[CIG-SHORT] update on benchmarks

[Bill Appelbe]

Hi all:

On 27/01/2007, at 12:20 AM, Marc Spiegelman wrote:

> Hi Brad/Charles,
>    just lurking on the CIG-SHORT list but this is very  
> interesting.  There was a very similar issue in the Subduction zone  
> benchmark (e.g. http://www.geo.lsa.umich.edu/~keken/ 
> subduction02.html) which was initially ambiguous about how to  
> describe the discontinuity in  velocity across the fault.  Same  
> issue: small differences in the assumptions of how to describe  
> discontinuities in contiuum models completely dominated the  
> solutions near the wedge corner.

My 2c worth (and I admit I'm a non-expert at the outset in numerical  
methods) is that we have to be very careful in any benchmarks with  
discontinuities... sometimes these discontinuities may not be obvious  
(especially if they are in boundary conditions).   Clearly, most  
discontinuities are not amenable to continuum numerical methods, so  
some "assumptions" needs to be made, and this is often customized for  
an individual code, leading to incompatibilities...

So if I had to draw some conclusions or suggestions here (for CIG and  
collaborators), they might include:
a)	Review all benchmarks closely for any discontinuities
	implicit or explicit (e.g., in boundary or initial conditions), note  
such
	discontinuity in the benchmark description, and how it is
	going to be dealt with (ideally remove it, unless it is part
	of the physics such as a fault)
b)	For solver developer, encourage them to report/document any
	discontinuities or numerical problems they find or have issues with  
in the
	benchmarks that may cause inconsistencies across different
	implementations...


>
> In retrospect, I suppose we shouldn't be surprised, but we would  
> never catch these things without solid benchmarking efforts.  They  
> consume an enormous amount of time but it's well worth it in the end.

Agreed, except that as noted, the earlier we can catch and document  
such problems, perhaps we might reduce the benchmarking efforts.  Do  
we have a standard format for benchmarks right now (if so, there  
needs to be a section titled something like "Discontinuities/ 
Numerical Instabilities")???

	My 2c worth - Bill
>
> Keep up the good work
> cheers
> marc
>
>
>
>
>
> On Jan 25, 2007, at 11:17 PM, Brad Aagaard wrote:
>
>> Hi all:
>>
>> Charles Williams and I have been spending some time scrutinizing  
>> the results from the strike-slip benchmark. We think we have  
>> finally zeroed in on a couple of problems. The most significant  
>> issue is that the benchmark description does not specify what to  
>> do where the two taper regions overlap. The supplied Matlab  
>> scripts for computing the analytical solution uses the minimum of  
>> the two tapers (maintaining a linear taper with respect to either  
>> y or z). Charles has been using a quadratic taper (product of the  
>> two linear tapers) which is why the GeoFEST results computed in  
>> late Nov have not been agreeing with the PyLith results. In the  
>> next few days, we should have new PyLith results with much better  
>> agreement for the strike-slip results.
>>
>> Unfortunately, this taper issue also affects the reverse-slip  
>> benchmark results for BOTH GeoFEST and PyLith (and perhaps COMSOL/ 
>> Femlab as well?). Greg Lyzenga used the PyLith input files in  
>> setting up the GeoFEST runs for this benchmark, and since Charles  
>> used a quadratic taper, the taper in the simulations doesn't match  
>> the taper in the analytical solution. So we should be able to  
>> reduce the misfit with respect to the analytical solution by  
>> redoing the benchmarks with the correct taper.
>>
>> Charles is working on fixing the PyLith input files for the two  
>> benchmarks. Once we run the strike-slip benchmark and verify  
>> everything is as we expect, we will need someone to fix the  
>> GeoFEST reverse-slip input files and rerun the reverse-slip  
>> benchmark for each of the three resolutions. Any volunteers?
>>
>> We all need to thanks Charles for staring at a bunch of numbers  
>> for more hours than he would have liked in order to get to the  
>> bottom of this!
>>
>> Brad
>> _______________________________________________
>> CIG-SHORT mailing list
>> CIG-SHORT at geodynamics.org
>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>
> ----------------------------------------------
> Marc Spiegelman
> Lamont-Doherty Earth Observatory &
> Dept. of Applied Physics/Applied Mathematics
> Columbia University
> SKYPE:   Tel: +1 (845) 363 4747
> http://www.ldeo.columbia.edu/~mspieg
> ----------------------------------------------
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short

         Cheers -- Bill

------------------------------------------------------------
Bill Appelbe                           +61 3 9925 4648
CEO, VPAC           (mobile)+61 0418 557115
                                       (fax)   +61 3 9925 4647
www.vpac.org                             bill at vpac.org

Jane Douglas                       +61 3 9925 4734
Executive Assistant         jdouglas at vpac.org



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://geodynamics.org/pipermail/cig-short/attachments/20070130/aac4ea55/attachment.htm