Evaluate proof aggregation schemes

[matthiasgoergens]

Look into all the different ways to aggregate and compress proves. What are the trade-offs we want to make?

Check whether the approach taken in this repository so far is actually in line with Ceno's goals.

Compare what SP1 and Risc0 etc are doing, and what design considerations they had---we might as well learn from them. (Assuming we can find their deliberations. But we can also ask, Risc0 folks have been very communicative when asked in the past.)

[matthiasgoergens]

I'm still learning about how our base VM works, so I can look more into recursive verification. @Jiangkm3 Please jump in as you see fit.

[matthiasgoergens]

See also https://scrollco.slack.com/archives/C07STVAHKPT/p1730169868030399

[Jiangkm3]

We have decided that it is necessary to finish the most rudimentary implementation of the recursive verifier for benchmark testing. We shift our focus to finish that implementation first.

[kunxian-xia]

Try to understand which proof system is optimal for verifying ceno risc-v proof.

• use goldilocks by default in spartan-parallel Feat/Switch Spartan Backend to Use Goldilocks Field #17
• bench: verifying N poseidon permutation over goldilocks field using spartan_parallel #48

[lispc]

conclusion Jan07:
(1) vm vs static: prefer static since it reduces cpu chip proving, which needs control flow (a lot of cycles) (3.5x in sp1 benches) // kunxian
(2) r1cs vs fri: