Multiwfn forum

lmangin

Member

Registered: 2020-02-19

Posts: 7

Convergence criterion and energy accuracy

Dear Pr. Lu,

I am working on a project with system/basis set that is too heavy for a proper SCF convergence, with 64G memory over 7 days. I tried it with the regular algorithm, with the qc, xqc and yqc algorithms.

In some cases, the normal algorithm has time end up with "convergence criterion not met" (after the default of 129 cycles), however, Gaussian ends up with a "Normal termination".

I noticed that the energy is quite similar and oscillating around a value, for which the difference at each iteration is small (< 0.00000001 Ha) thus doesn't seem significant for an PES of a chemical reaction with an accuracy of 0.1 kcal/mol. However, I am wondering if this electronic energy is actually meaningless since it did not fully converge, or if it could be used as a single point for the electronic single point calculations of the reaction surface.

Thanks in advance!

Last edited by lmangin (2021-03-17 23:17:58)

sobereva

Tian Lu (Multiwfn developer)

From: Beijing

Registered: 2017-09-11

Posts: 1,830

Website

Re: Convergence criterion and energy accuracy

If you are using G16 C.01, even if SCF convergence is not met, the Gaussian still normally terminates. (This is a very poor design)

The current energy can be safely used since energy variation is fully negligible.

Note that for calculating single point energy purpose, SCF=conver=6 is fully adequate, this can save cost about 1/3.

lmangin

Member

Registered: 2020-02-19

Posts: 7

Re: Convergence criterion and energy accuracy

Thanks alot for the advice!