OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

Appendix A Review of Essential Math

Here we review the math that is used in the text.

A.1 Differential Equations

Differential equations for physical systems establish a relationship between differentials. A differ-
df (t) df (x)
ential such as or represents the change in f for a small change in the independent variable
dt dx
d2 f (t) df (t)
for space (x) or time (t). A second order differential, like say , is then the change of for a
dt2 dt
d2 x(t) F(x)
small change in t. From simple mechanics, Newton’s law has the form 2 = where F(x) is
dt m
the force, m is the mass, and now x(t) (the position) is the dependent variable and t for time is the
dx(t)
independent variable. So ≡ ẋ is then the change in x for a small change in t and is the velocity.
dt
d2 x(t) dx(t)
And ≡ ẍ is the change in for a small change in t and is the acceleration. Newton’s law
dt2 dt
d2 x(t) F(x)
then says that the acceleration, , is where F is the force and m is the mass.
dt2 m
df (x) d2 f (x) df (t)
We are interested in linear differential equations, meaning that terms such as , , ,
dx dx2 dt
2
d f (t) df 2
may appear, but not terms that involve higher powers of these terms such as ( ) . The order of
dt2 dx
d d
the differential equation is given by the number of times a given operator such as or operates
dx dt
on the function f (x) or f (t), respectively. So1

df (x)
+ f (x) = 0 (A.1)
dx
is first order, while

d2 f (x) df (x)
2
+a + cf (x) = 0 (A.2)
dx dx
is second order.
When all the operators and the unknown function are on the left-hand side and the right-hand
side is zero such as

df (x)
+ f (x) = 0 (A.3)
dx

1 Note that in this discussion, there is just one independent variable for the sake of simplicity. That variable is usually
either space or time (x or t, respectively). Rather than give each discussion in what follows in terms of say first time and
then space, we will just arbitrarily pick x or t, assuming then that the reader can easily substitute the variable interest
df (x) df (t)
for their own problem. Language such as the word “gradient” for or “rate” for should be changed, as required.
dx dt
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

330 REVIEW OF ESSENTIAL MATH

or

d2 f (x) df (x)
2
+a + cf (x) = 0 (A.4)
dx dx
then the equations are homogeneous. When there is a function on the right such as

df (x)
+ f (x) = g(x) (A.5)
dx
or

d2 f (t) df (t)
2
+a + cf (t) = sin 𝜔t (A.6)
dt dt
the equations are inhomogeneous.
First order linear homogeneous differential equations have one solution, second order linear
differential equations have two solutions, etc. An inhomogeneous differential equation has the
particular solution, fp (t) or fp (x), associated with the function on the right-hand side, and also
with the solution or solutions to the homogeneous equation. So for Eq. A.6, the total solution is
f (t) = A1 f1 (t) + A2 f2 (t) + fp (t). In the case of the a time dependent problem, A1 and A2 are
determined by the initial conditions, two in the case of a second order differential equation.

Examples of Solutions to Homogeneous Differential Equations

1. The solution to a first order homogeneous equation of the form

df (x)
+ bf (x) = 0 (A.7)
dx
can be found by rearranging and integrating:

df (x)
= −bdx (A.8)
f (x)
df (x)
∫ = −b ∫ dx (A.9)
f (x)

ln f (x) = −bx + a (A.10)

f (x) = Ae−bx (A.11)

where A = ea is a constant. to be determined by the boundary conditions.

Note on boundary and initial conditions: In solving a differential equation, since the equation
represents a relationship between different derivatives of a function and the function, the solution
for the function is going to depend on the starting value of that function when the independent
variable (say x or t) is zero. So, in the above example, A is going to be determine by the value of
f (x) when x = 0. That is called a boundary condition. If the problem involved time instead of
space, like f (t), you would need to know f (t = 0) which is then an initial condition. If a differential
dn f (t) dn f (x)
equation is nth order, meaning the highest order derivative is n . or , then there must be n
dt dxn
initial or boundary conditions. Think of Newton’s second law, which is second order in time, then
dx(t)
the position x(t) and velocity v(t) = must be specified at t = 0.
dt
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.1 DIFFERENTIAL EQUATIONS 331

2. The solution to a second order homogeneous equation of the form

d2 f (t) df (t)
2
+a + bf (t) = 0 (A.12)
dt dt
can be found by assuming a solution of the form

f (t) = e𝛾t (A.13)

The same approach used in case 1 above would work, but the above approach is now a little
simpler. Substituting this form, taking the derivatives and dividing through by e𝛾x results in a
quadratic equation given by

𝛾2 + a𝛾 + b = 0 (A.14)

With the solution given by the quadratic formula

1
𝛾± = (−a ± √a2 − 4b) (A.15)
2
Note that when a = 0 and b is real and > 0, then

1
𝛾± = ±i √b (A.16)
2
where

i = √−1 (A.17)

The two solutions to the homogeneous equation are

fh1 = A+ e𝛾+ t (A.18)

and

fh2 = A− e𝛾− t (A.19)

The complete solution is

fh (t) = A+ e𝛾+ t + A− e𝛾− t (A.20)

where the constants are determined by the initial conditions (boundary conditions if the indepen-
dent variable is space).
In case

a2 − 4b = 0 (A.21)

then
𝛾−a
𝛾+ = 𝛾− = (A.22)
2
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

332 REVIEW OF ESSENTIAL MATH

and the general solution is then

fh (t) = A1 e𝛾t + A2 te𝛾t (A.23)

Solutions to most common and physically relevant differential equations are tabulated, meaning
they can be found in published tables.2

Examples of Solutions to Inhomogeneous Differential Equations

The solution to an inhomogeneous differential equation is called the particular solution as indicated
earlier.
1. To solve an inhomogeneous first order differential equation of the form

df (x)
+ bf (x) = g(x) (A.24)
dx
we introduce an integrating factor:

ebx (A.25)

Then we can recover the above inhomogeneous equation by noting that

d bx df (x)
(e f (x)) = ebx ( + bf (x)) = ebx g(x) (A.26)
dx dx

We can integrate this equation:

d bx
∫ dx (e f (x)) = ∫ dx ebx g (x) (A.27)
dx
′
ebx f (x) = ∫ dx′ ebx g (x′ ) (A.28)

where x′ has been substituted on the right to distinguish between the x that is in the integral and
the x on the left that is not in the integral, so that

′
fp (x) = ∫ dx′ e−b(x−x ) g (x′ ) (A.29)

The subscript p denotes the particular solution to the inhomogeneous equation.

As indicated earlier, the complete solution to an inhomogeneous first order differential equation
is then the sum of the homogeneous solution and the inhomogeneous solution:

′
f (x) = fh (x) + fp (x) = Ae−bx + e−bx ∫ dx′ ebx g (x′ ) (A.30)

and A is determined for the boundary condition for f (x) (not just for the homogeneous part).

2 For this and many more helpful mathematical relationships, see for example Murray R. Spiegel, Seymour Lipschutz
and John Liu, Schaum’s Outline of Mathematical Handbook of Formulas and Tables, 4th Edition, McGraw Hill (2013).
Also, Milton Abramowitz and Irene A. Stegun, Handbook of Mathematical Functions, National Bureau of Standards
Applied Mathematics Series 55, US Government Printing Office (1964).
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.2 PARTIAL DIFFERENTIAL EQUATIONS (PDE)—METHOD OF SEPARATION OF VARIABLES 333

2. To solve a second order inhomogeneous differential equation, the complete solution is of the
form

f (x) = fh (x) + fp (x) (A.31)

where the two constants in the solution fh (x) are again determined for the boundary conditions for
f (x). The steps for generally finding the particular solution, fp (x) are beyond the current discussion,
but this is often done with a Green’s function approach.3

A.2 Partial Differential Equations (PDE)—Method of Separation of Variables

Much of the analysis and models in this text are limited to one dimension in order to keep the math
simple and because in today’s technology, a one-dimensional system has technological advantages.
A one-dimensional differential equation of the form:

d2 f (x)
+ bf (x) = 0 (A.32)
dx2
may be generalized to two or three dimensions when appropriate. For example, the equation
for electromagnetic waves including light and radio is often a three-dimensional problem. Like
mechanical vibrations of a three-dimensional object, quantum systems like atoms, and thermal
transport for heat management, many devices require a solution to the equivalent three-dimensional
equations. We can deal with this by replacing the differential operator with the appropriate 𝛻
operator where, in three dimensions in Cartesian coordinates, for example:

𝛻 2 f (x, y, z) + bf (x, y, z) = 0 (A.33)

where

𝜕2 𝜕2 𝜕2
𝛻2 = + + (A.34)
𝜕x2 𝜕y2 𝜕z2

This equation is solved by the method of separation of variables and, since the three coordinates
are independent of each other, it must be that this equation holds only when

f (x, y, z) = fx (x)fy (y)fz (z) (A.35)

If we substitute this into the partial differential equation and then divide by f (x, y, z) we get

2
1 𝜕 2 fx (x) 1 𝜕 fy (y) 1 𝜕 2 fz (z)
+ + +b=0 (A.36)
fx (x) 𝜕x2 fy (y) 𝜕y2 fz (z) 𝜕z2

Since this has to hold for all values of x, y, and z, it means that

1 𝜕 2 fx (x)
= ax (A.37)
fx (x) 𝜕x2

3 George B Arfken, Hans J. Weber and Frank E. Harris Mathematical Methods for Physicists, 7th ed., Elsevier,
Amsterdam (2013).
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

334 REVIEW OF ESSENTIAL MATH

2
1 𝜕 fy (y)
= ay (A.38)
fy (y) 𝜕y2
1 𝜕 2 fz (z)
= az (A.39)
fz (z) 𝜕z2

where

ax + ay + az = −b (A.40)

A.3 Eigenvalue Problems

d2
In general, if  is an operator like or 𝛻 2 , then the equation
dx2

̂ a (x) = aua (x)

Au (A.41)

where a is a constant, is called an eigenvalue equation. Then ua (x) is the eigenfunction and a is the
d2
eigenvalue. So, if  = , then
dx2

d2
u (x) = −aua (x) (A.42)
dx2 a
The solution is

ua (x) = e±i√ax (A.43)

d2
If there are no constraints like boundary conditions, then a can take on any positive value., and 2
dx
is said to have a continuous spectrum of eigenvalues. Sometimes, in this case, the eigenfunction is
written as

ua (x) ≡ u (a, x) (A.44)

If there are constraints, such as that the function must satisfy periodic boundary conditions, e.g.,

e±i√ax = e±i√a(x+L) (A.45)

then this is only satisfied for specific values of a. Namely,

e±i√aL = 1

Limiting the values of a to

√aL = 2m𝜋 where m is a positive integer (A.46)

or
2
2m𝜋
a=( ) (A.47)
L
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.3 EIGENVALUE PROBLEMS 335

In this case, the spectrum of eigenvalues is discrete and we write ua (x) as

ua (x) ≡ um (x) (A.48)

For either a discrete or continuous spectrum of eigenvalues and assuming that the operators are
∞ ∞ ∗
Hermitian, meaning an operator P̂ is Hermitian if ∫0 dx g∗ (x)Pf̂ (x) = ∫0 dx(Pg(x)) ̂ f (x) for an
arbitrary and well behaved g(x) and f (x), there are an infinite number of eigenfunctions and the set of
all eigenfunctions for each case forms a complete set. If the boundary conditions allow −∞ < x < ∞,
then any f (x) can be expanded in terms of the eigenfunctions. In the case that the spectrum of
eigenvalues is discrete, it means that f (x) can be written as:

∞
f (x) = ∑ cm um (x) (A.49)
m=0

where cm is called an expansion coefficient. This is a countably infinite series. This is very similar to
the ideas learned in the study of the Fourier series. If the spectrum of eigenvalues is continuous, we
could then expand f (x) in terms of an uncountably infinite series,

∞
f (x) = ∫ da c(a)u (a, x) (A.50)
0

where again c(a) is an expansion coefficient.

It can be shown that, for the operators of interest in these notes and that are associated with
an observable (i.e., the operators are Hermitian), the eigenfunctions are orthogonal and can be
normalized (i.e., they are orthonormal), mathematically meaning that, for eigenfunctions with
discrete eigenvalues,

∞
1 n=m
∫ dx u∗n (x)um (x) ≡ (un (x)|um (x)) = 𝛿nm ≡ { (A.51)
−∞
0 n≠m

where 𝛿nm is a Kronecker delta. A shorthand notation has been introduced to simplify the writing
and calculations and looks similar to Dirac notation though it is definitely not Dirac notation,
which describes eigenvectors in a Hilbert space. For eigenfunctions with a continuous spectrum
of eigenvalues, the orthonormality is given by

∞
∫ dx u∗ (a′ , x) u (a, x) ≡ (u (a′ , x) |u (a, x)) = 𝛿 (a′ − a) (A.52)
−∞

∞
where 𝛿 (a′ − a) is the Dirac delta-function with the property that ∫−∞ dx f (x)𝛿 (x − x0 ) = f (x0 ),
discussed below.
We can use these results to find the expansion coefficient above. For the case of a discrete spectrum
of eigenvalues,

∞
f (x) = ∑ cm um (x) (A.53)
m=0
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

336 REVIEW OF ESSENTIAL MATH

We then multiply both sides by u∗n (x) and integrate over x:

∞ ∞ ∞ ∞ ∞
∫ dx u∗n (x)f (x) = ∫ dx ∑ cm u∗n (x)um (x) = ∑ cm ∫ dx u∗n (x)um (x)
−∞ −∞ m=0 m=0 −∞
∞
=∑ c 𝛿 = cn (A.54)
m=0 m nm

Therefore, with n reverting now to the symbol m:

∞
cm = ∫ dx u∗m (x)f (x) (A.55)
−∞

For the case of a continuous spectrum of eigenvalues,

∞
f (x) = ∫ da c(a)u (a, x) (A.56)
0

We then multiply both sides by u∗ (a′ , x) and again integrate over x:

∞ ∞ ∞
∫ dx u∗ (a′ , x) f (x) = ∫ dx∫ da c(a)u∗ (a′ , x) u (a, x)
−∞ −∞ 0
∞ ∞ ∞
= ∫ da c(a) ∫ dx u (a , x) u (a, x) ∫ da c(a) 𝛿 (a′ − a) = c (a′ )
∗ ′
(A.57)
0 −∞ 0

Therefore, after a′ reverting back to the symbol a,

∞
c(a) = ∫ dx u∗ (a, x) f (x) (A.58)
−∞

A.4 Complex Numbers and Euler’s Theorem

For real numbers x, and y, a complex number z is written as

z = x + iy (A.59)

where

i = √−1 (A.60)

Euler’s theorem says that

z = x + iy = R exp (i 𝜃) = R cos 𝜃 + iR sin 𝜃 (A.61)

where
y
R = √x2 + y2 and tan 𝜃 = (A.62)
x
Complex numbers are represented in the complex plane as shown in Fig. A.1
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.5 DIRAC DELTA-FUNCTION 337

iy

z
r
θ
x

Fig. A.1 Complex numbers are often represented in the complex plane.

A.5 Dirac Delta-Function

The Dirac delta-function has physical meaning only under an integral and defined by the property
that
∞
∫ dx f (x)𝛿(x) = f (0) (A.63)
−∞

Or, by changing variables, it is easy to show that

∞
∫ dx f (x)𝛿(x − x0 ) = f (x0 ) (A.64)
−∞

Likewise,
∞
1
∫ dx f (x)𝛿(ax) = f (0); |a| > 0 (A.65)
−∞
|a|

From the Fourier theorem it can be shown that

∞
′
∫ dk e−ik(x−x ) = 2𝜋𝛿 (x − x′ ) (A.66)
−∞

Proof: It follows from Fourier theory that if

∞
1
ℱ(k) = ∫ dx eikx f (x) (A.67)
√2𝜋 −∞

and
∞
1
f (x) = ∫ dk e−ikx ℱ(k) (A.68)
√2𝜋 −∞

Then the Fourier integral theorem says that

∞ ∞
1 ′
f (x) = ∫ dk e−ikx ∫ dx′ eikx f (x′ ) (A.69)
2𝜋 −∞ −∞
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

338 REVIEW OF ESSENTIAL MATH

Reorganizing, we get
∞ ∞
1 ′
f (x) = ∫ dx′ ( ∫ dk e−ik(x−x ) ) f (x′ ) (A.70)
−∞
2𝜋 −∞

This is only true if

∞
1 ′
∫ dk e−ik(x−x ) = 𝛿 (x − x′ ) (A.71)
2𝜋 −∞

Another form for the Dirac delta-function that is convenient to use is

k−k0 2
1 −( )
𝛿 (k − k0 ) = lim e a (A.72)
a→∞ a√𝜋

A.6 Gaussian Integral

2 2
The functional form e−𝛼x or e−𝛼x +bx is called a Gaussian. The second form is equivalent to
b b2 b2
displaced Gaussian since the exponent can be rewritten as 𝛼x2 − bx = a (x2 − + 2 ) − =
a 4a 4a
b 2 b2
a(x − ) −
2a 4a

∞
2 𝜋
∫ dx e−𝛼x = ; Re a > 0 (A.73)
−∞
√a
∞
2 𝜋 b2
∫ dx e−𝛼x +bx = e 4a ; Re a > 0 (A.74)
−∞
√a

∞ 2 ∞
Note that more complex integrals involve a Gaussian form, such as ∫−∞ dx xe−𝛼x +bx or ∫−∞ dx
2 ∞
x2 e−𝛼x +bx , which can be evaluated by differentiating under the integral. For example: ∫−∞ dx
b2
2 d ∞ 2 b 𝜋 ∞ 2 d ∞ 2 𝜋
xe−𝛼x +bx = ∫−∞ dx e−𝛼x +bx = e 4a and ∫−∞ dx x2 e−𝛼x = − ∫−∞ dx e−𝛼x +bx = √ 3 .
db 2a √ a da a

A.7 Linear Algebra: Matrices, Determinants, Permanents,

and the Eigenvector
Multiplication
Matrices are a two-dimensional array of numbers: n − rows × m − column. A square matrix, such
as frequently encountered in quantum problems, is one where the number of columns and rows are
the same. An n × n matrix  is written as:

a11 ⋯ a1n
 = [ ⋮ ⋱ ⋮] (A.75)
an1 ⋯ ann

The subscripts refer to the rows and columns, respectively. A square n × n matrix has order n × n
which is sometimes just n. Two matrices, Â and B̂ of order j × k and m × n can be multiplied together
if k = m to give a resulting matrix that is order j × n. Matrix multiplication is given by
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.7 LINEAR ALGEBRA: MATRICES, DETERMINANTS, PERMANENTS, AND THE EIGENVECTOR 339

m m
a11 ⋯ a1m b11 ⋯ b1n ∑i=1 a1i bi1 ⋯ ∑i=1 a1i bin
 ∗ B̂ = [ ⋮ ⋱ ⋮ ][ ⋮ ⋱ ⋮ ]=[ ⋮ ⋱ ⋮ ] (A.76)
m m
aj1 ⋯ ajm bm1 ⋯ bmn ∑i=1 aji bi1 ⋯ ∑i=1 aji bin

Three examples:

1 4 1 6 1∗1 + 4∗2 1∗6 + 4∗3 9 18

[ ][ ]=[ ]=[ ] (A.77a)
3 5 2 3 3∗1 + 5∗2 3∗6 + 5∗3 13 33
1 6
[1 2] [ ] = [1 ∗ 1 + 2 ∗ 2 1 ∗ 6 + 2 ∗ 3] = [5 12] (A.77b)
2 3
1 6 4 1∗4 + 6∗5 34
[ ][ ] = [ ]=[ ] (A.77c)
2 3 5 2∗4 + 3∗5 23

Note that when a scalar multiplies a matrix, it multiples every term in that matrix:

1 6 3 18
3∗ [ ]=[ ] (A.77d)
2 3 6 9

Determinants
The determinant of a square matrix is a scalar and is given by selecting any row or column in the
i+j
matrix, and for each element in that row or column, aij , multiply it by (−1) and the determinant of
the matrix formed by removing the row i and column j. As an example, we evaluate the determinant
of the square matrix  by removing the top row:

|a11 ⋯ a1n | |a21 ⋯ a2n | |a11 ⋯ a2n |

det  = || ⋮ ⋱ ⋮ || = a11 || ⋮ ⋱ ⋮ || − a21 || ⋮ ⋱ ⋮ || + ⋯ (A.78)
|an1 ⋯ ann | |an2 ⋯ ann | |an2 ⋯ ann |

where the determinant associated with the multiplying factor out front, aj1 , is missing the first
column and jth row of the original determinant. The process is continued with each resulting matrix
until the final result is a scalar. This can be written more succinctly as:
n
i+j
det  = ∑ (−1) aij Mij (A.79)
i or j

Mij is the first minor of the aij element and is computed by forming the sub-matrix of  by removing
i+j
row i and column j and calculating the determinant. Then (−1) Mij is called the cofactor of the
aij element.
For the matrix:

3 1 2
 = [4 2 3] (A.80)
2 5 1

the minor of Â12 (note that Â12 = a12 = 1, the second entry from the upper left) is

4 3
M̂ 12 = det [ ] (A.81)
2 1
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

340 REVIEW OF ESSENTIAL MATH

So for the determinant of A:̂

|3 1 2|
|2 3| |1 2| |1 2|
det ||4 2 3|| = 3 | | − 4| | + 2| |
|5 1| |5 1| |2 3|
|2 5 1|
= 3 (2 − 15) − 4 (1 − 10) + 2 (3 − 4) = −5. (A.82)

The rank of a square matrix is the order of the largest sub-matrix with linearly independent rows
(columns). This corresponds to the largest sub-matrix with a non-zero determinant. In the case
above with the matrix of order three, corresponding to the determinant with value −45, since the
determinant is non-zero, the rank and order are both three.
The determinant is used to preserve exchange symmetry for fermions.

Permanents
i+j
The permanent of a square matrix is identical to the determinant except that there is no (−1)
factor. So in the case of Eq. A.82, switch the minus sign in front of the 4 in the first equality to a plus
|1 2 | |1 2|
sign: i.e., −4 | | → +4 | |.
|2 5 | |2 5|
The permanent is used to preserve the exchange symmetry for bosons.

Adjoint, Hermiticity, and Unitarity

The adjoint of a matrix  is the complex-transpose of the original matrix. With the matrix  above
†
then the corresponding adjoint, Â , is given by

†
a∗11 ⋯ a∗n1
̂
A =[ ⋮ ⋱ ⋮] (A.83)
a∗1n ⋯ a∗nn

† † −1
If  =  , then the matrix is Hermitian. If  =  , then the matrix is unitary.

Vectors: Inner and Outer Products and Dirac Notation

A matrix V̂ with a single column is an n × 1 column vector:

v1
V̂ = [ ⋮ ] (A.84)
vn

and the complex transpose is a 1 × n row vector:

†
V̂ = [v∗1 ⋯ v∗n ] (A.85)

In Dirac notation, the ket can be represented as a column vector. So for a three-dimensional
Hilbert space,

v1
|V ⟩ = [v2 ] (A.86)
v3
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.7 LINEAR ALGEBRA: MATRICES, DETERMINANTS, PERMANENTS, AND THE EIGENVECTOR 341

The bra is the corresponding complex transpose (row vector) of the ket. The inner product is similar
to a dot product and produces a scalar (a matrix of order unity) that can be complex.

v1
⟨W|V ⟩ = [w∗1 w∗2 w∗3 ] [v2 ] = w∗1 v1 + w∗2 v2 + w∗3 v3 (A.87)
v3

The outer product of two vectors of the same size is a square matrix:

v1 v1 w∗1 v1 w∗2 v1 w∗3

|V ⟩ ⟨W| = [v2 ] [w∗1 w∗2 w∗3 ] = [v2 w∗1 v2 w∗2 v2 w∗3 ] (A.88)
v3 v3 w∗1 v3 w∗2 v3 w∗3

An arbitrary matrix can then be written out as

 = ∑ ⟨i|A|̂ j⟩ |i⟩ ⟨j| (A.89)

ij

where |i⟩ (⟨j| ) are unit vectors with 0’s in all the positions except position i (j), where there is a 1.

01
a11 ⋯ a1n ⎡⎢ ⋮⎥
⎤
⟨i|A|̂ j⟩ ≡ aij = [01 ⋯ 1i ⋯ 0n ] [ ⋮ ⋱ ⋮ ] ⎢ 1j ⎥ (A.90)
⎢ ⎥
an1 ⋯ ann ⎢ ⋮ ⎥
⎣0n ⎦

Example for an operator of order two:

̂ = [1 0] [a11
⟨1|A|2⟩
a12 0
] [ ] = [a11
0
a12 ] [ ] = a12 (A.91)
a21 a22 1 1

For consistency in notation, we note that

Âij = ⟨i|A|̂ j⟩ = aij (A.92)

The Identity Matrix and the Inverse Matrix

The identity matrix has zero’s in all the positions except along the diagonal where it has ones. For
example for a matrix of arbitrary order,

1 0 0
I ̂ = [0 1 0] (A.93)
0 0 ⋱

Therefore, for any matrix  of order n,

 ∗ I ̂ = I ̂∗  =  (A.94)

where I ̂ is the identity matrix of order n.

 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

342 REVIEW OF ESSENTIAL MATH

−1
If  is a matrix with a non-singular determinant, then there exists an inverse matrix  such
that
−1 −1
 ∗  =  ∗  = I ̂ (A.95)

To find the inverse of a matrix, we use Cramer’s rule:

−1 1 ̂T
 = C (A.96)
|A|̂

where

|A|̂ ≡ det  (A.97)

T
and Ĉ is the transpose of the matrix of cofactors (see A.79 and discussion) where the cofactor of
i+j
the aij element is again (−1) Mij , as discussed above. So for say a 4th order matrix:

1 4 7 5
⎡3 |1 4 7|
0 5 3⎤ ⎥ = (−1)2+4 |−1 9 2|
Ĉ24 ≡ (−1) 2+4
M̂24 = (−1) 2+4
det ⎢ | |
⎢−1 9 2 7⎥ | 2 −4 8|
⎣ 2 −4 8 1 ⎦
9 2 −1 2 −1 9
= 1[ ] − 4[ ] + 7[ ] = 80 + 48 − 98 = 30 (A.98)
−4 8 2 8 2 −4

Finally, with Ĉ being given by

c11 ⋯ c1n
Ĉ = [ ⋮ ⋱ ⋮] (A.99)
cn1 ⋯ cnn
c11 cn1
⋯
−1 1 ̂T ⎡ |A|̂ |A|̂ ⎤
 = C =⎢ ⋮ ⋱ ⋮⎥ (A.100)
|A|̂ ⎢ c1n
⋯
cnn ⎥
⎣ |A|̂ |A|̂ ⎦

1
The factor was taken inside to emphasize that when a number multiplies a matrix, it multiplies
|A|̂
every element of that matrix (A.77d).

Cramer’s Rule for Solving n Linear Inhomogeneous Algebraic Equations in

n Unknowns
Suppose we have n linear equations in n unknowns:

a11 x1 + a12 x2 + ⋯ + a1n xn = b1

a21 x1 + a22 x2 + ⋯ + a2n xn = b2
(A.101)
⋮
an1 x1 + an2 x2 + ⋯ + ann xn = bn
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.7 LINEAR ALGEBRA: MATRICES, DETERMINANTS, PERMANENTS, AND THE EIGENVECTOR 343

These can be written in matrix form by putting the coefficients in a matrix and converting the
unknowns, xi and the right-hand side into vectors. Specifically,

a a12 … a1n x1 b
⎡a11 a22 … a2n ⎤ ⎡x ⎤ ⎡b1 ⎤
⎢ 21 ⎥ ⎢ ⎥ = ⎢ 2⎥
2
(A.102)
⎢ ⋮ ⋮ ⋮ ⋮ ⎥ ⎢ ⋮⎥ ⎢ ⋮⎥
⎣an1 an2 … ann ⎦ ⎣xn ⎦ ⎣bn ⎦

Or written more compactly:

ÂX̂ = B̂ (A.103)

If one or more of the b′i s ≠ 0, the set of algebraic equations is inhomogeneous. In this case, one
can derive Cramer’s rule, which says that the value of ith entry of X̂ (a column vector) is:

det Âi
xi = ; i = 1, … , n (A.104)
det Â

where the matrix Âi is formed by taking the matrix  above and replacing the ith column with the
vector B.̂
For example

2 3 x1 −1
[ ][ ] = [ ] (A.105)
1 4 x2 3
2 3
det A = det [ ]=5 (A.106)
1 4
−1 3
det [ ]
det Â1 3 4 −13
x1 = = = (A.107)
det  2 3 5
det [ ]
1 4
2 −1
det [ ]
det Â2 1 3 7
x2 = = = (A.108)
det  2 3 5
det [ ]
1 4

The Eigenvector Problem

In the above case, a special case is had if

b x
⎡b1 ⎤ ⎡ x1 ⎤
⎢ ⎥ = 𝜆 ⎢ 2⎥
2
(A.109)
⎢ ⋮⎥ ⎢ ⋮⎥
⎣bn ⎦ ⎣xn ⎦

In this case,

ÂX̂ = 𝜆X̂ = 𝜆IX̂̂ (A.110)

 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

344 REVIEW OF ESSENTIAL MATH

This is called an eigenvector or eigenvalue equation. Then X̂ is the eigenvector and 𝜆 is the
eigenvalue.
Then

0
⎡01 ⎤
(A − 𝜆I) X̂ = 2 ⎥ = 0
̂ ̂ ⎢ (A.111)
⎢ ⋮⎥
⎣0n ⎦

We rewrite this as

a −𝜆 a12 … a1n x
⎡ 11a a22 − 𝜆 … a2n ⎤ ⎡x1 ⎤
⎢ 21 ⎥ ⎢ 2⎥ = 0 (A.112)
⎢ ⋮ ⋮ ⋮ ⋮ ⎥ ⎢ ⋮⎥
⎣ an1 an2 … ann − 𝜆⎦ ⎣xn ⎦

This is again a set of n equations in n unknowns but now the right-hand side is 0 and so the equations
are a set of homogeneous algebraic equations. In order for a solution to exist, the determinant of
coefficients must be 0:

|a11 − 𝜆 a12 … a1n |

| a21 a22 − 𝜆 … a2n |
| |=0 (A.113)
| ⋮ ⋮ ⋮ ⋮ |
| an1 an2 … ann − 𝜆|

Solving this equation will yield n different values for 𝜆 and requires solving an nth order polynomial.
We consider the case for two unknowns:

a b
 = [ ] (A.114)
c d
x
X̂ = [ 1 ] (A.115)
x2

and

a−𝜆 b x
[ ] [ 1] = 0 (A.116)
c d − 𝜆 x2

We require that

|a − 𝜆 b |
| |=0 (A.117)
| c d − 𝜆|

or

(a − 𝜆) (d − 𝜆) − cb = 0 (A.118)

and after rearranging

𝜆2 − (a + d ) 𝜆 + ad − cb = 0 (A.119)
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.7 LINEAR ALGEBRA: MATRICES, DETERMINANTS, PERMANENTS, AND THE EIGENVECTOR 345

Giving as expected two different eigenvalues,

1 2
𝜆± = ((a + d) ± √(a + d) − 4 (ad − cb)) (A.120)
2

The complete solution only gives one unknown, say x2 , in terms of x1 for each of two eigenvalues.
So, using the equation from above:

a − 𝜆± b x
[ ] [ 1] = 0 (A.121)
c d − 𝜆 ± x2

(a − 𝜆± ) x1 + bx2 = 0 (A.122)

or

(𝜆± − a)
x2 = x1 (A.123)
b

x 1
X̂𝜆± = [ 1 ] = x1 [ (𝜆± −a) ] (A.124)
x2
b

Assuming that the eigenvalues are not degenerate (i.e., 𝜆+ ≠ 𝜆− ), each eigenvector is different. You
∗
can also show that they are orthogonal; i.e., X̂𝜆+ · X̂𝜆− = 0. In quantum systems using Dirac notation,
it would be

1
X̂𝜆± → |𝜆± ⟩ = x1 [ (𝜆± −a) ] (A.125)
b

and x1 would be determined then by normalization:

1
x1 = (A.126)
2
(𝜆± −a)
√1 + ( b
)

⟨𝜆i |𝜆j ⟩ = 𝛿ij , i, j = +/− (A.127)

We note now that we have found that

ÂX̂𝜆± = 𝜆± X̂𝜆± (A.128)

If we converted
′𝜆 0
 →  = [ + ] (A.129)
0 𝜆−

then it is easy to show that

𝜆+ 0
[ ] X̂′ 𝜆± = 𝜆± X̂′ 𝜆± (A.130)
0 𝜆−
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

346 REVIEW OF ESSENTIAL MATH

where the prime represents the eigenstates for this form of the matrix; the corresponding eigenvec-
tors are

1 0
X̂′ 𝜆+ = [ ] and X̂′ 𝜆− = [ ] (A.131)
0 1

So, we say the matrix  has been diagonalized.

Matrix Diagonalization
In the above example of solving the eigenvalue problem we found two eigenvectors of the original
′
matrix and then saw how replacing  with the diagonal matrix,  , the new eigenvectors were the
′
unit vectors in that basis , meaning in terms of the eigenvectors of  . It is important to know how to
′
mathematically convert from  to  . For an arbitrary non-singular matrix of order n, A,̂ we assume
that we have found the eigenvalues 𝜆i and the corresponding normalized eigenvectors X̂𝜆i such that

ÂX̂𝜆i = 𝜆i X̂𝜆i (A.132)

Consider a square matrix where the columns are the eigenvectors,

S ̂ = [X̂𝜆1 X̂𝜆2 ⋯ X̂𝜆n ] (A.133)

Then

ÂS ̂ = [𝜆1 X̂𝜆1 𝜆2 X̂𝜆2 ⋯ 𝜆n X̂𝜆n ] (A.134)

We now form the adjoint of S,̂ which you recall is the complex transpose of S:̂
∗T
⎡X̂𝜆1 ⎤
⎢ ∗T ⎥
† ∗T ⎢X̂ ⎥
Ŝ = Ŝ = ⎢ 𝜆2 ⎥ (A.135)
⎢ ⋮⎥
⎢ ∗T ⎥
⎣X̂𝜆n ⎦

Then
∗T
⎡X̂𝜆1 ⎤
⎢ ∗T ⎥
∗T
̂ ̂ ̂
†
̂ ̂ ̂ ⎢X̂ ⎥
S AS = S AS = ⎢ 𝜆2 ⎥ [𝜆1 X̂𝜆1 𝜆2 X̂𝜆2 ⋯ 𝜆n X̂𝜆n ]
⎢ ⋮⎥
⎢ ∗T ⎥
⎣X̂𝜆n ⎦
∗T ∗T ∗
⎡X̂𝜆1 𝜆1 X̂𝜆1 X̂𝜆1 𝜆2 X̂𝜆2 ⋯ X̂𝜆1 𝜆n X̂𝜆n ⎤ 𝜆1 0 ⋯ 0
⎢ ∗T ∗T ∗T ⎥ ⎡ 𝜆2 ⋯
⎤
= ⎢X̂𝜆2 𝜆1 X̂𝜆1 X̂𝜆2 𝜆2 X̂𝜆2 ⋯ X̂𝜆2 𝜆n X̂𝜆n ⎥ = ⎢ 0 0⎥
(A.136)
⎢ ⋮ ⋮ ⋮ ⋮ ⎥ ⎢ ⋮ ⋮ ⋮ ⋮⎥
⎢ ∗T ∗T ∗T ⎥ ⎢ ⎥
X̂𝜆n 𝜆n X̂𝜆n ⎦ ⎣ 0 0 ⋯ 𝜆n ⎦
⎣X̂𝜆n 𝜆1 X̂𝜆1 X̂𝜆n 𝜆2 X̂𝜆2 ⋯
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

A.7 LINEAR ALGEBRA: MATRICES, DETERMINANTS, PERMANENTS, AND THE EIGENVECTOR 347

Where, by orthonormality
∗T
X̂𝜆i 𝜆k X̂𝜆k = 0 for i ≠ k (A.137)

and
∗T
X̂𝜆j 𝜆j X̂𝜆j = 𝜆j (A.138)

Unitary Transformation
Consider an operator (e.g., a matrix), A.̂ Then let Û be a unitary operator (A.83 and discussion,
† −1
meaning Û = Û ) of the same order as  in the same Hilbert space. Then a new operator,
†
B̂ = ÛÂÛ , is the result of a unitary transformation of A.̂ For example, let ÂX̂ = 𝜆X̂. Since
† † †
Û Û = ÛÛ = I,̂ then inserting Û Û between  and X̂ and then multiplying both sides by Û,
we get
†
ÛÂÛ ÛX̂ = 𝜆ÛX̂ (A.139)

†
Hence, since X̂ is an eigenvector of  with eigenvalue 𝜆, then ÛX̂ is an eigenvector of B̂ = ÛÂÛ
with the same eigenvalue. The same is also true for
† † †
Û ÂÛÛ X̂ = 𝜆Û X̂ (A.140)

†
The transformation above from a matrix  to the matrix S ̂ ÂS,̂ which is now diagonal, is also a
unitary transformation. It is easy to show that the magnitude of both eigenvectors is the same.

Vocabulary (page) and Important Concepts

• first order differential equations 329
• second order differential equations 329
• Homogeneous differential equations 330
• Inhomogeneous differential equations 330
• Method of separation of variables 333
• Eigenvalue problems 334
• Euler’s Theorem 336
• Dirac delta-function 337
• Gaussian integral 338
• Matrix multiplication 338
• First minor 339
• Determinant 339
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

348 REVIEW OF ESSENTIAL MATH

• cofactor 339
• Permament 340
• Hermitian 340
• Unitary 340
• Adjoint 340
• Row vector 340
• Column vector 340
• Inner product 341
• Outer product 341
• Identity matrix 341
• Inverse matrix 342
• Cramer’s Rule 342
• Inhomogeneous algebraic equations 343
• Eigenvalue 344
• Eigenvector 344
• Matrix diagonalization 346
• Unitary transformation 347
 OUP CORRECTED PROOF – FINAL, 26/3/2021, SPi

Appendix B Power Series for

Important Functions

1
= 1 + x + x2 + ⋯ − 1 < x < 1
1−x

1 x 3x2
=1− +
√1 + x 2 8

x x2
√1 + x = 1 + − +⋯
2 8

x2 x4
cos x = 1 − + −⋯
2! 4!

x3 2x5
tan x = x + + +⋯
3 15

x3 x5
sin x = x − +
3! 5!

x2
ex = 1 + x + +⋯
2!

x2 x3 x4
ln (1 + x) = x − + − +⋯ −1<x<1
2 3 4

Taylor series expansion of f (x) about a:

1 ′′ 2
f (x) = f (a) + f ′ (a) (x − a) + f (a)(x − a) + ⋯
2!
 OUP CORRECTED PROOF – FINAL, 26/3/2021, SPi

Appendix C Properties and

Representations of the Dirac
Delta-Function

Definition:
∞
∫ dx f (x)𝛿 (x − x0 ) = f (x0 )
−∞

Identities:
1
𝛿(ax) = 𝛿(x)
|a|

Representations:
∞
1
𝛿 (k − k0 ) = ∫ dx ei(k−k0 )x
2𝜋 −∞
2
1 1 −( k−k0 )
𝛿 (k − k0 ) = lim e a
√𝜋 a→+0 a
∞
1
𝛿 (k − k0 ) = ∫ dx cos x (k − k0 )
𝜋 0

1 sin a (k − k0 )
𝛿 (k − k0 ) = lim
𝜋 a→∞ (k − k0 )

1 a
𝛿 (k − k0 ) = lim
𝜋 a→0 (k − k )2 + a2
0
 OUP CORRECTED PROOF – FINAL, 26/3/2021, SPi

Appendix D Vector Calculus and

Vector Identities

Cartesian coordinates:

Δ 𝜕f 𝜕f 𝜕f
f= x̂ + ŷ + ẑ
𝜕x 𝜕y 𝜕z

Δ 𝜕Ax 𝜕Ay 𝜕Az

·A= + +
𝜕x 𝜕y 𝜕z

Δ 2 𝜕2 f 𝜕2 f 𝜕2 f
f= + +
𝜕x2 𝜕y2 𝜕z2

Cylindrical coordinates:

Δ 𝜕f 1 𝜕f 𝜕f
f= 𝜌̂ + 𝜑̂ + z ̂
𝜕𝜌 𝜌 𝜕𝜑 𝜕z

Δ 1 𝜕𝜌A𝜌 1 𝜕A𝜑 𝜕Az

·A= + +
𝜌 𝜕𝜌 𝜌 𝜕𝜑 𝜕z

Δ 2 1 𝜕 𝜕f 1 𝜕2 f 𝜕2 f
f= 𝜌 + 2 2 + 2
𝜌 𝜕𝜌 𝜕𝜌 𝜌 𝜕𝜑 𝜕z

Spherical coordinates:

Δ 𝜕f 1 𝜕f ̂ 1 𝜕f
f= r ̂+ 𝜃+ 𝜑̂
𝜕r r 𝜕𝜃 r sin 𝜃 𝜕𝜑

Δ 1 𝜕r2 Ar 1 𝜕 (sin 𝜃A𝜃 ) 1 𝜕A𝜑

·A= + +
r2 𝜕r r sin 𝜃 𝜕𝜃 r sin 𝜃 𝜕𝜑

Δ 2 1 𝜕 2 𝜕f 1 𝜕 𝜕f 1 𝜕2 f
f= r + sin 𝜃 +
r2 𝜕r 𝜕r r2 sin 𝜃 𝜕𝜃 𝜕𝜃 r2 sin2 𝜃 𝜕𝜑2
1 𝜕 2
1 𝜕 𝜕f 1 𝜕2 f
= r f + sin 𝜃 + 2
r 𝜕r2 r sin 𝜃 𝜕𝜃
2 𝜕𝜃 r2 sin 𝜃 𝜕𝜑2

Divergence theorem:

Δ
∫ dv · A (x) = ∫ ds n̂ · A (x)
Volume Surface

Stokes’ theorem:

Δ
∫ d𝓵 · A (x) = ∫ ds n̂ · × A (x)
Volume Surface
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

Appendix F Maxwell’s Equations in

Media, the Wave Equation, and Coupling
of a Field to a Two-Level System

In general, Maxwell’s equations describe the behavior of the electromagnetic field. Of interest here
is to show how the observables that were calculated in the text relate directly to Maxwell’s equations.
For this discussion, we assume that the field generated is classical though the source is an ensemble
of quantum systems. The charge and the electric dipole related to the displacement of the charge
are central to Maxwell’s equations. The electron spin is another quantum system that also couples
directly to Maxwell’s equations through the magnetic field. The discussion is nearly parallel.
We start by writing down Maxwell’s equations in their most general form:

𝜕B
𝛻×E+ =0 (F.1)
𝜕t
𝜕D
𝛻×H− =J (F.2)
𝜕t
𝛻·B=0 (F.3)
𝛻·D=𝜌 (F.4)

In the above, J is the current density and 𝜌 is the charge density. Charge is conserved by the
continuity relation given by:

𝜕𝜌
𝛻·J+ =0 (F.5)
𝜕t
The constitutive relationship between D and E is

D = 𝜀0 E + P (F.6)

and between B and H is

B = 𝜇 0 H + 𝜇0 M (F.7)

where P is polarization per unit volume (corresponding to charge displacement), M is the mag-
netization per unit volume (corresponding to the magnetic field resulting from extrinsic and
intrinsic angular momentum), 𝜀0 ≅ 8.85 × 10−12 farads/meter the permittivity of free space, and
𝜇0 ≅ 1.2566 × 10−6 henries/meter is the permeability of free space.
For a single charge, the source terms become
2
𝜌 (r, t) = q||𝜓(r, t)|| (F.8)
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

MAXWELL’S EQUATIONS IN MEDIA, THE WAVE EQUATION 355

The quantum current is

iℏ
J=− (𝜓∗ (r, t) 𝛻𝜓 (r, t) − 𝜓 (r, t) 𝛻𝜓 ∗ (r, t)) (F.9)
2m
In bulk media, however, with N being the number of quantum systems per unit volume, the
polarization is

P(t) = qN⟨𝜓 (r, t) |r|𝜓(r, t)⟩ (F.10)

and the magnetization is

q
M= ̂
N⟨𝜓 (r, t) |J|𝜓(r, t)⟩ (F.11)
2m

The angular momentum, J ̂ in Eq. F.11, is intended to be general here, meaning that if the magnetic
moment is due to intrinsic spin, it would be S and there would be a corresponding g-factor).
Here we are interested in the electron described by a two-level Hamiltonian. Hence, we focus
just on P(t) = qN ⟨𝜓 (r, t) |r|𝜓 (r, t)⟩ and set the other source terms to 0. Maxwell’s equations then
become
𝜕B
𝛻×E+ =0 (F.12)
𝜕t
1 𝜕E 𝜕P
𝛻×B− 2
= 𝜇0 (F.13)
c 𝜕t 𝜕t

𝛻·B=0 (F.14)
𝛻·E =0 (F.15)

Combining with the curl equation, we get

1 𝜕2 E 𝜕2 P
𝛻×𝛻×E+ 2
= −𝜇0 2 (F.16)
c 𝜕t 2 𝜕t
Assuming that the vector components are in the Cartesian coordinate system, then 𝛻 × 𝛻 × E =
𝛻𝛻 · E − 𝛻2 E = −𝛻2 E since 𝛻 · E. Substituting this result, we get a wave equation,

1 𝜕2 E 𝜕2 P
𝛻2 E − 2
= 𝜇0 2 (F.17)
c 𝜕t 2 𝜕t
with

P = N ⟨ 𝝁⟩ = 𝜀0 𝜒E (F.19)

and

⟨𝝁⟩ = Tr𝝁𝜌̂ = (𝝁12 𝜌21

̂ + 𝝁21 𝜌12
̂ ) (F.20)

where 𝜌̂ is the density matrix operator in Chapter 18 and 𝜒 is the electric susceptibility.
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

356 MAXWELL’S EQUATIONS IN MEDIA, THE WAVE EQUATION

To see the implication of the field-induced polarization and ignoring the vector nature of
the field as well as the transient response, we have from Chapter 18 that 𝜌21̂ ∼ E(z)eikz−i𝜔t and
̂ ∼ E∗ (z)e−ikz+i𝜔t . If we set P+ = N𝜇12 𝜌21
𝜌12 ̂ , then we can write P+ as
̂ and P− = N𝜇21 𝜌12

P+ = 𝜀0 (𝜒R + i𝜒I ) E(z)eikz−i𝜔t (F.21)

If the field amplitude, E(z), varies slowly on the scale length of a wavelength, then we can ignore
𝜕2
terms in the wave equation that go like 2 E(z) (there is no dependence in the field amplitude on x
𝜕z
or y for a transverse plane wave). We can then write the wave equation (Eq. F.17) in the form

𝜔2 𝜕 𝜔2
(−k2 + ) E(z) + ik E(z) = − (𝜒 + i𝜒I ) E(z) (F.22)
c2 𝜕z c2 R

Setting imaginary and real parts equal to each other, we get two equations:

𝜔2 𝜔2
(−k2 + 2
) = − 2 𝜒R (F.23)
c c

and

𝜕 𝜔2
k E(z) = − 2 𝜒I E(z) (F.24)
𝜕z c

where the first equation represents the linear dispersion relation with

𝜔2
k2 = (1 + 𝜒R ) (F.25)
c2
𝜔
k=n (F.26)
c

where n2 = (1 + 𝜒R ) is the index of refraction (the ratio of the speed of light in vacuum to the speed
of light in the medium)
The second equation can be solved as

𝛼
− z
E(z) = E (z = 0) e 2 (F.27)

where the absorption coefficient (gain if it is negative) is

2𝜔2 2𝜔
𝛼= 𝜒 = 𝜒 (F.28)
kc2 I n c I
To see how to relate the absorption cross section to fundamental parameters in the density
matrix, we start with Eq. F.20 and use the density matrix from Chapter 18 using first order
perturbation theory. From the density matrix,

1 d𝜌21 𝜇 Eẽ −i𝜔t

̇ =
i𝜚21 [H, 𝜌]21 − i( ) = (𝜔0 − i𝛾) 𝜌21 − 21 (𝜌11 − 𝜌22 ) (F.29)
ℏ d1 decoherence 2ℏ
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

MAXWELL’S EQUATIONS IN MEDIA, THE WAVE EQUATION 357

Take for 𝜌21 in the field interaction picture:

̃ e−i𝜔t
𝜌21 = 𝜌21 (F.30)

Then

𝜇21 Ẽ
̃ = 𝜔0 𝜌21
𝜔𝜌21 ̃ − i𝛾𝜌21
̃ − (𝜌11 − 𝜌22 ) (F.31)
2ℏ
𝜇21 Ẽ
̃ =−
𝜌21 (𝜌 − 𝜌22 ) (F.32)
2ℏ [(𝜔 − 𝜔0 ) + i𝛾] 11

At low power, 𝜌11 − 𝜌22 = 1.

𝜇21 E ̃
̃ =−
𝜌21 (F.33)
2ℏ [(𝜔 − 𝜔0 ) + i𝛾]

and so (to first order in E, linear response)

𝜇12 𝜇21 Ẽ
P̃ + = −N (F.34)
2ℏ [(𝜔 − 𝜔0 ) + i𝛾]

From this,

2𝜔2 2𝜔
𝛼= 𝜒 = 𝜒 ≡ N𝜎 (F.35)
c2 k I c I

where 𝜎 is the cross section for absorption and is defined by this relationship. Since this is linear
theory, we use the phasor e−i𝜔t as used in the Maxwell equation work and the quantum work, and
we have for the prefactor to the phasor,

𝜇12 𝜇21 Ẽ 𝜇 𝜇 Ẽ [(𝜔 − 𝜔0 ) − i𝛾]

P̃ + = 𝜖0 (𝜒R + i𝜒I ) NẼ = 𝜇12 𝜌21
̃ =− N = − 12 21 2
N (F.36)
2ℏ [(𝜔 − 𝜔0 ) + i𝛾] 2ℏ [(𝜔 − 𝜔0 ) + 𝛾2 ]
𝜇12 𝜇21 𝛾 𝜇12 𝜇21 𝛾 𝜇12 𝜇21
𝜒I = 2
N= 2
N= Nℒ (Δ) (F.37)
2ℏ𝜖0 [(𝜔 − 𝜔0 ) + 𝛾2 ] 2ℏ𝜖0 [(𝜔 − 𝜔0 ) + 𝛾2 ] 2ℏ𝜖0 𝛾

𝛾2
where ℒ (Δ) = 2 . Finally, from the form for 𝛼 above,
[(𝜔−𝜔0 ) +𝛾 2 ]

𝜔 𝜒I 𝜇 𝜇 𝜔 4𝜋r12 r21 𝜔
𝜎=2 = 2 12 21 ℒ (Δ) = 𝛼FS ℒ (Δ) (F.38)
c N 2ℏ𝜖0 c𝛾 𝛾

where the fine structure constant is

e2
𝛼FS = (F.39)
4𝜋𝜖0 ℏc

Note that there is a correction factor of order unity in Eq. F.38 associated with the radial matrix
elements in the last expression discussed in Appendix G. The fine structure constant is dimensionless
1
and is approximately 𝛼FS ∼ . The dependence on the fine structure constant is frequently cited
137
 OUP CORRECTED PROOF – FINAL, 27/3/2021, SPi

358 MAXWELL’S EQUATIONS IN MEDIA, THE WAVE EQUATION

as a hall-mark of optical interactions and it reflects the intrinsically weak interaction between light
and charged particles. However, at resonance,

4𝜋r12 r21 𝜔
𝜎0 = 𝛼FS (F.40)
𝛾

In the absence of pure dephasing,

1 1
𝛾= Γ = A (F.41)
2 sp.em. 2
where A is the Einstein A-coefficient, or the inverse radiative lifetime.

4 e2 r12 r21 𝜔3 4 𝛼FS r12 r21 𝜔3

A= = (F.42)
3 4𝜋𝜖0 ℏc3 3 c2

Substituting into the cross section at resonance,

6𝜋c2 6𝜋 3 2
𝜎0 = = 2 = 𝜆 (F.43)
𝜔2 k 2𝜋

There are many important results here for real device studies; however, an important piece of
fundamental physics that impacts technology is the following.
In quantum electrodynamics a famous result is that electromagnetic radiation interacts with
charged particles only weakly, resulting from the fine structure constant. This is misleading when
the transition is a resonance and the transition is lifetime broadened. Just for comparison, for the
optical wavelength, 𝜎0 ∼ 10−12 square meters, but for scattering from a free electron (no resonance,
Thompson scattering) 𝜎 ∼ 10−28 square meters.