Ion ground state doesn't converge

#  HCCI molecule, initial single point
# Default to atomic units
# Structure from Kraus et. al. 10.1126/science.aab2160 (SM 5), CCSD(T)/cc-pVQZ/ECP28MDF
CalculationMode = gs
FromScratch = yes
ExperimentalFeatures = true
# Simulation box and geometry
Dimensions = 3
PeriodicDimensions = 0
Spacing = 0.05*angstrom
BoxShape = parallelepiped
%Lsize
4.0 | 4.0 | 9.0
%
%BoxCenter
0.0000 | 0.0000 | -4.5000
%
%Coordinates
"H" | 0.000000 | 0.000000 | -7.247245 | yes
"C" | 0.000000 | 0.000000 | -5.236745 | yes
"C" | 0.000000 | 0.000000 | -2.949445 | yes
"I" | 0.000000 | 0.000000 |  0.832255 | yes
%
# Starting occupations
#%Occupations
#1 | 1 | 1 | 1 | 1 | 1 | 1 | 1
#1 | 1 | 1 | 1 | 1 | 1 | 1 | 0
#%
MixField=density
ExcessCharge=1.0
SpinComponents = polarized
PseudopotentialSet = hgh_lda
ConvRelDens=1.0e-9
EigensolverTolerance=1.0e-8
#Eigensolver=chebyshev_filter
Eigensolver=rmmdiis
MaximumIter=-1
#SmearingFunction=fermi_dirac
ExtraStates = 8
ExtraStatesToConverge = 8
%Output
  dos
  geometry | xyz
  density | cube+xcrysden
%

This ground state SCF won't converge, with either occupations explicitly specified or just with ExcessCharge=1, or with a non-default eigensolver, or with the MixField set to the KS potential. After a couple of iterations the occupations look right, but even after 500 SCF iterations it doesn't converge. Any clues why?