1

Quadrupolar susceptibility modeling of substrated

metasurfaces with application to the generalized
Brewster effect
Ville Tiukuvaara, Olivier J. F. Martin, and Karim Achouri
Nanophotonics and Metrology Laboratory, Swiss Federal Institute of Technology Lausanne (EPFL),
EPFL-STI-NAM, Station 11, CH-1015 Lausanne, Switzerland
email: [email protected]
arXiv:2302.13835v1 [physics.optics] 27 Feb 2023

Abstract
We derive generalized sheet transition conditions (GSTCs) including dipoles and quadrupoles, using generalized functions
(distributions). This derivation verifies that the GSTCs are valid for metasurfaces in non-homogeneous environments, such as
for practical metasurfaces fabricated on a substrate. The inclusion of quadrupoles and modeling of spatial dispersion provides
additional hyper-susceptibility components which serve as degrees of freedom for wave transformations. We leverage them to
demonstrate a generalized Brewster effect with multiple angles of incidence at which reflection is suppressed, along with an
“anti-Brewster” effect where transmission is suppressed.

I. I NTRODUCTION
The study of metamaterials, and metasurfaces in particular, has reached a level of maturity such that recent works present
increasingly elaborate applications. While at first, metasurfaces were used to provide simple wave transformations [1], [2] and
flat optics [3], they have now been used for sophisticated holography [4] and recently for computation and signal processing
[5], [6], [7], [8], [9]. This last application – where transfer functions are implemented in the Fourier domain – has required
intricate design of meta-atoms to achieve control of the angular scattering response [9], [10], [11].
To aid in the design of metasurfaces for these applications, several modeling techniques have established themselves [12].
One popular approach is to model a metasurface as an impedance sheet which supports electric and magnetic currents [12], [13],
[14]. This induces boundary conditions on the tangential parts of the electric and magnetic fields. However, the impedances
do not provide characteristic parameters to represent the metasurface since they depend on the incident fields [15]. A second
approach is to determine the polarizability of an isolated meta-atom and account for its coupling through the array to other
meta-atoms using Green’s functions in the so-called T-matrix approach [16]. This provides insights into the multipole moments
which are present, and how they couple together. However, it does not serve as a boundary condition; rather, it provides the
scattered fields when the incident field is specified. The last popular approach is the use of surface susceptibilities, which
represent the metasurface as a zero-thickness sheet of multipole moment densities [17], [18], [19], [20], [21]. Given these
moments along a surface, generalized sheet transition conditions (GSTCs) provide boundary conditions on the fields adjacent
to the surface [22]. These have been used to design metasurfaces [23] and also implemented in numerical methods to greatly
decrease the computational resources needed for their analysis [24], [25], [26].
Recently, it has become evident that susceptibility modeling, which was previously limited to the dipolar regime, should
include spatial dispersion (nonlocality) [27]. This was analyzed with “angle-dispersive” dipolar susceptiblities in [28] while
we considered higher-order multipoles in [29], [30]. These considerations are especially true for optical metasurfaces, which
generally have large meta-atoms with dimensions that approach the wavelength. Using GSTCs that were generalized to include
quadrupoles, we demonstrated an improvement in the modeling accuracy [30]. In addition to improving the accuracy, the
additional susceptibility components provide additional degrees of freedom for designing metasurfaces.
However, the derivations in [30] are limited since they assume the media below and above the metasurface to be identical.
Thus, it is not a priori obvious whether they would rigorously apply to practical metasurfaces which are usually fabricated on
a substrate. In this work, we overcome the limitation of [30] by deriving the GSTCs, but using a different approach based on
distributions (generalized functions) [31], inspired by the work of Idemen [32]. Ultimately, our derivation produces GSTCs
identical to those in [30], which demonstrates that the latter can indeed be used in the presence of a substrate.
To demonstrate the utility of these GSTCs, we demonstrate the full control of the Brewster angle, where reflection at a
dielectric interface is suppressed at a particular angle. By placing a metasurface at the dielectric interface, it is possible to
tune the Brewster angle, as shown in [33], [34]. We now leverage the higher-order susceptibility components to show that the
additional degrees of freedom allow for further control, such as multiple Brewster angles, and suppression of transmission at
particular angles—which we call “anti-Brewster” angles.
This paper is outlined as follows. First, we introduce generalized functions and derive the GSTCs in Section II. Next,
Section III presents considerations to enforce the physicality of the analysis: spatial dispersion, properties of the moment
 2

tensors, and spatial symmetries of meta-atoms. Then, several examples of controlling the Brewster and “anti-Brewster” angles
are presented in Section V. Finally, we conclude in Section VI.

II. GSTC S WITH Q UADRUPOLES

In this section, we will generalize the GSTCs to account for quadrupolar moments. Such a derivation was performed in [30],
but with a caveat: the derivation assumed the bulk media adjacent to the metasurface to be homogeneous, and identical on both
sides. This limitation arose from the use of the vector potential of the surface currents. We will overcome this limitation using
an alternative derivation which represents the fields using distributions (generalized functions), following the approach taken
by Idemen [32]. Using this approach, the bulk material properties can be arbitrary as they are embedded in the definitions of
the fields.
Distributions are ideal for modeling metasurfaces as zero-thickness discontinuities, since they formalize the notion of an
“impulse function”. For example, the electric polarization of a flat metasurface in the xy plane may be expressed as P(x, y, z) =
P0 (x, y)δ(z), where δ(z) is the Dirac delta distribution, which in turn is rigorously defined using test functions [32]. More
generally, any field quantity Λ can be decomposed into a continuous part and a discontinuous part:
∞
X
Λ(z) = {Λ(z)} + Λk δ (k) (z) , (1)
k=0

where {Λ(z)} represents the continuous part of Λ(z) and a summation of the Dirac distribution and its derivatives is used
to represent the discontinuity, as in Fig. 1. By interpreting Maxwell’s equations with all field quantities as distributions,
discontinuities in the fields are acceptable and treated rigorously, one arrives at a new set of equations called the universal
boundary conditions [32], [35].
Λ(z)

P∞
k=0 Λk δ (k) (z)
{Λ(z)}

Fig. 1: A arbitrary discontinuous function Λ(z) can be divided into a continuous part {Λ(z)} and the discontinuity expressed
as a summation of Dirac delta functions.

The starting point to derive the GSTCs are the universal boundary conditions and the following relations for the electric and
magnetic flux densities for media with multipolar responses [29], [36]:
1
D = 0 E + P − Q · ∇ (2a)
 2 
1
B = µ0 H + M − S · ∇ , (2b)
2

where P is the electric dipole density, Q is the electric quadrupolar moment density, M is the magnetic dipolar density, and
S is the magnetic quadrupolar density. For a metasurface, each of the quadrupolar moments may be represented by a single
sheet discontinuity (no continuous part and k = 0 from (23)) and so we re-write (2) as
1h i
D = 0 E + {P} + P0 δ (0) (z) − Q0 δ(z)(0) · ∇ (3a)
 2 
1 h i
B = µ0 H + {M} + M0 δ (0) (z) − S 0 δ (0) (z) · ∇ , (3b)
2
where the bulk polarization (with possibly different media on the two sides of the metasurface) is embedded within {P} and
{M} as well as within the fields E, H, D and B. Then, by simplifying (25) and substituting it into the universal boundary
conditions as shown in the supplementary information, one arrives at the following GSTCs:

k02  
z × ∆E = −jωµ0 Mt + ẑ × Q · ẑ
20  
1 1 jωµ0 h  i
− ẑ × ∇t Pz − (∇t ẑ + ẑ∇t ) : Q + S − Szz I · ∇t (4a)
0 2 2 t
 3

k02  
z × ∆H = jωPt + ẑ × S · ẑ
2  
1 jω h  i
− ẑ × ∇t Mz − (∇t ẑ + ẑ∇t ) : S − Q − Qzz I · ∇t , (4b)
2 2 t

where I is the identity matrix and the t subscript indicates the tangential components. These are in agreement with those
derived independently in [30]; however, our derivation proves that the GSTCs are independent of the media on either side
of the metasurface, since there is no restrictions on the bulk moments, {P} and {M}, in (25). Nevertheless, the information
regarding the material parameters of these media remains present in these equations as it is embedded within the definition of the
fields that interact with the metasurface. Additionally, our derivation provides boundary conditions on the normal components
of the fields, which have not yet been shown in the literature:
 
jωµ0 0    
z · ∆D = −∇t · Pt − ẑ × S · ẑ − Q − Qzz I · ∇t (4c)
2
 
jω    
z · ∆B = −∇t · µ0 Mt + ẑ × Q · ẑ − S − Szz I · ∇t . (4d)
2
Naturally, these simplify to the conventional textbook boundary conditions for a dielectric interface when there are no surface
polarization moments, which can be easily verified by setting P = M = Q = S = 0.

III. N ON -L OCAL C ONSIDERATIONS

The GSTCs (4) provide boundary conditions, but do not model how the polarization densities arise. These can be captured
using multipolar susceptibilities. In this section, we formulate constitutive relations, and discuss some considerations that must
be taken when the metasurface is placed in a non-uniform environment.

A. Acting Fields
The fields which excite the meta-atoms are defined as the average of the adjacent fields on either side of the metasurface.
As in [37], [30], [38], these average fields are usually defined as
1
Eav = (Ei + Er + Et )|z=0 (5a)
2
1
Hav = (Hi + Hr + Ht )|z=0 . (5b)
2
This works well if the metasurface is freestanding, but is inappropriate if the metasurface is placed between two different
media. Considering a thin slab placed between two different media, it can be shown (see the supplementary information or
[39]) that the normal components should be defined using the flux densities, Ez and µHz , which remain continuous at a
dielectric interface, such that the acting fields are
1
Eav = Eav,t + (1 Ei,z + 1 Er,z + 2 Et,z )|z=0 ẑ, (6a)
2
1
Hav = Hav,t + (µ1 Hi,z + µ1 Hr,z + µ2 Ht,z )|z=0 ẑ, (6b)
2
where 1 , µ1 and 2 , µ2 correspond to the material parameters of the media at the top and the bottom sides of the metasurface,
respectively.

B. Spatial Dispersion
Given the acting fields (6), the constitutive relations can be written out. If the metasurface response may be described in
terms of only dipolar components (e.g., when the metasurface unit cell is much smaller than the wavelength so that higher-order
multipolar components are negligible), the induced moments are related using [23], [30]
c−1
     
P 0 χee 0 χem Eav
= −1 · , (7a)
M η0 χme χmm Hav
where the constants (i.e. 0 , η0 ) are selected so the surface susceptibilities have the unit of length [38]. However, when the
metasurface unit cell becomes large, higher-order multipoles must be considered, starting with the quadrupoles. Including these
adds a plethora of additional hyper-susceptibility components [30]:
 4

0 0
0 χij 1 ij 0 ijk ijk
 1

ee c0 χem 2k0 χee 2c0 k0 χem
   
Pi  1 ij 0 0 Eav,j
 Mi   η0 χme χij 1
2η0 k0 χme
ijk 1
2k0 χmm 
ijk 
mm  Hav,j 
 =
Qil   0 Qilj ilj 0 0  ·
 ∇k Eav,j  ,
 (8)
 k0 ee
1
c0 k0 Qem
0
Q iljk
2k02 ee
1
Q iljk 
2c0 k02 em 
Sil ilj 1 ilj
0 0 ∇k Hav,j
1
η0 k0 Sme k0 Smm
1
S iljk
2η0 k02 me
1
S iljk
2k02 mm

where in addition to quadrupolar susceptibilities which depend on the fields directly (e.g. Qil ∝ Qilj
ee Eav,j ), there are components
0
ijk
which depend on the field gradients (e.g. Pi ∝ χee ∇k Eav,j ). This spatial dispersion or nonlocality is necessitated by
0
reciprocity, which connects certain terms together (e.g. Qilj jli
ee = χee ) [29]. At this point, for simplicity but without loss of
generality, we will consider TM-polarized plane-wave fields propagating in the xz plane. Then, as shown in [30], (8) simplifies
to
 T
Px Pz My Qxz Qxx Qzz Syz Syx ∝
 T
X · Eav,x Eav,z Hav,y ∂x Eav,x ∂x Eav,z + ∂z Eav,x ∂z Eav,z ∂z Hav,y ∂x Hav,y , (9)

where X is the hypersusceptibilty matrix shown in Fig. 2a. It is an 8 × 8 matrix with 64 terms in general, but can be simplified
by imposing conditions such as reciprocity and tracelessness, as we will do shortly. Before that, consider that it is non-sensical
to include the derivatives ∂z , since the values Eav,x , Eav,z and Hav,y are independent of z, as defined in (6). That is, they
are functions of x and y only, and so the derivative (∂z ) would be zero. However, we can overcome this issue by switching
the order of operations; that is, by performing differentiation first, and then averaging. Furthermore, though the derivatives
along z are still problematic as they may be discontinuous, we can transform them into tangential derivatives using Maxwell’s
equations, as explained next.
First, consider Faraday’s equation in either medium, ∇ × E = jωB, which becomes ∂z Ex = ∂x Ez − jωBy . Then, it follows
that the spatial average of the derivative along z of Ex may be obtained as
1
(∂z Ex )|av =(∂z Ei,x + ∂z Er,x + ∂z Et,x )|z=0
2
= ∂x Eav,z − jωBav,y
jω
= ∂x Eav,z − (µ1 Hi,y + µ1 Hr,y + µ2 Ht,y )|z=0 , (10a)
2
where we have eliminated ∂z . Next, consider Gauss’ equation, ∇ · D = 0 in either medium, that is, ∂z Dz = −∂x Dx . Then,
1
(∂z Ez )|av = −
(1 ∂x Ei,x + 1 ∂x Er,x + 2 ∂x Et,x )|z=0 . (10b)
2
Finally, consider Ampere’s equation, ∇ × H = jωD; that is, ∂z Hy = ∂y Hz − jωDx .
1
(∂z Hy )|av = (∂z Hi,y + ∂z Hr,y + ∂z Ht,y )|z=0
2
= ∂y Hav,z − jωDav,x
jω
= ∂y Hav,z − (1 Ei,x + 1 Er,x + 2 Et,x )|z=0 . (10c)
2
Now, with reference to (10), (9) is modified to
 T
Px Pz My Qxz Qxx Qzz Syz Syx ∝
 T
X · Eav,x Eav,z Hav,y ∂x Eav,x ∂x Eav,z + (∂z Ex )|av (∂z Ez )|av (∂z Hy )|av ∂x Hav,y . (11)

C. Tensor symmetries, tracelessness, and reciprocity

Linear time-invariant metasurfaces that are not biased by a time-odd quantity (such as a magnetic field) are reciprocal and
as such must satisfy reciprocity conditions [40], [41]. Furthermore, we note that all of the moments should be symmetrical;
e.g. Qij = Qji [42]. These symmetry and reciprocity properties of the moments constrain the susceptibility terms such that
they are related to one another as shown in [29]. Enforcing these relations reduces the 64 terms in Fig. 2a to 36.
In addition to being symmetrical, the tensors should be traceless.P Only if this is the case are the moments truly physically
meaningful, and are called irreducible [42]. In particular, note that i Qii = 0, which implies Qzz = −Qxx given that Qyy = 0
in our simplified problem. In contradiction to this condition, Qzz and Qxx are independent in (11).
To enforce tracelessness, note how the 6th row in Fig. 2a should be the negative of the 5th row (Qzz = −Qxx ), and can thus
be eliminated. However, reciprocity must still be enforced: the 5th row and 5th column are related by reciprocity and tensor
 5

symmetries. Given the relationship between the 4th and 5th rows, reciprocity is maintained by rewriting the 5th column as
the negative of the 4th column. Then, the 5th row can be eliminated and we arrive at the 7 × 8 matrix in Fig. 2b. This matrix
ensures tensor symmetries, reciprocity, and traceless, and contains 28 unique terms. After eliminating Qzz , (11) becomes
 T
Px Pz My Qxz Qxx Syz Syx ∝
 T
X · Eav,x Eav,z Hav,y ∂x Eav,x ∂x Eav,z + (∂z Ex )|av (∂z Ez )|av (∂z Hy )|av ∂x Hav,y . (12)

D. Spatial Symmetries
Neumann’s principle states that the material parameters of a system should exhibit the same symmetry properties as the
physical structure they describe. This implies that if the considered physical structure (metasurface) is invariant under certain
symmetry operations, then so should their material parameters (susceptibility tensors) [43], [44].
For example, consider the metasurface in Fig. 3a, with all possible symmetries: reflection (σx , σy , σz ) and rotation (C4,z ).
These symmetries can be used to write invariance conditions on the susceptibility tensors in (8) which eliminate incongruous
components. The invariance relations are given in [44] along with an algorithm to easily apply them. Following this algorithm,
the hypersusceptiblity matrix reduces to the 9 terms in Fig. 2c. Furthermore, if the unit cell is deeply subwavelength (p  λ1 ),
yy
then higher-order susceptibilities will be negligible such that the surface can be described using only χxx zz
ee , χmm , and χee . Also,
zz xx yy
if the metasurface is very thin, as in 3b, then χee may be negligible, such that only χee and χmm are necessary.
However, optical meta-atoms are generally large, such that the dipolar model is inappropriate [30]. Then, quadrupolar
susceptibilities are necessary, and these provide additional degrees of freedom for specifying wave transformations. To provide
even more additional degrees of freedom, spatial symmetries can be broken. For example, consider breaking σz symmetry, as
is the case for the meta-atom in Fig. 3c. By following the algorithm in [44], one arrives at the matrix with 14 terms in Fig. 2d.
This matrix allows for bianisotropy (e.g. χxy em ), and will be used later to demonstrate the utility of the additional degrees of
freedom for wave transformations.

IV. S YNTHESIS AND S CATTERING A NALYSIS

Given (4a-b) and (11), it is possible to calculate the fields that will be scattered from a metasurface. In this section, we
will consider how plane waves are scattered for a metasurface and how to engineer the susceptibilities to control the angular
scattering behaviour.
We will consider TM-polarized plane waves propgagting in the xz plane. With reference Fig. 4, we express the fields as
Ha = Sa H0 ŷe−jka ·r (13a)
ηa
Ea = Ha × ka (13b)
ka
1λ is the wavelength in the background, or the shorter of the two wavelengths in the top or bottom media.

0 0
 0 0 0

χxx χxz χxy
em χee
xxz
χeexxx χeexzz χem xyx
χem xyz
0 0 0 0 0
 ee ee
χxy xyx xyz
 
0 0 0 0 0 χxx χxz em χeexxz χeexxx −χeexxx χem χem
 χzx
ee χzz
ee χzy
em χee
zxz
χeezxx χeezzz χem zyx
χem zyz 
  xz
ee ee
zz zy 0
zxz 0
zxx 0
zxx
0
zyx 0
zyz 
 yx
 χme χyz yy 0
yxz 0
yxx 0
yzz 0
yyx 0
yyz 
  χee χee χem χee χee −χee χem χem 
me χmm χme χme χme χmm χmm  0 0 0 0 0
 −χxy zy
χyy yxz yxx yxx yyx yyz 

 0 em −χ χme χme −χme χmm χmm

 xzx xzz xzy 0 xzxz 0 xzxx 0 xzzz 0 xzyx 0 xzyz  em mm 
Qee Qee Qem Qee Q Q Qem Qem  0 0
yxz 0 0 0 0
yxzx 0 
yzzx  (b)
 xxx xxz xxy 0 xxxz 0eexxxx 0eexxzz 0 xxyx 0 xxyz   χ xxz χ zxz −χme
 ee ee Qeexzxz
Qeexxxz
−Qeexxxz
−Sme −Sme
Qee Qee Qem Qee Qee Qee Qem Qem   0 xxx zxx 0
yxx 0 xxxz 0 0 0
yxxx 0

yzxx 
xxxx xxxx
 Qzzx Qzzz Qzzy Q0 zzxz Q0 zzxx Q0 zzzz Q0 zzyx Q0 zzyz 
   χee χ −χme Qee Qee −Qee −Sme −Sme 
 ee ee em ee ee ee em em  xyx eezyx yyx 0
yxzx 0
yxxx 0
yxxx 0
yxyx 0
yxyz 
 yxx yxz yxy yxxz yxxx yxzz yxyx 0 yxyz 
0 0 0 0
 −χem −χem χmm Sme Sme −Sme Smm Smm 
 Sme Sme Smm Sme Sme Sme Smm Smm  xyz zyz yyz 0
yzzx 0
yzxx 0
yzxx 0
yxyz 0
yzyz
yzx yzz yzy 0
yzxz 0
yzxx 0
yzzz 0
yzyx 0
yzyz −χem −χem χmm Sme Sme −Sme Smm Smm
Sme Sme Smm Sme Sme Sme Smm Smm

(a) Full susceptibility matrix (8 × 8) Reciprocal & traceless (7 × 8)

 0   0 0 
χxx 0 0 0 0 0 0 χem xyz
χxx
ee 0 χxy
em χeexxz 0 0 0 χem xyz
 0ee χzz 0 0
0 0 0 0 0 0  0 χzz 0 0 χeezxx −χeezxx 0 0 
  
 ee  ee
0 0 0
 yy yxz   xy yy yxz yyz 
 0 0 χmm χme 0 0 0 0   −χem 0 χmm χme 0 0 0 χmm 
0 0
 χ xxz 0 −χ0me 0
(d) All
 yxz
  0 yxz 0

yzzx 
xzxz
 0
 0 −χme Qee 0 0 0 0 
  ee Qeexzxz 0 0 0 −Sme 
0 0 0 0
xxxx xxxx  0 χzxx xxxx xxxx
 0
 0 0 0 Qee −Qee 0 0 
  ee 0 0 Qee −Qee 0 0  
0 0
yxyx yxyx
 0 0 0 0 0 0 Smm 0 
   
 0 0 0 0 0 0 Smm 0 
0 0 0
−χxyz
em 0 0 0 0 0 0 Smm
yzyz
−χxyz
em 0 χ yyz
mm Sme
yzzx
0 0 0 Smm
yzyz

(c) All symmetries: σx , σy , σz , C4z symmetries but σz

Fig. 2: The general susceptibility matrix X, for TM-polarized fields propagating in the xz plane, is shown in (a) and has 64
terms. By enforcing reciprocity and tracelenssness, this is reduced to 28 terms as in (b). Subsequently, spatial symmetries of
the metasurface can be leveraged to further simplify the matrix, as shown for two examples in (c) and (d).
 6

z
p  λ0
p  λ0
p < λ0

x y

yy yy yy xy yzzx
(a) χxx zz
ee , χmm , and χee (b) χxx
ee and χmm (c) χxx
ee , χmm , χem , Sme , ...

Fig. 3: Possible unit cells that have given spatial symmetries: (a) and (b) have all the structural symmetries as in Fig. 2c but are
deeply subwavelength such that quadrupolar responses are negligible. Since the height of the particles in (b) is negligible, the
normal response χzzee is also negligible in this case. In (c), σz symmetry is broken, corresponding to Fig. 2d, and is furthermore
only slightly subwavelength, meaning that quadrupolar responses are possible.

k
r = 1 , µ1 , η1 , k1 2 , µ2 , η2 , k2 1 , µ1 , η1 , k1 2 , µ2 , η2 , k2 Ei
[k
x 0 Hi
−
k metasurface S12
z1 ] T T T
θr ] θi ]
z2 k z2
0k
Et −
Hr [k x 0
= kx
kt Ht [−
Er θt = S22
T ki
S11 ] T
k z1 Ht θt ] Er
0 k z1
− Hr
[k x Et 0
= kx
ki θi [− k
S21 = r = θr
kt [−
x x k
Hi x 0
k
Ei z2 ] T
z z
y y

(a) Forward illumination (b) Backward illumination

Fig. 4: A depiction of the TM-polarized plane waves incident along the xz plane onto a metasurface placed at an interface
between two different media at z = 0. The situations for both forward and backward illumination are depicted, with the
incident field approaching from the first and second media, respectively.

with a = i for the incident field (fields are normalized with Si = 1), a = r for the reflected field (Sr = −S11 , the reflection
coefficient), and a = t for the transmitted field (St = S21 , the transmission coefficient). For backwards illumination, one
replaces 1 ⇐⇒ 2 along with kx → −kx , and kz,{1,2} → −kz,{2,1} . This is shown in Fig. 4.
Now, these fields can be substituted into the GSTCs (4a-b) and the constitutive relations (11). In the case of forward
illumination, this provides equations to solve for the two unknowns S11 and S21 , and for backwards illumination one can solve
for S22 and S21 . However, the expressions for these S-parameters are very unwieldy, and so we will limit the analysis to some
yy
of the terms from Fig. 3. Considering the dipolar susceptibilities χxx zz
ee , χmm , χee , then
 
2 yy


kz,{1,2} α + kz,{2,1} 4jkz,{1,2} χxx 2 2 2 zz
ee + n{1,2} α + 4jn{1,2} kx χee + k0 χmm
S{11,22} (kx ) = −   (14a)
kz,{1,2} α + kz,{2,1} 4jkz,{1,2} χxx 2 2 2 zz 2 yy
ee + n{1,2} α + 4jn{1,2} (kx χee − k0 χmm )
2kz,{1,2} n1 n2 α
S{21,12} (kx ) =   (14b)
4jkz,{1,2} χxx 2 2 2 zz 2 yy
kz,{1,2} α + kz,{2,1} ee + n{1,2} α + 4jn{1,2} (kx χee − k0 χmm )

α = kx2 χxx zz 2 xx yy
ee χee + k0 χee χmm − 4 , (14c)
with the first subscripts selected for forward illumination (S11 and S21 ) and the second for backward illumination (S22 and
S12 ).
Meanwhile, for use later, we will also derive expressions for the scattering with χxy yzzx
em , Sme . We find

∓kz,1 η22 β ± ± kz,2 η12 β ∓

S{11,22} (kx ) = (15a)
kz,1 η2 β ± + kz,2 η1 β ∓
 7

Ei Ei,1 Er,1 6= 0
Er = 0
Ei,2 Er,2 = 0
θi,1 θr,1 θi,1 θr,1 θi,2 θr,2
PMS PMS

θt,1 θt,1 θt,2

Pbulk (k Er ) Pbulk Pbulk
Et Et,1
Et,2

(a) (b) (c)

Fig. 5: For ordinary dielectric media, the Brewster angle occurs when the reflected and refracted rays are orthogonal and thus
prohibit any reflection, as in (a). If a metasurface is introduced, the additional polarization at the surface contributes to the
reflected field and thus they can be non-zero, as in (b). However, the Brewster angle now can occur at other angle(s), when
the net scattering due to the substrate and metasurface interfere destructively, as in (c).

yzzx 2 yzzx
4kx4 (Sme ) + k04 (Sme + 4χxy 2
 
em )
2n1 n2 kz,{1,2} 2 2 yzzx yzzx xy

− 4k0 kx Sme (Sme + 4χem ) − 16
S{21,12} (kx ) = (15b)
kz,1 η2 β ± + kz,2 η1 β ∓
2
β ± = 8jk0 ± 2kx2 Sme
yzzx
∓ k02 (Sme
yzzx
+ 4χxy

em ) , (15c)
where the top sign is selected for forward illumination (S11 , S21 ) and the bottom sign is selected for backward illumination
(S22 , S12 ).
Given these expressions, one can solve for the susceptibilities required to control the angular scattering behaviour. For
0 0
example, to suppress reflection at some angle kx , one needs to solve |S11 (kx )| = 0.

V. I LLUSTRATION : T UNING B REWSTER ’ S A NGLE

The Brewster angle is defined as the angle of incidence of a plane wave at a dielectric interface where reflection is eliminated;
i.e., there is complete transmission. For ordinary (non-magnetic) materials, it only occurs for TM polarization, and can be
understood intuitively from Fig. 5a. When the angle of incidence is θi,1 and the angle of refraction is θt,1 , the wave vectors of
the refracted and reflected fields are orthogonal. Consequently, the bulk electric polarization in the substrate is orthogonal to
the electric field of the reflected field and it is impossible for the bulk polarization to produce a reflected field. This Brewster
angle θi = θB occurs at [45]
r
−1 2
θB = tan , (16)
1
which corresponds to kx = kB = k1 sin θB .
Using a metasurface, it is possible to tune θB [34]. In Fig. 5b, a metasurface has been added that has a surface polarization
which has a non-zero projection onto the reflected electric field at θi,1 and thus there reflection is no longer suppressed.
However, now there can be another angle θi,2 where the superposition of the scattered fields from the bulk and the metasurface
result in a suppression of the scattered fields, as in Fig. 5c. By adjusting the metasurface, this angle can be tuned. In fact, we
will show that it is possible to have multiple such Brewster angles. There is one caveat: metasurfaces are generally resonant
and thus limited in bandwidth unlike the broadband dielectric Brewster effect given by (16).
In this section, we will illustrate the use of the additional susceptibility components from Fig. 2d to achieve this generalized
Brewster effect.

A. Generalized Brewster Effect

yy
Consider the very simple case of a Huygen’s metasurface with χxx ee and χmm . Using these two terms, it is possible to achieve
a single wave transformation [23], which we would desire to be the suppression of reflection at a given angle of incidence.
This corresponds to impedance matching the two media at a given angle. We will ensure that the metasurface is lossless, which
will ensure that all power is transmitted. Setting |S21 | = 0 in (15) and solving for χxx
ee , we find

4η0 (k2 kz1 η1 − k1 kz2 η2 − jk0 k1 k2 χyy

mm )
χxx
ee = , (17)
k0 [k0 χyy
mm η0 (k2 kz1 η1 − k1 kz2 η2 ) − 4jη1 η2 kz1 kz2 ]

which shows that χxx

ee must in general be complex, implying a metasurface with loss and gain, as was noted in [34]. However,
yy
we will show that it is possible to avoid the need for loss or gain. In order for the metasurface to be lossless, χxx
ee and χmm
 8

={χxx −5 m) <{χxx −3 m)
ee } (×10 ee } (×10
4 10 4 1
3 3 1 |S11 |2

=0 (cos θ2 / cos θ1 )|S21 |2

2 xx } 5 2 0.75
−4 m)

−4 m)
χ ee
={ 0.5

Scattered power
1 1

kx = −0.6k0

kx = +0.6k0
mm } (×10

mm } (×10
0 0 0 0 0.5
-1 -1
<{χyy

<{χyy
-0.5
-2 -5 -2 0.25
-3 -3 -1
-4 -10 -4 0
-0.5

-0.5

-0.5
0

1
-1

0.5

-1

0.5

-1

0.5
kx /k0 kx /k0 kx /k0
(a) (b) (c)
={χzz −5 m) <{χzz −3 m)
ee } (×10 ee } (×10
1.5 10 1.5 1
kx = −0.6k0

kx = +0.6k0
6
1 1
4
5 0.75
−3 m)

−3 m)

={χzz
ee } = 0

Scattered power
0.5 0.5 2
ee } (×10

ee } (×10

0 0 0 0 0.5
<{χxx

<{χxx

-0.5 -0.5 -2
-5
-4
0.25
-1 -1
-6
-1.5 -10 -1.5 0
-0.5

-0.5

-0.5
0

1
-1

0.5

-1

0.5

-1

0.5
kx /k0 kx /k0

(d) (e) (f)

Fig. 6: Brewster angle control using dipolar susceptibilites, considering an interface with two media having 1 = 1 and 2 = 2
and at λ0 = c0 /(300 THz). (a-b) A depiction of the real of imaginary part of χxx ee which satisfy (17) given a desired kB
and χyy
mm ∈ R. (c) Reflected and transmitted power in the case of χxx
ee = 4.44 × 10 −4
m and χyy
mm = 2.28 × 10
−4
m, where
−4 −4
kx,B = 0.6k0 . (a-c) The same plots, but considering χxx ee and χ zz
ee , with χ xx
ee = 4.44 × 10 m and χzz
ee = 6.34 × 10 m in
(f).

must both be purely real. To this end, we will set χyy xx

mm to be a purely real number and then find χee from (17) which are also
purely real.
yy
Figure 6a shows ={χxx ee } plotted as a function of the angle of incidence (kx /k0 ) on the x-axis and χmm on the y-axis, for
a particular case where a plane wave is incident from air (1 = 1) onto a substrate (2 = 2). Consider in particular the black
contours along which ={χxx ee } = 0 Choosing a point along these contours corresponds to a lossless and gainless metasurface.
Next, Fig. 6b shows the same contours superimposed on a plot of <{χxx ee }. We see that the contours cover all incident angles
(−k0 < kx < k0 ). Thus, one can choose any angle of incidence and the necessary susceptibilities to achieve a Brewster effect
at the given angle. For example, to achieve kB = 0.6k0 , χxxee = 4.44 × 10
−4
m and χyymm = 2.28 × 10
−4
m. The corresponding
reflected and transmitted power is plotted in Fig. 6c, corroborating a Brewster angle at kB = 0.6k0 .2 To practically realize
such a metasurface, one could use the cell in Fig. 3b.
Now, we will consider other susceptibility terms to demonstrate how the additional degrees of freedom provide more control
2
over the Brewster effect. Motivated by the scattering caused by χzz zz 2 xx
ee , which is proportional to sin (θB )·χee (unlike cos (θB )·χee
yy zz xx
and 1 · χmm ), which can be deduced from (14), let us consider the combination of χee and χee . This could be achieved using
a meta-atom like that in Fig. 3a.3

the transmitted power, an angle-dependent factor is used in (cos θ2 / cos θ1 )|S21 |2 to project the Poynting vector to ẑ [38].
2 For

χyy
3 Though
mm may in general be present as well, it will typically have a Lorentzian wavelength dependence such that the metasurface can be designed to
operate at a frequency where it is negligible [38].
 9

In this case, setting |S21 | = 0 in (15) and solving for χzz

ee , we find

4k0 (k2 kz1 η0 η1 − k1 kz2 η0 η2 + jk0 kz1 kz2 η1 η2 χxx

ee )
χzz
ee = (18)
kx2 η0 [4jk1 k2 η0 + k0 (k2 kz1 η1 − k1 kz2 η2 )χxx
ee ]
for which the real and imaginary parts are plotted in Fig. 6d and e. In this case, designing for kB = 0.6k0 results in the
reflection plotted in Fig. 6f. Compared to the use of χyy xx zz
mm and χee , the use of χee results in a much sharper reflection minimum
around the Brewster angle.
Finally, we consider the use of quadrupolar susceptibilities. Using the meta-atom of Fig. 3c, many terms are possible. To
yzzx
demonstrate the versatility of these terms, we will apply two in particular: Sme and χxy 4
em . Once again, setting |S21 | = 0 in
xy
(15) and solving for χem , one finds

yzzx √ √
2
xy 2kx2 − k02 Sme k2 kz1 η1 ± k1 kz2 η2
χem = − (19)
4k02 2jk0 (k2 kz1 η1 − k1 kz2 η2 )
where, noting that the last term is imaginary, then χxy yzzx
em will be imaginary if Sme is imaginary. This corresponds to a lossless
metasurface [29], and so we have two solutions. These are plotted in Fig. 7a and b. To design for kx,B = 0.6k0 , we first
arbitrarily select χxy
em = 2j · 10
−5
, which is indicated with black contours. Then, Smeyzzx
is selected using the first solution
in Fig. 7a. However, there are two more angles predicted by the second solution, resulting into three Brewster angles. The
4 We assume all other terms but these two are negligible. We leave the design of such a metasurface, which is not a trivial task, as future work.

={χxy
em } (×10
−4 ) ={χxy
em } (×10
−4 )

1 1 1

kx = −0.522k0

kx = +0.522k0
2
kx = −0.6k0

kx = +0.6k0

0.75 0.75
} (×10−3 )

} (×10−3 )

0.5 0.5 0.5

1
xy xy
0.25 χem = 2j · 10−5 m 0.25 χem = 2j · 10−5 m

0 0 0 0
yzzx

yzzx

kx = −0.847k0

kx = +0.847k0
={Sme

={Sme

-0.25 -0.25
-1
-0.5 -0.5 -0.5

-0.75 -2 -0.75
-1 -1 -1
-0.5

-0.5
0.25

0.75

0.25

0.75
0

1
-0.75

-0.25

-0.75

-0.25
-1

0.5

-1

kx /k0 kx /k0 0.5

(a) (b)

1 1

|S11 |2

0.75 (cos θ2 / cos θ1 )|S21 |2 0.75

Scattered power

Scattered power

0.5 0.5

0.25 0.25

0
0.59

0.61
0.6
0.595

0.605

0
0

1
0.2

0.4

0.6

0.8

kx /k0 kx /k0

(c) (d)
Fig. 7: By using quadrupolar susceptibilities, multiple Brewester angles are predicted, with (22) having two solutions for χxy em
plotted in (a) and (b). Both susceptibilities χxy yzzx
em and Sme are purely imaginary; i.e. lossless, and the black contours correspond
to χxy
em = 2.00j×10
−5
. Next, selecting Smeyzzx
= −0.285j×10−3 m and χxy em = 2.00j×10
−5
m, the transmitted and reflected
power is plotted in (c) and (d). Note: 1 = 1, 2 = 2 and λ0 = c0 /(300 THz).
 10

scattered power is plotted in Fig. 7c and d, corroborating the presence of three Brewster angles. Note that the minimum in
reflection at 0.6k0 is very sharp.

B. Engineered Angular Reflection (“Anti-Brewster”)

In addition to suppressing reflection at a particular angle to control the Brewster angle, it is possible to suppress transmission
to create what we will call an “anti-Brewster” angle. We will consider the same sets of susceptibilities as in Section V. Starting
yy
with χxx
ee and χmm in (15), setting |S21 | = 0, and solving for the susceptibilities,
4
χxx
ee = − , (20)
k02 χyy
mm

which has no dependence on the angle of incidence (kx , kz,1 , or kz,2 ). Thus, if (20) is satisfied, the metasurface will behave
as a mirror with complete reflection, rather than the desired refection at a particular angle of incidence.
Thus, we again consider χzz xx
ee , due to its angular behaviour, along with χee . Then, the condition for complete reflection is
4
χxx
ee = − , (21)
kx2 χzz
ee
which has a dependence on kx . Designing for kx = 0.6k0 , Fig. 8a shows the reflected and transmitted powers, verifying the
“anti-Brewster” behaviour.

1 1

|S11 |2
0.75 0.75
(cos θ2 / cos θ1 )|S21 |2
Scattered power

Scattered power

0.5 0.5

0.25 0.25

0 0
0.58

0.59

0.61

0.62
0.6
0

1
0.2

0.4

0.6

0.8

kx /k0 kx /k0
(a) (b)
={χxy
em } (×10
−4 ) ={χxy
em } (×10
−4 )

1 1 -0.28
kx = 0.598k0

kx = 0.602k0
2 2
0.75 0.75
kx = −0.598k0

kx = −0.602k0
kx = +0.598k0

kx = +0.602k0

-0.282
} (×10−3 )

} (×10−3 )

} (×10−3 )

0.5 0.5
1 1
0.25 0.25 -0.284
0 0 0 0
yzzx

yzzx

yzzx
={Sme

={Sme

={Sme

-0.25 -0.25 -0.286

-1 -1
-0.5 -0.5
-0.288
-0.75 -2 -0.75 -2
-1 -1 -0.29
-0.5

-0.5
0

1
-1

0.5

-1

0.5

0.6

kx /k0 kx /k0 kx /k0

(c) First solution to (22) (d) Second solution to (22) (e)

−4 −4
Fig. 8: (a) An anti-Brewster angle is engineered at kx = 0.6k0 using χxx ee = −4.44 × 10 m and χzz
ee = 6.34 × 10 m,
predicted using (21). Next, (b-e) shows two solutions of (22), predicting two anti-Brewster angles. These are plotted in (b),
yzzx
for a case where Sme = −0.285j×10−3 m and χxy em = 2.00j×10
−5
m. Note: 1 = 1, 2 = 2 and λ0 = c0 /(300 THz).
 11

yzzx
Finally, we consider quadrupolar susceptibilities. With Sme and χxy
em , there are two solutions for suppressed transmission:
2 2

yzzx
2kx − k0 Sme 1
χxy
em = ± , (22)
4k02 2jk0
which are plotted in Fig. 8c and d. The two solutions are very close together, as seen in the magnified plot of Fig. 8e. Using
the same susceptibilities as in Fig. 7, the reflected and transmitted powers are plotted in Fig. 8b. We see that the very sharp
Brewster angle is straddled by two close “anti-Brewster” angles. Thus, this combination of susceptibilities allows for 3 Brewster
angles and 2 “anti-Brewster” angles.
Overall, we see that by adding more susceptibility terms – and terms relating to quadrupoles and spatial disperion in particular
– it is possible to have increasing control over the angular scattering response. While we have highlighted a few of the possible
terms in the general hypersusceptibity matrix (8), other susceptibilities could be considered for even more intricate control,
such as more Brewster or “anti-Brewster” angles.

VI. C ONCLUSION
In summary, we have derived GSTCs which include spatial dispersion and are valid for metasurfaces in non-homogeneous
environments, such as for practical metasurfaces fabricated on a substrate. We have shown how the susceptiblity tensor properties
(symmetries, reciprocity, tracelessness) and spatial symmetries of the metasurface can be used to simplify the susceptibility
tensors. Furthermore, we demonstrated how the new hyper-susceptibility terms can be used to produce multiple Brewster and
“anti-Brewster” angles. For example, with tuning of χxy em and Sme
yzzx
it is possible to achieve 3 Brewster angles and 2 “anti-
Brewster” angles. We expect this work to provide a fundamental advance for Fourier-domain signal processing, where tuning
of the angular response is paramount.

VII. F UNDING
We gratefully acknowledge funding from the Swiss National Science Foundation (project PZ00P2 193221).

VIII. S UPPLEMENTARY M ATERIAL

A. Derivation of quadrupolar GSTCs
Maxwell’s equations in their usual form are only valid for continuous regions and fields. For example, they can be applied
inside a dielectric sphere (with appropriate boundary conditions), or outside the sphere, but not in a region which encloses
the dielectric interface. Doing so can lead to physical inconsistencies [32]. However, Idemen showed that by interpretting
Maxwell’s equations using distributions, they become valid despite the presence of discontinuities. Considering a discontinuity
at z = 0, all fields quantities Λ(z) take the form
∞
X
Λ(z) = {Λ(z)} + Λk δ (k) (z), (23)
k=0

as described in the main text. For example,

∞
X
D = {D} + Dk δ (k) (z) (24a)
k=0
X∞
Q = {Q} + Qk δ (k) (z) etc... . (24b)
k=0

However, typically k is limited, and we will only keep k = 0 for the moment densities, corresponding to a single layer [32].
Then, as noted in the main text,
∞
X 1h i
D = 0 {E} + 0 Ek δ (k) + P0 δ (0) (z) − Q0 δ(z)(0) · ∇ (25a)
2
k=0
∞
!
X
(k) (0) 1h (0)
i
B = µ0 {H} + 0 Hk δ + M0 δ (z) − S 0 δ (z) · ∇ . (25b)
2
k=0

Now, we will expand the gradient on the right, using the chain rule:
h i    
Q0 δ(z) · ∇ = Q0 · ∇t δ (0) (z) + Q0 · ẑ δ (1) (z) (26a)
h i    
S 0 δ(z) · ∇ = S 0 · ∇t δ (0) (z) + S 0 · ẑ δ (1) (z) , (26b)
 12

 
where ∇t = ∂x ∂y 0 denotes the tangential gradient. Thus, we have
1 h    i
D = 0 {E} + (0 E0 + P0 )δ (0) (z) − Q0 · ∇t δ (0) (z) + Q0 · ẑ δ (1) (z) (27a)
 2
1 h    i
(0) (0) (1)
B = µ0 {H} + (H0 + M0 )δ (z) − S 0 · ∇t δ (z) + S 0 · ẑ δ (z) , (27b)
2
where we see the the z components of the quadrupolar moments for k = 0 have changed order to k = 1 due to the
differentiation.

B. Universal Boundary Conditions

Taking all quantities in Maxwell’s equations to be in the form of (23), and separating the different orders, one arrives at a
set of universal boundary conditions. For a discontinuity at z = 0, these are expressed [32], [35]
ẑ × ∆H = −∇ × H0 − jωD0 + J0 (28a)
ẑ × ∆E = −∇ × E0 + jωB0 (28b)
ẑ · ∆D = ρ0 − ∇ · D0 (28c)
ẑ · ∆J = ρ0 − ∇ · D0 . (28d)
The singular terms associated with a source (J0 and ρ0 ) would be known beforehand, but the singular terms in the fields (e.g.
E0 ) are unknown and can be found using compatibility relations:
∇ × Hk + ∇ × Hk−1 + jωDk = Jk (29a)
∇ × Ek + ∇ × Ek−1 − jωBk = 0 (29b)
∇ · Dk + ∇ · Dk−1 = ρk (29c)
∇ · Bk + ∇ · Bk−1 = 0 (29d)
∇ · Jk + ∇ · Jk−1 − jωρk = 0 , (29e)
for k ≥ 1. These provide an infinite set of equations. To make the problem tractable, one assumes that the orders above a
certain k are negligible.

C. Iterative substitution into the compatibility relations

In (25), we assumed that only k = 0 is present for the moments, and now we will assume that the terms k ≥ 2 are zero for
the fields. Then, re-expressing (27), we have
1h i
D1,t = P1,t − Q0 · ẑ (30a)
 2 t 
1 h i
B1,t = µ0 M1,t − S 0 · ẑ (30b)
2 t
1 h i
H1,z = ẑ · S 0 · ẑ − M1,z (30c)
2 
1 1 h i
E1,z = ẑ · Q0 · ẑ − P1,z . (30d)
0 2
Substituting (30) into (29) with k = 1,
1 h  i jωµ 
0

ẑ × E0 = −∇ × E1 − jωB1 = ẑ × ∇t ẑ · Q0 · ẑ + S 0 · ẑ (31a)
20 2 t
1 h  i jω  
ẑ × H0 = −∇ × H1 + jωD1 = ẑ × ∇t ẑ · S 0 · ẑ − Q0 · ẑ (31b)
2 2 t
1  
ẑ · D0 = −∇t · D1 = ∇t · Q0 · ẑ (31c)
2 t
µ0 
ẑ · B0 = −∇t · B1 = ∇t · S 0 · ẑ , (31d)
2 t

where we have made use of the identity ∇ × [A(x, y)ẑ] = −ẑ × ∇t A(x, y).
 13

Solving (31) for the tangential and normal parts of the fields,
1 h  i jωµ
0
 
E0,t = −ẑ × (ẑ × E0 ) = ∇t ẑ · Q0 · ẑ − ẑ × S 0 · ẑ (32a)
20 2 t
1 h  i jω  
H0,t = −ẑ × (ẑ × H0 ) = ∇t ẑ · S 0 · ẑ + ẑ × Q0 · ẑ (32b)
2 2 t
1  
D0,z = ∇t · Q0 · ẑ (32c)
2 t
µ0  
B0,z = ∇t · S 0 · ẑ . (32d)
2 t
Thus, from (27) with k = 0,
1h i
D0,t = 0 E0,t + P0,t − Q0 · ∇
2 t
1 h  i jωµ 
0 0
  1h i
= P0,t + ∇t ẑ · Q0 · ẑ − ẑ × S 0 · ẑ − Q0 · ∇ (33a)
2 2 t 2 t

 i
1h
B0,t = µ0 H0,t + M0,t − S0 · ∇
2 t
 i
1 h  i jω   1h
= µ0 M0,t + ∇t ẑ · S 0 · ẑ + ẑ × Q0 · ẑ − S0 · ∇ (33b)
2 2 t 2 t

1 1 h i 1   1 h i
H0,z = B0,z + ẑ · S 0 · ∇ − M0,z = ∇t · S 0 · ẑ + ẑ · S 0 · ∇ − M0,z (33c)
µ0 2 2 t 2
   
1 1 h i 1 1   1 h i
E0,z = D0,z + ẑ · Q0 · ∇ − P0,z = ∇t · Q0 · ẑ + ẑ · Q0 · ∇ − P0,z , (33d)
0 2 0 2 t 2
which, with some manipulation, can be re-written
1 jωµ0 0   1h i
D0,t = P0,t + ∇t Q0,zz − ẑ × S 0 · ẑ − Q0 · ∇ (34a)
 2 2 t 2 t
1 jω   1h i
B0,t = µ0 M0,t + ∇t S0,zz + ẑ × Q0 · ẑ − S0 · ∇ (34b)
2 2 t 2 t
1
H0,z = (∇t ẑ + ẑ∇) : S 0 − M0,z (34c)
2 
1 1
E0,z = (∇t ẑ + ẑ∇) : Q0 − P0,z . (34d)
0 2

D. Substitution into the universal boundary conditions

Finally, we are ready to substitute into the universal boundary conditions, (28).
 
1 1
[[z × E]] = ẑ × ∇t E0,z − jωB0,t = ẑ × ∇t (∇t ẑ + ẑ∇) : Q − Pz
0 2
 i
1 jω   1h
− jωµ0 Mt − ∇t Szz + ẑ × Q · ẑ − S·∇ (35a)
2 2 t 2 t

 
1
[[z × H]] = ẑ × ∇t H0,z + jωD0,t = ẑ × ∇t (∇t ẑ + ẑ∇) : S − Mz
2
 i
1 jωµ0 0   1h
+ jω Pt + ∇t Qzz − ẑ × S · ẑ − Q·∇ . (35b)
2 2 t 2 t

Note that we have dropped the ‘0’ subscripts (e.g. Q0 → Q) since there is only one term in these series. Given that ω 2 µ0 0 = k02
and ω 2 µ0 = k02 /0 , (35) can be re-expressed as
k02  
[[z × E]] = −jωµ0 Mt + ẑ × Q · ẑ
20  
1 1 jωµ0 h  i
− ẑ × ∇t Pz − (∇t ẑ + ẑ∇t ) : Q + S − Szz I · ∇t (36a)
0 2 2 t
 14

k02
 
  1
[[z × H]] = jωPt + ẑ × S · ẑ − ẑ × ∇t Mz − (∇t ẑ + ẑ∇t ) : S
2 2
jω h  i
− Q − Qzz I · ∇t . (36b)
2 t
Meanwhile, the normal components are governed by

[[z · D]] = −∇t · D0

 i
1 jωµ0 0   1h
= −∇t · Pt + ∇t Qzz − ẑ × S · ẑ − Q · ∇t
2 2 t 2 t
 
jωµ0 0    
= −∇t · Pt − ẑ × S · ẑ − Q − Qzz I · ∇t (37a)
2

[[z · B]] = −∇t · B0

 i
1 jω   1h
= −∇t · µ0 Mt + ∇t Szz + ẑ × Q · ẑ − S · ∇t
2 2 t 2 t
 
jω    
= −∇t · µ0 Mt + ẑ × Q · ẑ − S − Szz I · ∇t . (37b)
2

E. Normal field components

In the main text, it is asserted that the correct form for the average fields is
1
Eav = Eav,t + (1 Ei,z + 1 Er,z + 2 Et,z )|z=0 ẑ (38a)
2
1
Hav = Hav,t + (µ1 Hi,z + µ1 Hr,z + µ2 Ht,z )|z=0 ẑ , (38b)
2
which we will now show, as was done in [38] for the tangential components of the fields.
Consider a dielectric slab, extending from 0 < z < d between two different media, and having a relative electric permittivity
of d and a relative magnetic permeability of µd , as in Figure 9. The first medium (z < 0) is modelled with relataive values
(1 , µ1 ) while the second medium (z > d) is modelled with (2 , µ2 ).

Fig. 9: A dielectric slab extending from 0 < z < d is illuminated by an obliquely incident plane wave. The average electric
field within the slab is unknown and desired. This figure is taken from [38].

Now, with an oblique incident plane wave from the left side (wavevector kd in the slab), the fields within the slab can be
expressed as
Ed (z) = Ae−jβz + Be+jβz , (39)
 15

where β = kd · ẑ, and constrained by the boundary conditions

Ed,t (0) = E1,t (0) (40a)
Ed,t (d) = E2,t (d) (40b)
d Ed,z (0) = 1 E1,z (0) (40c)
d Ed,z (d) = 2 E2,z (d) , (40d)
following which one can solve for A = At + Az ẑ and B = Bt + Bz ẑ:
E2,t (d) − E1,t (0)ejβd
At = (41a)
e−jβd − ejβd
E2,t (d) − E1,t (0)e−jβd
Bt = − (41b)
e−jβd − ejβd
r,2 E2,z (d) − r,1 E1,z (0)ejβd
Az = (41c)
r,d (e−jβd − ejβd )
r,2 E2,z (d) − r,1 E1,z (0)e−jβd
Bz = − . (41d)
r,d (e−jβd − ejβd )
Finally, we want to find the averages of these fields. This is done by integrating across the slab and dividing by its thickness,

1 d
Z
Ek,av = Ed,t (z)dz (42a)
d 0
 
E1,t (0) + E2,t (d) βd
= tan , (42b)
βd 2
using which we take the average as d → 0 for a thin surface. Then,
E1,t (0) + E2,t (0)
lim Ed,av = . (42c)
d→0 2
Meanwhile, the average part of the z component is
1 d
Z
Ez,av = Ed,z (z)dz (43a)
d 0
 
1 E1,z (0) + 2 E2,z (d) d βd
= tan . (43b)
βd 2
From which
1 E1,z (0) + 2 E2,z (0)
lim Ez,av = . (43c)
d→0 2d
Now, for susceptibilities the dimensionless factor in the denominator d serves as a factor which can be absorbed by the
susceptibilities, and so we can set d = 1. Thus, noting that E1 = Ei + Er and E2 = Et , we arrive at (38a). By duality, the
same exercise can be carried out for H fields to find (38b).

R EFERENCES
[1] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar, and C. R. Simovski, “Metasurfaces: From microwaves to visible,” Physics Reports, vol. 634,
pp. 1–72, May 2016.
[2] H.-T. Chen, A. J. Taylor, and N. Yu, “A review of metasurfaces: Physics and applications,” Rep. Prog. Phys., vol. 79, p. 076401, June 2016.
[3] N. Yu and F. Capasso, “Flat optics with designer metasurfaces,” Nat. Materials, vol. 13, Apr. 2014.
[4] P. Genevet and F. Capasso, “Holographic optical metasurfaces: A review of current progress,” Rep. Prog. Phys., vol. 78, no. 2, p. 024401, 2015.
[5] S. Abdollahramezani, O. Hemmatyar, and A. Adibi, “Meta-optics for spatial optical analog computing,” Nanophotonics, vol. 9, pp. 4075–4095, Oct.
2020.
[6] W. Xue and O. D. Miller, “High-NA optical edge detection via optimized multilayer films,” J. Opt., vol. 23, p. 125004, Nov. 2021.
[7] C. Chen, W. Qi, Y. Yu, and X. Zhang, “On-chip optical spatial-domain integrator based on Fourier optics and metasurface,” Nanophotonics, June 2021.
[8] A. Babaee, A. Momeni, A. Abdolali, and R. Fleury, “Parallel Analog Computing Based on a $2\ifmmode\times\else\texttimes\fi{}2$ Multiple-Input
Multiple-Output Metasurface Processor With Asymmetric Response,” Phys. Rev. Applied, vol. 15, p. 044015, Apr. 2021.
[9] A. Momeni, H. Rajabalipanah, M. Rahmanzadeh, A. Abdolali, K. Achouri, V. Asadchy, and R. Fleury, “Reciprocal Metasurfaces for On-axis Reflective
Optical Computing,” IEEE Trans. Antennas Propag., pp. 1–1, 2021.
[10] A. Momeni, M. Safari, A. Abdolali, N. P. Kherani, and R. Fleury, “Asymmetric Metal-Dielectric Metacylinders and Their Potential Applications From
Engineering Scattering Patterns to Spatial Optical Signal Processing,” Phys. Rev. Applied, vol. 15, p. 034010, Mar. 2021.
[11] K. Achouri and O. J. F. Martin, “Angular Scattering Properties of Metasurfaces,” IEEE Trans. Antennas Propag., vol. 68, pp. 432–442, Jan. 2020.
[12] X. Liu, F. Yang, M. Li, and S. Xu, “Generalized Boundary Conditions in Surface Electromagnetics: Fundamental Theorems and Surface Characterizations,”
Appl. Sci., vol. 9, p. 1891, Jan. 2019.
 16

[13] X. Wang, A. Dı́az-Rubio, and S. A. Tretyakov, “Independent Control of Multiple Channels in Metasurface Devices,” Phys. Rev. Applied, vol. 14,
p. 024089, Aug. 2020.
[14] A. Monti, A. Alù, A. Toscano, and F. Bilotti, “Surface Impedance Modeling of All-Dielectric Metasurfaces,” IEEE Trans. Antennas Propag., vol. 68,
pp. 1799–1811, Mar. 2020.
[15] V. Tiukuvaara, T. J. Smy, K. Achouri, and S. Gupta, “Surface Susceptibilities as Characteristic Models of Reflective Metasurfaces,” IEEE Transactions
on Antennas and Propagation, pp. 1–1, 2022.
[16] A. Rahimzadegan, T. D. Karamanos, R. Alaee, A. G. Lamprianidis, D. Beutel, R. W. Boyd, and C. Rockstuhl, “A Comprehensive Multipolar Theory
for Periodic Metasurfaces,” Advanced Optical Materials, vol. n/a, no. n/a, p. 2102059, 2022.
[17] M. Dehmollaian, G. Lavigne, and C. Caloz, “Comparison of Tensor Boundary Conditions With Generalized Sheet Transition Conditions,” IEEE
Transactions on Antennas and Propagation, vol. 67, pp. 7396–7406, Dec. 2019.
[18] D. Zaluški, A. Grbic, and S. Hrabar, “Analytical and experimental characterization of metasurfaces with normal polarizability,” Phys. Rev. B, vol. 93,
p. 155156, Apr. 2016.
[19] C. L. Holloway, A. Dienstfrey, E. F. Kuester, J. F. O’Hara, A. K. Azad, and A. J. Taylor, “A discussion on the interpretation and characterization of
metafilms/metasurfaces: The two-dimensional equivalent of metamaterials,” Metamaterials, vol. 3, pp. 100–112, Oct. 2009.
[20] C. L. Holloway and E. F. Kuester, “A Homogenization Technique for Obtaining Generalized Sheet Transition Conditions (GSTCs) for a Metafilm
Embedded in a Magneto-Dielectric Interface,” IEEE Trans. Antennas Propagat., vol. 64, pp. 4671–4686, Nov. 2016.
[21] C. L. Holloway, E. F. Kuester, and A. Dienstfrey, “Characterizing Metasurfaces/Metafilms: The Connection Between Surface Susceptibilities and Effective
Material Properties,” IEEE Antennas Wirel. Propag., vol. 10, pp. 1507–1511, 2011.
[22] E. Kuester, M. Mohamed, M. Piket-May, and C. Holloway, “Averaged transition conditions for electromagnetic fields at a metafilm,” IEEE Trans.
Antennas Propag., vol. 51, pp. 2641–2651, Oct. 2003.
[23] K. Achouri, M. A. Salem, and C. Caloz, “General Metasurface Synthesis Based on Susceptibility Tensors,” IEEE Trans. Antennas Propag., vol. 63,
pp. 2977–2991, July 2015.
[24] T. J. Smy, S. A. Stewart, J. G. N. Rahmeier, and S. Gupta, “FDTD Simulation of Dispersive Metasurfaces With Lorentzian Surface Susceptibilities,”
IEEE Access, vol. 8, pp. 83027–83040, 2020.
[25] T. J. Smy, V. Tiukuvaara, and S. Gupta, “IE-GSTC Metasurface Field Solver using Surface Susceptibility Tensors with Normal Polarizabilities,”
arXiv:2105.05875 [physics], May 2021.
[26] T. J. Smy and S. Gupta, “Part 2 – Eigenfunction Expansion (EFE) Analysis of Cylindrical and Sectorial Metasurfaces,” July 2022.
[27] F. Bernal Arango, T. Coenen, and A. F. Koenderink, “Underpinning Hybridization Intuition for Complex Nanoantennas by Magnetoelectric Quadrupolar
Polarizability Retrieval,” ACS Photonics, vol. 1, pp. 444–453, May 2014.
[28] J. G. Nizer Rahmeier, T. J. Smy, J. Dugan, and S. Gupta, “Part I -Spatially Dispersive Metasurfaces: Zero Thickness Surface Susceptibilities & Extended
GSTCs,” IEEE Transactions on Antennas and Propagation, pp. 1–1, 2022.
[29] K. Achouri and O. J. F. Martin, “Extension of Lorentz reciprocity and Poynting theorems for spatially dispersive media with quadrupolar responses,”
Phys. Rev. B, vol. 104, p. 165426, Oct. 2021.
[30] K. Achouri, V. Tiukuvaara, and O. J. F. Martin, “Multipolar Modeling of Spatially Dispersive Metasurfaces,” IEEE Transactions on Antennas and
Propagation, pp. 1–1, 2022.
[31] I. J. Richards and H. K. Youn, The Theory of Distributions: A Nontechnical Introduction. New York, NY: Cambridge University Press, 1990.
[32] I. Mithat, Discontinuities in the Electromagnetic Field. Wiley-IEEE Press, 2011.
[33] R. Paniagua-Domı́nguez, Y. F. Yu, A. E. Miroshnichenko, L. A. Krivitsky, Y. H. Fu, V. Valuckas, L. Gonzaga, Y. T. Toh, A. Y. S. Kay, B. Luk’yanchuk,
and A. I. Kuznetsov, “Generalized Brewster effect in dielectric metasurfaces,” Nat Commun, vol. 7, p. 10362, Jan. 2016.
[34] G. Lavigne and C. Caloz, “Generalized Brewster effect using bianisotropic metasurfaces,” Opt. Express, OE, vol. 29, pp. 11361–11370, Mar. 2021.
[35] M. Idemen and A. H. Serbest, “Boundary conditions of the electromagnetic field,” Electronics Lett., vol. 23, no. 13, pp. 704–705, 1987.
[36] C. Simovski, Composite Media with Weak Spatial Dispersion. Jenny Stanford Publishing, 1st edition ed., Nov. 2018.
[37] M. Albooyeh, S. Tretyakov, and C. Simovski, “Electromagnetic characterization of bianisotropic metasurfaces on refractive substrates: General theoretical
framework,” Annalen der Physik, vol. 528, no. 9-10, pp. 721–737, 2016.
[38] K. Achouri and C. Caloz, Electromagnetic Metasurfaces: Theory and Applications. Hoboken, NJ: Wiley-IEEE Press, 1st edition ed., May 2021.
[39] E. F. Kuester and D. C. Chang, Electromagnetic Boundary Problems: Electromagnetics, Wireless, Radar, and Microwaves. Boca Raton: CRC Press,
Oct. 2015.
[40] S. Tretyakov, Analytical Modeling in Applied Electromagnetics. Norwood, MA, USA: Artech House, 2003.
[41] C. Caloz, A. Alù, S. Tretyakov, D. Sounas, K. Achouri, and Z.-L. Deck-Léger, “Electromagnetic Nonreciprocity,” Physical Rev. Appl., vol. 10, no. 4,
p. 047001, 2018.
[42] M. Riccardi, A. Kiselev, K. Achouri, and O. J. F. Martin, “Multipolar expansions for scattering and optical force calculations beyond the long wavelength
approximation,” Phys. Rev. B, vol. 106, p. 115428, Sept. 2022.
[43] K. Achouri and O. J. F. Martin, “Symmetries and Angular Scattering Properties of Metasurfaces,” in 2019 Thirteenth International Congress on Artificial
Materials for Novel Wave Phenomena (Metamaterials), pp. X–007–X–009, Sept. 2019.
[44] K. Achouri, V. Tiukuvaara, and O. J. F. Martin, “Spatial Symmetries in Multipolar Metasurfaces: From Asymmetric Angular Transmittance to Multipolar
Extrinsic Chirality,” Aug. 2022.
[45] C. A. Balanis, Advanced Engineering Electromagnetics. Hoboken, USA: John Wiley & Sons, 2012.