nielsenmb
commented
2 weeks ago - [x] Latest plots from George
- [x] Running on BlueBear
- [x] When using the mixed model with lots of g-modes where there is only 1-2 avoided crossings, the ng=1 mode can appear either correctly or at higher frequencies, i.e., the solution is off by 1-2 orders. This won't always cause a problem, but can occasionally place a g-mode in the envelope where there is no avoided crossing. Maybe a gentle prior nudging the g-modes mode toward lower frequencies, i.e. adding a very slight skew to the comb structure due to DPi1 in the likelihood.
- [x] Extending model to higher/lower frequencies. This runs the risk of extending the asymptotic relation beyond it's limits, but the benefit is lots of nice modes the help with the surface effect correction.
- [x] Any way to add modes to the mode ID set post hoc. Either by recycling the old 'good' solution in a new fit. Fit the residual? Or include samples from the previous run in a new one somehow?