EECS 730, Winter 2009 K.

Sarabandi

Dyadic Analysis1
In this section a brief review of dyadic analysis is presented. Dyadic operations and
theorems provide an effective tool for manipulation of field quantities. 2 Dyadic notation
was first introduced by Gibbs in 1884 which later appeared in literature.3 Consider a
vector function F having three scalar component fi with (i = 1,2,3) in a Cartesian
system, that is

3
X
F = fi x̂i
i=1

Now consider three different vector functions F j , given by

3
X
Fj = fij x̂i j = 1, 2, 3 (1)
i=1

=
which constitute a dyad F given by

= 3
X 3 X
X 3
F= F j x̂j = fij x̂i x̂j (2)
j=1 i=1 j=1

The doublets x̂i x̂j (i, j = 1, 2, 3) form the nine-unit dyad basis. In matrix notation
 
=   f11 f12 f13
F = F 1, F 2, F 3 =  f21 f22 f23 
 

f31 f32 f33

It should be emphasized that

x̂i x̂j 6= x̂j x̂i for i 6= j

1
Copyright K. Sarabandi, 1997.
2
Tai, C.T., “Dyadic Green’s Functions in Electromagnetic Theory,” New York: IEEE Press, 2nd ed.,
1993.
3
Gibbs, J.W., “The scientific papers of J. Willard Gibbs” Vol. 2, pp. 84-90, New York: Dover, 1961.

1
In general a dyad can be formed from the product of two arbitrary vectors ā and b̄ to
= =
form F = āb̄. Components of F can be obtained from a matrix product of ā denoted
by a 3 × 1 matrix with 1 × 3 matrix b̄. Note that the converse may not be necessarily
true. That is, a general dyad may not be expressable in terms of product of two vector
like ab.
= =
Transpose of F , denoted by [F ]T , is defined by

= X XX
[F ]T = xj F j = fij x̂j x̂i .
j i j

= =
In matrix form [F ]T is simply the transpose of matrix F . A symmetric dyad is a dyad
for which
= =
[Fs ]T =F s .
=
A particular case of a symmetric dyad is the “idemfactor” (I ) described by

fij = δij

where δij is the Kronecker delta function. The idemfactor is explicitly expressed by

= 3
X
I= x̂i x̂i .
i=1

An antisymmetric dyad is defined by

= =
[Fa ]T = − Fa

It is easy to show that any dyad can be decomposed into a symmetric and antisymmetic
dyad components

= =
= 1 = T 1 = T
F = [F + F ] + [F − F ]
2 2

1 The Scalar Products

The “anterior scalar product” of a vector and a dyad is defined by

= 3 
X  3 X
X 3
a· F = a · F j x̂j = ai fij x̂j (3)
j=1 i=1 j=1

2
which is a vector. The “posterior scalar product” is defined by

= 3
X 3 X
X 3
F ·a = F j (x̂j · a) = aj fij x̂i (4)
j=1 i=1 j=1

which is also a vector. It is obvious that for a symmetric dyad

= =
a· F s =F s ·a

and

= =
a· I =I ·a = a

2 The Vector Products

Like scalar products, there are two different vector products between a vector a dyad.
The “anterior vector product” is defined by

= 3
X
a× F = (a × F j )x̂j (5)
j=1

and the “posterior vector product” is defined by

= 3
X
F ×a = F j (xj × a). (6)
j=1

3 Vector Dyadic Products

=
Consider two vectors a and b and a dyad C with vector components C j . Since

a · (b × C j ) = −b · (a × C j ) = (a × b) · C j j = 1, 2, 3 (7)

it can easily be shown that

= = =
a · (b× C ) = −b · (a× C ) = (a × b)· C (8)

Equation (8) is obtained from the three equations given by (7) after juxtaposing x̂j at
the posterior position of each equation and adding them up.

3
4 Differential Operations on Dyadic Functions
Differential operators such as divergence, gradient, and curl are often encountered in
analysis of electromagnetic field quantities. Hence it is useful to define these operators
when applied to dyads. The divergence of a dyadic function is defined by:

= 3
X
∇· F = ∇ · F j x̂j (9)
j=1

The resulting quantity is a vector. In a similar manner to the anterior vector product,
the curl of a dyadic is defined by

= 3
X
∇× F = (∇ × F j )x̂j (10)
j=1

which forms a dyad. Gradient of a vector function form a dyadic function which is given
by:

3 3 X
3
!
X X ∂fj
∇F = (∇fj )x̂j = x̂i x̂j (11)
j=1 i=1 j=1 ∂xi

5 Vector-Dyadic Green’s Second Identity

The Green’s second identity for vector functions can be used to develop the vector-dyadic
version of the theorem. For any two vector functions P and Qj which together with their
first and second derivatives are continuous it can be shown that4

ZZ Z ZZ
[P · ∇ × ∇ × Qj − (∇ × ∇ × P ) · Qj ]dv = [Qj × ∇ × P − P × ∇ × Qj ] · n̂ds (12)
v
ZZ
= [(∇ × P × n̂) · Qj + P · (n̂ × ∇ × Qj )]ds
s

=
Suppose Qj is one of the constituent vectors of dyad Q. By juxtaposing x̂j at the
posterior position of (12) and adding the resulting equations, it can easily be shown that

= = = =
ZZ Z ZZ
[P · ∇ × ∇× Q −(∇ × ∇ × P )· Q]dv = [P · (n̂ × ∇× Q) − (n̂ × ∇ × P )· Q]ds (13)
v

4
Stratton, J.A., “Electromagnetic Theory,” pp. 250, New York: Mc Graw-Hill, 1941.

4
which is the statement of vector-dyadic Green’s second identity.

6 Dyadic-Dyadic Green’s Second Identity

Another useful form of the Green’s second identity is its dyadic-dyadic form. This form
can be obtained directly from (13). Let us consider three vectors P j for j = 1, 2, 3.
Noting that
= =
P j · (n̂ × ∇× Q) = −(n̂ × P j ) · ∇× Q ,

using P j in (13), and then transposing this resulting equation, we get

ZZ Z  = =

T T
[∇ × ∇× Q] · P j − [Q] · ∇ × ∇ × P j dv = (14)
v
= =
ZZ  
− [∇× Q]T · (n̂ × P j) + [Q]T · (n̂ × ∇ × P j ) ds
s

Juxtaposing x̂j at the posterior position of (14) and summing the results, the dyadic-
dyadic Green’s second identity can be obtained and is given by

ZZ Z  = =

= =
[∇ × ∇× Q]T · P −[Q]T · ∇ × ∇× P dv = (15)
v
ZZ  = =

= =
T T
− [∇× Q] · (n̂× P ) + [Q] · (n̂ × ∇× P ) ds
s

5
EECS 730 Winter 2009 c K. Sarabandi

Dyadic Green’s Function

As mentioned earlier the applications of dyadic analysis facilitates simple manipulation
of field vector calculations. The source of electromagnetic fields is the electric current
which is a vector quantity. On the other hand small-signal electromagnetic fields satisfy
the linearity conditions and therefore the behavior of the fields can be described in terms
of the system impulse response. Since both the input (excitation current) and the output
(field quantities) of the system are vector quantities, the impulse response of the system
must be a dyadic quantity. In what follows the derivation of dyadic Green’s function
(impulse response for free space) is presented. Then the Fourier representation of the
Green’s function is derived which expresses the fields of an infinitesimal current source in
terms of a continuous spectrum of plane waves. This form of the dyadic Green’s function
is useful for further development of dyadic Green’s functions for more complicated media
such as a dielectric half-space medium or a stratified (multi-layer) dielectric medium.
Consider an arbitrary time-harmonic electric current distribution J e in an unbounded ho-
mogeneous medium with permittivity ǫ and permeability µ. Starting from the Maxwell’s
equation, the vector wave equation for the electric field can be obtained and is given by:

∇ × ∇ × E(r) − k 2 E(r) = iωµJ e (r) (1)

where J e (r) is the impressed volumetric current distribution. As shown previously the
electric field is usually calculated indirectly from the electric Hertz potential and is given
by:
 
E(r) = k 2 + ∇∇· Πe (r) (2)

where

iZ
Z
Πe (r) = J e (r ′ )g(r, r′ )dv ′ (3)
k v

The electric field expression given by (2) is valid for all r in this medium including source
points. Here
′
1 eik|r−r |
g(r, r ′ ) =
4π |r − r ′ |

1
is the scalar Green’s function satisfying the scalar wave equation
 
∇2 + k 2 g(r, r′ ) = −δ(r − r ′ ) (4)

Let us now consider an infinitesimal current source along x̂ direction given by

x̂ x̂
J e (r) = δ(r − r ′ ) = δ(r − r′ ) (5)
iωµ ikZ
According to (3) and (2) the resulting electric field can be obtained from

1
 
x
G (r, r ) = 1 + 2 ∇∇· g(r, r′ )x̂
′
k
x
where G (r, r′ ) denotes the impulse response to an x-directed excitation. In a similar
manner the electric field in response to infinitesimal y-directed and z-directed currents
are given by

1
 
y
G (r, r ) = 1 + 2 ∇∇· g(r, r ′ )ŷ
′
k
1
 
z
G (r, r ′ ) = 1 + 2 ∇∇· g(r, r′ )ẑ
k

Using the compact dyadic notation, the electric field due to an arbitrary oriented
p̂
(along p̂) infinitesimal current ikZ δ(r − r′ ) can be obtained from:

p =
E (r, r′ ) =G (r, r′ ) · p̂
= x y z
where G (r, r′ ) = G (r, r′ )x̂ + G (r, r ′ )ŷ + G (r, r′ )ẑ is referred to as the dyadic Green’s
=
function of free-space. The explicit expression for G (r, r′ ) is given by

1
 
=
′ ′
G (r, r ) = 1 + 2 ∇∇· g(r, r )(x̂x̂ + ŷŷ + ẑ ẑ)
k
1
 
=
= 1 + 2 ∇∇· g(r, r′ ) I
k
 
= = =
where I is the unit dyad (idemfactor). Noting that ∇ · ψ I = ∇ψ· I = ∇ψ for any
differentiable scalar function ψ, the expression for the dyadic Green’s function is given
by

2
 = 1 =
 
′ ′
G (r, r ) = I + 2 ∇∇ g(r, r ) (6)
k
=  q

Referring to (1) each vector component of G (r, r ′ ) G (r, r′ ); q = x, y, z satisfies

q q
∇ × ∇ × G (r, r′ ) − k 2 G (r, r ′ ) = q̂δ(r − r ′ ) (7)

By juxtaposing a unit vector x̂, ŷ, or ẑ at the posterior position of the three vector
equations given by (7) and summing these equations, we obtain

= = =
∇ × ∇× G (r, r′ ) − k 2 G (r, r′ ) =I δ(r − r′ ) (8)

1 Derivation of Field Quantities From the Dyadic

Green’s Function
Consider a homogeneous medium bounded by a closed surface S which includes an
arbitrary electric current distribution J e (r). Using the vector wave equation (1) and (8)
in conjunction with the vector-dyadic Green’s theorem given by

ZZZ  = =

P · ∇ × ∇× Q −(∇ × ∇ × P ) · Q dv = (9)
v
ZZ  = =

− (n̂ × ∇ × P )· Q +(n̂ × P ) · ∇× Q ds .
s

an explicit expression for the electric field due to the impressed electric current can be
= =
obtained. By letting P = E(r) and Q=G (r, r′ ) it can easily be shown that

=
ZZZ
E(r ′ ) = ikZ Je (r)· G (r, r ′ )dv (10)
v
ZZ  
= =
′ ′
− (n̂ × ∇ × E(r))· G (r, r ) + (n̂ × E(r)) · ∇× G (r, r ) ds .
s

Noting that ∇ × E(r) = ikZH(r), (10) can be written as

=
ZZZ
′
E(r ) = ikZ J e (r)· G (r, r ′ )dv (11)
v
ZZ  
= =
− ikZ(n × H(r))· G (r, r′ ) + (n̂ × E(r)) · ∇× G (r, r ′ ) ds
s

3
To find an expression for the magnetic field, we start with the vector wave equation for
the magnetic field given by

∇ × ∇ × H(r) − k 2 H(r) = ∇ × J e (r) (12)

= =
Again by letting P = H(r) and Q=G (r, r ′ ) in (9), we obtain

=
ZZZ h i
′
H(r ) = ∇ × J e (r) · G (r, r′ )dv (13)
v
ZZ  
= =
′ ′
− (n̂ × ∇ × H(r))· G (r, r ) + (n̂ × H(r)) · ∇× G (r, r ) ds
s

Applying the dyadic identity

= = =
∇ · (a× b ) = ∇ × a· b −a · ∇× b ,

The volume integral in (13) can be written as

ZZZ    
= = =
ZZZ h i
′ ′ ′
∇ × Je (r) · G (r, r )dv = ∇ · Je (r)× G (r, r ) + Je (r) · ∇× G (r, r ) dv
v v

Using the divergence theorem

   
= =
ZZZ ZZ
′ ′
∇ · Je (r)× G (r, r ) dv = n̂ · Je (r)× G (r, r ) ds
v s
=
ZZ h i
= n̂ × Je (r) · G (r, r ′ ) ds
s

and Maxwell’s equation

∇ × H(r) = −ikY E(r) + Je (r)

in (13), the expression for the magnetic field reduces to

=
ZZZ
H(r ′ ) = J e (r) · ∇× G (r, r′ )dv (14)
v
= =
ZZ  h i 
′ ′
+ ikY n × E(r) · G (r, r ) − (n̂ × H(r)) · ∇× G (r, r ) ds
s

4
2 Field Quantities Generated from Magnetic and
Electric Currents
Equations (11) and (14) provide the electric and magnetic field quantities in a bounded
region originated from an electric current distribution and a certain surface field quan-
tities at the surface of this bounded region. In this section these results are extended
to allow for the existence of both electric and magnetic currents. This can easily be
done by first obtaining the field expressions using a magnetic current distribution as
the excitation. The duality relations can be employed to find the field quantities for
a magnetic current excitation from those given by (11) and (14). We first point out
that the magnetic dyadic Green’s function for an unbounded homogeneous medium is
the same as the electric one. Apply the duality relations to (11) and (14) the following
expressions are obtained

=
ZZZ
′
H m (r ) = ikY J m (r)· G (r, r′ )dv (15)
v
ZZ  
= =
− −ikY (n̂ × E m (r))· G (r, r ′ ) + (n̂ × H m (r)) · ∇× G (r, r′ ) ds
s
=
ZZZ
′
E m (r ) = − J m (r) · ∇× G (r, r ′ ) dv (16)
v
ZZ  
= =
− ikZ(n̂ × H m (r))· G (r, r′ ) + (n̂ × E m (r)) · ∇× G (r, r′ ) ds
s

Superposition of (11) and (16) and (14) and (15) provides the total fields within S and
are given by

= =
ZZZ  
′ ′ ′
E(r ) = ikZJ e (r)· G (r, r ) − J m (r) · ∇× G (r, r ) dv (17)
v
ZZ  
= =
− ikZ(n̂ × H(r′ ))· G (r, r ′ ) + (n̂ × E(r ′ )) · ∇× G (r, r′ ) ds
s
 
= =
ZZZ
′ ′ ′
H(r ) = J e (r) · ∇× G (r, r ) + ikY J m (r)· G (r, r ) dv (18)
v
ZZ  
= =
′ ′
+ ikY (n̂ × E(r))· G (r, r ) − (n̂ × H(r)) · ∇× G (r, r ) ds
s

3 Radiation Condition For Dyadic Green’s Function

The contribution from the surface integrals of (13) and (14) should vanish as the surface
approaches infinity according to the radiation condition first postulated by Sommerfeld.

5
The electric field far away from the source and observation points satisfies
n o
lim r ∇ × E(r) − ikr̂ × E(r) = 0 (19)
r→∞

x y
The magnetic field also satisfies an identical equation. Using (19) for G (r, r′ ), G (r, r ′ ),
z
and G (r, r′ ) and by juxtaposing unit vectors x̂, ŷ, and ẑ at the posterior position of
each equation respectively and then adding the three resulting equations we get
 
= =
′ ′
lim r ∇× G (r, r ) − ikr̂× G (r, r ) = 0 (20)
r→∞

which is known as the radiation condition for the free-space dyadic Green’s function.

4 Explicit Forms of The Dyadic Green’s Function

The compact form of the dyadic Green’s function which is given by
′
= = 1 eik|r−r |
 
′
G (r, r ) = I + ∇∇
k2 4π|r − r ′ |
can be expressed in any desired coordinate system. For example, in Cartesian coordinate
system, where

∂ ∂ ∂
∇= x̂ + ŷ + ẑ ,
∂x ∂y ∂z
the dyadic Green’s function can be represented, in matrix form, in the following manner

∂2 ∂2 ∂2
k2 +
 
∂x2 ∂x∂y ∂x∂z ′
=
′ ∂2 ∂2 ∂2 eik|r−r |
k 2 + ∂y
 
G (r, r ) =  2
 (21)
 ∂y∂x
∂2 ∂2
∂y∂z
∂2
 4πk 2 |r − r′ |
∂z∂x ∂z∂y
k 2 + ∂z 2

=
It is quite obvious from (21) that G (r, r′ ) is a symmetric dyad, i.e.
T
= =

′ ′
G (r, r ) = G (r, r ) (22)

Therefore, for any vector V , we have:

= =
V · G (r, r ′ ) =G (r, r′ ) · V

6
 =
Also noting that ∇ × ∇ = 0, ∇× G (r, r′ ) can easily be evaluated as follows

1
    
= = =
∇× G (r, r ) = ∇ × I + 2 ∇∇ g(r, r ′ ) = ∇ × I g(r, r′ )
′
k
=
′
= ∇g(r, r )× I

which in Cartesian coordinate system takes the following form

∂ ∂
0 − ∂z
 
∂y ik|r−r ′ |
∂  e
=
∇× G (r, r ′ ) =  ∂
0 − ∂x  (23)
∂z
4π|r − r ′ |

∂ ∂
− ∂y ∂x
0
=
which is obviously anti-symmetric (any dyad of the form C× I is anti-symmetric).
=
Another expanded form of G (r, r′ ) can be obtained by noting that

d 1
∇g(R) = g(R)∇R = (ik − )g(R)∇R
dR R
1

= ik − g(R)R̂ ,
R

where R = |r − r ′ | and

r − r′
R̂ = .
|r − r ′ |
Hence,

1 1
    
∇∇g(R) = ∇ ik − g(R) R̂ + ik − g(R)∇R̂ , (24)
R R
∇R̂ can be calculated easily noting that,
!
R ∇(R) 1
 
∇(R̂) = ∇ = + R∇
R R R
=
But ∇(R) =I and therefore

1
 
=
∇(R̂) = I −R̂R̂
R
After some algebraic manipulations it can be shown that

7
 = 3 3i i 1 =
    
′
G (r, r ) = 2 2
− − 1 R̂R̂ + 1 + − 2 2 I g(R) (25)
k R kR kR k R

5 Far Field Expression of Dyadic Green’s Function

In the evaluation of fields away from the sources where |r −r ′ | is much larger than typical
dimension of the source, simple expressions for the field quantities are usually obtained.
Keeping only the terms of the order of R1 ≃ 1r , the far-field expression for the free-space
dyadic Green’s function can be obtained from (25) and is given by

′
= eik|r−r |
=
 
′
G (r, r ) ≃ I −r̂r̂ (26)
4πr
 ikr
e

= ′
≃ I −r̂r̂ e−ikr̂·r
4πr

Equation (26) indicates that the field quantities do not possess a radial component.

6 Fourier Representation of The Free-Space Dyadic

Green’s Function
Another useful representation of the dyadic Green’s function is its Fourier Transform
where the field response to an impulse excitation is expressed in terms of a continuous
spectrum (angular) of plane waves. This expansion in terms of plane waves is useful
since the scattering solution of many problems to plane wave excitation is known. Using
the plane wave solution together with the superposition principle, the solution to any
arbitrary source can be obtained.
The starting point is equation (4). Let us assume, without loss of generality, that the
source point is at the origin. Then

∂2 ∂2 ∂2
!
+ + + k02 g(r) = −δ(r) (27)
∂x2 ∂y 2 ∂z 2

The Fourier transform of g(r), represented by g̃(k), is given by

+∞
ZZZ
g̃(k) = g(r)e−i(kx x+ky y+kz z) dx dy dz
−∞

8
Conversely g(r) in terms of its Fourier transform is obtained from

+∞
1
ZZZ
g(r) = g̃(k)ei(kx x+ky y+kz z) dkx dky dkz (28)
(2π)3 −∞

Substituting (28) in (27) and noting that

+∞
1
ZZZ
δ(r) = ei(kx x+ky y+kz z) dkx dky dkz
(2π)3 −∞

g̃(k) can be evaluated and is given by

1
g̃(k) = (29)
kx2 + ky2 + kz2 − ko2

Although the 3-dimensional Fourier

 transform can be used to express the field quantities
in terms of plane waves eik·r , it is not usually used because all three components of
the propagation vector are independent, that is, the frequencies of these plane waves are
not necessarily the same. To constrain the propagation vector k the integration with
respect to one of the variables must be carried out. We consider integration of (28) over
kz , that is

+∞
1 1
Z
I(z) = eikz z dkz ; h2 = ko2 − kx2 − ky2 (30)
2π −∞ kz2 −h2

Considering the behavior of g(r), we expect that I(z) approaches zero at z = ±∞. This
is justifiable if we let h to be complex with Im[h] > 0. Such assumption is common
and corresponds to a slightly lossy media. After evaluation of the integral, the lossless
condition is restored by allowing Im[h] → 0. With this assumption the locations of the
poles of the integrand (30) are shown in Figure 1. The contour of integration is assumed
to be along the real axis. For z ≥ 0 the contour can be closed in the upper half-plane with
a semi-circle of a large radius (R → ∞) noting that Im[kz ] > 0 (a radiation condition
requirement). In this case the integrand along the semi-circle contour is zero.
For z ≤ 0, the contour can be closed in the lower half-plane. Using Cauchy’s residue
theorem to the contour integrals, I(z) can easily be evaluated and is given by
(
i eihz z ≥ 0 i ih|z|
I(z) = = e
2h e−ihz z ≤ 0 2h
q
where h = ko2 − kx2 − ky2 . Replacing h with kz and keeping in mind that kz is no longer
an independent parameter, (28) takes the following form

9
 Im [kz]

Path of integration
for z > 0

kz = h

Re [kz]
kz = –h

Figure 1: kz -plane and the location of the poles of integrand (26).

ZZ+∞ +ik⊥ ·r⊥ +ikz |z|

i e dk ⊥
g(r) = 2
(31)
(2π) −∞ 2kz
where
k ⊥ = kx x̂ + ky ŷ
r ⊥ = xx̂ + y ŷ
q
kz = k 2 − kρ2 , Im[kz ] > 0
kρ2 = kx2 + ky2
=
To find G (r), derivatives of g(r) must be evaluated, however, it should be noted that
the derivative of g(r) with respect to z is discontinuous, that is,
 +∞ 
∂  −1 ZZ 1
ik ⊥ ·r⊥ +ikz |z|

g(r) = e dk ⊥ f (z)
∂z  (2π)2 2 
−∞

where f (z) is a step function

(
1 z>0
f (z) =
−1 z < 0

Further differentiation with respect to z will give a dirac delta function

ZZ+∞
 
∂2  1 
g(r) = − eik⊥ ·r⊥ dk⊥ δ(z) −
∂z 2  (2π)2
−∞


10
 ZZ+∞
 
 i kz 
eik⊥ ·r⊥ +ikz |z| dk⊥ f 2 (z)
 (2π)2 2 
−∞

+∞
2 1
eik⊥ ·r⊥ dk⊥ = δ(x)δ(y) and therefore
RR
But f (z) = 1 and 2
(2π) −∞

ZZ+∞
∂2 i kz i(k⊥ ·r⊥ +kz |z|)
2
g(r) = −δ(r) − e dkx dky (32)
∂z (2π)2 −∞
2
=
Substituting (31) in (6) and using (32) a simple expression for G (r) can be obtained.
∂ ∂
Interchanging the order of differentiation and integration and noting that ∂x , ∂y and
∂
∂z
can be replaced with ikz , iky and ±ikz ( + sign for z > 0 and − sign for z < 0 )
respectively it can easily be shown that

 +∞
=
 
i 1
I − kk eik·r d k⊥ z > 0
 RR
−ẑ ẑ

=  8π 2 −∞ kz k2
G (r) = δ(r) +  +∞ (33)
k2
 
=
i 1
− KK iK·r
RR


8π 2 −∞ kz I k2
e d k⊥ z < 0

where

k = kx x̂ + ky ŷ + kz ẑ
K = kx x̂ + ky ŷ − kz ẑ

Equation (33) is the expansion of the free-space dyadic Green’s function in terms of a
continuous spectrum of plane waves (monochromatic) propagating along the vectors k
and K which are in  general complex quantities.
 Propagation vector K is the mirror image
of k in x-y plane K = k − 2(k · ẑ)ẑ and represents plane waves propagating along the
−ẑ direction. Another useful representation appropriate for planar boundaries can be
=
obtained by decomposing the vectors in G (r) into TE and TM components. Recognizing
that k/k is a unit vector (k̂) the horizontal (TE) and vertical (TM) unit vectors are,
respectively, defined by

k̂ × ẑ ky x̂ − kx ŷ 1
ê(kz ) = = q = (ky x̂ − kx ŷ)
|k̂ × ẑ| kx2 + ky2 kρ
−kz kρ
ĥ(kz ) = ê × k̂ = (x̂kx + ŷky ) + ẑ
kkρ k
 
The triplet ĥ, ê, k̂ form an orthonormal system, and therefore

11
 =
I −k̂ k̂ = êê + ĥĥ (34)
K
A similar orthonormal system can be formed with K̂ = k
instead of k̂. In this system
the horizontal and vertical unit vectors are given by:

K̂ × ẑ
ê(−kz ) = = ê(kz )
|K̂ × ẑ|
ĥ(−kz ) = ê × K̂

and as before

=
I −K̂ K̂ = ê(kz )ê(kz ) + ĥ(−kz )ĥ(−kz ) (35)

Inserting (34) and (35) into (33) and translating the source from the origin to r ′ , the
free-space dyadic Green’s function takes the following form:

+∞
 h i
i 1 ′
êê + ĥ(kz )ĥ(kz ) eik·(r−r ) d k⊥ z > z′
 RR
−ẑ ẑ

=  2
8π −∞ kz
′ ′
G (r, r ) = 2 δ(r − r ) +  +∞ (36)
k 
 i RR 1
h
êê + ĥ(−kz )ĥ(−kz ) e
i
iK·(r−r′ )
d k⊥ z < z ′
8π 2 −∞ kz

This form of the dyadic Green’s function is usually not appropriate for numerical eval-
uation, especially when z − z ′ << λ. In this case the convergence rate of the integral
is very poor.

7 Dyadic Green’s Function for Two-Dimensional Prob-

lems
In some electromagnetic scattering problems where the geometry of the problem is in-
dependent of one coordinate variable the formulation of the problem can be made some-
what simpler. Without loss of generality let us assume that the scatterer geometry is
independent of z. In this case the scatterer is a cylinder of arbitrary cross-section whose
axis is along ẑ. Since there is no variation with respect to z, all field quantities take the
z dependence of the excitation. Suppose

J(r) = J(ρ)eikzi z

12
where ρ = xx̂ + y ŷ. The differential operator can also be made explicit with respect to
∂
∂z
which is replaced by ikzi, that is

∇ = ∇t + ikzi ẑ
∂ ∂
where ∇t = ∂x x̂ + ∂y ŷ. Since the dependence of J(r) with respect to z is explicit, in
field calculations the integral with respect to z can be carried out. Therefore the 2-D
dyadic Green’s function is given by
′
+∞ = 1 = 1 +∞ eik|r−r | ikzi z ′ ′
 Z
=
Z
′ ′ ikzi z ′ ′
G (ρ, ρ ) = G (r, r )e dz = I + 2 ∇∇ e dz
−∞ 4π k −∞ |r − r ′ |
Using the identity
√
|ρ−ρ′ |2 +(z−z ′ )2
+∞ eik
Z
′ (1)
q eikzi z dz ′ = iπH0 (kρ |ρ − ρ′ |)eikziz (37)
−∞ |ρ − ρ′ |2 + (z − z ′ )2

q
2
where kρ = k 2 − kzi , the 2-D dyadic Green’s function takes the following form:

= i = 1 
 
′ 2 (1) ′ ik z
G (ρ, ρ ) = I + 2 ∇t ∇t + ikzi∇t ẑ + ikzi ẑ∇t − kzi ẑẑ H0 (kρ |ρ − ρ |)e zi . (38)
4 k

In matrix form (38) becomes

1 ∂2 1 ∂2 ikzi ∂
 
1+ k 2 ∂x2 k 2 ∂x∂y k 2 ∂x
 
 
=
′

1 ∂2 1 ∂2 ikzi ∂
 i (1)
G (ρ, ρ ) = 1+ H (kρ |ρ − ρ′ |)eikzi z (39)
 
k 2 ∂y∂x k 2 ∂y 2 k 2 ∂y
4 0
 
 
 
 
ikzi ∂ ikzi ∂ kρ2
k 2 ∂x k 2 ∂y k02

8 Fourier Representation of 2-D Dyadic Green’s Func-

tion
A procedure similar to what was shown for 3-dimensional dyadic Green’s function can
be followed to obtain the Fourier representation of 2-D dyadic Green’s function. The
Fourier representation can also be obtained in a simpler way using the following identity

(1)
q 1 Z +∞ 1 ikx (x−x′ )+iky |y−y′ |
H0 (kρ (x − x′ )2 + (y − y)2 ) = e dkx (40)
π −∞ ky

13
 q
where ky = kρ2 − kx2 . Substituting (40) in (39) and after some algebraic manipulations
it can be shown that


=
 
ieikzi z R +∞ 1 ′ ′
I − kk ei[kx (x−x )+ky (y−y ) dkx y > y ′

−ŷ ŷ

=  4π −∞ ky k2
′ ′ ik z
G (ρ, ρ ) = 2 δ(ρ − ρ )e zi +  
R 1 =
 (41)
k ieikzi z
− KK
′ ′
ei[kx (x−x )−ky (y−y )] dkx y < y ′
ky I

4π k2


where

k = kx x̂ + ky ŷ + kzi ẑ
K = kx x̂ − ky ŷ + kzi ẑ .

As before k can be considered as the propagation vector of a plane wave going along
positive y direction and K is that of a wave going along −y direction. kzi is a fixed
known quantity.

9 Symmetrical Property of Dyadic Green’s Function

Symmetrical property of dyadic Green’s function allows for simple evaluation of the
dyadic Green’s function when the locations of source and observation points are inter-
changed. To demonstrate this property the dyadic-dyadic Green’s second identity given
by

= T
ZZZ    = T
= =
∇ × ∇× Q ·P − Q · ∇ × ∇× P dv = (42)
v
ZZ ( = T    = T
  )
= =
− Q · n̂ × ∇× P + ∇× Q · n̂× P ds
s

will be employed. Let us consider two situations where in each case the source location
is at r a and rb respectively. The dyadic Green’s functions for each case must satisfy

= = =
∇ × ∇× G (r, r a ) − k 2 G (r, r a ) = δ(r − ra ) I (43)
= = =
2
∇ × ∇× G (r, rb ) − k G (r, rb ) = δ(r − rb ) I (44)

= = = =
Substituting G (r, r a ) for Q and G (r, rb ) for P in (42) and using the radiation condition
at infinity and equations (43) and (44) it can readily be shown that

14
 T
= =

G (r a , r b ) = G (rb , ra ) (45)

That is, the dyadic Green’s function when the source is at rb and the observation point
is at r a is the transpose of the Green’s function when the source point is at ra and
the observation point is at r b . Although the proof is given for the free space Green’s
function, (40) is a general result. Equations (21) and (25) show that the free space
Green’s function is symmetric, i.e.,
T
= =

′ ′
G (r, r ) = G (r, r )

In view of (45) for free space Green’s function we have

= =
′ ′
G (r, r ) =G (r , r) (46)

However, it should be noted that (46) is not a general result.

10 Dyadic Green’s Function for Piece-Wise Homoge-

nous Media
In the previous sections we considered the properties of dyadic Green’s function for
homogeneous media. In practice however, the medium of interest is often complex
which may be composed of many homogeneous media such as the one shown in Figure
2.
For these problems it is usually desired to derive the expression for a dyadic Green’s
function which satisfies the necessary field boundary conditions at the interfaces.
The field quantities in response to a volumetric electric current distribution J e (r) in the
unbounded inhomogeneous medium are simply given by

=
ZZZ
′
E(r ) = ikn Zn J e (r)· G (r, r′ ) dv (47)
v
=
ZZZ
H(r ′ ) = ∇′ × [Je (r)· G (r, r ′ )]dv (48)
v

=
The simple form of (47) is obtained by imposing certain boundary conditions on G (r, r ′ ).
To derive these boundary conditions consider a simple medium composed of two homo-
geneous media. Suppose the source exists only in medium 1 where we have

15
 µ2 2

µ1 1

P.E.C.
µ3 3

Figure 2: A complex medium composed of a number of homogeneous media and perfect

electric conductors.

∇ × ∇ × E 1 (r) − k12 E1 (r) = iωµ1J 1 (r) (49)

and in the second medium

∇ × ∇ × E 2 (r) − k22 E2 (r) = 0 (50)

= (11)
Let us denote the expression for the dyadic Green’s function in medium 1 by G (r, r ′ )
which satisfies

= (11) = (11) =
∇ × ∇× G (r, r′ ) − k12 G (r, r ′ ) =I δ(r − r ′ ) (51)
= (21)
and the dyadic Green’s function in medium 2 by G (r, r ′ ) which satisfies

= (21) = (21)
∇ × ∇× G (r, r′ ) − k22 G (r, r ′ ) = 0 . (52)

The application of the vector-dyadic Green’s second identity to (49) and (51) gives:
= (11)
ZZZ
′
E 1 (r ) = iωµ1 J 1 (r)· G (r, r ′ )dv ′ (53)
v
= (11) = (11)
ZZ  
′ ′
+ iωµ1 (n̂1 × H 1 (r1 ))· G (r, r ) + (n̂1 × E 1 (r)) · ∇× G (r, r ) ds
s

where s is the boundary between the two media.

16
The application of the vector-dyadic Green’s second identity to (50) and (52) provides

= (21) = (21)
ZZ  
′ ′
iωµ2 (n̂1 × H 2 (r))· G (r, r ) + (n̂1 × E 2 (r)) · ∇× G (r, r ) ds = 0 (54)
s

Noting that

n̂1 × H 1 (r) = n̂1 × H 2 (r)

n̂1 × E 1 (r) = n̂1 × E 2 (r)

and using the following identities

= (21) = (21)
n̂1 × H 1 (r)· G (r, r′ ) = −H 1 (r) · (n̂1 × G (r, r′ ))
= (21) = (21)
n̂1 × E 1 (r) · ∇× G (r, r′ ) = −E 1 (r) · n̂1 × (∇× G (r, r ′ ))

the contribution from the surface integral of (53) can be shown to vanish if

= (11) = (21)
µ1 n̂× G (r, r′ ) = µ2 n̂× G (r, r′ ) (55)

and

= (11) = (21)
n̂ × ∇× G (r, r′ ) = n̂ × ∇× G (r, r′ ) (56)

Equations (55) and (56) are the necessary boundary conditions for the dyadic Green’s
function.
On the surface of perfect electric conductors

n̂(r) × E(r) = 0
=
which mandates n̂(r)× G (r ′ , r) = 0.
The symmetry property of the dyadic Green’s function can be shown easily by following
the same procedure outlined for the free space dyadic Green’s function. Let us consider
two experiments where in one experiment the source is placed at an arbitrary point ra
and the observation point is at r in the nth region. In the second experiment we place
the source point at an arbitrary point r b while keeping the observation point at the same
location r as shown in Figure 3. In both experiments the dyadic Green’s functions satisfy
(51) (without the superscripts). Applying the dyadic-dyadic Green’s second identity to

17
 rb

r r
ra

n th n th
region region
Experiment #1 Experiment #2

Figure 3: Locations of sources and observation points in a complex dielectric medium

for demonstrating the symmetry property of dyadic Green’s function.

these dyadic Green’s functions which satisfy the radiation condition, it can readily be
shown that
 T
= =
G (r a , rb ) = G (rb , ra ) . (57)

It should be noted that it is not very easy to find a dyadic Green’s function for a
general piece-wise homogeneous media with arbitrary boundaries. In such cases it is
more convenient to use the free space dyadic Green’s function with surface integrals
given by (11) and (14).
In view of (57), equation (47) can be written as

=
ZZZ
E(r′ ) = i kn Zn ′
G (r , r) · J e (r) dv
v

where we used the fact that

 T
= =
′ ′
J e (r)· G (r, r ) = G (r, r ) · J e (r)

and T
= =

′
G (r, r ) =G (r ′ , r).

Now interchanging r ′ with r and vice versa we get

=
ZZZ
′ ′ ′
E(r) = ikn Zn G (r, r ) · J e (r )dv (58)
v

18
11 Dyadic Green’s Function For Inhomogeneous Me-
dia
Consider an inhomogeneous isotropic medium whose permittivity and permeability are
functions of position and are, respectively, denoted by ǫ(r) and µ(r). As before we are
seeking simple expressions for electric and magnetic fields for an arbitrary source using
the impulse response of the medium. It should be emphasized that the evaluation of the
dyadic Green’s function for this type of problem, in general, is very complex; however,
here a formal analysis is provided1 . Taking the curl of the modified Amper’s law, it can
be shown that the vector wave equation takes the following form
" #
1
∇× ∇ × E(r) − ω 2 ǫ(r) E(r) = iωJ e (r) (59)
µ(r)

Comparing (1) and (59)for this problem one can define a Green’s function such that it
would satisfy
" #
1 = = 1 =
∇× ∇× G (r, r′ ) − ω 2ǫ(r) G (r) = ′
I δ(r − r ) (60)
µ(r) µ(r)

Noting that
! !
1 1 1
∇· P ×∇×Q = ∇×Q·∇×P −P ·∇× ∇×Q
µ(r) µ(r) µ(r)

The Green’s second vector identity can be written as

ZZZ ( " # " #)

1 1
P ·∇× ∇×Q −Q·∇× ∇ × P dv = (61)
v
µ µ
1n
ZZ o
= Q × ∇ × P − P × ∇ × Q · n̂ ds
s
µ

from which the vector-dyadic Green’s second identity can be obtained and is given by

ZZZ ( " # " # )

1 = 1 =
P ·∇× ∇× Q − ∇ × ∇ × P · Q dv (62)
v
µ(r) µ(r)
1 =  =
ZZ    
= P · n̂ × ∇× Q − n̂ × ∇ × P · Q ds
s
µ(r)
1
W.C. Chew, “ ”.

19
 = =
Substituting E for P and G for Q in (62), applying (59) and (60), and using the radiation
condition it can easily be shown that

=
ZZZ
E(r) = iωµ(r) J e (r′ )· G (r ′ , r)dv ′ (63)
v

The magnetic field can be obtained from the application of Faraday’s law (H(r) =
1
iωµ(r)
∇ × E(r)).

1
 
=
ZZZ
′
H(r) = J(r) · ∇ × µ(r) G (r , r) dv (64)
µ(r) v

Using the dyadic-dyadic Green’s second identity

 !#T 
ZZZ " !
1 = =
T
= 1 = 
∇× ∇× Q · P −[Q] · ∇ × ∇× P dv =
v
 µ(r) µ(r) 
(  )
1 = T = T
    
= =
ZZ
− Q · n̂ × ∇× P + ∇× Q · n̂× P ds
s
µ(r)

it can easily be shown that

 T
= =
′ ′ ′
µ(r ) G (r, r ) = µ(r) G (r , r) . (65)

20