Source-linked AI summary
A General Superconvergence Result for Cubature on Triangulated Polygonal Domains
Jonah A. Reeger
TL;DR
Cubature rules are widely used to approximate integrals over polygonal domains, but their convergence depends on polynomial exactness and mesh refinement. This paper develops a triangular-subdivision analysis showing that exactness at a fixed even degree yields an extra convergence order, and recommends even degrees when constructing such rules.
Problem
The paper addresses how cubature rules exact for bivariate polynomials through degree m converge when triangular subdomains are refined.
Method
The paper relates errors on a triangle to errors on four congruent subtriangles using transformed nodes, scaled weights, Taylor expansions, and polynomial exactness.
Results
An extra convergence order is achieved over polygonal domains when m is even and the domain uses triangular subdomains.
Takeaways & Limitations
Even m should be a primary consideration when constructing cubature rules over polygonal domains based on triangular subdomains.
Takeaways & Limitations
The result assumes that the cubature rule first subdivides the polygonal domain into triangular subdomains and satisfies the stated local exactness and weight conditions.
Abstract
from arXiv · showhide
Cubature rules, which approximate definite integrals as a linear combination of a set of function values, are ubiquitous and necessary for computational methods in the physical sciences. A superconvergence result for cubature rules on polygonal domains is developed, whereby a rule that is exact for all bivariate polynomials of a fixed even degree realize an extra order of convergence under a decrease in the spacing between nodes.
1. Introduction.
Cubature rules support computational methods that integrate functions over subdivided polygonal domains. The paper studies how exactness for polynomial spaces produces an extra convergence order under refinement, using triangular subdomains.
- Cubature rules approximate definite integrals as linear combinations of function values and support methods such as moments and finite elements.
- Subdivision commonly partitions the integration domain into triangular subdomains whose approximations are summed.
- Exactness for even-degree polynomials can yield an extra convergence order through favorable error cancellations rather than higher polynomial precision.The paper contrasts this two-dimensional mechanism with one-dimensional extra precision from integrating degree m+1 polynomials.
- The superconvergence result assumes that the polygonal domain is first subdivided into triangular subdomains.
2. Problem Statement.
The paper formulates local cubature error on triangular subdomains and ties convergence to polynomial exactness. Rules are constructed so their error vanishes on all bivariate polynomials through degree m.
- Each triangular subdomain is represented by three vertices and a characteristic spacing parameter tied to its area.
- The local integral is approximated by weighted function values at cubature nodes plus an error term.
- The target local error decays as O(h_k^ρ) for sufficiently differentiable functions, with ρ ≥ 1.
- Exactness through degree m requires the local error to vanish for every polynomial in the bivariate space P_m^2.
- Taylor expansion about the triangle center is used when the function has continuous mixed partial derivatives through order m+2.
3. Error in Approximating Definite Integrals on a Triangle.
The triangle-level rule is specified through nodes, weights, and polynomial exactness, with weights obtained from a linear system. The analysis uses scaled node geometry and Taylor-based error relations.
- The cubature rule uses a set of local nodes and weights to approximate the integral over each triangle.
- For a unisolvent node set, polynomial exactness is imposed through a full-rank matrix system with at least (m+1)(m+2)/2 nodes.
- When more nodes than polynomial basis functions are available, multiple exact weight sets exist and an objective such as the squared Euclidean norm can select one.
- Radial-basis-function constructions provide an alternative way to generate cubature nodes and weights through a quadratic program.
- The triangle is subdivided into four congruent triangles whose geometry supports comparing refined and original cubature errors.
5. Congruent Subdivision of a Triangle.
The triangle is subdivided into four congruent copies, and transformed versions of the original nodes and scaled weights are applied to each copy. A key lemma links refined monomial errors to the original error, enabling local superconvergence.
- 5. Congruent Subdivision of a Triangle.: Connecting the side midpoints divides a triangle into four congruent triangles with the same interior angles.
- 5. Congruent Subdivision of a Triangle.: Refined node sets are formed by shifting, rotating, and scaling the original nodes, while weights are multiplied by (1/2)^2.
- 5. Congruent Subdivision of a Triangle.: The transformed cubature rules remain exact for polynomials through degree m on each congruent triangle.
- 5. Congruent Subdivision of a Triangle.: A lemma relates the error for degree-m+1 monomials on each refined triangle directly to the original triangle's error.
- 5. Congruent Subdivision of a Triangle.: Summing the four refined errors and comparing them with the original error produces the local superconvergence result.
7. A Relationship Between Errors Under Refinement.
For cubature rules exact through degree m, the refinement error has a parity-dependent leading order, with even m yielding an extra convergence order. The result follows by comparing errors at spacing h_k and h_k/2 under the stated regularity and exactness assumptions.
- Theorem assumptions: For an exact degree-m cubature rule, the local error has a stated asymptotic form under the theorem’s smoothness and weight assumptions.The theorem assumes f is continuous with continuous mixed partial derivatives and requires polynomial exactness through degree m.
- Error refinement: The refinement ratio E_tk,0,hk/2[f]/E_tk,0,hk[f] equals (1/2)^ρ(1 + o(1)).This ratio compares the local errors before and after halving the triangle spacing.
- Error refinement: As h_k → 0, the exponent is ρ = m + 3 + 1.The exponent is obtained by expressing the local error as a power series in h_k and comparing the leading terms.
- Global implication: Summing local triangle errors preserves the extra convergence order when m is even, provided K = O(h^-2).The global conclusion applies when the theorem’s assumptions hold for every triangle and h is the maximum triangle spacing.
8. Total Error Over a Polygonal Domain.
Under suitable triangulation regularity, the total polygonal-domain error inherits the local superconvergence behavior. Numerical experiments on quasi-uniform and pseudo-random node sets show the predicted rates for several polynomial degrees and weight-construction methods.
- Total-error assumptions: Regular triangulations with K = O(h^-2) support the asymptotic total-error estimate over the polygonal domain.The stated setting includes Delaunay triangulations of quasi-uniform nodes and suitable shape-regular triangulations of randomly spaced nodes.
- Total-error rate: An extra order of convergence is realized for even m in the total error E_Ω,h[f].The parity effect follows from the factor 1 − (−1)^(m+1) in the asymptotic expression.
- Numerical experiments: Experiments average errors over 10 node sets and 30 values of N ranging from 10^3 to 10^5.The tests use quasi-uniformly or pseudo-randomly spaced nodes and triangulations of the unit square.
- Numerical experiments: The observed error rates match the predicted behavior for both weight-construction methods, with RBF-based cubature performing slightly better in this case.The comparison is reported for the quasi-uniform node sets shown in the numerical-results figures.
10. Conclusions.
The conclusion recommends choosing even polynomial exactness orders when constructing cubature rules over polygonal domains subdivided into triangles. Because the required local node count grows with the polynomial order, keeping m small is also important.
- Conclusion: Even m should be a primary design consideration for cubature rules over triangular subdomains.The recommendation combines the superconvergence result with the node-count cost of increasing m.
- Conclusion: Keeping m as small as possible is important because the number of required cubature nodes and weights depends on m.The conclusion links polynomial degree selection to the solvability requirements of the local linear system.