{"id":"f2469d10-0b5a-4f8d-b66b-fb21b8e09e1b","arxiv_id":"1908.06041","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A dichotomy for complex Robin eigenvalues at large boundary parameter, with new numerical range bounds, a Dirichlet-to-Neumann proof, and interval, rectangle and ball asymptotics.","lead":"This paper studies the eigenvalues of the Robin Laplacian, an operator with a complex boundary parameter, on bounded Lipschitz domains. It proves that as the boundary parameter grows, every analytic eigenvalue curve either converges to a Dirichlet eigenvalue or escapes to infinity, and it gives new numerical range bounds and explicit asymptotics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Global eigenvalue labelling in Thm 4.1(2) is not justified: self-adjoint holomorphic families can have branch points, so Thm 1.5's quantification over 'each analytic eigenvalue curve' needs a local reformulation.","rationale":"The reader's weakest assumption correctly identifies the global labelling of eigenvalue curves in Theorem 4.1(2) as the load-bearing condition for the central dichotomy. The specific typo in Theorem 4.1(1) (A(α)*=A(α) instead of A(α)*=A(\\bar α)) is easily corrected, but the deeper difficulty remains: Kato's self-adjoint-holomorphic-family theory does not imply global single-valued meromorphic eigenvalue branches. Algebraic branch points at non-real crossings are a standard phenomenon, as the 2x2 example shows. The paper gives no argument ruling out such branch points for the Robin Laplacian on a general Lipschitz domain, and the explicit Bessel-form representation for the ball makes a computational test feasible. If such a branch point is found, the statement of Theorem 1.5 is not false in spirit (the dichotomy for each local analytic branch should still follow from the Dirichlet-to-Neumann argument), but the paper's global formulation and Theorem 1.1(2) would need substantial revision. This does not change the reader's CONDITIONAL verdict, since the central DtN-based dichotomy appears sound and the issue is one of precision and proof of global structure rather than a demonstrated counterexample to the dichotomy itself.","tokens_in":46388,"tokens_out":20950,"duration_ms":214658,"concrete_test":"For the unit ball B^2, use the explicit Dirichlet-to-Neumann expression M_l(λ)=√λ J_{l+1}(√λ)/J_l(√λ) - l from (9.19) (with d=2). Numerically search for zeros of d M_l/dλ in C off the Dirichlet spectrum, for l=0,1,2,.... If a zero λ* exists, then the inverse eigenvalue branch λ(α) has a branch point at α*=M_l(λ*), contradicting the global meromorphic extension in Theorem 4.1(2). Repeat for d=3 with half-integer Bessel indices; a single confirmed zero falsifies the global labelling claim and shows Theorem 1.5 needs a local-branch formulation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 4.1(2) asserts that each eigenvalue λ_k(α) extends to a meromorphic function on all of C, with at most algebraic singularities at non-real crossing points. The proof invokes Kato's local eigenvalue continuation and then a compact-exhaustion argument. But Kato's theory for holomorphic families, even self-adjoint holomorphic families with the corrected identity A(α)*=A(\\bar α), only guarantees local analytic branches away from crossings; at crossings the branches can have square-root-type algebraic branch points, not poles. A simple self-adjoint holomorphic family A(κ)=[[κ,1],[1,-κ]] satisfies A(κ)*=A(\\bar κ) and has eigenvalues ±√(κ^2+1), with branch points at κ=±i, so no single-valued meromorphic extension to C exists. Thus the global labelling claimed in Theorem 4.1(2) does not follow from the cited Kato theory and is likely false for the Robin Laplacian whenever two eigenvalue curves collide at a non-real parameter value. This is load-bearing for Theorem 1.5: its statement 'each analytic eigenvalue curve' presupposes well-defined global curves. If branch points exist, the dichotomy must be reformulated for locally defined analytic branches, and the exhaustiveness argument in Remark 4.3 becomes path-dependent. The reader's correction of the adjoint identity to A(α)*=A(\\bar α) is necessary but not sufficient: it restores Kato's hypothesis but does not remove the branch-point obstruction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the Robin Laplacian on bounded Lipschitz domains with a complex Robin parameter α. It establishes basic operator-theoretic properties (m-sectoriality, holomorphic dependence of eigenvalues and eigenprojections, Abel/Riesz basis results and failure of orthonormal bases for non-real α), gives new numerical-range and trace bounds with explicit constants, and uses a Dirichlet-to-Neumann duality to prove a dichotomy for eigenvalue branches as |α|→∞: each branch either converges to a point in the Dirichlet spectrum or diverges to infinity. The final sections work out detailed asymptotic expansions for intervals, hyperrectangles, and balls, and the paper closes with a conjecture for general smooth domains.","tokens_in":46708,"tokens_out":17962,"duration_ms":193630,"significance":"If the central claims can be made fully rigorous, this would be a substantial contribution to the spectral theory of non-self-adjoint Robin Laplacians. The Dirichlet-to-Neumann duality is developed carefully and is genuinely useful; the numerical-range bounds appear new even for real negative α; and the explicit interval, rectangle, and ball computations provide concrete, testable asymptotics. The paper is essentially self-contained, and I see no circularity: the DtN duality is independently proved, and the numerical-range bounds rest on trace inequalities rather than on the conclusions being drawn. The main reservations concern the global labelling of eigenvalue curves and a few statements whose proofs are too terse or one-sided.","major_comments":[{"comment":"The asserted identity A(α)* = A(α) is false for non-real α. The adjoint of the Robin Laplacian with parameter α is the Robin Laplacian with parameter Ā, so the correct self-adjoint-holomorphic condition is A(α)* = A(Ā). This matters because the subsequent proof of Theorem 4.1(2) relies on Kato's theory of self-adjoint holomorphic families; the reader must supply the corrected identity to even invoke that theory. Please correct the statement and make explicit that self-adjointness holds only for real α.","section":"Theorem 4.1(1), §4.1"},{"comment":"The claim that each λ_k(α) extends to a meromorphic function on all of C, with only algebraic singularities at non-real crossings, is not justified by the cited Kato theory and is likely false in general. Kato's theorem gives local analytic branches away from crossing points, but at crossings the branches can have square-root-type branch points. For example, the family A(κ)=[[κ,1],[1,-κ]] satisfies A(κ)*=A(Ā) and has eigenvalues ±√(κ²+1), with branch points at κ=±i; no single-valued meromorphic extension to C exists. The compact-exhaustion argument in the proof does not address this monodromy obstruction. Since Theorem 1.5 and Remark 4.3 quantify over 'each analytic eigenvalue curve', the global-labelling issue is load-bearing: the dichotomy must either be proved for locally defined analytic branches along paths, or an additional no-branching property must be established for the Robin Laplacian.","section":"Theorem 4.1(2) and proof, §4.1"},{"comment":"The statement that for every λ∈C there exist α_k with |α_k|→∞ such that λ is itself a Robin eigenvalue for every k is stronger than the preceding accumulation statement, and the proof for λ in the Dirichlet spectrum is not supplied. The one-sentence appeal to a 'multivalued operator' in Remark 7.3(2) does not define the eigenvalue equation for that multivalued operator or show that it has solutions α_k with |α_k|→∞. If only an accumulation statement is intended, the 'more precisely' sentence should be weakened; if the exact statement is intended, it needs a real proof, especially at points of σ(−Δ_D).","section":"Theorem 1.6(2) and §8"},{"comment":"The proof assumes the inclusion σ(A(α)) ⊂ {Im z ≥ 0} and then that all eigenvalues have Im λ_k ≥ 0. This is valid only for Im α > 0. For Im α < 0 the numerical range lies in the opposite half-plane and the contradiction as written does not apply. Since the theorem claims the result for all α∈C\\R, the proof must cover both half-planes, for example by arguing with the adjoint or by replacing α by Ā after justifying that the assumed orthonormal basis property transfers.","section":"Theorem 5.1, §5"}],"minor_comments":[{"comment":"The phrase 'meromorphic function with at most algebraic singularities' is internally contradictory: algebraic branch points are not poles of a single-valued meromorphic function. Please rephrase the assertion in terms of analytic continuation and Puiseux series.","section":"Theorem 4.1(2)"},{"comment":"The sector notation is unnecessarily confusing, and the sentence beginning 'if π/2<θ′<π, we set define T+...' is incomplete. Also, the statement that λ=0 is a Dirichlet eigenvalue of the interval is inconsistent with the standard Dirichlet spectrum (π²j²/(4a²), j≥1); please clarify whether 0 is being treated only as a singular point of the DtN representation.","section":"Definition 2.2 and §2.1"},{"comment":"The index of the Bessel function in the statement and proof appears inconsistent: formula (9.19) and the definition of m_l in the proof point to m = d/2 + l − 1, while the displayed result uses d/2 − l + 1. Please check and correct.","section":"Theorem 9.8"},{"comment":"There are numerous typos in this section, including 'tideous' and 'ourselved'. The phrase 'α→∞ in the left half-plane away from the imaginary axis' should presumably be 'Re α→−∞'. These do not affect the mathematics but should be cleaned up.","section":"§9.1"}],"recommendation":"major_revision","confidential_remarks":"The global-branching issue is the main correctness risk: Theorem 4.1(2) overstates what Kato's theory provides, and Theorem 1.5 inherits that fragility. The remaining technical apparatus, especially the DtN duality and the explicit examples, appears sound and valuable. I do not see a circularity problem. The paper should not be accepted in its present form, but the concerns seem addressable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a serious referee. The DtN duality argument is the real contribution: it gives the dichotomy for eigenvalue branches, the numerical-range bound, and the new real-negative-α bound of Corollary 1.4. The explicit interval/rectangle/ball analyses and the accumulation result are also valuable. This is not a repackaging of earlier work.\n\nSoft spots, roughly in order of seriousness.\n\nFirst, Theorem 4.1(1) states A(α)* = A(α), which is false for non-real α. The correct self-adjoint-holomorphic condition is A(α)* = A(ᾱ). This is probably a typo, but it matters because the subsequent Kato arguments depend on the right formulation.\n\nSecond, the proof of Theorem 5.1 silently assumes Im α > 0. That can be repaired by symmetry, but as written the numerical range argument only covers one half-plane.\n\nThird, Theorem 9.8 has an index error: the Bessel order should be d/2 + l − 1, not d/2 − l + 1. Again minor, but it makes the formula confusing.\n\nThe larger issue is the global eigenvalue labelling in Theorem 4.1(2) and Remark 4.3. Kato's theory for holomorphic families gives local analytic branches; it does not give global single-valued branches. At non-real crossing points, square-root-type branch points can occur, as in the 2×2 example A(κ)=[[κ,1],[1,-κ]]. The compact-exhaustion argument in the proof does not rule out monodromy. Consequently, the phrase 'each analytic eigenvalue curve' in Theorem 1.5 is ambiguous. The dichotomy itself is salvageable: the DtN proof shows that any bounded sequence on a locally defined analytic branch converges to a point in the Dirichlet spectrum, so a local version of the theorem is true. But the global statement as written is not justified. I would ask the authors to either prove global single-valuedness or reformulate Theorem 1.5 and the surrounding discussion in terms of locally defined analytic branches.\n\nThe citation pattern is fine. The only self-citation is the survey [18], used for context and open problems, not as load-bearing input.\n\nBottom line: the main mathematical ideas are new and likely correct, and the paper deserves peer review after fixing the labelling issue and the smaller slips. I would recommend acceptance with major/minor revisions, and I would certainly want to see the revised version.","headline":"A genuinely new paper on complex-α Robin eigenvalues with a solid DtN core, but the global eigenvalue-labelling theorem overreaches and should be reframed.","tokens_in":47223,"tokens_out":4235,"would_cite":true,"duration_ms":43990,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35J05","35J25","35P10","35P15","35S05","47A10","81Q12"],"pacs":[],"model":"deepseek-v4-flash","headline":"For the Robin Laplacian with a complex boundary parameter, every analytic eigenvalue curve either converges to an eigenvalue of the Dirichlet Laplacian or diverges to infinity in the complex plane.","keywords":["Robin Laplacian","complex Robin parameter","Dirichlet-to-Neumann operator","eigenvalue asymptotics","non-self-adjoint spectral theory","analytic eigenvalue curves","numerical range","Abel basis"],"falsifier":"On a bounded Lipschitz domain, numerically follow a simple Robin eigenvalue curve λ(α) along a sequence α_k→∞ with Re α_k≥0; if a finite accumulation point outside the Dirichlet spectrum of -Δ_D is found, then Theorem 1.5 (and the accumulation-point theorem) would be false, and the Dirichlet-to-Neumann pole argument would have to break down.","tokens_in":46188,"feed_emoji":"📐","tokens_out":10114,"duration_ms":87424,"temperature":0.7,"pith_summary":"This paper studies the Robin Laplacian on a bounded Lipschitz domain when the boundary parameter α is complex, and it establishes that the large-α behaviour of every eigenvalue is fully dichotomous: each analytic eigenvalue curve λ(α) either converges to an eigenvalue of the Dirichlet Laplacian or diverges to infinity in C. The proof replaces the variational min-max techniques that work for real α with a duality between the Robin eigenvalue problem and the Dirichlet-to-Neumann operator, whose eigenvalues as functions of λ are exactly the Robin parameters. This duality gives a short proof of the dichotomy and, together with new numerical-range bounds, shows that when α escapes to infinity away from the negative real semi-axis all accumulation points are Dirichlet eigenvalues, whereas arbitrary complex accumulation points become possible if α is allowed to approach the negative real semi-axis. Explicit interval, hyperrectangle and ball examples yield divergent eigenvalue curves with leading behaviour -$α^{2}$, and the paper formulates a conjecture that on smooth domains in dimension at least two, all eigenvalues converge to the Dirichlet spectrum whenever Re α remains bounded below.","feed_headline":"Complex Robin eigenvalues either diverge or hit Dirichlet spectrum","feed_subtitle":"Using Dirichlet-to-Neumann duality, every eigenvalue curve of the Robin Laplacian is classified as α→∞ in C.","key_machinery":"The load-bearing object is the Dirichlet-to-Neumann operator M(λ), which maps given Dirichlet data g for solutions of -Δu=λu to the negative outer normal derivative of that solution. The duality is that λ is a Robin eigenvalue with parameter α if and only if α is an eigenvalue of M(λ), so the roles of α and λ are exchanged: instead of following λ(α), one follows the eigenvalue branches α(λ) of a meromorphic family whose poles are exactly the Dirichlet eigenvalues. That meromorphy, together with the fact that an eigenvalue branch of a holomorphic family cannot appear or disappear without crossing a singularity, is what forces the dichotomy of Theorem 1.5. A secondary set of tools supports this: sectoriality of the associated form (its numerical range lies in a fixed sector), a derivative formula λ'(α)=∫∂Ωψ²dσ/∫Ωψ²dx for simple eigenvalues, the numerical-range estimate that confines eigenvalues to a parabolic region, and the basis theorem for forms with a self-adjoint principal part showing the eigenfunctions form an Abel basis but never an orthonormal basis for non-real α.","core_discovery":"The central discovery is that the spectrum of the Robin Laplacian with complex parameter is globally organised into analytic eigenvalue curves, and as α→∞ in C each such curve either has all its finite accumulation points in the Dirichlet spectrum or leaves every compact set. This is proved by studying the Dirichlet-to-Neumann operator M(λ), which is meromorphic in λ with poles exactly at the Dirichlet eigenvalues; boundedness of an eigenvalue branch forces λ to approach a pole, and divergence of λ forces the corresponding parameter α to diverge. The same machinery yields a complementary classification of accumulation points: if α diverges in a sector away from the negative real semi-axis, all Robin eigenvalues have only Dirichlet eigenvalues as accumulation points, regardless of curve choice, while for arbitrary α_k→∞ near the negative real semi-axis every λ∈C can be realised as a Robin eigenvalue. For the interval, hyperrectangle and ball, the dichotomy is made quantitative, giving asymptotic expansions whose leading divergent term is -$α^{2}$ in all three cases.","pith_inferences":["If the global analytic eigencurve labelling is repaired by using the correct self-adjoint-holomorphic identity A(α)*=A(ar α) in place of the paper's stated A(α)*=A(α), the dichotomy and all subsequent results should survive unchanged, since the misstatement does not affect the meromorphy argument.","The Dirichlet-to-Neumann method should transfer directly to other non-self-adjoint boundary-value problems, including quantum graphs with complex δ-couplings, where the same duality can turn a spectral asymptotics problem into a pole analysis of a meromorphic matrix family.","The conjecture that all eigenvalues converge when Re α is bounded below, if true, implies that large imaginary parts of Robin eigenvalues can only be generated by α approaching the negative real semi-axis; this gives a concrete spectral-instability criterion for impedance-type boundary conditions.","For variable boundary coefficients α∈L∞(∂Ω), the numerical-range estimates persist but the analytic-curve dichotomy is likely to fail, and one could test this by constructing a boundary function that creates eigenvalue crossings or non-meromorphic parameter dependence."],"forward_implications":["For large complex impedance on a bounded Lipschitz domain, spectral accumulation cannot occur outside the Dirichlet spectrum as long as Re α is bounded below or |Re α/Im α| is controlled.","The dichotomy applies to real α as well, closing a previously listed open problem about whether Robin eigenvalues can converge to non-Dirichlet limits as α→∞.","On an interval, α→∞ in a left half-plane sector produces exactly two divergent eigenvalues with λ(α)=-α²+O(α²e^{2a Re α}), while all other eigenvalues converge to Dirichlet eigenvalues.","On hyperrectangles there are infinitely many divergent eigenvalue curves at levels -jα² for j=1,...,d-1 and exactly 2^d curves at level -dα², and on the ball the divergent curves satisfy λ(α)=-α²+(d-1)α+O(1).","The numerical-range bound Re λ≥-(C₁²/4)|Re α|²-C₂|Re α| is new even for real negative α on general Lipschitz domains and gives a two-sided bound on the principal eigenvalue as α→-∞."],"supporting_citations":[{"why":"Supplies the holomorphic-family formalism and the eigenvalue/eigenprojection continuation theorem on which Theorem 4.1 and the global eigencurve labelling rest.","marker":"[40]"},{"why":"Establishes the duality between Robin eigenvalues and Dirichlet-to-Neumann eigenvalues on domains, used as Theorem 7.6.","marker":"[7]"},{"why":"Provides the sectoriality and resolvent properties of the Dirichlet-to-Neumann operator on rough domains used in Lemma 7.2.","marker":"[8]"},{"why":"Gives the perturbation formula M(λ)=M(0)+λP(0)*(I+λ(-Δ_D-λI)^{-1})P(0), from which the meromorphy of M(λ) and Lemma 7.4 follow.","marker":"[11]"},{"why":"Supplies the theorem on Bari, Riesz and Abel bases of root vectors that yields Theorem 5.7.","marker":"[3]"},{"why":"Provides the analogous application of the basis theorem in dimension one and on metric graphs, supporting the Riesz-basis statement for d=1.","marker":"[37]"},{"why":"Survey of the real-α Robin problem whose open problems (4.11, 4.17, 4.20) define the baseline that the paper answers or partially answers.","marker":"[18]"},{"why":"Supplies Lemma 8.1 on bounded Robin eigenfunctions with bounded H¹ norms converging to Dirichlet eigenfunctions, used in the accumulation-point theorem.","marker":"[19]"},{"why":"Establishes the principal eigenvalue asymptotics -α²+o(α²) on C¹ domains, used as the comparison for the new numerical-range bound on C² and Lipschitz domains.","marker":"[52]"}],"fun_headline_variants":["Complex Robin eigenvalues: either diverge or land on Dirichlet spectrum","Robin Laplacian with complex α: eigenvalue fate decided by duality","Every Robin eigenvalue curve either escapes or hits Dirichlet spectrum","Dichotomy for complex Robin eigenvalues: blow-up or Dirichlet limit","As α→∞, complex Robin eigenvalues either blow up or approach Dirichlet"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole framework rests on the global labelling of eigenvalues into analytic curves λ_k(α) on all of C, which the paper derives from self-adjoint-holomorphic family theory; the stated identity A(α)*=A(α) is false for non-real α, so the correct identity A(α)*=A(ar α) must be used for that derivation to go through.","fun_headline_variants_meta":{"raw":{"variants":["Complex Robin eigenvalues: either diverge or land on Dirichlet spectrum","Robin Laplacian with complex α: eigenvalue fate decided by duality","Every Robin eigenvalue curve either escapes or hits Dirichlet spectrum","Dichotomy for complex Robin eigenvalues: blow-up or Dirichlet limit","As α→∞, complex Robin eigenvalues either blow up or approach Dirichlet"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000363,"raw_usage":{"total_tokens":2018,"prompt_tokens":1069,"completion_tokens":949,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":685,"completion_tokens_details":{"reasoning_tokens":858}},"tokens_in":685,"tokens_out":949,"duration_ms":9478,"temperature":1.0,"reasoning_tokens":858,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:57:39.016510+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a bounded Lipschitz domain, numerically follow a simple Robin eigenvalue curve λ(α) along a sequence α_k→∞ with Re α_k≥0; if a finite accumulation point outside the Dirichlet spectrum of -Δ_D is found, then Theorem 1.5 (and the accumulation-point theorem) would be false, and the Dirichlet-to-Neumann pole argument would have to break down.","supporting_citations":[{"cited_title":"Perturbation theory for linear operators , second ed","cited_arxiv_id":null,"evidence_quote":"Supplies the holomorphic-family formalism and the eigenvalue/eigenprojection continuation theorem on which Theorem 4.1 and the global eigencurve labelling rest."},{"cited_title":"Friedlander’s eigenvalue inequalities and the Dirichlet-to-Neumann semigroup","cited_arxiv_id":null,"evidence_quote":"Establishes the duality between Robin eigenvalues and Dirichlet-to-Neumann eigenvalues on domains, used as Theorem 7.6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the sectoriality and resolvent properties of the Dirichlet-to-Neumann operator on rough domains used in Lemma 7.2."},{"cited_title":"An inverse problem of Calder´ on type with partial data.Comm","cited_arxiv_id":null,"evidence_quote":"Gives the perturbation formula M(λ)=M(0)+λP(0)*(I+λ(-Δ_D-λI)^{-1})P(0), from which the meromorphy of M(λ) and Lemma 7.4 follow."},{"cited_title":"S.On series in root vectors of operators defined by forms with a selfadjoint principal part.Funktsional","cited_arxiv_id":null,"evidence_quote":"Supplies the theorem on Bari, Riesz and Abel bases of root vectors that yields Theorem 5.7."},{"cited_title":"Non-self-adjoint graphs","cited_arxiv_id":null,"evidence_quote":"Provides the analogous application of the basis theorem in dimension one and on metric graphs, supporting the Riesz-basis statement for d=1."},{"cited_title":"The Robin problem","cited_arxiv_id":null,"evidence_quote":"Survey of the real-α Robin problem whose open problems (4.11, 4.17, 4.20) define the baseline that the paper answers or partially answers."},{"cited_title":"On the asymptotics of a Robin eigenvalue problem","cited_arxiv_id":null,"evidence_quote":"Supplies Lemma 8.1 on bounded Robin eigenfunctions with bounded H¹ norms converging to Dirichlet eigenfunctions, used in the accumulation-point theorem."},{"cited_title":"A singularly perturbed linear eigenvalue problem inC1 domains","cited_arxiv_id":null,"evidence_quote":"Establishes the principal eigenvalue asymptotics -α²+o(α²) on C¹ domains, used as the comparison for the new numerical-range bound on C² and Lipschitz domains."}],"review_version":1}