-
Notifications
You must be signed in to change notification settings - Fork 26
Teleconferences minutes
Taking part were:
- Christian Brinch (NBI, Denmark)
- Michiel Hogerheijde (U Leiden, Netherlands)
- Jes Jørgensen (NBI, Denmark)
- Attila Juhasz (U Cambridge, UK)
- Rolf Kuiper (U Tübingen, Germany)
- Tuomas Lunttila (Chalmers UT, Sweden)
- Sebastien Maret (U Grenoble Alpes, France)
- Sergey Parfenov (Ural Federal University, Russia)
- Reinhold Schaaf (U Bonn, Germany)
- Anika Schmiedeke (U Köln, Germany)
- Ian Stewart (U Leiden, Netherlands)
- Wouter Vlemmings (Chalmers UT, Sweden)
Thanks to SM for his hard work in curating the github project. IMS volunteered to take some of the load of reviewing PRs off SM.
Support for a systematic and generic format for storing all available grid point information seemed lukewarm at best. People clearly preferred HDF5 to FITS as a file format, and good arguments were presented in support of it. Nobody has yet volunteered to write a HDF5 interface however. The desirability of retaining the present VTK-readable output was emphasised, but it was left open whether this should continue to be implemented as a direct option in LIME or as a separate conversion from a generic format.
TL has now mostly implemented all the projected fixes to the algorithm, most recently the velocity projection fix.
IMS should generate a PR on a new user function to specify grid point number density. LIME should default in its absence to the present algorithm via the default mechanism CB originally put in place. Re the smoothing algorithm: there was general agreement that the present number (20) of iterations was excessive. Re triangulation: there was significant support for development of a bespoke triangulation routine. This could be faster, and would release LIME from reliance on qhull. This would amount to a fairly major software project however and it was clear that nobody contemplated working on it in the near term.
Not discussed.
People agreed we should leave review/merging of this til after the next release.
TL reports good progress.
AJ will raise a PR for his proposed scheme.
IMS to raise a PR with same photon directions retained for all iterations. MH informed the meeting about a related algorithm in use in RATRAN, which was felt to merit consideration for LIME as well. Various adaptive schemes for choice of photon direction were discussed. Re choice of photon frequency: TL and IMS proposed to remove the random element from this; CB maintained on the contrary that the present algorithm was optimal. IMS suggested the best way was to try and see.
People felt that there were probably advantages in doing this, but there was agreement that in view of the high work/benefits ratio that detailed consideration of this should be postponed at least until after the next release.
There was agreement on a tentative date of July 2016 for this. There was no discussion however of what should go into it. I guess proposals can be circulated later by mail. SM emphasized the need to check before release by repeating the 2002 benchmark. MH will try to locate the original data from the 2002 run.
There was no appetite for discussing this at present.