Heat flow and repeated differentiation of polynomials with i.i.d. roots

Abstract

How do the zeros of a polynomial evolve under the holomorphic heat flow or repeated differentiation? In this work, we develop a unified probabilistic proof to three conjectures on the evolution of polynomial zeros under holomorphic heat flow and repeated differentiation. We study these two evolutions for random polynomials with i.i.d.~roots $z_1,\dots ,z_n\sim {\mu }_0$, where ${\mu }_0$ on $\mathbb{C}$ satisfies suitable $2+\delta$ logarithmic moment conditions. For the heat flow of small time $t>0$ and Lipschitz continuous Stieltjes transform of ${\mu }_0$, we identify the limit distribution ${\mu }_t$ as explicit push-forward of ${\mu }_0$ under a transport map. For rotationally invariant ${\mu }_0$, we characterize the limit after $\lfloor tn\rfloor$ differentiations through its radial quantiles. For instance, the heat flow evolves the circular law into the elliptic and semicircle law, while differentiation moves surviving mass towards the origin. We also show that $o(n)$ derivatives preserve the initial distribution ${\mu }_0$. In fact, all convergences hold almost surely. The common proof strategy crucially relies on recursion identities from leaving out a root, and concentration inequalities, which lead to self-consistent equations for Stieltjes transforms. This brings a method familiar from random matrix theory to polynomial evolutions. Moreover, it allows for generalizations beyond rotational symmetric distributions and quantitative stability estimates for $o(n)$ differentiations and vanishing heat-flow times.

arXiv comment28p, 1fig, comments always welcome! v2: density assumption removed, compact support weakened to a logarithmic moment condition, and minor changes