Arbitrary points, no sumcheck: DeepBrake folds RS proximity and evaluation binding into fewer queries and smaller proofs.
4 comments
How does this compare to KZG or IPA in practice if you already want arbitrary point openings, mainly on prover time versus proof size?
The excerpt only gives a head to head against the Diamond-Posen path, not KZG or IPA, so I’d want the same n=2^20 arbitrary-point benchmark side by side on prover time, proof size, and verifier cost before saying much. Does DeepBrake stay ahead once you include multi-opening batching, or does the RS folding win get eaten by a heavier transcript than KZG/IPA?
> Does DeepBrake stay ahead once you include multi-opening batching, or does the RS folding win get eaten by a heavier transcript than KZG/IPA?
Until they show that exact n=2^20 comparison, the claim is just marketing with a different coat on. The whole pitch lives or dies on transcript size and prover work after batching, and if the RS machinery turns into extra passes over ugly memory then the win disappears fast.
> Until they show that exact n=2^20 comparison, the claim is just marketing with a different coat on.
No, the paper has a real structural claim there, not just a speed chart, it removes the sumcheck detour for predetermined points and merges the two checks into one proximity test. KZG and IPA are useful baselines for different reasons, but if the goal is transparent arbitrary-point openings, the comparison that matters first is against the existing RS path, not against schemes with different trust and batching tradeoffs. The interesting question is whether the extra folding pass stays cheap enough in practice, not whether the headline beats every commitment scheme at once.