[mesa-users] Fwd: Re: Core resolution: non-convergent behavior increasing mesh_delta_coeff

[Matteo Cantiello]

This is great work, Mathieu.
Thanks for sharing!

-M

Matteo Cantiello  |  http://matteocantiello.com/
Postdoctoral Fellow,  Kavli Institute for Theoretical Physics 
Chief Scientist, Authorea  |  http://authorea.com/

On Aug 24, 2015, at 10:20 AM, Mathieu wrote:

> Hi everyone,
> 
> This is a followup email to describe a solution for improving the convergence of massive stars structures (computed with mass loss) differing only in the the spatial resolution (mesh_delta_coeff).
> This solution has been found thanks to the help of Frank Timmes. You can find below my original description of the problem.
> 
> To solve this issue, three steps seem to be needed:
> 
> * take smaller timesteps, by setting varcontrol_target=1d-4, dX_nuc_drop_limit = 1d-4, and dX_nuc_drop_limit_at_high_T = 1d-4 (N.B: for silicon burning you might need to loosen some of these, look at the inlists of the MESA 3 paper);
> 
> * set the maximum relative mass of cells in the model to max_dq=0.5d-4 in the inlist. This makes the baseline grid have *at least* 20000 zones;
> 
> * resolve properly the *outer* portion of the star, to make sure the amount mass lost at each timestep does not depend on the mesh_delta_coeff adopted. The way I implemented this is in the routine routine other_mesh_fcn_data, in my run_star_extras.f (see attachment). In particular using:
>     
>                             vals(k,2) = 100*weight * min(s%m(1)/Msun,max(s%m(1)/Msun-0.5, s%m(k)/Msun))
> 
> where weight=10, I impose a minimum number of cells in the outermost 0.5Msun of the star, which allows to properly resolve the mass loss rate, and consequently its effects on the location of the burning       shells, and dimensions of the various cores at oxygen depletion. This was inspired by the R_function2* entries in the controls namelist.
> 
> The attached plots show the evolutionary tracks on the (rho_c,T_c)-plane, and the abundance profiles at oxygen depletion for two models computed respectively with mesh_delta_coeff=1.0 and 0.5, with the attached inlists and run_star_extras.f. This are, so far, my best proof of the convergence of these models. 
> 
> I think this shows how important is to check the convergence when changing the resolution of MESA, and that sometimes problems in very different regions of the stars might be strictly connected: even if you just care about the core structure, you might want to be sure to resolve properly the outermost layers of the star!
> 
> If anyone has encountered similar problems, and could suggest better and/or faster (less cells or larger timesteps) ways to obtain "converged" solutions, I would be very happy to know. Any other suggestion is very welcome too.
> 
> Best regards,
> 
> Mathieu
> 
> On 07/24/2015 12:27 PM, Mathieu wrote:
>> Hi everyone,
>> 
>> I am trying to run models of massive stars (15 <= M_zams/Msun <= 30) until oxygen depletion (X(o16)<0.04 in the center), but I am finding a problem with the spatial resolution (mesh gridding) of the inner regions, especially during and after neon burning. The problem is illustrated in the attached plots.
>> 
>> I compare 30 solar mass models computed starting on the ZAMS with the mesa_45.net nuclear reaction network. All models have the same physical settings (see attached inlists and run_star_extras.f), except for my varying mesh_delta_coeff. 
>> 
>> Increasing the mesh_delta_coeff produces qualitatively and quantitatively different tracks on the (central density,central temperature)-plane, and significantly different masses of the various (He, C/O) cores (see attached plots). The  number of cells at the end of each computation shown in the attached plots is, respectively:
>>                      
>>                             mesh_delta_coeff = 1     =>  nz = 31096
>>                             mesh_delta_coeff = 0.5  =>  nz = 65763
>>                             mesh_delta_coeff = 0.2  =>  nz = 142096
>> 
>> I also ran tests without my "customized" mass loss, mesh refinements and timestep controls, and different initial masses, both with the MESA release 6794 and 7624, and I have found that the discrepancies were worse than or comparable to what I am showing here. 
>> Using my timestep controls (in the routine extras_finish_step) and mesh refinements (other_mesh_fcns_data) improves the behavior, but only up to neon burning. 
>> 
>> I have uploaded here a minimal working example with models and photos to restart the runs with mesh_delta_coeff=1 and mesh_delta_coeff=0.5, ready to be compiled and run (with the release 7624).
>> 
>> This non-convergent behavior when changing the resolution makes a detailed study of the core structure in late stages very questionable, and I would be grateful if you could help me constrain the mesh intelligently, so that the core structure does not depend so sensitively on the spatial resolution (provided that enough cells are used, of course).
>> 
>> Thanks,
>> 
>> Mathieu
>> 
>> 
> 
> 
> 
> <outer_res+max_dq_ab_M.png><outer_res+max_dq_Tc_rhoc.png><inlist.txt><inlist_massive_defaults.txt><inlist_project.txt><run_star_extras.f>------------------------------------------------------------------------------
> _______________________________________________
> mesa-users mailing list
> mesa-users at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mesa-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20150824/b587a1d9/attachment.html>