Pith. sign in

REVIEW 5 minor 42 references

Occupation Uncertainty Relations

T0 review · 0 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Every ergodic Markov jump process obeys a reference-free lower bound on occupation-time fluctuation rates.

desk verdict A genuine reference-free matrix bound on occupation-time fluctuations, derived cleanly from level-2.5 large deviations; minor presentational slips don't undermine the core result. read the letter →

arxiv 2411.15118 v1 pith:BCHUEKNB submitted 2024-11-22 cond-mat.stat-mech physics.comp-ph

Katarzyna Macieszczak This is my paper · ORCID
classification cond-mat.stat-mechphysics.comp-ph MSC 60J2760F1082C31 PACS 05.40.-a05.10.Ln
keywords occupationuncertaintyrelationsMarkovjumpprocesseslevel-2.5largedeviationstimesstochasticsimulationdynamicexponentdetailedbalanceirreversiblechains
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

Stochastic trajectories spend fluctuating amounts of time in each configuration of a Markov process, and this paper asks how small those fluctuations can be. The answer it proposes is a matrix inequality: for any ergodic continuous-time Markov jump process, the asymptotic covariance matrix of occupation times is bounded from below by a matrix built only from the stationary distribution and the lifetimes of individual configurations. That bound is independent of any reference or observable choice, and it implies ordinary observable uncertainties as special cases. If correct, it gives a universal error benchmark for simulations that estimate stationary averages by occupation times, and it ties dynamic correlations to static ones, making dynamic exponents nonnegative.

What carries the argument

The load-bearing mechanism is the level-2.5 large-deviation rate function for occupation times and transition fluxes of a continuous-time Markov chain. Expanding that rate function to second order around the stationary averages and using the standard identity that the second derivative of the rate function is the inverse asymptotic variance rate, the paper reduces the fluctuation bound to a Cauchy-Schwarz optimization in the space weighted by $B$, the diagonal matrix of decay rates divided by stationary probabilities. The resulting matrix inequality is reference-free because optimizing over the reference distribution is equivalent to projecting with $Q_{\mathrm{ss}}$ and inverting $B$ on the orthogonal complement of the stationary vector.

What would settle it

Take any ergodic continuous-time Markov chain with three or more configurations, diagonalize or simulate it long enough to estimate the covariance matrix $C$ exactly, and check whether any observable $o$ gives $o^{\mathrm{T}} C o < (Q_{\mathrm{ss}}o)^{\mathrm{T}} B^{-1}(Q_{\mathrm{ss}}o)$; one clean numerical counterexample would refute the bound.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the covariance matrix $C$ of occupation-time fluctuations obeys $C \ge Q_{\mathrm{ss}}^{\mathrm{T}} B^{-1} Q_{\mathrm{ss}}$, where $B = D^{-1}\Lambda = \Lambda D^{-1}$ is the diagonal matrix of configuration decay rates weighted by stationary probabilities and $Q_{\mathrm{ss}} = I - \mathbf{1} p_{\mathrm{ss}}^{\mathrm{T}}$ projects off the stationary distribution. Equivalently, for any observable $o$, the asymptotic variance rate satisfies $o^{\mathrm{T}} C o \ge (Q_{\mathrm{ss}}o)^{\mathrm{T}} B^{-1}(Q_{\mathrm{ss}}o)$. The bound is faithful for non-constant observables, invariant under constant shifts, and independent of the reference distribution that ordinary occupation uncertainty relations require. The paper proves that unidirectional cycles saturate the bound, that equilibrium dynamics generally cannot saturate it at finite size, and that detailed-balance processes can saturate it asymptotically in the thermodynamic limit, yielding zero dynamic exponents.

Load-bearing premise

The load-bearing premise is the standard large-deviation law for Markov jump processes saying that the rate at which each configuration is visited and each transition occurs obeys the known level-2.5 rate function, whose second derivative equals the inverse of the occupation-time variance rate.

Editorial extensions

If this is right

  • Occupation-time estimators of stationary averages have a universal lower bound on their asymptotic variance, fixed only by $p_{\mathrm{ss}}$ and the configuration lifetimes.
  • For a given stationary distribution and given decay rates, any dynamics saturating the matrix bound is an optimal sampler of the stationary distribution via occupation times; unidirectional cycles always achieve this.
  • The bound implies $o^{\mathrm{T}} C o \ge \lambda_o^{-1} o^{\mathrm{T}} \Delta^2 o$, so the dynamic variance rate measured in units of the local lifetime always exceeds the static variance.
  • Integrated autocorrelation times satisfy $\lambda_{\max} \tau_o \ge \lambda_o \tau_o \ge 1/2$, so dynamic exponents defined in the thermodynamic limit are nonnegative.
  • Equilibrium dynamics can saturate the bound only asymptotically, which corresponds to vanishing dynamic exponents and optimal simulation performance in that limit.

Reading between the lines

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

  • One could test the matrix bound directly in experiments or simulations by estimating the occupation-time covariance matrix $C$ from long trajectories of a Markov jump process and comparing it with $Q_{\mathrm{ss}}^{\mathrm{T}}B^{-1}Q_{\mathrm{ss}}$ computed from the generator; violations would signal non-Markovian effects.
  • Because the bound depends only on $p_{\mathrm{ss}}$ and lifetimes, it suggests a practical design criterion for irreversible samplers: tune the generator to saturate the inequality for a chosen set of observables, not just for the stationary distribution.
  • The same level-2.5 expansion might extend the matrix bound to time-dependent or periodically driven generators, replacing the stationary distribution by the instantaneous one, though the paper does not make that claim.
  • The inequality connecting global eigenmode timescales and local lifetimes could serve as a diagnostic for decoupling phenomena such as Stokes-Einstein breakdown, where the ratio of these timescales diverges.
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

0 major / 5 minor

Summary. The paper studies finite irreducible continuous-time Markov chains and derives lower bounds on the asymptotic variance of time-integrated system observables. Because a constant shift of a system observable changes the standard uncertainty, the authors define a reference-dependent 'relative uncertainty' and prove an observable-independent bound (Eq. (1)) from the level-2.5 large-deviation rate function for occupations and fluxes. Optimizing over references yields the central matrix occupation uncertainty relation (mOUR), Eq. (3): C >= Q_ss^T B^{-1} Q_ss, where C is the asymptotic covariance-rate matrix of occupation times, D is the diagonal stationary-probability matrix, and B = D^{-1} Lambda. The paper shows that unidirectional cycles saturate the bound, discusses implications for stochastic estimation of stationary averages, derives non-negative dynamic exponents, and gives two finite-size examples with complementary analytic calculations. The supplemental material contains the derivations, tighter reference-dependent bounds, and saturation conditions.

Significance. If correct, Eq. (3) is a genuinely reference-free and parameter-free limit: every ergodic continuous-time Markov chain has occupation-time correlation rates bounded below by Q_ss^T B^{-1} Q_ss, a matrix determined only by stationary probabilities and configuration lifetimes. This is both a fundamental statement and a practical benchmark for Monte Carlo sampling. The central derivation is standard level-2.5 large-deviation input followed by an elementary Cauchy-Schwarz optimization; I checked the pivotal algebra in the supplement, including the expansion in Eq. (S4), the Schur-complement step, and the change of variables that eliminates delta, and found no error. The saturation proof for unidirectional cycles is given in three independent ways, and the counting argument excluding equilibrium saturation for d >= 3 is convincing. The paper ships no code, but the proofs are explicit enough to be checked by hand. The only fragile-looking input is the identification of the Hessian of the level-1 rate function with the inverse asymptotic variance; for finite irreducible CTMCs this is an established result, so I do not regard it as a gap.

minor comments (5)
  1. [OURs, after Eq. (1)] The sentence 'with both the bound and the uncertainty vanishing as epsilon^2' should say 'diverging as 1/epsilon^2', since the denominator [(p_ss - p)^T o]^2 is of order epsilon^2. This is a typographical inversion, but it appears precisely where the tightness of the bound is first discussed.
  2. [OURs, same paragraph] The sentence 'Redefining p = p_ss + delta p, we then find Eq. (1) holds for any p...' is terse. The perturbative derivation gives a quadratic-form inequality in delta p after dividing by epsilon^2 and taking the limit; making this step explicit would prevent a reader from thinking Eq. (1) is an analytic continuation of an infinitesimal statement.
  3. [Supplement I.B, Eq. (S14)] 'As delta B is positive definite' should read 'positive semi-definite'; for unidirectional cycles delta B = 0, and only semi-definiteness is needed for the bounds to be tighter. The later text in Sec. II.B uses the correct wording.
  4. [Supplement III.B, end of 'Derivation of Eq. (S81)'] The sentence 'This yields Eq. (S69)' is a typo; it should refer to Eq. (S81).
  5. [Saturating OURs and mOUR] The claim that saturation of Eq. (3) forces breaking of local detailed balance is stated compactly and delegated to Refs. [26,27]. Since this is one of the paper's interpretive claims, a short precise statement of the external theorem and its hypotheses would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the mOUR follows from external level-2.5 large-deviation theory and a standard Cauchy-Schwarz optimization, with no fitted parameters or self-citation carrying the central claim.

full rationale

The central result, Eq. (3), is derived from the level-2.5 large-deviation principle for finite irreducible continuous-time Markov chains (external Refs. [17,18]) via the standard second-order identification of the rate function with the inverse asymptotic variance. The supplement expands the level-2.5 rate function around the stationary values, optimizes the auxiliary scaling variable, and then maximizes the resulting Rayleigh quotient over the unconstrained variable to obtain the matrix bound; this is a self-contained algebraic argument, not an assumed conclusion. No parameter is fitted to data, and no 'prediction' is a renamed input. The self-citation [24] is contextual, and the companion supplement [19] is included in the manuscript rather than invoked as an unverified external authority. Claims about saturation for unidirectional cycles are checked by direct calculation in the supplement, and the upper bounds from Refs. [26,27] are external and used only to contextualize optimality, not to define the mOUR. The paper's own limitation remarks (e.g., that Example II's optimality is trivial) are honestly stated and do not reveal circularity.

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

No free parameters and no invented entities. The central proof imports the level-2.5 large deviation principle and the central limit theorem for additive functionals as standard external results; the only paper-specific modeling choices are the finite ergodic Markov setting, reference distributions, and the independent-flux assumption used in the saturation analysis.

assumptions (6)
  • domain assumption Finite-state, continuous-time, time-independent Markov jump process with generator W; ergodic with unique stationary distribution pss and pss_j > 0 for all j.
    The whole derivation takes this setting; it enters at Eq. (1) where pss, C, B, and Lambda are defined. Non-ergodic dynamics or zero-probability configurations would break the resolvent and positive-definiteness arguments.
  • standard math Level-2.5 large deviation principle for occupation times and fluxes of Markov processes (Refs. [17,18] and [20,21]).
    Used in Supplement I.A to expand the rate function around the stationary average and to identify the inverse variance rate 1/(o^T C o).
  • standard math The asymptotic variance rate for a time-integrated observable is o^T C o with C = -(DR + R^T D), where R is the group inverse satisfying R W = W R = Q_ss.
    Standard central limit theorem for additive functionals of finite Markov chains; used in Eq. (1) and throughout the proofs.
  • domain assumption Reference distributions p are probability vectors positive on all configurations, and the projection Q = I - 1 p^T is used to remove constant shifts.
    Needed so the relative uncertainty is finite and the projected observables have the same variances and higher cumulants as the original observables.
  • standard math Detailed-balance variance ordering from Refs. [26-29]: among dynamics with the same stationary distribution and time-symmetric flux averages, equilibrium dynamics have the largest asymptotic variance rate for time-integrated observables.
    Used only for placing dynamic exponents relative to equilibrium and for interpreting Example II, not for deriving the matrix occupation uncertainty relation.
  • ad hoc to paper For saturation conditions, the transition-number rates are asymptotically independent random variables aside from the continuity equations, so the condition in Eq. (S47) is necessary unless further hidden constraints exist.
    Supplement III.A assumes no extra algebraic dependence among the d'' random flux variables. This underlies the saturation characterizations but is not needed for the main inequality.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Occupation Uncertainty Relations." pith.science (2026). https://pith.science/paper/BCHUEKNB

@misc{pith2026241115118,
  author       = {Pith},
  title        = {Pith review of: Occupation Uncertainty Relations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BCHUEKNB}},
  note         = {Machine review of arXiv:2411.15118}
}
read the original abstract

We introduce occupation uncertainty relations (OURs) for dynamics of a Markov process over discrete configurations. Those are lower bounds on uncertainties of system observables that are time-integrated along stochastic trajectories. The uncertainty is defined as the ratio of the variance to the square of the average, with the latter necessarily shifted by the average in a reference distribution. The derived bounds are observable independent, but rely on the reference choice. We show, however, that all OURs originate from a matrix bound on correlations between times spent in different system configurations, i.e., occupation times. This result is reference independent, and expressed only in terms of stationary probability and lifetimes of individual configurations. Any dynamics that saturates the matrix bound, is optimal in approximating stationary distribution by occupation times; this always occurs for unidirectional cycles. Furthermore, using this bound, dynamic correlations can be asymptotically bounded in terms of static correlations, in turn leading to the positivity of dynamic exponents in the thermodynamic limit. While at a finite size the matrix bound generally saturates away from equilibrium, we demonstrate detailed-balance dynamics can still saturate it asymptotically, by leading to vanishing dynamic exponents and optimal simulations in that context.

Figures

Figures reproduced from arXiv: 2411.15118 by the authors.

Figure 1
Figure 1. OUR and mOUR for Markov cycles. (a) The dynamics for d = 4 configurations in Example I. (b) The quality of the OUR in Eq. (1) with random references (empty markers; p uniformly sampled for each x) and optimal references [filled markers; coincides with the quality of Eq. (2)], for (o)j = cos(jπ), cos(j2π/d), sin(jπ/2) (red squares, blue circles, green diamonds) and λj ∝ 2 j . (c) The best and worst performance of the… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 32 canonical work pages

  1. [1]

    While the averages of system observables and their time-integrals are asymptotically determined by the sta- tionary distribution, limt→∞⟨O(t)⟩ = limt→∞⟨O(t)⟩/t = ⟨O⟩ss ≡ pT sso, for the variances this holds true only for observables at a final time, limt→∞ ∆2O(t) = oT∆2o with ∆2 ≡ D − psspT ss and (D)jk = δjk (pss)j. The asymptotic rates of the variances ...

    work page

  2. [2]

    Seifert, Stochastic thermodynamics, fluctuation the- orems and molecular machines, Rep

    U. Seifert, Stochastic thermodynamics, fluctuation the- orems and molecular machines, Rep. Prog. Phys 75, 126001 (2012)

    work page 2012

  3. [3]

    A. C. Barato and U. Seifert, Thermodynamic Uncer- tainty Relation for Biomolecular Processes, Phys. Rev. Lett. 114, 158101 (2015)

    2015

  4. [4]

    T. R. Gingrich, J. M. Horowitz, N. Perunov, and J. L. England, Dissipation Bounds All Steady-State Current Fluctuations, Phys. Rev. Lett.116, 120601 (2016)

    2016

  5. [5]

    Dechant and S.-i

    A. Dechant and S.-i. Sasa, Entropic bounds on currents in Langevin systems, Phys. Rev. E97, 062101 (2018)

    work page 2018

  6. [6]

    Dechant and S

    A. Dechant and S. ichi Sasa, Current fluctuations and transport efficiency for general Langevin systems, J. Stat. Mech. Theory Exp.2018, 063209 (2018)

    work page 2018

  7. [7]

    Ito and A

    S. Ito and A. Dechant, Stochastic Time Evolution, In- formation Geometry, and the Cramér-Rao Bound, Phys. Rev. X 10, 021056 (2020)

    work page 2020

  8. [8]

    K. Liu, Z. Gong, and M. Ueda, Thermodynamic Uncer- tainty Relation for Arbitrary Initial States, Phys. Rev. Lett. 125, 140602 (2020)

    work page 2020

Show all 42 references
  1. [9]

    Koyuk and U

    T. Koyuk and U. Seifert, Thermodynamic Uncertainty Relation for Time-Dependent Driving, Phys. Rev. Lett. 125, 260604 (2020)

    2020

  2. [10]

    J. P. Garrahan, Simple bounds on fluctuations and un- certainty relations for first-passage times of counting ob- servables, Phys. Rev. E95, 032134 (2017)

    2017

  3. [11]

    I. D. Terlizzi and M. Baiesi, Kinetic uncertainty relation, J. Phys. A Math. Theor.52, 02LT03 (2018)

    2018

  4. [12]

    Hasegawa and T

    Y. Hasegawa and T. Van Vu, Fluctuation Theorem Un- certainty Relation, Phys. Rev. Lett.123, 110602 (2019)

    2019

  5. [13]

    Ziyin and M

    L. Ziyin and M. Ueda, Universal thermodynamic uncer- tainty relation in nonequilibrium dynamics, Phys. Rev. Res. 5, 013039 (2023)

    2023

  6. [14]

    J. M. Horowitz and T. R. Gingrich, Proof of the finite- time thermodynamic uncertainty relation for steady- state currents, Phys. Rev. E96, 020103 (2017)

    2017

  7. [15]

    Sokal, Monte Carlo Methods in Statistical Mechan- ics: Foundations and New Algorithms, inFunctional In- tegration: Basics and Applications , edited by C

    A. Sokal, Monte Carlo Methods in Statistical Mechan- ics: Foundations and New Algorithms, inFunctional In- tegration: Basics and Applications , edited by C. DeWitt- Morette, P. Cartier, and A. Folacci (Springer US, Boston, MA, 1997) pp. 131–192

    1997

  8. [16]

    den Hollander, Large Deviation Theory (American Mathematical Society, 2000)

    F. den Hollander, Large Deviation Theory (American Mathematical Society, 2000)

    2000

  9. [17]

    Touchette, The large deviation approach to statistical mechanics, Phys

    H. Touchette, The large deviation approach to statistical mechanics, Phys. Rep.478, 1 (2009)

    2009

  10. [18]

    Maes and K

    C. Maes and K. Netočný, Canonical structure of dynam- ical fluctuations in mesoscopic nonequilibrium steady states, EPL 82, 30003 (2008)

    2008

  11. [19]

    Bertini, A

    L. Bertini, A. Faggionato, and D. Gabrielli, Large devi- ations of the empirical flow for continuous time Markov chains, Ann. inst. Henri Poincare, Probab. Stat.51, 867 (2015)

    2015

  12. [20]

    Supplemental Material

  13. [21]

    Bertini, A

    L. Bertini, A. Faggionato, and D. Gabrielli, Flows, currents, and cycles for Markov chains: Large devia- tion asymptotics, Stoch. Process. Their Appl.125, 2786 (2015)

    2015

  14. [22]

    A. C. Barato and R. Chetrite, A Formal View on Level 2.5 Large Deviations and Fluctuation Relations, J. Stat. Phys. 160, 1154 (2015)

    2015

  15. [23]

    We consider any inequalities between matrices according to Loewner order, i.e.,A ≥ B if A − B is positive semi- definite

  16. [24]

    While (C)jj ≥ (pss)j[1−2(pss)j]/(Λ)jj is a bound that is positive for(pss)j < 1/2, this bound cannot be saturated as the omitted non-local term is strictly positive. 7

  17. [25]

    Macieszczak, K

    K. Macieszczak, K. Brandner, and J. P. Garrahan, Uni- fied Thermodynamic Uncertainty Relations in Linear Re- sponse, Phys. Rev. Lett.121, 130601 (2018)

    2018

  18. [26]

    Shiraishi, Optimal Thermodynamic Uncertainty Re- lation in Markov Jump Processes, J

    N. Shiraishi, Optimal Thermodynamic Uncertainty Re- lation in Markov Jump Processes, J. Stat. Phys.185, 19 (2021)

    2021

  19. [27]

    Rey-Bellet and K

    L. Rey-Bellet and K. Spiliopoulos, Improving the Con- vergence of Reversible Samplers, J. Stat. Phys.164, 472 (2016)

    2016

  20. [28]

    Kaiser, R

    M. Kaiser, R. L. Jack, and J. Zimmer, Acceleration of Convergence to Equilibrium in Markov Chains by Break- ing Detailed Balance, J. Stat. Phys.168, 259 (2017)

    2017

  21. [29]

    Hwang, S.-Y

    C.-R. Hwang, S.-Y. Hwang-Ma, and S.-J. Sheu, Acceler- ating Diffusions, Ann. Appl. Probab.15, 1433 (2005)

    2005

  22. [30]

    A. B. Duncan, T. Lelièvre, and G. A. Pavliotis, Vari- ance Reduction Using Nonreversible Langevin Samplers, J. Stat. Phys.163, 457 (2016)

    2016

  23. [31]

    S. C. Kapfer and W. Krauth, Irreversible Local Markov Chains with Rapid Convergence towards Equilibrium, Phys. Rev. Lett.119, 240603 (2017)

    2017

  24. [32]

    We define left αL and right αR eigenvectors of a matrix A corresponding to an eigenvalue a so that (αL)†A = a(αL)† and AαR = aαR and the normali- sation (αL)†αR = 1holds

  25. [33]

    Y. Jung, J. P. Garrahan, and D. Chandler, Excitation lines and the breakdown of Stokes-Einstein relations in supercooled liquids, Phys. Rev. E69, 061205 (2004)

    2004

  26. [34]

    Y. Jung, J. P. Garrahan, and D. Chandler, Dynamical exchanges in facilitated models of supercooled liquids, J. Chem. Phys. 123, 084509 (2005)

    2005

  27. [35]

    Z. Lei, W. Krauth, and A. C. Maggs, Event-chain Monte Carlo with factor fields, Phys. Rev. E99, 043301 (2019)

    2019

  28. [36]

    A. C. Maggs and W. Krauth, Large-scale dynamics of event-chain Monte Carlo, Phys. Rev. E 105, 015309 (2022)

    2022

  29. [37]

    For systems such as D-dimensional lattices of vol- ume V = LD with n configurations per local site, so that d = nV, dynamic exponents are typically calculated with respect to V (cf

    We consider divergence with respect toln d for general- ity. For systems such as D-dimensional lattices of vol- ume V = LD with n configurations per local site, so that d = nV, dynamic exponents are typically calculated with respect to V (cf. Ref. [14]). Then,zo needs to be mu...

  30. [38]

    Dechant and S.-i

    A. Dechant and S.-i. Sasa, Fluctuation–response inequal- ity out of equilibrium, Proc. Natl. Acad. Sci.117, 6430 (2020)

    2020

  31. [39]

    A. M. Timpanaro, G. Guarnieri, J. Goold, and G. T. Landi, Thermodynamic Uncertainty Relations from Ex- change Fluctuation Theorems, Phys. Rev. Lett. 123, 090604 (2019)

    2019

  32. [40]

    Dechant, Thermodynamic constraints on the power spectral density in and out of equilibrium (2023), arXiv:2306.00417

    A. Dechant, Thermodynamic constraints on the power spectral density in and out of equilibrium (2023), arXiv:2306.00417

    2023 arXiv

  33. [41]

    Koyuk and U

    T. Koyuk and U. Seifert, Operationally Accessible Bounds on Fluctuations and Entropy Production in Pe- riodically Driven Systems, Phys. Rev. Lett.122, 230601 (2019)

    2019

  34. [42]

    Pietzonka, Classical Pendulum Clocks Break the Ther- modynamic Uncertainty Relation, Phys

    P. Pietzonka, Classical Pendulum Clocks Break the Ther- modynamic Uncertainty Relation, Phys. Rev. Lett.128, 130606 (2022). 1 Supplemental Material: Occupation Uncertainty Relations In this Supplemental Material proofs of Eqs. (1), (2), and (3) in the main text are given, toge...

    2022

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.