Should the non-adiabatic calculation be much (~10x) slower?

[warrick]

Hi,

So, first, I'm not sure this is really a *bug* but it wasn't behaviour I was expecting, so here it is anyway. Out of curiosity, I decided to try computing non-adiabatic frequencies of a Sun-like star, just for change from my day-to-day adiabatic stuff, I guess.

I've found that, especially at low frequencies, I guess where the modes are mixing, the calculation is phenomenally slow and needs a large number of iterations (often near or above 200) to converge. Am I doing something wrong? I've attached my input model, which is from the MESA solar calibration routine and contains 1923 points. I've also attached the inlist that I used for both the adiabatic and the non-adiabatic calculation. For l=0,1,2 and 3, the adiabatic calculation takes about 15s; the non-adiabatic takes around 160s. I realize that the underlying problem is much harder (six equations versus four, right?), but I didn't realise it could take so much longer.

This is GYRE 2.3 compiled with the MESA SDK (8 April 2013) on four cores, Scientific Linux 6.4 (Carbon).

Thanks,
Warrick

[rhtownsend]

Hi Warrick --

Yes and no. Yes, it should be somewhat slower, as we're dealing with more dependent variables in the pulsation equations, plus we're having to do complex root finding. But it shouldn't be this much slower. I'm guessing this is related to the ongoing issues with non-adiabatic calculations and the Magnus solvers -- I still haven't quite got to the bottom of why the solvers behave rather badly.

I can't offer any timeline for a fix, but it is something that I'm actively working on.

cheers,

Rich