[EXTERNAL] hi about the potential parameters of latte

Hi - Marc and Christian are the right folks to answer this,
as it's really a LATTE Q. I'm also CCing this to
the LAMMPS mail list, as I think the Q/A will be of
general interest to other LAMMPS users who want
to try out tight-binding and LATTE.

Marc/Christian - if you CC the mail list address
in your replies, the info should appear there as well.


Dear Zhigong,

Our ‘best’ set of parameters for DFTB spans C-H-N-O. I am working on developing others that I’ll release through the github site once they’re ready. We will shortly have a paper out in J. Chem. Theory Comput. that details how users can develop their own parameter sets. Stay tuned.


Thanks for all of your replies. Best wishes on this module!

Dear Zhigong,

I believe there are also parameter files for MgO. Don’t know if those are of interest to you.

And I also have another question of, I think, general interest to other users. As far as I know, while the LATTE interface is new, LAMMPS can be and in the past was already used with DFTB+ code for running TB simulations. From the user point of view, what are the main advantages of one code over the other?


16.10.2017, 20:00, “Steve Plimpton” <sjplimp@…3…>:

I’ll leave comparisons of LATTE and DFTB+ to the experts.

I’m CCing Nir Goldman (LLNL) who has coupled DFTB+ and LAMMPS

for modeling he has done. I believe what he did was add

some code to DFTB+ to call LAMMPS as a library. I’m not clear on

if that version of DFTB+ is publicly available - Nir can comment.

From a software point of view the coupling of LAMMPS to the 2

codes is the inverse of each other, though both accomplish the

same thing,

For LATTE, LAMMPS is the driver, if calls LATTE as a lib each timestep,

thru fix latte as a wrapper. It passes coords, and receives back forces.

For DFTB+, it is the driver, it calls LAMMPS as a library, and the

LAMMPS script that is run includes a fix external command, which does a callback

to DFTB+ each timestep to pass coords and receive back forces.

Nir added a new Fortran interface to the LAMMPS lib to facilitate

this - see examples/COUPLE/fortran_dftb.


Hi All,

To follow up, regarding DFTB+/LAMMPS, Steve is correct: in this case, DFTB+ is the driver. It calls LAMMPS as a library, which then “calls back” to DFTB+ each time step to pass in coordinates and lattice vectors and then receive the system forces and components of the stress tensor. I had to make some modifications to the DFTB+ code, which I would be happy to share with any interested parties.

Regarding a comparison of DFTB+ and LATTE, both are very similar in spirit as both are derived from a density functional tight binding formalism. I’ve never used LATTE, so I can’t comment on it, specifically. However, I would caution any potential user to thoroughly test either code to assess its accuracy and computational efficiency for a given system of interest. Density functional tight binding is a semi-empirical formalism, so there’s no guarantee that an off the shelf parameterization will yield usable results.

Hope that helps,