Levels of concentration between exponential and Gaussian (Q1612806)

Poincaré and log-Sobolev inequalities have the tensorization property for Borel probability measure \(\mu\) on the Euclidean space, i.e., if they hold for \(\mu\), then they automatically hold for all the product measures \(\mu^k\), \(k\geq 1\). The Poincaré inequality implies a concentration inequality of exponential type, whereas the log-Sobolev inequality implies Gaussian concentration [see \textit{M. Ledoux}, in: Séminaire de probabilités XXXIII. Lect. Notes Math. 1709, 120-216 (1999; Zbl 0957.60016)]. \textit{S. G. Bobkov} recently studied these inequalities for log-concave probability measures [Ann. Probab. 27, No. 4, 1903-1921 (1999; Zbl 0964.60013)]. Using the Prekopa-Leindler inequality, he proved that any log-concave probability measure satisfies a Poincaré inequality. Under the additional condition \(\int \exp (\varepsilon|x|^2) d \mu (x) < \infty\) for a positive \(\varepsilon\), the measure \(\mu\) satisfies a log-Sobolev inequality. As a corollary, he showed that a log-concave probability measure \(\mu\) on \(\mathbb R^n\) such that \(\mu \{ x : |x|> t \} \leq c e^{-(t/d)^2}\) holds for fixed \( c >1\), \(d >0\) and all positive \(t\), satisfies an infinite-dimensional concentration inequality. The above hypothesis is some very weak form of Gaussian concentration for \(\mu\) itself. In this paper, the author extends the above result to other levels of concentration, between exponential and Gaussian. In fact, he proves that if for fixed \( c >1\), \(d > 0\), \(r \in [1,2]\) and all positive \(t\), a log-concave probability measure \(\mu\) on \(\mathbb R^n\) satisfies \(\mu (\{ x : |x|> d \cdot t\}) \leq c \exp (-t^r)\), then \(\mu\) satisfies a family of Sobolev inequalities. Consequently, independently for \(k\), the product measure \(\mu^k\) has a concentration property on the model of \(\exp(-|t|^r)\).