Paper 2024/108

Some Improvements for the PIOP for ZeroCheck

Abstract

Most multivariate proof systems require, at some point, an algebraic check against the rows of the trace. One popular protocol for this is known as zerocheck which is a sumcheck based protocol which proves a constraint function is zero over the $n$-dimensional Boolean hypercube. One of the drawbacks of running zerocheck over a small field, is that it usually involves a large number of evaluations of the constraint polynomial over a cryptographically large extension field $\mathbb{G}$. We describe several improvements to the zerocheck protocol over a small field $$ which both reduce the number of constraint evaluations the prover needs to perform and shifts some computation from the extension field back into the base field $$. Specifically, for a table of length $$, integer parameter $$ and constraint function $$ of degree $$ with evaluation costs $$ we show the protocol can be performed with prover cost roughly $$ To check the proof, the verifier needs to perform a single interpolation of degree $$ and $$ interpolations of degree $$. Thus varying $$ allows for a tradeoff between prover and verifier cost.