Quantum Mechanics-II

1
S. Shankaranarayanan
Department of Physics, IIT Bombay, Powai, India

E-mail: shanki@[Link]

1
Homepage, Uses [Link]
Contents

I Time-dependent Perturbation 1
1 Introduction: The Need for Time-Dependent Theory 1

2 A 2-State Example: Ammonia Molecule (N H3 ) 2

2.1 Tunneling and Inversion 2
2.2 Interaction with an External Field 2

3 Mathematical Formulation: The Two-State System 3

3.1 The Unperturbed System 3
3.2 Introducing the Perturbation 3
3.3 The Time-Dependent Schrödinger Equation 4

4 Exact Solution: Harmonic Perturbation 5

5 Rabi Oscillations and Resonance 6

5.1 Physical Implications of the Rabi Formula 6

6 Introduction to Time-Dependent Perturbation Theory 7

6.1 Zeroth Order (No Perturbation) 7
6.2 First Order Approximation 7
6.3 Second Order Approximation 7

7 Application: Sinusoidal Perturbation 8

7.1 First-Order Transition Amplitude 8
7.2 The Rotating Wave Approximation (RWA) 8

8 Transition Probabilities 9
8.1 Case 1: Absorption (|1⟩ → |2⟩) 9
8.2 Case 2: Stimulated Emission (|2⟩ → |1⟩) 10

9 Application: The Physics of LASERs 10

9.1 The Equality of Transition Probabilities 10
9.2 The Necessity of Population Inversion 11
–i–
 9.3 Why a Two-Level Laser is Impossible 11
9.4 The Solution: 3-Level and 4-Level Systems 11

10 Summary of the Perturbative Expansion 12

11 TDPT for an Arbitrary Dimensional Hilbert Space 12

11.1 The Coupled Differential Equations 13

12 The Dirac (Interaction) Representation 13

12.1 Defining the Interaction Picture 13
12.2 Evolution of the State Vector 14

13 How the Interaction Picture results match with 2-State system results? 15

14 Summary of Quantum Representations 16

15 The Time-Evolution Operator in the Interaction Picture 16

16 Iterative Solution: The Dyson Series 17

17 Physical Interpretation: Virtual Transitions and Uncertainty 18

17.1 Connection to the Heisenberg Uncertainty Principle 18

18 The Time-Ordering Operator (T ) 19

18.1 Visualizing the Integration Domain 19
18.2 The Full Dyson Series 20

19 Defining the Transition Probability 20

19.1 Expansion via the Dyson Series 21

20 Application: A Constant Perturbation 21

20.1 Physical Features of the Probability Curve 22

21 Transition Rates and Fermi’s Golden Rule 23

22 The Physical Meaning of the Golden Rule 23

22.1 The Apparent Triviality of the Delta Function 23
22.2 The Resolution: Transitions to a Continuum 24

– ii –
23 Discrete to Continuum transition probability 24
23.1 Integrating the Probability 24
23.2 Fermi’s Golden Rule 25

24 Second-Order Transitions (Fermi’s Golden Rule No. 1) 27

25 Application of Second-Order Transitions: Raman Scattering 28

25.1 Why is Raman Scattering a Second-Order Process? 29
25.2 The Three Scattering Regimes 29
25.3 Connecting to Fermi’s Golden Rule No. 1 30
25.4 Modern Applications 30

II Wentzel, Kramers, Brillouin (WKB) Approximation 31

26 Introduction: WKB vs. Perturbation Theory 31
26.1 WKB vs. The Variational Method 31

27 Review: The Constant Potential 33

28 The WKB Ansatz for a Slowly Varying Potential 33

28.1 Expansion in Powers of ℏ 34
28.1.1 Constructing the Final Wave Function 36
28.2 Expansion in Powers of ℏ 36
28.2.1 Zeroth-Order Approximation 36
28.2.2 First-Order Approximation 37
28.3 The WKB Wave Function upto first order 37
28.4 Why can we Expand in ℏ? 37

29 Breakdown of WKB: Classical Turning Points 39

29.1 Linearizing the Potential: The Airy Equation 39
29.2 Asymptotic Behavior of Airy Functions 40

30 The WKB Connection Formulas 40

31 Understanding the Connection Formulas Mathematically 41

31.1 Transforming the Schrödinger Equation 42
31.2 Analytic Continuation Around the Turning Point 43

– iii –
31.3 Synthesizing the Connection Formula 45

– iv –
 PART

Time-dependent Perturbation I
Reference
Section ??:
For 2-state system, see Feynman Vol: 3, Chapter 9.
Table 1. Contents for Part
For N-state system, see Sakurai Sec. 5.5 and Shankar Chapter 17.
II

1 Introduction: The Need for Time-Dependent

Theory
In our previous studies, we focused on systems where the potential was purely
time-independent, i.e., V = V (x, y, z). The central questions we asked were
kinematical in nature:

• What are the energy eigenvalues and eigenstates of the system?

• How can we solve for them exactly, or approximately using methods like
the time-independent perturbation theory or variational method?

However, we also want to understand the dynamical properties of

quantum systems. For this, we must ask entirely different set of questions:

1. How does a system (like a Hydrogen atom) respond to a time-varying

external field or current?

2. Do quantum systems exhibit resonance, analogous to classical systems?

[

3. If a system is initially prepared in state |1⟩, and we apply a time-

dependent external potential for a short duration, what is the prob-
ability that the system transitions to another state |2⟩?

To keep our initial mathematical framework transparent and simple, we will

start by analyzing a 2-dimensional Hilbert space (a two-state system)
subjected to a time-dependent coupling. This foundational model has critical
applications in fields such as Nuclear Magnetic Resonance (NMR) and Laser
physics.
1
A 2-State Example: Ammonia Molecule (N H3 ) 2

2 A 2-State Example: Ammonia Molecule (N H3)

A classic example of a two-state system is the Ammonia molecule (N H3 ) (see
Feynman Lectures on Physics, Vol 3, Chapter 9).
As shown below, the molecule forms a tetrahedral structure. By symme-
try, one might expect the logical position for the Nitrogen atom to be exactly
in the center of the equilateral triangle formed by the three Hydrogen atoms.
If it were in the center, the H-N-H bond angles would be 120◦ .

However, observations show that the Nitrogen atom does not sit in the plane
of the Hydrogen atoms; the actual bond angle is ≈ 108◦ . The Nitrogen atom
sits either above the plane (|1⟩, the "top" configuration) or below the plane
(|2⟩, the "bottom" configuration).

2.1 Tunneling and Inversion

As we discussed in Q,Mech 1 course, the molecule can spontaneously flip
between the top and bottom configurations at a frequency of 24 GHz. For
the Nitrogen atom to move from the top to the bottom, it must push its
way through the plane of Hydrogen atoms. As it approaches the plane, it
experiences a repulsive energy barrier.
Classically, if the molecule lacks the thermal energy to overcome this
barrier, the flip cannot occur. Quantum mechanically, however, the Nitrogen
atom can tunnel through the barrier, leading to a natural oscillation between
states |1⟩ and |2⟩.

2.2 Interaction with an External Field

The valence electrons in N H3 tend to reside closer to the Nitrogen atom,
giving it a slight negative charge and leaving the Hydrogens slightly positive.
Consequently, the molecule possesses a permanent electric dipole moment µe
(charge × distance).
Mathematical Formulation: The Two-State System 3

If we place the ammonia molecule in an external time-dependent electric

field E(t), the interaction energy is given by:

V̂ (t) = −µe · E(t) (2.1)

This external interaction couples the two states, creating a time-dependent

perturbation that can drive transitions between them.

3 Mathematical Formulation: The Two-State Sys-

tem

3.1 The Unperturbed System

Consider a general two-state system governed by an unperturbed Hamiltonian

Ĥ0 . In the basis of its eigenstates |1⟩ and |2⟩, the matrix representation of
Ĥ0 is diagonal:  
E 1 0
Ĥ0 =   (3.1)
0 E2
where E1 and E2 are the unperturbed energy eigenvalues[cite: 144].

3.2 Introducing the Perturbation

We now add a time-dependent interaction potential V̂ (t), making the total

Hamiltonian:
 
E1 V11 (t) 
Ĥ(t) = Ĥ0 + V̂ (t) =  (3.2)
V21 (t) E2 + V22 (t)

Any arbitrary state of the system |Ψ(t)⟩ can be written as a time-dependent

superposition of the unperturbed eigenstates:

|Ψ(t)⟩ = c1 (t)e−iE1 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ |2⟩ (3.3)

where the phase factors e−iEn t/ℏ explicitly account for the standard time evo-
lution of the unperturbed states The time-varying coefficients c1 (t) and c2 (t)
contain the transition dynamics caused entirely by V̂ (t). Conservation of
probability requires the normalization condition:

|c1 (t)|2 + |c2 (t)|2 = 1 (3.4)

Mathematical Formulation: The Two-State System The Time-Dependent Schrödinger Equation 4

3.3 The Time-Dependent Schrödinger Equation

To find the evolution of the coefficients c1 (t) and c2 (t), we substitute our state
into the time-dependent Schrödinger equation:

∂
iℏ |Ψ(t)⟩ = [Ĥ0 + V̂ (t)]|Ψ(t)⟩ (3.5)
∂t
Evaluating the time derivative on the left-hand side using the product rule:

∂ ∂ h i
iℏ |Ψ(t)⟩ = iℏ c1 (t)e−iE1 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ |2⟩
∂t ∂t
iE1
 
= iℏ ċ1 (t)e−iE1 t/ℏ |1⟩ − c1 (t)e−iE1 t/ℏ |1⟩
ℏ
iE2
 
+ iℏ ċ2 (t)e−iE2 t/ℏ |2⟩ − c2 (t)e−iE2 t/ℏ |2⟩ (3.6)
ℏ
Notice that the terms involving the energies E1 and E2 perfectly reconstruct
the action of the unperturbed Hamiltonian Ĥ0 :

∂  
iℏ |Ψ(t)⟩ = iℏ ċ1 (t)e−iE1 t/ℏ |1⟩ + ċ2 (t)e−iE2 t/ℏ |2⟩
∂t
+ E1 c1 (t)e−iE1 t/ℏ |1⟩ + E2 c2 (t)e−iE2 t/ℏ |2⟩
 
= iℏ ċ1 (t)e−iE1 t/ℏ |1⟩ + ċ2 (t)e−iE2 t/ℏ |2⟩ + Ĥ0 |Ψ(t)⟩ (3.7)

Equating this result to the right-hand side of the Schrödinger equation

([Ĥ0 + V̂ (t)]|Ψ(t)⟩), the Ĥ0 |Ψ(t)⟩ terms exactly cancel. This leaves a simplified
equation governing purely the transition dynamics:
 
iℏ ċ1 (t)e−iE1 t/ℏ |1⟩ + ċ2 (t)e−iE2 t/ℏ |2⟩ = V̂ (t)|Ψ(t)⟩ (3.8)

To find the equations of motion for the coefficients c1 (t) and c2 (t), we take
the inner product of this equation with the basis bras ⟨1| and ⟨2|. Projecting
onto ⟨1| yields:

ih i
ċ1 (t) = − c1 (t)V11 + c2 (t)V12 e−i(E2 −E1 )t/ℏ (3.9)
ℏ
Projecting onto ⟨2| yields:

ih i
ċ2 (t) = − c2 (t)V22 + c1 (t)V21 ei(E2 −E1 )t/ℏ (3.10)
ℏ

where we have defined the matrix elements Vij ≡ ⟨i|V̂ (t)|j⟩. Note the Her-
miticity condition requires V12 = V21∗ .
Simplification: For many physical applications (like dipole transitions),
the diagonal matrix elements of the perturbation vanish, so we set V11 =
Exact Solution: Harmonic Perturbation 5

V22 = 0. Let us also define the natural Bohr frequency of the transition as
ω0 = E2 −E
ℏ
1
. Assuming E2 > E1 , ω0 > 0. The coupled equations simplify to:

i
ċ1 (t) = − V12 (t)e−iω0 t c2 (t) (3.11)
ℏ
i
ċ2 (t) = − V21 (t)eiω0 t c1 (t) (3.12)
ℏ
This is exactly what we derived earlier, however, in the current scenario the
external potential is time-dependent.

4 Exact Solution: Harmonic Perturbation

The differential equations above are exact, but they are generally impossible
to solve analytically for an arbitrary time-dependent potential V (t). However,
an exact solution exists for a monochromatic harmonic perturbation.
Consider an external field oscillating at frequency ω, such that the cou-
pling is:
V12 (t) = γℏeiωt =⇒ V21 (t) = γℏe−iωt (4.1)

where γ is a constant representing the coupling strength. Substituting this

into our coupled equations:

ċ1 (t) = −iγei(ω−ω0 )t c2 (t) (4.2)

ċ2 (t) = −iγe−i(ω−ω0 )t c1 (t) (4.3)

By differentiating the first equation and substituting the second, we de-

couple the system into a single second-order linear differential equation for
c1 (t):
c̈1 (t) − i(ω − ω0 )ċ1 (t) + γ 2 c1 (t) = 0 (4.4)

This is a standard ODE with constant coefficients and resembles a harmonic

oscillator in the presence of complex dissipative force! Assuming a solution of
the form c1 (t) = αeiλ1 t + βeiλ2 t , the roots of the characteristic equation are:
q
(ω − ω0 ) ± (ω − ω0 )2 + 4γ 2 ω − ω0
λ1,2 = = ±Ω (4.5)
2 2
where we have defined the generalized Rabi flopping frequency
q
Ω≡ γ 2 + (ω − ω0 )2 /4 .
Rabi Oscillations and Resonance 6

5 Rabi Oscillations and Resonance

Assuming the system is initially entirely in the ground state at t = 0, we

apply the initial conditions c1 (0) = 1 and c2 (0) = 0. Solving for the constants
α and β, we can determine c2 (t). The probability of finding the system in the
excited state |2⟩ at time t is P1→2 (t) = |c2 (t)|2 .
Repeating the calculations we have done many times earlier, we obtain
the famous Rabi Formula:

γ2
|c2 (t)|2 = sin2 (Ωt) (5.1)
Ω2

Rabi Oscillations at Resonance (ω = ω0 )

1 Ground State |c1 (t)|2

Excited State |c2 (t)|2
Probability

0.5

0
0 π π 3π 2π
2Ω Ω 2Ω Ω

Time (t)

5.1 Physical Implications of the Rabi Formula

1. Resonance: The maximum transition probability (the amplitude of
the oscillation) is strongly peaked when the driving frequency matches
the natural frequency (ω ≈ ω0 ). When exactly on resonance (ω = ω0 ),
the amplitude is exactly 1. The system oscillates fully back and forth
between |1⟩ and |2⟩.

2. Off-Resonance: If ω is far from ω0 , the amplitude drops off rapidly,

and the system almost never reaches state |2⟩.

3. Peak Width: If the coupling strength γ is reduced, the maximum

amplitude at resonance remains 1, but the width of the resonance peak
becomes narrower.

This periodically forced two-state system demonstrates the fundamental fea-

tures of absorption and stimulated emission in electromagnetic fields. When
ω ≈ ω0 , the system absorbs a quantum of radiation from the applied field.
Introduction to Time-Dependent Perturbation Theory 7

6 Introduction to Time-Dependent Perturbation

Theory

The exact solution we just derived is great, however, it is only possible for
very specific, simple time-dependent potentials (like a pure sine wave). If the
external perturbation V (t) is arbitrary, we cannot solve equations (3.11) and
(3.12) exactly. However, if the perturbation is weak (small), we can approx-
imate the solution using a systematic expansion known as Time-Dependent
Perturbation Theory (TDPT).
We expand our coefficients in a power series of a small parameter tracking
the order of the perturbation:

(0) (1) (2)

ci (t) = ci (t) + ci (t) + ci (t) + . . . (6.1)

6.1 Zeroth Order (No Perturbation)

If there is no perturbation (V = 0), the system will stay exactly as it started

forever. Assuming the system starts in the ground state |1⟩:

(0) (0)
c1 (t) = 1, c2 (t) = 0 (6.2)

6.2 First Order Approximation

To find the first-order correction, we substitute the zeroth-order values into

the right-hand side of our coupled differential equations:

(1) i (0)
ċ1 (t) = − V12 (t)e−iω0 t c2 (t) = 0 (6.3)
ℏ
(1) i (0) i
ċ2 (t) = − V21 (t)eiω0 t c1 (t) = − V21 (t)eiω0 t (6.4)
ℏ ℏ
Integrating with respect to time gives us the first-order probability amplitude
to transition into state 2:

(1) iZt ′
c2 (t) = − V21 (t′ )eiω0 t dt′ (6.5)
ℏ 0

6.3 Second Order Approximation

We repeat the iterative process. To find the second-order correction to state

1, we plug the first-order results into the differential equation:

(2) i (1)
ċ1 (t) = − V12 (t)e−iω0 t c2 (t) (6.6)
ℏ
Application: Sinusoidal Perturbation 8

(1)
Substituting our expression for c2 and integrating from 0 to t:
2 Z t Z t′′
i

(2) ′′ ′
c1 (t) = − dt′′ V12 (t′′ )e−iω0 t dt′ V21 (t′ )eiω0 t (6.7)
ℏ 0 0

(2) (1)
Note that c2 (t) = c1 (t).

7 Application: Sinusoidal Perturbation

Let us apply first-order TDPT to a highly practical scenario: an atom sub-
jected to a monochromatic oscillating field (such as a laser or microwave
radiation). The time-dependent potential is sinusoidal:

eiωt + e−iωt
!
V̂ (t) = V̂ cos(ωt) = V̂ (7.1)
2

where V̂ contains the spatial dependence of the perturbation, and ω is the

driving frequency. Let V21 = ⟨2|V̂ |1⟩ be the time-independent spatial matrix
element.

7.1 First-Order Transition Amplitude

Assume the system starts entirely in the ground state |1⟩ at t = 0. The
probability amplitude to find the system in the excited state |2⟩ at time t is
given by our first-order integral:

(1) iZt ′
c2 (t) =− V21 (t′ )eiω0 t dt′ (7.2)
ℏ 0
Substituting our sinusoidal potential:

(1) iV21 Z t  iωt′ ′

 ′
c2 (t) = − e + e−iωt eiω0 t dt′ (7.3)
2ℏ 0

(1) iV21 Z t h i(ω0 +ω)t′ ′

i
c2 (t) =− e + ei(ω0 −ω)t dt′ (7.4)
2ℏ 0
Performing the integration:

V21 ei(ω0 +ω)t − 1 ei(ω0 −ω)t − 1

" #
(1)
c2 (t) =− + (7.5)
2ℏ ω0 + ω ω0 − ω

7.2 The Rotating Wave Approximation (RWA)

We now analyze the two terms in the brackets:
Transition Probabilities 9

• The first term has a denominator of (ω0 + ω). Since both frequencies
are positive, this denominator is very large, making the term’s overall
contribution extremely small and rapidly oscillating.

• The second term has a denominator of (ω0 −ω). If the driving frequency
is close to the natural transition frequency (ω ≈ ω0 ), this denominator
approaches zero, making this the dominant resonance term.

Near resonance, we can safely ignore the first term. This is known as the
Rotating Wave Approximation (RWA). Keeping only the resonant term,
we factor out ei(ω0 −ω)t/2 :

(1) V21 ei(ω0 −ω)t − 1

c2 (t) ≈ −
2ℏ ω0 − ω
V21 i(ω0 −ω)t/2 ei(ω0 −ω)t/2 − e−i(ω0 −ω)t/2
" #
=− e
2ℏ ω0 − ω
V21 i(ω0 −ω)t/2 sin[(ω0 − ω)t/2]
= −i e (7.6)
ℏ ω0 − ω

Probability P1→2 (t)

Time Evolution (t → ∞):

|V21|2 t2 1. Central peak height grows as t2 .
4h̄2 2. Central peak width shrinks as 1/t.
3. Area under peak ω0 grows as t.
Approaches a Dirac delta function
enforcing energy conservation.

1
Width ∝ t

Driving Frequency (ω)

6π 4π 2π 2π 4π 6π
ω0 − t
ω0 − t
ω0 − t
ω0 + t
ω0 + t
ω0 + t

8 Transition Probabilities

8.1 Case 1: Absorption (|1⟩ → |2⟩)

Let us assume the atom is initially in the lower energy state |1⟩. The prob-
(1)
ability of transitioning to the higher energy state |2⟩ is P1→2 (t) = |c2 (t)|2 .
Squaring the amplitude derived above:

|V21 |2 sin2 [(ω0 − ω)t/2]

P1→2 (t) = (8.1)
ℏ2 (ω0 − ω)2
Application: The Physics of LASERs Case 2: Stimulated Emission (|2⟩ → |1⟩) 10

If the perturbation is an electric field E = ϵ0 cos(ωt) coupling to the dipole

moment µ, then V21 = −µ21 · ϵ0 . The probability becomes:

sin2 [(ω0 − ω)t/2]

2
µ21 ϵ0

P1→2 (t) = (8.2)
ℏ (ω0 − ω)2

This represents Absorption. The atom absorbs energy E2 − E1 = ℏω0 from

the electromagnetic field to jump to the higher state.

8.2 Case 2: Stimulated Emission (|2⟩ → |1⟩)

Now, assume the atom is initially in the higher energy state |2⟩ (so c2 (0) =
1, c1 (0) = 0). By repeating the exact same perturbative mathematical steps
(1)
to find c1 (t), we will find that the probability of transitioning down to the
lower state |1⟩ is identical:

sin2 [(ω0 − ω)t/2]

2
µ12 ϵ0

(1)
P2→1 (t) = |c1 (t)|2 = (8.3)
ℏ (ω0 − ω)2

This represents Stimulated Emission. By shining light of appropriate wave-

length on the atom, we stimulate it to release energy E1 − E2 = −ℏω0 into
the radiation field and drop to the lower state.

9 Application: The Physics of LASERs

The mathematical framework we just derived for a two-state system lies at the
very heart of how LASERS (acronym for Light Amplification by Stimulated
Emission of Radiation) operate, but it also reveals a fundamental limitation
that laser engineers must overcome.

9.1 The Equality of Transition Probabilities

Our time-dependent perturbation theory yielded an interesting result: the
probability of absorption exactly equals the probability of stimulated emis-
sion:
P1→2 (t) = P2→1 (t) (9.1)

Because the matrix elements are Hermitian conjugates (|V12 |2 = |V21 |2 ), an

incoming photon of frequency ω ≈ ω0 is just as likely to force an electron up
(absorbing the photon) as it is to force an electron down (creating a second,
identical photon). This is equivalent to saying the Einstein B coefficients for
absorption and stimulated emission are equal (B12 = B21 ).
Application: The Physics of LASERs The Necessity of Population Inversion 11

9.2 The Necessity of Population Inversion

For a beam of light to be amplified as it passes through a material, it must
trigger more stimulated emissions than absorptions. Let N1 be the number
of atoms in the ground state and N2 be the number of atoms in the excited
state. The net rate of photon production is proportional to:

Net Rate ∝ (N2 − N1 )P2→1 (t) (9.2)

Under normal thermal equilibrium, states follow the Boltzmann distribution,

meaning N1 ≫ N2 . To achieve light amplification, we must create a Popu-
lation Inversion, where N2 > N1 .

9.3 Why a Two-Level Laser is Impossible

Suppose we try to build a laser using our simple two-state system. We shine
an intense "pump" light on the atoms to excite them from |1⟩ to |2⟩ and build
up N2 .
Because P1→2 = P2→1 , as soon as the pump excites exactly half the
atoms (N1 = N2 ), the incoming pump photons will start causing stimulated
emission at the exact same rate they cause absorption. The system saturates.
A two-level system can never achieve N2 > N1 using optical pumping.

9.4 The Solution: 3-Level and 4-Level Systems

To bypass this quantum mechanical roadblock, real lasers utilize three or four
energy levels:

• 3-Level System (e.g., Ruby Laser): Atoms are pumped from the
ground state (E1 ) to a high-energy state (E3 ). They quickly undergo
a fast, non-radiative decay to a metastable intermediate state (E2 ).
Because the transition from E2 → E1 is relatively slow, atoms "pile up"
in E2 , achieving population inversion between E2 and E1 .

• 4-Level System (e.g., He-Ne Laser): This adds a fourth level above
the ground state. The lasing transition occurs between E3 and E2 .
Because E2 naturally decays very quickly to the ground state E1 , E2 is
essentially always empty (N2 ≈ 0). Thus, any population in E3 instantly
creates a population inversion (N3 > N2 ), making 4-level lasers highly
efficient.

By studying the 2-state system mathematically, we see exactly why higher-

level systems are required to engineer coherent light!
TDPT for an Arbitrary Dimensional Hilbert Space 12

10 Summary of the Perturbative Expansion

Before applying our theory to a specific potential, let us make a crucial ob-
servation about the structure of the Time-Dependent Perturbation Theory
(TDPT) expansion:

(0)
• Zeroth Order (ci ): Contains zero matrix elements of the perturba-
tion V̂ . The system remains in its initial state.

(1)
• First Order (ci ): Contains one matrix element of V̂ . This describes
direct, single-step transitions between states.

(2)
• Second Order (ci ): Contains two matrix elements of V̂ (a product of
integrations). This describes two-step processes, such as transitioning
to an intermediate state before reaching the final state.

This iterative method forms the bedrock of TDPT. In the next section,
we will generalize this logic from a 2-state system to an N-state system.

11 TDPT for an Arbitrary Dimensional Hilbert

Space
In the previous lecture, we built our intuition using a two-state system. We
now generalize this to an N -state system (where N can be infinite).
Let the total Hamiltonian be:

Ĥ(t) = Ĥ0 + λV̂ (t) (11.1)

where Ĥ0 is the time-independent unperturbed Hamiltonian, and λ is a dimen-

sionless parameter tracking the order of the perturbation. The unperturbed
eigenstates form a complete orthonormal basis:

Ĥ0 |n⟩ = En |n⟩ (11.2)

Just like in the two-state case, any arbitrary state vector |Ψ(t)⟩ can be
written as a linear superposition of the unperturbed eigenstates, with time-
dependent coefficients taking into account the phase evolution:

|Ψ(t)⟩ = cn (t)e−iEn t/ℏ |n⟩ (11.3)

X

n
The Dirac (Interaction) Representation The Coupled Differential Equations 13

11.1 The Coupled Differential Equations

To find the dynamics, we substitute this state into the time-dependent Schrödinger
equation:
∂
iℏ |Ψ(t)⟩ = [Ĥ0 + λV̂ (t)]|Ψ(t)⟩ (11.4)
∂t
Evaluating the time derivative using the product rule gives:
X iEn

ċn (t)e
−iEn t/ℏ
cn (t)e−iEn t/ℏ |n⟩ = cn (t)e−iEn t/ℏ [Ĥ0 +λV̂ (t)]|n⟩
X
iℏ |n⟩ −
n ℏ n
(11.5)
Since Ĥ0 |n⟩ = En |n⟩, the En terms on both sides perfectly cancel. We
are left with:

ċn (t)e−iEn t/ℏ |n⟩ = λ cn (t)e−iEn t/ℏ V̂ (t)|n⟩ (11.6)

X X
iℏ
n n

To isolate the differential equation for a specific state m, we take the

inner product with ⟨m|:

iℏċm (t)e−iEm t/ℏ = λ cn (t)e−iEn t/ℏ ⟨m|V̂ (t)|n⟩ (11.7)

X

Multiplying both sides by eiEm t/ℏ and defining the matrix element Vmn (t) ≡
⟨m|V̂ (t)|n⟩, we arrive at the exact system of coupled differential equations:

iλ X
ċm (t) = − Vmn (t)eiωmn t cn (t) (11.8)
ℏ n

where ωmn ≡ Em −En

ℏ
is the Bohr transition frequency.

12 The Dirac (Interaction) Representation

The appearance of the eiωmn t phase factors in the equation above is slightly
cumbersome. To simplify the mathematical structure of Time-Dependent
Perturbation Theory, it is extremely useful to change our "picture" of quantum
mechanics.
In quantum mechanics, expectation values (the actual observable physics)
must remain invariant, but we have the freedom to push the time dependence
into the state vectors, the operators, or a mixture of both.

12.1 Defining the Interaction Picture

In the Schrödinger Picture, the state vector |Ψ(t)⟩S carries all the time de-
pendence, while operators (like position and momentum) are generally static.
The Dirac (Interaction) Representation Evolution of the State Vector 14

To remove the trivial, fast-oscillating time dependence caused by Ĥ0 , we de-

fine a new state vector in the Interaction (Dirac) Picture:

|Ψ(t)⟩I ≡ eiĤ0 t/ℏ |Ψ(t)⟩S (12.1)

Notice what this does: if the perturbation V̂ (t) were zero, the Schrödinger
state would evolve simply as e−iĤ0 t/ℏ . The transformation eiĤ0 t/ℏ perfectly
"unwinds" this evolution. Therefore, in the Interaction picture, state vectors
only evolve if there is a perturbation.

At t = 0, all three (Schroedinger, Heisenberg and Dirac) pictures coincide:

|Ψ(0)⟩I = |Ψ(0)⟩S = |Ψ(0)⟩H (12.2)

12.2 Evolution of the State Vector

Let us find the equation of motion for |Ψ(t)⟩I . Taking the time derivative:

∂ ∂  iĤ0 t/ℏ 
iℏ |Ψ(t)⟩I = iℏ e |Ψ(t)⟩S (12.3)
∂t ∂t
!
∂ ∂
iℏ |Ψ(t)⟩I = −Ĥ0 eiĤ0 t/ℏ |Ψ(t)⟩S + eiĤ0 t/ℏ iℏ |Ψ(t)⟩S (12.4)
∂t ∂t
 
Substitute the Schrödinger equation iℏ ∂t
∂
|Ψ(t)⟩S = (Ĥ0 + V̂S )|Ψ(t)⟩S :

∂
iℏ |Ψ(t)⟩I = −Ĥ0 |Ψ(t)⟩I + eiĤ0 t/ℏ (Ĥ0 + V̂S )e−iĤ0 t/ℏ |Ψ(t)⟩I (12.5)
∂t

Since Ĥ0 commutes with itself, the Ĥ0 terms cancel perfectly, leaving:

∂
iℏ |Ψ(t)⟩I = V̂I (t)|Ψ(t)⟩I (12.6)
∂t

where the operator in the Interaction picture is defined as:

V̂I (t) ≡ eiĤ0 t/ℏ V̂S (t)e−iĤ0 t/ℏ (12.7)

This is a key result: The state vector in the Interaction picture

evolves according to a Schrödinger-like equation, but driven ONLY
by the perturbation V̂I , not the total Hamiltonian.
How the Interaction Picture results match with 2-State system results? 15

13 How the Interaction Picture results match

with 2-State system results?

To demystify the Interaction picture, let us explicitly apply this new formal
machinery to the 2-state system we solved in the previous lectures.
Recall that in the Schrödinger picture, we wrote the arbitrary state of
our 2-state system as a superposition that included the standard unperturbed
time evolution:

|Ψ(t)⟩S = c1 (t)e−iE1 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ |2⟩ (13.1)

Now, let us transform this state into the Dirac (Interaction) picture using
our new definition, |Ψ(t)⟩I = eiĤ0 t/ℏ |Ψ(t)⟩S :
h i
|Ψ(t)⟩I = eiĤ0 t/ℏ c1 (t)e−iE1 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ |2⟩
   
= c1 (t)e−iE1 t/ℏ eiĤ0 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ eiĤ0 t/ℏ |2⟩ (13.2)

Because |1⟩ and |2⟩ are exact eigenstates of Ĥ0 (with eigenvalues E1 and
E2 ), the operator exponential acts very simply:

eiĤ0 t/ℏ |1⟩ = eiE1 t/ℏ |1⟩ and eiĤ0 t/ℏ |2⟩ = eiE2 t/ℏ |2⟩ (13.3)

Substituting this back into our equation, the phase factors perfectly can-
cel!

|Ψ(t)⟩I = c1 (t)e−iE1 t/ℏ eiE1 t/ℏ |1⟩ + c2 (t)e−iE2 t/ℏ eiE2 t/ℏ |2⟩
|Ψ(t)⟩I = c1 (t)|1⟩ + c2 (t)|2⟩ (13.4)

Thus, the time-varying coefficients c1 (t) and c2 (t) we obtained in the case
of 2-state system are exactly the probability amplitudes of the state vector
in the Interaction picture.
Furthermore, the coupled differential equations we manually derived for
ċ1 (t) and ċ2 (t) by canceling out the Ĥ0 terms were just the component-by-
component expansion of the Interaction picture Schrödinger equation:

∂
iℏ |Ψ(t)⟩I = V̂I (t)|Ψ(t)⟩I (13.5)
∂t
Without naming it at the time, we have been using the Interaction picture
from the very beginning to solve the 2-state system!
The Time-Evolution Operator in the Interaction Picture 16

14 Summary of Quantum Representations

To provide a clear conceptual map, the table below contrasts how states
and operators handle time evolution in the three standard representations of
quantum mechanics.

Feature Schrödinger Heisenberg Dirac (In-

Picture Picture teraction)
Picture

State Vector |Ψ(t)⟩ Evolves with Static (Con- Evolves

time. stant in time). slowly due to
V̂ (t).

Operators Â(t) Static (un- Evolves Evolves

less explic- rapidly with rapidly due to
itly time- time due to unperturbed
dependent). total Ĥ. Ĥ0 .

Transformation (Base repre- |Ψ⟩H = |Ψ(t)⟩I =

sentation) eiĤt/ℏ
|Ψ(t)⟩S e iĤ0 t/ℏ
|Ψ(t)⟩S

State Evolution Eq. ∂

iℏ ∂t |Ψ⟩S = ∂
∂t
|Ψ⟩H =0 ∂
iℏ ∂t |Ψ⟩I =
Ĥ|Ψ⟩S V̂I (t)|Ψ⟩I

Operator Evolution Eq. dÂS

dt
= ∂ ÂS
∂t
dÂH
dt
= dÂI
dt
=
i
ℏ
[Ĥ, ÂH ] + i
ℏ
[Ĥ0 , ÂI ] +
∂ ÂH ∂ ÂI
∂t ∂t

By using the Interaction picture, the expansion coefficients cn (t) we de-

rived earlier are exactly the components of the Interaction state vector: |Ψ(t)⟩I =
n cn (t)|n⟩. The interaction picture mathematically formalizes the intuitive
P

approach we used to solve the two-state system!

15 The Time-Evolution Operator in the Interac-

tion Picture
In the previous lecture, we showed that the state vector in the Interaction
picture evolves according to a Schrödinger-like equation driven entirely by
Iterative Solution: The Dyson Series 17

the interaction Hamiltonian V̂I (t):

∂
iℏ |Ψ(t)⟩I = V̂I (t)|Ψ(t)⟩I (15.1)
∂t
Instead of solving for the state vector directly, it is highly advantageous to
define a Time-Evolution Operator ÛI (t) that propagates the initial state
forward in time:
|Ψ(t)⟩I = ÛI (t)|Ψ(0)⟩I (15.2)

Because the state at t = 0 is just itself, the initial condition must be the
identity operator:
ÛI (0) = Iˆ (15.3)

Substituting our definition of ÛI (t) into the differential equation yields the
operator equation:
d
iℏ ÛI (t) = V̂I (t)ÛI (t) (15.4)
dt
Integrating this with respect to time from 0 to t, and applying the initial
condition, we can rewrite the differential equation as an equivalent integral
equation:
ˆ iZt
ÛI (t) = I − V̂I (t′ )ÛI (t′ )dt′ (15.5)
ℏ 0

16 Iterative Solution: The Dyson Series

Equation (5) is exact, but ÛI appears on both sides. However, this form is
perfectly suited for an iterative approximation. By repeatedly substituting
the equation into itself, we generate a power series in the perturbation V̂I ,
known as the Dyson Series.

• Zeroth Order: If the perturbation is zero, the system doesn’t evolve

in the interaction picture.

ÛI (t) = Iˆ
(0)
(16.1)

• First Order: Substitute the 0th-order solution into the integral.

(1) iZt
ÛI (t) =− V̂I (t′ )dt′ (16.2)
ℏ 0

• Second Order: Substitute the 1st-order solution into the integral.

Note the dummy variables for time: the outer integral is at t′ , so the
Physical Interpretation: Virtual Transitions and Uncertainty 18

inner integral must go up to t′ , using t′′ .

2 Z t Z t′
i

(2)
ÛI (t) = − dt′ dt′′ V̂I (t′ )V̂I (t′′ ) (16.3)
ℏ 0 0

17 Physical Interpretation: Virtual Transitions and

Uncertainty
Let us pause the mathematics to understand the important physical meaning
of the second-order term. Suppose the system starts in state |i⟩ at t = 0 and
ends in state |f ⟩ at time t. To find the second-order transition amplitude,
(2)
we take the matrix element of ÛI and insert a complete set of intermediate
states m |m⟩⟨m| = Iˆ between the two potential operators:
P

XZ t Z t′
(2)
⟨f |ÛI (t)|i⟩ ∝ ′
dt dt′′ ⟨f |V̂I (t′ )|m⟩⟨m|V̂I (t′′ )|i⟩ (17.1)
m 0 0

From the above equation we see that:

1. At time t′′ , the perturbation causes the system to jump from the initial
state |i⟩ into any state |m⟩. This state need not be the initial or final
state.

2. Once it transitions to this state |m⟩, the system remains in this state!

3. At a later time t′ (t′ > t′′ ), the perturbation acts again, causing the
system to jump from |m⟩ to the final state |f ⟩.

17.1 Connection to the Heisenberg Uncertainty Principle

The sum over m includes all possible eigenstates of Ĥ0 , even those that heavily
violate energy conservation (Em ̸= Ei ). How is this physically possible?
These intermediate states are called Virtual States, and the transitions
are Virtual Transitions. Because the system only occupies state |m⟩ for a
very brief duration ∆t = t′ − t′′ , the energy of the system during this time is
fundamentally uncertain. According to the Heisenberg Uncertainty Principle:

ℏ
∆E∆t ≳ (17.2)
2
If the time spent in the intermediate state is extremely short (∆t → 0), the
energy uncertainty ∆E can be massive. Thus, quantum mechanics allows the
system to temporarily "borrow" energy to visit any virtual state |m⟩, provided
it "pays it back" by jumping to the final state |f ⟩ before the time limit expires.
The Time-Ordering Operator (T ) 19

18 The Time-Ordering Operator (T )

Look closely at the integration limits for the second-order term:
Z t Z t′
dt′
dt′′ =⇒ 0 ≤ t′′ ≤ t′ ≤ t (18.1)
0 0

The operators are rigorously ordered such that the earlier time (t′′ ) always acts
first (on the right). Because V̂I (t′ ) and V̂I (t′′ ) generally do not commute at
different times, we cannot casually swap their order or change the integration
limits.

18.1 Visualizing the Integration Domain

The integration limits restrict us to the lower triangle of a t × t square in the
t′ − t′′ plane.

t′′ (Earlier time)

t
Unphysical Region (t′′ > t′ )
(Effects precede causes)
t ′′
=
t′

Physical Region
0 ≤ t′′ ≤ t′ ≤ t
Area = 12 t2
t′ (Later time)
t

To make the math symmetric and easier to compute, we introduce the

Dyson Time-Ordering Operator T . This operator automatically sorts
any product of time-dependent operators so that the earliest times are placed
to the right: 
V̂ (t1 )V̂ (t2 )

if t1 > t2
T [V̂ (t1 )V̂ (t2 )] = (18.2)
V̂ (t2 )V̂ (t1 )

if t2 > t1
By placing T inside the integral, the integrand becomes perfectly sym-
metric across the diagonal line t′ = t′′ . We can now integrate over the entire
square (both triangles) and simply divide by 2 (or 2!):
Z t Z t′
1 Z t ′ Z t ′′ h i
dt′ dt′′ V̂I (t′ )V̂I (t′′ ) = dt dt T V̂I (t′ )V̂I (t′′ ) (18.3)
0 0 2! 0 0
Defining the Transition Probability The Full Dyson Series 20

18.2 The Full Dyson Series

Generalizing this logic to the n-th order (which involves an n-dimensional
hyper-triangle of volume 1/n!), we arrive at the elegant, compact formulation
of the full Dyson series:

∞ 
−i n 1 Zt Z t
 h i
ÛI (t) = dt1 · · · dtn T V̂I (t1 ) . . . V̂I (tn ) (18.4)
X

n=0 ℏ n! 0 0

This series is the foundation for all Feynman diagram calculations in Quantum
Field Theory!

19 Defining the Transition Probability

The formalism we developed using the time-evolution operator allows us to
rigorously calculate transition probabilities. Suppose our quantum system
starts exactly in an unperturbed eigenstate |i⟩ at time t = 0. We turn on a
time-dependent perturbation V̂ (t). What is the probability that the system
will be found in a different eigenstate |f ⟩ at a later time t?
Let the unperturbed states be defined by Ĥ0 |i⟩ = Ei |i⟩ and Ĥ0 |f ⟩ =
Ef |f ⟩. In the Schrödinger picture, the probability of transitioning from |i⟩ →
|f ⟩ is given by the squared magnitude of the amplitude:

Pi→f (t) = |⟨f |ΨS (t)⟩|2 (19.1)

In the Schrödinger picture, the probability amplitude to find the system in

the final state |f ⟩ at time t is:

cf (t) = ⟨f |ΨS (t)⟩ (19.2)

Recall the fundamental transformation connecting the Schrödinger state to

the Interaction state:
|ΨS (t)⟩ = e−iĤ0 t/ℏ |ΨI (t)⟩ (19.3)

Substituting this transformation into our amplitude equation gives:

cf (t) = ⟨f |e−iĤ0 t/ℏ |ΨI (t)⟩ (19.4)

Because the target final state |f ⟩ is an exact eigenstate of the unperturbed

Hamiltonian (Ĥ0 |f ⟩ = Ef |f ⟩), the operator acting to the left simply yields
the time-evolution phase factor corresponding to the final energy:

cf (t) = e−iEf t/ℏ ⟨f |ΨI (t)⟩ (19.5)

Application: A Constant Perturbation Expansion via the Dyson Series 21

To find the observable physical probability, we take the absolute square

of the amplitude. The overall phase factor completely disappears because
2
e−iEf t/ℏ = 1, leaving:

Pi→f (t) = |cf (t)|2 = |⟨f |ΨI (t)⟩|2 (19.6)

This confirms that the probability of transitioning between unperturbed

eigenstates is identical in both pictures. Finally, because the system starts
entirely in state |i⟩ at t = 0, we can express the time-evolved state using
the Interaction picture time-evolution operator as |ΨI (t)⟩ = ÛI (t)|i⟩. This
yields our foundational working equation for computing probabilities: Be-
cause |f ⟩ is an exact eigenstate of Ĥ0 , the transition probability calculated in
the Interaction picture is identical to the Schrödinger picture:

Pi→f (t) = |⟨f |ÛI (t)|i⟩|2 (19.7)

19.1 Expansion via the Dyson Series

We can substitute the Dyson series for ÛI (t) into our probability expression.
Using λ as a bookkeeping parameter for the perturbation order:

(1) (2) 2
Pi→f (t) = δf i + λUf i (t) + λ2 Uf i (t) + . . . (19.8)

For i ̸= f (a transition to a different state), the Kronecker delta δf i vanishes.

The leading-order behavior is strictly determined by the first-order term:

(1) 2 1 Zt 2
Pi→f (t) ≈ λ2 Uf i (t) = ⟨f |V̂I (t′ )|i⟩dt′ (19.9)
ℏ2 0

20 Application: A Constant Perturbation

Let us apply this to the simplest possible scenario: a constant perturbation
that is abruptly turned on at t = 0.

0

t<0
V̂ (t) = (20.1)
V̂

t≥0

where the spatial matrix element Vf i = ⟨f |V̂ |i⟩ is time-independent. The

first-order amplitude is:

(1) −i Z t i(Ef −Ei )t′ /ℏ −iVf i Z t iωf i t′ ′

Uf i (t) = e Vf i dt′ = e dt (20.2)
ℏ 0 ℏ 0
Application: A Constant Perturbation Physical Features of the Probability Curve 22

where ωf i = (Ef − Ei )/ℏ is the transition frequency. Performing the integra-

tion:
−Vf i  iωf i t −2iVf i iωf i t/2 ωf i t
  
(1)
Uf i (t) = e −1 = e sin (20.3)
ℏωf i ℏωf i 2

Squaring this amplitude yields the transition probability:

4|Vf i |2 sin2 ∆
!
ωf i t Vf i |2 t2
 
Pi→f (t) = 2 2 sin2 = (20.4)
ℏ ωf i 2 ℏ2 ∆2

where ∆ = ωf i t/2.

Probability Pi→f

As time t → ∞:
|Vf i |2 t2
h̄2 1. Peak height ∝ t2
2. Peak width ∝ 1/t
Function converges to
the Dirac delta δ(ωf i )

1
Width ∝ t

ωf i
− 6πt − 4πt − 2πt 2π
t
4π
t
6π
t

20.1 Physical Features of the Probability Curve

Let ∆E = Ef − Ei = ℏωf i . The probability shapes itself as a sinc-squared

function peaking at ∆E = 0.

• Peak Height and Width: As time t increases, the height of the

central peak grows proportionally to t2 , while its width shrinks propor-
tionally to 1/t.

• Energy-Time Uncertainty: Significant transitions occur only for fi-

nal states under the central peak. This implies that energy conser-
vation can be violated, but only by an amount |∆E| ≤ 2πℏ/t. This
perfectly aligns with the Heisenberg Energy-Time uncertainty principle
(∆E∆t ∼ ℏ); the shorter the time the perturbation has acted, the wider
the spread in allowable final energies.
The Physical Meaning of the Golden Rule 23

21 Transition Rates and Fermi’s Golden Rule

For times long enough that the perturbation has fully established itself, it is
more physically meaningful to define a Transition Rate Wi→f , which is the
probability of transition per unit time:

d Pi→f (t)
Wi→f = lim Pi→f (t) or equivalently, Wi→f = lim (21.1)
t→∞ dt t→∞ t
To evaluate this limit, we use the standard mathematical representation
of the Dirac delta function:

sin2 (αt)
lim = δ(α) (21.2)
t→∞ πtα2

Applying this to our probability expression where α = ∆E/2ℏ:

1 4|Vf i |2 ∆Et 2π
" !#
lim sin2 = |Vf i |2 δ(∆E) (21.3)
t→∞ t ∆E 2 2ℏ ℏ

This famous result is known as Fermi’s Golden Rule No. 21

2π
Wi→f = |Vf i |2 δ(Ef − Ei ) (21.4)
ℏ

22 The Physical Meaning of the Golden Rule

22.1 The Apparent Triviality of the Delta Function
At first glance, Fermi’s Golden Rule looks mathematically problematic or
"trivial." If the energy levels are strictly discrete (e.g., transitions between
bound states of an atom), this probability simply oscillates back and forth
with a recurrence frequency ω = |Ef − Ei |/ℏ. However, in many critical
physical situations, the system behaves very differently.
If we are considering a transition between two discrete, isolated energy
levels, the formula implies:

• If Ef ̸= Ei , the rate is strictly zero.

• If Ef = Ei , the rate is infinite.

Neither of these outcomes makes practical sense for calculating a mea-

surable, finite transition rate. How do we extract real physics from this?
1
The numbering is a historical quirk originating from Enrico Fermi’s 1950 textbook on
Nuclear Physics. Fermi’s Golden Rule No. 2 is the transition rate calculated to the first
order of perturbation theory. Fermi’s Golden Rule No. 1 is the transition rate calculated
to the second order (and higher) of perturbation theory.
Discrete to Continuum transition probability The Resolution: Transitions to a Continuum 24

22.2 The Resolution: Transitions to a Continuum

The resolution lies in understanding that Fermi’s Golden Rule is essen-
tially never used for transitions between two discrete, isolated states
(for that, we use the Rabi oscillation formalism we studied earlier).
Physically, we often encounter situations where the final state is not a
single, isolated discrete level, but rather part of a continuous band of states.
A classic example is photoionization: a perturbation knocks an electron out
of a bound state (Ei < 0) and into a free-particle state (Ef > 0). Because
a free particle can have any continuous value of kinetic energy, there are
infinitely many possible final states available. Other examples include, an
atom emitting a photon (the photon can occupy a continuum of momentum
states in free space) and a particle scattering off a potential step.
When the final states form a continuum, the probability of transitioning
to one exact, infinitely precise energy level is practically zero. Instead, we
must calculate the probability of transitioning into a range or band of final
states centered around Ef ≈ Ei .

23 Discrete to Continuum transition probability

When the final states form a continuum, we cannot ask for the probability
to transition to one exact, infinitely precise state. We must ask for the total
transition rate Γ into a band of available final states.
To perform this sum over the continuum, we introduce the Density of
States, ρ(E). This function represents the number of available quantum
states per unit energy interval. The number of states dN in a small energy
window between E and E + dE is:

dN = ρ(E)dE (23.1)

23.1 Integrating the Probability

To find the total probability of transitioning into this continuous band, we
must integrate the single-state probability over all available final energies:
Z Z
Ptotal (t) = Pi→f (t)dN = Pi→f (t)ρ(Ef )dEf (23.2)

Substituting our expression for the transition probability:

4|Vf i |2 2 (Ef − Ei )t
Z !
Ptotal (t) = sin ρ(Ef )dEf (23.3)
(Ef − Ei )2 2ℏ
Discrete to Continuum transition probability Fermi’s Golden Rule 25

2
The function inside the integral ( sinx(xt)
2 ) is highly peaked at x = 0 (where
Ef = Ei ) and falls off rapidly as we move away from resonance. If the time t is
sufficiently large, this peak becomes extremely narrow. Because the integral
is completely dominated by the region strictly around Ef ≈ Ei , we can make
two standard approximations:

1. The matrix element |Vf i |2 and the density of states ρ(Ef ) are roughly
constant over this tiny energy window, so we can pull them out of the
integral, evaluating them at Ef = Ei .

2. We can extend the limits of integration from −∞ to +∞ without accu-

mulating any significant extra area, because the tails of the sinc function
contribute almost nothing.

Pulling the constants out, we have:

Z ∞
sin2 [(Ef − Ei )t/2ℏ]
Ptotal (t) ≈ 4|Vf i |2 ρ(Ei ) dEf (23.4)
−∞ (Ef − Ei )2

(Ef −Ei )t
We perform a change of variables. Let x = 2ℏ
, which means dEf =
2ℏ
t
dx. The integral becomes:

2ℏ Z ∞ sin2 (x)
!
t
 
Ptotal (t) = 4|Vf i | ρ(Ei )
2
dx · t (23.5)
4ℏ2 t −∞ x2
R ∞ sin2 (x)
Using the standard definite integral −∞ x2
dx = π, the final result is:

2π
Ptotal (t) = |Vf i |2 ρ(Ei )t (23.6)
ℏ

23.2 Fermi’s Golden Rule

Notice the interesting shift in the physics! For a single discrete state, the
probability oscillated. But when transitioning into a continuum, the total
probability grows linearly with time.
Because the probability grows at a constant rate, it is highly useful to
define the Transition Rate Γ (the probability of transition per unit time):

dPtotal
Γi→f = (23.7)
dt
Taking the time derivative of our linear result yields the most famous
equation in Time-Dependent Perturbation Theory, known as Fermi’s Golden
Rule:
2π
Γi→f = |Vf i |2 ρ(Ei ) (23.8)
ℏ
Discrete to Continuum transition probability Fermi’s Golden Rule 26

This rule states that the transition rate is driven by two independent factors:
the strength of the quantum coupling (the matrix element |Vf i |2 ) and the
sheer number of final states available for the system to jump into (the density
of states ρ(Ei )).

The Limits of TDPT

A crucial conceptual question arises when we consider Time-Dependent Per-
turbation Theory over very long timescales: If energy is not conserved by a
time-dependent Hamiltonian, what happens if we leave a "small" perturbation
on for a very long time T ?
In principle, the total energy pumped into (or extracted from) the system
by the perturbation could eventually exceed the energy scale of the unper-
turbed Hamiltonian Ĥ0 . When this happens, the fundamental properties of
the system are completely altered, and Ĥ0 can no longer be viewed as the
dominant governing physics.

The Analogy
To build an intuition for this, consider the geological process of ocean waves
crashing against a massive coastal cliff.

• The Unperturbed System (Ĥ0 ): The solid rock of the cliff.

• The Perturbation (V̂ (t)): The gentle, rhythmic mechanical force of

the ocean waves.

On a timescale of days or years, the waves act as a tiny perturbation. The

cliff remains fundamentally unchanged, and we can easily predict the water’s
behavior using perturbation theory. However, if we integrate this tiny force
over a timescale of millions of years (T → ∞), the ocean completely erodes
the rock. The "small" perturbation has dominated the system, and treating
the rock as the unchanging background (Ĥ0 ) is physically catastrophic.

Mathematical Breakdown: The Violation of Unitarity

We can see this breakdown explicitly in the mathematics of our first-order
transition probability. Recall our result for a constant perturbation applied
at resonance (ωf i ≈ 0 or ∆E ≈ 0):

|Vf i |2 2
Pi→f (t) = t (23.9)
ℏ2
Second-Order Transitions (Fermi’s Golden Rule No. 1) 27

Because the probability grows quadratically with time, there will in-
evitably be a critical time tc at which the calculated probability exceeds 100%:

ℏ
Pi→f (tc ) > 1 for tc > (23.10)
|Vf i |

This is a gross violation of unitarity (the conservation of total probability).

Why does this happen? In first-order TDPT, we assumed that the ampli-
tude of the initial state remains approximately constant (ci (t) ≈ 1). However,
as transitions occur at a constant rate Γ (via Fermi’s Golden Rule), the initial
state is continuously depleted. By assuming ci (t) = 1 for all time, our math
essentially pretends the initial state has an infinite supply of probability and
energy to pump into the final states.

The Window of Validity

For Time-Dependent Perturbation Theory and Fermi’s Golden Rule to be
mathematically consistent and physically meaningful, the time t during which
the perturbation acts must satisfy a strict Goldilocks condition:

1 ℏ
≪t≪ (23.11)
ωf i |Vf i |

1. The Lower Bound (t ≫ 1/ωf i ): The time must be long enough for the
perturbation to complete several oscillations, allowing the sinc2 function
to narrow into a Dirac delta function and enforce energy conservation.

2. The Upper Bound (t ≪ ℏ/|Vf i |): The time must be short enough
that the total probability of transitioning out of the initial state remains
strictly much less than 1 ( Pi→f ≪ 1). The initial state must not be
P

significantly depleted.

If we wish to study the system beyond this upper time limit, simple
perturbation theory fails. We must transition to more exact methods, such
as solving the fully coupled differential equations (like Rabi flopping, which
naturally handles the depletion and repopulation of states) or using the non-
perturbative Dyson series to infinite order.

24 Second-Order Transitions (Fermi’s Golden Rule

No. 1)
What happens if the first-order matrix element is strictly zero (Vf i = 0) due
to a selection rule? Does the transition rate vanish completely?
Application of Second-Order Transitions: Raman Scattering 28

To answer this, we must look at the second-order term of the Dyson

series, which describes a two-step transition passing through an intermediate
"virtual" state |m⟩.
2 X Z t Z t′
−i

(2)
Uf i (t) = ′
dt dt′′ ⟨f |V̂I (t′ )|m⟩⟨m|V̂I (t′′ )|i⟩ (24.1)
ℏ m 0 0

Substituting the time-dependence of the interaction picture operators

V̂I (t) = eiĤ0 t/ℏ V̂S e−iĤ0 t/ℏ :
2 X Z t′
−i Z t

(2) ′ i(Ef −Em )t′ /ℏ ′′ /ℏ
Uf i (t) = Vf m Vmi dt e dt′′ ei(Em −Ei )t (24.2)
ℏ m 0 0

Performing the inner integration over t′′ :

Z t′
′′ /ℏ ℏ  ′

ei(Em −Ei )t dt′′ = ei(Em −Ei )t /ℏ − 1 (24.3)
0 i(Em − Ei )

Substituting this back into the outer integral over t′ and simplifying reveals
an effective, combined matrix element:

Vf m Vmi
Veff = (24.4)
X

m Ei − Em

If we square this second-order amplitude and apply the exact same long-
time continuum limit we used in Section 2, we arrive at the second-order
transition rate, historically called Fermi’s Golden Rule No. 1:

2
(2) 2π X Vf m Vmi
Γi→f = ρ(Ef ) (24.5)
ℏ m Ei − Em

This allows transitions that are "forbidden" in the first order (like two-photon
absorption or Raman scattering) to occur by temporarily borrowing energy
to pass through virtual state |m⟩.

25 Application of Second-Order Transitions: Ra-

man Scattering
To understand the physical necessity of the second-order transition formula
(Fermi’s Golden Rule No. 1), we look at one of the most powerful analytical
techniques in modern physics and chemistry: Raman Scattering.
When monochromatic light (like a laser) shines on a material, the vast ma-
jority of the scattered photons undergo elastic scattering, known as Rayleigh
 Why Scattering
Application of Second-Order Transitions: Raman is Raman Scattering a Second-Order Process? 29

scattering. The scattered photons have the exact same energy (and frequency)
as the incident photons.
Sample molecules Virtual State
Raman scattered light
Incident laser E > E0
Rayleigh scattered light
E = E0 h̄(ω0 − Ω) h̄ω0
E0
Raman scattered light h̄ω0 h̄ω0 h̄ω0 h̄(ω0 + Ω)
E < E0
Excited State (v = 1)

Ground State (v = 0)

Rayleigh Stokes Anti-Stokes

Scattering Raman Raman

However, a tiny fraction of the photons (about 1 in 10 million) undergo

inelastic scattering. The scattered photon emerges with a different energy
than the incident photon. The difference in energy is absorbed by (or ex-
tracted from) the vibrational or rotational energy levels of the molecule. This
is the Raman effect.

25.1 Why is Raman Scattering a Second-Order Process?

Raman scattering fundamentally involves two photons: one incident photon is
absorbed, and one scattered photon is emitted. Because the electromagnetic
perturbation V̂ (t) must act twice to change the state of two different photons,
a first-order transition amplitude is exactly zero.
We must use second-order Time-Dependent Perturbation Theory. The
system transitions from an initial state |i⟩ to a final state |f ⟩ via an interme-
diate, highly unstable virtual state |m⟩.

25.2 The Three Scattering Regimes

Let the molecule initially be in state E1 and transition to state E2 . Let the
incident photon have frequency ωi and the scattered photon have frequency
ωs . Energy conservation between the initial and final states requires:

E1 + ℏωi = E2 + ℏωs (25.1)

Depending on the final state of the molecule, three things can happen:
• Rayleigh Scattering (E1 = E2 ): The molecule returns to its exact
initial state. The photon energy is unchanged (ωs = ωi ).

• Stokes Raman Scattering (E2 > E1 ): The molecule absorbs some

energy, transitioning to a higher vibrational state. The scattered photon
loses energy and is red-shifted (ωs < ωi ).

• Anti-Stokes Raman Scattering (E2 < E1 ): The molecule starts in

an excited state and drops to a lower state, giving its energy to the
photon. The scattered photon gains energy and is blue-shifted (ωs >
ωi ).
 Connecting to Fermi’s Golden Rule No. 1
Application of Second-Order Transitions: Raman Scattering 30

25.3 Connecting to Fermi’s Golden Rule No. 1

The rate at which these Raman transitions occur is governed exactly by the
effective second-order matrix element we derived:

⟨f |V̂ |m⟩⟨m|V̂ |i⟩

Veff = (25.2)
X

m Ei − Em

Here, Ei is the total initial energy (molecule + incident photon). The sum over
m represents a sum over all possible virtual electronic states of the molecule.
The denominator (Ei − Em ) shows that if the incident laser frequency per-
fectly matches a real electronic transition of the molecule, the denominator
approaches zero, and the scattering rate blows up. This is a special, highly
efficient case known as Resonance Raman Scattering.

25.4 Modern Applications

Because every specific chemical bond (C-C, C=O, O-H, etc.) has unique,
quantized vibrational energy levels, the energy shift of the scattered Raman
photons (∆E = ℏωi − ℏωs ) acts as a highly specific optical "fingerprint" for
the material.

• Materials Science: Raman spectroscopy is heavily used to character-

ize 2D materials like graphene, measuring the number of layers and the
presence of defects.

• Chemistry and Biology: It allows for non-destructive, label-free

identification of chemical compounds, proteins, and even the detection
of explosives or narcotics at security checkpoints.

• Art and Archaeology: Because it requires no physical sample extrac-

tion (just bouncing light off the surface), Raman scattering is used to
identify pigments in ancient paintings and historical artifacts.
 PART

Wentzel, Kramers, Brillouin (WKB) II

Approximation
Reference
Section ??:
Landau & Lifshitz Vol 3, Chapter 7
Table 2. Contents for Part
II
26 Introduction: WKB vs. Perturbation Theory
As we approach the final topics of the course, we introduce a powerful new
tool: the WKB (Wentzel, Kramers, Brillouin) Approximation.
Earlier in the course, we developed Time-Independent Perturbation The-
ory (TIPT) to solve problems where the potential could be split as Ĥ =
Ĥ0 + λV̂ . TIPT is highly effective when the perturbing potential V̂ is "small"
in magnitude compared to Ĥ0 .
But what if the potential is very strong, but varies very slowly in space?
TIPT would fail entirely. The WKB approximation (also known as the semi-
classical approximation) is designed exactly for this regime. It is applicable
when the de Broglie wavelength of the particle (λ = h/p) is much smaller
than the characteristic distance over which the potential changes. In this
limit, the system behaves almost classically.

26.1 WKB vs. The Variational Method

We have already contrasted WKB with Perturbation Theory. But how does
WKB compare to the other major approximate technique we learned: the
Variational Method?
Recall that the Variational Method is based on the Variational Principle
(Etrial ≥ Eground ). It requires us to guess a parameterized trial wave function
and minimize the energy expectation value.
While both WKB and the Variational Method are used when the potential
Ĥ cannot be solved exactly and Perturbation Theory fails (because there is
no "small" parameter λ), they are designed for completely opposite physical
regimes:

1. Target States (Ground vs. Excited):

31
Introduction: WKB vs. Perturbation Theory WKB vs. The Variational Method 32

• The Variational Method is a mathematical tool for finding the

ground state energy (and perhaps the first excited state). How-
ever, it becomes incredibly cumbersome for highly excited states
because the trial wave function must be forced to be mathemati-
cally orthogonal to all lower states.
• The WKB Approximation shines exactly where the Variational
method fails: highly excited states. Because WKB is a semiclassical
approximation, it works best when the particle has high energy
(large quantum number n). High energy means a large momentum
p, which means a very short de Broglie wavelength (λ = h/p). The
shorter the wavelength, the better the "slowly varying potential"
assumption holds! WKB generally gives poor results for the ground
state.

2. The Core Philosophy:

• The Variational Method relies on your physical intuition to

guess the global shape of the wave function. It doesn’t care if
the potential varies rapidly or slowly, as long as you can calculate
the integral ⟨ψ|Ĥ|ψ⟩.
• The WKB Approximation does not require you to guess the
wave function. Instead, it systematically builds the wave function
based on the local classical momentum p(x). It only requires that
the potential is smooth.

Summary of Approximation Techniques:

Method Condition for Va- Best Used For Primary Output

lidity

Perturbation Theory Ĥ = Ĥ0 +λV̂ where V̂ Small deviations from ex- Corrections to En and
is very small. actly solvable systems. |n⟩ for all states.

Variational Method None (but requires a Complex potentials, Upper bound on the
good physical guess multi-electron atoms, ground state energy.
for ψ). molecules.

WKB Approximation λ ≪ VV ′(x)

(x)
(Potential Highly excited states Asymptotic wave
varies slowly). (large n) and quantum functions and tunnel-
tunneling. ing rates.
The WKB Ansatz for a Slowly Varying Potential 33

27 Review: The Constant Potential

To motivate the WKB ansatz, let us recall the exact solutions to the 1D
Time-Independent Schrödinger Equation (TISE) for a particle in a constant
potential V (x) = V0 :

ℏ2 d2 Ψ d2 Ψ 2m(E − V0 )
− + V0 Ψ = EΨ =⇒ + Ψ=0 (27.1)
2m dx2 dx2 ℏ2

Let p20 = 2m(E − V0 ). Depending on the energy of the particle relative

to the potential, we have three cases:

1. Case 1: E > V0 (Classically Allowed Region)

Here, p20 > 0. The particle has non-zero kinetic energy and moves freely.
The solutions are oscillatory plane waves:

Ψ(x) = C1 eip0 x/ℏ + C2 e−ip0 x/ℏ (27.2)

2. Case 2: E < V0 (Classically Forbidden Region)

Here, p20 < 0. The kinetic energy is negative, which is classically impos-
sible. Quantum mechanically, the wave function exhibits exponential
decay/growth (evanescent waves):

Ψ(x) = C1 eκ0 x/ℏ + C2 e−κ0 x/ℏ (27.3)

q
where κ0 = 2m(V0 − E).

3. Case 3: E = V0 (Classical Turning Point)

Here, p20 = 0. The solution is simply a straight line:

Ψ(x) = C1 x + C2 (27.4)

28 The WKB Ansatz for a Slowly Varying Po-

tential
Now, suppose V (x) is not strictly constant, but varies very slowly. We ex-
pect the wave function to lookqlocally like the constant-potential solutions,
but with a momentum p(x) = 2m(E − V (x)) that changes gradually with
position.
Instead of a simple phase (p0 x/ℏ), we postulate a generalized phase func-
tion S(x):
Ψ(x) = AeiS(x)/ℏ (28.1)
The WKB Ansatz for a Slowly Varying Potential Expansion in Powers of ℏ 34

V (x)

V (x)

⃗v
E
Forbidden Classically Allowed Forbidden
E < V (x) E > V (x) E < V (x)
Decaying Ψ Oscillatory Ψ Decaying Ψ
x
x1 x2

Figure 1. A global view of a particle in a potential well. The WKB approximation

uses oscillatory wavefunctions in the classically allowed region and exponentially
decaying wavefunctions in the forbidden regions.

We substitute this ansatz into the general TISE:

d2 Ψ p2 (x)
+ 2 Ψ=0 (28.2)
dx2 ℏ
Taking the derivatives:

i ′′ (S ′ )2
!
i
Ψ = S ′ AeiS/ℏ
′
and Ψ = ′′
S − 2 AeiS/ℏ (28.3)
ℏ ℏ ℏ

Substituting Ψ′′ back into the TISE and dividing out the exponential gives
the fundamental non-linear WKB equation:
!
ℏ
(S ) +
′ 2
S ′′ = p2 (x) (28.4)
i

where S ′ ≡ dS/dx.

28.1 Expansion in Powers of ℏ

Because we are working in the semiclassical limit where ℏ is considered "small"
(or more accurately, where the wavelength is small compared to the variation
of the potential), we expand the generalized phase S(x) as a power series in
ℏ:  !2 
(x) (x)
!
ℏ S1 ℏ S2
S(x) = S0 (x) 1 + + + ... (28.5)
i S0 (x) i S0 (x)
The above series expansion leads to:

S ′ (x) = S0′ − iℏS1′ − ℏ2 S2′ + . . . (28.6)

S (x) =
′′
S0′′ − iℏS1′′ −ℏ2
S2′′ + ... (28.7)
The WKB Ansatz for a Slowly Varying Potential Expansion in Powers of ℏ 35

The non-linear term (S ′ )2 term (up to first order in ℏ) leads to:

(S ′ )2 = (S0′ − iℏS1′ + . . . )(S0′ − iℏS1′ + . . . )

ℏ
= (S0′ )2 + 2 S0′ S1′ + O(ℏ2 ) (28.8)
i
Next, we evaluate the −iℏS ′′ term:

−iℏS ′′ = −iℏ(S0′′ − iℏS1′′ + . . . )

!2
ℏ ℏ
= S0′′ + S1′′ + . . . (28.9)
i i

We now substitute these expanded pieces back into the full non-linear
Schrödinger equation: (S ′ )2 − iℏS ′′ = p2 (x).
   
ℏ ℏ
(S0′ )2 + 2 S0′ S1′ + S0′′ + O(ℏ2 ) = p2 (x) (28.10)
i i

We can now cleanly group the terms by their powers of ℏ:

 
(S0′ )2 − iℏ 2S0′ S1′ + S0′′ +O(ℏ2 ) = p2 (x) (28.11)
| {z }
Order ℏ0
| {z }
Order ℏ1

For this equation to hold true for any arbitrary, small value of the ex-
pansion parameter ℏ, the coefficients of identical powers of ℏ on the left and
right sides of the equation must be strictly equal. The right-hand side, p2 (x),
has no ℏ dependence, meaning it belongs entirely to the ℏ0 order.
Matching the ℏ0 terms on both sides of Eq. (28.11) gives the classical
Hamilton-Jacobi equation:

(S0′ )2 = p2 (x) =⇒ S0′ (x) = ±p(x) (28.12)

Integrating this gives the purely classical accumulated phase:

Z
S0 (x) = ± p(x)dx (28.13)

If we stopped Rhere (the strictly ℏ → 0 limit), our wave function would be

i
Ψ(x) ≈ Ae± ℏ p(x)dx . This gives the correct oscillatory behavior but com-
pletely ignores how the probability amplitude must change as the particle
speeds up or slows down.
Let us now match the ℏ1 terms in Eq. (28.11). Since there are no ℏ1
terms on the right-hand side, we set the coefficient to zero:

2S0′ S1′ + S0′′ = 0 (28.14)

The WKB Ansatz for a Slowly Varying Potential Expansion in Powers of ℏ 36

This is a first-order differential equation for S1′ . We can easily solve it by

rearranging:
S ′′ 1 d
S1′ = − 0′ = − ln(S0′ ) (28.15)
2S0 2 dx
Since we already know from the zeroth-order that S0′ = ±p(x), we can sub-
stitute this in:
1 d
S1′ = − ln(p(x)) (28.16)
2 dx
Integrating both sides gives:
 
1 1 
S1 (x) = − ln(p(x)) = ln  q (28.17)
2 p(x)

28.1.1 Constructing the Final Wave Function

Substituting our solutions for S0 (x) and S1 (x) back into our original ansatz
S(x) = S0 + ℏi S1 :

i
 
Ψ(x) = exp S(x)
ℏ
   
i Z
ℏ 1
= exp  ± p(x)dx + ln  q 
ℏ i p(x)
  
iZ 1 
 
= exp ± p(x)dx exp ln  q
ℏ p(x)
1 iZ
 
=q exp ± p(x)dx (28.18)
p(x) ℏ

This explicitly demonstrates how the purely real first-order correction (S1 )
gracefully drops out of the complex exponent to become the position-dependent
amplitude!

28.2 Expansion in Powers of ℏ

Because we are working in the semiclassical limit where ℏ is considered "small",
we expand S(x) as a power series in ℏ:
! !2
ℏ ℏ
S(x) = S0 (x) + S1 (x) + S2 (x) + . . . (28.19)
i i

28.2.1 Zeroth-Order Approximation

Taking the limit ℏ → 0, Eq. (8) becomes:

(S0′ )2 = p(x)2 =⇒ S0′ (x) = ±p(x) (28.20)

The WKB Ansatz for a Slowly Varying Potential The WKB Wave Function upto first order 37

Integrating this gives the classical accumulated phase:

Z
S0 (x) = ± p(x)dx (28.21)
R
If we stop here, we get Ψ(x) ≈ Ae±i p(x)dx/ℏ
, which completely ignores the
variation in the amplitude.

28.2.2 First-Order Approximation

To capture the amplitude dynamics, we match terms to the first order in ℏ.

Substituting the expansion into Eq. (8) and keeping only the linear ℏ terms
yields:
2S0′ S1′ − S0′′ = 0 (28.22)

Solving for S1′ :

S0′′ p′ (x) 1 d
S1′ = = = ln(p(x)) (28.23)
2S0 ′
2p(x) 2 dx
Integrating this gives:

1
q 
S1 (x) = ln p(x) =⇒ e−S1 (x) = q (28.24)
p(x)

28.3 The WKB Wave Function upto first order

Combining the zeroth and first-order terms, our wave function in the classi-
cally allowed region (E > V (x)) takes the form:

C1 R C2 − i R p(x)dx
Ψ(x) ∼
i
=q e ℏ p(x)dx + q e ℏ (28.25)
p(x) p(x)

q
Physical Interpretation of the Amplitude: Notice the 1/ p(x) fac-
tor. The quantum mechanical probability density of finding the particle is
|Ψ|2 ∝ 1/p(x) ∝ 1/v(x). This is a beautiful recovery of classical mechan-
ics! Classically, a particle spends more time ∆t = ∆x/v in regions where
it is moving slowly. Therefore, you are more likely to find it there. The
WKB approximation mathematically proves that quantum probability natu-
rally mimics this classical residence time.

28.4 Why can we Expand in ℏ?

A very sharp student might ask: "Wait, ℏ is a fundamental constant of nature
(1.05 × 10−34 J·s). How can we treat it as a ’small’ expansion parameter like
the tunable λ in perturbation theory?"
The WKB Ansatz for a Slowly Varying Potential Why can we Expand in ℏ? 38

This is an important and relevant question. In physics, it is meaningless

to say a dimensionful quantity is "small." It can only be small compared to
something else.
When we say we are expanding in powers of ℏ, we are actually using ℏ
as a convenient bookkeeping device for a purely dimensionless expansion.
Let us see how.

The True Expansion Parameter

The core assumption of WKB is that the potential varies "slowly." But slowly
compared to what? It must be slow compared to the particle’s local de Broglie
wavelength, λ(x) = h/p(x) = 2πℏ/p(x).
Let L be the characteristic length scale over which the momentum p(x)
changes significantly. Mathematically, we can define this length scale via the
′ (x)|
derivative: L1 ≈ |pp(x) . The WKB regime (the "semiclassical limit") requires
that the wavelength is much smaller than this characteristic length:

ℏ p(x) p′ (x)
λ(x) ≪ L =⇒ ≪ ′ =⇒ ℏ ≪1 (28.26)
p(x) |p (x)| p(x)2

The true expansion parameter of the WKB approximation is the dimen-

sionless quantity:
p′ (x)
ϵ(x) = ℏ (28.27)
p(x)2
ℏ as a Derivative Tracker
Notice that our dimensionless parameter ϵ contains exactly one power of ℏ
and exactly one spatial derivative of the momentum (p′ ).
If you carry out the WKB expansion to higher orders, you will find a
beautiful pattern:
• The ℏ0 term (zeroth order) contains no derivatives of p(x).

• The ℏ1 term (first order) contains the first derivative p′ (x).

• The ℏ2 term (second order) contains terms like p′′ (x) and (p′ (x))2 .
Therefore, when we write the formal series S(x) = S0 +ℏS1 +ℏ2 S2 +. . . , we
are simply grouping terms by how many spatial derivatives of the potential
they contain. Because the potential varies "slowly", higher derivatives are
physically smaller. We track ℏ simply because it naturally tags along
with every spatial derivative of the momentum!
By truncating the series at the first order of ℏ, we are making the physical
statement that the rate of change of the momentum (p′ ) is small enough that
we can safely ignore the curvature of the momentum (p′′ ) and higher-order
fluctuations.
Breakdown of WKB: Classical Turning Points 39

29 Breakdown of WKB: Classical Turning Points

The WKB approximation is great, but q it harbors a fatal mathematical flaw.
Look closely at the amplitude factor 1/ p(x).
At locations where E = V (x), the momentum p(x) → 0. These are
the classical turning points (where a classical particle stops and turns
around). At these points, our WKB wave function (28.25) diverges to infinity
(Ψ → ∞). The assumption that the potential varies "slowly" compared to the
wavelength breaks down here because as p → 0, the de Broglie wavelength
λ = h/p → ∞.

Linearized V (x) ≈ E − F0 (x − a)
Energy
Exact V (x)

E
ψ(x) ∼ Ai(ξ)

Allowed Region Forbidden Region

E > V (x) E < V (x)

x
x=a

Figure 2. Zoomed-in view of the classical turning point x = a. The exact potential
is linearized to match the Airy differential equation, providing a bridge between
the oscillatory and decaying WKB solutions.

29.1 Linearizing the Potential: The Airy Equation

To fix this, we must solve the Schrödinger equation exactly in the immediate
vicinity of the turning point x = a. We Taylor expand the potential around
the turning point, keeping only the linear term:

V (x) ≈ V (a) + V ′ (a)(x − a) = E − F0 (x − a) (29.1)

where F0 = −V ′ (a) is the constant classical force at the turning point.

Substituting this linearized potential into the Schrödinger equation:

ℏ2 d2 ψ
− − F0 (x − a)ψ = 0 (29.2)
2m dx2
The WKB Connection Formulas Asymptotic Behavior of Airy Functions 40

To solve this, we introduce a dimensionless variable ξ. Let:

2mF0
x − a = β −2/3 ξ where β = (29.3)
ℏ2
Under this change of variables, the Schrödinger equation simplifies into the
famous Airy Equation:
d2 ψ
− ξψ(ξ) = 0 (29.4)
dξ 2
The solutions to this differential equation are the Airy Functions, Ai(ξ) and
Bi(ξ). They are given by:

1 Z +∞
!
k3
Ai(ξ) = cos kξ + dk
2π −∞ 3
1 Z +∞
 3
 !!
kξ− k3 k3
Bi(ξ) = e + sin kξ + dk
2π −∞ 3

29.2 Asymptotic Behavior of Airy Functions

To connect our exact turning-point solutions back to the WKB solutions far
away from the turning point, we must look at the asymptotic limits of the
Airy functions:
1. Far into the classically forbidden region (ξ ≫ 0):

1 2 3/2
Ai(ξ) ∼ √ 1/4 e− 3 ξ (Exponential decay) (29.5)
2 πξ
1 2 3/2
Bi(ξ) ∼ √ 1/4 e+ 3 ξ (Exponential growth) (29.6)
πξ

Because physically acceptable wavefunctions must not blow up at infinity, we

usually discard the Bi(ξ) solution by setting its coefficient to zero.
2. Far into the classically allowed region (ξ ≪ 0):

1 2 π
 
Ai(ξ) ∼ √ sin |ξ|3/2 + (29.7)
π|ξ|1/4 3 4
1 2 3/2 π
 
Bi(ξ) ∼ √ cos |ξ| + (29.8)
π|ξ|1/4 3 4

30 The WKB Connection Formulas

By matching the asymptotic forms of the Airy functions to the standard WKB
solutions, we can safely cross the turning point. The Airy function acts as
a "bridge" or "patch" between the oscillatory WKB region and the decaying
WKB region.
Understanding the Connection Formulas Mathematically 41

x = a (Turning Point)
Region I: Allowed Region II: Patch
Region III: Forbidden
(WKB Valid) (Airy Valid) (WKB Valid)

Ψ ≈ Ai(ξ) R
Ψ ≈ √1 e− |p|dx
R 
Ψ≈ √1
p
cos pdx − π
4 2 |p|

Figure 3. The three distinct spatial regimes of the WKB approximation. The
exact Airy function solution bridges the gap over the turning point where standard
WKB diverges.

Let us define the WKB phase integrals starting from the turning point
x = a:

• Classically Allowed Region (x < a): W (x) = 1 Ra

ℏ x
p(x′ )dx′

• Classically Forbidden Region (x > a): K(x) = 1 Rx

ℏ a
|p(x′ )|dx′

Matching the forms yields the important Connection Formulas. If

we have an exponentially decaying state in the forbidden region, it uniquely
dictates the phase of the oscillatory state in the allowed region:

1 1 π
 
e−K(x)
←→ cos W (x) − (30.1)
4
q q
2 |p(x)| p(x)

Note: Where does the −π/4 phase shift come from? It is the direct
signature of the Airy function Ai(ξ)! Classical mechanics predicts an abrupt
reflection at the turning point. Quantum mechanics smooths this reflection
out over the Airy patch region, which delays the wave slightly, resulting in an
exact π/4 phase loss. Details are in the following section.

31 Understanding the Connection Formulas Math-

ematically
We previously established that the WKB approximation breaks down at the
classical turning points where p(x) → 0. To connect the oscillatory solutions
 Transforming the Schrödinger Equation
Understanding the Connection Formulas Mathematically 42

in the allowed region to the exponentially decaying solutions in the forbidden

region, we must cross this singularity.
Instead of patching with the Airy functions (which we did conceptually
earlier), we will now derive the exact connection formulas by moving into the
complex plane. By analytically continuing the wave function around the
singular turning point via a semicircular path, we can bypass the breakdown
region entirely!

31.1 Transforming the Schrödinger Equation

Let us start with the 1D Time-Independent Schrödinger equation:

d2 Ψ
2
+ k 2 (x)Ψ = 0 (31.1)
dx
where we define the local wavenumber:

2m p2 (x)
k 2 (x) = [E − V (x)] = (31.2)
ℏ2 ℏ2

To manipulate this into a form suitable for the WKB amplitude, we define
a new phase variable σ(x) and substitute a trial function u(x):
Z x
u(x)
σ(x) = k(x′ )dx′ and Ψ(x) = q (31.3)
k(x)

By the chain rule, the spatial derivative transforms as:

d dσ d d
= = k(x) (31.4)
dx dx dσ dσ
The second derivative becomes:

d2 Ψ d2 Ψ
!
d dΨ dΨ
2
= k(x) = k ′ (x) + k 2 (x) 2 (31.5)
dx dx dσ dσ dσ
√
Now, let us evaluate the derivatives of our trial wave function Ψ = u/ k
with respect to σ:

dΨ d  −1/2  du 1 −3/2 dk
= k u = k −1/2 − k u (31.6)
dσ dσ dσ 2 dσ
2
Taking the second derivative ddσΨ2 and substituting it back into Eq. (31.5)
allows us to entirely eliminate the first derivative term, yielding a simplified
 Analytic Continuation Around the Turning Point
Understanding the Connection Formulas Mathematically 43

differential equation for u:

 !2 
d2 u  3 k′ 1 k ′′ 
+ 1+ − u=0 (31.7)
dσ 2 4 k2 2 k3

In the WKB limit (slowly varying potential), the derivatives of k(x) are very
small compared to k(x) itself. Thus, the terms involving k ′ and k ′′ can be
neglected, leaving:
d2 u
+u≈0 (31.8)
dσ 2
The solutions are simply u(σ) = e±iσ . Substituting back into Eq. (31.3), we
recover the standard WKB forms:
1 R
Ψ(x) ≈ q e ±i k(x)dx
(31.9)
k(x)

31.2 Analytic Continuation Around the Turning Point

Let x = a be a classical turning point where V (a) = E. Assume the potential

slopes upward linearly near the turning point:

V (x) ≈ E + V ′ (a)(x − a) (31.10)

where V ′ (a) > 0. Therefore, x < a is the classically allowed region (oscilla-
tory), and x > a is the classically forbidden region (decaying).

Near x = a, the local wavenumber behaves as:

2m 2mV ′ (a)
k 2 (x) = [E − V (x)] ≈ − (x − a) ≡ −α2 (x − a) (31.11)
ℏ2 ℏ2
Taking the square root to find k(x):

k(x) ≈ iα(x − a)1/2 (31.12)

Because WKB breaks down precisely at x = a, we will avoid the origin

by moving the coordinate x into the complex plane. We define a semicircular
contour around x = a in the upper-half complex plane:

x − a = ρeiθ (31.13)

where ρ is a small, constant radius large enough to stay in the valid WKB
regime, and we vary the angle θ from 0 to π.
 Analytic Continuation Around the Turning Point
Understanding the Connection Formulas Mathematically 44

Moving from the Forbidden to the Allowed Region

Let us start in the classically forbidden region (x > a), which corresponds to
θ = 0. Here, x − a = ρ > 0. The forbidden momentum is pure imaginary.
Let us track an exponentially decaying WKB solution:

1 Z x  
Ψdecay (x) = q exp − |k(x′ )|dx′ (31.14)
2 |k(x)| a

Now, let us trace how the amplitude and phase behave as we move along
the complex semicircle from θ = 0 (Right) to θ = π (Left).

k(x) = iα(ρeiθ )1/2 = iαρ1/2 eiθ/2 (31.15)

• At θ = 0 (Right side): k(x) = iαρ1/2 . This is pure imaginary, as

expected.

• At θ = π (Left side): k(x) = iαρ1/2 eiπ/2 = iαρ1/2 (i) = −αρ1/2 .

Notice that traversing the semicircle introduces a phase factor of eiπ/2 = i
to the momentum!

Evaluating the Amplitude Transformation

The WKB amplitude factor transforms as follows:

1 1 1
∝ = = ρ−1/4 e−iθ/4 (31.16)
(x − a) (ρe )1/4
q
1/4 iθ
k(x)

As θ goes from 0 → π, the amplitude picks up a complex phase:

1 1
√ −→ √ e−iπ/4 (31.17)
kx>a kx<a

Evaluating the Phase Integral Transformation

The integral in the exponent transforms similarly. For a small distance ρ near
the turning point:
Z x Z ρeiθ
2 2
|k(x )|dx ≈
′ ′
α(x′ )1/2 dx′ = α(ρeiθ )3/2 = αρ3/2 e3iθ/2 (31.18)
a 0 3 3
As θ goes from 0 → π:
e3iπ/2 = −i (31.19)

Thus, the argument of the exponent smoothly morphs from real (decaying)
to pure imaginary (oscillating):
Z x Z a
− |k(x′ )|dx′ −→ +i k(x′ )dx′ (31.20)
a x
 Synthesizing the Connection Formula
Understanding the Connection Formulas Mathematically 45

31.3 Synthesizing the Connection Formula

Combining the transformation of the amplitude (Eq. (31.17)) with the trans-
formation of the exponent, our initial decaying wave on the right side maps
to the left side as:
1
 Z a 
Ψ(x) −→ q e exp i −iπ/4
k(x )dx ′ ′
+ c.c. (31.21)
2 k(x) x

1
Z a
π

=q cos k(x′ )dx′ − (31.22)
k(x) x 4

(Note: We add the complex conjugate because the physical wave function must
be entirely real, which corresponds to also traversing the contour in the lower-
half complex plane from 0 → −π).

Summary of Connection Formulas:

If we define the phase integrals starting from the turning point x = a:

1 1
Z a
π
Rx 
|k|dx′
−
e a (forbidden) ←→ cos kdx −′
(allowed)
4
q q
2 |k(x)| k(x) x

(31.23)
Similarly, if the wavefunction is purely exponentially growing in the for-
bidden region, its analytic continuation picks up a +π/4 phase shift:

1 −1
Z a
π
Rx 
|k|dx′
+
e a (forbidden) ←→ sin kdx −′
(allowed)
4
q q
|k(x)| k(x) x

(31.24)
This complex plane derivation verifies the exact π/4 phase loss that we
previously justified visually using the Airy function!