Recursive proofs with Nova folding

Nova is a folding scheme, not a SNARK by itself. It takes two instances of the same NP relation and produces a single folded instance whose satisfaction implies both.

Chain that operation step by step and you get incrementally verifiable computation: proving a long trace becomes a sequence of tiny folding steps with no expensive proof at every step. The final fold is closed out with a conventional SNARK only once.

SuperNova generalized this to multiple instruction types, and HyperNova pushed the idea onto CCS constraint systems. Folding is the reason zkVMs that used to need heavy recursion per cycle can now run with a single lightweight prover hot path.

It is the current frontier of zk performance work.