Manual For MBXASPY

User Manual:

Open the PDF directly: View PDF .
Page Count: 10

Download
Open PDF In Browser	View PDF

TUTORIAL FOR MBXASPY
( Many-Body XAS with Python)
Yufeng Liang

Contents
1. Installation
2. Generating pseudopotentials and other auxiliary ;iles
3. Running MBXASPY

Installation
(1) Shirley_XAS
Shirley_XAS is a software package for simulating x-ray absorption spectra using the
excited-state-core-hole (XCH) approach and Shirley’s optimal basis functions for band
structure interpolation. It is modi;ied based Quantum Espresso simulation package and
maintained by David Prendergast’s group at the Molecular Foundry. The Shirley_XAS
provides the Kohn-Sham eigen-energies and wavefunctions as the input for the
determinant formalism of x-ray spectra.
Check out the latest commit on the overlap branch (needs) if you have access to
http://trac-foundry2.lbl.gov/git/Shirley:
git checkout overlap
git pull
make shirley

Or you may do module-load if you have access to local clusters of High-Performance
Computing at Berkeley Lab:
module load shirley_overlap/QE4.3

(2)

MBXASPY

Checkout the code on github:
git clone https://github.com/yufengliang/mbxaspy ~/mbxaspy

From now on you can follow my updates by using “git pull”. You will see many python
source codes with post;ix *.py are cloned to your speci;ied directory.
Python 2.7+ is required.

Generate the pseudos and other auxiliary Diles
To perform a typical core-hole calculation, one needs pseudopotentials for the unexcited
atom and also the core-excited one. Here is a list of codes that are relevant for generating
pseudopotentials and other auxiliary ;iles:
_______________________________________________________________________________________________________
1. Vanderbilt Ultra-Soft Pseudopotential code (based on Fortran, with D. P. Prendergast’s
developments for outputting intermediate data of atomic wavefunctions):
https://github.com/yufengliang/uspp-dgp
This is the main code for generating pseudopotentials.
After git-clone to local and enter the root folder:
cd uspp-dgp/Source
make clean
make

to compile the code.
2. shirley_QE4.3.overlap/upftools/uspp2upf.x (Fortran)
For converting the binary pseudo ;ile into a *.UPF that can be read by Qespresso.
cd shirley_QE4.3.overlap/upftools
make clean
make

3. shirley_QE4.3.overlap/corerepair_dipole/corevalence_position.x (Fortran)
This code is for generating the single-body matrix elements for the position operator.
cd shirley_QE4.3.overlap/corerepair_dipole
make clean
make

4. mbxaspy/sij.py (PYTHON)

_______________________________________________________________________________________________________
If you are only interested in elements that will NOT be core-excited, then the ;irst two will
do.
Here are steps for generating the core-hole pseudos (Using C 1s as an example):
1)

Obtain sample input ;iles from uspp-dgp/Work. In this folder, there are a number of
examples for generating pseudos for some common elements. For example:

cd uspp-dgp/Work/006-C
ls
drwxr-x--drwxr-x--drwxr-x--drwxr-x--drwxr-x---

2
2
2
2
2

yfliang
yfliang
yfliang
yfliang
yfliang

yfliang
yfliang
yfliang
yfliang
yfliang

512
512
4096
4096
512

Dec
Dec
Dec
Dec
Dec

15
15
15
15
15

11:18
11:18
11:28
11:18
11:18

006-C-ca--bm
006-C-gpbe--bm
006-C-gpbe--yufengl
006-C-gpbe-1s1--yufengl
006-C-gpw-n-campos

The naming convention is: “006-C-ca—bm” means the carbon pseudo generated with
functional “ca” (LDA) by the author “bm”.
2) Make a new working directory for your ground-state atom from the sample.
You may:
cp -r 006-C-ca—bm 006-C-pbe-your_name
cd 006-C-pbe-your_name
make clean

If you cannot ;ind the element you want in these samples, pick a similar element to start
with.
3) Edit the input ;iles as needed.
There are two input ;iles: c_ae_s2p2.adat and c_ps.adat. The former is for all-electron
calculation and the latter for pseudizing.
c_ae_s2p2.adat (all-electron)
1
2
3
4
5
6
7
8
9
10

1
0
0
0
3
1.50
-2.4
1.6
1.0d-11
1.0d-09
0.5
C
6.0
0.0
0.0
80.0
6.0
59.0
3
0
100 2.
-20.0
200 2.
-1.0
210 2.
-0.5

ifae,ifpsp,ifprt,ifplw,ilogd (5i5)
80 rlogd,emin,emax,nnt (3f10.5,i5)
0 thresh,tol,damp,maxit (2e10.1,f10.5,i5)
title (a20)
z,xion,exfact (f7.2,2f10.5)
rmax,aasf,bbsf (3f10.5)
ncspvs,irel (2i5)
nnlz,wwnl,ee (i4,f7.3,f14.6)
nnlz,wwnl,ee (i4,f7.3,f14.6)
nnlz,wwnl,ee (i4,f7.3,f14.6)

Instructions for the bolded parts:
line 4: must use *shorthand* notation for element names (case-sensitive !)
line 5: exfact de;ines the types of the functional: 0.0: LDA, 5.0: GGA-PBE (recommended)
line 6, 8-10: edit the z number and atomic shells as needed.
For more information about the all-electron (AE) part, please see:
http://physics.rutgers.edu/~dhv/uspp/uspp-736/Doc/INPUT_AE
c_ps.adat (generating pseudos (GEN))
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

C

0
2
1.50
1.0d-11

1
1
-2.4
1.0d-09

1
2
2
10.0
20.0
3
0
0.8
4
1.0
1.1
1.1
0
0 -1.2
0
0
0.2
1
0 -0.6
1
0
0.2
8 10.0
2
0
0.0
3
8 10.0

3
1.6
0.5
40.0
0.0
2
2
2
2
1

ifae,ifpsp,ifprt,ifplw,ilogd (5i5)
rlogd,emin,emax,nnt (3f10.5,i5)
thresh,tol,damp,maxit (2e10.1,f10.5,i5)
title (a20)
ncores,nvales,nang (3i5)
10.0
besrmax,besemin,besemax,besde (4f10.5)
keyps,ifpcor,rinner (2i5,f10.5)
nbeta,rcloc (i5,f10.5)
rc (3f10.5)
lll,keyee,eeread,iptype (2i5,f10.5,i5)
lll,keyee,eeread,iptype (2i5,f10.5,i5)
lll,keyee,eeread,iptype (2i5,f10.5,i5)
lll,keyee,eeread,iptype (2i5,f10.5,i5)
npf,ptryc (i5,f10.5)
lloc,keyee,eloc,iploctype (2i5,f10.5,i5)
ifqopt,nqf,qtryc (2i5,f10.5)
80
0

line 4: shorthand for title
line 5: ncores - # shells in the core, nvales - # valence shells, nang - #
line 10-13: edit these lines if generating for a different element (instructions as below)
line 16: must be 3 if GGA-PBE (exfact = 5.0 in the AE input.)
More instructions:
http://physics.rutgers.edu/~dhv/uspp/uspp-736/Doc/INPUT_GEN
4) Generate the pseudos:
Run:
make clean
make

../../../Bin/runatom.x c_ae_s2p2.adat c_ae_s2p2.out c_ae_s2p2.ae c_ae_s2p2.atwf c_ae_s2p2.logd
dummy
beginning execution pseudopotential program version 7.3.6
beginning the all electron calculation
completed self-consistent cycle
1 delta = 0.2499440D+01
completed self-consistent cycle
2 delta = 0.1027510D+01
……
completed self-consistent cycle 37 delta = 0.1375697D-08
completed self-consistent cycle 38 delta = 0.7350098D-09
all electron calculation completed
../../../Bin/runatom.x c_ps.adat c_ps.out c_ae_s2p2.ae c_ps.atwf c_ps.logd c_ps.uspp
beginning execution pseudopotential program version 7.3.6
beginning the all electron calculation
completed self-consistent cycle
1 delta = 0.6194983D-10
all electron calculation completed
constructing ibeta =
1

constructing ibeta =
2
constructing ibeta =
3
constructing ibeta =
4
construction of qqq and ddd complete
pseudizing qfuncs for beta= 1
pseudizing qfuncs for beta= 2
pseudizing qfuncs for beta= 3
pseudizing qfuncs for beta= 4
qfunc pseudized
transformation of q and d complete
transformation of qfunc complete
transformation of qfcoef complete
fourier transforming the qfunctions
solving the schroedinger equation
descreening the potential
writing the pseudopotential to file
fourier analysing pseudowavefunctions
computing the logarithmic derivatives
beginning search for possible ghost states
eigensolution in bessel basis with cutoff 20.00
eigensolution in bessel basis with cutoff 30.00
eigensolution in bessel basis with cutoff 40.00

And you should be able to see these ;iles:
-rw-r-----rw-r-----rw-r-----rw-rw----rw-rw----rw-rw----rw-r-----rw-rw----rw-rw----rw-rw----rw-rw----

1
1
1
1
1
1
1
1
1
1
1

yfliang
yfliang
yfliang
yfliang
yfliang
yfliang
yfliang
yfliang
yfliang
yfliang
yfliang

yfliang
2237 Dec 15 13:33
yfliang
1333 Dec 15 13:33
yfliang
652 Dec 15 13:40
yfliang 20572 Dec 15 14:03
yfliang
2040 Dec 15 14:03
yfliang 90430 Dec 15 14:03
yfliang
1104 Dec 15 13:47
yfliang 141328 Dec 15 14:03
yfliang
4080 Dec 15 14:03
yfliang 600089 Dec 15 14:03
yfliang 98088 Dec 15 14:03

Makefile
README
c_ae_s2p2.adat
c_ae_s2p2.ae
c_ae_s2p2.logd
c_ae_s2p2.out
c_ps.adat
c_ps.atwf
c_ps.logd
c_ps.out
c_ps.uspp

c_ae_s2p2.out and c_ps.out are the corresponding output for the c_ae_s2p2.adat and
c_ps.out. You may check errors and warnings in these ;iles if there is any. Intermediate
data can also be found therein.
c_ps.uspp is the binary pseudo ;ile and is not human-readable. Next we convert it into an
UPF.
5) Convert the uspp to UPF
Run uspp2upf.x as follow:
shirley_QE4.3.overlapo/upftools/usppupf.x

and you’ll see:
yfliang@edison09:~/PseudoP/uspp-dgp/Work/006-C/006-C-pbe-yufengl> ~/
shirley_QE4.3.overlap/upftools/uspp2upf.x
Input file > c_ps.uspp
Pseudopotential successfully read
Output PP file in UPF format : c_ps.uspp.UPF
*** PLEASE TEST BEFORE USING!!! ***
review the content of the PP_INFO fields

6) Edit the pseudo.table to change the default pseudo or add a new one
Edit SYMBOL and MASS array if adding an element.
In the pbe block (if [[ -z $1 || $1 = 'pbe' ]])
Edit the PSEUDO to add a new pseudo.
If you are not adding a pseudo for a core-excited atom, that’s it. Below is for adding a new
species of excited atom (update soon …)
*****************************************************************************************
Generate a ground-state and a core-excited pseudo for C routinely using executables in
uspp-dgp. Let’s say they are C.pbe-van.UPF and C.pbe-van-1s1.UPF respectively.
(2)
Generate the initial-state single-body matrix elements and name it as: C.pbe-van.pos
(3)
In the pseudo output ;ile *_ps.out, we have access to transformation waves in the
“transformed atomic waves - davegp” block. Copy every line with digits only in the
block into a ;ile called “valence-gs.dat” for the ground-state, and “valence-x.dat” for the
excited state.
(4)
Run:
python /your/path/to/mbxaspy/sij.py valence-gs.dat valence-x.dat
(5)
Then you’ll see a Sij.dat ;ile in plain text, which is the overlap matrix for the
transformation waves. Rename it as “C.pbe-van-1s1.sij” and place it in the XCH pseudo
library so that mbxaspy can ;ind it when the pseudo “C.pbe-van-1s1.UPF” is provided.
(6)
For sanity checks of the transformation waves, they are plot out to “wave_gs.png” and
“wave_x.png”. Use “eog ;ile_name” to visualize them if on a cluster.
(7)
If
you run sij.py for the same atom, then you will see Sij.dat contains essentially the same
values as Q_int in the atom’s UPF.

Running MBXASPY
Part 1 Extend the previous shirley_xas calculations
(1) Run xas.sh, ref.sh, and ana.sh routinely. You can use the shirley_xas on the
master branch without the many-body XAS development. The new many-body
code is compatible with older calculations.
(2) On the overlap branch, a new script XAS_mbxas.sh is added. Cook up a script
like this to run it:
. $SLURM_SUBMIT_DIR/Input_Block.in
$SHIRLEY_ROOT/scripts/arvid/XAS_mbxas.sh
If the previous xas step has ;inished, this calculation is not expensive anyway
and takes similar time as shirley_xas.x. So you may submit it to the debug
queue in most cases when all Shirley interpolations are done.
This step generates:
*.eigval
eigenvalue ;ile
*.eigvec
eigenvector ;ile < B_i | nk > (transposed)
*.proj
projectors 
*.xmat
single-particle matrix elements 
overlap.dat
the overlap matrix < B_i | ~B_j >
This calculation will be done automatically for all speci;ied excited atoms in
the system and the ground state. These ;iles will appear in the working
directories for the excited atoms. overlap.dat is also produced for GS-to-GS
transformation and ideally it should be an identity matrix.

You can even run mbxas.sh without any one of xas.sh, ref.sh, and ana.sh but
you won’t have access to *.xmat for all excited atoms (;inal states) which are
not needed for many-body XAS, and you don’t have access to energy shifts and
fermi levels.
To automatically schedule all these calculations, you can alternatively checkout
the xas_script on github:
git clone https://github.com/yufengliang/xas_script ~/xas_script
and run:
~/xas_script/setup_mbxas.sh

(on NERSC Edison)
~/xas_script/setup_mbxas_cori.sh
(on NERSC Cori)
to produce all 5 scripts: xas.sh, ref.sh, ana.sh, state.sh, and mbxas.sh.
Part 2 mbxaspy
(1) Cook up an input ;ile (normally called mbxapy.in) like this:

# initial(ground)-state
path_i
=
'XAS/tio2/GS' # path
mol_name_i =
'tio2'
# ;ile pre;ix
#nbnd_i
=
# number of initial orbitals (not used now)
# ;inal-state
path_f
=
'XAS/tio2/O5' # path
#path_f
=
'XAS/tio2/GS' # path
mol_name_f =
'tio2.O05-FCH'
# ;ile pre;ix
#mol_name_f =
'tio2'
# ;ile pre;ix
#nbnd_f
=
# number of ;inal orbitals (not used now)
xas_arg
=
5
gamma_only = True

# number of k points along one direction
#

Using gamma point only
nproc_per_pool =
2
;inal_1p
= True

# number of procs used to process one (spin, k)
# Need one-body ;inal-state spectrum

xi_analysis
= True
#do_paw_correction
= False

# print out an analysis of the xi matrix
#

do PAW corrections or not
spec0_only = False # only want one-body spectra
# spectral
maxfn = 1
I_thr = 1e-3

# calculate up to f^(maxfn) order
# throw away transitions that are below I_thr (fractional) of the strongest one.

ELOW
EHIGH
SIGMA
NENER
ESHIFT_FINAL

= -5
= 35
= 0.6
= 1000
= 517.976

# *** Please note that this is the ;inal ESHIFT

(2) Running mbxaspy
Command line:
Non-anaconda python:
python /your/path/to/mbxaspy/main.py < mbxaspy.in > mbxaspy.out &
Anaconda distribution of python (as on NERSC):
python /your/path/to/mbxaspy/main.py mbxaspy.in > mbxaspy.out &
MPI environment (Anaconda)
srun –n #proc python /your/path/to/mbxaspy/main.py mbxaspy.in >
mbxaspy.out
On NERSC, be sure to unload python 2.7 and load 3.5-anaconda, which is more
stable:
module load python/3.5-anaconda
For the f^(1) term for one (spin, k) tuple, the mbxaspy calculation is actually
quite fast and you can run it on command line. It will take couples of minutes at
most if the system contains < 100 atoms.
Multiple (spin, k) are supported and highly parallelizable but only individual
spectra are printed out at the end because we still don’t know how to combine
them for multiple k-points.
Only one ;inal-state is handled for each mbxaspy calculation. We can do more
using scripts in future.
(3) Check mbxaspy.out for progress and error messages

(4) If the code runs successfully, in the end you will have for each (spin,k) tuple:
spec0_i.dat
one-body initial-state spectrum
spec0_f.dat
one-body ;inal-state spectrum, if xas.sh is done and ;inal-state xmat has been
output.
spec_xas.dat
many-body XAS spectrum up to f^(maxfn) order
spec_xps.dat
many-body XPS spectrum up to f^(maxfn – 1) order
Columns in the spec ;iles are: energy axis, total, x, y, z
(5) To check if the PAW corrections are done correctly, check test_xi_eig.png or
test_xi_eig.dat, which contains all the eigenvalues of \xi and they should not be
larger than 1.0.

♣



---

Source Exif Data:

File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.3
Linearized                      : No
Page Count                      : 10
Title                           : tutorial
Producer                        : Mac OS X 10.12.6 Quartz PDFContext
Creator                         : Pages
Create Date                     : 2017:12:15 22:28:19Z
Modify Date                     : 2017:12:15 22:28:19Z

EXIF Metadata provided by EXIF.tools