[Wannier] numerical issues with atoms at the center of the unit cell

[Aron Szabo]

Dear Wannier90 users and developers,

I have experienced a strange behavior of Wannier90. I was using VASP to 
do the SCF simulation and generate the initial projections of a system, 
where I had scandium over MoS2. I was using 
Sc:sp3-1;Sc:l=2;Mo:l=2;S:l=1; as projections. And when running 
wannier90, most of the Wannier functions had reasonably small (2-3 A^2) 
spreads even at the first step, except those that were corresponding to 
the l=2 (d orbital) projections of the 3rd and 9th Sc atoms (>100 A^2), 
and they remained large during the minimization. It was suspicious that 
those atoms had direct z coordinates very close to 0.5, and I thought 
this error might have been related to this fact. So I have shifted all 
the atoms with 0.025 in the z direction, and calculated everything 
again, and now the spreads were as small for those orbitals as for every 
other, while the system itself was in fact the same, since a shift does 
not change anything in a periodic system. So it seems like there is some 
numerical instability when atom coordinates are very close to the half 
of a lattice vector. I don't know if it is related to VASP or wannier90. 
I can circumvent it easily now that I know it. I was just wondering if 
anyone else has experienced the same, or maybe have any guess where this 
error could originate from. I can provide additional details if anyone 
is interested. The original POSCAR file is below.

Best regards,
Aron Szabo

POSCAR:
MoS2+Sc
    3.18300000000000
      1.0   0.0    0.0
      0.0   1.73205080757   0.0
      0.0   0.0    9.5617407071622829
    Sc   Mo   S
      12     2     4
Direct
   0.0000000000000000 -0.0000000000000000  0.3265600783327190
   0.0  0.3333332999999996  0.4146745116137508
   0.0000000000000000 -0.0000000000000000  0.5006405198480147
   0.0  0.3333332999999996  0.5867734775290360
   0.0000000000000000 -0.0000000000000000  0.6736668439773074
   0.0  0.3333332999999996  0.7587831003021329
   0.5000000000000000 0.5000000000000000  0.3265600783327190
   0.5  0.8333332999999996  0.4146745116137508
   0.5  0.5000000000000000  0.5006405198480147
   0.5  0.8333332999999996  0.5867734775290360
   0.5  0.5000000000000000  0.6736668439773074
   0.5  0.8333332999999996  0.7587831003021329
  -0.0000000000000000 -0.0000000000000000  0.2132268304255010
   0.5000000000000000  0.5000000000000000  0.2132268304255010
   0.0  0.3333332999999996  0.2689120454016644
   0.0  0.3333332999999996  0.1626395725698760
   0.5  0.8333332999999996  0.2689120454016644
   0.5  0.8333332999999996  0.1626395725698760