Now, given a query we create a set as follows: we take the union of the champion lists for each of the terms comprising . We now restrict cosine computation to only the documents in . A critical parameter in this scheme is the value , which is highly application dependent. Intuitively, should be large compared with , especially if we use any form of the index elimination described in Section 7.1.2 . One issue here is that the value is set at the time of index construction, whereas is application dependent and may not be available until the query is received; as a result we may (as in the case of index elimination) find ourselves with a set that has fewer than documents. There is no reason to have the same value of for all terms in the dictionary; it could for instance be set to be higher for rarer terms.