Linear Codes are Optimal for Index-Coding Instances with Five or Fewer Receivers

Abstract: We study zero-error unicast index-coding instances, where each receiver must perfectly decode its requested message set, and the message sets requested by any two receivers do not overlap. We show that for all these instances with up to five receivers, linear index codes are optimal. Although this class contains 9847 non-isomorphic instances, by using our recent results and by properly categorizing the instances based on their graphical representations, we need to consider only 13 non-trivial instances to solve the entire class. This work complements the result by Arbabjolfaei et al. (ISIT 2013), who derived the capacity region of all unicast index-coding problems with up to five receivers in the diminishing-error setup. They employed random-coding arguments, which require infinitely-long messages. We consider the zero-error setup; our approach uses graph theory and combinatorics, and does not require long messages.