The combination technique and some generalisations (Q860989)

The paper is concerned with the combination technique, as an effective tool for high-dimensional approximation with sparse grid spaces, which provides the exact sparse grid solution when the partial projection operators commute. In this aspect, the authors analyse approximation problems in terms of partial projections and their commutators. They derive conditions for when combinations of projections form again a projection and also errors for a general combination technique. Numerical experiments are presented on an application from machine learning, related to the reconstruction of a function from sample evaluations.