Standard deviation is a strongly Leibniz seminorm (Q2439667): Difference between revisions

Leibniz seminorms are important because they suggest a way to define and to study noncommutative metric spaces. The paper starts by recalling the argument how metric spaces in the usual sense can be recovered from the Lipschitz constants of continuous functions which provide (on a subalgebra) an example of a seminorm which is strongly Leibniz. The definition is as follows: A seminorm \(L\) on a unital normed algebra is said to be Leibniz if \(L(1)=0\) and \(L(AB) \leq L(A) \|B\| + \|A\| L(B)\) for all \(A,B\) and strongly Leibniz if for all invertible \(A\) further \(L(A^{-1}) \leq \|A^{-1}\|^2 L(A)\). The known sources of strongly Leibniz seminorms are normed first-order differential calculi, in particular, spectral triples. Looking at specific choices of the corresponding Dirac operator and considering quotients it turns out that standard deviation, in the probabilistic sense, can be thought of as a Leibniz seminorm. This is a new point of view even in the case of classical probability spaces and it also works for noncommutative probability spaces, with a suitable extension of the definition of standard deviation for non-selfadjoint elements. This is analyzed in detail, with an open question about the strong Leibniz property if the state is not tracial. In the following sections the problem and the solutions are generalized for matricial seminorms and for (noncommutative) conditional expectations which involves the use of Hilbert modules. The final section provides the first known examples of Leibniz seminorms without the strong Leibniz property, first derived from the previous theory then in a simplified form related to Arveson's distance formula. Open questions are formulated. The paper is well written with careful motivations and mostly self-contained, so its reading can also be recommended as an introduction to this topic.