[CIG-SHORT] Three questions about pylith

[Brad Aagaard]

Ikuo Cho wrote:
> 1)About the error measure used in benchmark tests
> What factors or rates does Pylith overestimate or underestimate the true
> values? (Also, how much do the displacement vectors calculated by Pylith
> orient to wrong directions?) Can you tell me any information of
> published papers or reports on this problem (if you have)?

The errors are very much problem dependent. Solving relevant test
problems (e.g., benchmarks) should give you an idea.


>  I found that the User Manual has a section on Benchmark tests, where an
> error measure is defined by 
> 
> error_local = sqrt{integrate ((u^t_i-u^fem_i)**2 dv}/Volume. 
> 
> In this case, however, the errors take small values at points distant
> from a fault plane even when respective displacement vectors of rigorous
> theory and those calculated by Pylith orient to the directions
> completely different from each other.

This error metric measures the absolute error, not the relative error.
Assuming   the uncertainty of observations scales with the strength of
the observation (e.g., magnitude of the displacement at the site of a
GPS receiver), the absolute error is more relevant.


> 2) To do a benchmark test
>  I tried to do numerical calculations by Pylith to answer to the above
> question by myself.  I downloaded parameter files required for an
> example of strike slip via a web site 
> http://www.geodynamics.org/svn/cig/short/3D/PyLith/benchmarks/trunk/quasistatic/strikeslipnog. 
> However, this trial failed by a fetal error. So, I hope you can tell me
> what was wrong. (I suspect the difference in the version of Pylith: I
> used Pylith of ver. 1.1.2).

The benchmarks are kept in sync with the development version of PyLith.
We have only made minor bug fixes and feature enhancements since the
release of version 1.3.0, so the benchmarks should work with PyLith 1.3.0.

Brad