Weighted polynomial inequalities (Q1083631)

Weighted polynomial inequalities play an important role in estimating rates of weighted approximations, in investigations of orthogonal polynomials and many other areas. G. Freud has considered the analogues of the Markov-Bernstein inequality with weights \(W_{\lambda}(x)=\exp (- | x|^{\lambda})\) \(-\infty <x<\infty\), \(\lambda >0\). He proved the inequality (*) \(\| W_{\lambda}P'_{n}\|_ p\leq Kn^{1- 1/\lambda}\| W_{\lambda}P_ n\|_ p\) \(1\leq p\leq \infty\), \(\lambda\geq 2\) and \(P_ n\) is any polynomial of degree \(\leq n\). The constant K depends on p and \(\lambda\) but is independent of n and of the polynomial \(P_ n.\) Recently A. L. Levin and D. S. Lubinsky [J. Approximation Theory (to appear)] showed that (*) holds also for \(1<\lambda <2\). It was conjectured that in the singular case \(\lambda =1\) (*) holds with the power of n replaced by log n whereas for \(0<\lambda <1\), the constant is independent of n. The authors prove this conjecture making use of a result interesting for its own merit which claims that for \(0<\lambda <1\), there exists a sequence of polynomials \(P_ n\) such that \(P_ n(0)=1\) and \(| P_ n(x)| \leq K \exp (-| nx|^{\lambda}),\)-1\(\leq x\leq 1\) while there is no such sequence of polynomials for \(\lambda =1\). Other interesting applications to polynomial approximation in weighted \(L_ p\) are also presented.