material_nonlinearity example
Robert - in the new example I notice you are not using presolve in the solver definition or definition the history term for the displacement variable:
'ls' : ('ls.scipy_direct', {
'presolve' : True
}),
variables = { 'u' : ('unknown field', 'displacement', 0., 'previous'), 'v' : ('test field', 'displacement', 'u'), }
but consistently used these for other time step problems. Any reason? This really emphasizes my lack of understanding of the internals.
On 12/20/10 17:45, Andre Smit wrote:
Robert - in the new example I notice you are not using presolve in the solver definition or definition the history term for the displacement variable:
'ls' : ('ls.scipy_direct', { 'presolve' : True }),
presolve set to True means that, for linear problems ('problem' : 'linear' in newton configuration), the matrix is assembled and factorized before the time-stepping starts - the factors are then reused in each time step to quickly solve the linear systems.
It also assumes that the matrix is the same in each time step - this is not the case here, as 'mu' changes... So presolve does not have sense in this example.
variables = { 'u' : ('unknown field', 'displacement', 0., 'previous'), 'v' : ('test field', 'displacement', 'u'), }
but consistently used these for other time step problems. Any reason? This really emphasizes my lack of understanding of the internals.
For linear problems with constant matrix, the most efficient would be:
solvers = { 'ls' : ('ls.scipy_direct', {'presolve' : True }), 'newton' : ('nls.newton', { 'i_max' : 1, 'eps_a' : 1e-10, 'eps_r' : 1.0, 'problem' : 'linear'}), 'ts' : ('ts.simple', {'t0' : 0.0, 't1' : 1.0, 'dt' : None, 'n_step' : 5, 'quasistatic' : True, # or False }), }
I have just fixed a small bug with 'quasistatic' : True
r.
Robert
In the get_pars function you prepare the mu or val matrix:
if mode != 'qp': return
val = nm.empty((coors.shape[0], 1, 1), dtype=nm.float64)
val.fill(1e0)
I'm trying to make sense of this. Are you repeating the material parameter for each gauss point? So for example, for a quad, there are 4 gauss points and essentially mu = tile(mu,4) and reshaped?
a
On Mon, Dec 20, 2010 at 11:08 AM, Robert Cimrman cimr...@ntc.zcu.czwrote:
On 12/20/10 17:45, Andre Smit wrote:
Robert - in the new example I notice you are not using presolve in the solver definition or definition the history term for the displacement variable:
'ls' : ('ls.scipy_direct', { 'presolve' : True }),presolve set to True means that, for linear problems ('problem' : 'linear' in newton configuration), the matrix is assembled and factorized before the time-stepping starts - the factors are then reused in each time step to quickly solve the linear systems.
It also assumes that the matrix is the same in each time step - this is not the case here, as 'mu' changes... So presolve does not have sense in this example.
variables = {
'u' : ('unknown field', 'displacement', 0., 'previous'), 'v' : ('test field', 'displacement', 'u'),}
but consistently used these for other time step problems. Any reason? This really emphasizes my lack of understanding of the internals.
For linear problems with constant matrix, the most efficient would be:
solvers = {
'ls' : ('ls.scipy_direct', {'presolve' : True }), 'newton' : ('nls.newton', { 'i_max' : 1, 'eps_a' : 1e-10, 'eps_r' : 1.0, 'problem' : 'linear'}), 'ts' : ('ts.simple', {'t0' : 0.0, 't1' : 1.0, 'dt' : None, 'n_step' : 5, 'quasistatic' : True, # or False }), }
I have just fixed a small bug with 'quasistatic' : True
r.
-- You received this message because you are subscribed to the Google Groups "sfepy-devel" group. To post to this group, send email to sfepy...@googlegroups.com. To unsubscribe from this group, send email to sfepy-devel...@googlegroups.comsfepy-devel%...@googlegroups.com . For more options, visit this group at http://groups.google.com/group/sfepy-devel?hl=en.
participants (2)
-
Andre Smit -
Robert Cimrman