Tight Consistency Bounds for Bitcoin

Abstract

We establish the optimal security threshold for the Bitcoin protocol in terms of adversarial hashing power, honest hashing power, and network delays. Specifically, we prove that the protocol is secure if [ra < 1/Δ0 + 1/rh,,] where rh is the expected number of honest proof-of-work successes in unit time, ra is the expected number of adversarial successes, and no message is delayed by more than Δ0 time units. In this regime, the protocol guarantees consistency and liveness with exponentially decaying failure probabilities. Outside this region, the simple private chain attack prevents consensus. Our analysis immediately applies to any Nakamoto-style proof-of-work protocol; in the full version of this paper we also present the adaptations needed to apply it in the proof-of-stake setting, establishing a similar threshold there.