Some inequalities for multivariate characteristic functions

From MaRDI portal





Some new lower and upper bounds are obtained for characteristic functions (cf's) of multivariate distributions that can be useful in limit theorems and stability problems. Many are multivariate extensions of results reported by the first author [J. Math. Sci., New York 84, No. 3, 1179-1189 (1997; Zbl 0915.60027)]. Notation: \({\mathbf X}\) is a random vector (rv) in \(R^m\), \(m\geq 1\), with distribution function (df) \(F({\mathbf x})\), density \(p({\mathbf x})\), cf \(f({\mathbf t})\) and covariance matrix \(\Sigma\). \(\text{R} f({\mathbf t})\) is the real part of \(f({\mathbf t})\). \(\langle{\mathbf x}, {\mathbf y}\rangle\) is the scalar product of vectors \({\mathbf x}\) and \({\mathbf y}\); \(\|{\mathbf x}\|=\sqrt{(\langle {\mathbf x},{\mathbf x}\rangle)}\). \(h(z)=2(1-\cos z)/z^2\) and \(k(z)=(2/z^2) \log(2/(2-z^2))\). Theorem 1. For \(u>0\) and any \(a>m/\sqrt \pi:\int_{\|{\mathbf x}\|\geq a/u}dF(x)\leq c(a)/u^m \int_{ \|{\mathbf t}\|\leq u}[1-Rf ({\mathbf t})] d{\mathbf t}\). Here \(c(a)= \Gamma(m/2 +1)/ \pi^{m/2}[1-m/a \sqrt\pi]\). Theorem 2. If \(\|{\mathbf X} \|\leq c< \infty\) and \(\sup p({\mathbf x})\leq a<\infty\), then \(|f({\mathbf t}) |\leq (\sin d)/d\) for \(\|{\mathbf t}\|\leq\pi/2c\) and \(|f({\mathbf t}) |\leq (\sin D)/D\) for \(\|{\mathbf t}\|>\pi/2c\). Here \(b=\pi^{(m-1)/2} c^{m-1}a/ \Gamma((m+1)/2)\), \(d=\|{\mathbf t}\|/2b\), \(D=\pi/4bc\). Theorem 3. If \(\|{\mathbf X}\|\leq c\) and \(a\in[0,\pi/4]\), then \(|f({\mathbf t}) |\leq 1-h(a){\mathbf t}\Sigma{\mathbf t}'/2\) for \(\|{\mathbf t}\|\leq a/c\). Theorem 4. \(|f({\mathbf t})|\geq 1-{\mathbf t}\Sigma {\mathbf t}'/2\) for all \({\mathbf t}\in R^m\). Theorem 5. \(|f({\mathbf t})|\geq\exp \{-k(a) {\mathbf t}\Sigma {\mathbf t}'/2\}\) for any \(a\in(0,\sqrt 2)\) and all \({\mathbf t}\in R^m\) such that \({\mathbf t}\Sigma {\mathbf t}'\leq a^2\). Theorem 6. Let \(g(z)\), \(z\geq 0\), be a non-negative, unbounded, increasing function. Let \(a\) and \(d\) be finite real numbers. If \(\sup p({\mathbf x})\leq a\), \(Eg(\|{\mathbf X}\|)=d\) and \(\delta\in (0,1]\), then \(|f({\mathbf t})\leq 1-(1-\delta)^2 \|{\mathbf t} \|^2/3 \pi^2b^2\) for \(\|{\mathbf t}\|\leq\pi/2c\) and \(|f({\mathbf t}) |\leq 1-(1- \delta)^3/ 12b^2c^2\) for \(\|{\mathbf t}\|> \pi/2c\). Here \(c= g^{-1}(d/ \delta)\) and \(b\) is as in Theorem 2. Theorem 7. If \(\sup p({\mathbf x})\leq a<\infty\), then \(|f({\mathbf t})|\leq 1-c\|{\mathbf t}\|^2\) for \(\sqrt{{\mathbf t}\Sigma{\mathbf t}'}\leq\pi/4\) and \(|f({\mathbf t})|\leq 1-c\pi^2 \|{\mathbf t}\|^2/ 16({\mathbf t}\Sigma {\mathbf t}')\) for \(\sqrt{{\mathbf t}\Sigma{\mathbf t}'}> \pi/4\). Here \(\sigma^2= E\|{\mathbf X}\|^2\) and \(c=9[ \Gamma((m+1)/2)]^2/2^{m+8} \pi^{m+1}a^2(\sigma^2)^{m-1}\).











This page was built for publication: Some inequalities for multivariate characteristic functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1585962)