You are on page 1of 4

Problem 3

For this problem, we know that the independent set problem is np-complete, and we need to show
that the difficulty of the induced matching problem is also just as hard as the difficulty of the independent
set problem, which is np-complete.

We start by listing the requirements for an np-complete problem:

An np-complete problem is both NP verifiable and is np-hard. We’ll start by showing that the
problem is NP-hard. We use the following proof by contradiction, ensuring that our reduction will get us
a solution that is p-time:

Let us denote the Independent set problem as IS, and the induced matching problem as IM.
Assume that we have a black box that is able to solve IM. If we assume that IM is solvable in p-
time, then IS is also solvable in p-time, because we can use the subroutine of the black box for IM to get
the solution for IS. We can picture the relationship of IS and IM using the diagram below:

Input: G and k Output: solution


exists or it
Independent doesn’t
Set

Change G to G’, Change the


output
Induced
Pass G’ and k into from IM
Matching
format back IM Black blox
to IS format

We need to show a bidirectional relationship between solving IS and solving IM, that is:

If we know that an input for IS will find a solution, then that means that the transformed input for
IM will also find a solution. Likewise, if transforming the input and putting it into the IM will find a
solution, then putting the original input into IS will also find a solution.

We start with the first part of the proof:

Suppose that we have a graph G with a value k, and that we know that these inputs can be put into
the IS problem, and it will output a valid solution to the problem. Notice that the nodes in the IS set are
not adjacent to any other node in the set.
B

Example: A and D are the independent set, so k = 2.

We need to transform this graph to be used as input for the IM problem. We notice that we can add an
edge to all existing nodes, and attach a new node to the end of the newly added edges; this will not affect
the original graph nor the relationships existing between the original nodes. For each node X in the
original graph G, we let X’ be the new node attached to X in the new graph G’.

B
B ’

A
C

A
C

D
D

Example: Here, a valid induced matching would be (A, A’) and (D, D’)
This new graph G’ and the original k value is now passed into IM black box. The IM problem is
able to find a solution where there are k pairs in an induced matching. The way the IM black box finds a
solution is by simply matching each X node with its corresponding X’ node. Because there were k nodes
in the IS, there will be k pairs in the IM. We know that this is guaranteed because for each node in the IS,
there were no other nodes in the IS that were adjacent to it. We also know that due to the way we
transformed the graph G, all X’ nodes are not adjacent to any other node in the entire graph with the
exception of its corresponding X node. Because of these conditions, we know that these X, X’ pairs of
nodes will satisfy the constraints of an IM problem, that is, each pair in the IM will not be connected to
any other pair by any edge in the graph G’.

Thus, we’ve shown that if the IS problem finds a valid solution given G and k, the IM problem
will also find a valid solution given G’ and k.

Now we need to do the second part of the proof:

Suppose this time that we know that given a G’ graph (that has been modified the way we
specified above) and k value can be inputted into the IM problem to produce a valid solution. We need to
examine a couple of cases for the different types of matchings in G’:

We know that there cannot be an X’, Y’ pairing in G’ because of the constraints that we put on
the transformation of G. Thus, we don’t need to consider this case.

For any X, X’ pairing in G’, we know that in the original graph G, X will be in the IS.

For any X, Y pairing in G’, we know that they are both in the original graph G, so the IS problem
can just pick either one of the nodes to be in the IS.

Putting these all together, we construct the IS on the original graph G based on the IM pairs in G’.

Finding a valid solution for the IS problem, given G and k, is guaranteed because, upon reverting
G’ back to G, we notice that any nodes that are in different pairs in the IM are non-adjacent to each in G’,
and as a result, if those nodes are also in G, then they will continue to remain non-adjacent in G;
therefore, those nodes will also be in the IS of G.

Thus, we’ve shown that if the IM problem will find a valid solution given G’ and k, then the IS
problem will also find a valid solution given G and k.

We’ve proven both parts of our claim; hence, we’ve shown that if the IM problem can be solved
in p-time, then the IS problem can also be solved in p-time. However, we know that the IS problem
cannot be solved in p-time, so we’ve reached a contradiction. Thus, the IM problem can only be solved in
np-time, and it follows that the IM problem is np-hard.
To show that the IM problem is np, we need a verifier that shows that, given a set of nodes
produced from the IM problem, we can verify that the set is indeed an IM set in p time; that is, given any
random matching, it takes p-time to check that it satisfies the constraints of the IM problem.

If there are k pairs in the IM set of the graph G’, then for any pair in the IM set, the time to check
if any pair is adjacent to any other pair is of the order O(k) time. Doing this for all pairs gets us a runtime
of the order O(k^2). This is indeed polynomial time. Hence, we’ve shown that the problem is np.

Since the IM problem is both np and np-hard, the IM problem is also np-complete.

You might also like