acetone excitations

[Zahariev, Federico [CHEM]]

Hi Lucas,

I am computing excitations of acetone by QWalk. For this purpose, I am first

computing CI (GUGA) excitations in GAMESS and then converting the chosen

excitation to the QWalk format by gamessci2qwalk as described in the Manual.

gamessci2qmc-Cyence7 -csf -state 2 -wthresh 0.00001 acetG2.log

gamess2qmc-Cyence7 -virtual 194 acetG2h

As a sanity check I've tried to reproduce the GAMESS first excited state value 

by doing a VMC calculation in QWalk without using any Jastrow factors. The 

match was far from excellent:

-35.55804151 +/- 0.0006257942545 (sigma     2.518459196 ) 

-35.6677 Hartree (GAMESS) 

As suggested in the manual, I've also tried to use a lower wthresh value (0.001)

during the gamessci2qwalk conversion combined with 

  SLATER

  OPTIMIZE_DET

  ORBITALS {

   ....

in the *slater file and also putting in the *.vmc file

method { OPTIMIZE

MINIMIZE VARIENCE

 }

or ABSOLUTE, LORENTZ, ENERGY or MIXED instead of VARIANCE.

(I've also played with the different MIXING parameters for MIXED.)

In all of the above attempts the optimization was simply "blowing up" after a 

few steps, i.e. it was suddenly jumping to completely unreasonable values. 

I wander where I am going wrong. Please, give some hints on how to acchieve

a reasonable match with the GAMESS excitations.

I am currently trying 

method { LINEAR vmc_nstep 1000 }

as well. Unfortunately LINEAR is extremely slow.

I am attaching the most relevant GAMESS and QWalk files (triple-zeta ECP 

basis of Burkatzki was used with the adjustment to the hydrogen ECP that 

you suggested in our last communication). I could send any other information

as well.

Thanks for your time!

Federico

[Lucas Wagner]

Hi Federico,

Something has changed in recent years (either GAMESS or QWalk; I'm not sure), and occasionally the sign is translated incorrectly for a weight. I still haven't been able to work out what is wrong--perhaps someone could take a stab at it. Keep in mind that if you swap the occupation of a determinant, the sign changes, so if there are some swappings for efficiency, then the weights will come out wrong.

You cannot optimize determinant coefficients using variance optimization, and as is written in the manual, OPTIMIZE does not perform well for energy minimization. LINEAR is really the best way to optimize determinant coefficients. It shouldn't be that slow for acetone; I'm not sure what you mean by it being slow.

That said, you can often get a quite good answer by simply promoting electrons in from the ground state. I would try that first before doing any complicated multi-determinant business. See the tutorial here:
https://github.com/QWalk/mainline/wiki/Excited-states

Cheers,

Lucas

--
You received this message because you are subscribed to the Google Groups "QWalk user's group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qwalk-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Zahariev, Federico [CHEM]]

Hi Lucas,

Thanks for your insightful reply!

I will try figuring out what goes wrong. From what I see so far all the coefficients in front

of the determinants are converted correctly from the GAMESS log file to the QWalk *.slater

file. Maybe the problem is in the Walk code? Has the multi-determinant feature of Walk

worked with any other ab-initio code? Where should I focus my attention?

I will certainly try with a simple singe-determinant excitation.

As for the timing, I have been running the optimization of coefficients (OPTIMIZE_DET)

of 392 CSF's on 256 processors for the last 12 hours and  LINEAR made only 6 steps so far.

Cheers,

Federico

---

From: qwalk...@googlegroups.com <qwalk...@googlegroups.com> on behalf of Lucas Wagner <lucas....@gmail.com>
Sent: Tuesday, September 8, 2015 8:38 AM
To: qwalk...@googlegroups.com
Subject: Re: acetone excitations

 

[Zahariev, Federico [CHEM]]

Hi Lucas,

I have fixed the problem with the GAMESS/QWalk excitations!

 

The reason the signs were playing tricks so far was that GAMESS is printing out 

the molecular orbitals (MOs) to the *.dat file shortly after the self-consistent 

Hartree-Fock (or DFT) cycle ends BUT it does some "normalization" of MO signs 

afterwards. As a result the MOs that are passed to a post-HF code or TDDFT are 

actually having different signs! Please note that MOs are defined up to a sign.

The fix was simply printing out a $VEC group from the code computing

the actual excitations. For my convenience I compute CI excitations through

the TDDFT code by specifying the Tamm-Dancoff approximation (TAMMD=.T. in 

the $TDDFT group) and a pure HF run (no DFT functional selected). I subsequently  

convert the TDDFT output to the QWalk format by a Python script. The advantage

of the TDDFT code in GAMESS is that it is much more efficient and could handle

excitations of relatively large systems. Obviously one could also decide to add

an actual DFT functional.

Cheers,

Federico

 

---

From: qwalk...@googlegroups.com <qwalk...@googlegroups.com> on behalf of Lucas Wagner <lucas....@gmail.com>
Sent: Tuesday, September 8, 2015 8:38 AM
To: qwalk...@googlegroups.com
Subject: Re: acetone excitations

 

[Lucas Wagner]

Hi Federico,

That's great, thank you! That sounds like a reasonable approach. Do you think that you could write up your experience in a HOWTO-type format? We can then include it in the documentation so that other people can benefit.

Thanks,

Lucas

[Zahariev, Federico [CHEM]]

Hi Lucas,

Thanks for your reply! You are quite right!

Let me, please, think it over and I will be

back to you sometime next week.

Cheers,

Federico

---

From: qwalk...@googlegroups.com <qwalk...@googlegroups.com> on behalf of Lucas Wagner <lucas....@gmail.com>

Sent: Friday, September 25, 2015 9:09 AM

[Michal Bajdich]

Hi Frederico, 

which determinant weights are you taking from Gamess? The code assumes the last or first set of vectors (dont remember anymore) of the original orbitals. So delete the additional vectors from log file and it should work. 

best,

Michal 

[Zahariev, Federico [CHEM]]

Hi Michal,

The truth is that I probably used the wrong $VEC group the CI *.dat file at

the time I was using the CI part of the GAMESS code.

Later, I switched to using the TDDFT part of the GAMESS code as it is much

more efficient. (If TAMMD=.T.  is specified in the $TDDFT group as well

as DFTTYP=NONE in the $CONTR group it exactly reproduced the CIS 

results.) 

The TDDFT code is certainly not printing out the molecular orbitals (MOs)

that are actually used in the excited-state calculation. I have inserted

a few extra lines in the TDDFT code to print out the right MOs and also 

written a Python script making the conversion from the TDDFT output 

to the QWalk *.slater format. 

Thanks,

Federico

---

From: qwalk...@googlegroups.com <qwalk...@googlegroups.com> on behalf of Michal Bajdich <baj...@gmail.com>
Sent: Friday, September 25, 2015 11:45 AM