{"id":"32d05454-5605-4c3f-8b06-ddae771b285e","arxiv_id":"2504.17484","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves the arithmetic transfer conjecture for unitary Rapoport-Zink spaces in full generality by constructing comparison isomorphisms between absolute and relative local models and p-divisible group categories.","lead":"Shimura varieties for unitary groups are hard to define at primes where the number field ramifies. This paper builds new integral models there, compares them with known local models, and uses that comparison to prove the arithmetic transfer conjecture for all odd primes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The globalization step in §9.3 relies on [Mih22a, Thm 1.2] for local constancy of intersection numbers over RZ spaces whose base field may be ramified over Q_p; the paper asserts applicability via [Zha25a, Rem. 6.16] without stating hypotheses, and the whole Step 6 conclusion depends on it.","rationale":"The reader's weakest_assumption identifies the same globalization step: Step 1 of §9.3 invokes [Mih22a, Theorem 1.2] for local constancy, and the entire Fourier-coefficient argument in Step 6 depends on it. I agree that this is the most load-bearing point. The paper gives detailed proofs for the local-model comparison, the p-divisible group equivalence, and the lci property of global cycles, and those parts appear internally coherent. The arithmetic-transfer proof, however, is a global induction that imports substantial technical machinery from [Zha25a], and the specific step that must extend to ramified F0 is the local-constancy assertion. The paper's citation of [Zha25a, Remark 6.16] is not a proof; it is an assertion that the earlier unramified-over-Q_p result applies verbatim to the new RZ spaces. Because this is exactly the assumption the paper aims to remove, the applicability of [Mih22a, Thm. 1.2] should be verified before the central claim is accepted unconditionally. If the theorem does apply, the proof strategy appears sound and the ACCEPT verdict would be justified; if it does not, Theorem 9.5 as stated is unproved. I therefore recommend CONDITIONAL rather than UNCHANGED, since the concern is concrete and testable without being a demonstrated error.","tokens_in":71720,"tokens_out":12324,"duration_ms":127409,"concrete_test":"Read [Mih22a, Theorem 1.2] in the original and check its hypotheses against Definition 5.20 and Corollary 5.14 with F0 of ramification index e>1 over Q_p (e.g., p odd, F0=Q_p(√p), F=F0(√u) with u a nonsquare unit). Determine whether the theorem's statement or proof restricts to absolutely unramified F0 or uses Z_p-linear deformation data that fails for e>1. If the hypotheses cover all p-adic F0, the concern is resolved; if not, identify the exact failing condition and exhibit a pair (g,u) for which the v-adic constancy assertion of Step 1 is not covered.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 9.5 is a globalization argument. In Step 1 of §9.3, a local pair (g_v,u_v) is replaced by a v-adically close global pair (g_0,u_0) with (u_0,u_0) totally positive; the equivalence AT(F_v/F0,v,L_v,g_v,u_v) ⇔ AT(F_v/F0,v,L_v,g_0,u_0) is justified solely by 'local constancy of orbital integrals and intersection numbers [Mih22a, Theorem 1.2], which applies in our situation (see [Zha25a, Remark 6.16])'. This is the key passage where the paper moves from the absolutely-unramified setting of [Zha25a] to arbitrary p-adic F0: Steps 2–6, in particular the equality 2J_α(ξ0,Φ′)+Int^{K-B}(α,ξ0,Φ)+Int(α,ξ0,Φ)=0 at the end of Step 6 and the extraction of the local identity from it, all inherit this step. The paper does not state the hypotheses of [Mih22a, Thm. 1.2], does not check them against the RZ spaces N^[t] of Definition 5.20 when F0/Q_p is ramified, and does not prove the cited remark [Zha25a, Rem. 6.16]. If Mih22a's theorem requires the coefficient field to be absolutely unramified, the reduction to the maximal-order type α0 (Step 4) and the final Fourier-coefficient comparison (Step 6) fail for the new cases claimed in Theorem 9.5. This is a transferred-hypothesis risk, not an internal contradiction; a secondary citation dependency of the same kind appears in Step 4 for the very special 1-cycles (footnote 8 and [GLX23, Cor. 1.5]), but the local-constancy input is earlier and more global.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies unitary Shimura varieties at primes where the totally real base field ramifies over Q. It introduces an (A,B)-strictness condition formulated via Eisenstein ideals, proves comparison isomorphisms between absolute and relative local models (Theorems 3.6 and 3.17), lifts these comparisons to categories of O-displays and p-divisible groups with duality-compatible equivalences (Theorems 4.9, 5.4, 5.12), and derives comparison isomorphisms for Rapoport–Zink spaces and their Kottwitz–Rapoport strata. In Part 2, the paper constructs integral models of RSZ unitary Shimura varieties, defines special divisors and derived CM cycles, and then proves the arithmetic transfer conjecture of the third author (Theorem 9.5) for odd p, arbitrary p-adic base field F0, unramified quadratic extension F/F0, and maximal parahoric level at an inert place. The proof follows the globalization and double-modularity strategy of Zhang, Mihatsch–Zhang, and Zhang, extended by the new local comparison results.","tokens_in":72131,"tokens_out":8740,"duration_ms":91572,"significance":"If correct, Theorem 9.5 is a major result: it removes the absolute unramifiedness assumption on F0/Qp in the arithmetic transfer conjecture, covering all odd p and all p-adic fields F0 for unramified F/F0. The local model comparisons and the duality-compatible display equivalences are independently valuable and are proved in the text rather than assumed. The paper also gives explicit integral models in new ramified cases, which is likely to have further applications to Kudla–Rapoport cycles and arithmetic inner product formulas. I saw no circularity: Theorem 9.5 is not used in its own proof, and the comparison (1.4) is derived from the Eisenstein condition. However, the arithmetic transfer proof relies at several load-bearing points on external results whose hypotheses are not stated or checked in the manuscript; for this reason the unconditional status of the main theorem is not yet fully established by the text.","major_comments":[{"comment":"The reduction from a local pair (g_v,u_v) to a v-adically close global pair (g_0,u_0) with (u_0,u_0) totally positive is justified solely by the sentence “By local constancy of orbital integrals and intersection numbers [Mih22a, Theorem 1.2] which applies in our situation (see [Zha25a, Remark 6.16]).” The hypotheses of [Mih22a, Theorem 1.2] are not stated, and no verification is given that they hold for the Rapoport–Zink spaces N^[t] of Definition 5.20 when F0/Qp is ramified. This step is the only place where the argument passes from the absolutely unramified setting of [Zha25a] to arbitrary p-adic F0; Steps 2–6, including the key identity at the end of Step 6, inherit this step. If [Mih22a, Theorem 1.2] requires the base field to be absolutely unramified, the globalization and therefore Theorem 9.5 fail for the new cases. The cited [Zha25a, Remark 6.16] cannot serve as a substitute because it belongs to a paper whose main theorem is proved only for F0 unramified over Qp. Please state the precise conditions under which the local-constancy theorem applies and check them, or prove the needed local constancy directly.","section":"§9.3, Step 1"},{"comment":"The construction of the very special 1-cycles, which is needed to obtain the modified cycle CML(α,f)^mod in (9.13), depends on the claim that the reduced Kottwitz–Rapoport strata are irreducible in each connected component of F_v ⊗ M in the range 1 < t_v < n−1. The paper justifies this by “the axiom (4c) of Rapoport–He holds in our context by [GLX23, Corollary 1.5]” and then says that the proof of [Zha25a, Proposition 12.24] carries over. No hypotheses of [GLX23, Corollary 1.5] are checked for the present ramified-base-field setting, and the carry-over of [Zha25a, Propositions 12.16, 12.17, 12.24, 12.25] is asserted rather than proved. Footnote 8 explicitly concedes that the reduced KR Z-stratum is not the union of Balloon strata in N^[t]_{(n-1,1),red}; the statement that the construction “carries over directly to unramified extensions F/F0” is not a proof. Since the existence of these cycles is load-bearing for the double induction and for the final extraction of the local identity, this is a major unresolved dependency.","section":"§9.3, Step 4"},{"comment":"The final passage from the global identity 2J_α(ξ0,Φ′) + Int^{K−B}(α,ξ0,Φ) + Int(α,ξ0,Φ) = 0 to the local arithmetic transfer identities is itself a citation: “From this, the local arithmetic transfer identity AT(F_v/F0,v,L_v,g_v,u_v) follows after enlarging d and choosing a suitable transferring pair (Φ,Φ′) as in [MZ24, Proposition 10.2].” Since [MZ24] was written for F0 unramified over Qp, the applicability of its transfer-pair construction to arbitrary p-adic F0 must be checked. In particular, one needs to know that the local test functions f_L and f_{L∨} of Theorem 9.5 arise as local components of standard partial transfers in the sense of [Zha25a, Definition 14.7] for the global data chosen in Step 1. This is not demonstrated in the text, and it is essential for the conclusion that the Fourier coefficient computation actually yields the displayed identities in Q log q.","section":"§9.3, Steps 5–6"}],"minor_comments":[{"comment":"The abstract says the result covers “all p-adic local fields (p odd)” without explicitly adding “for unramified quadratic extensions F/F0”; the body states the correct scope. Please make the abstract match the theorem.","section":"Abstract and Theorem 1.6"},{"comment":"There is a typo in part (2): “not flat in genreal” should be “not flat in general.”","section":"Theorem 3.14"},{"comment":"The proof that ˙F_0 is a σ-linear epimorphism is reduced to a “simple Dieudonné module calculation” with no details. Since this is an essential step in the construction of Φ_S, a fuller argument or a precise statement of the analogous calculation in [KRZ24] would be helpful.","section":"Theorem 4.9 and Lemma 4.11"},{"comment":"The global data are introduced by “we choose” without justification: one needs a CM extension F/F0 with prescribed local behavior at v and split behavior at all other p-adic places, a CM type Φ, and a hermitian space V with a vertex lattice L restricting to L_v. This is likely standard local-global reasoning, but a sentence of justification would make the globalization argument easier to verify.","section":"§9.3, Step 1"},{"comment":"The notation “O ˘Eb⊗O ˘F” is used without definition; it presumably denotes the completed tensor product. Please define it at first occurrence in §5.","section":"§8.2, Corollary 8.14"}],"recommendation":"major_revision","confidential_remarks":"The central theorems are substantial and mostly proved in the text, but the arithmetic transfer proof depends at three successive stages on external results ([Mih22a, Thm. 1.2], [GLX23, Cor. 1.5], [MZ24, Prop. 10.2]) whose hypotheses are not stated or verified in the new ramified-base-field context. The editors may also wish to confirm that the cited works are publicly available in the versions used here, since several are preprints or forthcoming. I would recommend requiring the authors to state the relevant hypotheses and either verify them or provide self-contained proofs for the points listed in the major comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does what it says: it removes the ramification restriction in [Zha25a] and proves arithmetic transfer for all odd p and all p-adic F0. The local model isomorphism (1.4) and the display-theoretic equivalences in Sections 4 and 5 are genuinely new, and they are proved in detail, including duality compatibility. That part is worth the price of admission on its own. The global proof of Theorem 9.5 follows the structure of [Zha25a] but now goes through at ramified primes, and the authors are careful about non-regular integral models and define the global cycles as local complete intersections. No circularity appears: the comparison isomorphisms are derived, not assumed, and the final theorem is not hidden in the hypotheses.\n\nThe soft spots are real but manageable. The stress-test note identifies the right one: Step 1 of §9.3 invokes [Mih22a, Theorem 1.2] for local constancy of orbital integrals and intersection numbers, with the justification being a pointer to [Zha25a, Remark 6.16]. The hypotheses of that theorem are not stated, and the case where the RZ base field is ramified over Q_p is exactly the new territory. If that theorem does not apply, the reduction to the maximal-order case and the Fourier-coefficient extraction in Step 6 collapse. I read the paper's claim as plausible — [Zha25a, Rem. 6.16] presumably covers it — but this needs to be checked by a referee who knows both papers. The same is true for the very special 1-cycles imported from [GLX23, Cor. 1.5]; the paper cites rather than proves the needed smoothness/irreducibility of Kottwitz–Rapoport strata. These are transferred-hypothesis risks, not internal contradictions.\n\nI also note the proof leans heavily on earlier work: [Zha25a], [MZ24], and [GLX23] are used as black boxes. That is normal for a paper at this level, but it means the certificate of correctness is partly inherited rather than self-contained. The appendix on BT and Faltings duality is a genuine addition and, as far as I can tell, fills a real gap in the literature.\n\nWho is this for? Specialists in arithmetic intersection theory, the arithmetic Gan–Gross–Prasad program, and Kudla–Rapoport cycles. If the Mih22a check passes, this settles the conjecture in full generality. I would send it to a serious referee, with the explicit instruction to verify that [Mih22a, Thm 1.2] applies to the RZ spaces of Definition 5.20 when F0/Q_p is ramified. My own verdict is a clear accept once that point is confirmed.","headline":"Strong paper that resolves the arithmetic transfer conjecture in full generality; the only serious risk is a cited local-constancy theorem whose hypotheses are not checked against the ramified-base-field Rapoport–Zink spaces.","tokens_in":72709,"tokens_out":1140,"would_cite":true,"duration_ms":15643,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11G18","14G35","14L05"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every odd-p p-adic local field, arithmetic transfer holds at unramified maximal parahoric level: derivatives of orbital integrals equal, up to a sign and a log q factor, arithmetic intersection numbers on unitary Rapoport–Zink spaces.","keywords":["unitary Shimura varieties","arithmetic transfer","Eisenstein condition","Rapoport-Zink spaces","p-divisible groups","orbital integrals","local models","CM cycles"],"falsifier":"Fix an odd prime p, take F0 to be a ramified extension of Q_p, choose a vertex lattice L of type t=1 in a hermitian space of dimension n=2 or n=3, and compute both sides of the first identity of Theorem 9.5 for a single regular semi-simple matching pair (γ,u1,u2),(g,u) with nonzero orbital-integral derivative; equality in Q log q for that one pair would be a confirmation, and any discrepancy would refute the theorem. A more direct witness is the ξ0-th Fourier coefficient of the modular form Diff(h)^{mod} constructed in Step 6: the paper's argument shows it vanishes, so a computation giving a nonzero coefficient would identify exactly where the proof fails.","tokens_in":71475,"feed_emoji":"🧮","tokens_out":7159,"duration_ms":67165,"temperature":0.7,"pith_summary":"This paper establishes the arithmetic transfer conjecture for unramified maximal parahoric level in full generality: for every p-adic local field F0 with odd residue characteristic, and every unramified quadratic extension F/F0, the derivative of a certain orbital integral attached to the lattice L equals, up to the factor log q, the arithmetic intersection number of Kudla–Rapoport divisors against derived CM cycles on the associated unitary Rapoport–Zink space. The point is to remove the earlier restriction that F0 be unramified over Qp, which had blocked the use of these identities at places with bad reduction. To do that, the paper first constructs integral models of unitary Shimura varieties at ramified primes by reformulating the Eisenstein condition as an (A,B)-strictness condition, compares absolute and relative local models, and lifts those comparisons to categories of p-divisible groups and Rapoport–Zink spaces. If correct, the result supplies the local input needed to push the relative trace formula approach to the arithmetic Gan–Gross–Prasad conjecture through places of bad reduction.","feed_headline":"Arithmetic transfer proved for every odd-p local field","feed_subtitle":"Derived orbital integrals match intersection numbers on unitary Rapoport–Zink spaces at maximal parahoric level.","key_machinery":"The load-bearing mechanism is the (A,B)-strictness condition, a reformulation of the Rapoport–Zink Eisenstein condition: an OF⊗R-submodule F of a free module M is (A,B)-strict if J_B M ⊆ F ⊆ J_A M for certain Eisenstein ideals J_A and J_B determined by a CM type. On the level of p-divisible groups and displays, the paper defines a modification functor Φ_A that multiplies by a lifted Eisenstein element ẽ_S; it divides out the subset A and turns (A,B)-strict objects into ordinary strict or étale ones, preserving duality once the base is restricted to relatively biformal groups. This functor upgrades the comparison of local models to an equivalence between categories of p-divisible groups and hence to isomorphisms of the corresponding Rapoport–Zink spaces; the same comparison is used to transport Kottwitz–Rapoport strata, Kudla–Rapoport divisors and CM cycles between the absolute and relative formal schemes. The arithmetic-transfer proof then runs through the standard globalization scheme: uniformization of the integral model, KR strata and very special 1-cycles, arithmetic Picard groups, and a holomorphic modular form whose vanishing gives the local identity.","core_discovery":"The paper's central claim is Theorem 9.5: for any regular semi-simple matching pair (γ,u1,u2) and (g,u), the derived orbital integrals of the two test functions attached to a vertex lattice L satisfy ∂Orb((γ,u1,u2), f_L) = − fInt_Z(g,u) log q and ∂Orb((γ,u1,u2), f_{L∨}) = −(−1)^t fInt_Y(g,u) log q, where fInt_Z and fInt_Y are arithmetic intersection numbers on the unitary Rapoport–Zink space of signature (n−1,1) and the first identity determines the second by duality. This resolves the arithmetic transfer conjecture [Zha25a, Conjecture 6.4] for all p>2 and all p-adic base fields F0, with F/F0 an unramified quadratic extension, at maximal parahoric level. The proof is obtained by globalizing both sides: a global regular semi-simple pair is chosen close to the local one, the geometric side is expressed through the integral models and special cycles constructed in the first part of the paper, and a difference of holomorphic modular forms is shown to vanish by double modularity and simple modification once the maximal-order case is known.","pith_inferences":["A testable extension of the same mechanism would be to replace the vertex-lattice level with a general lattice chain and prove arithmetic transfer at Iwahori or deeper parahoric level; the comparison theorems of Sections 3–5 are formulated for lattice chains and appear designed for this step.","The (A,B)-strictness reformulation suggests that the local-model comparison is not specific to unitary groups: any PEL datum whose local model is controlled by Eisenstein ideals of the same shape should admit an analogous 'absolute = base change of relative' isomorphism and the corresponding categorical equivalence of displays.","If the cited local-constancy theorem fails in ramified cases, the paper's globalization strategy could likely be repaired by proving the needed constancy directly from the display-theoretic equivalence, since that equivalence is shown to be compatible with Hodge filtrations and duality; the paper does not provide this alternative.","The compatibility of Φ_A with duality, restricted to relatively biformal groups, points toward a full duality-compatible equivalence between absolute and relative special cycles on RZ spaces, and hence toward derived intersection numbers that are independent of the chosen global model."],"forward_implications":["The arithmetic transfer conjecture [Zha25a, Conjecture 6.4] is true at maximal parahoric level for every p-adic local field F0 with p odd and F/F0 unramified quadratic; the identities (1.8) hold for every matching regular semi-simple pair.","RSZ unitary Shimura varieties now have explicit flat, normal integral models at ramified primes with parahoric level, defined through the (A,B)-strictness condition; the previous restrictions on ramification of p in F are removed.","The comparison of Rapoport–Zink spaces via Φ_A is compatible with Kudla–Rapoport divisors, Kottwitz–Rapoport strata, and CM cycles, so intersection numbers and special cycles can be transported between absolute and relative spaces in ramified situations.","Applications that previously required F0 unramified over Q_p—arithmetic Siegel–Weil formulas, arithmetic inner product formulas, and modularity of arithmetic theta series—can now be run at such places, as the paper states in its introduction.","Since the paper also describes how to extend the moduli construction to lattice chains and Drinfeld level structures, the same mechanism is available for other parahoric level situations."],"supporting_citations":[{"why":"Supplies the conjecture (Conjecture 6.4) and the proof in the case F0 is unramified over Qp, including the double-modularity and simple-modification techniques reused here.","marker":"[Zha25a]"},{"why":"Proved the arithmetic fundamental lemma over Qp, which is the base case and the model for the global-local strategy.","marker":"[Zha21]"},{"why":"Extended the arithmetic fundamental lemma to general local fields and provides the framework for decomposing global intersection numbers into local orbital integrals.","marker":"[MZ24]"},{"why":"The local-constancy theorem for orbital integrals and intersection numbers invoked in Step 1 of the globalization argument.","marker":"[Mih22a]"},{"why":"Corollary 1.5 supplies the smoothness and irreducibility axiom for Kottwitz–Rapoport strata used to construct the very special 1-cycles.","marker":"[GLX23]"},{"why":"Defines the RSZ unitary Shimura varieties and their integral models, which the present paper generalizes to ramified primes.","marker":"[RSZ21]"},{"why":"Provides the original Eisenstein condition and the Drinfeld-moduli comparison conjecture, reformulated here as (A,B)-strictness.","marker":"[RZ17]"},{"why":"The Ahsendorf–Cheng–Zink equivalence between nilpotent O-displays and formal p-divisible O-modules underlies the categorical results in Part 1.","marker":"[ACZ16]"},{"why":"Previous cases of the p-divisible-group equivalence and the Ahsendorf-functor construction are used in the uniform treatment of the modification functor.","marker":"[KRZ24]"}],"fun_headline_variants":["Arithmetic transfer resolved for all odd p","Full arithmetic transfer proof for unitary Shimura varieties","Unitary Shimura varieties: transfer conjecture proved","Ramified primes mastered: arithmetic transfer proven","Odd p local fields: arithmetic transfer holds fully"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a local-constancy theorem for orbital integrals and intersection numbers, originally proved for unramified settings, applies to the Rapoport–Zink spaces over possibly ramified base fields used in the globalization step; if it does not, the reduction to the maximal-order case and the final Fourier-coefficient comparison collapse.","fun_headline_variants_meta":{"raw":{"variants":["Arithmetic transfer resolved for all odd p","Full arithmetic transfer proof for unitary Shimura varieties","Unitary Shimura varieties: transfer conjecture proved","Ramified primes mastered: arithmetic transfer proven","Odd p local fields: arithmetic transfer holds fully"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000551,"raw_usage":{"total_tokens":2634,"prompt_tokens":959,"completion_tokens":1675,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":575,"completion_tokens_details":{"reasoning_tokens":1606}},"tokens_in":575,"tokens_out":1675,"duration_ms":10964,"temperature":1.0,"reasoning_tokens":1606,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:39:20.820794+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix an odd prime p, take F0 to be a ramified extension of Q_p, choose a vertex lattice L of type t=1 in a hermitian space of dimension n=2 or n=3, and compute both sides of the first identity of Theorem 9.5 for a single regular semi-simple matching pair (γ,u1,u2),(g,u) with nonzero orbital-integral derivative; equality in Q log q for that one pair would be a confirmation, and any discrepancy would refute the theorem. A more direct witness is the ξ0-th Fourier coefficient of the modular form Diff(h)^{mod} constructed in Step 6: the paper's argument shows it vanishes, so a computation giving a nonzero coefficient would identify exactly where the proof fails.","supporting_citations":[],"review_version":1}