Introduction to Hartree-Fock

Molecular Orbital Theory

C. David Sherrill
School of Chemistry and Biochemistry
Georgia Institute of Technology

Preceding Material
These notes pick up from General
Introduction to Electronic Structure
Theory by the author.

What Hartree-Fock is for

A way to approximately solve the Electronic
Schrdinger equation

This gives us the electronic wavefunction,

from which we can extract dipole moment,
polarizability, etc!
The electronic energy Eel(R) is the potential
energy surface: gives equilibrium geometries,
reaction paths, etc.

The Slater Determinant

John Slater

Shorthand:

or

We will assume the electronic wavefunction can be written

as a single Slater Detrminant (this is an approximation). This
enforces an antisymmetric wavefunction. The Hartree-Fock
procedure will give us the orbitals.
Images from Wikipedia

Hartree-Fock Molecular Orbital Theory

1. Invoke the BornOppenheimer approximation
2. Express the electronic
wavefunction as a single Slater
Determinant
3. Solve for those orbitals
which minimize the electronic
energy (variational method)
This winds up being
mathematically equivalent to
assuming each electron

interacts only with the average

charge cloud of the other
electrons
Images from Wikipedia

Douglas Hartree

V. A. Fock

The Operators
One-electron operator: for electron i,
its KE and its attraction to all nuclei
Two-electron operator: for electrons
i and j, their Coulomb repulsion
Electronic Hamiltonian in terms of these operators:

The Hartree-Fock Energy

If the variational theorem says to minimize
the energy, what is the energy of a Slater
determinant?

Slaters Rules tell us how to get the matrix

elements of the electronic Hamiltonian using
Slater Determinants

The Hartree-Fock Energy

One-electron integral (4-dimensional):

Two-electron integral (8-dimensional):

Physical Meaning of the Terms

Each electron contributes a one-electron
integral

This looks like the expectation (average) value

of the operator for an electron in orbital , so
long as the orbital is normalized
Recall contains electron KE and potential of
attraction to all the nuclei
Sum over all orbitals i to get total electron KE
and attraction to nuclei

Physical Meaning of the Terms

Each pair of electrons (in orbitals i and j) has
a Coulomb integral:

Probability electron 1 in
orbital i is located at x1
Integrate over all possible
locations for the electrons

Probability electron 2 in
orbital j is located at x2

Coulomb repulsion between electron at

x1 and electron at x2

Overall this integral represents the Coulomb

repulsion between electron 1 in orbital i and
electron 2 in orbital j

Physical Meaning of the Terms

Each pair of electrons (in orbitals i and j) has
also has an Exchange integral:

This is like the Coulomb integral

Except two of the orbital indices have been

exchanged!
No direct physical meaning .. consequence of
Slater Determinant

Hartree-Fock Energy Example

Simple example: He atom

2 electrons, 1sa, 1s
Number the spin orbitals 1=1sa, 2=1s

We (typically) need a computer to evaluate these

integrals
Is this as simple as we can get it?
No! The last integral is zero. Why?

Spin Factorization and Spatial Orbitals

Recall each spin orbital (x) is a function of 4
coordinates: (x,y,z,)
We normally write each spatial orbital as a product of a
spatial part (r) and a spin part, which we might call
(), i.e., (x) = (r) () [recall r = {x,y,z}]
The operators in Hartree-Fock theory, and 1/r12, do not
depend on the spin coordinate
That means an integral over x can be factored into a
simple integral over the spin coordinate (no operators)
times a more complicated integral (involving operators)
over the spatial coordinates r, e.g.,

Spin Factorization and Spatial Orbitals

Recall each spin orbital (x) is a function of 4
coordinates: (x,y,z,)
We normally write each spatial orbital as a product of a
spatial part (r) and a spin part, which we might call
(), i.e., (x) = (r) () [recall r = {x,y,z}]
The operators in Hartree-Fock theory, and 1/r12, do not
depend on the spin coordinate
That means an integral over x can be factored into a
simple integral over the spin coordinate (no operators)
times a more complicated integral (involving operators)
over the spatial coordinates r, e.g.,

Spin Factorization and Spatial Orbitals

Recall each spin orbital (x) is a function of 4
coordinates: (x,y,z,)
We normally write each spatial orbital as a product of a
spatial part (r) and a spin part, which we might call
(), i.e., (x) = (r) () [recall r = {x,y,z}]
The operators in Hartree-Fock theory, and 1/r12, do not
depend on the spin coordinate
That means an integral over x can be factored into a
simple integral over the spin coordinate (no operators)
times a more complicated integral (involving operators)
over the spatial coordinates r, e.g.,

Spin Factorization and Spatial Orbitals

Recall each spin orbital (x) is a function of 4
coordinates: (x,y,z,)
We normally write each spatial orbital as a product of a
spatial part (r) and a spin part, which we might call
(), i.e., (x) = (r) () [recall r = {x,y,z}]
The operators in Hartree-Fock theory, and 1/r12, do not
depend on the spin coordinate
That means an integral over x can be factored into a
simple integral over the spin coordinate (no operators)
times a more complicated integral (involving operators)
over the spatial coordinates r, e.g.,

Spin Factorization of 2-elec Integrals

We can also factorize out the spin functions
in two-electron integrals

Spin Factorization of 2-elec Integrals

We can also factorize out the spin functions
in two-electron integrals

Spin Integration
Integrals over spin coordinates are usually
easy to do because the spin function () is
usually just a() or ()
The spin integration rules for a() and ()
are very easy and result in 1 or 0

Spin Integration General Results

One-electron integrals survive if both spin-orbitals

have the same spin

Two-electron integrals survive if spins i,j match on

the left and spins k,l match on the right

Spin Integration in Hartree-Fock

We just did the generic one- and twoelectron integrals; the ones in Hartree-Fock
are specific types

same index both sides

each index appears twice

(although in different places)

How do these repeats affect spin

integration?

Spin Integration in Hartree-Fock

One-electron Integrals

Spin Integration in Hartree-Fock

One-electron Integrals

Factor in brackets
is always = 1
(same spin function)

Spin Integration in Hartree-Fock

Coulomb Integrals

Spin Integration in Hartree-Fock

Coulomb Integrals
1

Spin Integration in Hartree-Fock

Coulomb Integrals
1

Coulomb integrals always survive spin integration!

Spin Integration in Hartree-Fock

Exchange Integrals

Spin Integration in Hartree-Fock

Exchange Integrals

Spin Integration in Hartree-Fock

Exchange Integrals

Exchange integrals sometimes survive spin integration!

Need spin orbitals i and j to have same spin

Back to our Hartree-Fock Energy

Example

Simple example: He atom

2 electrons, 1sa, 1s
Renumber the spin orbitals: 1=1sa, 1=1s

Exchange integrals
must have same spin
on i and j
Can we simplify this result? Yes! (normally)

Simplifying Spatial Integrals

Helium atom example
1

Spatial part is the same. Therefore after spin integration, spatial integrals in
terms of these two orbitals must be the same. Lets check.

Simplifying Spatial Integrals

Simplifying Spatial Integrals

Conclusion: if spin orbitals come in a, pairs with the same spatial part , then
after spin integration we can remove the overbar labels on the spatial orbital
integrals

Simplifying Spatial Integrals

Helium atom example
1

Restricted Orbitals
(Restricted Hartree-Fock, RHF)
Spin orbitals always come in (a,) pairs that
share the same spatial orbital

This is how we normally think about orbitals in

chemistry
Theres no reason not to use such restricted
orbitals in normal molecules in which all
electrons are paired (closed shell molecules)

Unrestricted Orbitals
(Unrestricted Hartree-Fock, UHF)
When not all electrons are paired (open-shell
molecules), we can sometimes get a lower
energy solution if we unrestrict the orbitals:
allow the spatial part to be different for the a
spin than for the spin:

Introduces spin contamination (e.g., mix

singlet and triplet); can sometimes cause
severe errors in properties
Can be easier to converge

Pseudo-Classical Interpretation of
Hartree-Fock Energies

Always survive spin integration

Nonzero only if i and j have same spin

Each electron contributes a term (i||j) = hij

Each unique pair of electrons contributes a
Coulomb repulsion (ii|jj) = Jij
Each unique pair of same spin electrons
contributes an exchange term (ij|ji) = -Kij

Hartree-Fock Energy Example:

Li atom
2

3 electrons: 1=1sa, 1=1s, 2=2sa

Connection to Hunds Rules

Why do Hunds rules say a high-spin state is
more stable than a low-spin state for a given
electron configuration?
We can use Hartree-Fock theory to understand
this
As an example, consider the p2 electron
configuration
1

EHF = h11 + h22 + J12 K12

Energy lower because

exchange integral is
subtracted!

vs

EHF = 2h11 + J11

Hartree-Fock Equations
Minimizing the Hartree-Fock energy with respect
to the orbitals leads to the Hartree-Fock
equations for the orbitals:

Problem: This is a very complicated integrodifferential equation!

Roothan to the Rescue!

If we introduce a basis set, we
convert the equation into a
much simpler linear algebra
problem
Clemens C. J. Roothan

Images from Wikipedia

Summary of Hartree-Fock-Roothan
Equations

Pseudo-eigenvalue equation
C collects the expansion coefficients for each
orbital expressed as a linear combination of
the basis functions (each column of C is a
molecular orbital)
Note: C depends on F, which depends on C!

Self-consistent-field procedure
1. Specify molecule, basis functions, and electronic
state of interest (i.e., singlet, triplet, etc)
2. Form overlap matrix S
3. Guess initial MO coefficients C
4. Form Fock matrix F
5. Solve FC=SC
6. Use new MO coefficients C to build new Fock
matrix F
7. Go to step 5; repeat until C no longer changes
from one iteration to the next

Forming the Fock Matrix

For closed-shell RHF

Computational Cost
Rate determining step normally computation of
O(n4) integrals
Integrals very small if basis functions centered
on atoms far from each other; can use
screening techniques to reduce to O(n2)
significant integrals
Fast multipole methods, etc., can reduce cost
further to linear scaling
Alternatively, can replace 4-index integrals by
3-index integrals using density fitting

So What Did We Get?

The electronic energy (one point on the
potential energy surface; PES can give
equilibrium geometry, reaction paths,
etc.)
The electronic wave function (can get
dipole moment, polarizability,
electrostatic potential, other properties)
Orbitals (can give insight into bonding)
Orbital energies

Orbital Energies
Occupied orbital energy (usually
negative) approximately gives negative
of energy required to remove an
electron from that orbital
Unoccupied orbital energy (usually
positive) very approximately gives
energy required to put an electron in
that orbital
Orbital energies do not sum to the
Hartree-Fock energy

Energy Units
Atomic unit of energy is the Hartree
(sometimes abbreviated au or Eh)
H atom energy in the Born-Oppenheimer
approximation (use electron mass not
reduced mass) defined as -1/2 Hartree
Its a big unit! 1 Hartree = 627.509 kcal/mol

Example of Orbital Energies

Comparison of Hartree-Fock to
Experiment for Cu+ (atomic units =
Hartree)

Data from McQuarrie, Quantum Chemistry

Practical considerations
Hartree-Fock self-consistent-field (HF SCF) usually converges
fairly well with a good initial guess
Stretched bonds, diradicals, transition metals, high-spin states,
etc., can cause problems for convergence
In high-symmetry cases, the program can guess the wrong
orbital occupations, and then have trouble recovering from this
to get the desired solution
Not guaranteed to land on a local minimum in C space; can
check by running a Hartree-Fock stability analysis (useful but
not commonly done). However, even this doesnt guarantee
youre not in some other local minimum (esp. for high-symmetry
cases)
User is responsible for making sure the orbital occupations are
reasonable and the spin state is correct. Many students dont
know that the ground state of O2 is a triplet, not a singlet. The
programs dont know about this!

Improving Convergence
Most codes use direct inversion of the
iterative subspace (DIIS) to improve
convergence (improves guess for the next
step)
The quality of the guess density makes a
big difference. Core Hamiltonian (no initial
density) is quite poor. Hckel and GWH
ok; superposition of atomic densities
(SAD) seems best when available
Using MO's from one geometry as guesses
for a nearby geometry (or neutral orbitals
as a guess for a cation or anion, or singlet
orbitals as a guess for a triplet) works well