Professional Documents
Culture Documents
Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF
Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF
Molecular Electromagnetism A Computational Chemistry Approach - Stephan P.A. Sauer PDF
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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)
= |Φ(0) ({R
~ K }, {~ri })|2
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|
4πǫ0 iK |~ri − R 4πǫ0 i<j |~ri − ~rj |
e2 X ZK ZL
+
~K − R
4πǫ0 K<L |R ~ L|
(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
Molecular Electromagnetism A Computational Chemistry Approach – p.9/47
2.2 The Born-Oppenheimer Approximation - 3
(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
(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
∂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
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)
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
~ˆi | Ψk ({~ri }, t)i
(0) (0)
X
= ∇ hΨk ({~ri }, t) | δ(~ri − ~r)p
∂t 2me i
N
!
e
~ˆi | Ψk ({~ri }, t)i∗
(0) (0)
X
+ hΨk ({~ri }, t) | δ(~ri − ~r)p
2me i
~ 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)
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.
~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
~ = m d~v
F
dt
~ 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
H(~r, p
~, t) = p
~ · ~v − L(~r, ~v , t)
G(T, P ) = H(S, P ) − T S
∂L(~r, ~v , t)
pα =
∂vα
~ r, t)
~ = me ~v − eA(~
p
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
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
∂L(~r, ~v , t) me vα
pα = = q − e Aα (~r, t)
∂vα 2
1 − ~vc2
me c2
H= q − e φ(~r, t)
2
1 − ~vc2
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
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
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
~ˆ r)
Let’s assume time-independent potentials φ̂E (~r) and A(~
and collect the time dependence of the wavefunction in a phase factor
|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
E + e φ̂(~r) + me c2 = 2 me c2 + E N R + e φ̂(~r)
Using only the first term, the equation for the large component reads
( )( )
1 X h i X h i
σ α p̂α + e Âα (~r) σ α p̂α + e Âα (~r) |ψ̄L 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
σ α p̂α + e Âα (~r) σ α p̂α + e Âα (~r) |ψ̄L 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
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.
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~vˆi = p ~ˆ ri )
~ˆi + eA(~
(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.
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.40/47
2.9 Gauge Transformations - 1
~ r, t)
∂ A(~
~ r , t)
E(~ = ~
−∇φ(~r, t) −
∂t
~ r , t)
B(~ = ~ × A(~
∇ ~ 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
Molecular Electromagnetism A Computational Chemistry Approach – p.41/47
2.9 Gauge Transformations - 2
~ˆ i
~ˆ i = −ı~∇
p
ˆi′ | Ψ′ i = hΨ | ˆi | Ψi
′
X X
hΨ | ~π ~π
i i
~ 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)
~ ′ (~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.
Molecular Electromagnetism A Computational Chemistry Approach – p.45/47
2.9 Gauge Transformations - 6
∂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) ~ˆi 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) ~ˆi (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
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
~ ) |Ψ0 (F)i
Ĥ(F ~ = E0 (F
~ ) |Ψ0 (F)i
~
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···
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···
~
∂ 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
(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
Molecular Electromagnetism A Computational Chemistry Approach – p.6/74
3.2 Time-Independent Perturbation Theory - 2
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.
(0)
The unperturbed wavefunction |Ψ0 i is normalized
(0) (0)
hΨ0 | Ψ0 i = 1
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.
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) ~
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
Molecular Electromagnetism A Computational Chemistry Approach – p.10/74
3.2 Time-Independent Perturbation Theory - 6
(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
(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
Molecular Electromagnetism A Computational Chemistry Approach – p.13/74
3.3 Time-Independent Response Theory - 1
~ ) | 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.
~ ) 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.
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···
~)
∂ 2 Ĥ(F FF FF
= Ôαβ··· + Ôβα···
∂Fβ··· ∂Fα···
Molecular Electromagnetism A Computational Chemistry Approach – p.16/74
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.17/74
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
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
~ = 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)
~ )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
∂ (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 ~
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
hΨ(0)
m | [P̂ , Ĥ
(0)
] | Ψ(0) (0) (0) (0) (0)
n i = (En − Em )hΨm | P̂ | Ψn i
~ˆ r , Ĥ (0) ] = ı~ O
Recalling the well-known commutator relation [O ~ˆ p ,
me
~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
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
Molecular Electromagnetism A Computational Chemistry Approach – p.26/74
3.8 The Interaction Picture - 2
we consider only time-dependent electric fields here and have no Ĥ (2) (t)
Perturbation expansion of the time dependent wavefunction
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···
∂ 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 ?
∂ 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
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~ −∞ −∞
(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.
(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~ −∞
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.39/74
3.10 Transition Probabilities and Rates - 4
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···
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··· .
Molecular Electromagnetism A Computational Chemistry Approach – p.40/74
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~ −∞
Molecular Electromagnetism A Computational Chemistry Approach – p.41/74
3.11 Time Dependent Response Theory - 2
~ | 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~ −∞
~ )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~
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···
′ 1
h i
I F,I ′ (0) I F,I ′ (0)
hh P̂ (t) ; Ôβ··· (t ) ii = Θ(t − t ) hΨ0 | P̂ (t), Ôβ··· (t ) | Ψ0 i
ı~
′ 1
h i
I F,I ′ (0) I F,I ′ (0)
hh P̂ (t) ; Ôβ··· (t ) ii = Θ(t − t ) hΨ0 | P̂ (t), Ôβ··· (t ) | Ψ0 i
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 !
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
~ | 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···
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
Molecular Electromagnetism A Computational Chemistry Approach – p.47/74
3.11 Time Dependent Response Theory - 7
∂ Ĥ( ~
F) X FF
~) =
P̂ (F F
= Ôα··· + FF
Ôαβ··· + Ôβα··· Fβ···
∂Fα···
β···
~ | 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···
+ ···
~ ) | 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···
~ ) | 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···
We can identify the static response function as the first derivative of the
first-order correction to an perturbed expectation value,
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~ −∞ −∞
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~
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~
~ )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···
+ ··· .
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
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
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
(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
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
]
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
Polarization propagator
ω † b (0) −1 ω
hh P̂ ; Ôβ··· iiω = P̂ | (~ω Ib − H ) Ôβ···
(0)
All other states |Ψn i of the system must be generated by operating on the
(0)
reference state |Ψ0 i.
Molecular Electromagnetism A Computational Chemistry Approach – p.63/74
3.12 Matrix Representation of the Propagator - 10
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
)
Inserting this in the superoperator binary products between the complete set
of operators arranged as a matrix, (ĥ | ĥT ),
and
ω (0) ω (0)
T j (Ôβ··· ) = hΨ0 | [ĥ†j , Ôβ··· ] | Ψ0 i
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
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
Molecular Electromagnetism A Computational Chemistry Approach – p.68/74
3.12 Matrix Representation of the Propagator - 15
(E − ~ωS) R = 0
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
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
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
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
~ = 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
+ ...
!
~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
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
∂ 2 φ E (~r)
1X
+ (rα − RO,α )(rβ − RO,β ) + ...
2 ∂rα ∂rβ ~r=R ~O
αβ
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
~ 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
~ O) −
E
X
~ ~ 1X ~ O )Eαβ (R
~ O) + . . .
E = q φ (R µα (RO )Eα (RO ) − Qαβ (R
α
2
αβ
~ O) ∂E
µα ( R = −
~ O)
∂Eα (R
~ O) ∂E
Qαβ (R = −2
~ O)
∂Eαβ (R
~ O) ∂E
Θαβ (R = −3
~ O)
∂Eαβ (R
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
We need the perturbing Hamiltonian operator Ĥ (1) = qr φ̂E (~r) for the
P
~ 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
~ 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)
∂ 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
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
~ = µα + µind
µα (E) ~
α (E)
~ = µα +
X 1X 1X
µα (E) ααβ Eβ + βαβγ Eβ Eγ + γαβγδ Eβ Eγ Eδ + . . .
β
2 βγ 6 βγδ
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
~
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
Similarly one defines two isotropic averages for the first hyperpolarizability
1X
βk = (βzαα + βαzα + βααz )
5 α
1X
β⊥ = (2βzαα − 3βαzα + 2βααz )
5 α
1 X
γ⊥ = (2γαββα − γααββ )
15 αβ
Molecular Electromagnetism A Computational Chemistry Approach – p.23/31
4.5 Quantum Mechanical Expressions for α - 1
∂ Ĥ (1) ~ O)
~ O ) + Ω̂Eα (R
= µ̂α (R
∂Eα
~ 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.
~ = hΨ0 (E)
µα (E) ~ | µ̂α | Ψ0 (E)i
~
~
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
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
~K = ZK e E~ρ (R
F ~K)
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
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)
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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.
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 |
~ · ~j = 0
for an arbitrary function f (~r) and a steady current distribution ∇
Choosing f = rα gives Z
jα (~r) d~r = 0
~
r
or
j ~ µ0 ~ −R
(R ~ GO )
~
A (R) = ~ ×
m + ... .
4π ~ ~
| R − RGO | 3
~ 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)
~ 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
+ ...
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
~ = −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
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
N N
sB
X ge e ge e X
Ôα = ôsB
i,α =− Ŝα = − ŝi,α
i
2me 2me i
Molecular Electromagnetism A Computational Chemistry Approach – p.8/52
5.3 Quantum Mechanical Expression for m
~ -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
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∗
~ˆ | Ψi = hΨ | L
hΨ | L ~ˆ ∗ | Ψi
~ˆ | Ψi = −hΨ | L
hΨ | L ~ˆ | Ψi
~ 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
~
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
~ = E (0) −
X 1X
E(B) mα Bα − ξαβ Bα Bβ + . . .
α
2 αβ
BB ~ e2 X h 2
~ GO ) δαβ − (ri,α − RGO,α )(ri,β − RGO,β )
i
Ôαβ (RGO ) = (~ri − R
8me i
Molecular Electromagnetism A Computational Chemistry Approach – p.16/52
5.5 Quantum Mechanical Expression for ξ - 2
(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
∂ 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
~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
~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
~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
where
N
FC ge eµ0 X ~ K )ŝi,α
ÔK,α =− δ(~ri − R
3me i
h i
ge e µ0 X sˆi · (~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
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 sˆi · (~
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
~ 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
KL h KL
Kαβ = J αβ
µ2N gK gL
the well know expression of NMR spectroscopy for the local field at nucleus K
Molecular Electromagnetism A Computational Chemistry Approach – p.27/52
5.7 Induced Magnetic Fields and NMR Parameters - 3
K 1X K
σ = σαα
3 αα
~K =
where m gK µN
~
I~K and the z-component of the nuclear spin is quantized,
Iz = mI K ~
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.28/52
5.7 Induced Magnetic Fields and NMR Parameters - 4
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.
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
~ 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
(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
(2)
XX mK mL
Ĥ = Ôαβ mK L
α mβ
KL αβ
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
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
j ~ ~ m
KL ∂Bα (RK , B, ~ L)
K =
∂mL β
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
1 h
ˆ
~ (Rµ ~K) × m ˆ (Rl ~ GO )
i
p
= hh O ~ ; Ôβ iiω=0
me c2 α
Molecular Electromagnetism A Computational Chemistry Approach – p.42/52
5.9 SOS Expression for Diamagnetic Terms - 2
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
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
Molecular Electromagnetism A Computational Chemistry Approach – p.44/52
5.9 SOS Expression for Diamagnetic Terms - 4
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
~ r) = ∇
B(~ ~ ×A
~ B (~r)
~ 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 ~
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
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.52/52
Molecular Electromagnetism
A Computational Chemistry Approach
Chapter 6
Stephan P. A. Sauer
sauer@kiku.dk
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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
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).
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
Molecular Electromagnetism A Computational Chemistry Approach – p.4/58
6.1 Molecular Rotation as Source for Magnetic Moments - 3
mJα (J)
~ = mJ,rig
α
~ + mJ,ind
(J) α
~
(J)
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
Molecular Electromagnetism A Computational Chemistry Approach – p.5/58
6.1 Molecular Rotation as Source for Magnetic Moments - 4
~ J)
~ = −
X J ~ =− µ N
X
∆E(B, Bα mα (J) Bα gJ,αβ Jβ
α
~ αβ
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.
~ 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
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
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
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 .
Molecular Electromagnetism A Computational Chemistry Approach – p.11/58
6.2 QM Expression for the Rotational g Tensor - 5
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
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
el 4mp me l ~ l ~
gJ,αβ = hh m̂α ( R CM ) ; m̂β (RCM ) iiω=0
e2 Iββ
Molecular Electromagnetism A Computational Chemistry Approach – p.15/58
6.2 QM Expression for the Rotational g Tensor - 9
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
Molecular Electromagnetism A Computational Chemistry Approach – p.17/58
6.3 Rotational g Tensor and Electric Dipole Moment - 2
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.
′ ′
gJ,αβ Iββ = gJ,αβ Iββ
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
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
′ ′
gJ,αβ Iββ = gJ,αβ Iββ
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.20/58
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
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)
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
Molecular Electromagnetism A Computational Chemistry Approach – p.24/58
6.5 Molecular Rotation as Source for Magnetic Fields - 3
~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
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
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.
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
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
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
6.6 QM Expression for the Spin Rotation Tensor - 6
K,el µN gK 2me OP
Cαβ = hh ÔK,α ; m̂lβ (R
~ K ) iiω=0
2πIββ e
Molecular Electromagnetism A Computational Chemistry Approach – p.32/58
6.6 QM Expression for the Spin Rotation Tensor - 7
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
dia ~ me gK ~ K K,nuc
σαβ = σαβ (R K ) + Cαβ − Cαβ
mp 2πIββ
~2 ∂ 2 1 h
ˆ ˆ
i2 1
~ˆi · 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.
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)
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
~ˆi · p
~ˆj | Ψ0 ({~ri }; R)i
(0) (0)
X
+ hΨ0 ({~ri }; R) | p
2(mA + mB ) i,j
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)
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
Molecular Electromagnetism A Computational Chemistry Approach – p.41/58
6.7 Non-adiabatic Reduced Masses - 9
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.42/58
6.7 Non-adiabatic Reduced Masses - 10
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)
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
Molecular Electromagnetism A Computational Chemistry Approach – p.44/58
6.7 Non-adiabatic Reduced Masses - 12
ZA m2B + ZB m2A
gnnuc = mp
(mA + mB )mA mB
where the rotational and vibrational g factors in atomic masses are defined as
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
(RB,z − RCM,z ) p
P̂zA = P̂R + Ôz
R
(RA,z − RCM,z ) p
P̂zB = P̂R + Ôz
R
where we have chosen L̂⊥ = L̂x . Molecular Electromagnetism A Computational Chemistry Approach – p.50/58
6.8 Partitioning of the g Factors - 3
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
The second terms are the electronic contributions to the rotational and
vibrational g factors, which we replace by the total g factors.
Molecular Electromagnetism A Computational Chemistry Approach – p.51/58
6.8 Partitioning of the g Factors - 4
The last three terms are static response functions involving the total electronic
momentum operator O ~ˆ
~ˆ p and another operator O.
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~
~ 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
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.
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.
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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.
~ω sin(~k · ~r − ωt)
~ r , t) = ı2A
A(~
~ 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
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
~
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.
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
~
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
+ ...
The Fourier components of the operator and the field are then given as
ω
Ôβ··· = − µ̂β
Fβ··· (ω) = Eβω1
~
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
which for ω1 = 0, i.e. for the static polarizability, reduces to the expression
obtained by static response theory or Rayleigh-Schrödinger perturbation
theory.
∆nr (ω1 ) = nL R
r (ω1 ) − nr (ω1 )
~ 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
Molecular Electromagnetism A Computational Chemistry Approach – p.15/52
7.3 Optical Rotation - 2
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
′ 1X ′
G (−ω1 ; ω1 ) = Gαα (−ω1 ; ω1 )
3 α Molecular Electromagnetism A Computational Chemistry Approach – p.17/52
7.3 Optical Rotation - 4
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 ).
~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)
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
+···
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
and can see that for ω1 = 0 the frequency-dependent mixed electric dipole
magnetic dipole polarizability tensor vanishes.
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
π e 2 (0) 2 X
(0) r (0) 2
(1) (0) ω 2
Wn0 = En − E0 hΨn | Ôα | Ψ0 i (Aαn0 )
2 ~ α
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 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
Molecular Electromagnetism A Computational Chemistry Approach – p.28/52
7.4 Excitation Energies & Transition Moments - 6
(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
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)
n ihΨn | µ̂β | Ψ0 i
~ 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
Molecular Electromagnetism A Computational Chemistry Approach – p.31/52
7.4 Excitation Energies & Transition Moments - 9
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
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
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
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.36/52
7.5 Dipole Oscillator Strength Sums - 2
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
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 )
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
for k = 0, 1, 2, 3, · · ·
Molecular Electromagnetism A Computational Chemistry Approach – p.40/52
7.5 Dipole Oscillator Strength Sums - 6
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, · · ·
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
ˆA 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
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
where the double sum runs over all complex states in which at least one of the
molecules is excited.
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
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
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
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
C6AB
E0disp
A ,0B
=−
|RAB |6
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 }.
Molecular Electromagnetism A Computational Chemistry Approach – p.51/52
7.6 van der Waals Coefficients - 9
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
An upper bound can be obtained either from the [n, n]α Padé approximant
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
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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′
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′
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
Φ(0)
nv ({~
~ K }) = Ψ(0)
ri }, {R n ({~
~ K }) Θ(0)
ri }; {R ~
v ({RK })
v
Question: What about ξαβ ?
(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
(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.
Molecular Electromagnetism A Computational Chemistry Approach – p.8/24
8.1 Sum-over-states treatment - 5
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′
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
~ 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
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
i
(0) (0)
+hΘ(0)
v Ψ0 | µ̂α + Ω̂Eα | Ψ0 Θ(1) ~
v (E)i
= 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
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 .
(0)
~ K }) is also expanded in a Taylor
in which the nuclear potential energy E0 ({R
series in the normal coordinates {Qa }
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
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
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
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.23/24
8.3 Vibrational and Thermal Averaging - 8
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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
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 Electromagnetism A Computational Chemistry Approach – p.3/46
9. Short Review of Electronic Structure Methods - 3
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.4/46
9. Short Review of Electronic Structure Methods - 4
Non-variational methods: MP or CC
Energy as an transition or asymmetric expectation value
∂E0SCF ({cµp })
=0
∂cµp Molecular Electromagnetism A Computational Chemistry Approach – p.5/46
9.1 Hartree-Fock Theory - 1
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
hψp | ψq i = δpq
Molecular Electromagnetism A Computational Chemistry Approach – p.6/46
9.1 Hartree-Fock Theory - 2
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
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 )
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
∂E0SCF ({cµp })
= 0
∂cµp
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
..
.
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
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
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
|ΦMCSCF
0 ({κpq }, {Sn0 })i = e−κ̂ e−Ŝ |ΦMCSCF
0 i
†
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
†
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.19/46
9.4 Configuration Interaction - 1
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
ĤC = E CI C
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
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 )
hΦ | Ĥ (0) | Φrs···
pq··· i = 0
Molecular Electromagnetism A Computational Chemistry Approach – p.22/46
9.4 Configuration Interaction - 4
hΦSCF
0 | Ĥ (0) | Φai i = hψi | fˆ| ψa i = 0
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 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
Molecular Electromagnetism A Computational Chemistry Approach – p.25/46
9.5 Møller-Plesset Perturbation Theory - 3
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
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
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
Ĥ (0) |ΦCC CC CC
0 i = E0 |Φ0 i
E0CC = hΦSCF
0 | Ĥ (0) | ΦCC
0 i
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.
hΦab
ij | e
−T̂
Ĥ (0) | ΦCC
0 i = 0 ... hΦSCF | d ĥNµ e−T̂ Ĥ (0) | ΦCC
0 i = 0
The coupled cluster amplitude equations are often collectively called the
coupled cluster vector function ei with elements,
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
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.
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
Molecular Electromagnetism A Computational Chemistry Approach – p.34/46
9.7 The Hellmann-Feynman Theorem for a Φ0 - 2
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
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.
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 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
The Lagrangian multipliers in the coupled cluster Lagrangian are just the λiµ
amplitudes of the dual bra or “Λ” state hΦΛ
0 |.
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
9.7 The Hellmann-Feynman Theorem for a Φ0 - 7
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.
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.
For a variational energy the last term vanishes again and the second derivative
of the energy for variational methods is given as
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
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]
(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
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
Sum-over-states expressions
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
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
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
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
M ij = ~ωS ij − E ij
(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
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
)
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
= −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
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
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.
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.
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
hΦMP
0
†
| [qai , [F̂ + V̂ , qbj ]] | ΦMP
0 i
(0)
hΦMP
0 | ΦMP
0 i
(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
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.
√ 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
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
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
10.3.2 Second Order PPA - 10
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.27/35
10.3.3 Higher-order and mixed methods - 2
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.29/35
10.3.4 Iterative and Non-iterative Doubles Correction - 2
Σ(0,2) 0 eX
n
= ~ωn
0 −Σ(0,2) dX
n
(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
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
Molecular Electromagnetism A Computational Chemistry Approach – p.32/35
10.3.4 Iterative and Non-iterative Doubles Correction - 5
The eigenvalues and thus excitation energies in the RPA(D) method are
defined as the sum of the zeroth- and second-order contributions
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
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
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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
~ is given as
An element of the perturbed Fock matrix Fpq (F)
~
Fpq (F) = hψp | fˆ(F)
~ | ψq i
(1) (2)
X X
ĥ (i) + ĥ (i) = Fα··· ôF
i,α··· + Fα··· ôFF
i,αβ··· Fβ···
α··· α,β,···
~ = 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
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
(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
= −hψa | ôF
α··· | ψi i
The two expressions in the parentheses are the RPA A(0,1) and B (1) matrices.
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
~ = 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
|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
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
(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
11.2 Multiconfigurational Linear Response Functions - 4
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.22/38
11.3 SOPPA - 1
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··· |
|ΦMP
0 (t)i = e
ıκ̂(t) ıŜ(t)
e |ΦMP
0 i
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.
hΦΛ CC
0 (t) | P̂ | Φ0 (t)i
|ΦCC
0 (t)i = e
T̂ (t)
|ΦSCF
0 i
h i
hΦ0 (t)| = hΦ0 | 1 + Λ̂(t) e−T̂ (t)
Λ SCF
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
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) + · · ·
Molecular Electromagnetism A Computational Chemistry Approach – p.28/38
(1)
11.4 Coupled Cluster Linear Response Functions - 4
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
∂ ∂
Aiµ jν = e iµ = hΦSCF
0 | d ĥiµ e−T̂ Ĥ (0) | ΦCC
0 i
∂tjν ∂tjν
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)
Molecular Electromagnetism A Computational Chemistry Approach – p.30/38
11.4 Coupled Cluster Linear Response Functions - 6
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
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
where the elements of the “property gradient" vectors T Λ (Ô) and T CC (Ô) for
an operator Ô are defined as
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
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.38/38
Molecular Electromagnetism
A Computational Chemistry Approach
Chapter 12
Stephan P. A. Sauer
sauer@kiku.dk
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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
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
Molecular Electromagnetism A Computational Chemistry Approach – p.7/22
12.2.1 First Derivatives - 3
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.8/22
12.2.1 First Derivatives - 4
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 )
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
i.e. one set of coupled Hartree-Fock equations with this second type of
Lagrangian, LMP2ai as right hand side
Molecular Electromagnetism A Computational Chemistry Approach – p.10/22
12.2.1 First Derivatives - 6
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
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
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···
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.
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
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.
∂Φ(t)
hΦ(t) | i=0
∂Fα··· (0)
|Φ(t)i = |ΦSCF
0 (t)i + |ΦMP1 (t)i
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).
Molecular Electromagnetism A Computational Chemistry Approach – p.20/22
12.3 Time Dependent Analytical Derivatives - 7
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.
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.
Department of Chemistry
University of Copenhagen
c Stephan P. A. Sauer 2011
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.
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.4/33
13.1 Basis Sets - 4
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.6/33
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.
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
Molecular Electromagnetism A Computational Chemistry Approach – p.9/33
13.1.1 Electric Properties - 3
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.10/33
13.1.1 Electric Properties - 4
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
~ ~r, B)
χζ,k,m,n (R, ~ = e 2~ { ~ GO −R)×
ıe (R ~ B~ }·~
r ~ ~r)
χζ,k,m,n (R,
~ ~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.
Molecular Electromagnetism A Computational Chemistry Approach – p.14/33
13.1.2 Magnetic Properties - 4
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.
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.
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
)
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
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 %
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.
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.
Molecular Electromagnetism A Computational Chemistry Approach – p.25/33
13.3.4 Indirect Nuclear Spin-Spin Coupling Constants - 1
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.