[aspect-devel] VTK output vs. actual solution

[Marie Kajan]

Thanks! I thought that it would be easier to write a script for ParaView, but from now on I'll definitely rely on ASPECT to output data like this directly.

I still want to use ParaView for flow field visuals etc., so it was a bit scary to see 200-600 K discrepancies in the average temperature. However, it seems like increasing the mesh refinement will improve things. These new graphs are with the initial global refinement set to 5. The error indicator is highest near the surface so that could be the main issue here.

Marie

> On Sun, Mar 19, 2017 at 8:29 PM, Elbridge Gerry Puckett <egpuckett at ucdavis.edu> wrote:
> Hi Marie,
> 
> There are several 'limitations' with Paraview that you should be aware of. 
> My understanding is that the data that you (ASPECT) is putting  in the *.vtk files is only at the (square?) cell nodes.  Paraview uses this data in some way to produce visualizations and manipulate your data in other ways.
> In particular, if you are using a 'high-order' finite element spacial discretization of your problem, the order of accuracy in Paraview will necessary be reduced, probably to the equivalent of a Q_1 element.
> There are others who are more familiar with the specific details of these issues, so some of these details may not be entirely correct.  (I'm sure you'll hear comments about them here.) However, personally I would not rely on Paraview to compute graphs, etc. that require a high-order of accuracy.
> The data that is put in the Paraview *.vtk file is truncated. In other words, even though in ASPECT you are obtaining values that are, in principal, are accurate to 1e-16, which is the limit of 64bit IEEE floating point accuracy (you don't really expect 16 digits of accuracy, but something small, say 12 digits) a certain additional number of digits of accuracy are lost when those values are written to the *.vtk file.  
> This may have been changed in the past year, but I don't think this is the case.
> Anyway, from a mathematician who has spent his career in HPC and is very concerned with accuracy this is, in broad brush strokes, what I think you need to be aware of if you plan to use Paraview to postprocess data.
> 
> I'm certain you'll receive comments from others on this issue.
> 
> Good Luck!
> 
> - Gerry Puckett
> On 03/19/2017 04:17 PM, Marie Kajan wrote:
>> Hi everyone,
>> 
>> I've read on here before about ParaView having certain limitations. So how much can the VTK output actually be relied upon to represent the solution that ASPECT comes up with? I wrote a python script to compute horizontal averages in ParaView with the Slice and IntegrateVariables filters, and it works great, but there are discrepancies between the "depth           average" profiles straight from ASPECT and what ParaView shows me. I'm attaching a couple of quick graphs of temperature and viscosity to show what I mean. The viscosity drop around 660km definitely doesn't show up anywhere in ParaView, for example. This is a 3D shell type model.
>> 
>> What causes this? There's no difference with the "interpolate output" option. The global mesh refinement here is only 4, though... I'm hoping that this won't be an issue with GR>4. An adaptive mesh would be great, but I haven't           had any luck with AMR and complex 3D models.
>> 
>> Thanks in advance for any insight!
>> 
>> Marie
>> 
>> 
>> _______________________________________________
>> Aspect-devel mailing list
>> Aspect-devel at geodynamics.org
>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
> 
> 
> -- 
> 
> 
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20170319/654fc6af/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: temp-gr5.png
Type: image/png
Size: 23181 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20170319/654fc6af/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: visc-gr5.png
Type: image/png
Size: 25075 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20170319/654fc6af/attachment-0003.png>