{"id":"c940fa67-9b49-42b5-bb5c-609e43336563","arxiv_id":"2502.06137","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"For every C2 hypersurface not lying in a hyperplane, there exist f and a weight w with the weighted extension integral at scale R exceeding the Mizohata-Takeuchi bound by log R.","lead":"A new construction shows the Mizohata-Takeuchi conjecture, a central open inequality in Fourier restriction theory, fails by a logarithmic factor for every curved C2 hypersurface. If correct, this closes a long-open question and blocks a proposed route to sharp multilinear restriction estimates.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4.4's containment of the C^2 graph points in the standard moment-curve boxes is false for d >= 3, so the reduction of the incidence lemma to the moment curve is not established.","rationale":"Read in good faith, the paper's high-level strategy is coherent: X-ray transform estimates, a subset-sum construction, and an incidence lemma would yield a log R counterexample if the incidence lemma held for arbitrary C^2 hypersurfaces. The reader's verdict identifies the weakest load-bearing step: Lemma 4.4, the only bridge from the moment curve to the general C^2 case. My own check of the containment confirms the reader's objection. The chosen points lie on a permuted version of the moment curve, not a perturbation of the standard one, and the box U^d has anisotropic widths that are far too small in the coordinates where the discrepancy is order t^2. This is an explicit false assertion in the proof, not a disagreement with consensus. It blocks the claimed reduction to Lemma 4.3 and therefore the counterexample is not established by this text. The rest of the argument may be salvageable by applying Lemma 4.3 to a coordinate permutation of M_d, but that repair is absent. Since the submitted proof has this gap in its central path, I do not see reason to move away from the reader's rejection; the verdict is UNCHANGED.","tokens_in":14608,"tokens_out":13158,"duration_ms":118098,"concrete_test":"Verify Lemma 4.4 coordinatewise in the smallest case d = 3. Take c = 100, b = 2, n = 10, so x_{c,n} = (100^{-10}, 100^{-20}, 100^{-30}). The point xi_n has second coordinate 100^{-30}, while U^3_{b,c,n} is centered at 100^{-20} and has second-coordinate half-width about 4 * 100^{-22}; clearly 100^{-30} is outside the interval. Equivalently, compare c^{-3n} - c^{-2n} with the box width O((b/c)^2 c^{-2n}) analytically; if the gap persists for c >> b, Lemma 4.4 is false and the reduction needs a different argument (e.g., a permuted moment curve).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4 reduces Lemma 4.1 for an arbitrary C^2 hypersurface to Lemma 4.3 by Lemma 4.4, which claims each xi_n = (c^{-n}, c^{-3n}, ..., c^{-dn}, Phi(c^{-n}, c^{-3n}, ...)) lies in the standard box U^d_{b,c,n} around M_d(c^{-n}) = (c^{-n}, c^{-2n}, ..., c^{-dn}). For d >= 3 this containment is false. In the second coordinate the discrepancy is c^{-3n} - c^{-2n} = -c^{-2n}(1 - c^{-n}), whereas the second side length of U^d_{b,c,n} is O((b/c)^2 c^{-2n}); since c >> b, the discrepancy is larger by a factor (c/b)^2. In the last coordinate the discrepancy is about c^{-2n}, while the box side is O((b/c)^d c^{-dn}), an even larger relative gap. The proof of Lemma 4.4 itself computes the difference as if Phi(xi_n) matched the second coordinate and ignores the fact that xi_n uses t^3, not t^2, in its second slot. Consequently the transfer from arbitrary curved hypersurfaces to the standard moment-curve incidence statement (Lemma 4.3) is not proved, and Lemma 4.1, hence Lemma 3.5 and the log R lower bound in Proposition 3.1, is unsupported. A coordinate permutation of M_d might repair the argument, but that is not what the manuscript proves.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims a counterexample to the Mizohata-Takeuchi conjecture: for every C2 hypersurface in R^d not contained in a hyperplane, there exist f in L^2(Sigma) and a nonnegative weight w such that the extension-operator energy on B_R has a log R lower bound relative to the squared L2 norm of f times the supremum of line integrals of w (Theorem 1.2). The proof has three parts: an L^p estimate for the X-ray transform (Theorem 2.1), a reduction of the desired weighted inequality to an incidence statement for lacunary points on Sigma (Lemma 3.5 via Proposition 3.1), and a proof of that incidence lemma by approximating any C2 hypersurface by the moment curve M_d(t)=(t,t^2,...,t^d) (Lemma 4.4) and proving a moment-curve incidence bound (Lemma 4.3).","tokens_in":1323,"tokens_out":1402,"duration_ms":97895,"significance":"If the main theorem were proved, this would be an important development: it would refute a long-standing conjecture, imply that Stein's conjecture as stated in (1.2) is false, and show that the CHV approach cannot yield endpoint multilinear restriction estimates. The paper also contains a clean and apparently correct X-ray transform estimate (Theorem 2.1) and a plausible induction proof of the moment-curve incidence lemma (Lemma 4.3). However, the transfer lemma that connects the moment-curve construction to a general C2 hypersurface is false as stated, and the main theorem is therefore not established by the argument given.","major_comments":[{"comment":"The containment xi_n in U^d_{b,c,n} asserted in Lemma 4.4 is false for d >= 3. By (4.4), U^d_{b,c,n} is centered at M_d(c^{-n})=(c^{-n}, c^{-2n}, ..., c^{-dn}) and has side length in the j-th coordinate comparable to (b/c)^j c^{-jn}. The point constructed before Lemma 4.4 is xi_n = (c^{-n}, c^{-3n}, c^{-4n}, ..., c^{-dn}, Phi(omega_n)). Its second coordinate differs from that of the center by c^{-3n} - c^{-2n}, whose size is c^{-2n}; since c is much larger than b, this is much larger than the allowed width (b/c)^2 c^{-2n}. Moreover, by (4.7), Phi(omega_n) is approximately c^{-2n}, while the d-th coordinate of the center is c^{-dn} and the allowed width there is (b/c)^d c^{-dn}; the discrepancy of size c^{-2n} again far exceeds the width. The proof's assertion that Mtilde_d - M_d lies only in the second-coordinate direction is contradicted by direct computation. Since Lemma 4.4 is the only step transferring Lemma 4.1 for arbitrary curved C2 hypersurfaces to the moment-curve Lemma 4.3, Lemma 4.1, and with it Proposition 3.1 and Theorem 1.2, are unsupported.","section":"Section 4, Lemma 4.4"},{"comment":"The proof of the lower bound (3.6) is incomplete. The step that if q, xi_i, xi_j are randomly chosen then there is a probability at least 1/4 that q + xi_i - xi_j will lie in Q is asserted without a precise probability model or proof, and the conclusion (3.8) is not derived from it. In addition, the deduction that every R^{-1}-ball contains at most O(1) points of Q from Lemma 3.5 is not immediate and requires an argument; Lemma 3.5 controls intersections with planes, not with balls. These gaps matter because (3.6) is the source of the log R factor. I list this as a separate concern, but it is secondary to the failure of Lemma 4.4.","section":"Section 3.4, Eqs. (3.8)-(3.9)"}],"minor_comments":[{"comment":"The display in Lemma 3.5(i) appears to have a typo: the summands are written as c1(xi_1-xi_0)+...+c_N(xi_1-xi_0), which would use xi_1 in every term; presumably the intended expression is c1(xi_1-xi_0)+...+c_N(xi_N-xi_0).","section":"Section 3.2, Lemma 3.5(i)"},{"comment":"The construction of eta in Section 3.4 should justify that there is a function eta with eta_hat supported in B(C), 1_{B(C^{-1})} <= eta_hat <= 1_{B(C)}, and eta*eta >= B_1(0) on a fixed ball while the negative part of eta*eta has sufficiently small integral; the current text states this as a choice without details.","section":"Section 3.4, definition of eta"},{"comment":"The 'White Lie' notation A is approximately B is used in displays that later enter the formal proof, e.g., in Eq. (3.3); the paper would benefit from stating exactly which approximations are valid and which are only heuristic, even after the rigorous Section 3.4.","section":"General presentation"}],"recommendation":"reject","confidential_remarks":"The paper is clearly written and the failure is well localized, but it is load-bearing: Lemma 4.4 is false, so the main theorem is not proved by the given argument. A correct proof would require a genuinely new reduction from arbitrary C2 hypersurfaces to a usable model curve, not a local correction. If the author can supply such an argument, the paper could become highly significant, but the current version is not acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThe headline: this paper claims a log R counterexample to Mizohata-Takeuchi, but the transfer step from the moment curve to a general C^2 hypersurface (Lemma 4.4) is wrong as written, so the main theorem is not established. The reader's assessment is correct, and the stress-test note lands.\n\nWhat's genuinely good: the X-ray transform L^p estimates in Section 2 are new and clean; the \"white lie\" construction is a nice expository device; the moment-curve incidence lemma (Lemma 4.3) is plausible and may be a real result; the paper is self-contained and engages honestly with the literature. If the moment-curve case holds up, that is already a meaningful partial result.\n\nThe soft spot is load-bearing. Lemma 4.4 claims that the points ξ_n = (ω_n, Φ(ω_n)) with ω_n = (c^{-n}, c^{-3n}, ..., c^{-dn}) lie in the standard boxes U^d_{b,c,n} around the moment curve M_d(t) = (t, t^2, ..., t^d). They do not. The second coordinate of ξ_n is c^{-3n}, not c^{-2n}; the difference is about c^{-2n}, which is larger than the second side length of the box, roughly (b/c)^2 c^{-2n}. The higher coordinates also mismatch. The proof of Lemma 4.4 treats the last coordinate Φ(ω_n) as though it were the second coordinate of the point; it is not. So the reduction of Lemma 4.1 to the moment-curve case is not proved, and Proposition 3.1 and Theorem 1.2 are unsupported.\n\nThis is not a cosmetic gap. A coordinate permutation or a different choice of ω_n might repair the argument, but that is not what the manuscript shows. The moment-curve version of the incidence lemma does not directly apply to the constructed points.\n\nWho this is for: people working on Fourier restriction, weighted estimates, and the MT conjecture. The paper is worth reading as a serious attempt with an interesting strategy, and the flaw is specific enough that a revision could be evaluated. My recommendation: send it to referees; the claim is major and the error is pinpointable, but expect heavy revision or a corrected transfer argument.","headline":"A serious counterexample attempt whose reduction to the moment curve fails on a coordinate mismatch; the main theorem is unproved as written, though the moment-curve lemma may survive.","tokens_in":15416,"tokens_out":6280,"would_cite":false,"duration_ms":48812,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42B10","42B25","44A12"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every C2 hypersurface that is not a hyperplane, the Mizohata–Takeuchi weighted extension inequality fails by a factor of $\\log R$.","keywords":["Mizohata–Takeuchi conjecture","Fourier extension operator","weighted L2 estimates","X-ray transform","multilinear restriction","incidence geometry","moment curve","logarithmic loss"],"falsifier":"Compute the third-coordinate discrepancy for $d=3$ and for $\\Sigma$ given near $0$ by $\\Phi(u,v)=u^2+v^2$: the chosen point is $(t,t^3,t^2+o(t^3))$, while the box around $(t,t^2,t^3)$ has third-direction half-width of order $t^3$. Since $t^2\\gg t^3$ for $t=c^{-n}$ small, the point lies outside the box; this direct check would settle whether the reduction to the moment-curve case holds.","tokens_in":14322,"feed_emoji":"📐","tokens_out":12434,"duration_ms":99273,"temperature":0.7,"pith_summary":"The paper claims that the Mizohata–Takeuchi conjecture, a central weighted $L^2$ estimate for Fourier extension operators, is false in every dimension for every $C^2$ hypersurface that is not contained in a hyperplane. The counterexample is quantitative: on a ball of radius $R$, the integral of $|Ef|^2$ against a nonnegative weight can exceed the conjectured bound by a factor of $\\log R$. If correct, this closes the route from Mizohata–Takeuchi to sharp endpoint multilinear restriction estimates, because that route would have implied the $R^{\\varepsilon}$-free bound the construction contradicts. It also disproves Stein's conjecture in the form stated in the paper. The proof's heart is a combinatorial incidence lemma about half-sums of points on the surface.","feed_headline":"Mizohata-Takeuchi fails for every curved surface with log R loss","feed_subtitle":"The weighted extension inequality loses a log factor on any C2 hypersurface that is not a plane.","key_machinery":"The load-bearing object is the half-sum set $Q$ together with the claim (Lemma 3.5, proved through Lemma 4.1) that the $N$ points $\\xi_i$ can be chosen on any curved $C^2$ hypersurface so that no hyperplane contains more than $2^{d-1}$ of the radius-$R^{-1}$ balls centered in $Q$. The proof of the lemma approximates the surface locally by the moment curve $M_d(t)=(t,t^2,\\ldots,t^d)$, places the points at dyadically spaced parameter values, and inducts on dimension with the projection $\\varphi(x_1,\\ldots,x_d)=(x_2/x_1,\\ldots,x_d/x_1)$, which sends hyperplanes to hyperplanes and sends $M_d$ to $M_{d-1}$. A separate estimate (Theorem 2.1) bounds the X-ray transform of the weight by an $L^2(L^1)$ norm of a Fourier slice of $h$, converting the combinatorial incidence control into the upper bound on the right-hand side.","core_discovery":"Theorem 1.2 asserts that for any compact $C^2$ hypersurface $\\Sigma \\subset \\mathbb{R}^d$ not lying in a hyperplane, with surface measure $d\\sigma$ and extension operator $Ef(x)=\\int_\\Sigma e^{-2\\pi i x\\cdot\\xi}f(\\xi)\\,d\\sigma(\\xi)$, there are $f\\in L^2(\\Sigma;d\\sigma)$ and a nonnegative weight $w$ such that $\\int_{B_R(0)}|Ef|^2\\,w \\gtrsim \\log R\\,\\|f\\|_{L^2}^2 \\sup_{\\ell\\subset\\mathbb{R}^d\\text{ a line}}\\int_\\ell w$. The conjectured inequality would instead give the right-hand side with no $\\log R$ factor, so the two sides are directly comparable. The construction takes $f$ to be a sum of $N\\approx\\log R$ surface caps at carefully selected separated points $\\xi_i$, and takes $h$ to be a counting measure on the lattice $Q=\\{\\sum_i c_i\\xi_i: c_i\\in\\{0,1\\},\\ \\sum_i c_i=\\lfloor N/2\\rfloor\\}$ convolved with a scale-$R^{-1}$ bump. The convolution $h\\ast f\\,d\\sigma$ accumulates to size roughly $N^2|Q|R^d$, while the incidence lemma guarantees that every hyperplane contains only $O(1)$ of the relevant $R^{-1}$-balls, which keeps the X-ray transform factor bounded by roughly $R|Q|$ and yields the $\\log R$ ratio.","pith_inferences":["If the containment gap in Lemma 4.4 is real, a natural repair is a permuted moment curve whose second coordinate carries the quadratic term $t^2$ and whose remaining coordinates are higher-order monomials; the theorem-level claim could survive if the incidence argument goes through under that reordering.","The strength of the failure ($\\log R$ rather than $R^{\\delta}$) suggests that the sharp universal obstruction for weighted extension is logarithmic, so the local $R^{\\varepsilon}$ bound may be the natural endpoint for Kakeya-free methods.","The half-sum lattice construction may transplant to other weighted Fourier inequalities, such as Bochner–Riesz or Stein-type maximal estimates, whenever an analogous plane-incidence lemma holds.","A concrete next test is to replace planes by cones or curved submanifolds in the incidence lemma and check whether the $\\log R$ loss persists for surfaces with flat components, which the current $C^2$ non-planar hypothesis does not cover."],"forward_implications":["For every $C^2$ hypersurface not contained in a hyperplane, the original Mizohata–Takeuchi inequality fails by a $\\log R$ factor.","The functional-analytic route from Mizohata–Takeuchi to endpoint multilinear restriction cannot produce $R^{\\varepsilon}$-free bounds; a different input would be needed to sharpen those estimates.","Stein's conjecture in the form (1.2), which implies Mizohata–Takeuchi, also fails as stated.","The local reformulation with an $R^{\\varepsilon}$ loss (Conjecture 1.5) is not contradicted by the construction and remains open.","The failure is uniform: it is measured by the supremum of the X-ray transform over all lines, not by a single selected direction."],"supporting_citations":[{"why":"Establishes that Mizohata–Takeuchi implies endpoint multilinear restriction without $R^{\\varepsilon}$ losses; this is the consequence Theorem 1.2 blocks.","marker":"[CHV23]"},{"why":"Supplies the endpoint multilinear Kakeya inequality used in that implication.","marker":"[Gut10]"},{"why":"Formulated the multilinear restriction conjecture and proved it with $R^{\\varepsilon}$ loss at the endpoint, the bound the Mizohata–Takeuchi route would have sharpened.","marker":"[BCT06]"},{"why":"Provides the best general Mizohata–Takeuchi-type bounds with $R^{(n-1)/(n+1)+\\varepsilon}$ loss and justifies taking the weight locally constant at scale 1 in the construction.","marker":"[CIW24]"},{"why":"Origin of the conjecture as a necessary condition for $L^2$ well-posedness of the perturbed Schrödinger Cauchy problem.","marker":"[Miz85]"}],"fun_headline_variants":["Mizohata-Takeuchi conjecture refuted for all curved surfaces","Log R counterexample disproves Mizohata-Takeuchi on curved C2","Mizohata-Takeuchi fails: curved surfaces force log R loss","Any nonplanar surface gives log R loss, defeating MT conjecture","MT conjecture false: log R gap on every C2 non-hyperplane"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction reduces to Lemma 4.4's claim that the points chosen on an arbitrary curved $C^2$ hypersurface, through the ansatz $(c^{-n},c^{-3n},\\ldots,c^{-dn})$, land inside the rectangular boxes built around the standard moment curve; if that containment fails in any coordinate, the incidence lemma and the $\\log R$ counterexample do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Mizohata-Takeuchi conjecture refuted for all curved surfaces","Log R counterexample disproves Mizohata-Takeuchi on curved C2","Mizohata-Takeuchi fails: curved surfaces force log R loss","Any nonplanar surface gives log R loss, defeating MT conjecture","MT conjecture false: log R gap on every C2 non-hyperplane"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000889,"raw_usage":{"total_tokens":3837,"prompt_tokens":951,"completion_tokens":2886,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":2787}},"tokens_in":567,"tokens_out":2886,"duration_ms":18054,"temperature":1.0,"reasoning_tokens":2787,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T16:39:22.306614+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the third-coordinate discrepancy for $d=3$ and for $\\Sigma$ given near $0$ by $\\Phi(u,v)=u^2+v^2$: the chosen point is $(t,t^3,t^2+o(t^3))$, while the box around $(t,t^2,t^3)$ has third-direction half-width of order $t^3$. Since $t^2\\gg t^3$ for $t=c^{-n}$ small, the point lies outside the box; this direct check would settle whether the reduction to the moment-curve case holds.","supporting_citations":[],"review_version":1}