mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
55 lines
2.2 KiB
TeX
55 lines
2.2 KiB
TeX
\section{Safety Under Adaptive Thresholding}
|
|
|
|
\label{sec:safety}
|
|
|
|
\begin{theorem}[Ray Safety]
|
|
\label{thm:ray-safety}
|
|
The Ray protocol with adaptive threshold $\beta^*(r)$ achieves safety probability
|
|
$\leq 2^{-128} + \delta$ where $\delta$ is the confidence parameter of the adversary
|
|
estimator.
|
|
\end{theorem}
|
|
|
|
\begin{proof}
|
|
Define event $A$: the adversary estimator correctly bounds the true fraction
|
|
($\hat{\gamma}^+ \geq \gamma$). By Lemma~\ref{lem:conservative}, $\Pr[A] \geq 1 - \delta$.
|
|
|
|
Conditioned on $A$, the threshold $\beta^*$ is sufficient, and by Theorem~\ref{thm:adaptive-safety},
|
|
the safety violation probability is $\leq 2^{-128}$.
|
|
|
|
Conditioned on $\overline{A}$ (estimator underestimates), the threshold may be too low,
|
|
but the minimum threshold $\beta_{\min} \geq 3$ still provides base safety of
|
|
$(q_0/q_1)^3 \leq 2^{-69}$ for typical parameters (generous upper bound).
|
|
|
|
By the law of total probability:
|
|
\[
|
|
\Pr[\text{safety violation}] \leq (1 - \delta) \cdot 2^{-128} + \delta \cdot 2^{-69}
|
|
\]
|
|
For $\delta = 2^{-60}$, this is $\leq 2^{-128} + 2^{-129} < 2^{-127}$, negligible.
|
|
\end{proof}
|
|
|
|
\subsection{Light-Speed Bound}
|
|
|
|
\begin{definition}[Light-Speed Confirmation]
|
|
The minimum possible confirmation latency for a consensus protocol with round time
|
|
$\Delta_r$ is $\beta_{\min} \cdot \Delta_r$, achievable only when the network has no
|
|
adversarial interference and all validators agree immediately.
|
|
\end{definition}
|
|
|
|
\begin{proposition}[Ray Achieves Light-Speed]
|
|
\label{prop:lightspeed}
|
|
Under zero Byzantine faults, the Ray protocol with $\beta_{\min} = 3$ achieves
|
|
confirmation in $3 \cdot \Delta_r$ time, matching the light-speed bound.
|
|
\end{proposition}
|
|
|
|
\begin{proof}
|
|
With $\gamma = 0$, the confidence velocity $\nu_c = 1$ (supermajority in every round
|
|
since all validators agree). The adaptive threshold becomes $\beta^* = \max(3, \lceil \lambda / 1 \rceil) = 3$
|
|
since $\lambda = 128 \ln 2 / \ln(q_1/q_0)$ and with $\gamma = 0$, $q_1 \approx 1$ and
|
|
$q_0 \approx 0$, making $\lambda$ very small (the ratio $q_1/q_0$ is astronomically large).
|
|
Thus finalization occurs after 3 rounds.
|
|
\end{proof}
|
|
|
|
%% -----------------------------------------------------------------------
|
|
%% 6. EVALUATION
|
|
%% -----------------------------------------------------------------------
|