Information of relative pairwise comparisons (Q795013)

The amount of information obtained from a set of pairwise comparisons is studied by means of two kinds of entropy introduced for a graph G defined on a finite set \(\Sigma\). The combinatorial entropy is introduced as the Shannon's entropy of a family of random variables deduced from independent normally distributed random variables \(X_{\sigma}\) linked with the elements of \(\Sigma\). The \(\lambda\)-Gaussian entropy of G is the relative information between \(X_{\Sigma}=\{X_{\sigma},\sigma \in \Sigma \}\) and a family of random variables \(Z_ G\) obtained by considering a family of independent random variables with same distribution N(0,1/\(\lambda)\). The paper gives partial solutions to the following problems: 1.) In a given class of graphs, find a graph which maximises the combinatorial or Gaussian entropy. 2.) For a graph G and a vertex \(\sigma\), find the minimum sufficient subgraph of G at \(\sigma\) in the combinatorial or Gaussian sense. 3.) Find relations between the combinatorial and the Gaussian entropies.