{"id":"d824a444-6a2b-4b36-a3a5-e5b8a1ca9179","arxiv_id":"1908.01239","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The tangential cone condition is verified for potential, diffusion, and nonlinear source identification in parabolic PDEs, establishing convergence of Landweber-type methods in suitable function spaces.","lead":"This paper proves that four standard inverse problems for parabolic PDEs satisfy the tangential cone condition, a structural property that guarantees convergence of iterative regularization methods like Landweber iteration. The verification covers both the reduced (parameter-to-output) and all-at-once (parameter-and-state) formulations, under full state observations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the full-observation and local-smallness limitations are explicit in the paper and the conditional theorem is supported.","rationale":"The reader correctly identified the full-observation assumption as the most restrictive condition in the paper, and I agree that it is the weakest point of the setup. However, the paper states this assumption explicitly and builds it into Theorem 3.2 and Proposition 4, so it is a scope limitation rather than a flaw in the central argument. The proof of the reduced TCC from the all-at-once condition is direct, and the concrete examples verify the necessary assumptions with the stated choices of function spaces. The only other point worth checking is the smallness of the constant, but Remark 4 already flags it and the examples allow it by taking rho small. Therefore the reader's ACCEPT verdict remains appropriate, and I would not adjust it.","tokens_in":55028,"tokens_out":23889,"duration_ms":239801,"concrete_test":"Independently re-derive the core composition estimate in Proposition 4 starting from the definitions of S, F = C∘S, and the all-at-once residual (31): verify that the source term Phi in (55)-(56) equals the negative of the residual in (R4) and that, using C as the embedding of V into Y, ||F(q)-F(q~)-F'(q)(q-q~)||_Y equals ||S(q)-S(q~)-S'(q)(q-q~)||_Y. Also check one concrete example, e.g., the potential problem with n=3, p=r=2, that the constants in (A.108) and (R3-dual) are finite and that choosing rho small forces c_tcc below 1/(2 C_lin). If either check fails, the theorem needs amendment.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The central claim, Theorem 3.2, is conditional on Assumption 3.1 and on C being the embedding of V into Y. The weakest point is that condition (31) requires full observations in the sense R(C(t)) = Y, stated immediately before (32), and all reduced-setting corollaries inherit this through (R4) and Proposition 4. This is an acknowledged scope restriction, not a hidden assumption: for partial observations such as boundary traces the proof would indeed break down, since the right side of (31) could vanish while the linearization residual remains nonzero. The other possible concern is the smallness of the reduced TCC constant c_red = C_lin c_tcc in (58); Remark 4 explicitly requires c_tcc to be sufficiently small, and for the examples this can be achieved by shrinking rho. I find no internal inconsistency in the derivations: the identity in Proposition 4, the regularity checks (R1)-(R3) in Sections 3.1-3.4, and the all-at-once estimates in the appendix are consistent with the stated conclusions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves the tangential cone condition for a class of time-dependent inverse problems governed by parabolic PDEs, in both the all-at-once and reduced formulations. The main result, Theorem 3.2, shows that under Assumption 3.1 (local Lipschitz continuity of the nonlinearity, well-definedness of the parameter-to-state map, stability of the linearized problem, and the all-at-once tangential cone condition) the reduced forward operator satisfies the tangential cone condition with a uniformly bounded derivative. The authors verify Assumption 3.1 for four benchmark problems: identification of a potential, a diffusion coefficient, a quadratic first-order source, and a cubic zero-order source. The verifications are carried out in the appendix through detailed Hölder, Sobolev embedding, and contraction estimates.","tokens_in":55206,"tokens_out":11865,"duration_ms":116163,"significance":"If correct, the paper provides the first systematic verification of the tangential cone condition for parabolic coefficient identification problems, a central structural assumption for the convergence of Landweber-type iterative regularization methods. The general framework and the explicit index conditions in Corollaries 1-4 are useful for future applications. The proofs are detailed and appear internally consistent; the all-at-once reduction in Proposition 4 is clean, and the example verifications convincingly establish the hypotheses. The main limitation, namely that the observation operator must provide full spatial observations of the state rather than boundary or partial observations, is explicitly stated in the assumptions and does not undermine the conditional claims of the paper.","major_comments":[],"minor_comments":[{"comment":"The sentence before (32) suggests that surjectivity of the observation operator (R(C(t)) = Y) is necessary for the all-at-once condition (31), but the examples verify (31) through the pointwise estimate (39) without relying on this surjectivity; clarifying the precise role of (32) would prevent a misleading reading.","section":"Section 2 (near (32))"},{"comment":"The remark states that c_tcc must be 'sufficiently small' for the reduced constant c_red to satisfy the convergence condition, but it does not state the explicit bound (e.g., c_red < 1/2 for Landweber iteration); adding the exact condition used by the cited convergence results would make the smallness requirement precise.","section":"Remark 4"},{"comment":"The verification of (R2) is partly delegated to [39, Proposition 4.2]; since this is a self-citation and a load-bearing step, restating the precise assumptions from [39] (or including a short proof sketch) would improve the paper's self-containedness.","section":"Section 3.4 (R2)"},{"comment":"The notation ≽ is used in (36) and earlier but defined only in the appendix; defining it at first use (or at the beginning of the notation list in the main text) would help the reader.","section":"Appendix/Notation"},{"comment":"There are a few typographical errors (e.g., 'exsistence' in Remark 6) and the formulas in the appendix contain OCR artifacts in the arXiv version; these should be corrected in the final typeset version.","section":"General presentation"}],"recommendation":"minor_revision","confidential_remarks":"The paper relies on [39] (by co-author Nguyen) for well-definedness and differentiability of the parameter-to-state map; the editor may wish to confirm that this reference is indeed published and that the dependence is not excessive. The paper fits the journal's scope well, and I see no concerns about novelty or citation integrity beyond the normal reliance on previous work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Fast take: this is a solid verification paper, not a flashy one. It supplies the first examples of the tangential cone condition for parabolic coefficient identification problems, in both reduced and all-at-once formulations, with complete proofs. The transfer theorem (Proposition 4) and the four corollaries are the main new content; the examples are potential, diffusion, quadratic source, cubic source. I read the estimates carefully and they hold up. The paper does what it claims: it gives the structural condition needed for Landweber-type convergence for these four parabolic models.\n\nThe all-at-once section is clean and useful because it avoids differentiability of the parameter-to-state map, and Proposition 4 shows how the all-at-once TCC yields the reduced TCC under a bounded-invertibility condition on the linearized state equation. The corollaries give concrete Lebesgue/Sobolev index conditions, including Hilbert-space settings for low dimensions. That is genuinely useful for people who want to apply Landweber iteration in practice.\n\nSoft spots, in proportion. The biggest one is full observations: condition (32) and the choice of C as the embedding V into Y are assumed throughout, so boundary-trace or partial-observation cases are not covered. This is explicit in the paper, not hidden, but it does limit the scope more than the title suggests. Second, the TCC constant in the reduced setting is C_lin times c_tcc, and the smallness needed for convergence proofs is only shown by shrinking the radius rho; there is no quantitative bound for a given ball. That is normal in this literature, but worth noting. Third, some well-posedness and differentiability results for the parameter-to-state map are quoted from a previous paper by one of the authors (Nguyen, Inverse Problems 2019). This is a mild self-citation, but the cited result appears in a peer-reviewed venue and the paper here supplies the reduction argument that makes it applicable. I do not see circularity.\n\nWho this is for: people working on iterative regularization for nonlinear parabolic inverse problems, and anyone who needs to verify TCC-style conditions for their own model. It has no numerics, so readers looking for practical algorithms should also look at the convergence theory literature. As a referee, I would send this out. The proofs are detailed, the claims are internally consistent, and the limitations are stated honestly. I would accept after minor revisions at most.","headline":"Solid, honest verification of the tangential cone condition for four parabolic coefficient identification problems; worth a careful referee, but scope is limited by the full-observation assumption.","tokens_in":55715,"tokens_out":3704,"would_cite":true,"duration_ms":35017,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65J20","65M32","35K55","47J06"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper verifies the tangential cone condition — the inequality that guarantees Landweber convergence — for four coefficient identification problems in parabolic PDEs, in both reduced and all-at-once form.","keywords":["tangential cone condition","Landweber iteration","parameter identification","parabolic PDE","all-at-once formulation","inverse problems","ill-posed problems","nonlinear evolution equations"],"falsifier":"A direct numerical test for the potential problem in one space dimension would settle the claim: fix a smooth $q_0$, draw random perturbations $q,\\tilde q$ inside a ball of radius $\\rho$, solve (4)–(6) for each, and compute the ratio $\\|F(q)-F(\\tilde q)-F'(q)(q-\\tilde q)\\|_Y / \\|F(q)-F(\\tilde q)\\|_Y$. Corollary 1 predicts the ratio stays bounded by a small constant independent of the pair; a single pair violating the inequality, or a ratio that grows with $\\rho$, would falsify it. Repeating the same experiment with boundary-only observations should make the inequality fail, confirming the paper's full-observation requirement.","tokens_in":54832,"feed_emoji":"🔍","tokens_out":15584,"duration_ms":128451,"temperature":0.7,"pith_summary":"The paper proves that the tangential cone condition holds for four benchmark inverse problems in time-dependent parabolic partial differential equations: recovering a potential, a diffusion coefficient, and two nonlinear source terms from observations of the full state. The tangential cone condition is the structural inequality that guarantees the Landweber iteration, a simple iterative method for ill-posed problems, converges and does not get trapped in local minima. The verification is carried out in two formulations: the classical reduced one, where the forward operator maps the parameter to the observations, and the all-at-once one, where parameter and state are unknowns together. Verifying this condition for time-dependent models has been a known bottleneck — earlier examples were static — and the paper supplies a reusable template: check an all-at-once estimate, then let the stability of the linearized state equation lift it to the reduced setting. If the paper is right, these four model problems join the short list of ill-posed equations on which iterative regularization provably works.","feed_headline":"Four parabolic inverse problems pass the Landweber convergence test","feed_subtitle":"The tangential cone condition that guarantees Landweber convergence now holds for four time-dependent benchmarks.","key_machinery":"The tangential cone condition is the central object: an inequality of the form $\\|F(q)-F(\\tilde q)-F'(q)(q-\\tilde q)\\|_Y \\le c_{\\rm tcc}\\|F(q)-F(\\tilde q)\\|_Y$ with a small constant $c_{\\rm tcc}$, the structural hypothesis under which Landweber-type iterations converge and the residual has no spurious local minima. The main technical mechanism is the all-at-once version (31), which bounds the residual of the model equation by the observation difference $\\|C(u-\\tilde u)\\|_Y$; because $C$ is the full embedding $V\\hookrightarrow Y$, this becomes a state-space estimate and bypasses the parameter-to-state map altogether. For the bilinear potential and diffusion coefficient problems the model residual reduces to the product $((B(q-\\tilde q))(t))(u-\\tilde u)$, controlled by product estimates in Sobolev spaces; for the nonlinear source problems it is controlled by the Hölder continuity conditions (45)–(46) on $\\Phi'$ and $\\Psi'$, with arbitrary growth exponents admitted as long as the state space is smooth enough (conditions (A.113)–(A.124)). The transformation $z=e^u$ converts the quadratic-gradient problem into a potential problem, so its verification inherits the potential-problem argument once positivity of $z$ is established.","core_discovery":"The paper's central claim is Theorem 3.2: if the all-at-once tangential cone condition (31) holds, the observation operator $C$ is the continuous embedding $V\\hookrightarrow Y$, and the parameter-to-state map is well defined with a boundedly invertible linearization (Assumption 3.1, items (R1)–(R4)), then on a sufficiently small ball around the initial guess the reduced forward operator has a uniformly bounded derivative and satisfies the reduced tangential cone condition (52) with a small constant. Four corollaries instantiate this: potential identification in the linear heat equation with the diffusive Malthus interpretation, diffusion coefficient identification in the groundwater flow model, a source term with quadratic gradient nonlinearity, and a source term with cubic zero-order nonlinearity covering Ginzburg–Landau, Allen–Cahn, Zel'dovich, Fisher, and Nagumo type equations. The proofs verify (31) by bounding the nonlinear model residual through pointwise-in-time Hölder estimates in which the observation difference $\\|u-\\tilde u\\|_Y$ enters; the all-at-once inequality then implies the reduced one through the stability of the linearized state equation (R3), without needing to differentiate the parameter-to-state map.","pith_inferences":["The Hölder framework of (45)–(46) suggests the same verification should extend to any source nonlinearity with Hölder-continuous derivative, such as general polynomials in $u$ and $\\nabla u$, by choosing the state space smooth enough; the paper itself shows only the quadratic and cubic cases.","The full-observation requirement (32) marks the boundary of the method: boundary-measurement or sparse-sensor versions would need a condition stronger than (31), plausibly an added smoothing estimate linking the model residual to the observed trace.","The positivity argument used to justify $z=e^u$ indicates a general recipe: any inverse problem transformable to a verified one by a state-dependent change of variables inherits the cone condition as long as the transform's positivity and regularity can be maintained.","Since the all-at-once inequality (31) does not involve the parameter-to-state map, it also supports convergence statements for the all-at-once iterative schemes cited in the paper's references, a consequence the paper leaves implicit."],"forward_implications":["Landweber iteration provably converges for the four inverse problems — potential, diffusion coefficient, quadratic-gradient source, and cubic source — whenever the full space-time state is observed.","Because the tangential cone condition enforces local convexity of the residual, the iteration cannot stall in local minima for these problems.","Each example admits a full Hilbert space setting with parameter and data spaces $L^2$ or $H^1$ as appropriate, so adjoint-based implementations are available, as noted in the remarks attached to each corollary.","The all-at-once condition carries over unchanged to wave equations and fractional diffusion by replacing the first time derivative, so those settings inherit the verified cone condition without new estimates.","The quadratic-gradient example is verified directly in its original variables, giving a second route beyond the $z=e^u$ transformation and different admissible function spaces."],"supporting_citations":[{"why":"Introduced the tangential condition as the sufficient condition for convergence of the Landweber iteration that the paper verifies for its examples.","marker":"[17]"},{"why":"Supplies the exact form of the tangential cone condition (26) and the convergence result it feeds into.","marker":"[44]"},{"why":"Provides the all-at-once versus reduced comparison framework and the Gâteaux differentiability result for the parameter-to-state map used in Proposition 5.","marker":"[39]"},{"why":"Supplies the existence, regularity, and embedding theorems used to verify the well-definedness and boundedness conditions (R2) for the parameter-to-state map.","marker":"[43]"},{"why":"Provides the parabolic regularity theorem used in the verification of the stability condition (R3) for the linearized problems.","marker":"[11]"},{"why":"The monograph-level convergence analysis of iterative regularization methods under the tangential cone condition that the verified examples attach to.","marker":"[25]"}],"fun_headline_variants":["Landweber convergence proven for four parabolic inverse problems","Tangential cone condition verified for four time-dependent benchmarks","Four inverse problems satisfy Landweber's key condition","Parabolic coefficient identification: Landweber works","New benchmarks validate Landweber iteration for parabolic PDEs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that observations cover the whole space-time state: the all-at-once inequality (31) is verified only with $C$ as the full embedding $V\\hookrightarrow Y$, so the right-hand side $\\|C(u-\\tilde u)\\|_Y$ is a full-state observation difference; with partial observations such as boundary traces, that right-hand side is too weak to dominate the model residual and the proof does not survive.","fun_headline_variants_meta":{"raw":{"variants":["Landweber convergence proven for four parabolic inverse problems","Tangential cone condition verified for four time-dependent benchmarks","Four inverse problems satisfy Landweber's key condition","Parabolic coefficient identification: Landweber works","New benchmarks validate Landweber iteration for parabolic PDEs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000403,"raw_usage":{"total_tokens":2038,"prompt_tokens":821,"completion_tokens":1217,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":437,"completion_tokens_details":{"reasoning_tokens":1141}},"tokens_in":437,"tokens_out":1217,"duration_ms":9025,"temperature":1.0,"reasoning_tokens":1141,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:19:27.630068+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct numerical test for the potential problem in one space dimension would settle the claim: fix a smooth $q_0$, draw random perturbations $q,\\tilde q$ inside a ball of radius $\\rho$, solve (4)–(6) for each, and compute the ratio $\\|F(q)-F(\\tilde q)-F'(q)(q-\\tilde q)\\|_Y / \\|F(q)-F(\\tilde q)\\|_Y$. Corollary 1 predicts the ratio stays bounded by a small constant independent of the pair; a single pair violating the inequality, or a ratio that grows with $\\rho$, would falsify it. Repeating the same experiment with boundary-only observations should make the inequality fail, confirming the paper's full-observation requirement.","supporting_citations":[{"cited_title":"H/a.pc/n.pc/k.pc/e.pc, A","cited_arxiv_id":null,"evidence_quote":"Introduced the tangential condition as the sufficient condition for convergence of the Landweber iteration that the paper verifies for its examples."},{"cited_title":"S/c.pc/h.pc/e.pc/r.pc/z.pc/e.pc/r.pc, Convergence criteria of iterative methods based on Landweb er iteration for nonlinear problems, J","cited_arxiv_id":null,"evidence_quote":"Supplies the exact form of the tangential cone condition (26) and the convergence result it feeds into."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the all-at-once versus reduced comparison framework and the Gâteaux differentiability result for the parameter-to-state map used in Proposition 5."},{"cited_title":"R/o.pc/u.pc/b.pc/iacute.pc/ccaron.pc/e.pc/k.pc, Nonlinear Partial Differential Equations with Application s, International Series of Numerical Mathematics, Basel","cited_arxiv_id":null,"evidence_quote":"Supplies the existence, regularity, and embedding theorems used to verify the well-definedness and boundedness conditions (R2) for the parameter-to-state map."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the parabolic regularity theorem used in the verification of the stability condition (R3) for the linearized problems."},{"cited_title":"K/a.pc/l.pc/t.pc/e.pc/n.pc/b.pc/a.pc/c.pc/h.pc/e.pc/r.pc, A","cited_arxiv_id":null,"evidence_quote":"The monograph-level convergence analysis of iterative regularization methods under the tangential cone condition that the verified examples attach to."}],"review_version":1}