Contents

1 Introduction to this book

2 Introduction to DFT
2.1 Background . . . . . . . . . . . . . . . . . . . . . .
2.2 Exchange correlation functionals . . . . . . . . . .
2.3 Basis sets . . . . . . . . . . . . . . . . . . . . . . .
2.4 Pseudopotentials . . . . . . . . . . . . . . . . . . .
2.5 Fermi Temperature and band occupation numbers
2.6 Spin polarization and magnetism . . . . . . . . . .
2.7 TODO Recommended reading . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

3 Molecules
3.1 Defining and visualizing molecules . . . . . . . . . . . . .
3.1.1 From scratch . . . . . . . . . . . . . . . . . . . . .
3.1.2 Reading other data formats into a calculation . . .
3.1.3 Predefined molecules . . . . . . . . . . . . . . . . .
3.1.4 Combining Atoms objects . . . . . . . . . . . . . .
3.2 Simple properties . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Getting cartesian positions . . . . . . . . . . . . .
3.2.2 Molecular weight and molecular formula . . . . . .
3.2.3 Center of mass . . . . . . . . . . . . . . . . . . . .
3.2.4 Moments of inertia . . . . . . . . . . . . . . . . . .
3.2.5 Computing bond lengths and angles . . . . . . . .
3.3 Simple properties that require single computations . . . .
3.3.1 Energy and forces . . . . . . . . . . . . . . . . . .
3.3.2 Visualizing electron density . . . . . . . . . . . . .
3.3.3 Dipole moments . . . . . . . . . . . . . . . . . . .
3.3.4 The density of states (DOS) . . . . . . . . . . . . .
3.3.5 Atom-projected density of states on molecules . . .
3.3.6 Electrostatic potential . . . . . . . . . . . . . . . .
3.3.7 Bader analysis . . . . . . . . . . . . . . . . . . . .
3.4 Geometry optimization . . . . . . . . . . . . . . . . . . . .
3.4.1 Manual determination of a bond length . . . . . .
3.4.2 Automatic geometry optimization with VASP . . .
3.4.3 Relaxation of a water molecule . . . . . . . . . . .
3.5 Vibrational frequencies . . . . . . . . . . . . . . . . . . . .
3.5.1 Manual calculation of vibrational frequency . . . .
3.5.2 Automated vibrational calculations . . . . . . . . .
3.6 Simulated infrared spectra . . . . . . . . . . . . . . . . . .
3.7 Thermochemical properties of molecules . . . . . . . . . .
3.8 Molecular reaction energies . . . . . . . . . . . . . . . . .
3.8.1 O2 dissociation . . . . . . . . . . . . . . . . . . . .
3.8.2 Water gas shift example . . . . . . . . . . . . . . .
3.8.3 Temperature dependent water gas shift equilibrium
3.9 Molecular reaction barriers . . . . . . . . . . . . . . . . .
3.9.1 Get initial and final states . . . . . . . . . . . . . .
3.9.2 Run band calculation . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

7
8
9
9
10
11
11
12

. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
constant
. . . . . .
. . . . . .
. . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

12
12
12
15
16
22
22
23
24
25
25
26
29
29
34
36
38
39
40
42
45
45
47
48
49
49
50
52
56
58
58
70
75
80
80
81

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

4 Bulk systems
4.1 Defining and visualizing bulk systems . . . . . . . . . . . . . . . . . . . . . . .
4.1.1 Built-in functions in ase . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.2 Using http://materialsproject.org . . . . . . . . . . . . . . . . . . .
4.2 Computational parameters that are important for bulk structures . . . . . . . .
4.2.1 k-point convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 Effect of SIGMA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 The number of bands . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Determining bulk structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 fcc/bcc crystal structures . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2 Optimizing the hcp lattice constant . . . . . . . . . . . . . . . . . . . .
4.3.3 Complex structures with internal degrees of freedom . . . . . . . . . . .
4.3.4 Effect of XC on bulk properties . . . . . . . . . . . . . . . . . . . . . . .
4.4 Cohesive energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 TODO Elastic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1 Al elastic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.2 Fe elastic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6 Bulk thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7 Effect of pressure on phase stability . . . . . . . . . . . . . . . . . . . . . . . .
4.8 Bulk reaction energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.1 Alloy formation energies . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8.2 Metal oxide oxidation energies . . . . . . . . . . . . . . . . . . . . . . .
4.9 Bulk density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.10 Atom projected density of states . . . . . . . . . . . . . . . . . . . . . . . . . .
4.10.1 Effect of RWIGS on ADOS . . . . . . . . . . . . . . . . . . . . . . . . .
4.11 Band structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.11.1 create example showing band dispersion with change in lattice constant
4.12 Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.12.1 Determining if a magnetic solution is energetically favorable . . . . . . .
4.12.2 Antiferromagnetic spin states . . . . . . . . . . . . . . . . . . . . . . . .
4.12.3 TODO NiO-FeO formation energies with magnetism . . . . . . . . . . .
4.13 TODO phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.14 TODO solid state NEB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

82
82
82
87
90
90
91
92
96
96
99
100
105
107
109
109
110
112
113
116
116
121
125
127
129
131
134
135
135
136
136
136
136

5 Surfaces
5.1 Surface structures . . . . . . . . . . . . . . . . . . .
5.1.1 Simple surfaces . . . . . . . . . . . . . . . . .
5.1.2 Vicinal surfaces . . . . . . . . . . . . . . . . .
5.2 Surface calculation parameters . . . . . . . . . . . .
5.3 Surface relaxation . . . . . . . . . . . . . . . . . . .
5.4 Surface reconstruction . . . . . . . . . . . . . . . . .
5.4.1 Au(110) missing row reconstruction . . . . .
5.4.2 Ag(110) missing row reconstruction . . . . .
5.4.3 Cu(110) missing row reconstruction . . . . .
5.5 Work function . . . . . . . . . . . . . . . . . . . . . .
5.6 Surface energy . . . . . . . . . . . . . . . . . . . . .
5.6.1 Advanced topics in surface energy . . . . . .
5.7 Dipole correction . . . . . . . . . . . . . . . . . . . .
5.7.1 Slab with no dipole correction . . . . . . . . .
5.7.2 Slab with a dipole correction . . . . . . . . .
5.7.3 Comparing no dipole correction with a dipole
5.8 Adsorption energies . . . . . . . . . . . . . . . . . . .
5.8.1 Simple estimate of the adsorption energy . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

154
154
154
155
158
158
162
162
164
166
167
168
170
170
170
170
171
172
172

. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
correction
. . . . . .
. . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

5.9
5.10

5.11
5.12
5.13

5.8.2 Coverage dependence . . . . . . . . . . . . .

5.8.3 Effect of adsorption on the surface energy . .
Adsorbate vibrations . . . . . . . . . . . . . . . . . .
5.9.1 Vibrations of the bridge site . . . . . . . . . .
Surface Diffusion barrier . . . . . . . . . . . . . . . .
5.10.1 Standard nudged elastic band method . . . .
5.10.2 Climbing image NEB . . . . . . . . . . . . .
5.10.3 Using vibrations to confirm a transition state
TODO Diffusion rates with transition state theory .
TODO Effects of electric fields on adsorbates . . . .
TODO Simulating STM images . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

181
183
184
186
186
186
187
188
188
188
189

6 Atomistic thermodynamics
199
6.1 Bulk phase stability of oxides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
6.2 Effect on adsorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.3 Atomistic therodynamics and multiple reactions . . . . . . . . . . . . . . . . . . . . . . . . 207
7 Advanced electronic structure methods
7.1 DFT+U . . . . . . . . . . . . . . . . . . . . . . . . .
7.1.1 Metal oxide oxidation energies with DFT+U
7.2 Hybrid functionals . . . . . . . . . . . . . . . . . . .
7.2.1 FCC Ni DOS . . . . . . . . . . . . . . . . . .
7.3 TODO DFT+D . . . . . . . . . . . . . . . . . . . .
7.3.1 PBE . . . . . . . . . . . . . . . . . . . . . . .
7.3.2 DFT-D2 . . . . . . . . . . . . . . . . . . . . .
7.3.3 Advanced vdW-DF functionals . . . . . . . .
7.4 ELF . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.5 TODO Charge partitioning schemes . . . . . . . . .
7.6 TODO Modeling Core level shifts . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

8 Acknowledgments

208
208
208
212
212
213
214
215
216
216
217
217
220

9 Appendices
9.1 Recipes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.1 Modifying Atoms by deleting atoms . . . . . . . . . . . . .
9.1.2 Advanced tagging . . . . . . . . . . . . . . . . . . . . . . .
9.1.3 Using units in ase . . . . . . . . . . . . . . . . . . . . . . .
9.1.4 Extracting parts of an array . . . . . . . . . . . . . . . . . .
9.1.5 Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.6 Curve fitting . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.7 Nonlinear curve fitting . . . . . . . . . . . . . . . . . . . . .
9.1.8 Nonlinear curve fitting by direct least squares minimization
9.1.9 Nonlinear curve fitting with confidence intervals . . . . . .
9.1.10 Interpolation with splines . . . . . . . . . . . . . . . . . . .
9.1.11 Interpolation in 3D . . . . . . . . . . . . . . . . . . . . . . .
9.1.12 Reading and writing data . . . . . . . . . . . . . . . . . . .
9.1.13 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.14 Numerical differentiation . . . . . . . . . . . . . . . . . . .
9.1.15 NetCDF files . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.16 Python modules . . . . . . . . . . . . . . . . . . . . . . . .
9.1.17 Writing and reading Excel files . . . . . . . . . . . . . . . .
9.1.18 TODO making movies . . . . . . . . . . . . . . . . . . . . .
9.2 Computational geometry . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

220
220
220
220
222
223
224
225
226
227
228
230
231
235
237
238
242
243
244
244
245

9.3

9.4

9.2.1 Changing coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

9.2.2 Simple distances, angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.2.3 Unit cell properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.2.4 d-spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
Equations of State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.3.1 Birch-Murnaghan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.3.2 Murnaghan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.3.3 Birch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.3.4 The Anton-Schmidt Equation of state 104 . . . . . . . . . . . . . . . . . . . . . . . 250
9.3.5 Fitting data to these equations of state . . . . . . . . . . . . . . . . . . . . . . . . 250
Miscellaneous jasp/VASP tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
9.4.1 Installing jasp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
9.4.2 Using a special setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
9.4.3 Running jasp in parallel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
9.4.4 Running multiple instances of jasp in parallel . . . . . . . . . . . . . . . . . . . . . 255
9.4.5 Exporting data json, xml, python, sqlite . . . . . . . . . . . . . . . . . . . . . . . . 259
9.4.6 Recommended values for ENCUT and valence electrons for different POTCAR files 266

10 Python
268
10.1 easy_install as a user . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.2 pip as a user . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.3 Integer division math gotchas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
11 References

269

12 GNU Free Documentation License

276

13 Index

285

Copyright 20122015\ John Kitchin

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU
Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation;
with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is
included in the section entitled "GNU Free Documentation License".

Introduction to this book

This book serves two purposes: 1) to provide worked examples of using DFT to model materials properties, and 2) to provide references to more advanced treatments of these topics in the literature. It is
not a definitive reference on density functional theory. Along the way to learning how to perform the
calculations, you will learn how to analyze the data, make plots, and how to interpret the results. This
book is very much "recipe" oriented, with the intention of giving you enough information and knowledge
to start your research. In that sense, many of the computations are not publication quality with respect
to convergence of calculation parameters.
You will read a lot of python code in this book, as the comprehension of code is an integral part of
its design. I believe that computational work should always be scripted. As a result, a written record of
everything you have done will be available, allowing you to reproduce your code or report the method
of its execution exactly at a later time.
This book makes heavy use of many computational tools including:
Python
Module index
Atomic Simulation Environment (ase)
numpy
scipy
matplotlib
emacs
org-mode This book is written in org-mode, and is best read in emacs in org-mode. This
format provides clickable links, easy navigation, syntax highlighting, as well as the ability to
interact with the tables and code. The book is also available in PDF.
git This book is available at https://github.com/jkitchin/dft-book
jasp
mod:jasp is a sophisticated wrapper to the mod:ase.calculators.vasp python interface to the VASP calculator. It was written by me to facilitate writing this book and to develop the best possible way to
run DFT calculations. The best way to learn to use jasp is from this book. It probably requires the
latest svn version of ase to work since I have been adding new functionality to mod:ase.calculators.vasp
as mod:jasp is developed. mod:jasp is available at https://github.com/jkitchin/jasp.
The goal in writing mod:jasp was to create a computational environment where you write one script
to create, run and analyze your calculations. The code is smart, and submits jobs to a queue system
when required, does not submit jobs more than once, allows many jobs to run in parallel, and gives you
answers when available. mod:jasp takes care of changing into VASP calculation directories, and then
changing back to where you started from. mod:jasp has an interface to sqlite, and to version control
(git) software. mod:jasp has some features that mod:ase.calculators.vasp does not have yet. They may
get incorporated into ase one day.
jaspsum is a utility program provided in jasp which prints a pretty representation of the state of a
calculation, plots the relaxation trajectory, or prints a code representation of a VASP directory.
It is not critical that you use mod:jasp. You should be able to copy the actual VASP input files
(which were generated by jasp) from a directory and simply run vasp. mod:jasp is mostly compatible
with mod:ase.calculators.vasp, so you can replace this kind of code:
1
2

with jasp(directory, lotsofkeywords, atoms=atoms) as calc:

#do stuff

Open the python script (dft-scripts/script-1.py).

with code similar to this:
1
2
3
4

CWD = os.getcwd()
os.chdir(directory)
calc=Vasp(lotsofkeywords)
atoms.set_calculator(calc)

5
6

try:

#do stuff
finally:
os.chdir(CWD)

8
9

Open the python script (dft-scripts/script-2.py).

The DFT code used primarily in this book is VASP.
VASP wiki
VASP Manual
Similar code would be used for other calculators, e.g. GPAW, Jacapo, etc. . . you would just have to
import the python modules for those codes, and replace the code that defines the calculator.
Exercise 1.1
Review all the hyperlinks in this chapter.

Introduction to DFT

A comprehensive overview of DFT is beyond the scope of this book, as excellent reviews on these subjects
are readily found in the literature, and are suggested reading in the following paragraph. Instead, this
chapter is intended to provide a useful starting point for a non-expert to begin learning about and using
DFT in the manner used in this book. Much of the information presented here is standard knowledge
among experts, but a consequence of this is that it is rarely discussed in current papers in the literature.
A secondary goal of this chapter is to provide new users with a path through the extensive literature
available and to point out potential difficulties and pitfalls in these calculations.
A modern and practical introduction to density functional theory can be found in Sholl and Steckel. 1
A fairly standard textbook on DFT is the one written by Parr and Yang. 2 The Chemists Guide to
DFT 3 is more readable and contains more practical information for running calculations, but both of
these books focus on molecular systems. The standard texts in solid state physics are by Kittel 4 and
Ashcroft and Mermin. 5 Both have their fine points, the former being more mathematically rigorous and
the latter more readable. However, neither of these books is particularly easy to relate to chemistry. For
this, one should consult the exceptionally clear writings of Roald Hoffman, 6,7 and follow these with the
work of Nrskov and coworkers. 8,9
In this chapter, only the elements of DFT that are relevant to this work will be discussed. An
excellent review on other implementations of DFT can be found in Reference 10 , and details on the
various algorithms used in DFT codes can be found in Refs. 11,12 .
One of the most useful sources of information has been the dissertations of other students, perhaps
because the difficulties they faced in learning the material are still fresh in their minds. Thomas Bligaard,
a coauthor of Dacapo, wrote a particularly relevant thesis on exchange/correlation functionals 13 and a
dissertation illustrating the use of DFT to design new alloys with desirable thermal and mechanical
properties. 14 The Ph.D. thesis of Ari Seitsonen contains several useful appendices on k-point setups,
and convergence tests of calculations, in addition to a thorough description of DFT and analysis of
calculation output. 15 Finally, another excellent overview of DFT and its applications to bimetallic alloy
phase diagrams and surface reactivity is presented in the PhD thesis of Robin Hirschl. 16

2.1

Background

In 1926, Erwin Schr\"odinger published the first accounts of his now famous wave equation. 17 He later
shared the Nobel prize with Paul A. M. Dirac in 1933 for this discovery. Schr\"odingers wave function
seemed extremely promising, as it contains all of the information available about a system. Unfortunately,
most practical systems of interest consist of many interacting electrons, and the effort required to find
solutions to Schr\"odingers equation increases exponentially with the number of electrons, limiting this
approach to systems with a small number of relevant electrons, N . O(10). 18 Even if this rough estimate
is off by an order of magnitude, a system with 100 electrons is still very small, for example, two Ru atoms
if all the electrons are counted, or perhaps ten Pt atoms if only the valence electrons are counted. Thus,
the wave function method, which has been extremely successful in studying the properties of small
molecules, is unsuitable for studies of large, extended solids. Interestingly, this difficulty was recognized
by Dirac as early as 1929, when he wrote "The underlying physical laws necessary for the mathematical
theory of a large part of physics and the whole of chemistry are thus completely known, and the difficulty
is only that the application of these laws leads to equations much too complicated to be soluble.". 19
In 1964, Hohenberg and Kohn showed that the ground state total energy of a system of interacting
electrons is a unique functional of the electron density. 20 By definition, a function returns a number
when given a number. For example, in f (x) = x2 , f (x) is the function, and itR equals four when

x = 2. A functional returns a number when given a function. Thus, in g(f (x)) = 0 f (x)dx, g(f (x))
is the functional, and it is equal to two when f (x) = sin(x). Hohenberg and Kohn further identified
a variational principle that appeared to reduce the problem of finding the ground state energy of an
electron gas in an external potential (i.e., in the presence of ion cores) to that of the minimization of
a functional of the three-dimensional density function. Unfortunately, the definition of the functional
involved a set of 3N-dimensional trial wave functions.
In 1965, Kohn and Sham made a significant breakthrough when they showed that the problem of
many interacting electrons in an external potential can be mapped exactly to a set of noninteracting
electrons in an effective external potential. 21 This led to a set of self-consistent, single particle equations
known as the Kohn-Sham (KS) equations:

1

2

2 + veff (r) -_j _j (r)=0, (1)
with
v
R

_eff(r) = v(r)+ { n(r0 )}{|rr0 |}dr0 +v _xc(r), (2)

where v(r) is the external potential and vxc (r) is the exchange-correlation potential, which depends
on the entire density function. Thus, the density needs to be known in order to define the effective
potential so that Eq. (2.1) can be solved. j (r) corresponds to the j th KS orbital of energy j .
The ground state density is given by:
n(r) =
PN
j=1 |_j(r)|2 (3)
To solve Eq. (2.1) then, an initial guess is used for j (r) which is used to generate Eq. (2.1), which
is subsequently used in Eq. (2.1). This equation is then solved for j (r) iteratively until the j (r) that
result from the solution are the same as the j (r) that are used to define the equations, that is, the
solutions are self-consistent. Finally, the ground state energy is given by:
E=

R
R
P
_j$\epsilon$_j + Exc [n(r)]- vxc (r) n(r) dr -1 2 {{

n(r)n(r0 )}{|rr0 |}dr0 r},(4)

where Exc [n(r)] is the exchange-correlation energy functional. Walter Kohn shared the Nobel prize in
Chemistry in 1998 for this work. 18 The other half of the prize went to John Pople for his efforts in wave
function based quantum mechanical methods. 22 Provided the exchange-correlation energy functional is
known, Eq. (2.1) is exact. However, the exact form of the exchange-correlation energy functional is not
known, thus approximations for this functional must be used.

2.2

Exchange correlation functionals

The two main types of exchange/correlation functionals used in DFT are the local density approximation (LDA) and the generalized gradient approximation (GGA). In the LDA, the exchange-correlation
functional is defined for an electron in a uniform electron gas of density n. 21 It is exact for a uniform electron gas, and is anticipated to be a reasonable approximation for slowly varying densities. In molecules
and solids, however, the density tends to vary substantially in space. Despite this, the LDA has been
very successfully used in many systems. It tends to predict overbonding in both molecular and solid
systems, 23 and it tends to make semiconductor systems too metallic (the band gap problem). 24
The generalized gradient approximation includes corrections for gradients in the electron density,
and is often implemented as a corrective function of the LDA. The form of this corrective function, or
"exchange enhancement" function determines which functional it is, e.g. PBE, RPBE, revPBE, etc. 25
In this book the PBE GGA functional is used the most. N{}rskov and coworkers have found that the
RPBE functional gives superior chemisorption energies for atomic and molecular bonding to surfaces,
but that it gives worse bulk properties, such as lattice constants compared to experimental data. 25
Finally, there are increasingly new types of functionals in the literature. The so-called hybrid functionals, such as B3LYP, are more popular with gaussian basis sets (e.g. in Gaussian), but they are
presently inefficient with planewave basis sets. None of these other types of functionals were used in this
work. For more details see Chapter 6 in Ref. 3 and Thomas Bligaards thesis on exchange and correlation
functionals. 13

2.3

Basis sets

Briefly, VASP utilizes planewaves as the basis set to expand the Kohn-Sham orbitals. In a periodic solid,
one can use Blochs theorem to show that the wave function for an electron can be expressed as the
product of a planewave and a function with the periodicity of the lattice: 5
nk (r) = exp(ik r)unk (r)

(5)

where r is a position vector, and k is a so-called wave vector that will only have certain allowed values
defined by the size of the unit cell. Blochs theorem sets the stage for using planewaves as a basis set,
because it suggests a planewave character of the wave function. If the periodic function unk (r) is also
expanded in terms of planewaves determined by wave vectors of the reciprocal lattice vectors, G, then
the wave function can be expressed completely in terms of a sum of planewaves: 11
X
i (r) =
ci,k+G exp(i(k + G) r).
(6)
G

where ci,k+G are now coefficients that can be varied to determine the lowest energy solution. This
also converts Eq. (2.1) from an integral equation to a set of algebraic equations that can readily be
solved using matrix algebra.
In aperiodic systems, such as systems with even one defect, or randomly ordered alloys, there is
no periodic unit cell. Instead one must represent the portion of the system of interest in a supercell,
which is then subjected to the periodic boundary conditions so that a planewave basis set can be used.
It then becomes necessary to ensure the supercell is large enough to avoid interactions between the
defects in neighboring supercells. The case of the randomly ordered alloy is virtually hopeless as the
energy of different configurations will fluctuate statistically about an average value. These systems were
9

not considered in this work, and for more detailed discussions the reader is referred to Ref. 26 . Once a
supercell is chosen, however, Blochs theorem can be applied to the new artificially periodic system.
To get a perfect expansion, one needs an infinite number of planewaves. Luckily, the coefficients of
the planewaves must go to zero for high energy planewaves, otherwise the energy of the wave function
would go to infinity. This provides justification for truncating the planewave basis set above a cutoff
energy. Careful testing of the effect of the cutoff energy on the total energy can be done to determine
a suitable cutoff energy. The cutoff energy required to obtain a particular convergence precision is also
element dependent, shown in Table 1. It can also vary with the "softness" of the pseudopotential. Thus,
careful testing should be done to ensure the desired level of convergence of properties in different systems.
Table 1 refers to convergence of total energies. These energies are rarely considered directly, it is usually
differences in energy that are important. These tend to converge with the planewave cutoff energy much
more quickly than total energies, due to cancellations of convergence errors. In this work, 350 eV was
found to be suitable for the H adsorption calculations, but a cutoff energy of 450 eV was required for O
adsorption calculations.
Table 1: Planewave cutoff energies (in eV) required for different convergence precisions for different
elements.
Precision Low High
Mo
168
293
O
300
520
O_sv
1066 1847
Blochs theorem eliminates the need to calculate an infinite number of wave functions, because there
are only a finite number of electrons in the unit (super) cell. However, there are still an infinite number
of discrete k points that must be considered, and the energy of the unit cell is calculated as an integral
over these points. It turns out that wave functions at k points that are close together are similar, thus
an interpolation scheme can be used with a finite number of k points. This also converts the integral
used to determine the energy into a sum over the k points, which are suitably weighted to account for
the finite number of them. There will be errors in the total energy associated with the finite number of
k, but these can be reduced and tested for convergence by using higher k-point densities. An excellent
discussion of this for aperiodic systems can be found in Ref. 26 .
The most common schemes for generating k points are the Chadi-Cohen scheme, 27 and the MonkhorstPack scheme. 28 The use of these k point setups amounts to an expansion of the periodic function in
reciprocal space, which allows a straight-forward interpolation of the function between the points that is
more accurate than with other k point generation schemes. 28

2.4

Pseudopotentials

The core electrons of an atom are computationally expensive with planewave basis sets because they are
highly localized. This means that a very large number of planewaves are required to expand their wave
functions. Furthermore, the contributions of the core electrons to bonding compared to those of the
valence electrons is usually negligible. In fact, the primary role of the core electron wave functions is to
ensure proper orthogonality between the valence electrons and core states. Consequently, it is desirable
to replace the atomic potential due to the core electrons with a pseudopotential that has the same effect on the valence electrons. 29 There are essentially two kinds of pseudopotentials, norm-conserving soft
pseudopotentials 29 and Vanderbilt ultrasoft pseudopotentials. 30 In either case, the pseudopotential function is generated from an all-electron calculation of an atom in some reference state. In norm-conserving
pseudopotentials, the charge enclosed in the pseudopotential region is the same as that enclosed by the
same space in an all-electron calculation. In ultrasoft pseudopotentials, this requirement is relaxed and
charge augmentation functions are used to make up the difference. As its name implies, this allows a
"softer" pseudopotential to be generated, which means fewer planewaves are required to expand it.
The pseudopotentials are not unique, and calculated properties depend on them. However, there are

10

standard methods for ensuring the quality and transferability (to different chemical environments) of the
pseudopotentials. 31
TODO PAW description
VASP provides a database of PAW potentials. 32,33

2.5

Fermi Temperature and band occupation numbers

At absolute zero, the occupancies of the bands of a system are well-defined step functions; all bands
up to the Fermi level are occupied, and all bands above the Fermi level are unoccupied. There is a
particular difficulty in the calculation of the electronic structures of metals compared to semiconductors
and molecules. In molecules and semiconductors, there is a clear energy gap between the occupied
states and unoccupied states. Thus, the occupancies are insensitive to changes in the energy that occur
during the self-consistency cycles. In metals, however, the density of states is continuous at the Fermi
level, and there are typically a substantial number of states that are close in energy to the Fermi level.
Consequently, small changes in the energy can dramatically change the occupation numbers, resulting in
instabilities that make it difficult to converge to the occupation step function. A related problem is that
the Brillouin zone integral (which in practice is performed as a sum over a finite number of k points)
that defines the band energy converges very slowly with the number of k points due to the discontinuity
in occupancies in a continuous distribution of states for metals. 12,34 The difficulty arises because the
temperature in most DFT calculations is at absolute zero. At higher temperatures, the DOS is smeared
across the Fermi level, resulting in a continuous occupation function over the distribution of states. A
finite-temperature version of DFT was developed, 35 which is the foundation on which one solution to this
problem is based. In this solution, the step function is replaced by a smoothly varying function such as
the Fermi-Dirac function at a small, but non-zero temperature. 12 The total energy is then extrapolated
back to absolute zero.

2.6

Spin polarization and magnetism

There are two final points that need to be discussed about these calculations, spin polarization and
dipole corrections. Spin polarization is important for systems that contain net spin. For example, iron,
cobalt and nickel are magnetic because they have more electrons with spin "up" than spin "down" (or
vice versa). Spin polarization must also be considered in atoms and molecules with unpaired electrons,
such as hydrogen and oxygen atoms, oxygen molecules and radicals. For example, there are two spin
configurations for an oxygen molecule, the singlet state with no unpaired electrons, and the triplet state
with two unpaired electrons. The oxygen triplet state is lower in energy than the oxygen singlet state,
and thus it corresponds to the ground state for an oxygen atom. A classically known problem involving
spin polarization is the dissociation of a hydrogen molecule. In this case, the molecule starts with no
net spin, but it dissociates into two atoms, each of which has an unpaired electron. See section 5.3.5 in
Reference 3 for more details on this.
In VASP, spin polarization is not considered by default; it must be turned on, and an initial guess
for the magnetic moment of each atom in the unit cell must be provided (typically about one Bohrmagneton per unpaired electron). For Fe, Co, and Ni, the experimental values are 2.22, 1.72, and 0.61
Bohr-magnetons, respectively 4 and are usually good initial guesses. See Reference 31 for a very thorough
discussion of the determination of the magnetic properties of these metals with DFT. For a hydrogen
atom, an initial guess of 1.0 Bohr-magnetons (corresponding to one unpaired electron) is usually good.
An oxygen atom has two unpaired electrons, thus an initial guess of 2.0 Bohr-magnetons should be used.
The spin-polarized solution is sensitive to the initial guess, and typically converges to the closest solution.
Thus, a magnetic initial guess usually must be provided to get a magnetic solution. Finally, unless an
adsorbate is on a magnetic metal surface, spin polarization typically does not need to be considered,
although the gas-phase reference state calculation may need to be done with spin-polarization.
The downside of including spin polarization is that it essentially doubles the calculation time.
11

TODO Recommended reading

2.7

This section needs cleaning up.

Original papers on DFT 20,21
Kohns Nobel Lecture, 18 Poples Nobel Lecture 22
7
Hoffman
PAW in GPAW 36
All calculations were performed using VASP 12,3739 with the projector augmented wave (PAW) potentials provided in VASP.
Monkhorst-Pack k-points 28

Molecules

In this chapter we consider how to construct models of molecules, how to manipulate them, and how to
calculate many properties of molecules. For a nice comparison of VASP and Gaussian see 40 .

3.1

Defining and visualizing molecules

We start by learning how to define a molecule and visualize it. We will begin with defining molecules
from scratch, then reading molecules from data files, and finally using some built-in databases in ase.
3.1.1

From scratch

When there is no data file for the molecule you want, or no database to get it from, you have to define
your atoms geometry by hand. Here is how that is done for a CO molecule (Figure 1). We must define
the type and position of each atom, and the unit cell the atoms are in.
1
2

from ase import Atoms, Atom

from ase.io import write

3
4
5
6
7

# define an Atoms object

atoms = Atoms([Atom(C, [0., 0., 0.]),
Atom(O, [1.1, 0., 0.])],
cell=(10, 10, 10))

8
9

print V = {0:1.0f} Angstrom^3.format(atoms.get_volume())

10
11

write(images/simple-cubic-cell.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-3.py).

V = 1000 Angstrom^3
There are two inconvenient features of the simple cubic cell:
1. Since the CO molecule is at the corner, its electron density is spread over the 8 corners of the box,
which is not convenient for visualization later (see Visualizing electron density).
2. Due to the geometry of the cube, you need fairly large cubes to make sure the electron density of
the molecule does not overlap with that of its images. Electron-electron interactions are repulsive,
and the overlap makes the energy increase significantly. Here, the CO molecule has 6 images due
to periodic boundary conditions that are 10 away. The volume of the unit cell is 1000 3 .
The first problem is easily solved by centering the atoms in the unit cell. The second problem can be
solved by using a face-centered cubic lattice, which is the lattice with the closest packing. We show the
results of the centering in Figure 2, where we have guessed values for b until the CO molecules are on
average 10 apart. Note the final volume is only about 715 3 , which is smaller than the cube. This
will result in less computational time to compute properties.
12

Figure 1: Image of a CO molecule with the C at the origin.

1
2

from ase import Atoms, Atom

from ase.io import write

3
4
5
6
7
8
9

b = 7.1
atoms = Atoms([Atom(C, [0., 0., 0.]),
Atom(O, [1.1, 0., 0.])],
cell=[[b, b, 0.],
[b, 0., b],
[0., b, b]])

10
11

print(V = {0:1.0f} Ang^3.format(atoms.get_volume()))

12
13
14

atoms.center() # translate atoms to center of unit cell

write(images/fcc-cell.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-4.py).

V = 716 Ang^3

13

Figure 2: CO in a face-centered cubic unit cell.

At this point you might ask, "How do you know the distance to the neighboring image?" The ag
viewer lets you compute this graphically, but we can use code to determine this too. All we have to do
is figure out the length of each lattice vector, because these are what separate the atoms in the images.
We use the mod:numpy module to compute the distance of a vector as the square root of the sum of
squared elements.
1
2

from ase import Atoms, Atom

import numpy as np

3
4
5
6
7
8
9

b = 7.1
atoms = Atoms([Atom(C, [0., 0., 0.]),
Atom(O, [1.1, 0., 0.])],
cell=[[b, b, 0.],
[b, 0., b],
[0., b, b]])

10
11
12
13

# get unit cell vectors and their lengths

(a1, a2, a3) = atoms.get_cell()
print |a1| = {0:1.2f} Ang.format(np.sum(a1**2)**0.5)

14

14
15

print |a2| = {0:1.2f} Ang.format(np.linalg.norm(a2))

print |a3| = {0:1.2f} Ang.format(np.sum(a3**2)**0.5)

Open the python script (dft-scripts/script-5.py).

|a1| = 10.04 Ang
|a2| = 10.04 Ang
|a3| = 10.04 Ang
3.1.2

Reading other data formats into a calculation

mod:ase.io.read supports many different file formats:

Known formats:
=========================
format
=========================
GPAW restart-file
Dacapo netCDF output file
Old ASE netCDF trajectory
Virtual Nano Lab file
ASE pickle trajectory
ASE bundle trajectory
GPAW text output
CUBE file
XCrySDen Structure File
Dacapo text output
XYZ-file
VASP POSCAR/CONTCAR file
VASP OUTCAR file
SIESTA STRUCT file
ABINIT input file
V_Sim ascii file
Protein Data Bank
CIF-file
FHI-aims geometry file
FHI-aims output file
VTK XML Image Data
VTK XML Structured Grid
VTK XML Unstructured Grid
TURBOMOLE coord file
TURBOMOLE gradient file
exciting input
AtomEye configuration
WIEN2k structure file
DftbPlus input file
CASTEP geom file
CASTEP output file
CASTEP trajectory file
ETSF format
DFTBPlus GEN format
CMR db/cmr-file
CMR db/cmr-file

===========
short name
===========
gpw
dacapo
nc
vnl
traj
bundle
gpaw-text
cube
xsf
dacapo-text
xyz
vasp
vasp_out
struct_out
abinit
v_sim
pdb
cif
aims
aims_out
vti
vts
vtu
tmol
tmol-gradient
exi
cfg
struct
dftb
cell
castep
geom
etsf.nc
gen
db
cmr
15

LAMMPS dump file

Gromacs coordinates
=========================

lammps
gro
===========

You can read XYZ file format to create mod:ase.Atoms objects. Here is what an XYZ file format
might look like:
#+include: molecules/isobutane.xyz
The first line is the number of atoms in the file. The second line is often a comment. What follows
is one line per atom with the symbol and Cartesian coordinates in . Note that the XYZ format does
not have unit cell information in it, so you will have to figure out a way to provide it. In this example,
we center the atoms in a box with vacuum on all sides (Figure 3).
1

from ase.io import read, write

2
3
4
5

atoms = read(molecules/isobutane.xyz)
atoms.center(vacuum=5)
write(images/isobutane-xyz.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-6.py).

3.1.3

Predefined molecules

mod:ase defines a number of molecular geometries in the mod:ase.data.molecules database. For example,
the database includes the molecules in the G2/97 database. 41 This database contains a broad set of
atoms and molecules for which good experimental data exists, making them useful for benchmarking
studies. See this site for the original files.
The coordinates for the atoms in the database are MP2(full)/6-31G(d) optimized geometries. Here
is a list of all the species available in mod:ase.data.g2. You may be interested in reading about some of
the other databases in mod:ase.data too.
1
2
3
4
5

from ase.data import g2

keys = g2.data.keys()
# print in 3 columns
for i in range(len(keys) / 3):
print {0:25s}{1:25s}{2:25s}.format(*tuple(keys[i * 3: i * 3 + 3]))

Open the python script (dft-scripts/script-7.py).

isobutene
COF2
CH3OH
PH3
O2
Be
C3H9N
CH3
C2H6CHOH
CH3O
P
OH
SiCl4
C3H7
CH2NHCH2
SiF4

CH3CH2OH
CH3NO2
CCH
Si2H6
BCl3
H2CCl2
CH3CH2OCH3
CH4
SiH2_s1A1d
H
C3H4_C3v
methylenecyclopropane
HCF3
CH3CH2O
SiH2_s3B1d
H2CCO
16

CH3COOH
CF3CN
CH3CH2NH2
O3
CH2_s1A1d
C3H9C
BF3
S2
H3CNH2
BeH
C2F4
F2O
HCCl3
AlF3
H2CF2
PH2

Figure 3: An isobutane molecule read in from an XYZ formatted data file.

OCS
SH2
CH3CH2Cl
HCOOH
2-butyne
HOCl
C
C4H4S
H2CCHCl
C2H4
C2Cl4
C6H6
H2CCHCN
HCl
Na
CH3CH2SH

HF
C3H4_C2v
isobutane
CH3ONO
SH
CS2
CH3S
S
C2H6
HCN
bicyclobutane
N2H4
H2CCHF
CH3OCH3
CH3SiH3
OCHCHO

NO2
H2O2
CH3COF
C5H8
NF3
P2
O
C3H7Cl
CH3CHO
C2H2
H2
C4H4NH
cyclobutane
Li2
NaCl
SiH4

17

C2H5
ClO
NO
HCO
CH3COCH3
CO
CH3COCl
Si
N2
F2
CH2OCH2
SO
ClF3
B
C3H6_Cs
LiF
LiH
C2H6SO
Na2
NH3
ClNO
CH3SCH3

SiH3
AlCl3
C2H3
CH3CONH2
C3H4_D2d
CN
N
C3H8
Cl2
CO2
H2O
HCOOCH3
Li
CH3SH
C2H6NH
H2COH
SiO
C5H5N
C4H4O
NH2
C3H6_D3h
H2CO

NH
CCl4
ClF
CH2SCH2
CH
F
CH3Cl
CS
NCCN
Cl
CH3CO
butadiene
PF3
CF4
N2O
cyclobutene
Si2
trans-butane
SO2
CH2_s3B1d
Al
CH3CN

isobutene
COF2
CH3OH
PH3
O2
Be
C3H9N
CH3
C2H6CHOH
CH3O
P
OH
SiCl4
C3H7
CH2NHCH2
SiF4
OCS
SH2
CH3CH2Cl
HCOOH
2-butyne
HOCl
C
C4H4S
H2CCHCl
C2H4
C2Cl4
C6H6
H2CCHCN

CH3CH2OH
CH3NO2
CCH
Si2H6
BCl3
H2CCl2
CH3CH2OCH3
CH4
SiH2_s1A1d
H
C3H4_C3v
methylenecyclopropane
HCF3
CH3CH2O
SiH2_s3B1d
H2CCO
HF
C3H4_C2v
isobutane
CH3ONO
SH
CS2
CH3S
S
C2H6
HCN
bicyclobutane
N2H4
H2CCHF

CH3COOH
CF3CN
CH3CH2NH2
O3
CH2_s1A1d
C3H9C
BF3
S2
H3CNH2
BeH
C2F4
F2O
HCCl3
AlF3
H2CF2
PH2
NO2
H2O2
CH3COF
C5H8
NF3
P2
O
C3H7Cl
CH3CHO
C2H2
H2
C4H4NH
cyclobutane

18

HCl
Na
CH3CH2SH
C2H5
ClO
NO
HCO
CH3COCH3
CO
CH3COCl
Si
N2
F2
CH2OCH2
SO
ClF3
B
C3H6_Cs
LiF
LiH
C2H6SO
Na2
NH3
ClNO
CH3SCH3

CH3OCH3
CH3SiH3
OCHCHO
SiH3
AlCl3
C2H3
CH3CONH2
C3H4_D2d
CN
N
C3H8
Cl2
CO2
H2O
HCOOCH3
Li
CH3SH
C2H6NH
H2COH
SiO
C5H5N
C4H4O
NH2
C3H6_D3h
H2CO

Li2
NaCl
SiH4
NH
CCl4
ClF
CH2SCH2
CH
F
CH3Cl
CS
NCCN
Cl
CH3CO
butadiene
PF3
CF4
N2O
cyclobutene
Si2
trans-butane
SO2
CH2_s3B1d
Al
CH3CN

Some other databases include the mod:ase.data.s22 for weakly interacting dimers and complexes, and
mod:ase.data.extra_molecules which has a few extras like biphenyl and C60.
Here is an example of getting the geometry of an acetonitrile molecule and writing an image to a file.
Note that the default unit cell is a 1 1 1 cubic cell. That is too small to use if your calculator
uses periodic boundary conditions. We center the atoms in the unit cell and add vacuum on each side.
We will add 6 of vacuum on each side. In the write command we use the option show_unit_cell =2
to draw the unit cell boundaries. See Figure 4.
1
2

from ase.structure import molecule

from ase.io import write

3
4

atoms = molecule(CH3CN)

5
6
7
8
9

atoms.center(vacuum=6)
print unit cell
print ---------
print atoms.get_cell()

10
11

write(images/ch3cn.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-8.py).

unit cell
--------[[ 13.775328
[ 0.
[ 0.

0.
13.537479
0.

0.
]
0.
]
15.014576]]

It is possible to rotate the atoms with func:ase.io.write if you wanted to see pictures from another
angle. In the next example we rotate 45 degrees about the x-axis, then 45 degrees about the y-axis.
Note that this only affects the image, not the actual coordinates. See Figure 5.
19

Figure 4: A CH3 CN molecule in a box.

1
2

from ase.structure import molecule

from ase.io import write

3
4

atoms = molecule(CH3CN)

5
6
7
8
9

atoms.center(vacuum=6)
print unit cell
print ---------
print atoms.get_cell()

10
11
12

write(images/ch3cn-rotated.png, atoms,
show_unit_cell=2, rotation=45x,45y,0z)

Open the python script (dft-scripts/script-9.py).

unit cell
--------[[ 13.775328
[ 0.

0.
13.537479

0.
0.

]
]
20

0.

0.

15.014576]]

Figure 5: The rotated version of CH3 CN.

If you actually want to rotate the coordinates, there is a nice way to do that too, with the func:ase.Atoms.rotate
method. Actually there are some subtleties in rotation. One rotates the molecule an angle (in radians)
around a vector, but you have to choose whether the center of mass should be fixed or not. You also must
decide whether or not the unit cell should be rotated. In the next example you can see the coordinates
have changed due to the rotations. Note that the write function uses the rotation angle in degrees, while
the rotate function uses radians.
1
2
3

from ase.structure import molecule

from ase.io import write
from numpy import pi

4
5
6
7

atoms = molecule(CH3CN)
atoms.center(vacuum=6)
p1 = atoms.get_positions()

8
9
10

atoms.rotate(x, pi/4, center=COM, rotate_cell=False)

atoms.rotate(y, pi/4, center=COM, rotate_cell=False)

11
12
13
14
15
16

write(images/ch3cn-rotated-2.png, atoms, show_unit_cell=2)

print difference in positions after rotating
print atom
difference vector
print --------------------------------------
p2 = atoms.get_positions()

17

21

18
19
20

diff = p2 - p1
for i, d in enumerate(diff):
print {0} {1}.format(i, d)

Open the python script (dft-scripts/script-10.py).

difference in positions after rotating
atom
difference vector
-------------------------------------0 [-0.65009456 0.91937255 0.65009456]
1 [ 0.08030744 -0.11357187 -0.08030744]
2 [ 0.66947344 -0.94677841 -0.66947344]
3 [-0.32532156 0.88463727 1.35030756]
4 [-1.35405183 1.33495444 -0.04610517]
5 [-0.8340703
1.33495444 1.2092413 ]

Figure 6: Rotated CH3 CN molecule

Note in this last case the unit cell is oriented differently than the previous example, since we chose
not to rotate the unit cell.
3.1.4

Combining Atoms objects

It is frequently useful to combine two Atoms objects, e.g. for computing reaction barriers, or other types
of interactions. In ase, we simply add two Atoms objects together. Here is an example of getting an
ammonia and oxygen molecule in the same unit cell. See Figure 7. We set the Atoms about three
apart using the func:ase.Atoms.translate function.
1
2

from ase.structure import molecule

from ase.io import write

3
4

atoms1 = molecule(NH3)

5
6
7

atoms2 = molecule(O2)
atoms2.translate([3, 0, 0])

8
9
10

bothatoms = atoms1 + atoms2

bothatoms.center(5)

11
12

write(images/bothatoms.png, bothatoms, show_unit_cell=2, rotation=90x)

Open the python script (dft-scripts/script-11.py).

3.2

Simple properties

Simple properties do not require a DFT calculation. They are typically only functions of the atom types
and geometries.

22

Figure 7: Image featuring ammonia and oxygen molecule in one unit cell.
3.2.1

Getting cartesian positions

If you want the (x, y, z) coordinates of the atoms, use the func:ase.Atoms.get_positions. If you are
interested in the fractional coordinates, use func:ase.Atoms.get_scaled_positions.
1

from ase.structure import molecule

2
3

atoms = molecule(C6H6)

# benzene

4
5
6
7
8
9
10
11
12
13

# access properties on each atom

print # sym
p_x
p_y
p_z
print ------------------------------
for i, atom in enumerate(atoms):
print {0:3d}{1:^4s}{2:-8.2f}{3:-8.2f}{4:-8.2f}.format(i,
atom.symbol,
atom.x,
atom.y,
atom.z)

14
15
16
17
18

# get
sym =
pos =
num =

all properties in arrays

atoms.get_chemical_symbols()
atoms.get_positions()
atoms.get_atomic_numbers()

19
20

atom_indices = range(len(atoms))

21
22
23
24
25

print
print # sym
at#
p_x
p_y
p_z
print -------------------------------------
for i, s, n, p in zip(atom_indices, sym, num, pos):

23

26
27
28

px, py, pz = p
print {0:3d}{1:>3s}{2:8d}{3:-8.2f}{4:-8.2f}{5:-8.2f}.format(i, s, n,
px, py, pz)

Open the python script (dft-scripts/script-12.py).

# sym
p_x
p_y
p_z
-----------------------------0 C
0.00
1.40
0.00
1 C
1.21
0.70
0.00
2 C
1.21
-0.70
0.00
3 C
0.00
-1.40
0.00
4 C
-1.21
-0.70
0.00
5 C
-1.21
0.70
0.00
6 H
0.00
2.48
0.00
7 H
2.15
1.24
0.00
8 H
2.15
-1.24
0.00
9 H
0.00
-2.48
0.00
10 H
-2.15
-1.24
0.00
11 H
-2.15
1.24
0.00
# sym
at#
p_x
p_y
p_z
------------------------------------0 C
6
0.00
1.40
0.00
1 C
6
1.21
0.70
0.00
2 C
6
1.21
-0.70
0.00
3 C
6
0.00
-1.40
0.00
4 C
6
-1.21
-0.70
0.00
5 C
6
-1.21
0.70
0.00
6 H
1
0.00
2.48
0.00
7 H
1
2.15
1.24
0.00
8 H
1
2.15
-1.24
0.00
9 H
1
0.00
-2.48
0.00
10 H
1
-2.15
-1.24
0.00
11 H
1
-2.15
1.24
0.00
3.2.2

Molecular weight and molecular formula

We can quickly compute the molecular weight of a molecule with this recipe. We use func:ase.Atoms.get_masses to get an array of the atomic masses of each atom in the Atoms object, and then just sum them
up.
1

from ase.structure import molecule

2
3
4

atoms = molecule(C6H6)
masses = atoms.get_masses()

5
6
7

molecular_weight = masses.sum()
molecular_formula = atoms.get_chemical_formula(mode=reduce)

8
9
10
11

# note use of two lines to keep length of line reasonable

s = The molecular weight of {0} is {1:1.2f} gm/mol
print s.format(molecular_formula, molecular_weight)

Open the python script (dft-scripts/script-13.py).

The molecular weight of C6H6 is 78.11 gm/mol
24

Note that the argument reduce=True for func:ase.Atoms.get_chemical_formula collects all the symbols to provide a molecular formula.
3.2.3

Center of mass

The center ofPmass (COM) is defined as:

mi ri
COM = P
mi

The center of mass is essentially the average position of the atoms, weighted by the mass of each
atom. Here is an example of getting the center of mass from an Atoms object using func:ase.Atoms.get_center_of_mass.
1
2

from ase.structure import molecule

import numpy as np

3
4
5

# ammonia
atoms = molecule(NH3)

6
7

print COM1 = {0}.format(atoms.get_center_of_mass())

# cartesian coordinates

8
9
10
11

# compute the center of mass by hand

pos = atoms.positions
masses = atoms.get_masses()

12
13
14
15
16

COM = np.array([0., 0., 0.])

for m, p in zip(masses, pos):
COM += m*p
COM /= masses.sum()

17
18

print COM2 = {0}.format(COM)

19
20
21

# one-line linear algebra definition of COM

print COM3 = {0}.format(np.dot(masses, pos) / np.sum(masses))

Open the python script (dft-scripts/script-14.py).

COM1 = [
COM2 = [
COM3 = [

0.00000000e+00
0.00000000e+00
0.00000000e+00

5.91843349e-08
5.91843349e-08
5.91843349e-08

4.75457009e-02]
4.75457009e-02]
4.75457009e-02]

You can see see that these centers of mass, which are calculated by different methods, are the same.
3.2.4

Moments of inertia

PN
The moment of inertia is a measure of resistance to changes in rotation. It is defined by I = i=1 mi ri2
where ri is the distance to an axis of rotation. There are typically three moments of inertia, although
some may be zero depending on symmetry, and others may be degenerate. There is a convenient function
to get the moments of inertia: func:ase.Atoms.get_moments_of_inertia. Here are several examples of
molecules with different types of symmetry.:
1

from ase.structure import molecule

2
3
4
5
6

print linear rotors: I = [0 Ia Ia]

atoms = molecule(CO2)
print CO2 moments of inertia: , atoms.get_moments_of_inertia()
print

7
8
9
10

print symmetric rotors (Ia = Ib) < Ic

atoms = molecule(NH3)
print NH3 moments of inertia: , atoms.get_moments_of_inertia()

11
12
13
14

atoms = molecule(C6H6)
print C6H6 moments of inertia:, atoms.get_moments_of_inertia()
print

15

25

16
17
18
19

print symmetric rotors Ia < (Ib = Ic)

atoms = molecule(CH3Cl)
print CH3Cl moments of inertia: , atoms.get_moments_of_inertia()
print

20
21
22
23
24

print spherical rotors Ia = Ib = Ic

atoms = molecule(CH4)
print CH4 moments of inertia: , atoms.get_moments_of_inertia()
print

25
26
27
28

print unsymmetric rotors Ia != Ib != Ic

atoms = molecule(C3H7Cl)
print C3H7Cl moments of inertia: , atoms.get_moments_of_inertia()

Open the python script (dft-scripts/script-15.py).

linear rotors: I = [0 Ia Ia]
CO2 moments of inertia: [

0.

44.45384271

44.45384271]

symmetric rotors (Ia = Ib) < Ic

NH3 moments of inertia: [ 1.71012426 1.71012548 2.67031768]
C6H6 moments of inertia: [ 88.77914641
88.77916799 177.5583144 ]
symmetric rotors Ia < (Ib = Ic)
CH3Cl moments of inertia: [ 3.20372189
spherical rotors Ia = Ib = Ic
CH4 moments of inertia: [ 3.19145621

37.97009644

3.19145621

unsymmetric rotors Ia != Ib != Ic
C3H7Cl moments of inertia: [ 19.41351508

37.97009837]

3.19145621]

213.18961963

223.16255537]

If you want to know the principle axes of rotation, we simply pass vectors=True to the function,
and it returns the moments of inertia and the principle axes.
1

from ase.structure import molecule

2
3
4
5
6

atoms = molecule(CH3Cl)
moments, axes = atoms.get_moments_of_inertia(vectors=True)
print Moments = {0}.format(moments)
print axes = {0}.format(axes)

Open the python script (dft-scripts/script-16.py).

Moments = [ 3.20372189
axes = [[ 0. 0. 1.]
[ 0. 1. 0.]
[ 1. 0. 0.]]

37.97009644

37.97009837]

This shows the first moment is about the z-axis, the second moment is about the y-axis, and the
third moment is about the x-axis.
3.2.5

Computing bond lengths and angles

A typical question we might ask is, "What is the structure of a molecule?" In other words, what
are the bond lengths, angles between bonds, and similar properties. The Atoms object contains an
func:ase.Atoms.get_distance method to make this easy. To calculate the distance between two atoms,
you have to specify their indices, remembering that the index starts at 0.

26

from ase.structure import molecule

2
3
4

# ammonia
atoms = molecule(NH3)

5
6
7
8
9

print atom symbol

print ===========
for i, atom in enumerate(atoms):
print {0:2d} {1:3s} .format(i, atom.symbol)

10
11
12
13

# N-H bond length

s = The N-H distance is {0:1.3f} angstroms
print s.format(atoms.get_distance(0, 1))

Open the python script (dft-scripts/script-17.py).

atom symbol
===========
0 N
1 H
2 H
3 H
The N-H distance is 1.017 angstroms
Bond angles are a little trickier. If we had vectors describing the directions between two atoms, we
could use some simple trigonometry to
the angle between the vectors: ~a ~b = |~a||~b| cos(). So we

 compute
~
a~b
, we just have to define our two vectors ~a and ~b. We compute
can calculate the angle as = arccos
|~
a||~b|

these vectors as the difference in positions of two atoms. For example, here we compute the angle H-N-H
in an ammonia molecule. This is the angle between N-H1 and N-H2 . In the next example, we utilize
functions in mod:numpy to perform the calculations, specifically the func:numpy.arccos function, the
func:numpy.dot function, and func:numpy.linalg.norm functions.
1

from ase.structure import molecule

2
3
4

# ammonia
atoms = molecule(NH3)

5
6
7
8
9

print atom symbol

print ===========
for i, atom in enumerate(atoms):
print {0:2d} {1:3s}.format(i, atom.symbol)

10
11
12

a = atoms.positions[0] - atoms.positions[1]
b = atoms.positions[0] - atoms.positions[2]

13
14
15

from numpy import arccos, dot, pi

from numpy.linalg import norm

16
17

theta_rad = arccos(dot(a, b) / (norm(a) * norm(b)))

# in radians

18
19

print theta = {0:1.1f} degrees.format(theta_rad * 180./pi)

Open the python script (dft-scripts/script-18.py).

atom symbol
===========
0 N
1 H
2 H
3 H
theta = 106.3 degrees

27

Figure 8: Schematic of the vectors defining the H-N-H angle.

Alternatively you could use func:ase.Atoms.get_angle. Note we want the angle between atoms with
indices [1, 0, 2] to get the H-N-H angle.
1
2
3
4

from ase.structure import molecule

from numpy import pi
# ammonia
atoms = molecule(NH3)

5
6

print theta = {0} degrees.format(atoms.get_angle([1, 0, 2]) * 180. / pi)

Open the python script (dft-scripts/script-19.py).

theta = 106.334624232 degrees
Dihedral angles There is support in ase for computing dihedral angles. Let us illustrate that for
ethane. We will compute the dihedral angle between atoms 5, 1, 0, and 4. That is a H-C-C-H dihedral
angle, and one can visually see (although not here) that these atoms have a dihedral angle of 60 (Figure
9).
1
2
3

# calculate an ethane dihedral angle

from ase.structure import molecule
import numpy as np

4
5

atoms = molecule(C2H6)

6
7
8
9
10

print atom symbol

print ===========
for i, atom in enumerate(atoms):
print {0:2d} {1:3s}.format(i, atom.symbol)

11
12
13

da = atoms.get_dihedral([5, 1, 0, 4]) * 180. / np.pi

print dihedral angle = {0:1.2f} degrees.format(da)

28

Open the python script (dft-scripts/script-20.py).

atom symbol
===========
0 C
1 C
2 H
3 H
4 H
5 H
6 H
7 H
dihedral angle = 60.00 degrees

Figure 9: Schematic of the calculated ethane dihedral angle.

In this section we covered properties that require simple calculations, but not DFT calculations, to
compute.

3.3

Simple properties that require single computations

There are many properties that only require a single DFT calculation to obtain the energy, forces, density
of states, electron denisty and electrostatic potential. This section describes some of these calculations
and their analysis.
3.3.1

Energy and forces

Two of the most important quantities we are interested in are the total energy and the forces on the
atoms. To get these quantities, we have to define a calculator and attach it to an mod:ase.Atoms object
so that ase knows how to get the data. After defining the calculator a DFT calculation must be run.
Here is an example of getting the energy and forces from a CO molecule. The forces in this case are
very high, indicating that this geometry is not close to the ground state geometry. Note that the forces
are only along the x-axis, which is along the molecular axis. We will see how to minimize this force in
Manual determination and Automatic geometry optimization with VASP.
Note:
This is your first DFT calculation in the book! See ISMEAR, incar:SIGMA, incar:NBANDS, and incar:ENCUT to learn more about these VASP keywords.
1
2
3
4

from ase import Atoms, Atom

from jasp import *
import numpy as np
np.set_printoptions(precision=3, suppress=True)

5
6
7
8

co = Atoms([Atom(C, [0, 0, 0]),

Atom(O, [1.2, 0, 0])],
cell=(6., 6., 6.))

9
10
11

with jasp(molecules/simple-co, # output dir

xc=PBE, # the exchange-correlation functional

29

12
13
14
15
16
17
18

nbands=6,
# number of bands
encut=350,
# planewave cutoff
ismear=1,
# Methfessel-Paxton smearing
sigma=0.01, # very small smearing factor for a molecule
atoms=co) as calc:
print energy = {0} eV.format(co.get_potential_energy())
print co.get_forces()

Open the python script (dft-scripts/script-21.py).

energy = -14.687906 eV
[[ 5.095 0.
0.
]
[-5.095 0.
0.
]]
Convergence with unit cell size There are a number of parameters that affect the energy and forces
including the calculation parameters and the unit cell. We will first consider the effect of the unit cell on
the total energy and forces. The reason that the unit cell affects the total energy is that it can change
the distribution of electrons in the molecule.
1
2
3
4

from jasp import *

from ase import Atoms, Atom
import numpy as np
np.set_printoptions(precision=3, suppress=True)

5
6
7

atoms = Atoms([Atom(C, [0, 0, 0]),

Atom(O, [1.2, 0, 0])])

8
9

L = [4, 5, 6, 8, 10]

10
11
12
13
14
15
16
17
18
19
20
21
22
23

volumes, energies = [], []

ready = True
for a in L:
atoms.set_cell([a, a, a], scale_atoms=False)
atoms.center()
with jasp(molecules/co-L-{0}.format(a),
encut=350,
xc=PBE,
atoms=atoms) as calc:
try:
energies.append(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
ready = False

24
25
26

if not ready:
import sys; sys.exit()

27
28
29
30
31
32
33

import matplotlib.pyplot as plt

plt.plot(L, energies, bo-)
plt.xlabel(Unit cell length ($\AA$))
plt.ylabel(Total energy (eV))
plt.savefig(images/co-e-v.png)
plt.show()

Open the python script (dft-scripts/script-22.py).

Here there are evidently attractive interactions between the CO molecules which lower the total
energy for small box sizes. We have to decide what an appropriate volume for our calculation is, and the
choice depends on the goal. We may wish to know the total energy of a molecule that is not interacting
with any other molecules, e.g. in the ideal gas limit. In that case we need a large unit cell so the electron
density from the molecule does not go outside the unit cell where it would overlap with neighboring
images.
It pays to check for convergence. The cost of running the calculation goes up steeply with increasing
cell size. Doubling a lattice vector here leads to a 20-fold increase in computational time! Note that
doubling a lattice vector length increases the volume by a factor of 8 for a cube. The cost goes up because
the number of planewaves that fit in the cube grows as the cube gets larger.

30

Figure 10: Total energy of a CO molecule as a function of the unit cell length.

from jasp import *

2
3

L = [4, 5, 6, 8, 10]

4
5
6
7

for a in L:
with jasp(molecules/co-L-{0}.format(a)) as calc:
print {0} {1} seconds.format(a, calc.get_elapsed_time())

Open the python script (dft-scripts/script-23.py).

4 1.601 seconds
5 3.932 seconds
6 6.543 seconds
8 19.678 seconds
10 36.013 seconds
Let us consider what the pressure in the unit cell is. In the ideal gas limit we have P V = nRT ,
which gives a pressure of zero at absolute zero. At non-zero temperatures, we have P = n/V RT . Let us
consider some examples. In atomic units we use kB instead of R.
1
2
3

from ase.units import kB, Pascal

import numpy as np
import matplotlib.pyplot as plt

4
5

atm = 101325 * Pascal

6
7

L = np.linspace(4, 10)

31

V = L**3

9
10

n = 1

# one atom per unit cell

11
12
13

for T in [298, 600, 1000]:

P = n / V * kB * T / atm

# convert to atmospheres

14
15

plt.plot(V, P, label={0}K.format(T))

16
17
18
19
20

plt.xlabel(Unit cell volume ($\AA^3$))

plt.ylabel(Pressure (atm))
plt.legend(loc=best)
plt.savefig(images/ideal-gas-pressure.png)

Open the python script (dft-scripts/script-24.py).

Figure 11: Ideal gas pressure dependence on temperature and unit cell volume.

Convergence of ENCUT The total energy and forces also depend on the computational parameters,
notably incar:ENCUT.
1
2
3
4

from ase import Atoms, Atom

from jasp import *
import numpy as np
np.set_printoptions(precision=3, suppress=True)

5
6
7
8
9

atoms = Atoms([Atom(C, [0, 0, 0]),

Atom(O, [1.2, 0, 0])],
cell=(6, 6, 6))
atoms.center()

10
11

ENCUTS = [250, 300, 350, 400, 450, 500]

32

12
13
14
15
16
17
18
19
20
21
22
23

energies = []
ready = True
for en in ENCUTS:
with jasp(molecules/co-en-{0}.format(en),
encut=en,
xc=PBE,
atoms=atoms) as calc:
try:
energies.append(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
ready = False

24
25
26

if not ready:
import sys; sys.exit()

27
28
29
30
31
32

import matplotlib.pyplot as plt

plt.plot(ENCUTS, energies, bo-)
plt.xlabel(ENCUT (eV))
plt.ylabel(Total energy (eV))
plt.savefig(images/co-encut-v.png)

Open the python script (dft-scripts/script-25.py).

Figure 12: Dependence of the total energy of CO molecule on ENCUT.

You can see in this figure that it takes a cutoff energy of about 400 eV to achieve a convergence level
around 10 meV, and that even at 500 meV the energy is still changing slightly. Keep in mind that we
are generally interested in differences in total energy, and the differences tend to converge faster than a
single total energy. Also it is important to note that it is usually a single element that determines the
rate of convergence. The reason we do not just use very high ENCUT all the time is it is expensive.
1

grep "Elapsed time (sec):" molecules/co-en-*/OUTCAR

33

Open the python script (dft-scripts/script-26.py).

molecules/co-en-250/OUTCAR:
molecules/co-en-300/OUTCAR:
molecules/co-en-350/OUTCAR:
molecules/co-en-400/OUTCAR:
molecules/co-en-450/OUTCAR:
molecules/co-en-500/OUTCAR:

Elapsed
Elapsed
Elapsed
Elapsed
Elapsed
Elapsed

time
time
time
time
time
time

(sec):
(sec):
(sec):
(sec):
(sec):
(sec):

3.590
6.016
6.014
7.883
12.421
15.167

Although here the times do not seem that unreasonable, note we increased the time by a factor of
three by simply doubling ENCUT. If the first time was a week, this is a significant increase in time!
3.3.2

Visualizing electron density

The electron density is a 3d quantity: for every (x, y, z) point, there is a charge density. That means we
need 4 numbers for each point: (x, y, z) and (x, y, z). Below we show an example (Figure 13) of plotting
the charge density, and we consider some issues we have to consider when visualizing volumetric data in
unit cells with periodic boundary conditions. We will use the results from a previous calculation.
#TODO does not work in canopy
1
2
3
4

from
from
from
from

jasp import *
enthought.mayavi.mlab import *
ase.data import vdw_radii
ase.data.colors import cpk_colors

5
6
7
8

with jasp(molecules/simple-co) as calc:

atoms = calc.get_atoms()
x, y, z, cd = calc.get_charge_density()

9
10

mlab.figure(1, bgcolor=(1, 1, 1)) # make a white figure

11
12
13
14
15
16
17
18
19
20
21

# plot the atoms as spheres

for atom in atoms:
mlab.points3d(atom.x,
atom.y,
atom.z,
scale_factor=vdw_radii[atom.number]/5., #this determines the size of the atom
resolution=20,
# a tuple is required for the color
color=tuple(cpk_colors[atom.number]),
scale_mode=none)

22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

# draw the unit cell - there are 8

a1, a2, a3 = atoms.get_cell()
origin = [0, 0, 0]
cell_matrix = [[origin, a1],
[origin, a2],
[origin, a3],
[a1,
a1 + a2],
[a1,
a1 + a3],
[a2,
a2 + a1],
[a2,
a2 + a3],
[a3,
a1 + a3],
[a3,
a2 + a3],
[a1 + a2, a1 + a2 +
[a2 + a3, a1 + a2 +
[a1 + a3, a1 + a3 +

corners, and 12 connections

a3],
a3],
a2]]

38
39
40
41
42
43

for p1, p2 in cell_matrix:

mlab.plot3d([p1[0], p2[0]], # x-positions
[p1[1], p2[1]], # y-positions
[p1[2], p2[2]], # z-positions
tube_radius=0.02)

44
45
46
47

# Now plot the charge density

mlab.contour3d(x, y, z, cd)
mlab.view(azimuth=-90, elevation=90, distance=auto)

48
49
50

mlab.savefig(images/co-cd.png)
mlab.show()

34

Open the python script (dft-scripts/script-27.py).

Figure 13: Charge density of a CO molecule that is located at the origin. The electron density that is
outside the cell is wrapped around to the other corners.
If we take care to center the CO molecule in the unit cell, we get a nicer looking result.
TODO broken in canopy
1
2
3
4
5

from
from
from
from
from

jasp import *
enthought.mayavi import mlab
ase.data import vdw_radii
ase.data.colors import cpk_colors
ase import Atom, Atoms

6
7
8
9

atoms = Atoms([Atom(C, [2.422, 0.0, 0.0]),

Atom(O, [3.578, 0.0, 0.0])],
cell=(10,10,10))

10
11

atoms.center()

12
13
14
15
16

with jasp(molecules/co-centered,
encut=350,
xc=PBE,
atoms=atoms) as calc:

17
18
19

atoms = calc.get_atoms()
x, y, z, cd = calc.get_charge_density()

20
21

mlab.figure(bgcolor=(1, 1, 1))

22
23
24
25

# plot the atoms as spheres

for atom in atoms:
mlab.points3d(atom.x,

35

atom.y,
atom.z,
scale_factor=vdw_radii[atom.number]/5.,
resolution=20,
# a tuple is required for the color
color=tuple(cpk_colors[atom.number]),
scale_mode=none)

26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

# draw the unit cell - there are 8

a1, a2, a3 = atoms.get_cell()
origin = [0, 0, 0]
cell_matrix = [[origin, a1],
[origin, a2],
[origin, a3],
[a1,
a1 + a2],
[a1,
a1 + a3],
[a2,
a2 + a1],
[a2,
a2 + a3],
[a3,
a1 + a3],
[a3,
a2 + a3],
[a1 + a2, a1 + a2 +
[a2 + a3, a1 + a2 +
[a1 + a3, a1 + a3 +

corners, and 12 connections

a3],
a3],
a2]]

49
50
51
52
53
54

for p1, p2 in cell_matrix:

mlab.plot3d([p1[0], p2[0]], # x-positions
[p1[1], p2[1]], # y-positions
[p1[2], p2[2]], # z-positions
tube_radius=0.02)

55
56
57
58

# Now plot the charge density

mlab.contour3d(x, y, z, cd, transparent=True)

59
60
61

# this view was empirically found by iteration

mlab.view(azimuth=-90, elevation=90, distance=auto)

62
63
64

mlab.savefig(images/co-centered-cd.png)
mlab.show()

Open the python script (dft-scripts/script-28.py).

3.3.3

Dipole moments

The dipole moment is a vector describing the separation of electrical (negative) and nuclear (positive)
charge. The magnitude of this vector is the dipole moment, which has units of Coulomb-meter, or more
commonly Debye. The symmetry of a molecule determines if a molecule has a dipole moment or not.
Below we compute the dipole moment of CO. We must integrate the electron density to find the center
of electrical charge, and compute a sum over the nuclei to find the center of positive charge.
1
2

from jasp import *

from ase.units import Debye

3
4
5
6

with jasp(molecules/co-centered) as calc:

atoms = calc.get_atoms()
atoms.get_potential_energy()

7
8

vcd = VaspChargeDensity()

9
10
11

cd = np.array(vcd.chg[0])
n0, n1, n2 = cd.shape

12
13
14
15

s0 = 1.0 / n0
s1 = 1.0 / n1
s2 = 1.0 / n2

16
17
18
19

X, Y, Z = np.mgrid[0.0:1.0:s0,
0.0:1.0:s1,
0.0:1.0:s2]

20
21
22
23

C = np.column_stack([X.ravel(),
Y.ravel(),
Z.ravel()])

36

Figure 14: Charge density of a CO molecule centered in the unit cell. Now the electron density is centered
in the unit cell.
24
25
26
27

atoms = calc.get_atoms()
uc = atoms.get_cell()
real = np.dot(C, uc)

28
29
30
31
32

#
x
y
z

now convert arrays back to unitcell shape

= np.reshape(real[:, 0], (n0, n1, n2))
= np.reshape(real[:, 1], (n0, n1, n2))
= np.reshape(real[:, 2], (n0, n1, n2))

33
34
35
36

nelements = n0 * n1 * n2
voxel_volume = atoms.get_volume() / nelements
total_electron_charge = -cd.sum() * voxel_volume

37
38
39
40
41
42

electron_density_center = np.array([(cd * x).sum(),

(cd * y).sum(),
(cd * z).sum()])
electron_density_center *= voxel_volume
electron_density_center /= total_electron_charge

43
44

electron_dipole_moment = -electron_density_center * total_electron_charge

45
46
47

# now the ion charge center. We only need the Zval listed in the potcar
from jasp.POTCAR import get_ZVAL

48
49
50

LOP = calc.get_pseudopotentials()
ppp = os.environ[VASP_PP_PATH]

51
52
53
54

zval = {}
for sym, ppath, hash in LOP:
fullpath = os.path.join(ppp, ppath)

37

55
56
57
58
59
60
61
62
63

z = get_ZVAL(fullpath)
zval[sym] = z
ion_charge_center = np.array([0.0, 0.0, 0.0])
total_ion_charge = 0.0
for atom in atoms:
Z = zval[atom.symbol]
total_ion_charge += Z
pos = atom.position
ion_charge_center += Z*pos

64
65
66

ion_charge_center /= total_ion_charge
ion_dipole_moment = ion_charge_center * total_ion_charge

67
68

dipole_vector = (ion_dipole_moment + electron_dipole_moment)

69
70

dipole_moment = ((dipole_vector**2).sum())**0.5 / Debye

71
72

print The dipole moment is {0:1.2f} Debye.format(dipole_moment)

Open the python script (dft-scripts/script-29.py).

The dipole moment is 0.10 Debye
Note that a function using the code above exists in jasp which makes it trivial to compute the dipole
moment. Here is an example of its usage.
1
2

from jasp import *

from ase.units import Debye

3
4
5
6
7

with jasp(molecules/co-centered) as calc:

dipole_vector = calc.get_dipole_moment()
dipole_moment = ((dipole_vector**2).sum())**0.5/Debye
print The dipole moment is {0:1.2f} Debye.format(dipole_moment)

Open the python script (dft-scripts/script-30.py).

The dipole moment is 0.10 Debye
3.3.4

The density of states (DOS)

The density of states (DOS) gives you the number of electronic states (i.e., the orbitals) that have a
particular energy. We can get this information from the last calculation we just ran without having to
run another DFT calculation.
Warning:
One way that jasp is unintuitive is that whenever you are "inside" (indented) a with
statement, the current working directory may be different than where your script started.
In this example that can lead to your figure not being saved where you expect it to be
saved!
1
2
3

from jasp import *

from ase.dft.dos import DOS
import matplotlib.pyplot as plt

4
5
6
7
8
9

with jasp(molecules/simple-co) as calc: # we already ran this!

dos = DOS(calc)
plt.plot(dos.get_energies(), dos.get_dos())
plt.xlabel(Energy - $E_f$ (eV))
plt.ylabel(DOS)

10
11
12
13

# make sure you save the figure outside the with statement, or provide
# the correct relative or absolute path to where you want it.
plt.savefig(images/co-dos.png)

Open the python script (dft-scripts/script-31.py).

38

Figure 15: Density of states for a CO molecule.

3.3.5

Atom-projected density of states on molecules

Let us consider which states in the density of states belong to which atoms in a molecule. This can
only be a qualitative consideration because the orbitals on the atoms often hybridize to form molecular
orbitals, e.g. in methane the s and p orbitals can form what we call sp3 orbitals. We can compute
atom-projected density of states in VASP, which is done by projecting the wave function onto localized
atomic orbitals. Here is an example. We will consider the CO molecule. To get atom-projected density
of states, we must set RWIGS for each atom. This parameter defines the radius of the sphere around the
atom which cuts off the projection. The total density of states and projected density of states information
comes from the DOSCAR file.
Note that unlike the DOS, here we must run another calculation because we did not specify the
atom-projected keywords above. Our strategy is to get the atoms from the previous calculation, and use
them in a new calculation. You could redo the calculation in the same directory, but you risk losing the
results of the first step. That can make it difficult to reproduce a result. We advocate our approach of
using multiple directories for the subsequent calculations, because it leaves a clear trail of how the work
was done.
Note:
The RWIGS is not uniquely determined for an element. There are various natural choices,
e.g. the ionic radius of an atom, or a value that minimizes overlap of neighboring spheres,
but these values can change slightly in different environments.
You can also get spin-polarized atom-projected DOS, and magnetization projected DOS.
See http://cms.mpi.univie.ac.at/vasp/vasp/DOSCAR_file.html#doscar for more
details.

39

1
2
3

from jasp import *

from ase.dft.dos import DOS
import matplotlib.pyplot as plt

4
5
6
7

# get the geometry from another calculation

with jasp(molecules/simple-co) as calc:
atoms = calc.get_atoms()

8
9
10
11
12
13
14

with jasp(molecules/co-ados,
encut=300,
xc=PBE,
rwigs=[1.0, 1.0],
# these are the cutoff radii for projected states
atoms=atoms) as calc:
calc.calculate()

15
16
17
18

# now get results

dos = DOS(calc)
plt.plot(dos.get_energies(), dos.get_dos() + 10)

19
20
21

ados = VaspDos(efermi=calc.get_fermi_level())
energies = ados.energy

22
23

plt.plot(energies, ados.dos + 8, label=ADOS)

# these are the total DOS

24
25
26
27
28

c_s
c_p
o_s
o_p

=
=
=
=

ados.site_dos(0,
ados.site_dos(0,
ados.site_dos(1,
ados.site_dos(1,

s)
p)
s)
p)

29
30
31

c_d = ados.site_dos(0, d)
o_d = ados.site_dos(1, d)

32
33
34
35
36
37
38
39
40
41
42
43

plt.plot(energies, c_s + 6, energies, o_s + 5)

plt.plot(energies, c_p + 4, energies, o_p + 3)
plt.plot(energies, c_d, energies, o_d + 2)
plt.xlabel(Energy - $E_f$ (eV))
plt.ylabel(DOS)
plt.legend([DOS, ADOS,
C$_s$, O$_s$,
C$_p$, O$_p$,
C$_d$, O$_d$],
ncol=2, loc=best)
plt.savefig(images/co-ados.png)

Open the python script (dft-scripts/script-32.py).

3.3.6

Electrostatic potential

This is an example of the so-called hole in a halogen bond. The coordinates for the CF3 Br molecule
were found at http://cccbdb.nist.gov/exp2.asp?casno=75638.
TODO mlab does not work
1
2
3

from jasp import *

from ase import Atom, Atoms
from ase.io import write

4
5
6
7

from enthought.mayavi import mlab

from ase.data import vdw_radii
from ase.data.colors import cpk_colors

8
9
10
11
12
13
14
15

atoms = Atoms([Atom(C, [ 0.0000,

Atom(Br, [ 0.0000,
Atom(F, [ 0.0000,
Atom(F, [ 1.0787,
Atom(F, [-1.0787,
cell=(10, 10, 10))
atoms.center()

0.0000,
0.0000,
1.2455,
-0.6228,
-0.6228,

-0.8088]),
1.1146]),
-1.2651]),
-1.2651]),
-1.2651])],

16
17
18
19
20
21

with jasp(molecules/CF3Br,
encut=350,
xc=PBE,
ibrion=1,
nsw=50,

40

Figure 16: Atom-projected DOS for a CO molecule. The total density of states and the s, p and d states
on the C and O are shown.
22
23
24
25
26

lvtot=True,
lvhar=True,
atoms=atoms) as calc:
calc.set_nbands(f=2)
calc.calculate()

27
28
29

x, y, z, lp = calc.get_local_potential()
x, y, z, cd = calc.get_charge_density()

30
31

mlab.figure(1, bgcolor=(1, 1, 1)) # make a white figure

32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

# plot the atoms as spheres

for atom in atoms:
mlab.points3d(atom.x,
atom.y,
atom.z,
scale_factor=vdw_radii[atom.number]/5.,
resolution=20,
# a tuple is required for the color
color=tuple(cpk_colors[atom.number]),
scale_mode=none)
# plot the bonds. We want a line from C-Br, C-F, etc...
# We create a bond matrix showing which atoms are connected.
bond_matrix = [[0, 1],
[0, 2],
[0, 3],
[0, 4]]

49
50
51
52
53
54

for a1, a2 in bond_matrix:

mlab.plot3d(atoms.positions[[a1,a2], 0], # x-positions
atoms.positions[[a1,a2], 1], # y-positions
atoms.positions[[a1,a2], 2], # z-positions
[2, 2],

41

tube_radius=0.02,
colormap=Reds)

55
56
57
58
59
60

mlab.contour3d(x, y, z, lp)
mlab.savefig(images/halogen-ep.png)
mlab.show()

Open the python script (dft-scripts/script-33.py).

Figure 17: Plot of the electrostatic potential of CF3 Br. TODO: figure out how to do an isosurface of
charge, colormapped by the local potential.
See http://www.uni-due.de/~hp0058/?file=manual03.html&dir=vmdplugins for examples of using VMD for visualization.
3.3.7

Bader analysis

Bader analysis is a charge partitioning scheme where charge is divided by surfaces of zero flux that
define atomic basins of charge. The most modern way of calculating the Bader charges is using the
bader program from Graeme Henkelmens group. 42,43 Let us consider a water molecule, centered in a
box. The strategy is first to run the calculation, then run the bader program on the results.
1
2
3
4

from jasp import *

from ase.structure import molecule
atoms = molecule(H2O)
atoms.center(vacuum=6)

5
6

with jasp(molecules/h2o-bader,

42

7
8
9
10

xc=PBE,
encut=350,
atoms=atoms) as calc:
calc.calculate()

11
12

os.system(bader -p all_atom -p atom_index CHG)

Open the python script (dft-scripts/script-34.py).

GRID BASED BADER ANALYSIS

(Version 0.27e 05/09/11)

OPEN ... CHG

VASP-STYLE INPUT FILE
VASP5 format
FFT-grid:
112 x 128 x 120
CLOSE ... CHG
RUN TIME:

0.74 SECONDS

CALCULATING BADER CHARGE DISTRIBUTION

0 10 25 50 75 100
PERCENT DONE: **********************
REFINING AUTOMATICALLY
ITERATION: 1
EDGE POINTS:
1400048
REASSIGNED POINTS: 111247
ITERATION: 2
CHECKED POINTS:
913174
REASSIGNED POINTS:
1424
ITERATION: 3
CHECKED POINTS:
15784
REASSIGNED POINTS:
362
ITERATION: 4
CHECKED POINTS:
3190
REASSIGNED POINTS:
176
ITERATION: 5
CHECKED POINTS:
1228
REASSIGNED POINTS:
99
ITERATION: 6
CHECKED POINTS:
688
REASSIGNED POINTS:
21
ITERATION: 7
CHECKED POINTS:
267
REASSIGNED POINTS:
1
ITERATION: 8
CHECKED POINTS:
26
REASSIGNED POINTS:
0
RUN TIME:

26.40 SECONDS

CALCULATING MINIMUM DISTANCES TO ATOMS

0 10 25 50 75 100
PERCENT DONE: **********************

43

RUN TIME:

1.29 SECONDS

WRITING BADER ATOMIC CHARGES TO ACF.dat

WRITING BADER VOLUME CHARGES TO BCF.dat
NUMBER OF BADER MAXIMA FOUND:
SIGNIFICANT MAXIMA FOUND:
NUMBER OF ELECTRONS:

5407
10
7.99999

WRITING ATOMIC VOLUMES

0 10 25 50 75 100
PERCENT DONE: **********************
RUN TIME:

3.96 SECONDS

WRITING BADER ATOMIC INDEX TO AtIndex.dat

1

cat ACF.dat

Open the python script (dft-scripts/script-35.py).

#
X
Y
Z
CHARGE
MIN DIST
ATOMIC VOL
-------------------------------------------------------------------------------1
6.0000
6.7632
6.5963
7.9990
2.2193
644.5831
2
6.0000
7.5265
6.0000
0.0007
1.6422
759.8654
3
6.0000
6.0000
6.0000
0.0003
1.6422
640.1559
-------------------------------------------------------------------------------VACUUM CHARGE:
0.0000
VACUUM VOLUME:
0.0000
NUMBER OF ELECTRONS:
8.0000
This script should work, but only if displacement is set to None or we convert the positions to Bohr.
Otherwise, the positions do not match because of a Bohr/Angstrom units problem in the code. Note the
sorting that has to be done, because the order of atoms in the Atoms object is not the same as in the
POSCAR file.
1
2
3

from jasp import *

from ase.io.bader import attach_charges
from ase.units import Bohr

4
5
6

with jasp(molecules/h2o-bader) as calc:

atoms = calc.get_atoms()

7
8
9
10

symbols = np.array(atoms.get_chemical_symbols())[calc.sort]
pos = atoms.positions[calc.sort] * Bohr
newatoms = Atoms(symbols, positions=pos, cell=atoms.get_cell())

11
12

attach_charges(newatoms, ACF.dat)

13
14
15
16
17
18
19

print #+tblname: bader

print #+caption: Bader charges for a water molecule
print | atom | Bader charge|
print |-
for atom in newatoms:
print |{0} | {1} |.format(atom.symbol, atom.charge)

Open the python script (dft-scripts/script-36.py).

44

Table 2: Bader charges for a water molecule

atom Bader charge
O
0.001
H
0.9993
H
0.9997

3.4

Geometry optimization

3.4.1

Manual determination of a bond length

The equilibrium bond length of a CO molecule is approximately the bond length that minimizes the
total energy. We can find that by computing the total energy as a function of bond length, and noting
where the minimum is. Here is an example in VASP. There are a few features to point out here. We want
to compute 5 bond lengths, and each calculation is independent of all the others. mod:jasp is set up to
automatically handle jobs for you by submitting them to the queue. It raises a variety of exceptions to
let you know what has happened, and you must handle these to control the workflow. We will illustrate
this by the following examples.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6

bond_lengths = [1.05, 1.1, 1.15, 1.2, 1.25]

energies = []

7
8

for d in bond_lengths:

# possible bond lengths

9
10
11
12

co = Atoms([Atom(C, [0, 0, 0]),

Atom(O, [d, 0, 0])],
cell=(6, 6, 6))

13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

with jasp(molecules/co-{0}.format(d), # output dir

xc=PBE,
nbands=6,
encut=350,
ismear=1,
sigma=0.01,
atoms=co):
try:
e = co.get_potential_energy()
energies.append(e)
print d = {0:1.2f} ang.format(d)
print energy = {0:1.3f} eV.format(e)
print forces = (eV/ang)\n {0}.format(co.get_forces())
print # blank line
except (VaspSubmitted, VaspQueued):
energies.append(None)
pass

31
32
33
34
35
36
37

if None not in energies:

import matplotlib.pyplot as plt
plt.plot(bond_lengths, energies, bo-)
plt.xlabel(rBond length ($\AA$))
plt.ylabel(Total energy (eV))
plt.savefig(images/co-bondlengths.png)

Open the python script (dft-scripts/script-37.py).

d = 1.05 ang
energy = -14.215 eV
forces = (eV/ang)
[[-14.903
0.
0.
[ 14.903
0.
0.

]
]]

45

d = 1.10 ang
energy = -14.720 eV
forces = (eV/ang)
[[-5.8 0.
0. ]
[ 5.8 0.
0. ]]
d = 1.15 ang
energy = -14.838 eV
forces = (eV/ang)
[[ 0.645 0.
0.
[-0.645 0.
0.

]
]]

d = 1.20 ang
energy = -14.688 eV
forces = (eV/ang)
[[ 5.095 0.
0.
[-5.095 0.
0.

]
]]

d = 1.25 ang
energy = -14.352 eV
forces = (eV/ang)
[[ 8.141 0.
0.
[-8.141 0.
0.

]
]]

To find the minimum we could run more calculations, but a simpler and faster way is to fit a
polynomial to the data, and find the analytical minimum. The results are shown in Figure 18.
1
2
3
4

from jasp import *

import numpy as np
import matplotlib.pyplot as plt
np.set_printoptions(precision=3, suppress=True)

5
6
7

bond_lengths = [1.05, 1.1, 1.15, 1.2, 1.25]

energies = []

8
9

for d in bond_lengths:

# possible bond lengths

10
11
12
13

with jasp(molecules/co-{0}.format(d)) as calc:

atoms = calc.get_atoms()
energies.append(atoms.get_potential_energy())

14
15
16
17

# Now we fit an equation - cubic polynomial

pp = np.polyfit(bond_lengths, energies, 3)
dp = np.polyder(pp) # first derivative - quadratic

18
19
20
21

# we expect two roots from the quadratic eqn. These are where the
# first derivative is equal to zero.
roots = np.roots(dp)

22
23
24
25

# The minimum is where the second derivative is positive.

dpp = np.polyder(dp) # second derivative - line
secd = np.polyval(dpp, roots)

26
27
28

minV = roots[secd > 0]

minE = np.polyval(pp, minV)

29
30

print The minimum energy is {0} eV at V = {1} Ang^3.format(minE, minV)

31
32
33
34

# plot the fit

x = np.linspace(1.05, 1.25)
fit = np.polyval(pp, x)

35
36
37
38
39

plt.plot(bond_lengths, energies, bo )
plt.plot(x, fit, r-)
plt.plot(minV, minE, m* )
plt.legend([DFT, fit, minimum], numpoints=1)

46

40
41
42
43

plt.xlabel(rBond length ($\AA$))

plt.ylabel(Total energy (eV))
plt.savefig(images/co-bondlengths.png)
plt.show()

Open the python script (dft-scripts/script-38.py).

The minimum energy is [-14.843] eV at V = [ 1.144] Ang^3

Figure 18: Energy vs CO bond length.

3.4.2

Automatic geometry optimization with VASP

It is generally the case that the equilibrium geometry of a system is the one that minimizes the total
energy and forces. Since each atom has three degrees of freedom, you can quickly get a high dimensional
optimization problem. Luckily, VASP has built-in geometry optimization using the IBRION and NSW tags.
Here we compute the bond length for a CO molecule, letting VASP do the geometry optimization for us.
Here are the most common choices for IBRION.
IBRION value
1
2

algorithm
quasi-Newton (use if initial guess is good)
conjugate gradient

47

Note:
VASP applies a criteria for stopping a geometry optimization. When the change in
energy between two steps is less than 0.001 eV (or 10*EDIFF), the relaxation is stopped.
This criteria is controlled by the incar:EDIFFG tag. If you prefer to stop based on forces,
set EDIFFG=-0.05, i.e. to a negative number. The units of force is eV/. For most
work, a force tolerance of 0.05 eV/ is usually sufficient.
1
2

from ase import Atom, Atoms

from jasp import *

3
4
5
6

co = Atoms([Atom(C,[0, 0, 0]),
Atom(O,[1.2, 0, 0])],
cell=(6, 6, 6))

7
8
9
10
11
12
13
14
15
16

with jasp(molecules/co-cg,
xc=PBE,
nbands=6,
encut=350,
ismear=1,
sigma=0.01, # this is small for a molecule
ibrion=2,
# conjugate gradient optimizer
nsw=5,
# do at least 5 steps to relax
atoms=co) as calc:

17
18
19
20

print Forces
print =======
print co.get_forces()

21
22
23
24

pos = co.get_positions()
d = ((pos[0] - pos[1])**2).sum()**0.5
print Bondlength = {0:1.2f} angstroms.format(d)

Open the python script (dft-scripts/script-39.py).

3.4.3

Relaxation of a water molecule

It is not more complicated to relax more atoms, it just may take longer because there are more electrons
and degrees of freedom. Here we relax a water molecule which has three atoms.
1
2

from ase import Atoms, Atom

from jasp import *

3
4
5
6
7

atoms = Atoms([Atom(H, [0.5960812, -0.7677068, 0.0000000]),

Atom(O, [0.0000000, 0.0000000, 0.0000000]),
Atom(H, [0.5960812, 0.7677068, 0.0000000])],
cell=(8, 8, 8))

8
9
10
11
12
13
14
15
16

with jasp(molecules/h2o_relax,
xc=PBE,
encut=400,
ismear=0, # Gaussian smearing
ibrion=2,
ediff=1e-8,
nsw=10,
atoms=atoms) as calc:

17
18
19
20

print "forces"
print =======
print atoms.get_forces()

Open the python script (dft-scripts/script-40.py).

forces
=======
[[ 0.
-0.
[-0.001 0.
[ 0.
0.

0.
0.
0.

]
]
]]
48

3.5

Vibrational frequencies

3.5.1

Manual calculation of vibrational frequency

The principle idea in calculating vibrational frequencies is that we consider a molecular system as masses
connected by springs. If the springs are Hookean, e.g. the force is proportional to the displacement, then
we can readily solve the equations of motion and find that the vibrational frequencies are related to the
force constants and the masses of the atoms. For example, in a simple molecule like CO where there is
only one spring,
the frequency is:
p
1
k/ where 1 = m1C + m1O and k is the spring constant. We will compute the value of k from
= 2
DFT calculations as follows:
2
k = xE2 at the equilibrium bond length. We actually already have the data to do this from Manual
determination. We only need to fit an equation to the energy vs. bond-length data, find the minimum
energy bond-length, and then evaluate the second derivative of the fitted function at the minimum. We
will use a cubic polynomial for demonstration here. Polynomials are numerically convenient because
they are easy to fit, and it is trivial to get the roots and derivatives of the polynomials, as well as to
evaluate them at other points using func:numpy.polyfit, func:numpy.polyder, and func:numpy.polyval.
1
2

from jasp import *

from ase.units import *

3
4
5

bond_lengths = [1.05, 1.1, 1.15, 1.2, 1.25]

energies = []

6
7
8
9
10

for d in bond_lengths:
with jasp(molecules/co-{0}.format(d)) as calc:
atoms = calc.get_atoms()
energies.append(atoms.get_potential_energy())

11
12
13
14
15

# fit the data

pars = np.polyfit(bond_lengths, energies, 3)
xfit = np.linspace(1.05, 1.25)
efit = np.polyval(pars, xfit)

16
17
18
19
20
21

# first derivative
dpars = np.polyder(pars)
# find where the minimum is. chose the second one because it is the
# minimum we need.
droots = np.roots(dpars)

22
23
24

# second derivative
ddpars = np.polyder(dpars)

25
26

d_min = droots[np.polyval(ddpars, droots) > 0]

27
28
29

# curvature at minimum = force constant in SI units

k = np.polyval(ddpars, d_min) / (J / m**2)

30
31
32
33
34

# mu, reduced mass

from ase.data import atomic_masses
C_mass = atomic_masses[6] / kg
O_mass = atomic_masses[8] / kg

35
36

mu = 1.0 / (1.0 / C_mass + 1.0 / O_mass)

37
38
39
40

frequency = 1. / (2. * np.pi) * np.sqrt(k / mu)

print The CO vibrational frequency is {0} Hz.format(*frequency)
print The CO vibrational frequency is {0} cm^{{-1}}.format(frequency/3e10)

41
42
43
44
45
46
47

import matplotlib.pyplot as plt

plt.plot(bond_lengths, energies, bo )
plt.plot(xfit, efit, b-)
plt.xlabel(Bond length ($\AA$))
plt.ylabel(Total energy (eV))
plt.show()

Open the python script (dft-scripts/script-41.py).

The CO vibrational frequency is 6.43082289837e+13 Hz
49

The CO vibrational frequency is [ 2143.608] cm^{-1}

This result is in good agreement with experiment. The procedure described above is basically how
many vibrational calculations are performed. With more atoms, you have to determine a force constant
matrix and diagonalize it. For more details, see. 44 In practice, we usually allow a packaged code to
automate this, which we cover in Automated vibrational calculations.
We now consider how much energy is in this vibration. This is commonly called zero-point energy
(ZPE) and it is defined as EZP E = 21 h for a single mode, and h is Plancks constant (4.135667516e-15
eV/s).
1
2

c = 3e10 # speed of light cm/s

h = 4.135667516e-15 # eV*s

3
4

nu = 2143.6076625*c

# 1/s

5
6

E_zpe = 0.5*h*nu

7
8

print E_ZPE = {0:1.3f} eV.format(E_zpe)

Open the python script (dft-scripts/script-42.py).

E_ZPE = 0.133 eV
This is a reasonable amount of energy! Zero-point energy increases with increasing vibrational frequency, and tends to be very important for small atoms.
A final note is that this analysis is in the "harmonic approximation". The frequency equation is the
solution to a harmonic oscillator. If the spring is non-linear, then there are anharmonic effects that may
become important, especially at higher temperatures.
3.5.2

Automated vibrational calculations

VASP has built-in capability for performing vibrational calculations. We access the capability by using a
new value for incar:IBRION. The values of 5 and 6 calculated the Hessian matrix using finite differences.
For IBRION=5, all atoms that are not constrained are displaced. For IBRION=6, only symmetry
inequivalent displacements are considered, which makes the calculations slightly cheaper. You can specify
the number of displacements with incar:NFREE. The default number of displacements is 2. You can
also specify the size of the displacement with incar:POTIM (the default is 0.015 ).
1
2
3
4
5

# <<water-vib>>
# adapted from http://cms.mpi.univie.ac.at/wiki/index.php/H2O_vibration
from ase import Atoms, Atom
from jasp import *
import ase.units

6
7
8
9
10
11

atoms = Atoms([Atom(H, [0.5960812,

Atom(O, [0.0000000,
Atom(H, [0.5960812,
cell=(8, 8, 8))
atoms.center()

-0.7677068,
0.0000000,
0.7677068,

0.0000000]),
0.0000000]),
0.0000000])],

12
13
14
15
16
17
18
19
20
21
22

with jasp(molecules/h2o_vib,
xc=PBE,
encut=400,
ismear=0,
# Gaussian smearing
ibrion=6,
# finite differences with symmetry
nfree=2,
# central differences (default)
potim=0.015, # default as well
ediff=1e-8,
# for vibrations you need precise energies
nsw=1,
# Set to 1 for vibrational calculation
atoms=atoms) as calc:

23
24
25
26

print Forces
print ======
print atoms.get_forces()

50

27
28
29
30
31
32

print
# vibrational energies are in eV
energies, modes = calc.get_vibrational_modes()
print energies\n========
for i, e in enumerate(energies):
print {0:02d}: {1} eV.format(i, e)

Open the python script (dft-scripts/script-43.py).

Forces
======
[[ 0.018 -0.032 -0.001]
[-0.035 0.
0.003]
[ 0.018 0.032 -0.001]]
energies
========
00: 0.475824705 eV
01: 0.461734211 eV
02: 0.19618999 eV
03: 0.007093876 eV
04: 0.002459836 eV
05: (0.000291856+0j)
06: (0.012642767+0j)
07: (0.012959493+0j)
08: (0.015883929+0j)

eV
eV
eV
eV

Note we get 9 frequencies here. Water has 3 atoms, with three degrees of freedom each, leading to 9
possible combinations of collective motions. Three of those collective motions are translations, i.e. where
all atoms move in the same direction (either x, y or z) and there is no change in the total energy of
the molecule. Another three of those motions are rotations, which also do not change the total energy
of the molecule. That leaves 3N-6 = 3 degrees of vibrational freedom where some or all of the bonds
are stretched, resulting in a change in the total energy. The modes of water vibration are (with our
calculated values in parentheses):
1. a symmetric stretch at 3657 cm-1 (3723)
2. an asymmetric stretch at 3756 cm-1 (3836)
3. and a bending mode at 1595 cm-1 (1583)
http://webbook.nist.gov/cgi/cbook.cgi?ID=C7732185&Mask=800#Electronic-Spec
The results are not too far off, and more accurate frequencies may be possible using tighter tolerance
on POTIM, or by using IBRION=7 or 8.
Let us briefly discuss how to determine which vectors are vibrations and which are rotations or translations. One way is to visualize the modes. The vibrations are easy to spot. The rotations/translations
are not always cleanly separable. This is an issue of accuracy and convergence. We usually do not worry
about this because these modes are usually not important.
1. mode 0 is an asymmetric stretch
2. mode 1 is a symmetric stretch
3. mode 2 is a bending mode
4. mode 3 is a mixed translation/rotation
5. mode 4 is a rotation
51

6. mode 5 is a translation
7. mode 6 is a rotation
8. mode 7 is a partial translation
9. mode 8 is a rotation
1
2
3
4
5
6

# <<h2o-vib-vis>>
from jasp import *
import numpy as np
with jasp(molecules/h2o_vib) as calc:
energies, modes = calc.get_vibrational_modes(mode=3, massweighted=True,
show=True)

Open the python script (dft-scripts/script-44.py).

See http://www.gaussian.com/g_whitepap/vib.htm for a more quantitative discussion of these
modes, identifying them, and a method to project the rotations and translations out of the Hessian
matrix.
Zero-point energy for multiple modes For a molecule with lots of vibrational modes the zero-point
energy is defined
P as the sum over all the vibrational modes:
EZP E = i 21 hi
Here is an example for water. Note we do not sum over the imaginary modes. We should also ignore
the rotational and translational modes (some of those are imaginary, but some are just small).
1
2
3
4

from jasp import *

import numpy as np
c = 3e10 # speed of light cm/s
h = 4.135667516e-15 # eV/s

5
6
7
8

# first, get the frequencies.

with jasp(molecules/h2o_vib) as calc:
freq = calc.get_vibrational_frequencies()

9
10
11
12
13
14
15

ZPE = 0.0
for f in freq:
if not isinstance(f, float):
continue # skip complex numbers
nu = f * c # convert to frequency
ZPE += 0.5 * h * nu

16
17

print The ZPE of water is {0:1.3f} eV.format(ZPE)

18
19
20
21

# one liner
ZPE = np.sum([0.5 * h * f * c for f in freq if isinstance(f, float)])
print The ZPE of water is {0:1.3f} eV.format(ZPE)

Open the python script (dft-scripts/script-45.py).

The ZPE of water is 0.572 eV
Note the zero-point energy of water is also fairly high (more than 0.5 eV). That is because of the high
frequency O-H stretches.

3.6

Simulated infrared spectra

At http://homepage.univie.ac.at/david.karhanek/downloads.html#Entry02 there is a recipe for

computing the Infrared vibrational spectroscopy intensities in VASP. We are going to do that for water
here. First, we will relax a water molecule.

52

1
2

from ase import Atoms, Atom

from jasp import *

3
4
5
6
7

atoms = Atoms([Atom(H, [0.5960812,

Atom(O, [0.0000000,
Atom(H, [0.5960812,
cell=(8, 8, 8))

-0.7677068,
0.0000000,
0.7677068,

0.0000000]),
0.0000000]),
0.0000000])],

8
9
10
11
12
13
14
15
16
17
18
19

with jasp(molecules/h2o_relax,
xc=PBE,
encut=400,
ismear=0, # Gaussian smearing
ibrion=2,
ediff=1e-8,
nsw=10,
atoms=atoms) as calc:
print Forces
print ===========================
print atoms.get_forces()

Open the python script (dft-scripts/script-46.py).

Forces
===========================
[[ 0.
-0.
0.
]
[-0.001 0.
0.
]
[ 0.
0.
0.
]]
Next, we instruct VASP to compute the vibrational modes using density functional perturbation
theory with IBRION=7. Note, this is different than in 3.5 where finite differences were used.
1

from jasp import *

2
3
4
5

# read in relaxed geometry

with jasp(molecules/h2o_relax) as calc:
atoms = calc.get_atoms()

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

# now define a new calculator

with jasp(molecules/h2o_vib_dfpt,
xc=PBE,
encut=400,
ismear=0, # Gaussian smearing
ibrion=7, # switches on the DFPT vibrational analysis (with
# no symmetry constraints)
nfree=2,
potim=0.015,
lepsilon=True, # enables to calculate and to print the BEC
# tensors
lreal=False,
nsw=1,
nwrite=3, # affects OUTCAR verbosity: explicitly forces
# SQRT(mass)-divided eigenvectors to be printed
atoms=atoms) as calc:
calc.calculate(atoms)

Open the python script (dft-scripts/script-47.py).

To analyze the results, this shell script was provided to extract the results.
1
2
3

#!/bin/bash
# A utility for calculating the vibrational intensities from VASP output (OUTCAR)
# (C) David Karhanek, 2011-03-25, ICIQ Tarragona, Spain (www.iciq.es)

4
5
6
7
8
9
10

# extract Born effective charges tensors

printf "..reading OUTCAR"
BORN_NROWS=grep NIONS OUTCAR | awk {print $12*4+1}
if [ grep BORN OUTCAR | wc -l = 0 ] ; then \
printf " .. FAILED! Born effective charges missing! Bye! \n\n" ; exit 1 ; fi
grep "in e, cummulative" -A $BORN_NROWS OUTCAR > born.txt

11

53

12
13
14
15
16
17
18
19

# extract Eigenvectors and eigenvalues

if [ grep SQRT(mass) OUTCAR | wc -l != 1 ] ; then \
printf " .. FAILED! Restart VASP with NWRITE=3! Bye! \n\n" ; exit 1 ; fi
EIG_NVIBS=grep -A 2000 SQRT(mass) OUTCAR | grep cm-1 | wc -l
EIG_NIONS=grep NIONS OUTCAR | awk {print $12}
EIG_NROWS=echo "($EIG_NIONS+3)*$EIG_NVIBS+3" | bc
grep -A $(($EIG_NROWS+2)) SQRT(mass) OUTCAR | tail -n $(($EIG_NROWS+1)) | sed s/f\/i/fi /g > eigenvectors.txt
printf " ..done\n"

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

# set up a new directory, split files - prepare for parsing

printf "..splitting files"
mkdir intensities ; mv born.txt eigenvectors.txt intensities/
cd intensities/
let NBORN_NROWS=BORN_NROWS-1
let NEIG_NROWS=EIG_NROWS-3
let NBORN_STEP=4
let NEIG_STEP=EIG_NIONS+3
tail -n $NBORN_NROWS born.txt > temp.born.txt
tail -n $NEIG_NROWS eigenvectors.txt > temp.eige.txt
mkdir inputs ; mv born.txt eigenvectors.txt inputs/
split -a 3 -d -l $NEIG_STEP temp.eige.txt temp.ei.
split -a 3 -d -l $NBORN_STEP temp.born.txt temp.bo.
mkdir temps01 ; mv temp.born.txt temp.eige.txt temps01/
for nu in seq 1 $EIG_NVIBS ; do
let nud=nu-1 ; ei=printf "%03u" $nu ; eid=printf "%03u" $nud ; mv temp.ei.$eid eigens.vib.$ei
done
for s in seq 1 $EIG_NIONS ; do
let sd=s-1 ; bo=printf "%03u" $s ; bod=printf "%03u" $sd ; mv temp.bo.$bod borncs.$bo
done
printf " ..done\n"

42
43
44
45
46
47
48
49
50
51
52
53
54

# parse deviation vectors (eig)

printf "..parsing eigenvectors"
let sad=$EIG_NIONS+1
for nu in seq 1 $EIG_NVIBS ; do
nuu=printf "%03u" $nu
tail -n $sad eigens.vib.$nuu | head -n $EIG_NIONS | awk {print $4,$5,$6} > e.vib.$nuu.allions
split -a 3 -d -l 1 e.vib.$nuu.allions temp.e.vib.$nuu.ion.
for s in seq 1 $EIG_NIONS ; do
let sd=s-1; bo=printf "%03u" $s; bod=printf "%03u" $sd; mv temp.e.vib.$nuu.ion.$bod e.vib.$nuu.ion.$bo
done
done
printf " ..done\n"

55
56
57
58
59
60
61
62
63

# parse born effective charge matrices (born)

printf "..parsing eff.charges"
for s in seq 1 $EIG_NIONS ; do
ss=printf "%03u" $s
awk {print $2,$3,$4} borncs.$ss | tail -3 > bornch.$ss
done
mkdir temps02 ; mv eigens.* borncs.* temps02/
printf " ..done\n"

64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86

# parse matrices, multiply them and collect squares (giving intensities)

printf "..multiplying matrices, summing "
for nu in seq 1 $EIG_NVIBS ; do
nuu=printf "%03u" $nu
int=0.0
for alpha in 1 2 3 ; do
# summing over alpha coordinates
sumpol=0.0
for s in seq 1 $EIG_NIONS ; do
# summing over atoms
ss=printf "%03u" $s
awk -v a="$alpha" (NR==a){print} bornch.$ss > z.ion.$ss.alpha.$alpha
# summing over beta coordinates and multiplying Z(s,alpha)*e(s) done by the following awk script
paste z.ion.$ss.alpha.$alpha e.vib.$nuu.ion.$ss | \
awk {pol=$1*$4+$2*$5+$3*$6; print $0," ",pol} > matr-vib-${nuu}-alpha-${alpha}-ion-${ss}
done
sumpol=cat matr-vib-${nuu}-alpha-${alpha}-ion-* | awk {sum+=$7} END {print sum}
int=echo "$int+($sumpol)^2" | sed s/[eE]/*10^/g | bc -l
done
freq=awk (NR==1){print $8} temps02/eigens.vib.$nuu
echo "$nuu $freq $int">> exact.res.txt
printf "."
done
printf " ..done\n"

87
88
89

# format results, normalize intensities

printf "..normalizing intensities"

54

90
91
92

max=awk (NR==1){max=$3} $3>=max {max=$3} END {print max} exact.res.txt

awk -v max="$max" {printf "%03u %6.1f %5.3f\n",$1,$2,$3/max} exact.res.txt > results.txt
printf " ..done\n"

93
94
95
96
97
98
99
100
101
102
103
104
105

# clean up, display results

printf "..finalizing:\n"
mkdir temps03; mv bornch.* e.vib.*.allions temps03/
mkdir temps04; mv z.ion* e.vib.*.ion.* temps04/
mkdir temps05; mv matr-* temps05/
mkdir results; mv *res*txt results/
let NMATRIX=$EIG_NVIBS**2
printf "%5u atoms found\n%5u vibrations found\n%5u matrices evaluated" \
$EIG_NIONS $EIG_NVIBS $NMATRIX > results/statistics.txt
# fast switch to clean up all temporary files
rm -r temps*
cat results/results.txt

Open the python script (dft-scripts/script-48.py).

..reading OUTCAR ..done
..splitting files ..done
..parsing eigenvectors ..done
..parsing eff.charges ..done
..multiplying matrices, summing ......... ..done
..normalizing intensities ..done
..finalizing:
001 3827.3 0.227
002 3713.0 0.006
003 1587.2 0.312
004 235.5 1.000
005
19.1 0.006
006
2.3 0.000
007
16.6 0.005
008
45.0 0.000
009 136.1 0.345
Note that the results above include the rotational and translational modes (modes 4-9). The following
shell script removes those, and recalculates the intensities. Note that it appears to just remove the last
6 modes and req compute the intensities. It is not obvious that will always be the right way to do it as
the order of the eigenvectors is not guaranteed.
1
2
3
4
5
6
7
8
9
10
11
12
13

#!/bin/bash
# reformat intensities, just normal modes: 3N -> (3N-6)
printf "..reformatting and normalizing intensities"
cd intensities/results/
nlns=wc -l exact.res.txt | awk {print $1} ; let bodylns=nlns-6
head -n $bodylns exact.res.txt > temp.reform.res.txt
max=awk (NR==1){max=$3} $3>=max {max=$3} END {print max} temp.reform.res.txt
awk -v max="$max" {print $1,$2,$3/max} temp.reform.res.txt > exact.reform.res.txt
awk -v max="$max" {printf "%03u %6.1f %5.3f\n",$1,$2,$3/max} temp.reform.res.txt > reform.res.txt
printf " ..done\n..normal modes:\n"
rm temp.reform.res.txt
cat reform.res.txt
cd ../..

Open the python script (dft-scripts/script-49.py).

..reformatting and normalizing intensities ..done
..normal modes:
001 3827.3 0.726
002 3713.0 0.019
003 1587.2 1.000
55

The interpretation of these results is that the mode at 3713 cm-1 would be nearly invisible in the
IR spectrum. Earlier we interpreted that as the symmetric stretch. In this mode, there is only a small
change in the molecule dipole moment, so there is a small IR intensity.
See also. 45 For HREELS simulations see. 46
The shell script above has been translated to a convenient python function in mod:jasp.
1
2
3
4
5

from jasp import *

with jasp(molecules/h2o_vib_dfpt) as calc:
print mode Relative intensity
for i, intensity in enumerate(calc.get_infrared_intensities()):
print {0:02d}
{1:1.3f}.format(i, intensity)

Open the python script (dft-scripts/script-50.py).

mode
00
01
02
03
04
05
06
07
08

Relative intensity
0.227
0.006
0.312
1.000
0.006
0.000
0.005
0.000
0.345

3.7

Thermochemical properties of molecules

mod:ase.thermochemistry can be used to estimate thermodynamic properties of gases in the ideal gas
limit. The module needs as input the geometry, the total energy, the vibrational energies, and some
information about the molecular symmetry. We first consider an N2 molecule.
The symmetry numbers are determined by the molecular point group. 47 Here is a table of the most
common ones.
Table 3: Symmetry numbers for common point groups
point group
examples
C1
1
Cs
1
C2
2
C2v
2 H2 O
C3v
3 NH3
Cv
1 CO
D2h
4
D3h
6
D5h
10
Dh
2 CO2 , H2
D3d
6
Td
12 CH4
Oh
24

1
2
3

from ase.structure import molecule

from ase.thermochemistry import IdealGasThermo
from jasp import *

4
5

atoms = molecule(N2)

56

atoms.set_cell((10,10,10), scale_atoms=False)

7
8
9
10
11
12
13
14
15

# first we relax a molecule

with jasp(molecules/n2-relax,
xc=PBE,
encut=300,
ibrion=2,
nsw=5,
atoms=atoms) as calc:
electronicenergy = atoms.get_potential_energy()

16
17
18
19
20
21
22
23
24
25
26
27

# next, we get vibrational modes

with jasp(molecules/n2-vib,
xc=PBE,
encut=300,
ibrion=6,
nfree=2,
potim=0.15,
nsw=1,
atoms=atoms) as calc:
calc.calculate()
vib_freq = calc.get_vibrational_frequencies() # in cm^1

28
29
30
31
32
33
34
35

#convert wavenumbers to energy

h = 4.1356675e-15 # eV*s
c = 3.0e10 #cm/s
vib_energies = [h*c*nu for nu in vib_freq]
print vibrational energies\n====================
for i,e in enumerate(vib_energies):
print {0:02d}: {1} eV.format(i,e)

36
37
38
39
40
41
42

# # now we can get some properties. Note we only need one vibrational
# energy since there is only one mode. This example does not work if
# you give all the energies because one energy is zero.
thermo = IdealGasThermo(vib_energies=vib_energies[0:0],
electronicenergy=electronicenergy, atoms=atoms,
geometry=linear, symmetrynumber=2, spin=0)

43
44
45

# temperature in K, pressure in Pa, G in eV

G = thermo.get_free_energy(temperature=298.15, pressure=101325.)

Open the python script (dft-scripts/script-51.py).

vibrational energies
====================
00: 0.29159234324 eV
01: 0.0169775241059 eV
02: 0.0169775241059 eV
03: 2.853610575e-09 eV
04: 8.68490175e-10 eV
05: 0.0 eV
Enthalpy components at T = 298.15 K:
===============================
E_elec
-16.478 eV
E_ZPE
0.000 eV
Cv_trans (0->T)
0.039 eV
Cv_rot (0->T)
0.026 eV
Cv_vib (0->T)
0.000 eV
(C_v -> C_p)
0.026 eV
------------------------------H
-16.388 eV
===============================
Entropy components at T = 298.15 K and P = 101325.0 Pa:
=================================================
S
T*S
57

S_trans (1 atm)
0.0015579 eV/K
0.464 eV
S_rot
0.0007870 eV/K
0.235 eV
S_elec
0.0000000 eV/K
0.000 eV
S_vib
0.0000000 eV/K
0.000 eV
S (1 atm -> P)
-0.0000000 eV/K
-0.000 eV
------------------------------------------------S
0.0023449 eV/K
0.699 eV
=================================================
Free energy components at T = 298.15 K and P = 101325.0 Pa:
=======================
H
-16.388 eV
-T*S
-0.699 eV
----------------------G
-17.087 eV
=======================
Let us compare this to what is in the Nist webbook via the Shomate equations.
1
2
3
4
5
6
7
8
9

import numpy as np
A = 28.98641
B = 1.853978
C = -9.647459
D = 16.63537
E = 0.000117
F = -8.671914
G = 226.4168
H = 0.0

10
11
12

T = 298.15
t = T/1000.

13
14
15

S = A*np.log(t) + B*t + C*t**2/2 + D*t**3/3 - E/(2*t**2) + G

print -T*S = {0:1.3f} eV.format(-T*S/1000/96.4853)

Open the python script (dft-scripts/script-52.py).

-T*S = -0.592 eV
This is reasonable agreement for the entropy. You will get different results if you use different exchange
correlation functionals.

3.8
3.8.1

Molecular reaction energies

O2 dissociation

The first reaction we consider is a simple dissociation of oxygen molecule into two oxygen atoms: O2
2O. The dissociation energy is pretty straightforward to define: it is the energy of the products minus
the energy of the reactant. D = 2 EO EO2 . It would appear that we simply calculate the energy of
an oxygen atom, and the energy of an oxygen molecule and evaluate the formula. Let us do that.
Simple estimate of O2 dissociation energy
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

atoms = Atoms([Atom(O, [5, 5, 5])],

cell=(10, 10, 10))

6
7
8

with jasp(molecules/O,
xc=PBE,

58

encut=400,
ismear=0,
atoms=atoms) as calc:

9
10
11
12

try:

13

E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

14
15
16
17
18
19
20

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5, 5, 5]),
Atom(O, [6.22, 5, 5])],
cell=(10, 10, 10))

21
22
23
24
25
26
27
28
29
30
31
32

with jasp(molecules/O2,
xc=PBE,
encut=400,
ismear=0,
ibrion=2,
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

33
34
35

if None not in (E_O, E_O2):

print O2 -> 2O D = {0:1.3f} eV.format(2 * E_O - E_O2)

Open the python script (dft-scripts/script-53.py).

O2 -> 2O

D = 8.521 eV

The answer we have obtained is way too high! Experimentally the dissociation energy is about 5.2
eV (need reference), which is very different than what we calculated! Let us consider some factors that
contribute to this error.
We implicitly neglected spin-polarization in the example above. That could be a problem, since the
O2 molecule can be in one of two spin states, a singlet or a triplet, and these should have different
energies. Furthermore, the oxygen atom can be a singlet or a triplet, and these would have different
energies. To account for spin polarization, we have to tell VASP to use spin-polarization, and give initial
guesses for the magnetic moments of the atoms. Let us try again with spin polarization.
Estimating O2 dissociation energy with spin polarization in triplet ground states To tell
VASP to use spin-polarization we use ISPIN=2, and we set initial guesses for magnetic moments on
the atoms with the magmom keyword. In a triplet state there are two electrons with spins of the same
sign.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

atoms = Atoms([Atom(O, [5, 5, 5], magmom=2)],

cell=(10, 10, 10))

6
7
8
9
10
11
12
13
14
15
16

with jasp(molecules/O-sp-triplet,
xc=PBE,
encut=400,
ismear=0,
ispin=2, # turn spin-polarization on
atoms=atoms) as calc:
try:
E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

17
18

print Magnetic moment on O = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

19
20
21
22
23

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5,
5, 5], magmom=1),
Atom(O, [6.22, 5, 5], magmom=1)],
cell=(10, 10, 10))

59

24
25
26
27
28
29
30
31
32
33
34
35
36

with jasp(molecules/O2-sp-triplet,
xc=PBE,
encut=400,
ismear=0,
ispin=2,
# turn spin-polarization on
ibrion=2, # make sure we relax the geometry
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

37
38
39

# verify magnetic moment

print Magnetic moment on O2 = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

40
41
42

if None not in (E_O, E_O2):

print O2 -> 2O D = {0:1.3f} eV.format(2 * E_O - E_O2)

Open the python script (dft-scripts/script-54.py).

Magnetic moment on O = 2.0000072 Bohr magnetons
Magnetic moment on O2 = 2.0000084 Bohr magnetons
O2 -> 2O D = 6.668 eV
This is much closer to accepted literature values for the DFT-GGA O2 dissociation energy. It is
still more than 1 eV above an experimental value, but most of that error is due to the GGA exchange
correlation functional. Some additional parameters that might need to be checked for convergence are
the SIGMA value (it is probably too high for a molecule), as well as the cutoff energy. Oxygen is a
"hard" atom that requires a high cutoff energy to achieve high levels of convergence.
TODO: show what "hard or soft" is in the pseudopotential.
Looking at the two spin densities In a spin-polarized calculation there are actually two electron
densities: one for spin-up and one for spin-down. We will look at the differences in these two through
the density of states.
1
2

from jasp import *

from ase.dft.dos import *

3
4
5
6
7
8

with jasp(molecules/O2-sp-triplet) as calc:

dos = DOS(calc, width=0.2)
d_up = dos.get_dos(spin=0)
d_down = dos.get_dos(spin=1)
e = dos.get_energies()

9
10
11
12
13

ind = e <= 0.0

# integrate up to 0eV
print number of up states = {0}.format(np.trapz(d_up[ind], e[ind]))
print number of down states = {0}.format(np.trapz(d_down[ind], e[ind]))

14
15
16
17
18
19
20
21

import pylab as plt

plt.plot(e, d_up,
e, -d_down)
plt.xlabel(energy [eV])
plt.ylabel(DOS)
plt.legend([up, down])
plt.savefig(images/O2-sp-dos.png)

Open the python script (dft-scripts/script-55.py).

number of up states = 5.95489738902
number of down states = 5.00001754136
You can see in Figure 19 that there are two different densities of states for the two spins. One has
7 electrons in it (the blue lines), and the other has 5 electrons in it (the green line). The difference of
60

Figure 19: Spin-polarized DOS for the O2 molecule.

two electrons leads to the magnetic moment of 2 which we calculated earlier. Remember that only peaks
in the DOS below the Fermi level are occupied. It is customary to set the Fermi level to 0 eV in DOS
plots. The peaks roughly correspond to electrons. For example, the blue peak between -25 and -30 eV
corresponds to one electron, in a 1s orbital, where as the blue peak between -5 and -10 eV corresponds
to three electrons.
Convergence study of the O2 dissociation energy
1
2
3

from jasp import *

from ase import Atom, Atoms
encuts = [250, 300, 350, 400, 450, 500, 550]

4
5
6
7
8

D = []
for encut in encuts:
atoms = Atoms([Atom(O, [5, 5, 5], magmom=2)],
cell=(10, 10, 10))

9
10
11
12
13
14
15
16
17
18
19

with jasp(molecules/O-sp-triplet-{0}.format(encut),
xc=PBE,
encut=encut,
ismear=0,
ispin=2,
atoms=atoms) as calc:
try:
E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

20
21
22
23

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5,
5, 5], magmom=1),
Atom(O, [6.22, 5, 5], magmom=1)],

61

cell=(10, 10, 10))

24
25

with jasp(molecules/O2-sp-triplet-{0}.format(encut),
xc=PBE,
encut=encut,
ismear=0,
ispin=2,
# turn spin-polarization on
ibrion=2, # this turns relaxation on
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

26
27
28
29
30
31
32
33
34
35
36
37
38

if None not in (E_O, E_O2):

d = 2*E_O - E_O2
D.append(d)
print O2 -> 2O encut = {0}

39
40
41
42

D = {1:1.3f} eV.format(encut, d)

43
44
45
46
47
48

import matplotlib.pyplot as plt

plt.plot(encuts, D)
plt.xlabel(ENCUT (eV))
plt.ylabel(O$_2$ dissociation energy (eV))
plt.savefig(images/O2-dissociation-convergence.png)

Open the python script (dft-scripts/script-56.py).

O2
O2
O2
O2
O2
O2
O2

->
->
->
->
->
->
->

2O
2O
2O
2O
2O
2O
2O

encut
encut
encut
encut
encut
encut
encut

=
=
=
=
=
=
=

250
300
350
400
450
500
550

D
D
D
D
D
D
D

=
=
=
=
=
=
=

6.696
6.727
6.708
6.668
6.650
6.648
6.650

eV
eV
eV
eV
eV
eV
eV

Based on these results (Figure 20), you could argue the dissociation energy is converged to about 2
meV at a planewave cutoff of 450 eV, and within 50 meV at 350 eV cutoff. You have to decide what
an appropriate level of convergence is. Note that increasing the planewave cutoff significantly increases
the computational time, so you are balancing level of convergence with computational speed. It would
appear that planewave cutoff is not the cause for the discrepancy between our calculations and literature
values.
1
2
3

encuts = [250, 300, 350, 400, 450, 500, 550]

print encut (eV)
Total CPU time
print --------------------------------------------------------

4
5
6
7
8
9
10
11

for encut in encuts:

OUTCAR = molecules/O2-sp-triplet-{0}/OUTCAR.format(encut)
f = open(OUTCAR, r)
for line in f:
if Total CPU time used (sec) in line:
print {0} eV: {1}.format(encut, line)
f.close()

Open the python script (dft-scripts/script-57.py).

encut (eV)
Total CPU time
-------------------------------------------------------250 eV:
Total CPU time used (sec):

67.926

300 eV:

Total CPU time used (sec):

121.296

350 eV:

Total CPU time used (sec):

150.808

62

Figure 20: Convergence study of the O2 dissociation energy as a function of ENCUT.

400 eV:

Total CPU time used (sec):

307.465

450 eV:

Total CPU time used (sec):

199.381

500 eV:

Total CPU time used (sec):

240.705

550 eV:

Total CPU time used (sec):

266.318

Illustration of the effect of SIGMA The methodology for extrapolation of the total energy to
absolute zero is only valid for a continuous density of states at the Fermi level. 12 Consequently, it should
not be used for semiconductors, molecules or atoms. In VASP, this means a very small Fermi temperature
(SIGMA) should be used. The O2 dissociation energy as a function of SIGMA is shown in Figure 21.
A variation of nearly 0.2 eV is seen from the default Fermi temperature of kb T = 0.2 eV and the value
of kb T = 0.0001 eV. However, virtually no change was observed for a hydrogen atom or molecule or for
an oxygen molecule as a function of the Fermi temperature. It is recommended that the total energy be
calculated at several values of the Fermi temperature to make sure the total energy is converged with
respect to the Fermi temperature.
We were not careful in selecting a good value for SIGMA in the calculations above. The default value of
SIGMA is 0.2, which may be fine for metals, but it is not correct for molecules. SIGMA is the broadening
factor used to smear the electronic density of states at the Fermi level. For a metal with a continuous
density of states this is appropriate, but for molecules with discrete energy states it does not make sense.
We are somewhat forced to use the machinery designed for metals on molecules. The solution is to use a

63

very small SIGMA. Ideally you would use SIGMA=0, but that is not practical for convergence reasons,
so we try to find what is small enough. Let us examine the effect of SIGMA on the dissociation energy
here.
1
2

from jasp import *

from ase import Atom, Atoms

3
4

sigmas = [0.2, 0.1, 0.05, 0.02, 0.01, 0.001]

5
6
7
8
9

D = []
for sigma in sigmas:
atoms = Atoms([Atom(O,[5, 5, 5], magmom=2)],
cell=(10, 10, 10))

10
11
12
13
14
15
16
17
18
19
20
21

with jasp(molecules/O-sp-triplet-sigma-{0}.format(sigma),
xc=PBE,
encut=400,
ismear=0,
sigma=sigma,
ispin=2,
atoms=atoms) as calc:
try:
E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

22
23
24
25
26

# now relaxed O2 dimer

atoms = Atoms([Atom(O,[5,
5, 5],magmom=1),
Atom(O,[6.22, 5, 5],magmom=1)],
cell=(10, 10, 10))

27
28
29
30
31
32
33
34
35
36
37
38
39
40

with jasp(molecules/O2-sp-triplet-sigma-{0}.format(sigma),
xc=PBE,
encut=400,
ismear=0,
sigma=sigma,
ispin=2,
# turn spin-polarization on
ibrion=2, # make sure we relax the geometry
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

41
42
43
44
45

if None not in (E_O, E_O2):

d = 2 * E_O - E_O2
D.append(d)
print O2 -> 2O sigma = {0}

D = {1:1.3f} eV.format(sigma, d)

46
47
48
49
50
51

import matplotlib.pyplot as plt

plt.plot(sigmas, D, bo-)
plt.xlabel(SIGMA (eV))
plt.ylabel(O$_2$ dissociation energy (eV))
plt.savefig(images/O2-dissociation-sigma-convergence.png)

Open the python script (dft-scripts/script-58.py).

Clearly SIGMA has an effect, but it does not move the dissociation energy closer to the literature
values!
Estimating singlet oxygen dissociation energy Finally, let us consider the case where each species
is in the singlet state.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

atoms = Atoms([Atom(O, [5, 5, 5], magmom=0)],

cell=(10, 10, 10))

6
7
8
9

with jasp(molecules/O-sp-singlet,
xc=PBE,
encut=400,

64

Figure 21: Effect of SIGMA on the oxygen dissociation energy.

ismear=0,
ispin=2,
atoms=atoms) as calc:

10
11
12
13

try:

14

E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

15
16
17
18

print Magnetic moment on O = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

19
20
21
22
23

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5, 5, 5], magmom=1),
Atom(O, [6.22, 5, 5], magmom=-1)],
cell=(10, 10, 10))

24
25
26
27
28
29
30
31
32
33
34
35
36

with jasp(molecules/O2-sp-singlet,
xc=PBE,
encut=400,
ismear=0,
ispin=2, # turn spin-polarization on
ibrion=2, # make sure we relax the geometry
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

37
38
39

# verify magnetic moment

print O2 molecule magnetic moment = , atoms.get_magnetic_moment()

40
41
42

if None not in (E_O, E_O2):

print O2 -> 2O D = {0:1.3f} eV.format(2*E_O - E_O2)

65

Open the python script (dft-scripts/script-59.py).

Magnetic moment on O = 1.9998232 Bohr magnetons
O2 molecule magnetic moment = 0.0
O2 -> 2O D = 5.650 eV
Interestingly, VASP still found a triplet spin state on the oxygen atom, even though we guessed an
initial magnetic moment of 0. This highlights a difficulty in computing magnetic moments: you provide
an initial guess and a solution is found. The magnetic moment of a singlet state is zero, so the molecule
is correct. Also interesting is that the dissociation energy is almost equal to the experimental value. This
is probably a coincidence, and may reflect the fact that the singlet oxygen state is less stable than the
triplet state. Let us directly compare their total energies:
1

from jasp import *

2
3
4

with jasp(molecules/O2-sp-singlet) as calc:

print singlet: {0} eV.format(calc.get_atoms().get_potential_energy())

5
6
7

with jasp(molecules/O2-sp-triplet) as calc:

print triplet: {0} eV.format(calc.get_atoms().get_potential_energy())

Open the python script (dft-scripts/script-60.py).

singlet: -8.830101 eV
triplet: -9.848238 eV
You can see here the triplet state has an energy that is 1 eV more stable than the singlet state.
Estimating triplet oxygen dissociation energy with low symmetry It has been suggested that
breaking spherical symmetry of the atom can result in lower energy of the atom. The symmetry is broken
by putting the atom off-center in a box. We will examine the total energy of an oxygen atom in a few
geometries. First, let us consider variations of a square box.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6

# square box origin

atoms = Atoms([Atom(O, [0, 0, 0], magmom=2)],
cell=(10, 10, 10))

7
8
9
10
11
12
13
14
15
16
17
18

with jasp(molecules/O-square-box-origin,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
print Square box (origin): E = {0} eV.format(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

19
20
21
22

# square box center

atoms = Atoms([Atom(O, [5, 5, 5], magmom=2)],
cell=(10, 10, 10))

23
24
25
26
27
28
29
30
31
32
33

with jasp(molecules/O-square-box-center,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
print Square box (center): E = {0} eV.format(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):

66

34

pass

35
36
37
38

# square box random

atoms = Atoms([Atom(O, [2.13, 7.32, 1.11], magmom=2)],
cell=(10, 10, 10))

39
40
41
42
43
44
45
46
47
48
49
50

with jasp(molecules/O-square-box-random,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
print Square box (random): E = {0} eV.format(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-61.py).

Square box (origin): E = -1.516623 eV
Square box (center): E = -1.516623 eV
Square box (random): E = -1.515359 eV
There is no significant difference in these energies. The origin and center calculations are identical in
energy. The meV variation in the random calculation is negligible. Now, let us consider some non-square
boxes.
1
2
3

# calculate O atom energy in orthorhombic boxes

from jasp import *
from ase import Atom, Atoms

4
5
6
7

# orthorhombic box origin

atoms = Atoms([Atom(O, [0, 0, 0], magmom=2)],
cell=(8, 9, 10))

8
9
10
11
12
13
14
15
16
17
18
19

with jasp(molecules/O-orthorhombic-box-origin,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
print Orthorhombic box (origin): E = {0} eV.format(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

20
21
22
23

# orthrhombic box center

atoms = Atoms([Atom(O, [4, 4.5, 5], magmom=2)],
cell=(8, 9, 10))

24
25
26
27
28
29
30
31
32
33
34
35

with jasp(molecules/O-orthorhombic-box-center,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
print Orthorhombic box (center): E = {0} eV.format(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

36
37
38
39

# orthorhombic box random

atoms = Atoms([Atom(O, [2.13, 7.32, 1.11], magmom=2)],
cell=(8, 9, 10))

40
41
42
43
44
45

with jasp(molecules/O-orthorhombic-box-random,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,

67

ispin=2,
atoms=atoms) as calc:

46
47
48

try:

49

print Orthorhombic box (random): E = {0} eV.format(atoms.get_potential_energy())

except (VaspSubmitted, VaspQueued):
pass

50
51

Open the python script (dft-scripts/script-62.py).

Orthorhombic box (origin): E = -1.8941 eV
Orthorhombic box (center): E = -1.894124 eV
Orthorhombic box (random): E = -1.494285 eV
This is a surprisingly large difference in energy! Nearly 0.4 eV. This is precisely the amount of energy
we were in disagreement with the literature values. Surprisingly, the "random" position is higher in
energy, similar to the cubic boxes. Finally, we put this all together. We use a non-symmetric box for the
O-atom.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

atoms = Atoms([Atom(O, [5.1, 4.2, 6.1], magmom=2)],

cell=(8, 9, 10))

6
7
8
9
10
11
12
13
14
15
16
17
18

with jasp(molecules/O-sp-triplet-lowsym,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
atoms=atoms) as calc:
try:
E_O = atoms.get_potential_energy()
print Magnetic moment on O = {0} Bohr
except (VaspSubmitted, VaspQueued):
E_O = None

magnetons.format(atoms.get_magnetic_moment())

19
20
21
22
23

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5,
5, 5], magmom=1),
Atom(O, [6.22, 5, 5], magmom=1)],
cell=(10, 10, 10))

24
25
26
27
28
29
30
31
32
33
34
35
36
37

with jasp(molecules/O2-sp-triplet,
xc=PBE,
encut=400,
ismear=0,
sigma=0.01,
ispin=2,
# turn spin-polarization on
ibrion=2, # make sure we relax the geometry
nsw=10,
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
# verify magnetic moment
print Magnetic moment on O2 = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

38
39
40

except (VaspSubmitted, VaspQueued):

E_O2 = None

41
42
43
44

if None not in (E_O, E_O2):

print E_O: , E_O
print O2 -> 2O D = {0:1.3f} eV.format(2 * E_O - E_O2)

Open the python script (dft-scripts/script-63.py).

Magnetic moment on O = 2.0000016 Bohr magnetons
Magnetic moment on O2 = 2.0000084 Bohr magnetons
E_O: -1.893571
O2 -> 2O D = 6.061 eV
68

This actually agrees within 30-50 meV of reported literature values, although still nearly an eV greater
than the experimental dissociation energy. Note that with a different "random" position, we get the lower
energy for the O atom. All the disagreement we had been seeing was apparently in the O atom energy.
So, if you do not need the dissociation energy in your analysis, you will not see the error. Also note that
this error is specific to there being a spherical atom in a symmetric cell. This is not a problem for most
molecules, which are generally non-spherical.
Verifying the magnetic moments on each atom It is one thing to see the total magnetic moment
of a singlet state, and another to ask what are the magnetic moments on each atom. In VASP you must
use LORBIT=11 to get the magnetic moments of the atoms written out.
1

from jasp import *

2
3
4

with jasp(molecules/O2-sp-singlet) as calc:

calc.clone(molecules/O2-sp-singlet-magmoms)

5
6
7
8
9

with jasp(molecules/O2-sp-singlet-magmoms) as calc:

calc.set(lorbit=11)
atoms = calc.get_atoms()
magmoms = atoms.get_magnetic_moments()

10
11
12
13
14

print singlet ground state

for i, atom in enumerate(atoms):
print atom {0}: magmom = {1}.format(i, magmoms[i])
print atoms.get_magnetic_moment()

15
16
17

with jasp(molecules/O2-sp-triplet) as calc:

calc.clone(molecules/O2-sp-triplet-magmoms)

18
19
20
21
22
23
24
25
26
27

with jasp(molecules/O2-sp-triplet-magmoms) as calc:

calc.set(lorbit=11)
atoms = calc.get_atoms()
magmoms = atoms.get_magnetic_moments()
print
print triplet ground state
for i, atom in enumerate(atoms):
print atom {0}: magmom = {1}.format(i, magmoms[i])
print atoms.get_magnetic_moment()

Open the python script (dft-scripts/script-64.py).

Note the atomic magnetic moments do not add up to the total magnetic moment. The atomic
magnetic moments are not really true observable properties. The moments are determined by a projection
method that probably involves a spherical orbital, so the moments may be over or underestimated.
Using a different potential It is possible we need a higher quality potential to get the 6.02 eV value
quoted by many in the literature. Here we try the O_sv potential, which treats the 1s electrons as
valence electrons. Note however, the ENMIN in the POTCAR is very high!
1

grep ENMIN $VASP_PP_PATH/potpaw_PBE/O_sv/POTCAR

Open the python script (dft-scripts/script-65.py).

ENMAX

= 1421.493; ENMIN

= 1066.119 eV

In the following calculation, we let VASP select an appropriate ENCUT value.

1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

atoms = Atoms([Atom(O, [4, 4.5, 5], magmom=2)],

cell=(8, 9, 10))

6
7

with jasp(molecules/O-sp-triplet-lowsym-sv,

69

xc=PBE,
ismear=0,
ispin=2,
sigma=0.01,
setups={O: _sv},
atoms=atoms) as calc:

8
9
10
11
12
13
14

try:

15

E_O = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O = None

16
17
18
19

print Magnetic moment on O = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

20
21
22
23
24

# now relaxed O2 dimer

atoms = Atoms([Atom(O, [5,
5, 5], magmom=1),
Atom(O, [6.22, 5, 5], magmom=1)],
cell=(10, 10, 10))

25
26
27
28
29
30
31
32
33
34
35
36
37
38

with jasp(molecules/O2-sp-triplet-sv,
xc=PBE,
ismear=0,
sigma=0.01,
ispin=2,
# turn spin-polarization on
ibrion=2, # make sure we relax the geometry
nsw=10,
setups={O: _sv},
atoms=atoms) as calc:
try:
E_O2 = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
E_O2 = None

39
40
41

# verify magnetic moment

print Magnetic moment on O2 = {0} Bohr magnetons.format(atoms.get_magnetic_moment())

42
43
44

if None not in (E_O, E_O2):

print O2 -> 2O D = {0:1.3f} eV.format(2*E_O - E_O2)

Open the python script (dft-scripts/script-66.py).

Magnetic moment on O = 2.0000001 Bohr magnetons
Magnetic moment on O2 = 1.9999996 Bohr magnetons
O2 -> 2O D = 6.299 eV
This result is close to other reported values. It is possibly not converged, since we let VASP choose
the ENCUT value, and that value is the ENMIN value in the POTCAR. Nevertheless, the point is that
a harder potential does not fix the problem of overbinding in the O2 molecule. That is a fundamental
flaw in the GGA exchange-correlation functional.
3.8.2

Water gas shift example

We consider calculating the reaction energy of the water-gas shift reaction in this example.
CO + H2 O  CO2 + H2
We define the reaction energy as the difference in energy between the products and reactants.
E = ECO2 + EH2 ECO EH2 O
For now, we compute this energy simply as the difference in DFT energies. In the next section we
will add zero-point energies and compute the energy difference as a function of temperature. For now,
we simply need to compute the total energy of each molecule in its equilibrium geometry.
1
2

from ase.structure import molecule

from jasp import *

3
4

# first we define our molecules. These will automatically be at the coordinates from the G2 database.

5
6
7

CO = molecule(CO)
CO.set_cell([8, 8, 8], scale_atoms=False)

8
9

H2O = molecule(H2O)

70

10

H2O.set_cell([8, 8, 8], scale_atoms=False)

11
12
13

CO2 = molecule(CO2)
CO2.set_cell([8, 8, 8], scale_atoms=False)

14
15
16

H2 = molecule(H2)
H2.set_cell([8, 8, 8], scale_atoms=False)

17
18
19
20
21
22
23
24
25
26
27
28
29

# now the calculators to get the energies

with jasp(molecules/wgs/CO,
xc=PBE,
encut=350,
ismear=0,
ibrion=2,
nsw=10,
atoms=CO) as calc:
try:
eCO = CO.get_potential_energy()
except (VaspSubmitted, VaspQueued):
eCO = None

30
31
32
33
34
35
36
37
38
39
40
41

with jasp(molecules/wgs/CO2,
xc=PBE,
encut=350,
ismear=0,
ibrion=2,
nsw=10,
atoms=CO2) as calc:
try:
eCO2 = CO2.get_potential_energy()
except (VaspSubmitted, VaspQueued):
eCO2 = None

42
43
44
45
46
47
48
49
50
51
52
53

with jasp(molecules/wgs/H2,
xc=PBE,
encut=350,
ismear=0,
ibrion=2,
nsw=10,
atoms=H2) as calc:
try:
eH2 = H2.get_potential_energy()
except (VaspSubmitted, VaspQueued):
eH2 = None

54
55
56
57
58
59
60
61
62
63
64
65

with jasp(molecules/wgs/H2O,
xc=PBE,
encut=350,
ismear=0,
ibrion=2,
nsw=10,
atoms=H2O) as calc:
try:
eH2O = H2O.get_potential_energy()
except (VaspSubmitted, VaspQueued):
eH2O = None

66
67
68
69
70
71
72
73

if None in (eCO2, eH2, eCO, eH2O):

pass
else:
dE = eCO2 + eH2 - eCO - eH2O
print Delta E = {0:1.3f} eV.format(dE)
print Delta E = {0:1.3f} kcal/mol.format(dE * 23.06035)
print Delta E = {0:1.3f} kJ/mol.format(dE * 96.485)

Open the python script (dft-scripts/script-67.py).

Delta E = -0.720 eV
Delta E = -16.614 kcal/mol
Delta E = -69.514 kJ/mol
We estimated the enthalpy of this reaction at standard conditions to be -41 kJ/mol using data from
the NIST webbook, which is a fair bit lower than we calculated here. In the next section we will examine
whether additional corrections are needed, such as zero-point and temperature corrections.
71

It is a good idea to verify your calculations and structures are what you expected. Let us print them
here. Inspection of these results shows the geometries were all relaxed, i.e., the forces on each atom are
less than 0.05 eV/.
1

from jasp import *

2
3
4
5
6
7
8

print **** Calculation summaries

print ***** CO
with jasp(molecules/wgs/CO) as calc:
print #+begin_example
print calc
print #+end_example

9
10
11
12
13
14

print ***** CO2

with jasp(molecules/wgs/CO2) as calc:
print #+begin_example
print calc
print #+end_example

15
16
17
18
19
20

print ***** H2
with jasp(molecules/wgs/H2) as calc:
print #+begin_example
print calc
print #+end_example

21
22
23
24
25
26

print ***** H2O

with jasp(molecules/wgs/H2O) as calc:
print #+begin_example
print calc
print #+end_example

Open the python script (dft-scripts/script-68.py).

Calculation summaries
CO
: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/molecules/wgs/CO
converged: True
Energy = -14.789536 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 8.000 0.000 0.000] 8.000
a1 [ 0.000 8.000 0.000] 8.000
a2 [ 0.000 0.000 8.000] 8.000
a,b,c,alpha,beta,gamma (deg):8.000 8.000 8.000 90.0 90.0 90.0
Unit cell volume = 512.000 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.005 0.005 0.005-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
O
[0.000
0.000
0.490] 0
0.01
T T T
1
C
[0.000
0.000
7.346] 0
0.01
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 9
ismear: 0
nsw: 10
72

ibrion:
encut:
magmom:
prec:
kpts:
reciprocal:
xc:
txt:
gamma:

2
350.0
None
Normal
[1, 1, 1]
False
PBE
False

Pseudopotentials used:
---------------------C: potpaw_PBE/C/POTCAR (git-hash: 2272d6745da89a3d872983542cef1d18750fc952)
O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
CO2
: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/molecules/wgs/CO2
converged: True
Energy = -22.959572 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 8.000 0.000 0.000] 8.000
a1 [ 0.000 8.000 0.000] 8.000
a2 [ 0.000 0.000 8.000] 8.000
a,b,c,alpha,beta,gamma (deg):8.000 8.000 8.000 90.0 90.0 90.0
Unit cell volume = 512.000 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.009 0.009 0.008-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
C
[0.000
0.000
0.000] 0
0.00
T T T
1
O
[0.000
0.000
1.177] 0
0.01
T T T
2
O
[0.000
0.000
6.823] 0
0.01
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 12
ismear: 0
nsw: 10
ibrion: 2
encut: 350.0
magmom: None
prec: Normal
kpts: [1, 1, 1]
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
73

---------------------C: potpaw_PBE/C/POTCAR (git-hash: 2272d6745da89a3d872983542cef1d18750fc952)

O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
H2
: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/molecules/wgs/H2
converged: True
Energy = -6.744001 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 8.000 0.000 0.000] 8.000
a1 [ 0.000 8.000 0.000] 8.000
a2 [ 0.000 0.000 8.000] 8.000
a,b,c,alpha,beta,gamma (deg):8.000 8.000 8.000 90.0 90.0 90.0
Unit cell volume = 512.000 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.000 0.000 0.000-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
H
[0.000
0.000
0.376] 0
0.00
T T T
1
H
[0.000
0.000
7.624] 0
0.00
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 5
ismear: 0
nsw: 10
ibrion: 2
encut: 350.0
magmom: None
prec: Normal
kpts: [1, 1, 1]
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------H: potpaw_PBE/H/POTCAR (git-hash: fbc0773b08b32f553234b0b50cc6ad6f5085c816)
H2O
: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/molecules/wgs/H2O
converged: True
Energy = -14.193569 eV
Unit cell vectors (angstroms)
x
y
z
length
74

a0 [ 8.000 0.000 0.000] 8.000

a1 [ 0.000 8.000 0.000] 8.000
a2 [ 0.000 0.000 8.000] 8.000
a,b,c,alpha,beta,gamma (deg):8.000 8.000 8.000 90.0 90.0 90.0
Unit cell volume = 512.000 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.005 0.004 0.005-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
O
[0.000
0.000
0.122] 0
0.06
T T T
1
H
[0.000
0.766
7.522] 0
0.04
T T T
2
H
[0.000
7.234
7.522] 0
0.04
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 8
ismear: 0
nsw: 10
ibrion: 2
encut: 350.0
magmom: None
prec: Normal
kpts: [1, 1, 1]
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------H: potpaw_PBE/H/POTCAR (git-hash: fbc0773b08b32f553234b0b50cc6ad6f5085c816)
O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
3.8.3

Temperature dependent water gas shift equilibrium constant

To correct the reaction energy for temperature effects, we must compute the vibrational frequencies of
each species, and estimate the temperature dependent contributions to vibrational energy and entropy.
We will break these calculations into several pieces. First we do each vibrational calculation. After
those are done, we can get the data and construct the thermochemistry objects we need to estimate the
reaction energy as a function of temperature (at constant pressure).
CO vibrations
1

from jasp import *

2
3
4
5

# get relaxed geometry

with jasp(molecules/wgs/CO) as calc:
CO = calc.get_atoms()

6
7
8
9
10
11
12
13
14
15

# now do the vibrations

with jasp(molecules/wgs/CO-vib,
xc=PBE,
encut=350,
ismear=0,
ibrion=6,
nfree=2,
potim=0.02,
nsw=1,

75

16
17
18
19
20

atoms=CO) as calc:
calc.calculate()
vib_freq = calc.get_vibrational_frequencies()
for i, f in enumerate(vib_freq):
print {0:02d}: {1} cm^(-1).format(i,f)

Open the python script (dft-scripts/script-69.py).

00:
01:
02:
03:
04:
05:

2115.528894 cm^(-1)
60.594878 cm^(-1)
60.594878 cm^(-1)
(0.987178+0j) cm^(-1)
(17.958586+0j) cm^(-1)
(17.958586+0j) cm^(-1)

CO has only one vibrational mode (3N-5 = 6 - 5 = 1). The other 5 modes are 3 translations and 2
rotations.
CO2 vibrations
1

from jasp import *

2
3
4
5

# get relaxed geometry

with jasp(molecules/wgs/CO2) as calc:
CO2 = calc.get_atoms()

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

# now do the vibrations

with jasp(molecules/wgs/CO2-vib,
xc=PBE,
encut=350,
ismear=0,
ibrion=6,
nfree=2,
potim=0.02,
nsw=1,
atoms=CO2) as calc:
calc.calculate()
vib_freq = calc.get_vibrational_frequencies()
for i, f in enumerate(vib_freq):
print {0:02d}: {1} cm^(-1).format(i, f)

Open the python script (dft-scripts/script-70.py).

00:
01:
02:
03:
04:
05:
06:
07:
08:

2352.901285 cm^(-1)
1316.689504 cm^(-1)
635.015913 cm^(-1)
635.015913 cm^(-1)
(0.344306+0j) cm^(-1)
(1.763867+0j) cm^(-1)
(1.763867+0j) cm^(-1)
(62.700411+0j) cm^(-1)
(62.700411+0j) cm^(-1)

CO2 is a linear molecule with 3N-5 = 4 vibrational modes. They are the first four frequencies in the
output above.
H2 vibrations
1

from jasp import *

2
3
4

# get relaxed geometry

with jasp(molecules/wgs/H2) as calc:

76

H2 = calc.get_atoms()

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

# now do the vibrations

with jasp(molecules/wgs/H2-vib,
xc=PBE,
encut=350,
ismear=0,
ibrion=6,
nfree=2,
potim=0.02,
nsw=1,
atoms=H2) as calc:
calc.calculate()
vib_freq = calc.get_vibrational_frequencies()
for i, f in enumerate(vib_freq):
print {0:02d}: {1} cm^(-1).format(i, f)

Open the python script (dft-scripts/script-71.py).

00:
01:
02:
03:
04:
05:

4281.917749 cm^(-1)
129.146855 cm^(-1)
129.146855 cm^(-1)
0.0 cm^(-1)
0.0 cm^(-1)
(1e-05+0j) cm^(-1)

There is only one frequency of importance (the one at 4281 cm1 ) for the linear H2 molecule.
H2 O vibrations
1

from jasp import *

2
3
4
5

# get relaxed geometry

with jasp(molecules/wgs/H2O) as calc:
H2O = calc.get_atoms()

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

# now do the vibrations

with jasp(molecules/wgs/H2O-vib,
xc=PBE,
encut=350,
ismear=0,
ibrion=6,
nfree=2,
potim=0.02,
nsw=1,
atoms=H2O) as calc:
calc.calculate()
vib_freq = calc.get_vibrational_frequencies()
for i, f in enumerate(vib_freq):
print {0:02d}: {1} cm^(-1).format(i, f)

Open the python script (dft-scripts/script-72.py).

00:
01:
02:
03:
04:
05:
06:
07:
08:

3782.062213 cm^(-1)
3672.1246 cm^(-1)
1586.23055 cm^(-1)
135.82763 cm^(-1)
16.280411 cm^(-1)
(0.208582+0j) cm^(-1)
(26.297061+0j) cm^(-1)
(106.869518+0j) cm^(-1)
(131.286732+0j) cm^(-1)

Water has 3N-6 = 3 vibrational modes.

77

Thermochemistry Now we are ready. We have the electronic energies and vibrational frequencies of
each species in the reaction.
1
2
3
4

from ase.thermochemistry import IdealGasThermo

from jasp import *
import numpy as np
import matplotlib.pyplot as plt

5
6
7
8
9

# first we get the electronic energies

with jasp(molecules/wgs/CO) as calc:
CO = calc.get_atoms()
E_CO = CO.get_potential_energy()

10
11
12
13

with jasp(molecules/wgs/CO2) as calc:

CO2 = calc.get_atoms()
E_CO2 = CO2.get_potential_energy()

14
15
16
17

with jasp(molecules/wgs/H2) as calc:

H2 = calc.get_atoms()
E_H2 = H2.get_potential_energy()

18
19
20
21

with jasp(molecules/wgs/H2O) as calc:

H2O = calc.get_atoms()
E_H2O = H2O.get_potential_energy()

22
23
24
25

# now we get the vibrational energies

h = 4.1356675e-15 # eV * s
c = 3.0e10 # cm / s

26
27
28
29

with jasp(molecules/wgs/CO-vib) as calc:

vib_freq = calc.get_vibrational_frequencies()
CO_vib_energies = [h * c * nu for nu in vib_freq]

30
31
32
33

with jasp(molecules/wgs/CO2-vib) as calc:

vib_freq = calc.get_vibrational_frequencies()
CO2_vib_energies = [h * c * nu for nu in vib_freq]

34
35
36
37

with jasp(molecules/wgs/H2-vib) as calc:

vib_freq = calc.get_vibrational_frequencies()
H2_vib_energies = [h * c * nu for nu in vib_freq]

38
39
40
41

with jasp(molecules/wgs/H2O-vib) as calc:

vib_freq = calc.get_vibrational_frequencies()
H2O_vib_energies = [h * c * nu for nu in vib_freq]

42
43
44
45
46
47

# now we make a thermo object for each molecule

CO_t = IdealGasThermo(vib_energies=CO_vib_energies[0:0],
electronicenergy=E_CO, atoms=CO,
geometry=linear, symmetrynumber=1,
spin=0)

48
49
50
51
52

CO2_t = IdealGasThermo(vib_energies=CO2_vib_energies[0:4],
electronicenergy=E_CO2, atoms=CO2,
geometry=linear, symmetrynumber=2,
spin=0)

53
54
55
56
57

H2_t = IdealGasThermo(vib_energies=H2_vib_energies[0:0],
electronicenergy=E_H2, atoms=H2,
geometry=linear, symmetrynumber=2,
spin=0)

58
59
60
61
62

H2O_t = IdealGasThermo(vib_energies=H2O_vib_energies[0:3],
electronicenergy=E_H2O, atoms=H2O,
geometry=nonlinear, symmetrynumber=2,
spin=0)

63
64
65
66
67
68
69
70

# now we can compute G_rxn for a range of temperatures from 298 to 1000 K
Trange = np.linspace(298, 1000, 20) # K
P = 101325. # Pa
Grxn = np.array([(CO2_t.get_free_energy(temperature=T, pressure=P)
+ H2_t.get_free_energy(temperature=T, pressure=P)
- H2O_t.get_free_energy(temperature=T, pressure=P)
- CO_t.get_free_energy(temperature=T, pressure=P)) * 96.485 for T in Trange])

71
72
73
74

Hrxn = np.array([(CO2_t.get_enthalpy(temperature=T)
+ H2_t.get_enthalpy(temperature=T)
- H2O_t.get_enthalpy(temperature=T)

78

- CO_t.get_enthalpy(temperature=T)) * 96.485
for T in Trange])

75
76
77
78
79
80
81
82
83

plt.plot(Trange, Grxn, bo-, label=$\Delta G_{rxn}$)

plt.plot(Trange, Hrxn, ro:, label=$\Delta H_{rxn}$)
plt.xlabel(Temperature (K))
plt.ylabel(r$\Delta G_{rxn}$ (kJ/mol))
plt.legend(loc=best)
plt.savefig(images/wgs-dG-T.png)

84
85
86

plt.figure()
R = 8.314e-3

# gas constant in kJ/mol/K

87
88
89
90
91
92
93
94

Keq = np.exp(-Grxn/R/Trange)
plt.plot(Trange, Keq)
plt.ylim([0, 100])
plt.xlabel(Temperature (K))
plt.ylabel($K_{eq}$)
plt.savefig(images/wgs-Keq.png)
plt.show()

Open the python script (dft-scripts/script-73.py).

Figure 22: Thermodynamic energies of the water gas shift reaction as a function of temperature.
You can see a few things here. One is that at near 298K, the Gibbs free energy is about -75 kJ/mol.
This is too negative compared to the experimental standard free energy, which we estimated to be about
-29 kJ/mol from the NIST webbook. There could be several reasons for this disagreement, but the most
likely one is errors in the exchange-correlation functional. The error in energy has a significant effect on
the calculated equilibrium constant, significantly overestimating it.

79

Figure 23: Temperature dependence of the equilibrium constant.

3.9

Molecular reaction barriers

We will consider a simple example of the barrier for NH3 inversion. We have to create an NH3 molecule in
the initial and inverted state (these have exactly the same energy), and then interpolate a band of images.
Then, we use the NEB method 48 to compute the barrier to inversion. The NEB class of methods are
pretty standard, but other algorithms for finding barriers (saddle-points) exist that may be relevant. 49
3.9.1
1
2
3
4
5
6

Get initial and final states

# compute initial and final states

from ase import Atoms
from ase.data.molecules import molecule
import numpy as np
from jasp import *
from ase.constraints import FixAtoms

7
8
9
10

atoms = molecule(NH3)
constraint = FixAtoms(mask=[atom.symbol == N for atom in atoms])
atoms.set_constraint(constraint)

11
12

Npos = atoms.positions[0]

13
14
15
16

# move N to origin
atoms.translate(-Npos)
atoms.set_cell((10, 10, 10), scale_atoms=False)

17
18
19

atoms2 = atoms.copy()
pos2 = atoms2.positions

20
21
22

for i,atom in enumerate(atoms2):

if atom.symbol == H:

80

23
24
25

# reflect through z
pos2[i] *= np.array([1, 1, -1])
atoms2.positions = pos2

26
27
28
29

#now move N to center of box

atoms.translate([5, 5, 5])
atoms2.translate([5, 5, 5])

30
31
32
33
34
35
36
37
38
39
40

with jasp(molecules/nh3-initial,
xc=PBE,
encut=350,
ibrion=1,
nsw=10,
atoms=atoms) as calc:
try:
calc.calculate()
except (VaspSubmitted, VaspQueued):
pass

41
42
43
44
45
46
47
48
49
50
51

with jasp(molecules/nh3-final,
xc=PBE,
encut=350,
ibrion=1,
nsw=10,
atoms=atoms2) as calc:
try:
calc.calculate()
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-74.py).

3.9.2

Run band calculation

Now we do the band calculation.

1
2
3

# Run NH3 NEB calculations

from jasp import *
from ase.neb import NEB

4
5
6

with jasp(molecules/nh3-initial) as calc:

atoms = calc.get_atoms()

7
8
9

with jasp(molecules/nh3-final) as calc:

atoms2 = calc.get_atoms()

10
11
12
13
14

# 5 images including endpoints

images = [atoms]
images += [atoms.copy() for i in range(3)]
images += [atoms2]

15
16
17

neb = NEB(images)
neb.interpolate()

18
19
20
21
22
23
24
25

with jasp(molecules/nh3-neb,
xc=PBE,
ibrion=1,
nsw=90,
spring=-5, debug=logging.DEBUG,
atoms=images) as calc:
images, energies = calc.get_neb()

26
27
28
29
30

calc.plot_neb(show=False)
import matplotlib.pyplot as plt
plt.savefig(images/nh3-neb.png)
plt.show()

Open the python script (dft-scripts/script-75.py).

81

Figure 24: Nudged elastic band results for ammonia flipping.

Bulk systems

See http://arxiv.org/pdf/1204.2733.pdf for a very informative comparison of DFT codes for computing different bulk properties.

4.1
4.1.1

Defining and visualizing bulk systems

Built-in functions in ase

As with molecules, mod:ase provides several helper functions to create bulk structures. We highlight a
few of them here. Particularly common ones are:
mod:ase.lattice.cubic.FaceCenteredCubic
mod:ase.lattice.cubic.BodyCenteredCubic
mod:ase.lattice.hexagonal.Graphite
mod:ase.lattice.compounds.NaCl
For others, see https://wiki.fysik.dtu.dk/ase/ase/lattice.html
We start with a simple example, fcc Ag. By default, mod:ase knows Ag is an fcc metal, and knows
the experimental lattice constant. We have to specify the directions (vectors along each axis) to get
something other than the default output. Here, the default fcc cell contains four atoms.

82

1
2

from ase.io import write

from ase.lattice.cubic import FaceCenteredCubic

3
4

atoms = FaceCenteredCubic(Ag)

5
6

write(images/Ag-fcc.png, atoms, show_unit_cell=2)

7
8

print atoms

Open the python script (dft-scripts/script-76.py).

Lattice(symbols=Ag4, positions=..., cell=[4.09, 4.09, 4.09], pbc=[True, True, True])
Note:
A mod:ase.lattice.bravais.Lattice object is returned! This is practically the same as as
an mod:ase.atoms.Atoms object.
Here we specify the primitive unit cell, which only has one atom in it.
1
2

from ase.io import write

from ase.lattice.cubic import FaceCenteredCubic

3
4
5
6

atoms = FaceCenteredCubic(Ag, directions=[[0, 1, 1],

[1, 0, 1],
[1, 1, 0]])

7
8

write(images/Ag-fcc-primitive.png, atoms, show_unit_cell=2)

9
10

print atoms

Open the python script (dft-scripts/script-77.py).

Lattice(symbols=Ag, positions=..., cell=[[2.892066735052979, 0.0, 0.0], [1.4460333675264898, 2.5046

We can use these modules to build alloy unit cells. The basic strategy is to create the base unit cell
in one element and then selectively change some atoms to different chemical symbols. Here we examine
an Ag3 Pd alloy structure.
1
2

from ase.io import write

from ase.lattice.cubic import FaceCenteredCubic

3
4
5
6
7
8
9

atoms = FaceCenteredCubic(directions=[[1, 0, 0],

[0, 1, 0],
[0, 0, 1]],
size=(1, 1, 1),
symbol=Ag,
latticeconstant=4.0)

10
11

write(images/Ag-bulk.png, atoms, show_unit_cell=2)

12
13
14
15

# to make an alloy, we can replace one atom with another kind

atoms[0].symbol = Pd
write(images/AgPd-bulk.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-78.py).

83

Figure 25: A simple fcc Ag bulk structure in the primitive unit cell.

Figure 27: A simple fcc Ag bulk structure in the traditional unit cell.

84

Figure 26: A simple fcc Ag bulk structure in the primitive unit cell.

Figure 28: A simple Ag3 Pd bulk structure.

To create a graphite structure we use the following code. Note that we have to specify the lattice
constants (taken from http://www.phy.ohiou.edu/~asmith/NewATOMS/HOPG.pdf) because mod:ase has
C in the diamond structure by default. We show two views, because the top view does not show the
spacing between the layers.
1
2

from ase.lattice.hexagonal import Graphite

from ase.io import write

3
4
5
6
7
8

atoms = Graphite(C, latticeconstant={a: 2.4612,

c: 6.7079})
write(images/graphite.png,
atoms.repeat((2, 2, 1)),
rotation=115x, show_unit_cell=2)

9
10
11
12

write(images/graphite-top.png,
atoms.repeat((2, 2, 1)),
show_unit_cell=2)

85

Open the python script (dft-scripts/script-79.py).

Figure 29: A top view of graphite.

To get a compound, we use the following code. We have to specify the basis atoms to the function
generating the compound, and the lattice constant. For NaCl we use the lattice constant at (http:
//en.wikipedia.org/wiki/Sodium_chloride).
1
2

from ase.lattice.compounds import NaCl

from ase.io import write

3
4
5

atoms = NaCl([Na, Cl], latticeconstant=5.65)

write(images/NaCl.png, atoms, show_unit_cell=2, rotation=45x,45y,45z)

Open the python script (dft-scripts/script-80.py).

mod:ase.spacegroup A final alternative to setting up bulk structures is mod:ase.spacegroup. This is
a concise way to setup structures if you know the following properties of the crystal structure:
1. Chemical symbols
2. Coordinates of the non-equivalent sites in the unit cell
3. the spacegroup
4. the cell parameters (a, b, c, alpha, beta, gamma)
1
2
3
4
5
6
7

from ase.lattice.spacegroup import crystal

# FCC aluminum
a = 4.05
al = crystal(Al, [(0, 0, 0)],
spacegroup=225,
cellpar=[a, a, a, 90, 90, 90])
print(al)

86

Open the python script (dft-scripts/script-81.py).

Atoms(symbols=Al4, positions=..., cell=[[4.05, 0.0, 0.0], [2.4799097682733903e-16, 4.05, 0.0], [2.4

Here is rutile TiO2 .
1

from ase.lattice.spacegroup import crystal

2
3
4
5
6
7

a = 4.6
c = 2.95
rutile = crystal([Ti, O], basis=[(0, 0, 0), (0.3, 0.3, 0.0)],
spacegroup=136, cellpar=[a, a, c, 90, 90, 90])
print rutile

Open the python script (dft-scripts/script-82.py).

Atoms(symbols=Ti2O4, positions=..., cell=[[4.6, 0.0, 0.0], [2.816687638038912e-16, 4.6, 0.0], [1.80

4.1.2

Using http://materialsproject.org

The Materials Project offers web access to a pretty large number of materials (over 21,000 at the time
of this writing), including structure and other computed properties. You must sign up for an account at
the website, and then you can access the information. You can search for materials with lots of different
criteria including formula, unit cell formula, by elements, by structure, etc. . . The website allows you to
download the VASP files used to create the calculations. They also develop the pymatgen project (which
requires python 2.7+).
For example, I downloaded this cif file for a RuO2 structure (Material ID 825).
1
2
3
4
5
6
7
8
9
10
11
12

#\#CIF1.1
##########################################################################
#
Crystallographic Information Format file
#
Produced by PyCifRW module
#
# This is a CIF file. CIF has been adopted by the International
# Union of Crystallography as the standard for data archiving and
# transmission.
#
# For information on this file format, follow the CIF links at
# http://www.iucr.org
##########################################################################

13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

data_RuO2
_symmetry_space_group_name_H-M
_cell_length_a
_cell_length_b
_cell_length_c
_cell_angle_alpha
_cell_angle_beta
_cell_angle_gamma
_chemical_name_systematic
_symmetry_Int_Tables_number
_chemical_formula_structural
_chemical_formula_sum
_cell_volume
_cell_formula_units_Z
loop_
_symmetry_equiv_pos_site_id
_symmetry_equiv_pos_as_xyz
1 x, y, z

P 1
3.13970109
4.5436378
4.5436378
90.0
90.0
90.0
Generated by pymatgen
1
RuO2
Ru2 O4
64.8180127062
2

32
33
34
35
36
37
38
39
40
41

loop_
_atom_site_type_symbol
_atom_site_label
_atom_site_symmetry_multiplicity
_atom_site_fract_x
_atom_site_fract_y
_atom_site_fract_z
_atom_site_attached_hydrogens
_atom_site_B_iso_or_equiv

87

42
43
44
45
46
47
48

_atom_site_occupancy
O O1 1 0.000000 0.694330 0.694330 0 . 1
O O2 1 0.500000 0.805670 0.194330 0 . 1
O O3 1 0.000000 0.305670 0.305670 0 . 1
O O4 1 0.500000 0.194330 0.805670 0 . 1
Ru Ru5 1 0.500000 0.500000 0.500000 0 . 1
Ru Ru6 1 0.000000 0.000000 0.000000 0 . 1

Open the python script (dft-scripts/script-83.py).

We can read this file in with func:ase.io.read. That function automatically recognizes the file type by
the extension.
1

from ase.io import read, write

2
3

atoms = read(bulk/Ru2O4_1.cif)

4
5

write(images/Ru2O4.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-84.py).

Figure 32: An RuO2 unit cell prepared from a cif file.

You can also download the VASP files. I have copied these files (INCAR, POSCAR, KPOINTS) to
a directory (bulk/Ru2O4), and now we can run a calculation like this:
1

from jasp import *

2
3
4
5
6
7

with jasp(bulk/Ru2O4,
xc=PBE,
setups={Ru: _pv}) as calc:
calc.calculate()
print calc

Open the python script (dft-scripts/script-85.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/Ru2O4
converged: True
Energy = -44.302690 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 3.137 0.000 0.000] 3.137
a1 [ 0.000 4.541 0.000] 4.541
a2 [ 0.000 0.000 4.541] 4.541
a,b,c,alpha,beta,gamma (deg):3.137 4.541 4.541 90.0 90.0 90.0
88

Unit cell volume = 64.694 Ang^3

Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
-0.002 -0.000 -0.000-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Ru [1.569
2.270
2.270] 0
0.00
T T
1
Ru [0.000
0.000
0.000] 0
0.00
T T
2
O
[0.000
3.153
3.153] 0
0.00
T T
3
O
[1.569
3.659
0.882] 0
0.00
T T
4
O
[0.000
1.388
1.388] 0
0.00
T T
5
O
[1.569
0.882
3.659] 0
0.00
T T
--------------------------------------------------

T
T
T
T
T
T

INCAR Parameters:
----------------nbands: 31
ismear: 1
nelmin: 3
icharg: 1
nelm: 100
nsw: 99
ibrion: 2
npar: 1
isif: 3
encut: 520.0
sigma: 0.2
ediff: 0.0003
lwave: True
magmom: [0.6, 0.6, 0.6, 0.6, 0.6, 0.6]
prec: Normal
algo: Fast
lreal: Auto
kpts: [8, 6, 6]
reciprocal: False
setups: {Ru: _pv}
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------Ru: potpaw_PBE/Ru_pv/POTCAR (git-hash: c29610ef9b7bfa353e710b09dfadcd2b0fb0d274)
O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Some notes: we had to specify the PBE functional. That is good habit, since the default may be
something else. We also had to specify the special setups used, since those were not defaults either.
We get exactly the same result as the website! Why is this helpful then? Well, we could use this as
a starting point to look at an equation of state, or to plot some interesting electronic structure, etc. . .
Or we can gain a little confidence that our version of VASP is working like the one they used.

89

4.2
4.2.1

Computational parameters that are important for bulk structures

k-point convergence

In the section on molecules, we learned that the total energy is a function of the planewave cutoff energy
(ENCUT) used. In bulk systems that is true also. There is also another calculation parameter you
must consider, the k-point grid. The k-point grid is a computational tool used to approximate integrals
of some property, e.g. the electron density, over the entire unit cell. The integration is performed
in reciprocal space (i.e. in the Brillouin zone) for convenience and efficiency, and the k-point grid is
where the property is sampled for the integration. The higher the number of sampled points, the more
accurately the integrals are approximated.
We will typically use a Monkhorst-Pack 28 k-point grid, which is essentially a uniformly spaced grid
in the Brillouin zone. Another less commonly used scheme is the Chadi-Cohen k-point grid. 27 The
Monkhorst-Pack grids are specified as n1 n2 n3 grids, and the total number of k-points is n1 n2 n3.
The computational cost is linear in the total number of k-points, so a calculation on a 4 4 4 grid
will be roughly 8 times more expensive than on a 2 2 2 grid. Hence, one seeks again to balance
convergence with computational tractability. Below we consider the k-point convergence of fcc Ag.
1
2

from ase.lattice.cubic import FaceCenteredCubic

from jasp import *

3
4

atoms = FaceCenteredCubic(Ag)

5
6

KPTS = [2, 3, 4, 5, 6, 8, 10]

7
8

TE = []

9
10
11
12
13
14
15
16
17
18
19
20

ready = True
for k in KPTS:
with jasp(bulk/Ag-kpts-{0}.format(k),
xc=PBE,
kpts=(k, k, k), # specifies the Monkhorst-Pack grid
encut=300,
atoms=atoms) as calc:
try:
TE.append(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
ready = False

21
22
23

if not ready:
import sys; sys.exit()

24
25

import matplotlib.pyplot as plt

26
27
28
29

# consider the change in energy from lowest energy state

TE = np.array(TE)
TE -= TE.min()

30
31
32
33
34
35

plt.plot(KPTS, TE)
plt.xlabel(number of k-points in each dimension)
plt.ylabel(Total Energy (eV))
plt.savefig(images/Ag-kpt-convergence.png)
plt.show()

Open the python script (dft-scripts/script-86.py).

Based on this figure, we need at least a 6 6 6 k-point grid to achieve a convergence level of at
least 50 meV. Note: the k-point convergence is not always monotonic like it is in this example, and
sometimes very dense grids (e.g. up to 20 20 20) are needed for highly converged properties such
as the density of states in smaller unit cells. Oscillations in the total energy are typical, and it can be
difficult to get high levels of convergence. The best practices are to use the same k-point sampling grid
in energy differences where possible, and dense (high numbers of k-points) otherwise. It is important to
check for convergence in these cases.
As unit cells get larger, the number of k-points required becomes smaller. For example, if a 1 1 1
fcc unit cell shows converged energies in a 12 12 12 k-point grid, then a 2 2 2 fcc unit cell would

90

show the same level of convergence with a 6 6 6 k-point grid. In other words, doubling the unit cell
vectors results in a halving of the number of k-points.
Sometimes you may see k-points described as k-points per reciprocal atom. For example, a 121212
k-point grid for a primitive fcc unit cell would be 1728 k-points per reciprocal atom. A 2 2 2 fcc unit
cell has eight atoms in it, or 0.125 reciprocal atoms, so a 6 6 6 k-point grid has 216 k-points in it, or
216/0.125 = 1728 k-points per reciprocal atom, the same as we discussed before.
In the k-point convergence example above, we used a 6 6 6 k-point grid on a unit cell with four
atoms in it, leading to 864 k-points per reciprocal atom. If we had instead used the primitive unit cell,
we would need either a 9 9 9 or 10 10 10 k-point grid to get a similar level of accuracy. In this
case, there is no exact matching of k-point grids due to the difference in shape of the cells.
4.2.2

Effect of SIGMA

In the self-consistent cycle of a DFT calculation, the total energy is minimized with respect to occupation
of the Kohn-Sham orbitals. At absolute zero, a band is either occupied or empty. This discrete occupation results in discontinuous changes in energy with changes in occupation, which makes it difficult
to converge. One solution is to artificially broaden the band occupancies, as if they were occupied at
a higher temperature where partial occupation is possible. This results in a continuous dependence of
energy on the partial occupancy, and dramatically increases the rate of convergence. SIGMA and ISMEAR
affect how the partial occupancies of the bands are determined.
Some rules to keep in mind:
1. The smearing methods were designed for metals. For molecules, semiconductors and insulators you
should use a very small SIGMA (e.g. 0.01).
2. Standard values for metallic systems is SIGMA=0.1, but the best SIGMA may be material specific.
The consequence of this finite temperature is that additional bands must be included in the calculation
to allow for the partially occupied states above the Fermi level; the number of extra bands depends on the
temperature used. An example of the maximum occupancies of the bands for an Cu bulk as a function
of SIGMA is shown in Figure 34. Obviously, as SIGMA approaches 0, the occupancy approaches a step
function. It is preferable that the occupancy of several of the highest bands be zero (or at least of order
1 108 ) to ensure enough variational freedom was available in the calculation. Consequently, it is
suggested that fifteen to twenty extra bands be used for a SIGMA of 0.20. In any case, it should be
determined that enough bands were used by examination of the occupancies. It is undesirable to have
too many extra bands, as this will add computational time.
Below we show the effect of SIGMA on the band occupancies.
1
2
3

from jasp import *

from ase import Atom, Atoms
import matplotlib.pyplot as plt

4
5
6
7
8
9

a = 3.61
atoms = Atoms([Atom(Cu, (0, 0, 0))],
cell=0.5 * a * np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]])).repeat((2, 2, 2))

10
11

SIGMA = [0.001, 0.05, 0.1, 0.2, 0.5]

12
13

for sigma in SIGMA:

14
15
16
17
18
19
20
21
22
23

with jasp(bulk/Cu-sigma-{0}.format(sigma),
xc=PBE,
encut=350,
kpts=(4, 4, 4),
ismear=-1,
sigma=sigma,
nbands=9 * 8,
atoms=atoms) as calc:
e = atoms.get_potential_energy()

24

91

nbands = calc.nbands
nkpts = len(calc.get_ibz_k_points())

25
26
27

occ = np.zeros((nkpts, nbands))

for i in range(nkpts):
occ[i,:] = calc.get_occupation_numbers(kpt=i)

28
29
30
31

max_occ = np.max(occ,axis=0) #axis 0 is columns

32
33

plt.plot(range(nbands), max_occ, label=$\sigma = {0}$.format(sigma))

34
35
36
37
38
39
40
41

plt.xlabel(band number)
plt.ylabel(maximum occupancy (electrons))
plt.ylim([-0.1, 2.1])
plt.legend(loc=best)
plt.savefig(images/occ-sigma.png)
plt.show()

Open the python script (dft-scripts/script-87.py).

4.2.3

The number of bands

In the last figure, it is evident that due to the smearing of the electronic states you need to have extra
bands to accommodate the electrons above the Fermi level, and the higher the incar:SIGMA value is,
the more bands you need. You need enough bands so that the highest energy bands are unoccupied, and
VASP will give you a warning that looks like this:
----------------------------------------------------------------------------|
|
| ADVICE TO THIS USER RUNNING VASP/VAMP
(HEAR YOUR MASTERS VOICE ...): |
|
|
|
Your highest band is occupied at some k-points! Unless you are
|
|
performing a calculation for an insulator or semiconductor, without
|
|
unoccupied bands, you have included TOO FEW BANDS!! Please increase
|
|
the parameter NBANDS in file INCAR to ensure that the highest band
|
|
is unoccupied at all k-points. It is always recommended to
|
|
include a few unoccupied bands to accelerate the convergence of
|
|
molecular dynamics runs (even for insulators or semiconductors).
|
|
Because the presence of unoccupied bands improves wavefunction
|
|
prediction, and helps to suppress band-crossings.
|
|
Following all k-points will be listed (with the Fermi weights of
|
|
the highest band given in paranthesis) ... :
|
|
|
|
6
(-0.01472)
|
|
8
(-0.01413)
|
|
13
(-0.01733)
|
|
14
(-0.01838)
|
|
|
|
The total occupancy of band no.
49 is -0.00932 electrons ...
|
|
|
----------------------------------------------------------------------------We tell VASP the number of bands to use with the incar:NBANDS keyword. VASP will set the
NBANDS automatically if you do not provide a value, but this is in general bad practice (even though it
is often done in this book!). There are a few general guidelines for setting NBANDS. First we recognize
that a band can only have two electrons in it (one spin up, and one spin down) in an calculation without
spin-polarization, or one electron per band for a spin-polarized calculation (note that spin-polarization
doubles the number of bands). There absolutely must be enough bands to accommodate all the electrons,
so the minimum number of bands is int(ceil(nelectrons/2)).
92

Here is an example of what this equation does.

1

import numpy as np

2
3
4

print int(np.ceil(50 / 2.))

print int(np.ceil(51 / 2.))

Open the python script (dft-scripts/script-88.py).

25
26
However, due to the smearing, the minimum number of bands is almost never enough, and we always
add more bands. The default behavior in VASP is:
non-spin polarized
spin-polarized

NELECT/2 + NIONS/2
0.6*NELECT + NMAGIONS

These do not always work, especially for small molecular systems where NIONS/2 may be only 1, or
transition metals where it may be necessary to add up to 2*NIONS extra bands.
To figure out how many bands you need, it is necessary to know how many electrons are in your
calculation. The func:jasp.get_valence_electrons provides this for you. Alternatively, you can look in
the Appendix for a table listing the number of valence electrons for each POTCAR file. Armed with this
information you can set NBANDS the way you want.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7

atoms = Atoms([Atom(Cu, [0.000, 0.000, 0.000])],

cell= [[1.818, 0.000, 1.818],
[1.818, 1.818, 0.000],
[0.000, 1.818, 1.818]])

8
9
10
11
12
13
14
15
16
17
18

with jasp(bulk/alloy/cu,
xc=PBE,
encut=350,
kpts=(13, 13, 13),
nbands=9,
ibrion=2,
isif=4,
nsw=10,
atoms=atoms) as calc:
print calc.get_valence_electrons()

Open the python script (dft-scripts/script-89.py).

11.0
For this calculation we need at least 6 bands (11/2=5.5 which is rounded up to 6) and we need to
include some extra bands. The default rule would only add half a band, which is not enough. We add
three additional bands. This system is so small it does not substantially increase the computational cost.
If you are too trifling to do that much work, you can use the func:jasp.set_nbands to automatically
set the number of bands. This function takes an argument N to set the number of bands to N, or an
argument f to set the NBANDS according to the formula nbands = int(nelectrons/2 + len(atoms) f ).
The default value of f is 1.5. If you want the default VASP behavior, set f=0.5. For transition metals,
it may be required that f=2. This function does not consider whether the calculation is spin-polarized
or not. Here is an example of using func:jasp.set_nbands.
1
2

from jasp import *

from ase import Atom, Atoms

93

4
5
6
7

atoms = Atoms([Atom(Cu, [0.000, 0.000, 0.000])],

cell=[[1.818, 0.000, 1.818],
[1.818, 1.818, 0.000],
[0.000, 1.818, 1.818]])

8
9
10
11
12
13
14
15
16
17
18

with jasp(bulk/alloy/cu,
xc=PBE,
encut=350,
kpts=(13, 13, 13),
ibrion=2,
isif=4,
nsw=10,
atoms=atoms) as calc:
calc.set_nbands(9)
print calc

Open the python script (dft-scripts/script-90.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/alloy/cu
converged: True
Energy = -3.723306 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 1.818 0.000 1.818] 2.571
a1 [ 1.818 1.818 0.000] 2.571
a2 [ 0.000 1.818 1.818] 2.571
a,b,c,alpha,beta,gamma (deg):2.571 2.571 2.571 60.0 60.0 60.0
Unit cell volume = 12.017 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.006 0.006 0.006-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Cu [0.000
0.000
0.000] 0
0.00
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 9
nsw: 10
ibrion: 2
isif: 4
encut: 350.0
magmom: None
prec: Normal
kpts: (13, 13, 13)
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------Cu: potpaw_PBE/Cu/POTCAR (git-hash: a44c591415026f53deb16a99ca3f06b1e69be10b)
1
2

from jasp import *

from ase import Atom, Atoms

94

4
5
6
7

atoms = Atoms([Atom(Cu, [0.000, 0.000, 0.000])],

cell=[[1.818, 0.000, 1.818],
[1.818, 1.818, 0.000],
[0.000, 1.818, 1.818]])

8
9
10
11
12
13
14
15
16
17
18

with jasp(bulk/alloy/cu-setnbands,
xc=PBE,
encut=350,
kpts=(13, 13, 13),
ibrion=2,
isif=4,
nsw=10,
atoms=atoms) as calc:
calc.set_nbands(f=3)
print calc

Open the python script (dft-scripts/script-91.py).

False
: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/alloy/cu-setnbands
converged: None
Energy = nan eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 1.818 0.000 1.818] 2.571
a1 [ 1.818 1.818 0.000] 2.571
a2 [ 0.000 1.818 1.818] 2.571
a,b,c,alpha,beta,gamma (deg):2.571 2.571 2.571 60.0 60.0 60.0
Unit cell volume = 12.017 Ang^3
Stress was not computed
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Cu [0.000
0.000
0.000] 0
nan
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 9
nsw: 10
ibrion: 2
isif: 4
encut: 350
magmom: None
prec: Normal
kpts: (13, 13, 13)
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------None: None (git-hash: None)
You are, of course, free to use any formula you want to set the number of bands. Some formulas I
have used in the past include:

95

1. NBANDS = 0.65*NELECT + 10
2. NBANDS = 0.5*NELECT + 15
3. etc. . .

4.3

Determining bulk structures

What we typically mean by determining bulk structures includes the following:

What is the most stable crystal structure for a material?
What is the lattice constant of fcc Cu?
What are the lattice parameters and internal atom parameters for TiO2 ?
All of these questions can often be addressed by finding the volume, shape and atomic positions
that minimize the total energy of a bulk system. This is true at 0K. At higher temperatures, one must
consider minimizing the free energy, rather than the internal energy.
4.3.1

fcc/bcc crystal structures

The fcc and bcc structures are simple. They only have one degree of freedom: the lattice constant. In
this section we show how to calculate the equilibrium volume of each structure, and determine which
one is more stable. We start with the fcc crystal structure of Cu. We will manually define the crystal
structure based on the definitions in Kittel 4 (Chapter 1).
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7
8
9
10
11
12

# fcc
LC = [3.5, 3.55, 3.6, 3.65, 3.7, 3.75]
fcc_energies = []
ready = True
for a in LC:
atoms = Atoms([Atom(Cu, (0, 0, 0))],
cell=0.5 * a * np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))

13
14
15
16
17
18

with jasp(bulk/Cu-{0}.format(a),
xc=PBE,
encut=350,
kpts=(8, 8, 8),
atoms=atoms) as calc:

19
20

try:

21

e = atoms.get_potential_energy()
fcc_energies.append(e)
except (VaspSubmitted, VaspQueued):
ready = False

22
23
24
25
26
27

if not ready:
import sys; sys.exit()

28
29
30
31
32
33

import matplotlib.pyplot as plt

plt.plot(LC, fcc_energies)
plt.xlabel(Lattice constant ($\AA$))
plt.ylabel(Total energy (eV))
plt.savefig(images/Cu-fcc.png)

34
35
36
37
38

print #+tblname: cu-fcc-energies

print r| lattice constant ($\AA$) | Total Energy (eV) |
for lc, e in zip(LC, fcc_energies):
print | {0} | {1} |.format(lc, e)

Open the python script (dft-scripts/script-92.py).

96

lattice constant ()
3.5
3.55
3.6
3.65
3.7
3.75

Total Energy (eV)

-3.649238
-3.696204
-3.719946
-3.723951
-3.711284
-3.68426

Exercise 4.2
Use the data in the table above to plot the total energy as a function of the lattice
constant. Fit a cubic polynomial to the data, and find the volume that minimizes the
total energy.
If you want to know the lattice constant that gives the lowest energy, you would fit an equation of
state to the data. Here is an example using mod:ase.utils.eos. See also the appendix equations of state.
1
2
3
4
5
6
7
8
9
10

from jasp import *

from ase.utils.eos import EquationOfState
LC = [3.5, 3.55, 3.6, 3.65, 3.7, 3.75]
energies = []
volumes = []
for a in LC:
with jasp(bulk/Cu-{0}.format(a)) as calc:
atoms = calc.get_atoms()
volumes.append(atoms.get_volume())
energies.append(atoms.get_potential_energy())

11
12
13

eos = EquationOfState(volumes, energies)

v0, e0, B = eos.fit()

14
15
16
17
18

print
v0 = {0} A^3
E0 = {1} eV
B = {2} eV/A^3.format(v0, e0, B)

19
20

eos.plot(images/Cu-fcc-eos.png)

Open the python script (dft-scripts/script-93.py).

v0 = 12.0167118546 A^3
E0 = -3.7246811445 eV
B = 0.859009528171 eV/A^3
3.63585568663
Before we jump into the bcc calculations, let us consider what range of lattice constants we should
choose. The fcc lattice is close-packed, and the volume of the primitive cell is V = 1/4a3 or about
11.8 3 /atom. The volume of the equilibrium bcc primitive cell will probably be similar to that. The
question is: what bcc lattice constant gives that volume? The simplest way to answer this is to compute
the answer. We will make a bcc crystal at the fcc lattice constant, and then compute the scaling factor
needed to make it the right volume.
1
2
3

from ase import Atom, Atoms

import numpy as np
a = 3.61 # lattice constant

4
5
6
7
8

atoms = Atoms([Atom(Cu, [0,0,0])],

cell=0.5 * a*np.array([[ 1.0, 1.0, -1.0],
[-1.0, 1.0, 1.0],
[ 1.0, -1.0, 1.0]]))

9
10

print BCC lattice constant = {0} Ang.format(a * (11.8 / atoms.get_volume())**(1./3.))

97

Open the python script (dft-scripts/script-94.py).

BCC lattice constant = 2.86838428403 Ang
Now we run the equation of state calculations.
1
2

from jasp import *

from ase import Atom, Atoms

3
4

LC = [2.75, 2.8, 2.85, 2.9, 2.95, 3.0]

5
6
7
8
9
10

for a in LC:
atoms = Atoms([Atom(Cu, [0, 0, 0])],
cell=0.5 * a * np.array([[ 1.0, 1.0, -1.0],
[-1.0, 1.0, 1.0],
[ 1.0, -1.0, 1.0]]))

11
12
13
14
15
16
17

with jasp(bulk/Cu-bcc-{0}.format(a),
xc=PBE,
encut=350,
kpts=(8, 8, 8),
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-95.py).

Finally, we will compare the two crystal structures.
1

from jasp import *

2
3
4
5
6
7
8
9
10
11

# bcc energies and volumes

bcc_LC = [2.75, 2.8, 2.85, 2.9, 2.95, 3.0]
bcc_volumes = []
bcc_energies = []
for a in bcc_LC:
with jasp(bulk/Cu-bcc-{0}.format(a)) as calc:
atoms = calc.get_atoms()
bcc_volumes.append(atoms.get_volume())
bcc_energies.append(atoms.get_potential_energy())

12
13
14
15
16
17
18
19
20
21

# fcc energies and volumes

fcc_LC = [3.5, 3.55, 3.6, 3.65, 3.7, 3.75]
fcc_volumes = []
fcc_energies =[]
for a in fcc_LC:
with jasp(bulk/Cu-{0}.format(a)) as calc:
atoms = calc.get_atoms()
fcc_volumes.append(atoms.get_volume())
fcc_energies.append(atoms.get_potential_energy())

22
23
24
25

import matplotlib.pyplot as plt

plt.plot(fcc_volumes, fcc_energies, label=fcc)
plt.plot(bcc_volumes, bcc_energies, label=bcc)

26
27
28
29
30

plt.xlabel(Atomic volume ($\AA^3$/atom))

plt.ylabel(Total energy (eV))
plt.legend()
plt.savefig(images/Cu-bcc-fcc.png)

31
32
33
34
35
36
37
38

# print table of data

print #+tblname: bcc-data
print #+caption: Total energy vs. lattice constant for BCC Cu.
print | Lattice constant (\AA$^3$) | Total energy (eV) |
print |-
for lc, e in zip(bcc_LC, bcc_energies):
print | {0} | {1} |.format(lc, e)

Open the python script (dft-scripts/script-96.py).

Exercise 4.3
Use the data for FCC and BCC Cu to plot the total energy as a function of the lattice
constant.
98

Table 4: Total energy vs. lattice constant for BCC Cu.

Lattice constant (3 ) Total energy (eV)
2.75
-3.587523
2.8
-3.668112
2.85
-3.708699
2.9
-3.716496
2.95
-3.697781
3.0
-3.657913
Note we plot the energy vs. atomic volume. That is because the lattice constants of the two crystal
structures are very different. It also shows that the atomic volumes in the two structures are similar.
What can we say here? The fcc structure has a lower energy than the bcc structure, so we can
conclude the fcc structure is more favorable. In fact, the fcc structure is the experimentally found
structure for Cu. Some caution is in order; if you run these calculations at a 4 4 4 k-point grid, the
bcc structure is more stable because the results are not converged!
Exercise 4.4
Compute the energy vs. volume for fcc and bcc Cu for different k-point grids. Determine
when each result has converged, and which structure is more stable.
What can we say about the relative stability of fcc to hcp? Nothing, until we calculate the hcp
equation of state.
4.3.2

Optimizing the hcp lattice constant

The hcp lattice is more complicated than the fcc/bcc lattices because there are two lattice parameters: a
and c or equivalently: a and c/a. We will start by making a grid of values and find the set of parameters
that minimizes the energy. See Figure 38.
1
2
3

from ase.lattice.hexagonal import HexagonalClosedPacked

from jasp import *
import matplotlib.pyplot as plt

4
5
6
7

atoms = HexagonalClosedPacked(symbol=Ru,
latticeconstant={a: 2.7,
c/a: 1.584})

8
9
10

a_list = [2.5, 2.6, 2.7, 2.8, 2.9]

covera_list = [1.4, 1.5, 1.6, 1.7, 1.8]

11
12
13
14

for a in a_list:
energies = []
for covera in covera_list:

15
16
17
18

atoms = HexagonalClosedPacked(symbol=Ru,
latticeconstant={a: a,
c/a: covera})

19
20

wd = bulk/Ru/{0:1.2f}-{1:1.2f}.format(a, covera)

21
22
23
24
25
26
27
28
29
30
31
32
33

with jasp(wd,
xc=PBE,
# the c-axis is longer than the a-axis, so we use
# fewer kpoints.
kpts=(6, 6, 4),
encut=350,
atoms=atoms) as calc:
try:
energies.append(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass
plt.plot(covera_list, energies, label=ra={0} $\AA$.format(a))

34

99

35
36
37
38
39

plt.xlabel($c/a$)
plt.ylabel(Energy (eV))
plt.legend()
plt.savefig(images/Ru-covera-scan.png)
plt.show()

Open the python script (dft-scripts/script-97.py).

It looks like there is a minimum in the a=2.7 curve, at a c/a ratio of about 1.6. We can look at
the same data in a contour plot which shows more clearly there is minimum in all directions near that
point (Figure 39).
1
2

from jasp import *

import matplotlib.pyplot as plt

3
4
5

x = [2.5, 2.6, 2.7, 2.8, 2.9]

y = [1.4, 1.5, 1.6, 1.7, 1.8]

6
7
8

X,Y = np.meshgrid(x, y)
Z = np.zeros(X.shape)

9
10
11

for i,a in enumerate(x):

for j,covera in enumerate(y):

12

wd = bulk/Ru/{0:1.2f}-{1:1.2f}.format(a,covera)

13
14

with jasp(wd) as calc:

atoms = calc.get_atoms()
try:
Z[i][j] = atoms.get_potential_energy()
except (VaspSubmitted, VaspQueued):
pass

15
16
17
18
19
20
21
22
23

cf = plt.contourf(X, Y, Z, 20,
cmap=plt.cm.jet)

24
25
26

cbar = plt.colorbar(cf)
cbar.ax.set_ylabel(Energy (eV))

27
28
29

plt.xlabel($a$ ($\AA$))
plt.ylabel($c/a$)

30
31
32
33

plt.legend()
plt.savefig(images/ru-contourf.png)
plt.show()

Open the python script (dft-scripts/script-98.py).

4.3.3

Complex structures with internal degrees of freedom

A unit cell has six degrees of freedom: the lengths of each unit cell vector, and the angle between each
vector. There may additionally be internal degrees of freedom for the atoms. It is impractical to try the
approach used for the hcp Ru on anything complicated. Instead, we rely again on algorithms to optimize
the unit cell shape, volume and internal degrees of freedom. It is usually not efficient to make a wild
guess of the geometry and then turn VASP loose on to optimize it. Instead, the following algorithm
works pretty well.
1. Find the volume (at constant shape, with relaxed ions) that minimizes the total energy (ISIF=2).
The goal here is to just get an idea of where the right volume is.
2. Using the results from step 1 as a starting point, perform a set of calculations at constant volume
around the minimum from step 1, but the shape and internal atom positions are allowed to change
(ISIF=4).
3. Finally, do a final calculation near the minimum energy allowing the volume to also change.
(ISIF=3).

100

This multistep process is pretty reasonable to get a converged structure pretty quickly. It is not
foolproof, however, and if you have materials such as graphite it may not work well. The problem with
graphite is that it is a layered compound that is held together by weak van der waal type forces which
are not modeled well by typical GGA functionals. Thus the change in energy due to a volume change is
larger in the plane of the graphite sheet than in the direction normal to the sheet. With a typical GGA,
the sheets may just move apart until they do not interact any more.
We will illustrate the process on a well-behaved system (rutile TiO2 ) which has two lattice parameters and one internal degree of freedom. There are a few subtle points to mention in doing these
calculations. The VASP manual recommends that you set incar:PREC to high, and that ENCUT be
set to 1.3*max(ENMAX) of the pseudopotentials. This is necessary to avoid problems caused by small
basis sets when the volume changes, and Pulay stress. It is important to ensure that the energies are
reasonably converged with respect to k-point grids. Hence, it can be a significant amount of work to do
this right! Let us start with determining the ENCUT value that is appropriate for TiO2 .
1
2

grep ENMAX $VASP_PP_PATH/POTPAW_PBE/Ti/POTCAR

grep ENMAX $VASP_PP_PATH/POTPAW_PBE/O/POTCAR

Open the python script (dft-scripts/script-99.py).

ENMAX
ENMAX

=
=

178.330; ENMIN
400.000; ENMIN

=
=

133.747 eV
300.000 eV

According to the manual, we should use ENCUT = 1.3*400 = 520 eV for good results.
Now we consider the k-point convergence. The lattice vectors of the rutile TiO2 structure are not
all the same length, which means it is not essential that we use the same number of k-points in each
direction. For simplicity, however, we do that here.
1
2
3
4
5

# step 1 frozen atoms and shape at different volumes

from ase import Atom, Atoms
import numpy as np
from jasp import *
import matplotlib.pyplot as plt

6
7
8
9
10
11
12
13
14

create a TiO2 structure from the lattice vectors at

http://cst-www.nrl.navy.mil/lattice/struk/c4.html
This site does not exist anymore.

a = 4.59 # experimental degrees of freedom.

c = 2.96
u = 0.3 # internal degree of freedom!

15
16
17
18
19

#primitive vectors
a1 = a * np.array([1.0, 0.0, 0.0])
a2 = a * np.array([0.0, 1.0, 0.0])
a3 = c * np.array([0.0, 0.0, 1.0])

20
21
22
23
24
25
26
27

atoms = Atoms([Atom(Ti, [0., 0., 0.]),

Atom(Ti, 0.5 * a1 + 0.5 * a2 + 0.5 * a3),
Atom(O, u * a1 + u * a2),
Atom(O, -u * a1 - u * a2),
Atom(O, (0.5 + u) * a1 + (0.5 - u) * a2 + 0.5 * a3),
Atom(O, (0.5 - u) * a1 + (0.5 + u) * a2 + 0.5 * a3)],
cell=[a1, a2, a3])

28
29
30

KPOINTS = [2, 3, 4, 5, 6, 7, 8]
energies = []

31
32
33
34
35
36
37
38
39

ready = True
for k in KPOINTS:
with jasp(bulk/tio2/kpts-{0}.format(k),
encut=520,
kpts=(k, k, k),
xc=PBE,
sigma=0.05,
atoms=atoms) as calc:

101

40

try:

41

energies.append(atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
ready = False

42
43
44
45
46

if not ready:
import sys; sys.exit()

47
48
49
50
51
52

plt.plot(KPOINTS, energies)
plt.xlabel(number of k-points in each vector)
plt.ylabel(Total energy (eV))
plt.savefig(images/tio2-kpt-convergence.png)
plt.show()

Open the python script (dft-scripts/script-100.py).

A k-point grid of 5 5 5 appears suitable for reasonably converged results. Now we proceed with
step 1: Compute the total energy of the unit cell allowing internal degrees of freedom to relax, but
keeping a constant cell shape.
1
2
3
4
5

# step 1 frozen atoms and shape at different volumes

from ase import Atom, Atoms
import numpy as np
from jasp import *
import matplotlib.pyplot as plt

6
7
8
9
10
11
12
13

create a TiO2 structure from the lattice vectors at

http://cst-www.nrl.navy.mil/lattice/struk/c4.html

a = 4.59 # experimental degrees of freedom.

c = 2.96
u = 0.3 # internal degree of freedom!

14
15
16
17
18

#primitive vectors
a1 = a * np.array([1.0, 0.0, 0.0])
a2 = a * np.array([0.0, 1.0, 0.0])
a3 = c * np.array([0.0, 0.0, 1.0])

19
20
21
22
23
24
25
26

atoms = Atoms([Atom(Ti, [0., 0., 0.]),

Atom(Ti, 0.5 * a1 + 0.5 * a2 + 0.5 * a3),
Atom(O, u * a1 + u * a2),
Atom(O, -u * a1 - u * a2),
Atom(O, (0.5 + u) * a1 + (0.5 - u) * a2 + 0.5 * a3),
Atom(O, (0.5 - u) * a1 + (0.5 + u) * a2 + 0.5 * a3)],
cell=[a1, a2, a3])

27
28
29

v0 = atoms.get_volume()
cell0 = atoms.get_cell()

30
31

factors = [0.9, 0.95, 1.0, 1.05, 1.1] #to change volume by

32
33

energies, volumes = [], []

34
35
36
37
38

ready = True
for f in factors:
v1 = f * v0
cell_factor = (v1 / v0)**(1. / 3.)

39
40

atoms.set_cell(cell0 * cell_factor, scale_atoms=True)

41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

with jasp(bulk/tio2/step1-{0:1.2f}.format(f),
encut=520,
kpts=(5,5,5),
isif=2, # relax internal degrees of freedom
ibrion=1,
nsw=50,
xc=PBE,
sigma=0.05,
atoms=atoms) as calc:
try:
energies.append(atoms.get_potential_energy())
volumes.append(atoms.get_volume())
except (VaspSubmitted, VaspQueued):
ready = False

56

102

57
58

if not ready:
import sys; sys.exit()

59
60
61
62
63

plt.plot(volumes, energies)
plt.xlabel(Vol. ($\AA^3)$)
plt.ylabel(Total energy (eV))
plt.savefig(images/tio2-step1.png)

64
65
66
67
68
69
70

print #+tblname: tio2-vol-ene

print #+caption: Total energy of TiO_{2} vs. volume.
print | Volume ($\AA^3$) | Energy (eV) |
print |-
for v, e in zip(volumes, energies):
print | {0} | {1} |.format(v, e)

Open the python script (dft-scripts/script-101.py).

Table 5: Total energy of TiO2 vs. volume.
Volume (3 ) Energy (eV)
56.1254184
-51.820427
59.2434972
-52.462241
62.361576
-52.761097
65.4796548
-52.801329
68.5977336
-52.64715
Now, we know the minimum energy volume is near 64 3. You could at this point fit an equation
of state to find that minimum. However, we now want to use these initial starting points for a second
round of optimization where we allow the unit cell shape to change, at constant volume: ISIF=4.
1

from jasp import *

2
3

factors = [0.9, 0.95, 1.0, 1.05, 1.1]

# to change volume by

4
5
6
7
8
9
10
11
12
13

energies1, volumes1 = [], [] # from step 1

energies, volumes = [], [] # for step 2
ready = True
for f in factors:
with jasp(bulk/tio2/step1-{0:1.2f}.format(f)) as calc:
atoms = calc.get_atoms()
energies1.append(atoms.get_potential_energy())
volumes1.append(atoms.get_volume())
calc.clone(bulk/tio2/step2-{0:1.2f}.format(f))

14
15
16
17
18
19
20
21
22
23

# now set ISIF=4 and run

with jasp(bulk/tio2/step2-{0:1.2f}.format(f),
isif=4) as calc:
atoms = calc.get_atoms()
try:
energies.append(atoms.get_potential_energy())
volumes.append(atoms.get_volume())
except (VaspSubmitted, VaspQueued):
ready = False

24
25
26

if not ready:
import sys; sys.exit()

27
28
29
30
31
32
33
34

import matplotlib.pyplot as plt

plt.plot(volumes1, energies1, volumes, energies)
plt.xlabel(Vol. ($\AA^3)$)
plt.ylabel(Total energy (eV))
plt.legend([step 1, step 2], loc=best)
plt.savefig(images/tio2-step2.png)
plt.show()

Open the python script (dft-scripts/script-102.py).

The take away message here is that the total energy slightly decreases when we allow the unit cell
shape to change, especially for the larger unit cell deformation. This has little effect on the minimum
103

volume, but would have an effect on the bulk modulus, which is related to the curvature of the equation
of state. At this point, you could fit an equation of state to the step 2 data, and estimate the volume at
the minimum volume, and recalculate the total energy at that volume.
An alternative is a final calculation with ISIF=3, which optimizes the unit cell volume, shape and
internal coordinates. It looks like the calculation at bulk/tio2/step2-1.05 is close to the minimum, so we
will use that as a starting point for the final calculation.
1

from jasp import *

2
3
4

with jasp(bulk/tio2/step2-1.05) as calc:

calc.clone(bulk/tio2/step3)

5
6
7
8
9
10

with jasp(bulk/tio2/step3,
isif=3) as calc:
calc.calculate()
atoms = calc.get_atoms()
print calc

11
12
13

from pyspglib import spglib

print \nThe spacegroup is {0}.format(spglib.get_spacegroup(atoms))

Open the python script (dft-scripts/script-103.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/tio2/step3
converged: True
Energy = -52.818610 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 4.661 0.000 0.000] 4.661
a1 [ 0.000 4.661 0.000] 4.661
a2 [ 0.000 0.000 2.970] 2.970
a,b,c,alpha,beta,gamma (deg):4.661 4.661 2.970 90.0 90.0 90.0
Unit cell volume = 64.515 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
-0.000 -0.000 -0.000-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Ti [0.000
0.000
0.000] 0
0.00
T T T
1
Ti [2.331
2.331
1.485] 0
0.00
T T T
2
O
[1.420
1.420
0.000] 0
0.00
T T T
3
O
[3.241
3.241
0.000] 0
0.00
T T T
4
O
[3.751
0.910
1.485] 0
0.00
T T T
5
O
[0.910
3.751
1.485] 0
0.00
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 20
nsw: 50
ibrion: 1
isif: 3
encut: 520.0
sigma: 0.05
magmom: None
prec: Normal
kpts: [5, 5, 5]
104

reciprocal:
xc:
txt:
gamma:

False
PBE
False

Pseudopotentials used:
---------------------O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Ti: potpaw_PBE/Ti/POTCAR (git-hash: a29a4f0f14083cf90544a3cf1ae7fc220b31829c)
The spacegroup is P4_2/mnm

(136)

This is the final result. You can see that the forces on all the atoms are less than 0.01 eV/, and
the stress is also very small. The final volume is close to where we expect it to be based on steps 1 and
2. The space group is still correct. The lattice vectors are close in length to the experimentally known
values, and the angles between the vectors has not changed much. Looks good!
As a final note, the VASP manual recommends you do not use the final energy directly from the
calculation, but rather run a final calculation with incar:ISMEAR set to -5. Here we examine the effect.
1

from jasp import *

2
3
4
5
6

with jasp(bulk/tio2/step3) as calc:

atoms = calc.get_atoms()
print default ismear: , atoms.get_potential_energy()
calc.clone(bulk/tio2/step4)

7
8
9
10
11
12

with jasp(bulk/tio2/step4,
ismear=-5,
nsw=0) as calc:
atoms = calc.get_atoms()
print ismear=-5:
, atoms.get_potential_energy()

Open the python script (dft-scripts/script-104.py).

default ismear:
ismear=-5:

-52.81861
-52.817345

The difference here is on the order of a meV. That does not seem significant here. I suspect the
recommended practice stems from early days when much smaller ENCUT values were used and changes
in the number of basis functions were more significant.
4.3.4

Effect of XC on bulk properties

The exchange correlation functional can significantly affect computed bulk properties. Here, we examine
the effect on the bulk lattice constant of Pd (exp. 3.881). An excellent review of this can be found in. 50
We examine several functionals. The xc keyword in jasp is used to select the POTCARs. The value of
xc can be LDA, PW91 or PBE. Let us consider the LDA functional first.
1
2
3

from jasp import *

from ase import Atom, Atoms
from ase.utils.eos import EquationOfState

4
5

LC = [3.75, 3.80, 3.85, 3.90, 3.95, 4.0, 4.05, 4.1]

6
7
8
9
10
11
12
13

volumes, energies = [], []

for a in LC:
atoms = Atoms([Atom(Pd, (0, 0, 0))],
cell=0.5 * a * np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))
with jasp(bulk/Pd-LDA-{0}.format(a),

105

encut=350,
kpts=(12, 12, 12),
xc=LDA,
atoms=atoms):

14
15
16
17
18

try:

19

e = atoms.get_potential_energy()
energies.append(e)
volumes.append(atoms.get_volume())
except (VaspSubmitted, VaspQueued):
pass

20
21
22
23
24
25
26
27
28

if len(energies) == len(LC):
eos = EquationOfState(volumes, energies)
v0, e0, B = eos.fit()
print LDA lattice constant is {0:1.3f} Ang^3.format((4*v0)**(1./3.))

Open the python script (dft-scripts/script-105.py).

LDA lattice constant is 3.855 Ang^3
For a GGA calculation, it is possible to specify which functional you want via the incar:GGA tag.
This tag was designed to use the LDA POTCAR files, but with a GGA functional. We will consider four
different functionals here.
1
2
3

from jasp import *

from ase import Atom, Atoms
from ase.utils.eos import EquationOfState

4
5

LC = [3.75, 3.80, 3.85, 3.90, 3.95, 4.0, 4.05, 4.1]

6
7
8
9
10

GGA = {AM:
PE:
PS:
RP:

AM05,
PBE,
PBEsol,
RPBE}

11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

for key in GGA:

volumes, energies = [], []
for a in LC:
atoms = Atoms([Atom(Pd, (0, 0, 0))],
cell=0.5 * a * np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))
with jasp(bulk/Pd-GGA-{1}-{0}.format(a, key),
encut=350,
kpts=(12, 12, 12),
xc=LDA,
gga=key,
atoms=atoms):
try:
e = atoms.get_potential_energy()
energies.append(e)
volumes.append(atoms.get_volume())
except (VaspSubmitted, VaspQueued):
pass

31
32
33
34
35
36
37
38
39

if len(energies) == len(LC):
eos = EquationOfState(volumes, energies)
v0, e0, B = eos.fit()
print {1:6s} lattice constant is {0:1.3f} Ang^3.format((4*v0)**(1./3.),
GGA[key])
else:
print energies, LC
print {0} is not ready.format(GGA[key])

Open the python script (dft-scripts/script-106.py).

PBEsol
AM05
RPBE
PBE

lattice
lattice
lattice
lattice

constant
constant
constant
constant

is
is
is
is

3.885
3.882
3.990
3.952

Ang^3
Ang^3
Ang^3
Ang^3
106

These results compare very favorably to those in. 50 It is typical that LDA functionals underestimate
the lattice constants, and that GGAs tend to overestimate the lattice constants. PBEsol and AM05 were
designed specifically for solids, and for Pd, these functionals do an exceptional job of reproducing the
lattice constants. RPBE is particularly bad at the lattice constant, but it has been reported to be a
superior functional for reactivity. 25

4.4

Cohesive energy

The cohesive energy is defined as the energy to separate neutral atoms in their ground electronic state
from the solid at 0K at 1 atm. We will compute this for rhodium. Rh is normally an fcc metal, so we
will use that structure and let VASP find the equilibrium volume for us.
1
2
3

from jasp import *

from ase.lattice.cubic import FaceCenteredCubic
from ase import Atoms, Atom

4
5
6
7
8
9
10

# bulk system
atoms = FaceCenteredCubic(directions=[[0, 1, 1],
[1, 0, 1],
[1, 1, 0]],
size=(1, 1, 1),
symbol=Rh)

11
12
13
14
15
16
17
18
19
20

with jasp(bulk/bulk-rh,
xc=PBE,
encut=350,
kpts=(4, 4, 4),
isif=3,
ibrion=2,
nsw=10,
atoms=atoms) as calc:
bulk_energy = atoms.get_potential_energy()

21
22
23
24

# atomic system
atoms = Atoms([Atom(Rh,[5, 5, 5])],
cell=(7, 8, 9))

25
26
27
28
29
30
31

with jasp(bulk/atomic-rh,
xc=PBE,
encut=350,
kpts=(1, 1, 1),
atoms=atoms) as calc:
atomic_energy = atoms.get_potential_energy()

32
33
34

cohesive_energy = atomic_energy - bulk_energy

print The cohesive energy is {0:1.3f} eV.format(cohesive_energy)

Open the python script (dft-scripts/script-107.py).

The cohesive energy is 6.187 eV
According to Kittel, 4 the cohesive energy of Rh is 5.75 eV. There are a few reasons we may have
discrepancy here:
1. The k-point grid used in the bulk state is not very dense. However, you can see below that the
total energy is pretty converged by a 6 6 6 k-point grid.
2. We did not check for convergence with the planewave cutoff.
3. We neglected spin on the atomic state. Rh in the atomic state has this electronic structure: [Kr]
4d8 5s1 and is a doublet.
First we consider the k-point convergence.

107

from jasp import *

2
3
4
5

with jasp(bulk/atomic-rh) as calc:

atoms = calc.get_atoms()
atomic_energy = atoms.get_potential_energy()

6
7
8

with jasp(bulk/bulk-rh) as calc:

atoms = calc.get_atoms()

9
10

kpts = [3, 4, 6, 9, 12, 15, 18]

11
12
13
14
15
16
17
18

for k in kpts:
with jasp(bulk/bulk-rh-kpts-{0}.format(k),
xc=PBE,
encut=350,
kpts=(k, k, k),
atoms=atoms) as calc:
e = atoms.get_potential_energy()

19
20

print ({0:2d}, {0:2d}, {0:2d}): cohesive energy = {1} eV.format(k,e-atomic_energy)

Open the python script (dft-scripts/script-108.py).

( 3, 3, 3): cohesive
( 4, 4, 4): cohesive
( 6, 6, 6): cohesive
( 9, 9, 9): cohesive
(12, 12, 12): cohesive
(15, 15, 15): cohesive
(18, 18, 18): cohesive

energy
energy
energy
energy
energy
energy
energy

=
=
=
=
=
=
=

-4.837472
-6.189573
-6.210668
-6.208576
-6.215367
-6.215319
-6.216032

eV
eV
eV
eV
eV
eV
eV

Using only 1 k-point for the bulk energy is a terrible approximation! It takes at least a 6 6 6
grid to get the total energy converged to less than 10 meV. Note we do not need to check the k-point
convergence of the atomic state because it is surrounded by vacuum on all sides, and so there should not
be any dispersion in the bands.
We will examine the magnetic state next.
1
2
3
4
5
6
7
8
9

from jasp import *

from ase.lattice.cubic import FaceCenteredCubic
from ase import Atoms, Atom
# bulk system
atoms = FaceCenteredCubic(directions=[[0, 1, 1],
[1, 0, 1],
[1, 1, 0]],
size=(1, 1, 1),
symbol=Rh)

10
11
12
13
14
15
16
17
18
19

with jasp(bulk/bulk-rh,
xc=PBE,
encut=350,
kpts=(4, 4, 4),
isif=3,
ibrion=2,
nsw=10,
atoms=atoms) as calc:
bulk_energy = atoms.get_potential_energy()

20
21
22
23

# atomic system
atoms = Atoms([Atom(Rh,[5, 5, 5], magmom=1)],
cell=(7, 8, 9))

24
25
26
27
28
29
30
31

with jasp(bulk/atomic-rh-sp,
xc=PBE,
encut=350,
kpts=(1, 1, 1),
ispin=2,
atoms=atoms) as calc:
atomic_energy = atoms.get_potential_energy()

32

108

33
34

cohesive_energy = atomic_energy - bulk_energy

print The cohesive energy is {0:1.3f} eV.format(cohesive_energy)

Open the python script (dft-scripts/script-109.py).

The cohesive energy is 5.887 eV
Again, the value in Kittel 4 is 5.75 eV which is very close to this value. Finally, it is also possible
there is a lower energy non-spherical atom energy; we did not check that at all (see Estimating triplet
oxygen dissociation energy with low symmetry).

TODO Elastic properties

4.5

See this reference. 51

We seek the elastic constant tensor that relates stress () and strain () via = c. The stress and
strain are six component vectors, so c will be a 6 6 symmetric matrix.
4.5.1
1
2

Al elastic properties

from jasp import *

from ase.lattice.cubic import FaceCenteredCubic

3
4

atoms = FaceCenteredCubic(symbol=Al)

5
6
7
8
9
10
11
12
13
14
15
16

with jasp(bulk/Al-bulk,
xc=PBE,
kpts=(12, 12, 12),
encut=350,
prec=High,
isif=3,
nsw=30,
ibrion=1,
atoms=atoms) as calc:
print atoms.get_potential_energy()
print atoms.get_stress()

Open the python script (dft-scripts/script-110.py).

-14.96989
[-0.127 -0.127 -0.127

0.

0.

0.

Ok, now with a relaxed geometry at hand, we proceed with the elastic constants. This is accomplished
with IBRION=6 and ISIF 3 in VASP. See this reference (from the VASP page).
1. Le Page and P. Saxe, Phys. Rev. B 65, 104104 (2002).
1
2

from jasp import *

from jasp.elastic_moduli import *

3
4
5

with jasp(bulk/Al-bulk) as calc:

calc.clone(bulk/Al-elastic)

6
7
8
9
10
11
12
13
14
15
16

with jasp(bulk/Al-elastic,
ibrion=6,
#
isif=3,
# gets elastic constants
potim=0.015, # displacements
nsw=1,
nfree=2) as calc:
atoms = calc.get_atoms()
print atoms.get_potential_energy()
EM = calc.get_elastic_moduli()
print EM

17
18
19
20
21

c11 = EM[0,0]
c12 = EM[0,1]
B = (c11 + 2 * c12) / 3.0
print B

109

Open the python script (dft-scripts/script-111.py).

-14.969509
[[ 98.944 66.153
[ 66.153 98.944
[ 66.153 66.153
[ 0.
0.
[ 0.
0.
[ 0.
0.
77.08309

66.153
66.153
98.944
0.
0.
0.

0.
0.
0.
1.687
0.
0.

0.
0.
0.
0.
1.687
0.

0.
]
0.
]
0.
]
0.
]
0.
]
1.687]]

This example shows the basic mechanics of getting the elastic constants. The C44 constant above
is too low, and probably we need to check these constants for convergence with respect to kpoints,
planewave cutoff, and maybe the value of POTIM.
4.5.2
1
2

Fe elastic properties

from jasp import *

from ase.lattice.cubic import BodyCenteredCubic

3
4
5
6

atoms = BodyCenteredCubic(symbol=Fe)
for atom in atoms:
atom.magmom = 3.0

7
8
9
10
11
12
13
14
15
16
17
18

with jasp(bulk/Fe-bulk,
xc=PBE,
kpts=(6, 6, 6),
encut=350,
ispin=2,
isif=3,
nsw=30,
ibrion=1,
atoms=atoms) as calc:
print atoms.get_potential_energy()
print atoms.get_stress()

Open the python script (dft-scripts/script-112.py).

-16.604623
[-0.001 -0.001 -0.001 -0.

-0.

-0.

Ok, now with a relaxed geometry at hand, we proceed with the elastic constants. This is accomplished
with IBRION=6 and ISIF 3 in VASP. See this reference (from the VASP page).
1. Le Page and P. Saxe, Phys. Rev. B 65, 104104 (2002).
1

from jasp import *

2
3
4

with jasp(bulk/Fe-bulk) as calc:

calc.clone(bulk/Fe-elastic)

5
6
7
8
9
10
11
12
13

with jasp(bulk/Fe-elastic,
ibrion=6,
#
isif=3,
# gets elastic constants
potim=0.05, # displacements
nsw=1,
nfree=2) as calc:
atoms = calc.get_atoms()
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-113.py).

-16.592189
110

Now, the results are written out to the OUTCAR file. Actually, three sets of moduli are written out
1) the elastic tensor for rigid ions, 2) the contribution from allowing the atoms to relax, and 3) the total
elastic modulus, all in kBar.
SYMMETRIZED ELASTIC MODULI (kBar)
Direction
XX
YY
ZZ
XY
YZ
ZX
-------------------------------------------------------------------------------XX
2803.5081
1622.6085
1622.6085
0.0000
0.0000
0.0000
YY
1622.6085
2803.5081
1622.6085
0.0000
0.0000
0.0000
ZZ
1622.6085
1622.6085
2803.5081
0.0000
0.0000
0.0000
XY
0.0000
0.0000
0.0000
866.8792
0.0000
0.0000
YZ
0.0000
0.0000
0.0000
0.0000
866.8792
0.0000
ZX
0.0000
0.0000
0.0000
0.0000
0.0000
866.8792
-------------------------------------------------------------------------------and
ELASTIC MODULI CONTR FROM IONIC RELAXATION (kBar)
Direction
XX
YY
ZZ
XY
YZ
ZX
-------------------------------------------------------------------------------XX
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
YY
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
ZZ
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
XY
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
YZ
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
ZX
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
--------------------------------------------------------------------------------

TOTAL ELASTIC MODULI (kBar)

Direction
XX
YY
ZZ
XY
YZ
ZX
-------------------------------------------------------------------------------XX
2803.5081
1622.6085
1622.6085
0.0000
0.0000
0.0000
YY
1622.6085
2803.5081
1622.6085
0.0000
0.0000
0.0000
ZZ
1622.6085
1622.6085
2803.5081
0.0000
0.0000
0.0000
XY
0.0000
0.0000
0.0000
866.8792
0.0000
0.0000
YZ
0.0000
0.0000
0.0000
0.0000
866.8792
0.0000
ZX
0.0000
0.0000
0.0000
0.0000
0.0000
866.8792
-------------------------------------------------------------------------------Let us write a small code here to extract the Total elastic moduli from the OUTCAR file. First we
get the line where the total elastic moduli start, then take the six lines that start three lines after that.
Finally we parse out the matrix elements and cast them as floats.
1
2

import numpy as np
EM = []

3
4
5
6
7
8
9
10

with open(bulk/Fe-elastic/OUTCAR) as f:
lines = f.readlines()
for i, line in enumerate(lines):
if line.startswith( TOTAL ELASTIC MODULI (kBar)):
j = i + 3
data = lines[j:j+6]
break

11
12
13

for line in data:

EM += [[float(x) for x in line.split()[1:]]]

14
15

print np.array(EM)

111

Open the python script (dft-scripts/script-114.py).

[[ 2803.5081
[ 1622.6085
[ 1622.6085
[
0.
[
0.
[
0.

1622.6085
2803.5081
1622.6085
0.
0.
0.

1622.6085
1622.6085
2803.5081
0.
0.
0.

0.
0.
0.
866.8792
0.
0.

0.
0.
0.
0.
866.8792
0.

0.
]
0.
]
0.
]
0.
]
0.
]
866.8792]]

TODO: compare to a normal EOS, and maybe literature. Further discussion of symmetry.
Fe is in a BCC crystal structure, which is high in symmetry. Consequently, many of the elements in
the matrix are equal to zero.
see http://www.nist.gov/data/PDFfiles/jpcrd34.pdf for a lot of detail

4.6

Bulk thermodynamics

We can predict temperature dependent thermodynamic properties of bulk materials without too much
effort. As with the thermochemical properties of ideal gases, we must use some simple models that we
parameterize by DFT. Here we follow the example in Reference 52 for computing the thermal coefficient
of expansion, heat capacity, enthalpy and entropy for Ni as a function of temperature.
We start by computing the equation of state for fcc Ni.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7
8
9
10
11

# fcc
LC = [3.5, 3.55, 3.6, 3.65, 3.7, 3.75]
volumes, energies = [], []
for a in LC:
atoms = Atoms([Atom(Ni, (0, 0, 0), magmom=2.5)],
cell=0.5 * a * np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))

12
13
14
15
16
17
18
19
20
21
22
23
24

with jasp(bulk/Ni-{0}.format(a),
xc=PBE,
encut=350,
kpts=(12,12,12),
ispin=2,
atoms=atoms) as calc:
try:
e = atoms.get_potential_energy()
energies.append(e)
volumes.append(atoms.get_volume())
except:
pass

25
26
27

if len(energies) != len(LC):
import sys; sys.exit()

28
29
30
31
32
33

import matplotlib.pyplot as plt

plt.plot(LC, fcc_energies)
plt.xlabel(Lattice constant ($\AA$))
plt.ylabel(Total energy (eV))
plt.savefig(images/Ni-fcc.png)

Open the python script (dft-scripts/script-115.py).

112

4.7

Effect of pressure on phase stability

So far we have only considered relative stability at a pressure of 0 Pa. We now consider the relative
stability of two phases under pressure. We
will consider TiO2 in the rutile and anatase phases.
E
The pressure is defined by: P = V
. So if we have an equation of state E(V ) we can calculate
T
the pressure at any volume, or alternatively, given a pressure, compute the volume. Pressure can affect
the energy of two phases differently, so that one may become stable under pressure. The condition where
a phase transition occurs is when the pressure in the two phases is the same, which occurs at a common
tangent.
To show this, we need Erutile (V ) and Eanatase (V ).
1
2
3

# run the rutile calculations

from jasp import *
from ase import Atom, Atoms

4
5
6
7
8

B = Ti; X = O; a = 4.59; c = 2.958; u = 0.305;

create a rutile structure from the lattice vectors at

http://cst-www.nrl.navy.mil/lattice/struk/c4.html

9
10
11
12
13
14

spacegroup: 136 P4_2/mnm

a1 = a * np.array([1.0, 0.0, 0.0])

a2 = a * np.array([0.0, 1.0, 0.0])
a3 = c * np.array([0.0, 0.0, 1.0])

15
16
17
18
19
20
21
22

atoms = Atoms([Atom(B, [0., 0., 0.]),

Atom(B, 0.5*a1 + 0.5*a2 + 0.5*a3),
Atom(X, u*a1 + u*a2),
Atom(X, -u*a1 - u*a2),
Atom(X, (0.5+u)*a1 + (0.5-u)*a2 + 0.5*a3),
Atom(X, (0.5-u)*a1 + (0.5+u)*a2 + 0.5*a3)],
cell=[a1, a2, a3])

23
24
25
26

nTiO2 = len(atoms) / 3.
v0 = atoms.get_volume()
cell0 = atoms.get_cell()

27
28

volumes = [28., 30., 32., 34., 36.]

#vol of one TiO2

113

29
30
31
32

for v in volumes:
atoms.set_cell(cell0 * ((nTiO2 * v / v0)**(1. / 3.)),
scale_atoms=True)

33
34
35
36
37
38
39
40
41
42
43
44
45
46
47

with jasp(bulk/TiO2/rutile/rutile-{0}.format(v),
encut=350,
kpts=(6, 6, 6),
xc=PBE,
ismear=0,
sigma=0.001,
isif=2,
ibrion=2,
nsw=20,
atoms=atoms) as calc:
try:
calc.calculate()
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-116.py).

1
2
3
4

# run the anatase calculations

from jasp import *
from ase import Atom, Atoms
# http://cst-www.nrl.navy.mil/lattice/struk/c5.html

5
6

B = Ti; X = O; a = 3.7842; c = 2*4.7573; z = 0.0831;

7
8
9
10

a1 = a * np.array([1.0, 0.0, 0.0])

a2 = a * np.array([0.0, 1.0, 0.0])
a3 = np.array([0.5 * a, 0.5 * a, 0.5 * c])

11
12
13
14
15
16
17
18

atoms = Atoms([Atom(B, -0.125 * a1 + 0.625 * a2 + 0.25 * a3),

Atom(B, 0.125 * a1 + 0.375 * a2 + 0.75 * a3),
Atom(X, -z*a1 + (0.25-z)*a2 + 2.*z*a3),
Atom(X, -(0.25+z)*a1 + (0.5-z)*a2 + (0.5+2*z)*a3),
Atom(X, z*a1 - (0.25 - z)*a2 + (1-2*z)*a3),
Atom(X, (0.25 + z)*a1 + (0.5 + z)*a2 + (0.5-2*z)*a3)],
cell=[a1,a2,a3])

19
20
21
22

nTiO2 = len(atoms) / 3.
v0 = atoms.get_volume()
cell0 = atoms.get_cell()

23
24

volumes = [30., 33., 35., 37., 39.]

#vol of one TiO2

25
26
27
28

for v in volumes:
atoms.set_cell(cell0 * ((nTiO2*v/v0)**(1./3.)),
scale_atoms=True)

29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

with jasp(bulk/TiO2/anatase/anatase-{0}.format(v),
encut=350,
kpts=(6, 6, 6),
xc=PBE,
ismear=0,
sigma=0.001,
isif=2,
ibrion=2,
nsw=20,
atoms=atoms) as calc:
try:
calc.calculate()
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-117.py).

Now we will fit cubic polynomials to the data.
1
2
3
4

# fit cubic polynomials to E(V) for rutile and anatase

from jasp import *
import matplotlib.pyplot as plt
import numpy as np

114

np.set_printoptions(precision=2)

6
7
8
9
10
11
12
13
14
15

# anatase equation of state

volumes = [30., 33., 35., 37., 39.] # vol of one TiO2 formula unit
a_volumes, a_energies = [], []
for v in volumes:
with jasp(bulk/TiO2/anatase/anatase-{0}.format(v)) as calc:
atoms = calc.get_atoms()
nTiO2 = len(atoms) / 3.0
a_volumes.append(atoms.get_volume() / nTiO2)
a_energies.append(atoms.get_potential_energy() / nTiO2)

16
17
18
19
20
21
22
23
24
25

# rutile equation of state

volumes = [28., 30., 32., 34., 36.] # vol of one TiO2
r_volumes, r_energies = [], []
for v in volumes:
with jasp(bulk/TiO2/rutile/rutile-{0}.format(v)) as calc:
atoms = calc.get_atoms()
nTiO2 = len(atoms) / 3.0
r_volumes.append(atoms.get_volume() / nTiO2)
r_energies.append(atoms.get_potential_energy() / nTiO2)

26
27
28
29

# cubic polynomial fit to equation of state E(V) = pars*[V^3 V^2 V^1 V^0]
apars = np.polyfit(a_volumes, a_energies, 3)
rpars = np.polyfit(r_volumes, r_energies, 3)

30
31
32
33
34
35
36
37

print E_anatase(V) = {0:1.2f}*V^3 + {1:1.2f}*V^2 + {2:1.2f}*V + {3:1.2f}.format(*apars)

print E_rutile(V) = {0:1.2f}*V^3 + {1:1.2f}*V^2 + {2:1.2f}*V + {3:1.2f}.format(*rpars)
print anatase epars: {0!r}.format(apars)
print rutile epars: {0!r}.format(rpars)
# get pressure parameters P(V) = -dE/dV
dapars = -np.polyder(apars)
drpars = -np.polyder(rpars)

38
39
40

print anatase ppars: {0!r}.format(dapars)

print rutile ppars: {0!r}.format(drpars)

41
42
43
44

print
print P_anatase(V) = {0:1.2f}*V^2 + {1:1.2f}*V + {2:1.2f}.format(*dapars)
print P_rutile(V) = {0:1.2f}*V^2 + {1:1.2f}*V + {2:1.2f}.format(*drpars)

45
46

vfit = np.linspace(28, 40)

47
48
49
50

# plot the equations of state

plt.plot(a_volumes, a_energies,bo , label=Anatase)
plt.plot(vfit, np.polyval(apars, vfit), b-)

51
52
53

plt.plot(r_volumes, r_energies,gs , label=Rutile)

plt.plot(vfit, np.polyval(rpars, vfit), g-)

54
55
56
57
58
59
60

plt.xlabel(Volume ($\AA^3$/f.u.))
plt.ylabel(Total energy (eV/f.u.))
plt.legend()
plt.xlim([25, 40])
plt.ylim([-27, -26])
plt.savefig(images/rutile-anatase-eos.png)

Open the python script (dft-scripts/script-118.py).

E_anatase(V) = -0.00*V^3 + 0.13*V^2 + -5.23*V + 42.47
E_rutile(V) = -0.00*V^3 + 0.14*V^2 + -5.33*V + 38.59
anatase epars: array([ -0. ,
0.13, -5.23, 42.47])
rutile epars: array([ -0. ,
0.14, -5.33, 38.59])
anatase ppars: array([ 0. , -0.26, 5.23])
rutile ppars: array([ 0. , -0.29, 5.33])
P_anatase(V) = 0.00*V^2 + -0.26*V + 5.23
P_rutile(V) = 0.00*V^2 + -0.29*V + 5.33
To find the conditions where a phase transition occurs, we have to find the common tangent line
between the rutile and anatase phases. In other words we have to solve these two equations:
115

(Eanatase (V 1) Erutile (V 2))/(V 1 V 2) = Panatase (V 1)

(Eanatase (V 1) Erutile (V 2))/(V 1 V 2) = Prutile (V 2)
This is a nonlinear algebra problem. We use the func:scipy.optimize.fsolve to solve this problem.
1
2

from ase.units import GPa

from numpy import array, linspace, polyval

3
4
5
6
7
8

# copied from polynomial fit above

anatase_epars = array([-1.06049246e-03,
4.25202869e+01])
rutile_epars = array([-1.24680208e-03,
3.85903670e+01])

1.30279404e-01,
1.42966536e-01,

-5.23520055e+00,
-5.33239733e+00,

9
10
11
12

# polynomial fits for pressures

anatase_ppars = array([3.18147737e-03, -2.60558808e-01,
rutile_ppars = array([3.74040625e-03, -2.85933071e-01,

5.23520055e+00])
5.33239733e+00])

13
14
15
16
17

def func(V):
V1 = V[0] # rutile volume
V2 = V[1] # anatase volume

18
19
20

E_rutile = polyval(rutile_epars, V1)

E_anatase = polyval(anatase_epars, V2)

21
22
23

P_rutile = polyval(rutile_ppars, V1)

P_anatase = polyval(anatase_ppars, V2)

24
25
26

return [(E_anatase - E_rutile) / (V1 - V2) - P_anatase,

(E_anatase - E_rutile) / (V1 - V2) - P_rutile]

27
28
29
30
31
32

from scipy.optimize import fsolve

x0 = fsolve(func, [28, 34])
print The solutions are at V = {0}.format(x0)
print Anatase pressure: {0} GPa.format(polyval(anatase_ppars, x0[1]) / GPa)
print Rutile pressure: {0} GPa.format(polyval(rutile_ppars, x0[0]) / GPa)

33
34
35

# illustrate the common tangent

import matplotlib.pyplot as plt

36
37
38
39
40
41
42
43
44
45

vfit = linspace(28, 40)

plt.plot(vfit, polyval(anatase_epars,vfit), label=anatase)
plt.plot(vfit, polyval(rutile_epars,vfit), label=rutile)
plt.plot(x0, [polyval(rutile_epars,x0[0]),
polyval(anatase_epars,x0[1])], ko-, label=common tangent)
plt.legend()
plt.xlabel(Volume ($\AA^3$/f.u.))
plt.ylabel(Total energy (eV/f.u.))
plt.savefig(images/eos-common-tangent.png)

Open the python script (dft-scripts/script-119.py).

The solutions are at V = [ 31.67490656
Anatase pressure: 4.52495142656 GPa
Rutile pressure: 4.5249514267 GPa

34.60893508]

At a pressure of 4.5 GPa, we expect that anatase will start converting into rutile. Along this common
tangent, a mixture of the two phases will be more stable than either pure phase.
TODO add literature discussion
there is some controversy about the most stable phase. add
discussion here.

4.8
4.8.1

Bulk reaction energies

Alloy formation energies

In this section we will consider how to calculate the formation energy of an fcc Cu-Pd alloy and how to
use that information to discuss relative stabilities. The kinds of questions we can easily answer are:
116

1. Is the formation of an alloy at a particular composition and structure energetically favorable?

2. Given two alloy structures at the same composition, which one is more stable?
3. Given a set of alloy structures at different compositions, which ones are stable with respect to
phase separation?
Each of these questions is answered by calculating the formation energy of the alloy from the parent
metals. Thus, we will need the total energies of fcc Cu and fcc Pd. To get started. We get those first.
Rather than compute a full equation of state for these, we will rely on the built in unit cell optimization
algorithm in VASP (ISIF=3).
Basic alloy formation energy
1
2
3

# get bulk Cu and Pd energies. <<pure-metal-components>>

from jasp import *
from ase import Atom, Atoms

4
5
6
7
8

atoms = Atoms([Atom(Cu, [0.000,

cell= [[ 1.818, 0.000,
[ 1.818, 1.818,
[ 0.000, 1.818,

0.000,
1.818],
0.000],
1.818]])

0.000])],

9
10
11
12
13
14
15
16
17
18
19

with jasp(bulk/alloy/cu,
xc=PBE,
encut=350,
kpts=(13, 13, 13),
nbands=9,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
cu = atoms.get_potential_energy()

20
21
22
23
24

atoms = Atoms([Atom(Pd, [0.000,

0.000,
cell=[[ 1.978, 0.000, 1.978],
[ 1.978, 1.978, 0.000],
[0.000, 1.978, 1.978]])

0.000])],

25
26
27
28
29
30
31
32
33
34
35

with jasp(bulk/alloy/pd,
xc=PBE,
encut=350,
kpts=(13, 13, 13),
nbands=9,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
pd = atoms.get_potential_energy()

36
37
38

print Cu energy = {0} eV.format(cu)

print Pd energy = {0} eV.format(pd)

Open the python script (dft-scripts/script-120.py).

Cu energy = -3.723352 eV
Pd energy = -5.18438 eV
Note that the Pd energy is more negative than the Cu energy. This does not mean anything significant.
We cannot say Pd is more stable than Cu; it is not like Cu could transmutate into Pd!
Next, we will consider a question like which of two structures with composition of CuPd is more stable.
These coordinates for these structures came from research of the author. The approach is pretty general,
you must identify the coordinates and unit cell of the candidate structure, and then run a calculation
to find the optimized geometry and unit cell. This may take some work, as previously described in the
multistep process for optimizing a bulk system. Here the geometry is pretty close to optimized, so we
can use the VASP optimization routines. We consider two structures with composition CuPd.
117

1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7
8

atoms = Atoms([Atom(Cu, [0.000,

0.000,
Atom(Pd, [-1.652,
0.000,
cell= [[0.000, -2.039, 2.039],
[0.000, 2.039, 2.039],
[-3.303, 0.000, 0.000]])

0.000]),
2.039])],

9
10
11
12
13
14
15
16
17
18
19

with jasp(bulk/alloy/cupd-1,
xc=PBE,
encut=350,
kpts=(12, 12, 8),
nbands=17,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
cupd1 = atoms.get_potential_energy()

20
21
22
23
24
25
26
27

atoms = Atoms([Atom(Cu, [-0.049,

Atom(Cu, [-11.170,
Atom(Pd, [-7.415,
Atom(Pd, [-3.804 ,
cell=[[-5.629, 3.701,
[-3.701, 5.629,
[-5.629, 5.629,

0.049,
11.170,
7.415,
3.804,
5.629 ],
5.629 ],
3.701 ]])

0.049]),
11.170]),
7.415]),
3.804])],

28
29
30
31
32
33
34
35
36
37
38

with jasp(bulk/alloy/cupd-2,
xc=PBE,
encut=350,
kpts=(8,8,8),
nbands=34,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
cupd2 = atoms.get_potential_energy()

39
40
41

print cupd-1 = {0} eV.format(cupd1)

print cupd-2 = {0} eV.format(cupd2)

Open the python script (dft-scripts/script-121.py).

cupd-1 = -9.148409 eV
cupd-2 = -17.979436 eV
Looking at these energies, you could be tempted to say cupd-2 is more stable than cupd-1 because its
energy is much lower. This is wrong, however, because cupd-2 has twice as many atoms as cupd-1. We
should compare the normalized total energies, that is the energy normalized per CuPd formula unit, or
as an alternative the number of atoms in the unit cell. It does not matter which, as long as we normalize
consistently. It is conventional in alloy calculation to normalize by the number of atoms in the unit cell.
1

from jasp import *

2
3
4
5

with jasp(bulk/alloy/cupd-1) as calc:

atoms = calc.get_atoms()
e1 = atoms.get_potential_energy()/len(atoms)

6
7
8
9

with jasp(bulk/alloy/cupd-2) as calc:

atoms = calc.get_atoms()
e2 = atoms.get_potential_energy()/len(atoms)

10
11
12

print cupd-1: {0} eV/atom.format(e1)

print cupd-2: {0} eV/atom.format(e2)

Open the python script (dft-scripts/script-122.py).

cupd-1: -4.5742045 eV/atom
cupd-2: -4.494859 eV/atom
118

After normalizing by number of atoms, we can see that cupd-1 is a more stable structure. However,
we are looking at total energies, and we might ask: is cupd-1 more stable than an unreacted mixture of
the parent compounds, fcc Cu and Pd? In other words, is the following reaction exothermic:
Cu + Pd CuPd for the two configurations we examined? Below, we show some pretty general
code that computes these formation energies, and normalizes them by the number of atoms in the unit
cell.
1

from jasp import *

2
3
4
5
6

# bulk energy 1
with jasp(bulk/alloy/cu) as calc:
atoms = calc.get_atoms()
cu = atoms.get_potential_energy()/len(atoms)

7
8
9
10
11

# bulk energy 2
with jasp(bulk/alloy/pd) as calc:
atoms = calc.get_atoms()
pd = atoms.get_potential_energy()/len(atoms)

12
13
14
15
16
17
18
19
20
21
22

with jasp(bulk/alloy/cupd-1) as calc:

atoms = calc.get_atoms()
e1 = atoms.get_potential_energy()
# subtract bulk energies off of each atom in cell
for atom in atoms:
if atom.symbol == Cu:
e1 -= cu
else:
e1 -= pd
e1 /= len(atoms) # normalize by number of atoms in cell

23
24
25
26
27
28
29
30
31
32

with jasp(bulk/alloy/cupd-2) as calc:

atoms = calc.get_atoms()
e2 = atoms.get_potential_energy()
for atom in atoms:
if atom.symbol == Cu:
e2 -= cu
else:
e2 -= pd
e2 /= len(atoms)

33
34
35

print Delta Hf cupd-1 = {0:1.2f} eV/atom.format(e1)

print Delta Hf cupd-2 = {0:1.2f} eV/atom.format(e2)

Open the python script (dft-scripts/script-123.py).

Delta Hf cupd-1 = -0.12 eV/atom
Delta Hf cupd-2 = -0.04 eV/atom
The answer is yes. Both structures are energetically more favorable than an equal composition
mixture of the parent metals. The heat of formation for both structures is exothermic, but the cupd-1
structure is more stable than the cupd-2 structure. This is shown conceptually in Figure 45.
We will now examine another structure at another composition and its stability.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7

# parent metals
with jasp(bulk/alloy/cu) as calc:
atoms = calc.get_atoms()
cu = atoms.get_potential_energy()/len(atoms)

8
9
10
11

with jasp(bulk/alloy/pd) as calc:

atoms = calc.get_atoms()
pd = atoms.get_potential_energy()/len(atoms)

12
13
14
15
16

atoms = Atoms([Atom(Cu,
Atom(Cu,
Atom(Cu,
Atom(Pd,

[-3.672,
[0.000,
[-10.821,
[-7.246,

3.672,
0.000,
10.821,
7.246,

3.672]),
0.000]),
10.821]),
7.246])],

119

17
18
19

cell=[[-5.464,
[-3.565,
[-5.464,

3.565,
5.464,
5.464,

5.464],
5.464],
3.565]])

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

with jasp(bulk/alloy/cu3pd-1,
xc=PBE,
encut=350,
kpts=(8, 8, 8),
nbands=34,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
e3 = atoms.get_potential_energy()
for atom in atoms:
if atom.symbol == Cu:
e3 -= cu
else:
e3 -= pd
e3 /= len(atoms)
print Delta Hf cu3pd-1 = {0:1.2f} eV/atom.format(e3)

Open the python script (dft-scripts/script-124.py).

Delta Hf cu3pd-1 = -0.02 eV/atom
The formation energy is slightly exothermic, which means the structure is more stable than a mixture
of the parent metals. However, let us consider whether the structure is stable with respect to phase
separation into pure Cu and the cupd-1 structure. We define the following quantities:
Hf,Cu = 0.0 eV/atom, x0 = 0, Hf,cupd1 = -0.12 eV/atom, x3 = 0.5.
The composition weighted average at xP d = 0.25 is:
x0x
Hf = Hf,Cu + x0x3
(Hf,cupd1 Hf,Cu )
1
2

x0 = 0.0; x3 = 0.5; x = 0.25;

Hf1 = 0.0; Hf3 = -0.12;

3
4

print Composition weighted average

5
6
7

= {0} eV.format(Hf1 +
(x0 - x)
/ (x0 - x3)
* (Hf3 - Hf1))

Open the python script (dft-scripts/script-125.py).

Composition weighted average

= -0.06 eV

We find the weighted composition formation energy of pure Cu and cupd-1 is more favorable than the
formation energy of cu3pd-1. Therefore, we could expect that structure to phase separate into a mixture
of pure Cu and cupd-1. Schematically what we are seeing is shown in Figure ??fig:alloy-phase-separation.
Finally, let us consider one more structure with the Cu3 Pd stoichiometry.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7

# parent metals
with jasp(bulk/alloy/cu) as calc:
atoms = calc.get_atoms()
cu = atoms.get_potential_energy()/len(atoms)

8
9
10
11

with jasp(bulk/alloy/pd) as calc:

atoms = calc.get_atoms()
pd = atoms.get_potential_energy()/len(atoms)

12
13
14
15
16
17

atoms = Atoms([Atom(Cu, [-1.867,

Atom(Cu, [0.000,
Atom(Cu, [0.000,
Atom(Pd, [-1.867,
cell=[[-3.735, 0.000,

1.867,
0.000,
1.867,
0.000,
0.000],

0.000]),
0.000]),
1.867]),
1.86])],

120

[0.000,
[0.000,

18
19

0.000,
3.735,

3.735],
0.000]])

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

with jasp(bulk/alloy/cu3pd-2,
xc=PBE,
encut=350,
kpts=(8, 8, 8),
nbands=34,
ibrion=2,
isif=3,
nsw=10,
atoms=atoms) as calc:
e4 = atoms.get_potential_energy()
for atom in atoms:
if atom.symbol == Cu:
e4 -= cu
else:
e4 -= pd
e4 /= len(atoms)
print Delta Hf cu3pd-2 = {0:1.2f} eV/atom.format(e4)

Open the python script (dft-scripts/script-126.py).

Delta Hf cu3pd-2 = -0.10 eV/atom
This looks promising: the formation energy is much more favorable than cu3pd-1, and it is below the
composition weighted formation energy of -0.06 eV/atom. Consequently, we conclude that this structure
will not phase separate into a mixture of Cu and CuPd. We cannot say, however, if there is a more
stable phase not yet considered, or if it might phase separate into two other phases. We also note here
that we have ignored a few other contributions to alloy stability. We have only considered the electronic
energy contributions to the formation energy. At temperatures above absolute zero there are additional
contributions including configurational and vibrational entropy, which may stabilize some structures
more than others. Finally, our analysis is limited to comparisons of the structures computed on the
fcc lattice. In fact, it is known that the CuPd alloy forms a bcc structure. We did not calculate that
structure, so we can not say if it is more or less stable than the obvious fcc structure we found.
The construction of alloy phase diagrams is difficult. You are always faced with the possibility that
there is a phase that you have not calculated that is more stable than the ones you did calculate. One
approach is to use a tool that automates the discovery of relevant structures such as the Alloy Theoretic
Automated Toolkit (ATAT) 53,54 which uses a cluster expansion methodology.
4.8.2

Metal oxide oxidation energies

We will consider here the reaction 2 Cu2 O + O2

4 CuO. The reaction energy is:
E = 4ECuO 2ECu2 O EO2 . We need to compute the energy of each species.
Cu2 O calculation
1
2
3

# run Cu2O calculation

from jasp import *
from ase import Atom, Atoms

4
5
6

# http://phycomp.technion.ac.il/~ira/types.html#Cu2O
a = 4.27

7
8
9
10
11
12
13

atoms = Atoms([Atom(Cu, [0, 0, 0]),

Atom(Cu, [0.5, 0.5, 0.0]),
Atom(Cu, [0.5, 0.0, 0.5]),
Atom(Cu, [0.0, 0.5, 0.5]),
Atom(O, [0.25, 0.25, 0.25]),
Atom(O, [0.75, 0.75, 0.75])])

14
15

atoms.set_cell((a, a, a), scale_atoms=True)

16
17
18

with jasp(bulk/Cu2O,
encut=400,

121

19
20
21
22
23
24
25
26
27

kpts=(8, 8, 8),
ibrion=2,
isif=3,
nsw=30,
xc=PBE,
atoms=atoms) as calc:
calc.set_nbands()
calc.calculate()
print calc

Open the python script (dft-scripts/script-127.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/Cu2O
converged: True
Energy = -27.204590 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 4.270 0.000 0.000] 4.270
a1 [ 0.000 4.270 0.000] 4.270
a2 [ 0.000 0.000 4.270] 4.270
a,b,c,alpha,beta,gamma (deg):4.270 4.270 4.270 90.0 90.0 90.0
Unit cell volume = 77.854 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.000 0.000 0.000-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Cu [0.000
0.000
0.000] 0
0.00
T T T
1
Cu [2.135
2.135
0.000] 0
0.00
T T T
2
Cu [2.135
0.000
2.135] 0
0.00
T T T
3
Cu [0.000
2.135
2.135] 0
0.00
T T T
4
O
[1.067
1.067
1.067] 0
0.00
T T T
5
O
[3.202
3.202
3.202] 0
0.00
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 37
nsw: 30
ibrion: 2
isif: 3
encut: 400.0
magmom: None
prec: Normal
kpts: (8, 8, 8)
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Cu: potpaw_PBE/Cu/POTCAR (git-hash: a44c591415026f53deb16a99ca3f06b1e69be10b)
CuO calculation
122

1
2
3

# run CuO calculation

from jasp import *
from ase import Atom, Atoms

4
5
6
7
8
9
10
11
12

# CuO
# http://cst-www.nrl.navy.mil/lattice/struk/b26.html
# http://www.springermaterials.com/docs/info/10681727_51.html
a = 4.6837
b = 3.4226
c = 5.1288
beta = 99.54/180*np.pi
y = 0.5819

13
14
15
16

a1 = np.array([0.5*a, -0.5*b, 0.0])

a2 = np.array([0.5*a, 0.5*b, 0.0])
a3 = np.array([c*np.cos(beta), 0.0, c*np.sin(beta)])

17
18
19
20
21
22

atoms = Atoms([Atom(Cu, 0.5*a2),

Atom(Cu, 0.5*a1 + 0.5*a3),
Atom(O, -y*a1 + y*a2 + 0.25*a3),
Atom(O, y*a1 - y*a2 - 0.25*a3)],
cell=(a1, a2, a3))

23
24
25
26
27
28
29
30
31
32
33
34

with jasp(bulk/CuO,
encut=400,
kpts=(8, 8, 8),
ibrion=2,
isif=3,
nsw=30,
xc=PBE,
atoms=atoms) as calc:
calc.set_nbands()
calc.calculate()
print calc

Open the python script (dft-scripts/script-128.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/CuO
converged: True
Energy = -19.509498 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.302 -1.776 0.046] 2.908
a1 [ 2.302 1.776 0.046] 2.908
a2 [-0.762 0.000 5.087] 5.144
a,b,c,alpha,beta,gamma (deg):2.908 2.908 5.144 95.8 95.8 95.8
Unit cell volume = 41.730 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.008 0.013 0.009-0.000 0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Cu [1.151
0.888
0.023] 0
0.00
T T T
1
Cu [0.770
-0.888
2.566] 0
0.00
T T T
2
O
[2.111
-0.168
1.318] 0
0.03
T T T
3
O
[1.730
0.168
3.861] 0
0.03
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 23
nsw: 30
ibrion: 2

123

isif:
encut:
magmom:
prec:
kpts:
reciprocal:
xc:
txt:
gamma:

3
400.0
None
Normal
(8, 8, 8)
False
PBE
False

Pseudopotentials used:
---------------------O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Cu: potpaw_PBE/Cu/POTCAR (git-hash: a44c591415026f53deb16a99ca3f06b1e69be10b)
Reaction energy calculation
1

from jasp import *

2
3
4
5
6
7
8

# dont forget to normalize your total energy to a formula unit. Cu2O

# has 3 atoms, so the number of formula units in an atoms is
# len(atoms)/3.
with jasp(bulk/Cu2O) as calc:
atoms = calc.get_atoms()
cu2o_energy = atoms.get_potential_energy()/(len(atoms)/3)

9
10
11
12

with jasp(bulk/CuO) as calc:

atoms = calc.get_atoms()
cuo_energy = atoms.get_potential_energy()/(len(atoms)/2)

13
14
15
16
17

# make sure to use the same cutoff energy for the O2 molecule!
with jasp(molecules/O2-sp-triplet-400) as calc:
atoms = calc.get_atoms()
o2_energy = atoms.get_potential_energy()

18
19
20

rxn_energy = 4.0*cuo_energy - o2_energy - 2.0*cu2o_energy

print Reaction energy = {0} eV.format(rxn_energy)

Open the python script (dft-scripts/script-129.py).

Reaction energy = -1.966119 eV
This is the reaction energy for 2 Cu2 O 4 CuO. In, 55 the experimental reaction is estimated to be
about -3.14 eV.
There are a few reasons why our number does not agree with the experimental reaction energy. One
reason is related to errors in the O2 dissociation energy, and another reason is related to localization of
electrons in the Cu 3d orbitals. 55 The first error of incorrect O2 dissociation error is a systematic error
that can be corrected empirically. 55 Fixing the second error requires the application of DFT+U (see
DFT+U).
The heat of reaction is reported to be 1000 J/g product at http://onlinelibrary.wiley.com/doi/
10.1002/er.4440130107/pdf for the reaction 2CuO
Cu2 O + 1/2 O2 .
1
2
3

from ase import Atoms

atoms = Atoms(Cu2O)
MW = atoms.get_masses().sum()

4
5
6

H = 1. # kJ/g
print rxn energy = {0:1.1f} eV.format(-2 * H * MW / 96.4)

# convert to eV

Open the python script (dft-scripts/script-130.py).

124

rxn energy = -3.0 eV

This is pretty close to the value in 55 and might need a temperature correction to get agreement at
298K.

4.9

Bulk density of states

The density of states refers to the number of electronic states in a particular energy range.
The solution to Eq. (2.1) yields a set of Kohn-Sham (K-S) orbitals and an associated set of eigenvalues
that correspond to the energies of these orbitals, neither of which have any known directly observable
meaning. 18 The sum of the squared K-S orbitals, however, is equal to the electron density (Eq. (2.1)), and
the sum of the eigenvalues is a significant part of the total energy (Eq. (2.1)). Thus, it seems reasonable
to suppose these quantities have other significant relationships to physical observables. Perdew et al.
showed that the highest occupied eigenvalue is equal to the ionization energy of a system within an exact
density functional theory, 24 but their interpretation has been vigorously debated in the literature, 5658
and is only true for the exact exchange/correlation functional, not the approximate ones used in practice. 3
Stowasser and Hoffmann discussed an approach to using the K-S orbitals in more traditional molecular
orbital interpretations, but the results were primarily qualitative. 59 More recently, a DFT analog of
Koopmans theorem has been developed that formally identifies the eigenvalues with vertical ionization
potentials, which can be measured with photoelectron spectroscopy. 60
Despite the arguments against ascribing physical meaning to the K-S orbitals and eigenvalues, it has
become fairly standard, especially for solids, to use them to calculate the density of states (DOS) 61 [Sec.
VI. B]. This has been found to yield reasonable results for the valence bands in metals, but poor results
for tightly bound orbitals and band gaps. 24 A highly technical discussion of this issue can be found in
Ref. 62 . The density of states can be calculated by a sum over the k-points: 15

()=

P
P
_\mathbf{k} _\mathbf{k} _i ( -i {k}) (7)

where k is the weight associated with the k-point, and is a broadening function, typically a
gaussian function, to account for the finite number of k-points used in the calculations. The amount of
broadening is arbitrary, and should tend to zero as the number of k-points approaches infinity.
1

from jasp import *

2
3
4

npoints = 200
width = 0.1

5
6
7
8

def gaussian(energies, eik):

x = ((energies - eik) / width)
return np.exp(-x**2) / np.sqrt(np.pi) / width

9
10

with jasp(bulk/pd-dos) as calc:

11
12
13

# kpt weights
wk = calc.get_k_point_weights()

14
15
16
17
18
19

# for each k-point there are a series of eigenvalues

# here we get all the eigenvalues for each k-point
e_kn = []
for i, k in enumerate(wk):
e_kn.append(calc.get_eigenvalues(kpt=i))

20
21

e_kn = np.array(e_kn) - calc.get_fermi_level()

22
23
24

# these are the energies we want to evaluate the dos at

energies = np.linspace(e_kn.min(), e_kn.max(), npoints)

25
26
27

# this is where we build up the dos

dos = np.zeros(npoints)

125

28
29
30
31
32

for j in range(npoints):
for k in range(len(wk)): # loop over all kpoints
for i in range(len(e_kn[k])): # loop over eigenvalues in each k
dos[j] += wk[k] * gaussian(energies[j], e_kn[k][i])

33
34
35
36
37

import matplotlib.pyplot as plt

plt.plot(energies, dos)
plt.savefig(images/manual-dos.png)
plt.show()

Open the python script (dft-scripts/script-131.py).

Here is a more convenient way to compute the DOS using mod:ase.dft.

1
2
3

from jasp import *

import matplotlib.pyplot as plt
from ase.dft import DOS

4
5
6
7
8

with jasp(bulk/pd-dos) as calc:

dos = DOS(calc, width=0.2)
d = dos.get_dos()
e = dos.get_energies()

9
10
11
12
13
14

import pylab as plt

plt.plot(e, d)
plt.xlabel(energy (eV))
plt.ylabel(DOS)
plt.savefig(images/pd-dos.png)

Open the python script (dft-scripts/script-132.py).

This DOS looks roughly like you would expect. The peak between -5 to 0 eV is the Pd d-band.
The VASP manual recommends a final run be made with ISMEAR=-5, which uses the tetrahedron
method with Bl\"ochl corrections.

126

1
2
3
4

from jasp import *

from ase.dft import DOS
with jasp(bulk/pd-dos) as calc:
calc.clone(bulk/pd-dos-ismear-5)

5
6
7
8

with jasp(bulk/pd-dos-ismear-5) as calc:

bulk = calc.get_atoms()
calc.set(ismear=-5)

9
10
11
12
13

bulk.get_potential_energy()
dos = DOS(calc, width=0.2)
d = dos.get_dos()
e = dos.get_energies()

14
15
16
17
18
19

import pylab as plt

plt.plot(e, d)
plt.xlabel(energy [eV])
plt.ylabel(DOS)
plt.savefig(images/pd-dos-ismear-5.png)

Open the python script (dft-scripts/script-133.py).

This not notably different to me.
We can test for convergence of the DOS. The k-points are most important.
1
2
3

from ase import Atoms, Atom

from jasp import *
import sys

4
5
6
7
8

import matplotlib.pyplot as plt

import numpy as np
from ase.dft import DOS
import pylab as plt

9
10
11
12
13
14
15

a = 3.9 # approximate lattice constant

b = a / 2.
bulk = Atoms([Atom(Pd, (0.0, 0.0, 0.0))],
cell=[(0, b, b),
(b, 0, b),
(b, b, 0)])

16
17

kpts = [8, 10, 12, 14, 16, 18, 20]

18
19
20
21
22
23
24

for k in kpts:
with jasp(bulk/pd-dos-k{0}-ismear-5.format(k),
encut=300,
xc=PBE,
kpts=(k, k, k),
atoms=bulk) as calc:

25

# this runs the calculation

try:
bulk.get_potential_energy()
dos = DOS(calc, width=0.2)
d = dos.get_dos() + k / 4.0
e = dos.get_energies()

26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

plt.plot(e,d, label=k={0}.format(k))
except:
pass
plt.xlabel(energy (eV))
plt.ylabel(DOS)
plt.legend()
plt.savefig(images/pd-dos-k-convergence-ismear-5.png)
plt.show()

Open the python script (dft-scripts/script-134.py).

4.10

Atom projected density of states

One major disadvantage of a planewave basis set is that it is difficult to relate the completely delocalized
planewaves to localized phenomena such as bonding. Much insight into bonding has been gained by
127

atomic/molecular orbital theory, which has carried over to the solid-state arena. 7 Consequently, several
schemes have been developed to project the one-electron Kohn-Sham wave functions onto atomic wave
functions. 6365 In VASP, the one electron wave functions can be projected onto spherical harmonic
orbitals. The radial component of the atomic orbitals extends to infinity. In a solid, this means that
the projection on one atom may overlap with the projection on a neighboring atom, resulting in double
counting of electrons. Consequently, a cutoff radius was introduced, beyond which no contributions are
included. It is not obvious what the best cutoff radius is. If the radius is too small, it might not capture
all of the electrons associated with the atom. However, if it is too large, it may include electrons from
neighboring atoms. One might want to use different cutoff radii for different atoms, which have different
sizes. Furthermore, the ideal cutoff radius for an atom may change in different environments, thus it
would require an iterative procedure to determine it. This difficulty arises because the orbital-band
occupations are not observable, thus how the electrons are divided up between atoms is arbitrary and,
as will be seen later, is sensitive to the cutoff radius (and in other DFT implementations, the basis
set). However, Mulliken orbital populations have been used successfully for many years to examine the
qualitative differences between similar systems, and that is precisely what these quantities are used for
here. Thus, a discussion of the analysis and results is warranted.
The s and p states in a metal are typically delocalized in space and more like free-electrons, whereas
the d-orbitals are fairly localized in space and have been treated successfully with tight-binding theories
such as extended H\"u ckel theory, 7 and linear muffin tin orbital theory. 66 Consequently, the remaining
discussion will be focused on the properties of the projected d-states.
In this example, we consider how to get the atom-projected density of states (ADOS). We are interested in properties of the d-band on Pd, such as the d-band center and d-band width. You must set
the RWIGS tag to get ADOS, and these are the Wigner-Seitz radii for each atom. By integrating the
projected d-band up to the Fermi level, the d-band filling can be determined. It is not obvious what the
electron count in the d-band should be for an atom in a metal. For a gas-phase, neutral metal atom in
the ground state, however, the d-orbital electron count is well defined, so it will be used as an initial
reference point for comparison. 4
A powerful method for characterizing distributions is to examine various moments of the distribution
(see Chapter 4 in Ref. 67 and Chapter 6 in Refs. 68 and 69 ). The nth order moment, n , of a distribution
of states () with respect to a reference o is defined by
R n
 ( o )d
R
(8)
n =
( o )d

In this work, the reference energy is always the Fermi level. The zeroth moment is just the total number of
states, in this case it will be normalized to unity. The first moment is the average energy of distribution,
analogous to the center of mass for a mass density distribution. The second moment is the mean squared
width of the distribution. The third moment is a measure of skewness and the fourth moment is related
to kurtosis, but these moments are rarely used, and only the first and second moments are considered in
this work.
It is important to note that these projected density of states are not physical observables. They
are the wavefunctions projected onto atomic orbitals. For some situations this makes sense, e.g. the d
orbitals are fairly localized and reasonably approximated by atomic orbitals. The s valence orbitals in a
metal, in contrast, are almost totally delocalized. Depending on the cutoff radius (RWIGS) you choose,
you can see very different ADOS.
1
2
3

from ase import Atoms, Atom

from jasp import *
from ase.calculators.vasp import VaspDos

4
5
6

import matplotlib.pyplot as plt

import numpy as np

7
8
9
10

a = 3.9 # approximate lattice constant

b = a / 2.
bulk = Atoms([Atom(Pd, (0.0, 0.0, 0.0))],

128

cell=[(0, b, b),
(b, 0, b),
(b, b, 0)])

11
12
13
14
15
16
17
18
19
20
21

with jasp(bulk/pd-ados,
encut=300,
xc=PBE,
lreal=False,
rwigs=[1.5], # wigner-seitz radii for ados
kpts=(8, 8, 8),
atoms=bulk) as calc:

22
23
24

# this runs the calculation

bulk.get_potential_energy()

25
26
27

# now get results

ados = VaspDos(efermi=calc.get_fermi_level())

28
29
30

energies = ados.energy
dos = ados.site_dos(0, d)

# energy is an attribute, not a function

# this is a function, get d-orbitals on atom 0

31
32
33

# we will select energies in the range of -10, 5

ind = (energies < 5) & (energies > -10)

34
35
36

energies = energies[ind]
dos = dos[ind]

37
38
39
40
41
42

Nstates = np.trapz(dos, energies)

occupied = energies <= 0.0
N_occupied_states = np.trapz(dos[occupied], energies[occupied])
# first moment
ed = np.trapz(energies * dos, energies) / np.trapz(dos, energies)

43
44
45

# second moment
wd2 = np.trapz(energies**2 * dos, energies) / np.trapz(dos, energies)

46
47
48
49
50

print
print
print
print

Total # states = {0:1.2f}.format(Nstates)

number of occupied states = {0:1.2f}.format(N_occupied_states)
d-band center = {0:1.2f} eV.format(ed)
d-band width = {0:1.2f} eV.format(np.sqrt(wd2))

51
52
53

# plot the d-band

plt.plot(energies, dos, label=$d$-orbitals)

54
55
56
57
58
59

# plot the occupied states in shaded gray

plt.fill_between(x=energies[occupied],
y1=dos[occupied],
y2=np.zeros(dos[occupied].shape),
color=gray, alpha=0.25)

60
61
62

plt.xlabel($E - E_f$ (eV))

plt.ylabel(DOS (arbitrary units))

63
64
65

plt.savefig(images/pd-ados.png)
plt.show()

Open the python script (dft-scripts/script-135.py).

Total # states = 9.29
number of occupied states = 8.18
d-band center = -2.00 eV
d-band width = 2.72 eV

4.10.1

Effect of RWIGS on ADOS

Here we examine the effect of changing RWIGS on the number of counted electrons, and properties of
the d-band moments.
1
2
3

from ase import Atoms, Atom

from jasp import *
from ase.calculators.vasp import VaspDos

129

import sys

5
6
7

import matplotlib.pyplot as plt

import numpy as np

8
9
10
11
12
13
14

a = 3.9 # approximate lattice constant

b = a / 2.
bulk = Atoms([Atom(Pd, (0.0, 0.0, 0.0))],
cell=[(0, b, b),
(b, 0, b),
(b, b, 0)])

15
16

RWIGS = [1.0, 1.1, 1.25, 1.5, 2.0, 2.5, 3.0, 4.0, 5.0 ]

17
18

ED, WD, N = [], [], []

19
20
21
22

for rwigs in RWIGS:

with jasp(bulk/pd-ados) as calc:
calc.clone(bulk/pd-ados-rwigs-{0}.format(rwigs))

23
24
25

with jasp(bulk/pd-ados-rwigs-{0}.format(rwigs)) as calc:

calc.set(rwigs=[rwigs])

26
27

try:

28

calc.calculate()
except (VaspSubmitted, VaspQueued):
continue

29
30
31
32
33

# now get results

ados = VaspDos(efermi=calc.get_fermi_level())

34
35
36

energies = ados.energy
dos = ados.site_dos(0, d)

37
38
39

#we will select energies in the range of -10, 5

ind = (energies < 5) & (energies > -10)

40
41
42

energies = energies[ind]
dos = dos[ind]

43
44
45
46
47
48

Nstates = np.trapz(dos, energies)

occupied = energies <= 0.0
N_occupied_states = np.trapz(dos[occupied], energies[occupied])
ed = np.trapz(energies * dos, energies) / np.trapz(dos, energies)
wd2 = np.trapz(energies**2 * dos, energies) / np.trapz(dos, energies)

49
50
51
52

N.append(N_occupied_states)
ED.append(ed)
WD.append(wd2**0.5)

53
54
55
56
57
58

plt.plot(RWIGS, N, bo, label=N. occupied states)

plt.legend(loc=best)
plt.xlabel(RWIGS ($\AA$))
plt.ylabel(# occupied states)
plt.savefig(images/ados-rwigs-occupation.png)

59
60
61
62
63
64
65
66

fig, ax1 = plt.subplots()

ax1.plot(RWIGS, ED, bo, label=d-band center (eV))
ax1.set_xlabel(RWIGS ($\AA$))
ax1.set_ylabel(d-band center (eV), color=b)
for tl in ax1.get_yticklabels():
tl.set_color(b)

67
68
69
70
71
72

ax2 = ax1.twinx()
ax2.plot(RWIGS, WD, gs, label=d-band width (eV))
ax2.set_ylabel(d-band width (eV), color=g)
for tl in ax2.get_yticklabels():
tl.set_color(g)

73
74
75

plt.savefig(images/ados-rwigs-moments.png)
plt.show()

Open the python script (dft-scripts/script-136.py).

130

You can see the number of occupied states increases approximately linearly here with RWIGS. This
is due to overcounting of neighboring electrons.
The d-band center and width also change.

4.11

Band structures

To compute a band structure we do two things. First, we compute the self-consistent band structure.
Then we compute the band structure at the desired k-points. We will use Si as an example (adapted
from http://bbs.sciencenet.cn/bbs/upload/20083418325986.pdf).
First, we get the self-consistent electron density in a calculation.
1
2
3

from jasp import *

from ase import Atom, Atoms
from ase.visualize import view

4
5
6
7

a = 5.38936
atoms = Atoms([Atom(Si, [0, 0, 0]),
Atom(Si, [0.25, 0.25, 0.25])])

8
9
10
11

atoms.set_cell([[a / 2., a / 2., 0.0],

[0.0, a / 2., a / 2.],
[a / 2., 0.0, a / 2.]], scale_atoms=True)

12
13
14
15
16
17
18
19
20
21
22

with jasp(bulk/Si-selfconsistent,
xc=PBE,
prec=Medium,
istart=0,
icharg=2,
ediff=0.1e-03,
kpts=(4, 4, 4), debug=logging.DEBUG,
atoms=atoms) as calc:
atoms.get_potential_energy()
print calc

Open the python script (dft-scripts/script-137.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/Si-selfconsistent
converged: True
Energy = -10.799983 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.695 2.695 0.000] 3.811
a1 [ 0.000 2.695 2.695] 3.811
a2 [ 2.695 0.000 2.695] 3.811
a,b,c,alpha,beta,gamma (deg):3.811 3.811 3.811 60.0 60.0 60.0
Unit cell volume = 39.134 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
-0.026 -0.026 -0.026-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Si [0.000
0.000
0.000] 0
0.00
T T T
1
Si [1.347
1.347
1.347] 0
0.00
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 8
istart: 0
icharg: 2
131

ediff:
magmom:
prec:
kpts:
reciprocal:
xc:
txt:
gamma:

0.0001
None
Medium
(4, 4, 4)
False
PBE
False

Pseudopotentials used:
---------------------Si: potpaw_PBE/Si/POTCAR (git-hash: ae2291fdf186a34272463039fb5a87cde0e50174)
Now, we run a new calculation along the k-point path desired. The standard VASP way of doing this
is to modify the INCAR and KPOINTS file and rerun VASP. We will not do that. Doing that results in
some lost information if you overwrite the old files. We will copy the old directory to a new directory,
using code to ensure this only happens one time.
1

from jasp import *

2
3

wd = bulk/Si-bandstructure

4
5
6

with jasp(bulk/Si-selfconsistent) as calc:

calc.clone(wd)

7
8
9
10
11

kpts = [[0.5, 0.5, 0.0],

[0, 0, 0],
[0, 0, 0],
[0.5, 0.5, 0.5]]

# L
# Gamma
# X

12
13
14
15
16
17
18
19

with jasp(wd,
kpts=kpts,
reciprocal=True,
kpts_nintersections=10,
icharg=11) as calc:
calc.calculate()
print(calc)

Open the python script (dft-scripts/script-138.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/bulk/Si-bandstructure
converged: True
Energy = -3.620909 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.695 2.695 0.000] 3.811
a1 [ 0.000 2.695 2.695] 3.811
a2 [ 2.695 0.000 2.695] 3.811
a,b,c,alpha,beta,gamma (deg):3.811 3.811 3.811 60.0 60.0 60.0
Unit cell volume = 39.134 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
-0.205 -0.205 -0.205-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Si [0.000
0.000
0.000] 0
0.00
T T T
1
Si [1.347
1.347
1.347] 0
0.00
T T T
--------------------------------------------------

132

INCAR Parameters:
----------------nbands: 8
istart: 0
icharg: 11
ediff: 0.0001
magmom: None
prec: Medium
kpts: [[0.5, 0.5, 0.0], [0, 0, 0], [0, 0, 0], [0.5, 0.5, 0.5]]
kpts_nintersections: 10
reciprocal: True
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------Si: potpaw_PBE/Si/POTCAR (git-hash: ae2291fdf186a34272463039fb5a87cde0e50174)
We will learn how to manually parse the EIGENVAL file here to generate the band structure. The
structure of the EIGENVAL file looks like this:
1

head -n 20 bulk/Si-bandstructure/EIGENVAL

Open the python script (dft-scripts/script-139.py).

2
2
1
1
0.1956688E+02 0.3810853E-09
1.000000000000000E-004
CAR
unknown system
8
20
8

0.3810853E-09

0.3810853E-09

0.5000000E+00 0.5000000E+00
1
-1.826311
2
-1.826306
3
3.153690
4
3.153716
5
6.744294
6
6.744321
7
16.392989
8
16.405965

0.0000000E+00

0.5000000E-01

0.4444444E+00 0.4444444E+00
1
-2.669062
2
-0.918013

0.0000000E+00

0.5000000E-01

We can ignore the first five lines.

1

f = open(bulk/Si-bandstructure/EIGENVAL, r)

2
3
4
5
6

line1
line2
line3
line4

=
=
=
=

f.readline()
f.readline()
f.readline()
f.readline()

133

0.5000000E-15

7
8

comment = f.readline()
unknown, nkpoints, nbands = [int(x) for x in f.readline().split()]

9
10

blankline = f.readline()

11
12

band_energies = [[] for i in range(nbands)]

13
14
15

for i in range(nkpoints):
x, y, z, weight = [float(x) for x in f.readline().split()]

16
17
18
19
20
21
22

for j in range(nbands):
fields = f.readline().split()
id, energy = int(fields[0]), float(fields[1])
band_energies[id - 1].append(energy)
blankline = f.readline()
f.close()

23
24

import matplotlib.pyplot as plt

25
26
27

for i in range(nbands):
plt.plot(range(nkpoints), band_energies[i])

28
29
30
31
32
33
34
35

ax = plt.gca()
ax.set_xticks([]) # no tick marks
plt.xlabel(k-vector)
plt.ylabel(Energy (eV))
ax.set_xticks([0, 10, 19])
ax.set_xticklabels([$L$, $\Gamma$, $X$])
plt.savefig(images/Si-bandstructure.png)

Open the python script (dft-scripts/script-140.py).

Next we will examine the connection between band structures and density of states. In this example,
we will compute the band structure of TiO2 using a function built into mod:jasp to do the analysis
described above.
1
2

from jasp import *

from jasp.jasp_bandstructure import *

3
4
5
6
7
8
9
10

with jasp(bulk/tio2/step3) as calc:

n, bands, p = calc.get_bandstructure(kpts_path=[($\Gamma$, [0.0, 0.0, 0.0]),
(X, [0.5, 0.5, 0.0]),
(X, [0.5, 0.5, 0.0]),
(M, [0.0, 0.5, 0.5]),
(M, [0.0, 0.5, 0.5]),
($\Gamma$, [0.0, 0.0, 0.0])])

11
12

p.savefig(images/tio2-bandstructure-dos.png)

Open the python script (dft-scripts/script-141.py).

4.11.1

create example showing band dispersion with change in lattice constant

In this section, we examine the effect of the lattice constant on the band structure. Since the lattice
constant affects the overlap of neighboring atoms, we expect that smaller lattice constants will show
more dispersion, i.e. broader bands. Larger lattice constants, in contrast, should show narrower bands.
We examine this in silicon.
1
2

from jasp import *

from jasp.jasp_bandstructure import *

3
4
5
6

from ase import Atom, Atoms

ready = True
for i,a in enumerate([4.7, 5.38936, 6.0]):

7
8
9

atoms = Atoms([Atom(Si, [0, 0, 0]),

Atom(Si, [0.25, 0.25, 0.25])])

10
11
12

atoms.set_cell([[a/2., a/2., 0.0],

[0.0, a/2., a/2.],

134

[a/2., 0.0, a/2.]], scale_atoms=True)

13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

with jasp(bulk/Si-bs-{0}.format(i),
xc=PBE,
prec=Medium,
istart=0,
icharg=2,
ediff=0.1e-03,
kpts=(4, 4, 4),
atoms=atoms) as calc:
try:
n,bands,p = calc.get_bandstructure(kpts_path=[(L, [0.5,0.5,0.0]),
($\Gamma$, [0, 0, 0]),
($\Gamma$, [0, 0, 0]),
(X, [0.5, 0.5, 0.5])],
kpts_nintersections=10)
except (VaspSubmitted, VaspQueued):
print not ready {0}.format(i)
ready = False

32
33
34

if not ready:
import sys; sys.exit()

35
36
37
38

p.savefig(images/Si-bs-{0}.png.format(i))
p.show()

Open the python script (dft-scripts/script-142.py).

not ready 0
You can see the band structure for a=6.0 is notably sharper than the band structure for a=4.0.

4.12

Magnetism

4.12.1

Determining if a magnetic solution is energetically favorable

We can force a total magnetic moment onto a unit cell and compute the total energy as function of
the total magnetic moment. If there is a minimum in the energy, then we know there is a lower energy
magnetic solution than a non-magnetic solution. We use incar:NUPDOWN to enforce the magnetic
moment in the cell. Note that NUPDOWN can only be an integer. You cannot set it to be an arbitrary
float.
1
2

from jasp import *

from ase.lattice.cubic import BodyCenteredCubic

3
4
5
6
7
8

atoms = BodyCenteredCubic(directions=[[1, 0, 0],

[0, 1, 0],
[0, 0, 1]],
size=(1, 1, 1),
symbol=Fe)

9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

NUPDOWNS
energies
for B in
with

= [0.0, 2.0, 4.0, 5.0, 6.0, 8.0]

= []
NUPDOWNS:
jasp(bulk/Fe-bcc-fixedmagmom-{0:1.2f}.format(B),
xc=PBE,
encut=300,
kpts=(4, 4, 4),
ispin=2,
nupdown=B,
atoms=atoms) as calc:
try:
e = atoms.get_potential_energy()
energies.append(e)
except (VaspSubmitted, VaspQueued):
pass

25
26
27

import matplotlib.pyplot as plt

plt.plot(NUPDOWNS, energies)

135

28
29
30
31

plt.xlabel(Total Magnetic Moment)

plt.ylabel(Energy (eV))
plt.savefig(images/Fe-fixedmagmom.png)
plt.show()

Open the python script (dft-scripts/script-143.py).

You can see here there is a minimum in energy at a total magnetic moment somewhere between 4
and 5. There are two Fe atoms in the unit cell, which means the magnetic moment on each atom must
be about 2.5 Bohr-magnetons. This is a good guess for a real calculation. Note that VASP recommends
you overestimate the magnetic moment guesses if you are looking for ferromagnetic solutions.
To run a spin-polarized calculation with initial guesses on each atom, we must set the magnetic
moment on the atoms. Here we set it through the magmom attribute on the atom. In the example after
this, we set it in the Atoms object.
1
2

from jasp import *

from ase.lattice.cubic import BodyCenteredCubic

3
4
5
6
7
8

atoms = BodyCenteredCubic(directions=[[1, 0, 0],

[0, 1, 0],
[0, 0, 1]],
size=(1, 1, 1),
symbol=Fe)

9
10
11
12

# set magnetic moments on each atom

for atom in atoms:
atom.magmom = 2.5

13
14
15
16
17
18
19
20
21
22
23
24
25
26

with jasp(bulk/Fe-bcc-sp-1,
xc=PBE,
encut=300,
kpts=(4, 4, 4),
ispin=2,
lorbit=11, # you need this for individual magnetic moments
atoms=atoms) as calc:
try:
e = atoms.get_potential_energy()
B = atoms.get_magnetic_moment()
magmoms = atoms.get_magnetic_moments()
except (VaspSubmitted, VaspQueued):
pass

27
28
29

print Total magnetic moment is {0:1.2f} Bohr-magnetons.format(B)

print Individual moments are {0} Bohr-magnetons.format(magmoms)

Open the python script (dft-scripts/script-144.py).

Total magnetic moment is 4.25 Bohr-magnetons
Individual moments are [ 2.134 2.134] Bohr-magnetons
4.12.2

Antiferromagnetic spin states

In an antiferromagnetic material, there are equal numbers of spin up and down electrons that align in a
regular pattern, but pointing in opposite directions so that there is no net magnetism. It is possible to
model this by setting the magnetic moments on each mod:ase.Atom object.
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5
6
7
8
9
10

atoms = Atoms([Atom(Fe, [0.00, 0.00, 0.00], magmom=5),

Atom(Fe, [4.3,
4.3,
4.3], magmom=-5),
Atom(O, [2.15, 2.15, 2.15], magmom=0),
Atom(O, [6.45, 6.45, 6.45], magmom=0)],
cell=[[4.3,
2.15,
2.15],
[2.15,
4.3,
2.15],
[2.15,
2.15,
4.3]])

11
12
13
14
15
16
17
18
19
20
21

with jasp(bulk/afm-feo,
encut=350,
prec=Normal,
ispin=2,
nupdown=0, # this forces a non-magnetic solution
lorbit=11, # to get individual moments
lreal=False,
136
atoms=atoms) as calc:
print Magnetic moments = , atoms.get_magnetic_moments()
print Total magnetic moment = , atoms.get_magnetic_moment()

Open the python script (dft-scripts/script-145.py).

Magnetic moments =

[ 2.652 -2.652

0.

0.

Figure 30: A side view of graphite.

137

Figure 31: A view of a NaCl crystal structure.

138

Figure 33: k-point convergence of the total energy of fcc Ag.

139

Figure 34: Effects of SIGMA on the occupancies of the Cu system.

140

Figure 35: Total energy vs. fcc lattice contant for Cu. It appears the minimum is near 3.65 .

Figure 36: Total energy vs. volume for fcc Cu with fitted cubic polynomial equation of state.

141

Figure 37: Comparison of energies between fcc and bcc Cu. The fcc structure is lower in energy.

142

Figure 38: Total energy vs. c/a for different values of a.

143

Figure 39: Contour plot of the total energy of hcp Ru for different values of a and c/a.

144

Figure 40: k-point convergence of rutile TiO2 .

145

Figure 41: Total energy vs. volume for rutile TiO2 in step 1 of the optimization.

146

Figure 42: Total energy vs. volume for step 2 of the unit cell optimization.

147

Figure 43: Equations of state (E(V)) for anatase and rutile TiO2 .

148

Figure 44: Illustration of the common tangent that shows the pressure where anatase and rutile coexist
before anatase converts to rutile.

149

Figure 45: Conceptual picture of two alloys with exothermic formation energies. The dashed line represents a composition weighted average energy of the parent metals. E4 and E3 are energies associated
with two different alloy structures at the same composition. Both structures are more stable than a
mixture of pure metals with the same composition, but E3 is more stable than E4.

150

Figure 46: Illustration of of an alloy structure with an exothermic formation energy that is not stable
with respect to phase separation. The solid line shows the composition weighted average energy of a
mixture of Cu and cupd-2. Since the energy of cu3pd-1 is above the solid line, it is less favorable than a
mixture of Cu and cupd-2 with the same composition.

151

Figure 47: Illustration that cu3pd-2 is more stable than cu3pd-1 and that is it is more stable than a
composition weighted mixture of Cu and cupd-1. The dotted line shows the energy of a composition
weighted average energy of a mixture of Cu and cupd-1. Since cu3pd-2 is below the dotted line, it is
more stable than the phase-separated mixture.

152

Figure 48: Total DOS for bulk Pd.

153

Surfaces

5.1

Surface structures

As with molecules and bulk systems mod:ase provides several convenience functions for making surfaces.
5.1.1

Simple surfaces

ase provides many utility functions to setup surfaces. Here is a simple example of an fcc111 Al surface.
There are built in functions for fcc111, bcc110, bcc111, hcp001 and diamond111.
1
2

from ase.lattice.surface import fcc111

from ase.io import write

3
4

slab = fcc111(Al, size=(2, 2, 3), vacuum=10.0)

5
6

write(images/Al-slab.png, slab, rotation=90x, show_unit_cell=2)

Open the python script (dft-scripts/script-146.py).

154

Figure 60: An Al(111) slab with three layers and 20 of vacuum.

5.1.2

Vicinal surfaces

The vast majority of surface calculations are performed on flat surfaces. This is partially because these
surfaces tend to have the lowest surface energies, and thus are likely to be experimentally observed. The
flat surfaces, also known as low Miller index surfaces, also have small unit cells, which tends to make
them computationally affordable. There are, however, many reasons to model the properties of surfaces

155

that are not flat. You may be interested in the reactivity of a step edge, for example, or you may use the
lower cooridnation of steps as a proxy for nanoparticle reactivity. Many stepped surfaces are not that
difficult to make now. The main idea in generating them is described here. mod:ase provides a general
function for making vicinal surfaces. Here is an example of a (211) surface.
1
2

from ase.lattice.surface import surface

from ase.io import write

3
4
5
6

# Au(211) with 9 layers

s1 = surface(Au, (2, 1, 1), 9)
s1.center(vacuum=10, axis=2)

7
8
9
10
11

write(images/Au-211.png,
s1.repeat((3, 3, 1)),
rotation=-30z,90x, # change the orientation for viewing
show_unit_cell=2)

Open the python script (dft-scripts/script-147.py).

156

Figure 61: An Au(211) surface constructed with ase.

157

5.2

Surface calculation parameters

There is one important parameter that is different for surfaces than for bulk calculations, the k-point
grid. Assuming you have followed the convention that the z-axis is normal to the surface, the k-point
grids for slab calculations always have the form of M N 1. To illustrate why, consider this example:
1
2
3
4
5

from
from
from
from
slab

ase.lattice.surface import fcc111

ase.io import write
jasp import *
jasp.jasp_bandstructure import *
= fcc111(Al, size=(1, 1, 4), vacuum=10.0)

6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

with jasp(surfaces/Al-bandstructure,
xc=PBE,
encut=300,
kpts=(6, 6, 6),
atoms=slab) as calc:
n, bands, p = calc.get_bandstructure(kpts_path=[(r$\Gamma$, [0, 0, 0]),
($K1$, [0.5, 0.0, 0.0]),
($K1$, [0.5, 0.0, 0.0]),
($K2$, [0.5, 0.5, 0.0]),
($K2$, [0.5, 0.5, 0.0]),
(r$\Gamma$, [0, 0, 0]),
(r$\Gamma$, [0, 0, 0]),
($K3$, [0.0, 0.0, 1.0])],
kpts_nintersections=10)

21
22
23

p.savefig(images/Al-slab-bandstructure.png)
p.show()

Open the python script (dft-scripts/script-148.py).

5.3

Surface relaxation

When a surface is created, the bulk symmetry is broken and consequently there will be forces on the
surface atoms. We will examine some consequences of this with a simple Al slab. First, we show there
are forces on the slab atoms.
1
2

from jasp import *

from ase.lattice.surface import fcc111

3
4

atoms = fcc111(Al, size=(1, 1, 4), vacuum=10.0)

5
6
7
8
9
10
11
12

with jasp(surfaces/Al-slab-unrelaxed,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
atoms=atoms) as calc:
atoms.get_forces()
print calc

Open the python script (dft-scripts/script-149.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/surfaces/Al-slab-unrelaxed
converged: True
Energy = -14.177069 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.864 0.000 0.000] 2.864
a1 [ 1.432 2.480 0.000] 2.864
a2 [ 0.000 0.000 27.015] 27.015
a,b,c,alpha,beta,gamma (deg):2.864 2.864 27.015 90.0 90.0 90.0
158

Figure 62: Band structure of an Al slab in the plane (path from Gamma to K1 to K2 to Gamma) and
normal to the surface (Gamma to K3). Note the bands are flat in the direction normal to the surface,
hence only one k-point is needed in this direction.
Unit cell volume = 191.872 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.007 0.007 0.002-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Al [0.000
0.000
10.000] 0
0.01
T T
1
Al [1.432
0.827
12.338] 0
0.18
T T
2
Al [2.864
1.653
14.677] 0
0.18
T T
3
Al [0.000
0.000
17.015] 0
0.01
T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 10
encut: 350.0
magmom: None
prec: Normal
kpts: (6, 6, 1)
reciprocal: False
xc: PBE
txt: gamma: False

159

T
T
T
T

Pseudopotentials used:
---------------------Al: potpaw_PBE/Al/POTCAR (git-hash: c8d9ecb0b6ebec0256c5f5072cee4de6a046dac2)
Some points to note. The forces on the atoms have symmetry to them. That is because the slab is
centered. Had the slab had an odd number of atoms, it is likely the center atom would have no forces
on it. Next we consider the spacing between each layer in the slab. We do this for comparison later.
1

from jasp import *

2
3
4
5

with jasp(surfaces/Al-slab-unrelaxed) as calc:

atoms = calc.get_atoms()
print Total energy: {0:1.3f} eV.format(atoms.get_potential_energy())

6
7
8

for i in range(1, len(atoms)):

print {0} deltaz = {1:1.3f} angstroms.format(i, atoms[i].z - atoms[i-1].z)

Open the python script (dft-scripts/script-150.py).

Total energy: -14.177 eV
1 deltaz = 2.338 angstroms
2 deltaz = 2.338 angstroms
3 deltaz = 2.338 angstroms
To reduce the forces, we can let VASP relax the geometry. We have to make some decisions about
how to relax the slab. One choice would be to relax all the atoms in the slab. If we do that, then there
will be no atoms with bulk like spacing unless we increase the slab thickness pretty dramatically. It is
pretty common to freeze some atoms at the bulk coordinates, and let the others relax. We will freeze
the bottom two layers (defined by tags 3 and 4) and let the first two layers relax. To do that we add
constraints to the slab.
Note: the ase constraints are only partially used by jasp. The mod:ase.constraints.FixAtoms constraint gets written to the POSCAR file, and is then used internally in VASP. The only other constraint
that VASP can use internally is mod:ase.constraints.FixScaled. The other constraints are not written to the POSCAR and are not used by VASP.
1
2
3

from jasp import *

from ase.lattice.surface import fcc111
from ase.constraints import FixAtoms

4
5

atoms = fcc111(Al, size=(1, 1, 4), vacuum=10.0)

6
7
8

constraint = FixAtoms(mask=[atom.tag >= 3 for atom in atoms])

atoms.set_constraint(constraint)

9
10
11
12
13
14
15
16
17
18
19

with jasp(surfaces/Al-slab-relaxed,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()
print calc

Open the python script (dft-scripts/script-151.py).

: ----------------------------VASP calculation from /home-research/jkitchin/dft-book2/surfaces/Al-slab-relaxed
converged: True
160

Energy = -14.181417 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.864 0.000 0.000] 2.864
a1 [ 1.432 2.480 0.000] 2.864
a2 [ 0.000 0.000 27.015] 27.015
a,b,c,alpha,beta,gamma (deg):2.864 2.864 27.015 90.0 90.0 90.0
Unit cell volume = 191.872 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.006 0.006 0.001-0.000 -0.000 -0.000
Atom# sym
position [x,y,z]tag rmsForce constraints
0
Al [0.000
0.000
10.000] 4
0.00
F F F
1
Al [1.432
0.827
12.338] 3
0.00
F F F
2
Al [2.864
1.653
14.637] 2
0.05
T T T
3
Al [0.000
0.000
16.982] 1
0.02
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 10
nsw: 10
ibrion: 2
isif: 2
encut: 350.0
magmom: None
kpts: (6, 6, 1)
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------Al: potpaw_PBE/Al/POTCAR (git-hash: c8d9ecb0b6ebec0256c5f5072cee4de6a046dac2)
You can see that atoms 2 and 3 (the ones we relaxed, because the have tags of 1 and 2, which are less
than 3) now have very low forces on them and it appears that atoms 0 and 1 have no forces on them.
That is because the FixAtoms constraint works by setting the forces on those atoms to zero. We can see
in the next example that the z-positions of the relaxed atoms have indeed relaxed and changed, while
the position of the frozen atoms did not change.
Note there are two versions of the forces. The true forces, and the forces when constraints are applied.
1

from jasp import *

2
3
4

with jasp(surfaces/Al-slab-relaxed) as calc:

atoms = calc.get_atoms()

5
6

print Constraints = True: , atoms.get_forces(apply_constraint=True)

7
8

print Constraints = False: , atoms.get_forces(apply_constraint=False)

Open the python script (dft-scripts/script-152.py).

Constraints = True:

[[ 0.

0.

0.

]
161

[ 0.
0.
0.
]
[ 0.
0.
-0.049]
[ 0.
0.
-0.019]]
Constraints = False: [[ 0.
[ 0.
0.
0.069]
[ 0.
0.
-0.049]
[ 0.
0.
-0.019]]
1
2

0.

-0.002]

from jasp import *

from ase.lattice.surface import fcc111

3
4
5
6

with jasp(surfaces/Al-slab-relaxed) as calc:

atoms = calc.get_atoms()
print Total energy: {0:1.3f}.format(atoms.get_potential_energy())

7
8
9
10

for i in range(1, len(atoms)):

print d_({0},{1}) = {2:1.3f} angstroms.format(i, i-1,
atoms[i].z - atoms[i-1].z)

Open the python script (dft-scripts/script-153.py).

Total energy: -14.181
d_(1,0) = 2.338 angstroms
d_(2,1) = 2.299 angstroms
d_(3,2) = 2.345 angstroms
Depending on the layer there is either slight contraction or expansion. These quantities are small,
and careful convergence studies should be performed. Note the total energy change from unrelaxed to
relaxed is not that large in this case (e.g., it is about 5 meV). This is usually the case for metals, where
the relaxation effects are relatively small. In oxides and semiconductors, the effects can be large, and
when there are adsorbates, the effects can be large also.

5.4

Surface reconstruction

We previously considered how relaxation can lower the surface energy. For some surfaces, a more extreme
effect can reduce the surface energy: reconstruction. In a simple surface relaxation, the basic structure
of a surface is preserved. However, sometimes there is a different

surface structure that may have a lower

surface energy. Some famous reconstructions include: Si- 7 7, Pt(100) hex reconstruction, 73,74 and
the Au(111) herringbone reconstruction.
We will consider the (110) missing row reconstruction. 75 For some metals, especially Pt and Au, it
is energetically favorable to form the so-called missing row reconstruction where every other row in the
surface is "missing". It is favorable because it lowers the surface energy. Let us consider how we might
calculate and predict that. It is straightforward to compute the energy of a (110) slab, and of a (110) slab
with one row missing. However, these slabs contain different numbers of atoms, so we cannot directly
compare the total energies to determine which energy is lower.
We have to consider where the missing row atoms have gone, so we can account for their energy. We
will consider that they have gone into the bulk, and so we to consider the energy associated with the
following transformation:
slab110 slabmissing row + bulk
Thus, if this change in energy: Ebulk + Eslabmissingrow Eslab110 is negative, then the formation of the
missing row is expected to be favorable.
5.4.1

Au(110) missing row reconstruction

We first consider the Au(110) case, where the reconstruction is known to be favorable.

162

Clean Au(110) slab

1
2
3
4

from
from
from
from

jasp import *
ase.lattice.surface import fcc110
ase.io import write
ase.constraints import FixAtoms

5
6
7
8

atoms = fcc110(Au, size=(2, 1, 6), vacuum=10.0)

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])
atoms.set_constraint(constraint)

9
10

write(images/Au-110.png, atoms.repeat((2, 2, 1)), rotation=-90x, show_unit_cell=2)

11
12
13
14
15
16
17
18
19
20

with jasp(surfaces/Au-110,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-154.py).

Missing row in Au(110)
1
2
3
4

from
from
from
from

jasp import *
ase.lattice.surface import fcc110
ase.io import write
ase.constraints import FixAtoms

5
6
7

atoms = fcc110(Au, size=(2, 1, 6), vacuum=10.0)

del atoms[11] # delete surface row

8
9
10

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])

atoms.set_constraint(constraint)

11
12
13
14
15

write(images/Au-110-missing-row.png,
atoms.repeat((2, 2, 1)),
rotation=-90x,
show_unit_cell=2)

16
17
18
19
20
21
22
23
24
25

with jasp(surfaces/Au-110-missing-row,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-155.py).

Bulk Au
1
2
3

from jasp import *

from ase.visualize import view
from ase.lattice.cubic import FaceCenteredCubic

4
5
6
7
8
9

atoms = FaceCenteredCubic(directions=[[0, 1, 1],

[1, 0, 1],
[1, 1, 0]],
size=(1, 1, 1),
symbol=Au)

10
11
12
13
14
15
16

with jasp(bulk/Au-fcc,
xc=PBE,
encut=350,
kpts=(12, 12, 12),
atoms=atoms) as calc:
calc.calculate()

163

Open the python script (dft-scripts/script-156.py).

Analysis of energies
1

from jasp import *

2
3
4
5

with jasp(surfaces/Au-110) as calc:

slab = calc.get_atoms()
eslab = slab.get_potential_energy()

6
7
8
9

with jasp(surfaces/Au-110-missing-row) as calc:

missingrow = calc.get_atoms()
emissingrow = missingrow.get_potential_energy()

10
11
12
13

with jasp(bulk/Au-fcc) as calc:

bulk = calc.get_atoms()
ebulk = bulk.get_potential_energy()

14
15
16
17

print natoms slab

= {0}.format(len(slab))
print natoms missing row = {0}.format(len(missingrow))
print natoms bulk
= {0}.format(len(bulk))

18
19

print dE = {0:1.3f} eV.format(emissingrow + ebulk - eslab)

Open the python script (dft-scripts/script-157.py).

natoms slab
= 12
natoms missing row = 11
natoms bulk
= 1
dE = -0.070 eV
The missing row formation energy is slightly negative. The magnitude of the formation energy is
pretty small, but just slightly bigger than the typical convergence errors observed, so we should cautiously
conclude that the reconstruction if favorable for Au(110). We made a lot of shortcuts in computing this
quantity, including using the experimental lattice constant of Au, not checking for convergence in kpoints or planewave cutoff, and not checking for convergence with respect to slab thickness or number
of relaxed layers.
5.4.2

Ag(110) missing row reconstruction

Clean Ag(110) slab

1
2
3
4

from
from
from
from

jasp import *
ase.lattice.surface import fcc110
ase.io import write
ase.constraints import FixAtoms

5
6
7
8

atoms = fcc110(Ag, size=(2, 1, 6), vacuum=10.0)

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])
atoms.set_constraint(constraint)

9
10
11
12
13
14
15
16
17
18

with jasp(surfaces/Ag-110,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-158.py).

164

Missing row in Ag(110)

1
2
3
4

from
from
from
from

jasp import *
ase.lattice.surface import fcc110
ase.io import write
ase.constraints import FixAtoms

5
6
7

atoms = fcc110(Ag, size=(2, 1, 6), vacuum=10.0)

del atoms[11] # delete surface row

8
9
10

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])

atoms.set_constraint(constraint)

11
12
13
14
15
16
17
18
19
20

with jasp(surfaces/Ag-110-missing-row,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-159.py).

Bulk Ag
1
2
3

from jasp import *

from ase.visualize import view
from ase.lattice.cubic import FaceCenteredCubic

4
5
6
7
8
9

atoms = FaceCenteredCubic(directions=[[0, 1, 1],

[1, 0, 1],
[1, 1, 0]],
size=(1, 1, 1),
symbol=Ag)

10
11
12
13
14
15
16

with jasp(bulk/Ag-fcc,
xc=PBE,
encut=350,
kpts=(12, 12, 12),
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-160.py).

Analysis of energies
1

from jasp import *

2
3
4
5

with jasp(surfaces/Ag-110) as calc:

slab = calc.get_atoms()
eslab = slab.get_potential_energy()

6
7
8
9

with jasp(surfaces/Ag-110-missing-row) as calc:

missingrow = calc.get_atoms()
emissingrow = missingrow.get_potential_energy()

10
11
12
13

with jasp(bulk/Ag-fcc) as calc:

bulk = calc.get_atoms()
ebulk = bulk.get_potential_energy()

14
15
16
17

print natoms slab

= {0}.format(len(slab))
print natoms missing row = {0}.format(len(missingrow))
print natoms bulk
= {0}.format(len(bulk))

18
19

print dE = {0:1.3f} eV.format(emissingrow + ebulk - eslab)

Open the python script (dft-scripts/script-161.py).

165

natoms slab
= 12
natoms missing row = 11
natoms bulk
= 1
dE = -0.008 eV
For Ag(110), the missing row formation energy is practically thermoneutral, i.e. not that favorable.
This energy is so close to 0eV, that we cannot confidently say whether the reconstruction is favorable or
not. Experimentally, the reconstruction is not seen on very clean Ag(110) although it is reported that
some adsorbates may induce the reconstruction. 76
5.4.3

Cu(110) missing row reconstruction

Clean Cu(110) slab

1
2
3

from jasp import *

from ase.lattice.surface import fcc110
from ase.constraints import FixAtoms

4
5
6
7

atoms = fcc110(Cu, size=(2, 1, 6), vacuum=10.0)

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])
atoms.set_constraint(constraint)

8
9
10
11
12
13
14
15
16
17

with jasp(surfaces/Cu-110,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-162.py).

Missing row in Cu(110)
1
2
3

from jasp import *

from ase.lattice.surface import fcc110
from ase.constraints import FixAtoms

4
5
6

atoms = fcc110(Cu, size=(2, 1, 6), vacuum=10.0)

del atoms[11] # delete surface row

7
8
9
10
11
12
13
14
15
16
17
18

constraint = FixAtoms(mask=[atom.tag > 2 for atom in atoms])

atoms.set_constraint(constraint)
with jasp(surfaces/Cu-110-missing-row,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
ibrion=2,
isif=2,
nsw=10,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-163.py).

Bulk Cu
1
2
3

from jasp import *

from ase.visualize import view
from ase.lattice.cubic import FaceCenteredCubic

4
5
6
7

atoms = FaceCenteredCubic(directions=[[0, 1, 1],

[1, 0, 1],
[1, 1, 0]],

166

size=(1, 1, 1),
symbol=Cu)

8
9
10
11
12
13
14
15
16

with jasp(bulk/Cu-fcc,
xc=PBE,
encut=350,
kpts=(12, 12, 12),
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-164.py).

Analysis
1

from jasp import *

2
3
4
5

with jasp(surfaces/Cu-110) as calc:

slab = calc.get_atoms()
eslab = slab.get_potential_energy()

6
7
8
9

with jasp(surfaces/Cu-110-missing-row) as calc:

missingrow = calc.get_atoms()
emissingrow = missingrow.get_potential_energy()

10
11
12
13

with jasp(bulk/Cu-fcc) as calc:

bulk = calc.get_atoms()
ebulk = bulk.get_potential_energy()

14
15
16
17

print natoms slab

= {0}.format(len(slab))
print natoms missing row = {0}.format(len(missingrow))
print natoms bulk
= {0}.format(len(bulk))

18
19

print dE = {0:1.3f} eV.format(emissingrow + ebulk - eslab)

Open the python script (dft-scripts/script-165.py).

natoms slab
= 12
natoms missing row = 11
natoms bulk
= 1
dE = -0.011 eV
It is questionable whether we should consider this evidence of a missing row reconstruction because
the number is small. That does not mean the reconstruction will not happen, but it could mean it is
very easy to lift.

5.5

Work function

To get the work function, we need to have the local potential. This is not written by default in VASP,
and we have to tell it to do that with the LVTOT and LVHAR keywords.
1
2

from jasp import *

import matplotlib.pyplot as plt

3
4
5

with jasp(surfaces/Al-slab-relaxed) as calc:

atoms = calc.get_atoms()

6
7
8
9
10
11
12
13

with jasp(surfaces/Al-slab-locpot,
xc=PBE,
kpts=(6, 6, 1),
encut=350,
lvtot=True, # write out local potential
lvhar=True, # write out only electrostatic potential, not xc pot
atoms=atoms) as calc:

14
15
16

calc.calculate()
ef = calc.get_fermi_level()

17

167

18

atoms = calc.get_atoms()

19
20

x, y, z, lp = calc.get_local_potential()

21
22

nx, ny, nz = lp.shape

23
24
25
26
27

axy = np.array([np.average(lp[:, :, z]) for z in range(nz)])

# setup the x-axis in realspace
uc = atoms.get_cell()
xaxis = np.linspace(0, uc[2][2], nz)

28
29
30
31
32
33

plt.plot(xaxis, axy)
plt.plot([min(xaxis), max(xaxis)], [ef, ef], k:)
plt.xlabel(Position along z-axis)
plt.ylabel(x-y averaged electrostatic potential)
plt.savefig(images/Al-wf.png)

34
35
36
37

ind = (xaxis > 0) & (xaxis < 5)

wf = np.average(axy[ind]) - ef
print The workfunction is {0:1.2f} eV.format(wf)

Open the python script (dft-scripts/script-166.py).

The workfunction is 4.04 eV
The workfunction of Al is listed as 4.08 at http://hyperphysics.phy-astr.gsu.edu/hbase/tables/
photoelec.html.

5.6

Surface energy

The easiest way to calculate surface energies is from this equation:

Nslab
Ebulk )
= 12 (Eslab N
bulk
where Eslab is the total energy of a symmetric slab (i.e. one with inversion symmetry, and where both
sides of the slab have been relaxed), Ebulk is the total energy of a bulk unit cell, Nslab is the number of
atoms in the slab, and Nbulk is the number of atoms in the bulk unit cell. One should be sure that the
bulk energy is fully converged with respect to k-points, and that the slab energy is also converged with
respect to k-points. The energies should be compared at the same cutoff energies. The idea is then to
increase the thickness of the slab until the surface energy converges.
Unfortunately, this approach does not always work. The bulk system is treated subtly different than
the slab system, particularly in the z-direction where the vacuum is (where typically only one k-point is
used in slabs). Consequently, the k-point sampling is not equivalent in the two systems, and one can in
general expect some errors due to this, with the best case being cancellation of the errors due to total
k-point convergence. In the worst case, one can get a linear divergence in the surface energy with slab
thickness. 77
A variation of this method that usually results in better k-point error cancellation is to calculate the
bulk unit cell energy using the slab unit cell with no vacuum space, with the same k-point mesh in the x
and y directions, but with increased k-points in the z-direction. Thus, the bulk system and slab system
have the same Brillouin zone in at least two dimensions. This maximizes the cancellation of k-point
errors, but still does not guarantee convergence of the surface energy, as discussed in. 77,78
For quick estimates of the surface energy, one of the methods described above is likely sufficient. The
advantage of these methods is the small number of calculations required to obtain the estimate, one
needs only a bulk calculation (which must be done anyhow to get the bulk lattice constant to create the
slab), and a slab calculation that is sufficiently thick to get the estimate. Additional calculations are
only required to test the convergence of the surface energy.
An alternative method for calculating surface energies that does not involve an explicit bulk calculation follows Ref. 78 The method follows from equation (ref{eq:se}) where for a N-atom slab, in the limit
of N ,
Nslab
Eslab 2 + N
Ebulk
bulk
Then, we can estimate Ebulk by plotting the total energy of the slab as a function of the slab thickness.
168

N
= limN 12 (Eslab
N EN )
N 1
N
where EN = Eslab Eslab
.
We will examine this approach here. We will use unrelaxed slabs for computational efficiency.

1
2
3

from jasp import *

from ase.lattice.surface import fcc111
import matplotlib.pyplot as plt

4
5
6
7

Nlayers = [3, 4, 5, 6, 7, 8, 9, 10, 11]

energies = []
sigmas = []

8
9

for n in Nlayers:

10
11
12

slab = fcc111(Cu, size=(1, 1, n), vacuum=10.0)

slab.center()

13
14
15
16
17
18
19
20
21
22
23

with jasp(bulk/Cu-layers/{0}.format(n),
xc=PBE,
encut=350,
kpts=(8, 8, 1),
atoms=slab) as calc:
calc.set_nbands(f=2) # the default nbands in VASP is too low for Al
try:
energies.append(slab.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

24
25
26
27
28
29
30

for i in range(len(Nlayers) - 1):

N = Nlayers[i]
DeltaE_N = energies[i + 1] - energies[i]
sigma = 0.5 * (-N * energies[i + 1] + (N + 1) * energies[i])
sigmas.append(sigma)
print nlayers = {1:2d} sigma = {0:1.3f} eV/atom.format(sigma, N)

31
32
33
34
35

plt.plot(Nlayers[0:-1], sigmas, bo-)

plt.xlabel(Number of layers)
plt.ylabel(Surface energy (eV/atom))
plt.savefig(images/Cu-unrelaxed-surface-energy.png)

Open the python script (dft-scripts/script-167.py).

nlayers
nlayers
nlayers
nlayers
nlayers
nlayers
nlayers
nlayers

= 3
= 4
= 5
= 6
= 7
= 8
= 9
= 10

sigma
sigma
sigma
sigma
sigma
sigma
sigma
sigma

=
=
=
=
=
=
=
=

0.551
0.399
0.590
0.319
0.578
0.322
0.592
0.403

eV/atom
eV/atom
eV/atom
eV/atom
eV/atom
eV/atom
eV/atom
eV/atom

One reason for the oscillations may be quantum size effects. 79 In 80 the surface energy of Cu(111) is
reported as 0.48 eV/atom, or 1.36 J/m2 . Here is an example showing a conversion between these two
units. We use ase to compute the area of the unit cell from the norm of the cross-product of the vectors
defining the surface unit cell.
1
2
3

from ase.lattice.surface import fcc111

from ase.units import J, m
import numpy as np

4
5
6

slab = fcc111(Cu, size=(1, 1, 3), vacuum=10.0)

cell = slab.get_cell()

7
8

area = np.linalg.norm(np.cross(cell[0], cell[1]))

9
10

sigma = 0.48

# eV/atom

11
12

print sigma = {0} J/m^2.format(sigma / area / (J / m**2))

169

Open the python script (dft-scripts/script-168.py).

sigma = 1.3628146074 J/m^2
5.6.1

Advanced topics in surface energy

The surface energies can be used to estimate the shapes of nanoparticles using a Wulff construction.
See 81 for an example of computing Mo2 C surface energies and particle shapes, and 82 for an example of
the influence of adsorbates on surface energies and particle shapes of Cu.
For a classic paper on trends in surface energies see. 83

5.7

Dipole correction

A subtle problem can arise when an adsorbate is placed on one side of a slab with periodic boundary
conditions, which is currently the common practice. The problem is that this gives the slab a dipole
moment. The array of dipole moments created by the periodic boundary conditions generates an electric
field that can distort the electron density of the slab and change the energy. The existence of this field
in the vacuum also makes the zero-potential in the vacuum ill-defined, thus the work function is not
well-defined. One solution to this problem is to use slabs with adsorbates on both sides, but then very
thick (eight to ten layers) slabs must be used to ensure the adsorbates do not interact through the slab.
An alternative solution, the dipole correction scheme, was developed by Neugebauer and Scheffler 84 and
later corrected by Bengtsson. 85 In this technique, an external field is imposed in the vacuum region that
exactly cancels the artificial field caused by the slab dipole moment. The advantage of this approach is
that thinner slabs with adsorbates on only one side can be used.
There are also literature reports that the correction is small. 86 Nevertheless, in the literature the use
of this correction is fairly standard, and it is typical to at least consider the correction.
Here we will just illustrate the effect.
5.7.1

Slab with no dipole correction

We simply run the calculation here, and compare the results later.
1
2
3
4
5

# compute local potential of slab with no dipole

from ase.lattice.surface import fcc111, add_adsorbate
from jasp import *
import matplotlib.pyplot as plt
from ase.io import write

6
7
8

slab = fcc111(Al, size=(2, 2, 2), vacuum=10.0)

add_adsorbate(slab, Na, height=1.2, position=fcc)

9
10
11

slab.center()
write(images/Na-Al-slab.png, slab, rotation=-90x, show_unit_cell=2)

12
13
14
15
16
17
18
19
20

with jasp(surfaces/Al-Na-nodip,
xc=PBE,
encut=340,
kpts=(2, 2, 1),
lvtot=True, # write out local potential
lvhar=True, # write out only electrostatic potential, not xc pot
atoms=slab) as calc:
calc.calculate()

Open the python script (dft-scripts/script-169.py).

5.7.2

Slab with a dipole correction

Note this takes a considerably longer time to run than without a dipole correction! In VASP there are
several levels of dipole correction to apply. You can use the IDIPOL tag to turn it on, and specify which
direction to apply it in (1=x, 2=y, 3=z, 4=(x, y, z)). This simply corrects the total energy and forces. It
170

does not change the contents of LOCPOT. For that, you have to also set the LDIPOL and DIPOL tags. It
is not efficient to set all three at the same time for some reason. The VASP manual recommends you first
set IDIPOL to get a converged electronic structure, and then set LDIPOL to True, and set the center
of electron density in DIPOL. That makes these calculations a multistep process, because we must run
a calculation, analyze the charge density to get the center of charge, and then run a second calculation.
1
2
3

# compute local potential with dipole calculation on

from ase.lattice.surface import fcc111, add_adsorbate
from jasp import *

4
5
6

slab = fcc111(Al, size=(2, 2, 2), vacuum=10.0)

add_adsorbate(slab, Na, height=1.2, position=fcc)

7
8

slab.center()

9
10
11
12
13
14
15
16
17

with jasp(surfaces/Al-Na-dip,
xc=PBE,
encut=340,
kpts=(2, 2, 1),
idipol=3,
# only along z-axis
lvtot=True, # write out local potential
lvhar=True, # write out only electrostatic potential, not xc pot
atoms=slab) as calc:

18
19

calc.calculate()

20
21
22
23
24
25

x, y, z, cd = calc.get_charge_density()
n0, n1, n2 = cd.shape
nelements = n0 * n1 * n2
voxel_volume = slab.get_volume() / nelements
total_electron_charge = cd.sum() * voxel_volume

26
27
28
29
30
31

electron_density_center = np.array([(cd * x).sum(),

(cd * y).sum(),
(cd * z).sum()])
electron_density_center *= voxel_volume
electron_density_center /= total_electron_charge

32
33
34

print electron-density center = {0}.format(electron_density_center)

uc = slab.get_cell()

35
36
37

# get scaled electron charge density center

sedc = np.dot(np.linalg.inv(uc.T), electron_density_center.T).T

38
39
40

# we only write 4 decimal places out to the INCAR file, so we round here.
sedc = np.round(sedc, 4)

41
42

calc.clone(surfaces/Al-Na-dip-step2)

43
44
45
46
47

# now run step 2 with dipole set at scaled electron charge density center
with jasp(surfaces/Al-Na-dip-step2,
ldipol=True, dipol=sedc) as calc:
calc.calculate()

Open the python script (dft-scripts/script-170.py).

electron-density center = [
5.7.3

4.212

2.419

10.657]

Comparing no dipole correction with a dipole correction

To see the difference in what the dipole correction does, we now plot the potentials from each calculation.
1
2

from jasp import *

import matplotlib.pyplot as plt

3
4
5

with jasp(surfaces/Al-Na-nodip) as calc:

atoms = calc.get_atoms()

6
7
8

x, y, z, lp = calc.get_local_potential()
nx, ny, nz = lp.shape

171

10
11
12
13

axy_1 = [np.average(lp[:, :, z]) for z in range(nz)]

# setup the x-axis in realspace
uc = atoms.get_cell()
xaxis_1 = np.linspace(0, uc[2][2], nz)

14
15

e1 = atoms.get_potential_energy()

16
17
18

with jasp(surfaces/Al-Na-dip-step2) as calc:

atoms = calc.get_atoms()

19
20
21

x, y, z, lp = calc.get_local_potential()
nx, ny, nz = lp.shape

22
23
24
25
26

axy_2 = [np.average(lp[:, :, z]) for z in range(nz)]

# setup the x-axis in realspace
uc = atoms.get_cell()
xaxis_2 = np.linspace(0, uc[2][2], nz)

27
28
29

ef2 = calc.get_fermi_level()
e2 = atoms.get_potential_energy()

30
31

print The difference in energy is {0} eV..format(e2-e1)

32
33
34
35
36
37
38
39

plt.plot(xaxis_1, axy_1, label=no dipole correction)

plt.plot(xaxis_2, axy_2, label=dipole correction)
plt.plot([min(xaxis_2), max(xaxis_2)], [ef2, ef2], k:, label=Fermi level)
plt.xlabel(z ($\AA$))
plt.ylabel(xy-averaged electrostatic potential)
plt.legend(loc=best)
plt.savefig(images/dip-vs-nodip-esp.png)

Open the python script (dft-scripts/script-171.py).

The difference in energy is 3.867888 eV.
The key points to notice in this figure are:
1. The two deep dips are where the atoms are.
2. Without a dipole correction, the electrostatic potential never flattens out. there is near constant
slope in the vacuum region, which means there is an electric field there.
3. With a dipole correction the potential is flat in the vacuum region, except for the step jump near
23 .
4. The difference between the Fermi level and the flat vacuum potential is the work function.
5. The difference in energy with and without the dipole correction here is small.

5.8
5.8.1

Adsorption energies
Simple estimate of the adsorption energy

Calculating an adsorption energy amounts to computing the energy of the following kind of reaction:
slab + gas-phase molecule slab_adsorbate + products
There are many variations of this idea. The slab may already have some adsorbates on it, the slab
may reconstruct on adsorption, the gas-phase molecule may or may not dissociate, and the products may
or may not stick to the surface. We have to decide where to put the adsorbates, i.e. what site to put
them on, and some sites will be more stable than others. We will consider the dissociative adsorption of
O2 on three sites of a Pt(111) slab. We will assume the oxygen molecule has split in half, and that the
atoms have moved far apart. We will model the oxygen coverage at 0.25 ML, which means we need to
use a 2 2 surface unit cell. For computational speed, we will freeze the slab, but allow the adsorbate
to relax.
Hads (eV /O) = Eslab+O Eslab 0.5 EO2

172

Calculations
clean slab calculation
1
2
3

from jasp import *

from ase.lattice.surface import fcc111
from ase.constraints import FixAtoms

4
5
6
7

atoms = fcc111(Pt, size=(2, 2, 3), vacuum=10.0)

constraint = FixAtoms(mask=[True for atom in atoms])
atoms.set_constraint(constraint)

8
9
10

from ase.io import write

write(images/Pt-fcc-ori.png, atoms, show_unit_cell=2)

11
12
13
14
15
16
17

with jasp(surfaces/Pt-slab,
xc=PBE,
kpts=(4, 4, 1),
encut=350,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-172.py).

-67.711178

fcc site
1

from jasp import *

2
3
4

from ase.lattice.surface import fcc111, add_adsorbate

from ase.constraints import FixAtoms

5
6

atoms = fcc111(Pt, size=(2, 2, 3), vacuum=10.0)

7
8
9

# note this function only works when atoms are created by the surface module.
add_adsorbate(atoms, O, height=1.2, position=fcc)

10
11
12

constraint = FixAtoms(mask=[atom.symbol != O for atom in atoms])

atoms.set_constraint(constraint)

13
14
15

from ase.io import write

write(images/Pt-fcc-site.png, atoms, show_unit_cell=2)

16
17
18
19
20
21
22
23

with jasp(surfaces/Pt-slab-O-fcc,
xc=PBE,
kpts=[4, 4, 1],
encut=350,
ibrion=2,
nsw=25,
atoms=atoms) as calc:

24
25

print atoms.get_potential_energy()

Open the python script (dft-scripts/script-173.py).

-73.663924

173

O atom on the bridge site

1
2
3

from jasp import *

from ase.lattice.surface import fcc111, add_adsorbate
from ase.constraints import FixAtoms

4
5

atoms = fcc111(Pt, size=(2, 2, 3), vacuum=10.0)

6
7
8

# note this function only works when atoms are created by the surface module.
add_adsorbate(atoms, O, height=1.2, position=bridge)

9
10
11

constraint = FixAtoms(mask=[atom.symbol != O for atom in atoms])

atoms.set_constraint(constraint)

12
13
14
15
16
17
18
19

with jasp(surfaces/Pt-slab-O-bridge,
xc=PBE,
kpts=(4, 4, 1),
encut=350,
ibrion=2,
nsw=25,
atoms=atoms) as calc:

20
21
22

calc.calculate()
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-174.py).

-73.663615

174

Figure 72: Initial geometry of the bridge site. It is definitely on the bridge.
hcp site
1
2
3

from jasp import *

from ase.lattice.surface import fcc111, add_adsorbate
from ase.constraints import FixAtoms

4
5

atoms = fcc111(Pt, size=(2, 2, 3), vacuum=10.0)

6
7
8

# note this function only works when atoms are created by the surface module.
add_adsorbate(atoms, O, height=1.2, position=hcp)

9
10
11

constraint = FixAtoms(mask=[atom.symbol != O for atom in atoms])

atoms.set_constraint(constraint)

12
13
14

from ase.io import write

write(images/Pt-hcp-o-site.png, atoms, show_unit_cell=2)

15
16
17
18
19
20
21
22
23

with jasp(surfaces/Pt-slab-O-hcp,
xc=PBE,
kpts=(4, 4, 1),
encut=350,
ibrion=2,
nsw=25,
atoms=atoms) as calc:
calc.calculate()

Open the python script (dft-scripts/script-175.py).

175

Analysis of adsorption energies

1
2

from jasp import *

from ase.io import write

3
4
5
6
7

with jasp(surfaces/Pt-slab) as calc:

atoms = calc.get_atoms()
e_slab = atoms.get_potential_energy()
write(images/pt-slab.png, atoms,show_unit_cell=2)

8
9
10
11
12

with jasp(surfaces/Pt-slab-O-fcc) as calc:

atoms = calc.get_atoms()
e_slab_o_fcc = atoms.get_potential_energy()
write(images/pt-slab-fcc-o.png, atoms,show_unit_cell=2)

13
14
15
16
17

with jasp(surfaces/Pt-slab-O-hcp) as calc:

atoms = calc.get_atoms()
e_slab_o_hcp = atoms.get_potential_energy()
write(images/pt-slab-hcp-o.png, atoms,show_unit_cell=2)

18
19
20
21
22

with jasp(surfaces/Pt-slab-O-bridge) as calc:

atoms = calc.get_atoms()
e_slab_o_bridge = atoms.get_potential_energy()
write(images/pt-slab-bridge-o.png, atoms,show_unit_cell=2)

23
24
25
26

with jasp(molecules/O2-sp-triplet-350) as calc:

atoms = calc.get_atoms()
e_O2 = atoms.get_potential_energy()

27
28
29
30

Hads_fcc = e_slab_o_fcc - e_slab - 0.5 * e_O2

Hads_hcp = e_slab_o_hcp - e_slab - 0.5 * e_O2
Hads_bridge = e_slab_o_bridge - e_slab - 0.5 * e_O2

31
32
33
34

print Hads (fcc)

= {0} eV/O.format(Hads_fcc)
print Hads (hcp)
= {0} eV/O.format(Hads_hcp)
print Hads (bridge) = {0} eV/O.format(Hads_bridge)

Open the python script (dft-scripts/script-176.py).

Hads (fcc)

= -1.0384925 eV/O
176

Hads (hcp)
= -0.5986145 eV/O
Hads (bridge) = -1.0381835 eV/O
You can see the hcp site is not as energetically favorable as the fcc site. Interestingly, the bridge
site seems to be as favorable as the fcc site. This is not correct, and to see why, we have to look at the
final geometries of each calculation. First the fcc (Figure 73 and hcp (Figure 74 sites, which look like we
expect.

Figure 73: Final geometry of the fcc site.

Figure 74: Final geometry of the hcp site.

The bridge site (Figure 75, however, is clearly not at a bridge site!

177

Figure 75: Final geometry of the bridge site. You can see that the oxygen atom ended up in the fcc site.

Let us see what the original geometry and final geometry for the bridge site were. The POSCAR contains the initial geometry (as long as you havent copied CONTCAR to POSCAR), and the CONTCAR
contains the final geometry.
1

from ase.io import read, write

2
3
4

atoms = read(surfaces/Pt-slab-O-bridge/POSCAR)
write(images/Pt-o-brige-ori.png, atoms, show_unit_cell=2)

5
6
7

atoms = read(surfaces/Pt-slab-O-bridge/CONTCAR)
write(images/Pt-o-brige-final.png, atoms, show_unit_cell=2)

Open the python script (dft-scripts/script-177.py).

Figure 76: Initial geometry of the bridge site. It is definitely on the bridge.

178

Figure 77: Final geometry of the bridge site. It has fallen into the fcc site.
You can see the problem. We should not call the adsorption energy from this calculation a bridge
site adsorption energy because the O atom is actually in an fcc site! This kind of result can happen with
relaxation, and you should always check that the result you get makes sense. Next, we consider how to
get a bridge site adsorption energy by using constraints.
Some final notes:
1. We did not let the slabs relax in these examples, and allowing them to relax is likely to have a
big effect on the adsorption energies. You have to decide how many layers to relax, and check for
convergence with respect to the number of layers.
2. The slabs were pretty thin. It is typical these days to see slabs that are 4-5 or more layers thick.
3. We did not consider how well converged the calculations were with respect to k-points or incar:ENCUT.
4. We did not consider the effect of the error in O2 dissociation energy on the adsorption energies.
5. We did not consider coverage effects (see Coverage dependence).
Adsorption on bridge site with constraints To prevent the oxygen atom from sliding down into
the fcc site, we have to constrain it so that it only moves in the z-direction. This is an artificial constraint;
the bridge site is only metastable. But there are lots of reasons you might want to do this anyway. One is
the bridge site is a transition state for diffusion between the fcc and hcp sites. Another is to understand
the role of coordination in the adsorption energies. We use a func:ase.constraints.FixScaled constraint
in ase to constrain the O atom so it can only move in the z-direction (actually so it can only move in
the direction of the third unit cell vector, which only has a z-component).
1

from jasp import *

2
3
4
5

from ase.lattice.surface import fcc111, add_adsorbate

from ase.constraints import FixAtoms, FixScaled
from ase.io import write

6
7

atoms = fcc111(Pt, size=(2, 2, 3), vacuum=10.0)

8
9
10
11
12
13
14

# note this function only works when atoms are created by the surface module.
add_adsorbate(atoms, O, height=1.2, position=bridge)
constraint1 = FixAtoms(mask=[atom.symbol != O for atom in atoms])
# fix in xy-direction, free in z. actually, freeze movement in surface
# unit cell, and free along 3rd lattice vector
constraint2 = FixScaled(atoms.get_cell(), 12, [True, True, False])

15
16
17
18

atoms.set_constraint([constraint1, constraint2])
write(images/Pt-O-bridge-constrained-initial.png, atoms, show_unit_cell=2)
print Initial O position: {0}.format(atoms.positions[-1])

179

19
20
21
22
23
24
25
26
27

with jasp(surfaces/Pt-slab-O-bridge-xy-constrained,
xc=PBE,
kpts=(4, 4, 1),
encut=350,
ibrion=2,
nsw=25,
atoms=atoms) as calc:
e_bridge = atoms.get_potential_energy()

28
29
30

write(images/Pt-O-bridge-constrained-final.png, atoms, show_unit_cell=2)

print Final O position : {0}.format(atoms.positions[-1])

31
32
33
34
35

# now compute Hads

with jasp(surfaces/Pt-slab) as calc:
atoms = calc.get_atoms()
e_slab = atoms.get_potential_energy()

36
37
38
39
40

with jasp(molecules/O2-sp-triplet-350) as calc:

atoms = calc.get_atoms()
e_O2 = atoms.get_potential_energy()

41
42

Hads_bridge = e_bridge - e_slab - 0.5*e_O2

43
44

print Hads (bridge) = {0:1.3f} eV/O.format(Hads_bridge)

Open the python script (dft-scripts/script-178.py).

Initial O position: [ 1.386
Final O position : [ 1.386
Hads (bridge) = -0.488 eV/O

0.
0.

15.726]
15.977]

You can see that only the z-position of the O atom changed. Also, the adsorption energy of O on the
bridge site is much less favorable than on the fcc or hcp sites.

Figure 78: Initial state of the O atom on the bridge site.

180

Figure 79: Final state of the constrained O atom, still on the bridge site.
5.8.2

Coverage dependence

The adsorbates on the surface can interact with each other which results in coverage dependent adsorption
energies. 87 Coverage dependence is not difficult to model; we simply compute adsorption energies in
different size unit cells, and/or with different adsorbate configurations. Here we consider dissociative
oxygen adsorption at 1ML on Pt(111) in an fcc site, which is one oxygen atom in a 1 1 unit cell.
For additional reading, see these references from our work:
Correlations of coverage dependence of oxygen adsorption on different metals 88,89
Coverage effects of atomic adsorbates on Pd(111) 90
Simple model for estimating coverage dependence 87
Coverage effects on alloys 91
clean slab calculation
1
2
3

from jasp import *

from ase.lattice.surface import fcc111
from ase.constraints import FixAtoms

4
5
6
7

atoms = fcc111(Pt, size=(1, 1, 3), vacuum=10.0)

constraint = FixAtoms(mask=[True for atom in atoms])
atoms.set_constraint(constraint)

8
9

write(images/Pt-fcc-1ML.png, atoms, show_unit_cell=2)

10
11
12
13
14
15
16
17

with jasp(surfaces/Pt-slab-1x1,
xc=PBE,
kpts=(8, 8, 1),
encut=350,
atoms=atoms) as calc:
slab_e = atoms.get_potential_energy()
print slab_e

Open the python script (dft-scripts/script-179.py).

-16.927696

181

fcc site at 1 ML coverage

1

from jasp import *

2
3
4
5

from ase.lattice.surface import fcc111, add_adsorbate

from ase.constraints import FixAtoms
from ase.io import write

6
7

atoms = fcc111(Pt, size=(1, 1, 3), vacuum=10.0)

8
9
10

# note this function only works when atoms are created by the surface module.
add_adsorbate(atoms, O, height=1.2, position=fcc)

11
12
13

constraint = FixAtoms(mask=[atom.symbol != O for atom in atoms])

atoms.set_constraint(constraint)

14
15
16

write(images/Pt-o-fcc-1ML.png, atoms, show_unit_cell=2)

17
18
19
20
21
22
23
24
25
26

with jasp(surfaces/Pt-slab-1x1-O-fcc,
xc=PBE,
kpts=(8, 8, 1),
encut=350,
ibrion=2,
nsw=25,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-180.py).

-21.941003

182

Adsorption energy at 1ML

1

from jasp import *

2
3
4
5

with jasp(surfaces/Pt-slab-1x1-O-fcc) as calc:

atoms = calc.get_atoms()
e_slab_o = atoms.get_potential_energy()

6
7
8
9
10

# clean slab
with jasp(surfaces/Pt-slab-1x1) as calc:
atoms = calc.get_atoms()
e_slab = atoms.get_potential_energy()

11
12
13
14

with jasp(molecules/O2-sp-triplet-350) as calc:

atoms = calc.get_atoms()
e_O2 = atoms.get_potential_energy()

15
16
17

hads = e_slab_o - e_slab - 0.5 * e_O2

print Hads (1ML) = {0:1.3f} eV.format(hads)

Open the python script (dft-scripts/script-181.py).

Hads (1ML) = -0.099 eV
The adsorption energy is much less favorable at 1ML coverage than at 0.25 ML coverage! We will
return what this means in Atomistic thermodynamics effect on adsorption.
5.8.3

Effect of adsorption on the surface energy

There is a small point to make here about what adsorption does to surface energies. Let us define a
general surface formation energy scheme like this:
183

Let us presume the surfaces are symmetric, and that each surface contributes half of the energy
change. The overall change in energy:
E = Eslab,ads Eads Ebulk
where the the energies are appropriately normalized for the stoichiometry. Let us rearrange the terms,
and add and subtract a constant term Eslab .
E = Eslab,ads Eslab Eads Ebulk + Eslab
1
(Eslab Ebulk ), and we defined Hads = Eslab,ads Eslab Eads for adsorption
We defined clean = 2A
on a single side of a slab. In this case, there are adsorbates on both sides of the slab, so Eslab,ads
Eslab Eads = 2Hads . If we normalize by 2A, the area for both sides of the slab, we get
Hads
E
2A = = clean + A
You can see here that the adsorption energy serves to stabilize, or reduce the surface energy, provided
that the adsorption energy is negative.
Some final notes about the equations above:
We were not careful about stoichiometry. As written, it is assumed there are the same number
of atoms (not including the adsorbates) in the slabs and bulk, and the same number of adsorbate
atoms in the slab and Eads . Appropriate normalization factors must be included if that is not true.
It is not necessary to perform a symmetric slab calculation to determine the effect of adsorption
on the surface energy! You can examine clean with knowledge of only the adsorption energies!

5.9

Adsorbate vibrations

Adsorbates also have vibrational modes. Unlike a free molecule, the translational and rotational modes
of an adsorbate may actually have real frequencies. Sometimes they are called frustrated translations or
rotations. For metal surfaces with adsorbates, it is common to only compute vibrational modes of the
adsorbate on a frozen metal slab. The rationale is that the metal atoms are so much heavier than the
adsorbate that there will be little coupling between the surface and adsorbates. You can limit the number
of modes calculated with constraints (func:ase.constraints.FixAtoms or func:ase.constraints.FixScaled) if
you use IBRION=5. The other IBRION settings (6, 7, 8) do not respect the selective dynamics constraints.
Below we consider the vibrational modes of an oxygen atom in an fcc site on Pt(111).
1

from jasp import *

2
3
4

with jasp(surfaces/Pt-slab-O-fcc) as calc:

calc.clone(surfaces/Pt-slab-O-fcc-vib)

5
6
7
8
9
10
11
12
13
14
15

with jasp(surfaces/Pt-slab-O-fcc-vib) as calc:

calc.set(ibrion=5,
# finite differences with selective dynamics
nfree=2,
# central differences (default)
potim=0.015, # default as well
ediff=1e-8,
nsw=1)
atoms = calc.get_atoms()
f, v = calc.get_vibrational_modes(0)
print Elapsed time = {0} seconds.format(calc.get_elapsed_time())
allfreq = calc.get_vibrational_modes()[0]

16
17
18
19

from ase.units import meV

c = 3e10 # cm/s
h = 4.135667516e-15 # eV*s

20
21
22
23
24
25
26

print
print
print
print
print
print

vibrational
vibrational
vibrational
vibrational

energy
energy
freq
freq

=
=
=
=

{0}
{0}
{0}
{0}

eV.format(f)
meV.format(f/meV)
1/s.format(f/h)
cm^{{-1}}.format(f/(h*c))

All energies = , allfreq

Open the python script (dft-scripts/script-182.py).

184

Elapsed time = 14269.345 seconds

vibrational energy = 0.061606647 eV
vibrational energy = 61.606647 meV
vibrational freq
= 1.48964216204e+13 1/s
vibrational freq
= 496.547387346 cm^{-1}
All energies =

[0.061606647, 0.044346362, 0.044344470000000004]

There are three modes for the free oxygen atom. One of them is a mode normal to the surface (the
one with highest frequency. The other two are called frustrated translations. Note that we did not
include the surface Pt atoms in the calculation, and this will have an effect on the result because the O
atom could be coupled to the surface modes. It is typical to neglect this coupling because of the large
difference in mass between O and Pt. Next we look at the difference in results when we calculate all the
modes.
1

from jasp import *

2
3
4

with jasp(surfaces/Pt-slab-O-fcc) as calc:

calc.clone(Pt-slab-O-fcc-vib-ibrion=6)

5
6
7
8
9
10
11
12
13

with jasp(surfaces/Pt-slab-O-fcc-vib-ibrion=6) as calc:

calc.set(ibrion=6, # finite differences with symmetry
nfree=2, # central differences (default)
potim=0.015, # default as well
ediff=1e-8,
nsw=1)
atoms = calc.get_atoms()
print Elapsed time = {0} seconds.format(calc.get_elapsed_time())

14
15
16

f, m = calc.get_vibrational_modes(0)
allfreq = calc.get_vibrational_modes()[0]

17
18
19
20

from ase.units import meV

c = 3e10 # cm/s
h = 4.135667516e-15 # eV*s

21
22
23
24
25
26
27
28

print
print
print
print
print
print
print

For mode 0:
vibrational energy
vibrational energy
vibrational freq
vibrational freq

=
=
=
=

{0}
{0}
{0}
{0}

eV.format(f)
meV.format(f / meV)
1/s.format(f / h)
cm^{{-1}}.format(f / (h * c))

All energies = , allfreq

Open the python script (dft-scripts/script-183.py).

Elapsed time = 77121.015 seconds
For mode 0:
vibrational energy = 0.063537929 eV
vibrational energy = 63.537929 meV
vibrational freq
= 1.53634035507e+13 1/s
vibrational freq
= 512.113451691 cm^{-1}
All energies =

[0.06353792899999999, 0.045628623, 0.045628623, 0.023701702, 0.023701702, 0.023223747

Note that now there are 39 modes, which is 3*N where N=13 atoms in the unit cell. Many of
the modes are low in frequency, which correspond to slab modes that are essentially phonons. The O
frequencies are not that different from the previous calculation (497 vs 512 cm1 . This is why it is
common to keep the slab atoms frozen.
Calculating these results took 39*2 finite differences. It took about a day to get these results on a
single CPU. It pays to use constraints to minimize the number of these calculations.

185

5.9.1

Vibrations of the bridge site

Here we consider the vibrations of an O atom in a bridge site, which we saw earlier is a metastable saddle
point.
1
2

from jasp import *

from ase.constraints import FixAtoms

3
4
5
6

# clone calculation so we do not overwrite previous results

with jasp(surfaces/Pt-slab-O-bridge-xy-constrained) as calc:
calc.clone(surfaces/Pt-slab-O-bridge-vib)

7
8
9
10
11
12
13
14
15
16
17

with jasp(surfaces/Pt-slab-O-bridge-vib) as calc:

calc.set(ibrion=5, # finite differences with selective dynamics
nfree=2,
# central differences (default)
potim=0.015, # default as well
ediff=1e-8,
nsw=1)
atoms = calc.get_atoms()
del atoms.constraints
constraint = FixAtoms(mask=[atom.symbol != O for atom in atoms])
atoms.set_constraint([constraint])

18
19
20

f, v = calc.get_vibrational_modes(2)
print calc.get_vibrational_modes()[0]

21
22
23
24

from ase.units import meV

c = 3e10 # cm/s
h = 4.135667516e-15 # eV*s

25
26
27
28
29

print
print
print
print

vibrational
vibrational
vibrational
vibrational

energy
energy
freq
freq

=
=
=
=

{0}
{0}
{0}
{0}

eV.format(f)
meV.format(f/meV)
1/s.format(f/h)
cm^(-1).format(f/(h*c))

Open the python script (dft-scripts/script-184.py).

[0.06691932, 0.047345270999999994, (0.020649715000000003+0j)]
vibrational energy = (0.020649715+0j) eV
vibrational energy = (20.649715+0j) meV
vibrational freq
= (4.99307909065e+12+0j) 1/s
vibrational freq
= (166.435969688+0j) cm^(-1)
Note that we have one imaginary mode. This corresponds to the motion of the O atom falling into
one of the neighboring 3-fold sites. It also indicates this position is not a stable minimum, but rather a
saddle point. This position is a transition state for hopping between the fcc and hcp sites.

5.10

Surface Diffusion barrier

See this review 92 of diffusion on transition metal surfaces.

5.10.1

Standard nudged elastic band method

Here we illustrate a standard NEB method. You need an initial and final state to start with. We will
use the results from previous calculations of oxygen atoms in an fcc and hcp site. then we will construct
a band of images connecting these two sites. Finally, we let VASP optimize the band and analyze the
results to get the barrier.
1
2
3

from jasp import *

from ase.neb import NEB
import matplotlib.pyplot as plt

4
5
6

with jasp(surfaces/Pt-slab-O-fcc) as calc:

initial_atoms = calc.get_atoms()

186

8
9

with jasp(surfaces/Pt-slab-O-hcp) as calc:

final_atoms = calc.get_atoms()

10
11
12
13
14
15
16

# here is our estimated transition state. we use vector geometry to

# define the bridge position, and add 1.451 Ang to z based on our
# previous bridge calculation. The bridge position is half way between
# atoms 9 and 10.
ts = initial_atoms.copy()
ts.positions[-1] = 0.5 * (ts.positions[9] + ts.positions[10]) + [0, 0, 1.451]

17
18
19
20
21

# construct the band

images = [initial_atoms]
images += [initial_atoms.copy()]
images += [ts.copy()] # this is the TS

22
23
24
25

neb = NEB(images)
# Interpolate linearly the positions of these images:
neb.interpolate()

26
27
28
29
30

# now add the second half

images2 = [ts.copy()]
images2 += [ts.copy()]
images2 += [final_atoms]

31
32
33

neb2 = NEB(images2)
neb2.interpolate()

34
35
36

# collect final band. Note we do not repeat the TS in the second half
final_images = images + images2[1:]

37
38
39
40
41
42

with jasp(surfaces/Pt-O-fcc-hcp-neb,
ibrion=1,
nsw=90,
spring=-5,
atoms=final_images) as calc:

43
44

try:
images, energies = calc.get_neb()
p = calc.plot_neb(show=False)

45
46
47
48
49
50
51
52
53

# remember you are in surfaces/Pt-O-fcc-hcp-neb, so to save in

# the images directory you need ../../ in the path to get you
# back up.
plt.savefig(../../images/pt-o-fcc-hcp-neb.png)
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-185.py).

Optimization terminated successfully.
Current function value: -0.575224
Iterations: 12
Function evaluations: 24
We should compare this barrier to what we could estimate from the simple adsorption energies in
the fcc and bridge sites. The adsorption energy in the fcc site was -1.04 eV, and in the bridge site was
-0.49 eV. The difference between these two is 0.55 eV, which is very close to the calculated barrier from
the NEB calculation. In cases where you can determine what the transition state is, e.g. by symmetry,
or other means, it is much faster to directly compute the energy of the initial and transition states for
barrier determinations. This is not usually possible though.
5.10.2

Climbing image NEB

One issue with the standard NEB method is there is no image that is exactly at the transition state.
That means there is some uncertainty of the true energy of the transition state, and there is no way
to verify the transition state by vibrational analysis. The climbing image NEB method 93 solves that
problem by making one image climb to the top. You set LCLIMB==True= in jasp to turn on the climbing
image method. Here we use the previous calculation as a starting point and turn on the climbing image
method.
187

1
2
3
4

# perform a climbing image NEB calculation

from jasp import *
with jasp(surfaces/Pt-O-fcc-hcp-neb) as calc:
calc.clone(surfaces/Pt-O-fcc-hcp-cineb)

5
6
7
8

with jasp(surfaces/Pt-O-fcc-hcp-cineb,
debug=logging.DEBUG) as calc:
calc.set(ichain=0, lclimb=True)

9
10
11
12
13
14

images, energies = calc.get_neb()

calc.plot_neb(show=False)
import matplotlib.pyplot as plt
plt.savefig(images/pt-o-cineb.svg)
plt.show()

Open the python script (dft-scripts/script-186.py).

Climbing image NEB. TODO This did not do exactly what I expected. I thought there would be
an image that had climbed to the top. Maybe this is an artifact of the spline fitting, or the top is within
the tolerance of the top two points. More band points would probably clarify that.
5.10.3

Using vibrations to confirm a transition state

A transition state should have exactly one imaginary degree of freedom which corresponds to the mode
that takes reactants to products. See Vibrations of the bridge site for an example.

5.11

TODO Diffusion rates with transition state theory

5.12

TODO Effects of electric fields on adsorbates

Electric fields can polarize adsorbates, i.e. shift their electron density around, which changes their
stability . We can model this to some extent in VASP. We have to set the incar:EFIELD parameter to
specify the field strength (in V/), and incar:LDIPOL to True, and finally specify which lattice vector
the field should be applied to with incar:IDIPOL. Usually, incar:IDIPOL will be set to 3 for surfaces so
it points in the direction of the third lattice vector, which is conventionally in the z-direction and normal
to the surface.
We will illustrate the effect using an example from 94 for the effect of electric field on the adsorption
energy of a CO molecule on Pt(111) at 0.25 ML. For simplicity, we will use a frozen slab, with the CO
molecule in the geometry reported in that reference.
1
2

from jasp import *

from ase.lattice.surface import fcc111

3
4
5
6

atoms = fcc111(Pt, size=(2, 2, 4), vacuum=10.0, a=3.986)

atoms.append(Atom(C, [atoms[12].x, atoms[12].y, atoms[12].z + 1.851]))
atoms.append(Atom(O, [atoms[-1].x, atoms[-1].y, atoms[-1].z + 1.157]))

7
8
9
10
11
12
13
14
15
16
17
18
19
20

for field in [-0.5, 0.0, 0.5]:

with jasp(surfaces/Pt-co-field-{0}.format(field),
xc=PBE,
encut=350,
kpts=(6, 6, 1),
efield=field, # set the field
ldipol=True,
# turn dipole correction on
idipol=3,
# set field in z-direction
atoms=atoms) as calc:
try:
print {0}: {1:1.3f}.format(field, atoms.get_potential_energy())
except (VaspSubmitted, VaspQueued):
pass

Open the python script (dft-scripts/script-187.py).

0.0: -108.077
188

These energies are crazy different, and the forces on the atoms with the field on are insanely high.
This section is on hold until I figure that out.

5.13

TODO Simulating STM images

http://cst-www.nrl.navy.mil/users/sullivan/stm_backup/stm2.html

189

Figure 63: The unreconstructed Au(110) surface viewed from the side.
190

Figure 64: Au(110) with the missing row reconstruction.

191

Figure 65: xy averaged local electrostatic potential of an Al(111) slab.

Figure 66: Schematic figure illustrating the calculation of a surface energy.

192

Figure 67: Surface energy of a Cu(111) slab as a function of thickness.

193

194

Figure 69: Comparison of the electrostatic potentials with a dipole correction and without it.

195

Figure 70: Schematic of an adsorption process.

196

Figure 71: Pt(111) fcc surface

Figure 80: Schematic of forming a surface with adsorbates. First we form two clean surfaces by cleaving
the bulk, then allow adsorption to occur on the surfaces.

197

Figure 81: Energy pathway for O diffusion from an fcc to hcp site with a spline fit to determine the
barrier.

198

Atomistic thermodynamics

Let us consider how much the Gibbs free energy of an O2 molecule changes as a function of temperature,
at 1 atm. We use the Shomate polynomials to approximate the temperature dependent entropy and
enthalpy, and use the parameters from the NIST Webbook for O2 .
1
2

from ase.units import *

K = 1.0

3
4

print J, mol, K

5
6

print 0.100 * kJ / mol / K

7
8

print 1 * eV / (kJ / mol)

Open the python script (dft-scripts/script-188.py).

6.24150636309e+18 6.0221367e+23 1.0
0.00103642721413
96.485308989
1
2
3
4

import numpy as np
import matplotlib.pyplot as plt
from ase.units import *
K = 1. # Kelvin not defined in ase.units!

5
6
7
8
9

#
A
D
G

Shomate parameters
= 31.32234; B = -20.23531; C = 57.86644
= -36.50624; E = -0.007374; F = -8.903471
= 246.7945; H = 0.0

10
11
12
13
14
15
16
17
18
19

def entropy(T):
entropy returned as eV/K
T in K

t = T / 1000.
s = (A * np.log(t) + B * t + C * (t**2) / 2.
+ D * (t**3) / 3. - E / (2. * t**2) + G)
return s * J / mol / K

20
21
22
23
24
25
26
27

def enthalpy(T):
H - H(298.15) returned as eV/molecule
t = T / 1000.
h = (A * t + B * (t**2) / 2. + C * (t**3) / 3.
+ D * (t**4) / 4. - E / t + F - H)
return h * kJ / mol

28
29

T = np.linspace(10, 700)

30
31

G = enthalpy(T) - T * entropy(T)

32
33
34
35
36

plt.plot(T, G)
plt.xlabel(Temperature (K))
plt.ylabel(r$\Delta G^\circ$ (eV))
plt.savefig(images/O2-mu.png)

Open the python script (dft-scripts/script-189.py).

This is clearly a big effect! Between 500-600K, the energy has dropped by nearly 1 eV.
Pressure also affects the free energy. In the ideal gas limit, the pressure changes the free energy by
kT ln P/P0 where P0 is the standard state pressure (1 atm or 1 bar depending on the convention chosen).
Let us see how this affects the free energy at different temperatures.
1
2
3

import matplotlib.pyplot as plt

import numpy as np
from ase.units import *

199

Figure 82: Effect of temperature on the Gibbs free energy of an O_2 molecule at standard state (1 atm).

4
5
6

atm = 101325 * Pascal #atm is not defined in units

K = 1 # Kelvin

7
8
9

# examine range over 10^-10 to 10^10 atm

P = np.logspace(-10, 10) * atm

10
11
12
13
14
15
16

plt.semilogx(P / atm, kB * (300 * K) * np.log(P / (1 * atm)), label=300K)

plt.semilogx(P / atm, kB * (600 * K) * np.log(P / (1 * atm)), label=600K)
plt.xlabel(Pressure (atm))
plt.ylabel(r$\Delta G$ (eV))
plt.legend(loc=best)
plt.savefig(images/O2-g-p.png)

Open the python script (dft-scripts/script-190.py).

None
Similarly, you can see that simply changing the pressure has a large effect on the Gibbs free energy
of an ideal gas through the term: kT ln(P/P0 ), and that this effect is also temperature dependent. This
leads us to the final formula we will use for the chemical potential of oxgyen:
DF T
ZP E
O2 = EO
+ EO
+ (T ) + kT ln(P/P0 )
2
2
We can use O2 in place of EO2 everywhere to include the effects of pressure and temperature on the
gas phase energy. If T=0K, and P=1 bar, we are at standard state, and this equation reduces to the
DFT energy (+ the ZPE).

200

Figure 83: Effects of pressure on the ideal gas Gibbs free energy of O2 .

6.1

Bulk phase stability of oxides

We will consider the effects of oxygen pressure and temperature on the formation energy of Ag2 O and
Cu2 O. For now, we neglect the effect of pressure and temperature on the solid phases. Neglecting pressure
is pretty reasonable, as the solids are not that compressible, and we do not expect the energy to change
for small pressures. For neglecting the temperature, we assume that the temperature dependence of the
oxide is similar to the temperature dependence of the metal, and that these dependencies practically
cancel each other in the calculations. That is an assumption, and it may not be correct.
2Cu + 1/2O2 Cu2 O
In atomistic thermodynamics, we define the free energy of formation as:
Gf = GCu2 O 2GCu 0.5GO2
We will at this point assume that the solids are incompressible so that pV 0, and that SCu2 O
2SCu 0, which leads to GCu2 O 2GCu ECu2 O 2ECu , which we directly compute from DFT. We
DF T
ZP E
express GO2 = O2 = EO
+ EO
+ (T ) + kT ln(P/P0 ). In this example we neglect the zero-point
2
2
energy of the oxygen molecule, and finally arrive at:
DF T
Gf ECu2 O 2ECu 0.5(EO
+ (T ) + kT ln(P/P0 ))
2
Which, after grouping terms is:
DF T
Gf ECu2 O 2ECu 0.5(EO
) 0.5 O2 (P, T )
2
with O2 (P, T ) = (T ) + kT ln(P/P0 ). We get (T ) from the Janaf Tables, or the NIST
Webbook.
we are explicitly neglecting all entropies of the solid: configurational, vibrational and electronic
we also neglect enthalpic contributions from temperature dependent electronic and vibrational
states
201

You will recognize in this equation the standard formation energy we calculated in Metal oxide
oxidation energies plus a correction for the non standard state pressure and temperature (O2 (P, T ) = 0
at standard state).
Gf Hf 0.5 O2 (P, T )
The formation energy of Cu2 O is -1.9521 eV/formula unit. The formation energy for Ag2 O is -0.99
eV/formula unit. Let us consider what temperature the oxides decompose at a fixed oxygen pressure of
11010 atm. We need to find the temperature where:
Hf = 0.5 O2 (P, T )
which will make the formation energy be 0.
1
2
3
4

import numpy as np
import matplotlib.pyplot as plt
from ase.units import *
from scipy.optimize import fsolve

5
6
7

K = 1. #not defined in ase.units!

atm = 101325 * Pascal

8
9
10
11
12

#
A
D
G

Shomate parameters valid from 100-700K

= 31.32234; B = -20.23531; C = 57.86644
= -36.50624; E = -0.007374; F = -8.903471
= 246.7945; H = 0.0

13
14
15
16
17
18
19
20
21
22

def entropy(T):
entropy returned as eV/K
T in K

t = T/1000.
s = (A * np.log(t) + B * t + C * (t**2) / 2.
+ D * (t**3) / 3. - E / (2. * t**2) + G)
return s * J / mol / K

23
24
25
26
27
28
29
30

def enthalpy(T):
H - H(298.15) returned as eV/molecule
t = T / 1000.
h = (A * t + B * (t**2) / 2. + C * (t**3) / 3.
+ D * (t**4) / 4. - E / t + F - H)
return h * kJ / mol

31
32
33
34
35
36
37
38
39

def DeltaMu(T, P):

returns delta chemical potential of oxygen at T and P

T in K
P in atm

return enthalpy(T) - T * entropy(T) + kB * T * np.log(P / atm)

40
41

P = 1e-10*atm

42
43
44
45
46

def func(T):
Cu2O
return -1.95 - 0.5*DeltaMu(T, P)

47
48
49

print Cu2O decomposition temperature is {0:1.0f} K.format(fsolve(func,

900)[0])

50
51
52
53

def func(T):
Ag2O
return -0.99 - 0.5 * DeltaMu(T, P)

54
55
56

print Ag2O decomposition temperature is {0:1.0f} K.format(fsolve(func,

470)[0])

57
58
59
60
61
62
63
64

T = np.linspace(100, 1000)
# Here we plot delta mu as a function of temperature at different pressures
# you have use \\times to escape the first \ in pyplot
plt.plot(T, DeltaMu(T, 1e10*atm), label=r1$\times 10^{10}$ atm)
plt.plot(T, DeltaMu(T, 1e5*atm), label=r1$\times 10^5$ atm)
plt.plot(T, DeltaMu(T, 1*atm), label=1 atm)

202

65
66

plt.plot(T, DeltaMu(T, 1e-5*atm), label=r1$\times 10^{-5}$ atm)

plt.plot(T, DeltaMu(T, 1e-10*atm), label=r1$\times 10^{-10}$ atm)

67
68
69
70
71

plt.xlabel(Temperature (K))
plt.ylabel(r$\Delta \mu_{O_2}(T,p)$ (eV))
plt.legend(loc=best)
plt.savefig(images/O2-mu-diff-p.png)

Open the python script (dft-scripts/script-191.py).

Cu2O decomposition temperature is 917 K
Ag2O decomposition temperature is 478 K

Figure 84: O2 (T,p) at different pressures and temperatures.

Now, let us make a phase diagram that shows the boundary between silver oxide, and silver metal in
P and T space.
1
2
3
4

import numpy as np
import matplotlib.pyplot as plt
from ase.units import *
from scipy.optimize import fsolve

5
6
7

K = 1. #not defined in ase.units!

atm = 101325*Pascal

8
9
10
11
12

#
A
D
G

Shomate parameters valid from 100-700K

= 31.32234; B = -20.23531; C = 57.86644
= -36.50624; E = -0.007374; F = -8.903471
= 246.7945; H = 0.0

13
14

203

15
16
17
18
19
20
21
22

def entropy(T):
entropy returned as eV/K
T in K

t = T/1000.
s = (A*np.log(t) + B*t + C*(t**2)/2.
+ D*(t**3)/3. - E/(2.*t**2) + G)
return s*J/mol/K

23
24
25
26
27
28
29
30

def enthalpy(T):
H - H(298.15) returned as eV/molecule
t = T/1000.
h = (A*t + B*(t**2)/2. + C*(t**3)/3.
+ D*(t**4)/4. - E/t + F - H)
return h*kJ/mol

31
32
33
34
35
36
37
38

def DeltaMu(T, P):

T in K
P in atm

return enthalpy(T) - T * entropy(T) + kB * T * np.log(P / atm)

39
40
41
42

P = np.logspace(-11, 1, 10) * atm

T = []
for p in P:

43
44
45
46

def func(T):
return -0.99 - 0.5 * DeltaMu(T, p)
T.append(fsolve(func, 450)[0])

47
48
49
50
51
52
53

plt.semilogy(T, P / atm)
plt.xlabel(Temperature (K))
plt.ylabel(Pressure (atm))
plt.text(800, 1e-7, Ag)
plt.text(600, 1e-3, Ag$_2$O)
plt.savefig(images/Ag2O-decomposition.png)

Open the python script (dft-scripts/script-192.py).

This shows that at high temperature and low pO2 metallic silver is stable, but if the pO2 gets high
enough, the oxide becomes thermodynamically favorable. Here is another way to look at it.
1
2
3

import numpy as np
import matplotlib.pyplot as plt
from ase.units import *

4
5
6
7

K = 1. # not defined in ase.units!

atm = 101325*Pascal
Hf = -0.99

8
9

P = 1 * atm

10
11

Dmu = np.linspace(-4, 0)

12
13

Hf = -0.99 - 0.5*Dmu

14
15
16
17
18
19

plt.plot(Dmu, Hf, label=Ag$_2$O)

plt.plot(Dmu, np.zeros(Hf.shape), label=Ag)
plt.xlabel(r$\Delta \mu_{O_2}$ (eV))
plt.ylabel($H_f$ (eV))
plt.savefig(images/atomistic-thermo-hf-mu.png)

Open the python script (dft-scripts/script-193.py).

None
This graph shows graphically the O2 required to make the metal more stable than the oxide.
Anything less than about -2 eV will have the metal more stable. That can be achieved by any one of the
following combinations (graphically estimated from Figure 84): About 500K at 11010 atm, 600K at
1105 atm, 900K at 1atm, etc. . .
204

Figure 85: Temperature dependent decomposition pressure for Ag2 O.

6.2

Effect on adsorption

We now consider the question: Given a pressure and temperature, what coverage would you expect on
a surface? We saw earlier that adsorption energies depend on the site and coverage. We lso know the
coverage depends on the pressure and temperature. Above some temperature, desorption occurs, and
below some pressure adsorption will not be favorable. We seek to develop a quantitative method to
determine those conditions.
We redefine the adsorption energy as:
Gads Eslab,ads Eslab ads
where again we neglect all contributions to the free energy of the slabs from vibrational energy
and entropy, as well as configurational entropy if that is relevant. That leaves only the pressure and
temperature dependence of the adsorbate, which we treat in the ideal gas limit.
We expand ads as Eads + (T, p), and thus:
Gads Eslab,ads Eslab Eads (T, p)
or
Gads Hads (T, p)
where Hads is the adsorption energy we defined earlier. Now we can examine the effect of (T, p)
on the adsorption energies. We will use the adsorption energies for the oxygen on Pt(111) system we
computed earlier:
1
2

import numpy as np
import matplotlib.pyplot as plt

3
4
5
6

fcc25 = -1.04
hcp25 = -0.60
bridge25 = -0.49

205

Figure 86: Dependence of the formation energy on the oxygen chemical potential.
Table 6: Adsorption site dependence of adsorption energies of oxygen on Pt(111).
system
H(eV /O)
fcc (0.25 ML)
-1.04
hcp (0.25 ML)
-0.60
bridge (0.25 ML)
-0.49
fcc(1ML)
-0.10

fcc1 = -0.10

8
9

Dmu = np.linspace(-4, 2)

10
11
12
13
14
15

plt.plot(Dmu,
plt.plot(Dmu,
plt.plot(Dmu,
plt.plot(Dmu,
plt.plot(Dmu,

np.zeros(Dmu.shape), label=Pt(111))
0.25 * (fcc25 - 0.5*Dmu), label=fcc - 0.25 ML)
0.25 * (hcp25 - 0.5*Dmu), label=hcp - 0.25 ML)
0.25 * (bridge25 - 0.5*Dmu), label=bridge - 0.25 ML)
1.0 * (fcc1 - 0.5*Dmu), label=fcc - 1.0 ML)

16
17
18
19
20
21

plt.xlabel(r$\Delta \mu O_2$ (eV))

plt.ylabel(r$\Delta G_{ads}$ (eV/O))
plt.legend(loc=best)
plt.savefig(images/atomistic-thermo-adsorption.png)
plt.show()

Open the python script (dft-scripts/script-194.py).

206

Figure 87: Effect of oxygen chemical potential on the adsorption energy.

6.3

Atomistic therodynamics and multiple reactions

In 82 we considered multiple reactions in an atomistic thermodynamic framework. Let us consider these

three reactions of dissociative adsorption of hydrogen and hydrogen disulfide, and consider how to compute the reaction energy for the third reaction.
1. H2 + 2  2H
2. H2 S + 2  H +SH
3. SH +  S +H
The reaction energy of interest is Erxn = S +HSH The question is, what are these chemical
potentials? We would like them in terms of pressures and temperature, preferrably of molecules that
can be approximated as ideal gases. By equilibrium arguments we can say that H = 12 H2 . It follows
that at equilibrium:
H + SH = H2 S and H + S = SH .
From the first equation we have:
SH = H2 S 12 H2
and from the second equation we have:
S = SH H = H2 S H2 .
Thus, the chemical potentials of all these three adsorbed species depend on the chemical potentials
of two gas-phase species. The chemical potentials

of each of these gases can be defined as:
gas (T, p) = Egas (0K) + + kT ln p/p0 , as we have defined before, so that only simple DFT
calculations are needed to estimate them.
207

Advanced electronic structure methods

7.1

DFT+U

VASP manual on DFT+U

It can be difficult to find the lowest energy solutions with DFT+U. Some strategies for improving
this are discussed in. 95
7.1.1

Metal oxide oxidation energies with DFT+U

We will reconsider here the reaction (see Metal oxide oxidation energies) 2 Cu2 O + O2
4 CuO. We
need to compute the energy of each species, now with DFT+U. In 55 they use a U parameter of 4 eV for
Cu which gave the best agreement with the experimental value. We will also try that.
Cu2 O calculation with U=4.0
1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

with jasp(bulk/Cu2O) as calc:

calc.clone(bulk/Cu2O-U=4.0)

6
7
8
9
10
11
12
13
14
15
16

with jasp(bulk/Cu2O-U=4.0) as calc:

calc.set(ldau=True,
# turn DFT+U on
ldautype=2, # select simplified rotationally invariant option
ldau_luj={Cu:{L:2, U:4.0, J:0.0},
O:{L:-1, U:0.0, J:0.0}},
ldauprint=1,
ibrion=-1, #do not rerelax
nsw=0)
calc.calculate()
print calc

Open the python script (dft-scripts/script-195.py).

: ----------------------------VASP calculation from /home/jkitchin/dft-org/bulk/Cu2O-U=4.0
converged: True
Energy = -22.228203 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 4.270 0.000 0.000] 4.270
a1 [ 0.000 4.270 0.000] 4.270
a2 [ 0.000 0.000 4.270] 4.270
a,b,c,alpha,beta,gamma (deg): 4.270 4.270 4.270 90.0 90.0 90.0
Unit cell volume = 77.854 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.018 0.018 0.018 -0.000 -0.000 -0.000
Atom# sym
position [x,y,z]
tag rmsForce constraints
0
Cu [0.000
0.000
0.000]
0
0.00
T T T
1
Cu [2.135
2.135
0.000]
0
0.00
T T T
2
Cu [2.135
0.000
2.135]
0
0.00
T T T
3
Cu [0.000
2.135
2.135]
0
0.00
T T T
4
O
[1.067
1.067
1.067]
0
0.00
T T T
5
O
[3.202
3.202
3.202]
0
0.00
T T T
-------------------------------------------------INCAR Parameters:
208

----------------nbands: 37
nsw: 0
ibrion: -1
ldautype: 2
isif: 3
ldauprint: 1
encut: 400.0
ldau: True
magmom: None
ldaul: [-1.0, 2.0]
ldauj: [0.0, 0.0]
ldauu: [0.0, 4.0]
ldau_luj: {O: {J: 0.0, U: 0.0, L: -1}, Cu: {J: 0.0,
U: 4.0, L: 2}}
prec: Normal
kpts: [8, 8, 8]
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Cu: potpaw_PBE/Cu/POTCAR (git-hash: a44c591415026f53deb16a99ca3f06b1e69be10b)
1

grep -A 3 "LDA+U is selected, type is set to LDAUTYPE" bulk/Cu2O-U=4.0/OUTCAR

Open the python script (dft-scripts/script-196.py).

LDA+U is selected,
angular momentum
U (eV)
J (eV)

type is set to LDAUTYPE =

for each species LDAUL =
for each species LDAUU =
for each species LDAUJ =

2
-1
0.0
0.0

2
4.0
0.0

CuO calculation with U=4.0

1
2

from jasp import *

from ase import Atom, Atoms

3
4
5

with jasp(bulk/CuO) as calc:

calc.clone(bulk/CuO-U=4.0)

6
7
8
9
10
11
12
13
14
15
16

with jasp(bulk/CuO-U=4.0) as calc:

calc.set(ldau=True,
# turn DFT+U on
ldautype=2, # select simplified rotationally invariant option
ldau_luj={Cu:{L:2, U:4.0, J:0.0},
O:{L:-1, U:0.0, J:0.0}},
ldauprint=1,
ibrion=-1, #do not rerelax
nsw=0)
calc.calculate()
print calc

Open the python script (dft-scripts/script-197.py).

209

: ----------------------------VASP calculation from /home/jkitchin/dft-org/bulk/CuO-U=4.0

converged: True
Energy = -16.870130 eV
Unit cell vectors (angstroms)
x
y
z
length
a0 [ 2.302 -1.776 0.046] 2.908
a1 [ 2.302 1.776 0.046] 2.908
a2 [-0.762 0.000 5.087] 5.144
a,b,c,alpha,beta,gamma (deg): 2.908 2.908 5.144 95.8 95.8 95.8
Unit cell volume = 41.730 Ang^3
Stress (GPa):xx,
yy,
zz,
yz,
xz,
xy
0.030 0.018 0.027 -0.000 -0.007 -0.000
Atom# sym
position [x,y,z]
tag rmsForce constraints
0
Cu [1.151
0.888
0.023]
0
0.00
T T T
1
Cu [0.770
-0.888
2.566]
0
0.00
T T T
2
O
[2.111
-0.168
1.318]
0
0.04
T T T
3
O
[1.730
0.168
3.861]
0
0.04
T T T
-------------------------------------------------INCAR Parameters:
----------------nbands: 23
nsw: 0
ibrion: -1
ldautype: 2
isif: 3
ldauprint: 1
encut: 400.0
ldau: True
magmom: None
ldaul: [-1.0, 2.0]
ldauj: [0.0, 0.0]
ldauu: [0.0, 4.0]
ldau_luj: {O: {J: 0.0, U: 0.0, L: -1}, Cu: {J: 0.0,
U: 4.0, L: 2}}
prec: Normal
kpts: [8, 8, 8]
reciprocal: False
xc: PBE
txt: gamma: False
Pseudopotentials used:
---------------------O: potpaw_PBE/O/POTCAR (git-hash: 9a0489b46120b0cad515d935f44b5fbe3a3b1dfa)
Cu: potpaw_PBE/Cu/POTCAR (git-hash: a44c591415026f53deb16a99ca3f06b1e69be10b)
Reaction energy calculation with DFT+U
1

from jasp import *

210

3
4
5
6
7
8

# dont forget to normalize your total energy to a formula unit. Cu2O

# has 3 atoms, so the number of formula units in an atoms is
# len(atoms)/3.
with jasp(bulk/Cu2O-U=4.0) as calc:
atoms = calc.get_atoms()
cu2o_energy = atoms.get_potential_energy()/(len(atoms)/3)

9
10
11
12

with jasp(bulk/CuO-U=4.0) as calc:

atoms = calc.get_atoms()
cuo_energy = atoms.get_potential_energy()/(len(atoms)/2)

13
14
15
16
17

# make sure to use the same cutoff energy for the O2 molecule!
with jasp(molecules/O2-sp-triplet-400) as calc:
atoms = calc.get_atoms()
o2_energy = atoms.get_potential_energy()

18
19
20
21

rxn_energy = 4.0*cuo_energy - o2_energy - 2.0*cu2o_energy

print Reaction energy = {0} eV.format(rxn_energy)
print Corrected energy = {0} eV.format(rxn_energy - 1.36)

Open the python script (dft-scripts/script-198.py).

Reaction energy = -1.663819 eV
Corrected energy = -3.023819 eV
This is still not in quantitative agreement with the result in, 55 which at U=4 eV is about -3.14 eV
(estimated from a graph). We have not applied the O2 correction here yet. In that paper, they apply
a constant shift of -1.36 eV per O2 . After we apply that correction, we agree within 0.12 eV, which is
pretty good considering we have not checked for convergence.
How much does U affect the reaction energy?
results are to the U parameter. We do that here.
1
2
3
4
5

It is reasonable to consider how sensitive our

from jasp import *

for U in [2.0, 4.0, 6.0]:
## Cu2O ########################################
with jasp(bulk/Cu2O) as calc:
calc.clone(bulk/Cu2O-U={0}.format(U))

6
7
8
9
10
11
12
13
14
15
16

with jasp(bulk/Cu2O-U={0}.format(U)) as calc:

calc.set(ldau=True,
# turn DFT+U on
ldautype=2, # select simplified rotationally invariant option
ldau_luj={Cu:{L:2, U:U, J:0.0},
O:{L:-1, U:0.0, J:0.0}},
ldauprint=1,
ibrion=-1, #do not rerelax
nsw=0)
atoms = calc.get_atoms()
cu2o_energy = atoms.get_potential_energy()/(len(atoms)/3)

17
18
19
20

## CuO ########################################
with jasp(bulk/CuO) as calc:
calc.clone(bulk/CuO-U={0}.format(U))

21
22
23
24
25
26
27
28
29
30
31

with jasp(bulk/CuO-U={0}.format(U)) as calc:

calc.set(ldau=True,
# turn DFT+U on
ldautype=2, # select simplified rotationally invariant option
ldau_luj={Cu:{L:2, U:U, J:0.0},
O:{L:-1, U:0.0, J:0.0}},
ldauprint=1,
ibrion=-1, #do not rerelax
nsw=0)
atoms = calc.get_atoms()
cuo_energy = atoms.get_potential_energy()/(len(atoms)/2)

32
33
34
35
36
37

## O2 ########################################
# make sure to use the same cutoff energy for the O2 molecule!
with jasp(molecules/O2-sp-triplet-400) as calc:
atoms = calc.get_atoms()
o2_energy = atoms.get_potential_energy()

38

211

39
40

rxn_energy = 4.0*cuo_energy - o2_energy - 2.0*cu2o_energy

print U = {0} reaction energy = {1}.format(U,rxn_energy - 1.99)

Open the python script (dft-scripts/script-199.py).

U = 2.0
U = 4.0
U = 6.0

reaction energy = -3.876906

reaction energy = -3.653819
reaction energy = -3.397605

In, 55 the difference in reaction energy from U=2 eV to U=4 eV was about 0.5 eV (estimated from
graph). Here we see a range of 0.48 eV from U=2 eV to U=4 eV. Note that for U=0 eV, we had a
(corrected reaction energy of -3.96 eV). Overall, the effect of adding U decreases this reaction energy.
This example highlights the challenge of using an approach like DFT+U. On one hand, U has a
clear effect of changing the reaction energy. On the other hand, so does the correction factor for the O2
binding energy. In 55 the authors tried to get the O2 binding energy correction from oxide calculations
where U is not important, so that it is decoupled from the non-cancelling errors that U fixes. See 96 for
additional discussion of how to mix GGA and GGA+U results.
In any case, you should be careful to use well converged results to avoid compensating for convergence
errors with U.

7.2
7.2.1

Hybrid functionals
FCC Ni DOS

This example is adapted from http://cms.mpi.univie.ac.at/wiki/index.php/FccNi_DOS

1
2
3

from jasp import *

from ase.lattice.cubic import FaceCenteredCubic
from ase.dft import DOS

4
5
6
7
8
9
10

atoms = FaceCenteredCubic(directions=[[0,1,1],
[1,0,1],
[1,1,0]],
size=(1,1,1),
symbol=Ni)
atoms[0].magmom = 1

11
12
13
14
15
16
17
18
19
20
21

with jasp(bulk/Ni-PBE,
ismear=-5,
kpts=(5,5,5),
xc=PBE,
ispin=2,lorbit=11,
atoms=atoms) as calc:
print PBE energy:
,atoms.get_potential_energy()
dos = DOS(calc, width=0.2)
e_pbe = dos.get_energies()
d_pbe = dos.get_dos()

22
23
24

calc.clone(bulk/Ni-PBE0)
calc.clone(bulk/Ni-HSE06)

25
26
27
28
29
30
31
32
33
34

with jasp(bulk/Ni-PBE0) as calc:

calc.set(lhfcalc=True,
algo=D,
time=0.4)
atoms = calc.get_atoms()
print PBE0 energy: ,atoms.get_potential_energy()
dos = DOS(calc, width=0.2)
e_pbe0 = dos.get_energies()
d_pbe0 = dos.get_dos()

35
36
37
38
39
40
41

with jasp(bulk/Ni-HSE06) as calc:

calc.set(lhfcalc=True,
hfscreen=0.2,
algo=D, time=0.4)
atoms = calc.get_atoms()
print HSE06 energy: , atoms.get_potential_energy()

212

dos = DOS(calc, width=0.2)

e_hse06 = dos.get_energies()
d_hse06 = dos.get_dos()

42
43
44
45
46
47
48
49
50
51
52
53

import pylab as plt

plt.plot(e_pbe, d_pbe, label=PBE)
plt.plot(e_pbe0, d_pbe0, label=PBE0)
plt.plot(e_hse06, d_hse06, label=HSE06)
plt.xlabel(energy [eV])
plt.ylabel(DOS)
plt.legend()
plt.savefig(images/ni-dos-pbe-pbe0-hse06.png)

Open the python script (dft-scripts/script-200.py).

PBE energy:
PBE0 energy:
HSE06 energy:

-5.530247
-6.848931
-6.293369

Figure 88: Comparison of DOS from GGA, and two hybrid GGAs (PBE0 ad HSE06).

7.3

TODO DFT+D

http://cms.mpi.univie.ac.at/vasp/vasp/DFT_D2_method_Grimme.html 97
Van der Waal forces can play a considerable role in binding of aromatic molecules to metal surfaces
(ref). Here we consider the effects of these forces on the adsorption energy of benzene on an Au(111)
surface.First, we consider the regular PBE functional.

213

7.3.1

PBE

gas-phase benzene
1
2

from jasp import *

from ase.data.molecules import molecule

3
4
5

benzene = molecule(C6H6)
benzene.center(vacuum=5)

6
7
8
9
10
11
12
13
14

with jasp(molecules/benzene-pbe,
xc=PBE,
encut=350,
kpts=(1,1,1),
ibrion=1,
nsw=100,
atoms=benzene) as calc:
print benzene.get_potential_energy()

Open the python script (dft-scripts/script-201.py).

-76.011625
clean slab
1
2
3
4

# the clean gold slab

from jasp import *
from ase.lattice.surface import fcc111, add_adsorbate
from ase.constraints import FixAtoms

5
6

atoms = fcc111(Au, size=(3,3,3), vacuum=10)

7
8
9
10

# now we constrain the slab

c = FixAtoms(mask=[atom.symbol==Au for atom in atoms])
atoms.set_constraint(c)

11
12

#from ase.visualize import view; view(atoms)

13
14
15
16
17
18
19
20
21

with jasp(surfaces/Au-pbe,
xc=PBE,
encut=350,
kpts=(4,4,1),
ibrion=1,
nsw=100,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-202.py).

-82.419513
benzene on Au(111)
1
2
3
4
5

# Benzene on the slab

from jasp import *
from ase.lattice.surface import fcc111, add_adsorbate
from ase.data.molecules import molecule
from ase.constraints import FixAtoms

6
7
8
9

atoms = fcc111(Au, size=(3,3,3), vacuum=10)

benzene = molecule(C6H6)
benzene.translate(-benzene.get_center_of_mass())

10
11
12
13
14
15
16

# I want the benzene centered on the position in the middle of atoms

# 20, 22, 23 and 25
p = (atoms.positions[20] +
atoms.positions[22] +
atoms.positions[23] +
atoms.positions[25])/4.0 + [0.0, 0.0, 3.05]

17

214

18
19

benzene.translate(p)
atoms += benzene

20
21
22
23

# now we constrain the slab

c = FixAtoms(mask=[atom.symbol==Au for atom in atoms])
atoms.set_constraint(c)

24
25

#from ase.visualize import view; view(atoms)

26
27
28
29
30
31
32
33
34

with jasp(surfaces/Au-benzene-pbe,
xc=PBE,
encut=350,
kpts=(4,4,1),
ibrion=1,
nsw=100,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-203.py).

7.3.2

DFT-D2

To turn on the van der Waals corrections 98 we set incar:LVDW to True.

gas-phase benzene
1
2

from jasp import *

from ase.data.molecules import molecule

3
4
5

benzene = molecule(C6H6)
benzene.center(vacuum=5)

6
7
8
9
10
11
12
13
14
15

with jasp(molecules/benzene-pbe-d2,
xc=PBE,
encut=350,
kpts=(1,1,1),
ibrion=1,
nsw=100,
lvdw=True,
atoms=benzene) as calc:
print benzene.get_potential_energy()

Open the python script (dft-scripts/script-204.py).

clean slab
1
2
3
4

# the clean gold slab

from jasp import *
from ase.lattice.surface import fcc111, add_adsorbate
from ase.constraints import FixAtoms

5
6

atoms = fcc111(Au, size=(3,3,3), vacuum=10)

7
8
9
10

# now we constrain the slab

c = FixAtoms(mask=[atom.symbol==Au for atom in atoms])
atoms.set_constraint(c)

11
12

#from ase.visualize import view; view(atoms)

13
14
15
16
17
18
19
20
21
22

with jasp(surfaces/Au-pbe-d2,
xc=PBE,
encut=350,
kpts=(4,4,1),
ibrion=1,
nsw=100,
lvdw=True,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-205.py).

215

benzene on Au(111)
1
2
3
4
5

# Benzene on the slab

from jasp import *
from ase.lattice.surface import fcc111, add_adsorbate
from ase.data.molecules import molecule
from ase.constraints import FixAtoms

6
7
8
9

atoms = fcc111(Au, size=(3,3,3), vacuum=10)

benzene = molecule(C6H6)
benzene.translate(-benzene.get_center_of_mass())

10
11
12
13
14
15
16

# I want the benzene centered on the position in the middle of atoms

# 20, 22, 23 and 25
p = (atoms.positions[20] +
atoms.positions[22] +
atoms.positions[23] +
atoms.positions[25])/4.0 + [0.0, 0.0, 3.05]

17
18
19

benzene.translate(p)
atoms += benzene

20
21
22
23

# now we constrain the slab

c = FixAtoms(mask=[atom.symbol==Au for atom in atoms])
atoms.set_constraint(c)

24
25

#from ase.visualize import view; view(atoms)

26
27
28
29
30
31
32
33
34
35

with jasp(surfaces/Au-benzene-pbe-d2,
xc=PBE,
encut=350,
kpts=(4,4,1),
ibrion=1,
nsw=100,
lvdw=True,
atoms=atoms) as calc:
print atoms.get_potential_energy()

Open the python script (dft-scripts/script-206.py).

7.3.3

Advanced vdW-DF functionals

There is a more sophisticated (i.e. more parameters) treatment of van der Waal forces in VASP (http:
//cms.mpi.univie.ac.at/vasp/vasp/vdW_DF_functional_Langreth_Lundqvist_et_al.html).

7.4

ELF

Need better intro here.

1
2
3
4

# compute ELF for CF4

from jasp import *
from ase.structure import molecule
from enthought.mayavi import mlab

5
6
7

atoms = molecule(CF4)
atoms.center(vacuum=5)

8
9
10
11
12
13
14
15
16
17

with jasp(molecules/cf4-elf,
encut=350,
prec=high,
ismear=0,
sigma=0.01,
xc=PBE,
lelf=True,
atoms=atoms) as calc:
calc.calculate()

18
19
20
21

x, y, z, elf = calc.get_elf()
mlab.contour3d(x, y, z, elf,contours=[0.3])
mlab.savefig(../../images/cf4-elf-3.png)

22
23

mlab.figure()

216

24
25

mlab.contour3d(x, y, z, elf,contours=[0.75])
mlab.savefig(../../images/cf4-elf-75.png)

Open the python script (dft-scripts/script-207.py).

None

./images/mlab-test.png

These images (Figure 89 and 90) are basically consistent with those in Reference. 99

7.5

TODO Charge partitioning schemes

7.6

TODO Modeling Core level shifts

217

Figure 89: ELF for an isosurface of 0.3 for CF4 .

218

Figure 90: ELF for an isosurface of 0.75 for CF4 .

219

Acknowledgments

I would like to thank Zhongnan Xu for sending me some examples on magnetism. Alan McGaughey and
Lars Grabow for sending me some NEB examples. Matt Curnan for examples of phonons.
Many thanks to students in my class who have pointed out typos, places of confusion, etc. . . These
include Bruno Calfa, Matt Curnan, Charlie Janini, Feng Cao, Gamze Gumuslu, Nicholas Chisholm,
Prateek Mehta, Qiyang Duan, Shubhaditya Majumdar, Steven Illes, Wee-Liat Ong, Ye Wang, Yichun
Sun, Yubing Lu, and Zhongnan Xu.

Appendices

9.1

Recipes

9.1.1

Modifying Atoms by deleting atoms

Sometimes it is convenient to create an Atoms object by deleting atoms from an existing object. Here
is a recipe to delete all the hydrogen atoms in a molecule. The idea is to make a list of indices of which
atoms to delete using list comprehension, then use list deletion to delete those indices.
1
2

import textwrap
from ase.data.molecules import molecule

3
4
5

atoms = molecule(CH3CH2OH)
print atoms

6
7
8
9

#delete all the hydrogens

ind2del = [atom.index for atom in atoms if atom.symbol==H]
print Indices to delete: ,ind2del

10
11

del atoms[ind2del]

12
13
14

# now print what is left

print atoms

Open the python script (dft-scripts/script-208.py).

Atoms(symbols=C2OH6, positions=..., cell=[1.0, 1.0, 1.0], pbc=[False, False, False])
Indices to delete: [3, 4, 5, 6, 7, 8]
Atoms(symbols=C2O, positions=..., cell=[1.0, 1.0, 1.0], pbc=[False, False, False])
9.1.2

Advanced tagging

We can label atoms with integer tags to help identify them later, e.g. which atoms are adsorbates, or
surface atoms, or near an adsorbate, etc. . . We might want to refer to those atoms later for electronic
structure, geometry analysis, etc. . .
The method uses integer tags that are powers of two, and then uses binary operators to check for
matches. & is a bitwise AND. The key to understanding this is to look at the tags in binary form. The
tags [1 2 4 8] can be represented by a binary string:
1
2
4
8

=
=
=
=

[1
[0
[0
[0

0
1
0
0

0
0
1
0

0]
0]
0]
1]

So, an atom tagged with 1 and 2 would have a tag of [1 1 0 0] or equivalently in decimal numbers, a
tag of 3.

220

1
2
3

adapted from https://listserv.fysik.dtu.dk/pipermail/campos/2004-September/001155.html

4
5
6
7
8

from
from
from
from

ase import *
ase.io import write
ase.lattice.surface import bcc111, add_adsorbate
ase.constraints import FixAtoms

9
10
11
12
13
14

# the bcc111 function automatically tags atoms

slab = bcc111(W,
a=3.92,
# W lattice constant
size=(2,2,6), # 6-layer slab in 2x2 configuration
vacuum=10.0)

15
16
17

#reset tags to be powers of two so we can use binary math

slab.set_tags([2**a.get_tag() for a in slab])

18
19
20
21
22
23
24

# we had 6 layers, so we create new tags starting at 7

# Note you must use powers of two for all the tags!
LAYER1 = 2
ADSORBATE = 2**7
FREE = 2**8
NEARADSORBATE = 2**9

25
26
27
28
29
30
31

# let us tag LAYER1 atoms to be FREE too. we can address it by LAYER1 or FREE
tags = slab.get_tags()
for i,tag in enumerate(tags):
if tag == LAYER1:
tags[i] += FREE
slab.set_tags(tags)

32
33
34
35

#create a CO molecule
co= Atoms([Atom(C,[0., 0., 0. ], tag=ADSORBATE),
Atom(O,[0., 0., 1.1], tag=ADSORBATE+FREE)]) #we will relax only O

36
37

add_adsorbate(slab,co,height=1.2,position=hollow)

38
39
40
41
42
43
44
45
46
47
48
49

#the adsorbate is centered between atoms 20, 21 and 22 (use

#view(slab)) and over atom12 let us label those atoms, so it is easy to
#do electronic structure analysis on them later.
tags = slab.get_tags() # len(tags) changed, so we reget them.
tags[12]+=NEARADSORBATE
tags[20]+=NEARADSORBATE
tags[21]+=NEARADSORBATE
tags[22]+=NEARADSORBATE
slab.set_tags(tags)
#update the tags
slab.set_tags(tags)

50
51
52
53

#extract pieces of the slab based on tags

#atoms in the adsorbate
ads = slab[(slab.get_tags() & ADSORBATE) == ADSORBATE]

54
55
56

#atoms in LAYER1
layer1 = slab[(slab.get_tags() & LAYER1) == LAYER1]

57
58
59

#atoms defined as near the adsorbate

nearads = slab[(slab.get_tags() & NEARADSORBATE) == NEARADSORBATE]

60
61
62

#atoms that are free

free = slab[(slab.get_tags() & FREE) == FREE]

63
64
65

#atoms that are FREE and part of the ADSORBATE

freeads = slab[(slab.get_tags() & FREE+ADSORBATE) == FREE+ADSORBATE]

66
67
68

#atoms that are NOT FREE

notfree = slab[(slab.get_tags() & FREE) != FREE]

69
70
71
72
73

constraint = FixAtoms(mask = (slab.get_tags() & FREE) != FREE)

slab.set_constraint(constraint)
write(images/tagged-bcc111.png, slab, rotation=-90x, show_unit_cell=2)
from ase.visualize import view; view(slab)

Open the python script (dft-scripts/script-209.py).

221

Figure 91: The tagged bcc(111) structure created above. Unfortunately, the frozen atoms do not show
up in the figure.
9.1.3

Using units in ase

mod:ase uses a base set of atomic units.These are eV for energy, for distance, seconds for time, and
amu for mass. Other units are defined in terms of those units, and you can easily convert to alternative
222

units by dividing your quantity in atomic units by the units you want.
Not too many units are defined: [A, AUT, Ang, Angstrom, Bohr, C, Debye, GPa, Ha,
Hartree, J, Pascal, Ry, Rydberg, alpha, cm, eV, erg, fs, kB, kJ, kcal, kg, m, meV,
mol, nm, s, second]
It is not that hard to define your own derived units though. Note these are only conversion factors.
No units algebra is enforced (i.e. it will be ok to add a m and a kg)!
1

from ase.units import *

2
3
4

d = 1*Angstrom
print d = {0} nm.format(d/nm)

5
6
7
8
9

print
print
print
print

1
1
1
1

eV
eV
eV
eV

=
=
=
=

{0}
{0}
{0}
{0}

Hartrees.format(eV/Hartree)
Rydbergs.format(eV/Rydberg)
kJ/mol.format(eV/(kJ/mol))
kcal/mol.format(eV/(kcal/mol))

10
11
12

print 1 Hartree = {0} kcal/mol.format(1*Hartree/(kcal/mol))

print 1 Rydberg = {0} eV.format(1*Rydberg/eV)

13
14
15
16

# derived units
minute = 60*s
hour = 60*minute

17
18
19

#convert 10 hours to minutes

print 10 hours = {0} minutes.format(10*hour/minute)

Open the python script (dft-scripts/script-210.py).

d = 0.1 nm
1 eV = 0.036749 Hartrees
1 eV = 0.073499 Rydbergs
1 eV = 96.485309 kJ/mol
1 eV = 23.060542 kcal/mol
1 Hartree = 627.509541 kcal/mol
1 Rydberg = 13.605698 eV
10 hours = 600.0 minutes
9.1.4

Extracting parts of an array

See http://www.scipy.org/Cookbook/BuildingArrays for examples of making numpy arrays.

When analyzing numerical data you may often want to analyze only a part of the data. For example,
suppose you have x and y data, (x=time, y=signal) and you want to integrate the date between a
particular time interval. You can slice a numpy array to extract parts of it. See http://www.scipy.
org/Cookbook/Indexing for several examples of this.
In this example we show how to extract the data in an interval. We have x data in the range of 0 to 6,
and y data that is the cos(x). We want to extract the x and y data for 2 < x < 4, and the corresponding
y-data. To do this, we utilize the numpy capability of slicing with a boolean array. We also show some
customization of matplotlib.
1
2
3
4
5

import numpy as np
import matplotlib as mpl
#http://matplotlib.sourceforge.net/users/customizing.html
mpl.rcParams[legend.numpoints] = 1 #default is 2
import matplotlib.pyplot as plt

6
7
8

x = np.linspace(0,6,100)
y = np.cos(x)

9
10

plt.plot(x,y,label=full)

11
12

ind = (x>2) & (x<4)

13

223

14
15

subx = x[ind]
suby = y[ind]

16
17
18
19
20
21

plt.plot(subx,suby,bo,label=sliced)
xlabel(x)
ylabel(cos(x))
plt.legend(loc=lower right)
plt.savefig(images/np-array-slice.png)

Open the python script (dft-scripts/script-211.py).

None

Figure 92: Example of slicing out part of an array. The solid line represents the whole array, and the
symbols are the array between 2 < x < 4.
The expression x > 2 returns an array of booleans (True where the element of x is greater than 2,
and False where it is not) equal in size to x. Similarly x < 4 returns a boolean array where x is less
than 4. We take the logical and of these two boolean arrays to get another boolean array where both
conditions are True (i.e. x < 2 and x > 4). This final boolean array is True for the part of the arrays
we are interested in, and we can use it to extract the subarrays we want.
9.1.5

Statistics

Confidence intervals mod:scipy has a statistical package available for getting statistical distributions.
This is useful for computing confidence intervals using the student-t tables. Here is an example of
computing a 95% confidence interval on an average.

224

1
2

import numpy as np
from scipy.stats.distributions import

3
4
5
6
7

n = 10 #number of measurements
dof = n - 1 #degrees of freedom
avg_x = 16.1 #average measurement
std_x = 0.01 #standard deviation of measurements

8
9

#Find 95% prediction interval for next measurement

10
11

alpha = 1.0 - 0.95

12
13

pred_interval = t.ppf(1-alpha/2., dof) * std_x * np.sqrt(1.+1./n)

14
15
16
17

s = [We are 95% confident the next measurement,

will be between {0:1.3f} and {1:1.3f}]
print .join(s).format(avg_x - pred_interval, avg_x + pred_interval)

Open the python script (dft-scripts/script-212.py).

We are 95% confident the next measurement will be between 16.076 and 16.124
9.1.6

Curve fitting

Linear fitting
1
2

#examples of linear curve fitting using least squares

import numpy as np

3
4
5

xdata = np.array([0.,1.,2.,3.,4.,5.,6.])
ydata = np.array([0.1, 0.81, 4.03, 9.1, 15.99, 24.2, 37.2])

6
7
8
9
10

#fit a third order polynomial

from pylab import polyfit, plot, xlabel, ylabel, show, legend, savefig
pars = polyfit(xdata,ydata,3)
print pars from polyfit: {0}.format(pars)

11
12
13
14
15
16

## numpy method returns more data

A = np.column_stack([xdata**3,
xdata**2,
xdata,
np.ones(len(xdata),np.float)])

17
18
19

pars_np,resids,rank,s = np.linalg.lstsq(A, ydata)

print pars from np.linalg.lstsq: {0}.format(pars_np)

20
21
22
23
24

we are trying to solve Ax = b for x in the least squares sense. There

are more rows in A than elements in x so, we can left multiply each
side by A^T, and then solve for x with an inverse.

25
26
27
28
29
30
31

A^TAx = A^Tb
x = (A^TA)^-1 A^T b

# not as pretty but equivalent!

pars_man= np.dot(np.linalg.inv(np.dot(A.T,A)), np.dot(A.T,ydata))
print pars from linear algebra: {0}.format(pars_man)

32
33
34
35
36

#but, it is easy to fit an exponential function to it!

# y = a*exp(x)+b
Aexp = np.column_stack([np.exp(xdata), np.ones(len(xdata), np.float)])
pars_exp=np.dot(np.linalg.inv(np.dot(Aexp.T, Aexp)), np.dot(Aexp.T, ydata))

37
38
39
40
41
42
43
44
45

plot(xdata, ydata, ro)

fity = np.dot(A, pars)
plot(xdata, fity, k-, label=poly fit)
plot(xdata, np.dot(Aexp, pars_exp), b-, label=exp fit)
xlabel(x)
ylabel(y)
legend()
savefig(images/curve-fit-1.png)

Open the python script (dft-scripts/script-213.py).

225

pars from polyfit: [ 0.04861111 0.63440476 0.61365079 -0.08928571]

pars from np.linalg.lstsq: [ 0.04861111 0.63440476 0.61365079 -0.08928571]
pars from linear algebra: [ 0.04861111 0.63440476 0.61365079 -0.08928571]

Figure 93: Example of linear least-squares curve fitting.

9.1.7
1
2

Nonlinear curve fitting

from scipy.optimize import leastsq

import numpy as np

3
4

vols = np.array([13.71, 14.82, 16.0, 17.23, 18.52])

5
6

energies = np.array([-56.29, -56.41, -56.46, -56.463,-56.41])

7
8
9
10
11
12
13

def Murnaghan(parameters,vol):
From Phys. Rev. B 28, 5480 (1983)
E0 = parameters[0]
B0 = parameters[1]
BP = parameters[2]
V0 = parameters[3]

14
15

E = E0 + B0*vol/BP*(((V0/vol)**BP)/(BP-1)+1) - V0*B0/(BP-1.)

16
17

return E

18
19
20
21
22

def objective(pars,y,x):
#we will minimize this function
err = y - Murnaghan(pars,x)
return err

23
24

x0 = [ -56., 0.54, 2., 16.5] #initial guess of parameters

226

25
26

plsq = leastsq(objective, x0, args=(energies,vols))

27
28

print Fitted parameters = {0}.format(plsq[0])

29
30
31

import matplotlib.pyplot as plt

plt.plot(vols,energies,ro)

32
33
34
35
36
37
38
39

#plot the fitted curve on top

x = np.linspace(min(vols),max(vols),50)
y = Murnaghan(plsq[0],x)
plt.plot(x,y,k-)
plt.xlabel(Volume)
plt.ylabel(energy)
plt.savefig(images/nonlinear-curve-fitting.png)

Open the python script (dft-scripts/script-214.py).

Fitted parameters = (array([-56.46839641,

0.57233217,

2.7407944 ,

16.55905648]), 1)

Figure 94: Example of least-squares non-linear curve fitting.

See additional examples at http://docs.scipy.org/doc/scipy/reference/tutorial/optimize.
html.
9.1.8
1
2

Nonlinear curve fitting by direct least squares minimization

from scipy.optimize import fmin

import numpy as np

3
4

volumes = np.array([13.71, 14.82, 16.0, 17.23, 18.52])

227

5
6

energies = np.array([-56.29, -56.41, -56.46, -56.463,-56.41])

7
8
9
10
11
12
13

def Murnaghan(parameters,vol):
From PRB 28,5480 (1983
E0 = parameters[0]
B0 = parameters[1]
BP = parameters[2]
V0 = parameters[3]

14

E = E0 + B0*vol/BP*(((V0/vol)**BP)/(BP-1)+1) - V0*B0/(BP-1.)

15
16

return E

17
18
19
20
21
22

def objective(pars,vol):
#we will minimize this function
err = energies - Murnaghan(pars,vol)
return np.sum(err**2) #we return the summed squared error directly

23
24

x0 = [ -56., 0.54, 2., 16.5] #initial guess of parameters

25
26

plsq = fmin(objective,x0,args=(volumes,)) #note args is a tuple

27
28

print parameters = {0}.format(plsq)

29
30
31

import matplotlib.pyplot as plt

plt.plot(volumes,energies,ro)

32
33
34
35
36
37
38
39

#plot the fitted curve on top

x = np.linspace(min(volumes),max(volumes),50)
y = Murnaghan(plsq,x)
plt.plot(x,y,k-)
plt.xlabel(Volume ($\AA^3$))
plt.ylabel(Total energy (eV))
plt.savefig(images/nonlinear-fitting-lsq.png)

Open the python script (dft-scripts/script-215.py).

Optimization terminated successfully.
Current function value: 0.000020
Iterations: 137
Function evaluations: 240
parameters = [-56.46932645
0.59141447
1.9044796

9.1.9
1
2
3
4

Nonlinear curve fitting with confidence intervals

# Nonlinear curve fit with confidence interval

import numpy as np
from scipy.optimize import curve_fit
from scipy.stats.distributions import t

5
6
7
8

fit this equation to data

y = c1 exp(-x) + c2*x

9
10
11
12

this is actually a linear regression problem, but it is convenient to

use the nonlinear fitting routine because it makes it easy to get
confidence intervals. The downside is you need an initial guess.

13
14
15

from Matlab
b =

16

4.9671
2.1100

17
18
19
20
21

bint =

22

4.6267
1.7671

23
24
25

5.3075
2.4528

26

228

16.59341303]

Figure 95: Fitting a nonlinear function.

27
28
29

x = np.array([ 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1. ])
y = np.array([ 4.70192769, 4.46826356, 4.57021389, 4.29240134, 3.88155125,
3.78382253, 3.65454727, 3.86379487, 4.16428541, 4.06079909])

30
31
32
33

# this is the function we want to fit to our data

def func(x,c0, c1):
return c0 * np.exp(-x) + c1*x

34
35

pars, pcov = curve_fit(func, x, y, p0=[4.96, 2.11])

36
37

alpha = 0.05 # 95% confidence interval

38
39
40

n = len(y)
# number of data points
p = len(pars) # number of parameters

41
42

dof = max(0, n-p) # number of degrees of freedom

43
44

tval = t.ppf(1.0-alpha/2., dof) # student-t value for the dof and confidence level

45
46
47
48
49
50

for i, p,var in zip(range(n), pars, np.diag(pcov)):

sigma = var**0.5
print c{0}: {1} [{2} {3}].format(i, p,
p - sigma*tval,
p + sigma*tval)

51
52
53
54
55
56
57
58

import matplotlib.pyplot as plt

plt.plot(x,y,bo )
xfit = np.linspace(0,1)
yfit = func(xfit, pars[0], pars[1])
plt.plot(xfit,yfit,b-)
plt.legend([data,fit],loc=best)
plt.savefig(images/nonlin-fit-ci.png)

Open the python script (dft-scripts/script-216.py).

229

c0: 4.96713966439 [4.62674476321

c1: 2.10995112628 [1.76711622067

5.30753456558]
2.45278603188]

Figure 96: Nonlinear fit to data.

9.1.10

Interpolation with splines

When you do not know the functional form of data to fit an equation, you can still fit/interpolate with
splines.
1
2
3
4
5

# use splines to fit and interpolate data

from scipy.interpolate import interp1d
from scipy.optimize import fmin
import numpy as np
import matplotlib.pyplot as plt

6
7
8

x = np.array([ 0,
y = np.array([ 0.,

1,
0.308,

2,
0.55,

3,
0.546,

4
])
0.44 ])

9
10
11

# create the interpolating function

f = interp1d(x, y, kind=cubic, bounds_error=False)

12
13
14
15
16

# to find the maximum, we minimize the negative of the function. We

# cannot just multiply f by -1, so we create a new function here.
f2 = interp1d(x, -y, kind=cubic)
xmax = fmin(f2, 2.5)

17
18

xfit = np.linspace(0,4)

19
20
21
22

plt.plot(x,y,bo)
plt.plot(xfit, f(xfit),r-)
plt.plot(xmax, f(xmax),g*)

230

23
24
25
26
27
28

plt.legend([data,fit,max], loc=best, numpoints=1)

plt.xlabel(x data)
plt.ylabel(y data)
plt.title(Max point = ({0:1.2f}, {1:1.2f}).format(float(xmax),
float(f(xmax))))
plt.savefig(images/splinefit.png)

Open the python script (dft-scripts/script-217.py).

Figure 97: Illustration of a spline fit to data and finding the maximum point.
There are other good examples at http://docs.scipy.org/doc/scipy/reference/tutorial/interpolate.
html
9.1.11

Interpolation in 3D

You might ask, why would I need to interpolate in 3D? Suppose you want to plot the charge density along
a line through a unit cell that does not correspond to grid points. What are you to do? Interpolate. In
contrast to an abundance of methods for 1D and 2D interpolation, I could not find any standard library
methods for 3D interpolation.
The principle we will use to develop an interpolation function in 3D is called trilinear interpolation,
where we use multiple linear 1D interpolations to compute the value of a point inside a cube. As
developed here, this solution only applies to rectangular grids. Later we will generalize the approach.
We state the problem as follows:
We know a scalar field inside a unit cell on a regularly spaced grid. In VASP these fields may be the
charge density or electrostatic potential for example, and they are known on the fft grids. We want to
estimate the value of the scalar field at a point not on the grid, say P=(a,b,c).

231

Solution: Find the cube that contains the point, and is defined by points P1-P8 as shown in Figure
??.

We use 1D interpolation formulas to compute the value of the scalar field at points I1 by interpolating
between P1 and P2, and the value of the scalar field at I2 by interpolating between P3 and P4. In these
points the only variable changing is x, so it is a simple 1D interpolation. We can then compute the
value of the scalar field at I5 by interpolating between I1 and I2. We repeat the process on the top of
the cube, to obtain points I3, I4 and I5. Finally, we compute the value of the scalar field at point P
by interpolating between points I5 and I6. Note that the point I5 has coordinates (a,b,z1) and I6 is at
(a,b,z2), so the final interpolation is again a 1D interpolation along z evaluated at z=c to get the final
value of the scalar field at P=(a,b,c).
1
2

from jasp import jasp

import numpy as np

3
4
5
6

with jasp(molecules/co-centered) as calc:

atoms = calc.get_atoms()
x, y, z, cd = calc.get_charge_density()

7
8
9
10
11

def interp3d(x,y,z,cd,xi,yi,zi):

interpolate a cubic 3D grid defined by x,y,z,cd at the point

(xi,yi,zi)

232

12

13
14
15
16
17
18
19
20
21
22

def get_index(value,vector):

assumes vector ordered decreasing to increasing. A bisection

search would be faster.

for i,val in enumerate(vector):

if val > value:
return i-1
return None

23
24
25
26

xv = x[:,0,0]
yv = y[0,:,0]
zv = z[0,0,:]

27
28

a,b,c = xi, yi, zi

29
30
31
32

i = get_index(a,xv)
j = get_index(b,yv)
k = get_index(c,zv)

33
34
35
36
37
38
39

x1
x2
y1
y2
z1
z2

=
=
=
=
=
=

x[i,j,k]
x[i+1,j,k]
y[i,j,k]
y[i,j+1,k]
z[i,j,k]
z[i,j,k+1]

u1
u2
u3
u4
u5
u6
u7
u8

=
=
=
=
=
=
=
=

cd[i, j, k]
cd[i+1, j, k]
cd[i, j+1, k]
cd[i+1, j+1, k]
cd[i, j, k+1]
cd[i+1, j, k+1]
cd[i, j+1, k+1]
cd[i+1, j+1, k+1]

40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57

w1 = u2
w2 = u4
w3 = w2
w4 = u5
w5 = u7
w6 = w4
w7 = w3
u = w7

+
+
+
+
+
+
+

(u2-u1)/(x2-x1)*(a-x2)
(u4-u3)/(x2-x1)*(a-x2)
(w2-w1)/(y2-y1)*(b-y2)
(u6-u5)/(x2-x1)*(a-x1)
(u8-u7)/(x2-x1)*(a-x1)
(w5-w4)/(y2-y1)*(b-y1)
(w6-w3)/(z2-z1)*(c-z1)

58
59

return u

60
61

pos = atoms.get_positions()

62
63
64

P1 = np.array([0.0, 5.0, 5.0])

P2 = np.array([9.0, 5.0, 5.0])

65
66

npoints = 60

67
68

points = [P1 + n*(P2-P1)/npoints for n in range(npoints)]

69
70

R = [np.linalg.norm(p-P1) for p in points]

71
72
73

# interpolated line
icd = [interp3d(x,y,z,cd,p[0],p[1],p[2]) for p in points]

74
75

import matplotlib.pyplot as plt

76
77
78
79
80
81
82
83
84
85

plt.plot(R, icd)
cR = np.linalg.norm(pos[0] - P1)
oR = np.linalg.norm(pos[1] - P1)
plt.plot([cR, cR], [0, 2], r-) #markers for where the nuclei are
plt.plot([oR, oR], [0, 8], r-)
plt.xlabel(|R| ($\AA$))
plt.ylabel(Charge density (e/$\AA^3$))
plt.savefig(images/CO-charge-density.png)
plt.show()

Open the python script (dft-scripts/script-218.py).

None
233

Figure 98: An example of interpolated charge density of a CO molecule along the axis of molecule.
To generalize this to non-cubic cells, we need to do interpolation along arbitrary vectors. The overall
strategy is the same:
Find the cell that contains the point (a,b,c). compute the scaled coordinates (sa,sb,sc) of the point
inside the cell. Do the interpolations along the basis vectors. Given u1 at P1(x1,y1,z1) and u2 at
P2(x2,y2,z2) where (P2-P1) is a cell basis vector a, u = u1 + sa*(u2-u1). There are still 7 interpolations
to do.
Below is an example of this code, using a the python library bisect to find the cell.
1
2
3

3D vector interpolation in non-cubic unit cells with vector

interpolation.

4
5
6
7
8
9
10

This function should work for any shape unit cell.

from jasp import *

import bisect
import numpy as np
from pylab import plot, xlabel, ylabel, savefig, show

11
12
13
14

with jasp(molecules/co-centered) as calc:

atoms = calc.get_atoms()
x,y,z,cd = calc.get_charge_density()

15
16

def vinterp3d(x,y,z,u,xi,yi,zi):

17
18

p = np.array([xi,yi,zi])

19
20
21
22
23

#1D arrays of cooridinates

xv = x[:,0,0]
yv = y[0,:,0]
zv = z[0,0,:]

24

234

25
26
27
28
29
30

#
#
#
i
j
k

we subtract 1 because bisect tells us where to insert the

element to maintain an ordered list, so we want the index to the
left of that point
= bisect.bisect_right(xv,xi) - 1
= bisect.bisect_right(yv,yi) - 1
= bisect.bisect_right(zv,zi) - 1

31
32
33
34
35
36

#points at edge of cell. We only need P1, P2, P3, and P5

P1 = np.array([x[i,j,k],y[i,j,k],z[i,j,k]])
P2 = np.array([x[i+1,j,k],y[i+1,j,k],z[i+1,j,k]])
P3 = np.array([x[i,j+1,k],y[i,j+1,k],z[i,j+1,k]])
P5 = np.array([x[i,j,k+1],y[i,j,k+1],z[i,j,k+1]])

37
38
39
40
41
42
43
44
45
46

#values of u at edge of cell

u1 = u[i,j,k]
u2 = u[i+1,j,k]
u3 = u[i,j+1,k]
u4 = u[i+1,j+1,k]
u5 = u[i,j,k+1]
u6 = u[i+1,j,k+1]
u7 = u[i,j+1,k+1]
u8 = u[i+1,j+1,k+1]

47
48
49
50
51

#cell basis vectors, not the unit cell, but the voxel cell containing the point
cbasis = np.array([P2-P1,
P3-P1,
P5-P1])

52
53
54

#now get interpolated point in terms of the cell basis

s = np.dot(np.linalg.inv(cbasis.T),np.array([xi,yi,zi])-P1)

55
56
57
58
59

#now s =
#next we
ui1 = u1
ui2 = u3

(sa, sb, sc) which are fractional coordinates in the vector space
do the interpolations
+ s[0]*(u2-u1)
+ s[0]*(u4-u3)

60
61
62

ui3 = u5 + s[0]*(u6-u5)
ui4 = u7 + s[0]*(u8-u7)

63
64
65

ui5 = ui1 + s[1]*(ui2-ui1)

ui6 = ui3 + s[1]*(ui4-ui3)

66
67

ui7 = ui5 + s[2]*(ui6-ui5)

68
69

return ui7

70
71
72
73

# compute a line with 60 points in it through these two points

P1 = np.array([0.0, 5.0, 5.0])
P2 = np.array([10.0, 5.0, 5.0])

74
75

npoints = 60

76
77

points = [P1 + n*(P2-P1)/npoints for n in range(npoints)]

78
79
80

# compute the distance along the line

R = [np.linalg.norm(p-P1) for p in points]

81
82

icd = [vinterp3d(x,y,z,cd,p[0],p[1],p[2]) for p in points]

83
84
85
86
87
88
89
90
91
92
93

plot(R,icd)
pos = atoms.get_positions()
cR = np.linalg.norm(pos[0]-P1)
oR = np.linalg.norm(pos[1]-P1)
plot([cR,cR],[0,2],r-) #markers for where the nuclei are
plot([oR,oR],[0,8],r-)
xlabel(|R| ($\AA$))
ylabel(Charge density (e/$\AA^3$))
savefig(images/interpolated-charge-density.png)
show()

Open the python script (dft-scripts/script-219.py).

None
9.1.12

Reading and writing data

235

Figure 99: Interpolated charge density for a CO molecule.

Built-in io modules pylab has two convenient and powerful functions for saving and reading data,
func:pylab.save and func:pylab.load.
1

pylab.save(pdat.dat,(x,y))

Open the python script (dft-scripts/script-220.py).

and later you can read these arrays back in with:
1

x,y = pylab.load(pdat.dat)

Open the python script (dft-scripts/script-221.py).

see also func:pylab.csv2rec and func:pylab.loadtxt and func:pylab.savetxt.
See http://www.scipy.org/Cookbook/InputOutput for examples of numpy io.
From scratch You can save data in many ways from scratch. Basically, just open a file and write data
to it. Likewise, any datafile that has some structure to it can probably be read by python.
Let us consider a datafile with these contents:
#header
#ignore these lines
john, 4
robert, 5
terry, 5
236

A standard approach would be to read in all the lines, skip the first two lines, split each line (remember
each line is a string) at the ,, and append the first field to one variable, and append the second field to
another variable as an integer. For example:
1
2
3

v1 = []
v2 = []
lines = open(somefile,r).readlines()

4
5
6
7
8

for line in lines[2:]: #skip the first two lines

fields = line.split(,)
v1.append(fields[0]) #names
v2.append(int(fields[1])) #number

Open the python script (dft-scripts/script-222.py).

Writing datafiles is easy too.
1
2
3
4
5
6
7
8

v1 = [john, robert, terry]

v2 = [4,5,6]
f = open(somefile, w) #note w = write mode
f.write(#header\n)
f.write(#ignore these lines\n)
for a,b in zip(v1,v2):
f.write({0}, {1}\n.format(a,b))
f.close()

Open the python script (dft-scripts/script-223.py).

Some notes:
1. opening a file in w mode clobbers any existing file, so do that
with care!
1. when writing to a file you have to add a carriage return to each line.
2. Manually writing and reading files is pretty tedious. Whenever
possible you should use the built-in methods of mod:numpy or mod:pylab.
9.1.13

Integration

Numerical integrations is easy with the numpy.trapz() method. Use it like this: func:numpy.trapz(y,x).
Note that y comes first. y and x must be the same length.
Integration can be used to calculate average properties of continuous distributions. Suppose for
example, we have a density of states, as a function of energy E. We can integrate the density of states
to find the total
R number of states:
Nstates = dE
or, in python:
1

Nstates = np.trapz(rho,E)

Open the python script (dft-scripts/script-224.py).

where rho is a vector that contains the density of states at each energy in the vector E (vector here
means a list of numbers).
The average
energy of distribution is:
R
EdE
Eavg = R
dE

or, in python:
1

e_avg = np.trapz(rho*E,E)/np.trapz(rho,E)

237

Open the python script (dft-scripts/script-225.py).

These last two examples are the zeroth and first moments of the density of states. The second moment
is related to the width squared of the distribution, and the third and fourth moements are related to
skewness and kurtosis of the distribution.
The nth
is defined by:
R moment
E n dE
mn = R
dE

To get the second moment of the density of states in python, we use::

1
2

n = 2
mom_2 = np.trapz(rho*E**n,E)/np.trapz(rho,E)

Open the python script (dft-scripts/script-226.py).

9.1.14

Numerical differentiation

numpy has a function called func:numpy.diff that is similar to the one found in Matlab. It calculates the
differences between the elements in your list, and returns a list that is one element shorter, which makes
it unsuitable for plotting the derivative of a function.
Simple loops to define finite difference derivatives Loops in python are pretty slow (relatively
speaking) but they are usually trivial to understand. In this script we show some simple ways to construct
derivative vectors using loops. It is implied in these formulas that the data points are equally spaced.
1
2
3

import numpy as np
import matplotlib.pyplot as plt
import time

4
5
6
7
8
9

These are the brainless way to calculate numerical derivatives. They

work well for very smooth data. they are surprisingly fast even up to
10000 points in the vector.

10
11
12
13
14
15
16
17
18

x = np.linspace(0.78, 0.79, 100) # 100 points between 0.78 and 0.79

y = np.sin(x)
dy_analytical = np.cos(x)

let us use a forward difference method:

that works up until the last point, where there is not
a forward difference to use. there, we use a backward difference.

19
20
21
22
23
24
25

tf1 = time.time()
dyf = [0.0]*len(x)
for i in range(len(y)-1):
dyf[i] = (y[i+1] - y[i])/(x[i+1]-x[i])
#set last element by backwards difference
dyf[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

26
27

print Forward difference took {0:1.1f} seconds.format(time.time() - tf1)

28
29
30
31
32
33
34
35

and now a backwards difference

tb1 = time.time()
dyb = [0.0]*len(x)
#set first element by forward difference
dyb[0] = (y[0] - y[1])/(x[0] - x[1])
for i in range(1,len(y)):
dyb[i] = (y[i] - y[i-1])/(x[i]-x[i-1])

36
37

print Backward difference took {0:1.1f} seconds.format(time.time() - tb1)

38
39
40
41
42
43

and now, a centered formula

tc1 = time.time()
dyc = [0.0]*len(x)
dyc[0] = (y[0] - y[1])/(x[0] - x[1])
for i in range(1,len(y)-1):

238

44
45

dyc[i] = (y[i+1] - y[i-1])/(x[i+1]-x[i-1])

dyc[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

46
47

print Centered difference took {0:1.1f} seconds.format(time.time() - tc1)

48
49
50
51

the centered formula is the most accurate formula here

52
53
54
55
56

plt.plot(x,dy_analytical,label=analytical derivative)
plt.plot(x,dyf,--,label=forward)
plt.plot(x,dyb,--,label=backward)
plt.plot(x,dyc,--,label=centered)

57
58
59

plt.legend(loc=lower left)
plt.savefig(images/simple-diffs.png)

Open the python script (dft-scripts/script-227.py).

Forward difference took 0.0 seconds
Backward difference took 0.0 seconds
Centered difference took 0.0 seconds
Obviously, all of these evaluations are very fast.

Figure 100: Comparison of different numerical derivatives.

Loops are usually not great for performance. Numpy offers some vectorized methods that allow us
to compute derivatives without loops, although this comes at the mental cost of harder to understand
syntax:

239

1
2

import numpy as np
import matplotlib.pyplot as plt

3
4
5
6

x = np.linspace(0,2*np.pi,100)
y = np.sin(x)
dy_analytical = np.cos(x)

7
8
9
10
11
12

# we need to specify the size of dy ahead because diff returns

#an array of n-1 elements
dy = np.zeros(y.shape,np.float) #we know it will be this size
dy[0:-1] = np.diff(y)/np.diff(x)
dy[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

13
14
15
16
17

calculate dy by center differencing using array slices

18
19
20
21
22

dy2 = np.zeros(y.shape,np.float) #we know it will be this size

dy2[1:-1] = (y[2:] - y[0:-2])/(x[2:] - x[0:-2])
dy2[0] = (y[1]-y[0])/(x[1]-x[0])
dy2[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

23
24
25
26
27
28
29

plt.plot(x,y)
plt.plot(x,dy_analytical,label=analytical derivative)
plt.plot(x,dy,label=forward diff)
plt.plot(x,dy2,k--,lw=2,label=centered diff)
plt.legend(loc=lower left)
plt.savefig(images/vectorized-diffs.png)

Open the python script (dft-scripts/script-228.py).

None
If your data is very noisy, you will have a hard time getting good derivatives; derivatives tend to
magnify noise. In these cases, you have to employ smoothing techniques, either implicitly by using a
multipoint derivative formula, or explicitly by smoothing the data yourself, or taking the derivative of a
function that has been fit to the data in the neighborhood you are interested in.
Here is an example of a 4-point centered difference of some noisy data:
1
2

import numpy as np
import matplotlib.pyplot as plt

3
4
5
6

x = np.linspace(0,2*np.pi,100)
y = np.sin(x) + 0.1*np.random.random(size=x.shape)
dy_analytical = np.cos(x)

7
8
9
10
11
12
13

#2-point formula
dyf = [0.0]*len(x)
for i in range(len(y)-1):
dyf[i] = (y[i+1] - y[i])/(x[i+1]-x[i])
#set last element by backwards difference
dyf[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

14
15
16

calculate dy by 4-point center differencing using array slices

17
18

\frac{y[i-2] - 8y[i-1] + 8[i+1] - y[i+2]}{12h}

19
20
21
22

y[0] and y[1] must be defined by lower order methods

and y[-1] and y[-2] must be defined by lower order methods

23
24
25
26

dy = np.zeros(y.shape,np.float) #we know it will be this size

h = x[1]-x[0] #this assumes the points are evenely spaced!
dy[2:-2] = (y[0:-4] - 8*y[1:-3] + 8*y[3:-1] - y[4:])/(12.*h)

27
28
29
30
31

dy[0] = (y[1]-y[0])/(x[1]-x[0])
dy[1] = (y[2]-y[1])/(x[2]-x[1])
dy[-2] = (y[-2] - y[-3])/(x[-2] - x[-3])
dy[-1] = (y[-1] - y[-2])/(x[-1] - x[-2])

32
33

plt.plot(x,y)

240

Figure 101: Comparison of different numerical derivatives.

34
35
36
37
38

plt.plot(x,dy_analytical,label=analytical derivative)
plt.plot(x,dyf,r-,label=2pt-forward diff)
plt.plot(x,dy,k--,lw=2,label=4pt-centered diff)
plt.legend(loc=lower left)
plt.savefig(images/multipt-diff.png)

Open the python script (dft-scripts/script-229.py).

None
The derivative is still noisy, but the four-point derivative is a little better than the two-pt formula.
FFT derivatives
1
2

It is possible to perform derivatives using fast fourier transforms (FFT):

import numpy as np
import matplotlib.pyplot as plt

3
4
5

N = 101 #number of points

L = 2*np.pi #interval of data

6
7

x = np.arange(0.0,L,L/float(N)) #this does not include the endpoint

8
9
10
11

#add some random noise

y = np.sin(x) + 0.05*np.random.random(size=x.shape)
dy_analytical = np.cos(x)

12
13
14

http://sci.tech-archive.net/Archive/sci.math/2008-05/msg00401.html

15
16

you can use fft to calculate derivatives!

241

Figure 102: Comparison of 2 point and 4 point numerical derivatives.

17

18
19
20
21
22

if N % 2 == 0:
k = np.asarray(range(0,N/2)+[0] + range(-N/2+1,0))
else:
k = np.asarray(range(0,(N-1)/2) +[0] + range(-(N-1)/2,0))

23
24

k *= 2*np.pi/L

25
26

fd = np.fft.ifft(1.j*k * np.fft.fft(y))

27
28
29
30
31

plt.plot(x,y)
plt.plot(x,dy_analytical,label=analytical der)
plt.plot(x,fd,label=fft der)
plt.legend(loc=lower left)

32
33

plt.savefig(images/fft-der.png)

Open the python script (dft-scripts/script-230.py).

This example does not show any major advantage in the quality of the derivative, and it is almost
certain I would never remember how to do this off the top of my head.
9.1.15

NetCDF files

NetCDF is a binary, but cross-platform structured data format. The input file and output file for Dacapo
is the NetCDF format. On creating a NetCDF file you must define the dimensions and variables before
you can store data in them. You can create and read NetCDF files in python using one of the following
modules:
mod:Scientific.IO.NetCDF (http://dirac.cnrs-orleans.fr/plone/software/scientificpython/)
242

Figure 103: Comparison of FFT numerical derivatives.

mod:netCDF3 (http://netcdf4-python.googlecode.com/svn/trunk/docs/netCDF3-module.html)
mod:pycdf (http://pysclint.sourceforge.net/pycdf/) this is a very low level module modelled
after the C-api. I am not sure it is completely bug-free (I have problems with character variables)
9.1.16

Python modules

The comma separated values (mod:csv) module in python allows you to easily create datafiles:
csv writing:
1

import numpy as np

2
3
4

x = np.linspace(0.0,6.0,100)
y = np.cos(x)

5
6
7
8

import csv
writer = csv.writer(open("some.csv", "w"))
writer.writerows(zip(x,y))

Open the python script (dft-scripts/script-231.py).

It is not so easy to read the data back in though because the module only returns strings, so you
must turn the strings back into floats (or whatever other format they should be).
csv reading:
1
2

import csv
reader = csv.reader(open("some.csv",r),delimiter=,)

3
4
5

x,y = [],[]
for row in reader:

243

6
7
8
9

#csv returns strings that must be cast as floats

a,b = [float(z) for z in row]
x.append(a)
y.append(b)

Open the python script (dft-scripts/script-232.py).

This is almost as much work as manually reading the data though. The module is more powerful
than I have shown here, so one day checkout pydoc csv.
The mod:pickle and mod:shelve modules of python also offer some data storage functionality. Check
them out some day too.
9.1.17

Writing and reading Excel files

Writing Excel files It is sometimes convenient to do some analysis in Excel. We can create Excel
files in python with mod:xlwt. Google this module if you need to do this a lot.
1
2

import numpy as np
import xlwt

3
4
5

wbk = xlwt.Workbook()
sheet = wbk.add_sheet(sheet 1)

6
7
8

volumes = np.array([13.72, 14.83, 16.0, 17.23, 18.52])

energies = np.array([-56.29, -56.41, -56.46, -56.46, -56.42])

9
10
11
12
13
14
15

for i, pair in enumerate(zip(volumes, energies)):

vol = pair[0]
energy = pair[1]
sheet.write(i,0,vol)
sheet.write(i,1,energy)
wbk.save(images/test-write.xls)

Open the python script (dft-scripts/script-233.py).

Reading Excel files We can also read Excel files (even on Linux!) with mod:xlrd. Let us read in the
data we just wrote. We wrote 5 volumes to column 0, and 5 energies to column 1.
1
2
3
4
5

import xlrd
wbk = xlrd.open_workbook(images/test-write.xls)
sheet1 = wbk.sheet_by_name(sheet 1)
print sheet1.col_values(0)
print sheet1.col_values(1)

Open the python script (dft-scripts/script-234.py).

[13.72, 14.83, 16.0, 17.23, 18.52]
[-56.29, -56.41, -56.46, -56.46, -56.42]
9.1.18

TODO making movies

1. using animate
2. using swftools (png2swf, pdf2swf)
#http://wiki.swftools.org/wiki/Main_Page#SWF_Tools_0.9.2_.28_Current_Stable_Version_.29_
Documentation

244

9.2

Computational geometry

9.2.1

Changing coordinate systems

Let A, B, C be the unit cell vectors

A = A1x + A2y + A3z

(9)

B = B1x + B2y + B3z

(10)

C = C1x + C2y + C3z

(11)

and we want to find the vector [s1, s2, s3] so that P = s1A + s2B + s3C
if we expand this, we get:

s1A1x + s1A2y + s1A3z

+ s2B1x + s2B2y + s2B3z
+ s3C1x + s3C2y + s3C3z = p1x + p2y + p3z
If we now match coefficients on x, y,

A1
A2
A3

and z, we can

B1 C1
B2 C2
B3 C3

write a set of linear equations as:

p1
s1
s2 = p2
p3
s3

(12)

or, in standard form:

AT s = p
and we need to solve for s as:
s = (AT )1 p
p must be a column vector, so we will have to transpose the positions provided by the atoms class,
and then transpose the final result to get the positions back into row-vector form:
s = ((AT )1 pT )T
Here we implement that in code:
1
2
3

from ase.lattice.surface import fcc111

import numpy as np
np.set_printoptions(precision=3,suppress=True)

4
5
6
7
8

slab = fcc111(Pd,
a=3.92,
# Pd lattice constant
size=(2,2,3), #3-layer slab in 1x1 configuration
vacuum=10.0)

9
10

pos = slab.get_positions() #these positions use x,y,z vectors as a basis

11
12
13

# we want to see the atoms in terms of the unitcell vectors

newbasis = slab.get_cell()

14
15
16

s = np.dot(np.linalg.inv(newbasis.T),pos.T).T
print Coordinates in new basis are: \n,s

17
18
19

# what we just did is equivalent to the following atoms method

print Scaled coordinates from ase are: \n,slab.get_scaled_positions()

Open the python script (dft-scripts/script-235.py).

Coordinates in new basis are:
[[ 0.167 0.167 0.408]
[ 0.667 0.167 0.408]
[ 0.167 0.667 0.408]
245

[ 0.667 0.667 0.408]

[-0.167 0.333 0.5 ]
[ 0.333 0.333 0.5 ]
[-0.167 0.833 0.5 ]
[ 0.333 0.833 0.5 ]
[ 0.
0.
0.592]
[ 0.5
0.
0.592]
[ 0.
0.5
0.592]
[ 0.5
0.5
0.592]]
Scaled coordinates from ase are:
[[ 0.167 0.167 0.408]
[ 0.667 0.167 0.408]
[ 0.167 0.667 0.408]
[ 0.667 0.667 0.408]
[ 0.833 0.333 0.5 ]
[ 0.333 0.333 0.5 ]
[ 0.833 0.833 0.5 ]
[ 0.333 0.833 0.5 ]
[ 0.
0.
0.592]
[ 0.5
0.
0.592]
[ 0.
0.5
0.592]
[ 0.5
0.5
0.592]]
The method shown above is general to all basis set transformations. We examine another case next.
Sometimes it is nice if all the coordinates are integers. For this example, we will use the bcc primitive
lattice vectors and express the positions of each atom in terms of them. By definition each atomic position
should be an integer combination of the primitive lattice vectors (before relaxation, and assuming one
atom is at the origin, and the unit cell is aligned with the primitive basis!)
1
2
3
4
5
6
7
8

from ase.lattice.cubic import BodyCenteredCubic

import numpy as np
bulk = BodyCenteredCubic(directions=[[1,0,0],
[0,1,0],
[0,0,1]],
size=(2,2,2),
latticeconstant=2.87,
symbol=Fe)

9
10
11
12
13

newbasis = 2.87*np.array([[-0.5, 0.5, 0.5],

[0.5, -0.5, 0.5],
[0.5, 0.5, -0.5]])

14
15

pos = bulk.get_positions()

16
17
18
19

s = np.dot(np.linalg.inv(newbasis.T),pos.T).T
print atom positions in primitive basis
print s

20
21
22
23

#let us see the unit cell in terms of the primitive basis too
print unit cell in terms of the primitive basis
print np.dot(np.linalg.inv(newbasis.T),bulk.get_cell().T).T

Open the python script (dft-scripts/script-236.py).

atom positions in primitive basis
[[ 0. 0. 0.]
[ 1. 1. 1.]
[ 0. 1. 1.]
[ 1. 2. 2.]
246

[ 1. 0. 1.]
[ 2. 1. 2.]
[ 1. 1. 2.]
[ 2. 2. 3.]
[ 1. 1. 0.]
[ 2. 2. 1.]
[ 1. 2. 1.]
[ 2. 3. 2.]
[ 2. 1. 1.]
[ 3. 2. 2.]
[ 2. 2. 2.]
[ 3. 3. 3.]]
unit cell in terms of the primitive basis
[[ 0. 2. 2.]
[ 2. 0. 2.]
[ 2. 2. 0.]]
9.2.2

Simple distances, angles

Scientific.Geometry contains several useful functions for performing vector algebra including computing lengths and angles.
1
2

import numpy as np
from Scientific.Geometry import Vector

3
4
5

A = Vector([1,1,1])
#Scientfic
a = np.array([1,1,1]) #numpy

6
7

B = Vector([0.0,1.0,0.0])

8
9
10
11

print |A| = ,A.length()

#Scientific Python way
print |a| = ,np.sum(a**2)**0.5 #numpy way
print |a| = ,np.linalg.norm(a) #numpy way 2

12
13
14

print ScientificPython angle = ,A.angle(B) #in radians

print numpy angle =
,np.arccos(np.dot(a/np.linalg.norm(a),B/np.linalg.norm(B)))

15
16
17
18

#cross products
print Scientific A .cross. B = ,A.cross(B)
print numpy A .cross. B
= ,np.cross(A,B) #you can use Vectors in numpy

Open the python script (dft-scripts/script-237.py).

9.2.3

Unit cell properties

The volume of a unit cell can be calculated from V = (a1 a2 ) a3 where a1 , a2 and a3 are the unit
cell vectors. It is more convenient, however, to simply evaluate that equation as the determinant of the
matrix describing the unit cell, where each row of the matrix is a unit cell vector.
V = | det(ucell)|
Why do we need to take the absolute value? The sign of the determinant depends on the handedness
of the order of the unit cell vectors. If they are right-handed the determinant will be positive, and if
they are left-handed the determinant will be negative. Switching any two rows will change the sign of
the determinant and the handedness. mod:ase implements a convenient function to get the volume of
an mod:Atoms object: func:ase.Atoms.get_volume.
Here are three equivalent ways to compute the unit cell volume.
1

import numpy as np

2
3

a1 = [2, 0, 0]

247

4
5

a2 = [1, 1, 0]
a3 = [0, 0, 10]

6
7

uc = np.array([a1, a2, a3])

8
9
10

print V = {0} ang^3 from dot/cross.format(np.dot(np.cross(a1,a2),a3))

print V = {0} ang^3 from det.format(np.linalg.det(uc))

11
12

from ase import Atoms

13
14
15

atoms = Atoms([],cell=uc) #empty list of atoms

print V = {0} ang^3 from get_volume.format(atoms.get_volume())

Open the python script (dft-scripts/script-238.py).

V = 20 ang^3 from dot/cross
V = 20.0 ang^3 from det
V = 20.0 ang^3 from get_volume
9.2.4

d-spacing

If you like to set up the vacuum in your slab calculations in terms of equivalent layers of atoms, you need
to calculate the d-spacing (which is the spacing between parallel planes of atoms) for the hkl plane you
are using. The script below shows several ways to accomplish that.
1
2

import numpy as np
from ase.lattice.cubic import FaceCenteredCubic

3
4
5
6
7
8
9

ag = FaceCenteredCubic(directions=[[1,0,0],
[0,1,0],
[0,0,1]],
size=(1,1,1),
symbol=Ag,
latticeconstant=4.0)

10
11
12

# these are the reciprocal lattice vectors

b1,b2,b3 = np.linalg.inv(ag.get_cell())

13
14
15

g(111) = 1*b1 + 1*b2 + 1*b3

16
17
18
19
20

and |g(111)| = 1/d_111

h,k,l = (1,1,1)
d = 1./np.linalg.norm(h*b1 + k*b2 + l*b3)

21
22

print d_111 spacing (method 1) = {0:1.3f} Angstroms.format(d)

23
24
25
26

#method #2
hkl = np.array([h,k,l])
G = np.array([b1,b2,b3]) #reciprocal unit cell

27
28
29

Gstar is usually defined as this matrix of dot products:

30
31
32
33

Gstar = np.array([[dot(b1,b1), dot(b1,b2), dot(b1,b3)],

[dot(b1,b2), dot(b2,b2), dot(b2,b3)],
[dot(b1,b3), dot(b2,b3), dot(b3,b3)]])

34
35
36

but I prefer the notationally more compact:

Gstar = G .dot. transpose(G)

37
38
39

then, 1/d_hkl^2 = hkl .dot. Gstar .dot. hkl

40
41

Gstar = np.dot(G,G.T)

42
43

id2 = np.dot(hkl,np.dot(Gstar,hkl))

44
45

print d_111 spacing (method 2) =,np.sqrt(1/id2)

46
47

# http://books.google.com/books?id=nJHSqEseuIUC&lpg=PA118&ots=YA9TBldoVH

248

48
49

# &dq=reciprocal%20metric%20tensor&pg=PA119#v=onepage
# &q=reciprocal%20metric%20tensor&f=false

50
51
52
53
54
55
56
57

Finally, many text books on crystallography use long algebraic

formulas for computing the d-spacing with sin and cos, vector lengths,
and angles. Below we compute these and use them in the general
triclinic structure formula which applies to all the structures.

from Scientific.Geometry import Vector

import math

58
59
60
61
62

unitcell = ag.get_cell()
A = Vector(unitcell[0])
B = Vector(unitcell[1])
C = Vector(unitcell[2])

63
64
65
66
67

#
a
b
c

lengths of the vectors

= A.length()#*angstroms2bohr
= B.length()#*angstroms2bohr
= C.length()#*angstroms2bohr

68
69
70
71
72

# angles between the vectors in radians

alpha = B.angle(C)
beta = A.angle(C)
gamma = A.angle(B)

73
74
75
76
77

print
print a
b
c
alpha beta gamma
print {0:1.3f} {1:1.3f} {2:1.3f} {3:1.3f} {4:1.3f} {5:1.3f}\n.format(a,b,c,
alpha,beta,gamma)

78
79

h,k,l = (1,1,1)

80
81

from math import sin, cos

82
83
84
85
86
87
88
89
90

id2 = ((h**2/a**2*sin(alpha)**2
+ k**2/b**2*sin(beta)**2
+ l**2/c**2*sin(gamma)**2
+2*k*l/b/c*(cos(beta)*cos(gamma)-cos(alpha))
+2*h*l/a/c*(cos(alpha)*cos(gamma)-cos(beta))
+2*h*k/a/b*(cos(alpha)*cos(beta)-cos(gamma)))
/(1-cos(alpha)**2-cos(beta)**2 - cos(gamma)**2
+2*cos(alpha)*cos(beta)*cos(gamma)))

91
92

d = 1/math.sqrt(id2)

93
94

print d_111 spacing (method 3) = {0}.format(d)

Open the python script (dft-scripts/script-239.py).

a
b
c
alpha beta gamma
4.000 4.000 4.000 1.571 1.571 1.571
d_111 spacing (method 3) = 2.30940107676

9.3

Equations of State

The module mod:ase.utils.eos uses a simple polynomial equation of state to find bulk unit cell equilibrium
volumes and bulk modulus. There are several other choices you could use that are more standard in
the literature. Here we summarize them and provide references to the relevant literature. In each of
these cases we show equations for the energy as a function of volume, although sometimes the volume is
transformed or normalized.
9.3.1

Birch-Murnaghan

This is probably the most common equation of state used most often, and is a modification of the original
Murnaghan EOS described below. A current description of the equation is in reference. 100 You can also
find the equations for the Vinet and Poirier-Tarantola equations of state in that reference.
249

Birch-Murnaghan EOS:
0 V0
( 2 1)2 (6 + B00 ( 2 1) 4 2 )
E() = E0 + 9B16
where = (V /V0 )1/3 , B0 and B00 are the bulk modulus and its pressure derivative at the equilibrium
volume V0 . You may find other derivations of this equation in the literature too.
Two other equations of state in that reference are the Vinet EOS:
2B0 V0
0
3(B00 1)(1)/2
E() = E0 + (B
)
0
2 (2 (5 + 3B0 ( 1)e
0 1)
and the Poirier-Tarantola
EOS:
2
E(%) = E0 + B0 V60 % (3 + %(B00 2))
with % = 3 ln().
9.3.2

Murnaghan

The equation mostoften used inthe Murnaghan 101 equation of state is described in 102 .
B0
V0 B0
0 /V ) 0
+1 B
E = ET + BB00V (VB
0 1
0 1
0

where V is the volume, B0 and B00 are the bulk modulus and its pressure derivative at the equilibrium
volume V0 . All of these are parameters that are fitted to energy vs. unit cell volume (V ) data. When
fitting data to this equation a guess of 2-4 for B00 is usually a good start.
9.3.3

Birch

The original Birch equation 103 is:

2


2
E = E0 + 98 B0 V0 VV0 3 1 +

9
0
16 B0 V0 (B0

4)



V
V0

2/3

3
1

The Anton-Schmidt Equation of state 104

 n+1 

BV0
1
V
E(V ) = E + n+1
ln VV0 n+1
V0
where E corresponds to the energy at infinite separation, although the model they use to derive
this equation breaks down at large separations so this is usually not a good estimate of the cohesive
energy. n is typically about -2.
9.3.4

9.3.5

Fitting data to these equations of state

To use these equations of state to find the equilibrium cell volume and bulk modulus we need a set of
calculations that give us the energy of the unit cell as a function of the cell volume. We then fit that
data to one of the above equations to extract the parameters we want. All of these equations of state are
non-linear in the cell volume, which means you have to provide some initial guesses for the parameters.
Here we describe a strategy for getting some estimates of the parameters using a linear least squares
fitting of a parabola to the data to estimate E0 , V0 , B and B00 which are used as initial guess for a
non-linear least squares fit of the equation of state to the data.
The following example illustrates one approach to this problem for the Murnaghan equation of state:
1

Example of fitting the Birch-Murnaghan EOS to data

2
3
4
5

import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import leastsq

6
7
8
9

# raw data from 2.2.3-al-analyze-eos.py

v = np.array([13.72, 14.83, 16.0, 17.23, 18.52])
e = np.array([-56.29, -56.41, -56.46, -56.46, -56.42])

10
11
12

#make a vector to evaluate fits on with a lot of points so it looks smooth

vfit = np.linspace(min(v),max(v),100)

13
14

### fit a parabola to the data

250

15
16

# y = ax^2 + bx + c
a,b,c = np.polyfit(v,e,2) #this is from pylab

17
18
19
20

the parabola does not fit the data very well, but we can use it to get
some analytical guesses for other parameters.

21
22
23
24
25

V0 = minimum energy volume, or where dE/dV=0

E = aV^2 + bV + c
dE/dV = 2aV + b = 0
V0 = -b/2a

26
27
28

E0 is the minimum energy, which is:

E0 = aV0^2 + bV0 + c

29
30

B is equal to V0*d^2E/dV^2, which is just 2a*V0

31
32
33

and from experience we know Bprime_0 is usually a small number like 4

34
35
36
37
38
39

#now
v0 =
e0 =
b0 =
bP =

here are our initial guesses.

-b/(2*a)
a*v0**2 + b*v0 + c
2*a*v0
4

40
41
42
43
44
45
46
47
48
49
50

#now we have to create the equation of state function

def Murnaghan(parameters,vol):

given a vector of parameters and volumes, return a vector of energies.

equation From PRB 28,5480 (1983)

E0 = parameters[0]
B0 = parameters[1]
BP = parameters[2]
V0 = parameters[3]

51
52

E = E0 + B0*vol/BP*(((V0/vol)**BP)/(BP-1)+1) - V0*B0/(BP-1.)

53
54

return E

55
56
57
58
59
60

# and we define an objective function that will be minimized

def objective(pars,y,x):
#we will minimize this function
err = y - Murnaghan(pars,x)
return err

61
62

x0 = [e0, b0, bP, v0] #initial guesses in the same order used in the Murnaghan function

63
64

murnpars, ier = leastsq(objective, x0, args=(e,v)) #this is from scipy

65
66
67
68
69
70
71
72

#now we make a figure summarizing the results

plt.plot(v,e,ro)
plt.plot(vfit, a*vfit**2 + b*vfit + c,--,label=parabolic fit)
plt.plot(vfit, Murnaghan(murnpars,vfit), label=Murnaghan fit)
plt.xlabel(Volume ($\AA^3$))
plt.ylabel(Energy (eV))
plt.legend(loc=best)

73
74
75
76
77
78
79
80
81

#add some text to the figure in figure coordinates

ax = plt.gca()
plt.text(0.4, 0.5, Min volume = {0:1.2f} $\AA^3$.format(murnpars[3]),
transform = ax.transAxes)
plt.text(0.4, 0.4, Bulk modulus = {0:1.2f} eV/$\AA^3$ = {1:1.2f} GPa.format(murnpars[1],
murnpars[1]*160.21773),
transform = ax.transAxes)
plt.savefig(images/a-eos.png)

82
83
84
85

np.set_printoptions(precision=3)
print initial guesses : , np.array(x0) #array for easy printing
print fitted parameters: , murnpars

Open the python script (dft-scripts/script-240.py).

initial guesses :
fitted parameters:

[-56.472
[-56.466

0.631
0.49

4.
4.753
251

16.79 ]
16.573]

Figure 104: Fitted equation of state for bulk data. The initial fitted parabola is shown to illustrate how
it is useful for making initial guesses of the minimum and bulk modulus.
You can see the Murnaghan equation of state fits the data better than the parabola.
Here is a comparison of the initial guesses and final parameters. You can see our guesses from the
parabola were actually pretty good, and are the main reason we converged to a solution. If you try other
guesses you will probably find the scipy.optimize.leastsq function does not converge.

9.4
9.4.1

Miscellaneous jasp/VASP tips

Installing jasp

You need to create an executable script named runvasp.py on your executable path. Here is an example
script that works for both serial and parallel versions of VASP. This script is located in jasp/bin.
1
2

#!/usr/bin/env python
import os

3
4
5

serial_vasp = /home/jkitchin/src/vasp/bin/vasp_serial_intel_mkl
parallel_vasp = /home/jkitchin/src/vasp/bin/vasp_openmpi_intel_mkl

6
7
8

if PBS_NODEFILE in os.environ:
NPROCS = len(open(os.environ[PBS_NODEFILE]).readlines())

9
10
11
12
13
14
15
16
17

if NPROCS == 1:
print NPROCS = ,NPROCS
exitcode = os.system(serial_vasp)
else:
print NPROCS = ,NPROCS
parcmd = mpirun -np %i %s % (NPROCS,parallel_vasp)
exitcode = os.system(parcmd)
else:

252

exitcode = os.system(serial_vasp)

18
19

#end

Open the python script (dft-scripts/script-241.py).

Then create an environment variable that points to this script. For example in .bashrc:
1

export VASP_SCRIPT=$HOME/kitchinpython/jasp/bin/run_vasp.py

Open the python script (dft-scripts/script-242.py).

jasp expects to find the POTCAR potentials in specific directories. You have two options:
1. Install the potentials in a directory and name the potential directories like this:
my_vasp_potentials
potpaw
potpaw_GGA
potpaw_PBE
The potpaw directory contains the LDA potentials, potpaw_GGA contains PW91 potentials, and potpaw_PBE contains the PBE potentials. Then, in your .bashrc file set an environment variable that points to
this directory. For example:
1

export VASP_PP_PATH=$HOME/vasp/my_vasp_potentials

Open the python script (dft-scripts/script-243.py).

If you already have potential directories setup with different names, you can always make symlinks
with the names above that point to your directories.
9.4.2

Using a special setup

VASP provides special setups for some elements. The following guidelines tell you what is in a potential:
No extension means the standard potential. The following extensions mean:
Table 7: Meaning of extensions on POTCAR files for special setups.
extension
_h
_s
_sv
_pv
_d

means the potential is harder than the standard (i.e. needs a higher cutoff energy)
means the potential is softer than the standard (i.e. needs a lower cutoff energy)
s and p semi-core states are treated as valence states
p semi-core states are treated as valence states
d semi-core states are treated as valence states

Here are some links to information in the VASP manual for the setups.
1st row elements
Alkali and alkali-earth metals
d-elements
p-elements
f-elements
Here we show how to select the O_sv potential in a calculation.

253

1
2

from ase import Atoms, Atom

from jasp import *

3
4
5

atoms = Atoms([Atom(O,[5, 5, 5], magmom=1)],

cell=(6, 6, 6))

6
7
8
9
10
11
12
13
14

with jasp(molecules/O_sv,
encut=300,
xc=PBE,
ispin=2,
ismear=0,
sigma=0.001,
setups={O:_sv}, # specifies O_sv potential
atoms=atoms) as calc:

15
16

print Total energy = {0} eV.format(atoms.get_potential_energy())

Open the python script (dft-scripts/script-244.py).

Total energy = -1578.61345 eV
How do you know you got the right special setup? We can look at the first line of the POTCAR file
in the calculation directory to see.
1

head -n 1 molecules/O_sv/POTCAR

Open the python script (dft-scripts/script-245.py).

PAW_PBE O_sv 05Jul2007
This shows we indeed used the O_sv setup.
9.4.3

Running jasp in parallel

jasp is smart. If you ask for more than one node, it will automatically try to run in parallel.
1
2
3
4
5

from jasp import *

JASPRC[queue.nodes]=4
from ase import Atom, Atoms
atoms = Atoms([Atom(O,[5,5,5],magmom=1)],
cell=(6,6,6))

6
7
8
9
10
11
12
13
14

with jasp(molecules/O_sv-4nodes,
encut=300,
xc=PBE,
ispin=2,
ismear=0,
sigma=0.001,
setups={O:_sv}, # specifies O_sv potential
atoms=atoms) as calc:

15
16

print calc.calculate()

Open the python script (dft-scripts/script-246.py).

None
How do you know it ran on four nodes?
1

head molecules/O_sv-4nodes/OUTCAR

Open the python script (dft-scripts/script-247.py).

254

vasp.5.2.12 11Nov11 complex

executed on
running on
4 nodes
distr: one band on

LinuxIFC date 2012.08.31

1 nodes,

09:10:10

4 groups

----------------------------------------------------------------------------------------------------9.4.4

Running multiple instances of jasp in parallel

mod:jasp was designed to enable asynchronous, parallel running processes through a queuing system.
This is ideal for submitting large numbers of independent calculations in one script. The design uses
exceptions to exit the script if the results are not available for subsequent analysis. The design expects
that you run the script often, and the results are analyzed only when they are finally available.
Sometimes it is convenient to run a set of calculations and then wait for them to finish so that a
second set of calculations that depend on the first results can be run. In this scenario, it is inconvenient
to have to rerun your script again after the first set of calculations is done. The challenge is how to tell
the computer to run a set of calculations in parallel, and wait for the calculations to finish. This can be
achieved using the mod:multiprocessing module in python.
The principle idea is to set up the calculations you want to run, and use mod:multiprocessing to
handle running them and waiting for you. To do this, you must instruct mod:jasp to use a "run mode",
and construct a script with a function that runs a calculation, and a section that only runs in the "main"
script.
1
2
3
4
5

import multiprocessing
from jasp import *
from ase import Atom, Atoms
from ase.utils.eos import EquationOfState
import numpy as np

6
7
8
9
10
11
12
13
14
15

# this is the function that runs a calculation

def do_calculation(calculator):
function to run a calculation through multiprocessing
with calculator as calc:
atoms = calc.get_atoms()
e = atoms.get_potential_energy()
v = atoms.get_volume()
return v, e

16
17
18
19

# this only runs in the main script, not in processes on other cores
if __name__ == __main__:
NCORES = 6 # number of cores to run processes on

20
21
22
23
24
25
26
27

# setup an atoms object

a = 3.6
atoms = Atoms([Atom(Cu,(0, 0, 0))],
cell=0.5 * a*np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))
v0 = atoms.get_volume()

28
29
30
31
32
33
34
35
36
37

# Step 1
COUNTER = 0
calculators = [] # list of calculators to be run
factors = [-0.1, 0.05, 0.0, 0.05, 0.1]
for f in factors:
newatoms = atoms.copy()
newatoms.set_volume(v0*(1 + f))
label = bulk/cu-mp/step1-{0}.format(COUNTER)
COUNTER += 1

38
39
40
41
42

calc = jasp(label,
xc=PBE,
encut=350,
kpts=(6,6,6),

255

isym=2,
atoms=newatoms)

43
44
45
46

calculators.append(calc)

47
48
49

# now we set up the Pool of processes

pool = multiprocessing.Pool(processes=NCORES)

50
51
52
53
54

# get the output from running each calculation

out = pool.map(do_calculation, calculators)
pool.close()
pool.join() # this makes the script wait here until all jobs are done

55
56
57
58

# now proceed with analysis

V = [x[0] for x in out]
E = [x[1] for x in out]

59
60
61
62

eos = EquationOfState(V, E)
v1, e1, B = eos.fit()
print step1: v1 = {v1}.format(**locals())

63
64
65
66
67
68
69

### ################################################################
## STEP 2, eos around the minimum
## #################################################################
factors = [-0.06, -0.04, -0.02,
0.0,
0.02, 0.04, 0.06]

70
71
72
73
74
75
76

calculators = [] # reset list

for f in factors:
newatoms = atoms.copy()
newatoms.set_volume(v1*(1 + f))
label = bulk/cu-mp/step2-{0}.format(COUNTER)
COUNTER += 1

77
78
79
80
81
82
83
84

calc = jasp(label,
xc=PBE,
encut=350,
kpts=(6,6,6),
isym=2,
atoms=newatoms)
calculators.append(calc)

85
86

pool = multiprocessing.Pool(processes=NCORES)

87
88
89
90

out = pool.map(do_calculation, calculators)

pool.close()
pool.join() # wait here for calculations to finish

91
92
93
94

# proceed with analysis

V += [x[0] for x in out]
E += [x[1] for x in out]

95
96
97

V = np.array(V)
E = np.array(E)

98
99

f = np.array(V)/v1

100
101
102

# only take points within +- 10% of the minimum

ind = (f >=0.90) & (f <= 1.1)

103
104
105
106
107

eos = EquationOfState(V[ind], E[ind])

v2, e2, B = eos.fit()
print step2: v2 = {v2}.format(**locals())
eos.plot(images/cu-mp-eos.png)

Open the python script (dft-scripts/script-248.py).

step1: v1 = 12.0218897111
step2: v2 = 12.0216094217

256

Figure 105: Equation of state for Cu using the multiprocessing module.

Note:
The first time you run this you will get all the VASP output. The second time you get
the smaller output above.
Also, I have not figured out how to integrate this method with the queue system. At the
moment, the runjasp.py script which ultimately runs VASP will run VASP in parallel, i.e.
one process on multiple nodes/cores instead of a single job that runs multiple processes
simultaneously on multiple nodes/cores.
Here is an example of running this through the queue. The main variations are you must set several
variables in JASPRC that indicate you want to use mod:multiprocessing, and you must save the script
and submit manually to the queue with matching parameters. This is not 100% satisfying, but it is the
best that I have found for now.
1
2
3
4
5
6

#!/usr/bin/env python
import multiprocessing
from jasp import *
from ase import Atom, Atoms
from ase.utils.eos import EquationOfState
import numpy as np

7
8

JASPRC[queue.nodes] = 1

9
10
11
12

# Here we will be able to run three MPI jobs on 2 cores at a time.

JASPRC[queue.ppn] = 6
JASPRC[multiprocessing.cores_per_process] = 2

13
14
15
16
17
18

# to submit this script, save it as cu-mp.py

# qsub -l nodes=1:ppn=6,walltime=10:00:00 cu-mp.py
import os
if PBS_O_WORKDIR in os.environ:
os.chdir(os.environ[PBS_O_WORKDIR])

19
20
21
22

# this is the function that runs a calculation

def do_calculation(calculator):
function to run a calculation through multiprocessing

257

23
24
25
26
27

with calculator as calc:

atoms = calc.get_atoms()
e = atoms.get_potential_energy()
v = atoms.get_volume()
return v, e

28
29
30

# this only runs in the main script, not in processes on other cores
if __name__ == __main__:

31
32
33
34
35
36
37
38

# setup an atoms object

a = 3.6
atoms = Atoms([Atom(Cu,(0, 0, 0))],
cell=0.5 * a*np.array([[1.0, 1.0, 0.0],
[0.0, 1.0, 1.0],
[1.0, 0.0, 1.0]]))
v0 = atoms.get_volume()

39
40
41
42
43
44
45
46
47
48

# Step 1
COUNTER = 0
calculators = [] # list of calculators to be run
factors = [-0.1, 0.05, 0.0, 0.05, 0.1]
for f in factors:
newatoms = atoms.copy()
newatoms.set_volume(v0*(1 + f))
label = bulk/cu-mp2/step1-{0}.format(COUNTER)
COUNTER += 1

49
50
51
52
53
54
55
56

calc = jasp(label,
xc=PBE,
encut=350,
kpts=(6,6,6),
isym=2,
debug=logging.DEBUG,
atoms=newatoms)

57
58

calculators.append(calc)

59
60
61

# now we set up the Pool of processes

pool = multiprocessing.Pool(processes=3) # ask for 6 cores but run MPI on 2 cores

62
63
64
65
66

# get the output from running each calculation

out = pool.map(do_calculation, calculators)
pool.close()
pool.join() # this makes the script wait here until all jobs are done

67
68
69
70

# now proceed with analysis

V = [x[0] for x in out]
E = [x[1] for x in out]

71
72
73
74

eos = EquationOfState(V, E)
v1, e1, B = eos.fit()
print step1: v1 = {v1}.format(**locals())

75
76
77
78
79
80
81

### ################################################################
## STEP 2, eos around the minimum
## #################################################################
factors = [-0.06, -0.04, -0.02,
0.0,
0.02, 0.04, 0.06]

82
83
84
85
86
87
88

calculators = [] # reset list

for f in factors:
newatoms = atoms.copy()
newatoms.set_volume(v1*(1 + f))
label = bulk/cu-mp2/step2-{0}.format(COUNTER)
COUNTER += 1

89
90
91
92
93
94
95
96
97

calc = jasp(label,
xc=PBE,
encut=350,
kpts=(6,6,6),
isym=2,
debug=logging.DEBUG,
atoms=newatoms)
calculators.append(calc)

98
99

pool = multiprocessing.Pool(processes=3)

100

258

101
102
103

out = pool.map(do_calculation, calculators)

pool.close()
pool.join() # wait here for calculations to finish

104
105
106
107

# proceed with analysis

V += [x[0] for x in out]
E += [x[1] for x in out]

108
109
110

V = np.array(V)
E = np.array(E)

111
112

f = np.array(V)/v1

113
114
115

# only take points within +- 10% of the minimum

ind = (f >=0.90) & (f <= 1.1)

116
117
118
119
120

eos = EquationOfState(V[ind], E[ind])

v2, e2, B = eos.fit()
print step2: v2 = {v2}.format(**locals())
eos.plot(images/cu-mp2-eos.png,show=True)

Open the python script (dft-scripts/script-249.py).

step1: v1 = 12.0218897111
step2: v2 = 12.0216189798

Figure 106: Second view of a Cu equation of state computed with multiprocessing.

9.4.5

Exporting data json, xml, python, sqlite

jasp has some capability for representing a calculation result in an archival format. The formats currently
under development are json, xml, python and sqlite. The main point of these methods is to make it easy
to create archive files that are machine readable for supplementary information in publications. These
are under development.

259

python This is code that should reconstruct the python code needed to run a particular calculation.
There are some limitations, e.g. it does not currently get magnetic moments on the atoms.
1

from jasp import *

2
3
4

with jasp(bulk/alloy/cu) as calc:

print calc.python

Open the python script (dft-scripts/script-250.py).

from numpy import array
from ase import Atom, Atoms
from jasp import *
atoms = Atoms([Atom(Cu,[0.0, 0.0, 0.0]),
cell = [[1.818, 0.0, 1.818],
[1.818, 1.818, 0.0],
[0.0, 1.818, 1.818]])
with jasp(bulk/alloy/cu,
nbands = 9,
nsw = 10,
ibrion = 2,
isif = 4,
encut = 350.0,
prec = Normal,
kpts = array([13, 13, 13]),
reciprocal = False,
xc = PBE,
txt = -,
gamma = False,
atoms=atoms) as calc:
# your code here
json
1

from jasp import *

2
3
4

with jasp(bulk/alloy/cu) as calc:

print calc.pretty_json

Open the python script (dft-scripts/script-251.py).

{
"INCAR": {
"addgrid": null,
"aexx": null,
"aggac": null,
"aggax": null,
"aldac": null,
"algo": null,
"amin": null,
"amix": null,
"amix_mag": null,
"bmix": null,
260

"bmix_mag": null,
"ddr": null,
"deper": null,
"dfnmax": null,
"dfnmin": null,
"dipol": null,
"drotmax": null,
"ebreak": null,
"ediff": null,
"ediffg": null,
"eint": null,
"emax": null,
"emin": null,
"enaug": null,
"encut": 350.0,
"encutfock": null,
"encutgw": null,
"falpha": null,
"falphadec": null,
"fdstep": null,
"ferdo": null,
"ferwe": null,
"fnmin": null,
"ftimedec": null,
"ftimeinc": null,
"ftimemax": null,
"gga": null,
"hfscreen": null,
"ialgo": null,
"iband": null,
"ibrion": 2,
"ichain": null,
"icharg": null,
"idipol": null,
"images": null,
"iniwav": null,
"invcurve": null,
"iopt": null,
"isif": 4,
"ismear": null,
"ispin": null,
"istart": null,
"isym": null,
"iwavpr": null,
"jacobian": null,
"kgamma": null,
"kpuse": null,
"kspacing": null,
"laechg": null,
"lasph": null,
"lasync": null,
"lbfgsmem": null,

261

"lcharg": null,
"lclimb": null,
"lcorr": null,
"ldau": null,
"ldau_luj": null,
"ldauj": null,
"ldaul": null,
"ldauprint": null,
"ldautype": null,
"ldauu": null,
"ldiag": null,
"ldipol": null,
"ldneb": null,
"lelf": null,
"lepsilon": null,
"lglobal": null,
"lhfcalc": null,
"llineopt": null,
"lmaxmix": null,
"lnebcell": null,
"loptics": null,
"lorbit": null,
"lpard": null,
"lplane": null,
"lscalapack": null,
"lscalu": null,
"lsepb": null,
"lsepk": null,
"ltangentold": null,
"lthomas": null,
"luse_vdw": null,
"lvdw": null,
"lvhar": null,
"lvtot": null,
"lwave": null,
"magmom": null,
"maxmix": null,
"maxmove": null,
"nbands": 9,
"nblk": null,
"nbmod": null,
"nelect": null,
"nelm": null,
"nelmdl": null,
"nelmin": null,
"nfree": null,
"ngx": null,
"ngxf": null,
"ngy": null,
"ngyf": null,
"ngz": null,
"ngzf": null,

262

"nkred": null,
"nkredx": null,
"nkredy": null,
"nkredz": null,
"nomega": null,
"nomegar": null,
"npar": null,
"nsim": null,
"nsw": 10,
"nupdown": null,
"nwrite": null,
"param1": null,
"param2": null,
"pomass": null,
"potim": null,
"prec": "Normal",
"precfock": null,
"ropt": null,
"rwigs": null,
"sdalpha": null,
"sdr": null,
"sigma": null,
"smass": null,
"snl": null,
"spring": null,
"stol": null,
"symprec": null,
"system": null,
"tebeg": null,
"teend": null,
"time": null,
"timestep": null,
"vdwgr": null,
"vdwrn": null,
"voskown": null,
"weimin": null,
"zab_vdw": null,
"zval": null
},
"atoms": {
"cell": [
[
1.818,
0.0,
1.818
],
[
1.818,
1.818,
0.0
],
[

263

0.0,
1.818,
1.818
]
],
"pbc": [
true,
true,
true
],
"positions": [
[
0.0,
0.0,
0.0
]
],
"symbols": [
"Cu"
],
"tags": [
0
]
},
"input": {
"gamma": false,
"kpts": [
13,
13,
13
],
"kpts_nintersections": null,
"reciprocal": false,
"setups": null,
"txt": "-",
"xc": "PBE"
}
}
xml This relies on the pyxser module.
1

from jasp import *

2
3
4

with jasp(bulk/alloy/cu) as calc:

print calc.xml

Open the python script (dft-scripts/script-252.py).

<?xml version="1.0" encoding="utf-8"?>

<pyxs:obj xmlns:pyxs="http://projects.coder.cl/pyxser/model/" version="1.0" type="vasp" module="jasp.
<pyxs:col type="dict" name="d">
<pyxs:col type="dict" name="INCAR">
<pyxs:prop type="str" name="prec">Normal</pyxs:prop>
<pyxs:prop type="float" name="encut">350.0</pyxs:prop>
264

<pyxs:prop type="int" name="nbands">9</pyxs:prop>

<pyxs:prop type="int" name="isif">4</pyxs:prop>
<pyxs:prop type="int" name="nsw">10</pyxs:prop>
<pyxs:prop type="int" name="ibrion">2</pyxs:prop>
</pyxs:col>
<pyxs:col type="dict" name="input">
<pyxs:col type="list" name="kpts">
<pyxs:prop type="int" name="kpts">13</pyxs:prop>
<pyxs:prop type="int" name="kpts">13</pyxs:prop>
<pyxs:prop type="int" name="kpts">13</pyxs:prop>
</pyxs:col>
<pyxs:prop type="bool" name="reciprocal">False</pyxs:prop>
<pyxs:prop type="str" name="xc">PBE</pyxs:prop>
<pyxs:prop type="str" name="txt">-</pyxs:prop>
<pyxs:prop type="bool" name="gamma">False</pyxs:prop>
</pyxs:col>
<pyxs:col type="dict" name="atoms">
<pyxs:col type="list" name="cell">
<pyxs:col type="list" name="cell">
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
<pyxs:prop type="float" name="cell">0.0</pyxs:prop>
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
</pyxs:col>
<pyxs:col type="list" name="cell">
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
<pyxs:prop type="float" name="cell">0.0</pyxs:prop>
</pyxs:col>
<pyxs:col type="list" name="cell">
<pyxs:prop type="float" name="cell">0.0</pyxs:prop>
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
<pyxs:prop type="float" name="cell">1.818</pyxs:prop>
</pyxs:col>
</pyxs:col>
<pyxs:col type="list" name="symbols">
<pyxs:prop type="str" name="symbols">Cu</pyxs:prop>
</pyxs:col>
<pyxs:col type="list" name="pbc">
<pyxs:prop type="bool" name="pbc">True</pyxs:prop>
<pyxs:prop type="bool" name="pbc">True</pyxs:prop>
<pyxs:prop type="bool" name="pbc">True</pyxs:prop>
</pyxs:col>
<pyxs:col type="list" name="positions">
<pyxs:col type="list" name="positions">
<pyxs:prop type="float" name="positions">0.0</pyxs:prop>
<pyxs:prop type="float" name="positions">0.0</pyxs:prop>
<pyxs:prop type="float" name="positions">0.0</pyxs:prop>
</pyxs:col>
</pyxs:col>
</pyxs:col>
</pyxs:col>
</pyxs:obj>

265

9.4.6

Recommended values for ENCUT and valence electrons for different POTCAR files

The incar:ENCUT tag and incar:PREC tag affect the accuracy/convergence of your calculations.
1
2
3

from jasp.POTCAR import *

from ase.data import chemical_symbols
import glob, os

4
5
6
7
8
9

print
print
print
print
print

#+tblname: POTCAR
#+caption: Parameters for POTPAW_PBE POTCAR files.
#+ATTR_LaTeX: longtable
| POTCAR | ENMIN | ENMAX | prec=high (eV) | # val. elect. |
|-

10
11
12

chemical_symbols.sort()
for symbol in chemical_symbols:

13
14
15

potcars = glob.glob({0}/POTPAW_PBE/{1}*/POTCAR.format(os.environ[VASP_PP_PATH],
symbol))

16
17

for potcar in potcars:

18
19
20
21
22
23
24

POTCAR = os.path.relpath(potcar,
os.environ[VASP_PP_PATH]+/POTPAW_PBE)[:-7]
ENMIN = get_ENMIN(potcar)
ENMAX = get_ENMAX(potcar)
HIGH = 1.3*ENMAX
ZVAL = get_ZVAL(potcar)

25
26

print |{POTCAR:30s}|{ENMIN}|{ENMAX}|{HIGH:1.3f}|{ZVAL}|.format(**locals())

Open the python script (dft-scripts/script-253.py).

266

Table 8: Parameters for POTPAW_PBE POTCAR files.

POTCAR
ENMIN ENMAX prec=high (eV) # val.
Ac
129.178
172.237
223.908
Ag_new
187.383
249.844
324.797
Ag
187.385
249.846
324.800
Ag_pv
223.399
297.865
387.225
Al
180.225
240.3
312.390
Am
191.906
255.875
332.637
Ar
199.795
266.393
346.311
As_d_GW
259.629
346.172
450.024
As_d
216.488
288.651
375.246
As_GW
156.526
208.702
271.313
As
156.51
208.68
271.284
At_d
199.688
266.251
346.126
At
121.073
161.43
209.859
Au_new
172.457
229.943
298.926
Au
172.461
229.948
298.932
Ba_sv
140.408
187.21
243.373
Be
185.658
247.544
321.807
Be_sv
231.563
308.75
401.375
B_h
500.0
700.0
910.000
Bi_d
182.138
242.851
315.706
Bi
78.777
105.037
136.548
Bi_pv
231.89
309.187
401.943
B
238.954
318.606
414.188
Br
162.198
216.264
281.143
B_s
201.934
269.245
350.019
Ba_sv
140.408
187.21
243.373
Be
185.658
247.544
321.807
Be_sv
231.563
308.75
401.375
Bi_d
182.138
242.851
315.706
Bi
78.777
105.037
136.548
Bi_pv
231.89
309.187
401.943
Br
162.198
216.264
281.143
Ca
77.067
102.755
133.582
Ca_pv
89.665
119.554
155.420
Ca_sv
199.939
266.586
346.562
C_d
310.494
413.992
538.190
Cd
205.757
274.342
356.645
Ce_3
135.964
181.286
235.672
Ce_h
224.925
299.9
389.870
Ce
204.781
273.042
354.955
C_GW
310.494
413.992
538.190
C_h_nr
556.263
741.684
964.189
C_h
500.0
700.0
910.000
Cl_h
306.852
409.136
531.877
Cl
196.854
280.0
364.000
Co_new
200.976
267.968
348.358
Co
200.977
267.969
348.360
Co_sv
292.771
390.362
507.471
C
300.0
400.0
520.000
Cr
170.311
227.082
295.207
Cr_pv_new
199.261
265.681
345.385
Cr_pv
199.262
265.683
345.388
267
Cr_sv_new
296.603
395.471
514.112
Cr_sv
296.603
395.471
514.112
C_s
205.426
273.901
356.071
Cs_sv
165.238
220.318
286.413
Cu_f
221.585
295.446
384.080
Cu_new
221.585
295.446
384.080

elect.
11.0
11.0
11.0
17.0
3.0
17.0
8.0
15.0
15.0
5.0
5.0
17.0
7.0
11.0
11.0
10.0
2.0
4.0
3.0
15.0
5.0
21.0
3.0
7.0
3.0
10.0
2.0
4.0
15.0
5.0
21.0
7.0
2.0
8.0
10.0
4.0
12.0
11.0
12.0
12.0
4.0
4.0
4.0
7.0
7.0
9.0
9.0
17.0
4.0
6.0
12.0
12.0
14.0
14.0
4.0
9.0
11.0
11.0

10

Python

10.1

easy_install as a user

easy_install is a python command-line utility that automatically installs python packages. Usually
you need root access to install a python package, but you can also tell easy_install where to install a
package. This usually works if the directory is on your PYTHONPATH
easy_install -d ~/lib/python2.6/site-packages/ pymatgen

10.2

pip as a user

pip is even better than easy_install these days.

1

pip install --user some-package

Open the python script (dft-scripts/script-254.py).

For me this installs here:
~/.local/lib/python2.7/site-packages
That may or may not be on your Python path.

10.3

Integer division math gotchas

It pays to be careful when dividing by integers because you can get unexpected results if you do not
know the integer division rules. In python 2.6, if you divide two integers, you get an integer! This is
usually not a problem if there is no emainder in the division, e.g. 6/3=2. But, if there is a remainder,
and that remainder is important, you will lose it. Here is an example of calculating the mole fraction
of a species from integer numbers of atoms in the unit cell. If you are not careful, you get the wrong
answer! You can convert (also called casting) a number to a float using the float command.
1
2
3
4

nPd = 4
nCu = 5
x_Cu = nCu/(nPd + nCu)
print x_cu = {0} (integer division).format(x_Cu)

5
6
7
8

# now cast as floats

x_Cu = float(nCu)/float(nPd + nCu)
print x_cu = {0} (float division).format(x_Cu)

Open the python script (dft-scripts/script-255.py).

x_cu = 0 (integer division)
x_cu = 0.555555555556 (float division)
Note that if one of the numbers is a float, python will automatically cast the integer as a float, and
return a float.
1
2

nPd = 4
nCu = 5

3
4
5
6

# now cast as floats

x_Cu = float(nCu)/(nPd + nCu)
print x_cu = {0}.format(x_Cu)

Open the python script (dft-scripts/script-256.py).

x_cu = 0.555555555556

268

Finally, you can tell python a number is a float by adding a decimal to it. You do not need to put a
0 after the decimal, but you can.
1
2

nPd = 4.
nCu = 5

# this is a float

3
4
5

x_Cu = nCu/(nPd + nCu)

print x_cu = {0}.format(x_Cu)

Open the python script (dft-scripts/script-257.py).

x_cu = 0.555555555556

11

References

References
[1] David S. Sholl and Janice A. Steckel. Density Functional Theory: A Practical Introduction. Wiley,
2009.
[2] R. G. Parr and W. Yang. Density-Functional Theory of Atoms and Molecules. Oxford Science
Publications, 1989.
[3] W. Koch and M. C. Holthausen. A Chemists Guide to Density Functional Theory. Wiley-VCH,
2 edition, 2001.
[4] Charles Kittel. Introduction to Solid State Physics. Wiley, 8th edition, 2005.
[5] N. W. Ashcroft and N. David Mermin. Solid State Physics. Saunders College Publishing, 1976.
[6] Roald Hoffmann. How chemistry and physics meet in the solid state. Angewandte Chemie International Edition in English, 26(9):846878, 1987. ISSN 1521-3773. doi: 10.1002/anie.198708461.
URL http://dx.doi.org/10.1002/anie.198708461.
[7] Roald Hoffmann. A chemical and theoretical way to look at bonding on surfaces. Rev. Mod. Phys.,
60:601628, Jul 1988. doi: 10.1103/RevModPhys.60.601. URL http://link.aps.org/doi/10.
1103/RevModPhys.60.601.
[8] B. Hammer and J.K. Nrskov. Theoretical surface science and catalysiscalculations and concepts.
In Helmut Knozinger Bruce C. Gates, editor, Impact of Surface Science on Catalysis, volume 45 of
Advances in Catalysis, pages 71 129. Academic Press, 2000. doi: 10.1016/S0360-0564(02)45013-4.
URL http://www.sciencedirect.com/science/article/pii/S0360056402450134.
[9] Jeff Greeley, Jens K. Nrskov, and Manos Mavrikakis. Electronic structure and catalysis on
metal surfaces. Annual Review of Physical Chemistry, 53(1):319348, 2002. doi: 10.1146/
annurev.physchem.53.100301.131630. URL http://www.annualreviews.org/doi/abs/10.1146/
annurev.physchem.53.100301.131630.
[10] A J Freeman and E Wimmer. Density functional theory as a major tool in computational materials
science. Annual Review of Materials Science, 25(1):736, 1995. doi: 10.1146/annurev.ms.25.080195.
000255.
[11] M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias, and J. D. Joannopoulos. Iterative minimization
techniques for ab initio total-energy calculations: molecular dynamics and conjugate gradients. Rev.
Mod. Phys., 64:10451097, Oct 1992. doi: 10.1103/RevModPhys.64.1045. URL http://link.aps.
org/doi/10.1103/RevModPhys.64.1045.

269

[12] G. Kresse and J. Furthmller. Efficiency of ab-initio total energy calculations for metals and
semiconductors using a plane-wave basis set. Computational Materials Science, 6(1):15 50, 1996.
ISSN 0927-0256. doi: 10.1016/0927-0256(96)00008-0. URL http://www.sciencedirect.com/
science/article/pii/0927025696000080.
[13] T. Bligaard. Exchange and Correlation Functionals - a study toward improving the precision of
electron density functional calculations of atomistic systems. Masters thesis, Technical University
of Denmark, 2000. http://www.fysik.dtu.dk/~{}bligaard/masterthesis/masterdirectory/
project/project.pdf.
[14] T. Bligaard. Understanding Materials Properties on the Basis of Density Functional Theory
Calculations. PhD thesis, Technical University of Denmark, 2003. http://www.fysik.dtu.dk/
~{}bligaard/phdthesis/phdproject.pdf.
[15] A. P. Seitsonen. Theoretical Investigations into adsorption and co-adsorption on transition-metal
surfaces as models to heterogeneous catalysis. PhD thesis, Technical University of Berlin, School of
Mathematics and Natural Sciences, 2000. http://edocs.tu-berlin.de/diss/2000/seitsonen_ari.pdf.
[16] R. Hirschl. Binary Transition Metal Alloys and Their Surfaces. PhD thesis, Institut fr Materialphysik, University of Vienna, 2002. http://www.hirschl.at/download/diss_part1.pdf and
http://www.hirschl.at/download/diss_part2.pdf.
[17] L. Pauling and E. B. Wilson, Jr. Introduction to Quantum Mechanics with Applications to Chemistry. Dover Publications, Inc., 1963.
[18] W. Kohn. Nobel lecture: Electronic structure of matter-wave functions and density functionals.
Rev. Mod. Phys., 71:12531266, Oct 1999. doi: 10.1103/RevModPhys.71.1253. URL http://
link.aps.org/doi/10.1103/RevModPhys.71.1253.
[19] P. A. M. Dirac. Quantum mechanics of many-electron systems. Proceedings of the Royal Society
of London. Series A, Containing Papers of a Mathematical and Physical Character, 123(792):pp.
714733, 1929. ISSN 09501207. URL http://www.jstor.org/stable/95222.
[20] P. Hohenberg and W. Kohn. Inhomogeneous electron gas. Phys. Rev., 136:B864B871, Nov 1964.
doi: 10.1103/PhysRev.136.B864. URL http://link.aps.org/doi/10.1103/PhysRev.136.B864.
[21] W. Kohn and L. J. Sham. Self-consistent equations including exchange and correlation effects.
Phys. Rev., 140:A1133A1138, Nov 1965. doi: 10.1103/PhysRev.140.A1133. URL http://link.
aps.org/doi/10.1103/PhysRev.140.A1133.
[22] John A. Pople. Nobel lecture: Quantum chemical models. Rev. Mod. Phys., 71:12671274,
Oct 1999. doi: 10.1103/RevModPhys.71.1267. URL http://link.aps.org/doi/10.1103/
RevModPhys.71.1267.
[23] M. Fuchs, M. Bockstedte, E. Pehlke, and M. Scheffler. Pseudopotential study of binding properties
of solids within generalized gradient approximations: The role of core-valence exchange correlation.
Phys. Rev. B, 57:21342145, Jan 1998. doi: 10.1103/PhysRevB.57.2134. URL http://link.aps.
org/doi/10.1103/PhysRevB.57.2134.
[24] John P. Perdew, Robert G. Parr, Mel Levy, and Jose L. Balduz. Density-functional theory for
fractional particle number: Derivative discontinuities of the energy. Phys. Rev. Lett., 49:1691
1694, Dec 1982. doi: 10.1103/PhysRevLett.49.1691. URL http://link.aps.org/doi/10.1103/
PhysRevLett.49.1691.
[25] B. Hammer, L. B. Hansen, and J. K. Nrskov. Improved adsorption energetics within densityfunctional theory using revised perdew-burke-ernzerhof functionals. Phys. Rev. B, 59:7413
7421, Mar 1999. doi: 10.1103/PhysRevB.59.7413. URL http://link.aps.org/doi/10.1103/
PhysRevB.59.7413.
270

[26] G. Makov and M. C. Payne. Periodic boundary conditions in ab initio calculations. Phys. Rev. B,
51:40144022, Feb 1995. doi: 10.1103/PhysRevB.51.4014. URL http://link.aps.org/doi/10.
1103/PhysRevB.51.4014.
[27] D. J. Chadi and Marvin L. Cohen. Special points in the brillouin zone. Phys. Rev. B, 8:57475753,
Dec 1973. doi: 10.1103/PhysRevB.8.5747. URL http://link.aps.org/doi/10.1103/PhysRevB.
8.5747.
[28] Hendrik J. Monkhorst and James D. Pack. Special points for brillouin-zone integrations. Phys.
Rev. B, 13:51885192, Jun 1976. doi: 10.1103/PhysRevB.13.5188. URL http://link.aps.org/
doi/10.1103/PhysRevB.13.5188.
[29] N. Troullier and Jos Luriaas Martins. Efficient pseudopotentials for plane-wave calculations. Phys.
Rev. B, 43:19932006, Jan 1991. doi: 10.1103/PhysRevB.43.1993. URL http://link.aps.org/
doi/10.1103/PhysRevB.43.1993.
[30] David Vanderbilt. Soft self-consistent pseudopotentials in a generalized eigenvalue formalism. Phys.
Rev. B, 41:78927895, Apr 1990. doi: 10.1103/PhysRevB.41.7892. URL http://link.aps.org/
doi/10.1103/PhysRevB.41.7892.
[31] E. G. Moroni, G. Kresse, J. Hafner, and J. Furthmller. Ultrasoft pseudopotentials applied to
magnetic fe, co, and ni: From atoms to solids. Phys. Rev. B, 56:1562915646, Dec 1997. doi:
10.1103/PhysRevB.56.15629. URL http://link.aps.org/doi/10.1103/PhysRevB.56.15629.
[32] P. E. Blchl. Projector augmented-wave method. Phys. Rev. B, 50:1795317979, Dec 1994. doi:
10.1103/PhysRevB.50.17953. URL http://link.aps.org/doi/10.1103/PhysRevB.50.17953.
[33] G. Kresse and D. Joubert. From ultrasoft pseudopotentials to the projector augmented-wave
method. Phys. Rev. B, 59:17581775, Jan 1999. doi: 10.1103/PhysRevB.59.1758. URL http:
//link.aps.org/doi/10.1103/PhysRevB.59.1758.
[34] M J Gillan. Calculation of the vacancy formation energy in aluminium. Journal of Physics:
Condensed Matter, 1(4):689, 1989. URL http://stacks.iop.org/0953-8984/1/i=4/a=005.
[35] N. David Mermin. Thermal properties of the inhomogeneous electron gas. Phys. Rev., 137:A1441
A1443, Mar 1965. doi: 10.1103/PhysRev.137.A1441. URL http://link.aps.org/doi/10.1103/
PhysRev.137.A1441.
[36] J. J. Mortensen, L. B. Hansen, and K. W. Jacobsen. Real-space grid implementation of the projector
augmented wave method. Phys. Rev. B, 71:035109, Jan 2005. doi: 10.1103/PhysRevB.71.035109.
URL http://link.aps.org/doi/10.1103/PhysRevB.71.035109.
[37] G. Kresse and J. Furthmller. Efficient iterative schemes for ab initio total-energy calculations using
a plane-wave basis set. Phys. Rev. B, 54:1116911186, Oct 1996. doi: 10.1103/PhysRevB.54.11169.
URL http://link.aps.org/doi/10.1103/PhysRevB.54.11169.
[38] G. Kresse and J. Hafner. Ab initio molecular-dynamics simulation of the liquid-metal/amorphoussemiconductor transition in germanium. Phys. Rev. B, 49:1425114269, May 1994. doi: 10.1103/
PhysRevB.49.14251. URL http://link.aps.org/doi/10.1103/PhysRevB.49.14251.
[39] G. Kresse and J. Hafner. Ab initio molecular dynamics for liquid metals. Phys. Rev. B, 47:558561,
Jan 1993. doi: 10.1103/PhysRevB.47.558. URL http://link.aps.org/doi/10.1103/PhysRevB.
47.558.
[40] Joachim Paier, Robin Hirschl, Martijn Marsman, and Georg Kresse. The PerdewBurkeErnzerhof
exchange-correlation functional applied to the G2-1 test set using a plane-wave basis set. The
Journal of Chemical Physics, 122(23):234102, 2005. doi: 10.1063/1.1926272. URL http://link.
aip.org/link/?JCP/122/234102/1.
271

[41] Larry A. Curtiss, Krishnan Raghavachari, Paul C. Redfern, and John A. Pople. Assessment of
gaussian-2 and density functional theories for the computation of enthalpies of formation. The
Journal of Chemical Physics, 106(3):10631079, 1997. doi: 10.1063/1.473182. URL http://link.
aip.org/link/?JCP/106/1063/1.
[42] Graeme Henkelman, Andri Arnaldsson, and Hannes Jnsson. A fast and robust algorithm for
bader decomposition of charge density. Computational Materials Science, 36(3):354 360, 2006.
ISSN 0927-0256. doi: 10.1016/j.commatsci.2005.04.010. URL http://www.sciencedirect.com/
science/article/pii/S0927025605001849.
[43] Thomas A. Manz and David S. Sholl. Chemically meaningful atomic charges that reproduce the
electrostatic potential in periodic and nonperiodic materials. Journal of Chemical Theory and
Computation, 6(8):24552468, 2010. doi: 10.1021/ct100125x. URL http://pubs.acs.org/doi/
abs/10.1021/ct100125x.
[44] Jr. E. Bright Wilson, J.C. Decius, and Paul C. Cross. Molecular Vibrations: The Theory of Infrared
and Raman Vibrational Spectra. Dover Publications, 1955.
[45] Paolo Giannozzi and Stefano Baroni. Vibrational and dielectric properties of c6 0 from densityfunctional perturbation theory. The Journal of Chemical Physics, 100(11):85378539, 1994. doi:
10.1063/1.466753. URL http://link.aip.org/link/?JCP/100/8537/1.
[46] David Karhnek, Tom Buko, and Jrgen Hafner. A density-functional study of the adsorption
of methane-thiol on the (111) surfaces of the Ni-group metals: II. vibrational spectroscopy. Journal
of Physics: Condensed Matter, 22(26):265006, 2010. URL http://stacks.iop.org/0953-8984/
22/i=26/a=265006.
[47] Antonio Fernndez-Ramos, Benjamin Ellingson, Rubn Meana-Paeda, Jorge Marques, and Donald Truhlar. Symmetry numbers and chemical reaction rates. Theoretical Chemistry Accounts:
Theory, Computation, and Modeling (Theoretica Chimica Acta), 118:813826, 2007. ISSN 1432881X. URL http://dx.doi.org/10.1007/s00214-007-0328-0. 10.1007/s00214-007-0328-0.
[48] Daniel Sheppard, Rye Terrell, and Graeme Henkelman. Optimization methods for finding minimum
energy paths. The Journal of Chemical Physics, 128(13):134106, 2008. doi: 10.1063/1.2841941.
URL http://link.aip.org/link/?JCP/128/134106/1.
[49] R. A. Olsen, G. J. Kroes, G. Henkelman, A. Arnaldsson, and H. Jonsson. Comparison of methods
for finding saddle points without knowledge of the final states. The Journal of Chemical Physics,
121(20):97769792, 2004. doi: 10.1063/1.1809574. URL http://link.aip.org/link/?JCP/121/
9776/1.
[50] Ann E. Mattsson, Rickard Armiento, Joachim Paier, Georg Kresse, John M. Wills, and Thomas R.
Mattsson. The am05 density functional applied to solids. The Journal of Chemical Physics, 128(8):
084714, 2008. doi: 10.1063/1.2835596. URL http://link.aip.org/link/?JCP/128/084714/1.
[51] Yvon Le Page and Paul Saxe. Symmetry-general least-squares extraction of elastic data for strained
materials from ab initio calculations of stress. Phys. Rev. B, 65:104104, Feb 2002. doi: 10.1103/
PhysRevB.65.104104. URL http://link.aps.org/doi/10.1103/PhysRevB.65.104104.
[52] Shun-Li Shang, Yi Wang, DongEung Kim, and Zi-Kui Liu. First-principles thermodynamics from
phonon and debye model: Application to ni and ni3al. Computational Materials Science, 47(4):
1040 1048, 2010. ISSN 0927-0256. doi: 10.1016/j.commatsci.2009.12.006. URL http://www.
sciencedirect.com/science/article/pii/S0927025609004558.
[53] A. van de Walle, M. Asta, and G. Ceder. The alloy theoretic automated toolkit: A user guide.
Calphad, 26(4):539 553, 2002. ISSN 0364-5916. doi: 10.1016/S0364-5916(02)80006-2. URL
http://www.sciencedirect.com/science/article/pii/S0364591602800062.
272

[54] Axel van de Walle. Multicomponent multisublattice alloys, nonconfigurational entropy and other
additions to the alloy theoretic automated toolkit. Calphad, 33(2):266 278, 2009. ISSN 0364-5916.
doi: 10.1016/j.calphad.2008.12.005. URL http://www.sciencedirect.com/science/article/
pii/S0364591608001314. <ce:title>Tools for Computational Thermodynamics</ce:title>.
[55] Lei Wang, Thomas Maxisch, and Gerbrand Ceder. Oxidation energies of transition metal oxides
within the GGA + U framework. Phys. Rev. B, 73:195107, May 2006. doi: 10.1103/PhysRevB.73.
195107. URL http://link.aps.org/doi/10.1103/PhysRevB.73.195107.
[56] Leonard Kleinman. Significance of the highest occupied Kohn-Sham eigenvalue. Phys. Rev. B, 56:
1204212045, Nov 1997. doi: 10.1103/PhysRevB.56.12042. URL http://link.aps.org/doi/10.
1103/PhysRevB.56.12042.
[57] John P. Perdew and Mel Levy. Comment on "Significance of the highest occupied Kohn-Sham
eigenvalue". Phys. Rev. B, 56:1602116028, Dec 1997. doi: 10.1103/PhysRevB.56.16021. URL
http://link.aps.org/doi/10.1103/PhysRevB.56.16021.
[58] Leonard Kleinman. Reply to "Comment on Significance of the highest occupied Kohn-Sham
eigenvalue ". Phys. Rev. B, 56:1602916030, Dec 1997. doi: 10.1103/PhysRevB.56.16029. URL
http://link.aps.org/doi/10.1103/PhysRevB.56.16029.
[59] Ralf Stowasser and Roald Hoffmann. What do the Kohn-Sham Orbitals and Eigenvalues Mean?
Journal of the American Chemical Society, 121(14):34143420, 1999. doi: 10.1021/ja9826892. URL
http://pubs.acs.org/doi/abs/10.1021/ja9826892.
[60] O. V. Gritsenko and E. J. Baerends. The analog of Koopmans theorem in spin-density functional
theory. The Journal of Chemical Physics, 117(20):91549159, 2002. doi: 10.1063/1.1516800. URL
http://link.aip.org/link/?JCP/117/9154/1.
[61] R. O. Jones and O. Gunnarsson. The density functional formalism, its applications and prospects.
Rev. Mod. Phys., 61:689746, Jul 1989. doi: 10.1103/RevModPhys.61.689. URL http://link.
aps.org/doi/10.1103/RevModPhys.61.689.
[62] J. P. Perdew and Alex Zunger. Self-interaction correction to density-functional approximations for
many-electron systems. Phys. Rev. B, 23:50485079, May 1981. doi: 10.1103/PhysRevB.23.5048.
URL http://link.aps.org/doi/10.1103/PhysRevB.23.5048.
[63] Daniel Sanchez-Portal, Emilio Artacho, and Jose M Soler. Projection of plane-wave calculations
into atomic orbitals. Solid State Communications, 95(10):685 690, 1995. ISSN 0038-1098. doi:
10.1016/0038-1098(95)00341-X. URL http://www.sciencedirect.com/science/article/pii/
003810989500341X.
[64] M. D. Segall, R. Shah, C. J. Pickard, and M. C. Payne. Population analysis of plane-wave electronic
structure calculations of bulk materials. Phys. Rev. B, 54:1631716320, Dec 1996. doi: 10.1103/
PhysRevB.54.16317. URL http://link.aps.org/doi/10.1103/PhysRevB.54.16317.
[65] M. D. Segall, C. J. Pickard, R. Shah, and M. C. Payne. Population analysis in plane wave electronic
structure calculations. Mol. Phys., 89(2):571577, 1996.
[66] A Ruban, B Hammer, P Stoltze, H.L Skriver, and J.K Nrskov. Surface electronic structure
and reactivity of transition and noble metals. Journal of Molecular Catalysis A: Chemical, 115
(3):421 429, 1997. ISSN 1381-1169. doi: 10.1016/S1381-1169(96)00348-2. URL http://www.
sciencedirect.com/science/article/pii/S1381116996003482.
[67] A. Cottrell. Introduction to the Modern Theory of Metals. The Institute of Metals, 1988.
[68] F. Ducastelle. Order and Phase Stability in Alloys. Elsevier Science Publishers, 1991.

273

[69] D. G. Pettifor and A. H. Cottrell, editors. Electon Theory in Alloy Design. The Institute of
Materials, 1992.
[70] Stefano Baroni, Stefano de Gironcoli, Andrea Dal Corso, and Paolo Giannozzi. Phonons and related
crystal properties from density-functional perturbation theory. Rev. Mod. Phys., 73:515562, Jul
2001. doi: 10.1103/RevModPhys.73.515. URL http://link.aps.org/doi/10.1103/RevModPhys.
73.515.
[71] Kyle J. Caspersen and Emily A. Carter. Finding transition states for crystalline solid/solid phase
transformations. Proceedings of the National Academy of Sciences of the United States of America,
102(19):67386743, 2005. doi: 10.1073/pnas.0408127102. URL http://www.pnas.org/content/
102/19/6738.abstract.
[72] Daniel Sheppard, Penghao Xiao, William Chemelewski, Duane D. Johnson, and Graeme Henkelman. A generalized solid-state nudged elastic band method. The Journal of Chemical Physics, 136
(7):074103, 2012. doi: 10.1063/1.3684549. URL http://link.aip.org/link/?JCP/136/074103/
1.
[73] G. Ritz, M. Schmid, P. Varga, A. Borg, and M. Rnning. Pt(100) quasihexagonal reconstruction:
A comparison between scanning tunneling microscopy data and effective medium theory simulation
calculations. Phys. Rev. B, 56:1051810525, Oct 1997. doi: 10.1103/PhysRevB.56.10518. URL
http://link.aps.org/doi/10.1103/PhysRevB.56.10518.
[74] Paula Havu, Volker Blum, Ville Havu, Patrick Rinke, and Matthias Scheffler. Large-scale surface
reconstruction energetics of pt(100) and au(100) by all-electron density functional theory. Phys.
Rev. B, 82:161418, Oct 2010. doi: 10.1103/PhysRevB.82.161418. URL http://link.aps.org/
doi/10.1103/PhysRevB.82.161418.
[75] Wei Chen, David Schmidt, William F. Schneider, and C. Wolverton. First-principles cluster expansion study of missing-row reconstructions of fcc (110) surfaces. Phys. Rev. B, 83:075415, Feb
2011. doi: 10.1103/PhysRevB.83.075415. URL http://link.aps.org/doi/10.1103/PhysRevB.
83.075415.
[76] J. W. M. Frenken, R. L. Krans, J. F. van der Veen, E. Holub-Krappe, and K. Horn. Missing-row
surface reconstruction of ag(110) induced by potassium adsorption. Phys. Rev. Lett., 59:2307
2310, Nov 1987. doi: 10.1103/PhysRevLett.59.2307. URL http://link.aps.org/doi/10.1103/
PhysRevLett.59.2307.
[77] J. C. Boettger. Nonconvergence of surface energies obtained from thin-film calculations. Phys.
Rev. B, 49:1679816800, Jun 1994. doi: 10.1103/PhysRevB.49.16798. URL http://link.aps.
org/doi/10.1103/PhysRevB.49.16798.
[78] J. C. Boettger, John R. Smith, Uwe Birkenheuer, Notker Rsch, S. B. Trickey, John R. Sabin, and
S. Peter Apell. Extracting convergent surface formation energies from slab calculations. Journal
of Physics: Condensed Matter, 10(4):893, 1998. URL http://stacks.iop.org/0953-8984/10/
i=4/a=017.
[79] Carlos Fiolhais, L.M. Almeida, and C. Henriques. Extraction of aluminium surface energies from
slab calculations: perturbative and non-perturbative approaches. Progress in Surface Science, 74
(18):209 217, 2003. ISSN 0079-6816. doi: 10.1016/j.progsurf.2003.08.017. URL http://www.
sciencedirect.com/science/article/pii/S0079681603000777.
[80] Fabien Tran, Robert Laskowski, Peter Blaha, and Karlheinz Schwarz. Performance on molecules,
surfaces, and solids of the Wu-Cohen GGA exchange-correlation energy functional. Phys. Rev. B,
75:115131, Mar 2007. doi: 10.1103/PhysRevB.75.115131. URL http://link.aps.org/doi/10.
1103/PhysRevB.75.115131.

274

[81] Jeong Woo Han, Liwei Li, and David S. Sholl. Density functional theory study of H and CO
adsorption on alkali-promoted Mo2 C surfaces. The Journal of Physical Chemistry C, 115(14):6870
6876, 2011. doi: 10.1021/jp200950a. URL http://pubs.acs.org/doi/abs/10.1021/jp200950a.
[82] Nilay nolu and John R. Kitchin. Atomistic thermodynamics study of the adsorption and the
effects of watergas shift reactants on cu catalysts under reaction conditions. Journal of Catalysis,
261(2):188 194, 2009. ISSN 0021-9517. doi: 10.1016/j.jcat.2008.11.020. URL http://www.
sciencedirect.com/science/article/pii/S0021951708004314.
[83] L. Vitos, A.V. Ruban, H.L. Skriver, and J. Kollr. The surface energy of metals. Surface Science,
411(12):186 202, 1998. ISSN 0039-6028. doi: 10.1016/S0039-6028(98)00363-X. URL http:
//www.sciencedirect.com/science/article/pii/S003960289800363X.
[84] Jrg Neugebauer and Matthias Scheffler. Adsorbate-substrate and adsorbate-adsorbate interactions
of Na and K adlayers on Al(111). Phys. Rev. B, 46:1606716080, Dec 1992. doi: 10.1103/PhysRevB.
46.16067. URL http://link.aps.org/doi/10.1103/PhysRevB.46.16067.
[85] Lennart Bengtsson. Dipole correction for surface supercell calculations. Phys. Rev. B, 59:12301
12304, May 1999. doi: 10.1103/PhysRevB.59.12301. URL http://link.aps.org/doi/10.1103/
PhysRevB.59.12301.
[86] Yoshitada Morikawa. Adsorption geometries and vibrational modes of C2 H2 on the si(001) surface.
Phys. Rev. B, 63:033405, Jan 2001. doi: 10.1103/PhysRevB.63.033405. URL http://link.aps.
org/doi/10.1103/PhysRevB.63.033405.
[87] Nilay nolu and John R. Kitchin. Simple model explaining and predicting coverage-dependent
atomic adsorption energies on transition metal surfaces. Phys. Rev. B, 82:045414, Jul 2010. doi:
10.1103/PhysRevB.82.045414. URL http://link.aps.org/doi/10.1103/PhysRevB.82.045414.
[88] Spencer D. Miller, Nilay Inoglu, and John R. Kitchin. Configurational correlations in the coverage
dependent adsorption energies of oxygen atoms on late transition metal fcc(111) surfaces. The
Journal of Chemical Physics, 134(10):104709, 2011. doi: 10.1063/1.3561287. URL http://link.
aip.org/link/?JCP/134/104709/1.
[89] Spencer D. Miller and John R. Kitchin. Relating the coverage dependence of oxygen adsorption
on Au and Pt fcc(111) surfaces through adsorbate-induced surface electronic structure effects.
Surface Science, 603(5):794 801, 2009. ISSN 0039-6028. doi: 10.1016/j.susc.2009.01.021. URL
http://www.sciencedirect.com/science/article/pii/S0039602809001186.
[90] John R. Kitchin. Correlations in coverage-dependent atomic adsorption energies on pd(111). Phys.
Rev. B, 79:205412, May 2009. doi: 10.1103/PhysRevB.79.205412. URL http://link.aps.org/
doi/10.1103/PhysRevB.79.205412.
[91] John R. Kitchin, Karsten Reuter, and Matthias Scheffler. Alloy surface segregation in reactive environments: First-principles atomistic thermodynamics study of Ag3 Pd(111) in oxygen
atmospheres. Phys. Rev. B, 77:075437, Feb 2008. doi: 10.1103/PhysRevB.77.075437. URL
http://link.aps.org/doi/10.1103/PhysRevB.77.075437.
[92] Anand Udaykumar Nilekar, Jeff Greeley, and Manos Mavrikakis. A simple rule of thumb for
diffusion on transition-metal surfaces. Angewandte Chemie International Edition, 45(42):7046
7049, 2006. ISSN 1521-3773. doi: 10.1002/anie.200602223. URL http://dx.doi.org/10.1002/
anie.200602223.
[93] Graeme Henkelman, Blas P. Uberuaga, and Hannes Jonsson. A climbing image nudged elastic band
method for finding saddle points and minimum energy paths. The Journal of Chemical Physics,
113(22):99019904, 2000. doi: 10.1063/1.1329672. URL http://link.aip.org/link/?JCP/113/
9901/1.
275

[94] P. Deshlahra, E. E. Wolf, and W. F. Schneider. A periodic density functional theory analysis of
co chemisorption on pt(111) in the presence of uniform electric fields. The Journal of Physical
Chemistry A, 113(16):41254133, 2009. doi: 10.1021/jp810518x. URL http://pubs.acs.org/
doi/abs/10.1021/jp810518x.
[95] B. Meredig, A. Thompson, H. A. Hansen, C. Wolverton, and A. van de Walle. Method for locating
low-energy solutions within DFT + u. Phys. Rev. B, 82:195128, Nov 2010. doi: 10.1103/PhysRevB.
82.195128. URL http://link.aps.org/doi/10.1103/PhysRevB.82.195128.
[96] Anubhav Jain, Geoffroy Hautier, Shyue Ping Ong, Charles J. Moore, Christopher C. Fischer,
Kristin A. Persson, and Gerbrand Ceder. Formation enthalpies by mixing GGA and GGA +
U calculations. Phys. Rev. B, 84:045115, Jul 2011. doi: 10.1103/PhysRevB.84.045115. URL
http://link.aps.org/doi/10.1103/PhysRevB.84.045115.
[97] Kyuho Lee, amonn D. Murray, Lingzhu Kong, Bengt I. Lundqvist, and David C. Langreth.
Higher-accuracy van der waals density functional. Phys. Rev. B, 82:081101, Aug 2010. doi: 10.
1103/PhysRevB.82.081101. URL http://link.aps.org/doi/10.1103/PhysRevB.82.081101.
[98] Stefan Grimme. Semiempirical gga-type density functional constructed with a long-range dispersion
correction. Journal of Computational Chemistry, 27(15):17871799, 2006. ISSN 1096-987X. doi:
10.1002/jcc.20495. URL http://dx.doi.org/10.1002/jcc.20495.
[99] B. Silvi and A Savin. Classification of chemical bonds based on topological analysis of electron
localization functions. Nature, 371:683686, 1994. URL http://dx.doi.org/10.1038/371683a0.
[100] M. Hebbache and M. Zemzemi. Ab initio study of high-pressure behavior of a low compressibility
metal and a hard material: osmium and diamond. Phys. Rev. B, 70:224107, Dec 2004. doi:
10.1103/PhysRevB.70.224107. URL http://link.aps.org/doi/10.1103/PhysRevB.70.224107.
[101] F. D. Murnaghan. The compressibility of media under extreme pressures. Proceedings of the
National Academy of Sciences of the United States of America, 30(9):pp. 244247, 1944. ISSN
00278424. URL http://www.jstor.org/stable/87468.
[102] C. L. Fu and K. M. Ho. First-principles calculation of the equilibrium ground-state properties
of transition metals: Applications to Nb and Mo. Phys. Rev. B, 28:54805486, Nov 1983. doi:
10.1103/PhysRevB.28.5480. URL http://link.aps.org/doi/10.1103/PhysRevB.28.5480.
[103] Michael J. Mehl, Barry M. Klein, and Dimitri A. Papaconstantopoulos. Intermetallic Compounds:
Principles and Principles, Volume I: Principles, volume I, chapter First principles calculations of
elastic properties of metals, pages 195210. John Wiley and Sons, 1995. URL http://cst-www.
nrl.navy.mil/users/mehl/papers/cij453.pdf.
[104] B. Mayer, H. Anton, E. Bott, M. Methfessel, J. Sticht, J. Harris, and P.C. Schmidt. Ab-initio
calculation of the elastic constants and thermal expansion coefficients of laves phases. Intermetallics, 11(1):23 32, 2003. ISSN 0966-9795. doi: 10.1016/S0966-9795(02)00127-9. URL
http://www.sciencedirect.com/science/article/pii/S0966979502001279.

12

GNU Free Documentation License

GNU Free Documentation License
Version 1.3, 3 November 2008

Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.
<http://fsf.org/>
276

Everyone is permitted to copy and distribute verbatim copies

of this license document, but changing it is not allowed.
0. PREAMBLE
The purpose of this License is to make a manual, textbook, or other
functional and useful document "free" in the sense of freedom: to
assure everyone the effective freedom to copy and redistribute it,
with or without modifying it, either commercially or noncommercially.
Secondarily, this License preserves for the author and publisher a way
to get credit for their work, while not being considered responsible
for modifications made by others.
This License is a kind of "copyleft", which means that derivative
works of the document must themselves be free in the same sense. It
complements the GNU General Public License, which is a copyleft
license designed for free software.
We have designed this License in order to use it for manuals for free
software, because free software needs free documentation: a free
program should come with manuals providing the same freedoms that the
software does. But this License is not limited to software manuals;
it can be used for any textual work, regardless of subject matter or
whether it is published as a printed book. We recommend this License
principally for works whose purpose is instruction or reference.

1. APPLICABILITY AND DEFINITIONS

This License applies to any manual or other work, in any medium, that
contains a notice placed by the copyright holder saying it can be
distributed under the terms of this License. Such a notice grants a
world-wide, royalty-free license, unlimited in duration, to use that
work under the conditions stated herein. The "Document", below,
refers to any such manual or work. Any member of the public is a
licensee, and is addressed as "you". You accept the license if you
copy, modify or distribute the work in a way requiring permission
under copyright law.
A "Modified Version" of the Document means any work containing the
Document or a portion of it, either copied verbatim, or with
modifications and/or translated into another language.
A "Secondary Section" is a named appendix or a front-matter section of
the Document that deals exclusively with the relationship of the
publishers or authors of the Document to the Documents overall
subject (or to related matters) and contains nothing that could fall
directly within that overall subject. (Thus, if the Document is in
part a textbook of mathematics, a Secondary Section may not explain
any mathematics.) The relationship could be a matter of historical
connection with the subject or with related matters, or of legal,
commercial, philosophical, ethical or political position regarding

277

them.
The "Invariant Sections" are certain Secondary Sections whose titles
are designated, as being those of Invariant Sections, in the notice
that says that the Document is released under this License. If a
section does not fit the above definition of Secondary then it is not
allowed to be designated as Invariant. The Document may contain zero
Invariant Sections. If the Document does not identify any Invariant
Sections then there are none.
The "Cover Texts" are certain short passages of text that are listed,
as Front-Cover Texts or Back-Cover Texts, in the notice that says that
the Document is released under this License. A Front-Cover Text may
be at most 5 words, and a Back-Cover Text may be at most 25 words.
A "Transparent" copy of the Document means a machine-readable copy,
represented in a format whose specification is available to the
general public, that is suitable for revising the document
straightforwardly with generic text editors or (for images composed of
pixels) generic paint programs or (for drawings) some widely available
drawing editor, and that is suitable for input to text formatters or
for automatic translation to a variety of formats suitable for input
to text formatters. A copy made in an otherwise Transparent file
format whose markup, or absence of markup, has been arranged to thwart
or discourage subsequent modification by readers is not Transparent.
An image format is not Transparent if used for any substantial amount
of text. A copy that is not "Transparent" is called "Opaque".
Examples of suitable formats for Transparent copies include plain
ASCII without markup, Texinfo input format, LaTeX input format, SGML
or XML using a publicly available DTD, and standard-conforming simple
HTML, PostScript or PDF designed for human modification. Examples of
transparent image formats include PNG, XCF and JPG. Opaque formats
include proprietary formats that can be read and edited only by
proprietary word processors, SGML or XML for which the DTD and/or
processing tools are not generally available, and the
machine-generated HTML, PostScript or PDF produced by some word
processors for output purposes only.
The "Title Page" means, for a printed book, the title page itself,
plus such following pages as are needed to hold, legibly, the material
this License requires to appear in the title page. For works in
formats which do not have any title page as such, "Title Page" means
the text near the most prominent appearance of the works title,
preceding the beginning of the body of the text.
The "publisher" means any person or entity that distributes copies of
the Document to the public.
A section "Entitled XYZ" means a named subunit of the Document whose
title either is precisely XYZ or contains XYZ in parentheses following
text that translates XYZ in another language. (Here XYZ stands for a

278

specific section name mentioned below, such as "Acknowledgements",

"Dedications", "Endorsements", or "History".) To "Preserve the Title"
of such a section when you modify the Document means that it remains a
section "Entitled XYZ" according to this definition.
The Document may include Warranty Disclaimers next to the notice which
states that this License applies to the Document. These Warranty
Disclaimers are considered to be included by reference in this
License, but only as regards disclaiming warranties: any other
implication that these Warranty Disclaimers may have is void and has
no effect on the meaning of this License.
2. VERBATIM COPYING
You may copy and distribute the Document in any medium, either
commercially or noncommercially, provided that this License, the
copyright notices, and the license notice saying this License applies
to the Document are reproduced in all copies, and that you add no
other conditions whatsoever to those of this License. You may not use
technical measures to obstruct or control the reading or further
copying of the copies you make or distribute. However, you may accept
compensation in exchange for copies. If you distribute a large enough
number of copies you must also follow the conditions in section 3.
You may also lend copies, under the same conditions stated above, and
you may publicly display copies.

3. COPYING IN QUANTITY
If you publish printed copies (or copies in media that commonly have
printed covers) of the Document, numbering more than 100, and the
Documents license notice requires Cover Texts, you must enclose the
copies in covers that carry, clearly and legibly, all these Cover
Texts: Front-Cover Texts on the front cover, and Back-Cover Texts on
the back cover. Both covers must also clearly and legibly identify
you as the publisher of these copies. The front cover must present
the full title with all words of the title equally prominent and
visible. You may add other material on the covers in addition.
Copying with changes limited to the covers, as long as they preserve
the title of the Document and satisfy these conditions, can be treated
as verbatim copying in other respects.
If the required texts for either cover are too voluminous to fit
legibly, you should put the first ones listed (as many as fit
reasonably) on the actual cover, and continue the rest onto adjacent
pages.
If you publish or distribute Opaque copies of the Document numbering
more than 100, you must either include a machine-readable Transparent
copy along with each Opaque copy, or state in or with each Opaque copy
a computer-network location from which the general network-using

279

public has access to download using public-standard network protocols

a complete Transparent copy of the Document, free of added material.
If you use the latter option, you must take reasonably prudent steps,
when you begin distribution of Opaque copies in quantity, to ensure
that this Transparent copy will remain thus accessible at the stated
location until at least one year after the last time you distribute an
Opaque copy (directly or through your agents or retailers) of that
edition to the public.
It is requested, but not required, that you contact the authors of the
Document well before redistributing any large number of copies, to
give them a chance to provide you with an updated version of the
Document.

4. MODIFICATIONS
You may copy and distribute a Modified Version of the Document under
the conditions of sections 2 and 3 above, provided that you release
the Modified Version under precisely this License, with the Modified
Version filling the role of the Document, thus licensing distribution
and modification of the Modified Version to whoever possesses a copy
of it. In addition, you must do these things in the Modified Version:
A. Use in the Title Page (and on the covers, if any) a title distinct
from that of the Document, and from those of previous versions
(which should, if there were any, be listed in the History section
of the Document). You may use the same title as a previous version
if the original publisher of that version gives permission.
B. List on the Title Page, as authors, one or more persons or entities
responsible for authorship of the modifications in the Modified
Version, together with at least five of the principal authors of the
Document (all of its principal authors, if it has fewer than five),
unless they release you from this requirement.
C. State on the Title page the name of the publisher of the
Modified Version, as the publisher.
D. Preserve all the copyright notices of the Document.
E. Add an appropriate copyright notice for your modifications
adjacent to the other copyright notices.
F. Include, immediately after the copyright notices, a license notice
giving the public permission to use the Modified Version under the
terms of this License, in the form shown in the Addendum below.
G. Preserve in that license notice the full lists of Invariant Sections
and required Cover Texts given in the Documents license notice.
H. Include an unaltered copy of this License.
I. Preserve the section Entitled "History", Preserve its Title, and add
to it an item stating at least the title, year, new authors, and
publisher of the Modified Version as given on the Title Page. If
there is no section Entitled "History" in the Document, create one
stating the title, year, authors, and publisher of the Document as
given on its Title Page, then add an item describing the Modified
Version as stated in the previous sentence.

280

J. Preserve the network location, if any, given in the Document for

public access to a Transparent copy of the Document, and likewise
the network locations given in the Document for previous versions
it was based on. These may be placed in the "History" section.
You may omit a network location for a work that was published at
least four years before the Document itself, or if the original
publisher of the version it refers to gives permission.
K. For any section Entitled "Acknowledgements" or "Dedications",
Preserve the Title of the section, and preserve in the section all
the substance and tone of each of the contributor acknowledgements
and/or dedications given therein.
L. Preserve all the Invariant Sections of the Document,
unaltered in their text and in their titles. Section numbers
or the equivalent are not considered part of the section titles.
M. Delete any section Entitled "Endorsements". Such a section
may not be included in the Modified Version.
N. Do not retitle any existing section to be Entitled "Endorsements"
or to conflict in title with any Invariant Section.
O. Preserve any Warranty Disclaimers.
If the Modified Version includes new front-matter sections or
appendices that qualify as Secondary Sections and contain no material
copied from the Document, you may at your option designate some or all
of these sections as invariant. To do this, add their titles to the
list of Invariant Sections in the Modified Versions license notice.
These titles must be distinct from any other section titles.
You may add a section Entitled "Endorsements", provided it contains
nothing but endorsements of your Modified Version by various
parties--for example, statements of peer review or that the text has
been approved by an organization as the authoritative definition of a
standard.
You may add a passage of up to five words as a Front-Cover Text, and a
passage of up to 25 words as a Back-Cover Text, to the end of the list
of Cover Texts in the Modified Version. Only one passage of
Front-Cover Text and one of Back-Cover Text may be added by (or
through arrangements made by) any one entity. If the Document already
includes a cover text for the same cover, previously added by you or
by arrangement made by the same entity you are acting on behalf of,
you may not add another; but you may replace the old one, on explicit
permission from the previous publisher that added the old one.
The author(s) and publisher(s) of the Document do not by this License
give permission to use their names for publicity for or to assert or
imply endorsement of any Modified Version.

5. COMBINING DOCUMENTS
You may combine the Document with other documents released under this
License, under the terms defined in section 4 above for modified

281

versions, provided that you include in the combination all of the

Invariant Sections of all of the original documents, unmodified, and
list them all as Invariant Sections of your combined work in its
license notice, and that you preserve all their Warranty Disclaimers.
The combined work need only contain one copy of this License, and
multiple identical Invariant Sections may be replaced with a single
copy. If there are multiple Invariant Sections with the same name but
different contents, make the title of each such section unique by
adding at the end of it, in parentheses, the name of the original
author or publisher of that section if known, or else a unique number.
Make the same adjustment to the section titles in the list of
Invariant Sections in the license notice of the combined work.
In the combination, you must combine any sections Entitled "History"
in the various original documents, forming one section Entitled
"History"; likewise combine any sections Entitled "Acknowledgements",
and any sections Entitled "Dedications". You must delete all sections
Entitled "Endorsements".

6. COLLECTIONS OF DOCUMENTS
You may make a collection consisting of the Document and other
documents released under this License, and replace the individual
copies of this License in the various documents with a single copy
that is included in the collection, provided that you follow the rules
of this License for verbatim copying of each of the documents in all
other respects.
You may extract a single document from such a collection, and
distribute it individually under this License, provided you insert a
copy of this License into the extracted document, and follow this
License in all other respects regarding verbatim copying of that
document.

7. AGGREGATION WITH INDEPENDENT WORKS

A compilation of the Document or its derivatives with other separate
and independent documents or works, in or on a volume of a storage or
distribution medium, is called an "aggregate" if the copyright
resulting from the compilation is not used to limit the legal rights
of the compilations users beyond what the individual works permit.
When the Document is included in an aggregate, this License does not
apply to the other works in the aggregate which are not themselves
derivative works of the Document.
If the Cover Text requirement of section 3 is applicable to these
copies of the Document, then if the Document is less than one half of
the entire aggregate, the Documents Cover Texts may be placed on
covers that bracket the Document within the aggregate, or the

282

electronic equivalent of covers if the Document is in electronic form.

Otherwise they must appear on printed covers that bracket the whole
aggregate.

8. TRANSLATION
Translation is considered a kind of modification, so you may
distribute translations of the Document under the terms of section 4.
Replacing Invariant Sections with translations requires special
permission from their copyright holders, but you may include
translations of some or all Invariant Sections in addition to the
original versions of these Invariant Sections. You may include a
translation of this License, and all the license notices in the
Document, and any Warranty Disclaimers, provided that you also include
the original English version of this License and the original versions
of those notices and disclaimers. In case of a disagreement between
the translation and the original version of this License or a notice
or disclaimer, the original version will prevail.
If a section in the Document is Entitled "Acknowledgements",
"Dedications", or "History", the requirement (section 4) to Preserve
its Title (section 1) will typically require changing the actual
title.

9. TERMINATION
You may not copy, modify, sublicense, or distribute the Document
except as expressly provided under this License. Any attempt
otherwise to copy, modify, sublicense, or distribute it is void, and
will automatically terminate your rights under this License.
However, if you cease all violation of this License, then your license
from a particular copyright holder is reinstated (a) provisionally,
unless and until the copyright holder explicitly and finally
terminates your license, and (b) permanently, if the copyright holder
fails to notify you of the violation by some reasonable means prior to
60 days after the cessation.
Moreover, your license from a particular copyright holder is
reinstated permanently if the copyright holder notifies you of the
violation by some reasonable means, this is the first time you have
received notice of violation of this License (for any work) from that
copyright holder, and you cure the violation prior to 30 days after
your receipt of the notice.
Termination of your rights under this section does not terminate the
licenses of parties who have received copies or rights from you under
this License. If your rights have been terminated and not permanently
reinstated, receipt of a copy of some or all of the same material does
not give you any rights to use it.

283

10. FUTURE REVISIONS OF THIS LICENSE

The Free Software Foundation may publish new, revised versions of the
GNU Free Documentation License from time to time. Such new versions
will be similar in spirit to the present version, but may differ in
detail to address new problems or concerns. See
http://www.gnu.org/copyleft/.
Each version of the License is given a distinguishing version number.
If the Document specifies that a particular numbered version of this
License "or any later version" applies to it, you have the option of
following the terms and conditions either of that specified version or
of any later version that has been published (not as a draft) by the
Free Software Foundation. If the Document does not specify a version
number of this License, you may choose any version ever published (not
as a draft) by the Free Software Foundation. If the Document
specifies that a proxy can decide which future versions of this
License can be used, that proxys public statement of acceptance of a
version permanently authorizes you to choose that version for the
Document.
11. RELICENSING
"Massive Multiauthor Collaboration Site" (or "MMC Site") means any
World Wide Web server that publishes copyrightable works and also
provides prominent facilities for anybody to edit those works. A
public wiki that anybody can edit is an example of such a server. A
"Massive Multiauthor Collaboration" (or "MMC") contained in the site
means any set of copyrightable works thus published on the MMC site.
"CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0
license published by Creative Commons Corporation, a not-for-profit
corporation with a principal place of business in San Francisco,
California, as well as future copyleft versions of that license
published by that same organization.
"Incorporate" means to publish or republish a Document, in whole or in
part, as part of another Document.
An MMC is "eligible for relicensing" if it is licensed under this
License, and if all works that were first published under this License
somewhere other than this MMC, and subsequently incorporated in whole or
in part into the MMC, (1) had no cover texts or invariant sections, and
(2) were thus incorporated prior to November 1, 2008.
The operator of an MMC Site may republish an MMC contained in the site
under CC-BY-SA on the same site at any time before August 1, 2009,
provided the MMC is eligible for relicensing.

284

13

Index

285

Index
adsorption energy, 168
atomistic thermodynamics, 195
bader, 38
band structure, 127
center of mass, 21
cohesive energy, 103
convergence
ENCUT, 57
KPOINTS, 103
DFT+U, 204
dipole correction, 166
dipole moment, 32
HSE06, 208
infrared intensity, 48
ISMEAR, 87
jasp, 2
molecular weight, 20
moment of inertia, 21
nudged elastic band, 77
reconstruction, 158
SIGMA, 87
thermochemistry, 52
vibrations, 45
work function, 163

286