Pith. sign in

REVIEW 3 major objections 4 minor 30 references

A Framework for Combined Transaction Posting and Pricing for Layer 2 Blockchains

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read An L2 sequencer's optimal posting policy is a threshold: post the whole queue once it exceeds $Q^*(P_t)$, a level set by the current L1 gas price.

desk verdict The framework is sensible and the threshold-policy result is a plausible target, but the supplied proof of Theorem 1 contains a false inequality under the paper's own concavity lemma, so the central result is not established as written. read the letter →

arxiv 2505.19556 v1 pith:36TQFXPZ submitted 2025-05-26 cs.GT

Shouqiao Wang , Davide Crapis , Ciamac C. Moallemi This is my paper · ORCID
classification cs.GT MSC 90C4093E3560K2591B26
keywords Layer2rollupstransactionpostingthresholdpolicydynamicfeemechanismbudgetbalancecongestioncontrolL1gaspricestochasticapproximation
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

This paper argues that an L2 rollup should manage its L1 posting and its L2 fee together, not as separate problems. It shows that the optimal posting decision is a simple threshold: when the pending queue exceeds a threshold that depends only on the current L1 gas price, post every pending transaction; otherwise post none. On the pricing side, it derives unique fees that balance the budget and control congestion, and proposes an adaptive fee-update rule that converges to those fees. The value of the unified picture is that a rollup can keep throughput high while covering volatile L1 posting costs and avoiding L2 congestion.

What carries the argument

The load-bearing object is the value function $J(Q,P)$ of the Markov decision process, with cost $a(Q-S)+(b_0+b_1S)P\,\mathbf{1}_{\{S>0\}}$. A discrete convexity lemma, $2J(Q+1,P) \ge J(Q,P)+J(Q+2,P)$, makes the cost difference monotone in $S$, collapsing the action space to $\{0,Q\}$ and producing the threshold $Q^*(P_t)$. On the pricing side, the mechanism is a projected stochastic-approximation update driven by two observables, per-period profit/loss $X(t;f)$ and the gap between target and actual arrivals $Y(t;p)$, with an adaptive switching rule that selects whichever constraint is binding; the convergence theorem uses Markovian switching and classical stochastic approximation results for decreasing step sizes.

What would settle it

Run the adaptive fee update with the paper's parameters under i.i.d. prices: Theorem 4 predicts $f_t \to f^*$ and $p_t \to p^*$ almost surely, so any systematic bias or non-convergence across repeated long runs would falsify it. Separately, run the same update under the AR(1) price process; lack of convergence there would show that the simulation-based robustness claim does not extend to the paper's own price model.

Watch

Extended reading notes

Core claim

The central claim is that the joint posting-and-pricing problem has a tractable optimal structure. The value function is discrete-convex in queue length, which forces the optimal action at every L1 block to be either "post all" or "post nothing"; the boundary is a threshold $Q^*(P_t)$ that moves with the L1 gas price. Given that policy, the L2 fee that maximizes long-run throughput subject to viability is $\max(f^*, p^*)$, where $f^*$ uniquely balances fee revenue against posting cost and $p^*$ uniquely throttles arrivals to the target rate. Finally, the paper's adaptive mechanism, which switches between budget-balance and congestion-control updates, keeps fees near these targets and converges almost surely to them in the i.i.d. gas-price case.

Load-bearing premise

The convergence theorem for the fee mechanism treats L1 gas fees as independent and identically distributed, but the model describes them as a mean-reverting AR(1) process; for the realistic AR(1) case the paper offers only simulation evidence.

Editorial extensions

If this is right

  • The optimal policy is binary: no partial batches; a sequencer only needs to compare the queue length with $Q^*(P_t)$ and either post everything or post nothing.
  • Policy iteration on the two-action policy runs orders of magnitude faster than prior Q-learning approaches, reducing a single run from about 72 hours to about 6 seconds.
  • Under the stated parameter inequalities there exists a unique budget-balance fee $f^*$ and a unique congestion-control fee $p^*$, and the throughput-maximizing viable fee is $\max(f^*, p^*)$.
  • With decreasing step sizes and i.i.d. L1 prices, the adaptive fee sequence converges almost surely to $(f^*, p^*)$, and the long-run fractions of time spent in each update mode converge to the stationary distribution of the limiting two-state chain.
  • When $f^* \neq p^*$, increasing the observation window $\kappa$ drives the mechanism to select the larger of the two fees with probability approaching 1, at a rate $O(1/\sqrt{\kappa})$.

Reading between the lines

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

  • The paper leaves open a formal convergence proof for the mean-reverting AR(1) gas-price model; a natural extension is to adapt the stochastic-approximation analysis to dependent, geometrically mixing price sequences.
  • The threshold's dependence on $P_t$ suggests an implementable heuristic: estimate the current L1 base fee, read $Q^*(P_t)$ from a table computed offline, and post the entire queue when $Q_t$ crosses it.
  • If real demand is not linear or fees are differentiated by transaction gas usage, the uniqueness of $f^*$ and $p^*$ would need re-derivation; the paper's robustness claims rest on the linear-demand assumption.
  • The constant-step-size, non-i.i.d. simulation regime is the closest to production rollups; the paper does not prove convergence there, so operators should expect bounded fluctuations around $f^*$ and $p^*$ rather than exact convergence.
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

3 major / 4 minor

Summary. The paper develops a discrete-time model of an L2 sequencer that decides whether to post its transaction queue to L1 in the presence of mean-reverting L1 gas prices, and sets L2 fees to balance budget recovery and congestion control. The main theoretical claims are: (i) an optimal posting policy has a threshold form (Theorem 1); (ii) budget-balance and congestion-control fees f* and p* exist and are unique under explicit demand conditions (Theorems 2 and 3); and (iii) an adaptive stochastic-approximation fee update converges almost surely to these fees, with the long-run frequency of the two update modes characterized (Theorem 4). The paper also reports numerical simulations, including settings for which no formal convergence theorem is claimed, such as non-i.i.d. and constant-step-size cases.

Significance. If the advertised results were fully established, the framework would be a useful unification of batch posting and fee design for L2 systems. The threshold-policy formulation is practically appealing and would constitute a substantial simplification over the Q-learning approach cited by the authors; the reported computational speedup is compelling. The fee mechanism extends EIP-1559-style adjustment ideas to the L2 context and the numerical experiments suggest robustness beyond the proved settings. The paper is clearly written and the modeling choices are generally sensible. However, the central proof of Theorem 1 contains a false inequality, and the proof of Theorem 4 does not verify the hypotheses of the stochastic-approximation theorem it invokes. These are load-bearing gaps: without them, the two headline theorems are not established as written.

major comments (3)
  1. [Appendix A, Proof of Theorem 1] The proof of the threshold property uses the inequality J(Q_t+A,P_{t+1}) - J(A,P_{t+1}) <= [Q_t/(Q_t+1)](J(Q_t+1+A,P_{t+1}) - J(A,P_{t+1})), claimed to follow from Lemma 1. Writing d_i = J(A+i,P') - J(A+i-1,P'), Lemma 1 gives only concavity d_1 >= ... >= d_{Q+1}; the asserted inequality is equivalent to (1/Q) sum_{i=1}^Q d_i <= d_{Q+1}, which is the reverse of what concavity implies unless all d_i are equal. The counterexample J(x)=sqrt(x), A=0, Q=1 gives 1 <= 1/sqrt(2), which is false. Since this step is exactly what is used to conclude S_t(Q+1,P)=Q+1 from S_t(Q,P)=Q, the existence of the threshold Q*(P) is not established by the argument as written.
  2. [Appendix A, Proof of Theorem 4] The proof of convergence of f_n and p_n consists essentially of the statement that convergence follows 'straightforwardly' from Theorem 2.1 at Page 127 of [7]. The update is a projected stochastic approximation with state-dependent noise, a random number of observations per update (the cycle length tau(g) depends on the current fee), and a projection onto [0, lambda_0/(2k)]. The boundedness, regression-function, and noise conditions required by the cited theorem are not verified. In the fourth part of the proof, the transition probabilities also appear inconsistent with the switching rule: when delta_t=0, the switch to delta_{t+1}=1 is triggered by X_kappa < 0, not by Y_kappa < 0, so the claimed P10 = P(Y_kappa<0) -> 0 and P00 -> 0 do not correspond to the stated dynamics. Thus both the almost-sure convergence and the stationarity-frequency conclusions of Theorem 4 remain unsupported.
  3. [Appendix A, Proof of Property 1] The strict monotonicity of the expected cost is used in Theorem 2 to prove uniqueness of the budget-balance fee, so the proof matters. The current proof asserts that for every t, P(c(S_t^(1),Q_t^(1),P_t;f1) - c(tilde S_t,tilde Q_t,P_t;f2) >= min{a,b1 P_t}) > 0. This statement is not justified: before any unmarked job has arrived, or in periods with no posting, the cost difference is zero. The intended conclusion can likely be repaired by arguing that with positive probability the first unmarked job eventually arrives and is later posted, producing a positive discounted cost difference, but the argument as written is not rigorous.
minor comments (4)
  1. [Section 4, Property 2] Property 2 is stated as an assertion with no proof and no citation to a proof; if it is not needed for the subsequent theorems it should be marked as a remark, and if it is part of the theoretical foundation it should be proved or removed.
  2. [Appendix A, Proof of Theorem 4] There is a typo on the last page of the proof: 'we can conclude taht' should be 'we can conclude that'.
  3. [Section 5.2] The simulation section reports that the non-i.i.d. and constant-step-size cases lack formal convergence theorems, which is an appropriate caveat; this limitation should be stated prominently in the main body where Theorem 4 is discussed, rather than only in the simulation section.
  4. [Appendix A, Proof of Property 1] The notation c(S,Q,P;f) is overloaded, since the cost function does not depend on f except through the policy and queue process; explicitly defining this dependence would remove ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the derivation chain is self-contained, with the fee mechanism a standard stochastic-approximation root finder and self-citations confined to the literature review.

full rationale

The paper's central claims do not reduce to their inputs by construction. Theorem 1 derives a threshold posting policy from the Bellman equation using concavity-type lemmas on the value function; the threshold is defined from the optimal action, not assumed. The budget-balance fee f* is defined as the root of E[A f - c(S*, Q, P; f)] = 0, and the congestion fee p* is the explicit root p* = (lambda0 - lambda_bar)/k. Existence and uniqueness (Theorems 2 and 3) are proven from monotonicity (Property 1) and explicit parameter conditions; these results do not rely on the convergence theorem. The adaptive fee mechanism is a Robbins-Monro stochastic approximation: ft+1 = projection(ft - a X) and pt+1 = projection(pt - b Y), and Proposition 1 shows E[X] = E[tau] E[A f - c] and E[Y] = E[tau] E[lambda_bar - A], so the updates target exactly the previously defined roots. Theorem 4's convergence proof invokes an external, machine-checkable-style reference (Kushner and Yin [7]) for stochastic approximation, not a self-citation, and does not assume the conclusion. The simulations calibrate parameters and test the mechanism's self-consistency against the same model; they are not presented as external predictions, so there is no fitted-input-called-prediction pattern. Self-citations ([4], [5], [6]) appear only in the literature review and are not load-bearing for the paper's theorems. The proof of Theorem 1 contains a questionable analytic inequality that may be a correctness gap, but that is not circularity under the specified criteria: no equation is defined in terms of the conclusion, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 11 free parameters · 6 assumptions · 0 invented entities

The framework rests on several modeling choices (linear demand, AR(1) gas prices, linear delay cost, compensation mechanism) and on the existence of stationary distributions. The specific parameter values for the cost function and discount factor are not reported, limiting empirical validation. No new physical or mathematical entities are introduced.

free parameters (11)
  • lambda_0 (maximum arrival rate at zero fee) = 180
    Calibrated from historical transaction counts and a fee cap; used in the demand model and simulations.
  • k (demand sensitivity) = 1.67e6
    Derived from the fee cap corresponding to zero arrivals and the typical L2 fee in the simulation setup.
  • mu (mean L1 gas fee) = 3.86e-8 ETH
    Calibrated from L1 base fee data in February 2024.
  • sigma (L1 fee volatility) = 1.93e-8 (i.i.d.) or 8.41e-9 (AR(1))
    Calibrated from the same L1 base fee data; different values for the two simulation scenarios.
  • theta (mean reversion speed) = 0.1
    Hand-chosen for the AR(1) gas price simulation.
  • a (per-block delay penalty)
    Appears in the cost and compensation functions; no value is reported in the paper.
  • b0 (fixed posting cost)
    Fixed cost in the posting cost function; no value reported.
  • b1 (per-transaction posting cost)
    Linear sensitivity of posting cost to gas price; no value reported.
  • gamma (discount factor)
    Discount factor in the objective function; no value reported.
  • lambda_bar (target arrival rate)
    Target for congestion control; not specified in the simulation setup.
  • step sizes a_fee and b_fee
    Step sizes for the fee update rules; not specified.
assumptions (6)
  • domain assumption Arrival process A(t;g) is Poisson with rate lambda(g) = lambda0 - k g.
    Assumed in Section 4 (Arrival Rate) to derive budget balance and congestion fees; supported only by a non-peer-reviewed blog post.
  • domain assumption L1 gas price follows an AR(1) mean-reverting process P_{t+1} = theta*mu + (1-theta)*P_t + sigma*omega_t.
    Stated in Section 2; used for cost expectations and in non-i.i.d. simulations. The theoretical convergence theorem instead assumes i.i.d. fees.
  • domain assumption Cost structure c(Q,S,P) = a(Q-S) + (b0+b1*S)*P*1{S>0}.
    Stated in Section 2; the linear delay penalty and fixed posting cost are modeling choices.
  • ad hoc to paper Users are compensated a per L1 block of delay, making utility independent of posting time.
    Introduced in Section 2 (Compensation Mechanism) to align user incentives with the cost objective.
  • domain assumption The objective is to minimize expected discounted cost and to achieve budget balance and congestion control.
    The objectives are stated as design goals; they are not derived.
  • domain assumption Stationary distribution of the queue exists under the optimal policy.
    Implicitly assumed when taking expectations E[c(S*,Q,P;f)] and when using the ergodic theorem in Proposition 1; not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Framework for Combined Transaction Posting and Pricing for Layer 2 Blockchains." pith.science (2026). https://pith.science/paper/36TQFXPZ

@misc{pith2026250519556,
  author       = {Pith},
  title        = {Pith review of: A Framework for Combined Transaction Posting and Pricing for Layer 2 Blockchains},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/36TQFXPZ}},
  note         = {Machine review of arXiv:2505.19556}
}
read the original abstract

This paper presents a comprehensive framework for transaction posting and pricing in Layer 2 (L2) blockchain systems, focusing on challenges stemming from fluctuating Layer 1 (L1) gas fees and the congestion issues within L2 networks. Existing methods have focused on the problem of optimal posting strategies to L1 in isolation, without simultaneously considering the L2 fee mechanism. In contrast, our work offers a unified approach that addresses the complex interplay between transaction queue dynamics, L1 cost variability, and user responses to L2 fees. We contribute by (1) formulating a dynamic model that integrates both posting and pricing strategies, capturing the interplay between L1 gas price fluctuations and L2 queue management, (2) deriving an optimal threshold-based posting policy that guides L2 sequencers in managing transactions based on queue length and current L1 conditions, and (3) establishing theoretical foundations for a dynamic L2 fee mechanism that balances cost recovery with congestion control. We validate our framework through simulations.

Figures

Figures reproduced from arXiv: 2505.19556 by the authors.

Figure 1
Figure 1. Illustration of the fee update mechanism under decreasing step size and i.i.d. L1 fees. The top row displays the evolution of the posting fees over time, with red dashed lines marking the theoretical optimal levels f ∗ and p ∗ . In the bottom row, the plots show the proportion of times each type of fee update mechanism (budget balance vs. congestion control) is selected. All four trajectories exhibit convergence beh… view at source ↗
Figure 2
Figure 2. Illustration of the fee update mechanism under decreasing step size and non￾i.i.d. L1 fees. The top row displays the evolution of the posting fees over time, with red dashed lines marking the theoretical optimal levels f ∗ and p ∗ . In the bottom row, the plots show the proportion of times each type of fee update mechanism (budget balance vs. congestion control) is selected. All four trajectories exhibit convergence… view at source ↗
Figure 3
Figure 3. Illustration of the fee update mechanism under constant step size and non-i.i.d. L1 fees. The top row shows time-series of the fee updates ft and pt under a mean￾reverting L1 fee process. In the middle row, the plots track the long-run fraction of times each update rule is chosen, indicating how often the budget-balance versus congestion￾control mechanism is active. The bottom row presents empirical histograms of ft… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 29 canonical work pages

  1. [7]

    Stochastic approximation and recursive algorithm and applications

    J Harold, G Kushner, and George Yin. Stochastic approximation and recursive algorithm and applications. Application of Mathematics, 35(10), 1997

    work page 1997

  2. [1]

    Optimal Publishing Strategies on a Base Layer

    Yogev Bar-On and Yishay Mansour. Optimal publishing strategies on a base layer. arXiv preprint arXiv:2312.06448, 2023

    work page Pith review arXiv 2023

  3. [2]

    Towards a Functional Fee Market for Cryptocurrencies

    Soumya Basu, David Easley, Maureen O’Hara, and Emin G¨ un Sirer. Towards a functional fee market for cryptocurrencies. arXiv preprint arXiv:1901.06830, 2019

    work page Pith review arXiv 1901

  4. [3]

    Arbitrum nitro: A second-generation optimistic rollup, 2022

    Lee Bousfield, Rachel Bousfield, Chris Buckland, Ben Burgess, Joshua Colvin, Ed- ward W Felten, Steven Goldfeder, Daniel Goldman, Braden Huddleston, H Kalon- der, et al. Arbitrum nitro: A second-generation optimistic rollup, 2022

    work page 2022

  5. [4]

    Eip-4844 fee market analysis

    Davide Crapis. Eip-4844 fee market analysis. https://ethresear.ch/t/ eip-4844-fee-market-analysis/15078 , March 2023

    work page 2023

  6. [5]

    EIP-4844 Economics and Rollup Strategies

    Davide Crapis, Edward W Felten, and Akaki Mamageishvili. Eip-4844 economics and rollup strategies. arXiv preprint arXiv:2310.01155, 2023

    work page Pith review arXiv 2023

  7. [6]

    Optimal Dynamic Fees for Blockchain Resources

    Davide Crapis, Ciamac C. Moallemi, and Shouqiao Wang. Optimal dynamic fees for blockchain resources. CoRR, abs/2309.12735, 2023

    work page Pith review arXiv 2023

  8. [8]

    Handbook of the poisson distribution, 1968

    SK Katti and A Vijaya Rao. Handbook of the poisson distribution, 1968

    work page 1968

Show all 30 references
  1. [9]

    Redesigning bitcoin’s fee market

    Ron Lavi, Or Sattath, and Aviv Zohar. Redesigning bitcoin’s fee market. ACM Transactions on Economics and Computation, 10(1):1–31, 2022

    2022

  2. [10]

    Optimality despite chaos in fee markets

    Stefanos Leonardos, Dani¨ el Reijsbergen, Barnab´ e Monnot, and Georgios Piliouras. Optimality despite chaos in fee markets. CoRR, abs/2212.07175, 2022

    2022 arXiv

  3. [11]

    Little’s law

    John DC Little and Stephen C Graves. Little’s law. Building intuition: insights from basic operations management models and principles, pages 81–100, 2008

    2008

  4. [12]

    Akaki Mamageishvili and Edward W. Felten. Efficient Rollup batch posting strategy on Base Layer. CoRR, abs/2212.10337, 2022. arXiv:2212.10337, doi: 10.48550/arXiv.2212.10337

    2022 doi arXiv

  5. [13]

    Gas fees on the ethereum blockchain: from foundations to derivative valuations.Frontiers in Blockchain, 7:1462666, 2024

    Bernhard K Meister and Henry CW Price. Gas fees on the ethereum blockchain: from foundations to derivative valuations.Frontiers in Blockchain, 7:1462666, 2024

    2024

  6. [14]

    Markov decision processes: discrete stochastic dynamic pro- gramming

    Martin L Puterman. Markov decision processes: discrete stochastic dynamic pro- gramming. John Wiley & Sons, 2014

    2014

  7. [15]

    Transaction fee mechanism design for the ethereum blockchain: An economic analysis of eip-1559

    Tim Roughgarden. Transaction fee mechanism design for the ethereum blockchain: An economic analysis of eip-1559. arXiv preprint arXiv:2012.00854, 2020

    2012 arXiv

  8. [16]

    Getting the pricing right in crypto, 2024

    Matt Stephenson and Ally Zach. Getting the pricing right in crypto, 2024. URL: https://panteracapital.com/ research-getting-the-pricing-right-in-crypto/

    2024

  9. [17]

    Correct cryptocurrency asic pricing: Are miners over- paying? In 5th Conference on Advances in Financial Technologies (AFT 2023)

    Aviv Yaish and Aviv Zohar. Correct cryptocurrency asic pricing: Are miners over- paying? In 5th Conference on Advances in Financial Technologies (AFT 2023). Schloss Dagstuhl-Leibniz-Zentrum f¨ ur Informatik, 2023

    2023

  10. [18]

    An incentive analysis of some bitcoin fee designs

    Andrew Chi-Chih Yao. An incentive analysis of some bitcoin fee designs. arXiv preprint arXiv:1811.02351, 2018. 18 Shouqiao et al. A Proofs Proof of Theorem 1. Proof. Consider the objective function J (Qt, Pt) = min {Ss} E  X s≥t γs−t [a (Qs − Ss) + (b0 + b1Ss) Ps1{Ss > 0}] ...

    2018 arXiv

  11. [19]

    Third, we show that the long-run average proportions i(t)/t → πf and j(t)/t → πp almost surely

    Similarly, we can also show that pt → p∗ almost surely. Third, we show that the long-run average proportions i(t)/t → πf and j(t)/t → πp almost surely. We have already established that the update rules for the fees satisfy ft − →f ∗, p t − →p∗, a.s. and that, when the fees are...

  12. [20]

    if δt = 1, P (δt+1 = 1|δt = 1, Ft) = P (δt+1 = 1|δt = 1, gt) ≥ P11 − α

  13. [21]

    This is because the sequence{ft}t>τ and {pt}t>τ will fall into very close neighborhoods of f ∗ and p∗ respectively, by the definition of almost surely convergence

    if δt = 0, P (δt+1 = 1|δt = 0, Ft) = P (δt+1 = 1|δt = 0, gt) ≥ P01 − α with probability 1. This is because the sequence{ft}t>τ and {pt}t>τ will fall into very close neighborhoods of f ∗ and p∗ respectively, by the definition of almost surely convergence. Let’s define Tl = inf{...

  14. [22]

    At step t + 1, we set δt+1 =    1, if ξl,1 ≤ P (δt+1 = 1|δt = 1, gt), 0, otherwise

  15. [23]

    Title Suppressed Due to Excessive Length 27

    If δt+1 = 1, the procedure stops for this row, and we move immediately to the next row ( {ξl+1,s}s≥1). Title Suppressed Due to Excessive Length 27

  16. [24]

    At step t + 2, we then use the next entry in this row, setting δt+2 = ( 1, if ξl,2 ≤ P (δt+2 = 1|δt+1 = 0, gt+1), 0, otherwise

    If δt+1 = 0, we remain on row l. At step t + 2, we then use the next entry in this row, setting δt+2 = ( 1, if ξl,2 ≤ P (δt+2 = 1|δt+1 = 0, gt+1), 0, otherwise

  17. [25]

    More generally, for each s ≥ 2, if we have not yet returned to state 1 by step t + s − 1, we continue using the subsequent entries of row l as follows: δt+s = ( 1, if ξl,s ≤ P (δt+s = 1|δt+s−1 = 0, gt+s−1), 0, otherwise

  18. [26]

    After returning to state 1, we increment the row index from l to l + 1 and repeat the entire process with the new row

    This process continues until the chain returns to the state 1. After returning to state 1, we increment the row index from l to l + 1 and repeat the entire process with the new row. Thus, the random time spent on row l is precisely the duration until the chain first returns to...

  19. [27]

    At step t + 1, set βt+1 =    1, if ξl,1 ≤ ˆP11, 0, otherwise

  20. [28]

    If βt+1 = 0, continue using the subsequent entry ξl,2 at step t + 2

    If βt+1 = 1, stop using row l and increment the row index tol+1. If βt+1 = 0, continue using the subsequent entry ξl,2 at step t + 2

  21. [29]

    For each subsequent step t + s with s ≥ 2, if βt+s−1 = 0, define βt+s =    1, if ξl,s ≤ ˆP01, 0, otherwise

  22. [30]

    After returning to state 1, increment the row index to l + 1 and repeat the procedure

    Continue this process until β returns to state 1. After returning to state 1, increment the row index to l + 1 and repeat the procedure. 28 Shouqiao et al. We also define the return time as ˆtl = inf{s ≥ 1 : βt+s = 1}. Since ∀t > τ, P (δt+1 = 1|δt = 1, gt) ≥ ˆP11, P (δt+1 = 1|...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.