Almost everywhere convergence of spline sequences (Q2218714)

The authors consider sequences \((t_i)\) of grids points in \((0,1)\) where each of them occurs at most \(k-1\) times for \(k\ge 2\), which are not necessarily assumed to be dense in \((0,1)\). They denote by \(S_n\) the space of splines of order \(k\) (degree \(k-1)\) corresponding to the grid and by \(P_n\) the orthogonal projection from \(L^2([0,1],\lambda)\) onto \(S_n\), where \(\lambda\) denotes the Lebesgue measure. Shadrin and the second named author showed that for any \(g\in L^1([0,1],\lambda)\) the sequence \(g_n=P_n g\) converges almost everywhere and also that, in the case that \((t_i)\) is dense in the unit interval, given a sequence \((g_n)\) bounded in \(L^1\) such that \(g_n=P_n(g_{n+1})\), then \((g_n)\) converges almost everywhere to \(g\in L^1\) and \(\|g\|_1\le \sup\|g_n\|_1\) (see [\textit{M. Passenbrunner} and \textit{A. Shadrin}, J. Approx. Theory 180, 77--89 (2014; Zbl 1290.41010)]). In the present paper, it is shown for vector-valued functions that if \(X\) has the RNP and \((g_n)\in L^1_X\) satisfies \(\sup_n\|g_n\|_{L^1_X}<\infty\) and \(g_m=P_m(g_{n})\) for \(m\le n\), then \((g_n)\) converges almost everywhere to \(g\in L_X^1\). The result seems to be new even in the scalar-valued case for non-dense sequences of grids. This result is the analogue to the well-known result for vector-valued martingales.