Brainstorming about limiting jumps in TJLF solutions

[orso82]

1. Hold the widths constant

   • Typically TJLF takes a width guess as input, and then runs the max2 function to find the best width
   • TJLF also has a flag to force use of the width input by the user
   • Can we try to make a input scan, where we let TJLF find the best width once, and then keep the same value for other runs in the scan?

2. We could try to introduce damping to the eigenvalue solver by adding a constant term that increases the diagonal dominance of the matrix. The values on the diagonal could be a small fraction of the values on the diagonal of the original matrix, ensuring that the damping is proportional to the scale of the system's terms.

[tomneiser]

The saturation rule and spectral shift are very sensitive to growth rate spectra. Branch jumps are very visible in these spectra, and I also see gaps in the spectra whenever the eigenvalue solver failed to find an actually unstable eigenvalue at a certain ky. So to mitigate the effect of these discontinuities in the spectra, we could make the spectra smooth with interpolation. At first maybe just start with smoothing the output (e.g. electron heat flux spectrum before integration over ky), and then try smoothing the growth rate spectra before calling the spectral shift and saturation rule routines