Topic: global2local
Hi,
In a relatively recent commit (6143c82ef3c8f0f8ce5c2eda01e9b2c5b516a37a "Fixing global2local implementation"),
fei2dtrquad.C, fei2dquadquad.C, fei3dtetquad.C och fei3dwedgelin.C, the global2local function was changed
from jac.solveForRhs(res, delta, true); to jac.solveForRhs(res, delta);
This change removes a transpose from the Jacobian.
This change destroys the convergence in global2local for a problem I am currently working with.
My question is therefore:
Has anyone verified that it was indeed correct to modify the transpose in the functions mentioned above?
Performing the derivation myself I get the Jacobian as
J_ab = sum_i x_a^i (dN^i/dxi)_b,
but in giveJacobianMatrixAt, the Jacobian is computed as
J_ab = (dN^i/dxi)_a sum_i x_b^i.
Hence, it seems to me that the transpose should actually be there.
Regards,
Erik