pyadic 0.3.0

p-Adic numbers and finite fields in Python

pyAdic

The pyadic library is Python 3 package that provides number types for finite fields $\mathbb{F}_p$ (ModP) and $p$-adic numbers $\mathbb{Q}_p$ (PAdic). The goal is to mimic the flexible behavior of built-in types, such as int, float and complex. Thus, one can mix-and-match the different number types, as long as the operations are consistent. In particular, ModP and PAdic are compatible with fractions.Fraction.

In addition to arithmetic operations, the pyadic library also provides the following functions:

• rationalise to perform rationalization ($\mathbb{F}_p\rightarrow \mathbb{Q}$ and $\mathbb{Q}_p \rightarrow \mathbb{Q}$);
• finite_field_sqrt and padic_sqrt to compute square roots (which may involve FieldExtension);
• padic_log to compute the $p$-adic logarithm.
• polynomial and rational function interpolation, see interpolation.py module.

A shout-out to galois for a very nice tool. It is recommented for vectorized finite field operations, unless type compatibility is an issue. For scalar operation this repo is recommended. See performance comparison below.

Installation

The package is available on the Python Package Index

pip install pyadic

Alternativelty, it can be installed by cloning the repo

git clone https://github.com/GDeLaurentis/pyadic.git path/to/repo
pip install -e path/to/repo

Requirements

pip will automatically install the required packages, which are

numpy, sympy

Additionally, pytest is needed for testing.

Testing

Extensive tests are implemented with pytest

pytest --cov pyadic/ --cov-report html tests/ --verbose

Quick Start

In [1]: from pyadic import PAdic, ModP
In [2]: from fractions import Fraction as Q

# 7/13 as a 12-digit 2147483647-adic number
In [3]: PAdic(Q(7, 13), 2147483647, 12)  
Out [3]: 1817101548 + 825955248*2147483647 + 1156337348*2147483647^2 + 330382099*2147483647^3 + 1321528398*2147483647^4 + 991146298*2147483647^5 + 1817101547*2147483647^6 + 825955248*2147483647^7 + 1156337348*2147483647^8 + 330382099*2147483647^9 + 1321528398*2147483647^10 + 991146298*2147483647^11 + O(2147483647^12)

# 7/13 in F_2147483647
In [4]: ModP(Q(7, 13), 2147483647)
Out [4]: 1817101548 % 2147483647

# Mapping back to rational numbers
In [5]: from pyadic.finite_field import rationalise
In [6]: rationalise(ModP(Q(7, 13), 2147483647))
Out [6]: Fraction(7, 13)
In [7]: rationalise(PAdic(Q(7, 13), 2147483647, 12))
Out [7]: Fraction(7, 13)

Perfomance comparison with galois for finite fields

Scalar instantiation and operations are faster in pyadic

import numpy
from galois import GF
from pyadic import ModP
from random import randint

GFp = GF(2 ** 31 - 1)
x = randint(0, 2 ** 31 - 1)

%timeit GFp(x)
2.84 µs ± 63.5 ns

%timeit ModP(x, 2 ** 31 - 1)
297 ns ± 0.876 ns

%timeit GFp(x) ** 2
30.1 µs ± 20.6 µs 

%timeit ModP(x, 2 ** 31 - 1) ** 2
2.23 µs ± 91.8 ns

while galois is faster for vectorized operations (the bigger the array the bigger the gain)

%timeit numpy.array([randint(0, 2 ** 31 - 1) for i in range(100)]).view(GFp) ** 2
65.6 µs ± 1.86 µs 

%timeit numpy.array([ModP(randint(0, 2 ** 31 - 1), 2 ** 31 - 1) for i in range(100)]) ** 2
351 µs ± 9.28 µs

However, galois requires everything to be appropriately typed, while pyadic performs type-casting on-the-fly

numpy.array([randint(0, 2 ** 31 - 1) for i in range(100)]).view(GFp) / 2
TypeError

numpy.array([ModP(randint(0, 2 ** 31 - 1), 2 ** 31 - 1) for i in range(100)]) / 2
array([...], dtype=object)

Citation

If you found this library useful, please consider citing it

@inproceedings{DeLaurentis:2023qhd,
    author = "De Laurentis, Giuseppe",
    title = "{Lips: $p$-adic and singular phase space}",
    booktitle = "{21th International Workshop on Advanced Computing and Analysis Techniques in Physics Research}: {AI meets Reality}",
    eprint = "2305.14075",
    archivePrefix = "arXiv",
    primaryClass = "hep-th",
    reportNumber = "PSI-PR-23-14",
    month = "5",
    year = "2023"
}