CS151 Complexity Theory Lecture 15 - PowerPoint Presentation

1. CS151Complexity TheoryLecture 15May 23, 2023

2. May 23, 2023Arthur-Merlin GamesDelimiting # of rounds:AM[k] = Arthur-Merlin game with k rounds, Arthur (verifier) goes firstMA[k] = Arthur-Merlin game with k rounds, Merlin (prover) goes firstTheorem: AM[k] (MA[k]) equals AM[k] (MA[k]) with perfect completeness. i.e., x L implies accept with probability 1proof on problem set CS151 Lecture 15

3. May 23, 2023Arthur-Merlin GamesTheorem: for all constant k 2AM[k] = AM[2].Proof:we show MA[2] AM[2]implies can move all of Arthur’s messages to beginning of interaction:AMAMAM…AM = AAMMAM…AM… = AAA…AMMM…M CS151 Lecture 15

4. May 23, 2023Arthur-Merlin GamesProof (continued):given L MA[2]x L m Prr[(x, m, r) R] = 1Prr[m (x, m, r) R] = 1 x L m Prr[(x, m, r) R] ε Prr[ m (x, m, r) R] 2|m|ε by repeating t times with independent random strings r, can make error ε < 2-tset t = m+1 to get 2|m|ε < ½.  order reversedCS151 Lecture 15

5. May 23, 2023MA and AMTwo important classes:MA = MA[2]AM = AM[2]definitions without reference to interaction:L MA iff poly-time language R x L m Prr[(x, m, r) R] = 1 x L m Prr[(x, m, r) R] ½ L AM iff poly-time language R x L Prr[ m (x, m, r) R] = 1 x L Prr[ m (x, m, r) R] ½ CS151 Lecture 15

6. May 23, 2023MA and AML AM iff poly-time language R x L Prr[ m (x, m, r) R] = 1 x L Prr[ m (x, m, r) R] ½Relation to other complexity classes:both contain NP (can elect to not use randomness)both contained in ∏2. L ∏2 iff R P for which: x L Prr[ m (x, m, r) R] = 1 x L Prr[ m (x, m, r) R] 1so clear that AM ∏2know that MA AM CS151 Lecture 15

7. May 23, 2023MA and AMPNPcoNPΣ2Π2AMcoAMMAcoMACS151 Lecture 15

8. May 23, 2023MA and AMTheorem: coNP AM PH = AM.Proof:suffices to show Σ2 AM (and use AM Π2)L Σ2 iff poly-time language R x L y z (x, y, z) R x L y z (x, y, z) RMerlin sends y1 AM exchange decides coNP query: z (x, y, z)R ?3 rounds; in AM CS151 Lecture 15

9. May 23, 2023MA and AMWe know Arthur-Merlin = IP.“public coins = private coins”Theorem (GS): IP[k] AM[O(k)]stronger resultimplies for all constant k 2,IP[k] = AM[O(k)] = AM[2]So, GNI IP[2] = AM CS151 Lecture 15

10. May 23, 2023Back to Graph IsomorphismThe payoff:not known if GI is NP-complete.previous Theorems: if GI is NP-complete then PH = AMunlikely! Proof: GI NP-complete GNI coNP-complete coNP AM PH = AM CS151 Lecture 15

11. May 23, 2023MA and AMTwo important classes:MA = MA[2]AM = AM[2]definitions without reference to interaction:L MA iff poly-time language R x L m Prr[(x, m, r) R] = 1 x L m Prr[(x, m, r) R] ½ L AM iff poly-time language R x L Prr[ m (x, m, r) R] = 1 x L Prr[ m (x, m, r) R] ½ CS151 Lecture 15

12. May 23, 2023MA and AMPNPcoNPΣ2Π2AMcoAMMAcoMACS151 Lecture 15

13. Derandomization revisitedL MA iff poly-time language R x L m Prr[(x, m, r) R] = 1 x L m Prr[(x, m, r) R] ½ Recall PRGs:for all circuits C of size at most s:|Pry[C(y) = 1] – Prz[C(G(z)) = 1]| ≤ ε May 23, 2023NPAMMAseedoutput stringGt bitsu bitsCS151 Lecture 15

14. Using PRGs for MAL MA iff poly-time language R x L m Prr[(x, m, r) R] = 1 x L m Prr[(x, m, r) R] ½ produce poly-size circuit C such thatC(x, m, r) = 1 (x,m,r) Rfor each x, m can hardwire to get Cx,m m Pry[Cx,m(y) = 1] = 1 (“yes”) m Pry[Cx,m(y) = 1] ≤ 1/2 (“no”) May 23, 2023CS151 Lecture 15

15. Using PRGs for MAcan compute Prz[Cx,m(G(z)) = 1] exactlyevaluate Cx,m(G(z)) on every seed z {0,1}trunning time (O(|Cx,m|)+(time for G))2tx L m [Prz[Cx,m(G(z)) = 1] = 1]x L m [Prz[Cx,m(G(z)) = 1] ½ + ] L NP if PRG with t = O(log n), < 1/2 Theorem: E requires exponential size circuits MA = NP.  May 23, 2023poly(n)poly-timeCS151 Lecture 15

16. May 23, 2023MA and AMPNPcoNPΣ2Π2AMcoAMMAcoMA(under a hardness assumption)CS151 Lecture 15

17. May 23, 2023MA and AMPNPcoNPΣ2Π2AMcoAMMA = = coMA(under a hardness assumption)What about AM, coAM? CS151 Lecture 15

18. May 23, 2023Derandomization revisitedTheorem (IW, STV): If E contains functions that require size 2Ω(n) circuits, then E contains functions that are 2Ω(n) –un-approximable by circuits.Theorem (NW): if E contains 2Ω(n)-unapp-roximable functions there are poly-time PRGs fooling poly(n)-size circuits, with seed length t = O(log n), and error < 1/4. CS151 Lecture 15

19. May 23, 2023Oracle circuitsA-oracle circuit Calso allow “A-oracle gates”circuit Cdirected acyclic graphnodes: AND (); OR (); NOT (); variables xi   x1x2   x3…xn A1 iff x A xCS151 Lecture 15

20. May 23, 2023Relativized versionsTheorem: If E contains functions that require size 2Ω(n) A-oracle circuits, then E contains functions that are 2Ω(n) -unapproximable by A-oracle circuits.Recall proof:encode truth table to get hard functionif approximable by s(n)-size circuits, then use those circuits to compute original function by size s(n)O(1)-size circuits. Contradiction.CS151 Lecture 15

21. May 23, 2023Relativized versions01001010m:01001010Enc(m):0001001100010R:01000DCf:{0,1}log k {0,1} f ’:{0,1}log n {0,1} small A-oracle circuit C approximating f’decoding procedurei {0,1}log k small A-oracle circuit computing f exactlyf(i)CS151 Lecture 15

22. May 23, 2023Relativized versionsTheorem: if E contains 2Ω(n)-unapp-roximable fns., there are poly-time PRGs fooling poly(n)-size A-oracle circuits, with seed length t = O(log n), and error < 1/4.Recall proof:PRG from hard function on O(log n) bitsif doesn’t fool s-size circuits, then use those circuits to compute hard function by size sn-size circuits. Contradiction. CS151 Lecture 15

23. Relativized versionsMay 23, 2023Gn(y)=flog n(y|S1)◦flog n(y|S2)◦…◦flog n(y|Sm)010100101111101010111001010flog n:doesn’t fool A-oracle circuit of size s:|Prx[C(x) = 1] – Pry[C( Gn(y) ) = 1]| > εimplies A-oracle circuit P of size s’ = s + O(m):Pry[P(Gn(y)1…i-1) = Gn(y)i] > ½ + ε/mCS151 Lecture 15

24. Relativized versionsMay 23, 2023Gn(y)=flog n(y|S1)◦flog n(y|S2)◦…◦flog n(y|Sm)010100101111101010111001010flog n:Poutput flog n(y ’) hardwired tablesy’A-oracle circuit approximating f CS151 Lecture 15

25. Using PRGs for AML AM iff poly-time language R x L Prr[ m (x, m, r) R] = 1 x L Prr[ m (x, m, r) R] ½produce poly-size SAT-oracle circuit C such thatC(x, r) = 1 m (x,m,r) Rfor each x, can hardwire to get Cx Pry[Cx(y) = 1] = 1 (“yes”) Pry[Cx(y) = 1] ≤ ½ (“no”) May 23, 20231 SAT query, accepts iff answer is “yes”CS151 Lecture 15

26. Using PRGs for AMx L [Prz[Cx(G(z)) = 1] = 1]x L [Prz[Cx(G(z)) = 1] ½ + ]Cx makes a single SAT query, accepts iff answer is “yes”if G is a PRG with t = O(log n), < ¼, can check in NP: does Cx(G(z)) = 1 for all z? May 23, 2023CS151 Lecture 15

27. May 23, 2023Relativized versionsTheorem: If E contains functions that require size 2Ω(n) A-oracle circuits, then E contains functions that are 2Ω(n) -unapproximable by A-oracle circuits. Theorem: if E contains 2Ω(n)-unapproximable functions there are PRGs fooling poly(n)-size A-oracle circuits, with seed length t = O(log n), and error < ½.Theorem: E requires exponential size SAT-oracle circuits AM = NP.  CS151 Lecture 15

28. May 23, 2023MA and AMPNPcoNPΣ2Π2MA = = coMA(under a hardness assumption)coAMAMCS151 Lecture 15

29. May 23, 2023MA and AMPNPcoNPΣ2Π2MA = = coMA(under a hardness assumption)= coAMAM =CS151 Lecture 15

30. May 23, 2023New topic(s)Optimization problems, Approximation Algorithms, and Probabilistically Checkable ProofsCS151 Lecture 15

31. May 23, 2023Optimization Problemsmany hard problems (especially NP-hard) are optimization problemse.g. find shortest TSP toure.g. find smallest vertex cover e.g. find largest cliquemay be minimization or maximization problem“opt” = value of optimal solutionCS151 Lecture 15

32. May 23, 2023Approximation Algorithmsoften happy with approximately optimal solutionwarning: lots of heuristicswe want approximation algorithm with guaranteed approximation ratio of rmeaning: on every input x, output is guaranteed to have value at most r*opt for minimization at least opt/r for maximizationCS151 Lecture 15

33. May 23, 2023Approximation AlgorithmsExample approximation algorithm:Recall:Vertex Cover (VC): given a graph G, what is the smallest subset of vertices that touch every edge?NP-complete CS151 Lecture 15

34. May 23, 2023Approximation AlgorithmsApproximation algorithm for VC:pick an edge (x, y), add vertices x and y to VCdiscard edges incident to x or y; repeat.Claim: approximation ratio is 2.Proof: an optimal VC must include at least one endpoint of each edge considered therefore 2*opt actual CS151 Lecture 15

35. May 23, 2023Approximation Algorithmsdiverse array of ratios achievablesome examples:(min) Vertex Cover: 2 MAX-3-SAT (find assignment satisfying largest # clauses): 8/7 (min) Set Cover: ln n(max) Clique: n/log2n(max) Knapsack: (1 + ε) for any ε > 0 CS151 Lecture 15

36. May 23, 2023Approximation Algorithms(max) Knapsack: (1 + ε) for any ε > 0 called Polynomial Time Approximation Scheme (PTAS)algorithm runs in poly time for every fixed ε>0poor dependence on ε allowedIf all NP optimization problems had a PTAS, almost like P = NP (!)CS151 Lecture 15

37. May 23, 2023Approximation AlgorithmsA job for complexity: How to explain failure to do better than ratios on previous slide?just like: how to explain failure to find poly-time algorithm for SAT...first guess: probably NP-hardwhat is needed to show this?“gap-producing” reduction from NP-complete problem L1 to L2CS151 Lecture 15

38. May 23, 2023Approximation Algorithms“gap-producing” reduction from NP-complete problem L1 to L2noyesL1L2 (min. problem)foptkrkCS151 Lecture 15

39. May 23, 2023Gap producing reductionsr-gap-producing reduction:f computable in poly time x L1 opt(f(x)) k x L1 opt(f(x)) > rk for max. problems use “ k” and “< k/r”Note: target problem is not a languagepromise problem (yes no not all strings)“promise”: instances always from (yes no)  CS151 Lecture 15

40. May 23, 2023Gap producing reductionsMain purpose:r-approximation algorithm for L2 distinguishes between f(yes) and f(no); can use to decide L1 “NP-hard to approximate to within r”noyesL1fkrkyesnoL1fk/rkL2 (min.)L2 (max.)yesnoyesnoCS151 Lecture 15

41. May 23, 2023Gap preserving reductionsgap-producing reduction difficult (more later)but gap-preserving reductions easierfkrkk’r’k’Warning: many reductions not gap-preservingyesnoyesnoL1 (min.)L2 (min.)CS151 Lecture 15

42. May 23, 2023Gap preserving reductionsExample gap-preserving reduction:reduce MAX-k-SAT with gap ε to MAX-3-SAT with gap ε’ “MAX-k-SAT is NP-hard to approx. within ε MAX-3-SAT is NP-hard to approx. within ε’ ”MAXSNP (PY) – a class of problems reducible to each other in this wayPTAS for MAXSNP-complete problem iff PTAS for all problems in MAXSNP constantsCS151 Lecture 15

43. May 23, 2023MAX-k-SATMissing link: first gap-producing reduction history’s guideit should have something to do with SATDefinition: MAX-k-SAT with gap εinstance: k-CNF φYES: some assignment satisfies all clausesNO: no assignment satisfies more than (1 – ε) fraction of clausesCS151 Lecture 15

44. May 23, 2023Proof systems viewpointk-SAT NP-hard for any language LNP proof system of form:given x, compute reduction to k-SAT: xexpected proof is satisfying assignment for xverifier picks random clause (“local test”) and checks that it is satisfied by the assignment x L Pr[verifier accepts] = 1 x L Pr[verifier accepts] < 1 CS151 Lecture 15

45. May 23, 2023Proof systems viewpointMAX-k-SAT with gap ε NP-hard for any language L NP proof system of form:given x, compute reduction to MAX-k-SAT: xexpected proof is satisfying assignment for xverifier picks random clause (“local test”) and checks that it is satisfied by the assignment x L Pr[verifier accepts] = 1 x L Pr[verifier accepts] ≤ (1 – ε)can repeat O(1/ε) times for error < ½ CS151 Lecture 15

46. May 23, 2023Proof systems viewpointcan think of reduction showing k-SAT NP-hard as designing a proof system for NP in which:verifier only performs local testscan 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 testsinvalidity of proof* evident all over: “holographic proof” and an fraction of tests notice such invalidity CS151 Lecture 15