{"id":"f0d186c4-97e5-4d20-8183-10621ec2b8bd","arxiv_id":"1908.00813","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A C2-smooth isogeometric collocation framework for Poisson's equation on bilinear multi-patch planar domains is constructed and numerically shown to converge at rates comparable to one-patch collocation.","lead":"This paper presents an isogeometric collocation method for Poisson's equation on planar multi-patch domains, using a globally C2-smooth spline space with a basis that is independent of the geometry. Numerical tests show convergence rates similar to one-patch collocation, which is useful for engineering simulations on composite CAD geometries.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The basis/dimension claim (Eq. 19) rests on an unproved direct-sum/linear-independence assertion in §2.4, which the new C4 vertex construction is not covered by the cited [27, Thm. 1]; if false, the collocation system is ill-posed.","rationale":"The paper presents a plausible and well-tested numerical method, and I do not see a demonstrated fatal flaw. However, the foundational basis/dimension claim is not fully established. Eq. (19) is used in Example 1 to argue that the number of collocation points and the number of degrees of freedom differ only by lower-order terms, and the full collection of functions is used to assemble the collocation system. The asserted linear independence is plausible: the vertex-function index ranges on each edge (j2 ≤ 4−j1) do not overlap with the edge-space index ranges (j2 ≥ 5−j1), and the Hermite-type systems (14) appear block triangular. But the paper provides no proof of these structural facts, and the citation to [27, Theorem 1] is not decisive because the vertex construction here is new. The numerical experiments in Section 4 are consistent with the claims, and the superconvergence results are explicitly presented as numerical observations, so the main unresolved risk is the basis/dimension statement rather than the convergence rates. This matches the reader's weakest-assumption identification, and supports keeping the CONDITIONAL verdict: the central claim is credible but should be verified by a rank check before acceptance.","tokens_in":27372,"tokens_out":19463,"duration_ms":204462,"concrete_test":"Implement the construction of §2.3–§2.4 for several random bilinear multi-patch configurations, including an inner vertex of valency 4, 5, and 6, and assemble the candidate basis functions from (5)–(18) for (p,r) = (5,2), (6,2), and (6,3) over several values of k. Compute the numerical rank of the global Gram matrix (or of the collocation matrix (21) with Greville points) and compare it with dim W from Eq. (19); additionally, solve the vertex interpolation system (14) and verify that the derivatives in (13) are reproduced. If the rank matches dim W, the direct-sum/independence claim is supported; if it is smaller, the basis and dimension claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the collocation system in §3.1 is a well-posed discretization built on the C2-smooth space W depends on the basis statement in §2.4. There the paper asserts that the functions from (5)–(18) are linearly independent, that (4) is a direct sum, and that this yields the dimension formula (19), citing [27, Theorem 1]. But the vertex subspaces constructed in §2.3.2 are new: they add C4 conditions at vertices and define the functions via the interpolation systems (13)–(14). The invertibility of these systems, the linear independence of the 15 vertex functions, and the absence of nontrivial intersections with the edge subspaces are asserted without a proof. If a vertex system (14) were singular for some valency or bilinear configuration, or if a vertex function were a nontrivial linear combination of edge functions with overlapping support near vertices, then dim W would be smaller than (19), the overdetermination-ratio argument in Example 1 would break, and the least-squares collocation matrix (21) would have the wrong rank. This is the weakest link in the central construction, and it is precisely the point on which the claim of a geometry-independent basis rests.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an isogeometric collocation framework for the Poisson equation on planar bilinearly parameterized multi-patch domains. The discretization space is a globally C2-smooth spline space W defined as the direct sum of patch, edge, and vertex subspaces; the vertex construction is modified by enforcing C4-smoothness at vertices so that the space dimension is independent of the initial geometry (Eq. 19). Two choices of collocation points are investigated: tensor-product Greville abscissae and clustered superconvergent points. Numerical experiments on several multi-patch domains (Section 4) report convergence rates of O(h^p), O(h^p), and O(h^{p-1}) in the L2, H1, and H2 norms for the superconvergent points, with one observed reduction to O(h^5) for p=5 in the L2 norm.","tokens_in":27644,"tokens_out":2727,"duration_ms":26554,"significance":"If the construction is sound, the paper makes a useful contribution to isogeometric collocation for multi-patch domains: it provides a uniform, locally supported basis for a C2-smooth space whose dimension is independent of the geometry, and it extends superconvergent-point collocation from one-patch to multi-patch settings. The numerical results support the claimed convergence rates for the tested configurations, and the authors are explicit about the one suboptimal case. The main weakness is that the theoretical foundation of the space construction—the linear independence and direct-sum decomposition in Section 2.4—is asserted rather than proved, and this is exactly the part that is new relative to the authors' prior work [27].","major_comments":[{"comment":"The claim that the collection of functions (5)–(18) forms a basis of W, and the resulting dimension formula (19), rests on the assertion that the decomposition (4) is a direct sum and that the individual functions are linearly independent. This is not proved for the new vertex subspaces constructed in §2.3.2. In particular, the cited [27, Theorem 1] does not cover the newly added C4 vertex conditions or the interpolation systems (13)–(14). The authors need to show that the systems (14) are nonsingular for all admissible valencies and bilinear parameterizations, that the 15 (or 18) vertex functions are linearly independent, and that no vertex function is a nontrivial linear combination of edge functions with overlapping support near the vertex. Without this, the dimension formula (19) and the well-posedness of the collocation system (21) are not established.","section":"§2.4"},{"comment":"The sentence after Eq. (14) states that the first set of interpolation conditions 'uniquely determine all coefficients a^{Γ(i_l)}_{j1,j2}' and that the second set determines all a^{(i_l)}_{j1,j2}. No argument is given for the count of independent conditions versus unknowns, nor for the invertibility of the resulting linear systems for the geometry-dependent coefficients α and β. Since this is the mechanism that defines the vertex basis functions, a proof or a reference to a published proof is needed; the assertion that the construction 'works' is precisely the point that the stress-test identifies as load-bearing.","section":"§2.3.2"},{"comment":"The superconvergent point polynomials (15x^4−12x^2+1 and 99x^5−130x^3+31x) are stated without derivation or reference for the regularity cases r=2 and r=3. These points are used for all multi-patch numerical experiments, so their derivation from the 1D model problem (23) should be documented, or a reference provided. This is not a flaw in the method itself, but it makes the convergence claims in Section 4 harder to verify independently.","section":"§3.2.3, Table 1"}],"minor_comments":[{"comment":"In the definition of the interface edge subspace, the index range 'j2 = 5−j1,...,n_{j1}+j1−6' appears to contain a typo; the subscript on n is unexplained and the same expression is garbled in §2.4 as 'n j1 +j1−6'.","section":"§2.3.1"},{"comment":"For boundary vertices of valency ν_i ≥ 3, the text says that g^{(i0)}_{Γ(i0);j1,j2}(ξ2,ξ1) is 'just defined as standard B-splines' N^{p,r}_{j1,j2}(ξ2,ξ1); this notation is confusing because the same symbol is used for interface edge functions in Eq. (8), and the argument ordering should be stated more carefully.","section":"§2.3.2"},{"comment":"The count of Greville collocation points, |J| = ν n^2 − ν n + 1, is stated without derivation; it would be helpful to explain how repetitions along interfaces and at the central vertex are counted.","section":"Example 1"},{"comment":"For (p,r)=(6,2), the text first says 'we have even to add two points' and then states that 'the set of all superconvergent points coincides with the set of clustered superconvergent points'; this phrasing is confusing and should be clarified.","section":"§3.2.3"}],"recommendation":"major_revision","confidential_remarks":"The paper's main novelty over [27] is the new vertex construction and the geometry-independent dimension claim, and the missing linear-independence/direct-sum proof is exactly in that new part. The numerical experiments are consistent with the claims and the method appears plausible, so the issue is fixable by adding the missing proof or by restricting the claims accordingly. I therefore recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, honest engineering paper. The genuinely new piece is the vertex construction that enforces C^4 smoothness at vertices to make the space dimension geometry-independent, and the paper shows how to use that space for multi-patch collocation without special interface handling. The numerics are extensive and support the claimed convergence rates. I did not find a fatal flaw.\n\nThe main soft spot is exactly what the stress-test note flags. Section 2.4 asserts linear independence and the direct sum decomposition for the new vertex functions without proof. That claim is load-bearing for the dimension formula (19) and for the nonsingularity/rank of the least-squares collocation system. The cited [27, Theorem 1] covered the old vertex space, not the new C^4-modified one. The invertibility of systems (14), the linear independence of the 15 vertex functions, and the absence of nontrivial edge-vertex intersections are all asserted rather than shown. I think the construction is probably correct—the numerical results would not be that clean if the dimension collapsed—but this should be proved or at least given a careful rank argument. It is a fixable rigor gap, not a reason to reject.\n\nOther soft spots are minor. The superconvergent point polynomials in Table 1 appear without derivation; a one-line note on how they follow from the 1D model problem would help. The convergence rates are read off log-log plots; tables of errors and measured rates would be better. There are a few typos in the index ranges in Sections 2.3.1 and 2.4. The geometry setting is bilinear, or bilinear near interfaces, and the paper mostly stays in that setting. The self-citations to [26, 27] are heavy but appropriate: those are published results that the construction directly builds on, and that is not a flaw.\n\nThe paper deserves a serious referee. I would send it to review with a request to prove or rigorously justify the basis/dimension claim, derive the superconvergent points, and report numerical convergence tables. A caveated accept or major-revision verdict seems right.","headline":"A useful, honest C^2 multi-patch collocation paper whose main construction is plausible and well tested, but whose dimension/basis claim needs proof before I'd fully trust it.","tokens_in":28166,"tokens_out":2235,"would_cite":true,"duration_ms":25183,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N35","65D17","68U07"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs a globally C2-smooth isogeometric spline space whose dimension is independent of the geometry, and uses collocation on it to solve the Poisson equation over planar bilinear multi-patch domains.","keywords":["isogeometric analysis","collocation","C2-smooth multi-patch spline space","superconvergent points","Poisson's equation","Greville abscissae","multi-patch domain"],"falsifier":"Compute, for a small concrete configuration (for example the three-patch domain of Fig. 6(a) with p=5, r=2, k=2), the actual rank of the matrix whose columns are the evaluations of the candidate basis functions at a sufficiently large set of sample points; if the rank is strictly less than the value from (19), the basis and direct-sum claim fails. Alternatively, rerun Example 3 with a manufactured solution and check whether the L2 error for p=5 with clustered superconvergent points decreases as O($h^{5}$); an O($h^{4}$) rate would indicate that the claimed convergence behavior is not reproduced.","tokens_in":27115,"feed_emoji":"🧮","tokens_out":5111,"duration_ms":46547,"temperature":0.7,"pith_summary":"The paper aims to solve the Poisson equation on planar domains split into several bilinearly parameterized patches by collocation, using a globally C2-smooth spline space as the discretization space. The construction splits the space into patch, edge, and vertex parts and enforces extra smoothness at vertices so that the space dimension is independent of the initial multi-patch geometry. Two collocation point strategies are compared: tensor-product Greville abscissae and clustered superconvergent points. The numerical experiments indicate that the method reproduces one-patch convergence behavior, with superconvergent points improving the odd-degree case.","feed_headline":"C2-smooth splines solve Poisson on multi-patch domains","feed_subtitle":"Space dimension is independent of the geometry; superconvergent points recover one-patch convergence rates.","key_machinery":"The load-bearing object is the direct-sum decomposition W = (sum of patch subspaces) ⊕ (sum of edge subspaces) ⊕ (sum of vertex subspaces) of the C2-smooth space, with each subspace spanned by locally supported functions. C2-smoothness across interfaces is enforced through the geometric-continuity conditions (1)-(3), expressed via the functions α and β that describe how adjacent bilinear parameterizations meet. The vertex subspaces are modified by requiring C4-smoothness at the vertex, which fixes their dimension to 15 (or 18 for valency-2 boundary vertices) and makes the total dimension (19) geometry-independent. The collocation system is assembled by mapping either Greville abscissae or clustered superconvergent points through each patch geometry, leading to an overdetermined linear system solved by least squares.","core_discovery":"The paper's central claim is that for any planar bilinearly parameterized multi-patch domain one can construct, by explicit local formulas or small linear systems, a basis of locally supported C2-smooth isogeometric functions whose dimension is given by formula (19) and is independent of the geometry. The basis comprises functions associated with patches, edges, and vertices; vertex functions are forced to be C4 at the corresponding vertex, which removes the geometry-dependent degrees of freedom present in earlier constructions. With this space as ansatz, collocation for the Poisson equation with Dirichlet boundary conditions yields the reported convergence rates under h-refinement, and the same rates as the one-patch case are observed for the tested degrees p=5 and p=6 and regularities r=2,3.","pith_inferences":["The C4-at-vertices trick is a general dimension-stabilization mechanism and could likely be reused to make C1 or C3 multi-patch spline spaces geometry-independent in the same fashion.","The polynomial roots for superconvergent points (for example 15x^4 - 12x^2 + 1 for p=5, r=2) may serve as building blocks for other reduced-regularity spline spaces, not only the three cases tested.","Since the collocation system is overdetermined, a natural next step not pursued in the paper is to search for a square subset of points that keeps the convergence rates, which would remove the least-squares solve.","The observed loss of one L2 order for p=5 with superconvergent points on multi-patch domains suggests that the vertex or interface coupling, rather than the point selection, is the limiting factor; testing a single-patch analog with the same reduced regularity would isolate the cause."],"forward_implications":["A C2-smooth collocation solve on multi-patch domains can be assembled without special treatment of interfaces, unlike methods that only couple patches with C0 continuity.","For the tested spaces (p,r) in {(5,2),(6,2),(6,3)}, Greville points give the same suboptimal odd-degree rates as in the one-patch case, while clustered superconvergent points recover O(h^p) in L2 and H1 and O(h^{p-1}) in H2.","The dimension formula (19) lets users predict the number of degrees of freedom before solving, regardless of the patch layout.","The method applies to any planar bilinear multi-patch domain and also to a bicubic three-patch geometry that is bilinear close to the patch interfaces, since the construction only needs bilinear behavior near interfaces."],"supporting_citations":[{"why":"Supplies the prior C2-smooth space construction and the triharmonic Galerkin framework that the new space extends.","marker":"[27]"},{"why":"Establishes the interface conditions (1)-(3) and the trace interpolation properties used to build the edge functions.","marker":"[26]"},{"why":"Describes the dimension and complex structure of the full C2-smooth space, the baseline that motivates taking a simpler subspace.","marker":"[24]"},{"why":"Introduces Greville-abscissae collocation and its one-patch convergence rates, which the multi-patch results are compared against.","marker":"[2]"},{"why":"Supplies superconvergent points and the least-squares approach for the overdetermined collocation system.","marker":"[1]"},{"why":"Provides the variational collocation formulation and an alternating selection of superconvergent points.","marker":"[15]"},{"why":"Gives the clustered selection of Galerkin superconvergent points that the paper generalizes to multi-patch domains.","marker":"[32]"}],"fun_headline_variants":["Uniform C2-smooth collocation for any planar multi-patch","Multi-patch Poisson solved with C2-smooth local basis","Geometry-independent C2 space for multi-patch collocation","Superconvergent collocation points beat Greville on multi-patch","C2 collocation: superconvergent points recover one-patch rates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the collection of locally supported functions built in Section 2.3 is genuinely a basis: linear independence of each family and trivial pairwise subspace intersections are asserted rather than proved, and the same holds for the C4-smoothness of the vertex functions; without this, the dimension count (19) and the nonsingularity of the least-squares system are not guaranteed.","fun_headline_variants_meta":{"raw":{"variants":["Uniform C2-smooth collocation for any planar multi-patch","Multi-patch Poisson solved with C2-smooth local basis","Geometry-independent C2 space for multi-patch collocation","Superconvergent collocation points beat Greville on multi-patch","C2 collocation: superconvergent points recover one-patch rates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000631,"raw_usage":{"total_tokens":2907,"prompt_tokens":929,"completion_tokens":1978,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":1897}},"tokens_in":545,"tokens_out":1978,"duration_ms":14643,"temperature":1.0,"reasoning_tokens":1897,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:32:01.502270+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a small concrete configuration (for example the three-patch domain of Fig. 6(a) with p=5, r=2, k=2), the actual rank of the matrix whose columns are the evaluations of the candidate basis functions at a sufficiently large set of sample points; if the rank is strictly less than the value from (19), the basis and direct-sum claim fails. Alternatively, rerun Example 3 with a manufactured solution and check whether the L2 error for p=5 with clustered superconvergent points decreases as O($h^{5}$); an O($h^{4}$) rate would indicate that the claimed convergence behavior is not reproduced.","supporting_citations":[{"cited_title":"Kapl and V","cited_arxiv_id":null,"evidence_quote":"Supplies the prior C2-smooth space construction and the triharmonic Galerkin framework that the new space extends."},{"cited_title":"Kapl and V","cited_arxiv_id":null,"evidence_quote":"Establishes the interface conditions (1)-(3) and the trace interpolation properties used to build the edge functions."},{"cited_title":"Kapl and V","cited_arxiv_id":null,"evidence_quote":"Describes the dimension and complex structure of the full C2-smooth space, the baseline that motivates taking a simpler subspace."},{"cited_title":"Auricchio, L","cited_arxiv_id":null,"evidence_quote":"Introduces Greville-abscissae collocation and its one-patch convergence rates, which the multi-patch results are compared against."},{"cited_title":"Montardini, G","cited_arxiv_id":null,"evidence_quote":"Gives the clustered selection of Galerkin superconvergent points that the paper generalizes to multi-patch domains."}],"review_version":1}