2

The Spin-Vector Calculus of Polarization

Spin-vector calculus is a powerful tool for representing linear, unitary trans-

formations in Stokes space. Spin-vector calculus attains a high degree of ab-
straction because rules for vector operations in Stokes space are expressed in
vector form; there is no a priori reference to an underlying coordinate system.
Absence of the underlying coordinate system allows for an elegant, compact
calculus well suited for polarization studies.
Spin-vector calculus is well known in quantum mechanics, especially re-
lating to quantized angular momentum. Aso, Frigo, Gisin, and Gordon and
Kogelnik have greatly assisted the optical engineering community by adopting
this calculus to telecommunications applications [1, 3–5]. The purpose of this
chapter is to bring together a complete description of the calculus as found in
a variety of disparate sources [2, 3, 5–8], and to tailor the presentation with
a vocabulary familiar to the electrical engineer. Tables 2.2 and 2.3 located at
the end of the chapter offer a summary of the principal relations.

2.1 Motivation
The purpose of this calculus is to build a geometric interpretation of polar-
ization transformations. The geometric interpretation of polarization states
was already developed in §1.4. The Jones matrix, while a direct consequence
of Maxwell’s equations when light travels through a medium, is a complex-
valued 2 × 2 matrix. This is hard to visualize. The Mueller matrix, however,
can be visualized as rotations and length-changes in Stokes space. The spin-
vector formalism makes a bilateral connection between the Jones and Mueller
matrices.
Of all the possible Jones matrices, two classes predominate in polarization
optics: the unitary matrix and the Hermitian matrix. The unitary matrix pre-
serves lengths and imparts a rotation in Stokes space. A retardation plate is
described as a unitary matrix. The Hermitian matrix comes from a measure-
ment, such as that of a polarization state. Since all measured values must be
38 2 The Spin-Vector Calculus of Polarization

real quantities, the eigenvalues of a Hermitian matrix are real. The projection
induced by a polarizer is described as a Hermitian matrix.
Based on the characteristic form (1.4.27) on page 19 of the Mueller matrix
for a unitary matrix, defined by U U † = I, one can write
 
1 0 0 0
0 
JU −→ MU =  0

 (2.1.1)
R
0

where R is a 3 × 3 rotation matrix having real-valued entries. Since the po-

larization transformation through multiple media is described as the product
of Jones matrices, one would expect a one-to-one correspondence between
multiple unitary matrices and multiple rotation matrices. This would lead to
 
1 0 0 0
0 
JU2 U1 −→ MU2 U1 =  0 R2 R1 
 (2.1.2)
0

This is indeed the case. Moreover, the Mueller matrix representing passage
of light through any number of retardation plates always keeps the form
of (2.1.1). Rotation matrix R is therefore a group closed under rotation.
Taking the abstraction one step further, any rotation has an axis of ro-
tation and an angle through which the system rotates. Instead of describing
the rotation R as a 3 × 3 matrix, it is more general to describe the rotation
as a vector quantity: R = f (r̂, ϕ), where r̂ is the rotation axis in Stokes space
and ϕ is the angle of rotation. The vector r̂ need not be resolved onto an
orthonormal basis to give r̂ = x̂ rx + ŷ ry + ẑ rz ; this operation may be post-
poned indefinitely. This is in contrast to writing R as a 3 × 3 matrix where
the underlying orthonormal basis is explicit. Accordingly, r̂ exists as a vector
in vector space and can undergo operations such as rotation, inner product,
and cross product with respect to other vectors.
In parallel to the unitary-matrix case, the Mueller matrix that corresponds
to a Hermitian matrix, defined by H = H † , one can write
 
 
JH −→ MH = 
 H̃ 
 (2.1.3)

This indeed is a tautology. As with the unitary matrices, products of Hermi-

tian matrices in Jones space result in products of Mueller matrices in Stokes
space. That is,
JH2 H1 −→ MH2 H1 = MH2 MH1 (2.1.4)
All Hermitian operations are closed within the 4 × 4 Mueller matrix.
 2.2 Vectors, Length, and Direction 39

As it appears, products of unitary-corresponding Mueller matrices change

only entries in the lower-right-hand 3 × 3 sub-matrix. Inclusion of even a single
Hermitian-corresponding Mueller matrix scatters those nine elements into all
sixteen matrix positions. This is a non-reversible process. There is, however, a
remarkable exception. A traceless Hermitian matrix H, defined by TrH = 0,
has a corresponding Mueller matrix of the form
 
1 0 0 0
0 
JH −→ MH =  0

 (2.1.5)
V
0
where V is a Stokes-space vector having a length and pointing direction. (Note
that a rotation operator has unit length, two angles that determine the vector
direction, and one angle of rotation. A Stokes vector has a length and two
angles that determine the vector direction. Both rotation operator and Stokes
vector have three parameters). Arbitrary products of unitary matrix U and
traceless Hermitian matrix H form an extended closed group in which entries
change only in the lower right-hand 3 × 3 sub-matrix of M.
Throughout this chapter and the chapters on polarization-mode disper-
sion, one looks for zero trace of Hermitian matrices. If this property is es-
tablished, then a calculus that includes lossless rotations of vectors can be
applied to the system. This calculus is called spin-vector calculus, and is the
topic of the present chapter.

2.2 Vectors, Length, and Direction

Physical systems can often be described by the state the system is in at a
particular time and position. The span of all possible states for a given system
is called a state space. Any particular state represents all the information that
one can know about the system at that time and position. Interaction between
a physical system and external influences, such as transmission through media
or applied force, can change the state. So, there are two categories of study:
the description of state, and the transformation of state.
A state that describes wave motion can be represented by a vector with
complex scalar entries. The dimensionality of the state vector is determined
by the number of states that are invariant to an external influence. That
is, the dimension of a state vector equals the number of eigenstates of the
system. For polarization, the dimensionality is two. The important properties
of a vector space are direction, length, and relative angles. These metrics will
form a common theme throughout the following development.

2.2.1 Bra and Ket Vectors

Bra and ket spaces are two equivalent vector spaces that describe the same
state space. Bra and ket spaces, or “bracket” space, is a formulation developed
40 2 The Spin-Vector Calculus of Polarization

by P. A. M. Dirac and used extensively in quantum mechanics. Bras and kets

are vectors with dimension equal to the state dimension. When a bra space
and ket space describe the same state vector, the bra and ket are duals of one
another. For a state vector a, the ket is written |a and the bra is written a |.
The entries in bra and ket vectors are complex scalar numbers. A ket vector
suitable for polarization studies is
 
ax
|a =   (2.2.1)
ay

where ax and ay are the components along an orthogonal basis. The entries
are complex and accordingly there are four independent parameters contained
in (2.2.1). Since the entries are complex, they have magnitude and phase:
   
|ax |ejφx |a |
|a =   = ejθ  
x
(2.2.2)
|ay |ejφy |ay |ejφ

where θ is a common phase and φ is the phase difference of the second row.
In the following the explicit magnitude symbols | · | will be dropped and the
intent of magnitude or complex number should be clear from the context. Bra
vector a | is said to be the dual of |a because they are not equal but they
describe the same state:
dual
|a ←−−−→ a |
The bra vector a | corresponding to |a is
 
a | = a∗x a∗y (2.2.3)

for every |a. The bra vector is the adjoint (†), or complex-conjugate transpose,
of the corresponding ket vector:
†
a | = (|a) (2.2.4)

Bra and ket vectors obey algebraic additive properties of identity, addition,
commutation, and associativity. Identity and addition rules for kets are

identity |a + |0 = |a

addition |a + |b = |γ
where |0 is the null ket. Commutation and associativity are straightforward
to prove using the matrix representation. A bra or ket vector can also be
multiplied by a scalar quantity c:

c |a = |a c (2.2.5)

Physically, the multiplication of a state vector by a scalar does not change the
state and therefore the two commute. Operations that have no meaning are
 2.2 Vectors, Length, and Direction 41

the multiplication of multiple ket vectors or bra vectors. For example, |b |a
is meaningless.
Finally, it should be understood that state vectors a | and |a are a more
general representation than column and row vectors (2.2.1) and (2.2.3). A
state vector is a coordinate-free abstraction that has the properties of length
and direction; a row or column vector is a representation of a state vector
given a choice of an underlying coordinate system.

2.2.2 Length and Inner Products

Bra and ket vectors have properties of length, phase, and pointing direction.
The length of a real-valued vector is a scalar quantity and is determined by
the dot product: |a|2 = a · a. For complex-valued bra-ket vectors, the inner
product is used to find length of a vector and is determined by multiplying its
bra representation a | with its ket representation |a: a2 = a |a, where  · 
is the norm of the vector.
More generally, one wants to measure the length of one vector as projected
onto another. The inner product of two different vectors is the product of the
bra form of one vector and the ket for of the other: b |a. For real-valued vec-
tors it is clear that b · a = a · b. However, for bra-ket vectors, having complex
entries, the order of multiplication dictates the sign of the resulting phase.
That is,

b |a = |b |a| ejγ (2.2.6a)

−jγ
a |b = |b |a| e (2.2.6b)

The two inner products are related by the complex conjugate:

∗
b |a = (a |b) (2.2.7)

The inner product of a bra and ket is a complex-valued scalar. Based on

(2.2.7) it is clear that the inner product of a vector onto itself yields a real
number, and since the inner product is a measure of length, the real number is
positive definite: a |a = real number ≥ 0. Only the null ket has length zero.
Any finite ket has a length greater than zero. Throughout the body of the
text, polarization vectors are taken to be unit vectors unless otherwise stated.
A unit vector has a direction, phase, and unity length. Any vector can be
converted to a unit vector by division by its norm:
1
|ã = |a (2.2.8)
a |a

so that
ã |ã = 1 (2.2.9)
In the following the tilde over the vectors will be dropped.
42 2 The Spin-Vector Calculus of Polarization

Two vectors are defined as orthogonal to one another when the inner
product vanishes:
b |a = 0 (2.2.10)
This is an essential inner product used regularly.
When two polarization vectors are resolved onto a common coordinate
system,
b |a = b∗x ax + b∗y ay (2.2.11)
Finally, the inner product in matrix representation of a normalized vector is
the sum of the component magnitudes squared:

a |a = |ax |2 + |ay |2 = 1 (2.2.12)

2.2.3 Projectors and Outer Products

The inner product measures the length of a vector or the projection of one
vector onto another. The result is a complex scalar quantity. In contrast,
the outer product retains a vector nature while also producing length by
projection. There are two outer product types to study: the projector, having
the form |pp|; and the outer product |pq|. The form |pq| is called a dyadic
pair because the vector pair has neither a dot nor cross product between them.
In quantum mechanics the projector |pp| is called the density operator for
the state.
Consider a projector that operates on ket |a:

|pp |a = |p (p |a) = c |p (2.2.13)

The quantity c = p |a is just a complex scalar and commutes with the ket.
Operating on |a the projector measures the length of |a on |p and produces
a new vector |p.
The effect of the projector is to point along the |p direction where the
length of |p is scaled by p |a. Projectors work equally well on bras, e.g.

a |p p | = c∗ p | (2.2.14)

so in fact it should be clear that

†
a |p p | = (|pp |a) (2.2.15)

The adjoint operator connects the bra and ket forms.

The behavior of the outer product |pq| is similar to the projector but
for the fact that the projection vector and resultant pointing direction differ.
The resultant pointing direction depends whether the outer product operates
on a ket or a bra. Acting on a ket, the outer product yields

|pq |a = (q |a) |p (2.2.16)

 2.2 Vectors, Length, and Direction 43

whereas acting on a bra of the same vector, the outer product yields

a |p q | = (a |p) q | (2.2.17)

The resultant pointing direction and projected length depends on whether the
outer product operates on a bra or ket vector.
In the study of polarization, the outer product is a 2 × 2 matrix with
complex entries:  
bx a∗x bx a∗y
|ba| =   (2.2.18)
by a∗x by a∗y
The determinant is
det (|ba|) = 0 (2.2.19)
and therefore the projector is non-invertible. The determinant of an outer
product of any dimension is likewise zero. That means the action of |ba|
on a ket is irreversible, which is reasonable because the original direction of
the ket is lost. So, while all outer products are operators not all operators
are outer products. Operators that are linear combinations of projectors are
reversible under the right construction.
In summary, the outer product follows these rules:
†
equivalence (|ba|) = |ab|
associative (|ba|) |γ = b | (a |γ)
trace Tr (|ba|) = a |b
irreversible det (|ba|) = 0

where Tr stands for the trace operation. The trace connects the outer product
to the inner product.

2.2.4 Orthonormal Basis

An orthonormal basis is a complete set of orthogonal unit-length axes on which

any vector in the space can be resolved. Consider a vector space with N di-
mensions and orthogonal unit vectors (|a1  , |a2  , . . . , |aN ). The orthogonality
requires
am |an  = δm,n (2.2.20)
where δm, n is the Kronecker delta function. Only a vector projected onto
itself yields a non-vanishing inner product. When the set is complete, the
outer products are closed, where closure is defined as

|an an | = I (2.2.21)

n
44 2 The Spin-Vector Calculus of Polarization

When a basis set, or group, is closed, any operation to a member of the group
results in another member within the group. Together, (2.2.20–2.2.21) are the
two conditions that define an orthonormal basis.
Given an orthonormal basis, any arbitrary vector can be resolved onto the
basis using (2.2.21). An arbitrary ket |s is resolved as
 

|s = |an an | |s = cn |an  (2.2.22)

n n

where the complex coefficients are given by cn = an |s. The inner prod-
uct s |s is the sum of the absolute-value squares of the coefficients cn :

s |s = |ca
|2 (2.2.23)

a


When |s is normalized a
|ca
|2 = 1.

2.3 General Vector Transformations

Interaction between a physical system and external influences can change the
state of a system. Left unperturbed, a state persists indefinitely. Operators
embody the action of external influences and are distinct from the state of
the system itself. The bra and ket vectors of the preceding section are two
equivalent spaces that describe the same state space. Operators also have two
distinct and equivalent spaces that describe the same state transformation.
While there is no special notation to represent a “ket” operator or a “bra”
operator, equivalence between operator spaces is maintained under

X |a ←−−−→ a | X †
dual
(2.3.1)
X † is said to be the adjoint operator of X. Care should be taken because the
action of X |a is not the same as a | X; these two results are different.

2.3.1 Operator Relations

Operators always act on kets from the left and bras from the right, e.g. X |a
or a | X. The expressions |a X and Xa | are undefined. An operator multi-
plying a ket produces a new ket, and an operator multiplying a bra produces
a new bra. In general, an operator changes the state of the system,
X |a = c |b (2.3.2)
where c is a scaling factor induced solely by X. Operators are said to be equal
if
X |a = Y |a ⇒ X = Y (2.3.3)
Operators obey the following arithmetic properties of addition:
 2.3 General Vector Transformations 45

commutative X +Y =Y +X
associative X + (Y + Z) = (X + Y ) + Z
distributive X (|a + |b) = X |a + X |b
Operators in general do not commute under multiplication. That is
XY = Y X (2.3.4)
In matrix form, only when X and Y are diagonal matrices does XY = Y X.
Other multiplicative properties are

identity I |a = |a

associative X(Y Z) = (XY )Z
distributive X (Y |a) = XY |a
All of the above arithmetic properties apply equally well to bra vectors.
The effect X has on state a is measured by
a |X| a
expectation value of X on a = (2.3.5)
a |a
In general, an inner product that encloses an operator gives a complex number:
b | (X |a) = b |X| a = complex number (2.3.6)
Consider dual constructions, first where X |a is left-multiplied by b |, and
second where the dual a | X † is right-multiplied by |b:
   ∗
b |X| a = a X †  b (2.3.7)
These two cases are duals of one another and are therefore complex conjugates.
In the study of polarization, operators are represented as 2 × 2 complex-
valued Jones matrices:  
a ejα b ejβ
X=  (2.3.8)
c ejγ d ejη
There are eight independent variables contained in the operator. If det X = 0,
then X is invertible and the action of X can be undone.
The properties of operators are summarized as follows:
dual
operator duality X |a ←−−−→ a | X †
change of state X |a = c |b
a | X † = c∗ b |
inner product with operator b |X| a = complex number
 
conjugate relation b |X| a = a X †  b∗
†
conjugate transpose (XY ) = Y † X †
46 2 The Spin-Vector Calculus of Polarization

Just as an arbitrary ket can be resolved onto an orthonormal basis, an

arbitrary operator X can be resolved onto a set of projection operators formed
on the orthonormal basis. Applying the closure relation (2.2.21) yields
   

X = |am am | X |an an |

m n

= |am  am |X| an an | (2.3.9)

n m

The indexing symmetry of (2.3.9) looks like a matrix with am |X| an  as
the (m, n) entry. For polarization, the matrix is 2 × 2 and looks like
 

a1 |X| a1  a1 |X| a2 

|am  am |X| an an | →   (2.3.10)
n m a2 |X| a1  a2 |X| a2 

The resolved form of X in (2.3.10) will become particularly simple in discus-

sion of Hermitian and unitary matrices.

2.4 Eigenstates, Hermitian and Unitary Operators

Many physical systems exhibit particular states that are not transformed by
interaction with the system. These invariant states are called eigenstates of
the system. In spin-vector calculus, operators embody the influence of a phe-
nomena. The eigenvectors of an operator are the eigenstates of the system.
When an operator X acts on its own eigenstate a,
X |a1  = a1 |a1  (2.4.1a)
†
a1 | X = a∗1 a1 | (2.4.1b)
the state of the system is unaltered but for a scaling factor a1 . The scale factor
is the eigenvalue of X associated with eigenstate a1 . Each eigenvector has an
associated eigenvalue, and a well-conditioned matrix has as many eigenvectors
as rows in the matrix or, equivalently, dimensions in the state space.
The eigenvectors of Hermitian and unitary operators are orthogonal when
the associated eigenvalues are distinct. The eigenvalues of a Hermitian op-
erator are real-valued scalars, and the eigenvalues of a unitary operator are
complex exponential scalars. A Hermitian or unitary operator X having N
eigenkets (|a1  , |a2  , . . . , |aN ) and associated eigenvalues (a1 , a2 , . . . , aN ) pro-
duces the series of inner products
 
am X † X  an  = |am |2 δm,n (2.4.2)
The operator X † X scales each axis by a different amount, but does not rotate
nor create reflection of the original basis. The eigenvalues of operator X are
related to the determinant and trace by
 2.4 Eigenstates, Hermitian and Unitary Operators 47

det(X) = a1 a2 · · · aN (2.4.3a)
Tr(X) = a1 + a2 + · · · + aN (2.4.3b)

Since the eigenvalues of a Hermitian matrix are real, its determinant and trace
are real.

2.4.1 Hermitian Operators

The defining property of a Hermitian operator is

H† = H (2.4.4)

The associated Hermitian matrix in polarization studies has only four inde-
pendent variables: three amplitudes and one phase. This contrasts with the
general Jones matrix (2.3.8) which has eight.
The eigenvectors of H form a complete orthonormal basis and the eigen-
values are real. That the eigenvalues are real is proved from the following
difference:
 
an  H † − H  am  = (an ∗ − am ) an |am 
= 0 (2.4.5)

Non-trivial solutions are found when neither vector is null. The eigenvectors
may be the same or different. Consider first when the eigenvectors are the
same. Since an |an  = 0, (a∗n − an ) = 0 and the eigenvalue is real. Consider
when the eigenvectors are different. Unless am = an , in which case the eigen-
vectors are not linearly independent, it must be the case that an |am  = 0.
All eigenvalues are therefore real. Hermitian operators H scale its own basis
set:  
am H † H  an  = a2m δm,n (2.4.6)
When det(H) = 0, H is invertible and the action of H on the state of a system
is reversible.
The expansion of H onto its own basis generates a diagonal eigenvalue
matrix. Under construction (2.3.9) the expansion yields

H = |am  am |H| an am |

n m

= am |am am | (2.4.7)

m

where am |H| an  = an δm,n . The orthonormal expansion is written in matrix

form as H = SΛS −1 , where S is a square matrix whose columns are the
eigenvectors of H and Λ is a diagonal matrix whose entries are the associated
eigenvalues. Schematically,
48 2 The Spin-Vector Calculus of Polarization
   
a1
 | | |   
   a2 
S=
 v1 v2 · · · vN
 , and Λ = 
  ..

 (2.4.8)
   . 
| | | aN

where |an  = vnT .

2.4.2 Unitary Operators

The defining property of a unitary operator is

T †T = I (2.4.9)

Acting on its orthogonal eigenvectors |an , the unitary operator preserves the
unity basis length:  
am T † T  an  = δm,n (2.4.10)
Taking the determinant of both sides of (2.4.9) gives det(T † T ) = 1. Since the
determinant of a product is the product of the determinants and the adjoint
operator preserves the norm, the determinant of T must be

det(T ) = ejθ (2.4.11)

Since the the determinant is the product of eigenvalues, the eigenvalues of T

must themselves be complex exponentials and, accounting for (2.4.10), they
must have unity magnitude. Therefore T acting on an eigenvector yields

T |an  = e−jαn |an  (2.4.12)

The eigenvalues of T lie on the unit circle in the complex plane.

A special form of T exists where the determinant is unity. This special
form is denoted U and is characterized by det(U ) = +1. To transform from T
to U , the common phase factor β = exp(jθ/N ) must be extracted from each
eigenvalue of T , where N is the dimensionality of the operator. The T and U
forms are thereby related:
T = ejβ U (2.4.13)
It should be noted that when det(U ) = −1, a reflection is present along an
odd number of axes in the basis set of U .
The eigenvalue equation for U is

U |an  = e−jφn |an  (2.4.14)

U expands on its own basis set in the same way H expands (2.4.7):

U= e−jφm |am am | (2.4.15)

m
 2.4 Eigenstates, Hermitian and Unitary Operators 49

Hy = H UyU = +1 =m

+1

<e <e
eig(H) eig(U)

Fig. 2.1. Eigenvalue loci of H and U . Left: eigenvalues of H lie on the real number
line. Right: eigenvalues of U lie on the unit circle in the complex plane.

This orthonormal expansion has the matrix analogue of U = S exp(−jΛ)S −1 ,

where the diagonal matrix is
 −jφ 
e 1
 
 e−jφ2 
exp (−jΛ) =  ..

 (2.4.16)
 . 
e−jφN
and S is the same form as (2.4.8).

2.4.3 Connection between Hermitian and Unitary Matrices

The connection between Hermitian and unitary operators is quite intimate.
Figure 2.1 illustrates the eigenvalue domains for H and U . The eigenvalues
of H lie on the real number line, while those of U lie on the unit circle in
the complex plane. Multiplying the eigenvalues of H by −j and taking the
exponential, one can construct the eigenvalues of U . Note that the eigenvalues
of U are cyclic, so only the real number line modulo 2π is significant.
Based on the operator expansions of the preceding sections, one has
H = SΛH S −1 (2.4.17a)
−jΛU −1
U = Re R (2.4.17b)
Since in general exp(SΛS −1 ) = S exp(Λ)S −1 , the H and U operators may be
connected as
U = e−jH =⇒ Se−jΛU S −1 = Se−jΛH S −1 (2.4.18)
For every Hermitian operator H there is an associated unitary operator U that
shares the same basis set and has eigenvalues related through the complex
exponential.

2.4.4 Similarity Transforms

Frequently one has Hermitian operator H and orthonormal basis |pn  that
are not aligned. That is, the eigenvectors |an  of H are not parallel to vec-
tors |pn . Expansion of H into |pn  using the expansion expression (2.3.10)
50 2 The Spin-Vector Calculus of Polarization

generates a matrix that is not diagonal. However, the expansion matrix can
be diagonalized by rotating basis |pn  into |an . The unitary matrix does this
operation. Taking advantage of U † U = 1, one can write
 
p |H| p = p U † U HU † U  p
 
= a U HU †  a
= a |HT | a (2.4.19)

Since (2.4.19) holds for any choice of initial basis |pn , the operators

HT = U HU † (2.4.20)

must be equal. Equation (2.4.20) is known as a similarity transform. Both the

determinant and trace of H are independent of basis; that is

det(HT ) = det(U ) det(H) det(U † )

= det(H) (2.4.21)

and
Tr(HT ) = Tr(U HU † ) (2.4.22)
The trace is always preserved under a similarity transform.

2.4.5 Construction of General Unitary Matrix

The characteristic property T † T = 1 of unitary matrices restricts the eight

independent variables generally available for a polarization operator (2.3.8).
Derivation of the restrictions generates a general form of the unitary matrix.
First consider U , where det(U ) = +1 and . Substitution of X (2.3.8) for U
in U † U = 1 generates the following requirements:

|a|2 + |b|2 = 1, |a|2 = |d|2 , |b|2 = |c|2 ,

(2.4.23)
ac e−j(α−γ) + bd e−j(β−η) = 0

The determinant requirement generates

ad ej(α+η) − bc ej(β+γ) = 1 (2.4.24)

The amplitude restrictions in (2.4.23) are satisfied by a = cos κ and b = sin κ.

There remains, however, a sign degeneracy in that c = ±b and d = ±a. This
degeneracy is insignificant in that any choice flows through the restriction
criteria and produces the same matrix form.
Combination of the last equation in (2.4.23) and (2.4.24) generates two
restrictions on the phase:
 2.4 Eigenstates, Hermitian and Unitary Operators 51

ej(γ+β) = −ej(α+η)
ejα = e−jη (2.4.25)
ejγ = −e−jβ

There are only two independent phases. Combining all of the above restric-
tions, the general matrix form of U is written
 
ejα cos κ −ejβ sin κ
U =  (2.4.26)
e−jβ sin κ e−jα cos κ

There are three independent variables in U : one amplitude and two phases.
The fourth independent variable has been suppressed because of the arbitrary
selection det(U ) = +1. The unitary matrix T includes the common phase:
 
e cos κ −e sin κ
jα jβ
T = ejφ   (2.4.27)
e−jβ sin κ e−jα cos κ

where there are now four independent variables: one amplitude and three
phases.
The Cayley-Klein form of U , using complex entries a and b, is
 
a b
U =  (2.4.28)
−b∗ a∗

The inverse of the unitary matrix is U −1 (a, b) = U (a∗ , −b).

2.4.6 Group Properties of SU(2)

For polarization studies, unitary operators are 2 × 2 square matrices with

complex entries. The group defined by multiplication operations of 2 × 2 uni-
tary matrices is called U(2), and the subgroup of unitary matrices where
det(U ) = +1 is called SU(2), “S” for special. The group properties for multi-
plication are

Identity UI = U
Closure U1 U2 = U3
Inverse U −1 U = I
Associativity (U1 U2 )U3 = U1 (U2 U3 )

where in all cases U1,2,3 ∈ SU(2). SU(2) is closed under these four operations.
52 2 The Spin-Vector Calculus of Polarization

2.5 Vectors Cast in Jones and Stokes Spaces

Thus far, state spaces and operators have been presented without restriction
on their dimensionality. The properties of these vectors and matrices have
been studied in general with passing observations about polarization-specific
points of interest. At this stage the scope of presentation will concentrate
on the study of polarization so that a formal connection between Jones and
Stokes space can be established. The tools developed in the preceding sections
are essential to make the bilateral connections that follow.
Recall from (1.4.5) on page 13 that a polarization vector is written as
 
cos χ
|s = Eo ejθ   (2.5.1)
sin χejφ

where Eo is real. There are two polar angles in (2.5.1): χ and φ. The common
phase exp(jθ) is lost on conversion to Stokes space.
There are seven measurements necessary to determine the polarization
ellipse uniquely. The first measurement is for the overall intensity and the
remaining measurements project the ellipse onto six different reference axes.
The formal construction of a projection matrix is necessary at this point.
Consider points along two orthogonal axes and their projection onto a
line L inclined by angle θ that passes through the origin. As illustrated in
Fig. 2.2, the coordinate (1, 0) is projected to point a on line L. The coordinates
of a as measured along the two orthogonal axes are (cos2 θ, sin θ cos θ). After
a similar analysis for the coordinate (0, 1), one can construct the projection
matrix P:  
cos2 θ sin θ cos θ
P=  (2.5.2)
sin θ cos θ sin2 θ
It is clear that det(P) = 0; P is non-invertable and its action is irreversible.
There is loss of information after projection. Moreover, P 2 = P, so once
the projection is taken, subsequent projections along the same line L do not
change the result.

2.5.1 Complete Measurement of the Polarization Ellipse

There are seven measurements necessary for complete determination of the

polarization ellipse. The first measurement is one of total intensity, the remain-
ing six measurements are projections. The projections are defined in pairs and
the difference values are associated with the Stokes coordinates. The result of
an intensity of the polarization ellipse is the inner product
 
1 0
s |s = s | |s (2.5.3)
0 1
 2.5 Vectors Cast in Jones and Stokes Spaces 53

L
(0, 1) h i
a = cos u cos u
a sin u
u
h i
b
b = sin u cos u
sin u
(1, 0)

Fig. 2.2. Projection of unit coordinates (1, 0) and (0, 1) onto line L, which is inclined
by angle θ and passes through the origin. The projected coordinates are tabulated
on the right. A second projection of a and b onto L does not change the coordinates
of a and b. The projection operator is non-invertable.

Without loss of generality, s |s = 1 in the following.

The first projection pair is θ = 0 and θ = π/2. The projection measure
comes from the inner products
   
1 0 0 0
P0 = s | |s , and Pπ/2 = s | |s (2.5.4)
0 0 0 1

The Stokes parameter s1 is defined as

s1 = P0 − Pπ/2 (2.5.5)

Substitution of (2.5.4) into (2.5.5) makes

 
1 0
s1 = s | |s (2.5.6)
0 −1

The second projection pair is θ = ±π/4. These projections produce

   
1 1 1 −1
P+π/4 = 2 s |
1
|s , and P−π/4 = 2 s |
1
|s (2.5.7)
1 1 −1 1

The Stokes parameter s2 is defined as

s2 = P+π/4 − P−π/4 (2.5.8)

which makes  
0 1
s2 = s | |s (2.5.9)
1 0
The last projection requires the measurement of the ellipse circularity. By
convention, right-hand circular polarization rotates in the counter-clockwise
(ccw) direction when observed along the −ẑ direction (looking into the light).
The right-hand circular polarization vector is
 
1
|s R = (2.5.10)
j
54 2 The Spin-Vector Calculus of Polarization

The ccw vector needs mapping to the θ = 0 axis; a unitary transform does
the rotation. The right- and left-hand projections are calculated via

PR = s | U † P0 U |s (2.5.11a)
PL = s | U † Pπ/2 U |s (2.5.11b)

The unitary matrix  

1 1 −j
U= (2.5.12)
2 −j 1
maps right-hand circular polarization to the θ = 0 axis:
    
1 1 −j 1 1
= (2.5.13)
2 −j 1 j 0

Substituting (2.5.2) and (2.5.12) into (2.5.11) produces

   
1 −j 1 j
PR = 1
s | |s , and PL = 1
s | |s (2.5.14)
4 j 1 4 −j 1

The Stokes parameter s3 is defined as

s3 = PR − PL (2.5.15)

which makes  
0 −j
s3 = s | |s (2.5.16)
j 0
From these seven measurements one can transform from a ket in Jones
space to three Stokes coordinates that lie on the unit sphere:

|s =⇒ ŝ (2.5.17)

where the vector ŝ is a column vector defined by ŝ = (s1 , s2 , s3 )T .

This completes the measurement of the polarization ellipse. From these
measurements the polarimetric angles χ and φ are uniquely determined. These
measurements combined with the definition of the Stokes parameters generate
the Pauli spin matrices. This is the topic of the next section.

2.5.2 Pauli Spin Matrices

The Pauli spin matrices connect Jones to Stokes spaces through the projection
measurements of the preceding section. The identity Pauli matrix is
 
1 0
σ0 = (2.5.18)
0 1
 2.5 Vectors Cast in Jones and Stokes Spaces 55

The Pauli spin matrices are1

     
1 0 0 1 0 −j
σ1 = ; σ2 = ; σ3 = (2.5.19)
0 −1 1 0 j 0

The spin matrices are both Hermitian and unitary:

σk † = σk and σk † σk = I (2.5.20)

The determinants of the spin matrices are −1 and the traces zero:

det(σk ) = −1 and Tr(σk ) = 0 (2.5.21)

A spin matrix multiplied by itself yields

σk σk = I (2.5.22)

and multiplied by other matrices gives

σi σj = −σj σi = jσk (2.5.23)

where the indices of the multiplication table (i, j, k) are cyclic permutations
of (1, 2, 3).
Each Stokes coordinate of a polarization state |s is calculated by inserting
the associated Pauli matrix into the inner product s | · | s. The individual
Stokes coordinates are
sk = s |σk | s (2.5.24)
This is shorthand for the projection-difference measurements of (2.5.6, 2.5.9,
2.5.16). Since the spin matrices are Hermitian, the Stokes coordinates sk are
real, signed quantities. Moreover, since det(σk ) = −1 and the Jones vector |s
is assumed to be normalized, sk is bounded by −1 ≤ sk ≤ +1. The proof that
the norm of ŝ is unity, |ŝ| = 1, is shown below.

2.5.3 The Pauli Spin Vector and the Bilateral Connection

Between Jones and Stokes Vectors

The Pauli spin vector condenses further the notation of (2.5.24). The spin
vector is defined as  
σ1
σ =  σ2  (2.5.25)
σ3
1
In physics texts the z direction is denoted by the σ1 spin matrix while here it is
denoted by σ3 . Historically, the Pauli spin matrices describe electron spin, which
is either up or down in the “z” direction. In polarization optics, one usually thinks
of a horizontal polarization state aligned to the “x” axis.
56 2 The Spin-Vector Calculus of Polarization

where σ is a vector of matrices. The vector of Stokes coordinates ŝ is derived

from the Jones vector |s using the spin vector:
   
s1 s |σ1 | s
 s2  =  s |σ2 | s  (2.5.26)
s3 s |σ3 | s

More concisely,
ŝ = s |σ | s (2.5.27)
This is the most compact way to map Jones vectors to Stokes vectors.
The reciprocal connection is made through an eigenvalue equation whose
parameters are the Stokes vector ŝ and the spin vector. First, observe that the
spin vector behaves both as a 3 × 1 vector and as a 2 × 2 matrix, depending on
the context. Above shows the spin vector acting as a 3×1 vector. Alternatively,
the dot product of ŝ with the spin vector yields

ŝ · σ = s1 σ1 + s2 σ2 + s3 σ3
 
s1 s2 − js3
=  (2.5.28)
s2 + js3 −s1

ŝ · σ in this case is a 2 × 2 Jones matrix and, since the coefficients sk are real,
ŝ · σ is Hermitian: (ŝ · σ ) † = (ŝ · σ ).
Next, recall from §2.2.3 that the trace operation connects the projector
with its inner product: Tr(|ss|) = s |s. Since the trace of each Pauli matrix
is zero it is also true that Tr (ŝ · σ ) = 0. For a normalized state vector such
that s |s = 1, one can construct the projector for ket |s in terms of the spin
vector:
1
|ss| = (I + ŝ · σ ) (2.5.29)
2
Subsequent multiplication on the right by |s generates the eigenvalue equation

ŝ · σ |s = |s (2.5.30)

This is the most compact way to map Stokes vectors to Jones vectors. The
eigenvector of ŝ · σ associated with eigenvalue +1 generates the Jones vector
|s from Stokes vector ŝ.

2.5.4 Spin-Vector Identities

Vector operations that include spin-vectors do not yield to the same intu-
ition one is accustomed to with “normal” vectors. For example, while one is
quite familiar with a · (b × a) = 0, since a is orthogonal to b × a, the spin-
vector analogue produces σ · (a × σ ) = −2j(a · σ ). The difference comes from
 2.5 Vectors Cast in Jones and Stokes Spaces 57

the cyclic multiplication table for spin-vectors (2.5.22–2.5.23), where the sign
of a product is determined by the order in which the spin-vectors appear.
The purpose of the following identity tabulation is to provide reductions
in the order k of (σ )k . For the following identities, a and b are real-valued 3×1
vectors and σ is the spin vector. Real vectors a and b are not interchangeable
with the spin vector σ .
Identities of order (σ )0 and (σ ):

a · a = a2 (2.5.31)
a · σ = σ · a (2.5.32)
a(a · σ ) = (a · σ )a (2.5.33)
2
Identities of order (σ ) :

σ · σ = 3I (2.5.34)
σ (a · σ ) = aI + ja × σ (2.5.35)
(a · σ )σ = aI − ja × σ (2.5.36)
(a · σ )(a · σ ) = a2 I (2.5.37)
(a · σ )(b · σ ) = (a · b)I + (ja × b) · σ (2.5.38)
[(a · σ ), σ ] = −2ja × σ (2.5.39)
{(a · σ ), σ } = 2a I (2.5.40)
 
(a · σ ), (b · σ ) = 2(ja × b) · σ (2.5.41)
 
(a · σ ), (b · σ ) = 2(a · b) I (2.5.42)

σ · (ja × σ ) = 2(a · σ ) (2.5.43)

(ja × σ ) · σ = −2(a · σ ) (2.5.44)
(ja × σ )(a · σ ) = a σ − a(a · σ )
2
(2.5.45)
(a · σ )(ja × σ ) = a(a · σ ) − a σ
2
(2.5.46)
where [A, B] = AB − BA is the commutator and {A, B} = AB + BA is the
anti-commutator.
Identities of order (σ )3 :

σ · ((a · σ )σ ) = (σ (a · σ )) · σ = −(a · σ ) (2.5.47)

σ · (σ (a · σ )) = ((a · σ )σ ) · σ = 3(a · σ ) (2.5.48)
(a · σ )σ (a · σ ) = 2a(a · σ ) − a σ
2
(2.5.49)
58 2 The Spin-Vector Calculus of Polarization

Identity of order (σ )n :


 an n even
(a · σ )n = (2.5.50)
 an−1 (â · σ ) n odd

Finally, there are identities that relate to inner products taken with various
forms of the spin vector. These identities are as follows:

s |a · σ | s = a · s |σ | s = a · ŝ (2.5.51)

s |a × σ | s = a × s |σ | s = a × ŝ (2.5.52)

s |Rσ | s = Rs |σ | s = Rŝ (2.5.53)
where a and b are arbitrary vectors in Stokes space, a is the length of a,
and R is a 3x3 matrix. Identity (2.5.53) is always a source of confusion, so it
is repeated explicitly:
    
r11 r12 r13 s |σ1 | s s |r11 σ1 + r12 σ2 + r13 σ3 | s
    
    
 r21 r22 r23   s |σ2 | s  =  s |r21 σ1 + r22 σ2 + r23 σ3 | s 
    
r31 r32 r33 s |σ3 | s s |r31 σ1 + r32 σ2 + r33 σ3 | s

where Rŝ on the left and s |Rσ | s on the right.

2.5.5 Conservation of Length

Expressions (2.5.27) and (2.5.30) complete the bilateral connection between

Jones and Stokes vectors. Length must be conserved, of course, and this is
verified now.
The Stokes-vector length is derived from the product ŝ · ŝ = s21 + s22 + s23 .
Consider one coordinate alone,

s2k = s |σk | ss |σk | s (2.5.54)

Substitution of the projector |ss|, (2.5.29), for the innermost term gives
1 1
s2k = s |s + s |σk (ŝ · σ )σk | s (2.5.55)
2 2
The sum of all three terms gives
3 1
s21 + s22 + s23 = s |s + s |σ · ((ŝ · σ )σ )| s (2.5.56)
2 2
The spin-vector identity (2.5.48) simplifies (2.5.56):
3 1
s21 + s22 + s23 = s |s − s |ŝ · σ | s = s |s (2.5.57)
2 2
 2.5 Vectors Cast in Jones and Stokes Spaces 59

Thus,  s 2 = s |s. Length is clearly preserved in this direction. For the

reverse mapping, construction of (2.5.30) without the assumption s |s = 1
produces
s · σ |s = s |s |s −→ ŝ · σ |s = |s (2.5.58)
That length is conserved over the bilateral connections is thus established.
There is, however, one piece of information that is lost in the mapping
from Jones to Stokes. Since Stokes coordinates are derived from intensity
measurements, the common phase of the Jones vector is lost. Transformation
from Stokes back to Jones does not reintroduce this phase. Any Jones vector
constructed from a Stokes vector is accurate to the “true” Jones vector to
within an arbitrary common phase. Physically this just means that the abso-
lute time it took for the light to travel from its source to the observer cannot
be determined from Stokes measurements.

2.5.6 Orthogonal Polarization States

For every polarization state |s+  there is a unique polarization state |s−  such
that s− |s+  = 0. These states |s+  and |s−  are orthogonal. Given |s+  how
does one can construct the orthogonal state |s−  and its Stokes equivalent?
From (2.5.30) on page 56 one writes
  
s− |s+  = s− | (ŝ− · σ )† (ŝ+ · σ ) |s+  = 0 (2.5.59)

Since (ŝ− · σ ) is Hermitian, (2.5.59) is rewritten using spin-vector iden-

tity (2.5.38) as

s− |s+  = (ŝ− · ŝ+ )s− |s+  + js− |(ŝ− × ŝ+ ) · σ | s+  (2.5.60)

As s− |s+  = 0, (2.5.60) requires that (ŝ− × ŝ+ ) = 0. There are two ori-
entations that produce (ŝ− × ŝ+ ) = 0: ŝ− · ŝ+ = ±1. If ŝ− · ŝ+ = +1, then
s− |s+  = 1, contradicting the orthogonality of the two states. Therefore it
must be the case that
ŝ− · ŝ+ = −1 (2.5.61)
The Stokes coordinates for any two orthogonal polarization states are on op-
posite sides of the Poincaré sphere: ŝ− = −ŝ+ . Specifically, a chord that con-
nects any two orthogonal states crosses through the origin of the sphere. The
polarimetric parameters χ and φ are related through the Stokes vectors as
   
cos 2χ− cos 2χ+
   
   
 sin 2χ− cos φ−  = −  sin 2χ+ cos φ+  (2.5.62)
   
sin 2χ− sin φ− sin 2χ+ sin φ+

A sufficient requirement to satisfy (2.5.62) is

60 2 The Spin-Vector Calculus of Polarization

a) b) S3
o ^
90 s+
jp- i

S2
a
2a
S1
jp+i
^
s-

Fig. 2.3. Orthogonal polarization states in Jones and Stokes space. a) The hand-
edness of the polarization ellipse is reversed and the major axis is rotated by π/2.
b) Points on opposite sides of the Poincaré sphere are orthogonal.

2χ− = 2χ+ + π (2.5.63a)

φ− = φ+ (2.5.63b)

Equations (2.5.61) and (2.5.63) show that orthogonal polarization states have
opposite handedness and perpendicular orientations of the respective elliptical
major axes. The Jones and Stokes representation of orthogonal polarization
pairs is illustrated in Fig. 2.3.

2.5.7 Non-Orthogonal Polarization States

The inner product magnitude between two polarization states may be calcu-
lated either in Jones or Stokes space. Consider two Jones vectors |p and |q
that are not normalized, and recall that Tr (|p q |) = p |q. In a manner sim-
ilar to (2.5.29), the inner product between the two Jones vectors is written
1
|pp| = (I + p · σ ) p |p (2.5.64)
2
Multiplication on the right by |q and on the left by q |, and some rearrange-
ment, makes
2
|p |q| 1
= (1 + p · q) (2.5.65)
p |p q |q 2
When |p and |q are normalized, the identity reduces to

2 1
|p |q| = (1 + p̂ · q̂) (2.5.66)
2
The magnitude of the inner product in Jones space is derived directly from
the Stokes vectors using this equation. What cannot be discerned, however, is
 2.5 Vectors Cast in Jones and Stokes Spaces 61

the phase of the inner product. To recover the phase, p |q must be calculated
explicitly in Jones space. To construct the Jones vectors, one must either solve
the eigenvalue equation (2.5.30) or make the Jones vector (1.4.19) on page 18
for both |p and |q

2.5.8 Pauli Spin Operators

A general operator can be constructed from the identity matrix and spin-
vector by the form

A = a0 I + a · σ
= a0 I + a1 σ1 + a2 σ2 + a3 σ3
 
a0 + a1 a2 − ja3
=  (2.5.67)
a2 + ja3 a0 − a1

where all ak are complex numbers. This matrix has the eight requisite inde-
pendent variables necessary for a general Jones matrix. The entries in A are
isolated by the trace:

1 1 (2.5.68)
a0 = 2 Tr(A) and a = 2 Tr (Aσ )

A Hermitian operator is a special case of A:

H = a0 I + a · σ , ak real (2.5.69)
 
The determinant is det(H) = a20 − a21 + a22 + a23 . Moreover, when the trace
of H is zero, the Hermitian matrix equals a spin-vector form

HTr=0 = a · σ (2.5.70)

Throughout much of this text, Hermitian operators with zero trace, and op-
erators that preserve that trace, are associated with the spin-vector form.
The general operator A can be decomposed into Hermitian and skew-
Hermitian matrices
A = Hr + jHi (2.5.71)
where the operator K = (jHi ) is skew-Hermitian: K † = −K. The eigenvalues
of skew-Hermitian matrices are purely imaginary. The matrices Hr and Hi
contain the real and imaginary parts of A, respectively. The decomposition
is taken further by separating the finite-trace component from the traceless
components. Writing the complex number a0 as a0 = a0 + ja0 and identifying
each traceless Hermitian matrix with a spin-vector form, one has

A = ar · σ + jai · σ + (a0 + ja0 ) I (2.5.72)

62 2 The Spin-Vector Calculus of Polarization

This is the most general spin-vector form of an arbitrary operator A. In prac-

tice, the decomposition matrices are derived from A as follows:
   
Hr = 1
2 A + A† , and Hi = − 2j A − A† (2.5.73)

The matrices Hr and Hi are made traceless by calculating a0 = 12 Tr(H) for

the real and imaginary components. The real-valued Stokes vectors ar and σi
can be read from the matrices ar,i · σ = Hr,i − ar,i,0 I.
The Pauli spin operator S produces the most compact way to describe op-
erators and concatenations of operators. The spin operator is a matrix expo-
nential form of A and can describe Hermitian, unitary, and general operators.
The matrix exponential is written

S = exp (A/2) (2.5.74)

The exponential is evaluated using its Taylor expansion. For instance,

1 1 1
exp(M ) = I + M + M 2 + M 3 + M 4 + . . .
2! 3! 4!
   
1 1 1 1
= I + M2 + M4 + . . . + M + M3 + M5 + . . .
2! 4! 3! 5!
(2.5.75)

For a Hermitian matrix, M = αo I + ( α · σ ) where αk are real. The nth -order

n
spin-vector identity (2.5.50) gives the necessary reductions for ( α · σ ) , which
gives
α · σ /2) = I cosh (α/2) + (α̂ · σ ) sinh (α/2)
exp ( (2.5.76)
where α
 = αα̂. The Pauli spin operator for a Hermitian operator is thus

SH = exp (α0 /2) exp (

α · σ /2)
 
= eα0 /2 I cosh (α/2) + (α̂ · σ ) sinh (α/2) (2.5.77)

One can interpret this operator as that for a partial polarizer: the common
loss is α0 /2 (which is a negative quantity for loss), the maximum and mini-
mum differential losses are 1 ± tanh α/2, and the Stokes direction of partial
polarization is α̂.
The Pauli spin operator for a unitary matrix is constructed by recalling
the connection between Hermitian and unitary operators (2.4.18) on page 49.
For coefficients βk real, the unitary form of M is M = −j(βo I + (β  · σ )). The
equivalent to (2.5.76) is
 
exp −j β  · σ /2 = I cos (β/2) − j(β̂ · σ ) sin (α/2) (2.5.78)

 = β β̂. The Pauli spin operator for a unitary operator is thus

where β
 2.6 Equivalent Unitary Transformations 63
 
ST = exp (−j β0 /2) exp −j β · σ /2
 
= e−j β0 /2 I cos (β/2) − j(β̂ · σ ) sin (β/2) (2.5.79)

One interprets this operator as a retardation plate: −β0 /2 is the common

phase, the full retardance is β, and the axis of retardation is β̂. When the
common phase is removed the unitary matrix U is recovered.
In the particular case where the axes of polarization and retardance co-
incide, the compound effect is expressed as the spin vector (−j β +α  ) · σ .
Following the form of (2.5.76) the Pauli spin operator expands to matrix form
as
 
+α
exp ((−j β0 + α0 )/2) exp (−j β  ) · σ /2 =
 
e(−jβ0 +α0 )/2 I cosh ((−jβ + α)/2) + (r̂ · σ ) sinh ((−jβ + α)/2) (2.5.80)

where r̂ is the axis of polarization and retardance.

Two relevant properties of matrix exponentials are
∂ At
e = AeAt = eAt A (2.5.81)
∂t
∂ At Bt Ct
e e e = AeAt eBt eCt + eAt BeBt eCt + eAt eBt eCt C (2.5.82)
∂t
while two non-intuitive results are
∂ At
e = teAt (2.5.83)
∂A
eAt eBt = e(A+B)t (2.5.84)

unless, for the second equation, A and B commute.

When one writes a compound operation such as

H1 U H2 = exp ((  · σ )/2) exp ((

α1 · σ )/2) exp(−j(β α2 · σ )/2) (2.5.85)

one is only figuratively expressing the series of operations on a polarization

state. The evaluation of H1 U H2 requires substitution of the matrix form for
each operator.

2.6 Equivalent Unitary Transformations

The preceding sections have established the bilateral connection between
Jones and Stokes vectors, and have constructed Hermitian, unitary, and gen-
eral Jones operators that act on Jones vectors. The connection between a
Hermitian matrix and an equivalent Stokes matrix is made with the Mueller
64 2 The Spin-Vector Calculus of Polarization

matrix, (1.4.22) on page 18 and figuratively (2.1.3) on page 38. Mueller ma-
trices operate on 4 × 1 Stokes vectors to create new, transformed 4 × 1 Stokes
vectors.
The connection between a unitary matrix and an equivalent Stokes matrix
is also made with the Mueller matrix, but as indicated by (2.1.1) on page 38,
only the lower right 3 × 3 sub-matrix R is relevant. Sub-matrix R maps
spherical coordinates (S1 , S2 , S3 ) into new spherical coordinates (S1 , S2 , S3 )
without change of length. Therefore one expects the existence of a rotation
operator R corresponding to matrix R that performs rotations on the Poincaré
sphere. The operator R does indeed exist and its derivation and properties are
so central to the description of retardance that this entire section is devoted
to its understanding.
Operators U and R are equivalent representation of the same transforma-
tion cast in two different vector spaces. The operators are called isomorphic
because they have similar, but not equal, effects. The isomorphism is two-to-
one since, as well be seen, there are two Jones operations that have the same
effect as every one Stokes operation.
Consider equivalent vectors |s and ŝ at the input of a system and equiv-
alent vectors |t and t̂ at the output. In Jones space a unitary transforma-
tion T , corresponding to the underlying Maxwell’s equations in anisotropic
media, links the input and output. In Stokes space the rotation operator R
links the input and output. The parallel transformations are
dual
|t = T |s ←−−−→ t̂ = R ŝ (2.6.1)

Expansion of the Stokes vectors on the right side of (2.6.1) into their corre-
sponding inner products gives the relation between R and T

t |σ | t = Rs |σ | s (2.6.2)

Replacing |t with T |s and applying identity (2.5.53) gives

 
s T †σ T  s = s |R σ | s (2.6.3)

Since (2.6.3) holds for any |s, the embedded operators must be equal. There-
fore
R σ = U †σ U (2.6.4)
where the common phase of T commutes with σ and is eliminated. Equa-
tion (2.6.4) has an unusual form; the interpretation is
    
σ1 U † σ1 U
    
    
 3 × 3   σ2  =  U † σ2 U  (2.6.5)
    
†
σ3 U σ3 U

That R is unitary is derived by multiplying (2.6.4) by its adjoint:

 2.6 Equivalent Unitary Transformations 65

σ † R† R σ = U †σ † U U †σ U (2.6.6)

The product σ †σ = 3I, so only if R† R = I does (2.6.6) hold. Therefore,
R† R = RR† = I (2.6.7)
Since R is unitary it embodies a pure rotation; there is no scaling or trans-
lation; t̂ is related to ŝ by a rotation in Stokes space. The group properties
of R and their correspondence to the group properties of U are listed in the
following section.
There are two ways to derive an explicit expression for R, either through
the matrix form of U , generating a matrix form of R; or through the Pauli
spin operator form of U , generating a vector form of R. The vector form
is more “lightweight” and powerful than the matrix form because successive
operations in Stokes space are evaluated purely in vector form without an
a priori choice of orthonormal basis. The vector form is also a template with
which to construct any matrix R without matrix multiplication.

2.6.1 Group Properties of SU(2) and O(3)

The group defined by multiplication operators of R is called O(3), where O

stands for orthogonal and (3) for the three rotational dimensions of R. The
group properties for multiplication mirror those for SU(2), cf. §2.4.6:

Identity UI = U RI = R
Closure U1 U2 = U3 R1 R2 = R3
−1
Inverse U U =I R−1 R = I
Associativity (U1 U2 )U3 = U1 (U2 U3 ) (R1 R2 )R3 = R1 (R2 R3 )

where in all cases U1,2,3 ∈ SU(2) and R1,2,3 ∈ O(3).

To confirm the entries in the above table, note that multiplication of suc-
cessive U and R operators form a one-to-one correspondence. For example,
dual
|t = T2 T1 |s ←−−−→ t̂ = R2 R1 ŝ (2.6.8)

Operator equality is generated through the expression

R2 R1σ = U1 † U2 †σ U2 U1 (2.6.9)
Making the correspondences R = R2 R1 and U  = U2 U1 , then (2.6.9) is equiv-
alent to (2.6.4). Given that U2 U1 ∈ SU(2) and U  ←→ R , one infers closure
on O(3).
There are twice as many entries in the group SU(2) as in O(3). For ev-
ery Uo , the corresponding Ro is Roσ = Uo †σ Uo . For every −Uo the corre-
sponding operator Ro is the same: Roσ = (−Uo )†σ (−Uo ). The isomorphism
between SU(2) and O(3) is two-to-one.
66 2 The Spin-Vector Calculus of Polarization

2.6.2 Matrix Entries of R in a Fixed Coordinate System

One way to generate R explicitly is through the matrix form of U . Earlier,

the generation of matrix entries for the Mueller matrix from entries in the
Jones matrix was given without derivation, (1.4.22) on page 18. That equation
created a 4 × 4 matrix; only for a unitary matrix is a distinct 3 × 3 sub-matrix
formed. The matrix entries for M are found as follows.
Start with the two relations t̂j = t |σj | t and t̂j = Tr (M σj ), equations
(2.5.24) on page 55 and (2.5.68) on page 61, respectively. The index j is
for j = 0, 1, 2, 3. Moreover, assume the system |t = J |s. Identification with
t |t = Tr(|tt|) gives

tj = t |σj | t
= Tr (|tt| σj )
 
= Tr J |ss| J † σj (2.6.10)

Next, the outer product |ss| is replaced with something close to the spin-
vector form (2.5.29) on page 56. Since the vector is not necessarily normalized,
the expression |ss| = 12 s |s (I + ŝ · σ ) will be used. Thus,
   
Tr J |ss| J † σj = 1
2 s |s Tr J (I + ŝ · σ ) J † σj
 
= 1
2 s |s Tr J(ŝ · σ )J † σj

1
 
3
= Tr Jσk J † σj sk (2.6.11)
2
k=0

where σ has been loosely indexed here to include σ0 , sk = s |s ŝk , and the
common phase in T commutes with (ŝ·σ ) and is eliminated. The last equation
has the matrix multiplication form

3
tj = Mj+1,k+1 sk (2.6.12)
k=0

Identification of the matrix entries gives the final expression

1  
Mj+1,k+1 = Tr Jσk J † σj (2.6.13)
2
The specialized case of J = U is of immediate interest. Substitution of U
for J in (2.6.13) shows that for k = 0, j = 0, and for j = 0, k = 0, the matrix
entries are identically zero. This is because Tr(σj ) = 0. Other than the M1,1
matrix entry, which is unity, only the sub-matrix R survives the trace. Ex-
plicitly, the matrix entries for R given a matrix form of U are
1  
Rj,k = Tr U σk U † σj (2.6.14)
2
 2.6 Equivalent Unitary Transformations 67

Table 2.1. Elementary Rotations in Jones and Stokes Space

 
  1
ejα  
 
κ=0 U =  R1 =  cos 2α sin 2α 
e−jα  
− sin 2α cos 2α

 
  cos 2κ sin 2κ
α=0 cos κ −j sin κ  
U =  R2 = 
 1


β = π/2 −j sin κ cos κ  
− sin 2κ cos 2κ

 
  cos 2κ − sin 2κ
cos κ − sin κ  
 
α=β=0 U =  R3 =  sin 2κ cos 2κ 
sin κ cos κ  
1

Using the general form of U given in (2.4.26) on page 51, R is resolved as

 
cos 2κ − cos(α − β) sin 2κ − sin(α − β) sin 2κ
R= cos(α + β) sin 2κ cos 2α cos2 κ − cos 2β sin2 κ sin 2α cos2 κ + sin 2β sin2 κ 
− sin(α + β) sin 2κ − sin 2α cos2 κ + sin 2β sin2 κ cos 2α cos2 κ + cos 2β sin2 κ

(2.6.15)

Calculation of the determinant gives

det(R) = 1 (2.6.16)

which verifies that R is invertible, unitary, and contains no reflections.

Table 2.1 gives the three elementary rotations about the Stokes axes
(S1 , S2 , S3 ) and their original unitary form. Notice that all angles which ap-
pear in the unitary matrices are doubled in the corresponding Stokes matrices.
Stokes angles are twice that of physical, or “laboratory” angles. This is also
why the Jones to Stokes isomorphism is two-to-one: rotation of π in Jones
space is invariant, and so is rotation by 2π in Stokes space.

2.6.3 Vector Expression of R in a Local Coordinate System

The vector form of R abstracts away any notion of an underlying, fixed coor-
dinate system. Rather, each operation R has its own local coordinate system
based on the eigenvectors and spin direction of R. The vectorial form of R
gives the highest level of geometric interpretation to transformation mechanics
in Stokes space.
68 2 The Spin-Vector Calculus of Polarization

The vector expression for R is derived from the vector form of U . The
operator U is resolved into its eigenvector-based projectors using (2.4.15) on
page 48; the resolution for a two-dimensional system gives

U = e−jϕ/2 |r+ r+ | + ejϕ/2 |r− r− | (2.6.17)

where the projectors are equated to the spin-vector form via

1
|r± r± | = (I ± r̂ · σ ) (2.6.18)
2
Note that r̂+ = −r̂− are orthogonal Stokes coordinates. In the following, r̂
will denote the eigenvector with the positive subscript. Substitution of the
spin-vector form of the projector into U gives

U = I cos(ϕ/2) − j(r̂ · σ ) sin(ϕ/2) (2.6.19)

Equation (2.6.19) is now in familiar form and can be mapped to the exponen-
tial equivalent as
U = e−j(ϕ/2)(r̂·
σ) (2.6.20)
where (ϕ/2)(r̂ · σ ) is the Hermitian operator associated with U .
Substitution of (2.6.19) into the equivalence relation (2.6.4), and applying
the spin-vector identities (2.5.35), (2.5.36), and (2.5.49) produces

R σ = cos ϕσ + (1 − cos ϕ)r̂(r̂ · σ ) + sin ϕ(r̂ × σ ) (2.6.21)

Since each term on the left- and right-hand sides of (2.6.21) operates on σ ,
one can extract the embedded relation for R

R = cos ϕI + (1 − cos ϕ)(r̂r̂·) + sin ϕ(r̂×) (2.6.22)

Grouping like terms gives

R = (r̂r̂·) + sin ϕ(r̂×) + cos ϕ(I − (r̂r̂·)) (2.6.23)

Recalling the vector identity a × a × c = b(a · a) − c(a · b), the last term on
the right-hand side is identified as

(r̂×)(r̂×) = (r̂r̂·) − I (2.6.24)

The final vectorial form of R is then

R = (r̂r̂·) + sin ϕ(r̂×) − cos ϕ(r̂×)(r̂×) (2.6.25)

Equation (2.6.25) is a beautifully compact expression for the action any uni-
tary operator has on a polarization vector. The vector r̂ points in the direction
of the positive eigenvector of U . The vector operators {(r̂r̂·), (r̂×), (r̂×)(r̂×)}
form a local orthonormal basis. The local basis requires a vector about which
 2.6 Equivalent Unitary Transformations 69

a) b)
^
r
^
^ t
r
^^
rr. ^ ^ ^
(rx)(rx) s

^
^ (rx)
s w

Fig. 2.4. Vector components of rotation operator R. a) The local orthonormal basis
{(r̂r̂·), (r̂×), (r̂×)(r̂×)} as resolved on ŝ. b) Transformation to t̂ from ŝ via precession
about r̂, travelling through precession angle ϕ.

the basis can be fully resolved; for instance, operation on state ŝ generates
the basis (r̂, r̂ × ŝ, r̂ × r̂ × ŝ). In the absence of being fully resolved, the local
basis has immutable properties that are independent of the resolving vector.
Figure 2.4(a) illustrates the local basis resolved by ŝ. Vector ŝ in relation
to r̂ defines a precession circle, the circle about which ŝ travels. Local axis (r̂r̂·)
always points parallel to r̂. The local axes (r̂×), (r̂×)(r̂×) define the plane
of the precession circle and are perpendicular to r̂. The local axis (r̂×) is
tangent to the precession circle and (r̂×)(r̂×) points to the origin of the
precession circle. The particular pointing directions of (r̂×) and (r̂×)(r̂×),
while always in the precession plane, are determined only after determination
of ŝ. Figure 2.4(b) illustrates transformation to state t̂ from ŝ about r̂. The
precession angle is ϕ and the precession direction follows the right-hand rule.
Since the motion of precession is so central in the description of polar-
ization transformation mechanics, Fig. 2.5 is included to describe precession
in a local coordinate system. Consider the input state ŝ and the precession
axis r̂. The precession axis can be the birefringent axis of a dielectric medium
or the principal-state-of-polarization axis used to describe polarization-mode
dispersion. In any case, the angle γ separates the two vectors. The motion of
precession is to turn ŝ about r̂ in a circle while keeping the angle γ fixed. This
is the same motion a gyroscope exhibits under gravitational influence. The
angle subtended by projections of states ŝ and t̂ onto the base circle is the
precession angle. The differential equation of motion can be deduced from R
in local-coordinate form. Consider state ŝ that undergoes a small change in
angle δϕ. The motion is
ŝ + δŝ = Rδϕ ŝ (2.6.26)
Taking R in the form (2.6.22) and simplifying for small angles,

ŝ + δŝ = I ŝ + δϕ r̂ × ŝ (2.6.27)
70 2 The Spin-Vector Calculus of Polarization
^
r
^
^
t
s

g g ^
d s = ^r x ^s
dw

ws-t

Fig. 2.5. Precessional motion of ŝ about r̂, passing through state t̂. Angle γ remains
fixed, while angle ϕ, as projected onto the base, is the degree of precession.

which is rewritten in differential form as

dŝ
= r̂ × ŝ (2.6.28)
dϕ
The r̂ × ŝ term dictates that ŝ moves perpendicular to r̂.
Equation 2.6.22 can be used as a template to construct a matrix represen-
tation for R given any r̂ and ϕ. The matrix representations for r̂r̂· and r̂×
are
   
r1 r1 r1 r2 r1 r3 0 −r3 r2
   
   
(r̂r̂·) =  r2 r1 r2 r2 r2 r3  , (r̂×) =  r3 0 −r1  (2.6.29)
   
r3 r1 r3 r2 r3 r3 −r2 r1 0

The first matrix is a projector, as verified by det(r̂r̂·) = 0. The second matrix is

derived from r̂× = r̂ × S1 + r̂ × S2 + r̂ × S3 , where S1,2,3 are the three Stokes
axes.

2.6.4 Select Vector Identities

Figure 2.6 illustrates two identities that have a simple geometric interpreta-
tion. The identity
r̂ · (R a ) = r̂ · a (2.6.30)
is illustrated by Fig. 2.6(a). The rotation of a generated by R about r̂ through
angle ϕ does not change the angle between r̂ and a. Thus the dot product
may be taken with or without the intermediate rotation. The identity
(R a ) · (R b ) = a · b (2.6.31)
is illustrated by Fig. 2.6(b). The rotation due to R does not change the angle
between vectors a and b, so again the dot product may be taken with or
without the intermediate rotation.
 2.6 Equivalent Unitary Transformations 71
^ ^
a) Ra
^ r b) Ra
^ r
^ ^
a a
^
Rb ^

g g b

d d

Fig. 2.6. Geometric representation of two rotational identities. a) r̂ · (R a ) = r̂ · a.

b) (R a ) · (R b ) = a · b.

2.6.5 Euler Rotations

The Euler rotations are an alternative method to construct the operator R in

matrix form (2.6.15). While there are several ways to establish the connection,
the fact that multiplication in Stokes space corresponds to multiplication in
Jones space is simple enough to construct the operator U in the form (2.4.26),
and to map the terms from Jones back to Stokes space. One can verify that
   
eju cos κ − sin κ ejv
U =    (2.6.32)
e−ju sin κ cos κ e−jv

where u = (α + β)/2 and v = (α − β)/2. Identification of the Jones operators

in (2.6.32) with the Stokes operators in Table 2.1 gives the equivalent Stokes
transformations
R = R1 (α + β) R3 (2κ) R1 (α − β) (2.6.33)
These rotations generate the general rotation matrix in (2.6.15).
Another way to view the unitary operator is to diagonalize the matrix. The
eigenvalues of U are complex exponentials that have unity magnitude and they
are conjugates of one another: U |r±  = exp(∓jϕ/2) |r± . The operator U can
be separated as
U = V ΛV † (2.6.34)
where the matrix V is a 2 × 2 unitary matrix with the two eigenvectors |r± 
of U entered as columns of V , and where the matrix Λ is a diagonal 2 × 2
matrix with the eigenvalues of U on the diagonal. The corresponding Stokes
operation is
R = RE R1 (ϕ)R† E (2.6.35)
where RE is the Euler rotation associated with V . Now, several observations
can be made. If the state at the input is |r± , then the action of U does not
72 2 The Spin-Vector Calculus of Polarization

change the state, only a phase is contributed. The state is invariant under U .
For every |r±  there are corresponding r̂± vectors in Stokes space. The be-
havior of R† E is to rotate r̂± to ±s1 ; R1 (ϕ) then pirouettes the state about
the s1 axis, and RE returns the state back to r̂± .
The decomposition of U as in (2.6.34) has much significance in relation to
propagation through birefringent media. For example, the propagation con-
stants for ordinary and extraordinary waves in a birefringent medium are
βo = ωno /c and βe = ωne /c. The eigenvalues of the propagation matrix are
exp(∓j(βe − βo )z/2). The polarization transformation in Stokes space is ac-
cordingly
R∆β = RE Rx (∆βz)R† E (2.6.36)
The inner matrix Rx creates precession about the s1 axis in Stokes space while
the Euler rotation and its adjoint transforms the eigenstates of the system
onto the s1 axis and then restores the pointing direction of the eigenstate.
The precession about s1 is transformed to precession about r̂.

2.6.6 Some Relevant Transformation Applications

Four examples are presented here to give some illustrative detail on how to
use the Stokes transformation operator R cast in local-coordinate form. First,
differential precession rules for a single homogeneous birefringent section are
written. Then, the polarization evolution through a concatenation of two mis-
aligned birefringent sections, as a function of length, is illustrated. Third, the
shortest distance between two polarization states is found. Finally, uniform
and biased polarization scattering examples are given.
For polarization studies, the axis r̂ is the extraordinary axis of a bire-
fringent medium. For a birefringent crystal the birefringent axis lies in the
equatorial plane in Stokes space. An input state ŝ precesses about r̂ as the
state propagates through the medium. The retardation of a birefringent plate
is just the angle ϕ through which the state processes: ϕ = ω∆nL/c, albeit
any two rotations of ϕ that are the same modulo 2π yield identical Stokes
transformations. The angle ϕ is called the birefringent phase.
The differential equation of motion for a state of polarization as it evolves
through a homogeneous birefringent material is the same for differentials in
either position or frequency. The retardation as a function of length z and
radial frequency ω is
ω∆nz
ϕ= (2.6.37)
c
Moreover, the birefringent axis Ω is parallel to r̂. Differentiation of (2.6.37)
with respect to z and substitution into (2.6.28) gives
dŝ
= Ω × ŝ (2.6.38)
dz
where Ω = (ω∆n/c)r̂. Equally possible is differentiation with respect to ω,
which gives
 2.6 Equivalent Unitary Transformations 73

a) S3 b) S3
^
r
c

S2 ^ S2
s
a

^
r2 S1 S1
^
r1 ws-t
^
t
w1 b
w2

Fig. 2.7. a) Polarization evolution through two misaligned birefringent sections.

Input state (a) precesses about r̂1 to state (b). That state enters the second stage,
precesses about r̂2 , and leaves as state (c). b) Construction of great circle through
states ŝ and t̂. The normal to the circle is r̂.

dŝ
= Ω × ŝ (2.6.39)
dω
The differential precession rule for a single homogeneous birefringent section
is the same whether the position or frequency changes. This simplicity is
quickly broken when two or more homogeneous sections are concatenated.
Birefringent concatenation is in the category of polarization-mode dispersion.
The polarization state evolution through two misaligned birefringent sec-
tions as a function of length can be evaluated using (2.6.25) in the following
way. Since the media are misaligned, their birefringent axes are not parallel;
that is, r̂1 = r̂2 . Figure 2.7(a) illustrates the polarization evolution through
the sections. The input state, arbitrarily selected, is located at position (a).
That state precesses about r̂1 through angle ϕ1 , dictated by the length and
birefringence of the section, as well as the input frequency. The output polar-
ization from the first section is located at position (b). That state then enters
the second section which transforms it about r̂2 . The polarization state now
traces a second circle that is different from the first. The output state is even-
tually located at position (c). The aggregate polarization transformation is
calculated by the concatenation of R2 and R1 . The compounded polarization
transformation is
 
R2 R1 = (r̂2 r̂2 ·) + sin ϕ2 (r̂2 ×) − cos ϕ2 (r̂2 ×)(r̂2 ×) (2.6.40)
 
(r̂1 r̂1 ·) + sin ϕ1 (r̂1 ×) − cos ϕ1 (r̂1 ×)(r̂1 ×) (2.6.41)

Each transformation is denoted by a unique index and the concatenation

is written right-to-left as is usual for matrix multiplication. Sometimes the
vector products offer simplifications that can reduce the complexity of the
overall motion.
74 2 The Spin-Vector Calculus of Polarization

a) S3 b) ^
so S3

S2 S2

S1 S1

Fig. 2.8. Uniform and biased scattering through operator R. a) Uniform scat-
tering. r̂ points in any direction with equal likelihood. ϕ is uniformly distributed.
b) Biased scattering, a = 0.05. R̃ is constructed along s3 and oriented toward ŝo .

As another application, the shortest distance between points ŝ and t̂ on

a unit sphere is along a great circle. The axis normal to the great circle is
evidently
ŝ × t̂
r̂ = (2.6.42)
|ŝ × t̂|
and the rotation angle between the two points is

cos ϕs−t = ŝ · t̂ (2.6.43)

Figure 2.7(b) illustrates the motion. r̂ is derived from the cross of ŝ and t̂.
Angle ϕs−t rotates ŝ through to t̂.
Uniform and biased polarization scattering is useful in connection with
polarization-mode dispersion fiber-modelling calculations. The scattering pro-
cess occurs between any two adjacent birefringent sections and it intended to
model the relative alignments of the respective birefringent axes. A uniform
scattering process sends the polarization state at the output of one section, ŝo ,
to any point on the Poincaré sphere with equal probability. That state, ŝi , is
then input to the next birefringent section. The biased scattering process
weights the scattering along a predetermined direction, often the direction
of ŝo . For either uniform or biased scattering an operator R needs to be con-
structed.
There are two variables contained in R, (2.6.25): pointing direction r̂ and
precession angle ϕ. Direction r̂ itself has two independent variables, the po-
lar angles of declination and azimuth. Combined, R has three independent
variables. The random process is derived using the unit deviate ũ. To have r̂
point in any direction on the unit sphere with equal likelihood, the azimuth
angle φ and position along the s3 axis are both uniformly distributed. Also, the
precession angle is uniformly distributed to generate precessions with equal
 2.6 Equivalent Unitary Transformations 75

likelihood. The random variable expressions are

r̃3 = 2ũ − 1 (2.6.44a)

φ̃ = (2ũ − 1) π (2.6.44b)
ϕ̃ = (2ũ − 1) π (2.6.44c)

Relating r̃3 to the polar angle as r̃3 = cos θ, the remaining coordinates are

r̃1 = sin θ cos φ̃ (2.6.45a)

r̃2 = sin θ sin φ̃ (2.6.45b)

The random variable R̃ can now be constructed. Since uniform scattering

is completely symmetric on the unit sphere, the pointing direction r̃ does
not need to be oriented toward ŝo . Figure 2.8(a) illustrates an output state
scattered on the Poincaré sphere.
Biased scattering is used to enhance the likelihood of rare events, rare
events being those where multiple birefringent axes are preferentially aligned,
misaligned, or some other construction. To preferentially align the birefringent
sections, r̂ should be biased to point toward ŝo . A simple way to generate the
bias is first to bias r̃ towards the s3 direction using the following formula:

r̃3 = 2ũ1/a − 1 (2.6.46)

For a ≤ 1 the bias is toward +s3 and for a ≥ 1 the bias is toward −s3 . The
scattering operator R is now biased toward s3 and is denoted R3 . Before R3
can be applied to ŝo the former needs to be rotated into the latter. Following
the previous example of the shortest distance between two points in Stokes
space, a deterministic operator R3−so is constructed to perform the required
rotation. Operator R3−so needs to be calculated only once. Figure 2.8(b) illus-
trates an output state scattered on the Poincaré sphere and biased toward ŝo .
76 2 The Spin-Vector Calculus of Polarization

Table 2.2. Jones and Stokes Equivalent Expressions

Jones expressions Stokes expressions

 
  s1
sx ejφx  
 
|s
=   ŝ =  s2 
sy ejφy  
s3
|t
= T |s
t̂ = R ŝ
|t
= T2 T1 |s
t̂ = R2 R1 ŝ
† †
TT = T T = I RR† = R† R = I
|s
= ŝ · σ |s
ŝ = s | σ | s

|s
s| = 1
2
(I + ŝ · σ ) ŝ = 1
2
Tr (|s
s| σ )
†
U σ U R σ
U = I cos(ϕ/2) − j(r̂ · σ ) sin(ϕ/2) R = (r̂r̂·) + sin ϕ(r̂×) − cos ϕ(r̂×)(r̂×)
 
a b  
U =  Rj,k = 12 Tr U σk U † σj
−b∗ a∗
 
  1
e jα  
 
U1 =   R1 =  cos 2αsin 2α 
−jα
e  
− sin 2α cos 2α

 
  cos 2κ sin 2κ
cos κ −j sin κ  
 
U2 =   R2 =  1 
−j sin κ cos κ  
− sin 2κ cos 2κ

 
  cos 2κ − sin 2κ
cos κ − sin κ  
 
U3 =   R3 =  sin 2κ cos 2κ 
sin κ cos κ  
1

d |s
dŝ
= −j/2(r̂ · σ ) |s
= r̂ × ŝ
dϕ dϕ
 2.6 Equivalent Unitary Transformations 77

Table 2.3. Spin-Vector Expressions

       
1 0 1 0 0 1 0 −j
σ0 =   ; σ1 =   ; σ2 =   ; σ3 =  
0 1 0 −1 1 0 j 0
 
σ1
 
 
σ =  σ2 
 
σ3
 
a0 + a1 a2 − ja3
a0 I + a · σ =  
a2 + ja3 a0 − a1

R = (r̂r̂·) + sin ϕ(r̂×) + cos ϕ(I − (r̂r̂·))

   
r1 r1 r1 r2 r1 r3 0 −r3 r2
   
   
r̂r̂· =  r2 r1 r2 r2 r2 r3 , r̂× =  r3 0 −r1 
   
r3 r1 r3 r2 r3 r3 −r2 r1 0

 
α · σ /2) = eα0 /2
H = exp (α0 /2) exp ( I cosh (α/2) + (α̂ · σ ) sinh (α/2)

   
· σ /2 = e−j β0 /2 I cos (β/2) − j(β̂ · σ ) sin (β/2)
T = exp (−j β0 /2) exp −j β
78 2 The Spin-Vector Calculus of Polarization

References
1. O. Aso, I. Ohshima, and H. Ogoshi, “Unitary-conserving construction of the Jones
matrix and its applications to polarization-mode dispersion analysis,” Journal of
the Optical Society of America A, vol. 14, no. 8, pp. 1988–2005, Aug. 1997.
2. D. M. Brink and G. R. Satchler, Angular Momentum, 3rd ed. Oxford: Oxford
Science Publications, 1999.
3. N. Frigo, “A generalized geometric representation of coupled mode theory,” IEEE
Journal of Quantum Electronics, vol. QE-22, no. 11, pp. 2131–2140, 1986.
4. N. Gisin and B. Huttner, “Combined effects of polarization mode dispersion and
polarization dependent losses in optical fibers,” Optics Communications, vol. 142,
pp. 119–125, Oct. 1997.
5. J. P. Gordon and H. Kogelnik, “PMD fundamentals: Polarization mode dispersion
in optical fibers,” Proceedings of National Academy of Sciences, vol. 97, no. 9,
pp. 4541–4550, Apr. 2000. [Online]. Available: http://www.pnas.org
6. M. Rose, Elementary Theory of Angular Momentum. New York: Dover Publi-
cations, 1995.
7. J. J. Sakurai, Modern Quantum Mechanics. New York: Addison–Wesley, 1985.
8. G. Strang, Linear Algebra and its Applications, 3rd ed. New York: Harcourt
Brace Jovanovich College Publishers, 1988.
http://www.springer.com/978-0-387-22493-0