Sudakov minoration principle and supremum of some processes (Q1380455): Difference between revisions

The author gives a method to estimate from below the r.v. \(E[\sup_{t\in T}\sum t_iX_i]\) where \(\{X_i\}\) are independent symmetric random variables with logarithmically concave tails and \(T\) is an index set of sequences. The method is a generalization of Sudakov's minoration principle. It provides geometric conditions over the index set \(T\), equivalent to boundedness of the process \((\sum t_iX_i)_{t\in T}\). The paper begins with a technical lemma that is important to prove the main result. This lemma consists in a minoration inequality for the expectation \(E[\sup_{t\in S}\sum_{i=1}^n t_iY_i+w(t)]\) where the \(Y_i\) are r.v. such that \(P\{| Y_i| \geq x\}\) is exponentially decreasing, in some range of the \(x\) variable, \(w\) is a function that could be equal to zero and \(S\) is a subset of \(l^2\). Using this lemma the author gives in Theorem 1 a generalization of Sudakov's minoration principle. The method can be viewed, in the author's own words, as a continuation of previous works done by Talagrand.