pith. sign in

arxiv: 2603.29804 · v3 · submitted 2026-03-31 · ❄️ cond-mat.stat-mech · cond-mat.soft

Optimal Control of a Mesoscopic Information Engine

Emanuele Panizon This is my paper

Pith reviewed 2026-05-13 23:25 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech cond-mat.soft
keywords information engineoptimal controlLQG dynamicsPOMDPoptical trapmesoscopic systemsfeedback protocolthermal fluctuations
0
0 comments X

The pith

LQG dynamics decouple optimal measurement scheduling from trap-position feedback in a mesoscopic information engine.

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

The paper solves the finite-time optimal control problem for an overdamped particle in an optical trap when each measurement carries a fixed cost. It formulates the task inside a POMDP and shows that the linear-quadratic-Gaussian structure lets the designer optimize when and how precisely to measure independently of how to move the trap. A reader would care because the separation produces explicit feedback laws, recovers the known open-loop protocol, and supplies concrete bounds on the net work that can be extracted from thermal noise before viscous drag dominates.

Core claim

By casting the engine inside the POMDP framework, the underlying LQG dynamics decouple the optimal measurement and driving protocols. The optimal feedback law for trap placement recovers the discontinuous Schmiedl-Seifert protocol in the open-loop limit and extends it to arbitrary measurement schedules. For a binary on/off sensor the paper derives the optimal measurement times, identifies deadline-induced blindness, and obtains exact periodic schedules in the infinite-horizon limit together with the macroscopic velocity envelopes beyond which the engine becomes net dissipative.

What carries the argument

Linear-Quadratic-Gaussian (LQG) dynamics inside the Partially Observable Markov Decision Process (POMDP) formulation, which separate the choice of measurement instants from the continuous feedback law that sets trap position.

If this is right

  • The optimal driving protocol reduces exactly to the discontinuous Schmiedl-Seifert protocol when measurements are absent.
  • Optimal finite-horizon measurement schedules exist for any fixed measurement cost and any deadline.
  • Deadline-induced blindness appears: all measurements cease as the deadline is approached regardless of cost.
  • Exact periodic measurement schedules are obtained in the infinite-horizon steady state as a function of cost.
  • Macroscopic velocity envelopes mark the boundary between net work extraction and net dissipation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same separation may let designers add variable-precision sensors without re-optimizing the entire feedback law.
  • Bounds on net gain supply a practical design rule for choosing trap stiffness and measurement rate in viscous environments.
  • The decoupling suggests that similar LQG information engines could be built with non-binary sensors without losing analytic tractability.
  • Testing the predicted blindness near deadlines in a real optical-trap setup would directly probe the finite-time theory.

Load-bearing premise

The particle obeys linear-quadratic-Gaussian dynamics under the overdamped approximation and measurements are binary with a fixed per-use cost.

What would settle it

Numerical simulation or experiment on an overdamped particle showing that the optimal measurement schedule changes when the feedback gain for trap position is altered would falsify the claimed decoupling.

Figures

Figures reproduced from arXiv: 2603.29804 by Emanuele Panizon.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: The Variable-Precision Sensor. Generalizing the binary sensor to a continuous variable-precision mech￾anism, we assume the thermodynamic observation cost depends on its precision: C(Σ+, Σ −) = c(1/Σ + − 1/Σ −). The reduction of variance can be quantified by the Kalman gain[44] Lk = Σ −−Σ + Σ− . Again, the optimal trap placement is the same as before, as it is fully independent from the measurement protocol… view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
read the original abstract

We analytically solve the finite-time control problem of driving an overdamped particle via an optical trap under costly measurement. By formulating this mesoscopic information engine within the Partially Observable Markov Decision Process (POMDP) framework, we demonstrate that the underlying Linear-Quadratic-Gaussian (LQG) dynamics decouple the optimal measurement and driving protocols. We derive the optimal feedback control law for the trap placement, which recovers the discontinuous Schmiedl-Seifert driving protocol in the open-loop limit and extends it to any measurement scheduling. For a costly, binary (on/off) sensor, we evaluate the optimal measurement protocol and derive physical bounds on the maximum net gain that can be extracted from thermal fluctuations. We show the emergence of deadline-induced blindness, a phenomenon where all measurements cease as the deadline approaches regardless of their cost. Taking the infinite-horizon limit, we find the exact periodic measurement schedules for the steady state as a function of the measurement cost $C$ and derive the macroscopic velocity envelopes beyond which viscous drag forces the engine into a net-dissipative regime. Finally, we generalize the results to a variable-precision sensor.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper analytically solves the finite-time optimal control problem for an overdamped colloidal particle driven by an optical trap under costly binary measurements. Formulating the system as a POMDP under LQG dynamics, it claims that the optimal measurement and driving protocols decouple, derives the feedback law for trap placement (recovering the discontinuous Schmiedl-Seifert protocol in the open-loop limit), evaluates optimal measurement schedules with net-gain bounds, identifies deadline-induced blindness, obtains exact periodic schedules in the infinite-horizon limit as a function of measurement cost C, derives velocity envelopes for net-dissipative regimes, and generalizes to variable-precision sensors.

Significance. If the decoupling holds rigorously, the work offers a valuable analytical framework for optimal operation of mesoscopic information engines, extending known protocols with exact results and physical bounds that could inform experiments in stochastic thermodynamics. The recovery of the Schmiedl-Seifert limit and derivation of velocity envelopes add concrete physical insight.

major comments (2)
  1. [Section 3 (derivation of optimal feedback control law)] The central decoupling claim (that LQG dynamics separate optimal measurement and driving protocols for discrete on/off measurements with fixed cost C) is load-bearing but requires explicit verification: the dynamic-programming recursion for the value function must be shown to factorize independently of the measurement schedule, as the belief-state update with additive cost typically couples mean/variance evolution to the horizon in a non-separable manner (standard LQG separation applies to continuous observations).
  2. [Section 6 (infinite-horizon limit)] The macroscopic velocity envelopes (beyond which viscous drag renders the engine net-dissipative) rest on the overdamped LQG approximation; their quantitative range of validity should be bounded against the point where inertial or nonlinear effects invalidate the linear dynamics assumption.
minor comments (3)
  1. [Abstract] The term 'deadline-induced blindness' is introduced in the abstract without definition; a one-sentence clarification on first use would improve accessibility.
  2. [Section 2] Belief-state notation (mean and variance) should be introduced with explicit symbols in the methods section before its repeated use in the POMDP formulation.
  3. A small number of equation cross-references in the text appear inconsistent with the numbering in the displayed equations; these should be checked for accuracy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. We address the two major comments point by point below, providing clarifications and indicating the revisions we will implement.

read point-by-point responses

  1. Referee: [Section 3 (derivation of optimal feedback control law)] The central decoupling claim (that LQG dynamics separate optimal measurement and driving protocols for discrete on/off measurements with fixed cost C) is load-bearing but requires explicit verification: the dynamic-programming recursion for the value function must be shown to factorize independently of the measurement schedule, as the belief-state update with additive cost typically couples mean/variance evolution to the horizon in a non-separable manner (standard LQG separation applies to continuous observations).

    Authors: We appreciate the referee's emphasis on making the separation explicit. In the LQG-POMDP setting with quadratic control costs and additive measurement penalties independent of state, the Bellman recursion factors such that the optimal feedback law for trap position depends only on the current belief mean (recovering the Schmiedl-Seifert form in the open-loop case) while the measurement schedule is optimized separately via the expected reduction in variance cost. The variance evolution enters only through an additive term in the value function that does not affect the argmin over control. To address the request for verification, we will insert a short appendix deriving the factorization of the dynamic-programming equation for the binary-measurement case, confirming independence from the horizon-dependent measurement policy. revision: yes

  2. Referee: [Section 6 (infinite-horizon limit)] The macroscopic velocity envelopes (beyond which viscous drag renders the engine net-dissipative) rest on the overdamped LQG approximation; their quantitative range of validity should be bounded against the point where inertial or nonlinear effects invalidate the linear dynamics assumption.

    Authors: We agree that bounding the regime of validity strengthens the physical interpretation. The envelopes are derived under the standard overdamped colloidal-particle model used throughout the literature on stochastic thermodynamics. In the revision we will add a paragraph in Section 6 that estimates the inertial relaxation time (m/γ) relative to the control and measurement intervals for representative experimental parameters (micron-scale beads in aqueous solution), showing that the envelopes lie well within the overdamped regime for typical trap stiffnesses and viscosities. We will also note the velocity scale at which trap anharmonicity or inertia would require extension beyond the linear model. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivations follow from standard LQG separation under explicit model assumptions.

full rationale

The paper formulates the mesoscopic engine as a POMDP with linear dynamics and quadratic costs, then applies the LQG separation principle to decouple measurement and control. This yields analytic optimal feedback laws that recover the known Schmiedl-Seifert protocol in the open-loop limit. No equation reduces a claimed prediction to a fitted parameter by construction, no uniqueness theorem is imported from self-citation, and no ansatz is smuggled via prior work. The binary measurement cost and deadline effects are derived directly from the value function under the stated overdamped approximation; the results remain falsifiable against the model equations without circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central results rest on the overdamped limit, LQG dynamics, and a fixed-cost binary sensor; measurement cost C is the only explicit free parameter introduced.

free parameters (1)
  • measurement cost C
    Scalar cost per binary measurement used to optimize schedules and compute net-gain bounds.
axioms (2)
  • domain assumption Overdamped Langevin dynamics for the particle
    Standard modeling choice for mesoscopic colloidal systems in viscous fluid.
  • domain assumption Linear-Quadratic-Gaussian control structure
    Enables exact decoupling of measurement and driving policies.

pith-pipeline@v0.9.0 · 5488 in / 1391 out tokens · 38876 ms · 2026-05-13T23:25:20.377350+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages

  1. [1]

    Jarzynski, Physical review letters78, 2690 (1997)

    C. Jarzynski, Physical review letters78, 2690 (1997)

    work page 1997

  2. [2]

    G. E. Crooks, Physical Review E60, 2721 (1999)

    work page 1999

  3. [3]

    Seifert, Reports on progress in physics75, 126001 (2012)

    U. Seifert, Reports on progress in physics75, 126001 (2012)

    work page 2012

  4. [4]

    Sagawa and M

    T. Sagawa and M. Ueda, Physical review letters104, 090602 (2010)

    work page 2010

  5. [5]

    Bauer, D

    M. Bauer, D. Abreu, and U. Seifert, Journal of Physics A: Mathematical and Theoretical45, 162001 (2012)

    work page 2012

  6. [6]

    Sagawa and M

    T. Sagawa and M. Ueda, Phys. Rev. E85, 021104 (2012)

    work page 2012

  7. [7]

    Ehrich and D

    J. Ehrich and D. A. Sivak, Frontiers in Physics11, 1108357 (2023)

    work page 2023

  8. [8]

    J. C. Maxwell,Theory of Heat(Longmans, Green, and Company, 1871)

    work page

  9. [9]

    Szilard, Zeitschrift f¨ ur Physik53, 840 (1929)

    L. Szilard, Zeitschrift f¨ ur Physik53, 840 (1929)

    work page 1929

  10. [10]

    Landauer, IBM journal of research and development 5, 183 (1961)

    R. Landauer, IBM journal of research and development 5, 183 (1961)

    work page 1961

  11. [11]

    J. M. Parrondo, J. M. Horowitz, and T. Sagawa, Nature physics11, 131 (2015)

    work page 2015

  12. [12]

    Toyabe, T

    S. Toyabe, T. Sagawa, M. Ueda, E. Muneyuki, and M. Sano, Nature physics6, 988 (2010)

    work page 2010

  13. [13]

    B´ erut, A

    A. B´ erut, A. Arakelyan, A. Petrosyan, S. Ciliberto, R. Dillenschneider, and E. Lutz, Nature483, 187 (2012)

    work page 2012

  14. [14]

    I. A. Mart´ ınez,´E. Rold´ an, L. Dinis, and R. A. Rica, Soft matter13, 22 (2017)

    work page 2017

  15. [15]

    Paneru and H

    G. Paneru and H. Kyu Pak, Advances in Physics: X5, 1823880 (2020)

    work page 2020

  16. [16]

    T. K. Saha, J. N. Lucero, J. Ehrich, D. A. Sivak, and J. Bechhoefer, Proceedings of the National Academy of Sciences118, e2023356118 (2021)

    work page 2021

  17. [17]

    T. K. Saha, J. Ehrich, M. Gavrilov, S. Still, D. A. Sivak, and J. Bechhoefer, Physical Review Letters131, 057101 (2023)

    work page 2023

  18. [18]

    Bensoussan, Cambridge, Cambridge (1992)

    A. Bensoussan, Cambridge, Cambridge (1992)

    work page 1992

  19. [19]

    Alvarado, E

    J. Alvarado, E. G. Teich, D. A. Sivak, and J. Bechhoefer, Annual Review of Condensed Matter Physics17(2026)

    work page 2026

  20. [20]

    Aurell, C

    E. Aurell, C. Mej´ ıa-Monasterio, and P. Muratore- Ginanneschi, Physical review letters106, 250601 (2011)

    work page 2011

  21. [21]

    Kamijima, A

    T. Kamijima, A. Takatsu, K. Funo, and T. Sagawa, Phys- ical Review Research7, 023159 (2025)

    work page 2025

  22. [22]

    A. T. Mohite and H. Rieger, arXiv preprint arXiv:2511.00974 (2025)

    work page arXiv 2025

  23. [23]

    Blaber, M

    S. Blaber, M. D. Louwerse, and D. A. Sivak, Physical Review E104, L022101 (2021)

    work page 2021

  24. [24]

    Schmiedl and U

    T. Schmiedl and U. Seifert, Physical review letters98, 108301 (2007)

    work page 2007

  25. [25]

    Abreu and U

    D. Abreu and U. Seifert, EPL (Europhysics Letters)94, 8 10001 (2011)

    work page 2011

  26. [26]

    Garcia-Millan, J

    R. Garcia-Millan, J. Sch¨ uttler, M. E. Cates, and S. A. Loos, Physical Review Letters135, 088301 (2025)

    work page 2025

  27. [27]

    Sch¨ uttler, R

    J. Sch¨ uttler, R. Garcia-Millan, M. E. Cates, and S. A. Loos, Physical Review E112, 024119 (2025)

    work page 2025

  28. [28]

    Whitelam, Physical Review X13, 021005 (2023)

    S. Whitelam, Physical Review X13, 021005 (2023)

    work page 2023

  29. [29]

    Bechhoefer, New Journal of Physics17, 075003 (2015)

    J. Bechhoefer, New Journal of Physics17, 075003 (2015)

    work page 2015

  30. [30]

    Biehl and N

    M. Biehl and N. Virgo, inInternational Workshop on Active Inference(Springer, 2022) pp. 16–31

    work page 2022

  31. [31]

    Meier, J

    L. Meier, J. Peschon, and R. Dressler, IEEE Transactions on Automatic Control12, 528 (1967)

    work page 1967

  32. [32]

    B. D. Anderson and J. B. Moore,Optimal control: linear quadratic methods(Courier Corporation, 2007)

    work page 2007

  33. [33]

    Bechhoefer,Control theory for physicists(Cambridge University Press, 2021)

    J. Bechhoefer,Control theory for physicists(Cambridge University Press, 2021)

    work page 2021

  34. [34]

    K. J. ˚Astr¨ om, Journal of mathematical analysis and ap- plications10, 174 (1965)

    work page 1965

  35. [35]

    L. P. Kaelbling, M. L. Littman, and A. R. Cassandra, Artificial intelligence101, 99 (1998)

    work page 1998

  36. [36]

    G. E. Uhlenbeck and L. S. Ornstein, Physical Review36, 823 (1930)

    work page 1930

  37. [37]

    D. T. Gillespie, Physical review E54, 2084 (1996)

    work page 2084

  38. [38]

    Bellman,Dynamic Programming(Princeton Univer- sity Press, Princeton, NJ, 1957)

    R. Bellman,Dynamic Programming(Princeton Univer- sity Press, Princeton, NJ, 1957)

    work page 1957

  39. [39]

    Bertsekas,Dynamic programming and optimal control: Volume I, Vol

    D. Bertsekas,Dynamic programming and optimal control: Volume I, Vol. 4 (Athena scientific, 2012)

    work page 2012

  40. [40]

    H. A. Simon, Econometrica, Journal of the Econometric Society24, 74 (1956)

    work page 1956

  41. [41]

    Bar-Shalom and E

    Y. Bar-Shalom and E. Tse, IEEE Transactions on Auto- matic Control19, 494 (1974)

    work page 1974

  42. [42]

    Sagawa and M

    T. Sagawa and M. Ueda, Physical review letters102, 250602 (2009)

    work page 2009

  43. [43]

    R. M. Corless, G. H. Gonnet, D. E. Hare, D. J. Jeffrey, and D. E. Knuth, Advances in Computational mathe- matics5, 329 (1996)

    work page 1996

  44. [44]

    Kalman, Journal of Basic Engineering82, 35 (1960)

    R. Kalman, Journal of Basic Engineering82, 35 (1960)

    work page 1960

  45. [45]

    L. L. Bonilla, Physical Review E100, 10.1103/phys- reve.100.022601 (2019)

    work page doi:10.1103/phys- 2019

  46. [46]

    L. K. Davis, arXiv preprint arXiv:2603.16778 (2026)

    work page arXiv 2026

  47. [47]

    Cocconi, J

    L. Cocconi, J. Knight, and C. Roberts, Physical Review Letters131, 188301 (2023)

    work page 2023

  48. [48]

    Casert and S

    C. Casert and S. Whitelam, Nature Communications15, 9128 (2024). A. Decoupling and Dimensional Reduction of the Cost Function To distinguish the state transitions within a sin- gle step, we denote the prior state before measure- ment as (µ−,Σ −), the posterior state after measurement as (µ +,Σ +), and the prior state at the next step as (µ′−,Σ ′−). In the...

    work page 2024

  49. [49]

    Since the trap stiff- nessκis constant, theE[x 2] terms cancel exactly: Ex[W] =−κµ +λ′ + 1 2 κ(λ′)2 +κµ +λ− 1 2 κλ2

    The expected thermodynamic work of instanta- neously translating a harmonic trap is indepen- dent of the spatial variance. Since the trap stiff- nessκis constant, theE[x 2] terms cancel exactly: Ex[W] =−κµ +λ′ + 1 2 κ(λ′)2 +κµ +λ− 1 2 κλ2

    work page

  50. [50]

    The transition of the prior mean is deterministic and invariant to the measurement uncertainty Σ +: µ′− =αµ + + (1−α)λ ′

    work page

  51. [51]

    Since Σ′− is independent ofλ ′, the future informational costg n−1(Σ′−) acts as a constant with respect to the in- ner minimization overλ ′ and factors out

    The information dynamics Σ ′− =α 2Σ+ + Σeq are deterministic and independent of the physical trap placementλ ′. Since Σ′− is independent ofλ ′, the future informational costg n−1(Σ′−) acts as a constant with respect to the in- ner minimization overλ ′ and factors out. We define the 9 remaining inner minimization as the intermediate physi- cal cost ˜Jn(µ+,...

    work page