[aspect-devel] Loading a large list of semi-unstructured point data into Aspect

Rene Gassmoeller rene.gassmoeller at mailbox.org
Mon Jul 2 09:35:33 PDT 2018


Hi Jonathan,

sorry for the late reply, we were all a bit busy and tired at the end of 
the hackathon. Let me start by asking what you exactly intend to do? 
Loading point-wise data from one model into another model of ASPECT is a 
complicated task for unstructured grids, and I wonder if there is a 
simpler way to achieve your purpose. We used Richardson Extrapolation in 
some of our models before, but did it for a global quantity (RMS 
velocity) instead of a local difference. If this would be an option it 
would make your task much simpler. Another option would be to do it as a 
postprocessing step, if you have the output of both models you could 
handle it in paraview or convert the output to text data and postprocess 
that further. This would also make reading the data unnecessary.

If both above options are not sufficient then you will indeed need to 
think about a way to read the data. As you mentioned reading 
unstructured data into a new model tends to be slow (in the best case 
using kd-trees it could be O(NlogN)), but if you can use the fact that 
the mesh structure is identical and find a way to for example serialize 
the old output and de-serialize it into the new model that might work. 
It is still significant work, so before you start anything like that, 
think carefully if there is another and simpler way.

Best,

Rene


On 06/28/2018 12:14 PM, Jonathan Robey wrote:
> I am working with a Richardson Extrapolation based convergence estimator
> for Aspect. Currently I need to find a better way to transfer large
> amounts of quadrature point data between runs, when the triangulation
> may not be consistent between the two. Do you have any advice you might
> offer?
>
> Currently, the plugin goes through the list of quadrature points and
> finds the active cell around each using GridTools to do the computation,
> but would you know of a more efficient method as, based on qualitative
> runtime costs, I think this scales non-linearly, and perhaps not even at
> log(n) per lookup.
>
> The code for the plugin is at
> https://github.com/class4kayaker/aspect/tree/Richardson_extrapolation_merge
> for only the FE portions of the plugin, and
> https://github.com/class4kayaker/aspect/tree/vof_richardson for the
> version with some specialized code for estimating VOF convergence rates.
>
> Thank you,
>
> Jonathan Robey
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

-- 
Rene Gassmoeller
https://gassmoeller.github.io/



More information about the Aspect-devel mailing list