Orthogonal polynomial projection error measured in Sobolev norms in the unit ball (Q2358633): Difference between revisions

Let \(B^{d}\) be the unit ball of \(\mathbb{R}^{d},\) \(\alpha > -1\) and let \(W_{\alpha} : B^{d} \to \mathbb{R},\) \(W_{\alpha}(x) = (1-\| x \|^{2})^{\alpha}\) be a weight function of generalized Gegenbauer form, where \(\| \cdot \|\) is the Euclidean norm. If \(l \geq 0\) is a given integer number, then \(H^{l}_{\alpha}\) denotes the weighted Sobolev space whose squared norm is \(\| u \|^{2}_{H^l_{\alpha}} := \sum_{k=0}^{l} \, \| \nabla_{k} u \|^{2}_{\alpha},\) where \(\| u \|^{2}_{\alpha} := \int_{B^d} \, | u(x) |^{2} W_{\alpha}(x) \, dx\) is the natural squared norm defined on the weighted Lebesgue space \(L^{2}(B^d, W_{\alpha}) := \{ W_{\alpha}^{-\frac{1}{2}} f \, | \, f \in L^{2}(B^d) \}.\) Let \(S_{N}^{\alpha}\) be the orthogonal projector mapping \(L^{2}_{\alpha}(B^d, W_{\alpha})\) onto the space of \(d\)-variate polynomials of degree less than or equal to N. The main result of the paper is the following theorem. Theorem. For all integers \(1 \leq r \leq l\) there exists \(C = C(d,\alpha,l,r)>0\) such that \[ \| u-S_{N}^{d}(u) \|_{H_{\alpha}^{r}} \leq C N^{-\frac{1}{2}+2r-l} \| u \|_{H_{\alpha}^{l}} \] for all \(u \in H_{\alpha}^{l}.\)