Pith. sign in

REVIEW 2 major objections 5 minor 34 references

Improving noisy free-energy measurements by adding more noise

T0 review · 2 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Adding carefully chosen extra noise while rescaling the potential energy leaves equilibrium thermodynamics unchanged and makes Jarzynski free-energy estimates far more precise.

desk verdict A clean, exact noise-injection trick for overdamped Langevin systems that measurably sharpens Jarzynski estimates in two models, held back by one data-entry error and an overbroad causal claim. read the letter →

arxiv 2502.03734 v3 pith:73KLB3AS submitted 2025-02-06 cond-mat.stat-mech

Stephen Whitelam This is my paper · ORCID
classification cond-mat.stat-mech
keywords Jarzynskiequalityfree-energyestimationnoiseinjectionLangevindynamicsworkfluctuationspotentialrescalinginformationerasurenonequilibriumthermodynamics
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

Estimating free-energy differences from nonequilibrium work measurements is hard when work fluctuations are large, because the Jarzynski exponential average is dominated by rare trajectories. This paper proposes the opposite of the usual remedy: instead of slowing the protocol or redesigning it, inject additional white noise into the dynamics and multiply the potential energy by a common factor $\lambda>1$. The two changes cancel thermodynamically, so the equilibrium free-energy difference is untouched, but the system relaxes $\lambda$ times faster—equivalently, it feels the driving protocol as if it had been run $\lambda$ times more slowly. In two model systems, a translated optical trap and a double-well information-erasure process, the modified dynamics lowers dissipated reduced work and cuts by orders of magnitude the number of trajectories needed to reach a target precision. The caveat is that the potential must be rescaled uniformly, which the paper argues is feasible mainly for optical traps and thermodynamic computers.

What carries the argument

The central object is the dual transformation of Eq. (11): multiply the energy function $U$ by $\lambda>1$ and add zero-mean Gaussian white noise of variance $\sigma^2 = 2k_{\rm B}T\mu(\lambda-1)$. Its load-bearing identity is that the combined noise has variance $2k_{\rm B}T\mu\lambda$, so the modified dynamics is exactly the original Langevin dynamics with $U_\lambda=\lambda U$ and temperature $T_\lambda=\lambda T$; after the time rescaling $t\to t/\lambda$ it becomes the original system driven by the slowed protocol $c(t/\lambda)$. This equivalence converts added noise into faster relaxation without changing the equilibrium distribution, which is what reduces dissipated reduced work and rare-event sampling cost.

What would settle it

Measure the reduced work $\beta_\lambda\langle W_\lambda\rangle_\lambda$ in the trap-translation experiment at fixed $t_f=1$ for several $\lambda$; the paper predicts $\beta_\lambda\langle W_\lambda\rangle_\lambda = w(\lambda) = c_f^2\lambda^{-2}(\lambda+e^{-\lambda}-1)$ from Eq. (19). If measured values deviate from this curve, or if the estimator $J_\lambda$ does not converge to $\Delta F=0$, the dual transformation is not exact in practice.

Watch

Extended reading notes

Core claim

The central claim is that for overdamped Langevin dynamics the replacement $U \to \lambda U$ together with added Gaussian white noise of variance $2k_{\rm B}T\mu(\lambda-1)$ yields a dual dynamics with identical equilibrium thermodynamics. The modified equation of motion is the original Langevin equation with potential $U_\lambda=\lambda U$ and effective temperature $T_\lambda=\lambda T$; rescaling time $t \to t/\lambda$ turns it into the original dynamics driven by the protocol $c(t/\lambda)$, applied $\lambda$ times more slowly for $\lambda$ times longer. As a result the Jarzynski equality still reads $\langle e^{-\beta_\lambda W_\lambda}\rangle_\lambda = e^{-\beta \Delta F}$, while the dissipated reduced work $\beta_\lambda\langle W_\lambda\rangle_\lambda - \beta\Delta F$ decreases with increasing $\lambda$, improving the convergence of the estimator $J_\lambda$. In the trap-translation model at fixed protocol time $t_f=1$, the estimator improves from $J=0.24\,k_{\rm B}T$ at $\lambda=1$ to $J\approx -0.00093\,k_{\rm B}T$ at $\lambda=30$; in the erasure model $J_\lambda$ approaches $\Delta F = k_{\rm B}T\ln 2$ as $\lambda$ grows.

Load-bearing premise

The load-bearing premise is that the experimenter can implement the dual modification exactly—multiply the entire potential by $\lambda$ and inject white noise of precisely the right variance—because any mismatch changes the effective temperature and breaks the Jarzynski relation used to extract the free energy.

Editorial extensions

If this is right

  • For any overdamped Langevin system whose potential can be uniformly rescaled, a fixed-duration driving protocol becomes effectively slower as $\lambda$ grows, so the Jarzynski estimator improves without lengthening the experiment.
  • In the Gaussian-work case the number of trajectories needed to estimate $\Delta F$ to a given precision decreases exponentially with $\lambda$, because estimator variance tracks dissipated reduced work.
  • In the information-erasure model, raising $\lambda$ increases erasure success to unity for $\lambda \gtrsim 3$ and drives the measured free-energy cost toward the Landauer bound $k_{\rm B}T\ln 2$.
  • For the trap-translation model the measured reduced work obeys $\beta_\lambda\langle W_\lambda\rangle_\lambda = w(\lambda)$, matching the mean work of the original dynamics run $\lambda$ times more slowly.
  • The method extends the notion of optimal control to include noise strength as a controllable resource, alongside the potential protocol.

Reading between the lines

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

  • Editorial inference: the identical dual recipe could be applied inside a numerical simulation—rescaling the Hamiltonian and adding auxiliary noise is trivial in software—so the same precision gain may carry over to path-sampling and free-energy estimators beyond the experimental setups the paper considers.
  • Editorial inference: the paper notes the added noise costs energy scaling as $\sigma^2$; a fair practical comparison of noise engineering against simply running the experiment longer should include this energetic overhead, which the paper does not optimize.
  • Editorial inference: for underdamped systems, the extra requirement of rescaling the damping coefficient might be approximated with feedback-based effective friction, potentially recovering some of the precision gain in setups where uniform potential rescaling is unavailable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes a noise-injection strategy for overdamped Langevin systems: scale the potential by a factor λ > 1 and add Gaussian white noise of variance 2 k_B T μ (λ − 1). The modified dynamics is equivalent to the original dynamics with the same protocol applied λ times more slowly for λ times longer (Eqs. (11)–(16)), so the equilibrium free-energy difference is unchanged while the reduced dissipated work decreases. The authors argue that this reduction in dissipation improves convergence of the Jarzynski estimator, and they demonstrate the effect numerically for a trap-translation model and an information-erasure model. The paper also discusses the practical limitations of the method, noting that uniform potential rescaling is feasible mainly for optical traps and thermodynamic computers.

Significance. The exact time-rescaling duality is elegant and cleanly derived, and the numerical confirmation that β_λ⟨W_λ⟩_λ = w(λ) (Eq. (22)) is a strong, parameter-free check. If the improved-convergence claim holds generally, the method could be practically useful in experimental settings where external noise and potential scaling are available. However, the manuscript's own Appendix A limits the rigorous justification of the convergence claim to Gaussian work statistics, and the reported N0.1 values contain a numerical inconsistency that affects the quantitative comparison. The paper is honest about the narrow regime of applicability, which is a credit, but the central claim is currently broader than its proof.

major comments (2)
  1. [Section III A, Eq. (10), footnote [27]] The reported values N0.1 = 1.5 × 10^7 (optimal protocol) and N0.1 = 0.4 × 10^7 (linear protocol) are inconsistent with Eq. (10) and the Gaussian relation (21). For the optimal protocol, Eq. (10) with Δf = 0.1 gives N0.1 = 100(e^{2β⟨W⟩} − 1) ≈ 1.7 × 10^9, exactly as stated in footnote [27]; for the linear protocol with β⟨W⟩ ≈ 9.20, the same formula gives N0.1 ≈ 9.8 × 10^9. The text values appear to omit the factor 1/(Δf)^2 = 100. Because N0.1 is used in Fig. 2(e) and in the quantitative comparison with the λ-modified dynamics, this error should be corrected and the affected numbers and figure replotted.
  2. [Section II, Appendix A] The central inference that reduced dissipated work 'leads to more accurate free-energy estimates' is established rigorously only for Gaussian work statistics. The exact time-rescaling identity (16) proves that the λ-modified dynamics reduces β_λ⟨W_λ⟩_λ for a given protocol, but Appendix A explicitly concedes that for non-Gaussian work distributions the minimum-mean-work protocol is not necessarily the protocol that gives the best convergence of Eq. (3). The two numerical demonstrations (trap translation, whose work distribution is Gaussian, and the erasure model, which is non-Gaussian) support the claim but do not prove the general statement made in the abstract. I recommend either softening the abstract and Section II to present the improved-convergence claim as a heuristic supported by numerical evidence, or adding an additional argument or broader numerical tests that directly characterize Var(e^{−β_λ W_λ}) as a function of λ for non-Gaussian cases.
minor comments (5)
  1. [Section II, Eq. (11)] The sentence 'The two noise terms in (12) can be considered...' should refer to Eq. (11), not Eq. (12).
  2. [Fig. 2 caption] The caption repeats panel label '(a)' for the first two panels; the second panel should be labeled '(b)'.
  3. [Section III A] The paragraph before Fig. 3 says 'In Fig. 1(f) we show...' but the referenced panel is Fig. 2(f).
  4. [Fig. 4 caption] The caption says 'we show the potential (17)' for the erasure model; the potential is defined in Eq. (24), not Eq. (17).
  5. [Section III B] The text contains the typo 'Jarzynksi' in 'the Jarzynksi free-energy estimator'; it should be 'Jarzynski'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central identity is an exact rescaling of the overdamped Langevin equation, benchmarked against analytic work formulas and known free-energy values, with no fitted parameters.

full rationale

The paper's derivation chain is self-contained and non-circular. Equations (11)-(12) define the modified dynamics by scaling U to λU and adding white noise of variance 2k_BT μ(λ−1); Eq. (16) follows by the exact change of variables t→t/λ, showing that the λ-modified dynamics is equivalent to the original dynamics with the same protocol run λ times more slowly. Equation (14) is the standard Jarzynski equality applied to this rescaled dynamics, and the identity βλWλ = βW makes the estimator (15) an unbiased estimator of βΔF, not a fit. The numerical demonstrations are checked against analytic expressions (for example, Eq. (22) reproduces the known work formula w(λ) of Eq. (19)) and against the exactly known free-energy changes ΔF=0 for the trap and βΔF=ln2 for erasure. No parameter is fitted to the target quantity, and no uniqueness or optimality claim is imported from the author's prior work. The only self-citation, Ref. [28], appears in peripheral remarks about the energy cost of added noise and about enacting a protocol faster, and it is not load-bearing for the main result. The paper itself flags the one genuinely heuristic step—that reduced mean dissipated work guarantees better convergence of the exponential average only for Gaussian work statistics (Appendix A)—so that limitation is disclosed rather than concealed. This is a correctness or generality caveat, not circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No fitted parameters; lambda is a user-chosen control parameter. The derivation uses standard Langevin and work relations. No new entities are introduced.

assumptions (4)
  • domain assumption The system evolves by overdamped Langevin dynamics with Gaussian white noise satisfying detailed balance (Eq. 1).
    The entire method is derived for this dynamics class; underdamped or non-white noise cases require additional control and are not covered.
  • standard math The Jarzynski equality and Crooks fluctuation theorem hold for the driven dynamics (Eqs. 2-3).
    These are established nonequilibrium work relations, used as external benchmark for the estimator.
  • standard math For Gaussian work fluctuations, beta sigma^2 = 2(<W> - Delta F) (Appendix A, Eq. A2).
    Used to relate work variance to mean dissipation and to justify the convergence heuristic.
  • domain assumption Slower driving (longer duration) reduces dissipated work and improves convergence of the exponential average (refs [7,25]).
    The claimed improvement relies on this scaling; the paper verifies it numerically for the two models but does not prove it generally.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving noisy free-energy measurements by adding more noise." pith.science (2026). https://pith.science/paper/73KLB3AS

@misc{pith2026250203734,
  author       = {Pith},
  title        = {Pith review of: Improving noisy free-energy measurements by adding more noise},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/73KLB3AS}},
  note         = {Machine review of arXiv:2502.03734}
}
read the original abstract

Estimating free-energy differences using nonequilibrium work relations, such as the Jarzynski equality, is hindered by poor convergence when work fluctuations are large. For systems governed by overdamped Langevin dynamics, we propose the counterintuitive approach of adding noise in order to increase the precision of such calculations. By introducing additional stochastic fluctuations to the system and rescaling its potential energy, we leave the thermodynamics of the system unchanged while increasing its relaxation rate. For a given time-dependent protocol this modification reduces dissipated work, leading to more accurate free-energy estimates. We demonstrate this principle using computer simulations applied to two model systems. However, the regime of applicability of this strategy is likely limited, because it requires control of the system's potential energy in a way that is feasible in only a few experimental settings.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 24 canonical work pages

  1. [27]

    Nu- merical calculation of the exponential average is highly imprecise for the pulling rate considered

    The work distribution for the optimal protocol is Gaus- sian, and so N0.1 = 100(e 2β⟨W ⟩ − 1) ≈ 1.7 × 109. Nu- merical calculation of the exponential average is highly imprecise for the pulling rate considered

    work page

  2. [1]

    Jarzynski, Physical Review Letters 78, 2690 (1997)

    C. Jarzynski, Physical Review Letters 78, 2690 (1997)

    1997

  3. [2]

    G. E. Crooks, Physical Review E 60, 2721 (1999)

    1999

  4. [3]

    Jarzynski, Physical Review E 73, 046105 (2006)

    C. Jarzynski, Physical Review E 73, 046105 (2006)

    work page 2006

  5. [4]

    Hummer and A

    G. Hummer and A. Szabo, Proceedings of the National Academy of Sciences 107, 21441 (2010)

    work page 2010

  6. [5]

    C. M. Rohwer, F. Angeletti, and H. Touchette, Physical Review E 92, 052104 (2015)

    work page 2015

  7. [6]

    The adjective reduced refers to the multiplication of a quantity by β

    work page

  8. [7]

    Schmiedl and U

    T. Schmiedl and U. Seifert, Physical Review Letters 98, 108301 (2007)

    work page 2007

Show all 34 references
  1. [8]

    Geiger and C

    P. Geiger and C. Dellago, Physical Review E 81, 021127 (2010)

    2010

  2. [9]

    Vaikuntanathan and C

    S. Vaikuntanathan and C. Jarzynski, Physical Review Letters 100, 190601 (2008)

    2008

  3. [10]

    Blaber and D

    S. Blaber and D. A. Sivak, The Journal of Chemical Physics 153 (2020)

    2020

  4. [11]

    Arrar, F

    M. Arrar, F. M. Boubeta, M. E. Szretter, M. Sued, L. Boechi, and D. Rodriguez, Journal of Computational Chemistry 40, 688 (2019)

    2019

  5. [12]

    S. A. Loos, S. Monter, F. Ginot, and C. Bechinger, Phys- ical Review X 14, 021032 (2024)

    2024

  6. [13]

    Zhong and M

    A. Zhong and M. R. DeWeese, Physical Review E 106, 044135 (2022)

    2022

  7. [14]

    Alvarado, E

    J. Alvarado, E. Teich, D. Sivak, and J. Bechhoefer, arXiv preprint arXiv:2504.08676 (2025)

    2025 arXiv

  8. [15]

    I. A. Martinez, E. Roldan, J. M. R. Parrondo, and D. Petrov, Physical Review E 87, 032159 (2012)

    2012

  9. [16]

    Chupeau, B

    M. Chupeau, B. Besga, D. Guery-Odelin, E. Trizac, A. Petrosyan, and S. Ciliberto, Physical Review E 98, 010104 (2018)

    2018

  10. [17]

    T. K. Saha, J. Ehrich, M. Gavrilov, S. Still, D. A. Sivak, and J. Bechhoefer, Physical Review Letters 131, 057101 (2023)

    2023

  11. [18]

    Aifer, K

    M. Aifer, K. Donatella, M. H. Gordon, S. Duffield, T. Ahle, D. Simpson, G. Crooks, and P. J. Coles, npj Unconventional Computing 1, 13 (2024)

    2024

  12. [19]

    D. A. Sivak, J. D. Chodera, and G. E. Crooks, Physical Review X 3, 011007 (2013)

    2013

  13. [20]

    H. Wu, J. K¨ ohler, and F. No´ e, Advances in neural infor- mation processing systems 33, 5933 (2020)

    2020

  14. [21]

    Zhong, B

    A. Zhong, B. Kuznets-Speck, and M. R. DeWeese, Phys- ical Review E 110, 034121 (2024)

    2024

  15. [22]

    N. G. Van Kampen, Stochastic processes in physics and chemistry, Vol. 1 (Elsevier, 1992)

    1992

  16. [23]

    Ciliberto, Physical Review X 7, 021051 (2017)

    S. Ciliberto, Physical Review X 7, 021051 (2017)

    2017

  17. [24]

    Jarzynski, The European Physical Journal B 64, 331 (2008)

    C. Jarzynski, The European Physical Journal B 64, 331 (2008)

    2008

  18. [25]

    Van Vu and Y

    T. Van Vu and Y. Hasegawa, Physical Review Letters 126, 010601 (2021)

    2021

  19. [26]

    S. Dago, S. Ciliberto, and L. Bellon, Proc. Nat. Acad. Sci. 120, e2301742120 (2023)

    2023

  20. [28]

    Whitelam, arXiv preprint arXiv:2410.12211 (2024)

    S. Whitelam, arXiv preprint arXiv:2410.12211 (2024)

    2024

  21. [29]

    Proesmans, J

    K. Proesmans, J. Ehrich, and J. Bechhoefer, Physical Review E 102, 032105 (2020)

    2020

  22. [30]

    P. R. Zulkowski and M. R. DeWeese, Physical Review E 89, 052140 (2014)

    2014

  23. [31]

    Barros, S

    N. Barros, S. Whitelam, S. Ciliberto, and L. Bellon, arXiv preprint arXiv:2409.15050 (2024)

    2024 arXiv

  24. [32]

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

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

    1961

  25. [33]

    C. H. Bennett and R. Landauer, Scientific American 253, 48 (1985)

    1985

  26. [34]

    Conte, E

    T. Conte, E. DeBenedictis, N. Ganesh, T. Hylton, J. P. Strachan, R. S. Williams, A. Alemi, L. Al- tenberg, G. Crooks, J. Crutchfield, et al., arXiv preprint arXiv:1911.01968 (2019)

    2019 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.