Open stenczelt opened 1 year ago
Does the SCF converge with 20.1 or is the error just not reported back to Python and the unconverted energy is returned?
How can I determine that when running the above snippet? Forces are returned, not sure if that's any indication.
You can increase the verbosity of the calculator or use the API provided in xtb.interface
.
Describe the bug
A single point calculation that completed on version
v20.1
cannot be evaluated withv20.2
andv22.1
. This is a specific example, a pair ofOH
separated by 4Å, seen as part of re-evaluating an earlier MD run which then worked nicely but re-evaluating it has failed. See a simplified input below with a specific example.To Reproduce
Run the following script with version
v20.1
andv20.2
/v22.1
On
v20.1
: (xtb6.3.3
)On
v20.2
&v22.1
: (xtb6.5.0
&6.5.1
)Both tested with the same ubuntu machine using Conda, see env files below, and re-tried on MacOS (Intel & M2 as well) for the latest version.
Expected behaviour
Would expect the calculation to be successful on the later versions as well.
Additional context
conda info:
xtb-20.1.yaml.txt xtb-20.2.yaml.txt xtb-22.1.yaml.txt