Research
Selfish Mining in Multi-Attacker Scenarios: An Empirical Evaluation of Nakamoto, Fruitchain, and Strongchain
Selfish mining lets a miner with less than 50% of the hash rate earn more than its fair share by withholding blocks and releasing them strategically to waste honest miners' work.

- arXiv
- 2601.02984
- Published
- 2026-01-06
- Authors
- Martin Perešíni, Tomáš Hladký, Jakub Kubík, Ivan Homoliak
AI summary
Overview
Research area: Blockchain security — Proof-of-Work consensus protocol design and incentive attacks, specifically selfish mining under multi-attacker conditions. Technical level: Intermediate. Readers should know basic PoW mining and fork-choice concepts; the protocol mechanics (fruits, weak headers) are explained well enough to follow without prior expertise. Scope: A stochastic simulation study that builds a unified framework for one or more selfish miners and compares Nakamoto consensus, Fruitchain and Strongchain under identical conditions, verifying published profitability thresholds and deriving new multi-attacker thresholds.
What This Paper Is About
Selfish mining lets a miner with less than 50% of the hash rate earn more than its fair share by withholding blocks and releasing them strategically to waste honest miners' work. Most prior work studies a single attacker on a single protocol, leaving it unclear how well selfish-mining defenses hold up when two or more attackers operate independently. The authors build one simulation framework that models Nakamoto consensus alongside the defenses Fruitchain and Strongchain, then measure the minimum mining power at which selfish mining becomes profitable for 1 to 7 attackers.
Key Contributions
- A unified, reusable simulation framework that models multiple consensus protocols under selfish mining attacks, with source code released on GitHub (fruitchain-sim and smasf repositories).
- Implementation and analysis of Fruitchain and Strongchain inside that framework, enabling direct side-by-side comparison with Nakamoto consensus.
- Extensive multi-attacker experiments (1 to 7 attackers) across values of the propagation factor γ, producing new profitability thresholds — the paper notes that all results marked N/A in the literature were obtained for the first time in this work.
- A cross-protocol comparison establishing a relative order of protection against selfish mining.
Main Findings
- Nakamoto thresholds reproduced: The simulation matches prior literature at γ = 0.5 (25% for one attacker) and γ = 1 (1%), and at γ = 0 (33%), confirming framework validity. For five and seven attackers the results deviate by 1% from prior work (14% simulated vs. 15% reported for five; 11% vs. 12% for seven).
- Two-attacker baseline: Nakamoto reaches the profitable threshold at 21% per attacker, i.e. 42% in total, matching Zhang et al.
- Strongchain is the most resistant protocol measured: Single-attacker threshold of 46% at γ = 0, versus 45% in the original Strongchain paper — a 1% difference the authors attribute to differing simulation parameters or stochasticity. Two attackers require 32% each, up from 21% on Nakamoto (Δ +11).
- Strongchain multi-attacker thresholds: 24% per attacker for three attackers (Δ +5), 17% for five (Δ +3), and 13% for seven (Δ +2).
- Fruitchain raises the single-attacker bar to about 38% to 39%: With the longest-chain rule, the threshold is 38% at γ = 0 (Δ +5), 38% at γ = 0.5 (Δ +13), and 38% at γ = 1 (Δ +37). Its multi-attacker values in the comparison table are 38% (two attackers), 25% (three), 17% (five) and 13% (seven).
- γ barely matters for the defenses: For Fruitchain, varying γ between 0 and 1 changes results by only about 1%, because the override action rarely occurs given fine-grained branch difficulty that also counts fruits. The authors note the same phenomenon occurs in Strongchain.
- More attackers, lower individual thresholds — but the defenses stay ahead: Across all protocols, the mining power each attacker needs falls as the number of independent attackers grows, yet Strongchain and Fruitchain keep thresholds considerably higher than Nakamoto.
- Reward structure shifts Fruitchain's behavior: Reducing the block reward from 50% to 9.09% (making fruits more valuable) widened the gap further and shifted the point of highest negative difference to a higher mining power.
- Attacker 2 suppresses Attacker 1's revenue but lowers his threshold: Increasing a rival attacker's power reduces Attacker 1's relative revenue while decreasing the power at which Attacker 1 becomes profitable — until a dominant rival can suppress that revenue entirely (the right-hand drop in the plots).
Methodology in Plain English
The authors replace real mining with a centralized "simulation manager" that picks the leader of each round at random in proportion to mining power; that leader, honest or selfish, produces the protocol's artifacts (blocks, fruits, weak headers or strong blocks). Selfish miners follow the Eyal–Sirer strategy — override, adopt, match, wait — adapted to each protocol's fork-choice rule, and other attackers can respond once an action becomes public. Network effects are collapsed into the propagation factor γ, the share of honest hash power that mines on the attacker's branch when chains tie; γ = 0 favors honest miners, γ = 0.5 is equal, γ = 1 favors the attacker. Multi-attacker runs fix γ = 0.5 so no single attacker is privileged, and attackers are assumed rational, independent and non-colluding. Each configuration was run 5 times with 100,000 rounds per run and results averaged; more than a thousand experiments were typically performed per protocol. A threshold is the smallest mining power α at which mean relative revenue reaches or exceeds fair share, with linear interpolation between straddling grid points and a 95% bootstrap confidence interval confirming the crossing. Transaction fees and accidental forks are excluded, and propagation/validation delays are treated as constant. The framework was built in Python 3.8 and TypeScript.
Why This Matters
Selfish mining is a profitability attack, not a correctness break: it silently transfers revenue to attackers and erodes the assumption that majority hash power is the only security requirement. This paper matters because it puts proposed mitigations on the same measuring stick for the first time under multi-attacker conditions, and it shows that defenses reported as effective for a lone attacker can degrade as attackers multiply — 46% for one attacker on Strongchain versus 13% for seven.
Real-world applications:
- Mining pool operators and large miners can use the thresholds to judge at what share of network hash rate withholding blocks becomes rational for them or for a competitor.
- Protocol designers and client teams evaluating fork-choice tweaks or inclusive reward schemes get comparable, reproducible numbers on defense strength rather than isolated claims.
- Exchanges, custodians and payment processors assessing confirmation-depth policy can factor in how low the profitable attack threshold drops when several miners act independently.
- Blockchain auditors and researchers get an open-source framework to plug in new protocols or attack strategies and cross-compare against already modeled ones.
Industry relevance: the result that γ has little effect on Strongchain and Fruitchain means their protection does not depend on optimistic connectivity assumptions, which is a more practical security claim than thresholds that only hold when the attacker is poorly connected.
Future Directions
- Extending the framework to other protocols the authors name, such as Subchain, and to alternative attack strategies such as stubborn mining.
- Adding transaction fees and accidental forks, both explicitly excluded here, and modeling block propagation time and network topology instead of collapsing them into γ.
- A deeper investigation of the economics of differing reward structures, particularly the fruit-versus-block reward split in Fruitchain, where the authors already saw block reward changes alter results.
- Applying the framework to DAG-based protocols such as PHANTOM/GOSTDAG and Prism under the assumption of transaction fees.
Target Audience
Blockchain protocol researchers and graduate students working on consensus security and incentive analysis; PoW protocol designers and client developers assessing selfish-mining resistance; mining pool operators and large miners interested in profitability thresholds; and security analysts or auditors at exchanges and custodians who need quantitative evidence on how much hash power a profitable withholding attack requires under different consensus designs.
Authors’ abstract
The aim of this work is to enhance blockchain security by deepening the understanding of selfish mining attacks in various consensus protocols, especially the ones that have the potential to mitigate selfish mining. Previous research was mainly focused on a particular protocol with a single selfish miner, while only limited studies have been conducted on two or more attackers. To address this gap, we proposed a stochastic simulation framework that enables analysis of selfish…
The full abstract on arXiv