1: \documentclass[aps,preprint]{revtex4}
2: \usepackage{psfig}
3: \usepackage{amsfonts,amssymb}
4: \usepackage{graphics}
5:
6: \begin{document}
7:
8: \title{Propagation of Surface Plasmons in Ordered and Disordered Chains of
9: Metal Nanospheres}
10: \author{Vadim A. Markel}
11: \affiliation{Departments of Radiology and Bioengineering, University
12: of Pennsylvania, Philadelphia, PA 19104}
13:
14: \author{Andrey K. Sarychev}
15: \affiliation{Ethertronics Inc., San Diego, CA 92121 }
16:
17: \begin{abstract}
18: We report a numerical investigation of surface plasmon (SP)
19: propagation in ordered and disordered linear chains of metal
20: nanospheres. In our simulations, SPs are excited at one end of a
21: chain by a near-field tip. We then find numerically the SP amplitude
22: as a function of propagation distance. Two types of SPs are
23: discovered. The first SP, which we call the ordinary or quasistatic,
24: is mediated by short-range, near-field electromagnetic interaction
25: in the chain. This excitation is strongly affected by Ohmic losses
26: in the metal and by disorder in the chain. These two effects result
27: in spatial decay of the quasistatic SP by means of absorptive and
28: radiative losses, respectively. The second SP is mediated by longer
29: range, far-field interaction of nanospheres. We refer to this SP as
30: the extraordinary or non-quasistatic. The non-quasistatic SP can not
31: be effectively excited by a near-field probe due to the small
32: integral weight of the associated spectral line. Because of that, at
33: small propagation distances, this SP is dominated by the quasistatic
34: SP. However, the non-quasistatic SP is affected by Ohmic and
35: radiative losses to a much smaller extent than the quasistatic one.
36: Because of that, the non-quasistatic SP becomes dominant
37: sufficiently far from the exciting tip and can propagate with little
38: further losses of energy to remarkable distances. The unique
39: physical properties of the non-quasistatic SP can be utilized in
40: all-optical integrated photonic systems.
41: \end{abstract}
42:
43: \date{\today}
44: \maketitle
45:
46: \section{Introduction}
47: \label{sec:intro}
48:
49: Surface plasmons (SPs) are states of polarization that can propagate
50: along metal-dielectric interfaces or along other structures without
51: radiative losses. Polarization in an SP excitation can be spatially
52: confined on scales that are much smaller than the free-space
53: wavelength. This property proved to be extremely valuable for
54: manipulation of light energy on subwavelength
55: scales~\cite{sarychev_00_1,stockman_04_2}, miniaturization of optical
56: elements~\cite{engheta_05_1}, and achieving coherent temporal control
57: at remarkably short times~\cite{podolskiy_02_2,stockman_04_1}. SP
58: excitations in ordered one-dimensional arrays of nanoparticles have
59: attracted significant attention in recent years due to numerous
60: potential application in
61: nanoplasmonics~\cite{quinten_98_1,brongersma_00_1,burin_04_1,quidant_04_1,simovski_05_1}.
62: A periodic chain of high conductivity metal nanospheres can be used as
63: an SP wave guide - an analog of an optical
64: waveguide~\cite{maier_03_1}. High-quality SP modes in ordered and
65: disordered chains may be utilized in random lasers~\cite{burin_04_1}.
66: Electromagnetic forces acting on linear chains of nanoparticles can
67: produce the effect of optical trapping~\cite{guillon_06_1}. Various
68: spectroscopic and sensing applications have also been
69: discussed~\cite{markel_93_1,zou_04_1,zou_06_1}.
70:
71: In this paper we study theoretically and numerically propagation of SP
72: excitations in long ordered and disordered chains of nanospheres.
73: Although, under ideal conditions, SP excitations can propagate without
74: loss of energy, in practical situations this is not so. There are two
75: physical effects that can result in decay of SP excitations as they
76: propagate along the chain. The first effect is Ohmic losses due to the
77: finite conductivity of the metal. The second effect is radiative
78: losses due to disorder in the chain (scattering from imperfections).
79: This effect is more subtle and is closely related to the phenomenon of
80: localization. In this paper we discuss both effects and illustrate
81: them with numerical examples.
82:
83: We first focus on decay due to Ohmic losses and show that it can be
84: suppressed at sufficiently large propagation distances. The main idea
85: is based on exploiting an exotic non-Lorentzian resonance in the chain
86: which originates due to radiation-zone interaction of
87: nanoparticles~\cite{markel_93_1,markel_05_2} and can not be understood
88: within the quasistatics, even when both the nanoparticles in the chain
89: and the inter-particle spacing are much smaller than the wavelength.
90: From the spectroscopic point of view, the non-Lorentzian resonances
91: are manifested by very narrow lines in extinction
92: spectra~\cite{markel_93_1,zou_04_1}. One of the authors (V.A.M.) has
93: argued previously that the small integral weight of the spectral lines
94: associated with these resonances precludes them from being excited by
95: a near-field probe~\cite{markel_05_2}. This property would make the
96: non-Lorentzian resonance a curiosity which is rather useless for
97: nanoplasmonics. However, numerical simulations shown below reveal
98: that the corresponding SP has relatively small yet nonzero amplitude
99: and is also characterized by very slow spatial decay. Therefore, in
100: sufficiently long chains, this SP becomes dominant and can propagate,
101: without significant further losses, to remarkable distance. We stress
102: that the non-Lorentzian SP is an excitation specific to discrete
103: systems; it does not exist, for example, in metal nanowires.
104:
105: But the above consideration applies only to ordered chains. Therefore,
106: we consider next the effects of disorder. To isolate radiative losses
107: due to scattering on imperfections from Ohmic losses, we consider
108: nanoparticles with infinite conductivity (equivalently, zero Drude
109: relaxation constant). Although such metals do not exist in nature, an
110: equivalent system can be constructed experimentally by embedding
111: metallic particles into a dielectric medium with positive gain. We
112: show that while the ordinary (defined more precisely in the text
113: below) SP excitations are very sensitive to off-diagonal (position)
114: disorder, the SP due to the non-Lorentzian resonance is not. Diagonal
115: disorder (disorder in nanoparticle properties) is also considered.
116:
117: The paper is organized as follows. In section~\ref{sec:theory}, we
118: describe the theoretical model and introduce basic equations.
119: Conditions for resonance excitation of SPs in a chain are considered
120: ins section~\ref{sec:disp}. Numerical results for propagation in
121: ordered and disordered chains are reported in
122: sections~\ref{sec:prop_ord} and~\ref{sec:prop_dord}, respectively.
123: Finally, section~\ref{sec:summary} contains a summary of obtained
124: results.
125:
126: \section{Theoretical Model}
127: \label{sec:theory}
128:
129: Consider a linear chain of $N$ nanospheres with radiuses $a_n$
130: centered at points $x_n$. We work in the dipole approximation which is
131: valid if $x_{n+1} - x_n \gtrsim (a_{n+1} + a_n) / 2$ and has been
132: widely used in the
133: literature~\cite{brongersma_00_1,burin_04_1,simovski_05_1,weber_04_1,citrin_05_1,citrin_06_1}.
134: The $n$-th nanosphere is then characterized by a dipole moment with
135: amplitude $d_n$ oscillating at the electromagnetic frequency $\omega$.
136: The dipole moments are coupled to each other and to external field by
137: the coupled-dipole equation~\cite{markel_93_1}
138:
139: \begin{equation}
140: \label{CDE}
141: d_n = \alpha_n \left[ E_n + \sum_{n^{\prime}\neq n} G_k(x_n,x_{n^{\prime}})
142: d_{n^{\prime}} \right] \ ,
143: \end{equation}
144:
145: \noindent
146: where $\alpha_n$ is the polarizability of the $n$-th nanosphere, $E_n$
147: is the external electric field at the point $x_n$, $k=\omega/c$ is the
148: free space wave number and $G_k(x,x^{\prime})$ is the appropriate
149: element of the free space, frequency-domain Green's tensor for the
150: electric field. The latter is translationally invariant with respect
151: to spatial variables, namely, $G_k(x,x^{\prime}) = G_k(x-x^{\prime},0)
152: = G_k(x^{\prime}-x,0)$. For an SP polarized perpendicular and parallel
153: to the chain, the respective functions $G_k^{\perp}$ and
154: $G_k^{\parallel}$ are given by
155:
156: \begin{eqnarray}
157: \label{G_perp}
158: G_k^{\perp}(x,0) = \left(\frac{k^2}{\vert x \vert} + \frac{ik}{\vert x
159: \vert^2} - \frac{1}{\vert x \vert^3} \right)\exp( i k \vert x \vert) \ , \\
160: \label{G_par}
161: G_k^{\parallel}(x,0) = \left( - \frac{2ik}{\vert x \vert^2} + \frac{2}{\vert
162: x \vert^3} \right)\exp(i k \vert x \vert) \ .
163: \end{eqnarray}
164:
165: \noindent
166: Polarizability of the $n$-th sphere is taken in the form
167:
168: \begin{equation}
169: \label{alpha_def}
170: \alpha_n = \frac{1}{1/\alpha_n^{\mathrm{(LL)}} - 2ik^3/3} \ ,
171: \end{equation}
172:
173: \noindent
174: where $\alpha_n^{\mathrm{(LL)}}$ is the Lorenz-Lorentz quasistatic
175: polarizability of a sphere of radius $a_n$ and $2ik^3/3$ is the first
176: non-vanishing radiative correction to the inverse polarizability;
177: account of this correction is important to ensure that the system
178: conserves energy~\cite{draine_88_1}. The Lorenz-Lorentz polarizability
179: is given, in terms of the complex permeability of the $n$-th
180: nanosphere $\epsilon_n$, by
181:
182: \begin{equation}
183: \label{alpha_LL}
184: \alpha_n^{\mathrm{(LL)}} = a_n^3 \frac{\epsilon_n - 1}{\epsilon_n + 2} \ .
185: \end{equation}
186:
187: \noindent
188: We further adopt, for simplicity, the Drude model for $\epsilon_n$:
189:
190: \begin{equation}
191: \label{Drude}
192: \epsilon_n = 1 - \frac{\omega_{\mathrm{p}n}^2}{\omega(\omega +
193: i\gamma_n)} \ ,
194: \end{equation}
195:
196: \noindent
197: where $\omega$ is the electromagnetic frequency,
198: $\omega_{\mathrm{p}n}$ is the plasma frequency, and $\gamma_n$ is the
199: Drude relaxation constant in the $n$-th nanosphere. The inverse
200: polarizability of the $n$-th nanosphere is then given by
201:
202: \begin{equation}
203: \label{Re_Im_alpha}
204: \mathrm{Re}\left(\frac{1}{\alpha_n}\right) = \frac{1}{a_n^3} \left[ 1 -
205: \left( \frac{\omega}{\omega_{\mathrm{F}n}} \right)^2 \right] \ , \ \
206: \mathrm{Im} \left(\frac{1}{\alpha_n}\right) = -i \left(\frac{2k^3}{3}
207: + \frac{1}{a_n^3} \frac{\omega\gamma_n}{\omega_{\mathrm{F}n}^2} \right) \ ,
208: \end{equation}
209:
210: \noindent
211: where $\omega_{\mathrm{F}n}=\omega_{\mathrm{p}n}/\sqrt{3}$ is the
212: Frohlich frequency. The SP resonance of an isolated $n$-th nanosphere
213: takes place when $\omega=\omega_{\mathrm{F}n}$. Polarizability
214: $\alpha_n$ at the Frohlich resonance is purely imaginary; if, in
215: addition, there are no Ohmic losses in the material ($\gamma_n=0$),
216: the resonance value of the polarizability becomes $\alpha_n =
217: \alpha_{\mathrm{res}}= -i3/2k^3$, irrespectively of the particle
218: radius.
219:
220: Suppose that SP is excited at a given site (say, $n=n_0$) by a
221: near-field probe. Then the external field can be set to
222: $E_n=E_0\delta_{n,n_0}$. Of course, this is an idealization: the field
223: produced even by a very small near-field tip is, strictly speaking,
224: non-zero at all sites. However, this approximation is physically
225: reasonable because of the fast (cubic) spatial decay of the dipole
226: field in the near-field zone. The solution with $E_n = \delta_{n,n_0}$
227: is, essentially, the Green's function for polarization. We denote this
228: Green's function by ${\mathcal D}_{k}(x_{n},x_{n_{0}})$. It satisfies
229:
230: \begin{equation}
231: \label{GFD}
232: {\mathcal D}_k(x_n,x_{n_0}) = \alpha_n \left[ \delta_{n,n_0} + \sum_{n^{\prime}\neq
233: n} G_k(x_n,x_{n^{\prime}}) {\mathcal D}_k(x_{n^{\prime}},x_{n_0})
234: \right] \ ,
235: \end{equation}
236:
237: \noindent
238: where either (\ref{G_perp}) or (\ref{G_par}) should be used for $G_k$,
239: depending on polarization of the SP. In the case of a finite or
240: disordered chain, one can find ${\mathcal D}_k(x_n,x_{n_0})$ by
241: solving Eq.~(\ref{CDE}) numerically. However, in infinite ordered
242: chains such that $\alpha_n=\alpha=\mathrm{const}$ and
243: $x_{n+1}-x_n=h=\mathrm{const}$, the following analytic solution is
244: obtained by Fourier transform~\cite{markel_05_2}:
245:
246: \begin{equation}
247: \label{d_n_Fur}
248: {\mathcal D}_k(x_n,0) = \int_{-\pi/h}^{\pi/h} {\frac{{\exp(iq x_n)} }{{1/\alpha -
249: S(k,q)}}} {\frac{h dq }{2\pi}} \ ,
250: \end{equation}
251:
252: \noindent
253: where $S(k,q)$ is the ``dipole sum'' given by
254:
255: \begin{equation}
256: \label{S_def}
257: S(k,q) = 2\sum_{n>0} G_{k}(0,x_n)\cos(q x_n) \ .
258: \end{equation}
259:
260: \noindent
261: Obviously, in infinite chains ${\mathcal D}_k(x_n,x_{n^{\prime}}) =
262: {\mathcal D}_k(x_n-x_{n^{\prime}},0)={\mathcal D}_k(x_{n^{\prime}} -
263: x_n,0)$. Note that the dipole sum (\ref{S_def}) is independent of
264: material properties. It can be shown~\cite{markel_93_1} that, for all
265: values of parameters, $\mathrm{Im} S(k,q) \geq -2k^3/3$. The equality
266: holds when $q>k$. This is a manifestation of the fact that SPs with
267: $q>k$ are non-radiating due to the light-cone
268: constraint~\cite{burin_04_1}. Non-radiating modes exist if $\pi/h>k$
269: or, equivalently, if $\lambda>2h$. Obviously, these SPs do not couple
270: to running waves but can be excited by a near-field probe. The
271: dimensionless radiative relaxation parameter can be defined as $Q(k,q)
272: = [\mathrm{Im}S(k,q) + 2k^3/3]/(2k^3/3)$; this factor is identically
273: zero for $q>k$.
274:
275: \section{Dispersion Relations and Resonant Excitation of SP}
276: \label{sec:disp}
277:
278: In this section, we consider periodic chains with $a_n=a$,
279: $\omega_{\mathrm{p}n}=\omega_{\mathrm{p}}$,
280: $\omega_{\mathrm{F}n}=\omega_{\mathrm{F}}$, $\gamma_n=\gamma$ and,
281: consequently, with the constant polarizability $\alpha_n=\alpha$.
282:
283: It follows from formula (\ref{d_n_Fur}) that the wave numbers $q$ of
284: SP excitations that can propagate effectively in an infinite periodic
285: chain are such that $1/\alpha \approx S(k,q)$. This condition must be
286: satisfied for some range of $q$ which is, in some sense, small.
287: Indeed, there is no effective interaction in the chain if $1/\alpha -
288: S(k,q) \approx \mathrm{const}$, in which case integration according to
289: (\ref{d_n_Fur}) yields~\cite{fn2} $d_n \propto \delta_{n,0}$. We now
290: examine the conditions under which the denominator of (\ref{d_n_Fur})
291: can become small. The dispersion relation in the usual sense, e.g.,
292: the dependence of the resonant SP frequency on its wavenumber is
293: obtained by solving the equation $1/\alpha(\omega) - S(\omega/c,q) =
294: 0$. This approach was adopted, for example, in
295: Refs.~\onlinecite{brongersma_00_1,weber_04_1,burin_04_1,simovski_05_1}.
296: Solution to the above equation depends on the model for
297: $\epsilon(\omega)$ and may result in several branches of the complex
298: function $\omega(q)$. Here we adopt a slightly different point of
299: view. Namely, we note that for real frequencies $ \omega$ and wave
300: numbers $q$, the real part of the denominator can change sign while
301: the imaginary part is always non-negative. Physically,
302: $\mathrm{Re}[1/\alpha(\omega) - S(\omega/c,q)]$ can be interpreted as
303: the generalized \emph{detuning} from a resonance while
304: $\mathrm{Im}[1/\alpha(\omega) - S(\omega/c,q)]$ gives total (radiative
305: and absorptive) losses. We thus define the resonance condition to be
306: $\mathrm{Re}[1/\alpha - S(k,q)]=0$ and view $\omega$ and $q$ as
307: independent purely real variables.
308:
309: Plots of the dimensionless function $h^3 S(k,q)$ are shown in
310: Fig.~\ref{fig:disp} for some typical sets of parameters and for two
311: orthogonal polarizations of the SP. Note that the imaginary part of
312: $h^3 S(k,q)$ is related to radiative relaxation parameter $Q(k,q)$
313: [shown in Fig.~\ref{fig:disp}(c,d)] by $h^3 \mathrm{Im}S(k,q) =
314: [2(kh)^3/3][Q(k,q)-1]$. Apart from the very narrow peaks appearing in
315: the case of perpendicular polarization and centered at
316: $q=k$,~\cite{fn3} the numerical values of $h^3 \vert \mathrm{Re}S
317: \vert$ do not exceed, approximately, $6$. On the other hand, according
318: to (\ref{Re_Im_alpha}), we have $h^3\mathrm{Re}(1/\alpha) = (h/a)^3[1
319: -(\omega/\omega_{\mathrm{F}})^2]$. The dipole approximation is valid
320: when $(h/a)^3 \gtrsim 64$. Therefore, if we stay within the range of
321: parameters in which the dipole approximation is valid, the real part
322: of the denominator in (\ref{d_n_Fur}) is relatively small only if
323: $\omega\approx \omega_{\mathrm{F}}$, i.e., if $\omega$ is near the
324: Frohlich frequency of an isolated sphere. This is the case that will
325: be considered below.
326:
327: Consider first oscillations polarized orthogonally to the chain and
328: let $\omega=\omega_{\mathrm{F}}$. The condition of resonant excitation
329: of SP is then $\mathrm{Re}S(k,q) = 0$. It can be seen from
330: Fig.~\ref{fig:disp}(a) that, for sufficiently small values of the
331: dimensionless parameter $kh$, there are two different values of $q$
332: that satisfy the above resonance condition.
333:
334: \begin{figure}[tbp]
335: \centerline{\input{disp_re_ort.tex} \input{disp_re_par.tex}}
336: \centerline{\input{disp_im_ort.tex} \input{disp_im_par.tex}}
337: \caption{(a,b): Real parts of the dipole sum $S(k,q)$ as a function of $q$,
338: for different values of $kh$, as indicated. Sharp peaks
339: corresponding to divergence of $\mathrm{Re}S(k,q)$ at the point
340: $q=k$ are not completely resolved. Quasistatic results obtained in
341: the limit $k=0$ are shown by curves labeled ``QS''. Only half of the
342: Brillouin zone is shown since $ S(k,-q)=S(k,q)$. Polarization of SP
343: is perpendicular (a) and parallel (b) to the chain. (c,d):
344: Dimensionless radiative relaxation parameter
345: $Q(k,q)=[\mathrm{Im}S(k,q) + 2k^3/3]/(2k^3/3)$ as a function of $q$
346: for the same sets of parameters as above. Polarization of SP is
347: perpendicular (c) and parallel (d) to the chain. Quasistatic result
348: is not shown since $Q$ is not defined in the quasistatic limit.}
349: \label{fig:disp}
350: \end{figure}
351:
352: The first solution is $q=q_1\approx 0.47\pi/h$ (for $kh=0.2\pi$). The
353: value of $q_1$ depends only weakly on $kh$, as long as $kh\lesssim
354: 0.2\pi$, and is approximately the same as in the quasistatic limit
355: $kh=0$, in which case $q_1\approx 0.46\pi/h$. We will refer to the SP
356: with this wave number as the ordinary, or the quasistatic SP. This is
357: because propagation of this SP is mediated by near-field interaction,
358: while the far-field interaction is suppressed by destructive
359: interference of waves radiated by different nanoparticles in the
360: chain. Since the wave number of the ordinary SP is greater than $k$,
361: it propagates without radiative losses. However, it may experience
362: spatial decay due to absorption in metal. The characteristic
363: exponential length of decay can be easily inferred from
364: (\ref{d_n_Fur}) by making the quasi-particle pole approximation.
365: Namely, we approximate $S(k,q)$ as
366:
367: \begin{equation}
368: \label{QPP_approx}
369: S(k,q) \approx \mathrm{Re} S(k,q_1) + (q-q_1) \left\vert \frac{\partial
370: \mathrm{Re} S(k,q)}{\partial q} \right \vert_{q=q_1} - i\frac{2k^3}{3} \ .
371: \end{equation}
372:
373: \noindent
374: We then extend integration in (\ref{d_n_Fur}) to the real axis and
375: obtain the following characteristic exponential decay length $\ell$:
376:
377: \begin{equation}
378: \label{ell_def}
379: \ell = \frac{1}{\delta} \left \vert \frac{\partial \mathrm{Re}S(k,q)}
380: {\partial q} \right \vert_{q=q_1} \ ,
381: \end{equation}
382:
383: \noindent
384: where
385:
386: \begin{equation}
387: \label{delta_def}
388: \delta = -\mathrm{Im}(1/\alpha) - 2k^3/3
389: \end{equation}
390:
391: \noindent is a positive parameter characterizing the absorption strength of
392: the nanosphere. In general, it can be shown that $\delta=0$ in non-absorbing
393: particles whose dielectric function $\epsilon(\omega)$ is purely real at the
394: given frequency $\omega$. For the Drude model adopted in this paper, we have
395: $\delta=\omega\gamma/a^3\omega_{\mathrm{F}}^2$. Thus, the ordinary SP can
396: decay exponentially due to absorption in metal with the characteristic scale
397: given by (\ref{ell_def}).
398:
399: The second solution is obtained at $q = q_2 \approx k$, when
400: $\mathrm{Re} S(k,q)$ has a narrow sharp peak as a function of $q$ (for
401: fixed $k$)~\cite{fn1}. This peak is explained by far-field interaction
402: in an infinite chain~\cite{markel_93_1,markel_05_2}. It does not
403: disappear in the limit $kh\rightarrow 0$, but becomes increasingly
404: (super-exponentially) narrow~\cite{markel_05_2}. Note that in the
405: above limit, this peak appears as a singularity of zero integral
406: weight which can not be obtained from the formal quasistatic
407: approximation. We will refer to this SP as extraordinary, or
408: non-quasistatic. It is mediated by the far-field interaction. The
409: latter is important in the case of extraordinary SP because of the
410: constructive interference of far field contributions from all
411: nanoparticles arriving at a given one. Obviously, the extraordinary SP
412: can be excited only in chains with $h<\lambda/2$, where $\lambda$ is
413: the wavelength of light in free space.
414:
415: It is interesting to consider radiative losses of the extraordinary SP. As
416: can be seen from Fig.~\ref{fig:disp}(c), the factor $Q(k,q)$ is
417: discontinuous at $q=k$: it is zero for $q>k$ but positive for $q<k$. Since
418: the extraordinary SP has $q\approx k$, it can experience some radiative
419: losses, although the exact law of its decay depends on parameters of the
420: problem in a complicated manner. This is, in part, related to
421: inapplicability of the quasi-particle pole approximation for evaluating the
422: integral (\ref{d_n_Fur}) in the vicinity of $q\approx q_2$.
423:
424: In the case of oscillations polarized along the chain, the resonance
425: condition can be satisfied only at $q=q_1\approx 0.45\pi/h$. This is
426: the wave number of an ordinary (quasistatic) SP which depends on $k$
427: only weakly, as long as $kh \lesssim 0.2\pi$. However, the
428: extraordinary (non-quasistatic) SP can be excited even for
429: longitudinal oscillations. Mathematically, this can be explained by
430: observing that $\partial \mathrm{Re} S(k,q)/\partial q$ diverges at
431: $q=k$ while $\partial \mathrm{Im} S(k,q)/\partial q$ is discontinuous
432: at $q=k$ and performing integration (\ref{d_n_Fur}) by parts. However,
433: the amplitude of the extraordinary SP is much smaller for longitudinal
434: oscillations than for transverse oscillations; this will be
435: illustrated numerically in the next section.
436:
437: Finally, we note, that when inter-sphere separations become smaller
438: than the radiuses, higher-order multipole resonances can be
439: excited~\cite{sansonetti_80_1,park_04_1,markel_04_3}. In this case,
440: resonant excitation of SP can become possible even at frequencies
441: which are far from the Frohlich resonance of an isolated sphere, e.g.,
442: in the IR part of the spectrum.
443:
444: \section{Propagation in Finite Ordered Chains}
445: \label{sec:prop_ord}
446:
447: We now turn to propagation of SP in ordered chains of finite length $N$. We,
448: however, emphasize that the finite size effects play a very minor role in
449: the computations shown below. Citrin~\cite{citrin_05_1} has studied
450: dispersion relations in finite chains and has found that the infinite-chain
451: limit is reached at $N\approx 10$ (although we anticipate that longer chains
452: are needed for accurate description of the extraordinary SP). In this and
453: following sections, we work with chains of $N\ge 1000$. In this limit,
454: propagation of both ordinary and extraordinary SP is not much different from
455: the case of infinite chains. In particular, we have verified numerically
456: that the Green's function ${\mathcal D}_k(x_n,x_{501})$ (for $kh=0.2\pi$) in a
457: chain of $N=1001$ particles does not differ in any significant way from that
458: in an infinite chain, except for values of $n$ very close to either end of
459: the finite chain. The Green's function $G_k(x_n,x_1)$ (here $n_0=1$ is the
460: end-point of the finite chain) differed by a trivial factor in finite and
461: infinite chains (results not shown). However, proper numerical evaluation of
462: integral (\ref{d_n_Fur}) required very fine discretization of $q$ and was a
463: more demanding and less stable procedure than direct numerical solution of
464: the system of equations (\ref{GFD}).
465:
466: In this section, we take $a_n=a=\mathrm{const}$,
467: $\alpha_n=\alpha=\mathrm{const}$ and $x_n = nh$, $n=1,\ldots,N$. We
468: also assume that $kh=0.2\pi$ and $ h=4a$. Practically, this can be
469: realized for silver particles in a transparent host matrix with
470: refractive index of approximately $n=1.4$ so that the wavelength at
471: the Frohlich frequency is $\lambda_{\mathrm{F}}=2\pi c /
472: \omega_{\mathrm{F}} \approx 400\mathrm{nm}$, the chain spacing is
473: $h=40 \mathrm{nm}=0.1\lambda_{\mathrm{F}}$ and the sphere radius is
474: $a=10\mathrm{nm}=h/4$. The dipole approximation is very accurate for
475: this set of parameters. We then obtain the SP Green's function
476: ${\mathcal D}_k(x_n,x_1)$ by solving Eq.~(\ref{GFD}) numerically.
477:
478: The absolute value of the normalized SP Green's function
479:
480: \begin{equation}
481: \label{F_def}
482: {\mathcal F}_k(x_n) = \frac{{\mathcal D}_k(x_n,x_1)}{{\mathcal D}_k(x_1,x_1)}
483: \end{equation}
484:
485: \noindent
486: in a chain of $N=1000$ nanospheres is shown in Fig.~\ref{fig:ssp_ord}
487: as a function of $x$ (sampled at $x=x_{n}$) for two orthogonal
488: polarizations. Here the frequency of SP was taken to be exactly equal
489: to the Frohlich frequency $\omega_{\mathrm{F}}$ and the Drude
490: relaxation constant was $\gamma =0.002\omega_{\mathrm{F}}$. It can be
491: seen from the figure that two different SP are excited in the system.
492: The first is the ordinary (quasistatic) SP that decays exponentially
493: as $\exp (-x/\ell )$, where $\ell $ is defined by (\ref{ell_def}).
494: Corresponding asymptotes are shown by dotted lines. Note that the
495: quasistatic SP in a finite chain (with the point of excitation
496: coinciding with one of the chain ends) is very well described by the
497: exponential decay formula, even though the latter was obtained for
498: infinite chains. When the amplitude of the ordinary SP becomes
499: sufficiently small, there is a crossover to the extraordinary SP. The
500: decay rate of the extraordinary SP is much slower. We have also
501: confirmed by inspecting the real and imaginary parts of ${\mathcal
502: D}_{k}(x_{n},x_{1})$ (data not shown) that it oscillates at the
503: spatial frequency corresponding to the ordinary SP in the
504: fast-decaying segments of the curves shown in Fig.~\ref{fig:ssp_ord}
505: and with the spatial frequency that corresponds to the extraordinary
506: SP in the slow-decaying segments.
507:
508: \begin{figure}[tbp]
509: %\centerline{\input{ssp_ord.tex}}
510: \centerline{\psfig{file=ssp_ord_PS.ps,bbllx=0bp,bblly=500bp,bburx=596bp,bbury=725bp,clip=t}}
511: \caption{Propagation of a SP in an ordered chain of $N=1000$ nanospheres for
512: orthogonal (ORT) and parallel (PAR) polarization of oscillations
513: with respect to the chain. Parameters: $\omega=\omega_{\mathrm{F}}$,
514: $\gamma/\omega_{\mathrm{F}}=0.002$, $\lambda=10h$, $h=4a$.}
515: \label{fig:ssp_ord}
516: \end{figure}
517:
518: Mathematically, the relatively slow decay of the extraordinary SP can
519: be understood as follows. First, note the the exponential decay of the
520: ordinary SP is, in fact, the result of superposition of an infinite
521: number of plane waves whose wave numbers are in the interval $\Delta
522: q\propto \delta$. The corresponding wave packet decays spatially on
523: scales $\ell \propto 1/\Delta q$. However, in the case of
524: extraordinary SP, $\Delta q$ can not be defined since the
525: corresponding resonance is non-Lorentizian. It can be, however, stated
526: that the extraordinary SP is a superposition of plane waves whose wave
527: numbers are very close to $k$. Since, however, the wavenumbers can
528: still slightly deviate from $k$, some spatial decay at large distances
529: can still occur.
530:
531: The conclusion we can make so far is that the ordinary SP experiences
532: exponential decay along the chain due to Ohmic losses in metal. This
533: decay is very accurately described by the quasi-particle pole
534: approximation. The extraordinary SP has, initially, much smaller
535: amplitude than the ordinary one. This is because the peaks in
536: Fig.~\ref{fig:disp} are very narrow. The quasi-particle pole
537: approximation is invalid for the extraordinary SP and its decay is
538: much less affected by Ohmic losses. As a result, the extraordinary SP
539: decays at a much slower rate and, at sufficiently large propagation
540: distance, begins to dominate. We also note that the extraordinary SP
541: can be excited even for longitudinally-polarized SP, although its
542: amplitude is smaller by some four orders of magnitude than for the
543: case of transverse oscillations.
544:
545: In Fig.~\ref{fig:ssp_ord_gamma}, we illustrate the influence of Ohmic
546: losses on SP propagation. Here we plot $\vert {\mathcal F}_k(x) \vert$
547: as a function of $x$ (sampled at $x=x_n$) for
548: $\omega=\omega_{\mathrm{F}}$ and different values of the ratio
549: $\gamma/\omega_{\mathrm{F}}$. First, in the absence of absorption
550: ($\gamma=0$), the ordinary SP propagates along the chain without
551: decay. Once we introduce absorption, the ordinary SP decays
552: exponentially with the characteristic length scale $\ell$ given by
553: (\ref{ell_def}). Note that, for the specific metal permeability model
554: (\ref{Drude}), $\ell\propto \omega_{\mathrm{F}}/\gamma$. Some
555: dependence of the rate of decay of the extraordinary SP on the ratio
556: $\gamma/\omega_{\mathrm{F}}$ is visible in the case of orthogonal
557: polarization [Fig.~\ref{fig:ssp_ord_gamma}(a)]. However, when the
558: polarization is longitudinal [Fig.~\ref{fig:ssp_ord_gamma}(b)], decay
559: of the extraordinary SP is dominated by radiative losses. In
560: particular, the slow-decaying segments of the curves for
561: $\gamma/\omega_{\mathrm{F}}=0.002$ and
562: $\gamma/\omega_{\mathrm{F}}=0.004$ in Fig.~\ref{fig:ssp_ord_gamma}(b)
563: coincide with high precision.
564:
565: \begin{figure}[tbp]
566: %\centerline{\input{ssp_ord_gamma_ort.tex}\input{ssp_ord_gamma_par.tex}}
567: \centerline{\psfig{file=ssp_ord_gamma.ps,bbllx=0bp,bblly=500bp,bburx=596bp,bbury=725bp,clip=t}}
568: \caption{Same as in Fig.~\ref{fig:ssp_ord_gamma} for different
569: ratios $\gamma/\omega_{\mathrm{F}}$ and for SP
570: polarized orthogonally (a) and parallel (b) to the chain.}
571: \label{fig:ssp_ord_gamma}
572: \end{figure}
573:
574: Next, we study SP propagation for different values of the ratio $
575: \omega/\omega_{\mathrm{F}}$. As noted above, we assume the parameters
576: $ kh=\omega h/c=0.2\pi$ and $h/a=4$ to be fixed. Thus,
577: $\omega/\omega_{\mathrm{F}}$ can vary either due to a change in
578: $\omega_{\mathrm{F}}$ or due to a simultaneous change in $\omega$, $h$
579: and $a$ such that $\omega h = \mathrm{const}$ and
580: $h/a=\mathrm{const}$. It follows from Fig.~\ref{fig:disp}(a) that, for
581: the selected set of parameters, the ordinary plasmon can be excited
582: for $-0.89 < h^3\mathrm{Re}(1/\alpha) <1.56$. This corresponds to
583: $\omega/\omega_{\mathrm{F}}$ lying in the interval
584: $0.988<\omega/\omega_{\mathrm{F}} < 1.007$. In Fig.~\ref{fig:omega},
585: we illustrate propagation of SP excitations for some values of
586: $\omega/\omega_{\mathrm{F}}$ inside this interval, exactly at the
587: lower and upper bounds of this interval, and slightly outside of the
588: interval. Results are shown for two values of absorption strength:
589: $\gamma/\omega_{\mathrm{F}}=0$ and $\gamma/\omega_{
590: \mathrm{F}}=0.002$.
591:
592: \begin{figure}[tbp]
593: %\centerline{\input{ssp_ord_o=0.984oF.tex}\hspace{-0.8cm}\input{ssp_ord_o=0.988oF.tex}}
594: %\vspace{-0.8cm}
595: %\centerline{\input{ssp_ord_o=1.000oF.tex}\hspace{-0.8cm}\input{ssp_ord_o=1.003oF.tex}}
596: %\vspace{-0.8cm}
597: %\centerline{\input{ssp_ord_o=1.007oF.tex}\hspace{-0.8cm}\input{ssp_ord_o=1.010oF.tex}}
598: \centerline{\psfig{file=omega.ps,bbllx=0bp,bblly=300bp,bburx=596bp,bbury=725bp,clip=t}}
599: \caption{Same as in Fig.~\ref{fig:ssp_ord_gamma} for different
600: ratios $\omega/\omega_{\mathrm{F}}$ and $\gamma/
601: \omega_{\mathrm{F}}$, as indicated, and for SP polarized
602: orthogonally to the chain.}
603: \label{fig:omega}
604: \end{figure}
605:
606: First, consider the two cases when $\omega/\omega_{\mathrm{F}}$ is
607: outside of the interval where the ordinary SP can be excited:
608: $\omega/\omega_{ \mathrm{F}} = 0.984$ and $\omega/\omega_{\mathrm{F}}
609: = 1.010$ (correspondingly, $h^3\mathrm{Re}(1/\alpha) = 2.03$ and
610: $h^3\mathrm{Re}(1/\alpha) = -1.29$), shown in
611: Fig.~\ref{fig:omega}(a,f). In the case $\omega/\omega_{\mathrm{F}} =
612: 0.984$, the ordinary SP exhibits very fast spatial decay, which is
613: characteristic for the noninteracting limit when $G_k(n,n_0) \propto
614: \delta_{n,n_0}$. After the initial decay of the ordinary SP, the
615: extraordinary SP becomes dominating. The extraordinary SP decays
616: slowly by means of radiative losses. It is interesting to note that
617: decay of the extraordinary SP is almost unaffected by Ohmic losses in
618: metal, although a noticeable dependence on $\gamma$ appears if we
619: further increase the parameter $\gamma/\omega_{\mathrm{F}}$ by the
620: factor of $10$ (data not shown). A qualitatively similar behavior is
621: obtained at $\omega/\omega_{\mathrm{F}}=1.010$. However, the decay of
622: extraordinary SP in this case is a little faster. Paradoxically, the
623: curve corresponding to $\gamma/\omega_{\mathrm{F}}=0.002$ is slightly
624: higher than the curve corresponding to $\gamma=0$ in
625: Fig.~\ref{fig:omega}(f) (similar peculiarity is seen in
626: Fig.~\ref{fig:omega}(e)).
627:
628: When the ratio $\omega/\omega_{\mathrm{F}}$ is inside the interval
629: where the ordinary SP can be excited [Fig.~\ref{fig:omega}(c,d)], SP
630: propagation is strongly influenced by absorptive losses. In the
631: absence of such losses, the ordinary SP propagates along the chain
632: indefinitely and dominates the extraordinary SP. However, in the
633: presence of even small absorption, the ordinary SP decays
634: exponentially so that, at sufficiently large propagation distances,
635: the extraordinary SP starts to dominate. A qualitatively similar
636: picture is also obtained for the borderline case
637: $\omega/\omega_{\mathrm{F}}=0.984$ [Fig.~\ref{fig:omega}(b)]. In the
638: second borderline case [Fig.~\ref{fig:omega}(e)], SP propagation is
639: more complicated. The wave numbers of both ordinary and extraordinary
640: SPs in this case are close to $k$, so that both can experience
641: radiative decay, as is evident in the case of zero absorption.
642:
643: To conclude this section, we note that propagation of SP excitations
644: in long periodic chains can be characterized by exponential decay.
645: This decay is caused either by absorptive or by radiative losses. At
646: small propagation distances, energy is transported by the ordinary SP
647: excitation, if the ordinary SP can be excited (e.g., if
648: $\mathrm{Re}(1/\alpha)$ is inside the appropriate interval). However,
649: at sufficiently large propagation distances, there is a cross over to
650: transport by means of the extraordinary SP. In this case, propagation
651: is mediated by far-zone interaction and is characterized by slow,
652: radiative decay which is affected by absorptive losses only weakly.
653:
654: We note that exponential decay in ordered chains, if exists, is not
655: caused by Anderson localization, since we have not, so far, introduced
656: disorder into the system. For example, as was discussed in
657: section~\ref{sec:disp}, a linear superposition of delocalized plane
658: wave modes of the form (\ref{d_n_Fur}) can exhibit exponential decay
659: with the characteristic length (\ref{ell_def}). The irreversible
660: exponential decay is, in fact, obtained because the delocalized modes
661: form a truly continuous spectrum (are indexed by a continuous variable
662: $q$). Of course, any superposition of \emph{discrete} delocalized
663: modes would result in Poincare recurrences.
664:
665: \section{Propagation in Disordered Chains}
666: \label{sec:prop_dord}
667:
668: The ordinary (quasistatic) SP propagates in ordered chains without
669: radiative losses due to the perfect periodicity of the lattice.
670: However, once this periodicity is broken, the quasistatic SP can
671: experience radiative losses and spatial decay even in the absence of
672: absorption. Dependence of the radiative quality of SP modes in finite
673: one-dimensional chains on disorder strength was studied in
674: Ref.~\onlinecite{burin_04_1}. It was shown that position disorder
675: tends to decrease the radiative quality factor of initially
676: non-radiating (``bound'') modes. In this section we study how the
677: disorder influences propagation of the SP along the chain and take a
678: separate look at the ordinary and extraordinary SPs. We also consider
679: two types of disorder: off-diagonal and diagonal. Off-diagonal
680: disorder is disorder in particle positions while all particles are
681: identical. Diagonal disorder arises due to differences in particle
682: properties, even if the particle positions are perfectly ordered.
683:
684: The exponential decay due to Ohmic losses in the material can mask the
685: effects of disorder. Therefore, we assume in this section that the
686: nanospheres are non-absorbing, i.e., set $\gamma_n=0$. Physically,
687: absence of absorption can be realized by embedding the chain of
688: nanoparticles in a transparent dielectric host medium with positive
689: gain~\cite{sudarkin_89_1,bergman_03_1,avrutsky_04_1,lawandy_04_1}.
690: Such medium has dielectric permeability $\epsilon_{\mathrm{h}}$ with
691: positive real and negative imaginary parts. The gain can be tuned so
692: that the effective permeability of inclusions
693: $\epsilon/\epsilon_{\mathrm{h}}$ is purely real and negative. We also
694: work in the regime $kh=0.2\pi $, $h=4a$.
695:
696: \subsection{Off-Diagonal Disorder}
697: \label{subsec:offdiag}
698:
699: Off-diagonal disorder is disorder in particle position. It is called
700: ``off-diagonal'' because it affects only off-diagonal elements of the
701: interaction matrix $G_k(x_n,x_{n^{\prime}})$. In the simulations shown
702: below, coordinates of particles in a disordered chain were taken to be
703: $ x_n=h(n+\xi_n)$ where $\xi_n$ is a random variable evenly
704: distributed in the interval $[-A,A]$. The random numbers $\xi_n$ are
705: taken to be mathematically independent. Therefore, the disorder is
706: uncorrelated.
707:
708: The first example concerns propagation in an non-absorbing chain of
709: $N=10,000$ nanospheres excited exactly at the Frohlich frequency
710: $\omega = \omega_{\mathrm{F}}$. Numerical results for different levels
711: of disorder are illustrated in Fig.~\ref{fig:ssp_dord_ort_a1} where we
712: plot $\vert {\mathcal F}_k(x) \vert$ as a function of $x$ (sampled at
713: $x=x_n$).
714:
715: \begin{figure}[tbp]
716: %\centerline{\input{ssp_dord_ort_a1=0.00.tex}}
717: %\vspace{-1cm}
718: %\centerline{\input{ssp_dord_ort_a1=0.01.tex}}
719: %\vspace{-1cm}
720: %\centerline{\input{ssp_dord_ort_a1=0.02.tex}}
721: %\vspace{-1cm}
722: %\centerline{\input{ssp_dord_ort_a1=0.04.tex}}
723: %\vspace{-1cm}
724: %\centerline{\input{ssp_dord_ort_a1=0.08.tex}}
725: %\vspace{-1cm}
726: %\centerline{\input{ssp_dord_ort_a1=0.16.tex}}
727: \centerline{\psfig{file=ssp_dord_ort_a1.ps,bbllx=100bp,bblly=200bp,bburx=496.bp,bbury=750bp,clip=t}}
728: \caption{Propagation of SP in a chain of $N=10,000$ non-absorbing ($
729: \gamma=0$) nanospheres for different levels of \emph{off-diagonal}
730: disorder $ A$, as indicated. SP polarization is orthogonal to the
731: chain. Other parameters: $kh=0.2\pi$, $\omega/\omega_{\mathrm{F}
732: }=1$, $h/a=4$.}
733: \label{fig:ssp_dord_ort_a1}
734: \end{figure}
735:
736: One obvious conclusion that can be made from inspection of
737: Fig.~\ref{fig:ssp_dord_ort_a1} is that disorder causes spatial decay
738: of SP. Since the system has no absorption, energy is lost to
739: radiation. However, the exact law of decay strongly depends on
740: particular realization of disorder. In
741: Figs.~\ref{fig:ssp_dord_ort_a1=0.01_R}
742: and~\ref{fig:ssp_dord_ort_a1=0.02_R}, we plot the function
743: $\vert{\mathcal F}_k(x) \vert$ for $A=0.01$ and $A=0.02$,
744: respectively, and for three different realizations of disorder
745: (without the use of logarithmic scale). Giant fluctuations in the
746: amplitude of transmitted SP are quite apparent. In all cases, after
747: some initial growth, the amplitude decays, although some random
748: recurrences (due to re-excitation) can take place. However, the
749: amplitude $\vert{\mathcal F}_k(x) \vert$ at a given site $x=x_n$
750: strongly depends on realization of disorder and can be very far from
751: its ensemble average, even at very large propagation distances.
752: Therefore, the function ${\mathcal F}_k(x)$ appears to be \emph{not
753: self-averaging}. In particular, any particular realization of the
754: intensity ${\mathcal I}_k(x)=\vert {\mathcal F}_k(x) \vert^2$ does not
755: satisfy either the radiative transport equation or the diffusion
756: equation. The absence of transport and self-averaging is
757: characteristic for Anderson localization. We note that the
758: localization discussed here is an interference (radiative) effect
759: which is different from localization of quasistatic polarization modes
760: studied in Refs.~\onlinecite{stockman_01_1,genov_05_1}. Localization
761: properties of the eigenmodes in disordered chains is a subject of
762: separate investigation and will be reported elsewhere.
763:
764: \begin{figure}[tbp]
765: %\centerline{\input{ssp_dord_ort_a1=0.01_R0.tex}}
766: %\vspace{-1cm}
767: %\centerline{\input{ssp_dord_ort_a1=0.01_R1.tex}}
768: %\vspace{-1cm}
769: %\centerline{\input{ssp_dord_ort_a1=0.01_R2.tex}}
770: \centerline{\psfig{file=ssp_dord_ort_a1=0.01_R.ps,bbllx=0bp,bblly=450bp,bburx=596bp,bbury=710bp,clip=t}}
771: \caption{Propagation of SPs in a chain with off-diagonal disorder of
772: amplitude $A=0.01$ for different random realizations of disorder.
773: Other parameters same as in Fig.~\ref{fig:ssp_dord_ort_a1}.}
774: \label{fig:ssp_dord_ort_a1=0.01_R}
775: \end{figure}
776:
777: \begin{figure}[tbp]
778: %\centerline{\input{ssp_dord_ort_a1=0.02_R0.tex}}
779: %\vspace{-1cm}
780: %\centerline{\input{ssp_dord_ort_a1=0.02_R1.tex}}
781: %\vspace{-1cm}
782: %\centerline{\input{ssp_dord_ort_a1=0.02_R2.tex}}
783: \centerline{\psfig{file=ssp_dord_ort_a1=0.02_R.ps,bbllx=0bp,bblly=450bp,bburx=596bp,bbury=710bp,clip=t}}
784: \caption{Same as in Fig.~\ref{fig:ssp_dord_ort_a1=0.01_R}, but for $
785: A=0.02$.}
786: \label{fig:ssp_dord_ort_a1=0.02_R}
787: \end{figure}
788:
789: The situation is complicated by the presence of two types of SP
790: excitations. One can argue that localization properties of these two
791: types of SPs might be different. To investigate this possibility, we
792: perform two types of numerical experiments. First, we repeat
793: simulations illustrated in Fig.~\ref{fig:ssp_dord_ort_a1} but for
794: $\omega/\omega_{\mathrm{F}}=0.948$. In this case, the ordinary SP is
795: not excited in ordered chains (see Fig.~\ref{fig:omega}(a)). Results
796: are shown in Fig.~\ref{fig:ssp_dord_ort_a1_o=0.984} . It can be
797: concluded from the figure that off-diagonal disorder does not result
798: in additional decay of the extraordinary SP. In fact, the curves with
799: $A=0$, $A=0.1$ and $A=0.04$ are indistinguishable. This should be
800: contrasted with the case $\omega/\omega_{\mathrm{F}}=1$, when, at the
801: level of disorder $A=0.02$, spatial decay is already well manifested.
802: The conclusion one can make is that the extraordinary SP excitations
803: do not experience localization due to uncorrelated off-diagonal
804: disorder in the chain. Therefore, spatial decay seen in
805: Figs.~\ref{fig:ssp_dord_ort_a1} and~\ref{fig:ssp_dord_ort_a1=0.02_R}
806: is decay of the ordinary SP. At very large propagation distances,
807: ${\mathcal F}_k(x)$ continues to oscillate around the baseline of the
808: extraordinary SP.
809:
810: \begin{figure}[tbp]
811: \centerline{\input{ssp_dord_ort_a1_o=0.984.tex}}
812: \caption{Same as in Fig.~\ref{fig:ssp_dord_ort_a1} but for
813: $\omega/\omega_{\mathrm{F}} = 0.984$ and for different levels $A$ of
814: off-diagonal disorder, as indicated. Note that the curves for $A=0$
815: and $ A=0.01$ and $A=0.04$ are indistinguishable.}
816: \label{fig:ssp_dord_ort_a1_o=0.984}
817: \end{figure}
818:
819: The second numerical experiment that can elucidate the influence of
820: disorder on ordinary and extraordinary SPs is simulation of an
821: extinction measurement when the chain is excited, instead of a
822: nera-field tip, by a plane wave $E_0 \exp (iqx)$. Namely, we will look
823: at the dependence of the specific (per one nanosphere) extinction
824: cross section $\sigma_e$ on $q$. There are two different
825: experimental setups that can be used to measure $\sigma_e(q)$. When
826: $q$ is in the interval $0\leq q<k$, this experiment can be carried out
827: simply by varying the angle between the incident beam and the chain.
828: However, values $q>k$ are not accessible in this experiment. In this
829: case, the chain can be placed on a dielectric substrate and excited by
830: evanescent wave originating due to the total internal reflection of
831: the incident beam. The maximum longitudinal wave number of the
832: evanescent wave is $nk$, where $n>1$ is the refractive index of the
833: substrate. We note that it is not realistically possible to access all
834: wave numbers up to $q=\pi /h$ in this way because, in the particular
835: case $kh=0.2\pi $, this would require the refractive index $n=5$.
836: Refractive indices of such magnitude are not achievable in the optical
837: range. However, in a numerical simulation, we can assume that a
838: hypothetical transparent substrate with $n=5$ exists. Besides, all
839: wave numbers $q$ in the first Brillouin zone of the lattice can be
840: accessible for a different choice of parameters (particularly, for
841: larger values of $kh$).
842:
843: The specific extinction $\sigma_e(q)$ is given by the following formula:
844:
845: \begin{equation}
846: \label{sigma_e_def}
847: \sigma_e(q) = \frac{4\pi k}{N\vert E_0 \vert^2} \sum_{n=1}^N E_0^*\exp(-i q
848: x_n) d_n \ ,
849: \end{equation}
850:
851: \noindent
852: where $d_n$ is the solution to Eq.~(\ref{CDE}) with the right-hand
853: side $E_n=E_0 \exp(i q x_n)$. In Fig.~\ref{fig:qspect_a1}, we plot the
854: dimensionless quantity $h^{-2}\sigma_e(q)$ as a function of $q$ for
855: transverse oscillations in partially disordered non-absorbing chains.
856: Two peaks are clearly visible in the spectrum. The first peak at
857: $q=q_1\approx 0.47\pi/h$ corresponds to excitation of the ordinary SP.
858: The second peak at $ q=q_2\approx k$ corresponds to excitation of the
859: extraordinary SP. In infinite, periodic and non-absorbing chains, the
860: spectrum has a simple pole at $q=q_1$.~\cite{fn4} In the finite chain
861: with $N=10,000$, the singularity is replaced by a very sharp maximum.
862: Introduction of even slight disorder tends to further broaden and
863: randomize this peak. Obviously, this broadening, as well as the
864: randomization of the spectrum in the vicinity of $ q=q_1$, result in
865: spatial decay of the ordinary SP excitated by a near-field tip.
866: However, the peak corresponding to the extraordinary SP is almost
867: unaffected by disorder. Consequently, the extraordinary SP does not
868: experience localization and related spatial decay when off-diagonal
869: disorder is introduced into the system.
870:
871: \begin{figure}[tbp]
872: \centerline{\input{qspect_a1.tex}}
873: \caption{Specific extinction $\sigma_e$ as a function of lateral
874: wave number of incident wave, $q$, for different levels of
875: off-diagonal disorder, $A$.}
876: \label{fig:qspect_a1}
877: \end{figure}
878:
879: The physical reason why the extraordinary SP is not affected by
880: disorder is quite straightforward. This SP is mediated by
881: electromagnetic waves in the far (radiation) zone that arrive at a
882: given nanosphere from all other nanospheres. The synchronism condition
883: (that all these secondary waves arrive in phase) is not affected by
884: disorder as long as the displacement amplitudes $\xi_n$ are small
885: compared to the wavelength. In the numerical examples of this section,
886: wavelength is ten times larger than the inter-particle spacing, which
887: is, in turn, much larger than the displacement amplitudes.
888: Additionally, effects of disorder are expected to be averaged if the
889: amplitudes $\xi_n$ are mathematically independent because the
890: scattered field at a given nanosphere is a sum of very large number of
891: secondary waves. In contrast, the ordinary SP is mediated by
892: near-field interactions which are very sensitive to even slight
893: displacements of nanospheres. Besides, the scattered field at a given
894: nanosphere (when the ordinary SP propagates in the chain) is a rapidly
895: converging sum of secondary fields, so that only a few terms in this
896: sum are important and no effective averaging takes place.
897:
898: \subsection{Diagonal Disorder}
899: \label{subsec:diag}
900:
901: Diagonal disorder is disorder in the properties of nanospheres. There
902: are several possibilities for introducing such disorder.
903:
904: First, the spheres can be polydisperse, i.e., have different radiuses $a_n$.
905: This is, however, not a truly diagonal disorder. Indeed, polarizability of $
906: n $-th nanosphere can be written in this case as $f_n \langle \alpha \rangle$
907: where $f_n = (a_n /\langle a \rangle)^3$ and $\langle a \rangle$ is the
908: average radius (note that the radiative correction to $1/\alpha_n$ can be
909: included into the dipole sum $S(k,q)$, so that this analysis remains valid
910: even when this correction is important). The factors $f_n$ are all
911: positive-definite which allows one to introduce a simple transformation of
912: Eq.~(\ref{CDE}) which removes the diagonal disorder~\cite{perminov_04_1}.
913: Then the disorder becomes effectively off-diagonal. More importantly, one
914: can retain a well-defined spectral parameter of the theory, $1/\langle
915: \alpha \rangle$. We, therefore, do not consider polydispersity in this
916: section.
917:
918: The second possibility is variation of the absorptive parameter $\gamma$.
919: This effect that can be practically important. Yet, we are interested in
920: propagation in the absence of Ohmic losses and, therefore, set $\gamma_n=0$.
921:
922: The third, and the most fundamental, reason for diagonal disorder is
923: variation of the Frohlich frequency of nanospheres. Namely, we take
924: the Frohlich frequency of the $n$-th particle to be
925: $\omega_{\mathrm{F} n}=\langle \omega_{\mathrm{F}} \rangle (1 +
926: \xi_n)$, where $\xi_n$ are statistically independent random variables
927: evenly distributed in the interval $[-A,A]$ and $\omega/ \langle
928: \omega_{\mathrm{F}} \rangle = 1$. The factors $f_n$ in this case are
929: no longer positive-definite and the transformation of
930: Ref.~\onlinecite{perminov_04_1} can not be applied. The most profound
931: consequence of introducing the diagonal disorder is that the spectral
932: parameter such as $1/\alpha$ is no longer well defined. As long as the
933: disorder amplitude is relatively small, one can view $1/\langle \alpha
934: \rangle$ as an approximate spectral parameter. However, as the
935: amplitude of disorder increases, this approach becomes invalid. We
936: have seen in section~\ref{sec:prop_ord} that variation of the ratio
937: $\omega/\omega_{\mathrm{F}}$ in the interval $0.988 <
938: \omega/\omega_{\mathrm{F}} < 1.007$ can result in dramatic changes in
939: the way an SP excitation propagates along the chain. For
940: $\omega/\omega_{\mathrm{F}}$ outside of this interval, ordinary SP
941: could not be effectively excited. However, in
942: section~\ref{sec:prop_ord}, variation of $\omega/\omega_{\mathrm{F}}$
943: applied to all nanospheres simultaneously. We now introduce random
944: uncorrelated variation of this ratio for each individual nanosphere.
945: The effects of such disorder are difficult to predict theoretically.
946: We can, however, expect that these effects become dramatic for
947: $A\gtrsim 0.01$ since, in this case, the ratio
948: $\omega/\omega_{\mathrm{F} n}$ can be outside of the interval $[0.988,
949: 1.007]$.
950:
951: We first adduce in Fig.~\ref{fig:ssp_dord_ort_a3} the same
952: dependencies as shown in Fig.~\ref{fig:ssp_dord_ort_a1} but for
953: different levels of \emph{diagonal} disorder. The results appear to
954: be, qualitatively, quite similar, although in the case $A=0.016$,
955: there is no visible trend for $x/h>3000$. We then investigate whether
956: the extraordinary SP remains insensitive to diagonal disorder. Data
957: analogous to those shown in Fig.~\ref{fig:ssp_dord_ort_a1_o=0.984},
958: but for diagonal disorder, are presented in
959: Fig.~\ref{fig:ssp_dord_ort_a3_o=0.984}. It can be seen that the
960: influence of diagonal disorder on the extraordinary SP is stronger
961: than that of the off-diagonal disorder. When the disorder amplitude
962: $A$ exceeds $0.01$, the influence becomes quite dramatic. But
963: paradoxically, at $A=0.016$, the average decay rate is much slower
964: than for $A=0.008$, although the amplitude of fluctuations is much
965: larger.
966:
967: \begin{figure}[tbp]
968: %\centerline{\input{ssp_dord_ort_a3=0.000.tex}}
969: %\vspace{-1cm}
970: %\centerline{\input{ssp_dord_ort_a3=0.001.tex}}
971: %\vspace{-1cm}
972: %\centerline{\input{ssp_dord_ort_a3=0.002.tex}}
973: %\vspace{-1cm}
974: %\centerline{\input{ssp_dord_ort_a3=0.004.tex}}
975: %\vspace{-1cm}
976: %\centerline{\input{ssp_dord_ort_a3=0.008.tex}}
977: %\vspace{-1cm}
978: %\centerline{\input{ssp_dord_ort_a3=0.016.tex}}
979: \centerline{\psfig{file=ssp_dord_ort_a3.ps,bbllx=100bp,bblly=200bp,bburx=496.bp,bbury=750bp,clip=t}}
980: \caption{Same as in Fig.~\ref{fig:ssp_dord_ort_a1}, but for
981: different levels $A$ of \emph{diagonal} disorder.}
982: \label{fig:ssp_dord_ort_a3}
983: \end{figure}
984:
985: \begin{figure}[tbp]
986: \centerline{\input{ssp_dord_ort_a3_o=0.984.tex}}
987: \caption{Same as in Fig.~\ref{fig:ssp_dord_ort_a3} but for $
988: \omega/\omega_{\mathrm{F}}=0.984$ and for different levels $A$ of
989: off-diagonal disorder, as indicated. Note that the curves for $A=0$
990: and $ A=0.01$ are indistinguishable.}
991: \label{fig:ssp_dord_ort_a3_o=0.984}
992: \end{figure}
993:
994: To see why this happens, we look at the extinction spectra
995: $\sigma_e(q)$. These are plotted in Fig.~\ref{fig:qspect_a3}. The
996: fundamental difference between the off-diagonal and diagonal disorder
997: is clearly revealed by comparing this figure to
998: Fig.~\ref{fig:qspect_a1}. Namely, the effect of diagonal disorder is
999: not only to broaden and randomize the peak at $q=q_1$, but also to
1000: shift it towards the peak corresponding to the extraordinary SP. At
1001: sufficiently large levels of disorder ($A \geq 0.016$), the separate
1002: peaks disappear and a broad structure emerges. At this point, ordinary
1003: and extraordinary SP can no longer be distinguished. Correspondingly,
1004: the ordinary and extraordinary SP are effectively mixed in the
1005: $A=0.016$ curve in Fig.~\ref{fig:ssp_dord_ort_a3_o=0.984}, while fo
1006: smaller amplitudes of the diagonal disorder, the extraordinary SP is
1007: excited predominatlly.
1008:
1009: \begin{figure}[tbp]
1010: \centerline{\input{qspect_a3.tex}}
1011: \caption{Specific extinction $\sigma_e$ as a function of lateral
1012: wave number of incident wave, $q$, for different levels of
1013: off-diagonal disorder, $A$.}
1014: \label{fig:qspect_a3}
1015: \end{figure}
1016:
1017: \section{Summary}
1018: \label{sec:summary}
1019:
1020: We have considered surface plasmon (SP) propagation in a linear chain
1021: of metal nanoparticles. Computer simulations reveal the existence of
1022: two types of plasmons: ordinary (quasistatic) and extraordinary
1023: (non-quasistatic) SPs.
1024:
1025: The ordinary SP is characterized by short-range interaction of
1026: nanospheres in a chain. The retardation effects are inessential for
1027: its existence and properties. The ordinary SP behaves as a quasistatic
1028: excitation. The ordinary SP can not radiate into the far zone in
1029: perfectly periodic chains because its wave number is larger than the
1030: wavenumber $k=\omega/c$ of free electromagnetic waves. However, it can
1031: experience decay due to absorptive dissipation in the material.
1032:
1033: The second, extraordinary, SP propagates due to long-range (radiation
1034: zone) interaction in a chain. Its excitation is possible due to the
1035: existence of the non-Lorentzian optical resonance in the chain
1036: introduced in Ref.~\onlinecite{markel_05_2}.
1037: The extraordinary SP may experience some radiative loss but is much
1038: less afected by absorptive dissipation and disorder. As a result, it
1039: can propagate to much larger distances along the chain. The
1040: extraordinary SP can be used to guide energy or information in
1041: all-optical integrated photonic systems.
1042:
1043: We have also considered the effects of disorder and localization of
1044: the ordinary and extraordinary SPs. Results of numerical simulations
1045: suggest that even small disorder in the position or properties of
1046: nanoparticles results in localization of the ordinary SP. However, the
1047: extraordinary SP appears to remain delocalized for all types and
1048: levels of disorder considered in the paper.
1049:
1050: The authors can be reached at vmarkel@mail.med.upenn.edu (VAM) and
1051: asarychev@ethertronics.com (AKS).
1052:
1053: \bibliographystyle{prsty}
1054: \bibliography{abbrev,master,local}
1055:
1056: \end{document}
1057: