GSoC Week 11

Sorry for the delayed post! Last week was extremely busy.

It's time to wrap up my work. The good new is that rs_series (I called it series_fast earlier) works well for taylor series. The speedups are impressive and it can handle all sorts of cases (so far!). Now, I need to make it work for laurent and puiseux series.

Given that ring_series functions work well for negative and fractional powers, ideally that shouldn't be difficult. However, my current strategy is to add variables as generators to the currently used ring. The backend of creating rings is in polys, which doesn't allow negative or fractional powers in the generators (that is the mathematical definition of polynomials). For example:

In [276]: sring(a**QQ(2,3))
Out[276]: (Polynomial ring in a**(1/3) over ZZ with lex order, (a**(1/3))**2)

In [277]: _[0].gens
Out[277]: ((a**(1/3)),)

Contrast this with:

In [285]: sring(a**2)
Out[285]: (Polynomial ring in a over ZZ with lex order, a**2)

Generators with negative or fractional powers are treated as symbolic atoms and not as some variable raised to some power. So these fractional powers will never simplify with other generators with the same base.

The easy way to fix this is to modify sring but that would mean changing the core polys. I am still looking for a better way out.

The polynomial wrappers PR had been lying dead for quite some time. It currently uses piranha's hash_set but it needs to work on unordered_set when piranha is not available. I am adding that here. It is mostly done, except for encode and decode functions. Once the wrappers are in, I can start porting ring_series functions.

Next Week

  • Make rs_series work for puiseux series.

  • Complete polynomial wrappers.

  • Port the low level ring_series functions.