Quantum Physics III

Swiss Federal Institute of Technology Lausanne

Master program in Physics

Prof. Claudio Scrucca

Institute for Theoretical Physics
CH-1015 Lausanne
Contents

1 The semiclassical approximation to quantum mechanics 1

1.1 The Hamilton-Jacobi description of classical mechanics . . . . . . . . . . . . 1
1.2 The classical limit of quantum mechanics . . . . . . . . . . . . . . . . . . . 7
1.3 Validity of the classical approximation . . . . . . . . . . . . . . . . . . . . . 10
1.4 The semiclassical expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 The semiclassical wave function . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.1 Classically allowed regions . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.2 Classically forbidden regions . . . . . . . . . . . . . . . . . . . . . . 14
1.5.3 Classical turning points . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.4 Matching strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Behavior in a linear potential . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.7 Matching conditions across turning points . . . . . . . . . . . . . . . . . . . 21
1.7.1 Turning points with increasing potential . . . . . . . . . . . . . . . . 22
1.7.2 Turning points with decreasing potential . . . . . . . . . . . . . . . . 23

2 One-dimensional problems in the semiclassical approximation 25

2.1 Energy levels in a generic potential well . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Evaluation of the semiclassical wave function . . . . . . . . . . . . . 26
2.1.2 Quantum spectrum and classical interpretation . . . . . . . . . . . . 29
2.1.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.4 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 Tunneling probability through a generic barrier . . . . . . . . . . . . . . . . 32
2.2.1 Evaluation of the semiclassical wave function . . . . . . . . . . . . . 33
2.2.2 Tunneling probability and classical interpretation . . . . . . . . . . . 36
2.2.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.4 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Lifetime of a metastable state . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 Evaluation of the semiclassical wave function . . . . . . . . . . . . . 40
2.3.2 Decay width and classical interpretation . . . . . . . . . . . . . . . . 43

1
 2.4 Multiple wells and barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.1 Multiple wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4.2 Multiple barriers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.4.3 Multiple metastable wells . . . . . . . . . . . . . . . . . . . . . . . . 51

3 Central problems and the semiclassical approximation 55

3.1 Separation of radial and angular parts . . . . . . . . . . . . . . . . . . . . . 55
3.2 Exact angular wave function . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Semiclassical angular wave function . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 Exact radial wave function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4.1 Asymptotic behavior in the origin . . . . . . . . . . . . . . . . . . . 60
3.4.2 Asymptotic behavior at infinity . . . . . . . . . . . . . . . . . . . . . 60
3.4.3 Relative asymptotic behaviors for vanishing potential . . . . . . . . 61
3.4.4 Relative asymptotic behavior for non-vanishing potential . . . . . . 61
3.5 Semiclassical radial wave function . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Localized stationary states . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.6.1 Spectrum and quantization rule . . . . . . . . . . . . . . . . . . . . . 65
3.6.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7 Delocalized stationary states . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.7.1 Scattering amplitude and cross section . . . . . . . . . . . . . . . . . 70
3.7.2 Classical interpretation . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.7.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.8 Metastable quasi-stationary states . . . . . . . . . . . . . . . . . . . . . . . 80
3.8.1 Energy and width . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.9 Resonance effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4 Approximation methods for central scattering problems 85

4.1 General behavior of the partial-wave expansion . . . . . . . . . . . . . . . . 85
4.2 Weak coupling scattering and the Born approximation . . . . . . . . . . . . 86
4.2.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.3 High energy scattering and the eikonal approximation . . . . . . . . . . . . 90
4.3.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.4 Low energy scattering and the threshold approximation . . . . . . . . . . . 94
4.4.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5 General operatorial formalism for scattering problems 103

5.1 The time-evolution operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 The scattering matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 The cross section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

2
 5.4 Systematics of the perturbative expansion . . . . . . . . . . . . . . . . . . . 111
5.5 Central problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.6 Relation to Green functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

6 Approximation methods for many-body problems 121

6.1 Individual wave functions approach . . . . . . . . . . . . . . . . . . . . . . . 121
6.1.1 The Hartree approximation . . . . . . . . . . . . . . . . . . . . . . . 122
6.1.2 The Hartree-Fock approximation . . . . . . . . . . . . . . . . . . . . 123
6.1.3 Self-consistent field solution . . . . . . . . . . . . . . . . . . . . . . . 125
6.2 Particle density approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2.1 The Thomas-Fermi approximation . . . . . . . . . . . . . . . . . . . 126
6.2.2 The Thomas-Fermi-Dirac approximation . . . . . . . . . . . . . . . . 129
6.2.3 Equilibrium particle density solution . . . . . . . . . . . . . . . . . . 130
6.3 Application to the structure of atoms . . . . . . . . . . . . . . . . . . . . . . 130

3
Chapter 1

The semiclassical approximation

to quantum mechanics

In this chapter, we will describe the general features of the classical limit of quantum
mechanics and present a systematic method of approximation that allows to account in a
very simple and efficient way for small quantum effects. This semiclassical approximation
allows to find a general approximate form of the wave function describing a quantum
system in a quasi classical regime, expressed in terms of quantities describing the classical
motion, even in situations where the exact quantum mechanical wave function cannot be
found exactly. This is true both in classically allowed regions, where the wave function is
an oscillating wave with a sizable amplitude, and in classically forbidden regions, where
the wave function is instead an exponentially damped function. The joining of these two
kinds of behaviors across classical turning points of the motion requires some care, and
the precise gluing prescription must be derived by studying the exact wave function in the
vicinity of such a turning point, where the potential can be linearized.

1.1 The Hamilton-Jacobi description of classical mechanics

In the Lagrangian formulation of mechanics, a system is described in terms of a Lagrangian
functional L depending on the variables q i (t) and the corresponding velocities q̇ i (t), besides
the time t:

L = L(q i , q̇ i , t) . (1.1.1)

In terms of this function, the equations of motions are rewritten as Euler-Lagrange equa-
tions, which are n ordinary differential equations of the second order in time:
 
d ∂L ∂L
i
− i = 0. (1.1.2)
dt ∂ q̇ ∂q
These equations can also be derived from a variational principle: the Hamilton principle.
This is based on an action functional obtained by integrating the Lagrangian for an arbi-
trary path q i (t) with the only constraint that its values at some initial and final times t1

1
and t2 are held fixed to some q1i and q2i :
Z t2
I= L dt , (1.1.3)
t1

The classical trajectory then corresponds to the configuration q i (t) that extremizes the
above action functional. More precisely, the Euler-Lagrange equations are equivalent to
the requirement that the action must be stationary with respect to a generic synchronous
variation δ of the path q i (t) connecting the two fixed end-points:

δI = 0 . (1.1.4)

From this variational formulation, it becomes obvious that the Lagrangian function is not
uniquely defined, even once a definite parametrization of the problem has been chosen.
Indeed, two Lagrangians that differ by the total time-derivative of an arbitrary function F
lead to actions that differ only by the difference of this function at the two end-points, and
since at those points the variation δ is defined to vanish, these two actions lead to the same
equations of motion. Such an ambiguity has however no really interesting consequences
in this formulation.
In the Hamiltonian formulation of mechanics, the system is instead described by a
Hamiltonian functional H, which is obtained by performing the Legendre transform of the
Lagrangian and depends on the variables q i (t) and the canonical momenta pi (t) conjugate
to the velocities q̇ i (t), besides the time t:

H = pi q̇ i − L = H(q i , pi , t) , (1.1.5)

with
∂L
pi = . (1.1.6)
∂ q̇ i
In terms of this function, the equations of motions are rewritten as Hamilton equations,
which are 2n ordinary differential equations of the first order in time:
∂H
q̇ i − = 0, (1.1.7)
∂pi
∂H
ṗi + i = 0 . (1.1.8)
∂q
These equations can again be derived from a variational principle: the modified Hamilton
principle. This is based on the same action functional as before, but now rewritten by
expressing the Lagrangian in terms of the Hamiltonian and interpreting it as a functional of
the variables q i (t) and the momenta pi (t), which are now to be considered as independent
variables:
Z t2
pi q̇ i − H dt ,


I= (1.1.9)
t1

The classical trajectory then corresponds to the configuration (q i (t), pi (t)) that extremizes
the above action functional. More precisely, the Hamilton equations are equivalent to the

2
requirement that the action must be stationary with respect to generic and independent
synchronous variations δ of both the path q i (t) connecting the two fixed end-points and
the momenta pi (t):

δI = 0 . (1.1.10)

Again, from this variational formulation, it becomes clear that the canonical momenta and
the Hamiltonian function are actually not uniquely defined, even once a definite choice for
the coordinates has been made. This is related to the fact that already the Lagrangian
was in fact defined modulo the total time-derivative of an arbitrary function F . In this
formulation, it turns out that one can actually exploit this fact in a rather interesting way.
One may at this point consider general reparametrizations of the phase space, which
correspond to switching from the variables (q i , pi ) to some new variables (Qi , Pi ):

Qi = Qi (q j , pj ) , (1.1.11)
Pi = Pi (q j , pj ) . (1.1.12)

The subset of these transformations that are of interest are the canonical transformations.
These are defined by the requirement that the new equations of motion for (Qi , Pi ) retain
the same form as the original Hamilton equations for (q i , pj ), but with some new function
K(Qi , Pi ) replacing the Hamiltonian H(q i , pi ):

∂K
Q̇i − = 0, (1.1.13)
∂Pi
∂K
Ṗ i + = 0. (1.1.14)
∂Qi
Recalling the variational formulation, we know that these new equations follow from the
stationarity of an action based on the functional Pi Q̇i − K, while the original Hamilton
equations follow from the stationarity of an action based on the Lagrangian pi q̇ i − H. In
order for the two sets of equations to coincide, the above two functionals must then differ
at most by the total time-derivative of an arbitrary function F :

K − Pi Q̇i − H − pi q̇ i = Ḟ .



(1.1.15)

The function F can depend only on half of all the variables q i , Qi , pi , Pi , in addition to
time, because it should be possible to interpret it both as a function of the old phase space
variables and of the new ones, separately. There are then four distinct possibilities, which
correspond to chose F to depend on the variables (q i , Qi ), (q i , Pi ), (pi , Qi ) or (pi , Pi ),
besides time. For each choice, one may make the form of the total time-derivative Ḟ more
explicit and plug it back into the constraint (1.1.15). One then finds an equation with two
terms that multiply the time derivative of each of the two variables on which F depends,
plus a term that does not depend on time derivatives of these variables. These three terms
must then vanish independently, and this fixes the relation between the pair of variables
on which F depends and the other pair, and also the relation between K and H. After
suitably parametrizing the function F in terms of some other generating functions F1 , F2 ,

3
F3 and F4 , plus some explicit quadratic functions of the involved variables, one finds that
these four types of canonical transformations take the following forms:
∂F1 ∂F1 ∂F1
F1 = F1 (q i , Qi , t) : pi = + , Pi = − , K=H+ , (1.1.16)
∂q i ∂Qi ∂t
∂F2 ∂F2 ∂F2
F2 = F2 (q i , Pi , t) : pi = + i , Qi = + , K=H+ , (1.1.17)
∂q ∂Pi ∂t
∂F3 ∂F3 ∂F3
F3 = F3 (pi , Qi , t) : q i = − , Pi = − i , K=H+ , (1.1.18)
∂pi ∂Q ∂t
∂F4 ∂F4 ∂F4
F4 = F4 (pi , Pi , t) : q i = − , Qi = − i , K=H+ . (1.1.19)
∂pi ∂P ∂t

In the Hamilton-Jacobi formulation, the dynamics of a mechanical system is described

in yet a different way, in terms of a principal function S depending only on the variables
and time, and not on the velocities or momenta. This approach proves to be illuminating
to understand the classical limit of quantum mechanics. The method to set it up consists
in looking for a canonical transformation generated by a function F = S such that the
new variables Qi and Pi are constant:

Qi (q i , pi ) = Qi0 , (1.1.20)
i
Pi (q , pi ) = P0i . (1.1.21)

This implies that Q̇i = 0 and Ṗi = 0, and the new Hamiltonian K should therefore be a
constant. Requiring this to be zero, without loss of generality, one must then have:
∂S
H(q i , pi , t) + = 0. (1.1.22)
∂t
In this formulation, the dynamics is then entirely contained in the canonical transforma-
tion, and solving the problem is equivalent to finding the explicit form of this transfor-
mation, that is the generating function S. To do so, one has to choose one of the four
concrete types of canonical transformations listed above. It is convenient to chose the sec-
ond one, and therefore choose the generating function to depend on the old non-constant
coordinates q i and the new constant momenta Pi : S = S(q i , Pi , t). One then finds that:

∂S
pi = . (1.1.23)
∂q i

Plugging this into eq. (1.1.22), one then deduces that the generating function S must
satisfy the Hamilton-Jacobi equation:
 
i ∂S ∂S
H q, ,t + = 0. (1.1.24)
∂qi ∂t

This is a single first-order partial differential equation in the n variables q i and time
t, for given values of the variables Pi which are constants. This fully determines the
dependence of S on the variables q i and time, for fixed values of the constant Pi . Besides
an irrelevant additive constant, the general solution then involves n integration constants

4
and n constant values of P i , which altogether can be mapped to the 2n initial values q0i
and p0i of the q i and the pi . One finally verifies that the functional form of S is formally
related to the indefinite action obtained by integrating L over time, once the trajectory is
known. Indeed, one finds:
dS ∂S ∂S
= i q̇ i + = pi q̇ i − H = L , (1.1.25)
dt ∂q ∂t
The principal function is then given by the following expression, interpreted as a function
of q i and t on the trajectory:
Z t
S= L dt′ . (1.1.26)
t0

In the Hamilton-Jacobi approach, the equations of motions are therefore rewritten as a

partial differential equation of a single functional S of the coordinates q i , and the momenta
pi are given by the gradient of this function.
In the cases where the Hamiltonian H does not dependent explicitly on time and
corresponds to the conserved energy E of the system, things further simplify. One then
has:

H(q i , pi ) = E . (1.1.27)

In this situation, the usual equations of motion can be derived by an alternative variational
principle: the Maupertuis or minimal action principle. This is based on a reduced action
functional obtained by integrating the function pi q̇ i over time between t1 and t2 with the
constraint that the considered path should correspond to a fixed constant value E of the
Hamiltonian:
Z t2
A= pi q̇ i dt . (1.1.28)
t1

The true trajectory with energy E then corresponds to the configuration that extremizes
the above functional for constant H = E. More precisely, the equations of motions are
equivalent to the requirement that the reduced action must be stationary with respect to
generic and independent asynchronous variations ∆ of both q i (t) and pi (t):

∆A = 0 . (1.1.29)

In this situation, the Hamilton-Jacobi equation also simplifies, and becomes:

 
i ∂S ∂S
H q, + = 0. (1.1.30)
∂qi ∂t

The time dependence can then be separated, and the general form of the solution for the
principal function S becomes:

S(q i , t) = W (q i ) − Et . (1.1.31)

5
The function W (q i ) is called the characteristic function and does not depend explicitly on
time. It satisfies the equation
 
i ∂W
H q, =E. (1.1.32)
∂qi

This is a single first-order partial differential equation in the n variables q i , for given values
of the constant Pi . This entirely determines the dependence of S on the original variables
q i . Besides an irrelevant additive constant there are now n − 1 integration constants and
n parameters, which can be mapped to the 2n − 1 initial values q0i and p0i of the q i and
the pi for a trajectory of given energy E. One finally verifies that the functional form of
W is formally related to the indefinite reduced action obtained by integrating pi q̇ i over
time, once the trajectory is known. Indeed, one finds:
dW ∂W i
= q̇ = pi q̇ i , (1.1.33)
dt ∂q i
The characteristic function is then given by the following expression, interpreted as a
function of q i on the trajectory:
Z t
W = pi q̇ i dt′ . (1.1.34)
t0

To conclude, let us write more explicitly the functions appearing in the various de-
scriptions reviewed in this section for the case of a particle of mass m moving in a d-
dimensional space with coordinates ~x, subject to a conservative force F~ (~x) = −∇V~ (x).
The Lagrangian and the Hamiltonian then take the usual form L = T − V and H = T + V ,
where the kinetic energy has the minimal form T = 12 m~x˙ 2 = p~ 2/(2m), while the potential
V = V (~x) is arbitrary. The canonical momentum is given by p~ = m~x˙ , and the Lagrangian
and Hamiltonian equations of motions are just the usual second order and the decomposed
first order versions of the Newtonian equations of motions, which read:

p~˙ + ∇V
~ = 0. (1.1.35)

When the potential is independent of time, the energy is conserved and the corresponding
conservation law reads:
p~ 2
+V =E. (1.1.36)
2m
In the Hamilton-Jacobi formulation, the momentum is given by the gradient of the prin-
cipal function S, and the dynamics is governed by the Hamilton-Jacobi partial differential
equation for this function, which reads:
∂S 1 ~ 2
+ (∇S) + V = 0 . (1.1.37)
∂t 2m
When the potential is independent of time, the energy is conserved and this leads to the
reduced Hamilton-Jacobi equation for the characteristic function W :
1 ~
(∇W )2 + V = E . (1.1.38)
2m

6
In this case, it is possible to verify explicitly that indeed (1.1.37) is equivalent to (1.1.35),
and similarly (1.1.38) is equivalent to (1.1.36). To do so, one needs to use the definition of
the momentum as the gradient of a field, ~p = ∇S ~ or ~p = ∇W
~ , and exploit the fact that one
also has p~ = m~x˙ . Taking the gradient of (1.1.37), one then reproduces (1.1.35). Indeed,
the first two terms give ∂/∂t (∇S)~ ~ · ∇)
+ (∇S ~ (∇S)/m
~ = [∂/∂t + (~x˙ · ∇)]
~ p~ = d/dt p~ = p~˙ ,
while the last term simply gives ∇V ~ . Equation (1.1.38) is instead directly equivalent to
(1.1.36). Indeed, the first term on the left-hand side just gives (∇W ~ )2 /(2m) = ~p 2 /(2m).

1.2 The classical limit of quantum mechanics

The semiclassical expansion of quantum mechanics is based on the idea that in certain
situations one may treat ~ as a small quantity and perform a formal expansion in powers
of it. This will be a good approximation whenever the relevant quantities in the problem
with the dimensions of an action are much larger than ~. This is perfectly analogous to
the non-relativistic limit of classical mechanics, in which on treats c as a large quantity
and performs an expansion in inverse powers of it. Such an expansion is in that case a
good approximation whenever the velocities in the problem are much smaller than c.
To study the classical limit of quantum mechanics, let us start from the Schrödinger
equation for the wave function ψ in the presence of a generic potential V , which we rewrite
as:
∂ψ ~2
i~ =− ∆ψ + V ψ . (1.2.39)
∂t 2m
Clearly we cannot simply take the brutal limit ~ → 0 of this equation, since by doing so it
would loose any sense. The physical reason for this is that in the formal limit ~ → 0 the
wave function of a quantum system becomes a rapidly oscillating function in classically
accessible regions and a strongly damped function in classically forbidden regions, and it
is only its squared norm that tends to a simple smooth function representing the density
of probability of finding the classical particle at a given position. One then has to proceed
in a slightly more sophisticated way, following a method that was developed by Wentzel,
Kramers and Brillouin and is often called the WKB approximation. This consists in
parametrizing the wave function in the following way:
i
ψ(~x, t) = A e ~ S(~x,t) . (1.2.40)

Plugging this parametrization back into the Schrödinger equation, one finds that the
complex function S must satisfy the following differential equation:
∂S 1 ~ 2 i~
− = (∇S) + V − ∆S . (1.2.41)
∂t 2m 2m
The brutal limit ~ → 0 of this equation makes now sense, and takes the real form
∂S 1 ~ 2
− = (∇S) + V . (1.2.42)
∂t 2m

7
 The equation (1.2.42) that we have obtained as the classical limit of Schrödinger’s
equation governing the evolution of the quantum mechanical wave function is now recog-
nized to be formally identical to the Hamilton-Jacobi equation
∂S
− = H(~x, p~) , (1.2.43)
∂t
where S is Hamiltons’s principal function and the momentum p~ is the gradient of it:

~ .
p~ = ∇S (1.2.44)

The formal solution for S is then the indefinite integral of the Lagrangian L over time, as
seen in previous section:
Z t
S= L(~x, ~x˙ , t′ ) dt′ , (1.2.45)
t0

We thus discover that in the formal limit ~ → 0, the phase S of the quantum-mechanical
wave function describing a generic state maps to the indefinite action S of the classical
trajectory. This implies that in this limit the quantum system can be described by classical
trajectories, which are orthogonal to the surfaces of constant S, since the momentum is
the gradient of the latter. We can then say that in the classical limit the behavior of
i
the quantum system defined by the wave function ψ = Ae ~ S is described by a classical
trajectory that is orthogonal to the surfaces of constant phase S.
The situation further simplifies for stationary configurations. Recall that for a station-
ary state of definite energy E, the wave function has a fixed time-dependence and can be
written as:
i
ψ(~x, t) = φ(~x)e− ~ Et . (1.2.46)

The stationary wave function φ(~x) then satisfies the time-independent Schrödinger equa-
tion:
~2
− ∆φ + V φ = Eφ (1.2.47)
2m
To discuss the classical limit, we now proceed as before and parameterize the stationary
wave function through a stationary phase, as:
i
φ(~x) = A e ~ W (~x) . (1.2.48)

Comparing (1.2.48) with (1.2.40), we see that

S(~x, t) = W (~x) − Et . (1.2.49)

Plugging now the parametrization (1.2.46) into (1.2.47), one finds the complex equation:

1 i~
(∇W )2 + V − ∆W = E . (1.2.50)
2m 2m

8
The brutal limit ~ → 0 then leads to the real equation
1
(∇W )2 + V = E . (1.2.51)
2m
The equation (1.2.51) corresponds to the reduced Hamilton-Jacobi equation for mo-
tions with a given energy E, that is
H(~x, p~) = E , (1.2.52)
where W is Hamilton’s characteristic function and the momentum p~ is the gradient of it:
~ .
p~ = ∇W (1.2.53)
The formal solution for W is then given by the integral of the phase-space differential, as
reviewed in previous section:
Z t
W = p~ · ~x˙ dt′ . (1.2.54)
t0
We thus see that in the formal limit ~ → 0, the phase W of the quantum-mechanical wave
function of a stationary state maps to the indefinite reduced action W of the classical
trajectory. As before, in this limit the quantum system can be described by classical
trajectories, which are orthogonal to the surfaces of constant W since the momentum is
the gradient of this function. In the classical limit, the behavior of the quantum system
i
defined by the wave function φ = Ae ~ W is thus described by a classical trajectory that is
orthogonal to the surfaces of constant phase W .
It should be emphasized that whenever ~ is non-zero, the Schrödinger equation is in-
trinsically complex and physical states are described by a complex wave function. This sug-
gests that even in the classical limit derived above, where the wave function is parametrized
i i
as ψ = e ~ S or φ = e ~ W and the functions S and W are determined by a real Hamilton-
Jacobi equation, one can nevertheless associate a meaning also to complex solutions for S
and W . More precisely, we know that the solutions for S and W are respectively given by
the classical indefinite action and the classical indefinite reduced action, and it turns out
that these functions of the coordinates are real in classically allowed regions and imagi-
nary in the classically forbidden regions. As a result, the wave function in the classical
limit behaves as a rapidly oscillating wave in the classically allowed regions and a strongly
suppressed exponential profile in the classically forbidden regions:
Classically allowed regions: rapidly oscillating wave function , (1.2.55)
Classically forbidden regions: strongly damped wave function . (1.2.56)
Of course, in the formal limit where one brutally sends ~ → 0, the wave function can
be approximated with a smooth average amplitude that is non-zero only in classically
allowed regions and vanishes in classically forbidden regions. To reliably compute the
position dependence of this amplitude, one needs however to take into account at least
the first subleading correction in the Schrödinger equation. Indeed, subleading corrections
to the quantities S or W of order ~ induce a non-trivial correction to the exponents ~S or
~W of the wave function that is independent of ~, and therefore modulate the amplitude
in a significant way.

9
1.3 Validity of the classical approximation
The above argumentation shows the transition from quantum mechanics to classical me-
chanics is formally perfectly analogous to the transition from wave optics to geometrical
optics. In both cases, a wave propagation is traded with a trajectory, the velocity be-
ing orthogonal to the wave fronts. We will now see that the classical limit of quantum
mechanics is essentially applicable in the regime in which the De Broglie wave length
λ̄ = ~/p is small compared to the length-scales characterizing the system, very much like
the geometrical limit of wave optics is applicable when the optical wave length λ is small
compared to the length scales of the problem.
In a generic situation, the condition for the classical approximation to be reliable is
that the last term in (1.2.41) that we have dropped to arrive to (1.2.42) should have a
negligible effect, in particular with respect to the first term. This implies the condition
~|∆S| ≪ (∇S) ~ 2 , which can be rewritten as ~|∇ ~ · p~| ≪ p2 since ~p = ∇S. ~ Similarly,
in a stationary situation, the condition for the classical approximation to be reliable is
similarly that the last term in (1.2.47) that we have dropped to arrive to (1.2.51) should
have a negligible effect, in particular with respect to the first term. This implies that
~|∆W | ≪ (∇W~ )2 , which can again be written as ~|∇ ~ · p~| ≪ p2 since p~ = ∇W
~ . A general
necessary condition for the validity of the semiclassical approximation is therefore that

~ ·~
~|∇ p| ≪ p2 . (1.3.57)

At the quantitative level, this requires in general that p is large on average and varies
slowly from point to point, implying that the De Broglie wave length λ is small and
slowly varying, as expected from the analogy with optics. At the quantitative level, these
conditions for the applicability of a classical approximation based on trajectories depend
on the form of the potential V (~x) in the quantum theory, very much like they depend on
the refraction index in optics.
For one-dimensional systems, the situation further simplifies and the above condition
simply reads
~ dp
≪ 1. (1.3.58)
p2 dx
The restriction (1.3.58) can then be rewritten as a condition involving only the De Broglie
wave length λ. Indeed, using the relations
~ dλ̄ ~ dp
λ̄ = , =− 2 , (1.3.59)
p dx p dx
one finds that the condition simply becomes the requirement that the De Broglie wave
length should vary slowly over space:
dλ̄
≪ 1. (1.3.60)
dx
For a system of finite size a in which λ̄ does not vary too rapidly, one has dλ̄/dx ∼ λ̄/a and
this condition reduces to the requirement that the wave length should be small compared

10
to the dimensions of the system:

|λ̄| ≪ a . (1.3.61)

The restriction (1.3.58) can also be rephrased in an alternative way as a condition involving
the potential. Indeed, the momentum p is related to the potential by the equation of
p
motion dp/dt = −dV /dx and also by the relation p = 2m(V − E), which both imply
that
dp m dV
=− . (1.3.62)
dx p dx

Substituting this into (1.3.58) one finds that the potential should vary slowly over space,
and more precisely
m~ dV
≪ 1. (1.3.63)
p3 dx

In terms of the force F = −dV /dx, this implies that the momentum p should be much
larger than |m~F |1/3 , or in other words that the particle should not be too slow:

|p| ≫ |m~F |1/3 . (1.3.64)

This finally implies that the De Broglie wave length should be smaller than a certain
quantum-mechanical length scale determined by the mass m, the force F and the Planck
constant ~:
−1/3
mF
|λ̄| ≪ . (1.3.65)
~2

1.4 The semiclassical expansion

Whenever the quantum effects correcting the behavior of the classical limit are small, one
may compute these by using a perturbative expansion. Since the last term in eq. (1.2.41)
that is responsible for these corrections is proportional to ~, this is formally equivalent to
solve eq. (1.2.41) through a series expansion in powers of ~.
From now on, we shall focus on stationary states of the quantum theory, which possess
a definite energy E. We shall then start from the stationary Schrödinger equation (1.2.50).
For convenience, we start by rewriting this equation by bringing the potential term on the
right-hand side and multiplying by 2m, in the form

~ )2 − i~∆W = 2m(E − V ) .
(∇W (1.4.66)

We then look for a solution for W taking the form of a series expansion in −i~:


2
W (~x) = W0 (~x) + −i~ W1 (~x) + −i~ W2 (~x) + · · · (1.4.67)

11
Plugging back into (1.4.66) and requiring the coefficients of all orders in −i~ to vanish,
one finds a infinite set of coupled equations for the coefficients Wn . The first ones read:


 ~ 0 )2 = 2m(E − V ) ,
(∇W


~ 1 + 1 ∆W0 = 0 ,

~ 0 · ∇W

 ∇W


2
(1.4.68)


 ( ~ 1 )2 + 2∇W
∇W ~ 0 · ∇W
~ 2 + ∆W1 = 0 ,





 ···

These equations for W0 , W1 , W2 , · · · can be solved in sequence and allow in principle

to determine the solution W up to any order in ~. This procedure defines what is called
the semiclassical expansion, and is clearly well-defined only under the hypothesis that the
series converges.
In practice, the semiclassical expansion is useful only when quantum effects are small
compared to classical effects. It is then usually enough to restrict to the first two terms
in the expansion and write:

W ≃ W0 − i~W1 . (1.4.69)

In such an approximation, W0 describes the classical behavior whereas W1 encodes the

leading quantum effects. The wave function then takes the following form:
i
φ(x) = AeW1 (~x) e ~ W0 (~x) . (1.4.70)

We see that, as already remarked, the leading term W0 controls the coefficient of the 1/~
exponent and therefore determines the length scale on which the wave function oscillates
or dies out, while the subleasing term W1 controls the amplitude of the wave function.
For simplicity, let us consider again a one-dimensional problem. Since the quantity
2m(E − V ) corresponds to the classical momentum p, we shall introduce the short hand
notation
p
p(x) = 2m(E − V (x)) . (1.4.71)

The two equations determining W0 and W1 then simply read:

2
 W0′ = p2 ,


(1.4.72)
 W ′ W ′ + 1 W ′′ = 0 .
0 1
2 0
The first equation directly implies that W0′ = ±p, whereas the second can be rewritten as
W1′ = − 12 W0′′ /W0′ and therefore implies, together with the first one, that W1′ = − 12 p′ /p.
The general solution is then given by:
 Z x
 W0 (x) = ± p(x′ )dx′ + constant ,

(1.4.73)
 p
W1 (x) = − ln p(x) + constant .


12
 i
The semiclassical stationary wave function φ = Ae ~ W then takes the following simple
form, where the two constants of integration have been reabsorbed into the arbitrary
normalization A:
i x ′
 
A
Z
′
φ(x) ≃ p exp ± p(x )dx . (1.4.74)
p(x) ~

The general structure of this result for φ is easy to interpret from the classical point of
view. The form of the phase arg φ is related to the classical motion of the particle. Indeed,
the gradient of this phase (arg φ)′ (x) must be proportional to the momentum p(x), and
Rx
this implies that arg φ(x) should be proportional to the reduced action p(x′ )dx′ . The
form of the amplitude |φ| is instead related to the classical probability distribution of
finding the particle. Indeed, the infinitesimal probability |φ(x)|2 dx of finding the particle
between x and x + dx must be proportional to the time dt spent by the particle in this
interval, which is given by dx/v(x) in terms of the velocity v(x) and is thus proportional
to dx/p(x) where p(x) is the momentum. This implies that |φ(x)| should be proportional
p
to 1/ p(x).

1.5 The semiclassical wave function

The result (1.4.74) represents the general form taken by the wave function in the semiclas-
sical approximation, whenever this approximation is defined. The two possible signs in the
exponent define two linearly independent solutions of the semiclassical Schrödinger equa-
tion. The general solution is then given by a linear combination of these two independent
solutions with arbitrary coefficients A+ and A− :
 Z x
i x ′
  
A+ i A−
Z
′ ′ ′
φ(x) = p exp p(x )dx + p exp − p(x )dx . (1.5.75)
p(x) ~ p(x) ~
p
From the definition of the classical momentum p(x) = 2m(E − V ), we see that the
semiclassical wave function has a radically different behavior depending on whether the
energy E is larger, smaller or equal to the potential V at the point under consideration, so
that the momentum is correspondingly real, imaginary or vanishing. Let us then consider
these three cases separately.

1.5.1 Classically allowed regions

In the regions of space x where E > V (x), which are classically allowed, the particle has a
positive kinetic energy and p(x) is real. The semiclassical wave function is correspondingly
an oscillating function and the density of probability of finding the particle in such a region
is sizable. It is then convenient to rewrite the wave function in terms of the real and
positive quantum-mechanical wave number k(x) that locally corresponds to the classical
momentum p(x), which is defined as
1 1p
k(x) = p(x) = 2m(E − V (x)) . (1.5.76)
~ ~

13
The general solution in this region can then be written as the superposition of progressive
and a regressive waves with amplitudes weighted by two arbitrary coefficients B+ and B− :
Z x   Z x 
B+ ′ ′ B− ′ ′
φ(x) = p exp i k(x )dx + p exp −i k(x )dx . (1.5.77)
k(x) k(x)
Alternatively, one may write this solution as a standing wave with amplitude weighted by
an arbitrary coefficient B and phase determined by an arbitrary angle δ:
Z x 
B ′ ′
φ(x) = p sin k(x )dx + δ . (1.5.78)
k(x)

1.5.2 Classically forbidden regions

In the regions of space x where E < V (x), which are classically forbidden, the particle
formally has a negative kinetic energy and p(x) is imaginary. The semiclassical wave func-
tion is correspondingly an exponentially localized function and the density of probability
of finding the particle in such a region is tiny, although not zero. It is then convenient to
rewrite the wave function in terms of a real and positive quantum-mechanical penetration-
number β(x) that locally corresponds to the classical momentum p(x), which is defined
as
i 1p
β(x) = − p(x) = 2m(V (x) − E) . (1.5.79)
~ ~
The general solution in this region can then be written as the superposition of decreasing
and increasing exponentials with amplitudes weighted by two arbitrary coefficients D↓
and D↑ :
 Z x  Z x 
D↓ ′ ′ D↑ ′ ′
φ(x) = p exp − β(x )dx + p exp β(x )dx . (1.5.80)
β(x) β(x)

1.5.3 Classical turning points

Finally, at the points x0 where E = V (x0 ), which are turning points of the classical
trajectories, the particle has vanishing kinetic energy and p(x0 ) is zero. The semiclassical
wave function is correspondingly ill defined and the probability of finding the particle in
such points cannot be computed. However, at such points the semiclassical approximation
clearly breaks down and the approximate result derived for the wave function is in fact
not reliable, since when p = 0 and F = −V ′ 6= 0 the condition (1.3.64) is clearly brutally
violated.

1.5.4 Matching strategy

We thus learn that the semiclassical approximation of the wave function is never reliable
in the neighborhood of turning points, whereas far away form these turning points it might
be reliable under suitable circumstances and behaves respectively as an oscillating wave
and a damped profile in classically allowed and classically forbidden regions sufficiently far

14
away from the turning points. At first sight, one would then be tempted to conclude that
this matter of fact represents a severe limitation on the application of the semiclassical
approximation to the standard problems encountered in quantum mechanics, since in those
problems the basic physical properties of the system precisely emerge from the constraints
imposed by the transition between oscillatory behavior in classically allowed regions and
damped behavior in classically forbidden regions across turning points.
Happily, there exists a general way out of this problem, which provides a method to
derive the glueing conditions between the semiclassical wave functions that are valid in
classically allowed and forbidden regions sufficiently apart from a turning point. The idea
consists in noticing that in the immediate neighborhood of a turning point, where the
semiclassical approximation is not reliable, it is possible to use another approximation to
derive in a reliable way the behavior of the wave function, at least under the assumption
that the potential is a sufficiently slowly varying function. This consists in approximating
the potential in such a neighborhood with the linear potential defined by its first order
expansion around the turning point:

V (x) ≃ V (x0 ) + V ′ (x0 )(x − x0 ) ≃ E − F (x0 )(x − x0 ) . (1.5.81)

It turns out that the Schrödinger equation in such a linear potential can be solved exactly,
as will be described in next subsection, and this allows to determine the true behavior of
the wave function in the neighborhood of a turning point.
The basic question is then whether it is possible to reliably glue the approximate
oscillatory and damped wave functions obtained in the semiclassical approximation in
classically allowed and forbidden regions away from turning points with the wave function
obtained in the linearized potential approximation close to the fixed point. For this to
be possible, there should be an overlap among the validity regions of these two different
approximation. It turns out that in most of the situations this is indeed the case. To
argue this, let us start by imagining to be sufficiently close to the turning point for the
linearization of the potential to be a reliable approximation. In such a region, we can
compute the size of the momentum p(x) as

|p(x)| ≃ |p′ (x)(x − x0 )| ≃ 2m|F (x)||x − x0 | .

p
(1.5.82)

Inserting this in the necessary condition (1.3.64) for the applicability of the semiclassical
approximation, we see that this requires that |x − x0 | ≫ 21 ~|m~F (x)|−1/3 ≫ ~/|p(x)|,
which in terms of the De Broglie wave length λ̄ = ~/p reads:
1
|x − x0 | ≫ |λ̄(x)| . (1.5.83)
2
On the other hand, the other necessary condition (1.3.61) for the validity of the semi-
classical approximation implies that λ̄ must be small. In that situation the condition
(1.5.83) for the applicability of the semiclassical approximation will then usually start to
be satisfied already very close to the turning point, where the linearization of V is still
comfortably reliable, and the regions of validity of the two approximations will indeed
sufficiently overlap to be able to match the two types of wave functions.

15
1.6 Behavior in a linear potential
Let us then study the wave function of a one-dimensional particle subject to a linear po-
tential with overal constant equal to the energy E of the particle and a slope corresponding
to an arbitrary force F0 = F (x0 ):

V (x) = E − F0 (x − x0 ) . (1.6.84)

The stationary Schrödinger equation for the wave function φ(x), which can be written as
φ′′ + 2m/~2 (E − V )φ = 0, then becomes:

d2 2mF0
φ(x) + (x − x0 )φ(x) = 0 . (1.6.85)
dx2 ~2
It is then convenient to change from the variable x, which has dimensions of a length and
where the turning point is located in x0 , to a new variable z, which is dimensionless and
and where the turning point is located in 0. Since the quantity 2mF0 /~2 has dimensions
of a length to the power −3, the appropriate change of variable is:
1/3
2mF0
z = − sign(F0 ) (x − x0 ) . (1.6.86)
~2
It follows that
−1/3
d 2mF0 d
= − sign(F0 ) . (1.6.87)
dz ~2 dx
In this new variable, the stationary Schrödinger equation reduces to the simple ordinary
differential equation
d2
φ(z) − zφ(z) = 0 . (1.6.88)
dz 2
The solution of this equation can be found in integral form by using the method of
Laplace. This consists in writing the wave function φ(z) depending on the real variable z
in terms of an integral along some path γ in the complex plane involving an other function
ϕ(s) depending on a complex variable s. More precisely, we look for solutions of the form
1
Z
φ(z) = ϕ(s)ezs ds . (1.6.89)
2πi γ

Inserting this into eq. (1.6.88), one deduces that

Z
(s2 − z)ϕ(s)ezs ds = 0 . (1.6.90)
γ

Rewriting now zezs = d(ezs )/ds in the integrand and integrating by parts, this can be
equivalently rewritten in the following form, where s1 and s2 denote the end-points of the
path γ:
Z  
d s2
ϕ(s) + s ϕ(s) ezs ds − ϕ(s)ezs = 0 .
2
(1.6.91)
γ ds s1

16
It is now clear that we can find solutions of this equation by choosing ϕ in such a way
that the integrand of the bulk term vanishes, and the path γ in such a way that the
boundary term vanishes. The first requirement implies that ϕ(s) should satisfy the first
order differential equation
d
ϕ(s) + s2 ϕ(s) = 0 , (1.6.92)
ds
whose general solution is determined modulo an arbitrary normalization A and reads:
3 /3
φ(s) = A e−s . (1.6.93)

The second requirement implies that the function ϕ(s)ezs should take the same value at
s1 and s2 :
s2
ϕ(s)ezs = 0. (1.6.94)
s1

Without loss of generality, we can focus on open infinite paths where the end-points s1 and
s2 are at infinity in some direction in the complex s plane, and require that the function
ϕ(s)ezs tends to 0 at the infinite end-points of γ. Using the explicit form derived for ϕ(s),
this implies that:
3 /3
lim ezs−s = 0. (1.6.95)
s→s1,2

Writing s1,2 = |s1,2 |eiα1,2 and taking the modulus |s1,2 | to infinity, this condition is realized
only if Re s31,2 > 0 and thus constrains the phase α1,2 to be such that cos(3α1,2 ) > 0. This
allows for three disjoints intervals of values for α1,2 within ]−π, π[ , namely:
i 5π π h i π πh i π 5π h
I− = - , - , I0 = - , , I+ = , . (1.6.96)
6 2 6 6 2 6
Putting everything together, we finally find that the general solution for the wave function
φ(z) is given by:

A
Z
3
φ(z) = ezs−s /3 ds , (1.6.97)
2πi γ

where γ is an open infinite path with starting point s1 and ending point s2 in any of the
regions of the complex plane defined by the intervals I− , I0 , I+ , as shown in fig. 1.1:

γ = infinite open path with end-points in any of the regions I− , I0 , I+ . (1.6.98)

It is straightforward to verify that there are as expected two independent solutions,

corresponding to two independent choices for γ. Indeed, a basis of possible choices for γ
is given by the paths γ−+ , γ−0 and γ+0 going respectively from I− to I+ , from I− to I0
and from I+ to I0 . However, this basis is not linearly independent, because the difference
of the paths γ−0 and γ+0 is equivalent to the path γ−+ . Schematically this means that

17
 Im s

I+

I0

0 Re s

I-

Figure 1.1: Plot of the three domains I0 , I+ and I− .

γ−+ ≃ γ−0 − γ+0 . A convenient choice for the two linearly independent paths is then given
by γ−+ and the sum of the paths γ−0 and γ+0 :

γA ≃ γ−+ , (1.6.99)
γB ≃ γ−0 + γ+0 . (1.6.100)

The corresponding solutions for the wave function are given by the so-called Airy functions
of first and second kind, which are more precisely defined as:
1
Z
3
Ai(z) = ezs−s /3 ds , (1.6.101)
2πi γ−+
1 1
Z Z
3 3
Bi(z) = ezs−s /3 ds + ezs−s /3 ds . (1.6.102)
2π γ−0 2π γ+0

A more explicit representation of these functions in terms of real integrals can be obtained
by choosing special representatives for the paths γ−+ , γ−0 and γ+0 in terms of straight
lines. Their asymptotic behaviors are instead most easily derived by using the saddle-
point or steepest-descent method. This consists in choosing the complex path defining
the function in such a way that this passes exactly through the special points where the
exponent zs − s3 /3 of the integrand becomes stationary and moreover with an orientation
which corresponds to the maximal rate of variation of the whole integrand. In this way,
the main contribution to the integral arises from regions close to these stationary points,
and when z → ∓∞ the approximate value of the integral can be obtained from a simple
Gaussian integral based on the quadratic approximation for the phase of the integrand

18
 √
around stationary points. In our case, there are always two stationary points at s = ± z,
but their location depends on the sign of z. For z → −∞, z is negative and the two saddle
p
points are on the imaginary axis at s = ±i |z|. For z → ∞, z is instead positive and the
p
two saddle points are on the real axis at s = ± |z|.
For the first solution Ai(z), one may choose the path γ−+ to tend to the imaginary
axis parametrized by s = it with t ∈ ] − ∞, +∞[ . One then arrives at the following
representation:
Z +∞
1 3
Ai(z) = ei(zt+t /3) dt
2π −∞
1 +∞
Z
cos zt + t3 /3 dt .


= (1.6.103)
π 0

The asymptotic behavior for z → −∞ can be obtained by choosing γ−+ to pass through
p
to both of the saddle points at s = ±i |z|. Dropping the details, one finds:
 
1 2 3/2 π
Ai(z) ≃ √ sin |z| + , z ≪ 0. (1.6.104)
π|z|1/4 3 4

The asymptotic behavior for z → ∞ can be obtained by choosing the path γ−+ in such
p
a way that it passes through one of the saddle points at s = ± |z|. Dropping again the
details, one finds:

1/2 − 2 z 3/2
Ai(z) ≃ √ e 3 , z ≫ 0. (1.6.105)
πz 1/4

The function Ai(z) is therefore an oscillating wave in the classical allowed region z < 0 and
an exponentially damped function in the classically forbidden region z > 0. Its behavior
across the turning point at z = 0 is shown in fig. 1.2.
For the second solution Bi(z), one may choose the paths γ±0 to be the union of the
positive real axis, parametrized by s = t with t ∈ ]0, +∞[ , and respectively the positive
and negative imaginary axis with suitable orientation, parametrized by s = ∓it with
t ∈ ]−∞, 0[ . One then arrives at the following expression:
+∞ 0
1 i
Z Z
zt−t3 /3 3 /3)
Bi(z) = e dt − e−i(zt+t dt
2π 0 2π −∞
Z +∞ Z 0
1 3 i 3
+ ezt−t /3 dt + ei(zt+t /3) dt
2π 0 2π −∞
Z +∞ Z +∞
1 3 1
ezt−t /3 dt + sin zt + t3 /3 dt .


= (1.6.106)
π 0 π 0

The asymptotic behavior for z → −∞ can be obtained by choosing γ−0 and γ+0 to
p
respectively pass through the saddle points at s = ∓i |z|. One finds:
 
−1 2 3/2 π
Bi(z) ≃ √ sin |z| − , z ≪ 0. (1.6.107)
π|z|1/4 3 4

19
The asymptotic behavior for z → +∞ can be obtained by choosing the paths γ−0 and γ+0
p
in such a way that they both pass through one of the saddle points at s = ± |z|. One
finds:
1 2 3/2
Bi(z) ≃ √ 1/4 e 3 z , z ≫ 0 . (1.6.108)
πz

The function Bi(z) is therefore an oscillating wave in the classical allowed region z < 0 and
an exponentially growing function in the classically forbidden region z > 0. Its behavior
across the turning point at z = 0 is shown in fig. 1.2.

ΦHzL

1.5
BiHzL

0.5

AiHzL
-6 -4 -2 0 2 z

-0.5

Figure 1.2: Plot of the two Airy functions Ai(z) and Bi(z).

Summarizing, the general solution for the wave function in the neighborhood of a
turning point is a generic linear combination of the two Airy functions Ai(z) and Bi(z)
with arbitrary coefficients CA and CB :

φ(z) = CA Ai(z) + CB Bi(z) . (1.6.109)

The asymptotic behaviors of such a solution for z → ∓∞ are moreover given by:
   
CA 2 3/2 π CB 2 3/2 π
φ(z) ≃ √ sin |z| + −√ sin |z| −
π|z|1/4 3 4 π|z|1/4 3 4
C+ 2 3/2 C − 2 3/2
≃ √ e−i 3 |z| + √ ei 3 |z| , z ≪ 0 , (1.6.110)
π|z|1/4 π|z|1/4
and
CA /2 2 3/2 CB 2 3/2
φ(z) ≃ √ 1/4 e− 3 z + √ 1/4 e 3 z , z ≫ 0 , (1.6.111)
πz πz
where we have defined
1 iπ π
 1  −i π π

C+ = e 4 CA + e−i 4 CB , C− = e 4 CA + ei 4 CB . (1.6.112)
2 2
An interesting check on the behaviors of the two independent solutions can be done
by recalling that the Wronskian φA φ′B − φ′A φB of two independent solutions φA and φB of

20
the Schrödinger equation must be constant. In our case, this Wronskian is equal to 1/π,
as a result of the following property of Airy functions:
1
Ai(z)Bi′ (z) − Ai′ (z)Bi(z) = . (1.6.113)
π
It is straightforward to checked that this relation is indeed satisfied by the above-derived
asymptotic behaviors of the Airy functions for z → ±∞.
It is worth mentioning that the two asymptotic behaviors in the classically allowed
region where z → −∞ and in the classically forbidden regions where z → +∞ are actually
related by an analytic continuation and both descend from the asymptotic behavior of the
wave function φ(z) for |z| → +∞ in the complex z plane. As a result, the matching
between these two asymptotic behaviors can actually be determined without ever needing
the exact behavior of the wave function close to the turning point. The idea is to move
from z = ρ to z = −ρ, with ρ ≫ 1 real and positive, on a semicircle avoiding the origin
in the complex z plane, parametrized by z = ρe±iφ with φ ∈ [0, π]. Taking this point of
view, it becomes clear that when moving from the classically forbidden region z ≫ 0 to the
classically allowed region z ≪ 0, the real exponents ± 23 z 3/2 map to the imaginary phases
±i 23 |z|3/2 and the prefactor z −1/4 maps to the prefactor |z|−1/4 times some additional
π
phase factors of the form e±i 4 . But determining the precise connection formulae between
the two kinds of asymptotic behaviors is not completely trivial even with this method,
because while phases factors of different signs are on the same footing, real exponents of
different signs have hierarchically different magnitudes.

1.7 Matching conditions across turning points

The results that we have obtained clearly show that the exact wave function around a
turning point has a behavior which as expected interpolates from an oscillatory behavior
in the classically allowed region to an exponentially damped behavior in the classically
forbidden region. It is now straightforward to verify that these behaviors away from the
turning point are precisely of the general form obtained for the wave functions in such
regions in the semiclassical approximation. This allows to derive matching conditions for
the semiclassical behaviors on the two different sides of a turning point.
To work out the details, we need to switch back from the dimensionless variable z to
the original coordinate x, and figure out what the quantities k(x) and β(x) look like. For
concreteness, we shall study separately the two cases of turning points with increasing
and decreasing potential, where F0 < 0 and F0 > 0. Moreover, it will be convenient to
introduce new normalization constants C0A , C0B related to CA , CB by the relations
1/6
1 2mF0
C0A,B = √ CA,B . (1.7.114)
π ~2
We also define in the same way as before
1 iπ π
 1  −i π π

C0+ = e 4 C0A + e−i 4 C0B , C0− = e 4 C0A + ei 4 C0B . (1.7.115)
2 2

21
1.7.1 Turning points with increasing potential

Consider first the case of increasing potential with F0 < 0, as shown in fig. 1.3. In this
case, the relation between z and x reads:
1/3
2mF0
z= (x − x0 ) . (1.7.116)
~2

VHxL

E + ÈF0 È Hx - x0 L
E

x0 x

Figure 1.3: Turning point with an increasing potential.

In the classically allowed region, where z < 0 and therefore x < x0 , one easily computes
that the wave number is locally given by
1/2
2mF0
k(x) = (x0 − x)1/2
~2
1/3
2mF0
= |z|1/2 , (1.7.117)
~2

and its integral yields

x0
2mF0 1/2 x0
2 2mF0 1/2
Z Z
′ ′
k(x )dx = (x0 − x′ )1/2 dx′ = (x0 − x)3/2
x ~2 x 3 ~2
2
= |z|3/2 . (1.7.118)
3
Using these results, we see that the wave function far inside the allowed region, which is
determined by the asymptotic behavior (1.6.110), can be rewritten in a way that manifestly
matches the general form (1.5.78) of the semiclassical wave function in a classically allowed
region:
 Z x0   Z x0 
C0A ′ ′ π C0B ′ ′ π
φ(x) ≃ p sin k(x )dx + −p sin k(x )dx − (1.7.119)
k(x) x 4 k(x) x 4
 Z x0   Z x0 
C0+ ′ ′ C0− ′ ′
≃ p exp −i k(x )dx + p exp i k(x )dx , x ≪ x0 .
k(x) x k(x) x

22
Notice that C0+ and C0− control in this case the progressive and regressive components
Rx
of the wave, since the quantity x 0 k(x′ )dx′ decreases when x increases.
In the classically forbidden region, where z > 0 and therefore x > x0 , one easily finds
that the penetration factor is locally given by
1/2
2mF0
β(x) = (x − x0 )1/2
~2
1/3
2mF0
= z 1/2 , (1.7.120)
~2

and its integral yields

x 1/2 Z x 1/2
2mF0 2 2mF0
Z
β(x′ )dx′ = (x′ − x0 )1/2 dx′ = (x − x0 )3/2
x0 ~2 x0 3 ~2
2
= z 3/2 . (1.7.121)
3
Using these results, we see that the wave function far inside the forbidden region, which
is determined by the asymptotic behavior (1.6.111), can be rewritten in a way that mani-
festly matches the general form (1.5.80) of the semiclassical wave function in a classically
forbidden region:
 Z x  Z x 
C0A /2 ′ ′ C0B ′ ′
φ(x) ≃ p exp − β(x )dx + p exp β(x )dx , x ≫ x0 .(1.7.122)
β(x) x0 β(x) x0

1.7.2 Turning points with decreasing potential

Consider next the case of decreasing potential with F0 > 0, as shown in fig. 1.4. In this
case, the relation between z and x reads:
1/3
2mF0
z= (x0 − x) . (1.7.123)
~2

In the classically allowed region, where z < 0 and therefore x > x0 , the wave number
is given by
1/2
2mF0
k(x) = (x − x0 )1/2
~2
1/3
2mF0
= |z|1/2 , (1.7.124)
~2

and its integral yields

x
2mF0 1/2 x
2 2mF0 1/2
Z Z
′ ′
k(x )dx = (x′ − x0 )1/2 dx′ = (x − x0 )3/2
x0 ~2 x0 3 ~2
2
= |z|3/2 . (1.7.125)
3

23
 VHxL

E - ÈF0 È Hx - x0 L

x0 x

Figure 1.4: Turning point with an decreasing potential.

Using these results, we see that the wave function far inside the allowed region, which is
determined by the asymptotic behavior (1.6.110), can be rewritten in a way that manifestly
matches the general form (1.5.78) of the semiclassical wave function in a classically allowed
region:
Z x  Z x 
C0A ′ ′ π C0B ′ ′ π
φ(x) ≃ p sin k(x )dx + −p sin k(x )dx − (1.7.126)
k(x) x0 4 k(x) x0 4
 Z x  Z x 
C0+ ′ ′ C0− ′ ′
≃ p exp −i k(x )dx + p exp i k(x )dx , x ≫ x0 .
k(x) x0 k(x) x0

Notice that C0+ and C0− control in this case the regressive and progressive components
Rx
of the wave, since the quantity x0 k(x′ )dx′ increases when x increases.
In the classically forbidden region, where z > 0 and therefore x < x0 , the penetration
factor is given by
1/2
2mF0
β(x) = (x0 − x)1/2
~2
1/3
2mF0
= z 1/2 , (1.7.127)
~2
and its integral yields

2mF0 1/2 x0 2 2mF0 1/2

Z x0 Z
′ ′ ′ 1/2 ′
β(x )dx = (x 0 − x ) dx = (x0 − x)3/2
x ~2 x 3 ~2
2
= z 3/2 . (1.7.128)
3
Using these results, we see that the wave function far inside the forbidden region, which is
determined by the asymptotic behavior (1.6.111) can be rewritten in a way that matches
the general form (1.5.80) of the semiclassical wave function in a classically forbidden
region:
 Z x0   Z x0 
C0A /2 ′ ′ C0B ′ ′
φ(x) ≃ p exp − β(x )dx + p exp β(x )dx , x ≪ x0 (.1.7.129)
β(x) x β(x) x

24
Chapter 2

One-dimensional problems in the

semiclassical approximation

In this chapter, we shall apply the general results derived in the previous chapter for the
semiclassical form of the wave function to study a number of standard one-dimensional
problems. We will first study the spectrum of localized bound states admitted by a generic
potential well and show how the Bohr-Sommerfed quantization rules naturally emerge
in the semiclassical approximation. We will then study the reflection and transmission
coefficients for delocalized scattering states for a generic potential barrier, and display how
the quantum mechanical tunnel effect is captured by the semiclassical approximation. We
will finally study the properties of the quasi stationary states allowed by a generic potential
well associated to a finite barrier, and derive the energy spectrum and the life time of such
states in the semiclassical approximation. In all these cases, we will moreover discuss how
the results can be understood in terms of the quantities characterizing the classical motion.
To conclude, we will also present a systematic approach to deal with more complicated
potentials involving several wells and barriers, and use it to illustrate some interesting
phenomena that can occur in these settings, like for instance level splitting in double wells
or resonance and opacity phenomena in double barriers.

2.1 Energy levels in a generic potential well

One of the most interesting and important applications of the semiclassical approximation
is the study of the energy levels of a generic one-dimensional potential well with arbitrary
shape. This directly leads to the Bohr-Sommerfeld quantization conditions, which were
postulated in the early days of quantum mechanics but are then understood to really
emerge and hold only in the semiclassical approximation.
Consider a particle moving in a one-dimensional potential well V (x), with some definite
energy E. There are then two classical turning points x1 and x2 with x1 < x2 where
V (x1 ) = V (x2 ) = E and V ′ (x1 ) < 0, V ′ (x2 ) > 0, as depicted in fig. 2.1. This defines
a finite central region II = ]x1 , x2 [ that is classically allowed and two infinite boundary

25
regions I = ]−∞, x1 [ and III = ]x2 , +∞[ that are classically forbidden. The semiclassical
wave function must be an oscillating wave in the classically allowed region II and an
exponentially suppressed function in the classically forbidden regions I and III. The
condition that the wave function should vanish at x → ±∞ selects one particular solution
in the two forbidden regions I and III. The continuation of these semiclassical behaviors
across the two turning points, into the region x ∈ ]x1 , x2 [ , can then be done as explained
in previous section. This selects one particular solution in the classically allowed region
and also imposes a quantization condition on the spectrum of allowed energies E, which
should take discrete values parametrized by an integer quantum number related to the
number of nodes of the wave function.

VHxL

I x1 II x2 III x

Figure 2.1: Turning points in a potential well.

2.1.1 Evaluation of the semiclassical wave function

By applying the reasoning of last section to the first classical turning point at x1 where
the potential is decreasing, we deduce that the wave function in the regions I and II
separated by this turning point takes the forms (1.7.129) and (1.7.126), with arbitrary
coefficients now renamed C1A and C1B . The boundary condition that the wave function
should vanish for x → −∞ implies that C1B = 0, and there is thus only one non-vanishing
coefficient C1A 6= 0, so that:
 Z x1 
C1A /2 ′ ′
φI (x) ≃ p exp − β(x )dx , (2.1.1)
β(x) x
Z x 
C1A ′ ′ π
φII (x) ≃ p sin k(x )dx + . (2.1.2)
k(x) x1 4

By applying similarly the reasoning of last section to the second turning point at x2 where
the potential is increasing, we deduce that the wave function in the regions II and III
separated by this turning point takes the forms (1.7.119) and (1.7.122), with arbitrary
coefficients now called C2A and C2B . The boundary condition that the wave function
should vanish for x → +∞ implies that C2B = 0, and there is thus only one non-vanishing

26
coefficient C2A 6= 0, so that:
 Z x2 
C2A ′ ′ π
φII (x) ≃ p sin k(x )dx + , (2.1.3)
k(x) x 4
 Z x 
C2A /2 ′ ′
φIII (x) ≃ p exp − β(x )dx . (2.1.4)
β(x) x2

We now have to require that the two different expressions derived above for φII should be
identical. The corresponding condition is most easily written down by writing:
Z x2 Z x
′ ′
k(x )dx = J − k(x′ )dx′ , (2.1.5)
x x1

where
Z x2
J= k(x′ )dx′ . (2.1.6)
x1

Rewriting then (2.1.3) and (2.1.2) in terms of exponentials and using the above relation,
we see that they match if
Z x   Z x 
1 −i π4 ′ ′ 1 i π4 ′ ′
C1A e exp i k(x )dx + C1A e exp −i k(x )dx
2 x1 2 x1
 Z x  Z x 
1 −i π4 iJ ′ ′ 1 i π4 −iJ ′ ′
= C2A e e exp −i k(x )dx + C2A e e exp i k(x )dx . (2.1.7)
2 x1 2 x1

This condition is identically satisfied for all x ∈ ]x1 , x2 [ if and only if the coefficients of the
progressive and regressive waves both match. This yields the following two conditions:
C1A
e−2iJ = −1 , = ie−iJ . (2.1.8)
C2A
For future reference, notice that these can also be rewritten in an equivalent real form by

−1
using the identities cot J = i 1 + e−2iJ 1 − e−2iJ and sin J = − 2i e−iJ e2iJ − 1 :

C1A
cot J = 0 , = sin J . (2.1.9)
C2A
The solutions of these conditions are labeled by a non-negative integer number n = 0, 1, · · ·
and read:
 
1 C1A
J = n+ π, = (−1)n , n = 0, 1, · · · . (2.1.10)
2 C2A
Summarizing, there is an infinite number of stationary modes labelled by an integer n,
whose wave functions are determined in terms of a single overall constant C. Choosing
the parametrization C1A = C and C2A = (−1)n C, these wave functions are given by
  Z x1 
C ′ ′
φI (x) ≃ p exp − β(x )dx ,



2 β(x)



 x
Z x 
C π


′ ′
φII (x) ≃ p sin k(x )dx + , (2.1.11)

 k(x) x1 4
 Z x
(−1)n C

 
′ ′

 φIII (x) ≃ p exp − β(x )dx ,



2 β(x) x2

27
with the energy quantization condition
Z x2  
′ ′ 1
k(x )dx = n + π. (2.1.12)
x1 2

It is straightforward to verify that the quantum number n corresponds to the number of

nodes possessed by the wave function in the classically allowed region.
The overall constant C in the wave function can be fixed by requiring this to be properly
normalized. Since the classically allowed region is finite, the wave function is normalizable.
In the semiclassical approximation, one can neglect the contribution coming from the
classically forbidden regions, since in those regions the wave function is exponentially
suppressed. One can moreover evaluate the contribution to the total probability from
the classically allowed region by replacing the square of the sine function by its average
value 1/2, since the wave length must be much smaller that the size of the well for the
semiclassical approximation to hold. With these approximations, one finds:
Z +∞ Z x2
1 x2 dx
Z
2 2
|φ(x)| dx ≃ |φII (x)| dx ≃ |C|2
−∞ x1 2 x1 k(x)
~2 ∂J
≃ |C|2 , (2.1.13)
2m ∂E
where in the last step we have used the fact that from the definition of k it follows that
∂k/∂E = m/(~2 k) and therefore:
∂J m x2 dx
Z
= 2 . (2.1.14)
∂E ~ x1 k(x)

Requiring the above result to be unity fixes the constant C to be, modulo an irrelevant
phase:
√ 
2m ∂J −1/2

C≃ . (2.1.15)
~ ∂E

In order to assess the validity range of the above general results, we have now to
recall that the semiclassical approximation is valid only at points whose distance from
classical turning points is much bigger than the local De Broglie wave length λ̄(x) = 1/k(x)
characterizing the wave function. This implies in particular that the wave function must
have a large number of nodes within the classically allowed region, meaning that the
quantum number n should be large:

n ≫ 1. (2.1.16)

It is only in this limit of large quantum number that the semiclassical approximation
is justified. In this regime, the 1/2 shift obtained in the quantization condition (2.1.12)
formally represents a small effect. In practice, however, it happens quite often that (2.1.12)
yields a rather accurate result even for the low-lying energy levels, with the 1/2 shift
playing a quantitatively important role. In some particular cases, it even happens that
the semiclassical approximation accidentally yields the exact spectrum of energy levels.

28
2.1.2 Quantum spectrum and classical interpretation

The way in which the system behaves at the quantum level admits a nice interpretation
based on the classical trajectory. Recall for this that the classical motion is periodic, and
a full cycle starts at x1 reaches x2 and goes back to x1 . During this cycle, the classically
allowed region II is therefore spanned twice. Since the classical momentum is given by
p(x) = mdx/dt, the period T and the angular frequency ω of this motion are given by
dx
I I
T = dt = m , (2.1.17)
p(x)
2π
ω= . (2.1.18)
T
The functions J and ∂J/∂E that determine the quantization condition and the normal-
ization of the semiclassical wave function can then be rewritten as follows in terms of
classical quantities:
1
I
J= p(x)dx , (2.1.19)
2~
∂J m dx T 1
I
= = = . (2.1.20)
∂E 2~ p(x) 2~ 2hω

Let us first discuss the interpretation of the quantization condition. Using (2.1.19),
it is straightforward to recognize that (2.1.12) can be rewritten as a Bohr-Sommerfeld
type quantization condition on the energy E, except for the shift by 1/2 in the quantum
number n:
 
1
I
p(x) dx = n + h. (2.1.21)
2
The 1/2 shift emerges automatically and is directly linked to the uncertainty principle
that is inherent to the exact formulation of quantum mechanics. It corresponds to a zero-
point energy and reflects the fact that for a bound quantum system even the ground state
corresponding to n = 0 cannot have an exactly vanishing momentum, because of the finite
uncertainty on its position. However, the semiclassical approximation is strictly speaking
valid only in the limit of large quantum numbers n ≫ 1, and in this limit the shift represent
a quantitatively unimportant effect that can be neglected. With this consideration, the
condition (2.1.21) allows to interpret quantum states as classical states with a particular
H
quantized value for the quantity p dx, which corresponds to the area delimited by the
periodic classical motion in phase space. The Bohr-Sommerfeld condition then means that
this area should be quantized in units of a minimal area in phase space determined by
the fundamental constant h. Moreover, since the number n corresponds to the number of
different quantum states with energies less that the energy E associated to the considered
H
value of p dx, we can conclude that to each distinct quantum state we can associate a
cell in phase space of area h. The number of quantum states associated to an area ∆p∆x
in phase space is therefore given by:
∆p∆x
≃ # of quantum states . (2.1.22)
h

29
The Bohr-Sommerfeld quantization condition (2.1.21) also allows to deduce some general
properties of the energy spectrum in terms of the quantities characterizing the classical
motion. In particular, the energy separation ∆E between two contiguous levels with
∆n = 1 is much smaller than their average energy E associated to some large n ≫ 1.
Applying then ∆E ∂/∂E = ∆n ∂/∂n to the two sides of the quantization condition, one
deduces that
∂p
I
∆E (x) dx ≃ h . (2.1.23)
∂E
But ∂E/∂p = p/m, so that
∂p dx 2π
I I
(x) dx = m =T = . (2.1.24)
∂E p(x) ω
Its then follows that:

∆E = ~ ω . (2.1.25)

This means that in the region of the spectrum where the semiclassical approximation holds
true, the separation ∆E between levels is approximately given by ~ times the frequency ω
associated to the classical motion with energy E. This implies in particular that transitions
between neighbor levels through the emission or absorption of electromagnetic radiation
will involve photons of frequency ω.
Let us next discuss the interpretation of the normalization of the wave function. Using
(2.1.20), we deduce that the normalization constant (2.1.15) of the semiclassical wave
function can be expressed as
r r
4m 4mω
C≃ ≃ . (2.1.26)
~T h
With this result we can now compute the classical density of probability ρ(x) for finding
the particle at the position x, which is obtained by averaging the semiclassical density of
probability |φ(x)|2 over distances much larger than the wavelength λ(x). In the forbidden
regions, the suppressed exponential can be neglected and ρI,III ≃ 0, while in the allowed
region the square of the sine function can be approximated with its average value 1/2 and
ρII ≃ 2~ |C|2 /p(x). Using the above value of C, one then finds:


 ρI (x) ≃ 0 ,
2m 1 mω 1


ρII (x) ≃ ≃ , (2.1.27)

 T p(x) π p(x)

 ρ (x) ≃ 0 .
III

As a final general comment, let us mention that there exists a general technique to
H
evaluate the phase space integral p(x) dx as a function of the energy, in order to derive the
explicit form of a spectrum for a given potential V (x). This is based on the observation
p
that the phase space integral is equal to twice the integral of p(x) = 2m(E − V (x))
between the turning points x1 and x2 where V (x1 ) = V (x2 ) = E and thus p(x1 ) =

30
p(x2 ) = 0. In the complex plane, where p(z) has a branch cut ranging between the two
turning points x1 and x2 on the real axis, this phase space integral can then be written as
p
the integral of p(z) = 2m(E − V (z)) along a closed contour c encircling the cut. The
integral can then be evaluated by using Cauchy’s theorem for the reversed contour, which
encircles the rest of the complex plane. Assuming that V (z) is analytic everywhere except
the cut, the result of this integral is then simply −2πi times the residue of the function
p(z) at infinity, which is equal to minus the coefficient of the 1/z term in the Laurent
expansion of this function around infinity. In summary, we thus have:
I I

p(x) dx = p(z) dz = −2πi Res p(z) z=∞ . (2.1.28)
c

Of course, in simple cases one may also directly evaluate the result by computing the
original real integral by means of ordinary techniques.

2.1.3 Example

To illustrate the general result derived in this section, let us apply it to the simple case of
the one-dimensional harmonic oscillator, defined by a potential of the form
1
V (x) = mω 2 x2 . (2.1.29)
2
For a given energy E, there are two classical turning points at ±x0 , such that V (±x0 ) = E,
with:
r
2E
x0 = (2.1.30)
mω 2
We can then rewrite
 x 2
V (x) = E . (2.1.31)
x0
The classically allowed region is given by the interval ]−x0 , x0 [ , and the momentum in
this region is given by
p
p(x) = 2m(E − V (x))
√
p
x20 − x2
= 2mE . (2.1.32)
x0
H
To compute the integral p(x) dx, we can deform it in the complex plane and use (2.1.28).
Expanding p(z) around z = ∞, one finds:
√
r
z  x 2
0
p(z) = i 2mE 1−
x z
 0
√
 
i i 1 1
= 2mE z − x0 + O 3 . (2.1.33)
x0 2 z z
It follows that:
 √ hi i
Res p(z) z=∞
= 2mE x0 . (2.1.34)
2

31
Using this in (2.1.28), one then finds:
I

p(x)dx = −2πi Res p(z) z=∞
√
= 2mE πx0 . (2.1.35)

The same result can also be obtained by evaluating directly the original real integral:
I √ Z x0 p 2
x0 − x2
p(x)dx = 2 2mE dx
−x 0
x 0
√
 q
x  x x0
= 2mE x20 − x2 + x0 arcsin
x0 x0 −x0
√
= 2mE πx0 . (2.1.36)

Recalling the definition of x0 , the integral is finally found to take the following form:
2πE
I
p(x)dx = . (2.1.37)
ω
Plugging this result into the condition (2.1.21), one finally deduces that the semiclassical
spectrum coincides in this case with the well-know exact spectrum, namely:
 
1
En = n + ~ω . (2.1.38)
2

2.1.4 Generalizations

The above semiclassical analysis of the spectrum of energy levels in a potential well can
be generalized from the case of one degree of freedom associated to the coordinate x to
an arbitrary number d of degrees of freedom associated to equally many coordinates xi ,
provided the potential V (xi ) is such that the classical motion is periodic and that the
problem is completely separable. After performing the separation of variable, one then
arrives at d independent quantization conditions, which are defined in terms of the vari-
ables xi and their canonically conjugate momenta pi and involve equally many quantum
numbers ni . These quantization conditions have the same form as the one derived above
for a single degree of freedom, but with more general zero-point shifts γi ∈ [0, 1]. For each
i = 1, · · · , d, one gets:
I


pi (x)dxi = ni + γi h . (2.1.39)

Again, the semiclassical approximation holds a priori true only for large quantum numbers,
ni ≫ 1, and in this limit the above conditions reduce to the Bohr-Sommerfeld quantization
rules.

2.2 Tunneling probability through a generic barrier

Another very relevant application of the semiclassical approximation is the study of the
tunneling probability through a generic one-dimensional potential barrier with arbitrary
shape. This leads to a very useful approximate general result for such a probability.

32
 Consider a particle moving in a one-dimensional potential barrier V (x), with some
energy E. There are then two classical turning points x1 and x2 with x1 < x2 where
V (x1 ) = V (x2 ) = E and V ′ (x1 ) > 0, V ′ (x2 ) < 0, as depicted in fig. 2.2. This defines a
finite central region II = ]x1 , x2 [ that is classically forbidden and where the semiclassical
wave function is exponentially suppressed and two infinite boundary regions I = ]−∞, x1 [
and III = ]x2 , +∞[ that are classically allowed and where the semiclassical wave function
is oscillating. By matching these semiclassical behaviors across the two turning points,
one finds a relation between the amplitudes of the waves in the two classically allowed
regions, which allows to compute the transmission coefficient T . More specifically, let
us assume that the particle comes from the left in region I and is then partly reflected
back to the left in region I and partly transmitted to the right in region III. This setup
specifies as usual a definite boundary condition, which allows to solve for the complete
wave function in the three regions in terms of a single overall normalization constant. To
do this, we again use the joining conditions to continue the semiclassical wave function
across turning points.

VHxL

I x1 II x2 III x

Figure 2.2: Turning points in a potential barrier.

2.2.1 Evaluation of the semiclassical wave function

By looking at the first classical turning point at x1 where the potential is increasing,
we deduce that the wave functions in the regions I and II take the form (1.7.119) and
(1.7.122), with arbitrary coefficients C1A and C1B or equivalently C1+ and C1− , which
respectively control the progressive and regressive components in the region I. In this
case there is no boundary condition on the wave function for x → −∞, since in that
asymptotic region there is both an incident and a reflected wave, and there are thus two
independent coefficients C1A 6= 0 and C1B 6= 0, or C1+ 6= 0 and C1− 6= 0, so that:
 Z x1   Z x1 
C1+ ′ ′ C1− ′ ′
φI (x) ≃ p exp −i k(x )dx + p exp i k(x )dx , (2.2.40)
k(x) x k(x) x
 Z x  Z x 
C1A /2 ′ ′ C1B ′ ′
φII (x) ≃ p exp − β(x )dx + p exp β(x )dx . (2.2.41)
β(x) x1 β(x) x1

33
By looking at the second classical turning point at x2 where the potential is decreasing, we
similarly deduce that the wave functions in the regions II and III take the form (1.7.129)
and (1.7.126), with arbitrary coefficients renamed as C2A and C2B or equivalently C2+
and C2− , which respectively control the regressive and progressive components in the
region III. In this case there is a definite boundary condition on the wave function for
x → +∞, since in that asymptotic region one should find only a transmitted wave, and
there is thus only one independent coefficient out of C2A and C2B , or equivalently C2+
and C2− . More precisely, one has to choose C2B = −iC2A , so that the coefficient of the
regressive wave vanishes, C2+ = 0, and the coefficient of the progressive wave is then given
π
by C2− = e−i 4 C2A . One then finds:
 Z x2   Z x2 
C2A /2 ′ ′ iC2A ′ ′
φII (x) ≃ p exp − β(x )dx − p exp β(x )dx , (2.2.42)
β(x) x β(x) x
π
e−i 4 C2A
Z x 
φIII (x) ≃ p exp i k(x′ )dx′ . (2.2.43)
k(x) x2

We must now require that the two different expressions derived for φII (x) coincide. The
corresponding condition is most easily analyzed by rewriting:
Z x2 Z x
′ ′
β(x )dx = K − β(x′ )dx′ (2.2.44)
x x1

where
Z x2
K= β(x′ )dx′ . (2.2.45)
x1

One must then have:

 Z x  Z x 
1
C1A exp − β(x′ )dx′ + C1B exp β(x′ )dx′
2 x1 x1
Z x   Z x 
1 −K ′ ′ K ′ ′
= C2A e exp β(x )dx − iC2A e exp − β(x )dx . (2.2.46)
2 x1 x1

This condition can be satisfied for all x ∈ ]x1 , x2 [ if and only if the coefficients of the
increasing and decreasing exponentials separately match. This leads to the following two
relations among the coefficients:
C1A C1B 1
= −2ieK , = e−K , (2.2.47)
C2A C2A 2
or equivalently
   
C1+ π 1 C1− π 1
= e−i 4 eK + e−K , = −ei 4 eK − e−K , (2.2.48)
C2A 4 C2A 4

Summarizing, there is an infinite number of stationary modes labelled by a continuous

value E for the energy, whose wave functions are determined in terms of a single constant

−1
C. Choosing the parametrization C1+ = C, so that C1− = −i 1− 41 e−2K 1+ 41 e−2K


C,

34
 π
−1 π
−1
C1A = 2 e−i 4 1 + 14 e−2K C, C1B = 21 ei 4 e−2K 1 + 14 e−2K C and for the last one
i π4 −K 1 −2K −1


C2A = e e 1 + 4e C, these wave functions take the form

1 − 14 e−2K
  Z x1   Z x1 
C

′ ′ ′ ′

 φI (x) ≃ p exp −i k(x )dx − i exp i k(x )dx ,
1 + 14 e−2K




 k(x) x x
π

e−i 4
  Z x 
C


 ′ ′

 φII (x) ≃ p exp − β(x )dx
β(x) 1 + 41 e−2K

x1


π (2.2.49)
1 ei 4 e−2K
 Z x 
 ′ ′
 + exp β(x )dx ,
2 1 + 14 e−2K


x1



 Z x
e−K
  
C


′ ′
φ (x) ≃ exp i k(x )dx .

 III

k(x) 1 + 41 e−2K

 p
x2

The overall constant C can be fixed by requiring that the wave function is properly
normalized. But since the classically allowed regions are infinite, the wave function is
not normalizable in the usual sense. One then has to use a δ-function normalization, as
usual with plane waves. We will however not do this explicitly, because it is not needed
to extract the main physical properties of the barrier.
To compute the reflection and transmission coefficients of the barrier, we can now
study the form of the wave function in the asymptotic regions x → ∓∞. In those regions
the potential is assumed to tend to zero, V (x) → 0, and the wave number then tends to
√ Rx Rx
a constant, k(x) → 2mE/~. Moreover, x 1 k(x′ )dx′ decreases and x2 k(x′ )dx′ increases
when x grows. One then deduces that the incident and reflected waves for x → −∞ and
the transmitted wave for x → +∞ are given by
 Z x1 
Ainc ′ ′
φinc (x) ≃ p exp −i k(x )dx , Ainc = C , (2.2.50)
k(x) x
1 − 14 e−2K
 Z x1 
Aref
φref (x) ≃ p exp i k(x′ )dx′ , Aref = −i C, (2.2.51)
k(x) x 1 + 14 e−2K
Z x
e−K

Atra ′ ′
φtra (x) ≃ p exp i k(x )dx , Atra = C. (2.2.52)
k(x) x2 1 + 14 e−2K

For each of these, we can now evaluate the density of probability current, which is defined
to be:
i~  ∗ ′ 
j(x) = − φ φ − φ∗′ φ (x) . (2.2.53)
2m
One can neglect the terms where the derivative acts on the prefactor of the wave function,
since this is proportional to k−1/2 (k−1/2 )′ ∼ k′ /k2 ∼ λ̄′ , and this must be small for the
semiclassical approximation to be valid. One then finds that the currents associated to
the incident, the reflected and the transmitted waves are constant and given by
~ ~ ~
jinc ≃ |Ainc |2 , jref ≃ − |Aref |2 , jtra ≃ |Atra |2 . (2.2.54)
m m m

35
It follows that the reflection and transmission coefficients associated to the potential bar-
rier are given by:

1 − 14 e−2K 2
2
e−K
  
jref jtra
R= ≃ , T = ≃ . (2.2.55)
jinc 1 + 14 e−2K jinc 1 + 14 e−2K

Notice that |jref | + |jtra | = |jinc |, reflecting the usual conservation of the density of proba-
bility current j, and one thus finds:

R + T ≃ 1. (2.2.56)

Notice finally that in most of the situations one has K ≫ 1, and in that limit the reflection
and transmission coefficients of the barrier further simplify to

R ≃ 1 − e−2K , T ≃ e−2K . (2.2.57)

To estimate the validity range of the above general results, we have now to check
under which instances the conditions for the validity of the semiclassical approximation
really hold true. It is not totally straightforward to determine what kind of constraints
this implies on the form of the barrier. On general grounds, one could expect that the
approximation is justified only whenever the wave function in the forbidden region is
strongly suppressed, and this usually implies that the exponent K should be large, as
assumed to arrive at the final expressions (2.2.57):

K ≫ 1. (2.2.58)

In addition, one must require that the potential varies sufficiently slowly. In situations
where instead the potential varies rapidly at a turning point, for instance, the above for-
mula is a priori not valid, because the general gluing rules derived in section 1.7 cannot be
applied. However, it turns out that the corrections to the semiclassical approximation are
in fact quite mild, and consist in an additional non-trivial power-law prefactor multiplying
the exponential suppression factor in the tunneling probability. An interesting case where
this can be verified explicitly is for instance that of barriers with sharp discontinuities.
In those cases, the semiclassical approximation breaks down, but actually only right at
the turning point, since away from that point the potential is slowly varying. It is then
usually possible to solve the problem of joining the oscillating and damped semiclassical
wave functions that are valid on the two sides of such a turning point by simply imposing
the continuity of the wave function and its first derivative. In this way, one can derive a
reliable result including also the prefactor.

2.2.2 Tunneling probability and classical interpretation

Once again, the general results derived by using the semiclassical approximation admit a
nice interpretation in terms of classical trajectories. However, since the tunneling process
is controlled by the behavior of the wave function in the classically forbidden region,
the relevant trajectories are in this case not true classical trajectories in the classically

36
allowed region but fake complex trajectories in the classically forbidden region, defined by
going to complex position x. The basic observation behind this is that the quantity β(x)
entering the definition of the exponential factor K controlling the tunneling probability
is actually defined out of the analytic continuation of the classical momentum to the
classically forbidden region. More precisely, we have defined β(x) = −(i/~)p(x), where
p
p(x) = 2m(E − V (x)) is now purely imaginary. Correspondingly, we can write:
Z x2
i x2 ′
Z
′ ′
K= β(x ) dx = − p(x ) dx′ . (2.2.59)
x1 ~ x1
We conclude from this expression that the real factor K is −i times the imaginary reduced
action associated to the complex trajectory that solves the classical equations of motion
in the classically forbidden region, after extension to the complex plane.
The general result (2.2.57) can be used to draw some general conclusions about the
opacity against tunneling of a generic barrier as a function of its typical height H and
with W . To do so, we parametrize the form of the effective barrier above the energy E in
terms of these typical scales and a dimensionless function f with values of order 1 between
x1 and x2 , in such a way that form:
x
V (x) − E = Hf ( ) (2.2.60)
W
We can then estimate the exponent K by switching to the new dimensionless variable
y = x/W , which is of order unity and ranges between y1 = x1 /W and y2 = x2 /W :
√
1 x2p 2mH x2p
Z Z
K = 2m(V (x) − E) dx = f (x/W ) dx
~ x1 ~ x1
√
2mHW y2p
Z
= f (y) dy . (2.2.61)
~ y1

The integral is now a number of order unity, and the scaling of K with H, W and m is
therefore given by:
m1/2 H 1/2 W
K∝ . (2.2.62)
~

2.2.3 Example

To illustrate the above results, let us apply them to the simple case of a square potential
barrier with height V0 and width L0 :
(
V0 , |x| ≤ L0 /2 ,
V (x) = (2.2.63)
0 , |x| > L0 /2 .
For a given energy E < V0 , the classical turning points are located at x1 = −L0 /2 and
x2 = L0 /2. One then computes
1 L0 /2p
Z
K = 2m(V0 − E) dx ,
~ −L0 /2
p
2m(V0 − E)L0
= . (2.2.64)
~

37
In this simple case, one can work out explicitly the constraints imposed on the validity
regime of this result by the two necessary conditions described in last chapter. Firstly,
the De Broglie wave length of the wave function should be much smaller than the typical
p
size characterizing the system. In this case, this means that |λ̄| ≃ ~/ 2m(V0 − E) should
be much smaller than L0 : |λ̄| ≪ L0 . This implies that K ≫ 1, as expect on more general
grounds. Secondly, the derivative of the potential at the turning points should not be
two large. In our case, this derivative is actually infinite, and the gluing prescriptions of
section 1.7 are therefore strictly speaking not applicable. However, one should neverthe-
less get the correct exponential suppression factor and miss only a relatively unimportant
multiplicative correction prefactor A. We can thus infer that the reflection and trans-
mission coefficients should be given by the following approximate formulae, with K given
by eq. (2.2.64) and A representing an unknown prefactor depending polynomially on the
parameters of the problem:

R ≃ 1 − A e−2K , T ≃ A e−2K . (2.2.65)

In this case, it is actually straightforward to compute the exact wave function and derive
from it the exact expressions for the reflection and transmission coefficients. These are
found to be given by:
1 1
R= −2 , T = . (2.2.66)
1 + (4/A)−1 sinh K 1 + (4/A) sinh2 K

where K is given by eq. (2.2.64) and A is found to be given by

 
E E
A = 16 1− . (2.2.67)
V0 V0

We see that for K ≫ 1, the above result does indeed take the approximate form (2.2.65),
and the prefactor A has as promised only a mild polynomial dependence on the parameters.

2.2.4 Generalizations

The above semiclassical analysis of the probability of tunneling through a potential barrier
can be generalized from the case of one degree of freedom associated to the coordinate x
to an arbitrary number d of degrees of freedom associated to equally many coordinates xi .
The general d-dimensional result has the same general form as the one-dimensional result
derived above, but with an exponent K given by:
Z ~
x2
K = min β(~x′ )dl(~x′ ) . (2.2.68)
~
x1

Pictorially, this means that in the semiclassical approximation a d-dimensional tunneling

process is dominated by a one-dimensional tunneling process that proceeds along the
section of the barrier along which it appears most transparent.

38
2.3 Lifetime of a metastable state
Another interesting application of the semiclassical approximation is the study of the
quasi-localized states arising in a potential well involving a finite rather infinite barrier
against escape at infinity. In such a situation, we expect that due to the possibility of
tunneling through the barrier, there will be no strictly localized states associated to the
well, but only quasi-localized states involving not only a localized component inside the
well but also a delocalized component escaping outside the barrier. If the tunneling effect
is small, we can interpret this situation as leading to metastable localized state with a
finite lifetime against decay through the barrier. The corresponding energy will then be
complex, with a real part controlling the energy of the level and the usual stationary
behavior in time, and the imaginary part controlling the decay width and a damped
behavior in time. Such states are then called quasi-stationary.
Consider a particle moving in one-dimensional potential V (x) consisting of a well with
an infinite barrier on the left but a finite barrier on the right, with some energy E. There
are then three turning points x1 , x2 and x3 with x1 < x2 < x3 where V (x1 ) = V (x2 ) =
V (x3 ) = E and V ′ (x1 ) < 0, V ′ (x2 ) > 0, V ′ (x3 ) < 0, as depicted in fig. 2.3. This defines two
classically allowed regions II = ]x1 , x2 [ and IV = ]x3 , +∞[ , and two classically forbidden
regions I = ]−∞, x1 [ and III = ]x2 , x3 [ . By matching the semiclassical behaviors across
the three turning points, one finds a relation between the amplitudes of the wave function
in the classically forbidden region on the far left and the classically allowed region on
the far right. In region I, the wave function must be exponentially suppressed, but in
region IV we allow the wave function to consist of a progressive wave resulting from
the tunneling of the localized state in the well. This setup specifies a definite boundary
condition, which allows to solve for the complete wave function in the four regions in terms
of a single overall normalization constant. The energy spectrum follows as before from a
quantization condition, but in this case the values are complex, even if the Hamiltonian
is real, simply because one of the boundary condition is complex.

VHxL

I x1 II x2 III x3 IV x

Figure 2.3: Turning points in a potential well with a finite barrier.

39
2.3.1 Evaluation of the semiclassical wave function

The semiclassical wave function can be derived by proceeding as in the cases of a single
well or barrier and exploiting the calculations that have already been done for those cases.
The wave function in the regions I and II around the first turing point, with the boundary
condition that it vanishes at x → −∞, has the same form as for the potential well:
 Z x1 
C ′ ′
φI (x) ≃ p exp − β(x )dx , (2.3.69)
2 β(x) x
  Z x  Z x 
C i π4 ′ ′ −i π4 ′ ′
φII (x) ≃ p e exp −i k(x )dx + e exp i k(x )dx . (2.3.70)
2 k(x) x1 x1

The wave function in the regions II, III and IV across the barrier defined by the turning
points x2 and x3 , with the boundary condition that it reduces to a progressive wave at
x → +∞, has instead the same form as for the potential barrier:

1 − 41 e−2K
  Z x2   Z x2 
D ′ ′ ′ ′
φII (x) ≃ p exp −i k(x )dx − i exp i k(x )dx ,(2.3.71)
k(x) x 1 + 14 e−2K x
π
e−i 4
  Z x 
D ′ ′
φIII (x) ≃ p exp − β(x )dx
β(x) 1 + 41 e−2K x2
i π4 −2K Z x 
1 e e ′ ′
+ exp β(x )dx , (2.3.72)
2 1 + 14 e−2K x2
Z x
e−K
 
D ′ ′
φIV (x) ≃ p exp i k(x )dx . (2.3.73)
k(x) 1 + 41 e−2K x3

where
Z x3
K= β(x′ ) dx′ . (2.3.74)
x2

The two different expressions for the wave function in the region II must now be matched.
The corresponding condition is most easily formulated after rewriting
Z x2 Z x
′ ′
k(x )dx = J − k(x′ )dx′ , (2.3.75)
x x1

where
Z x2
J= k(x′ )dx′ . (2.3.76)
x1

One must then require that:

 Z x  Z x 
1 iπ ′ ′ 1 −i π ′ ′
Ce exp −i k(x )dx + Ce
4 4 exp i k(x )dx
2 x1 2 x1
Z x 1 −2K  Z x
1 − 4e
 
−iJ ′ ′ iJ ′ ′
= De exp i k(x )dx − i D e exp −i k(x )dx . (2.3.77)
x1 1 + 14 e−2K x1

40
The above condition is identically satisfied for every x ∈ ]x1 , x2 [ if and only if the coeffi-
cients of the progressive and the regressive waves both match. These two equations yield
a quantization condition on J and a relation between the two constants C and D:

1 − 14 e−2K C i π4 −iJ
e−2iJ = − 1 −2K , D = 2 e e . (2.3.78)
1 + 4e

Equivalently, these may also be written as:

i C π sin J
cot J = e−2K , = 2 e−i 4 . (2.3.79)
4 D 1 + i cot J
In both of these forms, we see that the quantization condition cannot be satisfied with real
J but only with complex J. As expected, this implies complex values of the energy. The
general solution can again be parametrized in terms of a non-negative integer n = 0, 1, · · ·
and reads:
s
1 − 14 e−2K 1 − 41 e−2K
   
1 i C π
i4 n
J = n+ π + log , = 2 e (−1) . (2.3.80)
2 2 1 + 14 e−2K D 1 + 41 e−2K

Summarizing, the system admits an infinite set of quasi-localized modes with complex
energies E and wave function given in terms of a single overall coefficient C by:
 Z x1 
C

′ ′

 φI (x) ≃ p exp − β(x )dx ,
2 β(x)

x




 Z x 
C π

′ ′

φII (x) ≃ p sin k(x )dx + ,


4

k(x)



 x 1
  Z x
nC
 
 φ (x) ≃ (−1) 1


 ′ ′
III p p exp − β(x )dx
2 β(x) 1 − 1 −4K
e x2 (2.3.81)
 16

x
e−2K
  
i
 Z
β(x′ )dx′ ,

+ exp

 p
2 1 −4K

1 − 16 e x

2



 π
i x
(−1)n C e 4 e−K

   Z 

′ ′
φIV (x) ≃ p exp i k(x )dx ,


 p
2 k(x) 1 −4K
1 − 16 e


 x3

with the quantization condition:

1 − 41 e−2K
Z x2    
′ ′ 1 i
k(x )dx = n + π + log . (2.3.82)
x1 2 2 1 + 41 e−2K

As already explained, the allowed energies E are complex. They are conventionally
written in the following form:
Γ
E = E0 − i . (2.3.83)
2
The states are correspondingly not really stationary. This can be seen by computing the
i
probability density defined by the full wave function ψ(x, t) = e− ~ Et φ(x) describing a

41
state with such a complex energy. One finds:
i
ρ(x, t) = |ψ(x, t)|2 = |e− ~ Et |2 |φ(x)|2
1
= e− ~ Γt |φ(x)|2 . (2.3.84)

We see that the overall normalization decreases exponentially in time, as a result of the
non-zero imaginary part in the energy. This corresponds to the fact that a state that is
initially localized in the barrier can decay by tunneling through the barrier to a delocalized
state. One can then define the lifetime of such an initially localized state as the time scale
of this exponential decay, that is the time by which the probability to find the state
localized in the well decreases by a factor of e. From the above result for the density
of probability, we then conclude that the lifetime is related to the imaginary part of the
energy by the following relation:
~
τ= . (2.3.85)
Γ
The quantity Γ, which has the dimension of an energy, has also a direct physical inter-
pretation. This is related to the fact that when trying to measure the energy E of an
unstable state, by any kind of physical process, the uncertainty relation ∆E∆t ≥ ~2 sets a
limitation on how small an uncertainty ∆E on the result one can get depending on how
large an observation time ∆T one can afford: ∆E ≥ ~/(2∆T ). When the state has a finite
lifetime τ = ~/Γ, one is limited to ∆T < ∼ τ and therefore ∆E ≥ Γ/2. This means that
Γ represents an intrinsic uncertainty on the energy that can be associated to the state.
In a spectroscopy experiment in which the levels are measured from some peaks in the
response of a system to radiation, E0 will then represent the location of the peaks and Γ
their widths. For this reason, Γ is often called the width of the state.
As usual, the validity of the semiclassical approximation normally requires that the
tunneling probability through the barrier is small and therefore

K ≫ 1. (2.3.86)

In that case, the factor controlling the tunneling probability through the finite barrier
is small, e−2K ≪ 1, and one can approximate the right-hand side of the quantization
condition with the simpler expression
1 − 14 e−2K
 
i i
log 1 −2K
≃ − e−2K . (2.3.87)
2 1 + 4e 4
Correspondingly, the imaginary part of the energy is much smaller that the real part, and
one has:

Γ ≪ E0 . (2.3.88)

In such a situation, the state is a long-lived quasi-stationary state, and one can derive an
explicit expression for its lift time. To do this one may expand the left-hand side of the
quantization condition around E0 at first order in powers of Γ:
Γ ∂J
J E
≃J E0
−i . (2.3.89)
2 ∂E E0

42
Plugging back these expressions into the quantization condition, one finally deduces that
the real and imaginer parts of the energy are approximately determined by two indepen-
dent equations:
 
1
E0 : J E0 ≃ n + π, (2.3.90)
2
1 ∂J −1 −2K
 
Γ= e . (2.3.91)
2 ∂E E0

The lifetime is then

 
∂J
τ = 2~ e2K . (2.3.92)
∂E E0

2.3.2 Decay width and classical interpretation

In this case too, the results that we have obtained in the semiclassical approximation
admit a nice interpretation in terms of the quantities characterizing the classical trajectory.
Recall in particular that the period T and the frequency ω are given by
dx
I I
T = dt = m , (2.3.93)
p(x)
2π
ω= , (2.3.94)
T
and that the functions J and ∂J/∂E can then be rewritten as follows in terms of these
quantities:
1
I
J= p(x)dx , (2.3.95)
2~
∂J m dx T 1
I
= = = . (2.3.96)
∂E 2~ p(x) 2~ 2hω

We see that the energy E0 of the levels are determined by the same quantization
condition as for a normal well, namely the Bohr-Sommerfeld type of condition:
 
1
I
p(x) dx = n + h. (2.3.97)
2

The implications and the interpretation of this result are exactly the same as those already
discussed for a simple well.
The width Γ, which is the new physical quantity associated the finite barrier attached
to the well, is instead given by the following expression in terms of classical quantities:
~ −2K
Γ= e . (2.3.98)
T
The corresponding lifetime is then

τ = T e2K . (2.3.99)

43
This result can understood in the following way. We know from the study of the tunneling
through a potential barrier that the probability for a particle at the turning point to
tunnel through it and escape is equal to e−2K ≪ 1. Now, in the classical approximation
the particle undergoes a periodic motion with period T in the well, and during some
large time t ≫ T it therefore arrives at the relevant turning point a number of times
t/T ≫ 1. The probability that the particle decays during this large time is given by
−2K
1 − (1 − e−2K )t/T ≃ 1 − e−t/T e . The lifetime τ should then be equal to the time t for
which this probability becomes equal to 1 − 1/e. This reproduces the result τ = T e2K .

2.4 Multiple wells and barrier

The computations done in the last three sections can be generalized to one-dimensional
situations where the potential involves several wells and barrier. In these more complicated
cases, it is convenient to proceed in the following way. First one determines the general
form of the semiclassical wave function in terms of two independent coefficients. Then one
imposes on it the boundary conditions at infinity that are appropriate to the problem.
Finally, one extracts the relevant information.
The general computation that has to be done is to find the relation between the two
coefficient parametrizing the general form of the wave function for x → +∞ in terms of
the two coefficients parametrizing the general form of the wave function for x → −∞,
by suitably gluing these coefficients across each turning point. From the three examples
that were worked out in detail, it is clear that this can be done according to some simple
and universal general rules. More precisely, putting the pair of coefficients describing the
wave-function in each region into a two-dimensional vector, the gluing across each turning
point amounts to the action of two by two matrix on this. More precisely, we shall track
the coefficients controlling the two independent exponentials describing the solution, and
we therefore define the following notation:
! !
Ci+ CiA
Ciall = , Cifor = . (2.4.100)
Ci− CiB

When crossing a turning point xi , the coefficients in the two regions that it separates
are related by the expression linking the coefficients Ci+ , Ci− and the coefficients CiA ,
CiB . For a turning point xi with raising potential with V ′ (xi ) > 0, the coefficients Cifor in
the forbidden region on the right are linked to the coefficients Ciall in the allowed region
on the left by the relation

Cifor = M↑ · Ciall , (2.4.101)

where:
π π
e−i 4 ei 4
!
M↑ = π π
. (2.4.102)
ei 4 e−i 4

44
Similarly, for a turning point xi with falling potential with V ′ (xi ) < 0, the coefficients
Ciall in the allowed region on the right are linked to the coefficients Cifor in the for region
on the left by the relation

Ciall = M↓ · Cifor , (2.4.103)

where:
π π
ei 4 e−i 4
!
1
M↓ = π π
. (2.4.104)
2 e−i 4 ei 4

When moving from one turning point xi to the next turning point xi+1 , one must
Rx
then rewrite the wave function by expressing the integrals of the form xi in terms of
Rx
integrals of the form x i+1 , before proceeding to the next gluing. When xi and xi+1
Rx
are separated by a well with factor Ji = xii+1 k(x′ )dx′ , one can use the simple relation
Rx ′ ′ xi+1
k(x′ )dx′ , and the coefficients Ci+1
all describing the wave function
R
xi k(x )dx = Ji − x
as seen from the turning point xi+1 on its left are linked to the coefficients Ciall describing
the wave function as seen from the turning point xi on its right by the relation
all
Ci+1 = MWi · Ciall , (2.4.105)

where:
eiJi
!
0 Z xi+1
MWi = , Ji = k(x′ )dx′ . (2.4.106)
e−iJi 0 xi

Rx
Similarly, when xi and xi+1 are separated by a barrier with factor Ki = xii+1β(x′ )dx′ , one
Rx Rx
can use the relation xi β(x′ )dx′ = Ki − x i+1β(x′ )dx′ , and the coefficients Ci+1
for describing

the wave function as seen from the turning point xi+1 on its left are linked to the coefficients
Cifor describing the wave function as seen from the turning point xi on its right by the
relation
for
Ci+1 = MBi · Cifor , (2.4.107)

where:
2eKi
!
0 Z xi+1
MBi = , Ki = β(x′ )dx′ . (2.4.108)
1 −Ki
2e 0 xi

Using the above rules, one can then find the matrix relating the coefficients on the far
right of the potential to those on the far left of the potential by simply multiplying the
matrices associated to each raising or falling turning point and to each well or barrier.

2.4.1 Multiple wells

As a first application, let us consider localized states in a multiple well potential, where
both the asymptotic regions are classically forbidden. The coefficients C> for describing the

45
 for describing the wave function
wave function for x → +∞ are linked to the coefficients C<
for x → −∞ by some matrix M as
for for
C> = M · C< , (2.4.109)

The form of the physical wave function is now obtained by imposing the boundary con-
ditions that it is a falling exponential for x → −∞ and x → +∞. This implies that
C<A 6= 0, C<B = 0 and C>A 6= 0, C>B = 0. The gluing equation is then:
! ! !
C>A MAA MAB C<A
= · · (2.4.110)
0 MBA MBB 0
This implies the following two conditions, which are respectively the quantization condi-
tion and the relation between the two coefficients:
C>A
MBA = 0 , = MAA . (2.4.111)
C<A
For the case of a single well, which has already been studied, it is straightforward to
recover the known results. The relevant matrix is M = M↑ · MW · M↓ , and one easily
computes MBA = cos J. This gives back correct quantization condition cot J = 0, whose
solution is parametrized by one non-negative integer n and reads J = n + 21 h.

VHxL

I x1 II x2 III x3 IV x4 V x

Figure 2.4: Turning points in a double potential well with a finite barrier.

For the case of a double well of the type depicted in fig. 2.4, one can proceed exactly
in the same way, without much more effort. The relevant matrix is in this case:

M = M↑ · MW2 · M↓ · MB · M↑ · MW1 · M↓ . (2.4.112)

A straightforward computation yields:

1
MBA = 2 eK cos J1 cos J2 − e−K sin J1 sin J2 . (2.4.113)
2
It follows that the quantization condition is given by:
1
cot J1 cot J2 = e−2K . (2.4.114)
4

46
Assuming that e−2K ≪ 1, the allowed levels are then in first approximation those of
the two wells considered separately, plus some small real corrections suppressed as e−2K
induced by tunneling effects. In the particular case of a symmetric double well, for which
J1 = J2 = J, this effect is however enhanced and is only suppressed as e−K , due to the
fact that it lifts a two-fold degeneration of the levels. Indeed, the quantization condition
becomes:
1
cot J = ± e−K . (2.4.115)
2
Let us then write the energy as a main contribution E0 plus a small correction ∆E due
to tunneling effects:

E = E0 + ∆E (2.4.116)

One can then expand the left-hand side of the quantization condition as follows:
 
−2 ∂J
cot J E ≃ cot J E0 − sin J ∆E . (2.4.117)
∂E E0

The leading part of the quantization condition states that the spectrum for the leading
values E0 is twice degenerate satisfies the quantization condition of each single well taken
separately:

E0 : cot J E0
= 0. (2.4.118)

The general solution is, as already seen, labelled by a non-negative integer n and is defined
by:
 
1
E0 : J E0 = n + π. (2.4.119)
2

The subleading terms in the quantization condition then determine the correction ∆E to
these levels induced by the tunneling effect through the finite barrier separating them.
One finds:

1 ∂J −1 −K
 
∆E = ∓ e . (2.4.120)
2 ∂E E0

Each degenerate pair of levels therefore splits, as a result of the tunneling effect. The
lowest corresponds to a parity-even wave-function and the upper one to a parity-odd wave
function. Recalling finally that ∂J/∂E = T /(2~) in terms of the period T of the classical
motion within each well, the above result can also be written as
~ −K
∆E = ∓ e . (2.4.121)
T

47
2.4.2 Multiple barriers

As a second application, let us consider delocalized states in a multiple barrier potential,

where both the left and the right asymptotic regions are allowed. The coefficients C> all

describing the wave function for x → +∞ are linked to the coefficients C< all describing the

wave function for x → −∞ by some matrix M as

all all
C> = M · C< , (2.4.122)

Conservation of the current density of probability j implies that the two pairs of coefficients
satisfy

|C<+ |2 − |C<− |2 = |C>− |2 − |C>+ |2 . (2.4.123)

This guarantees that the above matrix M must satisfy the following general properties:

|M−+ |2 − |M++ |2 = 1 , (2.4.124)

2 2
|M+− | − |M−− | = 1 , (2.4.125)
∗ ∗
M++ M+− − M−+ M−− = 0. (2.4.126)

Combining these three equations, one also deduces that:

|M++ | = |M−− | , |M+− | = |M−+ | , | det M | = 1 . (2.4.127)

The form of the physical wave function is now obtained by imposing the boundary con-
ditions that it is an incident plus a reflected wave for x → −∞ and an transmitted wave
for x → +∞. This implies that C<+ 6= 0, C<− 6= 0 and C>+ = 0, C>− 6= 0. The gluing
equation is then:
! ! !
0 M++ M+− C<+
= · · (2.4.128)
C>− M−+ M−− C<−

This implies the following two conditions, which are the relation between the three coef-
ficients:
C<+ M+− C>− det M
=− , = . (2.4.129)
C<− M++ C<− M++

Using (2.4.127), it follows that the reflection and transmission coefficients are given by:
2 2
C<− M++
R= = , (2.4.130)
C<+ M+−
2
C>− −2
T = = M+− . (2.4.131)
C<+

Thanks to (2.4.124)–(2.4.126), these add up to one as they should:

R + T = 1. (2.4.132)

48
 VHxL

I x1 II x2 III x3 IV x4 V x

Figure 2.5: Turning points in a double potential barrier.

For the case of a single barrier, which has already been studied, it is straightforward
to recover the correct results. The relevant matrix is in this case M = M↓ · MB · M↑ ,
and one easily finds that M++ = i eK − 14 e−K and M+− = eK + 41 e−K . It follows that



2
−2
−2
R = 1 − 41 e−2K 1 + 14 e−2K and T = e−2K 1 + 14 e−2K .
For the case of a double barrier of the type depicted in fig. 2.5, one can proceed in the
same way. In that case one has to consider the following matrix:

M = M↓ · MB2 · M↑ · MW · M↓ · MB1 · M↑ . (2.4.133)

After a straightforward computation one finds:

 1  1  K1 −K2 
M++ = 2i eK1 +K2 − e−K1 −K2 cos J + e − e−K1 +K2 sin J , (2.4.134)
16 2
 1  i  
M+− = 2 eK1 +K2 + e−K1 −K2 cos J − eK1 −K2 + e−K1 +K2 sin J . (2.4.135)
16 2
For simplicity, let us from now on focus on the simplest case of a symmetric barrier with
K1 = K2 = K. In that case the above expressions reduce to:
 1 
M++ = 2i e2K − e−2K cos J , (2.4.136)
16

2K 1 −2K 
M+− = 2 e + e cos J − i sin J . (2.4.137)
16
The reflection and transmission coefficients are then found to be given by the following
expressions:

cos2 J
R= , (2.4.138)
1 −4K −2
cos2 J + 41 e−4K 1 −


16 e
1 −4K 1 −4K −2


4 e 1 − 16 e
T = . (2.4.139)
1 −4K −2
cos2 J + 41 e−4K 1 − 16


e

49
Under the conservative assumption that K ≫ 1, these results further simplify to

cos2 J
R≃ , (2.4.140)
cos2 J + 41 e−4K
1 −4K
4e
T ≃ . (2.4.141)
cos2 J + 41 e−4K

The above reflection and transmission coefficients display peaks of transparency, where
R = 0, T = 1, in correspondence to energies E0 for which cos J = 0. These are recognized
to be the energy levels of the well:
 
1
E0 : J E0 = n + π. (2.4.142)
2

We know that these levels are actually only quasi-stationary states, due to the possibility
of escaping the well by tunneling through any of the two finite barriers. We then expect
the width of these states also to play a role in the above formulae. This is indeed the
case. To see it, let us anticipate that this width is equal to twice the one computed in
section 1.10, due to the presence of two identical barriers, as we will confirm by an explicit
computation in next subsection. We are thus led to define

∂J −1 −2K
 
Γ= e . (2.4.143)
∂E E0

Using this definition, we can then write down a simple expression for the dependence on
the energy E of the reflection and transmission coefficients close to the special energies
E0 . To do so, we expand around E0 and compute:
 
2 2 ∂J
cos J E ≃ cos J E0 − 2 cos J sin J (E − E0 )
∂E E0

∂J 2 ∂ 2J
   
2 2
− cos J − sin J − cos J sin J (E − E0 )2
∂E ∂E 2 E0
≃ Γ−2 e−4K (E − E0 )2 . (2.4.144)

Plugging back into the expressions for R and T , one finally finds that for E close to E0
the behavior of the reflection and transmission coefficients is simply

(E − E0 )2
R≃ , (2.4.145)
(E − E0 )2 + Γ2 /4
Γ2 /4
T ≃ . (2.4.146)
(E − E0 )2 + Γ2 /4

These approximate formulae clearly display that the appearance of peaks is directly related
to the existence of quasi stationary states in the well between the two barriers, and that
the energy E0 and the width Γ determine the location and the width of these peaks. One is
then in presence of a resonance phenomenon, and the above formulae display the universal
Breit-Wigner behavior associated to such phenomena. In the situation studied here, this

50
phenomenon can be visualized as an enhancement factor e4K related to the resonance
phenomenon that exactly compensates the suppression factor e−4K related to the barriers
in the transmission coefficient. But this type of phenomenon can actually occur also in
more more general situations.
The above reflection and transmission coefficients also display peaks of opacity, where
R goes through a local maximum and T through a local minimum, in correspondence to
energies E1 for which cos J = 1, that is:

E1 : J E1
= nπ . (2.4.147)

These energies do not correspond to any physical state of the well, but rather to the
particular energies for which a maximal distructive interference of the scattering wave
occurs inside the well. At such points one gets
1
R ≃ 1 − e−4K , (2.4.148)
4
1
T ≃ e−4K . (2.4.149)
4
In the situation at hand, this suppression effect can be visualized as due to the expected
suppression factor e−4K related to the barriers, which is no-longer compensated by any
enhancement factor. But this type of suppression effect can occur also in more general
situations.

2.4.3 Multiple metastable wells

As a last application, let us consider quasi-localized states in a multiple metastable well

potential, where one or both of the left and right asymptotic regions are classically allowed.
Let us consider these two possible situations separately.
In the first general case, where the left region is forbidden and the right region is
allowed, the coefficients C>all describing the wave function for x → +∞ are linked to the
for describing the wave function for x → −∞ by some matrix M as
coefficients C<
all for
C> = M · C< , (2.4.150)

The form of the physical wave function is now obtained by imposing the boundary condi-
tions that it is a damped exponential for x → −∞ and a progressive wave for x → +∞,
resulting from the decay of the quasi-localized state in the well. This implies that C<A 6= 0,
C<B = 0 and C>+ = 0, C>− 6= 0. The gluing equation is then:
! ! !
0 M+A M+B C<A
= · · (2.4.151)
C>− M−A M−B 0

This implies the following two conditions, which are the quantization condition and the
relation between the two coefficients:
C>−
M+A = 0 , = M−A . (2.4.152)
C<A

51
 For the case of a single well followed by a single barrier, which has already been
studied, it is straightforward to recover the correct result. The relevant matrix is in this
π
case M = M↓ · MB · M↑ · MW · M↓ , and one easily finds M+A = ei 4 eK cos J − 4i e−K sin J .

This gives back the quantization condition cot J = 4i e−2K , whose solution for K ≫ 1 is a

−1
complex energy E = E0 − iΓ/2 where E0 : J|E0 ≃ n + 21 π and Γ = 21 ∂J/∂E e−2K .

In the second general case, where both the left and the right regions are allowed, the
coefficients C>all describing the wave function for x → +∞ are linked to the coefficients
all
C< describing the wave function for x → −∞ by some matrix M as
all all
C> = M · C< . (2.4.153)

The form of the physical wave function is obtained by imposing the boundary conditions
that it is a regressive wave for x → −∞ and a progressive wave for x → +∞, resulting from
the decay of the quasi-localized state in the well. This implies that C<+ = 0, C<− 6= 0
and C>+ = 0, C>− 6= 0. The gluing equation is then:
! ! !
0 M++ M+− 0
= · · (2.4.154)
C>− M−+ M−− C<−

This implies the following two conditions, which are the quantization condition and the
relation between the two coefficients:
C>−
M+− = 0 , = M−− . (2.4.155)
C<−

VHxL

I x1 II x2 III x3 IV x4 V x

Figure 2.6: Turning points in a well with two barriers.

In the case of a simple well surrounded by two barriers of the type depicted in fig. 2.6,
one has to consider the following matrix:

M = M↓ · MB2 · M↑ · MW · M↓ · MB1 · M↑ . (2.4.156)

One easily finds:

1 −K1 −K2
i  K1 −K2 
M+− = 2 eK1 +K2 + e cos J − e + e−K1 +K2 sin J . (2.4.157)
16 2

52
In the simplest case of a symmetric barrier with K1 = K2 = K, and under the assumption
that K ≫ 1, this reduces to

M+− ≃ 2e2K cos J − i sin J . (2.4.158)

The quantization condition then reads:

i
cot J ≃ e−2K . (2.4.159)
2
The general solution of this equation is again a complex energy E = E0 − iΓ/2, with the
same real part E0 as before but an imaginary part Γ that is twice as big:
 
1
E0 : J E0 ≃ n + π, (2.4.160)
2
∂J −1 −2K
 
Γ= e . (2.4.161)
∂E E0

Recalling finally that ∂J/∂E = T /(2~) in terms of the period T of the classical motion
within the well, the lifetime τ = ~/Γ is found to be twice smaller as in the case of a single
barrier, as expected:
1
τ= T e2K . (2.4.162)
2

53
54
Chapter 3

Central problems and the

semiclassical approximation

In this chapter we will consider three-dimensional spherically symmetric problems with a

central potential and apply the semiclassical approximation to the study of a number of
typical problems in this setting. We will start by reviewing how the radial and angular
parts of this kind of problems can be separated, and how the angular wave function
can be decomposed in terms of spherical harmonics. We will then discuss the general
features of the radial part of the wave function and describe how one may compute it
in the semiclassical approximation, by treating the radial problem as an effective one-
dimensional problem. We will finally study the various typical problems that can arise in
this context. These include localized bound states and their spectrum of levels, delocalized
states and their scattering amplitude, and metastable states and their life time.

3.1 Separation of radial and angular parts

We shall start by summarizing the basic tools to treat spherically symmetric problems,
where the potential depends only on the distance. Using polar coordinates r, θ and ϕ, the
problem is completely separable. The Hamiltonian takes the form
L2
 
1 2
H= p + 2 + V (r) , (3.1.1)
2m r r
where pr is the Hermitian radial momentum
 
∂ 1
pr = −i~ + , (3.1.2)
∂r r
and L2 is the square angular momentum
 2
1 ∂2

2 2 ∂ ∂
L = −~ + cot θ + . (3.1.3)
∂θ 2 ∂θ sin2 θ ∂ϕ2
These operators also commute with the third component Lz of the angular momentum:
∂
Lz = −i~ . (3.1.4)
∂ϕ

55
The Schrödinger equation for a stationary state of fixed energy E and wave function
i
ψ(r, θ, ϕ, t) = φ(r, θ, ϕ)e− ~ Et implies that the stationary wave function φ(r, θ, ϕ) should
be an eigenvector of H with eigenvalue E:

H φ(r, θ, ϕ) = E φ(r, θ, ϕ) . (3.1.5)

In general the eigenvalue E can be either quantized in terms of some quantum number
n = 0, 1, 2, · · · if the state is a localized bound state, or allowed to take a continuum of
values, if the state is a delocalized wave state:
(
quantized in terms of n = 0, 1, 2, · · · for localized states
E= . (3.1.6)
continuous for delocalized states

Moreover, since L2 and Lz commute with H and among themselves, one can also require
that the stationary wave function φ(r, θ, ϕ) should be an eigenvector of L2 with eigenvalue
λ2 and of Lz with eigenvalue λz :

L2 φ(r, θ, ϕ) = λ2 φ(r, θ, ϕ) , (3.1.7)

Lz φ(r, θ, ϕ) = λz φ(r, θ, ϕ) . (3.1.8)

From the algebraic properties of the angular momentum operators, we finally know that
the eigenvalues of L2 are quantized in terms of a first quantum number l = 0, 1, 2, · · · , and
those of Lz are quantized in terms of an other quantum number m = −l, · · · , l:

λ2 = l(l + 1)~2 , l = 0, 1, 2, · · · , (3.1.9)

λz = m~ , m = −l, · · · , l . (3.1.10)

We can then separate the problem into radial and angular parts by taking the following
factorized form for the wave function:

φ(r, θ, ϕ) = φl (r)Ylm (θ, ϕ) . (3.1.11)

3.2 Exact angular wave function

The angular part Ylm (θ, ϕ) of the wave function corresponding to given values of l and m
satisfies the eigenvalue equations

L2 Ylm (θ, ϕ) = ~2 l(l + 1)Ylm (θ, ϕ) , (3.2.12)

Lz Ylm (θ, ϕ) = ~mYlm (θ, ϕ) . (3.2.13)

The problem is separable and can be reduced to two independent one-dimensional prob-
lems by posing:

Ylm (θ, ϕ) = Nlm ulm (θ)ρm (ϕ) . (3.2.14)

The functions ulm (θ) and ρm (ϕ) then satisfy some ordinary differential equations, which
happen to be exactly solvable, while the constant Nlm is an arbitrary normalization.

56
 The differential equation for ρm (ϕ) arises from the eigenvalue equation for Lz and
takes the following simple form:

ρ′′m (ϕ) + m2 ρm (ϕ) = 0 . (3.2.15)

This is of the same form as the Schrödinger wave equation for a one-dimensional problem
with a constant wave number: k = m. Modulo an arbitrary overall normalization, the
solution of this equation is just a simple phase:

ρm (ϕ) = eimϕ . (3.2.16)

The differential equation for ulm (θ) arises from the eigenvalue equation for L2 and
takes the following form:
m2
 
′′ ′
ulm (θ) + cot θ ulm (θ) + l(l + 1) − ulm (θ) = 0 . (3.2.17)
sin2 θ
To eliminate the term involving the first derivative, one can redefine ulm by extracting a
√
factor of sin θ and posing
vlm (θ)
ulm (θ) = √ . (3.2.18)
sin θ
In this way one finds the following equation for vlm (θ):

1 2 1 − 4m2
  
′′
vlm (θ) + l + + vlm (θ) = 0 . (3.2.19)
2 4 sin2 θ
This is again of the same form as the Schrödinger wave equation for a one-dimensional
problem,p but now with a non-constant wave number that depends on the coordinate:
k(θ) = (l + 1/2)2 − (m2 − 1/4) sin−2 θ. Nevertheless, the regular solution of this equa-
tion can be found in closed form, and modulo an arbitrary overall normalization, it is
given by:
√
vlm (θ) = sin θ Plm (cos θ) , (3.2.20)

in the terms of the associated Legendre polynomials Plm (z) defined as

1 l+m
2 m/2 d

l
Plm (z) = z2 − 1 ,


l
1 − z l+m
(3.2.21)
2 l! dz
Finally, the normalization constant Nlm can be fixed to any convenient value. It is
conventionally chosen to be
s
2l + 1 (l − m)!
Nlm = (−1)m . (3.2.22)
4π (l + m)!

In this way, the angular part of the wave function is finally given by the following spherical
harmonic function:

Ylm (θ, ϕ) = Nlm Plm (cos θ)eimϕ , (3.2.23)

57
These functions satisfy the following completeness and orthonormality conditions:
∞ X
X l
Ylm (θ, ϕ) Ylm∗ (θ ′ , ϕ′ ) = δ(cos θ − cos θ ′ ) δ(ϕ − ϕ′ ) , (3.2.24)
l=0 m=−l
Z 1 Z 2π
′
Ylm∗ (θ, ϕ) Ylm
′ (θ, ϕ) d(cos θ)dϕ = δll′ δmm′ . (3.2.25)
−1 0

3.3 Semiclassical angular wave function

As we have seen in the previous section, the angular part Ylm (θ, ϕ) = Nlm ulm (θ)ρm (ϕ)
of the wave function is universal and can be determined exactly in closed form. There is
therefore no practical need to apply the semiclassical approximation in this case. However,
we will nevertheless shortly discuss what one would find for such a wave function by
applying the semiclassical approximation, and what is the regime of applicability of this
approximate result. This will allow us later on to discuss more precisely the classical limit
of the full three-dimensional problem.
The exact wave equation for ρm (ϕ) involves a constant wave number, which is related
to the quantum number m and is given by:

k(ϕ) = m . (3.3.26)

The De Broglie wave length λ̄ = 1/k is then also constant, and one finds |dλ̄/dϕ| = 0. This
is always identically vanishing and in particular small. The semiclassical approximation
is then formally applicable for any m:

m : arbitrary . (3.3.27)

Moreover, the semiclassical form of the wave function reproduces in this case accidentally
the exact form of the wave function.
The exact wave equation for ulm (θ) involves instead a non-constant wave number,
which depends on both quantum numbers l and m and reads:
q
k(θ) = (l + 1/2)2 − (m2 − 1/4) sin−2 θ . (3.3.28)

The De Broglie wave length λ̄ = 1/k is then also a non-trivial function of θ, and one
finds |dλ̄/dθ| = |m2 − 1/4|| cos θ|[(l + 1/2)2 sin2 θ − (m2 − 1/4)]−3/2 . This is small only if
|(l + 1/2)2 sin2 θ − (m2 − 1/4)| is large. A necessary condition for this to be possible is that
l sin θ ≫ 1 should be large, which implies that l ≫ 1 and θ, π − θ ≫ 1/l. The semiclassical
approximation is therefore in this case applicable only when l is large and θ is not too
close to the extreme values 0 or π:

l ≫ 1 , θ, π − θ ≫ l−1 . (3.3.29)

In this case, the semiclassical form of the wave function will only approximate the exact
form of the wave function.

58
3.4 Exact radial wave function
The radial part φl (r) of the wave function corresponding to given values of l and E satisfies
instead the eigenvalue equation

Heff φl (r) = E φl (r) , (3.4.30)

where, using the definition of p2r and the eigenvalue of L2 , the effective Hamiltonian is
found to be given by

Heff = Teff + Veff , (3.4.31)

with effective kinetic and potential energies given by

~2 d2 ~2 l(l + 1)
 
2 d
Teff = − + , V eff = V (r) + . (3.4.32)
2m dr 2 r dr 2mr 2

More explicitly, the radial wave equation then reads:

 
′′ 2 ′ 2m
l(l + 1)
φl + φl + E − V (r) − φl = 0 . (3.4.33)
r ~2 r2

To get rid of the term of first order in the radial derivative, it is sometimes convenient to
further redefine the radial wave function by extracting a factor of r:

φl (r) = r −1 χl (r) (3.4.34)

The new radial wave function χ(r) then satisfies an equation that is very similar to the
Schrödinger equation for a single degree of freedom associated to the radial coordinate r.
More precisely, one now finds

Heff χl (r) = E χl (r) , (3.4.35)

with

Heff = Teff + Veff , (3.4.36)

where now
~2 d2 ~2 l(l + 1)
Teff = − , V eff = V (r) + . (3.4.37)
2m dr 2 2mr 2
More explicitly, the reduced form of the radial wave equation is then:
 
′′ 2m
l(l + 1)
χl + E − V (r) − χl = 0 . (3.4.38)
~2 r2

The exact behavior of the radial wave function depends on the form of the potential
V (r) and can in general not be determined in closed form. However, its asymptotic
behaviors turn out to be universal, under some mild assumptions. More precisely, let us
assume that V (r) is less singular than 1/r 2 for r → 0 and that it tends to 0 for r → +∞.

59
In such a situation, the behavior for r → 0 is determined by the centrifugal potential and
is universal, while the behavior for r → ∞ is essentially free and thus also universal. For
concreteness we focus here on the case where the energy E is positive. The results can
then be expressed in terms of the real asymptotic wave number defined by:
√
2mE
k= . (3.4.39)
~
The results for negative energy E can be obtained by an analytic continuation and essen-
tially correspond to having an imaginary wave number, as usual. But we will not work
them out in detail here. We shall moreover discuss the asymptotic behaviors of φl (r). The
asymptotic behaviors of the corresponding χl (r) are easily deduced by using the relation
χl (r) = rφ(r).

3.4.1 Asymptotic behavior in the origin

Consider first the behavior in the region r → 0. In that region, the differential equation
for φl is approximately given by
2 l(l + 1)
φ′′l + φ′l − φl ≃ 0 . (3.4.40)
r r2
The solutions of this equation are proportional to the function φl (r) ≃ r α , where α
must satisfy α2 + α − l(l + 1) = 0, implying α = l, −l − 1. Introducing some peculiar
normalizations that will be convenient later on, and noting (2n+1)!! = (2n+1)(2n−1) · · · 1,
the two independent solutions can then be taken to be :
1
φl j (r) ≃ (kr)l , (3.4.41)
(2l + 1)!!
φl n (r) ≃ −(2l − 1)!! (kr)−l−1 . (3.4.42)

The first of these two solutions is regular in the origin and therefore physically acceptable,
while the second is singular in the origin and must therefore be discarded.

3.4.2 Asymptotic behavior at infinity

Consider next the behavior in the region r → +∞. In that region, the differential equation
for φl simplifies to
2
φ′′l + φ′l + k2 φl ≃ 0 . (3.4.43)
r
The solutions of this equation are proportional to the function φl (r) ≃ r −1 sin(kr + δ),
with arbitrary δ. Introducing again some specific normalizations that will be convenient
later on, the two independent solutions can then be taken to be
1  π
φl j (r) ≃ sin kr − l , (3.4.44)
kr 2
1  π
φl n (r) ≃ − cos kr − l . (3.4.45)
kr 2
Both of these solutions are bounded at infinity and therefore acceptable.

60
3.4.3 Relative asymptotic behaviors for vanishing potential

In the particular case of a free particle subject to no potential, the radial wave function
can be determined exactly. The differential equation for φl is in that case given by
 
′′ 2 ′ 2 l(l + 1)
φl + φl + k − φl ≃ 0 . (3.4.46)
r r2
This is the spherical Bessel equation, modulo a simple rescaling of the variable r by k.
The two independent solutions of this equation can be chosen to be the spherical Bessel
functions jl (kr) and nl (kr):

φl j (r) ≃ jl (kr) , (3.4.47)

φl n (r) ≃ nl (kr) . (3.4.48)

The functions jl (z) and nl (z) have a typical behavior that is displayed in fig. 3.1 and are
given by the following expressions:
 l
l 1 d sin z
jl (z) = (−z) , (3.4.49)
z dz z
 l
l 1 d cos z
nl (z) = −(−z) . (3.4.50)
z dz z
These satisfy the following property, which reflects the fact that zjl (z) and znl (z) are the
two independent solutions of a differential equation without first derivative term and must
therefore have a constant Wronskian:

jl (z)nl (z)′ − jl (z)′ nl (z) = 1 . (3.4.51)

The exact solutions (3.4.47) and (3.4.48) can be shown to have precisely the asymp-
totic behaviors reported above, namely (3.4.41) and (3.4.42) for r → 0 and (3.4.44) and
(3.4.45) for r → +∞, including the numerical coefficients, and the transition between
the oscillatory and the power behavior is seen form the differential equation to occur for
p
r ≃ l(l + 1)/k. We then conclude that only the first of these solution is acceptable,
while the second must be discarded, due to the behavior at the origin. The asymptotic
behaviors of this regular combination φl reg (r) will then be the following:
1
φl reg (r) ≃ (kr)l , r → 0 , (3.4.52)
(2l + 1)!!
1  π
φl reg (r) ≃ sin kr − l , r → +∞ . (3.4.53)
kr 2
In other words, once the regular behavior is chosen in the origin, the behavior at infinity
has a fixed amplitude and phase.

3.4.4 Relative asymptotic behavior for non-vanishing potential

In the case of a particle subject to a non-trivial potential, one can in general not find
the explicit form of the two independent solutions of the differential equation defining

61
 ΦHzL

jl HzL

0 z

nl HzL

Figure 3.1: Typical form for spherical Bessel functions jl (z) and nl (z).

the radial wave function. Still, the asymptotic behaviors for r → 0 and r → ∞ must
correspond to linear combinations of the universal asymptotic behaviors that we have
deduced above. However, due to the effect of the potential, it will no longer be true that
the solution that has the j-type or the n-type behavior for r → 0 will have the j-type
or the n-type behavior for r → +∞. In fact, the acceptable solution with the regular
j-type behavior at r → 0 will behave as some linear combination of the j-type and n-type
behaviors at r → +∞. The asymptotic behaviors of this regular combination φl reg (r) will
then have the following general form:
1
φl reg (r) ≃ (kr)l , r → 0 , (3.4.54)
(2l + 1)!!
Al  π 
φl reg (r) ≃ sin kr − l + δl , r → +∞ . (3.4.55)
kr 2
The only possible effect of the potential on the asymptotic behavior of the radial wave
function is therefore an amplitude rescaling Al and a phase shift δl , which depend on the
angular momentum quantum number l and tend respectively to 1 and 0 in the limit of
vanishing potential.

3.5 Semiclassical radial wave function

As discussed in the previous section, the radial part χl (r) of the wave function is de-
termined by a one-dimensional schrödinger-like equation involving an effective potential
Veff (r) = V (r) + ~2 l(l + 1)/(2mr 2 ), and can in general not be determined exactly in
closed form. It is then interesting to address the problem in the semiclassical approxima-
tion. To do so, we can simply apply the general formulae obtained for the semiclassical
wave function of one-dimensional problems, after substituting the wave number k and the
penetration number β in the classically allowed and forbidden regions with the following

62
effective expressions including the effect of the angular momentum:
1p
keff (r) = 2m(E − Veff (r)) , (3.5.56)
~
1p
βeff (r) = 2m(Veff (r) − E) . (3.5.57)
~
Let us now investigate under which conditions the semiclassical approximation is appli-
cable. For simplicity, we shall again suppose that the potential has a behavior which is
less singular than 1/r 2 for r → 0 and that it tends to 0 for r → +∞. For small values of
p
r, Veff (r) is then dominated by the centrifugal term and |keff | ≃ l(l + 1)/r. It follows
p p
that |λ̄eff | ≃ r/ l(l + 1) and |dλ̄eff /dr| ≃ 1/ l(l + 1). The condition |dλ̄eff /dr| ≪ 1 then
implies that

l ≫ 1. (3.5.58)

For large values of r, the whole Veff (r) tends to zero in a smooth way an the semiclas-
sical approximation is therefore always valid. Finally, for intermediate values of r, Veff
depends in a significant way on the potential, and one may get further restrictions on the
applicability of the semiclassical approximation.
By applying the semiclassical approximation as describe above, one obtains results
that work well in situations where the semiclassical approximation holds true, that is for
l ≫ 1. However, it turns out that a minor modification of the above formulae, called
Langer modification, allows to further improve the accuracy of the results. It consists in
replacing the coefficient l(l+1) in front of the effective centrifugal potential with the factor
(l + 1/2)2 , and therefore use the following modified effective potential for the evaluation of
the above wave number and penetration number to define the semiclassical wave function:

~2 (l + 1/2)2
Veff (r) = V (r) + . (3.5.59)
2mr 2
This modification is clearly quantitatively irrelevant in the limit of large quantum number
l, in which the semiclassical approximation is a priori expected to be reliable. However, it
significantly improves the accuracy of the results for small l. There exist various arguments
to justify the above modification and explain why it leads to an improvement. One of the
simplest and most transparent of these arguments is related to the asymptotic behavior
of the semiclassical wave function χl (r) for r → 0. Indeed, we shall see below that
with this modification the semiclassical wave function displays for r → 0 precisely the
universal asymptotic behavior that the exact wave function must have, while without this
modification its behavior would involve small deviations from the exact behavior that are
negligible only for large l.
The behavior of the semiclassical wave function close to r ≃ 0 is easily computed.
Notice for this that the neighborhood of the origin always corresponds to a classically
forbidden region, at least for not too singular potentials, since the effective potential
tends to +∞ at this point. The semiclassical wave function in that region must then be

63
a superposition of the following two expressions:
 Z 
Cl j ′ ′
χl j (r) ≃ p exp − βeff (r )dr , (3.5.60)
βeff (r) r
Z 
Cl n ′ ′
χl n (r) ∼ p exp βeff (r )dr . (3.5.61)
βeff (r) r

For r → 0, one finds that the effective potential with Langer’s modification behaves as
Veff (r) ≃ ~2 (l + 1/2)2 /(2mr 2 ), so that

l + 1/2
βeff (r) ≃ , (3.5.62)
Z r
βeff (r ′ )dr ′ ≃ −(l + 1/2) ln r + constant . (3.5.63)
r

It follows that the two linearly independent semiclassical solutions do indeed display re-
spectively the j-type and n-type universal asymptotic behaviors that the exact wave func-
tion χl (r) ∼ rφl (r) is allowed to have close to the origin, namely

χl j (r) ∼ r 1/2 e(l+1/2)ln r ∼ r l+1 , (3.5.64)

χl n (r) ∼ r 1/2 e−(l+1/2)ln r ∼ r −l . (3.5.65)

From these behaviors we also learn that the acceptable solution that is regular in the
origin is given by (3.5.60), while the other solution (3.5.61) is singular and must therefore
be discarded.

3.6 Localized stationary states

As a first application, let us consider the study of localized stationary states. To illustrate
the situation, we consider a generic attractive potential V (r), which is negative, tends
to 0 for r → +∞ and is negligible with respect to the centrifugal term proportional to
1/r 2 for r → 0. In this situation, there can be a competition between the two terms in
the effective potential Veff (r), and this can give rise to a finite well. There can then be
localized stationary states with a discrete spectrum of negative energies.
For a negative energy E there can be two classical turning points r1 and r2 in the radial
motion, where Veff (r1 ) = E, Veff (r2 ) = E and Veff ′ (r ) < 0, V ′ (r ) > 0, as depicted in
1 eff 2
fig. 3.2. This defines a classically allowed region II = ]r1 , r2 [ and two classically forbidden
regions I = [0, r1 [ and III = ]r2 , +∞[ . The parameter describing the well is:
Z r2
J= keff (r ′ )dr ′ . (3.6.66)
r1

One can then determine the semiclassical radial wave function by using the usual joining
prescriptions. Requiring that the wave function should vanish in the origin and proceeding
as in the case of a one-dimensional well, one then deduces that the semiclassical radial

64
wave function is given by
  Z r1 
Cl ′ ′
χl I (r) ≃ p exp − βeff (r )dr ,



2 βeff (r)



 r
Z r 
Cl π


′ ′
χl II (r) ≃ p sin keff (r )dr + , (3.6.67)

 keff (r) r1 4
 Z r
(−1)n Cl

 
′ ′

 χl III (r) ≃ p exp − βeff (r )dr ,


2 βeff (r)

r2

with the energy quantization condition

 
1
J = n+ π. (3.6.68)
2

VHrL

Ñ2 Hl + 1  2L2
2 m r2

0 I r1 II r2 III r

E
Veff HrL

VHrL

Figure 3.2: Turning points in an attractive central potential.

3.6.1 Spectrum and quantization rule

The above result has the same kind of classical interpretation as in the one-dimensional
case. In particular, the quantization condition can be rewritten in the form of a Bohr-
Sommerfeld quantization rule involving the phase space integral of the radial part of the
problem. More precisely, one can write:
 
1
I
peff (r)dr = n + h. (3.6.69)
2

where
r

~2 (l + 1/2)2
peff (r) = 2m E − V (r) − . (3.6.70)
r2
H
The integral peff (r)dr over the radial phase space that enters the above quantization
condition can be evaluated explicitly as a function of the energy E for a given potential
V (r) by using the same general technique already explained for the one-dimensional case.

65
This is again based on the observation that this integral is equal to twice the integral of
p
peff (r) = 2m(E − Veff (r)) between the points r1 and r2 where Veff (r1 ) = Veff (r2 ) = E
and thus peff (r1 ) = peff (r2 ) = 0. In the complex plane, where peff (z) has a branch cut
ranging between the two turning points r1 and r2 on the real axis, this phase space integral
p
can then be written as the integral of peff (z) = 2m(E − Veff (z)) along a closed contour c
encircling the cut. The integral can then be evaluated by using Cauchy’s theorem for the
reversed contour, which encircles the rest of the complex plane. Assuming that Veff (z) is
analytic everywhere, except at most in z = 0 where the centrifugal term is divergent, the
result of this integral is then simply −2πi times the sum of the residues of the function
p(z) at z = ∞ and at z = 0, which are respectively equal to minus and plus the coefficients
of the 1/z terms in the Laurent expansions of p(z) around these points. In summary, we
thus have:
I I h i
 
peff (r) dr = peff (z) dz = −2πi Res peff (z) z=∞ + Res peff (z) z=0 . (3.6.71)
c
In this case, it is usually rather difficult to directly evaluate the result by computing the
original real integral by means of ordinary techniques, due to the non-trivial form of the
centrifugal barrier.

3.6.2 Example

As an application of this result, let us work out the spectrum of bound states in a Coulomb
potential, of the type:
α
V (r) = − . (3.6.72)
r
The effective potential then reads:
α ~2 (l + 1/2)2
Veff (r) = − + . (3.6.73)
r 2m r 2
For a given negative energy E, there are two classical turning points at r1 and r2 , such
that V (r1 ) = V (r2 ) = E, with:
r
2E~2 (l + 1/2)2
 
α
r1,2 = − 1∓ 1+ . (3.6.74)
2E α2 m
These satisfy the relations
α √ ~(l + 1/2)
r1 + r2 = − , r1 r2 = √ (3.6.75)
E −2mE
We can then rewrite
   
r1 + r2 r1 r2 (r2 − r)(r − r1 )
Veff (r) = E − 2 =E 1+ . (3.6.76)
r r r2
The classically allowed region is the interval ]r1 , r2 [ , and the momentum in this region is
given by
p
peff (r) = 2m(E − Veff (r)
√
p
(r2 − r)(r − r1 )
= −2mE . (3.6.77)
r

66
 H
To compute the integral peff (x) dx, we can deform it in the complex plane and use
(3.6.71). Expanding peff (z) around z = ∞ and z = 0, and taking care of choosing the
correct definition of the square root, one respectively finds:

√
r
r1  r2 
peff (z) = i −2mE 1− 1−
z z
√
  1 
i 1
= −2mE i − (r1 + r2 ) + O 2 , (3.6.78)
2 z z
√ √
r1 r2 
r
z  z
peff (z) = −i −2mE 1− 1−
z r1 r2
√
 
√ 1 i r1 + r2
= −2mE − i r1 r2 + √ + O(z) . (3.6.79)
z 2 r1 r2

It follows that:
 √ hi i
Res peff (z) z=∞
= −2mE (r1 + r2 ,
) (3.6.80)
2
 √ h √ i
Res peff (z) z=0
= −2mE − i r1 2 .
r (3.6.81)

Using these results into the general formula (3.6.71), one the finds
I h i
 
peff (r)dr = −2πi Res peff (z) z=∞ + Res peff (z) z=0
√ h √ i
= −2mE π r1 + r2 − 2 r1 r2 . (3.6.82)

The same result can also be derived by evaluating explicitly the original real integral:
Z r2 p
√ (r2 − r)(r − r1 )
I
peff (r) dr = 2 −2mE dr
r r
 1
√
 
p 1 r1 + r2 − 2r
= 2 −2mE (r − r1 )(r2 − r) − (r1 + r2 ) arcsin
2 r2 − r1
 r2
√ (r1 + r2 )r − 2r1 r2
− r1 r2 arcsin
r(r2 − r1 ) r1
√ h √ i
= −2mE π r1 + r2 − 2 r1 r2 . (3.6.83)

Recalling the definitions of r1 and r2 , the integral is finally found to take the following
form:
r
−mα2
I 
peff (r) dr = 2π −~(l + 1/2) , (3.6.84)
2E

Plugging this result into the condition (3.6.69), one finally deduces that the semiclassical
spectrum accidentally coincides in this case with the well-known exact spectrum, namely:

mα2 1
E=− . (3.6.85)
2~ (n + l + 1)2
2

67
3.7 Delocalized stationary states
As a second application, let us next consider the study of delocalized stationary states.
To illustrate the situation, we consider now a generic repulsive potential V (r), which is
positive, tends to 0 for r → +∞ and is negligible with respect to the centrifugal term
proportional to 1/r 2 for r → 0. In such a situation, the effective potential Veff (r) always
displays a infinite barrier at small r. There can then be delocalized stationary states with
a continuous spectrum of positive energies.
For a given positive energy E there is a single classical turning point r0 in the radial
motion, where Veff (r0 ) = E and Veff ′ (r ) < 0, as depicted in fig. 3.3. This defines a
0
classically forbidden region I = [0, r0 [ and a classically allowed region II = ]r0 , +∞[ .
One can then determine the semiclassical radial wave function by using the usual joining
prescriptions. Requiring that the wave function should vanish at the origin, and allowing
both progressive and regressive waves at infinity, one finds:
  Z r0 
Cl ′ ′
χ (r) ≃ exp − β (r )dr ,

 lI
 p eff
 2 βeff (r) r
Z r  (3.7.86)
Cl ′ ′ π
χ (r) ≃ sin k (r )dr + .


 l II
 p eff
4
keff (r) r0

VHrL

VHrL Veff HrL

Ñ2 Hl + 1  2L2
2 m r2

0 I r0 II r

Figure 3.3: Turning point in a repulsive central potential.

We see that the semiclassical wave function that we have obtained describes a radial
wave coming in from r → +∞ and a radial wave going out to r → +∞. The reflection
is complete and no transmission by tunneling is allowed. The waves come in from large
r, bounce on the barrier at r0 and get out to large r again. However, the boundary
condition at the origin in the classically forbidden region imposes as expected a definite
value for the phase of the wave at infinity. In fact, we see that the asymptotic form of
the semiclassical wave function φl (r) = r −1 χl (r) defined by (3.7.86) does indeed have the
expected universal behavior (3.4.55), where k is the asymptotic wave number
√
2mE
k= . (3.7.87)
~

68
 In the free case with V (r) = 0, one may verify that the phase of the wave function
behaves for large r as expected from eq. (3.4.53). Indeed, in such a situation one has
r
(l + 1/2)2
keff (r) = k2 − . (3.7.88)
0 r2
The turning point where this vanishes is located at r0 |0 = (l + 1/2)/k. One may then
p
rewrite keff = k 1 − r02 /r 2 , and evalutate, for r ≫ r0 :
Z r Z rp ′2
′ ′ r − r02 ′
keff (r )dr = k dr
r0 0 r0 r′
q  r  
′2 2 r0 π
= k r − r0 − r0 arccos ′ ≃ k r − r0
r r0 2
π π
≃ kr − l − . (3.7.89)
2 4
It follows that the semiclassical wave function behaves in this case as:
 π
χl (r) ∝ sin kr − l , r ≫ r0 . (3.7.90)
0 2
In the interacting case with V (r) 6= 0, one gets instead a different behavior for large
r, which is of the form implied by eq. (3.4.55) and involves an extra shift δl in the phase
compared to the free case. Indeed, in such a situation one has:
r
2m (l + 1/2)2
keff (r) = k2 − 2 V (r) − . (3.7.91)
V ~ r2
The turning point where this vanishes is now located at some r0 |V 6= (l + 1/2)/k. One
then finds the following type of result, for r ≫ r0
Z r
π π
keff (r)dr ≃ kr − l − + δl . (3.7.92)
r0 V 2 4

It follows that the semiclassical wave function behaves in this case as

 π 
χl (r) ∝ sin kr − l + δl , r ≫ r0 . (3.7.93)
V 2
The phase shift δl induced by the presence of a non-trivial potential V can be computed
more explicitly in the semiclassical approximation as the following difference of integrals:
Z ∞ V
δl ≃ keff (r)dr
r0 0
Z ∞r

~2 (l + 1/2)2 V
1
≃ 2m E − V (r) − dr . (3.7.94)
~ r0 r2 0

We have here implicitly assumed that the difference of the two integrals converges and
thus sent r → ∞ in the upper extremum of the integration. We see that this is the
case when V (r) tends to zero faster that 1/r for large r. On the other hand, when
V (r) tends to zero slower that 1/r, even the difference of the two integrals diverges, and

69
one finds an infinite result for δl . However, since this divergence is controlled by the
potential and is independent of the centrifugal barrier, it is actually independent of l
and therefore universal. One may then rewrite δl as the sum of an l-independent and
potentially divergent part δ plus an l-dependent and always finite part δ̂l :

δl = δ + δ̂l . (3.7.95)

3.7.1 Scattering amplitude and cross section

Let us now see how one can characterize the scattering process for the original three-
dimensional problem. The typical experimental setting to study such a process consists
in sending in an incident beam of particles from a given direction and measuring the
scattered beam of particles in some other arbitrary direction. In such a setting, one may
then consider the so-called differential cross section σ(Ω) associated to the problem. This
is an observable quantity that is defined as follows for a general scattering problem. One
first considers the quantity dσ defined by the ratio of the number of particles scattered per
unit time within some infinitesimal solid angle dΩ and the number of incident particles
per unit time and unit surface. One then divides this quantity by the infinitesimal solid
angle element dΩ to define σ(Ω) = dσ/dΩ. Summarizing:

# scattered particles in dΩ per unit time divided by dΩ

σ(Ω) = . (3.7.96)
# incident particles per unit time and surface
This quantity has the dimension of a length squared and measures the effective transversal
section that particles scattered with a certain energy and in a certain direction feel for
the target that generates the potential. It is the closest three-dimensional analogue of the
reflection coefficient for one-dimensional problems. But its normalization is not directly
constrained by the conservation of the probability density current, and it can be non-
trivial even at the classical level. One may also define the total cross section σtot as the
integral of the differential cross section σ(Ω) over the solid angle:
Z
σtot = σ(Ω) dΩ . (3.7.97)

For general non-central problems, the scattering process depends both on θ and on ϕ.
One must then to look at θ in the interval [θ, θ + dθ] and ϕ in the interval [ϕ, ϕ + dϕ],
and use dΩ = d(cos θ)dϕ as infinitesimal solid angle element. This gives σ(Ω) = σ(θ, ϕ).
For central problems, on the other hand, one can choose the z axis to coincide with the
direction of the incident particles, and the scattering process then depends only on θ and
not on ϕ. One can then look at θ in the interval [θ, θ + dθ] and ϕ anywhere in the interval
[0, 2π], and use dΩ = 2π d(cos θ) as infinitesimal solid angle element. One then finds
σ(Ω) = σ(θ).
Let us now see how one can compute the differential cross section at the quantum
level for central problems. Choosing the direction of the incident wave to be parallel to
the z axis, the scattered wave can only depend on the angle θ, while the angle ϕ must

70
remain constant. In fact, in such a situation Lz = 0 and the whole wave function φ(r, θ, ϕ)
describing the complete scattering state must be independent of ϕ:

φ(r, θ) = full wave-function of the delocalized stationary state . (3.7.98)

The boundary condition that we would now like to impose on this three-dimensional wave
function is that for r → +∞ it should be the sum of an incident plane wave of the form
φinc (r, θ) = eikr cos θ , which travels along the z axis with some wave number k, and a
scattered spherical wave of the form φsca (r, θ) = f (θ)r −1 eikr , which has an amplitude
that falls off as 1/r and is modulated by some function f (θ) of the scattering angle. The
common overall normalization of these two asymptotic waves will not be important for
our purposes and for simplicity we conventionally set it to one. We then require that:
f (θ) ikr
φ(r, θ) ≃ eikr cos θ + e , r → +∞ . (3.7.99)
r
The fluxes of particles described by this asymptotic wave can be computed by using the
usual expression for the density of probability current:
i~  ∗ ~ 
~ ∗ .
~ = − φ ∇φ − φ∇φ (3.7.100)
2m
For the incident plane wave φinc (r, θ) = eikr cos θ , the current is oriented along the z
direction and has a modulus given by:
i~  ∗ ∂ ∂ 
jinc (r, θ) = − φinc φinc − φinc φ∗inc (r, θ)
2m cos θ ∂r ∂r
~k
= . (3.7.101)
m
For the scattered spherical wave φsca (r, θ) = f (θ)r −1 eikr , on the other hand, this current
is oriented radially and has a modulus given at large r by
i~  ∗ ∂ ∂ 
jsca (r, θ) = − φsca φsca − φsca φ∗sca (r, θ)
2m ∂r ∂r
~k |f (θ)|2
≃ . (3.7.102)
m r2
To compute the differential cross section σ(θ), we now observe that the number of incident
particles per unit time and surface is just jinc , while the number of scattered particles per
unit time in the solid angle element dΩ defining a surface element dS = r 2 dΩ is given by
jsca dS = jsca r 2 dΩ. It then follows that
jsca (r, θ) 2
σ(θ) = r (3.7.103)
jinc (r, θ)
Using the expressions (3.7.101) and (3.7.102), we see that the dependence on r cancels out
and the differential cross section σ(θ) is finally given simply by the squared norm of the
function f (θ) modulating the amplitude of the scattered wave relative to the amplitude
of the incident wave, namely:

σ(θ) = |f (θ)|2 . (3.7.104)

71
The corresponding total cross section is:
Z 1
σtot = 2π |f (θ)|2 d(cos θ) . (3.7.105)
−1

The above quantities can be expressed in terms of properties of the infinitely many
radial problems corresponding to each value of the angular momentum quantum number
l. To see this, we start by expanding the full wave function φ(r, θ) on the complete set
of spherical harmonics Ylm (θ, ϕ) with coefficients φl (r). Since there is no dependence on
the angle ϕ, the expansion actually involves only the simplest spherical harmonics with
arbitrary l but m = 0. These depend only on θ, through a Legendre polynomial Pl (cos θ)
of degree l in cos θ:
r
2l + 1
Yl0 (θ) = Pl (cos θ) . (3.7.106)
4π
The explicit form of the Legendre polynomials Pl (z) is:

1 dl 2
l
Pl (z) = z − 1 . (3.7.107)
2l l! dz l
They moreover satisfy the following completeness and orthonormality relations:
∞
X
(2l + 1)Pl (x)Pl (x′ ) = 2 δ(x − x′ ) , (3.7.108)
l=0
1
2
Z
Pl (x)Pl′ (x) dx = δll′ . (3.7.109)
−1 2l + 1

With a suitable redefinition of the normalization factor in the coefficients φl (r) of the
original expansion, one may then write:
∞
X
φ(r, θ) = il (2l + 1)φl (r)Pl (cos θ) . (3.7.110)
l=0

On the other hand, a plane wave propagating along the z axis with wave number k also
admits a similar expansion, since it is a solution of the Schrödinger equation for the
spherically symmetric setting of a free particle. Moreover we known that the coefficients
in this expansion must be proportional to the spherical Bessel functions jl (r), since these
determine the allowed form of the radial wave function with definite angular momentum for
a free particle. The proportionality constants can be fixed by multiplying this expansion
by Pl′ (cos θ), integrating over θ and using the orthogonality property of the Legendre
polynomials. The precise formula is found to be:
∞
X
eikr cos θ = il (2l + 1)jl (kr)Pl (cos θ) . (3.7.111)
l=0

Finally, a spherical wave with an amplitude modulated by a function f (θ) can also be
expanded in this way. The non-trivial part is the arbitrary function f (θ), which can be

72
expanded with some arbitrary coefficients fl as:
∞
X
f (θ) = (2l + 1)fl Pl (cos θ) . (3.7.112)
l=0

With this parametrization, the differential cross section σ(θ) can be expressed as follows
in terms of the coefficients fl :
∞
X 2
σ(θ) = (2l + 1)fl Pl (cos θ) . (3.7.113)
l=0

The total cross section σtot is then obtained by integrating the differential cross section
R1
σ(θ) over the full solid angle: σtot = 2π −1 σ(θ)d(cos θ). Using the expression (3.7.113)
and the orthonormality relation (3.7.109) for Legendre polynomials, one finds:
∞
X
σtot = 4π (2l + 1)|fl |2 . (3.7.114)
l=0

It turns out that the partial amplitudes fl defining σ(θ) and σtot are fully determined
by the knowledge of the phase shifts δl characterizing the asymptotic form of the radial
wave functions for arbitrary angular momentum l. To find the relation between fl and
δl , we have to look at the asymptotic behavior of the above expansions for r → +∞. By
comparing (3.7.110) with (3.7.111), we see that the normalization implied for φl (r) is such
that in the case of a free particle subject to no potential this reduces to jl (kr). With this
normalization, we then know from (3.4.55) that that the full radial wave function must
behave as φl (r) ≃ Al (kr)−1 sin kr − l π2 + δl for r → +∞. The asymptotic behavior of

the full wave function is thus determined by the following expression in terms of the phase
shifts δl :
∞  
X
l Al  π
φ(r, θ) ≃ i (2l + 1) sin kr − l + δl Pl (cos θ) , r → +∞ . (3.7.115)
kr 2
l=0

Similarly, we know that jl (kr) ≃ (kr)−1 sin kr −l π2 for r → ∞. The asymptotic behavior

of the sum of the incident and the scattered wave is then given by
∞ 
ikr cos θ f (θ) ikr X l h1  π fl ikr
e + e ≃ i (2l + 1) sin kr − l + l e Pl (cos θ) ,
r kr 2 ir
l=0
r → +∞ . (3.7.116)

These two expressions are seen to be equal, as required by the boundary condition (3.7.99),
if and only if:

Al (−i)l eiδl eikr − Al (i)l e−iδl e−ikr = (−i)l (1 + 2ikfl ) eikr − (i)l e−ikr . (3.7.117)

Matching the progressive and regressive waves implies the following two conditions:
1 2iδl
Al = eiδl , fl =


e −1 . (3.7.118)
2ik

73
The second of these relations provides the relation between fl and δl we were looking for.
It can be written in the following equivalent forms:
1 2iδl
1 1 1
fl = e − 1 = eiδl sin δl = (3.7.119)
2ik k k cot δl − i
From this we see that the modulus of this partial amplitude is bounded by above for a
fixed energy corresponding to a fixed wave number k:
1 1
|fl | = | sin δl | ≤ . (3.7.120)
k k
The scattering amplitude is then found to be expressed in terms of the phase shifts δl by
the following relation:
∞
1 X
(2l + 1) e2iδl − 1 Pl (cos θ)


f (θ) =
2ik
l=0
∞
1X
= (2l + 1) eiδl sin δl Pl (cos θ) . (3.7.121)
k
l=0

The first term in the bracket involving the factor e2iδI controls the deflected part of the
scattered wave in a generic direction, while the second term in the bracket involving the
factor −1 is instead relevant only for the undeflected part of the scattered wave in the
forward direction. For θ = 0 both terms contribute and the result can be simplified by
using the fact that Pl (1) = 1. For θ 6= 0, on the other hand, only the first term contributes,
while the second drops, as can be verified by using the identity (3.7.108) together with
Pl (1) = 1. In these two situations one then finds:
 ∞
1X
f (0) = (2l + 1) eiδl sin δl ,




 k
l=0
∞ (3.7.122)
 1 X 2iδl
 f (θ) = 2ik (2l + 1) e Pl (cos θ) , θ 6= 0 .



l=0

The differential cross section for arbitrary θ finally reads

∞ 2
1 X iδl
σ(θ) = (2l + 1) e sin δl Pl (cos θ) . (3.7.123)
k2
l=0

In the two cases where θ = 0 and θ 6= 0, this can also be rewritten as

 ∞ 2
1 X iδl
σ(0) = (2l + 1) e sin δ ,

l

k2



l=0
∞ 2
(3.7.124)
 1 X 2iδl
σ(θ) = (2l + 1) e P (cos θ) , θ 6= 0 .

l

k2


l=0

The total cross section σtot is finally given by the following simple expression involving a
single series of partial square amplitudes:
∞
4π X
σtot = (2l + 1) sin2 δl . (3.7.125)
k2
l=0

74
We see that this satisfies the following relation, which shows up here as a matter of fact
but actually has a much more general validity and is called optical theorem:
4π
σtot = Imf (0) . (3.7.126)
k
Let us finally analyze the physical relevance of the potentially divergent universal part
in the phase shifts δl = δ + δ̂l . We see that the differential cross section for θ 6= 0 does not
depend on the overall term δ, while the forward differential cross section for θ = 0 and
the total cross section do instead depend on the overall term δ. This means that only the
forward scattering at vanishing angle is sensitive to the potentially divergent δ, while the
scattering at any non-vanishing angle only depends on the finite δ̂l . This is compatible
with the fact that the quantum cross section can display a divergence related to small
angle scattering when the potential does not fall off sufficiently fastly at infinity. We shall
see below that the same type of divergence related to small angle scattering is actually
already displayed by the classical cross section, for any long range potential, and that it
admits a clear semiclassical interpretation.

3.7.2 Classical interpretation

To conclude, let us discuss the classical interpretation of these results and see how a
classical scattering trajectory can emerge in situations where the whole scattering process
is semiclassical. Recall in this respect that the radial part of the problem is semiclassical
when l ≫ 1, while the angular part of the problems is semiclassical when θl, (π − θ)l ≫ 1.
This means that in situations where the semiclassical approximation is applicable, the
infinite sum over l defining the scattering amplitude must be dominated by terms with
l ≫ 1, and moreover the scattering angle must be such that θ, π − θ ≫ 1/l. We thus
assume that we can restrict to

l ≫ 1 , θ, π − θ ≫ 1/l . (3.7.127)

To proceed, we may then substitute the Legendre polynomials Pl (cos θ) with their asymp-
totic form for l sin θ ≫ 1. This corresponds to use for the angular part of the wave
function Yl0 (θ) = (2l + 1)/(4π) Pl (cos θ) the approximate expression that one would
p

find in a semiclassical rather than an exact treatment. From our discussion of section 3.3,
we see that when l sin θ ≫ 1 and m = 0 the relevant wave number (3.3.28) is approxi-
mately constant and given by k ≃ l + 1/2, and this semiclassical angular wave function
must therefore involve a plane wave with this wave number. More precisely, one can show
that the asymptotic form to be used is given by:
  
r sin l + 1 θ + π
2 2 4
Pl (cos θ) ≃ √ , l sin θ ≫ 1 . (3.7.128)
π l sin θ
Plugging this expression into the scattering amplitude for θ 6= 0 given by the second of
(3.7.122) and rewriting the sine in the numerator in terms of exponentials, we then arrive

75
at the following expression:
∞ r   
X l 1  1 π
f (θ) ≃ exp i 2δl − l + θ−
2π sin θ k 2 4
l≫1
 
 1 π
− exp i 2δl + l + θ+ . (3.7.129)
2 4
When written in this semiclassical form, we see that the scattering amplitude is an infinite
sum over l of terms that involve large phases that rapidly oscillate when l is changed.
As a result, most of these terms cancel out from the infinite sum over l, and the main
contribution comes from values of l close to the special value for which the phase of one of
the two terms is extremal and stationary. This happens whenever the derivative of these
phases with respect to l vanishes, that is when 2 dδl /dl ∓ θ ≃ 0 or equivalently:
dδl θ
≃± . (3.7.130)
dl 2
Notice that this relation implies that δl ≃ ±θl/2 + constant, and since lθ ≫ 1 it follows
that the semiclassical phase shifts are large:

|δl | ≫ 1 . (3.7.131)

Since the phase shifts δl depend on the energy, the constraint (3.7.130) represents a sharp
relation between the energy, the angular momentum and the deflection angle, as required
by the emergence of a definite classical trajectory. Using the previously derived semiclassi-
cal form (3.7.94) of these phase shifts, it is straightforward to show that this relation indeed
corresponds to the one implied by the classical trajectory followed by the particle during
the scattering process. To see this, let us start form the semiclassical expression (3.7.94)
Rr
for δl . Since we know that r0 keff (r)dr|0 = kr − (l + 1/2) π2 , the term associated to the free
R +∞ R +∞
motion can be written more explicitly as r0 keff (r)dr|0 = r0 kdr + kr0 − (l + 1/2) π2 .
√
Recalling also that k = 2mE/~ and L ≃ ~(l + 1/2), one can then write
Z r
1 ∞
L2 √ π √
  
1
δl ≃ 2m E − V (r) − 2 − 2mE dr + L − 2mE r0 . (3.7.132)
~ r0 r ~ 2

The derivative dδl /dl can now be easily computed. Using dL/dl = ~ and noticing that
although dr0 /dl 6= 0 the two terms that this induces cancel out, one finds:

L2 −1/2
Z ∞  
dδl L π
≃ − 2
2m E − V (r) − 2 dr + . (3.7.133)
dl r0 r r 2

Plugging this into the condition (3.7.130) one finally deduces the following relation between
the energy E, the angular momentum L and the scattering angle θ:

L2 −1/2
Z ∞  
L π∓θ
2
2m E − V (r) − 2 dr ≃ . (3.7.134)
r0 r r 2
This is recognized to be the general classical relation between the energy E, the angular
momentum L and the asymptotic deflection angle θ for a particle moving in a central

76
potential V (r). This relation can also be rewritten in a different way in terms of the
impact parameter b. Indeed, for a classical particle one has L = p b and E = p2 /(2m),
where p is the momentum at infinity. For a given energy E, the impact parameter b is
then related to the angular momentum L by the following expression:
L
b= √ . (3.7.135)
2mE
Using this new variable, the formula (3.7.134) can be rewritten as a relation between the
deflection angle θ and the impact parameter b for fixed energy E:
∞
V (r) b2 −1/2
 
b π∓θ
Z
1− − 2 dr ≃ . (3.7.136)
r0 r2 E r 2

This leads to the classical picture of the scattering process illustrated in 3.4. The negative
sign applies to repulsive potentials and the positive sign to attractive potentials, as can
be seen by requiring that the argument of the square root should admit one root r0 . It
should however be recalled that this semiclassical picture holds true only when the original
assumptions that θ, π − θ ≫ 1/l is fulfilled. In the semiclassical limit l ≃ L/~, and these
conditions can therefore be rephrased as the requirement that

θ, π − θ ≫ ~/L . (3.7.137)

We can now understand that this restriction is related to the limitations imposed by the
quantum mechanical uncertainty relations. Indeed, a classical interpretation requires that
for an impact parameter b determined with a small uncertainty ∆b ≪ b one should be able
to associate a deflection angle θ with a small uncertainty ∆θ ≪ θ, π − θ. But denoting
with p the momentum and with ∆p its uncertainty in the transverse direction, one has
∆θ ∼ ∆p/p. The usual uncertainty principle then implies that ∆b∆p > ∼ ~ and therefore
> >
∆θ ∼ ~/(p∆b). This finally implies that θ, π − θ ≫ ∆θ ∼ ~/(p∆b) ≫ ~/(p b) ∼ ~/L.

b r0
Θ
0

Figure 3.4: Classical scattering in a central potential.

In the above fully semiclassical situation, we may also compute more explicitly the cross
section and compare it to the classical result. To do so, we have to evaluate explicitly
the infinite sum defining f (θ). We have seen that this receives its dominant contribution
from terms where l is close to a special value l0 , which depends on the energy E and the

77
scattering angle θ as implied by eq. (3.7.134) with L ≃ ~l0 . To evaluate this contribution,
we may now define the new variable l′ = l − l0 , and expand the phases appearing in
(3.7.129) up to quadratic order in l′ . Since l0 is defined by eq. (3.7.130), one deduces that
dδl /dl|l0 ≃ ±θ/2 and thus d2 δl /dl2 |l0 ≃ ±(dθ/dl)/2|l0 . One then finds:
   
 1 π  1 π 1 dθ ′2
2δl ∓ l + θ∓ ≃ 2δl0 ∓ l0 + θ∓ ± l + O(l′3 ) . (3.7.138)
2 4 2 4 2 dl0
Using this expression for the phases in (3.7.129) and setting l ≃ l0 everywhere else, one can
then compute the sum over the values l′ around 0. After rewriting dθ/dl0 = ~ dθ/dL and
introducing ξ = ~ l′ , one may think of ~ as being very small. This allows to convert the
sum into an integral over ξ with dξ = ~ and evaluate this with the saddle point method.
In this way, assuming that dθ/dL < 0 one finds the following expression:

1 ∞ 2π dL 1/2 ∓i π
    r
i dθ ′2 i dθ 2
X Z
I ≃ exp ± l ≃ dξ exp ∓ ξ = e 4
2 dl0 ~ −∞ 2~ dL ~ dθ
l ≫1
′

√ 1/2
dl0 π
≃ 2π e∓i 4 . (3.7.139)
dθ
The scattering amplitude then becomes
r  
l0 1  1 π
f (θ) ≃ ± exp i 2δl0 ∓ l0 + θ∓ I
2π sin θ k 2 4
l0 1 dl0 1/2
r  
 1 π
≃ ± exp i 2δl0 ∓ l0 + θ∓ . (3.7.140)
sin θ k dθ 2 2

The differential cross section σ(θ) = |f (θ)|2 is finally found to be:

l0 dl0 l0 dl0
σ(θ) = 2
=− 2 . (3.7.141)
k sin θ dθ k sin θ dθ
Noticing now that eq. (3.7.135) implies that b ≃ l0 /k and recalling that dΩ = 2π d(cos θ),
we may finally write this formula in the following way:
d(πb2 )
σ(θ) = . (3.7.142)
dΩ
This is recognized to be the classical result for σ(θ). Indeed, for a classical particle of
fixed energy, the impact parameter b and the deflection angle θ are sharply related. As
a result, we see from fig. 3.4 that the number of particles scattered per unit time in the
solid angle element dΩ must be equal to the number of incident particles per unit time
per unit surface times the surface element defined by impact parameters b in the interval
[b, b + db] and arbitrary angles ϕ in the interval [0, 2π], which is given by dS = d(πb2 ). The
differential cross section is then simply σ(θ) = dS/dΩ = d(πb2 )/dΩ. The total cross section
is straightforwardly evaluated by integrating the above expression for the differential cross
section over the solid angle. Calling bmax the maximal value of the impact parameter for
which the differential cross section is non-vanishing, one trivially finds:

σtot = πb2max . (3.7.143)

78
This shows that in the classical limit the total cross section is equal to the transversal sur-
face in which the incident particles undergo a deflection due to the effect of the potential,
no matter how big this deflection is. This quantity is therefore finite for potentials with a
finite range and infinite for potentials with an infinite range.

3.7.3 Example

As an important example of scattering process, let us consider the case of an attractive

Coulomb potential of the form
α
V (r) = − . (3.7.144)
r
In this case, r0 is determined by the equation 1 + (α/E) r0−1 − b2 r0−2 = 0 and is found to
be given by:
r
α α2
r0 = − + + b2 . (3.7.145)
2E 4E 2
It follows that:
V (r) b2 −1/2 b2 −1/2
Z ∞   Z ∞   
b b
2
1− − 2 dr = r − r0 r+ dr
r0 r E r r0 r r0
b
= 2 arctan . (3.7.146)
r0
The equation (3.7.136) then gives:
b π+θ
= tan . (3.7.147)
r0 4
Using the expression of r0 it follows that
α b r0 θ
= − = 2 tan . (3.7.148)
Eb r0 b 2
Finally, this leads to the following expression for the impact parameter in terms of the
deflection angle:
α θ
b= cot . (3.7.149)
2E 2
The differential cross section may now be computed according to the classical formula
bdb b db
σ(θ) = =− (3.7.150)
d(cos θ) sin θ dθ
One finds in this way the Rutherford formula:
 α 2 θ
σ(θ) = sin−4 . (3.7.151)
4E 2
The total cross section is in this case divergent:
σtot = +∞ . (3.7.152)
It turns out that the above semiclassical results for the differential and total cross sec-
tion accidentally coincide with the exact quantum result, which can derived by solving
the differential equation for the radial wave function exactly in terms of hypergeometric
functions.

79
3.8 Metastable quasi-stationary states
As a third application, let us consider the study of metastable quasi stationary states. To
illustrate the situation, we imagine a repulsive potential V (r), which is positive, tends to
0 for r → +∞ and remains smaller that 1/r 2 for r → 0. In such a situation, the effective
potential can give rise to a finite well followed by a finite barrier. There can then be
metastable quasi-stationary states with a discrete spectrum of complex energies.
For a given energy E there are three classical turning points r1 , r2 and r3 , where
′ (r ) < 0, V ′ (r ) > 0, V ′ (r ) < 0, as depicted
Veff (r1 ) = Veff (r2 ) = Veff (r3 ) = 0 and Veff 1 eff 2 eff 3
in fig. 3.5. This defines two classically allowed regions II = ]r1 , r2 [ , IV = ]r3 , +∞[ and
two classically forbidden regions I = ]0, r1 [ , III = ]r2 , r3 [ . The parameters describing the
well and the barrier are:
Z r2
J= keff (r ′ )dr ′ , (3.8.153)
r1
Z r3
K= βeff (r ′ )dr ′ . (3.8.154)
r2

One can then determine the semiclassical radial wave function by using the usual joining
prescriptions. Requiring that the wave function should vanish at the origin and consist
only of a progressive wave at infinity, and proceeding as in the case of a one-dimensional
well associated to a barrier and focusing on the case where K ≫ 1, one finds:
  Z r1 
Cl ′ ′
χl I (r) ≃ p exp − βeff (r )dr ,



2 βeff (r)



 r
  r 
Cl π

 Z
 ′ ′

 χl II (r) ≃ p sin keff (r )dr + ,
keff (r) 4

r1


 Z r Z r
(−1)n Cl
  
′ ′ i −2K ′ ′
χl III (r) ≃ p exp − βeff (r )dr + e exp βeff (r )dr ,


2




 2 βeff (r) r2 r2

nC  π Z r 
 χ (r) ≃ (−1)

l

i 4 −K ′ ′
e e exp i keff (r )dr ,

 l IV p (3.8.155)
2 keff (r)

r3



with the quantization condition cot J ≃ 4i e−2K or more explicitly:

 
1 i
J ≃ n+ π − e−2K . (3.8.156)
2 4
By proceeding as in the corresponding one-dimensional problem, we then deduce that
the complex energy levels E = E0 − 2i Γ have real and an imaginary parts determined as
follows:
 
1
J E0 = n + π, (3.8.157)
2
1 ∂J −1 −2K
 
Γ= e . (3.8.158)
2 ∂E E0

80
 VHrL

Veff HrL
E

Ñ2 Hl + 1  2L2

VHrL 2 m r2

0 I r1 II r2 III r3 IV r

Figure 3.5: Turning points in a peculiar central potential.

3.8.1 Energy and width

The classical interpretation of the above results is the same as in the one-dimensional case.
The real part E0 of the energy is determined by a Bohr-Sommerfeld quantization rule on
the phase space integral of the radial part of the problem:
 
1
I
peff (r)dr = n + h. (3.8.159)
2

The width Γ associated to the imaginary part of the energy can instead again be written
in terms of the period T of the classical motion:
~ −2K
Γ= e . (3.8.160)
T
The associated lifetime is:

τ = T e2K . (3.8.161)

3.9 Resonance effects

As a last application, let us consider the possibility of having resonance effects in delo-
calized stationary states describing a scattering process. To illustrate the situation, we
imagine again a repulsive potential V (r), which is positive, tends to 0 for r → +∞ and
remains smaller that 1/r 2 for r → 0. In such a situation, the effective potential can
give rise to a finite well followed by a finite barrier. There can then be resonance effects
in delocalized stationary states with a continuous spectrum of real energies, which arise
whenever the energy gets close to one of the discrete values of the energy associated to
metastable quasi stationary states admitted by the effective potential.
For a given energy E there are three classical turning points r1 , r2 and r3 , where
′ (r ) < 0, V ′ (r ) > 0, V ′ (r ) < 0, as depicted
Veff (r1 ) = Veff (r2 ) = Veff (r3 ) = 0 and Veff 1 eff 2 eff 3
in fig. 3.5. This defines two classically allowed regions II = ]r1 , r2 [ , IV = ]r3 , +∞[ and

81
two classically forbidden regions I = ]0, r1 [ , III = ]r2 , r3 [ . The parameters describing the
well and the barrier are as before:
Z r2
J= keff (r ′ )dr ′ , (3.9.162)
r1
Z r3
K= βeff (r ′ )dr ′ . (3.9.163)
r2

One can then once again determine the semiclassical radial wave function by using the
usual joining prescriptions. Requiring that the wave function should vanish at the origin
and allowing both a progressive and a regressive wave at infinity, and focusing on the case
where K ≫ 1, one finds after a straightforward computation:
  Z r1 
Cl ′ ′
χl I (r) ≃ p exp − βeff (r )dr ,



2 βeff (r)



 r
 Z r 
Cl π


 ′ ′

 χl II (r) ≃ p sin keff (r )dr + , (3.9.164)
keff (r) 4

r1




   Z r  Z r 

 Cl ′ ′ ′ ′
χl III (r) ≃ p sin J exp − βeff (r )dr + 2 cos J exp βeff (r )dr ,

 2 β eff (r) r 2 r 2
    Z r 
 χ (r) ≃ p Cl i

−i π4

K −K ′ ′
e cos Je + sin Je exp i keff (r )dr

 l IV



 keff (r) 4 r3

    Z r 

 i π
K i −K ′ ′
+ e 4 cos Je − sin Je exp −i keff (r )dr .


4


 r3

As a check, notice that the above expression reduces to (3.8.155) for complex energies sat-
isfying the quantization condition rule (3.8.156), which implies that cos J ≃ 4i (−1)n e−2K
and sin J ≃ (−1)n . For generic and real values of the energy, on the other hand, we see
that the last term is non-vanishing and there is not only a progressive wave but also a
regressive wave for large r. One is then in a situation where one has a delocalized scatter-
ing state consisting of a radial wave coming in from r → +∞ and a radial wave getting
out to r → +∞. The reflection is complete, as before, because there is nowhere else than
r → +∞ that the wave can end. However, in this case only part of the wave bounces back
at the classical turning point r3 that it meets. The other part tunnels through the barrier
and bounces back only at the classical turning point r1 .
The wave function is seen to have a behavior for large r which is of the expected type,
and can be rewritten in the following more convenient form:
Z r  
′ ′ π i −2K
χl (r) ∝ sin keff (r )dr + + arg cot J + e . (3.9.165)
r3 4 4

From this expression, we see that the phase shifts δl are the sum of two terms,

δl = δlnor + δlres . (3.9.166)

82
The first term is the one that one would get in the absence of tunneling and is given by
the normal expression:
Z r V
δlnor = keff (r ′ )dr ′ . (3.9.167)
r3 0
The second term is instead due to the possibility of tunneling and represents an extra
effect that can display a resonance phenomenon:
 
res i −2K
δl = arg cot J + e . (3.9.168)
4
Let us now consider the behavior of the phase shift for energies E close to the energy
E0 of a quasi-stationary state:
 
1
E0 : J E0 = n + π. (3.9.169)
2
We also recall that the width of the quasi-stationary states, which is expected to play a
role, is given by
1 ∂J −1 −2K
 
Γ= e . (3.9.170)
2 ∂E E0
One may now expand around E0 in powers of E−E0 . Working at first order and neglecting
terms involving both e−2K and E − E0 unless they come with a Γ−1 , one finds:
     
i −2K i −2K −2 ∂J
cot J + e ≃ cot J + e − sin J (E − E0 )
4 E 4 E0 ∂E E0
i 1
≃ e−2K − Γ−1 e−2K (E − E0 )
4 2
 
1 −1 −2K Γ
≃ Γ e −E + E0 + i . (3.9.171)
2 2
Using this expression and assuming that the normal part of the phase shifts is a slowly
varying function of E close to E0 , one may then approximate:

δlnor ≃ δlnor (E0 ) , (3.9.172)

 
res Γ
δl ≃ arg −E + E0 + i . (3.9.173)
2
The total phase shift is therefore:
 
nor Γ
δl ≃ δl + arg −E + E0 + i . (3.9.174)
2
When increasing E form values smaller than E0 to values larger than E0 , we see that the
resonance term changes from 0 to π, going very sharply through π/2 around E0 . To see
the impact of this behavior in the scattering amplitude, we compute:
 
2iδl 2iδlnor E − E0 − iΓ/2 iΓ nor
e ≃e ≃ 1− e2iδl , (3.9.175)
E − E0 + iΓ/2 E − E0 + iΓ/2
nor iΓ nor
e2iδl − 1 ≃ e2iδl − 1 − e2iδl . (3.9.176)
E − E0 + iΓ/2

83
If follows that the partial scattering amplitude fl = e2iδl − 1 /(2ik) can be written as the

sum of a normal term and a resonance term,

fl ≃ flnor + flres , (3.9.177)
where:
1  2iδnor 
flnor = e l −1 , (3.9.178)
2ik
1 Γ/2 nor
flres ≃− e2iδl . (3.9.179)
k E − E0 + iΓ/2
We discover that the partial scattering amplitude fl (E) has a pole in the complex energy
plane, which is located at the complex energy E0 − iΓ/2 of the quasi-stationary state. For
real values of E considered in scattering processes, this does not lead to any singularity, but
since Γ was assumed to be small it results in a strong enhancement in the neighborhood
of E0 . Indeed, for E close to E0 the resonance term tends to dominate over the normal
term and in such a situation one then finds a Breit-Wigner behavior:
Γ2 /4 1
|fl (E)|2 ≃ . (3.9.180)
(E − E0 ) + Γ /4 k2
2 2

We see from this expression that for E ≃ E0 the square modulus of the partial amplitude
for the value of l for which the metastable state exists takes the maximal possible value:
|fl (E0 )|2 ≃ 1/k2 . If for some reason the contribution from this partial wave dominates over
the others, this peak may also show up in the cross section. This is a resonance effect that
is perfectly analogous to the one encountered in section 2.4 for the transmission coefficient
associated to a one-dimensional potential consisting of a symmetric double barrier, where
peaks of maximal transparency arise in correspondence of the energies of metastable states
allowed by the well between the two barriers. Here the resonance effect concerns the radial
problem associated to a definite partial wave with fixed l, and although there is a single
barrier this has to be crossed twice by the wave: once when it comes in and once when
gets out. The two situations are thus indeed very similar.
An other extreme behavior of the phase shifts occurs for energies E close to some
particular energy E1 satisfying:
E1 : J E1
= nπ . (3.9.181)
In such a situation, we see that cot J + 4i e−2K tends to ±∞ and therefore its argument is
equal to 0 modulo π. One then finds:
δlnor ≃ δlnor (E1 ) , (3.9.182)
δlres ≃ 0 mod π . (3.9.183)
The total phase shift is then equal to the normal one, modulo π:
δl ≃ δlnor mod π . (3.9.184)
If then follows that the resonance effect is totally absent and the partial amplitude is equal
to the normal one:
fl = flnor . (3.9.185)

84
Chapter 4

Approximation methods for

central scattering problems

In this chapter, we will present several alternative approximation methods for computing
the scattering amplitude, which can be applied in different regimes and become helpful in
many situations where a semiclassical approach is not possible and there are important
quantum effects. We will start with some general considerations concerning the compu-
tation of the scattering amplitude. We will then study the case of weak coupling, where
one can use a perturbative expansion where the potential is treated as a small correction
to the energy. We will next consider the case of high energy scattering, where the wave
length is much smaller than the typical range of the potential and the scattering process
displays a classical behavior in its radial part and a simple small angle behavior in its
angular part. We will finally consider the case of low-energy scattering, where the wave
length is much larger than the typical range of the potential and the scattering process
displays a threshold behavior dominated by quantum effects.

4.1 General behavior of the partial-wave expansion

We have seen in the previous chapter that the scattering amplitude defining the differential
and the total cross sections is given for central problems by an infinite sum over partial
waves. This takes the general form
+∞
X
f (θ) = (2l + 1)fl Pl (cos θ) , (4.1.1)
l=0
where the partial amplitudes fl are related to the phase shifts δl displayed by the radial
wave function χl (r) in the presence of the potential as compared to the free wave function
χ0 (r) = rjl (kr) that would arise in the absence of the potential:
1 2iδl
1 1 1
fl = e − 1 = eiδl sin δl = (4.1.2)
2ik k k cot δl − i
The determination of the phase shifts δl and the partial amplitudes fl requires in principle
the knowledge of the radial wave functions χl (r) solving the radial Schrödinger equation.

85
 An important point about the above partial wave expansion is the range of values of l
that significantly contributes. This strongly depends on the situation that is considered. In
situations where the semiclassical approximation can be applied, we have seen in previous
chapter that for given wave number k and scattering angle θ the sum over l is dominated by
values close to the special value l0 associated to the definite angular momentum L ≃ ~l0 or
impact parameter b ≃ l0 /k corresponding to the classical trajectory. But to look at all the
possible scattering angles, one needs in general to include all the possible angular momenta
and impact parameters, and a priori all the values of l can contribute. In situations where
quantum effects are important, the situation is yet more complicated, and even for given
wave number k and scattering angle θ, a wider range of values of l can give significant
contributions. And in order to look at all the possible scattering angles, one needs once
again to include all the possible values of l, in general.
To gain some intuition on which values of l can really give a significant contribution,
let us suppose for concreteness that the potential has a finite range a beyond which it is
strongly suppressed and below which it is significant. When looking at all the possible
scattering angles but some definite wave number k, we then expect partial waves to sig-
nificantly contribute only when l is smaller than or comparable to but not much larger
than a certain maximal value given by
lmax ≃ ka . (4.1.3)
In situations where the semiclassical approximation is valid, this corresponds to the fact
that if l ≫ ka then b ≫ a and the particle cannot be significantly deflected because it does
never pass at distances r <∼ a within the range of action of the potential. In situations
where quantum effects are important, this can instead be motivated through the following
observation. If l ≫ ka, the region of r <
∼ a where the potential can has a significant effect
is such that kr ≪ l. In such a region, the free radial wave function φ0 (r) = jl (kr) is then
well approximated by a power low of the form χ0 (r) ∼ (kr)l , and is thus tiny and damped.
It is then reasonable to expect that it should be very little affected by the presence of the
potential. This implies that that the phase shift δl should correspondingly be very small,
and there should thus be no significant contribution to the scattering amplitude.
Armed with these general considerations, we shall now study three different general
situations where it is possible to find a concrete approximate expression for the scattering
amplitude in terms of the potential.

4.2 Weak coupling scattering and the Born approximation

In many cases, the effects of the potential can be considered as small. In such a situation,
the scattering amplitude will also be small, although still a non-trivial function of the
wave number and the scattering angle. One can then compute it by taking into account
the effects of the potential in a perturbative way. The systematics of such a perturbative
expansion will be described in detail later on. But it turns out that the leading first order
approximation can be easily obtained with very elementary considerations, and we will
therefore discuss it here.

86
 The basic idea is that when the potential is small, the exact radial wave function
χl (r) in the presence of the potential differs only mildly from the free radial wave function
χ0l (r) = rjl (kr). One may then try to exploit this to find an approximate expression for
the phase shifts δl . To do so, let us start from the two differential equations satisfied by
√
these two functions. Denoting as usual with k = 2mE/~ the asymptotic momentum,
these can be written in the following form:
 
′′ 2 l(l + 1) 2m
χl (r) + k − 2
χl (r) = 2 V (r)χl (r) , (4.2.4)
r ~
 
l(l + 1) 0
χ0l ′′ (r) + k2 − χl (r) = 0 . (4.2.5)
r2

Multiplying the first of these equations by χ0l (r) and the second by χl (r), and taking the
difference, one deduces that:
2m
χ′′l (r)χ0l (r) − χ0l ′′ (r)χl (r) = V (r)χl (r)χ0l (r) . (4.2.6)
~2
We now see that the left-hand side of this equation is in fact a total derivative, since one
has χ′′l χ0l − χ0l ′′ χl = (χ′l χ0l − χ0l ′ χl )′ . Integrating over r ∈ [0, +∞[, and using the fact that
the behavior at r → 0 is universal and therefore identical for χl (r) and χ0l (r), one then
deduces that:
h i 2m Z +∞
′ 0 0′
lim χl (r)χl (r) − χl (r)χl (r) = 2 V (r)χl (r)χ0l (r)dr . (4.2.7)
r→∞ ~ 0

Using the asymptotic forms χ0l (r) ≃ k−1 sin(kr − l π2 ) and χl (r) ≃ k−1 eiδl sin(kr − l π2 + δl )
valid for large r, we further deduce after a trivial computation that the left-hand side of
this expression is directly related to the partial scattering amplitude:
h i 1
lim χ′l (r)χ0l (r) − χ0l ′ (r)χl (r) = − eiδl sin δl = −fl . (4.2.8)
r→∞ k
Using the expression χ0l (r) = rjl (kr) in the right-hand side, it finally follows that:
+∞
2m
Z
 
fl = − 2 V (r)χl (r) rjl (kr) dr . (4.2.9)
~ 0

This is an exact relation between the partial scattering amplitude fl and the corresponding
exact radial wave function χl (r).
In the case where the effect of the potential on the radial wave function χl is small, we
may determine this in a perturbative expansion in powers of the potential. This will yield
a result of the form χl = χ0l + χ1l + · · · , where the n-th term χnl will be suppressed by n
powers of the potential. In first approximation, we may then take χl (r) ≃ χ0l (r) = rjl (kr).
Plugging this in the above formula, we then deduce the following approximate result for
the partial scattering amplitude, which goes under the name of Born approximation:

2m +∞
Z
 2
fl ≃ − 2 V (r) rjl (kr) dr . (4.2.10)
~ 0

87
Notice that this approximation requires that |fl | ≪ 1, which implies that δl ≪ 1. In this
situation we therefore have the approximate relation δl ≃ kfl , and the above formula also
directly yields an approximate expression for the phase shifts:
2mk +∞
Z
 2
δl ≃ − 2 V (r) rjl (kr) dr . (4.2.11)
~ 0

Using the above expressions, we can now compute the scattering amplitude f (θ). This is
found to be given by:
+∞
X
f (θ) = (2l + 1)fl Pl (cos θ)
l=0
+∞ +∞
X 
2m
Z
2
(2l + 1) jl (kr) Pl (cos θ) r 2 dr .

= − 2 V (r) (4.2.12)
~ 0 l=0

It turns out that this result can be most conveniently expressed in terms of the momentum
that is transferred from the potential (or more precisely the body generating it) to the
particle during the scattering process:

~q = ~ksca − ~kinc . (4.2.13)

Notice that the modulus of the momentum of the particle is the same before and after the
scattering, since the potential tends to zero at infinity: ksca = kinc = k. But its orientation
can change, and this results in a non-zero ~q, whose modulus is related to the scattering
angle θ. Since θ is the angle between ~ksca and ~kinc , one has ~ksca · ~kinc = k2 cos θ. It follows
that:

q 2 = ksca
2 2
+ kinc − 2 ~ksca · ~kinc = 2k2 (1 − cos θ)
θ
= 4k2 sin2 , (4.2.14)
2
and therefore that
θ
q = 2k sin . (4.2.15)
2
The infinite sum appearing in the (4.2.12) can now be evaluated explicitly without too
~ ~
much difficulty. To do so, one may start from the plane wave ei~q·~r = eiksca ·~r e−ikinc ·~r ,
reexpress each of the two factors on the right hand side by an expansion of the type
(3.7.111) and integrate over the orientation of ~r relative to ~q. In this way one finds an
expression for the sum (4.2.12) in terms of the integral of ei~q·~r over the orientation of ~r
relative to ~
q . The result turns out to be given by the following expression in terms of q:
+∞
X  2 sin(qr)
(2l + 1) jl (kr) Pl (cos θ) = , (4.2.16)
qr
l=0

It then follows that the scattering amplitude is given as a function of q by

2m +∞ sin(qr) 2
Z
f (q) ≃ − 2 V (r) r dr . (4.2.17)
~ 0 qr

88
We can now easily recognized that the integral appearing in this expression is proportional
to the Fourier transform of the potential evaluated at the momentum transfer. Indeed:
Z Z +∞ Z 1 
V (r) e−i~q·~r d3~r = 2π V (r) e−iqr cos θ d(cos θ) r 2 dr
0 −1
Z +∞  
sin(qr) 2
= 4π V (r) r dr . (4.2.18)
0 qr
One finally finds the following simple expression for the scattering amplitude in the Born
approximation:
m
Z
f (q) ≃ − V (r) e−i~q·~r d3~r . (4.2.19)
2π~2
The corresponding differential cross section is given by:
 m 2 Z 2
−i~
q ·~
r 3
σ(q) = V (r) e d ~
r . (4.2.20)
2π~2
Notice finally that since q 2 = 2k2 (1 − cos θ) one has dq 2 = −2k2 d(cos θ), and the total
cross section can be written as an integral over q 2 :
Z 2
π 4k 2
σtot = 2 σ(q ) dq 2 . (4.2.21)
k 0
It should also be emphasized that the Born approximations for f (θ) and σtot do not satisfy
the optical theorem exactly, but rather in an approximate perturbative way. Indeed, the
first order expression for f (θ) is real and an imaginary part can arise only at second order
in perturbation theory, corresponding to the fact that the leading effect in σtot is quadratic
in the potential.

4.2.1 Example

As an example, let us consider the case of a screened repulsive Coulomb potential of the
type:
α
V (r) = e−r/a . (4.2.22)
r
The scattering amplitude in the Born approximation is given by (4.2.17), which reads in
this case:
2m α +∞−r/a
Z
f (q) ≃ − 2 e sin(qr) dr (4.2.23)
~ q 0
The integral that appears in this expression is easily calculated:
Z +∞
1 +∞ −(1/a−iq)r
Z  
−r/a −(1/a+iq)r
e sin(qr) dr ≃ e −e dr
0 2i 0
 
1 1 1
≃ −
2i 1/a − iq 1/a + iq
q
≃ 2 . (4.2.24)
q + a−2

89
One then finds:
2mα 1
f (q) ≃ − . (4.2.25)
~ q + a−2
2 2

The differential cross section then reads:

 2mα 2 1
σ(q 2 ) ≃ 2
2 . (4.2.26)
~ q 2 + a−2

Finally, the total cross section is

4k 2 4k 2
π π  2mα 2 dq 2
Z Z
σtot ≃ σ(q 2 ) dq 2 =
2 . (4.2.27)
k2 0 k2 ~2 0 q 2 + a−2

The integral in this expression is evaluated to be:

4k 2 4k2
dq 2

1 1 1
Z

2 = − 2 =− + −2
0 q 2 + a−2 q + a−2 0 4k2 +a −2 a
4k2 a4
= (4.2.28)
1 + 4k2 a2
It follows that:
 4mα 2 a4
σtot ≃= π . (4.2.29)
~2 1 + 4k2 a2
Notice that in the limit a → ∞, in which one recovers the case of an unscreened Coulomb
potential, the differential cross section reduces to the Rutherford cross section, while the
total cross section diverges. In that situation the born approximation accidentally yields
the exact result.

4.3 High energy scattering and the eikonal approximation

Another situation in which there is a significant simplification is that of the scattering
of very high energy particles from a potential with a finite range a. More precisely, let
us suppose that V is strongly suppressed for r larger that a, and that the energy of
the incoming particles is such that E ≫ |V | and k ≫ 1/a, meaning that λ̄ ≪ a. This
is a situation where according to our discussion of section 4.1 lmax ≃ ka ≫ 1. The
main contribution to the scattering amplitude therefore comes from partial waves with
large angular quantum number l and is restricted to small scattering angles. We may
then assume that l ≫ 1 and θ ≪ 1. This implies that the radial part of the problem
is semiclassical. One can then use the semiclassical approximation for the phase shifts
derived in previous chapter. The angular part of the problem is instead not semiclassical
but rather dominated by quantum effects. One can then not use the semiclassical result
for the cross section, but one may derive another simple result for it by exploiting the fact
that the scattering angle is very small.

90
 To derive an approximate formula for the scattering amplitude in the above situation,
let us then start from the semiclassical approximation for the phase shifts, which is valid
for l ≫ 1 and can be rewritten in the following way:
Z ∞r 2
Z ∞r
2m l l2
δl ≃ k2 − 2 V (r) − 2 dr − k2 − 2 dr . (4.3.30)
r0 ~ r r0 r
In the situation at hand, the potential term has a small effect in this semiclassical formula.
In particular, the turning point r0 is dominantly determined by the centrifugal barrier and
can be taken in both integrals to be approximately given by
l
r0 ≃ . (4.3.31)
k
Moreover, one may expand the integrand of the first integral in powers of V (r). The
zeroth order term then just cancels against the second integral, while the first order term
gives a good approximation of the result:

m ∞ l2 −1/2
Z  
δl ≃ − 2 V (r) k2 − 2 dr
~ l/k r
r02 −1/2
Z ∞  
m
≃ − 2 V (r) 1 − 2 dr . (4.3.32)
~ k r0 r
Finally, one may rewrite this integral in a more convenient way as an integral over the z
p p p
coordinate. Writing r = r02 + z 2 one has z = r 2 − r02 and dz = r/ r 2 − r02 dr, and
since r0 ≃ l/k it follows that:
Z ∞ p
m

δl ≃ − 2 V z 2 + l2 /k2 dz . (4.3.33)
2~ k −∞

This simple result can be interpreted in the following way. In a truly semiclassical
situation, the phase of the wave function would be given by S/~, where S is the classical
action evaluated along the classical trajectory. In the regime of very high energy scattering
we are considering, the situation is not really semiclassical, but it turns out that one may
nevertheless use in a meaningful way the same formula with S evaluated on a trajectory
that is a straight line with fixed impact parameter b and z going from −∞ to +∞,
corresponding to the fact that the scattering angle must be very small. The asymptotic
form for z → +∞ of the true scattering wave function in the presence of the potential
and the free wave-function that one would have in the absence of the potential are then
related by a factor e2iδ(b) where δ(b) = (S(b)|V − S(b)|0 )/(2~). The scattering amplitude
must then involve the factor e2iδ(b) − 1, and this suggests that it should be possible to
relate δ(b) to the phase shifts δl computed in this situation. As a matter of fact, one finds
that δ(b) is given by the following expression in the high-energy limit:

1 +∞ 2 2m p 2 1 +∞
Z r Z


δ(b) ≃ k − 2V 2
z + b dz − k dz
2 −∞ ~ 2 −∞
Z +∞ p
m

≃ − 2 V z 2 + b2 dz . (4.3.34)
2~ k −∞

91
We can now make contact with our previous expression (4.3.33) by taking b ≃ r0 ≃ l/k.
We then see that one indeed finds


δl ≃ δ b , (4.3.35)

with b given by
l
b≃ . (4.3.36)
k
To continue and derive an approximate expression for the scattering amplitude, we
now have to plug the above approximate expression for the phase shifts into the sum over
partial waves defining the scattering amplitude and evaluate this sum. To do so, we may
use the approximate behavior of the Legendre polynomials for small angles θ ≪ 1 and
large angular momenta l ≫ 1. This can be obtained by solving the angular wave equation
in the limit θ ≪ 1 and for l ≫ 1, which reads u′′l (θ) + 1/θ u′l (θ) + l2 ul (θ) ≃ 0. In this way
one deduces that


Pl (cos θ) ≃ J0 lθ , θ ≪ 1 , l ≫ 1 , (4.3.37)

in terms of the zeroth order Bessel function J0 (z), which is defined by the differential
equation J0′′ (z) + 1/z J0′ (z) + J0 (z) = 0 and admits the following integral representation:
2π
1
Z
J0 (z) = e−iz cos ϕ dϕ . (4.3.38)
2π 0

Substituting this in the infinite sum defining the scattering amplitude, one may now
evaluate the latter more concretely. To do so, we switch to the new variable b ≃ l/k and
rewrite δl = δ(b) and J0 (lθ) = J0 (kθb). One may then treat k as a large quantity and
approximate the infinite sum as an integral over b with db ≃ 1/k. Proceeding in this way
one finds:
+∞ +∞
1 X 1 X 2iδl
(2l + 1) e2iδl − 1 Pl (cos θ) ≃



f (θ) ≃ e − 1 J0 (lθ) l
2ik ik
l≫1 l≫1
Z +∞
e2iδ(b) − 1 J0 (kθb) b db .


≃ −ik (4.3.39)
0

Using the integral representation (4.3.38) for J0 (kθb), one may finally write this as:
+∞Z 2π
k
Z
e2iδ(b) − 1 e−ikθb cos ϕ b db dϕ .


f (θ) ≃ (4.3.40)
2πi 0 0

This result can be rewritten in a yet more convenient way in terms of the transferred
momentum ~ q . Indeed, for a very high energy scattering process involving a small deflection
angle θ ≪ 1, this lies approximately in the xy plane, like the impact parameter ~b, and has
a modulus that is simply given by kθ:

~q ⊥ ẑ , q ≃ k θ . (4.3.41)

92
 We may then identify the angle ϕ in (4.3.40) with the arbitrary angle between the vectors
~b and ~q in the xy plane. In this way, we have ~q · ~b ≃ kθ b cos ϕ and d2~b = b db dϕ, and
the scattering amplitude (4.3.40) can be rewritten as a two-dimensional integral over the
impact parameter vector, namely:
k
Z
~
e2iδ(b) − 1 e−i~q·b d2~b .


f (q) ≃ (4.3.42)
2πi

The result (4.3.42), or equivalently (4.3.39), is called the eikonal approximation to the
scattering amplitude, in reference to the fact that it is determined by a semiclassical
phase δ(b), which we recall is given by
Z +∞ p
m

δ(b) ≃ − 2 V z 2 + b2 dz . (4.3.43)
2~ k −∞

One can show that the eikonal approximation satisfies the optical theorem exactly. As a
result, one can exploit the optical theorem to compute the total cross section out of the
imaginary part of the forward scattering amplitude:
4π
σtot ≃ Imf (0)
kZ
≃ 4 sin2 δ(b) d2~b . (4.3.44)

Notice that in the special situations where the interaction is weak and the potential
can be treated as small, the phase δ(b) is small, since it is linear in the potential. One
may then make the further expansion e2iδ(b) − 1 ≃ 2iδ(b), and the eikonal approximation
reduces to the Born approximation:
Z +∞Z
m p ~
z 2 + b2 e−i~q·b d2~b dz


f (q) ≃ − 2
V
2π~ −∞
m
Z
≃ − V (r) e−i~q·~r d3~r . (4.3.45)
2π~2
As already mentioned, when doing this further approximation the optical theorem holds
only approximately.

4.3.1 Example

As a simple example, let us consider the high-energy scattering by a spherical barrier of

the form:
(
V0 , r < a ,
V (r) = (4.3.46)
0, r > a.

In such a situation, the eikonal phase is given by:

p
δ(b) = −ν0 1 − b2 /a2 θ(a − b) . (4.3.47)

93
in terms of the following parameter:
mV0 a
ν0 = . (4.3.48)
~2 k
It follows that the scattering amplitude in the eikonal approximation is given by the
following integral:
Z a  p  
f (θ) ≃ −ik 2 2
exp −2iν0 1 − b /a − 1 J0 (kθb) b db (4.3.49)
0

In the totally forward direction θ = 0, one can use J0 (0) = 1 and performing the change
of variable x = 1 − b2 /a2 , dx = −a−2 / 1 − b2 /a2 b db one therefore finds:
p p

Z 1   
2
f (0) ≃ −ika exp −2iν0 x − 1 x dx
0
ka2
 
1 −2iν0 i  −2iν0
≃ i+ e − 2 e −1
2 ν0 2ν0
ka2
   
cos(2ν0 ) sin(2ν0 ) 1 sin(2ν0 ) cos(2ν0 )
≃ − + i 1 + − − .(4.3.50)
2 ν0 2ν02 2ν02 ν0 2ν02

Using the optical theorem, one finally deduces that the total cross section is given by:
4π
σtot = Imf (0)
k  
2 1 sin(2ν0 ) cos(2ν0 )
≃ 2πa 1 + 2 − − . (4.3.51)
2ν0 ν0 2ν02

In the case of a very low barrier with ν0 ≪ 1, this simplifies to the following result (which
coincides with what one would obtain by using the Born approximation and taking the
high energy limit ka ≫ 1):

σtot ≃ 2πa2 ν02 , ν0 ≪ 1 . (4.3.52)

In the case of a very high barrier with ν0 ≫ 1, one finds instead the following very simple
result:

σtot ≃ 2πa2 , ν0 ≫ 1 . (4.3.53)

This is a factor of 2 bigger than the classical result πa2 for an unpenetrable spherical
barrier. The extra contribution can be interpreted as a diffraction effect related to the
shadow that must arise behind the spherical barrier in the forward direction.

4.4 Low energy scattering and the threshold approximation

Another situation in which there is a significant simplification is that of the scattering
of very low energy particles from a potential with a finite range a. More precisely, let
us suppose that V is strongly suppressed for r larger that a, and that the energy of
the incoming particles is such that k ≪ 1/a, meaning that λ̄ ≫ a. This is a situation

94
where the scattering process is totally dominated by quantum effects and does not admit
a semiclassical picture. However, there is an important simplification: according to our
discussion of section 4.1 lmax ≃ ka ≪ 1. All the partial waves with l 6= 0 then give a
suppressed contribution, and the scattering amplitude is dominated by the l = 0 term,
also called s-wave. Intuitively, the reason for this can also be understood as follows.
At distances r > ∼ a, the potential is negligible and Veff is therefore dominated by the
centrifugal barrier. In particular, at r ∼ a one has Veff (a) ∼ ~2 l(l + 1)/(2ma2 ). In order
for a radial mode with l 6= 0 to penetrate down to this distance and feel the potential,
it would then need to have an energy E ∼ ~2 l(l + 1)/(2ma2 ), corresponding to a wave
p
number k ∼ l(l + 1)/a. If instead k ≪ 1/a, the radial mode cannot penetrate to small
distances and significantly feel the potential. As a consequence, the radial wave function
is essentially the free one and the phase shifts approximately vanish. The only exception
to this arises for the l = 0 mode, for which the centrifugal barrier is absent and the effect
of the potential can then be felt significantly.
A more quantitative information about the low-energy behavior of the phase shifts,
also called threshold behavior, can be obtained by studying the radial wave equation,
which reads:
 
′′ 2 ′ 2 l(l + 1) 2m
φl (r) + φl (r) + k − 2
φl (r) = 2 V (r)φl (r) . (4.4.54)
r r ~

Recall that in general the behaviors for r → 0 and for r → +∞ are universal and inde-
pendent of the potential, while the phase shift observed at large r depends instead on the
effect of the potential in the whole region of finite r. In the situation at hand, however,
the potential affects the wave function only for r much smaller than a. For r much larger
than a, the potential is negligible and the wave function must therefore be a linear com-
bination of the two independent solution of the free radial wave equation. We may write
this general form of the wave function in the following way:
h i
φl (r) ≃ Al cos δl jl (kr) − sin δl nl (kr) , r ≫ a . (4.4.55)

Notice now that since k ≪ 1/a, the distance of order 1/k where this solution changes
its behavior from power low to oscillatory is much larger than a. Using the asymptotic
behaviors (3.4.41), (3.4.42) and (3.4.44), (3.4.45) one then finds:

(kr)l
 
(2l − 1)!!
φl (r) ≃ Al cos δl + sin δl , a ≪ r ≪ 1/k , (4.4.56)
(2l + 1)!! (kr)l+1
Al  π 
φl (r) ≃ sin kr − l + δl , r ≫ 1/k . (4.4.57)
kr 2
The constant δl that has been used to parametrize the relative weight of the two linearly
independent solutions is now recognized to be precisely the asymptotic phase shift. It
is determined by the matching across the region of r ∼ a the above free wave function
valid for r ≫ a to the potential-influenced wave function valid for r ≪ a. The precise
numerical value of δl can be determined only once the exact behavior of the wave function
in the region where the potential is relevant is known. However, one may deduce the

95
general way it can depend on k by a simple argument. The basic observation is that in
the region a ≪ r ≪ 1/k the k2 term in the differential equation can still be neglected,
and the solution should therefore not depend on k. This means that the coefficients
Al cos δl kl and −Al sin δl k−l−1 of the two independent solutions r l and r −l−1 in (4.4.56)
should be independent of k, modulo an overall normalization. In particular, their ratio
− tan δl k−2l−1 should be a constant αl depending only on l and not on k. This implies
that tan δl = −αl k2l+1 , and the phase shifts δl can therefore be parametrized in terms of
the constants αl as:

δl ≃ −arctan αl k2l+1 .


(4.4.58)

Notice now that since k is assumed to be small, it is rather natural to have δl ≪ 1. In that
case, the arctangent can be linearized to give δl ≃ −αl k2l+1 , and the partial scattering
amplitude is simply fl ≃ δl /k ≃ −αl k2l . However, it may also happen that some of the
αl are big, and that δl ∼ 1. In that case, one has to keep the arctangent in the above
expression for δl to compute the partial scattering amplitude, and one finds:
1 1
fl ≃ ≃ −1 −2l
k cot δl − ik −αl k − ik
αl k2l
≃ − . (4.4.59)
1 + iαl k2l+1
In this general expression, we can now safely approximate the denominator with 1, since
the extra term is further suppressed by an extra factor of k, even when αl k2l is sizable.
So finally:

fl ≃ −αl k2l . (4.4.60)

As expected, in a generic situation with small k the dominant effect is represented by the
l = 0 partial amplitude, while the l 6= 0 partial amplitudes are smaller and smaller when
l is increased. One may then perform a kind of multipole expansion where one retains a
finite number of partial waves according to the accuracy that one wants to achieve. In this
respect, it should be noticed that for small but finite k, the qualities αl = − tan δl /k2l+1
are actually only approximately constant and can be expanded in powers of k2 . But
the contributions of different partial waves are nevertheless distinguished by the fixed
dependence on θ through Pl (cos θ) that they induce in the full scattering amplitude f (θ).
The s-wave has partial amplitude f0 = −α0 and gives an isotropic contribution since
P0 (cos θ) = 1. The p wave has partial amplitude f1 = −α1 k2 and gives an anisotropic
contribution controlled by P1 (cos θ) = cos θ. The d wave has partial amplitude f2 = −α2 k4
and gives an anisotropic contribution controlled by P2 (cos θ) = 3/2 cos2 θ − 1/2. And so
on. In situations where k is very small, one may in first approximation restrict to the s
wave and write f (θ) ≃ f0 , or:

f (θ) ≃ −α0 . (4.4.61)

The scattering amplitude is thus approximately constant and independent of the angle at
very low energies. The number α0 , which has the dimension of a length and depends on

96
the form of the potential, is called the scattering length. Its general definition is:
tan δ0
α0 = − lim . (4.4.62)
k→0 k
The differential cross section is then given by

σ(θ) ≃ α20 , (4.4.63)

and finally the total cross section is simply:

σtot ≃ 4πα20 . (4.4.64)

It is worth emphasizing that the optical theorem, which implies that σtot = 4π/k Imf (0),
cannot be sensibly applied with these formulae for the threshold approximation, since one
has both k → 0 and Imf (0) → 0.
Whenever the potential is weak, the quantities αl are small and can be easily computed
at first order in perturbation theory. The result can be obtained immediately by starting
from the Born approximation (4.2.11). Replacing the spherical Bessel function with its
behavior for small argument, one deduces that as expected δl ≃ −αl k2l+1 , with:
Z +∞
2m 1
αl ≃ 2 V (r) r 2l+2 dr . (4.4.65)
~ [(2l + 1)!!]2 0

In particular, the approximate scattering length at weak coupling is given by

2m +∞
Z
α0 ≃ 2 V (r) r 2 dr . (4.4.66)
~ 0

This formula shows that in the weak coupling limit α0 is positive for repulsive potentials
and negative for attractive potentials, and small in absolute value.
In general situations where the potential is not weak, the quantities αl can be computed
by determining the exact behavior of the wave function in the region r < ∼ a and then
>
matching this to the general free form (4.4.56) for r ∼ a to determine the parameters
of the latter. One interesting situation where this matching can be carried out more
explicitly is that of potentials which exactly vanish for r larger than some finite range a.
The matching of the wave function can then be done by requiring the continuity of its
logarithmic derivative in r = a. More precisely, suppose that we known the exact wave
function φl (r) for r ≤ a. We may then compute its logarithmic derivative in r = a and
get some number γl :

rφ′l rχ′l
= − 1 = γl . (4.4.67)
φl a χl a

We must then compare this with the logarithmic derivative of the free wave function
φ0l (r) = Al [cos δl jl (kr) − sin δl nl (kr)] valid for r ≥ a evaluated in r = a, which is given by:

rφ0l ′ rχ0l ′ cos δl jl′ (ka) − sin δl n′l (ka)

= − 1 = ka . (4.4.68)
φ0l a χ0l a cos δl jl (ka) − sin δl nl (ka)

97
Matching these two expressions one deduces that the phase shifts δl are related to the
logarithmic derivatives γl of the exact wave function by the following relation:

ka jl′ (ka) − γl jl (ka)

 
δl = arctan . (4.4.69)
ka n′l (ka) − γl nl (ka)

For small k, one may now use the asymptotic behavior of the spherical Bessel functions
for small argument. In this way, one arrives at an expression for δl which has indeed the
expected form

δl ≃ −arctan αl k2l+1 ,


(4.4.70)

with αl given by the following expression:

γl − l a2l+1
αl = . (4.4.71)
γl + l + 1 (2l + 1)!!(2l − 1)!!

This means that the partial amplitudes behave as fl ∝ (ka)2l a, showing once again that
the contribution of higher and higher partial waves is a priori more and more suppressed
for ka ≪ 1, unless the dimensionless coefficient controlling them grows large. Looking at
the dominant l = 0 case, one deduces in particular that the scattering length is given by
γ0
α0 ≃ a. (4.4.72)
1 + γ0

There exist in this case an interesting pictorial interpretation for the scattering length
α0 , which is based on the form of the wave function for k → 0. In such a extreme
low-energy limit, 1/k → +∞, and therefore there are essentially only two regions to be
considered. In the region r ∈ [0, a[ the wave function depends on the potential. In the
region r ∈ [a, +∞[, on the other hand, it is universal and given by the expression (4.4.56)
with δ0 ≃ −arctan(α0 k), from which we deduce that φ0 (r) ∝ 1 + tan δ0 /(kr) = 1 − α0 /r
and therefore that χ0 (r) = rφ0 (r) has a simple linear behavior:

χ0 (r) ∝ r − α0 , r > a . (4.4.73)

This shows that α0 corresponds to the value of r where the extrapolation of this behavior
for the wave function crosses zero, that is the intercept of the free wave function evaluated
at r = a:

α0 = intercept of the wave function at r = a . (4.4.74)

It should be emphasized that depending on the value of γ0 , which is the value of the
logarithmic derivative of the s-wave at the boundary of the region where the potential
is non-vanishing, the scattering length α0 may differ significantly from the range of the
potential a. To be more precise, let us distinguish the two situations of repulsive and
attractive potentials. For increasingly strong repulsive potentials, the wave function in
the region r ≤ a is more and more damped, as shown in fig. 4.1. The scattering length
α0 then monotonically increases from 0 up to the maximal value a, which is reached

98
when γ0 = +∞. For increasingly strong attractive potentials, one the contrary, the wave
function in the region r ≤ a is more and more enhanced and can even develop one or several
nodes, as shown in fig. 4.2. The scattering length α0 then monotonically decreases from
0 to −∞, jumps from −∞ to +∞ when the first node appears and γ0 = −1, decreases
monotonically from +∞ to −∞, jumps again from −∞ to +∞ when the second node
appears and γ0 = −1, and so on and so forth.

Χl HrL

a r

Figure 4.1: Pictorial representation of the scattering length for more

and more repulsive potentials.

Χl HrL

a r

Figure 4.2: Pictorial representation of the scattering length for more

and more attractive potentials.

The above picture allows to give simple and physical interpretation to the fact that
α0 → ∞ whenever it happens that γ0 → −1:

α0 → ∞ , when γ0 → −1 . (4.4.75)

Recall for this that γ0 corresponds to the logarithmic derivative of φ0 (r) at r ≃ a, and
therefore 1 + γ0 similarly corresponds to the logarithmic derivative of χ0 (r) at r ≃ a. The
case where γ0 ≃ −1 thus corresponds to a situation in which the logarithmic derivative

99
of χ0 (r) in r ≃ a vanishes and the wave function tends to a small constant at infinity.
This signals the fact that in such a situation the attractive potential develops a bound
state with very small negative energy, and the absolute value of the scattering length then
takes its maximal value, namely |α0 | → +∞, due to a resonance phenomenon involving
this bound state. This effect is perfectly analogous to the resonance on a metastable state
of positive energy E0 and finite width Γ, which appears when E ≃ E0 . In this case, the
resonance is on a stable state of negative but very tiny energy E0 → 0− and vanishing
width Γ → 0, which appears when E → 0+ .
In fact, whenever a very large and positive scattering length αres
0 appears due to the
resonance on a bound state with very small negative energy E0 , one can relate αres 0 to
E0 by comparing the wave functions of the scattering state and the bound state. For
r ≤ a, the two wave functions must approximately coincide, since the former essentially
corresponds to E ≃ 0+ and the latter essentially to E ≃ 0− . For r ≥ a, on the other
hand, the two wave functions are parametrized by different quantities. For the scattering
state we have χ0 (r) ∝ 1 − r/α0 , where α0 is the scattering length. For the bound state
√
we have instead χ0 (r) ∝ e−β0 r , where β0 = −2mE0 /~ is the damping factor associated
to the bound state energy. Comparing these two expressions for r ∼ a and assuming that
β0 a ≪ 1 so that we can expand the exponential at first order, we deduce that α0 ≃ β0−1 ,
that is:
~
αres
0 ≃ √ . (4.4.76)
−2mE0

4.4.1 Example

As a simple example, let us consider the low-energy scattering by a spherical barrier of

the form:
(
V0 , r < a ,
V (r) = (4.4.77)
0, r > a.
The wave equation for r < a is:
 
′′ 2 ′ 2 l(l + 1)
φl (r) + φl (r) + − β − φl (r) = 0 . (4.4.78)
r r2
where:
r √
2m 2mV0
β= −k2 + 2 V0 ≃ . (4.4.79)
~ ~
The general solution is given by spherical Bessel functions with imaginary argument iβr,
and the one that is regular in the origin is:

φl (r) = Al jl (iβr) , r ≤ a . (4.4.80)

From this, we compute that the logarithmic derivative of the wave function at r = a is
given by:
jl′ (iβa)
γl = iβa (4.4.81)
jl (iβa)

100
For the case l = 0, we can use j0 (z) = sin z/z and obtain:

γ0 = βa coth(βa) − 1 . (4.4.82)

It then follows from (4.4.72) that the scattering length and the total low-energy cross
section are given by the following monotonic functions:

tanh(βa) 2 2
   
tanh(βa)
α0 = 1 − a , σtot = 4 1 − πa . (4.4.83)
βa βa
For a very low barrier with small V0 such that βa ≪ 1, representing an easily penetrable
sphere, one finds the following result (which coincides with what one would find by using
the Born approximation and taking the low-energy limit ka ≪ 1):
1 4
α0 ≃ (βa)2 a , σtot ≃ (βa)4 πa2 , for βa ≪ 1 . (4.4.84)
3 3
For a very high barrier with large V0 such that βa ≫ 1, representing an almost unpene-
trable sphere, one finally finds instead:

α0 ≃ a , σtot ≃ 4πa2 , for βa ≫ 1 . (4.4.85)

This behavior is very similar to the one displayed by the low-energy behavior of the reflec-
tion coefficient R for a one-dimensional square barrier, which monotonically increases from
the minimal value 0 to the maximal classical value 1 when the height V0 is increased. In
this three-dimensional case, however, we see that the total cross section σtot monotonically
increases from the minimal value 0 to a maximal value 4πa2 , which is four times larger
than the classical maximal cross section πa2 . This reflects the fact that in the low-energy
regime the scattering process is not semiclassical at all. The enhancement factor of 4 is
due to a maximal quantum interference between the incoming and scattered waves, which
is possible in the three-dimensional setting but not in the one-dimensional setting.
An other simple and interesting example, which can be studied in a very similar way,
is that of the low-energy scattering by a spherical well of the form:
(
−V0 , r < a ,
V (r) = (4.4.86)
0, r > a.

The wave equation for r < a is:

 
′′ 2 ′ 2 l(l + 1)
φl (r) + φl (r) + α − φl (r) = 0 . (4.4.87)
r r2
where:
r √
2m 2mV0
α= k2 + 2 V0 ≃ . (4.4.88)
~ ~
The general solution is given by spherical Bessel functions with real argument αr, and the
one that is regular in the origin is:

φl (r) = Al jl (αr) , r ≤ a . (4.4.89)

101
From this, we compute that the logarithmic derivative of the wave function at r = a is
given by:

jl′ (αa)
γl = αa (4.4.90)
jl (αa)

For the case l = 0, we can use j0 (z) = sin z/z and obtain:

γ0 = αa cot(αa) − 1 . (4.4.91)

It then follows from (4.4.72) that the scattering length and the total low-energy cross
section are given by the following jumping functions:

tan(αa) 2 2
   
tan(αa)
α0 = 1− a , σtot = 4 1 − πa . (4.4.92)
αa αa

For a very low well with small height V0 such that αa ≪ 1, one finds the following result
(which coincides with what one would find by using the Born approximation and taking
the low-energy limit ka ≪ 1):
1 4
α0 ≃ − (αa)2 a , σtot ≃ (αa)4 πa2 , for αa ≪ 1 . (4.4.93)
3 3
For wells with a height V0 such that tan(αa) = αa, one finds a total suppression of the
scattering length and the cross section:

α0 = 0 , σtot = 0 , for tan(αa) = αa . (4.4.94)

For wells with a height V0 such that tan(αa) = +∞, on the contrary, one finds a maximal
enhancement of the scattering length and the cross section:

α0 = +∞ , σtot = +∞ , for tan(αa) = +∞ . (4.4.95)

This behavior is very similar to the one displayed by the low-energy behavior of the
reflection coefficient R for a one-dimensional square well, which displays minima and
maxima for certain discrete values of V0 . In this three-dimensional case, we see that the
total cross section σtot displays minima with the smallest possible value 0 and maxima with
the largest possible value +∞. The first effect is called Ramsauer-Townsend effect and
just corresponds to a situation with maximal distractive interference. The second effect
is due to a resonance phenomenon involving bound states with approximately vanishing
energy. Indeed, it can be easily verified that such bound states with vanishing energy
√
occur when tan(αa) = +∞, where α ≃ 2mV0 /~, that is when the well is such that
√
2mV0 a/~ ≃ (n + 1/2)π.

102
Chapter 5

General operatorial formalism for

scattering problems

In this chapter, we will set up a more general formalism to study scattering problems which
can be applied to any potential, even without spherical symmetry. It is an operatorial
formalism based on the reinterpretation of a scattering process as a transition between
asymptotic free-particle states as a consequence of the action of the interaction potential
during the time evolution of the system. The basic ingredient in this formalism is the
time-evolution operator and its infinite-time limit, which defines the scattering operator.
We will describe how to compute the scattering amplitude and the cross section in this
formalism, and use this to discuss more systematically the perturbative weak-coupling
expansion. We will finally see how this formalism reduces to the one based on phase shifts
and partial waves in the particular case of central problems, and investigate its relation
to the use of Green functions to solve the stationary Schrödinger equation defining the
problem.

5.1 The time-evolution operator

A scattering process can be viewed as a time-dependent process, where the incoming
particle starts as a free particle far away from the interaction region, then feels the potential
only for the limited time during which it is in the vicinity of its center, and finally ends
up again as a free particle far away from the interaction region. It must then be possible
to describe the scattering amplitude in terms of the transition rate between two free
particle states with generically different momenta as a result of the application of the
time dependent perturbation represented by the effect of the potential. Once this is done,
it is possible to compute the scattering amplitude by using time-dependent perturbation
theory, where the crucial ingredient is the time-evolution operator.
To set up the problem, let us start from a general Hamiltonian H which is the sum of
the free kinetic energy H0 = p~ 2 /(2m) and a time-independent potential V (~r):

H = H0 + V . (5.1.1)

103
In the usual Schrödinger picture, the time evolution of the states is determined by the full
Hamiltonian while the time evolution of the operators is trivial:
d
i~ |ψS (t)i = H|ψS (t)i , (5.1.2)
dt
d
i~ OS (t) = 0 . (5.1.3)
dt
The finite form of the time evolution is:
i
|ψS (t)i = e− ~ H(t−t0 ) |ψS (t0 )i , (5.1.4)
OS (t) = OS (t0 ) . (5.1.5)

The usual description in terms of a time-dependent wave function ψS (~r, t) satisfying the
Schrödinger equation can be obtained by using the coordinate representation |~ri of the
Hilbert space and defining:

ψS (~r, t) = h~r|ψS (t)i . (5.1.6)

One may switch from the Schrödinger picture to the interaction picture by performing a
norm-preserving change of basis in the Hilbert space generated by the unitary transfor-
mation ei/~ H0 t . More precisely, we have:
i
|φI (t)i = e ~ H0 t |ψS (t)i , (5.1.7)
i
H t − ~i H0 t
OI (t) = e ~ 0 OS (t) e . (5.1.8)

In the interaction picture, the time evolution of the states is dictated by the potential, while
the time evolution of the operators is dictated by the free Hamiltonian. More precisely,
one easily shows that the time evolution of the states and the operators is determined by
the following differential equations:
d
i~ |φI (t)i = VI (t)|φI (t)i , (5.1.9)
dt
d
i~ OI (t) = [OI (t), H0 ] . (5.1.10)
dt
The finite form of these are given by:
i i i
|φI (t)i = e ~ H0 t e− ~ H (t−t0 ) e− ~ H0 t0 |φI (t0 )i , (5.1.11)
i
H (t−t0 ) − ~i H0 (t−t0 )
OI (t) = e ~ 0 OI (t0 )e . (5.1.12)

Finally, a description in terms of a time-dependent wave function φI (~r, t), which however
reduces to a stationary wave function associated to a given energy in the absence of
interaction potential and has more in general a time dependence that is directly related
to the effect of the potential, can be obtained by using the coordinate representation |~ri
of the Hilbert space and defining:

φI (~r, t) = h~r|φI (t)i . (5.1.13)

104
 The interesting feature of the interaction picture is that the effect of the interac-
tion potential is entirely encoded in the evolution of the states. Indeed, we see form
eq. (5.1.11) that this time evolution can be written in terms of the unitary operator
i i i
UI (t, t0 ) = e ~ H0 t e− ~ H (t−t0 ) e− ~ H0 t0 , which reduces to the identity in the absence of any
potential:
|φI (t)i = UI (t, t0 )|φI (t0 )i . (5.1.14)
However, in this picture the potential is effectively described by the time-dependent op-
erator VI (t), whose form is dictated by the definition (5.1.8):
i i
VI (t) = e ~ H0 t V e− ~ H0 t (5.1.15)
We observe now that since [H0 , V ] 6= 0 in the general situation that we are considering,
UI (t, t0 ) 6= e−i/~ V (t−t0 ) and similarly VI (t) 6= V . However, it turns out that that there
Rt
exists a simple formal relation between UI (t, t0 ) and t0 VI (t′ )dt′ . This can be derived
by finding out the differential equation satisfied by UI (t, t0 ) and then solving it. The
searched differential equation can be deduced by taking a time derivative of (5.1.14) and
using (5.1.9). In this way one deduces that:
d
i~ UI (t, t0 ) = VI (t)UI (t, t0 ) . (5.1.16)
dt
This must now be solved with the boundary condition that
UI (t0 , t0 ) = 1 . (5.1.17)
The simplest way to proceed is to convert this into an integral equation, by integrating it
over time between t0 and t:
i t
Z
UI (t, t0 ) = 1 − VI (t′ ) UI (t′ , t0 ) dt′ . (5.1.18)
~ t0
This equation can be formally solved by iteratively replacing the UI operator appearing
under the integral with the expression defined by the above equation. In this way one
finds:
 2 Z t Z t′
i t −i
Z
′ ′
UI (t, t0 ) = 1 − VI (t ) dt + VI (t′ ) VI (t′′ ) dt′ dt′′ + · · · . (5.1.19)
~ t0 ~ t0 t0

Finally, one may rewrite the multiple integrals in such a way that they all range over
the full time interval from t0 to t, by using the symmetry properties of integration. One
standard way of doing this is to use the time-ordering operator T , which orders the
operators to which it is applied in antichronological order. This allows to formally sum
up the whole series, but it is not directly useful for computing the cross section. An other
trick that one can use, which proves to be more useful in this context, it to introduce an
explicit step function to constrain each successive integration, as we shall see.
At this point, the above result derived for the time-evolution operator in the interac-
tion picture can be used to obtain a similar result for the time-evolution operator in the
Schrödinger picture. Indeed, from the relation between the two pictures it follows that
i i
US (t, t0 ) = e− ~ H0 t UI (t, t0 )e ~ H0 t . (5.1.20)

105
5.2 The scattering matrix
Let us now investigate how one can define a scattering amplitude and a cross section
from the operatorial point of view, where the scattering process is viewed as a transition
induced by a potential. For definiteness we use the interaction picture, but drop the
indices I specifying this choice for simplicity. We moreover start by temporarily putting
the whole system in a box of finite volume, in such a way to work with a discrete spectrum
of free particle states, and postpone for a while the discussion of the continuous spectrum
emerging in the infinite volume limit.
To describe the scattering process, we consider all the possible initial free-particle states
|ii and all the possible final free-particle states |f i, both chosen among the complete set
of free particle states:

|ii = initial free-particle state , (5.2.21)

|f i = final free-particle state . (5.2.22)

We use the canonical normalization for these free particles states in terms of Kronecker
δ-function, so that the orthonormality and completeness relations read:
X
hi|ji = δij , |nihn| = 1 . (5.2.23)
n

The key quantity to describe the dynamics of the scattering process is then the time-
evolution operator U (tf , ti ), which relates the full scattering state |φ(ti )i at some initial
time ti to the full scattering state |φ(tf )i at some final time tf :

|φ(tf )i = U (tf , ti )|φ(ti )i . (5.2.24)

More precisely, we are interested in studying this time evolution for initial times ti → −∞
in the far past and final times tf → +∞ in the far future, because we know that at those
times the state must be a free particle state. We are then led to define the scattering
operator or matrix S as the limit of the evolution operator U (tf , ti ) for these asymptotic
times:

S = U (+∞, −∞) . (5.2.25)

According to the formula (5.1.19), this is given by:

i +∞ i H0 t − i H0 t
Z
S = 1− e~ Ve ~ dt + · · · . (5.2.26)
~ −∞

Now, if the scattering state is chosen in such a way that |φ(−∞)i = |ii at ti → −∞, then
P
it will evolve to |φ(+∞)i = S|φ(−∞)i = S|ii = f hf |S|ii|f i at tf → +∞. This shows
that the probability that from a definite initial state |ii we get a definite final state |f i
is given by the square norm of the matrix element of the S operator between these two
states:

Pi→f = |hf |S|ii|2 . (5.2.27)

106
The scattering operator S is by construction unitary, since the evolution operator U (tf , ti )
is so:

S †S = 1 . (5.2.28)

This ensures that the probability that a given initial state |ii is seen to transform in any
of the possible final states |f i is unity, as it should:
X X
Pi→f = hi|S † |f ihf |S|ii = hi|S † S|ii = 1 (5.2.29)
f f

Notice moreover that in the interaction picture the scattering operator consists of a diago-
nal part which is just the identity and corresponds to the possibility that no diffusion takes
place at all, plus an off-diagonal part which describes the possible non-trivial diffusions.
It is then convenient to parametrize the scattering operator S in terms of a transition
operator T as follows:

S = 1 − iT . (5.2.30)

The unitarity of S implies that T should satisfy the following property:

i(T − T † ) = T † T . (5.2.31)

The explicit expression for T descends from that of S, and one has:

1 +∞ i H0 t − i H0 t
Z
T = e~ Ve ~ dt + · · · . (5.2.32)
~ −∞

In the relevant case where the final state |f i is not identical to the initial state |ii, we can
compute the transition probability by considering the matrix element of the T operator
rather than the S operator:

Pi→f = |hf |T |ii|2 . (5.2.33)

Let us now consider the real situation in which the free particle states form a continuum
and are a labelled by three-dimensional wave vector ~k. In such a situation, one may use
the canonical normalization involving a Dirac δ-function, so that the orthonormality and
completeness relations read:
Z
(3) ~ ~ |nihn| d3~kn = 1 .


hi|ji = δ ki − kj , (5.2.34)

In this situation, we may more sensibly compute the differential probability for the tran-
sition from an initial state |ii with definite wave vector ~ki to a final state |f i with wave
vector ~kf specified with an infinitesimal accuracy d3~kf :

dPi→f = |hf |T |ii|2 d3~kf . (5.2.35)

The evaluation of the matrix element hf |T |ii is in general a difficult task, and the best
that one can do is usually to take a perturbative approach and compute it in the weak

107
potential approximation by retaining only the first few terms of the expansion for the
operator T . The systematics of this expansion will be discussed in a forthcoming section.
There is however a simple and important generic feature that emerges for it, which we
can anticipate: the matrix element hf |T |ii will vanish whenever Ef 6= Ei . To see this,
let us for a moment return to a situation where ti = −τ /2 and tf = τ /2 with a total
time τ that is very large but finite, and check what happens when τ is sent to infinity.
When evaluating the matrix element hf |T |ii, one always encounters an integral over the
time during which the potential is active of the phase ei/~ (Ef −Ei )t that is induced by the
operators e±i/~ H0 t involved in the expression for U (tf , ti ) when they act on the initial
and final states. This integral yields a delta function enforcing energy conservation in the
limit of infinite time τ , as suggested by the uncertainty relation ∆E∆t > ∼ ~:
1 τ /2 i (Ef −Ei )t
 
sin (Ef − Ei )τ /(2~)
Z
e~ dt = → 2πδ(Ef − Ei ) . (5.2.36)
~ −τ /2 (Ef − Ei )/2

We then expect that the general structure for the matrix element hf |T |ii should consist of
the above factor enforcing energy conservation times a residual matrix element for definite
energy:

hf |T |ii = 2π δ(Ef − Ei ) TfEi . (5.2.37)

When squaring this matrix element, one encounters the square of the factor (5.2.36). This
can be evaluated in the following way:
2
sin2 (Ef − Ei )τ /(2~)
 Z τ /2  
1 i
(Ef −Ei )t 2πτ
e ~ dt = 2
→ δ(Ef − Ei ) . (5.2.38)
~ −τ /2 (Ef − Ei ) /4 ~

We then conclude that we can identify 2πδ(0) = τ /~ and:

2π
|hf |T |ii|2 = δ(Ef − Ei ) |TfEi |2 τ . (5.2.39)
~
Because of the special role played by the energy in the computation of the matrix element
appearing in (5.2.35), it is convenient to similarly decompose the differential d3~kf appear-
ing in this quantity. To do so, we observe that a free particle state with wave number ~kf
can be labeled by the modulus of this vector, which is related to the energy Ef , and the
orientation of this vector, which is labelled by a solid angle element Ωf . Recalling that
Ef = (~kf )2 /(2m), we then find that d3~kf = kf2 dkf dΩf = mkf /~2 dEf dΩf . We may
then write

d3~kf = ρ(Ef ) dEf dΩf , (5.2.40)

in terms of the density of states with a given energy Ef , defined as:

mkf
ρ(Ef ) = . (5.2.41)
~2
Using these results, we can finally rewrite (5.2.35) in the following form:
2πτ
dPi→f = δ(Ei − Ef )|TfEi |2 ρ(Ef ) dEf dΩf . (5.2.42)
~

108
Dividing this differential probability by the total time τ , we finally conclude that the
differential rate of transition per unit time is constant and given by
2π
dWi→f = δ(Ef − Ei ) |TfEi |2 ρ(Ef ) dEf dΩf . (5.2.43)
~
We now recognize that this is simply the generalization of Fermi’s golden rule to all orders
in perturbation theory.

5.3 The cross section

Let us now see how one can relate the above differential transition rate defined by the ma-
trix elements of the transition operator T to the differential cross section σ(Ω). According
to its definition (3.7.96), σ(Ω) is given by the ratio of the number of particles that are
scattered per unit time and unit solid angle divided by the number of incoming particles
per unit time and unit transverse surface.
To start, we compute the number of scattered particles per unit time through the solid
angle element dΩf . This is given by (5.2.43) integrated over the final energy Ef , which is
fixed by the δ-function, namely
Z +∞
dWi→f 2π E 2
dW̄i→f = dEf = |T | ρ(Ef ) dΩf (5.3.44)
0 dE f ~ fi
We then divide this by the volume element dΩf itself, to find the number of scattered
particles per unit time and unit solid angle, which is the numerator of the definition of
the differential cross section:
dW̄i→f 2π E 2 2πmk E 2
wsca (Ω) = = |Tf i | ρ(Ef ) = |Tf i | . (5.3.45)
dΩf ~ ~3
We next compute the number of incident particles per unit time and unit surface. This is
just given by the absolute value of the probability density current:

~jinc = − i~ φ∗i ∇φ
 
~ i − φi ∇φ
~ ∗i . (5.3.46)
2m
The wave function φi denotes the stationary wave function describing the initial state |ii.
This must be normalized compatibly with our convention (5.2.34), and is given by:
1 ~
φi (~r) = h~r|ii = 3/2
eiki ·~r . (5.3.47)
(2π)
The modulus of the incident flux, which is the denominator in the definition of the differ-
ential cross section, is then computed to be
~k
jinc = . (5.3.48)
m(2π)3
Finally, the differential cross section is given by the ratio between (5.3.45) and (5.3.48),
and reads:
wsca (Ω) (2π)4 m2 E 2
σ(Ω) = = |Tf i | . (5.3.49)
jinc ~4

109
As a last step, we may now write this as the squared norm of some scattering amplitude
f (Ω), to compare with the approach followed in chapters 3 and 4:

σ(Ω) = |f (Ω)|2 . (5.3.50)

The phase of f (Ω) is not fixed by this reasoning, but we will see that the choice that
precisely corresponds to our previous definition of the scattering amplitude is simply:
(2π)2 m E
f (Ω) = − Tf i . (5.3.51)
~2
The general expression (5.3.51) is the main formula in this formalism. It relates the
scattering amplitude f (Ω) to the matrix elements of the transition operator T . It can now
be proven that the unitarity of the scattering matrix S = 1 − iT implies a more general
version of the optical theorem, valid for completely general static potentials. To prove
this, let us recall that the unitarity of the S operator implies the following relation for the
T operator:

i(T − T † ) = T † T . (5.3.52)

Taking the matrix element of this relation between a free-particle state |ii and a free-
particle state hj|, and inserting a complete set of free-particle states |f i between the two
operators in the right-hand side, we deduce that:
Z
i Tji − Tij∗ = Tf∗j Tf i d3~kf .


(5.3.53)

In the special case where |ji = |ii, one then finds:

1
Z
|Tf i |2 d3~kf .


Im Tii = − (5.3.54)
2
We may now use the expression (5.2.37) for Tii with the identification 2πδ(0) → τ /~ and
the expression (5.2.39) for |Tf i |2 , to rewrite:
τ E
Tii = T , (5.3.55)
~ ii
2πτ
|Tf i |2 = δ(Ef − Ei )|TfEi |2 . (5.3.56)
~
We may then also use (5.2.40) and (5.2.41) to write:
mk
d3~kf = dEf dΩf . (5.3.57)
~2
Plugging back these expressions into (5.3.54), dividing by τ /~ and integrating over Ef , it
the follows that:
πmk
Z
Im TiiE = − 2 |TfEi |2 dΩf .


(5.3.58)
~
Finally, using the relation (5.3.51) between TfEi and f (Ω), one deduces that:

k
Z
Im f (0) = |f (Ω)|2 dΩ . (5.3.59)
4π

110
This generalizes the optical theorem (3.7.126) that we derived with an other method for
the special case of central potentials to generic potentials:
4π
σtot = Im f (0) . (5.3.60)
k

5.4 Systematics of the perturbative expansion

To compute the matrix element defining the scattering amplitude, we may now use a
perturbative expansion in powers of (−i/~)V truncated to some finite order. This will be
justified when the interaction potential is weak and this quantity is small. Notice that
since an inverse power of ~ appears, this expansion is in some sense the opposite of the
semiclassical expansion.
The crucial step that allows to simplify the time integrations defining the series ex-
pansion of the transition operator T and to explicitly factorize an energy-conservation
delta-function to define T E is to introduce the following free time-evolution operator re-
stricted to the future:
i
U0+ (t1 − t2 ) = e− ~ H0 (t1 −t2 ) θ(t1 − t2 ) . (5.4.61)

The operator T may then be rewritten as

∞
X
T = T (n) , (5.4.62)
n=1

where the operators T (n) are given by

  Z +∞
(1) −i i i
T = i e ~ H0 t V e− ~ H0 t dt , (5.4.63)
~ −∞
 2 Z Z +∞
−i i i ′
T (2)
= i e ~ H0 t V U0+ (t − t′ )V e− ~ H0 t dt dt′ , (5.4.64)
~ −∞
 3 Z Z Z +∞
−i i i ′′
T (3)
= i e ~ H0 t V U0+ (t − t′ )V U0+ (t′ − t′′ )V e− ~ H0 t dt dt′ dt′′ , (5.4.65)
~ −∞

···

To evalutate the matrix elements of the various contributions T (n) to the operator T , one
can now use the following Fourrier integral representation of U0+ (t1 − t2 ), where the small
positive number ǫ defines a prescription on how to perform the integration in the vicinity
of H0 , which is crucial to reproduce the θ(t1 − t2 ) factor in U0+ (t1 − t2 ):
+∞
i 1
Z
i
U0+ (t1 − t2 ) = e− ~ E(t1 −t2 ) dE . (5.4.66)
2π −∞ E − H0 + iǫ

One may verify this expression by extending the integral to the complex E plane. One
i
can then close the integration contour at complex ∞ in such a way the factor e− ~ E(t1 −t2 )
vanishes on the added part of the contour and compute the integral with the help of

111
Cauchy’s theorem. When t1 − t2 > 0, one has to close the contour in the lower half plane,
and the result of the integral is non-vanishing, since the integrand has a pole at E = H0 −iǫ
in such a half plane. When instead t1 − t2 < 0, one has to close the contour in the upper
half plane, and the result vanishes, since the integrand has no poles in such a half plane.
One may then also interpret the above expression in terms of a real integral. Indeed, from
that perspective one has to split the integral into a real principal part plus an imaginary
localized contribution, and this amounts to using the following prescription:
1 . 1
=P − iπδ(E − H0 ) . (5.4.67)
E − H0 + iǫ E − H0
We will now see that by using the representation (5.4.66), one may compute more explicitly
the matrix elements hf |T (n) |ii and bring them in the form that we have already anticipated
to compute the scattering amplitude, namely:

hf |T (n) |ii = 2πδ(Ef − Ei ) TfE(n)

i . (5.4.68)

One may then compute the matrix element TfEi controlling the scattering amplitude as
the following series:
∞
X
TfEi = TfE(n)
i . (5.4.69)
n=1

The first-order contribution TfE(1)

i is defined through the matrix element of the operator
T (1) given by (5.4.63). This is computed to be
1 +∞ i (Ef −Ei )t
Z
(1)
hf |T |ii = e~ dt hf |V |ii
~ −∞
= 2πδ(Ef − Ei ) hf |V |ii . (5.4.70)

It follows that:

TfE(1)
i = hf |V |ii . (5.4.71)

With a shorter notation, we may write this simply as:

TfE(1)
i = Vf i . (5.4.72)

At first order in perturbation theory, the transition amplitude on the fixed-energy shell is
thus simply proportional to the matrix element of the interaction potential. This is the
analogue of the fact that in time-independent perturbation theory the first order correction
to the energy levels is given by the expectation value of the interaction potential.
The second-order contribution TfE(2) i is defined through the matrix element of the
operator T (2) given by (5.4.64). This is computed to be
Z +∞ Z +∞
1 1 1 +∞ i (E−Ei )t′ ′ 1
Z
i
(2) (Ef −E)t
hf |T |ii = e ~ dt e~ dt hf |V V |ii dE
2π −∞ ~ −∞ ~ −∞ E − H0 + iǫ
Z +∞
1 1
= 2πδ(Ef − E) 2πδ(E − Ei ) hf |V V |ii dE
2π −∞ E − H0 + iǫ
1
= 2πδ(Ei − Ef )hf |V V |ii . (5.4.73)
Ei − H0 + iǫ

112
It follows, denoting now the conserved energy by Ei = Ef = E, that:

1
TfE(2)
i = hf |V V |ii . (5.4.74)
E − H0 + iǫ
Inserting a complete set of free-particle states between each operator, we may finally write
this in the following form, with Ek′ = (~2 k′ )2 /(2m):

Vf k′ Vk′ i
Z
TfE(2)
i = d3~k′ . (5.4.75)
E − Ek′ + iǫ
At second order in perturbation theory, the transition amplitude on the fixed-energy
shell thus receives a contribution involving two matrix elements of the interaction po-
tential, where the transition proceeds via an arbitrary virtual intermediate state with
unconstrained wave number. This is the analogue of the fact that in time-independent
perturbation theory the second order correction to the energy levels involves two matrix
elements of the interaction potential and an intermediate state with unconstrained energy.
The n-order contribution TfE(n)
i is defined through the matrix element of the operator
T (n) .Proceeding in a similar way as before, this is computed to be
1 1
TfE(n)
i = hi|V V V · · · |f i . (5.4.76)
E − H0 + iǫ E − H0 + iǫ
Inserting a complete set of free-particle states between each operator, and using the no-
tation Ek′ = (~2 k′ )2 /(2m), Ek′′ = (~2 k′′ )2 /(2m), · · · , we may finally write this as:

Vf k′ Vk′ k′′ Vk′′ k′′′ · · ·

ZZ
E(n)
Tf i = ···

d3~k′ d3~k′′ · · · . (5.4.77)
E − Ek′ + iǫ Ek′ − Ek′′ + iǫ · · ·

At the n-th order in perturbation theory, the transition amplitude on the fixed-energy shell
thus receives a contribution involving n matrix elements of the interaction potential, where
the transition proceeds via n − 1 arbitrary virtual intermediate states with unconstrained
wave numbers. This is again perfectly analogous to what happens in time-independent
perturbation theory for the n-th order correction to the energy levels.
Before evaluating more explicitly the form of the various contributions TfE(n) i , let us
E E
notice that from the above results for the matrix element Tf i = hf |T |ii for arbitrary free
particles states it follows that the operator T E is given by the following expression:
1 1 1
TE = V + V V +V V V + ···
E − H0 + iǫ E − H0 + iǫ E − H0 + iǫ

h 1 1 1
= V +V + V + ··· V
E − H0 + iǫ E − H0 + iǫ E − H0 + iǫ
1
= V +V V (5.4.78)
E − H0 − V + iǫ

Let us now evaluate more explicitly the first few contributions to the matrix element
defining the scattering amplitude. To do so, we first compute the matrix element of the
potential between two generic free particle states. With the normalization that we have

113
chosen, this is found to be:
1
Z Z
~ ~
Vij = hi|V |ji = hi|~riV (~r)h~r|ji d3 ~r = e−i(ki −kj )·~r V (~r) d3~r
(2π)3
1
= Ṽ (~ki − ~kj ) . (5.4.79)
(2π)3
It follows that
Vf k ′ Vk ′ i
Z
TfEi = Vf i + d3~k′ + · · · (5.4.80)
E − Ek′ + iǫ
1 1 Ṽ (~kf − ~k′ )Ṽ (~k′ − ~ki ) 3~ ′
Z
= Ṽ (~
k f − ~
k i ) + d k + ··· .
(2π)3 (2π)6 E − Ek′ + iǫ
We conclude from this that the scattering amplitude is given by:

Ṽ (~kf − ~k′ )Ṽ (~k′ − ~ki ) 3~ ′

 
m 1
Z
~ ~
f (ki , kf ) = − ~ ~
Ṽ (kf − ki ) + d k + · · · . (5.4.81)
2π~2 (2π)3 E − Ek′ + iǫ

We see that at first order we recover the Born approximation for the differential cross
section, which depends only on the transferred momentum ~q = ~kf − ~ki and is given by the
following expression, which coincides with the result (4.2.19) that we have already derived
with an other method for the special case of central potentials:
m
f (~q) ≃ − Ṽ (~
q) . (5.4.82)
2π~2
Moreover, we can now explicitly verify that the optical theorem is satisfied, provided one
works at the appropriate order in the perturbative expansion to evaluate its left-hand and
right-hand sides. Indeed, the leading contribution (5.4.82) to the scattering amplitude is
real at q = 0, and the first contribution to the imaginary part of the scattering amplitude
at Ω = 0 therefore arises from the second-order term, as a consequence of the prescription
(5.4.67). By decomposing d3 k′ = mk/~2 dEk′ dΩ′ , one easily shows that:
k  m 2
Z
Im f (0) ≃ q )|2 dΩ .
|Ṽ (~ (5.4.83)
4π 2π~2
The expressions (5.4.82) and (5.4.83) show that the relation (5.3.59) and thus the optical
theorem is indeed approximately satisfied.

5.5 Central problems

In the special case of central problems with a spherical symmetry, one may choose the
states as eigenstates of L2 and Lz labeled by the quantum numbers l and m. To do so,
one may start from the usual description of free particle states in terms of a wave vector ~k,
which treats the three dimensions on equal footing, then switch to a description in terms
of a wave number k and a solid angle direction Ω, which refer to the radial and angular
parts of the problem, and finally switch to a description in terms of an energy E for the
radial part and an angular momentum specified by L2 and Lz for the angular part.

114
 To perform the above change of basis properly, one must be careful with normaliza-
tions. To this purpose, let us recall that the asymptotic states |ii have been chosen with
a normalization such that hi|ji = δ(3) (~ki − ~kj ). When switching from the description in
terms of ~ki to the description in terms of Ei and Ωi , one may then write:

hi|ji = ρ−1 (Ei )δ(Ei − Ej )δ(2) (Ωi − Ωj ) , (5.5.84)

where
mk
ρ(E) = . (5.5.85)
~2
From this we deduce that we may describe the full states |ii in terms of a radial part |Ei i
and an angular part Ωi as
1
|ii = √ ρ−1/2 (Ei )|Ei i ⊗ |Ωi i , (5.5.86)
2π
where the states |Ei i and |Ωi i are normalized in the following way:

hEi |Ej i = 2πδ(Ei − Ej ) , (5.5.87)

hΩi |Ωj i = δ(2) (Ωi − Ωj ) , (5.5.88)

With this decomposition, the expectation value of an operator O at fixed energy, as defined
E , can be rewritten more explicitly in the following
by the relation Oij = 2πδ(Ei − Ej )Oij
way:

E 1 −1
Oij = ρ (Ei )hΩi |O|Ωj i . (5.5.89)
2π
At this point, we may perform any orthonormal change of basis we like in the angular
part. In particular, we may switch from the states |Ωi i to the states |l, mi by writing
+∞ X
X l
|Ωi i = Ylm∗ (Ωi )|l, mi , (5.5.90)
l=0 m=−l

where the states |l, mi are canonically normalized:

hl, m|l′ , m′ i = δl,l′ δm,m′ , (5.5.91)

and the spherical harmonic functions Ylm (Ωi ) are defined by the overlap

Ylm (Ωi ) = hΩi |l, mi . (5.5.92)

In the new basis |l, mi, the expectation value of an operator O which is invariant under
rotations and commutes with L2 and Lz is diagonal and depends only on l and not on m,
since the choice of the z axis is clearly arbitrary:

hl, m|O|l′ , m′ i = δll′ δmm′ Ol . (5.5.93)

115
It then follows that the expectation value of such an operator in the original basis |Ωi i
can be decomposed as follows:
l
+∞ X +∞ X
l ′
X X ′
hΩi |O|Ωj i = Ylm∗ (Ωi )Ylm ′ ′
′ (Ωj )hl, m|O|l , m i

l=0 m=−l l′ =0 m′ =−l′

+∞
X l
X
= Ylm∗ (Ωi )Ylm (Ωj )Ol . (5.5.94)
l=0 m=−l

Using at this point the relation

l
X 2l + 1
Ylm∗ (Ωi )Ylm (Ωj ) = Pl (cos θ) , (5.5.95)
4π
m=−l

where θ is the relative angle between Ωi and Ωj , this matrix element can be written as a
partial wave expansion:
+∞
1 X
hΩi |O|Ωj i = (2l + 1)Ol Pl (cos θ) . (5.5.96)
4π
l=0

Finally, the expectation value of a rotation-invariant operator O at fixed energy can there-
fore be written in the following form:
+∞
E ~2 X
Oij = (2l + 1)Ol Pl (cos θ) . (5.5.97)
8π 2 mk
l=0

At this point, we may use the above results to rewrite the scattering amplitude as a
partial wave expansion:
+∞
1 X
f (Ω) = − (2l + 1)Tl Pl (cos θ) . (5.5.98)
2k
l=0

Recalling that T = i(S − 1), this may also be written as:

+∞
1 X

f (Ω) = (2l + 1) Sl − 1 Pl (cos θ) , (5.5.99)
2ik
l=0

Comparing with the expression of the scattering amplitude in terms of phase shifts, we
conclude that the scattering matrix elements Sl are related to the phase shifts δl by the
following very simple relation:

Sl = e2iδl . (5.5.100)

Notice finally that in this basis the unitarity of the scattering matrix S implies the condi-
tion |Sl | = 1, which translates into the fact that the phase shifts for an elastic scattering
process must be real: Im δl = 0.

116
5.6 Relation to Green functions
As a final remark concerning the general formalism presented in this section, let us explore
the connection with Green functions and the tool that these offer for solving wave equations
like the Schrödinger equation through a perturbative expansion. We have seen in previous
section that a crucial ingredient in the perturbative expansion of the scattering amplitude
is the operator 1/(E − H0 + iǫ). It turns out that the matrix elements of this operator
are directly related to Green functions.
The matrix elements of the operator 1/(E − H0 + iǫ) in wave-vector space are trivially
computed. Denoting E = (~k)2 /(2m) and Ek′ = (~k′ )2 /(2m), one finds:
1
h~k′ | |~k′′ i = G̃(~k′ ) δ(3) (~k′ − ~k′′ ) , (5.6.101)
E − H0 + iǫ

where the function G̃ is given by

1 2m 1
G̃(~k′ ) = = 2 2 . (5.6.102)
E − Ek′ + iǫ ~ k − k′2 + iǫ
This trivially satisfies the algebraic equation:

~2 2
E − Ek′ G̃(~k′ ) = k − k′2 G̃(~k′ ) = 1 .



(5.6.103)
2m
The matrix elements of this same operator in configuration space can instead be computed
by inserting twice a complete set of states. One finds:
1 1
ZZ
h~r ′ | |~r ′′ i = h~r ′ |~k′ ih~k′ | |~k′′ ih~k′′ |~r ′′ i d3~k′ d3~k′′
E − H0 + iǫ E − H0 + iǫ
1 1
Z
~ ′ ′ ′′
= eik ·(~r −~r ) d3~k′
(2π) 3 E − Ek′ + iǫ
Z +∞Z 1
m ik ′ |~
r ′−~r ′′ | cos θ k′2
= e d cos θ dk′
2π 2 ~2 0 −1 k 2 − k ′2 + iǫ
Z +∞
m 1 k′
sin k′ |~r ′ − ~r ′′ | dk ′ (5.6.104)


= 2 2 ′ ′′ 2 ′2
.
π ~ |~r − ~r | 0 k − k + iǫ
To evaluate the last integral explicitly, we may now use complex integration techniques,
promoting k′ to a complex variable z. We notice for this that the integrand is even in z
and has simple poles in z = ±k ± iǫ/(2k). Using the short-hand notation r = |~r ′ − ~r ′′ |,
we may then write:
Z +∞
z 1 +∞ z
Z
2 2
sin(zr) dz = sin(zr) dz
0 k − z + iǫ 2 −∞ k − z 2 + iǫ
2

i z i z
Z Z
izr
= − 2 2
e dz + e−izr dz
4 c+ k − z + iǫ 4 c− k − z 2 + iǫ
2

π n z eizr o π n z e−izr o
= Res 2 + Res
2 k − z 2 z=k 2 k2 − z 2 z=−k
π
= − eikr (5.6.105)
2

117
It finally follows that
i
h~r ′ | |~r ′′ i = G(~r ′ − ~r ′′ ) , (5.6.106)
E − H0 + iǫ
where the function G is by construction the three-dimensional Fourier antitransform of G̃
and has the explicit form
′ ′′
2m eik|~r −~r |
G(~r ′ − ~r ′′ ) = − . (5.6.107)
~2 4π|~r ′ − ~r ′′ |
This satisfies the following differential equation, which defines the Green function of the
Schrödinger wave operator and represents the configuration space version of (5.6.103):
~2
−H0′ + E G(~r ′ − ~r ′′ ) = ∆′ + k2 G(~r ′ − ~r ′′ ) = δ(3) (~r ′ − ~r ′′ ) .



(5.6.108)
2m
We may now reexpress the scattering amplitude (5.4.81) in terms of the potential and
the above defined Green function. Using the wave-vector space versions Ṽ and G̃ of the
potential and the Green function, one simply has

~ ~ m
f (ki , kf ) = − Ṽ (~kf − ~ki )
2π~2

1
Z
+ ~ ~ ′ ~ ′ ~ ′ ~ 3~ ′
Ṽ (kf − k )G̃(k )Ṽ (k − ki ) d k + · · · . (5.6.109)
(2π)3

Using instead the configuration space versions V and G̃ of the potential and the Green
function, this expression becomes:
Z
~ ~ m ~ ′ ~ ′
f (ki , kf ) = − 2
e−ikf ·~r V (~r ′ ) eiki ·~r d3~r ′ (5.6.110)
2π~
Z 
−i~kf ·~r′ ′ ′ ′′ ′′ i~ki ·~
r ′′ 3 ′ 3 ′′
+ e V (~r )G(~r − ~r )V (~r ) e d ~r d ~r + · · · .

Let us now come back to the scattering problem we want to solve. This consists in
finding the solution with appropriate boundary conditions of the stationary Schrödinger
equation, which can be rewritten in the following form:
~2
∆ + k2 φ(~r) = V (~r)φ(~r) .



−H0 + E φ(~r) = (5.6.111)
2m
Using the Green function G, the general solution of this equation can now be formally
written in terms of a generic solution φ0 of the free equation as:
Z
φ(~r) = φ0 (~r) + G(~r − ~r ′ )V (~r ′ )φ(~r ′ )d3~r ′ . (5.6.112)

This is an integral equation for φ(~r), which can be solved in the usual way by iterating it.
In this way one finds:
Z
φ(~r) = φ0 (~r) + G(~r − ~r ′ )V (~r ′ )φ0 (~r ′ ) d3~r ′ (5.6.113)
ZZ
+ G(~r − ~r ′ )V (~r ′ )G(~r ′ − ~r ′′ )V (~r ′′ )φ0 (~r ′′ ) d3~r ′ d3~r ′′ + · · · .

118
The choice of the free solution φ0 (~r) is related to the boundary conditions for the true
solution φ(~r) at large r. More precisely, it must corresponds to the incident wave, since
φ(~r) → φ0 (~r) when V (~r) → 0. We then take φ0 (~r) to be a progressive wave with wave
vector ~ki = kẑ defining the incident wave vector and a conventional normalization equal
to unity:
~
φ0 (~r) = eiki ·~r . (5.6.114)

The remaining part of the solution φ(~r) must then correspond to the scattered wave. To
extract the scattering amplitude, we must then study the asymptotic behavior of this part
for positions ~r with large modulus r. We expect to find a progressive wave with a wave
vector ~kf = kr̂ defining the scattered wave number and an amplitude suppressed by 1/r.
To see that this is indeed what emerges, we can go back to eq. (5.6.112) and imagine that
the potential has a finite range a, beyond which it is negligible. In such a situation, the
integration over ~r ′ may be safely limited to r ′ <
∼ a. One can then consider the asymptotic
region where r ≫ a and thus r ≫ r ′ . In such a situation, ~r − ~r ′ differs only little from ~r,
and denoting by α the angle between ~r and ~r ′ , one computes, at leading order in r ′ /r:
r
p r ′ r ′2
|~r − ~r | = r + r − 2rr cos α = r 1 − 2 cos α + 2 ≃ r − cos α r ′
′ 2 ′2 ′
r r
≃ r − r̂ · ~r ′ . (5.6.115)

One may then approximate the Green function with the following expression, after recall-
ing that kr̂ is interpreted as the scattered wave vector ~kf :
m 1 ikr −i~kf ·~r ′
G(~r − ~r ′ ) ≃ − e e . (5.6.116)
2π~2 r
In this region of large r, the integral equation (5.6.112) then implies that the exact scat-
tering solution φ(~r) has indeed the expected type of asymptotic behavior, namely

~ f (~ki , ~kf ) ikr

φ(~r) ≃ eiki ·~r + e , (5.6.117)
r
with a scattering amplitude identified with the following expression:
m
Z
~ ~ ~ ′
f (ki , kf ) = − e−ikf ·~r V (~r ′ )φ(~r ′ )d3~r ′ (5.6.118)
2π~2
~ ′
Using finally the formal solution (5.6.113) for φ(~r) with φ0 (~r ′ ) = eiki ·~r , one deduces that
the scattering amplitude f (~ki , ~kf ) can be written in the form (5.6.110), or equivalently as
(5.6.109) after switching to wave-vector space.
We therefore see that the general result (5.6.109) or (5.6.110) for the scattering am-
plitude obtained in the operatorial approach can also be derived by using just the Green
function of the Schrödinger wave equation. The operatorial formalism has however the
advantage over the Green function approach of being more easily and directly general-
izable to describe also more complicated situations, like for instance inelastic scattering
processes.

119
120
Chapter 6

Approximation methods for

many-body problems

In this final chapter, we shall briefly describe two general classes of approximation methods
that can be used to face problems with many identical particles. The first method is based
on the idea that a many-body system might be effectively described in terms of a collection
of individual particles that behave independently but are subject to an effective potential,
which takes into account the presence of all the particles and their mutual interaction.
This effective potential is then self-consistently determined in such a way to minimize the
energy of the full system. The second method is based on the idea that a many-body
system might be effectively described in terms of a density of particles. This density is
then self-consistently determined in such a way to minimize the energy of the full system.
We shall then discuss the application of these methods to many-electron atoms.

6.1 Individual wave functions approach

Let us consider the problem of a system of N identical particles with coordinates ~ri and
spins si , where each particle is subject to the action of a potential V (~ri ), which depends
only on the position ~ri of each particle relative to the center, and each pair of particles is
subject to a mutual interaction W (~ri − ~rj ), which depends only on the relative position
~ri − ~rj of the two particles. The Hamiltonian describing the full system then takes the
following form:
X  ~2 
1 X
H= − ∆i + V (~ri ) + W (~ri − ~rj ) . (6.1.1)
2m 2
i i,j diff.

Under suitable circumstances, one may expect that such a system could be reasonably
well described in terms of individual wave functions, whose form take to some extent into
account the interaction with the other particles. To determine the best approximation of
this kind, we may consider a family of multi-particle wave functions constructed in terms
of such individual wave functions, and then apply the variational method to determine
the best approximation to the true ground state within this reduced set of wave functions.

121
6.1.1 The Hartree approximation

The simplest possibility is to use a family of trial wave functions taking simply the form
of a product of individual stationary wave functions for each particle, ignoring for the
moment their statistics:

φ(~r1 , s1 , · · · , ~rn , sn ) = φ1 (~r1 )χ1 (s1 ) · · · φN (~rN )χN (sn ) . (6.1.2)

The expectation value of H on such a trial function is easily found to be:

~2 X
Z XZ
∗ 3
hφ|H|φi = − φi (~ri ) ∆i φi (~ri ) d ~ri + V (~ri ) |φi (~ri )|2 d3~ri
2m
i i
1 X
ZZ
+ W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj . (6.1.3)
2
i,j diff.

The norm of each individual wave function is instead given by:

Z
hφi |φi i = |φi (~ri )|2 d3~ri . (6.1.4)

We can now determine the individual wave functions φi by minimizing the above expec-
tation value hφ|H|φi for the energy. In doing so, we impose the constraint that each
of the individual wave functions φi should be normalized to unity, namely hφi |φi i = 1.
This ensures not only that the full wave function is properly normalized, but also that it
can be effectively interpreted as the product of independent wave functions for individual
particles. This constrained minimization problem can be set up with the usual method of
Lagrange multipliers. Since we have N independent constraints we need to introduce N
Lagrange multipliers Ei and extremize the following functional:
X  
E = hφ|H|φi − Ei hφi |φi i − 1 . (6.1.5)
i

Using the expressions (6.1.3) and (6.1.4), we are then led to the following functional:

~2 X
Z XZ
E = − φ∗i (~ri ) ∆i φi (~ri ) d3 ~ri + V (~ri ) |φi (~ri )|2 d3~ri
2m
i i
1 X
ZZ
+ W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj
2
i,j diff.
X Z 
2 3
− Ei |φi (~ri )| d ~ri − 1 . (6.1.6)
i

The stationarity equations following from the invariance of E with respect to arbitrary
small variations of φ∗i give the following set of N coupled non-linear differential equations,
which determine the φi for given values of Ei :

~2
 XZ 
2 3
− ∆i + V (~ri ) + W (~ri − ~rj ) |φj (~rj )| d ~rj φi (~ri ) = Ei φi (~ri ) . (6.1.7)
2m
j6=i

122
Notice that these equations have the structure of N stationary Schrödinger equations for
the individual wave functions φi with energies Ei , but with some effective potentials which
themselves depend on the form of the other wave functions φj with j 6= i and take into
account the mutual interaction energy with the density of the remaining particles. The
values of Ei are then fixed by the stationary conditions for E with respect to variations of
Ei , which are noting but the constraints that the individual wave functions φi should be
properly normalized:
Z
|φi (~ri )|2 d3~ri = 1 . (6.1.8)

Multiplying eq. (6.1.7) by φ∗i (~ri ) and integrating over ~ri , one finds:

~2
Z Z
Ei = − φi (~ri ) ∆i φi (~ri ) d ~ri + V (~ri ) |φi (~ri )|2 d3~ri
∗ 3
2m
XZ Z
+ W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj . (6.1.9)
j6=i

Once the wave functions φi and the numbers Ei have been derived in this way, the total
energy of the system is given by the value of the functional E at this extremum. This is
found to be given by:
1 X
X ZZ
E= Ei − W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj . (6.1.10)
2
i i,j diff.

An important shortcoming of this approach, which is called the Hartree method, is that it
does not properly implement the quantum statistics of identical particles. In particular, for
fermionic particles it does not incorporate Pauli’s exclusion principle. One way to partly
remediate to this problem is to keep the same kind of trial function but to somehow
impose the additional requirement that the individual wave functions φi should satisfy
some orthogonality constraints, in such a way to ensure that all the individual fermions
are in a different state. This does however not come out automatically and it is not entirely
straightforward to implement this procedure. In particular, two individual wave functions
φi and φj are in general not orthogonal, even when Ei and Ej are different, because they
do not represent two different eigenvectors of a single differential operator but rather the
eigenvectors of two different differential operators.

6.1.2 The Hartree-Fock approximation

A more specific possibility to describe systems of identical fermions is to start from a trial
function that is completely antisymmetric under permutations, in such a way to properly
incorporate the fermonic statistics. This can be taken to be of the standard form of a
Slater determinant built out of individual wave functions:
φ1 (~r1 )χ1 (s1 ) · · · φ1 (~rN )χ1 (sN )
1 ·· ··
φ(~r1 , s1 , · · · , ~rn , sn ) = √ · · . (6.1.11)
N!
φN (~r1 )χN (s1 ) · · · φN (~rN )χN (sN )

123
The expectation value of H on such a trial function can now be computed. But this re-
quires slightly more care than before, to take into account the effect of all the permutation
terms building up the determinant, besides the one coming from the main diagonal. The
result can be derived in a systematic way by rewriting the determinant in terms of a sum
over permutations, and going through a sequence of simple manipulations on this sum of
permutations. The final result is that one simply finds an additional exchange term for
each pair of individual particles with identical spins:

~2 X
Z XZ
∗ 3
hφ|H|φi = − φi (~ri ) ∆i φi (~ri ) d ~ri + V (~ri ) |φi (~ri )|2 d3~ri
2m
i i
1 X
ZZ
+ W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj (6.1.12)
2
i,j diff
1 X
ZZ
− δsi ,sj W (~ri − ~rj ) φ∗i (~ri ) φj (~ri ) φ∗j (~rj ) φi (~rj ) d3~ri d3~rj .
2
i,j diff

The norm of each individual wave function is instead given by:

Z
hφi |φi i = |φi (~ri )|2 d3~ri . (6.1.13)

We can now proceed exactly as before and determine the individual wave functions φi by
minimizing the above expectation value hφ|H|φi for the energy. In doing so, we impose
again the constraint that each of the individual wave functions φi should be normalized to
unity, namely hφi |φi i = 1. To set up this constrained minimization problem we introduce
as before N Lagrange multipliers Ei and extremize the following functional:
X  
E = hφ|H|φi − Ei hφi |φi i − 1 . (6.1.14)
i

Using the expressions (6.1.12) and (6.1.13), and extending the sums over i, j diff to run
over all the values of i, j by exploiting the fact that when i = j the two W -dependent
terms cancel each other, we are then led to the following functional:

~2 X
Z XZ
E = − φ∗i (~ri ) ∆i φi (~ri ) d3 ~ri + V (~ri ) |φi (~ri )|2 d3~ri
2m
i i
1X
ZZ
+ W (~ri − ~rj ) |φi (~ri )| |φj (~rj )|2 d3~ri d3~rj
2
2
i,j
1X
ZZ
− δsi ,sj W (~ri − ~rj ) φ∗i (~ri ) φj (~ri ) φ∗j (~rj ) φi (~rj ) d3~ri d3~rj
2
i,j
X Z 
2 3
− Ei |φi (~ri )| d ~ri − 1 . (6.1.15)
i

The stationarity equations following from the invariance of E with respect to arbitrary
small variations of φ∗i give the following set of N coupled non-linear differential equations,

124
which determine the φi for given values of Ei :

~2
 XZ 
2 3
− ∆i + V (~ri ) + W (~ri − ~rj ) |φj (~rj )| d ~rj φi (~ri )
2m
j
X Z 
∗ 3
− δsi ,sj W (~ri − ~rj ) φj (~rj ) φi (~rj ) d ~rj φj (~ri ) = Ei φi (~ri ) . (6.1.16)
j

These equations have again a structure that is similar to N stationary Schrödinger equa-
tions for the individual wave functions φi with energies Ei , but with some effective po-
tentials which depend on the form of all the wave functions φi and take into account not
only the mutual interaction energy but also an exchange term due to the statistics. The
values of Ei are then fixed by the stationary conditions for E with respect to variations of
Ei , which reduce as before to the constraints that the φi should be properly normalized:
Z
|φi (~ri )|2 d3~ri = 1 . (6.1.17)

Multiplying eq. (6.1.16) by φ∗i (~ri ) and integrating over ~ri , one finds:

~2
Z Z
Ei = − φ∗i (~ri ) ∆i φi (~ri ) d3~ri + V (~ri ) |φi (~ri )|2 d3~ri
2m
XZ Z
+ W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj
j
X ZZ
− δsi ,sj W (~ri − ~rj ) φ∗i (~ri ) φj (~ri ) φ∗j (~rj ) φi (~rj ) d3~ri d3~rj . (6.1.18)
j

Once the wave functions φi and the numbers Ei have been derived in this way, the total
energy of the system is given by the value of the function E at this extremum. This is
found to be given by:
1X
X ZZ
E = Ei − W (~ri − ~rj ) |φi (~ri )|2 |φj (~rj )|2 d3~ri d3~rj
2
i i,j
1X
ZZ
+ δsi ,sj W (~ri − ~rj ) φ∗i (~ri ) φj (~ri ) φ∗j (~rj ) φi (~rj ) d3~ri d3~rj . (6.1.19)
2
i,j

A crucial advantage of this approach, which is called the Hartree-Fock method, is that it
properly implements the fermionic statistics and Pauli’s exclusion principle. In fact, two
individual wave functions φi and φj are automatically orthogonal to each other whenever
Ei and Ej are different, because they correspond to two different eigenvectors of one and
the same differential operator, although non-linear.

6.1.3 Self-consistent field solution

The above approach, based on individual wave functions, allows to approximate the so-
lutions of the original single linear differential equation involving 3N different variables
with the solutions of a system of N coupled non-linear equations each depending only on 3

125
variables. With this approximation, an analytical determination of the solution is in gen-
eral still impossible, but a reliable numerical solution becomes possible. A typical method
to solve the non-linear equations (6.1.7) or (6.1.16) is to start with an educated guess for
the functions φi , and then try to determine their true form by making an iterative use of
the equations. More precisely, one may start by setting in first approximation the φi to be
equal to the functions that would solve the equations (6.1.7) or (6.1.16) in the linearized
approximation. One may then use this solution to evaluate the non-linear W -dependent
terms, and solve the full non-linear equation with these terms fixed to the computed value,
to determine a better form for φi . Iterating this process one may then determine the true
solution of the non-linear equations, whenever it converges to a definite result. This then
determines self-consistent solutions for the wave functions φi and the energies Ei , which
can be associated to individual particles, and the effective potential that they feel.
The precision of this approach is intrinsically limited by the restriction of the set
of trial wave functions to those parametrized in terms of uncorrelated individual wave
functions. To improve the precision, one needs to include gradually more general wave-
functions allowing for some correlation between these individual modes, and this is not
easy to implement in a systematic way. However, this type of approximation method has
an interesting general property: being based on a variational approach, it always yields
upper bounds on the true ground state energy.

6.2 Particle density approach

Let us consider again the same problem of a system of N identical particles with coordi-
nates ~ri , where each particle is subject to the action of a potential V (~ri ), which depends
only on the position ~ri of each particle relative to the center, and each pair of particles is
subject to a mutual interaction W (~ri − ~rj ), which depends only on the relative position
~ri − ~rj of the two particles. The Hamiltonian describing the full system then takes the
same form as before, namely:
X  ~2 
1 X
H= − ∆i + V (~ri ) + W (~ri − ~rj ) . (6.2.20)
2m 2
i i,j diff.

Under favorable circumstances, it may happen that such a system could be reasonably
well described in the semi-classical approximation and in a statistical way, in terms of
a density of particles. To determine the best approximation of this kind, we may then
work out the total energy as a functional of this density of particles and require this to
be extremal.

6.2.1 The Thomas-Fermi approximation

Let us start by assuming that the potential can be considered as slowly varying and
that the semiclassical approximation is justified. We may then reason in classical terms
and think of the ground state of the system as a configuration where all the lowest-
lying quantum states are optimally occupied, compatibly with Pauli’s exclusion principle.

126
Locally, in the neighborhood of some position ~r, this means that electrons can have energies
up to some maximal value EF (~r) and thus momenta up to some maximal momentum pF (~r),
the relation between these two maximal quantities being:
q

pF (~r) = 2m EF (~r) − V (~r) . (6.2.21)

In such a neighborhood, we may then treat the particles in first approximation as a non-
interacting Fermi gas. This allows to relate the density of these particles to the above
maximal energy or momentum, at each point ~r. To find this relation, we may use the
semiclassical picture according to which the density of quantum states is given by the
action in full phase space divided by hd , where d is the number of degrees of freedom. In
our three-dimensional case, the phase space volume dS corresponding to particles with
momentum less than pF and position within some volume dV is given by dS = 43 πp3F dV .
The corresponding number of quantum states is obtained by diving this by h3 and is
thus given by d# = 43 π(pF /h)3 dV , or equivalently d# = 6π1 2 kF dV . The total number of
particles dn in the volume dV is instead given by dn = ρdV , where ρ is the density of
particles. Taking into account that each quantum state can be occupied by two particles
with opposite spins, we conclude that one must have 2d# = dn, and this implies that:
1 3
ρ(~r) = k (~r) , (6.2.22)
3π 2 F
or viceversa that:

1/3
kF (~r) = 3π 2 ρ1/3 (~r) . (6.2.23)

Let us now calculate the density t of kinetic energy in the neighborhood of the point ~r.
Observing that the density of free-particle quantum states with momentum between p and
p + dp and spin up or down is given by 2 d( 43 πp3 )/h3 = 8πp2 /h3 dp = π12 k2 dk, this is given
by:
Z kF (~r) 2 2 2 Z kF (~r)
~ k k dk ~2 4 ~2
t(~r) = = k dk = k5 (~r) . (6.2.24)
0 2m π 2 2π 2 m 0 10π 2 m F
Expressing this result in terms of the density of particles ρ(~r) and integrating it over the
whole space, one finally deduces that the kinetic energy is given by:
3~2
Z
2 2/3
ρ5/3 (~r) d3 ~r ,


Ekin = 3π (6.2.25)
10m
The potential energy due to the interaction with the external field V and the mutual
interaction W is instead simply given by:
1
Z ZZ
3
Epot = V (r) ρ(~r) d ~r + W (~r − ~r ′ ) ρ(~r) ρ(~r ′ ) d3~r d3~r ′ . (6.2.26)
2
Finally, the particle density ρ(~r) is subject to the constraint that its integral over the
whole space should be equal to the number of particles N :
Z
ρ(~r) d3~r = N . (6.2.27)

127
The total energy of the full system is given by E = Ekin + Epot and is a functional of
the density ρ(~r). The equilibrium configuration for the latter can then be determined
by minimizing this functional, compatibly with the constraint (6.2.27). This constrained
minimization problem can be handled by introducing one Lagrange parameter µ and
extremizing the following functional:

3~2
Z Z
2 2/3
ρ (~r) d ~r + V (r) ρ(~r) d3 ~r
5/3 3


E = 3π
10m
Z 
1
ZZ
+ W (~r − ~r ′ ) ρ(~r) ρ(~r ′ ) d3~r d3~r ′ − µ ρ(~r) d3~r − N . (6.2.28)
2

The stationarity of this functional E with respect to arbitrary variations of ρ(~r) implies
the following differential equation for ρ, for fixed numerical value of µ:

~2
Z
2 2/3 2/3
ρ (~r) + V (r) + W (~r − ~r ′ ) ρ(~r ′ ) d3~r ′ = µ .


3π (6.2.29)
2m

The stationarity with respect to µ implies instead the constraint (6.2.27), which determines
the number µ:
Z
ρ(~r) d3~r = N . (6.2.30)

Multiplying eq. (6.2.29) by ρ(~r)/N and integrating over ~r, one finds:

1 ~2
 Z Z
2 2/3
ρ (~r) d ~r + V (r) ρ(~r) d3~r
5/3 3


µ = 3π
N 2m
ZZ 
′ ′ 3 3 ′
+ W (~r − ~r ) ρ(~r) ρ(~r ) d ~r d ~r . (6.2.31)

Once the solution for the density profile ρ(~r) and the chemical potential µ has been found,
the total energy of the system is given by the functional E evaluated on this solution.
This can be written in the following way:

~2 1
Z ZZ

2/3
E = Nµ − 3π 2 ρ5/3 (~r) d3~r − W (~r − ~r ′ ) ρ(~r) ρ(~r ′ ) d3~r d3~r ′ .(6.2.32)
5m 2
This method is called the Thomas-Fermi method. It has the advantage of being simple
and of correctly implementing the fermionic statistics and Pauli’s exclusion principle. To
compare this method with the Hartree approach, we may identify the density of particles
ρ(~r) with the sum of all the probability densities defined by the individual wave functions
r )|2 . By comparing (6.2.28) with (6.1.6), we see that the kinetic
P
φi (~r): ρ(~r) = i |φi (~
energy is described in a different way, which exploits the semiclassical limit, but the
potential is essentially the same, at least in the statistical limit of many particles. This
comparison also shows that the Thomas-Fermi approach neglects the effects due to the
exchange energy.

128
6.2.2 The Thomas-Fermi-Dirac approximation

The exchange energy can be modeled also within this method of approximation. By using
a statistical approach that locally treats the system as a Fermi gas, very much as was done
above to derive an expression for the kinetic energy in terms of the particle density, one
can start from the exchange term in the energy as derived in the Hartree-Fock approach
and turn it into a functional of the particle density ρi (~r). More concretely, this can be
~
done by using free-particle plane-waves φi (~ri ) = (2π)−3/2 eiki ·~ri and any sum over i with
an integral over a wave vector ~k constrained to satisfy k ≤ kF . In this way one finds a
local expression of the general form:
Z
Eexc = fW [ρ(~r)] d3 ~r . (6.2.33)

The form of the functional fW [ρ] depends on the form of the mutual interaction W (~r −~r′ ).
In the simplest case of a Coulomb mutual interaction W (~r − ~r′ ) = α/|~r − ~r ′ |, one finds
for instance that fW [ρ(~r)] = − 34 (3/π)1/3 ρ4/3 (~r). For more general mutual interactions
W (~r − ~r′ ), one instead finds a more complicated fW [ρ(~r)]. Taking into account this
additional contribution describing the exchange energy, the total energy of the full system
is given by E = Ekin + Epot + Eexc and is again a functional of the density ρ(~r). The
functional to be minimized to determine the equilibrium configuration then becomes:

3~2
Z Z
2 2/3
ρ (~r) d ~r + V (r) ρ(~r) d3 ~r
5/3 3


E = 3π
10m
1
ZZ Z
+ W (~r − ~r ) ρ(~r) ρ(~r ) d ~r d ~r + fW [ρ(~r)] d3~r
′ ′ 3 3 ′
2
Z 
3
−µ ρ(~r) d ~r − N . (6.2.34)

The stationarity with respect to ρ(~r) now implies the following differential equation:

~2
Z

2/3 2/3
3π 2 ρ (~r) + V (r) + W (~r − ~r′ ) ρ(~r ′ ) d3~r ′ + fW
′
[ρ(~r)] = µ (6.2.35)
2m
The stationarity with respect to µ implies instead as before that:
Z
ρ(~r) d3~r = N . (6.2.36)

Multiplying eq. (6.2.35) by ρ(~r)/N and integrating over ~r, one finds:

1 ~2
 Z Z
2 2/3
ρ (~r) d ~r + V (r) ρ(~r) d3~r
5/3 3


µ = 3π
N 2m
ZZ Z 
′ ′ 3 3 ′ ′ 3
+ W (~r − ~r ) ρ(~r) ρ(~r ) d ~r d ~r + fW [ρ(~r)] ρ(~r) d ~r . (6.2.37)

Once the solution for the density profile ρ(~r) and the chemical potential µ has been found,
the total energy of the system is given by the functional E evaluated on this solution.

129
This can be written in the following way:

~2 1
Z ZZ
2 2/3 5/3 3
W (~r − ~r ′ ) ρ(~r) ρ(~r ′ ) d3~r d3~r ′


E = Nµ − 3π ρ (~r) d ~r −
5m 2
Z  
′
+ fW [ρ(~r)] − ρ(~r)fW [ρ(~r)] d3~r . (6.2.38)

This refined approach that includes also the exchange term is called the Thomas-Fermi-
Dirac approach.

6.2.3 Equilibrium particle density solution

The above approach, based on a particle density, allows to approximate the solutions of the
original single linear differential equation involving 3N different variables with the solution
of a single non-linear equation depending only on 3 variables. With this approximation,
an analytical determination of the solution is in general not possible, but a numerical
solution is usually totally straightforward to implement, after imputing an appropriate
boundary condition. This then determines the equilibrium particle density ρ. Once this is
known, one can compute the form of the effective potential felt by a probe particle placed
into the system, and compute the allowed stationary wave functions φi and the associated
energy levels Ei . These can finally be associated to the individual particles forming the
system.
The precision of this approach is intrinsically limited by the semiclassical and statistical
approximations that have been used. However, it can be proven rather easily that the
use of a single particle density to describe the system does not represent any limitation,
and one may in principle describe the exact solution in terms of an exact single-particle
density distribution rather than in terms of a multi-particle wave function. The difficulty
then reduces to find an as accurate energy functional as possible. This has led to the
so-called density functional theory.

6.3 Application to the structure of atoms

One of the most relevant applications of the above described approximation methods is
that of many-electron atoms. In that case, the external potential is given by the electric
field created by the nucleus, which is much heavier than the electrons and can therefore
by approximately considered to be at rest. The mutual interaction potential is instead
the electric repulsion between electrons. We therefore have in this case:

Ze2 e2
V (~r) = − , W (~r − ~r′ ) = . (6.3.39)
r |~r − ~r′ |

In principle, a many electron atom must be treated as a whole. The allowed configura-
tions are then determined by a multi-particle stationary state. The problem has an overall
spherical symmetry, and the total orbital angular momentum L ~ tot is therefore conserved.
Moreover, multi-particle states with definite symmetry properties under permutations of

130
the individual particles correspond to a definite value of the total spin S ~tot , which is also
conserved. The atomic energy levels can then be parametrized in terms of two quantum
numbers L and S, characterizing the sizes of the orbital and spin angular momenta, and
have a (2L + 1)(2S + 1)-fold degeneracy, corresponding to the arbitrary orientation of
these angular momenta. However, when relativistic corrections are included, only the
total angular momentum J~tot = L ~ tot + S
~tot remains conserved. The levels are then char-
acterized by a single quantum number J and have only a (2J + 1)-fold degeneracy. This
means that relativistic effects partially lift the original degeneracy of energy levels, and
a so-called fine-structure appears. More precisely, since J runs from |L − S| to L + S,
a state with given L and S, which was (2L + 1)(2S + 1) times degenerate, splits into
2 max(L, S) + 1 levels labelled by the allowed values of J, which are each 2J + 1 times
degenerate. The standard symbology to denote atomic energy levels is to write a capital
letter corresponding to the value of L = S, P, D, · · · , with a lower-right index denoting the
value of J = 0, 1/2, 1, · · · and an upper-left index denoting the value of 2S + 1 = 1, 2, · · · :
2S+1
LJ : atomic level . (6.3.40)

In some approximation, we may however describe a many electron atom as a system

composed of individual electrons in single-particle stationary states. As explained above,
the effect of the mutual interaction between electrons can be taken into account through
the fact that each of them feels a specific effective potential. As already discussed, all the
effective potentials felt by the individual electrons must be determined self-consistently,
by solving some nonlinear equations. But once this has been done, each of the electrons
is described by a one-particle stationary wave function, determined by an effective Hamil-
tonian. One can then talk about orbital configuration of each of the electrons separately.
To very good accuracy, each one-particle problem has an independent spherical symmetry,
and the allowed stationary states therefore have a definite individual angular momentum
~ Moreover, in each of these problems the spin does not matter and there is therefore a
L.
~ The individual electronic energy levels can then be parameter-
definite individual spin S.
ized in terms of the quantum numbers l and s = 1/2, and are (2l + 1)(2s + 1) = 2(2l + 1)
times degenerate. Again, relativistic effects partially lift this degeneracy, and only the
total angular momentum J~ = L ~ +S ~ is conserved. The resulting states are then describe
by a single quantum number j and have only a (2j + 1)-fold degeneracy. But we shall
ignore this here. Finally, we may label all the possible one-electron stationary states of
given l and s = 1/2 and higher and higher energies with a principal quantum number
n ≥ l + 1. More precisely, n − l − 1 corresponds to the number of nodes possessed by
the radial wave function describing this one-particle state. This is very similar to what
occurs in the Hydrogen atom, except that here the energy depends both on n and l. The
reason for this is that these states describe single-electrons moving in the effective po-
tential generated by all the other electrons, and this potential departs from the simple
1/r potential that was leading to this accidental degeneracy. The standard symbology to
denote individual electronic states, which are also called orbitals, is to write a lower-case
letter corresponding to the value of l = s, p, d, · · · , preceded by a number reporting the

131
value of n:

nl : electronic orbitals . (6.3.41)

One can then specify the electronic structure of a many-electron atom by specifying the
orbital of each electron. More and more complicated atoms with growing atomic number
Z will then correspond to filling the allowed single-electron orbitals in the energetically
most favorable way and compatibly with Pauli’s exclusion principle, which requires that
each electron should be in a different state. In this respect, we observe that owing to the
2(2l + 1) degeneracy of the single-electron energy levels of given n and l, there may be
as many as 2(2l + 1) electrons with the same n and l, and thus in first approximation
the same energy. For this reason, the group of 2(2l + 1) orbitals with definite n and l,
which have the same energy and are therefore said to be equivalent, is called a shell. The
electronic configuration of a many-electron atom may then be more simply characterized
by specifying the numbers of electrons that are present in each different shell. If a number
#nl of electrons occur in a shell with given n and l, this is written simply as nl#nl :

1s#1s 2s#2s 2p#2p · · · : electronic configuration . (6.3.42)

Notice finally that in order to full specify an atomic state, one needs to specify both the
atomic level and the electronic configuration. Indeed, for a given electronic configuration,
meaning given numbers of electrons in each nl shell, one may get different though usually
comparable atomic energy levels. This is due to the fact that the total energy depends on
the relative orientation of angular momenta of the electrons in each shell. The energetically
most favorable configuration can be determined according to Hund’s rule: it is the one that
has the largest possible value of S in absolute and the largest value of L compatible with
this value of S. A qualitative explanation for this statement can be given by considering
the resulting symmetry properties of the angular part of the multiparticle wave function
describing the full atom and arguing that the above rules lead to a maximization of
the average inter-electron distance and therefore a minimization of the repulsion energy
between different electrons.
In the above-described framework, the crucial dynamical question is now to understand
in which order the atomic shells are filled when the atomic number is increased. For excited
states of the Hydrogen atoms, the energy does not depend on l and increases when n is
increased. If one could fill theses levels without affecting their energies, one would thus
fill the shells in order of growing n, and independently of l. For actual many-electron
atoms, however, the filling of each orbital modifies the energies of the following ones, and
one has to study the full problem at once to determine the hierarchy of orbital energies.
One can then try to compare the spectrum of orbitals that is observed in spectroscopy
experiments with the one that can be computed using the Hartree-Fock or Thomas- Fermi
approximations. It turns out that one finds a fair agreement. The Hartree-Fock approach
is particular suited for small atomic numbers, while the Thomas-Fermi approach is more
suited for large atomic numbers. The structure of these spectra of orbitals allows moreover
to understand the structure of the periodic table of elements. In particular, it emerges

132
that different orbital states subdivide into distinct groups, which get progressively filled
when the number of electrons is increased, and correspond to the rows of the periodic
table:

1s 2 electrons
2s, 2p 8 electrons
3s, 3p 8 electrons
4s, 3d, 4p 18 electrons (6.3.43)
5s, 4d, 5p 18 electrons
6s, 4f, 5d, 6p 32 electrons
7s, 6d, 5f ... electrons

This observed filling order is in good agreement with the hierarchy of energy levels that
can be computed at the theoretical level. As expected, these energies depend not only on
the value of n but also on the value of l. A first characteristic of the spectrum is that for
a given n the energy increases monotonically when l increases. A qualitative explanation
for this is that for a given n the states with smaller l are on average closer to the nucleus
and deeper inside the cloud formed by the other electrons, and can therefore feel in a
stronger and less shielded way the attraction from the nucleus. An other characteristic
that emerges is that in some cases there can be a competition between states with different
values of n, and it may be energetically more favorable to increase n rather than l.

133
Bibliography

[1] H. Goldstein,
Classical mechanics,
Addison Wesley (1950).

[2] E. Merzbacher,
Quantum mechanics,
John Wiley & Sons (1970).

[3] J. J. Sakurai and J. Napolitano,

Modern quantum mechanics,
Addison Wesley (1994).

[4] L. D. Landau and E. M. Lifshitz,

Quantum mechanics: non-relativistic theory,
Pergamon (1977).

[5] A. Messiah,
Quantum mechanics,
North-Holland (1975).

[6] B. R. Holstein,
Topics in advanced quantum mechanics,
Addison Wesley (1992).

[7] C. Rossetti,
Introduzione alla meccanica quantistica,
Levrotto & Bella (1990).

[8] G. Baym,
Lectures on quantum mechanics,
Benjamin (1973).

[9] H. A. Bethe and R. Jackiw,

Intermediate quantum meccanics,
Benjamin (1986)