Thursday, April 18, 2024

Favorite Theorems: Quantum Provers

March Edition

For our next favorite theorem, we look at the surprising power of provers who share entangled bits. If you can prove something to an arbitrarily computable verifier, then two entangled provers can convince a polynomial-time verifier.

MIP* = RE
Zhengfeng Ji, Anand Natarajan, Thomas Vidick, John Wright and Henry Yuen

A quick tour:

  • A powerful prover convincing a skeptical computable deterministic verifier is another way of capturing computably enumerable (traditionally called recursively enumerable or RE). You can convince a verifier that a program halts by giving the halting time, and the verifier can simulate the machine for that many steps.
  • A powerful prover convincing a skeptical polytime deterministic verifier is another way of capturing the class NP, like giving a 3-coloring of a graph that can be easily checked.
  • If you allow the verifier to ask random questions, you can convince the verifier with high confidence that a graph is not colorable, or more generally any PSPACE problem.
  • If you add two separated provers that a skeptical probabilistic verifier can play off each other, the provers can convince the verifier that any problem in NEXP, non-deterministic exponential time.
One of many quantum variations of interactive proofs, MIP* has two provers that cannot communicate but have entangled quantum bits. This change could go either way: 
  • The provers to coordinate their answers better and so they wouldn't convince the verifier for all the languages in NEXP
  • The verifier could ask more complex questions to the provers which they could answer using the entanglement, allowing the provers to convince the verifier for even more complex languages.
Turns out it's the later in a very strong way.

Ito and Vidick showed that you can create a protocol that prevents the provers coordinating better, recovering all problems in NEXP. Natarajan and Wright showed you can ask more questions, showing that provers with entangled bits can convince a verifier of everything in NEEXP, non-deterministic double exponential time (\(2^{2^{n^c}}\)), already a proof too large for the verifier to even point into. The MIP* = RE paper takes that all the way to the computably enumerable sets, all the languages you would get with a classical prover convincing a deterministic verifier unrestricted by time.

3 comments:

  1. Slight typo: "Vidick and Wright" should be "Natarajan and Wright."

    ReplyDelete
  2. Last link should be Natarajan and Wright.

    ReplyDelete