Quantum Mechanics: A Mathematical Introduction

Quantum Mechanics
A Mathematical Introduction
ANDREW LARKOSKI
SLAC NATIONAL ACCELERATOR LABORATORY
Copyright 2023. This publication is in copyright. Subject to statutory exception and
to the provisions of relevant collective licensing agreements, no reproduction of any
part may take place without the written permission of Cambridge University Press.
Contact me at larkoski@slac.stanford.edu for any errors, corrections, or suggestions.

Contents
2 Linear Algebra 1
2.1 1
2.2 6
2.3 6
2.4 8
2.5 9
2.6 10
2.7 11
2.8 11
3 Hilbert Space 14
3.1 14
3.2 15
3.3 16
3.4 16
3.5 18
3.6 18
3.7 19
3.8 20
4 Axioms of Quantum Mechanics and Their Consequences 22

4.1 22
4.2 22
4.3 24
4.4 26
4.5 27
4.6 29
4.7 30
4.8 31
5 Quantum Mechanical Example: The Inﬁnite Square Well 34

5.1 34
5.2 36
5.3 36
5.4 39
5.5 41
iii
iv Contents
5.6 42
5.7 45
5.8 47
6 Quantum Mechanical Example: The Harmonic Oscillator 49

6.1 49
6.2 50
6.3 51
6.4 52
6.5 54
6.6 56
6.7 57
6.8 59
7 Quantum Mechanical Example: The Free Particle 62

7.1 62
7.2 63
7.3 65
7.4 66
7.5 68
7.6 70
7.7 72
7.8 73
8 Rotations in Three Dimensions 75

8.1 75
8.2 77
8.3 78
8.4 78
8.5 79
8.6 80
8.7 80
8.8 81
9 The Hydrogen Atom 84

9.1 84
9.2 85
9.3 87
9.4 91
9.5 91
9.6 93
9.7 93
9.8 95
v Contents
10 Approximation Techniques 97
10.1 97
10.2 98
10.3 99
10.4 100
10.5 103
10.6 105
10.7 105
10.8 108
10.9 110
11 The Path Integral 112

11.1 112
11.2 113
11.3 115
11.4 115
11.5 116
11.6 117
11.7 119
11.8 120
12 The Density Matrix 122

12.1 122
12.2 125
12.3 127
12.4 129
12.5 130
12.6 132
12.7 134
12.8 134
12.9 137
2 Linear Algebra
Exercises
2.1 (a) The 2 × 2 and 3 × 3 discrete derivative matrices can actually be read off from
the general form provided there:
1

0
D2×2 = 2∆x , (2.1)
− 2∆x
1
0
 1 
0 2∆x 0
D3×3 =  − 2∆x 1
0 1 
2∆x . (2.2)
0 − 2∆x1
0
(b) To calculate the eigenvalues, we construct the characteristic equation where

1
det(D2×2 − λ I) = 0 = λ 2 + . (2.3)
4∆x2
Then, the eigenvalues of the 2 × 2 derivative matrix are
i
λ =± . (2.4)
2∆x
For the 3 × 3 derivative matrix, its characteristic equation is
 
−λ 1
2∆x 0
det(D3×3 − λ I) = 0 = det  − 2∆x1
−λ 1 
2∆x (2.5)
0 − 1
−λ
2∆x
1 λ
= (−λ ) λ 2 + − .
4∆x2 4∆x2
One eigenvalue is clearly 0, and the other eigenvalues satisfy

1
λ2 + = 0, (2.6)
2∆x2
and so the 3 × 3 derivative matrix has eigenvalues
i
λ = 0, ± √ . (2.7)
2∆x
All non-zero eigenvalues are exclusively imaginary numbers.
1
2 2 Linear Algebra
(c) For the 2 × 2 derivative matrix, the eigenvector equation can be expressed as
1

0 a i a
2∆x =± , (2.8)
− 2∆x
1
0 b 2∆x b
for some numbers a, b. This linear equation requires that
b = ±ia , (2.9)
and so the eigenvectors can be expressed as

1 1
⃗v1 = a , ⃗v2 = a . (2.10)
i −i
To ensure that they are unit normalized, we require that

∗ 1
⃗v1 ·⃗v1 = a (1 − i)
2
= 2a2 , (2.11)
i
√
or that a = 1/ 2 (ignoring a possible overall complex phase). Thus, the
normalized eigenvectors are

1 1 1 1
⃗v1 = √ , ⃗v2 = √ . (2.12)
2 i 2 −i
Note that these are mutually orthogonal:

1 1
⃗v1∗ ·⃗v2 = (1 − i) = 1−1 = 0. (2.13)
2 −i
For the 3 × 3 derivative matrix, we will first determine the eigenvector
corresponding to 0 eigenvalue, where
 1    
0 2∆x 0 a 0
 − 1 0 1 
b  =  0 . (2.14)
2∆x 2∆x
0 − 2∆x
1
0 c 0
Performing the matrix multiplication, we find that
   
b 0
 c−a  =  0  . (2.15)
−b 0
This then enforces that b = 0 and c = a. That is, the normalized eigenvector
with 0 eigenvalue is (again, up to an overall complex phase)
 
1
1  
⃗v1 = √ 0 . (2.16)
2
1
Next, the eigenvectors for the non-zero eigenvalues satisfy
 1    
0 2∆x 0 a a
 − 1 1   i  
2∆x 0 2∆x b = ± √ b . (2.17)
2∆x
0 − 2∆x
1
0 c c
3 Exercises
Performing the matrix multiplication, we find

   
b a
1  i
c−a  = ±√  b , (2.18)
2∆x 2∆x
−b c
or that
√ i
b = ± 2i a , c = ± √ b = −a . (2.19)
2
Then, the other two eigenvectors are
   
√1 √1
⃗v2 = a  2i  , ⃗v3 = a  − 2i  . (2.20)
−1 −1
The value of a can be determined by demanding that they are normalized:

 
√ √1
⃗v2 ·⃗v2 = a 1 − 2i − 1  2i  = 4a2 ,
∗ 2
(2.21)
−1
or that a = 1/2. That is,

   
1 1
1 √  1 √ 
⃗v2 = 2i , ⃗v3 = − 2i . (2.22)
2 2
−1 −1
All three eigenvectors, ⃗v1 ,⃗v2 ,⃗v3 , are mutually orthogonal. For example,
 
1 √ √1
1−2+1
⃗v2∗ ·⃗v3 = 1 − 2i − 1  − 2i  = = 0. (2.23)
4 4
−1
(d) To determine how the exponentiated matrix

∞
∆xn n
M = e∆xD = ∑ D, (2.24)
n=0 n!
acts on an eigenvector ⃗v with eigenvalue λ , let’s use the Taylor expanded

form. Note that
Dn⃗v = λ n⃗v , (2.25)
and so
∞ ∞
∆xn n (λ ∆x)n
M⃗v = ∑ D ⃗v = ∑ ⃗v = eλ ∆x⃗v . (2.26)
n=0 n! n=0 n!
We can then just plug in the appropriate eigenvalues and eigenvectors.

4 2 Linear Algebra
(e) Now, we are asked to determine the matrix form of the exponentiated 2 × 2
and 3 × 3 derivative matrices. Let’s start with the 2 × 2 matrix and note that
we can write
∞ n ∞ n
∆xn 0 1 1 0 1
M2×2 = e∆xD2×2 = ∑ 2∆x = ∑ n .
n=0 n!
− 2∆x
1
0 n=0 2 n!
−1 0
(2.27)
So,
the problem
is reduced to establishing properties of the matrix
0 1
. Note that the first few powers of the matrix are:
−1 0
0
0 1 1 0
= = I, (2.28)
−1 0 0 1
1
0 1 0 1
= , (2.29)
−1 0 −1 0
2
0 1 0 1 0 1 1 0
= =− = −I . (2.30)
−1 0 −1 0 −1 0 0 1
This pattern continues, and it can be compactly expressed as

2n 2n+1
0 1 0 1 0 1
= i2n I , = (−i)i2n+1 , (2.31)
−1 0 −1 0 −1 0
for n = 0, 1, 2, . . . . Then, the 2 × 2 exponentiated matrix can be expressed as

∞ ∞
i2n 0 1 i2n
M2×2 = I ∑ 2n + ∑ (2.32)
n=0 2 (2n)!
−1 0 n=0 2 2n+1 (2n + 1)!

1 1 0 1 cos 21 sin 12
= cos I + sin = .
2 2 −1 0 − sin 12 cos 21
A similar exercise can be repeated for the 3 × 3 derivative matrix. We want

to evaluate the sum
 1 n
∞ 0 0
∆x n
 − 1
2∆x
1 
M3×3 = e∆xD3×3 = ∑ 2∆x 0 2∆x (2.33)
n!
n=0 0 − 2∆x
1
0
 n
∞ 0 1 0
1 
=∑ n −1 0 1  .
2 n!
n=0 0 −1 0
Now, note that even powers of the matrix take the form
 0
0 1 0
 −1 0 1  = I, (2.34)
0 −1 0
5 Exercises
 2  
0 1 0 1 0 1
 −1 0 1  = 0 0 0  − 2I , (2.35)
0 −1 0 1 0 1
 4  
0 1 0 1 0 1
 −1 0 1  = 2 0 0 0  + 4I , (2.36)
0 −1 0 1 0 1
and so the general result takes the form
 2n  
0 1 0 1 0 1
 −1 0 1  = (−2)n−1  0 0 0  + (−2)n I , (2.37)
0 −1 0 1 0 1
for n = 1, 2, . . . . Products of odd powers of the matrix take the form
 1  
0 1 0 0 1 0
 −1 0 1  =  −1 0 1  , (2.38)
0 −1 0 0 −1 0
 3  
0 1 0 0 1 0
 −1 0 1  = −2  −1 0 1  , (2.39)
0 −1 0 0 −1 0
 5  
0 1 0 0 1 0
 −1 0 1  = 4  −1 0 1  . (2.40)
0 −1 0 0 −1 0
The general form is then
 2n+1  
0 1 0 0 1 0
 −1 0 1  = (−2)n  −1 0 1 , (2.41)
0 −1 0 0 −1 0
for n = 0, 1, 2, . . . . Putting these results together, the 3 × 3 exponentiated
matrix is
   
∞ n 1 0 1
(−2)  1 
M3×3 = I + ∑ 2n − 0 0 0  + I (2.42)
n=1 2 (2n)! 2
1 0 1
 
∞ 0 1 0
(−2)n  −1 0 1 
+ ∑ 2n+1
n=0 2 (2n + 1)!
0 −1 0
   
1 0 1 1 0 −1 ∞
1 1 (−1)n
=  0 0 0  +  0 2 0  ∑ √ 2n
2 2 n=0 2 (2n)!
1 0 1 −1 0 1
 
0 1 0 ∞
1 (−1)n
+ √  −1 0 1  ∑ √ 2n+1
2 (2n + 1)!
0 −1 0 n=0 2
6 2 Linear Algebra
     
1 0 1 cos √12 1 0 −1 sin √12 0 1 0
1
=  0 0 0  +  0 2 0  + √  −1 0 1 .
2 2 2
1 0 1 −1 0 1 0 −1 0
2.2 If we instead defined the derivative matrix through the standard asymmetric
difference, the derivative matrix would take the form
 .. .. .. 
..
. . . . ···
 
 ··· − 1 1
0 ··· 
 ∆x ∆x 
D=  ··· 0 − ∆x 1 1
∆x ···  . (2.43)
 ··· 0 0 − 1
· · · 
 ∆x 
.. .. .. .. ..
. . . . .
Such a matrix has no non-zero entries below the diagonal and so its characteristic
equation is rather trivial, for any number of grid points. Note that
 .. .. .. 
..
. . . . ···
 
 ··· − 1 −λ 1
0 ··· 
 ∆x ∆x 
det(Dn×n − λ I) = det   ··· 0 − ∆x
1
−λ 1
∆x ··· 
 ··· 0 0 − ∆x
1
−λ ··· 
 
.. .. .. .. ..
. . . . .
n
1
= 0 = − −λ . (2.44)
∆x
Thus, there is but a single eigenvalue, λ = −1/∆x.
2.3 (a) We are asked to express the quadratic polynomial p(x) as a linear combina-
tion of Legendre polynomials. So, we have
p(x) = ax2 + bx + c = d0 P0 (x) + d1 P1 (x) + d2 P2 (x) (2.45)

r r !
1 3 5 2
= d0 √ + d1 x + d2 (3x − 1) ,
2 2 8
for some coefficients d0 , d1 , d2 . First, matching the coefficient of x2 , we must

have
r
5
3 d2 = a , (2.46)
8
or that
r
1 8
d2 = a. (2.47)
3 5
Next, matching coefficients of x, we have
r
3
d1 = b, (2.48)
2
7 Exercises
or that
r
2
d1 = b. (2.49)
3
Finally, matching coefficients of x0 , we have
r
1 5 1 a
d0 √ − d2 = d0 √ − = c , (2.50)
2 8 2 3
or that
√
√ 2
d0 = 2c + a. (2.51)
3
Then, we can express the polynomial as the linear combination
√ ! r r
√ 2 2 1 8
p(x) = 2c + a P0 (x) + bP1 (x) + aP2 (x) , (2.52)
3 3 3 5
or as the vector in the space of Legendre polnomials

 √ √
2

2c
q + 3 a
 
p(x) = 

2
b .
 (2.53)
q3
1 8
3 5 a
(b) Let’s now act on this vector with the derivative matrix we constructed:
√  √ √   √ 
  2c + 32 a 2b
0 3 √0  q  q
d
p(x) =  0 0 15  
2 = 
  3a 
b 8 . (2.54)
dx  q3
0 0 0 1 8 0
3 5a
Then, re-interpreting this as a polynomial, we have that

r
d √ 8
p(x) = 2bP0 (x) + aP1 (x) = b + 2ax , (2.55)
dx 3
which is indeed the derivative of p(x) = ax2 + bx + c.
(c) Let’s first construct the second derivative matrix through squaring the first
derivative matrix:
 √  √   √ 
2 0 3 √0 0 3 √0 0 0 3 5
d d d
= = 0 0 15   0 0 15  =  0 0 0 .
dx2 dx dx
0 0 0 0 0 0 0 0 0
(2.56)
By contrast, the explicit matrix element in the i j position would be
2 Z 1
d d2
= dx Pi−1 (x) Pj−1 (x) . (2.57)
dx2 i j −1 dx2
8 2 Linear Algebra
For the second derivative to be non-zero, it must act on P2 (x), and so j = 3.

Further, by orthogonality of the Legendre polynomials, the only non-zero
value is i = 1. Then, the one non-zero element in the matrix is
2 Z 1 Z 1 r
d d2 1 5 √
2
= dx P0 (x) 2 P2 (x) = √ dx 3 = 3 5, (2.58)
dx 13 −1 dx 2 −1 2
which agrees exactly with just squaring the derivative matrix.
(d) To calculate the exponential of the derivative matrix, we need all of its pow-
ers. We have already calculated the first and second derivative matrices, but
what about higher powers? With only the first three Legendre polynomials,
it is easy to see that the third and higher derivative matrices are all 0:
 √  √   
0 3 √0 0 0 3 5 0 0 0
d3 d d2
= = 0 0 15   0 0 0 = 0 0 0 .
dx3 dx dx2
0 0 0 0 0 0 0 0 0
(2.59)
Thus, the Taylor expansion of the exponentiated derivative matrix termi-
nates after a few terms:
 √   √ 
0 3 √0 0 0 3 5
∆x 2
M = e∆x dx = I + ∆x  0 0 15  +  0 0 0  . (2.60)
d
2
0 0 0 0 0 0
The action of this matrix on the polynomial as expressed as a vector in
Legendre polynomial space is
 √ √   √ √   √ 
2c + 32 a 2c + 32 a  √ 
 q   q  q2b 8a
 
M
2
b 
 =  2
b 
 + ∆x  3a 
8 + ∆x 2
0 
q3 q3
1 8 1 8 0 0
3 5a 3 5a
= ax2 + bx + c + (b + 2a)∆x + 2a∆x2 (2.61)

= a(x + ∆x) + b(x + ∆x) + c ,
2
which is indeed a translation of the polynomial, as expected.

2.4 (a) To determine if integration is linear, we need to verify the two properties.
First, for two functions f (x) and g(x), anti-differentiation acts on their sum
as:
Z
dx ( f (x) + g(x)) = F(x) + G(x) + c , (2.62)
where we have
d d
F(x) = f (x) , G(x) = g(x) , (2.63)
dx dx
and c is an arbitrary constant. Linearity requires that this equal the sum of
their anti-derivatives:
Z Z
dx f (x) + dx g(x) = F(x) + G(x) + 2c . (2.64)
9 Exercises
Note that each integral picks up a constant c. So, the only way that anti-
differentiation can be linear is if the integration constant c = 0. Note also
that integration, i.e., the area under a curve, is independent of the integration
constant.
Next, we must also demand that multiplication by a constant is simple for
linearity. That is, for some constant c, we must have that
Z
dx c f (x) = c F(x) , (2.65)
which is indeed true by the Leibniz product rule.

(b) For two vectors ⃗f ,⃗g related by the action of the differentiation matrix,
D⃗f = ⃗g , (2.66)
we would like to define the anti-differentiation operator A as
⃗f = D−1⃗g = A⃗g . (2.67)
However, this clearly is only well-defined if the derivative operator is inverti-
ble; or, that it has no 0 eigenvalues. For the 3 × 3 derivative matrix, we found
that it had a 0 eigenvalue, and this property holds for higher dimensional
differentiation matrices. Thus, some other restrictions must be imposed for
anti-differentiation to be well-defined as a matrix operator.
2.5 (a) As a differential equation, the eigenvalue equation for the operator Ŝ is
d
f (x) = λ fλ (x) ,
−ix (2.68)
dx λ
for some eigenvalue λ and eigenfunction fλ . This can be rearranged into
d fλ dx
= iλ , (2.69)
fλ x
and has a solution
fλ (x) = c xiλ = c eiλ log x , (2.70)
for some constant c. Note that the logarithm only makes sense if x ≥ 0, so
we restrict the domain of the operator Ŝ to x ∈ [0, ∞). (This can be relaxed,
but you have to define what you mean by logarithm of a negative number
carefully.) For the eigenfunctions to be bounded, we must require that λ is
real-valued.
(b) Integration of the product of two eigenfunctions with eigenvalues λ1 ̸= λ2
yields
Z ∞ Z ∞ Z ∞
dx fλ1 (x)∗ fλ2 (x) = dx x−i(λ1 −λ2 ) = dx e−i(λ1 −λ2 ) log x , (2.71)
0 0 0
where we restrict to the domain x ∈ [0, ∞), as mentioned above. Now, let’s
change variables to y = log x, and now y ∈ (−∞, ∞), and the integral becomes
Z ∞ Z ∞
−i(λ1 −λ2 ) log x
dx e = dy ey e−i(λ1 −λ2 )y . (2.72)
0 −∞
10 2 Linear Algebra
Note the extra factor ey in the integrand; this is the Jacobian of the change
of variables x = ey , and so dx = ey dy. If this Jacobian were not there, then
the integral would be exactly like we are familiar with from Fourier trans-
forms. However, with it there, this integral is not defined. We will fix it in
later chapters.
(c) A function g(x) with a Taylor expansion about 0 can be expressed as
∞
g(x) = ∑ an x n , (2.73)
n=0
for some coefficients an . If we act Ŝ on this function, we find

∞ ∞
d n
Ŝg(x) = −i ∑ an x x = −i ∑ nan xn . (2.74)
n=0 dx n=0
Therefore, if m powers of Ŝ act on g(x), it returns

∞ ∞
d m n
Ŝm g(x) = ∑ an −ix x = ∑ (−in)m an xn . (2.75)
n=0 dx n=0
Now, the action of the exponentiated operator on g(x) is

∞ ∞ ∞
(iα )m m (iα )m
eiα Ŝ g(x) = ∑ Ŝ g(x) = ∑ ∑ (−in)m an xn (2.76)
m=0 m! m=0 n=0 m!
∞ ∞ ∞ ∞
(nα )m
= ∑ an xn ∑ m! = ∑ enα
an x n
= ∑ an (eα x)n
n=0 m=0 n=0 n=0
= g (eα x) .
That is, this operator rescales the coordinate x.

2.6 (a) For the matrix M, its characteristic equation is

a−λ b
det(M − λ I) = det = (a − λ )(d − λ ) − bc (2.77)
c d −λ
= λ 2 − (a + d)λ + (ad − bc) = 0 .
We can write this in a very nice way in terms of the trace and the determinant
of M:
det(M − λ I) = λ 2 − (tr M) λ + det M = 0 . (2.78)
(b) Solving for the eigenvalues, we find

q
tr M ± (tr M)2 − 4 det M
λ= . (2.79)
2
So, the eigenvalues are real iff (tr M)2 ≥ 4 det M, or, in terms of the matrix
elements,
(a + d)2 ≥ 4(ad − bc) . (2.80)

11 Exercises
(c) With the determinant equal to 1 and the trace equal to 0, the eigenvalues
are
√
0± 0−4
λ= = ±i , (2.81)
2
which are clearly not real-valued.
2.7 (a) The other elements of the matrix M we need in the ⃗u-vector basis are

⊺ M11 M12 cos θ
⃗u1 M⃗u1 = (cos θ sin θ ) (2.82)
M21 M22 sin θ
= M11 cos2 θ + M21 cos θ sin θ + M12 cos θ sin θ + M22 sin2 θ ,

M11 M12 cos θ
⃗u⊺2 M⃗u1 = (− sin θ cos θ ) (2.83)
M21 M22 sin θ
= −M11 cos θ sin θ + M21 cos2 θ − M12 sin2 θ + M22 cos θ sin θ ,

M11 M12 − sin θ
⃗u⊺2 M⃗u2 = (− sin θ cos θ ) (2.84)
M21 M22 cos θ
= M11 sin2 θ − M21 cos θ sin θ − M12 cos θ sin θ + M22 cos2 θ .
(b) In exercise 2.6, we had shown that the characteristic equation for a 2 × 2
matrix can be written as
det(M − λ I) = λ 2 − (tr M) λ + det M = 0 , (2.85)
where tr M is the sum of the diagonal elements of M. In the ⃗v-vector basis,
this characteristic equation is
λ 2 − (tr M) λ + det M = 0 = λ 2 − (M11 + M22 ) λ + (M11 M22 − M12 M21 ) .
(2.86)
Now, in the⃗u-vector basis, we can calculate the trace and determinant. First,
the trace:
tr M = ⃗u⊺1 M⃗u1 +⃗u⊺2 M⃗u2 (2.87)

= M11 cos θ + M21 cos θ sin θ + M12 cos θ sin θ + M22 sin θ
2 2

+ M11 sin2 θ − M21 cos θ sin θ − M12 cos θ sin θ + M22 cos2 θ
= M11 + M22 ,
exactly the same as in the ⃗v basis. The determinant is
det M = ⃗u⊺1 M⃗u1⃗u⊺2 M⃗u2 −⃗u⊺1 M⃗u2⃗u⊺2 M⃗u1 . (2.88)
Plugging in the explicit values for the matrix elements, one finds the identical
value of the determinant in either the ⃗v-vector or ⃗u-vector basis. Therefore,
the characteristic equation is independent of basis.
2.8 (a) Verifying orthonormality is straightforward. For m ̸= n, we have
Z 2π Z 2π
1
∗
d θ fm (θ ) fn (θ ) = d θ ei(n−m)θ (2.89)
0 2π 0
i
=− e2π i(n−m) − 1 = 0 ,
2π (n − m)
12 2 Linear Algebra
because n − m is still an integer. If instead n = m, we can use l’Hôpital’s rule

to find
i 2π i 2π i(n−m)
lim − e2π i(n−m) − 1 = lim −i e = 1. (2.90)
m→n 2π (n − m) m→n 2π
(b) The eigenvalue equation for the operator D̂ is
d
D̂ fλ (θ ) = −i f (θ ) = λ fλ (θ ) . (2.91)
dθ λ
The solutions are
fλ (θ ) = ceiλ θ , (2.92)
for some constant c. These eigenfunctions must be single-valued on θ [0, 2π )

and periodicity requires that
fλ (θ + 2π ) = ceiλ (θ +2π ) = fλ (θ ) = ceiλ θ . (2.93)
That is, we must require that λ is an integer, so that
e2π iλ = 1 . (2.94)
Thus, the eigenfunctions of D̂ are exactly the orthonormal complex expo-

nentials from part (a).
(c) We want to show that the sinusoidal form of the basis elements are also
orthonormal. First, for normalization, the integral of the square of cosine
is
Z 2π Z 2π
cos2 (nπ ) 1
dθ = d θ (1 + cos(2nθ )) = 1 , (2.95)
0 π 2π 0
because the integral of cosine over its entire domain is 0, and we used the
double angle formula. Essentially the same calculation follows for sine. For
orthogonality, we will again just show one explicit calculation, the product
of cosine and sine:
Z 2π Z 2π
1 1
d θ cos(nπ ) sin(mπ ) = d θ [sin ((n + m)θ ) − sin ((n − m)θ )]
π 0 2π 0
= 0, (2.96)
using the angle addition formulas and noting again that the sum and differ-
ence of two integers is still an integer, and integral of sine over its domain
√ is
0. Finally, the normalized basis element if n = 0 is just the constant, 1/ 2π ,
as identified in the exponential form.
(d) We are asked to determine the matrix elements of D̂ in the sinusoidal basis.
To do this, we sandwich the operator between the basis elements and inte-
13 Exercises
grate over the domain. First, note that cos-cos or sin-sin matrix elements
vanish. For example,
Z 2π Z 2π
1 in
d θ cos(mθ )D̂ cos(nθ ) = d θ cos(mπ ) sin(nπ ) = 0 , (2.97)
π 0 π 0
by orthogonality. The same argument holds for the sin-sin contribution.

Therefore, the only non-zero matrix elements are mixed sin-cos, like
Z 2π Z 2π
1 in
d θ cos(mθ )D̂ sin(nθ ) = − d θ cos(mθ ) cos(nθ ) = −inδmn .
π 0 π 0
(2.98)
For the opposite order, the matrix element is
Z 2π Z 2π
1 in
d θ sin(mθ )D̂ cos(nθ ) = d θ sin(mθ ) sin(nθ ) = inδmn . (2.99)
π 0 π 0
3 Hilbert Space
Exercises
3.1 (a) If the matrix M is normal, then MM† = M† M. The Hermitian conjugate of
M is
∗
a c∗
M† = . (3.1)
b∗ d ∗
The product MM† is

a b a∗ c∗ |a|2 + |b|2 ac∗ + bd ∗
MM† = = . (3.2)
c d b∗ d∗ a∗ c + b∗ d |c|2 + |d|2
The opposite product is

a∗ c∗ a b |a|2 + |b|2 a∗ c + b∗ d
M M=
†
= . (3.3)
b∗ d∗ c d ac∗ + bd ∗ |c|2 + |d|2
Then, for the matrix M to be normal, we must require that every element of
MM† and M† M are equal. Therefore, we must enforce that
a∗ c + b∗ d = ac∗ + bd ∗ . (3.4)
(b) A general upper triangular matrix takes the form

 
M11 M12 M13 ···
 0 M22 M23 ··· 
 
M= 0 0 M33 ··· . (3.5)
 
.. .. .. ..
. . . .
Its Hermitian conjugate is

 ∗ 
M11 0 0 ···
 M12∗ M22∗ 0 ··· 
 
M† =  M13∗ M23∗ M33∗ ··· . (3.6)
 
.. .. .. ..
. . . .
14
15 Exercises
To show that this matrix is not normal, we will just focus on the 11 entry of
the products with its Hermitian conjugate. Note that
  ∗ 
M11 M12 M13 · · · M11 0 0 ···
 0 M22 M23 · · ·   ∗ ∗ ··· 
   M12 M22 0 
MM† =  0 0 M33 · · ·   M13 M23 M33 · · · 
  ∗ ∗ ∗ (3.7)
 
.. .. .. .. .. .. .. ..
. . . . . . . .
!
|M11 |2 + |M22 |2 + |M33 |2 + · · · · · ·
= .. .. .
. .
By contrast, the product of matrices in the other order is
 ∗  
M11 0 0 ··· M11 M12 M13 ···
 M∗ M∗ 0 · · ·  0 M22 M23 ··· 
 12 22  
M† M =  M ∗ M ∗ M ∗ · · ·   0 0 M33 ···  (3.8)
 13 23 33  
.. .. .. .. .. .. .. ..
. . . . . . . .
!
|M11 | · · ·
2
= .. .. .
. .
Therefore, for a general upper triangular matrix it is not true that
|M11 |2 = |M11 |2 + |M22 |2 + |M33 |2 + · · · , (3.9)
and so upper triangular matrices are not normal.
3.2 (a) We want to calculate the exponentiated matrix
∞
(iϕ )n n
A = eiϕ σ3 = ∑ σ3 . (3.10)
n=0 n!
Note that this sum can be split into even and odd components noting that

1 0 1 0 1 0
σ3 =
2
= = I. (3.11)
0 −1 0 −1 0 1
Thus, we have
∞ ∞ ∞
(iϕ )n n (iϕ )2n (iϕ )2n+1
∑ σ 3 = I ∑ + σ 3 ∑ = I cos ϕ + iσ3 sin ϕ (3.12)
n=0 n! n=0 (2n)! n=0 (2n + 1)!
iϕ
cos ϕ + i sin ϕ 0 e 0
= = .
0 cos ϕ − i sin ϕ 0 e−iϕ
(b) Now, we are asked to evaluate the exponentiated matrix
∞
(iϕ )n
B = eiϕ (σ1 +σ3 ) = ∑ n!
(σ1 + σ3 )n . (3.13)
n=0
Note that

1 1 1 1 2 0
(σ1 + σ3 )2 = = = 2I. (3.14)
1 −1 1 −1 0 2
16 3 Hilbert Space
Therefore, this sum again splits into even and odd parts where
∞ ∞ ∞
(iϕ )n 2n (iϕ )2n 2n (iϕ )2n+1
∑ (σ1 + σ3 )n = I ∑ + (σ1 + σ3 ) ∑ (3.15)
n=0 n! n=0 (2n)! n=0 (2n + 1)!
∞
√ √
(i 2ϕ ) 2n σ1 + σ3 ∞ (i 2ϕ )2n+1
=I∑ + √ ∑
n=0 (2n)! 2 n=0 (2n + 1)!
√ σ + σ √
1 3
= I cos 2ϕ + i √ sin 2ϕ .
2
This can be written in usual matrix form, but we won’t do that here.
3.3 (a) To verify that the rotation matrix is unitary, we just multiply its Hermitian
conjugate with itself:

cos θ sin θ cos θ − sin θ
MM = †
(3.16)
− sin θ cos θ sin θ cos θ

cos2 θ + sin2 θ cos θ sin θ − cos θ sin θ
= = I,
− cos θ sin θ + cos θ sin θ cos2 θ + sin2 θ
and so M is indeed unitary.
(b) We are now asked to determine which of the Pauli matrices are exponenti-
ated to generate this rotation matrix. First, note that each element of this
rotation matrix is real, but the form of the exponent is iθ σ j , which is in
general complex. To ensure that exclusively real-valued matrix elements are
produced, the Pauli matrix must have only imaginary elements and there is
only one Pauli matrix for which that is true:

0 −i
σ2 = . (3.17)
i 0
Then, its exponentiation is
∞
(iθ )n n
eiθ σ2 = ∑ σ2 . (3.18)
n=0 n!
Note that the square of the Pauli matrix is

0 −i 0 −i 1 0
σ2 =
2
= = I. (3.19)
i 0 i 0 0 1
Then, the sum splits up into even and odd parts:
∞ ∞ ∞
(iθ )n n (iθ )2n (iθ )2n+1
∑ σ2 = I ∑ + σ2 ∑ (3.20)
n=0 n! n=0 (2n)! n=0 (2n + 1)!

cos θ sin θ
= I cos θ + iσ2 sin θ = .
− sin θ cos θ
3.4 (a) The Hermitian conjugate of the vector ⃗u is

⃗u† = e−iϕ1 cos u e−iϕ2 sin u . (3.21)
17 Exercises
Then, the inner product of ⃗u with itself is

eiϕ1 cos u
⃗u†⃗u = e−iϕ1 cos u e−iϕ2 sin u = cos2 u + sin2 u = 1 , (3.22)
eiϕ2 sin u
and so ⃗u is normalized. The same calculation follows for⃗v with appropriate

relabeling.
(b) Now, we would like to determine the unitary matrix U that maps ⃗u to ⃗v.
With the hint in the problem, we write
iθ iϕ
e 1 cos v a b e 1 cos u
⃗v = U⃗u → = (3.23)
eiθ2 sin v c d eiϕ2 sin u
iϕ
ae 1 cos u + beiϕ2 sin u
= .
ceiϕ1 cos u + deiϕ2 sin u
Let’s focus on the first entry. For this to equal the first entry of ⃗v, we must
have the exponential phase factor to reduce to eiθ1 . So, we can set
a = ei(θ1 −ϕ1 ) |a| , b = ei(θ1 −ϕ2 ) |b| . (3.24)
The exact same idea follows for the second entries, and so
c = ei(θ2 −ϕ1 ) |c| , d = ei(θ2 −ϕ2 ) |d| . (3.25)
Inserting these expressions into the matrix product above, we then find the
reduced linear equation that

cos v |a| cos u + |b| sin u
= . (3.26)
sin v |c| cos u + |d| sin u
This form might be recognizable as an angle addition formula. For example,

note that
cos v = cos(u + v) cos u + sin(u + v) sin u . (3.27)
Then, we can set
a = ei(θ1 −ϕ1 ) cos(u + v) , b = ei(θ1 −ϕ2 ) sin(u + v) . (3.28)
Note also that
sin v = sin(u + v) cos u − cos(u + v) sin u . (3.29)
Then,
c = ei(θ2 −ϕ1 ) sin(u + v) , d = −ei(θ2 −ϕ2 ) cos(u + v) . (3.30)
Putting it all together, the unitary matrix is

i(θ −ϕ )
e 1 1 cos(u + v) ei(θ1 −ϕ2 ) sin(u + v)
U= . (3.31)
ei(θ2 −ϕ1 ) sin(u + v) −ei(θ2 −ϕ2 ) cos(u + v)
18 3 Hilbert Space
(c) The determinant of this matrix is

i(θ −ϕ )
e 1 1 cos(u + v) ei(θ1 −ϕ2 ) sin(u + v)
det U = det (3.32)
ei(θ2 −ϕ1 ) sin(u + v) −ei(θ2 −ϕ2 ) cos(u + v)

= −ei(θ1 +θ2 −ϕ1 −ϕ2 ) cos2 (u + v) + sin2 (u + v) = −ei(θ1 +θ2 −ϕ1 −ϕ2 ) .
This can never equal 0, and so the matrix U is non-degenerate.

3.5 By definition of living on the Hilbert space, both kets are normalized:
⟨v|v⟩ = 1 , ⟨u|u⟩ = 1 . (3.33)
Thus, the simplest operator that maps |u⟩ to |v⟩ is
Uuv = |v⟩⟨u| . (3.34)
Then, we have that
|v⟩ = Uuv |u⟩ = |v⟩⟨u|u⟩ = |v⟩ , (3.35)
by the normalization of the kets. Note however that this operator is not unitary,
for
Uuv U†uv = |v⟩⟨u|u⟩⟨v| = |v⟩⟨v| ̸= I . (3.36)
However, the fix to make the operator unitary is pretty simple: we just keep adding
outer products of vectors that are mutually orthogonal and orthogonal to |u⟩ until
they form a complete basis of the Hilbert space. Except for the fact that this basis
includes |u⟩, it is otherwise very weakly constrained, so we won’t write down an
explicit form.
3.6 (a) The sum of the outer products of these vectors is
iϕ
1 e 1 sin θ
|v1 ⟩⟨v1 | + |v2 ⟩⟨v2 | = (1 0) + ϕ e−iϕ1 sin θ e−iϕ2 cos θ
0 e cos θ
i 2
(3.37)

1 0 sin θ2
ei(ϕ1 −ϕ2 ) sin θ cos θ
= + −i(ϕ −ϕ .
0 0 e 1 2 ) sin θ cos θ cos2 θ
This clearly does not equal the identity matrix and therefore does not satisfy
the completeness relation for general θ . However, it does equal the identity
matrix if θ = 0, for which the off-diagonal entries vanish, and so does not
depend on the phases ϕ1 , ϕ2 .
(b) Recall that the first three Legendre polynomials are
1
P0 (x) = √ , (3.38)
2
r
3
P1 (x) = x, (3.39)
2
r
5 2
P2 (x) = (3x − 1) . (3.40)
8
19 Exercises
Then, the outer product of Legendre polynomials is

1 3 5
P0 (x)P0 (y) + P1 (x)P1 (y) + P2 (x)P2 (y) = + xy + (3x2 − 1)(3y2 − 1) .
2 2 8
(3.41)
(c) Now, we want to integrate this “identity matrix” against an arbitrary quad-
ratic polynomial. What this will do is effectively take the matrix or dot
product of the polynomial (like a vector) with the identity matrix and so
it should return exactly the polynomial. Let’s see if this is true. To do this
we need the integrals:
Z 1
1 a
dx (ax2 + bx + c) = +c, (3.42)
−1 2 3
Z 1
3
dx (ax2 + bx + c) xy = by , (3.43)
−1 2
Z 1
5 a
dx (ax2 + bx + c) (3x2 − 1)(3y2 − 1) = − + ay2 . (3.44)
−1 8 3
Then, the sum of these terms is
Z 1
dx (ax2 + bx + c)I = ay2 + by + c = p(y) . (3.45)
−1
And so, indeed, the outer product of Legendre polynomials P0 (x)P0 (y) +
P1 (x)P1 (y) + P2 (x)P2 (y) acts as an identity matrix.
3.7 (a) We are asked to calculate the expectation value of the Hamiltonian Ĥ on the
time-dependent state. First, the time-dependent state is
E1 t E2 t
|ψ (t)⟩ = α1 e−i h̄ |1⟩ + α2 e−i h̄ |2⟩ . (3.46)
Then, the action of the Hamiltonian on this linear combination of energy
eigenstates is
E1 t E2 t
Ĥ|ψ (t)⟩ = α1 E1 e−i h̄ |1⟩ + α2 E2 e−i h̄ |2⟩ , (3.47)
because Ĥ|1⟩ = E1 |1⟩, for example. Then, the expectation value is, assuming
orthonormality of the energy eigenstates,
E1 t E2 t E1 t E2 t
⟨ψ (t)|Ĥ|ψ (t)⟩ = α1∗ ei h̄ ⟨1| + α2∗ ei h̄ ⟨2| α1 E1 e−i h̄ |1⟩ + α2 E2 e−i h̄ |2⟩
= |α1 |2 E1 + |α2 |2 E2 . (3.48)
This is independent of time.
(b) In complete generality, we can express Ô as a linear combination of outer
products:
Ô = a|1⟩⟨1| + b|1⟩⟨2| + c|2⟩⟨1| + d|2⟩⟨2| , (3.49)
for some complex numbers a, b, c, d. In this form, we can act it on |1⟩ to find:
Ô|1⟩ = a|1⟩ + c|2⟩ = |1⟩ − |2⟩ , (3.50)
20 3 Hilbert Space
using orthonormality and the constraint provided in the problem. Then, it

is clear that a = 1, c = −1. Instead, acting on |2⟩, we have
Ô|2⟩ = b|1⟩ + d|2⟩ = −|1⟩ + |2⟩ . (3.51)
We then see that b = −1, d = 1. It then follows that the operator is

1 −1
Ô = |1⟩⟨1| − |1⟩⟨2| − |2⟩⟨1| + |2⟩⟨2| = . (3.52)
−1 1
This is indeed a Hermitian operator as all elements are real and it is
symmetric.
(c) Now, we want to determine the expectation value of the operator Ô on the
time-dependent state. First, note that
E1 t E2 t
Ô|ψ (t)⟩ = Ô α1 e−i h̄ |1⟩ + α2 e−i h̄ |2⟩ (3.53)
E1 t E2 t
= α1 e−i h̄ (|1⟩ − |2⟩) + α2 e−i h̄ (−|1⟩ + |2⟩)
E1 t E2 t E2 t E1 t
= α1 e−i h̄ − α2 e−i h̄ |1⟩ + α2 e−i h̄ − α1 e−i h̄ |2⟩ .
It then follows that the expectation value is

E1 t E1 t E2 t E2 t E2 t E1 t
⟨ψ (t)|Ô|ψ (t)⟩ = α1∗ ei h̄ α1 e−i h̄ −α2 e−i h̄ +α2∗ ei h̄ α2 e−i h̄ −α1 e−i h̄
(E1 −E2 )t (E1 −E2 )t
= |α1 |2 + |α2 |2 − α1∗ α2 ei h̄ − α1 α2∗ e−i h̄ . (3.54)
By contrast to the expectation value of the Hamiltonian, the expectation

value of Ô does depend on time.
If the coefficients α1 , α2 are real, then the expectation value simplifies to
(E1 −E2 )t (E1 −E2 )t
⟨ψ (t)|Ô|ψ (t)⟩ = |α1 |2 + |α2 |2 − α1 α2 ei − α1 α2 e−i
h̄ h̄ (3.55)

(E1 − E2 )t
= |α1 | + |α2 | − 2α1 α2 cos
2 2
.
h̄
3.8 (a) To determine the electron neutrino after an elapsed time T , we just augment
its expression with the time-evolution phase factors:
E1 t E2 t
|νe (T )⟩ = cos θ e−i h̄ |ν1 ⟩ + sin θ e−i h̄ |ν2 ⟩ . (3.56)
(b) We can isolate the probability for measuring a muon neutrino by taking the
inner product with the time-evolved electron neutrino state. The probability
amplitude is
E1 t E2 t
⟨νµ |νe (T )⟩ = (− sin θ ⟨ν1 | + cos θ ⟨ν2 |) cos θ e−i h̄ |ν1 ⟩ + sin θ e−i h̄ |ν2 ⟩
(3.57)
E2 t E1 t
= sin θ cos θ e−i h̄ − e−i h̄
21 Exercises

(E2 +E1 )t (E −E )t (E −E )t
= e−i 2h̄ sin θ cos θ e i 1 2h̄ 2
−e−i 1 2h̄ 2
(E2 +E1 )t (E1 − E2 )t

= ie−i 2h̄ sin(2θ ) sin .
2h̄
Then, the probability for an electron neutrino to transform into a muon
neutrino after time T is
(E1 − E2 )t
Pµ = |⟨νµ |νe (T )⟩|2 = sin2 (2θ ) sin2 . (3.58)
2h̄
By unitarity of time evolution, the electron neutrino stays normalized for
all time, so the probability that the electron neutrino transforms into an
electron neutrino is just 1 minus this:
(E1 − E2 )t
Pe = 1 − Pµ = 1 − sin2 (2θ ) sin2 . (3.59)
2h̄
(c) This is called neutrino oscillation because the probability to observe neutri-
nos of different flavors oscillates with the time over which the initial neutrino
is allowed to evolve.
Axioms of Quantum Mechanics and Their
4
Consequences
Exercises
4.1 This question has numerous interpretations. First, the variance of a non-
Hermitian operator isn’t even well-defined. To be interpretable as a property of a
probability distribution, the variance should be a real number, but even the expec-
tation value of a non-Hermitian operator ⟨Ĉ⟩ is not real. So, in some sense this is
a non-starter because eigenvalues of non-Hermitian operators do not correspond
to physical observables.
However, a more useful way to interpret the uncertainty principle is right-to-
left, as a consequence of a non-trivial commutation relation. If two operators
Ĉ and D̂ have a non-zero commutation relation, this means that it is impossi-
ble for them to be simultaneously diagonalizable. So, the eigenstates of Ĉ and
D̂ must be different if their commutation relation is non-zero. Correspondingly,
if their eigenspace is distinct, then an eigenstate of one operator is necessarily a
non-trivial linear combination of the others eigenstates. So there is still an “uncer-
tainty” in the decomposition of any vector in terms of eigenstates of Ĉ and D̂. A
probabilistic interpretation is lost (or at the very least not obvious), but a lack of
commutation means that there is no basis in which both operators take a simple
form.
4.2 (a) Recall that the 3 × 3 derivative matrix we constructed earlier is
 1 
0 2∆x 0
D =  − 2∆x
1
0 1
2∆x
. (4.1)
0 − 2∆x
1
0
Then, because the momentum operator is related to the derivative operator

as P = −ih̄D, we find
 1 
0 2∆x 0
P = −ih̄  − 2∆x
1
0 1
2∆x
. (4.2)
0 − 2∆x
1
0
(b) The position operator X has eigenvalues of the positions on the grid and in
position space, those are its diagonal entries:
22
23 Exercises
 
x0 0 0
X= 0 x1 0 . (4.3)
0 0 x2
(c) Now, let’s calculate the commutator of the position and momentum opera-
tors on this grid. First, we have the product
  1 
x0 0 0 0 2∆x 0
XP = −ih̄  0 x1 0   − 2∆x
1
0 1
2∆x
 (4.4)
0 0 x2 0 − 2∆x
1
0
 x0 
0 2∆x 0
= −ih̄  − 2∆x
x1
0 x1 
2∆x .
x2
0 − 2∆x 0
The other order of the product is
 1  
0 2∆x 0 x0 0 0
PX = −ih̄  − 2∆x
1
0 1 
2∆x 0 x1 0  (4.5)
0 − 2∆x
1
0 0 0 x2
 x1 
0 2∆x 0
= −ih̄  − 2∆x
x0
0 x2 
2∆x .
x1
0 − 2∆x 0
Their commutator is therefore

 x0   x1 
0 2∆x 0 0 2∆x 0
[X, P] = −ih̄  − 2∆xx1
0 x1
2∆x
 + ih̄  − x0
2∆x 0 x2
2∆x

x2 x1
0 − 2∆x 0 0 − 2∆x 0
 x1 −x0 
0 2∆x 0
= ih̄  x1 −x0
2∆x 0 x2 −x1 
2∆x . (4.6)
x2 −x1
0 2∆x 0
Now, on the grid, note that x1 − x0 = x2 − x1 = ∆x, so this simplifies to
 1 
0 2 0
[X, P] = ih̄  1
2 0 1
2
. (4.7)
1
0 2 0
This is a bit strange from the point of view of the canonical commutation
relation, because we would expect that their commutator is [X, P] = ih̄I, pro-
portional to the identity matrix. This is close, but this can make more sense
when taking a limit of ∆x → 0.
24 4 Axioms of Quantum Mechanics and Their Consequences
(d) As ∆x → 0, these matrices and the corresponding commutator grow larger

and larger, and we would anticipate that it takes the form
 . . . 
..
. .. .. .. · · ·
 
 ··· 0 1 0 ··· 
 2 
[X, P] = ih̄ 
 ··· 2 0 2 ···  .
1 1  (4.8)
 ··· 0 1 0 ··· 
 2 
.. .. .. .. . .
. . . . .
As ∆x decreases, there are never any non-zero entries on the diagonal, but
immediately off the diagonal are factors of 1/2. Relative to the size of the
matrix, these factors of 1/2 get closer and closer to being on the diagonal.
Continuous-dimensional spaces like position space are a bit strange because
infinitesimals can be defined, and so as the grid spacing ∆x → 0, we expect
that these off-diagonal entries merge and sum into the identity matrix.
4.5 (a) For the exponential to make sense, the quantity in the exponent must be
unitless. Therefore, the units of the prefactor for the position operator must
have dimensions of inverse length:
"r #
2π c
= L−1 . (4.9)
h̄
Rearranging, the units of c are then
[c] = [h̄]L−2 = MT −1 , (4.10)
that is, mass per unit time. Therefore, c quantifies a flow or current of mass
of the system, in a similar way that electric current quantifies the flow of
electric charge per unit time.
(b) To calculate the commutation relation of the operators X̂ and P̂, we need to
use a test function f (x) to ensure that the manipulations we perform use the
linearity of the derivative. We then have
q q
i 2π c x 2π h̄ d
[X̂, P̂] f (x) = e h̄ , e ch̄ dx f (x) (4.11)
q q q q
i 2π c x 2π h̄ d 2π h̄ d i 2π c x
=e h̄ e f (x) − e ch̄ dx e h̄ f (x)
ch̄ dx
q r ! q q r !
i 2π c x 2π h̄ i 2πh̄ c x+ 2πc h̄ 2π h̄
=e h̄ f x+ −e f x+ .
c c
On the second line, we have expressed the momentum and position opera-
tors in position space, and then on the third line, note that the exponentiated
derivative operator is the translation operator. Further, note that the second
remaining exponential factor simplifies:
q q
2π c
q q
i h̄ x+ 2πc h̄ i 2π c x+2π i i 2π c x
e =e h̄ =e h̄ , (4.12)
25 Exercises
because e2π i = 1. Therefore, these operators actually commute:

q r ! q r !
i 2πh̄ c x 2π h̄ i 2πh̄ c x 2π h̄
[X̂, P̂] f (x) = e f x+ −e f x+ = 0 . (4.13)
c c
If two operators commute, this means that they can be simultaneously

diagonalized and have the same eigenstates.
(c) By the periodicity of the imaginary exponential, we have that, for example,
the two operators
r r
2π c 2π c
x̂ , x̂ + 2π (4.14)
h̄ h̄
produce the same operator X̂. The range of x̂ that produces unique values
for X̂ is then
r
2π h̄
0 ≤ x̂ < , (4.15)
c
where this inequality is understood as a relationship of eigenvalues. A simi-
lar relationship can be established for momentum p̂, where the unique range
of eigenvalues is
√
0 ≤ p̂ < 2π ch̄ . (4.16)
(d) The eigenvalue equation for the exponentiated momentum operator is
q r !
2π h̄ d 2 π h̄
P̂ fλ (x) = λ fλ (x) = e ch̄ dx fλ (x) = f x + . (4.17)
ch̄
The only way that a translated function can equal itself up to a multiplicative
factor is if that function is periodic, and so can be expressed as a complex
exponential. Let’s write
fλ (x) = eibx , (4.18)
for some real constant b. This must satisfy
q
ib x+ 2πc h̄
λ eibx = e , (4.19)
or that
q
ib 2π h̄
λ =e c . (4.20)
Through the property of periodicity of the imaginary exponential, we can
limit b in the range
r
2π c
0≤b< . (4.21)
h̄
That is, eigenvalues of the operator P̂ lie on the unit circle, as expected
because P̂ is a unitary operator.
(e) Because the commutator [X̂, P̂] = 0, the eigenstates of X̂ are identical to the
eigenstates of P̂.
(f) This may seem like we have created position and momentum operators that
commute, and therefore have somehow skirted the Heisenberg uncertainty
principle. However, as we observed in part (c), the fact that these exponen-
tiated position and momentum operators are unitary means that in each the
position or momentum is only defined as modulo 2π . Therefore, the uncer-
tainty has migrated from the non-zero commutator into an uncertainty in
the operators themselves. A given eigenvalue for X̂, for example, does not
define a unique position, and in fact is consistent with a countable infinity
of positions!
4.4 First, let’s write down the relevant Taylor expansions for the exponentiated
matrices, to third order. We have
Â2 Â3
eiÂ = I + iÂ − −i +··· , (4.22)
2 6
B̂2 B̂3
eiB̂ = I + iB̂ − −i +··· , (4.23)
2 6
(Â + B̂)2 (Â + B̂)3
e i(Â+B̂)
= I + i(Â + B̂) − −i +··· . (4.24)
2 6
Then, to third order in the matrices Â and B̂, the product of the exponentials is

Â2 Â3 B̂2 B̂3
eiÂ eiB̂ = I + iÂ − −i +··· I + iB̂ − −i +··· (4.25)
2 6 2 6
Â2 B̂2 Â3 B̂3
= I + i(Â + B̂) − − − ÂB̂ − i − i
2 2 6 6
ÂB̂2 B̂Â2
−i −i +··· .
2 2
Now, we can associate terms and complete squares and cubes. Note that
Â2 B̂2 Â2 B̂2 ÂB̂ B̂Â ÂB̂ B̂Â

− − − ÂB̂ = − − − − − + (4.26)
2 2 2 2 2 2 2 2
(Â + B̂)2 1
=− − [Â, B̂] .
2 2
Also, note that the expansion of the cubic binomial is
(Â + B̂)3 = Â3 + B̂3 + ÂB̂2 + B̂Â2 + ÂB̂Â + B̂ÂB̂ + Â2 B̂ + B̂2 Â , (4.27)
as always being careful with the order of multiplication.

With these results, the difference between the exponential matrices is
[Â, B̂]
eiÂ eiB̂ − ei(Â+B̂) = −
2
i
− 2ÂB̂2 + 2B̂Â2 − ÂB̂Â − B̂ÂB̂ − Â2 B̂ − B̂2 Â + · · · (4.28)
6
27 Exercises
The terms in the parentheses at cubic order can be re-associated into
2ÂB̂2 + 2B̂Â2 − ÂB̂Â − B̂ÂB̂ − Â2 B̂ − B̂2 Â

= [Â, B̂]B̂ − [Â, B̂]Â + [Â, B̂2 ] − [Â2 , B̂] . (4.29)
Thus, at least through cubic order in the Taylor expansion, all of the terms that
differ between the product of exponentiated matrices to the exponential of the
sum of the matrices can be expressed in terms of commutators of matrices Â and
B̂. Note that if [Â, B̂] = 0, then all of these residual differences vanish.
4.5 (a) If the operator Ô acts as
Ô|n⟩ = |n + 1⟩ , (4.30)
then it can be expressed in outer product form as

∞
Ô = ∑ |n + 1⟩⟨n| . (4.31)
n=−∞
Its Hermitian conjugate is then

∞
Ô† = ∑ |n⟩⟨n + 1| . (4.32)
n=−∞
The product of these two operators is then

∞ ∞ ∞ ∞
Ô† Ô = ∑ ∑ |m⟩⟨m + 1|n + 1⟩⟨n| = ∑ ∑ |m⟩⟨n| δmn
m=−∞ n=−∞ m=−∞ n=−∞
∞
= ∑ |n⟩⟨n| . (4.33)
n=−∞
By the assumed completeness of the winding number states, this final outer
product is the identity matrix. Hence, Ô† Ô = I, which indeed proves that Ô
is unitary.
(b) As a unitary operator, the eigenvalue equation for Ô is
Ô|ψ ⟩ = eiθ |ψ ⟩ . (4.34)
We can express the eigenstate as a linear combination of winding states:

∞
|ψ ⟩ = ∑ βn |n⟩ , (4.35)
n=−∞
and on them the hopping operator acts as

∞ ∞ ∞
Ô|ψ ⟩ = ∑ βn Ô|n⟩ = ∑ βn |n + 1⟩ = ∑ βn−1 |n⟩ . (4.36)
n=−∞ n=−∞ n=−∞
Then, the eigenvalue equation becomes

∞ ∞
∑ βn−1 |n⟩ = ∑ βn eiθ |n⟩ . (4.37)
n=−∞ n=−∞
For this to be an equality, by the orthogonality of the winding states, we

must enforce that each coefficient is equal:
βn eiθ = βn−1 . (4.38)
Up to an overall normalization, this is satisfied by the coefficients
βn = e−inθ . (4.39)
Therefore, the eigenstate is

∞
|ψ ⟩ = ∑ e−inθ |n⟩ . (4.40)
n=−∞
(c) The winding states are eigenstates of the Hamiltonian, so we can express the
Hamiltonian in outer product form as
∞
Ĥ = ∑ |n|E0 |n⟩⟨n| . (4.41)
−∞
With this form, we can just calculate the commutator directly. We have
∞ ∞ ∞ ∞
Ĥ Ô = ∑ ∑ |m|E0 |m⟩⟨m|n + 1⟩⟨n| = ∑ ∑ |m|E0 |m⟩⟨n| δm,n+1
m=−∞ n=−∞ m=−∞ n=−∞
∞
= ∑ |n + 1|E0 |n + 1⟩⟨n| . (4.42)
n=−∞
The opposite order of the product is

∞ ∞ ∞ ∞
ÔĤ = ∑ ∑ |n|E0 |m + 1⟩⟨m|n⟩⟨n| = ∑ ∑ |n|E0 |m + 1⟩⟨n| δmn
m=−∞ n=−∞ m=−∞ n=−∞
∞
= ∑ |n|E0 |n + 1⟩⟨n| . (4.43)
n=−∞
The difference between these operator products is

∞
[Ĥ, Ô] = ∑ (|n + 1| − |n|) E0 |n + 1⟩⟨n| (4.44)
n=−∞
∞ −1
= ∑ (n + 1 − n) E0 |n + 1⟩⟨n| + ∑ (−n − 1 + n) E0 |n + 1⟩⟨n|
n=0 n=−∞
∞ −1
= ∑ E0 |n + 1⟩⟨n| − ∑ E0 |n + 1⟩⟨n| .
n=0 n=−∞
Note that this is clearly non-zero.

(d) Let’s just act the exponentiated Hamiltonian directly on the eigenstate of
the hopping operator:
29 Exercises
∞ ∞ |n|E0 t
e−i h̄ |ψ ⟩ = ∑ e−inθ e−i h̄ |n⟩ = ∑ e−inθ e−i
Ĥt Ĥt
h̄ |n⟩ (4.45)
n=−∞ n=−∞
∞
E t
−in θ +sgn(n) h̄0
= ∑ e |n⟩ ,
n=−∞
where the sign function is

n
sgn(n) = . (4.46)
|n|
4.6 (a) First, let’s think about the Cauchy–Schwarz inequality again. The more
mundane version of it is just a consequence of the properties of the vector
dot product. For two real-valued vectors⃗u,⃗v, the only way that⃗u2⃗v 2 = (⃗u·⃗v)2
is if the vectors are proportional: ⃗v = α⃗u, for some constant α . There-
fore, if the Cauchy–Schwarz inequality is saturated, the two vectors are
proportional:
(x̂ − ⟨x̂⟩)|ψ ⟩ = α ( p̂ − ⟨ p̂⟩)|ψ ⟩ . (4.47)
Using this, we can then further use the saturated Heisenberg uncertainty
principle. The saturated uncertainty principle enforces the relationship that
h̄2
σx2 σ p2 = . (4.48)
4
Next, the variance in position, say, is just the square of the vector in the
Cauchy–Schwarz inequality:
σx2 = ⟨ψ |(x̂ − ⟨x̂⟩)2 |ψ ⟩ = |α |2 ⟨ψ |( p̂ − ⟨ p̂⟩)2 |ψ ⟩ = |α |2 σ p2 . (4.49)
With these two relationships, we can solve for both α in terms of σx . We

have
2σx2
|α | = . (4.50)
h̄
Finally, using the fact that [x̂, p̂] = ih̄, we can replace the factor of i in α
that was divided in the uncertainty principle (see Eq. 4.99 on page 74 of the
textbook):
2iσx2
α =− . (4.51)
h̄
Then, the vectors that saturate the Cauchy–Schwarz inequality satisfy
2iσx2
(x̂ − ⟨x̂⟩)|ψ ⟩ = − ( p̂ − ⟨ p̂⟩)|ψ ⟩ . (4.52)
h̄
(b) In position space, the relationship between the action of position and
momentum on the saturated state is

2iσx2 d
(x − ⟨x̂⟩)ψ (x) = − −ih̄ − ⟨ p̂⟩ ψ (x) . (4.53)
h̄ dx
This can be rearranged into a more familiar organization where

d ⟨ p̂⟩ x − ⟨x̂⟩
−i + ψ (x) = 0 . (4.54)
dx h̄ 2σx2
This is a linear, first-order, homogeneous differential equation whose solu-
tion takes an exponential form. We have (ignoring overall normalization)
(x−⟨x̂⟩)2
⟨ p̂⟩x −
ψ (x) = ei h̄ e 4σx2 . (4.55)
(c) Now, we are asked to Fourier transform this solution to momentum space.
One can put this expression into the usual Fourier integral, but we’ll do
something different here. Using the fact that [x̂, p̂] = ih̄, we can instead
express the position operator as a momentum derivative:
d
x̂ = ih̄ , (4.56)
d p̂
and so the relationship between the vectors can be expressed in momentum
space as

d 2iσx2
ih̄ − ⟨x̂⟩ ψ (p) = − (p − ⟨ p̂⟩) ψ (p) . (4.57)
dp h̄
Because we are in momentum space, it makes more sense to use the
momentum variance, so we replace
h̄2
σx2 = , (4.58)
4σ p2
finding

d ih̄
ih̄ − ⟨x̂⟩ ψ (p) = − 2 (p − ⟨ p̂⟩) ψ (p) . (4.59)
dp 2σ p
Again, putting everything to one side, we have
!
d ⟨x̂⟩ p − ⟨ p̂⟩
+i + ψ (p) = 0 . (4.60)
dp h̄ 2σ p2
This is of essentially an identical form to the position space equation, so its

solution is
(p−⟨ p̂⟩)2
⟨x̂⟩p −
ψ (p) = e−i h̄ e 4σ p2
. (4.61)
4.7 (a) If these two equations of motion are supposed to be equal, then we must
have

dV (x̂) dV (⟨x̂⟩)
= . (4.62)
d x̂ d⟨x̂⟩
This is a particular instantiation of the relationship
⟨ f (x̂)⟩ = f (⟨x̂⟩) , (4.63)
31 Exercises
for some function f . This can only be satisfied for the expectation value on
a general state if the function f is linear:
f (x) = a + bx , (4.64)
for some constants a, b. This can be argued for a large class of functions by
assuming it has a Taylor expansion about x = 0 and noting that it is not true
in general that
⟨x̂n ⟩ = ⟨x⟩n , (4.65)
for n > 1. If this was true, then for example, the variance would always van-
ish. Thus, if the derivative of the potential must be linear, the potential must
be quadratic, and so the potential takes the form
V (x̂) = ax̂2 + bx̂ + c . (4.66)
By an appropriate choice of coordinates and a zero energy point, this can

always be reduced to the simple harmonic oscillator, which we will see in a
couple of chapters.
(b) For this power-law potential, the two derivatives are

dV (⟨x̂⟩) dV (x̂)
= 2nk⟨x̂⟩2n−1
, = 2nk⟨x̂2n−1 ⟩ . (4.67)
d⟨x̂⟩ d x̂
Note that ⟨x̂2n−1 ⟩ weights larger values of position more than does ⟨x̂⟩2n−1
and so we expect that

dV (x̂) dV (⟨x̂⟩)
≥ . (4.68)
d x̂ d⟨x̂⟩
(c) Now, we flip the interpretation and are asked to find the state |ψ ⟩ on which
⟨x̂2n−1 ⟩ = ⟨x̂⟩2n−1 . This can only be true if there is a unique value of x̂ for
which the state has a non-zero value; namely it is a position eigenstate. That
is, |ψ ⟩ = |x⟩, which is also not in the Hilbert space. Therefore, for any state
on the Hilbert space, it is actually impossible for ⟨x̂2n−1 ⟩ = ⟨x̂⟩2n−1 .
4.8 (a) The integral of the exponential probability distribution is
Z ∞
N
1=N dx e−λ x = . (4.69)
0 λ
Therefore, the normalization constant N = λ .
(b) The expectation value of x can be calculated through explicit integration,
but we will do another approach here. Let’s instead take the derivative of the
normalization of the probability distribution with respect to the parameter
λ . We have
Z ∞ Z ∞ Z ∞
d d
1=0= λ dx e−λ x = dx e−λ x − λ dx x e−λ x . (4.70)
dλ dλ 0 0 0
Note that the first integral is 1/λ and the second integral is exactly the
expectation value ⟨x⟩. Therefore, we have that
1
0= − ⟨x⟩ , (4.71)
λ
or that
1
⟨x⟩ = . (4.72)
λ
(c) The second moment can be calculated in exactly the same way, just through
taking the second derivative of the normalization integral. We have
Z ∞ Z ∞ Z ∞
d2 d2 d d
1=0= λ dx e−λ x = dx e−λ x − λ dx x e−λ x
dλ 2 dλ 2 0 dλ 0 dλ 0
Z ∞ Z ∞ Z ∞
=− dx x e−λ x − dx x e−λ x + λ dx x2 e−λ x
0 0 0
⟨x⟩
= −2 + ⟨x2 ⟩ . (4.73)
λ
Therefore, we find that
2
⟨x2 ⟩ = . (4.74)
λ2
The variance of the exponential distribution is then
1
σ 2 = ⟨x2 ⟩ − ⟨x⟩2 = , (4.75)
λ2
and then the standard deviation is
1
σ= . (4.76)
λ
The area of the distribution within one standard deviation of the expectation
value is
Z 2
λ
λ dx e−λ x = −e−2 + 1 ≃ 0.864665 . (4.77)
0
This is indeed greater than 1/2 and so most of the distribution is within 1
standard deviation.
(d) Now, let’s calculate this generalized spread about the mean. We have
Z ∞
⟨(x − ⟨x⟩)n ⟩ = λ dx (x − ⟨x⟩)n e−λ x . (4.78)
0
Let’s now make the change of variables y = x − ⟨x⟩, so that the integral
becomes
Z ∞ Z ∞
⟨(x − ⟨x⟩)n ⟩ = λ dy yn e−λ (y+⟨x⟩) = λ e−1 dy yn e−λ y . (4.79)
−⟨x⟩ − λ1
33 Exercises
Now, using integration by parts, we have

Z ∞ ∞ Z ∞
n −λ y yn n n−1 −λ y
dy y e = − e−λ y + dy y e (4.80)
− λ1 λ −1/λ −1/λ λ
Z ∞
n
= (−1)n λ −n−1 e + dy yn−1 e−λ y .
λ − λ1
That is, the nth moment is

Z
n −1 ∞
⟨(x − ⟨x⟩)n ⟩ = (−λ )−n +
λe dy yn−1 e−λ y (4.81)
λ − λ1
n
= (−λ )−n + ⟨(x − ⟨x⟩)n−1 ⟩ .
λ
This defines a recursion relation that can be solved for any n knowing the
boundary condition. Note that if n = 1, we have
⟨(x − ⟨x⟩)1 ⟩ = 0 . (4.82)
We can verify that this recursion relation is correct, or at least consistent,
for n = 2. Then,
1
⟨(x − ⟨x⟩)2 ⟩ = σ 2 = , (4.83)
λ2
and the recursion relation would predict
2 1
⟨(x − ⟨x⟩)2 ⟩ = (−λ )−2 + ⟨(x − ⟨x⟩)1 ⟩ = 2 , (4.84)
λ λ
exactly as expected.
(e) Finally, we are asked to calculate the median. The median is the point xmed
where half of the integral lies to the left (and therefore half lies to the right):
Z x
1 med
=λ dx e−λ x = 1 − e−λ xmed . (4.85)
2 0
We can then solve for xmed as
log 2
xmed =
. (4.86)
λ
Note that log 2 ≃ 0.693147, and so the median lies at a smaller value than
the mean.
Quantum Mechanical Example: The Infinite
5
Square Well
Exercises
5.1 (a) We can write the state |ψ ⟩ as

|ψ1 ⟩ + 3|ψ2 ⟩
|ψ ⟩ = , (5.1)
α
for some normalization constant α . The absolute square of this state is
⟨ψ1 | + 3⟨ψ2 | |ψ1 ⟩ + 3|ψ2 ⟩ ⟨ψ1 |ψ1 ⟩ + 9⟨ψ2 |ψ2 ⟩ 10
1 = ⟨ψ |ψ ⟩ = = = ,
α∗ α |α |2 |α |2
(5.2)
using the orthonormality of
√the energy eigenstates of the infinite square well.
Then, we can choose α = 10, and so the normalized state is
|ψ1 ⟩ + 3|ψ2 ⟩
|ψ ⟩ = √ . (5.3)
10
(b) In position basis, the state takes the form
r
1 2 πx 2π x
ψ (x) = √ sin + 3 sin (5.4)
10 a a a

1 πx 2π x
=√ sin + 3 sin .
5a a a
The expectation value of the position on this state is then
Z a
⟨ψ |x̂|ψ ⟩ = dx ψ ∗ (x)x ψ (x) (5.5)
0
Z a
1 πx 2π x πx 2π x
= dx sin + 3 sin x sin + 3 sin
5a 0 a a a a

1 16
=a − ≃ 0.3919a .
2 15π 2
(c) The expectation value of the Hamiltonian is very easy to calculate on this
state because it is expressed as a linear combination of energy eigenstates.
We have
(⟨ψ1 | + 3⟨ψ2 |) Ĥ (|ψ1 ⟩ + 3|ψ2 ⟩) E1 + 9E2 37 π 2 h̄2
⟨ψ |Ĥ|ψ ⟩ = = = . (5.6)
10 10 10 2ma2
34
35 Exercises
(d) To include time dependence, we just need to augment the wavefunction with
exponential phase factors with the appropriate energy eigenvalues. We have
1 h −i E1 t E2 t i
ψ (x,t) = √ e h̄ ψ1 (x) + 3e−i h̄ ψ2 (x) . (5.7)
10
The expectation value of the energy or the Hamiltonian is identical when
time dependence is included again because the wavefunction is expressed in
terms of energy eigenstates. However, the expectation value of the position
will change in time. To evaluate this, let’s first just look at the wavefunction
times its complex conjugate:
1 h i E1 t ∗ E2 t i
ψ ∗ (x,t)ψ (x,t) = √ e h̄ ψ1 (x) + 3ei h̄ ψ2∗ (x) (5.8)
10
1 h −i E1 t E2 t i
×√ e h̄ ψ1 (x) + 3e−i h̄ ψ2 (x)
10

1 (E2 −E1 )t
= |ψ1 (x)|2 + 9|ψ2 (x)|2 + 3e−i h̄ ψ1∗ (x)ψ2 (x)
10

(E2 −E1 )t
+3ei h̄ ψ1 (x)ψ2∗ (x) .
We have chosen the energy eigenstate wavefunctions to be real-valued func-

tions, and so, for example, ψ1∗ (x) = ψ1 (x), and so the square simplifies
to

1 (E2 − E1 )t
ψ ∗ (x,t)ψ (x,t) = ψ1 (x)2 + 9ψ2 (x)2 + 6 cos ψ1 (x)ψ2 (x)
10 h̄

3 (E2 − E 1 )t
= ψ (x)2 − 1 − cos ψ1 (x)ψ2 (x) . (5.9)
5 h̄
That is, the expectation value of x is given temporal dependence by the
interference of the energy eigenstates. Note that we already calculated the
expectation value on the time-independent state ψ (x), so we will just focus
on the modification here. That is, we write the time-dependent expectation
value as
⟨x̂(t)⟩ = ⟨x̂⟩ + ∆⟨x̂(t)⟩ , (5.10)
where
Z a
3 (E2 − E1 )t
∆⟨x̂(t)⟩ = − 1 − cos dx x ψ1 (x)ψ2 (x) (5.11)
5 h̄ 0

16a (E2 − E1 )t
= 1 − cos .
15π 2 h̄
Thus, the time-dependent expectation value of the position on this state is

a 16a (E2 − E1 )t
⟨x̂(t)⟩ = − cos . (5.12)
2 15π 2 h̄
36 5 Quantum Mechanical Example: The Infinite Square Well
(5.2) We will first verify that this state is normalized. Note that
px px
∗ e−i h̄ ei h̄ 1
ψ (x)ψ (x) = √ √ = , (5.13)
a a a
independent of position x. Then, the nth moment of this distribution on the
infinite square well is
Z a Z a
1 an
⟨x̂n ⟩ = dx ψ ∗ (x) xn ψ (x) = dx xn = . (5.14)
0 a 0 n+1
The normalization corresponds to n = 0, for which we do indeed find the value
1, so this is unit normalized. The variance is the difference between n = 2 and the
square of n = 1 moments:

2 1 1 a2
σx = ⟨x̂ ⟩ − ⟨x̂⟩ = a
2 2 2
− 2 = . (5.15)
3 2 12
On the other hand, for the variance of momentum on this state, note that this
state is an eigenstate of the momentum operator:
px px
d ei h̄ ei h̄
p̂|ψ ⟩ = p|ψ ⟩ ⇒ −ih̄ √ =p√ . (5.16)
dx a a
As an eigenstate, the variance necessarily vanishes: σ p2 = 0. This would seem to

imply that the Heisenberg uncertainty principle is violated, because
h̄2
σx2 σ p2 = 0 < . (5.17)
4
However, this state is actually not in the Hilbert space of the infinite square
well. The reason for this is the following. In earlier chapters, we showed that the
momentum operator is Hermitian only if all states in the Hilbert space vanish
at the boundaries of the physical space. This momentum eigenstate is non-zero
at the phase space boundaries of the infinite square well, and therefore violates
this property. That is, if this state were allowed to be in the Hilbert space, then
momentum would not be Hermitian and therefore momentum would also not
be observable! Our derivation of the Heisenberg uncertainty principle required
that the operators were Hermitian, so this result does not contradict what we had
proved earlier.
5.3 (a) First, the energy eigenvalues are just
n2 π 2 h̄2
En = , (5.18)
2mL2
where L is the width of the well. Now, L = π , and so the energy eigenvalues
are
n2 h̄2
En = . (5.19)
2m
37 Exercises
The energy eigenstates will be some linear combination of imaginary expo-

nentials that vanish at both boundaries and have opposite momentum:
pn x pn x
ψn (x) = α ei h̄ + β e−i h̄ , (5.20)
for some constants α , β and momentum eigenvalue pn > 0. If this vanishes

at x = −π /2, then we must enforce
π pn π pn
ψn (−π /2) = 0 = α e−i 2h̄ + β ei 2h̄ . (5.21)
That is,
π pn
β = −α e−i h̄ . (5.22)
Then, the energy eigenstate wavefunction takes the form

pn x pn (x+π )

ψn (x) = α ei h̄ − e−i h̄ . (5.23)
Next, if the wavefunction vanishes at the upper endpoint, where x = π /2, we

must enforce
π pn 3π pn
π pn
2π pn

ψn (π /2) = 0 = α ei 2h̄ − e−i 2h̄ = α ei 2h̄ 1 − e−i h̄ . (5.24)
Thus, the momentum eigenvalues are
pn = nh̄ , (5.25)
for n = 1, 2, 3, . . . . Then, the energy eigenstate wavefunction is

ψn (x) = α einx − e−in(x+π ) = α einx − (−1)n e−inx . (5.26)
Finally, we fix α by demanding that this wavefunction is normalized on the

well:
Z π /2
1= dx ψn∗ (x)ψn (x) (5.27)
−π /2
Z π /2
= |α |2 dx e−inx − (−1)n einx einx − (−1)n e−inx
−π /2
Z π /2
= |α |2 dx 2 − (−1)n e2inx − (−1)n e−2inx = 2π |α |2 .
−π /2
Then, the normalization constant can be chosen to be

r
1
α= , (5.28)
2π
and the normalized energy eigenstate wavefunction is
1
ψn (x) = √ einx − (−1)n e−inx . (5.29)
2π
(b) To determine the time dependence of this initial wavefunction, we expand

in energy eigenstates:
|ψ ⟩ = ∑ βn |ψn ⟩ . (5.30)
n=1
Taking the inner product with the bra ⟨ψm | isolates the coefficient βm :
∞
⟨ψm |ψ ⟩ = ∑ βn ⟨ψm |ψn ⟩ = βm . (5.31)
n=1
So, we just need to calculate

Z r Z π /4
π /4 2 1
βn = ⟨ψn |ψ ⟩ = dx ψn (x) = dx einx − (−1)n e−inx .
−π /4 π π −π /4
(5.32)
This vanishes if n is even. For odd n, we have
Z π /4
1 4 nπ
βn = dx einx + e−inx = sin . (5.33)
π −π /4 πn 4
Thus, the initial wavefunction can be expressed in the basis of energy
eigenstates as
r (2n−1)π
4 2 ∞ sin 4
ψ (x) = ∑
π π n=1 2n − 1
cos ((2n − 1)x) . (5.34)
To determine the wavefunction at a later time, we simply multiply each term

by the exponential energy phase factor, where
En t n2 h̄t
e−i h̄ = e−i 2m . (5.35)
Thus, the wavefunction at a later time t is
r (2n−1)2 h̄t
4 2 ∞ e−i 2m (2n − 1)π
ψ (x,t) = ∑ 2n − 1 sin 4 cos ((2n − 1)x) .
π π n=1
(5.36)
(c) The inner product ⟨χ |ψ ⟩ is

√ (2n−1)π Z π /2
4 2 ∞ (2n−1)2 h̄t sin
⟨χ |ψ ⟩ = 2 ∑ e−i 2m 4
dx cos ((2n − 1)x) (5.37)
π n=1 2n − 1 −π /2
√ (2n−1)π
8 2 ∞ (2n−1)2 h̄t sin
= 2 ∑ (−1)n e−i 2m 4
.
π n=1 (2n − 1)2
(d) We can then take the derivative and set t = 0 and then find
√ (2n−1)π
d⟨χ |ψ ⟩ 8 2 ∞ (2n − 1)2 h̄ sin 4
= −i 2 ∑ (−1)n (5.38)
dt t=0 π n=1 2m (2n − 1)2
√
4 2 h̄ ∞ (2n − 1)π
= −i 2
π m n=1∑ (−1)n sin
4
39 Exercises
√
4 2 h̄ 1 1 1 1
= −i 2 −√ + √ + √ − √ +··· .
π m 2 2 2 2
This sum actually vanishes. The sine factor just oscillates between + √12 and
− √12 , and so the sum keeps canceling itself at higher terms in the series.
From the Schrödinger equation, the time derivative is determined by the
Hamiltonian, which, for the infinite square well, is just the squared momen-
tum operator:
p̂2 d
Ĥ = so ih̄ |ψ ⟩ = Ĥ|ψ ⟩ . (5.39)
2m dt
The momentum operator is a spatial derivative, and at t = 0, the initial wave-
function is piecewise constant over the well. This is then projected onto a
uniform wavefunction on the well. The second derivatives are 0 almost eve-
rywhere, except
p at the points where the initial wavefunction changes value
from 0 to 2/π , but the values of the spatial second derivatives are oppo-
site of one another. (Roughly, about the point x = −π /4, the wavefunction
is “concave-up”, and around x = π /4, it is “concave-down.”) Once projected
onto the uniform wavefunction, these second derivatives cancel, rendering
the time-derivative at t = 0 0.
(e) Let’s first consider the expectation value of momentum on this state. Recall
from Ehrenfest’s theorem that the time dependence of the expectation values
are
d⟨ p̂⟩ i
= ⟨[Ĥ, p̂]⟩ . (5.40)
dt h̄
Because the Hamiltonian is purely dependent on momentum, Ĥ = p̂2 /(2m),

the commutator vanishes and ⟨ p̂⟩ is therefore constant in time. We had
already argued that the expectation value of momentum at t = 0 was 0, and
therefore this is true for all time: ⟨ p̂⟩ = 0.
For the expectation value of position, note that the expectation value of the
commutator is
d p̂2 h̄
⟨[Ĥ, x̂]⟩ = ih̄⟨ψ | |ψ ⟩ = i ⟨ p̂⟩ = 0 , (5.41)
d p̂ 2m m
where we used the expression of the position operator as a momentum deriv-

ative and recall that the expectation value of momentum was 0 on this state.
Then, the expectation value of position is also constant in time, d⟨x̂⟩/dt = 0,
and initially the wavefunction was symmetric about x = 0. Therefore, for all
time, ⟨x̂⟩ is 0.
5.4 (a) In position space, the momentum operator is of course
d
p̂ = −ih̄ . (5.42)
dx
Now, with position x = aϕ on the ring, the momentum operator is simply

h̄ d
p̂ = −i , (5.43)
a dϕ
in angle space.
To establish the Hermiticity of this representation of momentum, we cal-
culate its matrix element ( p̂)i j in angle space. For some orthonormal basis
{ψi }i on ϕ ∈ [0, 2π ), we have
Z
h̄ 2π d
( p̂)i j = −i d ϕ ψi∗ (ϕ ) ψ j (ϕ ) (5.44)
a 0 dϕ
Z 2π
h̄ ∗ ∗ d ∗
= −i ψi (0)ψ j (0) − ψi (2π )ψ j (2π ) − d ϕ ψ j (ϕ ) ψi (ϕ )
a 0 dϕ
Z
h̄ 2π d ∗
=i d ϕ ψ j (ϕ ) ψi (ϕ ) = ( p̂)∗ji .
a 0 dϕ
On the second line, we used integration by parts, and by the periodicity of
the states on the ring, note that
ψi∗ (0)ψ j (0) = ψi∗ (2π )ψ j (2π ) . (5.45)
Therefore, this momentum operator is indeed Hermitian on the Hilbert
space of the states on the ring.
(b) To find the energy eigenstates on the ring, we first note that energy eigen-
states are momentum eigenstates because there is no position-dependence
of the potential. Therefore, we can write the eigenstate wavefunctions as
apn ϕ
ψn (ϕ ) = α ei h̄ , (5.46)
for some momentum pn and normalization constant α . By periodicity,
ψn (ϕ + 2π ) = ψn (ϕ ), and so we must enforce that
apn (ϕ +2π ) apn ϕ
ei h̄ = ei h̄ , (5.47)
or that 2π ah̄ pn
= 2π n, for some integer n. Therefore, the allowed momentum
eigenvalues are
nh̄
pn = . (5.48)
a
The eigenenergies are then
p2n n2 h̄2
En = = . (5.49)
2m 2ma2
Unlike for the infinite square well, the state with n = 0 is allowed on the ring
because it is normalizable. Thus the minimal energy of the particle on the
ring is actually 0, for which the eigenstate wavefunction is just a constant.
(c) Because the ring is of finite size, σϕ is always finite. However, on the ground
state where n = 0, the momentum is clearly 0 and so σ p = 0, which would
seem to invalidate the Heisenberg uncertainty principle. However, recall
41 Exercises
what the generalized uncertainty principle actually stated. In this case, we

have
⟨[ϕ̂ , p̂]⟩
σϕ σ p ≥ . (5.50)
2i
We have established that on the ground state p̂|ψ0 ⟩ = 0 and so
⟨ψ0 |[ϕ̂ , p̂]|ψ0 ⟩ = 0. Thus, this is entirely consistent with the uncertainty prin-
ciple on the ring, because a state of 0 energy is on the Hilbert space. So, the
uncertainty principle isn’t so useful now.
5.5 (a) Recall that the energy eigenvalues of the infinite square well are
n2 π 2 h̄2
En = . (5.51)
2ma2
The Schwarzschild radius of an energy eigenstate is then
2GN 2GN n2 π 2 h̄2
Rs = 4
En = 4 . (5.52)
c c 2ma2
If this equals the width of the well a, then the level n of the eigenstate is
ma3 c4
n2 = . (5.53)
π 2 GN h̄2
(b) Rearranging the expression for the Schwarzschild radius, the energy when
it is the size of the well is
ac4
E= . (5.54)
2GN
In SI units, c = 3 × 108 m/s, GN = 6.67 × 10−11 m3 kg−1 s−2 , and we have
said that the size of the well is a = 10−15 m. Then, this energy is
E ≃ 6 × 1028 J , (5.55)
which is enormous! Everyday energies are maybe hundreds of Joules, so this
is orders and orders of magnitude greater.
(c) With the mass of the pion given as mπ = 2.4 × 10−28 kg, the energy level n
at which the infinite square well becomes a black hole is (using the result of
part (a))
n2 ≃ 3 × 1038 , (5.56)
or that n ≃ 1.7 × 1019 . By contrast, the energy at which the pion is traveling
at the speed of light would be equal to setting its kinetic energy equal to its
bound state energy:
1 n2 π 2 h̄2
mπ c2 = , (5.57)
2 2mπ a2
or that
mπ ca
n= ≃ 0.23 . (5.58)
π h̄
So, even the ground state of this infinite square well is relativistic. One thing
to note is that clearly if the pion is traveling near the speed of light, its
kinetic energy is not as simple as mπ v 2 /2. However, setting v = c estab-
lishes an energy at which the pion is definitely relativistic, and at which the
non-relativistic expression for the energy is no longer applicable.
5.6 (a) The matrix elements of the momentum operator are
h̄ mn
( p̂)mn = −2i 1 − (−1)m+n . (5.59)
a m −n
2 2
The matrix elements of the squared momentum operator are found by

matrix multiplication of the momentum operator with itself. We have
∞
( p̂2 )mn = ∑ ( p̂)ml ( p̂)ln (5.60)
l=1
4h̄2 ∞ ml ln
=− ∑
a2 l=1 m2 − l 2
1 − (−1) m+l
l 2 − n2
1 − (−1)l+n

4h̄2 mn ∞ l 2 1 − (−1)m+l 1 − (−1)l+n
=− 2 ∑ .
a l=1 (m2 − l 2 )(l 2 − n2 )
(b) Note that terms in the sum vanish if either (−1)m+l = 1 or (−1)l+n = 1. So,
to simply ensure that every term in the sum vanishes, we just need to force
one of these relationships for every value of l. This can be accomplished if
m and n differ by an odd number, call it k. If n = m + k, then either m + l is
even and n + l is odd, or vice-versa, for every value of l. Thus,
( p̂2 )m(m+k) = 0 , (5.61)
if k is odd.
If instead n = m + k and k is even, the sum can be expressed as
∞ l 2 1 − (−1)m+l
∞ l 2 1 − (−1)m+l

1 − (−1)l+m+k 1 − (−1)m+l
∑ (m2 − l 2 )(l 2 − (m + k)2 )
=∑
(m2 − l 2 )(l 2 − (m + k)2 )
l=1 l=1
∞ 2l 2 1 − (−1)m+l

=∑ 2 2 2 . (5.62)
l=1 (m − l )(l − n )
2
This follows by distributing the product and noting that (−1)2(m+l) = 1. Fur-
ther, note that the value of 1 − (−1)m+l is either 0 or 2, and so every non-zero
term in the sum has the same sign, at least from this factor.
Now, let’s partial fraction the denominators. Note that

1 1 1 1
= − , (5.63)
m2 − l 2 2l m − l m + l
and similar for the other denominator factor. Then, the sum becomes
1
∞ 2l 2 1−(−1)m+l
1 ∞ 1 1 1
∑ 2 2 2 2 2∑ = 1−(−1) m+l
− − .
l=1 (m −l )(l −n ) l=1 m−l m + l n+l n − l
(5.64)
43 Exercises
We can keep partial fractioning the product denominators when distributed.

We have

1 1 1 1 1
= + , (5.65)
m−l n+l m+n m−l n+l

1 1 1 1 1
= − , (5.66)
m−l n−l n−m m−l n−l

1 1 1 1 1
= − , (5.67)
m+l n+l n−m m+l n+l

1 1 1 1 1
= + . (5.68)
m+l n−l m+n m+l n−l
Using this, the product of denominators becomes

1 1 1 1 2m 1 1
− − = 2 − (5.69)
m−l m+l n+l n−l m − n2 l + m l − m

2n 1 1
+ 2 − .
n − m2 l + n l − n
Thus the expression of the squared momentum as an infinite sum breaks
into two telescoping series:
∞ 2l 2 1 − (−1)m+l

∑ 2 2 2 2 (5.70)
l=1 (m − l )(l − n )
1 ∞ m m n n

= 2 ∑
m − n2 l=1
1 − (−1)m+l
− − +
l +m l −m l +n l −n
.
The series that we need to evaluate is then of the form

∞
1 1
∑ −
l −m
, (5.71)
l=1 l + m
where the values of l that are summed over are either just odd or just even,
depending on m. If m is odd, then l is even and we can write l = 2k, and the
sum takes the form
∞
∑ (a2k+m − a2k−m ) , (5.72)
k=1
where ai is the placeholder for the corresponding term in the sum. Note that
by the telescoping nature, once k is greater than or equal to k + 1, everything
cancels between the first term and the second. The only sum that remains is
just of the second term, up to k = m:
∞ m m
1
∑ (a2k+m − a2k−m ) = − ∑ a2k−m = − ∑ 2k − m . (5.73)
k=1 k=1 k=1
Let’s see what this evaluates to for some low values of m. If m = 1, we find
1
1
−∑ = −1 . (5.74)
k=1 2k − 1
If m = 3, we find
3
1 1 1
−∑ = 1−1− = − . (5.75)
k=1 2k − 3 3 3
We could keep going to larger m, but this suggests that it evaluates to −1/m.
Let’s prove this with induction. We have already verified low m; let’s now
assume it is true for m and prove it is true for m + 2. We consider the sum
m+2 m+1
1 1 1
− ∑ 2k − m − 2
=− ∑
2k − m − 2
−
2m + 4 −m−2
(5.76)
k=1 k=1
m+1
1 1
=− ∑ −
2k − m − 2 m + 2
k=1
m+1
1 1
=− ∑ −
2(k − 1) − m m + 2
k=1
m
1 1
=−∑ −
k=0 2k − m m+2
m
1 1 1 1
= −∑ − =− .
m k=1 2k − m m + 2 m+2
This completes the induction step and proves that

∞
1 1 1
∑ l +m − l −m = −m , (5.77)
l=1
if m is odd.
Next, if m is even, then l must be odd in the sum and can be written as
l = 2k − 1 and the sum takes the form
∞ m m
1
∑ (a2k−1+m − a2k−1−m ) = − ∑ a2k−1−m = − ∑ 2k − 1 − m , (5.78)
k=1 k=1 k=1
by the telescoping nature of the sum, as established earlier. Now, let’s

evaluate this again for some low m. First, for m = 2 we have
2
1
−∑ = 1−1 = 0. (5.79)
k=1 2k − 1−2
For m = 4, we have
4
1 1 1
−∑ = +1−1− = 0. (5.80)
k=1 2k − 1 − 4 3 3
This suggests that this sum simply vanishes! Let’s again prove this with
induction, assuming it is true for m. The sum for m + 2 is
m+2 m+1
1 1 1
− ∑ 2k − 1 − m − 2
=− ∑ −
k=1 k=1 2(k − 1) − 1 − m 2m + 4 − 1 − m − 2
45 Exercises
m
1 1
=−∑ − (5.81)
k=0 2k − 1 − m m + 1
m
1 1 1
= −∑ − = 0,
m + 1 k=1 2k − 1 − m m + 1
which proves the induction step.
Inputting these results into Eq. 5.70 proves that all off-diagonal terms in the
matrix representation of p̂2 vanish.
(c) Let’s now consider just the ( p̂2 )11 element. From our earlier partial fraction-
ing, this term can be expressed as
2
4h̄2 ∞ 1 1
( p̂2 )11 = 2 ∑ + (5.82)
a k=1 2k − 1 2k + 1

4h̄2 ∞ 1 1 2
= 2 ∑ + +
a k=1 (2k − 1)2 (2k + 1)2 (2k − 1)(2k + 1)
" #
∞ ∞
4h̄2 π 2 1 1 1
= 2 +∑ +∑ −
a 8 k=1 (2(k + 1) − 1)2 k=1 2k − 1 2k + 1
" #
∞
4h̄2 π 2 π 2 1 1
= 2 + −1+ ∑ −
a 8 8 k=1 2k − 1 2k + 1
2 2
4h̄ π π 2
= 2 + −1+1
a 8 8
π 2 h̄2
= .
a2
In these expressions, we used our knowledge about telescoping series and the
given value of the series presented in the exercise. This is indeed the value of
the squared momentum in the first energy eigenstate.
(d) The solution to this problem is explicitly worked out in J. Prentis and B. Ty,
“Matrix mechanics of the infinite square well and the equivalence proofs of
Schrödinger and von Neumann,” Am. J. Phys. 82, 583 (2014).
5.7 (a) Let’s now calculate the matrix elements of the position operator in the
energy eigenstate basis. We have
Z mπ x nπ x
2 a
(x̂)mn = ⟨ψm |x̂|ψn ⟩ = dx sin x sin (5.83)
a 0 a a
4mn (1 − (−1)m+n )
=− a.
π 2 (m2 − n2 )2
We have just presented the result from doing the integral, which is a standard
exercise in integration by parts.
(b) To evaluate the commutator of position and momentum, we need to take
their matrix product in two orders. First,
∞ ∞ l 2 1 − (−1)m+l

8ih̄ 1 − (−1)l+n
(x̂ p̂)mn = ∑ (x̂)ml ( p̂)ln = 2 mn ∑ . (5.84)
l=1 π l=1 (m2 − l 2 )2 (l 2 − n2 )
The opposite order of the product is

∞ ∞ l 2 1 − (−1)m+l

8ih̄ 1 − (−1)l+n
( p̂x̂)mn = ∑ ( p̂)ml (x̂)ln = 2 mn ∑ . (5.85)
l=1 π l=1 (m2 − l 2 )(l 2 − n2 )2
Their difference, the matrix elements of the commutator of position and
momentum is then
∞ l 2 (2l 2 − m2 − n2 ) 1 − (−1)m+l

8ih̄ 1 − (−1)l+n
([x̂, p̂])mn = 2 mn ∑ .
π l=1 (m2 − l 2 )2 (n2 − l 2 )2
(5.86)
Now, just like we observed for the square of momentum, all terms in this
sum explicitly vanish if m and n are even and odd (or vice-versa). The sum is
only non-zero if they differ by an even number. In this case, the numerator
simplifies to
∞ l 2 (2l 2 − m2 − n2 ) 1 − (−1)m+l

16ih̄
([x̂, p̂])mn = 2 mn ∑ (5.87)
π l=1 (m2 − l 2 )2 (n2 − l 2 )2
∞ l 2 1 − (−1)m+l

16ih̄ 1 1
= − 2 mn ∑ 2 2 2 2 + .
π l=1 (m − l )(n − l ) m − l
2 2 n2 − l 2
From exercise 5.6, we had shown that the denominator partial fractions into

1 1 1 1 1 1
= m − −n − .
(m2 −l 2 )(n2 −l 2 ) 2l 2 (m2 −n2 ) l −m l +m l −n l +n
(5.88)
The matrix element of the commutator simplifies to
([x̂, p̂])mn (5.89)
∞
8ih̄ mn 1 1 1 1
= 2 2 ∑
π m − n2 l=1
1 − (−1)m+l
m −
l −m l +m
− n −
l −n l +n

1 1
× 2 + .
l − m2 l 2 − n 2
(c) On the diagonal, the value of the commutator’s matrix elements are:

32ih̄ 2 ∞ l 2 1 − (−1)n+l
([x̂, p̂])nn = 2 n ∑ . (5.90)
π l=1 (l 2 − n2 )3
Even just including the first 20 terms in the sum, the result is within 1% of
ih̄ for n = 1, 2, 3, 4, 5, 6, 7.
(d) Let’s take the trace of the commutator as a general operator expression. We
have
tr[x̂, p̂] = tr(x̂ p̂) − tr( p̂x̂) = tr(x̂ p̂) − tr(x̂ p̂) = 0 , (5.91)
using the linearity and cyclicity of the trace. We had just shown that every
diagonal element of the commutator was ih̄ and yet the sum of diagonal
47 Exercises
elements must vanish, which seems to be at odds with one another. Our intu-
ition for the trace as the sum of diagonal elements of a matrix works well for
finite-dimensional matrices, just like our intuition for finite sums. However,
we know that infinite sums or series can have very surprising properties and
the trace of the commutator of position and momentum is an infinite sum.
So it’s like for any finite n, the diagonal element ([x̂, p̂])nn = ih̄, but the “infi-
nite” diagonal element is an enormous negative number ensuring that the
trace vanishes.
5..8 (a) The normalization of this wavefunction is
Z a
a5
1 = N2 dx x2 (a − x)2 = N 2 . (5.92)
0 30
Therefore, the normalized wavefunction is
r
30
ζ1 (x) = x(a − x) . (5.93)
a5
(b) The overlap of this state with the ground state of the infinite square well is
√ Z √
60 a π x 8 15
⟨ζ1 |ψ1 ⟩ = 3 dx x(a − x) sin = . (5.94)
a 0 a π3
To evaluate this integral is a standard exercise in integration by parts, but we
suppress details here. The fraction of ζ1 (x) that is described by the ground
state of the infinite square well is then the square of this:
960
|⟨ζ1 |ψ1 ⟩|2 = ≃ 0.998555 . (5.95)
π6
As illustrated in the picture, these wavefunctions are very, very similar!
(c) We can rearrange the Hamiltonian’s eigenvalue equation to solve for the
potential. We then find
h̄2 1 d 2
V (x) = ζ1 (x) + E1 . (5.96)
2m ζ1 (x) dx2
The term involving the second derivative of the wavefunction becomes
1 d2 1 d2 2
ζ 1 (x) = x(a − x) = − . (5.97)
ζ1 (x) dx 2 x(a − x) dx 2 x(a − x)
Therefore, the potential is
h̄2 1
V (x) = − + E1 . (5.98)
m x(a − x)
By setting
4h̄2
E1 = , (5.99)
ma2
we can plot this potential and compare to the infinite square well. This is
shown in Fig. 5.1.
t
Fig. 5.1 Comparison of the new potential (solid black) to the infinite square well. The new potential diverges at the points
x = 0, a, which is also where the wavefunctions must vanish.
(d) The thing we know about the first excited state with respect to the ground
state is that it is orthogonal. Further, the ground state is symmetric about the
potential, and so to easily ensure that the first excited state is orthogonal, we
just make it anti-symmetric on the well. So, we expect that the first excited
state wavefunction takes the form
ζ2 (x) = Nx(a − x)(a − 2x) , (5.100)
for some normalization constant N (that we won’t evaluate here). Plugging
this into the eigenvalue equation for the Hamiltonian with the potential
established above, the first excited state energy is
h̄2 1 d 2 2h̄2 1
E2 = − ζ 2 (x) +V (x) = E1 + . (5.101)
2m ζ2 (x) dx 2 m x(a − x)
Note that this ζ2 (x) is actually not an eigenstate, because the “energy”
E2 for this potential is position dependent. However, we can estimate the
excited state energy level by noting that the minimum value of the position
dependent piece is when a/2 where
1 4
= . (5.102)
x(a − x) x=a/2 a2
So, the second energy eigenvalue is approximately
8h̄2
E2 ≃ E1 +
. (5.103)
ma2
For comparison, in the infinite square well, note that
3π 2 h̄2 h̄2
E2 = E1 + ≃ E1 + 14.8 . (5.104)
2ma2 ma2
Quantum Mechanical Example: The Harmonic
6
Oscillator
Exercises
6.1 (a) The inner product of two coherent states as represented by the action of the
raising operator on the harmonic oscillator ground state is
|λ |2 |η |2 ∗
⟨ψλ |ψη ⟩ = e− e− ⟨ψ0 |eλ â eη â |ψ0 ⟩ .
†
2 2 (6.1)
Now, in the basis of the raising operator, â is a derivative and so its
exponential is the translation operator. That is,
∗ † +λ ∗
eλ â eη â = eη (â
†
) = eλ ∗ η eη â† . (6.2)
Then, the inner product of the two coherent states is
|λ |2 |η |2 ∗ |λ |2 |η |2 ∗η
⟨ψλ |ψη ⟩ = e− e− ⟨ψ0 |eλ â eη â |ψ0 ⟩ = e− e− eλ ⟨ψ0 |eη â |ψ0 ⟩
† †
2 2 2 2
|λ |2 |η |2 ∗η
= e− 2 e− 2 eλ . (6.3)
For any finite values of λ , η this is non-zero, and so two generic coherent
states are not orthogonal to one another.
(b) Recall that the nth energy eigenstate of the harmonic oscillator can be
expressed through the action of the raising operator on the ground state:
(â† )n
|ψn ⟩ = √ |ψ0 ⟩ . (6.4)
n!
A coherent state with eigenvalue λ can then be expanded in energy eigen-
states as
∞
|λ |2 |λ |2 λ n (â† )n
|ψλ ⟩ = e− eλ â |ψ0 ⟩ = e− ∑ √n!
†
2 2 √ |ψ0 ⟩ (6.5)
n=0 n!
∞
|λ |2 λn
= e− 2
∑ √n! |ψn ⟩ .
n=0
Then, the integral representing the sum over all coherent states is
Z ∞Z ∞
dRe(λ ) dIm(λ ) |ψλ ⟩⟨ψλ | (6.6)
−∞ −∞
Z ∞Z ∞ ∞
λ n (λ ∗ )m
dRe(λ ) dIm(λ ) e−|λ | ∑
2
= √ |ψn ⟩⟨ψm | .
−∞ −∞ n,m=0 n!m!
49
50 6 Quantum Mechanical Example: The Harmonic Oscillator
To determine what this operator is in the basis of energy eigenstates, we

need to do two things. First, we will establish that all off-diagonal elements
vanish. At row n and column m, for n ̸= m, the element is
Z ∞Z ∞ ∗ m
2 λ (λ )
n
dRe(λ ) dIm(λ ) e−|λ | √ (6.7)
−∞ −∞ n!m!
Z ∞ Z 2π
1
d ϕ e−|λ | |λ |n+m+1 ei(n−m)ϕ ,
2
=√ d|λ |
n!m! 0 0
where we have expressed the complex number λ = |λ |eiϕ in polar coordi-

nates on the right. The integral over argument ϕ vanishes, and therefore
indeed, the operator formed from the sum over all coherent states is diagonal
in the basis of energy eigenstates of the harmonic oscillator.
Now, we just need to determine what the diagonal entries are. Restricting to
the element of the nth energy eigenstate, we have
Z ∞Z ∞ Z 2π Z
|λ |2n 1 ∞
dRe(λ ) dIm(λ ) e−|λ | d|λ | |λ |2n+1 e−|λ |
2 2
= dϕ
−∞ −∞ n! n! 0 0
Z ∞
2π
d|λ | |λ |2n+1 e−|λ | .
2
= (6.8)
n! 0
Now, we can change of variables to |λ |2 = x for which the remaining integral
becomes
Z ∞ Z ∞
1 n!
d|λ | |λ |2n+1 e−|λ | = dx xn e−x =
2
. (6.9)
0 2 0 2
Then, the nth diagonal entry of the coherent state sum is
Z ∞Z ∞
|λ |2n 2π n!
dRe(λ ) dIm(λ ) e−|λ |
2
= =π. (6.10)
−∞ −∞ n! n! 2
Thus, the sum over coherent states is overcomplete by a factor of π :
Z ∞Z ∞
dRe(λ ) dIm(λ ) |ψλ ⟩⟨ψλ | = π I . (6.11)
−∞ −∞
6.2 (a) Demanding that the ground state wavefunction is normalized, we have
Z ∞ r
2 − mh̄ω x2 2 π h̄
1=N dx e =N . (6.12)
−∞ mω
Therefore, the L2 -normalized ground state wavefunction of the harmonic
oscillator is
mω 1/4 mω 2
ψ0 (x) = e− 2h̄ x . (6.13)
π h̄
(b) We had determined that the first excited state wavefunction is
r
2mω − mω x2
ψ1 (x) = N x e 2h̄ . (6.14)
h̄
51 Exercises
The second excited state is defined through the action of the raising operator
as
(â† )2 â†
ψ2 (x) = √ ψ0 (x) = √ ψ1 (x) (6.15)
2 2
r r ! r
h̄ d mω mω − m ω x 2
= − + x N x e 2h̄
2mω dx 2h̄ h̄

N 2mω 2 mω 2
=√ x − 1 e− 2h̄ x .
2 h̄
The third excited state is
(â† )3 â†
ψ3 (x) = √ ψ0 (x) = √ ψ2 (x) (6.16)
3! 3
r r !
h̄ d mω N 2mω 2 mω 2
= − + x √ x − 1 e− 2h̄ x
2mω dx 2h̄ 6 h̄
r
N 2mω 3 2mω − mω x2
=√ x − 3x e 2h̄ .
6 h̄ h̄
6.3 (a) The matrix elements of the raising operator â† in the basis of energy eigen-
states are found in the usual way, from sandwiching with the eigenstates.
That is,
âm (â† )n
(â† )mn = ⟨ψm |â† |ψn ⟩ = ⟨ψ0 | √ â† √ |ψ0 ⟩ . (6.17)
m! n!
As established in this chapter, this matrix element is non-zero only if m =
n + 1, so there are an equal number of raising and lowering operators. In
this case, note that
r
ân+1 (â† )n+1 1 d n+1 (n + 1)!
p √ =p † n+1
† n+1
(â ) = (6.18)
(n + 1)! n! n!(n + 1)! d(â ) n!
√
= n+1.
Therefore, expressed as an outer product of the energy eigenstates, the

raising operator is
∞ √
â† = ∑ n + 1 |ψn+1 ⟩⟨ψn | . (6.19)
n=0
As a matrix, this looks like

 
0 0 0 ···
 1 ··· 

â† =  √0 0 
0 2 0 ··· . (6.20)
 
.. .. .. ..
. . . .
(b) Note that the lowering operator is just the Hermitian conjugate of the
raising operator, and so
∞ √
â = (â† )† = ∑ n + 1 |ψn ⟩⟨ψn+1 | . (6.21)
n=0
We had already established the expressions for the position and momentum
operators in terms of the raising and lowering operators, and so we can
write:
r r
h̄ h̄ ∞ √
x̂ =
2mω
†
â + â = ∑ n + 1 (|ψn ⟩⟨ψn+1 | + |ψn+1 ⟩⟨ψn |) ,
2mω n=0
(6.22)
r r
mh̄ω † mh̄ω ∞ √
p̂ = i
2
â − â = i ∑ n + 1 (|ψn+1 ⟩⟨ψn | − |ψn ⟩⟨ψn+1 |) .
2 n=0
(6.23)
(c) The raising operator has exclusively non-zero entries just off the diagonal,
so its determinant is 0. The determinant is basis-independent, so in any basis
in which the raising operator is expressed, its determinant is 0.
(d) The general Laurent expansion of a function of the raising operator does
not exist. Because its determinant is 0, the inverse of â† does not exist, so
for g(â† ) to exist, we must assume that it is analytic; i.e., it has a Taylor
expansion.
6.4 (a) Using the result of Eq. 6.118, we can note that the only effect of time evo-
lution of a coherent state is to modify its eigenvalue under the lowering
operator as λ → λ e−iω t . This remains a general complex number for all
time, and one can simply replace the value of λ in the beginning of the anal-
ysis in section 6.4 to demonstrate that neither the variances of position nor
momentum are modified by including time dependence. Therefore, we still
find that
h̄ h̄mω
σx2 = , σ p2 = . (6.24)
2mω 2
Further, the Heisenberg uncertainty principle is saturated for a coherent
state for all time.
(b) For a classical harmonic oscillator, if you start at rest from position x = ∆x,
then the time-dependent position is sinusoidal:
x(t) = ∆x cos(ω t) , (6.25)
by the definition of ω as the frequency of oscillation. Correspondingly, the

classical momentum p(t) is
dx
p(t) = m = −mω ∆x sin(ω t) . (6.26)
dt
53 Exercises
From the results of the beginning of section 6.4, the time-dependent expec-
tation value of position on a coherent state is
r
h̄
⟨x̂⟩ = (λ e−iω t + λ ∗ eiω t ) . (6.27)
2mω
If we demand that at time t = 0 this is ∆x, we have that
r
h̄
∆x = (λ + λ ∗ ) . (6.28)
2mω
Correspondingly, the time-dependent expectation value of momentum is
r
mh̄ω ∗ iω t
⟨ p̂⟩ = i (λ e − λ e−iω t ) . (6.29)
2
At time t = 0, we assume that the particle is at rest, and so we enforce
that λ = λ ∗ , or that the eigenvalue of the lowering operator is initially
real-valued. Then,
r
2h̄
∆x = λ, (6.30)
mω
and
r r
h̄ 2h̄
⟨x̂⟩ = λ (e−iω t + eiω t ) = λ cos(ω t) = ∆x cos(ω t) . (6.31)
2mω mω
The expectation value of momentum is
r
mh̄ω √
⟨ p̂⟩ = i λ (eiω t − e−iω t ) = − 2mh̄ωλ sin(ω t) = −mω ∆x sin(ω t) .
2
(6.32)
These expectation values are identical to the classical oscillation!
(c) A coherent state of the raising operator would satisfy the eigenvalue equa-
tion
â† |χ ⟩ = η |χ ⟩ . (6.33)
Assuming completeness of the energy eigenstates of the harmonic oscillator,
this can be expanded as
∞ ∞
(â† )n
|χ ⟩ = ∑ βn |ψn ⟩ = ∑ βn √n! |ψ0 ⟩ . (6.34)
n=0 n=0
Then, the eigenvalue equation would imply that

∞
(â† )n+1 ∞ √ ∞
∑ n n! 0 ∑ n
β √ |ψ ⟩ = β n + 1|ψn+1 ⟩ = ∑ βn η |ψn ⟩ . (6.35)
n=0 n=0 n=0
By orthogonality of the energy eigenstates, we must enforce the recursive

relationship:
√
βn−1 n = ηβn . (6.36)
Further, note that there is no contribution to the ground state |ψ0 ⟩ on the
left, so we must require that β0 = 0. However, with the recursion relation,
this then implies that β1 = 0, β2 = 0, etc. So, there is actually no state on
the Hilbert space of the harmonic oscillator that can satisfy the eigenvalue
equation for the raising operator. The reason for this is essentially the same
as why we must forbid negative eigenvalues of the Hamiltonian. If there
was one eigenstate of the raising operator, then there must be an arbitrary
number of them, with decreasingly negative expectation values of energy.
6.5 (a) Let’s just evaluate the product Â† Â with the information given

i i
Â† Â = − √ p̂ +W (x̂) √ p̂ +W (x̂) (6.37)
2m 2m
p̂2 i
= + √ [W (x̂), p̂] +W (x̂)2 .
2m 2m
Now, recall that
dW
[W (x̂), p̂] = ih̄ , (6.38)
d x̂
so we have
p̂2 h̄ dW (x̂) p̂2
Â† Â = −√ +W (x̂)2 = Ĥ − E0 = +V (x̂) − E0 . (6.39)
2m 2m d x̂ 2m
Here we used the fact that the Hamiltonian is the sum of the kinetic and
potential energies. We can then rearrange and solve for the potential:
h̄ dW (x̂)
V (x̂) = − √ +W (x̂)2 + E0 . (6.40)
2m d x̂
(b) Note that the commutator is
[Â, Â† ] = ÂÂ† − Â† Â . (6.41)
We already evaluated the second order of the product, and the first is

i i
ÂÂ = √ p̂ +W (x̂)
†
− √ p̂ +W (x̂) (6.42)
2m 2m
p̂2 i
= − √ [W (x̂), p̂] +W (x̂)2
2m 2m
p̂2 h̄ dW (x̂)
= +√ +W (x̂)2 .
2m 2m d x̂
The commutator is then
r
† 2 dW (x̂)
[Â, Â ] = h̄ . (6.43)
m d x̂
55 Exercises
(c) With these results, the anti-commutator is

p̂2 h̄ dW (x̂) p̂2 h̄ dW (x̂)
{Â, Â† } = +√ +W (x̂)2 + −√ +W (x̂)2
2m 2m d x̂ 2m 2m d x̂
p̂2
= + 2W (x̂)2 , (6.44)
m
which is the sum of squares.
(d) The potential of the harmonic oscillator is
mω 2 2 h̄ dW (x̂)
V (x̂) = x̂ = W (x̂)2 − √ + E0 . (6.45)
2 2m d x̂
Recall that the ground state energy of the harmonic oscillator is
h̄ω
E0 = . (6.46)
2
Now, we make the ansatz that the superpotential is linear in x̂, where
W (x̂) = α x̂ , (6.47)
for some constant α . Plugging this into the expression for the potential we
have
h̄ dW (x̂) h̄α h̄ω mω 2 2
W (x̂)2 − √ + E0 = α 2 x̂2 − √ + = x̂ . (6.48)
2m d x̂ 2m 2 2
Demanding that the constant terms cancel each other this sets α to be
r
m
α =ω . (6.49)
2
Note that this is also consistent with the value of α from matching the
quadratic terms.
Further, note that with this superpotential, the operators Â and Â† for the
harmonic oscillator are proportional to the raising and lowering operators
â and â† . The anti-commutator of these operators is
p̂2 p̂2 mω 2 2
{Â, Â† } = + 2W (x̂)2 = +2 x̂ = 2Ĥ . (6.50)
m m 2
So, we didn’t need the anti-commutator for the harmonic oscillator because
it is proportional to the Hamiltonian itself.
(e) The potential of the infinite square well is 0 in the well and so the superpo-
tential satisfies:
h̄ dW (x̂)
−√ +W (x̂)2 + E0 = 0 . (6.51)
2m d x̂
The ground state energy of the infinite square well is
π 2 h̄2
E0 = , (6.52)
2ma2
for width a. The differential equation can be massaged as

√
dW (x̂) 2m
= W (x̂)2 + E0 , (6.53)
d x̂ h̄
or that
√
dW 2m
= d x̂ . (6.54)
W 2 + E0 h̄
Both sides can be integrated to find
√
1 W 2m
√ tan−1 √ = x̂ + c0 , (6.55)
E0 E0 h̄
for some integration constant c0 . That is, the superpotential is
√
p 2mE0 p
W (x̂) = E0 tan x̂ + c0 E0 (6.56)
h̄

π h̄ π π h̄
=√ tan x̂ + c0 √ .
2ma a 2ma
To fix the integration constant, note that the the potential diverges at x =
a, and so we demand that the superpotential diverges there, too. Tangent
diverges where its argument is π /2, so setting x̂ = a (in position basis), we
fix
π π h̄ π
a + c0 √ = , (6.57)
a 2ma 2
or that
r
ma
c0 = − . (6.58)
2 h̄
Inserting this into the expression for the superpotential, we have
π h̄ π π π h̄ π
W (x̂) = √ tan x̂ − = −√ cot x̂ . (6.59)
2ma a 2 2ma a
6.6 (a) Recall that the expression for the eigenstate of the phase operator in terms
of the eigenstates of the number operator is
∞
|θ ⟩ = β0 ∑ einθ |n⟩ . (6.60)
n=0
We can determine the action of the number operator N̂ on this state, noting
that
N̂|n⟩ = n|n⟩ . (6.61)
That is,
∞ ∞
d
N̂|θ ⟩ = β0 ∑ einθ N̂|n⟩ = β0 ∑ n einθ |n⟩ = −i d θ |θ ⟩ . (6.62)
n=0 n=0
57 Exercises
Now, acting on the number operator eigenstate, we have

Z 2π Z 2π
d
N̂|n⟩ = n|n⟩ = d θ cθ N̂|θ ⟩ = d θ cθ −i |θ ⟩ (6.63)
0 0 dθ
Z 2π
dcθ
= dθ i |θ ⟩ .
0 dθ
On the right, we have used integration by parts and the Hermitivity of the
number operator N̂, so we know that the boundary terms cancel. Then, for
this to be an eigenstate, we must require the differential equation
dcθ
i = ncθ , (6.64)
dθ
which has a solution
cθ = c0 e−inθ , (6.65)
for some normalization constant c0 . Therefore, the number eigenstate

expressed in terms of the eigenstates of the phase operator is
Z 2π
|n⟩ = c0 d θ e−inθ |θ ⟩ . (6.66)
0
(b) We just showed that, in the phase eigenstate basis, the number operator is a
derivative. Therefore, we immediately know that the commutator is
[Θ̂, N̂] = i . (6.67)
Now, this is a bit quick because it ignores the n = 0 subtlety. On the state
|0⟩, the phase θ is ill-defined, or, the representation of the number operator
as a derivative breaks down. Specifically, on |0⟩, the phase can be anything
and has no relationship to the properties of the number operator state. Cor-
respondingly, the variances of the number and phase operators on this state
are completely unrelated: |0⟩ is an eigenstate of the number operator and
so has 0 variance, while it is flat or uniform in the eigenstates of the phase
operator. As such the product of variances can be 0, and there exists no
non-trivial uncertainty relation.
6.7 (a) To prove that f (â† ) is not unitary, all we have to do is to show that there
is one counterexample. For concreteness, let’s just take f (â† ) = â† and the
state |ψ ⟩ = |ψ1 ⟩, the first excited state of the harmonic oscillator. Then, we
have
√
f (â† )|ψ ⟩ = â† |ψ1 ⟩ = (â† )2 |ψ0 ⟩ = 2|ψ2 ⟩ , (6.68)
√
a factor of 2 larger than the second excited state. A unitary operator main-
tains normalization, but in general â† does not, as exhibited in this example.
Therefore, a general operator of the form of an analytic function of the
raising operator f (â† ) is not unitary.
(b) We would like to construct a unitary operator exclusively from the raising
and lowering operators that leaves the action on the ground state of the
harmonic oscillator unchanged. To do this, we will start from the knowl-
edge that we can always write a unitary operator Û as the exponential of a
Hermitian operator T̂ :
Û = eiT̂ . (6.69)
Now, we would like to construct a Hermitian operator T̂ from the raising

and lowering operators such that its action on the ground state |ψ0 ⟩ exclu-
sively is of the form of a Taylor series in raising operator â† . This can easily
be accomplished by expressing T̂ as a sum of terms, each of which is a prod-
uct of raising and lowering operators, with the lowering operators to the
right:
∞
T̂ = ∑ ∗
βmn (â† )m ân + βmn (â† )n âm . (6.70)
m,n=0
Note that this operator is indeed Hermitian because T̂ † = T̂ , and when

acting on the ground state we have
∞
T̂ |ψ0 ⟩ = ∑ ∗
βmn (â† )m ân + βmn (â† )n âm |ψ0 ⟩ (6.71)
m,n=0
" #
∞ ∞
= ∑ βm0 (â )† m
+∑ ∗
β0n (â† )n |ψ0 ⟩ .
m=0 n=0
(c) Now, for the operator that generates coherent states from acting on the
ground state. Again, the eigenvalue equation for the coherent state |χ ⟩ is
â|χ ⟩ = λ |χ ⟩ . (6.72)
Let’s now express the coherent state as a unitary operator Û acting on the
ground state, where
|χ ⟩ = Û|ψ0 ⟩ = eiT̂ |ψ0 ⟩ , (6.73)
where T̂ is a Hermitian operator. Then, the eigenvalue equation becomes

∞ ∞ ∞
(iT̂ )n [â, (iT̂ )n ] (iT̂ )n
âeiT̂ |ψ0 ⟩ = ∑ â n!
|ψ0 ⟩ = ∑
n!
|ψ0 ⟩ = ∑ λ
n!
|ψ0 ⟩ , (6.74)
n=0 n=1 n=0
where we have used â|ψ0 ⟩ = 0. Matching terms at each order in n, this then
implies the recursion relation
[â, (iT̂ )n+1 ] (iT̂ )n
|ψ0 ⟩ = λ |ψ0 ⟩ . (6.75)
(n + 1)! n!
By linearity of all of the operators, this rearranges to
[â, T̂ n+1 ]|ψ0 ⟩ = −iλ (n + 1)T̂ n |ψ0 ⟩ . (6.76)

59 Exercises
The commutator therefore acts exactly like a derivative, which suggests that
T̂ is formed from the raising operator, as established earlier:
T̂ ∼ −iλ â† , (6.77)
however, this is not Hermitian. This can be fixed easily, by including the
appropriate factor of â:
T̂ = −iλ â† + iλ ∗ â . (6.78)
We can prove the recursion relation with induction. First, it works if n = 0,
for which

[â, T̂ ]|ψ0 ⟩ = â −iλ â† + iλ ∗ â |ψ0 ⟩ = −iλ |ψ0 ⟩ . (6.79)
Now, assuming it is true for n − 1, for n we have

âT̂ n+1 |ψ0 ⟩ = [â, T̂ ]T n + T̂ âT̂ n |ψ0 ⟩ = −iλ T̂ n + T̂ (−iλ nT n−1 ) |ψ0 ⟩
= −iλ (n + 1)T̂ n |ψ0 ⟩ , (6.80)
which is what we wanted to prove. Therefore the unitarized operator that
produces coherent states from the ground state is:
† −λ ∗ â
Û = eiT̂ = eλ â . (6.81)
This is typically called the displacement operator.
6.8 (a) This slightly displaced state |ψϵ ⟩ is still normalized, and so we must enforce
⟨ψϵ |ψϵ ⟩ = 1 = (⟨ψ | + ϵ ⟨ϕ |)(|ψ ⟩ + ϵ |ϕ ⟩) (6.82)
= ⟨ψ |ψ ⟩ + ϵ (⟨ψ |ϕ ⟩ + ⟨ϕ |ψ ⟩) + O(ϵ 2)
= 1 + ϵ (⟨ψ |ϕ ⟩ + ⟨ϕ |ψ ⟩) + O(ϵ 2) .
Therefore, for this equality to hold at least through linear order in ϵ , we must
demand that |ψ ⟩ and |ϕ ⟩ are orthogonal:
⟨ψ |ϕ ⟩ = 0 . (6.83)
(b) To take this derivative, we need to evaluate the variances on the state |ψϵ ⟩
through linear order in ϵ . Here, we will present the most general solutions
with non-zero values for the expectation values. First, for two Hermitian
operators Â, B̂, we have (ignoring terms beyond linear order in ϵ )

⟨ψϵ |Â|ψϵ ⟩⟨ψϵ |B̂|ψϵ ⟩ = ⟨ψ |Â|ψ ⟩ + ϵ (⟨ϕ |Â|ψ ⟩ + ⟨ψ |Â|ϕ ⟩) (6.84)

× ⟨ψ |B̂|ψ ⟩ + ϵ (⟨ϕ |B̂|ψ ⟩ + ⟨ψ |B̂|ϕ ⟩)
= ⟨Â⟩⟨B̂⟩ + 2 ϵ ⟨Â⟩Re(⟨ϕ |B̂|ψ ⟩) + 2 ϵ ⟨B̂⟩Re(⟨ϕ |Â|ψ ⟩) ,
where ⟨Â⟩ = ⟨ψ |Â|ψ ⟩, for example. Then, in the derivative, the first term
cancels, leaving
lim ⟨ψϵ |Â|ψϵ ⟩⟨ψϵ |B̂|ψϵ ⟩ − ⟨ψ |Â|ψ ⟩⟨ψ |B̂|ψ ⟩ (6.85)
ϵ →0
= 2⟨Â⟩Re(⟨ϕ |B̂|ψ ⟩) + 2⟨B̂⟩Re(⟨ϕ |Â|ψ ⟩) .

(c) Now, inserting this into the expression for the derivative and demanding it
vanishes enforces the constraint for arbitrary |ϕ ⟩ that

⟨Â⟩B̂ + ⟨B̂⟩Â |ψ ⟩ = λ |ψ ⟩ . (6.86)
We note that this is an eigenvalue equation because we had established that
|ψ ⟩ and |ϕ ⟩ are orthogonal. So for the derivative to vanish, all we need is
that the action of the Â and B̂ operators produces something proportional
to |ψ ⟩ again.
Now, we can insert the corresponding position and momentum operators in
for Â and B̂. The operators are
Â = (x̂ − ⟨x̂⟩)2 , B̂ = ( p̂ − ⟨ p̂⟩)2 . (6.87)
Note also that
h̄2
⟨Â⟩ = σx2 , ⟨B̂⟩ = σ p2 = , (6.88)
4σx2
using the saturated uncertainty principle. The eigenvalue equation can then
be expressed as

σ p2 (x̂ − ⟨x̂⟩)2 + σx2 ( p̂ − ⟨ p̂⟩)2 |ψ ⟩ = λ |ψ ⟩ . (6.89)
The eigenvalue λ can be determined by using the saturated uncertainty
principle, noting that
h̄2
⟨ψ | σ p2 (x̂ − ⟨x̂⟩)2 + σx2 ( p̂ − ⟨ p̂⟩)2 |ψ ⟩ = 2σx2 σ p2 = = λ ⟨ψ |ψ ⟩ = λ .
2
(6.90)
Then, the eigenvalue is
h̄2
. λ= (6.91)
2
This is a familiar equation that we might want to factorize. Let’s call
b̂ = σ p (x̂ − ⟨x̂⟩) + iσx ( p̂ − ⟨ p̂⟩) , b̂† = σ p (x̂ − ⟨x̂⟩) − iσx ( p̂ − ⟨ p̂⟩) . (6.92)
Now, note that the product b̂† b̂ is
b̂† b̂ = [σ p (x̂ − ⟨x̂⟩) + iσx ( p̂ − ⟨ p̂⟩)] [σ p (x̂ − ⟨x̂⟩) − iσx ( p̂ − ⟨ p̂⟩)] (6.93)
= σ p2 (x̂ − ⟨x̂⟩)2 + σx2 ( p̂ − ⟨ p̂⟩)2 − iσx σ p [x̂, p̂]
h̄2
= σ p2 (x̂ − ⟨x̂⟩)2 + σx2 ( p̂ − ⟨ p̂⟩)2 + .
2
Note that up to a rescaling, the operators b̂, b̂† are the familiar raising and
lowering operators, just translated by their expectation values. Then, the
eigenvalue equation can be expressed as

h̄2 h̄2
b̂† b̂ + |ψ ⟩ = |ψ ⟩ , (6.94)
2 2
61 Exercises
or simply b̂|ψ ⟩ = 0. That is, this eigenvalue equation is exactly the coherent
state equation, translated in space away from ⟨x̂⟩ = 0 and in momentum
away from ⟨ p̂⟩ = 0. The solutions of this eigenvalue equation are therefore
the coherent states with appropriate expectation values.
(d) This becomes very clear when we consider ⟨x̂⟩ = ⟨ p̂⟩ = 0. In this case, the
eigenvalue equation reduces to
h̄2
(σ p2 x̂2 + σx2 p̂2 )|ψ ⟩ =|ψ ⟩ . (6.95)
2
This is exactly of the form of the eigenvalue equation for the Hamiltonian of
the harmonic oscillator. We can verify this by dividing everything by 2mσx2 :
! 2
σ p2 2 p̂2 h̄ p̂2 h̄2
x̂ + |ψ ⟩ = x̂ 2
+ |ψ ⟩ = |ψ ⟩ . (6.96)
2mσx 2 2m 8mσx 4 2m 4mσx2
Now, fixing the coefficient of x̂2 to be mω 2 /2 we have

h̄
σx2 = . (6.97)
2mω
Thus, the eigenvalue would be
h̄2 h̄ω
= , (6.98)
4mσx2 2
exactly the ground state energy of the harmonic oscillator.
(e) Non-zero values for the expectation of position and momentum can eas-
ily be calculated using the translation property of the variance, where we
replace ⟨x̂2 ⟩ → ⟨(x̂ − ⟨x̂⟩)2 ⟩ and ⟨ p̂2 ⟩ → ⟨( p̂ − ⟨ p̂⟩)2 ⟩. The resulting differen-
tial equation is exactly that established for coherent states, i.e., those states
that saturate the Heisenberg uncertainty principle.
7 Quantum Mechanical Example: The Free Particle
Exercises
7.1 (a) The lowering operator can be expressed in the momentum basis as
r r
i mω i mh̄ω d
â = √ p̂ + x̂ = √ p+i . (7.1)
2mh̄ω 2h̄ 2mh̄ω 2 dp
Acting on the wavefunction in momentum space we have
r !
i mh̄ω d
âg(p) = √ p+i g(p) = λ g(p) . (7.2)
2mh̄ω 2 dp
This is a homogeneous differential equation that can be solved by separating

as
r !
dg 2 p
= −iλ − g, (7.3)
dp mh̄ω mh̄ω
or in differential form
r !
dg 2 p
= −iλ − dp. (7.4)
g mh̄ω mh̄ω
This has a solution

" r #
2 p2
g(p) = c exp −iλ p− , (7.5)
mh̄ω 2mh̄ω
for some normalization constant c.

(b) The speed of the center-of-probability of the coherent state is
d i ⟨ψ | p̂|ψ ⟩
⟨ψ |x̂|ψ ⟩ = ⟨ψ |[Ĥ, x̂]|ψ ⟩ = . (7.6)
dt h̄ m
For the free particle, the Hamilonian is purely a function of momentum,
and so Ĥ and p̂ commute for all time. So, all we need to note is that the
momentum operator in terms of the raising and lowering operators is
r
mh̄ω †
p̂ = i (â − â) . (7.7)
2
62
63 Exercises
Then, on the coherent state

r r
mh̄ω mh̄ω √
⟨ψ | p̂|ψ ⟩ = i ⟨ψ |â − â|ψ ⟩ = −i
†
(λ − λ ∗ ) = 2mh̄ω Im(λ ) .
2 2
(7.8)
Therefore, the speed of the center-of-probability on a free-particle coherent
state is
r
d 2h̄ω
⟨ψ |x̂|ψ ⟩ = Im(λ ) . (7.9)
dt m
Note that this is constant in time, so the acceleration is 0, as expected for a
particle experiencing no force.
(c) In position space, an eigenstate of the raising operator would satisfy
r r r !
i mω h̄ d mω
â ψ (x) = − √
†
p̂ + x̂ ψ (x) = − + x ψ (x)
2mh̄ω 2h̄ 2mω dx 2h̄
= λ ψ (x) . (7.10)
The solution of this differential equation is

" r #
mω 2 2mω
ψ (x) ∝ exp x −λ x . (7.11)
2h̄ h̄
This is not normalizable on x ∈ (−∞, ∞), and so no eigenstates of the raising

operator live on the Hilbert space.
7.2 (a) The reflection and transmission amplitudes have divergences when their
denominators vanish. This occurs when, for the reflection and transmission
amplitudes respectively,
p ap
k2 − mV0 + ik k2 − 2mV0 cot k2 − 2mV0 = 0 ,
p ap h̄a p
k k − 2mV0 cos
2 k − 2mV0 − i(k − mV0 ) sin
2 2
k2 − 2mV0 = 0 .
h̄ h̄
(7.12)
We can divide the expression
for the transmission amplitude divergences by
p
−i sin ah̄ k2 − 2mV0 which produces
p ap
k2 − mV0 + ik k2 − 2mV0 cot k2 − 2mV0 = 0 , (7.13)
h̄
exactly the expression for the location of the poles in the reflection ampli-
tude.
(b) To more easily understand the limitations of the poles in the amplitudes, we
will replace k = ip, for some real-valued momentum p. Then, the location
of the poles satisfies
p ap
−p2 − mV0 − p −p2 − 2mV0 cot −p2 − 2mV0 = 0 . (7.14)
h̄
64 7 Quantum Mechanical Example: The Free Particle
Further, bound states can only exist if V0 < 0, so we can replace V0 = −|V0 |,
where
q q
a
m|V0 | − p2 − p 2m|V0 | − p2 cot 2m|V0 | − p2 = 0 . (7.15)
h̄
That is,
q p
a p 2m|V0 | − p2
tan 2m|V0 | − p2 = . (7.16)
h̄ m|V0 | − p2
Tangent diverges whenever its argument is an odd multiple of π /2, so we
would expect in general that there are many momentum solutions p to this
equation. However, for a very shallow potential, |V0 | → 0, it’s less clear.
To see what happens for a shallow potential, note that in this limit, the
argument of tangent is also very small, so we can approximate
q q
a a
lim tan 2m|V0 | − p2 = 2m|V0 | − p2 . (7.17)
|V0 |→0 h̄ h̄
Then, in this limit, the bound state solutions satisfy
q p
a p 2m|V0 | − p2
2m|V0 | − p2 = , (7.18)
h̄ m|V0 | − p2
or that
h̄
m|V0 | − p2 = p. (7.19)
a
Now, this is just a quadratic equation so we know how to explicitly solve it,
but I want to do something a bit different here. We are considering the limit
in which |V0 | → 0, and that can be imposed by rescaling |V0 | → λ |V0 |, and
then taking λ → 0, but keeping |V0 | fixed. We are also assuming that mass m
and width a are just finite parameters, so they do not scale with λ . However,
the bound state momentum p must scale with λ , so that the particle indeed
stays in the well. A priori, we do not know this scaling, so we will just make
the replacement p → λ β p, for some β > 0. With the introduction of λ , the
equation becomes
h̄
mλ |V0 | − λ 2β p2 = λ β p . (7.20)
a
Now, in the λ → 0 limit, we can just keep terms that could possibly
contribute at leading order. So, we ignore the p2 term and have
h̄
mλ |V0 | = λ β p , (7.21)
a
which requires that β = 1. Then, the (magnitude of the) bound state
momentum is
ma|V0 |
p= . (7.22)
h̄
Regardless of the shallowness of the potential, this bound state always exists.
65 Exercises
(c) Now, if instead |V0 | → ∞, we expect that there is a solution to the bound
state equation whenever tangent diverges. That is, whenever
q
a π
2m|V0 | − p2 = (2n − 1) , (7.23)
h̄ 2
for n = 1, 2, 3, . . . . This can be rearranged to produce
p2 (2n − 1)2 π 2 h̄2

En ≡ |V0 | − = . (7.24)
2m 2ma2
These are exactly the odd energy eigenstates of the infinite square well.
7.3 (a) For a matrix A, construct the two Hermitian matrices
A + A† A − A†
H1 = , H2 = . (7.25)
2 2i
Then, note that we can reconstruct A as
A = H1 + iH2 , (7.26)
exactly what we were asked to prove.

(b) Recall that the interaction matrix satisfies the optical theorem
M̂† M̂ = 2 Im(M̂) . (7.27)
In terms of the Hermitian matrices X and Y this relationship is
M̂† M̂ = X2 + Y2 = 2 Im(M̂) = 2Y . (7.28)
Then, the eigenvalues satisfy
xn2 + y2n − 2yn = 0 , (7.29)
or that
xn2 + (yn − 1)2 = 1 . (7.30)
This is the equation for a circle centered at (xn , yn ) = (0, 1) with radius 1.
(c) The interaction matrix for the narrow potential is
maV0
1 1
M̂ = − h̄
. (7.31)
p + i maV0 1 1
h̄
One eigenvalue of this matrix is simply 0, while the non-trivial eigenvalue λ

is
2maV0 2maV0 2m2 a2V02
p h̄2
λ =− h̄
=− h̄
+i . (7.32)
p + i maV m a2V 2 m2 a2V 2
0 2
h̄ p2 + h̄2 0 2
p + h̄2 0
Now, let’s verify that these two eigenvalues lie on the Argand diagram.
Indeed it is true that a zero eigenvalue satisfies xn2 + (yn − 1)2 = 1, but only
is a single point on the circle, where (xn , yn ) = (0, 0). For the non-trivial
eigenvalue we can write it in a more compact form where
2p0 p 2p2
λ =− 2 2
+i 2 0 2 , (7.33)
p + p0 p + p0
where
maV0
p0 = . (7.34)
h̄
This eigenvalue can indeed take any value on the Argand circle for p ∈
(−∞, ∞).
7.4 (a) L2 -normalization for this wave packet in momentum space is
Z ∞ Z ∞
−
(p−p0 )2
1 σ p2
d p |g(p)| = dp q
2
e . (7.35)
−∞ −∞ πσ p2
To massage this integral, we can first translate p by p0 , which has no effect

on the bounds of integration:
Z ∞ −
(p−p0 )2 Z ∞ − p2
2 Z ∞ − 2 p2
1 σ p2 1 2
dp q e = dp q e σp
= dp q e σp .
−∞ πσ p2 −∞ πσ p2 0 πσ p2
(7.36)
We then make the change of variables
p2
x= , (7.37)
σ p2
and then the differential element is
√
2p 2 x
dx = 2 d p = dp. (7.38)
σp σp
The integral is then
Z ∞ − p2
2 Z ∞
2 1
dp q e σp
=√ dx x−1/2 e−x = 1 , (7.39)
0 πσ p2 π 0
by the integrals presented in Appendix A of the textbook.

(b) The position space wavefunction is
Z ∞ Z ∞ −
(p−p0 )2
dp px 1 1 p(x+x0 )
ψ (x) = √ g(p) e−i h̄ = √ dpe −i h̄ e 2σ p2
−∞ 2π h̄ 2π h̄ (πσ p2 )1/4 −∞
(7.40)
p2
− 02 Z ∞ (x+x0 )σ p2
2σ p − 12 p2 −2 p0 +i
1 e h̄ p
=√ dpe 2σ p
.
2π h̄ (πσ p2 )1/4 −∞
67 Exercises
We can then complete the square in the exponent, where

! !!2
(x + x0 )σ p2 (x + x0 )σ p2
p − 2 p0 + i
2
p = p − p0 + i (7.41)
h̄ h̄
!2
(x + x0 )σ p2
− p0 + i
h̄
!!2
(x + x0 )σ p2
= p− p0 + i − p20
h̄
(x + x0 )σ p2 (x + x0 )2 σ p4
− 2ip0 + .
h̄ h̄2
As earlier, we can translate the momentum p freely which does not effect the
boundaries of integration. With this translation, we have
(x+x0 )p0 (x+x0 )2 σ p2
i
2h̄2
− Z ∞ 2
− p2
1 e h̄
ψ (x) = √ dpe 2σ p
(7.42)
2π h̄ (πσ p )
2 1/4 −∞
r (x+x0 )p0 (x+x0 )2 σ p2
σ p i h̄ −
= e 2h̄2 .
π 2 h̄
This is still of the form of a Gaussian function, centered at the position
x = −x0 .
(c) The wave packet has a momentum of p0 , and so the velocity of the center
of probability, where there is no potential, is
d⟨x̂⟩ p0
= , (7.43)
dt m
simply from Ehrenfest’s theorem.
(d) From Example 7.2, the transmitted and reflected wavefunctions in momen-
tum space are
px0 (p+p0 )2
−i maV
h̄
0
ei h̄ −
2σ p2
gR (p) = e , (7.44)
p + i maV
h̄
0 (πσ p )
2 1/4
px0 (p−p0 )2
p ei h̄ −
2σ p2
gT (p) = e . (7.45)
p + i maV
h̄
0 (πσ p2 )1/4
Let’s focus on the reflected wavefunction first. Note that

maV x
i p+i h̄ 0 h̄0 (p+p0 )2
maV0 x0 −i maV
h̄
0
e −
2σ p2
gR (p) = e h̄2 e (7.46)
p + i maVh̄
0 (πσ p2 )1/4

maV x
Z i p+i h̄ 0 h̄0
−
(p+p0 )2
maV0 maV0 x0 e 2σ p2
= 2 e h̄2 dx0 e .
h̄ (πσ p2 )1/4
Then, using the result for the Fourier transform of the initial wave packet,
we then have
r Z (x+x0 )p0 maV0 x0 (x+x0 )2 σ p2
maV0 σ p maV20 x0 −i − 2 −
ψR (x) = 2 e h̄ dx 0 e h̄ h̄ 2h̄2 . (7.47)
h̄ π h̄
2
The integral that remains cannot be expressed in terms of elementary func-

tions, but this is sufficient for our needs here and in the next part. A similar
technique can be applied to the transmitted wavefunction, but we suppress
the details here.
(d) In the limit that σ p → 0, the reflected wavefunction becomes
Z
maV0 maV0 x0 −i
(x+x0 )p0 maV0 x0
− 2 −i maV 0 p0 x0
ψR (x) ∝ 2 e h̄2 dx0 e h̄ h̄ ∝ h̄
e−i h̄ , (7.48)
h̄ −p0 + i maV
h̄
0
where we have only retained the p0 dependence in the expressions. This is

exactly of the same form that we derived from translating the wavefunctions
directly in position space, for reflected momentum p = −p0 .
7.5 (a) To determine the differential equation that the translation operator satisfies,
note that the Dyson-like series for it can be re-expressed in a recursive form:
Z
i x+∆x ′
Û(x, x + ∆x) = 1 + dx p̂(x′ ) Û(x, x′ ) (7.49)
h̄ x
Z
i ∆x
= 1+ dy p̂(x + y) Û(x, x + y) .
h̄ 0
Now, we can differentiate both sides with respect to ∆x and find
Z
d d i ∆x
Û(x, x + ∆x) = dy p̂(x + y) Û(x, x + y) (7.50)
d∆x d∆x h̄ 0
i
= p̂(x + ∆x)Û(x, x + ∆x) .
h̄
(b) The translation operator acts to the right, and so to translate from x0 < 0 to
x1 > a on the step potential we can factor it as
(x1 −a) p̂ x0 p̂
Û(x0 , x1 ) = Û(a, x1 )Û(0, a)Û(x0 , 0) = ei h̄ Û(0, a)e−i h̄ . (7.51)
Here, we have written the explicit form for the translation operators where
the potential is 0. Where the potential is non-zero for rightward translation
(p > 0), we have
Z
i a ′p
Û(0, a) = 1 + dx 2m(E −V0 ) (7.52)
h̄ 0
2 Z a Z x′
i p p
+ dx′ 2m(E −V0 ) dx′′ 2m(E −V0 ) + · · ·
h̄ 0 0
p 2 2
a 2m(E −V0 ) 1 i p
= 1+i + a 2m(E −V0 ) + · · ·
h̄ 2 h̄
√
a 2m(E−V0 )
i
=e h̄ .
69 Exercises
√
Then, with p = 2mE where the potential is zero, the translation operator
across the potential is
√ √ √
(x1 −a) 2mE a 2m(E−V0 ) x0 2mE
Û(x0 , x1 ) = ei h̄ ei h̄ e−i h̄ . (7.53)
(c) Taking this δ -function potential limit, we can simplify the translation
operator in a few ways. First, for fixed and finite energy E, aE → 0 and
so
√ √ √
(x −x ) 2mE a 2ma(V0 −E)
i 1 0h̄ −
lim Û(x0 , x1 ) = lim e eh̄ (7.54)
a→0,V0 →∞ a→0,V0 →∞
√
(x −x ) 2mE
i 1 0h̄
=e .
This would seem to suggest that the δ -function potential has no effect on
the translation operator of a free particle. However, we know this cannot be
true from our analysis of the S-matrix for this system.
So, let’s try another route. Let’s go back to the differential equation for the
translation operator and let’s actually take the second derivative:
d2 p̂(x + ∆x)2
Û(x, x + ∆x) = − Û(x, x + ∆x) . (7.55)
d∆x 2
h̄2
We will consider x = 0 and ∆x = a, for very small displacement a, so that we
focus right on the region where the potential is. Right around the potential,
the squared momentum for a fixed energy E state is
p̂2 (x + ∆x) = 2m(E −V (x)) = 2m (E − aV0 δ (x)) → −2maV0 δ (x) , (7.56)
because a finite energy E is smaller than the infinite potential where x = 0.

Then, the differential equation for the translation operator is
d2 2maV0
Û(x, x + ∆x) = δ (x)Û(x, x + ∆x) . (7.57)
d∆x2 h̄2
That is, we are looking for a function that produces a δ -function after two
derivatives. Note that the Heaviside Θ-function or the step function has a
derivative that is a δ -function:
d
Θ(x) = δ (x) . (7.58)
dx
The Heaviside Θ-function Θ(x) is 0 for x < 0 and 1 for x > 0, so it is finite
everywhere, but discontinuous at x = 0. So, we just need to integrate once
more. Note that the function (called a rectifier or ReLU)

0, x < 0
ϒ(x) = (7.59)
x, x > 0
has a second derivative that is a δ -function. Therefore, right around x = 0,

the translation operator when acting on a plane wave of fixed energy and
momentum p > 0 takes the form

2maV0
Û(0, x) = c0 + c1 x + ϒ(x) , (7.60)
h̄2
where c0 , c1 are integration constants. They can be fit to match with the
translation operator that acts on the free particle, but we won’t do that here
(and it was effectively done in section 7.3.2 of the textbook).
7.6 (a) By definition, the Hausdorff length L of any curve must be independent of
resolution ∆x. So, we can take a derivative of both sides of the Hausdorff
dimension definition and find
d d
L = 0 = ∆xD−1 l + l(D − 1)∆xD−2 . (7.61)
d∆x d∆x
A smooth curve has a finite, fixed length as the resolution scale ∆x → 0, just
from the assumption that the derivative of a smooth curve exists. Therefore,
the length l is independent of resolution scale
d
l = 0. (7.62)
d∆x
Then, the only way that the the Hausdorff length can also be independent
of resolution scale ∆x is if D = 1.
(b) Note that at the nth step of constructing the Koch snowflake, the resolution
we need to see all of those kinks is
n
1
∆x = . (7.63)
3
The length l at this level of resolution is then
n
4
l= , (7.64)
3
because at each level, the length increases by a factor of 4/3. Then, the
Hausdorff length can be expressed as
n n(D−1)
4 1
L = lim l(∆x)D−1 = lim . (7.65)
∆x→0 n→∞ 3 3
Now, the Hausdorff length is supposed to be independent of resolution ∆x
or recursion level n and so we can take a derivative:
" #
d d 4 n 1 n(D−1)
L=0= (7.66)
dn dn 3 3
n n(D−1)
4 1 4 1
= log + (D − 1) log .
3 3 3 3
We can then immediately solve for the Hausdorff dimension D and find
log 4
D= ≃ 1.26186 . (7.67)
log 3
71 Exercises
(c) With this insight, let’s determine the Hausdorff dimension of the trajectory
of a quantum mechanical particle with 0 expectation value of momentum,
⟨ p̂⟩ = 0. With this expectation value, the saturated Heisenberg uncertainty
principle is
h̄2
⟨ p̂2 ⟩σx2 = . (7.68)
4
Note that the variance of position σx2 is a measure of the squared step size ∆x
away from the mean position, so we set σx2 = ∆x2 . Further, the characteristic
squared momentum depends on the spatial and temporal step sizes:
∆x2
⟨ p̂2 ⟩ = m2 . (7.69)
∆t 2
Then, the Heisenberg uncertainty principle becomes
∆x h̄
m ∆x = . (7.70)
∆t 2
Next, let’s multiply and divide by the total number of steps N. We then have
N∆x l h̄
m δ x = m ∆x = , (7.71)
N∆t T 2
using the relationship T = N∆x and that the total, resolution-dependent
length, is l = N∆x. Thus, this can be rearranged into
T h̄
≡ L = l∆x . (7.72)
2m
Everything on the left is independent of resolution ∆x, and so comparing to
the definition of the Hausdorff length, we find that the Hausdorff dimension
of the trajectory of a quantum mechanical particle is D = 2. That is, the
trajectory is an area-filling curve.
(d) With a non-zero expectation value of momentum, the variance is now
σ p2 = ⟨ p̂2 ⟩ − ⟨ p̂⟩2 = ⟨ p̂2 ⟩ − p20 . (7.73)
Just as earlier, the expectation value of the squared momentum is deter-
mined by the magnitude of the fluctuations:
∆x2 m2 l 2
⟨ p̂2 ⟩ = m2 = . (7.74)
∆t 2 T2
With the saturated Heisenberg uncertainty principle, note that the variance
of momentum is
h̄2 h̄2
σ p2 = = . (7.75)
4σx2 4∆x2
Then, we have that
h̄2 m2 l 2
σ p2 = = ⟨ p̂2
⟩ − p2
0 = − p20 , (7.76)
4∆x2 T2
or solving for total length l,

s
T h̄2
l= p20 + . (7.77)
m 4∆x2
In the limit where p0 ≪ h̄/∆x, this reduces to what we had found earlier,
where
T h̄ 1
lim l = , (7.78)
p0 ≪h̄/∆x 2m ∆x
for which the trajectory has dimension 2. However, in the opposite limit
where p0 ≫ h̄/∆x, we have
T p0
lim l = , (7.79)
p0 ≫h̄/∆x m
which is independent of resolution ∆x. Thus, in the high-momentum/small
h̄ limit, the Hausdorff dimension of the trajectory is 1, corresponding to a
smooth curve, exactly as we would expect in classical mechanics of a point
particle.
(e) The formulation of this problem from free-particle wavefunctions is worked
out in detail in: L. F. Abbott and M. B. Wise, “The dimension of a quantum
mechanical path,” Am. J. Phys. 49, 37–39 (1981).
7.7 (a) The S-matrix represents the probability amplitude for the scattering of a
momentum eigenstate off of some potential. As probability amplitudes,
when absolute squared and summed over all momentum you must get a total
probability of 1. A pole in the S-matrix at a real value of momentum would
mean that the probability amplitude at that value of momentum diverges
or that there is infinite probability for that momentum eigenstate to scatter.
This violates the axioms of probability, and so cannot exist.
Note that this says nothing about poles at complex values of momentum
because complex values of momentum are not observable. Unitary of the S-
matrix is only a statement about the conservation of probability of allowed
observable values of momentum.
(b) Again, the S-matrix represents the probability amplitude for momentum
eigenstates to scatter. Thus, when integrated against an arbitrary L2 -
normalizable function g(p), the result must be finite. Namely, the total
probability for a wave packet g(p) to scatter off of a localized potential and
transmit, say, is
Z ∞
PT = d p |AT |2 |g(p)|2 < ∞ , (7.80)
−∞
where AT is the transmission amplitude entry in the S-matrix. Therefore,

the S-matrix elements cannot affect the integrability of a wavepacket, and
therefore cannot scale like any positive power of momentum p as |p| → ∞.
Thus, the worst that the S-matrix elements can scale at large momentum is as
73 Exercises
a constant: AT → constant as |p| → ∞, for example. This result is consistent

with what we observed with the δ potential.
7.8 (a) We can immediately write down most entries of this S-matrix because the
infinite potential barrier prohibits transmission and any waves from the
right. So, the S-“matrix” is just a single value and as it must be unitary,
it can be expressed as a point on the unit circle,
S = AR = eiϕ , (7.81)
for some phase ϕ to be determined. The transmission amplitude is 0.

(b) We can determine this phase by our usual technique of matching across
boundaries. We can express the momentum state in the region to the left
of the potential in terms of two momentum eigenstates with initial p and
reflected p:
px px px px ϕ
px ϕ px ϕ

ψI (x) = ei h̄ + AR e−i h̄ = ei h̄ + e−i h̄ +iϕ = ei 2 ei h̄ −i 2 + e−i h̄ +i 2 .
(7.82)
AR is the reflection amplitude and we used our result from part (a) to write
it as an effective phase. ϕ is just some overall, x-independent phase, so is
irrelevant for fixed momentum scattering, so we can ignore the overall phase
factor. That is, we can consider

px ϕ
ψi (x) = 2 cos − . (7.83)
h̄ 2
In the region of finite potential, we can express the momentum state as
√ √
x 2m(E−V0 ) x 2m(E−V0 )
ψII (x) = α ei h̄ + β e−i h̄ , (7.84)
where the energy E is set by the momentum in the region with no potential
p2
E= . (7.85)
2m
Because of the infinite potential barrier at x = 0, we must enforce that the
momentum state vanish there (as usual, by Hermiticity of momentum). This
then enforces that β = −α , and so
√ √ p
x 2m(E−V0 )
−i
x 2m(E−V0 ) x 2m(E −V0 )
ψII (x) = α e i h̄ −e h̄ = 2iα sin . (7.86)
h̄
Now, we can match the momentum states at x = −a:
p !
ap ϕ a p2 − 2mV0
ψI (−a) = ψII (−a) → cos + = −iα sin .
h̄ 2 h̄
(7.87)
Next, their derivatives must also match:

d ψI (x) d ψII (x)
= (7.88)
dx x=−a dx x=−a
p p
p ap ϕ p2 − 2mV0 a p2 − 2mV0
→ sin + = iα cos .
h̄ h̄ 2 h̄ h̄
From matching the wave values at x = −a, we have that

ϕ
cos aph̄ + 2
α =i √ . (7.89)
a p2 −2mV0
sin h̄
Plugging this into the equation matching derivatives, we have

p p !
ap ϕ p2 − 2mV0 a p2 − 2mV0
tan + =− cot . (7.90)
h̄ 2 p h̄
This can easily be solved for phase ϕ where

p p !!
2ap −1 p2 − 2mV0 a p2 − 2mV0
ϕ =− − 2 tan cot . (7.91)
h̄ p h̄
There are some interesting limits to study. First, if p → 0, then the argument
of arctangent diverges, corresponding to a value of π /2. Thus the phase ϕ
in this limit is
lim ϕ = −π , (7.92)
p→0
exactly as expected for a wave on a rope, say, fixed to a wall.

8 Rotations in Three Dimensions
Exercises
8.1 (a) To show that the determinant of a rotation matrix is 1, we make the following
observations. The determinant is basis-independent, and just the product of
eigenvalues of the matrix. For the 2 × 2 matrix we are studying here, there
are therefore two eigenvalues, λ1 , λ2 for which
det U = λ1 λ2 . (8.1)
Now, we have expressed the unitary matrix U in the form of an exponentiated

Hermitian matrix, so we can extract the exponent by taking the logarithm:
i
log U = θx Ŝx + θy Ŝy + θz Ŝz . (8.2)
h̄
Correspondingly, taking the logarithm of the determinant returns:
log det U = log λ1 + log λ2 , (8.3)
which is just the sum of the eigenvalues of the logarithm of U. The sum of
the eigenvalues of a matrix is its trace, and therefore
i
log det U = tr log U = tr θx Ŝx + θy Ŝy + θz Ŝz . (8.4)
h̄
The spin operators Ŝi are proportional to the Pauli matrices, and all Pauli
matrices have 0 trace:
tr = σi = 0 , (8.5)
for i = 1, 2, 3. Therefore, we also have that
log det U = 0 , (8.6)
or that det U = 1, as we are asked to prove.

(b) We can show that these vectors are unit normalized by simply taking the
inner product with their Hermitian conjugate. For example,
eiξ1 cos θ
⃗v1†⃗v1 = e−iξ1 cos θ e−iξ2 sin θ = cos2 θ + sin2 θ = 1 . (8.7)
eiξ2 sin θ
75
76 8 Rotations in Three Dimensions
A similar calculation follows for ⃗v2 . Next, we are asked to fix the phases
ξ1 , ξ2 , ξ3 , ξ4 so that ⃗v1 and ⃗v2 are orthogonal. That is,
−eiξ3 sin θ
0 =⃗v1†⃗v2 = e−iξ1 cos θ e−iξ2 sin θ (8.8)
eiξ4 cos θ

= −e−i(ξ1 −ξ3 ) + e−i(ξ2 −ξ4 ) sin θ cos θ .
To ensure that this vanishes, we then require that

ξ1 − ξ3 = ξ2 − ξ4 mod 2π . (8.9)
Therefore, with this assignment, the now-orthogonal vectors can be
expressed as
iξ
e 1 cos θ −eiξ3 sin θ
⃗v1 = , ⃗
v = . (8.10)
eiξ2 sin θ ei(ξ2 +ξ3 −ξ1 ) cos θ
2
Now, we can establish the unitarity of the matrix formed from vectors⃗v1 ,⃗v2
by matrix multiplication. We multiply
! !
⃗v1† ⃗v1†⃗v1 ⃗v1†⃗v2 1 0
U U=
†
(⃗
v ⃗
v
1 2 ) = = , (8.11)
⃗v2† ⃗v2†⃗v1 ⃗v2†⃗v2 0 1
so this is indeed a unitary matrix.

(c) In terms of the elements of the vectors we constructed above, the determi-
nant of the matrix U is
det U = eiξ1 cos θ ei(ξ2 +ξ3 −ξ1 ) cos θ + eiξ2 sin θ eiξ3 sin θ = ei(ξ2 +ξ3 ) . (8.12)
Therefore, we must enforce that
ξ3 = −ξ2 mod 2π . (8.13)
The vectors are then
iξ
e 1 cos θ −e−iξ2 sin θ
⃗v1 = , ⃗v2 = . (8.14)
eiξ2 sin θ e−iξ1 cos θ
(d) Here, we will just establish the Cartesian components of the vector ⃗v1 , and
a similar procedure will follow for ⃗v2 . The first component of ⃗v1 is
eiξ1 cos θ = cos ξ1 cos θ + i sin ξ1 cos θ = a11 + ib11 , (8.15)
and so
a11 = cos ξ1 cos θ , b11 = sin ξ1 cos θ . (8.16)
Similarly, the second component of ⃗v1 is
eiξ2 sin θ = cos ξ2 sin θ + i sin ξ2 sin θ = a21 + ib21 , (8.17)
and so
a21 = cos ξ2 sin θ , b21 = sin ξ2 sin θ . (8.18)
77 Exercises
(e) Unit normalization of the vector ⃗v1 in this complex Cartesian expression
implies that
⃗v1†⃗v1 = 1 = (a11 − ib11 ) (a11 + ib11 ) + (a21 − ib21 ) (a21 + ib21 ) (8.19)
= a211 + b211 + a221 + b221 .
Thus, this describes all points in a four-dimensional space equidistant from

the origin; i.e., a three-dimensional sphere S3 . The Hopf fibration expresses
this three-sphere as a product of a two-sphere and a circle: S3 ≃ S2 × S1 .
8.2 (a) Let’s consider the nested commutator of elements of the Lie algebra of
rotations
3 3 3
[L̂i , [L̂ j , L̂k ]] = ih̄ ∑ ϵ jkl [L̂i , L̂l ] = −h̄2 ∑ ∑ ϵ jkl ϵ ilm L̂m . (8.20)
l=1 l=1 m=1
The other commutators are just cyclic permutations of this:

3 3
[L̂k , [L̂i , L̂ j ]] = −h̄2 ∑ ∑ ϵ i jl ϵ klm L̂m , (8.21)
l=1 m=1
3 3
[L̂ j , [L̂k , L̂i ]] = −h̄2 ∑ ∑ ϵ kil ϵ jlm L̂m .
l=1 m=1
The sum of these three nested commutators is
[L̂i , [L̂ j , L̂k ]] + [L̂k , [L̂i , L̂ j ]] + [L̂ j , [L̂k , L̂i ]] (8.22)

3 3
= −h̄2 ∑ ∑ ϵ jkl ϵ ilm + ϵ i jl ϵ klm + ϵ kil ϵ jlm L̂m .
l=1 m=1
To evaluate this, we can pick specific values for i, j, k and note that the result
is completely invariant to permutations of i, j, k. First, if i = j = k, every
term in the sum has a factor of ϵ iil = 0, and so we find 0. If j = k, but i ̸= j,
then the sum reduces to
[L̂i , [L̂ j , L̂ j ]] + [L̂ j , [L̂i , L̂ j ]] + [L̂ j , [L̂ j , L̂i ]] (8.23)

3 3
= −h̄2 ∑ ∑ ϵ j jl ϵ ilm + ϵ i jl ϵ jlm + ϵ jil ϵ jlm L̂m
l=1 m=1
3 3
= −h̄2 ∑ ∑ ϵ i jl ϵ jlm + ϵ jil ϵ jlm L̂m
l=1 m=1
3 3
= −h̄2 ∑ ∑ ϵ i jl ϵ jlm − ϵ i jl ϵ jlm L̂m = 0 ,
l=1 m=1
because ϵ jil = − ϵ i jl , by the total anti-symmetry. Finally, if all of i, j, k are

distinct, then the only way that ϵ jkl ̸= 0 is if l = i. However, such a factor
multiplies ϵ ilm = ϵ iim = 0, and similar for all other terms. Therefore, any
possible configuration of rotation operators satisfies the Jacoby identity.
(b) If the Lie bracket is the commutator, then the cyclic sum of the nested
commutations is
[Â, [B̂, Ĉ]] + [Ĉ, [Â, B̂]] + [B̂, [Ĉ, Â]] (8.24)
= [Â, B̂Ĉ − ĈB̂] + [Ĉ, ÂB̂ − B̂Â] + [B̂, ĈÂ − ÂĈ]
= ÂB̂Ĉ − ÂĈB̂ − B̂ĈÂ + ĈB̂Â + ĈÂB̂ − ĈB̂Â
− ÂB̂Ĉ + B̂ÂĈ + B̂ĈÂ − B̂ÂĈ − ĈÂB̂ + ÂĈB̂
= 0,
because all terms cancel pairwise.
8.3 First, note that the trace of the Casimir of a D dimensional representation is

(R) 2 (R) 2 (R) 2
tr(CR ID ) = DCR = tr L̂x + L̂y + L̂z . (8.25)
Because we can always perform a unitary transformation to diagonalize a rota-

tion matrix, the trace of the squares of the rotation matrices must all be equal.
Then,

(R) 2 DCR
tr L̂i = , (8.26)
3
for any i = x, y, z. This, however, is exactly the Killing form, and so
h i DCR
(R) (R)
tr L̂i L̂ j = kR δi j = δi j . (8.27)
3
For a representation of spin ℓ, the dimension D is
D = 2ℓ + 1 , (8.28)
which can be easily input into the expression for the Killing form.
8.4 If a state |ψ ⟩ is a coherent state of the angular momentum lowering operator L̂− ,
then it satisfies
L̂− |ψ ⟩ = λ |ψ ⟩ , (8.29)
for some complex-valued λ . Using the completeness of the states in the spin-ℓ
representation of angular momentum, we can express |ψ ⟩ as the linear combina-
tion:
ℓ
|ψ ⟩ ∑ βm |ℓ, m⟩ , (8.30)
m=−ℓ
for z-component of angular momentum m and coefficients βm . Now, on this

expression, the coherent state satisfies
ℓ ℓ
L̂− |ψ ⟩ = ∑ βm L̂− |ℓ, m⟩ = ∑ λ βm |ℓ, m⟩ . (8.31)
m=−ℓ m=−ℓ
By the definition of the lowering operator, the state L̂− |ℓ, m⟩ ∝ |ℓ, m − 1⟩, when
m > −ℓ. Therefore, the highest state in the sum on the left is |ℓ, ℓ − 1⟩, while the
79 Exercises
highest state on the right is |ℓ, ℓ⟩. Therefore, we must enforce that λ βℓ = 0. There
are two possibilities. If we set βℓ = 0, then the recursive nature of this relationship
sets βℓ−1 = βℓ−2 = · · · = β−ℓ+1 = 0. Thus, the coherent state would simply be the
state with the lowest angular momentum eigenvalue, |ψ ⟩ = |ℓ, −ℓ⟩ with λ = 0. By
contrast, if we fix λ = 0, then we again find this state. Therefore, the only coherent
state of angular momentum is the state of the lowest value of z-component.
Note that in the case of the harmonic oscillator coherent states, there is no
upper limit to the energy eigenvalue, so there was no issue with the matching of
states in the coherent state eigenvalue equation. The finite number of angular
momentum states forbids the existence of coherent states.
8.5 (a) The Hamiltonian for a spin-1/2 particle immersed in a magnetic field was
eB0
Ĥ = Ŝz . (8.32)
me
The commutator of the Hamiltonian and the spin operator Ŝx is then
eB0 eh̄B0
[Ĥ, Ŝx ] = [Ŝz , Ŝx ] = i Ŝy . (8.33)
me me
Therefore, the uncertainty principle that the energy and x-component of
angular momentum satisfy is
2
⟨[Ĥ, Ŝx ]⟩ e2 h̄2 B20
σE2 σS2x ≥ = |⟨Ŝy ⟩|2 . (8.34)
2i 4m2e
(b) When the expectation value of the y-component of angular momentum van-
ishes, ⟨Ŝy ⟩ = 0, the lower bound in the uncertainty principle is 0. Having an
expectation value of 0 means that the state |ψ ⟩ is in a linear combination of
up and down spin with coefficients of equal magnitude:
1
|ψ ⟩ = √ eiϕ | ↑y ⟩ + e−iϕ | ↓y ⟩ , (8.35)
2
for some phase ϕ . In terms of eigenstates of Ŝz , we can express eigenstates
of Ŝy as
1 1
| ↑y ⟩ = √ (| ↑⟩ + i| ↓⟩) , | ↓y ⟩ = √ (| ↑⟩ − i| ↓⟩) . (8.36)
2 2
The state |ψ ⟩ in terms of the eigenstates of Ŝz is then
|ψ ⟩ = cos ϕ | ↑⟩ − sin ϕ | ↓⟩ . (8.37)
The variance of the Hamiltonian on this state only vanishes if it is an eigen-

state of Ŝz ; i.e., ϕ = 0, π /2, π , 3π /2. The variance of Ŝx vanishes when the
magnitudes of the coefficients of this state are equal, corresponding to an
eigenstate of Ŝx : ϕ = π /4, 3π /4, 5π /4, 7π /4. At all other values of ϕ , the
lower bound in the uncertainty principle is 0 on this state, but the product
of variances is non-zero.
8.6 To prove the Schouten identity, we will work in the basis of eigenstates of the spin
operator Ŝz . First, we can always rotate the z-axis to lie along the direction of one
of the spins. So we can, without loss of generality, choose |ψ ⟩ = |↑⟩. For the other
spinors, we can write
|ρ ⟩ = a1 | ↑⟩ + b1 | ↓⟩ , (8.38)
|χ ⟩ = a2 | ↑⟩ + b2 | ↓⟩ , (8.39)
|η ⟩ = a3 | ↑⟩ + b3 | ↓⟩ , (8.40)
for some complex coefficients ai , bi . The two products of inner products of spinors
on the left of the Schouten identity are then:
⟨ψ |ρ ⟩⟨χ |η ⟩ = a1 (a∗2 a3 + b∗2 b3 ) , ⟨ψ |η ⟩⟨χ |ρ ⟩ = a3 (a∗2 a1 + b∗2 b1 ) . (8.41)
Then, we need to determine how the Pauli matrix acts on spinors. Note that

0 1
iσ2 = . (8.42)
−1 0
Then, the action of this matrix on the spinors is:
iσ2 |η ⟩ = b3 | ↑⟩ − a3 | ↓⟩ , iσ2 |χ ⟩ = b2 | ↑⟩ − a2 | ↓⟩ . (8.43)
Then, the final product of inner products is
(⟨ρ |)∗ iσ2 |η ⟩⟨ψ |iσ2 (|χ ⟩)∗ = (a1 b3 − b1 a3 )b∗2 . (8.44)
Putting this together with the other spinor product on the right side of the
equation, we find
⟨ψ |η ⟩⟨χ |ρ ⟩ + (⟨ρ |)∗ iσ2 |η ⟩⟨ψ |iσ2 (|χ ⟩)∗ = a3 (a∗2 a1 + b∗2 b1 ) + (a1 b3 − b1 a3 )b∗2
= a1 (a∗2 a3 + b∗2 b3 ) = ⟨ψ |ρ ⟩⟨χ |η ⟩ ,
(8.45)
proving the Schouten identity.

8.7 (a) We can take the trace of the commutator, multiplied by either operator Â or
B̂. Note that

tr Â[Â, B̂] = tr Â2 B̂ − ÂB̂Â = 0 = iα trÂ2 + iβ tr ÂB̂ . (8.46)
The trace of the commutator is 0 by the cyclic property of the trace. By the
orthogonality of Â and B̂, the trace of their product is 0, and so this requires
that
iα trÂ2 = 0 . (8.47)
By assumption, trÂ2 ̸= 0, and so we must fix α = 0. A similar analysis holds

for the trace of B̂[Â, B̂], which requires that β = 0. Therefore, we must enforce
that [Â, B̂] = 0, or that a two-dimension Lie algebra is necessarily Abelian.
81 Exercises
(b) Consider the operator

Ĉ = B̂ + α Â , (8.48)
for some constant Â. The trace of this operator with Â is

tr ÂĈ = tr ÂB̂ + α trÂ2 . (8.49)
If we demand that this vanishes, then α is fixed to be

tr ÂB̂
α =− . (8.50)
trÂ2
Thus, if tr[ÂB̂] ̸= 0, we can construct

tr ÂB̂
Ĉ = B̂ − Â , (8.51)
trÂ2
and then tr[ÂĈ] = 0.
8.8 (a) The matrix that rotates about the z-axis is the simplest, because it is identical
to the familiar two-dimensional rotation matrix, just with an added diagonal
entry to represent that the z-component of the vector is unaffected:
 
cos θ − sin θ 0
Uz (θ ) =  sin θ cos θ 0  . (8.52)
0 0 1
The rotation matrices about the x and y axes are similar, just with compo-
nents permuted:
   
1 0 0 cos θ 0 sin θ
Ux (θ ) =  0 cos θ − sin θ  , Uy (θ ) =  0 1 0 .
0 sin θ cos θ − sin θ 0 cos θ
(8.53)
(b) We will just construct the Lie algebra element L̂x explicitly, and then just
state the results for L̂y , L̂z . We have
 
1 0 0
θ L̂x
Ux (θ ) =  0 cos θ − sin θ  = ei h̄ . (8.54)
0 sin θ cos θ
Taylor expanding the explicit matrix and the exponential each to linear
order in θ , we find
 
1 0 0
θ L̂x 
I+i = 0 1 −θ  , (8.55)
h̄
0 θ 1
or that
 
0 0 0
L̂x = h̄  0 0 i  . (8.56)
0 −i 0
Correspondingly, the operators L̂y and L̂z are

   
0 0 −i 0 i 0
L̂y = h̄  0 0 0  , L̂z = h̄  −i 0 0  . (8.57)
i 0 0 0 0 0
(c) Let’s first construct the matrices that implement rotation by π /2 about the
x and y axes. We have
   
1 0 0 0 0 1
Ux (π /2) =  0 0 −1  , Uy (π /2) =  0 1 0  . (8.58)
0 1 0 −1 0 0
If we first rotate about x and then y, the resulting vector is

   
0 0 1 1 0 0 0
Uy (π /2)Ux (π /2)⃗v =  0 1 0   0 0 −1   0  (8.59)
−1 0 0 0 1 0 1
    
0 0 1 0 0
=  0 1 0   −1  =  −1 .
−1 0 0 0 0
Performing the rotation in the opposite way, we find

   
1 0 0 0 0 1 0
Ux (π /2)Uy (π /2)⃗v =  0 0 −1   0 1 0   0  (8.60)
0 1 0 −1 0 0 1
    
1 0 0 1 1
=  0 0 −1   0  =  0  .
0 1 0 0 0
The difference of the resulting vectors is then

 
−1
⃗vyx −⃗vxy =  −1  . (8.61)
0
(d) By contrast, we could have calculated the commutator of the rotation

matrices. Note that their products in different orders are
    
0 0 1 1 0 0 0 1 0
Uy (π /2)Ux (π /2) =  0 1 0   0 0 −1  =  0 0 −1  ,
−1 0 0 0 1 0 −1 0 0
(8.62)
    
1 0 0 0 0 1 0 0 1
Ux (π /2)Uy (π /2) =  0 0 −1   0 1 0  =  1 0 0  .
0 1 0 −1 0 0 0 1 0
83 Exercises
Their commutator is thus

 
0 1 −1
Uy (π /2)Ux (π /2) − Ux (π /2)Uy (π /2) =  −1 0 −1  . (8.63)
−1 −1 0
The action of this matrix product difference on the vector ⃗v is then
  
0 1 −1 0
(Uy (π /2)Ux (π /2) − Ux (π /2)Uy (π /2))⃗v =  −1 0 −1   0 
−1 −1 0 1
 
−1
=  −1  , (8.64)
0
exactly as expected.
9 The Hydrogen Atom
Exercises
9.1 (a) The potential of this hydrogenized atom is now

Ze2 1
V (r) = − , (9.1)
4π ϵ 0 r
so effectively this is just the hydrogen atom potential with the replace-
ment e2 → Ze2 . Correspondingly, in the expression for the energy levels,
we can make this replacement to determine the new, modified energy levels.
Therefore, the energy eigenstates of this atom are
me Z 2 e4 1
En = − , (9.2)
2(4π ϵ 0 )2 h̄2 n2
for n = 1, 2, 3, . . . .
(b) To estimate the speed v of the electron in the ground state, we note that the
expectation value of the kinetic energy is
2
p̂r Ze2 1 1
= E1 + ≡ me v 2 . (9.3)
2me 4π ϵ 0 r̂ 2
The expectation value of the inverse radius on the ground state is
Z
1 4 ∞ 1 Ze2 me
= 3 dr r e−2r/a0 = = , (9.4)
r̂ a0 0 a0 4π ϵ 0 h̄2
the inverse of the new Bohr radius. Then, the expectation value of the kinetic
energy in the ground state is
2
p̂r me Z 2 e 4 Ze2 Ze2 me me Z 2 e 4
=− + = . (9.5)
2me 2(4π ϵ 0 )2 h̄2 4π ϵ 0 4π ϵ 0 h̄2
2(4π ϵ 0 )2 h̄2
The expectation value of the squared velocity of the ground state is then
Z 2 e4
v2 = . (9.6)
(4π ϵ 0 )2 h̄2
Setting this equal to v = c/2, the atomic number when the speed is compa-
rable to the speed of light is
4π ϵ 0 h̄c 1
Z= = ≃ 69 . (9.7)
2e 2 2α
84
85 Exercises
This element is thulium.

(c) The relativistic kinetic energy is
! s
q
|⃗p|2 c2
K= m2e c4 + |⃗p|2 c2 − me c2
= me c 1+ 2 4 −1 2
(9.8)
me c

|⃗p| c
2 2 |⃗p| c
4 4
= me c2 1 + 2 4 − 4 8 + · · · − 1
2me c 8me c
|⃗p|2 |⃗p|4
= − 3 2 +··· .
2me 8me c
So indeed the lowest order term in the small-velocity limit reproduces the
familiar non-relativistic kinetic energy. The first relativistic correction is
negative, and so relativistic corrections tend to decrease the kinetic energy
as compared to the non-relativistic expression.
9.2 (a) To calculate the Poisson bracket of the angular momentum and the Ham-
iltonian, we recall the form of the Poisson bracket. For an individual
component of the angular momentum vector Lk , the Poisson bracket is
3
∂ H ∂ Lk ∂ H ∂ Lk
{H, Lk } = ∑ − . (9.9)
l=1 ∂ rl ∂ pl ∂ pl ∂ rl
For this part, we will simply assume that the Hamiltonian takes the form
3 p2
H= ∑ 2mke +V (r) , (9.10)
k=1
p
where r = x2 + y2 + z2 , the radial distance. The kth component of the
angular momentum vector can be expressed as
3
Lk = ∑ ϵ i jk ri p j . (9.11)
i, j=1
Then, we can evaluate derivatives. We have

∂H ∂V rl ∂ V ∂H pl
= = , = , (9.12)
∂ rl ∂ rl r ∂r ∂ pl me
∂ Lk 3
∂ Lk 3
∂ rl
= ∑ ϵ l jk p j , ∂ pl
= ∑ ϵ jlk r j .
j=1 j=1
Then, the Poisson bracket is

3
rl r j ∂ V p j pl
{H, Lk } = ∑ jlk r ∂ r l jk me = 0 ,
ϵ − ϵ (9.13)
j,l=1
because
3 3
∑ ϵ jlk rl r j = ∑ ϵ l jk p j pl = 0 , (9.14)
j,l=1 j,l=1
86 9 The Hydrogen Atom
and positions and momenta commute amongst themselves. Therefore,

angular momentum is conserved in a central potential.
(b) For the Poisson bracket with the Laplace–Runge–Lenz vector, we need to
specify the explicit form of the potential of the Hamiltonian, and not just the
fact that it is only dependent on radial distance. Then, the spatial derivative
of the Coulomb potential is
∂H ∂V rl ∂ V e2 rl ∂ 1 e2 rl
= = =− = . (9.15)
∂ rl ∂ rl r ∂r 4π ϵ 0 r ∂ r r 4π ϵ 0 r 3
We had already evaluated the derivatives of the Laplace–Runge–Lenz vector

component Ak in the chapter, finding
∂ Ak e2 rk rl − r2 δkl |⃗p|2 δkl − pk pl

= + , (9.16)
∂ rl 4π ϵ 0 r3 me
∂ Ak 2rk pl − rl pk −⃗r ·⃗p δkl
= ,
∂ pl me
where we have inserted the explicit expression for the Hamiltonian. Then,
the Poisson bracket of the Hamiltonian and this component of the Laplace–
Runge–Lenz vector is

3
e2 rl 2rk pl − rl pk −⃗r ·⃗p δkl
{H, Ak } = ∑ 4π ϵ 0 r 3 me
(9.17)
l=1
2
pl e rk rl − r2 δkl |⃗p|2 δkl − pk pl
− +
me 4π ϵ 0 r3 me
!
3
1
=− |⃗p|2 pk − pk ∑ p2l
m2e l=1
" #
e2 1 3
+
4π ϵ 0 me r3 ∑ rk rl pl − pk rl2 −⃗r ·⃗p rk + r pk
2
l=1
= 0,
where we have used, for example, the dot product
3
⃗r ·⃗p = ∑ rl pl . (9.18)
l=1
Therefore, the Laplace–Runge–Lenz vector is also conserved in the hydro-

gen atom.
(c) Now onto the Poisson bracket of two components of the Laplace–Runge–
Lenz vector. Using the derivatives established above, the Poisson bracket
is
87 Exercises

3
e2 ri rk − r2 δik |⃗p|2 δik − pi pk 2r j pk − rk p j −⃗r ·⃗p δ jk
{Ai , A j } = ∑ 4π ϵ 0 r3
+
me me
k=1
(9.19)

e2 r j rk − r2 δ jk |⃗p|2 δ jk − p j pk 2ri pk − rk pi −⃗r ·⃗p δik
− +
4π ϵ 0 r3 me me
!
3
e2 1
= (ri p j − r j pi ) 3r2 − ∑ rk2
4 π ϵ 0 me r 3
k=1
!
3
1
− 2 (ri p j − r j pi ) 3|⃗p| − 2 ∑ pk
2 2
me k=1
2
2 |⃗p| 2
e 1 2
= − (ri p j − r j pi ) − = − H(ri p j − r j pi ) ,
me 2me 4π ϵ 0 r me
the Hamiltonian for the hydrogen atom. Now, all that remains is to interpret
ri p j − r j pi . This looks like a component of angular momentum if i ̸= j, so
we must enforce that. Note that
3 3 3
∑ ϵ i jk Lk = ∑ ϵ i jk ϵ lmk rl pm = ∑ δil δ jm − δim δ jl rl pm (9.20)
k=1 k,l,m=1 l,m=1
= ri p j − r j pi ,
where we used the result established in the chapter for the sum over a product
of totally anti-symmetric objects. Using this result, we then find
3
2
{Ai , A j } = − H ∑ ϵ Lk , (9.21)
me k=1 i jk
exactly as claimed.
9.3 (a) Because angular momentum is conserved, the two particles’ trajectories
must lie in a plane and must orbit with a common angular frequency ω .
Then, the total energy of the system can be expressed as
1 1 n
E = m1 r12 ω 2 + m2 r22 ω 2 + krn , (9.22)
2 2 |n|
where r is the distance between the masses m1 and m2 and r1 , r2 is their
respective distance from their common orbiting point. Note that r1 + r2 = r.
Given this expression for the potential energy, we can also write down New-
ton’s second law. Note that the magnitude of the conservative force from the
potential is
d
F= U(r) = |n|krn−1 . (9.23)
dr
Then, the Newton’s second laws for both masses are
|n|krn−1 = m1 r1 ω 2 , |n|krn−1 = m2 r2 ω 2 . (9.24)

Then, we note that this enforces

m1
r2 = r1 , (9.25)
m2
and then with r1 + r2 = r, the distances to the common center are
m2 m1
r1 = r, r2 = r. (9.26)
m1 + m2 m1 + m2
With these relationships, the total energy and Newton’s second laws enforce
1 m1 m2 2 2 n m1 m2
E= r ω + krn , |n|krn−1 = rω 2 . (9.27)
2 m1 + m2 |n| m1 + m2
Note the appearance of the reduced mass in both expressions.
Now, we will massage Newton’s second law to relate the kinetic and potential
energies. First, we can multiply both sides by distance r:
m1 m2 2 2
|n|krn = r ω . (9.28)
m1 + m2
Then, the right side is nearly the kinetic energy K, up to a factor of 2:
|n| n 1 m1 m2 2 2
kr = r ω = K. (9.29)
2 2 m1 + m2
Next, the right side the magnitude of the potential energy U up to a factor
of |n|/2:
|n| n |n|
kr = U . (9.30)
2 2
Therefore, the kinetic energy is related to the potential energy as
|n|
K= U. (9.31)
2
(b) Recall that the time dependence of an expectation value of an operator Ô is
d i
⟨Ô⟩ = ⟨[Ĥ, Ô]⟩ . (9.32)
dt h̄
On an energy eigenstate, Ĥ|ψ ⟩ = E|ψ ⟩, the energy E, and so the expectation
value of the commutator is
⟨[Ĥ, Ô]⟩ = E⟨Ô − Ô⟩ = 0 . (9.33)
Note that this argument is independent of what the operator Ô actually is.
With this observation and the fact that the Hamiltonian can be expressed as
kinetic plus potential operators
Ĥ = K̂ +V (r̂) , (9.34)
we require that
⟨[Ĥ, Ô]⟩ = 0 , (9.35)

89 Exercises
or that
⟨[K̂, Ô]⟩ = −⟨[V (r̂), Ô]⟩ . (9.36)
Let’s first consider the commutator of the potential with Ô. In position
space, the operator is
∂ ∂ ∂
Ô = −ih̄x − ih̄y − ih̄z . (9.37)
∂x ∂y ∂z

n ∂ ∂ ∂
[Ô,V (r̂)] = −ih̄ k x +y +z (x2 + y2 + z2 )n/2 . (9.38)
|n| ∂x ∂y ∂z
Note that the derivative is
∂ 2
(x + y2 + z2 )n/2 = nx2 (x2 + y2 + z2 ) 2 −1 ,
n
x (9.39)
∂x
and so the sum of the derivatives is

∂ ∂ ∂ n
x +y +z (x2 + y2 + z2 )n/2 = n(x2 + y2 + z2 ) 2 . (9.40)
∂x ∂y ∂z
Therefore, the commutator is
[Ô,V (r̂)] = −ih̄nV (r̂) . (9.41)
Now, for the kinetic energy. In momentum space, the operator is

∂ ∂ ∂
Ô = 3ih̄ + p̂x x̂ + p̂y ŷ + p̂z ẑ = ih̄ 3 + px + py + pz . (9.42)
∂ px ∂ py ∂ pz

h̄ ∂ ∂ ∂
[Ô, K̂] = i px + py + pz (p2x + p2y + p2z ) = 2ih̄K̂ . (9.43)
2m ∂ px ∂ py ∂ pz
Note that the constant term in the operator Ô from re-ordering the posi-
tion and momentum operators does not affect the commutator. Using these
results, the expectation values of the commutators are related as
⟨[K̂, Ô]⟩ = −⟨[V (r̂), Ô]⟩ → −2ih̄⟨K̂⟩ = −inh̄⟨V (r̂)⟩ . (9.44)
That is,
n
⟨K̂⟩ = ⟨V (r̂)⟩ , (9.45)
2
exactly the classical virial theorem, by Ehrenfest’s theorem.
(c) What makes the operator Ô useful for this task is the following observation.
In position space, the operator contains a term of the form
∂ ∂
Ô ⊃ −ih̄x = −ih̄ . (9.46)
∂x ∂ log x
Thus, the job of the operator Ô is to count the powers of x that appear in
the potential, for example. The fact that the potential and kinetic energies
are related by a factor of n/2 is a reflection of the n powers of position in the
potential and 2 powers of momentum in the kinetic energy.
(d) On the ground state of the hydrogen atom, recall that the potential is
e2 1
V (r̂) = − . (9.47)
4π ϵ 0 r̂
Its expectation value on the ground state wavefunction is
Z ∞
e2 4 1 −2r/a0 e2 1 me e 4
⟨V (r̂)⟩ = − dr r2 e =− =− .
4π ϵ 0 a30 0 r 4π ϵ 0 a0 (4π ϵ 0 )2 h̄2
(9.48)
Also, recall that the energy of the ground state is
me e4
E0 = − , (9.49)
2(4π ϵ 0 )2 h̄2
and so the expectation value of the kinetic energy must be their difference
me e4
⟨K̂⟩ = E0 − ⟨V (r̂)⟩ = . (9.50)
2(4π ϵ 0 )2 h̄2
Thus, we see that
1
⟨K̂⟩ = |⟨V (r̂)⟩| , (9.51)
2
exactly as predicted classically for a 1/r potential and Ehrenfest’s theorem.
(e) For the harmonic oscillator, recall that its eigenenergies were

1
En = n + h̄ω , (9.52)
2
for n = 0, 1, 2, . . . . The potential of the harmonic oscillator is
mω 2 2 mω 2 h̄ 2 h̄ω † 2
V (x̂) = x̂ = â† + â = â + â . (9.53)
2 2 2mω 4
The expectation value on an energy eigenstate is then
h̄ω 2
⟨ψn |V (x̂)|ψn ⟩ = ⟨ψ0 |ân â† + â (â† )n |ψ0 ⟩ (9.54)
4n!
h̄ω
= ⟨ψ0 |ân â† â + ââ† (â† )n |ψ0 ⟩
4n!
h̄ω
= ⟨ψ0 |ân [â† , â] + 2ââ† (â† )n |ψ0 ⟩
4n!
h̄ω h̄ω
= ⟨ψ0 |ân+1 (â† )n+1 |ψ0 ⟩ − ⟨ψ0 |ân (â† )n |ψ0 ⟩
2n! 4n!
h̄ω h̄ω h̄ω 1
= (n + 1) − = n+ .
2 4 2 2
91 Exercises
This is exactly half of the energy eigenvalue, and so we find that, for the
harmonic oscillator,
⟨K̂⟩ = ⟨V (x̂)⟩ , (9.55)
on an energy eigenvalue, again, exactly as expected classically. This is

another way which exhibits that momentum and position in the harmonic
oscillator are treated effectively identically.
9.4 For this question, we will just focus on the 0 angular momentum states. Then, in
position space, the eigenvalue equation for the Hamiltonian is
2
h̄ 1 d h̄2 d 2 k
− − − ψ (r) = E ψ (r) , (9.56)
m r dr 2m dr2 rn
using the results from section 9.2 in the textbook for relating squared Cartesian
momentum to the radial momentum. Let’s take the same ansatz for the ground
state of this problem as we did with the hydrogen atom, with
ψ (r) ∝ rb e−r/a , (9.57)
for two constants a and b. The eigenvalue equation then implies that

h̄2 b(b + 1) 2(b + 1) 1 k
− − + − n =E. (9.58)
2m r2 ar a2 r
Immediately, it is clear that if n > 2, there is no way to make this relationship con-
sistent; the k/rn term dominates as r → 0. If n = 2, we must enforce a cancellation
of the two terms that scale like r−2 :
h̄2
− b(b + 1) = k . (9.59)
2m
However, we must also enforce that the single term that scales like r−1 vanishes by
itself, or that b = −1. These requirements are inconsistent, and so only for n < 2
can there possibly be bound states.
9.5 (a) In evaluating the ground state and first excited state of the hydrogen atom,
for ℓ = 0, we had constructed specific forms for their ansatze. In particular,
these wavefunctions took the form
ψ0 (r) ∝ e−r/a0 , ψ1 (r) ∝ (b0 + b1 r)e−r/a1 , (9.60)
for constants a0 , a1 , b0 , b1 . So, this suggests that for the nth energy eigen-
state, the prefactor of the exponential suppression at large r, is an nth order
polynomial:
n
ψn (r) = e−r/an ∑ ci rn , (9.61)
i=0
for some constants an , ci . In general, such a polynomial will have n nodes, n

roots, as we expect for the nth excited state.
(b) The operator that corresponds to the k component of the Laplace–Runge–

Lenz vector is
3
p̂i L̂ j − L̂i p̂ j e2 r̂k
Âk = ∑ ϵ i jk
2me
−
4π ϵ 0 r̂
. (9.62)
i, j=1
If this acts on a state of ℓ = 0 angular momentum, then L̂i |n, 0, 0⟩ = 0, so the

operator simplifies to
3
L̂i p̂ j e2 r̂k
Âk = − ∑ ϵ i jk −
2me 4π ϵ 0 r̂
. (9.63)
i, j=1
Recall that the commutator of the momentum and angular momentum is

3
[L̂i , p̂ j ] = ih̄ ∑ ϵ p̂k , (9.64)
k=1 i jk
so we can switch the operators as

3
L̂i p̂ j = [L̂i , p̂ j ] + p̂ j L̂i = p̂ j L̂i + ih̄ ∑ ϵ p̂l , (9.65)
l=1 i jl
and on an ℓ = 0 state only the commutator survives. The Laplace–Runge–

Lenz operator then reduces to
3
h̄ e2 r̂k h̄ e2 r̂k
Âk = −i ∑ ϵ ϵ p̂l −
2me i, j,l=1 i jk i jl 4π ϵ 0 r̂
= −i p̂k −
me 4π ϵ 0 r̂
, (9.66)
as expected.
(c) The first excited state wavefunction is again

1 r
ψ2,0,0 (r) = q 1− e−r/2a0 . (9.67)
8π a0
3 2a 0
Let’s first determine the ψ1,1,0 (⃗r) wavefunction. The operator Âz is, in
position space,

h̄ e2 ẑ h̄2 ∂ 1 z
Âz = −i p̂z − =− + . (9.68)
me 4π ϵ 0 r̂ me ∂ z a0 r
The derivative acting on the wavefunction is

∂ ∂r d z r 1
ψ2,0,0 (r) = ψ2,0,0 (r) = − e−r/2a0 . (9.69)
∂z ∂ z dr r 4a20 a0
Then, the wavefunction of interest is

z r 1 −r/2a0 1 z r
ψ1,1,0 (⃗r) ∝ Âz ψ2,0,0 (r) ∝ − e + 1 − e−r/2a0
r 4a20 a0 a0 r 2a0
z
∝ − e−r/2a0 . (9.70)
a0
93 Exercises
In these expressions, we have dropped overall factors because we only care

about proportionality and can fix normalization later.
This result also allows us to immediately write down the other wavefunctions
immediately:
x + iy −r/2a0
ψ1,1,1 (⃗r) ∝ (Âx + iÂy )ψ2,0,0 (r) ∝ e , (9.71)
a0
x − iy −r/2a0
ψ1,1,−1 (⃗r) ∝ (Âx − iÂy )ψ2,0,0 (r) ∝ e .
a0
8.6 The largest possible energy of an emitted photon from hydrogen would be pro-
duced from the transition from a very high energy level, with Ei ≈ 0 down to the
ground state with E f = −13.6 eV. The wavelength of a photon with energy 13.6
eV is
hc
λ= . (9.72)
E
The value of Planck’s constant times the speed of light is approximately
hc ≈ 1240 eV·nm , (9.73)
and so the wavelength of this most energetic light is

1240
λ∞→1 ≈ nm ≈ 91 nm , (9.74)
13.6
a bit too small to be observed by eye. However, if instead hydrogen transitioned
from a very high energy level to the first excited state, the wavelength would be
four times this value, because the first excited state has an absolute value of energy
that is four times less than the ground state. That is,
λ∞→2 ≈ 364 nm , (9.75)
which is on the bluer end of the visible spectrum. It is unlikely that the electron is
in an extremely high energy state because any errant radiation will knock it out
of its bound state. So, on the other end, the transition from the second to the first
excited state would produce a photon with wavelength of
1240
λ3→2 ≈ nm ≈ 656 nm , (9.76)
13.6
22
− 13.6
32
on the redder end of the spectrum. Indeed, hydrogen’s strongest spectral lines are
red, with many lines at the blue end of the spectrum, but much weaker brightness.
9.7 (a) Recall that the Biot-Savart law is
Z
⃗B(⃗r) = µ0 I d⃗ℓ × (⃗r − ⃗ℓ)
, (9.77)
4π |⃗r − ⃗ℓ|3
where ⃗ℓ is the position vector on the current, ⃗r is the position of interest,

and I is the current. For a charged particle traveling around a circle, we are
interested in the magnetic field at the center of the circle,⃗r = 0, to determine
the magnetic moment. As such, a point on the circle can be expressed as the
vector
ℓ = r(cos ω t, sin ω t) , (9.78)
where r is the radius and ω is the angular frequency of the orbit. The
differential distance is
d⃗ℓ = rω (− sin ω t, cos ω t) dt , (9.79)
which we note is orthogonal to the position vector ℓ. In each full orbit, an
electric charge e travels around, and therefore the current is
eω
I= . (9.80)
2π
With these results, the magnitude of the magnetic field at the center of the
loop is
I
µ0 eω r2 ω dt
|⃗B(⃗0)| = , (9.81)
8π 2 r3
and the integral runs over the total period T = 2π /ω of the orbit. Then, the
magnetic field is
µ0 eω 1
|⃗B(⃗0)| = . (9.82)
4π r
Now, for a particle of mass me traveling with angular frequency ω around a
circle of radius r, its angular momentum is
L = me r 2 ω . (9.83)
Then, the magnetic field, in terms of angular momentum, is
µ0 e 1 µ0 2
|⃗B(⃗0)| = L= |⃗m| , (9.84)
4π me r 3 4π r 3
where ⃗m is the magnetic dipole moment. Thus, the magnetic dipole moment
is
e
|⃗m| = L, (9.85)
2me
and the gyromagnetic moment is easily read off,
e
|γclass | = . (9.86)
2me
When the magnitude is removed, we pick up a minus sign because the
electron is negatively charged.
(b) We consider the external magnetic field along the z-axis as in the Zeeman
effect example, ⃗B = B0 ẑ. This magnetic field only affects the energy through
the z-component of angular momentum of the electron as it orbits the
proton. The new energy operator Û that represents this effect is
eB0
Û = −⃗m · ⃗B = L̂z , (9.87)
2me
95 Exercises
where we assume for now that the g-factor is 1. On an eigenstate of hydrogen

|n, ℓ, m⟩, this induces a change in the energy ∆E of
eB0
∆E|n, ℓ, m⟩ = Û|n, ℓ, m⟩ = m|n, ℓ, m⟩ . (9.88)
2me
With an external magnetic field, the energy levels of hydrogen are no longer
independent of m.
9.8 (a) Recall that the operators T̂ and Ŝ are
r r
1 me 1 me
T̂i = L̂i + − Âi , Ŝi = L̂i − − Âi , (9.89)
2 2E 2 2E
when acting on an energy eigenstate of hydrogen. When acting on the
ground state of hydrogen, |1, 0, 0⟩, this has 0 angular momentum, and so
is annihilated by L̂i . Further, the Laplace–Runge–Lenz operator Âi changes
the angular momentum of the state without affecting its energy. However,
the ground state of hydrogen is unique, with degeneracy 1, so there is no
other state that Âi can transform the ground state to. Therefore, the ground
state is also annihilated by Âi . Therefore, the ground state is annihilated by
both T̂i and Ŝi .
(b) Note that the action of the raising and lowering angular momentum oper-
ators on a state |n, ℓ, m⟩ produce states
L̂+ |n, ℓ, m⟩ ∝ |n, ℓ, m + 1⟩ , L̂− |n, ℓ, m⟩ ∝ |n, ℓ, m − 1⟩ , (9.90)
assuming that m < ℓ and m > −ℓ. The action of the raising and lowering
Laplace–Runge–Lenz operators on this state is
Â+ |n, ℓ, m⟩ ∝ |n − 1, ℓ + 1, m + 1⟩ , Â− |n, ℓ, m⟩ ∝ |n − 1, ℓ + 1, m − 1⟩ ,
(9.91)
assuming that ℓ < n. Therefore, the raising operators T̂+ and Ŝ+ produce a
linear combination of states:
T̂+ |n, ℓ, m⟩ = α |n, ℓ, m + 1⟩ + β |n − 1, ℓ + 1, m + 1⟩ , (9.92)
Ŝ+ |n, ℓ, m⟩ = γ |n, ℓ, m + 1⟩ + δ |n − 1, ℓ + 1, m + 1⟩ ,
for example, for some complex constants α , β , γ , δ . The lowering opera-
tors produce the corresponding linear combination of lowered states. What
makes the T̂ and Ŝ formalism powerful is that the states generated by, say,
T̂+ and Ŝ+ are orthogonal. Note that, for example, (Ŝ+ )† = S− , and so
†
Ŝ+ |n, ℓ, m⟩ T̂+ |n, ℓ, m⟩ = ⟨n, ℓ, m|Ŝ− T̂+ |n, ℓ, m⟩ (9.93)
= (⟨n, ℓ, m − 1|η + ⟨n + 1, ℓ − 1, m + 1|χ )
× (α |n, ℓ, m + 1⟩ + β |n − 1, ℓ + 1, m + 1⟩)
= 0,
where η , χ are other complex coefficients, because states with any different
quantum numbers are orthogonal.
(c) Doing the same exercise for T̂z and Ŝz , the states that are produced are
T̂z |n, ℓ, m⟩ = α |n, ℓ, m⟩ + β |n − 1, ℓ + 1, m⟩ , (9.94)
and Ŝz produces the orthogonal linear combination.
(d) To determine the form of the operators T̂i and Ŝi in position space, we
can first simplify the expression for the Laplace–Runge–Lenz operator Âk .
Noting that the angular momentum operator is
3
L̂k = ∑ ϵ r̂i p̂ j , (9.95)
i, j=1 i jk
we note that
3 3
∑ ϵ ( p̂i L̂ j − L̂i p̂ j ) = ∑ ϵ ϵ p̂i r̂l p̂m − ϵ r̂l p̂m p̂ j (9.96)
i, j=1 i jk i, j,l,m=1 i jk lm j lmi
3
= ∑ (δim δkl − δil δkm ) p̂i r̂l p̂m
i,l,m=1
3
+ ∑ δkm δ jl − δkl δ jm r̂l p̂m p̂ j
j,l,m=1
3
= ∑ p̂i r̂k p̂i − p̂i r̂i p̂k + r̂i p̂k p̂i − r̂k p̂2i
i=1
3
= ih̄ ∑ (1 − δik ) p̂k = −2ih̄ p̂k .
i=1
Therefore, the Laplace–Runge–Lenz operator is

h̄ e2 r̂k
Âk = −i p̂k − , (9.97)
me 4π ϵ 0 r̂
exactly what we had found for its action on the reduced space of ℓ = 0 states.
The prefactor of this operator in the T̂ or Ŝ operators is
r
me 4π ϵ 0 h̄
− = (n + ℓ) . (9.98)
2En+ℓ e2
Then, the T̂z operator, for example, in position space is
r
1 me
T̂z = L̂z + − Âz (9.99)
2 2E

1 ∂ ∂ ∂ z
= −ih̄x + ih̄y − (n + ℓ)h̄ a0 + .
2 ∂y ∂x ∂z r
10 Approximation Techniques
Exercises
10.1 (a) The complete Hamiltonian for this simple system is

E0 + ϵ ϵ
Ĥ = . (10.1)
ϵ E1 + ϵ
Its eigenvalues λ satisfy the characteristic equation:
(E0 + ϵ −λ )(E1 + ϵ −λ ) − ϵ 2 = 0 , (10.2)
or that
s
E0 + E1 E1 − E0 ϵ2
λ= ± 1−4 +ϵ . (10.3)
2 2 (E1 − E0 )2
Taylor expanding to first order in epsilon, we find the eigenvalues
λ = E0 + ϵ , E1 + ϵ . (10.4)
(b) From quantum mechanical perturbation theory, the first correction to the
energies is the expectation value of the perturbation Hamiltonian on the
unperturbed eigenstates. Because the unperturbed Hamiltonian is already
diagonal, the ground state is

1
|ψ0 ⟩ = . (10.5)
0
The first correction to the ground state is then
∆E0 = ⟨ψ0 |Ĥ ′ |ψ0 ⟩ = ϵ . (10.6)
We find the exact same correction for the first excited state energy by the
symmetry of the matrix Ĥ ′ . Thus indeed, to lowest order the correction to
the energies is to just increase them by ϵ , which agrees with the expansion
of the exact result from part (a).
√
(c) The Taylor series of the function 1 − x converges for |x| < 1 and so the
Taylor expansion of the square root in the exact result converges for
4ϵ2
1> , (10.7)
(E1 − E0 )2
97
98 10 Approximation Techniques
or for
E1 − E0
ϵ< . (10.8)
2
That is, the perturbation had better be less than the difference in energy
between the two states to honestly be a “perturbation.”
10.2 (a) The normalization of the wavefunction is
Z a
16a5 2
1 = N2 dx (a2 − x2 )2 = N . (10.9)
−a 15
That is,
r
15
N= . (10.10)
16a5
(b) In position space, the Hamiltonian is
h̄2 d 2 mω 2 2
Ĥ = − 2
+ x , (10.11)
2m dx 2
and then its action on this wavefunction is

h̄2 d 2 mω 2 2
Ĥ ψ (x; a) = − + x N(a2 − x2 ) (10.12)
2m dx2 2
2
h̄ mω 2 2 2
=N + x (a − x2 ) .
m 2
Then, the expectation value of the Hamiltonian on this wavefunction is
Z a 2
15 h̄ mω 2 2 2
⟨ψ |Ĥ|ψ ⟩ = dx (a − x )
2 2
+ x (a − x )
2
(10.13)
16a5 −a m 2
mω 2 a 2 5h̄2
= + .
14 4ma2
(c) Now, we want to determine the value of a that minimizes this expression.
Taking the derivative and demanding that it vanish, we find

d mω 2 a 2 5h̄2 mω 2 a 5h̄2
+ = 0 = − , (10.14)
da 14 4ma2 7 2ma3
which has the solution of
r
235 h̄
a = . (10.15)
2 mω
Inserting this into the expression for the expectation value, we find the
variational estimate of the ground state energy to be
m ω 2 a2 5h̄2
⟨ψ |Ĥ|ψ ⟩min = + q (10.16)
14 4ma2 a2 = 35 h̄
2 mω
r r ! r
1 35 5 2 5
= + h̄ω = h̄ω
14 2 4 35 14
≈ 0.5976 . . . h̄ω ,
99 Exercises
slightly larger than the exact result of h̄ω /2.

10.3 (a) Recall that the form of the ansatz we had made for the ground state of the
infinite square well was
ψ1 (x; α ) ∝ xα (1 − x)α . (10.17)
If this is orthogonal to the ansatz for the first excited state, we must find
Z 1 Z 1
dx ψ1 (x; α )ψ2 (x; β ) ∝ dx xα +β (1 − x)α +β (1 − 2x) = 0 , (10.18)
0 0
because the integrand is odd about the center of the well.

(b) The normalization of the first excited state wavefunction is
Z 1
1 = N2 dx x2β (1 − x)2β (1 − 2x)2 (10.19)
0
Z 1
= N2 dx x2β (1 − x)2β (1 − 4x + 4x2 )
0
= N 2 (B(2β + 1, 2β + 1) − 4B(2β + 2, 2β + 1) + 4B(2β + 3, 2β + 1)) ,
in terms of the Beta function. In terms of the Gamma function, we have
B(2β + 1, 2β + 1) − 4B(2β + 2, 2β + 1) + 4B(2β + 3, 2β + 1) (10.20)

Γ(2β Γ(2β + 2)Γ(2β + 1)
+ 1)2 Γ(2β + 3)Γ(2β + 1)
= −4 +4
Γ(4β + 2) Γ(4β + 3) Γ(4β + 4)

2β + 1 (2β + 1)(2β + 2) Γ(2β + 1)2
= 1−4 +4
4β + 2 (4β + 2)(4β + 3) Γ(4β + 2)
1 Γ(2β + 1)2
= .
4β + 3 Γ(4β + 2)
Therefore, the normalization factor is

p
(4β + 3)Γ(4β + 2)
N= . (10.21)
Γ(2β + 1)
(c) We need the action of the Hamiltonian on this ansatz wavefunction. In

position space, the Hamiltonian is just the second derivative, and so we
have
h̄2 d 2 β
Ĥ ψ2 (x; β ) = −N x (1 − x)β (1 − 2x) (10.22)
2m dx2
h̄2 β −2
=N β x (1 − x)β −2 (1 − 2x)(1 − β + 2(1 + 2β )(1 − x)x) .
2m
The expectation value of the Hamiltonian is then

⟨ψ2 |Ĥ|ψ2 ⟩ (10.23)
2 Z 1
h̄
= N2 β dx x2β −2 (1 − x)2β −2 (1 − 2x)2 (1 − β + 2(1 + 2β )(1 − x)x)
2m 0
h̄2 6β (3 + 4β )
= .
2m 2β − 1
(d) Minimizing over β by taking the derivative and setting it to 0, we find
d 6β (3 + 4β ) 6
=0= (8β 2 − 8β − 3) , (10.24)
d β 2β − 1 (2β − 1)2
or that (taking the larger root),
r
1 5
β= + ≈ 1.29 . (10.25)
2 8
Then, the minimum expectation value is
h̄2 6β (3 + 4β ) √ h̄2
⟨ψ2 |Ĥ|ψ2 ⟩min = q = (21 + 6 10) (10.26)
2m 2β − 1 β = 12 + 58 2m
h̄2
≈ 39.9736 . . ..
2m
The exact expression for the energy of the first excited state would be
h̄2 h̄2
E2 = 4π 2
≈ 39.4784 . . . , (10.27)
2m 2m
so our estimate is only very sightly above the true result.
(e) The precise answer to this problem depends on the form of the ansatz made
for the second excited state wavefunction, so we will not present a solution
here. However, the fundamental property that this state must satisfy is that
it is orthogonal to both the ground state and the first excited state. An
example of a form of a wavefunction that is orthogonal to the first two
energy eigenstates is
ψ2 (x; γ ) ≃ x(1 − x)(x − γ )(1 − γ − x) , (10.28)
where γ is a parameter that can be fixed to ensure orthogonality with the
ground state ansatz. Note also that this form of a wavefunction has two
nodes exclusive of the well’s boundaries, exactly as expected for the second
excited state.
10.5 (a) With the ansatz expression for the inverse Hamiltonian, the product that
we must demand produces the identity operator is
∞
1
Ĥ Ĥ −1 = I = h̄ω (1 + â† â)
h̄ω ∑ βn (â† )n ân (10.29)
n=0
∞ ∞
= ∑ βn (â† )n ân + ∑ βn â† â(â† )n ân
n=0 n=0
101 Exercises
∞ ∞
= ∑ βn (â† )n ân + ∑ βn â† (â† )n â + [â, (â† )n ] ân
n=0 n=0
∞
= ∑ (1 + n)βn (â† )n ân + βn (â† )n+1 ân+1 .
n=0
Then, demanding that all terms with positive powers of raising and lower-
ing operators cancel, we must enforce the recursion relation
(1 + n)βn + βn−1 = 0 , (10.30)
or that
βn−1
βn = − , (10.31)
1+n
with β0 = 1. The solution of this recursion relation is
(−1)n
βn = . (10.32)
(n + 1)!
Then, the inverse Hamiltonian of the harmonic oscillator is
∞
1 (−1)n
Ĥ −1 =
h̄ω ∑ (n + 1)! (â† )n ân . (10.33)
n=0
(b) The expectation value of the Hamiltonian on a coherent state is
⟨χ |Ĥ|χ ⟩ = h̄ω ⟨χ |1 + â† â|χ ⟩ = h̄ω (1 + |λ |2 ) , (10.34)
because ⟨χ |â† = ⟨χ |λ ∗ .
(c) We would like to now evaluate the ratio of expectation values
⟨χ |Ĥ −1 |χ ⟩
. (10.35)
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩
Let’s first evaluate the numerator expectation value. We have
∞ ∞
1 (−1)n 1 (−1)n
⟨χ |Ĥ −1 |χ ⟩ =
h̄ω ∑ (n + 1)! ⟨χ |(â† )n ân |χ ⟩ = h̄ω ∑ (n + 1)! |λ |2n
n=0 n=0
1 − e−|λ |
2
1
= . (10.36)
h̄ω |λ |2
Now, let’s evaluate the denominator with the hint. We have
∞ ∞
1 1
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩ = ∑ ⟨χ |Ĥ −1 |ψn ⟩⟨ψn |Ĥ −1 |χ ⟩ = (h̄ω )2 ∑ (n + 1)2 |⟨χ |ψn ⟩|2 .
n=0 n=0
(10.37)
Now, the normalized energy eigenstate is
(â† )n
|ψn ⟩ = √ |ψ0 ⟩ , (10.38)
n!
and the normalized coherent state is

|λ |2
|χ ⟩ = e− eλ â |ψ0 ⟩ ,
†
2 (10.39)
where |ψ0 ⟩ is the ground state of the harmonic oscillator. We can now
evaluate their inner product straightforwardly as
|λ |2 |λ |2 |λ |2
e− 2 e− 2 λ n e− 2
⟨ψn |χ ⟩ = √ ⟨ψ0 |ân eλ â |ψ0 ⟩ = √ λ n ⟨ψ0 |eλ â |ψ0 ⟩ = √
† †
.
n! n! n!
(10.40)
Then, the denominator becomes

∞
e−|λ | ∞
2
1 1 |λ |2n
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩ = ∑
(h̄ω ) n=0 (n + 1)
2 2
|⟨ χ |ψ n ⟩|2
= ∑
(h̄ω ) n=0 n!(n + 1)2
2
e−|λ | ∞
2
|λ |2n
= ∑
(h̄ω ) n=0 (n + 1)!(n + 1)
2
. (10.41)
The factor of n + 1 in the denominator of the terms in the series is a bit

awkward, but note that
Z ∞ ∞
1 |λ |2n |λ |2n
|λ |2
d|λ |2 ∑ (n + 1)! = ∑ (n + 1)!(n + 1) , (10.42)
n=0 n=0
and the sum in this new integrand is

∞
e|λ | − 1
2
|λ |2n
∑ (n + 1)! = |λ |2 . (10.43)
n=0
Thus, the expectation value in the denominator can be expressed as

Z
e−|λ | 1 e|λ | − 1
2 2
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩ = d|λ |2 . (10.44)

(h̄ω )2 |λ |2 |λ |2
Putting this together, we then find that

2
1 1−e−|λ |
⟨χ |Ĥ −1 |χ ⟩ e|λ | − 1
2
h̄ω |λ |2
E0 ≤ = R = h̄ω R ,
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩ 2
e−|λ | 1 |λ |2
d|λ |2 e |λ |−1
|λ |2
dx e x−1
x
(h̄ω )2 |λ |2 2 0
(10.45)
as claimed.
(d) To determine the λ → 0 limit, we can use l’Hôpital’s rule. The derivative of
the numerator evaluated at |λ |2 = 0 is
d |λ |2
e − 1 = 1. (10.46)
d|λ |2 |λ |2 =0
103 Exercises
The derivative of the denominator, from the fundamental theorem of

calculus, is
Z |λ |2
e|λ | − 1
2
d ex − 1
dx = lim = 1. (10.47)
d|λ |2 0 x |λ |2 →0 |λ |
2
|λ |2 =0
Therefore, for λ = 0 corresponding to the true ground state of the harmonic

oscillator, we find that the estimate of the energy is h̄ω , the exact result. If
instead λ = 1, the estimate is
⟨χ |Ĥ −1 |χ ⟩ e−1
= h̄ω R 1 ex −1 ≈ 1.3038 h̄ω . (10.48)
⟨χ |Ĥ −1 Ĥ −1 |χ ⟩ 0 dx x
From part (b), the initial expectation value of the coherent state with λ = 1
would have been
⟨χ |Ĥ|χ ⟩ = h̄ω (1 + |λ |2 ) = 2h̄ω . (10.49)
So, indeed, we see that the power method improved the lower bound
estimate of the ground state of the harmonic oscillator.
10.5 (a) From the formalism of perturbation theory, the first-order correction to
the nth energy eigenstate is the expectation value
∆En = ⟨ψn |λ x̂4 |ψn ⟩ , (10.50)
where |ψn ⟩ is the nth energy eigenstate of the harmonic oscillator. Using
the raising and lowering operators, we can express this expectation value
as
h̄2
∆En = λ ⟨ψn |(â + â† )4 |ψn ⟩ . (10.51)
4m2 ω 2
Now, we can evaluate this by noting that the action of the raising and
lowering operators is
(â† )n+1 √
â† |ψn ⟩ = √ |ψ0 ⟩ = n + 1|ψn+1 ⟩ , (10.52)
n!
d (â† )n √
â|ψn ⟩ = † √ |ψ0 ⟩ = n|ψn−1 ⟩ .
d â n!
Because the expectation value is a diagonal element of x̂4 , there will only
be non-zero contributions when the number of raising and lowering oper-
ators are equal. That is, contributions like â(â† )3 necessarily vanish on this
expectation value. So, we only need to consider products of raising and
lowering operators that arise from expanding the product that have the
same number of both:
(â + â† )4 ⊃ â2 (â† )2 + ââ† ââ† + â(â† )2 â + â† ââ† â + â† (â)2 â† + (â† )2 â2 .
(10.53)
Further, note that
â† â|ψn ⟩ = n|ψn ⟩ , ââ† |ψn ⟩ = (n + 1)|ψn ⟩ . (10.54)
With these results, the matrix element of four of the terms is easily
evaluated:
⟨ψn |ââ† ââ† + â(â† )2 â + â† ââ† â + â† (â)2 â† |ψn ⟩ = (2n + 1)⟨ψn |ââ† + â† â|ψn ⟩
= (2n + 1)2 . (10.55)
The two other matrix elements are

√ √
⟨ψn |â2 (â† )2 + (â† )2 â2 |ψn ⟩ = ⟨ψn | n + 1â2 â† |ψn+1 ⟩ + ⟨ψn | n(â† )2 â|ψn−1 ⟩
(10.56)
= (n + 1)⟨ψn+1 |ââ† |ψn+1 ⟩ + n⟨ψn−1 |â† â|ψn−1 ⟩
= (n + 1)(n + 2) + n(n − 1) = 2n2 + 2n + 2 .
Putting this all together, we have

h̄2 h̄2
⟨ψn |x̂4 |ψn ⟩ = ⟨ψ n |(â + â† 4
) |ψ n ⟩ = (2n + 1)2 + 2n2 + 2n + 2
4m ω
2 2 4m ω
2 2
h̄2 2
= 6n + 6n + 2 . (10.57)
4m2 ω 2
Therefore, the correction to the nth energy eigenstate from the anharmonic
oscillator is
h̄2 2
∆En = λ 6n + 6n + 2 . (10.58)
4m2 ω 2
(b) With λ < 0, there will be some energy level n for which En + ∆En < 0.
Assuming that |λ | is very small, this n will be very large, so we will approx-
imate the harmonic oscillator energy level as En ≈ nh̄ω , and the correction
as
3h̄2 n2
∆En ≈ λ . (10.59)
2m2 ω 2
Then, if ∆En ≈ −En , we have
3h̄2 n2
λ ≈ nh̄ω , (10.60)
2m2 ω 2
or that
2m2 ω 3
n≈ . (10.61)
3h̄λ
(c) The correction to the ground state wavefunction |ψ0 ⟩ is
∞
λ ⟨ψn |x̂4 |ψ0 ⟩
|∆ψ0 ⟩ = −
h̄ω ∑ n
|ψn ⟩ . (10.62)
n=1
105 Exercises
Due to the form of the quartic position operator x̂4 in terms of raising and
lowering operators, only the states n = 2, 4 can contribute to the sum. Thus,
this simplifies to
∞
λ ⟨ψn |x̂4 |ψ0 ⟩
|∆ψ0 ⟩ = −
h̄ω ∑ n
|ψn ⟩ (10.63)
n=1

λ h̄ ⟨ψ4 |(â† )4 |ψ0 ⟩
=− 2 3 |ψ4 ⟩
4m ω 4

⟨ψ2 |â(â† )3 + â† â(â† )2 + (â† )2 ââ† |ψ0 ⟩
+ |ψ2 ⟩
2
√ !
λ h̄ 6 √
=− 2 3 |ψ4 ⟩ + 3 2 |ψ2 ⟩ .
4m ω 2
(d) The expectation value of a Hermitian operator T̂ to first order in λ on the

anharmonic oscillator’s ground state is
(|ψ0 ⟩ + |∆ψ0 ⟩)† T̂ (|ψ0 ⟩ + |∆ψ0 ⟩) = ⟨ψ0 |T̂ |ψ0 ⟩ + ⟨ψ0 |T̂ |∆ψ0 ⟩ + ⟨∆ψ0 |T̂ |ψ0 ⟩ .
(10.64)
With T̂ = x̂ and the modification only involving |ψ2 ⟩ and |ψ4 ⟩, the expec-
tation value of x̂ on the modified state still vanishes. So, the variance of the
ground state is still determined by the second moment: σx2 = ⟨ψ |x̂2 |ψ ⟩. Fur-
ther, the contribution from the state |ψ4 ⟩ is 0 on the expectation value of
x̂2 because x̂2 can only relate energy eigenstates that differ by 2h̄ω . There-
fore, the change in the variance of the ground state from the anharmonic
oscillator is
√
3 2λ h̄
∆σx = ⟨ψ0 |T̂ |∆ψ0 ⟩ + ⟨∆ψ0 |T̂ |ψ0 ⟩ = − 2 3 ⟨ψ2 |x̂2 |ψ0 ⟩ + ⟨ψ0 |x̂2 |ψ2 ⟩
2
√ 4m ω
3 2λ h̄2
=− ⟨ψ2 |(â + â† )2 |ψ0 ⟩ + ⟨ψ0 |(â + â† )2 |ψ2 ⟩
√ ω
8m 3 4
3 2λ h̄2 3λ h̄2
=− ⟨ψ 2 |(â † 2
) |ψ0 ⟩ + ⟨ψ 0 |â2
|ψ2 ⟩ = − . (10.65)
8m3 ω 4 2m3 ω 4
This is negative, and so the ground state wavefunction of the anharmonic
oscillator is narrower than the ground state of the harmonic oscillator. This
makes sense because the anharmonic oscillator’s potential is narrower than
the harmonic oscillator (see Fig. 10.7).
10.6 The solutions to this problem highly vary depending on what system the stu-
dent chooses to study. We refer to T. Sulejmanpasic and M. Ünsal, “Aspects
of perturbation theory in quantum mechanics: The BenderWu Mathematica
® package,” Comput. Phys. Commun. 228, 273–289 (2018) [arXiv:1608.08256
[hep-th]], for their worked examples and instructions to use the program.
10.7 (a) The first thing we need to do for the Bohr-Sommerfeld condition is to
establish what the classical turning points are. The classical turning points
for a given energy En correspond to the positions where the kinetic energy
is 0, so the total energy is just potential:
mω 2 2
En = x , (10.66)
2
or that
r
2En
xmin,max = ∓ . (10.67)
mω 2
Then, the Bohr-Sommerfeld condition for the harmonic oscillator is
q s
Z 2En
1 mω 2 ′ mω 2 ′2
q dx 2m En − x (10.68)
h̄ − 2En2 2
mω
q s
Z 2En
2 mω 2 ′ mω 2 ′2
= dx 2m En − x = nπ .
h̄ 0 2
Let’s now change variables to

r
2En
x′ = sin θ , (10.69)
mω 2
so that
r
′ 2En
dx = cos θ d θ , (10.70)
mω 2
and the integral becomes
q s
Z 2En
2 mω 2 ′ mω 2 ′2
dx 2m En − x (10.71)
h̄ 0 2
Z p
4En π /2
= d θ cos θ 1 − sin2 θ
h̄ω 0
Z Z
4En π /2 2En π /2
= d θ cos2 θ = d θ [1 + cos(2θ )]
h̄ω 0 h̄ω 0
π En
= = nπ .
h̄ω
That is,
En = nh̄ω , (10.72)
which differs from the correct energy eigenvalues of the harmonic oscillator
by the ground state energy, h̄ω /2.
(b) With the Maslov correction, all we need to change is the value of the
quantized integral, to

π En 1
= n+ π, (10.73)
h̄ω 2
107 Exercises
or that the energy levels are

1
En = n + h̄ω , (10.74)
2
exactly the correct result.
(c) Once the hydrogen atom Hamiltonian is expressed in terms of the radius,
we effectively must impose that r > 0, which makes r = 0 a hard boundary,
like that of the infinite square well. No Maslov correction is needed for
the infinite square well. Further, as r → ∞, the potential actually vanishes,
and does not diverge like the harmonic oscillator. This significantly softens
the suppression of the wavefunction at large radii compared to the infinite
square well, again rendering a Maslov correction irrelevant.
(d) With a potential of the form V (x) = k|x|α , the classical turning points
correspond to
En = k|x|α , (10.75)
or that
1/α
En
xmin,max = ∓ . (10.76)
k
Then, the quantization condition is
Z ( En )1/α
2 p 1
dx′
k
2m(En − kx′α ) = n + π, (10.77)
h̄ 0 2
by the symmetry about x = 0. Let’s now change variables to
kx′α = En u , (10.78)
and so
1/α
En
x′ = u1/α . (10.79)
k
The integration measure is
1/α
1′ En
u α −1 du ,
1
dx = (10.80)
α k
so the quantization becomes
Z ( En )1/α √ Z
2 p
′ 2 2mEn En 1/α 1
du u α −1 (1 − u)1/2
k 1
dx 2m(En − kx′α ) =
h̄ 0 h̄ α k 0
√ 1/α
2 2mEn En Γ(1/α )Γ(3/2)
=
h̄ α k Γ(1/α + 3/2)

1
= n+ π, (10.81)
2
where we have used the fact that the integral that remains is in the form of
the Euler Beta function. First, for general α , the energies scale with n for
large n like
1+ 1
En2 α
∝ n, (10.82)
or that
2α
En ∝ n 2+α . (10.83)
Note that if α = 2 we indeed reproduce the linear scaling of energies with

level n. If instead α → ∞, we first note that the Gamma function has the
limit
lim Γ(1/α ) = α , (10.84)

α →∞
and so
Γ(1/α )Γ(3/2)
lim = 1. (10.85)
α →∞ α Γ(1/α + 23 )
In this limit, the quantization condition becomes
2p
2mEn = nπ , (10.86)
h̄
and we have turned off the Maslov correction because α → ∞ is a singular
limit. The solution of this equation is
n2 π 2 h̄2
En = . (10.87)
2m · 22
As α → ∞, the potential diverges beyond |x| = 1 and is 0 for −1 < x <
1. Thus, this is an infinite square well with width a = 2, and the Bohr-
Sommerfeld quantization condition predicts that exactly correctly.
10.8 (a) The power method would predict that ground state is approximately
proportional to
|1⟩ ≃ Ĥ N |χ ⟩ . (10.88)
To estimate the ground state energy, we take the expectation value of the
Hamiltonian on this state and then ensure it is properly normalized. So,
the ground state energy is bounded as
⟨χ |Ĥ N Ĥ Ĥ N |χ ⟩ ⟨χ |Ĥ 2N+1 |χ ⟩
E1 ≤ = . (10.89)
⟨χ |Ĥ N Ĥ N |χ ⟩ ⟨χ |Ĥ 2N |χ ⟩
From the given form of the state |χ ⟩, the action of the Hamiltonian is
∞ 2N+1 ∞
me e4 βn
Ĥ 2N+1 |χ ⟩ = ∑ β Ĥ 2N+1 |n⟩ = − ∑ n4N+2 |n⟩ .
n=1 2(4 π ϵ 0 )2 h̄2
n=1
(10.90)
109 Exercises
Therefore, the upper bound of the ground state energy is

∞ |βn |2
me e 4 ∑n=1 n4N+2
E1 ≤ − . (10.91)
2 2 ∞ |βn |2
2(4π ϵ 0 ) h̄ ∑n=1 4N n
(b) Let’s now consider the state

∞
1
| χ ⟩ = β1 ∑ n |n⟩ , (10.92)
n=1
where β0 is a normalization constant. Its value can be determined by

demanding that
∞
1 π2
1 = ⟨χ |χ ⟩ = |β1 |2 ∑ n2 = 6
|β1 |2 . (10.93)
n=1
Thus, the properly normalized state is

√ ∞
6 1
|χ ⟩ = ∑
π n=1 n
|n⟩ . (10.94)
The expectation value of the Hamiltonian on this state is then

me e4 6 ∞ 1
⟨χ |Ĥ|χ ⟩ = − ∑ .
2(4π ϵ 0 )2 h̄2 π 2 n=1 n4
(10.95)
The value of the sum that remains is π 4 /90 and so the expectation value is

me e4 π2 me e 4
⟨χ |Ĥ|χ ⟩ = − ≈ 0.65794 − (10.96)
2(4π ϵ 0 )2 h̄2 15 2(4π ϵ 0 )2 h̄2
which is indeed larger than the ground state energy, as expected.
(c) From the result of part (a), the estimate of the ground state energy after N
applications of the power method on this state is
me e 4 ∑∞ 1
n=1 n4N+4 me e 4 ζ (4N + 4)
E1 ≤ − 2 ∞
= − 2 ζ (4N + 2)
. (10.97)
2(4π ϵ 0 ) h̄ ∑n=1 n4N+2
2 1
2(4π ϵ 0 ) h̄
2
(d) As N → ∞, the difference between 4N + 4 and 4N + 2 is negligible, and

the ratio of the zeta function terms becomes unity, thus reproducing the
expected ground state energy. For N = 1, the ratio of the zeta function terms
is
ζ (8)
≈ 0.98696 . (10.98)
ζ (6)
For N = 2, the ratio is
ζ (12)
≈ 0.999252 , (10.99)
ζ (10)
well within one part in a thousand of the true ground state energy.
10.9 (a) Note that we can write the kinetic energy as

1/2
|⃗p|2
K = me c 2 1 + 2 2 − me c2 (10.100)
me c
1/2
|⃗p|2 |⃗p|4
= me c 2 1 + 2 2 − 4 4 + · · · − me c2
2me c 8me c
|⃗p|2 |⃗p|4
= − 3 2 +··· .
2me 8me c
This can then be trivially promoted to a Hermitian operator on Hilbert
space, which takes the form in position space of
2 2 2
h̄2 ∂ ∂2 ∂2 h̄4 ∂ ∂2 ∂2
K̂ = − + + − + + +··· .
2me ∂ x2 ∂ z2 ∂ z2 8m3e c2 ∂ x2 ∂ z2 ∂ z2
(10.101)
(b) For this and the next part of this problem, we are going to cheat a little bit.
First, we note that on the ground state, the speed of the electron is about
v ≃ α c, less than 1/100th the speed of light. So, the relativistic correction
will be very small and so we can treat it as a perturbation, even in the power
method. To see how this works, we will expand the power method approxi-
mation to first order in the relativistic correction to the Hamiltonian. That
is,
⟨ψ |(Ĥ0 + Ĥrel )3 |ψ ⟩
E0,rel ≃ (10.102)
⟨ψ |(Ĥ0 + Ĥrel )2 |ψ ⟩
⟨ψ |Ĥ03 + Ĥ02 Ĥrel + Ĥ0 Ĥrel Ĥ0 + Ĥrel Ĥ02 + · · · |ψ ⟩
= ,
⟨ψ |Ĥ02 + Ĥ0 Ĥrel + Ĥrel Ĥ0 + · · · |ψ ⟩
where the ellipses hide operators that are quadratic or higher order in Ĥrel .
Using the fact that |ψ ⟩ is the ground state of the unperturbed Hamiltonian,
Ĥ0 |ψ ⟩ = E0 |ψ ⟩, this ratio simplifies
E03 + 3E02 ⟨ψ |Ĥrel |ψ ⟩ + · · ·
E0,rel ≃ ≃ E0 + ⟨ψ |Ĥrel |ψ ⟩ + · · · , (10.103)
E02 + 2E0 ⟨ψ |Ĥrel |ψ ⟩ + · · ·
from Taylor expanding the denominator. That is, the lowest-order correc-
tion to the ground state energy due to relativistic effects is the expectation
value of the relativistic Hamiltonian on the unperturbed ground state. This
is exactly what we derived from perturbation theory, but here established
it from the power method.
With this result, we can then move to calculating the expectation value.
Recall that the ground state wavefunction of the hydrogen atom exclusively
has radial dependence. On a function f of radius r, we had shown in the
previous chapter that
2
∂ ∂2 ∂2 2 d f d2 f
+ + f (r) = + . (10.104)
∂x 2 ∂z2 ∂z2 r dr dr2
111 Exercises
Now, the expectation value of the relativistic correction can be expressed

as
! !
1 |⃗p|
ˆ2 |⃗p|
ˆ2
⟨ψ |Ĥrel |ψ ⟩ = − ⟨ψ | |ψ ⟩ . (10.105)
2me c2 2me 2me
In the expectation value is just the unperturbed kinetic energy, and on the
unperturbed ground state, we can use the fact that
|⃗p|
ˆ2 e2 1
= E0 + . (10.106)
2me 4π ϵ 0 r̂
Then, the expectation value of the relativistic correction is
2
1 e2 1
⟨ψ |Ĥrel |ψ ⟩ = − ⟨ψ | E0 + |ψ ⟩ (10.107)
2me c2 4π ϵ 0 r̂

1 e2 1 e4 1
=− E 2
+ E0 ⟨ψ | |ψ ⟩ + ⟨ψ | |ψ ⟩ .
2me c2 0
2π ϵ 0 r̂ (4π ϵ 0 )2 r̂2
On the ground state, the expectation values are
Z
1 4 ∞ 1
⟨ψ | |ψ ⟩ = 3
dr r e−2r/a0 = , (10.108)
r̂ a0 0 a0
Z
1 4 ∞ 2
⟨ψ | 2 |ψ ⟩ = dr e−2r/a0 = 2 .
r̂ a30 0 a0
Then, the expectation value of the relativistic correction is

1 e2 2e4
⟨ψ |Ĥrel |ψ ⟩ = − E 2
+ E0 + (10.109)
2me c2 0
2π ϵ 0 a0 (4π ϵ 0 )2 a20
E02 5E02
=− (1 − 4 + 8) = − ,
2me c2 2me c2
where I have used the expression for the ground state energy.
(c) As observed in the previous part, the lowest-order expansion of the power
method exactly reproduces the first-order correction to the ground state
energy from perturbation theory.
11 The Path Integral
Exercises
11.1 (a) For a free particle that travels a distance x in time T , its velocity is
x
v= . (11.1)
T
Further, the energy E of the particle in terms of momentum is
p2
E= , (11.2)
2m
and so the difference is
p2 p2 p2 p2
px − ET = pvT − T = T− T= T, (11.3)
2m m 2m 2m
which is indeed the kinetic energy times the elapsed time; the classical
action of a free particle.
(b) In terms of the angular velocity ω , the elapsed angle over time T is
θ = ωT , (11.4)
and the angular momentum L is
L = Iω , (11.5)
for moment of inertia I. The rotational kinetic energy E is

1 L2
E = Iω 2 = , (11.6)
2 2I
and so the difference is
L2 L2 L2 L2
Lθ − ET = Lω T − T = T− T = T, (11.7)
2I I 2I 2I
again the rotational kinetic energy times the elapsed time, or the classical
action for a particle experiencing no torque.
(c) Assuming that the total energy is a constant value E, including time
dependence is trivial:
Zx
i p ET
ψ (x,t) ≈ exp dx′ 2m(E −V (x′ )) − i ψ (x0 ) , (11.8)
h̄ x0 h̄
112
113 Exercises
where T is the total elapsed time. We can change the integration variable
to the time x′ = vt = mp t, where v is the velocity and p is the momentum of
the particle. In general, the momentum varies with position or time, but, as
with the original WKB approximation, we assume that variation is small,
so the integration measure is
p
dx′ = dt . (11.9)
m
Then, the wavefunction is approximately
ZT
i 2m(E −V ) ET
ψ (x,t) ≈ exp dt −i ψ (x0 ) , (11.10)
h̄ 0 m h̄
p
because the momentum with fixed energy is p = 2m(E −V ). This can be
rearranged to:
ZT
i
ψ (x,t) ≈ exp dt [2(E −V ) − E] ψ (x0 ) (11.11)
h̄ 0
ZT
i
= exp dt (E − 2V ) ψ (x0 )
h̄ 0
ZT
i
= exp dt (K −V ) ψ (x0 ) ,
h̄ 0
where we have used that the total energy is the sum of the kinetic and poten-
tial energies: E = K +V . Thus, what appears in the exponent is indeed the
classical action, the time integral of the kinetic minus the potential energy.
The general exact expression for the wavefunction at any later time requires
the infinity of integrals of the path integral, but the WKB approximation
assumes that the potential varies slowly enough that those infinity of inte-
grals just produce an overall normalization, and only the first “step” of the
path integral is needed.
11.2 (a) The Euler-Lagrange equation from a Lagrangian L is
∂L d ∂L
− = 0. (11.12)
∂ x dt ∂ ẋ
For the harmonic oscillator’s Lagrangian, the necessary derivatives are
∂L d ∂L d
= −mω 2 x , − = − mẋ = −mẍ . (11.13)
∂x dt ∂ ẋ dt
Therefore, the Euler-Lagrange equations are
−mω 2 x − mẍ = 0 , (11.14)
as claimed.
(b) Simplifying the equations of motion, we have
ẍ + ω 2 x = 0 , (11.15)
for which the solutions are linear combinations of sines and cosines:
x(t) = a cos(ω t) + b sin(ω t) , (11.16)
114 11 The Path Integral
where a, b are integration constants. At time t = 0, the position is x(0) = xi

and so a = xi . At t = T , the position is x f , and so the constant b is found
from
x f = xi cos(ω T ) + b sin(ω T ) , (11.17)
and so
x f − xi cos(ω T )
b= . (11.18)
sin(ω T )
Thus, the general solution for the trajectory that satisfies the boundary
conditions is
x(t) = xi cos(ω t) + sin(ω t) . (11.19)
sin(ω T )
(c) The classical action is
Z
m T mZ T d
S[x] = dt ẋ − ω x =
2 2 2
dt (xẋ) − xẍ − ω x
2 2
(11.20)
2 0 2 0 dt
Z
m T d
= dt (xẋ) − x(ẍ + ω 2 x) .
2 0 dt
In these expressions, we have used integration by parts to move one of the
time derivatives. Then, ẍ + ω 2 x = 0 by the equations of motion and so the
classical action is
Z T
m d m
S[x] = dt (xẋ) = (x(T )ẋ(T ) − x(0)ẋ(0)) . (11.21)
2 0 dt 2
Note that the first time derivative of the trajectory is
ẋ(t) = −ω xi sin(ω t) + ω cos(ω t) . (11.22)
sin(ω T )
At t = 0 and t = T , we have
ẋ(0) = ω , (11.23)
sin(ω T )
ẋ(T ) = −ω xi sin(ω T ) + ω cos(ω T ) .
sin(ω T )
Then, the value of the classical action is

m x f −xi cos(ω T ) x f −xi cos(ω T )
S[x] = −ω xi x f sin(ω T )+ω x f cos(ω T )−ω xi
2 sin(ω T ) sin(ω T )

m x f −xi cos(ω T ) x f −xi cos(ω T )
= −ω xi x f sin(ω T )+ω x f cos(ω T )−ω xi
2 sin(ω T ) sin(ω T )
mω (x f +xi ) cos(ω T )−xi x f (1+ sin (ω T )+ cos (ω T ))
2 2 2 2
=
2 sin(ω T )
mω (x f +xi ) cos(ω T )−2xi x f
2 2
= . (11.24)
2 sin(ω T )
115 Exercises
11.3 (a) Note that we can rearrange the eigenvalue equation to

2
d
+ (ω − λ ) fλ (t) = 0 ,
2
(11.25)
dt 2
for which the solutions are
p p
fλ (t) = a cos(t ω 2 − λ ) + b sin(t ω 2 − λ ) , (11.26)
for integration constants a, b. Demanding that fλ (0) = 0 just sets a = 0,
while fλ (T ) = 0 enforces that
p
T ω 2 − λ = nπ , (11.27)
for n = 1, 2, 3, . . . . Rearranging, we have

n2 π 2 n2 π 2 ω 2T 2
ω2 − 2 = − 2 1 − 2 2 = λ . (11.28)
T T n π
(b) Note with the eigenvalues, we can evaluate the determinant as the product
of all eigenvalues. This is
2 ∞
d n2 π 2 ω 2T 2
det + ω = det A = ∏ − 2
2
1− 2 2 (11.29)
dt 2 n=1 T n π
2 2
sin(ω T ) ∞ n π
= ∏
ω T n=1
− 2
T
.
Interestingly, this still involves an infinite product of increasingly divergent

factors.
(c) This can be understood when comparing to the result we found in the
methods introduced in section 11.5.4 of the textbook. We had found the
determinant of the operator A to be

N N sin(ω T ) sin(ω T ) N3
det A = lim (−1) = lim − 2 . (11.30)
N→∞ ∆t 2 ω T ω T N→∞ T
If these two expressions are to agree, the remaining infinite product must
be equal to the limit:
∞
n2 π 2 N3
∏ −
T2
= lim
N→∞
−
T2
. (11.31)
n=1
Of course, the equality is only formal, because both sides are infinite.
11.4 (a) The harmonic oscillator path integral is
r s
S[x
i class
] m ωT
Z = e h̄ . (11.32)
2π ih̄T sin(ω T )
Taking the ω → 0 limit of the right-most square-root factor produces
s r
ωT ωT
lim = lim = 1. (11.33)
ω →0 sin(ω T ) ω →0 ω T
Thus, the free-particle path integral is

r
i
S[xclass ] m
Z=e h̄ , (11.34)
2π ih̄T
in agreement with the example.
(b) The path integral expression from equation 11.79 is
∞
∑ e−i
T En
Z= h̄ ψn (x f )ψn∗ (xi ) . (11.35)
n=1
Passing this to continuous momentum eigenstates, the free particle path

integral becomes
Z ∞ (x f −xi )
dp
e−i 2mh̄ p ei
T 2 p
Z= h̄ , (11.36)
−∞ 2π h̄
using the classical energy for a free particle, E = p2 /2m. Note the factor of
2π h̄ in the measure; this is to ensure that the dimensions of the path integral
are still inverse distance. Now, we will make the change of variables to
T 2
u2 = i p , (11.37)
2mh̄
or that
r
2mh̄
p= u. (11.38)
iT
The path integral then becomes
r Z ∞ q
m 1 −u2 +i(x f −xi ) ih̄T
2m u
Z= √ du e (11.39)
2π ih̄T π −∞
r Z ∞ q
m 1 −u2 +i(x f −xi ) ih̄T
2m u−(x −x )2 im +(x −x )2 im
= √ du e f i 2h̄T f i 2h̄T
2π ih̄T π −∞
r 2
m (x f −xi ) Z
∞ 2r
m ei 2 h̄T m (x f −xi ) m
du e−u = ei 2 h̄T
2
= √ ,
2π ih̄T π −∞ 2π ih̄T
where we completed the square in the second line. This is exactly the free
particle path integral that we calculated in the example.
11.5 Note that the path integral for the three-dimensional harmonic oscillator takes
the form
Z Z
i m 2 m 2 m 2 mω 2 2 mω 2 2 mω 2 2
Z3D = [dx][dy][dz] exp dt ẋ + ẏ + ż − x − y − z
h̄ 2 2 2 2 2 2
(11.40)
Z Z Z Z
i m 2 mω 2 2 i m 2 mω 2 2
= [dx] exp dt ẋ − x [dy] exp dt ẏ − y
h̄ 2 2 h̄ 2 2
Z Z
i m 2 mω 2 2
× [dz] exp dt ż − z
h̄ 2 2
117 Exercises
Z Z 3
i m 2 mω 2 2
= [dx] exp dt ẋ − x
h̄ 2 2
3
= Z1D .
That is, the path integral of the three-dimensional harmonic oscillator is just
the cube of the path integral for the one-dimensional harmonic oscillator, by
the linearity of integration.
11.6 (a) For the free-particle action, we need the first time derivative of the trajec-
tory. For this trajectory, we have
2 ′ 2
ẋ = (x − xi )Θ(T /2 − t) + (x f − x′ )Θ(T − t)Θ(t − T /2) , (11.41)
T T
where the Heaviside theta function Θ(x) is 1 if x > 0 and 0 if x < 0. The
action of this trajectory is therefore
Z
m T 4 ′ 4 ′ 2
S[x] = dt (x −x i )2
Θ(T /2−t) + (x f −x ) Θ(T −t)Θ(t−T /2)
2 0 T2 T2
m
= (x′ − xi )2 + (x f − x′ )2 , (11.42)
T
because the integrand is time independent (except for the Heaviside func-
tions). This is just a parabola in the intermediate point x′ with a minimum
where
d
′
(x′ − xi )2 + (x f − x′ )2 = 0 = 2(x′ − xi ) − 2(x f − x′ ) , (11.43)
dx
or that
x f + xi
x′ = . (11.44)
2
At this point, the trajectory of the particle is a straight line for all time
0 < t < T , as expected for the physical, classical free particle.
(b) The time derivative of this trajectory is now
1 πA πt
ẋ = (x f − xi ) + cos . (11.45)
T T T
The action of this trajectory is then
Z
m T 1 πA πt 2
S[x] = dt (x f − xi ) + cos (11.46)
2 0 T T T
Z T
m 1 π A
2 2
2 πt
= dt (x f − xi ) + 2 cos
2
2 0 T2 T T

m 1 π A2 2
= (x f − xi )2 + .
2 T 2T
This is again a parabola, this time in the amplitude A, and it is clearly
minimized when A = 0, when there is no sinusoidal oscillation on the
straight-line trajectory.
(c) For this trajectory, the velocity is
1 nπ nπ t
ẋ = (x f − xi ) + (x f − xi ) cos . (11.47)
T T T
The action is then
Z T 2
m 1 nπ nπ t
S[x] = dt (x f − xi ) + (x f − xi ) cos (11.48)
2 0 T T T
Z T
m 1 n2 π 2 nπ t
= 2
(x f − xi )2 + 2 (x f − xi )2 cos2
dt
2
0 T T T

m 1 n π
2 2
= (x f − xi )2 + (x f − xi )2 ,
2 T 2T
yet again a parabola, this time in the frequency n. And again, the action is
minimized when n = 0, eliminating the sinusoidal oscillation on top of the
classical trajectory.
(d) Repeating the exercise for the harmonic oscillator with the same model tra-
jectories is a very different analysis than for the free particle. The biggest
difference is that there is no limit of the parameter in these model tra-
jectories for which it reduces to the classical trajectory of the harmonic
oscillator, so the trajectory that minimizes the action will be some approx-
imation to the classical trajectory. Here, we will just present one of these
calculations; the kinked trajectory from part (a). We have already cal-
culated the time integral of the kinetic energy. The time integral of the
potential energy in the action is

Z T
mω 2 2t ′
S[x] ⊃ − dt xi + (x − xi ) Θ(T /2 − t) (11.49)
2 0 T
2
2
+ x′ + (t − T /2)(x f − x′ ) Θ(T − t)Θ(t − T /2)
T
Z T
" 2
mω 2 2t ′
=− dt xi + (x − xi ) Θ(T /2 − t)
2 0 T
2 #
′ 2 ′
+ x + (t − T /2)(x f − x ) Θ(T − t)Θ(t − T /2)
T
mω 2 T 2
=− x f + xi2 + (x f + xi )x′ + 2x′2 .
12
Then, the complete value of the action for this harmonic oscillator trajec-
tory is
m mω 2 T 2
S[x] = (x′ − xi )2 + (x f − x′ )2 − x f + xi2 + (x f + xi )x′ + 2x′2 .
T 12
(11.50)
119 Exercises
This is still a parabola in the intermediate position x′ . However, now its

minimum is where

d m ′ ′ 2
mω 2 T 2 ′ ′2

(x − x i )2
+ (x f − x ) − x + x 2
+ (x f + x i )x + 2x
dx′ T 12 f i
2m mω 2 T
=0=− (x f + xi − 2x′ ) − (x f + xi + 4x p ) (11.51)
T 12
or that
1 + ω24T x f + xi
2 2
x′ = . (11.52)
1− ω T
2 2
2
12
What is interesting about this point is that it is always at a position larger
than (xi + x f )/2 as long as the frequency is non-zero, ω > 0. Further, this
trajectory is only a minimum of the action (for this parametrization) if the
second derivative with respect to x′ is positive. The second derivative of the
action is

d2 m ′ ′ 2
mω 2 T 2 ′ ′2

(x − xi ) + (x f − x ) −
2 2
x f + xi + (x f + xi )x + 2x
dx′2 T 12
4m mω 2 T
= − > 0. (11.53)
T 3
This inequality can only be satisfied if
12
ω2 < . (11.54)
T2
So, for sufficiently long high enough frequencies, this trajectory is not even
a reasonable approximation to the classical trajectory.
11.7 (a) Let’s first consider the path integral of the particle that passes through the
upper slit. We can use the formalism established in example 11.2 to do this.
First, in time τ , the particle travels from the origin to the upper slit, located
at ⃗xm = (d, s/2). The path integral for this is
2 2
r
m d +(s/2) m
Z1 = ei 2 h̄τ , (11.55)
2π ih̄τ
where we note that the squared distance that the particle traveled was
s2
⃗x2f = d 2 + . (11.56)
4
Then, the particle travels from the slit to the point on the screen ⃗x f = (d +
l, h) in time T − τ . Note that the squared distance traveled now is
(⃗x f −⃗xm )2 = (l, h − s/2)2 = l 2 + (h − s/2)2 . (11.57)
This path integral is
l 2 +(h−s/2)2
r
i m2 h̄(T −τ )
m
Z2 = e . (11.58)
2π ih̄(T − τ )
Thus, the path integral for the whole trajectory from the origin to the screen
is
l 2 +(h−s/2)2 d 2 +(s/2)2 m 1
i m2 +i m2
Ztop slit = Z1 Z2 = e h̄(T −τ ) h̄τ p . (11.59)
2π ih̄ τ (T − τ )
The path integral for the particle passing through the bottom slit is almost
identical, we just replace s/2 → −s/2 to the position of the bottom slit.
Then,
l 2 +(h+s/2)2 d 2 +(s/2)2 m 1
i m2 +i m2
Zbottom slit = Z1 Z2 = e h̄(T −τ ) h̄τ p . (11.60)
2π ih̄ τ (T − τ )
(b) We can then sum together the two path integrals of the particle passing
through the top and bottom slits to get the total path integral Z:
Z = Ztop slit + Zbottom slit (11.61)
l 2 +(h−s/2)2 l 2 +(h+s/2)2

d 2 +(s/2)2 m 1 im im
i m2
=e h̄τ p e 2 h̄(T −τ ) + e 2 h̄(T −τ )
2π ih̄ τ (T − τ )
d 2 +(s/2)2 l 2 +h2 +(s/2)2

i m2 +i m2 m 1 i m hs −i m hs
=e h̄τ h̄(T −τ ) p e 2 h̄(T −τ ) + e 2 h̄(T −τ )
2π ih̄ τ (T − τ )
d 2 +(s/2)2 l 2 +h2 +(s/2)2

i m2 +i m2 m 1 ms
=e h̄τ h̄(T −τ ) p cos h .
π ih̄ τ (T − τ ) 2h̄(T − τ )
That is, the probability amplitude varies sinusoidally with the distance h
from the center of the screen. The center of the screen is bright (h = 0),
and then there is a dark spot, then light, etc., as |h| increases away from 0.
This analysis is restricted to a fixed time T − τ for which the particle travels
from the slit to the screen and in general we would need to integrate over τ
because we perform no measurement to determine the time τ at which the
particle passes through the slit. However, that integral would just modulate
the amplitude of the interference pattern on the screen, but have no effect
its broad features.
11.8 (a) The Euler-Lagrange equations that we would find for this Lagrangian
are simply augmented by accounting for the spatial derivative of the
wavefunction:
∂L ∂ ∂L ∂ ∂L
− − = 0. (11.62)
∂ψ ∗ ∂ t ∂ ψ̇ ∗ ∂ x ∂ ψ ∗′
From the given Lagrangian, the derivatives are:
∂L ∂L ∂ ∂L ∂ h̄2 ′ h̄2 ′′
= ih̄ψ̇ −V (x)ψ , = 0, = − ψ = − ψ .
∂ ψ∗ ∂ ψ̇ ∗ ∂ x ∂ ψ ∗′ ∂ x 2m 2m
(11.63)
Then, the Euler-Lagrange equations are
∂L ∂ ∂L ∂ ∂L h̄2 ′′
∗
− ∗
− ∗ ′ = 0 = ih̄ψ̇ −V (x)ψ + ψ , (11.64)
∂ψ ∂ t ∂ ψ̇ ∂x ∂ψ 2m
121 Exercises
or that
h̄2 ′′
ih̄ψ̇ = −
ψ +V (x)ψ , (11.65)
2m
exactly the Schrödinger equation in position space.
(b) The variation of this Lagrangian with respect to ψ produces the Euler-
Lagrange equations:
∂L ∂ ∂L ∂ ∂L
− − = 0. (11.66)
∂ ψ ∂ t ∂ ψ̇ ∂ x ∂ ψ ′
The derivatives are now
∂L ∂ ∂L ∂ ∂L h̄2 ∗ ′′
= −V (x)ψ ∗ , = ih̄ψ̇ ∗ , = − ψ . (11.67)
∂ψ ∂ t ∂ ψ̇ ∂ x ∂ ψ′ 2m
The Euler-Lagrange equation for ψ ∗ is then
∂L ∂ ∂L ∂ ∂L ∗ ∗ h̄2 ∗ ′′
− − = 0 = −V (x)ψ − ih̄ψ̇ + ψ , (11.68)
∂ ψ ∂ t ∂ ψ̇ ∂ x ∂ ψ ′ 2m
or that
h̄2 ∗ ′′
−ih̄ψ̇ ∗ = −
ψ +V (x)ψ ∗ , (11.69)
2m
which is clearly the complex conjugate of the Schrödinger equation.
(c) If ψ is an energy eigenstate, then it necessarily satisfies the Schrödinger
equation. In that case, we can re-write the action using integration by parts
as
Z T Z ∞
∗ h̄2 ∗ ′ ′ ∗
S[ψ ] = dt dx −ih̄ψ ψ̇ − ψ ψ −V (x)ψ ψ (11.70)
0 −∞ 2m
Z T Z ∞
h̄2 ∂ h̄2 ∗ ′′
= dt dx − ψ ∗ ψ ′ − ih̄ψ ∗ ψ̇ + ψ ψ −V (x)ψ ∗ ψ
0 −∞ 2m ∂ x 2m
2 Z T Z ∞
h̄ ∂
=− dt dx ψ ∗ψ ′ ,
2m 0 −∞ ∂x
where we have eliminated terms that vanish assuming the Schrödinger
equation. Then, by the fundamental theorem of calculus, the action
Z T Z ∞ Z T
h̄2 ∂ h̄2
ψ ∗ψ ′ = − dt ψ ∗ ψ ′
x=∞
S[ψ ] = − dt dx = 0.
2m 0 −∞ ∂x 2m 0
x=−∞
(11.71)
Now, with the assumption that ψ is an energy eigenstate wavefunction, it
must be L2 -normalizable and therefore have compact support in position
x. Thus, it and its first derivative must vanish as x → ±∞. So, the action on
an energy eigenstate is 0.
12 The Density Matrix
Exercises
12.1 (a) For two independent systems A and B, their combined density matrix is the
tensor product of their individual density matrices:
ρAB = ρA ⊗ ρB . (12.1)
Inserting this into the definition of the Rényi entropy, we have
α log tr (ρAα ⊗ ρBα )
(α ) log tr ρAB log tr (ρA ⊗ ρB )α
SAB = = = (12.2)
1−α 1−α 1−α
log tr ρAα + log tr ρBα
=
1−α
(α ) (α )
= SA + SB ,
so indeed Rényi entropies for independent systems satisfy subadditivity.
(b) Again, the density matrix that demonstrated violation of subadditivity for
Rényi entropies was
1 1 1
ρ12 = | ↑1 ↑2 ⟩⟨↑1 ↑2 | + | ↑1 ↓2 ⟩⟨↑1 ↓2 | + | ↓1 ↑2 ⟩⟨↓1 ↑2 | . (12.3)
2 4 4
Its partial traces are
3 1
ρ1 = tr2 ρ12 = ⟨↑2 |ρ12 | ↑2 ⟩ + ⟨↓2 |ρ12 | ↓2 ⟩ = | ↑1 ⟩⟨↑1 | + | ↓1 ⟩⟨↓1 | , (12.4)
4 4
3 1
ρ2 = tr1 ρ12 = ⟨↑1 |ρ12 | ↑1 ⟩ + ⟨↓1 |ρ12 | ↓1 ⟩ = | ↑2 ⟩⟨↑2 | + | ↓2 ⟩⟨↓2 | .
4 4
All of these density matrices are diagonal, so calculating the Rényi
entropies is simple. For the combined system, we find
α

(α ) log tr ρ12 log 21α + 41α + 41α log 21α + 42α
S12 = = = . (12.5)
1−α 1−α 1−α
The entropies of the two subsystems are equal to each other and are
α
3 1
log tr ρ1α log α + α
(α ) (α ) 4 4
S1 = S2 = = . (12.6)
1−α 1−α
Saturating the subadditivity inequality means that
(α ) (α ) (α )
S12 = S1 + S2 , (12.7)
122
123 Exercises
or that

1 2 3α 1
log α
+ α = 2 log α
+ α . (12.8)
2 4 4 4
Exponentiating both sides, we then have
α
1 2 3 1 2 9α 3α 1
α
+ α
= α
+ α
= α
+ 2 α
+ α. (12.9)
2 4 4 4 16 16 16
Rearranging, this is
8α + 2 · 4α = 9α + 2 · 3α + 1 . (12.10)
This is a transcendental equation for which the solution α cannot be
expressed in closed form. Nevertheless, we can approximate it by plot-
ting both sides of the equation and looking where the curves intersect. The
approximate value of α at which this equality is true is α ≈ 1.6.
(c) Let’s consider the density matrix of the form
ρ12 = (1 − 2 ϵ )| ↑1 ↑2 ⟩⟨↑1 ↑2 | + ϵ | ↑1 ↓2 ⟩⟨↑1 ↓2 | + ϵ | ↓1 ↑2 ⟩⟨↓1 ↑2 | , (12.11)
for some ϵ ∈ [0, 1/2]. The Rényi entropy of this density matrix, and the sum
of its two partial traces are
(α ) log ((1 − 2 ϵ )α + 2 ϵ α ) (α ) (α ) log ((1 − ϵ )α + ϵ α )2

S12 = , S1 + S2 = .
1−α 1−α
(12.12)
We are interested in violating subadditivity for α just greater than 1, so we
want to satisfy
(α ) (α ) (α )
S12 > S1 + S2 , (12.13)
or that
((1 − ϵ )α + ϵ α )2 > (1 − 2 ϵ )α + 2 ϵ α . (12.14)
Taking ϵ small (and importantly smaller than α −1), we can Taylor expand
in ϵ to yield the inequality:
1 − 2α ϵ +α (2α − 1) ϵ 2 + · · · > 1 − 2α ϵ +2α (α − 1) ϵ 2 + · · · , (12.15)
where we assume that α > 1, which establishes that, for example, ϵ > ϵ α .
Rearranging, this becomes
2α − 1 > 2α − 2 , (12.16)
which is indeed true. Therefore, for sufficiently small ϵ , the Rényi entropy
of the density matrix
ρ12 = (1 − 2 ϵ )| ↑1 ↑2 ⟩⟨↑1 ↑2 | + ϵ | ↑1 ↓2 ⟩⟨↑1 ↓2 | + ϵ | ↓1 ↑2 ⟩⟨↓1 ↑2 | , (12.17)
violates subadditivity for all α > 1.
124 12 The Density Matrix
(d) With an appropriate choice of basis, the most general density matrix of two
spins is
ρ12 = a| ↑1 ↑2 ⟩⟨↑1 ↑2 | + b| ↑1 ↓2 ⟩⟨↑1 ↓2 | + c| ↓1 ↑2 ⟩⟨↓1 ↑2 | + d| ↓1 ↓2 ⟩⟨↓1 ↓2 | ,

(12.18)
where 0 ≤ a, b, c, d ≤ 1 and a + b + c + d = 1. The Rényi entropy of this

density matrix is
(α ) 1
S12 = log (aα + bα + cα + d α ) . (12.19)
1−α
The sum of the Rényi entropies of the partial traces are then
(α ) (α ) 1
S1 + S2 = [log ((a + b)α + (c + d)α ) + log ((a + c)α + (b + d)α )]
1−α
(12.20)
Violation of subadditivity means that

(α ) (α ) (α )
S12 > S1 + S2 , (12.21)
or that, for 0 < α < 1,
aα + bα + cα + d α > ((a + b)α + (c + d)α ) ((a + c)α + (b + d)α ) . (12.22)
Let’s now take b = c = d = ϵ , and so the inequality becomes
(1 − 3 ϵ )α + 3 ϵ α > ((1 − 2 ϵ )α + (2 ϵ )α )2 (12.23)

2α 2α α α
= (1 − 2 ϵ ) + (2 ϵ ) + 2(2 ϵ ) (1 − 2 ϵ ) .
Now, let’s consider ϵ very small and so we can Taylor expand both sides of
this inequality to lowest order in ϵ , assuming that α < 1 and 1 − α is much
larger than ϵ . Taylor expanding, we find
1 − 3α ϵ +3 ϵ α > 1 − 4α ϵ +4α ϵ 2α +21+α ϵ α . (12.24)
2α is clearly larger than α , so for small enough ϵ , we can ignore the ϵ 2α

term. Rearranging, we have
α ϵ +(3 − 21+α ) ϵ α > 0 . (12.25)
Note that the second term is positive as long as 3 − 21+α > 0 or that
log 3
α< − 1 ≈ 0.585 . (12.26)
log 2
Therefore, for sufficiently small ϵ , this inequality is satisfied and sub-
additivity is violated for α ≲ 0.585. This example can be extended and
generalized to establish violation of subadditivity for all Rényi entropies
with 0 < α < 1.
125 Exercises
12.2 (a) If the systems A, B, and C are all independent, then the entropy of the joint
system ABC, for example, is the sum of the individual entropies:
ABC A B
SvN = SvN + SvN + SCvN . (12.27)
Using this result, the strong subadditivity condition for independent sys-
tems just reduces to
ABC
SvN B
+ SvN A
= SvN B
+ 2SvN + SCvN ≤ SvN
AB BC
+ SvN A
= SvN B
+ 2SvN + SCvN , (12.28)
which is indeed true.

(b) ABC = 0.
If the combined state ABC is pure, then its entropy vanishes: SvN
Then, the strong subaddivity condition becomes
B
SvN ≤ SvN
AB BC
+ SvN . (12.29)
Next, if ABC is pure, then the entropies of B and AC are equal. So, the
inequality becomes
AC
SvN ≤ SvN
AB BC
+ SvN . (12.30)
Then, continuing with this argument, the entropies of AB and C are equal,
as are BC and A. Then,
AC
SvN ≤ SvN
AB BC
+ SvN = SCvN + SvN
A
, (12.31)
but this is just the condition of subadditivity for systems A, C, and AC.
(c) First, checking the normalization of this pure state, we have
1
⟨ψ |ψ ⟩ = (⟨↑1 |⟨↓2 |⟨↓3 | + ⟨↓1 |⟨↑2 |⟨↓3 | + ⟨↓1 |⟨↓2 |⟨↑3 |)
3
× (| ↑1 ⟩| ↓2 ⟩| ↓3 ⟩ + | ↓1 ⟩| ↑2 ⟩| ↓3 ⟩ + | ↓1 ⟩| ↓2 ⟩| ↑3 ⟩)
1
= (⟨↑1 | ↑1 ⟩⟨↓2 | ↓2 ⟩⟨↓3 | ↓3 ⟩ + ⟨↓1 | ↓1 ⟩⟨↑2 | ↑2 ⟩⟨↓3 | ↓3 ⟩
3
+⟨↓1 | ↓1 ⟩⟨↓2 | ↓2 ⟩⟨↑3 | ↑3 ⟩)
= 1, (12.32)
as expected.
The reduced density matrices are
ρ12 = ⟨↑3 |ψ ⟩⟨ψ | ↑3 ⟩ + ⟨↓3 |ψ ⟩⟨ψ | ↓3 ⟩ (12.33)

1
= [| ↓1 ⟩| ↓2 ⟩⟨↓1 |⟨↓2 | + (| ↑1 ⟩| ↓2 ⟩ + | ↓1 ⟩| ↑2 ⟩) (⟨↑1 |⟨↓2 | + ⟨↓1 |⟨↑2 |)]
3
1 1 1
= | ↓1 ⟩| ↓2 ⟩⟨↓1 |⟨↓2 | + | ↑1 ⟩| ↓2 ⟩⟨↑1 |⟨↓2 | + | ↑1 ⟩| ↓2 ⟩⟨↓1 |⟨↑2 |
3 3 3
1 1
+ | ↓1 ⟩| ↑2 ⟩⟨↑1 |⟨↓2 | + | ↓1 ⟩| ↑2 ⟩⟨↓1 |⟨↑2 | .
3 3
The density matrix ρ23 is just a relabeling of this:

1 1 1
ρ23 = | ↓3 ⟩| ↓2 ⟩⟨↓3 |⟨↓2 | + | ↑3 ⟩| ↓2 ⟩⟨↑3 |⟨↓2 | + | ↑3 ⟩| ↓2 ⟩⟨↓3 |⟨↑2 |
3 3 3
(12.34)
1 1
+ | ↓3 ⟩| ↑2 ⟩⟨↑3 |⟨↓2 | + | ↓3 ⟩| ↑2 ⟩⟨↓3 |⟨↑2 | .
3 3
Finally, we can evaluate the density matrix ρ2 by partial tracing ρ12 , for
example,
1 2
ρ2 = ⟨↑1 |ρ12 | ↑1 ⟩ + ⟨↓1 |ρ12 | ↓1 ⟩ = | ↑2 ⟩⟨↑2 | + | ↓2 ⟩⟨↓2 | . (12.35)
3 3
123 = 0 and the statement
(d) Because the initial state of the three spins is pure, SvN
of strong subadditivity is
2
SvN ≤ SvN
12 23
+ SvN . (12.36)
The density matrix ρ2 is diagonal, so its von Neumann entropy is simple

to evaluate:
1 1 2 2 2
2
SvN = − log − log = log 3 − log 2 . (12.37)
3 3 3 3 3
In matrix form, we can express the density matrix ρ12 as, for example,
 1 
3 0 0
ρ12 =  0 13 13  , (12.38)
0 13 13
where we have ignored the completely 0 rows and columns. The eigenvalues
of this matrix are 2/3, 1/3 and 0. Therefore, its von Neumann entropy is
1 1 2 2 2
12
SvN = − log − log = log 3 − log 2 . (12.39)
3 3 3 3 3
By the simple, symmetric form of the initial pure state, the entropy of the
system 23 is identical:
2
23
SvN 12
= SvN = log 3 − log 2 . (12.40)
3
Then, the inequality we find is

2 2
log 3 − log 2 ≤ 2 log 3 − log 2 , (12.41)
3 3
or that
2
0 ≤ log 3 − log 2 ≈ 0.636514 , (12.42)
3
which is true.
127 Exercises
12.3 (a) Just considering the first three energy eigenstates, the partition function for
the hydrogen atom is
Z = ∑ e−β En ≈ e−β E0 + 4e−β E1 + 9e−β E2 , (12.43)

n
where E0 is the ground state and E1 and E2 are the first and second excited
states. The factors of 4 and 9 are the degeneracy of the excited states: there
are 4 orthogonal states on the Hilbert space that each have energy E1 ,
for example. Additionally, from the structure of the energy spectrum of
hydrogen, we know that
E0 E0
E1 = , E2 = . (12.44)
4 9
Thus, the partition function is approximately
E0 E0
Z ≈ e−β E0 + 4e−β 4 + 9e−β 9 . (12.45)
Demanding that 50% of the hydrogen is in the ground state means that the
Boltzmann factor for the ground state is half of the partition function:
1 e−β E0 e−β E0 1
= ≈ E0 E0 = 3E0 8E0 , (12.46)
2 Z e−β E0 + 4e−β 4 + 9e−β 9 1 + 4eβ 4 + 9eβ 9
or that
3E0 8E0
1 ≈ 4eβ 4 + 9eβ 9 . (12.47)
For right now, we will measure β in units of the inverse ground state energy,
β = −x/E0 , for some unitless number x (remember that E0 < 0). Then, the
equation to solve for 50% of the hydrogen in the ground state is
1 ≈ 4e− 4 x + 9e− 9 x .
3 8
(12.48)
To approximately solve for x here, we can plot both sides of the equa-
tion and then see where they intersect. We find that x ≈ 3, or that the
temperature T at which half of the hydrogen is in the ground state is
E0 13.6
kB T ≈ − ≈ eV ≈ 4.5 eV . (12.49)
3 3
This is a temperature of about 50,000 K, a bit less than our naive estimate
of the temperature of recombination from chapter 9. If you include more
and more energy eigenstates in the evaluation of the partition function, this
temperature further decreases.
(b) By the ratio test, an infinite series converges if the ratio of its terms an is
less than 1 as n → ∞:
an+1
lim < 1. (12.50)
n→∞ an
For the Planck–Larkin partition function, we will first simplify the terms
in the large n-limit. Note that the leading contribution of the term an for
n → ∞ is
!
E
−β 20 E0 E0 1 E0 2 E0
lim n e n − 1 + β 2 = lim n 1 − β 2 +
2 2
β 2 −1+β 2
n→∞ n n→∞ n 2 n n
β 2 E02
= + O(n−4 ) . (12.51)
2n2
Then, in the limit that n → ∞, the ratio test produces
an+1 n2 2
lim = lim = 1− < 1, (12.52)
n→∞ an n→∞ (n + 1)2 n
and so the Planck–Larkin partition function indeed converges. (We tech-
nically used a small extension to the ratio test developed by Gauss.)
(c) As T → 0, β → ∞. The bound state energies of hydrogen all have negative
energy, E0 < 0, and so the partition function is exponentially dominated
by the n = 1 term:
∞ E

−β 20 E0
lim ZP-L = lim ∑ n e n − 1 + β 2 → e−β E0 .
2
(12.53)
β →0 β →0 n=1 n
That is, as β → ∞, the only state that is occupied is the ground state of
hydrogen.
(d) In the high-temperature limit, T → ∞, β → 0, so we can Taylor expand each
term in the sum of the partition function. We have

E
−β 0 E0 β 2 E02 1 π2 2 2
lim lim n2 e n2 − 1 + β 2 = lim 2 = β E0 . (12.54)
β →0 n→∞ n 2 n→∞ n 12
In the high temperature limit, electrons should not be bound into hydro-
gen, but should instead be free particles. There are a continuous infinity of
free, scattering states, and so the probability that any bound state is occu-
pied should be 0. This seems to somewhat be represented with this partition
function, through the fact that it vanishes. However, taking a naive ratio
to determine the occupancy of the ground state in this limit is very strange.
The ratio of the n = 1 term to the full partition function should be a meas-
ure of the occupancy of the ground state, and with this prescription we
have
β 2 E02
6
2
= ≈ 0.608 , (12.55)
π2 2 2
β E0 π2
12
which is very large. If instead we took the ratio of the naive ground state
Boltzmann factor with this partition function, we get an even stranger
result. Note that
lim e−β E0 = 1 , (12.56)
β →0
129 Exercises
and so
e−β E0
lim → ∞. (12.57)
β →0 ZP-L
These results suggest that the Planck–Larkin partition function is a good

approximation only in the small temperature limit.
12.4 (a) Recall that the definition of the path integral was
Zpath = ⟨x f |e−i
T Ĥ
h̄ |xi ⟩ , (12.58)
the transition amplitude from an initial position xi to a final position x f
over time T . Now, we note that the partition function has no complex num-
bers in it, so let’s eliminate the i in the exponent by complexifying time. We
will set τ = iT so that
τ Ĥ
Zpath → ⟨x f |e− h̄ |xi ⟩ . (12.59)
Now, let’s take the product of the complexified path integral with its
complex conjugate:
τ Ĥ τ Ĥ τ Ĥ τ Ĥ
†
Zpath Zpath → ⟨x f |e− h̄ |xi ⟩⟨x f |e− h̄ |xi ⟩† = ⟨x f |e− h̄ |xi ⟩⟨xi |e− h̄ |x f ⟩ .
(12.60)
The path integral encodes transition amplitudes between pairs of initial
and final points, so we should think of it as a matrix. Matrix multiplication
means that we need to integrate over the intermediate rows and columns:
Z
τ Ĥ τ Ĥ
†
Zpath Zpath = dx f ⟨xi |e− h̄ |x f ⟩⟨x f |e− h̄ |xi ⟩ (12.61)
2τ Ĥ
≡ ⟨x|e− h̄ |x⟩ ,
where we have used the fact that the position basis is complete and
removed the subscript i in the final line because it holds for any posi-
tion x. This describes the transition amplitude from an initial position x
back to itself around a complex time loop of circumference 2τ . To turn
this into the trace of the exponential-suppressed Hamitonian operator, we
simply integrate or sum over all possible positions x. That is, the partition
function corresponds to the sum over all positions of the complex-time
loops of circumference 2τ . We identify this circumference with the inverse
temperature: β = 2τ .
(b) First, note that the partition function is a sum of Boltzmann factors over
energy eigenstates:
Zpart = ∑ e−β En . (12.62)
n
Next, we had expressed the path integral as a sum over energy eigenstate
wavefunctions, too, where
Zpath = ∑ e−i
T En
h̄ ψn (x f )ψn∗ (xi ) , (12.63)
n
where the elapsed time is T , the initial position is xi and the final position is
x f . As demonstrated in part (a), we just need to use completeness relations
a couple of times. First, if we complexify time iT = τ , then the path integral
is
τ En
Zpath → ∑ e− h̄ ψn (x f )ψn∗ (xi ) . (12.64)
n
Next, the product of the path integral and its complex conjugate is
Z
τ En τ Em
†
Zpath Zpath → dx f ∑ e− h̄ e− h̄ ψm (xi )ψm∗ (x f )ψn (x f )ψn∗ (xi )
m,n
− 2τh̄En
= ∑e |ψn (xi )|2 , (12.65)
n
using orthonormality of the energy eigenstates

Z
dx f ψm∗ (x f )ψn (x f ) = δmn . (12.66)
Then, integrating over position xi renders the product of the path integral
and its complex conjugate a sum or trace of Boltzmann factors over all
energy eigenstates.
(c) The harmonic oscillator’s path integral is, from the results of chapter 11,
r s
S[x
i class
] m ωT
Zpath = e h̄ . (12.67)
2π ih̄T sin(ω T )
The product of the path integral with its complex conjugate is then
r s r s
† m ωT m ωT
Zpath Zpath = (12.68)
2π ih̄T sin(ω T ) 2π (−i)h̄T sin(ω T )
mω 1
= .
2π h̄ sin(ω T )
Now, if we complexify time, setting
T̃
T =i , (12.69)
2
the product of path integrals becomes
mω 1 mω 1
†
Zpath Zpath → = , (12.70)
2π h̄ sin i ω T̃ π ih̄ 2 sinh ω T̃
2 2
which is indeed the partition function of the harmonic oscillator, up to a

normalization factor.
12.5 (a) In thermal equilibrium, the relative probability of an energy eigenstate is
determined by the Boltzmann factor. Therefore, the average energy is
1 1 dZ d log Z
⟨E⟩ = ∑
Z n
En e−β En = −
Z dβ
=−
dβ
, (12.71)
131 Exercises
where the partition function is
Z = ∑ e−β En . (12.72)
n
(b) The entropy is
1 e−β En 1
S = − ∑ pn log pn = − ∑ e−β En log = ∑ e−β En (β En + log Z)
n Z n Z Z n
= β ⟨E⟩ + log Z , (12.73)
because
1
Z∑
e−β En = 1 , (12.74)
n
by definition.
(c) The exponential factor in the path integral that contains the energy of the
particle, in analogy to the Boltzmann factor, is of the form
tEn
e−β En → e−i h̄ , (12.75)
for some elapsed time t. Then, with this generous analogy, we can identify
the inverse temperature in the path integral with
t
β →i , (12.76)
h̄
or that inverse temperature is proportional to elapsed time. Let’s see if this
makes any sense with the free particle partition function derived in the
previous chapter. Recall that the path integral is
(x −x )2
r
i m2 f h̄T i m
Z=e , (12.77)
2π ih̄T
for total elapsed time T in traveling from position xi to x f . The expectation
value of the energy, within this analogy, would be

h̄ d log Z d m (x f − xi )2 1 2π ih̄T
⟨E⟩ → − = ih̄ i − log (12.78)
i dT dT 2 h̄T 2 m

m (x f − xi )2 1 m (x f − xi )2 ih̄
= ih̄ −i 2
− = − .
2 h̄T 2T 2 T2 2T
This result is quite fascinating. Note that the ratio in the first term is just
the squared velocity v 2 of the classical free particle:
(x f − xi )2
v2 = , (12.79)
T2
and so the first term is exactly the expected kinetic energy of a classical, free
particle. The second term spoils this interpretation, but it is proportional
to h̄, and so seems to exclusively be due to quantum mechanical effects.
12.6 (a) To find the pure states of two spin-1/2 particles that produce the maximal
entropy of entanglement, we just write the post general pure state and take
the partial trace. This most general state is
|ψ ⟩ = a| ↑1 ↑2 ⟩ + b| ↑1 ↓2 ⟩ + c| ↓1 ↑2 ⟩ + d| ↓1 ↓2 ⟩ , (12.80)
for some complex numbers a, b, c, d. Taking the partial trace of particle 2
to generate the density matrix of particle 1, we have
ρ1 = ⟨↑2 |ψ ⟩⟨ψ | ↑2 ⟩ + ⟨↓2 |ψ ⟩⟨ψ | ↓2 ⟩ (12.81)
∗ ∗
= |a| + |b| | ↑1 ⟩⟨↑1 | + |c| + |d| | ↓1 ⟩⟨↓1 | + (ac + bd )| ↑1 ⟩⟨↓1 |
2 2 2 2
+ (a∗ c + b∗ d)| ↓1 ⟩⟨↑1 | .

For the entropy of entanglement to be log 2, the two eigenvalues of the
partial trace density matrix must both be 1/2. Note that by normalization
of the state, the trace is 1 and its determinant is

det ρ1 = |a|2 + |b|2 |c|2 + |d|2 − (ac∗ + bd ∗ )(a∗ c + b∗ d) (12.82)
= (ad − bc)(a∗ d ∗ − b∗ c∗ ) ,
by expanding out |a|2 = aa∗ , for example. Therefore, the characteristic
equation for the eigenvalues λ is
λ 2 − λ + |ad − bc|2 = 0 , (12.83)
which has the solution
q
1 1
λ= ± 1 − 4|ad − bc|2 . (12.84)
2 2
Thus, for maximal entropy of entanglement, we must enforce
1
|ad − bc| = , (12.85)
2
so that the discriminant vanishes.
(b) If instead the partial trace of the pure state density matrix of two particles
describes a pure state, then one of the eigenvalues of the resulting den-
sity matrix is 0. This requires that ad − bc = 0, or that ad = bc. To make
this a bit more transparent, we can re-express the coefficients in a way that
automatically satisfies the normalization or trace constraint. We can write:
a = eiα cos θ cos χ , b = eiβ cos θ sin χ , (12.86)
iγ iδ
c = e sin θ cos η , d = e sin θ sin η ,
for angles θ , χ , η and phases α , β , γ , δ . This automatically satisfies |a|2 +
|b|2 + |c|2 + |d|2 = 1. But now, the singular determinant constraint would
require
ad = bc → ei(α +δ ) cos θ sin θ cos χ sin η = ei(β +γ ) cos θ sin θ sin χ cos η .
(12.87)
133 Exercises
This is implicitly two equations, for the phase and the magnitude, but let’s
just focus on the equation for the magnitude:
cos χ sin η = sin χ cos η , (12.88)
and we must require that 0 ≤ χ , η ≤ π /2 so that the phases encode all sign
information. On this restricted range both cosine and sine are invertible,
and so this can only hold true if χ = η . In this case, the original state can
be expressed as
|ψ ⟩ = a| ↑1 ↑2 ⟩ + b| ↑1 ↓2 ⟩ + c| ↓1 ↑2 ⟩ + d| ↓1 ↓2 ⟩ (12.89)
iα iβ iγ
= e cos θ cos χ | ↑1 ↑2 ⟩ + e cos θ sin χ | ↑1 ↓2 ⟩ + e sin θ cos χ | ↓1 ↑2 ⟩
+ eiδ sin θ sin χ | ↓1 ↓2 ⟩

= eiα cos θ | ↑1 ⟩ cos χ | ↑2 ⟩ + ei(β −α ) sin χ | ↓2 ⟩

+ eiγ sin θ | ↓1 ⟩ cos χ | ↑2 ⟩ + ei(δ −γ ) sin χ | ↓2 ⟩ .
Note that the determinant vanishing requirement enforced that the phases
satisfied β − α = δ − γ . Further, note that the state
|ψ2 ⟩ = cos χ | ↑2 ⟩ + ei(β −α ) sin χ | ↓2 ⟩ , (12.90)
is just a definite spin pointed along an axis defined by the angle χ and the
phase β − α . Let’s call this spin up along the χ axis: | ↑2,χ ⟩. Then, the state
can be expressed as

|ψ ⟩ = eiα cos θ | ↑1 ⟩ + eiγ sin θ | ↓1 ⟩ | ↑2,χ ⟩ . (12.91)
Next, the state
|ψ1 ⟩ = eiα cos θ | ↑1 ⟩ + eiγ sin θ | ↓1 ⟩ , (12.92)
is again a definite spin state long an axis defined by angle θ and phase
difference γ − α . Let’s call this spin up along the θ axis: | ↑1,θ ⟩. Finally, if
the partial traced state is pure, then the initial state takes the form
|ψ ⟩ = eiα | ↑1,θ ⟩| ↑2,χ ⟩ . (12.93)
This is clearly a separable state of two spins pointed up around particular

axes.
(c) If we now have three spin-1/2 particles, the logic continues. For example,
if the initial state of the three spins is pure, and the partial-traced state of
just the spins 1 and 2 is pure and maximally entangled, then necessarily the
initial state |ψ ⟩ was a separable state of the form:
|ψ ⟩ = |ψ12 ⟩| ↑3,θ ⟩ , (12.94)
where |↑3,θ ⟩ is the state of particle 3 pointed up along the θ axis, and |ψ12 ⟩
is some pure state of particles 1 and 2. We can continue this logic. Now,
let’s assume that particles 1 and 3 form a pure state after partial tracing.
With this already separable form, particle 2 must also be separated:
|ψ ⟩ = | ↑1,χ ⟩| ↑2,η ⟩| ↑3,θ ⟩ , (12.95)
which is a completely separable state of three spins pointed along three
different axes. Thus, entropy is monogamous: only two particles can be
maximally entangled at a time.
12.7 (a) For the detected state |11⟩, where each detector is hit by an identical pho-
ton, there are two possible paths that the photons can take. For one, both
photons can be transmitted through the beam splitting and then hit the
detectors. We can assume that the probability amplitude for such a path is
some value A. Next, the two photons could instead have both reflected off
of the beam splitter. The probability of reflecting or transmitting is identi-
cal, and so the absolute value of the probability amplitude for this double
reflection is still |A|. However, each photon now picks up a relative phase
of π /2 as compared to the transmitted photons. Therefore, the probability
amplitude for generating the reflected contribution to |11⟩ is
A11 ⊃ eiπ /2 eiπ /2 A = −A , (12.96)
because there are two photons that each pick up a phase of π /2. Thus,
when the transmitted and reflected amplitudes are summed coherently, we
actually find that the probability for the state |11⟩ is 0.
(b) By contrast, the states |20⟩ and |02⟩ each have a unique path that can con-
tribute: one photon is transmitted and the other is reflected. Therefore,
quantum mechanically, they are both observed. This is exactly the opposite
measurements that would be made classically.
12.8 (a) The first thing we will do is to recall that the partition function of the
harmonic oscillator is
1
Z= . (12.97)
2 sinh β h̄2ω
Then, the thermal average of the Hamiltonian in the harmonic oscillator
is
1 dZ β h̄ω d 1
⟨⟨Ĥ⟩⟩β = ⟨E⟩ = − = −2 sinh (12.98)
Z dβ 2 d β 2 sinh β h̄ω
2
h̄ω β h̄ω
= coth .
2 2
(b) To show that the thermal expectation values of the momentum and posi-
tion are 0, it is sufficient to note that both x̂ and p̂ are some linear
combinations of the raising and lowering operators â† , â. In the evaluation
of the orthonormalization of the eigenstates of the harmonic oscillator, we
had established that
⟨ψn |â† |ψn ⟩ = ⟨ψn |â|ψn ⟩ = 0 . (12.99)
135 Exercises
Therefore, the expectation values of the position and momentum on a

thermal state are both 0.
(c) Because the thermal expectation values are only sensitive to diagonal
entries in the operators x̂2 and p̂2 , their only contributions come from terms
expressed with equal numbers of raising and lowering operators. That is,
h̄ h̄
⟨ψn |x̂2 |ψn ⟩ =
⟨ψn |ââ† + â† â|ψn ⟩ = ⟨ψn |(n + 1) + n|ψn ⟩
2mω 2mω
h̄
= (2n + 1) . (12.100)
2mω
For momentum, we have
mh̄ω mh̄ω
⟨ψn | p̂2 |ψn ⟩ = ⟨ψn |ââ† + â† â|ψn ⟩ = (2n + 1) . (12.101)
2 2
Then, the thermal expectation value of the squared position operator is
1 ∞ −β h̄ω (n+1/2) h̄
⟨⟨x̂2 ⟩⟩ = ∑e
Z n=0
(2n + 1)
2mω
(12.102)

h̄ β h̄ω − β h̄ω ∞ −β h̄ω n
= sinh e 2 ∑e (2n + 1)
mω 2 n=0
∞
h̄ β h̄ω − β h̄ω 2 d
= sinh e 2 − + 1 ∑ e−β h̄ω n
mω 2 h̄ω d β n=0

h̄ β h̄ω − β h̄ω 2 d 1
= sinh e 2 − +1
mω 2 h̄ω d β 1 − e−β h̄ω

h̄ β h̄ω
= coth .
2mω 4
The thermal expectation value for p̂2 is the same, up to an overall factor:

mh̄ω β h̄ω
⟨⟨ p̂2 ⟩⟩ = coth . (12.103)
2 4
Then, the uncertainty principle on this thermal state is

h̄ β h̄ω mh̄ω β h̄ω h̄2 β h̄ω h̄2
σx2 σ p2 = coth coth = coth2 ≥ .
2mω 4 2 4 4 4 4
(12.104)
Note that the range of hyperbolic cotangent is

2 β h̄ω
coth ≥ 1, (12.105)
4
and it only equals 1 if its argument diverges, β → ∞. The β → ∞ limit is
correspondingly the low temperature limit, when only the ground state has
a non-zero probability to be occupied. Additionally, note that in the β → 0
or high-temperature limit, coth takes the form

β h̄ω 16
lim coth2 = 2 2 2. (12.106)
β →0 4 β h̄ ω
In this limit, the uncertainty principle is

h̄2 16 4 h̄2
= ≥ . (12.107)
4 β 2 h̄2 ω 2 β 2ω 2 4
(d) The classical probability distribution of position is inversely proportional
to the speed: it is more likely for the particle to be found at positions where
the speed is small (and so spends a lot of time there) then where the speed
is large. For a fixed energy E, note that the kinetic energy is
m 2 mω 2 2
v =E− x , (12.108)
2 2
or that the speed is
r
2E
v= − ω 2 x2 . (12.109)
m
The turning points are located where the speed vanishes or
r
2E
x± = ± . (12.110)
mω 2
Then, the classical probability distribution is
N
p(x) = q , (12.111)
2E
m − ω 2 x2
for some normalization constant N. It is normalized between the turning

points:
Z x+
dx Nπ
1=N q = . (12.112)
x− 2E
− ω 2 x2 ω
m
Then, the classical probability distribution for the position is

1
p(x) = q . (12.113)
π mω 2 − x 2
2E
For the probability distribution of momentum, in analogy with position,

it is inversely proportional to the time derivative of momentum, or, by
Newton’s second law, the force. Note that
dp
= −mω 2 x , (12.114)
dt
and with fixed energy the potential energy is
mω 2 2 p2
x =E− , (12.115)
2 2m
or that the absolute position is
r
2E p2
x= − . (12.116)
mω 2 m2 ω 2
137 Exercises
Therefore, the normalized classical distribution of momentum is

1
p(p) = p , (12.117)
π 2mE − p2
using analogous results from the position distribution. Both the position
and momentum distributions have expectation value 0, ⟨x⟩ = ⟨p⟩ = 0,
because the probability distributions are symmetric about the position or
momentum origin. The second moment of position, or the variance, is
Z x+
x2 E
⟨x2 ⟩ = dx q = . (12.118)
x− π mω 2 − x
2E 2 m ω2
The second moment of momentum is correspondingly

Z p+
p2
⟨p2 ⟩ = dp p = mE . (12.119)
p− π 2mE − p2
Therefore, the classical product of the variances of position and momen-
tum in the harmonic oscillator are
E2
σx2 σ p2 = . (12.120)
ω2
From part (a), the thermal expectation value of the energy was

h̄ω β h̄ω 1
⟨E⟩ = coth → , (12.121)
2 2 β
in the high-temperature (small β ) limit. Thus, with this energy in the
classical harmonic oscillator, the product of variances is
E2 1
σx2 σ p2 = = 2 2, (12.122)
ω 2 β ω
which is the same as the high-temperature limit of the quantum oscillator
(up to a factor of 4).
12.9 (a) The operator Ĉ is
Ĉ = Ŝb (Ŝb̄ + Ŝb̄′ ) + Ŝb′ (Ŝb̄ − Ŝb̄′ ) . (12.123)
Its square is then

Ĉ2 = Ŝb (Ŝb̄ + Ŝb̄′ ) + Ŝb′ (Ŝb̄ − Ŝb̄′ ) Ŝb (Ŝb̄ + Ŝb̄′ ) + Ŝb′ (Ŝb̄ − Ŝb̄′ ) (12.124)
= Ŝb2 (Ŝb̄ + Ŝb̄′ )2 + Ŝb′2 (Ŝb̄ − Ŝb̄′ )2 + Ŝb Ŝb′ (Ŝb̄ + Ŝb̄′ )(Ŝb̄ − Ŝb̄′ )
+ Ŝb′ Ŝb (Ŝb̄ − Ŝb̄′ )(Ŝb̄ + Ŝb̄′ )
h̄2 2 h̄2
= (Ŝb̄ + Ŝb̄′2 + Ŝb̄ Ŝb̄′ + Ŝb̄′ Ŝb̄ ) + (Ŝb̄2 + Ŝb̄′2 − Ŝb̄ Ŝb̄′ − Ŝb̄′ Ŝb̄ )
4 4
+ Ŝb Ŝb (Ŝb̄ − Ŝb̄ + Ŝb̄ Ŝb̄ − Ŝb̄ Ŝb̄′ ) + Ŝb′ Ŝb (Ŝb̄2 − Ŝb̄′2 − Ŝb̄′ Ŝb̄ + Ŝb̄ Ŝb̄′ )
′ 2 ′2 ′
h̄4
− Ŝb Ŝb′ (Ŝb̄ Ŝb̄′ − Ŝb̄′ Ŝb̄ ) + Ŝb′ Ŝb (Ŝb̄ Ŝb̄′ − Ŝb̄′ Ŝb̄ )
=
4
h̄4
= − [Ŝb , Ŝb′ ][Ŝb̄ , Ŝb̄′ ] .
4
(b) This partially entangled state is
√ √
|ψ ⟩ = w| ↑b ↓b̄ ⟩ + 1 − w| ↓b ↑b̄ ⟩ . (12.125)
As observed in eq. 12.121 on page 325 of the textbook, the operator Ĉ only
contributes to the expectation value on the state |ψ ⟩ if there is a spin-flip.
A spin-flip correlates the two entangled states in |ψ ⟩, and
ptheir correlation
is proportional to the product of their relative fractions, w(1 − w). So, all
we need to do to account for this difference from thepcalculation in the text
is to renormalize the result of equation 12.124 by 2 w(1 − w). The factor
of 2 cancels the case when w = 1 − w = 1/2, and so
p
⟨ψ |Ĉ|ψ ⟩ = 2w(1 − w) h̄2 cos(ϕ − π /4) . (12.126)
At the optimized angle between spin measurements, ϕ = π /4, violation of
Bell’s inequalities is the statement that
p h̄2
⟨ψ |Ĉ|ψ ⟩ = 2w(1 − w) h̄2 > . (12.127)
2
This is satisfied if
1
2w(1 − w) > , (12.128)
4
or that

1 1 1 1
1− √ <w< 1− √ . (12.129)
2 2 2 2
(c) Indeed, there are entangled states that do not violate Bell’s inequalities. In
this case, if either

1 1 1 1
0<w≤ 1− √ , 1+ √ ≤ w < 1. (12.130)
2 2 2 2

Quantum Mechanics: A Mathematical Introduction

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quantum Mechanics: A Mathematical Introduction

Uploaded by

Copyright:

Available Formats

Quantum Mechanics

Contact me at larkoski@slac.stanford.edu for any errors, corrections, or suggestions.

4 Axioms of Quantum Mechanics and Their Consequences 22

5 Quantum Mechanical Example: The Inﬁnite Square Well 34

6 Quantum Mechanical Example: The Harmonic Oscillator 49

7 Quantum Mechanical Example: The Free Particle 62

8 Rotations in Three Dimensions 75

9 The Hydrogen Atom 84

11 The Path Integral 112

12 The Density Matrix 122

(b) To calculate the eigenvalues, we construct the characteristic equation where

One eigenvalue is clearly 0, and the other eigenvalues satisfy

Performing the matrix multiplication, we find

The value of a can be determined by demanding that they are normalized:

or that a = 1/2. That is,

(d) To determine how the exponentiated matrix

acts on an eigenvector ⃗v with eigenvalue λ , let’s use the Taylor expanded

Dn⃗v = λ n⃗v , (2.25)

We can then just plug in the appropriate eigenvalues and eigenvectors.

This pattern continues, and it can be compactly expressed as

for n = 0, 1, 2, . . . . Then, the 2 × 2 exponentiated matrix can be expressed as

A similar exercise can be repeated for the 3 × 3 derivative matrix. We want

p(x) = ax2 + bx + c = d0 P0 (x) + d1 P1 (x) + d2 P2 (x) (2.45)

for some coefficients d0 , d1 , d2 . First, matching the coefficient of x2 , we must

or as the vector in the space of Legendre polnomials

Then, re-interpreting this as a polynomial, we have that

For the second derivative to be non-zero, it must act on P2 (x), and so j = 3.

= ax2 + bx + c + (b + 2a)∆x + 2a∆x2 (2.61)

which is indeed a translation of the polynomial, as expected.

which is indeed true by the Leibniz product rule.

for some coefficients an . If we act Ŝ on this function, we find

Therefore, if m powers of Ŝ act on g(x), it returns

Now, the action of the exponentiated operator on g(x) is

That is, this operator rescales the coordinate x.

det(M − λ I) = λ 2 − (tr M) λ + det M = 0 . (2.78)

(b) Solving for the eigenvalues, we find

(a + d)2 ≥ 4(ad − bc) . (2.80)

because n − m is still an integer. If instead n = m, we can use l’Hôpital’s rule

(b) The eigenvalue equation for the operator D̂ is

for some constant c. These eigenfunctions must be single-valued on θ [0, 2π )

fλ (θ + 2π ) = ceiλ (θ +2π ) = fλ (θ ) = ceiλ θ . (2.93)

That is, we must require that λ is an integer, so that

Thus, the eigenfunctions of D̂ are exactly the orthonormal complex expo-

by orthogonality. The same argument holds for the sin-sin contribution.

The product MM† is

The opposite product is

(b) A general upper triangular matrix takes the form

Its Hermitian conjugate is

Note that the square of the Pauli matrix is

Then, the inner product of ⃗u with itself is

and so ⃗u is normalized. The same calculation follows for⃗v with appropriate

a = ei(θ1 −ϕ1 ) |a| , b = ei(θ1 −ϕ2 ) |b| . (3.24)

c = ei(θ2 −ϕ1 ) |c| , d = ei(θ2 −ϕ2 ) |d| . (3.25)

This form might be recognizable as an angle addition formula. For example,

cos v = cos(u + v) cos u + sin(u + v) sin u . (3.27)

Then, we can set

a = ei(θ1 −ϕ1 ) cos(u + v) , b = ei(θ1 −ϕ2 ) sin(u + v) . (3.28)

Note also that

sin v = sin(u + v) cos u − cos(u + v) sin u . (3.29)

c = ei(θ2 −ϕ1 ) sin(u + v) , d = −ei(θ2 −ϕ2 ) cos(u + v) . (3.30)

Putting it all together, the unitary matrix is

(c) The determinant of this matrix is

This can never equal 0, and so the matrix U is non-degenerate.