[mesa-users] Too much mesh points

[Mathieu Renzo]

Hi Eshan,

how much RAM memory do you use for these runs?
You can increase more the max_allowed_nz (how much depends mostly on the 
nuclear network you use, since this is the other thing that determines 
the amount of memory used). The limiting factor on the amount of memory 
you can use is in star_netwon.f90 get_netwon_work_sizes and is:

lwork=num_work_params + neq*(ndiag + 9)

which must be less than the max value of a Integer(4) -- 2^31.
In the above formula:

neq = nvar*nz !(number variables * number of cells)
ndiag = 3*nvar
nvar=num_isos+5 !(num_isos is the number of isotopes in your nuclear 
network)

So the memory usage scales as ~ 
((num_isos+5)*num_zones)*(3*num_isos+24). (thanks to Frank Timmes and 
Rob Farmer who pointed me in the right direction when I was looking for 
this).

Using ~45 isotopes I am able to run stars with up to 10^5 mesh points, 
but this of course requires a lot of patience, and you have to beware of 
spurious spikes in some gradients.

I understand that increasing max_allowed_nz is not a practical solution, 
but maybe it will allow you to do some experiments?

Cheers,
Mathieu


On 08/05/2016 12:07 PM, Ehsan Moravveji wrote:
> Dear MESA users,
>
> I would like to share an issue I’m facing (excessive number of mesh 
> points), and ask for your inputs.
>
> I’m setting up an inlist to compute models from 1.5 Msun to 35 Msun, 
> for later seismic modelling. Life goes pretty happy in the “lower” 
> end, while on the “upper” end, some of my models fail due to the 
> number of mesh exceeding the limit max_allowed_nz=20,000. This happens 
> after ~370 steps, when centre_h1~0.69.
> I have attached my inlist (v.8845) for a 35Msun model, with 
> exponential overshoot f=0.04, and min_D_mix=1d9. This is an extreme 
> case, I know; but, once this case is solved, the more relaxed cases 
> will hopefully work too.
>
> The point is: if I stick to the default mesh adjustment, the evolution 
> proceeds with ~950 mesh points, without buttlenecks. However, this is 
> too poor resolution for calling GYRE afterwards. I need around 3,000 
> mesh points across my model. For that purpose, I customise the mesh 
> distribution, and put quite some weight around the convective and 
> overshooting boundaries. That is when the problems begin, as soon as 
> the mu-gradient layers develop.
>
> My “bucket list”, here, is slim: I “wish” experimenting with high mass 
> models, including intensive mixing, and having reasonable number of 
> mesh points.
> Do you have suggestions how to reach that in one go?  ;-)
>
> Best regards,
> Ehsan.
>
>
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> mesa-users mailing list
> mesa-users at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mesa-users


-- 
Mathieu Renzo <https://staff.fnwi.uva.nl/m.renzo/>
PhD student
Anton Pannekoek Institute,
University of Amsterdam

Contacts:

Address: PO Box 94249, 1090 GE Amsterdam
Office: C4.128a
Telephone: (+31) 020-5258352
Skype name: m4thr3n
Mail: m.renzo at uva.nl <mailto:m.renzo at uva.nl>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20160805/2d38b484/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/png
Size: 122949 bytes
Desc: not available
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20160805/2d38b484/attachment.png>