Hanson-Wright inequality in Banach spaces (Q2028940)

The Hanson-Wright inequality states that for any sequence of independent mean zero \(\alpha \)-subgaussian random variables \({X_1},{X_2},\ldots,{X_n}\) and any symmetric matrix \(A = {({a_{ij}})_{i,j \le n}}\) one has \(P\left( {\left| {\sum\limits_{i,j = 1}^n {{a_{ij}}({X_i}{X_j} - E({X_i}{X_j})} } \right| \ge t} \right) \le 2\exp \left( { - \frac{1}{C}\min \left\{ {\frac{{{t^2}}}{{{\alpha ^4}{{\left\| A \right\|}_{HS}}}},\frac{t}{{{\alpha ^2}{{\left\| A \right\|}_{op}}}}} \right\}} \right)\). The main objective of the article is to provide estimates on vector-valued quadratic forms which can be applied more easily and are of optimal form. The extension to general subgaussian variables is then obtained in Banach spaces.