Adding Isolated Vertices Makes some Online Algorithms Optimal

TL;DR

The paper reveals that adding isolated vertices to certain graphs makes natural greedy online algorithms optimal for problems like independent set, vertex cover, and dominating set, with implications for their performance measures.

Contribution

It demonstrates that isolated vertices can render some greedy online algorithms optimal, and establishes NP-hardness for computing related online parameters.

Findings

Greedy algorithms are online optimal on Freckle Graphs.

NP-hardness of computing online independence, vertex cover, and domination numbers.

Existence of graphs where greedy algorithms are less effective despite worst case optimality.

Abstract

An unexpected difference between online and offline algorithms is observed. The natural greedy algorithms are shown to be worst case online optimal for Online Independent Set and Online Vertex Cover on graphs with 'enough' isolated vertices, Freckle Graphs. For Online Dominating Set, the greedy algorithm is shown to be worst case online optimal on graphs with at least one isolated vertex. These algorithms are not online optimal in general. The online optimality results for these greedy algorithms imply optimality according to various worst case performance measures, such as the competitive ratio. It is also shown that, despite this worst case optimality, there are Freckle graphs where the greedy independent set algorithm is objectively less good than another algorithm. It is shown that it is NP-hard to determine any of the following for a given graph: the online independence number, the online vertex cover number, and the online domination number.