Page 1 of 1

Parameter Settings during AIMD

Posted: Tue Nov 11, 2025 11:05 am
by qingyu_wang

Hello,
I plan to perform ab initio molecular dynamics (AIMD) simulations with the NPT ensemble (with machine learning enabled), but I am not fully clear about the following parameter settings, and the explanations provided in the wiki are rather ambiguous. For the simulation of the diffusion behavior of Ag in SiC that I aim to conduct, how should I configure these parameters?
Looking forward to your reply!
Best wishes!


Re: Parameter Settings during AIMD

Posted: Wed Nov 12, 2025 8:27 am
by marie-therese.huebsch

Hi Wang,

There are instructions on how to setup such a calculation on the Wiki: NPT ensemble
Did you try to follow the instructions?

Please don't hesitate to point out if a sentence is unclear. Also, feel free to ask if there is a specific parameter that you don't know how to set.


Re: Parameter Settings during AIMD

Posted: Wed Nov 12, 2025 10:10 am
by qingyu_wang

Dear marie-therese.huebsch,
First and foremost, thank you very much for your reply. I apologize for omitting the parameters I intended to inquire about. I have questions regarding the settings of the following three parameters:LANGEVIN_GAMMA = 10.0 10.0 10.0LANGEVIN_GAMMA_L = 10PMASS = 1000
I have carefully reviewed the corresponding explanations on the wiki, but I still do not know how to configure them appropriately.
Looking forward to your reply!
Best wishes!


Re: Parameter Settings during AIMD

Posted: Thu Nov 13, 2025 7:41 am
by marie-therese.huebsch

I see. Indeed we can try to add a practical guide to the Wiki in order to find the appropriate settings for LANGEVIN_GAMMA, LANGEVIN_GAMMA_L and PMASS. There are probably many ways one can arrive at reosanable settings. By reasonable seetings I mean that the instantanious temperature and instantanious pressure fluctuations are within the expected range according to the system size and stiffness and there should be no systematic drifts. There are some statistical tests you can find in literature, but I am not an expert in that area and it is rather a scientific question, I'd say.

Step 1: Start with an NVT ensemble to reduce the adjustable parameters. Thermalize the structure and plot the temperature as a function of MD steps. If you see a drift, you may need to decrease POTIM, or choose tighter settings for the ab initio calculation (ENCUT, PREC, etc). If you see basically no fluctuations, the ions are not coupling to the thermal bath. Hence increase the coupling by reducing the dampning (set a smaller LANGEVIN_GAMMA).

Step 2: Release the lattice degrees of freedom by setting ISIF=3 and choosing LANGEVIN_GAMMA_L and PMASS. The lattice vectors are now entering as degrees of freedom with there own ficticious mass (the lighter the more dynamic) and dampning (larger LANGEVIN_GAMMA_L yields less fluctuation). Usually you only adjust LANGEVIN_GAMMA_L and not PMASS, but thats a matter of taste. Again plot the pressure as a function of MD steps and find the best LANGEVIN_GAMMA_L to get no drifts and reaosanable fluctuations.

Mind that the goal is not top have no fluctuations. That goes along with bad ergodicity, i.e., you will have difficulty to visit all of the phase space.

Can you use this to setup a calculation and let me know if you obtain good MD trajectories?