Orthogonal polynomials for exponential weights \(x^{2\rho} e^{-2Q(x)}\) on \([0,d)\). II. (Q2369530)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Orthogonal polynomials for exponential weights \(x^{2\rho} e^{-2Q(x)}\) on \([0,d)\). II.
scientific article

    Statements

    Orthogonal polynomials for exponential weights \(x^{2\rho} e^{-2Q(x)}\) on \([0,d)\). II. (English)
    0 references
    0 references
    0 references
    22 May 2006
    0 references
    This paper continues the study of so called exponential weights by the authors as initiated in their paper in [J. Approximation Theory 134, No. 2, 199--256 (2005; Zbl 1079.42017)]. Let \(I=[0,d),\;0<d\leq\infty\) and \(Q: I\rightarrow [0,\infty)\) a continuous function. If for \(W(x)=\exp{(-Q(x))}\) all moments \(\int_I\,x^nW(x)\,dx\;(n\geq 0)\) exist, \(W\) is called an exponential weight on \(I\). The orthonormal polynomial (OP) of degree \(n\) for \(W^2\) is denoted by \(p_n(W^2;x)\) or, if confusion does not arise, \(p_n(x)\). Along with these weights and OP's, one uses \(W_{\rho}(x):=x^{\rho}W(x) \;(\rho>-1/2)\) and \(p_n(W^2_{\rho},x)=p_{n,\rho}(x)\). To fix the OP's: \[ \int_I\,p_{n,\rho}(x)p_{m,\rho}(x)x^{2\rho}W^2_{\rho}(x)\,dx=\delta_{m,n},\;p_{n,\rho}(x)=\gamma_{n,\rho}x^n+\cdots,\;\gamma_{n,\rho}\equiv \gamma_n(W^2_{\rho})>0. \] Introduce a.o. 1. \(T(x):=xQ'(x)/Q(x),\;x\in (0,d)\) (assumed to be quasi-increasing in \((0,d)\) with \(T(x)\geq \Lambda>1/2)\) and the zeros of \(p_{n,\rho}(x)\) by \[ x_{n,n}<x_{n-1,n}<\cdots <x_{2,n}<x_{1,n}. \] 2. The Mhaskar-Rakhmanov-Saff constants \(a_t\) for \(t>0\) as the positive root of \[ t={1\over\pi}\,\int_0^1\,{a_t uQ'(a_tu)\over\sqrt{u(1-u)}}\,du. \] 3. The number \(\eta_t=\left(tT(a_t)\right)^{-2/3}\) and \[ \varphi_t(x):=\begin{cases} \sqrt{x+a_tt^{-2}}(a_{2t}-x)\over t\sqrt{a_t-x+a_t\eta_t} & x\in [0,a_t]\\ \varphi_t (a_t) & x>a_t \\ \varphi_t(0) & x<0\end{cases}. \] 4. The fundamental polynomials at the zeros of \(p_{n,\rho}\) by \(\ell_{jn}\in{\mathcal P}_{n-1}\) satisfying \(\ell_{jn}(x_{k,n})=\delta_{k,j}\). Then under some technical conditions on \(W\), indicated by \(W\in{\mathcal L}(C^2)\) or \(W\in{\mathcal L}(C_{+}^2)\), the main results are A. Let \(\rho>-1/2,\;0<\beta<1,\;W\in{\mathcal L}(C^2_{+})\), then uniformly for \(n\geq 1\) \[ \sup_{x\in I}\,| p_{n,\rho}(x)| W(x)\left(x+{a_n\over n^2}\right)^{\rho}\sim \left({n\over a_n}\right)^{1/2}, \] and \[ \sup_{x\in [a_{\beta n},d)}\,| p_{n,\rho}(x)| W(x)\left(x+{a_n\over n^2}\right)^{\rho}\sim a_n^{-1/2}\left(nT_n(a_n)\right)^{1/6}. \] If \(W\in{\mathcal L}(C^2)\), the estimates hold with \(\sim\) replaced by \(\leq C\). B. Let \(W\in{\mathcal L}(C_{+}^2),\;\rho>-1/2\), then there exists an \(n_0\) such that uniformly for \(n\geq n_0,\;1\leq j\leq n\): \[ \left| p'_{n,\rho}W_{\rho}\right| (x_{jn}\sim \varphi_n(x_{jn})^{-1}[x_{jn}(a_n-x_{jn}]^{-1/4}.\tag{a} \] \[ \left| p_{n-1,\rho}W_{\rho}\right| (x_{jn})\sim a_n^{-1}[x_{jn}(a_n-x_{jn}]^{1/4}.\tag{b} \] \[ \max_{x\in I}\,\left| \ell_{jn}(x)W(x)\left(x+{a_n\over n^2}\right)^{\rho}\right| W_{\rho}^{-1}(x_{jn})\sim 1.\tag{c} \] \[ \left| p_{n,\rho}W_{\rho}\right| (x)\sim \min\{| x-x_{jn}| ,| x-x_{j+1,n}| \}\varphi_n(x_{jn})^{-1} [x_{jn}(a_n-x_{jn})]^{-1/4}.\tag{d} \] If \(W\in{\mathcal L}(C^2)\), then (c) holds with \(\sim\) replaced by \(\leq C\) and (b) with \(\sim\) replaced by \(\geq C\). C. Let \(W\in{\mathcal L}(C_{+}^2),\;\rho>-1/2\). Uniformly for \(n\geq 1,\;1\leq j\leq n\): \[ x_{jn}-x_{j+1,n}\sim\varphi_n(x_{jn}). \] D (Bernstein inequality). Let \(W\in{\mathcal L}(C^2),\;0<p\leq\infty,\;\beta>-1/p\) if \(p<\infty\) and \(\beta\geq 0\) if \(p=\infty\). For \(n\geq 1, P\in{\mathcal P}_n\) and some \(C\not= C(n,P)\): \[ \| (PW)'(x)\varphi_n^{\#}(x)x^{\beta}\| _{L_p(I)} \leq C\| (PW)(x)x^{\beta}\| _{L_p(I)}. \] E (Markov inequality). Let \(W\in{\mathcal L}(C^2),\;0<p\leq\infty,\;\beta>-1/p\) if \(p<\infty\) and \(\beta\geq 0\) if \(p=\infty\). For \(n\geq 1, P\in{\mathcal P}_n\): \[ \begin{aligned} \| (P'W)(x)x^{\beta}\| _{L_p(I)}&\leq C{n^2\over a_n}\| (PW)(x)x^{\beta}\| _{L_p(I)},\\ \| (P'W)(x)x^{\beta}\| _{L_p[a_{\gamma n},d)}&\leq C{n\over a_n}\sqrt{T(a_n)}\| (PW)(x)x^{\beta}\| _{L_p(I)}, \end{aligned} \] The (rather) intricate proofs are given in detail, along with several examples.
    0 references
    0 references
    orthogonal polynomials
    0 references
    exponential weight
    0 references
    zero spacing
    0 references
    Lagrange interpolation
    0 references
    weighted Markov inequality
    0 references
    weighted Bernstein inequality
    0 references
    0 references