1: %\documentclass[prl,twocolumn,showpacs]{revtex4}
2: %\usepackage{epsfig}
3:
4: \documentclass[superscriptaddress,twocolumn,showpacs,preprintnumbers,prb]{revtex4}
5: \usepackage{graphicx}
6: \usepackage{times}
7: \usepackage{epsfig}
8:
9:
10: \begin{document}
11: \title{Unexpected Conductance Dip in the Kondo Regime \\
12: of Linear Arrays of Quantum Dots}
13: \author{C.A. B\"usser, Adriana Moreo and Elbio Dagotto}
14: \affiliation{National High Magnetic Field Lab and Department of Physics,
15: Florida State University, Tallahassee, FL 32306}
16: \date{\today}
17:
18: \begin{abstract}
19: Using exact-diagonalization of small clusters and Dyson equation
20: embedding techniques, the conductance $G$
21: of linear arrays of quantum dots is investigated. The Hubbard interaction
22: induces Kondo peaks at low temperatures for an odd number of dots.
23: Remarkably, the Kondo peak is split in half
24: by a deep minimum, and the conductance
25: vanishes at one value of the gate voltage. Tentative explanations for
26: this unusual effect are proposed, including an
27: interference process between two channels contributing to $G$, with one more
28: and one less particle than the exactly-solved
29: cluster ground-state. The Hubbard interaction
30: and fermionic statistics of electrons also appear to be
31: important to understand this phenomenon. Although most of the calculations
32: used a particle-hole symmetric Hamiltonian and formalism,
33: results also presented
34: here show that the conductance dip exists even when this symmetry is
35: broken. The conductance
36: cancellation effect obtained using numerical techniques
37: is potentially interesting, and other many-body techniques should be used to
38: confirm its existence.
39: \end{abstract}
40:
41: %%%%%%%%%%%%%%%%%%%%%%%%%%% Pacs %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
42: \pacs{73.63.Kv,73.23.-b,73.21.La,75.75.+a}
43:
44: \maketitle
45:
46: %%%%%%%%%%%%%%%%% Main text begins here %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
47:
48:
49: %-----------------------------------
50:
51: \section{Introduction}
52:
53: The possibility of destructive interference between
54: two or more wave functions is among the most remarkable phenomena
55: predicted by quantum mechanics. The effect can be observed when
56: electronic beams are split and then brought together after traveling
57: paths of different lengths, or in a Aharonov-Bohm (AB) geometry -- such
58: as a ring -- where two equal-length paths nevertheless can carry
59: different phase factors in the presence of a magnetic flux.
60: Recent advances in nanotechnology have made possible the
61: fabrication of quantum dots \cite{kastner} -- analogous to artificial
62: atoms or molecules -- where these effects can be tested.
63: In fact, the AB effect was recently observed using a quantum
64: dot embedded in a ring in the Coulomb blockade regime \cite{AB}.
65: Another example of conductance cancellations are the
66: well-known Fano resonances\cite{fano} that occur
67: when charge can circulate through two paths:
68: one with a discrete level and the other with a continuum of states.
69: Many physical realizations of Fano resonances are known. For example,
70: when an atom is deposited on a metallic surface,
71: a scanning tunneling microscope (STM) tip probes transmission to the tip either
72: through the atom or directly from the surface, leading to current
73: cancellations \cite{tip}. In addition, theoretical studies predict that
74: Fano resonances should appear in ``T-shaped'' geometries
75: where an active dot -- connected to
76: left and right electrodes -- is also side-connected
77: to another dot \cite{selman1}.
78: A similar conductance cancellation
79: has been predicted using double quantum-dot
80: molecules attached to leads \cite{claro} and in a
81: 2$\times$2 quantum-dot array \cite{claro2}. All these cancellations
82: are caused by destructive interference among two different
83: paths between conductors. Related cases
84: correspond to multiple-level dots with noninteracting
85: electrons \cite{selman2}, which can also be
86: rephrased as a many-dot problem connected to leads in ring-like geometries,
87: leading naturally to conductance cancellations.
88:
89: \begin{figure}
90: \begin{center}
91: \epsfig{file=sistema.eps,width=8.0cm}
92: \end{center}
93: \caption{Schematic geometry and hopping amplitudes
94: of the quantum-dot linear array studied here. The exactly-solved
95: cluster -- with $n_{\rm cl}$=9 sites and $N$=3 dots in this example --
96: includes some sites of the leads.
97: }
98: \label{Fig1}
99: \end{figure}
100:
101: It is the purpose of this paper to report an unexpected
102: conductance cancellation found in linear arrays of $N$ quantum
103: dots ($N$ odd and keeping only
104: one level per dot). These arrays do not have any obvious
105: real-space paths
106: that may lead to explicit AB or Fano interferences to rationalize the
107: conductance zeros found here -- all electrons travel through the same
108: chain -- and in this respect the interference is exotic.
109: In addition,
110: the reported cancellations occur only in the presence
111: of Coulomb interactions, and in the previously believed to be well-understood
112: Kondo regime \cite{glazman}. In fact, {\it the Kondo peaks are here found to be
113: split in half}, a curious effect that may be observable experimentally
114: at very low temperatures.
115: Although transport through many quantum dots\cite{coupled}
116: or STM-engineered atomic
117: systems\cite{crommie} has received
118: considerable attention in recent years, experiments
119: using 3 dots or atoms have not been sufficiently accurate to address the
120: effect found in this paper.
121:
122:
123: Note that conductance dips have also been observed experimentally
124: and theoretically
125: at high magnetic fields in a two-level single dot \cite{S-T1,S-T2}.
126: This effect was ascribed to transitions between total spin S=1 and S=0
127: dot states, a situation that does not seem to apply to our case with
128: an odd number of electrons in the ground state \cite{oguri}.
129:
130:
131: The paper is organized as follows. In Sec.\ref{sec:model}, the
132: Hamiltonian and many-body technique are described. In Sec. \ref{sec:results}
133: the main results are presented, with emphasis on the conductance
134: cancellation for an odd number of dots. The case of an even number of
135: dots is also described, and in this situation there is no cancellation.
136: The dependence of the results with parameters in the model is presented
137: in this section as well.
138: In Sec.\ref{sec:1dot}, results for 1, 4, and 5 dots
139: are briefly described. In Sec.\ref{sec:explanations}, possible explanations
140: of the conductance
141: dip effect are presented. They include interference between conduction
142: processes with one more and one less particle in the cluster, as well
143: as mappings into systems with the T-shape geometry that are known
144: to lead to interference.
145: In Sec.\ref{sec:conclusions}, conclusions are
146: presented.
147:
148: %Regarding the theoretical aspects of the problem,
149: %it would be important
150: %to carry out further studies of arrays of dots using many-body techniques
151: %different from those employed in the present manuscript
152: %in order to confirm the subtle
153: %conductance cancellation found in our investigations.
154:
155:
156: Throughout the paper it is emphasized that confirmation
157: of our results using other
158: techniques is important. Although the numerical
159: studies presented below do not seem to be severely affected by size effects,
160: size dependences are sometimes very subtle. Thus, further work is
161: needed to confirm the exotic conductance dip found here numerically.
162: If this confirmation occurs, the effect unveiled
163: in the present investigations -- an unexpected quantum interference process
164: in linear chains of dots -- should be searched for experimentally. The
165: effort should be carried out at
166: sufficiently low temperatures such that the dip structure becomes
167: visible.
168:
169:
170: \section{Model and Technique} \label{sec:model}
171:
172: Keeping one level per dot, the Hamiltonian for our $N$-dot system coupled
173: to leads is
174: $H$=$H_{\rm dots}$+$H_{\rm leads}$+$H_{\rm int}$, where
175: \begin{equation}
176: H_{\rm dots} = -t^{''} \sum_{i\sigma}(c^\dagger_{i\sigma} c_{i+1\sigma}
177: + h.c.)+ U\sum_{i} n_{i\uparrow} n_{i\downarrow},
178: \label{Eq1}
179: \end{equation}
180: \noindent represents the electronic hopping and Hubbard interaction
181: in the dots subsystem ($i$ labels the dots) using a standard notation.
182: A gate voltage
183: $V_{\rm g}\sum_{i\sigma} n_{i\sigma}$ of equal strength for
184: the $N$ dots is also included. The term $H_{\rm leads}$ represents
185: the non-interacting electrons in the leads, with a nearest-neighbors
186: hopping amplitude $t$, while $H_{\rm int}$ is the hopping from the
187: ideal leads to the dots and its amplitude is $t^{'}$. Figure \ref{Fig1}
188: illustrates the geometry used in the study and conventions followed.
189: The Hamiltonian discussed here becomes particle-hole symmetric for
190: the case $V_{\rm g}$=$-U/2$, precisely the gate voltage needed for
191: the conductance cancellation reported below. However,
192: in subsection \ref{subsec:symmetries} other less symmetric models
193: were studied as well, and the zero in the conductance survives. Thus,
194: the dip reported in this paper does not seem to originate from a highly
195: symmetric Hamiltonian but its origin is more robust.
196:
197:
198:
199:
200: The zero-temperature, $T$=0, Green function $G_{\rm LR}(\omega)$ to
201: transfer charge from sites L to R (Fig.\ref{Fig1}) can be obtained by an
202: exact-diagonalization (Lanczos)
203: solution\cite{Elbio} of a cluster with $n_{\rm cl}$ sites
204: containing the $N$ dots. The exact
205: information about the cluster under study is supplemented by an
206: embedding procedure between the leads, already
207: discussed in previous literature \cite{meth1,meth2}.
208: To reproduce the one-dot Kondo effect it is crucial that the
209: exactly-solved cluster contains also a small
210: portion of the lead\cite{meth2}, assumed also
211: in a linear arrangement for
212: simplicity. The cluster size $n_{\rm cl}$ is chosen such that
213: $n_{\rm cl}$=$N$+2$n_{\rm odd}$, where $n_{\rm odd}$=1,3,5,...
214: As discussed before \cite{meth2},
215: with this convention the portion of the leads in the cluster contains
216: a zero-energy state that induces the Kondo effect already at the cluster
217: level, reducing finite-size effects.
218: The rest of the contacts is
219: incorporated using the Dyson equation
220: $\hat{G}$=$\hat{g} + \hat{g} ~\hat{t} ~\hat{G}$,
221: where $\hat{g}$ is the exactly-known
222: Green function matrix of the cluster, $\hat{G}$
223: is the dressed Green-function matrix across the cluster from L to R,
224: and $\hat{t}$ is the matrix of hopping elements
225: connecting the cluster and leads.
226: %%
227: In the present study, the total
228: $z$-component of the spin is either 1/2 or -1/2
229: for an odd number of sites in
230: the exactly-solved cluster. To respect particle-hole symmetry at
231: every step in the calculation, the cluster ground-state is here
232: taken as the sum (divided by $\sqrt{2}$) of the ground states of the
233: subspaces with total spin $z$-component 1/2 and -1/2.
234: This leads to Green functions for the ``up'' and ``down'' spins
235: that are identical.
236: Other conventions, discussed in the Appendix, lead to qualitatively
237: similar results regarding the presence of internal structure in
238: the conductance Kondo peak.
239: %%
240:
241: To consider charge fluctuations, the cluster Green functions
242: $\hat{g}_m$ for $m$ and $m$+1
243: electrons are combined. The mixed Green function $\hat{g}$
244: is written as $\hat{g}$=$(1-p) \hat{g}_m + p~\hat{g}_{\rm m+1}$.
245: With the dressed Green function $\hat{G}$ from the Dyson eq.,
246: the total cluster charge is obtained
247: as $Q$=$-1/\pi \int_{-\infty}^{E_{\rm F}}
248: \sum_j \mbox{Im} G_{jj}(\omega)$ (the sum in $j$ runs over the
249: cluster sites and $E_{\rm F}$ is the Fermi energy, assumed 0 in the
250: numerical calculations discussed below).
251: On the other hand, the charge at the cluster
252: in the mixed ($m$/$m$+1) state is
253: $q$=$(1$-$p)m$+$p(m$+$1)$ and, then, $p$ can be
254: found self-consistently to satisfy
255: $Q$=$q$ (in the region emphasized in the next section, with a $G$ cancellation,
256: $q$$\sim$$n_{\rm cl}$).
257: Finally, using the Keldish formalism
258: the conductance $G$ is written as \cite{meth2}
259: ${G}$=$(e/h)^2 t^2 |G_{\rm LR}(E_{\rm F})|^2 [\rho_{\rm leads}(E_{\rm F})]^2$.
260: The leads density-of-states (DOS) is $\rho_{\rm leads}(\omega)$, assumed here
261: to be a semi-circle from -2$t$ to 2$t$ (the results are weakly
262: dependent on this assumption).
263:
264: Previous studies showed that this formalism -- combination
265: of exact-diagonalization and embedding -- is sufficiently powerful to
266: reproduce the Kondo effect in electronic transport
267: across one dot \cite{meth2}, and for this reason there is no {\it a priori}
268: reason why it would fail for more dots. Nevertheless, to be cautious
269: in our discussion below statements are included to alert the reader on
270: possible subtle size effects that could alter our conclusions.
271: %
272: As already mentioned,
273: in the Appendix alternative conventions to our choice of spin
274: quantum numbers for the exactly-solved cluster ground state are also
275: discussed. These alternative conventions also lead to dips in the conductance
276: Kondo peak, as in the results presented in the following sections.
277: %
278: Although more work is needed to confirm the existence of the
279: conductance dip reported below,
280: the numerical results appear to be sufficiently robust that the effect
281: could even be observed experimentally at very low temperatures.
282:
283:
284: %--------------------------------------------------------
285: \begin{figure}
286: \begin{center}
287: \epsfig{file=F2-1.eps,width=8.0cm}
288: \end{center}
289: \caption{Conductance (in units of $e^2/h$) across an array of quantum dots
290: vs. $V_{\rm g}/U$, illustrating
291: the cancellation reported in this paper. The couplings are
292: $U/t$=1 and $t^{'}/t$=0.3 (a) corresponds to $N$=3 dots,
293: at $t^{''}/t$=0.2.
294: The solid (dashed) lines corresponds to an exactly-solved
295: cluster of $n_{\rm cl}$=5 (9) sites. The size dependence suggests
296: that this effect will survive the bulk limit.
297: (b) is for 2 dots ($n_{\rm cl}$=4),
298: same $t^{''}/t$ as in (a). (c-f) are results
299: in a wider range of $V_{\rm g}/U$ and varying $t^{''}$. For
300: very small $t^{''}$, (c), the central dot is virtually decoupled and
301: no Kondo effect is observed in the scale used. With increasing
302: $t^{''}$ a central peak is found, always split as in (a).
303: Applying the method outlined in the text to a cluster that only
304: has the dots (no extra lead sites), and then
305: incorporating the effect of the leads through the Dyson equation,
306: the Kondo peak is effectively eliminated (the exactly-solved cluster
307: does not have states near the leads Fermi energy). By this
308: procedure, just the Coulomb blockade peaks are found, roughly
309: representing the high-temperature solution of the problem. This result
310: is shown with dashed lines in (d) for completeness.
311: }
312: \label{Fig2}
313: \end{figure}
314:
315: \section{Results} \label{sec:results}
316:
317: \subsection{Conductance Dip for an Odd Number of Dots}
318:
319: The technique described in the previous paragraph was applied here to
320: the case of $N$$>$1 quantum dots forming a linear array. Our
321: original motivation was the study of transport in the regime of
322: large $t^{''}$ where $N$ odd (even) would lead to a quantum-dot
323: subsystem with spin 1/2 (0) and, as a consequence, the presence (absence) of
324: the Kondo effect as indeed occurs. However, studies at
325: intermediate couplings and hoppings regimes led to surprises.
326: The most unexpected
327: result of the present effort is shown in Fig.~\ref{Fig2}a where the
328: conductance across 3 dots is shown for $V_{\rm g}$ near $-U/2$ (inducing
329: one electron per dot), at relatively small $t^{''}$.
330: The shape of the broad peak (without the dip) resembles previous
331: Kondo-like results for one dot\cite{meth2}. Following standard
332: arguments, this Kondo effect is obtained
333: when the state of $N$ (odd) electrons carrying a net spin couples
334: to the leads \cite{glazman,meth2}.
335: However, the peak is found to be
336: split in half by an unexpected
337: zero in the conductance at exactly $V_{\rm g}$=$-U/2$.
338: This cancellation is absent at $U$=0, where
339: $G$/$(e^2/h)$=1 at $V_{\rm g}$=0 since the exactly-solved cluster has a
340: zero-energy state aligned with the leads Fermi-energy (assumed at 0).
341: As $U$ increases,
342: Coulomb blockade and Kondo peaks are generated, but
343: the latter is always split
344: by a zero at $V_{\rm g}$=$-U/2$. As a consequence, the effect
345: appears to
346: originate in correlation effects induced by a nonzero $U$.
347: Our study for
348: increasing $N$ suggests that the effect is present for any odd $N$, while
349: for $N$ even (2 as example, Fig.~\ref{Fig2}b) there is no conductance
350: cancellation
351: (the two peaks in Fig.~\ref{Fig2}b
352: are related to the Kondo splitting -- without a zero -- of double
353: quantum dots, previously discussed \cite{meth2,split}).
354: The $t^{''}$-dependence shown in Fig.~\ref{Fig2}c-f suggests
355: that there is an intermediate hopping range where the zero conductance
356: effect could be observed\cite{comment3}, while
357: both at very small and large $t^{''}$, its experimental observation will be
358: difficult.
359:
360: The conductance cancellation Fig.~\ref{Fig2}a is unexpected
361: since there are no obvious real-space multiple paths that can
362: lead to interference. Electrons here travel through a one
363: dimensional geometry. Note also that one level per dot is kept
364: in our analysis, and cancellations as in Ref.\onlinecite{selman2} are not
365: obviously present.
366:
367:
368: To gain insight on the origin of
369: the reported
370: phenomenon, Fig.~\ref{Fig3} shows the real (Re) and imaginary (Im)
371: components of the {\it cluster} Green function from one extreme
372: of the cluster to the other (denoted
373: $g_{\rm cl}$), for 2 and 3 dots. The overall
374: conductance emerges
375: from the behavior of $g_{\rm cl}$ at $\omega$=0, in the embedding procedure.
376: Clearly, the results for 2 and 3 dots
377: have different symmetry properties under $\omega$$\rightarrow$$-\omega$:
378: while for 2 dots Re($g_{\rm cl}$) is even, for 3 dots it is odd generating
379: a zero at $\omega$=0. Since both imaginary parts cancel at $\omega$=0,
380: then Re($g_{\rm cl}$) and Im($g_{\rm cl}$)
381: are zero (Re nonzero) for odd (even) number of dots
382: (this rule was verified numerically beyond the 2- and 3-dots example shown).
383: If $g_{\rm cl}$=0, the Dyson-equation embedding procedure cannot generate
384: a nonzero conductance.
385: If $U$$\rightarrow$0, the two peaks closest to $\omega$=0
386: in Fig.~\ref{Fig3} (right upper panel)
387: merge and the cancellation does not occur.
388:
389: \begin{figure}
390: \begin{center}
391: \epsfig{file=F3-1.eps,width=8.0cm}
392: \end{center}
393: \caption{Real and imaginary parts of the
394: cluster Green function
395: $g_{\rm cl}$
396: (from the first to the last site of the cluster) used to calculate $G$
397: through the Dyson equation. $\omega$ is in units of $t$.
398: Shown are exact results for $N$=2 dots ($n_{\rm cl}$=4 cluster)
399: and $N$=3 dots ($n_{\rm cl}$=5
400: cluster). The couplings are as in Fig.~\ref{Fig2}a. Note that
401: for 3 dots the real part vanishes at $\omega$=0, while for 2 dots
402: it is finite. The different behavior under $\omega$$\rightarrow$$-\omega$
403: for odd and even number of dots causes the cancellation of the conductance
404: of the former, discussed in the text.}
405: \label{Fig3}
406: \end{figure}
407:
408: \subsection{Analysis of Size Effects}
409:
410: An important aspect of the methodology discussed here, and in
411: previous literature, involves the exact solution of a cluster
412: followed by an embedding procedure. From the cluster size
413: dependence it is possible to infer whether a particular feature
414: under study will survive the bulk limit or not. Unfortunately, the
415: CPU time rapidly grows with the cluster size since the cluster
416: Green functions at all distances are needed for the Dyson equations,
417: and each Green function is calculated with approximately
418: one hundred steps in the continued-fraction procedure \cite{Elbio}.
419: This limits our detailed study of the conductance dip to clusters with
420: 5 and 9 sites (while a few values of the gate voltage can still be investigated
421: using 13 sites). In Fig.\ref{Fig7}, results for $n_{\rm cl}$=5 and 9
422: are presented at a small value of $t^{''}$. This small hopping was used
423: to amplify the region where the dip dominates (as $t^{''}$/$t$ increases,
424: the dip width is reduced as shown in Fig.\ref{Fig2}c-f). The results in
425: the figure show that the dip {\it survives} the increase of the cluster size,
426: and the maximum in the conductance actually is located even further away
427: from $V_{\rm g}$/$U$=-0.5 as the $n_{\rm cl}$ grows. While this is not
428: a definite proof, it is strongly suggestive that the conductance dip is
429: not an artifact of the many-body procedure and cluster-size used, but
430: it may be a real effect present in the bulk. Nevertheless, it is desirable
431: to have independent tests of our results using other many-body methods
432: to fully confirm our conclusions.
433:
434: \begin{figure}
435: \begin{center}
436: \epsfig{file=cnvg.eps,width=8.0cm}
437: \end{center}
438: \caption{
439: Conductance at $U$=1, $t$=1, and $t^{'}$/$t$=0.3, as used in
440: Fig.\ref{Fig2}a, and for $t^{''}$/$t$=0.075. Results for 3 dots and
441: cluster sizes $n_{\rm cl}$ of 5 and 9 sites are indicated. The maximum in the
442: conductance does not seem to move toward $V_{\rm g}$/$U$=-0.5 as $n_{\rm cl}$
443: grows, suggesting that the dip will survive the bulk limit.
444: %$~~~OPTIONAL$ A few results for 13 sites are also indicated.
445: }
446: \label{Fig7}
447: \end{figure}
448:
449:
450: \subsection{Survival of the Dip Reducing the Symmetries of the Hamiltonian}
451: \label{subsec:symmetries}
452:
453: In real quantums dots, the electron-hole symmetry of the Hamiltonian
454: used in previous sections cannot be achieved
455: since the Fermi level lies just a small fraction of eV
456: above the bottom of the conduction band.
457: However, it is always possible to find a gate potential where the main
458: levels involved -- $V_{\rm g}$ and $V_{\rm g}$+$U$ -- are symmetrically
459: located around $E_F$.
460: Since all the energy scales important for the Kondo effect ($U$ and
461: $t^{'}$) are of the order of meV \cite{Golhaber2} these levels are
462: very close of $E_F$, achieving an approximate
463: electron-hole symmetry.
464:
465: In addition to the previous argument, it is possible to repeat the
466: calculations presented before for cases where the particle-hole
467: symmetry is not present. This can be achieved, for instance, by merely
468: adding small random components to the on-site Hubbard energies
469: and, in addition, introducing small site energies. To further break symmetries
470: of the problem, the hopping $t^{'}$ from the 3-dot region to the rest can also
471: be made different on the right and the left, and even the two internal
472: dot-dot hopping amplitudes $t^{''}$ can be made different as well.
473: One representative result of this study
474: is shown in Fig.\ref{Fig8}. It is interesting to observe that the
475: dip in the conductance survives the breaking of symmetries in the model.
476: As a consequence, the effect appears to be robust and independent of fine
477: details in the analysis.
478:
479: \begin{figure}
480: \begin{center}
481: \epsfig{file=pertub.eps,width=8.0cm}
482: \end{center}
483: \caption{Conductance for the 3-dot system, with 5 sites in the
484: exactly-solved cluster (solid line). In this figure
485: the on-site Hubbard $U$ couplings, as well
486: as the on-site energies $\epsilon$, of the three interacting sites
487: are given random values of amplitude 0.01$t$ in addition
488: to the uniform values used in Fig.\ref{Fig2} (namely $U$=1 and $\epsilon$=0).
489: The hopping amplitudes $t^{'}$ and $t^{''}$ are also varied
490: as follows: from the left lead to the first dot $t^{'}$ is 0.3 (in
491: units of $t$), the next hopping amplitude is $t^{''}$=0.2, then
492: $t^{''}$=0.12, and finally $t^{'}$=0.35 for the connection between the
493: last dot and the
494: right lead. It is observed that the vanishing of the conductance still
495: occurs although with a small shift in its position.
496: For comparison, the result of Fig.\ref{Fig2}a in the particle-hole symmetric
497: case is also shown (dashed line).
498: }
499: \label{Fig8}
500: \end{figure}
501:
502:
503: %----------------------------1dot----------------------------
504:
505: \section{Results for Number of Dots 1, 4, and 5} \label{sec:1dot}
506:
507: The case of one dot is special.
508: If a ${n_{\rm cl}}$ cluster with an odd number of sites
509: (e.g., +-o-+) is solved exactly, the degeneracy
510: between ${n_{\rm cl}}$-1, ${n_{\rm cl}}$, and ${n_{\rm cl}}$+1
511: {\it remains} even for nonzero $U$.
512: This is an accidental degeneracy that avoids the dip splitting of the
513: Kondo peak found with 3 or more (odd) dots. However, by simply adding
514: on-site energies $\delta$ and $-\delta$ at the first and last sites
515: of the cluster, the accidental degeneracy at $U$$\neq$0
516: is removed and now a conductance cancellation
517: occurs as in the other cases (Fig.~\ref{Fig4}a) \cite{comment4}. The
518: dip phenomenon appears to be general and robust.
519: %
520: %
521: \begin{figure}
522: \begin{center}
523: \epsfig{file=F4-1.eps,width=8.0cm}
524: \end{center}
525: \caption{
526: (a) $G$/$(e^2/h)$ vs. $V_{\rm g}/U$ for
527: 1 dot, $t^{'}/t$=0.3, introducing on-site energies
528: $\pm\delta$=0.1 (see text). (b-c) are results for
529: clusters with 5 and 4 dots, showing a
530: rich structure. Parameters are as in Fig.~\ref{Fig2}a.
531: }
532: \label{Fig4}
533: \end{figure}
534:
535: Our study also extended beyond the 2- and 3-dot cases. For example,
536: Fig.~\ref{Fig4}b illustrates results obtained for 5 dots (7-site cluster).
537: Here, once again, $G$=0 at $V_{\rm g}$=$-U/2$. In addition, a rich
538: structure if observed at higher frequencies with multiple conductance
539: cancellations that resemble Fano resonances. Their origin is similar
540: to those discussed for 3 dots and emerge from cancellations between
541: ``competing'' poles at close distance in the Green function. Even
542: 4 dots (Fig.~\ref{Fig4}c) shows a highly nontrivial structure, also
543: with cancellations away from $V_{\rm g}$=$-U/2$. The
544: richness unveiled in the conductance properties of linear-dot chains
545: once unbiased accurate many-body techniques are used is
546: remarkable.
547:
548:
549: %----------------------------------------
550:
551: \section{Possible Explanations of Conductance Cancellation} \label{sec:explanations}
552:
553: \subsection{Interference Between States with ${n_{\rm cl}\pm 1}$ particles}
554:
555: The results of the previous paragraph suggest that simple
556: symmetry arguments involving just two states of the
557: entire Hilbert space -- those closest to $\omega$=0 -- should
558: be sufficient to understand the effect. With this in mind, consider
559: the $n_{\rm cl}$-site-cluster Green-function of interest expanded
560: in the basis of Hamiltonian eigenstates as
561: \begin{eqnarray}
562: & g_{\rm cl}(\omega) = \sum_l
563: {{\langle 0,{n_{\rm cl}} |c_1|l,{n_{\rm cl}}+1\rangle
564: \langle l,{n_{\rm cl}}+1 |c^\dagger_{n_{\rm cl}}|0,{n_{\rm cl}}\rangle}\over{\omega + E_l -E_0 +i\epsilon}}
565: \nonumber \\
566: &+ \sum_m{{\langle 0,{n_{\rm cl}} |c^\dagger_{n_{\rm cl}}|m,{n_{\rm cl}}-1\rangle
567: \langle m,{n_{\rm cl}}-1 |c_1|0,{n_{\rm cl}}\rangle}\over{\omega + E_0 -E_m +i\epsilon}},
568: \label{Eq2}
569: \end{eqnarray}
570: where $\epsilon$$\rightarrow$0 (10$^{-7}$ in practice, and 10$^{-2}$ for
571: the DOS).
572: In $|j,n\rangle$, $j$ labels states
573: of $n$ particles.
574: The Hubbard Hamiltonian is particle-hole (p-h) symmetric if
575: $V_{\rm g}$=$-U/2$. The explicit p-h transformation is
576: $c^\dagger_{i\sigma}$$\rightarrow$$(-1)^{i}{c}_{i\sigma}$.
577: The empty state is mapped into the fully occupied state.
578: It can be shown that
579: $\langle 0,{n_{\rm cl}} |c_1|l,{n_{\rm cl}}+1\rangle$$\rightarrow$$(-1)^p \langle m,{n_{\rm cl}}-1 |c_1|0,{n_{\rm cl}}\rangle$
580: for ${n_{\rm cl}}$=$2p$+$1$ or ${n_{\rm cl}}$=2$p$, with $p$=integer. In addition,
581: $\langle l,{n_{\rm cl}}+1 |c^\dagger_{n_{\rm cl}}|0,{n_{\rm cl}}\rangle$$\rightarrow$
582: $(-1)^{r}\langle 0,{n_{\rm cl}} |c^\dagger_{n_{\rm cl}}|m,{n_{\rm cl}}-1\rangle$,
583: where $r$=$p$ if ${n_{\rm cl}}$=$2p$+$1$, and $r$=$p$+$1$ if ${n_{\rm cl}}$=$2p$.
584: Isolating a pair of states $|l,{n_{\rm cl}}+1\rangle$ and $|m,{n_{\rm cl}}-1\rangle$
585: connected by the p-h transformation -- namely with equal energies
586: relative to $E_0$ --
587: this leads to a simple contribution to $g_{\rm cl}(\omega)$
588: of the form
589: $[{{AB}/(\omega+\omega_l+i\epsilon)}] \pm
590: [{{AB}/(\omega-\omega_l+i\epsilon)}]$,
591: where $A,B$ are numbers and $\omega_l$=$E_l$-$E_0$, assumed nonzero.
592: The $+$ $(-)$
593: sign corresponds to an {\it odd} ({\it even})
594: number ${n_{\rm cl}}$ of cluster sites. Clearly, for ${n_{\rm cl}}$=odd,
595: $Re[g_{\rm cl}(\omega)]$ is odd under
596: $\omega$$\rightarrow$$-\omega$, and, thus, it cancels at $\omega$=0.
597: For the other case, ${n_{\rm cl}}$=even, there is no cancellation since
598: the real part is even. Since the embedding process cannot generate
599: a nonzero conductance if the cluster has a vanishing Green function,
600: then the overall conductance is zero
601: for $V_{\rm g}$=$-U/2$ and ${n_{\rm cl}}$ odd.
602:
603: The proof of the results of the previous paragraph has been
604: mainly computational,
605: using the entire Hilbert space for small values of ${n_{\rm cl}}$. However,
606: a simpler qualitative understanding
607: can be obtained for example considering ${n_{\rm cl}}$=3 and using
608: ($|\uparrow \downarrow \uparrow \rangle$+
609: $|\downarrow \uparrow \downarrow \rangle$)/$\sqrt{2}$ as a
610: simplified ${n_{\rm cl}}$-particle
611: ground-state $|0,{n_{\rm cl}}\rangle$. For $| l,{n_{\rm cl}}+1\rangle$,
612: $c^\dagger_{-1,\uparrow}$$|0,{n_{\rm cl}}\rangle$ can be used,
613: and $c_{-1,\downarrow}$$|0,{n_{\rm cl}}\rangle$
614: for $| l,{n_{\rm cl}}-1\rangle$.
615: For these simplified states, it can be easily shown
616: that $|l,{n_{\rm cl}}+1\rangle$ transforms under
617: p-h to $|l,{n_{\rm cl}}-1\rangle$ for ${n_{\rm cl}}$=odd,
618: and to $-|l,{n_{\rm cl}}-1\rangle$ for ${n_{\rm cl}}$=even.
619: After simple algebra, recalling
620: that the matrix elements are real, and being careful with
621: the signs arising
622: from fermionic anticommutations, the p-h transforms of the matrix
623: elements are found,
624: completing the proof. This ${n_{\rm cl}}$=3 derivation can easily
625: be extended to arbitrary ${n_{\rm cl}}$.
626:
627:
628: %--------------------intuitive explanation ------------------
629:
630: The previous explanation of the anomalous zero conductance
631: emphasizes the competition, and eventual
632: interference, between two states that contribute to the
633: cluster Green function. The key aspect is the relative
634: sign of the matrix elements for the two poles, which leads
635: to interference for an odd number of cluster sites.
636: Let us discuss these aspects more intuitively,
637: and also explain why at $U$=0 the effect is not present.
638: Consider as example a 5-site cluster with 3 dots
639: (schematically +-o-o-o-+, o=dot, +=lead site). At $U$=0 this cluster, and any cluster with a total
640: number of sites odd and Hamiltonian Eq.~\ref{Eq1}, has a zero
641: energy eigenvalue. This implies a {\it degeneracy} between
642: the lowest-energy states with 4, 5, and 6 particles
643: (or ${n_{\rm cl}}$-1, ${n_{\rm cl}}$, ${n_{\rm cl}}$+1
644: particles for an ${n_{\rm cl}}$ (odd) cluster) since populating
645: the zero-energy state has no energy cost. With the Fermi energy
646: of the metal at 0 as well, there is a direct channel for conductance
647: through the dots and no cancellation. However, when $U$ is switched
648: on, the degeneracy is removed since there is a penalization
649: for having a site with two electrons (directly related to an empty site
650: by p-h symmetry). The doubly-occupied site, inevitable for ${n_{\rm cl}}$+1
651: electrons on ${n_{\rm cl}}$ sites, is located with more probability
652: outside the set of
653: $N$ dots, i.e. in the lead segments included in the exactly-solved cluster.
654: This produces a finite but small splitting $\Delta E$,
655: substantially smaller than $U$.
656: As $U$ increases, a Kondo peak is formed at $V_{\rm g}$=$-U/2$, as
657: previously discussed\cite{meth2}, but with a dip of width $\Delta E$
658: in the middle. Note that for ${n_{\rm cl}}$ even, there is no zero
659: in the cluster and no dip in the conductance. However,
660: states approach zero energy as ${n_{\rm cl}}$ (even)
661: increases and eventually
662: as ${n_{\rm cl}}$$\rightarrow$$\infty$ a common limit of zero conductance
663: for both ${n_{\rm cl}}$ odd and even is expected.
664:
665: The $G$ cancellation arises from the $U$-induced splitting of the
666: ${n_{\rm cl}}$+1 and ${n_{\rm cl}}$-1
667: states from the ${n_{\rm cl}}$ ground state. More intuitively,
668: for charge to transport through a cluster or molecule there are
669: two basic processes, that here interfere. In one case an electron
670: first jump to the cluster, leading to ${n_{\rm cl}}$+1 particles inside, and
671: then an electron exits. In the other case, first an electron leaves the cluster
672: (i.e (${n_{\rm cl}}$-1)
673: electrons in the cluster intermediate state), and then another
674: gets in. These two intermediate states corresponds to two ``paths''
675: in a quantum-mechanical formulation, and they do not need actual different
676: real-space trajectories to interfere with one another
677: (Fig.~\ref{Fig6b}).
678: %\cite{comment}.
679:
680: \begin{figure}
681: \begin{center}
682: \epsfig{file=nivs2.eps,width=8.0cm}
683: \end{center}
684: \caption{The two ``paths'' that lead to the conductance cancellation
685: involve intermediate states of ${n_{\rm cl}}$+1 and ${n_{\rm cl}}$-1 electrons.
686: }
687: \label{Fig6b}
688: \end{figure}
689:
690: \subsection{Two-Paths Interference in a One-dimensional Multidot
691: System}
692:
693: An alternative explanation of the conductance-cancellation effect
694: described in this paper is the following.
695: The ${n_{\rm cl}}$-site cluster, with ${n_{\rm cl}}$ odd,
696: has reflexion symmetry around
697: the central dot (here denoted by 0). This suggests a change of
698: basis defined by
699: $d_{\alpha i\sigma}$=$(c_{i\sigma}+c_{-i\sigma})/\sqrt{2}$
700: and
701: $d_{\beta i\sigma}$=$(c_{i\sigma}-c_{-i\sigma})/\sqrt{2}$,
702: where the sites $i$=1,2,... (-1,-2,...) are on the right (left)
703: of the central dot, which is left invariant by this transformation.
704: It can be shown that for just one dot, $N=1$, the system in the
705: new basis is equivalent to one-dot at the end of a semi-infinite
706: chain coupled to the $\alpha$ band, and decoupled from a $\beta$ band
707: (as sketched in Fig.~\ref{Fig5f}a, for $\delta$=0).
708: This geometry corresponds to a
709: one-channel Kondo problem, with a concomitant peak in electronic
710: transport.
711: In this new basis, the Green function used to calculate the
712: conductance can be written as
713: $G_{LR}(\omega)$=1/2($G_{\alpha\alpha}(\omega)$-$G_{\beta\beta}(\omega))$.
714: Then, an interference in the conductance occurs when
715: $G_{\alpha\alpha}(E_F)$=$G_{\beta\beta}(E_F)$.
716:
717: As discussed in Sec.\ref{sec:1dot},
718: consider now a diagonal energy $\delta$
719: at site $i$=+1 and $-\delta$ at $i$=-1
720: (i.e. immediately to the right and left of the active dot).
721: Transforming the operators, the previously decoupled
722: bands $\alpha$ and $\beta$
723: are now effectively connected by a hopping term of strength $2\delta$
724: (Fig.~\ref{Fig5f}a). In this representation, it is easy to visualize
725: a possible interference between processes involving
726: the direct hopping $\alpha$$\rightarrow$$\beta$ and those where jumping
727: to and from the dot is part of the path. This abstract-space representation
728: establishes a connection with the real-space $T$-geometry
729: interference previously discussed\cite{selman1}.
730:
731: \begin{figure}[h]
732: \begin{center}
733: \epsfig{file=figi-ch2.eps,width=5.0cm}
734: \end{center}
735: \caption{Illustration of the transformation to the $\alpha$-$\beta$
736: basis described in the text. (a) corresponds to
737: 1-dot with diagonal energies $\pm\delta$ in the
738: sites next to the dot. (b) corresponds to 3-dots.
739: }
740: \label{Fig5f}
741: \end{figure}
742:
743:
744: For $N$$>$1 the results are not as conclusive, but still suggestive.
745: Consider as an example $U$=0 and $N$=3. In this case,
746: the two channels $\alpha$ and $\beta$, each supplemented by
747: one dot, decouple from one another. The central dot is coupled to
748: the $\alpha$ channel (Fig.~\ref{Fig5f}b).
749: At $U$$\neq$0 a ``many-body'' coupling proportional to $U/2$
750: links the two channels. This
751: coupling -- too cumbersome to write here explicitly --
752: contains ``spin-flip'' contributions between $\alpha$
753: and $\beta$, density-density interactions, and even a two-electron
754: hopping term. The $T$-geometry appears once again, suggesting possible
755: interferences, but now with an
756: $\alpha$$\rightarrow$$\beta$ effective ``hopping''
757: which is very complicated. Nevertheless, this is sufficiently illustrative
758: since for $N$ even none of the two channels $\alpha$-$\beta$ have
759: an extra dot attached, and the entire system is effectively
760: linear with no
761: obvious sources of interference, as indeed observed numerically.
762: Then, {\it $N$ odd and even are fundamentally different
763: in this representation},
764: with the odd having possible sources of interference.
765:
766:
767:
768:
769: \section{Conclusions} \label{sec:conclusions}
770:
771: Using numerical techniques, in this paper it has been argued
772: that the conductance Kondo peak of an odd number
773: of quantum dots forming a linear array presents nontrivial internal structure in the form
774: of a dip. The calculation has passed many tests, but the authors
775: acknowledge that the reported result is quite unexpected and for
776: this reason other theoretical
777: techniques should be employed to test our predictions.
778: If the present results are confirmed in the near future,
779: the search for the ``Kondo dip'' in experiments should be carried out.
780: The experimental observation of the conductance dip
781: reported here may require considerable effort.
782: Realizations of the linear-array
783: geometry using atoms and employing STM techniques to measure conductances
784: are difficult. For instance,
785: atoms attract, and three of them on a metallic surface tend to form triangles
786: rather than chains \cite{crommie-private}. In addition, finite temperature
787: effects will likely tend to fill the dip in $G$, and
788: temperatures even lower than
789: usually employed will be needed to see the effect. At present,
790: the characteristic energy regulating the width of the dip is still
791: unknown since the method used in the paper works only at zero temperature.
792: In spite of these caveats, the
793: interference discussed here is sufficiently novel and interesting that its
794: experimental confirmation and theoretical extension to other types of
795: arrays should be actively pursued.
796:
797:
798: \section{acknowledgments}
799:
800: This work was supported by the NSF grants DMR-0303348 and DMR-0122523.
801: Conversations with Y. Meir, E. Anda, J. Verg\'es, G. Chiappe,
802: S. Ulloa, L. Glazman, and S. Hershfield are gratefully acknowledged.
803:
804: \section {Appendix: Results using other conventions}
805:
806: As discussed in Sec.\ref{sec:model}, the state representing the cluster
807: with an odd number of sites (and dots)
808: used in this effort is the equal-weight sum of the states with total
809: $z$-component of spin, $S^{\rm z}_{\rm tot}$,
810: equal to 1/2 and -1/2. By this procedure the
811: particle-hole symmetry is respected at every step in the calculation.
812: However, other conventions have been used in recent literature.
813: For example, related work in Ref.~\onlinecite{willy} considers only
814: the cluster state with $S^{\rm z}_{\rm tot}$=1/2. With this convention
815: the cluster Green functions for spins 1/2 and -1/2 are different
816: (although their sum is independent of the relative weights
817: of the two states). If these
818: ``up'' and ``down'' Green functions are independently
819: dressed through the Dyson formalism,
820: conductances for the ``up'' and ``down'' channels are obtained.
821: Individually, {\it each of these conductances contains a zero} quite similar
822: to the results shown in the bulk of the present paper. However, the location
823: of the zero is different for the two channels,
824: slightly shifted left and right
825: from $V_{\rm g}/U$=-0.5 (see Fig.\ref{Fig9b}) due to the breaking
826: of the particle-hole symmetry. When the two conductances
827: are added (and dividing by 2),
828: the zeros are no longer present since the ``up'' and ``down''
829: contributions have the cancellation at different gate voltages. However,
830: even accepting this symmetry-breaking alternative convention to carry out the
831: calculation, two dips are clearly found in the overall result, as also
832: shown in Fig.\ref{Fig9b}.
833: It is expected that the two procedures (with and without explicit particle-hole
834: symmetry in the calculation) will
835: lead to the same result in the bulk limit, and only further work
836: can clarify which convention is the best given the inevitable
837: size constraints of the present numerical
838: technique.
839:
840: %\epsfig{file=F9a.eps,width=7.0cm}
841:
842: \begin{figure}[h]
843: %\begin{top}
844: \epsfig{file=F9b.eps,width=7.0cm}
845: %\end{center}
846: \caption{
847: Conductance vs. gate voltage calculated using a 5 sites (3 dots)
848: cluster state
849: with total spin $z$-component equal to 1/2 (as opposed to the
850: equal-weight 1/2 and -1/2 used in the rest of the paper). Shown
851: with dashed lines are the dressed conductances (divided by 2) for the ``up''
852: and ``down'' channels, each showing a zero at values of the
853: gate voltage close to -0.5$U$. The solid line is the sum.
854: In the sum, two dips can be observed. Although the result is
855: quantitatively different
856: using one convention or the other for the spin of the cluster state,
857: the fact that the Kondo peak has internal structure in the form of
858: dips is qualitatively the same in both cases.
859: }
860: \label{Fig9b}
861: \end{figure}
862:
863:
864:
865: Finally, to avoid the cluster ground-state ambiguity
866: problem an alternative geometry sketched in the
867: inset of Fig.\ref{Fig10} can be used. In this new system, one extra dot is laterally
868: coupled to the central dot with a hopping amplitude $t^{'''}$.
869: Clearly, when $t^{'''} \to 0$ this system is equivalent to the 3 dots system
870: described in the bulk of the paper. For $V_{\rm g}/U$=-0.5
871: and one particle per site in the cluster, the
872: extra dot adds one electron to the system studied before
873: giving an $S^{\rm z}_{\rm tot}$=0 for the (nondegenerate)
874: ground state. This trick eliminates the ``up'' and ``down'' degeneracy
875: at all finite values of $t^{'''}$ (while at $t^{'''}$=0, the degeneracy is
876: recovered). Repeating the calculation as in Fig.\ref{Fig2}a, the
877: conductance for the new system once again presents a symmetric dip for all the
878: values of $t^{'''}$ investigated, even including the very small $t^{'''}$ regime
879: where the extra dot and linear array are nearly decoupled. Then, once again
880: it is concluded that different procedures to carry out the
881: calculations lead all to the same qualitative conclusions. Both Figs.\ref{Fig9b}
882: and \ref{Fig10} show that the main point of the present paper remains the same irrespective
883: of the convention: The Kondo peak of quantum-dot arrays with
884: an odd number of dots appears to have
885: internal structure in the form of dips as the gate voltage is varied.
886:
887:
888: \begin{figure}[h]
889: %\begin{top}
890: \epsfig{file=F10.eps,width=7.0cm}
891: %\end{center}
892: \caption{
893: Conductance vs. gate potential for the $T$-geometry
894: system (shown in the inset) proposed to avoid the ground-state
895: degeneracy at $V_{\rm g}/U$=$-0.5$. The parameters $U$, $t$,
896: $t^{'}$, and $t^{''}$
897: are the same as in Fig.~\ref{Fig2}a. The point line corresponds
898: to results at $t^{'''}/t^{''}$=0.005,
899: the dashed line at $t^{'''}/t^{''}$=0.5, and the solid line
900: at $t^{'''}/t^{''}$=1. The dip is present in all cases.
901: }
902: \label{Fig10}
903: \end{figure}
904:
905:
906:
907:
908: %%%%%%%%%%%%%%%%%%%%%%%% References %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
909: %\begin{references}
910: \begin{thebibliography}{99}
911:
912: \bibitem{kastner} M. A. Kastner, Rev. Mod. Phys. {\bf 64}, 849 (1992);
913: W. G. van der Wiel, S. De Franceschi, J. M. Elzerman, T. Fujisawa,
914: S. Tarucha, and L. P. Kouwenhoven, Rev. Mod. Phys. {\bf 75}, 1 (2003).
915:
916:
917: \bibitem{AB}
918: A. Yacoby, M. Heiblum, D. Mahalu, and Hadas Shtrikman,
919: Phys. Rev. Lett. {\bf 74}, 4047 (1995).
920:
921: \bibitem{fano} U. Fano, Phys. Rev. {\bf 124}, 1866 (1961).
922:
923: \bibitem{tip}
924: V. Madhavan, W. Chen, T. Jamneala, M. F. Crommie, and N. S. Wingreen,
925: Science {\bf 280}, 567 (1998).
926:
927: \bibitem{selman1} T-S. Kim and S. Hershfield, Phys. Rev. B{\bf 63},
928: 245326 (2001). See also A. A. Aligia and C. R. Proetto,
929: Phys. Rev. B{\bf 65}, 165305 (2002).
930:
931: \bibitem{claro} M. L. Ladr\'on de Guevara, F. Claro, and P. A. Orellana,
932: Phys. Rev. B{\bf 67}, 195335 (2003).
933:
934: \bibitem{claro2} Z. Y. Zeng, F. Claro, and A. P\'erez, Phys. Rev.
935: B{\bf 65},085308 (2002). See also Z. Y. Zeng, F. Claro, and W. Yan,
936: cond-mat/0105194.
937:
938: \bibitem{selman2} T-S. Kim and S. Hershfield, Phys. Rev. B{\bf 67},
939: 235330 (2003). See also Alessandro Silva, Yuval Oreg, and Yuval Gefen,
940: Phys. Rev. B {\bf 66}, 195316 (2002).
941:
942: \bibitem{glazman} L. Glazman and R. Raikh, JETP Lett. {\bf 47}, 452 (1988);
943: T. K. Ng and P. A. Lee, Phys. Rev. Lett. {\bf 61}, 1768 (1988); Y. Meir
944: and N.S. Wingreen, Phys. Rev. Lett. {\bf 68}, 2512 (1992).
945:
946: \bibitem{coupled}
947: L. P. Kouwenhoven, F. W. J. Hekking, B. J. van Wees, and C. J. P. M. Harmans,
948: C.E. Timmering,C.T. Foxon,
949: Phys. Rev. Lett. {\bf 65}, 361 (1990);
950: F. R. Waugh, M. J. Berry, D. J. Mar, R. M. Westervelt,
951: K. L. Campman and A. C. Gossard,
952: Phys. Rev. Lett. {\bf 75}, 705 (1995); T. H. Oosterkamp, T. Fujisawa,
953: W. G. van der Wiel, K. Ishibashi, R. V. Hijman, S. Tarucha, and
954: L. P. Kouwenhoven, Nature {\bf 395}, 873 (1998);
955: W. Chen, T. Jamneala, V. Madhavan, and M. F. Crommie,
956: Phys. Rev. B{\bf 60}, R8529 (1999).
957:
958: \bibitem{crommie} T. Jamneala, V. Madhavan, and M. F. Crommie,
959: Phys. Rev. Lett. {\bf 87}, 256804 (2001); and references therein.
960:
961: \bibitem{S-T1} W. G. van der Wiel, S. De Franceschi, J. M. Elzerman,
962: S. Tarucha, and L. P. Kouwenhoven, J. Motohisa, F. Nakajima, T. Fukui,
963: Phys. Rev. Lett. {\bf 88}, 126803 (2002).
964:
965: \bibitem{S-T2} M. Pustilnik and L. I. Glazman, Phys. Rev. Lett. {\bf 87},
966: 216601 (2001); W. Hofstetter and H. Schoeller, Phys. Rev. Lett. {\bf 88},
967: 016803 (2001).
968:
969: \bibitem{oguri}
970: Also note that some previous calculations using second order (in the
971: Hubbard $U$) perturbation theory have not found the
972: subtle effects reported in this paper (see
973: A. Oguri, Phys. Rev. B{\bf 63}, 115305 (2001); and
974: references therein). The origin of the discrepancy may simply
975: arise from the different accuracy in the many-body techniques employed,
976: but this issue remains to be investigated. However, note that our
977: cancellation of conductance for an odd number of dots
978: leads in a natural manner to the Hubbard insulator limit expected
979: at density one electron per site, as the number of dots grows.
980:
981:
982: \bibitem{Elbio} E. Dagotto, Rev. Mod. Phys {\bf 66}, 763 (1994).
983:
984: \bibitem{meth1} V. Ferrari, G. Chiappe, E.V. Anda, M.A. Davidovich,
985: Phys Rev Lett {\bf 82}, 5088, (1999).
986:
987: \bibitem{meth2}
988: C.A. B\"usser, E.V. Anda, A.L. Lima, Maria A. Davidovich
989: and G. Chiappe, Phys. Rev. B {\bf 62}, 9907 (2000).
990:
991: %\bibitem{comment5} The dip in Fig.~\ref{Fig2}a was shown to be
992: %stable adding small random site-dependent components to (i) the Hubbard
993: %couplings of the dots and (ii) the dot's on-site energies.
994:
995: \bibitem{split}
996: H. Jeong, A.M. Chang, M.R. Melloch, Science {\bf 293} 2221 (2001).
997:
998: \bibitem{comment3} $t^{''}$ can be modified by controlling
999: the tunneling to and from the central dot using other appropriate voltages.
1000:
1001:
1002: %\bibitem{Meir} A. Georges, Y. Meir. Phys. Rev. Lett. {\bf 82}, 3508, (1999).
1003:
1004:
1005: %\bibitem{experiments} D. Goldhaber-Gordon, Hadas Shtrikman, D. Mahalu, David
1006: %Abusch-Magder, U. Meirav, M.A. Kastner. Nature {\bf 39}, 156, (1998);
1007: %S.M. Cronenwet, T.H. Oosterkamp, L.P. Kouwenhoven.
1008: %Science {\bf 281}, 540, (1998).
1009: % 2 coupled QDots - Experiment.
1010:
1011: \bibitem{Golhaber2}
1012: D. Golhaber-Gordon, J. G\"ores, Hadas Shtrikman, D. Mahalu,
1013: U. Meirav, M.A. Kastner, contribution to
1014: {\it 24$^{th}$ International Conference on the
1015: Physics of Semiconductors}, D. Gershoni ed., Jerusalem, Israel (1998);
1016: and references therein.
1017:
1018:
1019: \bibitem{comment4} $\delta$ here could represent a bias potential
1020: that splits the Kondo peak, similarly as observed by S. De Franceschi,
1021: R. Hanson, W. G. van der Wiel, J. M. Elzerman, J. J. Wijpkema,
1022: T. Fujisawa, S. Tarucha, and L. P. Kouwenhoven,
1023: cond-mat/0203146. Finite-$T$ effects likely reduce the dip depth.
1024:
1025: \bibitem{crommie-private} M. F. Crommie, private communication.
1026:
1027: \bibitem{willy} G. Chiappe and J. A. Verg\'es, {\it Transport through
1028: an interacting system connected to leads}, preprint.
1029:
1030:
1031: \end{thebibliography}
1032:
1033: %\end{references}
1034:
1035: \end{document}
1036: