Initial guess of the density matrix

[Simone Piccinin]

Hi,

thanks a lot, now I think I understand how to use this option.
If this doesn't solve my problem I'll try Flo's suggestion.

Best wishes,
Simone


On 12 Feb, 11:13, marci <marc... at pci.uzh.ch> wrote:
> Hi
>
> you are almost right. Actually what the keywords say in this case is
> that for Cu1 1 el. 4s is taken away in spin up, while 1 4s and 2 3d
> are taken in spin down. The misunderstanding comes from the fact that
> cp2k is working for each spin independently, always using as a
> reference the full electronic configuration, and if LSD is active a
> factor 2 is applied afterwards.
> Then Cu is 4s-1 3d-10. The change imposed for the bias leads to Cu1
> ALPHA 4s-0 3d-10 . After the factor 2 is 4s-0 3d-5.
> Cu1 BETA instead becomes 4s-0 3d-4. At the end Cu1 has one unpaired
> spin-up.
>
> It should become clearer by reading the extended output that appears
> with print level MEDIUM.
> Example
> ##########
>  Spin 1
>
>  Guess for atomic kind: Cu1
>
>  Electronic structure
>     Total number of core
> electrons                                         18.00
>     Total number of valence
> electrons                                      10.00
>     Total number of
> electrons                                              28.00
>     Multiplicity                                                   not
> specified
>     S   [  2.00  2.00  2.00]
>     P   [  6.00  6.00]
>     D     10.00
> #########
>
> Indeed, for this system the counting of electrons after the bias must
> be readjusted, because each spin channel turns out to have one
> electron too much. In the output you can read:
> #######
> Re-scaling the density matrix to get the right number of electrons for
> spin 1
>                   # Electrons              Trace(P)
> Scaling factor
>                            33
> 33.999                        0.971
> ######
>
> and later on
>
> ######
> Re-scaling the density matrix to get the right number of electrons for
> spin 2
>                   # Electrons              Trace(P)
> Scaling factor
>                            33
> 33.999                        0.971
> #####
>
> Hope this helps,
> Marcella
>
> On Feb 12, 9:27 am, Simone Piccinin <piccini... at gmail.com> wrote:
>
> > Dear Marcella,
>
> > thanks a lot for the answer.
> > I'm not sure I understand how to use these feature, though.
>
> > The relevant section of the cu2cl6_m1_clp6cud9.inp file in the test/QS/
> > regtest-bs folder is this:
>
> >     &KIND  Cu1
> >       ELEMENT Cu
> >       BASIS_SET  DZVP-MOLOPT-SR-GTH
> >       POTENTIAL  GTH-BLYP-q11
> >       &BS
> >         &ALPHA
> >           NEL -1   0
> >           L    0   2
> >           N    4   3
> >         &END
> >         &BETA
> >           NEL -1  -2
> >           L    0   2
> >           N    4   3
> >         &END
> >       &END
> >     &END
>
> > Does this mean that we are removing, from the alpha channel of Cu1, 1
> > electron from the 4s and in the beta channel 1 electron from the 4s
> > and 2 electrons from the 3d?
> > If so, Cu1 now has 4 electrons less than the neutral atom. A similar
> > treatment is done for Cu2. For Cl, on the other hand, it seems that we
> > are adding 2 alpha
> > electrons and 2 beta electrons in the 3p orbitals. Since we have 6 Cl
> > and 2 Cu atoms in the molecule, the overall charge should be 2*4-6*4 =
> > -16, resulting
> > in a wrong number of electrons. Is this what is done, or did I
> > misunderstand something?
>
> > Thanks a lot.
> > Best wishes,
> > Simone
>
> > On 11 Feb, 22:59, marci <marc... at pci.uzh.ch> wrote:
>
> > > Hi
>
> > > something in the direction of a biased initial guess is already
> > > present in QS.
> > > The ATOMIC guess is normally based on the construction of a first
> > > density matrix
> > > by occupation of atomic orbitals according to the electronic
> > > configuration
> > > of each element.
> > > However, there is the possibility to modify this procedure by
> > > subtracting or adding
> > > valence electrons (NEL) from specific channels (identified by N and
> > > L ) of selected elements.
> > > This can be done independently for spin up (ALPHA) and spin down
> > > (BETA).
> > > Have a look at the reference inputs in cp2k/tests/QS/regtest-bs.
> > > This might provide a better initial guess, also for charged systems.
> > > As a safety check, if the resulting density matrix amounts to the
> > > wrong number of  electrons,
> > > it is rescaled by  a proper factor, and the calculation proceeds.
> > > This trick is only used for the initial guess, no further constraints
> > > are applied afterwards.
> > > Hence, there is no guarantee that the selected configuration is kept
> > > during the SCF.
>
> > > Kind regards
> > > Marcella
>
> > > On Feb 11, 9:12 pm, Simone Piccinin <piccini... at gmail.com> wrote:
>
> > > > Dear CP2K users,
>
> > > > I'm studying a large molecule containing 4 Ru atoms. These are
> > > > arranged as to form two pairs. I'd like to study the molecule in a
> > > > singlet state, and see if the coupling within each pair is
> > > > ferromagnetic (FM) or antiferromagnetic (AF). Is there a way in
> > > > quickstep to guide the scf cycle toward the FM or AF solution, for
> > > > example through an appropriate initial guess of the density matrix?
>
> > > > Thanks for your help.
>
> > > > Best wishes,
>
> > > > Simone Piccinin
> > > > CNR-IOM Democritos (Trieste)