Topics in Complexity Theory

Transcription

1 Topics in Complexity Theory

2 Announcements Final exam this Friday from 12:15PM-3:15PM Please let us know immediately after lecture if you want to take the final at an alternate time and haven't yet told us. Final exam review session today after lecture in lieu of normal problem session. Problem Set 7 graded, but unfortunately we can't hand them back today. We'll get them returned ASAP.

3 Approximation Algorithms

4 Decision vs. Function Problems All of the problems we have encountered this quarter are decision problems with yes/no answers. Many interesting questions do not have yes/no answers: What is 1 + 1? How many steps does it take this TM to halt on this string? What is the shortest path from u to v? These questions are called function problems and require some object to be found.

5 NP-Hard Function Problems Some function problems are NP-hard: there is a polynomial-time reduction from any problem in NP to those problems. Need a slightly different definition of a reduction in this case, but we'll skip that for now. Examples: What is the largest independent set in a graph? What is the length of the longest path in a graph? What is the minimum number of colors required to color a graph? A polynomial-time solver for any of these could be used to build a polynomial-time decider for any problem in NP.

6 NP-Hard Function Problems The maximum independent set problem (MAX-INDSET) is Given a graph G, find an independent set S in G of the largest possible size. MAX-INDSET is NP-hard by a reduction from independent set: M = On input G, k : Find a maximum independent set in G, call it S. If S k, accept; otherwise, reject.

7 NP-Hard Function Problems Because they can be used to solve any problem in NP, NP-hard function problems are believed to be computationally infeasible. If any NP-hard function problem has a polynomialtime solution, then P = NP. Since the P = NP question is still open, no NP-hard function problems are known to have polynomial-time solutions.

8 Approximation Algorithms An approximation algorithm is an algorithm for yielding a solution that is close to the correct solution to a problem. The definition of close depends on the problem: Maximum independent set: Find an independent set that is not much smaller than the maximum independent set. Longest path: Find a long path that is not much smaller than the longest path. Graph coloring: Find a coloring of the graph that does not use many more colors than the optimal coloring.

9 How Good is an Approximation? Approximation algorithms are only useful if there is some connection between the approximate answer and the real answer. We say that an approximation algorithm is a k-approximation if its answer is always within a factor of k of the optimal solution. A 2-approximation to the graph coloring problem always finds a coloring that uses at most twice the optimal number of colors. A 2/3-approximation to the longest path problem always finds a path that is at least 2/3 as long as the optimal path.

10 Why Approximations Matter A classic NP-hard problem is job scheduling. Given a set of tasks and a set of workers to perform the tasks, how do you distribute tasks to workers to minimize the total time required (assuming the workers work in parallel?) Applications to manufacturing, operating systems, etc. Although finding an optimal solution is NP-hard, there are polynomial-time algorithms for finding 4/3-approximate solutions. If we just need a good enough answer, the NPhardness of job scheduling is not a deterrent to its use.

11 A Hamiltonian cycle in an undirected graph G is a simple cycle that visits every node in G.

12 A Hamiltonian cycle in an undirected graph G is a simple cycle that visits every node in G.

13 A Hamiltonian cycle in an undirected graph G is a simple cycle that visits every node in G.

14 A Hamiltonian cycle in an undirected graph G is a simple cycle that visits every node in G.

15 A Hamiltonian cycle in an undirected graph G is a simple cycle that visits every node in G.

16 Hamiltonian Cycles The undirected Hamiltonian cycle problem is Given an undirected graph G, does G contain a Hamiltonian cycle? As a formal language: UHAMCYCLE = { G G is an undirected graph containing a Hamiltonian cycle } Important fact: UHAMCYCLE is NP-complete. Reduction from UHAMPATH.

17 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

18 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

19 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

20 Cost: Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

21 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

22 Cost: Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

23 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

24 Cost: Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

25 Cost: 39 (This is the optimal solution) 15 Given a complete, undirected, weighted graph G, the traveling salesman problem (TSP) is to find a Hamiltonian cycle in G of least total cost.

26 TSP, Formally Given as input A complete, undirected graph G, and a set of edge weights, which are positive integers, the TSP is to find a Hamiltonian cycle in G with least total weight. Note that since G is complete, there has to be at least one Hamiltonian cycle. The challenge is finding the least-cost cycle.

27 TSP There are many variations on TSP: Directed versus undirected graphs. Complete versus incomplete graphs. Finding a path versus finding a cycle. Nonnegative weights versus arbitrary weights. All of these problems are known to be NP-hard. The best-known exact algorithms have horrible runtimes: O(n 2 2 n ).

28 TSP There are many variations on TSP: Directed versus undirected graphs. Complete versus incomplete graphs. Finding a path versus finding a cycle. Nonnegative weights versus arbitrary weights. All of these problems are known to be NP-hard. The best-known exact algorithms have horrible runtimes: O(n 2 2 n ).

29 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP.

30 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP.

31 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP.

32 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP.

33 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP. Cost 1 Cost 2

34 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP. Cost 1 Cost 2 If there is a Hamiltonian cycle in the original graph, there is a TSP solution of cost n in the new graph (where n is the number of nodes) If there is no Hamiltonian path, the TSP solution has cost at least n.

35 TSP is NP-Hard To show that TSP is NP-hard, we reduce the undirected Hamiltonian cycle problem to TSP. Cost 1 Cost 2 If there is a Hamiltonian cycle in the original graph, there is a TSP solution of cost n in the new graph (where n is the number of nodes) If there is no Hamiltonian cycle, the TSP solution has cost at least n + 1.

36 Answering TSP TSP has great practical importance, and we want to be able to answer it. Determining the shortest subway routes to link stations in a ring. Determining the fastest way to move a mechanical drill so it can drill in the appropriate locations. Because TSP is NP-hard, obtaining an exact answer is impractically hard. Can we approximate it?

37 Approximating TSP An approximation algorithm to TSP is a k-approximation if it finds a Hamiltonian cycle whose cost is at most k times the optimal solution. If the optimal solution has cost 10, a 2-approximation would return a Hamiltonian cycle of cost between 10 and 20, inclusive. If the optimal solution has cost 10, a 3-approximation would return a Hamiltonian cycle of cost between 10 and 30, inclusive. For what values of k is there an efficient k-approximation to TSP?

38 Theorem: If P NP, then there is no polynomial-time k-approximation to TSP for any natural number k.

39 Hardness of Approximation The proof that TSP is hard to approximate is based on a beautiful construction: If we could obtain a k-approximation to TSP in polynomial-time, we would have a polynomial-time algorithm for UHAMCYCLE. Since UHAMCYCLE is NP-complete, this is a contradiction if P NP.

40 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k.

41 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k.

42 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k.

43 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. Cost 1 Cost??

44 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the original graph of n nodes has a Hamiltonian cycle, the TSP solution has cost n. The k-approximation algorithm thus must hand back a Hamiltonian cycle of cost at most kn. Cost 1 Cost?? What if we made the cost of the red edges so large that any solution using them must cost more than this?

45 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the original graph of n nodes has a Hamiltonian cycle, the TSP solution has cost n. The k-approximation algorithm thus must hand back a Hamiltonian cycle of cost at most kn. Cost 1 Cost?? What if we made the cost of the red edges so large that any solution using them must cost more than this?

46 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the original graph of n nodes has a Hamiltonian cycle, the TSP solution has cost n. The k-approximation algorithm thus must hand back a Hamiltonian cycle of cost at most kn. Cost 1 Cost?? What if we made the cost of the red edges so large that any solution using them must cost more than this?

47 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the original graph of n nodes has a Hamiltonian cycle, the TSP solution has cost n. The k-approximation algorithm thus must hand back a Hamiltonian cycle of cost at most kn. Cost 1 Cost kn + 1 What if we made the cost of the red edges so large that any solution using them must cost more than this?

48 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. Cost 1 Cost kn + 1

49 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the k-approximation hands back a solution of cost at most kn, the original graph has a Hamiltonian cycle. Cost 1 Cost kn + 1 If the k-approximation hands back a solution of cost at least kn + 1, the original graph has no Hamiltonian cycles.

50 The Construction Proof Sketch: Suppose, for the sake of contradiction, that there is a k-approximation to TSP for some k. If the k-approximation hands back a solution of cost at most kn, the original graph has a Hamiltonian cycle. Cost 1 Cost kn + 1 If the k-approximation hands back a solution of cost at least kn + 1, the original graph has no Hamiltonian cycles.

51 What Just Happened?

52 Cost of optimal TSP solution What Just Happened?

53 What Just Happened? Cost of optimal TSP solution Cost of a yes answer.

54 What Just Happened? Cost of the best possible no answer. Cost of optimal TSP solution Cost of a yes answer.

55 What Just Happened? Cost of the best possible no answer. Cost of optimal TSP solution k times the cost of a yes answer. Cost of a yes answer.

56 What Just Happened? Cost of the best possible no answer. Cost of optimal TSP solution k times the cost of a yes answer. Cost of a yes answer.

57 What Just Happened? Cost of the best possible no answer. Cost of optimal TSP solution k times the cost of a yes answer. Cost of a yes answer.

58 What Just Happened? Create an enormous gap between the TSP answer for yes and no instances of the Hamiltonian cycle problem. Make the gap so large that the worst possible k-approximate answer to a yes instance is distinguishable from the best possible no instance. Decide whether there is a Hamiltonian cycle by measuring this difference.

59 The PCP Theorem

60 Approximating 3SAT 3SAT is one of the canonical NP-complete problems. What would it mean to approximate a 3SAT answer?

61 Approximating 3SAT The MAX-3SAT problem is Given a 3CNF formula φ, find a satisfying assignment the maximizes the number of satisfied clauses. Idea: If we can satisfy the entire formula, then MAX- 3SAT finds a satisfying assignment. If not, MAX-3SAT finds the best assignment that it can. There is a known randomized 7/8-approximation algorithm for MAX-3SAT. Is it possible to do better?

62 The 3-CNF Value If φ is a 3-CNF formula, the value of φ (denoted val(φ)) is the fraction of the clauses of φ that can be simultaneously satisfied by any assignment. If φ is satisfiable, val(φ) = 1. All of the clauses are satisfiable. If φ is unsatisfiable, val(φ) < 1. Some (but not all) clauses can be satisfied.

63 The PCP Theorem For any language L NP, There exists a poly-time reduction f from L to MAX-3SAT so For any string w: If w L, then val(f( w)) = 1. If w L, then val(f( w)) 7/8.

64 The PCP Theorem For any language L NP, There exists a poly-time reduction f from L to MAX-3SAT so For any string w: If w L, then val(f( w)) = 1. If the original answer is yes, the 3-CNF formula is satisfiable. If w L, then val(f( w)) 7/8.

65 The PCP Theorem For any language L NP, There exists a poly-time reduction f from L to MAX-3SAT so For any string w: If w L, then val(f( w)) = 1. If the original answer is yes, the 3-CNF formula is satisfiable. If w L, then val(f( w)) 7/8. If the original answer is no, at most 7/8 of the clauses can be satisfied.

66 What Does This Mean? Our proof that (unless P = NP) TSP cannot be efficiently approximated works by building up a gap between yes answers and no answers. The PCP theorem states that any problem in NP can be reduced to MAX-3SAT such that All of the clauses are satisfiable if the original answer is yes. At most 7/8 of the clauses are satisfiable if the answer is no. This gap of 1/8 of the clauses is large enough to preclude approximations beyond the known 7/8-approximation.

67 Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT.

68 MAX-3SAT is Inapproximable Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT.

69 MAX-3SAT is Inapproximable Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT. Fraction of Satisfiable Clauses

70 MAX-3SAT is Inapproximable Fraction satisfiable in a yes answer Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT. Fraction of Satisfiable Clauses

71 MAX-3SAT is Inapproximable Fraction satisfiable in a yes answer 7/8 Fraction satisfiable in a no answer Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT. Fraction of Satisfiable Clauses

72 MAX-3SAT is Inapproximable Fraction satisfiable in a yes answer r times the fraction satisfiable in a yes answer 7/8 Fraction satisfiable in a no answer Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT. Fraction of Satisfiable Clauses

73 MAX-3SAT is Inapproximable Fraction satisfiable in a yes answer r times the fraction satisfiable in a yes answer 7/8 Fraction satisfiable in a no answer Theorem: If P NP, then for any r > 7/8, there is no polynomial-time r-approximation to MAX-3SAT. Fraction of Satisfiable Clauses

74 Effects on Approximability Assuming P NP, there is a limit to how well we can approximate 3SAT. Look at our reduction from 3SAT to INDSET: ( x y z ) ( x y z ) ( x y z )

75 Effects on Approximability Assuming P NP, there is a limit to how well we can approximate 3SAT. Look at our reduction from 3SAT to INDSET: ( x y z ) ( x y z ) ( x y z )

76 Effects on Approximability Assuming P NP, there is a limit to how well we can approximate 3SAT. Look at our reduction from 3SAT to INDSET: ( x y z ) ( x y z ) ( x y z ) y y y x z x z x z

77 Effects on Approximability Many reductions preserve the size of some difficult-toapproximate quantity. Assuming if P NP: Because MAX-3SAT is not efficiently 7/8-approximable, MAX-INDSET is not efficiently 7/8-approximable either. Because MAX-INDSET is not efficiently 7/8-approximable, MAX-CLIQUE is not efficiently 7/8-approximable. Because MAX-INDSET is not efficiently ρ-approximable, MAX-SETPACK is not efficiently 7/8-approximable. Not all reductions have this property; some NP-hard problems can be efficiently approximated to very high precision.

78 Oh, and a proof of the PCP theorem?

79 Oh, and a proof of the PCP theorem?

80 Summary of Approximability While many NP-hard problems can be efficiently approximated, certain NP-hard problems cannot unless P = NP. Efficiently approximating some NP-hard problems too well would solve NP-complete problems. The PCP theorem states that the gap between yes and no answers can be so large that approximating the solution would essentially solve the problem.

81 Beyond P and NP

82 co-np P NP

83 The Class co-np The complexity class co-np is the set of languages whose complements are in NP. Formally, L co-np iff L NP. Intuition: A language is in NP iff it can be verified efficiently. A language is in co-np iff it can be refuted efficiently.

84 Problems in co-np The problem Is φ satisfiable? (SAT) is a classic NP problem. We can easily verify that φ is satisfiable given a satisfying assignment. Finding that satisfying assignment might be hard. The problem Is φ a tautology? (TAUT) is a classic co-np problem. We can easily refute that φ is a tautology given an unsatisfying assignment. Finding that unsatisfying assignment might be hard.

85 Problems in co-np The problem does a graph contain a k-clique? is in NP. Can just check that a candidate clique is a k-clique. Finding that k-clique might be hard. The problem does a graph not contain a k- clique? is in co-np. If we are given a k-clique in a graph, we can refute that a graph does not contain a k-clique. Finding that k-clique might be hard.

86 NP co-np It is unknown whether NP = co-np. If P = NP, then NP = co-np. If NP co-np, then P NP. If P NP, it is possible that NP = co-np. co-np P NP

87 Beyond P, NP, and co-np

88 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

89 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

90 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

91 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

92 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

93 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

94 L NP iff poly-time verifier V for L. What is a poly-time verifier? w L iff V is a poly-time verifier for L iff poly-sized x. V accepts w, x in poly-time. Notation: p x means poly-sized x. V is a poly-time verifier for L iff w L iff p x. V accepts w, x in poly-time. Since V runs in polynomial time, this means that L( V) P. Consequently: L NP iff L' P. w L iff p x. w, x L'.

95 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

96 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

97 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

98 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

99 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

100 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

101 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

102 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. L co-np iff L' P. w L iff p x. w, x L'

103 L co-np iff L NP. Consequently: L co-np iff L' P. w L iff p x. w, x L' Equivalently: L co-np iff L' P. w L iff p x. w, x L' P is closed under complementation. Thus L co-np iff L' P. w L iff p x. w, x L'

104 L P iff there is a poly-time decider for L. L NP iff L' P. w L iff p x. w, x L' L co-np iff L' P. w L iff p x. w, x L'

105 The Polynomial Hierarchy Let Σ 0 P = P Let Π 0 P = P P Σ n+1 = { L L' P Π n+1 = { L L' Π np. w L iff p x. w, x L' } Σ np. w L iff p x. w, x L' } Examples: Σ 1 P = NP Π 1 P = co-np

106 The Polynomial Hierarchy Let Σ 0 P = P Let Π 0 P = P P Σ n+1 = { L L' P Π n+1 = { L L' Π np. w L iff p x. w, x L' } Σ np. w L iff p x. w, x L' } Examples: Σ 1 P = NP Π 1 P = co-np

107 The Polynomial Hierarchy Let Σ 0 P = P Let Π 0 P = P P Σ n+1 = { L L' P Π n+1 = { L L' Π np. w L iff p x. w, x L' } Σ np. w L iff p x. w, x L' } Examples: Σ 1 P = NP Π 1 P = co-np

108 The Polynomial Hierarchy

109 The Polynomial Hierarchy P

110 The Polynomial Hierarchy P Σ 1 P

111 The Polynomial Hierarchy Π 1 P P Σ 1 P

112 The Polynomial Hierarchy Π 1 P P Σ 1 P Σ 2 P

113 The Polynomial Hierarchy Π 2 P Π 1 P P Σ 1 P Σ 2 P

114 The Polynomial Hierarchy Π 2 P Π 1 P P Σ 1 P Σ 2 P Σ 3 P

115 The Polynomial Hierarchy Π 3 P Π 2 P Π 1 P P Σ 1 P Σ 2 P Σ 3 P

116 Um, what? Consider the following decision problem: Is there exactly one k-clique in a graph G? This problem is not known to be in either NP nor co-np. However, it is in Σ 2 P and Π 2P. G, k L iff clique. sets of k nodes. The clique is a valid k-clique and no other set is. G, k L iff sets of k nodes. clique. The clique is a valid k-clique and no other set is.

117 Generalizing Our Questions Under our new terminology: P P NP is the same as Σ 0 Σ 1P. P P co-np is the same as Π 0 Π 1P. P NP co-np is the same as Σ 1 Π 1P. More generally, for any natural number n, we do not know the following: Is Σ n P = Σ n+1p? Is Π n P = Π n+1p? Is Σ n P = Π np?

118 There are many infinite hierarchies of problems that we can solve. Complexity theory is all about the connections between those classes.

119 Next Time The Big Picture A quick recap of everything. Where to Go from Here What's next in CS theory?