[mesa-users] Weirdness w alpha(MLT) close to 1.

[Bill Paxton]

On Mar 19, 2015, at 3:48 PM, Michal Simon wrote:

> Hi Bill and Rob,
> 
> Thanks very much for looking into this.  I used a = 1.00 bcz I've been 
> brought up to believe that value is appropriate to a fully convective star, 
> as stars less massive than about 0.5 Msun are.  But I surely do't mind
> using a = 1.06.   We should however figure out why a 0.2Msun star 
> with a =1.00 goes weird at 1.42  My.... just for our peace-of-mind.
> 
> Cheers, Mike 


Hi Mike,

Here's a follow up.  Turns out your 0.2 Msun case is falling in the transition region between the OPAL eos and SCVH eos.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: plot_T_RHO_eos.pdf
Type: application/pdf
Size: 54434 bytes
Desc: not available
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20150319/8b071a89/attachment.pdf>
-------------- next part --------------

The quality of the partial derivatives suffers a little in that transition area, and we happen to hit a particularly bad spot when using alpha = 1.0.

Even in the bad spot are very close to meeting the standard convergence tolerances for the newton solver, but near isn't good enough, so the runs dies.

A bit of experimentation shows that by relaxing the correction tolerances from the defaults
      tol_correction_norm = 3d-5
      tol_max_correction = 3d-3
to 
      tol_correction_norm = 5d-5
      tol_max_correction = 5d-3
we breeze through the bad spot with no glitches.  (these changes go in the &controls section of your inlist)

The long term solution is to have a single EOS that does everything.  Don't hold your breath for that.

A more realistic alternative is to use these looser tolerances and carefully check that your results don't change significantly when you dial up increased resolution in grid and timesteps.   If you have trouble getting good results with that, you might retighten tolerances after you get past the rough spot.

Good luck!

-Bill