CHAPTER 7: APPROXIMATION METHODS FOR TIME-DEPENDENT

PROBLEMS

(From Cohen-Tannoudji, Chapter XIII)

A. STATEMENT OF THE PROBLEM

Consider a system with Hamiltonian Ĥ0; its eigenvalues and eigenvectors are

Ĥ0|ϕni = En|ϕni (7.1)

(Ĥ0 is discrete and non-degenerate for simplicity.)

At t = 0, a perturbation is applied

Ĥ(t) = Ĥ0 + W(t) = Ĥ0 + λŴ(t) (7.2)

where λ  1, and Ŵ(t) = 0 for t < 0:

t<0 t=0 t>0

stationary state W(t) final state
|ϕii evolution starts |ψ(t)i
eigenstate of Ĥ0 (|ϕii is not eigenstate of Ĥ )
What is the probability P f i(t) of finding the system in another eigenstate |ϕ f i of Ĥ0
at time t?

Treatment: solve the Schrödinger equation (S. E.)

d h i
i~ |ψ(t)i = Ĥ0 + λŴ(t) |ψ(t)i (7.3)
dt

with the initial condition |ψ(0)i = |ϕii

2
⇒ P f i(t) = hϕ f |ψ(t)i (7.4)

In generally this problem is not rigorously soluble!

⇒ we need APPROXIMATION METHODS
B. APPROXIMATE SOLUTION OF THE SCHRÖDINGER EQUATION
1. The Schrödinger equation in the {|ϕni} representation
We will use the {|ϕni} representation which is convenient as |ϕii and |ϕ f i are eigen-
states of Ĥ0, and obtain the differential equations for the components of the state
vector
X
|ψ(t)i = cn(t)|ϕni (7.5)
n
cn(t) = hϕn|ψ(t)i (7.6)
Ŵnk (t) = hϕn|Ŵ(t)|ϕk i (7.7)
and hϕn|Ĥ0|ϕk i = Enδnk (7.8)
We will project both sides of S.E. onto |ϕni (and use k |ϕk ihϕk | = 1̂):
P

d h i
i~ |ψ(t)i = Ĥ0 + λŴ(t) |ψ(t)i (7.9)
dt
d X
⇒ i~ cn(t) = Encn(t) + λŴnk (t)ck (t) (7.10)
dt
k
Changing functions
If λŴ(t) = 0 then the equations decouple
d
i~ cn(t) = Encn(t) (7.11)
dt
and yield simple solution

cn(t) = bne−iEnt/~ (7.12)

where bn is a constant depending on the initial conditions.

If λŴ(t) , 0 and λ  1, we expect the solutions cn(t) of the full equations to be very
close to the solution above (for λŴ(t) = 0), and thus if we perform the change of
function

cn(t) = bn(t)e−iEnt/~ (7.13)

we can predict that bn(t) will be slowly varying functions of time.

Substituted into the equation gives
d
i~e−iEnt/~ bn(t) + Enbn(t)e−iEnt/~
dt X
−iE
= Enbn(t)e n + t/~ λŴnk (t)bk (t)e−iEk t/~ (7.14)
k
E −E
Multiplying both sides by eiEnt/~ and introducing the Bohr frequency ωnk = n ~ k
gives
d X
i~ bn(t) = λ eiωnk t Ŵnk (t)bk (t) (7.15)
dt
k
2. Perturbation equations

In general, the solution is not known exactly and, for λ  1, we try to determine this
solution in the form of a power series in λ
(0) (1) (2)
bn(t) = bn (t) + λbn (t) + λ2bn (t) + . . . (7.16)

and substitute it into the equation, and set equal the coefficients of λr on both sides
of the equation
d (0)
i) r = 0 : i~ bn (t) = 0 (7.17)
dt
d (r) X
(r−1)
ii) r , 0 : i~ bn (t) = eiωnk t/~Ŵnk (t)bk (t) (7.18)
dt
k
RECURRENCE!
3. Solution to the first order in λ

a. The state of the system at time t

t<0: |ϕii i.e. bi(t) , 0, bk (t) = 0 ∀k , i (7.19)

t=0: Ĥ0 → Ĥ0 + λŴ and solution of S.E. is continuous at t = 0 (7.20)
⇒ bn(t = 0) = δni ∀λ (7.21)
(0)
⇒ bn (t = 0) = δni (7.22)
(r)
⇒ bn (t = 0) = 0 if r ≥ 1 (7.23)
(0)
d b (t) = 0 we get
and with i~ dt n

(0)
0th-order solution: bn (t) = δni for all t > 0
 d (1) X
1st − order: i~ bn (t) = eiωnk t Ŵnk (t)δki (7.24)
dt
k
= eiωnit Ŵni(t) (7.25)
Z t
(1) 1 0
Ŵni t0 dt0
 
By integration bn (t) = e iω ni t (7.26)
i~ 0
 (0) (1)
cn(t) = bn(t)e−iEnt/~ ≈ bn (t) + λbn (t) e−iEnt/~

(7.27)

to the first order time-dependent perturbation theory we get the state of the system
at time t calculated to the first order:
X
|ψ(t)i ≈ cn(t)|ϕni (7.28)
n
b. The transition probability Pi f (t)

2 2
c f (t) =
hϕ f |ψ(t)i = Pi f (t) (7.29)
c f (t) = b f (t)e−iE f t/~ (7.30)
2
⇒ Pi f (t) = b f (t) (7.31)
(0) (1)
where b f (t) = b f (t) + λb f (t) + . . .

Let us assume |ϕii and |ϕ f i are different (i.e. we are concerned only with transition
induced by λŴ between two distinct stationary states of Ĥ0):
(0)
b f (t) = 0 and consequently

2 (1) 2
Pi f (t) = λ b f (t) (7.32)
 (1)
and using the formula for bn (t) we get
2
Z t
1 iω f i t0
W f i t0 dt0
 
Pi f (t) = 2 e (7.33)
~ 0 | {z }
W(t)=λŴ

Consider the function W̃ f i(t0) which is zero for t0 < 0 and T 0 > t and is equal to
W f i(t0) for 0 ≤ t0 ≤ t.
W̃ f i(t0) is the matrix element of the perturbation “seen” by the system between the
time t = 0 and the measurement time t, when we try to determine if the system is in
the state |ϕ f i.
Pi f (t) is proportional to the square of the modulus of the Fourier transform of the
perturbation actually “seen” by the system, W̃ f i(t).
C. SPECIAL CASE: A SINUSOIDAL OR CONSTANT PERTURBATION

Ŵ(t) = Ŵ sin ωt or
Ŵ(t) = Ŵ cos ωt
Ŵ is a time independent observable and ω a constant angular frequency.

(Example: electromagnetic wave of angular frequency ω.

Pi f (t) is the probability, induced by monochromatic radiation, of a transition between
the initial state |ϕii and the final state |ϕ f i.)

Ŵ f i  iωt −iωt

Ŵ f i(t) = Ŵ f i sin ωt = e −e (7.34)
2i
Ŵ f i is a time independent complex number and
Z th
(1) Ŵni i(ω +ω)t 0 i(ω −ω)t 0i 0
bn (t) = − e ni − e ni dt (7.35)
2~ 0
 i(ω +ω)t i(ω −ω)t
 
Ŵni  1 − e ni 1−e ni
=

− (7.36)
ωni + ω ωni − ω
 
2i~
The transition probability becomes
    2
Wfi
2 i ω f i +ω t i ω f i −ω t
2 (1) 2 1−e 1−e
Pi f (t; ω) = λ b f (t) = − (7.37)
4~2 ωfi + ω ωfi − ω

(Pi f depends on the frequency of the perturbation)

If Ŵ f i(t) = Ŵ f i cos ωt,

    2
Wfi
2 i ω f i +ω t i ω f i −ω t
1−e 1−e
Pi f (t; ω) = + (7.38)
4~2 ωfi + ω ωfi − ω
Constant perturbation ω = 0
2 2
Wfi iω f i t 2 W fi  
Pi f (t; ω) = 1−e = F t; ω f i (7.39)
~ ωfi
2 2 ~2
   2
   sin ω f it/2 
F t; ω f i =   (7.40)
ω f i/2 
2. Sinusoidal perturbation which couples discrete states: resonance
a. Resonant nature of the transition probability
When t is fixed, Pi f (t; ω) is a function of one variable ω. This function has a maximum
for ω ' ω f i or ω ' −ω f i; this is a resonance phenomenon (choose ω ≥ 0)
 2
   
Ŵ f i
2 i ω +ω t i ω −ω t
1 − e fi 1 − e fi
Pi f (t; ω) = − (7.41)
4~2 ω i+ω ω i−ω
| f{z } | f{z }
A+ A−

h  i
 
i ω f i +ω t/2
sin ω f i + ω t/2
A+ = −ie   (7.42)
ω f i + ω /2
| {z }
goes to zero for ω=−ω f i
This term is anti-resonant for ω = ω f i (and resonant for ω = −ω f i)
Resonant term
h  i
  sin
i ω f i −ω t/2
ω f i − ω t/2
A− = −ie   (7.43)
ω f i − ω /2
Consider the case ω − ω f i  ω f i (this is the resonant approximation):
1st order transition probability:
2
Wfi  
Pi f (t; ω) = F t; ω − ω f i (7.44)
4~2
 h  i 2
 sin ω f i − ω t/2 
  
 
F t; ω − ω f i = 
 
(7.45)

 
ω f i − ω /2 

 

| {z }  
sinc function
b. The resonance width and time-energy uncertainty relation

The most of the resonant peak is concentrated around the resonant frequency ω f i,
2
 
ω−ω f i t 3π W f i t2
for example at 2 = 2 we get the transition probability which is approx-
9π2 ~2
imately 5% of the transition probability at the resonance.

We can define the width of the resonant peak as the difference between the frequen-
cies of the minima of Pi f around the resonant frequency, see the figure, then

4π
∆ω ' (7.46)
t
which is analogous to the time-energy uncertainty relation ∆E = ~∆ω ' ~t
c. Validity of the perturbation treatment
a) Discussion of the resonant approximation
A+ has been neglected relative to A−:
|A−(ω)|2 sinc function
 2
|A+(ω)|2 = |A−(−ω)|2  A− ω f i (7.47)

The resonant approximation is justified on the condition

2 ωfi >> ∆ω (7.48)

that is
1 1
t >> ' (7.49)
|{z}
ωfi ω
duration of the perturbation |{z}
oscillation period
b) Limits of the first-order calculations
If t becomes too large, the first-order approximation can cease to be valid (i.e. giving
infinit transition probability which is physically a nonsense):
2
  Wfi 2 = ∞
lim Pi f t; ω = ω f i = lim t (7.50)
t→∞ t→∞ 4~2
For the first-order approximation to be valid at resonance, Pi f (t; ω = ω f i)  1:

~
t  (7.51)
Wfi
3. Coupling with the states of the continuum

E f belongs to a continuous part of the spectrum of Ĥ0

⇓
We cannot measure the probability of finding the system in a well-defined state |ϕ f i
at time t
⇓
2
We have to integrate over probability density hϕ f |ψ(t)i over a certain group of final
states.
a. Integration over a continuum of final states; density of states

a) Example
– spinless particle of mass m
– scattering by a potential W(~r)

E = ~p2/2m, |ψ(t)i can be expanded in terms of |~pi

The corresponding wavefunctions are plane waves

1 3/2 i~p·~r/~
!
h~r|~pi = e (7.52)
2π~
The probability density
2
h~p|ψ(t)i (7.53)
Detector gives a signal when the particle is scattered with the momentum ~p f but
since it has a finite aperture it really gives the signal when the particle has momentum
in a domain D f of ~p-space around ~p f (δΩ f , δE f )
Z
  2
δP ~p f , t = d3 ~p h~p|ψ(t)i (7.54)
~p f ∈D f

d3 ~p = p2dp dΩ
|{z} = ρ(E)
|{z} dEdΩ
solid angle around ~p f density of final states

2 dp 2 m √
ρ(E) = p =p =m 2mE (7.55)
dE p
Z
  2
δP ~p f , t = dΩdE ρ(E) h~p|ψ(t)i (7.56)
Ω∈δΩ f ,E∈δE f
b) The general case
Eigenstates of Ĥ0, labeled by a continuous set of indices

hα|α0i = δ(α − α0) (7.57)

at time t: |ψ(t)i
  Z
δP α f , t = dα |hα|ψ(t)i|2 (7.58)
α∈D f
Change variables and introduce density of final states

dα = ρ(β, E)dβdE (7.59)

  Z
δP α f , t = dβdE ρ(β, E) |hβ, E|ψ(t)i|2 (7.60)
β∈δβ f ,E∈δE f
Fermi’s Golden Rule

Let |ψ(t)i be the normalized state vector of the system at time t.

Consider a system which is initially in an eigenstate |ϕii of Ĥ0 (in discrete part of
spectrum)
 
δP ϕi, α f , t = ? (7.61)

The calculations for the case of a sinusoidal or constant perturbation remain valid
when the final state of the system belongs to the continuous spectrum of Ĥ0
For W constant
1  E−E 
i
|hβ, E|ψ(t)i|2 = 2 |hβ, E|W|ψ(t)i|2 F t; (7.62)
~ ~
E – energy of the state |β, Ei
Ei – energy of the state |ϕii
Z  E−E 
  1 i
δP ϕi, α f , t = 2 dβdE ρ(β, E) |hβ, E|W|ψ(t)i|2 F t;
~ β∈δβ f ,E∈δE f ~
(7.63)
 E−E 
F t; ~ i varies rapidly about E = Ei; for sufficiently large t, this function can be
approximated, to within a constant factor, by the δ-fucntion δ (E − Ei):
 E−E  E − E 
i i
lim F t; = πtδ = 2π~tδ (E − Ei) (7.64)
t→∞ ~ 2~
The function ρ(β, E) |hβ, E|W|ψ(t)i|2 varies much more slowly with E . We will assume
that t is sufficiently large for the variation of this function over an energy interval of
width 4π~/t centered at E = Ei to be negligible.

 E−E 
⇒ We can replace F t; ~ i by 2π~tδ (E − Ei) which allows us to integrate over E
immediately.
If, in addition, δβ f is very small, integration over β is unnecessary and we get
(a) Ei ∈ δE f
  2π 2  
δP ϕi, α f , t = δβ f t hβ f , E f = Ei|W|ϕii ρ β f , E f = Ei (7.65)
~
(b) Ei < δE f
 
δP ϕi, α f , t = 0 (7.66)

⇒ A constant perturbation can induce transitions only between states of equal ener-
gies, and thus (b) holds.
The probability (a) increases linearly with t.
⇒ We can define

 
• transition probability per unit time δW ϕi, α f
  d  
δW ϕi, α f = δP ϕi, α f , t (7.67)
dt
which is time independent

• transition probability density per unit time and per unit interval of the variable β f
 
  δW ϕi, α f
w ϕi , α f = (7.68)
δβ f
Fermi’s Golden Rule
  2π 2  
w ϕi , α f = hβ f , E f = Ei|W|ϕii ρ β f , E f = Ei (7.69)
~

Assume that W is a sinusoidal perturbation which couples a state |ϕii to the contin-
uum of states |β f , E f i with energies E f close to Ei + ~ω. We can carry out the same
procedure as above:
  π 2  
w ϕi , α f = hβ f , E f = Ei + ~ω|W|ϕii ρ β f , E f = Ei + ~ω (7.70)
2~