Linear Programming

LP

Dual

ILP

Knapsack

Relation

Dual

Independent Set

ILP

Relaxation

Dual

Let for each edge .

Minimum Vertex Cover

ILP

Relaxation

Dual

Let for each edge .

Clique

Let be the complement graph, where

ILP

Relaxation

Dual

Let for each edge .

Hitting Set

Given universe and set family .

ILP

Relaxation

Dual

Let for each set .


Approximation

Minimum Vertex Cover

-approximation

Let be the number of edges, and let be the size of an optimal vertex cover.

At any point, suppose there are uncovered edges remaining. Since an optimal cover of size can cover all these edges, by averaging, some vertex in the optimal solution is incident to at least of the remaining edges. Therefore the maximum-degree vertex among the remaining graph is incident to at least uncovered edges as well.

If the algorithm picks such a highest-degree vertex, then after one step the number of remaining uncovered edges is at most After steps, the number of uncovered edges is at most To make this less than , it is enough to have that is, Hence the algorithm picks at most vertices, so it is an approximation.

Note: this proof is for the greedy algorithm that repeatedly picks the vertex covering the largest number of uncovered edges.

-approximation

Let be a maximal matching in the graph, and let be the set of all endpoints of edges in . Then

We first show that is a vertex cover. Suppose not. Then there exists an edge such that neither nor is in . That means neither nor is matched by any edge in . But then we could add to , contradicting the maximality of . Hence is a vertex cover.

Now compare with the optimum vertex cover . Since the edges in are pairwise disjoint, any vertex cover must cover each edge of using at least one distinct endpoint. Therefore every vertex cover, in particular an optimal one, must contain at least vertices. So

Combining the two inequalities,

Therefore, taking all endpoints of a maximal matching yields a for Minimum Vertex Cover.

Knapsack

-approximation

Sort items by non-increasing density . Let be the set obtained by greedily taking items in that order as long as they fit. Let be the first item that does not fit. Also consider the single feasible item of maximum value, and return the better of these two choices.

Let be the value of the fractional knapsack optimum. Since fractional knapsack follows the same density order, it takes all items in and then possibly a fraction of item . Therefore Hence Since every integral solution is also feasible for the fractional relaxation, Therefore the returned solution has value at least So this algorithm is a -approximation.

Hitting Set

-approximation

Assume every set has size at most , i.e.

Consider the LP relaxation: Let be an optimal LP solution. The rounding algorithm selects every element such that

Feasibility

Fix any set . Since and , it is impossible that every element in has because then a contradiction. So each set contains at least one selected element, and the rounded solution is a valid hitting set.

Approximation ratio

Let be the rounded set. For every , Thus Since the LP optimum is at most the integer optimum, Therefore So the rounding algorithm is an