MODULE 2:

QUANTUM MECHANICS
Principles and Theory
 You are
here

http://www.lbl.gov/CS/html/exascale4energy/nuclear.html 2
 Short Review of Quantum Mechanics
Why do we need quantum mechanics?
• Bonding and structure
• Electronic, magnetic and optical properties of materials
• Chemistry and reactions
Standard model of matter
• Matter consists of atoms
• Atoms consist of
‣ Massive, point-like nuclei (protons + neutrons)
‣ That are surrounded by tightly bound core electrons
‣ And held together in molecules, liquid and solids by the bonds formed
by valence electrons

3
 Short Review of Quantum Mechanics
Wave-particle duality
• Planck’s uncertainty relationship
·p=h

Schrödinger equation
• Time dependent

2
⇥ (r, t)
⇥ 2
(r, t) + V (r, t) (r, t) = i
2m ⇥t

4
 Solutions of the Schrödinger Equation (1)
Free particle
• V(r) = 0 ⇒ Solutions are plane waves
k (r, t) = exp (i k · r t)
2 2
k
Ek = =
2m

5
 Solutions of the Schrödinger Equation (2)
Infinite square well
• Plane waves that vanish at the boundary
!4
⇤ 16
n=4
0 if 0 < x < a
V (x) = 14
otherwise

Energy En/E1
12
⌅
n x⇥
!3
2 10

!(x)
n=3
⇥n (x) = sin 8
a a
2 2 2 6
n n=2 !2
En = 4
2ma2 2 n=1
!1

0 a
0
x

6
 Solutions of the Schrödinger Equation (3)
Metal surface
Metal Surfaces (II)
Metal Surfaces (I)
• Potential step ⇒ Plane wave inside metal, exponential decay outside

Feb 15 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari

7
 Solutions for the Coulomb Potential
Spherical Symmetry
⇤2 ⇤2 ⇤2
2
= + +
⇤x2 ⇤y 2 ⇤z 2
⇥ ⇥
1 ⇤ 2 ⇤ 1 ⇤ ⇤ 1 ⇤2
= r + 2 sin + 2 2
2
r ⇤r ⇤r r sin ⇤ ⇤ r sin ⇤⇥2

Separation of variables
nlm (r) = Rnl (r) · Ylm ( , ⇥)
Equation for radial wave functions
⇤ ⇥ ⌅
2
d 2
2 d l(l + 1) 2
+ + + V (r) R nl (r) = E R nl (r)
2m dr2 r dr 2mr2

8
The Periodic System of Elements

9
http://www.orbitals.com/orb/orbtable.htm
 Basis Set Expansion (Matrix Formulation)
k
Expand wave function in set of n orthogonal functions | = cn |⇥n
n=1

Plugging this solution into the Schrödinger equation yields and multiplying with
function ϕm yields
⇥⇥m |H| ⇤ = E⇥⇥m | ⇤
k
cn ⇥⇥m |H|⇥n ⇤ = E cm
n=1
k
Hmn cn = E cm
n=1
⇥ ⇥ ⇥
H11 ··· H1k c1 c1
⇧ .. .. ⌃ · ⇧ .. ⌃ = E ⇧ .. ⌃
⇤ . . ⌅ ⇤ . ⌅ ⇤ . ⌅
Hk1 ··· Hkk ck ck
10
 Variational Principle

⇥ |H| ⇤
E[ ] = ⇥ | ⇤ E0
E[ ] = E0 ⇥ If is the groundstate

Example: The hydrogen atom

⇥ = c · exp( r)
⌅ |H| ⇧
E[ ] =
⌅ | ⇧

11
 The Hydrogen Atom
Ansatz: ⇥ = c · exp( r)
⌅ |H| ⇧
E[ ] =
⌅ | ⇧
Calculate:
1 2 1
| ⇥ ⇥ | ⇧ | ⇤ ⇥ | | ⇤
2 r
Use: n!
...|...⇥ = 4 r ...d r
2 3
r exp( ar) =
n
0 0 an+1

⇤2 ⇤2 ⇤2
2
= 2
+ 2+ 2
⇤x ⇤y ⇤z
⇥ ⇥
1 ⇤ 2 ⇤ 1 ⇤ ⇤ 1 ⇤2
= r + 2 sin + 2 2
2
r ⇤r ⇤r r sin ⇤ ⇤ r sin ⇤⇥2

12
 The Hydrogen Atom
Result:
c2 1 2 c2 1 c2
⇥ | ⇤=⇥ , ⇥ | ⇧ | ⇤=⇥ , ⇥ | | ⇤=⇥
3 2 2 r 2

Write out the energy as a function of α and minimize with

respect to α:
c2 c2
⇥2 ⇥ 2 1
E( ) = = 2
⇥ c2
3
2
d
E( ) = 1=0
d
13
 Atomic Units

Quantity Name Symbol SI value

Energy Hartree energy Ha 4.359 744 17(75)×10-18 J
Length Bohr radius a0 5.291 772 108(18)×10-11 m
Mass Electron rest mass me 9.109 3826(16)×10-31 kg
Electric charge Elementary charge e 1.602 176 53(14)×10-19 C
Electrostatic force Coulombs constant 1/4πε0 8.9875516×109 C-2Nm2
constant

14
http://www.orbitals.com/orb/orbtable.htm
 The Many-Electron Problem
⇥
n
⇧ ⇧n n
⇧ 1 ⇧ n
⇤ 1 Z 1 ⌅ (r1 , . . . , rn ) = E (r1 , . . . , rn )
⇤2i +
2 i=1 i=1
r i i=1 j=i+1
|ri rj |

Example: Fe atom
• Fe has 26 electrons ⇒ wave function has 3×26 = 78 variables
• Store wave function on a grid
• Use a coarse grid of only 10 points along each direction
• To store wave function would require storage of 1078 numbers
• Single precision 1 number = 4 Bytes

• Compare that to all the data stored worldwide 1 zettabyte = 1021 Bytes

15
 The Hartree Method

Independent electron method

• Assume that electrons move independently of each other
• Each electron moves in an effective potential that consists of
- Attraction of nuclei
- Average repulsive interaction of other electrons

• Many-body wave function as product of single-particle orbitals

(r1 , r2 , . . . , rn ) = ⇥1 (r1 ) ⇥2 (r2 ) . . . ⇥n (rn )

16
 The Hartree Method
⇥
⇧ ⌃
⇧ 1 ⌥ ⌥ 1 ⌃
⇧ 2 3 ⌃
⇧ ⇤i +
2
V (RI ri ) + |⇥j (rj )| d rj ⌃ ⇥i (ri ) = ⇥i (ri )
⇧ 2 |ri rj | ⌃
⇤ I j=i ⌅
↵ ⌦
Hartree potential

Douglas Hartree and

Arthur Porter’s Meccano
differential analyzer built
in 1934 at a cost of £20.

It achieved an accuracy of
about 2%.


17
 Illustration of Electron Correlations
Uncorrelated Correlated
Cars are smeared out Cars avoid each other

Uncorrelated Correlated
Electrons described by their Electrons avoid each other to due
independent density, electrons to the Coulomb interaction
can get arbitrary close between them
http://www.digital-photography-school.com/how-to-shoot-light-trails http://www.flickr.com/photos/88943727@N00/101166668/

18
 The Hartree-Fock Method

What is missing in the Hartree approximation

• Wave function is not antisymmetric
• Does not include electron correlation

Antisymmetry for Fermions

• Exchanging two identical (indistinguishable) fermions changes the sign of
the wave function
(r1 , r2 , . . . , rj , . . . , rk , . . . , rn ) = (r1 , r2 , . . . , rk , . . . , rj , . . . , rn )

Pauli Exclusion Principle

• Two electrons cannot be in the same quantum state
• Consequence of the antisymmetry

19
 Slater Determinants

Slater determinant
• Antisymmetric product of single particle orbitals

⇥1 (r1 ) ⇥2 (r1 ) ··· ⇥n (r1 )

1 ⇥1 (r2 ) ⇥2 (r2 ) ··· ⇥n (r2 )
(r1 , r2 , . . . , rn ) = ⇥ .. .. ..
n! . . .
⇥1 (rn ) ⇥2 (rn ) · · · ⇥n (rn )

• Swapping rows in a determinant changes the sign

20
 The Hartree-Fock Method

Hartree-Fock equation for orbitals φλ

• Use of variational principle leads to set of equations for φλ
⇤ ⌅
1 2 ⇧
⇤i + V (RI ri ) ⇥ (ri ) +
2
I
⇤ ⌃ ⌅
⇧ 1
⇥µ (rj ) ⇥µ (rj )d rj ⇥ (ri )
3

µ
|r i r j |
⇧ ⌃ ⇥
1
⇥µ (rj ) ⇥ (rj )d rj ⇥µ (ri ) = ⇥ (ri )
3

µ
|ri rj |

21
 ⇤ ⌃ ⌅
⇧ 1
µ (rj )Exchange ⇥µ (rj )d rj ⇥ (ri )
3
⇥The Term
µ
|r i r j |
⇧ ⌃ ⇥
1
⇥µ (rj ) ⇥ (rj )d rj ⇥µ (ri ) = ⇥ (r
3

µ
|ri rj |

• Describes effect of exchange of electrons

• Cannot be written in the form
x
V (ri )
x
(ri )
• Instead it is of the form

x
V (ri , rj )
x
(rj )d rj
3

• This is called a non-local potential

22
 Successes and Limitations of Hartree-Fock
Successes
• Good for atomic properties
• Self-interaction free
• Good starting point for correlated-electron methods
Limitations
• Schrödinger equation: H Ψ = E Ψ ⇒ Εexact
• Hartree-Fock equations: F φi = εi φi ⇒ ΕHF
• Any effect beyond HF is called correlation Ecorr = Eexact - EHF
• Size of correlation energy Ecorr < 1% of Eexact

Example: N2 molecule: Ecorr = 14.9 eV < 1% of Eexact
• However: binding energy N2 → N + N is 

De(Hartree-Fock) = 5.1 eV, De(exp) = 9.9 eV

Thus, there are large contribution from the correlation energy

to relative energies, i.e. chemical
23
reaction energies.
 Beyond Hartree-Fock

Hartree Fock configuration

• HF Slater determinant is built from lowest energy 1-e orbitals
0
HF =| 1 2 ... K|
• Slater determinant is also called a configuration since it refers to
certain filled orbitals

Configuration interaction method

• Add additional configurations to the wave functions that mix in
“excited” states
• Excite electron from orbital i to orbital K+1
1
HF =| 1 2 ... K+1 . . . K|

CI = c0 0 + c1 1 + c2 2 + ...
24
 Density Functional Theory
• Theory for the ground state energy of a system as a function of the
electron density instead of the wave function

• Walter Kohn received the Nobel prize in 1998 for his development of
density functional theory

Motivation (r1 , r2 , . . . , rn ) Function of 3N variables

(r) Function of 3 variables
Electron density

Locally
homogeneous
electron gas
Position
25
 The Hohenberg-Kohn Theorems

• The external potential and the number of electrons define the problem
• Schrödinger’s equation in principle uniquely determines the wave functions
• All system properties follow from the wave functions
⇒ HK Thm I: The energy and everything else
(incl. the density) is a determined by Vext and Nel

Hohenberg-Kohn
Theorem
Vext(r) ! !0(r)
Schrödinger Integration of |!0|2 over
# $
equation N-1 electron coordinates
!i(r1, ... rN) " !0(r1, ... rN)

26
 The Universal Functional
• Since the ground state density determines all properties, the ground state
energy and its components are a functional of the density
0
Etot = Ekin [ ] + Vext [ ] + Vel
0 0
el [
0
]
• Vext is known

Vext [ 0 ] = Vext (r) (r)d3 r

• For the kinetic energy and the e-e interaction

F [ ] = Ekin [ ] + Vel
0 0
el [
0
]

• Form of this functional is the same for any molecule or solid

⇒ HK Thm II: A universal energy functional can be defined in terms of the
density, F[ρ], and the exact ground state is the functional minimum

(But, the functional form is unknown!)
27
 2nd Hohenberg-Kohn Theorem

• The groundstate energy can be obtained variationally.

• The density that minimizes the total energy is the exact groundstate density.

E[ (r)] = F [ (r)] + Vext (r) (r)d r3

E0

28
 The Kohn-Sham Equations
Mapping to a non-interacting system
N
(r) = |⇥i (r)|2
i=1
• Why? The kinetic energy of the non-interacting system is well defined.
N ⇥
1
TS [ (r)] = ⇥i (r)⇥2 ⇥i (r)
2 i=1
• Universal functional now takes the form

F [ (r)] = TS [ (r)] + EH [ (r)] + Exc [ (r)]

• Electron-electron interaction is separated into two terms, the Hartree term
(Coulomb interaction) and the unknown(!) exchange-correlation energy term
1 (r) (r ) 3 3 1 (r) (r
] + EH [ E
(r)] +
= E
H F [ (r)][ (r)] d r d r
xc = TS [ (r)] + EH [ (r)] + Exc [ (r)]
EH =
2 |r r | 2 |r r
29
 Euler-Lagrange Equations
Minimize energy with respect to variations of the density
⇤ ⇧ ⇧ ⇥⌅
⇥ F [⇥(r)] + Vext (r)⇥(r)d3 r µ ⇥(r)d3 r N =0
F [⇥(r)
⇥(r) + Vext = µ
• Resulting equations have a similar form as the Schrödinger equation
and are known as the Kohn-Sham equations

⇥
1 2
⇥ + VH (r) + Vxc (r) + Vext (r) ⇥i (r) = i ⇥i (r)
2
⇧ ⌅⇤ ⌃
HKS

30
 Summary

Hohenberg-Kohn Hohenberg-Kohn theorem

theorem applied to non-interacting electrons

Vext(r) ! !0(r) !0(r) " Vext(r)

Kohn-Sham
# $ $ #

!i(r1, ... rN) " !0(r1, ... rN) !i=1,N(r) ! !i(r)

N ⇥ ⇥
1
E[{⌅i (r)}] = ⌅i (r)⌅2 ⌅i (r)d3 r + EH [⇤(r)] + Exc [⇤(r)] + Vext (r)⇤(r)d3 r
i=1
2
N ⇥
Exc [⇤]
= ⇥i VH [⇤(r)] + Exc [⇤(r)] ⇤(r)d3 r
⇤(r)
But exchange-correlation functional is still not known!
i=1

31
What about Exc? The Local Density Approximation

D. M. Ceperley and B. J. Alder, “Ground State of the Electron Gas by a Stochastic Method” Phys. Rev. Lett. 45, 566 (1980). http://link.aps.org/abstract/PRL/v45/p566

32
 The Phases of Silicon
Under pressure Si displays 11 crystal phases
Silicon Crystal Phases
• LDA correctly predicts the energetic order of all these phases

Compression
11 GPa 13 GPa 16 GPa 36 GPa 42 GPa 79 GPa

Si(I) diamond Si(II) ! −tin Si(XI) Imma Si(V) hexagonal Si(VI) orthorhombic Si(VII) hcp Si(X) fcc
Z=4 Z=6 Z=6 Z=8 Z=10 Z=12 Z=12

Decompression
Slow pressure release Fast pressure release

9 GPa 2 GPa >480 K

Si(VIII) and Si(IX) tetragonal

Si(II) !−tin Si(XII) R8 Si(III) BC−8 Si(IV) hex. diamond

Z=6 Z=4 Z=4 Z=4

Phys. Rev. B 24, 7210 (1981), ibid. 49,33

5329 (1994), ibid. 69, 134112 (2004)
 Exchange-Correlation Functionals
Local density approximation (LDA)
• Based on Ceperley & Alder’s calculations for the uniform electron gas
by quantum Monte Carlo (a stochastic method for quantum particles)
Generalized gradient approximations (GGA: PW91, PBE)
• Gradients of the density are introduced
• Preserve analytic scaling features of the unknown exact functional
Meta-GGA (TPSS)
• Include information about curvature of the density
Hybrid density functionals (B3LYP, HSE)
• Based on GGA or meta-GGA approximations
• Add some non-local Hartree-Fock exchange to the functional

34
 Density Functional Theory in Practice

1. Remove tightly bound core electrons: the pseudopotential approach

2. Represent orbitals with a basis (plane waves or Gaussians)
3. Calculate total energy for trial orbitals
- Kinetic and Hartree energy in reciprocal space
- Exchange-correlation energy and external potential in real space
- Method can take advantage of Fast Fourier Transformations
- Sum over all states: BZ integrations
4. Minimize energy and iterate charge density to self-consistency

35
 Pseudopotentials
Electrons in the inner shells do not contribute to bonding
• Core electrons are effectively frozen
• Replace Coulomb potential between electrons and nuclei

with effective potential, the pseudopotential
3p -2.7 eV  Valence 2p -2.7 eV

3s -7.8 eV  states 1s -7.8 eV

Al 2p -69.8 eV  Pseudo Al

Z = 13 2s -108 eV  Core Z=3

 states

1s -1512 eV 

⇥ ⇥
1 2 1 2 pseudo
⇥ + Veff ⇥i = i ⇥i ⇥ + Veff ⇥pseudo = ⇥
i i
pseudo
2 2 i

36
 Pseudopotentials

The pseudopotential and


the wave function
• Real potential and wave

function are shown in blue

• Pseudopotential and

pseudo wave function in red

• Outside the cutoff region


(vertical black line)

the two are identical

37
 Bloch Theorem — G’s and k’s
Basis set choices
• For molecules: use atomic orbitals, or localized functions like Gaussians
• For solids, periodic functions such as sines and cosines (i.e., plane waves)
• Use Bloch Theorem for periodic solids:
wavevector k ~ electron momentum
(k point in 1st Brillouin Zone)
plane waves

reciprocal lattice vector G

R]
[H,
= [H, ]R]=
0TTRTR⇥
[H, ]==000(r)
⇥ (r)
= uunk(r)
(r) ===u=
exp(ik(r)
(r)
(r) exp(ik
exp(ik
exp(ik
r) r)
· r)
··(G)
nk r) exp(i(r
⇥ ⇥ nk (r) uu
nknk
nk · c G+· r)
R)==2⇡ (r)
m · exp
|G| Gmax
Hamiltonian and translation unk (r) = cnk (G) exp(i G · r)
operator commute |G| Gmax

function with periodicity

wavefunction of electron n at of xtal lattice expanded
wavevector k in Brillouin Zone in PW basis set
(Wigner-Seitz cell in reciprocal space) unk(r(r)
+=
+ (r)(r)
= unk
R) = = (G)
· cexp
nk (ik ·
38
|G| G |G|
 G’s: Plane Wave Basis Set
• Superposition of plane waves to represent orbitals with xtal lattice periodicity
unk (r) = cnk (G) exp(i G · r)
|G| Gmax

• Increase size of basis set to approach completeness (i.e., convergence)

39
 k’s: 1st Brillouin Zone (1BZ)
• Crystal structures defined by Bravais lattice {ai} and basis
al am
• Reciprocal lattice
 bk = 2 ·
ak · (al am )

• Brillouin zone is the Wigner-Seitz cell of the reciprocal lattice

40
 k’s: 1st Brillouin Zone (1BZ)
• The wavefunction (and energy) of each electron depends on both its quantum
number n and its position k (i.e., its momentum) within the 1BZ

• We must solve a different Kohn-Sham equation for each n and k

• Bloch Thm has allowed us to replace analysis of ~infinite number of electrons in the
xtal with only those in BZ but at an infinite number of k points!

• Payoff is that wavefunctions change smoothly in k so in practice consider finite

number of k-points — sampling over a k-space grid, refine grid until convergence

• Real-space quantities are computed by a discrete sum over n and integration over
k within the Brillouin Zone (approximated over a grid at finite k-points)

• e.g. density, n(r)

41
 k’s: Band Structure Plot — εn,k
Copper atom 0 size 0.20
• Copper: Band structure calculated with Wien2k
2.0
band (i.e., n)
0.0 EF

Energy (eV)
-2.0

-4.0 s-d hybridization

-6.0

-8.0

-10.0
W L ! " # X Z W K

path in BZ (i.e., k)
R ] [H,
= [H,
0TRT]
⇥
R =
] 0
= 0 ⇥
(r) =
⇥ u (r)
nk
=
(r) =
u (r)
exp(ik
u
nknk (r)
·exp(ik
exp(ik
r) · r)
· r)

• Nearly free electron s-band dominates at low and high energies

• Hybridization of nearly-free s and atomic-like d orbitals at intermediate energies
42
 Successes and Failures of DFT

Structural and elastic properties

• Lattice parameters are typically within a few percent of experimental
values and often accurate to better than 1%
• The bulk modulus and other elastic constants are usually within 10%
[00"] [0"1] [0""] ["""]
Vibrational Properties 10 Ni
• Forces = 1st derivative of energy
 fcc

with respect to atomic displacement 8

Frequency [THz]
• Forces are accurate to better than

6
10% (similar to elastic constants)
• Vibrational frequencies are the 2nd
 4
derivatives
EAM (NRL)
• Their accuracy is about 1/2 the
 2 EAM (Voter&Chen)
DFT (PW91)
accuracy of the forces or about 5% Neutron diffr. (296K)

0
! X W X K ! L
43
 Successes and Failures of DFT

Defect energies
• In many cases such as for metals, vacancy and interstitial energies are
highly accurate (within 0.1 eV)
• In some cases such as interstitial defects in silicon, DFT is too low by
about 1 eV predicting a 3–3.5 eV formation energy instead of the
4.5 eV of experiments and QMC [Phys. Rev. B 74, 121102(R) (2006)]
Excited states and gaps
• Local density approximation fails for excited states
• Bandgaps in LDA and GGA are usually underestimated by 20–50%
• In some cases such as Ge, LDA predicts a metallic instead of
semiconducting state
• Hybrid functionals (e.g. B3LYP and HSE) improve the accuracy to
about 10%

44
 -5

E-EFer
Wien2k

-10
The bandgap problem of DFT
Crystal03

Band
Example: L structure
Band structure
Bandstructure
Γ of
of InAs
Xof
InAs
W K with
InAs Γ Crystal
with Γ 03
Crystal
L 03 X W K Γ

LDA,
B3LYP nonogap:
LDA,
10%, gap:-0.42
gap: 0.06eVeV
-0.42 PBE,
PBE,
B3LYP, nonogap:
15%, gap:-0.13
gap: 0.30eV
-0.13 eV
eV B3LYP 20%, gap: 0.54 eV
55 5

00 0
E-EFermi [eV]
E-EFermi [eV]

-5-5
-5
Wien2k
Wien2k
Crystal03
Crystal03
-10
-10
-10

LL L ΓΓ XX W
WWK
KK ΓΓ ΓLL L ΓΓ Γ XX XWWW
KK K Γ ΓL
Γ Γ X W K Γ

B3LYP10%,
3LYP 10%,gap:
gap:0.06
0.06eVeV B3LYP,15%,
15%,gap:
gap:0.30
0.30eVeV B3LYP20%,
20%,gap:
gap:0.54
0.54eV
eV
• Experimental bandgap:B3LYP,
0.41 eV B3LYP

Band gap problem: LDA and GGA yield a metallic ground state!
• Practical solution: Hybrid functionals B3LYP & HSE (0.39 eV)
• Better solution: GW approximation or QMC methods
45
Beyond DFT – The GW approximation
Energy gap (eV)
-2 0 2 4 6 8 10
Si
Ge
diamond
3C SiC
LiCl
!-C3N4
BN
BP
BAs
AlN
AlP
AlAs
AlSb
GaN LDA
GaP
GaAs GWA
GaSb
InP Expt., indirect gap
InAs
InSb Expt., direct gap
Al0.5Ga0.5As
In0.53Ga0.47As
GaAs0.5N0.5
GaAs0.75N0.25
ZnO
ZnS
ZnSe
ZnTe
CdS
CdSe
CdTe
MgO
MnO
NiO
CaCuO2
Li2O
ZrO2
SnO2
46
 Summary of Density Functional Theory

LDA
• Lattice constants: 1-3% too small
• Cohesive Energies: 5-20% too strongly bound
• Bulk Modulus: 5-20% (largest errors for late TM)
• Bandgaps: too small
GGA
• Improves cohesive energies
• Often but not always better for lattice parameters
• Important for magnetic systems
Hybrid functionals
• Improved band gaps, often very accurate

Always check the accuracy of the computational method by

benchmarking against experimental data or more accurate theory.
47