Fluctuation-driven Coulomb drag in interacting quantum dot systems

TL;DR

This paper theoretically investigates Coulomb drag in a double quantum-dot system, revealing how charge fluctuations and Kondo effects influence drag behavior and offering predictions for experimental verification.

Contribution

It introduces a detailed theoretical model of Coulomb drag driven by charge fluctuations, including the effects of Kondo correlations and zero-bias anomalies in quantum dot systems.

Findings

Drag is driven by charge fluctuations in the drive quantum dot.

Kondo correlations suppress drag at low bias.

Maximum drag occurs when interdot interaction matches lead couplings.

Abstract

Coulomb drag between nanoscale conductors is of both fundamental and practical interest. Here, we theoretically study drag in a double quantum-dot (QD) system consisting of a biased drive QD and an unbiased drag QD coupled via a direct interdot Coulomb interaction. We demonstrate that the Coulomb drag is driven by the charge fluctuations in the drive QD, and show how the properties of the associated quantum noise allow to distinguish it from, e.g., shot-noise driven drag in circuits of weakly interacting quantum conductors. In the strong-interaction regime exhibiting an orbital ("pseudospin") Kondo effect, the drag is governed by charge fluctuations induced by pseudospin-flip cotunneling processes. The quenching of pseudospin-flip processes by Kondo correlations are found to suppress the drag at low bias and introduce a zero-bias anomaly in the second-order differential transconductance. Finally, we show that the drag is maximized for values of the interdot interaction matching the lead couplings. Our findings are relevant for the understanding of drag in QD systems and provide experimentally testable predictions in different transport regimes.

Figures

Equations

$$P_2(\omega )=\int dt{e}^{i\omega t}⟨{\bar{n}}_2(0){\bar{n}}_2(t)⟩,$$

$$I_{\alpha j}$$

$$\times \left[{\Sigma }_j^{>}(\omega ){\Sigma }_{\alpha j,\mathrm{tun}}^{<}(\omega )-{\Sigma }_j^{<}(\omega ){\Sigma }_{\alpha j,\mathrm{tun}}^{>}(\omega )\right],$$

$${\Sigma }_{j,\text{xc}}(\tau ,{\tau }^{\prime })=U^2{G}_j(\tau ,{\tau }^{\prime })P_{\bar{j}}(\tau ,{\tau }^{\prime }),\bar{j}\ne j,$$

$$I_{\text{drag}}$$

$$\times \left[{\Gamma }_{L1}(\omega -{\omega }^{\prime }){\Gamma }_{R1}(\omega )-{\Gamma }_{L1}(\omega ){\Gamma }_{R1}(\omega -{\omega }^{\prime })\right]$$

$$\times f_1(\omega -{\omega }^{\prime })\left[1-f_1(\omega )\right]P_2^{<}({\omega }^{\prime }),$$

$$I_{\text{drag}}$$

$$\times [f_1(\omega )-f_{\beta 2}(\omega )],$$

Full text

Introduction.—In recent years, systems of closely-spaced quantum-dots (QDs) or nanoscale conductors have been demonstrated to be hosts of novel transport mechanisms which can be exploited in, e.g., quantum information Burkard et al. (1999), thermoelectrics Harman et al. (2002); Koski et al. (2015); Thierschmann et al. (2016), and energy harvesting Thierschmann et al. (2015). Phenomena of fundamental importance such as, e.g., orbital Kondo physics Borda et al. (2003); Jarillo-Herrero et al. (2005); Amasha et al. (2013) and attractive electron-electron interactions Hamo et al. (2016) have been demonstrated. For such effects, the Coulomb interaction between the constituents is essential, and its appreciable size in nanoscale systems has driven experiments into hitherto inaccessible regimes.

This development has led to a revival of the phenomenon of Coulomb drag Narozhny and Levchenko (2016) in nanoscale systems with several reports of drag currents—i.e., a current induced in an unbiased drag system by its Coulomb interaction with a biased current-carrying drive system—in Coulomb-coupled double quantum dot (DQD) systems in 2DEGs Shinkai et al. (2009); Keller et al. (2016) as well as in graphene and carbon nanotubes Bischoff et al. (2015); Volk et al. (2015). Theoretically, drag in quantum conductors Mortensen et al. (2001, 2002); Levchenko and Kamenev (2008); Borin et al. (2019) and QD systems Sánchez et al. (2010); Kaasbjerg and Jauho (2016); Keller et al. (2016); Lim et al. (2018) has been studied thoroughly, with indications of an intimate link Aguado and Kouwenhoven (2000); Levchenko and Kamenev (2008) between drag and the quantum noise of nonequilibrium fluctuations Nazarov and Blanter (2009); Clerk et al. (2010).

So far, theoretical works on drag in QD systems have been based on master-equation approaches in the Coulomb-blockade regime $\Gamma<k_{B}T,U$ Sánchez et al. (2010); Kaasbjerg and Jauho (2016); Keller et al. (2016); Lim et al. (2018), which do not apply to quantum coherent transport and provide no direct interpretation in terms of quantum noise. In addition, experiments have demonstrated Coulomb drag across different transport regimes Bischoff et al. (2015); Keller et al. (2016), why further theoretical investigations may advance our understanding of Coulomb drag Bischoff et al. (2015); Keller et al. (2016) and related transport effects Harman et al. (2002); Koski et al. (2015); Thierschmann et al. (2015), as well as the link to quantum noise.

In this work, we apply the Keldysh nonequilibrium Green’s function (NEGF) formalism Haug and Jauho (1998) to the description of Coulomb drag in QD systems across interaction regimes, covering weak ($\Gamma>U$), to intermediate ($\Gamma\sim U$), and strong ($\Gamma<U$) Coulomb interactions $U$, where $\Gamma$ is the overall lead coupling.

As we here demonstrate, the leading contribution to the drag is governed by the nonsymmetrized quantum noise spectrum Nazarov and Blanter (2009) of the nonequilibrium charge fluctuations in the biased drive system,

[TABLE]

where $\bar{n}_{2}=\hat{n}_{2}-\langle\hat{n}_{2}\rangle$ is the occupation of the drive dot relative to its mean value (see also Fig. 1). This is in stark contrast to the drag induced between two coherent conductors (e.g., QPCs) by a circuit environment Levchenko and Kamenev (2008), which is driven by the quantum shot noise in the drive system; i.e., current fluctuations whose noise characteristics are distinctly different from those of charge fluctuations Rothstein et al. (2009). This points to a fundamental difference between drag mediated by, respectively, direct Coulomb interactions (this study) and a circuit environment (Ref. Levchenko and Kamenev, 2008).

For strongly interacting DQD systems, higher-order processes involving tunneling events in both QDs become important leading to an orbital analog Borda et al. (2003); Jarillo-Herrero et al. (2005); Amasha et al. (2013) of the conventional spin Kondo effect Hewson (1993); Ralph and Buhrman (1994); van der Wiel et al. (2000); Rosch et al. (2001) where the QD levels play the role of a pseudospin. To lowest order in the effective exchange coupling we find that the drag is dominated by pseudospin-flip cotunneling processes corresponding to simultaneous charge fluctuations in both QD systems. The quenching of charge fluctuations in the Kondo regime is found to suppress the drag at low bias voltages, and leads to a zero-bias anomaly in the second-order derivative of the drag current with respect to the drive voltage.

Model and theory.—We consider a spinless DQD system consisting of a drag ($j=1$) and drive ($j=2$) dot coupled via an interdot Coulomb interaction $U$ and connected to separate sets of source and drain contacts as depicted in Fig. 1. The total Hamiltonian takes the form $\mathcal{H}=\mathcal{H}_{\text{leads}}+\mathcal{H}_{\text{DQD}}+\mathcal{H}_{\text{tun}}$, where the Hamiltonian of the two Coulomb-coupled QDs is $\mathcal{H}_{\text{DQD}}=\sum_{j}\varepsilon_{j}d_{j}^{\dagger}d_{j}+Ud_{1}^{\dagger}d_{1}d_{2}^{\dagger}d_{2}$. Here, $\varepsilon_{j}$ is the position of the gate-controlled energy level in the $j$th QD, and $U$ is the interdot Coulomb interaction. The contacts are described by noninteracting reservoirs, $\mathcal{H}_{\text{leads}}=\sum_{\alpha jk}\xi_{\alpha jk}c^{\dagger}_{\alpha jk}c_{\alpha jk}$, where $c^{\dagger}_{\alpha jk}$ ($c_{\alpha jk}$) creates (annihilates) an electron in state $k$ of lead $\alpha=\{L,R\}$ and system $j$ with energy $\xi_{\alpha jk}$. In the drive system, the chemical potentials of the reservoirs are given by the applied bias voltage, $\mu_{\alpha}\equiv\mu_{\alpha 2}=\varepsilon_{F}\pm eV_{\alpha}$, while the reservoirs of the drag system are kept in equilibrium with $\mu_{\alpha 1}=\varepsilon_{F}=0$. Finally, the dot-reservoir tunneling is represented by $\mathcal{H}_{\rm tun}=\sum_{\alpha jk}\mathcal{V}_{\alpha jk}c^{\dagger}_{\alpha jk}d_{j}+{\rm h.c.}$ where $\mathcal{V}_{\alpha jk}$ are the tunnel couplings.

We describe the QD system using NEGF where the contour-ordered dot Green’s function (GF) $G_{j}(\tau,\tau^{\prime})=-i\langle T_{c}{d_{j}(\tau)d_{j}^{\dagger}(\tau^{\prime})}\rangle$ is given by the usual Dyson equation with the irreducible self-energy, $\Sigma_{j}=\Sigma_{j,\mathrm{tun}}+\Sigma_{j,\mathrm{int}}$, having contributions from (i) the tunnel couplings to the leads, $\Sigma_{j,\mathrm{tun}}(\tau,\tau^{\prime})=\sum_{\alpha}\Sigma_{\alpha j,\mathrm{tun}}(\tau,\tau^{\prime})=\sum_{\alpha k}\left|\mathcal{V}_{\alpha jk}\right|^{2}g_{\alpha jk}(\tau,\tau^{\prime})$ where $g_{\alpha jk}$ is the unperturbed Green’s function of the lead $\alpha j$, and (ii) the interdot Coulomb interaction, $\Sigma_{j,\mathrm{int}}=\Sigma_{j,H}+\Sigma_{j,\mathrm{xc}}$. The latter is split into separate Hartree ($H$) and exchange-correlation (xc) parts described in further detail below. Analytic continuation onto the real-time axis is performed with the Langreth rules Haug and Jauho (1998).

The current in lead $\alpha j$ is defined as $I_{\alpha j}=-e\tfrac{dN_{\alpha j}}{dt}$, where $N_{\alpha j}$ is the total occupation of lead $\alpha j$, can be expressed in terms of the dot GF and self-energies in Fourier space as Haug and Jauho (1998)

[TABLE]

where $G^{r/a}_{j}(\omega)=[\omega-\varepsilon_{j}-\Sigma_{j}^{r/a}(\omega)]^{-1}$ are the retarded/advanced dot Green’s functions, and $\Sigma_{\alpha j,\mathrm{tun}}^{r/a}(\omega)=\Lambda_{\alpha j}(\omega)\mp i\Gamma_{\alpha j}(\omega)/2$ are the retarded/advanced tunneling self-energies, $\Lambda_{\alpha j}$ gives the shift of the levels due to the tunnel coupling and $\Gamma_{\alpha j}$ is the lead hybridization function. The greater/lesser components are given by $\Sigma_{\alpha j,\mathrm{tun}}^{<}(\omega)=-i\Gamma_{\alpha j}(\omega)f_{\alpha j}(\omega)$ and $\Sigma_{\alpha j,\mathrm{tun}}^{>}(\omega)=i\Gamma_{\alpha j}(\omega)[1-f_{\alpha j}(\omega)]$, where $f_{\alpha j}(\omega)=1/[1+\exp{(\omega-\mu_{\alpha j})/k_{B}T}]$ is the Fermi-Dirac distribution function.

From Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems) we note that the interacting self-energy is, not surprisingly, instrumental for a nonzero drag current zer . More specifically, it is the dynamic xc part of the interaction self-energy which is responsible for the drag since the Hartree part has $\Sigma_{j,H}^{>/<}=0$ Thygesen and Rubio (2008). On the other hand, the retarded/advanced Hartree self-energy, $\Sigma_{j,H}^{r,a}=Un_{\bar{j}}$, where $n_{j}=-i\int\!\tfrac{d\omega}{2\pi}\,G^{<}_{j}(\omega)$ is the occupation of dot $j$, $\bar{j}\neq j$, and $G^{<}_{j}=G^{r}_{j}\Sigma^{<}_{j}G^{a}_{j}$, is essential for capturing the Coulomb coupling between the dots as it introduces an interaction-induced shift of the QD levels, $\varepsilon_{j}\rightarrow\varepsilon_{j}+Un_{\bar{j}}$.

The effect of the Coulomb coupling is evident from the stability diagram in Fig. 2(a), which shows the sum of the linear conductances $\mathcal{G}_{1}+\mathcal{G}_{2}$, $\mathcal{G}_{j}=dI_{j}/dV_{j}$, as a function of the level positions $\varepsilon_{j}$ for a DQD system in the regime $\Gamma\sim U$. As indicated in Fig. 2(a), the stability diagram maps out the energetically most favorable occupations of the QDs, and is characterized by the Coulomb-induced honeycomb vertex (dashed lines) connecting the triple points at $\varepsilon_{j}=-U$ and $\varepsilon_{j}=0$ van der Wiel et al. (2002).

For the xc part of the interaction self-energy, we adopt the single-bubble approximation illustrated by the Feynman diagram in Fig. 2(b), and given by

[TABLE]

where $P_{j}(\tau,\tau^{\prime})=G_{j}(\tau,\tau^{\prime})G_{j}(\tau^{\prime},\tau)$ is the nonequilibrium polarization bubble of the $j$ system. The greater and lesser components of the self-energy are $\Sigma_{j,\text{xc}}^{>/<}(t,t^{\prime})=U^{2}G_{j}^{>/<}(t,t^{\prime})P_{\bar{j}}^{>/<}(t,t^{\prime})$, where $P_{j}^{>/<}(t,t^{\prime})=G_{j}^{>/<}(t,t^{\prime})G_{j}^{</>}(t^{\prime},t)$. Generalization to more complicated self-energies, such as, e.g., the $GW$ approximation Thygesen and Rubio (2008), is possible Moldoveanu and Tanatar (2009); Zhou and Guo (2019). However, the $GW$ approximation is more relevant for extended systems where screening effects are important, and is therefore not expected to affect our findings below.

In the following, we pursue both analytic perturbative and numerical nonperturbative calculations Lunde et al. (2009). In the numerical calculations, the xc self-energy in Eq. (3) is obtained from the Hartree GF. Since the latter itself depends on the dot occupations, the Hartree GF must be calculated self-consistently. We should stress that this procedure yields a conserving approximation with overall charge conservation respected sup , $\sum_{\alpha}I_{\alpha j}=0$, allowing us to define the drive and drag currents as $I_{j}\equiv(I_{Lj}-I_{Rj})/2$.

The drag current.—We start by analyzing the drag arising from the self-energy in Eq. (3). Inserting in the general expression for the current in Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems), the drag current can be written as sup

[TABLE]

where $\mathcal{A}_{j}(\omega)=-(1/\pi)\mathrm{Im}\,G^{r}_{j}(\omega)$ is the dot spectral function and $\Gamma_{j}=\Gamma_{Lj}+\Gamma_{Rj}$.

Several important observations can be made from Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems). First, in order for a nonzero drag current, the lead couplings in the drag QD must fulfill $\Gamma_{L1}(\omega)\Gamma_{R1}(\omega^{\prime})-\Gamma_{L1}(\omega^{\prime})\Gamma_{R1}(\omega)\neq 0$; i.e., they cannot be proportional $\Gamma_{L1}(\omega)\neq C\Gamma_{R1}(\omega)$. This result is consistent with previous works in the Coulomb blockade regime Kaasbjerg and Jauho (2016); Keller et al. (2016). Secondly, the drag can be understood as arising from interaction-mediated creation and annihilation of electron-hole pair excitations in the drag and drive systems, and is driven by the finite-bias correlator $P_{2}^{<}$ of the drive QD which can be identified as the quantum charge noise $P_{2}(\omega)$ in Eq. (1). At low temperature, i.e., $k_{B}T\ll eV$, where the drag QD predominantly absorbs energy from the drive QD due to Pauli blocking of emission processes, the quantum nature of the noise manifests itself in the fact that the drag is governed by the emission noise of the drive QD Gavish et al. (2000); Nazarov and Blanter (2009) given by $P_{2}(\omega>0)$ sup . Lastly, the drag in Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems) can be viewed as a rectification of the quantum charge noise in the drive system, which discerns it from the shot-noise driven drag of Ref. Levchenko and Kamenev (2008) mentioned in the introduction.

Weak-interaction regime.—Having established the general properties of the drag, we next examine its behavior for weak interactions ($U\lesssim\Gamma$).

In the small-$U$ limit, $U\ll\Gamma$, an explicit expansion of the drag current to second order in $U$ applies, and amounts to replacing $G_{j}\rightarrow G_{j}^{0}$, where $G_{j}^{0}$ is the noninteracting dot GF, in Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems). In this limit, our DQD system behaves similarly as two Coulomb-coupled single-channel QPCs with transmission coefficients $T_{j}=4\Gamma_{jL}\Gamma_{jR}/\Gamma_{j}^{2}$, and the charge noise scales as $P_{2}\sim T_{2}$ for $k_{B}T\ll eV\ll\Gamma$ sup . This should be contrasted with the qualitatively different $S_{2}\sim T_{2}(1-T_{2})$ scaling of the shot noise Blanter and Büttiker (2000), thus providing a means to distinguish between drag due to, respectively, direct Coulomb interactions and a circuit environment Levchenko and Kamenev (2008) by tuning the conductance $\mathcal{G}_{2}=\tfrac{e^{2}}{h}T_{2}$ of the drive system.

Next, we turn to numerical calculations of the drag using parameters mimicking the experiment in Ref. Keller et al. (2016). In order to fulfil the conditions for a nonzero drag, we model for convenience the lead couplings $\Gamma_{\alpha j}$ by semielliptic bands with bandwidth $D_{\alpha j}$ and a relative shift between the bands in the left and right contacts of the drag system as illustrated in Fig. 3(c) sup .

Figure. 3(a) and 3(b) show the calculated drag and drive currents for the level positions marked with the colored dots in Fig. 2(a). At low bias, the level of the drive dot is off resonance with respect to the chemical potentials and both the drive and drag currents are small. With increasing bias voltage, the level of the drive dot enters the conduction window and the onset of the drive current induces a current in the drag system. While the direction of the drive current follows the sign of the applied bias voltage, the direction of the drag is governed by the lead couplings. From Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems), this can be traced back to the fact that $I_{\mathrm{drag}}$ has no linear dependence on $V$ when the drive contacts have proportional lead couplings, $\Gamma_{L2}\propto\Gamma_{R2}$ sup . As demonstrated in Fig. 3(d), the drag current at low bias thus increases as $I_{\mathrm{drag}}\sim V^{2}$, emphasizing the inherent nonlinear nature of Coulomb drag in quantum dot systems Kaasbjerg and Jauho (2016).

The dependence of the currents on the QD levels at a fixed bias voltage of $V=0.5\,\mathrm{mV}$ is shown in Fig. 4. Here, the drive current in Fig. 4(a) shows a weak dependence on the drag level, $\varepsilon_{1}$, due to the interaction-induced shift of the drive level upon changing the occupation of the drag QD. The drag current in Fig. 4(b) is only significant near the honeycomb vertex where the alignment between the QD levels and the chemical potentials allows for simultaneous interaction-induced electron-hole pair processes in the drive and drag system [cf. Fig. 1(b)]. Our results in Figs. 2(a) and 4 are in good qualitative agreement with the experimental measurements reported in Ref. Keller et al. (2016) [their Figs. 2(b)-(c) and 2(e)-(f)].

Strong-interaction and Kondo regime.—In the regime of strong interdot interaction, $U\gg\Gamma$, the DQD Hamiltonian can be mapped onto the Kondo Hamiltonian $\mathcal{H}_{\rm K}=\mathcal{H}_{\rm dir}+\mathcal{H}_{\rm ex}$ by a Schrieffer-Wolff transformation Schrieffer and Wolff (1966); sup with the two QD states acting as a pseudospin ($\hat{S}$) exhibiting an orbital Kondo effect Borda et al. (2003); Jarillo-Herrero et al. (2005); Amasha et al. (2013) at $T<T_{K}$ where $T_{K}\sim\sqrt{\Gamma U}\exp(-\pi U/4\Gamma)$ is the Kondo temperature Kaminski et al. (2000). Here, $\mathcal{H}_{\rm dir}=\sum_{\alpha\beta jkq}\mathcal{K}_{\alpha j,\beta j}c^{\dagger}_{\alpha jk}c_{\beta jq}^{\phantom{\dagger}}$ is the potential scattering term responsible for elastic pseudospin-conserving cotunneling transitions, whereas the exchange interaction $\mathcal{H}_{\rm ex}=\sum_{\alpha\beta ijkq}\mathcal{J}_{\alpha i,\beta j}\hat{S}_{l}s^{l}_{ij}c^{\dagger}_{\alpha ik}c_{\beta jq}^{\phantom{\dagger}}$ accounts for, e.g., inelastic pseudospin-flip cotunneling processes, and $\mathcal{K}_{\alpha i,\beta j}$ and $\mathcal{J}_{\alpha i,\beta j}$ are the effective couplings Hewson (1993); sup .

The importance of pseudospin (i.e., charge) fluctuations inherent to the orbital Kondo effect is evident already from the leading-order $I_{\text{drag}}\propto\lvert\mathcal{J}\rvert^{2}$ contribution to the drag. Performing a perturbative calculation of the drag at the particle-hole symmetric point ($\varepsilon_{1}=\varepsilon_{2}=-U/2$) Kaminski et al. (2000), we find sup ,

[TABLE]

where $\Delta\Gamma_{j}=\Gamma_{Lj}-\Gamma_{Rj}$. This can be viewed as drag due to nonlocal cotunneling processes where the pseudospin of the DQD is flipped in one coherent processes via virtual intermediate empty and filled states Kaasbjerg and Jauho (2016); Lim et al. (2018).

To get an indication of the corrections to the drag in Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems) due to Kondo correlations, we show in Fig. 5 numerical results [based on Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems)] for the drag at different temperatures at the particle-hole symmetric point. The spectral function $\mathcal{A}_{1}(\omega)$ of the drag dot in Fig. 5(a) clearly shows a low-temperature feature at the Fermi energy which resembles a Kondo peak Horvatić et al. (1987); Sun and Guo (2002); Bao et al. (2014). Overall, the corresponding drag currents in Fig. 5(b) only show marginal changes relative to the perturbative result in Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems) (dashed line). However, at low temperature and low bias, Kondo correlations quench the charge fluctuations driving the drag which results in a suppression relative to the perturbative result. Due to the low-bias $I_{\mathrm{drag}}\sim V^{2}$ behavior of the drag, the effect of Kondo correlations is more noticeable in the second-order differential transconductance $\mathcal{G}_{\rm tr}^{(2)}=d^{2}I_{\rm drag}/dV^{2}$ shown in Fig. 5(c), which features a pronounced zero-bias anomaly in $\mathcal{G}_{\rm tr}^{(2)}$ at low $T$. Finally, Fig. 5(d) shows the dependence of the drag on the strength of the interdot interaction at different bias voltages. For large values of $U$, the numerical results follow the $I_{\rm drag}\sim 1/U^{2}$ behavior of the perturbative result of Eq. (Fluctuation-driven Coulomb drag in interacting quantum dot systems) (dashed lines), whereas the $I_{\rm drag}\sim U^{2}$ behavior of the leading-order contribution to the drag in Eq. (3) is observed for small values of $U$ (dotted line). The two regimes are bridged by an intermediate region, $\Gamma\sim U$, with an optimal value of $U$ where $|I_{\rm drag}|$ is maximized.

Conclusions.—We have studied Coulomb drag across interaction regimes in Coulomb-coupled QD systems in the framework of the Keldysh NEGF technique. In agreement with previous works Sánchez et al. (2010); Kaasbjerg and Jauho (2016); Keller et al. (2016); Lim et al. (2018), we find that drag is an inherently nonlinear effect and that energy-dependent lead couplings are instrumental for the generation of a drag current. As we demonstrate, the drag is driven by the nonequilibrium charge fluctuations of the drive QD, and we discuss how the characteristics of the quantum noise allows to differentiate the drag mechanism discussed here from drag induced by a circuit environment Levchenko and Kamenev (2008) experimentally. In the case of strong interdot interactions, the charge fluctuations are quenched by orbital Kondo correlations at low temperature and bias, which suppresses the drag current with respect to the lowest-order cotunneling-only drag Kaasbjerg and Jauho (2016). In addition, we predict a clear signature of Kondo correlations in the second-order differential transconductance of the drag. Overall, our findings show that the Coulomb drag between quantum conductors is highly dependent on the interaction mediating the drag, and open the opportunity for further experimental studies of drag in QD systems.