[Wien] Weird wannier functions coming out from wien2wannier

Pablo Villar Arribi pablo.villar at esrf.fr
Wed Apr 6 18:14:47 CEST 2016


Dear Elias,

First of all, sorry for the very late reply. Let me answer point by 
point.

> 1. By way of general advice, I suggest to start with a coarse r-mesh
> (e.g. 10×10×10 points) and a larger plot region than you think you 
> need,
> to figure out the correct plotting window.  You said you checked this,
> but it can really be tricky, what with different settings both in
> Wannier90 (translate_home_cell) and wplot2xsf (-noshift).  Once you 
> have
> that, you can scale up the precision.

I've double checked this and I'm pretty sure that I'm plotting the right 
region. I've been using 8x8x8,10x10x10 and 12x12x12 k-meshes to perform 
all my tests.



> 2. All parts of the calculation (mainly, case.chk and case.vector) have
> to match.  If, for some reason, you ran another lapw1 after wannier90,
> wplot will not work.  Therefore, it sometimes helps simply to repeat 
> the
> whole procedure (lapw1 && w2w && wannier90 && wplot), just to be sure
> that there is no inconsistency.

I've repeated the procedure several times from scratch and every time I 
get the same result.



> 3. Did you check if the real-space representation of the WF is real
> (i.e., case.psink should be 0 or ±π up to numerical precision)?  
> Because
> the xsf format does not accomodate complex numbers, wplot2xsf by 
> default
> writes |w(r)|² sgn Re w(r).  This could lead to “strange” isosurfaces 
> if
> there is a phase.

When I open the *psink files I see something like this (this is only the 
head of the file)

-------------------------------------------------------
3D    NP         abs(X)   ang(X1,X)   ang(X2,X)
       15      7.2641070
       15      7.2641070    90.00000
       15     26.2974260    90.00000    90.00000
order: (((psi(ix,iy,iz),iz=1,nz),iy=1,ny),ix=1,nx)
   6.18314155E-05  3.08656339E-03  2.66355611E-06  2.69500392E-08  
5.04621879E-08  1.37216569E-06  8.48431916E-05  2.12029520E-04  
1.10324475E-04  4.31439119E-07
   2.98617091E-08  1.29545184E-09  9.81248903E-08  1.85362152E-04  
5.62558916E-04  4.05611462E-01  1.39474115E-03  2.53339043E-06  
3.22816113E-08  4.33206965E-08
   1.00373845E-06  2.69807803E-04  6.89898991E-01  1.00181080E-04  
9.45113631E-06  1.26330365E-07  1.64959367E-08  2.47125345E-06  
1.71181513E-04  2.03070344E-02
   9.35951297E-02  2.99484270E-04  1.98104817E-06  3.63551589E-08  
1.41456246E-08  4.69129054E-07  2.54101577E-04  1.58105216E-01  
3.92024553E-05  7.41021780E-05
--------------------------------------------------------

I guess that the problem may be here since I cannot see 0, +-Pi or 
multiples of Pi anywhere. However, as I told you, all the hopping 
integrals in the real-space Hamiltonian are real, which for me indicates 
that the calculation with Wannier90 has converged to the 
Maximally-Localized Wannier Functions. I'm asuming that having real 
Wannier functions implies having real hoppings, but not the opposite. Do 
you have any clue on how could I improve/modify this? I've tried with 
"rephase case" but it does not work, I get an error of the type

  read-in initial values for Wannier functions...
forrtl: No such file or directory
forrtl: severe (29): file not found, unit 15, file "my_path"/t1/t1_ 
1.psiarg
Image              PC                Routine            Line        
Source
rephase            00000000004A1F6E  Unknown               Unknown  
Unknown
rephase            00000000004A0A06  Unknown               Unknown  
Unknown
rephase            0000000000459B42  Unknown               Unknown  
Unknown
rephase            00000000004155FC  Unknown               Unknown  
Unknown
rephase            0000000000414B1C  Unknown               Unknown  
Unknown
rephase            00000000004284D8  Unknown               Unknown  
Unknown
rephase            000000000040448F  MAIN__                    136  
rephase.f
rephase            000000000040333C  Unknown               Unknown  
Unknown
libc.so.6          00007FC01202BEAD  Unknown               Unknown  
Unknown
rephase            0000000000403219  Unknown               Unknown  
Unknown



> 4. If the above does not help and you just want your plots, it could be
> an option to simply use the ones from Quantum Espresso.  In my view,
> real-space WFs are a qualitative rather than a quantitative tool 
> (unless
> you are very careful …), therefore it could be okay to use the QE
> real-space WFs as a visualization for your Wien2k WFs, as long as H(R)
> and the band structure are “the same”.  Obviously, this depends on what
> you want to do with them.

Both the H(R) and the bandstructures are the same but I need the shape 
of the Wannier orbitals from Wien2k, which is meant to coincide with the 
shape of the Wannier orbitals from Quantum Espresso in the outer part, 
but not necessarily in the inner region close to the nucleus since one 
is an all-electron code and the other one is a pseudopotential one.



> 5. On the other hand, if you want to get it right with Wien2k (and
> possibly help me find a bug), the next thing I would try is a different
> version of wplot (I am assuming that you tried the one included in
> Wien2k 14.2).  This is one part of the package that I changed quite a
> lot recently and invariably, bugs creep in.  You could either try the
> newest version from GitHub (caveat: the format of case.inwplot changed)
> where some of those bugs are fixed; or the  “old” one (0.97, available
> at
> http://www.ifp.tuwien.ac.at/forschung/arbeitsgruppen/cms/software-download/wien2wannier/),
> which has a different set of bugs.

I've been using the 14.1 version, not 14.2. Should I change to 14.2 or 
is it OK with 14.1?



> 6. Relatedly, what crystal structure do you have?  Wien2Wannier has had
> some problems with more “difficult” lattice types in the past.

I'm working with the 122 family of the iron-based superconductors.



Thank you very much for your advice.

Best regards,

Pablo


-------------------
Pablo Villar Arribi
PhD student at ESRF
Grenoble, France


More information about the Wien mailing list