[molpro-user] accu=n bug in DFT

[Hans-Joachim Werner]

Dear Matthias,
that accu=16 was not correctly working in geometry optimizations was indeed a bug which has now been fixed. It should appear in the next patch. For the time being please use the accu directive rather than the option (option accu=1.d-16 should also work).

Properties such as dipole moments and 2-electron energies depend in first order on errors of the MO coefficients, while the variational energy only in second order. Therefore it might indeed be necessary to tighten the threshold if you want very accurate properties. The default value of accu depends on the energy threshold, and is also automatically tightened in geometry optimizations or frequency calculations to ensure smooth convergence. 

Best wishes
Joachim
---
Prof. Dr. Hans-Joachim Werner
Institut für Theoretische Chemie
Universität Stuttgart
Pfaffenwaldring 55
70569 Stuttgart, Germany
e-mail: werner at theochem.uni-stuttgart.de




> Am 22.11.2017 um 03:54 schrieb Matthias Heger <heger at ualberta.ca>:
> 
> I just noticed a strange input bug in my DFT calculations. According to the manual, "accu=n" and "accu=1e-n" should equivalent for n>1. This works well in single point calculations, where
> {df-ks,(func),accu=14}
> produces the same output as
> {df-ks,(func),accu=1e-14}
> However, the first format does not seem to do anything if the ks program is followed by forces or optg; then "accu=n" is simply ignored. Only the second format or "accu,14" on a new line works.
> 
> I came across this when I noticed that the 2-electron energies and dipole moments at the end of my DFT iterations were not converged. I am now using "accu=1e-16" and "dftthresh,total=1e-8" and that seems to be a bit better. Are there any other settings I could use in this case?
> 
> Best,
> Matthias
> 
> 
> -- 
> Dr. Matthias Heger
> Department of Chemistry, University of Alberta
> 11227 Saskatchewan Drive NW
> Edmonton, AB T6G 2G2, Canada
> 
> Office: CCIS 4-151
> Lab: CCIS 4-108
> Email: heger at ualberta.ca
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user