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

Natalie Austin natalie... at
Tue Nov 10 19:14:06 UTC 2015

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 



On Tuesday, November 10, 2015 at 8:39:45 AM UTC-5, Marcella Iannuzzi wrote:
>   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
> On Monday, November 9, 2015 at 3:17:59 PM UTC+1, Natalie Austin wrote:
>> 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list