CS614. Advanced Algorithms. L07 Quiz.

CS614. Advanced Algorithms.

L07 Solutions

Acknowledgement

The questions in this problem set are adapted from the textbook on Parameterized Algorithms by Marek Cygan, Fedor V. Fomin, Łukasz Kowalik, Daniel Lokshtanov, Dániel Marx, Marcin Pilipczuk, Michał Pilipczuk, and Saket Saurabh.

Problem 1. Cluster Vertex Deletion

In the Cluster Vertex Deletion problem, we want to know if a simple undirected graph G has a subset S of at most k vertices such that G \setminus S is a disjoint union of cliques.

Problem 1.1 A Branching Algorithm

Design a 3^k \cdot n^{\mathcal{O}(1)} algorithm for Cluster Vertex Deletion.

Problem 1.2 Decreasing the threshold

Design a 3-approximation algorithm for Cluster Vertex Deletion.

Solution

As discussed in class, the induced path on three vertices is a forbidden substructure for a cluster graph. We state and prove this fact here for completeness.

Claim. A graph G is a disjoint union of cliques if and only if it does not contain a path on three vertices as an induced subgraph.

Proof (sketch). Suppose G is a disjoint union of cliques, and for the sake of contradiction, suppose it has an induced path on vertices x,y,z with the edges being between x and y, and y and z. Note that since this is an induced path, there is no edge between x and z. Since every component of G is a clique, we know that x and z must be in different components. However, there is a path from x to z via y, which is a contradiction.

Suppose G does not contain a path on three vertices as an induced subgraph. Again, for the sake of contradiction, suppose G has a connected component that is not a clique. Let (u,v) be a non-edge in this component. Let P be a shortest path between u and v consisting of the vertices:

P := \{u, w_1, \ldots, w_t, \ldots v\}.

Notice that t \geqslant 1, otherwise (u,v) is an edge. Further, notice that u, w_1, w_2 forms an induced path of length three¹ (if (u,w_2) was an edge then we have a shorter path by omitting w_1, contradicting our assumption that P is a shortest path between u and v). This contradicts our assumption.

Based on this, we have the following algorithm:

CVD(G,k):
    If k <= 0 and G has an induced P3 - RETURN NO
    If k >= 0 and G is a cluster graph - RETURN YES

    Let a,b,c be vertices such that ab and bc are edges and ac is not an edge.

    Return (CVD(G-a,k-1) OR CVD(G-b,k-1) OR (G-c,k-1))

One can obtain a 3-approximation by enumerating a maximal collection of disjoint induced P_3’s and including all vertices from the collection in the solution. If the collection has size t, we know that any solution (and in particular, the optimal one) must contain at least t vertices and the output has at most 3t vertices. The algorithm is summarized below:

CVD-3-Approx(G):
    Init S = emptyset

    While there is an induced P3 = (x,y,z) in G:
        include (x,y,z) in S
        G = G - (x,y,z)

    return S

Problem 2. Don’t Satisfy Too Much!

In the MIN-2-SAT problem, we are given a 2-CNF formula \phi and an integer k, and the objective is to decide whether there exists an assignment for \phi that satisfies at most k clauses.

Show that MIN-2-SAT can be solved in time 2^k n^{\mathcal{O}(1)}.

Solution

If there is a variable x that occurs only positively in \phi, we claim that there exists an optimal assignment that sets it to 0. Indeed, let \tau be an assignment that sets x to 1. Let \tau_x be the assignment obtained from \tau by flipping the value of x from 1 to 0. Note that the clauses that do not contain the variable x are either satisfied or falsified in both \tau and \tau_x. For clauses that contain x, it is possible that they are satisfied by \tau but not by \tau_x, but not vice versa. Therefore, \tau_x falsifies at least as many clauses as \tau, and we are done.

Based on this, our algorithm proceeds as follows:

if there is a variable x that occurs only as a positive literal:
    set x to 0
if there is a variable x that occurs only as a negated literal:
    set x to 1

The argument for the negated occurrences is symmetric to the one we have for positive literals.

Once we perform this preprocessing, assuming that have clauses remaining, we have the following guarantee:

Every variable has at least one positive and one negated occurrence.

Now we can branch exhuastively on the settings of variables; with the promise that either setting of the variable reduces our budget by at least one. The overall algorithm is summarized in the following pseudocode:

MINSAT(phi,k):
    if there is a variable x that occurs only as a positive literal:
        set x to 0
    if there is a variable x that occurs only as a negated literal:
        set x to 1

    if phi is empty:
        return YES
    if phi is not empty and k <= 0:
        return NO

    Let x be any variable that occurs in phi.
    return MINSAT(phi|[x = TRUE],k-1) OR MINSAT(phi|[x = FALSE],k-1)

Footnotes

Note that it is possible that w_2 = v.↩︎