“…In this paper, we restrict our attention to layer potentials and assume that the volume forcing term F (x, t) = 0. (Volume integrals are discussed, for example, in [14,25].) For the Dirichlet problem (α = 1 and β = 0 in (1.3)), taking the limit as a point x ∈ Ω(t) approaches a point x o ∈ Γ(t), we obtain the integral equation…”
Abstract. We discuss the numerical evaluation of single and double layer heat potentials in two dimensions on stationary and moving boundaries. One of the principal difficulties in designing high order methods concerns the local behavior of the heat kernel, which is both weakly singular in time and rapidly decaying in space. We show that standard quadrature schemes suffer from a poorly recognized form of inaccuracy, which we refer to as "geometrically induced stiffness," but that rules based on product integration of the full heat kernel in time are robust. When combined with previously developed fast algorithms for the evolution of the "history part" of layer potentials, diffusion processes in complex, moving geometries can be computed accurately and in nearly optimal time.
“…In this paper, we restrict our attention to layer potentials and assume that the volume forcing term F (x, t) = 0. (Volume integrals are discussed, for example, in [14,25].) For the Dirichlet problem (α = 1 and β = 0 in (1.3)), taking the limit as a point x ∈ Ω(t) approaches a point x o ∈ Γ(t), we obtain the integral equation…”
Abstract. We discuss the numerical evaluation of single and double layer heat potentials in two dimensions on stationary and moving boundaries. One of the principal difficulties in designing high order methods concerns the local behavior of the heat kernel, which is both weakly singular in time and rapidly decaying in space. We show that standard quadrature schemes suffer from a poorly recognized form of inaccuracy, which we refer to as "geometrically induced stiffness," but that rules based on product integration of the full heat kernel in time are robust. When combined with previously developed fast algorithms for the evolution of the "history part" of layer potentials, diffusion processes in complex, moving geometries can be computed accurately and in nearly optimal time.
“…Once σ or µ is known, (1.7), (1.9) can be used to evaluate the solution at time t = ∆t. This yields a one-step marching method for the heat equation that is both stable and robust (see, for example, [1,5,7,8,13,18,21,30,31]). …”
Section: Introductionmentioning
confidence: 99%
“…Some notable prior work on continuous (volume) fast transforms includes [27], which describes a triangulation-based adaptive refinement method, and [31], which makes use of a high-order, adaptive, quad-tree-based discretization. As in [31], our approach relies on an adaptive quadtree with high-order Chebyshev grids on leaf nodes, but we carry out a modified version of the FGT on the quad-tree itself. This requires a somewhat more complicated implementation, following that of the hierarchical fast multipole method (FMM) [14].…”
Abstract.A variety of problems in computational physics and engineering require the convolution of the heat kernel (a Gaussian) with either discrete sources, densities supported on boundaries, or continuous volume distributions. We present a unified fast Gauss transform for this purpose in two dimensions, making use of an adaptive quad-tree discretization on a unit square which is assumed to contain all sources. Our implementation permits either free-space or periodic boundary conditions to be imposed, and is efficient for any choice of variance in the Gaussian.
“…Discrete sums of the form (1) are encountered in a variety of disciplines including computational physics, machine learning, computational finance and computer graphics [1], [2], [3], [4], [5]. The motivation for the present work comes from potential theory [6] applied to solving linear constant-coefficient parabolic partial differential equations (PDEs).…”
We present fast adaptive parallel algorithms to compute the sum of N Gaussians at N points. Direct sequential computation of this sum would take O(N 2 ) time. The parallel time complexity estimates for our algorithms are O N np for uniform point distributions and O N np log N np + np log np for nonuniform distributions using np CPUs. We incorporate a planewave representation of the Gaussian kernel which permits "diagonal translation". We use parallel octrees and a new scheme for translating the plane-waves to efficiently handle nonuniform distributions. Computing the transform to six-digit accuracy at 120 billion points took approximately 140 seconds using 4096 cores on the Jaguar supercomputer at the Oak Ridge National Laboratory.Our implementation is kernel-independent and can handle other "Gaussian-type" kernels even when an explicit analytic expression for the kernel is not known. These algorithms form a new class of core computational machinery for solving parabolic PDEs on massively parallel architectures.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.