[CIG-LONG] Gale at AGU

[Walter Landry]

Greetings Everyone,

I am attaching the abstract of the poster I will be presenting at AGU
on the morning of Wednesday, Dec 16.  If anyone wants to meet with me
to discuss any issues they have with Gale, please send me a note.  I
will be arriving on Sunday and leaving Thursday night.

Cheers,
Walter Landry
walter at geodynamics.org


ID# DI31A-1597
Location: Poster Hall (Moscone South)
Time of Presentation: Dec 16 8:00 AM - 12:20 PM

Robust, Scalable Methods for Incompressible Stokes Flow with Yielding Rheologies
W. Landry

Many finite element codes developed for crustal scale problems use
Q1-P0 elements. This scheme has a well known checkerboard
instability. Generally, this instability does not cause problems. When
the problem becomes strongly non-linear, such as with yielding
rheologies, then the method breaks down. One fix for this is to use a
penalty method. Unfortunately, this makes the matrix very
ill-conditioned and requires direct solvers. While there are parallel
direct solvers, they do not scale up to extremely large problems.

There have been a number of other proposed fixes. Most of them require
higher order elements, macro-elements, calculating higher order
derivatives, or mesh-dependent stabilization parameters. Dohrmann &
Pachev (2000) introduced a relatively simple scheme using Q1-Q1
elements and a stabilization term that looks like a
compressibility. For models the size of a sandbox, this method works
quite well. Once it is scaled up to the lithosphere, the stabilization
term creates new problems that are only seen with strongly non-linear
yielding rheologies. We discuss the steps we have taken to overcome
these problems and the performance characteristics of the code on
large, 3D problems.

http://geodynamics.org/cig/software/packages/long/gale/