That's what I got out of it too (well, substituting "natural centrality measures" for "means"), except it bothers me that for n dimensions, it looks like he use

That's what I got out of it too (well, substituting "natural centrality measures" for "means"), except it bothers me that for n dimensions, it looks like he uses L₂ for the vector norm. I'd much prefer argmin 𝐱( sum 𝐬∈Sⁿ sum j≤n |𝐬ⱼ-𝐱ⱼ|^i )