Poll

computational time difference

The first approach is to use a CIF file for the primitive cell, which is triclinic and contains 114 atoms. I used the UnitCells 3 3 3 command to create a supercell with 27 primitive cells (total of 3078 atoms).
0 (0%)
The second approach used a CIF file containing a large cubic supercell containing 3648 atoms. This simulation used UnitCells 1 1 1.
0 (0%)
The second simulation took about 10x longer to run, even though the size of the simulation cell was not that different. Why is there such a big difference in the simulation times for these two systems? Is it less efficient to have a large CIF file?
0 (0%)

Total Members Voted: 0

Author Topic: Larger computational time with supercell atoms in the CIF file  (Read 601 times)

pbs13

  • Newbie
  • *
  • Posts: 2
    • View Profile
Larger computational time with supercell atoms in the CIF file
« on: January 12, 2021, 01:44:30 AM »
I am computing N2 adsorption isotherms in UiO-66 using two different methods of specifying the UiO-66 framework.

The first approach is to use a CIF file for the primitive cell, which is triclinic and contains 114 atoms. I used the UnitCells 3 3 3 command to create a supercell with 27 primitive cells (total of 3078 atoms).

The second approach used a CIF file containing a large cubic supercell containing 3648 atoms. This simulation used UnitCells 1 1 1.

The second simulation took about 10x longer to run, even though the size of the simulation cell was not that different. Why is there such a big difference in the simulation times for these two systems? Is it less efficient to have a large CIF file and use a single unit cell compared with having a small CIF file and replicate the cell?
 

Christopher

  • Newbie
  • *
  • Posts: 16
    • View Profile
Re: Larger computational time with supercell atoms in the CIF file
« Reply #1 on: January 12, 2021, 11:34:08 AM »
First of all you have to look at the number of adsorbate molecules. A monte carlo cycle in raspa is defined as min(20, N) where N is the number of adsorbate molecules. So if there are more molecules in the simulation cell than you simply have more moves per cycle.

Then you say, that computational expense of one cycle should scale approx. with N^2 because you have to evaluate the energies.

Does this help?

David Dubbeldam

  • Full Member
  • ***
  • Posts: 185
    • View Profile
Re: Larger computational time with supercell atoms in the CIF file
« Reply #2 on: January 13, 2021, 06:58:09 PM »
Also, orthorhombic is much faster than triclinic. For triclinic structure, the distance computation is a matrix multiplication to get fractional coordinates, then apply periodic boundary conditions, and then convert back to Cartesian space.