Practical QUantum

Practical Quantum Chemistry
Script of a crash course held by
Florian M uller-Plathe
March 16th20th, 1998
Lecture notes prepared by
Patrick Ahlrichs, Roland Faller, Oliver Hahn, Mathias P utz, Dirk Reith, Yannick Rouault, Heiko Schmitz and Thomas Soddemann
Contents
1 Introduction 2 Quantum Mechanics 2.1 2.2 2.3 Some useful postulates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Born-Oppenheimer approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Units, Symbols and Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 2.3.2 2.3.3 Atomic units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1 2 2 3 3 3 3 4 4 4 5 5 6 6 7 7 8 9 9
Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3.1 2.3.3.2 Dirac bras and kets: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Two electron integrals notation: . . . . . . . . . . . . . . . . . . . . . . .
3 Hartree-Fock 3.1 3.2 3.3 3.4 3.5 3.6 Hartree approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Paulis antisymmetry principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hartree-Fock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hartree-Fock methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hartree-Fock-Roothaan equations (for RHF) . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1 Digression: How to solve the matrix equation? . . . . . . . . . . . . . . . . . . . .
3.7 3.8 3.9
Matrix elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Iterative scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Hartree-Fock-Roothaan implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.9.1 3.9.2 3.9.3 Classical SCF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Direct SCF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Speed up Hartree-Fock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.10 Koopmans theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4 Basis functions - basis sets 4.1 13
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Contents
4.1.1 4.2 4.3 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Slater-type orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Gaussian-type orbitals 4.3.1 4.3.2 4.3.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Radial part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Angular part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Advantages and disadvantages of GTOs . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 4.5 4.6 4.7 4.8 4.9
Application of basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 CGTO (contracted Gaussians) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Basis set size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Polarization functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Gaussian exponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Basis-set superposition error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 22
5 Electron correlation 5.1 5.2 5.3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Non-dynamic correlation effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Dynamic correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 5.3.1 5.3.2 5.3.3 How much of the electron correlation is described by a single determinant? . . . . . 25 Correlation energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 How to account for electron correlation . . . . . . . . . . . . . . . . . . . . . . . . 26 28
6 Conguration Interaction 6.1 6.2 6.3
Properties of CI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Matrix elements between Slater determinants . . . . . . . . . . . . . . . . . . . . . . . . . 29 Properties of the CI expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 32
7 Review of Perturbation Theory 7.1 7.2 7.3 7.4
Auxiliary theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 How to calculate n-th order energy? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Treating electron correlation via perturbation theory . . . . . . . . . . . . . . . . . . . . . . 35 Second-Order Energy (MBPT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 38
8 Size Consistency 8.1 8.2 8.3
Size Consistency of HF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Size Consistency of MBPT2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Trouble: Double CI Size Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 42
9 Derivatives of the Energy
ii
Contents
9.1 9.2 9.3 9.4 9.5 The Potential energy surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 How to nd stationary points? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Physical meaning of some derivatives of the potential energy . . . . . . . . . . . . . . . . . 44 The Hellmann-Feynman theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Gradients for RHF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 49
10 Density functional theory (DFT)
10.1 An illustrative example: The Thomas Fermi model . . . . . . . . . . . . . . . . . . . . . . 49 10.2 The Hohenberg-Kohn theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 10.3 The Kohn-Sham (KS) method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 10.4 Exchange (and) correlation functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 10.5 DFT in a Gaussian orbital basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 11 The Car-Parrinello Method 57
11.1 Plane wave basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 11.2 The Kohn-Sham equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 11.3 Pseudopotential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 11.4 CP Molecular dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
iii
1 Introduction
There have been many method developed to describe atoms and molecules in term of quantum mechanis. Due to the fact that it is nearly impossible to calculate the electron structure of molecules analytically correct, all those methods are approximations. To the most known and used mehtods belong ab initio methods Hartree Fock electron correlation (conguration interaction, many body perturbation theory ) semiempirical methods density functional theory In the following the range of the practial applicability of the above methods will be discussed. The choice of a basis sets as well as the chemist jargon (direct SCF, MP2, MNDO 6-31G**) will no longer be a mysterium to us. The gained knowlegde can than be applied to critical reading of theoretical (and experimental) literature. In this lecture, the following topics have not been covered foundations of quantum mechanics (recommended) thorough derivation of approximations advanced theoretical methods latest developments
2 Quantum Mechanics
2.1 Some useful postulates
Postulate 1 The state of the system is described by a wave function
(r1 r2 : : : rn t):
is single valued, continuous, and quadratically integrable:
j j2 dr =
dr
= nite
Postulate 2
^. To nd the operator, To every physical observable A there corresponds a linear Hermitian operatorA express A in cartesian coordinates and momenta, and substitute
r ;! r ^ p ;! p ^ = ;i~r
Examples:
= i qi ri , ^ = i qir ^i P ^ = ; Pi ~2 r2 kinetic energy K = i p2 i =2mi , K i =2mi

dipole moment Postulate 5
^ is obtained as The expectation value of A ^i = hA

Postulate 6
h h
^ A
i i
The time development of a system is given by the time dependent Schr odinger equation
^ is the Hamiltonian operator where H

of the stationary system.
^ (t) i~@t (t) = H

2
^ =V ^ (r) ; ~ r2 H 2m
The time independent Schr odinger equation is given by the famous eigenvalue equation
^ =E : H
Postulate 7 (Spin) Pauli exclusion (anti symmetry) principle for Fermions.
2.2 The Born-Oppenheimer approximation
2.2
The Born-Oppenheimer approximation
Quantum chemistry is the quantum mechanical treatment of molecules. We are dealing here with the Born Oppenheimer approximation which separates electronic and nuclear degrees of freedom. With V el = V (R) being the electronical potential and Vnuc = Eel the nuclear one, the wave function separates within the Born Oppenheimer approximation into two parts:
(R r) = (R)
(r)
R are the nuclear coordinates and r the electronic ones. This approximation makes sense because the nuclei are much heavier than the electrons, therefore the lighter electrons will instantaneously adapt to the nuclear conformation.
But the Born Oppenheimer Aprroximation has restrictions. It breaks down if
Eel ' Evib rot

photoionisation or electron-molecule scattering takes place, the temperature is high. Within the Born-Oppenheimer approximation the nuclei are often described as classical objects.
2.3
2.3.1
Units, Symbols and Notations

Atomic units
length energy mass charge angular momentum electric dipole moment electric dipole polarisability electric eld wave function
e (elementary charge) ea0 2 e a2 0 =EH EH =ea0 ;3=2 q0

~
me electron mass
a0 bohr EH hartree EH = 4 e2 0 a0
5:29177 10;11 m 4:3598 1018 J 9:1095 1:6022 1:0546 8:4784 1:6488 5:1423 2:5978 1031 kg 10;19 C 1034 Js 10;30 Cm 10;41 C 2 m2 J 10;11 V=m 1015 m;3=2
2.3.2
Symbols
wave function in general Slater determinant molecular orbital (MO) spin MO, often i (1) (1) = i , i (1) (1) = i AO (basis function) vector matrix, tensor position of electron i spin of electron i (= (i) (i)) spin coordinate of electron i, i.e. xi = (~ ri ! i )
~ a A r ~i !i x ~i
2 Quantum Mechanics
2.3.3
Notations
2.3.3.1 Dirac bras and kets:

j ii h ij A j ii h ij A h ij j i h ij A j j i
= = = = = =
^ A
i i
^ Ri A j R iA ^ i j.
2.3.3.2 Two electron integrals notation: physicists notation

hij jkli
ZZ
~1 ) spin orbital i, occupied by electron 1. i (x

chemists notation
(~ x1) j (~ x2 ) r1 k (~ x1) l (~ x2 )d3 x1d3 x2 12
ij jkl] =
ZZ
x1 ) l (~ x2 )d3 x1 d3 x2 (~ x1 ) j (~ x2 ) r1 k (~ 12 ij jkl] = hikjjli
antisymmetrised 2-el. integrals

hij jjkli
= hij jkli ; hij jlki = ikjjl] ; iljjk]
spatial orbitals ' chemists notation
(ij jkl) =
index permutation symmetries
hij jkli = hjijlk i hij jkli = hkljij i ij jkl] = kljij ] ij jkl] = jijkl] =
ZZ
(~ x1 ) j (~ x2 ) r1 k (~ x1 ) l (~ x2 )d3 x1d3 x2 12
complex orbitals
ij jlk] = jijlk]
real or complex real
3 Hartree-Fock
The electronic Schr odinger Equation is given by
H =E
where H is the all electron Hamiltonian
(3.1)
H = H1 + H2 + H3 H1 is the one electron hamiltonian consisting of kinetic energy and nuclear attraction H= H2 is th 2-electron interaction (repulsion) H2 =
N X N X i j>i rij N X i
; r2 ;
(3.2)
1 2
X ZA
A
riA
(3.3)
(3.4)
and H3 is the repulsion of nuclei in molecules (mostly treated clasically)
H3 =
X X ZAZB
A B>A
RAB
(3.5)
= (~ r1 ~ r2 ~ r3 : : : ~ rN ) is the all electron wave function and E denotes the total (electronic) energy.
3.1 Hartree approximation
In the one electron electron theory the, following approximation is made:
= (~ r1 ~ r2 ~ r3 : : : ~ rN ) =
r1 ) 2 (~ r2 ) 3 (~ r3 ) 1 (~
rN ) N (~
(3.6)
ri ) are the one electron wave which is a separation ansatz and named The Hartree Approximation. i (~ functions, the orbital functions. The operator h(i) can be dened to act only on the electron i and the problem can be rewritten as a system of N one electron wave functions hi i (~ ri ) = ri ) i i(~
(3.7)
3 Hartree-Fock
3.2
Paulis antisymmetry principle
When 2 electrons (which are of course Fermions) are exchanged, changes sign. The aim is now to derive a approximation which is as easy to handle as the Hartree one and on the other side obeys Paulis principle. The solution is to use Slater determinants. An exchange of two electrons in a 2 spin orbital can be written as
2 1 p (permutation of 1 and 2:) 2

Written as a (Slater) determinant
1 p
1 (1) 2 (2) ; 1 (2) 2 (1)] 1 (2) 2 (1) ; 1 (1) 2 (2)] 1 (1) 1 (2) 2 (1) 2 (2)
=;
1 =p 2
(3.8)
The exchange of two electrons is equal to the exchange of two rows in the determinant which results in a change of sign. In general:
1 =p 2
Shorthand:
1 (1) 1 (2) 1 (N )
. . .
::: ::: . . . ::: ( N ) ::: 2

2 (1) 2 (2)
N (1) N (2) N (N )
. . .
(3.9)
=j
1 2 : : : Nj
3.3
Hartree-Fock
1. restrict trial wave function to one Slater determinant 2. effective one electron Hamiltonian (a)
hi = ;1=2r2 ;
(b) mean-eld approximation: one electron moves in the average eld of all other electrons ) Hartree-Fock / self-consitent eld (SCF) HF is variational ) HF energy > true energy HF gives the Slater determinant with the lowest possible energy (HF gives the best one determinant wave function) Derivation: see Szabo/Ostlund result for i = 1
P Z =r + modications ) H uckel, extended H uckel 6= Hartree-Fock A a iA
: : : N:
f i=
i i
Here, f is the Fock operator
N X ZA X 1 2 ^j (1) ; K ^ j (1) f (1) = ; 2 r + r + J iA j =1 A
(3.10)
3.4 Remarks
^j (1) is the Coulomb operator, effective one-electron operator dened by its action on i (1) J
h
^j (1) j i (1)i = i (1)j J
i (1) i (1) r
12
r1 d~ r2 j (2) j (2)d~
= iijjj ] = hij jij i
^ j (1) exchange operator K

h
Coulomb integral
^ j (1) j i (1)i = i (1)j K
j (1) i (1) r
12
r1 d~ r2 = i (2) j (2)d~
ij jij ] = hij jjii
exchange integral
3.4
Remarks
Coulomb integral can be interpreted as repulsion betweeen two charge densities
Exchange integral has no obvious interpretation: Has nothing to do with physical exchange of electrons. Non local potential: Schr odinger equation ! operator is known ! nding of wave functions and energies. HF equations: operator dpends implucitly on wave functions ) iterative solution, self consitent eld (SCF) method.
3.5
Hartree-Fock methods
There are different types of Hartree-Fock methods restricted Hartree-Fock (RHF) closed-shell singlett spatial MOs doubly occupied (spin , spin )
x ? "# ? ? "# i? ? "# ? ? "# x ? " ? ? " ? ? i ? "# ? "# ? ?

"#
restricted open-shell Hartree Fock (ROHF)
spatial MOs singly or doubly occupied
unrestricted Hartree-Fock (UHF) " " # i " # different spatial MOs for " # " #
x ? ? ? ? ? ? ? ? ?
and
^2 spins not E.F. of S
3 Hartree-Fock
Example: Li atom ROHF: j1s2 2sj doublet UHF: j1s
1s 2s
j lower energy, but not pure doublet
3.6
Hartree-Fock-Roothaan equations (for RHF)
HF equations are integro-differential equations which can be solved numerically for atoms (and diatomics) only. Here, a basis set expansion is introduced (mostly atomic orbitals) RHF equations for closed-shell singlets
^ (1) i (1) = F
i are the spatial MOs.
i i (1)
(i = 1 : : : Ne =2) ^j (1) ; K ^ j (1)) (2J
(3.11)
^ (1) + ^ (1) = h F
i=
X
j
(3.12)
X
q
Ciq
(i q = 1 : : : N )
(3.13)
~=C~
^ F
(3.14)
X
q
Ciq q =
X
q
Ciq
p
(3.15)
^ pF
X
q
Ciq q d =
X
q
Ciq q d
p qd
(3.16)
X Z
q
Ciq
^ q d = i X Ciq pF
q
(3.17)
X
q
Ciq Fpq =
X
q
Ciq Spq
(3.18)
F C = S C (diag) (3.19) where F is the Fock matrix, C the coefcient matrix, S the overlap matrix and (diag) the orbital energies. C yF C = C yS C
(3.20)
non-symmetric matrix eigenvalue problem. Transformation to a symmetric matrix ! diagonalise ! , C
3.7 Matrix elements
3.6.1
Digression: How to solve the matrix equation?
F C = C (diag) C F C = C y C (diag) = (diag) (3.21) notice that C y C = E . From a diagonalisation of F one obtains C and (diag). Solution: make S = E (in general S 6= E ) by choosing a new orthogonal basis: L odwin symmetric ; 1 ; 1 = 2 ; 1 = 2 SS =E orthogonalisation S S = E also S How can we obtain S ;1=2 ? 1. diagonalize S : U y S U = s (diag) ;1=2 2. with sii ! sii one obtains s ;1=2 3. undiagonalize s ;1=2 with the back transformation U s ;1=2 U y = S ;1=2 Check: S ;1=2 S S ;1=2 = U s ;1=2 U y S U s ;1=2 U y = U s ;1=2 s s ;1=2 U y = U U y = E transform RHF equations C 0 = S 1=2 C , C = S ;1=2 C 0 F S ;1=2 C 0 = S S ;1=2 C 0 (diag) S ;1=2 F S ;1=2 C 0 = S ;1=2 S S ;1=2 C 0 (diag) F 0 C 0 = C 0 (diag)
y
If S were the identity matrix E , then we would have
3.7
Matrix elements
overlap matrix: Fock matrix:
Fpq =
=
Spq = R Fpq =
Now
P j = r cjr
= hpq + 2
j 2 Z X hpq + 2 p (1) 4 j XZ j r , so
^ (1) + p(1) h
( ) = hpjqi p q ( )F p ^ q = hpj F jqi
^j (1) ; K ^ j (1))] q (1) d (2J

j (2) r
12
3 j (2)5
q (1) ;
p(1) j (2) r j (2) q (1) d ; 12 j
XZ
2 X p (1) d 4
j p(1) j (2) r
j (2) r
12
3 j (1)5
q (2) d
12
j (1) q (2) d
Fpq = hpq + 2
;
X
jrs
X
jrs
cjr cjs
cjr cjs
p(1) r (2) r
p (1) r (2) r
12
s (2) q (1) d
We introduce now the denity matrix R
= C yC , Rrs = j cjr cjs X Fpq = hpq + Rrs 2(pqjrs) ; (psjrq)]

rs
12
s (1) q (2) d
(3.22)
3 Hartree-Fock
3.8
Iterative scheme
initial guess diagonalize ;! F ;! C ;! R converged? ;! stop
iterate
{z
3.9
3.9.1
Hartree-Fock-Roothaan implementation
Classical SCF
Algorithm: 1. Calculate integrals Spq , hpq , (pq j rs) and store the results on disk 2. Calculate S;1=2 .
^ , smaller basis) 3. Initial guess at MO coefcients C (semiempirical, only h

=
4. Construct R = C y C 5. Constuct the Fock matrix F = This is done integral-driven: a batch of integrals is read from disk and their contributions summed up into F . The storage for the matrices S S ;1=2 C R and F is of O (N 2 ).
= =
= = = = 6. Transform the Fock matrix: F 0 = S ;1=2 F S ;1=2 = = ==

7. Diagonalize the transformed Fock matrix: C 0yF C 0 = diag ;! C = S ;1=2 C 0
= ==
8. Convergence-Test: if energy change and density matrix change can be tolerated accept the actual matrices, else go back and repeat steps 4-8.
3.9.2
Direct SCF
The algorithm of the direct SCF implementation is slightly modied, specically in the steps 1 and 5: 1. Calculate integrals Spq and hpq and keep them in the memory. 2. Calculate S;1=2 3. Initial guess at MO coefcients C
4. Construct R = C y C 5. Constuct all (pq j rs) orbitals and sum them into F
= =
6. Transform the Fock matrix:
;1=2 F0 = S F S ;1=2 = ==
7. Diagonalize the transformed Fock matrix
10
3.10 Koopmans theorem

8. Convergence-Test: if energly change and density matrix change can be tolerated , accept the actual matrices, else go back and repeat steps 4-8. The changes have one minor disadvantage and several advantages. The disadvantage is, that one has to calculate the 2-el integrals Niter times. Nowadays, one typically has value of Niter = 5 ; 8. The advantages 4 3 2 + 2N ) integrals where N is the are: rst, that there is no disk space neccesary for the 1 8 (N + 2N + 3N 8 number of basis functions. (Example: N = 200 corresponds to 2 10 integrals = 1.6 GB). Second, the execution time is sped up since the slow disk I/O is avoided. Third, the algorithm can be nicely parallelized. To summarize the difference between the classical and direct SCF, one can state: direct SCF is especially suitable if the target problem involves large basis sets (with N & 100) and the target computer architecture provides fast CPUs and slow I/Os. Since that is true in almost every case today, direct SCF is widely applied, in contrast to classical SCF.
3.9.3
Speed up Hartree-Fock
Symmetry arguments: transform AO basis to symmetry-adapted basis. Thus, S and F factorize into blocks and many inte= = grals disappear. Observe however, that most molecules of interest are non-symmetric objects and one cannot exploit symmetry. Convergence accelerators: 1. Direct inversion of the iterative subspace DIIS. 2. Treat E as a function of C and extrapolate.
= n +1 n 3. Calculate C from C , C n;1 , C n;2 , ... = = = =

Integral cutoff
2. Ignore whole batches of integrals even before the begin of the calculations, e.g. ignore all d4 orbitals= 68 = 162 integrals. The consequences of these actions are advantageous in both methods: in classical SCF, the storage and I/O time are reduced and in direct SCF, the cpu calculation time is reduced.
1. Neglect (pq j rs) if (pq j rs) < 10;5 ; 10;7 EH
3.10
Koopmans theorem
We assume an N-electron Hartree-Fock single determinant with occupied spin orbitals of energies o and virtual spin orbitals of energies v . Herein, the virtual orbitals are artifacts of the method. They are basis set dependent and not convergent when applying greater basis sets since there is only the orthogonality requirement which is not sufcient for convergence. However, Koopmans theorem gives us a way of calculating approximate ionization potentials and electron afnities. He assumes (frozen orbital approximation) that the spin orbitals in the (N 1)-electron states, i.e. the positive and negative ions are identical with those of the N -electron state. Clearly, this approximation neglects relaxation of the spin orbitals in the (N 1)-electron states.
11
3 Hartree-Fock
The theorem states, that the energy level of the highest occupied state can be interpreted as a rst approximation of the ionization potential of the N -electron state and that the energy level of the lowest virtual state can be interpreted as electron afnity. Optimizing the spin orbitals in the (N 1)-electron single determinants by performing a separate HartreeFock calculation on these states would be a more reasonable, but costly performance. Doing so, we would get lower energy levels, thus the neglect of relaxation in Koopmans theorem tends to produce too positve ionization potentials and too negative electron afnities. Generally, the electron afnities obtained by Koopmans theorem are worse than the ionization potentials since the virtual states do not converge.
12
4 Basis functions - basis sets

4.1 Introduction
Modern quantum chemistry uses the expansion theorem to represent the wave function j i of a molecular systems in terms of basis functions. The following nomenclature is used:
j i j ii
= Ci j i i
i
(4.1)
^j =A
1 2 ::: n i
(4.2) (4.3)
j ki j'k i
= j'k !k i = Ck
(4.4)
where i is the Slater determinant of the many-electron basis, k are the spin orbitals, 'k are the spatial is the one-electron basis function (e.g. often molecular orbitals, !k is the spin function ( or ) and atomic orbitals).
f g denotes the basis set. Observe, that if the basis (both one-electron and many electron) is complete, the wave function would be exact. However, a complete basis involves an innite number of basis functions which is impossible to realize. Thus, there is always a basis set truncation error.
4.1.1
History
In the theory of the hydrogen atom, the following variables and equations are of interest:
nlm
= Rnl (r) lm (#) m (')

m (')
(4.5)
= 21 eim'
(4.6)
lm (#)
(2l + 1)(l ; jmj) 1=2 P jmj(cos#) l 2(l + jmj)!
(4.7)
Rnl(r) =
"
;
na0
2Z
(n ; l ; 1)! 2nf(n + l)!g3
#1=2
l+1 l ; =2 L2 n+l ( ) e
(4.8)
13

y
Figure 4.1: Denition of spherical coordinates
2Z r and a0 = h (Bohr radius), P the associate Legendre polynomials, L2l+1 the assowhere = na l n+l 4 2 e2 0 ciate Laguerre polynomials. Hydrogen functions have the important property that they satisfy the following orthonormality relation:
2
jmj
nlmjn l m
0 0
Z 1Z Z 2
0 0 0
nlm (r
# ')
0 0
n l m (r
0 0 0
# ')r2 sin #d'd#dr

(4.9)
= nn ll mm
0
The advantage of this method is that S = E due to the orthonormality relation, i.e. many integrals vanish. = = But on the other hand, there are several severe disadvantages: rst, the orthonormality relation only holds for one-atom systems and second, the integral evaluation is nearly impossible. Hence, the practical use is limited to atomic systems or to one-centre expansions which are exotic in molecular systems.
4.2
Slater-type orbitals
This method is a variation of hydrogen-like functions. The angular part lm (#) m (') is still evaluated like hydrogen orbitals, hence we keep the angular orthogonality. What is new is a simplication of the radial part: one applies some polynomial of simple power instead of the Laguerre polynomials:
n+1=2 p)(2n)! rn;1 exp(; r) Rn(r) = (2
(4.10)
;s is the Slater exponent with screening s and the effective radial quantum number n . There where = Zn are empirical rules (Slater rules) to determine the concrete values for s and n . As consequence, the radial functions are no longer orthogonal. On the other hand, the two-center integrals are now tractable and hence, the primary use of this method are calculations of diatomic molecules.
14
4.3 Gaussian-type orbitals
4.3
Gaussian-type orbitals
This method is one of the major inventions in Quantum Chemistry. It goes back to Boys in 1950.
4.3.1
Radial part
The main idea is to substitute polynomials or exponential functions of r with a Gaussian function:
exp(; r) ;! exp(; r2 )
It follows, that the functional form of all s-orbitals is the same. Specically, the function for the 1s-orbital reads as:
3 g1s (r) = 8 3
4.3.2 Angular part
1=4
exp(; r2 )
(4.11)
Generally, there are two possibilities to calculate the angular part of the integral: either like in the hydrogen atom by integrating over lm (#) m (') or by executing the integral over cartesian coordinates xl ym zn exp(; r2 ). The latter would e.g. involve the following functions for a p- and d-orbital:
g2px (r) = ( 1283 )1=4 x exp(; r2 )

1=4 2 g3dxy (r) = ( 2048 3 ) xy exp(; r )
In practice, one often proceeds as follows: Evaluate integrals over cartesian Gaussians, since that is fairly easy.
(4.12)
(4.13)
Transform to spherical Gaussians Observe: Cartesian d and higher sets give rise to spurious functions We give an example: 3d-orbitals: x2 ; y2 r2 ; z 2 xy xz yz (5 functions). cartesian: x2 y 2 z 2 xy xz yz (6 functions). Hence, there is one redundant function in the Cartesian set. After symmetry adaption, that function can be identied with the 3s-orbital: x2 + y 2 + z 2 = r 2 3s. Redundant functions make the basis set unnecessarily big, and can cause linear-dependency problems.
4.3.3
Advantages and disadvantages of GTOs
Figure 4.2 illustrates the major disadvantage of GTOs: Several GTOs per STO are needed to describe the nuclear cusp and to mimic the long-range bahaviour accurately. The major advantage of GTOs is that the integral evaluation becomes easier, faster and works also for 3 or 4 centers. That is due to the following reasons:
15
nuclear "cusp"
GTO
STO
r
Figure 4.2: Sketch of an GTO and an STO
1. A product of two Gaussians is a Gaussian. Example: Unnormalized 1s GTOs
~ A)g1s (~ ~ B ) = exp g1s(~ r;R r;R ~A ; R ~B R

2
~A R
exp
~B R
2
= exp
exp
;(
~P + ) ~ r;R
(4.14)
~P with R
~ A+ R ~B = R . +
For general cartesian GTOs:
l )(y ; Y m )(z ; Z n ) exp ( x ; XA A A
~A ~ r;R
2 2 2 2
(4.15)
l0 )(y ; Y m0 )(z ; Z n0) exp ; ~ ~B ( x ; XB r;R B B Pl+l (x) Pm+m (y) Pn+n0 (z) ~A 2 + ~ ~A 2 + ~ ~A exp ; ~ x;X y;Y z ;Z
0 0
exp
~B + ~ ~B + ~ ~B ~ x;X y;Y z ;Z
i.e. the GTOs factorize into x, y and z parts. We can take advantage of that, as can be seen in the following example of 2x orbitals:
16
4.3 Gaussian-type orbitals

+ RP RA RB
Figure 4.3: Illustration of the product of two 1s Gaussians.
(x ; XA ) (x ; XB ) exp (x ; XA ) (x ; XB ) exp exp

; ;
~A ~ r;R
2
exp
;( ;( ;(
~B ~ r;R
2 2 2 2
= =
(4.16)
+ (xP ; xA + xP ; xB )(x ; XP ) exp +(xP ; xA )(xP ; xB ) exp
~A ; R ~B R
~A ; R ~B R
2
exp
~P + ) ~ r;R ~P + ) ~ r;R ~P + ) ~ r;R ~P + ) ~ r;R

2
(x ; XP )2 exp
;(
~. i.e. the product of two px orbitals becomes a sum of an s, a px , and a dxx orbital at the center P
2. Laplace transform The 1 r (Coulomb) operator can be transformed into a Gaussian.
1=p 1
Z1
0
exp 1
;;sr2 psds
;
(4.17)
e.g. nuclear attraction integral:
ZZZ
exp
~A ~ r;R
;(
~cj exp j~ r;R

2 p
~B ~ r;R
d~ r=
(4.18)
ZZZZ Z
exp
~P + ) ~ r;R ~A ~ r1 ; R
2
s exp ;s (r ; Rc)2 d~ rds ~B ~ r1 ; R

2 2
This is similar for electron-repulsion integrals (2-electron-integrals):
exp
;
exp
;
(4.19)
j~ r1 ; r ~2 j exp
~C ~ r2 ; R
exp
~D ~ r2 ; R
d~ r1 d~ r2
17

Thus, all integrals over GTOs can be calculated analytically except for a non-analytical 1-dimensional integral
Fm (W ) =
Z1
0
t2m exp(;Wt2 )dt
(4.20)
that has to be calculated numerically (asymptotic series, continued function, power series, Gauss integration, gamma function, table interpolation, Chebyshev polynomials,....)
4.4
Application of basis functions
In the following, we summarize the typical eld of applications for the different methods. Atoms STO and numerical methods (nite differences in r ) Diatomics GTO (STO) (numerical methods: 2D nite differences, partial waves, 2D nite elements) Polyatomics STO (semiempirical) GTO (ab-initio) Continuum (and solid state) Plane waves exp(;i~ r Exotic and historical One-center expansions (hydrids) lobe functions: s-GTOs, but off-centre Floating spherical GTOs: same, but makes position variational parameter.
p ~) connected to STOs
4.5
CGTO (contracted Gaussians)
The GTOs have some deciencies. They dont give the nuclear cusp in a satisfactory way and they are short range. If one doesnt want to use a huge basis it is not easy to overcome this. One possible solution is to x a certain linear combination of Gaussians, i.e. contract them
l m n X d exp(; r2 ): i i lmn = x y z i
(4.21)
The di and i are adjusted once and for all to satisfy
18
4.5 CGTO (contracted Gaussians)

Carbon Huzinaga DZ contraction coefcient S61 4232.61 634.882 146.097 42.4974 14.1892 1.9666 S11 5.1477 S11 0.4962 S11 0.1533 P41 18.1577 3.9864 1.1429 0.3594 P11 0.1146 0.002029 0.015535 0.075411 0.257121 0.59655 0.242517 1.0 1.0 1.0 0.018534 0.115442 0.386206 0.640089 1.0 0.626 S61 4563.24 682.024 154.973 44.44553 13.029 1.82773 SP 3 1 20.9642 4.80331 1.45933 SP 1 1 0.483456 SP 1 1 0.145585 SP 1 1 0.0438 D11 Carbon 6-311G contraction coefcient 0.0019665 0.015236 0.0761269 0.260801 0.616462 0.221006 0.11466 / 0.0402487 0.919999 / 0.237594 -0.00303068 / 0.815854 1.0 / 1.0 1.0 / 1.0 diffuse function 6-311G+ 1/1 polarization function 6-311G* 1.0
Table 4.1: Two different carbon basis sets as tabulated in quantum chemistry programs. On the left, the Huzinaga-Dunning double-zeta (DZ) set, consisting of 1 contracted s (consisting of 6 GTOs), 3 uncontracted s, 1 contracted p (consisting of 4 GTOs) and 1 uncontracted basis functions. On the right, the 6-311G basis set, consisting of 1 contracted s (consisting of 6 GTOs), 1 contracted sp and 2 uncontracted sp basis functions. Note that for the sp, is the same for s and p, leading to efcient computation of the prefactors in products of two Gaussians. Additionally, the diffuse function and the polarization function is given for the 6-311G case (cf. section 4.8)
least-square STO (! STO-3G [3 Gaussians for a Slater], STO-4G, ...)
atomic HF calculation (Dunning-Huzinaga)
atomic correlated calculations (ANO). This is the most recent method.
Typical applications are highly contracted (lots of Gaussians xed together) inner shells to model the nuclear cusp. These wont change in the calculation. The valence shells need more exibility so one uses not so strong contraction or no contraction at all. A typical example is 6-31G which means that the 1s orbital is modeled by a contraction of 6 primitive GTOs. For the 2s and 2p orbitals one uses 2 GTOs each one CGTO consisting of 3 primitive ones and one being a bare one. CGTOs sometimes come with notations like (9s5p)![4s2p] what means that for the s-orbital 9 primitives are contracted to 4 CGTOs and 5 for 2 for the p-orbitals respectively.
19
4.6
Basis set size
1. There are minimum basis sets but except for few semi-empirical questions they are nowadays out of use. One has one basis function per occupied atomic orbital. However, shells have to be completed: In the boron atom, for example, only one p orbital is occupied. The other two p orbitals have to be included into the basis set, in order to complete the description of a boron atom in a molecule. The ab initio names go with STO-3G, MINI or similar. But most use is in semi-empirical calculations. 2. The next (and probably minimal useful) case is the split valence case. For the inner shells one function per orbital and for the outer two per orbital are used. So names like 3-21G mean 1 for the innermost and two for the next orbitals. 3. More accuracy provide double-zeta, triple-zeta ... which means 2,3... functions per atomic orbital. The name is historically fundated cause in the days of STOs the zeta was in the exponent exp(; r ).
4.7
Polarization functions
1 0 0 1
unperturbed spherical
1 0 0 1
electric field, lower symmetry
Figure 4.4: Helium atom in spherical state or polarized
Polarization functions are, e.g., necessary in the following situations: The Helium atom in its ground state and in absence of external elds may well be described by a 1s system. But the lower (cylindrical symmetry) needs additional (2p) orbitals. Also to form bonds one needs p-orbitals as is seen in Li2 . The Lithium atom lives in a 1s2 2s state which is not adequate for a bonded Li. In Be (1s2 2s2 ) electron correlation plays a role and low lying p (and d) states mix into the ground state. Also the calculation of excitation spectra for singly excited states with high transition moments needs polarization functions. 2! H2 : g Usually one uses polarization functions only for the valence shell. The corresponding notation (e.g. in Gaussian) reads in the case of one polarization function per atom: 6-31G* : 1 set of d functions on atoms heavier than He, 6-31G** : like 6-31G* additionally 1 set of p functions on H. In the case of more than one polarization function per atom, you need to specify it explicitly. 6-311g(3df 2p) means in this context 3 d functions and one f function on Li and higher additionally 2 p functions on hydrogen.
20
4.8 Gaussian exponents
4.8
Gaussian exponents
i i+1
The exponents of the Gaussians often form a geometric series:
const:
(4.22)
Basis sets of this kind are often referred to as even tempered basis sets. This may be used if for some reason a basis set has to be continued i.e. an additional function is needed. For very polarizable molecules, negative ions or Rydberg states so-called diffuse functions are used. This means that additionally one s and one p functions are used. The respective Gaussian notation is 6-311+g or 6-311++g for heavy atoms only or for hydrogens also. An example for the resulting basis set is given in table 4.5
4.9
Basis-set superposition error
In the study of interacting atoms almost always one encounters the problem of basis-set superposition error (BSSE). If the basis of atom A is incomplete (i.e. always) the basis of B will improve the wave function at A and therefore lower the energy. This BSSE can come to the same order of magnitude as the interaction energy itself. As an example consider the water dimer:
E = E (H2 O)2 ; 2E (H2O):

The different methods come with the following energies method STO 4G 4-31G
E (SCF )=kJ mol;1

-26.4 -32.2 -19.8 -15.4
541=31]
BSSE/kJ mol;1 -24 4 3
HF limit
If one wants to correct for the BSSE the Counterpoise correction may be appropriate. One calculates A with its own basis and with Bs basis (but without B). This method has some rigor for HF. Little is proven for higher approximations but it is used rather extensively.
11 00 00 11
A
11 00 00 11
B
Figure 4.5: Illustration of basis set superposition error (BSSE)
21
5 Electron correlation
5.1 Introduction
1 = 1 potential. So each electron feels what all the others do instantaElectrons interact via a j~ r12 r1 ;~ r2 j neously (if we disregard relativistic delay). So the motion of the electrons is correlated.
In the Hartree-Fock approximation one electron only sees the average distribution of the others not their instantaneous position. So due to this mean-eld approximation the treatment of electron correlation is incomplete.
5.2
Non-dynamic correlation effects
So correlated electron methods are in general more reliable than HF. One now distinguishes between dynamic and non-dynamic correlation effects. The names are somewhat misguiding. The non-dynamic effects are just everything for which single-determinant HF is not appropriate. But this has nothing to do with correlation effects. This deciencies can be resolved typically by quite simple means. As an example we look on the dissociation of H2 in minimum basis HF 1 . Let us write down the wave function
RHF
1 (1) (2) ; (2) (1)] =p 2 + sB : := sAp 2
(5.1) (5.2)
The overbar denotes the opposite spin. Now we have to perform some algebra
RHF
= 1 2 (sA (1) + sB (1))(sA (2) + sB (2)) ; (sA (2) + sB (2))(sA (1) + sB (1))] = 1 2 sA (1)sA (2) + sA(1)sB (2) + sB (1)sA (2) + sB (1)sB (2) ;sA (2)sA (1) ; sA (2)sB (1) ; sB (2)sA (1) ; sB (2)sB (1)] = 1 2 sA (1)sA (2) ; sA(2)sA (1) + sB (1)sB (2) ; sB (2)sB (1) +sA (1)sB (2) ; sA (2)sB (1) + sB (1)sA (2) ; sB (2)sA (1)]
(5.3)
Due to the symmetry of the problem with regard to exchanging nucleus A and B the rst two terms and the last two terms of the last equation are identical, leading to
RHF
= sA (1)sA (2) ; sA(2)sA (1) + sA (1)sB (2) ; sA(2)sB (1):
(5.4)
At bonding distance this makes no problem. But if we now look at the molecule at innite separation. In this case the second term is what we expect, namely two separate nuclei carrying their respective electrons.
1
The minimum basis is not causing trouble here. Better basis sets wont resolve the problem.
22
5.3 Dynamic correlation

But the rst term describes two ions an H+ and an H; . The energy would be
1 (s s js s ) ERHF (r = 1) = 2E (H ) + 2 A A A A
(5.5)
which is not just the two hydrogens. This could be corrected for by 2 determinants (which would in this case mean full CI). Also the old valence bond idea could resolve this where one just uses
1 s (1)s (2) ; s (2)s (1)]: =p A B 2 A B

Also unrestricted Hartree-Fock gave no such problems.
(5.6)
5.3
Dynamic correlation
Dynamic correlation manifests itself in real effects of instantaneous interaction of electronic charges. As an example we discuss the dispersion interactions between atoms or molecules (London, van der Waals). They are not accounted for in the HF limit (even for a complete basis set). The interaction between two Argon atoms would be strictly repulsive, so Ar(liq) should not exist. The problem is that the mean-eld center of charge does not coincide with the instantaneous center of charge. Due to uctuations the dipole moment of the atom does not vanish for all times.
~ (t) 6= 0
h~ i = 0
(t) 6= 0
h i
:::
(5.7) (5.8)
If you put the two atoms together (R > overlap distance) they feel each other and polarize each other mutually. So they induce dipole moments (and higher multi-poles) upon each other. The interaction of the form
B ind (t) Einter = ~ A(t)~ R3

leads to a lowering in total energy.This interaction is typically written in a series like
(5.9)
Edisper
; 6; R
C6
C8 ; C10 + ::: R8 R10
(5.10)
If we want to describe dispersive interaction there is no way out from electron correlation, i.e. a single determinant cannot do the job. And if it does it does it for the wrong reason, namely BSSE. Dispersion interaction puts also some requirements to the function in the basis sets. If the unperturbed atom is spherically symmetric (e.g. rare gas) the uctuating or induced dipole moments may not be describable this way. So one has to use polarization functions. For dipolar interaction at least p functions, for quadrupolar interactions d functions are necessary. So Helium atoms have to be described by
(1s)2 ! (1s)2 + c1 (1s2p) + :::]:

Never do electron correlation without polarization functions. Dispersion usually falls off with R;6 as leading term. The repulsion (Paulis principle) between two shells goes roughly exponentially, because the relevant quantity is the overlap of electron clouds. The wave functions fall off exponentially away from the core to leading order.
23
A dipole moment at B can of course be more easily induced if B is more polarizable. One expresses the energy as Taylor expansion in electric dipole eld (static case).
~ ~ ~ ) = E (F ~ = 0) + ~ perm: F ~ + 1F E (F 2 F + ::: ~ = @E ~ = ~ perm: + ~ ind: + ::: @F ~ + ::: ~ ind: = F

Here is a 3
(5.11) (5.12) (5.13)
3-tensor. So the dispersion goes proportional to the polarizability.
The exact relation (Casimir-Polder, 1947) for the leading coefcient in the dispersion series reads
C6 =
Approximate relations come to
Z1
0
A (i!) B (i!)d!:
(5.14)
3 E (RAB ) = ; R1 6 2
AB
1 IA
A B + I1 B
IA IB :
1st ionization potentials
(5.15)
3 A B =; 1 6 2q A q B RAB nA + nB
(London relation)
nA nB : effective numbers of electrons

(Slater-Kirkwood relation)
(5.16)
boiling points
200.0
125.7
100.0
36.1 -0.5 C4 H10 -42.1 C3H8 -88.6 C5H 12
98.4 69 C6 H14 C7 H16
C8H18
0.0
100.0
-164 CH4
C2H6 Kr -157.3 -185.7 Ar -245.9 Ne
Xe -107.1
200.0
-268.9 He
300.0
Figure 5.1: Boiling points increase with polarizability. The boiling point is a measure of dispersion energy of non-polar uids.
24
5.3.1
How much of the electron correlation is described by a single determinant?
An example for the treatment of electron correlations by a single determinant wave function is the following: Lets look at two electrons. Their wave function is given by
(x1 x2 ) =k 1 (~ r1 ) 2 (~ r2 ) (1) (2) k
with x = (~ r
!)
The probability density of nding electrons at the points (~ r1 ~ r2 ) in real space is
p(~ r1 ~ r2 ) =
Z 1 = 2 d!1 d!2 j 1 (1) (1) 2 (2) (2) ; 1 (2) (2) 2 (1) (1)j2 Z 1 = 2 d!1 d!2 j 1 (1)j2 j (1)j2 j 2 (2)j2 j (2)j2 +j 2 (1)j2 j (1)j2 j 1 (2)j2 j (2)j2 ; 1 (1) (1) 2 (2) (2) 2 (1) (1) 1 (2) (2) ; c:c:] 2 2 = 1 j2 j 2 (1)j2 2 j 1 (1)j j 2 (2)j + j 1 (2)Z 1 (1) (2) (1) (2) d! d! (1) (1) (2) (2) + c:c: ; 1 2 2 1 2 2 1
(2)) the integral in the last equation vanishes, as antiparallel + j 1 (2)j2 j 2 (1)j2
d!1d!2j (x1 x2 )j2
If the electron spins are antiparallel ( (1) spin eigenfunctions are orthogonal, so that
p(~ r1 ~ r2 ) = 1 2 ~ r1 = ~ r2 =: ~ r
j 1 (1)j2 j 2 (2)j2
Since the joint probability is just the product of the independent (1-electron) probabilities, the positions of the electrons are uncorrelated. Let the electrons now be in the same orbital ( 1 = 2 ) and consider the point
p(~ r~ r) = j 1 (~ r)j4 6= 0
This means that there is a nite probability for the two electrons to be at the same place in space. In reality this is impossible as there is the Coulomb interaction between the electrons with its singularity at zero distance. This effect is an artifact of the mean-eld approximation, as the electrons are treated as charge clouds and not as point charges. Lets now discuss the case of parallel spins ( (1) (2)). Here the spin integral evaluates as 1, so that
1 j (1)j2 j (2)j2 + j (2)j2 j (1)j2 p(~ r1 ~ r2 ) = 2 1 2 1 2 ; 1 (1) 2 (2) 2 (1) 1 (2) ; c:c] In the special case ~ r1 = ~ r2 =: ~ r this results in
p(~ r~ r) = 1 2
j 1 j2 j 2 j2 + j 1 j2 j 2 j2 ; 1 2 2 1 ; c:c
=0
So electrons with parallel spins cannot be in the same point in space even if they are in different orbitals (Fermi hole). To put it together the single-determinant wave function
25
EHF
EHF limit E Ecorr Efull CI Eexact small basis better basis basis set size
Figure 5.2: Illustration of basis size dependance of EHF , and dention of Ecorr
complete basis
(partially) correlates electrons with parallel spins. does not correlate electrons with opposite spins. does not correlate electrons within doubly occupied orbitals.
5.3.2
Correlation energy
A rigorous working denition of the correlation energy was given by L owdin. Considering the fact that HF is based on a variational principle it should give the lowest energy if a complete basis set is used. This energy EHF;limit is the lowest energy that can result from a single slater determinant wave function. If the exact energy (Eex ) is known, for example from experimental values, subtracting relativistic contributions, the difference between them is dened to be the correlation energy
Ecorr := EHF;limit ; Eexact

The difference between EHF and EfullCI is basis set dependent. Ecorr is only well dened for complete basis sets. Never calculate approximations using Ecorr and EHF obtained with different basis sets, for example EHF from a large basis and Ecorr from a small basis. If EHF;limit is wrong because of inadequacy of a single RHF determinant (e.g. H 2 dissociation) the denition of Ecorr makes no physical sense.
5.3.3
How to account for electron correlation
There are several methods to deal with the effect of electron correlation:
26

1. Clever choice of wave functions The usual Slater determinant is the antisymmetrised product of one-electron wave functions. A better basis is formed by functions that are not orbitals but depend also on the distance between electrons:
r1 ~ r2 ~ r1 ; ~ r2 ) k = k (~
Functions of this type were used by Hylleraas in 1930 to calculate the He atom, by James and Coolidge in 1936 and by Kotos and Wolniewicz around 1960 to calculate the H 2 molecule. The latter is the best calculation known. 2. Density functional theory (DFT) DFT is based on the Kohn-Sham equations
E=2
occ X i
1 Z dr dr (~ r1 ) (~ r2 ) + E (~ hii + 2 1 2 j~ xc r) r1 ; ~ r2 j
(~ r)]
Exc is a functional of the electron density and its gradient.
It is called exchange correlation. The Kohn-Sham equation would be exact if this functional was known exactly. In real life E exc is known only approximately and there are many other different approximations. It gives the electron density instead of the wave function. 3. Valence-bond If the structure and the dissociation behavior of the molecule is known, the results can be improved by a good choice of the wave function, but this method is difcult to extend to large systems, as there are many spins. 4. More determinants Instead of using only one slater determinant of the occupied orbitals, one can construct the wave function from several determinants, in which occupied orbitals are substituted by virtual ones.
= c0 jj 1 ::: n;1 n jj + c1 jj 1 :: n;1 n+1 jj + :::

where n+1 is a virtual orbital. If only the coefcients of the matrices are optimized the method is called Conguration Interaction (CI), if the orbitals within each determinant are optimized as well one speaks of Multi-Conguration SCF (MCSCF). 5. Perturbation theory Starting from the HF determinant as 0th order wave function, the electron correlation is regarded as a perturbation. This is called Many-body Perturbation Theory (MBPT).
27
6 Conguration Interaction
In the CI method the trial wave function is a linear combination of many Slater determinants k , which are themselves composed of molecular orbitals. These can be taken for example from HF calculations, which have the additional advantage of producing orthogonal orbitals. So the eigenvalue equation for the wave function
CI
has to be solved.
X
k
Ck
6.1
Properties of CI
CI is trivial in theory: 1. Calculate an orthogonal set of MOs f i g from a given set of AOs (using for example HF). 2. Construct all possible Slater determinants f k g of the MOs (taking into account the spin part as well). 3. Calculate the matrix elements of the Hamilton operator H kl 4. Diagonalize H As a result one obtains the energies and wave functions of all states that can be described by the basis, including excited states. In reality, however, things are a bit more complicated due to the following points: ad 2. The number of determinants is gigantic. If the basis has got there are
= h k j H j l i.
N elements and there are n electrons,
N n
determinants. The problem is NP-complete, the number scales factorial with N . A practical exam52 ple is Benzene in a 6-31g** standard basis. Here is N = 288, n = 42, which gives about 10 determinants. ad 3. The calculation of the Hamilton matrix is even worse, as its dimension is the number of determinants. In the example, one would have to calculate a 1052 1052 matrix, which needs 1093 TByte. Even a 10000 10000 matrix needs almost 1 GByte. ad 4. The Hamiltonian matrix has to be diagonalized. This is a N3 problem. How can these difculties be overcome?
N! = n!(N ; n)!
28
6.2 Matrix elements between Slater determinants

ad 2. Number of determinants Full CI is only possible for very small systems, e.g. double-zeta H 2 O. In all other cases the number of determinants has to be restricted. Keeping in mind that the inuence of higher order excitation determinants on the ground state energy is small, the set of determinants can be restricted to single excitation determinants (CIS) or to singles and doubles (CISD). The energy contribution of every determinant in a set is estimated (perturbation theory) and all determinants the contribution of which lies below a given threshold are discarded. It is very important to make use of every symmetry of the problem. If one constructs conguration state functions (CSF), which are spin eigenfunctions, the H matrix is block-diagonalized. This makes use of spin symmetry. With those tricks one can get along with 103 to 109 determinants or CSFs. ad 3. Storage of the matrix As it is almost impossible to store H matrix, the matrix elements are recalculated every time they are needed by the diagonalization algorithm (direct CI). Only the coefcient vectors are stored in memory. ad 4. Diagonalization Usually only the lowest eigenvalues and eigenvectors are of interest. To obtain them it is not necessary to diagonalize the matrix fully. Instead a simple iteration algorithm may be used (Shavitt, Davidson). It can be improved by convergence accelerators. Another problem is how to store the possible combinations of MOs for the determinants. A very elegant solution has been found by Duch and Karwowski using graph theory. The electrons and MOs form vertices of a graph. The determinants are represented as paths in this graph.
6.2
Matrix elements between Slater determinants
The calculation of the matrix elements is simplied by some useful rules. The general matrix element is
Hkl = hK j H jLi
where
H = H 1 + H2 =
Case 1: jK i Case 2: jK i jLi Case 3: jK i jLi
X
i
hi +
X
ij
1: r
ij
If more than two orbitals are different in k and l the matrix element vanishes. The other cases are:
= jLi = = = =
hK j H1 jK i hK j H2 jK i j:::m:::i hK j H1 jK i j:::p:::i hK j H2 jK i j:::mn:::i hK j H1 jK i j:::pq:::i hK j H2 jK i
N hmj h jmi = P m 1 N hmnj jmni = 2 mn = h m j h jpi P hmnj jpni = N n = 0 = hmnj jpqi
29
6 Conguration Interaction
6.3
Properties of the CI expansion
The CI wave function can be written as an expansion in the number of excited states in the determinant. If lower indices denote occupied orbitals and upper indices mean virtual (=excited) orbitals, the CI ansatz up to second order can be written as
j 0 i = c0 j 0 i +
X
ar
cr aj
ri + a
abrs
crs ab j
rs i + ::: ab
where 0 is the Hatree-Fock determinant. One expects that the singly excited determinants are closest in energy and will contribute most to the ground state wave function, while higher orders contribute less and less. Consider now an expansion of only two terms
r j 0 i = c0 j 0 i + cr a j ai
The CI equation is given by the secular problem
h 0j H j 0i h 0j H j r h r aj H j 0i h aj H j
1
ri a ri a
c0 cr a
= E0
c0 cr a
= 0:
The mixing between states is determined by the off-diagonal elements.
X h r har jbbi ; habjbr i = haj F jr i = a ar a j H j 0 i = haj h jri + b
F denotes the Fock operator, which is of course diagonal in the basis of the HF orbitals, its eigenfunctions.
The generalization of this result is called Brillouins theorem: There is no direct interaction between the HF determinant and singly excited determinants. This is also true for triply and higher excited determinants (cf. chapter 3). Even though, there are of course indirect interactions between all determinants. Only doubly excited congurations do have direct interaction with the HF state, so that they contribute most to the correlation energy. In spite of this, in most calculations the singly excited determinants are used as well, because is not very expensive, they contribute by indirect interaction via doubles, when calculating one electron properties the matrix element can be huge and so cancel out the smallness of the coefcient (e.g. dipole moment). An example: Dipole moment of CO Method SCF SCF+138D SCF+200D SCF+138D+62S Experiment E -112.788 -113.016 -113.034 -113.018
-0.108 -0.068 -0.072 +0.030 +0.044
Another problem occurs when the symmetry of the problem changes in a reaction for example. Look at the reaction 2C2 H4 !C4 H8 . The MOs are as follows:
30
6.3 Properties of the CI expansion

Symm. AA: AS: SA: SS: MO (1A-1B-2A+2B) (1A-1B+2A-2B) (1A+1B-2A-2B) (1A+1B+2A+2B)
R = 1 R =short ( 1 ; 2 )) ( A ; B ) ( 1 + 2 )) ( A ; B ) ( 1 ; 2 )) ( A + B ) ( 1 + 2 )) ( A + B )
R = short
The energy levels can be depicted in gure 6.1
R = oo 1 2 1 + 2
1 2 1 + 2
+ +
Figure 6.1: Energy level crossing for the reaction 2C2 H4
!C H
4 8
31
7 Review of Perturbation Theory

The problem to be solved pertubativly is
Hj
We assume that the eigenvalue problem
i i = Ei j i i = (H0 + V ) j i i
(7.1)
H0
(0) i
= Ei(0) i(0)
(7.2)
can be solved with reasonable accuracy (or even exactly). is known to be much smaller than H0 . We formally introduce a scaled perturbation
V V
is called the perturbation Hamiltonian, which
H = H0 +
(7.3)
where we will later take ! 1. (0) = jii In the following we make use of the notation: i Now let us expand the eigenvectors and eigenvalues of our pertubation problem in a Taylor series in
(1) 2 (2) + : : : i + i (0) (1) (2) Ei = Ei + Ei + 2Ei + : : : (n) (n) We call the coefcients i n-th order wave functions and the Ei n-th order energies.
j ii
= jii +
(7.4) (7.5)
7.1
Auxiliary theorem
hijii
Let us assume that the jii form a properly normalized basis set
=1
(7.6)
We chose the normalisation of i such that

hij i i = 1
(7.7)
This intermediate normalisation is always possible, unless jii and j i i are orthogonal to each other (we wont discuss this case, it just complicates the formulas and can be looked up in standard text-books) This implies:
hij i i
= |{z} hijii + 1
ij
(1) i
D (2) E ij + ::: {z i }
2
32
7.1 Auxiliary theorem

If this is to be true for all values of , it follows
)
ij
(n) i
=0
for all n > 0,
(7.8)
i.e. higher-order wave functions are orthogonal to the zero-th-order wave function. Let us plug in the Taylor expansion (7.4) into the full Schr odinger equation (7.1):
(H0 +
V ) jii +
jii +
(1) i (1) i
= Ei(0) + Ei(1) + 2 Ei(2) + : : :

If this expression is to be true for all values of be equal:
2 2
(2) i (2) i
+ :::
+ ::: :
on both sides must
the coefcients of various orders of
H0 jii = Ei(0) which is just the unperturbed Schr odinger equation. H0 H0 :::
1: 2: (1) i (2) i
0:
E E
+ V jii = Ei(0) i(1) + Ei(1) jii + V i(1) = Ei(0) i(2) + Ei(1) i(1) + Ei(2) jii
We multiply these conditions form the left by hij

hij H0 jii = Ei (0) (0) ! Ei hijii = Ei (nothing new)
0:
(0)
+ hij V jii = Ei(0) ij i(1) + Ei(1) hijii D E (0) ij (1) + hij V jii = E (1) ! Ei i i D (1) E (1) ! Ei = hij V jii since ij i = 0.
hij H0
1:
(1) i
+ hij V i(1) = Ei(0) ij i(2) + Ei(1) ij i(1) + Ei(2) hijii D E E (0) ij (2) + hij V (1) = E (2) ! Ei i i E i (2) (1) ! Ei = hij V i
hij H0 ! Ei . . .
2:
(2) i
3:
(3)
= hij V i(2)
33
7.2
How to calculate n-th order energy?

(n)
1. solve for (n-1)th-order wave function 2. calculate Ei
= hij V i(n;1)
Example: 2nd-order energy
H0 Ei(0) ; H0
(1) i
+ V jii = Ei(0) i(1) + Ei(1) jii = V

; Ei
(1) i
(1) jii = (V ; hij V jii) jii
multiply by a different eigenfunction hnj 6= hij of H0 from the left:

hnj (Ei
(0) ; H
(1) 0) i
= hnj V jii ; hij V jii hnjii

(1) i (1) i
using orthogonality of hnj and hij we get
Ei(0) nj i(1) ; hnj H0 D (Ei(0) ; Ei(1) ) nj
E E
= hnj V jii = hnj V jii
D
Expand i
nj
(1) i
hnjV jii = E (0) ;E (1)
(7.9)
(1) in eigenfunction of H 0 (1) i
X
n
c(1) n jni
(7.10)
where the cn are simply given by (7.9):
(1)
D
which nally gives:
mj
(1) i
E E
= =
X
n (1) ci
c(1) n hmjni =
=0
X
n
c(1) n
mn = c(1) m
ij i(1)
(1) i
X
n6=i
c(1) n jni =
XD
n6=i
nj
(1) i jni =
X D
n6=i
jni
nj
(1) i
(7.11)
Now we can take this result for the i
(1) and plug it into the formula for the second order energies: (1) i
Ei (2) = hij V
=
n6=i
E X
=
n6=i
hij V jni
nj
(1) i
X hij V jni hnj V jii X j hij V jni j2 =

(0) Ei(0) ; En
(0) (0) n6=i Ei ; En
(7.12)
34
7.3 Treating electron correlation via perturbation theory

Summary: What is needed for 2nd-order energy? 1. hij V jni The integrals of perturbation Hamiltonian between all unpertubed eigenstates of H 0, 2.
(0) En
and all eigenvalues of the unperturbed Hamiltonian.
Similarly, but more complicated, this holds for higher orders: e.g. the 3rd-order energy:
Ei(3) =
X X hij V jni hnj V jmi hmj V jii

(0) (0) (0) n6=i m6=i (Ei ; En )(Ei ; Em )
(0) ; Ei
(1)
the effect is precalculatable , H : : :
(0) (0) n6=i (Ei ; En )2
j hij V jni j2
(7.13)
7.3
Treating electron correlation via perturbation theory
Many-body perturbation theory (MBPT) Rayleigh-Schrdinger perturbation theory (RSPT) Moller-Plesset perturbation theory (MPPT,MP2,MP4,...)
The unperturbed system is described by
H0 j H0 = E0 =
6=
0 i = E0 j 0 i
where H0 is the Hartree-Fock Hamiltonian (not the Fock-Operator).
N X N X i i
f (i)
i= N X
EHF =
X
a
hij H jii +
XX
i j
haj h jai ;
1 X X habj jabi 2 a b
hij j jij i
The perturbation is described by
V = H ; H0
= (
N X i N X
h(xi ) + h(xi) +
N X N X
= ( = =
i=1 j>i ij N X N X
N 1 );X f (xi) r
i N N XX
i=1 j>i ij i N N 1 ; X X V (x )) HF i i=1 j>i rij i i
1 r );(
i N X
h(xi ) +
N X i
VHF (xi ))
(total electron-electron repulsion)(Hartree-Fock electron repulsion)
35
VHF (xi ) j (xi ) =

=
X
b
hbj
1 jbi (x ) ; X hbj 1 jj i (xx ) j i b i rij rij

b
coulomb exchange
This is equivalently written in our stream-lined notation:

hij VHF jj i =
X
b
(hibjjbi ; hibjbj i) =
X
b
hibj jjbi
(7.14)
Thus the 1st order energy becomes
(1) = E0
h 0j V j 0i h 0j
(0) + E (1) E0 0
X X = 1 habj jabi ; haj VHF jai 2 ab a X = ;1 2 ab habj jabi X 1X habj jabi = EHF = a;2 a ab
i<j rij
j 0i ; h 0j
X
i
VHF (xi) j
0i
I.e. the rst order pertubation correction restores the HF solution !
7.4
Second-Order Energy (MBPT)
Our general formula gives:
(2) = E0
We know j0i = j 0 i, but what are the jni ? singly excited states
h 0j V j r ai
X j h0j V jni j2
(0) (0) n6=0 E0 + En
= = =
h 0 j (H ; H0 ) j r ai r h 0 j H j a i ; h 0 j H0 j r ai
| {z }
0
; h 0j f j r a i = haj f jri = a ar
(Brillouin theorem)
=0
since jai jr i are eigenfunctions of f .
36
7.4 Second-Order Energy (MBPT)

doubly excited states jni = j rs ab i
H0 j
rs i ab
(0) j rs i = (E (0) ; ; + + ) j rs i = Eabrs a b r s ab ab P 0
(2) = E0
XX jh
a<b r<s
0j
XX
= 4 abrs a + b ; r ; s
a<b r<s a + b ; r ; s 1 X j habj jrsi j2
a+ b; r; s j habj jrsi j2
1 j rs i j2 ab r i<j ij
triply- and higher-excited states do not interact directly with j 0 i MBPT2 singles doubles triples quadruples MBPT3 MBPT4
D E2
-
S E4 D + ED+ ED + ED + ED E2 3 2 3 4 T E4 (expensive) Q (dont need all) E4
Table 7.1: Contributions to correlation energy. Note: truncated schemes like MBPT4(SDQ) are also size consistent.
AO integrals SCF integral trans. MBPT4 S MBPT4 D MBPT4 T MBPT4 Q total
CO2 (63 basis fcts.) 20 10 27 1 7 11 6 82
CO3 (84 basis fcts.) 62 30 173 4 88 216 45 618
Table 7.2: Time effort [CPU min.] (after S. Canuto 1989)
37
8 Size Consistency
One desirable property of quantum-chemical methods has been brought into QC by solid-state people. In simple words, for a given calculated molecule A with energy E A one constructs a supermolecule AA consisting of two A molecules, and treats them as one non-interacting molecule, i.e one runs a calculation with two A molecules being largely separated. The method is said to be size-consistent if E AA = 2EA . In the next three paragraphs, we will check the size consistency for HF, DCI and MBPT2 on the simple example of A = H2 in the minimum basis.
8.1
Size Consistency of HF
Remember that the wave function for one H2 in the minimal basis HF solution is
A = j A Ai
where denotes the 1s-MOs. The energy of the molecule A is then
(8.1)
EA =
2 X
= 2 h Aj h j Ai + h A Aj j A Ai :
For the supermolecule AA one has for the wavefunction
i=1
h Aj h j Ai +
1 2
h A A j j A A i + h A A j j A A i]
(8.2)
AA = j A A B B i
and the energy
(8.3)
EAA =
= + + +
2 X
4 h Aj h j Ai + 1 2 h A Aj j A Ai + h A h A A j j A Ai + h A h B Aj j B Ai + h B h B Aj j B Ai + h B
i=1
h Aj h j Ai +
4 X
i=3
h Bj h j Bi +
4 1X 2 a b=1 habj jabi
(8.4)
Aj j A A i
+ h A Bj j A Bi + h A Bj j A Bi Aj j B Ai + h B B j j B B i + h B B j j B B i i j j i + h j j i + h j j i A B A B B B B B B B B
Aj j A Ai + h A B j j A B i + h A B j j A B i
The underlined terms are the self-interaction parts of the two electron integrals, which are all zero (Remember: hij j jkli = hij jkli ; hij jlk i in physicist notation). The overlined terms are also zero, because we
38
8.2 Size Consistency of MBPT2

consider two H2 with distance going to innity, which means there is no overlap between remaining non-zero contributions give:
A and B .
the
EAA = 4 h
1 4h = 4 h Aj h j Ai + 2 h j j
Aj h j Ai + 2
jj i
i]
(8.5)
Thus we showed on this simple example that the HF method is size consistent. In fact, one can show that this is a general result.
8.2
Size Consistency of MBPT2
It is sufcient to consider the second order correction energy only, as we already know that the lower order terms are equal to the HF terms, which is size consistent. Remember: The second order energy correction in MBPT2 has the form
;1 X X j h 0j P rij j =
(2) E0
a<b r<s
a+
rs i j2 ab ; ; r s b
i, rs ab jh
XX
)
j habj jrsi j2 a<b r<s a + b ; r ; s
(8.6)
which means in our example for H2 ( 0
=j
=
j ;
2= E0 2(
Looking at the H4 -supermolecule ( 0
) :
j2
(8.7)
= j A A B B i) we rst notice there are two double excited states: rs = or A A B B (8.8) A A B B ab
The two other possible combinations (cross excitations) die on symmetry. The corresponding second order energy correction is
(2) (AA) = E0
j h B Bj
2 B B j
B;
j ;
j h A Aj
= 2 2(
jh
) :
B j2
2 A A j
A
A;
(8.9)
Again good news: In this example, MBPT2 is size consistent. Even better news: One can show rigorously that MPBT is size consistent in all orders in all contributions (singles, doubles, ...)
8.3
Trouble: Double CI Size Consistency
To perform our simple example for this case, we notice that for one H 2 , the determinants with one virtual MO do not contribute directly due to Brillouins theorem. But, the indirect contribution is also zero because
39
8 Size Consistency
of symmetry arguments (the HF soulution has g whereas the singly excited terms have u ). Therefore survives in DCI. Remark: In this simple case, DCI is equivalent only the doubly excited term rs ab = to full CI. For the two non-interacting H2 , the same symmetry reasoning yields zero contribution for the singly and triply excited wave functions. Therefore we are left with the doubly excited states. But note that there also exists a quadruply excited state (j A A B B i). Knowing that, it is easy to see that DCI cannot preserve size consistency: The quadruple excited state is not included in DCI for the H 4 , therefore we do not get twice the energy of the DCI (=full CI) of H2 . For a rigorous proof see Szabo/Ostlund, p.261. In general, the message is: truncated CI is not size consistent, full CI is size consistent.
40
no dyn
llm He yn nd ex am an am
siz ec lc ica ynm -Fe is ons l ica ten . orr cor an t r. iati var ona l
ds cite tate s
HF + + + + +
mean-eld description, single determinant
only UHF
+ () + + +
(Koopman theorem) + + +
limited include explicitly excited determinants selected according + CI to some prescription MCSCF small CI with orbitals optimized for every determinant + evaluate contributions of excited determinants consistently MBPT through given order in perturbation theory evaluate contribution of selected excitation through innite CC order full CI include all possible determinants, exact within basis set +
Table 8.1: Properties of different ab-initio methods
+ only if based on UHF only if based on UHF +
8.3 Trouble: Double CI Size Consistency
41

9.1 The Potential energy surface
If the Born-Oppenheimer approximation
molecule
~~ R r =
nuclei
~ R
r) electrons (~
(9.1)
holds the effective potential energy for the nuclei as a function of their positions R is dened by
~ =h V R
~ ) j eli + el j Hel (R
X ZAZB
RAB
A<B
(9.2)
This equation denes a potential energy surface PES (really a hypersurface) for the nuclei. The electronic Hamiltonian depends parametrically of the nuclear coordinates
1 ; X ZA + X 1 ~ = Hel R ; r2 2 i A riA i<j rij i

A schematic PES is shown in gure 9.1. Features of the PES can be related to molecular properties: geometries: A,C minimum (molecular structure) B saddlepoint (transition state) energetics: energy differences between minima (reaction enthalpies, stabilities) energy differences between minima and saddlepoints (activation energies) curvature: vibrational frequencies, normal modes (IR, Raman, thermochemistry) higher derivatives: (accurate) ro-vibrational spectra minimum-energy path between minima (A ! B ! C ): reaction path?
X"
(9.3)
9.2
How to nd stationary points?
Minima on the PES represent molecular conformers, saddlepoints transition states. To locate them, one makes use of optimisation techniques from numerical mathematics which use derivatives. The PES is ~ 0, Taylor-expanded around a stationary point R
~ = V R
~0 + X @V V R i @Ri X X @2V +1 2 i j @Ri @Rj
;R ; R0 i i ~ R ;R ; R0 ;R ; R0 + : : : i j i j ~
0
(9.4)
R0
42
9.2 How to nd stationary points?
Figure 9.1: Illustration of a potential energy surface (PES)
where the sums run over all relevant degrees of freedom. The rst derivatives are components of the gradient vector
@V gradient ~ g : gi = @R i @V Hessian H : Hij = @R i @Rj

The Taylor expansion can be written more compactly
(9.5)
(Note that the negative of the gradient is equal to the force acting along this degree of freedom.) The mixed second derivatives form the Hessian matrix
(9.6)
~ =V R ~0 + ~ ~ ;R ~0 + 1 R ~ ~ V R g R 2 ; R0
At a stationary point, we have
~ ;R ~0 R
(9.7)
~ 0 = 0: ~ g R
If the stationary point is a minimum, H is positive denite: It has only positive eigenvalues after the ones belonging to translation and rotation of the molecule as a whole have been removed. If the stationary point is an n-th order saddlepoint, H has n negative eigenvalues. Physically most important are rst-order saddlepoints. Minima can be located by standard numerical techniques, such as simplex or Fletcher-Powell (uses only V , no derivatives); steepest-descent, conjugate-gradient or variable-metric (uses gradients); Newton-Raphson (uses second derivatives). See a textbook on numerical methods for details (e.g. Chapter 10 of Numerical Recipes). Which method is being used depends not only on the specic molecular system, but also on the computer budget, the desired accuracy and most of all on the availability of analytical rst and maybe second derivatives. We will see that they can be quite complicated to calculate. Hence, not all derivatives have been implemented for all electronic-structure methods.
43

The location of saddlepoints is not as straightforward. Methods include mode-walking (follow the softest eigenmode) and others.
9.3
Physical meaning of some derivatives of the potential energy
~ i or an electric Some derivatives of V with respect to an external perturbation, like the position of a nucleusR ~ eld E , have a direct physical interpretation (to within a numerical factor): ~i = ; @V f ~i @R 2 @V fij = @R i @Rj ~ = @V ~ @E 2 @V ij = @E @E i j @~ = @ 2 V ~ i @ E@ ~ R ~i @R 3 @ V = ~@ ~ ~ ~i @ Ri @ E@ E@ R
9.4
: : : : : :
force on atom i harmonic force constant dipole moment static electric dipole polarisability (3x3 Cartesian tensor) dipole derivative (IR intensity) polarisability derivative (Raman intensity)
The Hellmann-Feynman theorem

, a useful relationship can be proven. The electronic energy
For the exact electronic wave function given by
E is
E = h jH j i:
Its derivative with respect to some parameter is obtained via the product rule
jH j i + h jH j
(9.8)
Since H is Hermitian, E can be extracted from the last two terms
@E = h j @H j i + h @ @ @ @
@ @
i
(9.9)
@E = h j @H j i + E h @ @ @ @
Using the product rule backwards, we get
j i+h j
@ @
(9.10)
@E = h j @H j i + E @ @ @ @
Since
h j i:
(9.11)
is normalised to some constant, h j i is independent of
and the last term is zero, i.e. (9.12)
@E = h j @H j @ @
which is called the Hellmann-Feynman theorem.
44
9.5 Gradients for RHF

Therefore, once the wave function of a system has been found, derivatives of the energy can be calculated simply by calculating expectation values of the corresponding derivative of the Hamiltonian. Derivatives of the wave functions with respect to lambda are not needed. This simplies calculations. The perturbation lambda can be any external perturbation, for example the ones given in section 9.3. Two remarks concerning the Hellmann-Feynman theorem: 1. It is always possible to determine derivatives numerically, for instance by calculating = 0 + and = 0 ; . Then, one may use e.g. centered differences:
E for =
0,
@E @ @2E @ 2
E( 0 +
0 0
1 E( + 0 2
) ; E( 2
0;
)
0;
(9.13)
) + E(
) ; 2E ( 0 )]
Finite differences are convenient (no derivatives necessary) but the number of energy calculations quickly runs out of hand, if there are many degrees of freedom (or s): Imagine a calculation of a Hessian matrix for a molecule of 30 atoms. If analytical second derivatives are available the complete Hessian can be evaluated at a fraction of the cost. 2. The derivation above assumes that is the EXACT wave function. Of the approximate wave functions used in quantum chemistry some satisfy the Hellmann-Feynman theorem, others dont. The ones that do satisfy the Hellmann-Feynman theorem are those that have been optimised with respect to all possible parameters. The parameters have to include the coefcients of the basis functions (except for full CI which is the exact solution within the given one-electron basis). Hence, the methods satisfying Hellmann-Feynman are: exact wave function Hartree-Fock (and self-consistent density functional) MCSCF full CI The Hellmann-Feynman theorem is not strictly satised by, e.g. incomplete CI MBPT (except in innite order!) coupled-cluster many others For these wave functions, one may of course calculate analytic derivatives. However, the @ has to be properly taken into account.
=@
term
9.5
Gradients for RHF
Where the Hellmann-Feynman theorem is not applicable, the wavefunction derivatives must also be calculated. Simplest example: Calculation of the energy function derivative in closed shell RHF.
45

The Hamiltonian is
ESCF = 2
0 occ X @
i
hii +
occ X
where the one electron part hii , the Coulomb part Jij and the exchange part Kij are
= i = energy of orbital i
1 2(Jij ; Kij )A {z }
(9.14)
hii = Jij =
=RR =RR
2 P ZA i ;1 2 r ; rij A
d iijjj ] ij jij ] :
(9.16) (9.15)
( ) j ( 2) d 1 d 2 = i ( 1 ) j ( 2 ) r1 12 i 1 ( ) j ( 1) d 1 d 2 = i ( 1 ) j ( 2 ) r1 12 i 2
Kij =
To evaluate the energy derivative
occ occ X occ @ESCF = 2 X @ h +X @ (2J ; K ) ii ij ij @ @ @ i i j
we rst consider the one electron part:
@h = h@ @ ii @
=
pq
all X
all X
i jh j i i + h i j @ j i i + h i j hj @ i i all @ C h jhj iC + X @ h jhj i C ip p q qi ip @ @ p q pq | {z def : = hii
@h
(9.17)
Cqi
pq
Cip h
pj h j q i
@C @ iq
Here the rst and last term contain the -dependence of the matrix elements over atomic orbitals (which can be calculated once and forever), while the second term has to be determind in the calculation. The derivative of the wave function can be written as linear combination of all (occupied and unoccupied) molecular orbitals
where the coefcients Uik express the admixture of other molecular orbitals (! perturbation theory). Thus
all @ i =X Uik @ k
(9.18)
occ occ occ X all X X @ h = 2X h + 4 Uik hik ii @ ii
(9.19)
The derivative of the coulomb interaction and the exchange interaction can be rewritten in an analogous manner, e.g.
all @ J = @ iijjj ] = iijjj ] +4 X ij | {z } k Uki kijjj ] @ @ def : = Jij
(9.20)
46
9.5 Gradients for RHF

where again the derivative contained in Jij needs to be calculated only once:
Jij =
Thus
all X
pqrs
Cip Ciq Cjr Cjs @@
h p qj
r12 j
r si
(9.21)
occ occ X occ occ X all X X X @E = 2 h + 2 J ; K + 4 Uik hik ii ij ij @ SCF i i j i k
(9.22)
+4 =
occ X all occ X X
0 occ X @
1 2hii + 2Jij ; Kij A i j | {z } = (hii + i ) 0 1 occ X all occ X X +4 Uik @hik + (2 kijjj ] ; kj jij ])A i k j {z } |
occ X occ
Uik
(2 kijjj ] ; kj jij ])
X
i
hii +
+4
X
occ
= hij f^(i) jki = i ik
Uii
It remains to express the coefcients Uik in terms of derivatives of atomic orbitals. This may be done by introducing the projection operator for moelcular orbitals
S=
all X
j ii h ij
(9.23)
and expand it in atomic orbitals, in complete analogy to eq. (9.17):
Sij = h i j S j j i = h all P = Cip Cqj h pj

pq
The derivative yields
i j j i = ij qi
h ij Sj @ @ h ij @ @
(9.24)
@ @ @ Sij = 0 = h @ i jS j j i + = h @@ i j j i + = Uji +
) Uij
h i j @S j j i @
+ + +
ji ji
(9.25)
Sij Sij
Uij
= ;1 2 Sij
all X
where in the third step eq. (9.18) has been inserted, and
Sij =
def.
pq
Cip Cqj @@
h pj q i
(9.26)
47

Finally,
occ @ E =X @ SCF i hii +
;2
occ X
Sii
(9.27)
48

In the framework of density functional theory it can be shown, that the ground state properties of a quantum r) = j (~ r)2 j . It is thus not necessary mechanical system can already be obtained from the electron density (~ r) itself. The ground state density (~ r) minimizes the Kohn-Sham to know the the complex wavefunction (~ energy functional
Z 1 EKS ] = 2 hij h jii + 2 i

occ X
Z (~ ~0 ) 3 3 0 r) (r d rd r + Exc
~0 j j~ r;r
(10.1)
where Exc
] is the exchange-correlation functional.
Note the formal analogy to the HF energy equation
EHF = 2
occ occ X occ X X X hij h(i) jii +2 hij Jj jii ; ij jij ] j | i {z } | i {z } |i j {z }

1-electron Coulomb exchange
(10.2)
where the best single slater determinant for the ground state is obtained from the HF variational equations. However, the HF method gives an approximation to the wavefunction, while the Kohn-Sham equations would yield the exact ground state density, provided that the exact form exchange-correlation functional were known (for all interesting cases this it is unfortunately not known exactly).
10.1
An illustrative example: The Thomas Fermi model

3 as particles V =`
Consider an ideal electron gas. N non-interacting electrons occupy a cubic volume in a box. The energy of the electrons is then discretized according to
h2 (n2 + n2 + n2 ) h2 R2 E (nx ny nz ) = 8ml y z 2 x 8ml2
(for sufciently large n)
(10.3)
The Pauli exclusion principle implies that the number of electrons with energy < (at T the volume of a sphere octant
= 0K ) is given by
(10.4)
4 R3 = 8ml2 ( )=21 8 3 3 h2
3=2
The degeneracy factor 2 appears since two electrons with opposite spin are allowed to have the same n x, ny , nz . Thus, the density of states is
2 g( ) = ( + ) ; ( ) = 2 8ml h2
1=2
+ O ( )2 ]
(10.5)
49

Since for T
= 0K the sphere is completely lled up to the Fermi energy F , the total energy in V
E=
Taking into account that
3 N E = 5 F 3h2 3 = 10 m 8 =
def.
N = ( F ) and introducing the electron density = N= V

2=3
0 3=2 3 5=2 8 m = 5 2 ` F 2 h
RF g( )d
is just
(10.6)
we nd (10.7)
V V
5=3 5=3 in atomic units m = ~ = 1
= CF = 2:871
3 (3 2 )2=3 10 | {z }
If the electron density is allowed to vary from point to point (inhomogeneous electron gas), the total energy (still equal to the total energy)is given by
E = TTF ] = CF
5=3 (~ r)d3 r
(10.8)
Nuclear attraction and Coulomb repulsion are added for the more general case (but not exchange/correlation, which partially are accounted for in TTF ]):
ETF ] = CF
5=3 (~ r)dr3 ; Z
Z (~ r) d3 r + 1 Z Z (~ r) (~ r0 ) d3 rd3r0 r 2 j~ r ;~ r0 j
(~ r)d3 r ; N )
(10.9)
In order to determine the electron density it was then assumed (already in 1927!) that the correct minimizes ETF ] with the constraint of constant number of electrons, i.e.
(~ r)
0 =
!
TF
Z (~ Z r 0 ) d3 r ETF ] = 5 C (~ 2 = 3 r ) ; + F (~ r) 3 r ~ r ;~ r0
ETF ] ;
TF
with
N=
(~ r)d3 r
(10.10)
Note, however, that this variational principle contrary to DFT has not been derived rigorously from QM principles. It is rather based on physical intuition by noting the important role that variational principles play in physics. In general, Thomas Fermi theory gives in some sense averaged electron densities, i.e. radial oscillations showing up in more rigorous approaches (like HF) are missing. Numerous extensions to Thomas Fermi theory have been derived, e.g. the von Weizs acker kinetic energy functional (1935)
TTFW ] = CF
10.2
5=3 (~ r)d3 r +
8m
~2
jr
(~ r)j2 d3 r (~ r)
(10.11)
The Hohenberg-Kohn theorems
A QM system is completely determined by the number of electrons N and the external potential electrostatic potential induced by the nuclei, external elds, : : : ). Then it holds, that
v(~ r) (i.e.
50
10.2 The Hohenberg-Kohn theorems

HK I: The external potential is determined, within a trivial additive constant, by the electron density (~ r). (uniqueness) Proof: (for non-degenerate ground states) 0: Assume, that the same density (~ r) results for two different external potentials v and v
v(~ r) ! H ! E0 < h
Analogously
0j
and
v0 (~ r) ! H 0 !
(10.12)
The the variational principle for the ground state energy E 0 yields
Hj
0i
0 j 0i + h 0j H ; H 0 j 0i = h 0j H R 0 = E0 + (~ r) v(~ r) ; v0 (~ r)]d3 r :
(10.13)
0 < h j H0 j i = E + E0 0
R (~ r) v0 (~ r) ; v(~ r)]d3 r :
(10.14)
Summation of eq. (10.13) and eq. (10.14) leads to the contradiction

0 0 E0 + E0 < E0 + E0
(10.15)
which completes the proof. Consequences of HK I: Since
(~ r) determines uniquely v(~ r) (cf. HK I) (~ r) determines N =

In order to determine
R (~ r)d3 r
the system is completely described by
(~ r), an energy functional is needed:
(~ r). Consequently all properties can be derived from (~ r).
E ] = T ] + Vee ] + Z Vext ] = F (~ r)v(~ r)d3 r | HK {z }] +
{z
Hohenberg-Kohn functional (universal) with
(10.16)
system dependent ext. potential
FHK ] = T ] +
=
RR
J ] |{z}
Coulomb
+Exc ]
(10.17)
(~ r) (~ r) 3 3 0 j~ r;~ r j d rd r
0 0
where Exc ] is the non-classical exchange correlation functional, which is difcult to obtain and must be approximated in all interesting cases. The second Hohnberg Kohn theorem states
51

HK II: For a trial density (~ r) with (~ r) lower bound to the energy functional Ev ]:
E0
Proof: I (~ r) ! v(~ r) HK ;! H !
e Z Ev e] = FHK e] + e(~ r)v(~ r )d3 r
0 and N =
R e(~ r)d3 r, the true ground state energy E
0 is a
(10.18)
where the equality actually holds for the correct ground state density (~ r).
Ev e] = e H e
h j H j i = Ev
] = E0
(10.19)
Consequence of HK II: The energy functional is variationally stationary for the correct ground state density, i.e:
Ev ] ;
or
(~ r)d3 r ; N ) = 0
(10.20)
] = FHK ] + v(~ = Ev r) : (~ r) (~ r)
(10.21)
The latter equation tempts to interprete as a chemical potential in the sense of Paulings electronegativity (= arithmetic average of the electron afnity and the ionization potential), although a rigorous relationship cannot be established.
10.3
The Kohn-Sham (KS) method
We want a method to solve
E ] = T ] + J ] + EXC ] +
(~ r)v(~ r)d~ r
(10.22)
The three rst terms of the l.h.s. are FHK ] where the second and third terms can be summarized as before as Vee ]. The contribution of the terms to the energy is very different in magnitude, in particular T ] >> EX > EC where the exchange term EX is typically ten times greater than the correlation term EC . Note that the virial theorem gives simply < T >=< H >. The Thomas-Fermi approach consists of solving T ] directly from as an approximation to the true kinetic energy functional. The Kohn-Sham approach is to introduce an auxiliary wave function so that the largest part of T ] can be evaluated exactly and one worries about the rest later. The exact kinetic energy is
T=
N X i
ni h
i j ; 2 r2 j i i
(10.23)
where the s are called natural orbitals and are no more than the eigenfunctions of the reduced 1-electron density matrix and contain the many-electron effects. n i is an effective occupation numbers that may be fractional. The electron density is given thus by
(~ r) =
N X X i
ni
s=
j i (r
s)j2
(10.24)
52
10.4 Exchange (and) correlation functionals

The following simplied kinetic term TS
] is introduced :
N X i N X
h i j ; r2 j i i
TS ] =
(~ r) =
1 2
(10.25)
The s are now the molecular orbitals of the determinantal wave function S = p1 N det j 1 2 ::: N j with the occupation number 0 1 2. The s are eigenfunctions of the Hamiltonian without V ee .
i s=
j i (r
s)j2
1 i= i i 2 X X 1 TS ] = h sj (; 1 r2 )j s i = h i j ; r2 j i i 2 2 i i
hS i =
r) ; r2 + vs (~
Because TS ] = 6 T ] the KS exchange-correlation functional contains not only the usual terms (exchange,correlation,electronique repulsion) but also a kinetic energy correction term
E ] = TS ] + J ] + (Vee ] ; J ]) + (T ] ; TS ])] so that the terms in square brackets are the components of E XC ].
The Euler equation is given by
(10.26)
= veff (~ r ) + TS ]
with
(10.27)
veff (~ r) = v(~ r) + J ] + EXC ] Z Z (r ~0 ) ~0 rdr + vXC (~ r) = v(~ r) + ~0 j d~ j~ r;r vXC (~ r) being the exchange correlation potential. The KS equations for N orbitals are
; r2 + veff (~ r)
1 2
(~ r) =
XX
i s
i= i i
(10.28)
r i (~
s) i(~ r s)
with the constraint h i j j i = ij . Hence, it is possible to recalculate in a self consistent procedure veff (~ r) = f ( (~ r)) including the exchange-correlation potential from the electron density via the auxiliary wave function.
10.4
Exchange (and) correlation functionals
We have seen that the DFT is exact and KS is exact, if EXC estimated with the local-density approximation (LDA)
] can be calculated exactly. EXC ] can be

(10.29)
LDA = EXC ] EXC
(~ r) XC ]d~ r
53

(Note that exchange and correlation are not separated.) The philosophy is that the length scale of the exchange and correlation is smaller than the length scale of the variation of . (This should be compared with the wave function theory : The HF exchange term is not local and neither is the CI correlation term.) LDA The LDA is of course nearly exact for an uniform and less accurate for a strongly varying . v XC , vXC in the LDA approximation, is given by
LDA LDA = EXC ] = vXC (~ r)

The KS equation with the LDA becomes
XC
(~ r)] + (~ r) XC ]
(10.30)
Z (r ~0 ) 1 2 LDA (~ ; r + v (~ r ) + r + vXC r) i = i i (10.31) ~0 j d~ 2 j~ r;r For the uniform electron gas XC = X + C . X is known exactly : 1 3( 3 )1 3 3 (~ r). C is numerically well known. Today XC is nearly accurate for electron gas. X = ;4
3 3 (~ HFS (~ 3 vXC r) = ; 2 r)] 1
"
For atoms and molecules, the LDA is made using the Hartree-Fock-Slater (HFS) or X introduced by Slater in 1951. The idea is to replace the HF exchange operator by a local operator. In practice, the uniform electron gas results is applied to innitesimal portions of the system (10.32)
For Slater = 1 while Kohn and Sham proposed that X is the exchange term (without correlation) if 2 . Empirically, takes values between 0:7 and 0:75. In comparison with the Thomas-Fermi approach, =3 the difference is mainly in the kinetic energy but there is a large improvement. The local spin density (LSD) approximation is similar in spirit to the unrestricted HF method.
N (~ r) = (~ r) + (~ r) N =
The KS equation becomes
; r2 + veff (~ r)] i
(~ r)d~ rN =
(10.33)
(~ r)d~ r
N =N +N
1 2 = i i
i = 1 ::: N
(10.34)
veff (~ r) = v(~ r) +
~0 ) EXC (r ] d~ r 0 ~ (~ r) j~ r;r j
] + EC
~ Other terms could be added in the r.h.s. of the last equation like e.g. 2e r) to take into account the mc Bz (~ presence of a magnetic eld. We have as for the LDA case addition of the exchange and correlation terms
EXC
For the exchange term
] = EX
EX
4 3 3 1 Z ( )4 LSD = 2 1 3 ( )3 3 +( 3 ]d~ ) r ] EX 4 Z 4 1 4 4 3 )1 3 3 (1 + ) 3 + (1 ; ) 3 ]d~ = 23 3 ( ( ) r 4
(10.35)
54
10.5 DFT in a Gaussian orbital basis

with
LSD is an ugly and complicated term. being the spin polarisation. EC The exchange correlation terms can also been estimated by gradient correction. In this case the exchange correlation term is improperly called non-local although it is in fact still mathematically local. Some of them are listed here:
One example is the von Weizs acker correction to Thomas-Fermi. Langreth and Mehl:
LM EXC
LDA + 2:14 ] = EXC
10 ;3
jr j2
4 3
2 exp(;0:26) r7
7 ]d~ 9 r
(10.36)
Perdew and Yue have developped the generalised gradient approximation (GGA)
GGA ] = ; 3 3 EX 4
1 3
4 3
F s]d~ r
2
(10.37)
S = 2kjr (j~ r)
F
and
kF = (3
)3
F (S ) = (1 + 1:296 S 2 + 14 S 4 + 0:2 S 6 ) 15
B EX HFS ] = EX
Becke: It is the most useful numerically.
];
X Z
=
jr
7 3
j2
(1 + B jr
8 3
j2
);1 d~ r
where A and B are parameters. Combinations of gradient correction have been developped :BP, BLYP, BVWN ... There is also the possibility to combine exchange-correlation functionals with a HF exchange term (e.g. B3LYP): HF ] + (1 ; a) E B ] + bE LY P ] EXC ] = a EX X C
10.5
DFT in a Gaussian orbital basis
The KS equation in orbital basis is
H =
KS
; iS =
~ ci = 0
(10.38)
'i =
N bas X i
Cip p(~ r)
pj q i
(10.39)
where Nbas is the number of basis functions. We have then
Spq = h HKS pq = h
1
(10.40)
r)j q i + h pjJ j q i + h p jXC j q i p j ; 2 r2 + v(~
(10.41)
55

where the second and third term of the r.h.s. is the coulomb and exchange term respectively. The Coulomb term can be expressed in various ways. In the DFT language it is
h p jJ j q i =
ZZ
r) q (~ r) p(~
j~ r;r j
~0 ) ~0 (r rdr ~0 d~
(10.42)
In the quantum chemistry language, the MO are explicitely introduced

h p jJ j q i =
occ X 1 'j (2)'j (2)]d p(1) q (1) r
occ N bas X X j rs
Cjr Cjs
12 j
(10.43)
p (1) r (2) r
12
q (1) s (2)d
The calculation are expensive : the Coulomb and HF exchange term are of the order of N bas 4 . Hence, r) (Gaussian) are used : instead charge tting functions gt (~
(~ r) =
N fit X t
ct gt (~ r)h pjJ j
qi =
N fit X t
ct
p (1) q (1) r
1 g (2)d t
12
This is computationally cheaper, the order being decreased to N bas 2 Nfit / Nbas 3 The problem is now to nd the charge tting coefcients ct The simplistic method (in practice not done) would use the following minimisation :
min ( (~ r) ; fit (~ r))2 ! min min

where with the constraint
Z X
i
r) i (~ r) ; i (~
(r ~2 )d~ r1 d~ r2
c g (~ r) t t t
!2
d~ r
A better way is to minimize the residual Coulomb energy
ZZ
(r ~1 ) r1
12
ZX
h p jXC j q i =
(~ r) = (~ r) ;
c g (~ r) t t t
c g (~ r)d~ r=N t t t
p (1) q (1) r
We can also t functions to the exchange-correlation potential
ZZ
1 v ( (r r1 d~ r2 XC ~2 ))d~
12
vXC is local (even with gradient corrections) and is calculate on a grid

gridpoints
0 " #2 1 X X min @ vXC (~ r) ; Cu gu(~ r) A

u
3 The exchange correlation scales also with the order N bas
gu(~ r) being a gaussian function
KS 2 h Nbas = 3 J Nbas = 3 XC Nbas =
HF 2 Nbas 4 (N3 possible) Nbas bas 4 K : N bas (reduction impossible) =
56
11 The Car-Parrinello Method

The Car-Parrinello (CP) technique is the combination of electronic structure calculation, giving the energy and the forces and a molecular (dynamics) simulation. It has the advantages, from the quantum side to be accurate to use no force-eld the chemisty (reactions ...) is possible and from the molecular simulation side the environment temperature entropy dynamics statistical mechanics The disadvantages are of course that it is less accurate than a quantum chemistry calculation and still expensive in computation time with regard to force-eld simulation (good statistics only for a few picoseconds, and only few hundreds of atoms) The ingredients are DFT for electrons local density approximation or gradient correction plane wave basis set (faster than Gausssians) pseudopotentials in lieu of core electrons. (This is mandatory to use because of the preceding ingredient) (classical) molecular dynamics for the nuclei ctious dynamics for electrons. This is the important trick of the method parallel implementation
57
11 The Car-Parrinello Method
11.1
Plane wave basis
The plan wave basis functions are given by
r) = i (~
~ ):~ k+G r ~ exp i(~ k+G ~ Ci ~ G
(11.1)
~ is the reciprocal lattice vector dened by G ~ where G

11.2 The Kohn-Sham equations
~e2 2 r +V
~ l = 2 m with ~ l a lattice vector and m an integer.
2m
r) + VH (~ r) + VXC (~ r) ion (~
XC
r) = i i (~ r) i (~
@ (~ r)
with Vion (~ r)
VH (~ r) = e2
potential.
v(~ r) where the subscript ion means external potential due to the atomic cores (ions), R = (r~ ) dr r)] the exchange-correlation ~0 the Hartree Coulomb potential and V (~ r) = @EXC (~
~j j~ r;r
0 0
The KS equation is rewritten in reciprocal space and with help of the plane waves
~ G
~e2
0
~ ~2 ~G ~ 0 ) + VH (G ~ 0) + VXC (G ~ 0) Ci ~ ~ ;G ~ ;G ~ ;G ~ + Vion (G ~ = i Ci ~ ~ k+G k+G 2m jk + Gj G (11.2)

0
2 is replaced by a The plane waves are orthonormal, i.e. there is no overlap matrix. The Laplace operator r 2 e ~ ~2 multiplication operation in Fourier space. The energy cutoff ~ 2m jk + Gj < Ecut / 100eV .
11.3
Pseudopotential
The valence orbitals are orthogonal to the inner(core)-shell orbitals of same symmetry. It implies radial nodes close to the nucleus, rapidly oscillating function. They cannot be represented feasibly by plane waves. The inner electrons and small r part of valence orbitals are not changed by chemical bonding. The rst solution is all electron calculations in plane-wave basis with sui table atom-centered basis functions (LMTO,LAPW,PAPW, ...). The second solution is the use of a pseudo-potentials. They leave the outer part of the valence orbitals unchanged. The radial nodes are removed and they replace the inner electrons. This should be compared to atom-centered basis (e.g. Gaussians) in molecular quantum chemistry: The effective core potentials (ECPs) are used for speeding up the calculation with fewer electrons and keep the relativistic effects (e.g. for heavy elements). ECPs are optional.
11.4
CP Molecular dynamics
The equation of motion for the nuclei is
X ZAZB d2 R @ E (R ~ ~ MA dt ( t ) = ; ( t ) ( t )) + A KS i 2 ~A @R A>B RAB

58
"
#
(11.3)
11.4 CP Molecular dynamics

and for the electrons
X d2 (t) = ; @ E (R ~ ( t ) ( t )) + i KS i j dt2 @ i
i k+G
ij j
(11.4)
The last term in the r.h.s. is the orthonormality constraint while is a ctious mass, a parameter. It implies that the acceleration of the coefcient of the plane wave C ~ ~ is given by
X ~ ~0 d2 C ~e2 ~ ~ 2 ~ 0 )VXC (G ~ 0)]Ci ~ ~ ;G ~ ;G = ; ( j k + G j ; ) C ; Vion (G ; G ) + VH (G i ~ ~ ~ k +G i~ k +G k+G dt2 i ~ 2m ~ (11.5)

0
~ 0) is the contribution of the nuclei and i = ii = h i jH j i i is an approximate ~ ;G where Vion (G constraint coming from the subsequent iterative orthogonalisation of the i s like, e.g.
i= i;2
0
1 Xh
j 6=i
j j ii j
Viewed from MD, the CP-MD is an extended-system method. Viewed from DFT, CP-MD is avoiding a diagonalisation of the Hamiltonian.
59

Practical QUantum

Uploaded by

Copyright:

Available Formats

Practical QUantum

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Practical QUantum

Uploaded by

Copyright:

Available Formats

Practical Quantum Chemistry

Script of a crash course held by

March 16th20th, 1998

Lecture notes prepared by

3.7 3.8 3.9

3.10 Koopmans theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4 Basis functions - basis sets 4.1 13

Slater-type orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Gaussian-type orbitals 4.3.1 4.3.2 4.3.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Radial part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Angular part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Advantages and disadvantages of GTOs . . . . . . . . . . . . . . . . . . . . . . . 15

4.4 4.5 4.6 4.7 4.8 4.9

5 Electron correlation 5.1 5.2 5.3

6 Conguration Interaction 6.1 6.2 6.3

7 Review of Perturbation Theory 7.1 7.2 7.3 7.4

8 Size Consistency 8.1 8.2 8.3

Size Consistency of HF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Size Consistency of MBPT2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Trouble: Double CI Size Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 42

9 Derivatives of the Energy

10 Density functional theory (DFT)

= i qi ri , ^ = i qir ^i P ^ = ; Pi ~2 r2 kinetic energy K = i p2 i =2mi , K i =2mi

^ is obtained as The expectation value of A ^i = hA

^ is the Hamiltonian operator where H

^ (t) i~@t (t) = H

Postulate 7 (Spin) Pauli exclusion (anti symmetry) principle for Fermions.

2.2 The Born-Oppenheimer approximation

The Born-Oppenheimer approximation

Eel ' Evib rot

Units, Symbols and Notations

e (elementary charge) ea0 2 e a2 0 =EH EH =ea0 ;3=2 q0

2.3.3.1 Dirac bras and kets:

2.3.3.2 Two electron integrals notation: physicists notation

~1 ) spin orbital i, occupied by electron 1. i (x

(~ x1) j (~ x2 ) r1 k (~ x1) l (~ x2 )d3 x1d3 x2 12

x1 ) l (~ x2 )d3 x1 d3 x2 (~ x1 ) j (~ x2 ) r1 k (~ 12 ij jkl] = hikjjli

antisymmetrised 2-el. integrals

= hij jkli ; hij jlki = ikjjl] ; iljjk]

spatial orbitals ' chemists notation

real or complex real

and H3 is the repulsion of nuclei in molecules (mostly treated clasically)

In the one electron electron theory the, following approximation is made:

Paulis antisymmetry principle

2 1 p (permutation of 1 and 2:) 2

::: ::: . . . ::: ( N ) ::: 2

P Z =r + modications ) H uckel, extended H uckel 6= Hartree-Fock A a iA

Here, f is the Fock operator

N X ZA X 1 2 ^j (1) ; K ^ j (1) f (1) = ; 2 r + r + J iA j =1 A

^j (1) j i (1)i = i (1)j J

= iijjj ] = hij jij i

^ j (1) exchange operator K

^ j (1) j i (1)i = i (1)j K

ij jij ] = hij jjii

x ? "# ? ? "# i? ? "# ? ? "# x ? " ? ? " ? ? i ? "# ? "# ? ?

restricted open-shell Hartree Fock (ROHF)

spatial MOs singly or doubly occupied

^2 spins not E.F. of S

j lower energy, but not pure doublet

Hartree-Fock-Roothaan equations (for RHF)

(i = 1 : : : Ne =2) ^j (1) ; K ^ j (1)) (2J

non-symmetric matrix eigenvalue problem. Transformation to a symmetric matrix ! diagonalise ! , C

3.7 Matrix elements

Digression: How to solve the matrix equation?

If S were the identity matrix E , then we would have