Q. T. Le Gia,William McLean
Abstract:We propose a method to construct numerical solutions of parabolic equations on the unit sphere. The time discretization uses Laplace transforms and quadrature. The spatial approximation of the solution employs radial basis functions restricted to the sphere. The method allows us to construct high accuracy numerical solutions in parallel. We establish $L_2$ error estimates for smooth and nonsmooth initial data, and describe some numerical experiments.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: solving parabolic partial differential equations on the unit sphere. Specifically, the author proposes a new numerical method to construct high - precision numerical solutions of parabolic equations on the unit sphere. This method combines Laplace transforms for time discretization and radial basis functions (RBFs) for space discretization.
### Specific problem description
1. **Time discretization**: Traditional methods usually use the time stepping method, while this paper proposes using the Laplace transform to transform the problem in the time domain into a problem in the complex frequency domain, and then approximately solve it by numerical integration.
2. **Space discretization**: For space discretization, this paper adopts spherical radial basis functions (SRBFs) instead of the traditional finite element method. SRBFs have better flexibility and adaptability when dealing with problems on Riemannian manifolds such as the sphere.
3. **Error estimation**: The author establishes \( L^2 \) error estimates for smooth and non - smooth initial data and describes the results of some numerical experiments to verify the effectiveness and accuracy of the proposed method.
### Mathematical model
Consider the initial value problem:
\[
\frac{\partial u}{\partial t}+A u = f(t),\quad t > 0,\quad u(0)=u_0,
\]
where \( A \) is a linear, self - adjoint, positive semi - definite second - order elliptic partial differential operator on the unit sphere. Typically, \(-A\) is the Laplace - Beltrami operator. The source term \( f(t)\) may depend on the space variable, but this dependence is omitted in the notation, and \( f(t)\) is regarded as an element of the function space on the sphere.
### Solution
The method proposed by the author includes the following steps:
1. **Laplace transform**: Perform the Laplace transform on the time variable \( t \) to obtain the formal solution:
\[
(zI + A)\hat{u}(z)=g(z):=u_0+\hat{f}(z),
\]
where \( I \) represents the identity operator.
2. **Numerical integration**: By choosing an appropriate integration path \(\Gamma\), use the inverse Laplace transform formula to recover the solution in the time domain:
\[
u(t)=\frac{1}{2\pi i}\int_{\Gamma}e^{zt}\hat{u}(z)dz.
\]
3. **Space discretization**: Use spherical radial basis functions for space discretization to construct high - precision numerical solutions.
4. **Error analysis**: Establish strict error estimates for smooth and non - smooth initial data to ensure the accuracy and stability of the numerical solutions.
### Summary
The main contribution of this paper is to provide a novel and efficient numerical method for solving parabolic equations on the unit sphere, which is especially suitable for application scenarios requiring high - precision and parallel computing.