1: \documentclass[prb,twocolumn,showpacs,superscriptaddress,preprintnumbers,amssymb]{revtex4}
2: \usepackage{graphicx}% Include figure files
3: \usepackage{dcolumn}% Align table columns on decimal point
4: \usepackage{bm}% bold math
5: %\usepackage{wasysym}
6: \usepackage{amssymb}
7: \usepackage{amsfonts}
8: \usepackage{amsmath}
9: \begin{document}
10: %\usepackage{ulem}
11: \title{Optimal thermoelectric figure of merit of
12: a molecular junction}
13: \author{Padraig Murphy}
14: \affiliation{Department of Physics, University of California,
15: Berkeley, CA 94720}
16: \author{Subroto Mukerjee}
17: \affiliation{Department of Physics, University of California,
18: Berkeley, CA 94720} \affiliation{Materials Sciences Division,
19: Lawrence Berkeley National Laboratory, Berkeley, CA 94720}
20: \author{Joel Moore}
21: \affiliation{Department of Physics, University of California,
22: Berkeley, CA 94720} \affiliation{Materials Sciences Division,
23: Lawrence Berkeley National Laboratory, Berkeley, CA 94720}
24: \date{\today}
25: \begin{abstract}
26: We show that a molecular junction can give large values of the
27: thermoelectric figure of merit $ZT$, and so could be used as a
28: solid state energy conversion device that operates close to the
29: Carnot efficiency. The mechanism is similar to the Mahan-Sofo
30: model for bulk thermoelectrics --- the Lorenz number goes to zero
31: violating the Wiedemann-Franz law while the thermopower remains
32: non-zero. The molecular state through which charge is transported
33: must be weakly coupled to the leads, and the energy level of the
34: state must be of order $k_B T$ away from the Fermi energy of the
35: leads. In practice, the figure of merit is limited by the phonon
36: thermal conductance; we show that the largest possible
37: $ZT\sim(\tilde{G}_{th}^{ph})^{-1/2}$, where $\tilde{G}_{th}^{ph}$
38: is the phonon thermal conductance divided by the thermal
39: conductance quantum.
40: \end{abstract}
41:
42: \pacs{73.23.-b, 73.23.Hk, 72.15.Jf, 85.65.+h}
43:
44: \maketitle
45: \def\beq{\begin{equation}}
46: \def\eeq{\end{equation}}
47:
48: %\section{Introduction}
49: %\label{sec:intro}
50:
51:
52: %Thermoelectrics: \cite{mahan_phys_today},\cite{disalvo_review},\cite{majumdar_review},\cite{mahan_sofo, humphrey}; \cite{mukerjee_moore};\cite{peterson_2007};
53: %$U$ in benzene: \cite{kivelson}, \cite{gammel}, \cite{baeriswyl},
54: %\cite{kivelson_et_al_reply};
55: %Molecule experiments: \cite{quek};
56: %Thermal conductance of self-assembled monolayers:\cite{wang_segalman_majumdar}.
57: %Thermal transport in molecules (Nitzan)\cite{nitzan_2007}, \cite{galperin_2007}.
58: %Transport: \cite{ng_lee}, \cite{meir_wingreen_1992},
59: %\cite{meir_wingreen_lee-1993}, \cite{meir_wingreen_lee-1991}.
60:
61: %\cite{mehta},
62: %Kondo screening cloud: \cite{affleck_simon}, \cite{hand}, \cite{sorensen_affleck},
63: %Failures of DFT for nanoscale transport: \cite{wang}, \cite{koentopp}, \cite{toher},
64: %Kondo thermopower: \cite{dong_lei}, \cite{kim_hershfield}, \cite{scheibner-2005},
65: %Polygamma function: \cite{dingle}
66:
67: Electrical transport in mesoscopic and nanoscale systems has been a major
68: focus of recent work in physics, chemistry, and materials science. This interest has been motivated both by the potential for applications in electronics and by the opportunity to study novel transport phenomena such as conductance quantization~\cite{vanwees} and Coulomb blockade.
69: %, and the plateau in the electrical conductance associated with the Kondo effect \cite{ng_lee,meir_wingreen_lee}.
70: %\cite{ng_meir_wingreen_lee}.
71: Compared to electrical transport, thermal transport has been much
72: less intensively studied, despite a number of fascinating
73: results including the observation in 2000 of the quantum of thermal
74: conductance~\cite{schwab,rego_1997}.
75: %\cite{schwab,chiatti,rego_1997}.
76: %, and its universality among fermions, bosons and anyons\cite{rego_1999}.
77: Thermal and thermoelectric transport are also relevant to technological questions of great
78: importance, such as the construction of solid-state energy-conversion
79: devices.
80:
81: The efficiency of such devices is determined by the
82: thermoelectric figure of merit~\cite{mahan_phys_today} $ZT$,
83: defined as
84: \[
85: ZT = \frac{G S^2}{ G^{el}_{th}/T + G^{ph}_{th}/T}.
86: \]
87: Here $G$ is the charge conductance, $S$ the thermopower,
88: $G^{el}_{th}$ the electron thermal conductance, and
89: $G^{ph}_{th}$ the phonon thermal conductance.
90: As $ZT\rightarrow\infty$, the device attains the Carnot efficiency.
91: The figure of merit of the
92: semiconductors typically used for thermoelectric applications
93: is approximately 1; $ZT$ values of $2$ or $3$ would
94: lead to a considerable increase in commercial utility.
95: Increased interest in $ZT$ has given additional motivation to work on
96: the thermal and thermoelectric properties of unusual materials:
97: nanowires~\cite{hicks_dresselhaus_nanowires,murphy_moore},
98: nanostructured materials~\cite{harman}, strongly correlated
99: materials~\cite{terasaki_1997,cavaong,mukerjee_moore}, and molecular
100: junctions~\cite{selzer_allara,pramod}. Of these, molecular
101: junctions are particularly promising, as it is hoped that the
102: phonon thermal conductance will be small due to a density of
103: states mismatch between the phonons in the leads and the phonon
104: modes of the molecule.
105: %Unfortunately, experimental work on even the electrical conductance has proven difficult, primarily due to metal-molecule contacts.
106: Improvements of standard density functional methods are needed to give unambiguous predictions
107: for the transport properties of molecular systems~\cite{quek,toher}, especially when
108: correlations between electrons are significant.
109: It is therefore valuable to have a thorough understanding of transport in
110: simplified models, such as the Anderson model (reviewed below), that describe
111: intramolecular correlations, both as a check on the accuracy of new {\it ab initio} methods
112: and to guide both experiment and numerics to where high $ZT$ may occur.
113:
114:
115: In this paper we find the figure of merit of a molecular junction. We start from the
116: Anderson model for the molecule and leads, which is valid under the
117: assumption that a single molecular level dominates transport. If the
118: lifetime $\tau$ of an electron in the molecular state satisfies $\tau
119: \ll \hbar/(k_BT)$, it is generally true~\cite{sakano,unpub} that $ZT \ll
120: 1$. We concentrate therefore on the regime $\tau \gg \hbar/(k_BT)$.
121: Physically, this corresponds to a wavefunction for the level that is
122: strongly localized at the center of the molecule, so that the state is
123: weakly coupled to the leads. In reference (\onlinecite{tsaousidou}) it was
124: shown that for the Anderson model in this limit of weak coupling to the
125: leads, with strong interactions, and ignoring the phonon contribution to
126: the thermal conductance, the figure of merit becomes infinite. Here we
127: find that the phenomenon is surprisingly general --- it is also true in
128: the limit of weak electron-electron interactions. We clarify the
129: mechanism in both cases by comparing with the work of Mahan and Sofo
130: \cite{mahan_sofo} on bulk thermoelectrics. (The mechanism, a violation
131: of the Wiedemann-Franz law, has previously been studied in quantum dots
132: in different parameter regimes in references (\onlinecite{boese_fazio,krawiec,kubala}).)
133: Also, by carrying the calculation for the electron-only $ZT$ to higher
134: order in the coupling to the leads, we find the optimal value of $ZT$ once
135: the phonon contribution is included. Finally, we suggest a candidate
136: molecule where high $ZT$ might be observed.
137:
138:
139: The Anderson model is defined by the Hamiltonian
140: \begin{align}
141: \label{eqn:anderson}
142: H &= -t\sum_{j>0, j<-1, \sigma}( {c^{\dagger}}_{j+1, \sigma} c_{j, \sigma} + hc)
143: \nonumber \\
144: &-\sum_{\sigma}\left\{t'_R({c^{\dagger}}_{1,\sigma} c_{0,\sigma} + hc) +
145: t'_L({c^{\dagger}}_{0,\sigma} c_{-1,\sigma} + hc)\right\}\\
146: &+ \varepsilon_d (n_{0\uparrow} + n_{0\downarrow} ) +
147: Un_{0\uparrow} n_{0\downarrow} .\nonumber
148: \end{align}
149: The label $\sigma$ represents the spin of the electrons; $j$ is the site index,
150: with the molecule represented by the site at $j=0$. The amplitude
151: to hop from the molecule to the right (left) lead is given by $t'_{R(L)}$, and the leads
152: are represented by the sites
153: at $j>0$ and $j<0$. The energy to place a single electron on the
154: molecule is $\varepsilon_d$, which can be interpreted as a gate
155: voltage. The electron-electron repulsion within the molecule
156: results in an additional energy cost of $U$ to add two electrons to the
157: molecular site. The hopping within the leads is $t$, which gives a
158: band structure
159: %\cite{ashcroft_and_mermin}
160: of
161: $E(k) = - 2t\cos ka $, where $a$ is the lattice spacing.
162:
163: As a first step, we find the transport properties with no
164: electronic repulsion ($U=0$), when the transmission function of the impurity
165: can be found by solving the single-particle Schr\"odinger
166: equation. The result is
167: \[
168: {\cal T}_{\sigma} (E) = \frac{\Gamma_L\Gamma_R(1-E^2/(4t^2)) }{
169: \overline\Gamma^2(1-E^2/(4t^2)) + (E(1-\overline\Gamma/(2t))-\varepsilon_d)^2},
170: \]
171: where $E$ is the energy of the incident electron, $\Gamma_{L(R)} = 2{t'}_{L(R)}^{2}/t$,
172: and $\overline\Gamma = (\Gamma_L+\Gamma_R)/2$.
173: For electrons close to the center of the band, and $\Gamma_{L, R} \ll t$,
174: we get the familiar Lorentzian transmission
175: \[
176: {\cal T}_{\sigma} (E) = \frac{\Gamma_L\Gamma_R }{ \overline\Gamma}
177: \frac{\overline\Gamma }{
178: \overline\Gamma^2 + (E-\varepsilon_d)^2} .
179: \]
180: The hybridization energy, $\overline\Gamma$, determines the mean
181: lifetime, $\tau$, of an electron on the dot, through $\tau =
182: \hbar/\overline\Gamma$. We will assume this form of ${\cal
183: T}_\sigma(E)$ in the study of the non-interacting case that
184: follows.
185:
186: Within the Landauer formalism the transmission function determines
187: the charge current $I$ and the heat current due to the electrons
188: $I^{el}_Q$ through the relations
189: \begin{align*}
190: I &= -\frac{e}{ h} \int_{-2t}^{2t} dE
191: ({\cal T}_\uparrow (E)+{\cal T}_\downarrow (E)) (f_L^0(E) - f_R^0(E) ), \\
192: I^{el}_Q &= \frac{1}{ h} \int_{-2t}^{2t} dE (E-\mu)
193: ({\cal T}_\uparrow (E)+{\cal T}_\downarrow (E))
194: (f_L^0(E) - f_R^0(E) ).
195: \end{align*}
196: Here $-e$ is the electron charge, and the
197: function $f_L^0$ is defined by $f_L^0(E) = (e^{(E-\mu_L)/(k_B
198: T_L)} + 1)^{-1}$, where $\mu_L$ and $T_L$ are the chemical
199: potential and temperature in the left lead (and similarly for
200: $f_R^0$). At linear response, and for a chemical potential away
201: from the band edges, we have
202: \begin{align}
203: \label{eqn:curr1}
204: I &= -\frac{2e }{ h}\frac{\gamma_L\gamma_R }{\overline\gamma}
205: \left({\cal F}_0(\overline\gamma, \delta) eV + {\cal F}_1(\overline\gamma, \delta) k_B \Delta T\right),\\
206: \label{eqn:curr2}
207: I^{el}_Q & = \frac{2 k_B T }{ h}\frac{\gamma_L\gamma_R }{\overline\gamma}
208: \left( {\cal F}_1(\overline\gamma, \delta) eV + {\cal F}_2(\overline\gamma, \delta) k_B \Delta T \right),
209: \end{align}
210: where $\gamma_{L(R)} = \Gamma_{L(R)}/(k_BT)$, $\overline\gamma = \overline\Gamma/(k_BT)$, $\delta = (\varepsilon_d-\mu)/(k_BT)$,
211: $eV=\mu_L-\mu_R$, and $\Delta T=T_L-T_R$.
212: The functions ${\cal F}_n$ can be found from equations (\ref{eqn:curr1}) and (\ref{eqn:curr2}) by setting
213: $x = (E-\mu)/(k_BT)$; the result is
214: \[
215: {\cal F}_n (\overline\gamma, \delta) =
216: \int_{-\infty}^\infty dx \frac{x^n}{ (2\cosh(x/2))^2}
217: \frac{\overline\gamma }{ \overline\gamma^2 + (x-\delta)^2}.
218: \]
219: ${\cal F}_0$, ${\cal F}_1$ and ${\cal F}_2$ can then be expressed as
220: \begin{align*}
221: {\cal F}_0 & = \frac{1}{ 4\pi}
222: \left\{ \psi \left( \frac{\pi + w}{ 2\pi} \right)
223: +\psi \left( \frac{\pi + w^*}{ 2\pi} \right)\right\},\\
224: {\cal F}_1 & = \frac{1 }{ 4\pi i}
225: \left\{ w\psi \left( \frac{\pi + w}{ 2\pi} \right)-w^*
226: \psi \left( \frac{\pi + w^*}{ 2\pi} \right)\right\},\\
227: {\cal F}_2 &= \overline\gamma -\frac{1}{ 4\pi}
228: \left\{
229: w^2\psi \left( \frac{\pi + w}{ 2\pi} \right)
230: +w^{*2}\psi \left( \frac{\pi + w^*}{ 2\pi} \right)
231: \right\},
232: \end{align*}
233: where $w = \overline\gamma +i\delta$. The function $\psi$ is the
234: trigamma function $\psi(z) = \sum_{n=0}^\infty (z+n)^{-2}$.
235:
236: We define the limit of a long-lived molecular state
237: as $\overline\Gamma \ll k_B T$, i.e.~$\overline\gamma \ll 1$.
238: Ignoring the phonon contribution
239: to the heat current (to be restored below),
240: the figure of merit is given by
241: $ZT = \left({{\cal F}_0 {\cal F}_2/{\cal F}_1^2} - 1\right)^{-1}$.
242: %and is plotted in Fig.~\ref{fig:ZTofG_U=0}.
243: For $\overline\gamma$ fixed, $ZT$ is largest when
244: $\delta \simeq \pm 2.4$. If one holds
245: $\delta$ fixed, and allows $\overline\gamma\rightarrow 0$,
246: the figure of merit
247: diverges as
248: \[
249: ZT= \frac{1}{ \overline\gamma}\frac{\pi \delta^2}{ (2\cosh(\delta/2))^2} +
250: {\cal O}(\overline\gamma^0).
251: \]
252: (This follows from expanding ${\cal F}_n$ to first order in
253: $\overline\gamma$.)
254: As shown in Fig.~\ref{fig:ZTofdelt_U=0}, the
255: thermopower is finite as $\overline\gamma\rightarrow 0$,
256: and the divergence in $ZT$ is due to a violation of the Wiedemann-Franz law,
257: with the thermal conductance
258: vanishing faster than the electrical
259: conductance.
260:
261: %\begin{figure}
262: %\includegraphics[width=3in,height=2.2in]{opt515_gr23.eps}
263: %\caption{The figure of merit, $ZT$, in the non-interacting case $(U=0)$, and
264: %without including the phonon contribution to the thermal conductance. In the main
265: %figure, it is shown as a function of $\overline\gamma = \overline\Gamma/(k_BT)$,
266: %with $\delta=(\varepsilon_d-\mu)/(k_BT)=2$;
267: %in the inset, $\overline\gamma=0.1$, and $ZT$ is plotted as a function of
268: %$\delta$.
269: %}
270: %\label{fig:ZTofG_U=0}
271: %\end{figure}
272:
273: \begin{figure}
274: \includegraphics[width=3in,height=2.2in]{opt515_gr22.eps}
275: \caption{The Lorenz number (main figure) and the thermopower (inset) plotted
276: as a function of $\overline\Gamma/(k_BT)$, with $(\varepsilon_d-\mu)/(k_BT)=2$ in both cases.
277: Only the electron contribution to the thermal conductance is included in the
278: calculation of the Lorenz number.
279: }
280: \label{fig:ZTofdelt_U=0}
281: \end{figure}
282:
283: In practice, the figure of merit of the junction will be limited by the
284: phonon thermal conductance. This can be parametrized in terms of the fraction
285: $\tilde{G}_{th}^{ph}$
286: of a thermal conductance quantum transmitted by the molecule, i.e.
287: $
288: G_{th}^{ph} = \tilde{G}_{th}^{ph} \frac{\pi^2 }{ 3} \frac{k_B^2}{ h} T.
289: $
290: Using
291: \[
292: GS^2 = \frac{\gamma_L\gamma_R }{ \overline\gamma} \frac{2k_B^2}{ h}
293: \frac{\pi \delta^2}{ (2\cosh(\delta/2))^2},
294: \]
295: %(this can be derived by expanding ${\cal F}_n$ to zeroth order in $\overline\gamma$)
296: we include the phonon contribution to get
297: %\[
298: %{1\over ZT}
299: %= {G_{th}^{e}/T \over G S^2} + {G_{th}^{ph}/T \over G S^2}
300: % \simeq
301: %{\overline\gamma\over 1.32} + {1.25\tilde{G}_{th}^{ph} \over\overline\gamma}
302: %{\overline\gamma^2\over\gamma_L\gamma_R}.
303: %\]
304: %($\Pi = GS^2$ is plotted in Fig.~\ref{fig:PFofG_U=0}.)
305: \[
306: \frac{1}{ZT}
307: = \frac{G_{th}^{e} }{ G S^2T} + \frac{G_{th}^{ph} }{ G S^2T}
308: \simeq
309: \frac{(2\cosh(\delta/2))^2 }{ \pi \delta^2} \left( \overline\gamma
310: + \frac{\pi^2 \tilde{G}_{th}^{ph}\overline\gamma }{ 6 \gamma_L\gamma_R}
311: \right).
312: \]
313: Assuming $\gamma_L=\gamma_R=\gamma$ and minimising $1/ZT$ with respect to $\gamma$
314: and $\delta$
315: gives, for small $\tilde{G}_{th}^{ph}$, an optimal value of
316: \[
317: ZT \simeq 0.51/(\tilde{G}_{th}^{ph})^{1/2}.
318: \]
319: The figure of merit becomes infinite as $\tilde{G}_{th}^{ph}$ goes to zero, as we found
320: before.
321: \begin{figure}
322: \includegraphics[width=3in,height=2.2in]{opt_gr19.eps}
323: \caption{
324: $1/ZT$, without the phonon contribution to the thermal conductance,
325: is plotted for $\Gamma/(k_BT) \rightarrow 0$ and $(\varepsilon_d-\mu)/(k_B T)=2$.
326: }
327: \label{fig:ZTU}
328: \end{figure}
329:
330: There is a simple physical picture that gives us the correct expressions for
331: ${\cal F}_0$, ${\cal F}_1$, and ${\cal F}_2$ to order $\overline\gamma^0$,
332: and thus gives $1/ZT = 0 + {\cal O}(\overline\gamma^0)$.
333: This approximation
334: is known as ``sequential tunneling'', and has been used to study transport in
335: quantum dots and molecules \cite{beenakker-1991,beenakker_staring,koch,zianni}.
336: Suppose the molecular level has energy $\varepsilon_d$. Then, in this approximation,
337: an electron in the leads can tunnel onto the level only if it too has energy
338: $\varepsilon_d$. Energy is therefore conserved throughout the transport process.
339: This allows for a current proportional to $\Gamma_L\Gamma_R/\overline\Gamma$. In
340: quantum-mechanical tunneling
341: by contrast, the intermediate state is virtual; away from
342: resonance, the current is
343: proportional to the probability to tunnel onto the level multiplied by
344: the probability to tunnel off,
345: and thus to $\Gamma_L\Gamma_R$.
346:
347: In the sequential tunneling limit, each electron passing
348: through the junction carries the same amount of heat, $Q =
349: \varepsilon_d-\mu$. To understand why the Wiedemann-Franz law is
350: violated, it is essential to bear in mind that the thermal
351: conductance is defined as the heat current divided by the
352: temperature difference {\it under the condition of zero charge
353: current}. The condition $I=0$ implies that the flux of left-moving
354: electrons is the same as the flux of right-moving electrons; since
355: each left or right moving electrons carries exactly the same
356: amount of heat $Q$, the thermal current $I^{el}_Q$ is also zero.
357: This is the same mechanism that Mahan and Sofo used to model the
358: bulk material with the best thermoelectric figure of
359: merit~\cite{mahan_sofo,humphrey}. For long-lived
360: molecular states, unlike bulk materials (except in the atomic limit~\cite{mukerjee_moore}), the theoretical analysis
361: can be extended to include electron-electron repulsion, as now
362: done.
363:
364: We now consider whether, once electron-electron repulsion
365: in the molecule is added,
366: the electronic figure of merit $ZT$ still scales like $1/\overline\gamma$ in the limit of
367: a long-lived molecular state.
368: Define the probability of the molecular level being empty as $P(0)$,
369: of there being an up spin as $P(\uparrow)$, a down spin as $P(\downarrow)$,
370: and of it being doubly occupied as $P(2)$. We consider the
371: currents where the molecule meets the left lead.
372: An example of a process that contributes to the currents at this point
373: is the tunneling of an up electron with energy $\varepsilon_d$ from the left
374: lead onto the empty level.
375: %The contribution of this process to the currents is
376: %\begin{align*}
377: %I &= \cdots-e \tau^{-1} P(0) f_{L\uparrow}^0 (\varepsilon_d) +\cdots,\\
378: %I^{el}_Q &= \cdots (\varepsilon_d-\mu)
379: %\tau^{-1} P(0) f_{L\uparrow}^0 (\varepsilon_d) +\cdots.
380: %\end{align*}
381: To find the total currents, we consider all possible tunneling processes, both onto and
382: off of the molecule.
383: %We assume that
384: %there is no magnetic field, which implies that
385: %$f_{L\uparrow}^0 = f_{L\downarrow}^0=f_{L}^0$, and
386: %similarly for the right lead.
387: For
388: $\delta>0$, ie. $\varepsilon_d>\mu$,
389: the charge and heat currents at the left junction
390: are then given by
391: \begin{align*}
392: I &= -e\tau_L^{-1} \left( P(0) 2 f_L(\varepsilon_d) -P(1)(1 - f_L(\varepsilon_d) ) \right. \\
393: &\left.\qquad +P(1)f_L(\varepsilon_d+U) - 2P(2) (1 - f_L(\varepsilon_d+U))\right),\\
394: I^{el}_Q &= \tau_L^{-1} k_B T \left( \delta P(0) 2 f_L(\varepsilon_d) -\delta P(1)(1 - f_L(\varepsilon_d) ) \right. \\
395: & \qquad +(\delta+u)P(1)f_L(\varepsilon_d+U) \\
396: &\left.\qquad - (\delta+u)2P(2) (1 - f_L(\varepsilon_d+U))\right),
397: \end{align*}
398: where $P(1) = P(\uparrow)+P(\downarrow)$,
399: $u= U/(k_B T)$, and, as before, $\delta = (\varepsilon_d-\mu)/(k_BT)$.
400: The probabilities $P(0)$, $P(\uparrow)$, $P(\downarrow)$, and $P(2)$ are determined
401: from the steady state condition.
402: The final result for the linear response transport coefficients is
403: \begin{align*}
404: G &=
405: \frac{e^2}{ \hbar}\frac{\gamma_L \gamma_R }{\overline\gamma}\frac{e^{\delta +u} ( 1 + e^\delta(2+e^{\delta+u}))}{ (1+ e^{\delta} )(1+e^{\delta +u} )( 1 + e^{\delta+u}(2+e^{\delta}))},\\
406: S &= -\frac{k_B}{e} \frac{(\delta (1 +e^\delta(2+e^{\delta+u})) + u(1+e^\delta))
407: }{
408: ( 1 + e^\delta(2+e^{\delta+u}))},\\
409: \frac{G_{th}^{el} }{ T} &= \frac{k_B^2}{ \hbar} \frac{\gamma_L \gamma_R }{\overline\gamma}
410: \frac{u^2 e^{2\delta + u} }{ (1+2e^{\delta+u} + e^{2\delta +u})
411: (1 + 2e^\delta + e^{2\delta +u}) } .
412: \end{align*}
413: (It is straight-forward to confirm that this result, with $u=0$, agrees with the
414: exact result for the non-interacting case in the limit of $\gamma_{L,R}\ll 1$, if
415: one makes the identification $\tau_{L(R)}^{-1} = \Gamma_{L(R)}/\hbar$.
416: This confirms that only sequential tunneling is important in the limit
417: of a long-lived state.)
418:
419: %Setting $u=0$, the conductance is
420: %We confirm that this model, with $u=0$, agrees with the exact result for the non-interacting case in the limit of $\gamma\ll 1$.
421: %Setting $u=0$, the conductance is
422: %\[
423: %G = {e^2 \tau^{-1} \over k_B T} {e^\delta (1+e^\delta)^2 \over
424: %(1+e^\delta)^4} = {e^2 \tau^{-1} \over k_B T} {1\over (2\cosh(\delta/2))^2}.
425: %\]
426: %Identifying $\tau^{-1} = \Gamma/\hbar$ gives
427: %\[
428: %G = \gamma {e^2 \over h/(2\pi)}{1\over (2\cosh(\delta/2))^2}
429: % = {2e^2 \over h} {\gamma \pi \over (2\cosh(\delta/2))^2},
430: %\]
431: %which is the correct result.
432: %{\bf INCLUDE $\kappa$, $S$}
433:
434: %Returning to the interacting case, the figure of merit is given by
435:
436: %\lookhere
437: The figure of merit in the sequential tunneling
438: approximation (again ignoring phonons) is given by
439: \[
440: ZT = \frac{e^{-\delta} }{ u^2} \frac{ (\delta(1+2e^{\delta} + e^{2\delta+u}) + u(1+e^\delta))^2
441: }{ (1+e^\delta) (1+e^{\delta+u})}.
442: \]
443: (An approximate version of this formula was derived in \onlinecite{tsaousidou}.)
444: $1/ZT$ is plotted in Fig.~\ref{fig:ZTU} as a function of $u$ for
445: $\delta=2$ and $\gamma \rightarrow 0$. At $u=0$, the
446: non-interacting case, $ZT$ is divergent, as before. For $u \gg 1$,
447: $1/ZT$ goes to zero like $e^{-u}$. This can be understood within
448: the picture presented earlier. Even though, in contrast to the
449: non-interacting case there are now two levels (of energy
450: $\varepsilon_d$ and $\varepsilon_d+U$) on the molecule through which
451: transport can occur, the relative probability of electrons in the
452: leads having the corresponding energies goes as $e^{-u}$. As $U$
453: becomes large there is thus only an exponentially small
454: probability that there will be an electron in the leads with
455: energy $\varepsilon_d+U$. In this case, each electron that passes
456: through the junction carries heat $\varepsilon_d-\mu$ and the
457: situation is similar to the non-interacting case where transport
458: takes place through just one level on the molecule. Thus, the
459: thermal conductance in this case is also zero. (For convenience,
460: let us consider the symmetric case, with $\gamma_L=\gamma_R
461: =\gamma$.) Since the sequential tunneling model is correct only to
462: first order in $\gamma$, we can conclude that the thermal
463: conductance vanishes to first order in $\gamma$, and so must be of
464: order $\gamma^2$. For $\delta \simeq 2.4$ and $u\gg 1$, we have $G
465: S^2 \simeq \gamma 0.4k_B^2/\hbar$, and so $GS^2$ scales like
466: $\gamma$ as in the non-interacting case. Adding in once more the
467: phonon contribution to the heat current, we can again conclude
468: that $ZT$ will scale like $(\tilde{G}^{ph}_{th})^{-1/2}$.
469: In the limit of $u \rightarrow \infty$, the Green function method
470: \cite{meir_wingreen_1992,meir_wingreen_lee} can be used to find the
471: transport coefficients to higher order in $\gamma$, and so the
472: coefficient in the scaling of $ZT$.
473: The result is that, including the phonon contribution as before,
474: the optimal value of $ZT$ is $0.42 (G_{ph}^{th})^{-1/2}.$
475:
476: %\section{Conclusion}
477: %\label{sec:conclusion}
478:
479: We have shown that transport through a single molecular level can lead
480: to large values of the thermoelectric figure of merit $ZT$.
481: Within the Anderson model for the level, this occurs if
482: the molecular state is long-lived
483: (i.e., for the case of a symmetric junction,
484: $\gamma \rightarrow 0 $) and the energy
485: scale $U$ of the electron-electron
486: interactions within the molecule satisfies either
487: $U\ll k_BT$ or $U\gg k_BT$.
488: The increase in $ZT$ is due to a violation of the Wiedemann-Franz law ---
489: the charge conductance scales like $\gamma$, whereas the electron thermal
490: conductance divided by the temperature scales like $\gamma^2$.
491: %This violation is similar to that found in the work of Mahan and Sofo\cite{mahan_sofo} on bulk thermoelectric materials.
492: % They found that a material with a delta-function density of states for the conduction band has a similar anomaly in the Wiedemann-Franz law, and so can give large values of $ZT$.
493: While similarly large figures of merit may be achieved for optimal band structures in bulk materials~\cite{mahan_sofo}, the advantage of molecular systems is that they may have only a small phonon contribution to the thermal conductance, so that their
494: maximum achieved value of $ZT$ once phonons are included
495: may be higher than in bulk materials.
496:
497: There are some practical challenges in achieving dramatically
498: enhanced $ZT$ in a real device by the mechanism presented here. The wavefunction of the level must be localized at the center of the molecule to give a small overlap with the states
499: in the leads, and so a long lifetime; in addition, the energy of
500: the level must lie a distance of about $2k_BT$ from the Fermi
501: level of the leads. These requirements may be satisfied by the
502: Co(tpy-(CH$_2$)$_5$-SH)$_2$ molecules studied by Park et al
503: \cite{park_review,park_2002}. The electronic transport in these
504: systems is through a cobalt atom that sits at the center of the
505: molecule; the alkyl chains separate the cobalt from the gold
506: leads, creating a state localized away from the leads. It would be
507: of great interest to see if indeed the Wiedemann-Franz law is
508: violated in these molecules, as implied by the results in this paper.
509:
510: \acknowledgments
511:
512: The authors thank A. Majumdar, P. Reddy, and R.
513: Segalman for useful discussions and acknowledge support from the
514: Division of Materials Sciences and Engineering of the Department
515: of Energy.
516:
517: \bibliographystyle{./apsrev}
518: \bibliography{./ZTbib}
519: \end{document}
520: