{"id":"6d179af1-ed54-4431-bae9-b9bb88b4b38c","arxiv_id":"1909.02535","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Ancient mean curvature flows that converge rapidly to a compact self-shrinker at time minus infinity must have the same codimension as the shrinker, and if the convergence is super-exponential the flow equals the shrinker.","lead":"Mathematicians showed that if a shrinking shape's past behavior approaches a known self-shrinking shape fast enough, the flow cannot hide in extra dimensions and may be completely rigid. The work extends a recent theorem about cylinders to all compact self-shrinkers, giving new tools for classifying ancient solutions of the mean curvature flow.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.3's proof depends on an unstated bound on the coefficient vector of the extra linear mode; until that estimate is derived, the codimension-bound argument is incomplete.","rationale":"The reader's verdict is CONDITIONAL with moderate confidence, and the reader already flagged the unstated spectral-gap arguments, including the decay bound on V_{r1+1} in Theorem 1.3. My stress-test agrees that the central theorems are plausible and that the convergence-rate hypothesis is the sharp dividing line, as Remark 3.4 shows. However, I would put the weight not on the rate itself (which is an explicit hypothesis) but on the underived coefficient-vector estimate in the proof of Theorem 1.3. That estimate is load-bearing because it enters the error term in Lemma 3.7 and hence the final contradiction. The paper does not provide the derivation, and it is not a purely cosmetic omission: the natural normalization bound on U is not automatically compatible with the required vanishing of the extra term. I found no evidence of a counterexample or of circularity, and the Carleman-based rigidity theorem in Section 4 appears fixable despite a compressed regularity estimate. Since the identified gap can likely be filled by a direct spectral computation and the reader already requires such filling, the verdict should remain CONDITIONAL; it should not be upgraded to ACCEPT without the missing estimate, but no downgrade is warranted.","tokens_in":17472,"tokens_out":30635,"duration_ms":332941,"concrete_test":"Independently derive the bound on V_{r1+1} in the proof of Theorem 1.3. Start from the orthonormalization at t = -Ω^{m_q+1}, write v_{r1+1} = ⟨x, U⟩, and use the C^1-graph closeness together with orthogonality to ψ_1, ..., ψ_{r1} to track exponents through the Gaussian norms. If the correct estimate is |U| ~ 1/(√(-t)ε(t)) rather than Ω^{m_q(-λ1+ρ)}, substitute that into the term -t1 κ in Lemma 3.7 and check whether the assumed decay of J_t(|Φ|, 1) still makes the right side of (3.34) tend to 0 for all λ1 > 0. If it fails when λ1 > 1/2, Theorem 1.3 requires a revised argument.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 1.3, after orthonormalizing {1, v_1, ..., v_{r1+1}} at t = -Ω^{m_q+1}, the authors assert without derivation that |V_{r1+1}| ≤ C Ω^{m_q(-λ1+ρ)}. This bound is load-bearing because V_{r1+1} is the coefficient vector of the linear function v_{r1+1} = ⟨x, V_{r1+1}⟩, and it controls the term -C t1 κ in Lemma 3.7 through κ = sup J_t(|Φ||U|, 1). The final contradiction from (3.32)-(3.34) requires this term to vanish. A naive normalization argument only gives |U| ~ (-t1)^{-1/2} for a generic orthonormal linear function, and orthogonality to the transplanted eigenfunctions ψ_i does not automatically improve this bound when λ1 > 1/2. The analogous unstated step in Theorem 1.1, that extra coordinate functions lie in P_{2δ'}, is easily justified from |u| ≤ C√(-t)ε(t) = o((-t)^δ), but the coefficient-vector estimate needed in Theorem 1.3 is not derived. These are not refutations of the theorems, but they are genuine gaps: without the coefficient-vector estimate, the upper bound in Lemma 3.7 and the contradiction in Theorem 1.3 do not follow. The convergence-rate hypothesis is explicit and sharp (Remark 3.4); the concern is not the rate itself but the underived spectral and normalization bookkeeping that converts the rate into a usable estimate.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies ancient mean curvature flows in arbitrary codimension whose rescaled flow at -∞ is a C^1-graph over a compact shrinker Σ. Theorem 1.1 proves that if the graph error decays as (-t)^{1/2-δ}ε(t)→0 with δ<λ_1, where λ_1 is the first nonzero drift-Laplacian eigenvalue on Σ, then the ambient codimension of the flow equals that of Σ. Corollary 1.2 applies this to the m-covered circle, giving rigidity under a type I assumption. Theorem 1.3 weakens the convergence-rate assumption at the cost of extra spectral assumptions and proves codim(M_t) ≤ codim(Σ)+r_1. Theorem 1.4 uses a Carleman inequality to show that sufficiently fast exponential convergence forces the rescaled flow to equal the shrinker exactly. The paper is motivated by the explicit torus-curve examples from [AAAW13], which are used to test sharpness of the rates.","tokens_in":17738,"tokens_out":13387,"duration_ms":135989,"significance":"If the proofs are completed, the paper gives a meaningful extension of Colding-Minicozzi's codimension-complexity paradigm from round cylinders to arbitrary compact shrinkers, with a sharp rate in Theorem 1.1 and an explicit sharpness example via torus curves. The entropy computation for torus curves and the discussion around Conjecture 2.4 are valuable contributions, and the Carleman-based rigidity in Theorem 1.4 is a clean and potentially useful tool. However, the proof of Theorem 1.3 contains an underived coefficient-vector estimate that is load-bearing for the main weaker-convergence result, and the statement of Theorem 1.4 needs a uniformity clarification. These are local but necessary repairs rather than refutations of the central strategy.","major_comments":[{"comment":"The estimate |V_{r_1+1}| ≤ C Ω^{m_q(-λ_1+ρ)} is asserted without derivation, and it is load-bearing: it is what makes the κ-term in Lemma 3.7 vanish in (3.34), and the final contradiction in Theorem 1.3 depends on that. The preceding facts do not imply it automatically. Orthonormality at t=-Ω^{m_q+1} with v_{r_1+1}=⟨x,V_{r_1+1}⟩ gives |V| of order Ω^{-(m_q+1)/2} on the natural scaling, while orthogonality to the transplanted eigenfunctions ψ_i does not by itself improve this to a power with exponent -λ_1, especially when λ_1>1/2. The authors need to supply the missing spectral/normalization argument, including the exact scaling convention for V_{r_1+1}; without it, the upper bound (3.34) and the contradiction in Theorem 1.3 are not established.","section":"§3.3, proof of Theorem 1.3, after (3.33)"},{"comment":"The hypothesis is written as limsup_{t→-∞} φ^2 e^{-αt}=0 for a function on Σ×(-∞,0), but the proof requires that ∫_Σ φ^2(·,T_1)e^{-αT_1}→0 as T_1→-∞. Pointwise convergence in x on the compact manifold Σ does not imply the integral of the boundary term converges to zero without uniformity in x. The statement should clarify that the limsup is uniform in x, e.g. limsup_{t→-∞} sup_Σ φ^2 e^{-αt}=0, or the proof should establish the needed uniform decay from the graph equation. As written, the passage from the Carleman inequality to φ≡0 has a gap.","section":"§4, Theorem 4.1 and its proof"}],"minor_comments":[{"comment":"The assertion that an additional coordinate function lies in P_{2δ'}(M_t) is stated without derivation; a short justification using |u|≤C√(-t)ε(t)=o((-t)^δ) and the graph closeness to Σ would make the proof self-contained.","section":"§3.1, proof of Theorem 1.1"},{"comment":"The proof says 'Applying Lemma 3.1 to v', but the correct reference is Lemma 3.6, since v is constructed to be orthogonal to 1 and to the ψ_i and the bound uses λ_{l+1} rather than λ_1.","section":"§3.3, proof of Lemma 3.7"},{"comment":"The notation in the assumption ||ψ_i/s_i(t)-φ_i||_{C^1} is ambiguous because φ_i is originally an eigenfunction on Σ while the norm is taken on M_t; the text should specify that φ_i denotes the transplanted function, and likewise in Theorem 1.3.","section":"§3.3, Lemma 3.7"}],"recommendation":"major_revision","confidential_remarks":"The paper has a sound overall architecture and the torus-curve sharpness discussion is a genuine strength. The missing coefficient-vector estimate in Theorem 1.3 is the main obstacle; it appears repairable, but it needs to be written out rather than asserted. The Theorem 1.4 uniformity issue is smaller but should also be fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Good paper, worth refereeing carefully. The main results are real: Theorem 1.1 extends the Colding-Minicozzi cylinder codimension bound to arbitrary compact shrinkers, with a sharp rate assumption; Theorem 1.3 gives a more flexible bound using the modified space P-tilde_d of linear caloric functions; and the rigidity theorem (Theorem 4.1) is a clean Carleman argument under super-exponential convergence. The torus curve discussion is also useful, and the authors are honest that the motivating example comes from AAAW13.\n\nThe proof strategy is coherent and mostly careful. The genuine soft spot is in the proof of Theorem 1.3: after orthonormalizing the extra functions at t = -Omega^{m_q+1}, the authors assert |V_{r1+1}| <= C Omega^{m_q(-lambda1+rho)} with no derivation. This bound is load-bearing: it is exactly what makes the kappa term in Lemma 3.7 go to zero against the decay hypothesis. A generic orthonormal linear function only gives |U| ~ (-t)^{-1/2} = Omega^{-m_q/2}, which is weaker when lambda1 > 1/2. The text says the bound follows from the proof of Theorem 1.1, but that isn't immediate; the normalization is done at a different time, and the orthogonal decomposition into eigenfunction directions is not written out. I don't think it's a fatal gap; it's the kind of spectral and normalization bookkeeping a referee can ask to be filled in. The analogous step in Theorem 1.1 is easier and likely fine.\n\nTwo other remarks. First, the assumptions of Theorem 1.3 are heavy (psi_i, monotone scaling, second-order Phi decay), and as the authors admit, the torus curve doesn't satisfy the Phi decay condition, so the motivating example is not covered by the stronger theorem. That is a limitation but not a flaw. Second, the citation pattern looks fine: Corollary 1.2 uses Baldauf-Sun as an external entropy bound, and the AAAW13 example is properly credited.\n\nNet: the paper is a solid contribution, clearly written, and the soft spots are local. I would send it to a serious referee. The referee should push for a full derivation of the coefficient-vector estimate in Theorem 1.3 and a cleaner statement of the spectral assumption.","headline":"A serious extension of the Colding-Minicozzi codimension program to general compact shrinkers, with one underived but likely repairable estimate in the proof of Theorem 1.3.","tokens_in":18300,"tokens_out":3331,"would_cite":true,"duration_ms":31125,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53C44"],"pacs":[],"model":"deepseek-v4-flash","headline":"An ancient mean curvature flow that converges quickly enough to its tangent flow at $-\\infty$ is forced to have the same codimension as that tangent flow, and with sufficiently fast convergence it must be exactly equal to it.","keywords":["ancient mean curvature flow","codimension","self-shrinker","tangent flow at -∞","drift Laplacian","caloric functions","curve shortening flow","rigidity"],"falsifier":"Build or find an ancient mean curvature flow whose rescaled flow is a $C^1$ graph over a compact shrinker with error decaying at the rate $(-t)^{1/2-\\delta}\\varepsilon(t)\\to0$ for some $\\delta<\\lambda_1$, but whose codimension is strictly larger than that of the shrinker; the theorem predicts no such flow exists. At the borderline rate the torus-curve example already shows the conclusion is false, so the decay gap is the only place to attack.","tokens_in":17216,"feed_emoji":"🌀","tokens_out":6438,"duration_ms":61365,"temperature":0.7,"pith_summary":"This paper establishes that an ancient mean curvature flow that, after rescaling by $\\sqrt{-t}$, converges to a compact shrinker fast enough is constrained in two ways: it cannot live in a higher-dimensional ambient space than the shrinker does, and with sufficiently rapid convergence it must be exactly the shrinker. The threshold is quantitative: convergence at rate $(-t)^{1/2-\\delta}$ for some $\\delta$ below the first nonzero eigenvalue $\\lambda_1$ of the drift Laplacian forces equal codimension, while convergence at merely the borderline rate permits higher codimension, as the paper's example of torus curves shows. A separate theorem shows that if the graph function decays faster than $e^{-\\alpha t}$ in an exponential sense, the flow is rigidly identical to the shrinker. These results matter because ancient solutions model singularity formation, and knowing that their geometry is controlled by their behavior at $-\\infty$ gives a practical way to classify them.","feed_headline":"Convergence rate at -∞ decides an ancient flow's codimension","feed_subtitle":"A sharp threshold: faster than the first drift eigenvalue forces equality; at borderline rates higher codimension can appear.","key_machinery":"The central objects are the drift Laplacian on the limiting shrinker and the caloric functions on the moving flow. The first eigenvalue $\\lambda_1$ of the drift Laplacian sets the critical convergence rate: the codimension bound holds only when the graph error decays faster than $(-t)^{-1/2+\\lambda_1}$. The proof controls the dimension of spaces of ancient solutions to the heat equation along the flow (caloric functions) with polynomial growth, using Gaussian-weighted norms; coordinate functions of the immersion are caloric, so bounding that dimension bounds the ambient dimension. Two inequalities do the work: a Poincar\\'e-type upper bound comparing Gaussian norms of caloric functions to gradients via the Rayleigh quotient on the shrinker, and a lower bound from orthonormalizing functions at times $-\\Omega^m$. For the rigidity theorem, a Carleman inequality for functions on $\\Sigma\\times(-\\infty,0)$ converts the assumed exponential decay of the graph function into vanishing of the graph function.","core_discovery":"The central discovery is a sharp codimension rigidity theorem for ancient mean curvature flows. If an ancient flow $M_t^n \\subset \\mathbb{R}^N$ has rescaled flow $M_t/\\sqrt{-t}$ forming a $C^1$ graph over a fixed compact shrinker $\\Sigma$ with error $\\varepsilon(t)$ satisfying $\\lim_{t\\to-\\infty}(-t)^{1/2-\\delta}\\varepsilon(t)=0$ for some $\\delta<\\lambda_1$, then $\\operatorname{codim}(M_t)=\\operatorname{codim}(\\Sigma)$. The proof works by showing that the space of slowly growing caloric functions on the flow has the same dimension as on the shrinker: constants plus coordinate functions, so no extra ambient directions can appear. The same machinery, with eigenfunction-transplantation assumptions and a spectral gap, yields a weaker bound $\\operatorname{codim}(M_t)\\le\\operatorname{codim}(\\Sigma)+r_1$ when convergence is only at the borderline rate. Finally, for graphs over a closed shrinker, a Carleman inequality proof shows that if $\\limsup_{t\\to-\\infty}\\varphi^2e^{-\\alpha t}=0$, then $\\varphi\\equiv0$, so the rescaled flow is exactly the shrinker.","pith_inferences":["The same caloric-function machinery likely extends to noncompact shrinkers, since the proofs work at the level of spectral gaps and Gaussian weights; the paper explicitly leaves that setting open.","The borderline torus-curve examples suggest a testable dichotomy: any ancient flow converging exactly at the borderline rate should either have extra codimension or be accounted for by eigenfunctions below $1/2$; one could search for other solutions realizing the bound $\\operatorname{codim}(\\Sigma)+r_1$ in Theorem 1.3.","The Carleman rigidity argument is local in the graph function and likely extends to higher codimension by tracking normal-bundle components, replacing the scalar $\\varphi$ by a section of the normal bundle.","The sharp constant conjecture for the entropy bound could be probed computationally by enumerating torus-curve-like solutions with different winding numbers and checking whether any ancient curve shortening flow in high codimension beats $\\sup_t\\lambda(M_t)\\ge (N/2)\\lambda(S^1)$."],"forward_implications":["An ancient flow that converges to a compact shrinker faster than the $\\lambda_1$ threshold cannot escape into higher codimension; its entire history lies in the same Euclidean subspace as the shrinker.","For ancient curve shortening flows, if the rescaled flow is a $C^1$ graph over the multiplicity-$m$ circle with the stated rate and only type I singularities, the flow is forced to be the shrinking circle with multiplicity $m$.","The torus-curve solutions show the rate is sharp: at the borderline rate $(-t)^{1/2-\\lambda_1}\\varepsilon(t)\\le C$, one can have tangent flow a multiplicity-$k_m$ circle yet live in arbitrarily high codimension.","Under faster-than-exponential convergence in the graph sense, an ancient mean curvature flow is identical to its tangent flow at $-\\infty$, not merely close to it.","The type I assumption in the circle rigidity result is necessary: there exist nontrivial rescaled ancient flows converging to $mS^1$ exponentially fast but with type II singularities."],"supporting_citations":[{"why":"Supplies the torus-curve ancient solutions whose limiting behavior shows the convergence rate in Theorem 1.1 is sharp and motivates the eigenfunction-assumption in Theorem 1.3.","marker":"[AAAW13]"},{"why":"Establishes the cylinder codimension-bound paradigm, the caloric-function spaces, the Gaussian inner product, the Poincar\\'e-type lemmas, and the orthonormalization lower bound that the paper generalizes.","marker":"[CM19b]"},{"why":"Provides the sharp entropy bound for plane curves used to force the multiplicity-$m$ circle rigidity in Corollary 1.2.","marker":"[BS18]"},{"why":"Supplies the graph evolution equation and normal component computation used in the proof of the rapid-convergence rigidity theorem.","marker":"[Wan14]"},{"why":"Provides the construction showing the type I hypothesis is necessary by giving a nontrivial exponentially converging rescaled flow with type II singularities.","marker":"[CHH18]"},{"why":"Supplies the weighted monotonicity formula used repeatedly in the caloric-function upper-bound estimates.","marker":"[Eck04]"}],"fun_headline_variants":["Fast convergence to shrinker forces equal codimension","Drift eigenvalue threshold controls ancient flow's codimension","Rapidly converging ancient flow is its tangent flow","Codimension rigidity: faster than first drift eigenvalue, none extra"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole codimension result rests on the assumption that the rescaled flow converges to the shrinker at a rate faster than the borderline rate set by the first nonzero eigenvalue: $(-t)^{1/2-\\delta}\\varepsilon(t)\\to0$ for some $\\delta<\\lambda_1$, and the paper's own torus-curve example shows that at the borderline rate $(-t)^{1/2-\\lambda_1}\\varepsilon(t)\\le C$, higher codimension does occur.","fun_headline_variants_meta":{"raw":{"variants":["Fast convergence to shrinker forces equal codimension","Drift eigenvalue threshold controls ancient flow's codimension","Rapidly converging ancient flow is its tangent flow","Codimension rigidity: faster than first drift eigenvalue, none extra"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000672,"raw_usage":{"total_tokens":3032,"prompt_tokens":890,"completion_tokens":2142,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":506,"completion_tokens_details":{"reasoning_tokens":2078}},"tokens_in":506,"tokens_out":2142,"duration_ms":16989,"temperature":1.0,"reasoning_tokens":2078,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:48:04.520250+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build or find an ancient mean curvature flow whose rescaled flow is a $C^1$ graph over a compact shrinker with error decaying at the rate $(-t)^{1/2-\\delta}\\varepsilon(t)\\to0$ for some $\\delta<\\lambda_1$, but whose codimension is strictly larger than that of the shrinker; the theorem predicts no such flow exists. At the borderline rate the torus-curve example already shows the conclusion is false, so the decay gap is the only place to attack.","supporting_citations":[],"review_version":1}