Random variables with completely independent subcollections (Q875094)

The author investigates ``the algebra and geometry of the independence conditions on discrete random variables''. This leads to the study of ``a special case of what are called `independence ideals' in the algebraic statistics''. More precisely, he considers the generic \(a_1\times\cdots\times a_n\) table \(A= (x_{i_1\cdots i_n})\) over an arbitrary field \(K\) and a collection \(\Delta\) of subsets of \(\{1,\dots, n\}\). For \(J= \{j_1,\dots, j_m\}\in\Delta\), \(A_J\) is an \(a_{j_1}\times\cdots\times a_{j_m}\) table, the entries of which are certain sums of the variables \(x_{i_1\cdots i_n}\). Let \(I(A_J)\) be the ideal in the polynomial ring \(K[x_{i_1\cdots i_n}\mid 1\leq i_\nu\leq a_\nu]\) generated by the generalized \(2\times 2\)-minors of \(A_J\). (A generalized \(2\times 2\)-minor of a \(b_1\times\cdots\times b_m\) table \(B= (y_{k_1\cdots k_m})\) is a minor of the form \[ \text{det}\left(\begin{matrix} y_{k_1\cdots k_m}\\ y_{k_1\cdots k_{t-1}, l_t, k_{t+1}\cdots k_m}\end{matrix} \begin{matrix} y_{l_1\cdots l_{t-1}, k_t, l_{t+ 1}\cdots l_m}\\ y_{l_1\cdots l_m}\end{matrix}\right), \] \(1\leq k_\mu\), \(l_\mu\leq b_\mu\).) Set \(I_\Delta= I_\Delta(A)= \sum_{J\in\Delta} I(A_J)\). \(A\) is called \(\Delta\)-independent, if \(A\) lies on the variety defined by \(I_\Delta\). The main part of the article is devoted to the primary decomposition of the determinantal ideal \(I_\Delta\). The author shows that there is exactly one minimal prime of \(I_\Delta\) which does not contain the sum of all the variables, and is therefore ``the only minimal prime that corresponds to probability distributions''. Furthermore he describes the other minimal primes of \(I_\Delta\). In case \(\Delta\) is a simplicial complex with at most three facets, he proves that \(I_\Delta\) is a radical ideal. An example is given which shows that, in general, \(I_\Delta\) fails to be radical.