[aspect-devel] log file

[Austermann, Jacqueline]

Hi everyone,

Yes S40RTS is in spherical harmonics and does the depth interpolation through a spline so it takes very long to read in. In fact, for each point of your mesh you iterate 21 * 41 * 42 / 2 = 18081 times to sum over the coefficients. The reason for implementing it this way was to leave the data in their original form. Any resampling can introduce a mismatch to the original data. For practical purposes however it definitely makes sense to precompute the grid and interpolate. For all global calculations I've done I've used a 1 deg x 1 deg grid and interpolated, as Thorsten said. I was actually thinking at the Hackathon that an initial condition that takes a 1 x 1 grid would be a nice addition.

Cheers,
Jacky


On Jul 10, 2015, at 2:55 PM, Wolfgang Bangerth <bangerth at tamu.edu> wrote:

> On 07/10/2015 04:49 PM, Thorsten Becker wrote:
>> As Timo said, the code is presumably loading a spherical harmonic expansion
>> via a library package to evaluate a single point, each time for each
>> integration point. Besides efficiency questions, there's no point in doing
>> that, of course, if the initial models are not parametrized as spherical
>> harmonics. I think it would be preferable to load a 3d grid once and then
>> have a simple (eg bi linear) interpolation routine spit out values at
>> desired locations. Even with a clunky importation (eg via slices of Netcdf
>> grids, read by each processor, as we do for citcom), this shouldn't take
>> more than a few tens of seconds after the I/O  for a ~1million elements.
> 
> Right. And there are deal.II classes that can already do all of that as long as you give them the x-, y-, and z-coordinates of the grid and the values.
> 
> I do think I recall that the S40RTS model is indeed described in terms of spherical harmonics as it is derived from a seismic model (where, unfortunately, using spherical harmonics seems to be prevalent -- although it makes absolutely no sense to do so). It may be worthwhile for efficiency reasons to re-sample these data sets differently.
> 
> Best
> W.
> 
> -- 
> ------------------------------------------------------------------------
> Wolfgang Bangerth               email:            bangerth at math.tamu.edu
>                                www: http://www.math.tamu.edu/~bangerth/
> 
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel