The \(K\)-level crossings of a random algebraic polynomial with dependent coefficients (Q654502)

This is a very interesting paper on the general area of polynomials with random coefficients. There are many known results concerning the mathematical expectation of the number of real zeros of a random algebraic polynomial \[ P_n(x, \omega) \equiv P_n(x)= \sum_{k=0}^{n}X_k(\omega) x^k, \] where \(\omega \in \Omega\) is an elementary event in the fixed probability space \((\Omega, {\mathcal A}, \Pr)\). The results so far are mainly for identically, independent as well as dependent, Gaussian distributed coefficients \(\{X_k\}_{k=0}^{n-1}\), but also for a constant \(K\), possibly allowed to be a function of \(n\) the expected number of real roots of the equation \(P_n(x)=K\) -- known as \(K\)-level crossings of \(P_n\). Earlier results on this subject were discussed in [\textit{A. T. Bharucha-Reid} and \textit{M. Sambandham}, Random polynomials. Orlando etc.: Academic Press, Inc. (Harcourt Brace Jovanovich, Publishers) (1986; Zbl 0615.60058); the reviewer, Topics in Random Polynomials. Harlow: Addison Wesley Longman (1998; Zbl 0949.60010)]. The authors in this paper consider a new class of dependency for the coefficients of a polynomial and obtain for \(n\) sufficiently large asymptotic results for the expected number of \(K\)-level crossings. Both cases of \(K\) being constant and \(K=o\left( \sqrt{n/\log \log n}\right)\) are given. The coefficients \(\{X_k\}_{k=0}^n\) are assumed to be a stationary sequence of normal random variables with mean zero and covariance function \(\Gamma(k)= \text{E}(X_0 X_k)\) with \(\Gamma(0)=1\). Similar to the independent case it is shown that the above expected number is the same as the expected number of zeros as long as \(K\) remains bounded. For large \(K\), with \(K=o\left( \sqrt{n/\log \log n}\right)\), there is a reduction to the number of real roots of \(P_n(x)=K\). However, this reduction occurs in the interval \((-1, 1)\). The expected number of real roots in \((-\infty, -1) \cup (1, \infty)\) remains the same. The covariance function expressed as \(\Gamma(k)=\int_{-\pi}^\pi \exp(-k\phi)f(\phi)d \phi\), where \(f(\phi)\) is the spectral density of the covariance function, is used. The paper should initiate a new direction in this research area by looking at other types of polynomials with the above dependency.