\input{template}
\input{macros}

\usepackage{color, graphicx}
\usepackage{amssymb, amsmath}
\usepackage{epsfig}


\begin{document}
\lecture{19}{Nonbipartite Matching}{Neha Singh}
%% -----------------------------------------------------------------------------
%%                          Beginning of Section 1.
%% -----------------------------------------------------------------------------

\section{Recap}
In last class we had seen how to find the matching of maximum size
in a bipartite graph. This is done by starting with any single
edge as a matching and then iteratively finding augmenting paths
and increasing the size of the matching by one at each step until
there is no augmenting path in the graph w.r.t that matching. We
had proved the following theorems last time:
\begin{Thm}
 The matching M of the graph G is maximum iff there is no augmenting path in G w.r.t M.
\end{Thm}
\begin{Thm}
 For bipartite graphs, if there exists an augmenting path in the graph G then the we can find it by using the modified bfs algorithm.
\end{Thm}

Now let us proceed to finding maximum matching in a non-bipartite graph.

\section{Nonbipartite Matching}
Till now we have seen how to find the maximum matching for a
bipartite graph. The same algorithm for finding an augmenting path
in a bipartite graph will be extended for non-bipartite graphs.
However modifications need to be made to the algorithm.\newline As
before, we start with an unmatched even vertex, find all its
neighboring odd vertices. If any of these odd vertices were
unmatched then we have found an augmenting path else we continue
from these odd vertices by following the matched edges incident on
them. This yields a new set of even vertices and so on. In case of
the a bipartite graph edges can be only between even and odd
vertices. Thus while following the path, we keep alternating
between even and odd vertices. However in the case of a
non-bipartite graph, we can have odd-odd and even-even edges also
which will give rise to odd circuits in the graph. We consider the
problems arising due to even-even edges only, as in the modified
bfs algorithm, after reaching an odd vertex, we follow the matched
edge and there can be only one matched edge incident on a vertex.
\newline

\begin{figure}
 \centering
 \includegraphics[width=.25\textwidth]{blossom2}
 \caption{Blossom}
 \label{blossom}
\end{figure}

For example consider the figure below. Here there is an even-even
edge between $v_1$ and $v_2$. We observe that $v_1$ and $v_2$ have
a common ancestor which has to be necessarily even. This is
because in our modified bfs algorithm, we branch off at only the
even vertices. At odd vertices, we only follow the matched edge.
If we come across an even-even edge between two vertices while
searching for augmenting path, the graph must have an odd cycle
where the even-even edge is a part of the odd cycle. In the
modified algorithm for finding augmenting paths in non-bipartite
graphs, whenever we come across such an odd cycle, we shrink it to
a single vertex (say $v_o$) such that all the vertices adjacent to
any of the vertex in the cycle are now adjacent to $v_o$. Such an
odd length cycle having a matched and unmatched edges incident on
all but one vertex (the base where both incident edges are
unmatched) is called a blossom.\newline

\begin{Def}
 A blossom is a cycle of length 2k+1 with k edges matched . The base of the blossom is the one and the only vertex in the cycle with both incident edges unmatched.
\end{Def}

Naturally for shrinking to represent a valid operation while searching for augmenting path, we must show that by shrinking a blossom does not add or omit augumenting paths in the graph. The exact method and the proof of correctness will be discussed later. First we prove the following lemma.

\begin{Lem}
 Let M be a matching. Let B be a blossom such that the base is unmatched in G. Let $G^`$ be the graph obtained by shrinking B. Let $M^`$ be the edges of M outside B. There is an augumenting path in G w.r.t M iff there is an augumenting path in $G^`$ w.r.t $M^`$.
\end{Lem}

\begin{proof}
\emph{Part 1:} ($\implies$)  Given an augmenting path in G w.r.t M
to show that there exist an augmenting path in $G^`$ w.r.t
$M^`$.\newline

\begin{enumerate}
\item Case (i): The augmenting path in G does not intersect the
blossom. So the same augmenting path exists in $G^`$ w.r.t $M^`$.
\item Case (ii): The augmenting path intersects the cycle in G.
\newline \emph{Claim:} We can find an augumenting path in $G^`$ if
we stop as soon as we hit the shrunk vertex(say $v_o$) while
travelling along the augmenting path of G. \newline \emph{Proof:}
This is because of the fact that $v_o$ in unmatched in $G^`$. Note
that all the vertices in the blossom except the base already have
an incident matching edge inside the blossom and the base is
assumed to be unmatched.
\newline
\end{enumerate}




\emph{Part 2:}($\Leftarrow$)  Given an augmenting path in $G^`$
w.r.t $M^`$ to show that there exist an augmenting path in G w.r.t
M.

\begin{enumerate}
\item Case (i): The augmenting path in $G^`$ does not contain
$v_o$. Then the same path is an augmenting path in G also. \item
Case (ii): The augmenting path contain $v_o$. Since $v_o$ is an
unmatched vertex in $G^`$, the augmenting path can either start of
end at $v_o$.

On expanding the blossom, this path will intersect the blossom in
the graph G. Thus the path will contain some edge incident on some
vertex of the blossom. The vertex through which the path enters
the blossom can be either the base or any other vertex of the
blossom.

\begin{enumerate}
\item If the path contains the edge incident on the base of the blossom then we have found an augmenting path in G as the base in unmatched.
\item If the path contains edge incident on any other vertex of the cycle, i.e., other than the base, then extend this path by moving along the cycle to reach the unmatched base and we will have an augmenting path  in G.
\end{enumerate}

\end{enumerate}

\end{proof}

We note that in the figure above the base is unmatched only when
the common ancestor of $v_1$ and $v_2$ is the vertex from which we
started searching for an augmenting path. To use this lemma, we
need to do something more once we identify a blossom. This we will
see in the next lecture.

\end{document}