Moses' lookup table approach "ITALU" has some things
in common with what we did with TILU. However he never
got it beyond about 10 integrals or so, as I recall.
TILU would look at a product and find the most
complicated part and use that part as the primary
index. The other parts would be secondary indexes etc.
There are a number of subtle issues requiring that
a pattern be indexed in multiple ways, though.
If this pointer is to Moses' whole PhD dissertation,
that actually makes for some interesting reading.
RJF
Wolfgang Jenkner wrote:
> Richard Fateman <fateman@cs.berkeley.edu> writes:
>
>
>>I have a number of tables that are out of copyright.
>>But I suggest that we make extra integral problems by
>>making a list of integrand "parts" e.g.
>>
>>p1=x^2-a^2
>>p2=sin(a*x+b)
>>p3=exp(a*x+b)
>>
>>etc/
>>Then compose integrands by exhaustion: pi*pj, pi/pj,
>>pi^2 etc.
>>
>>make a table.
>
>
> Is this piece of advice meant for implementing something like the
> pattern matching + hash code approach suggested by Moses, Symbolic
> Integration, AI Memo 97, p. 15,
>
> ftp://publications.ai.mit.edu/ai-publications/0-499/AIM-097.ps
>
> or for doing something else?
>
> Wolfgang
>