Closed nickhale closed 5 years ago
Although the computed coefficients of the solution in the following match to machine precision, the evaluated solutions are very different (presumably with error in the evaluation of the quad version):
p = 60; rhs = -1; N = chebop2(@(u) lap(u)); N.bc = 0; sol = N\rhs; % Solve using rectangles R = ultraSEMDomain.rectangle; S = ultraSEM(R, {1,0,0}, rhs, p); sol1 = S\0; figure [xx, yy] = getGrid(sol1); err = abs(feval(sol, xx, yy) - feval(sol1, xx, yy)); surf(xx, yy, log10(err)); % Solve using quads R2 = ultraSEMDomain.quad([-1 -1; 1 -1 ; 1 1 ; -1 1]); S = ultraSEM(R2, {1,0,0}, rhs, p); sol2 = S\0; figure [xx, yy] = getGrid(sol1); err = abs(feval(sol, xx, yy) - feval(sol2, xx, yy)); surf(xx, yy, log10(err)); alignfigs err_coeffs = norm(sol2.u{1} - sol1.u{1}, inf)
Fixed in 00bf7652d114e6820281fda397913b79d0c4dc2c.
Although the computed coefficients of the solution in the following match to machine precision, the evaluated solutions are very different (presumably with error in the evaluation of the quad version):