Skip to content
Crypto.world

Research

Decentralized Equilibrium for Bitcoin Mining

Bitcoin's prescribed mining rule ("always extend the longest public chain and publish immediately") is known not to be an equilibrium: Eyal and Sirer (2018) showed that a miner can profit by…

Decentralized Equilibrium for Bitcoin Mining
arXiv
2604.06092
Published
2026-04-07
Authors
Manuel Mueller-Frank, Minghao Pan, Omer Tamuz

AI summary

Overview

Research area: Game theory and mechanism design applied to proof-of-work (PoW) blockchains, specifically the incentive structure of Bitcoin mining; filed under cs.CR (cryptography and security).

Technical level: Advanced. The paper builds a formal infinite-horizon stochastic game with Bayesian information (each miner sees only its own mined blocks and the public blocks) and proves an equilibrium theorem using a combinatorial accounting argument.

Scope in one sentence: The paper constructs and proves the existence of a mining protocol — "inertial mining" — that is a Nash equilibrium of the standard Bitcoin mining game, yields a single longest chain, and gives each miner an asymptotic block share equal to its share of computational power, as long as no miner controls one half or more of the computational power.

What This Paper Is About

Bitcoin's prescribed mining rule ("always extend the longest public chain and publish immediately") is known not to be an equilibrium: Eyal and Sirer (2018) showed that a miner can profit by withholding blocks ("selfish mining"), so Bitcoin's intended decentralized outcome had no known equilibrium foundation. The authors ask whether a fair, decentralized equilibrium exists — one with a single longest chain in which every miner's share of blocks equals its share of computational power — without redesigning Bitcoin's consensus mechanism or blockchain architecture. They answer yes by proposing a new mining strategy, inertial mining, and prove it is an equilibrium whenever max_i α_i < 1/2.

Key Contributions

  1. A new protocol, inertial mining. Miners append to the last block of the current longest chain when there is a single chain, but if competing chains exist they do not switch to a rival chain unless it is longer by at least a parameter I > 0 blocks. The parameter I must be chosen large enough as a function of max_i α_i; the closer max_i α_i is to one half, the larger I must be.

  2. An equilibrium existence theorem (Theorem 1). For any mining power profile with max_i α_i < 1/2, inertial mining is an equilibrium of the mining game for sufficiently large I. Because inertial mining coincides with Nakamoto's protocol on the equilibrium path, the equilibrium outcome is a single longest chain that satisfies what the authors call fair decentralized consensus: each miner i's asymptotic share of blocks equals α_i.

  3. A negative result for majority miners (Proposition 2). If α_1 > 1/2, then player 1 asymptotically owns all blocks in the longest chain in every equilibrium — so no decentralized equilibrium exists in that regime. This gives a game-theoretic counterpart to the familiar one-half threshold from the Byzantine fault tolerance perspective.

  4. A proof technique for a rich deviation space. Since the utility is asymptotic, no one-shot deviation principle applies. The authors introduce an accounting system that assigns each killed block of the honest player to a "killer" block of the deviator, and show that no strategically mined block can be expected to kill enough honest blocks to offset the risk that it never enters the longest chain.

A further contribution is Proposition 1: whenever there is a unique largest miner (α_1 > α_i for all i > 1, which is generically the case), a centralized equilibrium exists in which u_1 = 1 and u_i = 0 for all i > 1 — even without a majority.

Main Findings

  • Nakamoto's protocol is not an equilibrium. Eyal and Sirer (2018) show a selfish miner with power α_i earns more than α_i when α_i > (1-γ)/(3-2γ), where γ is the probability others extend the selfish miner's block on a tie. In this paper's model, uniform tie-breaking gives γ = 1/2, so the threshold is α_i > 1/4.

  • Inertial mining kills selfish mining. With I large enough, withholding and selective disclosure no longer pays, and the paper's main technical work is showing that no other deviation pays either.

  • On-path equivalence. Under inertial mining all blocks are published as soon as they are mined, so the equilibrium path is identical to the outcome of the standard Bitcoin protocol; the two protocols differ only off path, after a deviation is observed.

  • Fairness holds almost surely (Claim 1). Under inertial mining there is a unique chain containing every mined block, so by the strong law of large numbers each player's utility is almost surely α_i.

  • Majority implies centralization. When α_1 > 1/2, player 1 can always deviate to mine its own chain and guarantee utility u_1 = 1, forcing centralization in every equilibrium.

  • Centralization is generically an equilibrium even below one half (Proposition 1). If one miner is strictly the largest, everyone mining only on their own chain and ignoring others is an equilibrium in which the largest miner asymptotically owns all blocks. The authors note this is not realistic, since other miners would have no reason to remain in the game.

  • Implementation requires no protocol change. Inertial mining can be deployed in Bitcoin without modifying its consensus mechanism or blockchain architecture, unlike Freshness Preferred, ZeroBlock, or FruitChains, which redesign the mechanism or architecture.

  • The magnitude of I is not reported. The paper proves existence for "sufficiently large" I and explicitly leaves an explicit calculation for future work, conjecturing that I is small enough to be practical provided max_i α_i is not extremely close to one half.

  • Comparison point on restricted models. Kiayias et al. (2016), in a complete-information game where each miner may build at most one block at each depth, show the standard protocol is an equilibrium when all α_i ≤ 0.308.

Methodology in Plain English

The authors model mining as a discrete-time game. Each miner i holds a fraction α_i of total computational power (the α_i sum to one). Blocks are pairs of labels drawn uniformly from [0,1], so labels are almost surely unique, and everything descends from a genesis block (0,0). In each period nature picks a miner with probability equal to its mining power; that miner attaches a new block to whichever parent it chose, and may publish any subset of its private blocks. Miners see only their own blocks plus the public ones, and a public randomization device is available for tie-breaking. A miner's payoff is the limit inferior of its asymptotic fraction of blocks in the longest public chain (averaged over longest chains if there are several) — a long-lived-miner, relative-reward assumption standard in this literature and inherited from Eyal and Sirer (2018).

An equilibrium is a profile in which no miner gains by switching strategy. The proof fixes a single deviator i and lumps all other miners into one honest player j with power 1 - α_i, which is without loss of generality because all non-deviators choose the same target block.

The heart of the argument is bookkeeping. At any time horizon T, some blocks of the honest player are "killed" (not on the longest chain). Each killed block is matched to a cousin — the block on the longest chain at the same depth — and the authors show (Claim 3) that a cousin must have been dishonestly mined, because honestly mined blocks never share a depth (Claim 2). Each killed block is then charged to a specific dishonestly mined "killer" block, using a window parameter J between 1 and I - 1. The proof shows that, for J large and I - J large, a deviating block cannot expect to kill enough honest blocks to compensate for the probability that it never makes it into the longest chain. Since the honest players' guaranteed payoff is α_i, the deviator gains nothing.

Why This Matters

The paper closes a gap that has been open since Eyal and Sirer (2018): Bitcoin's decentralization promise had no known equilibrium foundation, and the absence of such an equilibrium was cited in discussions around Ethereum's move away from PoW (Buterin, 2017; Hall, Shiaeles, and Li, 2024). The authors show the promise can be kept within the existing architecture.

Real-world applications:

  • Mining pool and client policy: a deviation-detection rule that stays on the current branch unless a rival chain leads by I blocks could be implemented by pools or node clients without touching consensus rules.
  • Empirical forensics on PoW chains: the paper cites Li, Campajola, and Tessone (2024), who find evidence consistent with selfish mining in several PoW blockchains, especially Monacoin and Bitcoin Cash — inertial mining gives a concrete countermeasure to test against such data.
  • Security parameter reasoning for exchanges and custodians: the majority threshold now has an equilibrium interpretation, not just a fault-tolerance one, which informs confirmation-depth and counterparty-risk policies.
  • Guidance for new PoW chain launches: any chain whose mining power distribution approaches a single dominant miner faces the centralization result of Proposition 2 as a design constraint.

Industry relevance: miners, pool operators, and protocol developers gain a strategy that is individually rational and preserves proportional rewards; the paper also notes that in over sixteen years of Bitcoin operation little selfish mining has been observed, so the practical value is as much about robustness guarantees as about observed attacks.

Future Directions

  • An explicit bound on I. The theorem only asserts existence for sufficiently large I; the authors leave the calculation for future work and conjecture that I is small enough to be practical when max_i α_i is not extremely close to one half. Determining the exact trade-off is the most direct open problem.

  • Robustness to other utility specifications. The paper uses the asymptotic fraction of blocks in the longest chain and expects "similar results" for standard discounted utilities, but does not prove this.

  • Richer environments. The model assumes discrete time, independent per-period mining success, and a public randomization device (which the authors suggest could in practice be replaced by secret-sharing on block labels). Behavior under network outages and propagation delays — the paper notes that competing chains can arise inadvertently through network outages — and under non-stationary mining power remain outside the analysis.

  • Implementation and empirical testing. Since the protocol needs no change to Bitcoin's consensus mechanism, a natural next step is deploying or simulating the I-switch rule and measuring its effect on fork rates and stale blocks, alongside the empirical selfish-mining detection literature the paper cites.

Target Audience

Researchers in algorithmic game theory, mechanism design, and blockchain security; economists working on cryptoasset market structure (the paper situates itself in the literature of Biais et al. 2019, Schilling and Uhlig 2019, Leshno and Strack 2020, Huberman et al. 2021, Pagnotta 2022, and Bahrani and Weinberg 2024); and protocol engineers and mining-pool operators who need a rigorous, implementation-compatible answer to whether proportional-reward, single-chain mining can be incentive compatible. Readers should be comfortable with formal game-theoretic definitions; the proof's combinatorial accounting is the most demanding part.

Authors’ abstract

Cryptocurrencies such as Bitcoin are defined by protocols that specify how participants record transactions and create new currency units. These protocols are not enforced by law or any central entity and instead are intended to be incentive compatible. However, the Bitcoin mining protocol proposed by Nakamoto (2008) and implemented in practice is known not to constitute an equilibrium (Eyal and Sirer, 2018). This leaves open the question of whether the decentralized outcome…

The full abstract on arXiv
Read the original paper