CS151 Complexity Theory

Transcription

1 Gap producing reductions CS151 Complexity Theory Lecture 15 May 18, 2015 L 1 f L 2 (min.) rk k Main purpose: r-approximation algorithm for L 2 distinguishes between f() and f(); can use to decide L 1 NP-hard to approximate to within r L 1 f L 2 (max.) k k/r May 18, MAX-k-SAT Missing link: first gap-producing reduction history s guide it should have something to do with SAT Definition: MAX-k-SAT with gap ε instance: k-cnf φ YES: some assignment satisfies all clauses NO: assignment satisfies more than (1 ε) fraction of clauses May 18, Proof systems viewpoint can think of reduction showing k-sat NP-hard as designing a proof system for NP in which: verifier only performs local tests can think of reduction showing MAX-k-SAT with gap ε NP-hard as designing a proof system for NP in which: verifier only performs local tests invalidity of proof* evident all over: holographic proof and an fraction of tests tice such invalidity May 18, PCP Probabilistically Checkable Proof (PCP) permits vel way of verifying proof: pick random local test query proof in specified k locations accept iff passes test fancy name for a NP-hardness reduction PCP PCP[r(n),q(n)]: set of languages L with p.p.t. verifier V that has (r, q)-restricted access to a string proof V tosses O(r(n)) coins V accesses proof in O(q(n)) locations (completeness) x L proof such that Pr[V(x, proof) accepts] = 1 (soundness) x L proof* Pr[V(x, proof*) accepts] ½ May 18, May 18,

2 PCP Two observations: PCP[1, poly n] = NP proof? PCP[log n, 1] NP proof? The PCP Theorem (AS, ALMSS): PCP[log n, 1] = NP. May 18, PCP Corollary: MAX-k-SAT is NP-hard to approximate to within some constant. using PCP[log n, 1] protocol for, say, VC enumerate all 2 O(log n) = poly(n) sets of queries construct a k-cnf φ i for verifier s test on each te: k-cnf since function on only k bits YES VC instance all clauses satisfiable NO VC instance every assignment fails to satisfy at least ½ of the φ i fails to satisfy an = (½)2 -k fraction of clauses. May 18, The PCP Theorem Elements of proof: arithmetization of 3-SAT we will do this low-degree test we will state but t prove this self-correction of low-degree polymials we will state but t prove this proof composition we will describe the idea The PCP Theorem Two major components: ( outer verifier ) we will prove this from scratch, assuming lowdegree test, and self-correction of low-degree polymials NP PCP[n 3, 1] ( inner verifier ) we will t prove May 18, May 18, Proof Composition (idea) NP PCP[log n, polylog n] ( outer verifier ) NP PCP[n 3, 1] ( inner verifier ) composition of verifiers: reformulate outer so that it uses O(log n) random bits to make 1 query to each of 3 provers replies r 1, r 2, r 3 have length polylog n Key: accept/reject decision computable from r 1, r 2, r 3 by small circuit C May 18, Proof Composition (idea) NP PCP[log n, polylog n] ( outer verifier ) NP PCP[n 3, 1] ( inner verifier ) composition of verifiers (continued): final proof contains proof that C(r 1, r 2, r 3 ) = 1 for inner verifier s use use inner verifier to verify that C(r 1,r 2,r 3 ) = 1 O(log n)+polylog n randomness O(1) queries tricky issue: consistency May 18,

3 Proof Composition (idea) NP PCP[log n, 1] comes from repeated composition PCP[log n, polylog n] with PCP[log n, polylog n] yields PCP[log n, polyloglog n] PCP[log n, polyloglog n] with PCP[n 3, 1] yields PCP[log n, 1] many details omitted The inner verifier Theorem: NP PCP[n 2, 1] Proof (first steps): 1. Quadratic Equations is NP-hard 2. : proof = all quadratic functions of a soln. x verification = check that a random linear combination of equations is satisfied by x (if prover keeps promise to supply all quadratic fns of x) May 18, May 18, Quadratic Equations quadratic equation over F 2 : i<j a i,j X i X j + i b i X i + c = 0 language QUADRATIC EQUATIONS (QE) = { systems of quadratic equations over F 2 that have a solution (assignment to the X variables) } Quadratic Equations Lemma: QE is NP-complete. Proof: clearly in NP; reduce from CIRCUIT SAT circuit C an instance of CIRCUIT SAT QE variables = variables + gate variables g i z g i z 1 z 2 g i z 1 z 2 and g out = 1 g i z = 1 g i z 1 z 2 = 0 g i (1-z 1 )(1-z 2 ) = 1 May 18, May 18, Quadratic Functions Code intended proof: F the field with 2 elements given x 2 F n, a solution to instance of QE f x : F n! F 2 all linear functions of x f x (a) = i a i x i : F n x n! F 2 includes all quadratic fns of x (A) = i, j A[i,j]x i x j KEY: can evaluate any quadratic function of x with a single evaluation of f x and May 18, If prover keeps promise to supply all quadratic fns of x, a solution of QE instance Verifier s action: query a random linear combination R of the equations of the QE instance Completeness: obvious Soundness: x fails to satisfy some equation; imagine picking coeff. for this one last Pr[x satisfies R] = 1/2 May 18,

4 To enforce promise, verifier needs to perform: linearity test: verify f, g are (close to) linear self-correction: access the linear f, g that are close to f, g [so f = Had(u) and g = Had(V)] consistency check: verify V = u u May 18, Linearity test: given access to h:f m! F pick random a,b; check if h(a) + h(b) = h(a+b); repeat O(1) times do this for functions f and g supplied by prover Theorem [BLR]: h linear)prob. success = 1; prob. success 1 ± ) 9 linear h s.t. Pr a [h (a) = h(a)] 1 O(±) May 18, Self-correction: given access to h:f m!f close to linear h ; i.e., Pr a [h (a) = h(a)] 1 O(±) to access h (a), pick random b; compute h(b) + h(a+b) with prob. at least 1 2 O(±), h(b) = h (b) and h(a+b) = h (a+b); hence we compute h (a) May 18, Consistency check: given access to linear functions f = Had(u) and g = Had(V) pick random a, b 2 F n ; check that f (a)f (b) = g (ab T ) completeness: if V = u u f (a)f (b) = ( ia i u i )( ib i u i ) = i,j a i b j V[i,j] = g (ab T ) May 18, Consistency check: given access to linear functions f = Had(u) and g = Had(V) soundness: claim that if V u u Pr[( ia i u i )( ib i u i ) = i,j a i b j V[i,j] ] 3/4 9 i,j s.t. uu T and V differ in entry (i,j); 9 i s.t. (uu T )b and pick b j last Vb differ in entry i; Pr[(uu T )b Vb] pick 1/2 a i last Pr[aT(uu T )b a T Vb] ½ ½ = ¼ May 18, The outer verifier Theorem: Proof (first steps): define: Polymial Constraint Satisfaction (PCS) problem prove: PCS gap problem is NP-hard May 18,

5 MAX-k-SAT given: k-cnf output: max. # of simultaneously satisfiable clauses generalization: MAX-k-CSP given: variables x 1, x 2,, x n taking values from set S k-ary constraints C 1, C 2,, C t output: max. # of simultaneously satisfiable constraints May 18, algebraic version: MAX-k-PCS given: variables x 1, x 2,, x n taking values from field F q n = q m for some integer m k-ary constraints C 1, C 2,, C t assignment viewed as f:(f q ) m F q output: max. # of constraints simultaneously satisfiable by an assignment that has deg. d May 18, MAX-k-PCS gap problem: given: variables x 1, x 2,, x n taking values from field F q n = q m for some integer m k-ary constraints C 1, C 2,, C t assignment viewed as f:(f q ) m F q YES: some degree d assignment satisfies all constraints NO: degree d assignment satisfies more than (1- ) fraction of constraints May 18, Lemma: for every constant 1 > ε > 0, the MAX-k-PCS gap problem with t = poly(n) k-ary constraints with k = polylog(n) field size q = polylog(n) n = q m variables with m = O(log n / loglog n) degree of assignments d = polylog(n) gap is NP-hard. May 18, t = poly(n) k-ary constraints with k = polylog(n) field size q = polylog(n) n = q m variables with m = O(log n / loglog n) degree of assignments d = polylog(n) check: headed in right direction O(log n) random bits to pick a constraint query assignment in O(polylog(n)) locations to determine if it is satisfied completeness 1; soundness 1- (if prover keeps promise to supply degree d polymial) Proof of Lemma reduce from 3-SAT 3-CNF φ(x 1, x 2,, x n ) can encode as :[n] x [n] x [n] x {0,1} 3 {0,1} (i 1, i 2, i 3, b 1, b 2, b 3 ) = 1 iff φ contains clause (x i1 b 1 xi2 b 2 xi3 b 3) e.g. (x 3 x 5 x 2 ) (3,5,2,1,0,1) = 1 May 18, May 18,

6 pick H F q with {0,1} H, H = polylog n pick m = O(log n/loglog n) so H m = n identify [n] with H m :H m x H m x H m x H 3 {0,1} encodes φ assignment a:h m {0,1} Key: a satisfies φ iff i 1,i 2,i 3,b 1,b 2,b 3 (i 1,i 2,i 3,b 1,b 2,b 3 ) = 0 or a(i 1 )=b 1 or a(i 2 )=b 2 or a(i 3 )=b 3 :H m x H m x H m x H 3 {0,1} encodes φ a satisfies φ iff i 1,i 2,i 3,b 1,b 2,b 3 (i 1,i 2,i 3,b 1,b 2,b 3 ) = 0 or a(i 1 )=b 1 or a(i 2 )=b 2 or a(i 3 )=b 3 extend to a function :(F q ) 3m+3 F q with degree at most H in each variable can extend any assignment a:h m {0,1} to a :(F q ) m F q with degree H in each variable May 18, May 18, :(F q ) 3m+3 F q encodes φ a :(F q ) m F q s.a. iff (i 1,i 2,i 3,b 1,b 2,b 3 ) H 3m+3 (i 1,i 2,i 3,b 1,b 2,b 3 ) = 0 or a (i 1 )=b 1 or a (i 2 )=b 2 or a (i 3 )=b 3 define: p a :(F q ) 3m+3 F q from a as follows p a (i 1,i 2,i 3,b 1,b 2,b 3 ) = (i 1,i 2,i 3,b 1,b 2,b 3 )(a (i 1 ) - b 1 )(a (i 2 ) - b 2 )(a (i 3 ) - b 3 ) a s.a. iff (i 1,i 2,i 3,b 1,b 2,b 3 ) H 3m+3 p a (i 1,i 2,i 3,b 1,b 2,b 3 ) = 0 May 18,