Thermoelectric efficiency of three-terminal quantum thermal machines

Thermoelectric efficiency of three-terminal quantum thermal machines

Francesco Mazza NEST, Scuola Normale Superiore, and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy    Riccardo Bosisio Service de Physique de l’ État Condensé (CNRS URA 2464), IRAMIS/SPEC, CEA Saclay, 91191 Gif-sur-Yvette, France    Giuliano Benenti CNISM and Center for Nonlinear and Complex Systems, Universitá degli Studi dell’ Insubria, via Valleggio 11, 22100 Como, Italy Istituto Nazionale di Fisica Nucleare, Sezione di Milano, via Celoria 16, 20133 Milano, Italy    Vittorio Giovannetti NEST, Scuola Normale Superiore, and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy    Rosario Fazio NEST, Scuola Normale Superiore, and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy    Fabio Taddei NEST, Scuola Normale Superiore, and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy
Abstract

The efficiency of a thermal engine working in linear response regime in a multi-terminal configuration is discussed. For the generic three-terminal case, we provide a general definition of local and non-local transport coefficients: electrical and thermal conductances, and thermoelectric powers. Within the Onsager formalism, we derive analytical expressions for the efficiency at maximum power, which can be written in terms of generalized figures of merit. Furthermore, using two examples, we investigate numerically how a third terminal could improve the performance of a quantum system, and under which conditions non-local thermoelectric effects can be observed.

pacs:
72.20.Pa, 05.70.Ln, 73.23.-b

1 Introduction

Thermoelectricity has recently received enormous attention due to the constant demand for new and powerful ways of energy conversion. Increasing the efficiency of thermoelectric materials, in the whole range spanning from macro- to nano-scales, is one of the main challenges, of great importance for several different technological applications [1, 2, 3, 4, 5, 6]. Progress in understanding thermoelectricity at the nanoscale will have important applications for ultra-sensitive all-electric heat and energy transport detectors, energy transduction, heat rectifiers and refrigerators, just to mention a few examples. The search for optimisation of nano-scale heat engines and refrigerators has hence stimulated a large body of activity, recently reviewed by Benenti et al. [7].

While most of the investigations have been carried out in two-terminal setups, thermoelectric transport in multi-terminal devices just begun to be investigated [8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22] since these more complex designs may offer additional advantages. An interesting perspective, for instance, is the possibility to exploit a third terminal to “decouple” the energy and charge flows and improve thermoelectric efficiency [9, 10, 11, 12, 16, 17, 18, 19]. Furthermore, fundamental questions concerning thermodynamic bounds on the efficiency of these setups has been investigated[13, 14, 15, 20, 21], also accounting for the effects of a magnetic field breaking the time-reversal symmetry[23]. In most of the cases studied so far, however, all but two-terminal were considered as mere probes; i.e. no net flow of energy and charge through them was allowed. In other works a purely bosonic reservoir has been used, only exchanging energy (and not charge) current with the system [9, 10, 11, 12].

Figure 1: Three-terminal thermal machine. A scattering region is connected to 3 different fermionic reservoirs, each of these is able to exchange heat and particles with the system. Reservoir is taken as the reference for measuring temperature and energy: ; . The reservoirs and have small variations in temperature and chemical potential: , . With we denote a generic coherent scattering region.

A genuine multi-terminal device will however offer enhanced flexibility and therefore it might be useful to improve thermoelectric efficiency. A full characterization of these systems is still lacking and motivates us to tackle this problem. Here we focus on the simplest instance of three reservoirs, which can exchange both charge and energy current with the system. A sketch of the thermal machine is shown in Fig.1, where three-terminal are kept at different temperatures and chemical potentials connected through a scattering region. Our aim is to provide a general treatment of the linear response thermoelectric transport for this case, and for this purpose we will discuss local and non-local transport coefficients. Note that non-local transport coefficients are naturally requested in a multi-terminal setup, since they connect temperature or voltage biases introduced between two-terminal to heat and charge transport among the remaining terminals. We will then show that the third terminal could be exploited to improve thermoelectric performance with respect to the two-terminal case. We will focus our investigations on the efficiency at maximum power [24, 25, 26, 27, 28, 29, 30, 31, 32, 33], i.e. of a heat engine operating under conditions where the output power is maximized. Such quantity, central in the field of finite-time thermodynamics [34], is of great fundamental and practical relevance to understand which systems offer the best trade-off between thermoelectric power and efficiency.

The paper is organized as follows. In Section 2 we briefly review the linear response, Onsager formalism for a generic three-terminal setup. We will discuss the maximum output power and trace a derivation of all the local and non-local transport coefficients. In Section 3 we extend the concept of Carnot bound at the maximum efficiency to the three-terminal setup and we derive analytical formulas of the efficiency at maximum power in various cases, depending on the flow of the heat currents. These expressions are written in terms of generalized dimensionless figures of merit. Note that the expressions derived in Section 2 and 3 are based on the properties of the Onsager matrix and on the positivity of the entropy production. Therefore they hold for non-interacting as well as interacting systems. This framework will then be applied in Section 4 to specific examples of non-interacting systems in order to illustrate the salient physical picture. Namely, we will consider a single quantum dot and two dots in series coupled to the three-terminal. Finally Section 5 is devoted to the conclusions.

2 Linear response for 3-terminal systems

The system depicted in Fig. 1 is characterized by three energy and three particle currents ( and , respectively) flowing from the corresponding reservoirs, which have to fulfill the constraints:

(1)

(positive values being associated with flows from the reservoir to the system). In what follows we will assume the reservoir 3 as a reference and the system to be operating in the linear response regime, i.e. set and write with and for , and is the Boltzmann constant. Under these assumptions the relation between currents and biases can then be expressed through the Onsager matrix of elements via the identity:

(2)

where and are the generalized forces, and where are the heat currents of the system, the corresponding currents to reservoir 3 being determined from and via the conservation laws of Eq. (1). In our analysis we take to be symmetric (i.e. ) by enforcing time reversal symmetry in the problem. We also remind that, due to the positivity of the entropy production rate, such matrix has to be semi-positive definite (i.e. ) and that it can be used to describe a two-terminal model connecting (say) reservoir 1 to reservoir 3 by setting for all .

2.1 Transport coefficients

For a two-terminal model the elements of the Onsager matrix can be related to four quantities which gauge the transport properties of the system under certain constraints. Specifically these are the electrical conductance and the Peltier coefficient (evaluated under the assumption that both reservoirs have the same temperature), and the thermal conductance and the thermopower (or Seebeck coefficient) (evaluated when no net charge current is flowing through the leads). When generalized to the multi-terminal model these quantities yield to the introduction of non-local coefficients, which describe how transport in a reservoir is influenced by a bias set between two other reservoirs.

2.1.1 Thermopower

For a two-terminal configuration the thermopower relates the voltage that develops between the reservoirs to their temperature difference under the assumption that no net charge current is flowing in the system, i.e. . A generalization of this quantity to the multi-terminal scenario can be obtained by introducing the matrix of elements

(3)

with local () and non-local () coefficients, being the electron charge. In this definition, which does not require the control of the heat currents, we have imposed that the particle currents in all the leads are zero (the voltages are measured at open circuits) and that all but one temperature differences are zero (of course this last condition is not required in a two-terminal model). It is worth observing that Eq. (3) differs from other definitions proposed in the literature. For example in Ref. [35] a generalization of the two-terminal thermopower to a three-terminal system, was proposed by setting to zero one voltage instead of the corresponding particle current. While operationally well defined, this choice does not allow one to easily recover the thermopower of the two-terminal case (in our approach instead this is rather natural, see below). Finally in the probe approach presented in Refs. [8, 13, 14, 15, 20, 21] it was possible to study a multi-terminal device by using an effective two-terminal system only, because the heat and particle currents of the probe terminals are set to vanish by definition. Therefore, within this approach, there are no chances of having non-local transport coefficients.

In the three-terminal scenario we can use Eq. (2) to rewrite the elements of the matrix (3). In particular introducing the quantities

(4)

we get (see  A for details)

(5)
(6)

which yields, correctly, as the only non-zero element, by taking the two-terminal limits detailed at the end of the previous section.

2.1.2 Electrical conductance

In a two-terminal configuration the electric conductance describes how the electric current depends upon the bias voltage between the two-terminal under isothermal conditions, i.e. . The generalization to many-terminal systems is provided by the following matrix:

(7)

Using the three-terminal Onsager matrix (2) we find

(8)

which, in the two-terminal limit where reservoir 2 is disconnected from the rest, gives as the only non-zero element.

2.1.3 Thermal conductance

The thermal conductance for a two-terminal is the coefficient which describes how the heat current depends upon the temperature imbalance under the assumption that no net charge current is flying through the system, i.e. . In the multi-terminal scenario this generalizes to

(9)

where one imposes the same constraints as those used for the thermopower matrix (3), i.e. no currents and for all terminals but the -th. For a three-terminal case, using Eq. (4) this gives

(10)
(11)

and

(12)

Once more, in the two-terminal limit where the reservoir 2 is disconnected from the rest, the only non-zero element is .

2.1.4 Peltier coefficient

In a two-terminal configuration the Peltier coefficient relates the heat current to the charge current under isothermal condition, i.e. . For multi-terminal systems this generalizes to the matrix

(13)

which can be shown to be related to the thermopower matrix (3), through the Onsager reciprocity equations, i.e. ( being the magnetic field on the system),[36, 37] from which, using Eqs. (5) and (6), one can easily derive for the three-terminal case the dependence upon the Onsager matrix .

3 Efficiency for 3-terminal systems

In order to characterize the properties of a multi-terminal system as a heat engine we shall now analyze its efficiency. Generalizing the definition for the efficiency of a two-terminal machine [7, 36], we define the steady state heat to work conversion efficiency , for a three-terminal machine, as the power generated by the machine (which equals to the sum of all the heat currents exchanged between the system and the reservoirs), divided by the sum of the heat currents absorbed by the system, i.e.

(14)

where the symbol in the denominator indicates that the sum is restricted to positive heat currents only, and where in the last expression we used Eq. (1) to express in terms of the other two independent currents[41].

The definition (14) applies only to the case in which is positive. Since the signs of the heat currents are not known a priori (they actually depend on the details of the system), the expression of the efficiency depends on which heat currents are positive. For the three-terminal system depicted in Fig. 1 we set for simplicity and focus on those situations where is negative (positive values of being associated with regimes where the machine effectively works as a refrigerator which extract heat from the coldest reservoir of the system). Under these conditions the efficiency is equal to

(15)

when both and are positive, or

(16)

when for or only is positive.

3.1 Carnot efficiency

The Carnot efficiency represents an upper bound for the efficiency and is obtained for an infinite-time (Carnot) cycle. For a two-terminal thermal machine the Carnot efficiency is obtained by simply imposing the condition of zero entropy production, namely . If the two reservoirs are kept at temperatures and (with ), from the definition of the efficiency, Eq. (14), one gets the two-terminal Carnot efficiency . The Carnot efficiency for a three-terminal thermal machine is obtained analogously by imposing the condition of zero entropy production, when a reservoir at an intermediate temperature is added. If only is positive as in Eq. (16), one obtains

(17)

where . Note that Eq. (17) is the sum of the two-terminal Carnot efficiency and a term whose sign is determined by . Since , and , it follows that is always reduced with respect to its two-terminal counterpart . Analogously if only is positive, one obtains

(18)

which again can be shown to be reduced with respect to , since , , , and . We notice that this is a hybrid configuration (not a heat engine, neither a refrigerator): the hottest reservoir absorbs heat, while the intermediate-temperature reservoir releases heat. However, the heat to work conversion efficiency is legitimately defined since generation of power () can occur in this situation. Finally, if both and are positive as in Eq. (15) one obtains

(19)

Since , the term that multiplies is positive so that is reduced with respect to the two-terminal case.

It is worth noticing that, in contrast to the two-terminal case, the Carnot efficiency cannot be written in terms of the temperatures only, but it depends on the details of the system. Moreover, note that the Carnot efficiency is unchanged with respect of the two-terminal case if in (17) or if in (19). Indeed, in this situation the quantities are equal to one, making the extra terms in Eqs. (17) or (19) to vanish.

The above results for the Carnot efficiency could be generalized to many-terminal systems. In particular, we conjecture that, given a system that works between and (with ) and adding an arbitrary number of terminals at intermediate temperatures will in general lead to Carnot bounds smaller than . On the other hand, adding terminals at higher (or colder) temperatures than and will make increase.

Notice that within linear response and via Eq. (2) we can express the Carnot efficiencies (17)-(19) in terms of the generalized forces .

3.2 Efficiency at Maximum Power

The efficiency at maximum power is the value of the efficiency evaluated at the values of chemical potentials that maximize the output power of the engine. In the two-terminal case the efficiency at maximum power can be expressed as [28]

(20)

where is a dimensionless figure of merit which depends upon the transfer coefficient of the system. The positivity of the entropy production imposes that such quantity should be non-negative (i.e. ), therefore is bounded to reach its maximum value only in the asymptotic limit of (Curzon-Ahlborn limit [24, 25, 26, 27] within linear response [28]).

For the three-terminal configuration the output power is a function of the four generalized forces introduced in Eq. (2), i.e.

(21)

In the linear regime this is a quadratic function which can be maximized with respect to and while keeping and constant (the existence of a maximum being guaranteed by the the positivity of the entropy production). The resulting expression is

(22)

where is a positive semi-definite matrix, see  B, whose elements depends on the Onsager coefficients via the identities

(23)

Indicating with the eigenvalues of we can then further simplify Eq. (22) by writing it as

(24)

where is the geometric average of system temperatures, while the angle identify the rotation in the , plane which defines the eigenvectors of . We call the parameter

(25)

three-terminal power factor. It relates the maximum power to the temperature difference: by construction it fulfills the inequality , the maximum being achieved for (i.e. by ensuring that coincides with the eigenvector of associated with its largest eigenvalue ). Note that in the two-terminal limit we have , , , so that the usual two-terminal power factor is recovered.

Exploiting Eq. (24) we can now write the efficiency at maximum power for the three cases detailed in Eqs. (15) and (16). Specifically we have

(26)
(27)

and

(28)

where we have defined the parameters , and and introduced the following generalized coefficients:

(29)

The efficiencies (26), (27) and (28) can also be expressed in terms of the corresponding Carnot efficiencies given in Eqs. (17), (18) and (19), obtaining the following equations which mimic Eq. (20) of the two-terminal case:

(30)
(31)
(32)

where we have introduced the constants

(33)
(34)
(35)

and the combinations of figures of merit

(36)

Notice also that in writing Eq. (32) we retained only the leading order neglecting contributions of order or higher.

The above expressions can be used to provide a generalization of the Curzon-Ahlborn limit efficiency for a multi-terminal quantum thermal device. Indeed using the Cholesky decompositions on the Onsager matrix, we can prove that the constants , defined in Eqs. (33), (34) are positive, see  B for details. This fact together with the positivity of the quantities , that we have checked numerically, implies that the efficiencies are always upper bounded by half of the associated Carnot efficiencies, i.e.

(37)

the inequality being saturated when the generalized coefficients (29) diverge. An analogous conclusion can be reached also for (32), yielding

(38)

In this case is no longer guaranteed to be positive due to the presence of . Still the inequality (38) can be derived by observing that the quantities and entering in the rhs of Eq. (32) have always the same sign.

4 Examples

In this Section we shall apply the theoretical framework developed so far to two specific non-interacting systems attached to three terminals. Namely, we will discuss the case of a single dot and the case of two coupled dots, in the absence of electron-electron interaction (which cannot be dealt within the Landauer-Büttiker formalism). Our aim is to show that one can easily find situations where the efficiency and output power are enhanced with respect to the two-terminal case. Furthermore, through the example of the single dot, we find the conditions that guarantee the non-local thermopowers to vanish.

The coherent flow of particles and heat through a non-interacting conductor can be described by means of the Landauer-Büttiker formalism. Under the assumption that all dissipative and phase-breaking processes take place in the reservoirs, the electric and thermal currents are expressed in terms of the scattering properties of the system [38, 39, 40]. For instance, in a generic multi-terminal configuration the currents flowing into the system from the -th reservoir are:

(39)
(40)

where the sum over is intended over all but the -th reservoir, h is the Planck’s constant, is the transmission probability for a particle with energy to transit from the reservoir to reservoir , and where finally is the Fermi distribution of the particles injected from reservoir (notice also that we are considering currents of spinless particles). In what follows we will use the above expressions in the linear response regime where and , and compute the associated Onsager coefficients (2), see  C.

4.1 Single dot

In this section we study numerically a simple model consisting of a quantum dot with a single energy level , coupled to three fermionic reservoirs, labeled , , and , see Fig. 2. For simplicity, the coupling strength to electrodes 1 and 3 are taken equal to , while the coupling strength to electrode 2 is denoted by . In particular we want to investigate how the efficiencies, output powers and transport coefficients evolve when the system is driven from a two-terminal to a three-terminal configuration, that is by varying the ratio . The two-terminal configuration corresponds to and the third terminal is gradually switched on by increasing . As detailed in C, the transmission amplitudes between each pair of terminals can be used to evaluate the Onsager coefficients – the resulting expression being provided in Eqs. (86). Once the matrix is known, all the currents flowing through the system, efficiencies, output powers and transport coefficients can be calculated within the framework developed in the previous Sections.

Figure 2: (Color online) Sketch of the single dot model used in the numerical simulations: a quantum dot with a single energy level is connected to three fermionic reservoirs , , and . The chemical potential and temperature of the reservoir 3 are assumed as the reference values and . The constants and represent the coupling between the system and the various reservoirs (see  C.1 for details). A zero value of corresponds to disconnecting the reservoir 2 from the system: in this regime the model describes a two-terminal device where reservoirs 1 and 3 are connected through the single dot.

4.1.1 Efficiencies and maximum power

In Fig. 3 we show how the Carnot efficiency depends on the temperature differences and , when the chemical potentials are chosen to guarantee maximum output power, i.e., fixing the generalized forces in order to maximize . As we can see, increases linearly along any “radial” direction defined by a relation , where is a constant. In particular, the dashed lines corresponding to , , and separate the different regimes discussed in Sec. 3.1: for the system absorbs heat only from reservoir (if ) or from and (if ); for the system absorbs heat from reservoirs from and (if ) or from only (if ); finally, for and the system absorbs heat only from reservoir (if ) or from and (if ). In the case when only one heat flux is absorbed the Carnot efficiency is given by Eq. (17) or Eq. (18), while it is given by Eq. (19) if two heat fluxes are absorbed.

Figure 3: (Color online) Carnot efficiency (density plot) of the three-terminal system depicted in Fig. 2, as a function of the gradients of temperature in reservoirs and (the chemical potentials and being chosen to guarantee maximum output power ). The coupling with the reservoirs have been set to have a symmetric configuration with respect to and (i.e. ). Note that increases linearly along any radial direction defined by a relation , where is a constant. In particular, the dashed lines corresponding to , and separate different regimes discussed in Sec. 3.1. The numbers in each region identify the reservoirs from which the heat is absorbed. Parameter values: , .

In Figs. 4 and 5, we show how the efficiency Eq. (14), the output power Eq. (21), the efficiency at maximum output power Eqs. (26)-(28) and the maximum output power Eq. (22), vary when the system is driven from a two-terminal to a three-terminal configuration, i. e. by varying the ratio . We set opposite signs for and , so that the system absorbs heat only from the hottest reservoir , and , in such a way that the Carnot efficiency coincides with that of a two-terminal configuration, namely . Interestingly, we proved that increasing the coupling to the reservoir may lead to an improvement of the performance of the system. In particular, as shown in Fig. 6, the efficiency and the output power can be enhanced at the same time at small couplings , exhibiting a maximum around and , respectively.

In Fig. 5 we show results for the same quantities but at the maximum output power [ and ]. In this case, while still increases with , the corresponding efficiency decreases approximately linearly.

Figure 4: (Color online) Left panel: Efficiency , normalized over the associated Carnot limit computed as in Sec. 3.1, as a function of the coupling to the reservoir . Note that as is switched on, the efficiency increases until it reaches a maximum around , and then it decreases. Right panel: Output power extracted from the system, as a function of the coupling to the reservoir . Parameters: , , , , and .
Figure 5: (Color online) Left panel: Efficiency at maximum power , normalized over the Carnot limit, as a function of the coupling to the reservoir . Right panel: Maximum output power extracted from the system, as a function of the coupling to the reservoir . Parameters: , , , and .

In Figs. 6 and 7 we show the same quantities as in Figs. 4 and 5, but as a function of the coupling for two values of ( and ). From Fig. 6 we can see that at small the coupling to a third terminal can enhance both the efficiency (for ) and the power (for ). In Fig. 7 we note that, both for the two- and the three-terminal system, the efficiency at maximum power tends to in the limit , while the output power vanishes. For two-terminal the result is well known, since a delta-shaped transmission function leads to the divergence of the figure of merit  [42, 43, 44]. Correspondingly, the efficiency at maximum power saturates the Curzon-Ahlborn bound . The same two-terminal energy-filtering argument explains the three-terminal result. Indeed, we found numerically that for the chemical potentials optimizing the output power are such that . Since also the temperatures are chosen so that , we can conclude that terminals and can be seen as a single terminal.

Figure 6: (Color online) Left panel: Efficiency , normalized over the Carnot limit, as a function of the coupling energy . Right panel: Output power extracted from the system, as a function of the coupling energy . In both cases, the full red curves correspond to a three-terminal configuration with , while the dashed blue curve refer to the two-terminal case (). Parameters: , , , and .
Figure 7: (Color online) Left panel: Efficiency at maximum power , normalized over the Carnot limit, as a function of the coupling energy . Right panel: Maximum output power extracted from the system, as a function of the coupling energy . In both cases, the full red curves correspond to a three-terminal configuration with , while the dashed blue curves refer to the two-terminal case (). Parameters: , , and .

4.1.2 Thermopowers

In this section we show analytically that the non-local thermopowers are always zero in this model, while the local ones are equal. We consider a general situation, with three different coupling parameters: , and , with . Under these assumptions, the transmissions are given at the end of C. Substituting these expressions in Eqs. (5) and (6), we find:

(41)

This result is a direct consequence of the factorization of the energy dependence of the transmission probabilities, which are all proportional to the same function , as shown in Eq. (76). Such factorization allows us to rewrite the Onsager’s coefficients as in Eq. (86) and derive Eq. (41). The fact that the non-local thermopowers, for example , are zero can be understood as follows. Consider first the case in which and terminal behaves as a voltage probe. If so, from the condition we derive . Due to the factorization of the energy dependence in the transmissions we obtain . Hence, does not depend on the coupling . If in particular we consider , because of the symmetry of the system under exchange of the terminal and we have . We can therefore conclude that, independently of the coupling , the probe voltage condition for terminal implies . It can be shown that such result remains valid even when but , as requested in the calculation of the thermopower . As a result, . The same argument can be repeated for the current with the terminal acting as a voltage probe, leading to .

4.2 Double Dot

Let us now consider a system made of two quantum dots in series, each with a single energy level, coupled to three fermionic reservoirs. This system is described by the Hamiltonian:

(42)

We call the hopping energy between the dots, and we assume that dot is coupled to the left lead (), dot is coupled to the right lead () and that both are coupled to a third lead () (see Fig.8). The self energies describing these couplings are:

(43)

In the wide-band approximation, we assume that these quantities are energy-independent and they can be written as purely imaginary numbers . The self energies thus become:

(44)

The retarded Green’s function of the system is then:

(45)
(46)

with

Figure 8: (Color online) Sketch of the double dot model used in the numerical simulations: two quantum dots with a single energy level are connected in series to three fermionic reservoirs , and . The chemical potential and temperature of reservoir are assumed as the reference values and . A two-terminal configuration is obtained in the case in which the coupling to reservoir (equal for both the dots) vanishes ().
(47)

Let us now define the broadening matrices as :

(48)

The matrix of transmission probability between each pair of reservoirs is then given by the Fisher-Lee formula [39, 46]

(49)

For the system under consideration we obtain

(50)
(51)
(52)

At this point, it is clear that the energy dependence of the transmission matrix cannot be factorized as for the single dot case. This model is hence the simplest in which we can observe finite non-local thermopowers and an increase of both the power and the efficiency of the corresponding thermal machine. We find that the behavior of such quantities as functions of the various parameters is qualitatively very similar to the case of the single dot, thus confirming that a third terminal could improve the performance of a quantum machine.

Since in this system all the transport coefficients are different from zero, it is instructive to study the behavior of the generalized figures of merit defined in Eq. (29). In Fig. 9 we show, in the configuration with only one positive heat flux (), (dotted line), (dashed line) and (full line). We investigate their behavior as a function of the coupling and of the total coupling . Note that in the two-terminal limit () reduces to the standard thermoelectric figure of merit , while and tend to zero. When we turn on the interaction with the reservoir (left panel), we notice that the figure of merit decreases, while the figures of merit and increase their absolute values. From the behavior as a function of the total coupling we can see that in the limit of -shaped transmission function (), the figures of merit diverge, leading to the Carnot efficiency, while in the limit of broad transmission window (), all the figures of merit go to zero and we recover the case of zero efficiency.

Figure 9: Various figures of merit (dotted line), (dashed line) and (full line) as a function of the coupling to the bottom reservoir (left panel), and as a function of the total coupling (right panel). Parameter values: , , (left panel) and (right panel).
Figure 10: (Left panel) Non local thermopowers as a function of the coupling to lead 2. The full red line corresponds to , while the dashed (blue) line corresponds to . (Right panel) Local thermopowers as a function of the coupling to lead 2. The full (red) line corresponds to , while the dashed (blue) line corresponds to . Parameter values as in Fig. 9.

4.2.1 Thermopowers

As mentioned before, the fact that the energy-dependence of the transmission matrix for the double dot cannot be factorized is sufficient to guarantee finite non-local thermopowers, as shown in the left panel of Fig. 10. As a function of , starts from zero, while starts from a finite value. This different behavior for the two non-local thermopowers is due to the different role played by in the two cases. As far as is concerned, when we set a temperature difference in lead 2, a chemical potential difference develops in lead 1 to annihilate the current that flows out of the lead 2. When the coupling goes to zero, that current goes to zero and so does the chemical potential difference . This argument does not hold for , because the temperature difference is set in lead 1, and does not control the current anymore. Therefore when the coupling approaches zero the current still have a finite value, and so the chemical potential difference needed to annihilate it. Furthermore, the local thermopowers are no more equal, as shown in the right panel of Fig. 10.

5 Conclusions

In this paper we have developed a general formalism for linear-response multi-terminal thermoelectric transport. In particular, we have worked out analytical expressions for the efficiency at maximum power in the three-terminal case. By means of two simple non-interacting models (single- and double-dot), we have shown that a third terminal can be useful to improve the thermoelectric performance of a system with respect to the two-terminal case. Moreover, we have discussed conditions under which non-local thermopowers could be observed. Our analysis could be extended also to cases in which time-reversal symmetry is broken by a magnetic field or including bosonic or superconducting terminals. It is an interesting open problem to understand in such instances both thermoelectric performance in realistic systems and fundamental bounds on efficiency for power generation and cooling.

Acknowledgements

The authors would like to acknowledge H. Linke, M. Molteni and S. Valentini for the useful discussions. This work has been supported by EU project ThermiQ and by MIUR-PRIN: Collective quantum phenomena: from strongly correlated systems to quantum simulators.

Appendix A Calculation of the transport coefficients and thermopowers

To compute the multi-terminal thermopowers defined in Eqs. (5) and (6) we have to express one of the temperatures as a function of a thermal current. For example let us start from the inversion between and . In the Onsager’s formalism this can be expressed as:

(53)

where is a permutation matrix that switches and , and are column vectors with components and