[Mesa-users] About noises in the mass transfer rate curve

[Pablo Marchant]

Hi Xiang,

forwarding to mesa-users as I forgot to include them in one of the previous
replies.

If you are satisfied with your new setup, please share your inlists, so
they can serve as reference for future users going through the mailing list.

Cheers

On Wed, Aug 23, 2017 at 9:59 PM, 贺祥 <hexiang at ynao.ac.cn> wrote:

> Hi Marchant,
>     With the help of the suggestion you support, I try to adjust some
> parameters of implicit scheme, such as min_change_factor ,max_change_factor
> and so on.
> When I adjust the min_change_factor and max_change_factor,   I found the
> noise decrease substantially as the same as you. See the attached figures
> (1.pdf /2.pdf/1chang.pdf/2change.pdf).
> A numerical problem can be drawn from the figures. Well thanks for your
> help.
> Cheers,
> Xiang He
>
> -----原始邮件-----
> *发件人:* "Pablo Marchant" <pamarca at gmail.com>
> *发送时间:* 2017年8月23日 星期三
> *收件人:* "贺祥" <hexiang at ynao.ac.cn>
> *抄送:*
> *主题:* Re: About noises in the mass transfer rate curve
>
>
> Here you have an example that provides a slight improvement. Did not run
> the entire thing, but it's enough to see mass transfer rates come out with
> less noise, and that the general evolution remains almost the same. As you
> can see in the attached mass_transfer.pdf plot, where your inlists
> correspond to the "lower res" line and my run with higher resolution is
> labeled as "higher res", the noise is completely removed in the earlier
> phases, but it does not work that well through the entire run.
>
> In particular, I increased resolution using
>
> mesh_delta_coeff = 0.8
> max_dq = 0.001
>
> in inlist1, and in inlist_project I used
>
> implicit_scheme_tolerance = 5d-5
> min_change_factor = 1.01
> max_change_factor = 1.2
>
> which reduces the tolerance of the mass transfer solver, and also reduces
> the iteration-to-iteration changes in mass transfer rates that the solver
> takes. If you want you can keep experimenting with these options, and it's
> likely reducing the timesteps using varcontrol_target will help as well.
>
> Attached are the inlists I used, let us know if you find a set of options
> that produce better results! However, as I was saying previously, the
> general results are more or less consistent, you will get very similar
> final periods and masses despite the noise. So it's a "choice your poison
> situation". Do you want the mass transfer rate to be noise-free and take a
> hit in performance from using a much higher spatial and temporal
> resolution? Is that the case even if the final results are more or less the
> same?
>
> FYI, you should do a detailed convergence test on your own. Do not simply
> trust my words that the final results are very similar despite increasing
> the resolution.
>
> cheers
>
> On Tue, Aug 22, 2017 at 11:40 AM, Pablo Marchant <pamarca at gmail.com>
> wrote:
>
>> Hi Xiang, please include mesa-users in these emails.
>>
>> I have to say that I have no simple solution for this issue, at the late
>> stages of mass transfer the radius is very sensitive to the mass transfer
>> rate, and can result in these numerical issues.  You can see that the
>> change in radius when you have that noise in mass transfer rate is
>> associated to noise in the stellar radius of less than 0.1%. You could try
>> reducing the value of implicit_scheme_tolerance, though at 1d-4 I'm still
>> seeing plenty of noise.
>>
>> Despite this, you can see there is a consistent behavior in the mass
>> transfer rate, and in particular many people doing calculations very
>> similar to this average the mass transfer rate through multiple steps in
>> order to plot their results.
>>
>> The main thing you need to be concerned with is convergence. As you use
>> smaller timesteps and higher resolution I'm pretty sure you'll see your
>> results are consistent, despite the presence of that noise. Unless there is
>> a fundamental reason for you to make that curve continuous, I would not
>> spent too much time on it.
>>
>> Cheers
>>
>> On Tue, Aug 22, 2017 at 1:40 AM, 贺祥 <hexiang at ynao.ac.cn> wrote:
>>
>>> Dear professor  Pablo Marchant,
>>>     Recently , I tried to use MESA to compute the evolution of binary
>>> stars.
>>> I found that the mass transfer rate curve is always not smooth in my
>>> calculation.
>>> See the attached figures (1.pdf  /2.pdf/3.pdf/4.pdf).  I have tried to
>>> vary parameters (e.g. fm, fr) in the inlist,
>>> but the noise in the curve is still there. I would think the noise is
>>> due to some numerical reason.
>>> Is there anyone who can tell me how to reduce the noise in the curve and
>>> make the curve smooth?
>>> Thanks in advance!  The inlist files are attached.
>>>
>>>
>>> Cheers,
>>> Xiang He
>>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>> --
>> Pablo Marchant Campos
>> M.Sc on Astrophysics, Universidad Católica de Chile
>> PhD on Astrophysics, Argelander-Institut für Astronomie, Universität Bonn
>>
>
>
>
> --
> Pablo Marchant Campos
> M.Sc on Astrophysics, Universidad Católica de Chile
> PhD on Astrophysics, Argelander-Institut für Astronomie, Universität Bonn
>
>
>
>
>
>


-- 
Pablo Marchant Campos
M.Sc on Astrophysics, Universidad Católica de Chile
PhD on Astrophysics, Argelander-Institut für Astronomie, Universität Bonn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20170823/702f750e/attachment.html>