{"id":"5e7dd633-e499-4c38-b004-9124c0607ce1","arxiv_id":"2506.16132","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For tensors of any fixed order, geometric rank is bounded by a function of subrank, and for order-three tensors the bound is quadratic, over every field.","lead":"This mathematics paper proves new upper bounds connecting two important measures of large multidimensional arrays, called tensors. It shows that the geometric rank is controlled by the subrank, settling open questions in algebraic complexity and combinatorics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5.1's contradiction skips the field-extension transfer: g∈(g_1,...,g_t) only gives GR_{L_m}(T'_{m+1})≤2I, so bounding GR_K(span_K{T_i}) needs Prop 4.6 with M_1>J(d,c,2I), not the stated M_1>2I.","rationale":"The paper's two headline results—Theorem 1.4 for order-three tensors over arbitrary fields and Theorem 1.2 over algebraically closed fields—appear sound on inspection. The order-three proof uses Lemma 2.2 directly, and the contrapositive plus Proposition 3.19 and Proposition 4.2 supply the quadratic bound; the finite-field case uses the Grassmannian count and Proposition 4.4 correctly. Theorem 1.2 avoids the field-extension transfer by working over an algebraically closed field, where the counterpart Lemma 6.1 needs only the Ananyan–Hochster Theorem 2.5 and no Prop 4.6 step. The genuine weakness is in Lemma 5.1, which supports Theorem 1.1: the proof asserts a direct bound on GR(span_K{T_i}) from ideal membership of a form over an extension L_m, but Proposition 4.6 only gives a bound via the function J. Since the manuscript defines M_1 merely with M_1>2I, the displayed contradiction is missing its last link. This is not a fatal flaw in the central order-three result, but it is a real gap in one of the three main theorems, so the CONDITIONAL verdict is appropriate and no change to the reader's disposition is needed.","tokens_in":25664,"tokens_out":34107,"duration_ms":333706,"concrete_test":"Re-derive the Step 1.1 contradiction in Lemma 5.1 with M_1(d,c) redefined as J(d,c, 2I(d, c^d−c(c−1)^{d−1}+(d−1)c)) + 1, where J is the function from Prop 4.6 and I from Prop 3.14; check that the full chain g∈(g_1,...,g_t) ⇒ str_{L_m}(g)≤t ⇒ GR_{L_m}(T'_{m+1})≤2I ⇒ GR_{\\bar K}(span)≤2I ⇒ GR_K(span)≤J(d,c,2I) now contradicts M_1. If this repaired bound goes through while the paper's literal choice M_1>2I does not, the gap is confirmed as a real but fixable omission in Theorem 1.1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Lemma 5.1, Step 1.1, the proof assumes that if g∈(g_1,...,g_t), then t≥str(g), and concludes: 'Proposition 4.6 and Proposition 4.5 would lead to a contradiction: I(d,N+(d−1)m) ≥ GR(span_K{T_1,...,T_c})/2 ≥ M_1(d,c)/2.' The displayed inequality is unjustified. From g∈(g_1,...,g_t) one gets str_{L_m}(g)≤t≤I, hence by Prop 4.5, GR_{L_m}(T'_{m+1})≤2I. But T'_{m+1} lies in span_{L_m}{T_1,...,T_c}, not necessarily in span_K{T_1,...,T_c}; to contradict assumption (iii) one must bound GR_K(span_K{T_i}). Prop 4.6 gives only GR_K(V)≤J(d,c,2I), not ≤2I, and the chosen M_1(d,c)>2I(...) does not ensure M_1>J(d,c,2I(...)). The proof therefore omits a necessary stability step. The gap is repairable by defining M_1 via J∘I, so the theorem may still be true, but as written the induction in Lemma 5.1 is incomplete. Since Lemma 5.1 is the engine for Theorem 1.1, this is a load-bearing proof-level concern, though it does not appear to affect Theorems 1.2 or 1.4.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the relation between geometric rank GR(T) and subrank Q(T) for tensors of fixed order d over a field K. Its three main theorems are: (Theorem 1.1) under a characteristic condition, GR(T) is bounded by a function of the subrank over a field extension of bounded degree; (Theorem 1.2) over algebraically closed fields, GR(T) ≤ C(d, Q(T)) for a function C, resolving an open problem of Kopparty–Moshkovitz–Zuiddam; and (Theorem 1.4) for order-three tensors over arbitrary fields, GR(T) = O(Q(T)^2), with an explicit quadratic bound for infinite fields. The paper derives several corollaries concerning stability of subrank, direct sums, de-bordering of border subrank, bounds on partition and analytic rank, and a gap theorem. The proofs are built on polynomial-ideal results in the style of Ananyan–Hochster, rational-point lemmas in bounded-degree extensions, and structural lemmas about geometric rank.","tokens_in":25951,"tokens_out":20479,"duration_ms":199406,"significance":"If the proofs are correct, this is a substantial contribution. The order-three quadratic bound is optimal for generic tensors, and the algebraic-closure result answers a named open problem. The applications to border subrank, partition rank, and analytic rank are natural and have not, to my knowledge, appeared before. The paper is generally careful and self-contained, and it gives explicit functions in several places, including the clean bound GR(T) ≤ 2Q(T)^2 + 3Q(T) for infinite fields.","major_comments":[{"comment":"The sentence \"Clearly, an R1-sequence of forms in K[x_1,...,x_n] is a prime sequence\" is false. For the form xy in K[x,y], the quotient K[x,y]/(xy) satisfies Serre's condition (R1) but is not a domain. This implication is used in Proposition 3.16 to conclude that (f_1,...,f_m) is prime and that f_1,...,f_m,g form a regular sequence, and in Lemma 3.15 to treat the sequence produced by Proposition 3.14 as regular. Since Proposition 3.16 is invoked in Step 1.2 of Lemma 5.1 and is similarly reused in Lemma 6.1, Theorems 1.1 and 1.2 are not supported as written. The authors should either prove that the specific R1-sequences produced by Propositions 3.13 and 3.14 and Theorem 2.5 are prime sequences, or replace this construction by a theorem that produces prime sequences.","section":"§2.2, definition of R1-sequence"},{"comment":"The displayed contradiction \"I(d,N+(d−1)m) ≥ GR(span_K{T_1,...,T_c})/2 ≥ M_1(d,c)/2\" is unjustified. From g ∈ (g_1,...,g_t) one obtains str_{L_m}(g) ≤ t ≤ I, hence GR_{L_m}(T'_{m+1}) ≤ 2I by Proposition 4.5; however T'_{m+1} lies in span_{L_m}{T_i}, not necessarily in span_K{T_i}. To contradict assumption (iii) one must bound GR_K(span_K{T_i}), and Proposition 4.6 gives GR_K(span_K{T_i}) ≤ J(d,c,2I), not ≤ 2I. The condition M_1(d,c) > 2I(...) does not imply M_1(d,c) > J(d,c,2I(...)). The argument can likely be repaired by defining M_1 through J ∘ I, but as written the induction in Lemma 5.1 is incomplete.","section":"§5, Lemma 5.1, Step 1.1"},{"comment":"The definition of the integer s is off by one. The contrapositive of Lemma 5.1 with c = s requires Q_F(T) ≤ s−1 for every extension F/K with [F:K] ≤ M_2(d,s), but the text says \"degree M_2(d,s−1)\" and then uses the same index in the subsequent minimality argument. Replacing M_2(d,s−1) with M_2(d,s) in the definition appears to fix the step, but as written the conclusions [L:K] = M_2(d,s−1) ≤ B(d,GR(T)) and GR(T) ≤ A(d,Q_L(T)) do not follow from the stated condition.","section":"§5, proof of Theorem 1.1"}],"minor_comments":[{"comment":"The notation Q_K(T) is not defined; if it denotes subrank over the algebraic closure, the equality GR_K(T) = GR(T) needs a justification or a reference, and the notation should be introduced explicitly.","section":"§1.3, Corollary 1.6"},{"comment":"The expression \"∪_{m}^{c=0}\" should read \"∪_{c=0}^{m}\".","section":"§4, proof of Proposition 4.2"},{"comment":"The set X is defined as a subset of (K^{n_3})^* although the tensor has order d; it should be (K^{n_d})^*, and the condition should refer to GR(T_u), not rank(T_u).","section":"§6, proof of Theorem 1.2"},{"comment":"In the reference list, \"Disrete Analysis\" should be \"Discrete Analysis\", and \"Theroem B\" should be \"Theorem B\".","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper's results, if repaired, would be a strong contribution to the area. The most serious issue is not a mere typo: the claimed implication from R1-sequences to prime sequences is false as stated, and it is used in a load-bearing way in Proposition 3.16 and hence in the main proofs. The transfer gap in Lemma 5.1 is more routine but still needs a fix. I recommend that the authors be asked to address both points before the paper can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the order-three result is real and likely correct; the general-d theorem is plausible but Lemma 5.1 as written has a gap. I would send it to a serious referee.\n\nWhat is new: for fixed order d, the paper proves GR(T) is bounded by a function of Q(T) over algebraically closed fields, answering an open problem of Kopparty, Moshkovitz and Zuiddam. For d=3, it gives the explicit bound GR(T) <= 2Q(T)^2 + 3Q(T) over infinite fields, and a similar quadratic bound over finite fields. The six corollaries, especially stability of subrank, the first de-bordering result for border subrank, and the first bounds PR(T)=O(Q(T)^2), AR(T)=O(Q(T)^2), are genuine additions.\n\nWhat it does well: the order-three proof is clean, uses Lemma 2.2 and Proposition 4.2 directly, and does not depend on the more delicate induction. The polynomial-ideal machinery is applied carefully, and the paper is honest about which constants are explicit and which are merely existential. It also correctly notes that the quadratic bound is best possible generically.\n\nSoft spots, in proportion: the stress-test concern about Lemma 5.1 is correct. In Step 1.1, if g lies in (g1,...,gt), then strength over L_m is at most I, so Proposition 4.5 gives GR over L_m of T'_{m+1} at most 2I. But T'_{m+1} lives in span_{L_m}{T1,...,Tc}, while assumption (iii) is about GR_K(span_K{T1,...,Tc}). To get the contradiction you need Proposition 4.6, which gives only GR_K(V) <= J(d,c,2I), not <= 2I. The displayed chain I(d,...) >= GR(span_K)/2 >= M1(d,c)/2 skips that transfer, and M1 > 2I does not imply M1 > J(d,c,2I). This is fixable by taking M1(d,c) larger than J(d,c,2I(...)), but as written the induction in Lemma 5.1 is incomplete. The issue is load-bearing for Theorem 1.1 only. Theorem 1.4 does not use Lemma 5.1, and Lemma 6.1 works over algebraically closed fields with no field-extension transfer, so those two appear fine.\n\nThere is also a minor indexing ambiguity in Theorem 1.1's proof: the minimality definition of s and the claimed extension of degree M2(d,s-1) with Q_L(T)=s-1 need tightening. I read it as a small fix, not a substantive flaw.\n\nCitation pattern is fine. The finite-field branch uses the analytic-rank/geometric-rank equivalence from [5,14,46], including the authors' own [14], but that is an external quoted result, and I see no circularity.\n\nWho this is for: tensor-rank researchers in combinatorics, algebraic geometry, and complexity theory. The order-three theorem is the part I would rely on immediately.\n\nRecommendation: seriously referee, not desk reject. Ask for a repaired Lemma 5.1 and a cleaned-up s-step in Theorem 1.1; after that, the paper is publishable.","headline":"Genuinely new bounds for geometric rank in terms of subrank; the order-three theorem is solid, but the proof of the general large-field theorem has a repairable but real gap in Lemma 5.1.","tokens_in":26548,"tokens_out":7575,"would_cite":true,"duration_ms":60950,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["15A69","05D05","13P10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Subrank bounds geometric rank for fixed-order tensors — and quadratically in order three.","keywords":["geometric rank","subrank","border subrank","partition rank","analytic rank","field extension","R1-sequence","strength of forms"],"falsifier":"Enumerate order-three tensors of small format (for example, 4×4×4 over Q or F_2), compute Q(T) by exhaustive restriction search and GR(T) by the codimension of the zero set of the associated multilinear form, and check the claimed quadratic inequality; a single tensor with GR(T) > 2Q(T)^2+3Q(T) over an infinite field would refute Theorem 1.4.","tokens_in":25400,"feed_emoji":"🧮","tokens_out":10622,"duration_ms":99674,"temperature":0.7,"pith_summary":"The paper proves that geometric rank, a tensor invariant measuring the codimension of the set of directions where a tensor's multilinear form vanishes, is always dominated by subrank, the size of the largest identity tensor that can be embedded into the tensor. For order-three tensors over any field the domination is quadratic: an explicit bound GR(T) ≤ 2Q(T)^2 + 3Q(T) holds over infinite fields, with an explicit quadratic bound over finite fields as well. For tensors of arbitrary fixed order, geometric rank is bounded by a function of subrank over algebraically closed fields, and over fields of characteristic zero or characteristic larger than the order the same conclusion holds after passing to a field extension of bounded degree. These results close an open problem on whether subrank controls geometric rank, and they imply new bounds on partition rank, analytic rank, and border subrank.","feed_headline":"Subrank controls geometric rank: quadratic in order 3","feed_subtitle":"Subrank is easy to estimate in examples; now it provably limits geometric rank, even across field extensions.","key_machinery":"The machinery is a two-layer reduction. First, Proposition 4.2 expresses GR(T) as min_c(codim X_c + c), where X_c is the set of d-th slices of T having geometric rank c, so the entire tensor's geometric rank is controlled once the slices' ranks are controlled. Second, the slices are controlled by commutative algebra: repeatedly applying Proposition 3.14 (bounded-length R1-sequences of forms, i.e. sequences whose successive quotient rings stay regular in codimension one) and Proposition 3.16 (rational points in bounded-degree extensions) produces vectors that avoid the derivative ideal of the slice span while retaining one prescribed nonzero value, forcing the Kronecker-delta contractions that build an identity tensor of size Q(T). The quantitative strength bounds underlying these propositions, and their transfer under field extensions, are what turn an existence argument into an explicit GR(T) ≤ A(d, Q(T)) bound. In the order-three case, Lemma 2.2 replaces the heavier algebra: c linearly independent slices whose nonzero linear combinations all have rank at least 2c(c−1) already force Q(T) ≥ c, so the proof reduces to a Grassmannian counting estimate.","core_discovery":"The central claim is that for tensors of fixed order, subrank is a genuine measure of complexity: it controls geometric rank quantitatively. Theorem 1.4 gives GR(T) ≤ 2Q(T)^2 + 3Q(T) for every order-three tensor over an infinite field, and an explicit quadratic polynomial for finite fields, so GR(T) = O(Q(T)^2). The quadratic growth is optimal, since for a generic n×n×n tensor one has Q(T) = Θ($n^{{1/2}}$) and GR(T) = n. For algebraically closed fields, Theorem 1.2 supplies a function C(d, Q(T)) with GR(T) ≤ C(d, Q(T)) for every order d, answering the open problem raised in [35, Section 9]; Theorem 1.1 is the analogue for large characteristic and bounded-degree extensions. The proof reworks the geometry of slices: GR(T) is read off from the geometric ranks of its d-slices, and high geometric rank of the span of the slices is forced to produce the contractions that realize a large identity tensor.","pith_inferences":["The same slice-avoidance technique may yield a polynomial bound of degree d−1 for all orders if the base-field transfer in Propositions 4.5 and 4.6 can be made effective over arbitrary fields; the paper itself proposes this as Conjecture 8.1, and the order-three case is its first nontrivial confirmation.","Because geometric, partition, and analytic rank are linearly equivalent over large fields, the quadratic-in-subrank bound effectively promotes a combinatorial lower bound on subrank into a geometric lower bound on all three ranks; cap-set-style constructions that bound subrank from below may now yield new extremal results.","The explicit constants in Theorem 1.4 are small enough that the proof is close to algorithmic: for a tensor with subrank r, a certificate of geometric rank could in principle be exhibited as a slice of the form built by the induction."],"forward_implications":["For order-three tensors, subrank is stable under field extensions: Q_K(T) = O(Q(T)^2), generalizing known real-field stability to arbitrary fields (Corollary 1.6).","Subrank of a direct sum is controlled by the sum of the bounding functions: Q(S⊕T) ≤ C(d,Q(S)) + C(d,Q(T)), and in order three the maximal quadratic gap between Q(S⊕T) and Q(S)+Q(T) is attained (Corollary 1.8).","Border subrank is de-bordered: Q(T) ≤ Q̄(T) ≤ C(d,Q(T)) over algebraically closed fields, with a quadratic bound in order three (Corollary 1.10).","For order-three tensors, partition rank and analytic rank are each O(Q(T)^2), the first upper bounds of this kind (Corollaries 1.11 and 1.12).","A gap theorem follows for the subrank of Kronecker powers over algebraically closed fields: either every power has subrank 1, or subrank grows at least exponentially in k/2 (Corollary 1.13)."],"supporting_citations":[{"why":"Defines geometric rank and subrank, proves Q(T) ≤ GR(T), and poses the open problem answered by Theorem 1.2.","marker":"[35]"},{"why":"Gives generic subrank Θ(n^{1/(d−1)}), which makes the quadratic bound optimal, and the direct-sum gap generalized by Corollary 1.8.","marker":"[23]"},{"why":"Supplies the R1-sequence and strength bounds (Theorems 2.5 and 2.7) that give explicit constants for the polynomial-ideal arguments.","marker":"[4]"},{"why":"Provides strength and partition-rank transfer under field extensions used in Proposition 4.6 and Theorem 2.8.","marker":"[10]"},{"why":"Provides the partition–analytic rank equivalence and the bound PR ≤ (2^{d−1}−1)GR used to convert geometric rank bounds to partition rank.","marker":"[18]"},{"why":"Establishes stability of ranks under field extensions and GR ≍ AR, used for the analytic-rank corollary.","marker":"[14]"},{"why":"Provides Lemma 2.2, the slice-rank counting lemma that powers the order-three theorem.","marker":"[12]"},{"why":"Gives the generic border-subrank growth Θ(n^{1/2}) that the de-bordering corollary complements.","marker":"[6]"}],"fun_headline_variants":["Subrank squared bounds geometric rank in order 3","Quadratic control: subrank limits geometric rank","Order-three tensor rank gap closed by subrank bound","Subrank determines geometric rank for fixed order"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on a quantitative trade-off: the geometric rank of the space spanned by the tensor's slices must be large enough relative to the number of low-degree forms needed to generate the ideal they avoid, and this transfer must survive passing to a field extension; if that gap closes at any induction step, the construction that forces a large identity restriction collapses.","fun_headline_variants_meta":{"raw":{"variants":["Subrank squared bounds geometric rank in order 3","Quadratic control: subrank limits geometric rank","Order-three tensor rank gap closed by subrank bound","Subrank determines geometric rank for fixed order"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000854,"raw_usage":{"total_tokens":3783,"prompt_tokens":1093,"completion_tokens":2690,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":709,"completion_tokens_details":{"reasoning_tokens":2630}},"tokens_in":709,"tokens_out":2690,"duration_ms":19147,"temperature":1.0,"reasoning_tokens":2630,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:29:12.580513+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate order-three tensors of small format (for example, 4×4×4 over Q or F_2), compute Q(T) by exhaustive restriction search and GR(T) by the codimension of the zero set of the associated multilinear form, and check the claimed quadratic inequality; a single tensor with GR(T) > 2Q(T)^2+3Q(T) over an infinite field would refute Theorem 1.4.","supporting_citations":[{"cited_title":"Kopparty, G","cited_arxiv_id":null,"evidence_quote":"Defines geometric rank and subrank, proves Q(T) ≤ GR(T), and poses the open problem answered by Theorem 1.2."},{"cited_title":"Derksen, V","cited_arxiv_id":null,"evidence_quote":"Gives generic subrank Θ(n^{1/(d−1)}), which makes the quadratic bound optimal, and the direct-sum gap generalized by Corollary 1.8."},{"cited_title":"Ananyan and M","cited_arxiv_id":null,"evidence_quote":"Supplies the R1-sequence and strength bounds (Theorems 2.5 and 2.7) that give explicit constants for the polynomial-ideal arguments."},{"cited_title":"Cohen and G","cited_arxiv_id":null,"evidence_quote":"Provides the partition–analytic rank equivalence and the bound PR ≤ (2^{d−1}−1)GR used to convert geometric rank bounds to partition rank."},{"cited_title":"Bri¨ et, M","cited_arxiv_id":null,"evidence_quote":"Provides Lemma 2.2, the slice-rank counting lemma that powers the order-three theorem."},{"cited_title":"Biaggi, C.-Y","cited_arxiv_id":null,"evidence_quote":"Gives the generic border-subrank growth Θ(n^{1/2}) that the de-bordering corollary complements."}],"review_version":1}