The real zeros of a random algebraic polynomial with dependent coefficients (Q444713)

The average number of real zeros of a polynomial \(P_n(x)=\sum_{k=0}^{n} X_k x^k \) of degree \(n\), where the coefficients \(\{X_k\}_{k=0}^n\) are a sequence of independent identically distributed Gaussian random variables, is well-known. This average number, denoted by \(\operatorname{E}N_n(a, b)\) over the interval \((a, b)\), is asymptotic to \(\operatorname{E}N_n(-\infty, \infty) \sim (2/\pi)\log n \) for all sufficiently large \(n\). The problem is more complicated when the coefficients are assumed to be dependent. Different cases of dependency were studied in original work by \textit{M. Sambandham} [J. Indian Math. Soc., New Ser. 41, 83--97 (1977; Zbl 0437.60049); Pac. J. Math. 81, 207--215 (1979; Zbl 0411.60066)]. It is shown that, in the case where there is a reasonable size of dependency between \(X_j\) and \(X_k\), \(j,k= 0, 1, \dotsc, n\), there is a reduction by half of \(\operatorname{E}N_n(-\infty, \infty)\) to \((1/\pi)\log n\). However, if the dependency decreases significantly fast, there is no change in the value of this expected value. A summary of these results is given in [\textit{K. Farahmand}, Topics in random polynomials. Pitman Research Notes in Mathematics Series. 393. Harlow: Addison Wesley Longman. (1998; Zbl 0949.60010)]. The present paper assumes a similar assumption of dependency as Sambandham above for the coefficients, i.e., \(\operatorname{E}[X_0X_k]=\rho^k\), where \(\rho \in (0, 1/2)\), as well as some mild assumptions on the spectral density function; then the error term involved in the asymptotic formula for \(\operatorname{E}N_n(-\infty, \infty)\) is explicitly evaluated. The proof is based on a formula known as the Kac-Rice formula; however, the application of this formula is more complicated for this case and requires that the interval \((-\infty, \infty)\) be broken up into several subintervals.