The concentration of measure on the cube $ [0, 1]^n $ equipped with uniform probability measure $\mu_{\infty}$, states that for any $A \subset [0, 1]^n $ with $ \mu_{\infty}(A) \geq \frac{1}{2} $, we have: \begin{align} 1 - \mu_{\infty}(A_{\epsilon}) \leq e^{- \pi \epsilon^2 }, \epsilon > 0 \end{align}

where $A_{\epsilon} = \{ x \in [0, 1]^n; d(x, A) < \epsilon \}$, and $d = L_2$ is the Euclidean distance. Similarly, let $B^n_{2}$ denote the Euclidean ball with normalized uniform measure. Then there is a constant $C$, for any $A \subset B^n_{2} $, with $ \mu_2(A) > 0 $:

$$ 1 - \mu_2(A_{\epsilon}) \leq \frac{1}{\mu_2(A)} e^{-n C\epsilon^{2} } $$ where $A_{\epsilon} = \{ x \in B^n_2; d(x, A) < \epsilon \}$, and $d = L_2$ is again the $L^2$ norm.

The concentration rate on $ [0, 1]^n $ is slower by a factor of $n$ compared to $B^n_2$. I'm wondering if there is a family of convex bodies $\Gamma(t)$, $ t \in [0, 1] $ with the following properties:

  • $\Gamma(0) = [0, 1]^n $ and $\Gamma(1) = B^n_2 $,
  • $ \Gamma(t) $ are all equipped with normalized uniform measures, and Euclidean $L_2$ metric,
  • each $ \Gamma(t) $ has a concentration rate $\gamma_t(n, \epsilon)$ such that $\gamma_0(n, \epsilon) = \pi \epsilon^2 $ (for the cube) and $\gamma_1(n, \epsilon) = n C \epsilon^2 $ (for the $L^2$ ball),
  • As $t$ goes from $0$ to $1$, $\gamma_t(n)$ "concentrates faster". So we get a family of concentration inequalities, interpolating between the ones on $[0, 1]^n $ and $B^n_2$.

A natural example is the concentration of measures on $L^p$ balls, but the version I know is equiped with $d = L_p$ metric, instead of $L_2$. Roughly, I want to know if it is true that the more symmetric a convex body is, the faster its concentration rate.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.