A mathematical analysis of kinetic proofreading examines how the layer carrying an energy-driven transition relates to the accuracy and effective depth of a modeled cascade. For a driven transition at layer j, the attainable error floor has exponent K−j+1 relative to equilibrium error, and every finite drive remains strictly above that floor.
Drive strength shows a conditional pattern in the same model. Error initially decreases as m departs from one and remains below equilibrium for every m > 1. If Q2 is positive, the curve has a unique global minimum and then increasing error; if Q2 is zero or negative, it decreases monotonically toward the model’s stated limit. These are exact results within the specified graph model.
A network for testing molecular choices
The study represents proofreading as a finite-state, continuous-time Markov process: a directed graph with reversible edges in a strongly connected graph and a unique steady state. Its main example is a core “butterfly” network, with two mirrored wings joined at a free-enzyme root. Each wing contains K checkpoint states, and the ratio of the two exit probabilities defines the modeled error, written as ϵ = πē/πe.
Before driving, admissible positive rates satisfy detailed balance and give an equilibrium error of ϵeq = α−1. The comparison leaves graph vertices and edges fixed while multiplying the rates of the selected layer and its symmetry-related partner by m > 1. The corresponding drive affinity is Δμ = ln m.
Downstream cycles and modeled depth
For a driven transition at layer j, the model’s steady-state error is strictly above a floor whose exponent relative to equilibrium error is K−j+1. When the rate constants and drive strength vary, that same floor is the infimum—the lowest limiting value approached by the model.
The analysis describes the exponent in topological terms: it is one plus the first Betti number of the downstream subgraph. Here, that term means the number of independent cycles in that part of the network. Cycles upstream of the driven layer do not increase the effective proofreading depth.
The two extreme placements have different modeled capacities. At the first layer, all K−1 independent cycles cooperate, and the model has the full K-fold discriminatory capacity. At the last layer, the cascade performs no better than a two-checkpoint scheme, regardless of the cascade’s depth.
The response to stronger driving is conditional
An exact formula gives the full dependence of error on drive strength. The drive appears only through c = m−1, while the remaining expression is compressed into four equilibrium quantities. It is an algebraic result calculated from the model rather than fitted to measurements.
The theorem rules out antiproofreading in this model: for every admissible rate choice, error initially decreases when m departs from one and stays below equilibrium error for every m > 1. The theorem gives two later patterns according to Q2. If Q2 is positive, the curve has one global minimum and then increasing error; if Q2 is zero or negative, error decreases monotonically toward the stated limit.
A numerical illustration used K = 4 checkpoints and α = 3, with the analytic curves checked against a direct numerical solution of a nine-state master equation. Each curve started at equilibrium, reached its best modeled error and then rose as the drive increased.
What the result does—and does not—cover
The position-law bound is sharp as an approached infimum. A constructed one-parameter family approaches it as δ tends to zero, but equality is not attained at finite parameter values. The result therefore describes a limiting capacity of the model, rather than a finite setting that necessarily reaches the floor.
The position result is established for the core butterfly topology under the stated conditions: reversible edges in a strongly connected graph, strictly positive rates, detailed balance before driving and one symmetry-related driven pair. This comparison concerns that architecture and drive design; it does not establish the same position rule for arbitrary network topologies or multiple-drive designs.
The document is an arXiv version-1 preprint dated 26 August 2026. Its conclusions are exact within the stated mathematical assumptions, and no statistical uncertainty estimates are reported.
Paper data and sources
Original title: Where Energy Is Spent Sets the Depth of Kinetic Proofreading
Authors: Uğur Çetiner
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text