Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF

Molecular Electromagnetism
A Computational Chemistry Approach

Chapter 1
Stephan P. A. Sauer
sauer@kiku.dk
Department of Chemistry
University of Copenhagen

c Stephan P. A. Sauer 2011
Molecular Electromagnetism A Computational Chemistry Approach – p.1/5

1. Introduction
Experiments carried out on molecules can be understood as interaction of the

molecules with electromagnetic fields
Fields can be treated as perturbations apart from the case of
Chemical reactions, where molecule changes identity
Intense laser fields, where field and molecule have to be treated on equal
footing
Possible electromagnetic fields
Static external electric field in a capacitor
Static external electric field due to another molecule nearby
Static external magnetic field in an NMR or ESR spectrometer
Internal magnetic fields due to the magnetic moments of nuclei with spin
Internal electric field due to the electric quadrupole moment of a nucleus
Oscillating electric and magnetic fields of electromagnetic radiation

1. Introduction - 2
Interactions are often described in terms of molecular properties

Electric dipole moment
Frequency dependent polarizability
Chemical shift and indirect nuclear spin-spin coupling constant of NMR
hyperfine coupling constant of ESR
Molecular properties
are intrinsic properties of a particular state of a molecule
are independent of the strength of the fields.
can be used to describe the response of a molecule to an arbitrary field.

1. Introduction - 3
Molecular properties play thus an important role in the interpretation of

numerous experimental phenomena
refractive index
Stark and Zeeman effects
Kerr effect
nuclear and electric magnetic resonance spectra
and many more
Even long-range interactions between molecules can be understood in terms
of molecular electric moments

1. Introduction - 4
Molecular properties can be determined to high accuracy in experiments

Calculations of molecular properties are also important
Unknown compounds or molecular configurations can be identified by their
calculated properties.
Candidates for new materials with desired properties can fast and
inexpensively be screened by calculating the respective properties.
The value of an electromagnetic property of a particular molecule or the
changes of a property in a series of molecules can often be explained by
analyzing all the terms which contribute to it.
Theoretical calculations of molecular electromagnetic properties supplement
experiments in addition to calculations of the energetics and structure of
molecules.

Chapter 2
Stephan P. A. Sauer
sauer@kiku.dk


2. The Schrödinger Equation in the Presence of Fields
A complete quantum mechanical treatment of the interaction of molecules and

fields would required quantum electrodynamics (QED).
QED is probably the most successful theory, which was ever derived.
But it is not yet regularly employed in the calculation of electromagnetic
properties of molecules and the effects are expected to be very small.
Therefore we will make a series of approximations to this approach.
We will only use quantum mechanics for the description of the molecule and
use classical electrodynamics for the electromagnetic fields.
In this semi-classical approach the perturbing fields and nuclear moments
are considered to be unaffected by the molecular environment, the
so-called minimal coupling approximation.

2. The Schrödinger Equation in the Presence of Fields - 2
We use only non-relativistic quantum mechanics, i.e. the Schrödinger

equation. This is justified,
if we restrict ourselves to atoms of the first three rows of the periodic table,
for which relativistic effects are generally unimportant
if we are not interested in discussing properties, which include interactions
with the spin of the electrons such as NMR and ESR coupling constants.
The Schrödinger equation is in principle a spin-free theory contrary to the
Dirac equation.
However, the necessary operators for the interaction with the electron spin
can be derived from the Dirac equation and then added to the Schrödinger
Hamiltonian in an ad hoc fashion.
Finally, the Born-Oppenheimer approximation is applied in order to separate
the nuclear and electronic wavefunctions.

2.1 The Time-Dependent Schrödinger Equation - 1
The total non-relativistic Hamiltonian for a molecule with N electrons and M

nuclei is in the absence of fields
M ˆ2
(0) 1Xp ~K e2 X ZK ZL
Ĥnuc,e = +
2 K mK ~K − R
4πǫ0 K<L |R ~ L|
N NM
1 X ˆ2 e2 X ZK e2 X 1
+ ~i −
p +
2me i ~K|
4πǫ0 iK |~ri − R 4πǫ0 i<j |~ri − ~rj |
where p~ˆ K = −ı~∇ ~ˆ K and p ~ˆ i are the operators for the canonical
~ˆ i = −ı~∇
momentum of nucleus K with mass mK and atomic number ZK at R ~ K and
electron i at ~ri , respectively.
The state of a molecule is completely described by the time-dependent
~ K }, {~ri }, t). The wavefunction is the solution to the
wavefunction Φ(0) ({R
time-dependent Schrödinger equation
(0)
Ĥnuc,e ~ K }, {~ri }, t)i = ı~ ∂ |Φ(0) ({R
|Φ(0) ({R ~ K }, {~ri }, t)i
∂t

When the Hamiltonian does not depend explicitly on time, we can separate the
~ K } and {~ri }.
time variable t from the spatial coordinates {R
We write the time-dependent wavefunction Φ(0) ({R ~ K }, {~ri }, t) as the product
~ K }, {~ri }) and a time-dependent
of a time-independent wavefunction Φ(0) ({R
phase factor ϑ(t).
Φ(0) ({R
~ K }, {~ri }, t) = Φ(0) ({R
~ K }, {~ri }) ϑ(t)
With this trial-solution, the time-dependent Schrödinger equation separates in

the time-independent Schrödinger equation
(0)
Ĥnuc,e |Φ(0) ({R
~ K }, {~ri })i = E (0) |Φ(0) ({R
~ K }, {~ri })i
and an equation for the time-dependent phase factor ϑ(t).

For the time-dependent phase factor one obtains
ı E (0) t
−~
ϑ(t) = e

The time-independent Schrödinger equation is a second-order partial

differential equation.
(0)
Ĥnuc,e |Φ(0) ({R
~ K }, {~ri })i = E (0) |Φ(0) ({R
~ K }, {~ri })i
It can be interpreted as an eigenvalue equation.

The time-dependence of the wavefunction corresponds to a rotation in the
complex plane
ıE (0)
(0) ~ (0) ~ −~ t
Φ ({RK }, {~ri }, t) = Φ ({RK }, {~ri }) e
The probability density of the time-dependent wavefunction is constant in time.

ı E (0) t ∗ ı E (0) t

(0) 2 (0)
~ K }, {~ri }, t)| = Φ ({R ~ K }, {~ri })e ~
− (0) ~ K }, {~ri })e ~
−
|Φ ({R Φ ({R
= |Φ(0) ({R
~ K }, {~ri })|2
The eigenfunctions of the time-independent Schrödinger equation are

therefore called stationary states.

The average measured value < P >t of a physical observable P obtained in a

series of measurements on a large ensemble of molecules, which are all in the
~ K }, {~ri }, t)i, can be calculated as expectation value of
same state |Φ(0) ({R
the corresponding quantum mechanical operator P̂
~ K }, {~ri }, t) | P̂ | Φ(0) ({R

hΦ(0) ({R ~ K }, {~ri }, t)i
< P >t ≡
~ K }, {~ri }, t) | Φ(0) ({R
hΦ(0) ({R ~ K }, {~ri }, t)i
For the stationary states the expectation value of a time-independent

quantum mechanical operator P̂ becomes also time-independent
~ K }, {~ri }) | P̂ | Φ(0) ({R

hΦ(0) ({R ~ K }, {~ri })i
< P >≡
~ K }, {~ri }) | Φ(0) ({R
hΦ(0) ({R ~ K }, {~ri })i

2.2 The Born-Oppenheimer Approximation - 1
The masses of the nuclei, mK , are at least three orders of magnitude larger
than the mass of an electron.
We assume that the electrons will instantaneously adjust to a change in the
positions of the nuclei and that we can find a wavefunction for the electrons for
each arrangement of nuclei.
(0)
The total molecular Hamilton operator Ĥnuc,e is partitioned in the kinetic
P p~ˆ 2K
energy operator of the nuclei, 12 M K mK , and a molecular field free
electronic Hamiltonian Ĥ (0) for fixed nuclear coordinates {R ~ K }, defined as
N NM
(0) 1 X ˆ2 e2 X ZK e2 X 1
Ĥ = ~i −
p +
2me i ~K|
e2 X ZK ZL
+
~K − R
4πǫ0 K<L |R ~ L|

With this operator we obtain the time-independent field-free electronic

Schrödinger equation for a given set of nuclear coordinates {R~K}
(0) ~ K })i = E (0) ({R

~ K }) |Ψ(0) ({~ri }; {R
Ĥ (0) |Ψk ({~ri }; {R k k
~ K })i
(0)
~ K }), is a function of the nuclear
The total electronic energy, Ek ({R
coordinates {R~ K }.
(0) ~ K }) is a function of the electronic
The electronic wavefunction Ψk ({~ri }; {R
coordinates {~ri } and depends parametrically on the nuclear coordinates {R ~K}
~ K } the electronic wavefunction shows a
For each set of nuclear coordinates {R
different functional dependence on the electronic coordinates {~ri }.
In the Born-Oppenheimer approximation the total time-independent
wavefunction is approximated by the product of an electronic wavefunction and
a nuclear wavefunction
(0) ~ K }, {~ri }) = Ψ(0) ({~ri }; {R
~ K }) Θ(0) ({R
~ K })
Φk,v,J ({R k v,J
~K}
which depends only on the nuclear coordinates {R
When we insert this approximate trial wavefunction in the time-independent

Schrödinger equation, neglect two small terms and make use of the
time-independent field free electronic Schrödinger equation, we obtain the
nuclear Schrödinger equation
" M #
1Xp ~ˆ K
2
(0) ~ K })i = E (0) |Θ(0) ({R
~ K }) |Θ(0) ({R ~ K })i
+ Ek ({R v,J k,v,J v,J
2 K mK
(0) ~ K }), fulfills the role of the potential energy

The total electronic energy, Ek ({R
for the motion of the nuclei and is therefore often called a potential energy
surface.
Neither the nuclear Schrödinger equation nor the parametrical dependence of
the electronic wavefunction on the nuclear coordinates will be considered until
much later.
We will therefore also omit the dependence on the nuclear coordinates,
({R~ K }), in the notation for the electronic energies and wavefunctions until
later.
2.3 Electron Charge and Current Density - 1
Max Born’s interpretation of the wavefunction of an electron :
|ψ(~r)|2 = ψ ∗ (~r) ψ(~r)
is the probability density of finding the electron at the point ~r.

Generalized to the N -electron case
2
(0)
Ψk (~r1 , ~r2 , · · · , ~rN ) d~r1 d~r2 · · · d~rN

(0)∗ (0)
= Ψk (~r1 , ~r2 , · · · , ~rN )Ψk (~r1 , ~r2 , · · · , ~rN ) d~r1 d~r2 · · · d~rN
is the probability of finding electron 1 in the volume element d~r1 at ~r1 and
electron 2 in the volume element d~r2 at ~r2 and so forth
Probability just for finding electron 1 in the volume element d~r1 at point ~r1 is
obtained by integration over the coordinates of the other electrons
Z Z 2
(0)
P1 (~r1 ) d~r1 = ··· Ψk (~r1 , ~r2 , · · · , ~rN ) d~r2 · · · d~rN d~r1

~
r2 ~
rN

Electrons are indistinguishable and the probability of finding electron 2 in a

volume element at the same point is identical.
The probability of finding an electron in the volume element d~r at a point ~r
is therefore N times the probability of finding electron 1 at this point and is thus
given as
Z Z 2
(0)
P (~r) d~r = N ··· Ψk (~r, ~r2 , · · · , ~rN ) d~r2 · · · d~rN d~r

~
r2 ~
rN
where P (~r) is the electron density.

Using the properties of the Dirac δ function, δ(~r0 ), in three dimensions
Z
δ(~r − ~r0 )f (~r) d~r = f (~r0 )
~
r
we can extend the integration to include all electrons and write

Z Z 2
(0)
P (~r) = N ··· δ(~r1 − ~r) Ψk (~r1 , · · · , ~rN ) d~r1 · · · d~rN

~
r1 ~
rN

Reversing the indistinguishability of the electrons argument we can write the

electron density as expectation value
N Z Z 2
X (0)
P (~r) = ··· δ(~ri − ~r) Ψk (~r1 , · · · , ~rN ) d~r1 · · · d~rN

i ~
r1 ~
rN
(0) (0)
= hΨk ({~ri }) | D̂(~r) | Ψk ({~ri })i
PN
of the density operator D̂(~r) defined as D̂(~r) = i δ(~ri − ~r)
Reduced one electron density matrix
Z Z
(0) (0)∗
P (~r, ~r ′ ) = N · · · Ψk (~r, ~r2 , · · · , ~rN ) Ψk (~r ′ , ~r2 , · · · , ~rN ) d~r2 · · · d~rN
a generalization of the definition of the electron density P (~r)

Charge density ρel (~r) of an N-electron system
(0) (0)
ρel (~r) = −e hΨk ({~ri }) | D̂(~r) | Ψk ({~ri })i

Using the continuity equation of classical electromagnetism
∂ρ(~r, t) ~ · ~j(~r)
= −∇
∂t
we can derive an expression for the current density of the electrons, ~j el (~r)
Time derivative of the time-dependent charge density ρel (~r, t) of the electrons
el Z Z
∂ρ (~r1 , t)
2
∂

(0)
= −eN ··· Ψk (~r1 , ~r2 , · · · , ~rN , t) d~r2 · · · d~rN

∂t ∂t ~
r2 ~
rN
The time derivative of the wavefunction is given by the time-dependent

Schrödinger equation
el Z Z ( N
∂ρ (~r1 , t) −ıeN
~ˆ 2i Ψk ({~ri }, t)
(0) (0)∗
X
= ··· Ψk ({~ri }, t) p
∂t 2me ~ ~r2 ~
rN i=1
N
)
~ˆ i Ψ ({~ri }, t) d~r2 · · · d~rN
(0)∗
X 2 (0)
−Ψ ({~ri }, t)
k p k
i=1

Using that
(0) ˆ 2 Ψ(0)∗ ({~ri }, t)
Ψk ({~ri }, t)∇ k
ˆ ˆ ˆ ˆ
h i h ih i
(0)
~ · Ψ ({~ri }, t)∇Ψ ~ (0)∗ (0)
~ · Ψ ({~ri }, t) ∇ ~ ·Ψ (0)∗
=∇ k k ({~ri }, t) − ∇ k k ({~ri }, t)
and that the terms in the summation for i = 2 · · · N vanish we arrive at

∂ρel (~r1 , t)
Z Z h
ı~eN ~ˆ (0) ~ˆ 1 Ψ(0)∗ ({~ri }, t)
= ∇1 · ··· Ψk ({~ri }, t)∇ k
∂t 2me ~r2 ~
rN
ˆ
i
(0)∗ (0)
~ 1 Ψ ({~ri }, t) d~r2 · · · d~rN
−Ψk ({~ri }, t)∇ k
Changing the variable ~r1 to ~r and introducing the Dirac δ function gives
N
∂ρel (~r, t) ~ˆ · e
~î | Ψk ({~ri }, t)i
(0) (0)
X
= ∇ hΨk ({~ri }, t) | δ(~ri − ~r)p
∂t 2me i
N
!
e
~î | Ψk ({~ri }, t)i∗
(0) (0)
X
+ hΨk ({~ri }, t) | δ(~ri − ~r)p
2me i

Comparison with the continuity equation leads to the current density

N
e
~î | Ψk ({~ri }, t)i
(0) (0)
X
~j el (~r) = hΨk ({~ri }, t) | δ(~ri − ~r)p
2me i
N
e
~î | Ψk ({~ri }, t)i∗
(0) (0)
X
+ hΨk ({~ri }, t) | δ(~ri − ~r)p
2me i
For a stationary state this reduces to

N
e
~î | Ψk ({~ri })i
(0) (0)
X
~j el (~r) = hΨk ({~ri }) | δ(~ri − ~r)p
2me i
N
!
~î | Ψk ({~ri })i∗
(0) (0)
X
+hΨk ({~ri }) | δ(~ri − ~r)p
i

2.4 The Force due to Electromagnetic Fields - 1
~ of an electric E(~
The force F ~ r, t) and magnetic field B(~~ r , t) on a particle with
charge q, mass m and velocity ~v = d~ r
dt
is called the Lorentz force
h i
F~ (~r, t) = q E(~
~ r, t) + ~v × B(~
~ r, t)
~ r, t)
The scalar potential, φ(~r, t), and vector potential, A(~
~ r, t)
∂ A(~
~ r , t)
E(~ = ~
−∇φ(~r, t) −
∂t
~ r , t)
B(~ = ~ × A(~
∇ ~ r, t)
are often more convenient to work with.

Electric and magnetic phenomena are coupled in the case of a
~ B (~r, t).
time-dependent vector potential due to the time-derivative of A
Using the potentials the Lorentz force becomes
( )
~ r, t)
∂ A(~ h i
~ ~
F = q −∇φ(~r, t) − + ~v × ∇~ × A(~
~ r, t)
∂t
2.4 The Force due to Electromagnetic Fields - 2
One possible set of expressions for the potentials of a static and

homogeneous electric field E~ and magnetic induction B ~ is
φ E (~r) = −~r · E~
~ B (~r) 1 ~
A = B × ~r
2
However, the potentials are not uniquely. Given an arbitrary scalar function
χ(~r, t), called gauge function, the following transformations of the potentials
~ r, t) → A
A(~ ~ ′ (~r, t) = ~ r , t) + ∇χ(~
A(~ ~ r , t)
∂χ(~r, t)
φ(~r, t) → φ ′ (~r, t) = φ(~r, t) −
∂t
~ r, t) and B(~
leave the fields, E(~ ~ r, t), unchanged.
These transformations are called gauge transformations

2.5 Minimal Coupling - Non Relativistically - 1
Semi-classical approach for the interaction between electromagnetic fields or

nuclear electromagnetic moments and molecules
The electrons are treated by quantum mechanics.
The fields or nuclear moments
are treated classically.
are not part of the system, which is treated quantum mechanically.
are merely considered to be perturbations which do not respond to the
presence of the molecule.
enter the molecular Hamiltonian in terms of external potentials similar to
the coloumb potential due to the charges of the nuclei.
Here only derivation of the Hamiltonian operator for the motion of a single
electron in the presence of external fields.
Generalization to the many-electron case is simple.

Standard approach for construction of the Schrödinger equation

classical Hamiltonian H for the system ( e.g. a field-free particle):
a function of the position coordinates ~r and the canonical (generalized or
conjugate) momenta p ~ of the particles.
~2
p
H=
2m
Substitutions rules
p
~ → p ~ˆ
~ˆ = −ı~∇
∂
H → ı~
∂t
Operators act on the wavefunction of the given system:
the time-dependent Schrödinger equation
∂ ~ˆ 2
p
ı~ |ψ(~r, t)i = |ψ(~r, t)i
∂t 2m

We have to derive the classical Hamiltonian function for an electron in the

presence of electromagnetic fields
This is normally done from the classical Lagrangian L
L(~r, ~v , t) = T (~r, ~v ) − U (~r, ~v , t)
a function of generalized position coordinates ~r and their time derivatives, i.e.

the generalized velocities ~v .
T (~r, ~v ) is the kinetic energy and U (~r, ~v , t) is a generalized potential.
U (~r, ~v , t) is chosen in such a way that the Euler-Lagrange equations

d ∂L ∂L
− =0
dt ∂vα ∂rα
reproduce Newton’s second law
~ = m d~v
F
dt

In our case Newton’s second law involves the Lorentz force

( )
~ r, t)
∂ A(~ h i d~v
~
−e −∇φ(~r, t) − + ~v × ∇~ × A(~
~ r , t) = me
∂t dt
Guess for the generalized potential U (~r, ~v , t)
~ r, t)
U (~r, ~v , t) = −e φ(~r, t) + e ~v · A(~
generalized because it depends not only on time and on the position of the
electron but also on the velocity of the electron.
Lagrangian
2
me~v ~ r, t)
L(~r, ~v , t) = T (~r, ~v ) − U (~r, ~v , t) = + e φ(~r, t) − e ~v · A(~
2
inserted into the Euler-Lagrange equations gives Newton’s second law for an
electron in external fields

Classical Hamiltonian function,
H(~r, p
~, t) = p
~ · ~v − L(~r, ~v , t)
a function of time, of the generalized position coordinates ~r and of their

conjugated generalized momenta p ~.
Classical Lagrangian and Hamiltonian are Legendre transforms. Velocities
and conjugated momenta exchange role as independent variables.
Legendre transformation between enthalpy H(S, P ) and Gibbs free energy
G(T, P ) = H(S, P ) − T S
Hamilton’s and Lagrange’s generalizations of classical mechanics are

essentially the same theory as Newton’s formulation. But
they are more elegant
they are often computationally easier to use.
they are form invariant under a change of coordinates. The Euler-Lagrange
equations have the same form for all types of coordinates.
Components of the generalized momentum vector p

~, i.e. the canonical
conjugate to the position vector ~r
∂L(~r, ~v , t)
pα =
∂vα
For the Lagrangian of on an electron in external fields this gives
~ r, t)
~ = me ~v − eA(~
p
The canonical momentum p ~ is in the presence of electromagnetic fields no

longer equal to the product of mass and velocity.
The product of mass and velocity the is called the kinematical or mechanical
momentum ~π
~π ≡ me ~v = p ~ r , t)
~ + e A(~
because the kinetic energy is defined as
1 1
T = me~v 2 = ~π 2
2 2me
Classical Hamiltonian
me ~v · ~v
H= − e φ(~r, t)
2
or in terms of the canonical momentum p~, and not the velocity ~v
1 h i2
~ r, t) − e φ(~r, t)
H= p~ + eA(~
2me
Application of substitution rules gives non-relativistic Hamiltonian operator Ĥ

for a single, spin-less particle
1 hˆ i2
ˆ
~ r , t) − e φ̂(~r, t)
Ĥ = ~ + eA(~
p
2me
~ ·A
or in the Coulomb gauge, ∇ ~=0
1 ˆ2 1 1 i2
ˆ ˆ
h
Ĥ = ~ +
p e A(~ ~ˆ +
~ r, t) · p 2
e A(~~ r, t) − e φ̂(~r, t)
2me me 2me

2.6 Minimal Coupling - Relativistically - 1
Electrons have spin

Schrödinger Hamiltonian operator is spin-free
Spin is a natural ingredient of the relativistic Dirac Hamiltonian operator
We need to derive the relativistic Dirac Hamiltonian operator for a single
particle in the presence of an electromagnetic field
The Lorentz force is unchanged in special relativity
Newton’s second law
~ = d (mr ~v )
F
dt
is changed due to the velocity dependence of the relativistic mass mr
m
mr = q
v2
~
1− c2
where m is the rest mass.

Lagrangian which will yield Newton’s second law

r
2 ~v 2 ~ r, t)
L(~r, ~v , t) = −me c 1 − 2 + e φ(~r, t) − e ~v · A(~
c
Components of the canonical momentum vector
∂L(~r, ~v , t) me vα
pα = = q − e Aα (~r, t)
∂vα 2
1 − ~vc2
Classical Hamiltonian from the Legendre transform of the Lagrangian
me c2
H= q − e φ(~r, t)
2
1 − ~vc2
in terms of the canonical momentum

r h i2
H= m2e c4 + c2 p ~ r, t)
~ + e A(~ − e φ(~r, t)

r h i2
The square root m2e c4 + c2 p ~ r, t) prevents transition to quantum
~ + e A(~
mechanics
Dirac proposed to write the kernel of the square root as perfect square
( )2
h i2 X
m2e c4 + c2 p ~ r, t) = β me c2 + c
~ + e A(~ αµ [pµ + e Aµ (~r, t)]
µ=x,y,z
where the ab priori unknown α′ s and β have to fulfill the conditions
α2µ = β 2 = 1 for µ = x, y, z
αµ αν + αν αµ = 0 for µ 6= ν
αµ β + βαµ = 0 for µ = x, y, z

Simplest solutions are set of 4 × 4 matrices

   
I2 02 02 σµ
β=   , αµ =  
02 −I 2 σµ 02
where I 2 is a 2 × 2 unit matrix and the σ µ are the Pauli spin matrices
     
0 1 0 −ı 1 0
σx =   , σy =   , σz =  
1 0 ı 0 0 −1
The classical Hamiltonian can therefore be written as

2
X
H = β me c + c αµ [pµ + e Aµ (~r, t)] − e φ(~r, t)I 4
µ=x,y,z

Finally substitution rules and application on the time-dependent wavefunction

of an electron, |ψ(~r, t)i, gives the time-dependent Dirac equation
( )
∂ X h i
ı~ |ψ(~r, t)i = c αµ p̂µ + e Âµ (~r, t) − e φ̂(~r, t)I 4 + β me c2 |ψ(~r, t)i
∂t µ=x,y,z
The wavefunction ψ consists of four components - a four-component spinor

 
  ψ1 
ψL ψ2 
ψ= = 
 
ψS ψ3 
 
ψ4
ψL and ψS are called the large and small component of the wavefunction
Dirac equation is a set of four coupled differential equations

In terms of the large and small components, the Dirac equation can be written
as two coupled two-component equations
" #
X
2
∂
c σ α p̂α + e Âα |ψS i + −e φ̂ + me c |ψL i = ı~ |ψL i
α=x,y,z
∂t
" #
X
2
∂
c σ α p̂α + e Âα |ψL i + −e φ̂ − me c |ψS i = ı~ |ψS i
α=x,y,z
∂t
We are not interested in a relativistic treatment of molecular properties here.

We only want to derive the operators for the interaction with the electronic spin.
We need to reduce the Dirac equation to a non-relativistic two-component
form: elimination of the small component

2.7 Elimination of the Small Component - 1
Time-dependent Dirac equation

( )
∂ X h i
ı~ |ψ(~r, t)i = c αµ p̂µ + e Âµ (~r, t) − e φ̂(~r, t)I 4 + β me c2 |ψ(~r, t)i
∂t µ=x,y,z
~ˆ r)
Let’s assume time-independent potentials φ̂E (~r) and A(~
and collect the time dependence of the wavefunction in a phase factor
|ψ(~r, t)i = |ψ̄(~r)i e−ıEt/~

∂
ψ(~r, t) is then an eigenfunction of ı~ ∂t with eigenvalue E
This gives the time-indepedent Dirac equation
( )
X h i
2
c σ α p̂α + e Âα (~r) |ψ̄S i = E + e φ̂(~r) − me c |ψ̄L i
α=x,y,z
( )
X h i
2
c σ α p̂α + e Âα (~r) |ψ̄L i = E + e φ̂(~r) + me c |ψ̄S i
α=x,y,z

Time-independent Dirac equation

( )
X h i
2
c σ α p̂α + e Âα (~r) |ψ̄S i = E + e φ̂(~r) − me c |ψ̄L i
α=x,y,z
( )
X h i
2
c σ α p̂α + e Âα (~r) |ψ̄L i = E + e φ̂(~r) + me c |ψ̄S i
α=x,y,z
|ψ̄S i can be expressed in terms of the large component |ψ̄L i (2nd equation)
( )
c X h i
|ψ̄S i = σ α p̂α + e Âα (~r) |ψ̄L i
E + e φ̂(~r) + me c 2
α=x,y,z
Inserted into the 1st equation gives a two component equation for |ψ̄L i
( ) ( )
2
X h i c X h i
σ α p̂α + e Âα (~r) σ α p̂α + e Âα (~r) |ψ̄L i
E + e φ̂(~r) + me c 2
α=x,y,z α=x,y,z

2
= E + e φ̂(~r) − me c |ψ̄L i

Defining the non-relativistic energy as E N R = E − me c2 , we can write
E + e φ̂(~r) + me c2 = 2 me c2 + E N R + e φ̂(~r)
Since 2 me c2 is of the order of MeV we can assume that

E N R + e φ̂(~r) << 2 me c2 and thus expand
 
c2 1  1
= 
E + e φ̂(~r) + me c2 2 me 1 + E N R +e φ̂(~
r )
2 me c 2
!
NR
1 E + e φ̂(~r)
= 1− + ...
2 me 2 me c2
Using only the first term, the equation for the large component reads
( )( )
1 X h i X h i
2 me α=x,y,z α=x,y,z
h i
NR
= E + e φ̂(~r) |ψ̄L i

Large component
( )( )
1 X h i X h i
2 me α=x,y,z α=x,y,z
h i
NR
= E + e φ̂(~r) |ψ̄L i
Can be simplified using a relation for the Pauli spin matrices and two general
~ˆ and D
vector operators C ~ˆ with components Ĉα and D̂α
! 
ˆ ˆ ˆ ˆ
X X X
σ α Ĉα  ~ ·D
σ β D̂β  = C ~ I2 + ı σα C~ ×D
~
α
α=x,y,z β=x,y,z α=x,y,z
~ˆ a general vector
and a relation which holds for the gradient operator ∇,
~ˆ and a scalar function ψ
operator C
ˆ ˆ ˆ

~ × Cψ
∇ ~ ~ × ∇ψ
= −C ~ + ∇~ ×C
~ ψ

The non-relativistic Schrödinger-Pauli equation becomes then

 
 1 h i2 e~
ˆ ˆ ˆ
h i 
ˆ
X
p ~
~ + e A(~
r) I 2 + ~ ~
σ α ∇ × A(~
r) − e φ̂(~
r)I 2 |ψ̄L i = E N R |ψ̄L i
 2 me 2 me α
α=x,y,z

Comparison with the Schrödinger Hamiltonian : additional Zeeman term
Zeeman e~ X h
ˆ ˆ
~ r)
~ × A(~
i
Ĥ = σα ∇
2 me α=x,y,z
α
The electron spin operator ~sˆ is related to the Pauli spin matrices σ
~
~
~sˆ = σ~
2
The electron spin Zeeman operator is therefore
2 e ˆ h ~ˆ ˆ
i g e e h
ˆ ˆ
i
Ĥ Zeeman
= ~ r ) or Ĥ
~s · ∇ × A(~ Zeeman
= ~sˆ · ∇~ × A(~
~ r)
2 me 2 me
where ge ≈ 2.002 is the QED electron g-factor.

2.8 The Molecular Electronic Hamiltonian - 1
Now we collect all terms and generalize for the many-electron case
In the minimal coupling approximation the vector potential enters the
mechanical momentum of electron i
π̂i = me~vî = p ~ˆ ri )
~î + eA(~
No contribution of the fixed nuclei in the Born-Oppenheimer approximation

Scalar potential times the charges of the particles
N
X M
X
−e φ̂(~ri ) + e ~K)
ZK φ̂(R
i K
Zeeman term of the Pauli Hamiltonian

N
Zeeman
X ge e ˆ ~ˆ ˆ
~ ri )

Ĥ = ~si · ∇ × A(~
i
2me

Molecular electronic Hamiltonian operator Ĥ in the presence of an

electromagnetic field
Ĥ = Ĥ (0) + Ĥ (1) + Ĥ (2)

XN X N
X N
X
= ĥ(0) (i) + (0)
ĝ(i, j) + Ĥnuc + ĥ(1) (i) + Ĥnuc
(1)
+ ĥ(2) (i)
i i<j i i
Ĥ (0) is the unperturbed Hamiltonian and contains one-electron ĥ(0) (i),

(0)
two-electron ĝ(i, j) and nuclear Ĥnuc contributions
M
1 ˆ2 e2 X ZK
ĥ(0) (i) = ~i −
p
2me ~K|
4πǫ0 K |~ri − R
e2 1
ĝ(i, j) =
4πǫ0 |~ri − ~rj |
(0) e2 X ZK ZL
Ĥnuc =
4πǫ0 ~K − R
|R ~ L|
K<L

(1)
Ĥ (1) includes all one-electron, ĥ(1) (i), and nuclear, Ĥnuc , operators, which are
linear in the perturbing field and thus first-order
(1) e ~ˆ ˆ ge e ˆ h ~ˆ ˆ
i
~ ri ) − e φ̂(~ri )
ĥ (i) = A(~ri ) · p
~i + ~si · ∇ × A(~
me 2me
XM
(1) ~K)
Ĥnuc = e ZK φ̂(R
K
~ ·A
where we have assumed the Coulomb gauge, i.e. ∇ ~ = 0, again.
Ĥ (2) contains the one-electron operators quadratic in the perturbations and is

thus second-order
(2) e2 ~ˆ 2
ĥ (i) = A (~ri )
2me

Separation of the perturbation Hamiltonian operators

(1) (2)
X F X FF
Ĥ + Ĥ = Ôα··· Fα··· + Fα··· Ôαβ··· Fβ···
α··· α,β,···
F FF
in interaction or perturbation operators Ôα··· and Ôαβ··· and a general field
with tensor components Fα···
~ with three components
The notation Fα··· we will cover both vector fields, F,
Fα as well as second rank tensor fields, F , with nine components Fαβ , etc.
F F
Ôα··· denotes cartesian components of a vector operator Ôα as well as
FF
components of a second rank tensor operator Ôαβ , etc.
FF FF FFF
Ôαβ··· stands for second Ôαβ , third Ôαβγ and etc. rank tensor operators
Perturbation operators as sum over all electrons
N
X N
X
F
Ôα··· = ôF
i,α···
FF
Ôαβ··· = ôFF
i,αβ···
i i
where ôF FF
i,α··· and ôi,αβ··· are perturbation operators acting on electron i alone.
2.9 Gauge Transformations - 1
~ ri , t) and φ(~ri , t) are not uniquely

The vector and scalar potentials, A(~
~ ri , t) and B(~
determined by their relations to the fields, E(~ ~ ri , t)
~ r, t)
∂ A(~
~ r , t)
E(~ = ~
−∇φ(~r, t) −
∂t
~ r , t)
B(~ = ~ × A(~
∇ ~ r, t)
A gauge transformation of the two potentials with a gauge function χ(~r, t)
~ r , t)
A(~ → ~ ′ (~r, t) = A(~
A ~ r, t) + ∇χ(~
~ r, t)
∂χ(~r, t)
φ(~r, t) → φ′ (~r, t) = φ(~r, t) −
∂t
~ ri , t) and B(~
changes the potentials but leaves the fields E(~ ~ ri , t) unchanged.
All equations describing the physics of a system must be form invariant under
this gauge transformation. This applies in particular to the time-dependent
Schrödinger equation
∂
ı~ |Ψ(t)i = Ĥ |Ψ(t)i
∂t
~ ri , t) in the Hamiltonian Ĥ by φ′ (~ri , t)

Replacing the potentials φ(~ri , t) and A(~
~ ′ (~ri , t) yields a new gauge transformed Hamiltonian Ĥ ′ .
and A
N N
i2 X
1 X hˆ ~ˆ ri , t) + e∇χ(~
~ ri , t) − ∂χ(~
r i , t)
Ĥ ′ = ~i + eA(~
p eφ̂(~ri , t) − e
2me i i
∂t
NM
e2 X ZK e2 X 1
− +
~K|
This Hamiltonian Ĥ ′ can also be obtained by the transformation

P
P
′ ∂ e
−ı i ~ χ(~
ri ,t) ∂ e χ(~
ı i ~ ri ,t)
Ĥ − ı~ =e Ĥ − ı~ e
∂t ∂t
The time-dependent Schrödinger equation is form invariant if

P e
′
|Ψ(t)i → |Ψ (t)i = e −ı i ~ χ(~
ri ,t)
|Ψ(t)i
The time-independent Schrödinger equation is form invariant under

P e
transformations with i ~ χ(~ri )
The form invariance of the Schrödinger equation leads to gauge invariant

expectation values of the Hamiltonian.
But expectation values of arbitrary operators are not gauge invariant.
In particular expectation values of the canonical momentum operator
~ˆ i
~ˆ i = −ı~∇
p
are not gauge invariant.

Expectation values of the mechanical or kinematical momentum operator ~π
~πi ≡ me ~vi = p ~ B (~ri , t)

~i + e A
are gauge invariant
î′ | Ψ′ i = hΨ | î | Ψi
′
X X
hΨ | ~π ~π
i i
The mechanical or kinematical momentum operator is therefore sometimes

also called the gauge invariant momentum operator.
An important gauge transformation in the context of the calculation of

molecular properties is given by the gauge function
1 ~ ~
χ(~ri ) = − B × RGO · ~ri
2
~ GO is the arbitrary gauge origin.
where R
This gauge function implies that
~ ri ) = − 1 B
∇χ(~ ~×R
~ GO
2
and that the vector potential for a uniform magnetic induction
~ B′ 1 ~ ~ GO )
A (~ri ) = B × (~ri − R
2
~ GO
becomes a linear function of the arbitrary gauge origin R
This has important consequences for all magnetic properties.

For time-dependent properties let us consider the case that φ(~r, t) = 0 and
that the vector potential depends only on time
~
∂ A(t)
~
E(t) = − ~ r, t) = ∇
B(~ ~ × A(t)
~ =0
∂t
This implies that the magnetic field vanishes and that the time-dependent,
spatially uniform electric field enters the Hamiltonian via the vector potential.
In the length gauge one chooses a gauge function
~ · ~ri
χ(~ri , t) = −A(t)
The transformed potentials become then
~ ′ (~ri , t)
A = ~ + ∇χ(~
A(t) ~ ri , t) = 0
∂χ(~ri , t) ~ · ~ri
∂ A(t)
′
φ (~ri , t) = φ(~ri , t) − = ~ · ~ri
= −E(t)
∂t ∂t
The time-dependent electric field enters the Hamiltonian via the scalar
potential and is coupled with the position vectors of the electrons.
In the velocity gauge one chooses a gauge function

Z t
e ~ 2 (t′ )dt′
χ(t) = − A
2me 0
This gives rise to a scalar potential
∂χ(t) e ~2
φ′ (t) = − = A (t)
∂t 2me
which cancels the second-order contribution to the molecular Hamiltonian.
The time-dependent, spatially uniform electric field enters the molecular
~ˆ · p
Hamiltonian only via the linear A(t) ~î term.
It is coupled through its vector potential to the canonical momentum or velocity
of the electrons.
A time-dependent, spatially uniform electric field can enter the Hamiltonian
~ˆ · p
~ · ~ri (length gauge) or as A(t)
either as −E(t) ~î (velocity gauge).

In the Lorenz gauge the gauge function χ(~ri , t) is chosen in such a way that
′
~ · A (~ri , t) +
~′ 1 ∂φ (~ri , t)
∇ 2
=0
c ∂t
With this gauge, the third and fourth Maxwell equations in vacuum
~ r , t)
∂ B(~
~ ~
∇ × E(~r, t) = −
∂t
~ r, t)
1 ∂ E(~
~ ~
∇ × B(~r, t) = 2
c ∂t
take a simple form in terms of the potentials
2 1 ∂ 2 φ(~r, t)
∇ φ(~r, t) − =0
c ∂t2
2~
~ r , t)
1 ∂ 2 A(~
∇ A(~r, t) − =0
c ∂t2

Chapter 3
Stephan P. A. Sauer
sauer@kiku.dk


3. Perturbation Theory
So far we have derived the molecular electronic Hamiltonian Ĥ in the presence

of static electromagnetic fields.
Now we want to obtain the energy E0 (F) ~ and wavefunction |Ψ0 (F)i~ of a
~.
particular state k in the presence of static electromagnetic fields F
i.e. We want to solve the time-independent Schrödinger equation
~ ) |Ψ0 (F)i
Ĥ(F ~ = E0 (F
~ ) |Ψ0 (F)i
~
We consider only electromagnetic fields which cause a slight change in the

nuclear and electronic structure of the molecule
We treat the fields as mere perturbations of the wavefunction and energy.
We are often more interested in the small changes in the wavefunction and
energy of the system than in the final state of the system.
Therefore we will not attempt to solve the time-independent Schrödinger
equation directly but rather apply time-independent perturbation theory.

3.1 The Hellmann-Feynman Theorem - 1
Let us consider the case of a Hamiltonian Ĥ(F~ ) with eigenfunctions |Ψ0 (F)i
~
and eigenvalues E0 (F~ ) which all depend on a general electromagnetic field F~
~ ) |Ψ0 (F)i
Ĥ(F ~ = E0 (F)
~ |Ψ0 (F)i
~
~
and assume that the eigenfunctions are normalized for all values of F
~ ) | Ψ0 (F)i
hΨ0 (F ~ =1
~
First derivative of the energy with respect to a component Fα··· of the field F
~
dE0 (F) d
= hΨ0 (F) ~ | Ĥ(F)
~ | Ψ0 (F)i
~
dFα··· dFα···
~)
∂ Ĥ(F ~
dΨ0 (F) ~
dΨ0 (F)
~
= hΨ0 (F ) | ~
| Ψ0 (F)i+h ~ ~ ~ ~
| Ĥ(F ) | Ψ0 (F)i+hΨ0 (F) | Ĥ(F) | i
∂Fα··· dFα··· dFα···
!
~ ~ ~
~ ) | ∂ Ĥ(F ) | Ψ0 (F)i
= hΨ0 (F ~ + E0 (F)~ h
dΨ0 (F) ~ + hΨ0 (F
| Ψ0 (F)i ~ ) | dΨ0 (F) i
∂Fα··· dFα··· dFα···
~)
∂ Ĥ(F d
~
= hΨ0 (F ) | ~ + E0 (F)
| Ψ0 (F)i ~ ~ | Ψ 0 (F
hΨ0 (F) ~ )i
∂Fα··· dFα···
~ is normalized for any value of Fα···

The last term vanishes, because Ψ0 (F)
The first derivative of the energy with respect to the component Fα··· can be
calculated as an expectation value of the derivative of the Hamiltonian
~
dE0 (F) ~
∂ Ĥ(F)
~
= hΨ0 (F) | ~ )i
| Ψ0 (F
dFα··· ∂Fα···
Derivative of the Hamiltonian leads to a possibly field-dependent operator
~
∂ Ĥ(F) X
~
P̂ (F ) = = F
Ôα··· + FF
Ôαβ··· + FF
Ôβα··· Fβ···
∂Fα··· β···
= P̂ (0) + P̂ (1) (F
~)
where in the following we use often the symbol P̂ for P̂ (0) = Ôα···
F
Often one is interested in the derivative of the energy evaluated for |F| = 0
!
~
∂E0 (F)

(0)
~)
∂ Ĥ(F (0) (0) (0)
= hΨ0 | | Ψ0 i = hΨ0 | P̂ | Ψ0 i
∂Fα··· ∂Fα···

|F|=0 |F|=0
which is called the Hellmann-Feynman theorem

The Hellmann-Feynman theorem establishes the equivalence between the

first derivative of the energy and the expectation value of an operator, given as
the corresponding derivative of the Hamiltonian.
In the derivation of the Hellmann-Feynman theorem we do not assume that we
~ as we
have solved the Schrödinger equation for the system without the field F
are going to do in the following sections on perturbation theory.

3.2 Time-Independent Perturbation Theory - 1
We make the assumptions that

The Hamiltonian is partitioned
Ĥ = Ĥ (0) + λĤ (1) + λ2 Ĥ (2)
The unperturbed Schrödinger has been solved exactly
Ĥ (0) |Ψ(0) (0) (0)

n i = En |Ψn i
(0) (0)
i.e. the unperturbed energies En and wavefunctions |Ψn i are known.
In reality this is of course not possible; however we will ignore this fact right
now.
(0)
The functions {|Ψn i} are the eigenfunctions of the Hermitian operator
Ĥ (0) and therefore form a complete orthonormal set.
The eigenfunction |Ψ0 (F ~ )i and eigenvalue E0 (F)
~ of the full Hamiltonian Ĥ
are close to those of the unperturbed Hamiltonian Ĥ (0) , i.e. the perturbation
~ is indeed small.
by F
(0)
The state |Ψ0 i is not degenerate
We expand the perturbed wavefunction and energy |Ψ0 (F)i ~ and E0 (F)
~ in a
(0) (0)
power series in λ around the exact solutions En and |Ψn i of the
unperturbed Hamiltonian Ĥ (0)
~ (0) (1)
~ ) + λ2 E (F (2)
~) + · · ·
E0 (F) = E0 + λE0 (F 0
~ (0) (1)
~ + λ2 |Ψ (F)i
~ + ··· (2)
|Ψ0 (F)i = |Ψ0 i + λ|Ψ0 (F)i 0
(m) ~ and |Ψ (m) ~ )i are called the m-th order correction to the energy and
E0 (F) 0 (F
wavefunction
(2) (2)
~ )i or E (F)|Ψ
~ (1)
~
Terms like Ĥ (1) |Ψ0 (F 0 0 (F)i are third-order terms
Assuming that the perturbations are small means that with increasing order m
the corrections to the energy and wavefunction become systematically smaller.

intermediate normalization of the perturbed wavefunction |Ψ0 i

(0) ~ )i
1 = hΨ0 | Ψ0 (F
(0) (0) (0) (1)
~ )i + λ2 hΨ | Ψ (F)i (0)
~ + ··· (2)
= hΨ0 | Ψ0 i + λhΨ0 | Ψ0 (F 0 0
(0)
The unperturbed wavefunction |Ψ0 i is normalized
(0) (0)
hΨ0 | Ψ0 i = 1
conditions on the higher-order (n > 0) corrections to the wavefunction

(0) (n)
~ =0
hΨ0 | Ψ0 (F)i
This is very convenient for the derivation of the higher-order corrections to the
energy
~ is not normalized, if one truncates the
However, the wavefunction |Ψ0 (F)i
power series at a finite order.
The truncated wavefunction has to be renormalize in each order.

Inserting the the power series expansions in the Schrödinger equation,

(0) (1) 2 (2) (0) (1) ~ 2 (2) ~
Ĥ + λĤ + λ Ĥ |Ψ0 i + λ|Ψ0 (F )i + λ |Ψ0 (F)i + · · ·

(0) (1) ~ 2 (2) ~ (0) (1) ~ 2 (2) ~
= E0 + λE0 (F) + λ E0 (F) + · · · |Ψ0 i + λ|Ψ0 (F)i + λ |Ψ0 (F)i + · · ·
Terms of the same order, i.e. power of λ, have to be equal on both sides
A series of equations, where the first- and second-order equations are
(1) (0) (0) (1) (1) (0)
~ + Ĥ (1) |Ψ i
Ĥ (0) |Ψ0 (F)i 0 = E0 |Ψ0 (F)i 0
~
~ + E (F)|Ψ
0 i
(2) (1) (0) (0) (2) (1) (1)
Ĥ (0) |Ψ0 (F
~ )i + Ĥ (1) |Ψ (F)i
0
~ + Ĥ (2) |Ψ i
0 = E0 |Ψ0 (F)i 0
~
~ + E (F)|Ψ ~
0 (F)i
(2)
~ )|Ψ i (0)
+ E0 (F 0
and the general m-th order equation for m > 2 can be written as
(m) (m−1) (m−2)
Ĥ (0) |Ψ0 ~ )i + Ĥ (1) |Ψ
(F 0
~ + Ĥ (2) |Ψ
(F)i 0
~
(F)i
m−1
(0) (m) ~ (i) (m−i) (m) (0)
X
= E0 |Ψ0 (F )i + ~ )|Ψ
E0 (F ~ +E
(F)i ~
(F)|Ψ
0 0 0 i
i=1

(m) ~
Expansion of the m-th order correction to the wavefunction |Ψ0 (F)i in the
(0)
complete set of eigenfunctions {|Ψn i} of the unperturbed Hamiltonian Ĥ (0)
(m) ~ (0) (m)
X
| Ψ0 (F)i = ~
|Ψn i Cn0 (F)
n6=0
(m) ~
The expansion coefficients Cn0 (F) are defined as the projection of the m-th
(0)
order correction to the wavefunction against the basis functions hΨn |
(m) ~ = hΨ(0) (m) ~ )i
Cn0 (F) n | Ψ0 (F
(m)
~ ) = 0 due to intermediate normalization
The n = 0 term is excluded, as C00 (F
Inserting the expansion of the m-th order correction to the wavefunction gives
2
(0) (m) ~ (0) (m−i)
X X X
(0) (i) ~
Ĥ |Ψk i Ck0 (F) + Ĥ |Ψk i Ck0 (F)
k6=0 i=1 k
m
(0) (0) (m) ~ (i) ~ (0) (m−i)
X X X
= E0 |Ψk i Ck0 (F) + E0 (F) |Ψk i Ck0 ~)
(F
k6=0 i=1 k
(0) ~ (−1) ~
if we define Ck0 (F ) = δk0 and Ck0 (F) =0
(m) ~
Expressions for the coefficients Cn0 (F) are derived by projecting these
(0)
equations against the corresponding basis functions hΨn |.
First-order coefficients
(0) (0)
(1) ~ hΨn | Ĥ (1) | Ψ0 i
Cn0 (F) = (0) (0)
E0 − En
First-order correction to the wavefunction

(0) (1) (0)
(1) ~ (0) hΨn | Ĥ | Ψ0 i
X
| Ψ0 (F )i = | Ψn i (0) (0)
n6=0 E0 − En
Coefficients of the m-th order correction to the wavefunction

2 X (0) (0) m−1 (m−i)
(m) ~
X hΨn | Ĥ (i) | Ψk i (m−i) ~
X (i) ~ Cn0 ~
(F)
Cn0 (F) = (0) (0)
Ck0 (F) − E0 (F ) (0) (0)
i=1 k E0 − En i=1 E0 − En
(0) ~ (−1) ~
with Ck0 (F ) = δk0 and Ck0 (F) =0

(m) ~
Expressions for the corrections to the energy E0 (F ) are derived by
(0)
projecting the equations against the the unperturbed ground state hΨ0 |
(1) (0)
~ ) = hΨ | Ĥ (1) | Ψ i (0)
E0 (F 0 0
(2) (0) (0)
~ ) = hΨ | Ĥ (2) | Ψ i + hΨ | Ĥ (1) | Ψ (F
~ )i (0) (1)
E0 (F 0 0 0 0
(m) (0)
~ ) = hΨ | Ĥ (2) | Ψ (m−2) (0)
~ )i + hΨ | Ĥ (1) | Ψ (m−1) ~
E0 (F 0 0 (F 0 0 (F)i
(1)
~ is independent of the changes in the wavefunction.
Note: E0 (F)
(m) ~ one gets the energy up to order 2m + 1
In general with Ψ0 (F)
(2m+1) ~ (m) ~ ) | Ĥ (2) | Ψ(m−1) ~ + hΨ (m−1) ~ ) | Ĥ (2) | Ψ (m) ~ )i
E0 (F) = hΨ0 (F 0 (F)i 0 (F 0 (F
(m) ~ (m) ~
+ hΨ0 (F ) | Ĥ (1) | Ψ0 (F )i
m
(2m+1−i−j) ~ (i) ~ (j) ~
X
− E0 (F )hΨ0 (F ) | Ψ0 (F )i
i,j=1
which is the well known 2m + 1 rule

Inserting the first-order correction to the wavefunction we get for the

second-order correction to the energy
(2) ~ (0) (0)

X hΨ(0) | Ĥ (1) | Ψ(0) (0)
n ihΨn | Ĥ
(1) (0)
| Ψ0 i
E0 (F ) = hΨ0 | Ĥ (2) | Ψ0 i + 0
(0) (0)
n6=0 E0 − En
(2)
~ consists of two terms:
E0 (F)
a ground state expectation value over the second-order Hamiltonian Ĥ (2)
a so-called sum-over-states term, which involves a summation over all
excited states of the system and transition moments between the ground
state and these excited states with the first-order Hamiltonian Ĥ (1) .
Inserting the expressions for the first- and second-order perturbation
Hamiltonians we can write it as
 
(2)
X (0) (0)
X hΨ(0)
0 | Ô F | Ψ(0) ihΨ(0) | Ô F | Ψ(0) i
α··· n n β··· 0
~
E0 (F) = FF
Fα··· hΨ0 | Ôαβ··· | Ψ0 i +
  Fβ···
(0) (0)
α,β,··· n6=0 E0 − En
(2) ~ ~.
showing the quadratic dependence of E0 (F) on components of the field F
3.3 Time-Independent Response Theory - 1
Perturbation theory applied to an expansion of an expectation value in the

presence of external and internal fields
~ ) in the presence of the
Expectation value of a field-dependent operator P̂ (F
~
field F
~ ) | P̂ (F)
hΨ0 (F ~ | Ψ0 (F)i
~ = hΨ0 (F)
~ | P̂ | Ψ0 (F ~ | P̂ (1) (F)
~ )i + hΨ0 (F) ~ | Ψ0 (F
~ )i
~ to first order
Expanding the field-dependent wavefunction |Ψ0 (F)i
(0) (0) (0) ~ ) | Ψ(0) i
~ | P̂ (F)
hΨ0 (F) ~ | Ψ0 (F
~ )i = hΨ0 | P̂ | Ψ0 i + hΨ0 | P̂ (1) (F 0
(0) (1)~ )i + hΨ (F (1)
~ ) | P̂ | Ψ i + · · · (0)
+ hΨ0 | P̂ | Ψ0 (F 0 0
The last three terms are linear in the components of the field and represent the
(0) (0) ~.
linear response of the expectation value hΨ0 | P̂ | Ψ0 i to the field F
Therefore we call this time-independent response theory.

3.3 Time-Independent Response Theory - 2
~ ) gives
Inserting the expressions for the first-order wavefunction and for P̂ (1) (F
(0) (0)
X (0) FF (0)
~ ~ ~
hΨ0 (F) | P̂ (F) | Ψ0 (F )i = hΨ0 | P̂ | Ψ0 i + FF
hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 iFβ···
β···

X X hΨ(0) F (0) (0) F (0)
0 | Ôα··· | Ψn ihΨn | Ôβ··· | Ψ0 i
+ 
(0) (0)
β··· n6=0 E0 − En

X hΨ(0) F (0) (0) F (0)
0 | Ôβ··· | Ψn ihΨn | Ôα··· | Ψ0 i
+ (0) (0)
 Fβ···
n6=0 E0 − En
We could have derived the terms linear in Fβ··· also as first derivative of
(2) ~
E0 (F) with respect to Fα··· .
However, later we are going to extend our treatment of response theory to the
case of time-dependent fields, where the energy is not the eigenvalue of the
Hamiltonian and where we only can work with expectation values.

3.4 Second Derivatives of the Energy - 1
Second derivatives of the energy with respect to two components Fα··· and
Fβ··· without relying on perturbation theory.
According to the Hellmann-Feynman theorem, the second derivative of the
energy is equal to the first derivative of the expectation value of the derivative
of the Hamiltonian for a non-zero value of the field
~
d 2 E0 (F) d ~
∂ Ĥ(F)
= ~
hΨ0 (F) | ~ )i
| Ψ 0 (F
dFβ··· dFα··· dFβ··· ∂Fα···
~ ∂ Ĥ(F)
dΨ0 (F) ~ ~ dΨ0 (F)
∂ Ĥ(F) ~
= h | ~ ~
| Ψ0 (F)i + hΨ0 (F) | | i
dFβ··· ∂Fα··· ∂Fα··· dFβ···
~
∂ 2 Ĥ(F)
~ |
+hΨ0 (F) ~
| Ψ0 (F)i
∂Fβ··· ∂Fα···
Inserting the first and second derivative of the Hamiltonian

~)
∂ Ĥ(F X
F FF FF
= Ôα··· + Ôαβ··· + Ôβα··· Fβ···
∂Fα··· β···
~)
∂ 2 Ĥ(F FF FF
= Ôαβ··· + Ôβα···
∂Fβ··· ∂Fα···
3.4 Second Derivatives of the Energy - 2
we obtain for the second derivative of the energy evaluated at zero field
strength

~)
d 2 E0 (F d ∂ Ĥ( ~
F)

= hΨ0 (F)~ | ~
| Ψ0 (F)i
dFβ··· dFα··· dFβ··· ∂Fα···
|F|=0 |F|=0

~
dΨ0 (F) dΨ 0 ( ~
F)
= h F ~
| Ôα··· | Ψ0 (F)i + hΨ0 (F) ~ | Ô F
α··· | i

dFβ··· dFβ···
|F|=0 |F|=0
(0) FF FF (0)
+ hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 i
Using anyway perturbation theory for the perturbed wavefunction |Ψ0 (F)i ~ one
obtains

2 ~ ) (1) ~ (1) ~
d E0 (F (0) F dΨ0 (F ) dΨ0 (F ) F (0)
= hΨ 0 | Ôα··· | i + h | Ôα··· | Ψ0 i
dFβ··· dFα··· dFβ··· dFβ···

|F|=0
(0) FF FF (0)
+ hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 i
Inserting the expression for the first-order wavefunction, would bring us back to
the expression for the second-order energy correction from perturbation theory.
3.5 Density Matrices - 1
P
Expectation value of a spin free one-electron operator Ô = i ô(i)
(0) (0)
hΨ0 | Ô | Ψ0 i
Z Z
(0)∗ (0)
X
= ··· Ψ0 (~ x2 , · · · , ~xN )
x1 , ~ ô(i) Ψ0 (~x1 , ~x2 , · · · , ~xN ) d~
x1 · · · d~
xN
i
P
Electrons are indistinguishable; each term in i ô(i) gives the same
(0) (0)
hΨ0 | Ô | Ψ0 i
Z Z
(0)∗ (0)
= N ··· Ψ0 (~ xN ) ô(1) Ψ0 (~
x1 , ~x2 , · · · , ~ x1 , ~x2 , · · · , ~
xN ) d~
x1 · · · d~
xN
The operator Ô is spin free and we can therefore integrate over the spin
(0) (0)
hΨ0 | Ô | Ψ0 i
Z Z
(0)∗ (0)
= N ··· Ψ0 (~r1 , ~r2 , · · · , ~rN ) ô(1) Ψ0 (~r1 , ~r2 , · · · , ~rN ) d~r1 · · · d~rN

Preparing the kernel for the reduced one-electron density matrix

(0) (0)
hΨ0 | Ô | Ψ0 i
Z Z
(0) (0)∗
= N ··· ô(1) Ψ0 (~r1 , ~r2 , · · · , ~rN ) Ψ0 (~r ′1 , ~r2 , · · · , ~rN ) d~r1 · · · d~rN
r ′1 =~
~ r1
The subscript ~r ′1 = ~r1 on the integral means that ~r ′1 should be set equal to ~r1
(0)
after ô(1) has acted on Ψ0 (~r1 , ~r2 , · · · , ~rN ) but before the integration is carried
out.
Using the definition of the reduced one-electron density matrix gives
Z
(0) (0)
hΨ0 | Ô | Ψ0 i = ô(1) P (~r1 , ~r ′1 ) d~r1
r ′1 =~
~ r1

~ also the electron density P (~r) and

In the presence of the perturbing field F
reduced one electron density matrix P (~r, ~r ′ ) become field dependent
~ = hΨ0 ({~ri }, F)
P (~r, F) ~ | D̂(~r) | Ψ0 ({~ri }, F)i
~
Z Z
~ = N · · · Ψ0 (~r, ~r2 , · · · , ~rN , F
P (~r, ~r ′ , F) ~ d~r2 · · · d~rN
~ ) Ψ∗0 (~r ′ , ~r2 , · · · , ~rN , F)
Expanding them in a perturbation series

(1)
X
P (~ ~ = P (~
r, F) r) + Pα (~
r) Fα + · · ·
α
(1)
X
P (~ ′~ = P (~
r , F)
r, ~ r ′) +
r, ~ Pα (~ r ′ ) Fα + · · ·
r, ~
α
where the first-order corrections are given as

(1) (1) (0) (0) (1)
Pα (~r) = hΨ0 ({~ r) | Ψ0 ({~
ri }) | D̂(~ ri })i + hΨ0 ({~ ri }) | D̂(~ r) | Ψ0 ({~ ri })i
Z Z
(1) ′ (1) (0)∗
Pα (~r, ~r ) = N · · · Ψ0 (~ r2 , · · · , ~
r, ~ rN ) Ψ0 (~ r ′, ~
r2 , · · · , ~
rN ) d~
r2 · · · d~
rN
Z Z
(0) (1)∗
+N · · · Ψ0 (~ r, ~ rN ) Ψ0 (~
r2 , · · · , ~ r ′, ~
r2 , · · · , ~
rN ) d~
r2 · · · d~
rN

The first-order correction to a field dependent expectation value of spin free

one-electron operator Ô can be written with the first-order reduced density
(1)
matrix Pα (~r, ~r ′ ) as
~ )i(1)
~ ) | Ô | Ψ0 (F (0) (1) ~ (1) ~ (0)
hΨ0 (F = hΨ0 | Ô | Ψ0 (F )i + hΨ0 (F ) | Ô | Ψ0 i
X Z
= Fα ô(1) Pα(1) (~r1 , ~r ′1 ) d~r1
α r ′1 =~
~ r1

3.6 The Ehrenfest Theorem - 1
Let us study the time evolution of an expectation value by deriving an

expression for the time derivative
d (0) (0)
hΨ0 (t) | P̂ | Ψ0 (t)i
dt
(0) (0)
∂Ψ0 (t) (0) (0) ∂ P̂ (0) (0) ∂Ψ0 (t)
= h | P̂ | Ψ0 (t)i + hΨ0 (t) | | Ψ0 (t)i + hΨ0 (t) | P̂ | i
∂t ∂t ∂t
(0) (0)
hΨ0 (t) | ∂∂tP̂ | Ψ0 (t)i = 0 if P̂ 6= P̂ (t)
Time derivative of the wavefunction ??

3.6 The Ehrenfest Theorem - 2
Let us study the time evolution of an expectation value by deriving an

expression for the time derivative
d (0) (0)
hΨ0 (t) | P̂ | Ψ0 (t)i
dt
(0) (0)
∂Ψ0 (t) (0) (0) ∂ P̂ (0) (0) ∂Ψ0 (t)
= h | P̂ | Ψ0 (t)i + hΨ0 (t) | | Ψ0 (t)i + hΨ0 (t) | P̂ | i
∂t ∂t ∂t
(0) (0)
hΨ0 (t) | ∂∂tP̂ | Ψ0 (t)i = 0 if P̂ 6= P̂ (t)
Time derivative of the wavefunction ⇐ time-dependent Schrödinger equation
∂ (0) ı (0)
|Ψ0 (t)i = − Ĥ |Ψ0 (t)i
∂t ~
∂ (0) ı (0)
hΨ0 (t)| = hΨ0 (t)|Ĥ
∂t ~
Ehrenfest theorem
d (0) (0) ı (0) (0)
hΨ0 (t) | P̂ | Ψ0 (t)i = − hΨ0 (t) | [P̂ , Ĥ] | Ψ0 (t)i
dt ~

3.7 The Off-Diagonal Hypervirial Theorem - 1
Application of the Ehrenfest theorem to two different stationary states and an

unperturbed Hamiltonian instead of a general time-dependent wavefunction
d
ı~ hΨ(0) (0) (0)
m (t) | P̂ | Ψn (t)i = hΨm (t) | [P̂ , Ĥ
(0)
] | Ψ(0)
n (t)i
dt
ı (0)
(0) (0)
Being stationary states, |Ψm (t)i = |Ψm ie− ~ Em t
and
(0)
(0) (0) − ı En t
|Ψn (t)i = |Ψn ie ~ are eigenfunctions of Ĥ (0) and we can write
d
ı~ hΨ(0) (0) (0) (0) (0) (0)
m (t) | P̂ | Ψn (t)i = (En − Em )hΨm (t) | P̂ | Ψn (t)i
dt
Combining both equations gives for m = n the hypervirial theorem
hΨ(0)
m | [P̂ , Ĥ
(0)
] | Ψ(0)
m i = 0
and for m 6= n the off-diagonal hypervirial theorem
hΨ(0)
m | [P̂ , Ĥ
(0)
] | Ψ(0) (0) (0) (0) (0)
n i = (En − Em )hΨm | P̂ | Ψn i

3.7 The Off-Diagonal Hypervirial Theorem - 2
Important example: the sum of the position operators of all electrons

N
X
r
P̂ = Ôα = r̂i,α
i
~ˆ r , Ĥ (0) ] = ı~ O
Recalling the well-known commutator relation [O ~ˆ p ,
me
where O~ˆ p is the total canonical momentum operator of the electrons

N
X
p
Ôα = p̂i,α
i
~ˆ p
we obtain the off-diagonal hypervirial relation for transition moments of O

ˆ r (0) ı~ ~ˆ p (0)
En(0) − (0)
Em hΨ(0)
m | ~
O | Ψ n i = hΨ(0)
m | O | Ψn i
me
~ˆ p
and the hypervirial relation for expectation values of O
hΨ(0) ~ˆ p (0)
m | O | Ψm i = 0

3.8 The Interaction Picture - 1
(normal) Schrödinger picture

Ô are time independent (apart from time dependent perturbations)
|Ψ0 (t)i carries the time dependence
Natural choice for time independent systems
Interaction or Dirac picture
Useful for a time dependent Hamiltonian
Ĥ(t) = Ĥ (0) + Ĥ (1) (t)
Wavefunctions |ΨI0 (t)i and operators Ôα

I
(t) in the interaction picture
ı (0)
|ΨI0 (t)i = e ~ Ĥ t
|Ψ0 (t)i
I ı (0) t ı (0) t
Ôα (t) = e ~ Ĥ Ôα e− ~ Ĥ
Differ in the form of the operators and wavefunctions: e.g. time dependence
But hΨ0 | Ô | Ψ0 i and hΨ0 | Ψn i are always the same
Other example: gauge transformations of the vector and scalar potentials
Consider Ĥ (1) (t) = 0
Time dependent wavefunction in Schrödinger picture

ıE (0) ı Ĥ (0) t
(0) −~ t (0) −~ (0)
|Ψ0 (t)i =e 0 |Ψ0 i =e |Ψ0 i
Time dependence is what ?

Consider Ĥ (1) (t) = 0
Time dependent wavefunction in Schrödinger picture

ıE (0) ı Ĥ (0) t
(0) −~ t (0) −~ (0)
|Ψ0 (t)i =e 0 |Ψ0 i =e |Ψ0 i
Time dependence is a rotation in the complex plane
Time dependent wavefunction in interaction picture

(0),I ı (0) t (0) ı (0) t ı (0) t (0) (0)
|Ψ0 (t)i = e ~ Ĥ |Ψ0 (t)i = e ~ Ĥ e− ~ Ĥ |Ψ0 i = |Ψ0 i
(0)
Time dependence of |Ψ0 (t)i is removed
Rotation due to Ĥ (0) is frozen out
(0)
Interaction picture switches to rotating frame of |Ψ0 (t)i
Time evolution of |ΨI0 (t)i is governed by Ĥ (1) (t) alone

3.9 Time-Dependent Perturbation Theory - 1
Looking for solutions to the time dependent Schrödinger equation

∂ h
(0) (1)
i
~ = Ĥ + Ĥ (t) |Ψ0 (t, F
~ )i
ı~ |Ψ0 (t, F)i
∂t
with a time dependent first order perturbation Hamiltonian Ĥ (1) (t)

(1)
X F
Ĥ (t) = Ôβ··· Fβ··· (t)
β···
we consider only time-dependent electric fields here and have no Ĥ (2) (t)
Perturbation expansion of the time dependent wavefunction
~ = |Ψ(0) (t)i + |Ψ(1) (t, F

|Ψ0 (t, F)i ~ )i + |Ψ(2) (t, F
~ )i + · · ·
0 0 0

F
In length gauge and dipole approximation Ôβ··· is the electric dipole operator
N
X
F
Ôβ··· = −e (ri,β − RO,β )
i
Time-dependent field Fβ··· (t) for monochromatic linear polarized radiation

1 ıωt −ıωt
Fβ··· (t) = Fβ··· (ω) cos(ωt) = Fβ··· (ω) e + e
2
or for a general pulse of coherent but polychromatic radiation
Z ∞ Z ∞
1 1
dω Fβ··· (ω) eıωt + e−ıωt = dω Fβ··· (ω) e−ıωt

Fβ··· (t) =
2 0 2 −∞
is the Fourier transform of Fβ··· (t) into Fourier components 21 Fβ··· (ω)
The perturbation Hamiltonian expressed in terms of Fourier components
Z ∞ Z ∞
1 X
Ĥ (1) (t) = dω Ĥ (1) (ω) = ω
Ôβ··· dω Fβ··· (ω) e−ıωt
−∞ 2 −∞
β···

Interaction or Dirac representation of the time dependent wavefunction

ı Ĥ (0) t
|ΨI0 (t, F
~ )i =e ~ ~ )i
|Ψ0 (t, F
∂
Time derivative ı~ ∂t ~ ?
of |ΨI0 (t, F)i


ı Ĥ (0) t
|ΨI0 (t, F
~ )i =e ~ ~ )i
|Ψ0 (t, F
∂
Time derivative ı~ ∂t ~
of |ΨI0 (t, F)i
∂ I ~ ı Ĥ (0) t
(0) ~ ~ ı Ĥ (0) t ∂ ~ )i
ı~ |Ψ0 (t, F )i = −Ĥ e |Ψ0 (t, F )i + e ~ ı~ |Ψ0 (t, F
∂t ∂t
And now ?


ı Ĥ (0) t
|ΨI0 (t, F
~ )i =e ~ ~ )i
|Ψ0 (t, F
∂
Time derivative ı~ ∂t ~
of |ΨI0 (t, F)i
∂ I ~ ı Ĥ (0) t
(0) ~ ~ ı Ĥ (0) t ∂ ~ )i
ı~ |Ψ0 (t, F )i = −Ĥ e |Ψ0 (t, F )i + e ~ ı~ |Ψ0 (t, F
∂t ∂t
Use time dependent Schrödinger equation for ∂ ~ )i

|Ψ0 (t, F
∂t
~ in the interaction picture
Equation of motion for |ΨI0 (t, F)i
∂ I ~ = Ĥ (1),I (t) |ΨI0 (t, F

~ )i
ı~ |Ψ0 (t, F)i
∂t
Time dependence is governed by Ĥ (1),I (t) alone ⇐ Interaction picture

Integration
Z t
~ − |ΨI0 (−∞, F
|ΨI0 (t, F)i ~ )i = 1 dt′ Ĥ (1),I (t′ ) |ΨI0 (t′ , F)i
~
ı~ −∞

Limit t → −∞ of the Interaction picture wavefunction ?

Limit of the perturbed Schrödinger wavefunction |Ψ0 (t)i
(0) (0) ı Ĥ (0) t
−~ (0)
|Ψ0 (−∞)i = lim |Ψ0 (t)i =e |Ψ0 i
t→−∞
⇒ Limit t → −∞ of the Interaction picture wavefunction

(0) t
~ = lim e ~ı Ĥ
|ΨI0 (−∞, F)i ~ = |Ψ(0) i
|Ψ0 (t, F)i 0
t→−∞
Time dependent wavefunction Ψ0 (t) in the Schrödinger picture

Z t
ı (0)
~ = e− ~ Ĥ t |Ψ(0) i + 1 ı Ĥ (0) (t−t′ )
|Ψ0 (t, F)i 0 dt ′ −~
e Ĥ (1) (t′ ) |Ψ0 (t′ , F
~ )i
ı~ −∞
However, this is only the integral version of

∂ I ~ = Ĥ (1),I (t) |ΨI0 (t, F
~ )i
ı~ |Ψ0 (t, F)i
∂t

Expression for time dependent wavefunction Ψ0 (t)

Z t
ı
− ~ Ĥ (0) t (0) 1 ı Ĥ (0) (t−t′ )
~
|Ψ0 (t, F)i = e |Ψ0 i + ′ −~
dt e Ĥ (1) (t′ ) |Ψ0 (t′ , F
~ )i
ı~ −∞
can be solved how ?

Expression for time dependent wavefunction Ψ0 (t)

Z t
ı
− ~ Ĥ (0) t (0) 1 ı Ĥ (0) (t−t′ )
~
|Ψ0 (t, F)i = e |Ψ0 i + ′ −~
dt e Ĥ (1) (t′ ) |Ψ0 (t′ , F
~ )i
ı~ −∞
has to be solved iteratively

~ )i = |Ψ(0) (t′ )i gives
First iteration |Ψ0 (t′ , F 0
Z t
ı (0)
~ = e− ~ Ĥ t |Ψ(0) i+e− ~ Ĥ t ı (0) 1 ′ ~ı Ĥ (0) t′ (1) ′ ı Ĥ (0) t′
−~ (0)
|Ψ0 (t, F)i 0 dt e Ĥ (t ) e |Ψ 0 i+· · ·
ı~ −∞
and thus the first order correction to the time-dependent wavefunction is

Z t
(1) ı (0) 1 (0)
~
|Ψ0 (t, F)i = e− ~ Ĥ t dt′ Ĥ (1),I (t′ ) |Ψ0 i
ı~ −∞
~ )i = |Ψ(0) (t′ )i + |Ψ(1) (t′ , F)i

Second-order iteration : |Ψ0 (t′ , F ~ gives
0 0
2 Z t Z t′
(2) ı Ĥ (0) t 1 (0)
~ =e
|Ψ0 (t, F)i −~
dt′ dt′′ Ĥ (1),I (t′ ) Ĥ (1),I (t′′ ) |Ψ0 i
ı~ −∞ −∞

3.10 Transition Probabilities and Rates - 1
For the interpretation of the time-dependent wavefunctions it is useful to

expand them in a complete set of unperturbed wavefunctions
(1)
X ı Ĥ (0) t (1)
~ )i
|Ψ0 (t, F = e −~
|Ψ(0) ~
n i Cn0 (t, F)
n6=0
where the time-dependent first-order coefficients are defined as

(1) ı Ĥ (0) t (1)
~)
Cn0 (t, F = he −~
Ψ(0)
n | Ψ ~
0 (t, F)i
(1) ~ )|2 is the probability for finding the system in the stationary
The norm |Cn0 (t, F
(0)
state |Ψn i at time t.
(0)
Before the perturbation was turned on, the system was in the state Ψ0 .
(1) ~ )|2 is the probability for the transition from Ψ to Ψn , i.e. a (0) (0)
|Cn0 (t, F 0
transition probability.

Inserting the first-order wavefunction

Z t
(1) ı Ĥ (0) t ı Ĥ (0) t 1 (0)
~)
Cn0 (t, F = hΨ(0)
n |e
~ e −~
dt′ Ĥ (1),I (t′ ) | Ψ0 i
ı~ −∞
Z t
1 ı (0) ′ ı (0) ′ (0)
= dt′ hΨ(0)
n |e
~
Ĥ t
Ĥ (1) (t′ )e− ~ Ĥ t
| Ψ0 i
ı~ −∞
(0) (0)
Using that Ψ0 and Ψn are eigenfunctions of Ĥ (0) and defining a transition
angular frequency ωn0
(0) (0)
En − E0
ωn0 =
~
we can write the first-order probability amplitude as
Z t
(1) ~ 1 ′ (0) (1) ′ (0) ıωn0 t′
Cn0 (t, F ) = dt hΨn | Ĥ (t ) | Ψ0 ie
ı~ −∞

With the expressions for the perturbation Hamiltonian it becomes

Z ∞
(1) ~) = 1 X (0) ω (0)
Cn0 (t, F hΨn | Ôβ··· | Ψ0 i dω Fβ··· (ω)
2ı~ β··· 0
Z t Z t
′ ′
× dt′ eı(ωn0 −ω)t + dt′ eı(ωn0 +ω)t
−∞ −∞
Integration over t′ gives then
(1) ~ 1 X (0) ω (0)

Cn0 (t, F) = hΨn | Ôβ··· | Ψ0 i
2ı~
β···
Z ∞ ı(ωn0 −ω)t ı(ωn0 +ω)t

e −1 e −1
× dω Fβ··· (ω) +
0 ı(ωn0 − ω) ı(ωn0 + ω)
The first term is negligible unless ω ≃ ωn0 , the second term unless ω ≃ −ωn0 .
The first term is the absorption of a photon of energy ~ω.
The second term is the emission of a photon of energy ~ω.
We can consider the two processes separately.
The transition probability for absorption is

2 1
2 Z ∞

(1) ~
2
1 X
(0) ω (0) 2 2 sin 2
(ωn0 − ω)t
Cn0 (t, F ) = 2 hΨn | Ôβ··· | Ψ0 i dω Fβ··· (ω)
2~ 0 (ωn0 − ω)2
β···
2 sin2 ( 1
2
(ωn0 −ω)t)
As a function of the frequency ω the term (ωn0 −ω)2
has one sharp
maximum for ω ≃ ωn0 , which dominates the integral.
We approximate Fβ··· (ω) by its value Fβ··· (ωn0 ) at ωn0 and integrate
(1) ~ 2
2
π X (0) ω

(0) 2
Cn0 (t, F ) = 2 t hΨn | Ôβ··· | Ψ0 i Fβ··· (ωn0 )
2~
β···
The transition probability to first-order increases linear with time.

(1)
Defining a transition rate, Wn0 , as time-derivative of the transition probability
d (1) ~ 2
2
π X (0) ω

(1) (0) 2
Wn0 = Cn0 (t, F ) = 2 hΨn | Ôβ··· | Ψ0 i Fβ··· (ωn0 )
dt 2~ β···
(0)ω (0)
The matrix element hΨn | Ôβ··· | Ψ0 i is called a transition moment Mn0,β··· .
3.11 Time Dependent Response Theory - 1
~ ) | P̂ | Ψ0 (t, F)i
Perturbation expansion of an expectation value hΨ0 (t, F ~ in the
~ (t)
presence of a general, time-dependent field F
~ ) | P̂ | Ψ0 (t, F)i
hΨ0 (t, F ~
(0) (0) (0) (1) ~ )i + hΨ(1) (t, F) ~ | P̂ | Ψ(0) (t)i + · · ·
= hΨ0 (t) | P̂ | Ψ0 (t)i + hΨ0 (t) | P̂ | Ψ0 (t, F 0 0
Z t
(0) (0) 1 (0) ı (0) (0)
= hΨ0 (t) | P̂ | Ψ0 (t)i + dt′ hΨ0 (t) | P̂ e− ~ Ĥ t Ĥ (1),I (t′ ) | Ψ0 i
ı~ −∞
Z t
1 (0) ı (0) (0)
− dt′ hΨ0 | Ĥ (1),I (t′ ) e ~ Ĥ t P̂ | Ψ0 (t)i + · · ·
ı~ −∞
Using
(0) ı (0) t (0)
|Ψ0 (t)i = e− ~ Ĥ |Ψ0 i
commutator [Â, B̂] = ÂB̂ − B̂ Â
we can write this more compact as
~ ) | P̂ | Ψ0 (t, F)i
hΨ0 (t, F ~
Z t
(0) (0) 1 (0) (0)
= hΨ0 | P̂ | Ψ0 i + dt′ hΨ0 | [P̂ I (t), Ĥ (1),I (t′ )] | Ψ0 i + · · ·
ı~ −∞
Extending the integration limit to ∞
~ | P̂ | Ψ0 (t, F
hΨ0 (t, F) ~ )i
Z ∞
(0) (0) 1 (0) (0)
= hΨ0 | P̂ | Ψ0 i + dt′ Θ(t − t′ ) hΨ0 | [P̂ I (t) , Ĥ (1),I (t′ )] | Ψ0 i + · · ·
ı~ −∞
with Heaviside step function Θ(t) = {1 for t ≥ 0 and 0 for t < 0}

Using definition of the first-order perturbation Hamiltonian
~ )i = hΨ(0) | P̂ | Ψ(0) i
~ | P̂ | Ψ0 (t, F
hΨ0 (t, F) 0 0
Z ∞
1 ′ ′ (0) I
X F,I ′ (0)
+ dt Θ(t − t ) hΨ0 | [P̂ (t) , Ôβ··· (t )Fβ··· (t′ )] | Ψ0 i + · · ·
ı~ −∞
β···
(0) (0)
= hΨ0 | P̂ | Ψ0 i
XZ ∞ ′
1 (0) (0)

+ dt Fβ··· (t′ ) Θ(t − t′ ) hΨ0 | [P̂ I (t) , Ôβ···
F,I ′
(t )] | Ψ0 i + · · ·
β··· −∞
ı~
The second term is the time-dependent linear response of the expectation

(0) (0)
value hΨ0 | P̂ | Ψ0 i to the time-dependent field.
n o
′ 1 (0) I F,I ′ (0)
Θ(t − t ) ı~ hΨ0 | [P̂ (t) , Ôβ··· (t )] | Ψ0 i expresses how susceptible the
expectation value is to changes by the time-dependent field.
It is called the linear response function in the time domain and is given its
F,I ′
own symbol hh P̂ I (t) ; Ôβ··· (t ) ii defined as
I F,I ′ ′ 1 (0)
h
I F,I ′
i
(0)
hh P̂ (t) ; Ôβ··· (t ) ii = Θ(t − t ) hΨ0 | P̂ (t), Ôβ··· (t ) | Ψ0 i
ı~
It is often also called the polarization propagator in the time domain , while
mathematically it is a double-time Green’s function.
The expansion of the time-dependent, perturbed expectation value as
~ | P̂ | Ψ0 (t, F
~ )i (0) (0)
hΨ0 (t, F) = hΨ0 | P̂ | Ψ0 i
Z ∞ X
+ dt ′
hh P̂ I (t) ; Ôβ···
F,I ′
(t ) iiFβ··· (t′ ) + · · ·
−∞ β···

Is the propagator a function of two times t and t′ ?
′ 1
h i
I F,I ′ (0) I F,I ′ (0)
ı~

Is the propagator a function of two times t and t′ ?
′ 1
h i
I F,I ′ (0) I F,I ′ (0)
ı~
′ 1
h ı (0) ı Ĥ (0) t
i
(0) Ĥ t − F,I ′ (0)
= Θ(t − t ) hΨ0 | e ~ P̂ e ~ , Ôβ··· (t ) | Ψ0 i
ı~
′ 1 ı Ĥ (0) t ı Ĥ (0) t
h i
(0) −~ F,I ′ (0)
= Θ(t − t ) hΨ0 | P̂ , e Ôβ··· (t ) e ~ | Ψ0 i
ı~
′ 1
h i
(0) F,I ′ (0)
= Θ(t − t ) hΨ0 | P̂ , Ôβ··· (t − t) | Ψ0 i
ı~
F,I ′
= hh P̂ ; Ôβ··· (t − t) ii
Physics says : NO !

Fourier transform of the time-dependent polarization propagator

Z ∞
ω F,I
hh P̂ ; Ôβ··· iiω = hh P̂ ; Ôβ··· (t) ii e−iωt dt
−∞
is the frequency-dependent polarization propagator

Carrying out the Fourier transform leads to the spectral representation
ω
X hΨ(0)
0 | P̂ | Ψ
(0)
n ihΨ
(0)
n | Ô ω
β··· | Ψ
(0)
0 i
hh P̂ ; Ôβ··· iiω = (0) (0)
n6=0 ~ω + E0 − En
X hΨ(0)
0 | Ô ω
β··· | Ψ
(0)
n ihΨ
(0)
n | P̂ | Ψ
(0)
0 i
+ (0) (0)
n6=0 −~ω + E0 − En
Expansion of the time-dependent, perturbed expectation value becomes
~ | P̂ | Ψ0 (t, F
~ )i (0) (0)
hΨ0 (t, F) = hΨ0 | P̂ | Ψ0 i
1 ∞
Z X
−ıωt ω
+ dω e hh P̂ ; Ôβ··· iiω Fβ··· (ω) + · · ·
2 −∞
β···

General symmetry property of the polarization propagator

ω ω
hh P̂ ; Ôβ··· iiω = hh Ôβ··· ; P̂ ii−ω
with the special cases

Two real, hermitian operators or for two purely imaginary, hermitian operators
ω ω
hh P̂ ; Ôβ··· iiω = hh P̂ ; Ôβ··· ii−ω
One real hermitian and one imaginary hermitian operator

ω ω
hh P̂ ; Ôβ··· iiω = −hh P̂ ; Ôβ··· ii−ω
In the static case, ω = 0,
F F
X hΨ(0) F (0) (0) F (0)
0 | Ôα··· | Ψn ihΨn | Ôβ··· | Ψ0 i
hh Ôα··· ; Ôβ··· iiω=0 = (0) (0)
n6=0 E0 − En
X hΨ(0) F (0) (0) F (0)
0 | Ôβ··· | Ψn ihΨn | Ôα··· | Ψ0 i ∂ 2 E0
(2)
+ (0) (0)
=
E0 − En ∂Fα ∂Fβ···
n6=0
For magnetic properties we must choose the operator P̂ to be the perturbation

~
dependent operator P̂ (F)
∂ Ĥ( ~
F) X FF
~) =
P̂ (F F
= Ôα··· + FF
Ôαβ··· + Ôβα··· Fβ···
∂Fα···
β···
The expansion of the expectation value becomes then
~ | P̂ (F
hΨ0 (t, F) ~ )i = hΨ(0) | P̂ | Ψ(0) i
~ ) | Ψ0 (t, F
0 0
Z ∞
1 −ıω1 t
X (0) FF FF (0) ω

+ dω1 e hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 i + hh P̂ ; Ôβ··· iiω Fβ··· (ω)
2 −∞
β···
+ ···
In the static case, ω = 0, this is reduced to
~ ) | P̂ (F)
hΨ0 (F ~ = hΨ(0) | P̂ | Ψ(0) i
~ | Ψ0 (F)i
0 0
X (0) FF FF (0) F

+ hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 i + hh P̂ ; Ôβ··· iiω=0 Fβ··· + · · ·
β···

In the static case, ω = 0
~ ) | P̂ (F)
hΨ0 (F ~ = hΨ(0) | P̂ | Ψ(0) i
~ | Ψ0 (F)i
0 0
X (0) FF FF (0) F

+ hΨ0 | Ôαβ··· + Ôβα··· | Ψ0 i + hh P̂ ; Ôβ··· iiω=0 Fβ··· + · · ·
β···
this is another way of writing

(0) (0) (0) ~ ) | Ψ(0) i
~ | P̂ (F)
hΨ0 (F) ~ | Ψ0 (F
~ )i = hΨ0 | P̂ | Ψ0 i + hΨ0 | P̂ (1) (F 0
(0) (1)
~ )i + hΨ (F (1)
~ ) | P̂ | Ψ i + · · · (0)
+ hΨ0 | P̂ | Ψ0 (F 0 0
We can identify the static response function as the first derivative of the
first-order correction to an perturbed expectation value,
F ∂ (0) (1) ~ (1) ~ (0)

hh P̂ ; Ôβ··· iiω=0 = hΨ0 | P̂ | Ψ0 (F)i + hΨ0 (F ) | P̂ | Ψ0 i
∂Fβ···
(1)
or relate it to the first-order reduced density matrix Pα (~r, ~r ′ )
Z
F (1)
hh P̂ ; Ôβ··· iiω=0 = p̂(1) Pβ (~r1 , ~r ′1 ) d~r1
r ′1 =~
~ r1 Molecular Electromagnetism A Computational Chemistry Approach – p.49/74
Collecting all second-order terms in the time dependent expectation value

(1) ~ | P̂ | Ψ(1) (t, F)i
~ + hΨ(2) (t, F
~ ) | P̂ | Ψ(0) (t)i + hΨ(0) (t) | P̂ | Ψ(2) (t, F)i
~
hΨ0 (t, F) 0 0 0 0 0
inserting the corrections to the wavefunction

2 Z t Z t′
1 ′ ′′ (0)
hh
I (1),I ′
i
(1),I ′′
i
(0)
dt dt hΨ0 | P̂ (t) , Ĥ (t ) , Ĥ (t ) | Ψ0 i
ı~ −∞ −∞
or
2 Z ∞ Z ∞
1 ′ ′ ′′ ′ ′′ (0)
hh
I (1),I ′
i
(1),I ′′
i
(0)
dt Θ(t−t ) dt Θ(t −t )hΨ0 | P̂ (t), Ĥ (t ) , Ĥ (t ) | Ψ0 i
ı~ −∞ −∞
inserting the first-order perturbation Hamiltonian

Z ∞ Z ∞ X
′ ′′
dt dt Fβ··· (t′ )Fγ··· (t′′ )
−∞ −∞ β···,γ···
( 2 )
1 (0)
hh
I
i i
(0)
× Θ(t − t′ ) Θ(t′ − t′′ ) F,I ′ F,I ′′
hΨ0 | P̂ (t), Ôβ··· (t ) , Ôγ··· (t ) | Ψ0 i
ı~

Quadratic response function
hh P̂ I (t) ; Ôβ···
F,I ′ F,I ′′
(t ), Ôγ··· (t ) ii
2
′ ′ ′′ 1 (0)
hh
I F,I ′
i
F,I ′′
i
(0)
= Θ(t − t )Θ(t − t ) hΨ0 | P̂ (t), Ôβ··· (t ) , Ôγ··· (t ) | Ψ0 i
ı~
The quadratic response function depends on two time differences t1 = t′ − t

and t2 = t′′ − t
hh P̂ I (t) ; Ôβ···
F,I ′ F,I ′′
(t ), Ôγ··· F,I
(t ) ii = hh P̂ ; Ôβ··· F,I
(t1 ), Ôγ··· (t2 ) ii
2
1 (0)
hh
F,I
i
F,I
i
(0)
= Θ(−t1 )Θ(−t2 ) hΨ0 | P̂ , Ôβ··· (t1 ) , Ôγ··· (t2 ) | Ψ0 i
ı~

Quadratic response function in the frequency domain

ω1 ω2
hh P̂ ; Ôβ··· , Ôγ··· iiω1 ,ω2

X  hΨ(0) (0) (0) ω1 (0) (0) ω2 (0)
0 | P̂ | Ψn ihΨn | Ôβ··· | Ψm ihΨm | Ôγ··· | Ψ0 i
= P (ω1 ↔ ω2 )
 E (0) − E (0) − ~(ω + ω ) E (0) − E (0) − ~ω
n6=0,m6=0 n 0 1 2 m 0 2
(0) ω1 (0) (0) (0)

ω2 (0) (0)
hΨ0 | Ôγ··· | Ψn ihΨn | P̂ | Ψm ihΨm | Ôβ··· | Ψ0 i
+
(0) (0) (0) (0)
En − E0 + ~ω1 Em − E0 − ~ω2

(0) ω1 (0) (0) ω2 (0) (0) (0)
hΨ0 | Ôβ··· | Ψn ihΨn | Ôγ··· | Ψm ihΨm | P̂ | Ψ0 i 
+
(0) (0) (0) (0)
En − E0 + ~ω1 Em − E0 + ~(ω1 + ω2 ) 
(0) (0) (0) (0) (0) (0)

where hΨn | P̂ | Ψm i ≡ hΨn | P̂ − hΨ0 | P̂ | Ψ0 i | Ψm i, matrix element of a
fluctuation operator.

Expansion of the time-dependent expectation value in terms of linear and

quadratic response functions in the time domain
Z ∞ X
hΨ0 (t, F) ~ )i = hΨ(0) | P̂ | Ψ(0) i +
~ | P̂ | Ψ0 (t, F dt ′
hh P̂ ; Ô F,I ′
(t − t) iiFβ··· (t ′
)
0 0 β···
−∞ β···
Z ∞ Z ∞
1 ′ ′′
X F,I ′ F,I ′′
+ dt dt hh P̂ ; Ôβ··· (t − t), Ôγ··· (t − t) iiFβ··· (t′ ) Fγ··· (t′′ ) + · · ·
2 −∞ −∞ βγ···
or in the frequency domain
~ )i = hΨ(0) | P̂ | Ψ(0) i
~ | P̂ | Ψ0 (t, F
hΨ0 (t, F) 0 0
Z ∞
1 −ıωt
X ω
+ dω1 e hh P̂ ; Ôβ··· iiω1 Fβ··· (ω)
2 −∞
β···
1 ∞
Z Z ∞ X
−ıω1 t −ıω2 t ω1 ω2
+ dω1 e dω2 e hh P̂ ; Ôβ··· , Ôγ··· iiω1 ,ω2 Fβ··· (ω1 )Fγ··· (ω2 )
8 −∞ −∞ β··· ,γ···
+ ··· .

3.12 Matrix Representation of the Propagator - 1
Derivation of an alternative expression for the polarization propagator

Time derivative of a polarization propagator
d F,I
ı~ hh P̂ ; Ôβ··· (t) ii
dt
dΘ(−t) 1 (0)
h
F,I
i
(0) d (0)
h
F,I
i
(0)
= ı~ hΨ0 | P̂ , Ôβ··· (t) | Ψ0 i + Θ(−t) hΨ0 | P̂ , Ôβ··· (t) | Ψ0 i
dt ı~ dt
(0)
h
F,I
i
(0) 1 (0)
h h
(0) F,I
ii
(0)
=−δ(−t)hΨ0 | P̂ , Ôβ··· (t) | Ψ0 i − Θ(−t) hΨ0 | P̂ , Ĥ , Ôβ··· (t) | Ψ0 i
ı~
h i
(0) (0)
= −δ(t)hΨ0 | P̂ , Ôβ··· | Ψ0 i − hh P̂ ; [Ĥ (0) , Ôβ···
F,I
(t)] ii
Equation of motion for the polarization propagator in the time domain

where it was used that
d F,I 1
Ô (t) = F,I
[Ôβ··· (t) , Ĥ (0) ]
dt β··· ı~
Z t
Θ(t) = δ(t′ ) dt′
−∞
F,I F,I
δ(t) Ôβ··· (t) = δ(t − 0) Ôβ··· (t = 0) = δ(t) Ôβ···
Equation of motion for the polarization propagator in the time domain

d F,I
ı~ hh P̂ ; Ôβ··· (t) ii
dt
h i
(0) (0)
= −δ(t)hΨ0 | P̂ , Ôβ··· | Ψ0 i − hh P̂ ; [Ĥ (0) , Ôβ···
F,I
(t)] ii
Equation of motion in frequency domain

ω (0) ω (0)
~ωhh P̂ ; Ôβ··· iiω = hΨ0 | [P̂ , Ôβ··· ] | Ψ0 i + hh P̂ ; [Ĥ (0) , Ôβ···
ω
] iiω
derived by using the inverse Fourier transform of the polarization propagator

Z ∞
1 ω
F,I
hh P̂ ; Ôβ··· (t) ii = dω hh P̂ ; Ôβ··· iiω eiωt
2π −∞
and another representation of the Dirac δ function

Z ∞
1
δ(t) = eiωt dω
2π −∞

Equation of motion in frequency domain derived from spectral representation
ω
X hΨ(0) (0) (0) ω (0)
0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i
hh P̂ ; Ôβ··· iiω = (0) (0)
n6=0 ~ω + E0 − En
X hΨ(0) ω (0) (0) (0)
0 | Ôβ··· | Ψn ihΨn | P̂ | Ψ0 i
+ (0) (0)
n6=0 −~ω + E0 − En
ab ab d ab
using that c+d
= c
− c c+d
in the first term gives
(0) (0) ω (0) (0) (0) (0) ω (0) (0)

hΨ0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i hΨ0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i
(0) (0)
=
~ω + E0 − En ~ω

(0) (0) (0) ω (0) (0) (0)
hΨ0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i E0 − En
−
(0) (0)
~ω ~ω + E0 − En

Using the same trick also for the second term
ω 1 X n (0) (0)
hh P̂ ; Ôβ··· iiω = hΨ0 | P̂ | Ψ(0) (0) ω
n ihΨn | Ôβ··· | Ψ0 i
~ω n
o
(0) ω (0) (0) (0)
−hΨ0 | Ôβ··· | Ψn ihΨn | P̂ | Ψ0 i
 (0)
(0) (0) ω (0) (0) (0)
1 X  hΨ0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i E0 − En
− (0) (0)
~ω n6=0
 ~ω + E0 − En

(0) ω (0) (0) (0) (0) (0)
hΨ0 | Ôβ··· | Ψn ihΨn | P̂ | Ψ0 i E0 − En 
− (0) (0)
−~ω + E0 − En 
where we can use the resolution of the identity

X (0)
1= |Ψn ihΨ(0)n |
n
(0) (0)
and that |Ψn i are eigenfunctions of the Hamiltonian Ĥ (0) with eigenvalue En

leading to
ω 1 (0) ω (0)
hh P̂ ; Ôβ··· iiω = hΨ0 | [P̂ , Ôβ··· ] | Ψ0 i
~ω
( (0) (0) (0) (0)
(0) ω
1 X hΨ0 | P̂ | Ψn ihΨn | [Ĥ , Ôβ··· ] | Ψ0 i
+ (0) (0)
~ω n ~ω + E0 − En
(0) (0) (0) (0)
)
hΨ0 | [Ĥ (0) , Ôβ···
ω
] | Ψn ihΨn | P̂ | Ψ0 i
+ (0) (0)
−~ω + E0 − En
or
ω 1 (0) ω (0) 1
hh P̂ ; Ôβ··· iiω = hΨ0 | [P̂ , Ôβ··· ] | Ψ0 i + hh P̂ ; [Ĥ (0) , Ôβ···
ω
] iiω
~ω ~ω
Taking the zero frequency limit of the equation of motion in the frequency
domain, we obtain the following relation between a polarization propagator and
a ground state expectation value
(0) ω (0) ω
hΨ0 | [P̂ , Ôβ··· ] | Ψ0 i = hh P̂ ; [Ôβ··· , Ĥ (0) ] iiω=0


ω (0) ω (0)
ω
] iiω
Iterating on this equation gives what ?


ω (0) ω (0)
ω
] iiω
Iterating on this equation:

Moment expansion of the polarization propagator

ω 1 (0) ω (0)
hh P̂ ; Ôβ··· iiω = hΨ0 | [P̂ , Ôβ··· ] | Ψ0 i
~ω
2
1 (0)
h h
(0) ω
ii
(0)
+ hΨ0 | P̂ , Ĥ , Ôβ··· | Ψ0 i
~ω
3
1 (0)
h h
(0)
h
(0) ω
iii
(0)
+ hΨ0 | P̂ , Ĥ , Ĥ , Ôβ··· | Ψ0 i
~ω
+ ···

Superoperator formalism
A superoperator acts on operators in a space of operators
Superoperator binary product
ω (0) ω (0)
(P̂ | Ôβ··· ) = hΨ0 | [P̂ † , Ôβ··· ] | Ψ0 i
b (0)
Superoperator Hamiltonian H
b (0) Ôβ···
H ω
= [Ĥ (0) , Ôβ···
ω
]
Superoperator identity operator Ib

ω ω
Ib Ôβ··· = Ôβ···

Moment expansion of the polarization propagator

ω 1 (0) ω (0)
hh P̂ ; Ôβ··· iiω = hΨ0 | [P̂ , Ib Ôβ··· ] | Ψ0 i
~ω
2
1 (0)
h i
b Ôβ··· | Ψ(0) i
(0) ω
+ hΨ0 | P̂ , H 0
~ω
3
1 (0)
h
(0) 2 ω
i
(0)
+ hΨ0 | P̂ , (H ) Ôβ··· | Ψ0 i + · · ·
b
~ω
or
2
ω 1 ω 1 b (0) Ôβ···
ω
hh P̂ ; Ôβ··· iiω = (P̂ † | Ib Ôα··· ) + (P̂ † | H )
~ω ~ω
3
1 b (0) )2 Ôβ···
ω
+ (P̂ † | (H )
~ω
+ ···
1
This looks like the series expansion of 1−x

Superoperator resolvent defined by the series expansion

( ∞
!n )
−1 1 (0)

b (0)
X Hb
~ω Ib − H = Ib +
~ω n=1
~ω
Polarization propagator

ω † b (0) −1 ω
hh P̂ ; Ôβ··· iiω = P̂ | (~ω Ib − H ) Ôβ···
This is only a cosmetic change: superoperator resolvent is an inverse operator

We need a matrix representation of the superoperator resolvent
We need a complete set of basis vectors in "superoperator space":
a complete set of operators: {ĥn } consists of a complete set of excitation and
(0)
de-excitation operators with respect to the reference state |Ψ0 i.
(0)
ĥn |Ψ0 i = |Ψ(0)
n i
(0)
All other states |Ψn i of the system must be generated by operating on the
(0)
reference state |Ψ0 i.
Resolution of the superoperator identity using the complete set of

excitation and de-excitation operators {ĥn }, arranged either as column vector
ĥ or as row vector ĥT
Ib = | ĥT )(ĥ | ĥT )−1 (ĥ |
Matrix representation of the superoperator resolvent can be obtained via the
inner projection technique
Step 1: Insert the resolution of the superoperator identity twice in

ω † (0) −1 ω
hh P̂ ; Ôβ··· iiω = P̂ | (~ω Ib − H
b ) Ôβ···
which gives
ω
hh P̂ ; Ôβ··· iiω = (P̂ † | ĥT )(ĥ | ĥT )−1 (ĥ | (~ω I−
b Hb (0) )−1 | ĥT )(ĥ | ĥT )−1 (ĥ | Ôβ···
ω
)

b (0) )−1 | ĥT ).

Step 2: Find an alternative expression for (ĥ | (~ω Ib − H
We start with the definition of the superoperator resolvent, i.e.
b (0) )−1 (~ω Ib − H

(~ω Ib − H b (0) ) = Ib
Inserting this in the superoperator binary products between the complete set
of operators arranged as a matrix, (ĥ | ĥT ),
b (0) )−1 (~ω Ib − H

(ĥ | (~ω Ib − H b (0) ) | ĥT ) = (ĥ | ĥT )
Inserting the resolution of the identity
b (0) )−1 | ĥT )(ĥ | ĥT )−1 (ĥ | (~ω Ib − H

(ĥ | (~ω Ib − H b (0) ) | ĥT ) = (ĥ | ĥT )
b (0) ) | ĥT )−1 and

Multiplying this equation from the right first by (ĥ | (~ω Ib − H
then by (ĥ | ĥT ) we arrive at the desired alternative expression
b (0) )−1 | ĥT ) = (ĥ | ĥT )(ĥ | (~ω Ib − H

(ĥ | (~ω Ib − H b (0) ) | ĥT )−1 (ĥ | ĥT )

Exact matrix representation of the polarization propagator in the superoperator

formalism
ω
hh P̂ ; Ôβ··· iiω = (P̂ † | ĥT )(ĥ | (~ω Ib − H
b (0) ) | ĥT )−1 (ĥ | Ôβ···
ω
)
contains the inverse of matrix representations of the operators.

Inserting the definitions gives our final expression for a matrix representation
of the polarization propagator
ω
hh P̂ ; Ôβ··· iiω = T T (P̂ ) (~ωS − E)−1 T (Ôβ···
ω
)
where T T (P̂ ) and T (Ôβ···

ω
) are respectively row and column vectors and are
called property gradient vectors with elements
(0) (0)
T Ti (P̂ ) = hΨ0 | [P̂ , ĥi ] | Ψ0 i
and
ω (0) ω (0)
T j (Ôβ··· ) = hΨ0 | [ĥ†j , Ôβ··· ] | Ψ0 i

The combination (~ωS − E) is also called the principal propagator and

consists of the electronic Hessian matrix E with the elements
(0) (0)
E ij = hΨ0 | [ĥ†i , [Ĥ (0) , ĥj ]] | Ψ0 i
and the overlap matrix S with the elements

(0) (0)
S ij = hΨ0 | [ĥ†i , ĥj ] | Ψ0 i
The matrix representation is exact as long as

the set of operators {ĥn } is complete
(0)
|Ψ0 i is an eigenfunction of Ĥ (0)
The matrix representation involves no excited states of the system, but only
(0)
the ground state |Ψ0 i or in the general case the reference state.
Ideal starting point for approximate polarization propagator methods
(0)
by making an approximation to the exact wavefunction |Ψ0 i
by truncating the otherwise infinite set of operators {ĥn }.

The complete set of operators {ĥn } consists in general of excitation {e ĥn } and
de-excitation {d ĥn } operators.
The principal propagator matrix can therefore be divided in four blocks
   
ee ed ee ed
S S E E
M = ~ωS − E = ~ω   −  
de dd de dd
S S E E
and correspondingly the property: gradient vectors

T e T d T
T (P̂ ) = T (P̂ ) T (P̂ )
The spectral representation can be recovered choosing the set of operators

{ĥn } to be state transfer operators
n o n o
e d (0) (0) (0) (0)
{ĥn } = ĥn , ĥn = | Ψn ihΨ0 |, | Ψ0 ihΨn |
The two off-diagonal blocks ed E and de E of the electronic Hessian matrix and
ed
S and de S of the overlap matrix become zero
The propagator can be written as

ω e
hh P̂ ; Ôβ··· iiω = T T (P̂ ) (~ω ee S − ee E)−1 e T (Ôβ···
ω
)
−1
d T dd dd d ω
+ T (P̂ ) ~ω S − E T (Ôβ··· )
which leads to the spectral representation because

†

ee (0) e (0) (0)
E nm = hΨ0 | ĥn , [Ĥ (0) , e ĥm ] | Ψ0 i = δnm En(0) − E0
†

dd (0) (0) (0)
E nm = −hΨ0 | [Ĥ (0) , d ĥm ], d ĥn | Ψ0 i = δnm En(0) − E0
ee (0) † e (0)
S nm = hΨ0 | e ĥn ĥm | Ψ0 i = δnm
dd (0) † (0)
S nm = −hΨ0 | d ĥm d ĥn | Ψ0 i = −δnm
e (0) (0) (0)
T Tn (P̂ ) = hΨ0 | P̂ e ĥn | Ψ0 i = hΨ0 | P̂ | Ψ(0)
n i
d (0) (0) (0)
T Tn (P̂ ) = −hΨ0 | d ĥn P̂ | Ψ0 i = −hΨ(0)
n | P̂ | Ψ0 i
e ω (0) (0) (0)
T n (Ôβ··· ) = ω
hΨ0 | e ĥ†n Ôβ··· | Ψ0 i = hΨ(0) ω
n | Ôβ··· | Ψ0 i
d ω (0)
ω d † (0) ω (0)
T n (Ôβ··· ) = −hΨ0 | Ôβ··· ĥn | Ψ0 i = −hΨ0 | Ôβ··· | Ψ(0)
n i

3.13 Pseudo-Perturbation Theory - 1
(0) (0)
Electronic vertical excitation energies of the system (En − E0 ) are the poles
of the polarization propagator
In the matrix representation a polarization propagator has a pole, if the
principal propagator matrix (E − ~ωS) becomes singular.
This leads to the homogenous linear equations
(E − ~ωS) R = 0
and thus to the generalized eigenvalue problem
ERn = ~ωn SRn

(0) (0)
where ~ωn = (En − E0 ) is an eigenvalue of the electronic Hessian matrix
and equal to a vertical excitation energy
Rn is the corresponding eigenvector.

For an asymmetric Hessian matrix E we have a right eigenvector Rn and a

left eigenvector Ln
Ln E = Ln S~ωn
The eigenvectors are normally orthonormalized with the overlap matrix S as
metric
Lm SRn = δmn
For symmetric Hessian matrices the left and right eigenvectors are the same.
Often one knows or can easily obtain the eigenvalues and eigenvectors of an
approximation to the original eigenvalue problem
E (0) R(0) (0) (0) (0)

n = ~ωn S Rn
L(0)
n E (0)
= L(0) (0)
n S ~ωn
(0)

Pseudo-perturbation theory is applying the techniques of perturbation

theory to this eigenvalue problem.
The Hessian and overlap matrices are then partitioned into the zeroth-order
parts E (0) and S (0) and a remainder which is treated as first and second order
E = E (0) + E (1) + E (2) + · · ·

S = S (0) + S (1) + S (2) + · · ·
The eigenvalues and eigenvectors of E are then also expanded in a

perturbation series as
ωn = ωn(0) + ωn(1) + ωn(2) + · · ·

Rn = R(0) (1) (2)
n + Rn + Rn + · · ·
Ln = L(0) (1) (2)

n + Ln + Ln + · · ·

It will be convenient to choose the first-order matrices E (1) and S (1) in such a
way that their contribution to first-order is zero, i.e.
L(0)
n E
(1) (0)
Rn = 0
L(0)
n S
(1) (0)
Rn = 0
Inserting the expansions in the eigenvalue problem and separating orders

gives then first- and second-order equations

(1) (0) (0) (1) (1) (0) (0) (1)
E Rn + E Rn = ~ωn S + ~ωn S R(0) (0) (0) (1)
n + ~ωn S Rn
E (2) R(0)
n + E (1) R(1)
n +E
(0) (2)
Rn

(2) (0) (0) (2) (1) (1)
= ~ωn S + ~ωn S + ~ωn S R(0)
n

(1) (0) (0) (1)
+ ~ωn S + ~ωn S R(1)
n + ~ωn S
(0) (0) (2)
Rn

Projected against the zeroth-order left eigenvector leads to first- and

second-order corrections to the eigenvalues

(1) (0) (1) (0) (1)
~ωn = Ln E − ~ωn S R(0)
n = 0

(2) (0) (2) (0) (2) (0) (0) (1) (0) (1)
~ωn = Ln E − ~ωn S Rn + Ln E − ~ωn S R(1)
n
Rearranging the first-order equation give the first-order right eigenvector

−1
R(1)
n =− E (0)
− ~ωn(0) S (0) E (1)
− ~ωn(0) S (1) R(0)
n
and thus finally for the second-order correction to the eigenvalue

(2) (0) (2) (0) (2)
~ωn = Ln E − ~ωn S R(0)
n
−1
(0) (1) (0) (1) (0) (0) (0)
−Ln E − ~ωn S E − ~ωn S

(1) (0) (1)
× E − ~ωn S R(0)
n
Same structure as in regular time-independent perturbation theory.

Chapter 4
Stephan P. A. Sauer
sauer@kiku.dk


4. Definition of Electric Properties
So far we have derived

the Hamiltonian for the interaction of molecules with electromagnetic fields
perturbation theory expression for the corrections to the energy and
wavefunction of a molecule due to the interaction with fields
Now we can define many different time-independent molecular properties and
derive quantum mechanical expressions for them.
Some of the properties are well known from pre-quantum mechanical physics
and chemistry:
electric dipole moment of a distribution of positive and negative charges
magnetic dipole moment of rotating charges
For these properties we will start of with their classical definitions and then
translate them to quantum mechanical expressions.

4.1 Electric Multipole Expansion - 1
The electric charges in a molecule give rise to an electric field

~
The field can be represented by the associated electrostatic potential φρ (R).
Other molecules in the neighborhood will experience and react to this field
⇒ long-range intermolecular interactions.
~ for a continuous distribution of charges with
Electrostatic potential φρ (R)
charge density ρ(~r)
Z
~ = 1
φρ (R)
ρ(~r)
d~r
~ − ~r |
4πǫ0 ~r | R
An exact but not particularly useful expression

~
Integration necessary for each observation point R
Complete knowledge of the charge distribution ρ(~r) is required
~ which requires neither
We will derive an alternative expression for φρ (R)
separate integration for every observation point R~ nor complete knowledge of
the charge density ρ(~r).

Expansion of 1 ~ O inside of ρ(~r)

in a Taylor series around origin R
~
|R−~ r|
!
1 1 ∂ X 1
= + (rα − RO,α )
~ −~
|R r| ~ −R
|R ~O | ~ −~
∂rα | R r|
α ~ ~O
r =R
!
1 X ∂ 2 1
+ (rα − RO,α )(rβ − RO,β ) + · · ·
~ −~
2 αβ ∂rα ∂rβ | R r| ~ ~
r =RO
~ = ρ(~
r)
Inserted in the electrostatic potential φρ (R) 1
R
4πǫ0 ~ ~ r|
r |R−~
d~r
"
~ = 1 1
Z
φρ (R) r) d~
ρ(~ r
4πǫ0 ~ −R
|R ~O | ~
r
!
∂ 1
X Z
+ r) (rα − RO,α ) d~
ρ(~ r
~ −~
∂rα | R r| ~
r
α ~ ~O
r =R
!
1X ∂2 1
Z
+ r) (rα − RO,α )(rβ − RO,β ) d~
ρ(~ r
2 αβ ~ −~
∂rα ∂rβ | R r| ~
r
~ ~O
r =R

+ ...

xn f (x) dx are the n-th order moments of the function f (x).

R
The integrals
Moments of the charge distribution ρ(~r) or electric moments
Z
q = ρ(~r) d~r
~r
Z
~ O) =
µα ( R (rα − RO,α ) ρ(~r) d~r
Z~r
Qαβ (R~ O) = (rα − RO,α ) (rβ − RO,β ) ρ(~r) d~r
~
r
The zeroth-order electric moment q is the total charge

µα are components of the first-order or electric dipole moment
Qαβ are components of the second-order electric moment tensor
We have removed or at least hidden the integration over the charge distribution
in the Taylor expansion of the electrostatic potential

 !
~ρ 1  q X ∂ 1 ~ O)
φ (R) = + µα ( R
~ −R
4πǫ0 | R ~O | ~ − ~r |
∂rα | R
α ~ ~O
r =R
! 
1X ∂2 1 ~ O) + . . . 
+ Qαβ (R
2 ~ − ~r |
∂rα ∂rβ | R
αβ ~ ~O
r =R
Evaluation of the derivatives gives the multipole expansion of the

electrostatic potential
ρ ~ = 1 q 1 X ~ O ) Rα − RO,α
φ (R) + µα ( R
~ −R
4πǫ0 | R ~O | 4πǫ0 α ~ −R
|R ~ O |3
1 X 3 (R α − R O,α ) (R β − R O,β ) − δ αβ ( ~ −R
R ~ O )2
+ ~ O)
Qαβ (R + ...
8πǫ0 αβ ~ ~
| R − RO | 5
~ for any point R

We can calculate the electrostatic potential φρ (R) ~ if we know
the electric multipole moments of the corresponding charge distribution ρ(~r)

The first non-vanishing moment of a charge distribution is independent of the

~ O.
choice of the origin R
All the higher moments depend on this origin
The dipole moment of a neutral molecule or the quadrupole moment of a
~O
neutral and unpolar molecule are both independent of the origin R
Dipole molecule of an ion or the quadrupole moment of a neutral but polar
molecule depend on the origin R ~O
Definition of a traceless quadrupole moment tensor Θαβ with only five

independent elements is defined as
Z h
~ O) = 1 2
i
~ O ) ρ(~r) d~r
Θαβ (R 3(rα − RO,α ) (rβ − RO,β ) − δαβ (~r − R
2 ~r
It measures the deviation of the charge distribution ρ(~r) from spherical

symmetry.
Qαβ is called the second electric moment and Θαβ the electric quadrupole
moment.

4.2 Potential Energy in an Electric Field - 1
Electric multipole moments are important for the description of interactions

between molecules and external electric fields.
The potential energy E of a distribution of charges ρ(~r) immersed in an
external static electric field E~ with scalar potential φ E (~r)
Z
E(E,~ E) = ρ(~r) φ E (~r) d~r,
~
r
Exact expression but evaluation requires knowledge of charge density ρ(~r)

and the electric potential φ E (~r) for all ~r.
A more useful expression by expanding the scalar potential in a Taylor series
~O
around R
E

E E ~
X ∂φ (~r)
φ (~r) = φ (RO ) + (rα − RO,α )
α
∂rα ~r=R ~O
∂ 2 φ E (~r)

1X
+ (rα − RO,α )(rβ − RO,β ) + ...
2 ∂rα ∂rβ ~r=R ~O
αβ

The derivatives of φ E are the components of the electric field vector Eα

E

~ ∂φ (~r)
Eα ( R O ) = −
∂rα ~r=R ~O
and of the electric field gradient tensor Eαβ

2 E

~ ∂ φ (~r)
Eαβ (RO ) = −
∂rα ∂rβ ~ ~O
r =R
Potential energy becomes then

Z X Z
~ E) = φ E (R
E(E, ~ O ) ρ(~r) d~r − ~ O ) (rα − RO,α ) ρ(~r) d~r
Eα ( R
~
r α ~
r
Z
1X ~ O ) (rα − RO,α ) (rβ − RO,β ) ρ(~r) d~r + . . .
− Eαβ (R
2 ~r
αβ
The integrals over ρ(~r) are the electric moments again

The Interaction energy between a charge distribution and a static electric field
can be expressed in terms of the electric moments of the charge distribution
~ E) = q φ E (R
~ O) −
X
~ ~ 1X ~ O )Eαβ (R
~ O) + . . .
E(E, µα (RO )Eα (RO ) − Qαβ (R
α
2
αβ
Alternatively using the quadrupole moment tensor Θ
~ E) = q φ (R
~ O) −
E
X
~ O )Eα (R
~ O) − 1 X
~ O )Eαβ (R
~ O) + . . .
E(E, µα ( R Θαβ (R
α
3 αβ
The electric multipole moments of a charge distribution ρ(~r) can be used to

express
the electrostatic potential φρ due to this charge distribution ρ(~r)
the interaction energy E of the same charge distribution ρ(~r) with an
external scalar potential φE

Expression for the interaction energy
~ O) −
E
X
~ ~ 1X ~ O )Eαβ (R
~ O) + . . .
E = q φ (R µα (RO )Eα (RO ) − Qαβ (R
α
2
αβ
leads to alternative definitions of the electric moments
~ O) ∂E
µα ( R = −
~ O)
∂Eα (R
~ O) ∂E
Qαβ (R = −2
~ O)
∂Eαβ (R
~ O) ∂E
Θαβ (R = −3
~ O)
∂Eαβ (R
in addition to the definitions as moments of ρ(~r)

Z
~ O) =
µα ( R (rα − RO,α ) ρ(~r) d~r
Z~r
Qαβ (R ~ O) = (rα − RO,α ) (rβ − RO,β ) ρ(~r) d~r
~
r
4.3 Quantum Mechanical Expressions for ~
µ-1
Transition to Quantum Mechanics can use three ways

Electric moments as integrals over the charge density ρ(~r)
⇒ need quantum mechanical expression for the charge density
Electric moments as derivatives of the interaction energy with fields
⇒ need quantum mechanical expression for the energy
Hellmann-Feynman theorem: derivative of the energy is an expectation
value of the corresponding derivative of the Hamiltonian
⇒ need quantum mechanical property operator
Let’s start with the integrals over the charge density
In the Born-Oppenheimer approximation
the electrons form a continuous charge distribution ρel (~r)
the discrete nuclear charges are located at fixed points R ~K
Expression for the α-component of the electric dipole moment
Z X
~ el
µα (RO ) = (rα − RO,α ) ρ (~r) d~r + ZK e (RK,α − RO,α )
~
r K

µ-2
(0)
Quantum mechanical expression for the charge density in a state |Ψ0 i
N
(0) (0)
X
el
ρ (~
r) = −e hΨ0 | ri − ~
δ(~ r ) | Ψ0 i
i
Expression for the α-component of the electric dipole moment

Z N
(0) (0)
X
~O)
µα (R = (rα − RO,α ) (−e)hΨ0 | ri − ~
δ(~ r) | Ψ0 i d~
r
~
r i
X
+ ZK e (RK,α − RO,α )
K
Definition of the Dirac delta function δ(x − a) here generalized to 3 dimensions

Z
f (~
r ) δ(~a − ~
r) d~
r = f (~a)
~
r
The quantum mechanical expression for the electric dipole moment is

N
(0) (0)
X X
~ O) =
µα (R hΨ0 | −e (ri,α − RO,α ) | Ψ0 i + ZK e (RK,α − RO,α )
i K

µ-3
Analogeously for the components of the second electric moment tensor Qαβ
and the quadrupole moment tensor Θαβ
N
(0) (0)
X
~O)
Qαβ (R = hΨ0 | −e (ri,α − RO,α )(ri,β − RO,β ) | Ψ0 i
i
X
+ ZK e (RK,α − RO,α )(RK,β − RO,β )
K
N
~O) 1 (0) X
~ O )2 | Ψ(0) i
Θαβ (R = hΨ0 | −e 3(ri,α − RO,α ) (ri,β − RO,β ) − δαβ (~
ri − R 0
2 i
1X h
~K − R~O) 2
i
+ ZK e 3(RK,α − RO,α ) (RK,β − RO,β ) − δαβ (R
2 K

µ-4
2nd approach: moments as derivatives of the energy of a molecule in the

presence of an electric field
We need an expression for the energy of a molecule in the presence of an
inhomogeneous electric field
⇒ Perturbation theory
(1) (0)
~ = hΨ | Ĥ (1) | Ψ i (0)
E0 (E) 0 0
We need the perturbing Hamiltonian operator Ĥ (1) = qr φ̂E (~r) for the
P
interaction with an electrostatic potential due to an electric field
~ O ) · E(R
φ̂E (~r) = − (~r − R ~ O)

1X 1 ~ O )2 Eαβ (R
~ O)
− (rα − RO,α )(rβ − RO,β ) − δαβ (~r − R
2 αβ 3
First-order perturbing Hamiltonian

(1)
Xh E i Xh ∇E i
Ĥ = − Ôα (R E ~ O ) Eα ( R
~ O ) + Ω̂α (R ~ O )− Ôαβ (R ∇E ~ O ) Eαβ (R
~ O ) + Ω̂αβ (R ~ O)
α αβ

µ-5
Perturbation operators for electrons

N
X
E ~ ~ O)
Ôα (RO ) = −e (ri,α − RO,α ) ≡ µ̂α (R
i
N
∇E ~ e X 1 ~ O )2
Ôαβ (R O ) = − (ri,α − RO,α )(ri,β − RO,β ) − δαβ (~ri − R
2 i 3
1 ~ O)
≡ Θ̂αβ (R
3
Perturbing operators for nuclear charges
E ~
X
Ω̂α (RO ) = ZK e(RK,α − RO,α )
K

~ 1 X 1 ~ O )2
~K − R
Ω̂∇E
αβ ( R O ) = ZK e (R K,α − R O,α )(R K,β − R O,β ) − δαβ (R
2 K 3

µ-6
Moments are first derivatives :

we only need to consider the first-order energy correction
(1)
~
∂E0 (E) ∂ (0) (0)
~ O)
µα ( R = − =− hΨ0 | Ĥ (1) | Ψ0 i
~ O)
∂Eα (R ~ O)
∂Eα (R
(1)
~
∂E0 (E) ∂ (0) (0)
~ O)
Θαβ (R = −3 = −3 hΨ0 | Ĥ (1) | Ψ0 i
~ O)
∂Eαβ (R ~ O)
∂Eαβ (R
Alternatively we can use the Hellmann-Feynman theorem:

we obtain the moments as expectation values of the derivatives of the total
Hamiltonian
~ O) (0) ∂ Ĥ (0)
µα ( R = hΨ0 | − | Ψ0 i
~ O)
∂Eα (R
~ O) (0) ∂ Ĥ (0)
Θαβ (R = 3 hΨ0 | − | Ψ0 i
∂Eαβ (R~ O)

µ-7
However, the first derivatives of the first-order perturbation Hamiltonian and of

the full molecular electronic Hamiltonian are the same
∂ Ĥ ∂ Ĥ (1) ~ O)
~ O ) − Ω̂Eα (R
= = −µ̂α (R
~ O)
∂Eα (R ∂Eα (R~ O)
∂ Ĥ ∂ Ĥ (1) 1 ~ O ) − Ω̂∇E ~
= = − Θ̂αβ (R αβ (RO )
∂Eαβ (R~ O) ∂Eαβ (R ~ O) 3
Quantum-mechanical expressions for the electric moments
~ O) (0) ~ O ) | Ψ(0) i + Ω̂Eα (R

~ O)
µα ( R = hΨ0 | µ̂α (R 0
~ O) (0) ~ O ) | Ψ(0) i + 3 Ω̂∇E ~
Θαβ (R = hΨ0 | Θ̂αβ (R 0 αβ (RO )
Same results as from the expression as integrals over the charge density ρ(~r)
This shows that the different definitions are indeed equivalent as along as we
know the exact solutions to the unperturbed Schrödinger equation.
This will no longer be the case for approximate wavefunctions

4.4 Induced Electric Moments and Polarizabilities - 1
A non-fixed charge distribution will redistribute itself in the presence of the

external electric field in such a way that the total energy is minimized.
The electric moments of the charge distribution will change and their value will
depend on the strength of the field.
Field dependent moments µind ~
α (E) are induced by the external field E
~ = µα + µind
µα (E) ~
α (E)
Traditional expansion of the field dependence of the electric moments in

powers of the field strength
~ = µα +
X 1X 1X
µα (E) ααβ Eβ + βαβγ Eβ Eγ + γαβγδ Eβ Eγ Eδ + . . .
β
2 βγ 6 βγδ
Polarizabilities and hyperpolarizabilities as response functions

Polarizabilities and hyperpolarizabilities as first and higher derivatives of the
field dependent moments
Question: What is the energy?
Energy in the presence of the external field:

infinitesimal change in energy dE due to an infinitesimal change in the fields
X
dE = − ~ dEα = −~
µα (E) ~ · dE~
µ(E)
α
The energy can now obtained by integration

Z ~
E Z Ex Z Ey Z Ez
~ −E (0) ~′ ~′
E(E) =− ~ (E ) · dE = −
µ ~ (E~′ ) · dEz′ · dEy′ · dEx′
µ
0 0 0 0
 
~
E
XZ X 1X 1X
= − µα + ααβ Eβ′ + βαβγ Eβ′ E ′
γ + γαβγδ Eβ′ Eγ′ Eδ′ + . . .  dEα
′
α 0 β
2 βγ 6 βγδ
Line integral over vector field ~ µ(E) in the space defined by the components of
the electric field
R ~r
A line integral 0 A(~~ r′ ) · d~r′ is independent of the path when the vector function
~ r) is the gradient of a scalar single-valued field with continuous derivatives
A(~

Integration in three steps: from E~′ = (0, 0, 0) to (Ex , 0, 0), from (Ex , 0, 0) to
(Ex , Ey , 0) and finally to (Ex , Ey , Ez ).
~ = E (0)
E(E)
X 1X 1X 1 X
− µ α Eα − ααβ Eα Eβ − βαβγ Eα Eβ Eγ − γαβγδ Eα Eβ Eγ Eδ
α
2 6 24
αβ αβγ αβγδ
One can calculate the change in energy of a charge distribution due to an

external electric field of arbitrary strength, if one knows the various
polarizabilities and hyperpolarizabilities.
The induction energy of intermolecular interactions is determined by the
polarizabilities of the molecules
We can define the polarizabilities and hyperpolarizabilities as derivatives of the
energy

Table 1: Definitions of tensor components of the electric polarizabilities and

hyperpolarizabilities as derivatives of components of the field dependent
~ or of the energy E(E)
electric dipole µα (E) ~
~
µα (E) ~
E(E)
∂ ∂2
ααβ −
∂Eβ ∂Eβ ∂Eα
∂2 ∂3
βαβγ −
∂Eγ ∂Eβ ∂Eγ ∂Eβ ∂Eα
∂3 ∂4
γαβγδ −
∂Eδ ∂Eγ ∂Eβ ∂Eδ ∂Eγ ∂Eβ ∂Eα

Isotropic averages and anisotropy of dipole polarizability

sP
1X αβ (3ααβ ααβ − ααα αββ )
α= ααα ∆α =
3 α 2
Similarly one defines two isotropic averages for the first hyperpolarizability
1X
βk = (βzαα + βαzα + βααz )
5 α
1X
β⊥ = (2βzαα − 3βαzα + 2βααz )
5 α
where the molecular z-axis is parallel to µ

~
Two isotropic averages for the second hyperpolarizability
1 X
γ = γk = (γααββ + γαβαβ + γαββα )
15
αβ
1 X
γ⊥ = (2γαββα − γααββ )
15 αβ
4.5 Quantum Mechanical Expressions for α - 1
We have define the electric dipole polarizability ααβ

~ in the presence of a field
as derivative of the energy E(E)
~
as derivative of the perturbation dependent electric dipole µα (E)
Let’s start with the definition as derivatives of the energy
We use the perturbation theory expression for the perturbed energy and
differentiate it twice with respect to the appropriate components of the field

2
2 (2)
∂ E0 ∂ E0
ααβ = − =−
∂Eα ∂Eβ |E|=0 ∂E ∂E

~ α β
~
|E|=0

2 (0) (1) (0) (0) (1) (0)
∂ X hΨ0 | Ĥ | Ψn ihΨn | Ĥ | Ψ0 i
= − (0) (0)
∂Eα ∂Eβ E0 − En

n6=0 ~
|E|=0

The derivative of the first-order perturbation Hamiltonian, Ĥ (1) , with respect to

a component of the electric field Eα is the electric dipole operator
∂ Ĥ (1) ~ O)
~ O ) + Ω̂Eα (R
= µ̂α (R
∂Eα
The electric dipole polarizability is then
X hΨ(0) | µ̂α | Ψ(0) ihΨ

(0)
| | Ψ
(0)
0 n n µ̂ β 0 i
ααβ = − (0) (0)
n6=0 E0 − En
X hΨ(0) | µ̂β | Ψ(0) (0) (0)
n ihΨn | µ̂α | Ψ0 i
0
− (0) (0)
n6=0 E0 − En
Question: Where is the nuclear contribution?

Question: Does it depend on the origin?

~ O)
There is no contribution from the nuclear operators Ω̂Eα (R
(0) ~ O ) | Ψ(0) E ~ (0) (0)
hΨ0 | Ω̂Eα (R n i = Ω̂α (RO )hΨ0 | Ψn i = 0
~ O.
The polarizability is independent of the origin R
N
(0) (0) (0)
X
~ O ) | Ψ(0)
hΨ0 | µ̂α (R n i = hΨ0 | −e ri,α | Ψ(0) (0)
n i + hΨ0 | eRO,α | Ψn i
i
N
(0) (0)
X
= hΨ0 | −e ri,α | Ψ(0) (0)
n i + eRO,α hΨ0 | Ψn i
i
N
(0)
X
= hΨ0 | −e ri,α | Ψ(0)
n i
i
~ O ) nor R
Neither Ω̂Eα (R ~ O act on the electronic wavefunctions and the
(0) (0)
unperturbed states are orthogonal hΨ0 | Ψn i = 0.

2n d approach: first derivative of the field-dependent electric dipole moment

Perturbation-dependent dipole moment
~ = hΨ0 (E)
µα (E) ~ | µ̂α | Ψ0 (E)i
~
Response theory expression using expansion in response functions

∂ ~ ~

ααβ = hΨ0 (E) | µ̂α | Ψ0 (E)i = −hh µ̂α ; µ̂β iiω=0
∂Eβ ~
|E|=0
~
Response theory expression using perturbation theory expansion of Ψ0 (E)

(1) ~ (1) ~
(0) ∂Ψ0 (E) ∂Ψ0 (E)
(0)
ααβ = hΨ0 | µ̂α | i +h | µ̂α | Ψ0 i
∂Eβ ~ ∂Eβ ~
|E|=0 |E|=0
Perturbation theory expression for the first-order correction to the wavefunction

(0) (1) ~ (0)
(1) ~ P (0) hΨn |Ĥ (E)|Ψ0 i
| Ψ (E)i =
0 | Ψn i
n6=0 (0) (0)
E0 −En
gives the same expression as before

4.6 Molecular Electric Fields and Field Gradients - 1
The electric fields arising from a distribution of charges are also important,
~ of a distribution of charges
Derivatives of the electrostatic potential φρ (R)
ρ
molecular electric field Eαρ and field gradient Eαβ
Z
1 R α − rα
Eαρ (R)
~ = ρ(~r) d~r
4πǫ0 ~r ~
| R − ~r |3
Z " #
ρ ~ = 1 δαβ (Rα − rα )(Rβ − rβ )
Eαβ (R) ρ(~r) −3 d~r
4πǫ0 ~
r ~ − ~r |3
|R |R~ − ~r |5
~ acting on the charges in the

The molecular electric field gives rise to a force F
charge distribution.
The force acting on a nucleus K in a molecule is
~K = ZK e E~ρ (R
F ~K)
This force will be zero if the molecule is in its equilibrium geometry.

~ or E ρ (R)
Question: Can Eαρ (R) ~ be measured?
αβ

Fields can be probed by dipole and field gradients by quadrupole moments

We cannot bring dipole moments into a molecule, but nuclei with a spin
quantum number I K > 1 possess an electric quadrupole moment ΘK
ρ ~ K ),
The molecular electric field gradient at the positions of the nuclei, Eαβ (R
can be studied via the quadrupole coupling with the nuclear electric
quadrupole moment defined via an effective spin Hamiltonian
1 X ΘK X K ρ
Ĥ spin
= Iˆα Eαβ (R
~ K )IˆβK
2~ K IK (2IK − 1)
αβ
nuclear quadrupole coupling constant

ρ ~
ΘK Ezz (RK )/~
can be obtained from

the hyperfine structure of rotational spectra
the quadrupole the splitting of the lines in Mössbauer spectra
the linewidth of the lines in NMR spectra
Asymmetry parameter ηK defined as

ρ ~ ρ ~
Eaa (RK ) − Ebb (R K )
ηK = ρ ~
Ecc (R K )
ρ ~ ρ ~ ρ ~
where Eaa (RK ), Ebb (RK ) and Ecc (RK ) are the three eigenvalues of the electric
field gradient tensor with
ρ ~ ρ ~ ρ ~
|Ecc (RK )| ≥ |Ebb (RK )| ≥ |Eaa (RK )|
The expression for the electric field gradient is analog to the electric moments
N
" #
ρ e (0)
X (ri,α − RK,α )(ri,β − RK,β ) δαβ (0)
~
Eαβ (RK ) = hΨ0 | 3 − | Ψ0 i
4πǫ0 ~ K |5
| ~ri − R ~ K |3
| ~ri − R
i
" #
1 X (RL,α − RK,α )(RL,β − RK,β ) δαβ
− ZL e 3 −
4πǫ0 |R ~L − R ~ K |5 |R~L − R~ K |3
L6=K

Θ Θ
Defining electric field gradient operators, Ô and Ω̂ ,
N
X
Θ ~
Ôαβ (R) = ôΘ ~
i,αβ (R)
i
N
" #
e X (ri,α − RK,α )(ri,β − RK,β ) δαβ
= 3 −
4πǫ0 ~ K |5
| ~ri − R ~ K |3
| ~ri − R
i
" #
1 X (RL,α − RK,α )(RL,β − RK,β ) δαβ
Ω̂Θ ~
αβ (R) = − ZL e 3 −
4πǫ0 L6=K ~L − R
|R ~ K |5 |R~L − R~ K |3
we can write the quantum mechanical expression for the electric field gradient
~ more simply as
at an arbitrary point R
ρ (0) Θ ~ (0)
Eαβ ~K)
(R = hΨ0 | Ôαβ (R) | Ψ0 i + Ω̂Θ ~
αβ (R)

Chapter 5
Stephan P. A. Sauer
sauer@kiku.dk


5. Definition of Magnetic Properties
So far
We have defined electric properties as derivatives of the energy of a charge
distribution ρ(~r) in the presence of an electric field or field gradient
Some properties could also be defined as derivatives of the electric
moments.
We have derive quantum mechanical expressions for all these properties
using perturbation theory or static response theory
In the present chapter
We will define magnetic properties and derive quantum mechanical
expression for them quite analogously
However, there are some important differences.
there will be more types of properties to be studied, because in addition
to an external magnetic field we are also interested in the interaction with
nuclear magnetic dipole moments.
magnetic properties exhibit a greater complexity than electric properties.

5.1 Magnetic Multipole Expansion - 1
A dynamic system of charges gives rise to a current density
~j(~r) = ρ(~r) ~v (~r)
where ~v (~r) is the velocity distribution and ρ(~r) the charge density
Vector potential due to this current density
Z ~j(~r)
A ~ = µ0
~ j (R) d~r
4π ~
r ~
| R − ~r |
analogous to the scalar potential of a charge distribution [4-vectors]

Using the Taylor expansion of 1 ~ GO
around an origin R
~
|R−~ r|
Z
µ0 1
Ajα (R)
~ = jα (~r) d~r
~ −R
4π | R ~ GO | ~r
Z
µ0 X Rβ − RGO,β
+ jα (~r) (rβ − RGO,β ) d~r + . . .
4π ~ ~ 3
| R − RGO | ~r
β
Electric quadrupole approximation: We neglect higher terms

From the divergence theorem of vector calculus one can derive

Z h i
~ (~r) · ~j(~r) d~r = 0
∇f
~
r
~ · ~j = 0
for an arbitrary function f (~r) and a steady current distribution ∇
Choosing f = rα gives Z
jα (~r) d~r = 0
~
r
Magnetic monopole moments do not exist.

Using now f = (rα − RGO,α )(rβ − RGO,β ) shows that
Z
[(rβ − RGO,β ) jα (~r) + (rα − RGO,α ) jβ (~r)] d~r = 0
~
r
This is the symmetric part of the dipole term

Z
µ 0
X R β − R GO,β
Ajα (R)
~ = (r − RGO,β ) jα (~r) d~r + . . .
4π β | R ~ −R ~ GO |3 ~r β

Only the asymmetric part of the dipole term is non-zero

Z h ~ −R~ GO )
~ = µ0 1 (R
i
~ j (R)
A ~ ~
(~r − RGO ) × j(~r) × d~r + . . .
4π 2 ~
r ~ ~
| R − RGO | 3
or
j ~ µ0 ~ −R
(R ~ GO )
~
A (R) = ~ ×
m + ... .
4π ~ ~
| R − RGO | 3
where the magnetic dipole moment is defined as

Z Z
1 ~ GO ) × ~j(~r) d~r = 1 ~ GO ) × ~v (~r) d~r
m~ = (~r − R ρ(~r)(~r − R
2 ~r 2 ~r
~ GO as a
The magnetic dipole moments are independent of the gauge-origin R
direct consequence of the absence of magnetic monopole moments.
Closed shell molecules do not posses a permanent magnetic moment.
Open shell molecules have permanent magnetic moments ⇒ paramagnetism
Nuclei with non zero spin have a magnetic moment (NMR, ESR)

5.2 Potential Energy in an Magnetic Induction - 1
The potential energy of a current distribution in an external magnetic induction

Z
E(B)~ = − ~j(~r) · A ~ B (~r) d~r
~
r
~ B (~r) is the vector potential of the magnetic induction B

where A ~
~ r) = ∇
B(~ ~ ×A
~ B (~r)
Expanding the vector potential in a Taylor series around the gauge origin R~ GO
B

B ~
X ∂Aα (~r)
ABα (~
r ) = Aα ( R GO ) + (r β − R GO,β ) + ...
β
∂r β ~ ~ GO
r =R
leads to
X Z
~ = −
E(B) B ~
α Aα (RGO ) jα (~r) d~r
~
r
X ∂AB
Z
α (~
r)
− (rβ − RGO,β )jα (~r) d~r + . . .
∂rβ ~ ~ GO
r =R ~
r
αβ

5.2 Potential Energy in an Magnetic Induction - 2
Monopole term and symmetric part of dipole term vanish

X ∂AB α (~r)

1
Z
~ = −
E(B) [(rβ − RGO,β ) jα (~r) − (rα − RGO,α ) jβ (~r)] d~r
αβ
∂r β ~ ~
r =RGO 2 ~
r
+ ...
or in vector notation
Z h
~ =− 1 i h
~ GO ) × ~j(~r) · ∇~ ×A
~ (~r)
B
i
E(B) (~r − R d~r + . . .
2 ~r ~ ~ GO
r =R
Using the definition of the magnetic dipole moment
~ = −m
E(B) ~ R
~ · B( ~ GO ) + . . .
The magnetic dipole moment can also be defined as the derivative of the
potential energy with respect to the field induction Bα
∂E(B)~
mα = −
~ GO )
∂Bα (R

5.3 Quantum Mechanical Expression for m
~ -1
The vector potential at the position of electron i for a static and homogenous
magnetic induction B~
~ˆB (~ri ) = 1 B
A ~ × (~ri − R
~ GO )
2
With this this vector potential we obtain the first-order perturbing Hamiltonian
(1)
X h lB sB
i
Ĥ = − Ôα (R~ GO ) + Ôα Bα
α
where the perturbation operators are defined as

N
X
lB ~
Ôα (RGO ) = ôlB ~ l ~
i,α (RGO ) ≡ m̂α (RGO )
i
N N h
e ~ GO ) = − e X
~ GO ) = − e X
ˆ
~ GO ) × p
i
= − L̂α (R l̂i,α (R (~ri − R ~i
2me 2me i 2me i α
N N
sB
X ge e ge e X
Ôα = ôsB
i,α =− Ŝα = − ŝi,α
i
2me 2me i
~ -2
The perturbation operators are again the first derivatives of the molecular
electronic Hamiltonian

∂ Ĥ ∂ Ĥ (1)

lB sB
= = −Ôα − Ôα
∂Bα ~ ∂Bα

|B|=0
Derivation of magnetic dipole moment as first derivative of the perturbed

energy
(0) lB sB (0)
mα = hΨ0 | Ôα + Ôα | Ψ0 i
=
(0) ~ GO ) | Ψ(0) i − ge e hΨ(0) | Ŝα | Ψ(0) i
hΨ0 | m̂lα (R 0 0 0
2me
e (0) ~ GO ) + ge Ŝα | Ψ(0) i
= − hΨ0 | L̂α (R 0
2me
A derivation starting from the classical definition of the magnetic dipole

moment as an integral over the current density gives only the orbital magnetic
moment.
~ = 0 for closed shell molecules?
Question: Why is m
~ -3
Proof that closed shell molecules have not permanent magnetic moment
The angular momentum operator is a hermitian operator
~ˆ | Ψi = hΨ | L
hΨ | L ~ˆ | Ψi∗
For real wavefunctions Ψ the right hand side becomes
~ˆ | Ψi = hΨ | L
hΨ | L ~ˆ ∗ | Ψi
~ˆ is a purely imaginary operator

L
~ˆ | Ψi = −hΨ | L
hΨ | L ~ˆ | Ψi
~ˆ is zero for real wavefunctions

⇒ the expectation value of L
⇒ the orbital angular momentum is quenched
For molecules in orbitally non-degenerate states we can always choose the
wavefunctions to be real and therefore they do not have a permanent orbital
magnetic moment
~ -4
(0) (0)
For singlet states the expectation value hΨ0 | Ŝ | Ψ0 i vanishes
⇒ the molecule has neither spin nor orbital permanent magnetic moment.
Question: Are there nuclear contributions to the magnetic moment?

~ -4
(0) (0)
For singlet states the expectation value hΨ0 | Ŝ | Ψ0 i vanishes
⇒ the molecule has neither spin nor orbital permanent magnetic moment.
Question: Are there nuclear contributions to the magnetic moment?
No nuclear contributions to the magnetic moment

Nuclear spin magnetic moments are much smaller
⇒ we will consider them as perturbations of the electronic structure
Nuclear orbital magnetic moments require an angular motion of the nuclei
and thus a rotation of the nuclear framework of the molecule.
For a fixed nuclear geometry we are missing this contribution
We will consider this contribution and other couplings with molecular
rotation later

5.4 Induced Magnetic Moment and Magnetizability - 1
A non-rigid distribution of moving charges will redistribute itself in the presence

of the external magnetic field in such a way that the total energy is minimized
The magnetic moments of the charge distribution will change
~ ind are induced by the external field B
Field dependent moments m
Another important source of magnetic induction in molecules is the magnetic
dipole moments m~ K of nuclei K
~ K })
~ {m
mα (B, = mα + mind ~ ~ K })
α (B, {m
X XX K K
~ K })
~ {m
mα (B, = mα + ξαβ Bβ − σβα mβ + . . .
β K β
ξαβ is the dipole magnetizability:

the linear response of m
~ to an external magnetic induction
K
σβα is the nuclear magnetic shielding tensor:
the linear response of m
~ to the magnetic dipole moment of a nucleus K
K
σβα is the molecular property behind the chemical shift of NMR

~
Energy of a polarizable current distribution in an magnetic induction B
Infinitesimal change in the energy dE
X
dE = − ~ dBα = −m(
mα (B) ~ · dB
~ B) ~
α
Inserting the expansion of the magnetic dipole moment and integration

 
X BZ ~ X
~ (0)
E(B) − E = −  mα + ′
ξαβ Bβ′ + . . .  dBα
α 0 β
Line integration as in the electric case
~ = E (0) −
X 1X
E(B) mα Bα − ξαβ Bα Bβ + . . .
α
2 αβ
Including the contribution from the nuclear magnetic moment mK

X 1X XX K K
~ K }) = E (0) −
~ {m
E(B, mα Bα − ξαβ Bα Bβ + σβα mβ Bα + . . .
α
2 αβ K αβ
The magnetizability and nuclear magnetic shielding tensor of a nucleus K as

first derivatives of the electronic magnetic dipole moment

~
∂mα (B)
ξαβ =
∂Bβ ~

|B|=0

K
K ∂mα (m ~ )
σβα = −
∂mK

β
K
|m
~ |=0
The magnetizability and nuclear magnetic shielding tensor of a nucleus K as

second derivatives of the energy

2 ~
∂ E(B)
ξαβ = −
∂Bα ∂Bβ ~

|B|=0

K
2 ~
∂ E(B, m K
~ )
σαβ =
∂Bβ ∂mK

α ~ K
|B|=|m
~ |=0

5.5 Quantum Mechanical Expression for ξ - 1
Magnetizability defined as derivative of the energy

(2)

2
∂ 2 E0

∂ E0
ξαβ = − = −
∂Bα ∂Bβ |B|=0
~ ∂Bα ∂Bβ ~
|B|=0
 
(0) (1) | Ψ (0) (0) (1) | Ψ (0)
∂2

hΨ | Ĥ n ihΨ n | Ĥ i
hΨ(0) | Ĥ (2) | Ψ(0) i −
X
0 0

= − 0 0 (0) (0)

∂Bα ∂Bβ n6=k E − En

0 ~
|B|=0
~ˆ contributes both to Ĥ (1) and to Ĥ (2)

A vector potential A
Therefore the full expression for the second order energy correction
Second-order perturbing Hamiltonian Ĥ (2) for an external magnetic induction
(2)
X BB
Ĥ = Ôαβ (R~ GO ) Bα Bβ
αβ
where the perturbation operator is defined as
BB ~ e2 X h 2
~ GO ) δαβ − (ri,α − RGO,α )(ri,β − RGO,β )
i
Ôαβ (RGO ) = (~ri − R
8me i
sum-over-states expression for the components of the magnetizability tensor
(0) BB ~ BB ~ (0)
ξαβ = − hΨ0 | Ôαβ (RGO ) + Ôβα (RGO ) | Ψ0 i
X hΨ(0)
0 | m̂l ~
α ( R GO ) | Ψ
(0)
n ihΨ
(0)
n | m̂l ~
β ( R GO ) | Ψ
(0)
0 i
− (0) (0)
n6=0 E0 − En
X hΨ(0)
0 | m̂l ~
β ( R GO ) | Ψ
(0)
n ihΨ
(0)
n | m̂l ~
α ( R GO ) | Ψ
(0)
0 i
− (0) (0)
n6=0 E0 − En
~ : hΨ(0) | Ŝα | Ψ(0)

No contribution from the electron spin S n i = 0
0

2nd approach: magnetizability as first derivative of the magnetic

field-dependent moment using the magnetic dipole moment operator in the
presence of a magnetic induction
∂ Ĥ X h BB i
~ˆ (R
=m l ~ GO , B) l ~ GO ) −
~ = m̂α (R Ôαβ (R BB ~ GO ) Bβ
~ GO ) + Ôβα (R
∂Bα
β
leads to

~

∂mα (B) ∂

~ l ~ ~ ~

ξαβ = = hΨ0 (B) | m̂α (RGO , B) | Ψ0 (B)i
∂Bβ ~ ∂Bβ

~
|B|=0
|B|=0

~
∂Ψ0 (B)
l ~ (0) (0) l ~
~
∂Ψ0 (B)
= h | m̂α (RGO ) | Ψ0 i + hΨ0 | m̂α (RGO ) | i
∂Bβ ∂Bβ

~
~
|B|=0 |B|=0
h i
(0) BB ~ BB ~ (0)
− hΨ0 | Ôαβ (RGO ) + Ôβα (RGO ) | Ψ0 i
Insertion of the perturbation theory expression for the first-order correction to

the wavefunction gives the same final expression

Magnetizability as linear response function or polarization propagator

∂ ~ l ~ ~ ~

ξαβ = hΨ0 (B) | m̂α (RGO , B) | Ψ0 (B)i
∂Bβ ~
|B|=0
h i
(0) BB ~ BB ~ (0)
= −hΨ0 | Ôαβ (RGO ) + Ôβα (RGO ) | Ψ0 i
~ GO ) ; m̂lβ (R
−hh m̂lα (R ~ GO ) iiω=0
Insertion of the spectral representation of the polarization propagator gives the

same final expression

There are two different contributions to the magnetizability

dia ~ para ~
ξαβ = ξαβ (RGO ) + ξαβ (RGO )
dia
The diamagnetic contribution ξαβ is an expectation value with the
(0)
unperturbed wavefunction Ψ0 of the system. It is negative.
para
The paramagnetic contribution ξαβ involves a sum over all the other
(0)
unperturbed states {Ψn }. This contribution is positive.
dia ~ para ~
For most molecules ξαβ (RGO ) > ξαβ (RGO ) : the molecule is diamagnetic
dia ~ para ~
In a few closed shell molecules ξαβ (RGO ) < ξαβ (RGO ) : the molecule is
paramagnetic (TIP)
the dia- and paramagnetic contributions both depend quadratically on
~ GO but their sum, the magnetizability, is independent
the gauge origin R
The separation in a dia- and paramagnetic contribution is arbitrary and that no
physical meaning should be assigned to the two terms individually

5.6 Molecular Magnetic Fields and ESR Parameters - 1
~j (R)
The current density ~j(~r) gives rise to a molecular magnetic induction B ~
~j (R)
B ~ = ~ ×A
∇ ~ j (R)
~
Z ~
µ0 (R − ~r) × ~j(~r)
= − d~r
4π ~r | R ~ − ~r |3
µ0
Z ~ − ~r) × ~v (~r)
(R
= − ρ(~r) d~r .
4π ~r ~
| R − ~r |3
Only the value at the position of the nuclei can be probed experimentally
~K
Interaction energy with a nuclear magnetic dipole moment m
K
X K j
~ )=−
E(m mα Bα (R~K)
α
The molecular magnetic induction at the position of some nucleus K can be

defined as derivative
j ~ ∂E(m ~ K)
Bα (RK ) = −
∂mK α

~j,s (R
Experimentally interesting is the spin molecular magnetic induction B ~K)
at the position of a nucleus K with spin IˆK and nuclear magnetic moment
gK µN ~K
~K =
m I
~
It gives rise to the hyperfine coupling tensor aK
αβ of electron spin
resonance (ESR) spectroscopy of radicals, defined in terms of an effective
spin Hamiltonian
spin 2π X ˆK K
Ĥ = Iα aαβ Ŝβ
~
αβ
An expectation value of this operator for a particular spin state is the energy
X K j
E=− mα Bα (R ~K)
α
leading to an expression for the hyperfine coupling tensor

j,s ~
gK µN Bα (R K )
aK
αβ =−
2π hΨ(0) (0)
0 | Ŝβ | Ψ0 i
(0) (1) (0)
The molecular magnetic induction can be expressed as hΨ0 | ∂∂m
Ĥ
K | Ψ0 i
α
~K
Using the vector potential of a nuclear magnetic moment m
K µ0 K ~K)
(~ri − R
~
A (~ri ) = ~ ×
m
4π ~ K |3
| ~ri − R
The first-order perturbing Hamiltonian becomes

(1)
X l mK s mK

K gK µN X l mK s mK

Ĥ = − Ôα + Ôα mα = − Ôα + Ôα IαK
α
~ α
with the orbital paramagnetic or

paramagnetic (nuclear) spin (electron) orbit operator
N
lmK K
X
Ôα = ôlm OP
i,α ≡ ÔK,α
i
N N
!
e µ0 ~K)
l̂i,α (R e µ0 ~K
~ri − R
~î
X X
= − =− ×p
me 4π | ~ri − R ~ K |3 me 4π | ~ri − R~ K |3
i i α

and the Fermi-contact and spin-dipolar operators

smK
X smK FC SD
Ôα = ôi,α ≡ ÔK,α + ÔK,α
i
where
N
FC ge eµ0 X ~ K )ŝi,α
ÔK,α =− δ(~ri − R
3me i
 h i 
ge e µ0 X sî · (~ri − R
N 3 ~ ~ K ) (ri,α − RK,α )
ŝi,α

SD
ÔK,α =− −
2me 4π  ~ K |5
| ~ri − R | ~ri − R ~ K |3 
i

The permanent molecular magnetic induction at the position of nucleus K

j ~ j,l ~ j,s ~
Bα (R K ) = Bα (RK ) + Bα (R K )
(0) OP (0) (0)
FC SD (0)
= hΨ0 | ÔK,α | Ψ0 i + hΨ0 | ÔK,α + ÔK,α | Ψ0 i
Only the spin contribution is related to the ESR hyperfine coupling tensor
j,s ~ (0) FC SD (0)
gK µN Bα (RK ) gK µN hΨ0 | ÔK,α + ÔK,α | Ψ0 i
aK
αβ = − =−
2π hΨ(0) | Ŝβ | Ψ(0) i 2π (0) (0)
hΨ0 | Ŝβ | Ψ0 i
0 0
N
µ0 ege gK µN (0)
X
~ K )ŝi,α | Ψ(0) i
= hΨ 0 | δ(~
r i − R 0
6πme hΨ(0) | Ŝβ | Ψ(0) i i
0 0
h i
µ0 ege gK µN sî · (~
N 3 ~ ~ K ) (ri,α − RK,α )
ri − R ŝi,α
(0) (0)
X
+ hΨ | − | Ψ0 i
16πme hΨ(0) | Ŝβ | Ψ(0) i 0 i |~ ~ K |5
ri − R |~
ri − R ~ K |3
0 0
There are two contributions - one isotropic from the Fermi contact and one
anisotropic from the spin-dipolar operator

5.7 Induced Magnetic Fields and NMR Parameters - 1
~ or
Interaction of a charge distribution with an external magnetic induction, B,
with other nuclear magnetic moments, {m ~ L }, leads to an induced current
density ~j ind (~r).
~j,ind
~j ind (~r) gives rise to an induced molecular magnetic induction B
j ~ ~
Bα ~ L })
(R, B, {m = j ~
Bα j,ind ~ ~
(R) + Bα ~ L })
(R, B, {m
j ~ ~
X XX L
Bα ~ L })
(R, B, {m = j ~
Bα (R) − ~
σαβ (R) Bβ − ~ mL
Kαβ (R) β + ...
β L β
~ is the magnetic shielding tensor field

σαβ (R)
Generalization of the magnetic shielding tensor to an arbitrary point R~
~ Bβ is the induced molecular magnetic induction Bα
σαβ (R) j,ind ~ ~
(B) at point R
due to the external magnetic field B
K
σαβ = σαβ (R~ K ) is the nuclear magnetic shielding tensor of nucleus K of
NMR spectroscopy.

L ~
Kαβ (R) is the reduced indirect nuclear spin-spin coupling tensor field
L ~
Kαβ (R) mL j,ind
β is the induced molecular magnetic induction Bα ~ L ) at
(m
~ due to the the magnetic moment m
point R ~L
L ~
Kαβ (R) is the generalization of the reduced indirect nuclear spin-spin
coupling tensor to an arbitrary point R ~
KL L ~
Kαβ = Kαβ (RK ) is proportional to the indirect nuclear spin-spin
KL
coupling tensor Jαβ between nuclei K and L of NMR spectroscopy
KL h KL
Kαβ = J αβ
µ2N gK gL
~ loc which a nuclear K experiences

The local magnetic induction B
X X X
loc ~ j,ind ~
Bα (RK ) = Bα + Bα ~ m
(RK , B, ~ L) = Bα − K
σαβ Bβ − KL
Kαβ mL
β + ...
β L6=K β
X X X
K KL
= δαβ − σαβ Bβ − Kαβ mL
β + ...
β L6=K β
the well know expression of NMR spectroscopy for the local field at nucleus K
The nuclear magnetic shielding constant σ K of liquid or gas-phase NMR
K 1X K
σ = σαα
3 αα
The resonance frequency ν K of nucleus K is the frequency of an allowed

transition between two nuclear spin states of nucleus K.
~loc,K
Energy of a nuclear spin in the presence of a local magnetic induction B
EmI K = −mK loc ~

z Bz (RK )
~K =
where m gK µN
~
I~K and the z-component of the nuclear spin is quantized,
Iz = mI K ~
Resonance frequency of an allowed transition (∆mI K = ±1)
K
|EmI K +1 − EmI K | gK µN loc ~ gK µN K

ν = = Bz (RK ) = 1−σ Bz
h h h
for a molecule in the gas or liquid phase.
σ K cannot be obtained from an NMR spectrum

The measured parameter is the chemical shift δ
ν K − ν K,ref 6
δ= × 10
ν K,ref
The chemical shift written in terms of the shielding constants as
σ K,ref − σ K 6

K,ref K

6
δ= × 10 ≈ σ − σ × 10
1 − σ K,ref
The nuclear magnetic shielding σ K is often also called the absolute nuclear
magnetic shielding
A negative (positive) chemical shift implies that the nucleus is more (less)
shielded than in the reference molecule.

Spin-spin coupling constant is defined through an effective spin Hamiltonian

X gK µN X K 2π X X K KL
Ĥ NMR
=− Iˆα δαβ − σαβ Bβ +
K
Iˆα Jαβ + Dαβ IˆβL
KL
K
~ ~
αβ K6=L αβ
KL
Dαβ is the direct through space dipolar nuclear spin-spin coupling tensor
ˆ
All the corrections terms to − gK~µN I~K · B
~ involve interactions with the
electrons and can be obtained from corrections to the electronic energy
The interaction of the permanent and induced molecular magnetic induction
j ~
~ K of the nuclei: E = − α mK
P
with the magnetic moment m α Bα (RK )
K
XX K j XX K K X X KL K L
~
∆E(B, {m ~ }) = − mα Bα + σαβ mα Bβ + Kαβ mα mβ + . . .
K α K αβ KL αβ
 
X gK µN
ˆ
X K j X K K
ˆ 2π X X KL ˆK ˆL
= − Iα Bα + σαβ Iα Bβ +
 Jαβ Iα Iβ + . . .
K
~ α
~ KL
αβ αβ
L ~ KL
Shows that Kαβ (RK ) ∝ Jαβ

Table 1: Definitions of various magnetic properties as derivatives of the per-

turbed energy E(B, ~ m ~ L ) or of components of the perturbed magnetic
~ K, m
dipole moment mα (B, ~ m ~ B,
~ K ) and molecular magnetic induction Bβj (R; ~ m~ L)
~ m
mα (B, ~ K) Bβj (R;
~ B,
~ m
~ L) Bβj (R
~ K ; B,
~ m~ L) ~ m
E(B, ~ K, m
~ L)
∂ ∂2
ξαβ — — −
∂Bβ ∂Bβ ∂Bα
~ ∂
σβα (R) — − — —
∂Bα
K ∂ ∂ ∂2
σβα − — −
∂mKβ ∂Bα ∂mK
β ∂Bα
L ~ ∂
Kβα (R) — − — —
∂mLα
KL ∂ ∂2
Kβα — — −
∂mLα ∂mK L
β ∂mα

5.8 QM Expression for NMR Parameters - 1
K
σαβ as derivative of the perturbed energy with respect to the nuclear magnetic
moment mK α and the magnetic induction Bβ
(2) ~

K ∂ 2 E0 (B, ~ K )
m
σαβ =
∂mK

α ∂B β ~
~ K |=0
|B|=|m
 
(0) (1)| Ψ(0) ihΨ(0) |Ĥ (1)| Ψ(0) i
∂2 hΨ(0) |Ĥ (2)| Ψ(0) i+
X hΨ 0 |Ĥ n n 0
= |! K 0 0 (0) (0)

∂mα ∂Bβ n6=0 E − E n

0 ~
~ K |=0
|B|=|m
KL
Kαβ as derivative of the perturbed energy with respect to the nuclear
magnetic moments mK α and mβ
L
(2)

KL ∂ 2 E0 (m ~ K, m
~ L )
Kαβ = L
∂mK

α ∂m β
K
~ |=|mL |=0
|m
 
(0) (1)| Ψ(0) ihΨ(0) |Ĥ (1)| Ψ(0) i
∂2 hΨ(0) |Ĥ (2)| Ψ(0) i+
X hΨ 0 |Ĥ n n 0
= K L 0 0 (0) (0)

∂mα ∂mβ n6=0 E − En

0
~ K |=|m
|m ~ L |=0
full expression for the second order energy including the diamagnetic terms

First order perturbing Hamiltonians for the interaction

~ˆ R
with an external magnetic induction: m( ~ GO )
~ˆ OP , O
with a nuclear magnetic moment: O ~ˆ F C and O
~ˆ SD
Second order perturbing Hamiltonians Ĥ (2)

bilinear in the external magnetic induction and nuclear magnetic moment for
K
σαβ
KL
bilinear in two nuclear magnetic moments for Kαβ
Vector potentials
~ˆB (~ri ) = 1 B
for an external field: A ~ × (~ri − R
~ GO )
2
(~
ri −R~K)
~ K (~ri ) =
for a nuclear magnetic moment: A µ0
~K×
m
4π ri −R
|~ ~ K |3

Two second-order perturbing Hamiltonians

(2)
X X mK B K
Ĥ = Ôαβ mα Bβ
K αβ
(2)
XX mK mL
Ĥ = Ôαβ mK L
α mβ
KL αβ
Diamagnetic shielding operator

" #
e2 ~ (ri,β − RK,β )
KB µ0 ~ GO ) · (~
ri − RK )
X
m
Ôαβ = (~
ri − R δαβ − (ri,α − RGO,α )
2me 4π i |~
ri − R~ K |3 |~ ~ K |3
ri − R
Orbital diamagnetic or diamagnetic (nuclear) spin (electronic) orbit

operator
mK mL OD
Ôαβ = Ôαβ
" #
e2 µ 2 X
0 (~ ~ L)
ri − R (~ ~K)
ri − R (ri,α − RL,α ) (ri,β − RK,β )
= · δαβ −
2me 4π |~
ri − R~ L |3 | ~
ri − R~ K |3 |~
ri − R~ L |3 | ~ ~ K |3
ri − R
i

The nuclear magnetic shielding tensor
K (0) mK B (0)
X hΨ(0) OP (0) (0) ~ (0)
0 | ÔK,α | Ψn ihΨn | m̂β (RGO ) | Ψ0 i
σαβ = hΨ0 | Ôαβ | Ψ0 i + (0) (0)
n6=0 E0 − En
X hΨ(0) ~ (0) (0) OP (0)
0 | m̂β (RGO ) | Ψn ihΨn | ÔK,α | Ψ0 i
+ (0) (0)
n6=0 E0 − En
K,dia ~ K,para ~
= σαβ (RGO ) + σαβ (RGO )
Two contributions
paramagnetic contribution is negative
diamagnetic contribution is positive
K,para ~ K,dia ~ ~loc,K > B
~
For σαβ (RGO ) > σαβ (RGO ) the nucleus is de-shielded: B
The two contributions also depend on the gauge origin but their sum, the
shielding tensor, is independent
The separation in a dia- and paramagnetic contribution is arbitrary

The reduced indirect nuclear spin-spin coupling tensor

KL KL,OD KL,OP KL,F C KL,SD KL,F C/SD
Kαβ = Kαβ + Kαβ + Kαβ + Kαβ + Kαβ
where
KL,OD (0) mK mL (0)
Kαβ = hΨ0 | Ôαβ | Ψ0 i
KL,OP /F C/SD
X hΨ(0)
0 | Ô
OP /F C/SD
K,α | Ψ
(0)
n ihΨ
(0)
n | Ô
OP /F C/SD
L,β | Ψ
(0)
0 i
Kαβ = (0) (0)
n6=0 E0 − En
X hΨ(0)
0 | Ô
OP /F C/SD
L,β | Ψ
(0)
n ihΨ
(0)
n | Ô
OP /F C/SD
K,α | Ψ
(0)
0 i
+ (0) (0)
n6=0 E0 − En
KL,F C/SD KL
Kαβ is purely anisotropic and does not contribute to the trace of Kαβ
KL,F C KL,F C KL,F C
Kαβ is isotropic: Kαβ = δαβ Kαβ and
KL,F C KL,F C KL,F C
Kxx = Kyy = Kzz

KL,OP /F C/SD
X hΨ(0) OP /F C/SD
0 | ÔK,α
(0) (0) OP /F C/SD
| Ψn ihΨn | ÔL,β
(0)
| Ψ0 i
Kαβ = (0) (0)
n6=0 E0 − En
X hΨ(0)
0 | Ô
OP /F C/SD
L,β | Ψ
(0)
n ihΨ
(0)
n | Ô
OP /F C/SD
K,α | Ψ
(0)
0 i
+ (0) (0)
n6=0 E0 − En
~ˆ F C and O
O ~ˆ SD contain the electron spin operator Ŝ
(0)
Ŝ|Ψ0 i gives give a linear combination of triplet states
(0) (0) ~ˆ F C/SD (0)
|Ψn i have to be triplet states, otherwise hΨn | O | Ψ0 i = 0
For K KL,F C , K KL,SD and K KL,F C/SD are the excited states triplet states
~ˆ OP is spin free
O
(0)
The excited states have the same spin symmetry as |Ψn i
For K KL,OP are the excited states singlet states

Linear response expression for NMR parameters

j ~ ~ m ~ m
K ∂Bα (RK , B, ~ L) ∂mβ (B, ~ K)
σαβ = =
∂Bβ ∂mK α
j ~ ~ m
KL ∂Bα (RK , B, ~ L)
K =
∂mL β
The operator for the molecular magnetic moment in the presence of

nuclear magnetic moments
K lB sB
X X mK B
~ GO , {m
m̂α (R ~ GO ) + Ôα −
~ }) = Ôα (R Ôβα (R~ GO ) mK
β
K β
The operator for the molecular magnetic induction in the presence of an

~ and nuclear magnetic moments {m
external magnetic induction B ~ L}
lmK smK mK B ~ K mL
j ~
X XX
B̂α ~ L })
~ {m
(RK , B, = Ôα +Ôα − Ôαβ (RGO ) Bβ −2 m
Ôαβ mL
β
β L6=K β

Nuclear magnetic shielding tensor of nucleus K as first derivative of the

j ~ ~ at the position of nucleus K in the
molecular magnetic induction Bα (RK , B)
presence of an external magnetic induction

j ~ ~

K ∂Bα (RK , B) ∂ ~ | B̂ j (R
~ K , B)
~ | Ψ0 (B)i
~

σαβ = − =− hΨ0 (B) α
∂Bβ ~ ∂Bβ
~
|B|=0
|B|=0

(0)
~
∂Ψ0 (B) ~
∂Ψ0 (B) (0)
OP OP
= −hΨ0 | ÔK,α | i−h | ÔK,α | Ψ0 i
∂Bβ ~ ∂Bβ ~
|B|=0 |B|=0
(0)m KB (0)
+hΨ0 | Ôαβ ~ GO ) | Ψ i
(R 0
Alternatively as first derivative of a component of the molecular magnetic

dipole moment mα (m ~ K ) in the presence of a magnetic nucleus K

K ∂mα ~ K)
(m ∂ K ~ GO , m K K

σβα = − = − hΨ ( m
~ ) | m̂ ( R ~ ) | Ψ ( m
~ )i

0 α 0
∂mK K

β
∂m β
K
~ K |=0
|m |m
~ |=0

(0) l ~ ~ K )
∂Ψ0 (m ∂Ψ0 (m ~ K ) l ~ (0)
= −hΨ0 | m̂α (RGO ) | K i − h K | m̂ α ( R GO ) | Ψ 0 i
∂mβ K
∂mβ K
|m
~ |=0 |m
~ |=0
(0)m KB (0)
+hΨ0 | Ôβα ~ GO ) | Ψ i
(R 0
In both expressions we have assumed that we are looking at closed-shell

~ˆ sB and
molecules only and have therefore not included the contributions from O
O~ˆ smK .
The reduced indirect nuclear spin-spin coupling constant of nuclei K and L as

j ~
first derivative of the molecular magnetic induction Bα (R K , m ~ L ) at the position
of nucleus K in the presence of a magnetic nucleus L

j ~ L
∂B ( R , m
~ ) ∂

KL α K L j ~ L L
Kαβ = − = − hΨ ( m
~ ) | B̂ ( R , m
~ ) | Ψ ( m
~ )i

0 α K 0
∂mL ∂m L

β β
L
~ L |=0
|m |m
~ |=0

∂Ψ0 (m ~ L)

(0) OP FC SD
= −hΨ0 | ÔK,α + ÔK,α + ÔK,α | i
∂mL

β

|m~ L |=0

L
∂Ψ0 (m ~ ) OP FC SD (0)
−h L
| ÔK,α + ÔK,α + ÔK,α | Ψ 0 i
∂mβ

L
|m
~ |=0
(0) mK mL (0)
+2hΨ0 | Ôαβ | Ψ0 i

Using response functions we finally can write for the nuclear magnetic
shielding tensor

K ∂ ~ j ~ ~ ~

σαβ = − hΨ0 (B) | B̂α (RK , B) | Ψ0 (B)i
∂Bβ ~
|B|=0

∂ K ~ K K

= − hΨ0 (m
~ ) | m̂β (RGO , m ~ ) | Ψ0 (m ~ )i
∂mK α ~ K |=0
|m
(0) mK B ~ (0) OP
= hΨ0 | Ôαβ (RGO ) | Ψ0 i + hh ÔK,α ; m̂lβ (R
~ GO ) iiω=0
and for the reduced indirect nuclear spin-spin coupling constant of nuclei K
and L

∂

KL L j ~ L L
Kαβ = − L
~ ) | B̂α (RK , m
hΨ0 (m ~ ) | Ψ0 (m
~ )i
∂mβ L
|m
~ |=0
(0) mK mL (0) OP OP
= 2hΨ0 | Ôαβ | Ψ0 i + hh ÔK,α ; ÔL,β iiω=0
FC SD FC SD
+hh ÔK,α + ÔK,α ; ÔL,β + ÔL,β iiω=0

5.9 SOS Expression for Diamagnetic Terms - 1
dia ~ K,dia ~
One can rewrite the diamagnetic contributions ξαβ (RGO ) and σαβ (RGO ) as
a sum-over-states or linear response functions denoted with a superscript “∆"
h i
hΨ
(0)
| ˆ (R
µ
~ ~ GO ) × m ˆ (R
~ l ~ GO ) | Ψ(0) (0) p
n ihΨn | Ôβ | Ψ0 i
(0)
∆ ~ 1 X 0
α
ξαβ (RGO ) = − (0) (0)
2me E 0 − E n
n6=0
h i
hΨ
(0)
| Ô p
| Ψ
(0)
n ihΨ
(0)
n | ˆ (R
µ
~ ~ GO ) × m ˆ l (R
~ ~ GO ) | Ψ(0) i
1 X 0 β
α
0
− (0) (0)
2me n6=0 E0 − En
1 h i
= − hh µ ~ˆ(R~ GO ) × m ~ˆ (R
l ~ GO ) ; Ôβp iiω=0
2me α
ˆ
h i
hΨ
(0)
| O~ (R
µ ~K) × m ˆ (R
~ l ~ GO ) | Ψ(0) (0) p
n ihΨn | Ôβ | Ψ0 i
(0)
K,∆ ~ 1 X 0
α
σαβ (RGO ) = 2 (0) (0)
me c n6=0 E0 − En
ˆ
h i
hΨ
(0)
| Ô p
| Ψ
(0) (0) ~ ~
n ihΨn | O (RK ) × m
µ ˆ (R
~ l ~ GO ) | Ψ(0) i
1 X 0 β
α
0
+
me c2 E
(0)
0 − E n
(0)
n6=0
1 h
ˆ
~ (Rµ ~K) × m ˆ (Rl ~ GO )
i
p
= hh O ~ ; Ôβ iiω=0
me c2 α
KL,OD
and Kαβ as
~ˆ µ (R ~ˆ L
h i
(0) (0) (0) (0)
1 X hΨ0 | O ~K) × O OP
| Ψn ihΨn | Ôβp | Ψ0 i
KL,∆ α
Kαβ =
me c2 E0 − En
(0) (0)
n6=0
ˆ ˆ
h i
(0) p (0) (0) ~ (R
µ ~L
~K) × O OP (0)
1 X hΨ0 | Ôβ | Ψn ihΨn | O | Ψ0 i
α
+
me c2 n6=0
(0)
E0 −
(0)
En
1 h
ˆ µ
~ (R ˆ
~K) × O
~LOP
i
p
= 2
hh O ; Ôβ iiω=0
me c α
ˆ, m
where ~
µ ~ˆ µ are the electric and magnetic dipole and electric field
~ˆ l and O
~ˆ p is the total canonical momentum operator of the electrons.
operators, while O

In order to prove this we rewrite the operators of the diamagnetic terms as

dia ˆ ˆ ˆ ˆ

Ô =f O ~1 · O ~2
~1 ⊗ O
~ 2 I3 − O
where O ~ˆ 1 O
~ˆ2 = O
~ˆ 1 ⊗ O ~ˆ 2T is the outer or dyadic product of two vectors, which
ˆ ˆ

gives a 3 × 3 matrix with elements O ~1 ⊗ O~2 = Ô1,α Ô2,β .
αβ
~ˆ 1 and O
The constants f and the operators O ~ˆ 2 are
f Ô1 Ô2
e2
ξ dia
− ~ˆ(R
µ ~ GO ) ~ˆ(R
µ ~ GO )
4me
1 ~ˆ µ (R
σ K,dia − ~ˆ(R
µ ~ GO ) O ~K)
2me c2
1 ~ˆ µ (R
~ L) ~ˆ µ (R
~K)
K KL,dia O O
2me c4
The diamagnetic operators in can be written as the following commutator

ˆ ˆ ˆ
~1 ⊗ O
~ 2 I3 − O
~1 · O ˆ ı
~2 = − O
h
ˆ

~2 × O ˆ
~1 × O ˆ p

~ , (Oˆ r
~ ) T
i
O
~
~ˆ r is the sum of the position operators of all electrons.

where O
The diamagnetic contributions can therefore be written as
(0) dia (0) f (0)

h
ˆ

~2 × O ˆ
~1 × Oˆ p
~ , (O

ˆ
i
~ ) | Ψ(0) i
r T
hΨ0 | Ô | Ψ0 i = hΨ0 | O 0
ı~
f n (0) h ~ˆ
ˆ
~1 × O ~ˆ p
i
~ˆ r )T | Ψ(0) i
= hΨ0 | O2 × O (O 0
ı~
iT
ˆ r ~ˆ ~ˆ 1 × O ~ˆ p
h
(0) ~ (0)
−hΨ0 | O O2 × O | Ψ0 i

P (0) (0)
We can then insert the resolution of the identity, n |Ψn ihΨn | = 1, between
ˆ ˆ ˆ ~ˆ r leading to

~2 × O
the operators O ~
~1 × O p
and O
(0) dia (0) f X n (0) ~ˆ

ˆ ˆ

~ | Ψ(0)
~1 × O p (0) ~ˆ r T (0)
hΨ0 | Ô | Ψ0 i = hΨ0 | O2 × O n ihΨn | (O ) | Ψ0 i
ı~ n
iT
ˆ r (0) ~ˆ ~ˆ ~ˆ p
h
(0) ~ (0)
−hΨ0 | O | Ψn ihΨ(0)
n | O2 × O1 × O | Ψ0 i
Using the off-diagonal hypervirial relation, we arrive at the SOS expression
ˆ ˆ ˆ ~ˆ p )T | Ψ(0) i

(0) ~ ~1 × O~ | Ψ(0)
p (0)
dia f hΨ 0 | O 2 × O n ihΨ n | ( O 0
(0) (0)
X
hΨ0 | Ô | Ψ0 i = (0) (0)
me E0 − En
n6=0
iT
ˆ p (0) ˆ ˆ ˆ
h
(0) ~ (0) ~2 × O ~1 × O ~ p (0)
f X hΨ 0 | O | Ψ n ihΨ n | O | Ψ 0 i
+ (0) (0)
me E 0 − E n
n6=0
(0) ~ ˆ p (0)
where n = 0 was removed from the summation, because hΨ0 | O | Ψ0 i = 0
according to the hypervirial theorem Molecular Electromagnetism A Computational Chemistry Approach – p.46/52
5.10 The Gauge Origin Problem - 1
Vector potential is not uniquely defined by
~ r) = ∇
B(~ ~ ×A
~ B (~r)
~ B (~r) but not B(~

Gauge transformation changes A ~ r)
′
~ B (~r) → A
A ~ B (~r) = A
~ B (~r) + ∇χ(~
~ r)
Changes also the Hamiltonian

P e P e
′
Ĥ = e −ı i ~ χ(~
ri )
Ĥ e ı i ~ χ(~
ri )
But transforming also the wavefunction

P e
|Ψi → |Ψ i = e′ −ı i ~ χ(~
ri )
|Ψi
restores gauge invariance of Schrödinger equation

All properties have to be invariant under gauge transformations

A particularly important gauge function

1 ~ ~
χ(~ri ) = − B × RGO · ~ri
2
~ GO
⇒ vector potential depends on an arbitrary gauge origin R
As a consequence the dia- and paramagnetic contributions to ξ depend
~ GO
quadratically on R
dia ~ para ~
ξαβ = ξαβ (RGO ) + ξαβ (RGO )
~ GO
The two contributions to σ K depend linearly on R
K K,dia ~ K,para ~
σαβ = σαβ (RGO ) + σαβ (RGO )
~ GO
But the sum of both contributions must be independent of R
~ GO + D)
ξ(R ~ = ξ(R ~ ξ (R
~ GO ) + C ~ GO )D ~ Cξ D
~ +D ~
1 2
~ GO
σ(R ′ ~
+ D) = σ(R ~ 1σ D
~ GO ) + C ~

The linear gauge origin dependence tensor for ξ


(0) p (0) (0) l ~ (0)
ξ e X X hΨ 0 | Ôβ | Ψ n ihΨ n | m̂γ ( R GO ) | Ψ 0 i
~
C1,α (RGO ) = − ǫαβγ 
3me (0) (0)
βγ n6=0 E0 − En

X hΨ(0)
0 | m̂l ~
β ( R GO ) | Ψ
(0)
n ihΨ
(0)
n | Ôγ
p
| Ψ
(0)
0 i
− (0) (0)
n6=0 E0 − E n
e (0) (0)
− hΨ0 | µ̂α (RGO ) | Ψ0 i
3me
e X
p l ~ l ~ p

=− ǫαβγ hh Ôβ ; m̂γ (RGO ) iiω=0 − hh m̂β (RGO ) ; Ôγ iiω=0
6me βγ
e (0) (0)
− hΨ0 | µ̂α (RGO ) | Ψ0 i
3me
where ǫαβγ is the Levi-Civita symbol


 +1 if αβγ is an even permutation of x, y, z


ǫαβγ = −1 if αβγ is an odd permutation of x, y, z


0 otherwise

The quadratic gauge origin dependence tensor for ξ


2 (0) p (0) (0) p (0)
ξ e X X hΨ 0 | Ôγ | Ψ n ihΨ n | Ôγ | Ψ 0 i
C2,αβ =−  δ αβ
12m2e γ
(0)
E0 − En
(0)
n6=0

X hΨ(0)
0 | Ôα
p
| Ψ
(0)
n ihΨ
(0)
n | Ôβ
p (0)
| Ψ0 i e 2
− (0) (0)
− N δαβ
n6=0 E 0 − E n
6me
!
2
e X e2
=− δαβ hh Ôγp ; Ôγp iiω=0 − hh p
Ôα ; Ôβp iiω=0 − N δαβ
24m2e γ
6me

The gauge origin dependence tensors for σ K


(0) p (0) (0) OP (0)
σ e X X hΨ 0 | Ôβ | Ψ n ihΨ n | ÔK,γ | Ψ 0 i
C1,α = − ǫαβγ 
(0) (0)
6me E0 − En
βγ n6=0

(0) OP (0) (0) (0)
X hΨ0 | ÔK,β | Ψn ihΨn | Ôγp | Ψ0 i
− (0) (0)

n6=0 E0 − En
e (0) µ ~ (0)
− hΨ | Ôα ( R K ) | Ψ 0 i
3me c2 0
e X
p OP OP p

=− ǫαβγ hh Ôβ ; ÔK,γ iiω=0 − hh ÔK,β ; Ôγ iiω=0
6me βγ
e (0) µ ~ (0)
− hΨ 0 | Ôα ( R K ) | Ψ 0 i
3me c2
They are all zero in the case of exact unperturbed wavefunctions

In the case of approximate methods they can serve as a measure of the gauge
origin dependence

(CTOCD-DZ) approach by Lazzeretti and co-workers:

continuous transformation of origin of the current density whereby the
diamagnetic contribution to the current density is set to zero

(0) p (0) (0) CT OCD−DZ (0)
K,∆
X X hΨ 0 | Ôγ | Ψ n ihΨ n | Ô K,δα | Ψ 0 i
σαβ = ǫβγδ  (0) (0)
γδ n6=0 E 0 − E n

(0) CT OCD−DZ (0) (0) (0)
X hΨ0 | ÔK,δα | Ψn ihΨn | Ôγp | Ψ0 i
+ (0) (0)

n6=0 E0 − En
X
= CT OCD−DZ
ǫβγδ hh Ôγp ; ÔK,δα iiω=0
γδ
where
CT OCD−DZ 1 h OP OP
i
ÔK,δα = µ̂δ (RGO ) ÔK,α + ÔK,α µ̂δ (RGO )
4me
σ K,dia reformulated as a sum over all states in such a way that the gauge
origin dependence disappears exactly from the diamagnetic and paramagnetic
contribution.
Chapter 6
Stephan P. A. Sauer
sauer@kiku.dk


6. Properties Related to Nuclear Motion
So far
we have restricted ourselves to the situation that the nuclei are fixed in
space
we have considered molecular properties or contributions to the molecular
properties which can be obtained from the electronic Schrödinger equation
In this and the following chapter
we will finally lift this restriction and allow the nuclei to move again.
we will look at properties which arise or at least have contributions due to a
breakdown of the Born-Oppenheimer approximation.
we have to take into account the coupling of nuclear and electronic motion,
i.e. some of the terms which are neglected in the Born-Oppenheimer
approximation

6.1 Molecular Rotation as Source for Magnetic Moments - 1
When a molecule rotates around its centre of nuclear masses RCM , there are
rotating charges, which give rise to an additional current density ~j J (~r).
~j J (~r) = ρ(~r) ~v (~r) = ρ(~r) ω ~ CM )

~ × (~r − R
= ~ × (~r − R
ρ(~r)(I −1 J) ~ CM )
ω = I −1 J~ is the angular velocity of the rotating charges, J~ the angular

where ~
momentum of the rotation and I the moment of inertia tensor.
This rotational current density will lead to a magnetic moment m ~ J , called the
rotational magnetic moment,
Z
J,rig ~ 1 h
~ CM ) × (I J)
−1 ~ × (~r − R
i
~ CM ) d~r
m~ (J) = ρ(~r)(~r − R
2 ~r
We have assumed that the charges rotate rigidly. This means in particular that
the electrons move rigidly with the nuclear frame and do not lag behind.

However, the electronic charge is influenced by the fact that the nuclei rotate
with angular momentum J. ~
The coupling between nuclear rotational motion and the motion of the
electrons
~ˆ I −1 J~ˆ
Ĥ (1) = L
is neglected in the Born-Oppenheimer approximation.
The changes in the rotational magnetic moment and induction are thus a
manifestation of the breakdown of the Born-Oppenheimer approximation.
Going to the rotating frame of the nuclei leads according to Larmor’s theorem
to an apparent magnetic induction B ~J acting on the electrons
~J = − 2me I −1 J~
B
e
with associated vector potential
~ J = − me I −1 J~ × (~r − R
A ~ CM )
e
This magnetic induction leads to an induced magnetic moment m ~ J,ind in

addition to the magnetic moment of the rigidly rotating charge distribution
mJα (J)
~ = mJ,rig
α
~ + mJ,ind
(J) α
~
(J)
The rigid and induced or Born-Oppenheimer breakdown contribution is

combined in one property, the rotational g tensor g J , defined as
proportionality tensor between the rotational magnetic moment m ~ and the
~ J (J)
rotational angular momentum J~
J ~ µN X µN X rig ind
mα (J) = gJ,αβ Jβ = (gJ,αβ + gJ,αβ ) Jβ
~ ~
β β
e~
where µN = 2mp
is the nuclear magneton again.
Defining the rotational g tensor as first derivative of the rotational magnetic
moment

J ~
~ ∂mα (J)
gJ,αβ =
µN ∂Jβ ~

|J|=0
The rotational magnetic moment m ~ will interact with an external magnetic

~ J (J)
~ like any other magnetic moment
induction B
~ J)
~ = −
X J ~ =− µ N
X
∆E(B, Bα mα (J) Bα gJ,αβ Jβ
α
~ αβ
Defining the rotational g tensor as second derivative of the corresponding

interaction energy

2 ~ J)
~ ∂ E(B, ~
gJ,αβ = −
µN ∂Bα ∂Jβ ~ ~

|J|=|B|=0
However, the energy here is in principle the solution to the full

time-independent Schrödinger equation including the nuclear kinetic energy
terms. Consequently, only the induced contributions can be obtained as
derivatives of the electronic energy in the presence of the Born-Oppenheimer
breakdown operator given in.

Already in 1933 rotational magnetic moments and thus the rotational g tensor
were measured by deflection of molecular beams in inhomogenous magnetic
fields.
Alternatively, one can study the changes in the rotational energies due to an
external magnetic field, the so-called rotational Zeeman effect, which is
normally expressed in terms of an effective rotational Hamiltonian as
rot
ˆx2
J ˆy2
J Jˆz2 µN ~ ~ˆ
Ĥ ~
(B) = + + − BgJ J
2Ixx 2Iyy 2Izz ~
and allows for the experimental determination of the rotational g tensor.

6.2 QM Expression for the Rotational g Tensor - 1
An expression for the rigid contribution to the rotational magnetic moment is

most easily derived in analogy to the electric dipole moment by translating the
classical expression to quantum mechanics.
We make use of Lagrange’s formula for a vector triple product

~× B
A ~ ×C ~ =B ~ A ~·C ~ − A ~·B ~ C ~
and rewrite the classical expression for the rigid contribution
~ J,rig (J)
m ~
Z
1 h
~ CM ) · (~r − R
~ CM ) I 3 − (~r − R
~ CM ) ⊗ (~r − R
i
~ CM ) (I −1 J)
~ d~r
= ρ(~r) (~r − R
2 ~r
We can identify a component of the rigid contribution to the rotational g tensor

as
Z i 1
rig m p
h
2
~ CM ) δαβ − (rα − RCM,α )(rβ − RCM,β )
gJ,αβ = ρ(~r) (~r − R d~r
e ~r Iββ

The charge distribution ρ(~r) consists of the discrete nuclear charges located at
~ K and the continuous charge distribution ρel (~r) of the electrons
the points R
rig
The rigid contribution to the rotational g tensor gJ,αβ has thus a nuclear
X h i 1
nuc ~K − R 2
~ CM ) δαβ − (RK,α − RCM,α )(RK,β − RCM,β )
gJ,αβ = mp ZK (R
K
Iββ
and an electronic contribution

rig,el 2mp (0) JJ ~ (0) 1
gJ,αβ =− hΨ0 | Ôαβ (RCM ) | Ψ0 i
me Iββ
JJ ~
where the electronic perturbation operator Ôαβ (RCM ) is defined as
N
JJ ~ me X h 2
~ CM ) δαβ − (ri,α − RCM,α )(ri,β − RCM,β )
i
Ôαβ (RCM ) = (~ri − R
2 i
N
X
= ôJJ ~
i,αβ (RCM )
i

The derivation of the induced contribution, on the other hand, is very similar to
the derivation for the magnetizability.
We could start from the definition of the rotational g tensor as first derivative of
the rotational magnetic moment, which is then the induced contribution to it
Here we will make use of the definition as second derivative of the energy.
Applying this to the electronic energy in the presence of an external magnetic
induction B~ and the internal Born-Oppenheimer breakdown perturbation we
will obtain directly the induced contribution to the rotational g tensor.
We have a first-order perturbation Hamiltonian operator from the vector
potential due to the coupling with the rotation
(1)
X lJ
Ĥ = Ôα (R ~α
~ CM )(I −1 J)
α
and second-order operator which is bilinear in the external magnetic induction

and the coupling with the rotation
(2)
X BJ
Ĥ = ~ CM , R
Ôαβ (R ~ GO ) Bα (I −1 J)
~β
αβ Molecular Electromagnetism A Computational Chemistry Approach – p.10/58
where the first and second-order perturbation operators are given as

N N
ˆ ~ CM ) = 2me m̂lα (R
lJ ~
X X
Ôα (RCM ) = ôlJ ~
i,α (RCM ) = − li,α (R ~ CM )
i i
e
and
N
X
BJ ~ ~ GO ) ~ ~
Ôαβ (RCM , R = ôBJ
i,αβ (RCM , RGO )
i
N
e Xh ~ CM ) · (~ri − R
~ GO )δαβ − (ri,α − RCM,α )(ri,β − RGO,β )
i
=− (~ri − R
2 i
One should note that both perturbation operators are similar to the
corresponding operators for the magnetizability tensor, but not equal.
Apart from constant factors, they differ such that for the rotational g tensor the
first-order operator as well as one of the factors in the second-order operator
are defined with respect to the nuclear centre of masses R ~ CM and not with
respect to the arbitrary gauge origin R~ GO .
Using again perturbation theory for the perturbed energy we can obtain the
second derivative of the electronic energy directly from the second-order
correction to the energy

2 (2) ~ ~
ind ~ ∂ E0 (B, J)
gJ,αβ = −
µN ∂Bα ∂Jβ ~ ~

|J|=|B|=0

(0) l ~ (0) (0) l ~ (0)
1  4me X hΨ0 | m̂α (R GO ) | Ψn ihΨn | m̂β (RCM ) | Ψ0 i
= mp
Iββ e2 (0)
E0 − En
(0)
n6=0
(0)
!
~ CM ) | Ψ(0)
hΨ0 | m̂lβ (R
(0) l ~ (0)
n ihΨn | m̂α (RGO ) | Ψ0 i
+ (0) (0)
E0 − En

2 (0) BJ ~ ~ GO ) | Ψ(0) i
− hΨ0 | Ôαβ (RCM , R 0
e
The induced contribution consists therefore of a paramagnetic or

sum-over-states contribution and a diamagnetic or ground state expectation
value term.

Combining these with the contribution from the rigid charges yields
para dia nuc
gJ,αβ = gJ,αβ + gJ,αβ + gJ,αβ
(0) ~ GO ) | Ψ(0) (0) l ~ (0)
mp 4me X hΨ0 | m̂lα (R n ihΨn | m̂β (RCM ) | Ψ0 i
=
Iββ e2 (0)
E0 − En
(0)
n6=0
(0)
!
~ CM ) | Ψ(0)
hΨ0 | m̂lβ (R
(0) l ~ (0)
n ihΨn | m̂α (RGO ) | Ψ0 i
+ (0) (0)
E0 − En
mp (0) X h ~ ~ CM ) · (~ri − R
~ CM )δαβ
− hΨ0 | (RGO − R
Iββ i
#
(0)
− (ri,α − RCM,α )(RGO,β − RCM,β ) | Ψ0 i
mp X h
~K − R 2
~ CM ) δαβ − (RK,α − RCM,α )(RK,β − RCM,β )
i
+ ZK (R
Iββ K

This is one expression for the rotational g tensor consisting of three terms: a
paramagnetic term, a new diamagnetic-like term and a nuclear contribution.
The first term can then also be expressed in terms of linear response functions
according to leading to
para 4mp me l ~ l ~
gJ,αβ = hh m̂α ( R GO ) ; m̂β (RCM ) iiω=0
e2 Iββ
However the new diamagnetic contribution can also be written as a sum over
all states
(0)
Xh i
(0)
hΨ0 | ~ ~ ~
ri − RCM )(RGO − RCM )δαβ − (ri,α − RCM,α )(RGO,β − RCM,β ) | Ψ0 i
(~
i
(0) P ~ GO − R
~ CM ) × p (0) (0) P ~ CM ) | Ψ(0) i
1 X hΨ0 | i [( R ~ i ] α | Ψ n ihΨ n | i l i,β ( R 0
=− (0) (0)
me n6=0 − EnE0
(0) P (0) (0) P (0)
!
~ ~ ~ ~i ]α | Ψ0 i
hΨ0 | i li,β (RCM ) | Ψn ihΨn | i [(RGO − RCM ) × p
+ (0) (0)
E0 − En

Adding this reformulated diamagnetic contribution replaces the dependence of

one of the orbital angular momentum operators on the gauge origin RGO in the
paramagnetic term by a dependence on the centre of nuclear masses RCM .
An alternative expression for the rotational g tensor is thus
nuc el
gJ,αβ = gJ,αβ + gJ,αβ
mp X h
~K − R 2
~ CM ) δαβ − (RK,α − R ~ CM,α )(RK,β − R ~ CM,β )
i
= ZK (R
Iββ K
(0) P ~ (0) (0) P ~ CM ) | Ψ(0) i
mp X hΨ0 | i li,α (RCM ) | Ψn ihΨn | i li,β (R 0
+ (0) (0)
me Iββ n6=0 E0 − En
(0) P (0) (0) (0)
!
~ CM ) | Ψn ihΨn |
hΨ0 | i li,β (R
P ~
i li,α (RCM ) | Ψ0 i
+ (0) (0)
E0 − En
It consists of a nuclear and an electronic sum-over-states contribution, which

can be expressed as linear response function
el 4mp me l ~ l ~
gJ,αβ = hh m̂α ( R CM ) ; m̂β (RCM ) iiω=0
e2 Iββ
The equivalence between the two expressions is based on the reformulation of

the diamagnetic contribution as a sum-over-states or linear response function
and only exact, if we are dealing with the exact eigenstates of the unperturbed
Hamiltonian.
The electronic contribution is proportional to the paramagnetic contribution to
the magnetizability
el 4mp me para ~
gJ,αβ =− 2 ξ (RCM )
e Iββ αβ
if one chooses RCM , as the gauge origin for the magnetizability.
This is frequently used for the experimental determination of the paramagnetic
contribution to the magnetizability from measured rotational g tensors
Combined with a calculated diamagnetic contribution to the magnetizability
one can thus obtain semi-experimental values of the magnetizability
dia ~ e2 Iββ nuc

ξαβ = ξαβ (RCM ) − gJ,αβ − gJ,αβ
4mp me
This relation only holds for a given fixed nuclear geometry.

6.3 Rotational g Tensor and Electric Dipole Moment - 1
′
We consider a component of the rotational g tensor, gJ,αβ , of one isotopologue
with moment of inertia tensor I ′ and centre of nuclear masses, R ~ CM
′
, which is
~ =R
shifted by a vector D ~ CM
′
−R ~ CM from the centre of nuclear masses of the
second isotopologue with moment of inertia tensor I
X h i
′ ′ ~K − R ′ 2
~ CM ) δαβ − (RK,α − R ~ CM,α )(RK,β − R
′ ~ CM,β )
′
gJ,αβ Iββ = mp ZK (R
K
mp X ~ ′
X
~ CM
′
+ hh [(~ri − RCM ) × p
~i ]α ; [(~ri − R )×p
~i ]β iiω=0
me i i
Using the relation between the two centres of mass we can rewrite this as
′ ′
gJ,αβ Iββ = gJ,αβ Iββ
m X
p
ǫαγδ ǫβζη Dγ Dζ hh Ôδp ; Ôηp iiω=0
X
+ mp ~ ~
ZK D · D δαβ − Dα Dβ +
K
me γδζη
X h i
− mp ZK 2D ~ · (R
~K − R
~ CM )δαβ − Dα (RK,β − RCM,β ) − Dβ (RK,α − RCM,α )
K
!
mp X
ǫαγδ hh Ôδp ; ~ CM ) ; Ôp iiω=0
X X
− Dγ ~ CM ) iiω=0 + ǫβγδ hh
li,β (R li,α (R δ
me γδ i i
We are going to rewrite the three linear response functions as ground state
expectation values. We will use the equation-of-motion of the polarization
propagator for zero frequencies.
′ ′
X h i
− mp ZK 2D ~ · (R
~ K −R
~ CM )δαβ −Dα (RK,β −RCM,β )−Dβ (RK,α −RCM,α )
K
mp X (0)
X
~ CM )] | Ψ(0) i
+ Dγ ǫαγδ hΨ0 | [Ôδr , li,β (R 0
ı~ i
γδ
!
(0) ~ CM ), Ôδr ] | Ψ(0) i
X
−ǫβγδ hΨ0 | [ li,α (R 0
i
X
+ mp ZK ~ ·D
D ~ δαβ − Dα Dβ
K
mp X (0) (0)
+ ǫαγδ ǫβζη Dγ Dζ hΨ0 | [Ôδp , Ôηr ] | Ψ0 i
ı~
γδζη

Evaluating the commutators we obtain for the expectation values

(0) (0)
hΨ0 | [Ôδr , Ôηp ] | Ψ0 i = ı~N δδη
(0) ~ CM )] | Ψ(0) i (0) (0)
r
X X
hΨ0 | [Ôδ , li,β (R 0 = ı~ǫδβζ hΨ0 | (ri,ζ − RCM,ζ ) | Ψ0 i
i i
where N is the total number of electrons,

and thus for the relation between the rotational g tensors
′ ′
X h i
− mp ZK 2D ~ · (R
~ K −R
~ CM )δαβ −Dα (RK,β −RCM,β )−Dβ (RK,α −RCM,α )
K
~ · hΨ(0) | ~ CM ) | Ψ(0) iδαβ

X
+ 2mp D 0 (~ri − R 0
i
(0) (0) (0) (0)
X X
− mp Dα hΨ0 | (ri,β −RCM,β ) | Ψ0 i−mp Dβ hΨ0 | (ri,α −RCM,α ) | Ψ0 i
i i
X
+ mp ~ ·D
ZK D ~ δαβ − Dα Dβ − mp N D
~ ·D
~ δαβ − Dα Dβ
K

Recalling the definition of the electric dipole moment, introducing the total
charge q of the molecule
X
q=e ZK − eN
K
~ =R
and using the definition of D ~ CM
′
−R ~ CM we arrive finally at the relation
between the rotational g tensor of two isotopologues of a molecule and its
dipole moment
′ ′
mp ~ ′ ~ CM ) · µ ~ CM ) δαβ
−2 (RCM − R ~ (R
e
mp ′ ~ CM ) + mp (RCM,β
′ ~ CM )
+ (RCM,α − RCM,α ) µβ (R − RCM,β ) µα (R
e e
mp h
~ CM − R
′ 2
~ CM ) δαβ − (RCM,α − RCM,α )(RCM,β − RCM,β )
′ ′
i
+ q (R
e
This relation allows us to determine experimentally the electric dipole moment
by simply measuring the rotational g tensor of two isotopologues.
However, the expression holds only for a given fixed nuclear geometry.
6.4 Rotational g Tensor & Electric Quadrupole Moment - 1
The rotational g tensor can also be related to the electric quadrupole moment
tensor Θ, defined as
1
~ CM ) = hΨ | (0) 1 h
2 2
i
~ CM ) | Ψ(0) i
~ CM ) − µ̂2z (R
~ CM ) + µ̂y (R
Θxx (R 0 µ̂x (R 0
e 2

X
~ K,z − R 1
~ CM,z )2 − [(R ~ K,x − R~ CM,x )2 + (R~ K,y − R ~ CM,y )2 ]
+ ZK e (R
K
2
with the otherwise arbitrary origin of the coordinate system at the centre of
nuclear masses, R~ CM
The nuclear contribution to the quadrupole moment tensor is expressible in

nuc
terms of components of the nuclear contribution to the rotational g tensor, gJ,αβ
The electronic contribution is expressible in terms of components of the
dia ~
diamagnetic contribution to the magnetizability tensor, ξαβ (RCM ), as

~ CM ) = e 1 nuc nuc nuc
Θxx (R Ixx gxx,J + Iyy gyy,J − Izz gzz,J
mp 2
h
4me 1 dia ~ dia ~
i
dia ~
− ξxx (RCM ) + ξyy (RCM ) − ξzz (RCM )
e 2
6.4 Rotational g Tensor & Electric Quadrupole Moment - 2
Using the relation between the paramagnetic contribution to the

magnetizability and the electronic contribution to the rotational g tensor, we
can expressed the quadrupole moment also in terms of the total
magnetizability and g tensor as

~ CM ) = e 1
Θxx (R (Ixx gxx,J + Iyy gyy,J ) − Izz gzz,J
mp 2

4me 1
− (ξxx + ξyy ) − ξzz
e 2
with corresponding relations for the other components.

Since the quadrupole moment of a polar molecule depends on the origin of the
coordinate system it is important to remember that the centre of nuclear
masses is automatically chosen as origin here.
The relation holds for the values of these properties for a given fixed nuclear
geometry.

6.5 Molecular Rotation as Source for Magnetic Fields - 1
We will now look at the effect of molecular rotation on the magnetic field at the
position R̂K of a magnetic nucleus K.
The current density of the rigidly rotating charges gives rise to an additional
contribution, a rotational magnetic induction B ~j,J (R)
~
h i
Z ~ K ) × (I J)
(~r − R −1 ~ × (~r − R
~K)
~j,J,rig (R
~ K , J)
~ = µ 0
B ρ(~r) d~r
4π ~r ~
| ~r − RK | 3
where the position vectors are now with respect to the nucleus of interest.
Again we have assumed that the charges rotate rigidly and therefore added
the superscript “rig".
The coupling between the rotational motion of the nuclei and the motion of the
~j,J,ind , so that the total
electrons leads to an additional induced contribution, B
rotational magnetic induction at the position of nucleus K can be written as
j,J ~ ~ j,J,rig ~ j,J,ind ~
~ + Bα ~
Bα (RK , J) = Bα (RK , J) (RK , J)

One defines a spin rotation tensor C K as proportionality tensor between the

~j,J (R
rotational magnetic induction B ~ K , J)
~ and the rotational angular
momentum J~
j,J ~ ~ 2π X K 2π X K,rig K,ind
Bα (RK , J) = Cαβ Jβ = (Cαβ + Cαβ ) Jβ
µN gK µN gK
β β
e~
where µN = 2mp
and gK are the nuclear magneton and the nuclear g-factor of
nucleus K.
The induced contribution to the spin-rotation tensor is again a consequence of
the breakdown of the Born-Oppenheimer approximation.
The individual cartesian components of the spin-rotation tensor can thus be
defined as first derivatives of the rotational magnetic induction at the position
of nucleus K with respect to a component of the rotational angular momentum
J~ of the nuclei

j,J ~ ~
µ g ∂B ( R ; J)

K N K α K
Cαβ =

2π ∂Jβ

~
|J |=0
~j,J (R
The rotational magnetic induction, B ~ K , J),
~ at the position of a nucleus K
can be probed by the magnetic moment of this nucleus, m ~K
The change in energy is
K
X 2π X K K
∆E(m ~
~ , J) = − mK
α
j,J ~
Bα ~
(RK ; J) =− mα Cαβ Jβ
α
µN gK
αβ
Spin-rotation tensor defined as second derivative of the energy with respect to

a component of the rotational angular momentum J~ of the nuclei and a
component of the nuclear magnetic moment m ~K

K
~ m
µN gK ∂ 2 E(B, ~ K)

Cαβ = −
2π ∂mK ∂J

α β ~ K
|J|=|m
~ |=0
Taking the derivative of the electronic energy gives only the induced or
Born-Oppenheimer breakdown contribution to the spin-rotation tensor.

The spin-rotation tensor, CK , is the coupling tensor for the coupling of the
rotational angular moment of the molecule J~ with the spin I~K of the nuclei
{K}, which gives rise to an additional contribution to the rotational energy.
This is normally expressed in terms of an effective rotational Hamiltonian as
rot Jˆx2 Jˆy2 Jˆz2 2π X ~ˆK K ~ˆ

Ĥ = + + − I C J
2Ixx 2Iyy 2Izz ~ K

6.6 QM Expression for the Spin Rotation Tensor - 1
The derivation of the quantum mechanical expressions for the spin rotation
tensor is completely analogous to the one for the rotational g tensor.
The rigid contribution consists again of a nuclear and an electronic term
 "
K,rig µN gK e µ0
X ~L − R
(R ~K)
Cαβ = ~
ZL (RL − RK ) ~ δαβ
2πIββ 4π  ~ ~
| RL − RK | 3
L6=K
#
(RL,β − RK,β )
−(RL,α − RK,α )
~L − R
|R ~ K |3
N
" # )
(0)
X (~ri − R~K) (ri,β − RK,β ) (0)
− hΨ0 | ~
(~ri − RK ) δαβ − (ri,α − RK,α ) | Ψ0 i
~ K |3
| ~ri − R ~ K |3
| ~ri − R
i
For the derivation of the induced contribution we will start from the definition as
second derivative of the electronic energy in the presence of a nuclear
magnetic moment and the molecular rotation.

The corresponding vector potentials lead to two first-order perturbation

Hamiltonians and a new second-order order perturbation Hamiltonian
(2)
X mK J
Ĥ = Ôαβ (R ~ K ) mK
~ CM , R α (I
−1 ~
J)β
αβ
where the perturbation operator is given as

N
mK J mK J ~
X
Ôαβ ~ CM , R
(R ~K) = ~K)
ôi,αβ (RCM , R
i
" #
eµ0 X ~K)
(~ri − R (ri,β − RK,β )
= − ~
(~ri − RCM ) · δαβ − (ri,α − RCM,α )
4π i ~ K |3
| ~ri − R ~ K |3
| ~ri − R
This operator is very similar to the one for the diamagnetic contribution to the
nuclear magnetic shielding tensor but with one of the electronic position
vectors defined with respect to the nuclear centre of masses R ~ CM and not with
respect to the arbitrary gauge origin R~ GO

The second-order perturbation theory expression for a component of the spin

rotation tensor becomes then

K,ind
2
µN gK ∂ E(B,~ m K
~ )

Cαβ = −
2π ∂mK α ∂Jβ ~

~ K |=0
|J |=|m

(0) OP (0) (0) ~ CM ) | Ψ(0) i
µN gK X hΨ0 | ÔK,α | Ψn ihΨn | L̂β (R 0
= − (0) (0)
2πIββ E 0 − E n
n6=0
(0) (0) (0) (0)

!
~
hΨ0 | L̂β (RCM ) | Ψn ihΨn | ÔK,α | Ψ0 iOP
+ (0) (0)
E0 − En
i
(0) mK J ~ ~K)|Ψ i (0)
+ hΨ0 | Ôαβ (RCM , R 0
The induced contribution consists again of a paramagnetic or sum-over-states

contribution and a diamagnetic or ground state expectation value term

They can be combined with the contribution from the rigid charges
K K,para K,dia K.nuc
Cαβ = Cαβ + Cαβ + Cαβ
 h i
(0) OP (0) (0) P ~ (0)
 hΨ0 | ÔK,α | Ψn ihΨn | i (~ ~i | Ψ0 i
ri − RCM ) × p
µN gK eµ0  4π X β
= − (0) (0)
2πIββ 4π  eµ0 n6=0 E − En
0
h i
(0) P ~ CM ) × p (0) (0) OP | Ψ(0) i
hΨ0 | i (~
ri − R ~i | Ψn ihΨn | ÔK,α 0
4π X β
+ (0) (0)
eµ0 n6=0 E0 − En
N
" #
~K) (ri,β −RK,β )
(0) ~ CM ) (~ ri − R (0)
X
−hΨ0 | ~ K −R
(R δαβ −(RK,α −RCM,α ) | Ψ0 i
|~ ~ K |3
ri − R |~ ~ K |3
ri − R
i
" #
X ( ~L − R
R ~K) (R L,β − RK,β ) 
− ZL (R ~L − R~K) δ − (RL,α − RK,α )
| R~L − R ~ K |3 αβ |R~L − R~ K |3 
L6=K
These are again a paramagnetic term, a new diamagnetic like term and a
nuclear contribution

The paramagnetic can be expressed as a linear response function
K,para µN gK 2me OP
Cαβ = hh ÔK,α ; m̂lβ (R
~ CM ) iiω=0
2πIββ e
The new diamagnetic contribution can again be written as a sum over all
states
N
" #
(0)
X (~ri − R ~K) (ri,β − RK,β ) (0)
hΨ0 | ~ ~
(RK − RCM ) δαβ − (RK,α − RCM,α ) | Ψ0 i
| ~ri − R~ K |3 | ~ri − R ~ K |3
i
 h i
(0) OP (0) (0) P ~ ~ (0)
 hΨ 0 | Ô K,α | Ψ n ihΨ n | i ( R K − R CM ) × p
~ i | Ψ 0 i
4π X

β
= (0) (0)
eµ0  E − En 0
n6=0
h i 
(0) P ~K − R
~ CM ) × p (0) (0) OP (0)
hΨ0 | i (R ~i | Ψn ihΨn | ÔK,α | Ψ0 i 

β
+ (0) (0)
E0 − En 

This replaces in the paramagnetic term the dependence of the orbital angular
momentum operator on the nuclear centre of mass RCM by a dependence on
the position of nucleus K. Molecular Electromagnetism A Computational Chemistry Approach – p.31/58
An alternative expression for the spin rotation tensor is thus

K K,nuc K,el
Cαβ = Cαβ + Cαβ
"
µN gK eµ0 X ~L − R
(R ~K)
= ~ ~
ZL (RL − RK ) δαβ
2πIββ 4π ~ ~
| RL − RK |3
L6=K
#
(RL,β − RK,β )
−(RL,α − RK,α )
~L − R
|R ~ K |3
"
µN gK X hΨ(0)
0 | Ô OP
K,α | Ψ
(0)
n ihΨ
(0)
n | L̂β ( ~ K ) | Ψ(0) i
R 0
− (0) (0)
2πIββ E0 − En
n6=0
(0)
#
~ K ) | Ψ(0)
hΨ0 | L̂β (R
(0) OP (0)
n ihΨn | ÔK,α | Ψ0 i
+ (0) (0)
E0 − En
which consists of a nuclear and an electronic sum-over-states contribution.

The electronic term can be expressed in terms of linear response functions
K,el µN gK 2me OP
Cαβ = hh ÔK,α ; m̂lβ (R
~ K ) iiω=0
2πIββ e
The expression for the electronic contribution to the spin rotation tensor is
proportional to the paramagnetic contribution to the nuclear magnetic shielding
~ GO = R
tensor, if R ~K
K,el me gK ~ K,para ~
Cαβ = σ (R K )
mp 2πIββ αβ
This is of great importance for NMR spectroscopy because it allows us to

determine the absolute shielding tensor σ K from a measured spin rotation
tensor and a calculated diamagnetic contribution
dia ~ me gK ~ K K,nuc

σαβ = σαβ (R K ) + Cαβ − Cαβ
mp 2πIββ
This is the only way to determine semi-experimental, absolute shielding

constants, as one can only obtain differences in the shielding constants, i.e.
chemical shifts, from NMR spectra.
However, one has to take care of the vibrational corrections in the measured
spin rotation tensors and secondly NMR spectra are normally measured in the
liquid phase so that solvent effects would have to be considered as well.
6.7 Non-adiabatic Reduced Masses - 1
So far we have studied Born-Oppenheimer-breakdown corrections to

molecular properties, i.e. the effect of the coupling between nuclear and
electronic motion on the electronic energies.
Now we will turn our attention to the effect of this coupling on the motion of the
nuclei and will discuss Born-Oppenheimer-breakdown corrections to the
rotational and vibrational energies.
We will illustrate it for a diatomic molecule AB, where there is only one
vibrational mode which involves changes in the internuclear distance
R = |R ~A − R ~ B | and which we have placed along the z-axis.
After transformation to nuclear centre of mass coordinates and separation of

the translation of the whole molecule the field-free Hamiltonian of a diatomic
molecule in an electronic ground state of symmetry 1 Σ+ becomes
(0)
Ĥnuc,e = Ĥ (0) + Ĥ ′
Ĥ (0) is the Born-Oppenheimer molecular field-free electronic Hamiltonian with

all position vectors defined relative to the centre of nuclear masses RCM .
The remaining three terms in the Hamiltonian are:
~2 ∂ 2 1 h
ˆ ˆ
i2 1
~î · p
~ˆj
X
′
Ĥ = − + ~ ~ ~
J − L(RCM ) + p
2µn ∂R 2 2µn R 2 2(mA + mB ) i,j
the kinetic energy operator for the vibrational motion of the two nuclei of
masses mA and mB
the kinetic energy operator for the rotation of the nuclei about their centre of
mass
the mass polarization term.
The nuclear reduced mass µn is defined as
mA mB
µn =
mA + mB
ˆ
J~ is the angular momentum operator for rotation of the whole molecule about
~ˆ R
the molecular centre of mass and L( ~ CM ) is the operator for total angular
momentum of the electrons but now with respect to the nuclear centre of mass.

The eigenfunctions of the Hamiltonian, the molecular wavefunctions

(0)
Φk,vJ (R, θ, φ, {~ri }), are functions of both the electronic coordinates ~ri , the
internuclear distance R and the two rotation angles θ and φ.
Approximations to them are normally obtained by an approximate separation
of nuclear and electronic coordinates.
In the first step one solves the electronic Schrödinger equation,
(0) (0) (0)
Ĥ (0) |Ψk ({~ri }; R)i = Ek (R) |Ψk ({~ri }; R)i
(0)
which yields a complete set of electronic wave functions {Ψk ({~ri }; R)}.
The molecular wave functions can then be expanded in this complete set
(0)
X (0) k,(0)
Φk,v,J (R, θ, φ, {~ri }) = Ψk ({~ri }; R) Θv,J (R, θ, φ)
k
Vibration-rotation spectra are normally interpreted in terms of an effective

Hamiltonian for vibration-rotational motion of the nuclei, obtained by taking the
(0)
expectation value of the molecular Hamiltonian Ĥnuc,e over the corresponding
(0)
electronic state |Ψ0 ({~ri }; R)i. Molecular Electromagnetism A Computational Chemistry Approach – p.36/58
However, the electronic energies depend on the internuclear distance R, which

is the coordinate of the vibrational motion.
Therefore the Hamiltonian is first subjected to a unitary transformation
ˆ = e−ıλŜ ĤeıλŜ = H̃
H̃ ˆ (0) + λH̃ ˆ (2) + · · ·
ˆ (1) + λ2 H̃
The hermitian operator Ŝ is chosen such that the transformed Hamiltonian H̃ ˆ

does not couple different electronic wave functions through first order, i.e.
(0) ˆ (1) | Ψ(0) ({~r }; R)i = 0
hΨ0 ({~ri }; R) | H̃ n i
In zeroth order one obtains the Born-Oppenheimer nuclear Hamiltonian

The second order effective vibration-rotational Hamiltonian for the electronic
ground state includes four additional contributions:
ef f ~2 ∂ 1 ∂ 1 ~ˆ 2
Ĥ = − [1 + β(R)] + [1 + α(R)] J
2 ∂R µn ∂R 2µn R2
(0)
+E0 (R) + E ad (R) + E nad (R)

The adiabatic contribution to the potential energy, E ad (R), is a first order

contribution to the energy and thus an expectation value over the wave
function of the considered electronic state, here the ground state
ad ~2 (0) ∂2 (0)
E (R) = − hΨ0 ({~ri }; R) | | Ψ 0 ({~ ri }; R)i
2µn ∂R2
1 (0) 2 2 (0)
+ hΨ ({~
r i }; R) | Lx + L y | Ψ 0 ({~ ri }; R)i
2µn R2 0
1
~î · p
~ˆj | Ψ0 ({~ri }; R)i
(0) (0)
X
+ hΨ0 ({~ri }; R) | p
2(mA + mB ) i,j
The non-adiabatic, E nad (R), contribution to the potential energy is a second

correction and involves thus a sum over other electronic, i.e. excited states.
The effect of these corrections is that the potential energy of the nuclei does
not only depend on the internuclear distance R but also on the relative
momenta of the nuclei.
Both energy correction terms, E ad (R) and E nad (R), are not related to any
molecular properties and will therefore not be discussed further.
The last two terms, β(R) and α(R), are non-adiabatic correction terms for the
nuclear reduced masses in the nuclear kinetic energy operators. They involve
also sum over excited states
2
(0) ∂ (0)
2 X hΨ0 ({~ri }; R) | −ı~ ∂R | Ψn ({~ri }; R)i
β(R) = − (0) (0)
µn n6=0 E0 (R) − En (R)
D E2
(0) (0)
~ CM ) Ψn ({~ri }; R)
X Ψ0 ({~ri }; R) L̂⊥ (R

2
α(R) =
µn R 2 (0)
E0 (R) − En (R)
(0)
n6=0
where it is assumed that the molecule is aligned along the z axis and therefore
L̂x = L̂y , which is denoted as L̂⊥ .
Both terms are second order in the coupling between the nuclear and the
electronic motion.

The rotational correction, α(R), is due to the coupling between the nuclear and
electronic angular momentum, whereas the vibrational correction, β(R) arises
due to a similar coupling between the linear momenta.
They are thus a consequence of the breakdown of the Born-Oppenheimer
approximation and are therefore often also called Born-Oppenheimer
breakdown (BOB) parameters.
Physically they represent the contribution of the electrons to the reduced
masses or the lagging behind of the electrons and one can therefore define
effective reduced masses for vibration and for rotation as
µn
µef
v,n
f
=
1 + β(R)
µn
µef
J,n
f
=
1 + α(R)
These corrections to the vibrational and rotational reduced nuclear masses

can be related to other molecular electromagnetic properties.

∂
Recall that −ı~ ∂R is a hermitian operator, i.e.
(0) (0)
(0) ∂Ψn ({~ri }; R) ∂Ψ0 ({~ri }; R) ∗
hΨ0 ({~ri }; R) | −ı~ i = hΨ(0)
n ({~
ri }; R) | −ı~ i
∂R ∂R
(0)
∂Ψ0 ({~ri }; R)
= −hΨ(0)
n ({~
ri }; R) | −ı~ i
∂R
where in the second step was used that the wavefunctions are real.
We can thus rewrite the non-adiabatic vibrational correction alternatively as
(0) ∂ (0) (0) ∂ (0)
2~2 X hΨ0 ({~
ri }; R) | ∂R | Ψn ({~ri }; R)ihΨn ({~ri }; R) | ∂R | Ψ0 ({~
ri }; R)i
β(R) = − (0) (0)
µn n6=0 E (R) − En (R)
0
Like all sum over states expressions we can also express the non-adiabatic
corrections to the reduced masses as linear response functions
~2 ∂ ∂
β(R) = − hh ; iiω=0
µn ∂R ∂R
1 ~ CM ) ; L̂⊥ (R
~ CM ) iiω=0
α(R) = 2
hh L̂⊥ (R
µn R
The non-adiabatic correction to the rotational reduced mass is proportional to

el
the electronic contribution to the rotational g factor gJ,n (R)
me el
α(R) = gJ,n (R)
mp
where we refer to the xx or yy component of the rotational g tensor of a

diatomic molecule aligned along the z axis as the rotational g factor gJ (R), i.e.
gJ (R) ≡ gJ,xx (R) = gJ,yy (R)
The relation might at first sight be surprising, because α(R) is quadratic in the
coupling between rotation of the nuclei and electronic motion, whereas gJ (R)
is only bilinear in the coupling with rotation and the interaction with an external
magnetic induction.
However, rotation and apparent magnetic fields are interrelated.
A corresponding magnetic property for the non-adiabatic vibrational correction
β(R) does not exist, as molecules do not acquire a magnetic moment during
their vibrations.
el
Nevertheless one defines the electronic contribution, gv,n (R), to a vibrational
g factor as proportional to the non-adiabatic vibrational correction to the
reduced mass
me el
β(R) = gv,n (R)
mp
The effective Hamiltonian for vibration-rotational motion of nuclei contains thus
effective reduced masses
µn
µef f
v/J,n = me el
1+ m g
p v/J,n
(R)
in which µn is the reduced mass in terms of nuclear masses.

Sometimes it is more convenient to express the effective Hamiltonian in terms
of atomic masses MA and MB .
This can approximately be achieved, if one assumes that the difference
between an atomic and a nuclear mass is equal to the total mass of the
electrons of the given atom, i.e. ZA me or ZB me , where ZA and ZB are the
atomic numbers of the two atoms.
The inverse of an atomic reduced mass µ can then be expressed in terms of

nuclear masses and atomic numbers as
1 mA + ZA me + mB + ZB me mA + ZA me + mB + ZB me
= = h i
µ (mA + ZA me )(mB + ZB me ) Z Z
mA mB 1 + me mAA + mBB + m2e ZA ZB
Neglecting the term proportional to m2e and expanding the denominator as

−1
ZA ZB ZA ZB
1 + me + ≈ 1 − me +
mA mB mA mB
leads to
ZA m2B + ZB m2A

1 mA + mB ZA + ZB ZA ZB
≈ 1 − me − m2e +
µ mA mB (mA + mB )mA mB mA mB mA mB
Neglecting once more the term proportional to m2e one obtains an approximate
relation between the atomic and nuclear reduced masses (error ≤ 10−7 u)
ZA m2B + ZB m2A

1 1
≈ 1 − me
µ µn (mA + mB )mA mB
The correction is proportional to the nuclear contribution to the rotational g

factor of a diatomic molecule expressed in terms of nuclear masses,
ZA m2B + ZB m2A
gnnuc = mp
(mA + mB )mA mB
which is independent of internuclear distance.

The change from nuclear to atomic masses “introduces" thus at term like the
nuclear contribution to the rotational g factor of diatomic molecules.
We define therefore also a “total vibrational g factor" of a diatomic molecule
el
gv,n = gv,n + gnnuc
where the subscript n indicates again the use of nuclear masses

The effective reduced masses in µef f
v/J,n is then approximated by the atomic
reduced mass µ and a correction from the rotational or vibrational g factor as
1 1 1 me
≈ + gv/J,n
µef f
v/J,n
µ µn mp
The correction terms still depends on nuclear masses:

1
It contains an obvious factor µn
.
1
The electronic contributions to the g factors include a second factor µn
.
The electronic contribution to the rotational g factor depends on the
masses, because the angular momentum operator L( ~ˆ R
~ CM ) is defined with
respect to the centre of nuclear masses.
This dependence is ignored here, although it allows the determination of
the electric dipolar moment from the rotational g factors of isotopic variants
as discussed in section 6.3.
The nuclear contribution to the g factors is defined in terms of the nuclear
masses.
nuc
2
gn µ
However, gnuc ≈ µn and µµn ≈ 1 and we can write
" 2 #
1 1 µ me 1 1 me
≈ 1+ gv/J ≈ ef f ≈ 1+ gv/J
µef f
v/J,n
µ µn mp µv/J,n µ mp

The effective vibration-rotational Hamiltonian for the electronic ground state of

a diatomic molecule, using atomic masses, can then finally be written as
2

~ ∂ 1 m e ∂ 1 m e ~ˆ 2
Ĥ ef f = − 1+ gv (R) + 1 + g J (R) J
2 ∂R µ mp ∂R 2µR2 mp
(0)
+E0 (R) + E ad (R) + E nad (R)
where the rotational and vibrational g factors in atomic masses are defined as
gv/J (R) = g nuc + gv/J

el
(R)
The common nuclear contribution is independent of the internuclear distance

nuc mp 2 2
g = ZA (RA,z − RCM,z ) + ZB (RB,z − RCM,z )
µR2
2 2
ZA MB + ZB MA
= mp
(MA + MB )MA MB

The electronic contributions are response functions

(0) ∂ (0) (0) ∂ (0)
mp 2~2 X hΨ0 (R) | ∂R | Ψn (R)ihΨn (R) | ∂R | Ψ0 (R)i
gvel (R) = − (0) (0)
me µ E0 (R) − En (R)
n6=0
mp ~2 ∂ ∂
= − hh ; iiω=0
me µ ∂R ∂R
2
(0) (0)
~ CM ) | Ψn (R)i
mp 2 X hΨ 0 (R) | L̂⊥ ( R
gJel (R) =
me µR2 (0)
E0 (R) − En (R)
(0)
n6=0
mp 1 ~ CM ) ; L̂⊥ (R
~ CM ) iiω=0
= 2
hh L̂⊥ (R
me µR

6.8 Partitioning of the g Factors - 1
In order to fit vibration-rotation spectra of several isotopologues of a diatomic

molecule AB it is convenient to partition the g factors into two isotopically
independent terms which are associated with one or the other nucleus
µ A µ B
gJ (R) = gJ (R) + gJ (R)
MA MB
µ A µ B
gv (R) = gv (R) + g (R)
MA MB v
A/B A/B
Physically gJ and gv correspond to the hypothetical situations, where the
molecule rotates around one of the atoms, i.e. the axis of rotation goes
through this atom, or where only one of the atoms moves during the vibration.
Mathematically they are obtained by changing the origin of the angular or
linear momentum operators.
A/B ~ CM as origin of the electronic
For gJ one replaces the centre of mass R
angular momentum operator by the position vectors of the two atoms, i.e. R~A
~ B.
or R
In the following we assume that the molecule AB is placed along the z-axis.
A/B
For gv one replaces the canonical momentum operator for vibration,
∂
P̂R = −ı~ ∂R , by two isotopically invariant operators
(RB,z − RCM,z ) p
P̂zA = P̂R + Ôz
R
(RA,z − RCM,z ) p
P̂zB = P̂R + Ôz
R

where O
The isotopically invariant contributions to the g factors are thus given as
mp mp
gJA (R) = ZA + 2
~ B ) ; L̂x (R
hh L̂x (R ~ B ) iiω=0
µ me µR
mp mp
gJB (R) = ZB + 2
~ A ) ; L̂x (R
hh L̂x (R ~ A ) iiω=0
µ me µR
mp mp
gvA (R) = ZA + hh P̂zA ; P̂zA iiω=0
µ me µ
mp mp
gvB (R) = ZB + hh P̂zB ; P̂zB iiω=0
µ me µ
where we have chosen L̂⊥ = L̂x . Molecular Electromagnetism A Computational Chemistry Approach – p.50/58
In the following we will derive this partitioning of the g factors.

~ CM in Lx (R
For gJB (R) we add and subtract R ~ B)
mp mp
gJB (R) = ZB + 2
hh L̂x (R~ CM ) ; L̂x (R ~ CM ) iiω=0
µ me µR
mp
p ~ CM ) iiω=0 +hh L̂x (R p
~ CM ) ; Ôy iiω=0

− 2
(RCM,z−RA,z ) hh Ôy ; L̂x (R
me µR
mp 2 p p
+ 2
(R CM,z −R A,z ) hh Ôy ; Ôy iiω=0
me µR
for gvB (R) we insert the defintion of P̂zB

mp mp
gvB (R) = ZB + hh P̂R ; P̂R iiω=0
µ me µ
mp
p p

− (RCM,z − RA,z ) hh Ôz ; P̂R iiω=0 + hh P̂R ; Ôz iiω=0
me µR
mp 2 p p
+ (R CM,z − R A,z ) hh Ôz ; Ôz iiω=0
me µR2
The second terms are the electronic contributions to the rotational and
vibrational g factors, which we replace by the total g factors.
The last three terms are static response functions involving the total electronic
momentum operator O ~ˆ
~ˆ p and another operator O.
These can be replaced by ground state expectation values of commutators of

~ˆ and the sum of the position operators of the electrons, O
the operator O ~ˆ r ,
mp mp
gJB (R) = gJ (R) + ZB − 2
ZA (RA,z − RCM,z ) + ZB (RB,z − RCM,z ) 2
µ µR2
2mp 1 (0) r ~ CM )] | Ψ(0) i
+ (R CM,z − R A,z ) hΨ | [Ô y , L̂ x ( R
µR2 ı~ 0 0
mp 2 1 (0) r p (0)
− (R CM,z − R A,z ) hΨ 0 | [Ô y , Ô y ] | Ψ 0 i
µR2 ı~
mp mp
gvB (R) = gv (R) + ZB − Z A (R A,z − R CM,z ) 2
+ Z B (R B,z − R CM,z ) 2
µ µR2
!
2mp ∂ X X ∂
− (RCM,z − RA,z ) h Ψ0 | r̂i,z | Ψ0 i + hΨ0 | r̂i,z | Ψ0 i
µR ∂R i i
∂R
mp 2 1 (0) r p (0)
− (R CM,z − R A,z ) hΨ 0 | [Ôy , Ôy ] | Ψ 0 i
µR2 ı~

Evaluating the commutators we can rewrite this as

" 2 2 #
mp RA,z − RCM,z RB,z − RCM,z
gJB (R) = gJ (R) + ZB − ZA − ZB
µ R R
2mp (0)
X (0)
− (R CM,z − R A,z )hΨ 0 | (r i,z − R CM,z ) | Ψ 0 i
µR2 i
mp 2
− (R CM,z − R A,z ) N
µR2
and
" 2 2 #
mp RA,z − RCM,z RB,z − RCM,z
gvB (R) = gv (R) + ZB − ZA − ZB
µ R R
2mp ∂
− (RCM,z − RA,z ) hΨ0 | Ôzr | Ψ0 i
µR ∂R
mp 2
− 2
(R CM,z − R A,z ) N
µR

Recalling the definition of the z-component of the electric dipole moment

~ CM , R) for the internuclear distance R and with the origin of the
µz ( R
coordinate system at the centre of mass
~ CM , R)
µz ( R = e [ZA (RA,z −RCM,z ) + ZB (RB,z −RCM,z )]
X
−eh0 | (ri,z −RCM,z ) | 0i
i
and using q for the total charge of the molecule wecan write
2
2mp ~ CM , R) RA,z − RCM,z + mp q RA,z − RCM,z
gJB (R) = gJ (R) − µz ( R
eµR R µ R
and
2
2mp ∂ ~ CM , R) RA,z − RCM,z + mp q RA,z − RCM,z
gvB (R) = gv (R)− µz ( R
eµ ∂R R µ R

Choosing the coordinate system such that RA,z − RCM,z = −R µ/MA , which
implies for a molecule of polarity + AB− that µz < 0, we can finally write
~ CM , R)
2mp µz (R µ
gJB (R) = gJ (R) + + mp q 2
eR MA MA
and
2mp ∂ ~ CM , R) + mp q µ
gvB (R) = gv (R) + µz ( R 2
eMA ∂R MA
The analogous derivations for nucleus A gives
~ CM , R)
2mp µz (R µ
gJA (R) = gJ (R) − + mp q 2
eR MB MB
and
2mp ∂ ~ CM , R) + mp q µ
gvA (R) = gv (R) − µz ( R 2
eMB ∂R MB
For a neutral diatomic molecule, i.e. q = 0, one can then trivially prove the
partitioning of the rotational and vibrational g factors.

On the other hand simply averaging the two isotopically invariant g factors
leads to another partitioning of the g factors

mp 1 1 mp µ µ
gJ (R) = gJirr (R) − ~ CM , R)
µz ( R − − q 2
+ 2
eR MA MB 2 MA MB

m p ∂ 1 1 m p µ µ
gv (R) = gvirr (R) − ~ CM , R)
µz ( R − − q 2
+ 2
e ∂R MA MB 2 MA MB
where we have defined two “irreducible” non-adiabatic contributions

irr 1h B A
i
gJ (R) = gJ (R) + gJ (R)
2

mp 1 ~ A ) ; L̂x (R
~ A ) iiω=0
= ZA + ZB + 2
hh L̂x (R
2µ me R

1 ~ B ) ; L̂x (R
~ B ) iiω=0
+ 2
hh L̂x (R
me R
irr 1h B A
i
gv (R) = gv (R) + gv (R)
2

mp 1
= ZA + ZB + hh P̂zA ; P̂zA iiω=0 + hh P̂zB ; P̂zB iiω=0
2µ me
These relations give some physical insight in the g factors. There are three
contributions:
from the overall charge of the molecule
one due to the electric dipole moment or gradient of the dipole moment with
respect to the internuclear distance
one irreducible non-adiabatic contribution consisting of the average of the
isotopically invariant g factors.
Interestingly there seems to be a parallelism to the gross selection rules for
rotation and vibration spectra, where a permanent electric dipole moment or a
change in the electric dipole moment under vibration is required.

Finally by subtracting the isotopically invariant g factors we can obtain

expressions
for the electric dipole moment
eRµ h B A
i
µz (R) = g (R) − gJ (R)
2mp J
and the gradient of the electric dipole moment with respect to the
internuclear distance R
∂ ~ CM , R) = eµ h
B A
i
µz ( R gv (R) − gv (R)
∂R 2mp
From a computational point of view these expressions are not interesting, but
they offer an alternative route to the experimental determination of the electric
dipole moment and its gradient.
The analysis of vibration-rotation spectra recorded without external electric
fields via a fit to the effective Hamiltonian will produce values for the isotopically
invariant g factors which then can be used in the expressions above.

Chapter 7
Stephan P. A. Sauer
sauer@kiku.dk


7. Frequency-Dependent and Spectral Properties
So far static electric and magnetic fields

Molecules are exposed to time-dependent fields in the interaction with
electromagnetic radiation
Some of the properties in this chapter like the frequency-dependent
polarizability are generalizations to time- or frequency-dependent fields of the
static properties
Other spectral properties like the vertical excitation energies, transition dipole
moments and properties derived from them cannot be defined as derivatives
of the ground state energy.

7.1 Time-Dependent Fields - 1
Solving Maxwell’s equations for the vector potential of a plane or linear

polarized electromagnetic wave oscillating with angular frequency ω gives
ω ı(~
k·~
r −ωt) ω∗ −ı(~
~ ~
A(~r, t) = A e + A e k·~r−ωt)
~
The wave or propagation vector ~k points in the direction of the propagation of

the wave and has length
ω 2πnr (ω)
|~k| = nr (ω) =
c λ
where c is the speed of light, λ the wavelength of the electromagnetic wave in
vacuum and nr (ω) the refractive index of the medium through which the wave
propagates.
The refractive index is the ratio of the speed of electromagnetic radiation in
vacuum to the speed in a medium.
The dependence of the refractive index on the frequency is called dispersion.
In vacuum the refractive index is therefore equal to 1.

The amplitude A~ω is in principle a complex vector perpendicular to the

propagation vector ~k
It can be chosen to be purely imaginary by an appropriate choice of origin of
the time variable.
The time dependence of the vector potential reduces then to
~ω sin(~k · ~r − ωt)
~ r , t) = ı2A
A(~
showing that the vector potential is then real.

The electric and magnetic fields of this plane monochromatic electromagnetic
wave are then
~ r, t)
E(~ = E~ω cos(~k · ~r − ωt)
~ r, t)
B(~ = ~ω cos(~k · ~r − ωt)
B
~ ω ⊥ ~k and B
For the amplitudes holds that E~ω = ı2ω A ~ω = ı2~k × A
~ω ⊥ ~k and
~ω ⊥ E~ω
B

A general pulse of coherent polychromatic electromagnetic radiation can be

described as superposition of monochromatic plane waves.
The vector potential and fields of such a pulse are then given as
Z ∞ Z ∞ h ~ i
~
~ r, t) =
A(~ ı2A~ sin(~k · ~r − ωt)dω =
ω
A ω
~ e ı( k·~
r −ωt)
−e −ı( k·~
r −ωt)
dω
0 0
Z ∞ Z ∞
ω 1 ω
h ~
ı( k·~
r ~k·~
r
i
~ r, t) =
E(~ E~ cos(~k · ~r − ωt)dω = E~ e −ωt)
+e −ı( −ωt)
dω
0 2 0
Z ∞ Z ∞
~ r, t) = ω
~ cos(~k · ~r − ωt)dω = 1 ω
~ e
h ~
ı( k·~
r −ωt) −ı( ~
k·~
r −ωt)
i
B(~ B B +e dω
0 2 0
Typical molecules have diameters of 1 to 100 Å which is thus the order of

magnitude of the maximal length of the vector ~r
The propagation vector ~k has length 2π
λ
The product ~k · ~r ≪ 1 for all types of electromagnetic radiation with longer

wavelengths than X-rays.

~
We can expand eık·~r in a power series and approximate it
~ 1
eık·~r = 1 + ı~k · ~r + (ı~k · ~r)2 + · · · ≈ 1
2!
This is the same as setting ~k = ~0 and implies that we ignore the spatial
variation of the vector potential across a molecule.
We can therefore write for the vector potential
Z ∞ Z ∞
~ 2 ~ ω ~ ω −ıωt ıωt
A(t) = A sin(ωt)dω = A e −e dω
0 ı 0
and for the electric field

Z ∞ Z ∞
~ ~ ω 1 ~ ω −ıωt ıωt
E(t) = E cos(ωt)dω = E e +e dω
0 2 0
~
The magnetic induction B(t), however, vanishes because it is the curl of the
~ω = ı2~k × A
vector potential or the amplitude B ~ ω is zero for ~k = ~0.

In this long-wavelength approximation with this purely time-dependent vector

potential we can then choose the velocity or length gauge for the vector.
Because the length gauge implies that the perturbation Hamiltonian becomes
−µ~ˆ · E(t)
~ and thus involves the electric dipole moment operator, the
long-wavelength approximation is often called dipole approximation.
~
Retaining also the second term ı~k · ~r in the expansion of eık·~r leads to a
spatially non-uniform vector potential and electric field as well as to a
non-vanishing magnetic induction
The wave interacts with molecules also via the magnetic dipole and electric
quadrupole operators.

The electric field vectors of left-circularly polarized and right-circularly

polarized are given as as
L L
ωn (ω) ωn (ω)
E~L (~r, t) = E ω ~ei cos r
~ek · ~r − ωt + ~ej sin r
~ek · ~r − ωt
c c
R R
ωn (ω) ωn (ω)
E~R (~r, t) = E ω ~ei cos r
~ek · ~r − ωt − ~ej sin r
~ek · ~r − ωt
c c
The two unit vectors ~ei and ~ej are perpendicular to each other and to the
direction of the propagation, i.e. ~ei ⊥ ~ek , ~ej ⊥ ~ek and ~ei ⊥ ~ej , where ~ek is a
unit vector in the direction of ~k.
The associated magnetic induction vectors are
L L
~L (~r, t) = Bω −~ei sin ωnr (ω) ~ek · ~r − ωt − ~ej cos ωnr (ω) ~ek · ~r − ωt
B
c c
R R
~R (~r, t) = Bω ~ei sin ωnr (ω) ~ek · ~r − ωt − ~ej cos ωnr (ω) ~ek · ~r − ωt
B
c c

For right-circularly polarized radiation the electric field vector E~R rotates
clockwise when looking into the oncoming wave, i.e. at the source of the
radiation.
Circular polarization of photons corresponds to the two possible projections of
the photon’s spin on the direction of propagation, Sz , called helicity.
Right-circularly polarized photons have ms = −1 and thus Sz = −~, while
left-circularly polarized photons have ms = 1.
Plane polarized radiation can then be expressed as a superposition of left- and
right-circulary polarized waves with the same refractive index,
nr (ω) = nR L
r (ω) = nr (ω), i.e.

~ r, t) = E~L (~r, t) + E~R (~r, t) = 2E ω ~ei cos ωn r (ω)
E(~ ~ek · ~r − ωt
c

7.2 Frequency-Dependent Polarizability - 1
Expansion of a component of the time-dependent electric dipole moment in

the presence of a pulse of coherent polychromatic electromagnetic radiation
within the dipole approximation to
~
µα (E(t)) = µα + µind ~
α (E(t))
XZ ∞
~
µα (E(t)) = µα + ααβ (−ω1 ; ω1 )Eβω1 cos(ω1 t) dω1
β 0
1X ∞ ∞
Z Z
+ βαβγ (−ω1 − ω2 ; ω1 , ω2 )
2 0 0
βγ
× Eβω1 cos(ω1 t)Eγω2 cos(ω2 t) dω1 dω2
+ ...
where ααβ (−ω1 ; ω1 ) and βαβγ (−ω1 − ω2 ; ω1 , ω2 ) are components of the

frequency-dependent electric dipole polarizability also called the dynamic
electric dipole polarizability and first hyperpolarizability tensors,
respectively.

The isotropic frequency-dependent electric dipole polarizability α(−ω1 ; ω1 ) is

the trace of the polarizability tensor
1X
α(−ω1 ; ω1 ) = ααα (−ω1 ; ω1 )
3 α
α(−ω1 ; ω1 ) is the molecular property underlying the refractive index nr (ω1 ) of

a macroscopic sample with number density N .
For a non-polar molecule, i.e. without permanent electric dipole moment, the
relation between them can be shown to be
v
N
u 1 + 2α(−ω1 ; ω1 ) 3ǫ
u
0 N
nr (ω1 ) = t N
≈ 1 + α(−ω1 ; ω1 )
1 − α(−ω1 ; ω1 ) 3ǫ 2ǫ0
0
We can derive a quantum mechanical expression for α(−ω1 ; ω1 ) using

time-dependent response theory.

We evaluate the time-dependent expectation value of the electric dipole

~
operator hΨ0 (E(t)) ~
| µ̂α | Ψ0 (E(t))i in the presence of a time-dependent electric
field.
Employing the length gauge implies that the time-dependent electric field
enters the Hamiltonian via the scalar potential
The perturbation Hamilton operator for the periodic and spatially uniform
electric field of the electromagnetic wave is then given as
(1)
X E
Ĥ (t) = − Ôβ Eβ (t)
β
∞ Eβω1 −ıω1 t ∞ Eβω1 −ıω1 t
X Z X Z
ıω1 t
= − µ̂β e +e dω1 = − µ̂β e dω1
0 2 −∞ 2
β β
The Fourier components of the operator and the field are then given as
ω
Ôβ··· = − µ̂β
Fβ··· (ω) = Eβω1

Insertion of these operators yields for the expansion of the time-dependent

dipole moment
~
hΨ0 (E(t)) ~
| µ̂α | Ψ0 (E(t))i
∞ Eβω1 −ıω1 t
Z
(0) (0)
X
= hΨ0 | µ̂α | Ψ0 i + hh µ̂α ; −µ̂β iiω1 e dω1 + · · ·
−∞ β
2
XZ ∞
(0) (0)
= hΨ0 | µ̂α | Ψ0 i − hh µ̂α ; µ̂β iiω1 Eβω1 cos (ω1 t) dω1 + · · ·
β 0
We can then identify the frequency-dependent polarizability tensor as linear

response function or polarization propagator
ααβ (−ω1 ; ω1 ) = −hh µ̂α ; µ̂β iiω1

Using the spectral representation of the polarization propagator we can

alternatively write
X hΨ(0) | µ̂α | Ψ(0) (0) (0)

n ihΨn | µ̂β | Ψ0 i
0
ααβ (−ω1 ; ω1 ) = − (0) (0)
n6=0 ~ω1 + E0 − En
X hΨ(0) | µ̂β | Ψ(0) ihΨ
(0)
| | Ψ
(0)
0 n n µ̂ α 0 i
− (0) (0)
n6=0 −~ω1 + E0 − En
X (En(0) − E (0) )hΨ(0) | µ̂α | Ψ(0) (0) (0)
0 0
= 2 (0) (0)
n6=0 (En − E0 )2 − ~2 ω12
which for ω1 = 0, i.e. for the static polarizability, reduces to the expression
obtained by static response theory or Rayleigh-Schrödinger perturbation
theory.

7.3 Optical Rotation - 1
Plane or linear polarized radiation can be expressed as the superposition of

left-circularly polarized E~L and right-circularly polarized waves E~R with the
same refractive index.
If the refractive indices for left- and right-circularly polarized radiation differ by
∆nr (ω1 ) = nL R
r (ω1 ) − nr (ω1 )
the superposition will still be a plane polarized wave
~ r, t)
E(~ = E~L (~r, t) + E~R (~r, t)
L !
R

ω1 nr (ω1 ) + nr (ω1 )
= 2E ω1 cos ~ek · ~r − ω1 t
2c

ω1 ∆nr (ω1 ) ω1 ∆nr (ω1 )
× ~ei cos ~ek · ~r + ~ej sin ~ek · ~r
2c 2c
but with a plane of polarization which is rotated compared to the original wave
by an angle
ω1 ∆nr (ω1 )
∆θ = ~ek · ~r
2c
This is called optical rotation.

A medium which has the property that left- and right-circularly polarized waves
propagate with different velocity, i.e. that their refractive indices differ, is called
a circularly birefringent medium.
Chiral molecules have this property, because they experience the spatial
variation of the electric field vector of left- and right-circularly polarized waves
as being of the same or opposite handedness as their own structure.
In the derivation of the molecular properties, which give rise to this effect, we
have to take the spatial variation of the electric field vector into account and
can thus not make the dipole approximation.
This implies that we have to include a contribution from the interaction with the
~ × E(~
curl of the time-dependent electric field, ∇ ~ r, t), to the expansion of the
induced dipole moment of a molecule in

However, Maxwell’s third equation relates the curl of the electric field vector to
~ r, t)/∂t of the magnetic induction
the time derivative ∂ B(~
~ r , t)
∂ B(~
~ ~
∇ × E(~r, t) = −
∂t
We can alternatively expand the induced dipole moment in the electric field
and the time-derivative of the magnetic induction of a monochromatic wave as
µL/R
α (E~L/R (t), B
~L/R (t))
= µα + µL/R,ind
α (E~L/R (t), B
~L/R (t))
X 1 ′ ~ L/R (t)
∂B
L/R β
X
= µα + ααβ (−ω1 ; ω1 ) Eβ (t) + Gαβ (−ω1 ; ω1 ) + ...
ω1 ∂t
β β
where G′ (−ω1 ; ω1 ) is the mixed frequency-dependent electric dipole

magnetic dipole polarizability tensor, whose isotropic value G′ (−ω1 ; ω1 ) is
again the trace of the tensor
′ 1X ′
G (−ω1 ; ω1 ) = Gαα (−ω1 ; ω1 )
3 α Molecular Electromagnetism A Computational Chemistry Approach – p.17/52
The refractive indices of a macroscopic sample with number density N for left-
and right-circularly polarized waves becomes then
N N
nL/R
r (ω1 ) ≈ 1 + α(−ω1 ; ω1 ) ∓ G′ (−ω1 ; ω1 )
2ǫ0 2cǫ0
We can express the difference in refractive indices
N
∆nr (ω1 ) = −G′ (−ω1 ; ω1 )
cǫ0
and the angle of rotation
ω1 G′ (−ω1 ; ω1 )N
∆θ = − ~ek · ~r
2ǫ0 c2
in terms of the molecular mixed frequency-dependent electric dipole magnetic
dipole polarizability G′ (−ω1 ; ω1 ).

In order to derive a quantum mechanical expression for the mixed dynamic

electric dipole magnetic dipole polarizability tensor we have to evaluate the
time-dependent expectation value of the electric dipole operator
hΨ0 (t) | µ̂α | Ψ0 (t)i in the presence of the time-dependent magnetic induction of
left- or right-circularly polarized radiation
~L/R (t)
B = Bω1 [∓ ~ei sin(ω1 t) − ~ej cos(ω1 t)]
Bω1 −ıω1 t ıω1 t Bω1 −ıω1 t ıω1 t
= ± ~ei e −e − ~ej e +e
2ı 2
For a closed-shell molecule the perturbation Hamiltonian operator becomes
then to first order
(1)
ĤL/R (t) = ~ˆ lB · B
−O ~L/R (t)
ˆ l Bω1 −ıω1 t ıω1 t ˆ l Bω1 −ıω1 t ıω1 t

= ∓m
~ · ~ei e −e +m
~ · ~ej e +e
2ı 2

The Fourier components of the operator and the field are given as
~ˆ1ω = ∓m
O ~ˆ l · ~ei
B ω1
F1 (ω1 ) = [δ(ω − ω1 ) − δ(ω + ω1 )]
2ı
~ˆ2ω = m
O ~ˆ l · ~ej
B ω1
F2 (ω1 ) = [δ(ω − ω1 ) + δ(ω + ω1 )]
2
Insertion of these operators yields for the expansion of the time-dependent
dipole moment
(0) (0)
hΨ0 (t) | µ̂α | Ψ0 (t)iL/R = hΨ0 | µ̂α | Ψ0 i
B ω1
+ hh µ̂α ; ∓m ~ˆ · ~ei iiω1 e 1 − hh µ̂α ; ∓m
l −ıω t
~ˆ · ~ei ii−ω1 e 1
l ıω t
2ı
B ω1
+ hh µ̂α ; m ~ˆ · ~ej iiω1 e
l −ıω 1 t
~ˆ · ~ej ii−ω1 e
+ hh µ̂α ; m l ıω 1 t
2
+···

The electric dipole moment operator µ ~ˆ is a hermitian and real operator

whereas m~ˆ l is hermitian and purely imaginary.
The linear response function of such operators is thus purely imaginary and
antisymmetric with respect to a change in the sign of the frequency ω1 .
We can therefore rewrite the expansion as
hΨ0 (t) | µ̂α | Ψ0 (t)iL/R

ω1
(0) (0) ˆ l −ıω1 t ıω1 t B
= hΨ0 | µ̂α | Ψ0 i∓ hh µ̂α ; m
~ · ~ei iiω1 e +e
2ı
ω1
B
~ˆ · ~ej iiω1 e 1 − e 1
l −ıω t ıω t
+ hh µ̂α ; m + ···
2
~ˆ l · ~ei iiω cos(ω1 t) ı Bω1
(0) (0)
= hΨ | µ̂α | Ψ i ± hh µ̂α ; m
0 0 1
~ˆ l · ~ej iiω1 sin(ω1 t) ı Bω1 + · · ·

− hh µ̂α ; m

Using the fact that the time derivative of the magnetic induction becomes
~L/R (t)
∂B
= ω1 Bω1 [∓ ~ei cos(ω1 t) + ~ej sin(ω1 t)]
∂t
we finally obtain for the expansion of the time-dependent dipole moment
ı ~L/R (t)
∂B
hΨ0 (t) | µ̂α | Ψ0 (t)i L/R
=
(0) (0)
hΨ0 | µ̂α | Ψ0 i − ˆ
hh µ̂α ; m l
~ iiω1 · +·· ·
ω1 ∂t
We can identify the frequency-dependent mixed electric dipole magnetic dipole

polarizability tensor as linear response function or polarization propagator
G′αβ (−ω1 ; ω1 ) = −ıhh µ̂α ; m̂lβ iiω1
As mentioned before the hh µ̂α ; m̂lβ iiω1 response function is purely

imaginary, but the G′ tensor is real.

Using the spectral representation of the polarization propagator we can

alternatively write
X hΨ(0) (0) (0) l (0)

0 | µ̂α | Ψn ihΨn | m̂β | Ψ0 i
G′αβ (−ω1 ; ω1 ) = −ı (0) (0)
n6=0 ~ω1 + E0 − En
X hΨ(0) l (0) (0) (0)
0 | m̂β | Ψn ihΨn | µ̂α | Ψ0 i
−ı (0) (0)
n6=0 −~ω1 + E0 − En
X ω1 hΨ(0) (0) (0) l (0)
0 | µ̂α | Ψn ihΨn | m̂β | Ψ0 i
= 2ı~ (0) (0)
n6=0 (En − E0 )2 − ~2 ω12
and can see that for ω1 = 0 the frequency-dependent mixed electric dipole
magnetic dipole polarizability tensor vanishes.

7.4 Excitation Energies & Transition Moments - 1
The polarization propagator has a singularity or pole, if the frequency ω of the

perturbation takes one of the following values
(0) (0)
En − E0
ω=±
~
(0) (0)
However, En − E0 = ∆En0 is the vertical electronic excitation energy
(0) (0)
from the unperturbed state Ψ0 to the unperturbed state Ψn .
Finding the poles of the polarization propagator is thus a way of directly
calculating the vertical electronic excitation energies of a system.
(0) (0)
The residuum corresponding to a pole, ~ωn0 = En − E0 , defined as
ω ω (0) ω (0)
lim ~(ω − ωn0 )hh Ôα··· ; Ôα··· iiω = hΨ0 | Ôα··· | Ψ(0) (0) ω
n ihΨn | Ôα··· | Ψ0 i
ω→ωn0
(0)
= |hΨ(0) ω
n | Ôα··· | Ψ0 i|
2
is the square of the norm of the electronic transition moment Mn0,α··· of

ω (0) (0)
operator Ôα··· from state |Ψ0 i to state |Ψn i.

ω
In the following we want to derive expressions for the operators Ôα··· .
For the general vector potential the first-order perturbation Hamiltonian takes
the form
N
e
~î
X
Ĥ (1) (t) = ~ ri , t) · p
A(~
me i
and the transition rate for absorption of a photon becomes

2 X
(1) π e
Wn0 = |Mn0,α··· |2 (Aω n0 2
α )
2 ~me α
2 X 2
N
π e

ı~ (0)
(0) X k·~
ri
= hΨn | e p̂i,α | Ψ0 i (Aω n0 2
α )
2 ~me α

i

However, instead of evaluating directly the transition moments of the

PN ı~k·~ri ˆ
i e ~i interaction operator one expands the exponential.
p

The first term gives again the dipole approximation, i.e. one ignores the spatial
variation of the vector potential. This reduces the expression for the transition
rate to 2 X
π e (0) p (0) 2

(1) ω 2
Wn0 = hΨn | Ôα | Ψ0 i (Aαn0 )
2 ~me α

where O
According to the off-diagonal hypervirial relation we can replace the transition
moment of O~ˆ p by a transition moment of O~ˆ r and obtain for the transition rate
π e 2 (0) 2 X
(0) r (0) 2

(1) (0) ω 2
Wn0 = En − E0 hΨn | Ôα | Ψ0 i (Aαn0 )
2 ~ α
or in terms of the electric dipole moment operator ~ ˆ

µ
π (0) 2 X 2
(1) (0) (0) (0) ω 2
Wn0 = 2 En − E0 hΨn | µ̂α | Ψ0 i (Aαn0 )
2~ α

In the dipole approximation one arrives at a transition moment of the dipole

(0) ˆ (0)
operator, hΨn | ~ µ | Ψ0 i, which is called the electric dipole transition
moment M ~ n0
E1
.
Going beyond the dipole approximation we consider now the next term in the
expansion ı~k · ~r. The next contribution to the transition moment becomes then
N
~k · ~ri p̂i,α | Ψ(0) i
X
hΨ(0)
n |ı 0
i
This contribution depends not only on the direction α of the polarization of the
radiation but also on the direction ~k of the propagation of the wave.
For the derivation of the detailed form of the operators in this contribution, we
will consider radiation traveling along the z-axis whose vector potential is
polarized in the x-direction, which implies that α = x and that the propagation
vector is ~k = (0, 0, ωn0
c
).

The contribution to the transition moment becomes then

N
ıωn0 (0) X (0)
Mn0,zx = hΨn | zi p̂i,x | Ψ0 i
c i
The operator zi p̂i,x is one half of the y component of the electronic angular
momentum operator.
We can write it as the sum of its symmetric and antisymmetric part
N N N
X 1X 1X
zi p̂i,x = (zi p̂i,x + xi p̂i,z ) + (zi p̂i,x − xi p̂i,z )
i
2 i 2 i
where the second, antisymmetric part is the y component of the total orbital
angular momentum operator of the electrons L.~ˆ
Recalling that xi and p̂i,z commute and using the commutator relations
between [~ri , Ĥ (0) ] we can rewrite the first, symmetric part as well giving
N N
X 1 me X (0) (0)
1
zi p̂i,x = zi [xi , Ĥ ] + [zi , Ĥ ] xi + L̂y
i
2 ı~ i 2
The contribution to the transition moment becomes then

N
ωn0 me (0) X (0) (0)

(0) ıωn0 (0) (0)
Mn0,zx = hΨn | zi xi Ĥ − Ĥ zi xi | Ψ0 i+ hΨn | L̂y | Ψ0 i
2c ~ i
2c
(0) (0)
However, the states Ψ0 and Ψn are eigenstates of the Hamiltonian Ĥ (0)
N
ωn0 me (0) (0)
X (0) ıωn0 (0) (0)
Mn0,zx = − En − E0 hΨ(0)
n | z x
i i | Ψ 0 i + hΨn | L̂y | Ψ0 i
2c ~ i
2c
2 N
ωn0 me (0) X (0) ıωn0 (0) (0)
= − hΨn | zi xi | Ψ0 i + hΨn | L̂y | Ψ0 i
2c ~ i
2c
rr
Defining a second electric moment operator Ôαβ as
N
X
rr
Ôαβ = r̂i,α r̂i,β
i
and using the definition of the magnetic dipole moment operator m̂lα

We can finally write a general αβ element of this contribution to the transition

moment as
E2
X M1
Mn0,αβ = Mn0,αβ + ǫαβγ Mn0,γ
γ
2
ωn0 me (0) rr (0)
X ıωn0 me (0) l (0)
= − hΨn | Ôαβ | Ψ0 i − ǫαβγ hΨn | m̂γ | Ψ0 i
2c ~ γ
2c e
We have defined implicitly a component of the electric quadrupole transition

moment ME2 ~ M1
n0 and of the magnetic dipole transition moment Mn0 .
All these transition moments, M~ n0

E1
, ME2 ~ M1
n0 , and Mn0 can be obtained as
residua of the appropriate polarization propagators, i.e.
the frequency-dependent dipole α(−ω; ω)
the quadrupole polarizability C(−ω; ω)
the frequency-dependent paramagnetic contribution to the magnetizability
ξ p (−ω; ω).

The intensity of a measured absorption band is usually reported in terms of

the dimensionless dipole oscillator strength which is defined in terms of the
electric dipole transition moments as
l 2 me (0) (0)

(0) ˆ (0) 2 2 me (0) (0)

(0) ~ˆ r (0) 2
fn0 = En −E 0 |hΨ n | µ
~ | Ψ 0 i| = En −E0 |hΨ n | O | Ψ0 i|
3 ~ 2 e2 3 ~2
Due to the appearance of the position operator, this is called the dipole
oscillator strength in the length representation.
One can consider the oscillator strength as the trace of a tensor of cartesian
components
l me (0) (0)

(0) (0)
fn0,αβ = 2 2 2 En − E0 hΨ0 | µ̂α | Ψ(0) (0)
~ e
Using the off-diagonal hypervirial relation one can define two alternative
formulations of the oscillator strength, a velocity representation
(0) ~ ˆ p (0) 2
v 2 1 |hΨn | O | Ψ0 i|
fn0 = (0) (0)
3 me En − E0
and a mixed representation
m 2 1 ˆ p (0)
(0) ~ ˆ (0) 2 1 (0) ~ˆ p (0) ˆ r (0)
fn0 = hΨ0 | O | Ψn ihΨ(0)
n | µ
~ | Ψ 0 i = hΨ 0 | O | Ψ n ihΨ (0) ~
n | O | Ψ0 i
3 ı~e 3 ı~
The mixed representation is particular interesting because it does not involve
the excitation energies explicitly. It can alternatively also be written in the
following two ways
m 2 1 (0) ˆ ˆ p (0)
fn0 = − µ | Ψ(0)
hΨ0 | ~ (0) ~
n ihΨn | O | Ψ0 i
3 ı~e
1 1 h (0) ~ˆ p (0) (0) ˆ (0) (0) ˆ (0) ˆ p (0)
(0) ~
i
= hΨ0 | O | Ψn ihΨn | ~ µ | Ψ0 i − hΨ0 | ~
µ | Ψn ihΨn | O | Ψ0 i
3 ı~e
For optically active, i.e. chiral, molecules the intensity of the bands of circular
dichroism (CD) spectra is expressed in terms of a rotational strength
(0) ˆ (0) ˆ l (0)
µ | Ψ(0)
Rn0 = ıhΨ0 | ~ n ihΨn | m
~ | Ψ0 i
It can be calculated as residuum of the mixed frequency-dependent electric

dipole magnetic dipole polarizability tensor G′ (−ω; ω) and thus of the
hh µ̂α ; m̂lβ iiω polarization propagator. Molecular Electromagnetism A Computational Chemistry Approach – p.32/52
The calculation of electronic vertical excitation energies ∆En0 = ~ωn0 and

corresponding transitions moments or oscillator strengths fn0 from the linear
response functions or polarization propagators is a very interesting alternative
to the usual approach because it is done in a direct way.
(0) (0)
Neither the wavefunctions |Ψ0/n i nor the energies E0/n of the initial state 0 or
final state n have to be calculated explicitly in order to obtain these spectral
properties.
∆En0 and fn0 are obtained directly as poles and residues of the polarization
propagator.
The response theory approach is therefore predestinate to the calculation of
electronic spectra.

Finding the poles of the polarization propagator

implies
finding the values of
the frequency ω for which the matrix E [2] − ~ωS [2] becomes singular.
This could in principle be done by a pole search where one tries to determine
the frequency of the pole by repeatedly evaluating the response function.
However, this is cumbersome and unnecessary because singularity of this
matrix is also the necessary condition for that the set of linear equations

[2] [2]
E − ~ωS X=0
has a non-trivial solution for X, i.e. X 6= 0.

This, on the other hand, is nothing else than the generalized eigenvalue
equation for the electronic hessian matrix E [2]
E [2] X = ~ωS [2] X

Finding the poles of the propagator corresponds therefore to solving the

generalized eigenvalue problem for the electronic Hessian matrix or the
principal propagator, which is written out here in more detail
   
(0) † (0) (0) (0) † (0)
hΨ0 | [ĥi , [Ĥ , ĥj ]] | Ψ0 i · · · hΨ0 | [ĥi , ĥj ] | Ψ0 i · · ·
 − ~ωn0 
.. .. .. ..
 
. . . .
 
n0
Xj
× .  = 0
..
The vertical excitation energies ∆En0 are thus obtained as eigenvalues ~ωn0
and {Xjn0 } are the elements of the corresponding eigenvectors.
The transition moments, finally, can be calculated from the eigenvectors
(0) (0)
{Xjn0 } and the property gradient vectors T j (Ôα ) = hΨ0 | [h†j , Ôα ] | Ψ0 i as
(0) (0) (0)

X
hΨ(0)
n | Ôα | Ψ0 i = Xjn0 hΨ0 | [ĥ†j , Ôα ] | Ψ0 i
j

7.5 Dipole Oscillator Strength Sums - 1
The set of dipole oscillator strengths {fn0 } is often called the dipole oscillator
strength distribution (DOSD).
Summed over all excited states, bound as well as continuum states, they are
related to several other molecular properties as will be shown in the following.
One defines two types of energy-weighted moments of the dipole oscillator
strength distribution also called dipole oscillator strength sums
X k
(0)
S(k) = En(0) − E0 fn0
n6=0
X k
(0) (0)
L(k) = En(0) − E0 ln(En(0) − E0 )fn0
n6=0
Depending on whether one sums the oscillator strengths in their length, mixed
or velocity representation one obtains thus the sums in the three
representations.
We will only distinguish between the three representations when necessary by
adding the superscripts l, m or v.
One can define sums for the cartesian components of the dipole oscillator
strengths as
X k
(0)
Sαβ (k) = En(0) − E0 fn0,αβ
n6=0
X k
(0) (0)
Lαβ (k) = En(0) − E0 ln(En(0) − E0 )fn0,αβ
n6=0
Several dipole oscillator strength sums are related to other molecular

properties by so-called dipole oscillator strength sum rules.
The best known is the Thomas-Reiche-Kuhn sum rule which relates the S(0)
sum to the number of electrons N of the system, i.e.
X
S(0) = fn0 = N
n6=0

The frequency-dependent polarizability can be written in terms of the oscillator

strengths
l
~ 2 e2 X fn0,αβ
ααβ (−ω; ω) =
me n6=0 (En(0) − E0(0) )2 − ~2 ω 2
l
~ 2 e2 X fn0,αβ 1
= (0) (0) 2 ~2 ω 2
me (En − E0 ) 1− (0) (0) 2
n6=0 (En −E0 )
For
frequencies smaller
than the lowest excitation energy, i.e.
(0) (0)
~ω/(En − E0 ) < 1, we can expand the last term in a Taylor series

∞ l
~ 2 e2 X 2k
X fn0,αβ
ααβ (−ω; ω) = (~ω) (0) (0) 2k+2
me k=0 n6=0 (En − E0 )
or in terms of the dipole oscillator strength sums

∞
e2 ~ 2 X
ααβ (−ω; ω) = (~ω)2k Sαβ
l
(−2k − 2)
me
k=0
This is often called the Cauchy moment expansion of the

frequency-dependent polarizability
The sums S(k) for even but negative values of k are called Cauchy moments.
S(−2) in particular turns out to be proportional to the static polarizability
e2 ~ 2 l
ααβ (0; 0) = Sαβ (−2)
me
which is another well-known example for a dipole oscillator strength sum rule.
The other Cauchy moments, i.e. even and negative sums, S(−4), S(−6), · · · ,
describe the frequency dependence or dispersion of the frequency-dependent
polarizability

They can be defined either as even derivatives of the frequency-dependent

polarizability
m
l me 1 d
Sαβ (−m − 2) = ααβ (−ω; ω)
e2 ~2 ~m m! dω m ω=0
for m = 2, 4, 6, · · · being even positive numbers

or as both odd and even derivatives of the frequency-dependent polarizability
k
l me 1 1 d
Sαβ (−2k − 2) = lim ααβ (−ω; ω)
e2 ~2 2k ~2k k! ω→0 ω dω
for k = 1, 2, 3, · · · being positive numbers.

The positive even dipole oscillator strength sums can also be obtained as
derivatives of the frequency-dependent polarizability
k
l m e 1 d
Sαβ (2k) = (−1)k−1 2 2 k 2k lim ω 3 ω 2 ααβ (−ω; ω)
e ~ 2 ~ k! ω→∞ dω
for k = 0, 1, 2, 3, · · ·
Recalling −α(−ω; ω) = hh µ̂α ; µ̂β iiω , the even dipole oscillator strength
sums can also be expressed as derivatives of this polarization propagator, i.e.
k
l me 1 3 d
Sαβ (2k) = (−1)k lim ω ω 2 hh µ̂α ; µ̂β iiω
e2 ~2 2k ~2k k! ω→∞ dω
for k = 0, 1, 2, 3, · · ·
k
l me 1 1 d
Sαβ (−2k − 2) = − 2 2 k 2k lim hh µ̂α ; µ̂β iiω
e ~ 2 ~ k! ω→0 ω dω
for k = 1, 2, 3, · · ·
Similar relations between the even sums in mixed, S m , and velocity

representation, S v , and the hh µ̂α ; Ôβp iiω and hh Ôα
p
; Ôβp iiω polarization
propagators can also be derived.

The dipole oscillator strengths play also an important role in the description of
the interaction of molecules with beams of charged particles, i.e. ions.
A beam of ions with charge Z passing with velocity v through matter is
scattered by the medium molecules and loses part of its kinetic energy Ekin .
This is normally expressed in terms of the linear stopping power or energy
loss per unit path length x defined as
1 dEkin
S(v) = −
N dx
where N is the density of molecules in the target.
In the case of fast ions moving through a medium the main contribution to the
energy loss comes from the inelastic collision with the electrons of the
molecules in the medium, which will be exited or even ionized.

The simplest expression describing this process is the Bethe formula derived
via first-order perturbation theory
Ze2 N e2 2me v 2
S(v) = ln( )
me v 2 4πǫ20 I(0)
where N is the number of electrons in the target molecules and I(0) is called
the mean excitation energy of the target molecules.
The mean excitation energy is not a simple mean value of all electronic
(0) (0)
excitation energies (En − E0 ), but is defined in terms of the energy
weighted moments or sums of the dipole oscillator strength distribution as
P (0) (0)
L(0) n6=0 ln(En − E0 )fn0
ln I(0) = = P
S(0) n6=0 fn0

7.6 van der Waals Coefficients - 1
The purely classical, electrostatic contribution to the interaction energy

between two molecules separated by a large distance is given in terms of the
electric moments of molecules.
The contribution from quantum mechanical dispersion or London forces, i.e.
disp
the dispersion energy EAB , can be related to the frequency-dependent
polarizabilities of the two interacting molecules.
This is in line with the physical interpretation of the dispersion forces as arising
from the interaction of induced dipole moments, which implies that both charge
distributions are perturbed by their interaction.
The dispersion energy can be derived by perturbation theory.
The perturbation Hamiltonian consists of the interaction potential of the two
charge distributions.
One expands both charge distributions in multipole series and keeps the first
term.

For two uncharged molecules A and B separated by a distance |R ~ AB | the

perturbation Hamiltonian consists then of the dipole-dipole interaction term
" #
(1) 1 ˆ ˆ ~Â · R
3(µ ~ AB )(R ~ˆB )
~ AB · µ
ĤAB = ~A · µ
µ ~B − 2
4πǫ0 |RAB |3 RAB
Â and µ
where ~
µ ~ˆB are the dipole moment operators of the two molecules.
The unperturbed Hamiltonian of the complex is then the Hamiltonian of two
non-interacting molecules, meaning the sum of the Hamiltonian operators of
the two separate molecules
(0)
The unperturbed complex energies EnA ,mB are the sum of the energies of the
separate molecules
En(0)
A ,mB
= En(0)
A
(0)
+ Em B
(0)
The unperturbed complex wavefunctions ΨnA ,mB are the product of the
corresponding molecular wavefunctions
Ψ(0) (0) (0)

nA ,mB = ΨnA ΨmB

The first-order correction to the energy of the ground state of the complex is
(1) (0) (1) (0) (1)
E0A ,0B = hΨ0A ,0B | ĤAB | Ψ0A ,0B i = hΨ(0) (0) (0) (0)
nA ΨmB | ĤAB | ΨnA ΨmB i
or after inserting the perturbation Hamiltonian

" #
(1) 1 3(~ ~ ~
µA · RAB )(RAB · µ
~B)
E0A ,0B = µ
~ A · µ
~ B − 2
4πǫ0 |RAB |3 RAB
This is the electrostatic interaction between the permanent dipole moments ~

µA
and µ
~ B of the two molecules.
The second-order energy correction becomes
(0) (1) (0) (0) (1) (0)
(2)
X hΨ0A ,0B | ĤAB | ΨnA ,mB ihΨnA ,mB | ĤAB | Ψ0A ,0B i
E0A ,0B = (0) (0)
nA ,mB 6=0A ,0B E0A ,0B − EnA ,mB
where the double sum runs over all complex states in which at least one of the
molecules is excited.

If only one molecule, e.g. molecule A, is excited, we obtain
X hΨ(0) (1) (0) (0) (1) (0)

0A ,0B | ĤAB | ΨnA ,0B ihΨnA ,0B | ĤAB | Ψ0A ,0B i
E0ind
A ,0B
= (0) (0)
nA 6=0A E0A ,0B − EnA ,0B
This is the induction energy contribution to the intermolecular forces and can
be shown to consist of the static polarizability of molecule A and the
permanent electric dipole moment of molecule B. We will not consider this
term any further here.
If both molecules are excited, we obtain
(0) (1) (0) (0) (1) (0)
X X hΨ0A ,0B | ĤAB | ΨnA ,mB ihΨnA ,mB | ĤAB | Ψ0A ,0B i
E0disp
A ,0B
= (0) (0)
nA 6=0A mB 6=0B E0A ,0B − EnA ,mB
Inserting the unperturbed complex energies and wavefunctions this gives

(0) (0) (1) (0) (0) (0) (0) (1) (0) (0)
X X hΨ0A Ψ0B | ĤAB | ΨnA ΨmB ihΨnA ΨmB | ĤAB | Ψ0A Ψ0B i
E0disp
A ,0B
= (0) (0) (0) (0)
nA 6=0A mB 6=0B E0A − EnA + E0B − EmB
(1)
The product of transition moments of ĤAB can be written more explicitly as
2
(0) (0) (1) (0) (0)
hΨ0A Ψ0B | ĤAB | ΨnA ΨmB i
1

~T (0) ˆ (0) (0) ˆ (0)
= R AB hΨ 0 | µ
~ A | Ψ n i · hΨ 0 | µ
~ B | Ψ mB i I 3
(4πǫ0 )2 |RAB |10 A
A B
2
(0) ˆ (0) (0) ˆ (0) ~ AB
− 3hΨ0A | ~ µA | ΨnA i ⊗ hΨ0B | ~ µB | Ψ mB i R
However, in the gas or liquid phase the molecules can have all possible
orientations with respect to each other.
Therefore, for the isotropic interaction between the two molecules one has to
average over all molecular orientations which reduces the absolute square of
the transition matrix element to
2 2
(0) ˆ (0) (0) ˆ (0)
µA | ΨnA i hΨ0B | ~
hΨ0A | ~ µB | Ψ mB i
2
(0) (0) (1) (0) (0)
hΨ0A Ψ0B | ĤAB | ΨnA ΨmB i =
24π 2 ǫ20 |RAB |6

The dipole-dipole contribution to the isotropic dispersion energy between two

neutral molecules is thus given as
2 2
(0) ˆ (0) (0) ˆ (0)
disp 1 X X hΨ0A | ~ µA | ΨnA i hΨ0B | ~ µB | Ψ mB i
E0A ,0B = −
24π 2 ǫ20 |RAB |6 (En
(0)
− E
(0)
0A ) + (E
(0)
m − E
(0)
0B )
nA 6=0A mB 6=0B A B
In order to evaluate this contribution one needs only all excitation energies and
corresponding transition dipole moments for molecule A and also for molecule
B. Both can be obtained from the poles and residues of a polarization
propagator for molecule A and separately for molecule B.
However, it is preferable to avoid the simultaneous summation over all states
and express the dispersion energy in terms of molecular properties.
This can be achieved by using the following integral transform
2 ∞
Z
1 x y
= 2 2 2 2
dz
x+y π 0 x +z y +z
for the denominator of the dispersion energy.

This gives
2
(0) (0) (0) ˆ (0)
1
Z ∞ X (EnA − E0A ) hΨ0A | ~
µA | Ψ n A i
E0disp
A ,0B
=− dz
12π 3 ǫ20 |RAB |6 0
(0)
(EnA − E0A )2 + z 2
(0)
nA 6=0A
2
(0) (0) (0) ˆ (0)
X (EmB − E0B ) hΨ0B | ~
µB | Ψ mB i
× (0) (0)
mB 6=0B (EmB − E0B )2 + z 2
√
Choosing z = −1~ω = ı~ω we can see that the summations correspond to
frequency-dependent polarizabilities for imaginary frequencies, giving
Z ∞
3~
E0disp
,0 = − dω α A
(−ıω; ıω) α B
(−ıω; ıω)
A B
16π 3 ǫ20 |RAB |6 0
Commonly the dipole-dipole contribution to the dispersion energy is written as
C6AB
E0disp
A ,0B
=−
|RAB |6

where the C6 or van der Waals dispersion coefficient is then defined as

Z ∞
3~
C6AB = 2
dω α A
(−ıω; ıω) α B
(−ıω; ıω)
16π 3 ǫ0 0
which is often called the Casimir-Polder formula.

This expression is for the isotropic interaction.
Similar but more complicated expressions have been derived for the situation,
where the orientation of the two molecules to each other is important.
The components of the polarizability for imaginary frequencies can be
obtained from correspondingly complex linear response functions
ααβ (−ıω; ıω) = −hh µ̂α ; µ̂β iiıω
Alternatively one can make use of the fact, that the frequency dependence of
the polarizability can be expressed in terms of dipole oscillator strength sums.
This expansion, however, converges only for frequencies below the first
(0) (0)
excitation energy, i.e. ~ω < min{En − E0 }.
The expansion can be extended beyond this convergence radius and into the
complex plane by using well known analytical continuation techniques based
on Padé approximants [n, m]α to the frequency-dependent polarizability α.
In particular the [n, n − 1]α Padé approximant to α(ıω) expressed by the dipole
oscillator strength sums can be used as lower bound
α(ıω) ≥ [n, n − 1]α
An upper bound can be obtained either from the [n, n]α Padé approximant
α(ıω) ≤ [n, n]α
or via the same type of Padé approximant [n, n − 1] as for the lower bound but
now to S l (0) − ω 2 α(ıω) instead of to α(ıω).
This approximant is usually denoted as [n, n − 1]β and an upper bound to
α(ıω) is then given as
S l (0) − [n, n − 1]β

α(ıω) ≤
ω2
Chapter 8
Stephan P. A. Sauer
sauer@kiku.dk


8 Vibrational Contributions to Molecular Properties
The molecular properties were derived for fixed nuclei

This is not a realistic description of a molecule: even at 0 K a molecule vibrates
Temperature dependence and isotope shifts are solely due to nuclear motion
corrections
The static polarizability will be used to illustrate this
We have to go back to the Hamiltonian which includes kinetic energy operators
for the nuclei
The corresponding eigenfunctions are the so-called vibronic wavefunctions
(0) ~ K }) with energy E (0)
Φkv ({~ri }, {R kv
(0) ~ K }) as unperturbed wavefunctions in the
We should have used Φkv ({~ri }, {R
derivation of expression for the molecular properties
However, we still want to make use of the Born-Oppenheimer approximation
and have therefore the choice of applying it before or after the effect of the
external perturbation is introduced in Hamiltonian.

8.1 Sum-over-states treatment - 1
The effects of the perturbation on the electronic and vibrational part of the
wavefunction are treated simultaneously
Perturbation theory is applied to the vibronic wavefunctions.
Vibronic wavefunction through first order in the presence of a perturbation Eβ
~ (0) (1) ~
Φ0v (E) = Φ0v + Φ0v (E)
(0) P
(0)
hΦnv′ |− β µ̂β + Ω̂Eβ Eβ | Φ0v i
(0)
X
= Φ0v + |Φnv′ i (0) (0)
nv ′ 6=0v E0v − Env′
Static polarizability using second order perturbation Theory
X hΦ(0)
0v | µ̂ α + Ω̂ E
α | Φ
(0)
ihΦ
(0)
| µ̂ β + Ω̂ E
β | Φ
(0)
0v i
nv ′ nv ′
ααβ = −2 (0) (0)
nv ′ 6=0v E0v − Env′

The summation can be broken in two parts

Summation over all vibrational states v ′ 6= v of the same electronic state
n = 0: a pure vibrational contribution
X hΦ(0) | µ̂ α + Ω̂ E
α | Φ
(0)
ihΦ
(0)
| µ̂ β + Ω̂ E
β | Φ
(0)
0v i
αvαβ 0v 0v ′ 0v ′
= −2 (0) (0)
v ′ 6=v E0v − E0v′
Summation over all other electronic states n 6= 0 and all their vibrational levels
v ′ : an electronic-vibrational polarizability
X hΦ(0) E (0) (0) E (0)

0v | µ̂α + Ω̂α | Φnv ′ ihΦnv ′ | µ̂β + Ω̂β | Φ0v i
αe,v
αβ = −2 (0) (0)
n6=0,v ′ E0v − Env′
We use now the Born-Oppenheimer approximation for the unperturbed

vibronic wavefunctions
Φ(0)
nv ({~
~ K }) = Ψ(0)
ri }, {R n ({~
~ K }) Θ(0)
ri }; {R ~
v ({RK })

One obtains for the two contributions

X X hΘ(0) (0) (0) (0) (0) (0) (0) (0)
v | hΨ0 | µ̂α | Ψn i | Θv ′ ihΘv ′ | hΨn | µ̂β | Ψ0 i | Θv i
αe,v
αβ = −2 (0) (0)
v ′ n6=0 E0v − Env ′
X hΘ(0) (0) E (0) (0) (0) (0) E (0) (0)

v | hΨ0 | µ̂α + Ω̂α | Ψ0 i | Θv ′ ihΘv ′ | hΨ0 | µ̂β + Ω̂β | Ψ0 i | Θv i
αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v ′
Question: Can we write the second contribution more compact?


X X hΘ(0) (0) (0) (0) (0) (0) (0) (0)
αe,v
αβ = −2 (0) (0)
v ′ n6=0 E0v − Env ′
X hΘ(0) (0) E (0) (0) (0) (0) E (0) (0)

αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v ′
Recognizing the permanent electric dipole moment
X hΘ(0) ~ (0) (0) ~ (0)

v | µα ({RK }) | Θv ′ ihΘv ′ | µβ ({RK }) | Θv i
αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v′
v
Question: What about ξαβ ?


X X hΘ(0) (0) (0) (0) (0) (0) (0) (0)
αe,v
αβ = −2 (0) (0)
v ′ n6=0 E0v − Env ′
X hΘ(0) (0) E (0) (0) (0) (0) E (0) (0)

αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v ′
Recognizing the permanent electric dipole moment
X hΘ(0) ~ (0) (0) ~ (0)

αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v′
No pure vibrational contribution to the magnetic properties of closed shell

molecules due to the quenching of the angular momentum.

For the pure vibrational polarizability
X hΘ(0) ~ (0) (0) ~ (0)

αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v′
(0)
one needs the energies, E0v′ , of all vibrational states of the electronic
ground state and the corresponding vibrational dipole transition moments.
This requires knowledge of the potential energy and electric dipole moment
surface of this single electronic state
For the electronic-vibrational polarizability
X X hΘ(0) (0) (0) (0) (0) (0) (0) (0)

αe,v
αβ = −2 (0) (0)
v ′ n6=0 E0v − Env ′
(0)
one needs to know not only all excited electronic states, Ψn and the
electronic dipole transition moments to them but also all the vibrational
(0)
states, Θv′ , of these excited states.
This makes this approach rather difficult to apply in actual calculations.
However, we can make the approximation that the differences between the
vibrational energies are much smaller than the differences between the
electronic energies, i.e.
(0) (0) (0) (0)
E0v − Env′ ≈ E00 − En0
This removes the dependence on the vibrational states from the denominator.
Consequently we can use in the numerator that the vibrational wavefunctions
form a complete set, i.e. that
X (0) (0)
1= |Θv′ ihΘv′ |
v′
We obtain a simplified expression for the electronic-vibrational polarizability
X hΘ(0) (0) (0) (0) (0) (0)

v | hΨ0 | µ̂α | Ψn ihΨn | µ̂β | Ψ0 i | Θv i
αe,v
αβ ≈ −2 (0) (0)
n6=0 E00 − En0

8.2 Clamped-nucleus treatment - 1
The effect of the perturbation on the electronic and nuclear motion is treated
sequentially
Firstly
The Born-Oppenheimer approximation is applied to the vibronic
~ K }),
wavefunction Φ0v ({~ri }, {R
~ K }) and
It is expressed a product of an electronic wavefunction Ψ0 ({~ri }; {R
a vibrational wavefunction Θv ({R~ K })
~ K }) = Ψ0 ({~ri }; {R
Φ0v ({~ri }, {R ~ K }) Θv ({R
~ K })
Secondly
The perturbation Eβ enters the electronic Hamiltonian.
Using perturbation theory one obtains a first-order electronic wavefunction
(0) P E

(0)
hΨn | − β µ̂β + Ω̂β Eβ | Ψ0 i
~ = Ψ(0) +Ψ(1) (E)
~ = Ψ(0) +
X
Ψ0 (E) 0 0 0 | Ψ(0)
n i (0)
E ({ ~ K }) − En(0) ({R
R ~ K })
n6=0 0

and a first order electronic energy
~ K }, E)
~ (0) ~ K }) + E (1) ({R
~ K }, E)
~
E0 ({R = E0 ({R 0
(0) (0) (0)
X X E
= ~
E0 ({RK }) + hΨ0 | − µ̂β Eβ | Ψ0 i − Ω̂β Eβ
β β
The last two terms are again the permanent electric dipole moment
E0 ({R ~ = E (0) ({R

~ K }, E) ~ K }) − µ ~ K }) · E~
~ ({R
0
This is the potential energy for the nuclear motion in the Born-Oppenheimer
approximation
" #
X 1
~ˆK2 + E0 ({R
p ~ K }, E)
~ |Θv (E)i
~ = E0v (E)|Θ
~ ~
v (E)i
K
2mK

The external electric field, Eβ , enters the nuclear Hamiltonian together with an
operator which is an expectation value over the electronic wavefunction but
depends on the nuclear position vectors {R ~ K}
" #
X 1
~ˆK2 + E0 ({R
(0) ~ K }) − ~ ~ K }) · E~ |Θv (E)i
~ = E0v (E)|Θ
~ ~
p µ({R v (E)i
K
2mK
The vibrational wavefunction is expanded in a perturbation series and is to

first-order given as
(0) ~ K }) · E~ | Θ(0)
~ = Θ(0) (1) ~ (0)
X (0) hΘv′ | −~
µ({R v i
Θv (E) v + Θv (E) = Θv + | Θv ′ i (0) (0)
v ′ 6=v E0v − E0v′

Polarizability using static response theory

∂ ~ (1)
~ | µ̂α + Ω̂Eα | Φ0v (E)i
ααβ = hΦ0v (E)
∂Eβ
∂ h (0) (1) ~ (0) (0) (0) (1) ~
= hΘv Ψ0 (E) | µ̂α | Ψ0 Θ(0) (0)
v i + hΘv Ψ0 | µ̂α | Ψ0 (E)Θv i
∂Eβ
+ hΘ(1) ~ (0) E (0) (0)

v (E)Ψ0 | µ̂α + Ω̂α | Ψ0 Θv i
i
(0) (0)
+hΘ(0)
v Ψ0 | µ̂α + Ω̂Eα | Ψ0 Θ(1) ~
v (E)i
where we have used that the first-order correction to a vibronic wavefunction is

given as
(1) ~ K }, E)
~ (1) ~ Θ(0)
~ K }, E) ~
Φ0v ({~ri }, {R = Ψ0 ({~ri }; {R v ({RK })
(0) ~ K }) Θ(1) ~ ~
+ Ψ0 ({~ri }; {R v ({RK }, E)

This gives two contributions: the pure vibrational polarizability
X hΘ(0) ~ (0) (0) ~ (0)

αvαβ = −2 (0) (0)
v ′ 6=v E0v − E0v′
and a vibrationally averaged electronic polarizability
X hΨ(0) | µ̂α | Ψ(0) (0) (0)

αe,av
αβ = −2 hΘ(0)
v |
0
(0) (0)
| Θ (0)
v i
~
E ({RK }) − En ({RK }) ~
n6=0 0
= hΘ(0) ~ ~ (0) ~
v ({RK }) | ααβ ({RK }) | Θv ({RK })i
The expression for the vibrational polarizability is thus the same as in the
sum-over-states treatment
The expression for the electronic contribution is the pure electronic
polarizability but averaged with the unperturbed vibrational wavefunction
(0) ~ K }) of the electronic ground state.
Θv ({R

The vibrationally averaged electronic polarizability
X hΨ(0) | µ̂α | Ψ(0) (0) (0)

αe,av
αβ = −2 hΘ(0)
v |
0
(0) (0)
| Θ (0)
v i
~
E ({RK }) − En ({RK }) ~
n6=0 0
differs from the approximate form of the electronic-vibrational polarizability of

the sum-over-states treatment,
X hΘ(0) (0) (0) (0) (0) (0)
v | hΨ0 | µ̂α | Ψn ihΨn | µ̂β | Ψ0 i | Θv i
αe,v
αβ ≈ −2 (0) (0)
n6=0 E00 − En0
where
the transition moments to each excited electronic state are averaged
(0) ~ K }) of
individually with the unperturbed vibrational wavefunction Θv ({R
the electronic ground state
the denominator consist of the difference between the energies of the
(0)
vibrational ground state of the excited electronic states En0 and of the
(0)
electronic ground state E00 .

8.3 Vibrational and Thermal Averaging - 1
For two-atomic molecules:

the vibrational or more precisely vibration-rotational averaging in the
clamped-nucleus treatment can be carried out numerically:
One obtains all vibrational wavefunctions and corresponding vibrational
energies supported by the potential energy surface of the electronic ground
state as numerical solution of the nuclear Schrödinger equation
2
2
~ d J(J + 1) (0) (0) (0)
− + + E 0 (R) | Θ v,J i = Ev,J | Θ v,J i
2µ dR2 R2
One calculates the polarizability pointwise as a function of the internuclear

distance R.
One calculates the electric dipole moment pointwise as a function of the
internuclear distance R.
One numerically calculates the vibrational ground state expectation value of
the polarizability and the vibrational transition moment of the dipole
moment.

For polyatomic molecules:

clamped-nucleus electronic polarizability expressed as equilibrium geometry
value and vibrational correction:
αe,av (0) ~ ~ (0) ~ ~ v

αβ = hΘv ({RK }) | ααβ ({RK }) | Θv ({RK })i = ααβ ({RK,e }) + ∆ααβ
Polarizability expanded in a power series in the normal coordinates {Qa }

X ∂ααβ 2
~ K }) = ααβ ({R
~ K,e }) + 1 X ∂ ααβ
ααβ ({R Qa + Qa Qb + · · ·
a
∂Q a 2 ∂Q a ∂Q b
ab
The vibrational correction is then obtained as a series of expectation values of

increasing powers of the normal coordinates {Qa }
2
X ∂ααβ 1 X ∂ ααβ
∆αvαβ = hΘ(0)
v | Q a | Θ (0)
v i+ hΘ(0) (0)
v | Qa Qb | Θv i+· · ·
a
∂Qa 2 ab ∂Qa ∂Qb

(0)
The field-free, vibrational wavefunctions Θv are found by solving the
unperturbed vibrational Schrödinger equation
" M #
1 ˆ
p
~K2
(0) ~ K })i = E (0) |Θ(0) ({R
~ K }) |Θ(0) ({R
X
+ Ek ({R ~ K })i
v,J k,v,J v,J
2 K mK
(0)
~ K }) is also expanded in a Taylor
in which the nuclear potential energy E0 ({R
series in the normal coordinates {Qa }
(0) ~ K }) (0) ~ K,e }) 1X 2 2 1X

E0 ({R = E0 ({R + ωa Qa + Kabc Qa Qb Qc + · · ·
2 a 6 abc
√
where ωa = Kaa and Kabc are the harmonic vibrational frequency and force
constant and the cubic force constant
(0) ~ K })
∂ 2 E0 ({R
Kab =
∂Qa ∂Qb
(0) ~ K })
∂ 3 E ({R
0
Kabc =
∂Qa ∂Qb ∂Qc
Terminating the expansion after the quadratic term gives a harmonic potential.
The vibrational Schrödinger equation in this harmonic approximation
2

1X ∂
−~2 + ωa
2 2
Q a |Θ (0,0)
v ({Q a })i = Ev
(0,0)
|Θ (0,0)
v ({Qa })i
2 a ∂Q2a
can then be separated in equations for each normal mode Qa and its
one-mode harmonic oscillator wavefunction ϑva (Qa )
2

1 ∂
−~2 + ωa
2 2
Qa |ϑva (Qa )i = Eva |ϑva (Qa )i
2 ∂Q2a
The total many-mode vibrational wavefunction and energy are given as

(0,0)
Y
Θv ({Qa }) = ϑva (Qa )
a
X 1

Ev(0,0) = va + ~ωa
a
2

Including the next term in the expansion

1X
Kabc Qa Qb Qc
6
abc
leads to an approximate anharmonic potential.

(0)
The field-free vibrational wavefunction Θv ({Qa }) is thus expanded in the
usual perturbation series
Θ(0) (0,0)
v ({Qa }) = Θv ({Qa }) + Θ(0,1)
v ({Qa }) + · · ·
(0,1)
The first-order correction Θv ({Qa }) is expanded in a complete set of
functions, which in general consists of all many-mode vibrational
wavefunctions obtained by exciting one, two, ... up to all one-mode harmonic
oscillator functions ϑva (Qa ) to all possible higher (or lower) vibrational levels.
In the following we will consider systems in the vibrational ground state and
thus calculate the zero point vibrational correction (ZPVC) to the
polarizability: ∆αZPVC = ∆αv=0

For the zero point vibrational correction we can restrict ourselves to corrections
(0,0)
to the ground state vibrational wavefunction Θv=0 ({Qa }),
We can restrict ourselves to many-mode vibrational wavefunctions
(0,0)
Θvb =1 ({Qa }), where only one of the one-mode harmonic oscillator function
ϑvb (Qb ) was excited to the vb = 1 level, while the other modes a 6= b remain in
the lowest level va = 0 , i.e.
(0,0)
Y
Θvb =1 ({Qa }) = ϑvb =1 (Qb ) ϑva =0 (Qa )
a6=b
The first-order correction to the ground state vibrational wavefunction is then

(0,1)
Θv=0 ({Qa })
(0,0) (0,0)
hΘvb =1 ({Qa }) | 61
P
(0,0) cde Kcde Qc Qd Qe | Θv ({Qa })i
X
=− |Θvb =1 ({Qa })i
b
~ ωb

Using the following expectation values of the normal coordinates over

one-mode harmonic oscillator wavefunctions
hϑva | Qa | ϑva′ i = 0 if va′ 6= va ± 1

r
~
hϑva | Qa | ϑva +1 i = (va + 1)
2ωa
~ 1
hϑva | Qa Qa | ϑva i = (va + )
ωa 2
we can evaluate the transition matrix element

r
1X (0,0) ~ ~ X Kbcc
Kcde hΘvb =1 ({Qa }) | Qc Qd Qe | Θ(0,0)
v ({Q a })i =
6 4 2ωb c ωc
cde
and obtain then for the first-order correction to the perturbed wavefunction
s
(0,1) 1 X (0,0) ~ X Kbcc
Θv=0 ({Qa }) = − |Θvb =1 ({Qa })i
4 b 2ωb3 c ωc

To the lowest non-vanishing order in perturbation theory the ZPVC is

X ∂α αβ (0) (0)
∆αZPVC
αβ = hΘv=0 | Qa | Θv=0 i(1)
a
∂Qa
2
1X ∂ ααβ (0) (0)
+ hΘv=0 | Qa Qb | Θv=0 i(0)
2 ab ∂Qa ∂Qb
Inserting the wavefunctions and evaluating the matrix elements the zero point
vibrational correction to the static polarizability tensor becomes
! 2
~ X 1 ∂α αβ
X K abb ~ X 1 ∂ α αβ
∆αZPVC
αβ = − +
4 a ωa2 ∂Qa ωb 4 a ωa ∂Q2a
b
The first term thus arises from the anharmonic term in the potential
The second term comes from the non-linear term in the expansion of the
polarizability
They are sometimes called the mechanical and electrical anharmonic
contributions.
The effect of temperature, T , can finally be included by Boltzmann averaging

the averaged electronic polarizability over several vibrational states of energy
(0)
Ev
P e,av −Ev(0)
v ααβ e
kT
e
ααβ (T ) = (0)
P −Ev
v e
kT
or after inserting the expansion of the vibrationally averaged polarizability

(0)
−Ev
P v
∆α αβ e kT
αeαβ (T ) ~ K,e }) +
= ααβ ({R v
(0)
P −Ev
v e kT
For the vibrational polarizability:

(0)
One needs to calculate the vibrational energies E0v
(0) (0)
And vibrational transition moments hΘv | µα | Θv′ i
Can be done by perturbation theory as described above

Chapter 9
Stephan P. A. Sauer
sauer.kiku@gmail.com


9. Short Review of Electronic Structure Methods - 1
A very brief review of closed-shell ab initio methods for the calculation of the
energy and wavefunction in the absence of perturbations will be given in order
to introduce the concepts and notation of the methods.
We will cover only those methods, whose application to the calculation of
electromagnetic properties will be discussed in the following sections.
We will discuss only unperturbed wavefunctions and energies and therefore
drop the superscript “(0) " for the unperturbed, i.e. field free, problem. It will be
only used for the field free Hamiltonian Ĥ (0) in this chapter.
In all these methods the approximations |Φ0 i to the ground state N -electron
(0)
wavefunction |Ψ0 i can be expressed as a linear combination of Slater
determinants {|Φn i}
(0)
X
|Ψ0 i ≈ |Φ0 i = |Φn i Cn0
n
The symbol Φ is used for approximate N -electron wavefunctions.

The approximate wavefunctions |Φ0 i are assumed to be properly normalized
always.
A Slater determinant is an antisymmetrized product

ψ1 (~
x1 ) ψ2 (~
x1 ) · · · ψN (~
x1 )

ψ (~
1 1 x2 ) ψ2 (~ x2 ) · · · ψN (~
x2 )
|Φn i = √ det . .. .. ..

N! ..

. . .

ψ1 (~
xN ) ψ2 (~
xN ) · · · ψN (~xN )
of one-electron molecular spin-orbitals {ψp (~

x)}
A molecular spin-orbital is the product of a spatial molecular orbital φp (~r)
and the appropriate abstract one-electron spin functions α(s) or β(s)

 α(s)
ψp (~
x) = φp (~r)
 β(s)
s denotes the abstract spin variable, ~x the spatial and spin variables together.
In restricted methods the same set of spatial molecular orbitals {φp (~r)} is
used for α and β spin-orbitals.
Molecular orbitals (MO) are often expanded in a basis of one-electron

functions, {χµ }, denoted by Greek indices and called atomic orbitals (AO)
X
φp = χµ cµp
µ
where {cµp } are the molecular orbital coefficients.

In the Mulliken or chemical notation an electron repulsion integral over spatial
orbitals is given as
2 Z ∗ ∗
φ (~
r 1 ) φj (~r2 ) φk (~r1 ) φl (~r2 )
Z
e i
φi (~r1 )φk (~r1 ) φj (~r2 )φl (~r2 ) =
d~r1 d~r2
4πǫ0 ~r1 ~r2 |~r1 − ~r2 |
The spin orbitals and therefore the Slater determinants depend on the
molecular orbital coefficients {cµp }.
The approximate methods differ then in how the energy is calculated and how
the molecular orbital coefficients, {cµp }, and the coefficients {Cn0 } in the
expansion in Slater determinants are determined.
We can distinguish between variational and non-variational methods.
Variational methods: SCF, MCSCF or full CI

Energy as an expectation value
E0 ({Cn0 }, {cµp }) = hΦ0 ({Cn0 }, {cµp }) | Ĥ (0) | Φ0 ({Cn0 }, {cµp })i
Wavefunction parameters {Cn0 } and {cµp } are obtained variationally
∂E0 ({Cn0 }, {cµp }) ∂E0 ({Cn0 }, {cµp })

= =0
∂Cn0 ∂cµp
Non-variational methods: MP or CC
Energy as an transition or asymmetric expectation value
E0 ({Cn0 }, {cµp }) = hΦ0 ′ ({Cn0 }, {cµp }) | Ĥ (0) | Φ0 ({Cn0 }, {cµp })i
Determinant expansion coefficients {Cn0 } by projecting the corresponding

Schrödinger equation against appropriate determinants
Molecular orbital coefficients {cµp } are variational at SCF level
∂E0SCF ({cµp })
=0
∂cµp Molecular Electromagnetism A Computational Chemistry Approach – p.5/46
9.1 Hartree-Fock Theory - 1
In closed-shell Hartree-Fock (HF) or self-consistent field (SCF) theory the

(0)
unperturbed many electron wavefunction |Ψ0 i is approximated by a single
Slater determinant
(0)
|Ψ0 i ≈ |ΦSCF
0 i
The spatial orbitals are solutions to the Hartree-Fock equations
fˆ(i) φp (~ri ) = ǫp φp (~ri )
The Hartree-Fock energy, E0SCF , is an expectation value
E0SCF = hΦSCF
0 | Ĥ (0) | ΦSCF
0 i
HF equations are derived from condition that E0SCF has to be stationary with
respect to a variation of the spin orbitals δψp
δE0SCF = 0
under the constraint that the orbitals have to remain orthonormal
hψp | ψq i = δpq
Hartree-Fock Hamiltonian F̂ is the sum of the one-electron Fock operators fˆ(i)
N N h i
fˆ(i) =
X X (0) HF
F̂ = ĥ (i) + v̂ (i)
i i
1 ˆ2 e2
PM ZK
where ĥ(0) (i) = p
~
2me i
− 4πǫ0 K |~ ~K|
ri −R
v̂ HF (i) is an effective one-electron potential, called the Hartree-Fock potential

occ Z
2 X
e 2 − P̂12
v̂ HF (i) = φ∗j (~r2 ) φj (~r2 ) d~r2
4πǫ0 j ~r2 |~ri − ~r2 |
where P̂12 is a permutation operator, which permutes electron 1 with 2.

v̂ HF (i) is the potential, which an electron experiences, when it moves in the
averaged field of all the other electrons.
In order to calculate this averaged field of the other electrons one needs
molecular orbitals φj (~r2 ) which describe the other electrons.
Obviously they are also solutions of the Hartree-Fock equations.
The Fock operator depends thus on its own eigenfunctions and the
Hartree-Fock equations have to be solved iteratively until self-consistency of
the Hartree-Fock potential v̂ HF (i) is obtained.
This is the reason that the Hartree-Fock method is also called the
self-consistent field method.
The eigenvalues of the Fock operator are the orbital energies ǫp
occ h
ǫp = hψp | fˆ| ψp i = hψp | ĥ (0)
X
| ψp i + ψp (~r1 ) ψp (~r1 ) ψj (~r2 ) ψj (~r2 )

j
i
− ψp (~r1 ) ψj (~r1 )ψj (~r2 ) ψp (~r2 )
or in terms of the spatial orbitals

occ h
ǫp = hφp | fˆ| φp i = hφp | ĥ(0) | φp i +
X
2 φp (~r1 ) φp (~r1 ) φj (~r2 ) φj (~r2 )

j
i
− φp (~r1 ) φj (~r1 )φj (~r2 ) φp (~r2 )

The N spin-orbitals with the lowest energy, or N/2 spatial orbitals, are then
used to construct the N -electron Slater determinant |ΦSCF
0 i, i.e. the
Hartree-Fock wavefunction or SCF determinant.
These spin or spatial orbitals are therefore called the occupied orbitals and
are denoted with the Latin indices i, j, k, . . ..
Solutions to the Hartree-Fock equation with higher orbital energies are called
unoccupied or virtual orbitals and are denoted by the indices a, b, c, . . . ,
General spatial orbitals have indices p, q, r, . . ..
The Hartree-Fock wavefunction |ΦSCF
0 i is an eigenfunction of the Hartree-Fock
Hamiltonian F̂
X N
SCF
F̂ |Φ0 i = ǫi |ΦSCF
0 i
i
The eigenvalue of this operator is the sum of the orbital energies of the
occupied orbitals and not the Hartree-Fock energy.

The Hartree-Fock energy is the expectation value of the full Hamiltonian Ĥ (0)
occ occ occ
X 1 XXh
E0SCF (0)

= hψi | ĥ | ψi i + ψi (~r1 ) ψi (~r1 ) ψj (~r2 ) ψj (~r2 )

i
2 i j
i
− ψi (~r1 ) ψj (~r1 )ψj (~r2 ) ψi (~r2 )
or using the definition of the orbital energies
occ occ
X 1 Xh i
E0SCF =

ǫi − ψi (~r1 ) ψi (~r1 )ψj (~r2 ) ψj (~r2 ) − ψi (~r1 ) ψj (~r1 )ψj (~r2 ) ψi (~r2 )
i
2 ij
In terms of the spatial orbitals the Hartree-Fock energy is given as

occ
X occ X
X occ h
E0SCF (0)

=2 hφi | ĥ | φi i + 2 φi (~r1 ) φi (~r1 )φj (~r2 ) φj (~r2 )
i i j
i
− φi (~r1 ) φj (~r1 )φj (~r2 ) φi (~r2 )
or using the definition of the orbital energies
occ occ h
X X i
E0SCF = 2

ǫi − 2 φi (~r1 )φi (~r1 )φj (~r2 )φj (~r2 ) − φi (~r1 )φj (~r1 )φj (~r2 )φi (~r2 )
i ij
In the Roothaan-Hartree-Fock approach the molecular orbitals are expanded

in atomic orbitals, {χµ }.
Solving the Hartree-Fock equations for this ansatz corresponds then to finding
the molecular orbital coefficients.
The variational condition for the Hartree-Fock energy is similarly then
∂E0SCF ({cµp })
= 0
∂cµp
which applied implies also

∂
|ΦSCF
0 ({cµp })i = 0
∂cµp

9.2 Excited Determinants and Excitation Operators - 1
Additional eigenfunctions of the Hartree-Fock Hamiltonian F̂ can be generated

by replacing some or all of the occupied orbitals in |ΦSCF
0 i by virtual orbitals.
They are normally classified according to their relation to the SCF determinant
|ΦSCF
0 i:
A determinant in which one of the orbitals in |ΦSCF
0 i, i.e. an occupied orbital
i, is replaced by another, unoccupied orbital a is called a singly excited
determinant |Φai i.
In doubly excited determinants |Φabij i two occupied orbitals i and j are
replaced by two unoccupied orbitals a and b.
and so forth up to N -tuply excited determinants, where N is the number of
electrons in the system.
These commonly used names are quite unfortunate, because a priori these
determinants are not related to the excited states of a molecule.
However, in the simplest possible treatment of excited states they can be
used as a crude approximation for the excited states of a molecules.

This can be expressed in terms of general excitation operators e ĥiµ , which act
on the Hartree-Fock determinant.
The excitation level is indicated by the subscript i. µ refers to a particular
operator of this general excitation level.
The whole set of excitation operators of level i is often collected in a column
vector denoted by ĥi .
Alternatively one often expresses excitation operators of a particular level in
†
terms of the single excitation or orbital rotation operators q̂ai , where the
subscript ai then refers to the involved virtual and occupied orbitals.
The effect of single, double, etc. excitation operators acting on the
Hartree-Fock determinant can then be expressed in both notations as
e
ĥ1µ |ΦSCF
0
†
i = q̂ai |ΦSCF
0 i = |Φai i
e
ĥ2µ |ΦSCF
0
† †
i = q̂ai q̂bj |ΦSCF
0 i = |Φab
ij i
..
.

The expansion of the general approximate N -electron wavefunction |Φ0 i in a

linear combination of Slater determinants can then be written as

|Φ0 i = C 1 + T̂1 + T̂2 + · · · + T̂i + · · · + T̂N |ΦSCF
0 i
where the excitation T̂i operators are defined as
a † † †
tab
P P
T̂1 = t
ai i q̂ai T̂2 = a>b ij q̂ai q̂bj
i>j
..
.
P e P e
T̂i = µ t i µ ĥiµ ... T̂N = µ t N µ ĥNµ
The effect of e.g. T̂1 acting on |ΦSCF

0 i is thus
SCF
X a † SCF X a a
T̂1 |Φ0 i = ti q̂ai |Φ0 i = ti |Φi i
ai ai
The expansion coefficients tai , tab

ij or in general tiµ are called correlation
coefficients or amplitudes depending on the method.

Analogously one expands also a general approximate N -electron bra state

hΦ0 | in a complete set of operators acting on the bra Hartree-Fock determinant

SCF
hΦ0 | = hΦ0 | 1 + Λ̂1 + Λ̂2 + · · · + Λ̂i + · · · + Λ̂N C
where the de-excitation operators Λ̂i are defined as
a
λab
P P
Λ̂1 = ai λi q̂ai Λ̂2 = a>b ij q̂ai q̂bj
i>j
..
.
d
λ Nµ d h Nµ
P P
Λ̂iµ = iµ λiµ hiµ ... Λ̂Nµ = Nµ
The q̂ai , q̂ai q̂bj or general d hiµ operators are the hermitian conjugate or adjoint
of the excitation operators
† †
q̂ai = (q̂ai )
† † †
q̂ai q̂bj = (q̂bj q̂ai ) ... d
ĥiµ = e ĥ†iµ

They are therefore the single de-excitation, double de-excitation and so

forth operators and their effect is best described by letting them act on the
Hartree-Fock determinant as a bra state
hΦSCF
0 | d ĥ1µ = hΦSCF
0 |q̂ai = hΦai |
hΦSCF
0 | d ĥ2µ = hΦSCF
0 |q̂bj q̂ai = hΦab
ij |
..
.
The expansion coefficients λiµ are normally just called “λ" amplitudes and will
not necessarily be related to the tiµ amplitudes.
The whole set of excitation operators {e ĥiµ } forms a complete set of operators,
meaning that acting on the Hartree-Fock wavefunction they generate all
excited determinants and form therefore a resolution of the identity
SCF SCF
Xe
1 = |Φ0 ihΦ0 | + ĥiµ |ΦSCF
0 ihΦSCF
0 | d ĥiµ
iµ

9.3 Multiconfigurational Self-Consistent Field Method - 1
The multiconfigurational self-consistent field method (MCSCF) is a

generalization of the Hartree-Fock method.
The wavefunction is now a linear combination of several Slater determinants or
configuration state functions, |Φn i.
(0) MCSCF
X
|Ψ0 i ≈ |Φ0 i= |Φn i Cn0
n
Configuration state functions are spin- or symmetry adapted linear

combinations of a few determinants.
The molecular orbital coefficients, {cµp }, as well as the configuration
expansion coefficients, {Cn0 }, are simultaneously determined variationally
∂E0MCSCF ({Cn0 }), {cµp })

= 0
∂cµp
∂E0MCSCF ({Cn0 }), {cµp })
= 0
∂Cn0

Alternatively one can make use of an exponential unitary transformation of the

orbitals and also of the configuration state functions in a given initial
wavefunction |ΦMCSCF
0 i
|ΦMCSCF
0 ({κpq }, {Sn0 })i = e−κ̂ e−Ŝ |ΦMCSCF
0 i
The operators Ŝ and κ̂ are defined as

X † ∗

Ŝ = Sn0 R̂n0 − Sn0 R̂0n
n6=0
X
† †
κ̂ = κpq q̂pq − q̂qp
p>q
†
where {R̂n0 } and {R̂0n } are state transfer operators to orthogonal
complement states |ΦMCSCF
n i to the |ΦMCSCF
0 i wavefunction
†
R̂n0 = |ΦMCSCF
n ihΦMCSCF
0 | and R̂0n = |ΦMCSCF
0 ihΦMCSCF
n |
†
and q̂pq are single excitation operators for now general orbitals p and q.

The same formulation can also be used for the single Slater determinant of the
Hartree-Fock wavefunction.
All coefficients Sn0 vanish then and only the orbitals are unitarily transformed
|ΦSCF
0 ({κai })i = e−κ̂ |ΦSCF
0 i
The antihermitian operator κ̂ is here defined as

X
† †
κ̂ = κai q̂ai − q̂ia
ai
†
where qai are the proper single excitation operators, which explains why they
are often called the orbital rotation operators.
Rotations between the virtual orbitals vanish obviously.
Rotations between the occupied orbitals leave |ΦSCF
0 i unchanged, because
they correspond to linear combinations of the columns in the determinant.
One of the advantages of this formulation is that the orthonormality of the
orbitals is always preserved due to the unitary transformation.
9.4 Configuration Interaction - 1
If one keeps the molecular orbital coefficients {cµp } fixed in an MCSCF

wavefunction and optimizes the energy only with respect to the configuration
coefficients {Cn0 }, i.e.
∂E0CI ({Cn0 })) ∂hΦCI

0 ({Cn0 }) | Ĥ
(0)
| ΦCI
0 ({Cn0 })i
= =0
∂Cn0 ∂Cn0
one obtains the configuration interaction (CI) method.
The wavefunction in the CI method takes then the same form as in MCSCF
with the difference that the molecular orbital coefficients {cµp } are kept fixed
(0) CI
X
|Ψ0 i ≈ |Φ0 i = |Φn i Cn0
n
Normally the CI wavefunction is expressed in terms of the Hartree-Fock

wavefunction and the excited determinants {|Φa···
i··· i}
CI SCF
X a a
X ab ab
X
|Φ0 i = |Φ0 iC0 + |Φi i Ci + |Φij i Cij + |Φabc abc
ijk i Cijk + · · ·
a a>b a>b>c
i i>j i>j>k

CI ({C
∂E0 n0 }))
Application of the variational condition, ∂Cn0
= 0, leads then to a set of
linear equations for the configuration coefficients, which are conveniently
written as the following matrix eigenvalue equation
ĤC = E CI C
Ĥ is the configuration interaction or CI matrix, i.e. the matrix of the

unperturbed molecular Hamiltonian Ĥ (0) in the basis of the Hartree-Fock and
excited determinants {|ΦSCF
0 i, |Φa···
i··· i}
C is the configuration coefficients collected in a column vector.

Solving the eigenvalue equation for the energy E CI one has to evaluate matrix
elements of the Hamiltonian Ĥ (0) between the Hartree-Fock Slater
determinant ΦSCF
0 and excited Slater determinants or between two excited
Slater determinants.

Here the so-called Slater-Condon rules become very useful, which state that
the matrix element between two Slater determinants,
which differ by only one spin-orbital, i.e. Φ and Φrp , where ψp is replaced by
ψr , is equal to
hΦ | Ĥ (0) | Φrp i = hψp | ĥ(0) | ψr i

Xh i
+ ψp (~r1 ) ψr (~r1 )ψs (~r2 ) ψs (~r2 ) − ψp (~r1 ) ψs (~r1 )ψs (~r2 ) ψr (~r2 )
s
where the summation over s runs over all spin-orbitals which are included in
both Slater determinants.
which differ by two spin-orbitals, i.e. Φ and Φrs
pq , where ψp is replaced by ψr
and ψq is replaced by ψs , is equal to
(0)
| Φrs

hΦ | Ĥ pq i = ψp (~r1 )ψr (~r1 ) ψq (~r2 )ψs (~r2 ) − ψp (~r1 )ψs (~r1 ) ψq (~r2 )ψr (~r2 )

which differ by more than two spin-orbitals, i.e. Φ and Φrs···

pq··· , vanish
hΦ | Ĥ (0) | Φrs···
pq··· i = 0
However, the matrix element hΦSCF

0 | Ĥ (0) | Φai i between the Hartree-Fock
determinant and any singly excited determinant is special and vanishes
hΦSCF
0 | Ĥ (0) | Φai i = hψi | fˆ| ψa i = 0
because it is equal to an occupied-virtual off-diagonal element of the Fock

matrix, which is zero, if the orbitals are solutions of the Hartree-Fock equations.
This is called the Brillouin theorem.

9.5 Møller-Plesset Perturbation Theory - 1
Møller-Plesset (MP) perturbation theory is an application of

Rayleigh-Schrödinger perturbation theory to the electron correlation problem.
The field free Hamiltonian Ĥ (0) is thereby partitioned in the Hartree-Fock
Hamiltonian F̂ and the so-called fluctuation potential V̂ as perturbation:
Ĥ (0) = F̂ + V̂
The latter is the difference between the correct electron-electron repulsion and
the effective electron repulsion given by the sum of the Hartree-Fock potentials
X X
V̂ = ĝ(i, j) − v̂ HF (i)
i<j i
The wavefunction is thus expanded in a perturbation series in V̂

(0) MP SCF MP1 MP2
|Ψ0 i ≈ |Φ0 i = C |Φ0 i + |Φ i + |Φ i + ···
where C is a normalization constant

The zeroth-order wavefunction is the single determinant SCF wavefunction
|ΦSCF
0 i
The energy is also expanded in a perturbation series in V̂

(0)
E0 ≈ E0MP = E MP0 + E MP1 + E MP2 + · · ·
= E0SCF + E MP2 + · · ·
The notation without and with the subscript “0" tries to distinguish between the
MP second-order correction to the energy E MP2 and the MP second-order
(MP2) energy E0MP2 = E0SCF + E MP2 .
The Hartree-Fock energy is the sum of E0SCF and E MP1
The zeroth-order energy is the eigenvalue of the Hartree-Fock Hamiltonian F̂
and thus only the sum of the orbital energies of the occupied orbitals.
One has to go at least to second order for the first correction to the
Hartree-Fock energy.
This can easily be seen remembering
E0SCF = hΦSCF
0 | Ĥ (0) | ΦSCF
0 i = hΦSCF
0 | F̂ + V̂ | ΦSCF
0 i
E MP1 = hΦSCF
0 | V̂ | ΦSCF
0 i
Corrections to the wavefunction are as usually expanded in eigenfunctions of

the unperturbed Hamiltonian F̂ as basis set of many-electron wavefunctions
These are the singly excited |Φai i, doubly excited |Φab
ij i, . . . N -tuply excited
determinants
The first-order correction to the wavefunction becomes then
MP1
X hΦn | V̂ | ΦSCF
0 i SCF
X ab ab
|Φ i= | Φn i SCF SCF
= T̂ 2 [1]|Φ0 i = t ij [1] |Φ ij i
n6=0
hΦ0 | F̂ | Φ0 i − hΦn | F̂ | Φn i a>b
i>j
because hΦn | V̂ | ΦSCF

0 i = 0 for all but doubly excited determinants |Φab
ij i.
The determinant expansion coefficients Cn0 of the approximate MP

wavefunction are called correlation coefficients. The “[1]" in T̂2 [1] or tab
ij [1]
indicates that they belong to the MP1 wavefunction.
The first-order doubles correlation coefficients are thus given as

φa (~r1 ) φi (~r1 ) φb (~r2 ) φj (~r2 )

tab
ij [1] =
ǫi + ǫj − ǫa − ǫb
The second-order MP correction to the energy is given as an asymmetric

expectation value
E MP2 = hΦSCF
0 | V̂ | ΦMP1 i = hΦSCF 0 | V̂ T̂2 [1] | ΦSCF
0 i
1 X n ab ab
o
= φi (~r1 ) φa (~r1 )φj (~r2 ) φb (~r2 ) 4 tij [1] − 2 tji [1]
2 ab
ij
The second-order MP correction to the wavefunction
MP2
X hΦn | V̂ | ΦMP1 i
|Φ i= |Φn i
n6=0
hΦSCF
0 | F̂ | ΦSCF
0 i − hΦn | F̂ | Φn i
!
MP1
hΦn | Φ i
−hΦSCF
0 | V̂ | ΦSCF
0 i
hΦSCF
0 | F̂ | ΦSCF
0 i − hΦn | F̂ | Φn i
consists of singly, doubly, triply and quadruply excited determinants

MP2
X a a
X ab ab
X abc abc
X
|Φ i= ti [2] |Φi i+ tij [2] |Φij i+ tijk [2] |Φijk i+ tabcd abcd
ijkl [2] |Φijkl i
ai a>b a>b>c a>b>c>d
i>j i>j>k i>j>k>l

The second-order singles correlation coefficients are given as


a 1 1 X n bc bc
o
ti [2] = √  φa (~r1 ) φb (~r1 )φj (~r2 ) φc (~r2 ) 4 tij [1] − 2 tji [1]
2 ǫi − ǫa jbc

X n ba ba
o
− φk (~r1 ) φi (~r1 )φj (~r2 ) φb (~r2 ) 4 tjk [1] − 2 tkj [1] 
jkb
The “[1]" in tab a

ij [1] or “[2]" in ti [2] indicate again that they belong to the MP1 or
MP2 wavefunction.

9.6 Coupled Cluster Theory - 1
In coupled cluster theory (CC) the N -electron wavefunction is given as

(0)
|Ψ0 i ≈ |ΦCC
0 i = e T̂
|Φ SCF
0 i
The cluster operator T̂ consists of single, double, . . . excitation operators
T̂ = T̂1 + T̂2 + · · · + T̂i + · · · + T̂N
The determinant expansion coefficients tai , tab

ij , tiµ and so forth of the singly,
doubly, i-tuply, etc. excited determinants are in coupled cluster theory called
singles, doubles, i-tuple etc. amplitudes.
Different CC methods are obtained by truncating the expansion of T̂ .
In the popular coupled cluster singles and doubles model (CCSD), the
cluster operator consists of T̂1 and T̂2 .
Inserting the coupled cluster wavefunction in the Schrödinger equation gives
the coupled cluster Schrödinger equation
Ĥ (0) |ΦCC CC CC
0 i = E0 |Φ0 i

The coupled cluster energy is obtained as a transition expectation value by

projecting the Schrödinger equation against the Hartree-Fock wavefunction
E0CC = hΦSCF
0 | Ĥ (0) | ΦCC
0 i
where it is used that |ΦSCF

0 i is orthogonal to all excited determinants.
Alternatively one could also have projected against hΦSCF
0 |e−T
E0CC = hΦSCF
0 | e−T̂ Ĥ (0) | ΦCC
0 i
Coupled non-linear equations for the amplitudes are obtained by projecting the
coupled cluster Schrödinger equation against hΦai |e−T̂ , hΦab
ij |e
−T̂
etc.
hΦai | e−T̂ Ĥ (0) | ΦCC

0 i = 0
hΦab
ij | e
−T̂
Ĥ (0) | ΦCC
0 i = 0 ... hΦSCF | d ĥNµ e−T̂ Ĥ (0) | ΦCC
0 i = 0
The right hand sides vanish because of
hΦSCF | d ĥiµ e−T̂ | ΦCC

0 i = hΦ | ĥiµ e−T̂ eT̂ | ΦSCF
SCF d
0 i = hΦSCF | d ĥiµ | ΦSCF
0 i=0
The coupled cluster amplitude equations are often collectively called the
coupled cluster vector function ei with elements,
eiµ = hΦSCF | d ĥiµ e−T̂ Ĥ (0) | ΦCC

0 i = 0
For the popular CCSD model the expression for the energy becomes then
E0CCSD = hΦSCF
0 | e−T̂ Ĥ (0) | ΦCCSD i = hΦSCF 0 | e−(T̂1 +T̂2 ) Ĥ (0) e−(T̂1 +T̂2 ) | ΦSCF
0 i
1 2
= E0SCF + hΦSCF
0 | V̂ ( T̂1 + T̂2 ) | ΦSCF
0 i
2
and the amplitude equations read
hΦai | e−(T̂1 +T̂2 ) Ĥ (0) | ΦCCSD

0 i = hΦai | [F̂ , T̂1 ] + V̂ T1 + [V̂ T1 , T̂2 ] | ΦSCF
0 i=0
hΦab
ij | e
−(T̂1 +T̂2 ) (0)
Ĥ | ΦCCSD
0 i
1
= hΦab
ij | [F̂ , T̂2 ] + V̂
T1
+ [V̂ T1 , T̂2 ] + [[V̂ T1 , T̂2 ], T̂2 ] | ΦSCF
0 i=0
2
The T̂1 transformed operators are defined as
ÔT1 = e−T̂1 Ô eT̂1

In the CC2 model these amplitude equations are approximated to second

order based on Møller-Plesset perturbation theory arguments.
However, the single excitations and thus the T̂1 operator are treated as being
of zeroth order contrary to MP perturbation theory, where they are of second
order, because they enter first in the second-order wavefunction.
In the presence of an external perturbation, however, the single excitations will
always enter the wavefunction already in zeroth-order.
Therefore they are treated as zeroth-order in CC2, which is contrary to MP2
constructed primarily for the calculation of molecular properties via the
response theory approach and not ground state energies.
On the other hand, the double excitation and thus the T̂2 operator are treated
as first-order like in Møller-Plesset perturbation theory.
The CC2 amplitude equations become then
hΦai | e−(T̂1 +T̂2 ) Ĥ (0) | ΦCC2

0 i = hΦai | [F̂ , T̂1 ] + V̂ T1 + [V̂ T1 , T̂2 ] | ΦSCF
0 i=0
hΦab
ij | e
−(T̂1 +T̂2 ) (0)
Ĥ | ΦCC2
0 i = hΦab
ij | [F̂ , T̂2 ] + V̂
T1
| ΦSCF
0 i=0
The next higher method is CCSDT, where triple excitations T̂3 are also
included in the wavefunction.
This is a rather expensive method and not yet employed on a regular basis.
However, one can make the same type of approximation to the equations for
the triples amplitudes, i.e. the triples coupled cluster vector function e3 , as
were made in CC2 to the doubles amplitude equations.
This leads then to the CC3 model.
Nevertheless one still has to solve the equations for the triples amplitudes
iteratively.
An non-iterative alternative is the CCSD(T) model, where the triples correction
to the CCSD energy is obtained from the triples contribution to the fourth-order
Møller-Plesset perturbation theory energy and from one fifth-order term
describing the coupling between singles and triples. Both contributions are,
however, evaluated with the CCSD amplitudes.

9.7 The Hellmann-Feynman Theorem for a Φ0 - 1
Approximate wavefunctions |Φ0 i depend on molecular orbital coefficients

{cµp } and possibly also on some kind of configuration or determinant
coefficients {Cn0 } which together are here denoted as {Ci }.
The energy of a molecule in all these approximate methods can be expressed
as an asymmetric expectation value
E0 = hΦ0 ′ | Ĥ (0) | Φ0 i
In the case of the variational methods, SCF, MCSCF and CI, |Φ0 ′ i = |Φ0 i and
we have the normal expectation value.
In non-variational methods (MP perturbation or CC theory), the energy is
calculated as a transition expectation value, where |Φ0 ′ i = |ΦSCF
0 i.
Let us now consider again the case of a Hamiltonian Ĥ(λ), which depends on
a perturbation symbolized by the real parameter λ.
Both sets of wavefunction coefficients and thus indirectly also |Φ0 ′ i and |ΦSCF
0 i
depend on λ
|Φ0 (λ)i = |Φ0 ({Ci (λ)})i
In addition to the wavefunction parameters, {cµp } and {C0k }, also the basis
functions χµ can depend on the perturbation. This will be the case, when the
perturbation corresponds to a change in the geometry or when perturbation
dependent basis functions such as the gauge including atomic orbitals (GIAO)
are used. We will ignore this here.
The derivative of the electronic energy E0 (λ, {Ci (λ)}) with respect to λ is

dE0 (λ, {Ci (λ)}) ∂E0 (λ, {Ci (λ)}) X ∂(E0 (λ, {Ci (λ)}) ∂Ci (λ)
= +
dλ ∂λ i
∂Ci (λ) ∂λ
Alternatively in terms of the (transition) expectation values
dE0 (λ, {Ci (λ), Ci′ (λ)}) ∂ Ĥ(λ)

= hΦ0 ′ ({Ci′ (λ)}) | | Φ0 ({Ci (λ)})i
dλ ∂λ
X ∂Φ0 ′ ({Ci′ (λ)})
∂Ci′ (λ)

+ h ′
| Ĥ(λ) | Φ0 ({Ci (λ)})i
i
∂C i (λ) ∂λ

X ∂Φ 0 ({C i (λ)}) ∂C i (λ)
+ hΦ0 ′ ({Ci′ (λ)}) | Ĥ(λ) | i
i
∂Ci (λ) ∂λ
The molecular orbital coefficients will actually be the same in |Φ0 ′ i and |Φ0 i,
i.e. c′µp = cµp for MPn and CC wavefunctions.
If the wavefunction is variationally optimized with respect to all parameters, i.e.
∂E0 (λ, {Ci (λ), Ci′ (λ)}) ∂E0 (λ, {Ci (λ), Ci′ (λ)})
= =0
∂Ci (λ) ∂Ci′ (λ)
and thus
∂ ∂ ′
|Φ0 ({Ci (λ)})i = ′
hΦ 0 ({C i (λ)})| = 0
∂Ci (λ) ∂Ci (λ)
the Hellmann-Feynman theorem is fulfilled again
dE0 (λ, {Ci (λ), Ci′ (λ)}) ∂E0 (λ, {Ci (λ), Ci′ (λ)})
=
dλ ∂λ
∂ Ĥ(λ)
= hΦ0 ′ ({Ci′ (λ)}) | | Φ0 ({Ci (λ)})i
∂λ
This is always the case for a SCF and MCSCF wavefunction, because they are
optimized with respect to all wavefunction parameters.

Truncated CI wavefunctions, by contrast, are not variationally optimized with

respect to the molecular orbital coefficients.
The Hellmann-Feynman theorem is therefore satisfied only in the limit of a full
CI wavefunction, when the molecular orbital coefficients are redundant.
In non-variational approaches such as Møller-Plesset perturbation theory or
coupled cluster methods the wavefunction is not at all variationally optimized.
However, one can choose hΦ0 ′ | in such a way that the Hellmann-Feynman
theorem is fulfilled and the transition expectation value still gives the energy.
Arponen defined such a transition expectation value for the CC energy
E0CC,Λ = hΦΛ
0 | Ĥ
(0)
| ΦCC
0 i
between the coupled cluster state and a dual bra or “Λ” state hΦΛ
0 | defined as
hΦΛ SCF
0 | = hΦ0 |(1 + Λ̂) e−T̂
The Λ̂ operator is defined as sum of the de-excitation operators

Λ̂ = Λ̂1 + Λ̂2 + · · · + Λ̂i + · · · + Λ̂N
The transition expectation value is an example for the much more general
concept of Lagrangians for non-variational wavefunctions.
A Lagrangian is the normal expression for the energy augmented with
constraints that are the equations, from which the wavefunction parameters
are determined, multiplied with Lagrangian multipliers.
The Lagrangian can thus be made stationary with respect to the wavefunction
parameters of non-variational wavefunctions contrary to the normal energy
expression.
In the case of the coupled cluster wavefunction the equations for the
wavefunction parameters, i.e. for the coupled cluster amplitudes tiµ , are
simply the equations for the coupled cluster vector function eiµ .
The constraints are then eiµ = 0 and the coupled cluster Langrangian LCC
0 is
X
LCC
0 = E0CC + λ iµ e iµ
iµ
where λiµ are Lagrangian multipliers.

Arponen’s transition expectation value is just an alternative way of writing the

coupled cluster Langrangian
E0CC,Λ = LCC 0
The Lagrangian multipliers in the coupled cluster Lagrangian are just the λiµ
amplitudes of the dual bra or “Λ” state hΦΛ
0 |.
The definition of the dual bra or “Λ” state hΦΛ

0 | is thus just a convenient way to
build these constraints into an expectation value expression.
The variational condition with respect to the λiµ amplitudes is then trivially
fulfilled for the coupled cluster Lagrangian
∂LCC
0 ∂E0CC,Λ
= =0
∂λiµ ∂λiµ
because the derivative with respect to the λiµ amplitudes is

∂hΦΛ
0 | Ĥ
(0)
| ΦCC i
= hΦSCF | d hiµ e−T̂ Ĥ (0) | ΦCC
0 i = e iµ = 0
∂λiµ
i.e. the coupled cluster vector function eiµ and thus the equations which the
coupled cluster amplitudes anyway fulfill. Molecular Electromagnetism A Computational Chemistry Approach – p.39/46
Equations for the CC λiµ amplitudes are obtained from the condition that the
coupled cluster Lagrangian is stationary with respect to the tiµ amplitudes, i.e.
∂LCC ∂E0CC,Λ ∂hΦΛ

0 | Ĥ
(0)
| ΦCC i
0
= = = hΦΛ
0 | [Ĥ
(0) e
, hiµ ] | ΦCC i = 0
∂tiµ ∂tiµ ∂tiµ
The transition expectation value E0CC,Λ , or coupled cluster Lagrangian LCC

0 , is
thus stationary with respect to the configuration or determinant coefficients
and satisfies therefore partially the Hellmann-Feynman theorem
∂E0CC,Λ ∂LCC ∂hΦΛ CC

0 (λ) | Ĥ(λ) | Φ0 (λ)i ∂ Ĥ(λ) CC
= 0
= = hΦΛ
0 | | Φ0 i
∂λ ∂λ ∂λ ∂λ
Identical expressions for first-order properties can be obtained as first
derivatives of the energy or as asymmetric expectation value with the Λ state.
Similar transition expectation values can also be defined for other
non-variational methods like Møller-Plesset perturbation theory, where one
defines a Lagrangian by adding the equations for the correlation coefficients as
extra conditions multiplied with Lagrangian multipliers to the respective MP
energy expression. Molecular Electromagnetism A Computational Chemistry Approach – p.40/46
This approach for the calculation of expectation values is called the unrelaxed
method, because the conditions for the molecular orbital coefficients were not
included as additional constraints in the coupled cluster Lagrangian.
The Hellmann-Feynman theorem is therefore not completely fulfilled.
A coupled cluster Lagrangian including orbital relaxation is the following
L0CC,relax = hΦSCF
0 | e−κ̂ Ĥ (0) eκ̂ eT̂ | ΦSCF
0 i
X
+ λiµ hΦSCF0 | d ĥiµ e−T̂ e−κ̂ Ĥ (0) eκ̂ eT̂ | ΦSCF
0 i
iµ
X h i
+ τpq hΦSCF
0 | †
qpq ,e −κ̂
Ĥ (0) κ̂
e | ΦSCF
0 i
pq
κ̂ is the orbital rotation operator and the τpq coefficients are the Lagrangian
multipliers for the conditions on the molecular orbital coefficients, i.e. the
Brillouin theorem.

The second derivatives of the energy for an approximate wavefunction is
d2 ∂ 2 E0 (λ, {Ci (λ)}) X ∂ 2 E0 (λ, {Ci (λ)}) ∂Ci

E0 (λ, {Ci (λ)}) = +2
dλ2 ∂λ2 i
∂λ ∂Ci ∂λ
X ∂ 2 E0 (λ, {Ci (λ)}) ∂Ci 2
+ 2
i
∂C i ∂λ
X ∂E0 (λ, {Ci (λ)}) ∂ 2 Ci
+ 2
i
∂C i ∂λ
For a variational energy the last term vanishes again and the second derivative
of the energy for variational methods is given as
d2 ∂ 2 E0 (λ, {Ci (λ)}) X ∂ 2 E0 (λ, {Ci (λ)}) ∂Ci

E0 (λ, {Ci (λ)}) = +2
dλ2 ∂λ2 i
∂λ ∂Ci ∂λ
X ∂ 2 E0 (λ, {Ci (λ)}) ∂Ci 2
+ 2
i
∂C i ∂λ

9.8 Approximate Density Matrices - 1
Finally we will derive approximations to the electron density P (~r) and reduced
one-electron density matrix P (~r, ~r ′ ) using two approximate wavefunctions:
the SCF wavefunction |ΦSCF
0 i
the Møller-Plesset perturbation theory wavefunction through second order:
|ΦMP1
0 i + |ΦMP2
0 i
This is most conveniently done in terms of the density matrices Dpq and Dµν
in the molecular orbital and atomic orbital basis.
They are the coefficients in the expansion of the electron density P (~r) in the
set of molecular orbitals {φp } or atomic orbitals {χµ }
X ∗
X
P (~r) = φp (~r) Dpq φq (~r) = χµ (~r) Dµν χ∗ν (~r)
pq µν
and in the expansion of the reduced one electron density matrix P (~r, ~r ′ )
′
X ∗ ′
X
P (~r, ~r ) = φp (~r) Dpq φq (~r ) = χµ (~r) Dµν χ∗ν (~r ′ )
pq µν

The electron density P SCF (~r) for an SCF wavefunction can be obtained as
expectation value of the density operator D̂(~r) and simple application of the
Slater-Condon rules
N/2 N/2
SCF
X X
P (~r) = hΦSCF
0 | D̂(~r) | ΦSCF
0 i = 2hφi (~r1 ) | δ(~r1 −~r) | φi (~r1 )i = 2 φ∗i (~r) φi (~r)
i i
The SCF density matrix in the molecular orbital basis is thus given as
SCF SCF SCF
Dij = 2 δij , Dia =0, Dab =0
Transforming the molecular orbitals to the atomic orbital basis

N/2 N/2
SCF
X XX
P (~r) = 2 φ∗i (~r) φi (~r) = 2 χ∗ν (~r) c∗νi χµ (~r) cµi
i i µν
gives the SCF density matrix in the atomic orbital basis

N/2
SCF
X
Dµν =2 c∗νi cµi
i Molecular Electromagnetism A Computational Chemistry Approach – p.44/46
The MP second-order correction to the electron density can be defined as
P M P 2 (~r) = hΦMP1 | P̂ (~r) | ΦMP1 i + hΦSCF

0 | P̂ (~r) | ΦMP2 i + hΦMP2 | P̂ (~r) | ΦSCF
0 i
X MP 2
= φp (~r) φ∗q (~r) Dpq
pq
This leads to the second-order correction to the density matrix in the molecular
orbital basis
X ab n o
MP2 ab ab
Dij = − tik [1] 4 tjk [1] − 2 tkj [1]
abk
X n o
MP2
Dab = tac
ij [1] 4 tbc
ij [1] −2 tbc
ji [1]
cij
MP2 MP 2
√ a
Dia = Dai = 2 ti [2]
This is often called the unrelaxed second-order correction to the density

matrix in order to distinguish it from the relaxed density matrix.

In the presence of a perturbation the electron density and reduced

one-electron density matrix are perturbed and expanded in a perturbation
series
(1)
X
P (~ ~ = P (~
r, F) r) + Pα (~
r) Fα + · · ·
α
(1)
X
P (~ ′
r, ~ ~ = P (~
r , F) r ′) +
r, ~ Pα (~ r ′ ) Fα + · · ·
r, ~
α
(1)
The first-order reduced one-electron density matrix Pα (~r, ~r ′ ) is then
correspondingly be expanded in molecular or atomic orbitals with elements of
(1) (1)
the first-order density matrices Dα,pq and Dα,µν as coefficients
(1) ′
X (1)
Pα (~r, ~r ) = φp (~r) Dα,pq φ∗q (~r ′ )
pq
X (1)
= χµ (~r) Dα,µν χ∗ν (~r ′ )
µν

Chapter 10
Stephan P. A. Sauer
sauer@kiku.dk


10. Approximations to exact PT Expressions
Approximations to the "exact" expressions derived with perturbation theory

Ground state expectation values
(0) (0)
< Ô >= hΨ0 | Ô | Ψ0 i
Sum-over-states expressions
X hΨ(0) | Ôα··· | Ψ(0) ihΨ

(0)
| | Ψ
(0)
0 n n Ôβ··· 0 i
hh Ôα··· ; Ôβ··· iiω=0 = −2 (0) (0)
n6=0 E0 − En
Polarization propagator or linear response function

(0) (0)
hh Ôα··· ; Ôβ··· iiω = hΨ0 | [Ôα··· , ĥi ] | Ψ0 i ···
 −1  
(0) (0) (0) (0)
hΨ0 |[ĥi , ~ω ĥj − [Ĥ (0) , ĥj ]]|Ψ0 i ··· hΨ0 | [ĥj , Ôβ··· ] | Ψ0 i
× 
   
. ..   . 
. . .
. .

10.1 Ground State Expectation Values
P
Expectation value of a spin free operator Ô = i ô(i) with an approximate
wavefunction |Φ0 i in terms of the reduced one-electron density matrix
Z
hΦ0 | Ô | Φ0 i = ô(1) P (~r1 , ~r ′1 ) d~r1
r ′1 =~
~ r1
Using the earlier derived expansion of the reduced one electron density matrix
P (~r, ~r ′ ) in the set of molecular orbitals {φp } or atomic orbitals {χµ }
′
X ∗ ′
X
P (~r, ~r ) = φp (~r) Dpq φq (~r ) = χµ (~r) Dµν χ∗ν (~r ′ )
pq µν
we obtain the expectation value as contraction with the MO density matrix

X Z X
∗
hΦ0 | Ô | Φ0 i = Dpq φq (~r1 )ô(1) φp (~r1 ) d~r1 = Dpq hφq | ô | φp i
pq pq
or alternatively with the AO density matrix

X
hΦ0 | Ô | Φ0 i = Dµν hχν | ô | χµ i
µν
10.1 Ground State Expectation Values - 2
And for the first-order correction to the field dependent expectation value
X Z
hΨ0 (F ~ )i(1) =
~ ) | Ô | Ψ0 (F Fα ô(1) Pα(1) (~r1 , ~r ′1 ) d~r1
α r ′1 =~
~ r1
(1)
Using the earlier derived expansion of the first-order density matrix Pα (~r, ~r ′ )
in the set of molecular orbitals {φp } or atomic orbitals {χµ }
(1) ′
X (1) ∗ ′
X (1)
Pα (~r, ~r ) = φp (~r) Dα,pq φq (~r ) = χµ (~r) Dα,µν χ∗ν (~r ′ )
pq µν
we obtain analogously an expression in terms of orbitals and the first-order

density matrix
(1)
X X (1)
~ ~
hΨ0 (F) | Ô | Ψ0 (F)i = Fα Dα,pq hφq | ô | φp i
α pq
X X (1)
= Fα Dα,µν hχν | ô | χµ i
α µν

10.2 Sum-over-States Methods - 1
We want to approximate the spectral representation of the polarization

propagator
ω
X hΨ(0) (0) (0) ω (0)
0 | P̂ | Ψn ihΨn | Ôβ··· | Ψ0 i
hh P̂ ; Ôβ··· iiω = (0) (0)
n6=0 ~ω + E0 − En
X hΨ(0) ω (0) (0) (0)
0 | Ôβ··· | Ψn ihΨn | P̂ | Ψ0 i
+ (0) (0)
n6=0 −~ω + E0 − En
(0) (0) (0) (0)

We need excitation energies En − E0 and transition moments hΨ0 |Ô|Ψn i
(0) (0)
Or ground state wavefunction |Ψ0 i and excited state wavefunctions |Ψn i
Uncoupled Hartree-Fock approximation (in the early days)
(0)
En(0) − E0 ≈ ǫa − ǫi
(0)
hΨ0 | Ô | Ψ(0)
n i ≈ hφi | ô | φa i

10.2 Sum-over-States Methods - 2
Nowadays are such sum-over-states expressions used for

Benchmark studies of two electron systems using explicitly correlated
wavefunctions
Hyperpolarizabilities of larger systems using semi-empirical methods
Analysis of contributions to a molecular property from excitations between
individual, typically localized, molecular orbitals (RPA or DFT)

10.3 Møller-Plesset PT Polarization Propagator - 1
Methods in which approximations are made

ω (0) (0)
hh P̂α ;
Ôβ··· iiω = hΨ0 | [P̂α , ĥi ] | Ψ0 i ···
 −1  
(0) (0) (0) ω ] | Ψ(0) i
hΨ0 |[ĥi , ~ω ĥj − [Ĥ (0) , ĥj ]]|Ψ0 i ··· hΨ0 | [ĥj , Ôβ··· 0
× 
   
. ..   . 
.. . ..
Exact as long as
{hn } is complete
(0)
|Ψ0 i is an eigenfunction of the unperturbed Hamiltonian
Approximate polarization propagator methods
Truncating the set of operators
Using an approximate reference state |Φ0 i
MCSCF and MPn wavefunctions are commonly employed

Complete set of operators:

all possible single excitation and de-excitation operators ĥ1 , all possible double
excitation and de-excitation operators ĥ2 up to all possible N -tuple excitation
and de-excitation operators ĥN with respect to the SCF wavefunction |ΦSCF 0 i
       
e e
ĥ1 q̂ † ĥ2 q̂ † q̂ †
ĥ1 =  = , ĥ2 =  = , ...
d d
ĥ1 q̂ ĥ2 q̂q̂
The matrix form of the polarization propagator can be written as

 −1  
M 11 M 12 · · · ω )
T 1 (Ôβ···
   
ω )

ω  M 21 M 22 · · ·  T 2 (Ôβ···
hh P̂α ; Ôβ··· iiω = TT
1 (P̂α ) TT
2 (P̂α ) ··· 
 
  
 . . .
.
  
.. .. .. ..
where the M ij matrices in the principal propagator are defined as
M ij = ~ωS ij − E ij

The property gradient vectors T Ti (P̂α ) and T j (Ôβ···

ω
) are given as

(0) (0) (0) (0)
TT
i (P̂α ) = e T T (P̂ ) d T T (P̂ )
i α i α = hΨ0 | [P̂α ,e hT
i ] | Ψ0 i hΨ0 | [P̂α ,d hT
i ] | Ψ0 i
   
(0) ω ] | Ψ(0) i
ω
eT ω
i (Ôβ··· ) hΨ0 | [e h†i , Ôβ··· 0
T i (Ôβ··· )= =
(0) d †

d T (Ô ω ) ω ] | Ψ(0) i
hΨ0 | [ hi , Ôβ···
i β··· 0
and the overlap S ij are defined as

   
(0) (0) (0) (0)
ee S
ij
ed S
ij hΨ0 | [e h†i ,e hT
j ] | Ψ0 i hΨ0 | [e h†i ,d hT
j ] | Ψ0 i
S ij =  =
(0) (0) (0) (0)

de S
ij
dd S
ij hΨ0 | [d h†i ,e hT
j ] | Ψ0 i hΨ0 | [d h†i ,d hT
j ] | Ψ0 i
and electronic Hessian matrices E ij are defined as

 
ee E ed E
ij ij
E ij =  
de E dd E
ij ij
 
(0) (0) (0) (0)
hΨ0 | [e h†i , [F̂ + V̂ ,e hT
j ]] | Ψ0 i hΨ0 | [e h†i , [F̂ + V̂ ,d hT
j ]] | Ψ0 i
= (0) (0) (0) (0)

hΨ0 | [d h†i , [F̂ + V̂ , e hT
j ]] | Ψ0 i hΨ0 | [d h†i , [F̂ + V̂ , d hT
j ]] | Ψ0 i

A perturbation series of methods of order n is obtained by requiring that the

matrix elements in S ij , E ij as well as T Ti (P̂α ) and T i (Ôβ···
ω
) are evaluated
through order n in the fluctuation potential.
However, one concentrates on the single excitations and considers the
higher-excited contributions only as corrections to the former.
This is done in a partitioned form of the principal propagator matrix M .
The inverse of a blocked matrix can be rewritten as
 −1  −1 −1 
−1 −1
U V U −VZ W W − ZV U
  = −1 −1 
−1 −1
W Z V − UW Z Z − WU V
 −1 −1 
−1 −1 −1
U −VZ W −U V Z − W U V
=  −1 −1 
−1 −1 −1
−Z W U − V Z W Z − WU V

With this relation one can rewrite the matrix representation of polarization
propagator in a partitioned form as
h i
ω T T −1
hh P̂α ; Ôβ··· iiω = T 1 (P̂α ) − T 2··· (P̂α )(M 2···2··· ) M 2···1
−1 −1
× M 11 − M 12··· (M 2···2··· ) M 2···1
h i
ω −1 ω
× T 1 (Ôβ··· ) − M 12··· (M 2···2··· ) T 2··· (Ôβ··· )
where the “2 · · · " means contributions from the h2 and higher operators
In a nth-order polarization propagator approximation one includes all matrix
element of order up to and including n
in the M 11 matrix and T T1 (P̂α ) and in the T 1 (Ôβ···
ω
) vectors
in the products T T2··· (P̂α )(M 2···2··· )−1 M 2···1 , M 12··· (M 2···2··· )−1 T 2··· (Ôβ···
ω
)
and M 12··· (M 2···2··· )−1 M 2···1
this implies up to order m = n − 2 in the matrix M 2···2···
and at the same time n−m 2
in the matrices M 12··· and M 2···1 and as well
as in the vectors T T2··· (P̂α ) and T 2··· (Ôβ···
ω
)

10.3.1 First Order and Zeroth Order PPA - 1
First-order polarization propagator approximation(FOPPA)

Reference state: the Hartree-Fock wavefunction |ΦSCF
0 i
Set of operators: hn = h1
(0) (1) (1)
h2 could contribute, because M 22 6= 0, M 12 6= 0 and M 21 6= 0, but
M 12 (M 22 )−1 M 21 is at least second order and thus not included.
It is known as time-dependent Hartree-Fock approximation (TDHF) or
random phase approximation (RPA)
The polarization propagator in the RPA is then given as
 
e
ω

e T (0) d T (0)
X β··· (ω)
hh P̂α ; Ôβ··· iiω = Pα Pα  
d
X β··· (ω)
with the so-called solution vector given as

      −1  
e (1)∗ ω(0)
X β··· (ω) 1 0 A(0,1) B e
O β···
  = ~ω  −   
d (1) (0,1)∗ d ω(0)
X β··· (ω) 0 −1 B A O β···

with the non-vanishing RPA matrix elements in terms of real spin-orbitals {ψp }
given as
(0,1) ee
= hΦSCF †
]] | ΦSCF

Aai,bj = E 11 ai,bj 0 | [qai , [F̂ + V̂ , qbj 0 i
= hΦai | F̂ + V̂ | Φbj i − δij δab hΦSCF

0 | F̂ + V̂ | ΦSCF
0 i
= (ǫa − ǫi )δij δab

+ ψa (~r1 ) ψi (~r1 ) ψj (~r2 ) ψb (~r2 ) − ψa (~r1 ) ψb (~r1 ) ψj (~r2 ) ψi (~r2 )

(1) de
E 11 ai,bj = hΦSCF | [qai , [F̂ + V̂ , qbj ]] | ΦSCF

Bai,bj = 0 0 i
= −hΦSCF
0 | Ĥ (0) | Φab
ij i

= ψa (~r1 ) ψj (~r1 ) ψb (~r2 ) ψi (~r2 ) − ψa (~r1 ) ψi (~r1 ) ψb (~r2 ) ψj (~r2 )

e (0) e T
Pα,ai = T 1 (P̂α ) ai = hΦSCF †
] | ΦSCF

0 | [P̂α , qai 0 i = hψi | p̂α | ψa i
d (0) d T
Pα,ai = T 1 (P̂α ) ai = hΦSCF | [P̂α , qai ] | ΦSCF

0 0 i = −hψa | p̂α | ψi i
e ω(0) e ω
Oβ··· ,ai = T 1 (Ôβ··· ) ai = hΦSCF ω
] | ΦSCF i = hψa | ôω

0 | [qai , Ôβ··· 0 β··· | ψi i
d ω(0) d ω
Oβ··· ,ai = T 1 (Ôβ··· ) ai = hΦSCF † ω
] | ΦSCF i = −hψi | ôω

0 | [qai , Ôβ··· 0 β··· | ψa i

The RPA polarization propagator can be written as

ω
X e d

hh P̂α ; Ôβ··· iiω = X β··· ,ai (ω)hψi | p̂α | ψa i − X β··· ,ai (ω)hψa | p̂α | ψi i
ai
Recalling Z
F (1)
hh P̂ ; Ôβ··· iiω=0 = p̂(1) Pβ (~r1 , ~r ′1 ) d~r1
r ′1 =~
~ r1
and
X
Pα(1) (~r, ~r ′ ) = (1)
φp (~r) Dα,pq φ∗q (~r ′ )
pq
we can identify the occupied-virtual and virtual-occupied blocks of the RPA

(1),RPA
first-order density matrix Dβ,pq as
(1),RPA e
Dβ,ai = Xβ··· ,ai (ω)
(1),RPA
Dβ,ia = −d X β··· ,ai (ω)

The structure of the RPA equations is the same as in time-dependent density

functional theory (TD-DFT), although they differ in the expressions for the
elements of the Hessian matrix E 22 .
A frequent approximation to the RPA, often called the Tamm-Dancoff
approximation, is obtained by setting the B (1) matrix zero.
It is often employed in TD-DFT.
It corresponds to mono-excited CI
The remaining A(0,1) matrix is the "singly excited - singly excited
determinants" part of the CI matrix of the Hamiltonian minus the
Hartree-Fock energy
The excited states are a linear combination of singly excited determinants
{Φai } while the ground state is the Hartree-Fock determinant.
Retaining only the zeroth-order contribution to the hessian matrix, i.e. A(0) ,
gives uncoupled Hartree-Fock sometimes also called the zeroth-order
polarization propagator approximation (ZOPPA).

The B (1) matrix is the “Hartree-Fock - doubly excited determinant" part of the
CI matrix
It introduces correlation in the ground state, which in the RPA is described
with the Hartree-Fock determinant plus a linear combination of all doubly
excited determinants.
However, the expansion coefficients in this linear combination are not
variationally optimized.
They can be derived from the condition that the off-diagonal hypervirial
relation

ˆ r (0) ı~ ˆ p (0)
En − Em hΨ(0)
(0) (0)
m | ~
O | Ψ n i = hΨ (0) ~
m | O | Ψn i
me
is fulfilled for the RPA transition moments.

10.3.2 Second Order PPA - 1
Second-order polarization propagator approximation(SOPPA)

Reference state: the Møller-Plesset perturbation theory wavefunction
(0) MP SCF MP1 MP2

|Ψ0 i ≈ |Φ0 i = C |Φ0 i + |Φ i + |Φ i + ···
Set of operators: hn = {h1 , h2 }
The polarization propagator in the SOPPA is then given as

hh P̂α ; ω
Ôβ··· iiω = e P T (0,2) d P T (0,2) e ΠT (1) d ΠT (1)
α α α α
    (1)
−1 
Σ (0,2)
0 0 0 A(0,1,2) B (1,2)∗
C̃ 0 e O ω(0,2)
β···
(1)∗  d O ω(0,2) 
 
(0,2)∗
    
 0 −Σ 0 0  (1,2) (0,1,2)∗
 B A 0 C̃ 
β···
× ω  −
    
e Ωω(1) 
  
 0 0 1 0  (1) 0 D (0) 0
 C
 
    β··· 
∗ ∗ d Ωω(1)
0 0 0 −1 0 C (1) 0 D (0) β···
For the second order contributions to the A-, B- and Σ-matrix of RPA one
needs only the first-order correction to the wavefunction |ΦMP1 i
(2) ω(2)
In the P α and O β··· vectors one needs also the single excitation part,
P a a MP2
ai ti [2] |Φi i, of the second-order correction |Φ i to the wavefunction.

ω
For two spin-free operators P̂α and Ôβ··· the elements of the A-matrix can in
terms of spatial orbitals {φp } be written as
(0,1,2) ee (0,1,2)
Aai,bj = E 11 ai,bj
= hΦMP
0
†
| [qai , [F̂ + V̂ , qbj ]] | ΦMP
0 i
(0,1,2)
= (ǫa − ǫi )δij δab

+ 2 φa (~r1 ) φi (~r1 ) φj (~r2 ) φb (~r2 ) − φa (~r1 ) φb (~r1 )φj (~r2 ) φi (~r2 )

1
MP2 MP2

+ (ǫb − ǫj ) δab Dij − δij Dba
2
1 X n cd cd
o
− δab φj (~r1 ) φc (~r1 )φk (~r2 ) φd (~r2 ) 4 tik [1] − 2 tki [1]
2
cdk
1 X
φk (~r1 ) φb (~r1 )φl (~r2 ) φc (~r2 ) {4 tac ac

− δij kl [1] − 2 tlk [1]}
2
ckl
The second-order correction to the A matrix here is not Hermitian. The

second-order correction to the A-matrix is therefore normally symmetrized
′(2) 1 (2) (2)

(2) 1 (2) (2)

Aai,bj = Aai,bj + Abj,ai = Aai,bj + Abj,ai − Aai,bj
2 2
The elements of the B-matrix can be written as

(1,2) de (1,2) MP MP (1,2)
Bai,bj = E 11
ai,bj
= hΦ 0 | [qai , [F̂ + V̂ , qbj ]] | Φ 0 i

= −2 φa (~r1 ) φi (~r1 ) φb (~r2 ) φj (~r2 ) + φa (~r1 ) φj (~r1 ) φb (~r2 ) φi (~r2 )

1 Xh n bc bc
o
+ φa (~r1 ) φj (~r1 )φk (~r2 ) φc (~r2 ) 4 tik [1] − 2 tki [1]
2
ck
ac ac
+ φb (~r1 ) φi (~r1 ) φk (~r2 ) φc (~r2 )
4 tjk [1] − 2 tkj [1]
n bc bc
o
+ φa (~r1 ) φc (~r1 )φk (~r2 ) φj (~r2 ) 4 tki [1] − 2 tik [1]
ac ac i
+ φb (~r1 ) φc (~r1 )φk (~r2 ) φi (~r2 ) 4 tkj [1] − 2 tjk [1]
1X n ab ab
o
− φk (~r1 ) φi (~r1 )φl (~r2 ) φj (~r2 ) 4 tkl [1] − 2 tlk [1]
2
kl
1X n cd cd
o
− φa (~r1 ) φc (~r1 )φb (~r2 ) φd (~r2 ) 4 tij [1] − 2 tji [1]
2 cd

Compared to the A- and B-matrix in RPA one obtains in second-order

additional contributions:
One consist of contractions of two-electron repulsion integrals with the
ab ab

first-order doubles correlation coefficients 4 tij [1] − 2 tji [1]
Both matrices have contributions from integrals with two occupied and two
virtual molecular orbitals.
The B-matrix includes also integrals with four occupied or four virtual
molecular orbitals.
The A-matrix has also a contribution from the second-order correction to the
density matrix.
This is a renormalization term, which can be written as
hΦMP
0
†
| [qai , [F̂ + V̂ , qbj ]] | ΦMP
0 i
(0)
hΦMP
0 | ΦMP
0 i
(2)
It arises because the Møller-Plesset perturbation theory wavefunction has

to be normalized at each order.
There is no renormalization term in the B-matrix, as the zeroth-order
B-matrix vanishes. Molecular Electromagnetism A Computational Chemistry Approach – p.20/35
The elements of the Σ-matrix can be written as

(0,2) ee (0,2)
Σai,bj = S 11 ai,bj
= hΦMP
0
†
| [qai , qbj ] | ΦMP
0 i
(0,2)
1 MP2 1 MP2
= δab δij + δab Dij − δij Dba
2 2
The second-order correction to the overlap matrix Σ consists of a similar term
as the renormalization contribution the A-matrix.
It contains the second-order correction to the density matrix.
This is to be expected as the overlap matrix is related to the norm of the
second-order Møller-Plesset perturbation theory wavefunction, which can be
expressed in terms of the second-order correction to the density matrix as
MP MP (2)
X MP2 X MP2
hΦ0 | Φ0 i = Daa − Dii
a i

There are three new matrices due to the ĥ2 operators:

The M 21 block contains the C-matrix
(1) ee (1) †
Caibj,ck = E 21 aibj,ck
= hΦSCF
0 | [qai qbj , [F̂ + V̂ , qck ]] | ΦSCF
0 i(1)
√
2 n
= δik φa (~
r1 ) φc (~
r1 ) φb (~
r2 ) +δjk φa (~
r2 ) φj (~ r1 ) φi (~
r1 ) φb (~
r2 ) φc (~
r2 )
(1 + δab δij )
o
− δac φk (~
r1 ) φi (~
r1 )φb (~ r2 ) −δbc φa (~
r2 ) φj (~ r1 ) φi (~
r1 )φk (~
r2 ) φj (~
r2 )
The M 12 block contains the C̃-matrix

(1) ee (1) † †
C̃ck,aibj = E 12 ck,aibj = hΦSCF 0 | [qck , [F̂ + V̂ , qai qbj ]] | ΦSCF
0 i(1)
1
n
o
= √ δik 2 φj (~ r1 ) φb (~
r1 )φc (~
r2 ) φa (~ r2 ) − φj (~ r1 ) φa (~r1 )φc (~r2 ) φb (~
r2 )
2
n o
+ δjk 2 φi (~r1 ) φa (~
r1 )φc (~r2 ) φb (~r2 ) − φi (~ r1 ) φb (~r1 )φc (~
r2 ) φa (~r2 )
n o
− δac 2 φi (~ r1 ) φj (~
r1 ) φk (~ r2 ) φb (~
r2 ) − φj (~ r1 ) φi (~
r1 ) φk (~ r2 ) φb (~
r2 )
n
o
− δbc 2 φj (~
r1 ) φk (~
r1 )φi (~
r2 ) φa (~
r2 ) − φi (~
r1 ) φk (~
r1 )φj (~
r2 ) φa (~
r2 )
The M 22 block contains the D-matrix

(0) ee (0)
Daibj,ckdl = E 22 aibj,ckdl
= hΦSCF
0
† †
| [qai qbj , [F̂ + V̂ , qck qdl ]] | ΦSCF
0 i(0)
1
= (ǫa + ǫb − ǫi − ǫj ) (δac δik δbd δjl + δad δil δbc δjk )
(1 + δab δij )
They are given in terms of spatial orbitals {φp } for two spin-free operators P̂α
ω
and Ôβ··· using a biorthogonal set of double excitation operators {q̂ † q̂ † , q̂q̂}.
(1)
The C (1) and C̃ matrices couple double with single excitations.
They consist of two-electron repulsion integrals like A(1) and B (1)
The D (0) matrix consists only of molecular orbital energy differences.
Pure double excitations are treated in SOPPA only in zeroth-order and thus
not more accurate than in a uncoupled Hartree-Fock calculation.
However, the effect of the doubles excitations on the singly excited states is
(1)
still correct through second order due to the C (1) and C̃ matrices, as one
can see from the partitioned form of the polarization propagator.

T (0) ω(0)
There are second-order corrections to the P α and O β··· property gradients
e (0,2) e (0,2) †
Pα,ai = TT
1 (P̂α )
ai
= hΦ MP
0 | [ P̂ α , qai ] | Φ MP (0,2)
0 i
√ 1 X MP2 MP2

= 2hφi | p̂α | φa i + √ hφj | p̂α | φa i Dji − hφi | p̂α | φj i Daj
2 j
1 X MP2 MP2

+√ hφb | p̂α | φa i Dbi − hφi | p̂α | φb i Dab
2 b
d ω(0,2) d ω
(0,2) †
Oβ··· ,ai = T 1 (Ôβ··· ) ai = hΦMP 0 | [qai ω
, Ôβ··· ] | ΦMP
0 i
(0,2)
√ ω 1 X ω MP2 ω MP2

= − 2hφi | ôβ··· | φa i − √ hφj | ôβ··· | φa iDji − hφi | ôβ··· | φj iDaj
2 j
1 X ω MP2 ω MP2

−√ hφb | ôβ··· | φa iDbi − hφi | ôβ··· | φb iDab
2 b
They consist of contractions of the property integrals with D MP2

Contrary to the A(2) - and S (2) -matrix here it is the occupied-virtual and
virtual-occupied off-diagonal blocks of the density matrix.
This is the only place, where the second-order correction to the
wavefunction contributes in SOPPA. Molecular Electromagnetism A Computational Chemistry Approach – p.24/35
T (1) ω(1)
And additional contributions Πα and Ωβ··· due to the ĥ2 operators
e (1) e T (1)
Πα,aibj = T 2 (P̂α ) aibj = hΦMP
0
† †
| [P̂α , qai qbj ] | ΦMP
0 i
(1)
1 X n
ab ab
o n
ab ab
o
=− hφi | p̂α | φk i 4 tkj [1] − 2 tjk [1] +hφj | p̂α | φk i 4 tik [1] − 2 tki [1]
2
k
1 X n
cb cb
o ac ac
+ hφc | p̂α | φa i 4 tij [1] − 2 tji [1] +hφc | p̂α | φb i 4 tij [1] − 2 tji [1]
2 c
d ω(1) d ω (1) MP † † ω MP (1)

Ωβ··· ,aibj = T 2 (Ôβ··· ) aibj
= hΦ 0 | [qai qbj , Ôβ··· ] | Φ 0 i
2 Xn ω ab ω ab
o
= hφi | ôβ··· | φk i tkj [1] + hφj | ôβ··· | φk i tik [1]
(1 + δab δij )
k
2 Xn ω cb ω ac
o
− hφc | ôβ··· | φa i tij [1] + hφc | ôβ··· | φb i tij [1]
(1 + δab δij ) c
T (1) ω(1)
The first-order Πα and Ωβ··· vectors consist of contractions of the property
integrals in molecular orbital basis with the first-order doubles correlation
coefficients. Molecular Electromagnetism A Computational Chemistry Approach – p.25/35
The SOPPA polarization propagator can not be written as the contraction of

property integrals of the operator P̂ with a SOPPA first-order density matrix:
T (2)
The P α contribution is not just property integrals in the molecular orbital
basis. However, one could
define a kind of MP2 correction to the molecular orbitals as
!
1 X X
φMP2
p = MP2
φi Dip − φa DapMP2
2 i a
require that the MP2 correction to the property integrals is linear
hφi | p̂α | φa iMP2 = hφi | p̂α | φMP2

a i + hφ MP2
i | p̂α | φa i
express the ĥ1 part of the propagator as a contraction of these correlated

property integrals with a SOPPA first-order density matrix.
T (1)
The ĥ2 contributions to the property gradients Πα as well as to the
solution vector Ξβ··· (ω) are both 4 index quantities
This prevents us from formulating their contribution as a the contraction of
property integrals with a first-order one particle density matrix.
10.3.3 Higher-order and mixed methods - 1
Third-order polarization propagator approximation (TOPPA)

All terms in the partitioned form of the polarization propagator are evaluated
through third order.
The expressions for all matrix elements have been derived but only parts
had so far been implemented.
Other SOPPA-like methods are based on the fact that a coupled cluster
wavefunction gives a better description than the Møller-Plesset first- and
second-order wavefunctions.
In the second-order polarization propagator with coupled cluster
singles and doubles amplitudes - SOPPA(CCSD) - method, the
(0)
reference state |Ψ0 i is approximated by a linearized CCSD wavefunction

(0)
|Ψ0 i ≈ 1 + T̂1 + T̂2 |ΦSCF
0 i
This keeps the structure of the SOPPA equations but replaces in all matrix
elements the first-order MP doubles and second-order MP singles
correlation coefficient by coupled cluster singles and doubles amplitudes.
10.3.3 Higher-order and mixed methods - 2
In the earlier coupled cluster singles and doubles polarization propagator

approximation (CCSDPPA) this was done only partially and in particular not
in the second-order correction to the density matrix D MP2 .
Recently a third method, SOPPA(CC2), was proposed in which the
Møller-Plesset correlation coefficients are replaced by the corresponding
CC2 amplitudes instead of the CCSD amplitudes.
This reduces the computational cost in the calculation of the amplitudes to
the same as in the SOPPA calculation.

10.3.4 Iterative and Non-iterative Doubles Correction - 1
In RPA:
excited states are described as linear combination of singly excited
determinants
In SOPPA:
the description of excited states is improved by including electron correlation
in the reference state and by including also double excitation operators ĥ2 .
The main purpose of the double excitations is to improve the description of the
same single excitation dominated states as in RPA.
This is a rather costly correction.
The dimension of the principal propagator is increased from two times the
number of single excitations squared to two times the number of single and
double excitations squared.
It is thus worthwhile to consider alternatives for including the double
excitation corrections to the single excitation dominated excited states
without having to diagonalize a matrix of the full dimension of single and
double excitations squared.
The original implementations of the SOPPA method were therefore based on

the partitioned form of the polarization propagator.
The eigenvalue equation for the partitioned principal propagator in SOPPA is
  
(0,1,2) (1) (0) −1 (1) (1,2) e
A + C̃ (~ωn − D ) C B X
  n
(1)

B (1,2) A(0,1,2) + C̃ (−~ωn − D (0) )−1 C (1) dX
n
  
Σ(0,2) 0 eX
n
= ~ωn   
0 −Σ(0,2) dX
n
The matrix is frequency-dependent and depends on its own eigenvalues ~ωn .

One has to solve iteratively for one excitation energy at a time by inserting the
eigenvalue obtained in one iteration as frequency in the principal propagator
for the next iteration until this eigenvalue remains the same and one has
obtained self-consistency.
Today SOPPA excitation energies are not calculated from the partitioned form
of the polarization propagator anymore.

In the doubles corrected random-phase approximation - RPA(D) approach

one does not include the doubles corrections in the principal propagator.
Based on the pseudo perturbation theory one corrects the RPA excitation
energies with a non-iterative doubles correction.
The SOPPA electronic Hessian matrix is split in three contributions
 
(0,1) (1)
A B 0 0
 
 B (1) A(0,1) 0 0 
(0)
E =
 
(0)


 0 0 D 0  
0 0 0 D (0)
(1)
   
(2) (2)
0 0 C̃ 0 A B 0 0

 0 (1)  
 B (2) (2)

(1)
0 0 C̃  A 0 0 
E =  and E (2) = 
   

 C (1) 0 0 0   0 0 0 0 
   
0 C (1) 0 0 0 0 0 0

The SOPPA overlap matrix is also split in three contributions
S = S (0) + S (1) + S (2)

     
(2)
1 0 0 0 0 0 0 0 Σ 0 0 0
     
 0 −1 0 (2)
0   0 0 0 0   0 −Σ 0 0 
= + +
     
 
 0 0 1 0   0 0 0 0   0 0 0 0 
     
0 0 0 −1 0 0 0 0 0 0 0 0
Choosing the zeroth-order eigenvectors to be the RPA eigenvectors

T
T (0) †

Ln = RTn (0) = e
X RPA
n
d
X RPA
n 0 0
implies that the zeroth-order eigenvalues are the RPA excitation energies
ωn(0) = ωnRPA
and that the E (1) and S (1) matrices do not contribute in first-order, because
L(0)
n E
(1) (0)
Rn = 0
The second-order correction to the eigenvalues becomes then

  
(2)

e X RPA † d X RPA †
A(2) − ~ωn
RPA Σ(2) B (2) e X RPA
n
~ωn = n n
  
(2) (2) RPA Σ(2) d X RPA
B A + ~ωn n
  −1
(1)

e X RPA † d X RPA †
C̃ 0 D (0) − ~ωn
RPA 0
− n n

(1)
 
(0) RPA
0 C̃ 0 D + ~ωn
  
C (1) 0 e X RPA
n
×  
0 C (1) d X RPA
n
The eigenvalues and thus excitation energies in the RPA(D) method are
defined as the sum of the zeroth- and second-order contributions
~ωnRPA(D) = ~ωnRPA + ~ωn(2)
Only the RPA eigenvectors are necessary and that thus only the RPA
eigenvalue problem has to be solved.
The second-order and doubles correction are evaluated only once:
RPA(D) has a non-iterative doubles correction. Molecular Electromagnetism A Computational Chemistry Approach – p.33/35
10.4 Multiconfigurational PPA - 1
Multiconfigurational random phase approximation (MCRPA):

a multiconfigurational polarization propagator approximation
Set of operators contains state transfer operators {R† , R} and non-redundant
single excitation q † and de-excitation q operators.
The expression for the polarization propagator in MCRPA can be obtained
(0)
from the SOPPA equations if one identifies h2 with {R† , R} and |Ψ0 i with
|ΦMCSCF
0 i.
ω
hh P̂α ; Ôβ··· iiω
= T T (P̂α ) (~ωS − E)−1 T (Ôβ···
ω
)
    −1 
e ω

e T d T
Σ ∆ A B T (Ôβ··· )
= T (P̂α ) T (P̂α ) ~ω   −   
d ω
−∆ −Σ B A T (Ôβ··· )

10.4 Multiconfigurational PPA - 2
where the MCRPA electronic Hessian and overlap matrices property as well as
gradient vectors are commonly defined as
 
hΦMCSCF
0 | [q, [Ĥ (0) , q †T ]] | ΦMCSCF
0 i hΦMCSCF
0 | [R† , [Ĥ (0) , q T ]] | ΦMCSCF
0 i
A=  
MCSCF (0) †T MCSCF MCSCF (0) †T MCSCF
hΦ0 | [R, [Ĥ , q ]] | Φ0 i hΦ0 | [R, [Ĥ , R ]] | Φ0 i
 
MCSCF (0) T MCSCF MCSCF (0) T MCSCF
hΦ0 | [q, [Ĥ , q ]] | Φ0 i hΦ0 | [R, [Ĥ , q ]] | Φ0 i
B=  
hΦMCSCF
0 | [R, [Ĥ (0) , q †T ]] | ΦMCSCF
0 i hΦMCSCF
0 | [R, [Ĥ (0) , RT ]] | ΦMCSCF
0 i
 
hΦMCSCF
0 | [q, q †T ] | ΦMCSCF
0 i hΦMCSCF
0 | [q, R†T ] | ΦMCSCF
0 i
Σ= 
hΦMCSCF
0 | [R, q †T ] | ΦMCSCF
0 i hΦMCSCF
0 | [R, R†T ] | ΦMCSCF
0 i
 
MCSCF T MCSCF MCSCF T MCSCF
hΦ0 | [q, q ] | Φ0 i hΦ0 | [q, R ] | Φ0 i
∆=  
MCSCF T MCSCF MCSCF T MCSCF
hΦ0 | [R, q ] | Φ0 i hΦ0 | [R, R ] | Φ0 i

e T MCSCF †T MCSCF MCSCF †T MCSCF
T (P̂α ) = hΦ0 | [P̂α , q ] | Φ0 i hΦ0 | [P̂α , R ] | Φ0 i

Chapter 11
Stephan P. A. Sauer
sauer@kiku.dk


11. PT and RT with Approximate Wavefunctions
In this chapter we will apply time-independent and time-dependent

perturbation theory to approximate solutions of the unperturbed molecular
Hamiltonian.
In particular we will illustrate this in the following for Hartree-Fock, MCSCF and
coupled cluster wavefunctions.

11.1 Coupled and Time-Dependent Hartree-Fock - 1
In the coupled Hartree-Fock method (CHF) the Hartree-Fock equations are

~
solved self-consistently in the presence of a perturbing field F
fˆ(1, F)
~ ψp (~r1 , F)
~ = ǫp (F)
~ ψp (~r1 , F)
~
~ )} remain
under the condition that the perturbed occupied spin-orbitals {ψi (F
orthonormal
~ ) | ψj (~r1 , F
hψi (~r1 , F ~ )i = δij
The perturbed occupied spin-orbitals are expanded in the set of

orthonormalized unperturbed molecular spin-orbitals {ψq }
all
X
~) =
ψi (~r1 , F ~)
ψq (~r1 ) Uqi (F
q
This is contrary to the unperturbed Hartree-Fock theory, where the molecular

orbitals are expanded in atomic one-electron basis functions.
One of the consequences of this choice is that the unperturbed Fock matrix
becomes diagonal in this basis.
Inserting this ansatz in the perturbed Hartree-Fock equations gives

all all
fˆ(1, F)
X X
~ ~ = ǫi (F)
ψq (~r1 ) Uqi (F) ~ ~
ψq (~r1 ) Uqi (F)
q q
and for the orthonormality condition

all
X ∗ ~ ~ ) = δji
Ujq (F ) Uqi (F
q
Projecting the perturbed Hartree-Fock equations against another basis

function ψr one obtains a matrix form of the perturbed Hartree-Fock equations
all
X all
X
~ Uqi (F)
Frq (F) ~ = ǫi (F)
~ ~)
δrq Uqi (F
q q

~ is given as
An element of the perturbed Fock matrix Fpq (F)
~
Fpq (F) = hψp | fˆ(F)
~ | ψq i
= hψp | ĥ(0) + ĥ(1) + ĥ(2) | ψq i

occ X
X all
∗ ~ ~

+ Usj (F ) ψp ψq ψs ψt − ψp ψt ψs ψq
Utj (F)
j st
The first and second-order one-electron perturbation Hamiltonians ĥ(1) + ĥ(2)

are scalar or tensor products of the general perturbation Fα··· and two
one-electron perturbation operators ôF FF
α··· and ôαβ···
(1) (2)
X X
ĥ (i) + ĥ (i) = Fα··· ôF
i,α··· + Fα··· ôFF
i,αβ··· Fβ···
α··· α,β,···

The perturbed Fock matrix in the basis of the unperturbed molecular

spin-orbitals, Fpq (F~ ), the perturbed orbital energies ǫi (F~ ) and the coefficients
Uqi (F~ ) of the perturbed orbitals are then expanded in orders of this perturbing
field
(0)
X (1)
Fpq (F~ ) = Fpq + Fα··· ,pq Fα··· + · · ·
α···
~ = ǫ(0) + (1)
X
ǫi (F) i ǫα··· ,i Fα··· + · · ·
α···
~ ) = U (0) + (1)
X
Uqi (F qi Uα··· ,qi Fα··· + · · ·
α···
The zeroth-order Fock matrix and coefficient matrix are both diagonal
(0)
Uqp = δqp
(0)
Fpq = ǫ(0)
p δpq
which is a consequence of the fact that we used the unperturbed molecular

spin-orbitals {ψq } as basis.
The perturbed spin-orbitals are thus expanded in orders of the perturbation

all
(1)
X X
~ = ψi (~r1 ) +
ψi (~r1 , F) ψq (~r1 ) Uα··· ,qi Fα··· + · · ·
q α···
CHF ~
Also the SCF density matrix is perturbed, Dµν (F ), and is expanded in orders
of the perturbation
SCF ~ SCF
X (1),CHF
Dµν (F) = Dµν + Dα··· ,µν Fα··· + · · ·
α···
SCF
From the definition of the unperturbed density matrix Dµν it follows that for a
closed-shell molecule the first-order correction to the SCF density matrix in the
atomic orbital basis is then
(1),CHF
X (1)∗ ∗ (1)

(1)∗ (1)

Dα··· ,µν = 2 cα··· ,µi cνi + cµi cα··· ,νi = 2 Uα··· ,µν + Uα··· ,νµ
i

Inserting these expansion in the perturbed Hartree-Fock equations and

separating orders, we obtain the first-order Hartree-Fock equations
all all all
(1) (0) (0) (1) (1) (0)
X (0) (1)
X X
(Frq Uα··· ,qi + Fα··· ,rq Uqi ) = ǫi δrq Uα··· ,qi + ǫα··· ,i δrq Uqi
q q q
and orthonomality condition

all
(0)∗ (1) (1)∗ (0)
X
Ujq Uα··· ,qi + Uα··· ,jq Uqi =0
q
Inserting the zeroth-order terms one obtains

(0) (1) (1) (1)
(ǫ(0)
r − ǫi ) Uα··· ,ri = ǫα··· ,i δri − Fα··· ,ri
(1) (1)∗
Uα··· ,ji + Uα··· ,ji = 0

For r 6= i one finally obtains for the expansion coefficients of the first-order
correction to the orbitals
(1)
(1) Fα··· ,ri
Uα··· ,ri =− (0) (0)
ǫr − ǫi
(1)
which must be solved iteratively since the first-order Fock matrix, Fα··· ,ri ,
(1)
depends on Uα···
(1)
Fα··· ,ri = hψr | ôF
α··· | ψi i
occ X
X (1)∗ (1)

+ ψ r ψ i ψ s ψ t − ψ r ψ t ψ s ψ i Uα··· ,sj δtj + δsj Uα··· ,tj
j st

In principle the perturbed occupied orbitals are expanded in all unperturbed

spin-orbitals.
However, mixing the occupied orbitals among themselves does not change the
total wavefunction.
Therefore we only need the virtual-occupied blocks of the matrices of the first-
(n)
and higher-order coefficients Uα··· ,ai .
The first-order Hartree-Fock equations then read
vir h
occ X
X (0) (0)
i (1)
(ǫa − ǫi ) δab δij + ψa ψi ψj ψb − ψa ψb ψj ψi Uα··· ,bj
j b
vir
occ X
X (1)∗
+ ψa ψi ψb ψj − ψa ψj ψb ψi Uα··· ,bj

j b
= −hψa | ôF
α··· | ψi i
The two expressions in the parentheses are the RPA A(0,1) and B (1) matrices.

The first-order Hartree-Fock equations can then be written as

vir h
occ X i
(0,1) (1) (1) (1)∗
X
Aai,bj Uα··· ,bj − Bai,bj Uα··· ,bj = −hψa | ôF
α··· | ψi i
j b
For real (−) or imaginary (+) perturbations we have that

(1)∗ (1) (1)∗ (1)
Uα··· ,bj = Uα··· ,bj or Uα··· ,bj = −Uα··· ,bj ,
respectively and the first-order Hartree-Fock equations simplify to two sets of
linear equations
vir
occ X
(0,1) (1) (1)
X
Aai,bj ∓ Bai,bj Uα··· ,bj = −hψa | ôF
α··· | ψi i
j b
Once the expansion coefficients of the first-order correction to the orbitals,

(1)
Uα··· ,bj , are obtained as solutions of these equations, we can calculate the
corrections to the Hartree-Fock energy and thus molecular properties.

~ the perturbed Hartree-Fock energy is

In the presence of the perturbing field F
occ
X
E0SCF (F)
~ = ~)
ǫ i (F
i
occ vir
1XX ∗ ~ ∗ ~
~ )Udj (F)
~
− Uai (F )Ucj (F ) ψa ψb ψc ψd − ψa ψd ψc ψb Ubi (F

2 ij
abcd
Inserting the perturbation series and separating orders the second-order

correction to the energy can be shown to be
occ
(2),SCF
X X
E0 = Fα··· hψi | ôFF
αβ··· | ψi i Fβ···
α,β,··· i
vir
occ X
(1)∗ (1)
X X
+ Fα··· Uα··· ,ai hψa | ôF
β··· | ψi i + hψi | ôF
β··· | ψa i Uα··· ,ai Fβ···
α,β,··· i a
This can be compared with the sum-over-states expression for the

second-order energy correction of exact wavefunctions.

Second-order molecular properties which are defined as second derivatives of

the energy are thus obtained at the coupled Hartree-Fock level as
(2),SCF occ
∂ 2 E0 X
= hψi | ôFF
αβ··· | ψi i
∂Fα··· ∂Fβ··· i
vir
occ X
(1)∗ (1)
X
+ Uα··· ,ai hψa | ôF
β··· | ψi i + hψi | ôF
β··· | ψa i Uα··· ,ai
i a

The relation to the random phase approximation (RPA) becomes clear, when
one considers also the complex conjugate equation of the first-order
Hartree-Fock equations
vir h
occ X i
(1) (1) (0,1) (1)∗
X
−Bai,bj Uα··· ,bj + Aai,bj Uα··· ,bj = −hψi | ôF
α··· | ψa i
j b
The molecular orbital integrals hψa | ôF

α··· | ψi i over the first-order perturbation
(1) ω(0)
operator ôα··· are the elements of the RPA property gradients e O α··· and
d ω(0)
O α···
Combining the first-order Hartree-Fock equations and their complex conjugate
in one matrix equation one obtains
    
(1) ω(0)
A(0,1) B (1) U α··· e
O α···
   = − 
(1)∗ ω(0)
B (1) A(0,1) −U α··· d
O α···

(1) (1)∗
or solving for U α··· and −U α···
   −1  
(1) ω(0)
U α··· A(0,1) B (1) e
O α···
  = −   
(1)∗ ω(0)
−U α··· B (1) A (0,1) d
O α···
For a static perturbation, the RPA solution vectors e X α··· (ω = 0) and

d (1) (1)∗
X α··· (ω = 0) are identical to the expansion coefficients U α and −U α of
the first-order correction to the orbitals.

In Time-dependent Hartree-Fock theory (TDHF) this derivation is

generalized to the time-dependent field of a monochromatic linear polarized
radiation in the dipole approximation.
The molecular orbitals are then also time-dependent and are again expanded
in the unperturbed orbitals
~ = ψi (~r1 )
ψi (~r1 , t, F)
1X X h
(1) ıωt (1) −ıωt
i
+ ψp (~r1 ) Fα··· (ω) Uα··· ,pi (ω)e + Uα··· ,pi (−ω)e + ···
2 p α···
(1) (1)
The frequency-dependent expansion coefficients Uα··· ,pi (ω) and Uα··· ,pi (−ω)
are then determined by inserting the time-dependent orbitals in the
time-dependent version of the Hartree-Fock equation,

~ ) − ı ∂ ψp (~r1 , t, F)
fˆ(1, F ~ = ǫp (F)
~ ψp (~r1 , t, F
~)
∂t
which can be derived from Frenkel’s variational principle .

In analogy to the derivation of the coupled Hartree-Fock equations one can

then derive the time-dependent Hartree-Fock equations for the first-order
expansion coefficients
        
∗ (1) ω(0)
1 0 A(0,1) B (1) U α··· (−ω) e
O α···
~ω  − ∗
   =  
(1) d ω(0)
0 −1 B (1) A(0,1) −U α··· (ω) O α···
which are just the RPA equations with

   
(1) e
U (−ω) X α··· (ω)
 α··· = 
(1) d
−U α··· (ω) X α··· (ω)

The variational nature of the Hartree-Fock wavefunction means that the

CHF/TDHF equations are equivalent to the RPA equations.
Historically, CHF, was favoured over RPA since it could be solved in the atomic
orbital basis, rather than requiring a transformation to the molecular orbital
basis.
Unlike RPA and its correlated extensions, however, an atomic-orbital based
solution of the iterative CHF equations cannot give excitation energies and
transition moments.
The need for an inverse Hessian in RPA/SOPPA also restricted the size of
system which could be studied.
However, the use of direct atomic orbital driven methods for RPA response
properties and for SOPPA, coupled with iterative methods for solving the
inverse Hessian, mean that they can now be applied as widely as CHF/TDHF
and provide far more information about excited states and properties.

11.2 Multiconfigurational Linear Response Functions - 1
In the application of response theory to an MCSCF wavefunction |ΦMCSCF0 i the

time-dependent MCSCF state |ΦMCSCF0 (t)i is usually expressed as
|ΦMCSCF
0 (t)i = e ıκ̂(t) ıŜ(t)
e |Φ MCSCF
0 i
where the time-dependent orbital rotation and state transfer operators are
defined as
Xh † ∗
i
κ̂(t) = κpq (t) q̂pq + κpq (t) q̂pq
p>q
Xh † ∗
i
Ŝ(t) = Sn0 (t) R̂n0 + Sn0 (t) R̂0n
n6=0
∗
For the SCF case we have Sn0 (t) = Sn0 (t) = 0.
The orbital rotation parameters κpq (t) and κ∗pq (t) become then equal to the
(1)
Fourier transforms of the TDHF expansion coefficients Uα,pi (ω) and
(1)
Uα,pi (−ω).

The orbital rotation and state transfer operators are collected in a vector
ĥT = ({q̂pq
† †
}, {q̂pq }, {R̂n0 }, {R̂0n })
and the time-dependent orbital rotation and state transfer parameters are
collected in a one column vector
∗
γ(t) = ({κpq (t)}, {κ∗pq (t)}, {Sn0 (t)}, {Sn0 (t)})T
The parameters are then expanded in orders of the time-dependent

perturbation Hamiltonian Ĥ (1) (t)
γ(t) = γ (1) (t) + γ (2) (t) + · · ·
But contrary to exact states the time-dependence of the MCSCF wavefunction

is not determined directly from the time-dependent Schrödinger equation in the
presence of the perturbation Ĥ (1) (t).

Instead, one applies the Ehrenfest theorem to the operators, which determine
the time-dependence of the MCSCF wavefunction, i.e. the operators {ĥj }
d MCSCF ı
hΦ0 (t) | ĥj | ΦMCSCF
0 (t)i+ hΦMCSCF
0 (t) | [ĥj , Ĥ (0) +Ĥ (1) (t)] | ΦMCSCF
0 (t)i = 0
dt ~
Inserting the expression for the time-dependent MCSCF wavefunction and the
perturbation expansion of the wavefunction parameters one can separate the
orders and finds for the first-order equation
d d
ı~hΦMCSCF
0 | [ĥj , κ̂(t)(1) + Ŝ(t)(1) ] | ΦMCSCF
0 i
dt dt
−hΦMCSCF
0 | [[ĥj , Ĥ (0) ], κ̂(t)(1) + Ŝ(t)(1) ] | ΦMCSCF
0 i
= −ıhΨMCSCF
0 | [ĥj , Ĥ (1) (t)] | ΨMCSCF
0 i
where the first-order operators are defined as

(1)
X h (1) † (1)∗
i
κ̂ (t) = κpq (t) q̂pq + κpq (t) q̂pq
p>q
(1)
X h (1) † (1)∗
i
Ŝ (t) = Sn0 (t) R̂n0 + Sn0 (t) R̂0n
n6=0 Molecular Electromagnetism A Computational Chemistry Approach – p.21/38
Using the definitions of the MCRPA Hessian E and overlap S matrices one
can write the first-order equation more compactly as
d (1)
ı~S γ (t) − Eγ (1) (t) = T (Ĥ (1) (t))
dt
This linear system of ordinary differential equations is normally solved by
Fourier transforming it to the frequency domain.
We define thus a Fourier transform of the first-order time-dependent
wavefunction parameters γ (1) (t)
Z ∞ Z ∞
1 X
γ (1) (t) = dω γ (1) (ω) = ω
dω X(Ôβ··· ) Fβ··· (ω) e−ıωt
−∞ 2 β··· −∞
and insert it together with the Fourier transform of Ĥ (1) (t) and obtain
ω ω
(~ωS − E) X(Ôβ··· ) = T (Ôβ··· )
These are the MCRPA response equations again, but now derived via
response theory applied to an approximated MCSCF wavefunction.
11.3 SOPPA - 1
SOPPA and higher-order Møller-Plesset perturbation theory polarization

propagator approximations can also be derived via approximate response
theory based on an exponential parametrization of the time-dependent
wavefunction and the Ehrenfest theorem.
Using this formulation one can also derive SOPPA quadratic and higher order
response functions, which has not yet been achieved with the superoperator
formalism.
The key step is to rewrite the Møller-Plesset perturbation theory wavefunction
in such a way that it resembles a MCSCF wavefunction
 
MP  SCF X a a X ab ab
|Φ0 i = C |Φ0 i + ti |Φi i + tij |Φij i + · · · 

ai a>b
i>j
 
X a a† X ab ab†
= C 1 + ti Ri + tij Rij + · · ·  |ΦSCF i
 
0
ai a>b
i>j

11.3 SOPPA - 2
The Møller-Plesset state transfer operators are defined with respect to the
Hartree-Fock wavefunction
a···†
R̂i··· = |Φa··· SCF
i··· ihΦ0 |
a···
R̂i··· = |ΦSCF
0 ihΦa···
i··· |
And the correlation coefficients are expanded in Møller-Plesset perturbation

theory orders of the fluctuation potential
ta··· a··· a···

i··· = ti··· [1] + ti··· [2] + · · ·
The time-dependent Møller-Plesset wavefunction |ΦMP

0 (t)i is then written as
|ΦMP
0 (t)i = e
ıκ̂(t) ıŜ(t)
e |ΦMP
0 i

11.3 SOPPA - 3
The operators governing the time-dependence of the wavefunction are then

defined as
Xh † ∗
i
κ̂(t) = κpq (t) q̂pq + κpq (t) q̂pq
p>q
X h ab··· ab···† ab···∗ ab···
i
Ŝ(t) = Sij··· (t) R̂ij··· + Sij··· (t) Rij···
ab···
ij···
One should note that single excitations and de-excitations are not included in
the time-dependent MP state transfer operator Ŝ(t) because they are already
included as orbital rotations κ̂(t).
With this ansatz for the wavefunction the Ehrenfest theorem becomes
d MP ı MP
hΦ0 (t) | ĥj | ΦMP
0 (t)i + hΦ0 (t) | [ĥj , Ĥ (0) + Ĥ (1) (t)] | ΦMP
0 (t)i = 0
dt ~
From this equation one can derive the equation for the SOPPA solution vector
following the derivation of the MCSCF response function.

11.4 Coupled Cluster Linear Response Functions - 1
Coupled Cluster response functions can be derived from a time-dependent

version of the coupled cluster transition expectation value
hΦΛ CC
0 (t) | P̂ | Φ0 (t)i
where the time-dependent coupled cluster state |ΦCC

0 (t)i and dual or “Λ" state
hΦΛ
0 (t)| are defined as
|ΦCC
0 (t)i = e
T̂ (t)
|ΦSCF
0 i
h i
hΦ0 (t)| = hΦ0 | 1 + Λ̂(t) e−T̂ (t)
Λ SCF
The time-dependent cluster and Λ operator consist of single, double and so

forth excitation operators
T̂ (t) = T̂1 (t) + T̂2 (t) + T̂3 (t) + · · · + T̂N (t)
Λ̂(t) = Λ̂1 (t) + Λ̂2 (t) + Λ̂3 (t) + · · · + Λ̂N (t)

with single-, double-, etc. excitation and de-excitation operators are defined as
P a † P a
T̂1 (t) = ai t i (t) q̂ai Λ̂1 (t) = ai i (t) q̂ai
λ
† †
tab λab
P P
T̂2 (t) = a>b ij (t) q̂ai q̂bj Λ̂2 (t) = a>b ij (t) q̂ai q̂bj
i>j i>j
.. ..
. .
Using the shorthand notation {e ĥiµ } and {d ĥiµ } for all excitation and
de-excitation operators one can write the time-dependent cluster and Λ
operator more compactly as
X
T̂ (t) = tiµ (t) e ĥiµ
iµ
X
Λ̂(t) = λiµ (t) d ĥiµ
iµ
where tiµ (t) and λiµ (t) are the corresponding time-dependent amplitudes.

The time-dependent amplitudes, tiµ (t) and λiµ (t), are then determined from
the coupled-cluster time-dependent Schrödinger equations
d
e−T̂ (t) ı~ |ΦCC 0 (t)i = e
−T̂ (t)
Ĥ(t) |ΦCC
0 (t)i
dt

d
−ı~ hΦΛ 0 (t)| e
T̂ (t)
= hΦΛ 0 (t)| Ĥ(t) e
T̂ (t)
dt
by projecting them on hΦSCF

0 |d ĥiµ and e ĥiµ |ΦSCF
0 i, respectively, yielding two
systems of ordinary linear differential equations
dtiµ (t)
ı~ = hΦSCF
0 | d ĥiµ e−T̂ (t) Ĥ(t) | ΦCC
0 (t)i
dt
dλiµ (t)
−ı~ = hΦΛ e CC
0 (t) | [Ĥ(t), ĥiµ ] | Φ0 (t)i
dt
In the presence of a time-dependent perturbation Ĥ (1) (t) the amplitudes tiµ (t)
and λiµ (t) are expanded in a perturbation series
(1)
tiµ (t) = tiµ + tiµ (t) + · · ·
(1)
Inserting these expansions in the differential equations and separating orders

one obtains in first-order
(1)
dtiµ (t) X (1)
ı~ = hΦSCF
0 | d ĥiµ e −T̂
Ĥ (1)
(t) | ΦCC
0 i + tjν (t) Aiµ jν
dt jν
(1)
dλiµ (t) X (1)
X (1)
−ı~ = hΦΛ
0 | [Ĥ (1)
(t), e
ĥi µ ] | Φ CC
0 i + tjν (t) Fiµ jν + λjν (t) Ajν iµ
dt jν jν
where |ΦCC Λ
0 i and hΦ0 | are the time-independent, unperturbed coupled cluster
and “Λ” state wavefunctions, respectively, and T̂ is the time-independent,
unperturbed cluster operator.
The elements of the coupled cluster Jacobian A matrix and of the F matrix
are defined as
Aiµ jν = hΦSCF
0 | d ĥiµ e−T̂ [Ĥ (0) , e ĥjν ] | ΦCC
0 i
Λ (0) e e CC
F iµ j ν = hΦ0 | [Ĥ , ĥiµ ], ĥjν | Φ0 i

The Jacobian matrix A is the first derivative of the time-independent coupled

cluster amplitude equations, i.e. the coupled cluster vector function e with
respect to the time-independent amplitudes tjν
∂ ∂
Aiµ jν = e iµ = hΦSCF
0 | d ĥiµ e−T̂ Ĥ (0) | ΦCC
0 i
∂tjν ∂tjν
The equations for the time-dependent amplitudes are solved by transforming

them to the frequency domain.
ω ω
We define therefore Fourier components Xiµ (Ôβ··· ) and Yiµ (Ôβ··· ) of the
first-order amplitudes as
Z ∞ Z ∞
(1) (1) 1 X ω
tiµ (t) = dω tiµ (ω) = dω Xiµ (Ôβ··· ) Fβ··· (ω) e−ıωt
−∞ 2 −∞
β···
Z ∞ Z ∞
(1) (1) 1 X ω
λiµ (t) = dω λiµ (ω) = dω Yiµ (Ôβ··· ) Fβ··· (ω) e−ıωt
−∞ 2 β··· −∞
They are inserted in the two differential equations together with the Fourier
transform of Ĥ (1) (t)
Evaluating the derivatives gives

XZ ∞
ω
dω ~ω Xiµ (Ôβ··· ) Fβ··· (ω) e−ıωt
β··· −∞
XZ ∞
= dω hΦSCF
0 | d ĥiµ e−T̂ Ôβ···
ω
| ΦCC
0 iFβ··· (ω) e
−ıωt
β··· −∞
XZ ∞ X
ω
+ dω Xjν (Ôβ··· ) Aiµ jν Fβ··· (ω) e−ıωt
β··· −∞ jν
XZ ∞
ω
− dω ~ω Yiµ (Ôβ··· ) Fβ··· (ω) e−ıωt
β··· −∞
XZ ∞
= dω hΦΛ ω e CC
0 | [Ôβ··· , ĥiµ ] | Φ0 i Fβ··· (ω) e
−ıωt
β··· −∞
XZ ∞ X
ω
+ dω Fiµ jν Xjν (Ôβ··· ) Fβ··· (ω) e−ıωt
β··· −∞ jν
XZ ∞ X
ω
+ dω Yjν (Ôβ··· ) Ajν iµ Fβ··· (ω) e−ıωt
β··· −∞ jν

Both equations have to be fulfilled for any frequency ω and strength of the field
component Fβ··· (ω) which implies that
ω SCF d −T̂ ω CC
X ω
~ω Xiµ (Ôβ··· ) = hΦ0 | ĥiµ e Ôβ··· | Φ0 i + Xjν (Ôβ··· ) Aiµ jν
jν
X X
ω
~ω Yiµ (Ôβ··· ) = −hΦΛ ω e CC
0 | [Ôβ··· , ĥiµ ] | Φ0 i−
ω
Fiµ jν Xjν (Ôβ··· )− ω
Yjν (Ôβ··· ) A j ν iµ
jν jν
ω ω
Collecting the unknown Fourier components Xiµ (Ôβ··· ) and Yiµ (Ôβ··· ) in two
ω ω
column vectors X(Ôβ··· ) and Y (Ôβ··· ) one obtains two sets of coupled linear
equations
ω
X(Ôβ··· ) = (~ω 1 − A)−1 hΦSCF 0 | d he−T̂ Ôβ···
ω
| ΦCC
0 i
h i
T ω Λ ω e T CC ω
Y (Ôβ··· ) = − hΦ0 | [Ôβ··· , h ] | Φ0 i + F X(Ôβ··· ) (~ω 1 + A)−1
where 1 is a unit matrix of the same dimension as A.

ω
One should note that the X(Ôβ··· ) amplitudes have to be determined before
ω
one can solve for the amplitudes Y (Ôβ··· ) of the “Λ" state.
These are the analogous equations to the response equations for

Møller-Plesset perturbation theory polarization propagators or MCSCF linear
response functions.
However, there are a few important differences:
In coupled cluster linear response theory one has to solve two sets of
equations: one for each type of amplitudes.
But the coupled cluster Jacobian matrix A is only half the size of
corresponding Hessian matrices in Møller-Plesset polarization propagator
methods, because the set of operators {ĥiµ } consists of only excitation or
de-excitation operators in coupled cluster response theory.
The coupled cluster Jacobian matrix A is inherently asymmetric, which
implies that the left and right eigenvectors of it will not be the same.
ω ω
Solving these equations for the Fourier components Xiµ (Ôβ··· ) and Yiµ (Ôβ··· )
of the first-order time-dependent amplitudes means that one has determined
the time-dependence of the coupled cluster and “Λ" wavefunctions to first
order.
We can insert the time-dependent amplitudes now in an expansion of the

time-dependent transition expectation value in orders of the perturbation
hΦΛ CC
0 (t) | P̂ | Φ0 (t)i = hΦΛ CC
0 | P̂ | Φ0 i
X (1)
+ tiµ (t) hΦΛ e CC
0 | [P̂ , hiµ ] | Φ0 i
iµ
(1)
X
+ λiµ (t) hΦSCF
0 | d hiµ e−T̂ P̂ | ΦCC
0 i + ···
iµ
and obtain
hΦΛ
0 (t) | P̂ | Φ CC
0 (t)i = hΦ Λ
0 | P̂ | Φ CC
0 i
1X ∞
Z X ω
+ dω Xiµ (Ôβ··· ) hΦΛ e CC
0 | [P̂ , hiµ ] | Φ0 i Fβ··· (ω) e
−ıωt
+ ···
2 β··· −∞ iµ
1X ∞
Z X ω
+ dω Yiµ (Ôβ··· ) hΦSCF
0 | d hiµ e−T̂ P̂ | ΦCC
0 i Fβ··· (ω) e
−ıωt
2 −∞ i
β··· µ

Comparison with the analogous expansion of an expectation value for exact

states shows that the coupled cluster linear response function is given as
ω
hh P̂ ; Ôβ··· iiω = T Λ,T (P̂ ) X(Ôβ···
ω
) + Y T (Ôβ···
ω
) T CC (P̂ )
where the elements of the “property gradient" vectors T Λ (Ô) and T CC (Ô) for
an operator Ô are defined as
TiΛµ (Ô) = hΦΛ e CC

0 | [Ô, hiµ ] | Φ0 i
TiCC
µ
( Ô) = hΦ SCF d
0 | h i µ e −T̂
Ô | Φ CC
0 i
ω ω
Inserting the two solutions vectors X(Ôβ··· ) and Y (Ôβ··· ) we finally obtain
ω
hh P̂ ; Ôβ··· iiω = T Λ,T (P̂ ) (~ω 1 − A)−1 T CC (Ôβ···
ω
)
+ T Λ,T (Ôβ···
ω
) (−~ω 1 − A)−1 T CC (P̂ )
h iT
+ F (~ω 1 − A) T (Ôβ··· ) (−~ω 1 − A)−1 T CC (P̂ )
−1 CC ω

The poles of the coupled cluster linear response function and thus the vertical
excitation energies are found as eigenvalues of the coupled cluster Jacobian
(A − ~ω1) R = 0
For the most common coupled cluster response function methods, CCSD and
CC2, the Jacobian can be derived as derivatives with respect to the tiν
amplitudes of the CCSD and CC2 vector functions e
ACCSD =
 
hΦSCF d h Ĥ T1 + [Ĥ T1 , T̂ ], e hT | ΦSCF i hΦSCF | d h1 [Ĥ T1 , e hT SCF i

0 | 1 2 1 0 0 2 ] | Φ0
 
hΦSCF | d h2 Ĥ T1 + [Ĥ T1 , T̂2 ], e hT SCF i hΦSCF | d h2 [Ĥ T1 , e hT SCF i

0 1 | Φ0 0 2 ] | Φ0
ACC2 =
 
hΦSCF | d h1 Ĥ T1 + [Ĥ T1 , T̂2 ], e hT | ΦSCF hΦSCF | d h1 [Ĥ T1 , e hT SCF i

0 1 0 i 0 2 ] | Φ0
 
hΦSCF
0 | d h2 [Ĥ T1 , e hT SCF i
1 ] | Φ0 hΦSCF
0 | d h2 [F̂ , e hT
2 ] | Φ0
SCF i
where Ĥ T1 = e−T̂1 Ĥ (0) eT̂1 is the T1 transformed unperturbed Hamiltonian

CC2 linear response theory as well as SOPPA are in principle both

second-order response function methods, although there are significant
differences.
Nevertheless one can compare the blocks of elements in the CC2 Jacobian
with the corresponding matrices in the SOPPA Hessian.
The “d h2 , e hT2 " block in CC2 and the D (0) matrix in SOPPA have essentially
the same elements.
Also the “d h1 , e hT2 " and “d h2 , e hT1 " blocks in CC2 contain elements similar
(1)
to the C̃ and C (1) matrices.
However, there are also additional contributions in CC2 due to the T1
transformed Hamiltonian.
The same holds also for the “d h1 , e hT1 " block in CC2 which corresponds to
the A0,1,2 matrix in SOPPA.

Response functions have also been derived for the CC3 model which is an
approximation to the CCSDT model in the same way as CC2 is to CCSD.
The immense number of triple excitations included in the CC3 Jacobian
makes it necessary to formulate it in a partitioned form.
The partitioned CC3 Jacobian depends then on its own eigenvalues and
one has to iterate on each eigenvalue separately.
An alternative approximation to CC3 has been developed by using the
pseudo-perturbation theory, where the triples excitations are included as
non-iterative corrections to CCSD excitation energies.
This CCSDR(3) called method closely reproduces the results of CC3
calculations for vertical excitation energies.
The equation-of-motion coupled cluster approach (EOM-CCSD) method is
related to the CCSD linear response function approach but derived differently.
The EOM-CCSD excitation energies are identical to the excitation energies
obtained from the CCSD linear response function, but the transition
moments and second-order properties differ somewhat.
Chapter 12
Stephan P. A. Sauer
sauer@kiku.dk


12. Derivative Methods
In this chapter we will finally follow the third approach:

We abandon the perturbation theory approach all together.
We go back to the definitions of the properties as derivatives of the energy
in the presence of the perturbation.
We will illustrate with a few examples how this approach can be applied to
approximate expressions for the energy in the presence of both static as well
as time-dependent perturbations.
The presentation is restricted to Møller-Plesset perturbation theory and
coupled cluster energies as nothing new is obtained for variational methods
compared to the response theory approaches.

12.1 The Finite Field Method - 1
The finite field method involves numerical evaluation of derivatives of

Energy
First order properties (expectation values)
Higher order properties
In general a property P
Numerical derivative by finite differences or by fitting to a Taylor expansion
For example: ααα
E
Add − Ôα Eα to the Hamiltonian
Perform calculations of µα (Eα ) or E(Eα ) for several values of Eα
Numerical derivatives of µα (Eα ) or E(Eα )
ααβ as second derivative of E(E)~ requires a 2-dimensional fit
P can be a frequency dependent property like α(−ω; ω)

β(−ω; ω, 0), γ(−ω; ω, 0, 0) as numerical derivatives of α(−ω; ω)

12.1 The Finite Field Method - 2
Finite field method is easy to implement for real perturbation operators

Adding a one-electron operator to the Hamiltonian is necessary
Can be applied to any kind of correlation method
Can be applied even if no wavefunction or energy is defined:
⇒ Finite field calculations on top of SOPPA properties
Disadvantages of the finite field method
Numerical differentiation requires a careful choice of the perturbation
strength
Higher-order properties or multiple perturbations are cumbersome
Not applicable to time / frequency dependent perturbations
Imaginary perturbation operators of magnetic properties require complex
arithmetic
Becomes obsolete due to implementation of many analytical derivatives
Is still very much used for geometrical derivatives of properties
Similar approach: finite point charge method
12.2 The Analytical Derivative Method
Approximate expressions for P within a given method are differentiated

analytically with respect to the perturbation
It does not suffer from the numerical problems of the finite field method
More difficult to apply to a new type of wavefunction than finite field
Expressions for first and second-order properties have been implemented for
most ab initio methods
Geometrical gradients have been typically the first analytical derivatives to be
implemented.

12.2.1 First Derivatives - 1
First derivative of the energy of a system with respect Fα is for most methods

~
dE(F)
X ∂ ĥ(1) X ∂ ĥ(1)
= Dpq hφp | | φq i = Dµν hχµ | | χν i
dFα··· ~ ∂F ∂F

α··· α···
|F|=0 pq µν
where Dpq and Dµ,ν are density matrices

Evaluating the derivative of the Hamiltonians

dE(F)~ X F
X
= Dpq hφp | ôα··· | φq i = Dµν hχµ | ôF
α··· | χν i
dFα··· ~

|F |=0 pq µν
Atomic orbitals χµ are are here assumed to be independent of the perturbation

For variational wavefunctions: the same density matrices as obtained from the
expectation value

Non-variational wavefunctions: relaxed or response density matrix

Relaxed CC density can be calculated as transition expectation value of the
density operator
CC,rel
X CC,rel
P (~r) = hΦΛ | D̂(~r) | ΦCC i = φp (~r) φ∗q (~r) Dpq
pq
Relaxed density matrix: SCF and correlation part
D rel = D SCF + D corr,rel
correlation part:
D corr,rel = D amp,rel + D orb,rel
D amp,rel contains amplitudes or correlation coefficients
D orb,rel solution of Z-vector equations
Relaxation of the orbitals for non-variational wavefunctions
Only occupied-virtual and virtual-occupied blocks are non-zero
⇐ Brillouin condition for the SCF ground state
Illustrated for the differentiation of the MP2 energy correction

MP2 ~
dE (F) X MP2 ∂ ĥ(1) X MP2 ∂ ĥ(1)
= Dab hφa | | φb i + Dij hφi | | φj i
dFα··· ~ ∂Fα··· ∂F

α···
F=0 ab ij
X MP2 (1)
+ Lai Uα··· ,ai
ai
X MP2
X MP2
= Dab hφa | ôF
α··· | φb i + Dij hφi | ôF
α··· | φj i
ab ij
(1)
X
+ LMP2
ai Uα··· ,ai
ai
MP2 MP2
where Dij and Dab are the occupied-occupied and virtual-virtual blocks of
the second-order correction to the density matrix
MP2 MP2
Dij and Dab are contributions to D amp,rel
(1)
Uα··· ,ai are the solutions of the coupled Hartree-Fock equations, which enter in
the expression for the derivative, because the molecular orbitals also depend
on the field.
LMP2
ai is also called a Lagrangian, although it differs from the Lagrangians
discussed in connection with the Hellmann-Feynman theorem
It is given as
X
LMP2 2tcb tcb

ai = −4 ij [1] − ji [1] φa (~r1 ) φc (~r1 )φb (~r2 ) φj (~r2 )
jbc
X
2tab tcb

−4 kj [1] − jk [1] φk (~r1 ) φi (~r1 ) φb (~r2 ) φj (~r2 )

jkb
1 X MP2
+ Dbc 4 φa (~r1 ) φi (~r1 ) φb (~r2 ) φc (~r2 )

2 bc

− φa (~r1 ) φb (~r1 ) φi (~r2 ) φc (~r2 ) − φa (~r1 ) φc (~r1 ) φb (~r2 ) φi (~r2 )

1 X MP2
+ Djk 4 φa (~r1 ) φi (~r1 ) φk (~r2 ) φj (~r2 )

2
jk

− φa (~r1 ) φk (~r1 ) φi (~r2 ) φj (~r2 ) − φa (~r1 ) φj (~r1 ) φk (~r2 ) φi (~r2 )


(1)
Inserting the expression for Uα··· ,ai the third term can be written
−1
(1)
X XX
LMP2
ai Uα··· ,ai =− LMP2
ai A (0,1)
−B (1)
hφb | ôF
α··· | φj i
ai,bj
ai ai bj
This equation can be evaluated by

(1)
either solving first for Uα··· ,ai ,
X (0,1) (1)

(1)
Abj,ai − Bbj,ai Uα··· ,ai = hφb | ôF
α··· | φj i
ai
i.e. the coupled Hartree-Fock equations for all components α of the field F,~
with the molecular orbital integrals hφb | ôF
α··· | φj i as right hand side
or solving first for the so-called Z-vector

X (0,1) (1)

Aai,bj − Bai,bj Zbj = −LMP2
ai
bj
i.e. one set of coupled Hartree-Fock equations with this second type of
Lagrangian, LMP2ai as right hand side
The last term can then be written as

X MP2 (1) X
Lai Uα··· ,ai = Zai hφa | ôF
α··· | φi i
ai ai
It consists also of a contraction of molecular orbital integrals of the

perturbation operator ôFα··· with another vector
It is analog to the other two contributions to the first derivative, but involves
summation over pairs of occupied and virtual molecular orbitals
MP2,rel
Definition of the occupied-virtual and virtual-orbital blocks Dia and
MP2,rel
Dai of the relaxed density matrix as being equal to the Z vector
MP2,rel
Dai = Zai
This is the D orb,rel contribution to the relaxed density matrix

The first derivative of the MP2 energy correction is then the contraction of all
molecular orbital integrals of the perturbation operator with the corresponding
elements of the relaxed density matrix

MP2,rel ~
dE (F ) X MP2
= Dpq hφq | ôF
α··· | φp i
dFα···

~
F =0 pq
This is quite analog to the expression for the unrelaxed density matrix
consistent through second-order
The difference between the unrelaxed density matrix consistent through
second-order and the relaxed density MP2 matrix is therefore in the
occupied-virtual and virtual-orbital blocks D orb,rel
The difference is the inclusion of orbital relaxation

12.2.2 Second Derivatives - 1
The second derivative of the energy can be written as

2 ~
d E(F )
X ∂ 2 ĥ(2) X (1) ∂ ĥ(1)
= Dµν hχµ | | χν i + Dβ··· ,µν hχµ | | χν i
dFα··· dFβ··· ~ ∂F β··· ∂Fα··· ∂F

α···
|F|=0 µν µν
X FF FF
X (1)
= Dµν hχµ | ôαβ··· + ôβα··· | χν i + Dβ··· ,µν hχµ | ôF
α··· | χν i
µν µν
where we have assumed again that the atomic orbitals are independent of the
perturbation.
The derivative of the relaxed density matrix, so called first-order relaxed
density matrix, in the atomic orbital basis is given as
∗
(1) ∂D µν
X ∂D pq
X ∂c µp ∂c νq
Dβ··· ,µν = = c∗µp cνq + Dpq cνq + c∗µp
∂Fβ··· pq
∂Fβ··· pq
∂Fβ··· ∂Fβ···
The derivatives of the molecular orbital coefficients {cνq } are obtained by

solving the coupled-perturbed Hartree-Fock equations.

12.2.2 Second Derivatives - 2
At the Hartree-Fock level this is the same as the SCF first-order density matrix
discussed previously.
The occupied-occupied and virtual-virtual blocks of the correlated first-order
density matrix contain derivatives of the amplitudes or correlation coefficients
They are obtained by straightforward differentiation of the equations defining
the amplitudes
The occupied-virtual and virtual-occupied part requires the solution of the
first-order Z-vector equations, i.e. the derivative of the Z-vector equations.

12.3 Time Dependent Analytical Derivatives - 1
Several attempts have been made to extend the analytical energy derivative
method also to the case of time-dependent perturbations:
Pseudo-energy derivative (PED) method of Rice and Handy
Quasi-energy derivative (QED) method of Sasagane, Aiga and Itoh
Time-dependent second-order Møller-Plesset method (TDMP2) method of
Hättig und Heß
Finally the time-averaged quasi-energy method of Christiansen, Jørgensen
and Hättig, which summarizes all previous developments.
Response functions as derivatives of a perturbed time-dependent
quasi-energy
~ = hΨ(t, F)
Q(t, F) ~ | Ĥ(t) − ı~ ∂ | Ψ(t, F
~ )i
∂t

ω ω ω
Linear response function hh Ôα··· ; Ôβ··· iiω of an operator Ôα··· in the
presence of a time-dependent field with Fourier components Fβ··· (ω) is
in the PED method obtained as
 
1 2 ~
d Q(t, F )) 2 ~
d Q(t, F))
ω ω
hh Ôα··· ; Ôβ··· iiω = +

2 dFβ··· (ω)dFα··· (0) dFα··· (0)dFβ··· (ω)

~ |=0
|F ~ |=0
|F
in the QED or TDMP2 method obtained as

~
d2 Q(t, F)
ω ω
hh Ôα··· ; Ôβ··· iiω = | ~
dFα··· (−ω)dFβ··· (ω) |F|=0
and in the time-averaged quasi-energy method obtained as

~ )}T
d2 {Q(t, F
ω ω
hh Ôα··· ; Ôβ··· iiω = | ~
dFα··· (−ω)dFβ··· (ω) |F|=0
where the time average over a period T of the perturbation, i.e.

Ĥ (1) (t + T ) = Ĥ (1) (t), is defined as
T
1
Z
2
~ T
{Q(t, F)} = ~ dt
Q(t, F)
T −T
2
The TDMP2 approach has only be applied to the linear response function.
PED expressions for linear and quadratic response functions have been
derived at the SCF and MP2 level
QED expressions for linear and quadratic response functions have been
presented SCF, MP2 and coupled cluster level.
QED expressions for cubic and quartic response functions have been
presented for SCF, MCSCF, full and truncated CI wavefunctions.
Time-averaged QED expressions for linear and higher response functions,
excitation energies and excited state properties have been presented for
general variational and non-variational theories and in particular for various
coupled cluster methods.
At the SCF level all methods lead to the RPA expressions for the response
functions.
The QED and time-averaged QED method at the MCSCF level was also shown
to yield the same expressions as obtained from propagator or response theory.

For non-variational wavefunctions the various methods differ.

At the MP2 level, they were constructed to give the correct static perturbation
limit, i.e. the same as obtained from taking derivatives of the time-independent
MP2 energy, but give different frequency dependent values.
The PED method started from the normal expression for the MP2 closed-shell
energy but expressed with the time-dependent perturbed molecular orbitals.
In addition it was required that the condition
∂Φ(t)
hΦ(t) | i=0
∂Fα··· (0)
is fulfilled for the first-order time-dependent MP wavefunction
|Φ(t)i = |ΦSCF
0 (t)i + |ΦMP1 (t)i
However, no time-dependent contribution was included in the first-order

correlation coefficients apart from the time-dependence of the molecular
orbitals.

In the TDMP2, QED-MP2 and time-averaged QED methods, the derivatives

are taken of an MP2 time-dependent and relaxed quasienergy Lagrangian

MP2,relax SCF ∂ κ̂(t)
n o
L0 ~ = hΦ
(t, F) 0 |e −κ̂(t)
Ĥ(t) − ı~ e 1 + T̂2 [1](t) | ΦSCF 0 i
∂t

X ∂
+ λ2µ [1](t)hΦSCF
0 | d ĥ2µ e−κ̂(t) Ĥ(t) − ı~ eκ̂(t) | ΦSCF
0 i
2µ
∂t

X ∂
+ λ2µ [1](t)hΦSCF
0 | d ĥ2µ e−κ̂(t) F̂ (t) − ı~ eκ̂(t) , T̂2 [1](t) | ΦSCF
0 i
2µ
∂t

X ∂
+ τpq (t)hΦSCF
0
†
| qpq , e−κ̂(t) Ĥ(t) − ı~ eκ̂(t) | ΦSCF0 i
pq
∂t
The first term is the generalization of the normal MP2 energy to the case of
time-dependent molecular orbitals and first-order doubles correlation
coefficients t2µ [1](t).
κ̂(t) is the time-dependent orbital rotation operator of TDHF / RPA / SCF linear
response theory.

The second and third term are the time-dependent version of the equations for
the t2µ [1](t) coefficients multiplied with their Lagrangian multipliers λ2µ [1](t).
These equations are not included in the PED approach.
The last term is the TDHF equations multiplied by corresponding
time-dependent Lagrangian multipliers τpq (t).
This Lagrangian is thus variational with respect to
the TDHF coefficients
the first-order doubles correlation coefficients
the Lagrangian multipliers for both types of coefficients.
The TDHF coefficients have to be obtained by solving the TDHF equations.
The first-order MP2 amplitudes as well as the Lagrangian multipliers for both
the TDHF coefficients and the first-order MP2 amplitudes are obtained by
solving appropriate response equations.
A constraint like in PED is not necessary in the QED method as a result of the
fact that the second derivative is with respect to Fα··· (−ω).
The PED and QED methods at the MP2 level and the SOPPA and CC2 linear
response function methods differ in the poles of the response functions.
The PED expression contains Hartree-Fock orbital energy differences as poles
and thus as excitation energies.
The QED method has the TDHF/RPA poles and double excitation poles
coming from the time-dependent first-order doubles correlation coefficients
t2µ [1](t).
This is a major drawback of this approach because it is not in agreement
with the pole structure of the exact response functions.
It is a consequence of including orbital relaxation in the Lagrangian, which
implies a two-step procedure, where first the TDHF equations are solved
and afterwards equations for the time-dependent t2µ [1](t) amplitudes.
The SOPPA and CC2 response functions are implicitly based on an
unrelaxed formalism and have the correct pole structure.

Cluster response functions defined as derivatives of a time-average

quasienergy Lagrangian including orbital relaxation have the same problem
with the pole structure.
Therefore one prefers to define coupled cluster response functions as
derivatives of a time-dependent quasienergy Lagrangian without orbital
relaxation

∂
LCC
0 (t, ~
F ) = hΦ SCF
0 | Ĥ(t) − ı~ eT̂ | ΦSCF
0 i
∂t

X ∂
+ λ2µ (t)hΦSCF
0 | d ĥ2µ e−T̂ Ĥ(t) − ı~ eT̂ | ΦSCF
0 i
2
∂t
µ
Taking the derivatives one obtains an expression for the coupled cluster
response function which is essentially the same as the one derived from
coupled cluster response theory but symmetrized with respect to the two
ω ω
property operators P̂ = Ôα··· and Ôβ··· and the sign of the frequency ω.

Chapter 13
Stephan P. A. Sauer
sauer@kiku.dk


13.1 Basis Sets - 1
The approximate electronic wavefunctions are built from molecular orbitals,

which are the solutions of the Hartree-Fock equations or their
multiconfigurational extension.
Molecular orbitals are expanded in a set of one-electron functions {χµ }, called
the basis set
X
φp (~r) = χµ (~r) cµp
µ
In most cases one uses Gaussian Type Orbitals (GTO)

~ 2
~ ~r) = C(x − Rx )k (y − Ry )m (z − Rz )n e−ζ(~r−R)
χζ,k,m,n (R,
here given as cartesian Gaussian type functions, where C is the normalization

~
constant and the basis function is placed at point R.
The sum of the exponents k, m and n is the angular momentum quantum
number l of this function.
Basis functions come in sets of functions having the same exponent ζ but all
possible combinations of the exponents k, m and n giving the same value of l.
13.1 Basis Sets - 2
Mathematically the set of functions only has to be complete apart from fulfilling
the same boundary conditions as the molecular orbitals.
This is not possible in practice, because it requires infinitely large basis sets.
Often one chooses therefore a more physical approach, where the basis
functions are placed at the atoms and are meant to simulate atomic orbitals of
the individual atoms in the molecule.
The expansion becomes then a linear combination of atomic orbitals (LCAO).
GTOs are not an ideal approximation to atomic orbitals, which would be
~ ~ 2
exponential functions e−ζ|~r−R| instead of Gaussian type functions e−ζ(~r−R) of
the distance from the atom R.~
GTOs fall off too quickly as compared to real atomic orbitals and have a
~ instead of the cusp of proper atomic
maximum at the nucleus, i.e. at ~r = R,
orbitals.
The most inner and outer parts of the atomic orbitals are therefore not well
described by GTOs.

13.1 Basis Sets - 3
The common solution to this problem is to combine several GTOs, which are
called primitive GTOs, in a contracted GTO, which is then meant to simulate
one atomic orbital.
The coefficients of this combination are preset and kept fixed during the
calculation of the molecular orbital coefficients cµp , i.e. the solution of the
Hartree-Fock equations.
The Gaussian type orbitals (contracted or not) are not atomic orbitals but just
basis functions, one still keeps the nomenclature and distinguishes between
valence orbitals: meant to describe the electrons in the most outer shell,
e.g. the 2s and 2p electrons in carbon or the 1s electron in hydrogen
core orbitals: meant to describe the inner electrons,
e.g. the 1s electrons in carbon
Minimal basis set: each core and valence orbital of an atom is represented
by a single primitive or contracted GTO
A minimal basis set is not suitable for any quantitative calculations, even if it
would consist of the proper atomic orbitals.
13.1 Basis Sets - 4
Proper atomic orbitals describe the electrons in an isolated atom with a

spherically symmetric potential.
This symmetry is broken in any molecule and the basis has to reflect this
leading to two different kinds of extension of a minimal basis set:
First, one represents each atomic orbital not by a single contracted or
primitive GTO, but by two, three or more GTOs with different exponents ζ.
They are called double zeta (DZ), triple zeta (TZ) and so forth basis sets.
Normally this is in particular important for the valence functions, whereas
the core orbitals are described by a single but then contracted GTO.
A basis set with a single contracted GTO for each core orbital and two,
three or more contracted or primitive GTOs for each valence orbital is called
a valence double zeta (VDZ), valence triple zeta (VTZ) ...
A typical example of a VDZ basis set is the Pople style basis sets 6-31G:
each core orbital is represented by a contraction of 6 primitive GTOs
the valence orbitals are represented by two basis functions: one
contraction of three primitive GTOs and a second but primitive GTO with a
smaller exponent ζ. Molecular Electromagnetism A Computational Chemistry Approach – p.5/33
13.1 Basis Sets - 5
However, this is often still not enough because it does not introduce the
required asymmetry in the basis set.
This can be achieved by including also basis functions with a higher angular
momentum quantum number l than the valence orbitals,
e.g. d-orbitals for carbon or p-orbitals for hydrogen.
They are called polarization functions and are important for smaller
molecules with high symmetry, whereas in large molecules with low symmetry
the atoms can partly “borrow" from each other.
In the Pople style basis sets this would lead to a 6-31G(d,p) or shorter 6-31G**
basis set, which has an additional set of d-type GTOs for all second and third
row atoms and an additional set of p-type GTOs for hydrogen.
A basis set can therefore be improved by both increasing the number of
valence orbitals as well as adding more polarization functions.
In a balanced basis set this should be done in such a way that both changes
lead to changes of equal importance, which normally means changes in the
electronic energy in the same order of magnitude.
13.1 Basis Sets - 6
The series of correlation consistent polarized basis sets cc-pVXZ (with the
cardinal number X = D, T, Q, 5, 6) by Dunning and co-workers and the series
of polarization consistent pc-n (with n = 0, 1, 2, 3, 4) basis sets by Jensen are
two prominent examples for this.
In the cc-pVXZ basis sets the criteria was the change in the energy of a
correlated wavefunction calculation
The pc-n basis sets are optimized for density functional theory calculations.
All these basis sets are optimized for the calculation of electronic energies and
are therefore able to represent the operators included in the field free
electronic Hamiltonian Ĥ (0) reasonably well.
In the calculation of molecular electromagnetic properties it is necessary also
to represent other operators such as the electric dipole operator, the electronic
angular momentum operator, the Fermi contact operator and more.
Most of these basis sets are a priori not optimized for this and have to be
extended.

13.1.1 Electric Properties - 1
The electric dipole moment operator contains the position vector ~r of the
electrons, which implies that the tail of the wavefunction becomes important.
This part is not well described in GTOs
It is therefore essential to include additional valence functions with very small
exponents ζ - so-called diffuse basis functions.
Pople style basis sets
6-31G+: one diffuse function is added for second and third row atoms
6-31G++ basis sets: one diffuse function is also added for hydrogen
Correlation consistent and polarization consistent basis sets:
one set of diffuse functions of each type present in the basis set is added in
the aug-cc-pVXZ and aug-pc-n version of these basis sets.
In the series of correlation consistent basis sets it is also possible to add
two or more sets of diffuse functions in the “d-aug", “t-aug" and so forth
versions.

The requirements on the basis set increase with the order of the molecular
property, i.e. from an electric dipole moment via the electric dipole
polarizability to the first dipole hyperpolarizability ...
Polarization functions, become increasingly important and the required
maximum angular momentum quantum number represented in the basis set
increases with the order of the property
A similar increase in the requirements on the basis set is observed, when one
goes from the electric dipole moment to the electric quadrupole moment and ...
The cc-pVXZ basis sets contain a systematically increasing amount of
polarization functions and increasing maximum angular momentum quantum
number which is always just one value smaller than the cardinal number X.
The cc-pVXZ basis sets are well suited for and frequently used in the
calculation of polarizabilities and hyperpolarizabilities.
Often one can observe a monotonic convergence of the results, which offer
the possibility to extrapolate to a complete basis set limit.
However, these basis sets become quickly very large with increasing X
The medium size polarized basis sets by Sadlej are a cheaper alternative.
They are essentially triple zeta basis sets with polarization functions.
The key to their success is the way the ζ exponents for the polarization
functions are chosen:
The polarization functions have the same exponents as some of the
valence orbitals with the highest angular momentum quantum number:
the d- and f-type functions on carbon have the same exponents as some of
the p-type functions.
The first-order correction to the perturbed orbitals of an atom in the
presence of an electric field would approximately consist of atomic orbitals
with the same exponent ζ but the angular momentum quantum number l
one higher and one lower than the original function.
The basis set includes already functions for the first-order correction to AOs
For larger molecules this leads sometimes to linear dependencies and
consequently convergence problems in the calculation of the molecular
orbitals.
The same problem can also be observed in calculations with the larger
correlation consistent basis sets.
Therefore, a modified version of the medium size polarized basis sets, called
reduced-size polarized (ZmPolX) basis sets, has recently been developed
which contains a smaller number of diffuse polarization functions.
A new version of the Karlsruhe valence double-, triple- and quadruple-zeta
basis sets variationally optimized for the calculation of polarizabilities was
presented

13.1.2 Magnetic Properties - 1
In the calculation of magnetizabilities and rotational g-factors one has the

magnetic dipole moment or angular momentum operators of the electrons.
Inclusion of enough diffuse functions is therefore equally important as for the
electric analogue, the polarizability.
It is absolutely necessary to include polarization functions with high enough
angular momentum.
The correlation consistent basis sets in their augmented form are frequently
used, but it is often necessary for convergence to go to higher cardinal
numbers than in the case of polarizabilities.
Furthermore one is faced with the gauge origin problem.
A popular solution to this problem is perturbation dependent basis sets.
For an external magnetic field as perturbation they are given as
~ ~r, B)
χζ,k,m,n (R, ~ = e 2~ { ~ GO −R)×
ıe (R ~ B~ }·~
r ~ ~r)
χζ,k,m,n (R,
called gauge including atomic orbitals (GIAO) or London orbitals because

Their first application to the calculation of NMR nuclear magnetic shielding

tensors, on the other hand, goes back to Hameka and Ditchfield.
The prefactor in a GIAO basis function corresponds actually to a unitary gauge
~ GO to the centre R
transformation, which moves the global gauge origin R ~ of
this basis function.
Using GIAO basis functions removes the dependence on the global origin from
the magnetizabilities and NMR nuclear magnetic shielding property integrals.
But introduces a dependence on the external magnetic field also in the
two-electron repulsion integrals.
The exponential prefactor in GIAOs introduces implicitly higher angular
momentum functions in the basis set as can be seen by expanding it
~ ~r, B)
~ = χζ,k,m,n (R,
~ ~r)+ ıe n
~ GO − R)
~ ×B
o
~ ·~r χζ,k,m,n (R,
~ ~r)+· · ·
χζ,k,m,n (R, (R
2~
Calculation of magnetic properties with GIAO basis sets exhibit a significantly
improved basis set convergence over regular basis sets.

The GIAO basis functions are thus just one example for the general idea of
distributed gauge origins.
Application of the same prefactor to molecular orbitals is the idea behind the
IGLO and LORG approaches to the calculation of magnetizabilities and NMR
nuclear magnetic shielding tensors.
The nuclear magnetic shielding tensor depends also the orbital paramagnetic
1
~ |3 dependence.
operator with its |~r −R
i K
Consequently also the inner region of the electron density, i.e. closer to the
nuclei, has to be described properly by the basis set.
It is therefore necessary to add additional functions with large exponents,
so-called tight functions.
This can partly be achieved by using the correlation consistent core-valence
cc-pCVXZ series of basis sets or their augmented version aug-cc-pCVXZ.
Jensen has also developed a modified version of his polarization consistent
basis sets, called pcS-n, which are optimized for DFT calculations of shielding
constants by including tight p-functions.
In the CTOCD-DZ approach for solving the gauge origin problem in shielding
calculations the diamagnetic contribution to the shielding tensor is
reformulated as a linear response function of the total electronic momentum
operator
The new CTOCD-DZ operator is essentially the electronic position vector times
the orbital paramagnetic operator.
Consequently it becomes important to add not only p-type functions with large
exponents as in the pcS-n basis sets but also d-type functions with large
exponents.
Sauer and co-workers have therefore developed a modified version of the
aug-cc-pCVTZ basis sets, called aug-cc-pCVTZ-CTOCD-uc, where such
functions were added.

13.1.3 Electron Spin Dependent Properties - 1
The NMR indirect nuclear spin-spin coupling constants and the ESR hyperfine
coupling constants are probably the extreme cases of additional basis set
requirements in molecular property calculations.
The Fermi contact operator includes a Dirac δ function which means that only
the electron density at the coupled nuclei contributes.
GTO basis functions, being Gaussian functions, have the fundamentally wrong
behavior in that region: a maximum instead of a cusp.
It is necessary important to have s-type GTOs with very large exponents in the
basis set.
This was already realized by Schulman and Kaufman or Kowalewski and later
by Oddershede and co-workers or Guilleme and San Fabián .
Several specialized spin-spin coupling constants basis sets have been
developed which are modifications of the correlation consistent basis sets.

Helgaker and co-workers:

In the cc-pVXZ-Cs series the cc-pVXZ basis sets were extended with the
core-valence s-type functions of the cc-pCVXZ basis sets.
In the cc-pVXZ-sun basis sets the s-type functions of the cc-pVXZ basis
sets were decontracted and augmented with a series of n s-type functions
with increasingly larger exponents.
The aug-cc-pVTZ-J basis sets by Sauer and co-workers:
The aug-cc-pVTZ basis set was totally uncontracted, 4 s-type functions with
increasingly larger exponents and for third row atoms also 3 d-type
functions with large exponents were added.
The finally basis set was recontracted using the molecular orbital
coefficients of suitable test molecules as contraction coefficients and the
most diffuse second polarization function was removed.
The resulting basis set was shown to reproduce results of DFT calculations
for spin-spin coupling constants obtained with much larger basis sets.

Jensen and co-workers:

generated modifications of the pc-n and cc-pVXZ basis sets, called
respectively pcJ-n and ccJ-pVXZ, where not only tight s-type but also tight
p- and d-type functions were added.
These basis sets are to be preferred over the basis sets with only tight
s-type functions, whenever the spin-dipolar and orbital paramagnetic
contributions to the indirect spin-spin coupling constants are equally or
more important than the Fermi contact contribution.

13.2 Reduced Linear Equations - 1
General structure of a propagator

ω ω ω
hh P̂α ; Ôβ··· iiω = T̃ (P̂α )(~ωS − E)−1 T (Ôβ··· ) = T̃ (P̂α )X(Ôβ··· )
ω
“property gradient” vectors T̃ (P̂α ) and T (Ôβ··· )
“Hessian” or principal propagator matrix (~ωS − E)
Inverse is never evaluated
ω
Set of coupled linear equations for the solution vector X(Ôβ··· )
ω ω
(~ωS − E) X(Ôβ··· ) = T (Ôβ··· )
solved iteratively by expanding X in a basis of trial vectors {bi }

ω
X
X(Ôβ··· ) = bi c i
i

Each iteration the linear equations are transformed to the basis of the
trialvectors {b1 · · · bn } to the reduced space
(~ωS R − E R ) X R (Ôβ···
ω
) = T R (Ôβ···
ω
)
Equations are solved in the reduced space

X R (Ôβ···
ω ω
) are the optimal coefficients {ci } for X(Ôβ··· )
Matrices in the reduced space
R
Sij = bi T S bj
R
Eij = bi T E bj
ω
TiR (Ôβ··· ) = bi T T (Ôβ···
ω
)

Convergence if the residual vector is smaller than a given threshold

ω ω
Rn = (~ωS − E) X n (Ôβ··· ) − T (Ôβ··· )
New trial vector from conjugate gradient method
bn+1 = (~ωS diag − E diag )−1 Rn
In each iteration: linear transformation of the new trial vector (~ωS − E)bi
Linear transformation are done directly
RPA and SOPPA: linear transformations in AO basis possible

13.3.1 Electric Dipole Moment
Molecule SCF unrelaxed MP2 relaxed MP2 Exp.

HF 0.7570 0.6994 0.7100 0.7094
HCl 0.4725 0.4455 0.4419 0.4305
HBr 0.3777 0.3375 0.3417 0.3219
CH3 F 0.8443 0.7521 0.7380 0.7312
CH3 Cl 0.8095 0.7271 0.7589 0.7461
CH3 Br 0.8456 0.7524 0.7462 0.7162
(in a.u. ≈ 8.478358 × 10−30 C m)
Results are clearly improved by unrelaxed MP2 correction to the density matrix
Correlation at this level reduces the dipole moments on average by 9 %.
RMS % deviation of the unrelaxed MP2 results is 3.6 %, maximum 5.0 %
minimum -1.4 %

13.3.2 Isotropic Dipole Polarizability - 1
RPA/SCF SOPPA SOPPA(CCSD) CCSD MP2 MP4 Exp.

HF 4.874 6.085 5.818 5.724 5.674 5.770 5.60
HCl 16.664 17.671 17.352 17.499 17.368 17.433 17.39
H2 O 8.492 10.319 9.939 9.824 9.792 9.866 9.64
H2 S 23.614 24.922 24.343 24.604 24.570 24.542 24.71
NH3 12.926 14.736 14.366 14.411 14.432 14.411 14.56
PH3 29.915 31.120 30.184 30.674 30.689 30.510 30.93
CH4 16.120 16.853 16.520 16.709 16.754 16.704 17.27
SiH4 29.960 31.414 30.742 31.467 31.035 31.216 31.90
F2 8.593 8.903 8.525 8.550 8.219 8.662 8.38
Cl2 29.886 31.346 30.556 30.905 30.556 30.707 30.42
C2 H4 28.303 28.329 27.482 27.534 27.793 27.635 27.70
CO2 15.841 19.444 18.726 18.013 17.884 17.846 17.51
SO2 23.653 28.659 27.407 26.444 26.174 26.343 25.61
in units of e2 a20 Eh−1 using Sadlej’s basis sets

13.3.2 Isotropic Dipole Polarizability - 2
The maximum difference between the results at the highest level, CCSD, and
the experimental results is 3% or 0.83 a.u. for SO2 .
However, one should remember that the calculated results are for a fixed
equilibrium geometry while the experimental results are for the vibrational
ground state.
The RPA polarization propagator results and SCF second derivative results
are identical illustrating the equivalence of these two approaches.
Concerning the effect of electron correlation one can see that at the CCSD
level the effect varies between 1% and 17% or -0.04 and 2.79 a.u.
The difference between the unrelaxed CCSD linear response and the relaxed
MP4 results is very small.
The agreement with the CCSD linear response results is in most cases better
in the SOPPA(CCSD) than in the SOPPA approach.

13.3.3 Nuclear Magnetic Shielding Constants
SCF MP2 MP3 CCSD CCSD(T) Experiment

19 F
HF σ 413.6 424.2 417.8 418.1 418.6 419.7 ± 6
17
H2 O σ O 328.1 346.1 336.7 336.9 337.9 337 ± 2
15 N
NH3 σ 262.3 276.5 270.1 269.7 270.7 273.3 ± 0.1
13 C
CH4 σ 194.8 201.0 198.8 198.7 198.9 198.4 ± 0.9
19
F2 σ F -167.9 -170.0 -176.9 -171.1 -186.5 -192.8
15 N
N2 σ -112.4 -41.6 -72.2 -63.9 -58.1 -59.6 ± 1.5
13 C
CO σ -25.5 20.6 -4.2 0.8 5.6 2.8 ± 0.9
17
σ O -87.7 -46.5 -68.3 -56.0 -52.9 -56.79 ± 0.59
in ppm calculated as analytical derivatives with GIAO basis functions.
For molecules with single bonds the differences between the MP3, CCSD and CCSD(T)
results are rather small
For molecules with multiple bonds there are still very large changes between the CCSD
and CCSD(T) results.
MP2, on the other hand, typically overestimates the correlation correction and the MP2
results are sometimes in not much better agreement with higher level methods or
experimental results than the SCF results.
13.3.4 Indirect Nuclear Spin-Spin Coupling Constants - 1
RPA MCRPA SOPPA SOPPA(CCSD) Experiment
N2 1 J 15 N −14 N -14.9 0.8 2.7 2.1 1.4 ± 0.6

CO 1 J 13 C−17 O -5.7 16.1 20.4 18.6 15.6 ± 0.1
C2 H2 1 J 13 C−13 C 409.5 188.1 189.3 188.7 185.04
HF 1 J 1 H−19 F 666.9 543.7 539.5 529.4 540
H2 O 1 J 1 H−17 O -103.4 -83.9 -82.4 -80.6 -83.04 ± 0.02
NH3 1 J 1 H−15 N -78.4 -62.2 -62.4 -62.1
CH4 1 J 1 H−13 C 156.9 135.7 126.9 122.3 120.87 ± 0.05
C2 H2 1 J 1 H−13 C 411.1 247.3 262.9 253.6 242.70
SiH4 1 J 1 H−29 Si -241.2 — -202.5 -192.1 -193.3 ± 0.6
H2 O 2 J 1 H−1 H -22.4 -9.6 -9.1 -8.8 -7.8 ± 0.7
NH3 2 J 1 H−1 H -24.4 -11.2 -11.9 -11.3
CH4 2 J 1 H−1 H -27.0 -20.8 -15.3 -14.0 -11.878 ± 0.004
C2 H2 2 J 1 H−13 C -49.9 53.0 52.6 51.7 53.82
SiH4 2 J 1 H−1 H -1.2 — 2.1 2.6 2.62 ± 0.08
C2 H2 3 J 1 H−1 H 84.9 11.2 12.2 11.3 10.89
in Hz
J OP RPA MCRPA SOPPA(CCSD)
N2 1 J 15 N −14 N 0.43 2.69 3.00

CO 1 J 13 C−17 O 11.81 12.89 14.11
C2 H2 1 J 13 C−13 C 15.05 6.67 6.34
HF 1 J 1 H−19 F 195.05 182.0 189.82
H2 O 1 J 1 H−17 O -12.27 -11.45 -11.51
NH3 1 J 1 H−15 N -3.07 -2.93 -2.97
CH4 1 J 1 H−13 C 1.47 1.48 1.50
C2 H2 1 J 1 H−13 C -3.60 -0.84 -0.85
SiH4 1 J 1 H−29 Si 0.47 — 0.44
H2 O 2 J 1 H−1 H 9.09 9.23 9.31
NH3 2 J 1 H−1 H 6.24 6.19 6.24
CH4 2 J 1 H−1 H 3.73 3.59 3.72
C2 H2 2 J 1 H−13 C 8.28 5.58 5.60
SiH4 2 J 1 H−1 H 2.35 — 2.34
C2 H2 3 J 1 H−1 H 5.54 4.80 4.81

J SD RPA MCRPA SOPPA JFC RPA MCRPA SOPPA

(CCSD) (CCSD)
N2 1 J 15 N −14 N -7.84 -1.95 -1.76 -7.49 -0.53 0.79

CO 1 J 13 C−17 O -9.07 -4.77 -4.37 -8.53 3.90 8.76
C2 H2 1 J 13 C−13 C 29.06 9.04 8.46 365.35 172.34 173.92
HF 1 J 1 H−19 F -11.73 -1.41 -0.94 483.62 363.2 340.50
H2 O 1 J 1 H−17 O -0.01 -0.41 -0.47 -91.12 -72.08 -68.56
NH3 1 J 1 H−15 N -0.02 -0.18 -0.18 -75.22 -58.98 -58.87
CH4 1 J 1 H−13 C -0.21 0.02 0.03 155.42 123.53 120.58
C2 H2 1 J 1 H−13 C 3.04 0.43 0.43 411.41 247.40 253.73
SiH4 1 J 1 H−29 Si 0.05 — -0.05 -241.78 — -192.43
H2 O 2 J 1 H−1 H 1.25 1.03 0.89 -25.50 -12.70 -11.87
NH3 2 J 1 H−1 H 0.91 0.68 0.67 -26.30 -12.82 -12.94
CH4 2 J 1 H−1 H 0.46 0.35 0.36 -27.68 -15.73 -14.53
C2 H2 2 J 1 H−13 C -1.52 1.02 0.98 -55.25 47.77 46.47
C2 H2 3 J 1 H−1 H 3.02 0.57 0.59 79.93 9.43 9.49
SiH4 2 J 1 H−1 H 0.09 — 0.07 -1.22 — 2.59
The experimental data are extrapolated to the equilibrium geometry by

subtracting calculated ro-vibrational corrections from the experimental values.
One-bond C-C, C-H, N-H, Si-H and to a lesser degree also O-H coupling
constants are dominated by the Fermi contact contribution.
For two-bond H-H coupling constants the Fermi contact contribution is still the
largest contribution. However, the orbital paramagnetic and diamagnetic
contributions are often of the same order of magnitude as the Fermi contact
contribution but partially cancel each other, because they have often opposite
signs.
Only in the case of coupling constants involving atoms with many lone-pairs
like F or couplings across double and triple bonds one should expect that the
orbital paramagnetic and/or the spin-dipolar contributions will make a
significant contribution to the total coupling constant.

The agreement of the SOPPA(CCSD) results with the experimental values is

very good for some of the indirect nuclear spin-spin coupling constants.
For most molecules the deviations are less than 4 Hz with a rms value of
1 1 H−13 C
2.3 Hz (5.1 Hz with HF and J in C2 H2 ), apart from HF and
1 1 13
J H− C in C2 H2 , where the deviations are about 10 Hz.
The rms value of the percentage deviations is for all molecules 18% (10 %
1 1 H−13 C
without HF and J in C2 H2 ).
Spin-spin coupling constants are more difficult to calculate than dipole
polarizabilities and nuclear magnetic shielding constants.
The correlation corrections for the molecules here vary between 20 % for HF
and 426 % for CO.
The spin-dipolar and the Fermi contact contribution, depend on excited triplet
states, which uncorrelated methods like RPA are often not able to describe
properly.
The large correlation effects in the spin-spin coupling constants are thus
normally due to the Fermi contact term.
13.4 Examples for Vibrational Averaging Effects - 1
ZPVC to nuclear magnetic shielding constants (in ppm)

Molecule Property result at Re ZPVC %
19
HF σ F 419.68 -10.01 2.4
1H
HF σ 29.01 -0.32 1.1
17 O
H2 O σ 343.94 -9.86 2.9
1
H2 O σ H 30.97 -0.48 1.6
19 F
F2 σ -187.84 30.90 16.5
13
C2 H2 σ C 128.89 -3.78 2.9
1H
C2 H2 σ 30.45 -0.80 2.6
13 C
CO σ 5.29 -1.82 34.5
17
CO σ O -53.5 -4.8 9.0
15 N
N2 σ -58.7 -3.5 5.9

ZPVC to indirect nuclear spin-spin coupling constants (in Hz)

Molecule Property result at Re ZPVC %
HF 1 J 1 H−19 F 526.4 -26.9 5.1
H2 O 1 J 1 H−17 O -81.555 3.963 4.9
H2 O 2 J 1 H−1 H -8.581 0.653 7.6
NH3 1 J 1 H−15 N -61.968 0.341 0.5
NH3 2 J 1 H−1 H -10.699 0.537 5.0
CH4 1 J 1 H−13 C 123.846 5.030 4.1
CH4 2 J 1 H−1 H -14.450 -0.686 4.7
SiH4 1 J 1 H−29 Si -129.059 -7.585 3.8
C2 H2 1 J 13 C−13 C 189.995 4.861 1.9
C2 H2 1 J 1 H−13 C 254.906 -9.212 4.9
C2 H2 2 J 1 H−13 C 51.727 -3.237 6.3
C2 H2 3 J 1 H−1 H 11.311 -1.184 10.5

ZPVCs for NMR properties can amount to 10% and are typically larger for the
coupling than for the shielding constants.
ZPVCs are smaller but not significantly smaller than correlation effects
They have to considered in a high level calculation of NMR parameters.
The large zero point vibrational corrections to the nuclear magnetic shielding
constants of F2 and CO are two extreme cases, well known in the literature.
1 1
The corrections to geminal hydrogen-hydrogen couplings, 2 J H− H , are
percent wise larger than the corrections to other couplings in the same
molecule.

Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF

Uploaded by

Copyright:

Available Formats

You might also like

Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF

Uploaded by

Document Information

Original Description:

Original Title

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

Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF

Uploaded by

Copyright:

Available Formats

Molecular Electromagnetism

A Computational Chemistry Approach

Molecular Electromagnetism A Computational Chemistry Approach – p.1/5

Experiments carried out on molecules can be understood as interaction of the

Molecular Electromagnetism A Computational Chemistry Approach – p.2/5

Interactions are often described in terms of molecular properties

Molecular Electromagnetism A Computational Chemistry Approach – p.3/5

Molecular properties play thus an important role in the interpretation of

Molecular Electromagnetism A Computational Chemistry Approach – p.4/5

Molecular properties can be determined to high accuracy in experiments

Molecular Electromagnetism A Computational Chemistry Approach – p.5/5

Molecular Electromagnetism A Computational Chemistry Approach – p.1/47

A complete quantum mechanical treatment of the interaction of molecules and

Molecular Electromagnetism A Computational Chemistry Approach – p.2/47

We use only non-relativistic quantum mechanics, i.e. the Schrödinger

Molecular Electromagnetism A Computational Chemistry Approach – p.3/47

The total non-relativistic Hamiltonian for a molecule with N electrons and M

Molecular Electromagnetism A Computational Chemistry Approach – p.4/47

With this trial-solution, the time-dependent Schrödinger equation separates in

and an equation for the time-dependent phase factor ϑ(t).

Molecular Electromagnetism A Computational Chemistry Approach – p.5/47

The time-independent Schrödinger equation is a second-order partial

It can be interpreted as an eigenvalue equation.

The probability density of the time-dependent wavefunction is constant in time.

The eigenfunctions of the time-independent Schrödinger equation are

Molecular Electromagnetism A Computational Chemistry Approach – p.6/47

The average measured value < P >t of a physical observable P obtained in a

~ K }, {~ri }, t) | P̂ | Φ(0) ({R

For the stationary states the expectation value of a time-independent

~ K }, {~ri }) | P̂ | Φ(0) ({R

Molecular Electromagnetism A Computational Chemistry Approach – p.7/47

Molecular Electromagnetism A Computational Chemistry Approach – p.8/47

With this operator we obtain the time-independent field-free electronic

(0) ~ K })i = E (0) ({R

When we insert this approximate trial wavefunction in the time-independent

(0) ~ K }), fulfills the role of the potential energy

Max Born’s interpretation of the wavefunction of an electron :

|ψ(~r)|2 = ψ ∗ (~r) ψ(~r)

is the probability density of finding the electron at the point ~r.

Molecular Electromagnetism A Computational Chemistry Approach – p.11/47

Electrons are indistinguishable and the probability of finding electron 2 in a

where P (~r) is the electron density.

we can extend the integration to include all electrons and write

Molecular Electromagnetism A Computational Chemistry Approach – p.12/47

Reversing the indistinguishability of the electrons argument we can write the

a generalization of the definition of the electron density P (~r)

Molecular Electromagnetism A Computational Chemistry Approach – p.13/47

Using the continuity equation of classical electromagnetism

The time derivative of the wavefunction is given by the time-dependent

Molecular Electromagnetism A Computational Chemistry Approach – p.14/47

and that the terms in the summation for i = 2 · · · N vanish we arrive at

Molecular Electromagnetism A Computational Chemistry Approach – p.15/47

Comparison with the continuity equation leads to the current density

For a stationary state this reduces to

Molecular Electromagnetism A Computational Chemistry Approach – p.16/47

are often more convenient to work with.

One possible set of expressions for the potentials of a static and