Author Topic: ultrafast Brenner potential accuracy  (Read 1680 times)

0 Members and 1 Guest are viewing this topic.

Offline Luis M. Villamagua C.

  • Heavy QuantumATK user
  • ***
  • Posts: 27
  • Country: it
  • Reputation: 0
    • View Profile
ultrafast Brenner potential accuracy
« on: December 17, 2013, 14:56 »
Guys,

Should I trust the geometry optimization carried out by the ultrafast "brenner potential" of which quantumWise enjoys?  According to this tutorial (http://www.youtube.com/watch?v=t3xJNqP4X04) Brenner potential works very well for carbon atoms (which is my case; I have a graphene nanoribbon with only one SW defect  inserted in the middle).

I tried to carry out geometry optimization as it is shown in this tutorial (http://quantumwise.com/publications/tutorials/mini-tutorials/190-reconstruction-of-the-si-100-surface-a-geometry-optimization-study-with-atk) but only leaving one atom to move at will is taking too long (half a day).

The enormous time difference of both methods is making me hesitate about using Brenner potentials.

Thanks in advance
Luis

Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5446
  • Country: dk
  • Reputation: 89
    • View Profile
    • QuantumATK at Synopsys
Re: ultrafast Brenner potential accuracy
« Reply #1 on: December 17, 2013, 16:12 »
Brenner is pretty ok for pure carbon, but actually Tersoff is better and just a little bit slower. Both are available in ATK. But as all classical potentials, it's quite limited to a narrow range of systems. For instance, it cannot correctly predict the asymmetric dimer of the Si 100 interface, since it's an effect originating in the electronic structure which the classical potential doesn't contain. So, one must use DFT for that, or perhaps DFTB or a Slater-Koster model, although they tend to be too short-ranged as well.

Calculation times are different on different machines etc. When I ran the Si 100 system myself it took 2 hours, and that was with an older version of ATK, but on a cluster using 4 MPI processes.

PS: The calculation will not necessarily run faster just because you leave fewer atoms to move.