Local Algorithms for Sparse Spanning Graphs

Transcription

1 Local Algorithms for Sparse Spanning Graphs Reut Levi Dana Ron Ronitt Rubinfeld Intro slides based on a talk given by Reut Levi

2 Minimum Spanning Graph (Spanning Tree)

3 Local Access to a Minimum Spanning Graph (Spanning Tree) Oracle

4 Local Access to a Minimum Spanning Graph (Spanning Tree) Consistency with the same tree

5 Local Algorithm for Minimum Spanning Graph (Spanning Tree) Given parameter, and query access to over vertices and maximum degree, the algorithm provides oracle access to a subgraph of, such that: 1. is connected 2. is a tree with probability 3. is determined by and internal randomness

6 Local Algorithm for Minimum Spanning Graph (Spanning Tree)? YES on every edge

7 Local Algorithm for Minimum Spanning Graph (Spanning Tree)? YES on every edge Requires samples

8 A relaxation. Allow a few extra edges

9 Local Algorithm for Sparse Spanning Graph Given parameters,, and query access to over vertices and maximum degree, the algorithm provides oracle access to a subgraph of, such that: 1. is connected 2. with probability 3. is determined by and internal randomness

10 The Algorithm We will present a sparse spanning graph algorithm for High Expansion Graphs Phase 1: Global algorithm Phase 2: Local algorithm

11 Expander Graphs Expanders are sparse, yet highly connected graphs. We define expansion as follows: For a subset, is the set of vertices from with a neighbor in. A graph is an vertex-expander if for all sets of size at most,

12 Sparse Spanning Graph Algorithm for High Expansion Graphs

13 Global Algorithm Given a parameter, we partition the graph (or some of it): Randomly sample vertices, denoted as Centers For each vertex we want to find its closest center (up to ) How? BFS of depth from each center, breaking ties by id

14 Global Algorithm The edges of the spanning graph are: Edges of the BFS-tree over closest vertices of a center Edges of vertices not close enough to any center (singletons) Single edges connecting adjacent components

15 Choosing the radius is the neighborhood of is the minimum distance needed for most vertices to see enough vertices in their neighborhood.

16 Correctness is sparse with probability is connected

17 Correctness: G' is Sparse Within centers: all BFS-trees of all centers contains edges. Between centers: edges added between pairs of centers are Singletons:, each has edges, in total edges.

18 Correctness: G' is Sparse Within centers: all BFS-trees of all centers contains edges. Between centers: edges added between pairs of centers are Singletons:, each has edges, in total edges. The spanning graph has edges.

19 Correctness: G' is Sparse For each for which, the probability that no vertex in is selected as a center is

20 Correctness: G' is Sparse For each for which, the probability that no vertex in is selected as a center is From union bound, every such vertex is close enough to some center with probability

21 Correctness: G' is Sparse For each for which, the probability that no vertex in is selected as a center is From union bound, every such vertex is close enough to some center with probability By choosing, only a small fraction will be singletons, with probability

22 Correctness: G' is Connected It's enough to show that each 2 centers in are connected. Why? Each vertex has a path to some center: Each vertex assigned to a center, has a path to a center Each vertex not assigned to any center, has all its edges in

23 Correctness: G' is Connected Each 2 centers in are connected: I. All vertices of the shortest path are assigned to those centers.

24 Correctness: G' is Connected Each 2 centers in are connected: I. All vertices of the shortest path are assigned to those centers. II. All vertices are either assigned to those centers, or singletons.

25 Correctness: G' is Connected Each 2 centers in are connected: I. All vertices of the shortest path are assigned to those centers. II. All vertices are either assigned to those centers, or singletons. III. Some vertex on the path is assigned to another center.

26 Local Algorithm Randomly sample centers. Given an edge, perform BFS to depth from both and

27 Local Algorithm Randomly sample centers. Given an edge, perform BFS to depth from both and I. If either or are singletons, return YES. II. If they share a center, run min-id BFS from the center. III. If the centers are different, check their min-id shortest path.

28 Local Algorithm Complexity Let be the maximum size of a neighborhood The query and time complexity is

29 Local Algorithm Complexity Let be the maximum size of a neighborhood The query and time complexity is Estimate with probability

30 Estimating the radius Randomly sample vertices For each from, set Assume, set By Chernoff, with probability

31 Local Algorithm Complexity Since is expander, we can bound using and We can now bound We obtain the complexity (where ) For high expansion graphs we get nearly

32 Conclusion Local algorithm for sparse spanning graph of expanders Other graph families? General graphs?