mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zooqueen
a10bf7bbc6
- canonical lux-corona-pq paper: rank-7 Module-LWE (A in R_q^{8x7}, q=48-bit), Module-SIS
- scrub residual Corona Ring-LWE/R-LWE labels across ethos-of-lux, pqsafe-il5,
lux-whitepaper, lux-quasar-composition, lp-020; Corona<->Pulsar = parameter regime,
both Module-LWE (not rank-1-vs-rank-k)
- LLM.md: drop stale 'Corona as rank-1' wobble note; align to code/Ringtail params
- rebuild affected PDFs
100 lines
4.7 KiB
TeX
100 lines
4.7 KiB
TeX
\section{Implementation}
|
|
|
|
\label{sec:implementation}
|
|
%% ============================================================================
|
|
|
|
Quasar 3.0 ships in the \texttt{github.com/luxfi/consensus} package on
|
|
25 December 2025. Implementation highlights:
|
|
|
|
\begin{itemize}[leftmargin=2em,nosep]
|
|
\item \textbf{Language}: Go 1.24. Single binary, explicit error handling,
|
|
no runtime panics on the consensus path.
|
|
\item \textbf{Wire protocol}: ZAP (\texttt{github.com/luxfi/zap}), a
|
|
zero-allocation binary protocol with optional PQ-TLS~1.3
|
|
(X25519MLKEM768) transport encryption~\cite{lux-zap}.
|
|
\item \textbf{Storage}: SQLite WAL mode via ZapDB for consensus state.
|
|
Epoch-based pruning removes state older than 6 epochs.
|
|
\item \textbf{Cryptography}: \texttt{luxfi/crypto/bls} for BLS12-381,
|
|
\texttt{luxfi/crypto/mldsa} for ML-DSA-44/65/87,
|
|
\texttt{luxfi/corona} for Module-LWE threshold signing,
|
|
\texttt{luxfi/zk} for the Z-Chain Groth16 prover/verifier.
|
|
\item \textbf{QuasarGPU adapter}: LP-132~\cite{lp132-quasargpu} provides
|
|
a GPU-native verification substrate that schedules BLS, Corona,
|
|
and Groth16 verifies onto the same kernel stream as EVM execution.
|
|
Verification consumes no incremental host-CPU budget on the hot
|
|
path.
|
|
\item \textbf{Three-lane signing}: \texttt{TripleSignRound1}
|
|
(\texttt{consensus/protocol/quasar/quasar.go}) dispatches
|
|
BLS, Corona, and ML-DSA signing in separate goroutines.
|
|
$\sigma_i^{\mathrm{ML}}$ is submitted to Z-Chain rather than
|
|
gossipped, so the mainnet message size remains
|
|
$\mathcal{O}(96 + 56\text{KB})$ rather than
|
|
$\mathcal{O}(N \cdot 3{,}309)$.
|
|
\item \textbf{Fail-closed Corona}: if Q-Chain's threshold round fails
|
|
(insufficient shares, rejection sampling failure, network
|
|
partition during DKG), the protocol is specified to fall back to
|
|
Mode 2 (BLS + MLDSAGroth16). The current wire layer
|
|
(\texttt{PolicyQuantum}) requires Corona by default
|
|
(\texttt{requireRT = true}); the Mode\,2 fallback is governed by
|
|
this flag and an out-of-band operator decision rather than an
|
|
automatic in-protocol degradation. It does not degrade to
|
|
BLS-only without explicit operator action.
|
|
\item \textbf{Z-Chain prover degradation}: if the Z-Chain prover misses
|
|
$T_{\mathrm{ZK}}$, the proposer broadcasts the raw
|
|
$\geq t$ ML-DSA-65 signatures (the post-quantum ``signal flare''
|
|
fallback referenced in Theorem~\ref{thm:pq-safety}) instead of the
|
|
Groth16 proof.
|
|
\item \textbf{FPC integration}: the Wave FPC threshold selector
|
|
(\texttt{protocol/wave/fpc/fpc.go}) uses SHA-256-based PRF for
|
|
deterministic phase-dependent threshold selection.
|
|
\item \textbf{DAG finalization}: pre-allocated ring buffers for vote
|
|
counting, frontier-set reuse for DAG traversal, zero GC pressure
|
|
during consensus rounds. Finalization check 13.75\,ns at depth 10,
|
|
113\,ns at depth 100 (sub-linear via frontier caching).
|
|
\end{itemize}
|
|
|
|
\subsection{Module Structure}
|
|
|
|
\begin{table}[ht]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{@{}ll@{}}
|
|
\toprule
|
|
\textbf{Package} & \textbf{Function} \\
|
|
\midrule
|
|
\texttt{protocol/photon/} & Proposal emission \\
|
|
\texttt{protocol/wave/} & FPC voting, confidence tracking \\
|
|
\texttt{protocol/wave/fpc/} & PRF-based threshold selector \\
|
|
\texttt{protocol/focus/} & Certificate-subject derivation \\
|
|
\texttt{protocol/ray/} & Three-lane signing dispatch \\
|
|
\texttt{protocol/quasar/} & QuasarCert construction \& verification \\
|
|
\texttt{protocol/nova/} & Linear-chain consensus mode \\
|
|
\texttt{protocol/nebula/} & DAG consensus mode \\
|
|
\texttt{protocol/prism/} & DAG geometry and cutting \\
|
|
\texttt{protocol/flare/} & Nebula proposer (DAG variant of Photon) \\
|
|
\texttt{protocol/horizon/} & Nebula DAG voting \\
|
|
\texttt{protocol/field/} & Nebula finalization \\
|
|
\texttt{core/dag/} & Horizon, flare, frontier algorithms \\
|
|
\texttt{engine/chain/} & Linear-chain engine \\
|
|
\texttt{engine/dag/} & DAG engine \\
|
|
\texttt{engine/pq/} & Post-Quasar certificate engine \\
|
|
\texttt{types/} & Bag, config, consensus types \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Package structure of the Quasar 3.0 implementation.}
|
|
\label{tab:packages}
|
|
\end{table}
|
|
|
|
\subsection{Configuration}
|
|
|
|
Quasar 3.0 auto-configures consensus parameters from validator count:
|
|
\texttt{GetConfig(nodeCount)} computes $k$, $\alpha$, and $\beta$ to
|
|
satisfy safety and liveness for the given $n$. Three presets:
|
|
\texttt{Config.SingleValidator()}, \texttt{Config.Testnet()}, and
|
|
\texttt{Config.Mainnet()}. The 25 December 2025 mainnet activates with
|
|
\texttt{Config.Mainnet()} at $n=21$ and Mode~$\mu = 3$ (Quantum, all
|
|
three lanes).
|
|
|
|
|
|
%% ============================================================================
|