{"id":"1c13f193-bfa5-4815-902c-b2103df657f9","arxiv_id":"1908.05618","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"T-IFISS is an open-source MATLAB/Octave toolbox that provides self-adaptive finite element algorithms with rigorous error control for deterministic and parametric elliptic PDEs.","lead":"T-IFISS is a new open-source MATLAB toolbox for adaptive finite element computation on triangular meshes, with built-in error estimation, goal-oriented refinement, and stochastic Galerkin methods for uncertain coefficients. It extends the established IFISS package to adaptive and parametric settings, giving researchers and students a ready-made laboratory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Effectivity indices are validated only against approximate reference solutions; this is a real validation gap, but not a demonstrated flaw in the toolbox's central claim.","rationale":"The reader's weakest assumption is the same issue I identify: the effectivity indices are validated against unquantified approximate reference solutions. I agree this is the weakest link in the numerical evidence. Nevertheless, I do not think it is load-bearing enough to change the verdict. The central claim of the paper is that T-IFISS is a working computational laboratory for adaptive FEM with a posteriori error control, and that claim is supported by the software's public availability, the standard nature of the implemented algorithms, and consistency of the case studies with known convergence theory. The effectivity indices are illustrative rather than the sole basis for correctness; the reliability theory is cited from prior publications. The reference solutions used are much more accurate than the working approximations, so the reported indices are likely in the right range. A surrogate-accuracy check would be a worthwhile addition and would remove the gap, but its absence does not undermine the ACCEPT verdict. Hence I recommend UNCHANGED, with the caveat that the authors should quantify reference errors in any revision or provide a manufactured-solution test.","tokens_in":20148,"tokens_out":9775,"duration_ms":98637,"concrete_test":"Recompute the three effectivity-index sequences using one additional level of reference refinement over the surrogates used in the paper: for Example 1, take the P2 tol=2e-5 solution and additionally solve on one uniform refinement of its final mesh; for Example 3, take three uniform refinements of the final GOAFEM mesh instead of two; for Example 4, recompute the reference with one further uniform spatial refinement of the final mesh while keeping the polynomial space P_P union Q_P. If the relative change in any reported effectivity index is less than about 10%, the reference-surrogate concern does not land; if the indices shift by more than that, the numerical validation of 'rigorous error control' would need to be re-presented with quantified surrogate error.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that T-IFISS provides self-adaptive FEM with rigorous a posteriori error control. The in-paper evidence for the effectiveness of the implemented estimators is the effectivity indices in Figures 5(b), 9(b), and 12(b), and all three are computed relative to approximate reference solutions rather than exact errors: Example 1 uses a P2 adaptive solution with tol=2e-5, Example 3 uses two uniform refinements of the final GOAFEM mesh, and Example 4 uses a P2 solution on the final mesh with the enriched polynomial space P_P union Q_P (37,020,322 dof). The manuscript does not quantify the error of these reference surrogates. If any surrogate were inaccurate, the reported effectivity indices could be biased, weakening the numerical demonstration of 'rigorous error control.' This is a genuine validation gap. It is not, however, a demonstrated error in the software: the estimators are built on established theory (e.g., [1,6,44] for EES1-EES3 and [10,11,15] for SGFEM), the code is publicly available, and the reference solutions are substantially higher fidelity than the approximations being tested, so the indices are likely to be trustworthy. The concern is about the strength of the evidence, not the correctness of the algorithms.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces T-IFISS, an open-source MATLAB/Octave toolbox for adaptive finite element computation for deterministic and parametric elliptic PDEs. It describes the modular implementation of the adaptive loop SOLVE–ESTIMATE–MARK–REFINE, three a posteriori error estimation strategies, goal-oriented error estimation with four marking strategies, and stochastic Galerkin FEM with spatial and parametric adaptivity. Four numerical case studies are presented: an anisotropic diffusion problem, a harmonic function in the L-shaped domain, a goal-functional problem on a slit domain, and a parametric diffusion problem with random coefficients. The paper emphasizes that the toolbox is designed both as a research platform for reproducible experiments and as a teaching tool, and it documents the software's directory structure and relationship to the earlier IFISS package.","tokens_in":20361,"tokens_out":5168,"duration_ms":52859,"significance":"The toolbox fills a useful niche by making a range of state-of-the-art adaptive FEM and stochastic Galerkin techniques available in a high-level, readable MATLAB/Octave environment, complementing IFISS and similar packages. The open-source release, modular design, and built-in test problems support reproducibility and classroom use. The numerical case studies provide evidence that the implemented estimators and adaptive strategies achieve optimal or near-optimal convergence rates in the tested examples, and the stochastic Galerkin module demonstrates a genuinely distinctive capability within the IFISS framework. The paper is honest about the distinction between implemented algorithms and their theoretical foundations, which are cited to prior peer-reviewed work.","major_comments":[{"comment":"The effectivity indices are computed against approximate reference solutions rather than exact errors: in Example 1 a P2 adaptive solution with tol=2e-5, in Example 3 a solution on two uniformly refined final meshes, and in Example 4 a P2 solution on the final mesh with the enriched polynomial space PP∪QP (37,020,322 dof). The paper does not quantify the accuracy of these surrogate references. If any of the surrogates is not sufficiently accurate, the reported effectivity indices could be biased, and the empirical support for the phrase 'rigorous error control' would be weakened. Please add a quantitative estimate or bound for the reference error in each case, or explicitly state that the indices are indicative rather than validated surrogates for the exact error.","section":"Sections 2.2, 3.3, 4.5; Figures 5(b), 9(b), 12(b)"},{"comment":"The effectivity indices for the EES3 estimator in Example 1 are consistently below 1 (roughly 0.5-0.7), and several indices in Example 4 also fall below 1. Since the reference solutions are presumably more accurate than the approximations being tested, these values suggest that the estimator is not a guaranteed upper bound in these examples. The text should clarify whether 'rigorous error control' means a provable upper bound up to a generic constant or merely asymptotic reliability of the estimator, and should state explicitly that the case studies demonstrate the latter if that is the intended interpretation.","section":"Figure 5(b), Figure 12(b)"}],"minor_comments":[{"comment":"Footnote 10 states that the error estimation module and the adaptive algorithm are 'only implemented for P1 approximation,' but Section 2.2 clearly describes adaptive P2 computation with EES1 using quartic bubble functions in a deterministic setting. Please clarify that this footnote refers to the stochastic Galerkin module, not to the deterministic part of the toolbox.","section":"Section 4.1, footnote 10"},{"comment":"The multi-indices in Table 3 are displayed as finite tuples without an explicit note that omitted entries are zero. Adding a sentence such as 'for clarity, trailing zero entries are omitted' would make the table easier to read.","section":"Table 3"},{"comment":"The convergence plot in Figure 11 includes a reference line O(N^{-1/3}) for the total degrees of freedom, but the text does not explain the expected rate for the full stochastic Galerkin approximation when both spatial and parametric dimensions contribute to N. A brief comment on how N mixes the two types of degrees of freedom would help interpret the plot.","section":"Figure 11 and Section 4.5"},{"comment":"The paper uses 'effectivity index' for the ratio of the error estimate to an error measured against an approximate reference solution. Since this is not the exact error, consider defining this as a 'surrogate effectivity index' at first use to avoid ambiguity.","section":"General terminology"}],"recommendation":"minor_revision","confidential_remarks":"The validation gap regarding approximate reference solutions is real but local, and it can be addressed by adding error bounds or hedging the wording. The paper's central claims rest primarily on previously published theoretical work, and the numerical experiments are consistent with those foundations. I see no concerns about citation practices or fit with the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nT-IFISS is a genuinely useful software contribution. The paper describes an open-source MATLAB/Octave toolbox that does what it says: adaptive FEM on triangular meshes, goal-oriented adaptivity, and stochastic Galerkin methods for parametric elliptic PDEs, with the code publicly available. It is best read as a software report, not an algorithms paper.\n\nWhat is new is the integration. The individual components—hierarchical estimators, Dörfler marking, NVB refinement, the GOAFEM marking strategies, and the SGFEM technology—all come from prior papers, several by the same authors. But no other package ties them together in this accessible way, particularly for parametric problems. The case studies are a real asset. They demonstrate optimal convergence rates and sensible mesh patterns, including a nice reproduction of the Trefethen–Gopal Laplace challenge to nine digits. The implementation details, from vectorized refinement to the Kronecker-product structure in the stochastic solver, are concrete and credible.\n\nThe soft spots are proportionate. The effectivity indices in Figures 5(b), 9(b), and 12(b) are computed against approximate reference solutions rather than exact errors. For the stochastic example the reference is a 37-million-dof P2 solution on an enriched parametric space; for the goal-oriented example it is two uniform refinements of the final mesh. The paper never quantifies how accurate these surrogates are. That is a real validation gap, but not a demonstrated flaw. The surrogates are substantially higher fidelity than the approximations being tested, and the plotted indices are stable. I would ask the authors to acknowledge the surrogate nature in the captions and, if possible, report the surrogate's own estimated error. That is a minor revision, not a reason to reject.\n\nThe citation pattern is fine. The authors cite their own prior work because they built on it, and the theoretical backing for the estimators is external and published. The paper would be improved by a short statement about what has not been validated, but the central claim—that T-IFISS provides a working environment for self-adaptive FEM with rigorous error control—is well supported by the evidence.\n\nBottom line: the paper deserves a serious referee and, with minor revisions, publication. I would cite it if I worked on adaptive FEM or uncertainty quantification and wanted a readable, reproducible teaching and research tool. Recommend accept with minor revisions.","headline":"A genuinely useful, well-documented adaptive FEM toolbox; algorithms are prior work but the integration is new, and the only real weakness is the unquantified approximate reference solutions in the effectivity plots.","tokens_in":20880,"tokens_out":3134,"would_cite":true,"duration_ms":27595,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["97N80","65N30","65N15","65N50","35R60","65C20","65N22"],"pacs":[],"model":"deepseek-v4-flash","headline":"T-IFISS implements self-adaptive finite element workflows with rigorous a posteriori error control for deterministic and parametric elliptic PDEs.","keywords":["finite elements","adaptive methods","a posteriori error estimation","goal-oriented adaptivity","stochastic Galerkin methods","parametric PDEs","random coefficients","mathematical software"],"falsifier":"Compute the same four case studies with reference solutions obtained from one additional uniform refinement of the final adaptive mesh (or from a P3/high-order enriched space), and check whether the effectivity indices in Figures 5(b), 9(b), and 12(b) change by more than a few percent. If they shift substantially, the estimators are tracking the reference surrogate rather than the true error.","tokens_in":19938,"feed_emoji":"🧮","tokens_out":9085,"duration_ms":84256,"temperature":0.7,"pith_summary":"T-IFISS is an open-source MATLAB/Octave toolbox whose aim is to make adaptive finite element computation a reproducible laboratory exercise: given an elliptic PDE, a user can run through solve, estimate, mark, and refine with a choice of three a posteriori error estimation strategies and see not just a solution but a controlled error estimate. The paper argues that the same loop extends faithfully to parametric elliptic PDEs with random coefficients, where the spatial mesh and the set of polynomial basis functions in the parameter domain are refined together and the linear algebra is handled by Kronecker-structured preconditioned iteration. Four case studies—an anisotropic diffusion problem, a near-singular harmonic point value, a goal functional on a slit domain, and a random-coefficient diffusion problem—show optimal algebraic convergence rates and bounded effectivity indices when the recommended estimator and marking strategies are used. If the claim is right, the toolbox lets researchers and students compare, validate, and extend adaptive finite element methodology without re-implementing the machinery.","feed_headline":"Adaptive finite element computation gets a self-auditing toolbox","feed_subtitle":"T-IFISS brings rigorous error control, goal-oriented adaptivity, and stochastic Galerkin solvers to triangular-grid FEM.","key_machinery":"The machinery that carries the paper is the adaptive loop SOLVE → ESTIMATE → MARK → REFINE, instantiated with hierarchical error estimators (EES1–EES3), edge- or element-based bulk-chasing marking, and newest-vertex bisection with nested finite element spaces. For goal functionals, the dual problem supplies a second Galerkin solution, and the reliable estimate $\\mu\\zeta$ for $|G(u)-G(u_h)|$ is formed from the product of the primal and dual energy-error estimates; four marking strategies (GO-MARK1–4) combine the primal and dual indicators. For parametric problems, the stochastic Galerkin space is the tensor product $X\\otimes P$, the system matrix is a Kronecker sum $A=G_0\\otimes K_0+\\sum_m G_m\\otimes K_m$, and a preconditioned minimum-residual solver works on the component matrices without assembling $A$. The parametric error estimator $\\eta$ splits into spatial and parametric contributions, and the adaptive SGFEM algorithm chooses which component to refine based on the dominant contribution or the larger predicted error reduction.","core_discovery":"The central discovery is that a single modular software architecture can host the entire adaptive finite element workflow—Galerkin solve, local or global hierarchical error estimation, bulk-chasing marking, and newest-vertex-bisection refinement—and extend it to parametric PDEs through stochastic Galerkin discretization. In the deterministic setting, the paper's key identity is the goal-oriented error estimate $|G(u)-G(u_h)|\\le \\|u-u_h\\|\\,\\|z-z_h\\|$, obtained by Galerkin orthogonality and the dual problem, so that the product of two energy-error estimates gives a rigorous and computable bound on a functional error. In the stochastic setting, the error estimator splits into a spatial part $\\|e_X\\|_0$ and a parametric part $\\|e_P\\|_0$, combined as $\\eta=(\\|e_X\\|_0^2+\\|e_P\\|_0^2)^{1/2}$, and the adaptive algorithm chooses between spatial refinement and enrichment of the polynomial index set by comparing predicted error reductions. The four case studies are presented as evidence that these mechanisms deliver optimal convergence rates and stable effectivity indices in practice.","pith_inferences":["The paper leaves implicit that its effectivity validation rests on unquantified reference surrogates; a direct extension is to recompute Figures 5, 9, and 12 with one additional uniform refinement as the reference and require the effectivity indices to remain stable.","The version-2 refinement-selection rule (choose the larger predicted error reduction) could be applied in the goal-oriented stochastic setting, steering the estimator toward a functional by comparing $\\mu\\zeta$ before and after candidate spatial and parametric refinements.","The same Kronecker-preconditioned machinery and hierarchical parametric estimators should carry over to non-affine coefficient representations, giving a path from this toolbox to more general random-field models such as lognormal coefficients.","The observed failure of elementwise EES1 on constant-coefficient problems suggests an automatic estimator-selection heuristic: monitor whether interior residuals are structurally zero and switch to an edge-based estimator mid-loop."],"forward_implications":["For the anisotropic diffusion problem, the edge-based two-level estimator EES3 with bulk-chasing marking gives an optimal $O(N^{-1/2})$ error-decay rate and essentially monotonic reductions, while the elementwise estimator EES1 stalls because interior residuals vanish for constant coefficients.","Goal-oriented adaptivity with the GO-MARK4 combination rule computes the point value $u(0.01,0.01)\\approx 1.02679192311$, accurate to nine digits, and keeps the goal-error effectivity index near 2.8.","In stochastic Galerkin FEM, refining both the spatial mesh and the parametric index set yields faster convergence than refining only one component; the representative run reached a $3\\times 10^{-3}$ tolerance in 19 iterations with 23 polynomials and 6 active parameters.","The matrix-free preconditioned solver keeps iteration counts below 20 regardless of mesh resolution and active parameter count, making stochastic Galerkin systems with millions of unknowns practical on a laptop.","Because all components are modular and exchangeable, the toolbox supports reproducible research: a user can replace an estimator or marking strategy and immediately compare convergence behavior."],"supporting_citations":[{"why":"Supplies the element-residual technique and hierarchical estimator methodology used in EES1.","marker":"[1]"},{"why":"Provides the global hierarchical estimation idea used by EES2 with piecewise linear bubble functions.","marker":"[6]"},{"why":"Defines the bulk-chasing marking criterion used by the MARK module.","marker":"[20]"},{"why":"Supplies the newest-vertex-bisection theory that guarantees nested finite element spaces and contraction in adaptive loops.","marker":"[45]"},{"why":"Proves optimal convergence rates for the goal-oriented marking strategies GO-MARK2–4 used in the toolbox.","marker":"[28]"},{"why":"Derives energy-norm a posteriori error estimators for parametric operator equations, the foundation of the stochastic error estimation.","marker":"[10]"},{"why":"Proves convergence of adaptive stochastic Galerkin FEM, supporting the version-2 adaptive algorithm.","marker":"[11]"},{"why":"Extends goal-oriented error estimation and adaptivity to elliptic PDEs with parametric inputs, underpinning the stochastic goal-oriented module.","marker":"[12]"},{"why":"Analyzes block-diagonal preconditioning for stochastic Galerkin systems, justifying the bounded iteration counts of the solver.","marker":"[47]"}],"fun_headline_variants":["Self-auditing FEM toolbox tames adaptive error control","T-IFISS: rigorous error control for adaptive FEM","Adaptive FEM with built-in error bounds, now in MATLAB","Toolbox that checks its own work: adaptive FEM with rigorous estimates","Error-controlled adaptive FEM for deterministic and parametric PDEs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported effectivity indices—the ratios that validate the error estimators—are computed against approximate reference solutions whose own errors are not quantified, so the demonstration of rigorous error control assumes those reference surrogates are accurate enough.","fun_headline_variants_meta":{"raw":{"variants":["Self-auditing FEM toolbox tames adaptive error control","T-IFISS: rigorous error control for adaptive FEM","Adaptive FEM with built-in error bounds, now in MATLAB","Toolbox that checks its own work: adaptive FEM with rigorous estimates","Error-controlled adaptive FEM for deterministic and parametric PDEs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1287,"prompt_tokens":852,"completion_tokens":435,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":468,"completion_tokens_details":{"reasoning_tokens":352}},"tokens_in":468,"tokens_out":435,"duration_ms":3884,"temperature":1.0,"reasoning_tokens":352,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:08:02.995940+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the same four case studies with reference solutions obtained from one additional uniform refinement of the final adaptive mesh (or from a P3/high-order enriched space), and check whether the effectivity indices in Figures 5(b), 9(b), and 12(b) change by more than a few percent. If they shift substantially, the estimators are tracking the reference surrogate rather than the true error.","supporting_citations":[{"cited_title":"Ainsworth and J","cited_arxiv_id":null,"evidence_quote":"Supplies the element-residual technique and hierarchical estimator methodology used in EES1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the global hierarchical estimation idea used by EES2 with piecewise linear bubble functions."},{"cited_title":"D ¨orfler, A convergent adaptive algorithm for Poisson’s equation , SIAM J","cited_arxiv_id":null,"evidence_quote":"Defines the bulk-chasing marking criterion used by the MARK module."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the newest-vertex-bisection theory that guarantees nested finite element spaces and contraction in adaptive loops."},{"cited_title":"Feischl, D","cited_arxiv_id":null,"evidence_quote":"Proves optimal convergence rates for the goal-oriented marking strategies GO-MARK2–4 used in the toolbox."},{"cited_title":"Bespalov, C","cited_arxiv_id":null,"evidence_quote":"Derives energy-norm a posteriori error estimators for parametric operator equations, the foundation of the stochastic error estimation."},{"cited_title":"Bespalov, D","cited_arxiv_id":null,"evidence_quote":"Proves convergence of adaptive stochastic Galerkin FEM, supporting the version-2 adaptive algorithm."},{"cited_title":"Bespalov, D","cited_arxiv_id":null,"evidence_quote":"Extends goal-oriented error estimation and adaptivity to elliptic PDEs with parametric inputs, underpinning the stochastic goal-oriented module."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Analyzes block-diagonal preconditioning for stochastic Galerkin systems, justifying the bounded iteration counts of the solver."}],"review_version":1}