270 likes | 401 Views
Has There Been Progress on the P vs. NP Question?. Scott Aaronson (MIT). P vs. NP : I Assume You’ve Heard of It. Frank Wilczek (Physics Nobel 2004) was recently asked: “If you could ask a superintelligent alien one yes-or-no question, what would it be?”.
E N D
Has There Been Progress on the P vs. NP Question? Scott Aaronson (MIT)
P vs. NP: I Assume You’ve Heard of It Frank Wilczek (Physics Nobel 2004) was recently asked: “If you could ask a superintelligent alien one yes-or-no question, what would it be?” His response: “P vs. NP. That basically contains all the other questions, doesn’t it?”
From the standpoint of P vs. NP, the last 50 years of complexity theory have taken us around in circles and been a complete waste of time. A Depressing Possibility… This talk: We might be nowhere close to a proof, but at least the depressing possibility doesn’t hold! We’ve found (and continue to find) nontrivial insights that will play a role in the solution, assuming there is one. The end is not in sight, but we’re not at the beginning either.
Achievement 1: Increased Confidence That P vs. NP Was The Right Question To Ask
NP-completeness Goal: Lift this box What NP-completeness accomplished… PCP SAT
A half-century of speculation about alternative computational models has taken us only slightly beyond P The Unreasonable Robustness of P “But can’t soap bubbles solve the Minimum Steiner Tree problem in an instant, rendering P vs. NP irrelevant?” Would-be PNP provers: don’t get discouraged! Likewise for spin glasses, folding proteins, DNA computers, analog computers…
Randomness:P = BPPunder plausible assumptions(indeed, assumptions that will probably have to be proved beforePNP)[NW94], [IW97], … Nonuniform Algorithms:P/poly is “almost the same as P,” for P vs. NP purposes [Karp-Lipton 82] Quantum Computing:BQP probably is larger than P. But even NPBQP doesn’t look like a “radically” different conjecture from PNP Quantum Gravity? What little we know is consistent with BQP being the “end of the line”E.g., topological quantum field theories can be simulated in BQP[FKLW’02] More serious challenges to the Polynomial-Time Church-Turing Thesis have also been addressed…
Achievement 2: Half a Century of Experience with Efficient Computation, Increasing One’s Confidence That PNP
Trivial Problems #P Problems with Miraculous Cancellation Determinant, counting planar perfect matchings, 3-regular-planar-mod-7-SAT… P Dynamic Programming #P Problems with Miraculous Positivity Test Matching, Littlewood-Richardson coefficients… Linear Programming Semidefinite/Convex Programming Polynomial Identity Testing (assuming P=BPP) Matrix Group Membership(modulo discrete log) Polynomial Factoring
We now have a fairly impressive “statistical physics understanding” of the hardness of NP-complete problems Experimental Complexity Theory [Achlioptas, Ricci-Tersenghi 2006] Known heuristic CSP algorithms fail when a large connected cluster of solutions “melts” into exponentially many disconnected pieces
Feynman apparently had trouble accepting that P vs. NP was an open problem at all! Claim: Had we been physicists, we would’ve long ago declared PNP a law of nature When people say: “What if P=NP? What if there’s an n10000 algorithm for SAT? Or an nlogloglog(n) algorithm?” Response: What if the aliens killed JFK to keep him from discovering that algorithm?
“But couldn’t you have said the same about Linear Programming before Khachiyan, or primality before AKS?” No. In those cases we had plenty of hints about what was coming, from both theory and practice. “But haven’t there been lots of surprises in complexity?”
Achievement 3: Knowing What A Nontrivial Lower Bound Looks Like
(A 5-line observation that everyone somehow missed?) Can P vs. NP Be Solved By A “Fool’s Mate?” Suppose P=NP. Then clearly PA=NPA for all oracles A. But this is known to be false; hence PNP. Fact[A.-Wigderson ’08]: Given a 3SAT formula , suppose a randomized verifier needs (polylog n) queries to to decide if is satisfiable, even given polylog(n) communication with a competing yes-prover and no-prover (both of whom can exchange private messages not seen by the other prover). Then PNP. Proof: If P=NP, then NEXP=EXP=RG (where RG = Refereed Games), and indeed NEXPA=EXPA=RGA for all oracles A.
Lower bounds on proof complexity Monotone lower bound for Clique[Razborov] So What Does A Real Chess Match Look Like? Lower bounds for constant-depth circuits [FSS, Ajtai, RS] Circuit lower bounds based on algebraic degree [Strassen, Mulmuley…] Lower bounds for specific algorithms (DPLL, GSAT…) nlog(n) lower bound on multilinear formula size [Raz] Bounds on spectral gaps for NP-complete problems [DMV, FGG] Time-space tradeoffs for SAT Circuit lower bounds for PP, MAEXP, etc. [BFT, Vinodchandran, Santhanam]
Metaquestion: Given how short these results fall of proving PNP, can we infer anything from them about what a proof of PNP would look like? • Yes! Any proof of PNP (or at least of NPP/poly, NPcoNP, etc.) will have to contain most of the known lower bounds as special cases • Analogy: We don’t have a quantum theory of gravity, but the fact that it has to contain the existing theories (QM and GR) as limiting cases constrains it pretty severely • This provides another explanation for why PNP is so hard, as well as a criterion for evaluating proposed approaches
Relativization [BGS’75]: Any proof of PNP (or even NEXPP/poly, etc.) will need to use something specific about NP-complete problems—something that wouldn’t be true in a fantasy universe where P and NP machines could both solve PSPACE-complete problems for free Algebrization [AW’08]: Any proof of PNP (or even NEXPP/poly, etc.) will need to use something specific about NP-complete problems, besides the extendibility to low-degree polynomials used in IP=PSPACE and other famous non-relativizing results Natural Proofs [RR’97]: Any proof of NPP/poly (or even NPTC0, etc.) will need to use something specific about NP-complete problems—some property that can’t be exploited to efficiently certify a random Boolean function as hard (thereby breaking pseudorandom generators, and doing many of very things we were trying to prove intractable) But don’t serious mathematicians ignore all these barriers, and just plunge ahead and tackle hard problems—their minds unpolluted by pessimism? The known barriers, in one sentence each If you like to be unpolluted by pessimism, why are you thinking about P vs. NP?
PEXP[Hartmanis-Stearns] PNP RELATIVIZATION ALGEBRIZATION MAEXP P/poly[BFT] Parity AC0[FSS, Ajtai] NATURAL PROOFS • NP AC0[Furst-Saxe-Sipser, Ajtai] • NPACC0 • NPTC0 • NPNC • NPP/poly • MAEXP P/poly[Buhrman-Fortnow-Thierauf] • NEXPP/poly • PSPACEP/poly • EXPP/poly • NPP/poly
One can define analogues of P and NP over an arbitrary field F When F is finite (e.g., F=F2), we recover the usual P vs. NP question When F=R or F=C, we get an interesting new question with a “mathier” feel All three cases (F=F2, F=R, and F=C) are open, and no implications are known among them But the continuous versions (while ridiculously hard themselves) seem likely to be “easier” than the discrete version The Blum-Cucker-Shub-Smale Model
Even Simpler: Permanent vs. Determinant [Valiant 70’s]: Given an nn matrix A, suppose you can’t write per(A) as det(B), where B is a poly(n)poly(n) matrix of linear combinations of the entries of A. Then AlgNCAlg#P. This is important! It reduces a barrier problem in circuit lower bounds to algebraic geometry—a subject about which there are yellow books.
To each (real) complexity class C, one can associate a (real) algebraic variety XC Mulmuley’s GCT Program:The String Theory of Computer Science Dream: Show that X#P(n) has “too little symmetry” to be embedded into XNC(m). This would imply AlgNCAlg#P. X#P(n) = “Orbit closure” of the nn Permanent function, under invertible linear transformations of the entries XNC(m) = “Orbit closure” of the mm Determinant function, for some m=poly(n)
But where do we get any new leverage? • Proposal: Exploit the “exceptional” nature of the Permanent and Determinant functions—the fact that these functions can be uniquely characterized by their symmetries—to reduce the embeddability problem to a problem in representation theory • (Which merely requires a generalization of a generalization of a generalization of the Riemann Hypothesis over finite fields) • Indeed, we already knew from Relativization / Algebrization / Natural Proofs that we’d have to exploit some special properties of the Permanent and Determinant, besides their being low-degree polynomials Mulmuley’s GCT Program:The String Theory of Computer Science
Metaquestion: Why should PNP be provable at all? • Indeed, people have speculated since the 70s about its possible independence from set theory—see [A.’03] • If PNP is a “universal mathematical statement”, why shouldn’t the proof require an infinite number of mathematical ideas? • More concretely: if the proof needs to “know” that Matching is in P, Linear Programming is in P, etc., what doesn’t it need to know is in P? • GCT suggests one possible answer: the proof would only need to know about “exceptional” problems in P (e.g., problems characterized by their symmetries)
A proof of PNP might have to be the greatest synthesis of mathematical ideas ever But don’t let that discourage you “Obvious” starting point is Permanent vs. Determinant My falsifiable prediction: Progress will come not by ignoring the last half-century of complexity theory and starting afresh, but by subsuming the many disparate facts we already know into something terrifyingly bigger Conclusions If nothing else, this provides a criterion for evaluating proposed P vs. NP attempts
P vs. NP • Use GCT (or pieces of it) to prove something new about computation • One natural place to look: Polynomial Identity Testing • Evade the algebrization and natural proofs barriers, by exploiting additional “magic properties” of NP- and #P-complete problems • Beyond the locality of 3SAT and the low degree / self-correctibility of the permanent • “Experimental complexity theory”: What else can we do? Open Problems