Printing MO eigenvalues

[Peter Tentscher]

Dear all,

I'm trying to get MO eigenvalues from a GPW calculation.
I tried using the following
&PRINT
&MO MEDIUM
EIGENVALUES
&END MO
&END PRINT
inside the &DFT section.

This does print MO eigenvalues, but they're all 0.
Is this not the way to get MO eigenvalues?
Any pointer to where I can find the appropriate information
will be warmly welcomed.

I'm also attaching a complete sample input file, in case the
problem lies somewhere else.

Best,

Peter

[Matt W]

Hi,

seems you're a victim of some not fully synchronized algorithms in the code.

"Controls the printing of the molecular orbitals.Note that this is only functional with diagonalization based methods, in particular not with OT (see MO_CUBES)"

Turn OT off. Or try pdos or MO_CUBES in the print section of DFT.

Matt

[Peter Tentscher]

Dear Matt,

thanks for that information.
Things are a bit more complicated than I thought.
Ultimately, what I want is the SOMO energy from
a ROKS calculation. That is, I do need OT in order
to converge to the right solution.
I tried using MO_CUBES in conjunction with OT/ROKS,
but what I get is
" Unclear how we define MOs in the restricted case ... skipping"
so this doesn't solve my problem.
I also tried to use the *-RESTART.wfn from an OT calculation
as a guess for the corresponding ROKS calculation without
OT. But the subsequent SCF procedure then leads away from
the former OT solution, i.e. raising the energy.

Is there any way to get orbital energies from a ROKS calculation
using OT?

Best,

Peter

[carlo antonio pignedoli]

Dear Peter,
I did not try with ROKS but this works for me.

You could also go inside the subroutine of MO_CUBES
and print the vector work

DO iset=1,nseta

ncoa = npgfa(iset)*ncoset(la_max(iset))
sgfa = first_sgfa(1,iset)

DO i=1,nsgfa(iset)
work(i,1)=eigenvector(offset+i)
! write(77,*) work(i,1)
ENDDO



in qs_collocate_density.F
SUBROUTINE calculate_wavefunction

Hope this could help.

carlo

The tight convergence criteria was set to get almost identical states
as produced by MO_CUBES
and by the MO section

> --
> You received this message because you are subscribed to the Google Groups
> "cp2k" group.
> To post to this group, send email to cp...@googlegroups.com.
> To unsubscribe from this group, send email to
> cp2k+uns...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/cp2k?hl=en.
>

[Matt W]

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 

[Peter Tentscher]

Dear Matt,

thanks for that hint.
Maybe I should explain what I'm after in the end, so that
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

[Matt W]

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 

[Bin Gu]

Hi, Peter

I just meet the same problem with you.
I need to calculate the eigenvalue of HOMO of a charged system with SIC.

What I have done is with both SCI and ADMM-hybrid methods, but both of them are based on OT.
As my system is a little big over 350 atoms. It seems PBE0 is too expensive.

Do you have any new ideas with this problem?

Thanks for any clues.


Bin GU

NUIST and QUB