8
$\begingroup$

They look like the same thing, but their definitions are confusing. Additionally, while it's possible to use 'tolerance'< 0.05 in AIMS easily, in ASE, some structures never optimize if you choose fmax < 0.05. If they are the same thing, one should use the same value for both codes for comparison of the results, right?

I put their explanations from their documentations. These definitions confuse me. In one it says per atom, and in other one, it says all individual atoms

relax_geometry (control.in) Usage: relax_geometry type tolerance Purpose: Specifies if a structure optimization (geometry relaxation) is requested, and which. type specifies the type of requested structure optimization. Default: none. tolerance: Specifies the maximum residual force component per atom (in eV/Å) below which the geometry relaxation is considered converged. - FHI-aims Manual

The convergence criterion is that the force on all individual atoms should be less than fmax: - https://wiki.fysik.dtu.dk/ase/ase/optimize.html

$\endgroup$
9
  • $\begingroup$ Are you sure that all the optimization options are same between those two? FHI-AIMS has an option of modified BFGS? If so, what happens when you initiate optimization from FHI-AIMS's optimized structure in ASE? $\endgroup$
    – mykd
    Aug 5 '20 at 20:21
  • $\begingroup$ I haven't tried your suggestion, but it sounds like a good idea. For now, my biggest concern isfmax and tolerance if they are exactly the same thing. Then I need to use same value for a reliable comparison, right? $\endgroup$
    – Alfred
    Aug 5 '20 at 20:32
  • $\begingroup$ Unfortunately, that might be difficult to answer without extensive analysis. Within these two codes, the implementation of the optimizer module might be different even if they are the theoretically same method. You might be correct that f_max=0.05 is the proper convergence value, but what if the optimizer in ASE is unable to reach this value without playing with step size or trust radius? That's why we usually do not compare optimized geometrical structures between different softwares (unless it is benchmarked), but different methods in the same software. $\endgroup$
    – mykd
    Aug 5 '20 at 20:45
  • $\begingroup$ But ASE has AIMS calculator. I am making calculations through this calculator. Because ASE has DFTD3 but AIMS not. The only difference is that it uses its own BFGS algorithm. Still can't I compare them? $\endgroup$
    – Alfred
    Aug 5 '20 at 20:53
  • 1
    $\begingroup$ In ASE documentation (wiki.fysik.dtu.dk/ase/ase/calculators/calculators.html) "For ASE, a calculator is a black box that can take atomic numbers and atomic positions from an Atoms object and calculate the energy and forces and sometimes also stresses." So this excludes the BFGS and other optimizer modules from being common between ASE and FHI-AIMS. $\endgroup$
    – mykd
    Aug 5 '20 at 21:02