PREPRINT Page 1 2/13/2002

Determination of Negative Permittivity and Permeability of Metamaterials from Reflection

and Transmission Coefficients
D. R. Smith, S. Schultz
Department of Physics, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0319

P. Markoš*, C. M. Soukoulis
Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011
(Submitted 26 November 2001)

We analyze the reflection and transmission coefficients calculated from transfer matrix
simulations on finite lengths of electromagnetic metamaterials, to determine the effective
permittivity (ε) and permeability (µ). We perform this analysis on structures composed of
periodic arrangements of wires, split ring resonators (SRRs) and both wires and SRRs. We find
the recovered frequency-dependent ε and µ are entirely consistent with analytic expressions
predicted by effective medium arguments. Of particular relevance are that a wire medium
exhibits a frequency region in which the real part of ε is negative, and SRRs produce a
frequency region in which the real part of µ is negative. In the combination structure, at
frequencies where both the recovered real parts of ε and µ are simultaneously negative, the real
part of the index-of-refraction is found also to be unambiguously negative.

It has been proposed that electromagnetic specify the scattering in terms of z and n, it is often
metamaterials—composite structured materials, formed more convenient to choose a second set of analytic
from either periodic or random arrays of scattering variables that carry a direct material interpretation.
elements—should respond to electromagnetic radiation These variables are the electric permittivity ε=n/z, and
as continuous materials, at least in the long wavelength the magnetic permeability µ=nz. Both n and z, and
limit [1,2]. In recent experiments and simulations [3,4], therefore ε and µ, are frequency-dependent complex
it has been demonstrated that certain metamaterial functions that satisfy certain requirements based on
configurations exhibit scattering behavior consistent causality. For passive materials, Re(z) and Im(n) must
with the assumption of approximate frequency- be greater than zero.
dependent forms for ε and µ. However, the techniques At a given frequency, any material that supports
applied in those studies probed the materials indirectly, only one propagating mode will generally exhibit a
and did not provide an explicit measurement that would well-defined refractive index n, whether the material is
assign values for ε and µ. It is our aim in this paper to continuous or not. Photonic band gap materials, for
show that the previous conjectures were indeed valid: example, are characterized by dispersion curves from
unambiguous values for ε and µ can be applied to which an effective index can extracted, even for bands
electromagnetic metamaterials. Our approach here well above the first band gap, and well-defined
utilizes the transmission and reflection coefficients (S- refraction phenomena can result [7,8,9]. However, it is
parameters, equivalently) calculated for a wave generally not possible to assign an impedance (z) to a
normally incident on a finite slab of metamaterial. We non-continuous material, except in those cases where
invert the scattering data to determine n and z, for the wavelength in the material is much larger than the
systems of several thicknesses, from which we obtain dimensions and spacing of the constituent scattering
self-consistent values for ε and µ. While we utilize elements that compose the medium. If z depends
simulation data in this study, the technique we describe strongly on the surface termination, or if z depends on
will be readily applicable to the experimental the overall size of the sample, then z is ambiguous, and
characterization of metamaterial samples whenever the it is not possible to assign intrinsic values for ε and µ.
scattering parameters are known. The transmission coefficient for waves incident
The common method of characterizing the normally to the face of a 1-D slab of continuous
electromagnetic scattering properties of a homogeneous material (in vacuum) with length d is related to n and z
material is to identify its impedance (z) and refractive by
index (n) [5,6]. While it is possible to completely

*
Permanent address: Institute of Physics, Slovak Academy of Sciences, Dúbravska cesta 9, 842 28 Bratislava, Slovakia.

1
PREPRINT Page 2 2/13/2002

 i  1  ikd solution for Im(n). Re(n), however, is complicated by

−1
t =  cos ( nkd ) −  z +  sin ( nkd )  e , (1) the branches of the arccosine function, or,
 2 z 

where k=ω/c is the wavenumber of the incident wave.

Re ( n ) = ± Re 
( 2t ′ ( ))
 cos−1 1 1− r 2 −t ′2  
  + 2π m , (6)
The incident wave is assumed to travel rightward along  kd  kd
the positive x-axis, with the origin defined as the first  
face of the material seen by the radiation. To improve
the clarity of the subsequent formulas, we introduce the where m is an integer. When d is large, these branches
normalized transmission coefficient t’=exp(ikd)t. The can lie arbitrarily close to one another, making the
reflection coefficient is also related to z and n by selection of the correct branch difficult in the case of
dispersive materials. For this reason best results are
r 1  1 obtained for the smallest possible thickness of sample,
= − i  z −  sin ( nkd ) (2) as has commonly been known in the analysis of
t′ 2  z
continuous materials. Even with a small sample, more
than one thickness must be measured to identify the
Eqs. (1) and (2) can be inverted to find n and z as correct branch(es) of the solution which yields
functions of t’ and r. Performing this inversion leads to consistently the same values for n. Note that the
the following expressions: requirement that Im(n)>0 uniquely identifies the sign of
Re(n), which is essential when the material may
cos( nkd ) =
1
2t ′ 
(
1 − r 2 − t ′2 ) potentially have regions which are left-handed—that is,
materials in which Re(n) may be negative.
(3) In order to demonstrate the validity of assigning
1
= Re   −
1
( A r + A2t ′)
 t′  2 t′ 2 1
bulk permeability and permittivity values to non
continuous metamaterials, we utilize simulation data
generated by the transfer matrix method (TMM) [10,11]
on three types of structures: a medium composed of a
and periodic array of wires, a medium composed of a
periodic array of split ring resonators (SRRs), and a
(1+ r )2 −t ′2 medium composed by interleaving the wire and the
z=± , (4) SRR arrays. We simulate one unit cell of a given
(1−r )2 −t ′2 system using periodic boundary conditions in the
directions perpendicular to the propagation direction.
The TMM represents a complete solution of Maxwell’s
where A1 = r * t ′ + t ′ * r and A2 = 1 − r − t ′ are both
2 2

equations, and differs from Eqs. (1) and (2) in that the
real-valued functions that go to zero in the absence of exact details of the scattering elements are taken into
material losses. account. In the embedding medium, all eigenvalues of
Note that while the expressions for n and z are the TM can be found analytically [12]. We find that in
relatively uncomplicated, they are complex functions the frequency range of interest and for the incoming
with multiple branches, the interpretation of which can wave perpendicular to the sample surface, there is only
lead to ambiguities in determining the final expressions one propagating mode. All other modes are evanescent,
for ε and µ. We can resolve these ambiguities by so that the systems considered are effectively one-
making use of additional knowledge about the material. dimensional and we are justified utilizing Eqs. (5) and
For example, if the material is passive, the requirement (6), in which t’ and r represent the transmission and
that Re(z)>0 fixes the choice of sign in Eq. 4. reflection for this propagating mode. In the calculations
Likewise, Im(n)>0 leads to an unambiguous result for presented here, cells were meshed with 15x11x15 grid
Im(n): points, in a manner described in Ref. 11. In the
formulation of the TMM used here, the metal dielectric

Im ( n ) = ± Im 
(
2t ′  (
 cos−1 1 1− r 2 −t ′2  
  )) (5)
function is approximated by ε=-1000 + 18,000i, for all
frequencies. A larger value of the imaginary part of the
 kd  dielectric function for the metal results in lowered
  absorption, but does not change the qualitative picture
presented here.
When we solve for the right-hand side of Eq. 5, we A square array of conducting wires, based on
select whichever of the two roots yields a positive effective medium arguments, is expected to exhibit the

2
PREPRINT Page 3 2/13/2002

ideal frequency-dependent plasmonic form of ε=1- A second ambiguity that arises is the effective unit cell
ωco2/ω2, where ωco is a cutoff frequency [1]. Fig. 1 length that will result in values of ε and µ that do not
shows the permittivity recovered from simulation data depend on the overall length of the structure.
for an array of wires, continuous along the E-polarized Empirically, we find in the three cases investigated that
direction of the incident wave, and with lattice constant the correct length corresponds to the number of unit
5 mm. The figure shows data for one, three, and five cells multiplied by the length of each unit cell. This can
layers of lattice in the propagation direction. A fit to the be seen in Fig. 1, where the analysis of different lengths
resulting curves indicates a cutoff frequency of of the wire medium results in the same permittivity;
ωp/2π=22.3 GHz, consistent with that determined by the only for a wire medium consisting of one layer is there
noting the onset of transmission (not shown), with a a noticeable deviation from the “bulk” behavior. It
frequency dependence indistinguishable from the three should be noted that the unit cells analyzed here are
and five row curves. The recovered permeability was square, and thus the unit cell length is implicit in the
found to be unity at all frequencies. calculation by virtue of the spacing between elements in
While there is no ambiguity in determining the the direction perpendicular to propagation. The fact
phase of the transmitted wave, the phase of the reflected that even one unit cell in the propagation direction
wave must be measured relative to a reference plane produces nearly the same scattering parameters as a
that is considered to be at the first face of the material. thicker material may therefore not be so unexpected.
Given that metamaterials generally do not have well- Fig. 2 shows the real and imaginary parts of the
defined surfaces, the question arises as to where this permeability for a medium composed of only SRRs, as
reference plane should be located. A definite answer configured in Ref. [11]. The dark solid curves are taken
can be obtained in the case of the wire structure, where from the simulation data, while the dashed curves are
it is reasonable to assume that the reference plane from transmission and reflection coefficients computed
coincides with the first plane of the wires. We found for a uniform material with a permeability
that with this choice of reference plane, the relation corresponding to
z(ω)=1/n(ω) holds, consistent with the response of the
wire medium being entirely electric (i.e., µ=1). For the ω mp
2

SRR medium, the choice of reference plane was not µ (ω ) = 1 − 2 (7)

found to be critical; for the combined structure, ω − ω 02 + iΓω
however, the wires dominate the scattering, and the
reference plane was again chosen to coincide with the where we have used ω0/2π=8.5 GHz, ωmp/2π=3.3 GHz
first plane of wires.

Figure 2: Permeability as a function of frequency for the

Figure 1: Permittivity as a function of frequency for a wire SRR medium, for one or more unit cell lengths. The solid
medium. The cutoff frequency for the structure, dark lines correspond to the real and imaginary parts of the
determined by the onset of propagation is 22.3 GHz. The permeability determined using the TMM data, while the
inset depicts the orientation of the wire with respect to the dashed lines were obtained from Eq. (7) by assuming
incident radiation. parameters chosen to approximate the SRR medium. Note
that at ~8.5 GHz the real part of the permeability goes
negative, with a bandwidth of ~0.5 GHz.

3
PREPRINT Page 4 2/13/2002

7 especially if the elements are resonant. In Fig. 3, we

plot the recovered frequency-dependent permittivity
6 corresponding to the SRR medium, which is significant
over the frequency range covered.
5 The result of combining a wire medium with an
SRR medium is shown in Fig. 4, in which we compare
Permittivity (ε)

4 the real parts of the recovered permittivity and

permeability from the combined structure with that
3 from either the wire or the SRR medium alone. The
solid curves are from the combined structure, while the
2 dashed curves are from either medium alone.
Consistent with what is observed in other calculations,
1 as well as experiments, the addition of the wires to the
SRR medium does not impact the permeability
0 properties of the SRR medium. However, the
2 4 6 8 10 12 14 combination of the SRRs and wires results in a
Frequency (GHz) permittivity that is less negative than that for wires
alone.
Figure 3: Real part of the Recovered permittivity as a
function of frequency for the SRR medium. It was first theoretically suggested by Veselago
[13] that an incoming wave incident on an interface
between vacuum and material with simultaneously
and Γ=2.0 GHz. The recovered form for the
permeability is in agreement with the ideal resonant negative Re(ε) and Re(µ) would be refracted to the
form shown in Eq. 7, indicating a region of negative same side of the normal to the interface. By
permeability over the range from ~8.5 GHz to 9 GHz. considering the analytic properties of the refractive
This is consistent with the region of attenuated index in such a medium, it was shown that the refractive
transmission found from a plot of the transmittance index n must, in fact, be negative wherever ε and µ are
versus frequency (not shown). both negative [14]. In a recent experiment [4], a
An array of metal scattering elements can be metamaterial formed by combining a wire medium with
expected to have a net polarization in response to an a SRR medium, qualitatively similar to those simulated
applied electric field, and it might be further anticipated here, was used to demonstrate negative refraction, and
that the effective permittivity will be dispersive, the angle of refraction was utilized to recover n as a
4 function of frequency. While a refraction experiment is
16 a useful method to measure the index, we can also
obtain this information directly and unambiguously
2 12
using Eq. 3, as shown in Fig. 5. If one has access to
8 phase information in the transmission and reflection
Permeability (µ)

coefficients, one can identify both the magnitude and

Permittivity (ε)

0
4
sign of the refractive index from direct incidence
0 measurements on a planar slab.
-2
We have demonstrated that the traditional
-4
procedure of obtaining material parameters from
-4 -8 transmission/reflection data can be successfully applied
to metamaterials, subject to the possible ambiguities of
-12
-6
the definition of the first surface of the measured
4 6 8 10 12 14 sample, and the sample length. The former ambiguity
Frequency
e can be resolved in certain cases by using additional
information about the material, as shown in the case for
Figure 4: Recovered Re(µ) (top, left-hand scale) and Re(ε) the wire medium above. In cases for which it is not
(bottom, solid curve, right-hand scale) as a function of
possible to determine the correct position of the
frequency for the combination SRR/wire medium. The flat
dashed lines indicate the zero values for the two axes scattering surface, the range of possible solutions
(offset for clarity). Note that while the addition of the SRR corresponding to the various possible surfaces
medium to the wire medium results in a more positive throughout the unit cell will place bounds on the
permittivity (solid line) than wires alone (dashed curve), possible values of the effective material parameters.
nevertheless there is a frequency region (~8.5-9.0 GHz) These bounds will become narrower as the unit cell
where both ε and µ are negative. becomes smaller in comparison with the wavelength.
The correct length of a unit cell can be determined

4
PREPRINT Page 5 2/13/2002

5 2. J. B. Pendry, A. J. Holden, D. J. Robbins, W. J.

Stewart, IEEE Trans. Microwave Theory Tech., 47,
4
2075 (1999).
3
3. D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat-
Nasser, S. Schultz, Phys. Rev. Lett., 84, 4184
2 (2000).
4. R. A. Shelby, D. R. Smith, S. Schultz, Science, 292,
Re(n)

1 79 (2001).
5. W. B. Weir, Proc. IEEE, 62, 33 (1974).
0
6. A. M. Nicolson and G. F. Ross, IEEE Trans. Instr.
-1 and Meas., IM-19, 377 (1970).
7. M. Notomi, Phys. Rev. B 62, 10696 (2000).
-2 8. B. Gralak, S. Enoch, G. Tayeb, J. Opt. Soc. Am. A
17, 1012 (2000).
-3
0 2 4 6 8 10 12 14 9. H. Kosaka, T. Kawashima, A. Tomita, M. Notomi,
T. Tamamura, T. Sato, S. Kawakami, J. Lightwave
Frequency
Tech., 17, 2032 (1999).
Figure 5: Refractive index (n) as a function of frequency 10. J. B. Pendry and A. MacKinnon, Phys. Rev. Lett.,
for the combination SRR/wire medium. The region where 69, 2772 (1992).
the index is negative corresponds to the frequency region 11. P. Markoš and C. M. Soukoulis, Phys. Rev. B. 65,
where both ε and µ are simultaneously negative. 033401 (2002); Phys. Rev. E, to be published.
empirically, by requiring a solution that is stable for 12. J. B. Pendry and P. M. Bell, in Photonic Band Gap
several analyzed lengths. Materials, Ed. C. M. Soukoulis (Kluwer, Dodrecht,
As a final point, we note that this analysis depends 1996).
entirely on our ability to measure structures for which 13. V. G. Veselago, Sov. Phys. Usp. 10, 509 (1968).
the electric polarization P depends only on the electric 14. D. R. Smith and N. Kroll, Phys. Rev. Lett., 85, 3966
field E, and the magnetization M depends only on the (2000).
magnetic field H. For a general scattering element, this 15. F. Mariotte, P. Pelet, N. Engheta, Prog. In
does not necessarily hold, and we may have, for Electromagnetics Res., 9, 311 (1994).
example, a constitutive relation such as
H=γE+B/µ [15]. A medium that responds in such a
manner will exhibit chirality, and a rotation of the
polarization. Indeed, in these calculations we do find
some degree of polarization conversion in the data for
the SRRs and the combined SRR/wire medium. This
evidence of chiral behavior suggests the analysis
presented here provides a good, but not exact,
characterization of the metamaterial. Further
improvements to the technique should allow a full
characterization to be implemented.

ACKNOWLEDGEMENTS:
We are grateful for helpful discussions with Art Vetter
(Boeing, PhantomWorks), and John Pendry (Imperial
College). This work was supported by DARPA,
through grants from AFOSR (Contract No. KG3523)
and ARO (Contract No. DAAD19-00-1-0525). CS also
thanks NATO and Ames Lab. (DOE Contract No. W-
7405-Eng-82) for support, and PM thanks the Slovak
Grant Agency for partial financial support.

REFERENCES:
1. J. B. Pendry, A. J. Holden, W. J. Stewart, I.
Youngs, Phys. Rev. Lett., 76, 4773 (1996).