Assume we have a probability measure $\mu$ on $\mathbb R^n$. Assume it satisfies $$ \mu(||x|| > u) \le Ce^{-au} \ \ \forall u > 0 $$ In other words, its tail is dominated by an exponential function. Then, how to prove Poincare inequality? $$ Var_{\mu}f \le K\int||\nabla f||^2d\mu $$

  • $\begingroup$ You can't, without additional assumptions. Suppose $\mu$ has discrete support, say for example it's a Poisson distribution on $\mathbb{R}$, and $f$ is constant on a neighborhood of each point in the support of $\mu$. $\endgroup$ – Mark Meckes Mar 8 '13 at 1:42

As Mark Meckes already pointed out, this is not enough to proof a Poncaré inequality.

If you are interested in the case where $\mu$ is absolutely continuous, then there are conditions formulated in terms of the log-density of $\mu$, i.e. $\mu$ is of the form $$ \mu(\mathrm{d}x) = \exp(-H(x)) $$ The minus sign is just convention.

There is a complete characterisation of the constant $K$ for the real line via the Muckenhoupt functional. Therefore let $m$ be the median of $\mu$, i.e. $\mu((-\infty,m))=\mu((m,\infty))=1/2$, then define $$ \begin{align} B_+ &= \sup_{x\geq m} \left( \int_{m}^x \exp(H(y))\;\mathrm{d}y\ \int_{x}^\infty \exp(-H(y)) \; \mathrm{d} y \right) \\\\ B_- &= \sup_{x\leq m} \left( \int_{x}^m \exp(H(y))\;\mathrm{d}y \ \int_{-\infty}^x \exp(-H(y)) \; \mathrm{d}{y} \right) \end{align}$$ Then, one has $$ \frac{1}{2} \max\{B_-,B_+\} \leq K \leq 4 \max\{ B_, B_+\} $$ I.e. the Poncaré constant can be recovered upto a factor of $8$.

On $\mathbb{R}^n$ there is the theory of Lyapunov functions, which give sufficient conditions for Poincaré inequalities. See A simple proof of the Poincaré inequality by Bakry et. al. There, two prominent assumptions on $H$ which both individually imply a Poincaré inequality are

  1. For some $\alpha>0$ and all $|x|\geq R>0$ holds $$ \langle x, \nabla H\rangle \geq \alpha |x| $$

  2. For some $\alpha>0 , c>0$ and all $|x|\geq R>0$ holds $$ \alpha |\nabla H(x)|^2 - \Delta H(x) \geq 0 $$

Then $\mu$ satisfies a Poincaré inequality with an upper bound on the constant $K$.

| cite | improve this answer | |

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.