{"id":"b1983f5b-a269-4411-9a68-9435531ef043","arxiv_id":"2502.08038","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Fubini-Study map is quantitatively injective: the Hilbert-Schmidt distance between two metrics is bounded by k^n times the W^{2,2} norm of the associated function.","lead":"A mathematician proves a sharp stability estimate for the Fubini-Study map: how close two quantum metrics are is controlled, up to a polynomial factor in the line bundle power, by the Sobolev size of their difference. The paper also repairs two earlier papers by the same author that contained flawed proofs.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2.2 asserts a pointwise lower bound on the second fundamental form that is algebraically impossible when the codimension exceeds the dimension; the proof of Theorem 1.2 collapses at this step.","rationale":"The paper's central claim is a quantitative injectivity theorem for the Fubini-Study map, and the proof strategy using the Bergman kernel expansion, the decomposition Lambda = alpha + beta, and a coercivity estimate for the normal part is coherent. The reader focused on the unproved perturbation claim in Lemma 2.1. My concern is more decisive: even granting Lemma 2.1, the proof of Lemma 2.2 contains a displayed pointwise inequality that is algebraically impossible when the codimension of X in the projective space exceeds its dimension, which is the case for all large k. The matrix M has n rows and m columns with m > n, so M^* M necessarily has kernel; the claimed inequality cannot hold for any positive lambda_min. Since Theorem 1.2 uses Lemma 2.2 to estimate ||pi_N xi_beta||, the main proof is unsupported as written. I am not asserting that the theorem is false, only that the present manuscript does not establish it. A revision would require a genuinely different proof of Lemma 2.2 or a replacement estimate, and a careful reassessment of whether the theorem remains true under that replacement. For this reason I recommend moving the verdict from CONDITIONAL to REJECT, with the understanding that the paper could be reconsidered if the author supplies a valid proof of the missing coercivity estimate.","tokens_in":9218,"tokens_out":32725,"duration_ms":307847,"concrete_test":"Take n = 1, for example X = P^1, L = O(1), and k = d = 3, so the embedding is the rational normal curve in P^3, with H_k = Hilb(h^k). Compute the local frame used in Lemma 2.2 at a point x and form the 1 x m matrix M_j = dbar phi_1j(x). Choose a nonzero b in C^m with M b = 0. The asserted pointwise inequality would give 0 >= lambda_min |b|^2, a contradiction. This directly falsifies the step from Lemma 2.1 to Lemma 2.2; a corrected proof must replace it with a coercivity estimate on the image of M, not on all of C^m.","verdict_should_be":"REJECT","load_bearing_attack":"In the proof of Lemma 2.2, after the local frame calculation, the paper asserts that for every x and every b = (b_j) in C^m, where m = N_k - 1 - n is the codimension, one has sum_i |sum_j b_j dbar phi_ij(x)|^2 >= lambda_min sum_j |b_j|^2(x), citing Lemma 2.1. Let M_ij = dbar phi_ij(x). The claimed inequality is |M b|^2 >= lambda_min |b|^2 for all b in C^m, i.e. M^* M >= lambda_min I_m on C^m. But M is an n x m matrix, so rank(M) <= n. Since N_k = V k^n + O(k^{n-1}), for all sufficiently large k the codimension m = N_k - 1 - n is strictly larger than n. Hence M^* M has a kernel, so its smallest eigenvalue is zero and the inequality is impossible for any lambda_min > 0. This is not a missing detail in Lemma 2.1: even if that lemma is granted, it cannot imply the stated bound for all normal vectors. The proof of Lemma 2.2 therefore collapses, and with it the bound on ||pi_N xi_beta|| used directly in the proof of Theorem 1.2.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims a quantitative injectivity estimate for the Fubini–Study map: for a polarised smooth projective variety (X, L) with Hermitian metric h, there exist k0 and C_h > 0 such that for all k ≥ k0 and all positive definite Hermitian forms A, B on H^0(X, L^k), ||A^{-1} - B^{-1}||^2_{HS(H_k)} ≤ C_h k^n ||f_k(A,B;H_k)||^2_{W^{2,2}(ω_h)}, where H_k = Hilb(h^k) is the Hilbert functional and f_k is the ratio of Fubini–Study metrics. The proof uses the Bergman kernel expansion, a claimed uniform lower bound on the second fundamental form of the Kodaira embedding (Lemma 2.1), an estimate for the normal component of Hamiltonian vector fields (Lemma 2.2), and a comparison of the tangent projection of the vector field with derivatives of f_k. Section 3 applies Theorem 1.2 to correct an argument in the author's previous paper [8].","tokens_in":9465,"tokens_out":7183,"duration_ms":65991,"significance":"If the theorem were established, it would be a substantial quantitative refinement of Lempert's injectivity theorem, with an explicit polynomial dependence on the level k and a natural Sobolev norm. It would also repair arguments in the author's earlier work. The paper is well-motivated and the choice of the W^{2,2} norm is plausible. However, the proof as written contains a load-bearing algebraic impossibility in Lemma 2.2, so the main theorem is not established. The manuscript would require a fundamentally different estimate for the normal component of the Hamiltonian vector field, not a local repair.","major_comments":[{"comment":"The pointwise inequality proved in Lemma 2.2, namely sum_i |sum_{j=1}^{m} b_j dbar φ_ij(x)|^2 ≥ λ_min sum_j |b_j|^2 for all b = (b_j) in C^m, is algebraically impossible when m > n. Here M_ij = dbar φ_ij(x) is an n × m matrix, so for m > n the Hermitian matrix M^*M has rank at most n and therefore has a kernel; its smallest eigenvalue is 0, so no λ_min > 0 can satisfy the inequality for all b. Since m = N_k - 1 - n and N_k = V k^n + O(k^{n-1}), one has m > n for all sufficiently large k. This is not a missing detail in Lemma 2.1: a lower bound on the endomorphism -A^*_{h,k} ∧ A_{h,k} of TX does not imply injectivity of the normal-to-tangent map A: N → T^*X when the normal bundle has rank larger than n. The bound on ||π_N ξ_β|| used in the proof of Theorem 1.2 therefore does not follow, and the main theorem collapses.","section":"Section 2, Lemma 2.2"},{"comment":"The proof of Lemma 2.1 asserts without proof that one can perturb the Hermitian metric h locally to h_ε such that h_ε(x)=h(x), g_ε(x)=g(x), the curvature tensor is perturbed by -ε δ^i_j (k g_h)_{l\\bar m} at x, and Hilb(h_ε^k) = Hilb(h^k) + O(ε^2). This is a nontrivial statement: a local perturbation of order ε, even with a rapidly decreasing cutoff, will generically change the L^2 inner product Hilb at first order in ε unless the first moment of the perturbation against all products of sections vanishes. No such cancellation is demonstrated. The O(ε^2) property is load-bearing because the subsequent comparison of -A^*_{h,k,ε} ∧ A_{h,k,ε} with -A^*_{h,k} ∧ A_{h,k} assumes that the Fubini–Study metric and the orthogonal projection π_T are unchanged to first order. Without a proof of this perturbation claim, Lemma 2.1's uniform lower bound for all k is unsupported.","section":"Section 2, Lemma 2.1"},{"comment":"The correction to [8] in Section 3 relies directly on Theorem 1.2 (see the estimate |d_i - 1| ≤ C_7 C_9 C_h k^{2n-m-2}), and the earlier step bounding |A^{-1} - B^{-1}|_{HS(H_{m,k})} also uses Theorem 1.2. Since Theorem 1.2 is not established due to the failure of Lemma 2.2, the claimed correction to [8] is likewise unsupported.","section":"Section 3, correction to [8]"}],"minor_comments":[{"comment":"There are numerous OCR-type typographical errors, e.g. 'Fub ini' in the title, 'K¨ ahler' and 'ωn h' spacing issues, and 'λmin > 0 > 0' in Lemma 2.2.","section":"Throughout"},{"comment":"The notation 'with N − 1 = n + m' is ambiguous; it should be 'N_k - 1 = n + m' to be consistent with the rest of the paper.","section":"Section 2, Lemma 2.2 proof"},{"comment":"The constant C_4 is defined as 2 max{C_3, 2V}, but the bracketing of the preceding display suggests the coefficient of the L^2 term should be max{2, 2C_3} or similar; the stated choice of C_4 does not obviously match the displayed inequalities.","section":"Section 2, proof of Theorem 1.2"},{"comment":"The notation '||∇∇f_k||' is not defined; presumably it denotes the L^2 norm of the Hessian (or of ∇^{0,1}df_k), but it should be made explicit.","section":"Section 2, proof of Theorem 1.2"}],"recommendation":"reject","confidential_remarks":"The stress-test concern about Lemma 2.2 is valid and is the primary reason for rejection: the pointwise inequality asserted there is impossible when the codimension exceeds the dimension, which occurs for all sufficiently large k. This is not a fixable gap but a flaw in the proof strategy. The perturbation assertion in Lemma 2.1 is also unproved. For a journal in this area, the paper's main theorem is therefore not established; rejection is appropriate, although the theorem might be true by a different argument."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe headline is that the paper's main theorem is not proved. The argument fails at a specific point: Lemma 2.2 asserts a pointwise lower bound on the second fundamental form that is algebraically impossible when the codimension exceeds the dimension. No repair of Lemma 2.1 can fix this.\n\nWhat the paper does well: Theorem 1.2 is a clean, natural statement – a quantitative injectivity estimate for the Fubini–Study map with polynomial dependence on k. The author is transparent about the gaps in his earlier work, and the correction to [8] is a serious attempt to salvage a result that depended on a flawed claim. The writing is clear, and the use of Phong–Sturm's framework is standard.\n\nThe soft spot is load-bearing. The proof of Lemma 2.1 asserts, without proof, the existence of a local metric perturbation that changes the Hilbert functional only at order ε². That is already a nontrivial claim. But the deeper problem is Lemma 2.2. The inequality there says that for every point x and every vector b ∈ C^m (where m = codimension of X in P^{N_k-1}), we have |M b|² ≥ λmin |b|², with M_ij = ∂̄φ_ij(x). M is n × m, and for large k, m ≈ V k^n − n − 1 > n. A linear map from C^m to C^n cannot be injective when m > n, so M has a nonzero kernel, and the inequality fails for any λmin > 0. This holds even granting Lemma 2.1, because the obstruction is pure dimension counting. Consequently, Lemma 2.2 collapses, and with it the bound on ‖π_N ξ_β‖ that feeds directly into Theorem 1.2.\n\nThe final estimate does not go through, and the correction to [8] relies on Theorem 1.2, so it also loses its basis.\n\nWho this is for: anyone working on balanced metrics or the Yau–Tian–Donaldson program will want to know about this paper, mainly as a cautionary example of how these injectivity estimates can fail. The author's honesty about his earlier errors is commendable.\n\nMy recommendation: I would not cite the main theorem. But I would still send it to a serious referee – a competent referee can catch this quickly, and the author might find a way to repair the argument by restricting the class of normal vectors or using a different norm. As it stands, the paper should not be accepted without substantial revision.","headline":"The main theorem is not established; a rank obstruction in Lemma 2.2 makes the central inequality impossible for large k.","tokens_in":10018,"tokens_out":4644,"would_cite":false,"duration_ms":41950,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14C20","32Q15","53C55","32A25","14N05"],"pacs":[],"model":"deepseek-v4-flash","headline":"For any two metrics on H^0(X,L^k), the inverse-matrix distance is controlled by a Sobolev norm of their induced function, with a k^n bound.","keywords":["Fubini-Study map","quantitative injectivity","Kodaira embedding","Bergman kernel expansion","second fundamental form","Hilbert-Schmidt norm","Sobolev norm W^{2,2}","Kähler geometry"],"falsifier":"Find a polarized variety, for instance a smooth projective curve of genus at least two, compute the second fundamental form of the Kodaira embedding for a sequence of levels k and points x_k, and check whether the minimum eigenvalue of -A^*_{h,k} ∧ A_{h,k} stays bounded away from zero; a sequence tending to zero would falsify Lemma 2.1. A more direct test is to verify the perturbation assertion in the proof of Lemma 2.1: construct h_ε from the Kähler potential ε|z|^2|z_l|^2 times a cutoff function and check whether Hilb(h_ε^k) = Hilb(h^k) + O($ε^{2}$) for all large k; a counterexample to this identity would break the proof of the uniform lower bound.","tokens_in":8990,"feed_emoji":"📐","tokens_out":6078,"duration_ms":46303,"temperature":0.7,"pith_summary":"This paper proves a quantitative version of the injectivity of the Fubini–Study map: for a polarised smooth projective variety (X,L) with ample L and a fixed hermitian metric h, the difference $A^{{-1}}$-$B^{{-1}}$ between the inverse matrices of any two Hermitian forms on $H^{0}$(X,L^k) is bounded above by a constant times k^n times the $W^{{2,2}}$ norm of the function f_k(A,B;H_k). The exponent n is the dimension of X, so the estimate is polynomial in the level k, not exponential. This matters because the Fubini–Study map connects Hermitian metrics on the space of sections to Kähler metrics on X, and a quantitative injectivity statement with polynomial growth is precisely what is needed to compare geometric constructions across different powers k. The paper also corrects the arguments in the author's previous papers that relied on an erroneous surjectivity claim for the Hilbert map.","feed_headline":"Fubini–Study map: injectivity with a k^n bound","feed_subtitle":"Inverse matrices of Hermitian forms on H^0(X,L^k) differ by at most C_h k^n times a W^{2,2} norm of f_k.","key_machinery":"The proof combines the Bergman kernel expansion, which gives ω_{h,k} = kω_h + O(1/k) and locates the centre of mass of the embedding near the identity, with a lower bound on the second fundamental form -A^*_{h,k} ∧ A_{h,k} of the Kodaira embedding (Lemma 2.1), the load-bearing nondegeneracy statement. It then follows the Phong–Sturm decomposition of the traceless part of Λ into the automorphism algebra and its orthogonal complement, bounding the normal component of the associated holomorphic vector field by its (0,1)-derivative through the second fundamental form. The trace part is handled separately by integrating the defining function against the Bergman kernel, yielding the final $W^{{2,2}}$ estimate with the k^n factor.","core_discovery":"Theorem 1.2 states that there exist constants k_0 in N and C_h > 0, depending only on h, such that for every A,B in B_k and every k ≥ k_0, ||$A^{{-1}}$-$B^{{-1}}$||^2_{HS(H_k)} ≤ C_h k^n ||f_k(A,B;H_k)||^2_{$W^{{2,2}}$(ω_h)}. In words, the Fubini–Study map is injective with a quantitative bound, and the bound is polynomial of degree n in the exponent k. The result is stated with respect to the reference metric H_k = Hilb(h^k), and the proof shows that the trace-free part of Λ = $A^{{-1}}$-$B^{{-1}}$ is controlled by the first and second derivatives of f_k, while the trace part is controlled by the $L^{2}$ norm of f_k via the Bergman kernel expansion.","pith_inferences":["A natural test of sharpness would be to replace W^{2,2} by W^{1,2} for the trace-free part, since the control of the normal component of the vector field uses the (0,1)-derivative of the projected field, a second-order object; lowering the exponent to n-1 may fail.","The same strategy should give a uniform constant for families of metrics h varying in a compact set, since the compactness argument in Lemma 2.1 is the only place where uniformity in h enters.","The failure of the earlier C^0 estimate and the success of the W^{2,2} norm here suggest that the right function space is tied to the second fundamental form: quantitative injectivity holds exactly where the Kodaira embedding is uniformly non-degenerate."],"forward_implications":["The distance between A and B in B_k is controlled by the W^{2,2} norm of f_k, not merely its C^0 norm, so quantitative injectivity holds with a fixed reference metric H_k = Hilb(h^k).","The bound is uniform in A,B and all k ≥ k_0, with constants depending only on the reference metric h, so it applies to arbitrarily large powers of the line bundle.","The trace part of Λ is controlled by the L^2 norm of f_k via the Bergman kernel, so the full Hilbert–Schmidt norm is equivalent to the W^{2,2} norm up to the factor k^n.","The correction in Section 3 yields a quantitative bound |d_i - 1| ≤ C k^{2n-m-2} for the eigenvalues in the extremal-metric quantisation argument, at the cost of a power of k.","Injectivity of the Fubini–Study map in the sense of Lempert is recovered in the limit: if f_k has zero W^{2,2} norm, then A = B."],"supporting_citations":[{"why":"Supplies the formula ω_{h,k} = kω_h - (√-1/2π)∂∂̄ log ρ_k and FS_k(H_k) = ρ̄_k^{-1}h_k, which anchor the reference metric and the trace estimate.","marker":"[12]"},{"why":"Provides the asymptotic expansion ρ_k = k^n + O(k^{n-1}) in C^m norm, used throughout to control ω_{h,k}, the centre of mass, and the Bergman function ρ̄_k.","marker":"[10]"},{"why":"Supplies the decomposition of holomorphic vector fields, the identity relating the second fundamental form to ambient minus restricted curvature, and the estimate ||Λ||^2 ≤ Ck||ξ_Λ||^2.","marker":"[11]"},{"why":"Proves injectivity of the Fubini–Study map and the non-closedness of its image, the qualitative statement that Theorem 1.2 makes quantitative.","marker":"[9]"},{"why":"Provides a counterexample to the surjectivity of the Hilbert map, showing why the earlier C^0 approach in [7] fails and motivating the W^{2,2} norm.","marker":"[13]"},{"why":"The author's earlier paper containing the erroneous surjectivity claim and the C^0 estimate that this paper corrects.","marker":"[7]"},{"why":"The earlier quantisation-of-extremal-metrics paper whose Section 3 argument is corrected using Theorem 1.2.","marker":"[8]"},{"why":"Provides a counterexample with redundant basis elements that disproves the old argument, supporting the choice of reference metric H_k.","marker":"[4]"}],"fun_headline_variants":["Fubini–Study map injectivity gets a polynomial bound","Quantitative injectivity: Fubini–Study map with k^n control","Polynomial injectivity for Fubini–Study maps, past proofs fixed","Fubini–Study map: k^n-bound injectivity, with corrections"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of Lemma 2.1 relies on an unproved assertion that the hermitian metric h can be locally perturbed to h_ε so that the curvature of the induced metric changes by -εδ at a point while the Hilbert functional Hilb(h_ε^k) changes only at order $ε^{2}$; if this perturbation property fails, the uniform lower bound on the second fundamental form, and with it Theorem 1.2, collapses.","fun_headline_variants_meta":{"raw":{"variants":["Fubini–Study map injectivity gets a polynomial bound","Quantitative injectivity: Fubini–Study map with k^n control","Polynomial injectivity for Fubini–Study maps, past proofs fixed","Fubini–Study map: k^n-bound injectivity, with corrections"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000511,"raw_usage":{"total_tokens":2391,"prompt_tokens":758,"completion_tokens":1633,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":374,"completion_tokens_details":{"reasoning_tokens":1553}},"tokens_in":374,"tokens_out":1633,"duration_ms":10032,"temperature":1.0,"reasoning_tokens":1553,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T11:04:26.406016+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a polarized variety, for instance a smooth projective curve of genus at least two, compute the second fundamental form of the Kodaira embedding for a sequence of levels k and points x_k, and check whether the minimum eigenvalue of -A^*_{h,k} ∧ A_{h,k} stays bounded away from zero; a sequence tending to zero would falsify Lemma 2.1. A more direct test is to verify the perturbation assertion in the proof of Lemma 2.1: construct h_ε from the Kähler potential ε|z|^2|z_l|^2 times a cutoff function and check whether Hilb(h_ε^k) = Hilb(h^k) + O($ε^{2}$) for all large k; a counterexample to this identity would break the proof of the uniform lower bound.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the formula ω_{h,k} = kω_h - (√-1/2π)∂∂̄ log ρ_k and FS_k(H_k) = ρ̄_k^{-1}h_k, which anchor the reference metric and the trace estimate."},{"cited_title":"Ma and G","cited_arxiv_id":null,"evidence_quote":"Provides the asymptotic expansion ρ_k = k^n + O(k^{n-1}) in C^m norm, used throughout to control ω_{h,k}, the centre of mass, and the Bergman function ρ̄_k."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the decomposition of holomorphic vector fields, the identity relating the second fundamental form to ambient minus restricted curvature, and the estimate ||Λ||^2 ≤ Ck||ξ_Λ||^2."},{"cited_title":"On the Bergman kernels of holomorphic vector bundles","cited_arxiv_id":"2109.08593","evidence_quote":"Proves injectivity of the Fubini–Study map and the non-closedness of its image, the qualitative statement that Theorem 1.2 makes quantitative."},{"cited_title":"On The Image Of The Hilbert Map","cited_arxiv_id":"2208.13407","evidence_quote":"Provides a counterexample to the surjectivity of the Hilbert map, showing why the earlier C^0 approach in [7] fails and motivating the W^{2,2} norm."},{"cited_title":"Hashimoto, Mapping properties of the Hilbert and Fubini-Study maps in K ¨ ahler geometry, Ann","cited_arxiv_id":null,"evidence_quote":"The author's earlier paper containing the erroneous surjectivity claim and the C^0 estimate that this paper corrects."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The earlier quantisation-of-extremal-metrics paper whose Section 3 argument is corrected using Theorem 1.2."}],"review_version":1}