On the numerical solution of the heat equation I: Fast solvers in free space
Introduction
The solution of the heat equation (the diffusion equation) in free space or in unbounded regions arises as a modeling task in a variety of engineering, scientific, and financial applications. While the most commonly used approaches are based on finite difference (FD) and finite element (FE) methods, these must be coupled to artificial (non-reflecting) boundary conditions imposed on a finite computational domain in order to simulate the effect of diffusion into an infinite medium. These boundary conditions are discussed, for example, in [5], [14], [15], [16], [17]. Here, we describe a mathematically much more straightforward approach, which we will refer to as the fast recursive marching (FRM) method. It is based on evaluating the exact solution of the governing equation, using convolution in space and time with the free-space Green’s function. One advantage of this approach is that essentially no convergence theory is required. The error in the solution is simply the quadrature error in evaluating the solution. In the present paper, we restrict our attention to the simplest setting, namely the isotropic inhomogeneous heat equation in :in the absence of physical boundaries, subject to the initial conditionThe functions and are assumed to be compactly supported in the box , centered at the origin. We also assume that and are k-times differentiable: and . In subsequent works we will consider complex geometry and discontinuous data.
From standard potential theory [13], [23], the solution can be written aswhere the fundamental solution for the heat equation in is
We will refer to the first integral in (3) as an initial potential and the second integral as a volume potential. There is a substantial literature on Green’s function methods for problems of diffusion (see for example [3]). However, straightforward discretization of the above integrals leads to an enormously expensive numerical scheme – the solution is dependent on the full space–time history of the diffusion process. With N points in the discretization of the domain and M time steps, it is easy to see that work is required. Thus, the obvious advantages of the approach (stability, robustness, and the correctness of the far-field behavior) appear to be overwhelmed by the problems of cost.
In recent years, however, several fast algorithms have been developed [10], [11], [12], [26] that lead to nearly optimal schemes, for which the work required is . Related methods can be found in [18], [20], [27]. They involve a fairly substantial amount of numerical and analytic machinery. In the present context, where we need to evaluate volume potentials with smooth data, a simpler method can be developed based entirely on the continuous Fourier transform. After outlining the algorithm itself, we illustrate its performance with some numerical examples from materials science, simulating dendritic solidification.
Section snippets
Fourier representation of the solution
While (3) describes the solution to the heat equations (1), (2), significant advantage can be obtained by considering its Fourier transform. For this, we letIt is obvious from (1), (2), and well known that satisfies the ordinary differential equationwhereAn elementary calculation shows thatwhereThus,
The numerical scheme
With the full complement of tools in place, we can now provide an informal description of the fast recursive marching (FRM) method. For kth order accuracy in time, we refer to the method as FRM(k).
The FRM(2) method
- Step 1:
Initialization
(a) Select time step and number of time steps M.
(b) Select precision ϵ for quadrature in Fourier domain.
(c) Obtain N1 quadrature nodes and weights in Fourier space for (18) according to Corollary 1.
(d) Compute weights W0, W1 from (10).
- Step 2:
Transform initial data
(a)
Numerical results
We first test the performance of the algorithm in two dimensions with a known exact solution , constructed so that , where V corresponds to an initial potential and W corresponds to a volume potential. For this, we letClearly V satisfies the homogeneous heat equation with initial dataWe let W satisfy the inhomogeneous heat equation
Conclusions and generalizations
We have described the fast recursive marching method, a simple Fourier-based method for the solution of the heat equation in free space with smooth initial data and a smooth source term. It allows for efficient and accurate long-time simulations without the need for artificial boundary conditions on a finite computational domain. The convergence theory can be stated trivially – the error in the solution is the quadrature error in computing the space–time integral (3). The CPU time of the method
Acknowledgment
This work was supported by the Applied Mathematical Sciences Program of the U.S. Department of Energy under Contract DEFG0288ER25053.
References (27)
Artificial boundary conditions for diffusion equations: numerical study
J. Comput. Appl. Math.
(1996)- et al.
Spectral approximation of the free-space heat kernel
Appl. Comput. Harmonic Anal.
(2000) - et al.
A class of artificial boundary conditions for heat equation in unbounded domains
Comput. Math. Appl.
(2002) - et al.
Exact and approximating boundary conditions for the parabolic problems on unbounded domains
Comput. Math. Appl.
(2002) - et al.
An integral equation method for epitaxial step-flow growth simulations
J. Comput. Phys.
(2006) - et al.
The type 3 nonuniform FFT and its applications
J. Comput. Phys.
(2005) - et al.
Simulations of experimentally observed dendritic growth behavior using a phase-field model
J. Cryst. Growth
(1995) - et al.
Adaptive phase-fieldnext term computations of dendritic crystal growth
J. Cryst. Growth
(1997) A fast method for solving the heat equation by layer potentials
J. Comput. Phys.
(2007)- John Adams, Paul Swarztrauber, Roland Sweet, FISHPACK: a package of FORTRAN subprograms for the solution of separable...
On the fast Fourier transform of functions with singularities
Appl. Comput. Harmonic Anal.
Boundary Element Techniques
Methods of Numerical Integration
Cited by (40)
Fast integral equation methods for Rothe's method applied to the isotropic heat equation
2011, Computers and Mathematics with ApplicationsStability and convergence analysis of high-order numerical schemes with DtN-type absorbing boundary conditions for nonlocal wave equations
2024, IMA Journal of Numerical AnalysisHow Does Thermal Pressurization of Pore Fluids Affect 3D Strike-Slip Earthquake Dynamics and Ground Motions?
2023, Bulletin of the Seismological Society of AmericaFOURIER REPRESENTATION OF THE DIFFUSION MRI SIGNAL USING LAYER POTENTIALS
2023, SIAM Journal on Applied MathematicsStochastic heat of the third type model under mean square calculus
2023, Numerical Heat Transfer, Part B: Fundamentals