[CIG-SHORT] Adaptive time stepping in PyLith

[Brad Aagaard]

Tabrez Ali wrote:
> I just want to confirm:
> 
> 1. If adaptive time stepping factors in the relaxation time.
> 2. It can be used safely when we have velocity BC's in place
> 
> Also,
> 
> - I just ran an example having velocity BC's (total_time=1.0*year)  
> with a fixed time step of 0.1*year and it runs fine.
> 
> - With adaptive time stepping (max_dt=1.0*year; adapt_skip=10) it  
> also runs fine and seems to take a time step of 0.088*year.
> 
> - But if I use max_dt=5.0*year (adapt_skip=10) then things go wrong  
> as the solution fails to converge. Even though it appears as if the  
> time step it takes is 0.088*year (from looking at the "Finishing  
> advancing solution from ..." info).
> 
> Now I have checked everything twice and am confused as to how is this  
> even possible. Why in one case (max_dt=1.0*year) the solution  
> converges and in the second case (max_dt=5.0*year) it doesnt even  
> when the time step PyLith seems to take appears to be the same  
> (0.088*year) in both cases.

This really sounds like a bug. I will look into it as soon as I get a 
chance.

Brad