Questions about fitting GLE thermostat parameters

[Bruno Duran]

Dear all,

I am trying to use a GLE thermostat in an iron and hydrogen system. I am interested in conserving the dynamical properties during the simulations but I am getting diffusion coefficient values 5-6 times higher than expected. It is my first time working with Langevin-based thermostats.

To choose a set parameters for GLE I:

1. Run a classical NVT simulation.
2. Calculate the vibrational frequency distribution of several simulations both for iron and hydrogen atoms
3. Choose an input for optimal sampling that covers the normal modes from http://gle4md.org/index.html

Now I am not sure how should I improve the model. My understanding it is that I should use a NVE ensemble or a global thermostat and repeat the process, but please correct me if I am wrong. I have several question nevertheless:

1. Should I use I-PI for the NVE simulation?
2. Why do I need to provide the temperature for a NVE ensemble in I-PI?
3. If I use one single bead  and the ‘langevin’ thermostat. Am I just running a classical molecular dynamics simulation with a classical Langevin thermostat?
4. I do not know if using TRPMD or diffusion coefficients as figure of merit would be useful. I do not know how to fit these parameters, is there any resource that I can use to get more information about this?


Thanks,

Bruno

[Michele Ceriotti]

Hi Bruno, 

many questions for one post :-) 

On Thursday, 9 November 2017 16:47:28 UTC+1, Bruno Duran wrote:

Dear all,

I am trying to use a GLE thermostat in an iron and hydrogen system. I am interested in conserving the dynamical properties during the simulations but I am getting diffusion coefficient values 5-6 times higher than expected. It is my first time working with Langevin-based thermostats.

 

So generally speaking any thermostat will alter dynamical properties, although there are reasons to think that a weak enough thermo might be a better choice than NVE -- in the latter case strictly speaking you have to launch multiple trajectories from an equilibrated NVT. However, in general, it's very hard to get faster diffusion on a thermostatted traj than on a NVE traj at the same temperature - I suspect that either your potential has intrinsically too high diffusivity, or you're doing something wrong. 

With these caveates out of the way: 

To choose a set parameters for GLE I:

1. Run a classical NVT simulation.
2. Calculate the vibrational frequency distribution of several simulations both for iron and hydrogen atoms
3. Choose an input for optimal sampling that covers the normal modes from http://gle4md.org/index.html

This is overkill - you can get a very rough idea of the vibrational frequencies from wherever you can (experimental data, a minimal v-v correlation spectrum, similar systems, etc) and pick a suitable optimal sampling thermo. 

 

Now I am not sure how should I improve the model. My understanding it is that I should use a NVE ensemble or a global thermostat and repeat the process, but please correct me if I am wrong. I have several question nevertheless:

In general yes. Note however that recently we have shown you can recover dynamical properties from a thermostatted run, you may want to have a look at that http://aip.scitation.org/doi/10.1063/1.4990536 

1. Should I use I-PI for the NVE simulation?

there is no reason not to 

2. Why do I need to provide the temperature for a NVE ensemble in I-PI?

this is because with path integrals the temperature enters the spring constant, so you have to define it even if running  RPMD

3. If I use one single bead  and the ‘langevin’ thermostat. Am I just running a classical molecular dynamics simulation with a classical Langevin thermostat?

yes 

 

4. I do not know if using TRPMD or diffusion coefficients as figure of merit would be useful. I do not know how to fit these parameters, is there any resource that I can use to get more information about this?

There are several tutorials and examples linked from the ipi-code.org page. Then obviously the paper. Then it depends on what you mean by "figure of merit"

Cheers

Michele

 

Thanks,

Bruno

[Bruno Duran]

Thanks for answer!  I will try to fit again the thermostat.

PD: With figure of merit I mean merit function.