Non-standard constructions in convex geometry: geometric means of convex bodies with an appendix by Alexander Magazinov (Q2406328)

The paper under review is an engaging discussion about how to construct new convex bodies (always containing the origin in their interior), having as starting point the polarity map \(K\to K^{\circ}\). An easy example of this kind of construction is the so-called \textit{harmonic mean} of convex bodies \(K,T\), \[ H(K,T)=\left(\frac{K^{\circ}+T^{\circ}}{2}\right)^{\circ}, \] which is proved to satisfy the natural relation with the arithmetic mean \(H(K,T)\subseteq(K+T)/2\). A more involved construction is given by the \textit{continued fraction of convex bodies}: for a sequence of convex bodies \(\{K_m\}_{m=1}^{\infty}\), it is defined as \([K_1,K_2,\dots,K_m]=(K_1+[K_2,\dots,K_m])^{\circ}\), setting \([K_1]=K_1^{\circ}\). Then it is discussed when the limit \(\lim_{m\to\infty}[K_1,\dots,K_m]\) exists (in the Hausdorff sense), as well as other consequences and properties. The \textit{geometric mean} of convex bodies is the main construction considered in the paper: setting \(A_0=K\) and \(H_0=T\), let \[ A_{n+1}=\dfrac{A_n+H_n}{2}\quad\text{ and }\quad H_{n+1}=H(A_n,H_n); \] then it is proved that the limits \(\lim_{n\to\infty}A_n\) and \(\lim_{n\to\infty}H_n\) exist and are equal, and this limit is defined as the geometric mean of \(K\) and \(T\), denoted by \(G(K,T)\). Several properties of this new construction are studied, as well as its relation with the so-called logarithmic mean of convex bodies which was introduced by \textit{K. Böröczky jun.} et al. [Adv. Math. 231, No. 3--4, 1974--1997 (2012; Zbl 1258.52005)]. They also discuss volume inequalities and possible extensions to the \(L_p\)-setting (i.e., the Minkowski sum is replaced by the \(p\)-addition) and to the functional case. The last section of the paper is devoted to state several appealing open questions regarding the geometric mean of convex bodies. The paper is supplemented with an appendix by A. Magazinov who presents a counterexample to the scaling property of the geometric mean: he proves that there exist a square \(K\) and a hexagon \(T\) such that \(G(K,T)\neq G((1+\varepsilon)K,(1+\varepsilon)^{-1}T)\) for \(\varepsilon\) small enough. For the entire collection see [Zbl 1377.52002].