search

deepmodeling / abacus-develop

An electronic structure package based on either plane wave basis or numerical atomic orbitals.
http://abacus.ustc.edu.cn
GNU Lesser General Public License v3.0
169 stars 129 forks source link

Convergence difficulty in calculation with VDW DFT-D2 correction #987

Closed gawcista closed 2 years ago

gawcista commented 2 years ago

As described in the title, by setting vdw_method d2, the geometry relaxation can not converge. The LARGEST GRAD would just diverge:

>grep "LARGEST GRAD" out.m
 LARGEST GRAD (eV/A)  : 4.451e-01
 LARGEST GRAD (eV/A)  : 5.810e-01
 LARGEST GRAD (eV/A)  : 5.597e-01
 LARGEST GRAD (eV/A)  : 8.053e-01
 LARGEST GRAD (eV/A)  : 1.157e+00
 LARGEST GRAD (eV/A)  : 3.025e+00
...

Structural description: Twisted bilayer SnS (728 atoms) The initial structure has been well relaxed (EDIFFG = -0.01 eV/Å) with the same input parameters with vasp.6.3.1 (PBE) The input files for ABACUS can be found here: pack.zip I guess this could result from a bug of dft-d2 method.

P.S. I noticed that someone has submitted similar issues before: https://github.com/abacusmodeling/abacus-develop/issues/134

caic99 commented 2 years ago

Fixed by https://github.com/abacusmodeling/abacus-develop/pull/151.