Expansion coefficient

General discussion of the Cambridge quantum Monte Carlo code CASINO; how to install and setup; how to use it; what it does; applications.
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Expansion coefficient

Post by Mike Towler »

Hi Samaneh,

OK, following some discussion with Pablo and Neil : emin can get very confused (as in it gives consistently rubbish optimized wave functions) if the fixed configurations it uses are not distributed according to the initial trial wave function. Using the "empty Jastrow" in VMC appears to be the lesser evil, it doesn't fail as often. Tests can be found where the the opposite is true, but when we've tried to find the best default behaviour this has been our conclusion based on tests for a moderate variety of examples.

So if you encounter behaviour like that you observed, then in general we have said that you want to start optimizing the Jastrow factor using a single varmin/madmin cycle, and then it should be OK to continue with emin.

Neil has now made a small change to the default behaviour that will help to stop the sort of behaviour you observed happening again:

When CASINO detects an 'empty Jastrow' when using the CASINO standard Jastrow factor, it is possible to define a default value for the alpha_0 coefficient of the u term such that the Kato cusp condition is satisfied (Exercise for the student: 'work out what that default value should be!). Empirically it is found that under some circumstances this provides a better starting wave function for the optimization process; under other circumstances it is better to use only the Slater part of the wave function and to ignore the Jastrow part completely for the first config generation run.

The new default behaviour (based on various tests) is (quoting from the newly-modified manual):

"If one is studying a homogeneous system, or a 3D-periodic system or is using energy minimisation using the CASINO standard Jastrow factor, a simple default for u will be chosen that satisfies the Kato cusp conditions; otherwise, only the Slater wave function will be used for the first configuration generation run when performing wave-function optimization" (varmin, madmin, or emin).

For those people who know what the new essentially-unadvertised CASINO general Jastrow factor is (gjastrow in parameters.casl), something like this was already the default behaviour.

So this change is now in the current 2.13.94 version available from the website. Let me know if that works for you.

Best wishes,
Mike
Post Reply