[petsc-dev] Fwd: flyer of PETSc for an atmospheric modeling workshop next week

Jed Brown jedbrown at mcs.anl.gov
Wed Aug 28 23:05:40 CDT 2013

Barry Smith <bsmith at mcs.anl.gov> writes:
>> We are hosting a workshop next Wed and Thur on modeling atmospheric
>> processes at large eddies and sub-grid scales. One of the strong
>> motivation is to leverage the high performance computer at ALCF and
>> computational libraries such as PETSc in resolving the small scales
>> in the atmosphere boundary layer, which are currently highly
>> parametrized in models.  Enclosed please find the Workshop abstract,
>> agenda, and list of attendees. More information is on
>> http://atmos.anl.gov/workshop

I don't have a flyer, but I can comment on where a solvers library like
PETSc could be useful.  The obvious start is in flexible elliptic solves
for low-Mach formulations.  Another place worth experimenting is in IMEX
time integration, especially when SSP and higher order accuracy is
required, but implicitness is also needed to handle vertical stiffness.
Since LES expects to resolve phase for horizontal advection, the more
difficult implicit solves associated with long time step or steady-state
are not needed.  PETSc has a few technologies that may be useful in
model spinup and variational data assimilation and experimental design.

While PETSc's structured-grid support has been used by Cai's group at CU
Boulder for global climate problems using implicit methods, it's not
ideal for that purpose and I can't recommend choosing it for LES.  If
using a spectral element or DG formulation, the unstructured grid
support should work fine, though you may want to specialize the local
kernels anyway.

In a different direction, we've recently add some of the latest
multiscale time integration algorithms.  Atmosphere tends not to have
clear scale separation, but these recent methods might be a piece of the
puzzle for constructing accurate upscaled models at lower cost than LES

PETSc also provides some useful debugging and profiling tools, a
convenient way to expose possibly-nested configuration, some reusable
interaction and monitoring features, and generally a lot of algorithmic
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130828/6cc4f80b/attachment.sig>

More information about the petsc-dev mailing list