{"id":"e64c5d08-b1cb-4ae5-b43c-326b3d2c67cb","arxiv_id":"2511.04265","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A space-time adaptive boundary element method with residual error indicators is shown numerically to achieve roughly twice the convergence rate of uniform refinement for 2D wave scattering with singularities.","lead":"A fully space-time adaptive boundary element method for 2D wave scattering is formulated and tested. It reports roughly double the energy-norm convergence rate of uniform meshes on singular problems, with memory savings.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed doubling of convergence rates rests on Richardson-extrapolated benchmark energies from the same uniform Galerkin scheme; if those benchmarks are biased, the improved-rate conclusion is not supported.","rationale":"The paper's contribution is a numerical demonstration, and its most valuable claim is the factor-two rate improvement in the singular examples. That claim is only as strong as the error measurements. In §5, for all non-smooth examples, the 'exact' energy is obtained by the same uniform Galerkin scheme via Richardson extrapolation (E^(i+1)-E^(i) ratios ~3.3372 and ~1.317). The latter ratio is far from 4, so the extrapolation is not in the asymptotic regime; the resulting benchmark is not trustworthy. This is not an internal inconsistency, but it is a correctness risk for the headline rate claims. I do not think it warrants rejection: the adaptive algorithm, the indicator formulas, and the qualitative mesh-refinement patterns are plausible and interesting, and the paper is candid about limitations (shape-regular refinements, no anisotropic theory). The σ=0 issue identified by the reader is real but secondary: theorem 3.1 is quoted for σ>0, and no quantitative dependence is provided; this affects theoretical backing but not the direct numerical comparison if the benchmark issue were resolved. Hence I agree with the CONDITIONAL verdict and would keep it.","tokens_in":14059,"tokens_out":6411,"duration_ms":66161,"concrete_test":"Recompute the energy-error curves for Examples 5.3 and 5.4 using an independent reference energy: e.g., solve the same BIE with a higher-order space-time Galerkin method (piecewise linear in t and x) on a very fine uniform mesh, or with convolution quadrature, and compare against the Richardson benchmarks E≈2.07339e+01 and E≈3.64917. If the inferred benchmark shifts by more than ~10%, recompute the slopes in Figures 17 and 23; if the adaptive/uniform rate ratio drops from ≈2 toward ≈1, the improved-rate claim is not established. Optionally, repeat the smooth Example 5.1 with σ=0 vs. small σ>0 to test whether the constant in (13) degrades as σ→0.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim — approximately doubled energy-norm convergence rates on adaptively refined meshes (Examples 5.3 and 5.4, Figures 17 and 23) — is measured against benchmark energies E≈2.07339e+01 and E≈3.64917 that are inferred by Richardson extrapolation from the same uniform Galerkin scheme, not from known exact solutions. For the singular problems, the uniform solutions are not in the asymptotic regime on the finest meshes used (Δx=1/320 and Δx=1/160, respectively), so the inferred benchmarks can be systematically biased. Since every plotted 'error' is a difference with respect to this benchmark, a biased benchmark can steepen the adaptive error curve and flatten the uniform curve, producing the reported rate ratio ≈2. The paper provides no independent reference (manufactured solution, higher-order computation, or convolution quadrature) to validate the benchmarks. A secondary issue: Theorem 3.1 (Eq. 13) is cited from [23] for σ>0 with constant ≲_σ, while all experiments set σ=0 (Remark, Section 2); no argument is given that the constant stays bounded as σ→0, so the 'theoretical' indicators are not guaranteed to bound the actual error in the computed setting.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces a fully space-time adaptive boundary element method for the two-dimensional time-domain wave equation with Dirichlet boundary conditions. Starting from the energetic Galerkin formulation, the authors use residual-type local indicators motivated by the a posteriori bound in Theorem 3.1, perform Dörfler marking, and refine marked space-time elements by halving in both time and space. The main contribution is algorithmic: implementation of local tensor-product refinement, assembly and update of the non-Toeplitz Galerkin matrix, and numerical evaluation of the residual indicators. Four numerical experiments (smooth solution, localized peak, endpoint singularity, endpoint plus temporal singularity) are presented. For the singular examples, the measured energy-norm convergence rates on adaptively refined meshes are reported to be about twice those on uniform meshes, and memory savings are documented.","tokens_in":14374,"tokens_out":6335,"duration_ms":65555,"significance":"If the observed rate doubling is robust, this is the first systematic demonstration of fully space-time adaptive BEM for the 2D wave equation and a useful step toward 3D and nonlinear problems. The paper's strengths include the detailed algorithmic description, the honest separation between the rigorous a posteriori estimate and the heuristic indicator, and the breadth of numerical experiments. However, the central convergence-rate claims rest on Richardson-extrapolated benchmark energies computed from the same uniform Galerkin scheme, with no independent validation; in the singular cases the extrapolation is not in the asymptotic regime. The mismatch between the sigma>0 theory and the sigma=0 computations is also not addressed. With independent benchmarks and a sensitivity analysis, the conclusions could become convincing.","major_comments":[{"comment":"All energy errors in the singular examples are measured against benchmark values E≈2.07339e+01 (Example 5.3) and E≈3.64917 (Example 5.4), inferred by Richardson extrapolation from the same uniform Galerkin energies E^{(i)}_{Δt,Δx}. The quoted quotients are 3.3372 and 1.317, far from the value ≈4 observed in the smooth examples; this indicates that the sequence is not in the asymptotic regime, so the extrapolated benchmark can be systematically biased. Since the plotted quantity is the difference between E and the discrete energy, any bias changes the slopes of both the uniform and the adaptive curves and can produce a spurious factor of two in the rate ratio. The manuscript provides no independent reference (manufactured solution, higher-order discretization, or convolution quadrature) to validate these benchmarks. This is load-bearing for the central claim of approximately doubled conve","section":"§5.3, §5.4; Figs. 17 and 23"},{"comment":"The residual bound (13) is stated with a constant depending on σ, and the cited theory in [23] is for fixed σ>0. All experiments set σ=0, justified only by the remark 'as usual'. No argument is supplied that the constant in (13) remains finite as σ→0, nor that the norm used in the plots is the correct limit. Consequently, the 'theoretical' indicators used for marking are not rigorously guaranteed to bound the error in the computed setting. Since the adaptive algorithm is advertised as based on a posteriori error estimates, this gap should be addressed at least by a limiting argument or by explicitly classifying the σ=0 indicator as heuristic.","section":"Remark, §2; Theorem 3.1, Eq. (13)"},{"comment":"In Example 5.1 the exact solution ψ(t,x)=xt is known, and the exact energy could be computed directly, yet the benchmark is still obtained by Richardson extrapolation. This is a missed opportunity to validate the benchmark procedure against a ground truth. Conversely, in Example 5.4 the datum f(t,x)=H(t)t^{2/3} is not in H^2_σ(R+, H^{1/2}(Γ)) as required by Corollary 2.3, because its second time derivative is not square-integrable near t=0. The paper does not discuss whether the weak formulation and the a posteriori bound remain valid for such data. These two issues together weaken the evidence for the claimed rates in the singular cases.","section":"§5.1 and §5.4"}],"minor_comments":[{"comment":"The formula for β_i contains the undefined exponent 'γ+1'; presumably a typo.","section":"Eq. (17)"},{"comment":"The function F is defined with a Heaviside factor H(t−τ) but the subsequent formula for ∂_tR would be clearer if the intended domain of the Heaviside factors were specified explicitly.","section":"Section 4.3"},{"comment":"Section 6 attributes the residual estimate to [19], while Theorem 3.1 cites [23]. Please unify the references.","section":"Section 6 vs Theorem 3.1"},{"comment":"The formula for f in the second branch appears to have a missing parenthesis: 'sin^4(4π(−t−2/8)' should likely read sin^4(4π(−t−2/8)) or similar.","section":"Example 5.2"},{"comment":"The norm ∥·∥_{0,0,S_j} is used without definition; please define it as the L^2(S_j) norm.","section":"Notation, Section 3"},{"comment":"The slopes of the adaptive and uniform error curves would be easier to assess if each figure included a table of the fitted convergence rates and the range of DoFs used for the fit.","section":"Figures 17 and 23"}],"recommendation":"major_revision","confidential_remarks":"The Richardson-extrapolated benchmarks are the main risk to the paper's headline claim. If the authors can supply an independent validation (for example, a manufactured exact energy for Example 5.1, a finer uniform reference, or a CQ-based reference for the singular cases) and a sensitivity analysis of the rates with respect to the benchmark value, the paper would be considerably stronger. The σ=0 gap in the theoretical indicator should also be addressed, at least as an explicit limitation. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look. The authors put together the first systematic fully space-time adaptive BEM loop for 2D acoustic scattering: local tensor-product refinements, matrix update strategy, explicit analytic formulas for the residual derivatives, and an honest separation between rigorous and heuristic indicators. The four test cases cover smooth data, a localized peak, endpoint singularities, and a combined space-time singularity, and the adaptive meshes look plausible.\n\nThe main claim — roughly doubled energy-norm convergence rate on adaptive meshes in Examples 5.3 and 5.4 — is real but not yet established. The benchmark energies are inferred by Richardson extrapolation from the same uniform Galerkin scheme. For singular solutions, the uniform solutions are likely not in the asymptotic regime on the finest meshes used (Δx=1/320 and 1/160), so the benchmark can be biased. Since every plotted error is measured against that benchmark, a biased reference can steepen the adaptive error curve and flatten the uniform one, producing the reported factor of two. There is no independent reference (manufactured solution, higher-order scheme, convolution quadrature). The authors should either provide one or soften the rate claims.\n\nSecond, Theorem 3.1 is quoted for σ>0 with constant depending on σ, but all experiments use σ=0. For a finite time interval the exponential weight is harmless, so this is probably a one-paragraph fix, but the paper doesn't make the argument. As written, the \"theoretical\" indicator operates outside the stated theorem.\n\nMinor points: the conclusion attributes the residual bound to [19] while the theorem comes from [23]; no code or data are provided, which makes the benchmark issue harder to check.\n\nOverall it's a real contribution with a useful algorithmic template. The method deserves a serious referee; the referee should ask for a benchmark robustness check (e.g., comparing with a higher-order reference or a known manufactured solution) and a sigma=0 statement.","headline":"First fully space-time adaptive BEM loop for the wave equation – worth a serious look, but the headline rate gains rest on self-extrapolated benchmarks.","tokens_in":14794,"tokens_out":2498,"would_cite":true,"duration_ms":25888,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M38","65M50","35L05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A fully space-time adaptive boundary element method for the 2D wave equation, driven by a residual error indicator, achieves roughly twice the energy-norm convergence rate of uniform meshes on solutions with spatial, temporal, or traveling","keywords":["space-time adaptive method","boundary element method","wave equation","residual error indicator","a posteriori error estimate","acoustic soft scattering","singular solutions","energy norm"],"falsifier":"On a sequence of adaptively refined meshes for a problem with a known exact solution (e.g., the endpoint-singular datum of Example 5.3), compute the ratio of the true squared energy error to the sum of theoretical indicators Σ_j η_j^2; if this ratio grows without bound as the number of degrees of freedom increases, the σ = 0 indicator is not reliable and the reported rates are not certified. Alternatively, if a smooth problem shows a super-linear adaptive rate contradicting the paper's Example 5.1, that would signal an artifact.","tokens_in":13949,"feed_emoji":"🌊","tokens_out":7749,"duration_ms":64660,"temperature":0.7,"pith_summary":"The paper claims that a boundary element method for the two-dimensional acoustic wave equation can be made fully adaptive in both space and time, and that when the solution carries singularities—at endpoints, in time, or along traveling wave fronts—the adaptive refinements recover roughly twice the energy-norm convergence rate of uniform meshes. The mechanism is a cellwise residual error indicator built from the local mesh size and the spatial and temporal derivatives of the residual of the boundary integral equation, combined with a marking-and-refinement loop that halves marked space-time cells in both variables. The claim is supported by numerical experiments on a flat scatterer using piecewise-constant local tensor-product discretizations, comparing a rigorous indicator and a cheaper heuristic variant. If correct, this makes space-time adaptive BEM a practical tool for wave-scattering problems with singular solutions, using fewer degrees of freedom and less memory.","feed_headline":"Space-time adaptive BEM doubles wave-scattering convergence","feed_subtitle":"Residual-guided refinement beats uniform meshes for singular acoustic waves","key_machinery":"The central object is the residual-based error indicator η_j^2 = max{Δt_j, Δx_j}(||∇R||^2 + ||∂_t R||^2), computed cellwise from the residual R = f − Vψ_h of the weakly singular integral operator V for the wave equation. This indicator directly steers the adaptive loop: all cells with η_j^2 > Θ η_max^2 are marked and refined by halving their time and space extents, producing four children per marked cell. Its computation relies on analytic formulas for the time and spatial derivatives of the single-layer potential on a flat screen Γ ⊂ R^2, evaluated by Gauss quadrature. The heuristic indicator η̃_j^2 = Δx_j ||∇R||^2 + ||∂_t R||^2 is introduced as a cheaper alternative.","core_discovery":"The paper introduces the first systematic fully space-time adaptive boundary element procedure for the wave equation in 2D for the acoustic soft-scattering problem. The adaptive loop, SOLVE–ESTIMATE–MARK–REFINE, is steered by the residual indicator η_j^2 = max{Δt_j, Δx_j}(||∇R||^2 + ||∂_t R||^2), where R is the residual of the weakly singular integral equation; marked cells are refined by halving their time and space extents, creating local tensor-product refinements. In numerical experiments on a screen, solutions with an endpoint singularity (Example 5.3) and with combined spatial and temporal singularities (Example 5.4) show energy-norm errors decaying roughly as O(DoF^{-1}) and O(DoF^{-1","pith_inferences":["Because the rigorous bound behind the indicator is stated for σ>0 but all experiments use σ=0, the certified reliability of the indicator in the computed regime remains an open point; a test of the ratio of true error to indicated error on refined meshes would settle whether the observed rates are also certified.","The roughly doubled rates on singular problems echo the classical benefit of adaptive mesh refinement for elliptic problems; the paper's examples suggest the same benefit carries to hyperbolic space-time, but the theory of optimality for non-shape-regular or anisotropic space-time meshes is left open (the paper itself flags this).","The success of the cheaper heuristic indicator hints that the full max-weight combination of gradients may be unnecessary in practice; a traveling-wave-front example in 3D could probe the limits of this simplification."],"forward_implications":["For singular solutions, adaptive space-time BEM attains approximately twice the energy-norm convergence rate of uniform refinements: O(DoF^{-1}) versus O(DoF^{-1/2}) in the endpoint-singular example, and O(DoF^{-1/2}) versus roughly O(DoF^{-1/4}) for combined spatial/temporal singularities.","At fixed accuracy, the adaptive method uses fewer degrees of freedom and, despite losing the block-Toeplitz structure of uniform time-stepping meshes, can consume less memory.","The heuristic indicator, which weights only the spatial gradient by Δx_j, converges at the same rate as the energy error, while the rigorous indicator estimates a weaker norm.","The approach is ready for extension to 3D scattering, where larger savings in DoFs and memory are expected, and to higher-order hp discretizations and nonlinear contact problems."],"fun_headline_variants":["Adaptive space-time BEM outperforms uniform meshes in wave scattering","Space-time adaptive BEM: residual indicators sharpen convergence","Wave scattering: adaptive BEM shows faster error decay than uniform","Residual-guided adaptive BEM doubles wave-solve convergence"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The residual error bound that justifies the indicator is proved for a positive weighting parameter σ, yet every numerical experiment sets σ = 0, and no argument is given that the bound's constant remains uniform as σ → 0.","fun_headline_variants_meta":{"raw":{"variants":["Adaptive space-time BEM outperforms uniform meshes in wave scattering","Space-time adaptive BEM: residual indicators sharpen convergence","Wave scattering: adaptive BEM shows faster error decay than uniform","Residual-guided adaptive BEM doubles wave-solve convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000764,"raw_usage":{"total_tokens":3171,"prompt_tokens":633,"completion_tokens":2538,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":377,"completion_tokens_details":{"reasoning_tokens":2468}},"tokens_in":377,"tokens_out":2538,"duration_ms":18596,"temperature":1.0,"reasoning_tokens":2468,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T23:42:24.936771+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a sequence of adaptively refined meshes for a problem with a known exact solution (e.g., the endpoint-singular datum of Example 5.3), compute the ratio of the true squared energy error to the sum of theoretical indicators Σ_j η_j^2; if this ratio grows without bound as the number of degrees of freedom increases, the σ = 0 indicator is not reliable and the reported rates are not certified. Alternatively, if a smooth problem shows a super-linear adaptive rate contradicting the paper's Example 5.1, that would signal an artifact.","supporting_citations":[],"review_version":1}