A first-principle electronic structure calculation package in f90, especially for the PMT-QSGW.

We have a tutorial course at CMD workshops held by Osaka university (every March and Sep). http://phoenix.mp.es.osaka-u.ac.jp/CMD/index_en.html

This is read me at https://github.com/tkotani/ecalj. (checked at Dec2014.This README is still away from perfect...)

1. All electron full-potential PMT method: a mixed basis method of two kinds of augmented waves, that is, L(APW+MTO). Relaxiation of atomic positions is possible in GGA/LDA and LDA+U. Our recent development shows that very localized MTO (damping factor is \sim 1 a.u), together with APW (cutoff is \sim 3 to 4 Ry) works well to get reasonable convergences. In principle, it is possible to perform default calculations just from atomic structures.

2. The PMT-QSGW method, that is, the Quasiparticle self-consistent GW method (QSGW) based on the PMT method. In addion, we can calculate dielectric functions, spectrum function of the Green's functions and so on. GW-related codes are in ~/ecalj/fpgw/. For paralellized calculations, we can use lmf-MPIK and mpi version of hvccfp0,hx0fp0_sc,hsfp0_sc. (although we still have so much room to improve it). The PMT allows us to perform the QSGW calculations virtually automatically (10.7566/JPSJ.83.094711). http://journals.jps.jp/doi/abs/10.7566/JPSJ.83.094711

 
!! CAUTION for know bug(or not) for spin susceptibility mode!!! (apr2105).
T.Kotani think epsPP\_lmfh\_chipm branch may/(or may not) have a bug
(because of symmetrization). It may be near
------------------
          if (is==nspinmx) then 
            symmetrize=.true.
            call x0kf_v4hz(npm,ncc,... 
-----------------
in fpgw/main/hx0fp0.m.F
(This bug may be from a few years ago, after I implemented EIBZ mode).
I think  "if (is==nspinmx.or.chipm) then" may be necessary
especially for cases with more than two atoms in the cell
(thus fe_epsPP_lmfh test may not work for this case...)
A possible test is remove symmetrization---> use eibzsym=F.
If necessary, let me know...

3. Wannier function generator and effective model generator (Maxloc Wannier and effective interaction between Wannier funcitons). This is mainly from Dr.Miyake,Dr.Sakuma, and Dr.Kino. See fpgw/Wannier/README.

We have another home page at http://pmt.sakura.ne.jp/wiki/, but not well-organized yet, little in English yet. We will renew it.

Utilities such as a converter between POSCAR(VASP) and our crystal strucrue file 'ctrls.*' are included. (slightly buggy; let T.Kotani know problems in it; note that we should supply numerically accurate atomic positions to judge crystal symmetry automatically).

For your publications, please make two citations directly to this homepage as;

[1] ecalj package at https://github.com/tkotani/ecalj/. Its one-body part is developed based on Ref.[2].
[2] LMsuit package at http://www.lmsuite.org/. Its GW part is adopted mainly from Ref.[1].

in addition to our papers (in the references).

Follow these steps explained below.
(0) Get ecalj package and get tools.
--- You can run folling (1)-(4) by a command InstallAll.foobar at ecalj/
(1) make for single-core LDA part,
(2) make for MPIK LDA part,
(3) make for MPIK GW part.
(4) Install test
(5) crystal structure tools (not necessary). In the following explanation, we assume gfortran in ubuntu. But we can also use ifort, and others in your environment with minimum changes in makefiles. For testing \ecalj\ for small systems such as Si and GaAs, we can use even Ubuntu + gfortran + note PC. When install procedure have finished, we have all required binaries and shell scripts in your \verb+~/bin/+ directory). (or somewhere else where BINDIR specified in InstallAll.*).

--- Let us assume you use ubuntu. --- You need following tools and libraries to be installed.

sudo apt-get install git #version control and to get source from github
sudo apt-get install gitk #git gui.
sudo apt-get install gfortran # GFORTRAN
sudo apt-get install openmpi-bin libopenmpi-dev # or openmpi-dev ? sudo apt-get install libfftw3-3 or something else # FFTW
sudo apt-get install libblas3gf or something else # BLAS
sudo apt-get install liblapack3gf or something else # LAPACK
sudo apt-get install csh bash tcsh gawk # shells

memo: I think etags is automatically installed when you install emacs in the latest ubuntu. etags is essentially needed only for developers.

Note that python 2.x is also assumed (usually already in ubuntu. Type >python (ctrl+D for quit)). Version ctrl is by git (which makes things easier, but not necessarily required for installation).

After you have installed git (version control software), you can get ecalj package by

git clone https://github.com/tkotani/ecalj.git # Get source code

for ecalj. or get it in the form *.zip from the page https://github.com/tkotani/ecalj (push zip button). I recommend you to use git, to check your changes (>git diff), know version id, and upgrade. After you did the above git clone command, a directory ecalj/ appears (under the directory at which you did git clone).

We can check history of ecalj code development by ">gik --all" at ecalj/ directory after you did git clone.

Following procedures, (1)-(4), are done automatically by a script, for example, InstallAll.ifort (in the case of intel fortran) as;

cd ecalj
./InstallAll.ifort
(To clean all, do ./CleanAll.ifort).

Please look into the script "InstallAll.ifort". It contains the setting of your BINDIR= directory, to which the InstallAll.ifort will copy all binaries and scripts. It internally uses three machine-comilar dependent files;
a.lm7K/MAKEINC/Make.inc.ifort (for single core version -->(1))
b.lm7K/MAKEINC/Make.inc.ifort_mpik (k-point paralell version -->(2)
c.fpgw/exec/make.inc.ifort (this is only for mpi-omp version -->(3)).
At the last stage of the script, it runs automatic tests. (You can neglect failure for nio_gwsc; it may show one-failure among two checks). The test may use ten minutes or more... Have a coffee!

InstallAll.ifort may not work for your environment. The you may prepare your own InstallAll.foobar, in which you have to set compilar, linker, compilar options.

When InstallAll.ifort works well, it will show OK! sign finally. (one last test (nio_gwsc) may fail in cases, but usually no problem). The skip following (1) thru (4). (5) is also not necessary.

Let us assume gfortran case. Move to ecalj/lm7K/, then do "make PLATFORM=gfortran LIBMATH=xxx". Then make process start. (LIBMATH= specify BLAS,Lapack, and fftw.) The main makefile is ecalj/lm7K/Makefile, which contains lines

PLATFORM=gfortran #default is PLATFORM=gfortran
...
include MAKEINC/Make.inc.$(PLATFORM)

This means that this Makefile uses ecalj/lm7K/MAKEINC/Make.inc.gfortran as a part of the Makefile. Thus we have to check settings in it in advance to run "make PLATFORM=...". LIBMATH= should contain path to the math libraries, FFTW, BLAS and LAPACK. An example is
LIBMATH="/usr/lib/libfftw3.so.3 /usr/lib/liblapack.so.3gf /usr/lib/libblas.so.3gf"
Compilar options FFLAGS=, FFLAGS_LESS=... can be choosed by your own manner. But usually Make.inc.gfortran works without changes (let me know your changes; I like to include it in ecalj).

Let us think about an ifort case. In this case, we run

make PLATFORM=ifort LIBMATH='-mkl'

There are several MAKEINC/Make.inc.ifort* (not _mpik*) with which we installed to machines. You can choose one of them or you can set your own Make.inc.ifort.* (compilar, compilar options, math library).

Warning messages like ": warning: ignoring old commands for target `/vxcnls.o'" is just because of problem of Makefile. you can neglect this. We will fix it..

Parallel make like

make -j24 PLATFORM=gfortran

can speed it up for multicore machines(24 core in this case). But it stops because of dependency is not well-described in our current Makefile. In such case, repeat it a few times, or repeat it without -j24.

Finally run

make PLATFORM=gfortran install

This just copy required files (binaries and scripts) to your ~/bin. (check it in Makefile). If you like to copy them to /xxx instead of/bin, make with BINDIR=xxx.

(For CMD workshop participants: run

make PLATFORM=ifort.cmd LIBMATH='-mkl' BINDIR=~/bin

• I saw that current ecalj with gfortran4.6 or 4.7 works fine with FFLAGS=-O2, but failed with FFLAGS=-O3. (I am not sure now).
• ifort12 may need FFLAGS=-O1 in MAKEINC/Make.inc.ifort. -->Try InstallAll.ifort12.
• We may need -heap-arrays 100 (when zero, we had a problem in a version of ifort). In cases, -heap-arrays option did not generate working binaries. However, I think "ulimit -s unlimited" before QSGW calculations and so on works OK. So, maybe we don't need -heap-arrays option.
• mpiifort works for liker, but mpif90 did not... (but opposite case maybe). Need to set this in lm7K/MAKEINC/Make.inc.ifort lm7K/MAKEINC/Make.inc.ifort_mpik and fpgw/exec/make.inc.ifort (FC and LK variables).

lmf-MPIK and lmfgw-MPIK are needed for gwsc (srcipt for QSGW). These are k-point parallel version of lmf, and gw driver lmfgw. To make it, do
"make PLATFORM=gfortran_mpik".
For ifort, set PLATFORM=ifort_mpik.
Then Makefile includes ecalj/lm7K/MAKEINC/Make.inc.ifort_mpik. You may need to add -heap-arrays 1 (for large calculations. Because we use large stacksize) to ecalj/lm7K/MAKEINC/Make.inc.ifort_mpi, but I am not so sure about this.

(For CMD workshop participants: run

make PLATFORM=ifort_mpik.cmd LIBMATH='-mkl'

which corresponds to MAKEINC/Make.inc.ifort_mpik.cmd)

*Clean up:
If something wrong. do "make clean" or "make cleanall" and start over. Look into Makefile if you like to know what they do. "make cleanall" removes all *.o *.a modules, and binaries.

• Move binaries to your bin by

make install at ecalj/lm7K. It just moves all requied binaries to your ~/bin. In advance, you have to make your bin, and add ~/bin to your path (e.g. "export PATH=$HOME/bin:$PATH" in .bashrc . Then login again or "source .bashrc")

At ecalj/fpgw/exec/ directory, you have to a softlink make.inc such as

lrwxrwxrwx 1 takao takao 17 Aug 25 13:18 make.inc -> make.inc.gfortran

For each machine you have to prepare your own make.inc.foobar (There are samples. Here is the case of make.inc.ifort.cmd), and do

ln -s make.inc.ifort.cmd make.inc

to make a soft like make.inc -> make.inc.cmd

• Q. What is soft link foo -> bar? A. "foo" is an alias of the file "bar".

Then you have to run

make
make install
make install2

Before this, you have to set blas and lapack in fpge/exec/make.inc. (for ifort, -mkl is enough. LIBMATH= should be the same as that in Make.inc.*. "make install" copy requied files to your ~/bin.

• Caution!: we often see "Segmentation fault"due to stacksize limit (See the size by a command "ulimit -a"). It is needed to run "ulimit -s unimited" in the job-submition script or before running GW jobs.

We have to check whether binaries works fine or not. Move to ecalj/TestInstall. Then type make (with no arguments). It shows help about how to do test. To test all of binaries, just do

make all

All tests may require ~10min or a little more. (nio_gwsc takes ~300sec) In cases, nio_gwsc fails, showing

FAILED: nio_gwsc QPU compared by ./bin/dqpu
PASSED: nio_gwsc diffnum

However, we do not need to care its failure sign. (so nio_gwsc test must be improved...). (numerically small differences).

Help of make (no arguments) at ecalj/TestInstall, shows

make lmall !tests only LDA part.
make gwall !tests only GW part.

• NOTE (nov19 2014 kino):
  In TestInstall/Makefile.define, we define
  LMF=lmf
  LMFP=lmf-MPIK
  (it is possible to use "LMFP=lmf-MPI" instead(for future development). If we set LMFP=$(LMF), tests are done with using lmf, not with using lmf-MPIK.

• NOTE: in principle, repeat make should do nothing when all binaries are correctly generated. However, because of some problem in makefile, you may see some make procedure is repeated. You can neglect it as long as "All test are passed!" is shown in the (4)install test.

In any calculations, we first have to supply crystal structure correctly. To help this, we have a converter between POSCAR(VASP's crystal structure file) and ctrls(that for ecalj). In addition, we have a simple tool to invoke crystal strucrure viewer. It is in \verb+ecalj/Structure/tool/.

In advance, install a viever of crystal structure for POSCAR. Here we use VESTA at http://jp-minerals.org/vesta/. Download it, and expand it to a directory. VESTA can handle kinds of format of crystal structure.

Then make a softlike by

ln -s ~/ecalj/StructureTool/viewvesta.py ~/bin/viewvesta
ln -s ~/ecalj/StructureTool/ctrl2vasp.py ~/bin/ctrl2vasp
ln -s ~/ecalj/StructureTool/vasp2ctrl.py ~/bin/vasp2ctrl

With this procedure we can run command viewvesta, ctrl2vasp, vasp2ctrl from console as long as you have ~/bin/ in the command search path. In my case, .bashrc have a line export PATH=$HOME/bin:$HOME/VESTA-x86_64:$PATH

It depends on your machine. (after editing .bashrc, you have to do "source ~/.bashrc" to reflect changes).

Set the variable of VESTA=, at the begining of ~/ecalj/StructureTool/viewvesta.py to let it know where is VESTA.

Be careful to do version up. It may cause another problem. But it is not so difficult to move it back to original version if you use git. An important things is keeping your changes by yourself. Especially your own Make.inc.* files (see InstalAll.ifort).

cd ecalj
git log

This shows what version you use now.

git diff > gitdiff_backup

This is to save your changes added to the original (to a file git_diff_backup ) for safe. I recommend you do take git diff >foobar as backup.

git stash also move your changes to stash.

git checkout -f
CAUTION!!!: this delete your changes in ecalj/. This recover files controlled by git to the original which was just downloaded.

git pull
This takes all new changes.

I think it is recommended to use

gitk --all

and read this document. Difference can be easily taken, e.g. by >git diff d2281:README 81d27:README (here d2281 and 81d27 are several digits of the begining of its version id).

git show 81d27:README is also useful.

###### Documents of ecalj #####################
We have three documents.

1. GetStarted:
ecalj/Document/Manual/EcaljGetstarted.pdf
Almost written OK.
It gives minimum instruction of ecalj.

2. Usage:
ecalj/Document/Manual/EcaljUsage.pdf
Not yet completed.

3. Details:
Two papers (but not yet submitted) are
  ecalj/Document/Manual/pmttheory17.pdf (for PMT)
  ecalj/Document/Manual/pmtqsgw13.pdf   (for PMT-QSGW)
A little extended versions in japanese are
ecalj/Document/Manual/ecalj_pmttheory17j.pdf (for PMT)
ecalj/Document/Manual/ecalj_pmqsgw13j.pdf   (for PMT-QSGW)
We will unify this...


#######  Usage minimum. (e.g, PMT-QSGW(gwsc) for si)  ###################
Read ecalj/Document/Manual/EcaljGetstarted.pdf
Here is very minimum.
-------------------------------------------
(1) Write structure file ctrls.si by hand 
    (you can have ctrls from POSCAR(VASP) with vasp2ctrl in
    ecalj/StructureTool/.)
(2) conver ctrls.si to ctrl.si by ctrlgen2.py si --nk=6 
   (without argument, it shows help). 
   Then you have default ctrl.si (rename ctrlgen2.ctr.si to ctrl.si). 
(3) Run "lmfa si" to prepare atom.

NOTE: If you like to skip them,  run ./job_materials.py Si at /home/takao/ecalj/MATERIALS.
 >cd Si
 >cp ../syml.si
 >job_band_nspin1 si
This shows you band by LDA.

(4) For PMT-QSGW, make GWinput.tmp by mkGWIN_v2 si.
    Copy GWinput.tmp as GWinput. (you supply three numbers for the
    command mkGIWN_V2.)
(5) Then run a script gwsc, e.g. "gwsc 2 si -np 3" 
    (2+1 iteration with 3 nodes).
(6) To continue calculation do "gwsc 5 si -np 3" again.
    (To start, you need ctrl.si rst.si QGpsi ESEAVR sigm.si)
    When you start from these files, 0th iteration is skipped
   ---thus we have just five iteration.
(7) For band, dos, and pdos plot, 
    we have scripts which almost automatically makes these plot in
    gnuplot. Thus easy to modify these plots at your desposal.
-------------------------------------------

###### PROCASR (VASP format) generator #########
PROCASR mode for lmf (not yet in lmf-MPIK)
Band weight decomposition. 
(Size of circles show the size of components. Superposed on band
plot). See ecalj/MATERIALS/MgO_PROCAR/README.
May19.2014



#####################################################
Usage problems, Q&A error message.

1.Bandplot for FSMOMMETHOD/=0 
Even when you use FSMMOMMETHOD/=0 in GWinput for gwsc, 
yuo need to set FSMOMMETHOD=0 (or comment it out) when you run job_band_nspin2.
[If you run job_band_nspin2 with FSMOMMETHOD/=0, it make a shift 
 (adding bias magnetic field).]

2.Note that ctrlgenM1.py automatically set this for --systype=molecule.
   Then we have 
       TETRA=0
       N=-1  #Negative is the Fermi distribution function W= gives temperature.
       W=0.001 #W=0.001 corresponds to T=157K as shown in console
   In addiiton, FSMOM (n_up-n_down) is needed (FSMOMMETHOD=1)if we
   have magnetic moment.

3. core>evalence message.
   Ecore is grater than Evalence.
   For save, we do not allow this.
   Complare ECORE file and valence levels, shown in log file or
   console output.

4. If you see a error message from lmf (e.g., internally called in the gwsc script).;
  Exit -1 rdsigm: Bloch sum deviates more than allowed tolerance (tol=5e-6)
You have to enlarge RSRNG so that lmf finsh normally.
See FAQ at the bottom of http://titus.phy.qub.ac.uk/packages/LMTO/gw.html
E.g. try RSRNGE=8 for Ni...

5. Back ground charge and fractional Z.
   You can use fractional numbers for ATOM_Z, and also can set
   valence charge by BZ_ZBAK (I removed BZ_VAL).
   You see console out put, e.g,
     "Charges:  valence    19.80000   cores     8.00000   nucleii   -28.00000
        hom background       .20000   deviation from neutrality:      0.00000
   . This is a case with BZ_ZBAK=.2.

   NOTE: at the first iteration, Charges: shows such as
     Charges:  valence     8.00000   cores    20.00000   nucleii   -28.00000
      hom background     0.12300   deviation from neutrality: 0.12300
      because of the initial condition by superposition of atoms. It show
      deviation seems nonzero. But charge should be conserved from the
      next iteration.

6. Not converged in metal. --->mixing may help
For example, if you try metal such as Bi2Sr2CuO6, it may fail at LDA/GGA level.
Then use ITER MIX=A2,b=.2. or something (.2 means it only mix 20% of output to give
new input for next iteration). Then I see convergence. (b is the
mixing parameter.

7. Use PZ or not.
   If spillout of core is not so small (more then 0.05 or something.),
   it is better to use PZ(lo). Treat the core as valecne.
   Bi4d is such a case. Maybe use PZ=0,0,4.9

8.Core treatment 
  See 10.1103/PhysRevB.76.165106 (Eq.35 and after).
  Now I usually not use CORE2 (CORE1 only).

9. ERROR EXIT! rgwina: 2nd wrong l valence
   This may be because you use wrong GWinput.
   Back it up. And run mkGWIN_lmf2 (any n1 n2 n3 is fine).

10. Known bug.
    Error occurs when system is anisotroic such as CuAlTe2. 
    Temporary fix is "Add token NPWPAD=100 in HAM category".
    (guess of used APW fails (more than expected)).
    CuAlTe2,CuGaTe2 cases.

11. Known bug
    a little unstable when metal GGA, especially when we have large
    empty regions.

==========================================
Wanneir Branch now under developing (imported from T.Miyake's Wannier code and H.Kino's).
   A. make at ecalj/fpgw/Wannier/ directory, and do make, and make install. 
      (need to check Makefile first). You first have to install fpgw/exec/ in advance.
   B. Samples are at these directories. 
      MATERIALS/CuMLWFs (small samples),
      MATERIALS/CuMLWF/
      MATERIALS/CuMLWFs/
      MATERIALS/FeMLWF/      
      MATERIALS/NiOMLWF/
      MATERIALS/SrVO3MLWF/
   C. With GWinput and ctrl.*, run 
      >genMLWF
      at these directories.
      In GWinput, we supply settings to generate Wannier funcitons. (Sorry,not documentet yet..)
   D. After genMLWF, do
      >grep Wan lwmatK*
      then compare these with Result.grepWanlwmatK
      These are onsite effective interactions (diagonal part only shown).
      *.xsf are for plotting the Maximally localized Wannier funcitons.
Anyway, documentaion on Wannier is on the way half.
Time consuming part (and also the advantage) is for effective interaction in RPA.
Look into the shell script genMLWF; you can skip last part if you don't need the effective interaction.


--------- this is the end for users -----------


##################################################################
######    LOG and MEMO for developers                 ############
##################################################################
=====
Real TIME and Memory usage command. (H.Kino). rev.jun23.2014
We set comment lines such as
----
!TIME0
  ... here lines of fortran codes ...
!TIME1 "foo bar"
----
in fortran files, e.g., in fpgw/main/hsfp0.sc.m.F.
This reports computational time at the end of output from each node.
(see stdout.{rankID}.* files).

For example, if we see a line (in stdout.*)
>timediff,id= 106 (foo_bar_____)    420.00 sec     4 times
, we used the CPU time 420 sec (roughly =4 x 105 sec) for
the lines between '!TIME0' and '!TIME1 "foo bar"'.

See fpgw/exec/makefile. At its bottom,
gawk replaces '!TIME*' with fortran codes (time_xxx.F are
generated automatically by this procedure).

--
NOTE: !TIME can be nested as
!TIME0
  ... code1 ...
!TIME0
  ... code2 ...
!TIME1 "foobar1"
!TIME1 "foobar2"


=====
1.branch x0mpi_merge (6b5e3, Aug2013): 
contains a try to paralellize x0kf
(imaginary part of dielectric function) more. It just show a
strategy, not efficient yet.

2. With FSMOM, Efermi is not uniquely given in job_band_nspin2*.
It is given by a bndfp-bzwtsf-bzwsf L300 block
   if ((.not. lfill) .or. (metal .and. (nkp .eq. 1))) then
   (bisection method to determine a middle of LUMO and HOMO).
It can give some energy between LUMO and HOMO. 
Small changes of computational condition can give large change. 
But no problem.

3. Line length for fortran; Add .emacs the following three lines.
(add-hook 'fortran-mode-hook
	  '(lambda ()
	     (setq fortran-line-length 132)))



======
2014 nov23:
job_band si -np 4 [options], where
options can be "-vso=1 -vnspin=2" for SO case.
--->Need to add documents to GetStarted.

===
--ssig option (ScaledSigma option).
Need explanation...

======
PDOS: sigm_fbz is required.
(when cp sigm,rst,GWinput ->LDA-like result.
 Then cp sigm_fbz ->it fails.
 Need to make new directory, and copy rst,sigm_fbz.)
And how to check it. (whether 


======
mixbeta:
takao@TT4:~/ecalj/fpgw$ grep mixbeta */*.F
main/hqpe.sc.m.F:      call getkeyvalue("GWinput","mixbeta",beta,default=1d0,status=ret)
mixing parameter on sigm file.
As the default beta is unitiy, mixsigm and mixsigma files are 


=======
Check convergecne on QSGW.
grep rms lqpe*


======
other to DO
(1)
 modify shell script to bash:
 use function. (learn git, push your new package to github).
(2) 
 fortran.
 Read dataflow and data structure.
 Look into gwsc and 

 You can insert your own note in souce code.

Doxygen:

At ecalj/fpgw, run doxygen. Because we have Doxyfile there, we can have doxygen html and pdfs.

--------
efermi.lmf:
this is generated by lmf(lmf-MPI), which is used in the job_band.

----------------
New band plot and effective mass calculation (curvature):
------------
We now read syml.* in lm7K/fp/bndfp.F. Thus job_band is changed
(nspin=1 or nspin=2 is automatically choosed by job_band command).
We do not use plbnd anymore. If necessary, you can modify
"writeband" subrouitne in lm7K/fp/bndfp.F by yourself.

Follow step to get effective mass...
1. 
New syml read label of k point. It is shown by the gnuscript
file "bandplot.glt". Type "gnuplot -p bandplot.isp1.glt" and so on,
when you like to remake band plot.

2. 
New syml allow a special input suitable to determine effective mass
for semiconductor. An example of new syml is (this is a case of GaAs)
------ syml example start ------------------
5  0 0 0   .5 .5  .5    Gamma  L
5  0 0 0    1.  0  0    Gamma  X
5  0 0 0   .75 .75 0    Gamma  K
#############
## resolution qinit qend iqinit iqend  etolv(Ry) etolc(Ry) symllabels
## mass*.spin* contains bands which is evaltop-etolv < eband paste Band005Syml004Spin1.mass Band006Syml004Spin1.mass >temp
>gnuplot -persisit massfit.glt
where, massfit.glt is
----------start of massfit.glt------------------
set grid
set ylabel "mass/mass(electron)"
set xrange [0.00: 0.02537]
f(x) = a+ x**2*(b + x**2*c) 
fit f(x) "temp" u ($6):(($7+$14)/2.0) via a,b, c
plot \
"Band005Syml004Spin1.mass" u ($6):($7) lt 5 pt  5 w lp ti "band= 17",\
"Band006Syml004Spin1.mass" u ($6):($7) lt 6 pt  6 w lp ti "band= 18",\
f(x) 
print 'mass=',f(0)
----------end of massfit------------------
Then I got mhh(111)=0.796 in QSGW.
Note that Syml004 means 4th line of syml file.
("257  0 0 0   .5 .5  .5   1     32     0.1 0.01    Gamma  L" in our example).

An example is at ~/ecalj/MATERIALS/GaAs_so.qsgw.mass
Please look into ./job and run it here. 
(We automatically have three gnuplot figures.
 If we change nk1=nk2=nk3=8, we have different fitting curve).
----
For your convenience, we have dE/dk in the bnd*.spin* files.
This is useful to determine the Fermi surface.
See the efermi.lmf to read the Fermi energy.

--------
1. Set  section.
2. Run gwsigma or 
   Stop sc calculation after dielectric funciton, and run 
   echo 4| mpirun -np 24 hsfp0.
   Then we have SEComg.UP (DN) files, Look for file handle, ifoutsec,
   for the file in fpgw/main/hsfp0.sc.F to see format for the file.
3. Be careful about dw and omg_c.
   (We may not have good accuracy at high energy).
   We calculate weight of imaginary part along imaginary axis.

There is an example MATERIALS/SiSigma/
(To generate accurate Sigma(omega),
  we need to enlarge n1n2n3, and maybe with denser mesh (setting of dw, omg_c).)
============================================================

----------------
See
file:///home/takao/ecalj/Document/BACKUP/MarksOriginalDoc/fp.html#ldaplusu
We need to add lines such as
  IDU= 0 0 2 2 UH= 0 0 0.1 0.632 JH= 0 0 0 0.055
for each SPEC_ATOM, And initial occnum.foobar file.
An example is in ecalj/MATERIALS/GdNldau/

When you just like to generate initial condition for gwsc,
you have to remove (or comment out) IDU before the 1st iteration
with sigm file, because sigm may already can contain LDA+U kind of effect.
(Thus you may need to modify gwsc or stop it at the 0th iteration,
and then remove IDU...)

Caution: 
We need the initial condition file such as occnum.gdn for LDA+U. 
(you may need to set "% real" at the begninig of the file).
Note that definition of spherical harmonics is in ecaljmanual.pdf.
(real harmonics is usual ones used in jobpdos).
Look for the keyword ldau in fp/lmfp.F -> sudmtu.F which read occnum.gdn.