I am new user to Lammps, and i am working on the decomposition of NH3 molecule, for which I first ran the ML to get the MLIP. Now, when I use that MLIP for MD simulation in Lammps using the umbrella sampling technique also with a small timestep of 0.0001 (fs) my units=real, i consistently getting an error of lossing out one atom from the simulation box, now I reduce the timestep to 0.00005, my question is about the computational cost, because latter I want to run this MD for nanosecond/s, so is there any other way to solve this issue other than reducing the timestep. Thanks in advance
A timestep of 0.0001fs is extremely small. There must be something wrong with your potential or your system geometry or other simulation parameters. You should be able to run simulations with at least 0.1 fs if not 0.2fs for the kind of elements you have.