The Sine-beta process is a mathematical object that describes how eigenvalues of large random matrices are distributed locally, for any positive parameter beta. A central question is how certain statistical quantities called correlation functions behave when multiple observation points are brought close together and eventually merged. This merging limit is called "fusion," and understanding it precisely matters for connecting random matrix theory to other areas of mathematics and physics, including quantum mechanics and number theory.
Previous work established the leading-order behavior of these correlation functions during fusion, showing they scale like a quantity called the Vandermonde factor raised to a power determined by beta. The present paper computes the next correction beyond this leading term, which is the first nontrivial refinement. This correction takes the form of the leading term multiplied by a simple explicit function involving the separation between the merging points, valid when a certain parameter combining the number of merging points and beta exceeds one. The result is clean and explicit, which is somewhat surprising given the complexity of the underlying machinery.
The proof draws on several technical tools: a symmetry identity for finite random matrix ensembles, exact formulas for moments of a particular family of probability measures called Hua-Pickrell measures, and careful analytic control of a random entire function called the stochastic zeta function. A bonus result is an explicit formula for a certain average involving the Hua-Pickrell measure. The condition that beta times the number of merging points exceeds one turns out to be a genuine boundary for this approach, because a pole appears right at that threshold, hinting that the correction formula changes character there and a different mathematical treatment will be needed beyond it.