Best approximation of functions by log-polynomials (Q2067052)

In [J. Geom. Anal. 28, No. 2, 1182--1201 (2018; Zbl 1406.52011)] the authors extended the famous John ellipsoid theorem to the setting of log-concave functions. \textit{J. B. Lasserre} [Math. Program. 152, No. 1--2 (A), 559--591 (2015; Zbl 1337.90049)] generalized the Löwner-John ellipsoid theorem not only from convex bodies to arbitrary compact sets, but also from ellipsoids to level sets of homogeneous polynomial of an arbitrary even degree. Here the authors combine two approaches for further extensions. More precisely, they consider two problems. Problem 1: Given \(f:\mathbb{R}^n\to [0,+\infty)\) with \(\Vert f\Vert_{\infty}=f(0)=1,\) and \(d\in\mathbb{N}\) even, minimize \[ \int_{\mathbb{R}^n}te^{-g(x)^{\frac{1}{d}}}dx \] among all homogeneous polynomials \(g\) of degree \(d\) and \(t\geq 1\) such that \[ f(x)\leq te^{-g(x)^{\frac{1}{d}}}. \] Problem 2 differs from Problem 1 by replacing exponent \(\frac{1}{d}\) to 1. The authors prove the existence of solution in Problem 1 for log-concave integrable functions \(f.\) For Problem 2 they prove existence and uniqueness of solution under an additional supposition on \(f\) (boundedness of \[ \bigcup_{\lambda\in (0,1)}\log(1/\lambda)^{-1/d}\{x\in\mathbb{R}^n\, :\, f(x)\geq\lambda\}). \] Moreover, a related interior minimization point being the unique solution of Problem 2 can be characterized by some touching conditions, via the Karush-Kuhn-Tucker conditions. For these conditions to hold, no hypothesis on the log-concavity is needed.