Re: [genius-list] Genius Future: Dr.Genius (Dr.Geo+Genius)

On Tue, 7 Sep 1999, Micah Yoder wrote:

> Yeah, I'm the guy with the semi-functional C++ classes for doing symbolic
> differentiation. I'd still like to see them merged in this, but my time for
> this type of thing has been pretty close to zero recently. :-(

Yeah, I knocked one up in scheme a while back to test the waters there. I
take my hat off to anyone who can write one in C++ though!

> Even worse, I'm currently stumped by a linker error when compiling the classes.
> If anyone is really familiar with C++, especially the linker side of things, and
> wants to take a look at it, let me know. If that could get solved, I could do
> some work on it again.

Chuck it over here, I'll have a look.

> But speaking of symbolic math, I thought I'd throw out a wild and probably nutty
> idea I recently had.

<Perl idea deleted>
Yes that is getting closer to what I had invisaged - but I think we really
need to develop an AI(modified A* still looks best) search type approach,
where the heuristics judges each method for their applicability.

> So it seems to me like object oriented programming is the perfect way to do
> symbolic differentiation, while something like this *might* be a good way to do
> integration.

Nah, OO is too lowlevel for this sort of thing - very quickly the classes
will tend to become giant chunkc of C.

> (No, I haven't done a serious viability study of this yet. I could be ***WAY***
> off base. But I thought it was an interesting idea.)

This idea is good, but I suspect to really do justice to this problem we
are going to need to implement a higher level interpreter which does the
sort of matching(more push-down rather than finite-state, and
non-deterministic) and implement the rules in that. BTW, where are you
getting your rules from?

Another point, I think the hardest part is not diff(as that is basically
mechanical) but rather simplification. Factorization is probably quite
hard too, integration is known to be hard. Things like pulling a term
into isolation, differentiation(including implicit), and possibly laplace
transform are all algorithmic anyway and probably not what needs focusing
on.

Maybe we(all those interested in symbolic math) should meet on #symgen on
irc.gimp.org sometime(I'm listening on that channel)

njh
Received on Tue Sep 07 1999 - 18:14:10 CDT