Effect of thermostat rescaling constant

[Karthik Ganeshan]

Hi i-pi community, 

I was wondering what the optimal setting for \tau for the thermostat should be in TRPMD. I came across a recent paper (Hellstrom et al, JPCB 2018 112(44), 10158-10171) that uses PILE-G thermostat with \tau =100 fs. 

However, is this an adequate setting or should it be increased, perhaps to 1000 fs to mimic NVE, when gathering dynamics data? Is it a fundamental requirement of TRPMD to have a reasonably low \tau? 

Thank you! 

Sincerely,
Karthik. 

[Michele Ceriotti]

This is an excellent question. Strictly speaking, one should thermalize using imaginary time (fully thermostatted PIMD) trajectories,

and then launch simulations with no thermostat on the centroid. In practice, for a large enough system (a few 10s of atoms) the impact

of a global centroid thermostat on dynamical properties is minimal, and simplifies much the workflow. 

I have run with pile-g and a centroid tau of 1fs, finding no appreciable difference in the spectra or diffusion coefficients, so 100fs is a pretty safe setting. 

[marian...@gmail.com]

Hey Karthik,

Just to complement the answer, your conclusion (a very high tau to mimic NVE) is thus completely correct, and the careful way to do it is to pre-thermalize with PIMD (using aggressive thermostats overall). 

It is just that as Michele says, these global thermostats tend to affect dynamics little, and sometimes it is practical to use a lower tau if you are sure your system is not affected -- or at least not affected in ways that matter to you. If it was me doing the simulations, I would nevertheless check that this is fine for my system. If it is a system that has been simulated 1000s of times (e.g. water or similar), then it is ok :-)

Cheers,

Mariana

[Karthik Ganeshan]

That makes a lot of sense. Thank you so much, both of you! :D

Sincerely,
Karthik.