Dear Kostas, Yes indeed, the sign is different between the two versions so that the sign of the residual have to be systematically changed transforming an old brick into a new one. The reason is that the new break system consider directly right hand sides (for linear bricks it is not still necessary to add to the residual the matrix times the value of the variable). So the vectors in the new brick systems have to be considered as right hand side (minus the residual for the newton algorithm). It is explained in the documentation : http://download.gna.org/getfem/doc/getfemuser/getfemuser_20.html#id1
Yves. Konstantinos Poulios <[email protected]> a écrit : > I have found some difficulties in nonlinear solutions in combination > with the new brick system. For a very simple nonlinear model that I > have converted from the old brick system, the classical_Newton > subroutine diverges. Trying to trace down the problem I realized that > in the first iteration step the residual of the model_pb object used > in the classical_Newton subroutine, has a different sign as the > corresponding one in the old brick system. I my test case the residual > I have checked was exclusively due to a source_term. Perhaps the > following two screenshots could be helpful to understand: > > http://files.ubuntu-gr.org/forum/logari81/getfem/getfem_old_brick_system.png > http://files.ubuntu-gr.org/forum/logari81/getfem/getfem_new_brick_system.png > > Was this sign difference to expect? Should this sign difference be > taken into consideration when migrating bricks from the old system to > the new one? Could it be a bug? My question affects version 4.0.0. > > Kostas > > _______________________________________________ > Getfem-users mailing list > [email protected] > https://mail.gna.org/listinfo/getfem-users > _______________________________________________ Getfem-users mailing list [email protected] https://mail.gna.org/listinfo/getfem-users
