difficulties in vmc optmising

General discussion of the Cambridge quantum Monte Carlo code CASINO; how to install and setup; how to use it; what it does; applications.
Post Reply
Mauro_Causa
Posts: 8
Joined: Sun Jun 23, 2013 1:41 pm

difficulties in vmc optmising

Post by Mauro_Causa »

i am not able to optimize the Jastrow factors of this simple LiI crystal.
The opt process diverges.
I have similar problems in other similar cases, while similar tests give reasonable results.
Probably i do something wrong but i am not able to understand.

All my cases gave reasonable results when i run vmc for getting HF results (vmc without Jastrow factors).

I tested madmin and varmin with the same results.
mixing AE and ECP atoms seems not influence the criticity
dropping the initial values of Jastrow parameter gives the same citicity than
using some initial guess taken by test cases.

I enclose all the files of my test case.

thanks
mauro causa'
Attachments
lii.zip
(121.37 KiB) Downloaded 979 times
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: difficulties in vmc optmising

Post by Mike Towler »

Hi Mauro,

I had a look at your input files. There are three things that I could see wrong, the first one being a calculation killer.

(1) The number of electrons defined in the input file is wrong (neu and ned) - type 'casinohelp neu' etc. to see the correct definition. neu (ned) is the total number of spin-up (spin-down) electrons referenced by the full many-body wave function, which for a periodic system means the number in the 2x2x2 simulation cell not the underlying primitive cell. You have a total number of 28 up and down spin electrons which you should have multiplied by 8 to give 224. Note the output file does actually say this:

Code: Select all

Warning : [GWFDET_SETUP] Number of up/down electrons in input file differs
 from that provided in gwfn.data. You may want to check the values of NEU and
 NED in the input file.
(2) You're almost certainly not using enough configurations (vmc_nconfig_write=5000) to do the variance minimization. It says in the manual 'we recommend using more than 10000 configurations in general..' and often it is advisable to use considerably more than that. Not using enough configs will result in badly behaved optimizations.

(3) It is usually recommended to use a value of vmc_decorr_period in the input file greater than the default value of 3 when doing optimizations, to ensure the configurations are really statistically independent (a value of 10 is stated in the manual).

So why don't you fix these things, and then let us know if you still have a problem?

Cheers,
Mike
Mauro_Causa
Posts: 8
Joined: Sun Jun 23, 2013 1:41 pm

Re: difficulties in vmc optmising

Post by Mauro_Causa »

thank you for the answer and sorry for my trivial errors.
Post Reply