Newest version outputs NaN at first time step

Issue #135 resolved
Rogerio Jorge created an issue

Up to now, I have been using GS2’s version 7 to make some linear runs for stellarator geometries. However, I am not able to compile version 6 or 7 anymore due to an issue with the 'externals' or 'utils' directories (not present anymore?) which include an apparently important file, mp.fpp. For that reason, I am trying to run GS2 linearly with version 8. However, I find that input files that ran without a problem in version 6 and 7 now just output NaN in version 8. From my understanding, GS2 reads the geometry coefficients and the problem lies in the initialization.

I am attaching two grid files: one that works and one that doesn’t. They are very similar, the only difference is that one has higher resolution than the other.

Comments (6)

  1. David Dickinson

    @Rogerio Jorge thank you for creating this issue. Versions 6 and 7 of GS2 are from the time that the code was stored at sourceforge with svn as the repository type. Version 8 came out when we moved to git at bitbucket. This transition meant that the handling of external repositories like utils changed and the use of older versions of GS2 is a bit more complicated. Originally I would recommend checking the code out from sourceforge if you wanted to use an older version, however it appears that the old code repository is not currently visible/available. If you wish use version 6/7 we should be able to help you get this up and running although I would recommend using the most recent release as this has a number of bug fixes. The most recent release is 8.0.5 and should automatically deal with utils etc.

    I’m not familiar with the use of these grid files so I’ll have to dig into this a bit, but I hope I’ll be able to find out why this is no longer working and will report back here.

  2. Rogerio Jorge reporter

    Hi @David Dickinson thank you for your feedback. I appreciate your help on this issue.

  3. David Dickinson

    I think I’ve identified what has changed between the versions which is leading to this change in behaviour. In version 8 a new pitch angle grid treatment was enabled by default. The weights of the pitch angle integration are calculated in le_grids::lgridset and in particular the weights for the “wfb” for the new Radau-Gauss grids are different. It seems that at least in the lower resolution case this wfb weight is coming out as NaN and this then leaks out into all our results. A workaround for now is to set radau_gauss_grids to .false. in the le_grids_knobs namelist.

  4. Log in to comment