Atomic virial calculation in lammps

[pobo95]

First of all, thank you for making a good code.

I recently study MACE and just test MD simulation in lammps.

Unfortunately, I found that it doesn't support vflag_atom which for atomic virial stress calculation.

I'm not farmiliar with cpp file but

if(vflag_atom) {
torch::Tensor atomic_virial_tensor = output.at("atom_virial").toTensor().cpu();
auto atomic_virial = atomic_virial_tensor.accessor<float, 3>();
for (int ii = 0; ii < ntotal; ii++)
{
int i = ilist[ii];
vatom[i][0] += atomic_virial[i][0][0]; // xx
vatom[i][1] += atomic_virial[i][1][1]; // yy
vatom[i][2] += atomic_virial[i][2][2]; // zz
vatom[i][3] += 0.5*(atomic_virial[i][0][1]+atomic_virial[i][1][0]); // xy
vatom[i][4] += 0.5*(atomic_virial[i][0][2]+atomic_virial[i][2][0]); // xz
vatom[i][5] += 0.5*(atomic_virial[i][1][2]+atomic_virial[i][2][1]); // yz
}
}

Will this code work? I haven't tested it yet, but I'd appreciate it if you could tell me what seems to be a problem

Thanks!

[wcwitt]

Hi - sorry for the slow reply. How important is this for you? It might make things much slower.

[pobo95]

To get atomic/local shear and bulk modulus. I'm looking at the correlation between atomic virial and phonon.

[bernstei]

BTW, if the atomic virial is the strain derivative of a well-defined atomic energy, then it's also sufficient for calculating the heat flux for Green-Kubo thermal conductivity.

[pobo95]

@bernstei Yes you're right! I'm considering thermal conductivity too but what I really want to see is phase transition.

[nickhine]

The Green-Kubo thermal conductivity calculation is something we would like to be able to do in our group too, applying it to 2D materials. @pobo95 did you make any progress with this? I saw a PR from you that then got closed - was that code working?

[pobo95]

@nickhine
Yes It works. You can check it in my repository (atomic virial) and you need this code to pair with LAMMPS. But it's still in progress, so you can't use it right now, you need more tests. Please feel free to contact me by e-mail.