Skip to first unread message
XIN CHEN
Jun 30, 2019, 5:46:13 PM6/30/19
to aenet
Dear Nong:
As you mentioned in the paper, the force acting on an atom i depends on the changes of the energies of all atoms j within the cutoff around atom i: F_i = -\sum_j d/dR_i E_j. I have a puzzle here, is atom i included in atoms j? I think we also need to include atom i in order to calculate the force.
Sincerely,
Xin
Nongnuch Artrith
Jun 30, 2019, 5:53:40 PM6/30/19
to XIN CHEN, aenet
Dear Xin,
Yes, that is right. Atom i is included in the index j. The change of E_i (energy of atom i) with the change of the coordinates of atom i (R_i) is a major part of the atomic force F_i.
Best,
Nong
XIN CHEN
Jun 30, 2019, 6:03:12 PM6/30/19
to aenet
Dear Nong,
That makes sense, but I just read a post in the forum saying the aenet package can also calculate how much force each neighboring atoms is acting on atom i. If the change of E_i (energy of atom i) with the change of the coordinates of atom i (R_i) is a major part of the atomic force F_i, what's the definition of this part? The force from atom i itself? Besides, we don't know the formula of atomic energy, so I think in order to calculate the force, we have to move the atom i in all three directions a little, and calculate the change of total energy to get the force. Does aenet uses this numerical solution instead of analytical solution as the force? Thank you very much.
Sincerelu,
Xin
Nongnuch Artrith
Jun 30, 2019, 6:15:40 PM6/30/19
to XIN CHEN, aenet
Dear Xin,
As I wrote before, aenet calculates the analytical derivatives, NOT a numerical derivative (i.e., no need to displace atom i).
Not sure I understood the rest of your question. The force that acts between atoms i and j consists of two parts: (1) The part coming from the change of E_i with change of R_j, and (2) the part coming from the change of E_j with change of R_i. Since ANN potentials are many-body potentials (and not pair potentials) it is better to think in terms of the change of the total energy.
Nong
XIN CHEN
Jun 30, 2019, 6:19:38 PM6/30/19
to aenet
Dear Nong,
I'm more clear now. It's unbelievable that you can get the analytical derivatives. Thank you.
Sincerely,
Xin
在 2019年6月30日星期日 UTC-5下午5:15:40，Nong Artrith写道：
在 2019年6月30日星期日 UTC-5下午5:15:40，Nong Artrith写道：
XIN CHEN
Jul 3, 2019, 2:51:22 PM7/3/19
to aenet
Dear Nong,
When you use the analytical derivatives to calculate the force, do you consider the restriction of periodic boundary condition? To be more specific, the simulation box in MD may not be very big, so the periodic atom i' may also be included in the neighborlist of atom i. The total energy of atoms in the neighborlist can be written as E(ri,ri',......,rj), due to the periodic boundary condition, ri' always equals to ri + n1a+n2b+n3c (a,b,c are three lattice vectors), so when when calculating dE/dri , we need to consider both ri and ri' because thees coordinates are related. Thank you.
Sincerely,
Xin
在 2019年6月30日星期日 UTC-5下午5:15:40，Nong Artrith写道：
在 2019年6月30日星期日 UTC-5下午5:15:40，Nong Artrith写道：
Dear Xin,
Reply all
Reply to author
Forward
0 new messages