Damage in sampled spatial networks

Abstract

Many real network systems are studied only up to a finite number of nodes. These systems are often derived from a sampling scheme and become a subset of a much larger network. Here we study the variation of sampling properties of such networks. In particular, we tested the sampled networks robustness to random and targeted attack. The damage done by each type of attack is compared between the sampled and true network. The results show that connection dependent sampling such as connected node sampling and targeted node sampling preserves the robustness of the true network. A node dependent sampling such as random node sampling cannot capture the true network properties especially for small values of sampled to true network size ratio p. Sampling a network does not always capture the exact property of the original network. We are also able to characterize the extent of damage a network suffers resulting to different connectivity states: fully connected, fragmented and isolated nodes network.