{"id":"48fc91d3-51e3-4004-929d-5bd74d3ef4c8","arxiv_id":"2505.12777","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper constructs, for every concave-function Bruhat-Tits group scheme over a discrete valuation ring, a unique smooth quasi-projective wonderful embedding whose generic fiber is the classical wonderful compactification, with a toroidal special fiber and a freely generated Picard group.","lead":"For reductive groups over discretely valued Henselian fields, this paper builds complete, smooth compactifications (wonderful embeddings) of the integral group schemes that Bruhat-Tits theory attaches to concavity-filtered subgroups. The construction is intrinsic and functorial, works for quasi-split and, by descent, for all adjoint groups, and matches the known behavior of classical wonderful compactifications.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The SU3 commutator integrality in Lemma 6.3 is the load-bearing computational step; a sign or valuation error would break Theorem 6.5 and the quotient construction.","rationale":"I read the paper in good faith and found no internal contradiction. The strategy is coherent: the intrinsic Artin–Weil-style quotient construction is a plausible way to avoid embedding into an ambient space, and the paper gives independent consistency checks (generic fiber is the classical wonderful compactification, special fiber is a toroidal embedding, Picard group has the expected generators). The main theorems are stated precisely, and the proof is detailed. However, the existence of the integral wonderful embedding in Theorem 1.1 depends on the rational action A in Theorem 6.5 being well defined on the specific o-points U_{Φ+,x,f+}(o)×T(o)×U_{Φ−,x,f+}(o). That domain statement is proved via the commutator integrality estimates of Lemma 6.3, especially the SU3 case, the distribution criterion Corollary B.4, and the induction in Proposition 6.4. These estimates are intricate, hand-verified, and not machine-checked. The SU3 computation is the most delicate: the displayed Gauss decomposition involves a denominator ε whose unit property is asserted somewhat tersely, and the distribution step relies on a differential inclusion that is easy to get wrong. The reader's verdict is CONDITIONAL with moderate confidence, and my read does not move it: the concern is real but not demonstrated to be a flaw. A concrete recomputation of the SU3 identity and unit claim would settle whether the load-bearing premise holds. I agree with the reader's weakest_assumption; my only sharpening is to point to the exact SU3 denominator/unit issue and the unverified distribution inclusion as the place to test.","tokens_in":44831,"tokens_out":12464,"duration_ms":135944,"concrete_test":"Independently recompute the SU3 case of Lemma 6.3 with a computer algebra system: verify that the displayed Gauss decomposition of y_+(u,v)·z(t)·y_−(u′,v′) is correct with the stated ε and with the stated powers of m_{−a}(t) and σ(m_{−a}(t)), and verify that ε is a unit in o_α whenever ω(u)>f(a)+γ, ω(u′)>f(−a)+γ, ω(v)>f(2a), ω(v′)>f(−2a), with uσ(u)=v+σ(v). If the identity or the unit claim fails in any valuation configuration, Theorem 6.5(1) and the construction of Section 6.3.1 collapse.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on Theorem 6.5: the o-rational action A of G_{x,f}×o G_{x,f} on the big cell Ω_{x,f} must have definition domain containing U_{Φ+,x,f+}(o)×T(o)×U_{Φ−,x,f+}(o). The proof of Theorem 6.5 passes through Lemma 6.3, and the SU3 case (Case 2) is the least secure part. There, the product χ_+(u,v)·t·χ_−(u′,v′) is rewritten as χ_−(A,B)·T_a(ε)·t·χ_+(C,D), with ε = 1 − m_{−a}(t)σ(u)u′ + m_{−a}(t)σ(m_{−a}(t))vv′, and the displayed formulas for A and C involve division by ε or σ(ε). The valuation estimates showing d_a ∈ Γ(U_{-a,x,f}×T×U_{a,x,f}) are plausible, but the stronger statement needed for condition (1), namely that ε is a unit in o_α when u,u′,v,v′ lie in the strictly larger f+ subgroups, is not written out. In the SU3 case this unit claim depends on the quadratic relation uσ(u)=v+σ(v), and a sign or valuation error there would destroy the invertibility of ε, hence the definition domain D(d_a). Moreover, Proposition 6.4(2) uses Corollary B.4 to convert a distribution inclusion into an o-integral morphism; this inference is only as reliable as the differential computation in the induction step, which is not machine-checked. If Lemma 6.3 fails, the definition domain of A in Theorem 6.5 is not established, and the quotient sheaf construction of Section 6.3.1, and with it Theorem 1.1, collapses. This is a fragility of the computational core, not an observed contradiction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper constructs, for a connected quasi-split adjoint semisimple group G over a strictly Henselian discretely valued field k and a concave function f on the relative root system, a canonical smooth quasi-projective integral model \\overline{G}_{x,f} over o of the wonderful compactification \\overline{G}, extending the Bruhat--Tits group scheme G_{x,f}. The construction is intrinsic: it avoids embedding into an ambient scheme and instead builds \\overline{G}_{x,f} as a quotient of G_{x,f}\\times\\Omega_{x,f}\\times G_{x,f} by an equivalence relation attached to a rational action, following the Artin--Weil method. The main technical step is Theorem 6.5, which asserts that the rational action has a large definition domain; its proof rests on commutator estimates in Lemma 6.3 and a distribution computation in Proposition 6.4. The paper also proves that the special fiber is an equivariant toroidal embedding when f(0)=0, that the boundary is a union of smooth o-relative normal-crossing divisors, that the Picard group is freely generated by the divisors D_\\alpha, that the construction is compatible with dilatation, and that it descends by \\'etale descent to non-quasi-split groups (Theorem 1.3).","tokens_in":45034,"tokens_out":11492,"duration_ms":135103,"significance":"If the main results are correct, this is a substantial bridge between Bruhat--Tits theory and the theory of wonderful compactifications: it provides a canonical integral model of the wonderful compactification for every concave-function group scheme, with the classical wonderful compactification as generic fiber and a toroidal embedding as special fiber. The intrinsic quotient construction is a genuine methodological departure from the usual closure-in-an-ambient-space approach. The paper also gives explicit structural results (boundary divisors, Picard group, dilatation compatibility, descent) and carefully benchmarks them against known statements: the classical wonderful compactification, the toroidal-embedding classification, and the standard integral models of Bruhat--Tits theory. The main risk is that the computational core, especially the SU3 commutator integrality in Lemma 6.3, contains a load-bearing verification that is not actually written out.","major_comments":[{"comment":"The proof establishes d_a\\in\\Gamma(U_{-a,x,f}\\times T\\times U_{a,x,f}) by showing that the quantity \\epsilon lies in o_\\alpha for f-level points. However, condition (1) of Lemma 6.3 requires U_{a,x,f+}(o)\\times T(o)\\times U_{-a,x,f+}(o)\\subset D(d_a), which means that Norm_{k_\\alpha/k}(\\epsilon) must be a unit on those f+-level points. No such unit estimate is supplied: the displayed valuation inequalities in Case 2 are only for f-level points, and the quadratic relation u\\sigma(u)=v+\\sigma(v) is not used to control \\epsilon modulo the maximal ideal. Since \\epsilon appears in the denominator in the definition of \\theta_a, and since D(d_a) is by definition the locus where Norm(\\epsilon) is invertible, this missing unit estimate is exactly what is needed for Theorem 6.5(1) and hence for the quotient construction in Section 6.3.1.","section":"Section 6.2.2, Lemma 6.3, Case 2 (SU3)"},{"comment":"The proof that \\theta_a extends to an o-morphism from D(d_a) is not carried out. The text says that by Proposition 6.1 it suffices to check an inclusion of o-points and that this follows from the definition together with Corollary 6.2, but the displayed quotients in \\theta_a have denominators \\epsilon and \\sigma(\\epsilon), and no valuation argument for these quotients is written. This is not a cosmetic omission, because the same estimates control whether \\beta_a maps U_{-a,x,f+}\\times T\\times U_{a,x,f+} into the required f+-subgroups, which is used in Lemma 6.3(3) and then in Proposition 6.4(2). The extension claim therefore needs a complete verification, not an appeal to the extension principle alone.","section":"Section 6.2.2, Lemma 6.3, Case 2 (extension of \\theta_a)"},{"comment":"The proof of Theorem 6.15 is written only for the case f(0)=0. Section 6.4 explicitly assumes f(0)=0 in its opening paragraph, Claim 6.17 uses the special-fiber decomposition obtained in that section, and the quoted local factoriality is based on Proposition 6.6, which is proved for f(0)=0. When f(0)>0 the special fiber is unipotent and the boundary geometry is different. Since Theorem 6.15 is stated without an f(0) restriction, the proof needs a separate argument for f(0)>0, or the statement should be explicitly restricted to f(0)=0.","section":"Section 6.9 and Theorem 6.15 (Picard group for f(0)>0)"},{"comment":"The descent step asserts that the morphism displayed in Equations (20)--(21) induces an O-automorphism of \\overline{G}_{x,\\tilde{f}}. Since \\overline{G}_{x,\\tilde{f}} is constructed as a quotient by the equivalence relation of Definition 5.1, one must prove that this morphism is compatible with the equivalence relation generated by the rational action A and that the induced endomorphism is invertible. The present text only states this compatibility. Because Theorem 1.3 depends on this step, a detailed proof or a precise reference is needed.","section":"Section 7.3, Theorem 7.1 (descent datum)"}],"minor_comments":[{"comment":"The symbol m'_{-a} is used in the displayed formula for \\theta_a and in the last sentence of Case 2 without being defined; it should presumably be m_{-a'} for the Galois-conjugate root a', and this should be stated explicitly.","section":"Section 6.2.2, Lemma 6.3, Case 2"},{"comment":"There is a typo in the sentence 'the kernel of the restriction ... dose not contain': 'dose' should be 'does'.","section":"Section 3.2"},{"comment":"The use of [BLR90, \\S2.5, Proposition 5] to conclude that the rational map \\tau is defined on all of \\overline{G}_1 is terse; a sentence explaining why the hypotheses of that descent statement apply to a rational map rather than to an ordinary morphism would improve the exposition.","section":"Section 6.6, uniqueness proof"},{"comment":"In the sentence 'it is nothing else but the hyperplane cut out by X_a', the justification is abbreviated: since the coordinate ring of \\Omega_{x,f} is a UFD, a reader can fill in the argument, but a short explanation of why the given irreducible stable hypersurface must be the zero locus of the coordinate X_a would make the proof more transparent.","section":"Section 6.5, Proposition 6.13"}],"recommendation":"major_revision","confidential_remarks":"The paper is a substantial contribution and I do not see a fatal internal contradiction. The main issue is that the computational core of the proof, especially the SU3 case of Lemma 6.3, contains a missing unit/integrality verification that is load-bearing for Theorem 6.5 and hence for the existence of \\overline{G}_{x,f} in the quasi-split case. This is likely fixable with a detailed valuation computation, but it needs to be written out. I would also ask the editor to confirm the status of the author's preprint [Li23], since the identification of the generic fiber with the classical wonderful compactification relies on it. The notation issue with m'_{-a} should be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Shang Li has written a serious paper. For concave-function group schemes G_{x,f} attached to a reductive group over a discretely valued field, it constructs a canonical smooth quasi-projective group scheme compactification whose generic fiber is the classical wonderful compactification and whose special fiber is a toroidal embedding, computes the Picard group, and extends the construction by étale descent beyond the quasi-split case. That set of results is new. The method is also genuinely different from the classical ambient-closure route: it forms a quotient sheaf from a rational action, in the style of Artin–Weil birational group laws. The proofs are detailed, and the paper is honest about what is open.\n\nThe reader's stress-test note points at the right place. Lemma 6.3, especially the SU3 case, is the critical computational step. The object ε = 1 − m_{−a}(t)σ(u)u′ + m_{−a}(t)σ(m_{−a}(t))vv′ must be a unit in the f+ range, and the paper does not fully write out that unit check; the quadratic relation uσ(u)=v+σ(v) is where this could fail. A referee should verify that computation line by line. The rest of the construction depends on it through Theorem 6.5 and the quotient sheaf in Section 6.3.1. Nothing in the paper is machine-checked, so this is genuine fragility, not an observed contradiction. Additionally, Theorem 1.1 and Theorem 1.3 rely on the author's previous preprint [Li23]. That is openly cited and not circular, but it does mean the full proof is not self-contained.\n\nI did not find an internal inconsistency, and the external benchmarks are used properly: the generic fiber is checked against the classical wonderful compactification, the special fiber against the toroidal embedding classification, and the Picard group against known behavior. The open questions in Remark 1.4 are stated as questions, not dressed up as theorems.\n\nWho should read this? People working on Bruhat–Tits integral models, wonderful compactifications, and nearby arithmetic geometry. It deserves a real refereeing process: the paper should be sent to a knowledgeable referee, with explicit instruction to check Lemma 6.3 and the dependence on [Li23]. If those hold, the result is an important advance.","headline":"A substantively new conditional construction in Bruhat–Tits theory; the unverified SU3 commutator step is the main fragility, and the paper deserves serious refereeing.","tokens_in":45734,"tokens_out":4984,"would_cite":true,"duration_ms":48757,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14L15","20G25","14M27"],"pacs":[],"model":"deepseek-v4-flash","headline":"Every concave-function Bruhat–Tits group scheme over a discretely valued field admits a canonical smooth quasi-projective wonderful compactification over the ring of integers, uniquely determined by its big cell.","keywords":["Bruhat–Tits theory","wonderful compactification","concave functions","integral models","group schemes","étale descent","toroidal embeddings","Picard group"],"falsifier":"Compute the element $\\epsilon=1-m_{-a}(t)\\sigma(u)u'+m_{-a}(t)\\sigma(m_{-a}(t))vv'$ appearing in Lemma 6.3 for a ramified quadratic extension, with $u,u',v,v'$ at the boundary of the root filtrations allowed by some concave function $f$; if for some such choice $\\omega(\\mathrm{Norm}(\\epsilon))<0$, then the proposed integral model $d_a$ is not integral, Lemma 6.3 is false, and the rational-action construction of Theorem 1.1 fails. This is a finite valuation computation once the pinning and uniformizers are fixed.","tokens_in":44419,"feed_emoji":"📐","tokens_out":12262,"duration_ms":115301,"temperature":0.7,"pith_summary":"This paper claims that every Bruhat–Tits group scheme attached to a concave function $f$ and a point $x$ in the apartment of an adjoint quasi-split reductive group admits a canonical wonderful compactification over the ring of integers. The compactification is uniquely determined by a big cell: it contains the group scheme and the big cell as open subschemes, and the double action of the group scheme on the big cell generates the whole model. The construction does not realize the compactification as a closure in an ambient projective space; instead it forms a quotient by a rational action, following the birational group-law method. This matters because these group schemes provide integral models of parahoric and higher congruence subgroups, and a wonderful compactification lets the geometry of wonderful varieties be used in their arithmetic and representation-theoretic study. For the quasi-split adjoint case the paper also shows the special fiber is an equivariant toroidal embedding and that the Picard group is freely generated by the affine analogs of the boundary divisors.","feed_headline":"Bruhat–Tits groups get canonical wonderful compactifications","feed_subtitle":"New integral compactifications bring wonderful-variety geometry to p-adic representation theory.","key_machinery":"The central object is the quotient of the triple product $G_{x,f}\\times_o\\Omega_{x,f}\\times_o G_{x,f}$ by the equivalence relation generated by a rational action $A: G_{x,f}\\times_o\\Omega_{x,f}\\times_o G_{x,f}\\dashrightarrow\\Omega_{x,f}$. This is the birational group-law mechanism: rather than taking a schematic closure in an ambient space, one proves that the rational double action on the big cell gives an algebraic space, then uses quasi-projectivity to conclude it is a scheme. The load-bearing input is Theorem 6.5, which asserts that the definition domain of $A$ contains $U_{\\Phi_+,x,f+}(o)\\times T(o)\\times U_{\\Phi_-,x,f+}(o)$, so that the generic group law extends integrally. That theorem is proved by explicit commutator estimates in the $\\mathrm{SL}_2$ and $\\mathrm{SU}_3$ cases, the extension principle for smooth affine models, and a distribution-theoretic criterion that converts integrality of the induced differential on distributions into an integral morphism.","core_discovery":"The central claim is that an integral wonderful compactification $\\bar G_{x,f}$ exists over $o$, is unique, and is pinned down by its big cell. For a quasi-split adjoint $G$ over a strictly Henselian discretely valued field $k$, the paper constructs a smooth quasi-projective $o$-scheme $\\bar G_{x,f}$ whose generic fiber is the wonderful compactification of $G$; the open immersion $G\\hookrightarrow \\bar G$ extends to a $(G_{x,f}\\times_o G_{x,f})$-equivariant open immersion $G_{x,f}\\hookrightarrow \\bar G_{x,f}$, the canonical big cell $\\Omega = U^{-}\\times_k\\prod_{\\Delta}\\mathbb{A}^1_k\\times_k U^+$ extends to $\\Omega_{x,f}=U^{-}\\times_o\\prod_{\\Delta}A^{f(0)}_{1,o}\\times_o U^+$, and $(G_{x,f}\\times_o G_{x,f})\\cdot\\Omega_{x,f}=\\bar G_{x,f}$. When $f(0)=0$, the special fiber is the equivariant toroidal embedding of the maximal reductive quotient determined by the cone $C$ inside the negative Weyl chamber $C_{x,f}$; the boundary is covered by smooth relative effective Cartier divisors $S_\\alpha$ with relative normal crossings; and the Picard group is freely generated by the divisors $D_\\alpha=\\overline{B\\,\\dot s_\\alpha\\,B^{-}}$, for $\\alpha\\in\\Delta$. The same construction descends étale-locally to arbitrary adjoint groups over a Henselian discretely valued field. If these theorems are correct, every concave-function Bruhat–Tits group scheme has a canonical wonderful compactification that behaves like the classical one.","pith_inferences":["A natural test of the construction is to compute the intersection theory of $\\bar G_{x,f}$ in small examples such as $\\mathrm{PGL}_2$, where the models are explicit ($\\mathbb{P}^3_o$ or a simple gluing); the Picard and boundary-divisor statements should be directly verifiable there.","The quotient-by-rational-action formalism is written for a general flat group scheme acting on a scheme, so it likely constructs integral wonderful models for broader classes of wonderful varieties over $o$, not only the group schemes themselves.","The paper raises the question of an affine analogue of the classification of $B\\times B$-orbit closures: describing the orbits of parahoric subgroups on $\\bar G_{x,f}(o)$ by discrete filtration data would extend the classical Bruhat picture to the integral setting.","If the toroidal special-fiber description is correct, the set of values $\\Gamma_\\alpha$ and the cone $C$ should control the boundary strata and their normal bundles, giving a combinatorial dictionary between concave functions and compactification geometry."],"forward_implications":["Every parahoric and higher-congruence Bruhat–Tits group scheme now carries a canonical integral wonderful compactification, so the tools of wonderful-variety geometry apply over the ring of integers.","In the equilevel case $f(0)=0$, the special fiber is an explicit equivariant toroidal embedding of the maximal reductive quotient, making the combinatorial data of the concave function and the point visible in the boundary geometry.","The Picard group of the compactification is free abelian of rank $|\\Delta|$ with generators $D_\\alpha$, so the classical color-divisor computation for wonderful compactifications has a direct arithmetic analogue.","The compactifications are compatible with dilatation: when $g\\le f\\le g+1$, $\\bar G_{x,f}$ is the dilatation of $\\bar G_{x,g}$, so varying the concave function produces a ladder of compactifications linked by blow-up-like operations.","Beyond the quasi-split case, the construction is obtained by étale descent and is projective over $o$ exactly when the underlying Bruhat–Tits group scheme is reductive."],"supporting_citations":[{"why":"Establishes the existence and uniqueness of the smooth affine integral model $G_{x,f}$ with its big cell, and supplies the commutator and valuation estimates adapted in Lemma 6.3.","marker":"[BT84a]"},{"why":"Provides the extension principle for smooth affine models and the birational-group-laws proof of the integral model, both used to build the rational action.","marker":"[Lan96]"},{"why":"Modern reference for concave functions, root filtrations, unipotent radicals of special fibers, and the integral-model facts used throughout.","marker":"[KP23]"},{"why":"Constructs $G_{x,f}$ by dilatation when $f(0)>0$, which the paper uses for the gluing case and for compatibility with dilatation.","marker":"[Yu15]"},{"why":"Supplies the intrinsic quotient-by-rational-action method for relative wonderful compactifications, reformulated in Section 5 for the $o$-rational action.","marker":"[Li23]"},{"why":"Gives the classification of toroidal embeddings and the classical Picard/color results that identify the special fiber and the divisors $D_\\alpha$.","marker":"[BK05]"},{"why":"Provides the dilatation, flatness, and lifting lemmas behind smoothness, quasi-projectivity, and boundary-divisor arguments.","marker":"[BLR90]"},{"why":"Defines concave functions on the relative root datum and the root filtrations from which the groups and their integral models are built.","marker":"[BT72]"}],"fun_headline_variants":["Intrinsic wonderful compactifications for Bruhat-Tits groups","Canonical wonderful compactifications without ambient closure","Intrinsic wonderful compactifications for p-adic groups","Functorial wonderful embeddings for Bruhat-Tits schemes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the rational double action of $G_{x,f}\\times_o G_{x,f}$ on the big cell $\\Omega_{x,f}$ extends over $o$ with a definition domain containing $U_{\\Phi_+,x,f+}(o)\\times T(o)\\times U_{\\Phi_-,x,f+}(o)$; that extension rests on intricate integrality estimates for commutators, especially the $\\mathrm{SU}_3$ case, and on the distribution criterion connecting generic differentials to integral morphisms, so if any of those estimates fails the quotient construction and Theorem 1.1 collapse.","fun_headline_variants_meta":{"raw":{"variants":["Intrinsic wonderful compactifications for Bruhat-Tits groups","Canonical wonderful compactifications without ambient closure","Intrinsic wonderful compactifications for p-adic groups","Functorial wonderful embeddings for Bruhat-Tits schemes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000785,"raw_usage":{"total_tokens":3556,"prompt_tokens":1126,"completion_tokens":2430,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":742,"completion_tokens_details":{"reasoning_tokens":2366}},"tokens_in":742,"tokens_out":2430,"duration_ms":20160,"temperature":1.0,"reasoning_tokens":2366,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:28:00.538928+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the element $\\epsilon=1-m_{-a}(t)\\sigma(u)u'+m_{-a}(t)\\sigma(m_{-a}(t))vv'$ appearing in Lemma 6.3 for a ramified quadratic extension, with $u,u',v,v'$ at the boundary of the root filtrations allowed by some concave function $f$; if for some such choice $\\omega(\\mathrm{Norm}(\\epsilon))<0$, then the proposed integral model $d_a$ is not integral, Lemma 6.3 is false, and the rational-action construction of Theorem 1.1 fails. This is a finite valuation computation once the pinning and uniformizers are fixed.","supporting_citations":[],"review_version":1}