Back-action effects in charge detection Sarath Sankar, Matan Lotem, Joshua Folk, Eran Sela, Yigal Meir Abstract Charge detection offers a powerful probe of mesoscopic structures based on quantum dots, but it also invariably results in measurement back-action (MBA). If strong, MBA can be detrimental to the physical properties being probed. In this work, we focus on the effects of MBA on an Anderson impurity model in which the impurity is coupled electrostatically to a detector. Introducing a novel non-perturbative method, we explore the interplay of coherent dynamics, strong correlations and non-equilibrium conditions. The effects of MBA can be seen most clearly in the temperature derivative of occupation. In the non-equilibrium case, we identify this as arising due to an energy flow from the detector to the impurity. Introduction. Charge measurement is a valuable experimental probe in mesoscopic systems. The standard technique [ 1 ] involves a nearby detector, usually a quantum point contact or a quantum dot, the current through which is sensitive to the charge of the system due to electrostatic interactions. Such charge detectors have been widely used to study a variety of physical phenomena such as Coulomb blockade [ 1 ] , dephasing in electron interference [ 2 ] , double-quantum-dot qubits [ 3 , 4 ] , full counting statistics [ 5 ] , charge Kondo screening [ 6 ] and non-equilibrium thermodynamics [ 7 , 8 , 9 , 10 ] . A recent addition to the above list is the measurement of thermodynamic entropy using a Maxwell relation (MR) [ 11 , 12 , 13 ] , which has several promising applications [ 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 ] . While such charge measurements may often be considered as “noninvasive” [ 1 ] , it has now been established that there are circumstances where the detector may have significant effects on the mesoscopic system via measurement back-action (MBA) [ 2 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 ] . The charge detector continuously measures the system and tends to suppress its unitary dynamics [ 36 , 37 , 38 , 34 ] , as in the quantum Zeno effect [ 39 , 40 , 41 ] . In addition, a voltage-biased detector can also serve as an energy bath, leading to detector-assisted dynamics [ 29 , 30 , 31 , 32 , 33 ] . The presence of the detector leads to a competition between strong correlations developing in the system itself (the Kondo effect in the example treated below), and the correlations and entanglement between the system and the detector, manifested, e.g., in the Anderson overlap catastrophe [ 42 , 24 ] . Here we look in detail at a simple example, a quantum dot (QD) coupled to a lead, and study how MBA due to the charge detector modifies its thermodynamic observables. We focus on the average charge of the QD, N , and its temperature derivative, dN/dT . The latter has escaped attention in previous MBA studies but encodes important thermodynamic information of the system. We show that dN/dT is particularly sensitive to MBA. This makes it a robust marker for experimental investigation of back-action effects. This problem serves as a paradigmatic example of the rich interplay between strong correlations, coherent dynamics, and out-of-equilibrium physics, which cannot at present be treated exactly using analytical or numerical methods. Our treatment of the MBA is non-perturbative in both the QD-lead tunneling and the system-detector interaction. The latter is necessary to capture the many-body effect of the Anderson orthogonality catastrophe. In contrast, previous MBA treatments are perturbative in the system-detector interaction. This includes the phenomenological P(E) theory [ 43 , 44 , 45 ] , which describes how fluctuations in the electric field due to the current noise in the detector [ 46 , 47 ] lead to a dephasing effect in the measured QD. Figure 1: (a) Schematic of an Anderson impurity model electrostatically coupled to a quantum dot detector (QDD). (b) For \Gamma\to 0 , we calculate the exact QD propagator including the interaction with the QDD (wiggly line), and then use it to carry out an hybridization expansion in \Gamma by the D-NCA method discussed in the main text. Model. We consider a QD emulating an Anderson impurity, capacitively coupled to a detector as shown in Fig. 1 (a). For concreteness, we take the detector to be a quantum-dot detector (QDD); with an eye to future experimental tests, this choice has the advantage that the strength of the MBA due to a QDD can readily be tuned by changing the QDD energy level and its coupling to the detector leads [ 35 ] . However, the treatment in this paper will be formulated in terms of universal MBA parameters that are not specific to the type of detector. The Hamiltonian is made up of parts representing the system, the detector, and the interaction between them, H=H_{\rm{sys}}+H_{\rm{det}}+H_{\rm{int}} . The system is described by \begin{split}H_{\rm{sys}}&=\sum_{\sigma}\epsilon_{d}d_{\sigma}^{\dagger}d_{\sigma}+Ud_{\uparrow}^{\dagger}d_{\uparrow}d_{\downarrow}^{\dagger}d_{\downarrow}\\ &+\sum_{k\sigma}\epsilon_{k}c^{\dagger}_{k\sigma}c_{k\sigma}+\sum_{k\sigma}\left(\gamma_{k}c^{\dagger}_{k\sigma}d_{\sigma}+{\rm{H.c.}}\right).\end{split} (1) Here, d^{\dagger}_{\sigma} ( c^{\dagger}_{k\sigma} ) creates an electron in the single-level QD (lead) with spin \sigma and energy \epsilon_{d} ( \epsilon_{k} ). For later convenience, the density of states (DoS) of the lead is taken to be a Lorentzian of width D , so the line width (hybridization) of the QD is \Delta(\omega)=2\pi\sum_{k}|\gamma_{k}|^{2}\delta(\omega-\epsilon_{k})=\Gamma\frac{D^{2}}{\omega^{2}+D^{2}} . We assume \epsilon_{d},\Gamma,T\ll D so that band-edge effects are negligible. The average charge in the QD is N=\sum_{\sigma}\langle d^{\dagger}_{\sigma}d_{\sigma}\rangle . The onsite interaction U is taken to be infinite such that double occupancy is prohibited. The detector consists of a quantum dot tunnel-coupled to two leads at bias voltage \pm V/2 , and described by H_{\rm{det}}=\sum_{i=1}^{m}\epsilon_{q}q^{\dagger}_{i}q_{i}+\sum_{ik\mu}\left[\epsilon_{k}\varphi_{ki\mu}^{\dagger}\varphi_{ki\mu}+v_{\mu}(\varphi_{ki\mu}^{\dagger}q_{i}+{\rm H.c.})\right]. (2) Here, q_{i}^{\dagger} ( \varphi_{ki\mu}^{\dagger} ) creates an electron in the detector dot (lead \mu=L,R ) with energy \epsilon_{q} ( \epsilon_{k} ) and i denotes additional conserved degrees of freedom, which can include spin and other channels; for simplicity, we treat it as a degeneracy factor with m flavors. The width of the detector QD level due to tunneling is \Gamma_{q}=2\pi\nu_{q}(v_{L}^{2}+v_{R}^{2}) where v_{\mu} represents the tunneling strength and \nu_{q} is the DoS of the detector leads. The interaction strength between the system and detector is parameterized by \lambda , and given by H_{\rm{int}}=\lambda\sum_{i=1}^{m}\left(q^{\dagger}_{i}q_{i}-\frac{1}{2}\right)\left(\sum_{\sigma}d^{\dagger}_{\sigma}d_{\sigma}-\frac{1}{2}\right). (3) The QDD model ignores intra-detector-dot interactions to elucidate the MBA effects in a simple and universal manner. A treatment of MBA effects due to an interacting detector will be considered in a future work. Key results. Our main result is a condition for the regime in which MBA on thermodynamic quantities can be neglected, and the identification of MBA effects that occur away from this regime. The strength of MBA is parameterized by two quantities: • \alpha : a dimensionless exponent that characterizes the strength of the Anderson orthogonality catastrophe. It is given by \alpha=m(\delta/\pi)^{2} , in terms of the change in the detector scattering phase shift, \delta=\arctan\left(\frac{\epsilon_{q}+\lambda/2}{\Gamma_{q}}\right)-\arctan\left(\frac{\epsilon_{q}-\lambda/2}{\Gamma_{q}}\right) , due to an electron tunneling in/out of the QD [ 35 ] . • \Gamma_{\varphi} : a dephasing rate that captures the inelastic effects of MBA due to voltage bias in the detector. It was introduced previously [ 24 , 26 ] in the context of dephasing of Aharonov-Bohm oscillations and is reproduced in Eq. ( 6 ). As we demonstrate below, MBA on thermodynamic observables is negligible provided that \alpha\ll 1 and \Gamma_{\varphi}\ll T . At equilibrium (V=0) , MBA is governed only by \alpha , which reduces the width of the charging curve and modifies the lineshape of dN/dT . Most of our calculations apply to the non-equilibrium case, (V\neq 0) , for which we develop a new non-perturbative approach to solve the Anderson model coupled to a detector, capturing large electrostatic coupling to the detector (resulting in a large \alpha) and also hybridization (\Gamma) to the lead. We apply this method to show that the charging curve broadens due to \Gamma_{\varphi} , though this effect is small when \Gamma_{\varphi}\ll\mathrm{max}(\Gamma,T) . Then, we confirm that the Maxwell relation based on dN/dT may be applied only when \Gamma_{\varphi}\ll T . Finally, we use the techniques developed in this paper to study the heat current that flows from a biased detector to the system, uncovering a strong but unexpected dependence of the heat current on occupation. Equilibrium MBA effects. Figure 2: Back action effects due to an unbiased detector. (a) At large \Gamma/T , the broad peak of the spectral function of the QD level shifts and narrows with increasing \alpha . (b) For any given \Gamma , the width of the charge transition narrows with increasing \alpha , similar to the effect of reducing \Gamma . (c) The peak of dN/dT as a function of N shifts left with increasing \alpha , again similar to the effect of reducing \Gamma . But the dN/dT peak due to increased \alpha shows an additional broadening compared to that due to a reduced \Gamma . The different dN/dT curves here are rescaled to same height for easy comparison. (d) The effect of \alpha on entropy change obtained using the Maxwell relation, for the same set of parameters as in (b). The curves in (b) and (d) are shifted such that \epsilon_{d}=0 corresponds to N=1/2 . Figure 2 highlights MBA effects at equilibrium. Numerical renormalization group methods are applied to calculate how two quantities describing the Anderson impurity depend on \alpha : the impurity spectral function, \nu(\omega) , and its average occupation, N=\int_{-\infty}^{\infty}d\omega f(\omega)\nu(\omega) , where f(\omega) is the Fermi function. At weak tunneling, \Gamma/T\ll 1 , N is not affected by \alpha [Fig. 2 (b)], even though the spectral function itself broadens with \alpha (not shown) [ 48 ] . At strong tunneling, \Gamma/T\gg 1 , \nu(\omega) has a broad peak centered near \omega=\epsilon_{d} , and a Kondo peak at \omega=0 [Fig. 2 (a)]. The width of the broad peak is \Gamma when \alpha=0 ; it renormalizes downwards as \alpha grows [ 34 , 49 ] , signaling an effective reduction of the tunnel strength, which consequently reduces the width of the charge transition, N(\epsilon_{d}) , see Fig. 2 (b). It may be difficult to identify the effect of \alpha on N in an experiment, since \Gamma is typically extracted from the charge transition itself. For example, the \{\Gamma/T=14,\alpha=0\} curve in Fig. 2 (b) (solid blue) is nearly indistinguishable from the \{20,0.5\} curve (dotted red), which has higher bare \Gamma/T but effective tunnel strength reduced by MBA. As seen in Fig. 2 (c), however, these charge transitions can be more easily distinguished by looking at the temperature-induced modifications to the charge transition, dN/dT . Although they peak at nearly the same value of N , the line shape of dN/dT(N) for \{20,0.5\} is significantly wider than for \{14,0\} , by an amount that should be straightforward to distinguish in an experiment. A corollary of the different line shapes of dN/dT(N) is that the entropy change \Delta S obtained by integration of the Maxwell relation, \Delta S(\epsilon_{d})=\int_{\epsilon_{d}}^{\infty}\frac{\partial N}{\partial T}|_{\epsilon_{d}^{\prime}}d\epsilon_{d}^{\prime}, (4) is slightly different for \{20,0.5\} than for \{14,0\} . We note that the determination of entropy change via the MR is still valid for large \alpha because the system is in thermal equilibrium. In fact, MR measures the entropy change of the combined system plus detector. But for the regime considered here, \Gamma_{q}\gg T , the entropy change of the detector itself is negligible. Non-equilibrium MBA effects. We now discuss MBA effects due to a voltage-biased detector. When the QD is decoupled from its lead, \Gamma\to 0 , the dynamics of the many-body state of the detector is governed by the Hamiltonian H_{n}=H_{\rm{det}}+H_{\rm{int}} . Here n=\sum_{\sigma}d^{\dagger}_{\sigma}d_{\sigma} , the total charge of the system QD, is conserved and is either 0 or 1 . All the MBA information is encapsulated in the functions A^{+/-}(t)={\rm{Tr}}\left[\rho_{0/1}e^{itH_{0}}e^{-itH_{1}}\right], (5) where \rho_{n}\propto e^{-H_{n}/T} denotes the non-equilibrium detector density matrix in which eV is incorporated as a difference of the chemical potential of electrons incoming from source and drain. The resulting functions, A^{\pm}(t) , describe the effect of a sudden change of the detector Hamiltonian that occurs due to a tunnel event into [A^{+}(t)] or out of [A^{-}(t)] the QD. These functions can be computed exactly at finite V because both H_{n=0} and H_{n=1} are non-interacting. At long times they decay in an analytically tractable way, as [ 50 , 24 ] A^{\pm}(t)\approx\left(\frac{\pi T}{i\xi\sinh(\pi Tt)}\right)^{\alpha}e^{-\Gamma_{\varphi}t}, (6) where \xi=\mathcal{O}(\Gamma_{q}) serves as a high-energy cutoff. While \alpha describes a power-law decay exponent of these functions, the detector bias induces a dephasing rate, \Gamma_{\varphi} . We emphasize that the pair of parameters \alpha and \Gamma_{\varphi} in Eq. ( 6 ) gives a universal characterization of MBA in general detectors. Figure 3: Back action effects due to a biased detector for weak tunneling, \Gamma\ll T , at the Anderson impurity. (a) Dependence of the dephasing rate, \Gamma_{\varphi} , on detector bias. Here the line width of the QDD is \Gamma_{q}/T=30 . For small \alpha , the weak MBA regime of \Gamma_{\varphi}\ll T is satisified even for V\gg T . The case \alpha=0.3 is considered in panels (b)-(d). (b) Broadening of the charge transition with detector bias. The curves are shifted such that \epsilon_{d}=0 corresponds to N=1/2 . The black dotted line shows that the bias broadened curve for even a large dephasing rate ( \Gamma_{\varphi}/T\approx 7 ) fits nicely to a Fermi function, f(\epsilon_{d},T_{eff}) , with a larger effective temperature ( T_{eff}/T\approx 4 ). (c) Suppression of dN/dT due to increasing detector bias. (d) The dependence of the energy current from the detector to the system QD, on detector bias and average charge of the QD. Figure 3 (a) illustrates the dependence of \Gamma_{\varphi} on V and the other parameters of the QD detector in our model, using the numerically exact methods for computing A^{\pm}(t) and its Fourier transforms A^{\pm}(\omega) reported in our previous work [ 35 ] . We find that \Gamma_{\varphi}\propto V for T\lesssim V\lesssim\Gamma_{q} and it saturates for V>\Gamma_{q} . \Gamma_{\varphi} itself increases with \alpha and vanishes when \alpha\to 0 . Importantly, for small \alpha , the weak MBA regime of \Gamma_{\varphi}\ll T is satisfied even for V\gg T . We now discuss the distinct implications of the dephasing rate on N and dN/dT in the weak tunneling case by treating finite \Gamma within a rate equations approach [ 35 , 48 ] . The width of the charge transition grows with V as soon as \Gamma_{\varphi} is larger than T [Fig. 3 (b)]. In this way, the presence of strong dephasing, \Gamma_{\varphi}\gtrsim T , can be easily detected in experiments, in the form of bias-dependent broadening of the charge transition. Because charge transitions broadened primarily by \Gamma_{\varphi} are consequently less affected by T , dN/dT is suppressed at high detector biases [Fig. 3 (c)]. This signals that the system is out of thermal equilibrium, and the application of thermodynamic relations is made unreliable by MBA. Interestingly, even when \Gamma_{\varphi}\gtrsim T , the broadened charge transition curves fit very well to thermal broadening lineshapes, but with a larger effective temperature T_{\rm{eff}}>T , as shown in Fig. 3 (b). As a result, a naive application of the MR can yield an entropy change that is approximately correct, albeit with a rescaled temperature T_{\rm{eff}}(T) of the QD lead. Detector non-crossing approximation (D-NCA). To treat non-equilibrium MBA effects at a stronger tunnel coupling, we develop an approximate non-perturbative method in which a hybridization expansion is performed in the slave-boson framework resorting to the non-crossing approximation (NCA), with the MBA information entering as modified bare (i.e. \Gamma\to 0 ) QD propagators, see Fig. 1 (b). We now briefly describe this method. A more detailed description is provided in [ 48 ] . In the slave-boson formalism, a different Hilbert space is considered for the system QD, including a (non-physical) vacuum |\Omega\rangle , an empty state created by the slave-boson operator b^{\dagger}|\Omega\rangle=|0\rangle , and the singly occupied states created by the pseudo-fermion operators f_{\sigma}^{\dagger}|\Omega\rangle=|\sigma\rangle . The original fermion creation operators are expressed in terms of these new operators as d_{\sigma}^{\dagger}=f_{\sigma}^{\dagger}b . The expression of H_{\rm{sys}} becomes H_{\rm{sys}}^{\rm SB}=\sum_{\sigma}\epsilon_{d}f^{\dagger}_{\sigma}f_{\sigma}+\sum_{k\sigma}\epsilon_{k}\hat{n}_{k\sigma}+\sum_{k\sigma}\gamma_{k}\left(c^{\dagger}_{k\sigma}b^{\dagger}f_{\sigma}+{\rm H.c.}\right). (7) For \lambda=0 , the NCA is known to be highly accurate, even well below the Kondo temperature [ 51 , 52 ] . Under NCA, tunneling is treated using a simple Dyson-like diagrammatic expansion in \Gamma [ 53 ] . Our detector-NCA (D-NCA) approach generalizes the NCA treatment to \lambda\neq 0 . A key difference is that the bare retarded and lesser pseudo-fermion propagators are modified due to the interaction with the detector using the A^{\pm} functions, G_{f\sigma,0}^{r}(\omega)=-iA^{+,r}(\omega-\epsilon_{d}),\,G_{f\sigma,0}^{<}(\omega)=iA^{-}(\omega-\epsilon_{d}), (8) where A^{+,r}(\omega) is the Fourier transform of \theta(t)A^{+}(t) , with \theta(t) denoting the Heaviside step function. Figure 4: Back action effects ( \alpha=0.3 ) due to a biased detector for the case of strong tunneling, \Gamma/T=30 . (a) The suppression of the Kondo peak at \epsilon_{d}=-2\Gamma with detector bias. (b) The bias has a very weak effect on the impurity charge for values of V/T shown, each of which satisfy \Gamma_{\varphi}\ll\Gamma . The curves here are shifted such that \epsilon_{d}=0 corresponds to N=1/2 . (c) dN/dT gets suppressed with increasing bias, but only for N\gtrsim 1/2 . (d) The energy current that flows from the detector to the system is skewed with the peak occurring at N close to 1 . The additional approximation in D-NCA is neglecting higher order mixed irreducible self-energies with both interaction and tunneling vertices, similar in spirit to NCA. The D-NCA approximation scheme becomes close to exact in two regimes: (i) \Gamma\ll T together with arbitrary \alpha , and (ii) \alpha\ll 1 together with arbitrary \Gamma . The D-NCA is expected to be a decent approximation for the intermediate regimes that are discussed in the paper, but might fail in the extreme simultaneous regime of both \alpha\gg 1 and \Gamma\gg T . MBA in the Kondo regime. We now proceed to discuss the results obtained using D-NCA in the strong tunneling regime, see Fig. 4 . As in Fig. 2 (a), a Kondo peak appears in \nu(\omega) for \Gamma\gg T . The peak is suppressed with increasing detector bias, similar to the findings of Ref. [ 54 ] , and as observed experimentally in Ref. [ 27 ] . The width of the charge transition is only weakly affected by the bias voltage because it is dominated by tunneling, \Gamma\gg\Gamma_{\varphi} [Fig. 4 (b)]. Of particular interest is the behavior of dN/dT in Fig. 4 (c). At V=0 the peak in dN/dT occurs at N>0.5 , consistent with the rightward shift of the peak in Fig. 2 (c). Upon increasing bias, dN/dT is suppressed, but in an N- dependent manner so the lineshape is changed. This contrasts with the weak tunneling case in Fig. 3 (c), where dN/dT versus N is rescaled by a constant factor but does not change shape. The essential conclusion from this analysis, including both the weak and strong tunneling cases, is that the necessary and sufficient measurement condition to minimize MBA for thermodynamic measurements is to ensure that \Gamma_{\varphi}\ll T . In the case of weak coupling, it is possible to make meaningful measurements even when \Gamma_{\varphi}\gtrsim T through the use of an effective temperature, but in the case of strong coupling, even that is not possible. Energy flow. To gain further insight into the difference in the nature of the suppression of dN/dT between weak and strong tunneling regimes, we calculate the energy flow from the non-equilibrium detector to the impurity QD. As a proxy to this quantity, we note that away from equilibrium the impurity distribution function differs from the Fermi function, and the difference between these is related to heat flow. At equilibrium, the lesser Green’s function satisfies G^{<}(\omega)=2\pi i\nu(\omega)f(\omega) . We define the non-equilibrium distribution function of the QD—which determines how the available states are occupied—as h(\omega)\equiv iG^{<}(\omega)/(2\pi\nu(\omega)) . The energy current from the QD to the lead can be expressed as [ 48 ] J_{\rm{en}}=2\Gamma\int_{-\infty}^{\infty}d\omega\,\omega\,\nu(\omega)(h(\omega)-f(\omega)). (9) The factor of 2 accounts for spin. In the steady state, the energy flow from the detector to the system’s QD equals the energy flow from the system’s QD to its lead, so Eq. ( 9 ) also represents the heat flow between detector and QD. We present results for heat flow based on D-NCA in Fig. 3 (d) and Fig. 4 (d), representing weak and strong tunneling respectively. The quantity \frac{J_{\rm{en}}}{\Gamma T} that is plotted represents how far the system is from thermal equilibrium. As long as \Gamma_{\varphi}\ll T , the system stays close to thermal equilibrium and \frac{J_{\rm{en}}}{\Gamma T}\ll 1 for all N. When \Gamma_{\varphi}\gtrsim T , the energy flow as a function of the average QD charge is qualitatively different for weak and strong tunnel couplings [Fig. 3 (d) compared to Fig. 4 (d)]. For weak tunneling, it peaks at N\sim 1/2 as might naively be expected given that charge fluctuations in a weakly-coupled QD would also be peaked at N=1/2 . In contrast, for strong tunneling, heat flow is large only close to N=1 . While we currently do not have a simple picture for the anomalous N dependence of the energy flow, we note that this dependence is similar to the corresponding N -dependent suppression of dN/dT , which is also peaked at N close to 1. At an intuitive level, this may be understood from the fact that both quantities directly relate to the extent of deviation of the system from thermal equilibrium. An experiment probing N -dependent heat flow directly in the Anderson impurity-detector setup would be especially insightful in developing a better understanding of the mechanism. Acknowledgements. We gratefully acknowledge support from the European Research Council (ERC) under the European Union Horizon 2020 research and innovation programme under grant agreement No. 951541. ARO (W911NF-20-1-0013). Y.M. acknowledges support from ISF Grants Nos. 359/20 and 737/24. J.F. acknowledges support from the Natural Sciences and Engineering Research Council of Canada; the Canadian Institute for Advanced Research; the Max Planck-UBC-UTokyo Centre for Quantum Materials and the Canada First Research Excellence Fund, Quantum Materials and Future Technologies Program. We thank Zhanyu Ma and Cheolhee Han for illuminating discussions. Numerical renormalization group calculations where carried out with the QSpace tensor library [ 55 , 56 , 57 ] .