[CP2K:3846] Re: Printing MO eigenvalues

Matt W MattWa... at gmail.com
Wed May 30 23:33:21 CEST 2012


The obvious suggestion is to switch to a hybrid functional. For your 
current system, and extensions with a few water molecules, the cost would 
be very small.

Path b is to convince Joost(I guess) that the eigenvalues of the ROKS 
calculations are meaningful... - the blurb from qs_scf_post_gpw.F which 
chucks up your error message is

!
! The issue is that one state is different from the others,
! so we certainly can't rotate all N+1 alpha states among themselves
! furthermore, alpha and beta orbitals do have the same shape,
! but not the same v_xc, therefore these seems to be no reason that
! they should have the same eigenvalues.
! So it looks like I need a convincing definition of an eigenstate in this 
case :-)

Matt

On Wednesday, May 30, 2012 12:14:17 PM UTC+1, Peter Tentscher wrote:
>
> Dear Matt, 
>
> thanks for that hint. 
> Maybe I should explain what I'm after in the end, so that The 
> someone may or may not add useful information. 
> I want to evaluate the shift in SOMO energy of an organic 
> radical cation molecule when adding water molecules around 
> it, in analogy to what Hunt&Sprik did in ChemPhysChem 
> 2005, 6, 1805 � 1808 (DOI: 10.1002/cphc.200500006) 
> for closed shell inorganic ions. 
> I need self-interaction correction because otherwise, the 
> spin density will significantly "spill out" of the organic when 
> adding water. SIC works fine with ROKS (with &OT). Whenever 
> I try to use &SIC with UKS, I get <S^2> > 1 and an integrated 
> spin density of 4 or so, and the resulting spin density is not 
> physically meaningful. (These values are close to the ideal 
> ones when not using SIC). 
> You are right that the eigenvalues may not be properly 
> defined for ROKS. I hope this is a systematic problem that 
> wouldn't matter in my case, as I'm interested in the change 
> of the eigenvalue rather than the eigenvalue itself. 
>
>  From what I've tried so far, it seems that &SIC is not 
> compatible with &DIAGONALIZATION, as SCF energies just 
> oscillate and do not converge to what I get with an equivalent 
> &OT calculation (no matter if ROKS or UKS). 
> So I'm somewhat stuck on if it's possible to get what I want. 
>
> I'll be grateful for any type of suggestion on this matter. 
>
> Best, 
>
> Peter 
>
>
> On 05/25/2012 10:46 AM, Matt W wrote: 
> > Hi, 
> > 
> > I think ROKS should work with diagonalization as well as OT. There's a 
> > short thread "ROKS error" where Matthias implies that at least high 
> > spin should be working. Whether the eigenvalues of the orbitals are 
> > really properly defined I don't know. 
> > 
> > Removing &OT and adding an &DIAGONALIZATION section I get energies and 
> > occupations printed from &MO 
> > 
> > Cheers, 
> > 
> > Matt 
> > 
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20120530/c8a58659/attachment.html>


More information about the CP2K-user mailing list