Non-transitive transfer of confidence: A perfect zero-knowledge interactive protocol for SAT and beyond

A perfect zero-knowledge interactive proof is a protocol by which Alice can convince Bob of the truth of some theorem in a way that yields no information as to how the proof might proceed (in the sense of Shannon's information theory). We give a general technique for achieving this goal for any problem in NP (and beyond). The fact that our protocol is perfect zero-knowledge does not depend on unproved cryptographic assumptions. Furthermore, our protocol is powerful enough to allow Alice to convince Bob of theorems for which she does not even have a proof. Whenever Alice can convince herself probabilistically of a theorem, perhaps thanks to her knowledge of some trap-door information, she can convince Bob as well without compromising the trap-door in any way. This results in a non-transitive transfer of confidence from Alice to Bob, because Bob will not be able to subsequently convince someone else that the theorem is true. Our protocol is dual to those of GMW1, BC.