[petsc-users] Fieldsplit with LSC for constrained elasticity/poroelasticity?

[Jed Brown]

[re-adding list to Cc]

Tabrez Ali <stali at geology.wisc.edu> writes:

> Jed
>
> The reason I did not nondimenionalize was because it didnt make gmres 
> with asm go any faster for my test cases.

The true residual spikes up by 15 orders of magnitude when you start
solving (with MUMPS on fieldsplit_0_).  That is a major problem and will
hopefully be resolved if you scale the fields reasonably.  The scales
should be balanced so that residuals (respectively errors) are balanced
when your sense of quality of solution is balanced.

Note that triangular fieldsplit is a nonnormal preconditioner, so this
is more important there.

If don't nondimensionalize, you can use Schur Complement Reduction, but
with a scale imbalance of 1e15, you can't combine with an outer
iteration (usually not desirable for SCR, but you may have to solve the
Schur complement problem very accurately to keep errors small).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20141028/9af21725/attachment.pgp>