[CIG-SHORT] Initial stresses for large model

[Brad Aagaard]

Demian,

It sounds like you want an optimized spatial database for the case of 
using output as input. This is not implemented.

I think the best you can do now is a dense SimpleGridDB that minimizes 
any interpolation errors.

Regards,
Brad


On 01/26/2017 10:14 AM, Demian Gomez wrote:
> Matt,
>
> There is no projection involved in my approach. I didn't follow Brad's
> suggestion of using a topocentric coordinate system. I have a spherical
> model constructed in a geocentric CS (ECEF) and my initial stress
> spatial database in lat lon depth (i.e. cs-data = geographic and
> is-geocentric = false). Therefore, the conversion from one type of
> coordinate to the other shouldn't have any effect.
>
> Demián
>
> On Wed, Jan 25, 2017 at 9:26 PM, Matthew Knepley <knepley at rice.edu
> <mailto:knepley at rice.edu>> wrote:
>
>     On Wed, Jan 25, 2017 at 6:55 PM, Demian Gomez <demiang at gmail.com
>     <mailto:demiang at gmail.com>> wrote:
>
>         Dear all,
>
>         A few months ago I asked a question about strategies to assign
>         the initial stresses for a large model. I was originally using a
>         SimpleDB file and Brad suggested that a SimpleDbGrid provides a
>         much faster interpolation. Here's the
>         thread: http://lists.geodynamics.org/pipermail/cig-short/2016-November/002612.html
>         <http://lists.geodynamics.org/pipermail/cig-short/2016-November/002612.html>
>
>         I am now calculating the initial stresses on a first run (as
>         explained
>         here: https://wiki.geodynamics.org/_media/software:pylith:tutorials:cdm2016:pylithtutorial2016_gravity.pdf
>         <https://wiki.geodynamics.org/_media/software:pylith:tutorials:cdm2016:pylithtutorial2016_gravity.pdf>)
>         and I want to use them to compute quasi-static co-seismic
>         deformation. I've looked at generate_statedb.py in the gravity
>         examples, which uses a SimpleDB file to input the stresses into
>         Pylith. This method, however, will generate a severe slow down,
>         as discussed in the previous thread.
>
>         Is there another strategy to input the initial stresses
>         calculated by the first Pylith run? I do not want to use a
>         SimpleGridDb because this will require two interpolations: a
>         first one to generate the grid and a second one to assign the
>         stresses to each cell. I'm afraid this will lead to slightly
>         unbalanced stresses generated by the two interpolations.