{"id":"6172d8c1-1032-443c-a7a8-848133e05dc5","arxiv_id":"2507.16133","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The general torus orbit closure in OG(n,2n+1) degenerates explicitly into Richardson varieties, giving [Z] = sum_{I in [n-1]} sigma_I sigma_{I^c}.","lead":"This paper constructs an explicit degeneration of general torus orbit closures in the maximal orthogonal Grassmannian into Richardson varieties, and derives a cohomology class formula from it. It matters because it connects torus orbit geometry with delta-matroids and provides a concrete tool for enumerative problems in type B.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of dominance of pr_r (Lemma 4.2.5) is the load-bearing step; the sparsity argument does not explicitly establish that the specific rational map has Zariski-dense image, and the recursive nicest-locus induction depends on it.","rationale":"The paper gives substantial independent support: explicit matrix charts for the relevant Richardson varieties, a birationality statement for the coordinate maps, a polytope-decomposition proof, and a lattice-index argument for multiplicity 1. I did not find an internal inconsistency elsewhere. The single least secure point is the dominance of pr_r and the resulting non-emptiness of the recursive nicest locus. The Reader identified the same point, so my agreement is 'agree'. I would not reject the paper: the dominance claim is plausible and likely fixable, and the other components of the proof are independently checkable. But because the nicest locus is the induction that creates every Richardson component, a gap here is not cosmetic. Hence I recommend CONDITIONAL acceptance pending the Jacobian/elimination check of Lemma 4.2.5.","tokens_in":40147,"tokens_out":29750,"duration_ms":331969,"concrete_test":"Implement pr_r from §4.2 for all non-saturated allowed pairs with n=4 (preferably n=5): choose generic non-zero +-entries in C^{d(I,I′)}, compute the output tuple in C^{d(I_+,I′)} by the prescribed zeroing and row-reduction, and evaluate the rank of the Jacobian ∂pr_r/∂a at that point. Full rank equal to d(I_+,I′) for every pair verifies dominance and confirms Lemma 4.2.5; alternatively, compute the Zariski closure of the image by elimination and compare dimensions. A rank drop or dimension deficiency would make the nicest-locus induction collapse and would invalidate the construction of the degeneration for general Λ.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The recursive construction in §4 can only start from a 'general' Λ because Definition 4.2.6 defines the nicest locus on Σ_{I,I′} by pulling back the nicest loci of both children. For this to be non-empty on the root and every intermediate vertex of T_n, both child maps must be dominant. pr_ℓ is a coordinate projection, but pr_r is a rational map, and its dominance is exactly Lemma 4.2.5. The proof of Lemma 4.2.5 passes to a j×(j+1) submatrix, assigns input/output sparsity patterns, and concludes dominance from the fact that a general (j−1)-sparse matrix is row-equivalent to an ℓ-sparse one by moving bottom rows to the top. Row-equivalence of sparsity classes is not the same as domination by the specific map pr_r, which first zeros prescribed entries and then applies the unique row reduction that normalizes rightmost non-zero entries; no Jacobian or elimination computation is given to show the image of this particular map is Zariski dense. The case j=0, which occurs for every type-(iv) allowed pair, is not really covered by the j×(j+1) argument. If the image of pr_r had dimension less than d(I_+,I′) for any pair appearing in T_n, the nicest locus would be empty on some Σ_{I,I′}, and no general Λ could seed the degeneration that produces all 2^{n−1} Richardson components in Theorem 1.0.2.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs an explicit, T-invariant, embedded degeneration of the closure Z_Λ of the general torus orbit in the maximal orthogonal Grassmannian OG(n,2n+1) into a union of Richardson varieties Σ_{I,I^c}, one for each I⊂[n−1], each appearing with multiplicity one. The degeneration is built iteratively from explicit matrix charts for Richardson varieties, and the absence of extra components and the multiplicity-one statement are proved via moment polytopes: the polytopes P(Λ_I) of the limit pieces cover the unit hypercube, and the associated lattices coincide. From this degeneration the authors deduce the cohomology class formula [Z_Λ]=Σ_{I⊂[n−1]} σ_I σ_{I^c} in H^{n(n−1)}(OG(n,2n+1)). The paper also proves, in an appendix, the equivalence of two standard definitions of the base polytope of a realizable delta-matroid.","tokens_in":40443,"tokens_out":40483,"duration_ms":405581,"significance":"The result is significant: it gives a new, explicit degeneration proof of a type-B analogue of the Berget–Fink/Anderson–Tymoczko formula for general torus orbit closures, and it connects the cohomological formula to a polyhedral decomposition of the hypercube previously studied by Chen–Sanchez–Veliz–Ying. The constructions are concrete and largely self-contained: the charts for Richardson varieties are proved by explicit matrix manipulation, the degeneration is written down entry-by-entry, and the polytope cover and multiplicity statements are proved from rank inequalities and lattice indices rather than imported from localization or assumed cohomology. I specifically checked the potential weak point flagged during review, the dominance of the rational map pr_r in Lemma 4.2.5; on inspection the row-move argument is valid, although the proof is terse and the case j=0 is not explicitly separated. I found no load-bearing mathematical errors; the remaining requests are for clarification and small corrections.","major_comments":[],"minor_comments":[{"comment":"The proof of dominance of pr_r is quite terse. The assertion that moving the bottom (j−1)−ℓ rows of a (j−1)-sparse matrix to the top produces an ℓ-sparse matrix is correct, but it deserves a short anti-diagonal calculation; without it the reader cannot easily verify that the new first ℓ anti-diagonals are zero and the (ℓ+1)-st is non-zero. In addition, the case j=0 is not covered by the j×(j+1) submatrix argument; for j=0 the map pr_r is a coordinate projection, and this should be stated explicitly.","section":"§4.2, Lemma 4.2.5"},{"comment":"The proof would also benefit from one sentence explaining why row-equivalence to an ℓ-sparse matrix implies dominance of the specific row-reduction map pr_r: after the initial zeroing step, pr_r is the canonical row-reduction map, so a preimage of a general target matrix is obtained by row-reducing an ℓ-sparse matrix row-equivalent to it.","section":"§4.2, Lemma 4.2.5"},{"comment":"The last leaf of the tree T_4 is typeset as \"({∅, {1, 2, 3}\" in the text; this appears to be a typo, and it should presumably read \"({1,2,3},∅)\" or \"(∅,{1,2,3})\" according to the intended leaf.","section":"Example 3.1.5"},{"comment":"The sentence \"Propositions 5.2.5 and 5.2.5 together imply Proposition 5.2.1\" contains a duplicated reference; the second should be Proposition 5.2.6.","section":"After Proposition 5.2.6"},{"comment":"The statement that the limit components of all intermediate steps must also appear with multiplicity 1 is compressed; adding one sentence explaining that a multiplicity greater than 1 at an intermediate degeneration would force the corresponding leaf multiplicity in the final special fiber to exceed 1 would improve readability.","section":"Proof of Theorem 1.0.2"}],"recommendation":"minor_revision","confidential_remarks":"The paper appears mathematically sound after my review. The only substantive request is to expand the proof of Lemma 4.2.5 and explicitly handle the j=0 case; the remaining issues are typographical. I do not see concerns about novelty, citation practice, or fit with the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main thing to know: this paper is worth taking seriously. It constructs an explicit T-invariant degeneration of general torus orbit closures in OG(n,2n+1) into Richardson varieties, with multiplicity one for each component, and derives the class formula [Z] = sum_I sigma_I sigma_{I^c}. The type A analog was in the literature, but the type B case requires controlling the quadratic form through the matrix charts, and Sections 3-4 do that concretely. The polytope cover in Section 5 and the lattice-index argument for multiplicity one are self-contained and, as far as I can tell, correct. The paper is also honest about the overlap with the forthcoming KSSB work.\n\nWhere I would press the authors: Lemma 4.2.5, the dominance of pr_r. The proof argues with sparsity patterns and row-equivalence, but pr_r is not a row-equivalence map: it first zeros prescribed entries and then applies row reduction. Showing that a general (j-1)-sparse matrix is row-equivalent to some ell-sparse input does not, by itself, show that this particular composite map is dominant. The stress-test note makes exactly this complaint, and I think it lands. The j=0 case is not a real concern—there pr_r is a coordinate projection—but for j>0 the proof is incomplete as written. I do not think the lemma is false; it is likely fixable by a direct elimination computation on the inner box, and the rest of the paper's structure gives me confidence. Still, because the non-emptiness of the nicest locus, and hence the whole degeneration, depends on this dominance, a referee should ask for a genuinely rigorous proof.\n\nBottom line: for people in Schubert calculus, torus orbit closures, or delta-matroids, this is a useful and citable paper. The main theorem is believable and well-supported except for the gap in 4.2.5. I would send it to peer review with a request for clarification rather than desk reject. If the authors fill that gap, it should be a clean acceptance.","headline":"A genuinely useful, explicit type-B degeneration with a class formula, but Lemma 4.2.5 (dominance of pr_r) is under-proved as written and needs referee attention.","tokens_in":40984,"tokens_out":6885,"would_cite":true,"duration_ms":75190,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14M15","14M25","05B35","14N15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs an explicit, T-invariant degeneration of a general torus orbit closure in the maximal orthogonal Grassmannian OG(n,2n+1) into a union of Richardson varieties, each appearing once, and derives the cohomology class of…","keywords":["delta-matroids","orthogonal Grassmannian","torus orbit closures","Richardson varieties","toric degenerations","Schubert calculus","moment polytopes","polyhedral decompositions"],"falsifier":"Compute the equivariant localization of $[Z_\\Lambda]$ on $\\mathrm{OG}(3,7)$ for a general $\\Lambda$ and compare it with $\\sum_{I\\subset\\{1,2\\}}\\sigma_I\\sigma_{I^c}$; any disagreement refutes Theorem 1.0.1. Equivalently, find a point $x\\in[0,1]^3$ whose associated subset $I$ from Definition 5.1.2 violates the inequality $x(S)\\le g_{\\Lambda_I}(S)$ for a nice $\\Lambda_I$, which would refute Proposition 5.1.1.","tokens_in":39927,"feed_emoji":"🧊","tokens_out":9578,"duration_ms":101183,"temperature":0.7,"pith_summary":"This paper establishes a type-B analogue of the classical story that torus orbit closures in Grassmannians degenerate into unions of Schubert and Richardson varieties. For a general point $\\Lambda$ of the maximal orthogonal Grassmannian $\\mathrm{OG}(n,2n+1)$, the closure $Z_\\Lambda$ of its $n$-dimensional torus orbit is shown to admit an explicit embedded degeneration into the union of the Richardson varieties $\\Sigma_{I,I^c}$ over all subsets $I\\subset[n-1]$, each appearing with multiplicity one. From this degeneration the authors deduce that the cohomology class of $Z_\\Lambda$ equals $\\sum_{I\\subset[n-1]}\\sigma_I\\sigma_{I^c}$, a Schubert calculus formula in $H^{n(n-1)}(\\mathrm{OG}(n,2n+1))$. The proof works by degenerating matrix entries one at a time while preserving isotropy with respect to the underlying quadratic form, and the moment-map images of the pieces give a polyhedral decomposition of the unit hypercube $[0,1]^n$ by delta-matroid base polytopes. If correct, the result provides the type-B counterpart of the Berget–Fink and Anderson–Tymoczko class formulas and makes the toric geometry of these orbit closures explicitly computable.","feed_headline":"General torus orbit in OG(n,2n+1) degenerates to Richardson varieties","feed_subtitle":"The degeneration yields the cohomology class as a sum of Schubert products and tiles the hypercube by delta-matroid polytopes.","key_machinery":"The central object is the rooted binary tree $T_n$ of allowed pairs $(I,I')$ of disjoint subsets of $[n-1]$ (Definition 3.1.4), together with the explicit coordinate matrices $M_{I,I'}$ that chart the Richardson variety $\\Sigma_{I,I'}$. Each non-saturated vertex has a left child $(I,I'_+)$ and a right child $(I_+,I')$; degenerating the active entry of a 'nicest' matrix to zero realizes the flat limit as the union of the torus orbit closures of general points of the two children. The birational maps that forget plus-entries and take row spans give coordinates on the Richardson varieties, and the moment polytope of each piece is the base polytope of its delta-matroid. These base polytopes cover the unit hypercube $[0,1]^n$, which rules out extra limit components, and a lattice-index comparison shows each component has multiplicity one.","core_discovery":"The central discovery is that the general torus orbit closure $Z_\\Lambda\\subset\\mathrm{OG}(n,2n+1)$ can be flattened, in an explicitly constructed one-parameter family, into a reduced union of exactly the Richardson varieties $\\Sigma_{I,I^c}$ for $I\\subset[n-1]$. Each limit component appears once. The degeneration is built by walking down a binary tree whose vertices are 'allowed pairs' $(I,I')$; at each step a single active matrix entry is scaled to zero, and the limit splits into two pieces whose row spans lie in the Richardson varieties attached to the two children of that vertex. Because the components' delta-matroid base polytopes tile the hypercube $P(\\Lambda)=[0,1]^n$, no other components can appear, and the polytope-cover criterion of the paper's own Corollary 2.4.1 identifies the flat limit. Consequently $[Z_\\Lambda]=\\sum_{I\\subset[n-1]}\\sigma_I\\sigma_{I^c}$ in $H^{n(n-1)}(\\mathrm{OG}(n,2n+1))$, and the same identity holds equivariantly.","pith_inferences":["The same tree-and-matrix degeneration is likely to adapt to the even orthogonal Grassmannian $\\mathrm{OG}(n,2n)$ and the Lagrangian Grassmannian $\\mathrm{LG}(n,2n)$, where lattice-path delta-matroids already govern hypercube-type decompositions; the paper stops at the odd orthogonal case.","The proof suggests a combinatorial criterion for reducedness of any toric degeneration of torus-orbit closures: if the base polytopes of the candidate components cover the general fiber's base polytope and each component's vertex-difference lattice is full, then all multiplicities equal one. Testing this criterion on other delta-matroids would be a direct extension.","Because the degeneration scales one matrix entry at a time, one could try to read the Schubert identity $\\sigma_I\\sigma_{I^c}$ as a statement about initial ideals and Gröbner degenerations of the Plücker ideal of the orbit closure, yielding a purely algebraic proof of the class formula."],"forward_implications":["The cohomology class of a general torus orbit closure is $[Z_\\Lambda]=\\sum_{I\\subset[n-1]}\\sigma_I\\sigma_{I^c}$ in $H^{n(n-1)}(\\mathrm{OG}(n,2n+1))$, and the same identity holds $T$-equivariantly.","The special fiber of the constructed degeneration has exactly the Richardson varieties $\\Sigma_{I,I^c}$ as components, each appearing with multiplicity 1.","The base polytopes $P(\\Lambda_I)$ of the delta-matroids attached to these pieces tile the hypercube $[0,1]^n=P(\\Lambda)$, giving a moment-map shadow of the degeneration that matches the Chen–Sanchez–Veliz–Ying hypercube decomposition.","For a general $\\Lambda$, the class of the structure sheaf of $Z_\\Lambda$ agrees with the delta-matroid $K$-class $y(D(\\Lambda))$, so the formula is compatible with delta-matroid invariants.","The class formula is proved by explicit degeneration rather than by equivariant localization, giving a geometric explanation for why every coefficient in the Schubert expansion equals 1."],"supporting_citations":[{"why":"Introduces the torus orbit closure $Z$ associated to a matroid and the viewpoint that its geometry encodes matroid invariants, which this paper extends to delta-matroids.","marker":"[Spe09]"},{"why":"Proved the type-A class formula for general torus orbit closures in $\\mathrm{Gr}(r,n)$ by equivariant localization, the formula this paper's degeneration re-proves in type B.","marker":"[BF17]"},{"why":"Established the same type of class formula for Hessenberg varieties in the full flag variety, another predecessor this type-B result extends.","marker":"[AT10]"},{"why":"Constructed the analogous explicit degeneration in the full flag variety that this paper adapts to $\\mathrm{OG}(n,2n+1)$.","marker":"[Lia24b]"},{"why":"Supplies the polytope-cover and lattice-index criterion used to identify flat limits and their multiplicities in toric degenerations.","marker":"[KSZ91]"},{"why":"Gives the coefficient formula $c_k=[\\Xi:\\Xi_k]$ for multiplicities in toric degenerations, used to prove each Richardson component appears once.","marker":"[Kap93]"},{"why":"Identifies the moment-map image of the torus orbit closure with the twice-dilated delta-matroid base polytope, connecting the degeneration to polytopes.","marker":"[EFLS24]"}],"fun_headline_variants":["Torus orbit flattens to Richardson varieties","Delta-matroid polytopes tile hypercube in degeneration","Schubert sum from flat limit of torus orbit","Explicit degeneration yields Schubert class sum","Delta-matroids decompose torus orbit closure"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction only reaches a general torus orbit because, at every step, the rational map $pr_r$ that produces the right-hand limit piece is dominant over the target Richardson variety; if dominance ever failed, the 'nicest' subspaces would not form a dense open locus and the iterative degeneration could miss some Richardson varieties.","fun_headline_variants_meta":{"raw":{"variants":["Torus orbit flattens to Richardson varieties","Delta-matroid polytopes tile hypercube in degeneration","Schubert sum from flat limit of torus orbit","Explicit degeneration yields Schubert class sum","Delta-matroids decompose torus orbit closure"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000153,"raw_usage":{"total_tokens":1181,"prompt_tokens":891,"completion_tokens":290,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":216}},"tokens_in":507,"tokens_out":290,"duration_ms":3832,"temperature":1.0,"reasoning_tokens":216,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:17:45.471497+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the equivariant localization of $[Z_\\Lambda]$ on $\\mathrm{OG}(3,7)$ for a general $\\Lambda$ and compare it with $\\sum_{I\\subset\\{1,2\\}}\\sigma_I\\sigma_{I^c}$; any disagreement refutes Theorem 1.0.1. Equivalently, find a point $x\\in[0,1]^3$ whose associated subset $I$ from Definition 5.1.2 violates the inequality $x(S)\\le g_{\\Lambda_I}(S)$ for a nice $\\Lambda_I$, which would refute Proposition 5.1.1.","supporting_citations":[],"review_version":1}