The class of problems solvable by a probabilistic polynomial-time verifier who can exchange a polynomial number of messages with two competing, computationally-unbounded provers -- one trying to convince the verifier that the answer is "yes," the other that the answer is "no." Note that the verifier can hide information from the provers. Public-coin RG amounts to SAPTIME, which equals PSPACE [Pap83].
RG is in EXP relative to any oracle [KM92]; they are equal, unrelativized [FK97b].
Same as RG, except that now the verifier is a BQP machine, and can exchange polynomially many quantum messages with the competing provers.
The two provers are computationally unbounded, but must obey the laws of quantum mechanics. Also, we can assume without loss of generality that the provers share no entanglement, because adversaries gain no advantage by sharing information.
Defined in [Gut05], where it was also shown that QRG is contained in NEXP ∩ coNEXP.
QRG trivially contains RG (and hence EXP), as well as SQG.
QRG is contained in EXP [GW07]. Hence QRG = RG = EXP and finding optimal strategies for zero-sum quantum games is no harder than finding optimal strategies for zero-sum classical games.
Same as RG, except that now the verifier exchanges exactly k messages with each prover where k is a polynomial-bounded function of the input length. Messages are exchanged in parallel. By definition, RG(poly) = RG. See also RG(1) and RG(2).
Other than trivial bounds, very little is known of RG(k) for intermediate values of k. For example, does RG(k) = PSPACE for each constant ?
Same as RG, except that now the verifier can exchange only two messages with each prover. Messages are exchanged in parallel, so the verifier cannot process the answer from one prover before preparing the question for the other. See also RG(k).
RG(2) is contained in PSPACE, and they are equal, unrelativized [FK97b].
