BPP (complexity)

BPP (complexity)
Name	BPP
Full name	Bounded-error Probabilistic Polynomial time
Type	Complexity class
Introduced	1970s
Notable contributors	Richard M. Karp, Michael O. Rabin, Leslie Valiant, Noam Nisan, Alexander Russell, Salil Vadhan, Oded Goldreich, Madhu Sudan, Ravi Kannan, Joan Feigenbaum, Adi Shamir, Shafi Goldwasser, Silvio Micali, Leonid Levin, Stephen Cook, Donald Knuth, Alan Turing, Alonzo Church, John von Neumann, Claude Shannon, Paul Erdős, Kurt Gödel, Andrew Yao, Ran Raz, Avi Wigderson, László Babai, Oded Goldreich, Janos Simon, Sanjeev Arora, Dana Angluin, Oded Regev, Moses Charikar, Sanjeev Arora, Michael Sipser, Neil Immerman, Eric Allender, Egon Balas, Robert Tarjan, Richard Lipton, Eugene Lawler, Jack Edmonds, Leonard Adleman, Moses Schönhage

BPP (complexity) is the class of decision problems solvable by probabilistic Turing machines in polynomial time with bounded two-sided error. It formalizes efficient randomized computation and sits centrally among classes like P (complexity), NP (complexity), PSPACE, ZPP, RP (complexity), co-RP (complexity), AM (complexity), MA (complexity), IP (complexity), EXP (complexity), NEXP (complexity). BPP has influenced theoretical work by figures associated with Princeton University, MIT, Bell Labs, IBM, Microsoft Research, Stanford University and Harvard University.

Definition and Formal Definition

Formally, a language L is in BPP if there exists a polynomial-time probabilistic Turing machine M and a polynomial p such that for every input x of length n: if x ∈ L then Pr[M(x) accepts] ≥ 2/3, and if x ∉ L then Pr[M(x) accepts] ≤ 1/3, where probabilities are over at most 2^{p(n)} coin flips used by M. Alternate but equivalent formalizations use uniform families of Boolean circuits with random input bits, or randomized RAM models; these equivalences are standard in expositions from Stanford University and University of California, Berkeley texts by authors like Michael Sipser and Richard Lipton. The constants 2/3 and 1/3 can be replaced by any constants r, s with r − s ≥ 1/poly(n), preserving class membership; proofs of equivalence are credited in literature associated with University of Chicago and Carnegie Mellon University.

Relationships to Other Complexity Classes

BPP contains P (complexity) and is contained in P/poly (complexity), and is known to be a subset of Σ2P under certain relativized constructions studied by researchers at Princeton University and Harvard University. Containment relations have been explored relative to oracles associated with work by Bennett and Gill, and interactive proof connections link BPP to AM (complexity) and MA (complexity), with separations or collapses discussed by investigators at MIT and UC Berkeley. Results by Nisan and Wigderson relate BPP to deterministic classes via pseudorandom generators based on hardness assumptions attributed to problems studied by Richard Karp and Leslie Valiant. Complexity-theoretic consequences connecting BPP to NP (complexity), co-NP, and classes like PH (polynomial hierarchy) are central open directions investigated at Institute for Advanced Study and Microsoft Research.

Algorithmic Techniques and Randomized Algorithms

Typical algorithms in BPP include randomized primality tests inspired by work at Bell Labs and later refined at MIT and Princeton University, randomized polynomial identity tests developed by scholars at Stanford University and Tel Aviv University, and hashing-based randomized algorithms for approximate counting and streaming studied at Carnegie Mellon University and EPFL. Techniques exploited include coin-flip simulation, amplification through repetition, pairwise and k-wise independent hashing linked to results from Paul Erdős and Alon, and constructions of pseudorandom generators motivated by cryptographic research at RSA Laboratories and Bell Labs. Practical randomized algorithms connected to BPP have applications in computational tasks explored by teams at Google and Amazon.

Error Reduction and Amplification

Error reduction in BPP is achieved via repetition and majority vote, with Chernoff bounds and tail inequalities originating in work by Herman Chernoff and Andrey Kolmogorov providing formal guarantees; these techniques were elaborated in texts associated with Princeton University and Columbia University. Amplification reduces two-sided error to exponentially small values with only polynomial overhead in time. Stronger amplification methods use pairwise independence and expander graphs, the latter developed in studies by Margulis, Lubotzky, Phillips, and Sarnak and applied in constructions by researchers affiliated with Harvard University and Microsoft Research.

Completeness and BPP-Complete Problems

Unlike NP (complexity), BPP is not known to have natural complete languages under standard polynomial-time many-one reductions; BPP-completeness is subtle because randomized reductions can trivialize completeness. Several candidates and relativized complete sets have been proposed in oracle models studied at IBM Research and UC San Diego, but consensus favors the view that BPP likely lacks complete problems analogous to SAT (Boolean satisfiability problem) for NP unless unlikely collapses occur in hierarchies examined by Stephen Cook and Leonid Levin. Alternative completeness notions consider promise problems and search variants, topics explored by groups at UCLA and University of Washington.

Derandomization and Open Questions

Central open questions ask whether randomness adds power: is BPP equal to P (complexity)? Foundations for derandomization connect to hardness assumptions about functions studied by Odlyzko, Ajtai, Impagliazzo, Wigderson, and Nisan; pseudorandom generator constructions linking worst-case hardness to average-case hardness are active research conducted at Princeton University, MIT, and Microsoft Research. Relativized separations and collapses involving oracles like those in work by Bennett and Gill and Baker, Gill, and Solovay demonstrate limitations of relativizing techniques; nonrelativizing methods such as interactive proofs and algebraic geometry techniques have been pivotal in complexity breakthroughs at University of California, Berkeley and IAS. The question whether BPP = P remains one of the most prominent unresolved problems in theoretical computer science, with implications echoed in cryptography communities at RSA Laboratories and NIST.

Category:Computational complexity theory