[mesa-users] numerical noise
Bill Paxton
paxton at kitp.ucsb.edu
Thu Jun 6 12:49:28 EDT 2013
On Jun 6, 2013, at 6:14 AM, chenhl wrote:
> Dear all,
>
> I am using MESA for binary evolution. But I found a lot of numerical noise in the calculation. For example, the mass transfer rate curve. Is there any way to reduce the noise and make the curve smooth?
Hi,
I think it might be a good exercise for you to study the details of how the mdot is set during RLO.
It is a tricky game! For the "explicit" version, mdot is set as an exponential function of the relative overflow
at the start of the step --- exp((R-RL)/Hp), R=radius of star, RL=roche radius, Hp = scale height.
So if the R has jumped a little, mdot will jump a lot! The "implicit" version of the binary code
iterates to find an mdot such that at the end of the step it will have produced a new R that is very
close to RL. It tries different mdots until it finds one that gives the desired match. But don't
expect this to give a smooth evolution for mdot! Try it and see. And spend some "quality time"
reading the routine rlo_check_model and the things it calls in star/binary_systems/rlo_routines.
You might try changing some of the control parameters that I explained yesterday
(email attached below). E.g., a smaller value of fr should cause the code to use smaller timesteps
during stages of evolution when the relative overflow is changing rapidly.
Good luck!
Bill
Begin forwarded message:
> From: Bill Paxton <paxton at kitp.ucsb.edu>
> Date: June 5, 2013 1:27:30 PM PDT
> To: Jaime Villaseñor <jaime.villasenor at postgrado.uv.cl>
> Subject: Re: [mesa-users] Evolving a pre-CV
>
>
> On Jun 5, 2013, at 1:10 PM, Jaime Villaseñor wrote:
>
>> Ok, great! Everything is working now as I wanted with the rlo test.
>>
>> Just two more questions: Where can I find the angular momentum loss equations that mesa uses?
>
> The code lives in star/binary_systems/rlo_routines.inc
>
> Our implementation was originally done with help from Saul Rappaport.
> So you won't be surprised to learn that we use the "classic" scheme from
> rappaport, verbunt, and joss. apj, 275, 713-731. 1983.
>
> You can find that plus the GR term and the term for mass ejected from the system
> in the function "get_jdot". There is even a term for tidal Jdot.
>
>
>> and, what is the fr parameter in the 'inlist_rlo' file?
>
> There are 4 cryptically named parameters that are used in setting timesteps.
> They are specifying target per-timestep fractional changes for various things.
>
> fm -- fractional change in mass of envelope
> fa -- fractional change in binary separation
> fr -- fractional change in relative gap between R and RL, (RL-R)/R
> fj -- fractional change in angular momentum
>
>
> here are the default values:
> fm = 0.01d0
> fa = 0.01d0
> fr = 0.05d0
> fj = 0.01d0
>
>
> cheers,
> Bill
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20130606/e9940b39/attachment.html>
More information about the Mesa-users
mailing list