[Pw_forum] starting_magnetization and magnetization optimization in pwscf

Stefano de Gironcoli degironc at sissa.it
Thu Apr 1 23:17:50 CEST 2010


Dear Jia Chen,
  the starting magnetization variable is used to define the initial 
charge density in a scf run in pwscf.
  This density is used to build the first KS potential which thus 
results to have a magnetically broken symmetry.
  The resulting spin-up and spin-down hamiltonians are diagonalized and 
the Fermi energy located from the constraint on the total number of 
electrons (the system must be treated as a metal and a number of empty 
states are present).
  This determines the output density and since the potential is 
"magnetically ordered" the density also is if the system likes so.
  Then the code mixes in and out densities and builds a new guess for 
the input density and hence the new KS potential and so on...
   So starting magnetization really enters only in PW/atomic_rho.f90  
which  is used to build the starting guess for the  input charge density as

   rho_up =  sum_s  (1.0+starting_magnetization_atom_s)/2 * 
unpolarized_rho_atom_s
   rho_dw = sum_s  (1.0-starting_magnetization_atom_s)/2 * 
unpolarized_rho_atom_s

Hope this helps,

stefano


jia chen wrote:
> Dear All,
>
> Pwscf has a very nice feature that can set initial magnetism of a
> certain kind of atom by starting_magnetization. It is working very
> well for me. But I would like to know more about how this is really
> working in cp code, during magnetization optimization calculations. It
> is too much for me to read the code to get the answer. So I would like
> to give a guess of basic idea of this process.
>
> When we start a calculation with initial atomic wave functions. Pwscf
> code can give atomic occupations of spin up and down whose ratio
> satisfy the value of starting_magnetization. During the calculation,
> pwscf can adjust the occupation number after each diagonal step
> according to orbitals energies because pwscf goes with a bunch of
> empty states. Then there can be "electron flow" between spin up and
> down. Basically, that's I guess how pwscf does magnetization
> optimization.
>
> Can you tell me what is the actual process pwscf goes though in this
> kind of magnetization optimizations? Any comments are very welcome.
> Thank you in advance.
>
> Best Wishes
>   




More information about the users mailing list