Autonomous Quantum Heat Engine

• Autonomous quantum heat engines are self-contained systems that operate without external time-dependent control, converting fixed thermal gradients into usable work.
• They encompass minimal models like the three-level maser and advanced architectures including superconducting circuits, quantum rotors, and optomechanical devices.
• Key analyses focus on energy conversion metrics, efficiency limits, and noise impacts, clarifying the trade-offs between coherence, dissipation, and internal control.

An autonomous quantum heat engine is a self-contained quantum thermal machine whose operation is not prescribed by an externally driven, time-dependent protocol. In open-system formulations, fixed Hamiltonians and steady couplings to thermal reservoirs generate continuous heat currents that are internally converted into directed rotation, electrical current, cavity fields, or other work-bearing outputs; in isolated formulations, an explicit quantum clock implements the control through a time-independent global Hamiltonian (Malabarba et al., 2014, Roulet et al., 2016). The literature therefore uses the term for a family of engines that includes steady-state masers, rotor and piston analogues, superconducting-circuit devices, optomechanical machines, and architectures in which internal electronic, mechanical, flux, or molecular degrees of freedom switch bath couplings without external timing (Tonekaboni et al., 2018, Uusnäkki et al., 16 Mar 2026).

1. Definitions and conceptual scope

The defining feature of autonomy is the absence of externally imposed, time-dependent control. In the clock-driven framework, the total Hamiltonian is time independent, an explicit quantum clock generates effective control on the engine, and any energy-conserving unitary on the engine can be implemented exactly without degrading the clock; with a quantum work storage device, arbitrary unitaries on the remaining subsystem can be approximated with controllable accuracy, and the framework attains work extraction and accuracy arbitrarily close to optimal limits with no intrinsic thermodynamic cost compared to externally controlled machines (Malabarba et al., 2014). In the open-system setting, the same term denotes engines whose dynamics are sustained by fixed couplings to hot and cold reservoirs and by internal interactions rather than by externally timed strokes (Verteletsky et al., 2019).

Within this broad class, the minimal continuous autonomous engine is the three-level maser. A three-level system is identified as the smallest quantum system capable of autonomous cycling, with one transition coupled to a hot bath, one to a cold bath, and one to a work channel (Deng et al., 2024). More elaborate machines enlarge the working medium or the controller: quantum rotors encode piston motion in angular variables, electron shuttles use electronic occupation as a switching coordinate, and superconducting circuits use SQUID- or Josephson-mediated frequency conversion to couple thermal bias directly to coherent electrical output (Roulet et al., 2016, Tonekaboni et al., 2018, Uusnäkki et al., 16 Mar 2026).

A boundary case is the single-temperature engine powered by quantum measurement. Its cycle is autonomous only in the restricted sense that no feedback control is used and the protocol is preprogrammed; the energy input is provided by non-selective measurement back-action rather than by heat absorbed from a hot reservoir (Yi et al., 2017). This distinguishes “autonomous” from “two-bath heat-driven” and shows that the literature attaches the term primarily to the absence of feedback or externally timed control, not to a single microscopic implementation.

2. Thermodynamic description and notions of work

Autonomous engines sharpen the distinction between energy, heat, and work because the global Hamiltonian is typically time independent while energy still accumulates in a piston, flywheel, resonator, or current-carrying degree of freedom. For a single bosonic mode with average photon number $n$ and energy quantum $\mathcal{E}$, one experimentally realized superconducting Otto-like engine writes the first law as

$$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$

with $\delta W$ associated with changes in the mode energy quantum and $\delta Q$ with changes in occupation (Uusnäkki et al., 16 Mar 2026).

In the autonomous superconducting-qubit engine based on two qubits and a voltage-biased Josephson junction, work and heat are explicitly treated as process quantities rather than observables. Power is defined from the steady-state current across the junction,

$P=\langle \hat I\rangle_{\mathrm{ss}} V,$

while the heat current between qubit $\alpha$ and its bath is

$$J_\alpha=\operatorname{Tr}\!\left[(\hbar\omega_\alpha \sigma_\alpha^+\sigma_\alpha^-)\,\mathcal L_\alpha\rho\right].$$

In steady state this gives the balance relation

$P=J_h-J_c,\qquad \eta=\frac{P}{J_h},$

so the thermodynamic output is reconstructed from steady-state observables and dissipators rather than from a Hermitian “work operator” of the engine itself (Verteletsky et al., 2019).

This problem becomes more acute when the output degree of freedom is itself quantum. In the three-level maser analysis of work concepts, the piston’s entropy change cannot be neglected, so the energy accumulated in the piston is decomposed into ergotropy, bound ergotropy, and thermal energy. The single-copy ergotropy is

$$\mathcal{E}(\rho,H)=\operatorname{Tr}(\rho H)-\min_U \operatorname{Tr}(U\rho U^\dagger H),$$

while non-equilibrium free energy quantifies work accessible through thermal processes at a specified bath temperature (Niedenzu et al., 2019). This yields several inequivalent efficiencies: an energetic efficiency that counts total piston energy growth, an ergotropic efficiency that counts only immediately usable work, and free-energy efficiencies that remain Carnot bounded.

A further refinement appears when the controller is internal and dynamical. In the electron-shuttle engine, the first law for the mechanical energy is

$\mathcal{E}$0

where $\mathcal{E}$1 is an explicit controller-associated energy flux arising from the position-dependent tunneling dissipators that switch the bath couplings (Tonekaboni et al., 2018). This bookkeeping shows that internal control is not thermodynamically free merely because it is autonomous.

3. Internal control, timing, and cycle generation

Autonomous engines require an internal mechanism that replaces the external clock used in stroke engines. One route is literal clock embedding: a quantum clock with Hamiltonian $\mathcal{E}$2 traverses an interaction region and implements the intended unitary on the engine while remaining undegraded apart from translation (Malabarba et al., 2014). In this construction, the clock does not act as a thermal resource; it provides timing inside a time-independent Hamiltonian.

More common in heat-engine models is dynamical self-timing through state-dependent couplings. In the electron-shuttle engine, the occupation of a single electronic level acts as a controller: when the nano-island carries an electron, the mechanical mode couples to hot Johnson noise, and when it is empty that coupling is suppressed. Position-dependent tunneling to source and drain, together with the half-harmonic potential, converts this switching into sustained oscillation via autonomous stochastic resonance (Tonekaboni et al., 2018). In the cQED single-piston engine with a quantum rotor, the resonator frequency depends on the rotor angle, and a filter cavity acts as a valve that opens the hot inlet only when the working mode is near resonance, producing a self-timed Carnot-like cycle (Roulet et al., 2018).

Several proposals realize an approximate Otto cycle through internal frequency modulation. In the non-Markovian optomechanical engine, a mechanical driving mode modulates the frequency of an optical working mode so that it sweeps across a hot reservoir centered at $\mathcal{E}$3 and a cold reservoir centered at $\mathcal{E}$4; provided the mechanical mode begins in a coherent state with sufficiently large amplitude, the cycle is generated without external drive (Rasola et al., 2024). The superconducting-circuit proposal and its later experimental realization use an analogous idea: a low-frequency resonator modulates the working-body frequency through a SQUID-mediated optomechanical-like coupling, while hot and cold filter resonators provide spectrally selective thermal contacts, producing a “sinusoidally modulated Otto” cycle rather than ideal separated strokes (Rasola et al., 12 Feb 2025, Uusnäkki et al., 16 Mar 2026).

At the molecular scale, the same timing role is played by hysteresis. A molecular switch in a plasmonic nanocavity creates a double-well potential whose radiation-pressure tilt depends on cavity occupation; the cavity frequency and damping depend on the molecular coordinate, so the system alternates autonomously between hot- and cold-dominated configurations through hysteretic feedback (Zhu et al., 2024). A different extension, the adaptive three-level engine, couples the working medium to a Brownian controller that shifts its level spacings and thereby changes the region of temperature space in which it functions as an engine; the enhanced version derives conditions for maximum power extraction for all hot- and cold-bath temperatures (Khanahmadi et al., 2021).

4. Representative physical architectures

The topic is best understood as a set of recurring architectures rather than a single canonical model. The following platforms recur across the literature.

Platform	Working medium and output	Autonomous mechanism
Superconducting-qubit/Josephson engine (Verteletsky et al., 2019)	Two superconducting qubits; DC electrical work across a voltage-biased Josephson junction	Resonant excitation exchange plus two local heat baths
Josephson thermoelectric engine (Marchegiani et al., 2016)	N–FI–S thermoelectric element and Josephson weak link; contactless AC power to an RL load	Static thermal gradient drives Josephson DC/AC conversion
Electron-shuttle engine (Tonekaboni et al., 2018)	Nanomechanical oscillator with single-electron level; mechanical power dumped to a cold bosonic bath	Electronic occupation switches the hot Johnson-noise coupling
Quantum-rotor engines (Roulet et al., 2016, Roulet et al., 2018)	Harmonic or cavity working mode plus rotor angle and angular momentum; work stored as directed rotation	Angle-dependent bath couplings or filter-cavity valve timing
Non-Markovian optomechanical engine (Rasola et al., 2024)	Optical mode as working fluid, mechanical mode as driving/output mode	Internal coherent modulation sweeps the working mode across structured baths
Molecular optomechanical engine (Zhu et al., 2024)	Plasmonic cavity mode and molecular switch; work deposited in the switch	Radiation-pressure-induced hysteresis alternates hot and cold coupling
Superconducting Otto-like resonator engine (Rasola et al., 12 Feb 2025, Uusnäkki et al., 16 Mar 2026)	Working-body resonator, low-frequency driving resonator, hot/cold filter resonators; coherent microwave output	SQUID-mediated dispersive modulation and filtered thermal noise

These architectures differ chiefly in where work is stored and how the internal timing variable is embodied. In the superconducting-qubit engine, each excitation transferred from the hot qubit to the cold qubit is accompanied by a Cooper-pair tunneling event against the voltage bias, delivering work $\mathcal{E}$5 (Verteletsky et al., 2019). In the thermoelectric Josephson device, the output is high-frequency AC electrical power inductively delivered to a generic RL load without galvanic contact (Marchegiani et al., 2016). In rotor engines, work is the growth of angular momentum or kinetic energy (Roulet et al., 2016, Roulet et al., 2018). In the newer superconducting resonator engines, the output is coherent microwave power emitted from a low-frequency resonator mode (Rasola et al., 12 Feb 2025, Uusnäkki et al., 16 Mar 2026).

5. Fluctuations, coherence, and hidden dynamics

A central result of autonomous-engine theory is that steady state does not imply featureless dynamics. In the superconducting-qubit Josephson engine, the quantum regression theorem gives two-time correlations

$\mathcal{E}$6

from which both work and heat fluctuations are computed. The work variance is determined by the current autocorrelation function, the cold-bath heat variance by a Glauber-type four-operator correlation, and the same correlation functions reveal hidden strokes: hot absorption, coherent transfer producing Josephson current, cold emission, and reset. In frequency space the corresponding spectra show peaks near $\mathcal{E}$7 with linewidths of order $\mathcal{E}$8, demonstrating that a cyclical sequence can be embedded in a time-independent steady state (Verteletsky et al., 2019).

Noise also differentiates classical and quantum performance. In the autonomous rotor heat engine, quantum simulations show consistently lower efficiency than the classical engine because of additional noise; backaction heating contributes positively to rotor kinetic-energy growth but not to useful work, and low-inertia quantum rotors lose phase stability much more rapidly than their classical counterparts (Roulet et al., 2016). This is a direct counterexample to the misconception that quantization alone improves engine performance.

Quantum coherence can also be the sole fuel. In the coherence-driven engine with no heat flow, the bath is a pure state with the same diagonal energy distribution as a Gibbs state, and the engine operates by charging qubits through energy-preserving Jaynes–Cummings interactions. The extractable coherent work is

$\mathcal{E}$9

and the efficiency is defined as $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$0. The extractable work is maximized when four copies of the charged qubits are used, while the highest efficiency is obtained for slightly lower temperatures and weaker system-bath coupling than those for optimal coherence charging (Aimet et al., 2022). This does not describe a heat-flow engine in the usual sense, but it clarifies how internal and external coherence, work locking, and system-bath correlations constrain autonomous work extraction.

Several platforms predict nonclassical output statistics. The non-Markovian optomechanical engine finds a steady-state mechanical Fano factor of approximately $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$1 and, for $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$2, a thermal fraction of about $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$3 in the mechanical steady state, indicating that most stored energy is coherent (Rasola et al., 2024). The molecular engine predicts sub-Poissonian photon statistics $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$4 and negative Wigner-function regions for sufficiently strong $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$5 and $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$6, together with a correlation-power term that is absent semiclassically (Zhu et al., 2024).

6. Experimental status, validation, and unresolved issues

Experimental work has proceeded along two distinct paths: digital validation of autonomous-engine dynamics and direct analog realization. A gate-level superconducting-circuit experiment encoded the steady-state dynamics of the Scovil–Schulz–DuBois three-level engine on IBM’s ibm_lagos backend. After error mitigation, the measured population trajectories and steady-state populations closely matched the GKLS prediction and the noiseless simulation, confirming that constant Hamiltonian and dissipative couplings suffice to drive continuous autonomous cycling to a stationary state (Deng et al., 2024).

Direct analog realization was reported in a superconducting circuit that implements an approximate Otto cycle using a working-body resonator, a low-frequency driving resonator, hot and cold filter resonators, and a SQUID-mediated dispersive coupling. With no coherent probe signal, the device exhibited a narrow Lorentzian emission peak near $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$7, a ring in the IQ plane, and Poissonian statistics of the emitted power, and the coherent output power extracted from the emission peak was approximately $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$8 (Uusnäkki et al., 16 Mar 2026). The same work reports a dramatic increase of the driving resonator’s internal quality factor near threshold and emission stable for hours, thereby converting a long-standing proposal for superconducting autonomous heat engines into an experimental platform (Rasola et al., 12 Feb 2025, Uusnäkki et al., 16 Mar 2026).

Earlier superconducting hardware had already established experimentally relevant scales. The Josephson quantum heat engine based on a N–FI–S thermoelectric junction and a Josephson weak link predicts power delivery up to $$dE = n\, d\mathcal{E} + \mathcal{E}\, dn = \delta W + \delta Q,$$9 to a load $\delta W$0, with numerical results of approximately $\delta W$1 for $\delta W$2 and an efficiency $\delta W$3 for the studied parameters because most electrical power is dissipated in the shunt resistor rather than reaching the load (Marchegiani et al., 2016). By contrast, the two-qubit Josephson engine operates in the femtowatt regime, with maximal power near $\delta W$4 for the parameter set quoted in the analysis (Verteletsky et al., 2019).

The principal unresolved issues recur across platforms. Weak-coupling, Born–Markov, rotating-wave, and local-master-equation assumptions underlie much of the analytical literature, and several papers explicitly note that strong coupling, dressed-state effects, or non-Markovian reservoirs can invalidate those descriptions (Verteletsky et al., 2019). Structured reservoirs with finite tails blur the separation between isochoric and adiabatic strokes, which is why the experimentally realized superconducting engines are described as approximate rather than ideal Otto cycles (Rasola et al., 12 Feb 2025, Uusnäkki et al., 16 Mar 2026). Work identification also remains task dependent: coherent emitted power, ergotropy stored in a piston, energy transferred to a load, and controller-associated energy flux are not interchangeable notions (Niedenzu et al., 2019, Tonekaboni et al., 2018). This suggests that the mature description of autonomous quantum heat engines will continue to combine microscopic dynamics, explicit resource accounting, and measurement-specific definitions of output rather than relying on a single universal work observable.