A. Simple Aspects of the Structure of Quantum Mechanicscds.cern.ch/record/1481599/files/978-1-4614-3951-6_Book... · 2014. 7. 18. · equivalent representations of quantum mechanics.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A. Simple Aspects of the Structureof Quantum Mechanics
In Chapters 2 to 16 we have used the formulation of quantum mechanics interms of wave functions and differential operators. This is but one of manyequivalent representations of quantum mechanics. In this appendix we shallbriefly review that representation and develop an alternative representation inwhich state vectors correspond to the wave functions and matrices to the op-erators. To keep things simple we shall restrict ourselves to systems with dis-crete energy spectra exemplified on the one-dimensional harmonic oscillator.
A.1 Wave Mechanics
In Section 6.3 the stationary Schrödinger equation(− h̄
2m
d2
dx2+ m
2ω2x2
)ϕn = Enϕn(x)
of the harmonic oscillator has been solved. The eigenvalues En were found tobe
En = (n + 12 )h̄ω
together with the corresponding eigenfunctions
ϕn(x) = 1
(√πσ02nn!)1/2
Hn
(x
σ0
)exp
{− x2
2σ 20
}, σ0 =
√h̄
mω.
Quite generally, we can write the stationary Schrödinger equation as an eigen-value equation
Hϕn = Enϕn ,
where the Hamiltonian H – as in classical mechanics – is the sum
414 A. Simple Aspects of the Structure of Quantum Mechanics
T = p̂2
2mand the potential energy
V = m
2ω2x2 .
The difference to classical mechanics consists in the momentum being givenin one-dimensional quantum mechanics by the differential operator
p̂ = h̄
i
d
dx
so that the kinetic energy takes the form
T = − h̄2
2m
d2
dx2.
Two eigenfunctions ϕn(x), ϕm(x), m = n belonging to different eigenvaluesEm = En are orthogonal, i.e.,1∫ +∞
−∞ϕ∗
m(x)ϕn(x)dx = 0 .
Conventionally, for m = n the eigenfunctions are normalized to one, i.e.,∫ +∞
−∞ϕ∗
n (x)ϕn(x)dx = 1 ,
so that we may summarize∫ +∞
−∞ϕ∗
m(x)ϕn(x)dx = δmn ,
where we have used the Kronecker symbol
δmn ={
1 , m = n0 , m = n
.
The infinite set of mutually orthogonal and normalized eigenfunctionsϕn(x), n = 0, 1, 2, . . . , forms a complete orthonormal basis of all complex-valued functions f (x) which are square integrable, i.e.,∫ +∞
−∞f ∗(x) f (x)dx = N 2 , N <∞ .
N is called the norm of the function f (x). Functions with norm N can benormalized to one,
1The functions ϕn(x) are real functions. We add an asterisk (indicating the complex con-jugate) to the function ϕm(x) in the integral, since in other cases one often has to deal withcomplex functions.
A.2 Matrix Mechanics in an Infinite Vector Space 415∫ +∞
−∞ϕ∗(x)ϕ(x)dx = 1 ,
by dividing them by the normalization factor N ,
ϕ(x) = 1
Nf (x) .
The completeness of the set ϕn(x), n = 0, 1, 2, . . . , allows the expansion
f (x) =∞∑
n=0
fnϕn(x) .
Because of the orthonormality of the eigenfunctions the complex coefficientsfn are simply
fn =∫ +∞
−∞ϕ∗
n (x) f (x)dx .
We also get
N 2 =∫ +∞
−∞f ∗(x) f (x)dx =
∞∑n=0
| fn|2 .
The superposition of two normalizable functions
f (x) =∞∑
n=0
fnϕn(x) , g(x) =∞∑
n=0
gnϕn(x)
with complex coefficients α, β may be expressed by
α f (x)+βg(x) =∞∑
n=0
(α fn +βgn)ϕn(x) .
Their scalar product is defined as∫ +∞
−∞g∗(x) f (x)dx =
∞∑n=0
g∗n fn .
A.2 Matrix Mechanics in an Infinite Vector Space
The normalizable functions f (x) form a linear vector space of infinite di-mensionality, i.e., each function f (x) can be represented by a vector f in thatspace,
f (x) → f .
416 A. Simple Aspects of the Structure of Quantum Mechanics
With the base vectors
ϕ0 =
⎛⎜⎜⎜⎝100...
⎞⎟⎟⎟⎠ , ϕ1 =
⎛⎜⎜⎜⎝010...
⎞⎟⎟⎟⎠ , ϕ2 =
⎛⎜⎜⎜⎝001...
⎞⎟⎟⎟⎠ , . . .
a general vector f takes the form
f =∞∑
n=0
fnϕn =
⎛⎜⎜⎜⎝f0
f1
f2...
⎞⎟⎟⎟⎠ .
The axioms of the infinite space of complex column vectors are the naturalextension of the ones for finite complex vectors:
(i) Linear superposition (α, β complex numbers):
αf+βg =
⎛⎜⎜⎜⎝α f0 +βg0
α f1 +βg1
α f2 +βg2...
⎞⎟⎟⎟⎠ .
(ii) Scalar product:
g+ · f = (g∗0 , g∗
1 , g∗2 , . . .)
⎛⎜⎜⎜⎝f0
f1
f2...
⎞⎟⎟⎟⎠ =∞∑
n=0
g∗n fn .
Here the adjoint g+ of the vector g has been introduced, g+ = (g∗0 , g∗
1 , g∗2 , . . .),
as the line vector of the complex conjugates g∗0 , g∗
1 , g∗2 , . . . of the components
of the column vector
g =
⎛⎜⎜⎜⎝g0
g1
g2...
⎞⎟⎟⎟⎠ .
Because of the infinity of the set of natural numbers an additional axiom hasto be added:
A.2 Matrix Mechanics in an Infinite Vector Space 417
(iii) The norm |f| of the vectors f is finite,
|f|2 = f+ · f =∞∑
n=0
f ∗n fn = N 2 , N <∞ ,
i.e., the infinite sum has to converge. Because of Schwartz’s inequality
|g+ · f| ≤ |g||f|all scalar products of vectors f, g of the space are finite.
As in the ordinary finite-dimensional vector spaces we call a linear trans-formation A of a function f (x) into a function g(x),
g(x) = A f (x) ,
the linear operator A. Examples of linear transformations are
• the momentum operator p̂ = −ih̄ d/dx ,
p̂ f = h̄
i
d f
dx(x) ,
• the Hamiltonian H = −(h̄2/2m)d2/dx2 + V (x),
H f = − h̄2
2m
d2 f (x)
dx2+ V (x) f (x) ,
• the position operator x̂ = x ,
x̂ f = x f (x) .
Linear operators can be represented by matrices. We show this by thefollowing argument. The function g is represented by the coefficients gm ,
g(x) =∞∑
m=0
gmϕm(x) , gm =∫ +∞
−∞ϕ∗
m(x)g(x)dx .
The image function g of f is given by
g(x) = A f (x) = A
( ∞∑n=0
fnϕn(x)
)=
∞∑n=0
Aϕn(x) fn ,
i.e., by a linear combination of the images Aϕn of the elements ϕn of theorthonormal basis. The Aϕn themselves can be represented by a linear com-bination of the basis vectors
418 A. Simple Aspects of the Structure of Quantum Mechanics
Aϕn =∞∑
m=0
ϕm(x)Amn , n = 0, 1, 2, . . . ,
with the coefficients
Amn =∫ +∞
−∞ϕ∗
m(x)Aϕn(x)dx .
Inserting this into the expression for g(x) we obtain
g(x) =∞∑
m=0
∞∑n=0
ϕm(x)Amn fn .
Comparing with the representation for g(x) we find for the coefficients gm theexpression
gm =∞∑
n=0
Amn fn .
We arrange the coefficients Amn like matrix elements in an infinite matrixscheme
It should be noted that the two descriptions by wave functions and opera-tors or by vectors and matrices are equivalent. The correspondence relations
ϕ(x) =∞∑
n=0
anϕn(x) ↔ ϕ =
⎛⎜⎜⎜⎝a0
a1
a2...
⎞⎟⎟⎟⎠with
an =∫ +∞
−∞ϕ∗
n (x)ϕ(x)dx
A.3 Matrix Representation of the Harmonic Oscillator 419
for wave functions and vectors work in both directions. For a given wave func-tion ϕ(x) we can uniquely determine the vector ϕ relative to the basis ϕn(x),n = 0, 1, 2, . . . . Conversely, for a given vector ϕ relative to the basis ϕn(x)we can reconstruct the wave function ϕ(x) as the above superposition of theϕn(x). The descriptions in terms of ϕ(x) and ϕ contain the same informationabout the state the system is in.
Thus, generally one does not distinguish the two descriptions and says thesystem is in the state ϕ, often denoted by the ket |ϕ〉 as introduced by Paul A.M. Dirac.
The wave function ϕ(x) or the vector ϕ are considered merely as tworepresentations out of which many can be invented. The same statementshold true for the representation of operators in terms of differential operatorsor matrices. Also these are only representations of one and the same lineartransformation called linear operator. The states ϕ like the wave functions orvectors ϕ form a linear vector space with scalar product. This general spaceis called Hilbert space. The linear operators transform a state of the Hilbertspace into another state.
A.3 Matrix Representation of the Harmonic Oscillator
Since the ϕn(x) are normalized eigenfunctions of the Hamiltonian, we find forits matrix elements
Hmn =∫ +∞
−∞ϕ∗
m(x)Hϕn(x)dx
= (n + 12 )h̄ω
∫ +∞
−∞ϕ∗
m(x)ϕn(x)dx = (n + 12)h̄ωδmn .
Thus, the matrix representation of the Hamiltonian of the harmonic oscillatorin its eigenfunction basis is diagonal:
H = h̄ω
⎛⎜⎜⎜⎝12 0 0 . . .
0 32 0 . . .
0 0 52 . . .
......
.... . .
⎞⎟⎟⎟⎠ .
The representation of the eigenfunctions ϕn(x) in their own basis are given bythe standard columns
ϕ0 =
⎛⎜⎜⎜⎝100...
⎞⎟⎟⎟⎠ , ϕ1 =
⎛⎜⎜⎜⎝010...
⎞⎟⎟⎟⎠ , ϕ2 =
⎛⎜⎜⎜⎝001...
⎞⎟⎟⎟⎠ , . . . .
420 A. Simple Aspects of the Structure of Quantum Mechanics
Of course, the eigenvector equation is recovered also in matrix representation,
Hϕn = (n + 12 )h̄ωϕn .
With the help of the recurrence relations for Hermite polynomials,
dHn(x)
dx= 2nHn−1(x) ,
Hn+1(x) = 2x Hn(x)−2nHn−1(x) ,
we find the matrix representations for the position operator x̂ and the momen-tum operator p̂ in harmonic-oscillator representation,
x̂ϕn = xϕn(x) = σ0
(√πσ02nn!)1/2
(nHn−1(x)+ 1
2 Hn+1(x))
exp
{− x2
2σ 20
}= σ0√
2
(√n ϕn−1(x)+√
n +1ϕn+1(x))
.
The coefficients xmn are given by
xmn =∫ +∞
−∞ϕ∗
m(x)xϕn(x)dx = σ0√2
(√n δm(n−1) +
√n +1 δm(n+1)
),
and the matrix representation of the position operator is
x = σ0√2
⎛⎜⎜⎜⎜⎜⎝0 1 0 0 . . .
1 0√
2 0 . . .
0√
2 0√
3 . . .
0 0√
3 0 . . ....
......
.... . .
⎞⎟⎟⎟⎟⎟⎠ .
For the momentum operator we obtain in a similar way the matrix represen-tation
p = h̄√2σ0
⎛⎜⎜⎜⎜⎜⎝0 i 0 0 . . .
−i 0 i√
2 0 . . .
0 −i√
2 0 i√
3 . . .
0 0 −i√
3 0 . . ....
......
.... . .
⎞⎟⎟⎟⎟⎟⎠ .
One easily verifies the commutation relation
[p, x] = px − xp = h̄
i
also for the infinite matrix representations of x and p̂. Both matrices for x andp̂ are Hermitean, i.e.,
x∗nm = xmn , p∗
nm = pmn .
A.4 Time-Dependent Schrödinger Equation 421
A.4 Time-Dependent Schrödinger Equation
The time dependence of wave functions is determined by the time-dependentSchrödinger equation
ih̄∂
∂tψ(x , t) = Hψ(x , t) .
The eigenstates ϕn(x) of the Hamiltonian are the space-dependent factors inan ansatz
ψn(x , t) = exp
{− i
h̄Ent
}ϕn(x) ,
and the eigenvalues En determine the time dependence of the phase factor.In the matrix representation of the harmonic oscillator the time-dependent
Schrödinger equation simply reads
ih̄d
dtψ(t) = Hψ(t) ,
where ψ(t) is a vector in Hilbert space,
ψ(t) =
⎛⎜⎜⎜⎝ψ0(t)ψ1(t)ψ2(t)
...
⎞⎟⎟⎟⎠ .
Because of the linearity of the Schrödinger equation any linear combina-tion
ψ(x , t) =∞∑
n=0
anψn(x , t) =∞∑
n=0
an exp
{− i
h̄Ent
}ϕn(x)
also solves the Schrödinger equation. In vectorial representation we have
ψ(t) =∞∑
n=0
an exp
{− i
h̄Ent
}ϕn , En = (n + 1
2)h̄ω .
The initial condition at t = 0 for the time-dependent Schrödinger equation isthe initial wave function
ψ(x ,0) = ψi(x) =∞∑
n=0
ψinϕn(x) .
In vector notation this is an initial state vector ψ i. Its decomposition intoeigenvectors ϕn,
422 A. Simple Aspects of the Structure of Quantum Mechanics⎛⎜⎜⎜⎝ψi0
ψi1
ψi2...
⎞⎟⎟⎟⎠ = ψ i =∞∑
n=0
anϕn =
⎛⎜⎜⎜⎝a0
a1
a2...
⎞⎟⎟⎟⎠ ,
directly provides the identification of the expansion coefficients an with thecomponents ψin of the initial vector ψ i,
an = ψin .
This way the time-dependent Schrödinger equation is solved by the ex-pression
ψ(t) =∞∑
n=0
ψin exp
{− i
h̄Ent
}ϕn
for the initial conditionψ(0) = ψ i .
The time-dependent vector ψn(t) corresponding to ψn(x , t) is
ψn(t) = exp
{− i
h̄Ent
}ϕn = exp
{− i
h̄Ht
}ϕn ,
where En is the energy eigenvalue corresponding to the eigenvector ϕn, i.e.,En = (n + 1
2 )h̄ω for the harmonic oscillator. The last equality is meaningful ifwe define the exponential of a matrix by its Taylor series
exp
{− i
h̄Ht
}=
∞∑n=0
1
n!
(− i
h̄Ht
)n
.
For the case of the diagonal matrix H the nth power is trivial,
H n =
⎛⎜⎜⎜⎝En
0 0 0 . . .
0 En1 0 . . .
0 0 En2 . . .
......
.... . .
⎞⎟⎟⎟⎠ ,
and the explicit matrix form is
exp
{− i
h̄Ht
}=
⎛⎜⎜⎜⎜⎜⎜⎝exp
{− i
h̄ E0t}
0 0 . . .
0 exp{− i
h̄ E1t}
0 . . .
0 0 exp{− i
h̄ E2t}. . .
......
.... . .
⎞⎟⎟⎟⎟⎟⎟⎠ .
A.5 Probability Interpretation 423
Using the operator representation of
ψn(x , t) = exp
{− i
h̄Ent
}ϕn = exp
{− i
h̄Ht
}ϕn
as derived above we may rewrite ψ(t) into the form
ψ(t) =∞∑
n=0
ψin exp
{− i
h̄En
}ϕn
= exp
{− i
h̄Ht
} ∞∑n=0
ψinϕn
= exp
{− i
h̄Ht
}ψ(0)
= UH (t)ψ(0) .
The operator
UH (t) = exp
{− i
h̄Ht
}is called the temporal-evolution operator.
A.5 Probability Interpretation
The eigenfunctions ϕn(x), equivalently the eigenvectors ϕn, describe a stateof the physical system with the energy eigenvalue En. Thus, a precise mea-surement of the energy of this system in the state ϕn should be devised toproduce as a result the value En. In order to preserve the reproducibility ofthe measurement it should not change the eigenstate ϕn of the system duringthe measurement, i.e., immediately after the energy measurement the state ofthe system should still be ϕn.
The question arises what result will be found in the same energy measure-ment carried out at a system in a state ϕ described by the wave function ϕ(x),or equivalently, by the vector ϕ being a superposition of eigenfunctions ϕn(x)or eigenvectors ϕn,
ϕ =∞∑
n=0
anϕn ,
with norm one, i.e.,∞∑
n=0
|an|2 = 1 .
424 A. Simple Aspects of the Structure of Quantum Mechanics
The single measurement of the energy will result in one of the energy eigen-values which we call Em . Reproducibility of the measurement then requiresthat the system is in the state ϕm after the measurement.
The absolute square |am|2 of the coefficients am in the superposition of theϕm defining ϕ is the probability with which the energy eigenvalue Em will bedetermined in the single measurement.
Let us assume that we prepare a large assembly of N identical systems,all in the same state ϕ. If we carry out single measurements on these vari-ous identical systems we shall measure the energy eigenvalue Em with theabundance |am|2 N .
Performing a weighted average over the results of all measurements yieldsthe expectation value of the energy
〈E〉 = 1
N
∞∑n=0
|an|2 N En =∞∑
n=0
|an|2 En .
Using the state-vector representation of ϕ,
ϕ =
⎛⎜⎜⎜⎝a0
a1
a2...
⎞⎟⎟⎟⎠ ,
we find that the energy expectation value is simply
ϕ+Hϕ = (a∗0 ,a∗
1 ,a∗2 , . . .)
⎛⎜⎜⎜⎝E0 0 0 . . .
0 E1 0 . . .
0 0 E2 . . ....
......
. . .
⎞⎟⎟⎟⎠⎛⎜⎜⎜⎝
a0
a1
a2...
⎞⎟⎟⎟⎠=
∞∑n=0
a∗n Enan = 〈E〉 .
Equivalently, in wave-function formulation, we have∫ +∞
−∞ϕ∗(x)Hϕ(x)dx =
∫ +∞
−∞ϕ∗(x)
∞∑n=0
Enanϕn(x)
=∞∑
n=0
Enan
∫ +∞
−∞ϕ∗(x)ϕn(x)dx
=∞∑
n=0
Enana∗n = 〈E〉 .
B. Two-Level System
In Appendix A the equivalence of wave-function and matrix representation ofquantum mechanics was shown. The simplest matrix structure is the one intwo dimensions, i.e., in a space with two base states:
η1 =(
1
0
), η−1 =
(0
1
).
The linear space consists of all linear combinations
χ = χ1η1 +χ−1η−1 =(χ1
χ−1
)of the base states with complex coefficients χ1 and χ−1. The two states η1 andη−1 form an orthonormal basis of this space, i.e.,
η+1 ·η1 = 1 , η+
−1 ·η−1 = 1 , η+1 ·η−1 = η+
−1 ·η1 = 0 .
For the linear combination χ to be normalized to one we have
χ+ ·χ = χ∗1χ1 +χ∗
−1χ−1 = |χ1|2 +|χ−1|2 = 1 .
This suggests a representation of the absolute values |χr |, r = 1,−1, of thecomplex coefficients by trigonometric functions:
|χ1| = cosΘ
2, |χ−1| = sin
Θ
2.
The use of the half-angle Θ/2 is a convention, the usefulness of which willbecome obvious in the sequel. The complex coefficients themselves are ob-tained by multiplication of the moduli |χr | with arbitrary phase factors:
χ1 = e−iΦ1/2 cosΘ
2, χ−1 = e−iΦ−1/2 sin
Θ
2.
Since a common phase factor is irrelevant the general form can be restrictedto
The operators corresponding to physical quantities are Hermitean matri-ces
A =(
A1,1 A1,−1
A−1,1 A−1,−1
).
The Hermitean conjugate of A is defined as
A+ =(
A∗1,1 A∗
−1,1
A∗1,−1 A∗
−1,−1
), i.e., A+
rs = A∗sr .
The condition of Hermiticity,
A+ = A , i.e., A∗sr = Ars ,
requiresA∗
1,1 = A1,1 , A∗1,−1 = A−1,1 ,
A∗−1,1 = A1,−1 , A∗
−1,−1 = A−1,−1 .
Thus, the diagonal elements A1,1, A−1,−1 are real quantities, the off-diagonalelements A1,−1, A−1,1 are complex conjugates of each other. Hermiticity of theoperator A ensures that the expectation value of A for a given general state isreal,
χ+ Aχ =∑
i , j=1,−1
χ∗i Ai jχj
= χ∗1 A1,1χ1 +χ∗
1 A1,−1χ−1 +χ∗−1 A∗
1,−1χ1 +χ∗−1 A−1,−1χ−1 .
All Hermitean matrices can be linearly combined as the superpositions
A = a0σ0 +a1σ1 +a2σ2 +a3σ3
(with real coefficients a0, . . ., a3) of the unit matrix
σ0 =(
1 0
0 1
)and the Pauli matrices
B. Two-Level System 427
σ1 =(
0 11 0
), σ2 =
(0 −ii 0
), σ3 =
(1 00 −1
),
since the four matrices σ0, . . . , σ3 are Hermitean. One directly verifies the re-lations
σ 2i = σ0 , i = 0, 1, 2, 3 ,
andσ1σ2 = iσ3 , σ2σ3 = iσ1 , σ3σ1 = iσ2 .
These yield the commutation relations
[σ1,σ2] = σ1σ2 −σ2σ1 = 2iσ3
and cyclic permutations.The three Pauli matrices can be grouped into a vector in three dimensions,
σ = (σ1,σ2,σ3) ,
with the squareσ 2 = σ 2
1 +σ 22 +σ 2
3 = 3σ 20 .
The base states η1, η−1 are eigenstates of the Pauli matrix σ3 and of the sumof their squares σ 2,
σ3ηr = rηr , σ 2ηr = 3σ0ηr = 3ηr , r = 1,−1 ,
since σ3 and σ0 are diagonal matrices.According to Section A the time-dependent Schrödinger equation reads
ih̄d
dtξ (t) = Hξ (t) ,
where the Hamiltonian is a Hermitean 2×2 matrix,
H =(
H1,1 H1,−1
H−1,1 H−1,−1
),
with real diagonal matrix elements H1,1, H−1,−1 and with off-diagonal ele-ments H−1,1 = H ∗
1,−1. It can be represented by a superposition of the σ matri-ces,
H = h0σ0 +h3σ3 +h1σ1 +h2σ2 ,
where
h0 = 1
2(H1,1 + H−1,−1) , h3 = 1
2(H1,1 − H−1,−1) ,
and
428 B. Two-Level System
h1 = 1
2(H1,−1 + H−1,1) = Re H1,−1 ,
h2 = i
2(H1,−1 − H−1,1) = −Im H1,−1 .
Introducing the hi (i = 0, 1, 2, 3) into the matrix H we obtain
H =(
h0 +h3 h1 − ih2
h1 + ih2 h0 −h3
).
Introducing the factorization
ξ r (t) = exp
{− i
h̄Er t
}χ r , r = 1,−1 ,
into the time-dependent phase factor and the stationary state χ r we obtain thestationary Schrödinger equation
Hχ r = Erχ r , r = 1,−1 ,
for the eigenstate χ r belonging to the energy eigenvalue Er . For the eigenval-ues we find
E±1 = h0 ±|h| , |h| =√
h21 +h2
2 +h23 .
Since there are only two eigenvalues our system is called a two-level system.The eigenstates are
χ1 = 1√2|h|
( √|h|+h3 e−iΦ/2√|h|−h3 eiΦ/2
),
χ−1 = 1√2|h|
( −√|h|−h3 e−iΦ/2√|h|+h3 eiΦ/2
),
with the phase factor determined by
ei2Φ = h1 + ih2
h1 − ih2.
Introducing the angle Θ by
cosΘ
2=
√|h|+h3
2|h| , sinΘ
2=
√|h|−h3
2|h|we may write the eigenstates in the form
B. Two-Level System 429
χ1 = e−iΦ/2 cosΘ
2η1 + eiΦ/2 sin
Θ
2η−1 ,
χ−1 = −e−iΦ/2 sinΘ
2η1 + eiΦ/2 cos
Θ
2η−1 .
They are normalized and orthogonal to each other.The eigenstates χ1, χ−1 of the two-level system exhibit a time dependence
which is given by a phase factor only,
ξ r (t) = exp
{− i
h̄Er t
}χ r , r = 1,−1 .
If initially the system is not in an eigenstate the state oscillates. We assumethat the initial state is
ϕ(0) = η−1 .
Decomposition into the eigenstates yields
η−1 = ζ1χ 1 + ζ−1χ−1
with
ζ1 = χ+1 ·η−1 = e−iΦ/2 sin
Θ
2,
ζ−1 = χ+−1 ·η−1 = e−iΦ/2 cos
Θ
2.
The time-dependent state is obtained as
ϕ(t) = ζ1ξ 1(t)+ ζ−1ξ−1(t)
= e−iΦ/2 sinΘ
2e−iω1tχ1 + e−iΦ/2 cos
Θ
2e−iω−1tχ−1
with the angular frequencies
ωr = Er/h̄ , r = 1,−1 .
The probability to find the system (originally in the state η−1) in the state η1
is given by
P1,−1 = sin2Θ sin2 |h|h̄
t ,
and, of course, the probability to find it in the state η−1 is
We consider a detector which is capable of measuring position and momen-tum of a particle simultaneously with certain accuracies. If the result of ameasurement is the pair xD, pD of values we may assume that the true val-ues x , p of the quantities to be measured are described by the uncorrelatedbivariate Gaussian (cf. Section 3.5) probability density
ρD(x , p, xD, pD) = 1
2πσxDσpDexp
{−1
2
[(x − xD)2
σ 2xD
+ (p − pD)2
σ 2pD
]}.
That is to say, the probability for the true values x , p of the particle to be theintervals between x and x +dx and between p and p +dp is
dP = ρD(x , p)dx dp .
The particle to be measured by the detector possesses position and mo-mentum values x and p. The particle may have been produced by a sourcewhich does not define exactly the values of x and p but according to a proba-bility density
ρS(x , p, xS, pS) = 1
2πσxSσpSexp
{−1
2
[(x − xS)2
σ 2xS
+ (p − pS)2
σ 2pS
]}.
This is an uncorrelated bivariate Gaussian probability density with the expec-tation values xS and pS and the variances σ 2
xS and σ 2pS.
We now describe how much information can at best be obtained about theprobability density ρS(x , p) using the above detector. The probability for aparticle prepared by the source to be detected within the intervals (xD, xD +dxD) and (pD, pD +dpD) is given by
p are obtained by summing up the variances ofthe detector and source distribution,
σ 2x = σ 2
xD +σ 2xS , σ 2
p = σ 2pD +σ 2
pS .
The quantity wcl(xD, pD, xS, pS) is the result of analyzing the phase-spaceprobability density of the source ρS(x , p, xS, pS) with the help of the phase-space probability density ρD(x , p, xD, pD). The function wcl(xD, pD, xS, pS) isitself a phase-space probability density and obtained through a process weshall call phase-space analysis.
This distribution can be measured in principle if the source consecutivelyproduces a large number of particles which are observed in the detector. Fora detector of arbitrarily high precision,
σxD → 0 , σpD → 0 ,
the distribution wcl approaches the source distribution ρS(x = xD, p = pD, xS,pS).
We have seen that with a detector of high precision and with a sufficientlyhigh number of measurements the source distribution can be measured witharbitrary high accuracy. We now assume that the minimum-uncertainty rela-tions,
σxDσpD = h̄
2, σxSσpS = h̄
2,
hold for the widths characterizing the detector and the source. Besides thisrestriction we stay within the framework of classical physics. Now it is nolonger possible to measure the source distribution exactly.1 However, we maystill measure the distribution in position alone or the distribution in momen-tum alone with arbitrary accuracy. To show this we construct the marginaldistributions of wcl in the variables xD − xS, and pD − pS, respectively,
wclx (xD, xS) = 1√
2πσx
exp
{−1
2
(xD − xS)2
σ 2x
}and
1We could, however, compute the source distribution ρS by unfolding it from wcl.
432 C. Analyzing Amplitude
Fig.C.1. Phase-space distributions ρS (top), ρD (middle), and their product ρSρD (bottom)together with the marginal distributions of ρS and ρD. The two columns differ only inthe spatial mean xS of ρS. Units are used in which h̄ = 1.
wclp (pD, pS) = 1√
2πσp
exp
{−1
2
(pD − pS)2
σ 2p
}.
The first distribution approaches the corresponding marginal position distri-bution of the source,
ρSx (x , xS) =∫ +∞
−∞ρS(x , p, xS, pS)dp = 1√
2πσxS
exp
{−1
2
(x − xS)2
σ 2xS
}in the case of σxD → 0. However, because of the minimum uncertainty rela-tions, σpD as well as σp approach infinity. Therefore, the second distribution
Fig.C.2. The phase-space distribution ρS (top), the distribution ρD for a particular point(xD, pD) of mean values (middle), and convolution of ρS with ρD for all possible meanvalues (bottom). Also shown are the marginal distributions. The two columns differ inthe widths of ρS. Units h̄ = 1 are used.
wclp becomes so wide – and actually approaches zero – that no information
about the momentum distribution can be obtained from it.Conversely, for σpD → 0 the momentum distribution of the source can be
measured accurately. However, then the information about the position distri-bution is lost.
We illustrate the concepts of this section in Figures C.1 and C.2. We beginwith the discussion of the result of a single measurement, yielding the pair xD,pD of measured values. In the two columns of Figure C.1 we show (from topto bottom) the probability density ρS(x , p) characterizing the particle as pro-
434 C. Analyzing Amplitude
duced by the source, the density ρD(x , p) characterizing the detector for thecase xD = 0, pD = 0, and the product function ρS(x , p)ρD(x , p). The integralover the product function is the probability density wcl. It is essentially dif-ferent from zero only if there is a region, the overlap region, in which both ρS
and ρD are different from zero. In the left-hand column of Figure C.1 ρS andρD were chosen to be identical, so that wcl is large. In the right-hand columnthe overlap is smaller.
By very many repeated measurements, each yielding a different resultxD, pD we obtain the probability density wcl(xD, pD). In the two columns ofFigure C.2 we show (from top to bottom) the probability density ρS(x , p)characterizing the particle, the density ρD(x , p) characterizing the detectorfor the particular set measured values xD = 0, pD = 0, and the probabilitydensity wcl(xD, pD) for measuring the pair of values xD, pD. Also shown arethe marginal distribution ρSx (x), ρDx (x), and wcl
x (xD) in position, and ρSp(p),ρDp(p), and wcl
p (pD) in momentum. Comparing in the left-hand column thediagram of ρS with the diagram of wcl we see that the latter distribution is ap-preciably broader than the former in both variables. In the right-hand column,however, the spatial width of the detector distribution σxD is very small at theexpense of the momentum width σpD = h̄/(2σxD), which is very large. Thedistribution wcl is practically identical to ρS what concerns its spatial varia-tion. The width in momentum of wcl is, however, very much larger than thatof ρS.
C.2 Analyzing Amplitude: Free Particle
Quantum-mechanically we describe a particle by the minimum-uncertaintywave packet
ϕS(x) = ϕS(x , xS, pS) = 1
(2π )1/4σ1/2xS
exp
{− (x − xS)2
4σ 2xS
+ i
h̄pS(x − xS)
}.
We consider this wave packet as having been prepared by some physical ap-paratus, the source. The question now arises how the phase-space analysisof the particle as discussed in the last section can be described in quantummechanics.
If, in a particle detector with position-measurement uncertainty σxD andmomentum-measurement uncertainty σpD = h̄/(2σxD), the values xD, pD aremeasured, we want to interpret the result as in Section C.1. The same proba-bility density
ρD(x , p, xD, pD) = 1
2πσxDσpDexp
{−1
2
[(x − xD)2
σ 2xD
+ (p − pD)2
σ 2pD
]}
C.2 Analyzing Amplitude: Free Particle 435
describes the probability density of position x and momentum p of the parti-cle. Quantum-mechanically this probability density is the phase-space distri-bution introduced by Eugene P. Wigner in 1932 of the wave packet
ϕD(x , xD, pD) = 1
(2π )1/4σ1/2xD
exp
{− (x − xD)2
4σ 2xD
+ i
h̄pD(x − xD)
}.
Therefore, ρD(x , p, xD, pD) is also called Wigner distribution of ϕD (cf. Ap-pendix D).
Let us construct the analyzing amplitude
a(xD, pD, xS, pS) = 1√h
∫ +∞
−∞ϕ∗
D(x , xD, pD)ϕS(x , xS, pS)dx
representing the overlap between the particle’s wave function ϕS and the de-tecting wave function ϕD. It turns out to be
a(xD, pD, xS, pS) = 1√2πσxσp
exp
{− (xD − xS)2
4σ 2x
− (pD − pS)2
4σ 2p
− i
h̄
σ 2xD pD +σ 2
xS pS
σ 2x
(xD − xS)
},
where, like in Section C.1,
σ 2x = σ 2
xD +σ 2xS , σ 2
p = σ 2pD +σ 2
pS .
The absolute square of the analyzing amplitude
|a|2 = 1
2πσxσpexp
{−1
2
[(xD − xS)2
σ 2x
+ (pD − pS)2
σ 2p
]}= wcl(xD, pD, xS, pS)
is identical to the probability density wcl(xD, pD, xS, pS) of Section C.1.We conclude that the probability amplitude analyzing the values xD and
pD of position and momentum of a particle as a result of the interaction of aparticle with a detector is given by
a(xD, pD, xS, pS) = 1√h
∫ +∞
−∞ϕ∗
D(x , xD, pD)ϕS(x , xS, pS)dx .
Here, ϕS is the wave function of the particle and ϕD the analyzing wave func-tion. The probability to observe a position in the interval between pD andpD +dpD is
dP = |a(xD, pD, xS, pS)|2 dxD dpD .
436 C. Analyzing Amplitude
In analogy to the classical case we may now ask whether we can stillrecover the original quantum-mechanical spatial probability density
ρS(x) = |ϕS(x)|2 = 1√2πσxS
exp
{−1
2
(x − xS)2
σ 2xS
}from |a|2. Information about the position of the particle only is obtained byintegrating |a|2 over all values of pD, i.e., by forming the marginal distributionwith respect to xD,
|a|2x =∫ +∞
−∞|a|2 dpD = wcl
x (xD, xS) .
The result is the same as in the classical case. Again in the limit σxD → 0we find that the function |a|2x approaches the quantum-mechanical probabilitydensity ρS(x) which is equal to the classical distribution ρSx (x).
In Figures C.3 and C.4 we demonstrate the construction of the analyzingamplitude using particular numerical examples. Each column of three plotsin the two figures is one example. At the top of the column the particle wavefunction is shown as two curves depicting ReϕS(x) and ImϕS(x) together withthe numerical values of the parameters xS, pS, σxS which define ϕ(x). Like-wise, the middle plot shows the detector wave function, given by ReϕD(x)and ImϕD(x). The bottom plot contains the real and imaginary parts of theproduct function
ϕ∗D(x)ϕS(x) ,
which after integration and absolute squaring yields the probability density
|a|2 = 1
h
∣∣∣∣∫ +∞
−∞ϕ∗
D(x)ϕS(x)dx
∣∣∣∣2
of detection. Also given in the bottom plot is the numerical value of |a|2.Four different situations are shown in the two figures. In each case the samedetector function ϕD is used. Only the particle wave function ϕS changes fromcase to case.
(i) In the left-hand column of Figure C.3 ϕS and ϕD are identical. Forthat case we know that the overlap integral is explicitly real and that∫ +∞
−∞ ϕ∗SϕS dx = 1, so that |a|2 = 1/h = 1/2π in the units h̄ = 1 used.
(ii) In the right-hand column of Figure C.3 the particle wave packet ismoved to a position expectation value xS = xD, but we still havepS = pD, σxS = σxD. By construction, the overlap function is differentfrom zero in that x region where both ϕS and ϕD are sizably differentfrom zero. As expected, the value of |a|2 is considerably smaller thanin case (i).
C.2 Analyzing Amplitude: Free Particle 437
Fig.C.3. Wave function ϕS (top) and ϕD (middle) and the product function ϕ∗DϕS (bottom).
Real parts are drawn as thick lines, imaginary parts as thin lines. The two columns differin the mean value xS of ϕS. Units h̄ = 1 are used.
(iii) In the left-hand column of Figure C.4 the position expectation valuesand the widths of particle and detector wave function are identical, xS =xD, σxS = σxD, but the momentum expectation values differ, pS = pD.As in case (i) the product function ϕ∗
DϕS is different from zero in theregion x ≈ x0 but due to the different momentum expectation values itoscillates. Therefore, the value of |a|2 is much smaller than in case (i)since positive and negative regions of the product function nearly cancelwhen the integration is performed.
438 C. Analyzing Amplitude
Fig.C.4. As Figure C.3 but for different functions ϕS. The two columns differ only in thevalue of σxS.
(iv) In the right-hand column of Figure C.4 the particle wave packet hasa larger width σxS > σxD. All other parameters are as in case (iii). Theproduct function is similar to that for case (iii) and is concentrated in theregion xD −σxD ≤ x ≤ xD +σxD where both wave functions are apprecia-bly different from zero. However, the amplitude of the product functionis smaller than in case (iii) since the amplitude of ϕS(x) is smaller in theoverlap region. Therefore, the value of |a|2 is also smaller.
C.3 Analyzing Amplitude: General Case
The lesson learned in the last section can be generalized to the analysis of anarbitrary normalized wave function ϕ(x) describing a single particle in termsof an arbitrary complete or overcomplete set of normalized wave functionsϕ(x) or ϕ(x ,q1, . . . ,qN ). The functions ϕn(x) can in particular be eigenfunc-tions of a Hermitean operator, e.g., the energy. Examples for a set of over-complete functions ϕ(x ,q1, . . . ,qn) are
• free wave packets ϕD(x , xD, pD) as in the last section,
• coherent states of the harmonic oscillator ϕ(x , x0, p0) as we shall studyin detail in the next section, and
• minimum-uncertainty states of a set of noncommuting operators likethe operators Lx , L y , Lz of angular momentum or Sx , Sy , Sz of spin asinvestigated in Sections 10.5 and 17.2.
The analyzing amplitude for the different cases is given by
a = 1
N1
∫ +∞
−∞ϕ∗
n (x)ϕ(x)dx
or
a = 1
N2
∫ +∞
−∞ϕ∗(x ,q1, . . . ,qN )ϕ(x)dx .
Of course, a mutual analysis of two sets of analyzing functions is also ofinterest, e.g.,
a = 1
N3
∫ +∞
−∞ϕ∗
n (x)ϕ(x ,q1, . . . ,qN )dx .
The normalization constants have to be individually determined for every typeof analyzing amplitude.
C.4 Analyzing Amplitude: Harmonic Oscillator
For the harmonic oscillator of frequency ω we have discussed in Sections 6.3and 6.4 two sets of states in particular:
(i) The eigenstates ϕn corresponding to the energy eigenvalues En = (n +12 )h̄ω,
ϕn(x) = (√
2π2nn!σ0)−1/2 Hn
(x
σ0
)exp
{− x2
2σ0
},
C.4 Analyzing Amplitude: Harmonic Oscillator 439
440 C. Analyzing Amplitude
with the ground-state width
σ0 = √2σx , σx =
√h̄
2mω.
Plots of the ϕn are shown in Figure 6.5.
(ii) The coherent states,
ψ(x , t , x0, p0) =∞∑
m=0
am(x0, p0)ϕm(x)exp
{− i
h̄Emt
},
where the complex coefficients an are given by
an(x0, p0) = zn
√n!
exp
{−1
2z∗z
}, n = 0, 1, 2, . . . .
The variable z is complex and a dimensionless linear combination ofthe initial expectation values x0 of position and p0 of momentum,
z = x0
2σx+ i
p0
2σp, σp = h̄
2σx.
Plots of the coherent states ψ(x , t) are shown in Figure 6.6c.
The set of energy eigenfunctions ϕn(x) is complete, the set of coherent statesis overcomplete. We can form four kinds of analyzing amplitudes.
Eigenstate – Eigenstate Analyzing Amplitude
We analyze the energy eigenfunctions using energy eigenfunctions as analyz-ing wave functions. Thus, we obtain as analyzing amplitude
amn =∫ +∞
−∞ϕm(x)ϕn(x)dx = δmn ,
which yields as probability
a2mn = δmn .
This result, based on the orthonormality of the eigenfunctions ϕn(x), is illus-trated in Figure C.5 which shows the functions ϕmϕn. Whereas ϕ2
m is non-negative everywhere so that the integral over ϕ2
m cannot vanish it is qualita-tively clear from the figure that the integral over ϕmϕn vanishes for m = n.
The analysis of an eigenstate ϕn(x) with all eigenstates ϕm(x) thus yieldswith probability a2
mn = 1 the answer that the original wave function was indeedϕn, and with probability amn = 0 the result that the original wave functionwas ϕm with m = n. Such an analysis can also be considered as an energydetermination which with certainty yields the energy eigenvalue En.
C.4 Analyzing Amplitude: Harmonic Oscillator 441
Fig.C.5. Product ϕm(x)ϕn(x) of the wave functions of the harmonic oscillator for h̄ω= 1.The long-dash curve indicates the potential energy V (x), the short-dash lines show theenergy eigenvalue En of the functions ϕn . These lines also serve as zero lines for theproduct functions.
Eigenstate – Coherent-State Analyzing Amplitude
The function to be analyzed is the time-dependent wave function ψ(x , t , x0,p0) of the coherent state. The analyzing function is the energy eigenfunctionϕn(x). As analyzing amplitude we obtain
a(n, x0, p0) =∫ +∞
−∞ϕn(x)ψ(x , t , x0, p0)dx
= zn
√n!
exp
{−1
2z∗z
}exp
{− i
h̄Ent
}.
The corresponding probability is given by
|a(n, x0, p0)|2 = (|z|2)n
n!e−|z|2 , |z|2 = x2
0
4σ 2x
+ p20
4σ 2p
.
The probabilities |a(n, x0, p0)|2 for fixed x0, p0 are distributed in the integer naccording to a Poisson distribution, cf. Appendix G. Its physical interpretationcan be understood if we express |z|2 in terms of the expectation value of thetotal energy of an oscillator with initial values x0 and p0,
442 C. Analyzing Amplitude
E0 = p20
2m+ m
2ω2x2
0 .
We find|z|2 = E0/(h̄ω) = n0 ,
i.e., |z|2 equals the number n0 of energy quanta h̄ω making up the energy E0
of the classical oscillator. This number, of course, need not be an integer. Forthe absolute square of the analyzing amplitude we thus find
|an(x0, p0)|2 = nn0
n!e−n0 .
It is the probability of a Poisson distribution for the number of energy quantan found when analyzing a coherent wave function with the eigenfunctions ϕn.It has the expectation value
〈n〉 = n0
and the variancevar(n) = n0 .
Coherent-State – Eigenstate Analyzing Amplitude
Analyzing the eigenstate wave functions ϕn(x) with the coherent state wavefunctions for t = 0,
ϕD(x , xD, pD) =∞∑
n=0
an(xD, pD)ϕn(x) ,
with the coefficients
an(xD, pD) = znD√n!
exp
{−1
2z∗
DzD
}, zD = xD
2σx+ i
pD
2σp,
we find as the analyzing amplitude
a(xD, pD,n) = 1√h
∫ +∞
−∞ϕ∗
D(x , xD, pD)ϕn(x)dx
= 1√h
znD√n!
exp
{−1
2z∗
DzD
},
and for its absolute square
|a(xD, pD,n)|2
= 1
2π (√
2σx )(√
2σp)
1
n!
(x2
D
4σ 2x
+ p2D
4σ 2p
)n
exp
{−
(x2
D
4σ 2x
+ p2D
4σ 2p
)}.
C.4 Analyzing Amplitude: Harmonic Oscillator 443
Fig.C.6. Absolute square |a(xD, pD,n)|2 of the amplitude analyzing the harmonic-oscil-lator eigenstate ϕn(x) with a coherent state of position and momentum expectation valuexD and pD.
For a given quantum number n of the eigenstate, |a|2 is a probability den-sity in the xD, pD phase space of the analyzing coherent state which is shownin Figure C.6 for a few values of n. It has the form of a ring wall with themaximum probability at
|zD|2 = x2D
4σ 2x
+ p2D
4σ 2p
= n .
In terms of the energy
ED = p2D
2m+ m
2ω2x2
D
of a classical particle of mass m with position xD and momentum pD in aharmonic oscillator of angular frequency ω, we have
|zD|2 = x2D
4σ 2x
+ p2D
4σ 2p
= nD ,
where nD is the average number of energy quanta h̄ω in the analyzing wavefunction ϕD(x , xD, pD). We find
444 C. Analyzing Amplitude
|a(xD, pD,n)|2 = 1
he−nD
nnD
n!.
For a given eigenstate ϕn(x) of the harmonic oscillator the probability den-sity in the xD, pD phase space of coherent states depends only on the averagenumber nD of quanta in the analyzing coherent state.
Using as analyzing wave functions the coherent states ϕD(x , xD, pD), the an-alyzing amplitude for the time-dependent coherent states ψ(x , t , x0, p0) turnsout to be
a(xD, pD, x0, p0, t)
= 1√h
∫ +∞
−∞ϕ∗
D(x , xD, pD)ψ(x , t , x0, p0)dx
= 1√h
exp
{−1
2
(z∗
DzD +2z∗Dz(t)+ z∗(t)z(t)
)}exp
{− i
2ωt
}with
z(t) = ze−iωt , z = x0
2σx+ i
p0
2σp.
The absolute square yields
|a(xD, pD, x0, p0, t)|2
= 1
2π (√
2σx )(√
2σp)exp
{−1
2
[(xD − x0(t))2
2σ 2x
+ (pD − p0(t))2
2σ 2p
]}with
x0(t) = x0 cosωt + p0
mωsinωt ,
p0(t) = −mωx0 sinωt + p0 cosωt
representing the expectation values of position and momentum of the coherentstate ψ(x , t , x0, p0) at time t . It is a bivariate Gaussian in the space of xD andpD centered about the classical positions x0(t), p0(t) of the oscillator. The
C.4 Analyzing Amplitude: Harmonic Oscillator 445
probability density |a(xD, pD, x0, p0, t)|2 shows the same behavior as that of aclassical particle. The expectation values of the position xD and of momentumpD are simply given by the classical values
〈xD〉 = x0(t) , 〈pD〉 = p0(t) .
The variances of xD and pD are
var(xD) = 2σ 2x ,
var(pD) = 2σ 2p .
This is twice the values of the ones of the coherent state itself, a consequenceof the broadening caused by the analyzing wave packet ϕD(x) having itselfthe variances σ 2
x and σ 2p .
The classical Gaussian phase-space probability density corresponding toψ(x , t , x0, p0) is of the same form as |a(xD, pD, x0, p0, t)|2. It possesses, how-ever, the widths σx and σp, and has the explicit form
ρcl(x , p, x0, p0, t)
= 1
2πσxσpexp
{−1
2
[(x − x0(t))2
σ 2x
+ (p − p0(t))2
σ 2p
]}.
By the same token the classical phase-space density corresponding to the de-tecting wave packet is
ρclD(x , p, xD, pD)
= 1
2πσxσpexp
{−1
2
[(x − xD)2
σ 2x
+ (p − pD)2
σ 2p
]}.
The functions ρcl and ρclD are equal to the Wigner distributions (cf. Ap-
pendix D) of ϕ and ϕD, respectively. The analyzing probability density |a|2can again be written as
|a(xD, pD, x0, p0, t)|2
=∫ +∞
−∞
∫ +∞
−∞ρcl
D(x , p, xD, pD)ρcl(x , p, x0, p0, t)dxD dpD ,
which once more shows the reason for the broadening of |a|2 relative to ρ.
D. Wigner Distribution
The quantum-mechanical analog to a classical phase-space probability den-sity is a distribution introduced by Eugene P. Wigner in 1932. In the sim-ple case of a one-dimensional system described by a wave function ϕ(x) theWigner distribution is defined by
W (x , p) = 1
h
∫ +∞
−∞exp
{i
h̄py
}ϕ(x − y
2)ϕ∗(x + y
2)dy .
For an uncorrelated Gaussian wave packet with the wave function
ϕ(x , x0, p0) = 14√
2π√σx
exp
{− (x − x0)2
4σ 2x
+ ip0
h̄(x − x0)
}it has the form of a bivariate normalized Gaussian:
W (x , p, x0, p0) = 1
2πσxσpexp
{−1
2
[(x − x0)2
σ 2x
+ (p − p0)2
σ 2p
]},
where σx and σp fulfill the minimum-uncertainty relation
σxσp = h̄/2 .
The expression obtained for W (x , p, x0, p0) coincides with the classical phase-space probability density for a single particle introduced in Section 3.6. Themarginal distributions in x or p of the Wigner distribution are
of the wave function ϕ(x), i.e., ϕ̃(p) is the wave function in momentum space.For the case of the Gaussian wave packet we find for the marginal distri-
butions
Wp(x) = 1√2πσx
exp
{−1
2
(x − x0)2
σ 2x
},
Wx (p) = 1√2πσp
exp
{−1
2
(p − p0)2
σ 2p
}.
An alternative representation for the Wigner distribution can be obtainedby introducing the wave function in momentum space into the expressiondefining W (x , p) with the help of
ϕ(x) = 1√2π h̄
∫ +∞
−∞exp
{ip
h̄x
}ϕ̃(p)dp .
We find
W (x , p) = 1
h
∫ +∞
−∞exp
{− i
h̄xq
}ϕ̃(
p − q
2
)ϕ̃∗
(p + q
2
)dq .
A note of caution should be added: For a general wave function ϕ(x) theWigner distribution is not a positive function everywhere. Thus, in general itcannot be interpreted as a phase-space probability density. It is, however, areal function
W ∗(x , p) = W (x , p) .
As an example we indicate the Wigner distribution W (x , p,n) for theeigenfunction ϕn(x) of the harmonic oscillator,
W (x , p,n) = (−1)n
π h̄L0
n
(x2
σ 2x
+ p2
σ 2p
)exp
{−1
2
(x2
σ 2x
+ p2
σ 2p
)}.
Here, the widths σx , σp are given by
σx = √h̄/(2mω) , σp = h̄/(2σp) ,
and L0n(x) is the Laguerre polynomial with upper index k = 0 as discussed in
Section 13.4.Figure D.1 shows the Wigner distributions for the lowest four eigenstates
of the harmonic oscillator n = 0, 1, 2, 3 plotted over the plane of the scaledvariables x/σx , p/σp. Accordingly, the plots are rotationally symmetric aboutthe z axis of the coordinate frame. The nonpositive regions of W (x , p,n) canbe clearly seen. The corresponding plots for the absolute square of the ana-lyzing amplitude are shown in Figure C.6.
448 D. Wigner Distribution
Fig.D.1. Wigner distributions W (x , p,n) of the harmonic-oscillator eigenstates ϕn(x),n = 0, 1, 2, 3.
The relation to the analyzing amplitude can easily be inferred from thefollowing observation. For an arbitrary analyzing wave function ϕD(x) weform the Wigner distribution
WD(x , p) = 1
h
∫ +∞
−∞exp
{i
h̄py′
}ϕD
(x − y′
2
)ϕ∗
D
(x + y′
2
)dy′ .
Then, the integral over x and p of the product of WD and W yields∫ +∞
−∞
∫ +∞
−∞WD(x , p)W (x , p)dx dp = 1
h
∣∣∣∣∫ +∞
−∞ϕ∗
D(x)ϕ(x)dx
∣∣∣∣2
= |a|2 .
This is to say, analyzing the Wigner distribution W (x , p) of a wave functionϕ(x) with the Wigner distribution WD(x , p) of an (arbitrary) analyzing wavefunction ϕD(x) yields exactly the absolute square of the analyzing amplitude
a = 1√h
∫ +∞
−∞ϕ∗
D(x)ϕ(x)dx
introduced in Appendix C.
D. Wigner Distribution 449
The temporal evolution of a Wigner distribution
W (x , p, t) = 1
h̄
∫ +∞
−∞e
ih̄ pyψ∗(x + y
2, t)ψ(x − y
2, t)dy
corresponding to a time-dependent solution ψ(x , t) of the Schrödinger equa-tion with the Hamiltonian H = p2/2m +V (x), p = (h̄/i)d/dx , is governed bythe Wigner–Moyal equation. It is the quantum-mechanical analog of the Li-ouville equation for a classical phase-space distribution. For potentials V (x)which are constant, linear, or quadratic in the coordinate x or linear combina-tions of these powers the two equations of Wigner and Moyal, and of Liouvilleare identical. For these types quantum-mechanical and classical phase-spacedistributions that coincide at one instant t in time, say the initial one, coincideat all the times.
E. Gamma Function
The gamma function �(z) introduced by Leonhard Euler is a generalizationof the factorial function for integers n,
n! = 1 ·2 ·3 · . . . ·n , 0! = 1! = 1 ,
to noninteger and eventually complex numbers z. It is defined by Euler’s in-tegral
�(z) =∫ ∞
0t z−1e−t dt , Re(z)> 0 .
By partial integration of ∫ ∞
0t ze−t dt = �(1+ z)
we find the recurrence formula
�(1+ z) = −t ze−t∣∣∞0
+ z∫ ∞
0t z−1e−t dt = z�(z)
valid for complex z.From Euler’s integral we obtain
�(1) = 1
and, thus, with the help of the recurrence relation for non-negative integer n,
�(1+n) = n! .
Euler’s integral can also be computed in closed form for z = 1/2,
�
(1
2
)= √
π ,
so that – again through the recurrence relation – it is easy to find the gammafunction for positive half-integer arguments.
For nonpositive integer arguments the gamma function has poles as canbe read off the reflection formula
�(1− z) = π
�(z) sin(π z)= π z
�(1+ z) sin(π z).
In Figure E.1 we show graphs of the real and the imaginary part of �(z)as surfaces over the complex z plane. The most striking features are the polesfor nonpositive real integer values of z. For real arguments z = x the gammafunction is real, i.e., Im(�(x)) = 0. In Figure E.2 we show �(x) and 1/�(x).The latter function is simpler since it has no poles. The gamma function forpurely imaginary arguments z = iy, y real, is shown in Figure E.3.
For complex argument z = x + iy an explicit decomposition into real andimaginary part can be given,
�(x + iy) = (cosθ + i sinθ)|�(x)|∞∏
j=0
| j + x |√y2 + ( j + x)2
,
where the angle θ is determined by
θ = yψ(x)+∞∑
j=0
[y
j + x− arctan
y
j + x
].
Here ψ(x) is the digamma function
ψ(x) = d
dx(ln�(x)) = �′(x)
�(x).
For integer n the following formula follows from the recurrence formula:
The gamma function of a purely imaginary argument can be obtained by spe-cialization of the argument x + iy to x = 0 in �(x + iy) to yield
�(iy) = (sinθ − i cosθ )√
π
y sinh y
with
θ = −γ y +∞∑
j=1
[y
j− arctan
y
j
],
where Euler’s constant γ is given by
γ = −ψ(1) = limn→∞
(n−1∑k=1
1
k− lnn
)= 0.5772156649 . . . .
452 E. Gamma Function
Fig.E.1. Real part (top) and imaginary part (bottom) of �(z) over the complex z plane.
E. Gamma Function 453
Fig.E.2. The functions �(x) and 1/�(x) for real arguments x .
454 E. Gamma Function
Fig.E.3. Real and imaginary parts of the gamma function for purely imaginary argu-ments.
F. Bessel Functions and Airy Functions
Bessel’s differential equation
x2 d2 Zν(x)
dx2+ x
dZν(x)
dx+ (x2 −ν2)Zν(x) = 0
is solved by the Bessel functions of the first kind Jν(x), of the second kind(also called Neumann functions) Nν(x), and of the third kind (also called Han-kel functions) H (1)
ν (x) and H (2)ν (x) which are complex linear combinations of
the former two. The Bessel functions of the first kind are
Jν(x) =(x
2
)ν ∞∑k=0
(−1)k
k!�(ν+ k +1)
(x2
4
)k
,
where �(z) is Euler’s gamma function.The Bessel functions of the second kind are
Nν(x) = 1
sinνπ[Jν(x)cosνπ − J−ν(x)] .
For integer ν = n one has
J−n(x) = (−1)n Jn(x) .
The modified Bessel functions are defined as
Iν(x) =(x
2
)ν ∞∑k=0
1
k!�(ν+ k +1)
(x2
4
)k
.
The Hankel functions are defined by
H (1)ν (x) = Jν(x)+ iNν(x) ,
H (2)ν (x) = Jν(x)− iNν(x) .
The following relations hold for the connections of the functions just dis-cussed and the spherical Bessel, Neumann, and Hankel functions, cf. Sec-tion 10.8. The spherical Bessel functions of the first kind are
In Figures F.1 and F.2 we show the functions Jν(x) and Iν(x) for ν =−1, −2/3, −1/3, . . . , 11/3. The features of these functions are simple to de-scribe for ν ≥ 0. The functions Jν(x) oscillate around zero with an amplitudethat decreases with increasing x , whereas the functions Iν(x) increase mono-tonically with x . At x = 0 we find Jν(0) = Iν(0) = 0 for ν > 0. Only for ν = 0we have J0(0) = I0(0) = 1. For ν > 1 there is a region near x = 0 in which thefunctions essentially vanish. The size of this region increases with increasingindex ν. For negative values of the index ν the functions may become verylarge near x = 0.
Closely related to the Bessel functions are the Airy functions Ai(x) andBi(x). They are solutions of the differential equation(
d2
dx2− x
)f (x) = 0
and are given by
Ai(x) =
⎧⎪⎪⎪⎨⎪⎪⎪⎩1
3
√x
{I−1/3
(2
3x3/2
)− I1/3
(2
3x3/2
)}, x > 0
1
3
√x
{J−1/3
(2
3|x |3/2
)+ J1/3
(2
3|x |3/2
)}, x < 0
,
and
Bi(x) =
⎧⎪⎪⎪⎨⎪⎪⎪⎩√
x
3
{I−1/3
(2
3x3/2
)+ I1/3
(2
3x3/2
)}, x > 0√
x
3
{J−1/3
(2
3|x |3/2
)− J1/3
(2
3|x |3/2
)}, x < 0
.
Graphs of these functions are shown in Figure F.3. Both functions oscillatefor x < 0. The wavelength of the oscillation decreases with decreasing x . Forx > 0 the function Ai(x) drops fast to zero whereas Bi(x) diverges.
F. Bessel Functions and Airy Functions 457
Fig.F.1. Bessel functions Jν(x).
458 F. Bessel Functions and Airy Functions
Fig.F.2. Modified Bessel functions Iν(x).
F. Bessel Functions and Airy Functions 459
Fig.F.3. Airy functions Ai(x) and Bi(x).
G. Poisson Distribution
In Section 3.3 we first introduced the probability density ρ(x), which is nor-malized to one, ∫ +∞
−∞ρ(x)dx = 1 .
We also introduced the concepts of the expectation value of x ,
〈x〉 =∫ +∞
−∞xρ(x)dx ,
and of the variance of x ,
var(x) = σ 2x = ⟨
(x −〈x〉)2⟩
.
We now replace the continuous variable x by the discrete variable k whichcan assume only certain discrete values, e.g., k = 0,1,2, . . . . In a statisticalprocess the variable k is assumed with the probability P(k). The total proba-bility is normalized to one, ∑
k
P(k) = 1 ,
where the summation is performed over all possible values of k.The average value, mean value, or expectation value of k is
〈k〉 =∑
k
k P(k) ,
and the variance of k is
var(k) = σ 2(k) = ⟨(k −〈k〉)2
⟩ = ∑k
(k −〈k〉)2 P(k) .
The simplest case is that of an alternative. The variable only takes thevalues
κ = 0,1 .
The process yields with probability p = P(1) the result κ = 1 and with prob-ability P(0) = 1 − p the result κ = 0. Therefore, the expectation value of κis
We now consider a process which is a sequence of n independent alter-natives each yielding the result κi = 0,1, i = 1,2, . . . ,n. We characterize theresult of the process by the variable
k =n∑
i=1
κi ,
which has the rangek = 0,1, . . . ,n .
A given process yields the result k if κi = 1 for k of the n alternatives andκi = 0 for (n − k) alternatives. The probability for the sequence
is pk(1 − p)n−k . But this is only one particular sequence leading to the resultk. In total there are (
n
k
)= n!
k!(n − k)!
such sequences where
n! = 1 ·2 ·3 · . . . ·n , 0! = 1! = 1 .
Therefore, the probability that our process yields the result k is
P(k) =(
n
k
)pk(1− p)n−k .
This is the binomial probability distribution. The expectation value can becomputed by introducing P(k) into the definition of 〈k〉 or, even simpler, from
〈k〉 =n∑
i=1
〈κi〉 = np .
In Figure G.1 we show the probabilities P(k) for various values of n butfor a fixed value of the product λ = np. The distribution changes drasticallyfor small values of n but seems to approach a limiting distribution for verylarge n. Indeed, we can write
P(k) = n!
k!(n − k)!
(λ
n
)k(1− λ
n
)n(1− λ
n
)k
462 G. Poisson Distribution
Fig.G.1. Binomial distributions for various values of n but fixed product np = 3.
= λk
k!
(1− λ
n
)n n(n −1) · . . . · (n − k +1)
nk(1− λ
n
)k
= λk
k!
(1− λ
n
)n(1− 1
n
)(1− 2
n
) · . . . · (1− k−1n
)(1− λ
n
)k .
In the limit n → ∞ every term in brackets in the last factor approaches one,and since
limn→∞
(1− λ
n
)n
= e−λ
we have
P(k) = λk
k!e−λ .
This is the Poisson probability distribution. It is shown for various values ofthe parameter λ in Figure G.2. The expectation value of k is
〈k〉 =∞∑
k=0
kλk
k!e−λ =
∞∑k=1
kλk
k!e−λ
=∞∑
k=1
λλk−1
(k −1)!e−λ = λ
∞∑j=0
λ j
j!e−λ = λ .
G. Poisson Distribution 463
Fig.G.2. Poisson distributions for various values of the parameter λ.
In a similar way one finds
〈k2〉 = λ(λ+1)
and therefore, also the variance of k is equal to λ,
var(k) = ⟨(k −〈k〉)2
⟩ = ⟨k2 −2k〈k〉+〈k〉2
⟩= 〈k2〉−2〈k〉2 +〈k〉2 = 〈k2〉−〈k〉2
= λ(λ+1)−λ2 = λ .
The Poisson distribution is markedly asymmetric for small values of λ. Forlarge λ, however, it becomes symmetric about its mean value λ and in thatcase its bell shape resembles that of the Gaussian distribution.