Mazor Steinberg .

Particle chains and clusters on non-Bravais lattices

Y. Mazor and Ben Z. Steinberg

School of Electrical Engineering, Tel Aviv University, Ramat-Aviv, Tel-Aviv 69978 Israel
(Dated: September 2011)
We study optical guiding and dispersion in chains of linear particle clusters. A formulation
governing the electrodynamics of chains on interlaced and non-Bravais lattices is developed, and
used to study clustered chains. It is shown that non-Bravais and clustered chains may oer new
degrees of freedom for dispersion synthesis and engineering, controlling parameters such as frequency
and group and phase velocities.
PACS numbers: 78.67.Bf,42.82.Et,71.45.Gm,41.20.Jb,42.25.Bs
Keywords: plasmonic waveguide, particle chains, particle arrays, particle clusters
I. INTRODUCTION
The electromagnetic properties of linear
chains of identical and equally spaced micro-
particles have been studied in a number of
previous publications [1]-[7]. If the inter-
particle distance d is much smaller than the
free space wavelength , these structures can
support the propagation of guided electromag-
netic modes along the chain that do not radi-
ate into the free space, with total width that
can be much smaller than . Hence the name
Sub-Diraction Chains (SDC). If the parti-
cles are identical isotropic spheres, three inde-
pendent electric dipole modes are supported;
one with longitudinal e-dipole polarization p
z
,
and two degenerate modes with transverse and
mutually orthogonal polarizations p
x
, p
y
[7]. In
the former the group and phase velocities v
g
, v
p
are of the same sign, while in the latter v
g
and
v
p
are of opposite signs. It has been shown
that if the chain particles are non-spherical
(e.g. ellipsoids) more elaborate mode-structures
may exist [8]-[10]. Generally, the interest in
SDCs is motivated by both theoretical and
practical points of view. Particle chains were
proposed as guiding structures and junctions
[1]-[7], as mechanism for coupling to surface
waves [8], as polarization sensitive waveguides
[9], and for one-way optical guiding [10]. Re-
cently, it has been shown that SDCs can sup-
port second harmonic generation and gain, with
a proper design of their dispersion character-
istics. The exibility in designing the chain
dispersion, however, is quite limited and it has
been achieved only via a microscopic design of
the particles shape [11]. Clearly these potential

[email protected]; - Corresponding author.

This research was supported by the Israel Science
Foundation (grant 1503/10)
applications could benet from a better control
of the SDC dispersion characteristics.
Previous studies, however, were restricted to
Bravais lattices; i.e. the chain consists of array
of discrete points, the locations of which z
n
are
generated by a single discrete translation op-
eration z
n
= nd. Furthermore, in these stud-
ies all lattice points are populated by identical
particles. Hence, the structure seen by each
particle is independent of the particle site.
The purpose of the present work is to study
the optical modes of linearly clustered periodic
particle chains. Schematic examples are shown
in Fig. 1. In our chains, particles are situated
on lattice points that generally cannot be ob-
tained by the single translation operation de-
scribed above; we deal with chains on non-
Bravais lattices. Furthermore, our formulation
can handle chains in which dierent particles
populate dierent sites within the cluster. Ac-
cordingly, even though the chain is still peri-
odic with period D, the structure seen by
each particle depends on its specic location
within the cluster. Compared to conventional
FIG. 1. Possible congurations of clustered chains.
Bravais (B) chains, non-Bravais (NB) chains
may oer optical modes with a much richer va-
riety of dispersions and polarizations, as well
2
as group and phase velocities control. For ex-
ample, in B-chains of plasmonic particles the
dispersion curves of the transverse polarization
modes always have an electromagnetic light-
line branch, and a trapped plasmonic wave
branch with opposite group and phase veloc-
ities [7]. The plasmonic NB-chains support
these traditional modes, but may oer addi-
tional transverse modes having dispersion with
no light line branch, and a trapped plasmonic
wave branch with v
g
v
p
0. Likewise, in B-
chains the longitudinal polarization modes al-
ways possess v
g
v
p
0, while the NB-chain may
oer additional longitudinal polarization modes
with v
g
v
p
0. Generally, the specic prop-
erties of these new modes are determined by
the cluster parameters such as the number of
particles per cluster M, the ratio between the
chain period D and the typical inter-particle
distance within the cluster d, and the specic
polarizabilities of the particles within a cluster.
These added degrees of freedom may oer bet-
ter exibility for dispersion synthesis and engi-
neering, with applications to guiding, ltering,
slow-light structures, and more.
To tackle the problem, we apply polarizabil-
ity theory and a discrete dipole approximation
(DDA). These are standard tools used in many
studies of particle chains. They hold well as
long as the particle diameter a is small com-
pare to the wavelength, and the smallest inter-
particle distance d
i
is large compare to a. Stud-
ies show, in fact, excellent agreement with exact
solutions even for d
i
1.5a [12]. We note that
in innite chains, the inter-particle interactions
result in a sum that converges extremely slowly
( 1/n), and cannot be used as is for reliable
computations of, say, the chain modes or dis-
persion relations. For B-chains, the interactions
can be summed analytically and expressed in
terms of the Polylogarithm functions [7], that
can be computed eciently (see appendix for
ecient summation formulas). Unfortunately,
however, for NB chains the summation is much
more involved, and its exact expression and/or
ecient evaluation need a specially tailored ex-
ponentiation procedure that will be developed
in the present work.
The structure of the paper is as follows.
The mathematical formulation is presented in
Sec. II, where the B and NB cases are con-
trasted, and the summation-enabling exponen-
tiation procedure is developed. Some spe-
cic examples of clustered chains are shown in
Sec. III, and their physical properties are dis-
cussed. Concluding remarks are provided in
Sec. IV.
II. FORMULATION
We start by characterizing the structure.
The chain period consists of M (possibly dif-
ferent) particles, and its length is D. Particles
may be linearly clustered within D, separated
by inter-particle distances that are not neces-
sarily equal. We make only one restriction for
the inter-particle distances: they all must be
rational fractions of D. This is needed only
to facilitate compact analysis of the chain dy-
namics; since the set of rational numbers is a
dense subset of the real numbers, this mathe-
matically formal restriction bares no limitation
on the general physical validity of our results.
Hence, there exists a fundamental interval d
and an integer N M such that D = Nd and
such that all particle locations are obtained by
some integer translations of d. Clearly, the con-
verse is not necessarily true: not every integer
translation of d is populated by a particle; some
points may be empty. See Fig. 1 for examples.
We dene a nite sequence of N 3 3 matri-
ces
n
, n = 1, . . . , N, representing the electric
polarizabilities that apply to all the N points
within a period, populated or not. To describe
an empty point we set the corresponding polar-
izability to zero.
We denote by p
nm
, n = 1, . . . , N, m Z the
electric dipole response at the n-th location in
the m-th cluster (some will be null). Under
the discrete dipole approximation (DDA), the
chain modes are governed by
p
nm
=
1
0

n

=1
nm
A(R
n

nm
)p
n

m
(1)
where

nm
means that the summation over
n

excludes the nm term. In the above,

R
n

nm
= |(mm

)D + (n n

)d| is the dis-

tance between the nm and n

sites. The 33
matrix A(z) is the free space electric eld radi-
ation from innitesimal electric dipole,
A(z) =
e
ik|z|
4 |z|
_
k
2
A
1
+
_
1
z
2

ik
|z|
_
A
2
_
(2)
here A
1
= diag(1, 1, 0), A
2
= diag(1, 1, 2).
Clearly, Eq. (1) is shift invariant with m. Since
the structure period is D, we assume a solution
of the form
p
nm
= p
n0
e
imD
, n = 1, . . . , N. (3)
Substituting Eq. (3) in Eq. (1) and rearranging,
we obtain the 3M3M matrix equation (rows
n and columns n

that correspond to empty

3
points are taken out a-priori)
_
S
0

1
n
_
p
n0
+
N

=1
n
S
n

n
p
n

0
= 0, (4)
n = 1, . . . , N,
where
n
= 6
0
k
3

n
is the normalized (di-
mensionless) polarizability. Again,

n
indi-
cates that the summation excludes the n

= n
term. Since n

and n that correspond to empty

points are also taken out, this summation is
done over M 1 matrices S

, = 0 only. Gen-
erally, S

are 3 3 matrices obtained by the

matrix summations,
S
0
=
6
k
3

m=0
A(mD)e
imD
(5)
S

=
6
k
3

m
A(mD + d)e
imD
(6)
1 || N 1.
The dispersion relation () is obtained by
looking for values of for which the determi-
nant of Eq. (4) vanishes. The corresponding
vectors are the dipole moments of the M parti-
cles in the reference 0-th cluster. Note that the
3 3 matrix S
0
is identical to the one obtained
in conventional B-chains with inter-particle dis-
tance D. It can be expressed analytically using
the Polylogarithm functions [7]. Specically
S
0
=
3
2
_
f
1
(kD, D)
kD
A
1

if
2
(kD, D)
(kD)
2
A
2
+
f
3
(kD, D)
(kD)
3
A
2
_
(7)
where
f
s
(x, y) = Li
s
[e
i(x+y)
] + Li
s
[e
i(xy)
] (8)
and where Li
s
(z) is the s-th order Polyloga-
rithm function, dened by
Li
s
(z) =

n=1
z
n
n
s
, Li

s
(z) = z
1
Li
s1
(z),
(9)
with Li
0
(z) = z/(1 z), Li
1
(z) = ln(1 z).
See [13] for discussions of Li
s
(z) analytic prop-
erties, and [7, 14] for their use in particle B-
chains. The series in Eq. (9) is strictly conver-
gent only for |z| 1, and due to the algebraic
behavior convergence is slow. However, in the
present work we look for values of Li
s
(z) only
on the unit circle in the complex z plan (i.e.
z = e
i
). Fast converging formulas for the eval-
uation of Li
s
in this domain are readily avail-
able - see appendix for details.
The matrices S

, = 0 can also be evaluated

as nite sums of Li
s
. This is shown in Sec.II B.
A. The special case of B-Chains
A conventional B-chain of period D
b
(iden-
tical particles and no clustering) can be man-
ifested in Eq. (4) via one of the following op-
tions.
The most straight-forward manifestation cor-
responds to D = D
b
and d = D, N = M = 1.
Then, since the summation over S

, = 0 ex-
cludes the n

= n term, the only surviving term

in Eq. (4) is S
0

1
. This is identical to pre-
viously derived formulations, hence the previ-
ously derived dispersion relations are immedi-
ately obtained.
In another option, one may set D = D
b
and
d = D/N with N > 1, but M = 1. That is: the
period consists of many points, but only one is
populated. Recall that in Eq. (4) we exclude
n, n

that correspond to empty sites. Therefore

again the surviving formulation is identical to
those derived previously.
The third option is about redundancy; set
d = D
b
, with arbitrary N = M. That is: de-
ne an articial fully populated period D =
ND
b
= Nd that consists of N periods of the
original chain. Clearly this is not the funda-
mental (or irreducible) period of the structure,
so the Brillouin zone in compressed by a factor
of N [|| /(ND
b
) instead of || /D
b
].
As a punishment to the unaware user, the en-
tire set S

, || N 1 must be included (see

next subsection for details on how to compute
these). The resulting dispersion curve is the
original curve of B-chains, folded N 1 times
within the compressed Brillouin zone.
B. NB-Chains
The general case of NB chains is manifested
by the presence of the M1 additional matrices
S

in Eq. (4) except for the standard S

0
. Un-
like S
0
, these matrices cannot be expressed as
the simple three-fold sum over pairs of Polylog-
arithm functions, as in Eqs. (7)(8). However,
due to the fact that we have restricted the anal-
ysis to cases for which particle locations are ra-
tional fractions of the chain period D, one can
still express S

, = 0 analytically by a nite
weighted sum of Li
s
that can be computed ef-
ciently. Towards this end, we re-write Eq. (6)
as
S

= h
1
[]A
1
ih
2
[]A
2
+ h
3
[]A
2
(10)
where (recall that D = Nd)
h
s
[] =
3
2
1
(kd)
s

m
e
ikd|mN+|+idmN
|mN + |
s
. (11)
4
1. Evaluation of hs[]
For 1 N 1, h
s
[] can be re-written as
h
s
[] =
3
2
e
id
(kd)
s

m=0
_
e
i(k+)d

mN+
(mN + )
s
+
+
3
2
e
id
(kd)
s

m=1
_
e
i(k)d

mN
(mN )
s
. (12)
We concentrate on the rst sum above. It has
the form
=

m=0
(e
ix
)
mN+
(mN + )
s
. (13)
This sum can be re-written as
=

n=1
(e
ix
)
n
n
s
a
n
() (14)
where a
n
() is an auxiliary periodic sequence of
period N, satisfying
a
n
() =
_
0, n = 1, . . . , N, n =
1, n =
. (15)
Clearly, the series periodicity implies the recur-
rence relation a
n+N
= a
n
, whose characteristic
polynomial p() =
N
1 has N distinct roots

N
1 = 0 {
r
}
N1
r=0
= e
i2r/N
, (16)
hence the innite sequence a
n
(), n can be
generated by the nite sum
a
n
() =
N1

r=0
C
r
()
n
r
=
N1

r=0
C
r
()e
i2rn/N
.
(17)
The coecients C
r
() can be determined us-
ing the N-initial conditions of the recurrence
relation [given by Eq. (15)] in Eq. (17). The
result is the N N Vandermonde matrix equa-
tion a() = C(). Here a() is a vector of N
entries, whose elements are given by Eq. (15),
C() is the vector of unknown coecients, and
is a Vandermonde matrix, whose n, r entry
is
nr
= e
i2rn/N
. Due to its specic structure,
is also a unitary transformation from the Eu-
clidean basis a
n
() (roam with ) to a Fourier
basis. Its inverse is its adjoint (normalize rst).
Hence
C
r
() = N
1
e
i2r/N

a
n
() =
1
N
N1

r=0
e
i2r(n)/N
, n. (18)
Substituting this result into Eq. (14) and ex-
changing the order of summation, we can ex-
press as a nite sum of Polylogarithms
= N
1
N1

r=0
e
i2r/N
Li
s
(e
ix+i2r/N
). (19)
Likewise, we may repeat essentially the same
procedure for the second sum in Eq. (12) (note
the lower summation bound; shift the index by
1, and at the end change r Nr

). The nal
result for h
s
[], 1 N 1, is
h
s
[] =
3
2N
e
id
(kd)
s
(20)

N1

r=0
e
i2r/N
f
s
(kd, d + 2r/N)
where f
s
(x, y) is dened in Eq. (8). Repeating
the same procedure for h
s
[] with 1
N 1, we obtain again a summation over N
terms,
h
s
[] =
3
2N
e
id
(kd)
s
(21)

N1

r=0
e
i2r/N
f
s
(kd, d + 2r/N).
With these nite expansions for h
s
, we use
Eq. (10) to evaluate the matrices S

, 1 ||
N 1 for the NB-chain formulation [Eq. (4)].
Finally, note that if a cluster consists of M
particles (recall that M N), the matrix size
in Eq. (4) is actually 3M3M. If
n
is diago-
nal for all particles, this can be reduced to three
independent sets of M M each. This gener-
ally yields M dispersion curves for each polar-
ization. It should be emphasized, however, that
the S

matrix entries, always consists of sum-

mation over N terms containing Polylogarithm
functions, independently of how many of the N
sites in the cluster are actually populated.
III. EXAMPLES
We rst discuss the simplest example of clus-
tered chains, shown in Fig. 2; a cluster that con-
sists of two identical spherical particles (M =
2). The particles material susceptibility is as-
sumed to satisfy the Drude model. The chain
periodicity is set at D =
p
/20, where
p
is the
wavelength corresponding to
p
. Two exam-
ples are considered; the rst with d
1
very close
to D/2 corresponding to weak clustering, or an
almost regular chain, and the second with d
1
considerably smaller than D/2, corresponding
5
FIG. 2. Chain geometry for two-particles cluster.
to a relatively strong clustering. Specically,
we chose d
1
= D/2.1 (d
1
= D/3), leading to
N = 21 (N = 3). Hence
n
= 0 n except
for n = 1, 11 (n = 1, 2). For these two exam-
ples we have computed the dispersion curves
and dipole mode structures. For the latter it
is most instructive to examine the phase dier-
ence between the rst and second dipoles of
the cluster,
p
n

0
= p
10
e
i
(22)
with n

= 11 (n

= 2).
The results for the transverse polarization are
shown in Fig. 3. Only the domain 0 D
is presented since the dispersions are even with
. The following observations can be made.
0 0.5 1
0.575
0.576
0.577
0.578
0.579

p
D /
1 0 1
0.575
0.576
0.577
0.578
0.579
/
0 0.5 1
0.575
0.576
0.577
0.578
0.579

p
D /
1 0 1
0.575
0.576
0.577
0.578
0.579
/
(a) (b)
(c) (d)
"inphase"
"antiphased"
FIG. 3. Mode solutions for two dierent ratios
d1/D in a 2-particle clustered chain, for transverse
polarizations. (a) dispersion curves for D = 2.1d1.
(b) The inter-cluster phase dierence dened in
Eq. (22). (c) and (d) are the same as (a) and (b)
but for D = 3d1.
For each chain, the number of independent
transverse modes (or dispersion curves) is 2 -
identical to the number of particles in a clus-
ter. The upper dispersion curve (shown in red)
has essentially the same characteristics of the
transverse modes in conventional B-chains (see,
e.g. [7]). It possesses opposite group and phase
velocities. It also possesses a light-line branch
only a section of which is shown in red dotted
linethat runs close to the light-line cone, sat-
isfying c. This light-line branch repre-
sents a modal solution whose electromagnetic
elds weakly interact with the chain, and are
nearly plane-waves. In this mode the dipoles
within a cluster are excited in phase as ev-
ident from the fact that for the entire upper
(red) dispersion curve we have . In
contrast, the lower dispersion curve (blue) pos-
sesses quite dierent characteristics. Its phase
and group velocities are parallel, and it pos-
sesses no light-line branch. In this mode the
dipoles within a cluster are excited essentially
in opposite phases, as for the entire
curve. This fact explains the absence of light-
line branch in the blue curves; this dipole dis-
tribution cannot couple at all to a free-space
plane wave. In fact, for strongly clustered
chains (d
1
D) this mode approaches the case
of quadrupole-particles chain with inter-particle
distance D.
Solving the same chain geometry for longi-
tudinal polarizations we obtain the dispersion
curves in Fig. 4. Again, we have two indepen-
dent modes, the same as the number of cluster
particles. The lower (red) curve is essentially
the same as the dispersion curve of conventional
B-chain in longitudinal polarization, and the
cluster dipoles are excited mainly in-phase.
The upper (blue) curve shows that clustering
enables us to get v
g
v
p
< 0 which cannot be
achieved in this polarization with conventional
B-chains. Here the cluster dipoles are excited
in opposite phases. Both curves do not possess
light-line branches.
0 0.5 1
0.572
0.574
0.576
0.578
0.58

p
D /
1 0 1
0.572
0.574
0.576
0.578
0.58
/
0 0.5 1
0.572
0.574
0.576
0.578
0.58

p
D /
1 0 1
0.572
0.574
0.576
0.578
0.58
/
(b) (a)
(c)
(d)
FIG. 4. The same as Fig. 3 but for the longitudinal
polarization.
The dispersion curves in Figs. 3-4 indicate
that clustering can be used to reduce the op-
6
tical modes group velocity v
g
, while keeping
constant the same number of particles per unit
distance (clearly, one can always reduce v
g
by
increasing the distance between the particles,
but this also increases the light-line gap and
loss). In Fig. 5 we show v
g
for chains with two
particles clusters and period D = d
1
. It is
seen that v
g
decreases as increases (tighter
clustering), and the eect is more profound for
the anti-phase modes.
0.5772 0.5774 0.5776 0.5778 0.578 0.5782 0.5784 0.5786 0.5788 0.579
6
4
2
0
x 10
4
In phase modes
/
p
V
g

/
(
D

p
)
0.5756 0.5758 0.576 0.5762 0.5764 0.5766 0.5768 0.577 0.5772 0.5774
0
2
4
6
x 10
4
Anti phase modes
/
p
V
g

/
(
D

p
)


=4
=3
=2.25
=2
(b)
(a)
FIG. 5. Transverse polarization modes group ve-
locity in chains of two particles clusters. The case
of = 2 corresponds to a conventional chain of pe-
riod D/2. (a) the in-phase mode (b) the anti-phase
mode
We turn to examine larger clusters, consist-
ing of M equally-spaced particles, with D =
d
1
where d
1
is the inter-particle spacing within
a cluster. If = M the structure is nothing
but a regular B-chain with inter-particle dis-
tance d
1
. For > M the gap opened between
two adjacent clusters is larger than d
1
, hence
each cluster indeed looks like a set of M iso-
lated particles. For M1 < < M, this gap is
smaller than d
1
and the structure can be viewed
as a chain of two-particles cluster with period
D, interlaced with another periodic chain. We
examine these cases for M = 5. Figures 6-
7 show the dispersion curves for the cases of
= 4.75 and = 5.25, respectively. The dis-
persion curves for the transverse polarization
are shown in Figs. 6a and 7a, with insets in the
top of these gures schematizing the respective
chain structures. The corresponding phase dis-
tributions within cluster are shown in Figs. 6b
and 7b. Note that the only dispersion curve
for which all the particles in a cluster possess
the same phase, is the upper one in each case
(this occurs at the curves upper edge). Accord-
ingly, we have observed that this curve is the
only one that possesses light-line branch (not
shown). Also note that the dispersion curves
for < M span over a larger range of frequen-
cies, when compared to > M, and also have
lower group velocities.
0 0.5 1
0.568
0.57
0.572
0.574
0.576
0.578
0.58
0.582
0.584
0.586
D /

p
1 0.5 0 0.5 1
/


p
2
/p
1
p
3
/p
1
p
4
/p
1
p
5
/p
1
M=5, =4.75
(a) (b)
FIG. 6. Transverse polarization dispersion curves
(a) and the corresponding phase distributions (b)
for a chain with 5-particles cluster, and = 4.75.
The upper curve is the only one that possesses light-
line branch (not shown)
0 0.5 1
0.568
0.57
0.572
0.574
0.576
0.578
0.58
0.582
0.584
0.586
D /

p
1 0.5 0 0.5 1
/


p
2
/p
1
p
3
/p
1
p
4
/p
1
p
5
/p
1
M=5, =5.25
(a) (b)
FIG. 7. Transverse polarization dispersion curves
(a) and the corresponding phase distributions (b)
for a chain with 5-particles cluster, and = 5.25.
The upper curve is the only one that possesses light-
line branch (not shown).
IV. CONCLUSIONS
A formulation governing the propagation of
optical modes in general linearly-clustered par-
ticle chains and linear non-Bravais chains has
been developed. The formulation can handle
periodic chains with non-identical particles as
well. It provides the modes dispersion in terms
7
of the roots of the chain determinant, where
the latter is expressed in terms of a nite sum
of Polylogarithm functions. Specic examples
of two-particle clusters and ve-particle clusters
are provided. It is shown that clustered chains
may oer new degrees of freedom in dispersion
engineering, such as control of operating fre-
quency, and the wave velocities. Strong reduc-
tion of group velocity, and also the inversion
of the group velocity for longitudinal as well as
transverse excitations are shown.
Appendix A: Evaluation of Lis(e
i
)
Li
s
(e
ix
) for real x, can be expressed as -
nite sums of Clausen integrals [13],[15]. The
latter can be computed using series that con-
verge much faster than the algebraic conver-
gence of the formal denition in Eq. (9). Note
that Li
1
(z) = ln(1 z). Furthermore, from
Eq. (9) it follows that
Li
s+1
(z) =
_
z
0
t
1
Li
s
(t) dt (A1)
Collecting the Clausen integrals components
that constitute our Li
s
and rearranging, we ob-
tain for s = 2 and |x|
Li
2
(e
ix
) =

2
6

x
4
(2 x) (A2)
i
_
xln x x
1
2

n=1
B
2n
x
2n+1
n(2n + 1)(2n)!
_
where B
2n
are the Bernoulli numbers absolute
value. With terms up to n = 5 in the series
above, the relative error for x = is in the order
of 10
5
. From the integral relation Eq. (A1) we
obtain
Li
3
(e
ix
) = (3) +
x
2
2
(ln x 3/2)

1
4

n=1
B
2n
x
2n+2
n(n + 1)(2n + 1)(2n)!
+ i
_

2
x
6

x
2
4
+
x
3
12
_
(A3)
where (3) 1.2020569. Higher order Polylog-
arithm functions are not needed in the present
work.
[1] M. Quinten, A. Leitner, J. R. Krenn, and
F. R. Aussenegg, Electromagnetic energy
transport via linear chains of silver nanopar-
ticles, Optics Letters, 23(17), 1331 (1998).
[2] S. A. Tretyakon and A. J. Viitanen, Line
of periodically arranged passive dipole scat-
terers, Electrical Engineering, 82, 353-361
(2000).
[3] M. L. Brongersma, J. L. Hartman, and
H. A. Atwater, Electromagnetic energy trans-
fer and switching in nanoparticle chain ar-
rays below the diraction limit, Phys. Rev. B,
62(24) R16356 (2000).
[4] W. H. Weber and G. W. Ford, Propagation
of optical excitations by dipolar interactions in
metal nanoparticle chains, Phys. Rev. B 70,
125429 (2004).
[5] A. F. Koenderink and A. Polman, Complex
response and polariton-like dispersion split-
ting in periodic metal nanoparticle chains,
Phys. Rev. B 74, 033402 (2006).
[6] T. Yang and K. B. Crozier, Dispersion and
extinction of surface plasmons in an array
of gold nanoparticle chains: inuence of the
air/glass interface, Optics Express 16(12),
8570 (2008).
[7] Andrea Alu and Nader Engheta, Theory of
linear chains of metamaterial/plasmonic parti-
cles as subdiraction optical nanotransmission
lines, Phys. Rev. B 74, 205436 (2006).
[8] D. V. Orden, Y. Fainman, and V. Lomakin,
Optical waves on nanoparticle chains coupled
with surfaces, Optics Letters, 34(4), 422-424
(2009).
[9] D. V. Orden, Y. Fainman, and V. Lomakin,
Twisted chains of resonant particles: opti-
cal polarization control, waveguidance, and
radiation, Optics Letters, 35(15), 2579-2581
(2010).
[10] Y. Hadad and Ben Z. Steinberg, Magne-
tized spiral chains of plasmonic ellipsoids for
one-way optical waveguides, Phys. Rev. Lett.,
105, 233904 (2010).
[11] Ben Z. Steinberg, Parametric plasmonics
and second harmonic generation in particle
chains, Optics Express, 19(27), 25843-25853
(2011).
[12] S. A. Maier, P. G. Kik, and H. A. At-
water, Optical pulse propagation in metal
nanoparticle chain waveguides, Phys. Rev. B,
67,205402 (2003).
[13] Leonard Lewin, Polylogarithms and Associated
Functions, Elsevier , New York, 1981.
[14] Y. Hadad and Ben Z. Steinberg, Greens func-
tion theory for innite and semi-innite parti-
cle chains, Phys. Rev. B 84, 125402 (2011).
[15] I. A. Stegun, Miscellaneous Functions,
Chapt. 27 in Handbook of Mathematical
Functions, edited by I. A. Stegun and
M. Abramowitz (Dover Publications, Inc.,
New York, 1970).