Assertion 'nlist_a[axis_idx[dd]] >= 0' failed.

[Yi-FanLi]

I am testing our recently developed lammps PIMD program with DeepMD-kit v1.3.3-33-g816b803 version on a 192-atom water system with 1 bead for an NVE run. However, after running for about 150 steps, the simulation crushed, showing the following error message:

lmp_mpi: /scratch/gpfs/yifanl/Softwares/deepmd-kit/source/op/descrpt.cc:451: void DescrptOp<Device, FPTYPE>::make_axis(std::vector&, std::vector&, c onst int&, const std::vector&, int, const std::vector&, const std::vector&, const std::vector&, const SimulationRegion&, bool) const [with Device = Eigen::ThreadPoolDevice; FPTYPE = double]: Assertion `nlist_a[axis_idx[dd]] >= 0' failed.

The same error does not error when using "fix nve", which should be equivalent to the 1-bead NVE pimd run. Thus, there must be something wrong with the currently developing PIMD program.

My question is, what will trigger this error message in DeepMD-kit? What does the assertion 'nlist_a[axis_idx[dd]] >= 0' require?

[njzjz]

Could you please provide the model and trajectory of the last step?

[Yi-FanLi]

I was using the wrong mode. Now I froze a correct, smooth-version model and it works. Thanks!

[amcadmus]

You are using the non-smooth version of Deep Potential. It does not make sense to use it with NVE, because it does not conserve the total energy.

[Yi-FanLi]

I am using type="se_a" for the descriptor in the json script. So I think I am using the smooth version.

[Yi-FanLi]

I was using the wrong mode. Now I froze a correct, smooth-version model and it works. Thanks!