[论文解读] The real zeros of a random polynomial with dependent coefficients

Jeffrey Matayoshi|arXiv (Cornell University)|Jun 10, 2009

Geometry and complex manifolds被引用 1

一句话总结

本文通过分析更广泛的协方差结构，扩展了Sambandham关于具有依赖系数的随机多项式的工作。利用谱密度的限制条件，证明在温和的依赖条件下，实根的期望数量仍保持在 2/π log n 的数量级——与独立情况一致，表明Kac的经典结果对某些形式的系数依赖具有鲁棒性。

ABSTRACT

Abstract. Mark Kac gave one of the first results analyzing random polynomial zeros. He considered the case of independent standard normal coefficients and was able to show that the expected number of real zeros for a degree n polynomial is on the order of 2 log n, as n → ∞. Several years later, Sam-π bandham considered two cases with some dependence assumed among the coefficients. The first case looked at coefficients with an exponentially decaying covariance function, while the second assumed a constant covariance. He showed that the expectation of the number of real zeros for an exponentially decaying covariance matches the independent case, while having a constant covariance reduces the expected number of zeros in half. In this paper we will apply techniques similar to Sambandham’s and extend his results to a wider class of covariance functions. Under certain restrictions on the spectral density, we will show that the order of the expected number of real zeros remains the same as in the independent case. 1. One of the earliest results on the expected number of real zeros of the random polynomial given by n∑ (1.1) Pn(x) = Xkx k k=0 came from Mark Kac [6]. Kac considered the case when the coefficients are assumed to be independent standard normal random variables and was able to show that the value of the expected number of zeros is on the order of 2 π log n, as n → ∞. More recently, Edelman and Kostlan [4] derived a similar result, but in doing so they gave a nice geometric argument and derived formulas that hold for a wider class of coefficients. A natural generalization of this problem is to assume some dependence among the coefficients. Let X0, X1,... be a stationary sequence of normal random variables, where the covariance function is given by Γ(k) = E[X0Xk], Γ(0) = 1. Under these assumptions, two important results came from Sambandham. The first assumes that Γ(k) = ρk, where ρ ∈ (0, 1 2) [8]. In this case, it was shown that the expected number of zeros is on the same order as when the coefficients are independent. The second result assumes the covariance function is constant; that is, Γ(k) = ρ for any k, where ρ ∈ (0, 1) [7, 9]. Here, it was shown that the order of the

研究动机与目标

• 将Sambandham关于具有依赖系数的随机多项式的结果，从指数衰减和恒定协方差的特定情况推广至更广泛的情形。
• 研究当系数依赖通过更一般的协方差函数建模时，实根期望数量的阶是否保持不变。
• 确定系数序列谱密度满足何种条件时，实根的渐近阶与独立情况一致。
• 将Kac关于独立标准正态系数的经典结果，推广至更广泛的平稳、正态分布系数序列，其依赖程度较弱。
• 确定实根期望数量是否对协方差函数结构的敏感性，超出此前研究的两种特殊情况。

提出的方法

• 采用与Sambandham相似的技术，分析从平稳正态随机变量序列中抽取系数的随机多项式的实根期望数量。
• 协方差函数定义为 Γ(k) = E[X₀Xₖ]，其中 Γ(0) = 1，分析聚焦于满足某些正则性条件的谱密度的序列。
• 该方法依赖于谱表示理论，通过其谱密度表征系数的依赖结构。
• 论文利用来自随机矩阵和随机过程理论的实根期望数量的积分表示，分析其渐近行为。
• 通过施加对谱密度的限制条件（例如有界性和光滑性），作者推导出实根期望数量的渐近界。
• 分析利用了实根期望数量可表示为涉及协方差结构的实轴上积分的事实，从而可与独立情况进行比较。

实验结果

研究问题

• RQ1 当系数表现出弱平稳依赖时，随机多项式的实根期望数量是否仍保持在 2/π log n 的数量级？
• RQ2 系数序列的谱密度如何影响实根期望数量的渐近行为？
• RQ3 能否将独立系数情况下的结果推广至超越指数衰减和恒定协方差的更广泛依赖系数结构？
• RQ4 在何种协方差函数或谱密度条件下，实根期望数量不会相对于独立情况减少？
• RQ5 当谱密度满足温和正则性条件时，依赖与独立系数情况下零点计数行为是否存在结构相似性？

主要发现

• 对于一大类满足温和正则性条件的谱密度的协方差函数，实根期望数量仍保持在 2/π log n 的数量级，与独立情况一致。
• 即使系数表现出弱依赖，只要谱密度在零点有界且连续，该结果依然成立。
• 当依赖结构不引起系数间强相关性时，实根期望数量的渐近阶得以保持。
• 分析证实，影响零点计数的关键因素并非依赖的存在本身，而是系数序列的谱行为。
• 本文确立了实根期望数量的数量级对某些形式的系数依赖具有鲁棒性，从而推广了Kac的经典结果。
• 研究结果推广了Sambandham的早期工作：独立情况下的阶不仅在指数衰减协方差下保持，也适用于具有合适谱特性的更广泛协方差函数类。