What is the expected output? What do you see instead?
When solving 1-D problems in time-stepping mode, the Newton / time-stepping
algorithm has an adjustable time-step size (or stride) which can increase or
decrease depending on the "ease" of obtaining solutions. Often the time-step
size has grown to large values when the steady-state solution has been reached.
This can be followed by grid refinement and a re-start of the solution
algorithm.
There is, however, a problem with the size of the time-step upon resumption of
the calculation. Currently, the step size is kept at the value that is used in
the last time-step, rather than being re-set to the initial value defined in
the simulation. When time-stepping resumes, the algorithm often spends the
first few (ten or so) calculations progressively reducing the time-step size
till a new, smaller value can be found which lies within the transient domain
of convergence. This can be a costly procedure requiring numerous Jacobian
calculations.
This behaviour is often fixed in other software (e.g. Twopnt) by re-setting the
time-step size back to the initial user-defined value after grid refinement. I
think that a similar fix should be implemented here. However, this adjustment
to the solution algorithm is a cosmetic one, so I leave it up to the developers
to make the decision.
What version of the product are you using? On what operating system?
All versions.
Please provide any additional information below.
Adding the line "dt = m_tstep;" in the procedure "Sim1D::solve" in the if
statement around line 332 would do the trick. The affected lines would look
something like this:
if (refine_grid) {
new_points = refine(loglevel);
dt = m_tstep; // <= reset time step to initial value
if (new_points < 0) {
writelog("Maximum number of grid points reached.");
new_points = 0;
}
} else {
if (loglevel > 0) {
writelog("grid refinement disabled.\n");
}
new_points = 0;
}
Original issue reported on code.google.com by watso...@gmail.com on 10 Apr 2013 at 6:17
Original issue reported on code.google.com by
watso...@gmail.com
on 10 Apr 2013 at 6:17