A variance-sensitive Gaussian concentration inequality (Q6583775)

Let \(Z = (Z_1, Z_2,\ldots, Z_n)\) be an \(n\)-dimensional standard Gaussian vector. It is well-known that if \(f:\, \mathbb{R}^n \to\mathbb{R}\) is a Lipschitz function with Lipschitz constant \(L\), then we have the following concentration inequalities, for all \(t >0\), \(P(f(Z) - Ef(Z) \le - t) \le \exp\left(-\frac{t^2}{2L^2}\right)\) and \(P(f(Z) - M \le - t) \le 1/2\, \exp\left(-\frac{t^2}{2L^2}\right)\), where \(M\) is a median for \(f(Z)\). A recent paper [\textit{G. Paouris} and \textit{P. Valettas}, Ann. Probab. 46, No. 3, 1441--1454 (2018; Zbl 1429.60022)] pointed out that the Gaussian concentration phenomenon also holds for the convex functions. The improvement lies in the fact that \(\mathrm{Var}(f(Z)) \le L^2\). For further results, see [\textit{P. Valettas}, J. Anal. Math. 139, No. 1, 341--367 (2019; Zbl 1444.60020)]. The main result of the present paper is to obtain a multidimensional version of the variance-sensitive concentration inequality from [\textit{G. Paouris} and \textit{P. Valettas}, Ann. Probab. 46, No. 3, 1441--1454 (2018; Zbl 1429.60022)]. This main result is proved in Section 2, which is the last paragraph of this paper.