Documentation

Complexity.Crypto.Computational.OneWay.BPP

One-way functions put NP outside BPP #

If one-way functions exist then NP ⊄ BPP Goldreich, Chapter 2: the preimage-prefix problem of a one-way function (Cslib.Crypto.preimagePrefix) is not in BPP (Cslib.Crypto.OneWay.preimagePrefix_notMem_BPP).

A decider with error at most 2 ^ (-(L + 1)) on instances of encoded length L exists for every problem in BPP (Cslib.Complexity.mem_BPP_iff_forall_decidesWithErrorExponent). The randomized search (Cslib.Crypto.randomizedPreimageSearch) samples one string of coins and runs the deterministic bit-by-bit search, answering each query with the decider on a prefix of the coins of the length the decider expects. The search with exact answers asks n fixed queries, each with encoded length at least n; whenever the decider answers these correctly, the randomized search follows the exact one (Cslib.Crypto.preimageSearch_congr) and finds a preimage. Each of them is answered wrongly with probability at most 2 ^ (-(n + 1)), so by a union bound the randomized search inverts with probability at least 1 / 2 (Cslib.Crypto.half_le_succProbability_inversionGame), and it is a uniform strict PPT inverter. Since BPP is closed under polynomial-time reductions (Cslib.Complexity.PolyTimeReducible.mem_BPP), the Cook–Levin theorem puts SAT outside BPP as well (Cslib.Crypto.OneWay.SAT_notMem_BPP).

Main definitions #

Main results #

References #

The coins read by the randomized search for a preimage of length q.1 of q.2: enough for the decider on every query, whose prefix has at most q.1 bits.

Instances For

    The randomized search: sample searchCoinLength p (n, y) coins once, and run the bit-by-bit search, answering every query with V on the prefix of the coins of the length p (L) that V expects on an instance of encoded length L.

    Instances For

      The randomized search inverts with probability at least 1 / 2, for a decider of the preimage-prefix problem with error at most 2 ^ (-(L + 1)) on instances of encoded length L. The union bound runs over the n queries of the exact search, each of encoded length at least n.

      Inversion from BPP. If the preimage-prefix problem of f is in BPP, a uniform strict PPT inverter, told the length of the input, inverts f with probability at least 1 / 2 at every length.

      One-way functions have preimage-prefix problems outside BPP.

      One-way functions imply NP ⊄ BPP, for instances of the preimage-prefix problem.

      One-way functions imply SAT ∉ BPP: by the Cook–Levin theorem the preimage-prefix problem reduces to SAT, and BPP is closed under polynomial-time reductions.