Axel, thanks a lot for your detailed reply.
As you said, compared to the berendsen barostat, the nose-hoover barostat is indeed a more robust choice, which has been mentioned in many researches.
The reason why I try to use a berendsen barostat is that it is mentioned in a paper that the velocity-rescaling thermostat and the berendsen barostat are used in the equilibrium stage, to reduce large fluctuations and quickly reach equilibrium (DOI: 10.1021/acs.jpcc.6b07780). As mentioned in the last mail, I tried using nose-hoover for equilibration, there’s no large fluctuations in the temperature while some exist in the pressure (log file named log.npt.equilibrium is atteched). In this way, I used berendsen barostat and the problems appeared.
Considering that the nose-hoover thermostat and barostat are nonergodic, some may choose to equilibrate the system using other methods, afterwards, run production phase using nose-hoover methods. I really would like to know what you think of this simulation setup?
p.s.: thanks again for your suggestions for the potential settings, which has been corrected. Before that, I thought that the lj/cut/tip4p/long can only be used for interactions with water molecules.