- From: Andrei Voronkov <voronkov@cs.man.ac.uk>
- Date: Tue, 7 Oct 2003 15:42:27 +0100
- To: Adrian Walker <adrianw@snet.net>
- Cc: "Enrico Franconi" <franconi@inf.unibz.it>, <www-rdf-logic@w3.org>, <www-rdf-interest@w3.org>, voronkov@cs.man.ac.uk
It is probably a discussion on a mailing list I am not a member os. What Adrian writes is true. Just a couple of subtle notes. > It's possible that an important point is overlooked when many of us say > things like "datalog can only compute things of polynomial complexity" It is right to say that Datalog can only EXPRESS things of polynomial complexity. "Compute" sounds too ambiguous. > Now it would seem that, if one is proposing a rule interpreter, then its > inputs include variable datalog programs P. So, the applicable complexity > would appear to be "complete in DEXPTIME". The use of data complexity as the main complexity measure in database theory is usually based on two assumptions: (1) The database may be huge, while queries (in this case datalog programs) are small. (2) The collection of queries is rather conservative and does not change a lot. This means you can optimise each query as a fixed one (but the database may change). It is my personal opinion but it seems both assumptions are questionable when non-relational data models are considered so the combined complexity (data+program, usually equivalent to the program complexity) may be more appropriate. Best, AV
Received on Tuesday, 7 October 2003 15:48:39 UTC