Discontinuous Dirac-Coulomb EOM-CC energy with continuous SCF energy.

35 views
Skip to first unread message

Xubo Wang

unread,
Apr 17, 2021, 1:41:46 AMApr 17
to dirac-users
Dear Dirac developers,
I am running a Bi2 potential energy curve with both X2Cmmf Hamiltonian and Dirac-Coulomb Hamiltonian. The curve for X2Cmmf Hamiltonian seems to be continuous while the energy from Dirac-Coulomb differs from each other by a huge amount.I am not using the full SSSS Hamiltonian which is the only potential reason that comes to my mind right now. I have attached the result from dirac-coulomb eom-cc. If you need I can also attach the results from x2cmmf eom-cc.
Another problem that I encountered out of the scope of this problem is that, when I try to do a X2C with amfi calculation, the amfi module is complaining that the atomic scf calculation reaches the max iteration, are there any keywords that can increase the max scf cycles in the atomic scf calculation?
Best regards,
Xubo

Xubo Wang

unread,
Apr 17, 2021, 1:44:04 AMApr 17
to dirac-users

Sorry that I forgot to post the output files.

Xubo

--
You received this message because you are subscribed to a topic in the Google Groups "dirac-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/dirac-users/cgynJIvDA08/unsubscribe.
To unsubscribe from this group and all its topics, send an email to dirac-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dirac-users/d93a2987-a9da-4e94-aa62-fd499cf7c202n%40googlegroups.com.
eom_ee_dc_bi_Bi2.out
eom_ee_dc_bi_Bi2_2.70.out

Peterson, Kirk

unread,
Apr 17, 2021, 11:18:44 AMApr 17
to dirac...@googlegroups.com

I'm not sure why (maybe the use of a contracted basis set?) but your SCF energy from the recomputed Fock matrix in the CCSD is very much different from what arises from the DHF calculation.

 

For the amfi question, you can increase the default charge on the atom(s) to get the amfi to converge (.AMFICH). See: http://www.diracprogram.org/doc/release-19/manual/amfi.html

 

regards,  -Kirk

--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dirac-users/fe7aea94-76d2-653a-379f-401b0e2bd92c%40colorado.edu.

Xubo Wang

unread,
Apr 17, 2021, 9:16:17 PMApr 17
to dirac-users
I just tried a calculation with uncontracted basis set, and there is the same disagreement between SCF energy from CCSD module and the converged SCF energy, the energy with uncontracted dyall-v2z set yields a very similar result as compared to contracted ones. As I was reading the 2018 paper on dirac's implementation of EOMCC, I noticed that the authors only used X2Cmmf Hamiltonian, has the EOMCC with 4c Hamiltonians also tested?
Best,
Xubo

Andre Gomes

unread,
Apr 18, 2021, 2:46:58 AMApr 18
to dirac...@googlegroups.com
Hi,

Yes, the EOM module works with the DC Hamiltonian (or the othe Hamiltonians we have available) like the rest of the CC module.

From your output, there are a number of red flags that should tell you something went wrong even before the EOM part : the recompiled orbital energies are different than the ones coming from the SCF calculation, and he CCSD iterations didn’t converge.

That said, at a first glance I’m can’t tell why this seems to be happening.

Best,

Andre



Ilias Miroslav, doc. RNDr., PhD.

unread,
Apr 18, 2021, 2:50:31 AMApr 18
to dirac...@googlegroups.com

Hi,

 

Ad: „the recompiled orbital energies are different than the ones coming from the SCF calculation

 

X2Cmmf seems to work properly with EOM-CC, need more tests.

 

Instead, try X2C Hamiltonian.

 

M.

Andre Gomes

unread,
Apr 18, 2021, 3:25:11 AMApr 18
to dirac...@googlegroups.com
Hi again,

Perhaps right now the focus should be on trying to understand what went wrong between the scf and the integral sorting steps in the CC module. Because for the recalculated orbital energies to be different from the ones in the scf, that’s where something went wrong

 Once that is sorted out, you can resume working with EOM.

Though it doesn’t look like an issue, try seeing if the disk space on the node you ran this can accommodate the files in the integral transformation. A big difference between  2 component and 4 component is the explicit handling of the small component basis in the transformation, which will take up more space.

If something was not written, or at least incorrectly, some integrals will be missing and the result will be wrong.

Best,

Andre

Stefan Knecht

unread,
Apr 18, 2021, 4:12:37 AMApr 18
to dirac...@googlegroups.com
Hi,

André is right, the failed recomputation of the orbital energies is what you need to figure out first. 
Note that with X2Cmmf neither the orbital energies are recomputed nor the SCF energy (since the two-electron part of the Hamiltonian is not picture-change transformed). 
You can (for testing) mimic this behavior in the 4c DC run with 
.NORECMP
.USEOE
in the CC section. Please check the manual where to place those keywords in the CC input. 

with best regards
Stefan

--

 

---------------------------------------------------------------------------
PD Dr. Stefan Knecht
---------------------------------------------------------------------------
GSI Helmholtzzentrum für Schwerionenforschung GmbH
Abteilung SHE Chemie
Planckstr. 1
64291 Darmstadt
Germany

room: SB3 3.163
phone: +49 6159 71 2588
web: http://stefanknecht.xyz
email: s.kn...@gsi.de
---------------------------------------------------------------------------
GSI Helmholtzzentrum für Schwerionenforschung GmbH
Planckstraße 1, 64291 Darmstadt, Germanywww.gsi.de

Commercial Register / Handelsregister: Amtsgericht Darmstadt, HRB 1528
Managing Directors / Geschäftsführung:
Professor Dr. Paolo Giubellino, Dr. Ulrich Breuer, Jörg Blaurock
Chairman of the GSI Supervisory Board / Vorsitzender des GSI-Aufsichtsrats:
Ministerialdirigent Dr. Volkmar Dietz
---------------------------------------------------------------------------
ETH Zuerich
Laboratory of Physical Chemistry
Vladimir-Prelog-Weg 2
CH-8093 Zuerich
Switzerland

---------------------------------------------------------------------------


On 18 Apr 2021, at 09:25, Andre Gomes <asp...@gmail.com> wrote:



Xubo Wang

unread,
Apr 18, 2021, 10:58:42 AMApr 18
to dirac-users
Thanks for all your patient replies.
1. For the disagreement between the recalculated orbital energy, I switched to a cluster with definately enough amount of scratch space, it still doesn't change, the discrepency remains. The calculation with the amf Hamiltonian even have this same problem.
2. I also increased the maxit in ccsd even to 1000, but the RMS is just oscillating. So I guess the previous problem has to be resolved to get proper ccsd convergency.
3. I also tried stefan's suggestion on force dirac not to recompute the fock matrix,  that also yields results quite comparable to my x2cmmf results.
So my question now would be first, what can be done to get rid of this mo energy problem, second, is  .NORECMP and .USEOE going to give physical results in my calculation? Since in the manual, it says .NORECMP "
"For closed shell systems the two expressions are identical" and Bi2 is a closed shell system. Would this assumption breaks the EOM results? In manual of .USEOE, it says "It should only change the outcome of the [T], (T) and -T energy corrections." and since I'm not quite interested in the T corrections, this keyword should also do no harm to my calculation, am I understanding it correct?
Best,
Xubo

Trond SAUE

unread,
Apr 18, 2021, 12:27:58 PMApr 18
to dirac...@googlegroups.com

HI Xubo,

I just started looking at this thread. I note the following in your input

.ACTIVE                                                                                             
energy -2.0 10.0 1.0       

What occupied orbitals do you aim to include ? How did you select the lower energy threshold ? Could it be that the the number of occupied orbitals included along the  potential curve varies ? Please check.

All the best,

   Trond

--
Trond Saue
Laboratoire de Chimie et Physique Quantiques
UMR 5626 CNRS --- Université Toulouse III-Paul Sabatier
118 route de Narbonne, F-31062 Toulouse, France

Phone : +33/561556361 Fax: +33/561556065
Mail : trond...@irsamc.ups-tlse.fr
Web : http://dirac.ups-tlse.fr/saue
DIRAC : http://www.diracprogram.org/
ESQC : http://www.esqc.org/
Book: Principles and Practices of Molecular Properties: Theory, Modeling, and Simulations

Xubo Wang

unread,
Apr 18, 2021, 1:06:05 PMApr 18
to dirac-users
I'm targeting at the 5 valence electrons and 10 inner d shell electrons resulting in 30 electrons correlated in total.
Xubo

Trond SAUE

unread,
Apr 18, 2021, 1:09:28 PMApr 18
to dirac...@googlegroups.com

HI again,

if I understand correctly is that your potential curves are discontinuous. Did you check that you have the same number of occupied (and even virtual) orbitals for each point of your potential curve ?

All the best,

   Trond

On 4/18/21 7:06 PM, Xubo Wang wrote:
I'm targeting at the 5 valence electrons and 10 inner d shell electrons resulting in 30 electrons correlated in total
--

Xubo Wang

unread,
Apr 18, 2021, 1:14:30 PMApr 18
to dirac-users

Yes, they do all contained the 30 electrons and 126 virtual spinors. Based on what's been discussed so far, the problem of discontinuous potential energy curve originate from the recomputed fock matrix gives a very different mo energies which leads to the ground state ccsd to not converge thus ruin the forthcoming eomcc calculations. With keyword .NORECMP, the energies look good. But I still have no idea why the recomputed fock matrix behaves so strange. 
Xubo

Johann Pototschnig

unread,
Apr 19, 2021, 4:20:59 AMApr 19
to dirac-users
In order to fix CC convergence you could try different threshold for the
.ACTIVE
energy -2.0 10.0 1.0
It might be that you are missing an important contribution.

First the recomputation needs to be fixed.
Which version of DIRAC are you using?

best,
Johann

Kenneth Dyall

unread,
Apr 19, 2021, 11:57:55 AMApr 19
to Dirac Users
Apart from the orbital problems, it might also be worth considering whether a single-configuration SCF is adequate. I wouldn't expect an SCF curve to dissociate properly. And due to the spin-orbit interaction, I would expect a fair amount of multireference character in this system. A spin-free and a spin-orbit calculation will show different behavior in this regard, with the latter showing bigger issues with multireference character.

--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.

Peterson, Kirk

unread,
Apr 19, 2021, 1:10:34 PMApr 19
to dirac...@googlegroups.com

Dear Xubo,

 

I went ahead and ran your job here, more or less the same input with a few differences (and I only asked for CCSD rather than EOM-CC). In particular for the 4-index transformation I used the default scheme 6 rather than 4.  My Dirac version is also built with 64-bit integers whereas I note yours was 32-bit.   As you can see from the attached, my run was just fine.

ccsd_vdz.out

Xubo Wang

unread,
Apr 19, 2021, 11:50:03 PMApr 19
to dirac-users
Thank you Kirk for trying out this calculation. I tried to recompile with 64bit integer, and everythings works just fine. The recomputed mo energies now looks pretty much the same as the mo energies from scf. So I guess the technical issue can be considered resolved.
For Kenneth's question, I'm actually doing this calculation to compare with the Bi2 curve presented in a 2018 EOM-CCSD paper by Lan Cheng et al since my curve from our in house 2c casscf results and from bagel's 4c casscf results are quite different. So I wanted to also use dirac's eom-cc module to reproduce his results and furthermore try out different Hamiltonians to see the influence of Hamiltonian on this problem. Speaking of the multireference nature caused by spin-orbit interaction, I'm also curious about how strong can the multireference effects be. Ny casscf results with 8 pair of spinors correlated and with 18 pair of spinors correlated are quite different. 
Best,
Xubo

Kenneth Dyall

unread,
Apr 23, 2021, 12:18:02 AMApr 23
to Dirac Users
I'm not too surprised at the difference between 8 and 18 Kramers pairs - correlation of the 5d is quite important in the first half of the 6p block.
Ken.

Reply all
Reply to author
Forward
0 new messages