Chapter 3: Proving NP-completeness Results

Transcription

1 Chapter 3: Proving NP-completeness Results Six Basic NP-Complete Problems Some Techniques for Proving NP-Completeness Some Suggested Exercises 1.1

2 Six Basic NP-Complete Problems 3-SATISFIABILITY (3SAT) 3-DIMENSIONAL MATCHING (3DM) VERTEX COVER (VC) CLIQUE HAMILTONIAN CIRCUIT (HC) PARTITION First, a tour 1.2

3 3-Satisfiability 3-SATISFIABILITY (3-SAT) INSTANCE: Collection C = {c 1, c 2,,c m } of clauses on a finite set U of variables such that {c j } = 3, for 1 j m. QUESTION: Is there a truth assignment for U that satisfies all the clauses in C? 1.3

4 3-Dimensional Matching 3-DIMENSIONAL MATCHING (3-DM) INSTANCE: A set M W X Y, where W, X and Y are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? Example #1: q = 3 (not part of input technically) W = {a 1, a 2, a 3 } X = {b 1, b 2, b 3 } Y = {c 1, c 2, c 3 } M = { (a 1, b 2, c 1 ), (a 2, b 1, c 3 ), (a 3, b 3, c 2 ), (a 1, b 1, c 3 ), (a 2, b 2, c 1 ) } Yes! The first 3 form a matching. Note that every element in each set will appear in exactly one triple in M. 1.4

5 3-Dimensional Matching Example #2: q = 3 (not part of input technically) W = {a 1, a 2, a 3 } X = {b 1, b 2, b 3 } Y = {c 1, c 2, c 3 } M = { (a 1, b 1, c 1 ), (a 1, b 1, c 2 ), (a 2, b 1, c 3 ), (a 3, b 1, c 1 ) } No! In fact b 2 and b 3 do not appear in any triple. 1.5

6 Vertex Cover, Independent Set and Clique VERTEX COVER (VC) INSTANCE: A Graph G = (V, E) and a positive integer K <= V. QUESTION: Is there a vertex cover of size K or less for G, that is, a subset V V such that V <= K and, for each edge {u,v} E, at least one of u and v belongs to V? K = 3 u v w x z 1.6

7 Vertex Cover, Independent Set and Clique CLIQUE INSTANCE: A Graph G = (V, E) and a positive integer J <= V. QUESTION: Does G contain a clique of size J or more? J = 3 u v w x z 1.7

8 Hamiltonian Circuit (HS) HAMILTONIAN CIRCUIT (HC) INSTANCE: A Graph G = (V, E). QUESTION: Does G have a Hamiltonian circuit, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that {v n, v 1 } E and {v i, v i+1 } E for all i, 1 i n. How is a Hamiltonian circuit different from a tour, as in TSP? TRAVELING SALESMAN INSTANCE: Set C of m cities, distance d(c i, c j ) Z + for each pair of cities c i, c j C positive integer B. QUESTION: Is there a tour of C having length B or less, I.e., a permutation <c (1), c (2),, c (m) > of C such that: m 1 i1 d( c ( i), c ( i 1)) d( c ( m), c (1)) B? 1.8

9 3-Satisfiability A restricted version of satisfiability in which all instances have exactly 3 literals per clause. 3-SATISFIABILITY (3-SAT) INSTANCE: Collection C = {c 1, c 2,,c m } of clauses on a finite set U of variables such that {c j } = 3, for 1 j m. QUESTION: Is there a truth assignment for U that satisfies all the clauses in C? Contrasted with the more general SAT: SATISFIABILITY (SAT) INSTANCE: Collection C of clauses on a finite set U of variables. QUESTION: Is there a truth assignment for U that satisfies all the clauses in C? 1.9

10 3-Satisfiability Theorem 3.1: 3-SAT is NP-Complete. 1) 3-SAT NP 2) SAT 3-SAT I = U,C SAT instance A I = U,C 3-SAT instance I is satisfiable iff I is 1.10

11 3-Satisfiability Each clause c j in I => a set C j of clauses with new variables U j Let c j = {z 1, z 2,,z k } where z i s are literals derived from variables in U Case 1) k=1 c j = {z 1 } U j = {y j1, y j2 } C j = { {z 1, y j1, y j2 }, {z 1, y j1, y j2 }, {z 1, y j1, y j2 }, {z 1, y j1, y j2 } } Case 2) k=2 c j = {z 1, z 2 } U j = {y j1 } C j = { {z 1, z 2, y j1 }, {z 1, z 2, y j1 } } 1.11

12 3-Satisfiability Case 3) k=3 c j = {z 1, z 2, z 3 } U j = {} C j = { {z 1, z 2, z 3 } } Case 4) k>=4 c j = {z 1, z 2,,z k } U j = {y j i 1 i k-3} C j = { {z 1, z 2, y j1 } } U {y ji, z i+2, y j i+1 } 1 i k-4} U {y j k-3, z k-1, z k } } Example: c j = {z 1, z 2, z 3, z 4, z 5, z 6 } U j = { {z 1, z 2, a}, {a, z 3, b}, {b, z 4, c}, {c, z 5, z 6 } } 1.12

13 3-Satisfiability Finally, let: U = U { U j 1 j m } C = { C j 1 j m } Claim: 1) I = U,C is satisfiable iff I = U,C is satisfiable. 2) I can be computed in polynomial time. 1.13

14 Six Basic NP-Complete Problems A review of basic logic and language A if and only if B Equates to: A if B A only if B The if part The only if part Similarly: A if B B only if A B => A A only if B B if A A => B 1.14

15 Six Basic NP-Complete Problems Lastly, to prove: X => Y We frequently show: ~Y => ~X 1.15

16 3-Satisfiability So lets go back to the basic claim: 1) I = U,C is satisfiable iff I = U,C is satisfiable. a) (if) I = U,C is satisfiable if I = U,C is b) (only if) I = U,C is satisfiable only if I = U,C is 1.16

17 3-Satisfiability 2) I can be computed in polynomial time: Let m = C, n = U case 1) creates 4 clauses case 2) creates 2 clauses case 3) creates 1 clause case 4) creates k-1 clauses, where k is the number of literals in the clause This gives a total of at most k-1 new clauses in I for each clause in I Therefore, there is a total of m*(k-1) clauses in I Since k<=2n, it follows that there are at most 2nm clauses in I, which is O(mn) 1.17

18 3-Satisfiability 3-SAT is an example of what is called a special case of SAT. Some special cases, like 3-SAT, are NP-complete Others are solvable in polynomial time (chapter 4) How about 4-SAT? 5-SAT? N-SAT, for any fixed N>=3? 1-SAT? 2-SAT? 1.18

19 Some More New Problems Not All Equal (NAE) 3-SAT INSTANCE: Collection C = {c 1, c 2,,c m } of clauses on a finite set U of variables such that {c j } = 3, for 1 j m. QUESTION: Is there a truth assignment for U such that each clause in C has at least one true literal and at least one false literal? Fact: NAE 3-SAT is NP-complete 1.19

20 Some More New Problems Examples: { (a, b, c), (a, b, c), (a, b, c) } a = T, b = T, c = F { (a, b, c), (a, b, c), (a, b, c), Not even satistfiable (a, b, c), (a, b, c), (a, b, c) (a, b, c), (a, b, c) } { (a, b, c), (a, b, c), (a, b, c), Satistfiable, but not with (a, b, c), (a, b, c), (a, b, c) at least one literal true and false (a, b, c) } per clause 1.20

21 More Problems, cont. Hypergraph 2-Colorability (H2C) INSTANCE: Hypergraph H = (V, E), where 2 e i 3, for all e i E. QUESTION: Is H 2-colorable? In other words, is there a function f : V {0, 1} such that for all e i E there exist vertices u,v e i such that f(u) f(v)? Examples: No Yes No 1.21

22 H2C is NP-Complete Theorem: H2C is NP-complete H2C NP Not All Equal 3-SAT H2C I = U,C NAE3SAT instance A H = (V, E) H2C instance 1.22

23 H2C is NP-Complete, cont. Suppose I = U,C where U = {u 0, u 1,,u n-1 } and C = {c 0, c 1,,c m-1 } Construct H = (V, E) where: V = { v i1 u i U} {v i2 u i U} v i1 corresponds to u i v i2 corresponds to the complement of u i E = E 1 E 2 where E 1 = { (v 1, v 2, v 3 ) c i C, c i = (z 1, z 2, z 3 ) and v 1, v 2, v 3 are the vertices corresponding to the literals z 1, z 2 and z 3, respectively} E 2 = { (v i1, v i2 ) 0 i n-1} 1.23

24 H2C is NP-Complete, cont. Example: U = {u 0, u 1, u 2, u 3 } C = { (u 0, u 1, u 2 ), (u 0,u 2, u 3 ), (u 1, u 2, u 3 ) } 1 0 v 0 v 0 v v v v v v 3 0 The resulting hyper-graph is 2-colorable IFF there is a NAE satisfying truth assignment for the Boolean expression. (if) Suppose there is a NAE assignment, then there is a 2-coloring. (only if) Suppose there is a 2-coloring, then there is a NAE satisfying assignment. 1.24

25 3-Dimensional Matching 3-DIMENSIONAL MATCHING (3-DM) INSTANCE: A set M W X Y, where W, X and Y are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? Example #1: q = 3 (not part of input technically) W = {a 1, a 2, a 3 } X = {b 1, b 2, b 3 } Y = {c 1, c 2, c 3 } M = { (a 1, b 2, c 1 ), (a 2, b 1, c 3 ), (a 3, b 3, c 2 ), (a 1, b 1, c 3 ), (a 2, b 2, c 1 ) } Yes! The first 3 form a matching. Note that every element in each set will appear in exactly one triple in M. 1.25

26 3-Dimensional Matching Example #2: q = 3 (not part of input technically) W = {a 1, a 2, a 3 } X = {b 1, b 2, b 3 } Y = {c 1, c 2, c 3 } M = { (a 1, b 1, c 1 ), (a 1, b 1, c 2 ), (a 2, b 1, c 3 ), (a 3, b 1, c 1 ) } No! In fact b 2 and b 3 do not appear in any triple. 1.26

27 3-Dimensional Matching Theorem: 3DM is NP-complete. Proof: 1) 3DM NP 2) 3-SAT 3DM I = U,C 3-SAT instance A I = W,X,Y,M 3DM instance 1.27

28 3-Dimensional Matching Suppose I = U,C where: U = {u 1, u 2,,u n } C = {c 1, c 2,,c m } Construct M as follows. M will consists of three types of components: Truth Setting and fan-out Satisfaction testing Garbage collection 1.28

29 3-Dimensional Matching Suppose I = U,C where: U = {u 1, u 2,,u n } C = {c 1, c 2,,c m } Construct M as follows. M will consists of three types of components: Truth Setting and fan-out Satisfaction testing Garbage collection 1.29

30 Truth Setting and Fan Out Components For each variable u i U, create the following (TS&FO) elements: T it = { (u i [j], a i [j], b i [j]) 1 j m } U T if = { (u i [j], a i [j+1], b i [j]) 1 j m-1 } U { (u i [m], a i [1], b i [m]) } a big blob for each Boolean variable (suppose m=4): 1.30

31 Truth Setting and Fan Out Components Observation #1: Creates 2mn elements in W, mn in X and mn in Y. u i [j] represents the fact that variable u i could occur in clause c j. Similarly for u i [j] Observation #2: a i [j] and b i [j] both occur in exactly two triples and nowhere else. Observation #3: This tells us that in any matching M M we must have all white or all shaded triples, in other words, all triples from T it or all triples from T i f This corresponds to setting the variable false or true, respectively. 1.31

32 Satisfaction Testing Components For each clause c j C, create (ST) triples: C j = { (u i [j], s 1 [j], s 2 [j]) u i c j } U { (u i [j], s 1 [j], s 2 [j]) u i c j } Observation #4: This adds m elements to X, and m elements to Y. Observation #5: For each j, 1 j m, s 1 [j] and s 2 [j] both appear in exactly three triples and nowhere else. Observation #6: Any matching must choose exactly one triple from C j This corresponds to making that literal true, thereby satisfying the clause. This gives a total of m triples selected from the ST triples. Observation #7: Any matching will contain exactly one triple from C j if and only if there is a satisfying truth assignment for the 3-SAT instance. 1.32

33 Garbage Collection Components For each clause c j C, create 2mn(n-1) (GC) triples: G = { (u i [j], g 1 [k], g 2 [k]), (u i [j], g 1 [k], g 2 [k]) 1 k m(n-1), 1 i n, 1 j m} Observation #8: This adds m(n-1) elements to X, and to Y. Observation #9: For each k, 1 k m(n-1), g 1 [k] and g 2 [k] both appear in exactly 2mn triples and nowhere else. Observation #10: Consequently, exactly m(n-1) triples from G must occur in any matching for M. Observation #11: W = X = Y = 2mn = q. 1.33

34 Garbage Collection Components The resulting instance I of 3DM has a matching IFF the 3-SAT instance I is satisfiable. (if) Suppose you have a satisfying truth assignment for I. How do you construct the matching? (only if) Suppose you have a matching. What is the satisfying truth assignment? 1.34

35 Vertex Cover, Independent Set and Clique VERTEX COVER (VC) INSTANCE: A Graph G = (V, E) and a positive integer K <= V. QUESTION: Is there a vertex cover of size K or less for G, that is, a subset V V such that V <= K and, for each edge {u,v} E, at least one of u and v belongs to V? K = 3 u v w x z 1.35

36 Vertex Cover, Independent Set and Clique INDEPENDENT SET (IS) INSTANCE: Graph G = (V, E) and positive integer J <= V. QUESTION: Does G contain an independent set of size J or more? J = 2 u v w x z 1.36

37 Vertex Cover, Independent Set and Clique CLIQUE INSTANCE: A Graph G = (V, E) and a positive integer J <= V. QUESTION: Does G contain a clique of size J or more? J = 3 u v w x z 1.37

38 Vertex Cover, Independent Set and Clique Lemma 3.1: For any graph G=(V,E) and subset V V, the following statements are equivalent: (a) V is a vertex cover for G (b) V-V is an independent set for G (c) V-V is a clique in the complement G c of G, where G c =(V,E c ) with E c ={{u,v} u,v V and {u,v} E} u v w u v w u v w x z x z x z 1.38

39 Vertex Cover, Independent Set and Clique Lemma 3.1: For any graph G=(V,E) and subset V V, the following statements are equivalent: (a) V is a vertex cover for G (b) V-V is an independent set for G (c) V-V is a clique in the complement G c of G, where G c =(V,E c ) with E c ={{u,v} u,v V and {u,v} E} u v w u v w u v w x z x z x z 1.39

40 Vertex Cover Theorem: VERTEX COVER is NP-complete Proof: 1) VC NP 2) 3SAT VC Suppose I = U,C is an instance of 3SAT where: U = {u 1, u 2,,u n } C = {c 1, c 2,,c m } We will construct a graph G=(V,E) and a positive integer K<= V such that G has a vertex cover of size K or less if and only if C is satistfiable. 1.40

41 Vertex Cover For each variable u i U, there is a truth-setting component T i =(V i,e i ) where: V i = {u i, u i } and E i = {{u i,u i }} *Note that any vertex cover will have to contain at least one of u i and its complement. For each clause c j C, create a satisfaction testing component S j = (V j, E j ): V j = { a 1 [j], a 2 [j], a 3 [j] } E j = { {a 1 [j], a 2 [j]}, {a 1 [j], a 3 [j]}, {a 2 [j], a 3 [j]} } *Note that any vertex cover will have to contain at least two vertices from V j in order to cover the edges in E j. 1.41

42 Vertex Cover For each clause c j C, let the three literals in c j be x j, y j, z j. Then add communication edges: E j = { {a 1 [j], x j }, {a 2 [j], y j }, {a 3 [j], z j } } Finally, let K = n + 2m. 1.42

43 Vertex Cover Example: U = {u 1, u 2, u 3, u 4 } C = { (u 1, u 3, u 4 ), (u 1, u 2, u 4 ) } K = n + 2m =

44 Vertex Cover Observations: The transformation can be performed in polynomial-time. G = (V, E) will have a vertex cover of size K or less IFF I = U,C is satisfiable. (if) Suppose I = U,C is satisfiable (only if) Suppose G = (V, E) has a vertex cover of size K or less 1.44

45 Vertex Cover, Independent Set and Clique Recall Lemma 3.1: Lemma 3.1: For any graph G=(V,E) and subset V V, the following statements are equivalent: (a) V is a vertex cover for G (b) V-V is an independent set for G (c) V-V is a clique in the complement G c of G, where G c =(V,E c ) with E c ={{u,v} u,v V and {u,v} E} What does the previous result say about the independent set and clique problems? 1.45

46 Homework Problems INDEPENDENT SET (IS) INSTANCE: Graph G = (V, E) and positive integer J <= V. QUESTION: Does G contain an independent set of size J or more? Prove IS NP-complete by giving a transformation from 3DM. 4-DIMENSIONAL MATCHING (4-DM) INSTANCE: A set M W X Y Z, where W, X, Y and Z are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? Prove 4-DM NP-complete by giving a transformation from 3DM. 1.46

47 Hamiltonian Circuit (HS) HAMILTONIAN CIRCUIT (HC) INSTANCE: A Graph G = (V, E). QUESTION: Does G have a Hamiltonian circuit, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that {v n, v 1 } E and {v i, v i+1 } E for all i, 1 i n. How is a Hamiltonian circuit different from a tour, as in TSP? 1.47

48 Hamiltonian Circuit (HS) HAMILTONIAN CIRCUIT (HC) INSTANCE: A Graph G = (V, E). QUESTION: Does G have a Hamiltonian circuit, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that {v n, v 1 } E and {v i, v i+1 } E for all i, 1 i n. VERTEX COVER (VC) INSTANCE: A Graph G = (V, E) and a positive integer K <= V. QUESTION: Is there a vertex cover of size K or less for G, that is, a subset V V such that V <= K and, for each edge {u,v} E, at least one of u and v belongs to V? 1.48

49 Hamiltonian Circuit (HC) Theorem: HC is NP-complete Proof: 1) HC NP 2) VC HC Let G = (V,E) and K <= V be an instance of VC. We will construct a graph G = (V,E ) such that G has a vertex cover of size K or less if and only if G has a Hamiltonian circuit. 1.49

50 Hamiltonian Circuit (HC) 1) Add selector vertices a 1, a 2,,a k to V 2) For each edge e = {u,v} in E, construct the following cover testing component: this side corresponds to u this side corresponds to v More specifically, add the following 12 vertices: V e = {(u,e,i), (v,e,i) : 1<=i<=6}} And 14 edges: E e = {{(u,e,i), (u,e,i+1)}, {(v,e,i), (v,e,i+1)} : 1<=i<=5 } {{(u,e,3), (v,e,1)}, {(v,e,3), (u,e,1)}} {{(u,e,6), (v,e,4)}, {(v,e,6), (u,e,4)}} 1.50

51 Hamiltonian Circuit (HC) 3) For each vertex v V, let the edges incident on v be: e v[1], e v[2],,e v[deg(v)] Add the following edges: E v = { {(v,e v[i],6),(v,e v[i+1],1)} : 1<=i<deg(v) } This creates a path in G that touches the cover-testing components corresponding to the edges adjacent to v in G. start This path corresponds to the vertex v from G. end 1.51

52 Hamiltonian Circuit (HC) 4) Finally, add edges to G that connect the first (s i ) and last (e i ) vertices from each of these paths to every one of the selector vertices: E = { {a i,(v,e v[1],1)}, {a i,(v,e v[deg(v)],6)} : 1<=i<=K, v V } s 1 a 1 e 1 a 2 s 2 e 2 : : : : a K s V e V 1.52

53 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 x e 5 z K =

54 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 x e 5 z a 1 K = 3 a 2 a

55 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 K = 3 e 2 a 2 e 3 a 3 e 4 e

56 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 v K = 3 e 2 a 2 e 3 a 3 e 4 e

57 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 e 2 a 2 e 3 a 3 e 4 e

58 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 e 2 a 2 e 3 a 3 x e 4 e

59 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e

60 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.60

61 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.61

62 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.62

63 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.63

64 Hamiltonian Circuit (HC) A complete example consider the following instance of vertex cover: e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 G as a vertex cover of size 3 If and only if G has a Hamiltonian circuit e 5 z 1.64

65 Hamiltonian Circuit (HC) (only if) e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.65

66 Hamiltonian Circuit (HC) (if) e 1 e 2 u v w e 3 e 4 e 1 x e 5 z a 1 u v K = 3 w e 2 a 2 e 3 a 3 x e 4 e 5 z 1.66

67 Hamiltonian Path (HP) HAMILTONIAN PATH (HP) INSTANCE: A Graph G = (V, E), and vertices u,v V. QUESTION: Is there a Hamiltonian path from vertex u to vertex v, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that u = v 1, v = v n, and {v i, v i+1 } E for all i, 1 i n. FACT: HP is NP-complete (see page 60). 1.67

68 HW Problem Solution #1 INDEPENDENT SET (IS) INSTANCE: Graph G = (V, E) and positive integer J <= V. QUESTION: Does G contain an independent set of size J or more? Theorem: IS is NP-complete 1) IS NP 2) 3DM IS I = M,W,X,Y 3DM instance A G = (V, E), J IS instance 1.68

69 HW Problem Solution #1 Proof: Let M be an instance of 3DM, and construct an instance of IS as follows. V = {v i t i M} E = {v i, v j i j and t i and t j agree on some coordinate} j = q Claims: 1) The transformation can be performed in polynomial-time 2) M contains a matching if and only if G=(V,E) contains an independent set of size j. 1.69

70 HW Problem Solution #1 (only if) Let M = {t 0, t 1,,t q-1 } be a matching, where M M. It can be easily verified that V = {v 0, v 1,,v q-1 } is an independent set of size j. (if) Let V = {v 0, v 1,,v q-1 } is an independent set of size j for G. Then it can be easily verified that M = {t 0, t 1,,t q-1 } is a matching. 1.70

71 HW Problem Solution #2 4-DIMENSIONAL MATCHING (4-DM) INSTANCE: A set M W X Y Z, where W, X, Y and Z are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? Theorem: 4DM is NP-complete 1) 4DM NP 2) 3DM 4DM I = M,W,X,Y 3DM instance A I = M,W,X,Y,Z 4DM instance 1.71

72 HW Problem Solution #2 Proof: Let M,W,X, and Y be an instance of 3DM, and construct an instance of 4DM as follows: W = W X = X Y = Y Z = {z i, where 0 i q-1, and q = W = X = Y } M = {(w i,x j,y k,z r ) (w i,x j,y k ) M and z r Z} In other words, for each triple in M and value in Z, add a quadruple in M. Claims: 1) The transformation can be performed in polynomial-time 2) M contains a matching if and only if M does. 1.72

73 HW Problem Solution #2 (only if) Let M = {t 0, t 1,,t q-1 } be a 3D matching, where M M. Create a 4D matching M by adding z i to triple t i, for all i, where 0 i q-1. It can easily be verified that M is a 4D matching for I. (if) Let M = {t 0, t 1,,t q-1 } be a 4D matching, where M M. Create a 3D matching by letting M = {(w i,x j,y k ) (w i,x j,y k,z r ) M }. It can easily be verified that M is a 3D matching for I. 1.73

74 Partition PARTITION INSTANCE: A finite set A and a size s(a) Z + for each a A. QUESTION: Is there a subset A A such that: 3-DIMENSIONAL MATCHING (3-DM) INSTANCE: A set M W X Y, where W, X and Y are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? 1.74

75 Partition Theorem: Partition is NP-complete Proof: 1) PARTITION NP 2) 3DM PARTITION Let W, X, and Y with W = X = Y = q, and M W X Y be an instance of 3DM. Also let: and: W = {w 1, w 2,, w q } X = {x 1, x 2,, x q } Y = {y 1, y 2,, y q } M = {m 1, m 2,, m k } where k = M 1.75

76 1.76 Partition We will construct a set A, and a size s(a) Z + for each a A. This set A will be such that A contains a subset A satisfying: if and only if M contains a matching. ' ' ) ( ) ( A a A A a a s a s

77 Partition First add k elements {a i :1 i k} to A, where a i corresponds with triple m i M, and the size s(a i ) in binary is given by: p log 2 ( k 1) w 1 w 2 w q x 1 x 2 x q y 1 y 2 y q Notes: A triple from M corresponds to a 3-bit binary number, where each bit is in the rightmost position of one sub-field. If all the binary numbers were added up for all the triples, no sub-field would overflow (because of the choice of p). 1.77

78 Partition Let: B 3q 1 i0 2 pj w 1 w 2 w q x 1 x 2 x q y 1 y 2 y q Then B corresponds to the binary number containing a 1 in every sub-field rightmost position. If a set M of triples formed a matching, then the corresponding binary numbers would add up to exactly B. Now add two more elements b 1 and b 2 to A, where: s( b s( b k 1 ) 2 s( ) i1 k 2 ) s( ) i1 a i B a i B 1.78

79 1.79 Partition Note that the sum of the sizes of all the elements in A is: Thus, any partition of A into equal-sizes subsets A and A A, the sum of the sizes of the elements in those two partitions must both equal: Furthermore, b 1 and b 2 cannot be in the same set. k i a i B s b s 1 1 ) ( 2 ) ( k i a i B s b s 1 2 ) ( ) ( k i i k i i k i i k i i k i i a s B a s B a s a s b s b s a s ) ( 4 ) ) ( ( ) ) ( (2 ) ( ) ( ) ( ) ( k i a i s 1 ) ( 3 k i a i s 1 ) ( 2

80 Partition Thus, in any partition A of A into two equal-sized subsets, b1 and b2 must be in different sets: s A A A k ( b1 ) 2 s( a i ) B s( b ) i1 For the sum of the sizes of the two sets A and A A to sum to: 2 k i1 s( a i k 2 s( ) i1 a i B We therefore need to add some subset of the remaining elements of A to A that sums to B. ) As noted previously, the corresponding triples would have to be a matching for M. Consequently, the set A can be partitioned into two equal-sized subsets IFF M contains a matching. 1.80

81 Techniques for Proving NP-Completeness There are several general types of transformations: Restriction Local Replacement Component Design 1.81

82 Restriction Proving a problem NP NP-complete by restriction consists of showing that contains another known NP-complete problem as a special case. What is a special case of a problem? Let I be all instances of a problem, and let I be all instances of a problem If I I and I Y IFF I Y then is said to be a special case of Furthermore, if and are both in NP, and if is NP-complete, then so is The heart of a proof by restriction is the specification of restrictions to be placed on the instances of so that the resulting problem will be identical to Note that a proof by restriction is still a transformation, i.e., it is not a different kind of NPcompleteness proof. 1.82

83 Directed Hamiltonian Circuit (DHS) DIRECTED HAMILTONIAN CIRCUIT (DHC) INSTANCE: A directed graph G = (V, E). QUESTION: Does G have a Hamiltonian circuit, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that {v n, v 1 } E and {v i, v i+1 } E for all i, 1 i n. HAMILTONIAN CIRCUIT (HC) INSTANCE: A Graph G = (V, E). QUESTION: Does G have a Hamiltonian circuit, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that {v n, v 1 } E and {v i, v i+1 } E for all i, 1 i n. Fact: HC is NP-complete. 1.83

84 Directed Hamiltonian Circuit (DHS) Theorem: DHC is NP-complete Proof: 1) DHC NP 2) HC DHC, by restriction. Proof: Consider those instances of DHC where there is an edge from u to v if and only if there is an edge from v to u. Recall that technically a restriction is, in fact, a transformation. In what sense is the above proof a transformation? 1.84

85 Exact Cover by 3-Sets (X3C) EXACT COVER BY 3-SETS (X3C) INSTANCE: A finite set X with X = 3q and a collection C of 3-element subsets of X. QUESTION: Does C contain an exact cover for X, that is, a sub-collection C C such that every element of X occurs in exactly one member of C? How does this differ from 3DM? 3-DIMENSIONAL MATCHING (3DM) INSTANCE: A set M W X Y, where W, X and Y are disjoint sets having the same number q of elements. QUESTION: Does M contain a matching, that is, a subset M M such at M = q and no two elements of M agree in any coordinate? 1.85

86 Exact Cover by 3-Sets (X3C) Theorem: X3C is NP-complete Proof: 1) X3C NP 2) 3DM X3C Let M W Y Z, where W, Y and Z are disjoint sets having the same number q of elements, be an instance of 3DM. Construct an instance of X3C as follows: Let X = W Y Z Let C = { {a,b,c} (a,b,c) M} It can easily be verified that C contains an exact 3-cover for X IFF M contains a matching. 1.86

87 Exact Cover by 3-Sets (X3C) Theorem: X3C is NP-complete Proof: 1) X3C NP 2) 3DM X3C Another way to view this is as a restriction from X3C to 3DM consider those instances of X3C where X can be partitioned into 3 disjoint sets W, Y, and Z, and where each 3-element set C contains exactly one element from each of these sets. 1.87

88 MINIMUM COVER EXACT COVER BY 3-SETS (X3C) INSTANCE: A finite set X with X = 3q and a collection C of 3-element subsets of X. QUESTION: Does C contain an exact cover for X, that is, a sub-collection C C such that every element of X occurs in exactly one member of C? 1.88

89 Hitting Set HITTING SET INSTANCE: A collection C of subsets of a set S, positive integer K. QUESTION: Does S contain a hitting set for C of size K or less, that is, a subset S S with S K and such that S contains at least one element from each subset in C? Proof: Restrict to VC by considering those instances having c = 2 for all c C. VERTEX COVER (VC) INSTANCE: A Graph G = (V, E) and a positive integer K <= V. QUESTION: Is there a vertex cover of size K or less for G, that is, a subset V V such that V <= K and, for each edge {u,v} E, at least one of u and v belongs to V? 1.89

90 Subgraph Isomorphism SUBGRAPH ISOMORPHISM INSTANCE: Two graphs G = (V 1, E 1 ) and H = (V 2, E 2 ). QUESTION: Does G contain a subgraph isomorphic to H, that is, a subset V V 1 and a subset E E 1 such that V = V 2, E = E 2, and there exists a one-to-one function f : V 2 V satisfying {u,v} E 2 if and only if {f(u),f(v)} E. Proof: Restrict to CLIQUE by considering those instances where H is a complete graph, that is, E 2 contains all possible edges joining two members of V 2. CLIQUE INSTANCE: A Graph G = (V, E) and a positive integer J <= V. QUESTION: Does G contain a clique of size J or more? 1.90

91 Bounded Degree Spanning Tree Let G=(V,E) be a graph and let E E. Then T=(V,E ) is said to be a spanning tree if T is connected and E = V -1 (i.e., T contains no cycles). BOUNDED DEGREE SPANNING TREE INSTANCE: A graph G = (V, E) and a positive integer K V -1. QUESTION: Is there a spanning tree for G in which no vertex has degree exceeding K, that is, a subset E E such that E = V -1, the graph G = (V, E ) is connected, and no vertex in V is included in more than K edges from E? Proof: Restrict to HAMILTONIAN PATH by considering only instances where K=2. HAMILTONIAN PATH (HP) INSTANCE: A Graph G = (V, E), and vertices u,v V. QUESTION: Is there a Hamiltonian path from vertex u to vertex v, that is an ordering < v 1, v 2,,v n > of the vertices of G, where n = V, such that u = v 1, v = v n, and {v i, v i+1 } E for all i, 1 i n. FACT: HP is NP-complete (see page 60). 1.91

92 Minimum Equivalent Digraph MINIMUM EQUIVALENT DIGRAPH INSTANCE: A directed graph G = (V, A) and a positive integer K A. QUESTION: Is there a directed graph G = (V, A ) such that A A, A K, and such that, for every pair of vertices u and v in V, G contains a directed path from u to v if and only if G contains a directed path from u to v. Proof: Restrict to DIRECTED HAMILTONIAN CIRCUIT by considering those instances where G is strongly connected, that is, contains a path from every vertex u to every vertex v, and K= V. Note that this is technically a restriction to DIRECTED HAMILTONIAN CIRCUIT FOR STRONGLY CONNECTED DIGRAPHS, but the NP-completeness of that problem follows immediately from the constructions given for HC and DIRECTED HC. 1.92

93 Knapsack KNAPSACK INSTANCE: A finite set U and a size s(u) Z + and a value s(v) Z + for each u U, a size constraint B Z +, and a value goal K Z +. QUESTION: Is there a subset U U such that: and Proof: Restrict to PARTITION by considering only those instances where s(u) = v(u) for all u U and: PARTITION uu ' s( u) B uu ' uu INSTANCE: A finite set A and a size s(a) Z + for each a A. QUESTION: Is there a subset A A such that: v( u) K B K 1/ 2 s( u) aa' s( a) aa A' s( a) 1.93

94 Knapsack KNAPSACK INSTANCE: A finite set U and a size s(u) Z + and a value s(v) Z + for each u U, and a value goal K Z +. QUESTION: Is there a subset U U such that: uu ' v( u) K 1.94

95 Multiprocessor Scheduling MULTIPROCESSOR SCHEDULING INSTANCE: A finite set A of tasks, a length l(a) Z + for each a A, a number m Z + of processors, and a deadline D Z +. QUESTION: Is there a partition A = A 1 A 2 A m of A into m disjoint sets such that: max{ a A i l( a) :1 i m} D? Proof: Restrict to PARTITION by considering only those instances where m=2 and: D 1/ 2 l( a) aa PARTITION INSTANCE: A finite set A and a size s(a) Z + for each a A. QUESTION: Is there a subset A A such that: aa' s( a) aa A' s( a) 1.95

96 Invalid Restriction Subset Sum INSTANCE: Integers a 1, a 2,,a n and integer B. QUESTION: Is there a sequence of 0 s and 1 s, x 1, x 2,,x n such that: Fact: Subset Sum is NP-complete n i1 a i x i B? Real Subset Sum INSTANCE: Integers a 1, a 2,,a n and integer B. QUESTION: Is there a sequence of real numbers x 1, x 2,,x n such that: n i1 a i x i B? 1.96

97 Invalid Restriction, Cont. Claim (from a Ph.D disseration): Subset Sum is just a special case of Real Subset Sum Therefore, Real Subset Sum is NP-complete The proof is in error since it restricts the question, and not the instances. In fact, the problem is actually trivially solvable in polynomial time simply taking x 1 = B/a 1 and x i = 0, for all i, where 2 i n. Also note that the restriction does not preserve yes and no instances (in fact, all are yes instances). 1.97

98 Other Invalid Restrictions Some other invalid restrictions: 4-DM is NP-complete because it contains 3-DM as a special case. 4-SAT is NP-complete because it contains 3-SAT as a special case. In these cases, the instances of one aren t even a subset of the other! 1.98

99 Homework #2 See page 75, the first group: Longest Path Set Packing Partition into Hamiltonian Subgraphs Largest Common Subgraph Minimum Sum of Squares Give an NP-completeness proof for each of the above by restriction. Hand in any two. 1.99

100 Local Replacement Transformations are: sufficiently non-trivial to warrant spelling out details (in contrast to restriction) still relatively simple First example transforming SAT to 3-SAT: each clause in SAT was locally replaced by 1 or more clauses in 3-SAT Replacement of one clause was independent of all other clauses 1.100

101 Partition into Triangles Partition into Triangles INSTANCE: A Graph G = (V, E), with V = 3q. QUESTION: Is there a partition of V into q disjoint sets V 1, V 2,,V q of three vertices each such that, for each V i = {v i[1], v i[2], v i[3] } the three edges {v i[1], v i[2] }, {v i[1], v i[3] } and {v i[2], v i[3] } all belong to E? Theorem: Partition into Triangles is NP-complete Proof: 1) Partition into Triangles NP 2) X3C Partition into Triangles 1.101

102 Local Replacement Let set X with X = 3q and let C be a collection of 3-element subsets of X. Construct a graph G = (V,E) as follows: For each set c i = {x i,y i,z i } in C, create the following vertices and edges: That s it! 1.102

103 3-Dimensional Matching Can you draw the graph? q = 3 (not part of input technically) X = {a 1, a 2, a 3, b 1, b 2, b 3, c 1, c 2, c 3 } M = { (a 1, b 2, c 1 ), (a 2, b 1, c 3 ), (a 3, b 3, c 2 ), (a 1, b 1, c 3 ), (a 2, b 2, c 1 ) } Note the first 3 form an exact 3-cover

104 Local Replacement The resulting graph G = (V,E) can be partitioned into triangles IFF there is a sub-collection C C such that every element of X occurs in exactly one member of C. (if) Suppose there is a sub-collection C C such that every element of X occurs in exactly one member of C. Consider each triple from C. in C not in C {x 1,y 1,z 1 } {x 2,y 2,z 2 } {x 3,y 3,z 3 } {x k,y k,z k } It is easy to see that this gives a partition of G = (V,E) into triangles

105 Local Replacement (only if) Suppose the resulting graph G = (V,E) can be partitioned into triangles. Note that each component must be partitioned into one of two ways. can t happen in a valid partition {x 1,y 1,z 1 } {x 2,y 2,z 2 } In the first case, put the corresponding triple in C, and in the second case, don t. It is easy to see that the resulting sub-collection C C is an exact cover for X

106 Component Design The most complicated of the 3 types of proofs. Constructs components in the target instance from components of the given instance: not a simple one-to-one mapping, in contrast to local replacement typically between different types of problems Components serve two purposes typically: making choices selecting vertices or truth assignments testing properties making sure all edges are covered or that clauses are satisfied Additionally, the choice-components are connected to the testing components in such as way to ensure choices satisfy required properties. Examples: 3 Dimensional Matching (slide 27) Vertex Cover (slide 40) or is this local replacement? Hamiltonian Circuit (slide 49) 1.106

107 Component Design MINIMUM TARDINESS SEQUENCING INSTANCE: A finite set T of tasks, each t T having length 1 and a deadline d(t) Z +, a partial order on T, and a non-negative integer K T. QUESTION: Is there a schedule σ:t {0,1,, T -1} such that σ(t) σ(t ) whenever t t, such σ(t) < σ(t ) whenever t t, and such that {t T: σ(t)+1>d(t)} K? Theorem: Minimum Tardiness Sequencing is NP-complete Proof: 1) Minimum Tardiness Sequencing NP 2) CLIQUE Minimum Tardiness Sequencing Let the graph G=(V,E) and positive integer J V be an arbitrary instance of CLIQUE. Construct an instance of Minimum Tardiness Sequencing as follows

108 Component Design 1.108

109 Component Design Observation #1: A schedule is an ordering of the tasks, from 0 to T -1. Consequently, any schedule for the tasks will look as follows: 0 J(J+1)/2 T = V + E (edge task deadline) We need to show that the G = (V,E) contains a clique of size J IFF the resulting set of tasks has a minimum tardiness schedule

110 Component Design (only if) Suppose the graph G contains a clique of size J. Construct a schedule as follows: clique vertices clique edges J vertex tasks J(J-1)/2 edge tasks V - J vertex tasks E - J(J-1)/2 edge tasks 0 J(J+1)/2 T = V + E (edge task deadline) 1.110

111 Component Design (if) Suppose there is a schedule σ:t {0,1,, T -1} such that σ(t) σ(t ) whenever t t, such σ(t) < σ(t ) whenever t t, and such that {t T: σ(t)+1>d(t)} K? Observation #1: At least J(J-1)/2 of the edge tasks must be scheduled before the edge deadline; otherwise, too many edge tasks would be late. J vertex tasks J(J-1)/2 edge tasks 0 J(J+1)/2 T = V + E (edge task deadline) Observation #2: At least J vertex tasks must also occur prior to the edge task deadline; this is because the minimum number of vertices on J(J-1)/2 edges is J

112 Component Design Observation #3: Since there are at least J vertex tasks, and at least J(J-1)/2 edge tasks prior to the deadline J(J+1)/2, it follows there must be exactly J vertex tasks and exactly J(J-1)/2 edge tasks prior to the deadline J(J+1)/2. Observation #4: The J vertex tasks must also occur before the J(J-1)/2 edge tasks because of the partial order J vertex tasks J(J-1)/2 edge tasks 0 J(J+1)/2 T = V + E (edge task deadline) It follows that the J vertex tasks and the J(J-1)/2 edge tasks form a clique in the graph G=(V,E)

113 Homework #3 Which category would these be? 3-SAT (slide 10) Hypergraph 2-colorability (slide 22) Clique (slide 45) Independent Set (slide 45) Independent Set (slide 68) 4-Dimensional Matching (71) Partition (slide 75) 1.113