[molpro-user] Unclear problem in EOM-CCSD calculations: "small eigenvalue of S-eff matrix ... This can lead to convergence problems!"

[Evgeniy Gromov]

Dear Tatiana,

Many thanks for your response. While struggling with the problem I 
increased the integral accuracy (to 1dE-20) as well as
the accuracy for the T1+T2 amplitudes (to 1dE-12) . However, this did 
not help. Concerning the Davidson, the default
20 vectors per state should be OK because the problematic state is 
actually the lowest one. It is in fact a doubly degenerate (E')
state, that is there are two states of A' (the molecule has C3h 
symmetry, the computational group is Cs).
I do not think the problem relates to degeneracy, as other, higher-lying 
degenerate states converge smoothly.
The problem occurs only at the EOM-CCSD level,  at ADC(2) everything is 
fine. I also tried EOM in CFOUR and it is
the same. I am stumped with what can be wrong with that state.
I wonder what one could look at (print out) to trace the problem back?

Best regards,
Evgeniy



On 06/07/2017 05:20 PM, Tatiana Korona wrote:
> Dear Evgeniy,
>
> This message can signify near-linear dependencies in the small space 
> in the Davidson procedure. One can try to a) increase the quality of 
> the integrals (for the case when this dependence is caused by 
> numerical noise of orbitals and CCSD amplitudes - note that if CCSD is 
> not a final step, a bit more accurate T1+T2 amplitudes are often 
> required); b) try to play with Davidson options, e.g. change MAXDAV.
>
> Another issue is a high energy of the virtual orbital. I presume you 
> want to obtain some high-lying states, what often means that there is 
> a large density of states having a significant component of the 
> excitation to this orbital and then the procedure will jump from one 
> state to another and back (or to yet another state). In this case 
> sometimes it helps if you ask for more states (but only sometimes). In 
> general finding high states is a difficult no-black-box task, 
> unfortunately :-(
>
> Best wishes,
>
> Tatiana
>
> On Tue, 6 Jun 2017, Evgeniy Gromov wrote:
>
>> Hello,
>>
>> I encountered some unclear problem when doing EOM-CCSD calculations
>> for an Ir complex. All steps till EOM went fine, but in EOM I get the
>> following warning:
>>
>> The smallest eigenvalue of S-eff matrix     0.8656157915E-13
>> This can lead to convergence problems!
>>
>> and two of the five roots do not seem to have converged; there is
>> no  "Converged" message for them after 50 iterations.
>>
>> I wonder what this could be caused by? Could it be due to a very high
>> in energy (63.5 hartree) virtual orbital?
>>
>> Best regards,
>> Evgeniy
>>
>> -- 
>> Dr. Evgeniy Gromov
>> Theoretische Chemie
>> Physikalisch-Chemisches Institut
>> Im Neuenheimer Feld 229
>> D-69120 Heidelberg
>> Germany
>>
>> Telefon: +49/(0)6221/545213
>> Fax: +49/(0)6221/545221
>> E-mail: evgeniy.gromov at pci.uni-heidelberg.de
>>
>> _______________________________________________
>> Molpro-user mailing list
>> Molpro-user at molpro.net
>> http://www.molpro.net/mailman/listinfo/molpro-user
>>
>
> Dr. Tatiana Korona http://tiger.chem.uw.edu.pl/staff/tania/index.html
> Quantum Chemistry Laboratory, University of Warsaw, Pasteura 1, 
> PL-02-093 Warsaw, POLAND
>

-- 
Dr. Evgeniy Gromov
Theoretische Chemie
Physikalisch-Chemisches Institut
Im Neuenheimer Feld 229
D-69120 Heidelberg
Germany

Telefon: +49/(0)6221/545213
Fax: +49/(0)6221/545221
E-mail:evgeniy.gromov at pci.uni-heidelberg.de

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20170621/e5df9757/attachment.html>