Skip to content
Crypto.world

The Ledger

Preprint proposes fee-memory rule for Layer 2 posting costs

A Nov. 16 preprint models Layer 2 posting fees as a vault that collects user fees and pays Layer 1 data-posting costs. Its authors report that their proposed rule reduced benchmark tracking error by about one third at the fee movement implied by Arbitrum’s current parameters.

Preprint proposes fee-memory rule for Layer 2 posting costs

A Nov. 16 preprint proposes a feedback rule for Layer 2 fees that would use a rollup’s accumulated balance to adjust what users pay for Layer 1 data posting. Authors Ulysse Pavloff and Lin Oshitani model that accounting position as a “vault” and report that their rule reduced benchmark tracking error by about one third at the fee movement implied by Arbitrum’s current parameters.

Source arxiv.org

A balance-based fee model

The paper treats user fees as deposits into the vault and Layer 1 posting costs as withdrawals. Because revenue-cost differences accumulate over time, the authors distinguish recovering each interval’s posting cost from keeping the vault near a target balance. A fee that merely passes through the current cost can leave an existing surplus or shortfall untouched.

The proposed rule keeps a fraction of the previous fee and adds a correction proportional to the current vault error. The retained fraction, which the paper calls the fee-memory coefficient, sets a trade-off between tracking the target balance and limiting fee changes. More fee memory makes fees smoother but, in the model, allows balance errors to persist longer.

The closed-form analysis assumes fixed, price-independent transaction volume, a correctly known and constant mean posting cost, unpredictable cost deviations and fees without a lower or upper bound. The historical replay instead applies a zero fee floor and no upper cap. The authors caution that clipping fees at zero removes the optimality guarantee derived for the unconstrained benchmark.

Comparison with existing controllers

The paper compares its rule with proportional and proportional-integral controllers and an adaptation of Arbitrum Nitro’s pricing rule. In the idealized benchmark, the authors say their optimal rule achieves about one third less tracking error than the Arbitrum-style rule at the fee movement implied by Arbitrum’s current parameters.

The historical evaluation uses seven months of Ethereum fee data and a modeled workload. At the middle setting, the proposed LQR controller recorded a vault root-mean-square error of 0.0249 ETH and fee movement of 4.9%. The proportional-integral controller’s corresponding error was 0.41 ETH, with fee movement of 7.63%; the paper links much of its error to a surplus retained after Oct. 10 by the anti-windup rule used in the test.

Those results do not establish an empirical optimum: the authors describe the replay as a stress test, not a validation of the benchmark. Their evaluation covers one Layer 1 price trace and one blob-fee regime, with fixed posting cadence and payload. It also uses a full-window mean cost as a retrospective calibration, which the paper notes would not be available at deployment.

Limits and open questions

The authors report that adjacent posting costs in the historical data had a correlation of 0.979, meaning the previous cost could help predict the next one.

Other limitations include fixed demand and the absence of a model for users changing activity in response to fees. The paper also says a fee cap can dominate the controller: in its capped test, the reserve size and cap determined downside outcomes during an October spike, rather than the choice of feedback rule.

Sources

  1. Original source · Preprint Optimal Dynamic Fees for L2s: A Vault Approach arXiv:2611.20838

Related stories

Explore

More articles