Talagrand's inductive method and isoperimetric inequalities involving random sets (Q927360)

Let \((\Omega,K,P)\) be a probability space, let \((\Omega_j,K_j)\), \(1\leq j\leq n\), be measurable spaces, and let \(Z_{ij}:\Omega\to\Omega_j\), \(0\leq i\leq N\), \(1\leq j\leq n\), be \((K,K_j)\)-measurable functions such that \(P(Z_{0j}=Z_{ij})>0\) for \(0\leq i\leq N\) and \(1\leq j\leq n\). Put \(\Omega^{(n)}=\Omega_1\times\dots\times\Omega_n\) and \(Z_i=(Z_{i1},\dots,Z_{in})\), \(0\leq i\leq N\). Assuming that the random vectors \((Z_{0j},\dots,Z_{Nj})\), \(1\leq j\leq n\), are independent in a certain sense, and the distance function \(d:\Omega^{(n)}\times\Omega^{(n)}\to[0,\infty[\) satisfies suitable conditions, the main result is the inequality \(P(Z_0\in\{Z_1,\dots,Z_n\})E[\exp d(Z_0\in\{Z_1,\dots,Z_n\})^2/4]\leq1\). The author applies this result to the specific cases in which \(d\) is the Hamming distance and the Talagrand distance, and to prove certain exponential inequalities.