Is it possible to specifiy the multiplicity of a metallic (or bimetallic) system AND include the Fermi keyword?

[Natalie Austin]

Hello,

I've been comparing the output files for the geometry optimization of Cu55 at different multiplicities (M= 2, 4, 6,8). What I noticed is that although I specify multiplicity the atomic population from Mulliken Population Analysis is not representative of the multiplicity that I included.

Specifically,
 
for the doublet case i specified 303 for alpha and 302 for beta
and in the muliken analysis the values were 302.499984   302.500016
ive included the values for the multiplicity 4, 6, and 8 also below

quartet
304   301

303.546324   301.453676

sextet

305  300
303.548877   301.451123

octet
306    299

303.550009   301.449991

So it seems that the multiplicity isn't reflected in the final result. I'm starting to believe that this might be attributed to the fermi keyword which I read in another thread does not constrain the multiplicity. If this is the case, is it not possible to set the multiplicity in metallic systems or are there other ways to converge metallic systems without using the fermi keyword?

I've included my input and output files for the doublet case below.

Thanks,

Natalie

[Matt W]

Hi Natalie,

sounds like FIXED_MAGNETIC_MOMENT in the SMEAR section is what you want.

Matt

[Natalie Austin]

Hello,

I added FIXED_MAGNETIC_MOMENT -1  to the SMEAR section for the multiplicity 4 calculation but the alpha and beta orbitals were still fractional values: 303.548262   301.451738

In the Non-integer total number of Alpha/Beta Electrons. post someone stated that "One uses the smearing for metallic systems, and the Fermi energy of the two spin channels has to be the same.

This is not the case if the electrons are not free to fill the lowest energy states, irrespective of the spin"

Does this mean by restricting the alpha and beta to integer numbers my system could be (or is always) higher in energy than a system where fractional occupations are allowed?

Generally, I'm a bit unsure about the standard for performing spin polarized calculations in CP2K, do you only need to set UKS TRUE? is setting the multiplicity just as important?

Thanks,

Natalie

[Natalie Austin]

Hello,

I was able to get FIXED_MAGNETIC_MOMENT to work for an isolated Cu55 cluster. It turns out that there isn't a significant difference in the total energy if I use this keyword or not. However when I incorporated a nickel atom into the cluster (Cu54Ni), the optimization process was really slow (3 steps in 31 hours). It seems that fixing the magnetic moment in this case is not probable.
So my question still stands, is setting the multiplicity important if it is not reflected in the final result when using the fermi keyword?

Any help with this would be much appreciated

Thanks,
Natalie

[Marcella Iannuzzi]

  Dear Natalie, 

  Ideally it should be always possible that the system converges to the lowest energy state, whereas by fixing the multiplicity the system is 

forced and keep the assigned number of electrons for the two spins up and down. 

 On the other hand, to estimate energy differences among different spin states, one can fix the multiplicity and try to prepare an initial guess as close as possible to the desired spin state. If that spin state is at least a metastable state, the wave function should converge there. 

As it should be, cp2k gives both the possibilities. By applying the Fermi-Dirac smearing the occupation of the two spin channels is adjusted in order to reach the lowest energy state. By fixing the multiplicity, the system is forced in a given state, and most probably, the Fermi energy is going to be different for the two spin channels. For a non metallic system this translates to having a different HOMO energy for spin up and spin down. 

Which are the stable states for your system is a problem not really related to cp2k itself, but more to the level of theory and the other approximations of the electronic structure calculations. Are you sure that DFT, or the selected functional, or the basis sets and PPs are adequate to describe the magnetic properties you are interested in? 

If the state you are searching is not a minimum for the model you are using, it will be difficult to converge there.

Anyway, the Mulliken population analysis is just a rough evaluation of the charge and spin distribution. I would not take the Mulliken values too strictly and I would use also other analysis tools.

From a more technical point of view, the initial guess obtained by setting the multiplicity and also by the broken symmetry approach can be useful to bias the convergence toward a given state. However, starting from an electronic configuration that is far from any stable state might cause lengthy optimisation procedure. It is also possible that the SCF converges to some wrong result, if it does not find its way back to a reasonable minimum.

kind regards,

Marcella

[Natalie Austin]

Hello Marcella,

Thank you for your response. To the best of my knowledge the level of theory I've implemented is sufficient for the properties that I'm looking into. Based on the papers I've come across papers in which structural, electronic, and magnetic properties of Ni and Cu were investigated using fixed multiplicities, I want to be able to estimate energetic difference among different spin states for my systems of interest. Therefore my issue at hand now would be converging the Cu54Ni system with the fixed multiplicities. Hopefully by increasing the electronic temperature(range 500-2000K) I can get my system to reach convergence. Can I safely increase temperatures up to 2000K as long as the electronic temperature contribution is sufficiently small so that extrapolation to zero temperature result is reliable? 

Thanks,

Natalie