On the best constants in the Khintchine inequality for Steinhaus variables (Q476490)

From MaRDI portal
scientific article
Language Label Description Also known as
English
On the best constants in the Khintchine inequality for Steinhaus variables
scientific article

    Statements

    On the best constants in the Khintchine inequality for Steinhaus variables (English)
    0 references
    2 December 2014
    0 references
    The classical Khintchine inequality says that there exist constants \(A_p\) and \(B_p\), \(0 < p < \infty\), such that for all \(x = (x_j)_{j=1}^n \in \mathbb{R}^n\) \[ A_p \|x\|_2 \leq \left( \mathbb{E} \left| \sum_{j=1}^n x_j r_j \right|^p \right)^{1/p} \leq B_p \|x\|_2, \] where \((r_j)_{j=1}^n\), \(r_j: [0,1] \to \{+1,-1\} \subset \mathbb{R}\), are the Rademacher functions. For the best possible constants \(A_p\) and \(B_p\) see [\textit{U. Haagerup}, Stud. Math. 70, 231--283 (1981; Zbl 0501.46015)]. The Rademacher functions constitute an i.i.d.\ sequence of Bernoulli random variables. The Steinhaus variables \(S_j\) are the complex counterpart of the Rademacher functions. They are an i.i.d.\ sequence of random variables \(S_j : (\Omega,P) \to S^1 \subset \mathbb{C}\) on a probability space \((\Omega,P)\) which are uniformly distributed on the circle \(S^1\). For \(0 < p < \infty\) define \[ a_p := \min \left( \Gamma \left(\frac{p}{2}+1\right)^{1/p}, \sqrt{2} \left( \frac{\Gamma(\frac{p+1}{2})}{\Gamma(\frac{p}{2}+1)\sqrt{\pi}} \right)^{1/p},1 \right) \] and \[ b_p := \max \left( \Gamma\left(\frac{p}{2}+1\right)^{1/p},1 \right). \] Let \((S_j)\) be an i.i.d.\ sequence of Steinhaus random variables. Then for any \(n \in \mathbb{N}\) and any sequence \((x_j)_{j=1}^n \in \mathbb{C}^n\), \[ a_p \|x\|_2 \leq \left( \mathbb{E} \left| \sum_{j=1}^n x_j S_j \right|^p \right)^{1/p} \leq b_p \|x\|_2. \] The constants \(a_p\) and \(b_p\) are the best possible. This result was known for \(1 \leq p < \infty\) and the author proves the remaining cases \(0 < p < 1\) (verifying a conjecture of U.\,Haagerup). The proof relies on real analysis techniques only, in particular estimates involving the Bessel functions \(J_0\) and \(J_1\). The above result implies that, for \(x \in \mathbb{C}^n\) with \(\|x\|_2 = 1\), \[ \mathbb{E} \ln \left| \frac{S_1+S_2}{\sqrt{2}}\right| \leq \mathbb{E} \ln \left| \sum_{j=1}^n x_j S_j \right|, \] answering a question of Baernstein and Culverhouse (see Equation (2.2) in [\textit{A. Baernstein II} and \textit{R. C. Culverhouse}, Stud. Math. 152, No. 3, 231--248 (2002; Zbl 0997.60009)]).
    0 references
    0 references
    Khintchine inequality
    0 references
    Steinhaus random variable
    0 references
    best constants
    0 references
    0 references

    Identifiers