The Sagawa–Ueda generalized second law bounds the work extractable from a thermodynamic system under discrete quantum feedback control by the free-energy change plus kBTk_B T times the generalized mutual information between the measured system and the controller's outcomes, but the 2008 derivation invokes — without computing — an unspecified energy cost of the controller's information processing to reconcile the result with the ordinary second law, and assumes macroscopic equilibration of the final state without microscopic justification. The problem asks for a complete information-to-work balance sheet: a tight minimum work cost for implementing a general quantum controller (measurement, coherent processing, storage, erasure), valid for all finite-dimensional controller models, composed with the feedback work gain into a single combined inequality for the total cycle; the paper's macroscopic equilibration assumption is retained as part of the setting rather than required to be resolved. Later literature has settled the principle-level question (no net second-law violation when the demon is included as a physical system) and supplied controller-side costs at the framework level — model- and state-dependent measurement and erasure bounds, and general process-cost bounds in the resource theory of thermal operations — but these do not transfer directly to the original Hamiltonian finite-bath feedback cycle, so a tight combined inequality for general quantum controllers in that setting remains open.

Background

In a discrete quantum feedback protocol, a thermodynamic system S (a quantum system with a time-dependent Hamiltonian, coupled to heat baths) is measured at some point of its evolution, and the subsequent dynamics is conditioned on the measurement outcome. The measurement is described by measurement operators (Kraus operators) {M^k}\{\hat M_k\} satisfying ∑kM^k†M^k=1^\sum_k \hat M_k^\dagger \hat M_k = \hat 1, a general formulation of quantum measurement (POVM); the feedback is a collection of outcome-conditioned unitary evolutions {U^k}\{\hat U_k\}. Maxwell's demon is the thought-experiment ancestor of such protocols: an agent who uses information about the microscopic state to extract work apparently at the expense of a single heat bath. Sagawa and Ueda (Phys. Rev. Lett. 100, 080403 (2008)) proved that under such feedback the maximum extractable work from an isothermal process is Wext≤−ΔF+kBT I(ρ^1:X)W_{\rm ext} \le -\Delta F + k_B T\, I(\hat\rho_1 : X), where ΔF\Delta F is the free-energy change, kBk_B Boltzmann's constant, TT the bath temperature, and I(ρ^1:X)I(\hat\rho_1 : X) a generalized (Holevo-type) quantum–classical mutual information between the post-measurement system state and the classical outcome XX; the extra term kBT Ik_B T\, I is the work value of the acquired information. In a two-bath heat cycle this permits an efficiency exceeding the Carnot value — consistent with the second law only because the controller itself must perform information processing at a cost. That cost is the subject of Landauer's principle (erasing one bit of memory at temperature TT dissipates at least kBTln⁡2k_B T \ln 2), but in 2008 it entered the argument only as a qualitative invocation: the paper computed no lower bound on the work needed to measure, process, store, or erase the outcomes for a general quantum controller, and offered no single inequality combining the feedback work gain kBT Ik_B T\, I with the controller's minimal energetic cost. In addition, the derivation assumed that by the end of the protocol the system and bath reach thermodynamic equilibrium only macroscopically — the final density operator ρ^f\hat\rho_f need not equal the exact canonical distribution ρ^fcan=e−βH^f/Z\hat\rho_f^{\rm can} = e^{-\beta \hat H_f}/Z (with β=1/(kBT)\beta = 1/(k_B T) and ZZ the partition function) — and the microscopic conditions and dynamical timescales under which this equilibration assumption holds were not analyzed. Subsequent developments define the modern context of the question. Sagawa and Ueda themselves (Phys. Rev. Lett. 102, 250602 (2009)) derived fundamental lower bounds on the work costs of quantum measurement (with errors) and of information erasure, finding that Landauer's bound is vindicated only in a special case and otherwise breaks down — the minimal cost is a model- and state-dependent quantity, not a universal function of the logical entropy. Reeb and Wolf (New J. Phys. 16, 103011 (2014)) proved a rigorous, model-independent quantum Landauer equality with finite-size bath corrections, treating the bath microscopically. The thermodynamics-of-information program (reviewed by Parrondo, Horowitz and Sagawa, Nat. Phys. 11, 131 (2015)) established that when the demon (controller) is included as a physical system, the total entropy production of system plus memory is nonnegative, so no net second-law violation occurs; autonomous-demon models such as Strasberg et al. (Phys. Rev. Lett. 110, 040601 (2013)) realize this with fully specified physical controllers, and experiments (Camati et al., Phys. Rev. Lett. 117, 240502 (2016); Cottet et al., PNAS 114, 2017) have measured both the extracted work and the entropy left in the demon's memory. On the processing side, the resource theory of quantum thermodynamics — thermal operations, whose allowed processes are energy-conserving unitaries on the system plus heat baths and therefore preserve the thermal equilibrium state — yielded general work-cost bounds for implementing arbitrary logical quantum processes (Faist and Renner, Phys. Rev. X 8, 021011 (2018), via the 'coherent relative entropy', a single-shot measure of the work cost of implementing one quantum process by another), and the quantum-measurement energetics literature (Elouard et al., npj Quantum Inf. 3, 9 (2017); Mohammady, Auffèves and Anders, Commun. Phys. 3, 89 (2020)) identified energetic costs of measurement backaction and decoherence beyond Landauer. On the equilibration side, typicality and finite-time equilibration theorems (e.g. Short and Farrelly, New J. Phys. 14, 013063 (2012)) give sufficient conditions — the state spreads over many non-degenerate energy gaps — and bounds on equilibration times, but the known bounds are extremely loose and a sharp general characterization remains an open program in quantum statistical mechanics. Against this background, what is still missing is a single tight, model-independent second-law statement for the original Sagawa–Ueda feedback cycle that quantifies the minimum total work to implement the general quantum controller — measurement, coherent processing of the outcomes, storage, and erasure — and composes it with the kBT Ik_B T\, I feedback gain; the source paper's macroscopic equilibration assumption is retained as part of the setting.

Problem Statement

For the discrete quantum feedback cycle of Sagawa and Ueda (Phys. Rev. Lett. 100, 080403 (2008)) — a finite-dimensional quantum system and heat baths evolving unitarily, a measurement described by general measurement operators {M^k}\{\hat M_k\}, outcome-conditioned feedback unitaries {U^k}\{\hat U_k\}, and a final relaxation stage assumed only to bring the system and bath to macroscopic equilibrium — the final state need not equal the exact canonical state ρ^fcan=e−βH^f/Z\hat\rho_f^{\rm can} = e^{-\beta \hat H_f}/Z but must be characterised by the bath temperature TT, with mean energy equal to that of ρ^fcan\hat\rho_f^{\rm can} (an assumption retained from the source and not required to be resolved) — the information-to-work balance sheet consists of four components: (i) a characterization of the minimum total work required to physically implement the feedback controller for a general quantum controller — acquiring the measurement outcome, coherently processing it, storing it, and erasing the memory; (ii) a combined second-law inequality for the complete cycle in which this minimal controller cost appears together with the feedback work gain kBT I(ρ^1:X)k_B T\, I(\hat\rho_1 : X), bounding the net work extractable from the total setup (thermodynamic system plus controller plus memory); (iii) recovery of the Sagawa–Ueda bound Wext≤−ΔF+kBT I(ρ^1:X)W_{\rm ext} \le -\Delta F + k_B T\, I(\hat\rho_1 : X) as the costless limit in which controller costs are excluded; and (iv) an explicit implementation achieving equality in the combined inequality for every controller in the declared class, establishing its tightness. Their standing is: (i) supplies the cost term that enters (ii); (ii) is the central result; (iii) is the consistency corollary of (ii) obtained in the costless limit rather than an independent deliverable; and (iv) is the achievability of (ii) for every controller in the declared class. The required result is the tight combined inequality — (ii) together with its achievability (iv) for every controller in the declared class. Here 'model-independent' means that the combined inequality is established as a theorem for all finite-dimensional controller models within the declared class, with the bound permitted to depend on the state and on the controller's physical specification (Hamiltonians, memory dimension, baths); it does not require a single universal cost expression as a function of the logical entropy, which Sagawa and Ueda (Phys. Rev. Lett. 102, 250602 (2009)) showed cannot hold in general. Since existing frameworks do not share a common work definition, the work notion is fixed to the average work, applied consistently to both sides of the combined inequality. The controllers in scope are the general class of finite-dimensional controller models, each fixed by its physical specification (memory dimension, controller Hamiltonians, baths and work-storage devices), on which the bound may depend. For each such controller, with its specification held fixed, the minimum total work to physically implement it is taken over the allowed implementation processes — energy-conserving unitary processes on the system together with the controller's declared baths, memory and work-storage devices that reproduce the controller's measurement and feedback — and the combined inequality holds for every controller in the class. The result is required for general finite-dimensional settings — general measurement operators {M^k}\{\hat M_k\}, outcome-conditioned feedback unitaries {U^k}\{\hat U_k\}, and physical controllers — rather than within a single restricted controller model.

References

  1. T. Sagawa and M. Ueda, 'Second Law of Thermodynamics with Discrete Quantum Feedback Control', Physical Review Letters 100, 080403 (2008). DOI: 10.1103/PhysRevLett.100.080403; arXiv:0710.0956; https://doi.org/10.1103/PhysRevLett.100.080403
  2. T. Sagawa and M. Ueda, 'Minimal Energy Cost for Thermodynamic Information Processing: Measurement and Information Erasure', Physical Review Letters 102, 250602 (2009). DOI: 10.1103/PhysRevLett.102.250602; arXiv:0809.4098; https://doi.org/10.1103/PhysRevLett.102.250602
  3. T. Sagawa and M. Ueda, 'Nonequilibrium thermodynamics of feedback control', Physical Review E 85, 021104 (2012). DOI: 10.1103/PhysRevE.85.021104; https://doi.org/10.1103/PhysRevE.85.021104
  4. J. M. R. Parrondo, J. M. Horowitz, and T. Sagawa, 'Thermodynamics of information', Nature Physics 11, 131–139 (2015). DOI: 10.1038/nphys3230; https://doi.org/10.1038/nphys3230
  5. D. Reeb and M. M. Wolf, 'An improved Landauer principle with finite-size corrections', New Journal of Physics 16, 103011 (2014). DOI: 10.1088/1367-2630-16-10-103011; arXiv:1306.4352; https://doi.org/10.1088/1367-2630-16-10-103011
  6. P. Strasberg, G. Schaller, T. Brandes, and M. Esposito, 'Thermodynamics of a Physical Model Implementing a Maxwell Demon', Physical Review Letters 110, 040601 (2013). DOI: 10.1103/PhysRevLett.110.040601; arXiv:1210.5661; https://doi.org/10.1103/PhysRevLett.110.040601
  7. P. A. Camati, J. P. S. Peterson, T. B. Batalhão, K. Micadei, A. M. Souza, R. S. Sarthour, I. S. Oliveira, and R. M. Serra, 'Experimental Rectification of Entropy Production by Maxwell's Demon in a Quantum System', Physical Review Letters 117, 240502 (2016). DOI: 10.1103/PhysRevLett.117.240502; arXiv:1605.08821; https://doi.org/10.1103/PhysRevLett.117.240502
  8. N. Cottet, S. Jezouin, L. Bretheau, P. Campagne-Ibarcq, Q. Ficheux, J. Anders, A. Auffèves, R. Azouit, P. Rouchon, and B. Huard, 'Observing a quantum Maxwell demon at work', Proceedings of the National Academy of Sciences USA 114, 7561–7564 (2017). DOI: 10.1073/pnas.1704827114; arXiv:1702.05161; https://doi.org/10.1073/pnas.1704827114
  9. P. Faist and R. Renner, 'Fundamental work cost of quantum processes', Physical Review X 8, 021011 (2018). DOI: 10.1103/PhysRevX.8.021011; arXiv:1709.00506; https://doi.org/10.1103/PhysRevX.8.021011
  10. C. Elouard, D. A. Herrera-Martí, M. Clusel, and A. Auffèves, 'The role of quantum measurement in stochastic thermodynamics', npj Quantum Information 3, 9 (2017). DOI: 10.1038/s41534-017-0008-4; https://doi.org/10.1038/s41534-017-0008-4
  11. M. H. Mohammady, A. Auffèves, and J. Anders, 'Energetic footprints of irreversibility in the quantum regime', Communications Physics 3, 89 (2020). DOI: 10.1038/s42005-020-0356-9; arXiv:1907.06559; https://doi.org/10.1038/s42005-020-0356-9
  12. A. J. Short and T. C. Farrelly, 'Quantum equilibration in finite time', New Journal of Physics 14, 013063 (2012). DOI: 10.1088/1367-2630-14-1-013063; arXiv:1110.5759; https://doi.org/10.1088/1367-2630-14-1-013063