Refinement of an inequality of P. L. Chebyshev (Q1046840): Difference between revisions

Let \(| \cdot| _{[-1,1]}\) denote the uniform norm on \([-1,1]\) and let \(P_n\) be the set of polynomials of degree \(\leq n\). The paper contains several refinements of the classical Chebyshev inequality \[ | p| _{[-1,1]} \geq \frac{| a_n (p)| }{2^{n-1}}\; , \] where \(p \in P_n\) with \(p(x) = \sum_{k=0}^n a_k (p)\, x^k\). More precisely, with \[ b_k (p) := \sum\limits_{j=0}^k \,\frac{a_{n-2j} (p)}{2^{n-2j}}\, {n-2j \choose k-j} \] for \(0 \leq k \leq n/2\), the authors prove \[ | p| _{[-1,1]} \geq \begin{cases}\frac{| a_n (p)| }{2^n} + | b_k (p)| & \text{for \(0\leq k < n/2\)},\\ \\ \geq \frac{| a_n (p)| }{2^n} + \frac{1}{\sqrt{2}} | b_{n/2} (p)| & \text{if \(n\) is even},\\ \\ \geq \big| \frac{a_n (p)}{2^n} + \varepsilon b_k (p)\big| + \big| \frac{a_n (p)}{2^n} + \overline{\varepsilon} b_k (p)\big| & \text{for \(0 < k \leq n/2\) and \(| \varepsilon| < \frac{1}{2}\)}.\end{cases} \] Moreover, they show that the first inequality is strict if \(k > 0\) and \(p\not= 0\), the second is strict if \(p\not= 0\), and the third is strict if \(p\) is not a multiple of the \(n\)-th Chebyshev polynomial.