On the width of transitive sets: bounds on matrix coefficients of finite groups (Q2178451)

Let \(f_{\mathbb{R}}(d)\) be the smallest function such that for every finite subgroup \(G\) of the orthogonal group \(O(d)\), and every \(v \in S^{d-1}\) (the unit sphere in \(\mathbb{R}^d\)), there is a point \(w \in S^{d-1}\) so that \(Sup_{g \in G} |\langle gv,w\rangle| \leq f_{\mathbb{R}}(d)\). Analogously, the function \(f_{\mathbb{C}}(d)\) is defined if we look at the unit sphere \(S^{2d-1}\) in \(\mathbb{C}^d\) and run over the finite subgroups of the unitary group \(U(d)\). Note that such functions exist because the spheres are compact. The main result of the paper asserts \[f_{\mathbb{R}}(d), f_{\mathbb{R}}(d) << \frac{1}{\sqrt{\log(d)}} \qquad \text{for } d \geq 2.\] This answers a question posed by Yufei Zhao. There is a geometric interpretation of the above bound for \(f_{\mathbb{R}}(d)\). If \(X \subset S^{d-1}\) is transitive for a finite subgroup \(G\) of \(O(d)\), the width of \(X\) is defined to be the minimum of the lengths of the orthogonal projections from \(X\) to unit vectors \(w \in S^{d-1}\). In terms of this notion, the above bound for \(f_{\mathbb{R}}(d)\) is equivalent to the bound: \[\mathrm{Width}(X) << \frac{1}{\sqrt{\log(d)}} \qquad \text{for } d \geq 2.\] As the author points out, a phenomenon that manifests in higher dimensions is very useful here. It shows that, in some sense. when \(d \rightarrow \infty\), ``most'' of \(S^{d-1}\) is contained in the ``slab'' \[\bigg \{x \in \mathbb{R}^d: \bigg|x_1 \bigg| \stackrel{<}{{\approx}} \frac{1}{\sqrt{\log(d)}} \bigg \}.\] The above result also solves, in affirmative, the conjecture below in the case of finite groups. Diameters of sphere quotients \(S^n/G\) -- more precisely, good lower bounds -- were investigated by \textit{W. D. Dunbar} et al. [Differ. Geom. Appl. 27, No. 2, 307--319 (2009; Zbl 1171.53025)] where they made the following conjecture: If \(G\) acts irreducibly on an even dimensional sphere \(S^n\), then for each \(\varepsilon >0\), each such action with the orbit space \(S^n/G\) having large enough dimension has the property that the diameter of \(S^n/G\) is within \(\varepsilon\) distance from \(\pi/2\). The result on width in the present paper, also proves this conjecture when \(G\) is finite. The author also shows that the bounds he obtains for \(f_{\mathbb{R}}(d), f_{\mathbb{C}}(d)\) and \(\mathrm{Width}(X)\) are sharp up to a constant multiple. A simple argument shows that \(f_{\mathbb{R}}(d) < \sqrt{2} f_{\mathbb{C}}(d)\) and therefore, the author studies \(f_{\mathbb{C}}(d)\) that is easier to apply representation theoretic techniques to. To prove the bound for the latter function, the author uses induction on \(d\) and on subgroups (while using representation theory). As the author points out, choosing the vector \(w \in S^{2d-1}\) randomly for the normalized Haar measure does not always work. So, he considers a symmetric probability measure on the sphere and a function which is an upper bound for \(f_{\mathbb{C}}(d)\) and is more amenable to apply induction. The study reduces to two other functions \(f_{\mathrm{prim}}(d)\) and \(f_{\mathrm{perm}}(d)\) that are related, respectively, to primitive subgroups and to subgroups consisting of permutation matrices of determinant \(1\). As is seen in many of the author's papers, he combines diverse ideas very cleverly. In particular, among other things, he also uses a consequence of the classification of finite simple groups which gives a good lower bound for the smallest dimension of a faithful representation.