Barton Willis wrote:
> In the context of hypergeometic functions, I've been thinking about
> this problem too. For every identity in Abramowitz and Stegun and
> Gradsteyn-Rhyzik do we want to invent a name for the rule? What would
> that look like and how useful would it be? How could we organize this
> so that it is useful and easy to use?
>
>
The rules have to be set up so that the patterns distinguish match
parameters and their properties from
"constants". i.e. the matchdeclare( ) stuff is important.
For many rules there should be 2 directions.
Applying the rules may need to be done by some selection mechanism (e.g.
pointing with a mouse).
Some of the same ideas that went into Tilu could be used for organizing
the rules, e.g. to tell which
ones are even possible to apply in a given case. A kind of hash/tree
organization.
The match declarations were done by setting up a global list of pattern
variables that were
used again and again. e.g. n,m,i,j,k were integers...
For hypergeometric reductions, there are some attempts at building
algorithms though.
RJf