Fast evaluation in the extrapolated regularization method

Authors: Joseph Siebor, Svetlana Tlupova

SUNY Campus: Farmingdale State College

Presentation Type: Poster

Location: Old Union Hall

Presentation #: 40

Timeslot: Session C 1:45-2:45 PM

Abstract: Boundary integral equation methods are widely used in the solution of many partial differential equations. The kernels that appear in these surface integrals are nearly singular when evaluated near the boundary, and straightforward numerical integration produces inaccurate results. In Beale and Tlupova (Adv. Comput. Math, 2024), an extrapolated regularization method was proposed to accurately evaluate the nearly singular single and double-layer surface integrals for harmonic potentials or Stokes flow. The kernels are regularized using a smoothing parameter, and then a standard quadrature is applied. The integrals are computed this way for three smoothing parameter choices to find the extrapolated value to fifth order accuracy. In this work, we apply several techniques to reduce the computational cost of the extrapolated regularization method. First, we use a straightforward OpenMP parallelization over the target points. Second, we note that the effect of the regularization is local and evaluate only the local component of the sum for three values of the smoothing parameter. The non-local component of the sum is only evaluated once and re-used in the other sums. This component is still the computational bottleneck as it is O(N^2), where N is the system size. Finally, we apply the kernel-independent treecode to these far-field interactions to reduce the CPU time. We carry out experiments to determine optimal parameters both in terms of accuracy and efficiency of the computations.