Asymptotics of Tracy-Widom distributions and the total integral of a Painlevé II function (Q934625)

The famous Tracy-Widom eigenvalue distribution functions for various ensembles of random matrices involve exponentials of integrals of the Hastings-McLeod solution to the second Painlevé equation. These integrals are taken from the positive infinity to the point \(x\), therefore, due to the exponential decay of the integrand, their asymptotics as \(x\to+\infty\) is straightforward. As \(x\to-\infty\), the Hastings-McLeod solution behaves like \(\sqrt{-x/2}\), and hence the asymptotics of the involved integrals can be computed up to constant terms. The authors of the paper express the distribution functions in terms of the properly regularized integrals starting from \(-\infty\). Such representations allow them to find exact asymptotics of the distribution functions as \(x\to-\infty\) and the (regularized) total integrals of the Hastings-McLeod solution to PII over the real axis. The proof is based on the known relation of the distribution functions with the \(n\times n\) Toeplitz determinants \(D_n(t)\) and orthogonal polynomials on the unit circle. The authors express the determinant as a product of leading coefficients \(k_j\) of the orthonormal polynomials \(p_j(z)\), \(D_n(t)=\prod_{j=1}^nk_{j-1}^{-2}(t)\). In the limit \(t\to\infty\) and \(n=2t+xt^{1/3}\), the product converges to an integral from \(-\infty\) to \(x\). The technical part of the paper is devoted to a proof of this fact by evaluation of the asymptotics of various contributions to the above product based on the study of the matrix Riemann-Hilbert problem for the orthogonal polynomials using the conventional stationary phase method as well as the steepest descent approach by Deift and Zhou.