I use Discover module and give the structure compressive stress along z direction to check its structure deformation. I use NPT ensemble, time step is 1 fs, and the energy deviation is 5000 Kcal/mol, the thermolstat is chosen to be Anderson, Barostat is chosen to be Parrinello and the cell mass is 20 amu. The stress is given a minus value in the ZZ. With this setup, the simulation is totally fine.

However, when I give the structure compressive stress along x direction, a minus value in XX, using the same setup as above, the system can only run for 2700 fs, then the whole structure will get corrupted. The failure information is : the energy difference between successive steps in dynamics is greater than the user defined variable deviation, 5000.00 Kcal/mol. If I change the energy deviation into very large number, like 50000000 Kcal/mol, it still has the same problem. 

I also have tried to give the structure tensile stress along z direction, which is given a positive value in ZZ, also with the same setup situation as above. The simulation also cannot last longer time, usually will get corrupted around 3 ps with the same energy deviation problem as above.

These problems make me confused for a long time, I have tried two different methods. One is to change the time step into 0.1 fs or 0.5 fs, and the other method is to change the energy deviation into large number. Both of the two methods seem cannot solve the problems.

Hope you can help me fix the problems. Thank you very much.

Wen Xu