Large deviations, moderate deviations, and the KLS conjecture (2003.11442v1)

Abstract: Having its origin in theoretical computer science, the Kannan-Lov\'asz-Simonovits (KLS) conjecture is one of the major open problems in asymptotic convex geometry and high-dimensional probability theory today. In this work, we establish a new connection between this conjecture and the study of large and moderate deviations for isotropic log-concave random vectors, thereby providing a novel possibility to tackle the conjecture. We then study the moderate deviations for the Euclidean norm of random orthogonally projected random vectors in an $\ell_p^n$-ball. This leads to a number of interesting observations: (A) the $\ell_1^n$-ball is critical for the new approach; (B) for $p\geq 2$ the rate function in the moderate deviations principle undergoes a phase transition, depending on whether the scaling is below the square-root of the subspace dimensions or comparable; (C) for $1\leq p<2$ and comparable subspace dimensions, the rate function again displays a phase transition depending on its growth relative to $n^{p/2}$.