{"id":"5374749d-5c3a-49cf-b165-87a2310356c9","arxiv_id":"1908.09087","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A corrected Crouzeix-Raviart finite element eigenvalue converges from below to the exact Steklov eigenvalue with variable coefficients, at the same order as the uncorrected approximation.","lead":"This mathematics paper proves a way to compute lower bounds for Steklov eigenvalues with variable coefficients using corrected finite element approximations. The method guarantees the corrected eigenvalue is below the exact one without losing convergence speed.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The d=3 case of the main theorem rests on Lemma 2.1, whose proof cites only an unpublished preprint [33] for the key boundary estimate; if (2.8) is not established there, Theorem 3.1 does not follow for d=3.","rationale":"The reader's CONDITIONAL verdict is appropriate, and the most load-bearing concern is the d=3 branch of Lemma 2.1, which is an explicit proof gap rather than merely an unverified hypothesis. The proof of Lemma 2.1 for Ω⊂R3 is a single sentence referring to Theorem 4 of [33], a preprint by the same authors; the current manuscript does not reproduce or independently justify that argument. The proof of Theorem 3.1 relies on (2.8) in (3.14) to control the boundary term 2λ_h b(u−I_h u, u_h), and without (2.8) the bound (3.15) and hence the sign of the right-hand side of (3.16) is not established. The manuscript also does not spell out how the cited d=2 or constant-coefficient estimates in [16,1,25] adapt to variable α and β in d=3. This is a gap in the written proof, not a demonstrated falsehood. The alternative concern about the hypothesis ‖u−u_h‖_h ≥ C h^{1+r/2} is less damaging: for a fixed eigenvalue and a non-piecewise-linear eigenfunction, CR interpolation error is typically O(h^r), which dominates h^{1+r/2}, so that assumption is plausibly automatic although not proved. Both issues were noted by the reader; my concrete check targets the d=3 gap because it is decisive for half of the dimensional scope of the advertised result.","tokens_in":11606,"tokens_out":25053,"duration_ms":265157,"concrete_test":"Retrieve arXiv:1808.01609 and inspect Theorem 4. Check whether it states (2.6)-(2.8) for d=3 and for variable coefficients α,β as in (1.1), and whether (2.8) is proved there or follows by the Nitsche argument cited for d=2. Then write out the d=3 proof of Lemma 2.1 in full, identifying exactly where Theorem 4 is used. If (2.8) cannot be derived, Theorem 3.1 and Theorem 3.2 should be restricted to d=2; if [33] is unavailable or does not contain the needed estimate, the d=3 result remains unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2, Lemma 2.1 is the foundation for Theorem 3.1, but for Ω⊂R3 its proof is a single sentence: \"using similar arguments ... as well as referring to Theorem 4 in [33]\", where [33] is an unpublished 2018 arXiv preprint by the same group. The lemma must supply (2.6)-(2.8), and in particular the boundary estimate (2.8), ‖u−u_h‖_{0,∂Ω} ≤ C h^r ‖u−u_h‖_h. This estimate is used critically in (3.14) to control the nonconforming consistency term 2λ_h b(u−I_h u, u_h) in the proof of λ ≥ λ_c^h. If [33, Thm 4] provides only the energy and eigenvalue estimates but not (2.8) for d=3, or if it assumes constant coefficients or stronger regularity than r∈(0,1/2), then the d=3 branch of the central theorem is not established by the manuscript. The cited [16,1,25] appear to concern d=2 or constant-coefficient settings, so no independent route is visible in the text. This is a gap in the written proof, not a demonstration that the theorem is false.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a correction formula for Crouzeix–Raviart finite element eigenvalue approximations for the Steklov eigenvalue problem (1.1) with variable coefficients α, β on polygonal domains in R² and R³. The corrected eigenvalue λ_c^h is defined in (3.5), and Theorem 3.1 states that, under the conditions of Lemma 2.1 and an additional lower-bound assumption on the discrete error, λ ≥ λ_c^h for sufficiently small mesh size. Theorem 3.2 gives the identity λ − λ_c^h = λ − λ_h + λ_h M/(λ_h + M) with M = O(h²), showing the corrected eigenvalue converges at the same order as the uncorrected CR eigenvalue. Numerical experiments on square, L-shaped, hexagonal, cube, and Fichera corner domains with α = β = 1 illustrate the theoretical convergence order.","tokens_in":11873,"tokens_out":9174,"duration_ms":84597,"significance":"If the main theorem is fully established, the paper would extend asymptotic lower-bound results for Steklov eigenvalues to variable coefficients and remove the earlier restrictions of singular eigenfunctions or sufficiently large eigenvalues. The correction formula is explicit and computationally cheap, and Theorem 3.2 is a clean algebraic statement. However, the d = 3 branch of the main theorem depends on an unpublished preprint by the same group for the key a priori estimates, and the numerical experiments use only constant coefficients, so the variable-coefficient claim is not tested experimentally. These issues limit the present significance and require verification before the result can be considered fully reliable.","major_comments":[{"comment":"For Ω ⊂ R³, the proof of Lemma 2.1 is a single sentence that refers to Theorem 4 of [33], an unpublished arXiv preprint by the same authors. The boundary estimate (2.8), which is used critically in (3.14)–(3.15) to control the nonconforming consistency term, must be established for variable coefficients and for all r ∈ (0, 1/2) in three dimensions. As written, the manuscript does not verify that [33, Thm 4] supplies this estimate under the present assumptions; the cited works [16, 1, 25] appear to address two-dimensional or constant-coefficient settings. The d = 3 branch of Theorem 3.1 is therefore not independently verifiable from the manuscript.","section":"§2, Lemma 2.1 and §3, Theorem 3.1"},{"comment":"All numerical experiments take α = β = 1, so the coefficient variation term (α − I_0 α) in the correction (3.5) vanishes identically. The central novelty of the paper is the variable-coefficient setting, and the correction's effectiveness for nonconstant α and β is not tested. The authors should include at least one example with genuinely variable α and/or β on a polygonal domain where the exact eigenvalue or a high-accuracy reference is available, and they should state the value of δ used in the correction.","section":"§4, Numerical experiments"},{"comment":"The abstract and introduction state that the corrected eigenvalues are lower bounds whether eigenfunctions are singular or smooth and whether eigenvalues are large or not, but Theorem 3.1 requires the additional assumption ‖u − u_h‖_h ≥ C h^{1+r/2}. This assumption is not mentioned in the abstract and is not proved from the problem data; it is likely automatic in typical cases, but it is a genuine hypothesis of the theorem and should be stated whenever the result is summarized.","section":"Abstract and §1"}],"minor_comments":[{"comment":"The title contains typographical errors: 'EIGENV ALUES' should be 'EIGENVALUES', and 'EIGENV ALUE' should be 'EIGENVALUE'.","section":"Title"},{"comment":"The power of h in the fourth term of (3.16) appears as C h^{1/2+r}, whereas the derivation just before it yields C h^{r/2}. The correct exponent should be r/2; this is a typo in the display, though the argument is unaffected because both terms are higher-order under the assumption.","section":"§3, text near (3.16)"},{"comment":"The entry '0.39329 159' contains an unintended space; it should be '0.39329159'.","section":"§4, Table 1"},{"comment":"The description of the Fichera corner domain reads '[−1, 1]^3 \\ (−1, )]^3', which appears to be a typesetting error; the omission should be corrected.","section":"§4.2"},{"comment":"The phrase 'knot that (3.9) is valid' at the end of the proof of Theorem 3.1 should read 'know that'.","section":"§1 and §4"},{"comment":"Reference [33] is an unpublished arXiv preprint; if it remains the sole support for the d = 3 estimates, the authors should either provide the proof in the manuscript or clearly indicate that [33] has been accepted for publication and give a citable version.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The d = 3 gap in Lemma 2.1 is the primary technical concern: the key boundary estimate rests on a self-cited unpublished preprint, and no independent argument is given. In addition, the numerical section tests only constant coefficients, which does not exercise the variable-coefficient correction. Both issues are fixable within the scope of the paper, so a major revision is appropriate rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this paper proves a useful extension — corrected Crouzeix-Raviart eigenvalue approximations give asymptotic lower bounds for the Steklov problem with variable coefficients in 2D and 3D, maintaining the same order as the uncorrected approximations. That's genuinely new; prior work was either constant-coefficient or lost order with the correction. The central identity (3.2) and the argument around it are clean and check out.\n\nWhat's new and good: the correction formula (3.5) is cheap, and the proof of Theorem 3.1 is largely elementary once you accept Lemma 2.1. Theorem 3.2 is a simple but useful observation that the correction doesn't degrade the rate. The explicit constants in Lemmas 2.2 and 2.3 are a nice touch.\n\nSoft spots, in proportion. The d=3 branch of Lemma 2.1 — and therefore of the whole lower-bound theorem — rests on a single sentence referencing Theorem 4 of [33], a 2018 arXiv preprint by the same group. That's not enough for a referee to verify the key boundary estimate (2.8) in 3D. If Thm 4 of [33] supplies it, fine, but as written the proof is incomplete in d=3.\n\nSecond, the assumption ||u-u_h||_h ≥ C h^{1+r/2} is buried in Theorem 3.1, not in the abstract. It's a lower bound on the discrete error, which is not guaranteed by the problem data. For the theorem to be useful, the authors should say when this holds or at least comment on it. As it stands, the abstract's claim \"whether the eigenfunctions are singular or smooth\" is stronger than what the theorem actually states.\n\nThird, the numerical section sets α=β=1 in every experiment. The variable-coefficient claim is tested nowhere. The method parameter δ is also not stated, so the tables can't be reproduced. These are minor-to-moderate omissions, but they matter for a paper whose headline is variable coefficients.\n\nThe citation pattern is mostly fine; relying on an unpublished preprint is the only awkward spot. I don't read the paper as overclaiming beyond those gaps. The central idea is sound and the extension is real.\n\nWho benefits: people working on guaranteed or asymptotic bounds for Steklov eigenvalues, especially with nonconforming methods. It's not field-shaping, but it's a solid incremental result.\n\nMy recommendation: send it to peer review. A referee should ask for the d=3 proof to be spelled out or for [33] to be made available, for a comment on the assumption on the error norm, and for at least one variable-coefficient numerical experiment. Those are addressable in revision.","headline":"Useful extension of corrected CR lower bounds to variable-coefficient Steklov problems; d=3 proof leans on an unpublished preprint and experiments never vary coefficients, but the core is sound enough to referee.","tokens_in":12388,"tokens_out":2464,"would_cite":true,"duration_ms":21169,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N25","65N30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that a simple correction to Crouzeix-Raviart finite element eigenvalues yields asymptotic lower bounds for the Steklov eigenvalue problem with variable coefficients, without requiring singular eigenfunctions or large…","keywords":["correction","Steklov eigenvalue problem","Crouzeix-Raviart finite element","lower bounds","convergence order","variable coefficients","nonconforming finite element","eigenvalue approximation"],"falsifier":"On a convex domain with smooth eigenfunction, compute the exact eigenvalue and the CR eigenvalue on successively refined meshes while monitoring $\\|u-u_h\\|_h$. If $\\|u-u_h\\|_h$ decays faster than $C h^{1+r/2}$, the proof of Theorem 3.1 does not apply; if the corrected eigenvalue $\\lambda_c^h$ ever lies above the exact eigenvalue for small $h$, the claimed lower-bound property fails as stated. A separate check is to compare Lemma 2.1 against a known exact solution on tetrahedral meshes to confirm the three-dimensional error rate and constants.","tokens_in":11368,"feed_emoji":"📉","tokens_out":8558,"duration_ms":71810,"temperature":0.7,"pith_summary":"This paper establishes that, with a small post-processing correction, the standard Crouzeix-Raviart finite element approximations to the Steklov eigenvalue problem with variable coefficients are asymptotic lower bounds for the true eigenvalues. Earlier work had proved such lower bounds only when the corresponding eigenfunctions are singular or when the eigenvalues are large enough; this paper removes those conditions. The correction is a simple multiplicative factor built from local oscillations of the coefficients and the computed eigenfunction, and it is shown to converge at the same rate as the uncorrected approximation. The argument covers two- and three-dimensional polygonal domains and is supported by numerical experiments on smooth and singular cases.","feed_headline":"Corrected eigenvalues converge to Steklov eigenvalues from below","feed_subtitle":"The Crouzeix-Raviart finite element method gets a cheap correction and keeps its convergence rate.","key_machinery":"The machinery is the correction formula (3.5): $\\lambda_c^h=\\lambda_h/(1+M/\\lambda_h)$, where $M=\\frac{\\delta}{\\alpha_0}\\sum_\\kappa(\\|(\\alpha-I_0\\alpha)\\nabla u_h\\|_{0,\\kappa}+C h_\\kappa\\|\\beta u_h\\|_{0,\\kappa})^2$, with $\\delta>1$ an arbitrary constant, $I_0$ the piecewise constant interpolant, $\\alpha_0$ the positive lower bound of $\\alpha$, and $C$ an explicit constant from Poincare- and trace-type inequalities. The correction is designed so that, inside the eigenvalue-difference identity (3.2), the boundary and interpolation terms are dominated by the positive energy term $(1-1/\\delta)\\alpha_0\\sum_\\kappa|u-u_h|^2_{1,\\kappa}+\\sum_\\kappa\\int_\\kappa\\beta(u-u_h)^2dx$. The identity (3.2) is the engine: it decomposes $\\lambda-\\lambda_h$ into a discrete energy error, a boundary term, and interpolation-orthogonality terms, and the assumption $\\|u-u_h\\|_h\\ge C h^{1+r/2}$ is exactly what the proof uses to control the boundary contribution.","core_discovery":"The central claim is Theorem 3.1: for the Steklov problem $-\\operatorname{div}(\\alpha\\nabla u)+\\beta u=0$ in $\\Omega$ with $\\alpha\\partial u/\\partial\\nu=\\lambda u$ on $\\partial\\Omega$, the corrected CR eigenvalue $\\lambda_c^h$ defined by (3.5) satisfies $\\lambda\\ge\\lambda_c^h$ for all sufficiently small mesh size $h$, provided the discrete error satisfies $\\|u-u_h\\|_h\\ge C h^{1+r/2}$. Theorem 3.2 then gives the exact identity $\\lambda-\\lambda_c^h=\\lambda-\\lambda_h+\\lambda_h M/(\\lambda_h+M)$ with $|M|\\le C h^2$, so the corrected eigenvalue keeps the same convergence order as the uncorrected CR eigenvalue. Taken together, the paper reads these results as removing the earlier restrictions that the eigenfunction be singular or the eigenvalue be large enough: a corrected lower bound now holds for smooth and singular eigenfunctions alike, on two- and three-dimensional polygonal domains.","pith_inferences":["The structure of the correction suggests a general recipe for variable-coefficient eigenvalue problems: subtract a term proportional to the local oscillation of the coefficients on each element, whenever the nonconforming space has the interpolation orthogonality property used in (2.10).","If a computable a posteriori bound could replace the assumption $\\|u-u_h\\|_h\\ge C h^{1+r/2}$, the corrected eigenvalue would become a guaranteed lower bound rather than an asymptotic one; the numerical results are consistent with the bound holding in all tested cases.","The identity in Theorem 3.2 separates the corrected error into the standard error plus an $O(h^2)$ term involving a computable quantity $M$, which points toward verified enclosure methods if a certified upper bound on $M$ is available."],"forward_implications":["Once the mesh is fine enough, the corrected eigenvalue $\\lambda_c^h$ is a lower bound for the exact eigenvalue in cases where the original CR approximation converges from above.","The corrected eigenvalue converges at the same asymptotic order as the uncorrected CR eigenvalue, so the lower-bound property does not cost convergence rate.","Since $M$ is built only from local quantities and the computed CR eigenfunction, the correction adds essentially no computational time beyond the original solve.","The same correction recipe applies to the enriched Crouzeix-Raviart element, extending the lower-bound property to variable-coefficient Steklov problems for that element as well.","In the numerical tests, the average $(\\lambda_h+\\lambda_c^h)/2$ gives a more accurate approximation than the corrected value alone and often no worse accuracy than the uncorrected value."],"supporting_citations":[{"why":"Supplies the asymptotic lower-bound theory for the Steklov problem with constant coefficients and the identity that Lemma 3.1 generalizes; also provides two-dimensional CR error estimates used in Lemma 2.1.","marker":"[16]"},{"why":"Provides the interpolation orthogonality property (2.10), which is used throughout the proof to eliminate the leading interpolation error terms.","marker":"[2]"},{"why":"Supplies the explicit trace inequality constants used in Lemma 2.3 to control the boundary terms in the eigenvalue-difference identity.","marker":"[34]"},{"why":"Provides the explicit Poincare-type constants used in Lemma 2.2 for the elementwise interpolation error estimates.","marker":"[22]"},{"why":"Supports the three-dimensional CR-element error estimates in Lemma 2.1; it is an unpublished preprint by the same authors, so it is load-bearing but not independently verifiable in this manuscript.","marker":"[33]"},{"why":"Used in the derivation of the spectral approximation estimates that underlie Lemma 2.1.","marker":"[1]"},{"why":"Supplies a posteriori error estimates for nonconforming Steklov approximations, used together with [1] to establish Lemma 2.1.","marker":"[25]"},{"why":"Supports the boundary-norm estimate (2.8) through a Nitsche-type technique.","marker":"[32]"}],"fun_headline_variants":["CR correction gives lower Steklov eigenvalue bounds","Variable-coefficient Steklov lower bounds from corrected CR","Corrected CR eigenvalues bound Steklov from below","Sharp lower bounds for Steklov with variable coefficients","Lower Steklov bounds via corrected Crouzeix-Raviart"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the error estimates of Lemma 2.1 hold in three dimensions with the stated rates and constants, and that the discrete error never decays faster than $C h^{1+r/2}$; the three-dimensional estimate rests on an unpublished preprint by the same authors, and the lower-bound assumption is not guaranteed by the problem data alone.","fun_headline_variants_meta":{"raw":{"variants":["CR correction gives lower Steklov eigenvalue bounds","Variable-coefficient Steklov lower bounds from corrected CR","Corrected CR eigenvalues bound Steklov from below","Sharp lower bounds for Steklov with variable coefficients","Lower Steklov bounds via corrected Crouzeix-Raviart"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000177,"raw_usage":{"total_tokens":1245,"prompt_tokens":849,"completion_tokens":396,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":465,"completion_tokens_details":{"reasoning_tokens":315}},"tokens_in":465,"tokens_out":396,"duration_ms":4127,"temperature":1.0,"reasoning_tokens":315,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:23:06.025806+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a convex domain with smooth eigenfunction, compute the exact eigenvalue and the CR eigenvalue on successively refined meshes while monitoring $\\|u-u_h\\|_h$. If $\\|u-u_h\\|_h$ decays faster than $C h^{1+r/2}$, the proof of Theorem 3.1 does not apply; if the corrected eigenvalue $\\lambda_c^h$ ever lies above the exact eigenvalue for small $h$, the claimed lower-bound property fails as stated. A separate check is to compare Lemma 2.1 against a known exact solution on tetrahedral meshes to confirm the three-dimensional error rate and constants.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the asymptotic lower-bound theory for the Steklov problem with constant coefficients and the identity that Lemma 3.1 generalizes; also provides two-dimensional CR error estimates used in Lemma 2.1."},{"cited_title":"Armentano, R.G","cited_arxiv_id":null,"evidence_quote":"Provides the interpolation orthogonality property (2.10), which is used throughout the proof to eliminate the leading interpolation error terms."},{"cited_title":"Guaranteed eigenvalue bounds for the Steklov eigenvalue problem","cited_arxiv_id":"1808.08148","evidence_quote":"Supplies the explicit trace inequality constants used in Lemma 2.3 to control the boundary terms in the eigenvalue-difference identity."},{"cited_title":"Liu : A framework of verified eigenvalue bounds for self-adjoint differential operators, Appl","cited_arxiv_id":null,"evidence_quote":"Provides the explicit Poincare-type constants used in Lemma 2.2 for the elementwise interpolation error estimates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the three-dimensional CR-element error estimates in Lemma 2.1; it is an unpublished preprint by the same authors, so it is load-bearing but not independently verifiable in this manuscript."},{"cited_title":"Alonso, A.D","cited_arxiv_id":null,"evidence_quote":"Used in the derivation of the spectral approximation estimates that underlie Lemma 2.1."},{"cited_title":"Russo, A.E","cited_arxiv_id":null,"evidence_quote":"Supplies a posteriori error estimates for nonconforming Steklov approximations, used together with [1] to establish Lemma 2.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the boundary-norm estimate (2.8) through a Nitsche-type technique."}],"review_version":1}