PaperlayerView exact revision on arXiv

Source-linked AI summary

Hybrid Reconstruction of Admissible Spline Spaces from Locally Modified Unclamped Patches for Isogeometric Analysis

C. G. Provatidis, I. Dimitriou

arXiv:2609.35788v1cs.GR

TL;DR

Classical tensor-product NURBS make local modifications non-local, while existing approaches maintain admissibility throughout construction or refinement. This paper decomposes representations into independently modified Active Sections, then reconstructs admissibility algebraically; one-dimensional benchmarks report optimal convergence rates and preserve exact geometry.

  • Problem

    Classical tensor-product NURBS modifications propagate beyond the region of interest, and existing approaches maintain admissibility throughout construction or refinement rather than separating local editing from its later recovery.

    Source: §1 · p. 4

  • Method

    The framework modifies independent Active Sections and reconstructs an admissible basis using linear programming, non-negative least squares, and hierarchical pairwise condensation.

    Source: §1 · p. 5, §7 · p. 53

  • Results

    One-dimensional multipatch benchmarks, including nonlinear diffusion with heterogeneous local polynomial degrees, confirm optimal convergence rates and exact geometry preservation.

    Source: §8 · p. 59

  • Takeaways & Limitations

    The algebraic merging procedure applies to local basis types including Bernstein polynomials and unclamped NURBS, requiring only interface evaluations of basis functions and derivatives.

    Source: §7 · p. 52

    The study focuses on one-dimensional settings; multidimensional interface topologies are deferred, and conditioning at very high degrees or on heavily refined meshes warrants further investigation.

    Source: §7.7 · p. 58

Abstract

from arXiv · show

We propose a decoupled reconstruction framework that temporarily decomposes a NURBS representation into independent local Active Sections, enabling arbitrary local knot insertion, degree elevation, and basis modifications while preserving exact CAD geometry. However, these independent modifications inevitably violate inter-patch continuity, requiring a robust algebraic recovery of global admissibility. We introduce a novel reconstruction methodology that constructs a positive Hybrid Reconstruction Operator: anchor degrees of freedom are identified via QR pivoting, after which a sequence of linear programming problems with distance-based regularization generates strictly non-negative nullspace vectors, while a subsequent non-negative least-squares solve enforces a precise partition of unity. Crucially, to ensure computational efficiency and preserve locality, we implement a hierarchical pairwise condensation strategy that freezes columns already satisfying new interface constraints, confining the optimization exclusively to the active degrees of freedom at each merge step. The resulting hybrid basis is, by construction, strictly non-negative, spans the exact constrained nullspace, preserves partition of unity to machine precision, and reproduces the original geometry exactly. Numerical benchmarks, including a nonlinear diffusion problem on multi-patch curves with heterogeneous local polynomial degrees, confirm optimal convergence rates and demonstrate the framework's ability to seamlessly combine local geometric flexibility with globally consistent approximation spaces. By entirely decoupling local spline editing from the enforcement of continuity, this methodology provides a highly efficient, mathematically principled, and universally applicable tool for adaptive Isogeometric Analysis.

Highlights

The framework decomposes spline representations into local, geometry-preserving sections and reconstructs admissible spaces after local modifications. Its reconstruction uses local interface updates and a minimal knot window.

Highlights

  • The local hybrid B-spline framework enables geometry-preserving decomposition, and null-space reconstruction exactly recovers the original geometry.The reconstruction also supports local refinement while preserving spline properties.Source: Highlights · p. 2
  • Active-frozen coordinate decomposition reduces reconstruction to local interface updates.Source: Highlights · p. 2
  • A minimal local knot vector for each active section contains 2p+2 knots.The minimality of this vector is proved.Source: Highlights · p. 2

1. Introduction

The introduction frames local spline editing as non-local in classical NURBS and asks whether independent edits can precede recovery of global admissibility. The proposed framework separates these stages while preserving exact geometry.

1. Introduction

  • Classical tensor-product NURBS modifications such as knot insertion and degree elevation propagate beyond the target region, limiting localized refinement and editing.Source: §1 · p. 4
  • The paper asks whether independent local spline manipulations can be followed by reconstruction of a globally admissible space.Existing approaches maintain admissibility throughout construction or refinement, coupling local manipulation with global enforcement.Source: §1 · pp. 4–5
  • Active Sections isolate local spline descriptions for independent knot insertion, degree elevation, or basis modification while preserving exact CAD geometry.Continuity constraints are then assembled to construct admissible basis functions with strict non-negativity and partition of unity.Source: §1 · p. 5
  • The framework uses distance-regularized linear programs and hierarchical pairwise condensation, freezing columns that already satisfy new constraints.Optimization is confined to degrees of freedom affected by new constraints at each merge.Source: §1 · p. 5
  • The reconstructed basis is strictly non-negative, forms a partition of unity, spans the exact constrained nullspace, and preserves geometric exactness to machine precision.Source: §1 · p. 5
  • The framework complements existing spline technologies, but this work focuses on one-dimensional settings and defers multidimensional interface-topology challenges.Source: §1 · p. 6

2. Mathematical Background

The background introduces knot vectors, B-spline and NURBS representations, and local support. It also motivates Active Sections as compact local descriptions whose interfaces are coupled through algebraic constraints.

2. Mathematical Background

  • A knot vector Ξ is a non-decreasing sequence of knots, and higher-order B-spline basis functions are obtained recursively through the Cox–de Boor relation.The vector is indexed using polynomial degree p and number of basis functions n.Source: §2 · p. 6
  • B-spline basis functions have local support and non-negativity, with continuity controlled by knot multiplicity.A knot of multiplicity m yields continuity C^(p−m) at that location.Source: §2 · p. 7
  • NURBS extend B-splines with positive weights and represent geometry using control points, preserving exact geometry for many engineering objects.Source: §2 · p. 7
  • Each degree-p basis function spans at most p+1 consecutive knot spans, and p+1 basis functions are nonzero inside each nonzero span.Source: §2 · p. 7
  • An Active Section is a compact description of one nonzero knot span, using a local knot vector that contains the data needed to evaluate its nonzero basis functions.Interactions between neighboring sections are introduced later through algebraic continuity constraints at their common interfaces.Source: §2 · p. 7
  • In IGA, the basis functions used for geometry representation also approximate the field, and admissibility depends on continuity between neighboring regions.Source: §2 · p. 8

3. Related Work

Related spline technologies support local refinement, extraction-based formulations, and multipatch coupling while maintaining admissibility. This work instead separates local manipulation from admissibility recovery through a generic algebraic reconstruction methodology.

3. Related Work

  • Hierarchical, truncated hierarchical, T-spline, and locally refined approaches support local refinement while maintaining admissible spline spaces.Hierarchical methods activate basis functions over selected regions; T-splines and LR splines use local topological or mesh modifications.Source: §3 · p. 8
  • Bézier extraction expresses spline basis functions in terms of Bernstein polynomials, while U-splines support analysis on unstructured spline meshes.These approaches facilitate finite-element infrastructure integration and broaden admissible spaces for geometric design and numerical analysis.Source: §3 · pp. 8–9
  • Mortar, Nitsche-type, weak-coupling, and domain-decomposition methods enforce continuity between independently parameterized patches.Other work constructs smooth spaces over multipatch domains, including geometrically continuous parameterizations and analysis-suitable C1 spaces.Source: §3 · p. 9
  • These approaches maintain an admissible approximation space throughout refinement, enrichment, or patch coupling.Source: §3 · p. 9
  • The proposed framework instead modifies independent Active Sections first and subsequently reconstructs global admissibility using continuity constraints and a positive Hybrid Reconstruction Operator.It complements existing technologies and applies to spline representations whose continuity constraints can be expressed algebraically.Source: §3 · p. 9
  • The contribution is a generic algebraic methodology for reconstructing globally admissible spaces while preserving positivity, partition of unity, and exact geometric representation.Source: §3 · p. 10

4. Proposed Methodology

The method decomposes a NURBS representation into local Active Sections so spline operations can be made locally while preserving exact geometry. Because independent modifications can disrupt global continuity, the framework then turns to reconstructing an admissible approximation space.

4.2. Local Support and Definition of Active Sections

  • An Active Section is the smallest local spline entity determining the basis functions on one knot span, using a local vector of 2p+2 knots.The two central knots define the span; the outer p knots on either side are buffers.Source: §4.2 · p. 12

4.3. Extraction of Local Unclamped NURBS Elements

  • Each nonzero span is extracted as an independent unclamped NURBS element with its own knot vector, weights, and inherited control points.The extraction changes the representation, not the geometric mapping.Source: §4.3 · p. 13

4.4. Geometry Preservation

  • Decomposition is an exact reparameterization of the spline description, so it does not change the curve or introduce geometric approximation.Local knot insertion and degree elevation are subsequently performed on individual local NURBS entries.Source: §4.4 · p. 13, §4.3 · p. 13

4.5.1. Local Knot Insertion

  • Local knot insertion acts on the central block of an Active Section’s knot vector while retaining the outer p knots on each side as buffers.The initial central block contains the original span endpoints; insertion uses the standard NURBS algorithm [12].Source: §4.5.1 · pp. 14–15

4.5.2. Local Degree Modification

  • Direct degree elevation of extracted unclamped elements is not implemented; instead, the parent is elevated globally and re-extracted before selected elements replace local sections.This permits neighboring sections of different degrees to represent the same underlying geometry.Source: §4.5.2 · p. 15

4.6. Consequences of Local Manipulation

  • Local modifications preserve exact geometry but can destroy the original continuity, leaving the modified sections without a globally admissible approximation space for analysis.Continuity must be re-established and a new admissible space constructed.Source: §4.6 · p. 16

4.7. From Geometry to Analysis

  • The framework separates local geometric manipulation from admissible-space construction, motivating algebraic reconstruction when neighboring sections differ substantially.Direct coupling approaches were found restrictive for sufficiently different local knot vectors and polynomial degrees.Source: §4.7 · pp. 16–17

5. LocaltoGlobal Reconstruction

The reconstruction stage restores prescribed continuity after independent local spline modifications by expressing interface conditions as algebraic constraints. It constructs a positive basis for the admissible space and assembles it through successive local merges.

5.1. The Reconstruction Problem

  • Reconstruction preserves local modifications while restoring admissibility by enforcing continuity conditions between neighboring Active Sections.Interface constraints are local: each involves only basis functions participating in the interaction across that interface.Source: §5.1 · p. 17, §5.2 · p. 18

5.3. Pairwise Admissible Coupling

  • The method builds the final space through recursive pairwise admissible couplings rather than one global reconstruction solve.Each reconstructed block is coupled with the next neighboring Active Section.Source: §5.3 · p. 19

5.4. Global Constraint Representation

  • Local interface constraints are assembled into a global operator whose null space defines the admissible space; the operator remains sparse because each constraint affects few variables.Each row represents a continuity constraint and each column a local degree of freedom.Source: §5.4 · pp. 20–21

5.5.4. Properties of the Reconstruction Operator

  • The Hybrid Reconstruction Operator spans the admissible space while satisfying the continuity constraints, remaining non-negative, and recovering partition of unity.The stated operator properties include Thybrid 1 ≈ 1.Source: §5.4 · p. 21, §5.5.4 · p. 23

5.5.1. Anchor Selection

  • Column-pivoted QR identifies dependent and free variables, with the remaining free variables providing the anchors for admissible generators.The first rank(Ctot) pivot columns correspond to dependent variables.Source: §5.5.1 · p. 21

5.5.2. Positive Admissible Generators

  • For each anchor, a linear program finds an admissible non-negative generator, penalizing coefficients farther from the anchor to encourage localization.The constraints impose Ctot tj = 0, fix the anchor coefficient at 1, and bound coefficients between 0 and 1.Source: §5.5.2 · p. 22

5.5.3. Partition of Unity Recovery

  • A non-negative least-squares scaling step recovers partition of unity from generators that already satisfy the continuity constraints.The method applies a second NNLS solve after optional column balancing.Source: §5.5.3 · pp. 22–23

5.6. Efficient Hierarchical Assembly via Active/Frozen Columns

  • Hierarchical assembly freezes columns that already satisfy each new interface and optimizes only active degrees of freedom as patches are merged.The sequential construction yields the global hybrid basis and confines the optimization burden to active degrees of freedom at each step.Source: §5.6 · pp. 23–24, §5.7 · p. 26

6. Validation and Numerical Examples

The examples examine how continuity constraints recover admissible global bases from independent or locally modified elements, including nonnegativity and partition of unity. They cover exact reproduction of the standard Bézier extraction operator, prescribed-C1 patch merging, and beam tests with local degree enrichment.

6. Validation and Numerical Examples

  • Four C0 interface constraints reduce five piecewise-linear elements from 10 local degrees of freedom to six global basis functions.The six resulting global basis functions are illustrated in Figure 2.Source: §6 · pp. 26–28, Figure 2 · p. 29
  • Direct null-space bases may fail to ensure partition of unity and can contain negative values, so physical basis properties must be imposed algebraically.In Example 2, the raw basis had partition of unity but included negative values; the paper notes that this is not guaranteed generally.Source: §6 · pp. 28, 32
  • A column-wise strategy uses support-restricted continuity systems to find minimal-support nonnegative rays, then scales them by QR-based least squares to enforce elementwise partition of unity.It avoids computing the full null-space basis, and its cost scales linearly with the number of basis functions.Source: §6 · pp. 37–38
  • For two quadratic Bézier elements, the construction recovers the four-function standard Bézier extraction basis, nonnegative and with elementwise partition of unity.The resulting basis coincides with the Cox–de Boor B-splines.Source: §6 · pp. 30, 35, 38, Figure 3 · p. 32
  • A nullspace validation example produces a nonnegative third basis column directly, requiring no additional transformation.The computed nullspace output is reported as exactly matching Eq. (79).Source: §6 · p. 39

EXAMPLE-3: A multipatch, multidegree NURBS curve with local hrefinements

  • The 14-function hybrid basis represents a quarter-circle, with the reconstruction described as accurate to machine precision.The example starts from a rational quadratic Bézier patch and projects its 14 control points onto the hybrid space.Source: EXAMPLE-3: A multipatch, multidegree… · pp. 40, 45
  • The hybrid basis remains sparse and local, with 2–5 active functions per element, and has a reported condition number of 1.23 × 10^3.The reported condition number is characterized as moderate and suitable for numerical computations.Source: EXAMPLE-3: A multipatch, multidegree… · pp. 42, 43, 45
  • The 22 locally constructed functions merge into 14 global basis functions that are nonnegative, satisfy partition of unity, and join with C1 continuity at all four interfaces.The eight-function reduction matches the two constraints imposed at each of four interfaces.Source: EXAMPLE-3: A multipatch, multidegree… · pp. 41–42, Figure 5 · p. 43

EXAMPLE-4: A twopatch, multidegree NURBS curve of a straight clamped beam

7. Discussion

The framework separates local spline editing from continuity recovery, retaining substantial local freedom while reconstructing globally admissible spaces. The discussion positions it as complementary to existing spline technologies and identifies adaptive refinement, multidimensional extension, and high-degree stability as open areas.

7. Discussion

  • The reconstruction procedure applies to local bases when interface continuity constraints can be expressed algebraically, and the same merging algorithm handles Bernstein and unclamped NURBS patches.Its inputs are local basis evaluations and derivatives at interfaces.Source: §7 · pp. 52–53

7.1. Adaptive Characteristics

  • Active Sections permit local knot insertion and degree elevation, but the paper does not provide a complete error-indicator- or solution-driven adaptive refinement strategy.Developing such refinement procedures remains future work.Source: §7.1 · p. 54

7.2. Relationship to Hierarchical Methods

  • The framework complements rather than replaces hierarchical methods, which could be used within individual Active Sections before reconstruction.Its unclamped sections preserve native spline structure and use buffer knots to retain reconstruction information.Source: §7.2 · pp. 54–55

7.3. Compatibility with Existing Spline Technologies

  • The framework does not replace existing analysis-suitable multi-patch technologies; it offers a different approach by recovering admissibility after local spline manipulation.Its applicability depends on expressing continuity constraints algebraically.Source: §7.3 · p. 55

7.4. Alternative Reconstruction Strategies

  • Null-space reconstruction is one possible realization, while constraint elimination, penalty, mortar, and weak-coupling strategies are also identified as alternatives.For very large degree-of-freedom counts, constraint elimination or reduced-order modeling may be more efficient.Source: §7.4 · pp. 55–56

7.5. Local Flexibility versus Global Admissibility

  • About 70% of disconnected local degrees of freedom remain available after reconstruction, while conditioning, partition of unity, geometric exactness, and continuity are preserved to machine precision.For the examples considered, the Hybrid Reconstruction Operators remain well conditioned.Source: §7.5 · pp. 56–57
  • Local modifications and admissibility recovery occur at different stages, so reconstruction preserves continuity without eliminating the local approximation freedom used during editing.The reported examples retain about 70% of disconnected local degrees of freedom.Source: §7.5 · pp. 56–57

7.6. Clamped versus Unclamped Reconstruction

  • The paper retains unclamped Active Sections because they preserve native spline structure, allow flexible local edits, and retain buffer-knot information for reconstruction.This choice is more involved to implement than clamped reconstruction.Source: §7.6 · p. 57

8. Conclusions

The conclusions describe a reconstruction framework that decomposes NURBS geometry into locally editable unclamped Active Sections and then recovers global admissibility. One-dimensional benchmarks report optimal convergence rates and preserved geometry, while multidimensional extensions and further stability analysis remain future work.

8. Conclusions

  • The method extracts unclamped Active Sections for independent local edits, then uses interface constraints and a Hybrid Reconstruction Operator to recover a strictly non-negative, partition-of-unity basis spanning the constrained nullspace.The operator uses anchor selection, distance-regularized linear programming, and non-negative least squares.Source: §8 · p. 58
  • One-dimensional multipatch benchmarks, including nonlinear diffusion with heterogeneous local degrees, confirm optimal convergence rates and combine local geometric flexibility with globally consistent approximation spaces.The benchmarks also include geometry preservation, continuity recovery, and local degree variation.Source: §8 · p. 59
  • The framework is presented as a generic algebraic procedure for spline representations with expressible continuity constraints, while multidimensional extensions, fully adaptive refinement, and high-degree stability remain future work.The stated future work also includes heavily refined meshes.Source: §8 · p. 59

APPENDICES

The appendix defines Active Sections as local, unclamped spline entities associated with individual knot spans and distinguishes them from the disconnected space formed by coupling neighboring sections. It establishes the local knot-window size and the dimension of that pairwise space.

A.1. Local Unclamped Decomposition and Pairwise Hybrid Reconstruction

  • An Active Section restricts to one knot span the p + 1 parent basis functions active there, together with their control data and inherited local knot information.It is associated with one element, not the union of elements on either side of an interface.Source: §A.2 · pp. 62–63

This local knot vector contains exactly

A.3. Exact geometry preservation under local unclamped decomposition

The appendix establishes that extracting local unclamped sections from parent basis functions, control data, and inherited knots preserves the NURBS mapping. It also defines the disconnected space used to couple neighboring sections and gives its dimension for equal-degree patches.

A.3. Exact geometry preservation under local unclamped decomposition

  • When an Active Section is built from the parent basis restrictions, corresponding homogeneous control points, and inherited local knots, its mapping equals the parent mapping on its knot span.Because the nonzero spans cover the parametric domain, the piecewise decomposed representation coincides with the parent geometry everywhere.Source: §A.3 · pp. 65–66
  • Exact degree elevation of the parent before extracting an Active Section preserves the geometric mapping.The exact-elevation representation can therefore be used in the local decomposition result.Source: §A.3 · p. 66

A.4. Pairwise interface coupling space

  • Before continuity constraints are imposed, two independent neighboring degree-p Active Sections form a disconnected interface space of dimension 2p + 2.The space contains the complete local approximation spaces of both sections.Source: §A.4 · p. 67

A.5. Null-space criterion for exact coupling

Exact coupling is characterized by a reconstruction space spanning the interface-constraint null space. Under this criterion, the decomposed geometry can be represented by hybrid control points.

A.5. Null-space criterion for exact coupling

  • When range(T) equals ker(C), T spans exactly the globally admissible degrees of freedom defined by the interface constraints.C is the global interface constraint matrix, and T is specified as full-column-rank.Source: §A.5 · p. 68
  • Under the exact-coupling criterion, a hybrid control-point matrix exists that represents the decomposed geometry.The hybrid basis is defined through the reconstruction operator, with Ploc = T Ph.Source: §A.5 · p. 68

A.6. Recursive locality-preserving pairwise coupling

Recursive pairwise coupling preserves exact geometry under stated interface and null-space conditions. It limits each merge to affected coordinates and permits hybrid control-point recovery by pseudoinverse.

A.6. Recursive locality-preserving pairwise coupling

  • For every coupling step, a hybrid control-point matrix preserves exact geometry if the decomposed geometry satisfies each interface condition and no required null-space direction is removed.The theorem applies for k = 2, . . . , m.Source: §A.6 · p. 70

A.7. Active and frozen coordinates

  • At each merge, only coordinates participating in the newly introduced interface constraints need reconstruction; frozen coordinates pass through unchanged.The active–frozen decomposition spans the complete reduced null space.Source: §A.7 · pp. 72–73

A.8. Recovery of hybrid control variables

  • Once the range inclusion is established, the Moore–Penrose pseudoinverse recovers hybrid control variables and the reconstructed geometry is exact.The pseudoinverse alone does not establish exact geometry preservation.Source: §A.8 · pp. 73–74

Appendix B. Proof of the Minimal Local Knot Vector Property

The appendix derives the minimal local knot vector for an Active Section from B-spline local support. Each nonzero knot span depends on p + 1 basis functions, and the required local vector has 2p + 2 entries.

Appendix B. Proof of the Minimal Local Knot Vector Property

  • The minimal local knot-vector result follows from B-spline local support and is independent of the reconstruction procedure.The appendix establishes the mathematical foundation for the local 2p + 2 knot vector associated with each Active Section.Source: Appendix B · p. 74

B.1. Active Basis over a Single Knot Span

  • A nonzero knot span has exactly p + 1 nonzero basis functions, and approximation over that span depends exclusively on them.Source: §B.1 · pp. 74–75

B.2. Local knot vector of an individual basis function

  • Each degree-p B-spline basis function is determined by its local knot vector.Source: §B.2 · p. 75

B.3. Minimal knot vector of a knot span

  • The union of the local knot vectors for the active functions contains 2p + 2 knot entries.For the selected span, the union runs from ξk−p through ξk+p+1.Source: §B.3 · p. 75

B.4. Minimality

  • Removing any knot prevents at least one active basis function from being evaluated correctly, so no shorter vector supports exact local representation.Source: §B.4 · p. 76

B.5. Relation to the Active Section

  • Each Active Section corresponds to a nonzero span and contains its p + 1 active basis functions; its description uses the minimal local knot vector.The 2p + 2 count is for one Active Section, not the larger neighborhood used when adjacent sections are coupled.Source: §B.5 · p. 76
Loading 2609.35788v1…