Inertial Mining: Equilibrium Implementation of the Bitcoin Protocol

Source: arXiv:2604.06092 · Published 2026-04-07 · By Manuel Mueller-Frank, Minghao Pan, Omer Tamuz

TL;DR

This paper addresses a fundamental incentive problem in Bitcoin's proof-of-work mining: the standard Bitcoin mining protocol is not a Nash equilibrium because miners can increase their payoff by deviating via selfish mining, which involves withholding and strategically releasing blocks. Selfish mining incentivizes creating forks, harming consensus stability and trust. The authors propose a new mining protocol called inertial mining, which enforces a stricter condition on when miners switch to a competing fork—miners switch only if the alternative chain is at least I blocks longer than their current chain. By choosing the parameter I sufficiently large (depending on mining power distribution), inertial mining forms a Nash equilibrium where no miner can improve payoff by deviating, effectively eliminating selfish mining and other profitable strategies. Importantly, it achieves the intended outcome of the original Bitcoin protocol (a unique longest chain) without changing Bitcoin's consensus mechanism or blockchain structure. The key technical challenge is showing that all potential deviation strategies fail to yield a higher payoff than honest mining, which is handled by a detailed probabilistic analysis assigning displaced blocks to "killer" blocks and bounding expected utility gains.

Key findings

• The standard Bitcoin mining protocol is not a Nash equilibrium due to selfish mining, profitable for miners with computational power α_i > 1/4 assuming tie-breaking probability γ = 1/2 (Eyal and Sirer 2018).
• Inertial mining introduces a threshold parameter I > 0 such that miners switch to competing chains only if they are at least I blocks longer; for sufficiently large I, inertial mining is a Nash equilibrium for any maximum mining power α_i < 1/2.
• Under inertial mining, miners always achieve a utility equal to their mining power share α_i almost surely (Claim 1).
• Selfish mining is no longer profitable since the requirement to switch only for chains longer by I blocks disincentivizes strategic withholding and disclosure.
• The proof technique assigns each displaced block to an initial "killer" dishonest block and shows no dishonest block can expect to displace enough honest blocks to outweigh its cost, bounding expected payoffs (Proposition 2).
• The analysis reduces to two-player games for notational convenience but extends naturally to multiple miners by symmetric reasoning.
• No other strategic deviations besides selfish mining can improve utility above mining power share once inertial mining is adopted.
• The protocol differs from standard Bitcoin only off-path (in case of forks) and can be implemented without changing Bitcoin's consensus or blockchain architecture.

Threat model

The adversary is a rational miner with less than half of total mining power who can withhold blocks, selectively publish blocks, and choose any mining strategy to maximize their fraction of blocks in the longest chain. The adversary sees public chain history and their private blocks but not others' private blocks. They cannot control network delivery, add hash power beyond their share, or break cryptographic primitives.

Methodology — deep read

1. Threat model & assumptions: The miners are rational strategic agents who aim to maximize their expected share of blocks in the longest chain, reflecting profits from block rewards. No miner controls half or more of the mining power, and all miners know the distribution of mining power. Adversaries can withhold, release blocks at chosen times, block on any chain, and observe their own and public blocks but not others' private blocks. Communications are via broadcast, and no changes are made to Bitcoin's consensus or blockchain design.

2. Data & setup: The model is a discrete-time stochastic mining game with a finite set of miners N, each with mining power α_i > 0 summing to 1. Blocks are labeled uniquely, mined probabilistically proportional to mining power per round. The state history visible to each miner consists of own and all public blocks. The protocol tracks mining actions, publishing decisions, and chain extensions. Utility is asymptotic fraction of own blocks on the unique longest chain if it exists (else zero).

3. Protocol and strategies: The baseline is the standard Bitcoin mining protocol where miners always mine on the longest chain and publish immediately. Eyal and Sirer's selfish mining deviation involves withholding blocks to generate a private chain lead and selective disclosure.

The new protocol, inertial mining, introduces a parameter I ≥ 1. Miners mine on the last block of the current longest chain when that chain is unique. In cases where there are multiple chains, miners switch to a competing chain only if it is longer by at least I blocks compared to their current chain; otherwise they randomize mining among top chains containing their current block.

All blocks are published immediately when mined (on path). The core difference is in tie-breaking and chain switching only in the presence of longer competing chains beyond threshold I.

4. Training / equilibrium analysis: The authors do not train a model but analyze equilibria of the repeated mining game under inertial mining, proving no miner with less than half mining power can profitably deviate for large enough I. They focus on pure strategies without randomization by deviator (Claim 2).

5. Evaluation and proofs: The core evaluation is a theoretical game-theoretic proof. They analyze the expected payoff of any deviation strategy for a single miner against others playing inertial mining. Key tools include defining honest vs dishonest blocks, the concept of "killed" blocks displaced from the longest chain, and associating every killed block with a "killer" dishonest block.

They show (Proposition 2) for suitable choices of parameters J and I-J, the expected payoff for the deviator is bounded above by their mining power share, implying no profitable deviation.

Random walks model differences in block production between deviator and honest miners, with hitting probabilities bounding chances of profitable chain overtakes.

6. Reproducibility: This is a theoretical equilibrium construction and proof. No code or experimental datasets are involved. The mining game model and strategies are fully derived mathematically, with all definitions and claims rigorously stated. Probabilistic computations rely on standard random walk hitting time formulas.

Example end-to-end: Consider a miner with mining power α_i < 1/2 playing inertial mining with parameter I. If another miner attempts selfish mining by withholding blocks, their forks are ignored unless they exceed the honest chain length by I blocks. Due to random chance and mining delays modeled as biased random walks, the selfish miner cannot achieve this lead sufficiently often to improve payoff beyond α_i, thus no incentive to deviate.

Technical innovations

• Propose inertial mining, a mining protocol that delays miners switching to a competing chain until it exceeds the current chain length by at least parameter I, eliminating profitable selfish mining.
• Prove via probabilistic game-theoretic analysis that inertial mining forms a Nash equilibrium for any mining power distribution with no majority miner, generalizing beyond known selfish mining deviations.
• Introduce the "killer" block accounting technique that assigns every displaced honest block to a particular dishonest "killer" block, allowing bounding of expected payoff from deviations.
• Show inertial mining achieves Nakamoto’s intended unique longest chain outcome on equilibrium, without modifying Bitcoin’s consensus mechanism or blockchain architecture.

Baselines vs proposed

• Standard Bitcoin protocol: miners with α_i > 1/4 can profitably perform selfish mining with payoff > α_i (Eyal and Sirer 2018) vs Proposed inertial mining: no profitable deviations for any α_i < 1/2 with sufficiently large I.
• Baseline honest mining under standard Bitcoin: utility = mining power α_i vs Proposed inertial mining: equilibrium utility = α_i with no profitable deviation.

Limitations

• The requirement for sufficiently large parameter I is shown theoretically but exact bounds for practical mining power distributions are left for future work.
• The equilibrium proof assumes no miner controls half or more of total mining power; majority miners break equilibrium trivially.
• The model abstracts away transaction fees and network latency effects, focusing on asymptotic fraction of blocks mined.
• The approach does not experimentally validate parameter choices or implementation feasibility in deployed Bitcoin clients or networks.
• The strategy space considered involves pure strategies by the deviator; mixed strategies and collusion scenarios are not addressed in detail.
• The public randomization device assumption may be non-trivial to realize in practice.

Open questions / follow-ons

• What exact values of I suffice for equilibrium under realistic mining power distributions in Bitcoin networks?
• How do network latency, block propagation delays, and miner communication protocols affect inertial mining’s practical deployment and effectiveness?
• Can the inertial mining protocol be efficiently integrated into existing Bitcoin clients with backward compatibility and how does this affect incentive dynamics during rollout?
• Is inertial mining resilient to colluding miners or mixed-strategy deviations beyond pure strategies considered?

Why it matters for bot defense

While this paper focuses on proof-of-work blockchain mining incentives rather than bot defense or CAPTCHAs, the underlying theme of incentive-aligned equilibria in adversarial, distributed systems is relevant. For bot-defense engineers, the inertial mining protocol exemplifies how modifying agent behavior only in rare off-path conditions (forks) can restore equilibrium and eliminate profitable malicious strategies without altering core protocol mechanics. Similarly, bot defense often seeks minimal protocol tweaks that deter adversaries strategically without degrading normal user experience.

Moreover, the techniques used—such as attributing adversarial actions to "killer" events to bound payoffs and probabilistic random walk models to analyze incentive feasibility—might inspire analogous analytical frameworks in CAPTCHA robustness or bot-detection strategy design, particularly in settings where adversaries delay or manipulate challenge responses. However, direct technical translation is limited given the domain differences.

Cite

@article{arxiv2604_06092,
  title={ Inertial Mining: Equilibrium Implementation of the Bitcoin Protocol },
  author={ Manuel Mueller-Frank and Minghao Pan and Omer Tamuz },
  journal={arXiv preprint arXiv:2604.06092},
  year={ 2026 },
  url={https://arxiv.org/abs/2604.06092}
}

Read the full paper

• arXiv:2604.06092 (abstract)
• arXiv:2604.06092 (PDF)