The probability simplex consists of vectors of nonnegative weights summing to one. It is the convex hull of the coordinate vectors. A linear function has maximum on this simplex; equality holds exactly when the weights are supported on maximizing indices.
New to topics? Read the docs here!