Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Quantum state-agnostic work extraction (almost) without dissipation

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An adaptive protocol extracts work from unknown qubit copies with cumulative dissipation that grows only polylogarithmically in the number of copies, beating the square-root scaling of tomography-first strategies.

desk verdict A promising RL-to-thermodynamics transfer that probably works after a technical fix, but the explicit ε_k schedule in Theorem 7 breaks the learning algorithm's reward channel. read the letter →

arxiv 2505.09456 v1 pith:LVMTTIW6 submitted 2025-05-14 quant-ph cs.AIcs.LG

Josep Lumbreras , Ruo Cheng Huang , Yanglin Hu , Mile Gu , Marco Tomamichel This is my paper · ORCID
classification quant-phcs.AIcs.LG
keywords unknownpurequbitworkextractionstate-agnosticthermodynamicscumulativedissipationmulti-armedbanditsadaptivetomographyreinforcementlearningquantum
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks how much energy is unavoidably wasted when work is extracted from many identical copies of an unknown pure qubit state, and answers that a fully adaptive protocol can make the waste almost negligible. Interleaving learning and extraction, it achieves cumulative dissipation $O(\beta^{-1}\ln^2(N)\ln(N/\delta))$ with probability at least $1-\delta$, an exponential improvement over the $\Omega(\sqrt{N})$ dissipation that any tomography-then-extract strategy must pay. The reason the improvement is possible is that the battery-energy measurement already produces the binary outcomes needed to learn the state, so every round contributes both power and information. A sympathetic reader would take the paper's contribution to be this reduction of a thermodynamic problem to a bandit-learning problem, plus an explicit algorithm that attains the polylogarithmic bound.

What carries the argument

The load-bearing mechanism is the identification of each work-extraction round with a bandit reward: the battery-energy measurement in the thermal protocol is, in the quasi-static limit, a projective measurement of the unknown state along the current guess $\psi_k$, with outcomes distributed by Born's rule. The paper couples this to the companion paper's learning algorithm (named LinUCB-VVN in Algorithm 1), which maintains a variance-weighted design matrix and a median-of-means estimator and guarantees per-round infidelity $1-|\langle\psi_k|\psi\rangle|^2 \le C\ln(N/\delta)/k$ with high probability. A second ingredient, Theorem 6 in this paper, converts that infidelity into the relative-entropy dissipation bound $D(\psi\parallel\Delta_{2\epsilon}(\hat\rho)) \le 16\epsilon(2+\ln(d/(2\epsilon)))$; choosing $\epsilon_k$ to match the infidelity guarantee makes the per-round dissipation sum to polylogarithmic order.

What would settle it

Run the proposed protocol for a fixed $N$ with the battery-energy rewards, choose the unknown pure state adversarially after reading the algorithm, and check whether the empirical cumulative dissipation $\sum_k D(\psi\parallel\Delta_{2\epsilon_k}(\psi_k))$ stays within $O(\beta^{-1}\ln^2(N)\ln(N/\delta))$ with probability $1-\delta$; a single explicit state and outcome sequence violating that scaling, or violating the per-round infidelity bound $1-|\langle\psi_k|\psi\rangle|^2 \le C\ln(N/\delta)/k$, would falsify the main result.

Watch

Extended reading notes

Core claim

The main result (Eq. 15) is an explicit protocol for a semi-classical battery model in which, at each round $k$, the agent picks a guessed direction $\psi_k$ and an accuracy $\epsilon_k$ from previous battery-energy rewards, runs a quasi-static thermal swap to charge the battery, and uses the measured energy as the next learning signal. The expected work lost at round $k$ is $\beta^{-1}D(\psi\parallel\Delta_{2\epsilon_k}(\psi_k))$, where $\Delta_\epsilon$ is the depolarizing channel; the paper sets $\epsilon_k = C\ln(N/\delta)/k$ and invokes the learning algorithm's per-round infidelity guarantee to convert this into $W_{\mathrm{diss}}(N) = O(\beta^{-1}\ln^2(N)\ln(N/\delta))$ with probability at least $1-\delta$. The same adaptive reward structure also gives $O(\omega\ln(N)\ln(N/\delta))$ dissipation in a ladder-battery model, and the Landauer erasure cost of the measurement memory remains polylogarithmic if fresh memory registers are used and erased quasi-statically at the end.

Load-bearing premise

The polylogarithmic dissipation claim rests on the companion paper's guarantee that every round's infidelity $1-|\langle\psi_k|\psi\rangle|^2$ is at most $C\ln(N/\delta)/k$ with high probability; this paper quotes that theorem rather than proving it, so if the guarantee fails in any round the dissipation bound collapses.

Editorial extensions

If this is right

  • With $N$ copies of an unknown pure qubit, the total energy lost to state ignorance can be brought down to polylogarithmic in $N$, so for large $N$ almost all of the non-equilibrium free energy is recovered.
  • Any strategy that first spends copies on tomography and then extracts from the remaining copies suffers at least $\Omega(\sqrt{N})$ cumulative dissipation, so integrated adaptive learning is strictly better asymptotically than the two-phase approach.
  • The same learning signal can drive a ladder-battery model, where the cumulative dissipation is $O(\omega \ln(N)\ln(N/\delta))$, showing the mechanism is not tied to the semi-classical weight model.
  • Including the thermodynamic cost of recording and erasing the measurement outcomes (Landauer's principle), the total dissipation remains polylogarithmic, provided the memory is erased quasi-statically at the end.
  • The per-round accuracy $\epsilon_k = C\ln(N/\delta)/k$ means early rounds carry most of the dissipation, while later rounds approach reversibility as the estimate sharpens.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Extrapolation: because the dissipation bound inherits the learner's per-round infidelity guarantee, a future learner with a tighter per-round regret bound would immediately tighten the thermodynamic bound; conversely, a lower bound on per-round infidelity for any adaptive measurement strategy would locate the true optimal dissipation.
  • The identification of battery readout with a reward measurement suggests the same exploration-exploitation trade-off governs extraction of other quantum resources, such as coherence, entanglement, or magic, but the paper only conjectures that extension.
  • The degenerate-Hamiltonian assumption sidesteps work-locking; extending the proof to a non-degenerate Hamiltonian would require paying for the rotation energy explicitly, so the stated polylog bound should not be assumed to carry over without modification.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper studies sequential work extraction from N identical copies of an unknown pure qubit state into a semi-classical battery, using thermal operations with a reservoir. The agent adaptively chooses a measurement direction ψ_k and an accuracy parameter ε_k based on previous battery energy measurements, and the paper analyzes the cumulative dissipation W_diss(N) relative to the optimal state-aware protocol. The central claim, Theorem 7 (Eq. 15), is that an explicit adaptive protocol based on the LinUCB-VVN learning algorithm of the companion paper [18] achieves W_diss(N) = O(β^{-1} ln^2(N) ln(N/δ)) with probability at least 1−δ, an exponential improvement over the Ω(√N) dissipation of tomography-based two-phase strategies. The paper also discusses the Jaynes-Cummings battery variant and the additional cost of measurement and memory erasure, claiming that both remain polylogarithmic.

Significance. If established, the result is significant: it connects state-agnostic quantum work extraction to the exploration-exploitation trade-off in reinforcement learning, and it shows that fully adaptive protocols can beat the √N dissipation barrier for pure qubit states. The dissipation-regret relation is conceptually clean, and the extension to a Jaynes-Cummings battery and to Landauer erasure costs adds value. The paper is also explicit in presenting algorithms and in identifying the learning-theoretic guarantee on which the bound rests. However, the proof of the central theorem currently has internal gaps in the explicit accuracy schedule and does not fully account for finite-M corrections, so the significance is conditional on those issues being resolved.

major comments (3)
  1. [Appendix B, Eq. (B56) and Theorem 7] The accuracy schedule ε_k = min(C ln(N/δ)/k, 1/2) sets ε_k = 1/2 for all k below k* ≈ 2C ln(N/δ). For ε = 1/2, Eq. (7) gives w_{k,0} = β^{-1}(D(ψ_k‖1/2)+ln(1/2)) = 0 and w_{k,1} = β^{-1}(D(ψ_k^⊥‖1/2)+ln(1/2)) = 0, because D(·‖1/2) = ln 2 for every pure qubit state. The battery measurement is then deterministic and uninformative (in fact ν(τ,1/2) = 0 for all τ), so the threshold rule (B37) cannot produce the Born-rule rewards r_k that Algorithm 1 requires. Consequently, the infidelity guarantee of Theorem 1 cannot be invoked for these rounds, and the proof of Theorem 7 does not apply to the protocol as stated. This is not a harmless asymptotic detail because the first k* rounds initialize the design matrix and the weighted least-squares estimators. The issue is fixable by capping ε strictly below 1/2, for example ε_k = min(C ln(N/δ)/k, 1/2 − c/k) for a suitable constant c, which keeps the early-round dissipation O(ln N) and restores distinct work values.
  2. [Appendix B, Eq. (B30) and Theorem 5] The expected-work formulas used in the main proof, namely Eq. (3), Eq. (7), and Theorem 5, are derived only in the M→∞ quasi-static limit. Theorem 4 (Eq. B30) shows that the expected work is w_i + O(1/M) for each round, and the noiseless reward correspondence (B39) is only a limit statement. The cumulative effect of the O(1/M) per-round correction is O(N/M). The paper suggests in Appendix B that M need grow only polylogarithmically, but that would leave N/M super-polylogarithmic and would destroy the claimed O(polylog N) scaling in Eq. (15). Theorem 7 neither specifies M nor bounds the finite-M correction. The authors should either state the main theorem as an M→∞ idealization, or choose M and prove that the cumulative finite-time correction is O(polylog N).
  3. [Appendix A, Theorem 1] The central bound Eq. (15) is a direct function of the per-round infidelity guarantee 1−|⟨ψ_k|ψ⟩|^2 ≤ C ln(N/δ)/k quoted from the companion preprint [18] and restated as Theorem 1 in Appendix A. No proof of this guarantee is included in the present manuscript. Because the guarantee is load-bearing for every round in the proof of Theorem 7, the paper should either include a self-contained proof of Theorem 1 (or a clear derivation of the per-round bound) or cite a peer-reviewed version of [18]; otherwise the main result remains conditional on an external result whose proof is not verified here.
minor comments (5)
  1. [Eq. (8)] The formula Pr(ΔW_k = w_{k,i}) = i + (−1)^i |⟨ψ|ψ_k⟩|^2 is concise but easy to misread; writing the two cases i = 0 and i = 1 explicitly would improve readability.
  2. [Main Result, paragraph after Eq. (15)] The text “Laudauer’s principle” should read “Landauer’s principle”.
  3. [References [28] and [30]] References [28] and [30] appear to describe the same paper (Finite-Time Quantum Landauer Principle and Quantum Coherence, Phys. Rev. Lett. 128, 010602) with different author order; please consolidate and correct the citation.
  4. [Proof of Theorem 7, Eq. (B56)] The phrase “C is the constant in Theorem 1” is ambiguous because Theorem 1 states four constants C1–C4; please refer explicitly to the constant appearing in the per-round infidelity bound.
  5. [Algorithm 3] Algorithm 3 lists as input a fixed sequence {ε_k}, but the proof of Theorem 7 chooses ε_k by a formula and the protocol is meant to be adaptive; please clarify whether the sequence is precomputed from N and δ or updated online.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the polylog dissipation bound is a modular consequence of an external regret/infidelity theorem and standard relative-entropy inequalities, not an equivalent restatement of its inputs.

full rationale

The main result (Eq. 15) is derived by bounding each round's dissipation W_k^diss = D(psi || Delta_{2 epsilon_k}(psi_k)) (Eq. 5) with the Flammia-O'Donnell inequality (Eq. 11 / Theorem 6), then choosing epsilon_k from the per-round infidelity guarantee of Theorem 1. The thermodynamic work values w_{k,i} (Eq. 7) and the reward distribution (Eqs. 8, 10, B37) are independently defined from the battery model; the dissipation is not fitted to the learning regret, nor is the regret defined in terms of dissipation. The quoted regret/infidelity theorem [18, Theorems 9 and 11] is parameter-free, assumes only sequential pure-state measurements with Born-rule rewards, and does not assume the target dissipation result, so citing it as a module is independent support rather than circularity. A separate technical gap, not a circularity, affects Theorem 7's explicit schedule: setting epsilon_k = min(C ln(N/delta)/k, 1/2) (Eq. B56) gives epsilon_k = 1/2 in early rounds, and for epsilon = 1/2 Eq. (7) yields w_{k,0} = w_{k,1} = 0 because D(psi_k || 1/2) = D(psi_k^perp || 1/2) = ln 2; the reward map (B37) is then undefined and the battery outcomes need not follow Eq. (10). This would break the proof of Eq. (15) as stated, but it is an internal consistency and correctness issue, not an equivalence by construction, and a cap strictly below 1/2 would restore well-defined rewards while preserving the polylogarithmic bound.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No free parameters or new entities are introduced; the accuracy schedule ε_k is constructed from the learning theorem's constants and does not fit data. All assumptions are modeling choices or inherited theorems.

assumptions (6)
  • domain assumption Infidelity bound for LinUCB-VVN (Theorem 1 of [18])
    The central dissipation bound assumes the adaptive learning algorithm achieves per-round infidelity ≤ C ln(N/δ)/k; this is not proved in the present paper but inherited from the authors' companion preprint [18].
  • domain assumption Work extraction protocol realizes a projective reward measurement in the M→∞ limit
    Theorem 4 and Eq. (B38) establish the reward distribution in the quasi-static limit; the main theorem uses finite M without fully tracking the O(1/M) work-error contribution.
  • domain assumption Degenerate system Hamiltonian removes work-locking
    The protocol assumes H_A = ω 1/2, so every basis is an energy eigenbasis; with a non-degenerate Hamiltonian, additional energy costs would appear (Section B.4).
  • domain assumption Semi-classical battery model
    The battery is a continuous weight with sharp energy; quantum coherences of the battery are assumed not to affect the measured work distribution.
  • standard math Theorem 6 (Flammia-O'Donnell) bounding D(ρ||Δ_{2ε}(ρ̂)) by 16ε(2+ln(d/2ε))
    Published result [7], used to convert infidelity into relative-entropy dissipation.
  • domain assumption Landauer erasure is implemented quasi-statically at the end (Appendix D)
    The measurement and memory costs are assumed to obey Landauer's bound, achievable only in a quasi-static reset at the end of the protocol.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum state-agnostic work extraction (almost) without dissipation." pith.science (2026). https://pith.science/paper/LVMTTIW6

@misc{pith2026250509456,
  author       = {Pith},
  title        = {Pith review of: Quantum state-agnostic work extraction (almost) without dissipation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LVMTTIW6}},
  note         = {Machine review of arXiv:2505.09456}
}
abstract

We investigate work extraction protocols designed to transfer the maximum possible energy to a battery using sequential access to $N$ copies of an unknown pure qubit state. The core challenge is designing interactions to optimally balance two competing goals: charging of the battery optimally using the qubit in hand, and acquiring more information by qubit to improve energy harvesting in subsequent rounds. Here, we leverage exploration-exploitation trade-off in reinforcement learning to develop adaptive strategies achieving energy dissipation that scales only poly-logarithmically in $N$. This represents an exponential improvement over current protocols based on full state tomography.

Figures

Figures reproduced from arXiv: 2505.09456 by the authors.

Figure 1
Figure 1. Sketch of the sequential work extraction protocol [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the repetitions of the thermal oper [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Time-ordered free energy in correlated quantum systems: An agentic approach

    quant-ph 2026-08 conditional novelty 6.0 of 10

    An online agent harvesting work from a correlated quantum state stream has a computable optimal policy, and its efficiency loss relative to unconstrained extraction equals the entropy increase from sequential measurem...

Reference graph

Works this paper leans on

47 extracted references · 45 canonical work pages · cited by 1 Pith paper

  1. [18]

    High- Precision Test of Landauer’s Principle in a Feedback Trap

    Y. Jun, M. c. v. Gavrilov, and J. Bechhoefer.“High- Precision Test of Landauer’s Principle in a Feedback Trap”. Phys. Rev. Lett.113: 190601 (2014)

    work page 2014

  2. [1]

    The agent receives a sample of the unknown qubit stateψ

    work page

  3. [2]

    This computation is done using the previously selected directions and measured battery energies{ψ s,µs}k−1 s=1

    Based on the outcomes from previous rounds, the agent selects a directionψ k on the Bloch sphere, sets an accuracyϵ k ∈[0,1] and defines a basis{ψ k,ψ⊥ k} for systemA. This computation is done using the previously selected directions and measured battery energies{ψ s,µs}k−1 s=1

    work page

  4. [3]

    We first implement a unitary on the system qubit in the form of Uk =|0⟩⟨ψk|+ 1 ψ⊥ k ,(1) satisfying [HA,Uk] = 0 which tries to diagonalize the system qubit in computational basis

    work page

  5. [4]

    Specifically, for each ofMrepetitions (indexed byτ∈[M]), the agent: •Sets the energy gap of the reservoir qubit toν(τ,ϵ k) (see Eq

    The agent then performs a thermal operation by re- peatedly appending a reservoir qubitR, applying an energy-conserving unitary on the combined system ABRto transfer energy fromRto the batteryBand discardingR. Specifically, for each ofMrepetitions (indexed byτ∈[M]), the agent: •Sets the energy gap of the reservoir qubit toν(τ,ϵ k) (see Eq. (6)) and gets a...

    work page

  6. [5]

    Are quan- tum agents more energetically efficient at making pre- dictions?

    After completing theMsteps, the agent measures the energy of the battery in its eigenbasis and records the energyµ k. For the above protocol, the expected work extracted at roundk∈[N], as a function of the chosen direction ψk and accuracy parameterϵ k, is given by E[∆Wk] =β−1 [D(ψ∥1/2)−D(ψ∥∆ 2ϵk(ψk))],(3) where ∆ϵ(ρ) = (1−ϵ)ρ+ϵ1/2 denotes the depolarizing...

    work page 1903

  7. [6]

    Catalytic Coherence

    J. ˚Aberg.“Catalytic Coherence”. Phys. Rev. Lett. 113: 150402 (2014)

    work page 2014

  8. [7]

    Truly work-like work extraction via a single- shot analysis

    J. ˚Aberg.“Truly work-like work extraction via a single- shot analysis”. Nature communications4(1): 1925 (2013)

    work page 2013

Show all 47 references
  1. [8]

    Maximal work extraction from finite quantum systems

    A. E. Allahverdyan, R. Balian, and T. M. Nieuwenhuizen. “Maximal work extraction from finite quantum systems”. Europhysics Letters67(4): 565 (2004)

    2004

  2. [9]

    Resource theory of quan- tum states out of thermal equilibrium

    F. G. Brandao, M. Horodecki, J. Oppenheim, J. M. Renes, and R. W. Spekkens.“Resource theory of quan- tum states out of thermal equilibrium”. Physical review letters111(25): 250404 (2013)

    2013

  3. [10]

    When Does Adaptivity Help for Quantum State Learning?

    S. Chen, B. Huang, J. Li, A. Liu, and M. Sellke.“When Does Adaptivity Help for Quantum State Learning?”. In 2023 IEEE 64th Annual Symposium on Foundations of Computer Science (FOCS), pages 391–404, (2023)

    2023

  4. [11]

    Efficient quan- tum measurement engines

    C. Elouard and A. N. Jordan.“Efficient quan- tum measurement engines”. Physical review letters 120(26): 260601 (2018)

    2018

  5. [12]

    Quantum chi-squared tomography and mutual information testing

    S. T. Flammia and R. O’Donnell.“Quantum chi-squared tomography and mutual information testing”. Quantum 8: 1381 (2024)

    2024

  6. [13]

    Nonequilib- rium Quantum Landauer Principle

    J. Goold, M. Paternostro, and K. Modi.“Nonequilib- rium Quantum Landauer Principle”. Phys. Rev. Lett. 114: 060602 (2015)

    2015

  7. [14]

    Sample-Optimal Tomography of Quantum States

    J. Haah, A. W. Harrow, Z. Ji, X. Wu, and N. Yu. “Sample-Optimal Tomography of Quantum States”. IEEE Transactions on Information Theory63(9): 5628– 5641 (2017)

    2017

  8. [15]

    Engines for predictive work ex- traction from memoryful quantum stochastic processes

    R. C. Huang, P. M. Riechers, M. Gu, and V. Narasimhachar.“Engines for predictive work ex- traction from memoryful quantum stochastic processes”. Quantum7: 1203 (2023)

    2023

  9. [16]

    Work Extrac- tion from Unknown Quantum Sources

    D. ˇSafr´ anek, D. Rosa, and F. C. Binder.“Work Extrac- tion from Unknown Quantum Sources”. Phys. Rev. Lett. 130: 210401 (2023)

    2023

  10. [17]

    Comparison of quantum and semiclassical radiation theories with application to the beam maser

    E. Jaynes and F. Cummings.“Comparison of quantum and semiclassical radiation theories with application to the beam maser”. Proceedings of the IEEE51(1): 89– 109 (1963)

    1963

  11. [19]

    The extraction of work from quantum coherence

    K. Korzekwa, M. Lostaglio, J. Oppenheim, and D. Jen- nings.“The extraction of work from quantum coherence”. New Journal of Physics18(2): 023045 (2016)

    2016

  12. [20]

    Lattimore and C

    T. Lattimore and C. Szepesv´ ari.Bandit algorithms. Cam- bridge University Press (2020)

    2020

  13. [21]

    Descrip- 1 tion of quantum coherence in thermodynamic processes requires constraints beyond free energy

    M. Lostaglio, D. Jennings, and T. Rudolph.“Descrip- 1 tion of quantum coherence in thermodynamic processes requires constraints beyond free energy”. Nature commu- nications6(1): 6383 (2015)

    2015

  14. [22]

    Multi-armed quantum bandits: Exploration versus ex- ploitation when learning properties of quantum states

    J. Lumbreras, E. Haapasalo, and M. Tomamichel. “Multi-armed quantum bandits: Exploration versus ex- ploitation when learning properties of quantum states”. Quantum6: 749 (2022)

    2022

  15. [23]

    Learn- ing pure quantum states (almost) without regret

    J. Lumbreras, M. Terekhov, and M. Tomamichel.“Learn- ing pure quantum states (almost) without regret”. arXiv preprint arXiv:2406.18370 , (2024)

    2024 arXiv

  16. [24]

    Linear bandits with polylogarithmic minimax regret

    J. Lumbreras and M. Tomamichel.“Linear bandits with polylogarithmic minimax regret”. InProceedings of Thirty Seventh Conference on Learning Theory, volume 247 of Proceedings of Machine Learning Research, pages 3644– 3682, (2024)

    2024

  17. [25]

    Quantum Fluctuations Hinder Finite-Time Information Erasure near the Landauer Limit

    H. J. D. Miller, G. Guarnieri, M. T. Mitchison, and J. Goold.“Quantum Fluctuations Hinder Finite-Time Information Erasure near the Landauer Limit”. Phys. Rev. Lett.125: 160602 (2020)

    2020

  18. [26]

    Efficient quantum tomog- raphy

    R. O’Donnell and J. Wright.“Efficient quantum tomog- raphy”. InProceedings of the Forty-Eighth Annual ACM Symposium on Theory of Computing, STOC ’16, pages 899–912, New York, NY, USA(2016)

    2016

  19. [27]

    Initial-state dependence of thermodynamic dissipation for any quantum process

    P. M. Riechers and M. Gu.“Initial-state dependence of thermodynamic dissipation for any quantum process”. Phys. Rev. E103: 042145 (2021)

    2021

  20. [28]

    Minimal Energy Cost for Thermodynamic Information Processing: Measurement and Information Erasure

    T. Sagawa and M. Ueda.“Minimal Energy Cost for Thermodynamic Information Processing: Measurement and Information Erasure”. Phys. Rev. Lett.102: 250602 (2009)

    2009

  21. [29]

    M. O. Scully and M. S. Zubairy.Atom-field interaction - quantum theory, page 193–219. Cambridge University Press (1997)

    1997

  22. [30]

    Work ex- traction and thermodynamics for individual quantum sys- tems

    P. Skrzypczyk, A. J. Short, and S. Popescu.“Work ex- traction and thermodynamics for individual quantum sys- tems”. Nature communications5(1): 4185 (2014)

    2014

  23. [31]

    Strang.Calculus

    G. Strang.Calculus. Wellesley-Cambridge Press (2019)

    2019

  24. [32]

    ¨Uber die Entropieverminderung in einem thermodynamischen System bei Eingriffen intelligenter Wesen

    L. Szilard.“ ¨Uber die Entropieverminderung in einem thermodynamischen System bei Eingriffen intelligenter Wesen”. Zeitschrift f¨ ur Physik53(11): 840–856 (1929)

    1929

  25. [33]

    Finite-Time Quantum Lan- dauer Principle and Quantum Coherence

    V. V. Tan and K. Saito.“Finite-Time Quantum Lan- dauer Principle and Quantum Coherence”. Phys. Rev. Lett.128: 010602 (2022)

    2022

  26. [34]

    Landauer versus Nernst: What is the true cost of cooling a quantum system?

    P. Taranto, F. Bakhshinezhad, A. Bluhm, R. Silva, N. Friis, M. P. Lock, G. Vitagliano, F. C. Binder, T. De- barba, E. Schwarzhans, et al.“Landauer versus Nernst: What is the true cost of cooling a quantum system?”. PRX Quantum4(1): 010332, (2023)

    2023

  27. [35]

    Finite-Time Quantum Lan- dauer Principle and Quantum Coherence

    T. Van Vu and K. Saito.“Finite-Time Quantum Lan- dauer Principle and Quantum Coherence”. Phys. Rev. Lett.128: 010602 (2022)

    2022

  28. [36]

    Black box work extrac- tion and composite hypothesis testing

    K. Watanabe and R. Takagi.“Black box work extrac- tion and composite hypothesis testing”. Physical Review Letters133(25): 250401 (2024)

    2024

  29. [37]

    Universal work ex- traction in quantum thermodynamics

    K. Watanabe and R. Takagi.“Universal work ex- traction in quantum thermodynamics”. arXiv preprint arXiv:2504.12373 , (2025)

    2025

  30. [38]

    Autonomous Quan- tum Devices: When Are They Realizable without Addi- tional Thermodynamic Costs?

    M. P. Woods and M. Horodecki.“Autonomous Quan- tum Devices: When Are They Realizable without Addi- tional Thermodynamic Costs?”. Phys. Rev. X13: 011016 (2023). Appendix A: Learning pure quantum states (almost) without regret In this Section we review the quantum state tomograp...

    2023

  31. [39]

    (A11) guarantees that ˆσ2 k∗ is a good upper bound for the variance of the outcomes rk∗,i,j that are sampled after performing a measurement on the directions given bya k∗,i

    The particular form ˆσ2 k∗ given by Eq. (A11) guarantees that ˆσ2 k∗ is a good upper bound for the variance of the outcomes rk∗,i,j that are sampled after performing a measurement on the directions given bya k∗,i. With this choice as argued in [18] it allows to obtain rigorous...

  32. [40]

    Apply the swap unitaryV ρ∗,τ =P ij|i⟩⟨j|A⊗|j⟩⟨i|R⊗Γ (i−j)ν(τ,ϵ) on the system, the battery and the reservoir qubit

    from the reservoir. Apply the swap unitaryV ρ∗,τ =P ij|i⟩⟨j|A⊗|j⟩⟨i|R⊗Γ (i−j)ν(τ,ϵ) on the system, the battery and the reservoir qubit. Discard the reservoir qubit. Measure the extracted work Measure the battery energy, obtain the battery energyµ ′ and compute the extracted wo...

  33. [41]

    Then there existsc∈(a,b)such that f(b)−f(a) =f ′(c)(b−a).(B10) Theorem 3.Letf: [a,b]→Rbe a continuous function on the closed interval[a,b]

    Work distribution In this section, we will use the Lagrange mean value theorem and the first mean value theorem for definite inte- grals [26] as follows: Theorem 2.Letf: [a,b]→Rbe a continuous on the closed interval[a,b]and differentiable on the open interval (a,b). Then there...

  34. [42]

    Extracted work for different inputs Theorem 5.Let{ϕ i}i=0,1 and{p i}i=0,1 be the eigenvectors and eigenvalues ofρ ∗ andw i be the value of work extracted defined in Theorem. 4. It holds that,

  35. [43]

    When applying the protocol to any stateρ, the probability of measuring∆W=w i is given by Pr(∆W=w i) =⟨ϕ i|ρ|ϕ i⟩.(B41)

  36. [44]

    Proof.We first observe that the off-diagonal term⟨ϕ i|ρ|ϕ j⟩|i⟩⟨j|A⊗φ(x)B of the join state in Eq

    When the extraction protocol is operated on a stateρwhereρ̸=ρ ∗, the expected work extracted is given by E[∆W] =β −1 [D(ρ∥1/2)−D(ρ∥ρ ∗)],(B42) where the second term can be defined as the the dissipation due to the agent’s imperfect knowledge ofρ. Proof.We first observe that th...

  37. [45]

    Cumulative dissipation In this section, we consider the setting where we have oracle sequential access to an unknown pure qubit stateψ, and our goal is to extract the maximal amount of work into a battery system. To achieve this, we can use Algorithm 1 with the rewards (B39) t...

  38. [46]

    Then we have D(ρ∥∆2ϵ(ˆρ))≤16ϵ 2 + ln d 2ϵ .(B54) Given the above bound we can use the fidelity guarantee of Algorithm 1 in Theorem 1 to prove a bound on the cumulative dissipation. Theorem 7.Given a finite time horizonN∈Nandδ∈(0,1)there exists an explicit sequence of accuracie...

  39. [47]

    In this case, there is a pre-defined energy eigenbasis, hence all the swapping operations will have to be done in such basis

    Non-degenerate Hamiltonian So far we have discussed the case for when the system Hamiltonian is degenerate, in general though we can consider Hamiltonian’s with energy gap ofω, i.e.,H A =ω|E 1⟩⟨E1|. In this case, there is a pre-defined energy eigenbasis, hence all the swapping...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.