Localization of virtual orbitals

[Matteo Guglielmi]

Does `$localize` work with virtual orbitals too?

Thanks.

[uwe]

Hello,

localization of orbitals (Boys or Pipek-Mezey) can only be done for occupied ones. This simply comes from the fact that the procedure is based on something like maximisation of Mulliken charges. A proper way would be to generate PAOs (projected atomic orbitals), but that is not implemented and not needed by any Turbomole program.

A simple trick is to occupy the orbitals of interest by hand ($closed shells keyword for RHF, $alpha shells and $beta shells for UHF in the control file) and then start dscf or ridft (or whatever method you are using) with the option -proper. This will skip the SCF procedure and directly perform the property analysis which includes the newly occupied orbitals.

Do not forget to redo the changes if you want to use your input afterwards, otherwise you are dealing with a highly charged system...

Regards,

Uwe

[Matteo Guglielmi]

Thanks Uwe!

I did calculate excitation energies via the ricc2 method
on a protonated molecule (closed shell system, RHF,
total charge = +1)

So what you suggest me is to take the control file after
the final ricc2 run and:

- modify the '$closed shells' keyword i.e. I have to increase
the number of doubly occupied orbitals
- run the 'ricc2 -proper' command
- run the 'tm2molden' command

Is that correct?

[uwe]

Hi,

I wonder what you intend to do. Excited states, localization and then usage of tm2molden sounds like you want to visualize the excitation, i.e. from which ground state orbital to which orbital the excitation happens.

First, there are no excited state orbitals, just an excited state density. You could visualize the difference density (excited state density minus the ground state density), and you will see where the electron comes from and where it goes to. There is a short chapter about that in the documentation, see chapter 7.3.3 'Visualization of densities'.

Another possibility is to look at the output file of ricc2. There you will find a list of orbitals and the percentage of their contribution to the excitation. If you just visualize the ground state canonical orbitals (with dscf -proper and $pointval mo <number>-<number>, see chapter 10.2 of the documentation), not the localized ones, you will also get a picture of the excitation.

Regards,

Uwe

[Matteo Guglielmi]

Hi,

I wonder what you intend to do. Excited states, localization and then usage of tm2molden sounds like you want to visualize the excitation, i.e. from which ground state orbital to which orbital the excitation happens.

First, there are no excited state orbitals, just an excited state density. You could visualize the difference density (excited state density minus the ground state density), and you will see where the electron comes from and where it goes to. There is a short chapter about that in the documentation, see chapter 7.3.3 'Visualization of densities'.

Another possibility is to look at the output file of ricc2. There you will find a list of orbitals and the percentage of their contribution to the excitation. If you just visualize the ground state canonical orbitals (with dscf -proper and $pointval mo <number>-<number>, see chapter 10.2 of the documentation), not the localized ones, you will also get a picture of the excitation.

Regards,

Uwe

Hi Uwe,

I know that there are no excited state orbitals but only excited state densities;
what I need is to have an "orbital's picture" of the character of the excited state
by looking at the most relevant virtual orbitals (>= 10%).

You may ask why I want to localize those relevant virtual orbitals... well, because
sometimes they look either to much delocalized or simply "weird" even if the
calculation was fully converged.

That's why I tought "localization of virtual orbitals" might help with that.


Let me show you an example:

#################### ricc2.out #######################
     +==============================================+
     | type: RE0                        symmetry: a                         state:    1    |
     +---------------------------+--------------------------+------------------------+
     | occ. orb.  index spin | vir. orb.  index spin |  coeff/|amp|     %   |
     +===============+===============+==============+
     |   54 a       54              |   55 a       55              |   0.79128      62.6  |
     |   53 a       53              |   55 a       55              |   -.22787       5.2    |
     |   54 a       54              |   65 a       65              |   0.22067       4.9   |
     |   54 a       54              |   59 a       59              |   0.18068       3.3   |
     |   54 a       54              |   62 a       62              |   0.17354       3.0   |
     |   54 a       54              |   60 a       60              |   -.16402       2.7    |
     |   52 a       52              |   55 a       55              |   0.16134       2.6   |
     |   54 a       54              |   56 a       56              |   -.12131       1.5    |
     |   54 a       54              |   63 a       63              |   0.11838       1.4   |
     |   54 a       54              |   57 a       57              |   -.10858       1.2    |
     |   54 a       54              |   58 a       58              |   -.10485       1.1    |
     |   54 a       54              |   69 a       69              |   0.09560       0.9   |
##################################################

now I would like to localize all the virtual orbitals which are
characterized by >= 3.0% i.e. #55, #65, #59 and #62.

The questions are:
- which module to use in order localize them after the ricc2 run?
- which kewords do I have to add/modify in the last control file modified by the ricc2 calculation?


PS: you mentioned "the ground state canonical orbitals"... I'm sorry but I did not understand what
these "different" orbitals are about... I'm still a newbie one to electronic structures :-)


Regards,
MG.

[uwe]

Hi,

You may ask why I want to localize those relevant virtual orbitals... well, because
sometimes they look either to much delocalized or simply "weird" even if the
calculation was fully converged.

there are and there have been many and long discussions about molecular orbitals and their 'meaning'. A molecular orbital is a linear combination of atomic orbitals (LCAO), which is part of the Ansatz to solve the Schroedinger equation approximatively. The resulting molecular orbitals after convergence are usually not as clear and concrete as many users would like to have them. Delocalized, 'smeared' over the hole molecule, not clear shaped like the atomic orbitals, etc. The funny thing about orbitals is that if you transform them with a unitary transformation, they are still solving your Hartree-Fock equations. So localization with Boys, Pipek-Mezey, etc. just search for a unitary transformation that e.g. maximize the Mulliken charges at the atoms. After that step, the orbitals look a little bit more like atomic orbitals, since they are less delocalized, but I doubt that this gives any insight into the physics of your system. It just gives nice pictures...

Now, if you look at the orbitals from the ricc2 run:

#################### ricc2.out #######################
     +==============================================+
     | type: RE0                        symmetry: a                         state:    1    |
     +---------------------------+--------------------------+------------------------+
     | occ. orb.  index spin | vir. orb.  index spin |  coeff/|amp|     %   |
     +===============+===============+==============+
     |   54 a       54              |   55 a       55              |   0.79128      62.6  |
     |   53 a       53              |   55 a       55              |   -.22787       5.2    |
     |   54 a       54              |   65 a       65              |   0.22067       4.9   |
     |   54 a       54              |   59 a       59              |   0.18068       3.3   |
     |   54 a       54              |   62 a       62              |   0.17354       3.0   |
     |   54 a       54              |   60 a       60              |   -.16402       2.7    |
     |   52 a       52              |   55 a       55              |   0.16134       2.6   |
     |   54 a       54              |   56 a       56              |   -.12131       1.5    |
     |   54 a       54              |   63 a       63              |   0.11838       1.4   |
     |   54 a       54              |   57 a       57              |   -.10858       1.2    |
     |   54 a       54              |   58 a       58              |   -.10485       1.1    |
     |   54 a       54              |   69 a       69              |   0.09560       0.9   |
##################################################

As you can see, the excitation is not of a 'simple' character like pi -> pi* . It would be sheer luck to find a localized orbital that is build from a similar combination of molecular orbitals than given in the output above. Here, the difference density should give you much more insight in the kind of excitation than to look at the orbitals - since the excitation does not come from one single orbital and goes into another orbital.

That's why I tought "localization of virtual orbitals" might help with that.

That would only be the case if the whole calculation was based on localized orbitals rather than on the canonical ones (i.e. the ones that diagonalize the Fock matrix and come out 'natively' from a Hartree-Fock calculation).

I guess that even if localized orbitals were used: Your excitation is not of a local character, so you would get a similar combination of localized molecular orbitals in the list above than you get now.

The questions are:
- which module to use in order localize them after the ricc2 run?
- which kewords do I have to add/modify in the last control file modified by the ricc2 calculation?

1. use dscf to localize
2. none, but you can get localized orbitals with $localize, see chapter 10 of the documentation.

PS: you mentioned "the ground state canonical orbitals"... I'm sorry but I did not understand what
these "different" orbitals are about... I'm still a newbie one to electronic structures :-)

I have tried to explain that a little bit in the sections above, and I hope that it helps a bit.

Regards,

Uwe

[Matteo Guglielmi]

Hi,
...

I have tried to explain that a little bit in the sections above, and I hope that it helps a bit.

Regards,

Uwe

Ok Uwe,

Thanks for your explanation.

You convinced me that thare's more physics in the picture of the excited state density
than in localized orbitals.

So now lets suppose I use the following input file for the calculation
of the first 6 singlet excited states of a molecule (ricc2):

Code: [Select]

bla bla bla...
$ricc2
  maxiter=  150
$cbas    file=auxbasis
$tmpdir /scratch
$excitations
  irrep=a  multiplicity=  1  nexc=  6  npre=  6  nstart=  6
  spectrum  states=all  operators=diplen,dipvel
  exprop  states=all  operators=diplen,qudlen
$response
  semicano
  fop relaxed  operators=diplen,qudlen
$end

than, after having had a look to the specrum, I decide to calculate the density of the third excited state:

Where and how do I have to modify the input file in order to calculate the excited state density I need?

Do I have to use the keyword 'xgrad states=(a 3)' into the $excitations section?

How can I restart from the previous ricc2 calculation?

Sorry but I find the turbomanual not so clear regarding this problem.

Best,
Matteo.

[christof.haettig]

Yes, if you want to have the density for the 3th state you should ask for first-order properties or gradients for the 3th state. (Similar as for escf.) The name of the file on which the density will be saved etc. is explained in the manual in  ' The general density analysis option:'  subsection of the RI-CC2 chapter.

Christof