Chapter 2

Matrix description of wave propagation

and polarization

Contents
2.1 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2–1
2.2 Matrix description of wave propagation in linear systems . . . . . . . . . . . . . . 2–4
2.3 Matrix description of light polarization . . . . . . . . . . . . . . . . . . . . . . . . . 2–12

2.1 Electromagnetic waves

An electromagnetic wave propagating in a dielectric material with permittivity ε and permeability

µ is described by Maxwell’s equations in the time domain, relating the electric field E(r,t) and
magnetic field H(r,t) in the absence of sources:
(
∇ × E(r, t) = −µ ∂H(r,t)
∂t (2.1)
∇ × H(r, t) = ε ∂E(r,t)
∂t

In the frequency domain, the field components can be represented using the phasor notation

F̃(r)ejωt + F̃∗ (r)e−jωt

F(r, t) = Re(F̃(r)ejωt ) = (2.2)
2

in which F is a field vector (either E or H) and F̃(r) is a complex valued vector, representing the
ω
amplitude and phase of the sinusoidal time variation of F (at a frequency f = 2π ).
Using the phasor notation, we can rewrite Maxwell’s equations 2.1 to get a set of differential
equations relating the complex phasors representing the real field vectors.


∇ × Ẽ(r) = −jωµH̃(r)
(2.3)
∇ × H̃(r) = jωεẼ(r)

2–1
If an electromagnetic wave propagates in a uniform medium (ε and µ constants) equation 2.3 can
be rewritten as (by applying the operator ∇ × · to these equations and using ∇ · εẼ = 0 and
∇ · µH̃ = 0)

∇2 Ẽ + ω 2 εµẼ = 0

(2.4)
∇2 H̃ + ω 2 εµH̃ = 0

called the Helmholtz equations. A general solution to these equations can be written as

F̃(r) = F̃+ e−jk·r + F̃− ejk·r (2.5)

in which F̃(r) can be either the electric field phasor or the magnetic field phasor. The wave vector
k can have an arbitrary orientation and is related to the frequency as

√
k = ω εµ1k (2.6)

and defines the wavelength (in the material) λ as

2π
|k| = (2.7)
λ

This is the well known plane wave solution. F̃+ represents a plane wave propagating in the
direction 1k , while F̃− represents a plane wave propagating in the opposite direction. This can be
seen by substituting the forward propagating part of equation 2.5 into equation 2.2 and assuming
propagation along the z-axis (k = |k| 1z ). The time variation of the field can then be written as
 
F (z, t) = F̃+  cos(ωt − kz + arg(F̃+ )) (2.8)
 

This means that planes of constant phase φ move along the positive z-axis with a phase velocity

ω
vph = (2.9)
k

In what follows, we will omit the ˜ and implicitly assume that we are dealing with phasors.
Inserting equation 2.5 in equation 2.3 we find that

−jk × E+ e−jk·r + jk × E− e+jk·r = −jωµ(H+ e−jk·r + H− e+jk·r )


(2.10)
−jk × H+ e−jk·r + jk × H− e+jk·r = jωε(E+ e−jk·r + E− e+jk·r )

Identifying corresponding terms results in a relation between electric and magnetic field


k × E+ = ωµH+
 × E− = −ωµH−
k
(2.11)
k × H+ = −ωεE+
k × H− = ωεE−

2–2
 Figure 2.1: Orientation of field vectors for the plane wave solution

This means that electric field vector, magnetic field vector and wave vector are all orthogonal as
shown in figure 2.1.
If we now consider the case of an isotropic z-invariant dielectric waveguide described by per-
mittivity ε(x, y) and permeability µ(x, y), we can suggest a forward propagating solution of the
Maxwell equations of the form

E(x, y, z) = [eT (x, y) + ez (x, y)1z ] e−jβz


(2.12)
H(x, y, z) = [hT (x, y) + hz (x, y)1z ] e−jβz

in which eT (x, y) and hT (x, y) are vectors in the xy plane. Substituting equation 2.12 into equation
2.3 we can write

βeT − j∇T ez = −ωµ1z × hT

βhT − j∇T hz = ωε1z × eT
(2.13)
∇T · (1z × eT ) = jωµhz
∇T · (1z × hT ) = −jωεez

∂ ∂
in which ∇T = 1x ∂x + 1y ∂y and β is still unknown. This set of equations forms an eigenvalue
equation, for which there is only a solution for certain values of β. The corresponding field profiles
{eT , ez 1z , hT , hz 1z } are called eigenmodes or optical modes of the waveguide. It is easy to see that if
{eT , ez 1z , hT , hz 1z } e−jβz is a solution of equation 2.13 then {eT , −ez 1z , −hT , hz 1z } ejβz is also a
solution.
If β is assumed to be purely real (the mode is propagating without loss) and the waveguide is
lossless (ε is real), then eT and hT are purely real and ez and hz are purely imaginary.
One can show that all waveguide modes are orthogonal or
Z Z
1
ei × hi0 · 1z dS = Cδii0 (2.14)
2

The fields ei and hi are defined as


ei = eT,i + ez,i 1z
(2.15)
hi = hT,i + hz,i 1z

2–3
 Figure 2.2: Schematics of an (optical) circuit

and correspond with an eigenvalue βi .

In the case of a lossless waveguide equation 2.14 can be rewritten, due to the above mentioned
properties of the fields, as
Z Z
1
ei × h∗i0 · 1z dS =Cδii0 (2.16)
2

2.2 Matrix description of wave propagation in linear systems

2.2.1 Introduction

In this section we will gain insight into the global transmission properties of electromagnetic cir-
cuits. The results are applicable to microwave circuits as well as to optical circuits. We will look at
the circuit as a black box, which exchanges energy with the outside through several physical out-
lets that can be optical waveguides or mere free space electromagnetic beams as shown in figure
2.2.
In figure 2.2 ai and bi represent the complex amplitude of the ingoing and outgoing normal-
ized electromagnetic mode (carrying unit power) using the phasor notation. This means that
the transversal electric and magnetic field at position z=0 (with position z=0 chosen to be at the
intersection of the outlet and the surface S of the black box) of port i can be written as

ET,i (x, y, t) = Re (ai + bi )eT,i (x, y)ejωt 

 
(2.17)
HT,i (x, y, t) = Re (ai − bi )hT,i (x, y)ejωt

2–4
with eT,i (x, y) and hT,i (x, y) being the electric and magnetic transversal field profile of the normal-
ized electromagnetic mode respectively.
The ports are considered to be lossless waveguides and the electrical field is assumed to be zero
everywhere on S outside the ports. Every outlet has its own propagation axis and it is essential to
assume that electromagnetic modes are well confined around these axes, so that effective outlets
can be defined outside which the fields are negligible.
In figure 2.2, the outlet ports are assumed to carry only one mode. This is not a restriction as dif-
ferent outlet ports may physically coincide to describe multi mode ports (due to the orthogonality
of the waveguide modes).
In the following matrix formalisms no attention is paid to the internal details of the circuit. The
only requirements are that the system has to be passive and linear. A circuit is passive if there are
no outgoing waves generated, when there are no incident waves.

2.2.2 Scattering matrix description

As the ports are considered to be lossless waveguides, the orthonormality relation for the fields
can be written as
Z Z
1
ei × h∗i0 · 1z dS =δii0 (2.18)
2

integrated over the surface S and the unit vector 1z directed inwards. The power absorbed in the
circuit is obtained by adding the net powers entering the various ports

X
P = (|ai |2 − |bi |2 ) = A† A − B† B (2.19)

with A being the column vector of incident field amplitudes and B the column vector with the
amplitudes of the outgoing fields. Here † represents the Hermitian conjugate of the vector A,
being the complex conjugate of the transpose.
The output wave vector B is completely determined by A otherwise there could be two output
vectors B and B’ for the same input vector A and the difference B - B’ would correspond to a zero
input field. This is forbidden by the assumption that the circuit is passive. Since we assumed the
N-port circuit to be linear, the relation between A and B can be described by an N xN scattering
matrix S:

B = SA (2.20)

with A = (a1 , a2 , ..., aN )T and B = (b1 , b2 , ..., bN )T .

If all the terms of A are zero except ai = 1, the output waves are given by the i-th column of S.
The diagonal term Sii is the reflection coefficient of mode i, the non-diagonal terms Ski are the
transmission coefficients from the mode i towards the mode k.

2–5
 Figure 2.3: Moving the port planes

General properties of S matrices

• Moving the port planes

The ports are changed when the surface S enclosing the circuit is changed. Let lµ be the
displacement of port µ counted positively if the port is moved towards the inside. The new
circuit is then described by the scattering matrix S’ related to the original scattering matrix
S by

S0 = XSX (2.21)

where X is the diagonal matrix formed with the elements exp(jkµ lµ ) with kµ the propagation
constant of mode µ.

• Lossless circuits
Since lossless circuits absorb no power, we must have P = 0 in equation 2.19, hence

A† (1 − S† S)A = 0 (2.22)

for all possible A. This implies that the scattering matrix of a lossless circuit satisfies

2–6
 Figure 2.4: A change of waveguide width

S†S = 1 (2.23)

A matrix satisfying equation 2.23 is called unitary.

In the case of lossy circuits we must have P > 0 which leads to

A† S† SA ≤ A† A (2.24)

Inserting aµ = δkµ (µ = 1 · · · N ) into equation 2.24 one obtains

N
X
|Sjk |2 ≤ 1 (2.25)
j=1

for k = 1 · · · N .

• Reciprocal circuits
If a circuit is made out of materials with symmetrical constitutive parameters (εT = ε and
µT = µ) the circuit is called reciprocal. This is the case for almost all materials, except for
magnetic materials in the presence of a magnetic field. From Maxwell’s equations one can
deduce that the scattering matrix of a reciprocal circuit is symmetrical. This means that the
transmission between port j and k does not depend on the propagation sense. Note that
reciprocal circuits do not have to be symmetric themselves! This conclusion can have impor-
tant consequences. A typical example is illustrated in Figure 2.4. Naively one might think
that the transmission from the smaller left port to the larger right port will always be larger
than the transmission from right to left. However, if both ports are monomodal waveguides,
and taking into account that the scattering matrix for reciprocal circuits should be symmet-
ric, we know that S12 = S21 and therefore the transmission from port 1 to port 2 should
be equal to the transmission from port 2 to port 1 ! (hereby we assume that all radiation is
absorbed within the closed volume)
Note that for symmetric circuits the diagonal elements of the scattering matrix should all be
equal.

2–7
 Figure 2.5: Beamsplitter in a 4 port circuit representation

Elementary applications in optics

• Beam-splitter
Consider the so called directional coupler shown in figure 2.5a. This is a 4-port circuit de-
scribed by a 4x4 scattering matrix. The operation of this device will be studied in detail in
the chapter on dielectric waveguides but here it is sufficient to know that if two dielectric
waveguides are brought closely together, light can couple between both waveguides. In ad-
dition we know that there is no transmission between port 1 and 2 and between port 3 and
4 and there are no reflections. If the system is reciprocal and lossless, the scattering matrix
can therefore be written as
 
0 0 a b
 0 0 c d 
S=
 a
 (2.26)
c 0 0 
b d 0 0

because of reciprocity (symmetry of the scattering matrix) and the properties of the direc-
tional coupler as described above. Because the system is lossless, S has to be unitary, so


 |a|2 + |b|2 = 1
|c|2 + |d|2 = 1




 2
|b| + |d|2 = 1

(2.27)

 |a|2 + |c|2 = 1
ac∗ + bd∗ = 0




 ba∗ + dc∗ = 0


hence |a| = |d| and |b| = |c|. It is not difficult to prove that the last equation of 2.27 is
redundant: from all the other equations of 2.27 one can derive the last one and therefore it

2–8
 Figure 2.6: Optical power divider

can be left out. With a proper choice of the position of the ports one can use symmetry to
simplify the problem. Consider a symmetry plane that images ports 1 and 2 onto 3 and 4
respectively and a symmetry point in the center of the device, which images ports 1 and 2
onto 4 and 3 respectively. In this case we get:

S13 = S24 = a = d
(2.28)
S14 = S32 = b = c

By positioning the port planes appropriately we can also choose a and d to be real. From
equation 2.27 it is clear that b and c are then purely imaginary. This means that a and c (and b
and d) have a quadrature phase relationship. From this we can conclude that splitting a light
beam in a lossless way always means that the resulting light beams have a quadrature phase
relationship. Exactly the same reasoning can be followed to derive the scattering matrix of
the beamsplitter of figure 2.5b.

• Optical power divider

The simplest power divider, as shown in figure 2.6 is a 3 port circuit that divides the power
entering port 1 equally into ports 2 and 3 with no reflection back to port 1. If the device is
made out of dielectric optical waveguides and the circuit is slowly varying along the z-axis
we can assume that there is no reflection at all when light comes from the right and no light
is coupled from port 2 to 3 and vice versa. From the fact that the circuit is reciprocal and
symmetrical in port 2 and 3, we can write down its scattering matrix.
 √ √ 
2 2
0
√ 2 2
S= 2 (2.29)
0 0 
 
√2
2
2 0 0

This matrix is not unitary: ideal optical power dividers are lossy circuits even if they are
made with lossless dielectric waveguides. The whole power can even be lost if port 2 and
3 are excited with waves having opposite phase. The reason is that the power is converted
into radiating modes, which are not included in this scattering matrix description.

• Plane waves

2–9
 Figure 2.7: Optical power divider

Above, the ”ports” of the system were eigenmodes of optical waveguides. It is possible to
repeat the same reasoning taking into account a set of plane waves as the ports of a given sys-
tem. Although these are not finite in extent, they are orthogonal and all properties derived
above for the scattering matrix remain valid. An example is given by the beamsplitter of
figure 2.7. The scattering matrix for this beamsplitter is identical to the one of the directional
coupler of figure 2.5a and given by equation 2.26.

2.2.3 Transfer matrix description

The transfer matrix of an electromagnetic circuit is a matrix that describes the relation between the
output quantities and the input quantities. This representation of a circuit becomes very useful
when cascading different circuits, as the transfer matrix of the entire system can easily be found
by multiplication of the transfer matrices of the individual circuits of the cascade. The ports of
the circuit are hereby divided into ports considered as input ports of the circuit (schematically
represented at the left side of the black box) and output ports (schematically represented at the
right side of the black box).
Let us again consider the N-port circuit shown in 2.2, and assume that N is even. We can choose
N/2 ports to be the left ports and N/2 ports to be the right ports as shown in figure 2.8. Incident
modes on the left [right] ports are represented by the N/2 × 1 column matrix al [ar ] and outgoing
modes are represented by bl (br ). As the system is linear we can relate the incident and reflected
waves at the input ports to the incident and reflected waves at the output port by a transfer matrix
T. Using block matrix representation we can write

    
bl T11 T12 ar
= (2.30)
al T21 T22 br

If we now consider a cascade of p N-port networks, the transfer matrix of this cascade is the
product of the transfer matrices of the individual N-ports. This can be easily understood for the
case of the cascade of two N-ports (and generalized to p N-ports) as shown in 2.9. Here the N/2

2–10
 Figure 2.8: N port circuit

Figure 2.9: Schematic representation of the cascade of two N ports

left and N/2 right ports are schematically represented by two single ports described by the column
vectors al , bl ,ar and br .
For this structure we can write

         
bI,l aI,r bII,l aII,r aII,r
= TI = TI = TI TII = Tcasc (2.31)
aI,l bI,r aII,l bII,r bII,r

Figure 2.10: Two port circuit

2–11
Relation to the scattering matrix description

As both the scattering matrix description and the transfer matrix description relate the same pa-
rameters using a different set of input and output parameters, both descriptions are strongly re-
lated. To find the matrices Tij from equation 2.30, we start from the scattering matrix description,
relating output to input:     
bl S11 S12 al
= (2.32)
br S21 S22 ar

Or, equivalently,

bl = S11 al + S12 ar (2.33)

br = S21 al + S22 ar (2.34)

If S21 is invertible, one can rewrite 2.34 to:

al = (S21 )−1 [br − S22 ar ] (2.35)

Substituting 2.35 into 2.33 and comparing to the matrix equation 2.30 yields the relationship be-
tween the the T and S matrices.
If we limit ourself to the case of a two port as shown in 2.10, the scattering matrix and transfer
matrix are related by

   
t11 t12 1 −(S11 S22 − S12 S21 ) S11
= (2.36)
t21 t22 S21 −S22 1

This relation is useful because for some circuits it is easy to derive the scattering matrix, but not
its transmission matrix. Equation 2.36 relates both representations.
If the circuit is reciprocal, then det(T)=1.

2.3 Matrix description of light polarization

2.3.1 Polarized waves

Polarization refers to the time dependence of the electric field vector E(r,t) at a point r in space.
The orientation of the electric wavevector changes in time at a certain point r. Once we know
the polarization state of this electric field vector E, the polarization of the three remaining field
vectors D, H and B can be found through Maxwell’s equations. Therefore, in the following we
will concentrate on the electric field vector E.
Assume a monochromatic wave at position r. The electric field vector is described by

E(r, t) = Ex (t)1x + Ey (t)1y + Ez (t)1z (2.37)

Each component (i=x,y,z) can be written as

2–12
Figure 2.11: Elliptical polarization state: (a) general and (b) with a appropriately chosen origin of the time
axis

Ei (t) = Ai cos(ωt + φi ) = (Ai cos(φi )) cos(ωt) − (Ai sin(φi )) sin(ωt) (2.38)

using equation 2.38, equation 2.37 becomes

X X
E(r, t) = ( Ai cos(φi )1i ) cos(ωt) − ( Ai sin(φi )1i ) sin(ωt) = v1 cos(ωt) + v2 sin(ωt) (2.39)
i i

From this equation it is clear that, for a strictly monochromatic wave, the most general polarization
state is the elliptical polarization state for which the end-point of the electric field vector describes
an ellipse in space. Such an ellipse is periodically described at a repetition rate equal to the optical
ω
frequency f = 2π . This is visualized in figure 2.11a. By appropriately choosing the origin of the
time axis t, the vectors v1 and v2 can be made to coincide with the main axes of the ellipse, as
shown in figure 2.11b, where the new vectors are called v01 and v02 . These new vectors are now
orthogonal to each other. If the x-axis and y-axis are also chosen along those directions (with new
axes x0 and y 0 ), the electric field can finally be written as

E(r, t) = v01 cos(ωt0 ) + v02 sin(ωt0 ) = Ex0 cos(ωt0 ) + Ey0 sin(ωt0 ) (2.40)

For a complete representation of the elliptical polarization we need to know

• the orientation in space of the plane of the ellipse of polarization

• shape and size of the ellipse

• sense of traveling the ellipse and absolute temporal phase (i.e. the orientation of the electric
field vector at t=0)

The shape of the ellipse is determined by

• the azimuth θ, being the angle between the major axis of the ellipse and the positive direction
of the X axis with − π2 ≤ θ < π2 .

• its ellipticity e, being the ratio of the length of the minor axis to the major axis

b
e= (2.41)
a

2–13
 Figure 2.12: Azimuth and ellipticity angle

Figure 2.13: Linear and circular polarization

• its handedness, determining the sense in which the ellipse is described. The polarization is
defined to be right handed (left handed) if the ellipse is traversed in a clockwise (counter-
clockwise) sense seen when looking against the direction of propagation (into the beam).

It is convenient to incorporate the handedness in the definition of the ellipticity by allowing the
ellipticity to assume positive (right-handed) and negative (left-handed) values. Therefore,

−1≤e≤1 (2.42)

Together with the ellipticity an ellipticity angle  is introduced (− π4 ≤  ≤ π4 )

tan() = e (2.43)

as shown in figure 2.12.

The circular and linear polarization state are special cases of the more general state of elliptical
polarization as shown in figure 2.13, where we assume that the beam propagates in the positive z
direction. For circular polarization e = ±1. For linear polarization e = 0. Circular polarizations
have undetermined azimuth and linear polarizations have no handedness.

2–14
2.3.2 Representation of polarization

Jones vector of a plane monochromatic wave

The electric field component of a plane monochromatic wave, propagating in the positive z direc-
tion of a right handed Cartesian coordinate system xyz, is described by

2π 2π
E(z, t) = Ex cos(ωt − z + φx )1x + Ey cos(ωt − z + φy )1y (2.44)
λ λ
Suppressing the fixed unit vectors 1x and 1y , and the temporal information of the signal (as the
frequency is known), choosing the reference plane at z=0 and using phasor notation, the plane
wave can be represented by the Jones vector

Ex ejφx
 
E= (2.45)
Ey ejφy

The Jones vector is a complex vector that is a mathematical representation of a real wave. The
intensity of the optical wave can be written as

I = |Ex |2 + |Ey |2 = E† E (2.46)

A wave of unit intensity is said to be normalized and its Jones vector is said to be normal. Such a
vector satisfies the condition

E† E = 1 (2.47)

A normalized wave with Jones vector

 
1
E= (2.48)
0

represents a linearly polarized wave along the x axis and vice versa for the linear polarization
along the y axis. Left- and right-handed circularly polarized waves are represented by

 
1 1
El = √ (2.49)
2 −j
 
1 1
Er = √ (2.50)
2 j

In general, the Jones representation of a normalized elliptically polarized beam with azimuth θ
and ellipticity angle  is given by

 
jδ cos(θ) cos() − j sin(θ) sin()
e (2.51)
sin(θ) cos() + j cos(θ) sin()

2–15
with ejδ an arbitrary phase factor.
Two waves E1 and E2 are said to be orthogonal if

E†1 E2 = E†2 E1 = 0 (2.52)

As every polarization state can be described as the superposition of two orthogonal (and normal-
ized) polarization states, the same is true for the Jones vector. Every Jones vector can be described
as the superposition of two orthonormal Jones vectors.

E = A1 E 1 + A2 E 2 (2.53)

with E†1 E2 = E†2 E1 = 0 and E†1 E1 = E†2 E2 = 1.

This means that E can be written as the superposition of any two linearly polarized waves with
orthogonal polarization, as the superposition of two circularly polarized waves (one right-handed
and one left-handed) and of any two orthogonal elliptical polarizations. Changing from one set
of basis Jones vectors to another set implies a linear transformation to get the new representation
of the same polarization state in the new set of basis vectors Eold = FEnew . The columns of F
represent the new basis vectors in the old coordinate system. To change from the cartesian basis
vectors to the circular basis vectors, we can write

    
Ex 1 1 1 El
=√ (2.54)
Ey 2 −j j Er

Poincaré sphere

As the polarization ellipse of a monochromatic plane wave propagating in a known direction is

completely determined by the azimuth θ and ellipticity angle , it is useful to represent the po-
larization state as a point on a sphere (with a radius equal to 1) as shown in figure 2.14. This
point is completely characterized by its longitude 2θ and its latitude 2. As the equator repre-
sents polarization states with  = 0 these states are all linear polarizations. Above (below) the
equator all polarization states are right-handed (left-handed). The north and south pole of the
sphere represent the right-handed and left-handed circular polarization. Diametrically opposite
points represent pairs of orthogonal polarization. This is easily verified using the Jones vector
representation 2.51 and the definition of orthogonality 2.52.

2.3.3 Quasi-monochromatic waves

If we now consider a quasi-monochromatic wave, both field strength and phase become slowly
varying functions in time

2π 2π
E(z, t) = Ex (t) cos(ωt − z + φx (t))x + Ey (t) cos(ωt − z + φy (t))y (2.55)
λ λ

2–16
 Figure 2.14: Poincaré sphere

The polarization state will still be an elliptical polarization but the shape and orientation of the
ellipse will vary in time. This means that the Jones vector becomes a function of time and also the
position on the Poincaré sphere will change with time.
If there is no correlation between the two components of the Jones vector, the wave is called un-
polarized and the polarization state will describe a random walk on the Poincaré sphere. Partially
polarized light describes a random walk around a particular point on the sphere. For completely
polarized light, the point on the sphere is fixed in time. This requires

Ex (t)
= constant (2.56)
Ey (t)

and

φx (t) − φy (t) = constant (2.57)

Quasi-monochronatic waves are mathematically treated using the Stokes parameters (S0 ,S1 ,S2 ,S3 )
defined as

S0 =< Ex2 (t) > + < Ey2 (t) >



S1 =< Ex2 (t) > − < Ey2 (t) >


(2.58)
S = 2 < Ex (t)Ey (t) cos(φx (t) − φy (t)) >
 2


S3 = 2 < Ex (t)Ey (t) sin(φx (t) − φy (t)) >

in which < v > represents the time average of v.

2–17
S0 represents the total intensity of the optical wave. Parameter S1 is large (in absolute value)
if the quasi-monochromatic wave has a preference for linear polarization along x or y, while S2
represents the tendency towards linear polarization along ±45◦ (Ex and Ey have equal amplitudes
and the phase difference between x and y direction is 0 or π ). S3 represents the preference for
circular polarization (Ex and Ey have equal amplitudes and there is a phase difference π2 or 3π
2 ).

This definition implies that unpolarized light is represented by the Stokes parameters (S0 ,0,0,0) be-
cause there is no preferential direction of polarization. Completely polarized light (still defined by
its ellipticity angle  and azimuth θ) is represented by (S0 ,S0 cos(2)cos(2θ),S0 cos(2)sin(2θ),S0 sin(2)).
This can be understood by writing down the Stokes parameters as a function of a time varying
Jones vector (using equation 2.51)

Ex (t)ejφx (t)
     
Ec,x jδ(t) cos(θ) cos() − j sin(θ) sin()
E(t) = = = A(t)e (2.59)
Ec,y Ey (t)ejφy (t) sin(θ) cos() + j cos(θ) sin()

as

S0 =< |Ec,x |2 > + < |Ec,y |2 >



S1 =< |Ec,x |2 > − < |Ec,y |2 >


(2.60)

 S2 = 2 < Re(Ec,x ∗ Ec,y ) >
S3 = 2 < Re(−j Ec,x ∗ Ec,y ) >


This representation implies that the Stokes parameters of a completely polarized beam with inten-
sity I = 1 is represented by the parameters (1, S1 , S2 , S3 ) where (S1 , S2 , S3 ) represent the cartesian
coordinates of the point P on the Poincaré sphere using the coordinate system as shown in figure
2.14.

P (x, y, z) = (cos(2) cos(2θ), cos(2) sin(2θ), sin(2)) = (S1 , S2 , S3 ) (2.61)

For completely polarized light the Stokes parameters are related by

S02 = S12 + S22 + S32 (2.62)

For partially or unpolarized light the inequality

S02 ≥ S12 + S22 + S32 (2.63)

is valid. These equations suggests that the general case of partially polarized light can be treated
by splitting that wave into two components, a totally polarized component and an unpolarized
component

S = Sun + Stp (2.64)

where

2–18
 S = (S0 , S1 , Sp
2 , S3 )
Sun = ([S
p0 − (S12 + S22 + S32 )], 0, 0, 0) (2.65)
Stp = ( (S12 + S22 + S32 ), S1 , S2 , S3 )

Note that each Stokes parameter of the original beam is obtained by adding the corresponding
parameters of the component beams.
From the equations above we can define the degree of polarization ℘ as being the ratio of the
intensity of the totally polarized component to the total intensity of the wave

p
(S12 + S22 + S32 )
℘= (2.66)
S0

In terms of the total intensity I, the degree of polarization ℘, the azimuth θ and ellipticity angle
 of the ellipse of polarization of the totally polarized component, the Stokes parameters of any
quasi-monochromatic wave can be cast in the form

S = I(1, ℘ cos(2) cos(2θ), ℘ cos(2) sin(2θ), ℘ sin(2)) (2.67)

A normalized partially polarized wave (I = 1) is therefore represented by a point inside the

Poincaré sphere with spherical coordinates (℘,2θ,2).
The interpretation of the Stokes parameters suggests a simple experiment by which they may be
measured for a given wave. Let I0 denote the total intensity of the wave and let Ix , Iy , I π4 ,I −π ,
4
Il and Ir represent the intensities transmitted by an ideal variable polarizer placed in the path of
the wave and adjusted to transmit the x,y, π4 , −π
4 linear polarizations and the left- and right-circular
polarizations, respectively. In terms of these intensities the Stokes parameters are given by


 S0 = (Ix + Iy ) = (Il + Ir ) = (I π4 + I− π4 )

S1 = Ix − Iy

(2.68)

 S 2 = I π4 − I− π4

S3 = Ir − Il

2.3.4 Propagation of polarized light through polarizing optical systems

The Jones-matrix formalism

Consider a uniform monochromatic plane wave that is incident onto a non-depolarizing optical
system. A non-depolarizing system is defined as an optical system in which the degree of po-
larization of the output beam ℘o is larger than or equal to the degree of polarization of the input
beam ℘i . In the case where the input beam is completely polarized, this means that the output
beam is also completely polarized.
As a result of the interaction between the incident wave and the optical system the polarization
of the plane wave may change. The optical system including the incident optical wave and one of
the outgoing waves is schematically represented in figure 2.15.

2–19
 Figure 2.15: Schematic representation of a polarized beam incident to an optical system

Two cartesian coordinate systems (xyz) and (uvw) are associated with the incident and outgoing
plane wave and the z and w directions are chosen parallel to the respective wavevectors k and l.
The choice of the origin of both coordinate systems is arbitrary.
Using these coordinate systems we can represent the polarization state of the input beam by its
Jones vector

 
Eix
Ei = (2.69)
Eiy

and the polarization of the output beam

 
Eou
Eo = (2.70)
Eov

In the absence of non-linear processes we can relate the components of the Jones vector of the
incident wave to that of the outgoing wave using the Jones matrix

    
Eou T11 T12 Eix
= (2.71)
Eov T21 T22 Eiy

Equation 2.71 expresses the law of interaction between the incident wave and the optical system as
a simple linear matrix transformation of the Jones vector representation of the wave. The elements
Tij of the Jones matrix are in general complex. If optical systems are cascaded then the interaction
with an incident beam and this combined system is represented by the matrix multipication of the
individual Jones matrices.

• Quarter wave plate

Let us consider an optical system that transforms a linear polarization state (azimuth angle
θ = π4 ) to a right handed circular wave. The system also has to satisfy the requirement that
linear polarization states along x and y direction remain unchanged. The Jones matrix then
has to satisfy

   " √1
#
1 1 A B
√ e−jφ = 2 (2.72)
2 j C D √1
2

where B and C have to be zero because of the requirement for x and y polarization invariance.
It’s easy to see that the Jones vector representation of both x and y linear polarization are
eigenvectors of the Jones matrix describing the optical system.

2–20
 The Jones matrix of the optical system then has to be of the form

e−jφ
   
A B 0
= −j(φ− π2 ) (2.73)
C D 0 e

This means that the optical system has to retard the x polarization state and y polarization
state differently (and up to a phase shift of π2 ). This can be achieved using a plate of uniaxially
anisotropic material in which x and y polarization experience a different refractive index (nx
and ny respectively). The Jones matrix describing the polarization change through a plate of
length z can be written as

e−jk0 nx z
 
0
−jk (2.74)
0 e 0 ny z

in which k0 is the free space wave vector. If the thickness L of the plate satisfies

π
k0 (nx − ny )L = (2.75)
2
equation 2.74 and 2.73 are equivalent and all conditions are satisfied. A uniaxially anisotropic
plate satisfying equation 2.75 is called a quarter wave plate. Notice that a quarter wave plate
is not a quarter wavelength thick.

• Optical activity
As a second example we will consider an optical system that rotates an arbitrary linear polar-
ization state over an angle θ. If we consider x and y polarization states we can immediately
write down the Jones matrix of the optical system as
 
cos(θ) − sin(θ)
(2.76)
sin(θ) cos(θ)

Due to the superposition principle this optical system will also rotate an arbitrary linear
polarization state over an angle θ.
This Jones matrix can be realized using a plate of optically active material. This is a crys-
talline material with a helical molecular structure, so that a left-handed circular wave and a
right-handed circular wave injected along the axis of the helix experiences a different phase
velocity ( kωl and kωr ) dependent on the rotation sense of the molecular structure. The Jones
matrix of a plate with thickness z must satisfy
    
−jkr z 1 A B 1
e = (2.77)
j C D j

and
    
1 A B 1
e−jkl z = (2.78)
−j C D −j

If we introduce the average wave number k and the specific rotary power ρ as

2–21
  kr +kl
k= 2
kr −kl (2.79)
ρ= 2

then we can solve equation 2.77 and 2.78 to find the Jones matrix
 
−jkz cos(ρz) − sin(ρz)
e (2.80)
sin(ρz) cos(ρz)

Comparison with equation 2.76 indicates that the polarization state in a plate of optically
active material, excited by a linear polarization state, is continuously rotating the linear po-
larization, determined by the specific rotary power.

The Muller-matrix formalism

In the previous section we dealt with the problem of describing the propagation of totally polar-
ized light through linear non-depolarizing systems. In the present section we will deal with the
case of propagation of unpolarized or partially polarized quasi-monochromatic light propagated
through a non-depolarizing optical system. As partially polarized light is represented by its Stokes
vector


S0
 S1 
S= 
 S2  (2.81)
S3

We will find a relationship, the 4x4 Muller matrix M, between the Stokes vector of the output beam
So and the Stokes vector of the input beam Si .

So = MSi (2.82)

As the considered system is non-depolarizing, the Jones matrix formalism can still be used, relat-
ing both the (time-dependent) Jones vectors of input and output beams

Eo (t) = TEi (t) (2.83)

Making the direct product of both sides of equation 2.83 with E∗o and time averaging this product
we get

< Eo (t) ⊗ E∗o (t) >=< TEi (t) ⊗ T∗ E∗i (t) >= (T ⊗ T∗ ) < Ei (t) ⊗ E∗i (t) > (2.84)

Here, the direct product of an mxn matrix A and a pxq matrix B is defined as the (mp)x(nq) matrix
C written in block matrix notation as

2–22
  
a11 B a12 B ··· a1n B
 a21 B a22 B ··· a2n B 
C=A⊗B= (2.85)
 
.. .. .. .. 
 . . . . 
am1 B am2 B · · · amn B

Equation 2.84 can then be written as

Jo = (T ⊗ T∗ )Ji (2.86)

where Jk equals (k=o,i)

 ∗ > 
< Ekx Ekx
∗ > 
 < Ekx Eky
Jk =< Ek (t) ⊗ E∗k (t) >= 
 < Eky E ∗ > 
 (2.87)
kx
∗ >
< Eky Eky

This vector Jk is linearly related to the Stokes parameters by equation 2.60

   
S0 1 0 0 1
 S1   1 0 0 −1 
 S2  =  0
    J = AJ (2.88)
1 1 0 
S3 0 j −j 0

From equation 2.88 and 2.84 we can conclude that the effect of a non-depolarizing optical system
on the Stokes parameters, can be described by

So = A(T ⊗ T∗ )A−1 Si
 
(2.89)

in which

M = A(T ⊗ T∗ )A−1 (2.90)

is a 4x4 real matrix, called the Muller matrix. Equation 2.90 relates the Muller and Jones matrix
for a non-depolarizing optical system. It can be shown that in such a case only seven out of the
sixteen elements are independent.
If we now consider a depolarizing optical system, equation 2.83 is no longer valid, as a depolarizing
system can not be represented by a Jones matrix. Therefore, equation 2.90 is no longer valid.
However, one can show that relation 2.82 is still valid for a depolarizing optical system. In this
case all the sixteen real elements of the Muller matrix can be independent.

• Quarter wave plate

As equation 2.74 describes the Jones matrix for propagation through an uniaxial material,
we can calculate the corresponding Muller matrix using equation 2.90 as

2–23
  
1 0 0 0
 0 1 0 0 
M=  (2.91)
 0 0 cos(∆kz) − sin(∆kz) 
0 0 sin(∆kz) cos(∆kz)
π
with ∆k = k0 (nx − ny ). For a quarter wave plate we have ∆kz = 2 and the Muller matrix
becomes
 
1 0 0 0
 0 1 0 0 
M =  (2.92)
 0 0 0 −1 
0 0 1 0

• Optical activity
In a similar way, we can calculate the Muller matrix of propagation through an optically
active material. Using equation 2.80 and 2.90 we get
 
1 0 0 0
 0 cos(2ρz) − sin(2ρz) 0 
M =
 0 sin(2ρz) cos(2ρz)
 (2.93)
0 
0 0 0 1

Graphical representation of polarization change

As completely polarized light can be represented by a point on the Poincaré sphere, we can graph-
ically track the change of polarization through an optical system by drawing the locus of polariza-
tion states throughout the optical system.
For the case of an uniaxially anisotropic material the Muller matrix is given by equation 2.91. From
this equation it is easy to understand that the point on the Poincaré sphere, will, when it propa-
gates through an uniaxially anisotropic material, describe a circle formed by the intersection of the
Poincaré sphere and a plane parallel to the yz-plane (due to the invariance of S1 ) that contains the
initial polarization point P as shown in figure 2.16.
For the case of optical activity the point will describe a circle formed by the intersection of the
Poincaré sphere and a plane parallel to the xy-plane that contains the initial polarization point P
as shown in figure 2.17.

Scattering matrix and transfer matrix description including polarization

As the scattering and transfer matrix description only assumes orthogonal modes in its ports, po-
larization changes through an optical system can also be described by a scattering and transfer
matrix description when using orthogonal polarization states. The resulting scattering and trans-
fer matrix will depend on the choice of orthogonal polarization states.

2–24
Figure 2.16: Representation of polarization change through an uniaxially anisotropic medium

Figure 2.17: Representation of polarization change through an optically active material

Figure 2.18: Operation of a quarter wave plate

2–25
 Figure 2.19: System description of a quarter wave plate

Figure 2.20: System description of an optically active material

• Quarter wave plate

Let us consider a quarter wave plate that operates as shown in figure 2.18. If we choose the
orthogonal polarization states as in figure 2.19, we can readily write the scattering matrix as
 
0 0 1 −j
1  0 0 −j 1 
√   (2.94)
2  1 −j 0 0 
−j 1 0 0

From the symmetry of the scattering matrix we see that a quarter wave plate is a reciprocal
circuit.

• Optical activity
Let us now consider an optical active material that rotates a linear polarization over an angle
θ = ρL, with ρ the specific rotary power and L the thickness of the optically active plate. We
choose the orthogonal polarization states to be circular and define the propagation constant
of a circular wave propagating along (opposite to) the rotation sense of the molecules kp
(km ). For the circuit represented in figure 2.20, the scattering matrix becomes

e−jkp L
 
0 0 0
 0 0 0 e−jkm L 
 e−jkp L
  (2.95)
0 0 0 
0 e−jkm L 0 0

Because the scattering matrix is symmetric, this device is also reciprocal.

2–26