Transfer Matrix

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Introduction to Transfer Matrices

Jonathan Hood
(Dated: July 14, 2015)

I. INTRODUCTION TO TRANSFER MATRICES

A. Introduction

The scattering matrix and transfer matrix can both be used to represent a scattering problem. The transfer matrix
M relates the amplitudes on left-hand side to the amplitudes on the right-hand side, while the scattering matrix S
relates the incoming amplitudes to the outgoing amplitudes. Both pictures have advantages and disadvantages. Since
the transfer matrix relates amplitudes on one side to the other side, concatenating multiple systems is conveniently
accomplished by multiplying their transfer matrices together. However, symmetries and energy conservation take a
much simpler form in the scattering matrix picture. In this section, I use the scattering picture to derive constraints
on the reflection and transmission coefficients, which are then used when we move to the transfer matrix picture to
greatly reduce the degrees of freedom.

B. Scattering Picture

The scattering matrix is defined as:


out S in
z }| { z }| { z + }| {
0

L (x = 0) r t L (x = 0)
= 0 (1)
+
R (x = l)
t r R (x = l)

,+
L,R is the amplitude of the traveling wave, and the subscript L (R) denotes left (right) side, and - (+) denotes
backward (forward) propagating. From this definition, it follows that the variables r and t are the reflection and
transmission coefficients of the left-going wave, and r0 and t0 of the right-going wave. The complex scattering matrix
S has 8 degrees of freedom (DOFs). I will next show that the number can be greatly reduced by using energy
conservation and symmetries.

C. Energy Conservation = Unitarity

First, we use energy conservation to reduce the number of DOFs of the scattering matrix to 4. If the amplitudes
are normalized such that ||2 is proportional to power, then energy conservation requires | 2 + 2 + 2
L | + |R | = |L | +
2
|R | . This can be written in vector notation as
out out = in in (2)
This equation can be used to show that S is unitary. First, take the complex conjugate of both sides of (1) to get
(out ) = (Sin ) out = in S
Using again that out = Sin , we multiply the LHS on the right by out and the RHS on the right by Sin , which
gives out out = in S Sin . We then use the energy conservation equation from (2) to get
in in = in S Sin .
Since this must be true for all matrices, we get that S is unitary for any lossless system:

SS = 1 (3)
A unitary matrix has 4 DOFs, so we have reduced the DOFs from 8 to 4.
2

D. Time-reversal symmetry

Next we will time-reversal symmetry to reduce the DOFs from 4 to 3. Given some input in , the scattering matrix
tells us the output out = Sin . If energy is conserved, then the time-reversed scenario must also be a solution. The
output and input are swapped. The time-reserved output out is the input and the time-reversed input in is the
output:

in = Sout (4)

The complex conjugate is necessary but subtle. The vectors represent the complex amplitudes of a traveling wave
Aeik(xx0 ) at a specific location. The time dependence eit is of course assumed. To get the time-reversed scenario,
we just take t t, but this can also be achieved by taking the complex conjugate of the first term.
Next we take the complex conjugate of the time-reversed equation from eq (4) and solve for out to get

out = (S )1 in (5)

Setting this equation equal to the original out = Sin gives in = SS in , or that

SS = S S = 1 (6)

By looking at the definition of S in eq (1), this demands that the transmission coefficients are the same from both
directions, t = t0 .
When combined with unitarity S S = 1, we get S = S T , or that S is symmetric. So are final conclusion is that if
the system conserves energy, then the scattering matrix S is a symmetric unitary matrix, which has only 3 DOF.

E. Inversion symmetry

One additional symmetry can be used for systems that look the same from left-to-right as from right-to-left.
Inversion symmetry requires that the reflection coefficient be the same from both sides, r = r0 . Inversion symmetry is
true even for lossy case. This symmetry is discussed in more detail in Bendickson, where it is called parity symmetry.
Many systems have inversion symmetry, such as dipole scatterer or a symmetric cavity. However the scattering from
a change of index boundary does not since the indices are different on each side. In that case r and r0 differ by a
phase.

F. The most general lossless scattering matrix

We found that the scattering matrix that represents a lossless system (one that conserves energy and obeys time-
reversal symmetry) must be a unitary symmetric matrix. The most general symmetric unitary matrix has 3 DOF
and can be parametrized by the 3 variables , t , and as
 0   
r t cos()ei(/2+) sin()
S= 0 = exp(it ) (7)
t r sin() cos()ei(/2)

Using this general form for a symmetric unitary matrix, we can convert the symmetric unitary condition for the
scattering matrix S into an equivalent set of requirements for the reflection and transmission coefficients (using the
convention x = arg x) :

|r| = |r0 | arg(r/t) = r t = /2 |r|2 + |t|2 = 1


t = t0 arg(r0 /r) = r0 r = 2 t = |t|eit (8)

If r, r0 , t, and t0 satisfy these requirements, then the scattering matrix S is symmetric and unitary. It is a necessary
and sufficient set of requirements for a system to be lossless. These expressions are an extremely convenient way for
reducing the DOFs of a lossless system. They tell us that the magnitudes of the transmission and reflection coefficient
must add to 1, |r|2 + |t|2 = 1. They tell us that the transmission coefficient from left-to-right and right-to-left are
always the same, t = t0 . They tell us that the magnitude of reflection coefficients r and r0 from each side are the
same, but that the phase can differ by some phase 2. And that the phase 2 also determines the phase relationship
between r and t.
3

To summarize, the lossless system only has three DOFs: t , , and . determines the magnitude of r and t
by tan() = |r/t|. t is just the phase of t. And can be thought of as the phase relationship between r and t. It
can also be thought of as how much the system violates inversion symmetry. If we include inversion symmetry, then
r = r0 , which gives = 0. However for the case of normal reflection from dielectric surface (which has no inversion
symmetry), = /2 if n1 > n2 , and = +/2 if n1 < n2 .

G. Transfer Matrix Picture

After having re-expressed the energy conservation and symmetry requirements in terms of requirements for the
reflection and transmission coefficients r, r0 , t, and t0 in eq (8), we can now switch to the transfer matrix picture and
use these requirements to reduce the number of DOFs.

H. Scattering Picture to Transfer Matrix Picture

The transfer matrix relates the amplitudes on the LHS to the amplitudes on the RHS: R = M L . The scattering
matrix elements corresponded to the reflection and transmission coefficients from the RHS ( r, t) and LHS ( r0 , t0 ).
It is possible to re-express the elements of a transfer matrix in terms of the elements of the scattering matrix, which
gives
R M L
z }| { z  }| z
{  }| {
+
R (x = l)
1 tt0 rr0 r +
L (x = 0)
= (9)

R (x = l) t r0 1
L (x = 0)

Therefore the definition of the transfer matrix in terms of the reflection and transmission coefficients is

1 tt0 rr0 r
 
M= (10)
t r0 1

This equation can be derived by the following method. First, we define U1 and U2 such that

U1 L = in , U2 R = out (11)

U1 and U2 are given by:


   
1 0 1 M2,1 M1,1
U1 = , U2 = (12)
M2,1 M2,2 det M 1 0

Using these two relationships, we can solve for M in terms of S. We start with R = M L and multiply on the LHS
by U2 to get U2 R = (U2 M U11 )U1 L . Then we use eq (11) to convert the LHS: out = (U2 M U11 )in . Finally,
since we already know that out = Sin , we must have

S = U2 M U11 (13)

Solving M in terms of S then gives the result from (9).


Once we have the transfer matrix for a system, it is often convenient to derive the reflection and transmission
coefficients. We get

M1,2 1 M2,1 M1,1 M2,2 M1,2 M2,1


r= , t= , r0 = , t0 = (14)
M2,2 M2,2 M2,2 M2,2
4

One very important and convenient property of the transfer matrix is that its determinant is always 1, even for a
lossy system. We can see this by just solving the for the determinant of M from (10), which gives
1
det(M ) = ((tt0 rr0 ) + rr0 ) = t0 /t
t2
In the previous section, we found t = t0 for a lossless system, which gives det(M ) = 1. But this identity actually
extends to lossy systems. Including the loss channels in the scattering picture results in a N N unitary scattering
matrix which also must be symmetric due to time reversal symmetry, requiring that Sn,m = Sm,n . Therefore, for any
system that can be expressed in terms a N N scattering matrix,

det(M ) = 1 (15)

I. Free propagation transfer matrix

The transfer matrix for free propagation depends on the convention of the assumed time dependence of the traveling
wave. The wave vector represents the amplitude of a traveling wave at a specific location. The full expression for
a right-going traveling wave is given by Aeikxit . But we could have also used +, in which case eikx+it is the
right-going traveling wave. I copy Markos convention and assume a time dependence eit for the traveling wave.
Looking at eq (9), it is obvious that the free space transfer matrix for traveling some phase = kx is then given by
 i 
e 0
L() = (16)
0 ei

However, if I assumed a e+it time dependence, then my traveling wave would be the complex conjugate of this.

J. Combining transfer matrices

The transfer matrix basis is convenient because it relates the amplitudes on the left side to the amplitudes on
the right side. The total transfer matrix of a system can be obtained by multiplying the transfer matrices of all
the individual components. For the system in the figure I J, the amplitude vector in between systems 3 and 2 is
3,2 = M3 L . If we continue, we get R = M1 M2 M3 L , which means that we can define a transfer matrix M for
the combined system

M = M1 M 2 M3 (17)

Note that systems that are lined up left-to-right are multiplied right-to-left. But also note that usually we will be
solving for r and t from eq (14) which are the transmission and reflection coefficients for waves coming from the RHS,
in which case M1 is the first system the wave sees.

K. Simplifying the lossless transfer matrix

So far, the only requirement for transfer matrix is that det M = 1. But that still leaves a cumbersome 6 DOF for
the matrix. In the scattering picture, we converted the requirements for energy conservation and symmetries into
requirements for the transmission and reflection coefficients t, t0 , r, and r0 (see eq (8)). The final conclusion was that
there are 3 DOFs for a system that conserves energy and obeys time-reversal symmetry. In this section, I use the
requirements to simplify the most general lossless transfer matrix.
We start with the most general transfer matrix from eq (10):
 0
(tt rr0 )/t r/t

M= (18)
r0 /t 1/t
5

Remember that r and t are for the RHS wave, and r0 and t0 are for the left LHS wave. We are usually going to work
with r and t, so I would like to simplify the M1,1 and M2,1 matrix elements.
Using eq (8), I solve for r and 0r in terms of and t :

r = t /2 and 0r = t /2 + (19)

First I consider the M1,1 matrix element. I take out all of the phases to get:
i(t +r +0r ) 0
M1,1 = e , where = |r||r0 | |t|2 ei(2t r r ) (20)
|t|

The variable is defined because it is just equal to 1. Using eq (19), the phase in simplifies to 2t r 0r =
2t (t /2 ) (t /2 + ) = . And since |r0 | = |r|, the variable becomes = |r|2 + |t|2 = 1.
Next we look at the remaining phase in M1,1 . Again using eq (19), we get t + r + 0r = t + (t /2 ) +
(t /2 + ) = t . Inserting this and = 1 into the matrix element, we get

M1,1 = ei(t ) /|t| = 1/t (21)

Next we look at the matrix element M2,1 . Using eq (19) and |r| = |r0 |, we get

r0 |r0 | i(0r t ) |r| |r|


M2,1 = = e = ei((t /2+)t ) = ei(/2+) (22)
t |t| |t| |t|

Solving for in the first equation of eq (19), we get = t r /2. Inserting this into the matrix element, we get

|r| i(/2+(t r /2)) r


M2,1 = e = (23)
|t| t

Now we put the two simplified matrix elements back into M to get the most general lossless transfer matrix:

1/t r/t
 
Mlossless = (24)
r /t 1/t

It is important to note that this form of the transfer matrix is only valid for lossless systems. It makes no sense for
lossy systems. Like the lossless scattering matrix, it only has three DOFs. One could think of those three DOFs as
, t , and as we did for the scattering matrix. But I often find it more convenient to think of the three DOFs as
|r/t|, arg(r/t), and t . If a system has inversion symmetry, then we further know that arg(r/t) = /2, and so we
are reduced to two DOF.
Using eq (??), we could further understand these DOFs by taking out the phases of r/t and t. I kind of want to
do that later, but I havent yet. I think phases will often cancel out.

II. 1D CAVITY

A. The symmetric cavity

The ends of a finite periodic system form a cavity. This section reviews the properties of a symmetric cavity which
are relevant to the finite periodic system. We use the transfer matrix model to calculate the reflection and transmission
of a cavity. The mirrors are represented by
 
1 t2 r 2 r
Mmirror = (25)
t r 1
p
Next I assume that r is real and I take t = i 1 r2 . I am assuming a lossless mirror: |t|2 + |r|2 = 1. The transfer
matrix for two mirror and a single pass phase accumulation = kL, where k is the wave vector and L is the length
of the cavity) between them is

Mcavity = Mmirror L[] Mmirror (26)


6

The complex cavity reflection coefficient rcavity and transmission coefficient tcavity is ( using r = M1,2 /M2,2 and
t = 1/M2,2 ) :

r(1 ei2 ) (1 r2 )ei


rcavity = , tcavity = , where = kL (27)
1 r2 ei2 1 r2 ei2
The cavity is on-resonance when = n, and off-resonance when = (n + 1/2). These expressions are used to
derive other expressions for the cavity.
We can introduce loss into the system by letting the wave-vector be complex + i, where then e2 is
the single pass loss. Taking the absolute magnitude of the transmission, we get an expression for the overall power
transmission through a symmetric cavity with lossless mirrors with reflectivity R and transmission T , and with single
pass power loss e2 :
1
T cavity = (28)
L + F sin2 []

where
2
1 R e2 4R
L= and F = (29)
e2 T 2 T2
The variable F is called the coefficient of Finesse (see Hecht or wiki). It is large for good cavities. The variable L
is called the coefficient of loss. For a lossless cavity, = 0, and it equals 1. The coefficient of loss L is always
greater than 1. When the cavity is on-resonant,
1
T cavity |on-res = (30)
L
and when the cavity is off-resonant,
1
T cavity |off-res = (31)
L+F
These two expressions show the convenience of the definitions of L and F . By just measuring the on and off-resonant
values, we can extract the loss and mirror reflectivity.
For a lossless cavity ( = 0), we would just have 1 Rcavity = T cavity . For a lossy cavity, the reflection from the
cavity is related to the transmission by

1 Rcavity = L F sinh2 [] T cavity



(32)

The term in front of T cavity does not depend on , so we can treat at it as just an overall scale factor. We can also
now calculate the percent of power loss by looking at 1 (Rcavity + T cavity ).

B. Linewidth, Finesse, and Q

Next we look at the linewidth (units angular frequency) of the resonance. The Finesse is defined as

2fFSR c
F= , with fFSR = (33)
2Lng

Traditionally, the linewidth is defined as the full width where T = 1/2 for a cavity with no loss (see Hecht). Setting
L = 1 and solving for T cavity = 1/2 gives the half-width in = kL space. A small width in k space k can be
converted to small width in frequency space by = (d/dk)k = vg k. This then gives
h i
|T =1/2 = 4fFSR arcsin 1/ F (34)

But this definition is not very useful for low reflectivity cavities, since then the transmission never equals 1/2
and then the Finesse is imaginary. It also doesnt include loss. A much more useful definition of the line-width is by
7

defining it at where the transmission is halfway between the minimum 1/(L+F ) and maximum 1/L of the transmitted
power. In this case, the line-width is exactly
"s #
1
= 4fFSR arcsin (35)
2 + F/L

and
4R e2
F/L = (36)
(1 R e2 )2
Note that in the limit of good cavities where F/L  1, the two definitions are the same since
"s #
1 p
arcsin L/F (37)
2 + F/L

Another advantage of this new definition of linewidth is that it is scale invariant - it doesnt require a definition
of where T = 1, which sometimes is not obvious in a lossy transmission signal. So this is the definition of line-width
that we will use.
The Finesse is indepedent of the FSR given by

F= hq i (38)
1
2 arcsin 2+F/L

The cavity Q is defined by


L
Q= = F= ng F (39)
2fFSR /2
where is the free-space wavelength.
So as a summary, the linewidth and Q depend on both the free-spectral range and F/L. The Finesse depends only
on F/L.

C. Internal Field

The maximum and minimum intensity I = |E|2 in the cavity (without loss) is given by:
cavity 1R 2
Imax,min = 2 (1 R) (40)
(1 R)2 + 4R sin ()
Therefore, when on-resonance and then off-resonance, the maximum and minimum internal intensity is:

cavity,on-res 1 2 1 R
Imax,min = (1 R) = (41)
1R 1 R
cavity,off-res 1R 1R
Imax,min = 2
(1 R)2 = (42)
(1 + R) (1 R)2

D. Derivative of transmitted phase

Some peopel define the group index of a cavity as the derivative of the tranmitted phase, adn I just wanted to check
this. Taking the derivative of the transmission coefficient phase t , we get:

1R when off-resonant ( = /2)
dt 1 1+R c
= with fFSR = (43)
d 2fFSR 1+R when on-resonant ( = ) 2Ln g
1R
d
Here the group index ng = c/vg , with vg = d . For any spatially symmetric lossless blackbox, we must always have
dr dt
t r = /2, so d is identical to d except for a phase discontinuity for the phase which occurs when R = 0,
as seen in fig 1.
8

FIG. 1: Black = Cavity Transmitted , Red = Cavity Reflected, Gray = t (transmitted phase), Pink = r (reflected phase).
This is shown for three different single reflector reflection coefficients R.

III. ELECTROMAGNETIC STACK

A. Electromagnetic Waves with Transfer Matrix Method

Typically in EM, the transfer matrix is defined by


 +  +
ER EL
= ME (44)
ER EL

where E is the electric field amplitude. However, |E|2 is not proportional to intensity, but instead I n|E |2 . If
the index before and after the transfer matrix are not the same, then the corresponding scattering matrix is not

unitary, and none of the energy conservation and symmetry arguments are valid. If instead we define L = nL EL
q
,
R =

nR ER , and M = nnRL ME , then the scattering matrix is unitary, and all the formalism above holds. If the
index before and after systems is the same, then M = ME .
For a plane wave at normal incident with a dielectric surface with n1 on the left and n2 on the right, the transfer
matrix is
 
1 1+z 1z n1
ME = , where z = . (45)
2 1 z 1 + z n2

B. The EM Stack

FIG. 2: A 4 layered EM stack system.

The EM stack is a system composed of alternating dielectric layers of different index n1 , n2 and lattice constant
a1 , a2 . This problem is very easy to solve with transfer matrices, and it provides a convenient system for studying
properties of finite 1D periodic systems, for example what tapering the edges does.
I am using the same convention as above, where light enters from the right, so fig 2 starts with n1 , and next
dielectric is n2 . First I define the dielectric surface transfer matrix for n1 on the right and n2 on the left.
 
1 1+z 1z
Mn1,n2 = where z = n2 /n1 (?).
2 1z 1+z
and the free space propagation matrix

ei 0
 
M () = .
0 ei
9

We want to define the transfer matrix for a system consisting of N layers of the stack. We have to be careful,
because an actual structure is embedded in some dielectric, so we have to include the beginning and final index ni , nf
in the total transfer matrix. First I work out the total transfer matrix for 3 stacks to see what this will look like.
Defining 1 , 2 = kn1 , 2a1 , 2, the total transfer matrix for 3 stacks is
Mcell Mcell
z }| {z }| {
Mtot = Mni ,n1 L(1 )Mn1 ,n2 L(2 ) Mn2 ,n1 L(1 )Mn1 ,n2 L(2 ) Mn2 ,n1 L(1 )Mn1 ,n2 L(2 ) Mn2 ,nf (46)
My convention for the definition of the unit cell Mcell = Mn2 ,n1 L(1 )Mn1 ,n2 L(2 ) is arbitrary. Mcell propagates
the field amplitudes from immediately before the n2 n1 boundary to immediately before the n2 n1 in the next
unit cell. It is now obvious that a N-cell system can be written as
Mtot = Mni ,n1 Mn1
2 ,n1
N
Mcell Mn2 ,nf (47)
Mcell just consists of a slab of index n1 followed by a free propagation in n2 :
Mslab
z }| {
Mcell = Mn2,n1 L(1 )Mn1,n2 L(2 )
Since Mslab starts and stops in n2 and is lossless, it must take the form of eq (24):

1/t r/t
 
Mslab = (48)
r /t 1/t

where
r/t = i/2 (z 1/z) sin(1 )ei2
1/t = (cos(1 ) + i/2 (z + 1/z) sin(1 )) ei2 (49)
Here 1 = n1 ka1 and 2 = n2 ka2 , and z = n2 /n1 (I need to check this).
Now that we have the transfer matrix for the unit cell, we can use all the techniques developed above to understand
the bandgap. For example, inserting Mcell into the dispersion equation Inserting Mcell into Tr M = 2 cos a gives the
dispersion function:
 
1 n1 n2
cos(a) = cos(kn1 a1 ) cos(kn2 a2 ) + sin(kn1 a1 ) sin(kn2 a2 ) (50)
2 n2 n1
In the bandgap, is imaginary, representing a decay of the field. For a quarter-wave stack, kn1 a1 = kn2 a2 /2,
and these expression simplify. The dispersion equation is then solvable for (k).

IV. APPROXIMATIONS NEAR THE BANDGAP (NEED TO EDIT)

The reflected signal for any finite 1D periodic system with N cells of length a and with no tapering has N
reflection minima before the first bandgap (see Sec. ??). In -space, reflection minima are given by N a = p,
or = px /N , where integer p denotes the pth reflection minimum from = 0. The reflection maxima are given by:
N a = (2p + 1)/2, or = (p + 1/2)x /N . So the reflection minima are spaced by: = x /N , where x = /a.
This spacing also holds for finite periodic system with tapers as long as the tapered length is much smaller than
the periodic length. Using the Van Hove model, we can find approximations for the frequency and group index at
the reflection minima spacing (we care about the reflection minima because that is where the internal intensity and
imaginary self-Greens function is maximum). In these notes, p counts the reflection minima from = 0, and m counts
them from the bandedge = x .

Reflection minima spacing in -space: = x /N

Reflection minima spacing in space far from bandedge: 1 /N

  
mth spacing m-1 m 1 1 m0 2
= (2m 1), A:  1,  0
normal spacing 1 /N N 2 N
10

Group Index at mth reflection minimum ng (dm ) N


[A: dm  ,  1 ]
Group index far from bandedge = cx /1 1 m

So the frequency between the bandedge and the first reflection minimum is 02 /(2 N 2 ), and the group index
there is ng (d1 ) c x N
02
. One way to think about both assumptions made for these last two expressions is that the
spacing ratio and group index ratio cannot be less than one. So when either of them approach one, the approximation
breaks down. q q
cx
Away from band-edge, vg 0 /x , and near the band-edge vg x0 2d (or the group index is n g = 1

2d ).
Defining
p 2
F= A1 N ,
0
we get:
N, on-res
)
Imax F/m
 
1  1 m 2
N, off-res
A:  1, d  2, 2  12 (51)
Imax 4(m + 1/2)/F, A N

tapering cavity Bloch function


z  }|  { z }|
 { p  2
PC, on-res
p N 2 N 2 N 2
Imax,min A1 = A1
m 1 m 1 m 1

 
1  1 m 2 2 2
A:  1, d  2,  1 ,
A1 N

 
Bloch N 2 m  1  h m1 i
Imax,min , A: 1 (52)
m 1 N 2 N

[1] Note on Hdd and Ldd in a toy model: waveguide.


[2] Atoms in 1D Periodic Systems, J. D. Hood. March. 4.
[3] Pedersen, J. G., Xiao, S., Mortensen, N. A. (2008). Limits of slow light in photonic crystals. Physical Review B - Condensed
Matter and Materials Physics, 78(2), 5.
[4] Thyrrestrups thesis, Lodahls group.

You might also like