The Hartree-Fock
Approximation
Shyue Ping Ong
Stationary Schrödinger Equation for a System of
Atoms
where
NANO266
2
Eψ = Hψ
H = −
h 2
2me
∇i
2
i
∑ −
h 2
2mk
∇k
2
−
e2
Zk
rikk
∑
i
∑ +
e2
rijj
∑
i
∑
k
∑ +
Zk Zle2
rkll
∑
k
∑
KE of electrons
KE of nuclei
Coulumbic attraction
between nuclei and
electrons
Coulombic repulsion
between electrons
Coulombic repulsion
between nuclei
Stationary Schrödinger Equation inAtomic Units
To simplify the equations a little, let us from
henceforth work with atomic units
NANO266
3
Dimension Unit Name Unit Symbol
Mass Electron rest mass me
Charge Elementary Charge e
Action Reduced Planck’s constant ħ
Electric constant Coulomb force constant ke
H = −
1
2
∇i
2
i
∑ −
1
2mk
∇k
2
−
Zk
rikk
∑
i
∑ +
1
rijj
∑
i
∑
k
∑ +
Zk Zl
rkll
∑
k
∑
TheVariational Principle
We can judge the quality of the wave functions by
the energy – the lower the energy, the better. We
may also use any arbitrary basis set to expand
the guess wave function.
How do we actually use this?
NANO266
4
φHφ dr∫
φ2
dr∫
≥ E0
Linear combination of atomic orbitals (LCAO)
NANO266
5
http://www.orbitals.com/
Solving the one-electron molecular system with
the LCAO basis set approach
In general, we may express our trial wave functions
as a series of mathematical functions, known as a
basis set.
For a single nucleus, the eigenfunctions are
effectively the hydrogenic atomic orbitals. We may
use these atomic orbitals as a basis set for our
molecular orbitals. This is known as the linear
combination of atomic orbitals (LCAO) approach.
NANO266
6
φ = aiϕi
i=1
N
∑
The Secular Equation
NANO266
7
E =
aiϕi
i=1
N
∑
"
#
$
%
&
'H aiϕi
i=1
N
∑
"
#
$
%
&
'dr∫
aiϕi
i=1
N
∑
"
#
$
%
&
'
2
dr∫
=
aiaj ϕiHϕj dr∫
ij
∑
aiaj ϕiϕj dr∫
ij
∑
=
aiaj Hij
ij
∑
aiajSij
ij
∑
Resonance integral
Overlap integral
The Secular Equation,contd
To minimize the energy,
Which gives
Or in matrix form
NANO266
8
∂E
∂ak
= 0, ∀k
ai (Hki − ESki )
i=1
N
∑ = 0, ∀k
H11 − ES11 H12 − ES12 ! H1N − ES1N
H21 − ES21 H22 − ES22 ! H2N − ES2N
" " # "
HN1 − ESN1 HN 2 − ESN 2 ! HNN − ESNN
"
#
$
$
$
$
$
%
&
'
'
'
'
'
a1
a2
"
aN
"
#
$
$
$
$
$
%
&
'
'
'
'
'
= 0
The Secular Equation,contd
Solutions exist only if
Procedure:
i.  Select a set of N basis functions.
ii.  Determine all N2 values of Hij and Sij.
iii.  Form the secular determinant and determine the N roots Ej.
iv.  For each Ej, solve for coefficients ai.
NANO266
9
H11 − ES11 H12 − ES12 ! H1N − ES1N
H21 − ES21 H22 − ES22 ! H2N − ES2N
" " # "
HN1 − ESN1 HN 2 − ESN 2 ! HNN − ESNN
= 0
HückelTheory
Basis set formed from parallel
C 2p orbitals
Overlap matrix is given by
Hii = Ionization potential of
methyl radical
Hij for nearest neighbors
obtained from exp and 0
elsewhere
NANO266
10
Sij =δij
The Born-OppenheimerApproximation
Heavier nuclei moves much more slowly than
electrons => Electronic relaxation is
“instantaneous” with respect to nuclear motion
Electronic Schrödinger Equation
NANO266
11
(Hel +VN )ψel (qi;qk ) = Eelψel (qi;qk )
Electronic energy
Constant for a set of
nuclear coordinates
Stationary Electronic Schrödinger Equation
where
NANO266
12
Eelψel = Helψel
Hel = −
1
2
∇i
2
i
∑ −
Zk
rikk
∑
i
∑ +
1
rijj
∑
i
∑
KE and nuclear attraction terms are separable
H = hi
i
∑ where hi = −
1
2
∇i −
Zk
rikk
∑
Hartree-ProductWave Functions
Eigen functions of the one-electron Hamiltonian is given
by
Because the Hamiltonian is separable,
NANO266
13
hiψi =εiψi
ψHP = ψi
i
∏
HψHP = hi
i
∑ ψk
k
∏
= εi
i
∑
#
$
%
&
'
(ψHP
The effective potential approach
To include electron-electron repulsion, we use a
mean field approach, i.e., each electron sees an
“effective” potential from the other electrons
NANO266
14
hi = −
1
2
∇i −
Zk
rikk
∑ +Vi, j
where
Vi, j =
ρj
rij
∫
j≠i
∑ dr
Hartree’s Self-Consistent Field (SCF)Approach
NANO266
15
Guess MOs
Construct
one-
electron
operations
hi
Solve for
new ψ
hiψi =εiψi
Iterate until energy
eigenvalues converge
to a desired level of
accuracy
E = εi
i
∑ −
1
2
ψi
2
ψj
2
rij
dri drj∫∫
What’s the purpose
of this term?
What about the Pauli Exclusion Principle?
Two identical fermions (spin ½ particles) cannot
occupy the same quantum state simultaneously
è Wave function has to be anti-symmetric
For two electron system, we have
NANO266
16
ψSD =
1
2
ψa (1)α(1)ψb (2)α(2)−ψa (2)α(2)ψb (1)α(1)[ ]
=
1
2
ψa (1)α(1) ψb (1)α(1)
ψa (2)α(2) ψb (2)α(2)
where α is the electron spin eigenfunction
Slater determinant
For many electrons…
NANO266
17
ψSD =
1
N!
χ1(1) χ2 (1) ! χN (1)
χ1(2) χ2 (2) ! χN (2)
! ! " !
χ1(N) χ2 (N) ! χN (N)
where χk are the spin orbitals
The Hartree-Fock (HF) Self-Consistent Field
(SCF) Method
NANO266
18
fi = −
1
2
∇i
2
−
Zk
rik
+Vi
HF
{j}
k
nuclei
∑
F11 − ES11 F12 − ES12 ! F1N − ES1N
F21 − ES21 F22 − ES22 ! F21 − ES2N
" " # "
FN1 − ESN1 FN 2 − ESN 2 ! FNN − ESNN
= 0
HF Secular Equation
Fµυ = µ |−
1
2
∇i
2
|υ − Zk µ |
1
rk
|υ + Pλσ
λσ
∑ (µυ | λσ )−
1
2
(µλ |υσ )
$
%&
'
()
k
nuclei
∑
Weighting of four-index integrals by density
matrix, P
Flowchart of HF SCF Procedure
NANO266
19
Limitations of HF
Fock operators are one-electron => All electron
correlation, other than exchange, is ignored
Four-index integrals leads to N4 scaling with
respect to basis set size
NANO266
20
Ecorr = Eexact − EHF
PracticalAspects of HF Calculations
Basis Sets
Effective
Core
Potentials
Open-shell
vs Closed-
shell
Accuracy
Performance
NANO266
21
Basis Set
Set of mathematical functions used to construct
the wave function.
In theory, HF limit is achieved by an infinite basis
set.
In practice, use finite basis sets that can
approach HF limit as efficiently as possible
NANO266
22
Contracted Gaussian Functions
Slater-type orbitals (STO) with
radial decay cannot be
analytically integrated
-> Use linear combination of
Gaussian-type orbitals (GTOs)
with radial decay to
approximate STOs
STO-3G
•  STO approximated by 3 GTOs
•  Known as single-ζ or minimal basis set.
NANO266
23
e−r2
e−r
Multiple-ζ and Split-Valence
Multiple-ζ
•  Adding more basis functions per atomic orbital
•  Examples: cc-pCVDZ, cc-pCVTZ (correlation-consistent polarized
Core and Valence (Double/Triple/etc.) Zeta)
Split-valence or Valence-Multiple-ζ
•  Still represent core orbitals with single, contracted basis functions
•  Valence orbitals are split into many functions (Why?)
•  Examples: 3-21G, 6-31G, 6-311G
NANO266
24
# of primitives in
core
# of primitives in
valence
Polarization and Diffuse Functions
Polarization functions
•  Description of MOs require more flexibility than provided by AOs,
e.g., NH3 is predicted to be planar if using just s and p functions
•  Additional basis functions of one quantum number of higher
angular momentum than valence, e.g., first row -> d orbitals
•  Notation: 6-31G* [old] or 6-31G(d) [new], 6-31(2d,p) [2d functions
for heavy atoms, additional p for H]
Diffuse functions
•  Highest energy MOs of anions, highly excited states tend to be
more diffuse
•  Augment standard basis sets with diffuse functions
•  Notation: 6-31+G, 6-311++G(3df, 2pd), aug-cc-pCVDZ
NANO266
25
Effective Core Potentials
Heavy atoms have many electrons
•  Intractable to model all of them, even with a minimal basis set
•  However, most of the electrons are in the core
Solution: Replace core electrons with analytical
functions (effective core potentials or ECPs) that
represent combined nuclear-electronic core to the
remaining electrons
Key selection decision: How many electrons to
include in the core?
NANO266
26
Open-shell vs closed-shell
Restricted HF (RHF)
•  Closed-shell systems, i.e., no unpaired electrons
Restricted open-shell HF (ROHF)
•  Use RHF formalism, but with density matrix for singly occupied
orbitals not multiplied by a factor of 2.
•  Wave functions are eigenfunctions of S2
•  But fails to account for spin polarization in doubly occupied orbitals
Unrestricted HF (UHF)
•  Includes spin polarization
•  Wave functions are not eigenfunctions of S2, i.e., spin contamination
NANO266
27
Accuracy
Energetics
•  In general, extremely poor; correlation is extremely important in
chemical bonding!
•  Protonation energies are typically ok (no electrons in H+)
•  Koopman’s Theorem: First IE is equal to the negative of the
orbital energy of the HOMO
Geometry
•  Typically relatively good ground state structures with basis sets of
modest size
•  But transition states (with partial bonding) can be problematic, as
well as some pathological systems
NANO266
28
Performance
Formal N4 scaling
But in reality, speedups can be achieved through:
•  Symmetry
•  Estimating upper bounds to four-index integrals
•  Fast multipole and linear exchange integral computations
For practical geometry optimizations, frequently
helps to first compute geometry with a smaller
basis set to provide a better initial geometry and a
guess for the Hessian matrix.
NANO266
29

NANO266 - Lecture 2 - The Hartree-Fock Approach

  • 1.
  • 2.
    Stationary Schrödinger Equationfor a System of Atoms where NANO266 2 Eψ = Hψ H = − h 2 2me ∇i 2 i ∑ − h 2 2mk ∇k 2 − e2 Zk rikk ∑ i ∑ + e2 rijj ∑ i ∑ k ∑ + Zk Zle2 rkll ∑ k ∑ KE of electrons KE of nuclei Coulumbic attraction between nuclei and electrons Coulombic repulsion between electrons Coulombic repulsion between nuclei
  • 3.
    Stationary Schrödinger EquationinAtomic Units To simplify the equations a little, let us from henceforth work with atomic units NANO266 3 Dimension Unit Name Unit Symbol Mass Electron rest mass me Charge Elementary Charge e Action Reduced Planck’s constant ħ Electric constant Coulomb force constant ke H = − 1 2 ∇i 2 i ∑ − 1 2mk ∇k 2 − Zk rikk ∑ i ∑ + 1 rijj ∑ i ∑ k ∑ + Zk Zl rkll ∑ k ∑
  • 4.
    TheVariational Principle We canjudge the quality of the wave functions by the energy – the lower the energy, the better. We may also use any arbitrary basis set to expand the guess wave function. How do we actually use this? NANO266 4 φHφ dr∫ φ2 dr∫ ≥ E0
  • 5.
    Linear combination ofatomic orbitals (LCAO) NANO266 5 http://www.orbitals.com/
  • 6.
    Solving the one-electronmolecular system with the LCAO basis set approach In general, we may express our trial wave functions as a series of mathematical functions, known as a basis set. For a single nucleus, the eigenfunctions are effectively the hydrogenic atomic orbitals. We may use these atomic orbitals as a basis set for our molecular orbitals. This is known as the linear combination of atomic orbitals (LCAO) approach. NANO266 6 φ = aiϕi i=1 N ∑
  • 7.
    The Secular Equation NANO266 7 E= aiϕi i=1 N ∑ " # $ % & 'H aiϕi i=1 N ∑ " # $ % & 'dr∫ aiϕi i=1 N ∑ " # $ % & ' 2 dr∫ = aiaj ϕiHϕj dr∫ ij ∑ aiaj ϕiϕj dr∫ ij ∑ = aiaj Hij ij ∑ aiajSij ij ∑ Resonance integral Overlap integral
  • 8.
    The Secular Equation,contd Tominimize the energy, Which gives Or in matrix form NANO266 8 ∂E ∂ak = 0, ∀k ai (Hki − ESki ) i=1 N ∑ = 0, ∀k H11 − ES11 H12 − ES12 ! H1N − ES1N H21 − ES21 H22 − ES22 ! H2N − ES2N " " # " HN1 − ESN1 HN 2 − ESN 2 ! HNN − ESNN " # $ $ $ $ $ % & ' ' ' ' ' a1 a2 " aN " # $ $ $ $ $ % & ' ' ' ' ' = 0
  • 9.
    The Secular Equation,contd Solutionsexist only if Procedure: i.  Select a set of N basis functions. ii.  Determine all N2 values of Hij and Sij. iii.  Form the secular determinant and determine the N roots Ej. iv.  For each Ej, solve for coefficients ai. NANO266 9 H11 − ES11 H12 − ES12 ! H1N − ES1N H21 − ES21 H22 − ES22 ! H2N − ES2N " " # " HN1 − ESN1 HN 2 − ESN 2 ! HNN − ESNN = 0
  • 10.
    HückelTheory Basis set formedfrom parallel C 2p orbitals Overlap matrix is given by Hii = Ionization potential of methyl radical Hij for nearest neighbors obtained from exp and 0 elsewhere NANO266 10 Sij =δij
  • 11.
    The Born-OppenheimerApproximation Heavier nucleimoves much more slowly than electrons => Electronic relaxation is “instantaneous” with respect to nuclear motion Electronic Schrödinger Equation NANO266 11 (Hel +VN )ψel (qi;qk ) = Eelψel (qi;qk ) Electronic energy Constant for a set of nuclear coordinates
  • 12.
    Stationary Electronic SchrödingerEquation where NANO266 12 Eelψel = Helψel Hel = − 1 2 ∇i 2 i ∑ − Zk rikk ∑ i ∑ + 1 rijj ∑ i ∑ KE and nuclear attraction terms are separable H = hi i ∑ where hi = − 1 2 ∇i − Zk rikk ∑
  • 13.
    Hartree-ProductWave Functions Eigen functionsof the one-electron Hamiltonian is given by Because the Hamiltonian is separable, NANO266 13 hiψi =εiψi ψHP = ψi i ∏ HψHP = hi i ∑ ψk k ∏ = εi i ∑ # $ % & ' (ψHP
  • 14.
    The effective potentialapproach To include electron-electron repulsion, we use a mean field approach, i.e., each electron sees an “effective” potential from the other electrons NANO266 14 hi = − 1 2 ∇i − Zk rikk ∑ +Vi, j where Vi, j = ρj rij ∫ j≠i ∑ dr
  • 15.
    Hartree’s Self-Consistent Field(SCF)Approach NANO266 15 Guess MOs Construct one- electron operations hi Solve for new ψ hiψi =εiψi Iterate until energy eigenvalues converge to a desired level of accuracy E = εi i ∑ − 1 2 ψi 2 ψj 2 rij dri drj∫∫ What’s the purpose of this term?
  • 16.
    What about thePauli Exclusion Principle? Two identical fermions (spin ½ particles) cannot occupy the same quantum state simultaneously è Wave function has to be anti-symmetric For two electron system, we have NANO266 16 ψSD = 1 2 ψa (1)α(1)ψb (2)α(2)−ψa (2)α(2)ψb (1)α(1)[ ] = 1 2 ψa (1)α(1) ψb (1)α(1) ψa (2)α(2) ψb (2)α(2) where α is the electron spin eigenfunction Slater determinant
  • 17.
    For many electrons… NANO266 17 ψSD= 1 N! χ1(1) χ2 (1) ! χN (1) χ1(2) χ2 (2) ! χN (2) ! ! " ! χ1(N) χ2 (N) ! χN (N) where χk are the spin orbitals
  • 18.
    The Hartree-Fock (HF)Self-Consistent Field (SCF) Method NANO266 18 fi = − 1 2 ∇i 2 − Zk rik +Vi HF {j} k nuclei ∑ F11 − ES11 F12 − ES12 ! F1N − ES1N F21 − ES21 F22 − ES22 ! F21 − ES2N " " # " FN1 − ESN1 FN 2 − ESN 2 ! FNN − ESNN = 0 HF Secular Equation Fµυ = µ |− 1 2 ∇i 2 |υ − Zk µ | 1 rk |υ + Pλσ λσ ∑ (µυ | λσ )− 1 2 (µλ |υσ ) $ %& ' () k nuclei ∑ Weighting of four-index integrals by density matrix, P
  • 19.
    Flowchart of HFSCF Procedure NANO266 19
  • 20.
    Limitations of HF Fockoperators are one-electron => All electron correlation, other than exchange, is ignored Four-index integrals leads to N4 scaling with respect to basis set size NANO266 20 Ecorr = Eexact − EHF
  • 21.
    PracticalAspects of HFCalculations Basis Sets Effective Core Potentials Open-shell vs Closed- shell Accuracy Performance NANO266 21
  • 22.
    Basis Set Set ofmathematical functions used to construct the wave function. In theory, HF limit is achieved by an infinite basis set. In practice, use finite basis sets that can approach HF limit as efficiently as possible NANO266 22
  • 23.
    Contracted Gaussian Functions Slater-typeorbitals (STO) with radial decay cannot be analytically integrated -> Use linear combination of Gaussian-type orbitals (GTOs) with radial decay to approximate STOs STO-3G •  STO approximated by 3 GTOs •  Known as single-ζ or minimal basis set. NANO266 23 e−r2 e−r
  • 24.
    Multiple-ζ and Split-Valence Multiple-ζ • Adding more basis functions per atomic orbital •  Examples: cc-pCVDZ, cc-pCVTZ (correlation-consistent polarized Core and Valence (Double/Triple/etc.) Zeta) Split-valence or Valence-Multiple-ζ •  Still represent core orbitals with single, contracted basis functions •  Valence orbitals are split into many functions (Why?) •  Examples: 3-21G, 6-31G, 6-311G NANO266 24 # of primitives in core # of primitives in valence
  • 25.
    Polarization and DiffuseFunctions Polarization functions •  Description of MOs require more flexibility than provided by AOs, e.g., NH3 is predicted to be planar if using just s and p functions •  Additional basis functions of one quantum number of higher angular momentum than valence, e.g., first row -> d orbitals •  Notation: 6-31G* [old] or 6-31G(d) [new], 6-31(2d,p) [2d functions for heavy atoms, additional p for H] Diffuse functions •  Highest energy MOs of anions, highly excited states tend to be more diffuse •  Augment standard basis sets with diffuse functions •  Notation: 6-31+G, 6-311++G(3df, 2pd), aug-cc-pCVDZ NANO266 25
  • 26.
    Effective Core Potentials Heavyatoms have many electrons •  Intractable to model all of them, even with a minimal basis set •  However, most of the electrons are in the core Solution: Replace core electrons with analytical functions (effective core potentials or ECPs) that represent combined nuclear-electronic core to the remaining electrons Key selection decision: How many electrons to include in the core? NANO266 26
  • 27.
    Open-shell vs closed-shell RestrictedHF (RHF) •  Closed-shell systems, i.e., no unpaired electrons Restricted open-shell HF (ROHF) •  Use RHF formalism, but with density matrix for singly occupied orbitals not multiplied by a factor of 2. •  Wave functions are eigenfunctions of S2 •  But fails to account for spin polarization in doubly occupied orbitals Unrestricted HF (UHF) •  Includes spin polarization •  Wave functions are not eigenfunctions of S2, i.e., spin contamination NANO266 27
  • 28.
    Accuracy Energetics •  In general,extremely poor; correlation is extremely important in chemical bonding! •  Protonation energies are typically ok (no electrons in H+) •  Koopman’s Theorem: First IE is equal to the negative of the orbital energy of the HOMO Geometry •  Typically relatively good ground state structures with basis sets of modest size •  But transition states (with partial bonding) can be problematic, as well as some pathological systems NANO266 28
  • 29.
    Performance Formal N4 scaling Butin reality, speedups can be achieved through: •  Symmetry •  Estimating upper bounds to four-index integrals •  Fast multipole and linear exchange integral computations For practical geometry optimizations, frequently helps to first compute geometry with a smaller basis set to provide a better initial geometry and a guess for the Hessian matrix. NANO266 29