starting a AI MD simulation from minimised structure

[jing liang]

Hi,

I created a simulation box manually and then started a minimisation process

by using geometry optimisation. My goal is not to get a fully optimised structure

but to start a MD simulation using the NVE ensemble. My question is if one needs

a fully optimised structure in order to start MD or if there is a criterion to say

that the minimised structure is already fine? 

I am coming from classic MD simulations where some software suggest to have

the forces converged up to some particular threshold. 

Thanks

[DMITRII Drugov]

This is a good question but it depends on what you are simulating. If it is non- equilibrium system then it is pointless to look at the energy, temperature, pressure etc.

If it is equilibrium system, it should be similar to classical MD but on a shorter time scale. There is a tutorial how to calculate IR spectra with cp2k, you can find steps there for equilibration and spectrum data acquisition. 

It is on cp2k website.

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/7c1fb302-ae85-4a70-88d6-8dca9be6849en%40googlegroups.com.

[jing liang]

I am trying to see how the structure evolves in time, and monitor where some bond rotation occurs. The long story short, I

could pack all relevant molecules in a box. I was doing some search for parameters for gromacs/namd/... but I found it

not straightforward, also the mixing of force fields is something I am not completely sure about. Then, I thought that pure

CP2K simulation could work, and it is doing a great deal in the minimisation process. But I don't know how long it could

take to converge completely. 

As I wrote before, the goal of this simulation would be to see if some bond rotation occurs in the long term some spectra

would be great but at this point just the dynamics. That's why I asked here if there is some standard/accepted criteria

for using a structure as the initial frame for an AI MD simulation regarding the energies/forces. Thanks for you comments. 

[DMITRII Drugov]

Ideally you need to use force field to optimize your structure classically. I do this for liquids.

Then, you can continue NVE, T by quick step CP2K. Equilibration with AibMD is very time consuming, you will unlikely be able to equilibrate random structure with >300-500 atoms, or you could but it will be very expensive and long simulation. I couldn't equilibrate NPT of random liquid of 16*31+32*18+32 atoms with 36-48 cores for more than a month.

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/8c294cd6-a99a-482c-8a84-304cbbbee083n%40googlegroups.com.

[jing liang]

Thank you for your useful comments, I am starting to think about returning to namd/gromacs. But before

I do it, I wanted to ask you what criteria do you use to find out whether the system is equilibrated or not. 

I found this page https://www.cp2k.org/howto:md?do=  From it, I understood that by looking at the column

"Cons Qty[a.u.]" from the .ener file, one could get the difference between the maximum and minimum of the

curve and divide it by the kinetic energy, for an equilibrated system this ratio should be < 1% Maybe I didn't

get the point? 

[Niharendu Choudhury]

First of all, how big (no. of atoms) is your system?

If it is fairly small (around 500 -700 atoms with some 80 processors) then you can try to attempt AIMD simulation.

Best thing, I found, to start an AIMD simulation is to run a few pico seconds (may be 100-200 ps) classical simulation with a reasonable force field.

Then start your AIMD from the last step configuration.

Best of luck.

N Choudhury

[jing liang]

there are about 900 atoms in the simulation box

[DMITRII Drugov]

Yes, there is an energy output file. If you fix some NVE you can look at temperature variation if NVT then on energy variation. NTP is unpractical within ab initio MD, though, it depends on your resources and patience. 

Usually if you select 0.5 fs time step with H masses or 1.0 fs with D masses you can wait for first 10 ps to consider equilibration process. Again if you simulate non equilibrium process then there is no equilibration at all and you can read your trajectory from step 0.

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/19407f8b-7e89-4661-a994-295982bd8c94n%40googlegroups.com.

[jing liang]

Hi,

one question related to the previous ones. I managed to equilibrate the system by using Gromacs. Then, I obtained a xyz file

but the coordinates are not wrapped into the main cell, thus parts of the molecules look out of the box. I used this structure

as the initial one in CP2K and I see that the temperature looks stable in NVE. My question is, should the coordinates in the

xyz be wrapped or could they be expanded? Thanks.