[Wannier] A question on the Hamiltonian with spinors from Wannier90

[Yan, Binghai]

Dear Wannier developers

I am new user of wannier90. I extract wannier functions from a DFT calculation (I use vasp) including spin orbit coupling (SOC). Then I switch on spinors=T and double the num_bands. 
My systems have no spin polarization, respecting the time reversal symmetry. A wannier function is expected to be represented by two functions, which have the same center but different spins. I expect these two functions have the same onsite energies and the hopping term between them are exactly zero.
However, I get different onsite energies and nonzero hopping terms, although the band structure is well reproduced. The wannier cenceters are also very different.

The following is an example of my inputs for bulk HgTe calculations. (zincblende structure)

==========
num_wann          =   16
num_bands         =   38
#exclude_bands     =   3-12

dis_win_max       =  15.0
dis_froz_max      =  9
dis_num_iter      =  2000
dis_mix_ratio     = 1.d0

num_iter          = 1000
num_print_cycles  = 10

Begin Projections
Hg : sp3
Te : sp3
End Projections

spinors = .true.

begin unit_cell_cart
     3.2300000     3.2300000     0.0000000
     0.0000000     3.2300000     3.2300000
     3.2300000     0.0000000     3.2300000
end unit_cell_cart

begin atoms_cart
Hg       2.4225000     2.4225000     2.4225000
Te       4.0375000     4.0375000     4.0375000
end atoms_cart

mp_grid =    12    12    12

begin kpoints
      0.000000000000      0.000000000000      0.000000000000
      0.083333333333      0.000000000000      0.000000000000
      0.166666666667      0.000000000000      0.000000000000
      0.250000000000      0.000000000000      0.000000000000
................
========================

The optimized wannier centers are :
 Cycle:   1000
  WF centre and spread    1  (  1.125081,  2.105459,  2.107134 )     7.50544434
  WF centre and spread    2  (  2.108857,  1.123510,  2.103942 )     7.50493864
  WF centre and spread    3  (  2.106035,  2.107794,  1.122418 )     7.50768700
  WF centre and spread    4  (  1.124096,  1.124214,  1.123910 )     7.52064335
  WF centre and spread    5  (  3.522963,  3.524601,  3.524535 )     4.48040046
  WF centre and spread    6  (  3.526018,  2.934456,  2.939037 )     4.47399171
  WF centre and spread    7  (  2.939672,  3.523541,  2.932669 )     4.46863560
  WF centre and spread    8  (  2.934153,  2.938291,  3.522491 )     4.47010302
  WF centre and spread    9  (  2.107155,  2.105137,  1.125272 )     7.64204572
  WF centre and spread   10  (  1.123133,  1.123929,  1.123907 )     7.64532973
  WF centre and spread   11  (  1.122550,  2.106696,  2.105047 )     7.63263662
  WF centre and spread   12  (  2.104217,  1.123646,  2.107754 )     7.63564195
  WF centre and spread   13  (  3.524860,  3.524017,  3.523515 )     4.51698438
  WF centre and spread   14  (  3.522838,  2.936705,  2.934278 )     4.51145886
  WF centre and spread   15  (  2.933391,  3.524025,  2.938085 )     4.50864047
  WF centre and spread   16  (  2.937251,  2.934720,  3.525198 )     4.51260402
  Sum of centres and spreads ( 38.762270, 38.760740, 38.759191 )    96.53718585

For example, the matrix elements between WF 8 and WF 16 are 

    0    0    0    8    8    0.863620    0.000000
-
    0    0    0   16   16    0.932080    0.000000
-
    0    0    0   16    8   -0.032312   -0.000905
-
    0    0    0    8   16   -0.032312    0.000905


Is this problem due to a bad projection? Actually, I used sp3 for non-SOC case, it seems pretty well. 
Is there a way to constrain different spins have the same wannier centers? 

Thank you very much in advance!
Binghai