Abstract
In recent years, there has been an increasing interest in nanoelectromechanical devices, current-driven quantum machines, and the mechanical effects of electric currents on nanoscale conductors. Here, we carry out a thorough study of the current-induced forces and the electronic friction of systems whose electronic effective Hamiltonian can be described by an archetypal model, a single energy level coupled to two reservoirs. Our results can help better understand the general conditions that maximize the performance of different devices modeled as a quantum dot coupled to two electronic reservoirs. Additionally, they can be useful to rationalize the role of current-induced forces in the mechanical deformation of one-dimensional conductors.
Export citation and abstract BibTeX RIS
1. Introduction
As it is nowadays clear, electric currents can induce mechanical forces in nanodevices. Such current-induced forces (CIFs) [1–6], sometimes dubbed 'electron wind forces', have attracted increasing attention in many areas of condensed matter physics, including nanoelectromechanics [7–13], electromigration [14, 15], molecular wires and nanojuctions [16–25], the development of current-driven nanodevices [26–31], and quantum thermodynamics [32–38]. CIFs also play an important role in the understanding and development of molecular Brownian motors [39–41]. Among the most interesting examples are those where the forces are nonconservative. These types of forces may result in phenomena such as cooling, heating, or amplification of mechanical motions [8, 11, 20, 22]. Furthermore, it is also the basis of exciting proposals such as adiabatic quantum motors [27]. The most simplified description of a quantum motor consists of a system connected to leads and where a current of quantum particles drives some mechanical degrees of freedom. This is the reverse of what happens in a quantum pump, where the driving Hamiltonian's parameters, which may arise from a mechanical or an external electrical manipulation, induces a current [1, 27, 29, 33, 35, 36, 42]. Understanding the rules that dictate the interplay between the electronic and the mechanical degrees of freedom is crucial for the design of optimal electromechanical devices, and it can even contribute to the emergence of devices with novel features.
Aside from the potential applications, the type of devices discussed above may result particularly attractive from the theoretical point of view. This is so as quantum motors, quantum pumps, and some nanoelectromechanical devices can be interpreted as macroscopic manifestations of quantum behavior. Thus, these systems may provide new insights into the transition between the classical and quantum worlds, both from a dynamical [2, 29] and a thermodynamical [35–38] perspective. Moreover, the effective Hamiltonian of open quantum systems turns out to be non-Hermitian, which adds an extra richness to the problem. For example, the non-Hermiticity makes the dynamics of electrons to be affected in a nontrivial way by the change of some parameters, especially close to the so-called quantum dynamical phase transitions (QDPTs) [43–52].
In this work, we perform an in-depth study of the CIFs and the electronic-friction tensor of devices where a single energy level is relevant and, hence, they can be described by one of the most common Hamiltonian models in quantum transport [53–55], see figure 1. The main motivation behind that is to find the general conditions that lead to the improvement of the performance of different forms of nanoelectromechanical devices and quantum machines. Moreover, since CIFs may lead to the mechanical failure of conducting devices such as nanowires, the present work may also contribute to a better understanding of the circumstances that increase the chances of such failures.
The work is organized as follows. In section 2 we discuss the main theoretical aspects of the studied system. These include, a brief introduction to the QDPTs and CIFs relevant to our problem, sections 2.1 and 2.2 respectively. In section 3 we present the results of this work and in section 4 we highlight the main findings and discuss their importance.
To avoid overwhelming the readers with the derivations of the many analytic expressions, we put all the nonessential comments and mathematical details in the appendices from
2. General theory
As we will see, the understanding of CIFs and electronic friction requires the comprehension of the different quantum dynamical regimes of the electrons. These regimes are separated by QDPTs, abrupt change in the decay dynamics of particles. As we will also see, this is intimately related to the maximization of the work done by CIFs. For that reason, in the first part of this section, we briefly review the QDPTs that our model Hamiltonian may undergo, before going to the theoretical description of CIFs.
2.1. Quantum dynamical phase transitions
The intuitive idea of dynamical phase transitions is better understood by considering the classical problem of a damped harmonic oscillator in the absence of external forces. There, the oscillator presents two well defined dynamical regimes, damped and overdamped motions, which typically can be reached by moving a single parameter [43, 56]. In this case, there is an analytical discontinuity in the plot of some dynamical observables, like the oscillation frequency, versus the control parameter. Due to its similarity with thermodynamical phase transitions, the phenomenon is known as dynamical phase transitions [43, 56].
In quantum mechanics, the same kind of phenomenon appears. There, the movement of a single parameter of the Hamiltonian may produce abrupt changes (non-analytical) in the decay dynamics of quantum particles or in their spectra [44, 45, 47–49, 52, 57].
For time-independent Hamiltonians, QDPTs are usually analyzed through the poles of the retarded (or advanced) Green function (or ) [44],
This is natural as all spectral and dynamical properties of the system can be obtained from the elements of G r . For instance, the local density of states (LDOS) at a site n of a tight-binging Hamiltonian is given by:
On the other hand, the survival probability of a particle in the nth site, , can be expressed as [44]
where Θ(t) is the Heaviside step function.
2.1.1. Tight-binding model
In this paper, we will consider the tight-binding system depicted in figure 1. It consists of a single energy level with energy Ed coupled, with hopping constants VL and VR , to two semi-infinite tight-binding chains, with site energy E0 and hopping V0. This Hamiltonian represents a minimum model for a quantum dot coupled to two reservoirs and is widely used in the context of quantum transport. The total Hamiltonian reads
where E(0) = Ed and E(n) = E0 for n ≠ 0, and the hopping constants V(−1) = VL , V(0) = VR , and V(n) = V0 for . Note that we have disregarded the spin degree of freedom, which implies that our model is only valid in the absence of magnetic fields. We will keep this assumption throughout the manuscript.
For convenience, we will define as the Hamiltonian of the system which includes the first sites of each semi-infinite chain representing the leads. Then, the three by three matrix H S contains the site energies E(−1) = E0, E(0) = Ed , and E(1) = E0, as well the couplings VL and VR .
The matrix representation of the eigenvalue equation , which in principle is of infinite dimension, can be reduced to an effective system of finite dimension by using the decimation technique [54, 58]. There, the sites connected to leads are corrected by a self-energy associated to the G r (ɛ)
The real (Δ) and imaginary (Γ) parts of Σ are
where ɛ − E0 ⩾ 2V0 for case A, ɛ − E0 ∈ [−2V0, 2V0] for case B, and ɛ − E0 ⩽ −2V0 for case C. The region where Γ ≠ 0 (case B), the band, corresponds to the energy regions where excitations propagate freely inside the leads. Using the above self-energy we can define the effective Hamiltonian , where is given by . In matrix form is
By means of equation (1), can be used to calculate G r , while . The resulting eigenvalue equation is now finite and exact. The price to be paid is that now the coefficients of are energy-dependent. Notice that, because , the effective Hamiltonian becomes non-Hermitian, and thus, its eigenvalues, or equivalently, the poles of the associated Green's function G r are not necessarily real. Importantly, these poles can change abruptly with the parameters of the Hamiltonian and, in consequence, also the LDOS [see panels from (c) to (e) in figure 2] and the dynamics of the system [see panels from (f) to (h) in figure 2].
Download figure:
Standard image High-resolution imageFinally, one last detail. Due to technical reasons, we perturbatively couple a third lead to the system, see Γη in equation (8). This reservoir is modeled by a self-energy in the wideband approximation Ση = −iΓη with Γη positive. The purpose of this extra reservoir is to add a small broadening to the LDOS and to guarantee an occupation to the system. Both effects are particularly important when the eigenenergies of the effective Hamiltonian lay outside the band or when VL = VR = 0. Despite this, in all calculations, we take the limit Γη → 0, and thus its effects on the poles structure is negligible.
2.1.2. Poles and dynamical phase transitions
The poles of the system correspond to the singular points of the Green's function. The specific tight-binding model described in the previous section has two poles given by the solutions of the secular equation . Only three variables are really necessary to describe such poles, Ed , VL , and VR . The effect of E0 is only to shift the poles' position while V0 gives the energy scale. For that reason, we define the dimensionless parameters
where and . The real and imaginary part of the dimensionless poles are, respectively [46, 47],
for v2 ≠ 1. The first case of and corresponds to the condition , while the other one corresponds to the condition .
In figures 2(a) and (b) we show the real and imaginary part of the poles, p,±, as a function of the dimensionless parameter d . In the lower panels of the figure, see panels (c)–(h), we show the typical spectral and dynamical characteristic of three types of poles defining the resonant state, the virtual state, and the localized state.
Resonant states. They correspond to poles with a nonzero imaginary part. See the green dot marked as (1) in figure 2(a). When the system presents such a pole, the LDOS is typically a Lorentzian peak centered at approximately Re(p ) and whose width is given by Im(p ). This description is particularly accurate when Re(p ) is close to the center of the band. However, when Re(p ) approaches one of the band edges the maximum of the LDOS is shifted toward the closest band edge and its shape deviated from a perfect Lorentzian function. After a short quadratic decay, the dynamics of such systems typically shows an exponential decay [59].
Localized states. They correspond to poles with Im(p ) = 0 and where the Re(p ) lays outside the band (|Re(p )| > 2). For instance, see the green dot marked as (3) in figure 2(b). When the system presents such a pole, its LDOS typically shows a very narrow peak outside the band (a Dirac's delta in the limit Γη → 0). The survival probability may show a small decay at short times but then it remains constant with time.
Virtual states. They also correspond to poles with Im(p ) = 0 and |Re(p )| > 2, see the green dot marked as (2) in figure 2(b). However, they are poles of the nonphysical Riemann sheet of Σr , see references [44, 46] for more details. Due to this, there is not a peak in the LDOS at Re(p ). However, the effect of these poles on the LDOS within the band is the same as that of localized states. There is an accumulation of states near the closest band edge, which grows as the pole approaches this band edge. See the examples shown in panels (d) and (e).
The three type of poles described above determine two QDPT namely the resonant-virtual and the virtual-localized transitions. The resonant-virtual QDPT is given by [46, 47]
while the virtual-localized QDPT is given by [46, 47]
A more detailed classification may lead to another QDPT as resonant states can be subdivided into two additional type of poles, see references [44, 46]. However, for the purpose of this work this subclassification is irrelevant and will be ignored.
One final remark is still in order. In more general scenarios of a quantum dot coupled to m identical semi-infinite chains with coupling Vi , the corresponding poles are still given by equation (10), but the quadratic coupling v2 should be replaced by
2.2. Current-induced forces
When a current of quantum particles is coupled to a mesoscopic mechanical device, the classical dynamics of the mechanical part can be described by an effective Langevin equation [1, 36]:
Here, on the left-hand side we have the terms corresponding to the classical degrees of freedom, where the coordinates are associated with masses Mν , and accounts for any external forces. The terms on the right-hand side of equation (14) give the CIFs, or the forces arising from the interaction between the mechanical device and the quantum particles (electrons in our case). The term ξν describes the quantum fluctuations of the CIFs while is the mean value of the force operator
where is the electronic Hamiltonian.
2.2.1. CIFs in the Keldysh formalism
The mean value of the force operator can be calculated in terms of Green's functions by
where and is the lesser Green's function. This function, which for shortness we omitted its energy and time dependence (), is indeed the Wigner transform of what generalizes the density matrix and is called the particle propagator [1, 53]. The closely related Green function is dubbed the hole propagator, since the order of the creation and annihilation operators are reversed. Both functions are directly linked to observables and kinetic properties, such as particle densities and currents [53, 60].
In complex scenarios like the one treated here, where the time-dependent Hamiltonians are slowly varying functions of some parameters, the function can be calculated by resorting to an adiabatic expansion in terms of the time-dependent parameters [53, 60]. There, is given, up to first order, by [1],
where G < is the frozen lesser Green's function, given by
Here, G r/a are given by equation (1) and Σ< is the lesser self-energy, . The latter contains the information of the reservoirs, where fα is the Fermi–Dirac distribution function of reservoir α with chemical potential μα and temperature Tα . The function Γα is the imaginary part of the self-energy of the reservoir α, and describes the escape rate from the region of the contact in units of ℏ. Importantly, in deriving equation (17) it was assumed that the self-energies Σ< and Σr/a are time-independent. However, as we included the first sites of the leads into the effective Hamiltonian, equation (8), a variation of the couplings VL and VR does not change the self-energies.
Then, under the adiabatic approximation, is
where the 'frozen' force or, from now on, simply the force, is
The terms γνν' are the components of the electronic-friction tensor γ . The diagonal elements of γ provide the mechanical dissipation while the nondiagonal components are Lorentz-like terms which allow energy transfer between modes. This tensor can be linearly decomposed into a symmetric and an antisymmetric contribution γ = γ s + γ a . The antisymmetric component is finite only for nonzero voltage biases or temperature gradients between the reservoirs. Thus, close to equilibrium conditions only the symmetric contribution of the electronic-friction tensor needs to be taken into account.
2.2.2. Expansion of the CIFs
In this work, we are interested only on the leading orders of the expansion of in terms of the different nonequilibrium sources δμ, δT, and ,
The equilibrium contribution of the force,
is conservative and thus can be written as the gradient of a potential, . As we show in appendix
where f0 is the equilibrium Fermi–Dirac distribution function given by μ0 and T0, the equilibrium chemical potential and temperature respectively.
Comparing equations (21) and (19), it is obvious that . Therefore, only the equilibrium contribution of the symmetric component of γ is relevant for the present work, where and
Here, are the equilibrium greater (>) and lesser (<) Green's functions. The above equation can be rewritten, see appendix
where A is the spectral function, . This equation shows that only states with energies close to the Fermi energy contribute to the equilibrium electronic-friction tensor. Those states are also responsible for the electric current [54] and the nonequilibrium contribution of the CIFs, as we will see afterward. Therefore, equation (25) highlights a fundamental physical connection between these quantities. In the low temperature limit, equation (25) becomes
Stochastic forces, in equation (14), can be obtained by using the above equations and the fluctuation–dissipation relation [1], given by
where kB is the Boltzmann constant and it is assumed that stochastic forces are described by .
At low voltage biases and temperature gradients, the nonequilibrium forces are given by
which in the low-temperature limit (see appendix
Here, δμα = μα − μ0, δTα = Tα − T0, and the expression is only valid for δTα ⩽ T0.
2.2.3. Work
One of the main motivations of this work is to study the general conditions that maximize the performance of different nanoelectromechanical devices and quantum machines. In this respect, one of the central quantities to evaluate is the work W done by CIFs during cyclic motions. If one is dealing with systems where only two or three parameters are required to describe their motion, then it is useful to study the curl of the CIFs, defined as
where ρ, ν, and ν' are the indexes of the coordinates and ν < ν'. Now, the work can be obtained from
where the integration is done on the surface S enclosed by the closed curve C that describes the cyclic motion.
Now inserting equation (29) into equation (30), gives
where we used , ∂ν' Λν = ∂ν Λν', and . Equation (32) gives the work done per unit area in the low temperatures limit and for small bias voltages or temperature gradients.
2.2.4. CIFs in our model Hamiltonian
Let us return to the tight-binding model described in section 2.1.1. For the analysis of the forces, especially for the equilibrium forces, it is crucial to have a definite occupation of the system, even for localized states or when VL = VR = 0. As discussed in section 2.1.2, when the system is at the localized state, the poles of G r , or the eigenenergies of H eff, lay outside the band. At this condition, the imaginary part of the self-energies, Γ, is zero. The same happens for VL = VR = 0. This implies the nonphysical condition that localized states do not contribute to the forces, or even worse, that isolated systems do not have forces at all. To solve this problem, we added the third reservoir (η) to the model, assuming its occupation is given by the equilibrium Fermi function f0 = (fL + fR )/2, and taking the limit Γη → 0 in all calculations.
Given the Hamiltonian of equation (8) and the definition of
G
r/a
(equation (1)) and
G
< (equation (18)), we obtain a closed-form expression for the Green functions, (see appendix
According to equation (16), the forces depend on how the elements of the electronic Hamiltonian are affected by the movement of the mechanical degrees of freedom, but this is system-dependent. Therefore, to gain generality, we take as 'mechanical' variables the same parameters of the Hamiltonian (Ed
, VL
, VR
), which leads to dimensionless forces Fν
, see appendix
where . Notice that while forces may depend on the chosen variables, the work in equation (31) is independent of them, as long as the curve C remains the same, see appendix
3. Results
3.1. Equilibrium potential
Using equation (23) and the expression for Gr/a
obtained in appendix
where
Here, we used and = (ɛ − E0)/V0. It is worth mentioning that equation (34) is also valid for systems coupled to more than two reservoirs where v2 is given by equation (13).
In figure 3 we show the potential energy for some representative values of the normalized Fermi energy . The green and cyan lines superimposed to the plot show the different QDPTs. As can be seen, there is not a clear correlation between them and the equilibrium potential. However, the figures show a strong dependence of Ueq with the Fermi energy. This is important since then, e.g., gate voltages can be used to control the equilibrium position of the system or, even up to some extent, its dynamics. In this regard, having a simple but general expression for Ueq may result particularly useful.
Download figure:
Standard image High-resolution image3.2. Curl and electronic friction
The matrices Λν , with our choice of mechanical coordinates, see section 2.2.4, are
The spectral function can be calculated from . Using this in equation (32), it is possible to obtain a simple expression for the elements of the curl of the forces (see appendices
Here we take μ0 = (μL + μR )/2, T0 = (TL + TR )/2, δμ = μL − μR , and δT = TL − TR . The transmittance between reservoirs L and R is , is the LDOS of the dot, and
The curl of the force represents the work per unit area where, in our case, the unit area is given in units of energy and therefore the curl has units of one over energy.
As there are three components of the curl (equation (37)) and two coefficients of the expansion (one for δT and one for δμ), there are a total of six components of the expansion of the curls. However, due to the symmetry of the problem, the coefficients of the expansion of are equivalent to the coefficients of the expansion of with vL and vR interchanged. Therefore, we are going to consider only four coefficients namely , , , and , where
When one multiplies the coefficients or (where ρ = Ed , VL , or VR ) by δμ or respectively, the result gives the low temperature limit of the work done per unit area for a cyclic movement of the other two variables.
Having a simple expression for the coefficients or may be important when studying different forms of nanoelectromechanical devices, adiabatic quantum motors, or quantum heat engines. Let us recall that due to Onsager's reciprocal relations, the coefficients of the expansion of the work in terms of δμ and δT gives, respectively, the charge pumped by adiabatic quantum pumps and the heat pumped by adiabatic quantum heat pumps [27, 32]. Therefore, the expressions found for the coefficients and the conclusions regarding their general behavior are also valid for such systems.
In figure 4 we show some examples of the behavior of the coefficients of the expansion of the curl. In general, their behavior is highly dependent on the Fermi energy, just as with the case of Ueq. This feature may be useful, e.g., to control the dynamics of a system by using a gate voltage as a knob. Indeed, it is possible to change, in a given region of the parameter's space, the sign of the coefficients, and thus the preferred direction of motion of the system by changing F .
Download figure:
Standard image High-resolution imageThe green solid lines and the cyan dashed lines indicate the QDPTs in figure 4. Although not obvious at first glance, we find that the behavior of the W coefficients is related to the type of poles of the Green's function describing the effective Hamiltonian, equation (37). This is so as the shape of the LDOS is determined by the type of pole, which, in turn, depends on the parameters of the Hamiltonian.
The coefficients depend on the square of the LDOS (or its derivative). Note that, in our system the transmittance is proportional to the LDOS within a wide band approximation [53]. Therefore it is expected a coincidence between the regions with the maximum value of the coefficients and the regions with a maximum value of the LDOS, see equation (37). According to the discussions of section 2.1, for resonant states (see the region enclosed by the cyan dashed line in figure 4), the maximum of the LDOS approximately coincides (for poles not too close to the band edges) with the real part of the pole, see figure 2(c). In such a case it is expected a maximum of the coefficients for F = Re(p ), or
where we used equation (10). The above equation is plotted in figures 4(b) and (d) as violet dashed lines. As can be seen, the regions with a maximum value of the coefficients are close to the line given by equation (40).
For the coefficients the same analysis can be done except that they depend on the square of the LDOS multiplied by the term ( − d ), which shifts the maximum of the function from Re(p ). In this case, the maximum value of the coefficients should be close to the regions where the following function is maximum
Here, we used the fact that in our system the transmittance is proportional to the LDOS. We assumed the pole is close to the center of the band, and thus Γ can be taken in a wide band approximation and the LDOS has a Lorentzian shape centered around Re(p ) and with a width given by Im(p ). And we applied all this to equation (37). Note that p is a function of d , vL , and vR , see equation (10). The above condition, numerically found, is plotted in figures 4(a) and (c) as violet dashed lines. As can be seen, the regions with a maximum value of the coefficients are close to this line.
All the above discussions are valid as long as |F | ≪ 2. Let us recall that only when p is far from the band edges, the LDOS can be described by a Lorentzian function centered around Re(p ). When Re(p ) approaches one of the band edges (|Re(p )| ≈ 2), the LDOS is largely distorted exhibiting a peak with a maximum shifted toward the closest band edge. If we move the pole even further, it will correspond to a virtual state (see figure 2(a)). For virtual states (|Re(p )| ⩾ 2 and Im(p ) = 0), the LDOS shows a maximum almost at the band edges ≈ ±2, see figure 2(d). The height of this maximum grows until the virtual-localized QDPT is reached (where the maximum is exactly at = ±2) and then it decreases. Therefore, in all these cases the maximum of the LDOS is expected to approximately coincide with one of the band edges. This is the reason why for |F | ≈ 2 the maximum value of the curl coincide approximately with the virtual-localized QDPT, see figure 5. For localized states, obviously there is also a maximum outside the band, see figure 2(e). However, as discussed in section 2.2.4, states with energies outside the band do not contribute to nonequilibrium forces. Note also that the transmittance is zero outside the band and thus according to equation (37).
Download figure:
Standard image High-resolution imageIn summary, as a rule of thumb when |F | ≪ 2, the regions of the parameter space with maximum values of the coefficients are given by equation (40), while for the coefficients, these regions are shifted and given by the maximum of equation (41). On the other hand, when |F | ≈ 2, the regions of the parameter space with the maximum values of both coefficients ( and ) approximately coincide with the virtual-localized QDPT, equation (12).
Here, we also obtained closed-form expressions for the elements of the electronic-friction tensor
γ
, see appendix
Download figure:
Standard image High-resolution imageOne may wonder, what are the regions in the parameter space that maximizes the W coefficients independently of F , i.e., if one could move F at will. This is shown in figures 7 and 8. We do not show the W(δT) coefficients as their behavior is approximately the same as that of the W(δμ) coefficients. As can be seen, the regions that maximize the W coefficients coincide in general with the virtual-localized QDPT. Note however that in figure 7 the region vL ≈ vR ≈ 0 presents values of similar to those of the virtual-localized QDPT. Regarding the elements of the electronic-friction tensor, they roughly follow the behavior of the W coefficients.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageFinally, to help better understand how the above results can be applied to real physical systems, we added a discussion about a concrete tight-binding example in appendix
4. Conclusions
When designing or studying a nanoelectromechanical device or a quantum machine, it is crucial to understand the effect of the system's parameters on their performance. Here, we carry out a thorough study of the CIF and the electronic-friction tensor within one of the most common Hamiltonian models in quantum transport [53–55]. It is important to point out that any single-particle Hamiltonian, where there is only one relevant state, can be reduced to the model used here by means of a decimation procedure, see, e.g. reference [54].
We derived analytical formulas for the equilibrium potential (equations (23) and (35)), the low-temperature limit of the electronic-friction tensor (equation (26)), and the low-temperature limit of the curl of the CIFs up to leading order in δT and δμ (equations (37) and (32)). We showed the connection between QDPTs, the curl of the CIFs, and the electronic-friction tensor. Using this as a guide, we found some rule of thumbs that can be useful to quickly identify the regions of the parameters space that should be enclosed to maximize the work done by the device. Moreover, because of the strategy we used to carry out the analysis, our results are independent of the details of how the mechanical and electronic degrees of freedom are coupled.
Again, we want to emphasize that, due to Onsager's reciprocal relations, our analysis of the coefficients of the expansion of the curl ( and ) and the formulas shown in equations (32) and (37), are also valid for adiabatic quantum pumps and quantum heat pumps [27, 32].
The present work may also help to connect the configuration of conducting devices with current-induced structural-failures. Basically, our simple Hamiltonian also represents a minimal model of an impurity in a one-dimensional conductor. Therefore, according to our results, some particular values of the parameters of the system should dramatically increase nonconservative forces and the curl of the force, which should ultimately lead to a mechanical failure. That reasoning, of course, depends on the vibrational modes involved and the dynamical feedback between them and the electronic currents [16]. However, it is an interesting direction for further research.
Other appealing directions of extending this work include analyzing other common Hamiltonian models, such as double quantum dots, and studying systems within the Coulomb blockade regime of quantum transport.
Acknowledgments
We acknowledge financial support by Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET); Secretaría de Ciencia y Tecnología de la Universidad Nacional de Córdoba (SECYT-UNC); and Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT, PICT-2018-03587).
Data availability statement
No new data were created or analysed in this study.
Appendix A.: Equilibrium potential
By writing the Fermi functions as fL/R = f0 + ΔfL/R , one can split the zero order term of the adiabatic expansion of the CIFs Fν into the equilibrium () and nonequilibrium () contributions
Given their definition, the following relation holds for the Green functions
Therefore, equilibrium force can be written as
Now, let us take a matrix U which diagonalize G r ,
Obviously, U also diagonalize . Then, using , one finds
The elements of D r are the eigenvalues of G r , thus
A similar argument can be used for G a . The equilibrium force results in
where one can identify the equilibrium potential Ueq as
Note that , and then one can write
Taking and considering χ as the principal value of the logarithm, gives
Using this, the equilibrium potential can be written as
Note that it is not necessary to integrate from −∞ to ∞ but only over energies where Im( G r ) ≠ 0 (where χ(ɛ) ≠ 0). As the function 'arctan' is bounded and monotonically increasing, the following inequalities hold
where the symbol ∗ means that the integrals only cover the energy ranges where Im( G r ) ≠ 0, or, in other words, the integrals only run over energy ranges where the leads support propagating states, or the system presents localized states. Then, the above equation shows that the equilibrium potential is bounded by the occupation of the reservoirs and the system.
Appendix B.: Equilibrium friction tensor
As mentioned, here we are interested only in the symmetric component of the electronic-friction tensor at equilibrium, equation (24). Under this condition, the lesser and greater Green's functions are given by [1, 53]:
where A is the spectral function, . Using this we can write equation (24) as
where we used the notation .
Employing the cyclic property of the trace of matrix multiplications, one can readily show
The above equation can be used to further prove
Combining equations (B2) and (B3) we find
Now, after some algebra and using equations (B2) and (B4) the following relation is obtained
Inserting equation (B5) into equation (B1) yields
As is bounded, and , then
Using this in equation (B6) and after some algebra we finally find
Appendix C.: Expansion of Fne
Let us consider the contribution of the α lead to the CIF (Fα ),
and define the compact support function such that
where fα is the Fermi distribution function of reservoir α and
Expanding Fα up to second order in μα and Tα and taking the small temperature limit of Tα yields
As in the main text, here we used δTα = Tα − T0 and δμα = μα − μ0, where T0 and μ0 are the average temperature and chemical potential of the reservoirs connected to the system. Note that the first two terms of the right-hand side of equation (C2) will contribute to the equilibrium force while the last two terms are second order in an expansion in terms of δμα and δTα . Therefore, by inserting equation (C1) into the above equation, taking only the first-order terms and summing up all Fα contributions, one readily arrives at equation (29).
Appendix D.: Green functions in our system
The determinant of the effective Hamiltonian given in equation (8) is
The retarded Green function is
For convenience, we changed the notation for the indexes of the elements of G , with respect to that of equation (4) in the main text. Thus, e.g., is the LDOS of the dot (site '0' according to equation (4)), while is the LDOS of the first site of the left chain (site '−1' according to equation (4)). In all the appendixes, we followed the present convention.
Using equation (D1) one can readily calculate the elements of . Written them in terms of the dimensionless quantities (, d , vL/R , , and ) they are
The advanced Green function is calculated from , while the lesser Green function is given by
where . We will split G < into two contributions ( and G η )
where
Here, fL/R = f0 ± Δf with f0 = (fL + fR )/2 and Δf = fL − fR . We also assumed the third lead is at equilibrium with fη = f0. Then, the elements of and are
and
where we identified the equilibrium (eq) and nonequilibrium (ne) contributions to . The equilibrium contributions only contains f0 terms while nonequilibrium contributions only contains Δf terms.
Therefore, the nonequilibrium G < is given by
while the equilibrium contribution is
Appendix E.: Mechanical variables and CIFs
Let and be two sets of different mechanical variables related through the expression , which we assume continuous but not necessarily lineal. Then, CIFs can be written in terms of as or in terms of as . Both descriptions are related through
where is the covariant derivative.
The work done by the forces over a closed loop C in the space of the variables is independent of the choice of the sets of mechanical variables used to describe the system, as can be readily shown
In the main text, we used this property to choose a set of convenient variables to become independent of the particular way in which mechanical variables affect the electronic Hamiltonian.
Appendix F.: General expression of the curl
Our goal here is to obtained a closed-form expression for . We start by evaluating the derivative of the force ∂ν' Fν = ∂Fν /∂Xν'
The derivative of the lesser Green function is
where we assume ∂ν' Σ< = 0 and used ∂ν' G r/a = G r/a Λν' G r/a . Then, the derivative of the trace is
where we used and to make . Using this, the derivative of the force can be written as
Now as ∂ν' Λν = ∂ν Λν', the following holds
where, to compact the notation, we define
Finally, the curl of the force yields
where ρμν is the Levi-Civita symbol.
Appendix G.: Elements of the curl in the tight-binding system
We start by defining the index order of the variables
With this order, the following relations hold
where .
Defining and ΣL/R = ΔL/R − iΓL/R , one can deduce the following
This relation will be useful afterward. Another formula that will be useful is the expression for the elements of G<. For simplicity, we will assume Γη = 0 from the beginning, as we are interested only in the nonequilibrium contribution of the force, see appendix D. Then, the lesser self-energy yields
It will be helpful to define . In our case, the elements of G ≲ are
Note that . In the following, we will use all the above properties and definitions to derive the expression for each of the elements of the curl.
G.1. Element ρ = Ed
This element of the curl is given by
where
Now, the terms of the right-hand side of the above equation are
and therefore
Here, we define to compact the notation. As the retarded Green functions is symmetric in our case, the following relations holds , a31 = a13, a32 = a23, and a21 = a12. Using this, we obtain
where we defined with given by equation (G3). After some algebra, the right-hand side terms of the above equation can be written as
Using the expression for G r given in appendix D and equation (G1) we finds
and
Now introducing the transmittance [53, 54]
and using equations (G5)–(G7) into equation (G4) gives
Using the expressions for the elements of G r , it is not difficult to prove:
Therefore, the element ρ = Ed of the curl can be written as
where Nd is the LDOS of the dot
G.2. Element ρ = VL
This element of the curl is given by
where
Now, the terms of the right-hand side of the above equation are
Therefore
Using this we obtain
where the elements of the matrix G≲ are given in equation (G3). After some algebra it is not difficult to show
Using the expression for G r given in appendix D and equation (G1) one finds
Using equations (G8)–(G10) into equation (G10), we obtain
Therefore, the element ρ = VL of the curl can be written as
G.3. Element ρ = VR
The expression for the element ρ = VR of the curl is obtained by following a similar procedure to that of but replacing VL by VR . The result is
G.4. Final expression of the curl
The results of appendices G.1–G.3, can be written in the compact form
where
Appendix H.: Electronic-friction tensor in the tight-binding system
The three by three equilibrium friction tensor is symmetric, thus only six elements are necessary to describe it. To derive the expression for the elements of γ we used the following relations
and
The above come from the definition of the spectral function, , and the matrices Λν , equation (36). To evaluate the elements of the above matrices one can use the explicit expressions for the elements of the retarded Green's function given in appendix D. After some algebra we find
Using the above formulas, and the expressions for the transmittance and the LDOS (equations (G8) and (G9) respectively), the elements of the electronic-friction tensor yield:
H.1. Element
where we used .
H.2. Element
H.3. Element
Due to the symmetry of the problem, this element is equal to but replacing VL by VR in the formulas.
H.4. Element
where we used .
H.5. Element
Due to the symmetry of the problem, this element is equal to but replacing VL by VR in the formulas.
H.6. Element
Appendix I.: A concrete tight-binding example
In this section, we will analyze a concrete tight-binding example to help the readers understand how to use our results in physical systems. Our example consists of an atomic wire with a bend where the corner atom is free to move in the x–y plane. This example was already studied in reference [26] where the authors found that nonconservative CIFs drive the movement of the corner atom in a periodic orbit. This kind of system can be considered as an example of an adiabatic quantum motor [27]. We will not enter into the experimental details of how to realize such a proposal as this is beyond the scope of the present work. However, a good candidate to further study the proposal could be a quantum dot mounted on a nanopillar and coupled to two leads forming an angle. Another possibility would be to replace the nanopillar with some appropriate organic molecule.
The electronic Hamiltonian of the example is the same as that shown in equation (4) and figure 1. The fact that leads form an angle does not change the electronic Hamiltonian. For simplicity, we will assume the angle α between the leads is 90°, where the L-lead goes along the 'y' axis while the R-lead goes along the 'x' axis. Typically the hoppings vary exponentially with the distance between atoms, but here we will take the linearized version of this, such as
where δx and δy are the deviations from the equilibrium position of the dot along the 'x' and 'y' axis respectively. Note that the above equation can be readily modify to take into account α different from 90°. Just to put some numbers, we will take the same tight-binding parameters as in reference [26]: E0 = 0 eV, Ed = 2 eV, V0 = −3.88 eV, Å−1, and ɛF = E0 (the leads' bands are half-filled). This corresponds to the renormalized parameters d = 2/3.88, F = 0, and . According to equation (33), the CIFs Fx and Fy acting on the x and y directions are related to the dimensionless forces and by
For small displacement from the equilibrium position, the work done by CIFs in a periodic orbit of the dot's position can be written as
where ∬dx dy is the area A enclosed by the movement of x and y (a figure-of-eight orbit gives a zero area). As highlighted in appendix E, W is independent of how the variables of the Hamiltonian (VL and VR ) depends on the movement of mechanical variables (δx and δy). Thus, assuming Ed is constant along the trajectory and taking the small temperature limit and for small bias voltages, the work yields
For the chosen parameters, we numerically found the optimal working point (values of vL and vR that maximizes for F = 0). These values are approximately vL = vR = 0.51, which corresponds to .
Taking F ≈ |2|, which dramatically increases according to the discussion of section 3, the optimal working point is near the virtual-localized QDPT, see equation (12). This occurs for
where the minus sign corresponds to F ≈ −2 (Fermi energy close to the lower band edge) while the plus sign corresponds to F ≈ 2 (Fermi energy close to the upper band edge).