Probing Pauli Blocking Factors in Quantum Pumps with Broken Time-Reversal Symmetry
Extracted source text ยท arXiv cond-mat/0006331
%% ****** uses RevTeX 4 beta 3 ****** %
%%
%% Filename: PauliFactors.tex;
%% version 5 (16.02.2000)
%%
%% Dr. Mathias Wagner
%% Hitachi Cambridge Laboratory
%% Cavendish Laboratory
%% Madingley Road
%% Cambridge CB3 0HE
%% UK
%%
%% Tel : +44 (0)1223 467944
%% Fax : +44 (0)1223 467942
%% E-mail: wagner@phy.cam.ac.uk
%%
%% ***********************************************************
%% * TeX version 3.14159 *
%% * LaTeX2e <1999/06/01> patch level 1 *
%% * RevTeX 4 beta 3 (16 December 1999) *
%% ***********************************************************
%%
%%
%\documentclass[aps,preprint,groupedaddress,showpacs]{revtex4}
\documentclass[aps,twocolumn,groupedaddress]{revtex4}
%\documentclass[aps,preprint,superscriptaddress,showpacs]{revtex4}
%\documentclass[aps,twocolumn,groupedaddress]{revtex4}
%\documentclass[aps,twocolumn,groupedaddress,footinbib]{revtex4}
\usepackage{amsmath}
\usepackage{graphics}
\begin{document}
%\bibliographystyle{apsrev}
\bibliographystyle{revtex}
\preprint{TP-CB-371}
\title{Probing Pauli Blocking Factors in Quantum Pumps with Broken Time-Reversal Symmetry}
\author{Mathias Wagner}
\email[]{wagner@phy.cam.ac.uk}
\affiliation{Hitachi Cambridge Laboratory, Madingley Road,
Cambridge CB3 0HE, United Kingdom}
\date{\today}
\begin{abstract}
A recently demonstrated quantum electron pump is discussed within the
framework of photon-assisted tunneling. Due to lack of time-reversal
symmetry, different results are obtained for the pump current depending
on whether or not final-state Pauli blocking factors are used when
describing the tunneling process. Whilst in both cases the current
depends quadratically on the driving amplitude for moderate pumping, a
marked difference is predicted for the temperature dependence. With
blocking factors the pump current decreases roughly linearly with
temperature until $k_{\rm B} T$ $\approx$ $\hbar\omega$ is reached,
whereas without them it is unaffected by temperature, indicating that
the entire Fermi sea participates in the electronic transport.
\end{abstract}
% insert suggested PACS numbers in braces on next line
\pacs{73.50.Pz, 73.20.Dx, 73.40.Gk, 72.40.+w}
% 73.50.Pz: Photoconduction and photovoltaic effects (surfaces)
% 73.20.Dx: Electron states in low-dimensional structures
% (superlattices, quantun-well structures and multilayers)
% 73.20.Jc: Delocalisation processes
% 73.40.Gk: Tunneling
% 72.20.Ht: High-field and nonlinear effects
% 72.40.+w: Photoconduction and photovoltaic effects (bulk)
% 42.50.Hz: Strong-field excitation of optical transitions in
% quantum systems; multiphoton processes; dynamic Stark
% shift;
% 03.65.Ge: Solution of wave equations: bound states
\maketitle
\begin{figure}[b]
\centering
\includegraphics{basic-barrier.eps}
\caption{Single-barrier scattering state with transmitted and
reflected amplitudes, populated according to a distribution
function $f^{\rm L}$ characterising the left side. In the Pauli
picture, the tunneling probability incurs an additional
final-state blocking factor $1-f^{\rm R}$ on the right.
}
\label{Single-Barrier}
\end{figure}
Tunneling of electrons through classically forbidden regions is one of
the major paradigms of quantum mechanics. Although our understanding
has greatly improved over the past decades, some vital aspects of this
process are still subject to fierce debates. For example, no consensus
has been reached at yet whether so-called final-state blocking factors
exist in the tunneling process across a barrier sandwiched between two
conductors (Fig.~\ref{Single-Barrier}), to enforce Pauli's exclusion
principle that no two electrons may occupy the same quantum state.
This seemingly innocent question is related to the question where in
the so-called Fermi sea of conducting electrons the current
actually flows, and thus has far-reaching consequences for our
understanding of quantum transport. There are two schools on how to
calculate the tunnel current, one that insists on using the blocking
factors, and another that rejects them \cite{Datta95a}. The dilemma is
that both schools almost always seem to give identical answers for the
current. In this Letter we point out that this result is
fundamentally related to time-reversal symmetry, and study a generic
system where this symmetry is broken to gain insight into the nature
of blocking factors.
Consider a tunneling barrier impeding the current flow as depicted in
Fig.~\ref{Single-Barrier}. According to the first school, we start
by calculating the {\it coherent} scattering states of the system
consisting of an incident plane wave, a reflected wave, and a
transmitted wave on the far side. These scattering states travel either
from left to right or the opposite way. In the stationary limit they
can simply be populated according to distribution functions
$f^{\rm L}$ and $f^{\rm R}$ governing the asymptotic contact regions
on either side. This yields for the current \cite{Datta95a}
\begin{eqnarray}
I_{\rm S} & = &
{2e\over h} \int dE dE' \,{\cal D}_\bot T^+(E',E) f^{\rm L}(E) \cr
& - &
{2e\over h} \int dE dE' \,{\cal D}_\bot T^-(E,E') f^{\rm R}(E') ,
\label{eq:Tsu-Esaki-Current}
\end{eqnarray}
where ${\cal D}_\bot$ is a density-of-states factor \cite{footnote1},
$T^+(E',E)$ is the transmission probability for scattering states
incident from the left at energy $E$ and emerging at the right at $E'$
($\ne E$ in general), and $T^-$ is defined in a similar manner for the
reverse direction. According to Eq.~(\ref{eq:Tsu-Esaki-Current})
{\it all} scattering states contribute to the current, even those far
below the Fermi surface --- as recently discussed for the Fermi pump
\cite{Wagner99a}. It is only when the transmission probabilities are
symmetric, $T^+(E',E)$ = $T^-(E',E)$, that the {\it net} current seems
to stem from electrons in the immediate vicinity of the Fermi surface
only, as in this case all electrons further down cancel each other.
An alternative and widely used recipe for calculating the tunneling
current in Fig.~\ref{Single-Barrier} is based on the
transfer-Hamiltonian formalism, originally put forward by Oppenheimer
\cite{Oppenheimer28a}, and later refined by Bardeen and others
\cite{Duke69a}. Here the system is split into two subsystems, the
left- and the right-hand side, with a common overlap in the central
barrier region, and the \ result for the current can be expressed as
\begin{eqnarray}
I_{\rm P} & = &
{2e\over h} \int dE dE' \,{\cal D}_\bot T^+(E',E)
[1-f^{\rm R}(E')]f^{\rm L}(E) \cr
& - &
{2e\over h} \int dE dE' \,{\cal D}_\bot T^-(E,E')
[1-f^{\rm L}(E)]f^{\rm R}(E')
\label{eq:Pauli-Current}
\end{eqnarray}
which differs from Eq.~(\ref{eq:Tsu-Esaki-Current}) by the Pauli
blocking factors $1-f$. These factors are introduced --- more or less
{\it ad hoc} --- using the intuitive argument that an electron
tunneling from one side to the other needs to find an empty final
state on the far side to tunnel into \cite{Sturman84a}.
Clearly, the philosophies underlying Eqs.~(\ref{eq:Tsu-Esaki-Current})
and (\ref{eq:Pauli-Current}) are entirely different: For the former we
assume that the left- and right-hand side of the barrier belong to the
{\it same} system and that the scattering state extends coherently
across the barrier, whereas in the latter we consider {\it real}
incoherent transitions between initial and final states on opposite
sides of the barrier belonging to {\it different} subsystems
\cite{footnote2}.
Both views have their merits. Certainly in the limit of a very
transparent or even vanishing barrier, one has to regard both sides of
the barrier as part of the same system, in which case there is no place
for blocking factors. But equally well, if tunneling is weak, one can
argue that the two sides do form two different subsystems, and that
final-state blocking factors are mandatory to guarantee Pauli's
exclusion principle and to prevent an ``overflow'' of the final state.
Crucially, this analysis also applies to a system with {\it local}
time-dependent driving forces as its leads, and hence the distribution
of incident electrons, are not affected by the driving field. For a
system with time-reversal symmetry, it can be shown that the
transmission probabilities are such
that any channel $E \leftrightarrow E'$ is traversed with the same
probability in either direction, i.e. that microreversibility holds,
$T^+(E',E)$ $\equiv$ $T^-(E,E')$. Under this condition the cross terms
$f^{\rm L}f^{\rm R}$ arising from the Pauli blocking factors cancel and
Eqs.~(\ref{eq:Tsu-Esaki-Current}) and (\ref{eq:Pauli-Current}) yield
identical answers for the {\it total} current, which is the quantity
usually measured in experiment. It is probably because of this
indecisive result that both schools have coexisted for so long. The
only difference between the two is their prediction {\it where} in
the Fermi sea the current flows: With blocking factors, the current is
forced to flow close to the Fermi surface, as in this picture
lower-lying final states are blocked, whilst in the formulation based
on scattering states the current flows in the entire Fermi sea.
However, if time-reversal symmetry does not hold,
Eqs.~(\ref{eq:Tsu-Esaki-Current}) and (\ref{eq:Pauli-Current}) yield
different answers even for the {\it magnitude} of the current itself
as the cross terms do not cancel any longer. This is very intriguing
as it gives us hope to come closer to an experimentally verifiable
answer regarding the existence of Pauli blocking factors.
\begin{figure}[bt]
\centering
\includegraphics{Pump.eps}
\caption{Quantum-dot pump with broken time-reversal symmetry.
[A] Schematic drawing; [B] simple 1D model based on driven
$\delta$-function barriers.
}
\label{Pump}
\end{figure}
Our analysis is inspired by a recent experiment by the Marcus group
\cite{Switkes99a}. A semiconductor quantum dot with source and drain
point contacts has two additional lateral gates to which ac voltages
of relative phase $\phi$ are applied (see Fig.~\ref{Pump} A).
Time-reversal symmetry is broken unless $\phi$ is an integer multiple
of $\pi$. Averaging over different dot configurations, they measured
fluctuations of the emf voltage generated between the source and drain
contacts. Previous theoretical studies employed the concept of
adiabatic pumping \cite{Altshuler99a}.
We take a complementary perspective and view the pump current (which
exists even in the {\it absence} of any applied source-drain bias) as
due to photon-assisted tunneling \cite{Wagner95a}.
As a model system of a driven dot strongly coupled to its leads we
consider two harmonically oscillating $\delta$-function barriers of
equal strength $V_{\rm ac}$ a distance $d$ apart in a 1D potential,
with a variable phase difference $\phi$ in the ac signals
as depicted in Fig.~\ref{Pump} B \cite{Hekking91a}. This is a very
simplified model of the Marcus pump, but nevertheless it turns out to
exhibit many of its characteristic properties. For calculating the
transmission probabilities $T(E',E)$ across the dot we take advantage
of the fact that we can split this problem into two parts: If we know
the transmission and reflection amplitudes for each barrier separately,
we can use the Fabry-Perot method of raytracing known in optics to
calculate the interference pattern due to multiple reflections between
the barriers. As in optics the partial interference amplitudes can be
summed up to all orders in a geometric series, yielding for the
transmission amplitudes at the far side \cite{Wagner97b}
\begin{eqnarray}
t & = & t^R (I - K)^{-1} Q t^L ,
\label{Fabry-Perot}
\end{eqnarray}
where $K$ = $Qr^LQr^R$ describes one full round trip of the electron
between the two barriers $L$ and $R$, starting and ending at the $R$
barrier. Each round trip an electron at energy $E_n$ = $E$ +
$n\hbar\omega$ picks up a phase factor $Q_n$ = $\exp[i(k_n d+\theta)]$
consisting of two parts: The phase $k_n d$ incurred after travelling a
distance $d$ with wave vector $k_n$ = $\sqrt{2mE_n}/\hbar$, and a
fixed but unknown phase $\theta$. This latter phase is introduced to
account for the random changes in magnetic field and dot
geometry employed in the experiment to perform ensemble averages. In
general, the phase $\theta$ depends on the electron energy, but for
simplicity we ignore this and assume $\theta$ to be equally
distributed. Under this condition ensemble averaging simply means
averaging over $\theta$ at the end of the calculation.
Due to the discreteness of the photon energy, electrons emerge on the
far side of the barriers at energies differing from their
original energy $E$ by multiples of $\hbar\omega$: $E'$ = $E$ +
$n\hbar\omega$, the so-called sidebands. Consequently, all amplitudes
in Eq.~(\ref{Fabry-Perot}) have to be interpreted as
{\it matrices}, and the transmission probability takes the form
$T(E',E)$ = $\sum_n$ $\tilde T_n(E) \delta(E+n\hbar\omega-E')$.
Before presenting numerical results for strong driving, it is
instructive to study the weak-driving limit, which can be solved
analytically. Defining $T_{\rm e}(k,\varphi)$ = $\eta^2$ $(k_0/ 4 k)$
$\cos^2\left[d(k_0 + k)/2 + \varphi/2\right]$ where $\eta$ =
$V_{\rm ac}k_0/E_0$ is the dimensionless pump amplitude and $k_0$ the
wave vector of the incident electron, we obtain for the transmission
probabilities in the first three sidebands up to $\eta^2$
\cite{footnote3}
\begin{eqnarray}
\tilde T^+_0(\phi,\theta) & = &
1 + T_{\rm e}(-k_1,-\phi) + T_{\rm e}(-k_{-1},\phi) \cr
& &
- T_{\rm e}(k_1,2\theta-\phi) - T_{\rm e}(k_{-1},2\theta+\phi) \cr
\tilde T^+_{\pm 1}(\phi,\theta) & = & - T_{\rm e}(-k_{\pm 1},\mp \phi) .
\label{eq:T01}
\end{eqnarray}
The corresponding transmission probabilities $\tilde T^-_n$ for the
reverse direction are obtained by substituting $(\phi,\theta)$ $\to$
$(-\phi,\theta)$. As expected, the net transmission probability
$\tilde T_{\rm net}$ $\equiv$ $\tilde T^+$ - $\tilde T^-$ vanishes for
$\phi$ = $n\pi$ when time-reversal symmetry holds, and is maximal at
$\phi$ = $\pi/2$. In agreement with experimental findings, the pump
current at zero dc bias, being proportional to $\tilde T_{\rm net}$,
scales with $\eta^2$ for weak driving. In the experiment the photon
energy is 6 orders of magnitude smaller than the Fermi energy.
Expanding Eq.~(\ref{eq:T01}) in this limit the current turns out to
be {\it linear} in the driving frequency, in perfect
agreement with the experimental results available \cite{footnote4}.
\begin{figure}[tb]
\centering
\includegraphics{phi.eps}
\caption{Dependence of $\langle I_{\rm S}^2 \rangle^{1/2}$ on the
phase shift $\phi$ for weak and strong driving fields.
Parameters: $d$ = 0.2\,$\mu$m, $f$ = 10\,MHz, $m$ = 0.067\,$m_0$,
$E_F$ = 12\,meV, $T$ = 0.1\,K.
}
\label{I2-Dependence-On-Phi}
\end{figure}
\begin{figure}[tb]
\centering
\includegraphics{TE80.eps}
\caption{$\langle I_{\rm S}^2(\pi/2) \rangle^{1/2}$
and $\langle V_{\rm S}^2(\pi/2) \rangle^{1/2}$,
both based on Eq.~(\protect\ref{eq:Tsu-Esaki-Current}),
as a function of the driving strength $V_{\rm ac}$ for
a range of temperatures.
}
\label{Dependence-On-Vac-TE}
\end{figure}
After integrating over $\theta$ the ensemble average
$\langle\tilde T_{\rm net}\rangle$ is found to be identically zero up
to order $\eta^2$. This implies that the direct current
$\langle I \rangle$ is orders of magnitude smaller than its fluctuations
for all but the strongest driving fields, and hence we will concentrate
on the mean square average $\langle I^2 \rangle^{1/2}$ to study the
fluctuations instead \cite{footnote5}.
Figure~\ref{I2-Dependence-On-Phi} shows the dependence of
$\langle I_{\rm S}^2 \rangle^{1/2}$ on the phase shift $\phi$, where
we have used Eq.~(\ref{Fabry-Perot}) with an adaptive number of photon
sidebands to determine the full non-linear transmission probability,
and Eq.~(\ref{eq:Tsu-Esaki-Current}) to finally calculate the current.
For small driving amplitudes we find a $\sin\phi$ behaviour in
agreement with Eq.~(\ref{eq:T01}), whilst for stronger driving
nonharmonic features appear, both of which are in qualitative
agreement with experiment. A similar behaviour is observed when using
the formula (\ref{eq:Pauli-Current}) based on Pauli blocking factors
instead, except that in this case the pump current is much smaller.
For the remainder of this Letter we will fix the ac phase shift $\phi$
to $\pi/2$, the point of maximal time-reversal asymmetry. The current
fluctuations $\langle I_{\rm S}^2(\pi/2) \rangle^{1/2}$ calculated
using the formula (\ref{eq:Tsu-Esaki-Current}) {\it without} blocking
factors are illustrated in Fig.~\ref{Dependence-On-Vac-TE} as a
function of $\eta$, taken at the Fermi surface. For small driving
amplitudes up to $\eta$ $\approx$ 0.05 the fluctuations increase
quadratically with $\eta$ as suggested by Eq.~(\ref{eq:T01}) before
eventually starting to decrease. In the Marcus experiment, rather than
measuring the pump current, the emf voltage generated was studied.
Defining the emf voltage as the difference in chemical potentials
between the left and right leads necessary to make the pump current
vanish, we find that its fluctuations, also shown in
Fig.~\ref{Dependence-On-Vac-TE} (right-hand axis), have virtually the
same dependence on $\eta$ as the current fluctuations. Both the
quadratic rise as well as the levelling off for stronger pumping have
been observed in experiment. The maximal emf fluctuations generated,
15\,nV, is only one or two orders smaller than in the experiment,
which is a quite reasonable agreement given the simplicity of the
model.
\begin{figure}[tb]
\centering
\includegraphics{P8.eps}
\caption{$\langle I_{\rm P}^2(\pi/2) \rangle^{1/2}$, based on
Eq.~(\protect\ref{eq:Pauli-Current}), as a function of the
driving strength $V_{\rm ac}$ for a range of temperatures.
}
\label{Dependence-On-Vac-P}
\end{figure}
The corresponding results based on the alternative formula
(\ref{eq:Pauli-Current}) for the current which does include the Pauli
blocking factors are illustrated in Fig.~\ref{Dependence-On-Vac-P}.
Similar to the case without blocking factors of
Fig.~\ref{Dependence-On-Vac-TE}, the current fluctuations
$\langle I_{\rm P}^2(\pi/2) \rangle^{1/2}$ (as well as the emf
fluctuations, not shown) display an $\eta^2$ behaviour first before
eventually starting to saturate. Yet, not only do the magnitudes of
these results differ substantially from Fig.~\ref{Dependence-On-Vac-TE}
for the same driving amplitude, but now there is a very pronounced
temperature dependence as well!
This temperature dependence is worked out in more detail in
Fig.~\ref{Dependence-On-T} using a fixed driving strength of $\eta$ =
0.47. Whilst without Pauli blocking factors the variation with
temperature is minimal, an almost linear dependence is observed when
they are included. This distinctly different behaviour in the
low-temperature regime is most easily understood when looking at the
number of states in phase space effectively available for transport:
The blocking factors force the current to flow within a few
$k_{\rm B} T$ of the Fermi surface (see Fig.~\ref{Single-Barrier}),
and as the temperature approaches zero, this range of active
current-carrying states eventually diminishes to a minimal width of a
few $\hbar\omega$, which in the experiment is much smaller than
$k_{\rm B} T$. However, since each state can only carry a certain
maximal load, it follows that the pump current must also decrease with
temperature, until it settles for a residual value once $k_{\rm B} T$
$\approx$ $\hbar\omega$ is reached --- with our parameters at
$\approx$ 0.5\,mK. On the other hand, without blocking factors there
are no phase-space restrictions, in which case the pump current flows
in the entire Fermi sea and thus is largely immune to changes at the
Fermi surface brought about by temperature.
In the Marcus experiment the pump current is found to {\it increase} when
lowering the temperature, and appears to level off at $\approx$ 0.1\,K,
where phase-breaking events become less important \cite{Switkes99a}.
Such events are not included in our theory, and we can therefore not
expect to reproduce the high-temperature behaviour. However, the
experimental finding of a {\it saturated} pump current in the
low-temperature limit is (if genuine and not due to thermal decoupling)
only consistent with our results based on the scattering-state approach,
Eq.~(\ref{eq:Tsu-Esaki-Current}), but {\it not} with the formulation
relying on Pauli blocking factors, i.e. Eq.~(\ref{eq:Pauli-Current}).
Being able to prove the (non)existence of Pauli blocking factors has
drastic consequences for deciding whether the current flows in the
entire Fermi sea or at its surface only. In this Letter we have
demonstrated that a powerful tool for studying this issue is to look
at the temperature dependence of the pump current when breaking
time-reversal symmetry. Although our model system is simple, the
conclusions about the low-temperature behaviour, being drawn from
phase-space considerations, are clearly of a much more universal
nature.
\begin{figure}[b]
\centering
\includegraphics{Temp.eps}
\vspace{-0.2cm}
\caption{Temperature dependence of $\langle I^2(\pi/2) \rangle^{1/2}$
with and without Pauli blocking factors.
}
\label{Dependence-On-T}
\end{figure}
The author acknowledges discussions with C.~Marcus, and
support by the EU (FMRX-CT98-0180).
\bibliography{PauliFactors,DrivenSystems}
\end{document}
%
% ****** End of file PauliFactors.tex ******
\begin{thebibliography}{10}
\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 url\endcsname\relax
\def\url#1{\texttt{#1}}\fi
\expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
\expandafter\ifx\csname bibinfo\endcsname\relax \def\bibinfo#1#2{#2}\fi
\expandafter\ifx\csname eprint\endcsname\relax \def\eprint#1{#1}\fi
\bibitem{Datta95a}
\bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Datta}},
\emph{\bibinfo{title}{Electronic Transport in Mesoscopic Systems}}
(\bibinfo{publisher}{Cambridge University}, \bibinfo{year}{1995}).
\bibitem{footnote1}
Depending on the energy $E_\bot$ $\equiv$ $E'_\bot$
perpendicular to the tunneling, which we assume to be conserved.
\bibitem{Wagner99a}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Wagner}},
\bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sols}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{83}},
\bibinfo{pages}{4377} (\bibinfo{year}{1999}).
\bibitem{Oppenheimer28a}
\bibinfo{author}{\bibfnamefont{J.~R.} \bibnamefont{Oppenheimer}},
\bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{31}},
\bibinfo{pages}{66} (\bibinfo{year}{1928}).
\bibitem{Duke69a}
See, for instance, the review by
\bibinfo{author}{\bibfnamefont{C.~B.} \bibnamefont{Duke}},
\emph{\bibinfo{title}{Tunneling in Solids}}, vol. \bibinfo{volume}{10
(Suppl.)} of \emph{\bibinfo{series}{Solid State Physics}}
(\bibinfo{publisher}{Academic}, \bibinfo{address}{New York},
\bibinfo{year}{1969}).
\bibitem{Sturman84a}
A discussion for the case of elastic collisions can be found in:
\bibinfo{author}{\bibfnamefont{B.~I.} \bibnamefont{Sturman}},
\bibinfo{journal}{Usp. Fiz. Nauk} \textbf{\bibinfo{volume}{144}},
\bibinfo{pages}{497} (\bibinfo{year}{1984}), \bibinfo{note}{[Transl. in Sov.
Phys. Usp. {\bf 27}, 881 (1984)]};
\bibinfo{author}{\bibfnamefont{F.} \bibnamefont{Sols}},
\bibinfo{journal}{Ann. Phys. (NY)} \textbf{\bibinfo{volume}{214}},
\bibinfo{pages}{386} (\bibinfo{year}{1992}).
\bibitem{footnote2}
This is the same reasoning as in Fermi's Golden Rule, and so by analogy
the transfer-Hamiltonian formalism only applies for small transmission
probabilities.
\bibitem{Switkes99a}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Switkes}},
\bibinfo{author}{\bibfnamefont{C.~M.} \bibnamefont{Marcus}},
\bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Campman}},
\bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Gossard}},
\bibinfo{journal}{Science} \textbf{\bibinfo{volume}{283}},
\bibinfo{pages}{1905} (\bibinfo{year}{1999}).
\bibitem{Altshuler99a}
\bibinfo{author}{\bibfnamefont{B.~L.} \bibnamefont{Altshuler}},
\bibinfo{author}{\bibfnamefont{L.~I.}
\bibnamefont{Glazman}}, \bibinfo{journal}{Science}
\textbf{\bibinfo{volume}{283}}, \bibinfo{pages}{1864} (\bibinfo{year}{1999});
\bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Zhou}},
\bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Spivak}},
\bibinfo{author}{\bibfnamefont{B.~L.} \bibnamefont{Altshuler}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{82}},
\bibinfo{pages}{608} (\bibinfo{year}{1999});
\bibinfo{author}{\bibfnamefont{I.~L.} \bibnamefont{Aleiner}},
\bibinfo{author}{\bibfnamefont{A.~V.} \bibnamefont{Andreev}},
\bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{81}},
\bibinfo{pages}{1286} (\bibinfo{year}{1998});
\bibinfo{author}{\bibfnamefont{Q.}~\bibnamefont{Niu}}, \bibinfo{journal}{Phys.
Rev. Lett.} \textbf{\bibinfo{volume}{64}}, \bibinfo{pages}{1812}
(\bibinfo{year}{1990}).
\bibitem{Wagner95a}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Wagner}},
\bibinfo{journal}{Phys. Rev. A} \textbf{\bibinfo{volume}{51}},
\bibinfo{pages}{798} (\bibinfo{year}{1995}).
\bibitem{Hekking91a}
A similar system has been studied by
\bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Hekking}},
\bibinfo{author}{\bibfnamefont{Y.~V.} \bibnamefont{Nazarov}},
\bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{44}},
\bibinfo{pages}{9110} (\bibinfo{year}{1991}).
\bibitem{Wagner97b}
\bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Wagner}},
\bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Zwerger}},
\bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{55}},
\bibinfo{pages}{10217R} (\bibinfo{year}{1997}).
\bibitem{footnote3}
Eq.~(\protect\ref{eq:T01}) is almost identical to Eq.~(4) of
Ref.~[\protect\onlinecite{Hekking91a}] which was derived within the WKB
approximation, except for the fact that their result does not depend
on the wave vector $k_n$ of the outgoing channels. Summing over all
channels $n$, Hekking {\it et al.} thus find the total transmission
probability, in the absence of Pauli blocking factors, not to depend
on the phase shift $\phi$ at all, and they conclude that there is no
pump current in this case. This is, however, not a conclusive result,
as it simply reflects their assumption of {\it unitary} total
transmission probability even in the presence of the driving field.
\bibitem{footnote4}
This linear relationship has been said to be a fingerprint of
adiabatic pumping, but as seen here photon-assisted tunneling
yields the same dependency.
\bibitem{footnote5}
To avoid spurious low-energy resonance effects in the
dot (which would not exist had we included potential fluctuations)
we impose a lower cutoff to the energy integration in
Eqs.~(\protect\ref{eq:Tsu-Esaki-Current}) and
(\protect\ref{eq:Pauli-Current}) at 1\,GHz. Changing this cutoff leads
to small quantitative but no qualitative changes in our results.
\end{thebibliography}