frac(x) immediate simplifications

[rwst]

As a followup to #21232 these automatic simplifications should be added:

sage: frac(frac(x))
frac(x)
sage: frac(floor(x))  # ceil
0
sage: assume(x, 'integer')
sage: frac(x)
0
sage: assume(x > 0)
sage: frac(tanh(x))
tanh(x)

CC: @mkoeppe

Component: symbolics

Author: Ralf Stephan

Branch/Commit: u/rws/frac_x__immediate_simplifications @ bf36286

Issue created by migration from https://trac.sagemath.org/ticket/21274

[rwst]

Branch: u/rws/frac_x__immediate_simplifications

[rwst]

Author: Ralf Stephan

[rwst]

New commits:

bf36286

21274: frac(x) immediate simplifications

[rwst]

Commit: bf36286

[mkoeppe]

comment:3

This all seems to work, but it's handling very specific cases and it would be much nicer if the system knew this:

sage: floor(x).is_integer()
True     # wishful

[rwst]

comment:4

Actually the mechanics for this exists (eg pynac/pynac#78) but floor, as it is not a GinacFunction, needs a different approach in Pynac than the ones implemented in that issue. Best I think is to first review #12121 and then after merge make floor a GinacFunction.

[tscrim]

comment:5

A very micro optimization:

sage: %timeit ZZ.zero()
The slowest run took 46.49 times longer than the fastest. This could mean that an intermediate result is being cached.
10000000 loops, best of 3: 61.5 ns per loop
sage: %timeit 0    # Preparsed as Integer(0) with the 0 being a python int
The slowest run took 40.29 times longer than the fastest. This could mean that an intermediate result is being cached.
10000000 loops, best of 3: 71 ns per loop
sage: z = int(0)
sage: %timeit Integer(z)
The slowest run took 53.93 times longer than the fastest. This could mean that an intermediate result is being cached.
10000000 loops, best of 3: 75.2 ns per loop

[videlec]

comment:6

Why is this part of the function frac? The function should not care about simplifying the result in any way. If you want a custom frac you would better consider having a frac method to symbolic expression (and having the function frac calling it).

[rwst]

comment:7

Very interesting. I completely forgot about these. First, if you look at functions/* you'll note a lot of expression arguments handled by custom code in eval() member functions, so called special values. We should write a Expression member function for all these functions? That would be an even larger class than it is now.

Secondly, I removed the Expression.sin() member and the only doctest that fails is the one explicitly calling it. So, it seems to me, other than taste or maybe convenience, there is no reason to have it. Didn't Occam demand not to multiply symbols beyond necessity?

I'm aware of the function eval() member being too general for expressions, and I'm constantly moving code into Pynac, so the special handling in eval() should be seen as a first step for testing code that is well maintainable because it's Python. If you insist I'll immediately move the frac code to C++. But please don't enlarge Expression any further and help with efforts to actually make it smaller.

[rwst]

comment:8

As a further data point SymPy does not have functions as expression members at all, not even .square() or .abs().