Mesoscopic Fluctuations in Quantum Dots in the Kondo Regime
Extracted source text ยท arXiv cond-mat/0306074
\documentclass[aps,prb,twocolumn,showpacs,tightenlines]{revtex4}
\usepackage{epsfig,graphicx,times}
\begin{document}
\title{Mesoscopic Fluctuations in Quantum Dots in the Kondo Regime}
\author{R. K. Kaul}
\author{Denis Ullmo}
\author{Harold U. Baranger}
\affiliation{Department of Physics, Duke University,
Durham, NC 27708-0305}
\date{\today}
\begin{abstract}
Properties of the Kondo effect in quantum dots depend sensitively on
the coupling parameters and so on the realization of the quantum dot
-- the Kondo temperature itself becomes a mesoscopic
quantity. Assuming chaotic dynamics in the dot, we use random matrix
theory to calculate the distribution of both the Kondo temperature and
the conductance in the Coulomb blockade regime. We study two
experimentally relevant cases: leads with single channels and leads
with many channels. In the single-channel case, the distribution of
the conductance is very wide as $T_{K}$ fluctuates on a logarithmic
scale. As the number of channels increases, there is a slow crossover
to a self-averaging regime.
\end{abstract}
\pacs{73.23.Hk, 72.15 Qm, 73.63.Kv}
\maketitle
%\section{Introduction}
Advances in fabrication of nanoscale devices have made possible unprecedented control over their properties.\cite{kouwenhoven} While the transport properties of quantum dots -- systems of electrons confined to small regions of space -- have been studied extensively for the last decade, it is only recently that their many-body aspects have been probed\cite{ralph,cobden} using the exquisite control now available. On the other hand, a continually fascinating aspect of nanophysics is the presence of quantum coherence and the interference ``fluctuations'' that it engenders.\cite{kouwenhoven,alhassid} It is natural, then, to ask how this classic mesoscopic physics effect influences the newly found many-body aspects.
Quantum dots are usually formed in one of two ways: either by depleting a two-dimensional electron gas (2DEG) at the interface of a GaAs/AlGaAs heterostructure -- a \textit{semiconductor} quantum dot (SQD)\cite{kouwenhoven} -- or by attaching leads to ultra-small \textit{metallic} grains (MQD).\cite{ralph} If the dot is weakly coupled to all leads used to probe it, there is an energy price to be paid for an excess electron to enter the dot, simply because of the energy required to localize the charge. This regime is known as the Coulomb Blockade (CB). In a SQD, one attains this regime by reducing the number of propagating channels in the leads; after the threshold of the lowest transverse mode becomes larger than the Fermi energy in the leads, an effectively 1D tunneling barrier is formed in this lowest mode. In a MQD, in contrast, tunneling junctions with the leads can be made by oxidizing the surface of the dot. The main difference for our purposes is that in the case of a MQD there are many channels propagating at $E_{F}$ whereas in SQD's there is only one relevant quantum channel.
By capacitively coupling to a metallic gate, one can control the dot's potential, allowing current to flow and bringing out mesoscopic effects.\cite{kouwenhoven} For certain values of the gate voltage $V_{g}$, the electrostatic energy difference between $N$ electrons in the dot and $N \!+\! 1$ electrons is balanced by the interaction with the gate: an electron can freely jump from the left lead onto the dot and then out into the other lead. This process produces a peak in the conductance at this $V_{g}$. By changing the gate voltage one can observe large peaks followed by valleys. Both the peak heights and peak spacings fluctuate as one varies $V_{g}$. The distribution and correlations of the peak heights and spacings have been studied experimentally, as well as theoretically using random matrix theory (RMT).\cite{alhassid} In general there is agreement between experiment and theory.\cite{alhassid,usaj_spacings2,usaj_heights}
Whether the temperature $T$ is larger or smaller than the typical width of the resonances $\Gamma$ has a large effect on the dot's conductance. As long as $T \!\gg\! \Gamma$ transport proceeds by either resonant tunneling (peaks) or the off-resonant process known as co-tunneling (valleys).\cite{abg} However, when $T$ is sufficiently small many-body effects become important; in particular, at $T \!=\! 0$ if $N$ is odd the conductance can be of order the conductance quantum $~{e}^{2}/h$ in the Coulomb blockade valleys where naively one expects a strong suppression of the conductance.\cite{glazman89,lee89} Only recently have experiments actually seen this enhancement.\cite{goldhaber,cronenwett,unitary,abspar}
The temperature scale at which these many body effects develop is called the Kondo temperature $T_{K}$.\cite{hewson} This scale is exponentially sensitive to the tunneling rates from the leads to the dot. Since these tunneling rates depend on the wavefunctions on the dot, $T_{K}$ shows mesoscopic fluctuations.
Experimentally, mesoscopic fluctuations of the Kondo effect in SQD's is observed. Perhaps the clearest example is in Ref. \onlinecite{unitary} where the conductance as a function of magnetic field in the Kondo regime is shown. In addition, the frequent observation of the Kondo effect in certain valleys but not others\cite{cronenwett} suggests that mesoscopics plays a role. Likewise, the observation of the Kondo effect in two adjacent valleys\cite{abspar} argues for the important role of fluctuations.\cite{cobden}
At low temperatures $T_{K}$ is the only important energy scale in the problem,\cite{hewson} and thus one can calculate the distribution of any property given the distribution of $T_{K}$. We will focus on the conductance as it is the most relevant experimentally.
Assuming chaotic dynamics in the dot, we use random matrix theory to calculate the distribution of $T_{K}$ in the CB valleys. We restrict our attention to the case of odd $N$ and $S=1/2$ and study how the fluctuations depend on the number of propagating channels. We go on to calculate the distribution of the conductance at $T=0$ and discuss the effect of finite temperature.
%\section{The Hamiltonian}
%\label{theham}
\textit{The Hamiltonian---} The Hamiltonian of the system, ${\hat
H}={\hat H_{\rm dot}}+{\hat H_{\rm leads}}+{\hat H_{\rm T}}$, consists
of the quantum dot's Hamiltonian, the Hamiltonian of the leads, and
the tunneling Hamiltonian which describes dot-lead couplings.
%\begin{eqnarray}
%{\hat H}={\hat H_{\rm dot}}+{\hat H_{\rm leads}}+{\hat H_{\rm T}}.
%\label{hamilt}
%\end{eqnarray}
${\hat H_{\rm dot}}$ has both a one-body part and an interaction
part. Impurities and/or scattering from the boundaries are taken into
account statistically by using a random matrix model for the one-body
Hamiltonian. The most important interactions of the electrons on the
dot are the charging effect and exchange.\cite{abg} Thus, we find
\begin{equation}
{\hat H_{\rm dot}} =
\sum_{{j}\sigma}
\varepsilon_{{j}}
{\hat c}_{{j}\sigma}^{\dagger}
{\hat c}_{{j}\sigma}
+
{E_{C}}{(\hat n -\cal{N})}^{2}
-
{J_{S}}{\hat S}^{2},
\end{equation}
where $\varepsilon_{j}$ is the single-particle energy spectrum on the dot; $\cal{N}$ is the dimensionless gate voltage used to tune the number of electrons on the dot; $\hat n$ is the number operator for excess electrons on the dot; $\hat S$ is the spin operator; and $J_{S}$ and $E_{C}$ are the exchange constant and charging energy, respectively.
The Hamiltonian of the leads differs in the SQD and MQD cases. In the SQD a tunneling junction is made by pinching until just after the last propagating mode is cut off -- the lead is effectively 1D. In the MQD, the leads are wide and many propagating channels can tunnel through the oxide barrier. In the general case of $N_{\rm L}$ and $N_{\rm R}$ channels, we can label all the states in the leads by a channel index and 1D momentum,
\begin{equation}
{\hat H_{\rm leads}} =
\sum_{m k \sigma}
(\epsilon_{k}+E_{m}) \;
{\hat c}_{mk\sigma}^{\dagger}
{\hat c}_{mk\sigma},
\end{equation}
where $\epsilon_{k} + E_{m}$ are the one-particle energies in the $m^{\rm th}$ channel with momentum $k$.
Finally, the Hamiltonian for the dot-lead coupling is
\begin{equation}
{\hat H_{\rm T}}=
\sum_{jmk\sigma}
(t_{mj} \;
{\hat c}_{mk\sigma}^{\dagger}{\hat c}_{j\sigma}
+ {\rm h.c.})
\label{tunnelingh}
\end{equation}
where $t_{mj}$ are the matrix elements for each of the $N_{\rm L}+N_{\rm R}$ quantum channels tunneling into the $j^{\rm th}$ state of the quantum dot. $|t_{mj}|^2$ is proportional to the intensity of the wavefunction $j$ in the dot. We assume that it is independent of $k$ in the lead since the typical energy scale for changing wavefunctions in the clean leads is much larger than other relevant energy scales.
We would like to rewrite this Hamiltonian in terms of energy rather
than momentum states in the leads. We define new operators ${\hat
c_{m\epsilon\sigma}^{\dagger}}={{\hat c_{mk\sigma}^{\dagger}}}
%%\sqrt{\vert d\epsilon/dk\vert}
\vert d\epsilon/dk\vert^{-1/2}$
that create states with energy $\epsilon=E_{m}+\hbar^{2}k^{2}/2m$ in the $m^{\rm th}$ channel. The normalization is chosen so that $[{\hat c_{k}},{\hat c_{k'}^{\dagger}}]=\delta(k-k')$ implies $[{\hat c_{\epsilon}},{\hat c_{\epsilon'}^{\dagger}}]=$ $\delta(\epsilon-\epsilon')$. In terms of these operators the Hamiltonian is
\begin{eqnarray}
{\hat H_{\rm L}}=\sum_{m\sigma }\int \!d\epsilon \; \epsilon \; ({\hat c_{m\epsilon \sigma}^{\dagger}}{\hat c_{m\epsilon \sigma}})\\
H_{\rm T}=\sum_{jm\sigma}\int \!d\epsilon \; [{\tilde t_{mj}} \; {\hat c_{m\epsilon \sigma}^{\dagger}}{\hat c_{j\sigma}} + {\rm h.c.}]
\end{eqnarray}
with ${\tilde t}_{mj}\equiv t_{mj} \; \vert d\epsilon/dk\vert^{1/2}$.
Note that even for the same $\epsilon$ the derivatives and the lower limits on the integrals
will be different for different channels.
%\section{Mapping to single channel Anderson model}
%\label{mapping}
\textit{Mapping to single channel Anderson model---} The Hamiltonian of the system bears close resemblance to
an $N$-channel impurity problem. However, since we are considering the $S=1/2$ odd $N$ valley Kondo regime,\cite{abg} only one of the spatial states in the dots is of consequence, namely the highest energy orbital which we shall refer to as $j_{\rm res}$. When the amplitude from this one level leaks out through the barriers it is intuitively plausible that it couples to only one linear combination of the transverse wave-functions in the leads. Hence we look for a rotation in the channel basis that chooses the correct wavefunction in the leads to couple to level $j_{\rm res}$ on the dot, denoting the new creation and destruction operators in the leads ${\hat z^{\dagger}}$ and ${\hat z}$. If we choose
\begin{eqnarray}
{\hat z_{1\epsilon \sigma}^{\dagger}}=\sum_{m\in L}({\tilde t_{mj_{\rm res}}}{\hat c_{m\epsilon \sigma}^{\dagger}})/t_{L}\nonumber\\
{\hat z_{2\epsilon \sigma}^{\dagger}}=\sum_{m\in R}({\tilde t_{mj_{\rm res}}}{\hat c_{m\epsilon \sigma}^{\dagger}})/t_{R}
\end{eqnarray}
where $t_{L,R}=(\sum_{m\in L,R}{\tilde t_{mj_{\rm res}}}^{2})^{1/2}$,
only one channel couples to the quantum dot on each lead. We may
choose the remaining channels in any way just making sure they are
orthonormal to $\hat z_{1,2}$. This defines a unitary rotation matrix
$U$ the first two rows of which we have specified.\cite{fn1} The
Hamiltonian has now been effectively reduced to two parts: (1) two
single channel leads connected to a quantum dot with tunneling matrix
elements $t_{L}$ and $t_{R}$, and (2) $N_{\rm L}+N_{\rm R}-2$
decoupled channels.\cite{fn2}
The decoupled channels do not contribute, and the solution to the
first problem is available in the literature:\cite{glazman89} By
making another rotation in $R$-$L$ space, one can reduce the problem
to a 1-channel Kondo problem plus a decoupled channel. Thus all the
thermodynamic properties of the quantum dot are those of a one-channel
Kondo problem. For the conductance, one must rotate back to the
original basis of leads. We may use the Kubo formula to calculate the
conductance at finite temperatures with the result\cite{glazman03}
\begin{eqnarray}
G_K &=& G_{0} \; F_K (T/T_K) \label{gk} \\
F_K(T/T_K)&=&\frac{1}{2}\int d\omega (-df/d\omega)
\sum_{\sigma}[-\pi \; {\text {Im}}\, \cal{T}_{\sigma}(\omega)] \nonumber
\end{eqnarray}
where $\cal{T}_{\sigma}(\omega)$ is the $T$-matrix of the scattering problem, $f$ is the Fermi function,
and $G_{0}$ is defined in terms of the level widths $\Gamma_{m}=2\pi\nu_{m}\vert t_{mj_{\rm res}}\vert^{2}$ by
\begin{equation}
G_{0}= 4 \frac{2e^2}{h}\frac{\sum\limits_{i \in L}\Gamma_{i}\sum\limits_{k \in R}\Gamma_{k}}{\big(\sum\limits_{i\in L,R}\Gamma_{i}\big)^{2}} \;.
\label{pre}
\end{equation}
So we may write the expression for the conductance as a product of a
prefactor $G_{0}$ and a universal function $F_K (T/T_K)$ which has
been calculated\cite{nrg,nrgtrans} using the numerical renormalization
group technique.
%\section{Fluctuations of $T_{K}$}
%\label{finite}
\begin{figure}[t]
\includegraphics[width=3.0in]{moment.eps}
\caption{The moments of the distribution of $T_K$ as a function of the
number of channels in the leads $N_{\rm ch}$ (Eq.~\ref{momeqn}). The
first three moments are shown -- average (dotted), root-mean-square
(dot-dashed), and asymmetry (solid). The moments indicate that the
distribution is badly behaved for small $N_{\rm ch}$ while it becomes
reasonable for $N_{\rm ch} > 10$.}
\label{moment}
\end{figure}
\textit{Fluctuations of $T_{K}$---} Because of the exponential
sensitivity of $T_{K}$ on the wavefunctions, fluctuations are expected
to be strong in the weak tunneling regime of a SQD. As one increases
the number of channels one obtains self-averaging. Our goal is to
study the crossover between these two limits.
For a simple Anderson model for the quantum dot, the
Kondo temperature from a scaling argument \cite{haldane} is
\begin{equation}
T_{K}=(U\Gamma)^{1/2}{\rm exp}
\left[\frac{\pi}{2}\frac{\epsilon_{0}(\epsilon_{0}+U)}{\Gamma U} \right]
\end{equation}
where $\epsilon_{0}$ is negative and measures the energy difference
between $E_{F}$ and the singly occupied level on the dot, $\Gamma$ is
the width of the level, and $U=2E_C$ is the on-site interaction. In
the case of a quantum dot, we have to correct for the fact that the
spectrum in the dot is dense, and thus the high energy cutoff for the
Kondo Hamiltonian is not given by $U$ but by the mean spacing
$\Delta$.\cite{abg,haldane} Varying $V_{G}$ adjusts $\epsilon_{0}$;
expressing this as a fraction of $U$ by choosing $\epsilon_{0}=-xU
(x>0)$, we write $T_{K}$ as
\begin{equation} \label{eq:TK}
T_{K}=\frac{\Delta}{\sqrt{x(1-x)}} \left( \frac{\Gamma}{U} \right)^{1/2}
{\rm exp}\left[-\frac{\pi x(1-x)}{2}\frac{U}{\Gamma}\right] \;.
\end{equation}
To calculate the distribution of the Kondo temperatures one needs an
appropriate distribution for $\Gamma$. For simplicity we assume that
the different channels in the lead couple to the dot wavefunction in a
similar way on average so that the average level widths for all the
channels are the same although they fluctuate independently. We allow,
however, for a different number of channels in the left and right
leads. From random matrix theory, it is known that the total level
width follows a $\chi^{2}$ distribution with $N_{\rm ch} \equiv \beta(
N_{\rm L}+N_{\rm R})$ degrees of freedom:\cite{alhassid}
\begin{equation}
P(\Gamma) = \left( \frac{\Gamma}{2\bar \Gamma} \right)^{\frac{N_{\rm ch}}{2}}
\frac{1}{(N_{\rm ch}/2 -1)!}\Gamma^{\frac{N_{\rm ch}}{2}-1}
e^{-N_{\rm ch} \Gamma/2\bar \Gamma}
\label{ptdist}
\end{equation}
where $\beta=1(2)$ with(out) time reversal symmetry.
The distribution of $t_{K} \equiv T_{K} \sqrt{ x(1-x)} / \Delta$
follows from Eqs.~(\ref{eq:TK}) and (\ref{ptdist}). While we have not
been able to obtain a closed-form expression for $P(t_{K})$, we can
calculate all its moments:
\begin{eqnarray}
\langle t_{K}^{n}\rangle &=& \frac{2(aN_{\rm ch})^{N_{\rm
ch}/2}}{(N_{\rm ch}/2-1)!}
\left(\frac{\pi x(1-x) n}{2aN_{\rm ch}}\right)^{\frac{n+N_{\rm ch}}{4}}\nonumber
\\ & \times & K_{\frac{n+N_{\rm ch}}{2}}(\sqrt{2\pi anN_{\rm ch} x(1-x)}\,)
\label{momeqn}
\end{eqnarray}
where $a=U/2\overline{\Gamma}$ and $K_{n}(z)$ is the modified Bessel
function of the second kind.
\begin{figure}[t]
\includegraphics[width=3.0in]{tkdist.eps}
\caption{Probability density of $T_K$, $P(T_K)$, calculated
numerically in four cases: $N_{\rm ch} \!=\!$ 1, 2, 4, and 50. The
main panels are on a linear scale while the insets are the
distribution of $\log T_K$. The distributions for small $N_{\rm ch}$
are very broad while that for $N_{\rm ch} \!=\! 50$ is reasonably
behaved, though not yet Gaussian. Note that applying a magnetic field
doubles the number of effective channels in the leads -- changing
$N_{\rm ch} \!=\! 2$ to 4, for instance -- and so increases the
probability of seeing a Kondo effect. We have used $\bar \Gamma=0.2U$
and $x=1/2$. }
\label{tkdist}
\end{figure}
Note, first, that when $P(\Gamma)$ is broad, as for $N_{\rm ch} =1$ or
$2$, the fluctuations of $T_K$ are on a logarithmic scale and so are
huge. On the other hand, the Kondo temperature depends on only the sum
of the decay widths. On adding more channels one gets a sharply peaked
function for $\Gamma$. Thus, the huge logarithmic fluctuations
inherent in the Kondo effect are reduced until something well-behaved
is obtained.
In Fig.~\ref{moment} we have plotted the average, rms, as well as
cubic deviation as a function of the number of channels using our
analytic formula. The number of channels at which there is a
crossover, from logarithmic fluctuations to a distribution well
characterized by its mean and variance, is about $N_{\rm ch} \simeq
10$. We also note that the cubic moment and rms decay only
algebraically and so the crossover to self-averaging is slow -- there
is no ``scale" associated with this crossover. The main features of
the dependence on $N_{\rm ch}$ is clearly borne out in the numerical
results for the full distribution shown in Fig.~\ref{tkdist}.
%\section{Zero Temperature}
%\label{zero}
\textit{Zero Temperature---} In the case of many channels coupled to
the quantum dot, $\langle T_{K}\rangle$ is a meaningful quantity and
we can define a $T=0$ regime. To calculate the distribution of
conductance at zero temperature, we use $G_0 = G_K(T=0)$ taken from
Eq. (\ref{pre}). This distribution is also a good approximation in the
case $N_{\rm L} \gg N_{\rm R}$ as $T_K$ depends on $N_{\rm L} + N_{\rm
R}$ and hence does not fluctuate as much as $G_0$.
Using again the random matrix theory result (\ref{ptdist}) for the
level widths, the above calculation gives a result in closed form:
\begin{eqnarray}
P(g)= \frac{1}{N}\frac{k(g)^{N_{\rm L}/2-1}+k(g)^{N_{\rm
R}/2-1}}{[k(g)-1][1+k(g)]^{\frac{N_{\rm L}+N_{\rm R}}{2}-3}}
\label{distribution}
\\
k(g)=-1+\frac{1}{2g}+\frac{1}{2}\sqrt{\frac{1}{g^{2}}-\frac{4}{g}} \nonumber
\end{eqnarray}
\begin{figure}[t]
\includegraphics[width=3.0in]{analsim.eps}
\caption{\label{distplot}
Probability density of the conductance for three different sets of $(N_{\rm L},N_{\rm R})$ in the $T=0$ regime.}
\end{figure}
where $g$ is defined by $G=4(2e^{2}/h)g$.
We have plotted $P(g)$ in Fig.~\ref{distplot} for three different
channel realizations to emphasize that the distribution function can
be changed quite dramatically by changing $N_{\rm L}$ and $N_{\rm
R}$. As expected, in the symmetric case the most probable conductance
is $2(e^2/h)$. By putting in some asymmetry we can obtain a variety of
distributions. Note in particular the highly asymmetric case in which
the distribution peaks at a conductance much smaller than
$2(e^2/h)$. Since the fluctuations of $T_{K}$ depend on the sum of the
$\Gamma_{i}$, we expect in a heavily asymmetric case that the zero
temperature result will show even at finite temperatures. Such a
situation may be realized in a metallic particle when one of the
``leads" is an STM tip.
%\section{Conclusions}
%\label{conclusions}
\textit{Conclusions---} Our main results are: (1) The Kondo enhanced
conductance is given by (\ref{gk}) for any number of channels in the
left and right lead. (2) We have calculated the distribution function
for both $T_{K}$ (Figs. 1 and 2) and the prefactor $G_0$
(Fig.~3). (3) At $T=0$, fluctuations are dominated by the prefactor,
and one should look at (\ref{distribution}) for the distribution.
Turning to a comparison with the experiments, we first note that most
of the experimental dots were not in the deep Coulomb blockade
regime,\cite{goldhaber,cronenwett,unitary,abspar} and so application
of our results is problematic because we have neglected charge
fluctuations. Nonetheless, we use our theory to make a few
experimentally relevant estimates. One of the difficulties in the
experiments is finding several odd valleys in a row which show the
Kondo effect.\cite{chang} The changing characteristics of the dot as a
function of gate voltages contribute to this difficulty, but the
mesoscopic fluctuations of $T_K$ will also play a role. RMT predicts
that neighboring energy levels have completely uncorrelated
wavefunctions and so that $T_K$ in sequential odd $N$ valleys are
uncorrelated. For the dots of Ref.~\onlinecite{unitary}, for
instance, we estimate that $T_K > T$ in about $0.4$ of the odd valleys
(using $\langle\Gamma/U\rangle=0.2$ and $T/\Delta=0.02$).
Fluctuations of $G_K$ can also be generated by continuous tuning parameters such as magnetic field, revealing a correlation scale. The expected scale is the semiclassical scale for changing chaotic wavefunctions. Interestingly, in Ref.~\onlinecite{unitary} these fluctuations are correlated on a scale which is an order of magnitude larger. This is consistent with measurements of the correlations in Coulomb blockade valleys in the co-tunneling regime.\cite{cotunneling} Neither measurement is understood at this time.
%\begin{acknowledgments}
\textit{Acknowledgments---} We thank P.~Brouwer, L.~Glazman, K.~Matveev, E.~Mucciolo, and G.~Usaj for useful discussions.
This work was supported in part by the NSF (DMR-0103003).
%\end{acknowledgments}
\bibliography{kondo.bib}
\end{document}
\begin{thebibliography}{20}
\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{Kouwenhoven et~al.}(1997)\citenamefont{Kouwenhoven,
Marcus, McEuen, Tarucha, Westervelt, and Wingreen}}]{kouwenhoven}
\bibinfo{author}{\bibfnamefont{L.~P.} \bibnamefont{Kouwenhoven}},
\bibinfo{author}{\bibfnamefont{C.~M.} \bibnamefont{Marcus}},
\bibinfo{author}{\bibfnamefont{P.~L.} \bibnamefont{McEuen}},
\bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Tarucha}},
\bibinfo{author}{\bibfnamefont{R.~M.} \bibnamefont{Westervelt}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{N.~S.}
\bibnamefont{Wingreen}}, in \emph{\bibinfo{booktitle}{Mesoscopic Electron
Transport, NATO ASI, Ser. E, Vol. 345}}, edited by
\bibinfo{editor}{\bibfnamefont{L.}~\bibnamefont{Sohn}},
\bibinfo{editor}{\bibfnamefont{L.}~\bibnamefont{Kouwenhoven}},
\bibnamefont{and} \bibinfo{editor}{\bibfnamefont{G.}~\bibnamefont{Sch\"on}}
(\bibinfo{publisher}{Kluwer, Dordrecht}, \bibinfo{year}{1997}), pp.
\bibinfo{pages}{105--214}.
\bibitem[{\citenamefont{von Delft and Ralph}(2001)}]{ralph}
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{von Delft}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Ralph}},
\bibinfo{journal}{Phys. Rep.} \textbf{\bibinfo{volume}{345}},
\bibinfo{pages}{61} (\bibinfo{year}{2001}).
\bibitem[{\citenamefont{Pustilnik et~al.}(2001)\citenamefont{Pustilnik,
Glazman, Cobden, and Kouwenhoven}}]{cobden}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Pustilnik}},
\bibinfo{author}{\bibfnamefont{L.~I.} \bibnamefont{Glazman}},
\bibinfo{author}{\bibfnamefont{D.~H.} \bibnamefont{Cobden}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{L.~P.}
\bibnamefont{Kouwenhoven}}, \bibinfo{journal}{Lecture notes in Physics}
\textbf{\bibinfo{volume}{3}}, \bibinfo{pages}{579} (\bibinfo{year}{2001}),
\bibinfo{note}{http://xxx.lanl.gov/abs/cond-mat/0010336}.
\bibitem[{\citenamefont{Alhassid}(2000)}]{alhassid}
\bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Alhassid}},
\bibinfo{journal}{Rev. Mod Phys.} \textbf{\bibinfo{volume}{72}},
\bibinfo{pages}{895} (\bibinfo{year}{2000}).
\bibitem[{\citenamefont{Usaj and Baranger}(2002)}]{usaj_spacings2}
\bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Usaj}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{H.~U.} \bibnamefont{Baranger}},
\bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{66}},
\bibinfo{pages}{155333} (\bibinfo{year}{2002}).
\bibitem[{\citenamefont{Usaj and Baranger}(2003)}]{usaj_heights}
\bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Usaj}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{H.~U.} \bibnamefont{Baranger}},
\bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{67}},
\bibinfo{pages}{121308} (\bibinfo{year}{2003}).
\bibitem[{\citenamefont{Aleiner et~al.}(2002)\citenamefont{Aleiner, Brouwer,
and Glazman}}]{abg}
\bibinfo{author}{\bibfnamefont{I.~L.} \bibnamefont{Aleiner}},
\bibinfo{author}{\bibfnamefont{P.~W.} \bibnamefont{Brouwer}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{L.~I.}
\bibnamefont{Glazman}}, \bibinfo{journal}{Phys. Rep.}
\textbf{\bibinfo{volume}{358}}, \bibinfo{pages}{309} (\bibinfo{year}{2002}).
\bibitem[{\citenamefont{Glazman and Raikh}(1988)}]{glazman89}
\bibinfo{author}{\bibfnamefont{L.~I.} \bibnamefont{Glazman}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Raikh}},
\bibinfo{journal}{JETP Lett.} \textbf{\bibinfo{volume}{47}},
\bibinfo{pages}{452} (\bibinfo{year}{1988}).
\bibitem[{\citenamefont{Ng and Lee}(1988)}]{lee89}
\bibinfo{author}{\bibfnamefont{T.~K.} \bibnamefont{Ng}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Lee}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{61}},
\bibinfo{pages}{1768} (\bibinfo{year}{1988}).
\bibitem[{\citenamefont{Goldhaber-Gordon
et~al.}(1998)\citenamefont{Goldhaber-Gordon, G\"ores, Kastner, Shtiktman,
Mahalu, and Meirav}}]{goldhaber}
\bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Goldhaber-Gordon}},
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{G\"ores}},
\bibinfo{author}{\bibfnamefont{M.~A.} \bibnamefont{Kastner}},
\bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Shtiktman}},
\bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Mahalu}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Meirav}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{81}},
\bibinfo{pages}{5225} (\bibinfo{year}{1998}).
\bibitem[{\citenamefont{Cronenwett et~al.}(1998)\citenamefont{Cronenwett,
Oosterkamp, and Kouwenhoven}}]{cronenwett}
\bibinfo{author}{\bibfnamefont{S.~M.} \bibnamefont{Cronenwett}},
\bibinfo{author}{\bibfnamefont{T.~H.} \bibnamefont{Oosterkamp}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{L.~P.}
\bibnamefont{Kouwenhoven}}, \bibinfo{journal}{Science}
\textbf{\bibinfo{volume}{281}}, \bibinfo{pages}{540} (\bibinfo{year}{1998}).
\bibitem[{\citenamefont{van~der Weil et~al.}(2000)\citenamefont{van~der Weil,
Franceschi, Fujisawa, Elzerman, Tarucha, and Kouwenhoven}}]{unitary}
\bibinfo{author}{\bibfnamefont{W.~G.} \bibnamefont{van~der Weil}},
\bibinfo{author}{\bibfnamefont{S.~D.} \bibnamefont{Franceschi}},
\bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Fujisawa}},
\bibinfo{author}{\bibfnamefont{J.~M.} \bibnamefont{Elzerman}},
\bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Tarucha}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{L.~P.} \bibnamefont{Kouwenhoven}},
\bibinfo{journal}{Science} \textbf{\bibinfo{volume}{289}},
\bibinfo{pages}{2105} (\bibinfo{year}{2000}).
\bibitem[{\citenamefont{Schmid et~al.}(2000)\citenamefont{Schmid, Weis, Eberl,
and Klitzing}}]{abspar}
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Schmid}},
\bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Weis}},
\bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Eberl}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{K.~v.} \bibnamefont{Klitzing}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{84}},
\bibinfo{pages}{5824} (\bibinfo{year}{2000}).
\bibitem[{\citenamefont{Hewson}(1993)}]{hewson}
\bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Hewson}}, in
\emph{\bibinfo{booktitle}{The Kondo Problem to Heavy Fermions}}
(\bibinfo{publisher}{Cambrige University Press},
\bibinfo{year}{1993}).
\bibitem{fn1} Note the statistical properties of
the tunneling matrix elements are not invariant under
this transformation because the random matrix model
does not apply to the leads.
\bibitem{fn2} While preparing this work for publication, we became
aware of two papers with a similar result: S.~Y.\ Cho, H.-Q. Zhou, and
R.~H.\ McKenzie, cond-mat/0302090; P.\ Simon and I.\ Affleck,
cond-mat/0305326.
\bibitem[{\citenamefont{Glazman and Pustilnik}(2002)}]{glazman03}
\bibinfo{author}{\bibfnamefont{L.~I.} \bibnamefont{Glazman}} \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Pustilnik}}, in
\emph{\bibinfo{booktitle}{Proceedings of the NATO-ASI "New Directions in
Mesoscopic Physics"}} (\bibinfo{publisher}{Erice}, \bibinfo{year}{2002}),
\bibinfo{note}{http://xxx.lanl.gov/abs/cond-mat/0302159}.
\bibitem[{\citenamefont{Wilson}(1975)}]{nrg}
\bibinfo{author}{\bibfnamefont{K.~G.} \bibnamefont{Wilson}},
\bibinfo{journal}{Rev. Mod. Phys.} \textbf{\bibinfo{volume}{47}},
\bibinfo{pages}{773} (\bibinfo{year}{1975}).
\bibitem[{\citenamefont{Costi et~al.}(1994)\citenamefont{Costi, Hewson, and
Zlati\'c}}]{nrgtrans}
\bibinfo{author}{\bibfnamefont{T.~A.} \bibnamefont{Costi}},
\bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Hewson}},
\bibnamefont{and} \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Zlati\'c}},
\bibinfo{journal}{J. Phys. C} \textbf{\bibinfo{volume}{6}},
\bibinfo{pages}{2519} (\bibinfo{year}{1994}).
\bibitem[{\citenamefont{Haldane}(1978)}]{haldane}
\bibinfo{author}{\bibfnamefont{F.~D.~M.} \bibnamefont{Haldane}},
\bibinfo{journal}{J. Phys. C: Solid State Phys.}
\textbf{\bibinfo{volume}{11}}, \bibinfo{pages}{5015} (\bibinfo{year}{1978}).
\bibitem[{\citenamefont{Chang et~al.}(2003)\citenamefont{Chang, Jeong, and
Melloch}}]{chang}
\bibinfo{author}{\bibfnamefont{A.~M.} \bibnamefont{Chang}},
\bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Jeong}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Melloch}}, in
\emph{\bibinfo{booktitle}{Electron Transport in Quantum Dots}}, edited by
\bibinfo{editor}{\bibfnamefont{J.~A.} \bibnamefont{Bird}}
(\bibinfo{publisher}{Kluwer, New York}, \bibinfo{year}{2003}).
\bibitem[{\citenamefont{Cronenwett et~al.}(1997)\citenamefont{Cronenwett,
Patel, Marcus, Chapman, and Gossard}}]{cotunneling}
\bibinfo{author}{\bibfnamefont{S.~M.} \bibnamefont{Cronenwett}},
\bibinfo{author}{\bibfnamefont{S.~R.} \bibnamefont{Patel}},
\bibinfo{author}{\bibfnamefont{C.~M.} \bibnamefont{Marcus}},
\bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Chapman}}, \bibnamefont{and}
\bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Gossard}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{79}},
\bibinfo{pages}{2312} (\bibinfo{year}{1997}).
\end{thebibliography}