|
|
\documentclass{article} \usepackage[utf8]{inputenc} \usepackage{amsfonts} \usepackage{amsthm} \usepackage{amsmath} \usepackage{mathtools} \usepackage{enumerate} \usepackage{hyperref} \usepackage{xcolor}
% prevent warnings of underfull \hbox:
\usepackage{etoolbox} \apptocmd{\sloppy}{\hbadness 4000\relax}{}{}
\theoremstyle{definition} \newtheorem{definition}{Def}[section] \newtheorem{theorem}[definition]{Thm}
% custom lemma environment to set custom numbers
\newtheorem{innerlemma}{Lemma} \newenvironment{lemma}[1] {\renewcommand\theinnerlemma{#1}\innerlemma} {\endinnerlemma}
\title{Notes on Nova} \author{arnaucube} \date{February 2023}
\begin{document}
\maketitle
\begin{abstract} Notes taken while reading Nova \cite{cryptoeprint:2021/370} paper.
Usually while reading papers I take handwritten notes, this document contains some of them re-written to $LaTeX$.
The notes are not complete, don't include all the steps neither all the proofs. \end{abstract}
\tableofcontents
\section{Folding Scheme for Committed Relaxed R1CS}
\subsection{R1CS modification}
Want: merge 2 instances of R1CS with the same matrices into a single one. Each instance has $z_i = (W_i,~ x_i)$ (public witness, private values resp.).
\paragraph{traditional R1CS} Merged instance with $z=z_1 + r z_2$, for rand $r$. But, since R1CS is not linear $\longrightarrow$ can not apply.
eg. \begin{align*} Az \circ Bz &= A(z_1 + r z_2) \circ B (z_1 + r z_2)\\ &= A z_1 \circ B z_1 + r(A z_1 \circ B z_2 + A z_2 \circ B z_1) + r^2 (A z_2 \circ B z_2)\\ &\neq Cz \end{align*}
$\longrightarrow$ introduce error vector $E \in \mathbb{F}^m$, which absorbs the cross-temrs generated by folding.
$\longrightarrow$ introduce scalar $u$, which absorbs an extra factor of $r$ in $C z_1 + r^2 C z_2$ and in $z=(W, x, 1+r\cdot 1)$.
\paragraph{Relaxed R1CS} \begin{align*} &u=u_1+r u_2\\ &E=E_1 + r (A z_1 \circ B z_2 + A z_2 \circ B z_1 - u_1 C z_2 - u_2 C z_1) + r^2 E_2\\ &Az \circ Bz = uCz + E,~~ with~ z=(W,~x,~u) \end{align*} where R1CS set $E=0,~u=1$.
\begin{align*} Az \circ Bz &= A z_1 \circ B z_1 + r(A z_1 \circ B z_2 + A z_2 \circ B z_1) + r^2 (A z_2 \circ B z_2)\\ &= (u_1 C z_1 + E_1) + r (A z_1 \circ B z_2 + A z_2 \circ B z_1) + r^2 (u_2 C z_2 + E_2)\\ &= u_1 C z_1 + \underbrace{E_1 + r(A z_1 \circ B z_2 + A z_2 \circ B z_1) + r^2 E_2}_\text{E} + r^1 u_2 C z_2\\ &= u_1 C z_1 + r^2 u_2 C z_2 + E\\ &= (u_1 + r u_2) \cdot C \cdot (z_1 + r z_2) + E\\ &= uCz + E \end{align*}
For R1CS matrices $(A,~B,~C)$, the folded witness $W$ is a satisfying witness for the folded instance $(E,~u,~x)$.
\vspace{20px} Problem: not non-trivial, and not zero-knowledge. Solution: use polynomial commitment with hiding, binding, succintness and additively homomorphic properties.
\paragraph{Committed Relaxed R1CS} Instance for a Committed Relaxed R1CS\\ $(\overline{E}, u, \overline{W}, x)$, satisfyied by a witness $(E, r_E, W, r_W)$ such that \begin{align*} &\overline{E} = Com(E, r_E)\\ &\overline{W} = Com(E, r_W)\\ &Az \circ Bz = uCz+E,~~ where~z=(W, x, u) \end{align*}
\subsection{Folding protocol}
V and P take two \emph{committed relaxed R1CS} instances \begin{align*} \varphi_1&=(\overline{E}_1, u_1, \overline{W}_1, x_1)\\ \varphi_2&=(\overline{E}_2, u_2, \overline{W}_2, x_2) \end{align*}
P additionally takes witnesses to both instances \begin{align*} (E_1, r_{E_1}, W_1, r_{W_1})\\ (E_2, r_{E_2}, W_2, r_{W_2}) \end{align*}
Let $Z_1 = (W_1, x_1, u_1)$ and $Z_2 = (W_2, x_2, u_2)$.
% \paragraph{Protocol}
\begin{enumerate} \item P send $\overline{T} = Com(T, r_T)$,\\ where $T=A z_1 \circ B z_1 + A z_2 \circ B z_2 - u_1 C z_2 - u_2 C z_2$\\ and rand $r_T \in \mathbb{F}$ \item V sample random challenge $r \in \mathbb{F}$ \item V, P output the folded instance $\varphi = (\overline{E}, u, \overline{W}, x)$ \begin{align*} &\overline{E}=\overline{E}_1 + r \overline{T} + r^2 \overline{E}_2\\ &u = u_1 + r u_2\\ &\overline{W} = \overline{W}_1 + r \overline{W}_2\\ &x = x_1 + r x_2 \end{align*} \item P outputs the folded witness $(E, r_E, W, r_W)$ \begin{align*} &E = E_1 + r T + r^2 E_2\\ &r_E = r_{E_1} + r \cdot r_T + r^2 r_{E_2}\\ &W=W_1 + r W_2\\ &r_W = r_{W_1} + r \cdot r_{W_2} \end{align*} \end{enumerate}
P uses a zkSNARK showing that knows the valid witness $(E, r_E, W, r_W)$ for the committed relaxed R1CS without revealing its value. Then, vie Fiat-Shamir transform we achieve non-interactivity.
\section{IVC proofs} \textbf{WIP}
\bibliography{paper-notes.bib} \bibliographystyle{unsrt}
\end{document}
|