SCF convergence problem for ferrite perovskite

[Ruihao Bi]

Hi all,

    I want to calculate the PDOS of bulk SrFeO3, however, I find out SCF convergence is related to the simulation cell in an interesting way.

    Strontium ferrite perovskite (SrFeO3) is a metallic oxide, with helical magnetic order. Therefore, I used UKS and set up ferromagnetic magnetic order (approximate helical). In addition, I used Traditional diagonalization with smearing for it's metallic. What's unusual is SCF converged for 3x3x4 and 3x4x3 suppercells but not for 3x3x3 and 4x4x4.

    As I'm very new to CP2K, I suspect I made some mistakes in my inputs but I couldn't put my finger on where exactly. Below are parts of my input. All inputs, outputs, and coordinate files are in the attachment. 

    Any suggestions are welcomed!

! +++ DFT setup +++!

 &DFT

  BASIS_SET_FILE_NAME  BASIS_MOLOPT

  POTENTIAL_FILE_NAME  GTH_POTENTIALS

  CHARGE 0

  UKS .TRUE.

  &MGRID

   REL_CUTOFF 60

   NGRIDS 5

   CUTOFF 800

  &END MGRID

  &POISSON

   PERIODIC XYZ

  &END POISSON

  &QS

   EXTRAPOLATION USE_GUESS

  &END QS

  &SCF

   SCF_GUESS ATOMIC

   EPS_SCF 1.0E-6

   MAX_SCF 50

   &OUTER_SCF

    EPS_SCF 1.0E-6

    MAX_SCF 10

   &END OUTER_SCF

   ADDED_MOS 200

   &SMEAR ON

    METHOD FERMI_DIRAC

    ELECTRONIC_TEMPERATURE [K] 300

   &END SMEAR

&DIAGONALIZATION

    ALGORITHM STANDARD

    EPS_ADAPT 0.01

   &END DIAGONALIZATION

   &MIXING

    METHOD BROYDEN_MIXING

    ALPHA 0.2

    BETA 1.5

    NBROYDEN 8

   &END MIXING

  &END SCF

  &XC

   &XC_FUNCTIONAL PBE

   &END XC_FUNCTIONAL

  &END XC

&PRINT

   &PDOS

    NLUMO -1

    COMPONENTS .TRUE.

   &END PDOS

  &END PRINT

 &END DFT

! +++ KIND setup for Fe +++!

&KIND Fe

  ELEMENT Fe

  POTENTIAL GTH-PBE-q16

  BASIS_SET DZVP-MOLOPT-SR-GTH

  &BS T

   &BETA

    N    4      3

    L    0      2

    NEL -2     -6

   &END BETA

   &ALPHA

    N    4      3

    L    0      2

    NEL -2     +2

   &END ALPHA

  &END BS

 &END KIND