complete-looking induction with globalization and partial Gaussian test functions. The local constancy theorem (Theorem 5.5) is proved from the Vollaard–Wedhorn stratification, and the global comparison (Theorem 14.6) reduces the AFL to the modularity of generating series plus a density argument. The paper is honest about its limitations: F0 = Q only, strongly regular semisimple rather than all regular semisimple, and the archimedean Gaussian test functions are only partially constructed. Self-citation is heavy because the conjecture and earlier cases are the author's own, but the derivation is independent.\n\nThe real soft spot is exactly the one the stress test flagged: Theorem 8.6, the modularity of the generating series of KR special divisors with coefficients in the arithmetic Chow group, is load-bearing. It is imported from Bruinier–Howard–Kudla–Rapoport–Yang in a single paragraph, adapting it to the non-maximal level structure at primes dividing d. If that adaptation fails, the q-expansion of the arithmetic intersection series could differ at coefficients supported on the bad primes, and the comparison with the analytic side would break. On reading, the reduction looks plausible—the paper works away from the bad set S and the moduli problem away from S is essentially the one in [6]—but it is a genuine gap in presentation, not a manufactured concern. A referee should ask for a more detailed proof of that theorem.\n\nA second, smaller issue is that the review copy is truncated at Section 14.3, so the final deduction of Theorem 15.1 is not fully visible. The structure is clear, but the last step deserves eyes on it. No code, no machine-checked proofs, but that is normal for this area.\n\nWho is this for? Arithmetic geometers and automorphic forms people working on the arithmetic Gan–Gross–Prasad program. It deserves a serious referee, not a desk reject. My recommendation: send it out, with a referee instructed to check Theorem 8.6 and the final globalization steps carefully.","headline":"A genuine proof of the AFL for odd p ≥ n, with one load-bearing modularity input that deserves careful referee scrutiny.","tokens_in":79101,"tokens_out":1798,"would_cite":true,"duration_ms":24658,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F27","11F67","11G40","14C25","14G35"],"pacs":[],"model":"deepseek-v4-flash","headline":"The arithmetic fundamental lemma for unitary groups holds for odd p≥n, on the strongly regular semisimple locus.","keywords":["arithmetic fundamental lemma","Jacquet–Rallis fundamental lemma","relative trace formula","Weil representation","Kudla–Rapoport divisors","derived CM cycles","unitary Shimura varieties","arithmetic Gan–Gross–Prasad"],"falsifier":"Compute both sides of Conjecture 3.8(b) for a single strongly regular semisimple pair at a place with $p=n$ in which $g$ generates a non-maximal order; this is the case the induction does not check directly and where the modularity argument is doing the work. A mismatch would refute the theorem, while agreement supplies the missing local check.","tokens_in":77719,"feed_emoji":"🧮","tokens_out":10846,"duration_ms":108739,"temperature":0.7,"pith_summary":"This paper proves the arithmetic fundamental lemma (AFL) conjecture for unitary groups in $n$ variables over $\\mathbb{Q}_p$, for odd $p\\ge n$ and strongly regular semisimple elements: the derivative of a normalized orbital integral of the characteristic function of $S_n(\\mathcal{O}_{F_0})$ equals the negative of a Kudla–Rapoport intersection number times $\\log q$. The same global method proves the Jacquet–Rallis fundamental lemma whenever the residue field has at least $n$ elements. The route is a partially linearized relative trace formula in which the Weil representation of $\\mathrm{SL}_2$ makes the relevant sums into holomorphic modular forms of weight $n$; the arithmetic side is matched by the modular generating series of special divisors paired with a derived CM cycle. A sympathetic reader would take the paper as establishing the local identities that the relative trace formula approach to the arithmetic Gan–Gross–Prasad conjecture needs, with the strongly regular restriction harmless because that locus is open dense.","feed_headline":"The arithmetic fundamental lemma for unitary groups holds for odd p≥n","feed_subtitle":"Global SL2 modularity plus derived CM cycles prove the Jacquet–Rallis fundamental lemma too.","key_machinery":"The load-bearing object is the semi-Lie algebra, or partially linearized, relative trace formula for the diagonal action of $U(V)$ on $U(V)\\times V$, matched with $\\mathrm{GL}_{n-1}$ acting on $S_{n-1}\\times V'_{n-1}$. Because of the linear factor, $\\mathrm{SL}_2(\\mathbb{A}_0)$ acts through the Weil representation, turning the distributions $I_\\alpha(\\omega(h)\\Phi)$ into automorphic functions; with Gaussian test functions at the archimedean places these become holomorphic modular forms of weight $n$ with finite-dimensional spectrum. On the arithmetic side the same modularity appears in the generating series $\\widehat{Z}^B(\\tau,\\varphi)$ of Kudla–Rapoport special divisors with automorphic Green functions, paired with the derived CM cycle, a thickened fixed-point locus of a Hecke correspondence of virtual dimension one. Equating the two modular forms and comparing Fourier coefficients where the polynomial $\\alpha$ is locally maximal isolates the local identities; the archimedean difference between two Green functions, controlled by Theorem 8.4, is subtracted first.","core_discovery":"On its own terms, the paper establishes Conjecture 3.8: for matching strongly regular semisimple data, $\\partial\\mathrm{Orb}((\\gamma,u'),1_{(S_{n-1}\\times V'_{n-1})(\\mathcal{O}_{F_0})})=-\\mathrm{Int}(g,u)\\,\\log q$ in the semi-Lie algebra version, together with the corresponding group version; this is Theorem 15.1, valid for odd $p\\ge n$. It also establishes the Jacquet–Rallis fundamental lemma, Conjecture 2.3, for residue field size $q\\ge n$ (Theorem 13.9). The proof is global and inductive: local matching identities for orbital integrals are recovered from the equality of two holomorphic modular forms, after subtracting archimedean Green-function corrections and using linear independence of logarithms of distinct primes to separate places.","pith_inferences":["The paper leaves implicit that the non-maximal-order cases at the bad set $B$ are forced by modularity rather than verified locally; this means the sharp threshold is really the vanishing lemma for Fourier coefficients, and strengthening that lemma would lower the bound $q\\ge n$.","A testable extension is to compute the unit-index Fourier coefficient at a bad place in the non-maximal case: the equality of modular forms predicts a specific value, giving a finite numerical check of the whole induction.","The same two-modular-forms strategy suggests a route to the arithmetic transfer conjecture, provided one can construct an arithmetic generating series for derived cycles under the second Hecke symmetry rather than only for special divisors."],"forward_implications":["For every odd prime $p\\ge n$, the arithmetic fundamental lemma holds on the strongly regular semisimple locus, so the remaining local input for the RTF approach to arithmetic Gan–Gross–Prasad is the arithmetic transfer conjecture.","The Jacquet–Rallis fundamental lemma is now known for all unramified quadratic extensions with residue field $q\\ge n$, extending the earlier large-$p$ results to the sharp range.","The group and semi-Lie algebra versions of the AFL are equivalent for $q\\ge n$ by Proposition 4.12, so the semi-Lie version proved here is not weaker in substance.","Since strongly regular semisimple elements form an open dense subset and the intersection numbers are locally constant by Theorem 5.5, the proved identity extends by continuity and is not an isolated equality."],"supporting_citations":[{"why":"Supplies Theorem 8.6, the holomorphic modularity of the arithmetic generating series of special divisors with automorphic Green functions, the key imported modularity.","marker":"[6]"},{"why":"Supplies Theorem 8.4 controlling the archimedean difference between the two Green functions used in the comparison.","marker":"[8]"},{"why":"Defines the Kudla–Rapoport special divisors $Z(u)$ on the Rapoport–Zink space that form the intersection numbers $\\mathrm{Int}(g,u)$.","marker":"[24]"},{"why":"Provides the integral models of unitary Shimura varieties and the non-archimedean uniformization along the basic locus used to globalize the local intersections.","marker":"[40]"},{"why":"Proved the Jacquet–Rallis fundamental lemma for large $p$; the present proof uses and extends this input.","marker":"[46]"},{"why":"Formulated the AFL conjecture, proved low-rank cases, and set up the relative trace formula framework this paper follows.","marker":"[47]"},{"why":"Proved the existence of smooth transfer for Jacquet–Rallis orbital integrals, including the Weil-representation compatibility needed for the global comparison.","marker":"[48]"},{"why":"Developed the Fourier–Jacobi relative trace formula that motivates the semi-Lie algebra version of the FL and AFL.","marker":"[29]"},{"why":"Provided the special case of the semi-Lie AFL when the order generated by $g$ is maximal, and the K-theoretic formalism for formal schemes.","marker":"[33]"}],"fun_headline_variants":["AFL for unitary groups proven globally for odd p≥n","Global modularity proof settles unitary arithmetic fundamental lemma","Weil representation approach proves AFL for unitary groups","Arithmetic fundamental lemma for unitary groups: odd p≥n proven","Unitary AFL and Jacquet–Rallis lemma via global SL2 modularity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof hinges on the arithmetic generating series of special divisors being a holomorphic modular form at the required level structure, a fact imported from [6] and adapted in a single paragraph; if that modularity fails, the comparison of the two modular forms collapses.","fun_headline_variants_meta":{"raw":{"variants":["AFL for unitary groups proven globally for odd p≥n","Global modularity proof settles unitary arithmetic fundamental lemma","Weil representation approach proves AFL for unitary groups","Arithmetic fundamental lemma for unitary groups: odd p≥n proven","Unitary AFL and Jacquet–Rallis lemma via global SL2 modularity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000706,"raw_usage":{"total_tokens":3073,"prompt_tokens":726,"completion_tokens":2347,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":342,"completion_tokens_details":{"reasoning_tokens":2262}},"tokens_in":342,"tokens_out":2347,"duration_ms":17712,"temperature":1.0,"reasoning_tokens":2262,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:44:49.108530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute both sides of Conjecture 3.8(b) for a single strongly regular semisimple pair at a place with $p=n$ in which $g$ generates a non-maximal order; this is the case the induction does not check directly and where the modularity argument is doing the work. A mismatch would refute the theorem, while agreement supplies the missing local check.","supporting_citations":[{"cited_title":"Yun, The fundamental lemma of Jacquet–Rallis in positive charac teristics, Duke Math","cited_arxiv_id":null,"evidence_quote":"Proved the Jacquet–Rallis fundamental lemma for large $p$; the present proof uses and extends this input."},{"cited_title":"Modularity of generating series of divisors on unitary Shimura varieties","cited_arxiv_id":"1702.07812","evidence_quote":"Supplies Theorem 8.6, the holomorphic modularity of the arithmetic generating series of special divisors with automorphic Green functions, the key imported modularity."},{"cited_title":"Ehlen, S","cited_arxiv_id":null,"evidence_quote":"Supplies Theorem 8.4 controlling the archimedean difference between the two Green functions used in the comparison."},{"cited_title":"Kudla and M","cited_arxiv_id":null,"evidence_quote":"Defines the Kudla–Rapoport special divisors $Z(u)$ on the Rapoport–Zink space that form the intersection numbers $\\mathrm{Int}(g,u)$."},{"cited_title":"Arithmetic diagonal cycles on unitary Shimura varieties","cited_arxiv_id":"1710.06962","evidence_quote":"Provides the integral models of unitary Shimura varieties and the non-archimedean uniformization along the basic locus used to globalize the local intersections."},{"cited_title":"Zhang, On arithmetic fundamental lemmas , Invent","cited_arxiv_id":null,"evidence_quote":"Formulated the AFL conjecture, proved low-rank cases, and set up the relative trace formula framework this paper follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proved the existence of smooth transfer for Jacquet–Rallis orbital integrals, including the Weil-representation compatibility needed for the global comparison."},{"cited_title":"2, 8, 11","cited_arxiv_id":null,"evidence_quote":"Developed the Fourier–Jacobi relative trace formula that motivates the semi-Lie algebra version of the FL and AFL."},{"cited_title":"3, 4, 13, 14, 32","cited_arxiv_id":null,"evidence_quote":"Provided the special case of the semi-Lie AFL when the order generated by $g$ is maximal, and the K-theoretic formalism for formal schemes."}],"review_version":1}