PBE0 converges differently between 2 Qbox versions

[dingpan]

Hi Francois,

Please see the output files attached.
I tested the same input file using Qbox 1.63.8 and 1.60.4.
It seems for the PBE0 calculation, the old version of Qbox converges much better than the new one. Why?
For the PBE calculation, they are the same.
Thank you!

Best wishes,
Ding

[fgygi]

Hi Ding,

The default tolerance used to control the iterative diagonalization algorithm in the bisection calculation has been modified in release 1.60.9 (i.e. between these two versions). The value used in 1.60.4 was 1.e-6, and the value used in 1.63.8 is 1.0, which is likely too large. This can be modified using the following command:

Code: Select all

set debug BISECTION_TOL 1.e-6

The large default value (1.0) causes the bisection algorithm to stop too early in its iterations, which results in slight changes in the truncation domains of orbitals between successive iterations. This causes small fluctuations in the energy, and it prevents the calculation from reaching full convergence. Using the value 1.e-6 fixes the problem and full convergence can be reached. The default value will be decreased in the next release.

Francois