Some notes on concentration for \(\alpha\)-subexponential random variables (Q6150874)

The author establishes various concentration inequalities for quantities related to an underlying random variable \(X_1\) which satisfies \[ \mathbb{P}(|X_1-\mathbb{E}X_1|\geq t)\leq2\exp\left(-t^\alpha/C\right) \] for all \(t\), for some \(\alpha\in(0,2]\) and constant \(C>0\). That is, \(X_1\) has a bounded Orlicz norm of order \(\alpha\in(0,2]\). For such a random variable, let \(X\) denote a vector containing \(n\) independent copies of \(X_1\). The author proves a generalised Hanson-Wright inequality giving a concentration bound for \(X^TAX\) for a symmetric matrix \(A\), and a concentration inequality for \(f(X)\) for \(f:\mathbb{R}^n\to\mathbb{R}\) convex and Lipschitz. A uniform Hanson-Wright inequality and convex concentration for random tensors are also established. For the entire collection see [Zbl 07730209].