John's ellipsoid and the integral ratio of a log-concave function (Q1651364)

In this paper, the authors extend the famous John ellipsoid theorem to the setting of log-concave functions. They define the ellipsoidal functions as those ones which can be expressed as \({\mathcal E}^a(x)=a\chi_{{\mathcal E}}(x)\), where \(a>0\) and \(\chi_{{\mathcal E}}\) is the characteristic function of an ellipsoid. Then it is shown that, given an integrable log-concave function \(f:{\mathbb R}^n\longrightarrow{\mathbb R}\), there exists a unique ellipsoid \({\mathcal E}^{t_0\|f\|_{\infty}}\), \(t_0\in[e^{-n},1]\), such that \({\mathcal E}^{t_0\|f\|_{\infty}}\leq f\) and \[ \int_{{\mathbb R}^n}{\mathcal E}^{t_0\|f\|_{\infty}}(x)\,dx=\max_{{\mathcal E}^a\leq f}\int_{{\mathbb R}^n}{\mathcal E}^a(x)\,dx. \] Next, the integral ratio of \(f\) is defined as the natural extension of the volume ratio for a convex body, namely, \[ I_{\text{rat}}(f)=\left(\frac{\int_{{\mathbb R}^n}f(x)\,dx}{\int_{{\mathbb R}^n}{\mathcal E}^{t_0\|f\|_{\infty}}(x)\,dx}\right)^{1/n}, \] and it is proved that \[ I_{\text{rat}}(f)\leq\begin{cases} I_{\text{rat}}\left(e^{-\|\cdot\|_{B^n_{\infty}}}\right) & \; \text{ if \(f\) is even,}\\ I_{\text{rat}}\left(e^{-\|\cdot\|_{\Delta_n-c}}\right) & \; \text{ otherwise} \end{cases} \] for any \(c\in\Delta_n\), where \(\Delta_n\) represents the regular centered simplex. Both bounds are shown to be the best possible. The authors use these results in order to get a reverse functional affine isoperimetric inequality.