Scribd
Upload
80 views

Matching Algorithm

The Hungarian algorithm finds a maximum weighted matching in a bipartite graph by iteratively adjusting a weighted cover (u, v) until the equality subgraph Gu,v contains a perfect matching M. It initializes u and v, then forms Gu,v and finds a maximum matching M in each iteration. If M is perfect, it stops; otherwise, it updates u and v to reduce excesses and iterates again, guaranteeing termination and an optimal solution.

Uploaded by

raghuannamalai
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
80 views

Matching Algorithm

The Hungarian algorithm finds a maximum weighted matching in a bipartite graph by iteratively adjusting a weighted cover (u, v) until the equality subgraph Gu,v contains a perfect matching M. It initializes u and v, then forms Gu,v and finds a maximum matching M in each iteration. If M is perfect, it stops; otherwise, it updates u and v to reduce excesses and iterates again, guaranteeing termination and an optimal solution.

Uploaded by

raghuannamalai
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

How to nd a maximum weight matching in a bipartite graph?

In the maximum weighted matching problem a nonnegative weight wi,j is assigned to each edge xiyj of Kn,n and we seek a perfect matching M to maximize the total weight w(M ) = eM w(e). With these weights, a (weighted) cover is a choice of labels u1, . . . , un and v1, . . . , vn, such that ui + vj wi,j for all i, j. The cost c(u, v) of a cover (u, v) is ui + vj . The minimum weighted cover problem is that of nding a cover of minimum cost. Duality Lemma For a perfect matching M and a weighted cover (u, v) in a bipartite graph G, c(u, v) w(M ). Also, c(u, v) = w(M ) iff M consists of edges xiyj such that ui + vj = wi,j . In this case, M and (u, v) are both optimal.

The algorithm
The equality subgraph Gu,v for a weighted cover (u, v) is the spanning subgraph of Kn,n whose edges are the pairs xiyj such that ui + vj = wi,j . In the cover, the excess for i, j is ui + vj wi,j . Hungarian Algorithm Input. A matrix (wi,j ) of weights on the edges of Kn.n with partite sets X and Y . Idea. Iteratively adjusting a cover (u, v) until the equality subgraph Gu,v has a perfect matching. Initialization. Let ui = max{wi,j : j = 1, . . . , n} and vj = 0.

Iteration. Form Gu,v and nd a maximum matching M in it.


IF M is a perfect matching, THEN

The Assignment Problem An example


stop and report M as a maximum weight matching and (u, v) as a minimum cost cover
ELSE

let Q be a vertex cover of size |M | in Gu,v . R := X Q T := Y Q := min{ui + vj wi,j : xi X \ R, yj Y \ T } Update u and v: ui := ui if xi X \ R vj := vj + if yj T Iterate

1 6 1 3 4

2 7 3 6 1

3 8 4 2 3

4 7 4 8 5

5 2 5 7 4

Excess Matrix
0 0 0 0 0

Equality Subgraph

5 8 5 8 5

4 3 2 1 0

2 1 0 1 6 4 2 1 1 0 5 2 6 0 1

Theorem The Hungarian Algorithm nds a maximum weight matching and a minimum cost cover.

1 4 2 0 1 T T T

=1
3 4

4 7 4 7 4

0 0 1 1 1

The Duality Lemma states that if w(M ) = c(u, v) for some cover (u, v), then M is maximum weight.
R

3 2 2 1 0

1 0 0 1 6 R 3 1 1 1 0 4 1 6 0 1

0 3 2 0 1 T T T

We found a maximum weight matching (transversal). The fact that it is maximum is certied by the indicated cover, which has the same cost:
T T T

=1 3 7 3 6 3
1 0 1 2 2

3 1 1 1 0

2 0 0 2 7 3 0 0 1 0 4 0 5 0 1

3 7 3 6 3

1 0 1 2 2

1 2 3 4 5

6 7 8 7 2 1 3 4 4 5 3 6 2 8 7

4 1 3 5 4

0 2 1 0 1

w(M ) = 5 + 7 + 4 + 8 + 4 = 28 = = 1 + 0 + 1 + 2 + 2+ 3 + 7 + 3 + 6 + 3 = c(u, v)

DONE!!

Hungarian Algorithm Proof of correctness


Proof. If the algorithm ever terminates and Gu,v is the equality subgraph of a (u, v), which is indeed a cover, then M is a m.w.m. and (u, v) is a m.c.c. by Duality Lemma. Why is (u, v), created by the iteration, a cover? Let xiyj E(Kn,n). Check the four cases. xi R, yj Y \ T ui and vj do not change. xi R, xi X \ R, xi X \ R, yj T yj T yj Y \ T ui does not change vj increases. ui decreases by , vj increases by . ui + vj wi,j by denition of .

Matchings in general graphs


An odd component is a connected component with an odd number of vertices. Denote by o(G) the number of odd components of a graph G. Theorem. (Tutte, 1947) A graph G has a perfect matching iff o(G S) |S| for every subset S V (G). Proof. Easy. (Lovasz, 1975) Consider a counterexample G with the maximum number of edges. Claim. G + xy has a perfect matching for any xy E(G).

Why does the algorithm terminate? M is a matching in the new Gu,v as well. So either (i) max matching gets larger or (ii) # of vertices reached from U by M -alternating paths grows. (U is the set of unsaturated vertices of M in X.)
7

Proof of Tuttes Theorem Continued


Dene U := {v V (G) : dG (v) = n(G) 1} Case 1. G U consists of disjoint cliques. Proof: Straightforward to construct a perfect matching of G. Case 2. G U is not the disjoint union of cliques. Proof: Derive the existence of the following subgraph.
w y

Corollaries
Corollary. (Berge,1958) For a subset S V (G) let d(S) = o(G S) |S|. Then 2 (G) = min{n d(S) : S V (G)}. Proof. () Easy. () Apply Tuttes Theorem to G Kd. Corollary. (Petersen, 1891) Every 3-regular graph with no cut-edge has a perfect matching. Proof. Check Tuttes condition. Let S V (G). Double-count the number of edges between an S and the odd components of G S. Observe that between any odd component and S there are at least three edges.

E(G) / E(G)

Obtain contradiction by constructing a perfect matching M of G using perfect matchings M1 and M2 of G+xz and G + yw, respectively.
9

10

Factors
A factor of a graph is a spanning subgraph. A k-factor is a spanning k-regular subgraph.

Every regular bipartite graph has a 1-factor. Not every regular graph has a 1-factor. But... Theorem. (Petersen, 1891) Every 2k-regular graph has a 2-factor. Proof. Use Eulerian cycle of G to create an auxiliary k-regular bipartite graph H, such that a perfect matching in H corresponds to a 2-factor in G.

11

You might also like