Imagine you have a vibrating drum or a resonating cavity. The solutions to the wave equation on such a shape, called eigenfunctions, form standing wave patterns with nodal sets, which are the curves or surfaces where the wave is exactly zero. A famous open problem posed by the geometer Shing-Tung Yau asks: how does the total size (length, area, or volume) of these zero sets grow as the frequency of vibration increases? Yau conjectured that this nodal volume grows proportionally to the frequency, but proving tight bounds in general has remained a deep challenge. Rather than tackling a single fixed eigenfunction, this paper takes a probabilistic approach: it studies random superpositions of many eigenfunctions near a given high frequency, known as Riemannian random waves, and asks how much the nodal volume fluctuates around its expected value.
The main technical achievement is a sharp bound on the variance, meaning the typical squared deviation, of the nodal volume of these random waves. The authors show that on manifolds with chaotic geometry, such as negatively curved spaces where no two points are connected by multiple geodesics of the same length, the variance is dramatically smaller than previously known bounds, improving on prior work by more than a squared factor. A striking consequence is that what physicists and probabilists call Berry's cancellation phenomenon holds in this setting: positive and negative contributions to the nodal volume fluctuations cancel so thoroughly that the overall variance shrinks faster than naive estimates would suggest. This is analogous to a well-known prediction by the physicist Michael Berry about random waves in chaotic billiards.
The results rest on three pillars developed within the paper. First, the authors use the Kac-Rice formula, a classical tool for counting zeros of random functions by integrating local densities. Second, they apply a new decomposition of random variables into orthogonal components called Wiener-Ito chaos, which separates contributions at different orders of complexity. Third, and perhaps most originally, they prove a very precise estimate on how the wave equation's eigenfunctions are distributed locally in space, sharpening a classical result in spectral geometry called the Weyl law. Together these ingredients form a general framework that should be useful far beyond this specific problem, potentially applying to many other geometric and probabilistic questions about high-frequency waves on curved spaces.