Page 2 of 5

Re: molden2qmc

Posted: Fri May 01, 2015 8:50 am
by Vladimir_Konjkov
Hello, Katharina.

Your additional normalization factor (1/2, 3, 3, 3, 6) for the d-orbitals (l = 2) is realy magic. Thank you very much.

I think that the base normalisation factors for d,f,g MO orbital coefficients is product of m-dependent and m-independent
parts of normalisation constant, described in CASINO/examples/generic/gauss_dfg/README and additional magic normalization factor
only for d-orbitals. So I implement this in next version of the molden2qmc.py script.


Running short VMC task (100000 steps, CUSP CORRECTION ON) of the cubane molecule in basis dev2-QZVPP (up to g-orbitals) gave me:

ORCA FINAL SINGLE POINT ENERGY -307.503178932584

CASINO FINAL RESULT:

VMC energy (au) Standard error Correction for serial correlation

-307.508472714134 +/- 0.025454247944 No correction
-307.508472714134 +/- 0.049885753559 Correlation time method
-307.508472714132 +/- 0.052484947238 On-the-fly reblocking method

Sample variance of E_L (au^2/sim.cell) : 64.649711719656 +- 2.412110270919

Vladimir.

Re: molden2qmc

Posted: Fri May 01, 2015 9:05 am
by Mike Towler
Hi Vladimir,

As I said in my previous post, it's not an additional normalization factor, it's not magic, and it is mentioned in the gauss_dfg/README file if you look carefully. But - yes - the inconsistency it introduces is an unfortunate trap for the unwary.

Anyway, thanks for looking into this! If you have no objections perhaps we can now move towards putting the corrected version in the official CASINO distribution? We should probably coordinate with molden2qmc's author Mike Deible - I'll see if he's around.

Mike

Re: molden2qmc

Posted: Fri May 01, 2015 9:15 am
by Vladimir_Konjkov
Mike Towler wrote:Hi Vladimir,

As I said in my previous post, it's not an additional normalization factor, it's not magic, and it is mentioned in the gauss_dfg/README file if you look carefully. But - yes - the inconsistency it introduces is an unfortunate trap for the unwary.

Anyway, thanks for looking into this! If you have no objections perhaps we can now move towards putting the corrected version in the official CASINO distribution? We should probably coordinate with molden2qmc's author Mike Deible - I'll see if he's around.

Mike
Hello Mike.
I want to continue to work on this script
TODO:
1. implement TURBOMOLE, PSI4, C4.
2. implement pseudo potential
3. implement unrestricted wfn
4. verified g-orbitals normalisation
5. implement cartesian->spherical conversion

I can do it in github repository or in a repository CASINO, in recent cases, you have to give me access.
Also, since my computer is not powerful enough , I can not quite thoroughly test my script . However, in the case of issue I am ready to provide support.

Vladimir.

Re: molden2qmc

Posted: Fri May 01, 2015 9:35 am
by Mike Towler
Vladimir_Konjkov wrote: I can do it in github repository or in a repository CASINO, in recent cases, you have to give me access.
No need just for a single file (unless you really want to). When you're ready just email the script to me and I'll include it.
Vladimir_Konjkov wrote: Also, since my computer is not powerful enough , I can not quite thoroughly test my script . However, in the case of issue I am ready to provide support.
For starters, don't test it on cubane. Use the test suite in examples/generic/gauss_dfg - this is exactly what it is for.

It is annoying doing VMC-HF tests of course, since without the Jastrow factor the variance is huge and the jobs have to be run for a long time. If your computer really isn't very good you can send me your generated gwfn.data files - I'm happy to test them for you..

The test suite contains methane molecules with only d (d-ane), only f (f-ane), and only g (g-ane) basis functions. Once those are right, it's important to do all angular momenta together e.g. the avtz example, since the single-l examples only show it is correct up to an l-dependent scaling factor.

Once you have correct answers the numbers should be entered into the examples/generic/gauss_dfg/RESULTS file for permanent inclusion in the distribution, together with the gwfn.data files and i/o.

Thanks again for looking at this.

Cheers,
Mike

Re: molden2qmc

Posted: Fri May 01, 2015 8:35 pm
by Mdeible
Hi,

Thanks for taking a look at this Vladimir. I know that the original script that I wrote wasn't the prettiest piece of code, and I'm sure some updates would be helpful. If you want to send me wave functions files to run the tests I can also help out with that. If you want help checking off things on your to do list, let me know what I can help with. If there are any other codes that you think we can incorporate, such as NWCHEM or Q-Chem, and want me to generate some MOLDEN files if you don't have access to the codes, I can do that to. I think there is a smarter way to do the cartesian -> spherical conversions then I originally did it.

Mike D.

Re: molden2qmc

Posted: Sun May 03, 2015 3:19 am
by Vladimir_Konjkov
Mike Towler wrote:Hi Vladimir,

As I said in my previous post, it's not an additional normalization factor, it's not magic, and it is mentioned in the gauss_dfg/README file if you look carefully. But - yes - the inconsistency it introduces is an unfortunate trap for the unwary.

Anyway, thanks for looking into this! If you have no objections perhaps we can now move towards putting the corrected version in the official CASINO distribution? We should probably coordinate with molden2qmc's author Mike Deible - I'll see if he's around.

Mike
Realy, in the gauss_dfg/README, all described well:
One historical CASINO inconsistency which may be easily overlooked:

Constant numerical factors in the real solid harmonics e.g. the '3' in the 3xy
d function, or '15' in the (15x^3-45^xy2) f function, may be premultiplied into
the orbital coefficients so that CASINO doesn't have to e.g. multiply by 3
every time it evaluates that particular d function. In practice the CASINO
orbital evaluators do this only for d functions, but *not for f and g* (this
may or may not be changed in the future if it can be done in a.
backwards-consistent way)
I apologize for my careless reading the documentation.

Re: molden2qmc

Posted: Mon May 04, 2015 6:38 am
by Vladimir_Konjkov
Mdeible wrote:Hi,

Thanks for taking a look at this Vladimir. I know that the original script that I wrote wasn't the prettiest piece of code, and I'm sure some updates would be helpful. If you want to send me wave functions files to run the tests I can also help out with that. If you want help checking off things on your to do list, let me know what I can help with.
I have a personal license for CFOUR 2.00 beta and for many other programs too. Therefor it is not difficult to maintain CFour, just calculate the same molecule in the same basis with ORCA and CFOUR
and compare MO coefficients in a MOLDEN files, then ajust conversion parameters, but ORCA has a bit different general contraction scheme then CFOUR, i.e. if a basis set has an uncontacted
primitive function AND a contracted function with the same primitive, the primitive will be stripped off from the contraction in ORCA, but not in CFOUR.

For example cc-pVTZ basis for H.
In ORCA like https://bse.pnl.gov/bse/portal (check "Optimized General Contractions" box ON)

Code: Select all

basis "H_cc-pVTZ" SPHERICAL
H    S
     33.8700000              0.0060680
      5.0950000              0.0453080
      1.1590000              0.2028220
H    S
      0.3258000              1.0000000
H    S
      0.1027000              1.0000000
H    P
      1.4070000              1.0000000
H    P
      0.3880000              1.0000000
H    D
      1.0570000              1.0000000
end
In CFOUR like https://bse.pnl.gov/bse/portal (check "Optimized General Contractions" box OFF)

Code: Select all

s   5 1.00
   33.869999999999997        6.0679999999999996E-003
   5.0949999999999998        4.5308000000000001E-002
   1.1590000000000000        0.20282200000000000
   0.32579999999999998       0.50390299999999999    <---- this stripped in ORCA case
   0.10270000000000000       0.38342100000000001    <---- this stripped in ORCA case
s   5 1.00
   33.869999999999997        0.0000000000000000
   5.0949999999999998        0.0000000000000000
   1.1590000000000000        0.0000000000000000
   0.32579999999999998       1.0000000000000000
   0.10270000000000000       0.0000000000000000
s   5 1.00
   33.869999999999997        0.0000000000000000
   5.0949999999999998        0.0000000000000000
   1.1590000000000000        0.0000000000000000
  0.32579999999999998        0.0000000000000000
  0.10270000000000000        1.0000000000000000
p   2 1.00
   1.4070000000000000        1.0000000000000000
  0.38800000000000001        0.0000000000000000
p   2 1.00
   1.4070000000000000        0.0000000000000000
  0.38800000000000001        1.0000000000000000
d   1 1.00
   1.0569999999999999        1.0000000000000000

as result MO-coefficients is slightly different and can't be compared, Unfortunately, I can not find yet basis with f and g orbitals, identical in both programs.
So if you really want to help me, could you please provide me two MOLDEN files generated in CFOUR and ORCA for the same molecule in the same geometry and basis, considering all the above and containing f- and g-orbitals.
(in the case of d-orbitals def2-SVP basis is appropriate, and I can found conversion parameters).
Mdeible wrote: If there are any other codes that you think we can incorporate, such as NWCHEM or Q-Chem, and want me to generate some MOLDEN files if you don't have access to the codes, I can do that to. I think there is a smarter way to do the cartesian -> spherical conversions then
Mike D.
The above considerations show that the incorporation of other codes is not possible without a deep knowledge of there method of calculation. So I think that I can incorporate NWCHEM, but not Q-Chem, because I've never used them.

thanks in advance, Vladimir.

Re: molden2qmc

Posted: Mon May 04, 2015 10:56 am
by Kevin_Gasperich
...ORCA has a bit different general contraction scheme then CFOUR...I can not find yet basis with f and g orbitals, identical in both programs...
Vladimir,

It should be possible to use an explicitly-defined basis in either program.

from an ORCA reference website:
If you want to specify the basis set manually in the %basis block (here the def2-TZVP basis set is copy-pasted from the EMSL library, GAMESS-US format) :

Code: Select all

! BP86 def2-SVP TIGHTSCF printbasis
%basis
newgto C
S   6
  1  13575.3496820              0.22245814352E-03      
  2   2035.2333680              0.17232738252E-02      
  3    463.22562359             0.89255715314E-02      
  4    131.20019598             0.35727984502E-01      
  5     42.853015891            0.11076259931    
  6     15.584185766            0.24295627626    
S   2
  1      6.2067138508           0.41440263448    
  2      2.5764896527           0.23744968655    
S   1
  1      0.57696339419          1.0000000
...etc...
end
end #Don't forget the two "end" keywords!
Similar information can be found in section 6.3 of the ORCA manual.

and for CFOUR:
The use of non-standard basis sets often requires that a corresponding entry in the GENBAS file is made, as a set with the corresponding label must exit in the supplied GENBAS file (for adding a basis set to the GENBAS file, see section Basis-set file GENBAS).
Using these methods, you can explicitly input the same basis into both programs. It might be a good idea to start with the d, f, and g bases for methane found in /CASINO/examples/generic/gauss_dfg/. After you can successfully transform these bases that have a single l-value, you should test something like cc-pVQZ to verify that the relative scaling of each shell is correct.

Kevin

Re: molden2qmc

Posted: Mon May 04, 2015 11:05 am
by Vladimir_Konjkov
Kevin_Gasperich wrote:
The use of non-standard basis sets often requires that a corresponding entry in the GENBAS file is made, as a set with the corresponding label must exit in the supplied GENBAS file (for adding a basis set to the GENBAS file, see section Basis-set file GENBAS).
Using these methods, you can explicitly input the same basis into both programs. It might be a good idea to start with the d, f, and g bases for methane found in /CASINO/examples/generic/gauss_dfg/. After you can successfully transform these bases that have a single l-value, you should test something like cc-pVQZ to verify that the relative scaling of each shell is correct.

Kevin
Thank you Kevin, I just modify GENBAS definition of cc-pVTZ basis (Optimising General Contractions for it) and it works well, coefficients of s,p,d-orbitals are now similar in CFOUR and ORCA, but transformation of f-orbitals from cartesian to spherical still unclear for me.

Vladimir

Re: molden2qmc

Posted: Mon May 04, 2015 11:44 am
by Katharina Doblhoff
Your conversion for the d-orbitals is working? Could you post your transformation matrix?
Thanks,
Katharina