Re: [ESPResSo-users] is it possible to turn off inertia ?

[Ulf Schiller]

Hi Edvin,

To give you the short answer first, it would probably not be too
complicated to implement your suggestion (at least for your personal
testing purposes where you may be able to forgo ensuring compatibility
and no side-effects with the rest of ESPResSo).

However, I'm not sure what you would gain. If the relaxation times of
your system are long, they are long no matter what algorithm you use
(you can't beat physics). And your suggestion does not really imply huge
savings: you would still be calculating the velocity from the forces in
every time step. You may be able to use a larger time step, but if your
system is overdamped then I don't see why a larger time step would not
be possible with the velocity Verlet as well.

Another option is to eliminate the velocity entirely and solve only for
the positions (this is essentially Brownian dynamics for implicit
solvent and I think there's a solver for that in ESPResSo, you'd just
have to replace the Oseen/Rotne-Prager/... interaction if you want to
ignore correlations).

Of course it all depends on what the particular frequencies in your
system and the perturbations are, but at the end of the day, you'd still
have that same relaxation time of your system that you have to simulate.

Hope this helps,
Ulf

On 02/05/2017 04:01 PM, Edvin Memet wrote:
> Dear all,
> 
> Some of the projects I use Espresso for involve low Reynolds
> number/overdamped motion. In such cases, I have to make sure the system
> remains quasistatic, e.g. by allowing it to sufficient time to relax
> after applying a small perturbation. This can pose a problem when the
> relaxation times are long, sometimes resulting in unfeasibly long
> simulation times.
> 
> It seems to me that a lot of the computational effort is wasted when
> using the molecular dynamics framework in the case of overdamped motion.
> That is, if you could turn off inertia, you would just have to solve
> force = friction * v and not have to wait for the system to relax (?).
> 
> Making mass very small doesn't seem to help, since that introduces
> stability issues - you have to make the time step smaller, and you don't
> really gain anything. The solution (apart from writing your own code),
> as far as I can tell, is to change Espresso's code so it solves the
> force = friction * v equations of motions instead of doing Verlet
> integration (where inertia is implicit).
> 
> I'm wondering if people familiar with the code can assess whether this
> is feasible for me to do (does it just entail changing the code at one
> or a few places) or whether it's not worth trying because I would break
> and have to change too many things.
> 
> Best,
> Edvin

-- 
Dr. Ulf D. Schiller
Assistant Professor, Department of Materials Science and Engineering
Faculty Scholar, School of Health Research
Clemson University
161 Sirrine Hall
Clemson, SC 29634

Office: 299c Sirrine Hall
Phone: 1-864-656-2669
Fax: 1-864-656-5973