Temperature runaway during MD using npt_mttk

[Dankomaister]

Hi,

We are getting this temperature runaway during MD simulation of compression of C60 crystal along z-dir using GPUMD v3.9.5.
First noticed the problem with our own carbon NEP, but can also replicate it with 2024_Fan_C_GAP2020 NEP and Tersoff.
Seemingly works fine for smaller systems, but not with bigger ones (16 k atoms).

This is the GPUMD run file, attached is also the model.xyz

potential C_Tersoff_1989.txt
velocity 1000

ensemble npt_mttk temp 1000 1000 z 0 160
dump_exyz 1000 0 0
dump_thermo 1000
run 1000000

Target temperature is 1000 K but during the simulation temperature increases up to 20000 K, see attracted thermo.out.
Is this expected behavior or a bug ?

npt_mttk.zip

/Daniel

[psn417]

Hi, can you test it with lammps? I'm not sure if it is the problem of npt_mttk.

You can compare it with fix npt command of lammps.

[brucefan1983]

I guess it is due to too fast compression. So you roughly compress C60 by 100% in the $z$ direction during 1 ns. The Nose-Hoover chain thermostat in MTTK, with the default temperature coupling of $\tau_T=100$ timestep is not able to response so quickly. By reducing $\tau_T$ to 10 timestep,

ensemble npt_mttk temp 1000 1000 tperiod 10 z 0 160

I got more stable temperatures (see the Figure below), but I would suggest enlarge the simulation time too to reduce the rate of compression.