A constraint on extensible quadrature rules (Q5964943)

For \(\alpha >1\) and \(m>0\), let \({\mathcal F}\) be a class of integrable functions \(f:[0,1]^d \to {\mathbb R}\) with the following property of the worst case error \[ \sup_{f\in {\mathcal F}} | \frac{1}{n} \, \sum_{j=1}^n f(x_j) - \mu(f) | > m\,n^{-\alpha} \] for all \(n =1,\,2,\, \dots\) and for all \(x_j \in [0,\,1]^d\) \((j=1,\dots,n)\), where \(\mu(f)\) denotes the integral of \(f\). A sequence of points \(x_j \in [0,\,1]^d\) \((j = 1,2,\ldots)\) is called rate-optimal for \({\mathcal F}\) if there exists a sequence of integers \(n_k\) with \(n_1 < n_2 < \, \ldots\) and a constant \(M \geq m\) such that \[ \sup_{f\in {\mathcal F}} | \frac{1}{n} \, \sum_{j=1}^n f(x_j) - \mu(f) | \leq M\,n^{-\alpha} \] for all \(n \in \{n_1,\,n_2,\,\dots\}\). The authors shows that a rate-optimal sequence \(\{n_k\}\) must be spread out at least geometrically with \(n_{k+1}/n_k \geq \rho\). This extension bound \(\rho \in (1,\,2)\) depends on \(m/M\) and \(\alpha\). The same constraint holds in a root mean square error setting for any sequence of random points \(x_j \in [0,\,1]^d\).