Tight Bell inequality for d-outcome measurements correlations

In this paper we prove that the inequality introduced by Collins, Gisin, Linden, Massar and Popescu is tight, or in other words, it is a facet of the convex polytope generated by all local-realistic joint probabilities of d outcomes. This means that this inequality is optimal. We also show that, for correlation functions generalized to deal with three-outcome measurements, the satisfyability of this inequality is a necessary and sufficient condition for the existence of a local-realistic model accounting for them.