The billard theorem for multiple random Fourier series (Q515869)

Multiple random Fourier series are series of stochastic processes of the form \[ X(t) = \sum \limits _{n \in \mathbb{Z}^{d}} A_{n}\exp(i\Phi_{n})\exp(in.t), \] indexed by \(t \in \mathbb{T}^{d}\), and involving real valued random variables \((A_{n})_{n \in \mathbb{Z}^{d}}\) and \((\Phi_{n})_{n \in \mathbb{Z}^{d}}\). Understanding in which sense this infinite sum of random variables converges is an even more subtle question than its deterministic counterpart. For instance, two reasonable processes \(X,Y\) (say centred and weakly stationary) that are almost surely in \(L^{2}(\mathbb{T}^d)\) may differ in law even if their Fourier coefficients are equal. To enter this fascinating subject, one can start with the classic books [\textit{J. P. Kahane}, Some random series of functions. Cambridge: Cambridge University Press (1968; Zbl 0192.53801)] and [\textit{M. B. Marcus} and \textit{G. Pisier}, Random Fourier series with applications to harmonic analysis. Princeton: Princeton University Press (1981; Zbl 0474.43004)]. The paper under review deals with a special class of multiple random Fourier series by assuming that the \((A_{n})_{n \in \mathbb{Z}^{d}}\) are deterministic, non negative, even, and square summable, and that the \((\Phi_{n})_{n \in \mathbb{Z}^{d}}\) are uniformly distributed over \(\mathbb{R}\backslash 2\pi \mathbb{Z}\), such that \(\Phi_{-n}=-\Phi_{n}\) almost surely, and such that all \((\Phi_{n})_{n \in \mathcal{A}}\) are independent as long as \(\mathcal{A}\) does not intersect\(-\mathcal{A}\). The authors also point out that the assumptions on \((A_{n})_{n \in \mathbb{Z}^{d}}\) can a posteriori be weakened to allow for random amplitudes under some appropriate \(\ell^2\) summability and independence assumptions. In doing so, they allow, for instance, independent Gaussian variables as Fourier coefficients. Under these assumptions, the problem is still highly non trivial, but some pathological behaviour is ruled out (the Fourier series are determined by their coefficients for instance), and some major technical difficulties disappear. In particular, the authors show that if uniform convergence (or boundedness) holds almost surely for one (symmetrical) method of summation, then it also does for any other method of summation. This is a nice consequence of the Lévy-Itô-Nisio theorem, that ensures that a sum of independent symmetric random variables in a Banach space converge almost surely if and only if one of its subsequence does. The main theorem of the paper then shows that the following are equivalent: (1) Almost sure uniform boundedness for one/all method of summation. (2) Almost sure uniform convergence for one/all method of summation. (3) The series represents a continuous function almost surely. (4) For all methods of summation, the series converges everywhere almost surely. For \(d=1\), the theorem was proven in [\textit{P. Billard}, Stud. Math. 22, 309--329 (1963; Zbl 0214.43702)]. The authors also study properties of the processes obtained by summing such series, including their covariance structure, stationarity, and sub-Gaussian behaviour.