M. Shashkov et S. Steinberg, SOLVING DIFFUSION-EQUATIONS WITH ROUGH COEFFICIENTS IN ROUGH GRIDS, Journal of computational physics, 129(2), 1996, pp. 383-405
A finite-difference algorithm for the numerical solution of diffusion
problems in strongly heterogeneous and nonisotropic media is construct
ed for logically rectangular grids. The performance of this algorithm
is comparable to other algorithms for problems with smooth coefficient
s and regular grids, and it is superior for problems with rough coeffi
cients and/or skewed grids. The algorithm is derived using the support
-operators method, which constructs discrete analogs of the divergence
and flux operator that satisfy discrete analogs of the important inte
gral identities relating the continuum operators. This paper gives the
first application of this method to the solution of diffusion problem
s in heterogeneous an nonisotropic media. The support-operators method
forces the discrete analog of the flux operator to be the negative ad
joint of the discrete divergence in an inner product weighted by the c
onductivity, as in the differential case. Once this is accomplished, m
any other important properties follow; for example, the scheme is cons
ervative and the discrete analog of the variable material Laplacian is
symmetric and negative definite. In addition, on any grid, the discre
te divergence is zero on constant vectors and the discrete flux operat
or is exact for linear functions in case when K is piecewise constant.
Moreover, the discrete gradient's null space is the constant function
s, just as in the continuum. Because the algorithm is flux based, it h
as twice as many unknowns as more standard algorithms. However, the ma
trices that need to be inverted are symmetric and positive definite, s
o the most powerful linear solvers can be applied. Also, the scheme is
second-order accurate so, all things considered, it is efficient. For
rectangular grids, the discrete operators reduce to well-known discre
te operators and the treatment of discontinuous conductivity coefficie
nts in the case of isotropic media is equivalent to the well-known har
monic-averaging procedure. Comparison with standard schemes is present
ed. Numerical examples validate advantage of new method. (C) 1996 Acad
emic Press, Inc.