1: %% This document created by Scientific Word (R) Version 3.5
2:
3: \documentclass[aps,prb,showpacs]{revtex4}%
4: \usepackage{amsmath}
5: \usepackage{graphicx}
6: %\usepackage[LY1]{fontenc}
7: %\usepackage[dvipsone]{hyperref}
8: \usepackage{amsfonts}%
9: \usepackage{amssymb}%
10: %\usepackage[LY1]{mathtime}
11: %TCIDATA{OutputFilter=latex2.dll}
12: %TCIDATA{CSTFile=revtex4.cst}
13: %TCIDATA{Created=Sunday, July 08, 2001 07:46:04}
14: %TCIDATA{LastRevised=Monday, April 08, 2002 12:45:31}
15: %TCIDATA{<META NAME="GraphicsSave" CONTENT="32">}
16: %TCIDATA{<META NAME="DocumentShell" CONTENT="Articles\SW\REVTeX 4 (Test Version)">}
17: %TCIDATA{Language=American English}
18: \newtheorem{theorem}{Theorem}
19: \newtheorem{acknowledgement}[theorem]{Acknowledgement}
20: \newtheorem{algorithm}[theorem]{Algorithm}
21: \newtheorem{axiom}[theorem]{Axiom}
22: \newtheorem{claim}[theorem]{Claim}
23: \newtheorem{conclusion}[theorem]{Conclusion}
24: \newtheorem{condition}[theorem]{Condition}
25: \newtheorem{conjecture}[theorem]{Conjecture}
26: \newtheorem{corollary}[theorem]{Corollary}
27: \newtheorem{criterion}[theorem]{Criterion}
28: \newtheorem{definition}[theorem]{Definition}
29: \newtheorem{example}[theorem]{Example}
30: \newtheorem{exercise}[theorem]{Exercise}
31: \newtheorem{lemma}[theorem]{Lemma}
32: \newtheorem{notation}[theorem]{Notation}
33: \newtheorem{problem}[theorem]{Problem}
34: \newtheorem{proposition}[theorem]{Proposition}
35: \newtheorem{remark}[theorem]{Remark}
36: \newtheorem{solution}[theorem]{Solution}
37: \newtheorem{summary}[theorem]{Summary}
38: \newenvironment{proof}[1][Proof]{\textbf{#1.} }{\ \rule{0.5em}{0.5em}}
39:
40: \begin{document}
41:
42: \title{Surface scattering analysis of phonon transport in the quantum limit
43: using an elastic model}
44: \author{D.\ H.\ Santamore}
45: \author{M.\ C.\ Cross} \affiliation{Department of Physics,
46: California Institute of Technology 114-36, Pasadena, CA 91125}
47: \date{\today }
48:
49: \begin{abstract}
50: We have investigated the effect on phonon energy transport in mesoscopic
51: systems and the reduction in the thermal conductance in the quantum limit due
52: to phonon scattering by surface roughness using full 3-dimensional elasticity
53: theory for an elastic beam with a rectangular cross-section. At low
54: frequencies we find power laws for the scattering coefficients that are
55: strongly mode dependent, and different from the $\omega^{2}$ dependence,
56: deriving from Rayleigh scattering of scalar waves, that is often assumed. The
57: scattering gives contributions to the reduction in thermal conductance with
58: the same power laws. At higher frequencies the scattering coefficients becomes
59: large at the onset frequency of each mode due to the flat dispersion here. We
60: use our results to attempt a quantitative understanding of the suppression of
61: the thermal conductance from the universal value observed in experiment.
62:
63: \end{abstract}
64: \pacs{63.22.+m, 63.50.+x, 68.65.-k, 43.20.Fn}
65: \maketitle
66:
67: \section{Introduction}
68:
69: Landauer's formulation of quantum transport showed that when elastic
70: scattering dominates, the electrical conductance can be related to the
71: transmission coefficient of the electron waves\cite{L57}. In the ideal case of
72: no scattering, this leads to a universal conductance that is quantized in
73: units of $e^{2}/h$ at low temperatures, with an additional quantum of
74: conductance added as each channel or mode of the conductance pathway opens up.
75: The application of similar ideas to the phonon counterpart, namely thermal
76: conductance, was recently derived by a number of authors\cite{ACR98, RK98,
77: B99}, and is now recognized\cite{BV00} to be related to earlier work on the
78: entropy transport at low temperatures\cite{P83}. Rego and Kirczenow have
79: extended the concept of the universality of the thermal conductance to
80: particles of arbitrary statistics (anyons)\cite{RK99}.
81:
82: In the case of electrical resistance, the chemical potential or
83: the number of conducting modes can be varied at very low
84: temperatures, giving sharp jumps between various quantized values
85: of the resistance. On the other hand, thermal transport by phonons
86: necessarily requires nonzero temperatures to populate the modes of
87: the conducting pathway, and the width of the Bose distribution
88: function smears out the quantization of the conductance. Only at
89: very low temperatures, where just the modes of the conducting
90: pathway with zero frequency at long wavelengths contribute to the
91: thermal conductance, the quantization of the ideal conductance
92: becomes apparent in a universal thermal conductance $N_{0}K_{u}$
93: with $K_{u}=(\pi^{2}/3)k_{B}^{2}T/h$ the universal conductance per
94: mode, with $k_{B}$ Boltzmann's constant and $h$ Planck's constant,
95: and $N_{0}$ is the number of modes with zero frequency at long
96: wavelengths, which is four for a freely suspended elastic beam
97: connecting the two thermal reservoirs. Note that this value of the
98: low temperature conductance in the absence of scattering is
99: independent of the dimensions and elastic properties of the
100: thermal pathway.
101:
102: A low temperature thermal conductance consistent with the
103: predicted universal value was measured by Schwab et
104: al.\cite{SHWR00} in experiments on a lithographically defined
105: mesoscopic suspended beam (of dimensions about
106: $1\mathrm{\mu m}\times200\mathrm{nm}\times60\mathrm{nm}$). Whilst
107: their elegant experiment displays the universality of ballistic
108: phonon transport, the experiment also showed a \emph{decrease} in
109: the thermal conductance below the universal value in the
110: temperature range of $0.08\mathrm{K}<T<0.4\mathrm{K}$ that cannot
111: be explained by the ballistic theory, since in this theory an
112: increase in the thermal conductance is expected as the temperature
113: is raised and more modes are excited. The decrease in thermal
114: conductance is presumably associated with the scattering of the
115: thermal phonons, and can be understood using the ideas of Landauer
116: in terms of the scattering coefficient of the vibrational waves.
117: This is the topic of the present paper.
118:
119: In this paper we calculate the effect on the low temperature thermal
120: conductance of the scattering of the thermal phonons by surface roughness,
121: which is likely to be the major source of scattering in mesoscopic samples.
122: The scattering of scalar waves, described by the simple wave equation, in
123: waveguides with rough surfaces has been investigated by many workers,
124: including ourselves, using both numerical and analytic
125: methods\cite{S97,KFFL99,SFMY99,B99,SC00}. However, for the low frequency modes
126: of interest in the low temperature thermal conductance, the physical
127: vibrational waves have quite different properties than the waves in the scalar
128: model. For example the dispersion relations of the modes are different, with
129: two of the four modes with zero long-wavelength frequency having a quadratic
130: dispersion at small wave vectors, rather than the linear dependence given by
131: the simple scalar theory. To understand the experimental results
132: quantitatively, a more accurate treatment of the vibrational waves is needed.
133: At low temperatures the wavelengths of the thermally excited modes are large
134: compared with the atomic spacing, and so a treatment based on the equations of
135: macroscopic elasticity theory is appropriate. Blencowe\cite{B95} has
136: considered the scattering of elastic waves in a thin plate waveguide with
137: rough surfaces, but prior to our work, the scattering of elastic waves
138: confined in a beam-like wave guide with rough surfaces has not been considered.
139:
140: Previously, we have investigated the effect of surface scattering
141: on the low temperature thermal conductance using the scalar wave
142: model\cite{SC00}. In that paper we noted the apparent discrepancy
143: between the results of scalar model with a simple assumption for
144: the nature of the surface roughness and the data by Schwab et al.
145: below a temperature of $0.1\mathrm{K}$: the data seemed to show a
146: delay of the onset of scattering as the temperature increased that
147: was not predicted by the model. However, since the scalar model
148: does not properly account for the properties of the elastic waves,
149: it was not clear whether this discrepancy is due to an inadequate
150: modelling of the surface roughness, or the flaw in the description
151: of the waves themselves. To resolve this matter, and obtain a more
152: accurate account of the scattering of the waves by rough surfaces,
153: we develop a theory of based on the full elasticity equations, and
154: use this to calculate the thermal conductance at low temperatures.
155: A short version of this work has been previously
156: published\cite{SC01l}.
157:
158: In Section\ \ref{Sec_Formalism}, the scattering of elastic waves confined to a
159: beam of rectangular cross-section with rough surfaces is calculated using the
160: full three dimensional elasticity theory. We use a Green theorem approach, and
161: calculate the scattering coefficient to quadratic order in the amplitude of
162: the surface roughness. These results are quite general, but rather intractable
163: for further progress, since the structure of the modes in an elastic beam
164: cannot be determined in closed form. Thus in Section\ \ref{Sec_plate} we
165: reduce the expressions to a thin plate limit to provide a closed form for the
166: displacement fields, and to obtain analytical expressions for the scattering
167: behavior. In Section \ref{Sec_scattering} the general behavior of the
168: scattering and the effect on the thermal conductance is analyzed in detail,
169: using a simple description of the surface roughness, to investigate the
170: physical consequences of the novel features of the elastic waves. In Section
171: \ref{Sec_experiment} we use our theory to attempt to fit the data of Schwab et
172: al\cite{SHWR00} using more realistic descriptions of the surface roughness. A
173: number of the more difficult issues that arise in the elasticity theory are
174: described in appendices.
175:
176: Although our main interest is the scattering of thermally excited vibrational
177: waves in mesoscopic systems at low temperatures, the formulation of the
178: surface scattering is quite general, and can be applied to other situations,
179: such as the scattering of mechanically excited modes in macroscopic samples
180: for example.
181:
182: \section{General Formalism}
183:
184: \label{Sec_Formalism}
185:
186: \subsection{The model}
187:
188: The main focus of this paper is the effect of surface roughness on the low
189: temperature thermal conductance of mesoscopic structures. The geometry we
190: consider is a freely suspended elastic beam, which we call the bridge,
191: connecting two thermal reservoirs. We will consider a beam of rectangular
192: cross-section of dimensions width $W$ (in the $y$ direction) and depth $d$ (in
193: the $z$ direction). Mesoscopic structures are often produced lithographically
194: from epitaxially grown material. We choose a convention that the depth is the
195: dimension in the growth direction, and the width in the lithographically
196: defined transverse direction. We define the length of the rectangular beam of
197: nominally uniform cross section as $L$. In practice the bridge may be joined
198: to the reservoirs smoothly, by a portion of continuously growing width, to
199: eliminate or reduce the scattering of the vibration modes off a sharp
200: junction. We will suppose that the scattering by roughness is important only
201: in some narrower portion of length $L$.
202:
203: The thermal conductance is given by the expression\cite{ACR98, RK98, B99}%
204:
205: \begin{equation}
206: K={\frac{\hbar^{2}}{{k_{B}T^{2}}}}\sum_{m}{\frac{1}{{2\pi}}}\int_{\omega_{m}%
207: }^{\infty}\mathcal{T}_{m}(\omega){\frac{\omega^{2}e^{\beta\hbar\omega}%
208: }{{(e^{\beta\hbar\omega}-1)^{2}}}}d\omega, \label{eq:conductance}%
209: \end{equation}
210: where $\omega_{m}$ is the cutoff frequency of the $m$ th mode, $\beta
211: =1/(k_{B}T)$, and $T$ is the temperature. The integration is over the
212: frequency $\omega$ of the modes $m$ that propagate in the structure. The
213: transmission coefficient is unity for the ideal case. Any scattering reduces
214: the thermal conductance, and scattering of the lowest modes can reduce the
215: conductance below the universal value at low temperatures.
216:
217: To actually perform the scattering calculation we imbed the rough beam of
218: length $L$ in an infinite beam of the same cross section but with smooth
219: surfaces outside of the region of length $L$, Fig.\ \ref{modelstructure}. Thus
220: the mathematical calculation is the scattering of a wave incident from
221: $x=-\infty$ on a rough portion of the beam with surfaces at $y=\pm W/2\pm
222: f_{1}\left( x,z\right) $ and at $z=\pm d/2\pm f_{2}\left( x,y\right) $,
223: with the height functions $f_{1,2}$, defining the roughness, nonzero only in a
224: finite region $0<x<L$. Forward scattering is evaluated from the intensity of
225: waves as $x\rightarrow+\infty$, and backward scattering from the intensity of
226: waves as $x\rightarrow-\infty$.\begin{figure}[ptb]
227: \begin{center}
228: \includegraphics[ height=2.8037in, width=4.0612in ]{figure1.eps}
229: \end{center}
230: \caption{Top: Three dimensional elastic beam with rectangular cross-section.
231: The rough surfaces are on the top, bottom, and sides. Bottom: Side view of the
232: mathematical model of the structure actually used for the scattering
233: calculation.}%
234: \label{modelstructure}%
235: \end{figure}
236:
237: To calculate the scattering amplitude, we take a Green function approach
238: similar to our previous work on the scalar model\cite{SC00}.
239:
240: The displacement field $\mathbf{u}$ away from any sources satisfies the wave
241: equation:
242: \begin{equation}
243: \rho\partial_{t}^{2}u_{i}=\partial_{j}T_{ij} \label{wave}%
244: \end{equation}
245: where $\rho$ is the mass density, and%
246: \begin{equation}
247: T_{ij}=C_{ijkl}\partial_{k}u_{l} \label{stress strain}%
248: \end{equation}
249: is the stress tensor field with $C_{ijkl}$ the elastic modulus tensor. The
250: subscript $i$ runs over the $3$ Cartesian coordinates, we use the symbol
251: $\partial_{x}$ to denote the derivative $\partial/\partial x$ etc., and
252: repeated indices are to be summed over. The displacement field satisfies
253: stress free boundary conditions at the surfaces%
254: \begin{equation}
255: T_{ij}n_{j}|_{S}=0 \label{free-boundary}%
256: \end{equation}
257: where $S$ denotes the surface boundaries and $n_{j}$ is normal to the surface.
258: Assuming harmonic time dependence at frequency $\omega$, Eq.\ (\ref{wave})
259: becomes
260: \begin{equation}
261: \rho\omega^{2}u_{i}+C_{ijkl}\partial_{j}\partial_{k}u_{l}=0.
262: \label{helmholtz-u}%
263: \end{equation}
264:
265: We approximate the material of the system as an isotropic solid. Then the
266: elastic modulus tensor is
267: \begin{equation}
268: C_{ijkl}=\lambda\delta_{ij}\delta_{kl}+\mu\left( \delta_{ik}\delta
269: _{ji}+\delta_{il}\delta_{kj}\right) \label{elasticity-tensor}%
270: \end{equation}
271: where $\lambda$\ and $\mu$\ are Lam\'{e} constants ($\mu$ is also the shear modulus)%
272: \begin{equation}
273: \lambda = E \sigma / (1 + \sigma)(1 - 2\sigma)
274: ,\quad\mu=E/2\left( 1+\sigma\right)
275: \label{Lame}%
276: \end{equation}
277: with $E$ Young's modulus and $\sigma$\ the Poisson ratio.
278:
279: Even in a rectangular beam geometry the displacement fields in the propagating
280: modes yielded by these equations are complicated, and cannot be found
281: analytically. The modes can be grouped into four classes according to their
282: signature under the parity operations $y\rightarrow-y$ and $z\rightarrow-z$.
283: Some modes show regions of anomalous dispersion where the group velocity
284: $d\omega/dk$ is negative: these regions require a careful examination of the
285: notions of ``forward'' and ``backward'' scattering for the waves. The lowest
286: frequency mode of each class has a frequency that tends to zero at small wave
287: number. These four modes are the only ones excited at low enough temperature,
288: and are the ones contributing to the universal thermal conductance. The
289: structure of these modes at small wave numbers is simple and can be calculated
290: using familiar macroscopic arguments of elasticity theory: they are the
291: compression, torsion, and (two orthogonal) bending modes.
292:
293: We define a Green function $G_{iq}(\mathbf{x},\mathbf{x}^{\prime};t,t^{\prime
294: })$ to satisfy the wave equation with a source term $-\delta_{iq}%
295: \delta(\mathbf{x}-\mathbf{x}^{\prime})\delta(t-t^{\prime})$, and $\Gamma
296: _{ijq}$ to be the corresponding stress%
297: \begin{equation}
298: \Gamma_{ijq}\equiv C_{ijkl}\partial_{k}G_{lq}. \label{gamma}%
299: \end{equation}
300: It is convenient to introduce the frequency space version of the Green
301: function
302: \begin{equation}
303: G_{iq}(\mathbf{x};\mathbf{x}^{\prime};t,t^{\prime})=\int\,\frac{d\omega}{2\pi
304: }G_{iq}(\mathbf{x};\mathbf{x}^{\prime};\omega)e^{-i\omega\left( t-t^{\prime
305: }\right) }, \label{full-Green}%
306: \end{equation}
307: with a similar expression defining $\Gamma_{ijq}(\mathbf{x},\mathbf{x}%
308: ^{\prime};\omega)$. Inserting $G$, $\Gamma$, and the source term into
309: Eq.\ (\ref{helmholtz-u}) gives%
310: \begin{equation}
311: \rho\omega^{2}G_{iq}(\mathbf{x},\mathbf{x}^{\prime};\omega)+\partial_{j}%
312: \Gamma_{ijq}(\mathbf{x},\mathbf{x}^{\prime};\omega)=-\delta_{iq}\delta\left(
313: \mathbf{x}-\mathbf{x}^{\prime}\right) \label{helmholtz-G}%
314: \end{equation}
315: where $\mathbf{x}$ is the observation coordinate and $\mathbf{x}^{\prime}$ is
316: the source coordinate.
317:
318: Equations\ (\ref{helmholtz-u}) and (\ref{helmholtz-G}) lead to Green's theorem
319: expressing the displacement field at frequency $\omega$ in terms of a surface
320: integral%
321: \begin{equation}
322: u_{q}(\mathbf{x})=\int_{S^{\prime}}\left[ n_{j}^{\prime}T_{ij}\left(
323: \mathbf{x}^{\prime}\right) G_{iq}\left( \mathbf{x}^{\prime},\mathbf{x;\omega
324: }\right) -n_{j}^{\prime}u_{i}\left( \mathbf{x}^{\prime}\right) \Gamma
325: _{ijq}\left( \mathbf{x}^{\prime},\mathbf{x;\omega}\right) \right]
326: dS^{\prime}. \label{total field}%
327: \end{equation}
328: We are free to choose any closed integration surface $S^{\prime}$. One choice
329: is to use the physical rough surface thereby eliminating the first term in
330: Eq.\ (\ref{total field}) due to the boundary condition
331: Eq.\ (\ref{free-boundary}). However, the resulting integration over the rough
332: surface is not easy. Instead, we integrate over the smoothed surfaces at
333: $y=\pm W/2$ and $z=\pm d/2$ and impose the boundary conditions on the Green
334: function to be stress free on these smoothed surfaces%
335: \begin{equation}
336: \left. \Gamma_{ijq}n_{j}\right| _{S}=0, \label{Eq_Green-bc}%
337: \end{equation}
338: together with cross sections at $x^{\prime}\rightarrow\pm\infty$ to close the surface.
339:
340: The total field $\mathbf{u}$ can be written as the sum of incident and
341: scattered waves
342: \begin{equation}
343: \mathbf{u}=\mathbf{u}^{\mathrm{in}}+\mathbf{u}^{\mathrm{sc}}\text{.}%
344: \end{equation}
345: It can be shown (see Appendix \ref{Appendix_Separate}) that the integration
346: over the sections at $x^{\prime}\rightarrow\pm\infty$ on the right hand side
347: of Eq.\ (\ref{total field}) just gives $u_{q}^{\mathrm{in}}$. In the
348: integration over the smoothed surfaces at $y=\pm W/2$ and $z=\pm d/2$ the
349: second term in the integrand vanishes due to Eq.\ (\ref{Eq_Green-bc}). Thus we
350: find the expression for the scattered field%
351:
352: \begin{equation}
353: u_{q}^{\mathrm{sc}}(\mathbf{x})=\int_{S}\left[ n_{j}^{\prime}T_{ij}\left(
354: \mathbf{x}^{\prime}\right) G_{iq}\left( \mathbf{x}^{\prime},\mathbf{x;\omega
355: }\right) \right] dS^{\prime}, \label{scatterd-field-w/bd}%
356: \end{equation}
357: with the surface $S$ the smoothed surfaces $y=\pm W/2$ and $z=\pm d/2$. The
358: stress field $T_{ij}$ on the smoothed surface is evaluated by expanding about
359: its value on the \emph{rough} surfaces, where Eq.\ (\ref{free-boundary}) applies.
360:
361: The rest of the section goes as follows: firstly, we find an explicit
362: expression for the Green function with stress free boundary conditions; then
363: we apply the boundary perturbation method to project the stress at the rough
364: surfaces onto the smooth surfaces by expanding the stress free boundary terms
365: around the smooth surfaces using the small roughness as the expansion
366: parameter; and finally we evaluate the strength of the scattered waves to give
367: the scattering coefficient.
368:
369: \subsection{Green function}
370:
371: We evaluate $G_{iq}(\mathbf{x},\mathbf{x}^{\prime};\omega)$ as an expansion in
372: the complete orthonormal set of normal modes $\mathbf{u}^{\left( k,m\right)
373: }\left( \mathbf{x}\right) $ in the ideal geometry, which satisfy
374: Eq.\ (\ref{helmholtz-u}) and stress free boundaries at the smooth surfaces.
375: Here $k$ is the wave number in the $x$ direction, and $m$ labels the branch of
376: the dispersion curve. We define $\omega_{m}(k)$ as the frequency of the mode
377: $m$ at wave number $k$ in the ideal geometry. The modes satisfy the
378: completeness relation%
379: \begin{equation}
380: \sum_{m}\int\,\frac{dk}{2\pi}u_{i}^{(k,m)}(\mathbf{x}^{\prime})^{\ast}%
381: u_{j}^{(k,m)}(\mathbf{x})=\delta_{ij}\delta(\mathbf{x}-\mathbf{x}^{\prime}).
382: \label{completeness}%
383: \end{equation}
384: Substituting this expression on the right hand side of Eq.\ (\ref{helmholtz-G}%
385: ) leads to the expression for the Green function%
386: \begin{equation}
387: G_{iq}(\mathbf{x}^{\prime},\mathbf{x};\omega)=-\sum_{m}\frac{1}{2\pi}%
388: \int_{-\infty}^{\infty}dk\,\,\frac{\phi_{i}^{(k,m)}(y^{\prime},z^{\prime
389: })^{\ast}\phi_{q}^{(k,m)}(y,z)}{\rho\left[ \left( \omega+i\epsilon\right)
390: ^{2}-\omega_{m}^{2}(k)\right] }e^{ik\left( x-x^{\prime}\right) },
391: \label{GreenFT}%
392: \end{equation}
393: where we write%
394: \begin{equation}
395: u_{i}^{(k,m)}(\mathbf{x})=\phi_{i}^{(k,m)}(y,z)e^{ikx} \label{Eq_u-phi}%
396: \end{equation}
397: with$\ \phi_{i}^{(k,m)}$\ giving the transverse dependence\ of the
398: displacement field. In Eq.\ (\ref{GreenFT}) $\epsilon$ is a positive
399: infinitesimal number to incorporate causality, $G_{iq}(\mathbf{x}%
400: ,\mathbf{x}^{\prime};t,t^{\prime})=0$ for $t<t^{\prime}$.
401:
402: Equation\ (\ref{GreenFT}) can now be evaluated by contour integration. The
403: integrand has poles labelled by an index $n$ near values $k=k_{n}$ on the real
404: axis which are given by solutions to the dispersion relation $\omega_{m}%
405: (k_{n})=\omega$ for all branches $m$. (We take an incident wave with
406: $\omega>0$.) Note that for branches with regions of anomalous dispersion there
407: may be more than one solution to this equation for some $\omega$, so that the
408: index $n$ is not identical to the branch index $m$. The poles are shifted
409: slightly off the real axis by the infinitesimal $\varepsilon$ in
410: Eq.\ (\ref{GreenFT}), and are given by expanding about $k_{n}$%
411: \[
412: k=k_{n}+\frac{i\epsilon}{\omega_{n}v_{g}^{\left( n\right) }},
413: \]
414: with $v_{g}^{\left( n\right) }$\ the group velocity at the $n$th pole
415: $\left. d\omega_{m}/dk\right| _{k=k_{n}}$. Notice the poles are in the upper
416: half plane for $v_{g}^{(n)}>0$, and in the lower half plane for $v_{g}%
417: ^{\left( n\right) }<0$.
418:
419: Now we can perform the $k$ integration by complex integration. Consider first
420: the case, $x>x^{\prime}$. The contour must be closed in the upper half plane
421: so that the contribution from the semicircle at large $|k|$ vanishes. The
422: contour integration then picks up contributions from the poles in the upper
423: half plane, i.e. wave numbers with $v_{g}^{\left( n\right) }>0$. On the
424: other hand, for $x<x^{\prime}$ the contour must be closed in the lower half
425: plane and it is poles at wave numbers with $v_{g}^{\left( n\right) }<0$ that
426: give nonzero residue. Forward scattering or backscattering is thus seen to be
427: determined by the sign of the group velocity $v_{g}^{\left( n\right) }$
428: rather than by the sign of $k_{n}$, as indeed would be expected physically.
429:
430: Evaluating the residues gives the expression for the Green function:
431: \begin{equation}
432: G_{iq}(\mathbf{x}^{\prime},\mathbf{x};\omega)=i\left. \sum_{n}\right.
433: ^{\prime}\,\,\frac{u_{i}^{(n)}(\mathbf{x}^{\prime})^{\ast}u_{q}^{(n)}%
434: (\mathbf{x})}{2\rho\omega_{n}\,v_{g}^{\left( n\right) }},
435: \label{Green function}%
436: \end{equation}
437: where $u_{i}^{(n)}(\mathbf{x})$ is written for $u_{i}^{(k,m)}(\mathbf{x})$\ at
438: the value of the wave number $k=k_{n}$ satisfying $\omega_{m}(k_{n})=\omega$.
439: The prime on the sum is used to denote the fact that the sum runs over
440: intersections $n$ with $v_{g}^{(n)}>0$ for $x>x^{\prime}$, and over $n$ with
441: $v_{g}^{(n)}<0$ for $x<x^{\prime}$.
442:
443: The group velocity $v_{g}^{\left( n\right) }$ does not have an analytical
444: expression for a rectangular beam, and is obtained by differentiating the
445: dispersion curve which must be found numerically. Alternatively, to avoid
446: numerical differentiation, we can rewrite $v_{g}^{\left( n\right) }$ in
447: terms of the average power flow in mode $n$. Since $u_{i}^{\left( n\right)
448: }$\ is normalized, the power $P_{n}$ in mode $n$ can be written as
449: \begin{equation}
450: P_{n}=\frac{1}{2}\operatorname{Re}\int\int\left( -i\omega T_{ix}^{\left(
451: n\right) }u_{i}^{\left( n\right) \ast}\right) dydz=\frac{1}{2}\rho
452: \omega^{2}v_{g}^{\left( n\right) }, \label{power}%
453: \end{equation}
454: the first expression of the equality expressing the energy flux in terms of
455: the rate of work done across a section, and the second in terms of the group
456: velocity and the average energy density evaluated as twice the average kinetic
457: energy. Then $v_{g}^{\left( n\right) }$ can be evaluated in terms of $P_{n}$
458: as
459: \begin{equation}
460: v_{g}^{\left( n\right) }=2P_{n}/\rho\omega^{2} \label{vg-Power}%
461: \end{equation}
462: and $P_{n}$ has an expression directly in terms of displacement field given by
463: the first equality in Eq.\ (\ref{power})%
464: \begin{equation}
465: P_{n}=\frac{1}{2}\operatorname{Re}\int\int\left( -i\omega T_{ix}^{\left(
466: n\right) }u_{i}^{\left( n\right) \ast}\right) dydz. \label{Power_integral}%
467: \end{equation}
468: This expression for $v_{g}^{\left( n\right) }$ can also be derived directly
469: from the equations of motion \cite{A}.
470:
471: \subsection{Boundary perturbation}
472:
473: In this section we show the boundary perturbation technique for the rough
474: surfaces on the sides (i.e.\ the $xz$ boundary planes). We work out the
475: scattering coefficient explicitly for the surface near $y=W/2$. The surface
476: near $y=-W/2$ will give a similar contribution and, assuming uncorrelated
477: roughness on the two surfaces, is accounted for by multiplying the
478: single-surface scattering rate by $2$ at the end of the calculation. The
479: results for the top and bottom surfaces can be obtained by interchanging $y$
480: and $z$ whenever they occur in the indices in the displacement fields and
481: stress tensors in the calculation below.
482:
483: In order to calculate the stress on the smooth surface appearing in
484: Eq.\ (\ref{scatterd-field-w/bd}), we expand the stress $T_{ij}$\ in a Taylor
485: series about the flat surface, and impose stress free boundary conditions at
486: the rough surface which is the small distance $f_{1}$ away. We also assume
487: $f_{1}$\ is differentiable.
488:
489: The unit vector $\hat{n}$ normal to the rough boundaries to first order in
490: $f_{1}$ is
491: \begin{equation}
492: \hat{n}\simeq\hat{y}-\partial_{x}f_{1}\left( x,z\right) \hat{x}-\partial
493: _{z}f_{1}\left( x,z\right) \hat{z}.
494: \end{equation}
495: Then the stress free surface boundary conditions Eq.\ (\ref{free-boundary})
496: can be written%
497: \begin{equation}
498: \left[ T_{iy}-\partial_{x}f_{1}\left( x,z\right) T_{ix}-\partial_{z}%
499: f_{1}\left( x,z\right) T_{iz}\right] _{y=\frac{W}{2}+f}=0
500: \label{bondary-sides}%
501: \end{equation}
502: Now expanding Eq.\ (\ref{bondary-sides}) in the neighborhood of $y=W/2$ and
503: taking only the lowest order in $f_{1}$ and $f_{1}^{\prime}$, we obtain%
504: \begin{align}
505: \left. T_{xy}\right| _{y=\frac{W}{2}} & \simeq\left. \left( \partial
506: _{x}f_{1}\left( x,z\right) T_{xx}+\partial_{z}f_{1}\left( x,z\right)
507: T_{xz}-f_{1}\left( x,z\right) \partial_{y}T_{xy}\right) \right|
508: _{y=\frac{W}{2}}\label{xz-perturbed1}\\
509: \left. T_{zy}\right| _{y=\frac{W}{2}} & \simeq\left. \left( \partial
510: _{x}f_{1}\left( x,z\right) T_{zx}+\partial_{z}f_{1}\left( x,z\right)
511: T_{zz}-f_{1}\left( x,z\right) \partial_{y}T_{zy}\right) \right|
512: _{y=\frac{W}{2}}\label{xz-perturbed2}\\
513: \left. T_{yy}\right| _{y=\frac{W}{2}} & \simeq\left. -f_{1}\left(
514: x,z\right) \partial_{y}T_{yy}\right| _{y=\frac{W}{2}} \label{xz-perturbed3}%
515: \end{align}
516: where the first two expressions for $T_{xy}$ and $T_{zy}$ have been used to
517: simplify $T_{yy}$. Since the terms on the right hand side of
518: Eqs.\ (\ref{xz-perturbed1}-\ref{xz-perturbed3}) are explicitly first order in
519: the small parameter $f_{1}$, the stress field $T_{ij}$ on the right hand side
520: can be evaluated at zeroth order, i.e. for ideal smooth surfaces. These
521: results are used in Eq.\ (\ref{scatterd-field-w/bd}).
522:
523: \subsection{Scattering coefficient}
524:
525: We now evaluate the expression for the scattered field given by an integration
526: over the beam surfaces Eq. (\ref{scatterd-field-w/bd}). To calculate the
527: scattering coefficient, we consider an incident wave of unit amplitude in a
528: single mode $m$. Again in this section we will outline the calculation for the
529: scattering by the single surface at $y=W/2$, and will include the effects of
530: the other surfaces at the end. We therefore have%
531: \begin{equation}
532: u_{q}^{\text{sc}}\left( \mathbf{x}\right) =\int\int\left[ T_{iy}\left(
533: \mathbf{x}^{\prime}\right) G_{iq}\left( \mathbf{x}^{\prime},\mathbf{x}%
534: ;\omega\right) \right] _{y^{\prime}=\frac{W}{2}}dx^{\prime}dz^{\prime}.
535: \label{sc field}%
536: \end{equation}
537:
538: We can now evaluate the forward and backscattering amplitudes by using
539: Eq.\ (\ref{Green function}) for the Green function in Eq.\ (\ref{sc field}),
540: and evaluating the scattered wave at large positive and negative $x$%
541: \begin{equation}
542: u_{q}^{\mathrm{sc}}\left( x\rightarrow\infty,y,z\right) \simeq\int_{-\infty
543: }^{x}dx^{\prime}\int_{-\frac{d}{2}}^{\frac{d}{2}}dz^{\prime}\sum
544: _{n,v_{g}^{(n)}>0}\frac{i}{2\rho\omega\,v_{g}^{\left( n\right) }}\left[
545: T_{iy}\left( \mathbf{x}^{\prime}\right) u_{i}^{\left( n\right) }\left(
546: \mathbf{x}^{\prime}\right) ^{\ast}\right] _{y^{\prime}=\frac{W}{2}}%
547: u_{q}^{\left( n\right) }\left( \mathbf{x}\right) , \label{x-inf}%
548: \end{equation}%
549: \begin{equation}
550: u_{q}^{\mathrm{sc}}\left( x\rightarrow-\infty,y,z\right) \simeq\int
551: _{x}^{\infty}dx^{\prime}\int_{-\frac{d}{2}}^{\frac{d}{2}}dz^{\prime}%
552: \sum_{n,v_{g}^{(n)}<0}\frac{i}{2\rho\omega\,v_{g}^{\left( n\right) }}\left[
553: T_{iy}\left( \mathbf{x}^{\prime}\right) u_{i}^{\left( n\right) }\left(
554: \mathbf{x}^{\prime}\right) ^{\ast}\right] _{y^{\prime}=\frac{W}{2}}%
555: u_{q}^{\left( n\right) }\left( \mathbf{x}\right) . \label{x-minf}%
556: \end{equation}
557: The stress tensor $T_{ij}$ corresponding to the full displacement field of the
558: wave is evaluated from Eqs.\ (\ref{xz-perturbed1}-\ref{xz-perturbed3}). Since
559: these expressions explicitly include the small roughness amplitude $f_{1}$ on
560: the right hand side, to calculate the scattering at lowest order in the
561: roughness amplitude it is sufficient to replace all $T_{ij}$ on the right hand
562: side by the value $T_{ij}^{(m)}$ in the incident mode $m$. From
563: Eqs.\ (\ref{x-inf}), (\ref{x-minf}) we see that $\mathbf{u}^{\text{\textrm{sc}%
564: }}(\mathbf{x)}$ is expressed as a sum over modes $\mathbf{u}^{\left(
565: n\right) }(\mathbf{x)}$, and the coefficient of each mode is then the
566: scattering amplitude $t_{n,m}$ from incident mode $m$ into mode $n$, so that
567: \begin{equation}
568: t_{n,m}=\int_{-\infty}^{\infty}dx\int_{-\frac{d}{2}}^{\frac{d}{2}}%
569: dz\frac{i}{2\rho\omega\,v_{g}^{\left( n\right) }}\left[ T_{iy}^{(m)}\left(
570: \mathbf{x}\right) u_{i}^{\left( n\right) }\left( \mathbf{x}\right)
571: ^{\ast}\right] _{y=W/2}, \label{scat-amplitude}%
572: \end{equation}
573: where we can now extend the integration limit to $\pm\infty$ since $f_{1}$,
574: and so the integrand, is zero outside the domain of roughness $0<x<L$. Again
575: mode indices $n$ for which $v_{g}^{(n)}>0$ represent the forward-scattered
576: waves and those with $v_{g}^{(n)}<0$ the backward-scattered waves.
577:
578: Now use the expression for the stress tensor on the smooth surfaces obtained
579: in the previous section Eqs.\ (\ref{xz-perturbed1}-\ref{xz-perturbed3}) and
580: integrate the resulting expressions by parts with respect to $x$ or $z$ to
581: rewrite the terms in $\partial_{x}f_{1}$ and $\partial_{z}f_{1}$ as
582: integrations over $f_{1}$. After these manipulations we find $t_{n,m}$ can be
583: written
584: \begin{equation}
585: t_{n,m}=-\frac{i}{2\rho\omega v_{g}^{\left( n\right) }}\int_{-\infty
586: }^{\infty}dx\int_{-\frac{d}{2}}^{\frac{d}{2}}dz\,f_{1}\left( x,z\right)
587: \Gamma^{\left( m,n\right) }\left( x,z\right)
588: \end{equation}
589: where%
590: \begin{align}
591: \Gamma^{\left( m,n\right) }(x,z) & =\left[ \left( \partial_{x}%
592: T_{xx}^{\left( m\right) }+\partial_{y}T_{xy}^{\left( m\right) }%
593: +\partial_{z}T_{xz}^{\left( m\right) }\right) u_{x}^{\left( n\right)
594: \ast}+\left( \partial_{x}T_{zx}^{\left( m\right) }+\partial_{y}%
595: T_{zy}^{\left( m\right) }+\partial_{z}T_{zz}^{\left( m\right) }\right)
596: u_{z}^{\left( n\right) \ast}\right. \nonumber\\
597: & \left. +\partial_{y}T_{yy}^{\left( m\right) }u_{y}^{\left( n\right)
598: \ast}+T_{xx}^{\left( m\right) }\partial_{x}u_{x}^{\left( n\right) \ast
599: }+T_{zz}^{\left( m\right) }\partial_{z}u_{z}^{\left( n\right) \ast}%
600: +T_{zx}^{\left( m\right) }\left( \partial_{x}u_{z}^{\left( n\right) \ast
601: }+\partial_{z}u_{x}^{\left( n\right) \ast}\right) \right] _{y=W/2}.
602: \label{gamma-1}%
603: \end{align}
604: Applying the equations of motion Eq.\ (\ref{helmholtz-u}) and
605: remembering$\left. T_{iy}^{(m)}\right| _{y=W/2}=0$ for all $i$ and for all
606: $x,z$ leads to the somewhat simpler expression%
607: \begin{align}
608: \Gamma^{\left( m,n\right) }(x,z\mathbf{)} & =\left[ -\rho\omega
609: ^{2}\left( u_{x}^{(m)}u_{x}^{\left( n\right) \ast}+u_{y}^{(m)}%
610: u_{y}^{\left( n\right) \ast}+u_{z}^{(m)}u_{z}^{\left( n\right) \ast
611: }\right) \right. \nonumber\\
612: & \left. +T_{xx}^{\left( m\right) }\partial_{x}u_{x}^{\left( n\right)
613: \ast}+T_{zz}^{\left( m\right) }\partial_{z}u_{z}^{\left( n\right) \ast
614: }+T_{xz}^{\left( m\right) }\left( \partial_{z}u_{x}^{\left( n\right)
615: \ast}+\partial_{x}u_{z}^{\left( n\right) \ast}\right) \right] _{y=W/2}.
616: \end{align}
617: Notice that the scattering separates into a kinetic term (the first line)\ and
618: a stress term (the second line).
619:
620: The above form for $\Gamma^{(m,n)}$ is still neither instructive nor practical
621: for numerical evaluation. It can be further simplified using the expressions
622: Eqs.\ (\ref{stress strain}) and (\ref{Lame}) for the stress tensor in terms of
623: the displacements. First we use the boundary condition $T_{yy}^{(m)}=0$ for
624: the y-stress to give at $y=W/2$%
625: \begin{equation}
626: \partial_{y}u_{y}^{(m)}=-\frac{\sigma}{\left( 1-\sigma\right) }(\partial
627: _{x}u_{x}^{(m)}+\partial_{z}u_{z}^{\left( m\right) }). \label{bc1}%
628: \end{equation}
629: This can be used to simplify the expressions for the other components of the
630: stress tensors at $y=W/2$%
631: \begin{align}
632: T_{xx}^{(m)} & =\frac{E}{\left( 1-\sigma^{2}\right) }(\partial_{x}%
633: u_{x}^{(m)}+\sigma\partial_{z}u_{z}^{\left( m\right) }),\label{bc2}\\
634: T_{zz}^{(m)} & =\frac{E}{\left( 1-\sigma^{2}\right) }(\sigma\partial
635: _{x}u_{x}^{(m)}+\partial_{z}u_{z}^{\left( m\right) }),\label{bc3}\\
636: T_{xz}^{(m)} & =\frac{E}{2\left( 1+\sigma\right) }(\partial_{x}u_{z}%
637: ^{(m)}+\partial_{z}u_{x}^{(m)}). \label{bc4}%
638: \end{align}
639: Inverting these gives at $y=W/2$%
640: \begin{align}
641: \partial_{x}u_{x}^{\left( m\right) } & =\frac{1}{E}(T_{xx}^{\left(
642: m\right) }-\sigma T_{zz}^{\left( m\right) }),\label{bc5}\\
643: \partial_{z}u_{z}^{\left( m\right) } & =\frac{1}{E}(T_{zz}^{\left(
644: m\right) }-\sigma T_{xx}^{\left( m\right) }),\label{bc6}\\
645: \partial_{x}u_{z}^{(m)}+\partial_{z}u_{x}^{(m)} & =\frac{2(1+\sigma)}%
646: {E}T_{xz}^{\left( m\right) }. \label{bc7}%
647: \end{align}
648: We emphasize that Eqs.\ (\ref{bc1}-\ref{bc7}) are only true on the stress free
649: boundaries, and are not generally true in the bulk of the material.
650:
651: Using these results we get%
652: \begin{equation}
653: t_{n,m}=-\frac{i}{2\rho\omega\,v_{g}^{\left( n\right) }}\int_{-\frac{d}{2}%
654: }^{\frac{d}{2}}dz\,\tilde{f}_{1}\left( k_{m}-k_{n},z\right) \bar{\Gamma
655: }^{\left( m,n\right) }(z) \label{scattered-field-final}%
656: \end{equation}
657: with%
658: \begin{align}
659: \bar{\Gamma}^{\left( m,n\right) } & =\left\{ \rho\omega^{2}(\phi
660: _{x}^{(m)}\phi_{x}^{\left( n\right) \ast}+\phi_{y}^{(m)}\phi_{y}^{\left(
661: n\right) \ast}+\phi_{z}^{(m)}\phi_{z}^{\left( n\right) \ast})\right.
662: \label{numerator}\\
663: & \left. -\frac{1}{E}\left[ (\bar{T}_{xx}^{\left( m\right) }\bar{T}%
664: _{zz}^{\left( n\right) \ast}+\bar{T}_{zz}^{\left( m\right) }\bar{T}%
665: _{zz}^{\left( n\right) \ast})-(\sigma\bar{T}_{zz}^{\left( m\right) }%
666: \bar{T}_{xx}^{\left( n\right) \ast}+\bar{T}_{xx}^{\left( m\right) }\bar
667: {T}_{zz}^{\left( n\right) \ast})\right] -\frac{1}{\mu}\bar{T}_{xz}^{\left(
668: m\right) }\bar{T}_{zx}^{\left( n\right) \ast}\right\} _{y=W/2}\nonumber
669: \end{align}
670: where we have introduced the explicit $x$ dependence of $u_{i}^{(n)}%
671: (\mathbf{x})$ as in Eq.\ (\ref{Eq_u-phi}) and\ the stress tensor%
672: \begin{equation}
673: T_{ij}^{(n)}(\mathbf{x})=\bar{T}_{ij}(y,z)e^{ik_{n}x},
674: \end{equation}
675: so that the $x^{\prime}$ integration is just the Fourier transform $\tilde{f}$
676: of the roughness function, and $\bar{\Gamma}$ is a function of the $z$
677: coordinate only.
678:
679: Alternatively, using Eqs.\ (\ref{bc2}-\ref{bc4}) we can derive an expression
680: explicitly in the displacement fields, which is useful for numerical
681: evaluation,%
682: \begin{align}
683: \bar{\Gamma}^{\left( m,n\right) } & =\left\{ \rho\omega^{2}(\phi
684: _{x}^{(m)}\phi_{x}^{\left( n\right) \ast}+\phi_{y}^{(m)}\phi_{y}^{\left(
685: n\right) \ast}+\phi_{z}^{(m)}\phi_{z}^{\left( n\right) \ast})\right.
686: \nonumber\\
687: & -\frac{2\mu}{\left( 1-\sigma\right) }\left[ (k_{m}k_{n}\phi_{x}%
688: ^{(m)}u_{x}^{\left( n\right) \ast}+\partial_{z}\phi_{z}^{\left( m\right)
689: }\partial_{z}\phi_{z}^{\left( n\right) \ast})+\sigma(ik_{m}\phi_{x}%
690: ^{(m)}\partial_{z}\phi_{z}^{\left( n\right) \ast}-ik_{n}\partial_{z}\phi
691: _{z}^{\left( m\right) }\phi_{x}^{\left( n\right) \ast})\right] \nonumber\\
692: & \left. -\mu(ik_{m}\phi_{z}^{(m)}\partial_{z}\phi_{x}^{\left( n\right)
693: \ast}+k_{m}k_{n}\phi_{z}^{(m)}\phi_{z}^{\left( n\right) \ast}+\partial
694: _{z}\phi_{x}^{(m)}\partial_{z}\phi_{x}^{\left( n\right) \ast}-ik_{n}%
695: \partial_{z}\phi_{x}^{(m)}\phi_{z}^{\left( n\right) \ast})\right\}
696: _{y=W/2}. \label{gamma-U}%
697: \end{align}
698:
699: The scattering rate is given by multiplying $\left| t_{n,m}\right| ^{2}$ by
700: the ratio of the group velocities in the scattered and incident
701: waves\footnote{Note that the scattering amplitude normalized by the energy
702: flux $\bar{t}_{n,m}=\sqrt{\left| v_{g}^{\left( n\right) }\right| /\left|
703: v_{g}^{\left( m\right) }\right| }t_{n,m}$ can be seen from
704: Eq.\ (\ref{scattered-field-final}) to explicitly satisfy the reciprocity
705: relation $\bar{t}_{n,m}=\bar{t}_{-m,-n}^{\ast}$. For systems in which energy
706: is conserved, as in our case, reciprocity is equivalent to time-reversal
707: invariance\cite{P98}.}. We also treat the roughness of the surface
708: statistically, and take an ensemble average (denoted by angular brackets) to
709: give the final expression for the scattering rate $\gamma_{n,m}$ from mode $m$
710: to mode $n$ by the per unit length of single rough surface at $y=W/2$ given
711: by
712: \begin{equation}
713: \gamma_{n,m}L=\frac{v_{g}^{\left( n\right) }}{v_{g}^{\left( m\right) }%
714: }\left\langle \left| t_{n,m}\right| ^{2}\right\rangle =\frac{1}{4\rho
715: ^{2}\omega^{2}v_{g}^{\left( m\right) }v_{g}^{\left( n\right) }%
716: }\left\langle \left| \int_{-\frac{d}{2}}^{\frac{d}{2}}dz\tilde{f}_{1}\left(
717: k_{m}-k_{n},z\right) \bar{\Gamma}^{\left( m,n\right) }(z)\right|
718: ^{2}\right\rangle . \label{scattering-probabiity-nm}%
719: \end{equation}
720:
721: We are interested in the reduction of the phonon heat transport due to rough
722: surfaces. Only the backscattered waves (those with $v_{g}^{\left( n\right)
723: }<0$) reduce the amount of heat transmitted. Thus we define $\gamma_{m}$, the
724: thermal attenuation coefficient of mode $m$ per unit length, to be the sum of
725: the scattering rates from the incident mode $m$ to \emph{all} possible
726: backscattered modes, per unit length of rough surface. This can be written for
727: scattering off the single rough surface considered so far%
728: \begin{equation}
729: \gamma_{m}L=\sum\limits_{\substack{n\\v_{g}^{(n)}<0}}\gamma_{n,m}%
730: L=\sum_{\substack{n\\v_{g}^{(n)}<0}}\frac{1}{4\rho^{2}\omega^{2}v_{g}^{\left(
731: m\right) }v_{g}^{\left( n\right) }}\left\langle \left| \int_{-\frac{d}{2}%
732: }^{\frac{d}{2}}dz\tilde{f}_{1}\left( k_{m}-k_{n},z\right) \bar{\Gamma
733: }^{\left( m,n\right) }(z)\right| ^{2}\right\rangle .
734: \label{scattering-probability}%
735: \end{equation}
736:
737: To include the second rough side surface, assuming uncorrelated roughness, we
738: simply have to multiply the expression for $\gamma_{m}$ by a factor of $2$.
739: The expression for scattering off the top and bottom surfaces, if these are
740: rough too, can be derived in a similar manner, and the result may be obtained
741: by exchanging $y$ and $z$ in Eq.\ (\ref{scattering-probability}). The total
742: scattering rate is the sum of the scattering off all the surfaces.
743:
744: We have assumed that the amplitude of the surface roughness is small, allowing
745: us to use perturbation theory to derive the above expressions. In this case,
746: the attenuation coefficient gives an exponential decay of the energy flux, so
747: that the energy flux transmission coefficient is%
748: \begin{equation}
749: \mathcal{T}_{m}=\exp\left[ -\gamma_{m}L\right] .
750: \end{equation}
751:
752: \section{Thin Plate Limit}
753:
754: \label{Sec_plate}Although the expression in the previous section is general
755: and applicable to any rectangular waveguide with rough surfaces,
756: there are no closed-form expressions for the displacement fields
757: in general, and so a direct evaluation of the scattering would
758: have to be done completely numerically. Here, we instead use the
759: \emph{thin plate approximation} $d\ll W$ \cite{LF7,CL00}, which
760: yields closed form expressions for the displacement fields of the
761: modes (in terms of a dispersion curves $\omega_{m}(k)$ given by
762: numerical solution of a simple transcendental equation). The thin
763: plate limit captures the important properties of the elastic
764: modes, for example the quadratic dispersion of the bending modes
765: at small wave numbers, and regions of anomalous dispersion, as
766: well as providing analytical expressions enabling us to do further
767: analysis of the scattering. The thin plate theory is applicable
768: where the thickness of the sample is much less than the width and
769: the wavelengths are much greater than the thickness, which is the
770: case for many mesoscopic systems at low temperatures.
771:
772: The use of the thin plate limit for mesoscopic structures was
773: proposed in reference \cite{CL00}, where the calculation of the
774: structure of the modes is described in more detail. It is found
775: that the modes can be separated into two classes: \emph{in-plane
776: modes}, where the polarization of the displacement is largely in
777: the $xy$ plane (together with small strains in the $z$-direction
778: given by the Poisson effect) and the displacement field is
779: completely specified by giving the \emph{vertically averaged
780: horizontal displacement components }$\bar{u}_{x}(x,y)$ and
781: $\bar{u}_{y}(x,y)$; and \emph{flexural modes}, where the
782: displacement is primarily in the $z$ direction and is specified by
783: a vertical displacement field $\bar{u}_{z}(x,y)$. Within each
784: class we can further distinguish the modes by their parity under
785: $y\rightarrow-y$. For the in-plane modes we define the mode as
786: even if $\bar{u}_{x}\left( x,-y\right) =\bar{u}_{x}\left(
787: x,y\right) $\ and odd if $\bar{u}_{x}\left( x,-y\right)
788: =-\bar{u}_{x}\left( x,y\right) $. Similarly, the even flexural
789: modes have $\bar{u}_{z}\left( x,-y\right) =\bar{u}_{z}\left(
790: x,y\right) $ and the odd modes have $\bar{u}_{z}\left(
791: x,-y\right) =-\bar{u}_{z}\left( x,y\right) $. As in the general
792: case, there are four branches of the dispersion curves that tend
793: to zero frequency as the wave number goes to zero, corresponding
794: to one mode from each of these classes. The low frequency, even in
795: plane mode corresponds to the compression mode, and the odd mode
796: to a bending mode. The low frequency even, flexural mode
797: corresponds to the second bending mode, and the low frequency odd
798: flexural mode is the torsion mode.
799:
800: Explicit expressions for the displacement fields can be obtained using the
801: method described in reference \cite{CL00}. For the in-plane modes we find, up
802: to a normalization factor $A_{1}$ that is common to both even and odd parity
803: waves, the even modes%
804:
805: \begin{equation}
806: \bar{u}_{x}\left( x,y\right) =ikA_{1}\left[ \frac{k^{2}-\chi_{1}^{2}%
807: }{2k^{2}}\cos\left( \frac{\chi_{2}W}{2}\right) \cos\left( \chi_{1}y\right)
808: -\cos\left( \chi_{2}y\right) \cos\left( \frac{\chi_{1}W}{2}\right)
809: \right] e^{ikx} \label{even-ux-filed}%
810: \end{equation}%
811: \begin{equation}
812: \bar{u}_{y}\left( x,y\right) =A_{1}\left[ \frac{k^{2}-\chi_{1}^{2}}%
813: {2\chi_{1}}\cos\left( \frac{\chi_{2}W}{2}\right) \sin\left( \chi
814: _{1}y\right) +\chi_{2}\cos\left( \frac{\chi_{1}W}{2}\right) \sin\left(
815: \chi_{2}y\right) \right] e^{ikx}, \label{even-uy-field}%
816: \end{equation}
817: and the odd modes%
818: \begin{equation}
819: \bar{u}_{x}\left( x,y\right) =ikA_{1}\left[ \frac{k^{2}-\chi_{1}^{2}%
820: }{2k^{2}}\sin\left( \chi_{1}y\right) \sin\left( \frac{\chi_{2}W}{2}\right)
821: -\sin\left( \frac{\chi_{1}W}{2}\right) \sin\left( \chi_{2}y\right)
822: \right] e^{ikx} \label{odd-ux-field}%
823: \end{equation}%
824: \begin{equation}
825: \bar{u}_{y}\left( x,y\right) =-A_{1}\left[ \frac{k^{2}-\chi_{1}^{2}}%
826: {2\chi_{1}}\cos\left( \chi_{1}y\right) \sin\left( \frac{\chi_{2}W}%
827: {2}\right) +\chi_{2}\sin\left( \frac{\chi_{1}W}{2}\right) \cos\left(
828: \chi_{2}y\right) \right] e^{ikx}, \label{odd-uy-field}%
829: \end{equation}
830: where $\chi_{1}=\left( \omega^{2}/c_{t}^{2}-k^{2}\right) ^{1/2}$ and
831: $\chi_{2}=\left( \omega^{2}/c_{l}^{2}-k^{2}\right) ^{1/2}$,\ with $c_{t}%
832: $\ the transverse sound velocity and $c_{l}$\ the longitudinal
833: velocity \emph{in a large thin plate}
834: \begin{equation}
835: c_{t} = \sqrt{\frac{E}{2\rho \left( 1+\sigma \right) }},\quad
836: c_{l} = \sqrt{\frac{E}{\rho \left( 1-\sigma ^{2}\right) }},
837: \end{equation}
838: and $\omega$ and $k$ related by the dispersion curve which must be
839: found numerically. In the thin plate limit it is sufficient to
840: take for the in-plane modes
841: \begin{align}
842: u_{x}(x,y,z) & \simeq\bar{u}_{x}(x,y),\\
843: u_{y}(x,y,z) & \simeq\bar{u}_{y}(x,y),\\
844: u_{z}(x,y,z) & \simeq0.
845: \end{align}
846:
847: Similarly, the vertical displacement field for the even flexural modes is%
848: \begin{equation}
849: \bar{u}_{z}\left( x,y\right) =A_{2}\left[ \cosh\left( \frac{\chi_{-}W}%
850: {2}\right) \cosh\left( \chi_{+}y\right) -\frac{k^{2}\sigma-\chi_{+}^{2}%
851: }{k^{2}\sigma-\chi_{-}^{2}}\cosh\left( \frac{\chi_{+}W}{2}\right)
852: \cosh\left( \chi_{-}y\right) \right] e^{ikx}, \label{even_flex}%
853: \end{equation}
854: and for the odd flexural modes%
855: \begin{equation}
856: \bar{u}_{z}\left( x,y\right) =A_{2}\left[ \sinh\left( \frac{\chi_{-}W}%
857: {2}\right) \sinh\left( \chi_{+}y\right) -\frac{k^{2}\sigma-\chi_{+}^{2}%
858: }{k^{2}\sigma-\chi_{-}^{2}}\sinh\left( \frac{\chi_{+}W}{2}\right)
859: \sinh\left( \chi_{-}y\right) \right] e^{ikx}, \label{odd_flex}%
860: \end{equation}
861: where $\chi_{+}=\left( k^{2}+\sqrt{\rho d/D}\omega\right) ^{2}$ and
862: $\chi_{-}=\left( k^{2}-\sqrt{\rho d/D}\omega\right) ^{2}$, with
863: $D=Ed^{3}/12\left( 1-\sigma^{2}\right) $ the flexural rigidify, and again
864: $\omega$ and $k$ are related by the appropriate dispersion curve. In the
865: classical thin plate theory, the displacement fields are given in terms of
866: $\bar{u}_{z}$ by the expressions%
867: \begin{align}
868: u_{x}(x,y,z) & \simeq-z\partial_{x}\bar{u}_{z}(x,y),\\
869: u_{y}(x,y,z) & \simeq-z\partial_{y}\bar{u}_{z}(x,y),\\
870: u_{z}(x,y,z) & \simeq\bar{u}_{z}(x,y).
871: \end{align}
872: This approximation is adequate for evaluating the surface stress integrals in
873: Eq.\ (\ref{scattering-probability}) but turns out not to be sufficiently
874: accurate to evaluate the energy flux expression for the group velocity
875: Eq.\ (\ref{Power_integral}). We discuss this case in section
876: \ref{Subsec_Power_Flexural} below.
877:
878: \subsection{Ideal thermal conductance}%
879:
880: %TCIMACRO{\FRAME{ftbpFU}{3.8977in}{3.1522in}{0pt}{\Qcb{Mode frequency
881: %$\omega_{N}$ as a function of mode number $N$: crosses - thin plate theory;
882: %circles - xyz algorithm; solid line - bulk mode density of states calculation.
883: %A thickness to width ratio $d/W=0.38$ was used.}}{\Qlb{figure2}}%
884: %{figure2.eps}{\special{ language "Scientific Word"; type "GRAPHIC";
885: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
886: %width 3.8977in; height 3.1522in; depth 0pt; original-width 4.875in;
887: %original-height 3.9375in; cropleft "0"; croptop "1"; cropright "1";
888: %cropbottom "0"; filename 'figure2.eps';file-properties "XNPEU";}}}%
889: %BeginExpansion
890: \begin{figure}
891: [ptb]
892: \begin{center}
893: \includegraphics[
894: height=3.1522in,
895: width=3.8977in
896: ]%
897: {figure2.eps}%
898: \caption{Mode frequency $\omega_{N}$ as a function of mode number $N$: crosses
899: - thin plate theory; circles - xyz algorithm; solid line - bulk mode density
900: of states calculation. A thickness to width ratio $d/W=0.38$ was used.}%
901: \label{cut-off}%
902: \end{center}
903: \end{figure}
904: %EndExpansion
905: Since our quantitative calculation of the scattering\ coefficient relies on
906: the analytic expressions for the elastic modes available only in the thin
907: plate limit, it is essential to estimate the temperature range where the thin
908: plate limit is applicable for a given experimental structure. On the other
909: hand, as the wavelength becomes much smaller than the dimensions of the
910: structure, we should to be able to treat the waves in terms of separate
911: longitudinal and transverse waves in the bulk of the material, without
912: worrying too much about the complicated standing wave transverse mode
913: structure important for the long wavelength modes. In this regime, which we
914: refer to as the bulk mode limit, the counting of the modes is insensitive to
915: the details of the boundary conditions, and is the same as for a scalar wave
916: approximation. The ideal thermal conductance depends only on cut-off frequency
917: of the modes (see Eq.\ (\ref{eq:conductance})), and we can assess the
918: applicability of these simple limiting approximations by comparing the mode
919: cutoff frequencies with results from a numerical calculation of the full
920: elasticity theory. For the full elastic theory, we use the ``xyz
921: algorithm''\cite{NAW97}.
922:
923: For the bulk mode calculation, there are three polarizations (one
924: longitudinal and two transverse) with propagation velocities
925: $c_{3l}$ and $c_{t}$, respectively with $c_{t}$ as before and
926: \begin{equation}
927: c_{3l} = \sqrt{\frac{E\left( 1-\sigma \right) }{\rho \left(
928: 1+\sigma \right) \left( 1-2\sigma \right) }}.
929: \end{equation}
930: The precise details of the boundary conditions are unimportant in
931: the mode counting for large mode numbers. If we assume standing
932: waves in the transverse direction corresponding to zero normal
933: derivative boundary
934: conditions on the wave functions, the cutoff frequencies are%
935: \begin{equation}
936: \omega_{t,mn}=c_{t}\sqrt{\left( \frac{m\pi}{W}\right) ^{2}+\left(
937: \frac{n\pi}{d}\right) ^{2}},\text{\quad(two fold degenerate)}%
938: \end{equation}
939: for the transverse waves, and%
940: \begin{equation}
941: \omega_{l,mn}=c_{3l}\sqrt{\left( \frac{m\pi}{W}\right)
942: ^{2}+\left(
943: \frac{n\pi}{d}\right) ^{2}},\text{\quad(non degenerate)}%
944: \end{equation}
945: for the longitudinal waves, with $m,n=0,1,2\ldots$. For large $m,n$ we can use
946: the continuous form for the frequency $\omega_{N}$ of the $N$th mode%
947: \begin{equation}
948: N=\frac{dW}{4\pi}\omega_{N}^{2}\left( \frac{2}{c_{t}^{2}}+\frac{1}{c_{3l}^{2}%
949: }\right) . \label{bulk_mode}%
950: \end{equation}
951:
952: Figure \ref{cut-off} shows the cutoff frequencies as a function of mode number
953: for the thickness to width ratio $d/W=0.38$. The thin plate theory gives a
954: good approximation at lower frequencies. The accuracy of thin plate theory
955: becomes better as $d/W$ gets smaller. For example, in the case of $d/W=0.1$
956: (not shown), the error in the cutoff frequencies of the first $13$ modes is
957: less than $3\%$, whilst the error is as large as $5\%$ for the first $7$ modes
958: for the case $d/W=0.38$ shown in the figure. In terms of the ideal
959: (no-scattering) thermal conductance (Eq.\ (\ref{eq:conductance}) with the
960: transmission coefficient set to unity), we find that for $d/W=0.38$ the error
961: in the thermal conductance is less than $4\%$ up to $T\sim0.4$ K. Thus the
962: thin plate limit is adequate to examine the scattering effects in this
963: temperature range. At large frequencies $\omega W/c_{t}>30$ the elasticity
964: theory results approach closely the continuum bulk mode calculations,
965: Eq.\ (\ref{bulk_mode}). The thin plate approximation clearly fails in this
966: limit, since it predicts $N\propto\omega$ corresponding to a $2d$ structure.
967:
968: \subsection{Attenuation coefficient in the thin plate limit}
969:
970: The thin plate approximation is implemented by noticing that the
971: stress free boundary conditions imply that the stress components
972: $T_{iz}$ are zero on the top and bottom surfaces. For small
973: thickness this implies that the components $T_{iz}$ for any $i$
974: are small everywhere. In most situations these components can be
975: approximated as zero\cite{LF7}. This simplifies many of the terms
976: appearing in Eq.\ (\ref{scattered-field-final}). Also, at low
977: temperatures, only modes with no strong dependence on the $z$
978: coordinate will be excited, so that the mode sum extends over
979: modes with increasing numbers of nodes in the $y$ direction only.
980:
981: In this section we calculate the scattering of the elastic waves
982: by surface roughness for a thin plate. We assume that the
983: roughness is confined to the sides, since in the experiments
984: theses are prepared lithographically, whereas the top and bottom
985: surfaces are produced by the epitaxial growth process.
986:
987: For simplicity, we assume the roughness function $f_{1}$ has no
988: $z$ dependence. This is probably a reasonable description of the
989: roughness produced by a typical lithographic process of
990: anisotropic chemical etch\footnote{Without this assumption, we
991: would find slightly different $z$ averages of the roughness
992: function $\tilde{f}_{1}$ involved for the scattering of the
993: in-plane modes and of the flexural modes---in fact a direct
994: average for the inplane modes and an average weighted by $z^{2}$
995: for the flexural modes. In addition there would now be scattering
996: from in-plane to flexural modes, and \emph{vice versa}.}. Then the
997: Fourier transformed roughness function $\tilde{f}_{1}\left(
998: k_{m}-k_{n}\right) $\ can be pulled outside of the $z$ integral
999: in Eq.\ (\ref{scattering-probability}) and the statistical average
1000: over the roughness can be performed to give%
1001: \begin{equation}
1002: \left\langle \left| \tilde{f}_{1}\left( k\right) \right| ^{2}\right\rangle
1003: =\tilde{g}\left( k\right) L,
1004: \end{equation}
1005: where $\tilde{g}\left( k\right) $ is the Fourier transform of the roughness
1006: correlation function%
1007: \[
1008: \tilde{g}\left( k\right) =\int dxe^{-ikx}\left\langle f_{1}\left( x\right)
1009: f_{1}\left( 0\right) \right\rangle .
1010: \]
1011:
1012: Equation (\ref{scattering-probability}) leads to the back-scattering rate from
1013: mode $m$ to mode $n$
1014: \begin{equation}
1015: \gamma_{n,m}=\frac{\tilde{g}\left( k_{m}-k_{n}\right) }{2\rho^{2}\omega
1016: ^{2}v_{g}^{\left( m\right) }v_{g}^{\left( n\right) }}\left|
1017: \int_{-\frac{d}{2}}^{\frac{d}{2}}dz\bar{\Gamma}^{\left( m,n\right)
1018: }(z)\right| ^{2}. \label{attenuation-coefficient}%
1019: \end{equation}
1020: where Eq.\ (\ref{scattering-probability}) is multiplied by a factor of $2$ to
1021: account for the two surfaces at $y=\pm W/2$.
1022:
1023: With the closed forms of the displacement fields at hand, we can obtain the
1024: analytical expression for the attenuation coefficient. We first evaluate
1025: $\bar{\Gamma}^{(m,n)}$ from Eq.\ (\ref{numerator}). Since $T_{iz}^{(m)}%
1026: \simeq0$, the expression for $\bar{\Gamma}$ reduces to%
1027: \begin{equation}
1028: \bar{\Gamma}^{\left( m,n\right) }\simeq\left[
1029: \rho\omega^{2}\left( \phi_{x}^{(m)}\phi_{x}^{\left( n\right)
1030: \ast}+\phi_{y}^{(m)}\phi _{y}^{\left( n\right)
1031: \ast}+\phi_{z}^{(m)}\phi_{z}^{\left( n\right) \ast
1032: }\right) -\frac{1}{E}\left( \bar{T}_{xx}^{\left( m\right) }\bar{T}%
1033: _{xx}^{\left( n\right) \ast}\right) \right] _{y=W/2}. \label{thin-A}%
1034: \end{equation}
1035: In addition, putting $T_{zz}^{(m)}$ in Eq.\ (\ref{bc4}) at the stress free
1036: boundary to zero gives
1037: \begin{equation}
1038: \partial_{z}u_{z}^{\left( m\right) }=-\sigma\partial_{x}u_{x}^{(m)}%
1039: \end{equation}
1040: so that $T_{xx}^{(m)}$ from Eq.\ (\ref{bc2}) simplifies to%
1041: \begin{equation}
1042: T_{xx}^{(m)}=E\partial_{x}u_{x}^{(m)}.
1043: \end{equation}
1044: Now Eq.\ (\ref{attenuation-coefficient}) can be written as%
1045: \begin{equation}
1046: \gamma_{n,m}=\frac{\tilde{g}\left( k_{m}-k_{n}\right) }{2\rho^{2}\omega
1047: ^{2}v_{g}^{\left( m\right) }v_{g}^{\left( n\right) }}\left|
1048: \int_{-\frac{d}{2}}^{\frac{d}{2}}dz\left[ \rho\omega^{2}\phi_{i}^{(m)}%
1049: \phi_{i}^{\left( n\right) \ast}+Ek_{s}k_{n}\phi_{x}^{(m)}\phi_{x}^{(n)\ast
1050: }\right] _{y=\frac{W}{2}}\right| ^{2}, \label{gamma-thin}%
1051: \end{equation}
1052: where the index $i$ is summed over $x,y,z$. The scattering in the thin plate
1053: limit is seen to have two components: the kinetic term, the first term in the
1054: $[]$ in Eq.\ (\ref{gamma-thin}), which involves all components of the
1055: displacement; and the stress term, the second term, which just depends on the
1056: longitudinal displacement.
1057:
1058: To see how the scattering rate scales with the parameters it is useful to
1059: rewrite Eq.\ (\ref{gamma-thin}) as%
1060: \begin{multline}
1061: \gamma_{n,m}L=\frac{\tilde{g}\left( k_{m}-k_{n}\right) L}{2W^{4}}%
1062: \times\frac{W^{2}\omega^{2}}{v_{g}^{\left( m\right) }v_{g}^{\left(
1063: n\right) }}\label{gamma_thin_nonorm}\\
1064: \times\frac{\left| \int_{-d/2}^{d/2}dz\left[ \phi_{i}^{(m)}\phi_{i}^{\left(
1065: n\right) \ast}+\frac{Ek_{s}k_{n}}{\rho\omega^{2}}\phi_{x}^{(m)}\phi
1066: _{x}^{(n)\ast}\right] _{y=\frac{W}{2}}\right| ^{2}}{\left( \int
1067: _{-d/2}^{d/2}dz\int_{-W/2}^{W/2}\frac{dy}{W}\phi_{i}^{(m)}\phi_{i}^{(m)\ast
1068: }\right) ^{1/2}\left( \int_{-d/2}^{d/2}dz\int_{-W/2}^{W/2}\frac{dy}{W}%
1069: \phi_{i}^{(n)}\phi_{i}^{(n)\ast}\right) ^{1/2}}.
1070: \end{multline}
1071: The first factor is a dimensionless measure of the strength of the roughness;
1072: the second factor is a dimensionless ratio that depends, through the
1073: dispersion relation, only on the geometric ratio $d/W$ and the Poisson ratio
1074: $\sigma$; and the final factor involves integrals over the displacement
1075: fields, where we have introduced the explicit normalization factors in the
1076: denominator so that we may evaluate the ratio using convenient unnormalized
1077: expressions for the displacements.
1078:
1079: \subsection{Evaluating the group velocity}
1080:
1081: As we have seen in Eq.\ (\ref{vg-Power}), we can avoid evaluating the group
1082: velocity appearing in Eq.\ (\ref{gamma-thin}) via numerical differentiating
1083: the dispersion curve by instead relating the group velocity to the energy flux
1084: in the mode, which in turn can be written as an explicit integral
1085: Eq.\ (\ref{Power_integral}). Thus we need to evaluate the expression (we
1086: suppress the mode index in this section )%
1087: \begin{equation}
1088: P=-\frac{1}{2}\operatorname{Re}\left[ i\omega\int\int\left( T_{xx}%
1089: u_{x}^{\ast}+T_{yx}u_{y}^{\ast}+T_{zx}u_{z}^{\ast}\right) dydz\right] .
1090: \label{Power_integral_ex}%
1091: \end{equation}
1092: involving the displacement fields and their derivatives.
1093:
1094: In the thin plate limit the $z$ components of the stress are small. If we
1095: approximate $T_{zz}=0$ then expressions Eqs.\ (\ref{stress strain}) and
1096: (\ref{Lame}) can be used to evaluate the $z$-component of the strain%
1097: \begin{equation}
1098: \partial_{z}u_{z}=-\frac{\sigma}{(1-\sigma)}\left( \partial_{x}u_{x}%
1099: +\partial_{y}u_{y}\right) .
1100: \end{equation}
1101: This can be then used to simplify the in-plane components of the stress%
1102: \begin{align}
1103: T_{xx} & =\frac{E}{(1-\sigma^{2})}(\partial_{x}u_{x}+\sigma\partial_{y}%
1104: u_{y}),\label{Txx}\\
1105: T_{yy} & =\frac{E}{(1-\sigma^{2})}(\sigma\partial_{x}u_{x}+\partial_{y}%
1106: u_{y}),\label{Tyy}\\
1107: T_{yx} & =\frac{E}{2(1+\sigma)}(\partial_{x}u_{y}+\partial_{y}u_{x}).
1108: \label{Txy}%
1109: \end{align}
1110: These expression are used to evaluate the first two terms in the integrand in
1111: Eq.\ (\ref{Power_integral_ex}). The evaluation of the last term in the
1112: integrand turns out to depend on whether we are looking at the in-plane or
1113: flexural modes, and we now consider each case in turn.
1114:
1115: \subsubsection{In-plane modes}
1116:
1117: For the in-plane modes in the thin plate limit it is sufficiently
1118: accurate to approximate $T_{zx}\simeq0$, and we can evaluate the
1119: remaining
1120: terms in $P$ with the approximations $u_{x}\simeq\bar{u}_{x}$, $u_{y}%
1121: \simeq\bar{u}_{y}$ independent of $z$. This yields
1122: \begin{equation}
1123: P=\operatorname{Re}\left\{ -\frac{i\omega Ed}{4\left( 1-\sigma^{2}\right)
1124: }\int dy\left[ 2(\partial_{x}\bar{u}_{x}+\sigma\partial_{y}\bar{u}_{y}%
1125: )\bar{u}_{x}^{\ast}+(1-\sigma)\left( \partial_{x}\bar{u}_{y}+\partial
1126: _{y}\bar{u}_{x}\right) \bar{u}_{y}^{\ast}\right] \right\} .
1127: \label{Poynting-inp}%
1128: \end{equation}
1129:
1130: \subsubsection{Flexural modes}
1131:
1132: \label{Subsec_Power_Flexural}For the flexural mode the approximations
1133: $T_{zx}\simeq0$ and $u_{z}(x,y,z)\simeq\bar{u}_{z}(x,y)$ independent of $z$
1134: lead to the expressions for the horizontal displacements%
1135: \begin{align}
1136: u_{x}(x,y,z) & \simeq-z\bar{u}_{z}(x,y),\label{Flex_ux}\\
1137: u_{y}(x,y,z) & \simeq-z\bar{u}_{z}(x,y). \label{Flex_uy}%
1138: \end{align}
1139: Using these expressions with Eqs.\ (\ref{Txx}-\ref{Tyy}) shows that the first
1140: two terms in Eq.\ (\ref{Power_integral_ex}) are of order $d^{3}$, i.e.
1141: \emph{third order} in the expansion parameter of thin plate theory $d/W$. It
1142: turns out that to this order, we \emph{cannot} neglect the last term in
1143: $T_{zx}$, even though all $z$-components in the stress tensor are nominally
1144: ``small''. Indeed comparing the group velocity evaluated from
1145: Eq.\ (\ref{Power_integral_ex}) neglecting the term in $T_{zx}$ with those
1146: given by numerically differentiating the dispersion curve shows a clear
1147: discrepancy. This same problem comes up in deriving the wave equation for the
1148: flexural waves%
1149: \begin{equation}
1150: \rho d\omega^{2}\bar{u}_{z}=D\nabla_{\perp}^{4}\bar{u}_{z}. \label{flex_wave}%
1151: \end{equation}
1152: The term on the left hand side is the mass per unit area times the vertical
1153: acceleration, which is given by the integral over the depth of $\partial
1154: _{x}T_{zx}+\partial_{y}T_{zy}$. Clearly the components of $T_{zi}$ cannot be
1155: neglected completely. Their ``smallness'' is what leads to the unusual fourth
1156: order derivative appearing in this wave equation, with a coefficient again
1157: proportional to $d^{3}$.
1158:
1159: We have used two methods to arrive at the correct calculation of the energy
1160: flux integral for the flexural waves, which is then used to calculate the
1161: group velocity for the these waves. The first is to use an improved
1162: approximation to the expressions for the in-plane displacements Eqs.
1163: (\ref{Txx},\ref{Tyy}) and a nonzero $T_{zx}$ following the approach of
1164: Timoshenko \cite{T}. The second evaluates the energy flux in terms of the
1165: vertical displacement and an effective vertical force, and in addition the
1166: rotational displacement $\theta$ and corresponding torque $M$, as is used in
1167: the macroscopic derivation\cite{LF7} of the wave equation (\ref{flex_wave}).
1168: Either of these methods leads to the expression for the energy flux
1169: \begin{multline}
1170: P\simeq\frac{1}{2}\omega D\operatorname{Re}\left\{ \int dy\left[
1171: 2k^{3}\bar{u}_{z}\bar{u}_{z}^{\ast}+k(1-\sigma)(\partial_{y}\bar{u}_{z}%
1172: )\partial_{y}\bar{u}_{z}^{\ast}-k(1+\sigma)(\partial_{y}^{2}\bar{u}_{z}%
1173: )\bar{u}_{z}^{\ast}\right] \right. \label{Poynting-power}\\
1174: \left. +Dk\left[ (1-\sigma)(\partial_{y}\bar{u}_{z})\bar{u}_{z}^{\ast
1175: }\right] _{y=\frac{W}{2}}-Dk\left[ (1-\sigma)(\partial_{y}\bar{u}_{z}%
1176: )\bar{u}_{z}^{\ast}\right] _{y=-\frac{W}{2}}\right\}
1177: \end{multline}
1178: The derivations are displayed in Appendix \ref{App_TPP}. The comparison of the
1179: group velocity derived from Eq.\ (\ref{Poynting-power}) and from numerically
1180: differentiating the dispersion curve now shows agreement to high accuracy.
1181:
1182: \section{Scattering Analysis}
1183:
1184: \label{Sec_scattering}The thermal attenuation is calculated from
1185: Eq.\ (\ref{gamma-thin}) for normalized mode displacement fields or
1186: (\ref{gamma_thin_nonorm}) in general. The group velocity for each mode can be
1187: accurately evaluated numerically from the equality $v_{g}=2P/\rho w^{2}$, with
1188: the energy flux $P$ given by Eq.\ (\ref{Poynting-inp}) for the in plane modes
1189: and Eq.\ (\ref{Poynting-power}) for the flexural modes (both expressions are
1190: for normalized displacement fields). These are all explicit results in terms
1191: of the mode displacements, which are given by Eqs.\ (\ref{even-ux-filed}%
1192: -\ref{odd-uy-field}) for the in-plane modes, and Eqs.\ (\ref{even_flex}%
1193: ,\ref{odd_flex}) for the flexural modes.
1194:
1195: %Combining the interaction term, the power term, and the
1196: %attenuation term, we have%
1197: %\begin{equation}
1198: %\gamma_{m}=\sum_{n}\frac{\omega^{2}\tilde{g}(k_{m}-k_{n})}{8P_{n}P_{m}}\left|
1199: %\int_{-\frac{d}{2}}^{\frac{d}{2}}dz^{\prime}\left. \Gamma^{\left(
1200: %m,n\right) }\right| _{y=\frac{W}{2}}\right| ^{2}. \label{gamma-main}%
1201: %\end{equation}
1202: %where%
1203: %\begin{align}
1204: %\left. \Gamma^{\left( m,n\right) }\right| _{y=\frac{W}{2}} & =\left[
1205: %\left( \rho\omega^{2}-Ek_{s}k_{n}\right) u_{x}^{(s)}u_{x}^{(n)\ast}%
1206: %+\rho\omega^{2}u_{y}^{(s)}u_{y}^{\left( n\right) \ast}+\rho\omega^{2}%
1207: %u_{z}^{(s)}u_{z}^{\left( n\right) \ast}\right] _{y=\frac{W}{2}%
1208: %}\label{thin-gamma}\\
1209: %P_{n}^{\text{in-plane}} & =-i\omega\left[ \int\int dydz\frac{E}{\left(
1210: %1-\sigma^{2}\right) }\left[ \partial_{x}u_{x}^{\left( n\right) }%
1211: %+\sigma\partial_{y}u_{y}^{\left( n\right) }\right] u_{x}^{\ast}\right.
1212: %\label{thin-inp}\\
1213: %& +\left. \int\int dydz\frac{E}{2\left( 1+\sigma\right) }\left(
1214: %\partial_{x}u_{y}^{\left( n\right) }u_{y}^{\left( n\right) \ast}%
1215: %+\partial_{y}u_{x}^{\left( n\right) }u_{y}^{\left( n\right) \ast}\right)
1216: %\right] \nonumber\\
1217: %P_{n}^{\text{flexural}} & =\frac{\omega}{2}\left[ D\int dy\left[
1218: %2k^{3}ww^{\ast}+k\left( 1-\sigma\right) \left( \partial_{y}w\right)
1219: %\left( \partial_{y}w\right) ^{\ast}-k\left( 1+\sigma\right) \partial
1220: %_{y}^{2}ww^{\ast}\right] \right. \label{thin-flex}\\
1221: %& +\left. Dk\left[ \left( 1-\sigma\right) \left( \partial_{y}w\right)
1222: %w^{\ast}\right] _{y=\frac{W}{2}}\right] \nonumber
1223: %\end{align}
1224: %Now the above expressions can be evaluated easily once the displacement field
1225: %is known, which is followed next.
1226:
1227: Before analyzing the scattering behavior, we first need to have a good
1228: understanding of the dispersion relation of the modes, since the scattering
1229: rates are strongly dependent on this.
1230:
1231: \subsection{Dispersion relation and group velocity}
1232:
1233: \begin{figure}[ptb]
1234: \begin{center}
1235: \includegraphics[ height=2.7198in, width=3.704in ]{figure3.eps}
1236: \end{center}
1237: \caption{Dispersion relation for in-plane modes (solid) and flexural modes
1238: (dashed) for a geometry ratio $d/W=0.375$ and Poisson ration $0.24$. The wave
1239: numbers are scaled with the width $W$, and the frequencies by $W/c_{t}$ with
1240: $c_{t}=\sqrt{\mu/\rho\text{.}}$}%
1241: \label{dispersion}%
1242: \end{figure}
1243:
1244: \begin{figure}[tbh]
1245: \begin{center}
1246: \includegraphics[ height=2.7717in, width=3.6832in ]{figure4.eps}
1247: \end{center}
1248: \caption{Group velocity for in-plane modes for the same parameters as
1249: Fig.\ (\ref{dispersion}): dash dotted - in-plane bending mode; solid -
1250: compression mode. The wave numbers are scaled with the width $W$, and the
1251: group velocities by $c_{t}$ with $c_{t}=\sqrt{\mu/\rho}$.}%
1252: \label{inpvg}%
1253: \end{figure}The dispersion relations for a representative case are shown in
1254: Fig.\ (\ref{dispersion}). For this example we have used a Poisson
1255: ratio of $0.24$, and a depth to width ratio of $d/W=0.375$, values
1256: corresponding to the experimental work of Schwab et al.
1257: \cite{SHWR00}. As we have discussed, the modes fall into four
1258: classes, depending on their parity signatures. We label the lowest
1259: mode from each class, the one with zero frequency as the wave
1260: number goes to zero, as mode $0$, and the modes with successively
1261: higher cutoff frequencies in each class as mode $1$, mode $2$,
1262: etc., in that class.
1263:
1264: Notice that one of the curves in the figure, the one for the
1265: in-plane mode with cutoff frequency $\omega W/c_{t}\simeq5$, shows
1266: anomalous dispersion with the frequency \emph{decreasing} as the
1267: wave number increases up to about $3W^{-1}$. (This is actually an
1268: even mode, and some higher even and odd modes also show anomalous
1269: dispersion.) The dispersion curves for all modes $n>0$ have zero
1270: slope, and so zero group velocity, at onset. As we will see later,
1271: this results in a diverging scattering rate at each mode onset.
1272: For the $n=0$ modes, as $\omega\rightarrow0$ two of the modes (the
1273: compression and torsion modes) have linear dispersion, whilst the
1274: other two lowest modes (in-plane and flexural bending modes)
1275: exhibit quadratic dispersion. Figure (\ref{inpvg}) shows the group
1276: velocities $v_{g}$ for the four lowest in-plane modes. The group
1277: velocity of the bending mode approaches zero as
1278: $\omega\rightarrow0$ whilst that of the compression mode becomes
1279: constant. The group velocity of the compression mode suddenly
1280: drops to $\sim0.5c_{t}$ around $\omega W/c_{t}\sim4.6,$ then
1281: gradually recovers and approaches $0.9c_{t}$. These features of
1282: the dispersion curve will be reflected in the behavior of the
1283: scattering of the waves.
1284:
1285: \subsection{Scattering behavior}
1286:
1287: We first consider the scattering and reduction of the thermal transport by
1288: white noise roughness $\tilde{g}(k)=\tilde{g}(0)$. This allows us to focus on
1289: the role of geometry and the unusual mode structure of the elastic waves in
1290: the physics.
1291:
1292: \squeezetable
1293: \begin{table}[ptb]
1294: %\centering%
1295: \begin{ruledtabular}
1296: \begin{tabular}
1297: [c]{cccccc} & $\omega/\sqrt{E/\rho}$ & $v_{g}/\sqrt{E/\rho}$ &
1298: $\phi_{x}$ & $\phi_{y}$ & $\phi_{z}$\\\hline
1299: Extension & $k$ & $1$ & $1$ & $O(ky)$ & $O(kz)$\\
1300: In-plane bend & $(w/\sqrt{12})k^{2}$ & $(w/\sqrt{3})k$ & $-iky$ &
1301: $1$ &
1302: $O(kz)$\\
1303: Torsion & $\sqrt{2/(1+\sigma)}(d/w)k$ & $\sqrt{2/(1+\sigma)}(d/w)$
1304: & $O(kyz)$
1305: & $-z$ & $y$\\
1306: Flex-bend & $(d/\sqrt{12})k^{2}$ & $(d/\sqrt{3})k$ & $-ikz$ &
1307: $O(k^{2}yz)$ &
1308: $1$%
1309: \end{tabular}
1310: \end{ruledtabular}
1311: \caption{Dispersion relation, group velocity, and (unnormalized) transverse mode
1312: structure for the four modes with zero frequency at zero wave vector.}%
1313: \label{Table_Modes}%
1314: \end{table}
1315:
1316: In the low frequency limit the dispersion curve and the spatial dependence of
1317: the modes take on the simple analytic forms shown in Table \ref{Table_Modes},
1318: allowing us to make analytic predictions for the scattering at low
1319: frequencies, and then the thermal conductance at low temperatures. Since only
1320: small wave vector scattering is involved in these calculations, the results
1321: are true for a general roughness correlation function, providing $\tilde
1322: {g}(0)$ is nonzero. The mode structure in Table \ref{Table_Modes} may be
1323: calculated from Eqs. (\ref{even-ux-filed}-\ref{odd_flex}) taking
1324: $k\rightarrow0$ or from arguments of macroscopic elasticity theory.
1325:
1326: \squeezetable
1327: \begin{table}[ptb]
1328: %\centering
1329: \begin{ruledtabular}
1330: \begin{tabular}
1331: [c]{c|c}
1332: in-plane & flexural\\\hline%
1333: \begin{tabular}
1334: [c]{c|c|c}%
1335: cc & bb & bc,cb\\\hline
1336: $2\bar{\omega}^{2}$ & $\sqrt{3}\bar{\omega}$ & $\frac{3^{5/4}}{2^{3/2}}%
1337: \bar{\omega}^{3/2}$%
1338: \end{tabular}
1339: &
1340: \begin{tabular}
1341: [c]{c|c|c}%
1342: tt & bb & tb,bt\\\hline $\frac{9(1+\sigma)}{4}\left(
1343: \frac{W\bar{\omega}}{d}\right) ^{2}$ & O$\left[ \left(
1344: \frac{W\bar{\omega}}{d}\right) ^{3}\right] $ &
1345: $\frac{3^{5/4}(1+\sigma)^{1/2}}{4}\left(
1346: \frac{W\bar{\omega}}{d}\right)
1347: ^{3/2}$%
1348: \end{tabular}
1349: %\\\hline\hline
1350: \end{tabular}
1351: \end{ruledtabular}
1352: \caption{Scattering coefficients for the zero onset frequency modes at low
1353: frequencies: c denotes compression, b denotes bend, t denotes
1354: torsion, bb
1355: denotes bend to bend scattering etc. Values are quoted for $\gamma_{m}W^{4}%
1356: /\tilde{g}(0)$ as a function of scaled frequency
1357: $\bar{\omega}=\omega c_{E}/W$. For the flexural bend to bend
1358: scattering (bb) the terms in the braces in Eq.\ (\ref{gamma-thin})
1359: cancel to leading order resulting in very small
1360: $O(\bar{\omega}^{3})$ scattering. There is no scattering between
1361: in-plane and flexural modes for the z-independent roughness assumed.}%
1362: \label{Table_Scattering}%
1363: \end{table}
1364:
1365:
1366: %TCIMACRO{\FRAME{ftbpFU}{3.9998in}{3in}{0pt}{\Qcb{Attenuation coefficient
1367: %$\gamma_{m}W^{4}/\tilde{g}(0)$ for scattering from the two lowest $m=0$
1368: %inplane modes to any other mode as a function of scaled frequency $\omega
1369: %W/c_{t}$: solid line - inplane bend mode; dashed line - compression mode. The
1370: %insert shows an enlargement of the low frequency region, and compares with the
1371: %analytic low frequency expressions from Table \ref{Table_Scattering}: dotted -
1372: %analytic inplane bend mode; dash-dotted - analytic compression mode; other
1373: %lines as in the main figure.}}{\Qlb{Fig_scattinp0}}{figure5.ps}%
1374: %{\special{ language "Scientific Word"; type "GRAPHIC";
1375: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1376: %width 3.9998in; height 3in; depth 0pt; original-width 0pt;
1377: %original-height 0pt; cropleft "0"; croptop "1"; cropright "1";
1378: %cropbottom "0"; filename 'figure5.ps';file-properties "XNPEU";}}}%
1379: %BeginExpansion
1380: \begin{figure}
1381: [ptb]
1382: \begin{center}
1383: \includegraphics[
1384: height=3in,
1385: width=3.9998in
1386: ]%
1387: {figure5.eps}%
1388: \caption{Attenuation coefficient $\gamma_{m}W^{4}/\tilde{g}(0)$ for scattering
1389: from the two lowest $m=0$ inplane modes to any other mode as a function of
1390: scaled frequency $\omega W/c_{t}$: solid line - inplane bend mode; dashed line
1391: - compression mode. The insert shows an enlargement of the low frequency
1392: region, and compares with the analytic low frequency expressions from Table
1393: \ref{Table_Scattering}: dotted - analytic inplane bend mode; dash-dotted -
1394: analytic compression mode; other lines as in the main figure.}%
1395: \label{Fig_scattinp0}%
1396: \end{center}
1397: \end{figure}
1398: %EndExpansion
1399: The contributions to the thermal attenuation coefficient in the
1400: low frequency limit ($\omega W/c_{t}\ll1$) from the various
1401: scattering processes are shown in the Table \ref{Table_Scattering}
1402: \footnote{A more accurate expression for the scattering between
1403: the in-plane compression and bending modes is giving by keeping
1404: the next order term in $\bar{\omega}$ which is
1405: $O(\bar{\omega}^{1/2})$ for this scattering process. To include
1406: this correction multiply the expression in the table by
1407: $(1+\sqrt{\bar{\omega}}/12^{1/4})^{2}$.}. The expressions take on
1408: their simplest form if we introduce the frequency scaled with the
1409: velocity of the long wavelength compression mode
1410: $\bar{\omega}=\omega c_{E}/W$ with
1411: $c_{E}=\sqrt{E/\rho}=\sqrt{2(1+\sigma)}c_{t}$. The power laws can
1412: largely be understood from the prefactor in
1413: Eq.~(\ref{gamma_thin_nonorm}), $\gamma
1414: _{n,m}\propto\omega^{2}/v_{g}^{\left( m\right) }v_{g}^{\left(
1415: n\right) }$. The group velocity $v_{g}$ becomes a constant at
1416: small frequencies for the compression and torsion modes. Thus the
1417: torsion-torsion and compression-compression scattering shows the
1418: $\omega^{2}$ dependence corresponding to Rayleigh scattering in
1419: one dimension, and as was found for scalar waves with linear
1420: dispersion. On the other hand for the bending modes
1421: $v_{g}\propto\omega^{1/2}$. This has the important consequence
1422: that the in-plane bend-bend scattering increases more rapidly at
1423: low frequencies proportional to $\omega$, and the torsion-bend and
1424: compression-bend scattering have an $\omega^{3/2}$ frequency
1425: dependence. For the flexural bend-bend scattering the two terms in
1426: the braces in Eq.\ (\ref{gamma-thin}) cancel to leading order
1427: resulting in smaller scattering $O(\omega^{3})$ than given by the
1428: prefactor alone. Note that the expressions for the flexural modes
1429: involve additional factors of $W/d$, so that these modes will be
1430: scattered more strongly at a given $\omega$ in the thin plate
1431: limit. This is because these
1432: modes are softer, so that the scattering wave vectors are larger for the same frequency.%
1433:
1434:
1435:
1436:
1437: \begin{figure}
1438: [ptb]
1439: \begin{center}
1440: \includegraphics[
1441: height=3in, width=3.9998in
1442: ]%
1443: {figure6.eps}%
1444: \caption{Attenuation coefficient $\gamma_{m}W^{4}/\tilde{g}(0)$ for scattering
1445: from the two lowest $m=0$ flex modes to any other mode as a function of scaled
1446: frequency $\omega\sqrt{12(1-\sigma^{2})}(W/d)W/c_{E}$: solid line - the
1447: flex-bend mode; dashed line - torsion mode. The insert shows an enlargement of
1448: the low frequency region, and compares with the analytic low frequency
1449: expressions from Table \ref{Table_Scattering}: dotted line - analytic
1450: approximation for the flex-bend mode; dash-dotted: analytic expression for the
1451: torsion mode; other lines as in the main figure.}%
1452: \label{Fig_scatflex0}%
1453: \end{center}
1454: \end{figure}
1455:
1456: Numerical results for the attenuation coefficient $\gamma_{m}$ of
1457: the four lowest modes are shown in Fig.\ (\ref{Fig_scattinp0}) for
1458: the in-plane and Fig.\ (\ref{Fig_scatflex0}) for the flexural
1459: modes. The plot for the in-plane modes in particular shows
1460: interesting structure deriving from the complicated dispersion
1461: curves of Fig.\ \ref{dispersion}. Much of this structure can be
1462: understood from the product of group velocities in the denominator
1463: of Eq.\ (\ref{gamma-thin}). In particular there is a square root
1464: divergence in $\gamma_{m}$ at the onset frequency of each mode
1465: where the group velocity is zero. In addition, the large
1466: scattering around $\omega W/c_{t}=5$ derives from the region of
1467: anomalous dispersion, since the group velocity is small in this
1468: frequency range. The insert to Fig.\ (\ref{Fig_scattinp0})\ shows
1469: an expanded view of the low frequency behavior, using the results
1470: from Table \ref{Table_Scattering} together with the next order
1471: correction for the compression-bend scattering. The agreement for
1472: the compression mode is very good even up to $\omega
1473: W/c_{t}\sim3$, whereas for the bend mode the correspondence is
1474: only good for $\omega W/c_{t}\lesssim0.5$. The scattering for the
1475: flexural modes shows generally similar results, Fig.\
1476: (\ref{Fig_scatflex0}) although the behavior is simpler
1477: corresponding to the rather featureless dispersion curves. At low
1478: frequencies, (insert to Fig.\ (\ref{Fig_scatflex0})), the
1479: scattering of the flexural-bend mode is
1480: small, since the intramode scattering is reduced by the cancellation discussed above.%
1481:
1482: %TCIMACRO{\FRAME{ftbpFU}{4.587in}{3.9946in}{0pt}{\Qcb{Total scattering
1483: %$\sum_{m}\gamma_{m}W^{4}/\tilde{g}(0)$ for the in-plane modes on a log-log
1484: %plot. The dotted line shows the low frequency analytic expression from Table
1485: %\ref{Table_Scattering}, and the dashed line shows a power law $4$. (Note that
1486: %the heights of the peaks in the plot are not significant, depending on how
1487: %close the individual points, separated by $0.01$ in $\omega W/c_{t}$, used in
1488: %constructing the plot are to the mode onset frequencies where the scattering
1489: %diverges.)}}{\Qlb{scatt_loglog}}{figure7.eps}%
1490: %{\special{ language "Scientific Word"; type "GRAPHIC";
1491: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1492: %width 4.587in; height 3.9946in; depth 0pt; original-width 5.8072in;
1493: %original-height 5.0531in; cropleft "0"; croptop "1"; cropright "1";
1494: %cropbottom "0"; filename 'figure7.eps';file-properties "XNPEU";}}}%
1495: %BeginExpansion
1496: \begin{figure}
1497: [ptb]
1498: \begin{center}
1499: \includegraphics[
1500: height=3in, width=3.9998in
1501: ]%
1502: {figure7.eps}%
1503: \caption{Total scattering $\sum_{m}\gamma_{m}W^{4}/\tilde{g}(0)$ for the
1504: in-plane modes on a log-log plot. The dotted line shows the low frequency
1505: analytic expression from Table \ref{Table_Scattering}, and the dashed line
1506: shows a power law $4$. (Note that the heights of the peaks in the plot are not
1507: significant, depending on how close the individual points, separated by $0.01$
1508: in $\omega W/c_{t}$, used in constructing the plot are to the mode onset
1509: frequencies where the scattering diverges.)}%
1510: \label{scatt_loglog}%
1511: \end{center}
1512: \end{figure}
1513: %EndExpansion
1514:
1515: %\begin{figure}[ptb]
1516: %\begin{center}
1517: %\includegraphics[ height=3.103in, width=4.2272in ]{flex-mode.eps}
1518: %\end{center}
1519: %\caption{Attenuation coefficient for out-of plane modes to all possible
1520: %scattering modes. At low frequencies the scattering is dominated by inter-mode
1521: %scattering and obeys $\omega^{3/2}$ power law. Analytical expression for this
1522: %mode is plotted as a dashed line and the analytical sum of all modes are shown
1523: %as a dotted line .}%
1524: %\label{flex-mode}%
1525: %\end{figure}
1526:
1527: Figure\ (\ref{scatt_loglog}) shows the total scattering
1528: $\sum_{m}\gamma_{m}$ for the in-plane modes on a log-log plot,
1529: again with white noise roughness. At very low frequencies the
1530: scattering varies proportional to $\omega$ corresponding to the
1531: dominant intramode scattering of the compression mode at low
1532: frequencies (Table \ref{Table_Scattering}). For frequencies up to
1533: $\omega W/c_{t}\simeq3.5$, the first nonzero onset frequency of an
1534: in-plane mode, the analytic low-frequency expression given by
1535: summing the in-scattering expressions from Table
1536: \ref{Table_Scattering} (cc, cb, bc, and bb), shown as the dotted
1537: line in Fig.\ \ref{scatt_loglog}, gives a good approximation to
1538: the full results. At higher frequencies the total scattering
1539: increases rapidly, following a general trend proportional to
1540: $\omega^{4}$ (dashed line) together with divergent scattering at
1541: each mode onset frequency. The $\omega^{4}$ power law can be
1542: understood as the combination of the explicit $\omega^{2}$
1543: dependence of Eq. (\ref{gamma-thin}), together with two powers of
1544: $\omega$ coming from the number of modes available for scattering
1545: from and to.
1546:
1547: %The total scattering of the flexural modes is shown in Fig.\ (\ref{flex-mode}%
1548: %). At low frequencies the attenuation coefficient is dominated by
1549: %$\omega^{3/2}$ torsion-bending mode scattering (dashed line). For $\sqrt{\rho
1550: %d/D}\omega W^{2}<10$, the analytical expression for the total flexural
1551: %scattering from Table \ref{Table_Scattering} agrees well with the numerical
1552: %results. At higher frequency, $\sqrt{\rho d/D}\omega W^{2}>50$, the
1553: %attenuation coefficient again follows a trend proportional to $\omega^{4}$
1554: %(not shown in the figure).
1555:
1556: \subsection{Change in the thermal conductance}
1557:
1558: In the weak scattering limit the change in thermal conductance at low
1559: temperatures can be derived directly from the expressions for the scattering
1560: at low frequencies. If we write the thermal attenuation coefficient of mode
1561: $m$ as $\gamma_{m}L=A(\omega/\omega_{0})^{p}$, where $p$ is the power law
1562: obtained in the low frequency limit and $\omega_{0}$ some characteristic
1563: frequency, then the corresponding contribution of the suppression of the
1564: thermal conductance from this mode is
1565: \begin{equation}
1566: \delta K_{m}/K_{u}=AI_{p}(T/T_{0})^{p} \label{temp-power}%
1567: \end{equation}
1568: with $T_{0}=\hbar\omega_{0}/k_{B}$ \ the corresponding characteristic
1569: temperature and $K_{u}=\pi^{2}k_{B}^{2}T/3h$ the universal thermal
1570: conductance. The constant $I_{p}$ can be obtained evaluating the integral%
1571: \begin{equation}
1572: I_{p}=\frac{3}{\pi^{2}}\int_{0}^{\infty}dy\frac{y^{p+2}e^{y}}{(e^{y}-1)^{2}}.
1573: \end{equation}
1574: Thus the power law for the temperature dependence of the depression of the
1575: thermal conductivity is the same as the one for the low frequency behavior of
1576: the scattering coefficient.
1577:
1578: Figures (\ref{dK-inplane}) and (\ref{dK-flex}) show the thermal conductance
1579: depression scaled with the universal value $K_{u}$ as a function of the
1580: appropriate scaled temperature for the lowest in-plane and flexural modes,
1581: showing the deviation from the low temperature power laws as the temperature
1582: is raised. For the in plane modes we use the characteristic temperature
1583: $T_{E}=\hbar c_{E}/k_{B}W$ and for the flexural modes $T_{F}=\hbar
1584: c_{E}d/k_{B}W^{2}$. The individual plots are then independent of the geometry.
1585: To combine the contributions from the in-plane and flexural modes the ratio
1586: $d/W$ is needed to relate the two temperature scale factors. In the thin plate
1587: limit $T_{F}=(d/W)T_{E}\ll T_{E}$.%
1588: %TCIMACRO{\FRAME{ftbpFU}{3.8752in}{2.5832in}{0pt}{\Qcb{Reduction in the thermal
1589: %conductance scaled with the universal conductance $K_{u}$ for the lowest
1590: %in-plane modes as a function of scaled temperature $T/T_{E}$ with $T_{E}=\hbar
1591: %c_{E}/k_{B}W$: solid line - low temperature analytical expressions from
1592: %Table \ref{Table_Scattering}: points - full expression evaluated numerically.
1593: %The quantity plotted is $(\delta K_{c}+\delta K_{ib})/2K_{u}$ with $\delta
1594: %K_{c},\delta K_{ib}$ the depression of the contributions to the conductance by
1595: %the scattering for the compression and in-plane bending modes.}}%
1596: %{\Qlb{dK-inplane}}{figure8.eps}{\special{ language "Scientific Word";
1597: %type "GRAPHIC"; maintain-aspect-ratio TRUE; display "USEDEF";
1598: %valid_file "F"; width 3.8752in; height 2.5832in; depth 0pt;
1599: %original-width 0pt; original-height 0pt; cropleft "0"; croptop "1";
1600: %cropright "1"; cropbottom "0";
1601: %filename 'figure8.eps';file-properties "XNPEU";}}}%
1602: %BeginExpansion
1603: \begin{figure}
1604: [ptb]
1605: \begin{center}
1606: \includegraphics[
1607: height=3in, width=3.9998in
1608: ]%
1609: {figure8.eps}%
1610: \caption{Reduction in the thermal conductance scaled with the universal
1611: conductance $K_{u}$ for the lowest in-plane modes as a function of
1612: scaled temperature $T/T_{E}$ with $T_{E}=\hbar c_{E}/k_{B}W$:
1613: solid line - low temperature analytical expressions from Table
1614: \ref{Table_Scattering}: points - full expression evaluated
1615: numerically. The quantity plotted is $(\delta K_{c}+\delta
1616: K_{ib})/2K_{u}$ with $\delta K_{c},\delta K_{ib}$ the depression
1617: of the contributions to the conductance by the scattering for the
1618: compression
1619: and in-plane bending modes.}%
1620: \label{dK-inplane}%
1621: \end{center}
1622: \end{figure}
1623: %EndExpansion%
1624: %TCIMACRO{\FRAME{ftbpFU}{3.9167in}{2.6109in}{0pt}{\Qcb{Similar to
1625: %Fig.$\;$(\ref{dK-inplane}), $\delta K/2K_{u}$ for the lowest flexural modes
1626: %(torsion and flexural-bending) as a function of the scaled temperature
1627: %$T/T_{F}$ with $T_{F}=\hbar c_{E}d/k_{B}W^{2}$.}}{\Qlb{dK-flex}}%
1628: %{figure9.eps}{\special{ language "Scientific Word"; type "GRAPHIC";
1629: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1630: %width 3.9167in; height 2.6109in; depth 0pt; original-width 0pt;
1631: %original-height 0pt; cropleft "0"; croptop "1"; cropright "1";
1632: %cropbottom "0"; filename 'figure9.eps';file-properties "XNPEU";}}}%
1633: %BeginExpansion
1634: \begin{figure}
1635: [ptbptb]
1636: \begin{center}
1637: \includegraphics[
1638: height=3in, width=3.9998in
1639: ]%
1640: {figure9.eps}%
1641: \caption{Similar to Fig.$\;$(\ref{dK-inplane}), $\delta K/2K_{u}$ for the
1642: lowest flexural modes (torsion and flexural-bending) as a function of the
1643: scaled temperature $T/T_{F}$ with $T_{F}=\hbar c_{E}d/k_{B}W^{2}$.}%
1644: \label{dK-flex}%
1645: \end{center}
1646: \end{figure}
1647: %EndExpansion
1648:
1649: \section{Comparison with Experiment}
1650:
1651: \label{Sec_experiment}
1652:
1653: \subsection{Experimental geometry}
1654:
1655: Based on the SEM micrograph of the experimental
1656: structure\cite{Ssp}, we set the dimensions of the structure in the
1657: following way. In the experimental structure of Schwab et al.\ the
1658: thermal pathway was constructed with the shape function
1659: $W(x)=W\cosh(Ax)$ so that the beam width becomes large and joins
1660: smoothly to the thermal reservoirs at the ends, reducing the
1661: scattering due to the geometric imperfection at these junctions.
1662: Unfortunately this makes the calculation of the behavior of the
1663: elastic waves in the beams much harder. However both with and
1664: without the scattering off surface roughness, we expect the narrow
1665: portion of the beam to dominate the behavior. Thus we simplify the
1666: structure and model it as an elastic beam with rectangular cross
1667: section of width $W$, depth $d$, and effective length $L$. We
1668: estimate the width as the narrowest width of the structure,
1669: $W\simeq160nm$, and $L=1$ $\mu$m as the length over which the
1670: width is approximately constant. The thickness of the material was
1671: $d=60$ nm. The accuracy of the length estimation is not very
1672: critical, since the only length dependence in the scattering rate
1673: $\gamma$ appears in the combination $\delta^{2}L$ where $\delta$
1674: is the rms roughness which is a parameter of the model, so that
1675: any error in the assignment of $L$ will just change the value
1676: assigned to $\delta $. The width $W$ on the other hand plays a
1677: crucial role, for example determining the frequency cutoffs of the
1678: various modes, and so the temperature dependence of the thermal
1679: conductivity.
1680:
1681: \subsection{Roughness correlation function}
1682:
1683: Since the nature of the surface roughness on the experimental structure is not
1684: known, to fit the experimental data we need a sensible parameterization of the
1685: roughness. As a starting point we choose a Gaussian correlation function for
1686: the roughness, leading to the spectral density%
1687: \begin{equation}
1688: \tilde{g}\left( k\right) =\sqrt{\pi}a\delta^{2}\exp\left[ -\frac{a^{2}}%
1689: {4}k^{2}\right] . \label{Gaussian}%
1690: \end{equation}
1691: This parameterization of the roughness contains two parameters: $\delta$ the
1692: rms roughness and $a$ the correlation length.
1693:
1694: To analyze the data, we first quantify the amount of scattering by
1695: subtracting the data of Schwab et al. from the ideal thermal
1696: conductance obtained numerically using the ``xyz''
1697: algorithm\cite{NAW97}. Then we attempt to fit the data by
1698: adjusting the two parameters $a$ and $\delta^{2}L$.
1699: %TCIMACRO{\FRAME{ftbpFU}{3.8233in}{2.5486in}{0pt}{\Qcb{Attempts to fit the low
1700: %temperature data $T\lesssim0.2$ K using various values of $a\delta^{2}$: solid
1701: %line - $\sqrt{\pi}a\delta^{2}=0.1$; dotted line - $\sqrt{\pi}a\delta^{2}%
1702: %=0.05$; dashed line - $\sqrt{\pi}a\delta^{2}=0.02$; open circles -
1703: %from the experimental data of Schwab et al.}}{\Qlb{a-delta2}}{figure10.ps}%
1704: %{\special{ language "Scientific Word"; type "GRAPHIC";
1705: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1706: %width 3.8233in; height 2.5486in; depth 0pt; original-width 0pt;
1707: %original-height 0pt; cropleft "0"; croptop "1"; cropright "1";
1708: %cropbottom "0"; filename 'figure10.ps';file-properties "XNPEU";}}}%
1709: %BeginExpansion
1710: \begin{figure}
1711: [ptb]
1712: \begin{center}
1713: \includegraphics[
1714: height=2.5486in,
1715: width=3.8233in
1716: ]%
1717: {figure10.ps}%
1718: \caption{Attempts to fit the low temperature data $T\lesssim0.2$ K using
1719: various values of $a\delta^{2}$: solid line - $\sqrt{\pi}a\delta^{2}=0.1$;
1720: dotted line - $\sqrt{\pi}a\delta^{2}=0.05$; dashed line - $\sqrt{\pi}%
1721: a\delta^{2}=0.02$; open circles - from the experimental data of Schwab et al.}%
1722: \label{a-delta2}%
1723: \end{center}
1724: \end{figure}
1725: %EndExpansion
1726:
1727: The inadequacy of Eq.\ (\ref{Gaussian}) in fitting the experimental data is
1728: shown by the low temperature fits in Fig.\ (\ref{a-delta2}). At these low
1729: temperatures only small wave number modes are excited, so that the exponential
1730: term in Eq.\ (\ref{Gaussian}) can be approximated as unity and $\tilde
1731: {g}\left( k\right) \simeq\tilde{g}\left( 0\right) =\sqrt{\pi}a\delta^{2}$.
1732: Thus the roughness parameters only appear in the combination $a\delta^{2}$,
1733: and this quantity can be varied to attempt to fit the low temperature region.
1734: As seen from the figure, increasing $a\delta^{2}$ causes scattering that is
1735: systematically larger than the experimental data at the low temperatures,
1736: while decreasing $a\delta^{2}$ does not provide enough scattering in the range
1737: $0.1<T<0.2$ K.
1738:
1739: Although there is considerable scatter in the data over the range of the fit,
1740: the systematic differences between the predictions and the data lead us to
1741: propose a modified form of the roughness correlation that reduces the
1742: scattering at small wave numbers%
1743: \begin{equation}
1744: \tilde{g}\left( k\right) =\sqrt{\pi}a\delta^{2}\exp\left[ -\frac{a^{2}%
1745: \left( k-k_{0}\right) ^{2}}{4}\right] . \label{Gaussian-shift}%
1746: \end{equation}
1747: A nonzero value of the parameter $k_{0}$ leads to a roughness
1748: correlation function that is maximum at a length scale of order
1749: $k_{0}^{-1}$, and serves to reduce the scattering at long
1750: wavelengths. As mentioned in the introduction section, the same
1751: discrepancy (i.e., the overestimation of the scattering at long
1752: wavelengths in the theory compared with experiment) was found
1753: using the scalar model of the elastic waves\cite{SC00}. The full
1754: elasticity theory considered here actually makes the discrepancy
1755: worse, since the scattering at small frequencies now is predicted
1756: to increase more rapidly at small frequencies than the
1757: $\omega^{2}$ found in the scalar theory, varying as $\omega^{p}$
1758: with
1759: $p<2$ for most of the scattering processes, see Table \ref{Table_Scattering}.%
1760:
1761: %TCIMACRO{\FRAME{ftbpFU}{3.3252in}{2.4751in}{0pt}{\Qcb{Thermal conductance per
1762: %mode scaled with universal value $K_{u}$: solid line - fit using roughness
1763: %parameters $a/W=5.5$, $\delta/W=0.2$, and $k_{0}W=4.9$; circles - data of
1764: %Schwab et al. The dotted line shows the ideal value with no scattering.}%
1765: %}{\Qlb{fitK}}{figure11.ps}{\special{ language "Scientific Word";
1766: %type "GRAPHIC"; maintain-aspect-ratio TRUE; display "USEDEF";
1767: %valid_file "F"; width 3.3252in; height 2.4751in; depth 0pt;
1768: %original-width 5.3212in; original-height 3.9522in; cropleft "0";
1769: %croptop "1"; cropright "1"; cropbottom "0";
1770: %filename 'figure11.ps';file-properties "XNPEU";}}}%
1771: %BeginExpansion
1772: \begin{figure}
1773: [ptb]
1774: \begin{center}
1775: \includegraphics[
1776: height=2.4751in,
1777: width=3.3252in
1778: ]%
1779: {figure11.ps}%
1780: \caption{Thermal conductance per mode scaled with universal value $K_{u}$:
1781: solid line - fit using roughness parameters $a/W=5.5$, $\delta/W=0.2$, and
1782: $k_{0}W=4.9$; circles - data of Schwab et al. The dotted line shows the ideal
1783: value with no scattering.}%
1784: \label{fitK}%
1785: \end{center}
1786: \end{figure}
1787: %EndExpansion
1788: To fit the data of Schwab et al, we need to determine three
1789: parameters: $k_{0}$, $a$, and $\delta$. As well as looking at the
1790: fit by eye, we evaluate the quality of the fit by calculating the
1791: mean square deviation of the data from the theory curve over the
1792: temperature range up to $0.4K$. At higher temperatures many modes
1793: becoming excited, and the scattering of individual modes becomes
1794: strong, so that our theory is less reliable. Since the onset
1795: frequency of the scattering at low frequencies and the initial
1796: decrease in thermal conductance with increasing temperature near
1797: the onset is mainly determined by $k_{0}$, this parameter is the
1798: easiest to determine. We find the value $k_{0}W=4.9$ , rather
1799: insensitive to the values of $a$ and $\delta$. We have also
1800: investigated the fit with a delta function noise correlation
1801: function (i.e. $a\rightarrow0$ in Eq.\ (\ref{Gaussian-shift})).
1802: With this roughness, to get a maximum in the decrease in the
1803: thermal conductance at about $T\sim0.2K$ requires $k_{0}W$ values
1804: between $4.5$ and $5$, which agrees with the previous estimate.%
1805:
1806: %TCIMACRO{\FRAME{ftbpFU}{3.614in}{2.8253in}{0pt}{\Qcb{Same as in
1807: %Fig.\ (\ref{fitK}) but showing the decrease of $K/K_{u}$ from the ideal
1808: %value.}}{\Qlb{fitdeltaK}}{figure12.ps}%
1809: %{\special{ language "Scientific Word"; type "GRAPHIC";
1810: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1811: %width 3.614in; height 2.8253in; depth 0pt; original-width 5.028in;
1812: %original-height 3.9245in; cropleft "0"; croptop "1"; cropright "1";
1813: %cropbottom "0"; filename 'figure12.ps';file-properties "XNPEU";}}}%
1814: %BeginExpansion
1815: \begin{figure}
1816: [ptb]
1817: \begin{center}
1818: \includegraphics[
1819: height=2.8253in,
1820: width=3.614in
1821: ]%
1822: {figure12.ps}%
1823: \caption{Same as in Fig.\ (\ref{fitK}) but showing the decrease of $K/K_{u}$
1824: from the ideal value.}%
1825: \label{fitdeltaK}%
1826: \end{center}
1827: \end{figure}
1828: %EndExpansion
1829: The remaining fit parameters $a$ and $\delta$\ are not well
1830: determined. Visual inspection of the fit and a plot of the error
1831: as a function of $(a,\delta)$ shows a reasonable fit over the
1832: range $a/W=5,\delta/W=0.17$ to $a/W=6.7,\delta /W=0.32$. A choice
1833: of the roughness amplitude towards the lower end of this range
1834: seems most likely physically (e.g., $\delta/W=0.31\simeq50$ nm is
1835: unlikely for roughness generated by chemical etch). We therefore
1836: use $\delta/W=0.2$ and $a/W=5.5$ to generate the curves.
1837:
1838: A difficulty of fitting the data is the lack of data points at
1839: very low temperatures: it is in this range where only a few modes
1840: are involved that we have a very good understanding of the
1841: scattering. At higher temperatures many more modes become
1842: involved, and the scattering of individual modes becomes strong,
1843: so that the second order approximation used in calculating the
1844: scattering will not be good. A\ full test of the theory explaining
1845: the reduction in the thermal conductance in terms of the
1846: scattering off surface roughness requires more data below a
1847: temperature of about $0.08K$ for the type of geometry used by
1848: Schwab et al., or systems with smaller geometries where the
1849: effects can be measured at higher temperatures.
1850:
1851: \subsection{Individual mode contribution to the thermal conductance}%
1852:
1853: %TCIMACRO{\FRAME{ftbpFU}{3.8649in}{2.5763in}{0pt}{\Qcb{Individual mode
1854: %contribution to the thermal conductance. The lowest two flex modes and lowest
1855: %three in plane modes are shown. The contributions to $K/K_{u}$ from the four
1856: %modes with zero onset frequency tend to unity at low temperatures. The higher
1857: %modes only contribute at higher temperature. The modes are: dash-dotted -
1858: %in-plane bending; dashed - compression; dotted - torsion; dashed-dotted-dot -
1859: %out-of plane bending. The solid line shows the sum of all the mode
1860: %contributions, reduced by $4K_{u}$. Values of the roughness parameters used
1861: %were $a/W=5.5$, $\delta/W=0.2$, $k_{0}W=4.9$, and $d/W=0.375$.}}{\Qlb{Kmode}%
1862: %}{figure13.eps}{\special{ language "Scientific Word"; type "GRAPHIC";
1863: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "F";
1864: %width 3.8649in; height 2.5763in; depth 0pt; original-width 0pt;
1865: %original-height 0pt; cropleft "0"; croptop "1"; cropright "1";
1866: %cropbottom "0"; filename 'figure13.eps';file-properties "XNPEU";}}}%
1867: %BeginExpansion
1868: \begin{figure}
1869: [ptb]
1870: \begin{center}
1871: \includegraphics[
1872: height=2.5763in,
1873: width=3.8649in
1874: ]%
1875: {figure13.eps}%
1876: \caption{Individual mode contribution to the thermal conductance. The lowest
1877: two flex modes and lowest three in plane modes are shown. The
1878: contributions to $K/K_{u}$ from the four modes with zero onset
1879: frequency tend to unity at low temperatures. The higher modes only
1880: contribute at higher temperature. The modes are: dash-dotted -
1881: in-plane bending; dashed - compression; dotted - torsion;
1882: dashed-dotted-dot - out-of plane bending. The solid line shows the
1883: sum of all the mode contributions, reduced by $4K_{u}$. Values of
1884: the roughness parameters used were $a/W=5.5$, $\delta/W=0.2$,
1885: $k_{0}W=4.9$, and
1886: $d/W=0.375$.}%
1887: \label{Kmode}%
1888: \end{center}
1889: \end{figure}
1890: %EndExpansion
1891:
1892: It is interesting to investigate the contribution to the total
1893: thermal conductance of the individual modes with the roughness
1894: parameters used to fit the experimental data. This is shown in
1895: Fig.\ (\ref{Kmode}). The flex-bending mode shows a much smaller
1896: contribution to the reduction in $K$ at low temperatures for the
1897: reason we have already discussed. The modes with nonzero onset
1898: frequencies start to contribute significantly above about
1899: $T\simeq0.2K$, and this is the predominant cause for the increase
1900: in thermal conductivity above this temperature, since the recovery
1901: of the thermal conductance for the lowest mode occurs very slowly.
1902:
1903: \section{Conclusion}
1904:
1905: \label{Sec_conclusion}We have investigated the effect of surface roughness on
1906: the scattering of elastic waves in a rectangular beam or
1907: waveguide, and the resulting depression of the thermal conductance
1908: in the low temperature quantized limit, using full elasticity
1909: theory. Our formulation is quite general, but to obtain concrete
1910: results we have specialized to the thin-plate limit, which should
1911: be a reasonable approximation for many mesoscopic experiments
1912: where the depth of the structures is fixed by the epitaxial
1913: growth, whilst the width is determined lithographically. The thin
1914: plate limit preserves the peculiar features of the elastic waves
1915: in the full elastic theory, namely a quadratic dispersion at long
1916: wavelengths for two of the low frequency modes, and regions of
1917: negative dispersion in the spectra. A robust result is that the
1918: low frequency asymptotic dependence of the scattering by
1919: unstructured roughness of the modes that propagate at low
1920: frequencies (the ones that are important in the low temperature
1921: universal thermal conductance) depends on the structure of the
1922: modes and the dispersion relation, and is \emph{not} the simple
1923: $\omega^{2}$ dependence of Rayleigh scattering as found in the
1924: scalar approximation to the modes. We find different power laws
1925: for the various mode scattering processes that can be understood
1926: largely from the dispersion relations at: $\omega$ for intramode
1927: scattering for the in-plane bend mode (the flex-bend intramode
1928: scattering is anomalous because of a cancellation between leading
1929: order terms, and varies as $\omega^{3}$); $\omega^{3/2}$ for
1930: scattering between the bend modes and the modes with linear
1931: dispersion (torsion and compression modes); and the usual
1932: $\omega^{2}$ for the intramode scattering of the modes with linear
1933: dispersion. The current experimental data on the suppression of
1934: the low temperature thermal conductance below the universal value
1935: does not extend to low enough temperatures to provide a good test
1936: of these predictions. To investigate this prediction further, it
1937: would be interesting to extend the experiments to lower
1938: temperatures, or to smaller devices such as carbon nanotubes,
1939: where the characteristic temperature scales (when a typical
1940: thermally excited phonon has a wavelength comparable with the
1941: device dimensions) are higher.
1942:
1943: We have used our results to understand the data the data of Schwab
1944: et al.,\ who observed a depression of the thermal conductance
1945: below the universal value in the temperature range of $0.1K$ to
1946: $0.4K$. Although the scatter in the data is considerable at these
1947: low temperatures, the observations seem to show a delay in the
1948: onset of the depression scattering as the temperature is raised,
1949: beyond what can be fitted with our predictions for unstructured
1950: surface roughness. We tentatively resolve this delay by supposing
1951: that the surface roughness has a maximum amplitude at some nonzero
1952: length scale, which we parameterize by a shifted Gaussian
1953: correlation function. Due to the lack of data at low temperatures,
1954: a precise determination of the roughness parameters is not
1955: possible. However, we do obtain a fit to the data with parameters
1956: that do not look unreasonable when compared with electron
1957: micrographs of the actual devices.
1958:
1959: Our results are based on second order perturbation theory, and the
1960: thermal conductance is evaluated assuming the scattering over the
1961: length of the device is small. This is a good approximation at low
1962: temperatures, but the scattering becomes strong at higher
1963: temperatures, particularly for the new modes excited as the
1964: temperature is raised, which have a diverging scattering at onset
1965: due to the flat dispersion relation here. At higher temperatures
1966: multiple scattering and perhaps phonon localization will therefore
1967: become important. Kambili et al.\cite{KFFL99} and Sanchez-Gil et
1968: al. \cite{SFMY99}\ have numerically investigated the these effects
1969: in the simplified scalar wave approximation. It would be
1970: interesting in the future to extend their work to the full
1971: elasticity model.
1972:
1973: \begin{acknowledgments}
1974: This work was supported by NSF grant no. DMR-9873573. We thank Ruben
1975: Krasnopolsky for help on the numerical codes.
1976: \end{acknowledgments}
1977:
1978: \appendix
1979:
1980: \section{Incident and Scattered Fields}
1981:
1982: \label{Appendix_Separate}Using Green's theorem we have expressed the
1983: displacement field at frequency $\omega$ in terms of the surface integral%
1984: \begin{equation}
1985: u_{q}(\mathbf{x})=\int_{S^{\prime}}\left[ n_{j}^{\prime}T_{ij}\left(
1986: \mathbf{x}^{\prime}\right) G_{iq}\left( \mathbf{x}^{\prime},\mathbf{x}%
1987: \right) -n_{j}^{\prime}u_{i}\left( \mathbf{x}^{\prime}\right) \Gamma
1988: _{ijq}\left( \mathbf{x}^{\prime},\mathbf{x}\right) \right] dS^{\prime}
1989: \label{total field2}%
1990: \end{equation}
1991: Eq.\ (\ref{total field2}) involves the integration over a closed surface
1992: $S^{\prime}$, which we have chosen to be the smooth boundaries together with
1993: the cross sections at $x^{\prime}\rightarrow\pm\infty$. In this appendix we
1994: show that the integration over the sections at $\pm\infty$ simply yields the
1995: incident field $u_{q}^{\mathrm{in}}$, and this allows us to deduce the
1996: expression for the scattered field as an integration over the side surfaces.
1997: To deduce this result, we first need to derive what are known as reciprocity
1998: relations for the elastic modes\cite{A}.
1999:
2000: Let $\mathbf{u}^{(r)}$ and $\mathbf{u}^{(s)}$ be the displacement fields for
2001: modes $r$ and $s$ in the ideal beam, and $\mathbf{T}^{(r)}$, $\mathbf{T}%
2002: ^{(s)}$ the corresponding stress tensor fields. The modes satisfy the wave
2003: equation at frequency $\omega$, so that%
2004: \begin{equation}%
2005: \begin{array}
2006: [c]{c}%
2007: \rho\omega^{2}u_{i}^{(r)}+\partial_{j}T_{ij}^{(r)}=0\\
2008: \rho\omega^{2}u_{i}^{(s)}+\partial_{j}T_{ij}^{(s)}=0
2009: \end{array}
2010: \end{equation}
2011: Multiply the first equation by $u_{i}^{(s)\ast}$\ and the complex conjugate of
2012: the second by $u_{i}^{(r)}$, subtract the two equations, integrate over a
2013: volume of the beam between $x=x_{1}$ and $x=x_{2}$, and finally use the
2014: divergence theorem to find%
2015: \begin{equation}
2016: \int_{S}\left[ u_{i}^{(s)\ast}T_{ij}^{(r)}-u_{i}^{(r)}T_{ij}^{(s)\ast
2017: }\right] \hat{n}_{j}dS=0, \label{uttu}%
2018: \end{equation}
2019: where the integral is over the surface bounding the volume, consisting of the
2020: sides of the beam between $x_{1}$ and $x_{2}$, and the sections at $x_{1}$ and
2021: $x_{2}$. The integrations over the sides of the beam are zero by the stress
2022: free boundary conditions. For the integration over the sections introduce the
2023: explicit $x$-dependence $\mathbf{u}^{(r)}=\mathbf{\phi}(y,z)e^{ik_{r}x}$ and
2024: $\mathbf{T}^{(r)}=\mathbf{\bar{T}}^{(r)}(y,z)e^{ik_{r}x}$ with $k_{r}$ the
2025: wave number of mode $r$ at frequency $\omega$ etc. Then Eq.\ (\ref{uttu})
2026: reduces to
2027: \begin{equation}
2028: \left( 1-e^{i\left( k_{r}-k_{s}\right) \left( x_{1}-x_{2}\right)
2029: }\right) \int\int\left[ \phi_{i}^{(s)\ast}\bar{T}_{ix}^{(r)}-\phi_{i}%
2030: ^{(r)}\bar{T}_{ix}^{(s)\ast}\right] dydz=0
2031: \end{equation}
2032: and the integral is independent of $x$. Unless the prefactor is zero, this
2033: shows us that the integral over the section must be zero, and so
2034: \begin{equation}
2035: \int\int\left[ u_{i}^{(s)\ast}T_{ix}^{(r)}-u_{i}^{(r)}T_{ix}^{(s)\ast
2036: }\right] \,dydz=0,\quad k_{r}\neq k_{s}. \label{Eq_to-reciprocity}%
2037: \end{equation}
2038: This is one version of the reciprocity relations.
2039:
2040: For our purposes it is more convenient to express the condition for the
2041: reciprocity integral to be zero in terms of the group velocity rather than the
2042: wave number. To do so, we need to consider the dispersion curves. The
2043: condition for the reciprocity integral to be nonzero, $k_{r}=k_{s}$ for modes
2044: $r,s$ at the same frequency $\omega$, actually implies $r$ and $s$ are the
2045: \emph{same} mode, so that in fact $v_{g}^{(r)}=v_{g}^{(s)}$. The only other
2046: possibility is that $r$ and $s$ are modes with dispersion curves that cross at
2047: frequency $\omega$, $k=k_{r}=k_{s}$. However only modes of different $y,z$
2048: parity signatures can cross, and then the integration over the section for
2049: these different modes in Eq.\ (\ref{Eq_to-reciprocity}) is again zero. Thus we
2050: can rewrite the reciprocity relation as%
2051: \begin{equation}
2052: \int\int\left[ u_{i}^{(s)\ast}T_{ix}^{(r)}-u_{i}^{(r)}T_{ix}^{(s)\ast
2053: }\right] \,dydz=0,\quad v_{g}^{(r)}\neq v_{g}^{(s)}. \label{Eq_reciprocity_1}%
2054: \end{equation}
2055: If $r$ and $s$ are the same mode, the integral is related to the energy flux
2056: and hence to the group velocity (see Eq.\ (\ref{power}))
2057: \begin{equation}
2058: \int\int dydz\left( u_{i}^{(r)\ast}T_{ij}^{(r)}-u_{i}^{(r)}T_{ij}^{(r)\ast
2059: }\right) =2i\rho\omega v_{g}^{(r)}. \label{Eq_reciprocity_2}%
2060: \end{equation}
2061:
2062: We now use Eqs.\ (\ref{Eq_reciprocity_1},\ref{Eq_reciprocity_2}) to evaluate
2063: the contributions to Eq.\ (\ref{total field2}) from the integrations over the
2064: sections at $x^{\prime}\rightarrow\pm\infty$.
2065:
2066: Let us first consider $x^{\prime}\rightarrow\infty$. According to
2067: Eq.\ (\ref{Green function}) the $x^{\prime}$ dependence of the Green's
2068: function pair $\mathbf{G},\mathbf{\Gamma}$ consist of modes $\mathbf{u}%
2069: _{s}(x^{\prime})^{\ast}$ with $v_{g}^{(s)}<0$ since here $x^{\prime}>x$ for
2070: any finite $x$. On the other hand the field pair $\mathbf{u},\mathbf{T}$ are
2071: made up of the incident wave, and waves scattered from the roughness at finite
2072: $x$, and so consist of modes $\mathbf{u}_{r}(x^{\prime})$ with $v_{g}^{(r)}%
2073: >0$. The integral in Eq.\ (\ref{total field2}) over the section at $x^{\prime
2074: }\rightarrow\infty$ is therefore the sum of terms involving $\int\int\left[
2075: u_{i}^{(s)\ast}T_{ix}^{(r)}-u_{r}^{(r)}T_{ix}^{(s)\ast}\right] \,dydz$ with
2076: $v_{g}^{(r)}$ and $v_{g}^{(s)}$ of opposite sign. All these terms are zero by
2077: Eq. (\ref{Eq_reciprocity_1}), and so there is no contribution from the section
2078: at $x^{\prime}\rightarrow\infty$.
2079:
2080: Similar arguments apply to the section at $x^{\prime}\rightarrow-\infty$. The
2081: Green function is made up of modes with $v_{g}>0$. The scattered component of
2082: the field $\mathbf{u}$ consists of modes with $v_{g}<0$, and there is no
2083: contribution to the integral over the section from these modes. On the other
2084: hand the incident wave $\mathbf{u}^{\mathrm{in}}$ is mode $\mathbf{u}_{m}$
2085: with $v_{g}^{(m)}>0$, and there is the single term with $v_{g}^{(n)}%
2086: =v_{g}^{(m)}$ surviving in the sum over modes in the Green function. Using
2087: Eq.\ (\ref{Eq_reciprocity_2}) the integral just gives $u_{q}^{(m)}(x)$.
2088: Writing $\mathbf{u}=\mathbf{u}^{\mathrm{in}}+\mathbf{u}^{\mathrm{sc}}$ then
2089: leads to Eq.\ (\ref{scatterd-field-w/bd}) in the text.
2090:
2091: \section{Energy Flux for Flexural Modes}
2092:
2093: \label{App_TPP}The classical thin plate approximation of setting $T_{zi}=0$ is
2094: not sufficient to calculate the energy flux of the flexural modes using the
2095: integral Eq.\ (\ref{Power_integral_ex}). In this appendix we evaluate the
2096: correct expression for the energy flux by two different methods, first using
2097: the extended thin-plate theory of Timoshenko\cite{T} (see also Graff\cite{G}),
2098: and then using a method in terms of the energy of plate deformations\cite{LF7}
2099: that avoids these difficulties.
2100:
2101: In the extended thin plate approximation of Timoshenko the $z$-dependence of
2102: the in-plane displacements is still approximated as linear%
2103: \begin{align}
2104: u_{x}(x,y,z) & \simeq z\psi_{x}(x,y),\label{Eq_uxuy}\\
2105: u_{y}(x,y,z) & \simeq z\psi_{y}(x,y).
2106: \end{align}
2107: However, the $x,y$ dependence is no longer assumed to be given by the gradient
2108: of the mean vertical displacement $\bar{u}_{z}(x,y)$, but by the more general
2109: expression%
2110: \begin{equation}
2111: \mathbf{\psi}=-\mathbf{\nabla}_{\perp}\bar{u}_{z}+\mathbf{\nabla}_{\perp
2112: }S+\mathbf{\nabla}_{\perp}\times(\zeta\hat{z}) \label{Eq_def}%
2113: \end{equation}
2114: introducing the scalar potential $S(x,y)$ and vector potential $\zeta(x,y)$
2115: defining the corrections to the in-plane strain and rotation. Here
2116: $\mathbf{\nabla}_{\perp}=(\partial_{x},\partial_{y})$ is the horizontal
2117: gradient. In addition the vertically averaged stress $T_{zx}$ is taken to be%
2118: \begin{equation}
2119: T_{zx}\simeq\kappa^{2}\frac{E}{2\left( 1+\sigma\right) }(\partial
2120: _{x}\bar{u}_{z}+\partial_{z}u_{x}) \label{Eq_Tzx_kappa}%
2121: \end{equation}
2122: (with a similar expression for $T_{zy}$ given by replacing the subscript $x$
2123: with $y$ everywhere). Here the ``shear correction factor'' $\kappa$, a number
2124: of order unity, is introduced to take into account deviations of the in-plane
2125: displacements from the assumed linear dependence on $z$\cite{T}. In the usual
2126: thin plate approximation $T_{zi}$ are set to zero and $\psi=-\mathbf{\nabla
2127: }_{\perp}w$, so that $(u_{x},u_{y})=-z\mathbf{\nabla}_{\perp}w$.
2128:
2129: With the Timoshenko approximations, the equations of motion for the three
2130: components of displacement are now investigated.
2131:
2132: The equations of motion for the horizontal displacement lead to an equation
2133: relating $\mathbf{\psi}$ to $\bar{u}_{z}$\cite{G}%
2134: \begin{equation}
2135: \frac{D}{2}\left\{ (1-\sigma)\nabla^{2}\mathbf{\psi}+(1+\sigma)\mathbf{\nabla
2136: }_{\perp}\mathbf{\nabla}_{\perp}\cdot\mathbf{\psi}\right\} -\kappa^{2}\mu
2137: d\left( \mathbf{\psi}+\mathbf{\nabla}_{\perp}\bar{u}_{z}\right) =0
2138: \label{Psi}%
2139: \end{equation}
2140: (remember $D=Ed^{3}/12(1-\sigma^{2})$, with $E$ Young's modulus,
2141: and $\mu$ the shear modulus). The inertial terms
2142: $\partial_{t}^{2}\mathbf{\psi}$ \ turn out to be negligible in
2143: this equation. Using Eq.\ (\ref{Eq_def}),
2144: Eq.\ (\ref{Psi}) becomes%
2145: \begin{equation}
2146: D\mathbf{\nabla}_{\perp}\nabla_{\perp}^{2}(S-w)-\kappa^{2}\mu d\mathbf{\nabla
2147: }_{\perp}S+\frac{D}{2}(1-\sigma)\mathbf{\nabla}_{\perp}\times(\nabla_{\perp
2148: }^{2}\zeta\hat{z})-\kappa^{2}\mu d\mathbf{\nabla}_{\perp}\times(\zeta\hat
2149: {z})=0. \label{Eq_S_zeta}%
2150: \end{equation}
2151: Taking the vertical curl of Eq.\ (\ref{Eq_S_zeta}) gives%
2152: \begin{equation}
2153: \frac{D}{2}(1-\sigma)\nabla_{\perp}^{2}\Omega-\kappa^{2}\mu d\,\Omega=0
2154: \end{equation}
2155: with $\Omega=\hat{z}\cdot\mathbf{\nabla}_{\perp}\times\mathbf{\psi}%
2156: =-\nabla_{\perp}^{2}\zeta$ the rotation. For a wave disturbance $e^{ikx}$,
2157: this gives an exponential dependence on $y$, $e^{\pm\lambda y}$ with%
2158: \begin{equation}
2159: \lambda^{2}\simeq\frac{2\kappa^{2}\mu d}{D(1-\sigma)}\sim d^{-2}.
2160: \end{equation}
2161: Since $\lambda^{-1}\sim d\ll W$, the rotation will be large only
2162: over a boundary layer region with width of order $d$ near the
2163: edges $y=\pm W/2$,
2164: where the solution takes the form%
2165: \begin{equation}
2166: \Omega(x,y\simeq\pm W/2)\simeq\Omega(\pm W/2)e^{ikx}e^{-\lambda|y\mp W/2|}.
2167: \end{equation}
2168: The vector potential $\zeta$ has a similar solution, so that last two terms in
2169: Eq.\ (\ref{Eq_S_zeta}) cancel. This leaves for the scalar potential $S$%
2170: \begin{equation}
2171: \mathbf{\nabla}_{\perp}\left( D\nabla_{\perp}^{2}(S-\bar{u}_{z})-\kappa
2172: ^{2}\mu dS\right) =0
2173: \end{equation}
2174: which immediately gives%
2175: \begin{equation}
2176: D\nabla_{\perp}^{2}S-\kappa^{2}\mu dS=D\nabla_{\perp}^{2}\bar{u}_{z}.
2177: \label{Eq_Sw1}%
2178: \end{equation}
2179: (We are only interested in $\mathbf{\nabla}_{\perp}S$ and so do not need to
2180: keep track of the arbitrary gradient-free function that could be added to this equation.)
2181:
2182: The equation of motion for the vertical displacement is\cite{G}%
2183: \begin{equation}
2184: \kappa^{2}\mu d\nabla_{\perp}^{2}S=-\rho d\omega^{2}\bar{u}_{z}.
2185: \label{Eq_Sw2}%
2186: \end{equation}
2187: Together Eqs.\ (\ref{Eq_Sw1},\ref{Eq_Sw2}) give%
2188: \begin{equation}
2189: \rho d\omega^{2}(\bar{u}_{z}-\frac{D}{\kappa^{2}\mu d}\nabla_{\perp}%
2190: ^{2}\bar{u}_{z})=D\nabla_{\perp}^{4}\bar{u}_{z}.
2191: \end{equation}
2192: This is the usual fourth order wave equation, with a small correction term of
2193: order $(d/W)^{2}$ (the second term in the brackets on the left hand side).
2194: Note that solutions to this equation vary on the long scale of order $k^{-1}$
2195: or $W$, and not the small scale $\lambda^{-1}\sim d$, so that to a good
2196: approximation we have%
2197: \begin{subequations}
2198: \begin{align}
2199: \rho d\omega^{2}\bar{u}_{z} & =D\nabla_{\perp}^{4}\bar{u}_{z},\\
2200: S & =-\frac{D}{\kappa^{2}\mu d}\nabla_{\perp}^{2}\bar{u}_{z}.
2201: \label{Eq_S_approx}%
2202: \end{align}
2203: The first equation is now the standard fourth order wave equation.
2204: The second equation for $S$ shows it to be small compared with
2205: $\bar{u}_{z}$ by of order $(d/W)^{2}$.
2206:
2207: The boundary conditions at the edges are that all stresses are zero, so that
2208: in particular at $y=\pm W/2$%
2209: \end{subequations}
2210: \begin{equation}
2211: \int dzT_{zy}=\kappa^{2}\mu d(\partial_{y}\bar{u}_{z}+\psi_{y})=0.
2212: \end{equation}
2213: Substituting Eq.\ (\ref{Eq_def}) into this gives%
2214: \begin{subequations}
2215: \begin{equation}
2216: \partial_{y}S-\partial_{x}\zeta=0. \label{Eq_bc3}%
2217: \end{equation}
2218: Equation (\ref{Eq_bc3}) together with Eq.\ (\ref{Eq_S_approx}) tells us the
2219: size of the $\zeta$ correction, which at $y=\pm W/2$ takes the value%
2220: \end{subequations}
2221: \begin{equation}
2222: \zeta(x,y=\pm W/2)=-\frac{D}{\kappa^{2}\mu d}\frac{1}{ik}\left. \left(
2223: \partial_{y}\nabla_{\perp}^{2}\bar{u}_{z}\right) \right| _{y=\pm W/2}.
2224: \end{equation}
2225: This expression can be simplified using the boundary condition $T_{yy}=0$ at
2226: $y=\pm W/2$, which from Eq.\ (\ref{Tyy}) and Eqs.\ (\ref{Flex_ux}%
2227: ,\ref{Flex_uy}) gives at $y=\pm W/2$%
2228: \begin{equation}
2229: \partial_{y}^{2}\bar{u}_{z}=-\sigma\partial_{x}^{2}\bar{u}_{z}=\sigma
2230: k^{2}\bar{u}_{z},
2231: \end{equation}
2232: so that%
2233: \begin{equation}
2234: \zeta(x,y=\pm W/2)=-\frac{ikD(1-\sigma)}{\kappa^{2}\mu d}\left. \left(
2235: \partial_{y}\bar{u}_{z}\right) \right| _{y=\pm W/2}. \label{Eq_zeta}%
2236: \end{equation}
2237: The potential $\zeta$ is only large in the boundary layers near the edges
2238: where it takes the form%
2239: \begin{equation}
2240: \zeta(x,y\simeq\pm W/2)=-\frac{ikD(1-\sigma)}{\kappa^{2}\mu d}\left.
2241: (\partial_{y}\bar{u}_{z})\right| _{y=\pm W/2}e^{-\lambda|y\mp W/2|}
2242: \label{zeta-final}%
2243: \end{equation}
2244:
2245: Thus finally we have expressions for the horizontal displacement
2246: field, Eqs.\ (\ref{Eq_def}) and (\ref{Eq_uxuy}) together with
2247: Eqs.\ (\ref{Eq_S_approx}) and (\ref{zeta-final}) defining $S$ and
2248: $\zeta$, and Eq.\ (\ref{Eq_Tzx_kappa}). These can be used to
2249: calculate the additional contribution to the energy flux coming
2250: from the $T_{zx}$ term in Eq.\ (\ref{Power_integral_ex}). (The
2251: corrections to $u_{x}$ and $u_{y}$ derived here do not change the
2252: contributions from the first two terms in Eq.\
2253: (\ref{Power_integral_ex}) to the order we require, since these
2254: terms are already third order in the small parameter $d/W$.)
2255:
2256: We therefore need to evaluate%
2257: \begin{equation}
2258: \int\int T_{zx}u_{z}^{\ast}dydz\simeq\kappa^{2}\mu d\int dy(\partial
2259: _{x}S\,+\partial_{y}\zeta\,)\bar{u}_{z}^{\ast}. \label{z_flux}%
2260: \end{equation}
2261: Both terms in the integral give contributions at the same order. The first
2262: term, coming from the correction to the in-plane strain
2263: Eq.\ (\ref{Eq_S_approx}), is%
2264: \begin{equation}
2265: \kappa^{2}\mu d\int dy(\partial_{x}S)\,\bar{u}_{z}^{\ast}=-D\int(\partial
2266: _{x}\nabla_{\perp}^{2}\bar{u}_{z})\bar{u}_{z}^{\ast}\,dy.
2267: \end{equation}
2268: The second term in the integrand is only large in the boundary layer region
2269: near the edges and from Eq.\ (\ref{Eq_zeta}) evaluates to the edge
2270: contributions%
2271: \begin{equation}
2272: \kappa^{2}\mu d\int dy\,(\partial_{y}\zeta)\,\bar{u}_{z}^{\ast}=-ikD(1-\sigma
2273: )\left. [(\partial_{y}\bar{u}_{z})\bar{u}_{z}^{\ast}]\right| _{y=-W/2}%
2274: ^{y=W/2}.
2275: \end{equation}
2276:
2277: Combining these expressions for Eq.\ (\ref{z_flux}) with
2278: Eqs.\ (\ref{Poynting-inp}) together with (\ref{Flex_ux}) and (\ref{Flex_uy})
2279: yields the final expression%
2280: \begin{multline}
2281: P\simeq\frac{\omega kD}{2}\operatorname{Re}\left\{ \int dy\left[
2282: 2k^{2}\bar{u}_{z}\bar{u}_{z}^{\ast}+\left( 1-\sigma\right) \left(
2283: \partial_{y}\bar{u}_{z}\right) \left( \partial_{y}\bar{u}_{z}\right)
2284: ^{\ast}-\left( 1+\sigma\right) \left( \partial_{y}^{2}\bar{u}_{z}\right)
2285: \bar{u}_{z}^{\ast}\right] \right. \label{Timoshenko_power}\\
2286: \left. +\left[ \left( 1-\sigma\right) \left( \partial_{y}\bar{u}_{z}%
2287: \right) \bar{u}_{z}^{\ast}\right] _{y=-W/2}^{y=W/2}\right\} .
2288: \end{multline}
2289: which is identical to Eq.\ (\ref{Poynting-power}).
2290:
2291: An alternative approach to calculate the energy flux is to use the expression
2292: for the energy of distortions of the plate evaluated using the lowest order
2293: expressions Eqs.\ (\ref{Txx},\ref{Tyy}) and (\ref{Flex_ux},\ref{Flex_uy}%
2294: )\cite{LF7}%
2295: \begin{equation}
2296: F=\frac{1}{2}D\int\int\left[ \left( \nabla_{\perp}^{2}\bar{u}_{z}\right)
2297: ^{2}+2(1-\sigma)\left\{ \left( \frac{\partial^{2}\bar{u}_{z}}{\partial
2298: x\partial y}\right) ^{2}-\frac{\partial^{2}\bar{u}_{z}}{\partial x^{2}%
2299: }\frac{\partial^{2}\bar{u}_{z}}{\partial y^{2}}\right\} \right] dxdy.
2300: \end{equation}
2301: It turns out that the higher order corrections discussed above are not needed
2302: in this expression, and so we can derive the energy flux without these
2303: difficulties. The functional derivative of $F$ with respect to $\bar{u}_{z}$
2304: yields the vertical force per unit area in the interior of the plate, which
2305: can be used to derive the fourth order wave equation, as well as expressions
2306: for the energy flux into the plate across the boundaries. The latter
2307: expressions give us the result for the energy flux along the beam%
2308: \begin{equation}
2309: P=\frac{1}{2}\operatorname{Re}\left\{ -i\omega\left[ \int M_{x}\theta
2310: _{x}^{\ast}+V\bar{u}_{z}^{\ast}\,dy+\left. (F_{c}\bar{u}_{z}^{\ast}\right|
2311: _{y=W/2}\left. +F_{c}\bar{u}_{z}^{\ast}\right| _{y=-W/2})\right] \right\}
2312: \label{flux-CL}%
2313: \end{equation}
2314: where%
2315: \begin{equation}
2316: V=-D\partial_{x}[\partial_{x}^{2}\bar{u}_{z}+\left( 2-\sigma\right)
2317: \partial_{y}^{2}\bar{u}_{z}] \label{Eq_V}%
2318: \end{equation}
2319: is the effective vertical force that couples to the vertical displacement
2320: $\bar{u}_{z}$,%
2321: \begin{equation}
2322: M_{x}=-D\left( \partial_{x}^{2}\bar{u}_{z}+\sigma\partial_{y}^{2}%
2323: \bar{u}_{z}\right) \label{Eq_Mx}%
2324: \end{equation}
2325: is the torque that couples to the angular displacement $\theta_{x}%
2326: =\partial\bar{u}_{z}/\partial x$, and%
2327: \begin{equation}
2328: F_{c}(y=\pm W/2)=\pm\left. 2D(1-\sigma)\partial_{xy}^{2}\bar{u}_{z}\right|
2329: _{y=\pm W/2} \label{Eq_Fc}%
2330: \end{equation}
2331: is a vertical force localized at the edges of the plate.
2332:
2333: Substituting Eqs.\ (\ref{Eq_V}-\ref{Eq_Fc}) into Eq.\ (\ref{flux-CL}) gives
2334: \begin{multline}
2335: P=\frac{1}{2}\operatorname{Re}\left\{ \frac{\left( i\omega D\right) }%
2336: {2}\left[ \int dy\left( \partial_{x}^{2}\bar{u}_{z}+\sigma\partial_{y}%
2337: ^{2}\bar{u}_{z}\right) \left( -\partial_{x}\bar{u}_{z}\right) ^{\ast}+\int
2338: dy[\partial_{x}^{3}\bar{u}_{z}+\left( 2-\sigma\right) \partial_{x}%
2339: \partial_{y}^{2}\bar{u}_{z}]\bar{u}_{z}^{\ast}\right. \right.
2340: \label{free-energy-power}\\
2341: \left. \left. -\left. 2\left( 1-\sigma\right) \left( \partial
2342: _{x}\partial_{y}\bar{u}_{z}\right) \bar{u}_{z}^{\ast}\right| _{y=W/2}%
2343: +\left. 2\left( 1-\sigma\right) \left( \partial_{x}\partial_{y}%
2344: \bar{u}_{z}\right) \bar{u}_{z}^{\ast}\right| _{y=-W/2}\right] \right\}
2345: \end{multline}
2346: Evaluating $\partial_{x}=ik$, and using integration by parts, we again get
2347: Eq.\ (\ref{Poynting-power}).
2348:
2349: \begin{thebibliography}{9} %
2350:
2351: \bibitem {L57}R.~Landauer, IBM~J.~Res.~\textbf{1}, 223 (1957).
2352:
2353: %\bibitem {WTNPAFHPRJ88}D.A.\ Wharam, T.J.\ Thornton, R.\ Newbury, M.\ Pepper,
2354: %H.\ Ahmed, J.E.F.\ Frost, D.G.\ Hasko, D.C.\ Peacock, D.A.\ Ritchie,
2355: %G.A.C.\ Jones, J.\ Phys.\ C, \textbf{21}, L209 (1988).
2356: %
2357: %\bibitem {WHBWKMF88}B.J.\ van Wees, H.\ van Houten, C.W.J.\ Beenakker,
2358: %J.G.\ Williamson, L.P.\ Kouwenhoven, D.\ van der Marel, and C.T.\ Foxon,
2359: %Phys.\ Rev.\ Lett.\ \textbf{60}, 848 (1988).
2360:
2361: \bibitem {ACR98}D.~E.~Angelescu, M.~C.~Cross, and M.~L.~Roukes, Superlattices
2362: and Microstructures 23, 673 (1998).
2363:
2364: \bibitem {RK98}L.~G.~C.~Rego and G.~Kirczenow, Phys.~Rev.~Lett.~\textbf{81},
2365: 232 (1998).
2366:
2367: \bibitem {B99}M.~P.~Blencowe, Phys.~Rev.~B \textbf{59}, 4992 (1999).
2368:
2369: \bibitem {BV00}M.~P.~Blencowe and V.\ Vitalli, Phys.~Rev.~A \textbf{62}, 52104 (2000).
2370:
2371: \bibitem {P83}J.~B.~Pendry, J.~Phys.~A \textbf{16},2161 (1983).
2372:
2373: \bibitem {RK99}L.~G.~C.~Rego and G.~Kirczenow, Phys.~Rev.~B \textbf{59} (20)
2374: 13080 (1999).
2375:
2376: \bibitem {SHWR00}K.~Schwab, E.~A.~Henriksen, J.~M.~Worlock, M.~L.~Roukes,
2377: Nature \textbf{404}, 974 (2000).
2378:
2379: %\bibitem {UF72}E.I.\ Urazakov and L.A.\ Falkovskii, JETP \textbf{36} (6), 1214 (1972).
2380: %
2381: %\bibitem {TC97}V.I.\ Tatarskii, M.\ Charnotskii, Waves in Random Media
2382: %\textbf{8}, 29 (1998).
2383: %
2384: %\bibitem {FTB99}I.M.\ Fukes, V.I.\ Tatarskii, and D.E.\ Barrick, Waves in
2385: %Random Media \textbf{8}, 29 (1998).
2386: %
2387: %\bibitem {FMP95}R.M.\ Fitzgerald, A.A.\ Maradudin, F.\ Pincemin, Waves in
2388: %Random Media, \textbf{5}, 381 (1995).
2389:
2390: \bibitem {S97}M.~Spivack, J.\ Acoust.\ Soc.\ Am.\ \textbf{101} (3), 1250 (1997).
2391:
2392: \bibitem {KFFL99}A.~Kambili, G.~Fagas, V.~I.~Fal'ko, and C.~J.~Lambert,
2393: Phys.\ Rev.\ B \textbf{60}, 15593 (1999).
2394:
2395: \bibitem {SFMY99}J.A. Sanchez-Gil, V. Freilikher, A.A. Maradudin, I.V.
2396: Yurkevich, Phys. Rev. B \textbf{59}, 5915 (1999).
2397:
2398: \bibitem {SC00}D.~H.~Santamore and M.~C.~Cross, Phys.\ Rev.\ B \textbf{63}, 4306 (2001).
2399:
2400: \bibitem {B95}M.~P.~Blencowe, J.\ of Phys.\ Cond.\ Mat.\ \textbf{7} (27) 5177 (1995).
2401:
2402: %\bibitem {BGKP}S.V. Biryukov, Yu.V. Gulyaev, V.V. Krylov, V.P. Plessky,
2403: %\textit{Surface Acoustic Waves in Inhomogeneous Media}, Springer, Berlin, (1995).
2404: %
2405: %\bibitem {YKSIB94}S.\ Yu, K.W.\ Kim, M.A.\ Stroscio, G.J.\ Iafrate, and
2406: %A.\ Ballato, Phys.~Rev.~B \textbf{50} (3) 1733 (1994).
2407:
2408: \bibitem {SC01l}D.~H.~Santamore and M.~C.~Cross, Phys.\ Rev.\ Lett.\ \textbf{87},
2409: 115502 (2001).
2410:
2411: \bibitem {A}B.~A.~ Auld, Acoustic Fields and Waves in Solids Vol.\ 2, Robert
2412: E.~Krieger Publishing, Malabar, 1990.
2413:
2414: \bibitem {P98}Y.~Pomeau, Phil.\ Mag.\ B, \textbf{78} (2), 235, 1998.
2415:
2416: \bibitem {Ssp}K.~Schwab (private communication)
2417:
2418: \bibitem {CL00}M.~C.~Cross and R.~Lifshitz, Phys.\ Rev.\ B \textbf{64}, 085324 (2001).
2419:
2420: \bibitem {LF7}L.~D.~Landau and E.~M.~Lifshitz, \textit{Theory of Elasticity},
2421: Butterworth-Heinemann, Oxford, (1986).
2422:
2423: \bibitem {NAW97}N.~Nishiguchi, Y.~Ando, M.~N.~Wybourne, J.\ Phys.\
2424: Condens.\ Matter \textbf{9}, 5751, (1997).
2425:
2426: \bibitem {T}Timoshenko, \textit{Theory of Elastic Stability}, McGraw-Hill, New
2427: York, (1991).
2428:
2429: \bibitem {G}K.~Graff, \textit{Wave Motion in Elastic Solids}, Dover, New York, (1991).
2430: \end{thebibliography}
2431: \end{document}
2432: