Increased quadrature order in a term.

This is for the benefit of bi/tri-linear basis functions, as stated
in the comments.
This commit is contained in:
Atgeirr Flø Rasmussen 2013-01-16 13:45:15 +01:00
parent f539c4d183
commit 5288b84bdc

View File

@ -510,8 +510,15 @@ namespace Opm
// for jac_, i.e. rows cycling fastest.
{
// Even with ECVI velocity interpolation, degree of precision 1
// is sufficient for optimal convergence order for DG1.
const int deg_needed = 2*degree_ - 1;
// is sufficient for optimal convergence order for DG1 when we
// use linear (total degree 1) basis functions.
// With bi(tri)-linear basis functions, it still seems sufficient
// for convergence order 2, but the solution looks much better and
// has significantly lower error with degree of precision 2.
// For now, we err on the side of caution, and use 2*degree, even
// though this is wasteful for the pure linear basis functions.
// const int deg_needed = 2*degree_ - 1;
const int deg_needed = 2*degree_;
CellQuadrature quad(grid_, cell, deg_needed);
for (int quad_pt = 0; quad_pt < quad.numQuadPts(); ++quad_pt) {
// b_i (v \cdot \grad b_j)