This was one busy week again. I implemented polar numbers and changed over the integration and hyperexpand code to use them. This was more painful than I thought, but it seems to work now. Indeed the problems I mentioned before (other than matching) are gone now, as are the hacks. After this I cleaned up my branch to such an extent that I considered it ready for review. There are still some minor issues regarding numerical evaluation that I’m working to sort out, but this shouldn’t affect the review much.
Then I started working on adding lerchphi and polylogarithms to sympy. The goal of this is to incorporate them into hyperexpand(), so that a few more interesting series can be summed. This is good fun. Here are a few exampes. As you can see, all the standard things that one expects to work do work. And expand_func() can be used to reduce lerchphi to polylogarithms. In fact it can also reduce to hurwitz zeta functions in some cases but that is a mess. However, it is correct (tested numerically) and in a specific sense even simpler. In any case it’s nice to have it, even if it is not used much.
I also started extending hyperexpand to recognise lerch phi … this is slightly non-trivial (compared to normal table extensions) because lerch phi is actually not hypergeometric unless the parameter s is an integer, and even then the number of parameters of the hypergeometric function depends on s. Thus we need a special function to recognise such hypergeometric functions and generate formulae on the fly, this is what I am working on now. It will be finished on monday (so I have something cool to show off next week again *g*).
Finally it turns out that there are some subtle bugs in random numeric testing for hyperexpand. Since the code is in master now and we are about to release this is fairly bad of course. But luckily another gsoc student is setting up jenkins and configured it in such a way to currently run the relevant tests every five minutes. That should allow me to weed out all bugs.