{"id":"3fa19e71-d610-48e2-bd98-0b8214647adc","arxiv_id":"2412.01060","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A graded Buchweitz-Greuel-Schreyer rank conjecture is shown equivalent to a new sheaf-cohomology lower bound on non-Fano hypersurfaces, with a new exact Betti-number-to-cohomology formula as the bridge.","lead":"This paper connects a 1987 conjecture about matrix factorization ranks to a new conjecture about sheaf cohomology of non-Fano projective hypersurfaces. It proves the two statements are equivalent for Calabi-Yau hypersurfaces and supplies an exact formula translating Betti numbers into cohomology groups.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central implication in Theorem 1.4 is sound; the concrete defect is Example 4.7, where f=x_0^n+...+x_n^n has degree n, so a=1, not 0, invalidating the sharpness example as written.","rationale":"I read the paper in good faith and tried to stress-test the central equivalence. The proof of Theorem 3.3, which is the engine behind Theorem 1.4, appears sound: the combinatorial sum over j in part (2) exactly tiles all cohomological degrees, and the shift in Proposition 3.1, despite a sign typo in the proof display, is consistent with the final formula. A low-dimensional check for a plane cubic (n=2,d=3,a=0) gives a coherent Betti table with total rank 8, matching rho(O_X). The only definite mathematical error I found is in Example 4.7, where the polynomial has degree n, forcing a=1, not 0. This invalidates the sharpness example as stated but does not affect the conditional equivalence in Theorem 1.4. The reader identified the same Example 4.7 issue, so my concern agrees with the reader there; however, the reader's weakest_assumption focused on a possible off-by-one in Proposition 3.1, which my analysis did not confirm. I therefore mark agreement as partial. Since the existing CONDITIONAL verdict already covers the needed correction to Example 4.7 and the main theorem is unaffected, I recommend no change to the reader's verdict.","tokens_in":10781,"tokens_out":55922,"duration_ms":485584,"concrete_test":"Recompute Example 4.7 with n=2m+1 and f=x_0^{2m+2}+...+x_n^{2m+2}, so d=n+1 and a=0. Verify that the tensor product of m+1 rank-one matrix factorizations of the pairwise sums x_{2j}^{2m+2}+x_{2j+1}^{2m+2} has rank(F^0)=2^m=2^{floor(n/2)}, and use Theorem 3.3(2) to confirm rho(tilde M)=2^{m+1}. If this corrected construction works, the sharpness claim is recoverable and the central theorem is unchanged.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most concrete issue is Example 4.7. The stated polynomial f=x_0^n+...+x_n^n has degree d=n, so a=n+1-d=1, not 0. The example then invokes Theorem 2.3(3), the Calabi-Yau inverse equivalence, and claims the constructed matrix factorization achieves the bound in Conjecture 1.3. Both steps require a=0, so the example lies outside the hypotheses of Conjecture 1.3 and does not demonstrate sharpness as written. This does not invalidate Theorem 1.4 or Theorem 3.3, which are conditional on the graded BGS conjecture; the sharpness example is peripheral. The error is plausibly a typo: taking n odd and f=x_0^{n+1}+...+x_n^{n+1} gives d=n+1 and a=0, and the tensor product of (n+1)/2 rank-one factorizations has rank 2^{(n-1)/2}=2^{floor(n/2)}, recovering the claimed bound. I also checked the reader's flagged shift concern in Proposition 3.1: the displayed [n] in the proof appears to be a typo for [-n], but the final shift in Proposition 3.1 is consistent with Theorem 3.3, and for n=2,d=3 the formula produces a viable Betti table for O_X with total rank 8. Thus the load-bearing technical engine appears correct.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a graded version of the Buchweitz-Greuel-Schreyer rank conjecture for matrix factorizations and relates it to a new conjecture (Conjecture 1.3) about the total cohomology invariant ρ(C) of sheaves on non-Fano projective hypersurfaces. The main result, Theorem 1.4, shows that the graded BGS conjecture implies Conjecture 1.3 for all non-Fano hypersurfaces, and conversely in the Calabi-Yau case a=0. The technical core is Theorem 3.3, which uses Orlov's equivalence and graded Auslander duality to express the Betti numbers of a graded matrix factorization in terms of cohomology of P^n with coefficients in Ω^{r+a}(r+a), yielding a rank–cohomology equality. The paper also computes ρ for points, structure sheaves, and Fano counterexamples, and Example 4.7 purports to show sharpness of Conjecture 1.3.","tokens_in":10972,"tokens_out":27088,"duration_ms":235237,"significance":"If the main arguments are correct, the paper provides a clean bridge between two open conjectures: it reduces the graded BGS conjecture in the Calabi-Yau case to a concrete statement about sheaf cohomology, and it gives a new prediction about the total rank of the Beilinson spectral sequence. The explicit formula in Theorem 3.3 is checkable and is the main technical contribution. The paper is also transparent about the conditional nature of its main theorem, which depends on the still-open graded BGS conjecture. The flaw in Example 4.7 identified below does not invalidate the proof of Theorem 1.4 or Theorem 3.3, but it does affect the advertised sharpness of Conjecture 1.3.","major_comments":[{"comment":"Example 4.7 is internally inconsistent as written. The polynomial f=x_0^n+...+x_n^n has degree n, so a=n+1-n=1, not 0; therefore the line \"Since a=0\" is false, and Theorem 2.3(3) cannot be invoked. Since Conjecture 1.3 is only formulated for a≤0, the example falls outside the hypotheses of the conjecture and does not demonstrate sharpness. In addition, the tensor-product construction as written is problematic: one copy of F, which factorizes x_0^n, together with e copies of F', each of which factorizes x_0^n+x_1^n, would produce a factorization of (e+1)x_0^n+e x_1^n, not of f, unless the copies of F' are understood to be relabeled so as to act on disjoint pairs of variables. The example should be repaired, for instance by taking n odd and f=x_0^{n+1}+...+x_n^{n+1}, and then taking the tensor product of (n+1)/2 pair factorizations; this gives a=0 and recovers the claimed rank bound. Because the Introduction states that the lower bound in Conjecture 1.3 is sharp by citing this example, the correction is necessary before publication.","section":"Section 4, Example 4.7"}],"minor_comments":[{"comment":"The statement \"rank(F)=ρ(C)\" is ambiguous. Under the convention in Section 2.1, rank(F) means the common rank rank(F^0)=rank(F^1), but the proof computes rank(F)=rank(F^0)+rank(F^1)=Σ_j(b^0_j+b^1_j), and Theorem 1.4 uses the equality in the form 2·rank(F^0)=ρ(C). Please define rank(F) in Theorem 3.3(2) explicitly as the sum of the two ranks, or state the equality as 2·rank(F^0)=ρ(C).","section":"Theorem 3.3(2)"},{"comment":"The definition of the truncation G_{≺i} appears garbled: \"free summands of the form R(i) with i > 0\" cannot be right as written, since the index i is already the truncation parameter. Please clarify whether G_{≺i} consists of summands R(j) with j < i or j > i.","section":"Section 2.2"},{"comment":"In the chain of isomorphisms in the proof of Theorem 3.3, the reference \"(Theorem 3.1)\" should be \"(Proposition 3.1)\".","section":"Proof of Theorem 3.3"},{"comment":"The citation \"[OSS11] Chapter 1, Section 1.1]\" has a missing opening bracket or comma; it should read \"[OSS11, Chapter 1, Section 1.1]\".","section":"Example 4.3"}],"recommendation":"major_revision","confidential_remarks":"The flaw in Example 4.7 is substantial but appears repairable within the scope of the paper: the main theorems and the proof of Theorem 3.3 are not affected. If the authors cannot produce a correct sharpness example, they should remove the sharpness claim from the Introduction or replace it with a corrected statement. The paper is otherwise a solid contribution to the literature on matrix factorizations and derived categories."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe short version: this paper proves a clean bridge theorem between a graded form of the BGS rank conjecture and a new sheaf-cohomology rank conjecture, and the main argument looks sound. The one real defect I found is Example 4.7, which as printed does not achieve what it claims. That's fixable, but it needs fixing.\n\nWhat's actually new: Theorem 3.3 gives an exact formula expressing the graded Betti numbers of a reduced matrix factorization of f in terms of cohomology h^*(P^n, i_*C ⊗ Ω^{r+a}(r+a)), with rank(F) = ρ(C). This generalizes Pavlov's Koszul computation beyond the Calabi-Yau case to all non-Fano hypersurfaces. The bridge theorem, Theorem 1.4, shows that the graded BGS bound is equivalent to their Conjecture 1.3 in the Calabi-Yau case, and implies it in general. That is a genuinely useful reframing: it turns a local algebra question into a geometric statement about all sheaves on a hypersurface, and the Beilinson spectral interpretation is a nice touch.\n\nThe proof machinery is standard but used competently: Orlov, Auslander duality, Shamash/Koszul resolution. I cannot see a load-bearing error in the main computation. The shift concern raised by the stress-test is a red herring—the displayed [n] in the proof of Proposition 3.1 is a typo for [−n], and the final shift in the statement is consistent.\n\nWhere it's soft: Example 4.7 is internally inconsistent. For f = x_0^n + ... + x_n^n, deg f = n, so a = n+1−n = 1, not 0. The sharpness claim relies on a = 0 for both Conjecture 1.3 and the inverse equivalence, so as written the example lies outside the hypotheses. The likely fix is to take n odd and deg d = n+1, giving a = 0, and build rank 2^{(n-1)/2} factorizations; that recovers the claimed bound. The authors cite [BD20] for tensor products, which is appropriate and not a substitute for fixing the example.\n\nVerdict: the main theorem stands, but the example must be corrected before publication. A serious referee will catch this, and the paper deserves referee time. I'd bring it to our reading group—it's a good illustration of how to connect singularity categories with sheaf cohomology.","headline":"Solid bridge theorem between graded BGS and a new sheaf cohomology conjecture; the main argument holds, but Example 4.7 has a genuine a=1/0 error that must be fixed.","tokens_in":11641,"tokens_out":1656,"would_cite":true,"duration_ms":15047,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["13D02","14F08"],"pacs":[],"model":"deepseek-v4-flash","headline":"A graded rank conjecture for matrix factorizations and a new sheaf-cohomology conjecture are equivalent on Calabi-Yau hypersurfaces.","keywords":["matrix factorizations","rank conjecture","hypersurface singularities","sheaf cohomology","non-Fano hypersurfaces","derived categories","singularity categories","spectral sequences"],"falsifier":"Compute $\\Phi_0(k(\\ell))$ directly for a non-Fano hypersurface, say a quartic surface in $\\mathbb{P}^3$ or a quintic threefold in $\\mathbb{P}^4$, and compare the cohomological shift with $[2q+n-r-a-1]$; any discrepancy of one, or any computed value of $\\rho(C)$ below $2^{\\lfloor n/2\\rfloor+1}$ for a nonzero $C$ on a Calabi-Yau hypersurface, would disprove the bridge.","tokens_in":10463,"feed_emoji":"📐","tokens_out":15660,"duration_ms":117018,"temperature":0.7,"pith_summary":"The paper studies the graded version of a 1987 rank conjecture for matrix factorizations, which predicts that every nontrivial matrix factorization of a hypersurface singularity has rank at least $2^{\\lfloor n/2\\rfloor}$. It aims to show that, for smooth non-Fano projective hypersurfaces, this algebraic bound is really a geometric statement: the graded rank conjecture implies that every nonzero sheaf complex $C$ on $X$ satisfies $\\rho(C)\\ge 2^{\\lfloor n/2\\rfloor+1}$, where $\\rho$ counts cohomology against twisted differential forms on the ambient projective space. For Calabi-Yau hypersurfaces ($a=0$) the two statements are proved equivalent. A reader should care because this converts an open local algebra conjecture into a concrete lower bound on the cohomology of every sheaf, and the paper exhibits an example where the bound is attained.","feed_headline":"A 1987 rank conjecture now equals a sheaf-cohomology bound","feed_subtitle":"On Calabi-Yau hypersurfaces, an algebraic rank bound and a geometric cohomology bound are the same statement.","key_machinery":"The load-bearing mechanism is the Betti-number/cohomology dictionary of Theorem 3.3. For a reduced graded matrix factorization $F$ with $\\Psi_a(C)\\cong\\operatorname{coker}(F)$, writing $a-j=qd-r$ with $0\\le r<d$, it asserts $b^i_j=h^{r+a-2q-i+1}(\\mathbb{P}^n, i_*C\\otimes\\Omega^{r+a}_{\\mathbb{P}^n}(r+a))$, and hence $\\operatorname{rank}(F)=\\rho(C)$. The dictionary is built from an equivalence between the graded singularity category and $D^b(X)$, a duality isomorphism for the singularity category that turns the relevant Hom groups into Ext groups, and Proposition 3.1, which computes $\\Phi_0(k(\\ell))$ for $\\ell=qd-r$ as $i_*(\\wedge^{r+a}T_{\\mathbb{P}^n})(-r-a)[2q+n-r-a-1]$. That residue-field computation is the step that uses $a\\le 0$ through the inequality $\\ell-a+d>0$.","core_discovery":"For a smooth hypersurface $X=V(f)\\subseteq \\mathbb{P}^n_k$ of degree $d$ with $a=n+1-d\\le 0$, the central claim is: if every nontrivial graded matrix factorization of $f$ has rank at least $2^{\\lfloor n/2\\rfloor}$, then every nonzero object $C\\in D^b(X)$ satisfies $\\rho(C)\\ge 2^{\\lfloor n/2\\rfloor+1}$; and when $a=0$ the converse also holds. The bridge is Theorem 3.3, which identifies the Betti numbers of a graded matrix factorization with cohomology dimensions of $C$ twisted by $\\Omega^{r+a}_{\\mathbb{P}^n}(r+a)$, so that twice the rank of the factorization equals $\\rho(C)$. The paper does not settle either conjecture, but it reduces one to the other exactly, and it constructs a vector bundle on even-dimensional hypersurfaces given by a sum of $d$th powers that attains the bound.","pith_inferences":["One could test Conjecture 1.3 numerically on well-known vector bundles over Calabi-Yau threefolds; an indecomposable bundle with $\\rho<8$ would disprove it and, by Theorem 1.4, the graded rank conjecture in that case.","The dictionary may extend to Fano hypersurfaces if $\\rho$ is replaced by a truncated invariant, since the paper traces the failure on Fano examples precisely to line bundles.","The pattern suggests a general translation principle: rank statements about graded matrix factorizations correspond to total-cohomology statements about sheaves whenever the images of residue fields in the derived category are computable."],"forward_implications":["If the graded 1987 rank conjecture holds for a smooth non-Fano hypersurface, then every nonzero object in $D^b(X)$, including every vector bundle, satisfies $\\rho(C)\\ge 2^{\\lfloor n/2\\rfloor+1}$.","On Calabi-Yau hypersurfaces, Conjecture 1.3 and the graded rank conjecture are the same statement: proving either settles the other.","The bound is sharp: on even-dimensional hypersurfaces given by a sum of $d$th powers there is a reduced graded matrix factorization whose associated vector bundle has $\\rho=2^{\\lfloor n/2\\rfloor+1}$.","The invariant $\\rho(C)$ is the total rank of the $E_1$ page of the spectral sequence from Remark 3.4, so the conjecture says every nonzero sheaf complex on a non-Fano hypersurface needs at least $2^{\\lfloor n/2\\rfloor+1}$ total entries there.","The non-Fano condition $a\\le 0$ is necessary: on linear hypersurfaces, line bundles achieve $\\rho=2$, below the conjectured bound."],"supporting_citations":[{"why":"states the original rank conjecture whose graded version is the paper's target.","marker":"[BGS87]"},{"why":"supplies the equivalence between the graded singularity category and $D^b(X)$ used to connect ranks to sheaf cohomology.","marker":"[Orl09]"},{"why":"provides the construction and adjunction properties of the functors $\\Phi_i$ and $\\Psi_i$ used in Theorem 3.3.","marker":"[BS15]"},{"why":"gives the duality isomorphism for the singularity category that powers the Betti-number computation.","marker":"[KMVdB11]"},{"why":"supplies the Shamash construction of the minimal free resolution of the residue field used in Proposition 3.1.","marker":"[EP16]"},{"why":"provides the exterior-power/tangent-bundle isomorphism used in Proposition 3.1 and Proposition 4.1.","marker":"[Eis95]"},{"why":"supplies the lemma expressing graded Betti numbers as Homs in the singularity category, used in the proof of Theorem 3.3.","marker":"[Pav21]"},{"why":"gives the explicit cohomology of twisted differential forms on projective space used to verify the bound and its sharpness.","marker":"[OSS11]"},{"why":"identifies $\\rho(C)$ as the total rank of the $E_1$ page of the Beilinson spectral sequence, giving the conjecture its cohomological meaning.","marker":"[Be ˘ ı78]"}],"fun_headline_variants":["Matrix factorization ranks now tie to sheaf cohomology","On non-Fano hypersurfaces, rank bound equals cohomology bound","1987 rank conjecture and sheaf cohomology: the same bound","Rank conjecture meets sheaf cohomology on Calabi-Yau","Matrix factorizations bridge rank and sheaf cohomology"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument rests on one exact shift in a long calculation, and on the still-open graded 1987 rank conjecture for the first direction.","fun_headline_variants_meta":{"raw":{"variants":["Matrix factorization ranks now tie to sheaf cohomology","On non-Fano hypersurfaces, rank bound equals cohomology bound","1987 rank conjecture and sheaf cohomology: the same bound","Rank conjecture meets sheaf cohomology on Calabi-Yau","Matrix factorizations bridge rank and sheaf cohomology"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00053,"raw_usage":{"total_tokens":2481,"prompt_tokens":798,"completion_tokens":1683,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":414,"completion_tokens_details":{"reasoning_tokens":1593}},"tokens_in":414,"tokens_out":1683,"duration_ms":46975,"temperature":1.0,"reasoning_tokens":1593,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:45:08.300727+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\Phi_0(k(\\ell))$ directly for a non-Fano hypersurface, say a quartic surface in $\\mathbb{P}^3$ or a quintic threefold in $\\mathbb{P}^4$, and compare the cohomological shift with $[2q+n-r-a-1]$; any discrepancy of one, or any computed value of $\\rho(C)$ below $2^{\\lfloor n/2\\rfloor+1}$ for a nonzero $C$ on a Calabi-Yau hypersurface, would disprove the bridge.","supporting_citations":[],"review_version":1}