On differentiability and analyticity of positive definite functions (Q616718)

From MaRDI portal
scientific article
Language Label Description Also known as
English
On differentiability and analyticity of positive definite functions
scientific article

    Statements

    On differentiability and analyticity of positive definite functions (English)
    0 references
    0 references
    0 references
    12 January 2011
    0 references
    The paper deals with positive definite functions \(f:\mathbb{R} \to \mathbb{C}\), that is such that the inequality \(\sum_{j,k=1}^n f(x_j-x_k) \xi_j \overline{\xi_k} \geq 0\) holds for every choice of \(x_1,x_2,\ldots,x_n \in \mathbb{R}\) and \(\xi_1,\xi_2,\ldots,\xi_n \in \mathbb{C}.\) All results of the paper demonstrate the following effects: the global behavior of a smooth positive definite function is determined solely by the sequence of even-order derivatives at the origin. More precisely, the authors prove the following theorems. Theorem 1: Let \(f:\mathbb{R} \to \mathbb{C}\) be a positive definite function and suppose \(f\) is of class \(C^{2n}\) in a neighborhood of the origin for some positive integer \(n.\) If \(f^{(2m)}(0)=0\) for some positive integer \(m \leq n,\) then \(f\) is constant on \(\mathbb{R}\). Theorem 2: Let \(f:\mathbb{R} \to \mathbb{C}\) be a positive definite function and suppose \(f\) is of class \(C^{\infty}\) in a neighborhood of the origin. Then, if there exist positive constants \(M\) and \(D\) such that \[ 0\leq (-1)^n f^{(2n)}(0) \leq D\frac{(2n)!}{M^{2n}} \] for every non-negative integer \(n,\) we have: (i) \(f\) is analytic in \(\mathbb{R};\) (ii) Let \[ 1/h=\limsup \left({\frac{|f^{(2n)}(0)|}{(2n)!}}\right)^{\frac{1}{2n}}. \] There exist \(\alpha, \beta \in [h, +\infty]\) such that \(f\) holomorphically extends to the complex strip \(\{z\in \mathbb{C}: -\alpha < \mathrm{Im}(z) <\beta\},\) where \(\alpha\) and \(\beta\) are maximal with this property. Moreover, if \(h < \infty,\) the function \(f\) cannot be holomorphically extended to both the points \(z=ih\) and \(z=-ih\) simultaneously, implying in particular that \(h=\min\{\alpha,\beta\}\). The proof of these theorems are based on the fact that if a positive definite function \(f\) belongs to the class \(C^{2n}\) in a neighborhood of the origin for some positive integer \(n,\) then \(f \in C^{2n}(\mathbb{R})\) and for all integers \(0\leq m_1\leq n\), \(0\leq m_2\leq n\) and every \(x \in \mathbb{R}\) the inequality \[ |f^{(m_1+m_2)}(x)|^2 \leq (-1)^{(m_1+m_2)}f^{(2m_1)}(0)f^{(2m_2)}(0) \] is valid.
    0 references
    positive definite function
    0 references
    smooth positive definite function
    0 references
    positive definite kernel
    0 references
    positive definite matrix
    0 references
    positive semidefinite matrix
    0 references
    real-analytic function
    0 references

    Identifiers