[Feature Request] Restart with higher resolution

[Jason Parisi]

Would it be possible to implement a gs2 restart feature where the resolution is increased? For example, doubling the number of poloidal or radial wavenumbers? I have heard rumors that this has already been implemented in older versions of gs2.

[David Dickinson]

I think the easiest way to achieve this is probably outside of GS2. In other words a python script or similar that can take the existing restart files and expand the dimensions as appropriate. You would then just need to point GS2 to the updated restart files. A similar tool for changing the number of processors that can use used may also be useful and is slightly related in some sense.

I think this could be quite widely useful so I’d be keen to see something like this in repository. I’ve actually already started working on adding some extra information into the restart files (minor things like the layout string) to make such sorts of operations easier but there’s probably more to be done. The main challenge I expect we would run into is for large problems trying to read all the restart files at once to concatenate and then split is likely to lead us to run out of memory so we’d need to be careful and adopt a smarter approach.

[Jason Parisi]

I think the easiest way to achieve this is probably outside of GS2. In other words a python script or similar that can take the existing restart files and expand the dimensions as appropriate. You would then just need to point GS2 to the updated restart files.

Regarding this point, would it be as simple as padding the ’new’ wavenumbers with zeros, for example?

A similar tool for changing the number of processors that can use used may also be useful and is slightly related in some sense.

Agreed, this would be useful. Is this already possible when saving the restart files as a single file rather than many different files corresponding to each processor?

I think this could be quite widely useful so I’d be keen to see something like this in repository. I’ve actually already started working on adding some extra information into the restart files (minor things like the layout string) to make such sorts of operations easier but there’s probably more to be done. The main challenge I expect we would run into is for large problems trying to read all the restart files at once to concatenate and then split is likely to lead us to run out of memory so we’d need to be careful and adopt a smarter approach.

This is a good point. In some of the larger runs, I have seen 1000s of restart files, each larger than 100mb in size. However, parallelized python might be able to handle this well, especially on the clusters. Seems like it could get very painful, however, if done inefficiently!

Best wishes,

Jason