Until recently, the admirable and impressive eXist XML database sometimes failed to optimize queries with numerical predicates. For example, a search for $i/following::word[1] would retrieve all <word> elements on the following axis and only then apply the predicate as a filter to return only the first of them. This was enormously inefficient when $i pointed to a node near the beginning of a very large document, with many thousands of following <word> elements. As an end-user without the Java programming skills to write optimization code for eXist, the author describes two types of optimization in the more familiar XML, XPath, and XQuery, which reduced the number of nodes that needed to be accessed and thus improved response time substantially.

A subsequent optimization introduced by the eXist developers into the eXist code base is described in an addendum to this paper. Although this revision partially obviates the need for the work-arounds developed earlier, the analysis of the efficiency of various XPath approaches to a single problem continues to provide valuable general lessons about XPath.