Resetting solver after convergence failure.

[ethrelfall]

Dear community,

I'm running a parameter space scan for a convection simulation - two parameters, Rayleigh number Ra and Prandtl number Pr.

I'm doing a simple continuation method, outer loop picks a value for Pr and then inner loop over increasing Ra starting from 1. Let's say I iterate over Ra increasing until a convergence failure happens, then I want to increment Pr and restart the Ra from 1 (I do this by catching the exception and breaking out of the inner loop).

My question is how can I reset the solver after a convergence failure has occurred - at the moment, the next simulation after the convergence failure still seems to have memory of the failure, i.e. some bad values some how, (so it fails too, as a result of this) and I want to restart from a "clean" condition.

I've tried using interpolate to re-initialize the functions involved but this seems to have no effect.

Did search for matching discussion - nada.

Thanks,

Ed.

[ethrelfall]

I guess this wasn't a very good question - anyway I've made it work by simply instantiating a new solver for each new value of Pr.

[wence-]

I think it should be the case that reinitialising the initial guess to a safe value (say zero) should be enough. Can you show a minimal example that breaks?

[ethrelfall]

Thanks, there is example at

https://github.com/ethrelfall/Firedrake-questions/tree/main/10Feb2023

The inner loop runs until Ra gets too high and it fails - I then want to do the next pass of the inner loop but this fails because of the earlier failure. I tried to reset the fields using interpolate, but it still fails (me doing the reset wrong, probably?).

[Hasupama]

Hi
I am having the same issue. Have you found a solution for resetting the solver after a convergence failure has occurred? If so could you share it?