Dispersion-optimized quadrature rules for isogeometric analysis: modified inner products, their dispersion properties, and optimally blended schemes

Vladimir Puzyrev, Quanling Deng, Victor M. CaloarXiv 2016 · DOI · OpenReview

Abstract

This paper introduces optimally-blended quadrature rules for isogeometric analysis and analyzes the numerical dispersion of the resulting discretizations. To quantify the approximation errors when we modify the inner products, we generalize the Pythagorean eigenvalue theorem of Strang and Fix. The proposed blended quadrature rules have advantages over alternative integration rules for isogeometric analysis on uniform and non-uniform meshes as well as for different polynomial orders and continuity of the basis. The optimally-blended schemes improve the convergence rate of the method by two orders with respect to the fully-integrated Galerkin method. The proposed technique increases the accuracy and robustness of isogeometric analysis for wave propagation problems.

Introduction

In the simulation of wave propagation—spanning acoustics, electromagnetics, and seismology—numerical dispersion remains a primary source of error. Numerical dispersion occurs when the discrete representation of a wave travels at a velocity different from its physical counterpart. Over long durations or distances, these infinitesimal speed discrepancies accumulate into significant phase errors, potentially invalidating the simulation results.

Isogeometric Analysis (IGA), which utilizes smooth basis functions like B-splines from CAD environments, has demonstrated superior dispersion properties compared to traditional $C^0$ Finite Element Analysis (FEA). However, even IGA is not immune to dispersion. This research explores how strategically modifying the numerical integration (quadrature) used to form the discrete system can minimize these errors, effectively providing a "superconvergent" boost to accuracy without increasing the computational degrees of freedom.

Theoretical Framework: The Generalized Pythagorean Eigenvalue Theorem

To understand how modified integration affects the system, the authors extend the classical Pythagorean eigenvalue theorem originally proposed by Strang and Fix. In standard Galerkin methods, this theorem relates eigenvalue and eigenfunction errors to the total energy error.

By generalizing this theorem to account for "inexact integration" (modified inner products), the authors established a precise mathematical framework to quantify the "error budget." This allows researchers to see exactly how changes in the $L_2$ product and discrete energy norm—induced by different quadrature rules—interact with the underlying discretization error. This theoretical foundation is what enables the systematic design of "blended" rules that cancel out leading-order errors.

Methodology: Optimally-Blended Quadrature

The core innovation lies in the combination of two standard numerical integration techniques:

  1. Gauss-Legendre Quadrature: Typically results in a "phase lead," where numerical waves travel faster than physical waves.
  2. Gauss-Lobatto Quadrature: Often results in a "phase lag," where numerical waves travel slower than physical waves.

By creating a weighted average (blending) of these two rules, the authors can force the leading error terms to cancel each other out. The study identifies specific optimal blending ratios for different spline spaces:

  • $C^1$ Quadratic Elements: An optimal ratio of 2:1 (Lobatto to Gauss) was identified.
  • $C^2$ Cubic Elements: The authors discovered that a "non-convex" blending—utilizing a negative weight for one of the rules—is necessary to achieve peak accuracy.

Key Insights and Results

Superconvergence for "Free"

The most significant result is the achievement of $O(h^{2p+2})$ convergence for eigenvalues, compared to the $O(h^{2p})$ rate found in standard fully-integrated Galerkin IGA. Because the number of integration points remains comparable to standard methods, this two-order increase in accuracy is essentially "free" in terms of computational cost per degree of freedom.

Spectrum Stability

Unlike traditional FEA, which often suffers from "spikes" or optical branches in the error spectrum due to low continuity at element boundaries, the high continuity of IGA ($C^{p-1}$) provides a stable platform for these blending techniques. While the "optimal" blending is tuned for the low-frequency limit ($\Omega \to 0$), the authors also demonstrated that alternative blending parameters can be selected to minimize errors in the middle of the frequency spectrum, which is often more relevant for practical engineering applications.

Robustness Across Grids

The researchers tested the blended schemes against standard benchmarks, including 1D and 2D Laplace eigenvalue problems. The methods proved robust not only on uniform meshes but also on non-uniform and "stretched" grids. While the superconvergence effect is most mathematically perfect on uniform grids, the accuracy gains remain substantial in more complex, realistic mesh configurations.

Conclusion and Future Directions

Optimally-blended quadrature represents a significant step forward in making IGA a more robust tool for structural dynamics and wave physics. By balancing the error budget through modified inner products, the method achieves a level of precision previously requiring much finer meshes.

Future work will focus on streamlining the implementation. Currently, blending requires two separate integration passes; developing a single, non-standard quadrature rule (specific nodes and weights) that achieves the same result in one pass would further optimize the process. Additionally, extending these rules to handle variable continuity and complex 3D NURBS geometries remains an active area of development.

Summary updated

Abstract

This paper introduces optimally-blended quadrature rules for isogeometric analysis and analyzes the numerical dispersion of the resulting discretizations. To quantify the approximation errors when we modify the inner products, we generalize the Pythagorean eigenvalue theorem of Strang and Fix. The proposed blended quadrature rules have advantages over alternative integration rules for isogeometric analysis on uniform and non-uniform meshes as well as for different polynomial orders and continuity of the basis. The optimally-blended schemes improve the convergence rate of the method by two orders with respect to the fully-integrated Galerkin method. The proposed technique increases the accuracy and robustness of isogeometric analysis for wave propagation problems.
BibTeX
@misc{puzyrev2016dispersion,
  author = {Vladimir Puzyrev and Quanling Deng and Victor M. Calo},
  title = {{Dispersion-optimized quadrature rules for isogeometric analysis: modified inner products, their dispersion properties, and optimally blended schemes}},
  year = {2016},
  eprint = {1612.09097},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1612.09097},
}

Loading related content...

Figure