Wave localization in binary isotopically disordered one-dimensional harmonic chains with impurities having arbitrary cross section and concentration
Extracted source text ยท arXiv cond-mat/0403701
\documentclass[prb,twocolumn,showpacs,amsmath,amssymb]{revtex4}
\usepackage{graphicx,bm}
\newcommand{\E}{\mbox{e}}
\newcommand{\I}{\mbox{i}}
\newcommand{\SpringConstant}{\ensuremath{K}}
\newcommand{\EquilMass}{\ensuremath{m_{\circ}}}
\newcommand{\ImpurityMass}{\ensuremath{m_+}}
\newcommand{\UnitLength}{\ensuremath{a}}
\newcommand{\MassDensity}{\ensuremath{\mu}}
\newcommand{\YoungsModulus}{\ensuremath{E}}
\newcommand{\CrossSection}{\ensuremath{\sigma}}
\newcommand{\WaveNumber}{\ensuremath{k}}
\newcommand{\PulseSpeed}{\ensuremath{c_s}}
\newcommand{\Wavelength}{\lambda}
\newcommand{\Resistivity}{\rho}
\newcommand{\LocLength}{\xi}
\newcommand{\LocLengthDL}{\LocLength_\circ} % dilute limit
\newcommand{\LExp}{\gamma} % Lyapunov exponent
\newcommand{\MFP}{\Lambda} % mean free path
\newcommand{\VEC}[1]{{\bf #1}}
\newcommand{\SD}[1]{{s_{#1}}} % standard deviation
\newcommand{\VAR}[1]{{s^2_{#1}}} % variance
\newcommand{\COV}[1]{{\eta_{#1}}} % coefficient of variation
\newcommand{\CF}{\ensuremath{k}} % coverage factor
\newcommand{\Moment}{\Delta} % moment about the mean
% Roman Numerals
\newcommand{\RNone}{\uppercase\expandafter{\romannumeral1}}
\newcommand{\RNtwo}{\uppercase\expandafter{\romannumeral2}}
\newcommand{\RNthree}{\uppercase\expandafter{\romannumeral3}}
% time derivative
\newcommand{\dt}[1]{\dot{#1}}
\newcommand{\ddt}[1]{\ddot{#1}}
% FFT
\newcommand{\FT}[1]{\stackrel{~}{#1}}
\bibliographystyle{apsrev}
\begin{document}
\title{%
Wave localization in binary isotopically disordered one-dimensional harmonic chains
with impurities having arbitrary cross section and concentration
}
\author{K.A.~Snyder}
\affiliation{Materials and Construction Division,
National Institute of Standards and Technology}
\author{T.R.~Kirkpatrick}
\affiliation{Institute for Physical Science and Technology and
Department of Physics, University of Maryland}
\date{\today}
\pacs{%
% 43.20.+g, % general linear acoustics
45.05.+x, % general theory classical mech. of discrete systems
45.30.+s, % general linear dynamical systems
% 45.50.Jf, % Few- and many-body systems
46.40.Cd, % mechanical wave propagation
% 61.43.-j, % disordered solids
% 61.43.Bn, % structural modeling
62.40.+d % mechanical and elastic waves
% 62.20.Pw, % localized modes
% 63.50.+x % vibrational states in disordered systems
}
\begin{abstract}
The localization length for isotopically disordered harmonic one-dimensional
chains is calculated for arbitrary impurity
concentration and scattering cross section.
%%
The localization length depends on the
scattering cross section of a single scatterer, which is calculated for a
discrete chain having a wavelength dependent pulse propagation
speed.
%
% The cross section is shown to be symmetric about zero with respect to
% the mass either added to or subtracted from the background mass.
%
%
For binary isotopically disordered systems composed of many scatterers,
the localization length decreases with increasing impurity concentration,
reaching a mimimum before diverging toward infinity as the impurity
concentration approaches a value of one.
%
The concentration dependence of the localization length over the entire
impurity concentration range is
approximated accurately by the sum of the behavior at each limiting
concentration.
%
Simultaneous
measurements of Lyapunov exponent statistics indicate practical
limits for the
minimum system length and the number of scatterers to achieve representative
ensemble averages.
%
Results are discussed in the context of
future investigations of the time-dependent behavior of disordered
anharmonic chains.
%
% These results are applicable to polymeric systems and spectroscopy, and are
% necessary for understanding numerical experiments on
% discrete anharmonic systems that must exploit short wavelengths and
% strong scattering to be practical.
%
\end{abstract}
\maketitle
%
% INTRODUCTION
%
\section{Introduction}
The length scale over which Anderson localization \cite{Anderson58} occurs
in harmonic disordered chains can be expressed as a function of the
ensemble-averaged system resistivity
\cite{Landauer70,Anderson80}.
%
Using scaling arguments, \cite{Anderson80} the
ensemble-averaged resistivity of harmonic systems
having binary isotopic
disorder (single-valued impurities)
can be expressed as a function of the scattering
cross section of a single impurity and the impurity
concentration.
%
Binary isotopically disodered harmonic chains also have
the property that in the limit that the impurity
concentration approaches unity, the system is once
again ``pure,'' and the Anderson localization length
diverges toward infinity.
%
As a result, the Anderson localization length passes through a
minimum at intermediate concentrations \cite{Sheng86b}.
%
These properties of a binary isotopically disordered harmonic chain
(BIDHC) make these systems both tractable and interesting.
In this work, an approximation is developed for the localization
length of a BIDHC with arbitrary impurity cross section and
concentration.
%
In addition, the result incorporates the discrete nature of a
classical chain.
%
Similar studies have been performed previously, but with
important differences.
%
Bourbonnais and Maynard \cite{Bourbonnais90a} studied
energy transport in one-dimensional
systems having isotopic disorder, but the impurity masses
were not single-valued.
%
The results of Azbel and Soven \cite{Azbel82} on
quantum systems having binary isotopic disorder
were based on a continuum solution for impurity
locations constrained to exist on lattice sites.
%
Although the Azbel and Soven result
applied to short wavelengths that may exist between
the impurities, the results do not incorporate the additional
features of a discrete mechanical system.
%
This study serves as an introduction to future work on energy
transport in binary isotopically disordered anharmonic chains.
%
These anharmonic chains serve as a useful model for, among
other things, line width broadening
\cite{Held97,Rohmfeld01,Widulle02}.
%
To perform that work,
numerical integration will used to study the time-dependent
nature of these systems.
%
To be practical, the initial conditions will require sufficiently short
wavelengths and high impurity concentrations to keep integration
times manageable.
%
In addition, the results of the study will incorporate the discrete
nature of the mechanic chains so that these effects can
accounted for in the results.
In this study, the energy localization in a BIDHC
is studied for arbitrary displacement wavelength, impurity concentration and
scattering cross section.
%
Disorder is effected by changing randomly selected masses by a fixed amount.
%
A continuum Kronig-Penney (KP) model \cite{Kronig31}
is used to develop a general expression
for scattering cross section, and the continuum impurity impedance is
corrected for wavelength dependent pulse propagation speed in discrete
systems.
%
The resulting expression is verified by direct numerical integration.
%
The localization length of systems with strong scatterers is calculated using
both the continuum KP model and the MacKinnon and Kramer \cite{MacKinnon81}
(MK) method.
%
The distribution of Lyapunov exponents is studied using the continuum
KP model, and the minimum requirements are found for system length and
number of scatterers to achieve proper scaling statistics.
%
The localization length concentration dependence is studied using the MK method,
and an analytical expression is found for its behavior.
%
For arbitrary impurity concentration and impurity cross section,
the localization length in a BIDHC
can be approximated by invoking a simple {\em ansatz} based on an
analogy to electrical systems.
%
The result is accurate for systems having displacement wavelengths
at least four lattice spacings long.
%
%
% MODEL SYSTEM
%
\section{Model System}
The physical model used here is the harmonic version of the
Fermi-Pasta-Ulam (FPU) \cite{Fermi55} chain that is composed of discrete springs
and masses.
The masses $m_i$ are on a lattice with spacing $\UnitLength$ and interact via nearest
neighbor springs with force constant $K$.
Disorder is effected by changing the background mass $\EquilMass$ by a fixed amount
$\ImpurityMass$ with probability $c$.
To simplify the results, all lengths are scaled by the lattice spacing $\UnitLength$.
For a system composed of $N$ masses, each characterized by a displacement
$x_i$ about the equilibrium location and a momentum $p_i$, the Hamiltonian
is separable:
\begin{equation}
H = \frac{1}{2}\sum_i^N \frac{p_i^2}{m_i}
+ \SpringConstant\,\left(x_{i+1}-x_i\right)^2
\label{eqn:Hamiltonian}
\end{equation}
The real space equation of motion is
\begin{equation}
\frac{m_i}{K}\,\ddot{x}_i = x_{i+1} - 2x_i + x_{i-1}
\label{eqn:xeom}
\end{equation}
The Fourier transform leads to the corresponding equation of motion
for the energy eigenstate amplitudes $u_i(\omega)$:
\begin{equation}
u_{i+1} =
\left[2 - \frac{\omega^2 m_i}{\SpringConstant}\right]\,
u_i - u_{i-1}
\label{eqn:ueom}
\end{equation}
This is the corresponding Anderson tight-binding model for the
chain.
\subsection{Discrete Analysis}
The time-dependent properties of the system were determined by numerical
time integration of Eqn.~\ref{eqn:xeom}
using a fourth-order symplectic integrator algorithm (SIA4)
for separable Hamiltonians.
The coefficients were taken from Candy and Rozmus \cite{Candy91}, and
the time step was 1/200 of the natural period.
% $U$ and
% kinetic $T$ energies expressed in canonical coordinate $\VEC{q}$ and
% momentum $\VEC{p}$ variables:
% \begin{equation}
% H(\VEC{q},\VEC{p}) = U(\VEC{q}) + T(\VEC{p})
% \end{equation}
% The force $\VEC{F}$ and conjugate velocity $\VEC{V}$ are also separable:
% \begin{equation}
% \VEC{F}(\VEC{q}) = -\nabla_q U(\VEC{q})
% \hspace{1.0cm}
% \VEC{V}(\VEC{p}) = +\nabla_p T(\VEC{p})
% \end{equation}
% Starting from a state ($\VEC{q}^n_\circ,\VEC{p}^n_\circ$),
% the momenta and coordinates at time $(n+1)\Delta t$ are transformed by
% iterating the
% following expressions over the $1\leq m\leq 4$ index:
% \begin{eqnarray}
% \VEC{p}_m^n & = & \VEC{p}_{m-1}^{n} + b_m\, \VEC{F}(\VEC{q}_{m-1}^n)\,\Delta t
% \nonumber \\
% \VEC{q}_m^n & = & \VEC{q}_{m-1}^n + a_m\, \VEC{V}(\VEC{p}_{m}^n)\, \Delta t
% \end{eqnarray}
% The momenta and coordinates at $(n+1)\Delta t$ are equivalent to the last values
% of the previous calculations:
% \begin{equation}
% \VEC{p}^{n+1}_\circ = \VEC{p}^n_4 \\
% \hspace{1.5cm}
% \VEC{q}^{n+1}_\circ = \VEC{q}^n_4
% \end{equation}
The localization length was calculated from Eqn.~\ref{eqn:ueom} using the method
of MacKinnon and Kramer (MK) \cite{MacKinnon81}. This method exploits the
statistical properties of the $u_i$ so that periodic rescaling can be used to improve
overall statistics.
%
% CONTINUUM ANALYSIS
%
\subsection{Continuum Analysis}
A Kronig-Penney model \cite{Kronig31} is used to develop an expression for
the scattering cross section of an impurity and to study the statistics of the
scaling parameter \cite{Anderson80}. The continuum system analogous to the
discrete chain is a homogeneous elastic rod having mass density
$\MassDensity$ and Youngs modulus $\YoungsModulus$.
In the absence of impurities, a longitudinal displacement amplitude
$\psi(x,t|\omega)$ with angular velocity $\omega$ will propagate down the rod
with longitudinal velocity $c_l = \sqrt{\YoungsModulus/\MassDensity}$.
A harmonic oscillator impurity located at $x^\prime$ will give rise to a
reactive force due to the impurity impedance $Z$ for a wave with
angular velocity $\omega$:
\begin{equation}
\left[
\MassDensity\frac{\partial^2}{\partial t^2}
- \YoungsModulus \frac{\partial^2}{\partial x^2} =
-Z(\omega)\,\delta(x-x^\prime) \frac{\partial}{\partial t}
\right]\,\psi(x,t|\omega)
\label{eqn:continuum}
\end{equation}
We assume the solution has a time dependence given by an exponential of
angular velocity $\omega$ ($\psi=\phi(x|\omega)\,\E^{-\I\omega t}$)
to obtain
\begin{equation}
\left[\frac{\partial^2}{\partial x^2} + \kappa^2
= \frac{-\I\omega}{\YoungsModulus}\,
Z(\omega)\,\delta(x-x^\prime)
\right] \phi(x|\omega)
\label{eqn:Helmholtz}
\end{equation}
where $\kappa=\omega/c_l$.
Equation~\ref{eqn:Helmholtz} can be solved analytically
for a single scatterer or can be solved numerically for a particular
system composed of a number of
scatterers.
From this numerical solution, one can determine the system resistivity, which can
be used to calculate the localization length.
%
% CROSS SECTION
%
\section{Cross Section}
The scattering cross section $\CrossSection$ for a single impurity can be calculated from
the continuum system of Eqn.~\ref{eqn:Helmholtz}.
The Green's function for the 1-D Helmholtz equation \cite{Arfken70} can be used to solve for the admittance $\phi(x)$:
\begin{equation}
\phi(x) = \E^{\I \kappa x} -
\left(\frac{Z(\omega)}{Z(\omega)+ \sqrt{4\MassDensity\YoungsModulus}}
\right)\,
\E^{\I \kappa |x-x_i|}\,
\E^{\I \kappa x_i}
\end{equation}
The scattering cross section in
1-D is equivalent to the reflection probability:
\begin{equation}
\CrossSection = \frac{|Z(\omega)|^2}{|Z(\omega)|^2
+ 4\MassDensity\YoungsModulus}
\end{equation}
To apply this equation to the discrete chain, it must be converted from a continuum
description to the corresponding discrete description.
For an FPU system having masses spaced a distance $a$ apart, the
continuum coefficients can be expressed as a function of the discrete
properties in the limit $a\rightarrow 0$ \cite{Goldstein58}:
\begin{equation}
\MassDensity = \frac{\EquilMass}{a}
\hspace{1.0in}
\YoungsModulus = \SpringConstant a
\end{equation}
The scattering cross section can now be expressed as a function of the
discrete system components:
\begin{equation}
\CrossSection = \frac{|Z(\omega)|^2}{|Z(\omega)|^2
+ 4\SpringConstant\EquilMass}
\label{sigma:eqn}
\end{equation}
The impedance $Z(\omega)$
of a mass impurity along a one-dimensional chain
is proportional to the mass $\ImpurityMass$
that is added to the background mass $\EquilMass$
\cite{Morse68}:
\begin{equation}
Z(\omega) = -\I\,\omega \ImpurityMass / \PulseSpeed
\label{Z:eqn}
\end{equation}
This equation has been modified from the continuum relation to account
for the properties of a discrete one-dimensional chain. A displacement
wavelength has a corresponding wave number
$\WaveNumber = 2\pi/\Wavelength$. For the discrete chain, the
angular velocity $\omega$ and relative pulse propagation
speed $\PulseSpeed$ can be expressed as a function of the
displacement wave number $\WaveNumber$ \cite{Kittel86}:
\begin{equation}
\omega = 2\sin(\WaveNumber/2)
\hspace{2cm}
\PulseSpeed = \cos(\WaveNumber/2)
\label{c_s:eqn}
\end{equation}
Finally, substituting Eqn.~\ref{Z:eqn} into
Eqn.~\ref{sigma:eqn} gives the scattering cross section of a
mass impurity in a one-dimensional chain:
\begin{equation}
\CrossSection = \frac{(\ImpurityMass\, \omega / \PulseSpeed)^2}%
{(\ImpurityMass\, \omega / \PulseSpeed)^2 + 4\SpringConstant\EquilMass}
\label{cs:eqn}
\end{equation}
\begin{figure}
\includegraphics[angle=0,width=\columnwidth]{fig1}
\caption{Schematic of cross section numerical experiment;
each line represents the state of the system at the time of
measurement.
The time $t$, shown along left side, is expressed as a function
of the system length $L$ and pulse propagation speed $\PulseSpeed$.
The large dots denote the location of two impurities.}
\label{cs_schem:fig}
\end{figure}
A numerical experiment was performed to test the applicability of
Eqn.~\ref{cs:eqn} to FPU systems, and a schematic of the experiment
is shown in Fig.~\ref{cs_schem:fig}. The system had fixed ends and length
$L$. One impurity was located at $L/3$, and another impurity was
located at $2L/3$.
An initial displacement was made with wavelength $\Wavelength$ and total
length 16$\Wavelength$, and the hyperbolic tangent function
was used to taper the oscillation amplitude from $A$ to zero. Initially, the
pulse was located at one end of the system. It had length 8$\Wavelength$,
amplitude 2$A$, and zero initial velocity. The time-dependent behavior was
determined by numerical integration using the SIA4 algorithm.
The measured scattering
cross section was determined from the energy located in the
three regions separated by the impurities.
The pulse energy located in each three regions, denoted by
$E_1$, $E_2$, and $E_3$, was calculated at
four separate times: $t=0$, $L/6 \PulseSpeed$, $L/2 \PulseSpeed$, and
$5L/6 \PulseSpeed$.
At these times, the pulses are located near the center of a
region. Although redundant with respect to the initial total energy,
determining the energy at $t=L/6 \PulseSpeed$
provided a consistency check. In each case, the difference between
$E_1(t=0)$ and $E_1(t=L/6\PulseSpeed)$ was less than one part in 10$^4$.
The scattering cross section in FPU systems is the reflection coefficient.
In this experiment,
the reflection coefficient $R$ is calculated from the ratio of energies in the first
two intervals after one scattering event:
\begin{equation}
R = \frac{E_1(L/2\PulseSpeed)}{E_1(0)}
\label{R:eqn}
\end{equation}
Although the transmission coefficient $T$ could have been determined from
$E_2(L/2\PulseSpeed)$, a second scatterer was used as a more rigorous test
of the experiment design and numerical integrator.
The transmission coefficient $T$ for a single scatterer was
calculated using the energy after two scattering events:
\begin{equation}
T = \sqrt{\frac{E_3(5L/6\PulseSpeed)}{E_1(0)}}
\end{equation}
In all cases, the magnitude of $1-R-T$ was less than 10$^{-3}$.
\begin{figure}
\includegraphics[width=\columnwidth]{fig2}
\caption{Cross section $\CrossSection$ as a function of
$\ImpurityMass\omega/\PulseSpeed$ for different wavelengths $\Wavelength$.
The solid curve is the analytical result in
Eqn.~\protect{\ref{cs:eqn}}.
The inset shows additional data near zero.}
\label{cs:fig}
\end{figure}
A comparison of the estimated cross section $\CrossSection$
in Eqn.~\ref{cs:eqn}
to the measured reflection coefficient $R$ in Eqn.~\ref{R:eqn} is
shown in Fig.~\ref{cs:fig} for different displacement wavelengths
$\Wavelength$. The results demonstrate that Eqn.~\ref{cs:eqn} is
an accurate estimate for the scattering cross section $\CrossSection$
for displacement wavelengths as short as 4.
Moreover, the symmetry about zero for scattering cross section for negative
values of $\ImpurityMass$ is shown in the inset of Fig.~\ref{cs:fig}.
%
% LOCALIZATION LENGTH
%
\section{Localization Length}
The resistivity $\Resistivity_N$ of a system
composed of $N$ scatterers is
the ratio of the system reflection coefficient $R_N$ to the system
transmission coefficient $T_N$
\cite{Landauer70,Anderson80}:
\begin{equation}
\Resistivity_N = \frac{R_N}{T_N}
\label{eqn:Resistivity}
\end{equation}
The localization length of the system is defined in an averaged sense. To achieve proper scaling,
however, one defines its inverse, the Lyapunov exponent
$\LExp$:
\begin{equation}
\LExp = \frac{\ln\left(1+\Resistivity_N\right)}{L}
\end{equation}
More specifically, there is a distribution of values from an ensemble of systems. If the systems
composing the ensemble are sufficiently large, the distribution of $\LExp$ values will be normal.
The localization length $\LocLength$ for a system having length $L$ and $N$ scatterers is
defined from the ensemble averaged Lyapunov exponent:
\begin{equation}
\LocLength^{-1} = \langle\LExp\rangle
= \frac{\langle\ln\left(1+\Resistivity_N\right)\rangle}{L}
\label{LocLength:eqn}
\end{equation}
Unless otherwise noted, the symbol $\LExp$ shall imply the ensemble averaged quantity.
%
% STRONG SCATTERERS
%
\subsection{Strong Scatterers}
To perform numerical experiments on anharmonic systems of manageable length, the scatters
will need to be relatively strong. Therefore, weak scattering results will not be applicable.
Moreover, a means is needed to predict the averaged localization behavior of a system
using only single scatterer information.
Because the impurities are identical, the scaling law \cite{Anderson80} can be
exploited to express system behavior as a function of
the resistivity of a single scatterer $\Resistivity$:
\begin{equation}
\langle\ln(1+\Resistivity_N)\rangle = N\ln(1+\Resistivity)
\label{Nln:eqn}
\end{equation}
Substituting from Eqn.~\ref{LocLength:eqn} above yields an unbiased estimate for the
ensemble averaged localization length (dilute limite) as a function of impurity concentration $c$:
\begin{equation}
\LocLengthDL^{-1} = c \langle\ln(1+\Resistivity)\rangle
\label{eqn:LocLength}
\end{equation}
In the limit of weak scattering ($c,\Resistivity\rightarrow 0$),
one recovers the expected result
$\LocLengthDL^{-1} = c\CrossSection = \MFP^{-1}$, where $\MFP$ is the
classical mean free path.
\begin{figure}
\includegraphics[width=\columnwidth]{fig3}
\caption{Localization length $\LocLengthDL$, scaled by impurity concentration $c$, as a
function of impurity scattering cross section $\CrossSection$ for systems having length
$L$ and $N$ impurities.
The solid line is $\ln(1+\Resistivity)$ and the dashed line represents the mean free
path estimate.
The error bars represent the mean standard deviation.
(Many of the symbols lie upon one another and the error bars are
typically smaller than the symbols.)}
\label{xiN:fig}
\end{figure}
To demonstrate both the effect of strong scatterers ($\CrossSection\rightarrow 1$)
and the accuracy of Eqn.~\ref{Nln:eqn},
the ensemble average $\langle\ln(1+\Resistivity_N)\rangle$ was calculated
from 10~000 systems, each having length $L$ and $N$ impurities. The
displacement wavelength
was 32, and the scattering cross section of each impurity was $\CrossSection$.
The results from the calculation are shown in Fig.~\ref{xiN:fig} as a function of the
impurity cross section $\CrossSection$.
In the figure, the error bars represent the mean standard deviation, and many of the
symbols lie upon one another.
As can be seen, all the systems have the $\ln(1+\Resistivity)$ dependence that
deviates from the mean free path approximation ($\LocLengthDL^{-1}=c\CrossSection$)
for cross sections greater than approximately 0.2 for all combinations of system size and
number of scatterers.
%
% STATISTICS
%
\subsection{Statistics}
The results shown in Fig.~\ref{xiN:fig} demonstrate that, in the mean,
systems having a finite density
of scatterers have the expected behavior. Recent results
suggest that systems
having relatively few scatterers do not exhibit Gaussian behavior and, therefore, not
obey scaling laws.
For systems having sufficient length and number of scatterers,
the population of Lyapunov exponents is normally
distributed \cite{Anderson80}, with variance
$\VAR{\LExp}$\,\, \cite{Anderson80,Deych01}
\begin{equation}
\VAR{\LExp} = \frac{2}{L^2}\
\langle\ln(1+\Resistivity_N)\rangle
\label{variance:eqn}
\end{equation}
Returning to the data of Fig.~\ref{xiN:fig}, the population of Lyapunov exponents $\LExp$
was compared to the expectations of Eqn.~\ref{variance:eqn}. To assess the ``normality''
of the data, the intervals, both above and below the mean, having coverage factors
\cite{BIPM}
corresponding to one and two standard deviations were determined from the population.
In addition, the population standard deviation $\Moment$ so that it could be compared
to both its estimated value in Eqn.~\ref{variance:eqn} and to the corresponding coverage interval.
\begin{figure}
\includegraphics[width=\columnwidth]{fig4}
\caption{Lyapunov exponent $\LExp$ statistics as a function of
scattering cross section $\CrossSection$ for a system
with $L=16~384$, $N=128$, and $\Wavelength=32$.
Error bars represent coverage factors corresponding to one and two
standard deviations of a normal distribution.
The predicted intervals $\pm 1\SD{\LExp}$ and $\pm 2\SD{\LExp}$
(Eqn.~\protect{\ref{variance:eqn}})
are shown, along with observed population standard deviation $\Moment$.}
\label{stat:fig}
\end{figure}
The results of the Lyapunov exponent statistics calculation,
from an ensemble with a population of 10~000, are shown in Fig.~\ref{stat:fig} for
systems having length $L=16~384$, $N=128$ impurities, and displacement
wavelength $\Wavelength=32$. (Results for other systems were similar, and are
omitted for brevity.) The data are shown as a function of the single impurity
cross section $\CrossSection$.
The filled symbols are the average value, and the solid line is the estimated average value.
The error bars represent the intervals that have the same coverage factor as one and
two standard deviations in a normal distribution. The two pairs of dashed lines are the estimated standard deviations from Eqn.~\ref{variance:eqn}.
The dotted line is the population standard deviation $\Moment$.
For small scattering cross sections, the distribution of $\LExp$ is asymmetric, with zero as a
lower bound for the coverage intervals. Interestingly, only the outer intervals are asymmetric
about the mean. The inner intervals are nearly symmetric about the mean, and they
have a value nearly equal to the population standard deviation.
As the scattering cross section increases, the results begin to exhibit Gaussian
behavior. Above a 0.2 scattering cross section, the measured intervals, the predicted intervals
$\SD{\LExp}$, and the population standard deviation $\Moment$ all agree. Therefore,
agreement between the population standard deviation $\Moment$ and the estimated standard deviation $\SD{\LExp}$ is as much as measure
of ``normality'' as is a careful analysis of the population coverage intervals.
\begin{figure}
\includegraphics[width=\columnwidth]{fig5}
\caption{Lyapunov exponent coefficient of variation $\Moment/\LExp$ as a
function of the ratio of system length $L$ to the localization
length $\LocLengthDL$ for systems having different number of
scatterers $N$. The solid curve is Eqn.~\protect{\ref{cov:eqn}}.}
\label{cov:fig}
\end{figure}
Exploiting this relationship, the population standard deviation $\Moment$ for
all the data from the Lyapunov exponent statistics experiment
are shown in Fig.~\ref{cov:fig} as a function of the system length. For normally distributed
populations, the estimated coefficient of variation can be determined from
Eqn.~\ref{LocLength:eqn} and Eqn.~\ref{variance:eqn}, and has the following convenient form:
\begin{equation}
\frac{\SD{\LExp}}{\langle\LExp\rangle}
= \sqrt{\frac{2}{L/\LocLengthDL}}
\label{cov:eqn}
\end{equation}
This equation is shown as a solid line in Fig.~\ref{cov:fig}.
One can conclude from the figure that in order for the observed Lyapunov exponents to be
normally distributed, the system length must be at least 10 times the localization length, and
the number of scatterers must be greater than approximately 32.
%
% CONCENTRATED IMPURITIES
%
\section{Concentrated Impurities}
For dilute impurity concentrations, the localization length decreases with
increasing impurity concentration.
As the impurity concentration $c$ approaches unity, however,
the FPU system will become a pure system composed entirely of masses
$\EquilMass + \ImpurityMass$. At $c=1$, the system is once again devoid of
impurity and the localization length goes to infinity.
At intermediate impurity concentrations, the localization length passes
through a minimum. Therefore, the behavior of a system at arbitrary impurity
concentration cannot be fully characterized by the relation in Eqn.~\ref{eqn:LocLength}.
For dilute systems, the localization length $\LocLength_{c\rightarrow 0}$ is as before:
\begin{equation}
\LocLength_{c\rightarrow 0}^{-1}
= \LocLengthDL^{-1} = c \, \langle\ln (1+\Resistivity)\rangle
\end{equation}
At high concentrations, the localization length $\LocLength_{c\rightarrow 1}$ has a
concentration dependence that is proportional to $(1-c)$:
\begin{equation}
\LocLength_{c\rightarrow 1}^{-1} = (1-c)\langle\ln (1+\Resistivity^\prime)\rangle
\end{equation}
The adjusted resistivity $\Resistivity^\prime$ is for a system having
background mass $\EquilMass+\ImpurityMass$ and impurities with
mass $\EquilMass$.
By the nature of the solution using the MK method, and given that the systems
are harmonic, the frequency $\omega$ is
the same for both systems, but the wavelength and the corresponding
pulse speed are different:
\begin{eqnarray}
\WaveNumber^\prime & = & \WaveNumber\,\sqrt{\EquilMass+\ImpurityMass}\\
\PulseSpeed^\prime & = & \cos\left(\WaveNumber^\prime/2\right) \\
\Resistivity^\prime & = & \frac{(-\ImpurityMass\omega/\PulseSpeed^\prime)^2}{%
4\SpringConstant(\EquilMass+\ImpurityMass)}
\end{eqnarray}
The behavior of the system for all values of impurity concentration
is conjectured from the electrical analogy: as Lyapunov exponent is to
resistivity, localization length is to conductivity. If one assumes that
at some intermediate concentration the behavior is simultaneously
expressing itself as two systems with localization lengths
$\LocLength_{c\rightarrow 0}$ and $\LocLength_{c\rightarrow 1}$,
these two systems should contribute
independently to the overall behavior. By analogy to conductors,
the total localization length $\LocLength$ would be additive:
\begin{equation}
\LocLength = \LocLength_{c\rightarrow 0} \,+\, \LocLength_{c\rightarrow 1}
\label{loc_total:eqn}
\end{equation}
This equation represents a more complete estimate for the localization length that is
valid for both strong and concentrated scatterers.
% When compared to the results in Azbel and Soven, \cite{Azbel82}
% Eqn.~\ref{loc_total:eqn} is at least as accurate for
% displacement wavelengths greater than
% or equal to four lattice spacings.
%
% POSITIVE IMPURITY MASS
%
\subsection{Positive $\ImpurityMass$}
\begin{figure}
\includegraphics[width=\columnwidth]{fig6}
\caption{Localization length $\LocLength$ as a function of impurity
concentration $c$ in a discrete system having scatterers with cross section
0.2.
Solid lines is Eqn.~\protect{\ref{loc_total:eqn}},
dashed line is locus of minima, and
dotted line denotes equal contribution from $\LocLength_{c\rightarrow 0}$ and
$\LocLength_{c\rightarrow 1}$.
Inset shows same data as a function of (1-$c$).}
\label{phase:fig}
\end{figure}
The accuracy of this approximation is shown in Fig.~\ref{phase:fig} for
systems
having impurity scattering cross section $\CrossSection=0.20$ and displacement
wavelengths ranging from 8 to 128. The localization length $\LocLength$
for these systems was calculated using the MK solution. As can be seen,
the approximation in Eqn.~\ref{loc_total:eqn} is reasonably accurate.
The inset shows the same data, plotted as a function of $(1-c)$, highlighting
the separate behavior near $c\rightarrow 1$ for the different wavelengths.
Also shown in Fig.~\ref{phase:fig} are a dashed line and a dotted line. The
dashed line is the locus of localization length minima as a function of
displacement wavelength. Not only does the minimum localization length
increase with increasing wavelength, the concentration at which this happens
decreases with increasing wavelength. The dotted line in Fig.~\ref{phase:fig}
is the locus of points where
$\LocLength_{c\rightarrow 0}=\LocLength_{c\rightarrow 1}$.
This locus of points is meant to delineate the
cross-over point as the system passes from one dominate phase to the other.
The cross-over point has a stronger concentration dependence than the locus
of minima.
%
% NEGATIVE IMPURITY MASS
%
\subsection{Negative $\ImpurityMass$}
\begin{figure}
\includegraphics[width=\columnwidth]{fig7}
\caption{Localization length $\LocLength$ as a function of impurity
concentration $c$ for $\ImpurityMass = \pm$ 0.8.
Inset shows same data as a function of (1-$c$).}
\label{neg_cs_phase:fig}
\end{figure}
At a given oscillation wavelength and frequency,
the impurity scattering cross section depends upon the magnitude of the
change in mass $\ImpurityMass$. For small impurity concentrations, the
localization length for a system with impurity mass $\EquilMass+\ImpurityMass$ will be
equal to one with impurity mass $\EquilMass-\ImpurityMass$. At higher concentrations,
however, the behavior of systems will differ.
As an example, the localization concentration dependence was calculated for two systems
with displacement wavelength $\Wavelength=32$. The added impurity masses were
$+0.8$ and $-0.8$, and the localization length was determined by the MK method.
The results of the calculation are shown in Fig.~\ref{neg_cs_phase:fig} as a function of the impurity concentration $c$.
The estimate from Eqn.~\ref{loc_total:eqn} is shown as a solid line.
Within the inset are the data plotted as a function of $(1-c)$.
As expected, the behavior of the two systems diverge at higher concentrations.
% The divergence of behavior between positive and negative values of $\ImpurityMass$
% are also present in the rate of mode-mode interactions in anharmonic
% chains \cite{Snyder99}.
%
% AZBEL AND SOVEN COMPARISON
%
\subsection{Azbel and Soven Comparison}
\begin{figure}
\includegraphics[width=\columnwidth]{fig8}
\caption{Localization length $\LocLength$ as a function of
impurity concentration $c$ for two systems with
different wavenumber $\WaveNumber$.
Open symbols are calculations using the MK method,
and solid lines are Eqn.~\protect\ref{loc_total:eqn}.
The figure can be compared directly to
Figure 1 of Azbel and Soven\cite{Azbel82}}
\label{AS:fig}
\end{figure}
To better judge performance, Eqn.~\ref{loc_total:eqn} is
compared to the more rigorous result of Azbel and Soven
\cite{Azbel82} (AS).
The AS model contains quantum particles interacting
with delta function potentials that have strength $V$
and are located at random integer values of $x$.
For the AS systems, the value of $V$ is equal to -1, and
the scattering cross section
$\CrossSection_{\textrm{AS}}$ of each scatterer is
a function of the particle wave number $\WaveNumber$:
\begin{equation}
\CrossSection_{\textrm{AS}} = \frac{1}{1+16 \sin^2 (\WaveNumber/2)}
\end{equation}
This is sufficient to duplicate the AS numerical calculation.
Figure 1 of Azbel and Soven\cite{Azbel82} shows results from calculations
made for three values of wavenumber $\WaveNumber$: 0.02, 1.5, 3.13.
The MK method is used here to duplicate the numerical results for the two smaller values
of $\WaveNumber$, and the results are shown as open symbols in
Fig.~\ref{AS:fig}.
The solid lines in Fig.~\ref{AS:fig} are the corresponding estimate of
Eqn.~\ref{loc_total:eqn}.
Two things should be noted explicitly:
The AS definition of localization length corresponds to
twice the localization length defined here.
The definition of localization length in Azbel and Soven\cite{Azbel82} uses
the Landauer \cite{Landauer70} scaling parameter, while subsequent
work \cite{Azbel83a,Azbel83b} use the scaling of Anderson et al.~\cite{Anderson80}.
The results shown in Fig.~\ref{AS:fig} use the latter scaling definitions.
%
% c\lambda EFFECT
%
\subsection{$c\Wavelength$ Effect}
\begin{figure}
\includegraphics[width=\columnwidth]{fig9}
\caption{The ratio of the measured localization $\LocLength$ to the
calculated localization $\LocLength_\circ$ as a function of
product of impurity concentration $c$ and wavelength $\Wavelength$.
Filled symbols are calculated solutions for systems having impurity
cross section 0.2.
The dotted line is the limiting curve for $\Wavelength\rightarrow\infty$.}
\label{clambda:fig}
\end{figure}
As can be seen in Figs.~\ref{phase:fig} and
\ref{neg_cs_phase:fig}, the behavior
of the total localization length diverges from the dilute limit $\LocLengthDL$.
In Fig.~\ref{phase:fig}, the point at which $\LocLength$ begins to differ from
$\LocLengthDL$ is a function of the displacement wavelength. To more
clearly demonstrate this effect,
the ratio $\LocLength/\LocLengthDL$ is shown in Fig.~\ref{clambda:fig}
as a function of the product $c\Wavelength$. The data shown are those
appearing in Fig.~\ref{phase:fig}, with the addition of those
for $\Wavelength=512$. Also shown in the figure is a dotted line denoting
the long wavelength limiting behavior
for $\Wavelength\rightarrow\infty$.
For values of $c\Wavelength$ greater than 1, the dilute limit approximation
does not hold, and the observed localization length is greater than the
dilute limit estimation.
Therefore,
even though the effect is less dramatic for increasing $\Wavelength$, there
is a minimum effect, regardless of the wavelength.
%
% \beta-FPU IMPLICATIONS
%
\section{Anharmonic Chains}
\begin{figure}
\includegraphics[width=\columnwidth]{fig10}
\caption{The ratio of the measured localization $\LocLength$ to the
dilute limit localization length $\LocLength_\circ$, normalized by the
impurity concentration $c$, as a function of
displacement wavelength $\Wavelength$.}
\label{xi_lambda:fig}
\end{figure}
In an anharmonic system, phonon-phonon interactions will
lead to the creation of displacements with varying wavelength.
In time, very long wavelength displacements will be created.
Because the
impurity masses are fixed in time, the scattering cross section
for the displacement waves will decrease as $\Wavelength^{-2}$. In addition,
because the impurities are fixed in space, the product $c\Wavelength$
will increase over time. Based on the results from the last Section, one would expect that
the ratio $\LocLength/\LocLengthDL$ will diverge to infinity with increasing wavelength.
Using Eqn.~\ref{loc_total:eqn}, the ratio $\LocLength/\LocLengthDL$ was calculated
as a function of displacement wavelength for systems having constant impurity mass and
concentration. The results are shown in Fig.~\ref{xi_lambda:fig} for different values for
$\ImpurityMass$. For all values of impurity concentration $c$, the ratio
$\LocLength/\LocLengthDL$ asymptotes to a constant at long wavelength:
\begin{equation}
\lim_{\Wavelength\rightarrow\infty} \frac{\LocLength}{\LocLengthDL} =
1 + \frac{c}{1-c}\,\left(\EquilMass+\ImpurityMass\right)
\label{xi_lambda:eqn}
\end{equation}
Therefore, for anharmonic systems, the localization length is characterized, to within
a constant, by the dilute limit expression for $\LocLengthDL$ given in
Eqn.~\ref{eqn:LocLength}.
The same relationship applies to both positive and negative values for
$\ImpurityMass$. For negative values of $\ImpurityMass$, in fact, in the limit
$\ImpurityMass\rightarrow -1$, the
behavior of the system is accurately characterized
by $\LocLengthDL$ at long wavelengths.
\section{Conclusion}
The localization length for harmonic chains having binary disorder can be predicted accurately
over a wide range of wavelengths,
impurity cross section, and impurity concentration.
%
The ingredient needed for this prediction is the cross section
of a single scatterer, corrected for short wavelength displacements via the relative
pulse propagation speed.
%
The localization length over the entire impurity concentration range can then be estimated by
approximating the system as a sum of two independent systems, each accounting for the
behavior of the system at the two limits of impurity concentration.
The general result applies to systems in which the impurity mass is either larger or smaller
that the original mass.
%
Although the scattering cross section is symmetric about zero, with respect to the mass
added to the background value, the localization length behavior differs for negative and
positive changes in mass having the same scattering cross section.
%
This difference in behavior with respect to localization length, along with previous results
showing differences in the rate of phonon-phonon interactions, are discussed in the
context of numerical experiments on anharmonic systems.
The general results also suggest that the localization length of long
displacement wavelengths created by phonon-phonon interactions can be
approximated, to within a constant, from dilute limit calculations results.
%
For increasing concentration
and constant impurity concentration, the localization length, with respect to the dilute limit
prediction, will eventually diverge toward infinity.
%
This deviation occurs for all wavelengths, and is a universal function of the impurity
concentration and the displacement wavelength, for a constant cross section.
%
By contrast,
in a numerical experiment on an anharmonic system composed of fixed scatterers, the
scattering cross section decreases with increasing wavelength.
%
For these systems, the long wavelength behavior is, to within a constant, accurately predicted
by dilute limit predictions.
%
This conjecture will be confirmed in a future paper \cite{Snyder04}.
\begin{acknowledgments}
This work was supported by the NSF under grant number DMR-01-32726.
\end{acknowledgments}
\bibliography{/Users/jackal/jdb/j_abbrev,references}
\end{document}
\begin{thebibliography}{22}
\expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
\expandafter\ifx\csname bibnamefont\endcsname\relax
\def\bibnamefont#1{#1}\fi
\expandafter\ifx\csname bibfnamefont\endcsname\relax
\def\bibfnamefont#1{#1}\fi
\expandafter\ifx\csname citenamefont\endcsname\relax
\def\citenamefont#1{#1}\fi
\expandafter\ifx\csname url\endcsname\relax
\def\url#1{\texttt{#1}}\fi
\expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
\providecommand{\bibinfo}[2]{#2}
\providecommand{\eprint}[2][]{\url{#2}}
\bibitem[{\citenamefont{Anderson}(1958)}]{Anderson58}
\bibinfo{author}{\bibfnamefont{P.~W.} \bibnamefont{Anderson}},
\bibinfo{journal}{Phys.\ Rev.} \textbf{\bibinfo{volume}{109}},
\bibinfo{pages}{1492} (\bibinfo{year}{1958}).
\bibitem[{\citenamefont{Landauer}(1970)}]{Landauer70}
\bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Landauer}},
\bibinfo{journal}{Philos.\ Mag.} \textbf{\bibinfo{volume}{21}},
\bibinfo{pages}{863} (\bibinfo{year}{1970}).
\bibitem[{\citenamefont{Anderson et~al.}(1980)\citenamefont{Anderson, Thouless,
Abrahams, and Fisher}}]{Anderson80}
\bibinfo{author}{\bibfnamefont{P.~W.} \bibnamefont{Anderson}},
\bibinfo{author}{\bibfnamefont{D.~J.} \bibnamefont{Thouless}},
\bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Abrahams}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{D.~S.} \bibnamefont{Fisher}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{22}},
\bibinfo{pages}{3519} (\bibinfo{year}{1980}).
\bibitem[{\citenamefont{Sheng et~al.}(1986)\citenamefont{Sheng, White, Zhang,
and Papanicolaou}}]{Sheng86b}
\bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Sheng}},
\bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{White}},
\bibinfo{author}{\bibfnamefont{Z.-Q.} \bibnamefont{Zhang}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Papanicolaou}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{34}},
\bibinfo{pages}{4757} (\bibinfo{year}{1986}).
\bibitem[{\citenamefont{Bourbonnais and Maynard}(1990)}]{Bourbonnais90a}
\bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Bourbonnais}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Maynard}},
\bibinfo{journal}{Phys.\ Rev.\ Lett.} \textbf{\bibinfo{volume}{64}},
\bibinfo{pages}{1397} (\bibinfo{year}{1990}).
\bibitem[{\citenamefont{Azbel and Soven}(1982)}]{Azbel82}
\bibinfo{author}{\bibfnamefont{M.~Y.} \bibnamefont{Azbel}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Soven}},
\bibinfo{journal}{Phys.\ Rev.\ Lett.} \textbf{\bibinfo{volume}{49}},
\bibinfo{pages}{751} (\bibinfo{year}{1982}).
\bibitem[{\citenamefont{Held et~al.}(1997)\citenamefont{Held, Pfeiffer, and
Kuhn}}]{Held97}
\bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Held}},
\bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Pfeiffer}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Kuhn}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{55}},
\bibinfo{pages}{231} (\bibinfo{year}{1997}).
\bibitem[{\citenamefont{Rohmfeld et~al.}(2001)\citenamefont{Rohmfeld,
Hundhausen, Ley, Schulze, and Pensl}}]{Rohmfeld01}
\bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Rohmfeld}},
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Hundhausen}},
\bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Ley}},
\bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Schulze}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Pensl}},
\bibinfo{journal}{Phys.\ Rev.\ Lett.} \textbf{\bibinfo{volume}{86}},
\bibinfo{pages}{826} (\bibinfo{year}{2001}).
\bibitem[{\citenamefont{Widulle et~al.}(2002)\citenamefont{Widulle, Serrano,
and Cardona}}]{Widulle02}
\bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Widulle}},
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Serrano}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Cardona}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{65}},
\bibinfo{pages}{075206} (\bibinfo{year}{2002}).
\bibitem[{\citenamefont{de~L.~Kronig and Penney}(1931)}]{Kronig31}
\bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{de~L.~Kronig}}
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{W.~G.}
\bibnamefont{Penney}}, \bibinfo{journal}{Proc.\ Roy.\ Soc.\ Lond.\ A}
\textbf{\bibinfo{volume}{130}}, \bibinfo{pages}{499} (\bibinfo{year}{1931}).
\bibitem[{\citenamefont{MacKinnon and Kramer}(1981)}]{MacKinnon81}
\bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{MacKinnon}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Kramer}},
\bibinfo{journal}{Phys.\ Rev.\ Lett.} \textbf{\bibinfo{volume}{47}},
\bibinfo{pages}{1546} (\bibinfo{year}{1981}).
\bibitem[{\citenamefont{Fermi et~al.}(1955)\citenamefont{Fermi, Pasta, and
Ulam}}]{Fermi55}
\bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Fermi}},
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Pasta}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Ulam}}, \bibinfo{type}{Tech.
Rep.}, \bibinfo{institution}{Los Alamos Scientific Laboratory}
(\bibinfo{year}{1955}).
\bibitem[{\citenamefont{Candy and Rozmus}(1991)}]{Candy91}
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Candy}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Rozmus}},
\bibinfo{journal}{J.\ Comput.\ Phys.} \textbf{\bibinfo{volume}{92}},
\bibinfo{pages}{230} (\bibinfo{year}{1991}).
\bibitem[{\citenamefont{Arfken}(1970)}]{Arfken70}
\bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Arfken}},
\emph{\bibinfo{title}{Mathematical Methods for Physicists}}
(\bibinfo{publisher}{Academic Press}, \bibinfo{address}{New York},
\bibinfo{year}{1970}).
\bibitem[{\citenamefont{Goldstein}(1980)}]{Goldstein58}
\bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Goldstein}},
\emph{\bibinfo{title}{Classical Mechanics}}
(\bibinfo{publisher}{Addison-Wesley}, \bibinfo{address}{Reading, MA},
\bibinfo{year}{1980}), \bibinfo{edition}{2nd} ed.
\bibitem[{\citenamefont{Morse and Ingard}(1968)}]{Morse68}
\bibinfo{author}{\bibfnamefont{P.~M.} \bibnamefont{Morse}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{K.~U.} \bibnamefont{Ingard}},
\emph{\bibinfo{title}{Theoretical Acoustics}} (\bibinfo{publisher}{Princeton
University Press}, \bibinfo{year}{1968}).
\bibitem[{\citenamefont{Kittel}(1986)}]{Kittel86}
\bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Kittel}},
\emph{\bibinfo{title}{Introduction to Solid State Physics}}
(\bibinfo{publisher}{Wiley}, \bibinfo{year}{1986}), \bibinfo{edition}{sixth}
ed.
\bibitem[{\citenamefont{Deych et~al.}(2001)\citenamefont{Deych, Lisyansky, and
Altshuler}}]{Deych01}
\bibinfo{author}{\bibfnamefont{L.~I.} \bibnamefont{Deych}},
\bibinfo{author}{\bibfnamefont{A.~A.} \bibnamefont{Lisyansky}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{B.~L.}
\bibnamefont{Altshuler}}, \bibinfo{journal}{Phys.\ Rev.\ B}
\textbf{\bibinfo{volume}{64}}, \bibinfo{pages}{224202}
(\bibinfo{year}{2001}).
\bibitem[{BIP(1993)}]{BIPM}
\emph{\bibinfo{title}{Guide to the Expression of Uncertainty in Measurement}}
(\bibinfo{publisher}{International Organization for Standardization (ISO)},
\bibinfo{address}{Gen{\`e}ve, Switzerland}, \bibinfo{year}{1993}), ISBN
\bibinfo{isbn}{92-67-10188-9}.
\bibitem[{\citenamefont{Azbel and Soven}(1983)}]{Azbel83a}
\bibinfo{author}{\bibfnamefont{M.~Y.} \bibnamefont{Azbel}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Soven}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{27}},
\bibinfo{pages}{831} (\bibinfo{year}{1983}).
\bibitem[{\citenamefont{Azbel}(1983)}]{Azbel83b}
\bibinfo{author}{\bibfnamefont{M.~Y.} \bibnamefont{Azbel}},
\bibinfo{journal}{Phys.\ Rev.\ B} \textbf{\bibinfo{volume}{28}},
\bibinfo{pages}{4106} (\bibinfo{year}{1983}).
\bibitem[{\citenamefont{Snyder and Kirkpatrick}(2004)}]{Snyder04}
\bibinfo{author}{\bibfnamefont{K.~A.} \bibnamefont{Snyder}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{T.~R.} \bibnamefont{Kirkpatrick}}
(\bibinfo{year}{2004}), \bibinfo{note}{(to be submitted)}.
\end{thebibliography}