Modified Shephard's problem on projections of convex bodies (Q1001428)

Motivated by the successful applications of Fourier analysis to questions on central sections of symmetric convex bodies and by the often observed analogy between sections and projections (around the Busemann--Petty problem and the Shephard problem), this paper exhibits some instances where this analogy breaks down. Let \(K,L\) be \(0\)-symmetric convex bodies in \({\mathbb R}^n\) (\(n\geq 3)\), which are infinitely smooth and have positive Gauss curvature. Suppose that \[ (-\Delta)^{\alpha/2}\text{vol}_{n-1}(K|\theta^\perp) \geq (-\Delta)^{\alpha/2} \text{vol}_{n-1}(L|\theta^\perp) \qquad \text{for all }\theta\in S^{n-1},\tag{1} \] where \(\text{vol}_{n-1}(K|(\cdot)^\perp)\) is extended to \({\mathbb R}^n\) as a homogeneous function of degree \(1\). Whereas it is proved that this implies \(\text{vol}_n(K) \leq \text{vol}_n(L)\) if \(\alpha\in[n,n+1)\), examples \(K,L\) are found satisfying (1) with some \(\alpha \in[n-2,n)\) but \(\text{vol}_n(K) > \text{vol}_n(L)\) (and other examples \(K,L\) with \(\text{vol}_n(K) < \text{vol}_n(L)\)). Moreover, it is shown that the Fourier transform of \(\|x\|_K|x|_2^{-n+2}\) (where \(|\cdot|_2\) is the Euclidean norm) is not necessarily a positive definite distribution. This disproves a conjecture of A. Koldobsky.