PTAS for Matroid Matching

Transcription

1 PTAS for Matroid Matching Jon Lee 1 Maxim Sviridenko 1 Jan Vondrák 2 1 IBM Watson Research Center Yorktown Heights, NY 2 IBM Almaden Research Center San Jose, CA May 6, 2010 Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 1 / 19

2 Why matroid matching? Classical combinatorial optimization problems: Max-weight bipartite matching [Hungarian method, 1950 s] Max-weight independent set in a matroid [Rado, 1950 s] Max-weight non-bipartite matching [Edmonds, 1960 s] Max-weight independent set in the intersection of two matroids [Edmonds/Lawler 1970 s] Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 2 / 19

3 Why matroid matching? Classical combinatorial optimization problems: Max-weight bipartite matching [Hungarian method, 1950 s] Max-weight independent set in a matroid [Rado, 1950 s] Max-weight non-bipartite matching [Edmonds, 1960 s] Max-weight independent set in the intersection of two matroids [Edmonds/Lawler 1970 s] Matroid matching: proposed by Lawler as a common generalization Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 2 / 19

4 Matroids Definition A matroid on M = (N, I) is a system of independent sets such that 1 I 2 J I; I J I I 3 I, J I; I < J j J \ I; I {j} I. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 3 / 19

5 Matroids Definition A matroid on M = (N, I) is a system of independent sets such that 1 I 2 J I; I J I I 3 I, J I; I < J j J \ I; I {j} I. Examples: linear matroid (independent sets = linearly independent vectors) graphic matroid (independent sets = acyclic subgraphs) Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 3 / 19

6 Matroid Matching Given: Graph G = (V, E), matroid M = (V, I). Find: A matching M in G such that V (M) is independent in M. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 4 / 19

7 Matroid Matching Given: Graph G = (V, E), matroid M = (V, I). Find: A matching M in G such that V (M) is independent in M. Note: Matroid matching is equivalent to its special case, where G itself is a matching. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 4 / 19

8 Matroid parity Given: Matroid M = (N, I), N partitioned into disjoint pairs p 1,..., p n. Find: A subset I [n] such that i I p i is independent in M. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 5 / 19

9 Matroid parity Given: Matroid M = (N, I), N partitioned into disjoint pairs p 1,..., p n. Find: A subset I [n] such that i I p i is independent in M. Reduction from matroid matching: Given G = (V, E), replace each edge e = (u, v) by two unique elements (u e, v e ). For each vertex v, simulate the matching condition by defining {v e : v e} to be parallel copies of v in the matroid M. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 5 / 19

10 Special cases of matroid parity Matroid intersection: Given M 1, M 2, find the largest set independent in both matroids. M 1 M 2 Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 6 / 19

11 Special cases of matroid parity Matroid intersection: Given M 1, M 2, find the largest set independent in both matroids. M 1 M 2 Matching in non-bipartite graphs: obviously a special case of matroid matching. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 6 / 19

12 The matchoid problem Given: Graph G = (V, E), matroid M v = (E v, I v ) for each v V. Find: A set of edges F E such that for each vertex v V, the incident edges F E v are independent in M v. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 7 / 19

13 The matchoid problem Given: Graph G = (V, E), matroid M v = (E v, I v ) for each v V. Find: A set of edges F E such that for each vertex v V, the incident edges F E v are independent in M v. Note: The matchoid problem is a special case of matroid matching, and it still generalizes matroid intersection and non-bipartite matching. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 7 / 19

14 Complexity status overview matroid matching matroid parity matchoid non-b. matching matroid intersection bipartite matching Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 8 / 19

15 Complexity status overview matroid matching matroid parity NP-hard NP-hard matchoid NP-hard non-b. matching matroid intersection P bipartite matching P Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 8 / 19

16 Good news Definition A matroid M = (N, I) is linear if there are vectors {v i : i N) such that I I iff {v i : i I} are linearly independent. Lászlo Lovász (1980): Matroid matching of maximum cardinality can be found in polynomial time, if M is a linear matroid. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 9 / 19

17 Good news Definition A matroid M = (N, I) is linear if there are vectors {v i : i N) such that I I iff {v i : i I} are linearly independent. Lászlo Lovász (1980): Matroid matching of maximum cardinality can be found in polynomial time, if M is a linear matroid. Notes: There is also a randomized FPTAS in the weighted case for linear matroids. For general matroids given by an oracle, even unweighted matroid matching requires exponentially many queries to solve optimally. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 9 / 19

18 Approximation? Easy: The feasible solutions to a matroid matching problem form a 2-independence system: If A is feasible and {e} is a feasible edge, then there are edges a, b A such that (A \ {a, b}) {e} is feasible. Jenkyns (1976): For any 2-independence system, the greedy algorithm gives a 1/2-approximation (even in the weighted case). Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 10 / 19

19 Approximation? Easy: The feasible solutions to a matroid matching problem form a 2-independence system: If A is feasible and {e} is a feasible edge, then there are edges a, b A such that (A \ {a, b}) {e} is feasible. Jenkyns (1976): For any 2-independence system, the greedy algorithm gives a 1/2-approximation (even in the weighted case). Fujito (1993): 2/3-approximation for unweighted matroid matching. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 10 / 19

20 Approximation? Easy: The feasible solutions to a matroid matching problem form a 2-independence system: If A is feasible and {e} is a feasible edge, then there are edges a, b A such that (A \ {a, b}) {e} is feasible. Jenkyns (1976): For any 2-independence system, the greedy algorithm gives a 1/2-approximation (even in the weighted case). Fujito (1993): 2/3-approximation for unweighted matroid matching. More generally, matroid matching in k-uniform hypergraphs is a k-independence system = 1/k-approximation. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 10 / 19

21 What we did Theorem (Lee, Sviridenko, V.) 1 There is a PTAS for unweighted matroid matching. 2 In k-uniform hypergraphs, (2/k ɛ)-approximation for any ɛ > 0. Note: Special cases of k-uniform matroid matching are k-set packing (2/k ɛ known by Hurkens-Schrijver) and intersection of k matroids (only 1/k known until recently). Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 11 / 19

22 What we did Theorem (Lee, Sviridenko, V.) 1 There is a PTAS for unweighted matroid matching. 2 In k-uniform hypergraphs, (2/k ɛ)-approximation for any ɛ > 0. Note: Special cases of k-uniform matroid matching are k-set packing (2/k ɛ known by Hurkens-Schrijver) and intersection of k matroids (only 1/k known until recently). Open question: the weighted case. Theorem (Lee, Sviridenko, V.) A natural LP formulation of matroid matching has Ω(n) integrality gap. After r rounds of Sherali-Adams, still Ω(n/r) integrality gap. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 11 / 19

23 Our algorithm PTAS for matroid parity: Local search algorithm, try to add s + 1 new pairs and kick out s. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 12 / 19

24 Our algorithm PTAS for matroid parity: Local search algorithm, try to add s + 1 new pairs and kick out s. If we want to achieve a (1 ɛ)-approximation, we need s(ɛ) = 5 1/ɛ. Hence, running time n 51/ɛ. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 12 / 19

25 Our algorithm PTAS for matroid parity: Local search algorithm, try to add s + 1 new pairs and kick out s. If we want to achieve a (1 ɛ)-approximation, we need s(ɛ) = 5 1/ɛ. Hence, running time n 51/ɛ. Lemma If A, B are feasible solutions of matroid parity and ( A < 1 1 ) B 2t then there is a local improvement for A with s 5 t 1. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 12 / 19

26 Base case: t = 1 We want: If A < 1 2 B, then A can be extended by some pair from B. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 13 / 19

27 Base case: t = 1 We want: If A < 1 2 B, then A can be extended by some pair from B. B 0 A B Proof: Since A < 1 2 B, we can extend A to an independent set A B 0 such that B 0 B and B 0 > 1 2 B. B 0 > B \ B 0, so there must be a whole pair in B 0, which can be added to A. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 13 / 19

28 General case: t > 1 We assume A < (1 1 2t ) B. A 1 A 2 B 0 B 1 A B Extend A to A B 0 ; if B 0 contains a pair, we are done. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 14 / 19

29 General case: t > 1 We assume A < (1 1 2t ) B. A 1 A 2 B 0 B 1 A B Extend A to A B 0 ; if B 0 contains a pair, we are done. Let B 1 be paired up with B 0, and find A 1 A such that A 1 and B 1 can be swapped in the contracted matroid M/B 0. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 14 / 19

30 General case: t > 1 We assume A < (1 1 2t ) B. A 1 A 2 B 0 B 1 A B Extend A to A B 0 ; if B 0 contains a pair, we are done. Let B 1 be paired up with B 0, and find A 1 A such that A 1 and B 1 can be swapped in the contracted matroid M/B 0. If A 1 contains a pair, kick it out and add two pairs from B 0 B 1. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 14 / 19

31 General case: t > 1 We assume A < (1 1 2t ) B. A 1 A 2 B 0 B 1 A B Extend A to A B 0 ; if B 0 contains a pair, we are done. Let B 1 be paired up with B 0, and find A 1 A such that A 1 and B 1 can be swapped in the contracted matroid M/B 0. If A 1 contains a pair, kick it out and add two pairs from B 0 B 1. Otherwise, let A 2 be paired up with A 1, and recurse on A = A \ (A 1 A 2 ), B = B \ (B 0 B 1 ) in M/(B 0 B 1 ). Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 14 / 19

32 Inductive argument We have A < (1 1 2t 2 ) B. B 0 A 1 B 1 A 2 A B B Ã By induction there is a local improvement A \ Ã B in M/(B 0 B 1 ) such that B 5 t 1. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 15 / 19

33 Inductive argument We have A < (1 1 2t 2 ) B. B 0 A 1 B 1 A 2 A B B Ã By induction there is a local improvement A \ Ã B in M/(B 0 B 1 ) such that B 5 t 1. This set can be extended by pairs from A 1 A 2, but we miss up to 4 B pairs, as we must take whole pairs and keep independence. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 15 / 19

34 Inductive argument We have A < (1 1 2t 2 ) B. B 0 A 1 B 1 A 2 A B B Ã By induction there is a local improvement A \ Ã B in M/(B 0 B 1 ) such that B 5 t 1. This set can be extended by pairs from A 1 A 2, but we miss up to 4 B pairs, as we must take whole pairs and keep independence. However, whatever we miss in A 1 A 2, we can make up in B 0 B 1. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 15 / 19

35 Inductive argument We have A < (1 1 2t 2 ) B. B 0 A 1 B 1 A 2 A B B Ã By induction there is a local improvement A \ Ã B in M/(B 0 B 1 ) such that B 5 t 1. This set can be extended by pairs from A 1 A 2, but we miss up to 4 B pairs, as we must take whole pairs and keep independence. However, whatever we miss in A 1 A 2, we can make up in B 0 B 1. In total, we gain at least one pair and the swap size is 5 B 5 t. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 15 / 19

36 LP relaxations What to do about the weighted matroid matching problem? Local search - we are not sure how to analyze it. LP relaxations? Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 16 / 19

37 LP relaxations What to do about the weighted matroid matching problem? Local search - we are not sure how to analyze it. LP relaxations? Natural LP: max w e y e : e E x u r M (S); S V u S x u = y e ; u e E, x u, y e 0. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 16 / 19

38 Integrality gap Example: Pairs (u 1, v 1 ),..., (u n, v n ). Matroid M t : a set S is independent if it contains at most t pairs, and any number of singletons from the remaining pairs. i T max n y i : i=1 y i 1 ( T + t); T, 2 y i 0. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 17 / 19

39 Integrality gap Example: Pairs (u 1, v 1 ),..., (u n, v n ). Matroid M t : a set S is independent if it contains at most t pairs, and any number of singletons from the remaining pairs. i T max n y i : i=1 y i 1 ( T + t); T, 2 y i 0. Optimal solution: S = {(u 1, v 1 ), (u 2, v 2 ),..., (u t, v t )}, value t. Fractional solution: y e = 1/2 for each pair e = (u i, v i ), value n/2. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 17 / 19

40 Sherali-Adams Can we strengthen this LP? Sherali-Adams hierarchy: For some r 1, consider all disjoint I, J V such that I + J r. Multiply each constraint by i I y i j J (1 y j), expand the products and replace each monomial l L y k l l by y L. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 18 / 19

41 Sherali-Adams Can we strengthen this LP? Sherali-Adams hierarchy: For some r 1, consider all disjoint I, J V such that I + J r. Multiply each constraint by i I y i j J (1 y j), expand the products and replace each monomial l L y k l l by y L. We get constraints like: y l y i l T i I j J(1 y j ) 1 2 (t + T ) i I expand and replace monomials by y L. y i (1 y j ). j J Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 18 / 19

42 Sherali-Adams Can we strengthen this LP? Sherali-Adams hierarchy: For some r 1, consider all disjoint I, J V such that I + J r. Multiply each constraint by i I y i j J (1 y j), expand the products and replace each monomial l L y k l l by y L. We get constraints like: y l y i l T i I j J(1 y j ) 1 2 (t + T ) i I expand and replace monomials by y L. y i (1 y j ). Fractional solution: y L = 1/2 L is still feasible for this LP, for t = r. LP r n/2, while OPT = t = r. Hence, the gap is Ω(n/r). j J Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 18 / 19

43 Conclusion We showed a PTAS for unweighted matroid matching. Is the following true? For any ɛ > 0, there exists s(ɛ) such that local search with swap size s(ɛ) gives a (1 ɛ)-approximation for weighted matroid matching. Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 19 / 19

44 Conclusion We showed a PTAS for unweighted matroid matching. Is the following true? For any ɛ > 0, there exists s(ɛ) such that local search with swap size s(ɛ) gives a (1 ɛ)-approximation for weighted matroid matching. More generally, can we get (2/k ɛ)-approximation for weighted k-uniform matroid matching? Or at least for the weighted matchoid problem? (we have a 2/3-approximation for k = 2) Jan Vondrák (IBM Almaden) PTAS for Matroid Matching 19 / 19