Perfect Control of Reflection and Refraction Using Spatially

Dispersive Metasurfaces

V. S. Asadchy1,2 , M. Albooyeh1 , S. N. Tcvetkova1 ,

A. Dı́az-Rubio1 , Y. Ra’di1,3 , and S. A. Tretyakov1
arXiv:1605.02044v3 [physics.optics] 22 Aug 2016

1
Department of Radio Science and Engineering,
Aalto University, P. O. Box 13000, FI-00076 Aalto, Finland
2
Department of General Physics, Francisk Skorina
Gomel State University, 246019 Gomel, Belarus
3
Department of Electrical Engineering and Computer Science,
University of Michigan, Ann Arbor, Michigan 48109-2122, USA

Abstract
Non-uniform metasurfaces (electrically thin composite layers) can be used for shaping refracted
and reflected electromagnetic waves. However, known design approaches based on the generalized
refraction and reflection laws do not allow realization of perfectly performing devices: there are
always some parasitic reflections into undesired directions. In this paper we introduce and discuss
a general approach to the synthesis of metasurfaces for full control of transmitted and reflected
plane waves and show that perfect performance can be realized. The method is based on the use of
an equivalent impedance matrix model which connects the tangential field components at the two
sides on the metasurface. With this approach we are able to understand what physical properties
of the metasurface are needed in order to perfectly realize the desired response. Furthermore, we
determine the required polarizabilities of the metasurface unit cells and discuss suitable cell struc-
tures. It appears that only spatially dispersive metasurfaces allow realization of perfect refraction
and reflection of incident plane waves into arbitrary directions. In particular, ideal refraction
is possible only if the metasurface is bianisotropic (weak spatial dispersion), and ideal reflection
without polarization transformation requires spatial dispersion with a specific, strongly non-local
response to the fields.

1
I. INTRODUCTION

A metasurface is a composite material layer, designed and optimized in order to control

and transform electromagnetic fields. The layer thickness is negligible as compared to the
wavelength in the surrounding space. Conventional devices for wave transformations are
either bulky and heavy (e.g., reflector antennas or lenses) or complicated and require ac-
tive elements (transmitarray antennas, also called array lenses [1, 2]). Therefore, it is quite
tempting to become able to realize desired transformations (for example, focus or refract
wave beams) using extremely thin passive layers. Recently, there has been considerable in-
terest and progress in creating metasurfaces for controlling reflected and transmitted waves,
see recent reviews in [3–7]. Some limited manipulations of waves transmitted through a
thin metasurface can be accomplished due to specifically designed phase gradient over the
metasurface plane [8–11]. The required phase gradient is achieved by precise adjustment of
the phases of transmitted waves from each metasurface inclusion. Although this approach
has enabled realizations of transmitarrays even at optical frequencies, it suffers from very
low efficiency (less than 25% of transmitted power) and cannot provide control of polariza-
tion of the transmitted waves (in fact, it suffers from uncontrollable polarization rotation by
90◦ ). Subsequently, another approach based on generalized boundary conditions and the use
of symmetric metasurfaces was proposed by several researchers [12–15]. It provided more
efficient operation (more than 80% of transmitted power) and manipulation of polarization
[16]. However, even this approach cannot ensure ideal performance [7, 17], because these
symmetric layers cannot be matched to impedances of two propagating waves (incident and
transmitted) in different directions, and, therefore, they inevitably produce some reflec-
tions. Most recently, in Refs. [18, 19] it has been shown that the use of metasurfaces with
asymmetric response can open a possibility to realize metasurfaces for perfect refraction.

Known structures for manipulating reflection (both reflectarrays and metasurfaces) are
able to control reflection phase at each point of the reflector surface and nearly fully reflect
the incident power. Representative examples can be found in papers [20–27]. It has been
believed that these properties can allow full control over reflected waves. However, it is
not the case. As is shown in the submitted paper [28] and in this paper, lossless fully
reflecting metasurfaces designed to reflect a plane wave into another plane wave, always
produce parasitic beams in undesired directions. Without proper understanding of the

2
physical properties of metasurfaces responsible for refraction and reflection phenomena it is
not possible to create 100% efficient metasurfaces with desired properties.
Here we address this problem by introducing a general approach to the design of meta-
surfaces for arbitrary manipulations of plane waves, both in transmission and reflection. We
explain the main ideas of the proposed analytical approach to the synthesis of general func-
tional metasurfaces using simple but enough general examples of metasurfaces for refraction
or reflection of plane waves into arbitrary directions. In the first example, a metasurface
between two isotropic half-spaces (generally different) is designed so that a plane wave inci-
dent from one space (the incidence angle θi ) is fully refracted into a plane wave propagating
in the second space (the refraction angle θt ), without polarization change. We derive gen-
eral conditions on the equivalent circuit parameters of the metasurface to ensure perfect
refraction while the reflection coefficient is exactly zero (see Section II A). Subsequently, we
consider three different metasurface scenarios to satisfy these conditions (Sections II B, II C,
and II E). The latter scenario was independently considered in [18]. In the second exam-
ple, we show how to design metasurfaces which fully reflect plane waves into an arbitrary
direction (the reflection angle θr ). In this example, there are two plane waves coexisting
in the space in front of the reflecting metasurface. This issue complicates the study, but
the solution allows us to approach the problem of synthesis of metasurfaces for the most
general field transformations, where the main challenge is to account for interference be-
tween multiple plane waves. Indeed, any arbitrary field distribution can be represented as
a series of plane waves that interfere on both sides of the metasurface. In Section III A, we
examine conditions on the metasurface parameters for the perfect reflection regime. Similar
conditions were obtained independently in [28]. Next, in Sections III B, III C, and IV, we
consider different scenarios for metasurface realizations.
We show that perfect control over both refraction and reflection using lossless metasur-
faces requires careful engineering of spatial dispersion in the structure. To realize perfect
refraction, we need only weak spatial dispersion in form of the artificial magnetism and
bianisotropic omega coupling [29]. This effect is described by local relations between the
exciting electric and magnetic fields and the induced polarizations in the unit cells. Perfect
control over reflection using lossless metasurfaces appears to be possible only using strongly
non-local metasurfaces: part of the power received in one area of the surface should be
“channelled” and re-radiated at a different part of the surface. Lossless local-response meta-

3
surfaces (that is, conventional reflectarrays and earlier studied metamirrors) cannot create
a perfect reflected plane wave in any direction except the specular and retro directions.
The results clarify the necessary physical properties of metasurfaces for ideal wave re-
fraction and reflection and explain the limitations of earlier used design methods and earlier
studied realizations in form of electric and magnetic sheets, inhomogeneous high impedance
surfaces and reflectarrays. Possible routes towards realization of ideal and full control over
refraction and reflection are identified and discussed.

II. CONTROL OF TRANSMISSION: PERFECTLY REFRACTING METASUR-

FACE

As a first step we consider the problem of synthesis of metasurfaces for control of trans-
mitted waves. We require that a given plane wave is fully refracted into another plane wave,
without reflections or energy loss. The geometry of the problem is illustrated in Fig. 1. The

y z

Ei x
n θi
η1
η2
θt
Et

FIG. 1: Illustration of the desired performance of an ideally refracting metasurface.

metasurface is located in the yz-plane between two isotropic half-spaces with the charac-
teristic impedances η1 and η2 . We assume, without loss of generality of the approach, a
transverse electric (TE, with respect to the normal to the surface) incident plane wave. Our
approach can be used for waves of arbitrary polarizaitons, including arbitrary polarization
transformations, by using the dyadic parameters instead of the scalar ones.
Let us assume that the metasurface is illuminated from medium 1 by a plane wave (with

4
the wavenumber k1 and the electric field vector Ei ) at an angle θi . Requiring zero reflections,
the tangential field components Et1 and Ht1 on the illuminated side of the metasurface (at
x = 0) read
1
Et1 = Ei e−jk1 sin θi z , n × Ht1 = Ei cos θi e−jk1 sin θi z , (1)
η1
where z is the coordinate along the tangential component of the incident wavevector and
the unit vector n is orthogonal to the metasurface plane, pointing towards the source. The
time-harmonic dependency in form ejωt is assumed. We want to synthesize a metasurface
which will transform this incident wave into a refracted wave propagating in medium 2
(characterized by parameters k2 , η2 ) in some other direction, specified by the angle θt ,
without any loss of power. Therefore, the required tangential fields behind the metasurface
read
1
Et2 = Et e−jk2 sin θt z+jφt , n × Ht2 = Et cos θt e−jk2 sin θt z+jφt . (2)
η2
For generality, we assume that the refracted wave is phase-shifted by an arbitrary angle φt
with respect to the incident plane wave. With these notations, we can choose the origin of
the z-axis so that both Ei and Et will be real-valued vectors.
Obviously, the phase of the transmission coefficient

Φt (z) = ∠(Et2 /Et1 ) = −k2 sin θt z + φt + k1 sin θi z (3)

is not uniform over the surface, as long as k2 sin θt 6= k1 sin θi . Differentiating the above
equation, one can find the relation between the incidence and refraction angles in terms of
the transmission coefficient phase gradient:

dΦt (z)
k1 sin θi − k2 sin θt = . (4)
dz

This result suggests the simplest approach to the realization of refractive surfaces: Designing
a locally-periodical surface whose transmission coefficient is unity in the absolute value
(lossless Huygens’ sheet) and the phase of the transmission coefficient linearly changes in
accordance with (4). This method was used for a long time in antenna engineering (e.g.
[30]) and more recently in designs of metasurfaces, both in microwaves (e.g. [31]) and optics
(e.g. [8]). However, this approach does not lead to the desired perfect refraction [17], and
next we will explain how the desired performance can be realized exactly.

5
 A. Conditions on the equivalent circuit parameters

First, let us find the amplitude of the transmitted wave Et which corresponds to full power
transmission through the metasurface. Looking for possible realizations as metasurfaces with
local response, we equate the normal (to the metasurface) components of the Poynting vector
at each point of the metasurfaces, in the two media:
1 1
Re(Et1 × H∗t1 ) = Re(Et2 × H∗t2 ), (5)
2 2
and substitute the field values from (1) and (2). As a result, for metasurfaces with locally
full power transmission we obtain
r r
cos θi η2
Et = Ei . (6)
cos θt η1
Note that the amplitude of the transmitted wave can be larger or smaller than the amplitude
of the incident plane wave, although the metasurface is lossless and the power is conserved
in transmission. This result already tells about a limitation of the mentioned above simple
design approach based only on engineering the transmission phase according to (4).
Let us write the linear relations between the tangential fields at the two sides of the
metasurface in form of an impedance matrix:

Et1 = Z11 n × Ht1 + Z12 (−n × Ht2 ), (7)

Et2 = Z21 n × Ht1 + Z22 (−n × Ht2 ), (8)

and find such values of the Z-parameters which correspond to this particular field transfor-
mation. Knowing the Z-parameters of a metasurface, we will be able to determine suitable
topologies of constitutive elements (the unit-cell structures) which will realize the desired
functionality. Furthermore, the use of the equivalent T -circuit (Fig. 2) helps in understand-
ing what physical properties the metasurface should have in order to provide the desired
response.
Substituting the field values from (1), (2), and (6), we get the following equations for the
Z-parameters:
1
e−jk1 sin θi z = Z11 cos θi e−jk1 sin θi z
η1
1 p (9)
− Z12 √ cos θi cos θt e−jk2 sin θt z+jφt ,
η1 η2

6
 nxH t1 Z11– Z12 Z22– Z21 nxH t2

E t1 E t2
Z12=Z21

FIG. 2: Equivalent T -circuit of a reciprocal metasurface for the considered case of one
linear polarization (TE).

1 p
e−jk2 sin θt z+jφt = Z21 √ cos θi cos θt e−jk1 sin θi z
η1 η2
(10)
cos θt −jk2 sin θt z+jφt
− Z22 e .
η2

Obviously, there are infinitely many solutions for the unknown Z-parameters, because we
have only two conditions imposed on four complex parameters. Note that solutions with
complex values of impedance parameters mean that some of the components forming the
metasurface are either lossy or active, but all these solutions still correspond to the overall
lossless response of the metasurface, because the fields on the two sides of the metasurface
form plane waves carrying the same power in both upper and lower half-spaces.

This observation suggests that we can impose some restrictions on the values of the
equivalent parameters of the metasurface for a specific transformation and achieve different
realizations of metasurfaces which all perform the same operation on incident plane waves.
The possibility of multiple realizations of arbitrary metasurfaces using the susceptibility
model was discussed in Ref. [15].

B. Teleportation metasurface

Considering equations (9) and (10), we observe that while the left-hand sides are single
exponential functions (corresponding to either incident or transmitted wave), the right-hand
sides are sums of two different exponential functions. This property indicates that in general
the Z-parameters of the metasurface will depend on the coordinate z, that is, the metasurface
is, in general, not uniform. However, there is an interesting and conceptually simple solution
corresponding to a homogeneous metasurface. If we assume that Z12 = Z21 = 0, then both

7
equations are satisfied with

η1 η2
Z11 = , Z22 = − . (11)
cos θi cos θt

In this scenario, the metasurface is formed by a matched absorbing layer (the input resistance
Z11 ), a perfect electric conductor (PEC) sheet, and an active layer (an “anti-absorber” [32,
33]) on the other side. The incident plane wave is totally absorbed in the matched absorber.
The negative-resistance sheet (resistance Z22 ) together with the wave impedance of medium
2 forms a self-oscillating system whose stable-generation regime corresponds to generation
of a plane wave in the desired direction (the refraction angle θt ). Indeed, the sum of the
wave impedance of plane waves propagating at the angle θt and the input impedance of the
active layer is zero, and this is the necessary condition for stable generation. This structure
is similar to the “teleportation metasurface” introduced in [32, 33] for teleporting waves
without changing the propagation direction. As shown in [33], in that case if the reflector
separating the resistive and active layers is made at least slightly imperfect (parameters
Z12 = Z21 are very small but not exactly zero), the amplitude and phase of the transmitted
wave is synchronised with the incident field.
The teleportation metasurface is a theoretically perfect realization of the desired trans-
formation of plane waves in transmission. In particular, when the incidence angle equals
θi , the reflection coefficient is exactly zero. However, because the input resistance of the
metasurface seen from medium 2 is negative, reflections of waves coming from this medium
are very strong. Therefore, within the linear model of the negative resistance, the reflection
coefficient tends to infinity for waves coming from the direction θt . Moreover, the amplitude
of the field in medium 2 is established due to non-linear saturation of the negative resistance
device. Therefore, it is probably practically impossible to ensure that the negative resis-
tance saturates at exactly the desired amplitude of the generated wave. Next, we consider
an alternative realization, requiring perfect matching of the metasurface for waves coming
from medium 2.

C. Transmitarray

Let us consider alternative realizations demanding that the input impedance of the meta-
surface seen from medium 2 is matched to the wave impedance in medium 2, so that waves

8
coming from the direction θt will not produce any reflections. This requirement can be
satisfied if we demand that
η2
Z22 = . (12)
cos θt

Now we can find a realization of the metasurface as a nonreciprocal system where the ideal
voltage source in the output branch is defined by
√
2 η1 η2
Z21 =√ e−j(k2 sin θt −k1 sin θi )z+jφt , (13)
cos θi cos θt

as follows from (10). If the desired response for illumination from medium 1 is the only
η1
requirement, we can set Z12 = 0 and Z11 = cos θi
, so that for illuminations from medium
1 at the incidence angle θi the metasurface is acting as a matched absorber (matched re-
ceiving antenna array). This realization can be modeled by the corresponding nonreciprocal
equivalent circuit, shown in Fig. 3.

nxH t1 Z11 Z22 nxH t2

E t1 E t2
Z21 E t1

FIG. 3: Equivalent T -circuit of a nonreciprocal transmitarray realization of refractive

metasurfaces.

This realization reminds conventional transmitarrays [1]. The incident plane wave is
received by a matched antenna array on one side of the surface and the wave is launched into
medium 2 with a transmitting phase array antenna. In the ideal situation the transmitarray
is overall lossless, as the resistance seen from the illuminated side is in fact the radiation
resistance of the transmitting array (the two arrays need to be connected by matched cables).
The same model describes also the concept of field control and active cloaking using active
Huygens’ surfaces [13, 34]. In that scenario, there is no connection between the receiving
side (realized as a matched absorber) and the active array. The incident field is assumed
to be known and the amplitudes and phases of sources feeding the radiating array are set
accordingly.

9
 D. Symmetrical double current sheets

Within the metasurface paradigm, the simplest approach to realization of refractive meta-
surfaces is to assume that the refraction is controlled by engineering surface densities of
electric and magnetic current sheets, co-existing at the metasurface plane. It is obvious that
sheets of only electric or only magnetic currents cannot offer the desired functionality because
of the symmetry of the scattered fields in the forward and backward directions. Because
electric and magnetic surface current sheets are conveniently modeled by surface impedance
relations, it appears reasonable to model refractive metasurfaces by two impedance relations
which should hold both for the electric and magnetic surface current densities Je and Jm
[35–37]:
Et1 + Et2
Je = n × Ht1 − n × Ht2 = Ye Et = Ye , (14)
2
Ht1 + Ht2
Jm = −n × (Et1 − Et2 ) = Ym Ht = Ym . (15)
2
Here Et and Ht are the averaged tangential electric and magnetic fields at the metasurface
plane. Forming sums and differences of (7) and (8), it is easy to see that relations (14) and
(15) can hold only if the metasurface is symmetric and reciprocal, that is, when Z11 = Z22
and Z12 = Z21 . Under these assumptions,

2
Ye = , Ym = 2(Z11 − Z12 ). (16)
Z11 + Z12

Since we have only two unknown complex parameters Z11 and Z12 , the solution of (9) and
(10) becomes unique and it reads

η1 e−jΦt + ejΦt
Z11 = , (17)
cos θi e−jΦt − ηη1 cos θt jΦt
cos θ
e
2 i

√ η1 cos θt
+1
η1 η2 η2 cos θi
Z12 =√ , (18)
cos θi cos θt e−jΦt − ηη12 cos θt jΦt
cos θi
e

where Φt is defined by (3). We see that these parameters, as well as the electric sheet
admittance and magnetic sheet impedance (16), are complex numbers, which physically
means that the surface is either lossy or active at different values of z. For a special case of
refraction of a normally incident plane wave at 45◦ such solution for sheet parameters has
been published in [12, 38] and later on discussed in e.g. [7].

10
 Inspecting (17) and (18), we see that the metasurface parameters can be purely imaginary
for all z, corresponding to passive lossless realizations, only if

η1 cos θt
= 1, (19)
η2 cos θi

in which case

η1 1 η1
Z12 = j , Z11 = j cot Φt . (20)
cos θi sin Φt cos θi

Corresponding surface admittances, given by (16), are also purely imaginary and coincide
with those derived in [17] in an alternative way. Condition (19) physically means that the
impedance of the incident plane wave at the top side of the metasurface ( cosη1θi ) equals to the
impedance of the refracted wave at the bottom side of the surface ( cosη2θt ). It is, however, in
contradiction with the desired field structure: Equations (1) and (2) imply that the ratio of
the tangential field components (the wave impedance) must in general change if we require
perfect refraction. Thus, lossless double current sheets modeled by impedance relations (14)
and (15) cannot realize perfectly refractive metasurfaces.

In paper [17] the requirement for equal impedances (19) was derived in a different way,
demanding the absence of losses, and it was concluded that perfect refraction using lossless
metasurfaces was not possible without reflections. Indeed, it is clear that adding some
reflected field to (1), it is possible to make sure that the ratio of the tangential fields on top of
the metasurface is the same as at the bottom. This approach is followed nearly in all current
literature on lossless metasurfaces for refraction control: Nearly always only symmetric
metasurfaces have been considered and used (see [5, 7]) and the realization is thought in
form of a symmetric double-current sheet. This is the reason why earlier publications (see
the review in [7]) state that there must be at least small reflections or there is a need to use
active elements. The only known to us exception is the recent paper [18] where the problem
is attacked using the generalized scattering matrix.

Next we show that perfect refraction at an angle which is not equal to the incidence
angle is in fact possible using only lossless structures, but only if the surface is spatially
dispersive, exhibiting bianisotropic omega coupling. This result has been independently
obtained in [39].

11
 E. Metasurface formed by lossless elements

In the above example realizations, metasurfaces contained both lossy and active ele-
ments, which may require complicated and expensive realizations. It is therefore of interest
to consider if and how one can realize the same functionality using only reactive lossless
components.

1. Impedance matrix

To answer this question, we again consider the main set of requirements on the Z-
parameters of an ideal refractive metasurface (9) and (10) and look for a solution where
all the Z-parameters are purely imaginary (i.e., Zij = jXij ):

1
e−jk1 sin θi z = jX11 cos θi e−jk1 sin θi z
η1
1 p (21)
− jX12 √ cos θi cos θt e−jk2 sin θt z+jφt ,
η1 η2

1 p
e−jk2 sin θt z+jφt = jX21 √ cos θi cos θt e−jk1 sin θi z
η1 η2
(22)
cos θt −jk2 sin θt z+jφt
− jX22 e .
η2
This is a system of four real-valued equations for four real unknowns Xij , which has a unique
solution:
η1
X11 = cot Φt , (23)
cos θi
η2
X22 = cot Φt , (24)
cos θt
√
η1 η2 1
X12 = X21 = √ . (25)
cos θi cos θt sin Φt
For the case of zero phase shift (φt = 0) formulas (23)–(24) agree with the result of [18],
obtained using the generalized scattering parameters approach.
The metasurfaces modeled by (23)–(25) are reciprocal (X12 = X21 ). Indeed, the same
solution follows from (9)–(10) if we demand that a plane wave coming from the second
medium (the incidence angle θt ) is fully transmitted into the first medium in the direction
θi . The required physical properties of such metasurfaces can be understood from the cor-
responding equivalent T -circuit (see Fig. 2). The circuit is asymmetric, because X11 6= X22 .

12
This structure of the Z-matrix corresponds to bianisotropic omega layers, see a discussion
in [40, 41]. Possible appropriate topologies include arrays of Ω-shaped inclusions [42], arrays
of split rings, double arrays of patches (patches on the opposite sides of the substrate must
be different to ensure proper magnetoelectric coupling) [43–46], etc. A more complicated
set of three parallel reactive sheets was proposed in [18].
Previuosly, probably only in paper [18] asymmetric metasurfaces were used for transmis-
sion management (equations (23)–(25) also appear in [18] for the case when φt = 0). Note
also that the role of the omega-type bianisotropy of metasurfaces has been discussed in the
review paper [4], and omega layers have been successfully used in single-layer metamirrors
[42].
Comparing to the simple designs based on symmetrical metasurfaces (Section II D), we
again see from (23) and (24) that lossless symmetric realizations with X11 = X22 are possible
η1 η2
only if cos θi
= cos θt
, as we already saw from requirement (19). If media 1 and 2 are the same,
we can conclude that previously proposed symmetrical metasurfaces cannot provide perfect
refraction (without parasitic reflections or energy loss).

2. Unit-cell polarizabilities and appropriate topologies

Although the impedance matrix model provides a simple tool to design structures for de-
sired wave transformations, it is not directly applicable for identifying appropriate topologies
of the metasurface unit cells. Here we show how to determine what are the required proper-
ties of unit cells which realize ideally refractive metasurfaces. Knowing the polarizabilities
of each unit cell, we can identify what polarization response should be generated in unit cells
and what inclusions are needed to realize this response. So-called collective polarizabilities
[47] relate the tangential electric and magnetic dipole moments induced in the unit cell to
the fields of the incident wave. Knowing the Z-parameters of a metasurface is tantamount
to knowing reflection and transmission coefficients. Writing them also in terms of the col-
lective polarizabilities of unit cells, we can find the required polarizabilities which realize
the desired response. For the perfect refractive metasurfaces the collective polarizabilities
of unit cells read (see [48])
" s s ! #
S cos θi cos θt η1 cos θt η2 cos θi
α yy
bee = 2− + ejΦt (z) , (26)
jω η1 cos θt + η2 cos θi η2 cos θi η1 cos θt

13
 " s s ! #
S η1 η2 η1 cos θt η2 cos θi
α zz
bmm = 2− + ejΦt (z) , (27)
jω η1 cos θt + η2 cos θi η2 cos θi η1 cos θt

yz zy S η2 cos θi − η1 cos θt
α
bem = −b
αme = , (28)
jω η1 cos θt + η2 cos θi
yy zz yz zy
where S is the unit-cell area and α
bee , α
bmm , α
bem , α
bme are, respectively, electric, magnetic,
electromagnetic, and magnetoelectric polarizability components (coupling coefficients). The
yz zy
last two coefficients α
bem and α
bme imply so-called bianisotropic response in the unit cells
which models the effect of weak spatial dispersion [29]. In other words, the incident electric
(magnetic) field should induce also magnetic (electric) polarization in the unit cell. Here,
yz zy
α
bem = −b
αme , which is a typical characteristic of reciprocal omega inclusions [29].
As it can be expected, both the electric and magnetic polarizabilities in (26) and (27)
depend on z, and this dependence is the same for both of them. This result reflects the
requirement of zero reflection at any point of the metasurface, which demands the balance
of the induced electric and magnetic surface currents at any point (Huygens’ condition).
On the other hand, the omega coupling coefficient in (28) is constant with respect to z and
depends only on the impedances and angles. This result reflects the fact that bianisotropic
coupling of omega-type is necessary to ensure that the waves incident on both sides of the
metasurface see the same surface impedance, so that reciprocal full transmission is realized.
Since the impedances of the two waves depend only on the impedances of the media and on
the two angles, the coupling coefficient also depends only on these parameters. As expected,
we see that when the impedances of the incident and transmitted waves are the same, that
η1 η2
is, cos θi
= cos θt
, the required coupling coefficient vanishes.
Bianisotropic metasurfaces with the required properties defined by (26)–(28) can be re-
alized as arrays of low-loss particles with the appropriate symmetry. As it was mentioned,
for microwave applications, metallic canonical omega particles or double arrays of asymmet-
ric patches can be used. Multilayered topologies were proposed in paper [18]. For optical
applications, arrays of properly shaped dielectric particles were introduced as omega-type
bianisotropic metasurfaces [49, 50].
It is important to compare the polarizabilities (26)–(28) which are required for realizing
perfect refraction with the polarizabilities found in earlier works on wave transformations
in the transmitting regime (e.g., [8–11, 31]), where the design approach is based on the
geometrical optics model and the “generalized law of refraction” (3). In that theory, the

14
metasurface is assumed to be locally periodical, and the unit cells are designed so that
the transmission coefficient has unit amplitude and the desired phase at every point. These
requirements are satisfied if the electric and magnetic polarizabilities read (taking the earlier
considered special case of normal incidence and identical media at both sides [31])

1 zz S

α yy
bee = 2
α
bmm = 1 − ejΦt (z) , (29)
η jωη

yz
and the magnetoelectric coupling coefficient α
bem is zero. Periodical arrays formed by unit
cells having these collective polarizabilities have unit transmissivity and the transmitted
waves have the required phases Φt (z), but when the cells are assembled into a non-uniform
array, the performance becomes non-ideal. In other words, in order to ensure the desired
response of the non-uniform metasurface, properties of periodical arrays formed by its unit
cells must deviate from the simple geometrical-optics design recipe (3). This result is con-
sistent with that in [6]. We can conclude that in order to ensure perfect refraction, it is not
enough to make the metasuface bianisotropic (introducing asymmetry with respect to its
two sides). The electric and magnetic polarizabilities in the exact synthesis [see (26)–(28)]
solution are also different as compared to the conventional synthesis solution (29).

III. CONTROL OF REFLECTION: PERFECTLY REFLECTING METASUR-

FACE

In the previous case of refractive metasurfaces, there is only one single plane wave at
every point of space. In order to be able to synthesize metasurfaces for general field trans-
formations, we need to understand how to control several plane waves which propagate
and interfere in the same space. This problem can be solved at an example of a perfectly
reflective metasurface, which we consider next.
The geometry of the problem is shown in Fig. 4. The design goal is to fully reflect a plane
wave coming from a given direction θi into another plane wave propagating in a different and
also arbitrary direction θr . Here, we consider the case when the polarization of the reflected
wave is the same as that of the incident wave. Metasurfaces designed for full reflection were
called metamirrors in [42, 43]. In this scenario, the desired field distribution at the surface of
the metamirror is the superposition of two plane waves (the incident wave and the reflected

15
 y z
x
Ei
E
n θi θr
η1
η2
FIG. 4: Illustration of the desired performance of an ideally reflecting metasurface. TE
incidence is assumed and the metasurface is located in the yz-plane.

wave):
Et1 = Ei e−jk1 sin θi z + Er e−jk1 sin θr z+jφr ,
(30)
1 1
n × Ht1 = Ei cos θi e−jk1 sin θi z − Er cos θr e−jk1 sin θr z+jφr .
η1 η1
Here, Et1 and Ht1 are the tangential (to the metamirror plane) components of the total
electric and magnetic fields at the metamirror surface. For generality, we assume that the
reflected plane wave can have any desired phase shift φr with respect to the incident wave.
With these notations, we can choose the origin of the z-axis so that both Ei and Er will be
real-valued vectors.
Similarly to the refractive metasurface, we see that the phase of the reflection coefficient

Φr (z) = −k1 sin θr z + φr + k1 sin θi z (31)

depends on z, except the trivial case of specular reflection (θi = θr ). Differentiating, we

find the relation between the reflection and incidence angles in terms of the gradient of the
reflection coefficient phase:
dΦr (z)
.
k1 (sin θi − sin θr ) = (32)
dz
Analogously with the transmitting regime, this result suggests a simple design approach:
to realize a fully reflective surface (the amplitude of the reflection coefficient equals unity
at each point) but with a linearly varying reflection phase, according to (32). Reflecting
surfaces with engineered reflection phase are often called high impedance surfaces [51] or
reflectarrays [52]. Such an approach has been used, for example, in [20–27, 42] as well as in
all known designs of reflectarrays.

16
 However, similarly to refracting metasurfaces, in designing reflecting surfaces this sim-
plistic method also does not allow us to exactly realize the desired performance. Next, we
present the theory of perfect reflecting surfaces and explore various reflection scenarios, with
their advantages and limitations.

A. Power flow into the metamirror

Applying the same method as in analysing metasurfaces for transmission control, we start
from considering the power flow into the metamirror structure. The normal component of
the Poynting vector at the reflector surface reads

1
Pn = Re(Et1 × H∗t1 ). (33)
2

Substituting the required field distributions (30), we can write the normal component of the
Poynting vector as

1  
Pn = −Ei2 cos θi + Ei Er (cos θr − cos θi ) cos Φr (z) + Er2 cos θr , (34)
2η1

where the reflection phase Φr (z) is defined by (31).

If this quantity is identically zero at all points along z, the metasurface locally (at every
point) acts as a lossless reflector. Conventional realizations of non-uniform reflectors belong
to this class of locally responding reflectors. Examining the above expression, we see that
within this scenario, full transformation of an incident plane wave into a single reflected
plane wave of the same polarization is impossible, except the cases of specular or retro-
reflection, when θr = ±θi (this fact is proven also in [28]). Indeed, the expression for the
normal component of the Poynting vector (34) contains an oscillating term, proportional to
cos Φr (z), which can be zero only if cos θr = cos θi , that is, θr = ±θi . Therefore, any local,
passive and lossless non-uniform reflecting surface will create modulated reflected waves with
spatial dependence of the fields different from the design target (30).
The same expression (34) tells also that it is possible to reflect a plane wave into only
one plane wave along a specified direction if we allow energy loss in the metasurface. To
understand this conclusion, let us look for such constant amplitude of the reflected wave
Er which ensures that Pn ≤ 0 for all z (negative values of Pn corresponds to flow of power
into the surface, where it is absorbed). Obviously, this condition can be satisfied if Er = Ei ,

17
since with this amplitude of the reflected field we have

Ei2
Pn = (cos θr − cos θi ) [1 + cos Φr (z)] . (35)
2η1

Since 1 + cos Φr (z) is non-negative, Pn is negative or zero at all points of the metasurface
z if cos θr − cos θi ≤ 0. This realization scenario was introduced in [28]. For instance,
if the metamirror is excited by a normally incident plane wave (θi = 0), it is possible to
create a single reflected plane wave along any direction, because cos θr ≤ 1 for any θr .
However, as is seen from Eq. (35), the amount of power which is lost in the metasurface
increases with increasing difference between the incidence and reflection angles. In the limit
of cos θr − cos θi → −1, which corresponds to θi → 0 and θr → π/2, all incident power
is completely absorbed. Figure 5 (dashed line) shows the efficiency of this scenario as a
function of the reflection angle θr . The efficiency ζ is defined as the ratio of the plane-wave

0.8

0.6
ζ

0.4

0.2
Optimized lossless
Passive lossy
0
30 40 50 60 70 80 90
θr

FIG. 5: Comparison between the power efficiencies of the passive metamirror which reflects
a single plane wave [surface impedance (41), dashed curve] and the optimized metamirror
which minimizes reflections into non-desired directions [surface impedance (43), solid
curve] at normal incidence.

|Er |2
power carried into the desired direction Pr = 2η1
cos θr to the power of the incident plane
|Ei|2
wave Pi = 2η1
cos θi . As it is clear from (35) and this figure, increasing the reflection angle
results in decreasing the efficiency by a factor of cos θr / cos θi (notice that Er = Ei ).

18
 Actually, ideal reflection into a single plane wave without losing any power is possible, but
only if we allow periodical flow of power into the metamirror structure and back into space.
This conclusion is also evident from formula (34). Indeed, we see that if the amplitude of
the reflected plane wave equals √
cos θi
Er = √ Ei , (36)
cos θr
the normal component of the Poynting vector is a periodical function with zero average
value: √
Ei2 cos θi
Pn = √ (cos θr − cos θi ) cos Φr (z). (37)
2η1 cos θr
The metasurface performs the desired function perfectly, but the response must be strongly
non-local : the power which enters the metasurface structure in the areas where Pn < 0 must
be launched back from the areas where Pn > 0. Alternatively, the perfect reflection can
be achieved if the metasurface has active and lossy elements (being overall lossless in the
average over the surface area). We see again that there is no power flow into the metamirror
at any point only if θr = ±θi , in agreement with the previous conclusion.

B. Required surface impedance

Following the introduced synthesis approach based on the impedance matrix, we write
the linear relation between the tangential fields at the metamirror surface. Assuming that
the metamirror is a boundary and the fields behind it are zero (Et2 = 0, Ht2 = 0), we need
only one parameter of the Z-matrix (7)–(8), the input impedance Z11 :

Et1 = Z11 n × Ht1 . (38)

Substituting the desired field values from (30), we get the following equation for the unknown
input impedance Z11 :

Ei e−jk1 sin θi z + Er e−jk1 sin θr z+jφr

1
(39)
= Z11 Ei cos θi e−jk1 sin θi z − Er cos θr e−jk1 sin θr z+jφr .
η1
For the ideally performing non-local metasurface, which produced the reflected wave with
the amplitude given by (36), the corresponding input impedance reads
√ √
η1 cos θr + cos θi ejΦr (z)
Z11 = p p p . (40)
cos θi cos θr cos θi − cos θr ejΦr (z)

19
We see that the input impedance is a complex number, whose real part is a periodical
function of z. Figure 6 presents the required input impedance for the case when θi = 0◦
and θr = 70◦ . The real part of the input impedance periodically takes positive (loss) and

4
Z11 /η1

-2
ℜ{Z11 }
ℑ{Z11 }
-4
-1.5 -1 -0.5 0 0.5 1 1.5
z/λ

FIG. 6: The required normalized input impedance Z11 /η1 of the ideal metamirror for
θi = 0◦ , θr = 70◦ , φr = 0◦ .

negative (gain) values. The surface acts as if it is lossy close to the regions where the
reactive impedance is high (close to the regime of a perfect magnetic conductor, PMC)
and active in the areas where the reactance is small (close to a perfect electric conductor,
PEC). Importantly, this behaviour does not imply that the surface cannot be passive or
lossless. We stress that, on the contrary, properly tuned metasurface with strongly non-
local response can emulate such a metamirror: The power which passes through the input
surface in the “lossy” regions is not absorbed but it is re-radiated from the “active” regions.
Another possibility to realize the ideal performance dictated by impedance (40) could be a
metasurface with truly active and lossy elements where only the overall response is lossless.

As discussed in Section III A, it is possible to eliminate the need to realize active input
impedance (which increases the realization complexities), at the expense of losing some part
of the incident power in the metamirror. The surface impedance of such a lossy metasurface,
which creates a single plane wave in the desired direction, can be found from (39) upon

20
 5

Z11 /η1 0

ℜ{Z11 }
ℑ{Z11 }
-5
-0.5 0 0.5
z/λ

FIG. 7: The required normalized input impedance Z11 /η1 for passive metamirrors in the
case when θi = 0◦ , θr = 70◦ , φr = 0◦ . One period of the metamirror along the z-coordinate
is shown. The solid, dashed, and dotted lines correspond, respectively, to the lossy
metamirrors [Eq. (41)], the lossless metamirrors creating two reflected plane waves
[Eq. (43)], and the conventional non-uniform reflectors [Eq. (47)].

substitution of the reflected field amplitude Er = Ei . The result reads

e−jk1 sin θi z + e−jk1 sin θr z+jφr

Z11 = η1 . (41)
cos θi e−jk1 sin θi z − cos θr e−jk1 sin θr z+jφr

An example is plotted as a function of the coordinate in Fig. 7 (the solid lines). As it is

seen, the real part of the impedance is always non-negative, corresponding to the absorbed
power given by (35).

So far we have demonstrated that a surface having the input impedance (41) produces
a single (non-modulated) reflected wave in the desired direction if the power loss in the
metamirror is allowed [see Fig. 5]. However, depending on the application requirements, it
can be preferable to allow some modulation of the reflected wave but reduce the power loss.
In the next section, we present a scenario in which the metamirror is lossless at every point,
and at the same time the reflections into non-desired directions are reduced.

21
 C. Optimizing reflections from lossless and local metamirrors

It is possible to optimize the reactance Im{Z11 } profile of a lossless metamirror in order

to minimize reflections into non-desired directions based on particular optimization criteria.
As one example, we notice that there is an interesting lossless design, where all the power
which cannot be sent into the desired reflection direction θr is reflected into the specular
direction. To demonstrate this possibility, we consider the situation when the difference
between the incidence angle θi and θr is large, so that only three propagating plane waves
can exist in the Floquet spectrum of the propagating reflected field [53]
0
X
Er = En e−jk1 [(n+1) sin θr −n sin θi ]z . (42)
n=−2

For the normal illumination (θi = 0), this corresponds to θr larger than 30◦ . It is easy to
check that a set of three plane waves: the incident wave, the wave reflected into the desired
direction (n = 0), and the parasitic plane wave reflected into the specular direction (n = −1)
exactly satisfy the boundary condition (38) with a purely reactive impedance
η1
Z11 (z) = j cot [Φr (z)/2] , (43)
cos θr
if the wave reflected in the desired direction θr is given by
2 cos θi
E0 = Ei , (44)
cos θi + cos θr
and the wave reflected into the specular direction θi is
cos θi − cos θr
E−1 = Ei . (45)
cos θi + cos θr
The amplitude of the Floquet harmonic n = −2 is equal to zero, and the evanescent part
of the spectrum also vanishes. These amplitudes have been found by requiring that the
normal component of the Poynting vector is identically zero at the surface. In this case,
the metasurface is lossless and exhibits no strong spatial dispersion. Reciprocally, we can
conclude that 100% power reflection in the desired direction can be achieved by illuminating
the metasurface by two plane waves at once, properly selecting their relative amplitudes,
phases, and propagation directions.
It is interesting that the efficiency of transformation of the incident plane wave into the
desired reflected plane wave is much better than for the passive lossy scenario (presented in

22
Section III A) where the parasitic reflections were absent. This conclusion is illustrated in
Fig. 5 by comparing the efficiencies of these two cases.
The conventional approach for designing lossless non-uniform reflectors is based on the
“generalized reflection law” (32), which corresponds to a linear phase variation along the
metasurface. In that approach, the measurface is designed so that the local reflection co-
efficient at every point has unit amplitude and the phase as dictated by (32). The local
reflection coefficient is defined for an infinite uniform array, that is, the input impedance
can be found from
1

Ei + Er ejΦr (z) = Z11 Ei cos θi − Er cos θi ejΦr (z) , (46)
η1
where Er = Ei and Φr (z) is given by (31). The result reads
η1
Z11 = j cot [Φr (z)/2] , (47)
cos θi
and an example is plotted in Fig. 7. One can see that the required surface impedance in the
conventional reflectors is different from that of the lossless metamirror described by (43). In
the conventional reflectors, the reflected wave has a complex structure: Generally, several
propagating plane waves in different directions and some evanescent fields localized close to
the surface are excited. To study the field structure, one can use numerical simulations or
the theoretical technique exploited for the case of refraction in [54].

D. Unit-cell polarizabilities and appropriate topologies

Making use of the boundary conditions on the reflecting metasurface which tell that the
tangential electric and magnetic fields are equal, correspondingly, to the surface magnetic
and electric current densities, we can find relations between the surface impedance Z11 and
the collective polarizabilities of unit cells of the metamirror (see [48]):
 
η1 cos θi + cos θr jω η1 yy yz
= α
b +α bem , (48)
cos θi Z11 cos θr + η1 S cos θi ee
 
cos θi + cos θr jω cos θi zz yz
Z11 = α
bmm − αbem . (49)
Z11 cos θr + η1 S η1
Here S is the unit-cell area. Obviously, these equations have infinitely many solutions for
polarizabilities which realize the desired response. The metasurface can be either bian-
isotropic (omega coupling) or it can be a non-bianisotropic pair of electric and magnetic

23
current sheets. For the non-bianisotropic realization we set

yz zy
α
bem =α
bme = 0, (50)

and find the unique solution

yy S cos θi + cos θr
α
bee = , (51)
jω Z11 cos θr + η1
zz S Z11 η1 cos θi + cos θr
α
bmm = . (52)
jω cos θi Z11 cos θr + η1
We see that in the design of fully reflective metasurfaces, weak spatial dispersion effects are
necessary at least in form of the artificial magnetism. If we demand that both magnetic
polarizability and the bianisotropy coefficient are zero, the above equations have no solutions.
The use of bianisotropy offers additional design flexibilities.
Knowing the collective polarizabilities required for the desired performance we can imme-
diately see what are the appropriate topologies of unit cells. Since we need both electric and
magnetic polarizations, the physical thickness of the reflecting layer must be different from
zero, to allow formation of tangential magnetic moments in unit cells. For example, it is not
possible to realize the desired performance by any patterning of a single, infinitesimally thin
sheet of a perfect conductor. The non-bianisotropic realization scenario suggests the use
of a single array of small particles which are polarizable both electrically and magnetically,
such as small metal spirals as in [31]. A typical realization based on the bianisotropic route
is a high-impedance surface with a PEC ground plane (such as “mushroom layers” [51]).
An important advantage in using bianisotropic effects is the relaxed requirement on the
strength of the magnetic response. Especially for optical applications, it is easier to realize
strong bianisotropy (which is a first-order dispersion effect) as compared with the artificial
magnetism (which is a weaker, second-order effect) [29].

E. Ideal metamirrors

We have seen that all local lossless non-uniform reflectors modulate the reflected waves,
which reduces the power efficiency in the desired direction. The operation of conventional
planar reflectors (such as high impedance surfaces [51], reflective diffraction gratings [55],
and reflectarrays [52]) are similar in this respect. Next we discuss the potentials of ideal
metamirrors based on non-local and non-reciprocal surfaces.

24
 As shown above, it is possible to synthesise an overall lossless metamirror which would
create an unmodulated reflected wave into any desired direction, satisfying the requirement
(30) exactly, with a constant value of the reflected plane wave amplitude Er . This goal
can be achieved if we require that the normal component of the Poynting vector on the
metasurface is zero only in the average over the metamirror period, and not necessarily
is equal to zero at every point. In this case, the amplitude of the plane wave reflected
into the desired direction is given by (36), and the normal component of the Poynting
vector oscillates, according to Eq. (37). Realization of such metamirrors requires absorption
of power in some areas of the surface and generation of power in some other areas or,
alternatively, power channelling from one area to the other. Conceptually, this scenario of
balanced loss and gain can be realized using the same two approaches which were found
in the analysis of perfectly refractive metasurfaces: teleportation metasurface (Section II B)
and transmitarrays (Section II C). In the former approach, one can envisage a realization in
form of an array of small receiving antennas loaded by positive resistors in the areas where
the energy should be partially absorbed, and by negative resistors where the energy should
be launched back into space. This arrangement is similar to the teleportation metasurface
described in Section II B, where such arrays were positioned at the two opposite sides of a
metal screen. Alternatively, one can envisage a similar array of antennas, where the antennas
of the absorbing areas are connected by cables to the antennas of the active areas. Thus,
the power received at the absorbing areas is re-radiated by the active areas. It is important
to note that both these devices should be non-reciprocal, as the “active” antennas should
radiate power but not receive it back from space. Actual realization of both these concepts
is a challenging task. As to the teleportation approach, one needs non-reciprocal antennas,
which can be in principle realized using non-reciprocal materials like magnetized ferrites or
using active components. There is also an interesting possibility to use parametric circuits
for the same purpose [56]. The non-reciprocal transmitarray approach in principle can be
realized also in reflecting metasurfaces, using non-reciprocal circuits inside the metasurface,
but it appears that the use of spatial modulation of the surface impedance by external forces
(using unit cells equipped with varactors, for example), is more promising. Conceptually, the
desired performance can be achieved by modulating (for example) varactors in all unit cells
with the same amplitude but with different phases. Controlling the spatial distribution of the
modulation phase, one can possibly realize parameteric amplification or absorption according

25
to the design specifications. Initial work on space-time modulated metasurfaces [57, 58]
produced interesting and promising results, and we expect that developing this route may
lead to realizations of theoretically perfectly operating lossless non-uniform metasurfaces.

IV. PERFECTLY REFLECTING POLARIZERS

In the previous section, we considered metamirrors which reflect an incident plane wave
into a desired direction. However, we encountered either active-lossy realizations of the
metamirror or lossless reflection of modulated waves. In this section, we introduce a new
solution for a lossless metamirror which ideally reflects the incident wave into the desired
direction without any modulations. Since the main reason for modulations of reflected waves
is interference between the incident and reflected fields, we construct a metamirror which
reflects waves with the polarization orthogonal to that of the incident wave. As a simple
canonical example, we consider the transformation of a transverse electric (TE) wave with
the amplitude Eiy into a transverse magnetic (TM) wave with the amplitude Erz = Er cos θr ,
propagating in the desired direction. Figure 8 shows the problem configuration. It is clear

y z
x
Ei
Hr
n θi θr
η1
η2
FIG. 8: Illustration of the desired performance of an ideal metamirror which perfectly
transforms a TE incident wave into a TM reflected wave.

that in this case there is no interference between the incident and reflected waves. The
desired tangential electric and magnetic fields at the metamirror surface read

Et1 = ŷ Eiy e−jk1 sin θi z + ẑ Erz e−jk1 sin θr z+φr , (53)

and
cos θi y −jk1 sin θi z 1
n × Ht1 = ŷ Ei e − ẑ Erz e−jk1 sin θr z+φr , (54)
η1 η1 cos θr

26
respectively. Considering the metamirror as a boundary, the impedance relation between
the tangential electric and magnetic fields (7)–(8) in this case reads [47]

Et1 = Z 11 · n × Ht1 , (55)

where the impedance has the matrix form

 
yy yz
Z11 Z11 
 
Z 11 = . (56)
 
zy zz
Z11 Z11

Notice, in contrast with the previous case, we should consider the full-rank impedance
dyadics Z in order to account for any possible polarization transformation. Substituting
(53) and (54) into (55), we obtain the following matrix equation
   cos θi

yy yz
1 Z Z
   11 11   η1 
    
 =  , (57)
   



zy −1
Rzy ejΦr zz
Z11 Z11 Rzy ejΦr
η1 cos θr

where Rzy = Erz /Eiy and Φr is defined in (31). The solution of the above equation for the
lossless case (i.e., Re{Z 11 } = 0) is unique and reads as
   
η1 η1 cos θr 1
yy
Z Z yz
 cos θi cot Φ r
 11 11   Rzy sin Φr  
   .
 =j  (58)
   
zy zz η 1 1
Z11 Z11 Rzy η1 cos θr cot Φr
cos θi sin Φr
yy
Here, Z11 is the metamirror input impedance which is responsible for suppressing unwanted
yz
reflections in the specular direction. The proper values of the cross-components Z11 and
zy zz
Z11 ensure the polarization rotation, and, finally, Z11 is responsible for reflection with the
orthogonal polarization in the desired direction.
Next, we apply the condition for power conservation (we demand that the normal com-
ponent of the Poynting vector identically equals zero at the metasurface plane at each point
to ensure local response) to find the required reflection coefficient Rzy . This condition reads

2 1 1
− (Eiy ) cos θi + (Erz )2 = 0, (59)
2η1 2η1 cos θr

27
which defines the reflection coefficient for the perfect reflection regime:
p
Rzy = cos θi cos θr . (60)

As it is clear from (58), realization of this scenario is possible with purely lossless meta-
surface elements. Moreover, since the reflected field does not interfere with the incident
one, there is no field modulation. Therefore, the proposed metamirror provides an ideal and
single reflecting wave.

A. Unit-cell polarizabilities and appropriate topologies

Following the procedure outlined in Sections III D, we can find the relations for collective
polarizabilities of unit cells of the proposed metamirror in the case of perfectly reflecting
polarizers (see [48] for details):
 
jω cos θi zz zy
1= αbmm + α bme , (61)
S η1
 
zy jΦr jω cos θi yz yy
R e =− α
bmm + α bme , (62)
S η1
 
cos θi jω cos θi yz yy
= αbem + α
bee , (63)
η1 S η1
 
Rzy ejΦr jω cos θi zz zy
=− α
bem + αbee . (64)
η1 cos θr S η1
Obviously, these equations have infinitely many solutions for polarizabilities which realize
the desired response. Even restricting ourselves by reciprocal realizations, the metamirror
can be either bianisotropic (both omega and chiral couplings) or it can be non-bianisotropic
with anisotropic electric and magnetic responses. Here we show two simple design solu-
tions. In the first design, the metamirror is modeled by anisotropic electric and magnetic
polarizabilities. The non-zero polarizabilities read:

zz S η1
α
bmm = , (65)
jω cos θi

yz S η1
α
bmm =− Rzy ejΦr , (66)
jω cos θi
yy S cos θi
α
bee = , (67)
jω η1

28
 zy S 1
α
bee =− Rzy ejΦr . (68)
jω η1 cos θr
zy yy yz zz
In this design, the bianisotropic properties are excluded, that is, α
bme =α
bme =α
bem =α
bem =
yz
0. Notice that there is no limitations on the selection of the polarizability components α
bee ,
zz zy yy
α
bee ,α
bmm , and α
bmm (they can be chosen from considerations of reciprocity, for example).
Alternatively, another simple solution of system (61)–(64) can be found by suppressing
zy yz yz zy
the cross polarizability components (i.e., α
bme =α
bmm =α
bem =α
bee = 0 ). This implies that
the metamirror possesses chiral bianisotropic response:
zz S η1
α
bmm = , (69)
jω cos θi
yy S
α
bme = − Rzy ejΦr , (70)
jω
yy S cos θi
α
bee = , (71)
jω η1
zz S ejΦr
α
bem =− . (72)
jω Rzy
zz yy zz yy
while there is no limitation on α
bee ,α
bem ,αbme and α
bmm (they can be chosen from considerations
yy yy
of reciprocity). It can be shown that if we apply the reciprocity condition (b
αem = −b
αme and
zz zz
α
bme = −b
αem [29]) and choose
S
zz 1
α
bee = , (73)
jω η1 cos θi
yy S
α
bmm = η1 cos θi , (74)
jω
then the same metamirror dually operates both for TE and TM polarized incident waves.
One can note a similarity of the conditions on the polarizabilities (69)–(74) with those used
earlier for realizing polarization transformers [59] and absorbers [31, 60]. Here we see that
the amplitudes of the polarizabilities should be balanced (as shown in [59] for the normal
incidence), and the ideal reflector operation is ensured by proper adjustments of the chirality
parameter phase.
These solutions are only two possibilities, selected for their simplicity. Other solutions
are possible considering (61)–(64).

V. CONCLUSIONS AND DISCUSSIONS

In this paper we have introduced a general approach to the synthesis of metasurfaces for
arbitrary manipulations of plane waves. We have explained the main ideas of the method

29
on two canonical examples: A metasurface which perfectly refracts plane waves incident at
an arbitrary angle θi into plane waves propagating in an arbitrary direction defined by the
angle θt , and a metasurface which fully reflects a given plane wave into an arbitrary direction
θr . The general synthesis approach shows a possibility for alternative physical realizations,
and we have discussed different possible device realizations: self-oscillating teleportation
metasurfaces, non-local metasurfaces, and metasurfaces formed by only lossless components.
The crucial role of omega-type bianisotropy in the design of lossless-component realizations
of perfectly refractive surfaces has been revealed.
The conventional approach to realization of refractive and reflecting metasurfaces as well
as both transmitarray and reflectarray antennas is based on requiring full power transmis-
sion or reflection at each point of the surface and providing complete phase control over the
transmitted and reflected waves. We have clarified the role of modifications in the required
phase gradient for conventional planar refractive/reflective structures in gaining higher ef-
ficiencies. Moreover, we have revealed fundamental limitations of this classical technique
and showed how the ideal performance can be realized. For full control over transmission,
weak spatial dispersion in form of bianisotropic coupling is necessary, while ideal lossless
reflectarray operation calls for the use of structures with a strongly non-local response to
the incident fields or structures that transform polarization of reflected waves.
We think that the reason why the role of metasurface bianisotropy in controlling refraction
has not been appreciated earlier is that in this field transformation the wave polarization
should not change, and it appears natural to expect that bianisotropic effects, such as
chirality, are not needed. However, as we have shown here, omega coupling effects, which
do not change polarization, are crucial in engineering perfectly matched lossless refractive
metasurfaces.
In contrast to perfectly refracting metasurfaces, creation of perfectly reflecting surfaces
requires careful control over the interference of the incident and reflected waves. We have
shown that ideal transformation of an incident plane wave into a reflected plane wave prop-
agating at an angle different from what is dictated by the usual reflection law requires either
active structures or passive lossless non-local metasurfaces. We have discussed the structure
of reflected fields and proposed an optimal compromise realization using local and passive
metasurfaces.
In the last part of the paper we have shown that the requirement of strong spatial

30
dispersion or active inclusions for realization of perfect metamirrors can be lifted if the
polarization of the reflected wave is orthogonal to that of the incident field. In this case
there is no interference between the incident and reflected wave, and perfect reflection can
be realized using only weak spatial dispersion effects (artificial magnetism and chirality),
similarly to ideally refractive metasurfaces.
Since any exciting fields can be expressed in form of a plane-wave expansion, the developed
approach can be generalized to metasurfaces for the most general field transformations. We
hope that understanding of the physical requirements for perfect metasurface operation in
both transmission and reflection regime as well as the developed synthesis method will open
a way for design and realization of ultimately thin composite sheets for a broad range of
applications, such as lenses, antennas, sensors, etc.
We would like to note that during the review process of this paper a related preprint [61]
has been published, which describes a conceptual realization of perfectly reflecting lossless
metasurfaces in form of a set of three parallel reactive sheets. This structure exhibits the
required non-local properties (“channeling” energy in the transverse direction), according to
the theory presented here.

ACKNOWLEDGMENT

This work was supported in part by the Academy of Finland (project 287894).

[1] Z. Popović and A. Mortazawi, Quasi-Optical Transmit/Receive Front Ends, IEEE Trans. Mi-
crowave Theory Techn. 46, 11 (1998).
[2] S. V. Hum and J. Perruisseau-Carrier, Reconfigurable Reflectarrays and Array Lenses for
Dynamic Antenna Beam Control: A Review, IEEE Trans. Antennas Propag. 62, 1 (2014).
[3] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar, C. R. Simovski, Metasurfaces:
From microwaves to visible, Physics Reports, http://dx.doi.org/10.1016/j.physrep.2016.04.004
(2016).
[4] S. A. Tretyakov, Metasurfaces for General Transformations of Electromagnetic Fields, Phil.
Trans. R. Soc. A 373, 2049 (2015).

31
 [5] K. Achouri, B. A. Khan, S. Gupta, G. Lavigne, M. A. Salem, and C. Caloz, Synthesis of
Electromagnetic Metasurfaces: Principles and Illustrations, EPJ Appl. Metamat. 2, 12 (2015).
[6] N. M. Estakhri and A. Alù, Recent Progress in Gradient Metasurfaces, J. Optical Soc. of
America B 33, 2 (2016).
[7] A. Epstein and G. V. Eleftheriades, Huygens’ Metasurfaces via the Equivalence Principle:
Design and Applications, J. Optical Soc. of America B 33, 2 (2016).
[8] N. Yu, P. Genevet, M. A. Kats, F. Aieta, J.-P. Tetienne, F. Capasso, and Z. Gaburro, Light
Propagation with Phase Discontinuities: Generalized Laws of Reflection and Refraction, Sci-
ence 334, 333 (2011).
[9] A. V. Kildishev, A. Boltasseva, and V. M. Shalaev, Planar Photonics with Metasurfaces,
Science 339, 1232009 (2013).
[10] N. K. Grady, J. E. Heyes, D. R. Chowdhury, Y. Zeng, M. T. Reiten, A. K. Azad, A. J. Taylor,
D. A. R. Dalvit, and H.-T. Chen, Terahertz Metamaterials for Linear Polarization Conversion
and Anomalous Refraction, Science 340, 6138 (2013).
[11] M. I. Shalaev, J. Sun, A. Tsukernik, A. Pandey, K. Nikolskiy, and N. M. Litchinitser, High-
Efficiency All-Dielectric Metasurfaces for Ultracompact Beam Manipulation in Transmission
Mode, Nano Letters 15, 9 (2015).
[12] C. Pfeiffer and A. Grbic, Metamaterial Huygens Surfaces: Tailoring Wave Fronts with Reflec-
tionless Sheets, Phys. Rev. Lett. 110, 197401 (2013).
[13] M. Selvanayagam and G. V. Eleftheriades, Discontinuous Electromagnetic Fields Using Or-
thogonal Electric and Magnetic Currents for Wavefront Manipulation, Optics Express 21, 12
(2013).
[14] F. Monticone, N. M. Estakhri, and A. Alù, Full Control of Nanoscale Optical Transmission
with a Composite Metascreen, Phys. Rev. Lett. 110, 203903 (2013).
[15] K. Achouri, M. A. Salem, and C. Caloz, General Metasurface Synthesis Based on Susceptibility
Tensors, IEEE Trans. Antennas Propagat. 63, 7 (2015).
[16] C. Pfeiffer and A. Grbic, Millimeter-Wave Transmitarrays for Wavefront and Polarization
Control, IEEE Trans. Microwave Theory Techn. 61, 12 (2013).
[17] A. Epstein and G. V. Eleftheriades, Passive Lossless Huygens Metasurfaces for Conversion of
Arbitrary Source Field to Directive Radiation, IEEE Trans. on Antennas and Propagat. 62,
11 (2014).

32
[18] J. Wong, A. Epstein, and G. Eleftheriades, Reflectionless Wide-Angle Refracting Metasurfaces,
IEEE Antenn. Wireless Propag. Lett. 15, 1293 (2016).
[19] V. Asadchy, M. Albooyeh, S. Tcvetkova, Y. Ra’di, and S. A. Tretyakov, Metasurfaces for Per-
fect and Full Control of Refraction and Reflection, http://arxiv.org/abs/1603.07186 (2016).
[20] S. Sun, K.-Y. Yang, C.-M. Wang, T.-K. Juan, W. T. Chen, C. Y. Liao, Q. He, S. Xiao,
W.-T. Kung, G.-Y. Guo, L. Zhou, and D. P. Tsai, High-Efficiency Broadband Anomalous
Reflection by Gradient Meta-Surfaces, Nano Lett. 12, 6223 (2012).
[21] A. Pors, M. G. Nielsen, R. L. Eriksen, and S. I. Bozhevolnyi, Broadband Focusing Flat Mirrors
Based on Plasmonic Gradient Metasurfaces, Nano Lett. 13, 829 (2013).
[22] A. Pors and S. I. Bozhevolnyi, Plasmonic Metasurfaces for Efficient Phase Control in Reflec-
tion, Opt. Express 21, 27438 (2013).
[23] M. Farmahini-Farahani and H. Mosallaei, Birefringent Reflectarray Metasurface for Beam
Engineering in Infrared, Opt. Lett. 38, 462 (2013).
[24] M. Esfandyarpour, E. C. Garnett, Y. Cui, M. D. McGehee, and M. L. Brongersma, Metama-
terial Mirrors in Optoelectronic Devices, Nat. Nanotechnol. 9, 542 (2014).
[25] M. Kim, A. M. H. Wong, and G. V. Eleftheriades, Optical Huygens Metasurfaces with Inde-
pendent Control of the Magnitude and Phase of the Local Reflection Coefficients, Phys. Rev. X
4, 041042 (2014).
[26] M. Veysi, C. Guclu, O. Boyraz, and F. Capolino, Thin Anisotropic Metasurfaces for Simulta-
neous Light Focusing and Polarization Manipulation, Journ. Opt. Soc. Am. B 32, 2 (2015).
[27] Z. Li, E. Palacios, S. Butun, and K. Aydin, Visible-Frequency Metasurfaces for Broadband
Anomalous Reflection and High-Efficiency Spectrum Splitting, Nano Letters 15, 3 (2015).
[28] N. M. Estakhri and A. Alù, Wavefront Transformation with Gradient Metasurfaces, submitted
manuscript.
[29] A. N. Serdyukov, I. V. Semchenko, S. A. Tretyakov, and A. Sihvola, Electromagnetics of Bi-
anisotropic Materials: Theory and Applications (Gordon and Breach Science, Amsterdam,
2001).
[30] H. Steyskal, A. Hessel, and J. Shmoys, On the Gain-Versus-Scan Trade-offs and the Phase
Gradient Synthesis for a Cylindrical Dome Antenna, IEEE Trans. Antennas Propag. 27, 6
(1979).

33
[31] V. S. Asadchy, I. A. Faniayeu, Y. Ra’di, S. A. Khakhomov, I. V. Semchenko, and
S. A. Tretyakov, Broadband Reflectionless Metasheets: Frequency-Selective Transmission and
Perfect Absorption, Phys. Rev. X 5, 031005 (2015).
[32] R. Fleury, D. L. Sounas, and A. Alù, Negative Refraction and Planar Focusing Based on
Parity-Time Symmetric Metasurfaces, Phys. Rev. Lett. 113, 023903 (2014).
[33] Y. Ra’di, D. L. Sounas, A. Alu, and S. A. Tretyakov, Parity-Time-Symmetric Teleportation,
Phys. Rev. B 93, 235427 (2016).
[34] M. Selvanayagam and G. V. Eleftheriades, Experimental Demonstration of Active Electromag-
netic Cloaking, Phys. Rev. X 3, 041011 (2013).
[35] T. B. A. Senior, Combined Resistive and Conductive Sheets, IEEE Trans. Antennas Propag.
3, 5 (1985).
[36] M. Idemen and H. Serbest, Boundary Conditions of the Electromagnetic Field, Electron. Lett.
23, 13 (1987).
[37] E. F. Kuester, M. A. Mohamed, M. Piket-May, and C. L. Holloway, Averaged Transition
Conditions for Electromagnetic Fields at a Metafilm, IEEE Trans. Antennas Propag. 51, 10
(2003).
[38] M. Selvanayagam and G. V. Eleftheriades, Circuit Modeling of Huygens Surfaces, IEEE An-
tennas Wireless Propag. Lett. 12, 99 (2013).
[39] A. Epstein and G. V. Eleftheriades, Arbitrary Power-Conserving Field Transformations with
Passive Lossless Omega-Type Bianisotropic Metasurfaces, arxiv.org/1604.01100v1, 5.04.2016.
[40] J. Vehmas, S. Hrabar, and S. Tretyakov, Omega Transmission Lines with Applications to
Effective Medium Models of Metamaterials, J. Appl. Phys. 115, 134905 (2014).
[41] M. Albooyeh, R. Alaee, C. Rockstuhl, and C. Simovski, Revisiting Substrate-Induced Bian-
isotropy in Metasurfaces, Phys. Rev. B 91, 195304 (2015).
[42] V. S. Asadchy, Y. Ra’di, J. Vehmas, and S. A. Tretyakov, Functional Metamirrors Using
Bianisotropic Elements, Phys. Rev. Lett. 114, 095503 (2015).
[43] Y. Ra’di, V. S. Asadchy, and S. A. Tretyakov, Tailoring Reflections from Thin Composite
Metamirrors, IEEE Trans. Antennas Propag. 62, 7 (2014).
[44] Y. Ra’di, C. R. Simovski, and S. A. Tretyakov, Thin Perfect Absorbers for Electromagnetic
Waves: Theory, Design, and Realizations, Phys. Rev. Applied 3, 037001 (2015).

34
[45] M. Yazdi, M. Albooyeh, R. Alaee, V. Asadchy, N. Komjani, C. Rockstuhl, C. Simovski,
and S. Tretyakov, A Bianisotropic Metasurface with Resonant Asymmetric Absorption, IEEE
Trans. Antennas Propag. 63, 7 (2015).
[46] R. Alaee, M. Albooyeh, M. Yazdi, N. Komjani, C. R. Simovski, F. Lederer, and C. Rockstulh,
Magnetoelectric Coupling in Nonidentical Plasmonic Nanoparticles: Theory and Applications,
Phys. Rev. B 91, 115119 (2015).
[47] S. A. Tretyakov, Analytical Modeling in Applied Electromagnetics, (Artech House, Norwood,
MA, 2003).
[48] See Supplemental Material for additional information on the derivation of collective polariz-
abilities of metasurface unit cells.
[49] R. Alaee, M. Albooyeh, A. Rahimzadegan, M. S. Mirmoosa, Y. Kivshar, and C. Rockstuhl,
All-Dielectric Reciprocal Bianisotropic Nanoparticles, Phys. Rev. B 92, 245130 (2015).
[50] V. Asadchy, M. Albooyeh, and S. Tretyakov, Optical Metamirror: All-Dielectric Frequency-
Selective Mirror with Fully Controllable Reflection Phase, J. Optical Soc. Am. B 33, 2 (2016).
[51] D. Sievenpiper, L. Zhang, R. F. J. Broas, N. G. Alexópolous, and E. Yablonovitch, High-
Impedance Electromagnetic Surfaces with a Forbidden Frequency Band, IEEE Trans. Microw.
Theory Tech. 47, 2059 (1999).
[52] J. Huang and J. A. Encinar, Reflectarray Antennas (Wiley, New Jersey, 2008).
[53] J. C. Vardaxoglou, Frequency Selective Surfaces: Analysis and Design (Research Studies Press,
Taunton, 1997).
[54] A. Epstein and G. V. Eleftheriades, Floquet-Bloch Analysis of Refracting Huygens Metasur-
faces, Phys. Rev. B 90, 235127 (2014).
[55] I. Candezon, G. Granet, and P. N. Melezhik, Modern Theory of Gratings. Resonant Scattering:
Analysis Techniques and Phenomena (Springer, Dordrecht, Heidelberg, 2010).
[56] Y. Hadad, J. C. Soric, and A. Alù, Breaking Temporal Symmetries for Emission and Absorp-
tion, in Proceedings of the National Academy of Sciences (2016), p. 33471.
[57] Y. Hadad, D. L. Sounas, and A. Alù, Space-Time Gradient Metasurfaces, Phys. Rev. B 92,
100304 (2015).
[58] S. Taravati and C. Caloz, Space-Time Modulated Nonreciprocal Mixing, Amplifying and Scan-
ning Leaky-Wave Antenna System, in Proceedings of 2015 IEEE International Symposium
on Antennas and Propagation & USNC/URSI National Radio Science Meeting, (IEEE, New

35
 York, 2015), p. 639.
[59] T. Niemi, A. Karilainen, and S. Tretyakov, Synthesis of Polarization Transformers, IEEE
Trans. Antennas Propagation 61, 6 (2013).
[60] Y. Ra’di, V. S. Asadchy, and S. Tretyakov, Total Absorption of Electromagnetic Waves in
Ultimately Thin Layers, IEEE Trans. Antennas Propagation 61, 9 (2013).
[61] A. Epstein and G. V. Eleftheriades, Synthesis of Passive Lossless Metasurfaces Using Aux-
iliary Fields for Reflectionless Beam Splitting and Perfect Reflection, arxiv.org/1607.02954,
14.06.2016.

36