Page 1 of 1

orbital optimization doesn't find config data

Posted: Thu Nov 09, 2017 4:55 pm
by Vladimir_Konjkov
Hello CASINO developers.

I tried to calculate orbital optimized WFN starting from gaussian WFN. To do so I created following section in the correlation.data file (as described in p. 7.4.7 of manual)

Code: Select all

START MOLORBMODS
 Title
  Molecular orbital modifications for the b molecule
 START GAUSSIAN MO COEFFICIENTS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END GAUSSIAN MO COEFFICIENTS
 START GAUSSIAN EXPONENTS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END GAUSSIAN EXPONENTS
 START GAUSSIAN PRIMITIVE CORRECTIONS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END GAUSSIAN PRIMITIVE CORRECTIONS
END MOLORBMODS
I also set following variables in the input:
opt_orbitals : T
use_orbmods : T

but CASINO doesn't find START GAUSSIAN MO COEFFICIENTS in the correlation.data file.

Code: Select all

 
ERROR : READ_GWFMOLORB
 Expecting e.g. "START GAUSSIAN MO COEFFICIENTS" in correlation.data.
what am I doing wrong?

full inputs in the attachment.

Best Vladimir

PS: The error message is misleading. I just remove

Code: Select all

START GAUSSIAN PRIMITIVE CORRECTIONS
  Parameter values ; Optimizable (0=NO; 1=YES)
END GAUSSIAN PRIMITIVE CORRECTIONS
and calculations starts..... and finished with next error message:

--Job's stderr--

Code: Select all

[vladimir-Kubuntu-16:16340] *** An error occurred in MPI_Allreduce
[vladimir-Kubuntu-16:16340] *** reported by process [1949630465,1]
[vladimir-Kubuntu-16:16340] *** on communicator MPI_COMM_WORLD
[vladimir-Kubuntu-16:16340] *** MPI_ERR_TRUNCATE: message truncated
[vladimir-Kubuntu-16:16340] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[vladimir-Kubuntu-16:16340] ***    and potentially your MPI job)
PPS: Never give up! Okay, surrender today, but never again.

Re: orbital optimization doesn't find config data

Posted: Thu Nov 09, 2017 6:37 pm
by Pablo_Lopez_Rios
Hi Vladimir,

By inspection of the source, I would suggest replacing "PRIMITIVE CORRECTIONS" with "CONTRACTION COEFFICIENTS".

Best,
Pablo

Re: orbital optimization doesn't find config data

Posted: Fri Nov 10, 2017 1:59 am
by Vladimir_Konjkov
Pablo_Lopez_Rios wrote:Hi Vladimir,

By inspection of the source, I would suggest replacing "PRIMITIVE CORRECTIONS" with "CONTRACTION COEFFICIENTS".

Best,
Pablo
Hello, Pablo
Yes, following section is valid.

Code: Select all

START MOLORBMODS
 Title
  Molecular orbital modifications for the b molecule
 START GAUSSIAN MO COEFFICIENTS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END GAUSSIAN MO COEFFICIENTS
 START GAUSSIAN EXPONENTS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END GAUSSIAN EXPONENTS
 START CONTRACTION COEFFICIENTS
  Parameter values ; Optimizable (0=NO; 1=YES)
 END CONTRACTION COEFFICIENTS
END MOLORBMODS
But after first iteration I've got an issue

Code: Select all

--Job's stderr--
 
[vladimir-Kubuntu-16:10356] *** An error occurred in MPI_Allreduce
[vladimir-Kubuntu-16:10356] *** reported by process [1670774785,0]
[vladimir-Kubuntu-16:10356] *** on communicator MPI_COMM_WORLD
[vladimir-Kubuntu-16:10356] *** MPI_ERR_TRUNCATE: message truncated
[vladimir-Kubuntu-16:10356] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[vladimir-Kubuntu-16:10356] ***    and potentially your MPI job)
is optimization method EMIN valid in this case?

I think it's reasonable to optimize only MO COEFFICIENTS since GAUSSIAN EXPONENTS and CONTRACTION COEFFICIENTS are already optimized (in standard basis set).

Best, Vladimir.

Re: orbital optimization doesn't find config data

Posted: Fri Nov 10, 2017 9:49 am
by Pablo_Lopez_Rios
Hi Vladimir,

Yes, emin should be able to optimise these parameters (and any parameters you throw at it), and you can indeed delete the "GAUSSIAN EXPONENTS" and "CONTRACTION COEFFICIENTS" sub-blocks and just leave "GAUSSIAN MO COEFFICIENTS" if that is all you want to optimise. In a quick test I cannot reproduce your crash if I only optimise the MO coefficients. Your crash is a bug in CASINO related to mismatched MPI calls and needs to be fixed, but could you try to see if just optimising the MO coefficients works for you?

Best,
Pablo

Re: orbital optimization doesn't find config data

Posted: Fri Nov 10, 2017 10:21 am
by Vladimir_Konjkov
Pablo_Lopez_Rios wrote:Hi Vladimir,

Yes, emin should be able to optimise these parameters (and any parameters you throw at it), and you can indeed delete the "GAUSSIAN EXPONENTS" and "CONTRACTION COEFFICIENTS" sub-blocks and just leave "GAUSSIAN MO COEFFICIENTS" if that is all you want to optimise. In a quick test I cannot reproduce your crash if I only optimise the MO coefficients. Your crash is a bug in CASINO related to mismatched MPI calls and needs to be fixed, but could you try to see if just optimising the MO coefficients works for you?

Best,
Pablo
No, it doesn't, error message is the same with only GAUSSIAN MO COEFFICIENTS sub-block. But I've got it! When run in a single thread mode runqmc -p 1 all works without crashes.

Thanks, Vladimir.

PS. I ran Optimize orbital parameters for B atom and have got very strange result - VMC(energy) = -24.745(4) whereas exact energy = -24.65393. DMC(energy) even greater -24.786811794142 +/- 0.000799254606