On Gaussian marginals of uniformly convex bodies (Q1014053): Difference between revisions

Numerous results have been obtained of the following nature: let \(X\) denote a uniformly distributed vector inside a centrally-symmetric convex body \(K\) of volume 1 in \({\mathbb R}^n\). Let \(X_{\theta}=\langle X, \theta\rangle\) denote its marginal in the direction of \(\theta\in{\mathbb S}^{n-1}\), where \({\mathbb S}^{n-1}\) denotes the Euclidean unit sphere. Show that under suitable conditions on \(K\), the distribution of \(X_{\theta}\) is approximately Gaussian for most directions \(\theta\in{\mathbb S}^{n-1}\). Of course, the meaning of ``approximately'' and ``most'' need to be carefully defined, and vary among the different results. Motivated by different results, it was conjectured in [[1]: \textit{M. Anttila, K. Ball} and \textit{I. Perissinaki}, Trans. Am. Math. Soc. 355, No. 12, 4723--4735 (2003; Zbl 1033.52003) and [2]: \textit{U. Brehm} and \textit{J. Voigt}, Beitr. Algebra Geom. 41, No. 2, 437--454 (2000; Zbl 0983.52004)] that all convex bodies in \({\mathbb R}^n\) have at least one marginal which is approximately Gaussian, with the deviation tending to 0 as the dimension \(n\) tends to \(\infty\). This conjecture, referred to as the ``central limit problem for convex bodies'' has been confirmed to hold for various classes of convex bodies. \(K\) is called isotropic if it has volume 1 and satisfies that \(\text{Var}(X_{\theta})= L_K^2\) for all \(\theta\in{\mathbb S}^{n-1}\) and some constant \(L_K>0\), which is called the isotropic constant of \(K\) (\(\text{Var}(Y)\) denotes the variance of the random variable \(Y\)). Let us denote the density function of \(X_{\theta}\) by \(g_{\theta, K}(s) = \text{Vol}(K\cap\{s\theta+\theta^{\perp}\})\), and \(\phi_{\rho}(s)\) denote the Gaussian density with expectation 0 and variance \(\rho^2\). Recently, the central limit problem for arbitrary convex bodies was given an affirmative answer [[3]: \textit{B. Klartag}, Invent. Math. 168, 91--131 (2007; Zbl 1144.60021) and [4]: \textit{B. Klartag}, J. Funct. Anal. 245, No. 1, 284--310 (2007; Zbl 1140.52004)] in the following sense: For every isotropic convex body in \({\mathbb R}^n\) we have \(\sigma\{\theta\in{\mathbb S}^{n-1}: d_{TV}(g_{\theta, K}(s), \phi_{L_K}(s)) \leq \delta_n\}\geq 1 - \mu_n\), where \(d_{TV}(f, g) = \int_{-\infty}^{\infty} |f(s) - g(s)|\,ds\) is the total-variation metric between the measures given by the densities \(f\), \(g\), \(\delta\) and \(\mu_n\) are two series decreasing to 0, and \(\sigma\) is Haar probability measure on \({\mathbb S}^{n-1}\). The author of this paper focuses on showing the existence of approximately Gaussian marginals in a strong sense for a rather wide class of symmetric convex bodies. Following [[5]: \textit{B. Klartag} and \textit{E. Milman}, Isr. J. Math. 164, 221--249 (2008; Zbl 1151.52007)] but contrary to other approaches, the author proves that it turns out to be more useful to put the body \(K\) in some non-isotropic position (or affine image), for which \(P(|{{|X|}\over{\sqrt{n}}} - \rho|\geq \varepsilon \rho) \leq \varepsilon\), where \(0<\varepsilon<1/2\) and \(\rho>0\). Let us denote the average density over all possible directions by \(g_{\text{avg}}(s) =\int_{{\mathbb S}^{n-1}}g_{\theta}(s) \,d\sigma(\theta)\). In fact, Klartag obtains some of his results in [3] and [4] using Kolmogorov metric (for even densities) \(d_{\text{Kol}}\) and translates them to \(d_{TV}\). Hence all results stated in this paper for \(d_{\text{Kol}}\) can be easily translated to the total-variation metric. Klartag's approach to the central limit problem for convex bodies, being completely general, cannot exploit any good properties which certain classes of convex bodies posses. Consequently, certain results for concrete classes which preceded Klartag's solution, still give better quantitative bounds. The author extends results to arbitrary uniformly convex bodies with power type \(p\), for \(2\leq p< 4\). The paper is organized as follows. Section 2 titled ``Gaussian marginals in arbitrary position'' is dedicated to prove a result which was already used in [5] and is also used in Section 3. In Section 3, titled ``concentration of volume in uniformly convex bodies with good type'', the author extends and strengthens the results from [5] to \(p\)-convex bodies with ``good'' type. Section 4, titled ``concentration of volume in \(p\)-convex bodies for \(p<4\)'' is devoted to discussion of the interesting results from [[6]: \textit{S. G. Bobkov} and \textit{M. Ledoux}, Geom. Funct. Anal. 10, No. 5, 1028--1052 (2000; Zbl 0969.26019)].