[CIG-ALL] PyLith 1.0.2

Brad Aagaard baagaard at usgs.gov
Wed Dec 5 11:25:07 PST 2007


I am pleased to announce the 1.0.2 release of PyLith, a finite element
code designed to solve quasi-static viscoelastic problems in tectonic

This is a bugfix release, which fixes several bugs discovered since
the release of version 1.0.1 in July. We encourage all users of PyLith
1.0 to switch to this latest release. We have also made some
significant performance improvements. A binary package for Intel Macs
is now available too!

You can download the source code and binaries from


Installation instructions are in the bundled README and INSTALL  
files, as well as in the User Manual on the web page.


  * Performance optimizations have significantly reduced runtime and
    memory use relative to version 1.0.1. The default quadrature order
    for tetrahedral cells is now 1, which is appropriate for the
    default basis functions.

  * Tips

      - For most crustal deformation problems, we recommend using the
        Additive Schwartz preconditioner via the following PETSc

          + Command line arguments


          + pylithapp.cfg (or your other favorite .cfg file)

            pc_type = asm
            ksp_max_it = 100
            ksp_gmres_restart = 50
            ksp_rtol = 1.0e-08

      - If the solve takes more than a few hundred iterations for a
        large problem (use the --petsc.ksp_monitor=1 and
        --petsc.ksp_view=1 to see diagnostic information for the
        solver), this is an indication that something is wrong. Either
        the preconditioner is inappropriate for the type of problem you
        are solving (try other PETSc preconditioners) or there is an error in
        the problem setup.

  * Added checks to verify the compatibility of quadrature scheme for solid
    and cohesive cells.

  * Bug fixes

      - In some cases, cohesive cells were not inserted into the
        finite-element mesh properly. The cells mixed together vertices
        from the different sides of the fault. A more efficient
        procedure for creating cohesive cells fixed this problem.

      - Cell adjacency graph was created incorrectly which resulted in
        a poor quality of partitioning among processors.

      - VTK output for meshes with N faults included cohesive cells
        for N-1 faults. Since VTK output does not understand cohesive
        cells, we now remove all cohesive cells from the VTK output.

      - Using the SimpleDB in Spatialdata from Python limited
        interpolation to the "linear" scheme instead of allowing use of
        the "nearest" scheme. Setting the SimpleDB property to "nearest"
        and "linear" now works as expected.

      - The reader for Spatialdata coordinate systems information did not
        correctly putback characters in the input stream, resulting in
        reading errors. The putback routines were fixed.

      - Fault "up" and "normal" directions remained as string arrays
        when passed to the module, instead of being converted to float

  * Known issues

      - The VTK output contains the Lagrange vertices in the cohesive
        cells. These are forces, not displacements, which can cause
        confusion when visualizing the output. These will be removed
        when we improve the output in version 1.1.

      - Using the 32-bit linux binary on 64-bit linux systems

        The error message is ""merlin.DistributionNotFound:
        Cheetah". This error arises because one of the packages that
        PyLith uses does not know how to determine the system
        architecture at runtime. The workaround is:

        (1) Go to the lib/python2.5/site-packages directory.
        (2) Unzip merlin-1.3-py2.5.egg (if it is a file and not a directory).
        (3) Go to the merlin directory.
        (4) Edit __init__.py. Replace line 308
              plat = get_platform()
             plat = "linux-i686"
        (5) If merlin-1.3-py2.5.egg is a file, rezip merlin. Go to the
            site-packages directory and enter
            "zip -r merlin-1.3-py2.5.egg merlin".

More information about the CIG-ALL mailing list