new interactive example with time stepping

[Robert Cimrman]

Hi,

I have just updated the time stepping solvers in sfepy for interactive use, as
demonstrated in the new example [1]. For basic use, ignore the probing code -
the time stepper can be used as simply as:

tss = SimpleTimeSteppingSolver({'t0' : 0.0, 't1' : 100.0, 'n_step' : 11},
problem=problem)
tss.init_time()
for step, time, state in tss():
pass

r.

[1] http://sfepy.org/doc-devel/examples/diffusion/time_poisson_interactive.html

[Jan Heczko]

Hi,

I tried to use this approach ([1]) to solve a diffusion problem with a field-dependent material as in [2], but I got:

ValueError: material data not set! (call time_update())

By adding some print-calls to the material-defining function, I found out that the function gets called, but with mode='special' instead of 'qp'.
As a result None is returned instead of the values that correspond to the initial conditions.

My question(s):
- Is the SimpleTimeSteppingSolver usable for such problem? Is there an example?
- Should I use/implement a different time-stepper?

Thanks,
Jan

[2] http://sfepy.org/doc-devel/examples/diffusion/poisson_field_dependent_material.html

[Jan Heczko]

Update: There was a mistake in naming the materials. With things corrected, however, the problem remains.
I attach the file (oxygen_diffusion.py). Running it results in

ValueError: data of variable are not set! (q, step 0)

When the material parameters are set to constants, the computation runs.
The same happens when using the config-file with `ts` time-stepper, e.g. changing [2] to time-dependent problem (time_field_dependent.py).

J.

[Robert Cimrman]

Hi Jan,

SimpleTimeSteppingSolver just calls Problem.solve(), so it should update the
materials as usual (= also in QP). I will check it when I get to it.

r.

[Robert Cimrman]

Hi Jan,

the problem is, that the material update is called before the initial time
step, and the initial conditions (i.e., the values of DOFs in variables) are
not set yet. I will make a PR fixing that soon.

r.

[Jan Heczko]

Yes, that's probably the reason, why I added

    if ts.step == 0:
        state = problem.create_state()
        state.apply_ic()
        problem.equations.variables.set_data(state())

to `get_reaction_coefs`. But even with that piece of code, the solution fails in the second step. The error message says `step 0`, but it is actually the second time-step.

I was hoping to only get some clues to fix things myself, but you doing the work is ok as well :-)
J.

[Robert Cimrman]

Try [1], please. The time_field_dependent.py example seems to be working. With
oxygen_difusion.py I no longer get the exception, but NaNs are in the solution
- this might be unrelated to the issue, though.

r.

[1] https://github.com/sfepy/sfepy/pull/369

[Jan Heczko]

Works for me.
The NaNs are most probably a convergence-related issue, but that is what I actually wanted to work on.
Thanks a lot for fixing the solver!

[Robert Cimrman]

OK, I will merge the PR.

r.