Paper 2024/806

Resettable Statistical Zero-Knowledge for NP

Abstract

Resettable statistical zero-knowledge [Garg--Ostrovsky--Visconti--Wadia, TCC 2012] is a strong privacy notion that guarantees statistical zero-knowledge even when the prover uses the same randomness in multiple proofs. In this paper, we show an equivalence of resettable statistical zero-knowledge arguments for $NP$ and witness encryption schemes for $NP$. - Positive result: For any $NP$ language $L$, a resettable statistical zero-knowledge argument for $L$ can be constructed from a witness encryption scheme for $L$ under the assumption of the existence of one-way functions. - Negative result: The existence of even resettable statistical witness-indistinguishable arguments for $$ imply the existence of witness encryption schemes for $$ under the assumption of the existence of one-way functions. The positive result is obtained by naturally extending existing techniques (and is likely to be already well-known among experts). The negative result is our main technical contribution. To explore workarounds for the negative result, we also consider resettable security in a model where the honest party's randomness is only reused with fixed inputs. We show that resettable statistically hiding commitment schemes are impossible even in this model.