You are on page 1of 21

Version 0.

1 Eduardo Martin-Martinez - Amath 473

Block 4

Quantum Dynamics and Time-dependent


perturbation theory
4.1 Unitarity of quantum evolution

We are going to analyze time evolution in quantum mechanics. Let us begin with a time dependent Schrödinger
equation. This is a different approach than the local in time approach to the derivation of the Schrödinger
equation that we saw in detail in block 2 . To begin with, we will not assume that time evolution is characterized
by unitary operations. This will instead follow from our derivation. We start from

d −i
|ψ(t)i = Ĥ(t)|ψ(t)i (4.1.1)
dt }

Using the initial condition for |ψ(t)i at t = t0 , we can express the solution as some linear transformation
of the initial condition
|ψ(t)i = Û (t, t0 )|ψ(t0 )i (4.1.2)

Substituting that into Schrödinger equation we get that Û (t, t0 ) satisfies the same equation as |ψ(t)i

d −i
Û (t, t0 ) = Ĥ(t)Û (t, t0 ) (4.1.3)
dt }

with the initial condition Û (t0 , t0 ) = 11. Û (t, t0 ) is called the time evolution operator. If Ĥ were time
independent the solution to that equation would simply be

Û (t, t0 ) = e−i(t−t0 )Ĥ/} .

Which is a unitary operator. However, if Ĥ is time dependent, there is no simple closed form for Û (t, t0 ). Let
us see what we can say about Û analyzing how the value of U U † changes with time:

∂  †  ∂ ∂ i † † 
U U = (U † )U + U † (U ) = U H U − U † HU
∂t ∂t ∂t }

Since Ĥ is self-adjoint this yields


∂  † 
U U =0
∂t
which together with the initial condition means that U U † = 11 for all times. Hence, U (t, t0 ) is unitary
(Û † = Û −1 ) if Ĥ is Hermitian.

Time dependent perturbation theory 1 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

4.2 Schrödinger and Heisenberg pictures

In quantum mechanics, the quantities that can be ascribed physical meaning are probability distributions of
observables, not operators or vectors per se. Let us see how the expectation value of an observable evolves
with time. Consider the operator Ô representing the physical observable O. Its expectation value as a function
of time is D E
Ô = hψ(t)|Ô|ψ(t)i = hψ(t0 )|Û † (t, t0 ) Ô Û (t, t0 )|ψ(t0 )i (4.2.1)
t

From this expression there are two possible ways to understand the equation, we can either ascribe time
evolution to the vectors or to the operators:
We define the Schrödinger picture: this is the formalism we have been consistently applying so far
where the vectors (states) evolve with time while the operators are time independent:

|ψ(t)i = U (t, t0 )|ψ(t0 )i

The equation of motion in this case is (4.1.1) for the state vector |ψ(t)i.
There is, however, a different way of thinking about what evolves in time in equation (4.2.1). We can
ascribe time evolution to the operators instead of the vectors. We could equally define a time dependent
operator ÔH (t) as follows
ÔH (t) = U † (t, t0 ) Ô Û (t, t0 ) (4.2.2)
and then equation (4.2.1) can be written as
D E
Ô = hψ(t0 )|ÔH (t)|ψ(t0 )i. (4.2.3)
t

We call this the Heisenberg picture: In this picture the states are always time independent (do not evolve
in time) and the time evolution is inside the operator that represents an observable.
ÔH (t) is the operator in the Heisenberg picture corresponding to observable O. This in turn corresponds
to the operator Ô in the Schrödinger picture. Let us obtain the equation of motion for this operator: Differ-
entiating (4.2.2) and using (4.1.3) we obtain

dÔH i ∂ Ô
= Û † Ĥ ÔÛ − Û † ÔĤ Û + Û †


dt } ∂t
In the expression above, we are considering the possibility that Ô may be explicitly time-dependent. Now we
can rewrite this equation writing the operators Ĥ and Ô in the Heisenberg picture, obtaining
!
dÔH (t) i  ∂ Ô
= HH (t), ÔH (t) + (4.2.4)
dt } ∂t
H

where we have introduced HH (t) = U † HU following the definition (4.2.2). Notice that the last term
!
∂ Ô ∂ Ô
= Û † (t, t0 ) Û (t, t0 )
∂t ∂t
H

will only be non-zero if the operator Ô (in the Schrödinger picture) explicitly depends on time (i.e. other
than the time dependence given by dynamic evolution). An example of this is if it is coming from an external

Time dependent perturbation theory 2 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

potential that depends on time or the component of a tensor defined with respect to a mobile coordinate
system, but this does not include the time dependence generated via the system’s Hamiltonian.
The time derivative of the expectation value of an observable has a similar form in the two pictures. Let
us first compute it in the Schrödinger picture:
   
d ∂ ∂ ∂ Ô
hOit = hψ(t)| Ô|ψ(t)i + hψ(t)|Ô |ψ(t)i + hψ(t)| |ψ(t)i
dt ∂t ∂t ∂t

now using (4.1.1) we get

d i ∂ Ô
hOit = hψ(t)|(Ĥ Ô − ÔĤ)|ψ(t)i + hψ(t)| |ψ(t)i
dt } ∂t
Therefore
d i D E D ∂ Ô E
hOit = Ĥ, Ô +
dt } t ∂t t
where the operators are in the Schrödinger picture and the expectation values are taken with respect to the
time dependent state |ψ(t)i.
Now, in the Heisenberg picture, the states are time independent, and from (4.2.4), we can easily compute
the variation of the expectation value of O:
!
d dÔH i  ∂ Ô
hOit = hψ(t0 )| |ψ(t0 )i = hψ(t0 )| H, ÔH (t) |ψ(t0 )i + hψ(t0 )| |ψ(t0 )i
dt dt } ∂t
H

Therefore !
d i D E D ∂ Ô E
hOit = Ĥ, ÔH +
dt } t0 ∂t t0
H
where the operators are in the Heisenberg picture and the expectation values are taken with respect to the
time independent state |ψ(t0 )i.
These equations for the expectation of observables are the classical analogue to the equations of motion.
As one can see a naive comparison between the classical equations of motion and the quantum analogue sug-
gests that going from classical to quantum involves replacing variables representing observables by Hermitian
operators and Poisson brackets with commutators. This is, however, a deeper issue than it first appears due
to commutation ambiguities when guessing quantum Hamiltonians from the classical analogues.

4.3 The interaction (Dirac) picture

The Heisenberg and Schrödinger pictures are not the two only possible choices. We are going to see an evolution
picture where both states and observables will be time dependent: The interaction picture. The whole idea
behind the Interaction picture is too simplify the treatment of time dependent problems. Considering the full
Hamiltonian as
Ĥ = Ĥ0 + λV̂ (t)
where Ĥ0 is not time dependent. Let us denote as Û0 ≡ ei(t−tD )Ĥ0 , a unitary transformation such that
operators and states transform as

ÔD (t) = Û0 Ôs (t)Û0† , |ψ(t)iD = Û0 |ψs (t)i (4.3.1)

Time dependent perturbation theory 3 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

the parameter tD has no physical meaning and should not be confused with an initial time t0 . This parameter
reminds us of the global phase invariance of physical states. For simplicity, it is usually made equal to zero
and we are going to follow that usual convention, therefore we will assume from now on that

Û0 ≡ eiĤ0 t (4.3.2)

Let us obtain the form of Schrödinger equation in that picture

d
|ψs (t)i = −iH(t)|ψs (t)i
dt

let us introduce the unity Û0 Û0† = 11 in the equation

d  † 
i Û0 Û0 |ψ(t)i = Û0† Û0 H(t)Û0† Û0 |ψ(t)i
dt

Since ÔD (t) = Û0 Ôs (t)Û0† , and |ψ(t)iD = Û0 |Ψs (t)i we get

d  † 
i Û0 |ψ(t)iD = Û0† ĤD (t)|ψ(t)iD
dt

The time derivative of Û0† is easily computable:

dÛ0†
Û0† = e−iH0 t ⇒ = −iH0 e−iH0 t = −iH0 Û0†
dt

substituting this in Schrödinger equation and explicitly writing ĤD = Ĥ0D + V̂ we get
 
d

iÛ0 −iH0 |ψ(t)iD + |ψ(t)iD = Û0† (Ĥ0D + λV̂D )|ψ(t)iD (4.3.3)
dt

where we have not written the time dependence in V̂ (t) to lighten the notation.
 
Now since H0 , Û0 = 0 as Û0 is a function of Ĥ0 then Ĥ0 = Ĥ0D , in other words

Ĥ0D = Û0 Ĥ0 Û0† = Ĥ0 (4.3.4)

so the terms proportional to Û0† Ĥ0 |ψ(t)iD on both sides of equation (4.3.3) cancel out, yielding
 
d
iÛ0† |ψ(t)iD = Û0† λV̂D |ψ(t)iD
dt

Acting from the left with Û0 we finally get

d
i |ψ(t)iD = λV̂D |ψ(t)iD (4.3.5)
dt

which is the Schrödinger equation in the interaction picture. Notice that considering the kets in the inter-
action picture, their time evolution in this picture is generated only by the interaction part of the Hamiltonian.

Time dependent perturbation theory 4 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

How about the equation of motion for the expectation of an observable Ô. Let us begin with the equation
of motion in the Schrödinger picture:
!
d D E   ∂ Ô
Ô = hψ(t)| i H, Ô + |ψ(t)i
dt t ∂t

we introduce the identity in a similar fashion as before


 
d D E ∂  †
Ô = hψ(t)|Û0† Û0 i Û0† Û0 H Û0† Û0 , Û0† Û0 ÔÛ0† Û0 + Û0 Û0 ÔÛ0† Û0 Û0† Û0 |ψ(t)i
 
dt t ∂t
 
d D E  † † ∂  †
Û0 ÔD Û0 Û0† |ψ(t)iD

Ô = D hψ(t)|Û0 i Û0 HD Û0 , Û0 ÔD Û0 +
dt t ∂t
   
d D E  ∂  †
Ô = D hψ(t)| i HD ÔD − ÔD HD + Û0 Û0 ÔD Û0 Û0† |ψ(t)iD (4.3.6)
dt t ∂t
Using that
dÛ0†
Û0† = e−iH0 t ⇒ = −iH0 e−iH0 t = −iH0 Û0†
dt
dÛ0
Û0 = eiH0 t ⇒ = iH0 eiH0 t = iH0 Û0
dt
we know that  
∂  † 
† † ∂
Û0 ÔD Û0 = −iH0 Û0 ÔD (t)Û0 + Û0 ÔD (t) Û0 + iÛ0† ÔD (t)H0 Û0 (4.3.7)
∂t ∂t
substituting in (4.3.6)
    
d D E  ∂
Ô = D hψ(t)| i HD ÔD − ÔD HD − iH0 ÔD (t) + ÔD (t) + iÔD (t)H0 |ψ(t)iD (4.3.8)
dt t ∂t

where we have used that Ĥ0 commutes with any function of only Ĥ0 and in particular Û0 . Since we know that
ĤD = Ĥ0D + V̂D = Û0 Ĥ0 Û0† + VD = Ĥ0 + VD (4.3.9)
In other words Ĥ0 = Ĥ0D since Ĥ0 commutes with any function of only Ĥ0 . Substituting (4.3.9) into (4.3.8),
al the terms containing Ĥ0 cancel, yielding
   ∂ 
d D E
Ô = D hψ(t)| iλ VD ÔD − ÔD VD + ÔD (t) |ψ(t)iD (4.3.10)
dt t ∂t

or in other words, the equation of motion for the expectation of the observable Ô is identical to the Heisenberg
equation of motion only that the evolution is only generated by the interaction part:
  
d D E   ∂
Ô = D hψ(t)| iλ VD , ÔD + ÔD (t) |ψ(t)iD
dt t ∂t

Notice that the separation


 between
 the free evolution and the interaction part of the evolution cannot be
cleanly done: in general Ĥ0 , V̂ 6= 0, So even in the simplest case where the Hamiltonian is time independent
and therefore
   
Û |ψi = e−i(Ĥ0 +V̂ )t |ψi =
6 e−iĤ0 t e−iλV̂ t |ψi =
6 e−iλV̂ t e−iĤ0 t |ψi =
6 e−i(Ĥ0 +λV̂ )t |ψi

So one has to be careful when giving handwaving explanations as to what is the interaction picture in terms
of separating free evolution from the interaction part of the evolution.

Time dependent perturbation theory 5 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

4.4 Time-dependent perturbation theory: The Dyson series

As opposed to time independent perturbation theory, which dealt with the computation of eigenvalues and
eigenvectors of Hamiltonians, we are interested now in solving the fully time dependent Schrödinger equation
in order to obtain the time evolution of quantum states under a given Hamiltonian. In contrast to the block on
time-independent perturbation theory, to reduce the notational weigh, we are going to notate the eigenstates
and eigenvalues of H0 as
Ĥ0 |ni = En |ni.

In general, if the Hamiltonian involves complicated interaction terms, the problem of finding the state of a
quantum system at a given time t provided the state of the system at a time t0 is very complex. Much more
if we are dealing with a time dependent Hamiltonian. Let us assume we can separate the Hamiltonian into a
time independent and a time dependent one as follows:

Ĥ = Ĥ0 + λV̂ (t)

where we have made explicit that the second summand is multiplied by a small real constant λ. Similarly to
what we did in the Schrödinger picture, we can write the state in the interaction picture |ψ(t)iD as

|ψ(t)iD = ÛD (t, t0 )|ψ(t0 )iD

substituting this in equation (4.3.5), we end up with an equation for the time evolution operator in the
interaction picture
d
ÛD (t, t0 ) = −iλV̂D (t)ÛD (t, t0 ) (4.4.1)
dt
with the obvious initial condition
UD (t0 , t0 ) = 11 (4.4.2)

If the Hamiltonian commuted with itself at all times, this is, if V̂D (t), V̂D (t0 ) = 0, ∀t, t0 then the solution
 

to this equation can be simply written as


 Z t 
UD (t, t0 ) = exp −iλ dt V̂D (t)
t0

However, this will very rarely be the case. In general, if the Hamiltonian does not commute with itself at
different times this will not be correct.
In that case we lack a closed form for the time evolution operator. We can, however, write equation (4.4.1),
together with the initial condition (4.4.2), as an integral equation:
Z t
ÛD (t, t0 ) = 11 − iλ dt1 V̂D (t1 )ÛD (t1 , t0 )
t0

and we can recursively substitute the expression of Û (t, t0 ) into the right-hand side of the integral equation,
yielding
Z t  Z t1  Z t2 
ÛD (t, t0 ) = 11 − iλ dt1 V̂D (t1 ) 11 − iλ dt2 V̂D (t2 ) 11 − iλ dt3 V̂D (t3 ) (11 + . . . )
t0 t0 t0

Therefore, provided that the parameter λ  1, we can write the evolution operator as the following series
expansion:
(1) (2) (3)
ÛD (t, t0 ) = 11 + ÛD (t, t0 ) + ÛD (t, t0 ) + ÛD (t, t0 ) + . . . (4.4.3)

Time dependent perturbation theory 6 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

where
Z t
(1)
ÛD (t, t0 ) = −iλ dt1 V̂D (t1 ) (4.4.4)
t0
Z t Z t1
(2) 2
ÛD (t, t0 ) = −λ dt1 dt2 V̂D (t1 )V̂D (t2 ) (4.4.5)
t0 t0
Z t Z t1 Z t2
(3) 3
ÛD (t, t0 ) = iλ dt1 dt2 dt3 V̂D (t1 )V̂D (t2 )V̂D (t3 ) (4.4.6)
t0 t0 t0

We call the expansion (4.4.3) the Dyson series for the time evolution operator. This expression is often
shortened as follows:  Z t 
UD (t, t0 ) = T exp −iλ dt V̂D (t)
t0
where T notates the time ordering operator. This is a functional operator that acts on the following form:

fˆ(t)ĝ(s) t ≥ s

T[fˆ(t)ĝ(s)] =
ĝ(s)fˆ(t) s > t
Note that a version of the time ordering operator with an extra minus sign in the second region of the domain
is commonly employed when dealing with fermionic operators, but we will restrict to the usual definition for
all our purposes.

Example: A Harmonic oscillator coupled to N harmonic oscillators

Let us consider as a first example of application the case of a set of N + 1 harmonic oscillators coupled in
a way in which one of them, that we single out with the label subzero, is coupled to all the others in the
following way:
N N
!
P̂02 1 2 2
X P̂ 2
n 1 2 2
X
Ĥ = + m0 ω0 X̂0 + + mn ωn X̂n + κ X̂0 X̂n
2m0 2 2mn 2
n=1 n=1

where κ = λ 2m0 ω0 . Writing this Hamiltonian in terms of ladder operators
r r
mn ωn † 1
ân + â†n
 
P̂n = i ân − ân , X̂n =
2 2mn ωn

where â and ↠satisfy [ân , â†m ] = δnm , we get


1
Ĥ = ω0 â†0 â0 + ωn â†n ân + λ(â0 + â†0 )
X X
√ (ân + â†n ) + EZ 11 (4.4.7)
n n
2ωn mn

where EZ represents a constant term (zero point energy).


Although obviously not properly Lorentz covariant, this very simple model capture many features from
the fully quantum light-matter interaction in optical cavities. To extend the quantum optical analogy, let us
call the zeroth oscillator ‘detector’ and all the others ‘field modes’. One can think that the zeroth oscillator
as representing, for instance, the internal energy levels of an atom and the N other oscillators as the different
frequency modes of a quantum field living in an optical cavity.
Now, we can split the Hamiltonian into a free and an interaction part:

Ĥ = Ĥ0 + λV̂

Time dependent perturbation theory 7 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

where
1
Ĥ0 = ω0 â†0 â0 + V̂ = (â0 + â†0 )
X X
ωn â†n ân , √ (ân + â†n )
n n
2ωn mn
where we have conveniently ignored the zero point energy term into a renormalization of the origin of ener-
gies. Something that can be cleanly done if that energy is finite (N < ∞). Let us consider the interaction
Hamiltoinan V̂ in the interaction picture,
N N
! !
† iωn t â†n ân −iωn t â†n ân
Y Y
V̂D = Û0 V̂ Û0 = e V̂ e
n=0 n=0

Now, we can easily compute what would be the form of this operator using the Hadamard Lemma:

Â, Â, B̂ x2 Â, Â, Â, B̂ x3


      
x −xÂ
 
e Be = B̂ + Â, B̂ x + + + ...
2 6
† † â † † â
Let us compute its action on the creation and annihilation operators eiωn t â â â e−iωn t â and eiωn t â â a† e−iωn t â
using Hadamard’s Lemma

† †a (iωn t)2 † (iωn t)3 †


eiωn t a a a e−iωn t a = a + iωn t[a† a, a] + [a a, [a† a, a]] + [a a, [a† a, [a† a, a]]] + . . .
2! 3!
since the commutator [a, a† a] is
[a† a, a] = a† [a, a] + [a† , a]a = −a
We find
† †a (iωn t)2 † (iωn t)3 †
eiωn t a a a e−iωn t a = a − iωn a − [a a, a] − [a a, [a† a, a]] + . . .
2! 3!
(iωn )2 (iωn t)3 † (−iωn t)2 (−iωn t)3
= a − iωn t a + a+ [a a, a] + · · · = a − iωn t a + a+ a + ...
2! 3! 2! 3!
= e−iωn t a (4.4.8)

Therefore
1
V̂D = (â0 e−iω0 t + â†0 eiω0 t )
X
√ (ân e−iωn t + â†n eiωn t )
n
2ωn mn
Let us compute what would be the time evolution operator up to first order in λ.
Z t 
1 
dt1 â0 â†n e−i(ω0 −ωn )t1 + â†0 ân ei(ω0 −ωn )t1 + â0 ân e−i(ω0 +ωn )t1 + â†0 â†n ei(ω0 +ωn )t1
X
ÛD (t, t0 ) = 11 − iλ √
n
2ωn mn t0
(4.4.9)

plus corrections of order O(λ2 ). One first thing to notice is that there are, mainly, two kind of terms:
N N
O a† aj O
|00 i|1j i |0in −−0−→ |10 i |0in (4.4.10)
n6=j n
N N
O a†0 a†j O
|00 i |0in −−−→ |10 i|1j i |0in (4.4.11)
n n6=j

and their Hermitian conjugate analogues. We call the first kind of transitions Rotating-wave transitions, and
the second kind counter-rotating wave transitions.

Time dependent perturbation theory 8 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

The second kind has no classical analogue, it corresponds to quantum fluctuations and would be present
even if the initial state is the vacuum (least energy state of the free system). No phonons whatsoever (none of
the individual oscillators are excited) and still there is a chance that ‘detector’ gets excited out of the vacuum.
However, in most of the quantum optics experiments you have probably heard about, these kind of processes
(getting an atom excited by emitting a photon rather than capturing it) are not observed. An argument very
often used to justify this is that since those transitions ‘violate energy conservation’, they can only exist
for short times according to the uncertainty relationship between time and energy ∆E∆t ≥ 1/2. Although
it is true that those transtisions would only be observed for very short times, that naive explanation is far
from satisfying. To begin with, it is not true that energy is not conserved: Since the Schrödinger picture
Hamiltninan (4.4.7) is not time dependent, we know that energy is a conserved quantity.
Notice that the vacuum or the locally excited states portrayed in (4.4.10) and (4.4.11) are eigenstates of
the free Hamiltonian Ĥ0 but not eigenstates of the full Hamiltonian Ĥ. Therefore, when the interaction is
turned on, these states do not have a definite value of energy, and instead will have some expectation and
uncertainty in their energies.
Now, we can rigorously see why these kind of transitions only survive for short times as compared to the
inverse of the energy gap of the detector ω0 . The rotating-wave terms appear in the time evolution operators
weighed by the following integral over time
Z t
dt1 ei(ω0 −ωn )t1
t0

there are N terms like this, let us first consider the case of the non-resonant modes, if ωn 6= ω0 , let us see
what would be the behaviour for very long time evolution intervals. Defining ∆T = t − t0 and taking the limit
∆T → ∞, we get Z t D E 1
lim dt1 ei(ω0 −ωn )t1 ∼ ∆T e−i(ω0 −ωn )t1 ∼
t→∞ t0 (ω0 − ωn )
t0 →−∞
this is to say, the result is the time interval multiplied by the average of a complex exponential. This average
is obviously zero and, as the time interval grows this produces a bounded term. In contrast, for the resonant
mode, i.e. when ωj = ω0 we get
Z t
lim dt1 ei(ω0 −ωj )t1 ∼ ∆T
t→∞
t0 →−∞ t0

which is much bigger than the non-resoant contributions (that are bounded). This gives raise toa very common
approximation in quantum optics called the Single-mode approximation and which consists of neglecting the
contribution of the non-resonant modes, if the interaction lats long enough ∆T  ω0−1 .
Now, the counter-rotating wave transitions are weighed by the following integral over time
Z t
dt1 ei(ω0 +ωn )t1
t0

Notice that in this case, the phase will never be stationary for any values of ωn . In this case if we call
∆T = t − t0 , when ∆T → ∞
Z t D E 1
lim dt1 ei(ω0 +ωn )t1 ∼ ∆T ei(ω0 +ωn )t1 ∼
t→∞ t0 (ω0 + ωn )
t0 →−∞

So by the same token we could neglect ∆T  (ω0 + ωn )−1 , this is called the rotating-wave approximation. So
if we carry out the single mode approximation,

Time dependent perturbation theory 9 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

However, for times ∆T . ω0−1 none of the two approximations can be carried out and, indeed, vacuum
fluctuations would be observed. So if you look at the quantum vacuum for very short times, or with a very
powerful magnifying glass (small ω0 ) we will see particles.

4.5 Transition probability and transition rate

Considering again a time dependent Hamiltonian of the form

Ĥ = Ĥ0 + λV̂ (t) (4.5.1)

We want to known the probability of transition from an initial state |ψ(t0 )i to a final state |φi at a time t.
Born’s rule tells us that this is simply the modulus square of the transition amplitude
2
P|ψi→|φi = |hφ|ψ(t)i|2 = hφ|Û (t, t0 )|ψ(t0 )i

with states and operators in the Schrödinger picture.


Let us notate, as usual, |ni the eigenstate of Ĥ0 with energy En , such that

Ĥ0 |ni = En |ni,

a very relevant particular case when to consider when (4.5.1) represents a perturbation on a free Hamiltonian,
is the probability of transition between two different eigenstates of the free Hamiltonian Ĥ0 . In other words:
|ψ(t0 )i = |ni, |φi = |mi. In that case the calculation gets simplified by utilizing the interaction picture.
Let us first obtain the relationship between the Schrödinger and the interaction picture time evolution
operators for this we start from the interaction picture state |ψ(t)iD , recalling (4.3.2) and (4.3.1),

|ψ(t)iD = eiĤ0 t |ψ(t)i = eiĤ0 t Û (t, t0 )|ψ(t0 )i = eiĤ0 t Û (t, t0 )e−iĤ0 t0 |ψ(t0 )iD

and we know that on the other hand |ψ(t)iD = ÛD (t, t0 )|ψ(t0 )iD , therefore we get

ÛD (t, t0 ) = eiĤ0 t Û (t, t0 )e−iĤ0 t0 (4.5.2)

Although, obviously, the amplitude of probability computed in the two pictures would be the same, this is

hφ|Û (t, t0 )|ψ(t0 )i = D hφ|ÛD (t, t0 )|ψ(t0 )iD

we can maybe use a little trick to compute transition probabilities using the interaction picture time evolution
operator and the Scrödinger picture free eigenstates: it is trivial to see from (4.5.2) that the amplitude of
transition between eigenstates of Ĥ0 can be written as

hm|Û (t, t0 )|ni = e−i(Em t−En t0 ) hm|ÛD (t, t0 )|ni

therefore the probability of transition will be the same if we compute it using the trick of replacing the time
evolution operator by the interaction picture one:
2 2
Pn→m = hm|Û (t, t0 )|ni = hm|ÛD (t, t0 )|ni

Time dependent perturbation theory 10 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

Notice that this only works with eigenstates of the free Hamiltonian. If either |φi or |ψ(t0 )i are not eigenstates
of Ĥ0 the probability of transition can be still computed using the interaction time evolution operator, but
the result is not as simple. If we rewrite |φi and |ψ(t0 )i in the eigenbasis of Ĥ0 :
X X
|φi = Ck |ki, |ψ(t0 )i = Cj |ji
k j

we get that
2
X 2
hφ|Û (t, t0 )|ψ(t0 )i = e−i(Ek t−Ej t0 ) Ck∗ Cj hk|ÛD (t, t0 )|ji
j,k

Transition rate

For both practical and computational reasons, it is very often more interesting to calculate the transition rate
instead of the transition probability. This is, the transition probability per unit time:
d 2
Ṗ|ψi→|φi = hφ|Û (t, t0 )|ψi
dt
In the particular case where we compute the transition rate between two eigenstates |ni → |mi of Ĥ0 , we can
simply write
d 2 d  
Ṗn→m = hm|ÛD (t, t0 )|ni = hm|ÛD (t, t0 )|nihn|ÛD† (t, t0 )|mi
dt dt
Let us compute this at first order in perturbation theory: we know that
Z t
ÛD = 11 − iλ dt1 V̂D (t1 ) + O(λ2 )
t0

therefore, since we assume m 6= n,


Z t Z t 
2 d 0 † 0
Ṗn→m = λ dt1 dt1 hm|V̂D (t1 )|nihn|V̂D (t1 )|mi
dt t0 t0
 Z t Z t 
2 † †
= λ hm|V̂D (t)|ni dt1 hn|V̂D (t1 )|mi + hn|V̂D (t)|mi dt1 hm|V̂D (t1 )|ni + O(λ3 ) (4.5.3)
t0 t0

Example 1: Transition probabilities for a Constant perturbation

In the case where we switch on an constant interaction at time t0 = 0 and we keep it constant in time until
time t, this is 
0 t<0
V̂ (t) =
V̂ t ≥ 0
For the time interval when V̂ (t) is not zero, in the interaction picture

V̂D = eiĤ0 t V̂ e−iĤ0 t

The probability of transition between |ni and |mi at the first order in the perturbative parameter will be given
by Z t Z
2 t 2
2
Pn→m = λ2 dt1 hm|eiĤ0 t1 V̂ e−iĤ0 t1 |ni = λ2 Vmn dt1 ei(Em −En )t1 + O(λ3 )
0 0

Time dependent perturbation theory 11 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

where we notate the relevant V̂ matrix element as Vmn ≡ hm|V̂ |ni. The integral is trivially solvable
2 2 2
λ2 Vmn
 
2 Vmn  2 Vmn ωmn t
ei(Em −En )t −1 2
sin2 +O(λ3 ) (4.5.4)

Pn→m = = 2λ 1−cos(ωmn t) = 4λ
(Em − En )2 2
ωmn 2
ωmn 2

where ωmn = Em − En .

Example 2: Transition probabilities for a Harmonic perturbation

Another interesting case to consider is a perturbation that depends on time in a harmonic way:

V̂ (t) = V̂eiΩt + Vˆ† e−iΩt

In this case
hm|V̂D |ni = Vmn ei(ωmn +Ω)t + (V † )mn ei(ωmn −Ω)t

where we have followed the same notation as in the previous example (e.g., (V † )mn ≡ hm|Vˆ† |ni )
The transition probability will be then given by
Z t   2
Pn→m = λ 2
dt1 Vmn ei(ωni +Ω)t1 + (V † )mn ei(ωni −Ω)t1 (4.5.5)
0

which is very similar to the previous case. Solving the integrals


2
1 − ei(ωmn +Ω)t 1 − ei(ωmn −Ω)t †
Pn→m = λ2 Vmn + (V )mn (4.5.6)
ωmn + Ω ωmn − Ω

Transition probability to a subset of final states

If we want to compute the probability of transition to a number of different states, we need to add up the
probability, let us consider the set of final states F = {|φi i}i
2
X X
P|ψi→F = P|ψi→|φi i = hφi |Û (t, t0 )|ψ(t0 )i (4.5.7)
i i

This is especially useful to compute the transition probabilities. Let us go back to the first example of
a set of harmonic oscillators coupled according to the Hamiltonian (4.4.7). Let us assume that the initial
state is the ‘vacuum’ state, this is, the lowest energy state of the free Hamiltonian, which means that all the
oscillators are in the |0j i state.
We want to compute, up to first order in perturbation theory, the probability of the ‘detector’ (the zeroth
oscillator) clicking (transiting to the first excited state |00 i → |10 i) out of vacuum fluctuations. Since we do
not care about the the final state of all the other possible oscillators , we need to sum up over all the posible
final states for the N oscillators different from the zeroth one:.
2
X
P|00 i→|10 i = h10 , out|Û (t, t0 )|00 , 01 , . . . , 0N i
out

Time dependent perturbation theory 12 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

where the symbol |outi represents a generic final state for the set of N harmonic oscillators. Since the first
order contribution to Û (t, t0 ) given in (4.4.9) only contains elements of the kind a†0 an , a†0 a†n the only possible
final states will be first excitations of the different oscillators, therefore the expression above gets simplified to
N
2
X
P|00 i→|10 i = h10 , 1j , 0n6=j |Û (t, t0 )|00 , 01 , . . . , 0N i
j

where the notation O


|10 , 1j , 0n6=j i ≡ |10 i|1j i |0n i
n6=j

substituting (4.4.9) into the probability of transition we realize that neither the identity nor any of the terms
that do not contain two creation operators contribute to the probability of transition (all the other terms
annihilate the vacuum state on the right):

N X
N
! 2
Z t
1
h10 , 1j , 0n6=j |â†0 â†n |00 , 01 , . . . , 0N i p
X
P|00 i→|10 i = λ2 dt1 ei(ω0 +ωn )t1 + O(λ3 )
j=1 n=1
2ωj mj t0

Now obviously,

h10 , 1j , 0n6=j |â†0 â†n |00 , 01 , . . . , 0N i = h10 |â†0 |00 ih1j , 0n6=j |â†n |00 , 01 , . . . , 0N i = δjn

If, for simplicity and without loss of generality, we take t0 = 0, we readily obtain
N Z t 2 N
X 1 X 1 2
P|00 i→|10 i = λ2 dt1 ei(ω0 +ωj )t1 = λ2 ei(ω0 +ωj )t − 1
2ωj mj 0 2ωj mj (ω0 + ωj )2
j=1 j=1
N
X 1 − cos[(ω0 + ωj )t]
= λ2
ωj mj (ω0 + ωj )2
j=1

Therefore
N
X sin2 [ 1 (ω0 + ωj )t]
P|00 i→|10 i = 2λ2 2
+ O(λ3 )
ωj mj (ω0 + ωj )2
j=1

So we can see how the probability of seeing ‘detector clicks’ out of the vacuum oscillates in time very fast
(here you got the ‘soup’ of particles being created and destroyed that is often mentioned when talking about
the quantum vacuum).

4.6 Fermi’s Golden Rule

We have seen that if we are computing the transition probability into a set of final states we need to add up
the probability of transition to all the possible final states. Now consider that the transition is from an initial
state into a state part of a continuum spectrum of energies (or a set of states having a very similar energy).
Clearly our experimental resolution in energies will be finite and therefore the probability that we detect that
transition is going to be related with how good we are to distinguish between the possible different final states
that the system will have, i.e. of our energy resolution.

Time dependent perturbation theory 13 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

The number of states within an energy interval (E, E + dE) per energy unit is often called the density of
final states ρ(E). If the final states are part of a continuum of energies between E− and E+ , the sum (4.5.7)
gets transformed into an integral over the density of final states
Z E+
P|ψi→F = dE ρ(E)P|ψi→|Ei
E−

Fermi’s Golden Rule for constant perturbations

Let us now analyze the case of a constant perturbation. The probability of transition, at first order in λ,
from an eigenstate of Ĥ0 of energy En to an eigenstate of energy Em is given by the expression (4.5.4). Now
consider that we want to compute the probability of transition, under a constant perturbation, from a state of
energy En to a state into the set FEm of levels (maybe a continuum) of energies E− < Em < E+ distributed
according to some density of states ρ(E):

E+ E+ 2  
Vmn (Em − En )t
Z Z
2
Pn→FEm = dEm ρ(Em )Pn→m (Em ) = dEm ρ(Em )4λ sin2
E− E− (Em − En )2 2

It is especially interesting to consider the case when the energy of the initial state is very similar to the
energy of the set of final states. For example, one can imagine an
Let us make the following observation, let us consider the following observation: the weak limit of
a sinc2 (ax) when a → ∞ is proportional to the delta distribution:

1 sin2 (ax)
lim = δ(x)
a→∞ π ax2
with that in mind If we want to compute the probability of transition in the limit of very long times we obtain
Z E+
2 2
lim Pn→FEm ∼ 2πλ t dEm ρ(Em ) Vmn δ(Em − En )
t→∞ E−

This is telling us that when we take the limit of very long times, under constant perturbation, the transition
will only happen to states that have the same energy as the initial state. If we use the delta to integrate the
expression above we get
2
lim Pn→FEm 'En = 2πλ2 t Vmn ρ(Em )
t→∞

This is called Fermi’s Golden rule for constant perturbations. Here the limit t → ∞ has to be understood as
the limit for times much larger than the inverse of the energy scales involved in the problem.

Fermi’s Golden Rule for harmonic perturbations

Recalling the expression for the probability of transition from an eigenstate of Ĥ0 |ni to another eigenstate
|mi under a harmonic perturbation (4.5.6)
2
1 − ei(ωmn +Ω)t 1 − ei(ωmn −Ω)t †
Pn→m = λ2 Vmn + (V )mn
ωmn + Ω ωmn − Ω

Time dependent perturbation theory 14 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

remembering that ωmn = Em − En , it is easy to realize that the two summands inside the modulus squared
look very similar to the case of a constant perturbation, if we make the substitution

Em − En −→ Em − En + Ω, Em − En −→ Em − En − Ω (4.6.1)

for the first and second summands respectively.


Now let us think for a moment what happens in the limit of t → ∞. Very much along the lines of what
we saw for the example of a set of harmonic oscillators, the first summand will become an oscillatory function
bounded in modulus by (Em − En + Ω)−1 unless Em = En − Ω, in which case we see from (4.5.5) that the
first summand will be proportional to t . Conversely, the second summand will become an oscillatory function
bounded in modulus by (Em − En − Ω)−1 unless Em = En + Ω, in which case we see from (4.5.5) that the
second summand will be proportional to t.
These two conditions are, of course, mutually exclusive, so if we are looking at the limit of very large t,
only one of the summands will contribute to the probability of transition. The first summand corresponds to
the transition to a final state whose energy is the energy of the initial state minus the characteristic energy
of the perturbation (emission), and the second summand corresponds to the transition to a final state whose
energy is the energy of the initial state plus the characteristic energy of the perturbation (absorption).
In those regimes the expressions are identical to the case of the constant perturbation after the substitution
(4.6.1), therefore it is straightforward to write, in the limit of very large t, we can write the following result
Z E+
2
lim PEn →En −Ω ∼ 2πλ2 t dEm ρ(Em ) Vmn δ(Em − En + Ω) (4.6.2)
t→∞ E−
Z E+
2
lim PEn →En +Ω ∼ 2πλ2 t dEm ρ(Em ) (V † )mn δ(Em − En − Ω) (4.6.3)
t→∞ E−

So in this settings the harmonic perturbation can be seen as an inexhaustible source (or sink) of energy, and
where the ‘energy conservation’ deltas now account for the possible emission or absorption (respectively) of a
quantum of energy to the time dependent perturbation.
The integrals can be trivially carried out using the deltas to obtain
2
lim Pn→FEm =En −Ω = 2πλ2 t Vmn ρ(En − Ω) (4.6.4)
t→∞
2
lim Pn→FEm 'En +Ω = 2πλ2 t (V † )mn ρ(En + Ω) (4.6.5)
t→∞

This is called Fermi’s Golden rule for harmonic perturbations, and again, the limit t → ∞ has to be understood
as the limit for times much larger than the inverse of the energy scales involved in the problem.
It is convenient to define the differential transition rate wn→m as the transition rate per unit time, for every
individual possible final state. Integrating this over the number of final states,
R or in other words, over the
energies of the possible final states multiplied by the density of final states ( dEρ(E)) will give the transition
rate, this is
Z E+
Ṗn→FEm = dEm ρ(Em )wn→m
E−

From Fermi’s golden rule we get that the differential transition rate for very long times for a constant pertur-
bation is given by
2
wn→FEm = 2πλ2 Vmn δ(Em − En )

Time dependent perturbation theory 15 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

in the same fashion, for harmonic perturbations we obtain, in the limit of very long times, the following
differential rates for emission and absorption respectively
2
wEn →En −Ω ∼ 2πλ2 Vmn δ(Em − En + Ω) (4.6.6)
2 † 2
wEn →En +Ω ∼ 2πλ (V )mn δ(Em − En − Ω) (4.6.7)

which is the most common formulation of Fermi’s Golden Rule.

4.7 Stimulated emission and absorption of classical EM radiation by atoms

We are going to use what we learned to study the case of an atom (or more precisely, the electrons in the atoms)
with a classical electromagnetic radiation field. Assuming there is no stationary potentials, the interaction
Hamiltonian of a charge e of mass m interacting with an electromagnetic field is given by the minimal coupling
interaction
[p̂ − eA(x̂)]2 p̂2 e e2
Ĥ = + eφ(x̂) = − p̂·A(x̂) + [A(x̂)]2 + eφ(x̂) (4.7.1)
2m 2m m 2m
where we are using natural units ~ = c = 1. The last term in this equation correspond to a static electric
potential. In the case of an electron in an atom that would be the Coulomb potential. The term proportional
to A2 is very often neglected. There are, indeed, some times where this cannot be neglected (for example in
the case of the Casimir effect or for quantum phase transitions), but in general, if we are only interested in the
behaviour of the charge, the impact of this term on the dynamics of the electron will happen at third order
in the coupling strength e/m. Since we are gong to consider only up to second order contributions to time
evolution, we can safely ignore that term for now.
Notice that A(x) is a classical field, so it is not an Hermitian operator, but rather a vector field that
satisfies ∇·A = 0. In the Hamiltonian, x̂ corresponds to the position operator of the charged particle.
In general, that field would be a solution of Maxwell equations. Let us consider the particular case of a
monochromatic stationary wave:
A(x) = 2A0  cos(k·x − ωt)
where  is the polarization vector and the wave vector k, which gives the propagation direction, fulfills

|k| = ω, k· = 0

We can write the stationary wave as follows


 
A(x) = A0  ei(k·x−ωt) + e−i(k·x−ωt)

−e
which means that we can rewrite the term p̂·A(x̂) in (4.7.1) as
m
e  
− A0 p̂· ei(k·x̂−ωt) + e−i(k·x̂−ωt)
m
p̂2
Therefore we can treat this term like a harmonic perturbation over the atomic Hamiltonian Ĥ0 = 2m + eφ(x̂):

V̂ (t) = V̂eiωt + Vˆ† e−iωt

with
V̂ = −p̂· e−ik·x̂

Time dependent perturbation theory 16 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

eA0
and the coupling strength will be λ = .
m
As we learn from Fermi’s golden rule, we have two different processes that will dominate the dynamics in the
long time regimes: emission, where the electron reduces its energy by its interaction with the electromagnetic
field, and absorption, where the electron increases its energy from its interaction with the field.
Let us focus on the absorption case: since V̂ † = −p̂ ·  eik·x̂ we get that the differential rate of transition
from an eigenstate of the free Hamiltoinan |ni to another one |mi with energy En + ω is
e2 |A0 |2
wn→m = 2π |hm|eik·x̂ p̂·|ni|2 δ(Em − En − ω)
m2
A magnitude that is sometimes more useful that the transition rate is the differential cross section. The
differential cross section is defined as the ratio between the energy per unit time absorbed by the atom in the
transition |ni → |mi, and the total energy flux (energy per unit area per unit time) in the electromagnetic
field. Since the electromagnetic field considered is classical, this latter magnitude can be computed using
classical electromagnetism, the energy flux of the radiation field is simply:
1 E02 B02
 
1 2
ΦEM = + = ω |A0 |2
2 8π 8π 2π
where E0 and B0 are respectively the amplitudes of the electric and magnetic field, obtained from

E=− A, B =∇×A
∂t
In this fashion, the absorption cross section is given by
4π 2 e2
σabs = |hm|eik·x̂ p̂·|ni|2 δ(Em − En − ω)
m2 ω
As an exercise, let us, at least once, recover the full-dimensional expression of the cross section transforming
from natural units to fully dimensional units. We know that the result (a cross section) has units of area, so
we introduce the constants c, 4πε0 , ~ to obtain he correct dimensionality (units of length squared):
4π 2
σabs = ~α|hm|eik·x̂ p̂·|ni|2 δ(Em − En − ~ω)
m2 ω
where the fine structure constant is defined as
e2 1
α= '
4πε0 ~c 137

If we talk about atomic transitions, we need to explicitly write the electron wave function in the formula
above. We do that going into the position representation, something really easy R to do by introducing twice
the spectral decomposition of the identity in terms of the eigenstates of x̂: 11 = dx|xihx| we get
2
4π 2
Z Z
0 ik·x̂ 0 0
σabs = 2 ~α dx dx hm|xihx|e |x ihx |p̂·|ni δ(Em − En − ~ω)
m ω
we call ψn (x) = hx|ni the atomic electronic wavefunction of energy En and ψm (x) = hx|mi the atomic
wavefunction associated with the state of energy Em , using the orthogonality of that basis: hx|x0 i = δ(x − x0 ),
we readily get that:
2
4π 2
Z
ik·x ∗
σabs = 2 ~α  · dx e ψm (x)(−i~∇)ψn (x) δ(Em − En − ~ω)
m ω
where we have used that p̂ is the derivative in the position representation hx|p̂|ni = −i~∇ψn (x).

Time dependent perturbation theory 17 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

Multipole approximation

In order to evaluate the matrix element  · hm|eik·x̂ p̂|ni, if we are considering electronic transitions in atoms,
it is common to carry out the following approximation: the wavelength of the radiation field (λω ) interacting
with the atom is much larger than the localization of the atomic electrons (the size of the atom, typically). In
other words, λω ∼ |k|−1 = ω −1  ∆x, which in turn means that k will not vary very much over the spatial
extension of the atom.
If that hypothesis is correct, we can expand the exponential as follows:
1
eikn ·x = 11 + ikn · x̂ − (kn · x̂)2 + . . .
2
which we can write in an explicitly tensorial notation as
1
eikn ·x = 11 + ikα x̂α − kα kβ x̂α x̂β + . . .
2
Substituting term by term this into the matrix element we get, writing it in a complete tensor form
1
M = γ hm|(11 + ikα x̂α − kα kβ x̂α x̂β + . . . )pγ |ni
2
the multipole approximation consists of cropping this expansion to a finite number of terms. Let us consider
the first two contributions that give raise to the electric dipole and the electric quadrupole and magnetic dipole
transition amplitudes:

Order zero (electric dipole approximation):

This transition is often notated as E1. We can compute it if we just keep the zeroth order of expansion in the
exponential
M (0) = γ hm|p̂γ |ni
From the form of the free part of the Hamiltonian (4.7.1), Ĥ0 = p̂2 /2m, we get that

i
x̂γ , Ĥ0 = p̂γ ⇒ p̂γ = −im x̂γ , Ĥ0
   
(4.7.2)
m
therefore
M (0) = E1 = γ hm|p̂γ |ni = −im γ hm| x̂γ , Ĥ0 |ni = im(Em − En )γ hm|x̂γ |ni
 

Note that this is the matrix element of a vector operator, an object with angular momentum 1. Only transitions
where |ni and |mi have a difference in angular momentum equal to 1 will give a non-zero contribution to this
transition amplitude.

Order one (magnetic dipole and electric quadrupole approximation):

Considering the first order of expansion in the exponential we get M (1) = γ kα hm|x̂α p̂γ |ni. We can write this
matrix element as
1 h i
M (1) = γ kα hm|x̂α p̂γ |ni = γ kα hm|(x̂α p̂γ + p̂α x̂γ )|ni + hm|(x̂α p̂γ − p̂α x̂γ )|ni
2

Time dependent perturbation theory 18 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

The first summand represent the quadrupole electric transitions E2 and the second summand the dipole
magnetic transitions M1. To evaluate the first one, we use (4.7.2) again to substitute the momentum operator
by the commutator of Ĥ0 and x̂α :
−im  
α γ
  α  γ im im
γ kα hm| Ĥ0 , x̂α x̂γ |ni = (Em − En )hm|x̂α x̂γ |ni
 
E2 = γ kα hm| x̂ x̂ , Ĥ0 + x̂ , Ĥ0 x̂ |ni =
2 2 2
Note that this is the matrix element of a two indices tensor operator, an object with angular momentum 2.
Only transitions where |ni and |mi have a difference in angular momentum equal to 2 will give a non-zero
contribution to this transition amplitude.
Now the second summand can be written as the cross product between x̂ and p̂, namely,
1 1
M 1 = γ kα hm|(x̂α p̂γ − p̂α x̂γ )|ni = (k × ) · hm|(x̂ × p̂)|ni
2 2
The first term has the form of a B field if we remember that B = ∇ × A, whereas the second is the orbital
angular momentum. This is the magnetic dipolar contribution to the transition amplitude and it is, again, a
vector operator.
When we substitute this power expansion in the transition rates of emission and absorption of energy to
and from and electromagnetic field computed in the previous sections we obtain the electromagnetic selection
rules for the atomic quantum numbers in an electromagnetic transition.

4.8 The Photoelectric effect

We will now consider the process where an electron is ejected from an atom due to its interaction with a
classical electromagnetic field. This has the particularity that we are dealing with a transition, not between
different atomic energy levels, but between a bound state to a free electron state (which is in a continuum of
energies).
The first assumption we will make is that a free electron (in principle ignoring the electron spin degree of
freedom) is well described by an eigenstate of the free electron Hamiltoinan H0,free = p̂2 /2m. We notate the
state of a fee electron with momentum p as |pi. The position representation of that state is a plane wave.

hx|pi ≡ ψe (x) ∝ eip·x

however, we face the problem we discussed in Block 2: the eigenfunctions of the derivative operator are not
normalizable. Indeed these states are completely delocalized in space. To deal with this problem we are going
to use a very convenient trick. Let us consider, for now, that the electron is localized in a box of volume
V = L3 , on which we impose periodic boundary conditions (to allow for the existence of right-moving and
left-moving plane waves). Hopefully, we will be able to take the limit of infinite L and recover the free-space
case. Being in a box, now we are able to normalize the the plane wave states with the usual L2 inner product:
Z L Z L
1 ip·x
hx|pi = ψe (x) = e ⇒ hp|pi = dx hp|xihx|pi = dx |ψ(x)|2 = 1
L3/2 0 0

where the integration limits from 0 to L apply to the three spatial dimensions.
To apply Fermi golden rule, we need to obtain the density of final states. This can be done by counting
the different possible values of the electron’s momentum.

Time dependent perturbation theory 19 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

To do that we solve Schrödinger equation with periodic boundary conditions, which implies that for each
spatial component ψ(0) = ψ(L) ⇒ eipi 0 = 1 = eipi L ⇒ pi = 2ni π/L with ni ∈ Z. This means that

p= (nx , ny , nz ), ni ∈ Z
L
Imagine the possible electron momenta as points in a square lattice, to compute the density of states, we
are going to be taking the limit of very large L, or more concretely, |∆p|  2π/L, in that limit it is a
good approximation to treat n = (nx , ny , nz ) as a continuum. Let us consider, in that limit, a small volume
element such that the modulus of the (now continuous) vector n falls within |n| and |n| + d|n| and within the
differentially small solid angle dΩ. This is clearly

ρ(n)dn = |n|2 d|n|dΩ

Now, by assumption, the state of the free electron was an eigenstate of the fee Hamiltonian, so its energy
is just its eigenvalue, which we can easily compute
 2  2
p̂2 p2 p2 1 2π 2 L 1
|pi = |pi ⇒ E = = |n| ⇒ d|n| = m dE
2m 2m 2m 2m L 2π |n|

Therefore, writing the density of states in terms of final energies of free electrons:
 3  3
L √ L
ρ(E)dE = m 2mE dEdΩ = m|p| dEdΩ
2π 2π

so substituting in Fermi’s golden rule we now get that the differential cross-section per solid angle integrated
over all possible final states, is given by

4π 2 e2
Z Z
dΣabs dσabs
= ρ(E)dE = dE ρ(E) 2 |hp|eik·x̂ p̂·|ni|2 δ(E − En − ω)
dΩ dΩ m ω
Substituting the density of states
 3  3
4π 2 e2 4π 2 e2
Z
dΣabs L ik·x̂ 2 L
= dE m|p| 2 |hp|e p̂·|ni| δ(E − En − ω) = m|p| 2 |hp|eik·x̂ p̂·|ni|2
dΩ 2π m ω 2π m ω
p
where the energy conservation delta already fixed that |p| = 2m(En + ω) Let us insert the spectral decom-
position of the identity twice to write the initial atomic state and the final free electron case in the position
representation:
 3 2
4π 2 e2
Z Z
dΣabs L
= m|p| 2 · dx dx0 hp|xihx|eik·x̂ |x0 ihx0 |p̂·|ni
dΩ 2π m ω

This is to say,
3 2
4π 2 e2
 Z
dΣabs L
= m|p| 2 · dxeik·x ψe∗ (x)(−i∇)ψn (x)
dΩ 2π m ω
Given that the wavefunction of an electron in the ground state of an hydrogen atom and the wavefunction for
a free electron in a periodic box of length L are respectively:
s
1 −|x|/a0 1
ψn (x) = ψ1s (x) = 3 e , ψe (x) = 3/2 eip·x
πa0 L

Time dependent perturbation theory 20 Eduardo Martı́n-Martı́nez


Version 0.1 Eduardo Martin-Martinez - Amath 473

we get
3 s 2
4π 2 e2
 Z
dΣabs L 1 1 −|x|/a0
= m|p| 2 · dx eik·x 3/2 e−ip·x (−i∇) e
dΩ 2π m ω L πa30
which simplifying gives
2
e2
Z
dΣabs
= m|p| 2 3 2 · dx ei(k−p)·x (−i∇)e−|x|/a0 (4.8.1)
dΩ 2π a0 m ω

Notice first that the dependence on L cancels out, which tells us already that the result is independent of the
size of the periodic box we choose to be able to normalize the electron plane wave. in particular we can take
the limit of L → ∞ and we get the free space result.
Now we are going to play a little trick to simplify this integral, using Gauss theorem, it states that
Z I
dx ∇·F (x) = dS ·F (x) (4.8.2)
Σ

where the second integral is over the surface Σ that encloses the volume of integration. Now, if we make
F (x) = ψ1 (x)ψ2 (x), chain rule tells us that
 
∇· [ψ1 (x)ψ2 (x)] =  · [∇ψ1 (x)]ψ2 (x) + ψ1 (x)[∇ψ2 (x)]

Therefore, (4.8.2) gives


Z Z I
· dx ψ1 (x)∇ψ2 (x) = −· dx [∇ψ1 (x)]ψ2 (x) + · dS ·ψ1 (x)ψ2 (x)
Σ

in our particular case,


ψ1 (x) = ei(k−p)·x , ψ2 (x) = e−|x|/a0
And the surface interval is actually made over an infinitely big contour. The product of the two functions is
an L2 function, which means that the integral over the surface of constant radius |x| → ∞ is zero, due to the
e−|x| term. Therefore we obtain that (4.8.1) gets transformed into
2
e2 m|p|
Z
dΣabs h i
= 2 3 2 · dx (−i∇)ei(k−p)·x e−|x|/a0
dΩ 2π a0 m ω

and we can readily check that


−i·∇ei(k−p)·x = −(·p)ei(k−p)·x
where we used that  is perpendicular to k. Therefore the cross section results in

dΣabs e2 (·p)2
= m|p| 2 3 2 |F[ψ1s (x)]|2
dΩ 2π a0 m ω

where Z
F[ψ1s (x)] = dx ei(k−p)·x e−|x|/a0

is the Fourier transform of the atomic wavefunction of the 1s orbital of the hydrogen atom.

Time dependent perturbation theory 21 Eduardo Martı́n-Martı́nez

You might also like