[CIG-SHORT] Friction Solver optimization

[Romain Jolivet]

Dear Brad, Matt, Charles and other Pylith developers, 

I am currently trying to optimise the solver in an experiment I am running and I am having some difficulties getting my run to go faster...

The problem is 2D and has a fault cutting from top to bottom (with some angle).  Since I gravity is “on”, I have to prescribe adequate boundary conditions, particularly at the bottom of the model. As the fault cuts the entire domain, one of its node belongs as well to the bottom set of nodes. On this set of nodes, I therefore cannot impose DirichletBC (otherwise, I am fixing to many degrees of freedom on the node that also belongs to the fault). I therefore have imposed Neumann BC, with normal traction opposing gravity forces at the bottom of the domain.

When I try to use fieldsplit (to speed things up, following one of your tutorials, a few years back), the linear solver does not run since PETSc issues some NANORINF norm. When I remove the fault from the model, it runs fine. I guess this is because one node of my domain has 2 types of constraints (one Neumann BC and some frictional BC). 
Is that correct? Should I keep going that way or is it a bad idea in general (to mix Neumann and friction on the same node)?

Cheers, 
Romain  

—————————————————————————————————————
—————————————————————————————————————	
Romain Jolivet
Postdoctoral Fellow

University of Cambridge
Department of Earth Sciences
Bullard Labs
Madingley Rise 
Madingley Road
Cambridge CB3 0EZ
United Kingdom

email: rpj29 at cam.ac.uk
Phone: +44 1223 748 938
Mobile: +44 7596 703 148

France: +33 6 52 91 76 39
US: +1 (626) 560 6356
—————————————————————————————————————
—————————————————————————————————————