[Wien] LDA+U for 5f electron system

[Stefaan Cottenier]

>       I am trying to apply LDA+U (SIC) method to obtain the ground   
> state energy of a 5f electron system with an anti-ferromagnetic   
> arrangement. For lower values of U, say 1eV or 1.5 eV, scf cycle   
> does converge but if I use U = 3eV or 2.5 eV, some of the data   
> points do not converge even after running few hundred cycles, making  
>  it difficult to find lowest energy configuration. I am wondering if  
>  anyone else have faced similar non-convergence problem with LDA+U   
> method. Any comment will be appreciated.

With LDA, your 5f system is itinerant (for the lightest actinides, at  
least). With increasing U, you increase the localization. For such  
localized systems, it frequently happens that several self-consistent  
solutions exist (local minima in the born-oppenheimer surface, here  
they correspond to different ways to put the 5f electrons over the  
available f-orbitals). For specific values of U, it might happen that  
two or more of these solutions are degenerate, and during the scf  
cycle you keep shuffling around charge around the f-orbitals, from the  
occupation corresponding to one type of solution to another, without  
ever settling down in one of them.

One way to get around this, is to converge first with -orbc (this  
keeps the orbital potential that corresponds to the present density  
matrix fixed -- the density matrix will get adjusted selfconsistently  
to this fixed potential). This removes some of the freedom, and might  
help to stabilize one of the solutions if you release -orbc. By  
starting from another density matrix later on, you can stabilize  
another solution. I know no good general rules to select which of  
those solutions is the most physical one...

Stefaan


Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm