Floating point exception on a giant model

[earlbellinger]

Hello,

I ran GYRE 4.4 on some giant models produced by MESA r8118 with mesasdk 20160129 on a 1M solar-like track.

Most of the models end up fine, but some of the more evolved models failed with the following error:

Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.

Backtrace for this error:
#0 0x2AD8E03CE557
#1 0x2AD8E03CEB6E
#2 0x2AD8E12B211F
#3 0x48069A in __gyre_r_ext_MOD_exp_
#4 0x45FF9F in __gyre_r_magnus_ivp_MOD_shoot_
#5 0x49DB3F in __gyre_r_bvp_MOD_build_._omp_fn.1 at gyre_r_bvp.f90:?
#6 0x2AD8E09FA925
#7 0x2AD8E10650A3
#8 0x2AD8E1361FEC
#9 0xFFFFFFFFFFFFFFFF

I have attached an example FGONG file, the GYRE inlist that I used on it, and the MESA inlist that generated the model.

Any advice?

Best regards,
Earl Bellinger

[rhtownsend]

Hi Earl --

Apologies for not responding sooner -- been busy working on the next release of GYRE!

The floating point exception you've encountered appears to arise because you're looking for modes at very low frequencies. This isn't a problem for radial modes (as there are no g-modes when l=0), but causes numerical issues for l >= 1 because the g modes at these low frequencies have insanely large radial orders.

For instance, at the lower limit of your frequency scan, =0.01*the acoustic cutoff frequency, the radial order of l=1 g modes in your red giant model is n ~ 2,000,00!!!

I'm pretty sure that you don't want to consider such large radial orders. The fix, for the non-radial modes, is to increase freq_min to more reasonable values.

Another thing you should think about is whether your shooting grid has sufficient resolution. For high-order modes, the default shooting grid (which is just the stellar model grid) is almost certainly inadequate; you need to resample it to allow eigenfunctions to be properly resolved.

cheers,

Rich