Efficient Algorithms for Computing Multidimensional Integral Fractional Laplacians Via Spherical Means

Boxi Xu,Jin Cheng,Shingyu Leung,Jianliang Qian
DOI: https://doi.org/10.1137/19m1262358
IF: 2.968
2020-01-01
SIAM Journal on Scientific Computing
Abstract:We develop efficient algorithms for computing the multidimensional fractional operator (-Delta x)(alpha/2) in the form of hypersingular integral in the entire space, where the operator is the so-called integral fractional Laplacian when 0 < alpha < 2. By introducing polar coordinates, we reduce applying the multidimensional integral fractional operator to a function to applying the resulting one-dimensional fractional operator to the spherical mean of the underlying function. We propose two algorithms to compute spherical means of a given function: one by solving standard wave equations and the other by solving Darboux's equations. We further apply a finite difference numerical quadrature approach to compute the one-dimensional fractional operator. Our methodology is equally applicable to computing the action of the integral fractional Laplacian, the extended integral fractional Laplacian, and the Riesz potential operator, respectively, and the computational complexity for applying each of the three operators is O(Ln+1), where L denotes the number of mesh points in each spatial direction and n is the spatial dimension. Numerical examples, including algebraically decaying functions with varying regularities, demonstrate the performance and convergence rates of our new algorithms.
What problem does this paper attempt to address?