Graphite interlayer distance: Difference between revisions

From VASP Wiki
 
(11 intermediate revisions by 2 users not shown)
Line 1: Line 1:
{{Template:Bulk_systems}}
{{Template:Bulk_systems - Tutorial}}


== Task ==
== Task ==


Determine the interlayer distance of graphite in the stacking direction using the method  
In this example you will determine the interlayer distance of graphite in the stacking direction using the method of Tchatchenko and Scheffler to account for van der Waals interactions.
of Tchatchenko and Scheffler to account for van der Waals interactions.
 
Semilocal DFT at the GGA level underestimates long-range dispersion interactions.
This problem causes a bad overestimation of graphite lattice in the stacking direction: 8.84 Å (PBE) vs. 6.71 Å (exp).
 
In this example, the [[Tkatchenko-Scheffler method|dispersion correction method of Tchatchenko and Scheffler]] is used to cope with this problem.


== Input ==
== Input ==
Line 47: Line 51:
  0 0 0
  0 0 0


== Calcualtion ==
== Running this example ==
 
To run this example, execute the <code>run.sh</code> bash-script:
<pre>
#
# To run VASP this script calls $vasp_std
# (or posibly $vasp_gam and/or $vasp_ncl).
# These variables can be defined by sourcing vaspcmd
. vaspcmd 2> /dev/null
 
#
# When vaspcmd is not available and $vasp_std,
# $vasp_gam, and/or $vasp_ncl are not set as environment
# variables, you can specify them here
[ -z "`echo $vasp_std`" ] && vasp_std="mpirun -np 8 /path-to-your-vasp/vasp_std"
[ -z "`echo $vasp_gam`" ] && vasp_gam="mpirun -np 8 /path-to-your-vasp/vasp_gam"
[ -z "`echo $vasp_ncl`" ] && vasp_ncl="mpirun -np 8 /path-to-your-vasp/vasp_ncl"
 
#
# The real work starts here
#


Semilocal DFT at the GGA level underestimates
rm results.dat
long-range dispersion interactions.
This problem causes a bad overestimation of
graphite lattice in the stacking direction
(8.84 A (PBE) vs. 6.71 A (exp)).


In this example, dispersion correction method
for d in 6.5 6.6 6.65 6.7 6.75 6.8 6.9 7.0
of Tchatchenko and Scheffler (PRL 102, 073005
do
(2009)) is used to cope with this problem.
  cat>POSCAR<<!
graphite
1.0
1.22800000 -2.12695839  0.00000000
1.22800000  2.12695839  0.00000000
0.00000000  0.00000000  $d
4
direct
  0.00000000  0.00000000  0.25000000
  0.00000000  0.00000000  0.75000000
  0.33333333  0.66666667  0.25000000
  0.66666667  0.33333333  0.75000000
!
  $vasp_std
  cp OUTCAR OUTCAR.$d
  energy=$(grep "free  ene" OUTCAR.$d|awk '{print $5}')
  echo $d $energy >> results.dat
done
</pre>


Optimal length of the lattice vector c normal to  
The optimal length of the lattice vector ''c'' normal to the stacking direction is determined in a series of single point calculations with varied value of ''c'' (all other degrees of freedom are fixed at their experimental values).  
the stacking direction is determined in a series
of single point calculations with varied value  
of c (all other degrees of freedom are fixed at  
their experimental values).  


The computed c vs. energy dependence  
The computed ''c'' vs. energy dependence is written in the file <code>results.dat</code> and can be visualized e.g. using ''xmgrace''.
is written in the file results.dat and can be  
The optimal value can be obtained using the attached utility (python with numpy or Numeric is needed):
visualized e.g. using xmgrace. The optimal  
value can be obtained using the attached  
utility (python with numpy or Numeric is needed):


<tt>./utilities/fit.py results.dat</tt>
./utilities/fit.py results.dat
 
This should yield:


  200 iterations performed
  200 iterations performed
Line 80: Line 113:
         d0(A):  6.65603352689
         d0(A):  6.65603352689


The computed value of 6.66 A agrees well with  
The computed value of 6.66 &Aring; agrees well with experiment (6.71 &Aring;).
experiment (6.71 A).


Details of implementation of TS in VASP + a number
== Download ==
of tests:
[[Media:GraphiteDistance ts.tgz| graphiteDistance_ts.tgz]]
Bucko et al., PRB 87, 064110 (2013).


== Download ==
{{Template:Bulk_systems}}
[http://www.vasp.at/vasp-workshop/examples/graphiteDistance_ts.tgz graphiteDistance_ts.tgz]
----
[[VASP_example_calculations|To the list of examples]] or to the [[The_VASP_Manual|main page]]


[[Category:Examples]]
[[Category:Examples]]

Latest revision as of 13:25, 14 November 2019

Task

In this example you will determine the interlayer distance of graphite in the stacking direction using the method of Tchatchenko and Scheffler to account for van der Waals interactions.

Semilocal DFT at the GGA level underestimates long-range dispersion interactions. This problem causes a bad overestimation of graphite lattice in the stacking direction: 8.84 Å (PBE) vs. 6.71 Å (exp).

In this example, the dispersion correction method of Tchatchenko and Scheffler is used to cope with this problem.

Input

POSCAR

graphite
1.0
1.22800000 -2.12695839  0.00000000
1.22800000  2.12695839  0.00000000
0.00000000  0.00000000  7.0
4
direct
   0.00000000  0.00000000  0.25000000
   0.00000000  0.00000000  0.75000000
   0.33333333  0.66666667  0.25000000
   0.66666667  0.33333333  0.75000000

INCAR

IVDW = 20           
LVDW_EWALD =.TRUE. 
NSW = 1 
IBRION = 2
ISIF = 4
PREC = Accurate
EDIFFG = 1e-5
LWAVE = .FALSE.
LCHARG = .FALSE.
ISMEAR = -5
SIGMA = 0.01
EDIFF = 1e-6
ALGO = Fast
NPAR = 2

KPOINTS

Monkhorst Pack
0
gamma
16 16 8
0 0 0

Running this example

To run this example, execute the run.sh bash-script:

#
# To run VASP this script calls $vasp_std
# (or posibly $vasp_gam and/or $vasp_ncl).
# These variables can be defined by sourcing vaspcmd
. vaspcmd 2> /dev/null

#
# When vaspcmd is not available and $vasp_std,
# $vasp_gam, and/or $vasp_ncl are not set as environment
# variables, you can specify them here
[ -z "`echo $vasp_std`" ] && vasp_std="mpirun -np 8 /path-to-your-vasp/vasp_std"
[ -z "`echo $vasp_gam`" ] && vasp_gam="mpirun -np 8 /path-to-your-vasp/vasp_gam"
[ -z "`echo $vasp_ncl`" ] && vasp_ncl="mpirun -np 8 /path-to-your-vasp/vasp_ncl"

#
# The real work starts here
#

rm results.dat

for d in 6.5 6.6 6.65 6.7 6.75 6.8 6.9 7.0
do
  cat>POSCAR<<!
graphite
1.0
1.22800000 -2.12695839  0.00000000
1.22800000  2.12695839  0.00000000
0.00000000  0.00000000  $d
4
direct
   0.00000000  0.00000000  0.25000000
   0.00000000  0.00000000  0.75000000
   0.33333333  0.66666667  0.25000000
   0.66666667  0.33333333  0.75000000
!
  $vasp_std
  cp OUTCAR OUTCAR.$d
  energy=$(grep "free  ene" OUTCAR.$d|awk '{print $5}')
  echo $d $energy >> results.dat 
done

The optimal length of the lattice vector c normal to the stacking direction is determined in a series of single point calculations with varied value of c (all other degrees of freedom are fixed at their experimental values).

The computed c vs. energy dependence is written in the file results.dat and can be visualized e.g. using xmgrace. The optimal value can be obtained using the attached utility (python with numpy or Numeric is needed):

./utilities/fit.py results.dat

This should yield:

200 iterations performed
Ch-square: 4.30305519481e-09
---------

       E0(eV):         -37.433456779
       d0(A):  6.65603352689

The computed value of 6.66 Å agrees well with experiment (6.71 Å).

Download

graphiteDistance_ts.tgz