Computing the bit-size of the solutions of polynomial equations

Solving systems of polynomial equations is a foundational problem in mathematics with applications across engineering and the sciences. Beyond the algorithmic task of finding the solutions, a deeper theoretical goal is to predict the properties of these solutions directly from the equations.Take a system of polynomial equations with integer coefficients. Its solutions are points with algebraic coordinates, and as such they can be encoded in a computer using finitely many bits. A key question arises: can one predict the bit-size (or height) of these solutions without actually computing them?For arbitrary systems, the answer is no. Even a simple system of two linear equations in two variables with roots of unity as coefficients has a unique solution whose height varies unpredictably, as illustrated below.Yet a pattern emerges from the chaos. Numerical experiments reveal that a random choice of such coefficients produces a solution whose height approaches a specific value. We could confirm this prediction showing moreover that this value is a very special number related to the Riemann zeta function. This is achieved using techniques from the Arakelov geometry of toric varieties, linking randomness in geometry with arithmetic structures.