Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Metamaterial homogenization: extraction of effective

constitutive parameters
Chris Fietz and Gennady Shvets
University of Texas at Austin, Dept. of Physics, Austin TX 78712;

ABSTRACT
We present a numerical algorithm for extracting all 36 linear constitutive parameters of a metamaterial crystal as
a function of frequency and wavenumber based on driving a metamaterial with electric and magnetic charge and
current. We demonstrate how spatial dispersion can result in bianisotropy in a centrosymmetric crystal. Several
tests are performed on a 2D metamaterial crystal to validate these ”current driven” constitutive parameters.
Finally, we show how our method can be used to study spatial dispersion by studying constitutive parameters
for small k.
Keywords: metamaterial homogenization, constitutive parameters, metamaterial antenna

1. INTRODUCTION
Since the discovery of artificial materials with negative magnetic response about a decade ago,1 considerable
work has gone into the study of electromagnetic metamaterials (artificial materials engineered to have exotic
optical properties). In that time both analytic1, 2 and numerical3–6 methods have been suggested for determining
the constitutive parameters of such metamaterials. Although analytic expressions are useful for approximating a
metamaterial’s response and providing intuition, ultimately numerical methods are needed to accurately calculate
constitutive parameters. Currently, all proposed numerical methods are severely limited to extracting a small
number of the total 36 linear constitutive parameters. In this paper we outline a numerical procedure for
calculating all 36 constitutive parameters as a function of ω and k.
Our method can calculate constitutive parameters for frequencies and wavenumbers both on and off of the
dispersion curve. Up to now, research has only focused on the behavior of metamaterials on the dispersion curve.
However, applications such as metamaterial antennas require the calculation of constitutive parameters both on
and off the dispersion curve. To the best of the authors’ knowledge ours is the first work done calculating the
constitutive parameters of a metamaterial crystal away from the dispersion curve.
Essential to our extraction procedure is the idea of driving a metamaterial crystal with both electric and
magnetic charge and current. We will explain why this is necessary in Sec. 2 and how it allows us to extract
all 36 parameters. In addition to driving the crystal, a field averaging procedure is required. The method of
driving the crystal with electric and magnetic charge is separate from the averaging procedure used provided
the averaging procedure can handle driven fields. We present a new field averaging procedure in Sec. 3 that is
particularly good at providing approximately correct boundary conditions. In Sec. 4 we describe how spatial
dispersion can break the symmetry of a centrosymmetric crystal resulting in, among other things, bianisotropy.
In Sec. 5 we discuss an important ambiguity in the constitutive parameters that effects experimental attempts to
measure them. In Sec. 6 we present a number of tests for the current driven parameters of a metamaterial and
demonstrate these tests on a 2D metamaterial crystal. Finally, in Sec. 7 we study the constitutive parameters
of a 2D crystal for small k and show how this provides insights into spatial dispersion.
Corresponding author: cfietz@physics.utexas.edu

Metamaterials: Fundamentals and Applications II, edited by Mikhail A. Noginov, Nikolay I. Zheludev, Allan D. Boardman,
Nader Engheta, Proc. of SPIE Vol. 7392, 73920L · © 2009 SPIE · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.827031

Proc. of SPIE Vol. 7392 73920L-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


2. CURRENT DRIVEN METAMATERIAL HOMOGENIZATION
The equation we want to solve is
    
D  ξ E
= , (1)
B ζ μ H

where the field vectors D, B, E and H, are the known variables and the matrix elements (the constitutive
parameters) are the unknown variables. The matrix on the RHS of Eq. 1 is a 6x6 matrix known as the constitutive
matrix. By convention, this matrix is separated into 4 3x3 matrices known as , ξ, ζ and μ.7 Since there are
36 unknown matrix elements we need 36 equations of constraint. Eq. 1 provides six equations of constraint,
therefor we need six sets of Eq. 1. Specifically, we need six linearly independent sets of field vectors obeying Eq.
1. This gives us 36 equations of constraint to solve for 36 unknowns.
If we are limited to using only free waves we will have at most two linearly independent sets of fields (two
polarizations) for a particular ω and k, and then only if the two polarizations are degenerate. If they are not
degenerate, a free wave will only give us one set of linearly independent fields for a particular ω and k. We can
obtain three sets of linearly independent fields by driving our crystal with electric charge-current,6 but this is
not enough. A solution to our problem becomes apparent when we inspect this modified form of the Maxwell
equations.

1 ∂D 4π
∇ · D = 4πρ ∇×H− = J
c ∂t c
(2)
1 ∂B 4π
∇ · B = 4πφ −∇ × E − = I
c ∂t c
These are the familiar Maxwell equations with the addition of magnetic charge density φ and magnetic
current density I.8, 9 In a homogeneous medium an electric current J = J0 ei(ωt−k·x) and magnetic current
I = I0 ei(ωt−k·x) that are harmonic in time and space will generate an electromagnetic wave E(t, x) = E0 ei(ωt−k·x)
and H(t, x) = H0 ei(ωt−k·x) according to Eq. 2, which can be rearranged in ω and k space and combined with
the constitutive matrices to give us
    
ω/c −1 (k × +ωξ/c) E0 4πi −1 J0
−1 = . (3)
−μ (k × −ωζ/c) ω/c H0 c μ−1 I0

Here the four terms in the matrix on the left hand side are all 3x3 matrices and k× is a 3x3 matrix representing
the k cross operation. If ω and k are not on the dispersion curve of the system then the matrix on the LHS
of Eq. 3 is always invertible and we can solve for E0 and H0 . From this equation it is obvious that if we limit
ourselves to electric current only, we can obtain at most three linearly independent field vectors, thus, we would
be unable to solve Eq. 1 for the constitutive parameters. However, if we allow ourselves to drive our crystal with
both electric and magnetic current we can obtain six linearly independent field vectors and solve Eq. 1 for all
36 constitutive parameters.
Once one understands how to obtain six linearly independent field vectors, all that is required is to perform
six electromagnetic simulations, each one driving the metamaterial crystal with a different current polarization,
then average the microscopic fields returned by the simulation into macroscopic fields and solve Eq. 1 for the
constitutive parameters. Explicitly, the procedure for extracting the constitutive parameters is as follows: we
define the following 6x6 matrices

     
D(1) , D(2) , ... D(6) E(1) , E(2) , ... E(6) J(1) , J(2) , ... J(6)
D≡ , E≡ , J ≡ . (4)
B(1) , B(2) , ... B(6) H(1) , H(2) , ... H(6) I(1) , I(2) , ... I(6)

Each column of the matrices in Eq. 4 is associated with a single electromagnetic simulation. For example, in the
first simulation our driving current is Ji1 = (1, 0, 0, 0, 0, 0). That is, we drive our crystal with electric current in

Proc. of SPIE Vol. 7392 73920L-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


the x direction according to Eq. 2. We solve for the microscopic electromagnetic fields e, h, d and b using the
commercial finite element software package COMSOL Multiphysics. We then take the microscopic fields solved
for in our first simulation and average them into the macroscopic fields Di1 = (D(1) , B(1) ) and Ei1 = (E(1) , H(1) ).
In our second simulation we have Ji2 = (0, 1, 0, 0, 0, 0). We average the microscopic fields from the second
simulation into the macroscopic fields Di2 = (D(2) , B(2) ) and Ei2 = (E(2) , H(2) ). We follow the same steps for
the next four simulations, each one driving the crystal with a different current polarization so that the matrix
J has a nonzero determinant and as a result D and E also have nonzero determinants. Finally, we solve for our
constitutive parameters:
 
 ξ
C≡ = DE −1 . (5)
ζ μ

Since the six electromagnetic simulations are performed for a particular ω and k, our current driven constitutive
parameters are functions of ω and k, that is C = C(ω, k).
This method of driving our metamaterial crystal with electric and magnetic current is independent of the
averaging procedure used, provided that the averaging procedure works on Bloch waves driven by external electric
and magnetic currents. We have experimented with several averaging procedures we note that the averaging
procedure suggested by Pendry et. al.1 modified with with a correction factor provided by Smith et. al.10 returns
constitutive parameters that predict dispersion very well but do not provide the correct boundary conditions.
We now present a new averaging procedure that predicts dispersion correctly and provides approximately correct
boundary conditions.

3. FIELD AVERAGING
We present our field averaging procedure first in three dimensions and then two dimensions. Similar procedures
will work in one or four dimensions. Our goal is to take the microscopic electromagnetic Bloch wave fields e, h,
d and b solved for in an electromagnetic simulation, and average these microscopic fields into the macroscopic
plane waves E = E0 ei(ωt−k·x) , H = H0 ei(ωt−k·x) , D = D0 ei(ωt−k·x) and B = B0 ei(ωt−k·x) . If we take the
microscopic Maxwell equations that deal with the h and d fields in 3D we have the following equations:

∇ · d = 4πρ (6a)
4π ω
∇×h = J − i d. (6b)
c c

If we integrate these equations over the crystal unit cell volume Ω, the volume integrals of the left hand
sides of Eq. 6 become area integrals over the boundary of the unit cell ∂Ω. Our prescription for averaging the
microscopic fields is to force the boundary integrals over the macroscopic field to equal the boundary integrals
over the microscopic fields.


 3
 3
−ik · D0 SV = Ω d x∇ · D = Ω d x∇ · d = dn · d (7a)
∂Ω
 
−ik × H0 SV = Ω d3 x∇ × H = Ω d3 x∇ × h = dn × h (7b)
∂Ω

Here dn is an infinitesimal integration area normal to the boundary ∂Ω pointing out of the unit cell, k is the
Bloch wavenumber, and k× is a matrix cross product of the Bloch wavenumber
⎛ ⎞
0 −kz ky
k× = ⎝ kz 0 −kx ⎠ . (8)
−ky kx 0

Proc. of SPIE Vol. 7392 73920L-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


SV is an effective volume given by the equation

sin(kx ax /2) sin(ky ay /2) sin(kz az /2)


SV = · · (9)
kx /2 ky /2 kz /2
This effective volume ensures we get the correct macroscopic field if we average a plane wave in an already
homogeneous medium. Eq. 7a only restricts the ”longitudinal” component of D0 (the D field parallel to k∗ ,
where the ∗ indicates complex conjugation) and the ”transverse” components of H0 (the H field perpendicular
to k). The remaining components of D0 and H0 are determined from ”naive” averages over the components
orthogonal those determined in Eq. 7.
Eq. 7a determines the component of D0 parallel to k∗ . The two remaining undetermined components are
those perpendicular to k∗ , which we will call k1 and k2 (k1 · k∗ = k2 · k∗ = k1 · k∗2 = 0). We perform a simple
volume integral over the k1 · d and k2 · d components and determine D0 with the equation
⎛ i ⎞
∂Ω
dn · d
⎜ SV ⎟
⎛ ⎞−1 ⎜ ⎟
kx ky kz ⎜ ⎟
⎜ 1  3 ⎟
D0 ≡ ⎝ k1x k1y k1z ⎠ ⎜ d xk1 · d ⎟
⎜ SV Ω ⎟. (10)
k2x k2y k2z ⎜ ⎟
⎜ ⎟
⎝ 1  3 ⎠
d xk 2 · d
SV Ω
Notice that the first row in the matrix in Eq. 10 gives us Eq. 7a and the bottom two rows relate the two remaining
components of D0 to integrals over the volume of the unit cell Ω. The matrix in Eq. 10 is invertable as long as
k = 0 so solving for D0 is straightforward.
This leaves us to determine H0 . Obviously kt × kt = 0 so Eq. 7b determines two of the three components
of H0 , the exception being the component parallel to kt . Averaging this remaining component with a simple
volume integral we have

⎛ ⎞pinv ⎛ 1  ⎞
kx ky kz d3 xk · h
⎜ 0 ⎜ Ω ⎟
−kz ky ⎟ ⎜
S V

H0 ≡ ⎜ ⎟ ⎜ ⎟. (11)
⎝ kz 0 −kx ⎠ ⎝ i  ⎠
−ky kx 0 dn × h
SV ∂Ω
Again, the bottom three rows of the matrix in Eq. 11 gives us Eq. 7b and the first row determines the remaining
component of H0 by relating it to an integral over the volume of the unit cell. The 4x3 matrix in Eq. 11 has a
rank of 3 and we solve for H0 by multiplying the Moore-Penrose inverse11 (pinv) of the 4x3 matrix by the vector
on the right hand side of Eq. 11. Eq.s 10 and 11 completely determine D0 and H0 respectively.
The procedure for averaging the E and B fields is the same as averaging the H and D fields. When solving
for B0 take equation 10 and replace D0 and d with B0 and b and when solving for E0 take equation 11 and
replace H0 and h with E0 and e.
The examples we present are all two dimensional so we now define the 2D averaged fields which are derived
in a way very similar to Eqs. 10 and 11.
⎛ i ⎞
dn · dt
⎜ SA ∂Ω ⎟
⎛ ⎞−1 ⎜ ⎟
kx ky 0 ⎜ ⎟
⎜ 1  2 ⎟
D0 ≡ ⎝ 0 0 1 ⎠ ⎜ d x dz ⎟. (12)
⎜ SA Ω ⎟
ky∗ −kx∗ 0 ⎜ ⎟
⎜ ⎟
⎝ 1  2 ⎠
d x (ky∗ dx − kx∗ dy )
SA Ω

Proc. of SPIE Vol. 7392 73920L-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


⎛ ⎞
1  2
d x (kx hx + ky hy )
⎛ ⎜
⎞pinv ⎜ A
S Ω ⎟

kx ky 0 ⎜  ⎟
⎜ 0 ⎜ ⎟
0 ky ⎟ ⎜ i ⎟
H0 ≡ ⎜
⎝ 0
⎟ ⎜ dn hz ⎟. (13)
0 −kx ⎠ ⎜ SA ∂Ω ⎟
⎜ ⎟
−ky kx 0 ⎜  ⎟
⎝ i ⎠
dl · ht
SA ∂Ω

sin(kx ax /2) sin(ky ay /2)


SA = · (14)
kx /2 ky /2

Finally, we emphasize that due to the definitions in Eqs. 10,11,12 and 13, we cannot solve for the constitutive
parameters at the Γ point (k = 0).

4. BIANISOTROPY DUE TO SPATIAL DISPERSION


It is widely believed that a centrosymmetric crystal can not be bianisotropic (ξ and ζ must be zero), however, this
is only true if the constitutive parameters are only functions of ω. If the constitutive parameters are functions
of ω and k then spatial dispersion can cause bianisotropy in a centrosymmetric crystal.
Under any spatial reflection or rotation or combination thereof (we are limiting ourselves to transforma-
tion matrices with determinant equal to ±1), the pseudotensor ζ, when dependent only on ω, transforms like
ζ  (ω) = det(T)Tζ(ω)TT where T is the transformation matrix. For a centrosymmetric crystal, the consti-
tutive tensors (pseudotensors) should be unchanged by the inversion Tinv = diag(−1, −1, −1), implying that
ζ(ω) = ζ  (ω) = −Tinv ζ(ω)TT inv = −ζ(ω) = 0. However, if ζ (or ξ) is a function of k as well as ω, then we
have ζ(ω, k) = ζ  (ω, TT
inv k) = −Tinv ζ(ω, TT T
inv k)Tinv = −ζ(ω, −k). Instead of showing that ζ (and ξ) vanish for
a centrosymmetric crystal we instead find the important symmetry relation ζ(ω, k) = −ζ(ω, −k). There are
similar symmetry relations for the other constitutive matrices for a centrosymmetric crystal

(ω, k) = (ω, −k) ξ(ω, k) = −ξ(ω, −k)


(15)
ζ(ω, k) = −ζ(ω, −k) μ(ω, k) = μ(ω, −k)

We see that the k vector has broken the symmetry of the crystal. This can be seen more clearly if we
investigate how the pseudotensors ζ and ξ are restricted under spatial reflections. Consider the case when
k = kx x̂. If we perform a reflection across the x-z plane our transformation matrix is Ty = diag(1, −1, 1). Under

this transformation, for the zy component of ζ we find ζzy (ω, k) = ζzy (ω, TT T
y k) = ζzy (ω, k) since Ty k = k for
k = kx x̂. Thus we see that ζzy can have a nonzero value. The situation is opposite for ζzx as it transforms

like ζzx (ω, k) = ζzx (ω, TT
y k) = −ζzx (ω, k) = 0 for a centrosymmetric crystal. So constitutive tensors that are
functions of k as well as ω are restricted in different ways than constitutive tensors that are only functions of
ω. More generally, for wavenumbers that lie in both the x and y directions, the constitutive parameters for a
centrosymmetric crystal have the nonzero pattern
⎛ ⎞ ⎛ ⎞
xx xy 0 0 0 ξxy
(ω, k) = ⎝ yx yy 0 ⎠ ξ(ω, k) = ⎝ 0 0 ξyz ⎠
0 0 zz ξzx ξzy 0
⎛ ⎞ ⎛ ⎞ (16)
0 0 ζxz μxx μxy 0
ζ(ω, k) = ⎝ 0 0 ζyz ⎠ μ(ω, k) = ⎝ μyx μyy 0 ⎠.
ζzx ζzy 0 0 0 μzz

Finally, we note that in the limit of k = 0, the k vector no longer breaks the symmetry of the centrosymmetric
crystal, and ζ and ξ as well as the off diagonal terms in  and μ become zero.

Proc. of SPIE Vol. 7392 73920L-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


5. EXPERIMENTAL AMBIGUITY OF CONSTITUTIVE PARAMETERS
There are several tests that a good set of constitutive parameters should be able to pass. The simplest is
that constitutive parameters should reproduce the correct dispersion relation for a metamaterial crystal. The
dispersion relation of a homogeneous medium with the constitutive parameters of a metamaterial is calculated
by setting the determinant of the matrix in Eq. 3 equal to zero. Solutions to this dispersion relation should agree
with the dispersion curve calculated from an eigenvalue simulation of the metematerial.12
A second test is that the constitutive parameters should give correct boundary conditions. For example, if you
reflect a plane wave off of a metamaterial crystal, the transmission and reflection at the boundary should be well
approximated by calculations of reflection and transmission at the interface between vacuum and a homogeneous
medium with the constitutive parameters of the metamaterial.
Both of these tests are important, but both are limited by the fact that they rely on free waves that obey
the dispersion relation of the crystal. This has two disadvantages. First, a free wave is not effected by all of
the 36 constitutive parameters, therefore a test relying on free waves cannot truly test all of the constitutive
parameters. Second, we have the ability to extract constitutive parameters on as well as off of the dispersion
curve and it is important to test these parameters both on and off the curve. One way to do this is to simulate
driving a crystal with current J = J0 ei(ωt−k·x) and I = I0 ei(ωt−k·x) with an ω and k both on and off of the
dispersion curve and compare the averaged macroscopic fields to the fields predicted from a Green’s function
calculation for a homogeneous medium with the constitutive parameters of the metamaterial. This test is much
stronger than tests involving only free waves, however, all three tests mentioned are interesting and all three will
be demonstrated on a 2D crystal in Sec. 6.
Examining the left hand side Eq. 3 we see the 3x3 matrix μ−1 (k × −ωζ). Any experiment designed to test the
extracted constitutive parameters will be limited to using only electric current and not magnetic current. Without
magnetic current, Eq. 3 develops an ambiguity. We can transform between the current driven constitutive
parameters predicted by our theory to a second ”effective” set of parameters that obey the following equation.

μ−1 (k × −ωζ/c) = (μef f )−1 (k × −ωζ ef f /c) (17)

Eq. 3 predicts the fields generated by a source, but it is clear that this ambiguity also exists for the dispersion
of free waves. It is easy to see that a transformation in spatial coordinates changes both sides of Eq. 17 equally
so that Eq. 17 is still true after the transformation. Though more work is needed to explore this ambiguity, so
far we have found no experiment that can discriminate between our current driven parameters or any other set
of effective parameters obeying Eq. 17.
This issue at once both illustrates the essential role that using magnetic charge and current in our simulations
plays in restricting the possible values of the constitutive parameters and at the same time illuminates the problem
of the experimental ambiguity of these parameters. If an experiment (using only electric charge and current)
cannot discriminate between different sets of constitutive parameters that obey Eq. 17, then all such sets must be
considered experimentally equivalent. In the next section we’ll see that this ambiguity is important for comparing
our predicted constitutive parameters to those predicted by simpler, more established theories. Specifically, we’ll
see that the parameters predicted by S-Parameter retrieval3 do not agree with our theory unless we perform a
transformation according to Eq. 17.
Finally, we note that if one restricts one’s self to experiments that do not use electric charge and current then
there is a similar experimental ambiguity effecting  and ξ: −1 (k × +ωξ/c) = (ef f )−1 (k × +ωξ ef f /c), as can
be seen from the upper right part of the matrix in eq 3.

6. CONSTITUTIVE PARAMETERS FOR A SPOF


We now show results from a 2D metamaterial crystal known as a Strip Pair-One Film or a SPOF.13 A diagram
of the SPOF is given in Fig. 1. Our SPOF is a square crystal lattice with a thin Au film in the center of the
unit cell and two Au strips on both sides of the film. The permittivity of the Au is  = 1 − ωp2 /(ω(ω − iΓ)) with
ωp = 1.32 · 1016 /s and Γ = 1.2 · 1014 /s. The rest of the SPOF is dielectric with permittivity  = 1.562 .

Proc. of SPIE Vol. 7392 73920L-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


1

Re(kx ) (ax /π)


0.5
0
−0.5
−1
400 500 600 700 800 900 1000
λ0 (nm)
2

Im (kx ) (ax /π)


1
0
−1
−2
400 500 600 700 800 900 1000
λ0 (nm)

Figure 1. Left Side: The unit cell of the SPOF. The two strips and center film of the SPOF are Au with permittivity
 = 1 − ωp2 /(ω(ω − iΓ)) where ωp = 1.32 · 1016 /s and Γ = 1.2 · 1014 /s. The rest of the SPOF is made of dielectric with
permittivity  = 1.562 . Right Side: Real (upper plot) and imaginary (lower plot) parts of kx vs. λ0 where k = kx x̂.
Solid lines are dispersion curves for a p-polarized wave solved by an eigenvalue simulation.12 Dotted lines are dispersion
curves calculated from Eq. 18 using current driven constitutive parameters calculated from Eq. 5 by driving the SPOF at
ω and k = kx (ω)x̂ for two of the p-polarized eigenmodes. Note there is only one radiative mode.

In Fig. 1 we plot two different dispersion curves. First we plot the real and imaginary parts of the complex
wavenumber for p-polarized (electric field in the x-y plane) eigenmodes propagating through the SPOF along the
x-direction calculated from an eigenvalue simulation12 (solid lines). We calculated the full set of current driven
constitutive parameters as functions of ω and k = k(ω) for two of the eigenmodes as described in sections 2
and 3. It should be noted that instead of driving the crystal on the dispersion curve, it is important to drive the
crystal close to but slightly off of the dispersion curve so as to prevent the matrix in Eq. 3 from being singular.
We used these current driven constitutive parameters to calculate the real and imaginary parts of the complex
wavenumber for each eigenmode according to the dispersion relation of the SPOF (dotted lines), which for a
p-polarized wave propagating in the x-direction is
  
kx − ωζzy /c kx + ωξyz /c ω2
− = 0. (18)
μzz yy c2

Fig. 1 clearly shows that the current driven constitutive parmeters predict the correct dispersion for two of the
modes, one of which is radiative for some frequencies and evanescent for others and the other mode being always
evanescent. Though not shown in Fig. 1, the current driven constitutive parameters fail to predict the dispersion
of the third mode (red line).
We emphasize that testing the current driven constitutive parameters to see if they produce the correct
dispersion relation is an important physical test but does prove the accuracy of the parameters. Different
averaging procedures can pass this test while predicting slightly different constitutive parameters. For example,
as mentioned in Sec. 2, the averaging procedure in Smith10 passes the dispersion relation test but fails the
boundary condition test. This brings us to our next test of our current driven constitutive parameters.
In Fig. 2 we plot the yy , ζzy and μzz components of the SPOF extracted along the dispersion curve of the
”radiative” (blue line) p-polarized eigenmode in Fig. 1. Note that since the dispersion curve in Fig. 1 is for a
p-polarized wave propagating in the x direction and given the nonzero structure of the constitutive matrices for
propagation in the x-direction in a centrosymmetric crystal, only four of the total 36 constitutive parameters,
yy , ζzy , μzz and ξyz effect the p-polarized wave. Also, though we do not show ξyz in Fig. 2, the current driven

Proc. of SPIE Vol. 7392 73920L-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


20 3
Re(yy ) Re(ζzy /(kx ax ))
15 Im(yy ) Im(ζzy /(kx ax ))
2 Re(μzz )
10 Im(μzz )
1
5

0 0

−5
−1
−10
−2
−15

−20 −3
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
λ0 (nm) λ0 (nm)

Figure 2. Real and imaginary parts of the current driven parameters yy (left plot) and ζzy and μzz (right plot) calculated
on the dispersion curve k = kx (ω)x̂ shown in Fig. 1 according to Eq. 5. Notice that μzz is unity for all wavelengths. Also,
all components of ξ (not shown) are uniformly zero.

values of ξ turn out to be uniformly zero. We believe this is due to all inclusions of the SPOF having only an
electric response combined with the fact the SPOF is centrosymmetric.
Next to the the previously unexpected bianisotropy, the most interesting thing about Fig. 2 is that μzz is unity
for all wavelengths. This disagrees with S-parameter retrieval3 of  and μ for the SPOF structure. In fact the
unity of μzz and the unexpected bianisotropy of nonzero ζzy are connected and together with the experimental
ambiguity expressed in Eq. 17 explain the discrepancy with S-parameter retrieval. For a pseudotensor ζ(ω, k =
kx x) describing a centrosymmetric crystal, the p-polarized part of Eq. 17 can be rewritten as (kx −ωζzy /c)/μzz =
ef f
(kx − ωζzy /c)/μef f
zz . There is a second part of Eq. 17 that involves ζyz but this only effects s-polarized waves so
we ignore it here. Since we have two unknown effective parameters μef f ef f
zz and ζzy and one equation of constraint,
we have one degree of freedom in our effective parameters. Essential to S-parameter retrieval is the assumption
ef f
that ζ and ξ are zero, so we use our one degree of freedom to set ζzy = 0 giving us

μzz
μef f
zz = . (19)
ωζzy
1−
ckx

Fig. 3 shows a comparison of the current driven yy calculated from Eq. 5 and effective μef f
zz calculated from
Eq. 19 using current driven ζzy and μzz vs. yy and μzz determined with S-parameter retrieval3 of a SPOF slab
S S

five unit cells thick. We see a good agreement between the current driven yy and the S-parameter retrieved
Syy and we also see a good agreement between effective μef f S
zz and the S-parameter retrieved μzz . S-parameter
retrieval calculates  and μ from the the index of refraction n and impedance z of the matematerial where n and z
are inferred from reflection and transmission amplitudes through a metamaterial slab. The index is proportional
to the wavenumber, and as we’ve seen in Fig. 1 our current driven parameters predict the correct wavenumber
very well. The impedance is related to the boundary conditions at the interface between the slab and vacuum.
This is where the error in Fig. 3 comes from. It is the Maxwell boundary conditions (continuity of Etan , Htan ,
Dnorm and Bnorm ) that are less than perfectly described by the current driven parameters.
We see this in our next test shown in Fig. 4, which is a test of the current driven constitutive parameters to
provide the correct boundary conditions. The solid lines in Fig. 4 are reflection amplitudes (r = Hrefl inc
z /Hz ) of
p-polarized plane waves reflecting off of a SPOF with infinite thickness and the dotted lines are the reflection
amplitudes of p-polarized plane waves reflecting off of a homogeneous medium with the current driven parameters

Proc. of SPIE Vol. 7392 73920L-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


20 4
Re(yy )
15 Im(yy ) 3
Re(Syy )
10 Im(Syy ) 2

5 1

0 0

−5 −1

−10 −2 Re(μeff
zz )
Im(μeff
zz )
−15 −3 Re(μSzz )
Im(μSzz )
−20 −4
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
λ0 (nm) λ0 (nm)

Figure 3. Left Side: Real and imaginary parts of current driven parameters yy (solid lines) calculated with Eq. 5 and S
yy
(dotted lines) calculated with S-parameter retrieval3 for the SPOF. Right Side: Real and imaginary parts of μef f
zz (solid
lines) calculated from Eq. 19 with current driven ζzy and μzz calculated with Eq. 5 and μS
zz (dotted lines) calculated with
S-parameter retrieval for the SPOF. Note that between λ0 = 680nm and λ0 = 695nm both yy and μef f
zz are negative for
this mode giving us a negative index of refraction.

1 1
0◦ 45 ◦ 60◦ 75 ◦

0.5 0.5
Im(r)
Re(r)

0 0

−0.5 −0.5

−1 −1
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
λ0 (nm) λ0 (nm)

Figure 4. Real (left plot) and imaginary (right plot) parts of the reflection amplitude (r = Href inc
z /Hz ) of a plane wave with
incident angles 0◦ , 45◦ , 60◦ and 75◦ off of a SPOF (solid lines) and off of a homogeneous medium with the current driven
constitutive parameters of the SPOF (dotted line).

of the SPOF. Each point in Fig. 4 has a particular free-space wavelength λ0 and incidence angle θ. There is a
corresponding kx determined from an eigenvalue simulation12 and ky = 2π/λ0 sin(θ). We calculate the current
driven constitutive parameters for ω = 2π/λ0 and k = kx x̂ + ky ŷ and use them to calculate the reflection off of
a homogeneous medium.

Hrefl
z cos(θ) − zyz
r= = . (20)
Hinc
z cos(θ) + zyz

Proc. of SPIE Vol. 7392 73920L-9

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


1 π

π
0.75
|Hz | (arbitrary units)

arg(Hz )
λ0 = 600nm
0.5 0 λ0 = 800nm
λ0 = 1000nm

0.25 - π2

0 -π
M Γ X M Γ X

Figure 5. Absolute value (left plot) and phase (right plot) of Hz generated by driving current J = J0 ei(ωt−k·x) =
(−ky x + kx y)/|k|ei(ωt−k·x) for three frequencies: λ0 = 600nm, 800nm and 1000nm. We plot Hz0 averaged according to
Eq. 13 from h generated in the SPOF (solid line) and the Hz0 generated in a homogeneous medium with the current
driven constitutive parameters of the SPOF (dotted line). Note the resonances for the λ0 = 600nm and λ0 = 1000nm
lines correspond to free modes of the SPOF.

Here zyz = Etrans


y /Htrans
z and the transmitted fields Etrans y and Htrans z are given by the null vector of the matrix
in Eq. 3.
In Fig. 4 there are two bands 500nm < λ0 < 600nm and λ0 > 800nm where we see very good agreement
between the actual reflection off of the SPOF (solid line) and the reflection off of a homogeneous medium with
the current driven constitutive parameters (dotted line) of the SPOF. There is less agreement near the magnetic
resonance around λ0 ≈ 680nm and also near the electric resonance around λ0 ≈ 500nm. Curiously, there is quite
good agreement near the electric resonance near λ0 ≈ 900nm.
The final test is the Green’s function test. We will drive two different mediums with the same driving current
J = J0 ei(ωt−k·x) . One medium will be the SPOF, the other will be a homogeneous medium with the current
driven constitutive parameters of the SPOF. Since we want to do a test that could in principle be done in a real
experiment, we drive the crystal only with electric current. We are specifically interested in probing p-polarized
waves and so we drive our two mediums with electric current lying in the x-y plane and perpendicular to the
wavenumber J0 = (−ky x + kx y)/|k|. Solving Eq. 3 for p-polarized waves driven with only electric current gives
us

ky + ωζzx /c yy Jx − xy Jy kx − ωζzy /c yx Jx − xx Jy



4πi μzz xx yy − xy yx μzz xx yy − xy yx
Hz0 =− . (21)
c ω2 ky + ωζzx /c yy ky − xy kx kx − ωζzy /c xx kx − yx ky
− −
c2 μzz xx yy − xy yx μzz xx yy − xy yx

Fig. 5 shows the results. We have driven the SPOF and homogeneous medium at three different frequencies:
λ0 = 600nm, 800nm and 1000nm, and we varied k from the M point (k = π/ax x̂+π/ay ŷ) to the Γ point (k = 0) to
the X point (k = π/ax x̂). At each frequency and wavenumber we calculate the Hz field generated by the driving
current in the SPOF (solid line) and the homogeneous medium (dotted line). We see an excellent agreement
between the magnetic field generated in the SPOF and the homogeneous medium. Notice the resonances for the
λ0 = 600nm and λ0 = 1000nm lines. These occur when we drive the crystal near it’s dispersion relation as can
be seen by comparing the Γ − X branch in Fig. 5 to the dispersion curve in Fig. 1. The λ0 = 800nm line never
resonates because its free mode is too lossy and the SPOF is driven with a real valued k.

Proc. of SPIE Vol. 7392 73920L-10

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


7. SQUARE ARRAY
We now demonstrate our current driven extraction procedure on another 2D crystal shown in Fig. 6. It is a 2D
array of squares with length a/2 where a is the lattice constant. The squares have permittivity  = 200 − 5i
surrounded by a material with permittivity  = −1.14 Rather than test the current driven constitutive parameters
as we did for the SPOF, we will study the values of the current driven parameters for small k to learn how spatial
dispersion affects the constitutive parameters.

Re(kx ) (a/π)
0.5
0
−0.5
−1
5 7.5 10 12.5 15
λ0 (a)
1
Im (kx ) (a/π)

0.5
0
−0.5
−1
5 7.5 10 12.5 15
λ0 (a)

Figure 6. Left Side: The unit cell of the square array.14 The inner square has permittivity  = 200 − 5i and the
surrounding area has permittivity  = −1. Right Side: Real (upper plot) and imaginary (lower plot) parts of kx vs.
λ0 where k = kx x̂. Solid lines are dispersion curves for a p-polarized wave solved by an eigenvalue simulation.12 The
dotted line is a dispersion curve calculated from Eq. 18 using current driven constitutive parameters extracted along the
”radiative” (solid blue line) dispersion curve.

Using symmetry arguments similar to those in Sec. 4 we see that the components xx , and yy must be even
functions of kx and ky but components xy and yx must be odd functions of kx and ky . Likewise, ζzx and ζxz
must be even functions of kx but odd functions of ky and ζzy and ζyz must be odd functions of kx and even
functions of ky . Also, the square array is symmetric under a rotation of 90◦ and for  this symmetry manifests
as the symmetry relation (ω, k) = T90◦ (ω, TT T
90◦ k)T90◦ where T90◦ is a transformation matrix describing a 90

rotation of the x-y plane. There are similar rotational symmetry relations for ξ, ζ and μ.
In Fig. 7 we plot the current driven yy (ω, k = k(ω)) calculated from Eq. 5 and effective μef f
zz (ω, k = k(ω))
calculated from Eq. 19 along the dispersion curve for the ”radiative” mode in the square array (blue line Fig. 6).
We also plot the current driven yy (ω, k  0) and μef f
zz (ω, k  0) calculated for k·a  1. We see an antiresonance
in yy (ω, k = k(ω)) that does not appear in yy (ω, k  0) indicating the antiresonance is due to spatial dispersion
as has been suggested by Koschny et al.15 We also see that μef f
zz (ω, k = k(ω)) and μef f (ω, k  0) agree with
each other quite well suggesting that spatial dispersion does not strongly affect the magnetic resonance.

8. CONCLUSION
In conclusion we note that our theory is the only one that can numerically calculate all 36 linear electromagnetic
constitutive parmaters for a metamaterial crystal. Our current driven constitutive parameters show bianisotropy
due to spatial dispersion which has never before been predicted. The theory appears highly accurate at predicting
the dispersion of free waves in a metamaterial crystal. It is fairly accurate at providing the correct boundary
conditions at an interface between a metamaterial and vacuum. Our method can also calculate constitutive
parameters away from the dispersion curve which is useful characterizing a metamaterial for small k allowing
us to study spatial dispersion and for solving Green’s function problems. It is highly accurate at predicting the
fields generated by a driven current, a numerical test that has never before been performed on a metamaterial.

Proc. of SPIE Vol. 7392 73920L-11

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx


2 10
Re[yy (ω, k = k(ω))] Re[μeff
zz (ω, k = k(ω))]
Im[yy (ω, k = k(ω))] 7.5 Im[μeff
zz (ω, k = k(ω))]
Re[yy (ω, k = 0)] Re[μeff
zz (ω, k = 0)]
1 Im[yy (ω, k = 0)] 5 Im[μeff
zz (ω, k = 0)]

2.5

Imag
Real

0 0

−2.5

−1 −5

−7.5

−2 −10
5 7.5 10 12.5 15 5 7.5 10 12.5 15
λ0 (a) λ0 (a)

Figure 7. Left Side: Real and imaginary parts of the current driven yy (ω, k = k(ω)) (solid lines) extracted from the
square array along the dispersion curve of the ”radiative” mode (blue line Fig. 6) and current driven yy (ω, k  0)
(dashed lines) calculated for k · a  1. Note that the anti-resonance seen in yy (ω, k = k(ω)) is not seen in yy (ω, k  0).
Right Side: Real and imaginary parts of current driven μef f
zz (ω, k = k(ω)) (solid line) extracted from the square array
along the dispersion curve of the ”radiative” mode and current driven μef f (ω, k  0) (dashed line) calculated for k·a  1.

REFERENCES
[1] Pendry, J. B., Holden, A. J., Robbins, D. J., and Stewart, W. J., “Magnetism from conductors and enhanced
nonlinear phenomena,” IEEE Trans. Microwave Theory Tech. 47, 2075–2084 (1999).
[2] Marques, R., Medina, F., and Rafii-El-Idrissi, R., “Role of bianisotropy in negative permeability and left-
handed metamaterials,” Phys. Rev. B 65, 144440 (2002).
[3] Smith, D. R., Schultz, S., Markos, P., and Soukoulis, C. M., “Determination of effective permittivity and
permeability of metamaterials from reflection and transmission coefficients,” Phys. Rev. B 65, 195104 (2002).
[4] Smith, D. R. and Pendry, J. B., “Homogenization of metamaterials by field averaging,” J. Opt. Soc. Am.
B 23, 391–403 (2006).
[5] Lui, R., Cui, T. J., Huang, D., Zhao, B., and Smith, D. R., “Description and explanation of electromagnetic
behaviors in artificial metamaterials based on effective medium theory,” Phys. Rev. E 76, 026606 (2007).
[6] Silveirinha, M. G., “Metamaterial homogenization approach with application to the characterization of
microstructured composites with negative parameters,” Phys. Rev. B 75, 115104 (2007).
[7] Kong, J. A., [Electromagnetic Wave Theory], John Wiley and Sons, Inc (1986).
[8] Jackson, J. D., [Classical Electrodynamics], 273–274, John Wiley and Sons, Inc, Hoboken, 3rd ed. (1998).
[9] Moulin, F., “Magnetic monopoles and lorentz force,” Il Nuovo Cimento B 116B, 869–877 (2001).
[10] Smith, D. R., Vier, D. C., Kroll, N., and Schultz, S., “Direct calculation of permeability and permittivity
for a left-handed metamaterial,” Appl. Phys. Lett. 77, 2246–2248 (2000).
[11] Ben-Israel, A. and Greville, T., [Generalized Inverses], Springer, New York, 2nd ed. (2003).
[12] Davanco, M., Urzhumov, Y., and Shvets, G., “The complex bloch bands of a 2d plasmonic crystal displaying
isotropic negative refraction,” Optics Express 15, 9681–9691 (2007).
[13] Lomakin, V., Fainman, Y., Urzhumov, Y., and Shvets, G., “Doubly negative metamaterials in the near
infrared and visible regimes based on thin film nanocomposites,” Optics Express 14, 11164–11177 (2006).
[14] Felbacq, D. and Bouchitte, G., “Left-handed media and homogenization of photonic crystals,” Optics Let-
ters 30, 1189–1191 (2005).
[15] Koschny, T., Markos, P., Smith, D. R., and Soukoulis, C. M., “Resonant and antiresonant frequency depen-
dence of the effective parameters of metamaterials,” Phys. Rev. E 68, 065602(R) (2003).

Proc. of SPIE Vol. 7392 73920L-12

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

You might also like