The binomial ideal of the intersection axiom for conditional probabilities (Q633280): Difference between revisions

This paper concerns algebraic statistics, the intersection axiom of conditional probability is translated into an algebraic geometric setting. Consider three random discrete finite variables \(X_i, i=1,\dots,3\) taking values in the sets \( [r_i]:=\{ 1,\dots,r_i\}\) and let \(p_{ijk}\) be the unknown probability: \(P(X_1=i,X_2=j,X_3=k)\), Let define the following ideal \(I_{\mathcal M}\subset K[p_{ijk}]\); \[ I_{\mathcal M}=(p_{ijk}p_{i'j'k}-p_{ij'k}p_{i'jk}: i,i'\in [r_1], j,j'\in [r_2],k\in [r_3])+ \] \[ (p_{ijk}p_{i'jk'}-p_{ijk'}p_{i'jk}: i,i'\in [r_1], j\in [r_2],k,k'\in [r_3]). \] The main result of this paper is to give the primary decomposition of the ideal \(I_{\mathcal M}\) in terms of some graphs, answering a question by Cartwright and Engstrom. The reader interested in this paper should read the paper [\textit{J. Herzog}, et al. Binomial edge Ideals and conditional independence statement. \url{arXiv:0909.4717}].