Improve source handling in forward model

fsciortino / Aurora

Modern toolbox for impurity transport, neutrals and radiation modeling in magnetically-confined plasmas

https://aurora-fusion.readthedocs.io

MIT License

41 stars 24 forks source link

Improve source handling in forward model #37

Closed fsciortino closed 2 years ago

fsciortino commented 3 years ago

This PR collects works to improve the way sources are handled in the Aurora forward model of particle transport.

Main objectives:

resolve awkwardness of sources that are explicitly given as a 2D matrix in (space,time) -- these should be handled in the same way as sources for which radial and time dependencies of sources are separable.
make puffing into the divertor an independent variable -- no reason to follow the STRAHL input convention of divertor particle sources being a fraction of core sources.
allow specification of non-trivial recycling radial profile. This is an improvement wrt the assumption of recycling profiles being the same as the source radial profiles (which may also vary in time), since the latter could also represent a core impurity source (e.g. from a pellet).
allow clearer accounting of particle numbers, including neutral particles rather than just ionized particles.

This is work in progress (WIP).

tomodstrcil commented 3 years ago

Now I can run the code, but when I do it, in a few timesteps the density diverges and it is infinite. It is working for you? I have use the default setting in IMPRAD, default analytical profiles, default source (constant in time).

odstrcilt commented 3 years ago

@fsciortino still, there is some bug in recycling when I set R = 0. Any value of R > 0 works. Do you know where could be this bug?

odstrcilt commented 3 years ago

Another issue, if I need to vary a source during the inference, I will have to call the function 'setup_kin_profs_depts' or at least 'get_radial_source' in every iteration.

fsciortino commented 3 years ago

@odstrcilt I see the issue. I'm working on it. I'm going to clean up the particle conservation routine first, eliminate redundant use of xarray, and then move on from there.

fsciortino commented 3 years ago

Another issue, if I need to vary a source during the inference, I will have to call the function 'setup_kin_profs_depts' or at least 'get_radial_source' in every iteration.

@odstrcilt regarding this, I don't think that we can do much to improve this. Depending how what kind of source option you're interested in, I would recommend calling get_radial_source at every iteration, or just changing self.src_core via direct creation/interpolation of whatever source profile you have.

odstrcilt commented 3 years ago

@fsciortino I have solved the issue with varying source in the IMPRAD GUI. If the source is 1D function, it will call source_utils.get_source_time_history in each iteration and multiply it with source_rad_prof in aurora class to get src_core. And for any source (including 2D), there is a new variable imp+'_source_scale', which will just scale asim.src_core by some number.

odstrcilt commented 3 years ago

@fsciortino new issue, if the kinetics profiles are changing in time, particle conservation is not working. The recycling source will have to vary in time as well.

fsciortino commented 3 years ago

@odstrcilt OK, I updated rcl_rad_prof to be time-dependent. I also merged some things from master into this branch -- I'm kind of worried that there may be something wrong in atomic.interp_atom_prof at this point... Let me know how it looks to you.

odstrcilt commented 3 years ago

@fsciortino I have found two issues with interpolation. If the grid was too fine, it has not detected the non-equal spacing and it was crashed during extrapolation. But else it works fine, I have compared it with RectBivariateSpline and the results are nearly exactly the same.

fsciortino commented 2 years ago

Merging since we seem to have resolved all the raised issues. Glad that we got this done, @odstrcilt !