Ultimate fault tolerance for exponential delay model
Prove the conjectured ultimate fault tolerance threshold for Nakamoto consensus under the exponential network delay model, in which honest block propagation delays are i.i.d. Exp(μ1) and the mining process is Poisson with rate μ2, namely that transactions become eventually safe (the safety violation probability tends to zero as the confirmation depth k → ∞) if and only if β < (1 − β)/(1 + (1 − β)κ), where β is the adversarial mining power fraction, α = 1 − β is the honest fraction, and κ = μ2/μ1. Establish rigorously that any transaction is safe after waiting sufficiently long when the inequality holds and that no transaction is safe otherwise.
References
We provide our conjecture for the ultimate fault tolerance for exponential delay model in Appendix~\ref{app-c}. The ultimate fault tolerance for exponential delay model is, \begin{align} \beta<\frac{1-\beta}{1+(1-\beta)\kappa}.\label{ult-tol-conj-eqn} \end{align} In other words, any transaction is safe after waiting long enough if ult-tol-conj-eqn is satisfied and no transaction is safe if ult-tol-conj-eqn is not satisfied. We do not provide any proof for the observation above as it is out of the scope for this paper and requires another model to rigorously prove it.
The queue proof uses the strict inequality $\beta_c\lambda\tau>1$, rather than treating equality of average supply and consumption as sufficient for an infinite busy period. The sufficient condition $\lambda_a<\rho$ is equivalent to $\beta<\beta_{\mathrm{pa}}(\lambda\Delta)$, and $\lambda_a\tau<1$ is equivalent to $\beta<1/(\lambda\tau)$ when $\tau>0$. The stock and permanence theorems therefore give security strictly below the displayed upper boundary. The thresholds match in the fully decentralized regime. We do not settle behavior at equality.