number-theory

A small pure-Python library of exact and semi-exact number theory computations: Bernoulli numbers, Euler (zigzag) numbers, the Riemann zeta function at integers, and a primality/prime-generation toolkit that ties them together into an Euler-product approximation of π.

No dependencies beyond the standard library, except prime.py which uses pycryptodome for a fast Miller-Rabin primality check.

Table of Contents

• Modules
• Installation
• Usage
• How It Fits Together
• Examples
• Limitations
• License

Modules

Installation

git clone https://github.com/Abhrankan-Chakrabarti/number-theory.git
cd number-theory
pip install -e .

This installs the number_theory package in editable mode along with its one dependency, pycryptodome.

Usage

Import the public API directly from the package:

from number_theory import B, E, zeta, isPrime, NP, piprime

B(10)              # Fraction(5, 66)
E(4)               # 5
zeta(4)            # 1/90π⁴  (exact, as a formatted string)
isPrime(97)        # True
NP(97)             # 101
piprime(2, 1000)   # ≈ 3.14157...

Each module can also be run standalone via python -m:

python -m number_theory.bernoulli   # prints B(0) through B(60)
python -m number_theory.prime       # prompts for n, checks primality
python -m number_theory.nextPrime   # prompts for a prime, finds the next one
python -m number_theory.zeta        # prompts for n, prints ζ(n)
python -m number_theory.piprime     # prompts for n and iteration count, approximates π

How It Fits Together

number_theory/
├── __init__.py    (exposes the public API)
├── piprime.py
│   ├── nextPrime.py → prime.py → Crypto.Util.number (Miller-Rabin)
│   └── zeta.py
│       ├── B.py → bernoulli.py   (exact Bernoulli numbers via Fraction)
│       └── E.py → zigzag.py      (Euler zigzag numbers via DP table)

Bernoulli numbers (bernoulli.py) are generated with the Akiyama–Tanigawa-style triangle algorithm, yielding exact Fraction values lazily — no floating-point error, no recomputation from scratch for each index.

Zigzag numbers (zigzag.py) count alternating permutations and appear in the odd-argument case of ζ via their connection to tangent/secant numbers.

zeta(n) returns an exact closed form:

• Positive even n: ζ(n) = (−1)^(n/2+1) · B(n) · 2^(n−1) / n! · πⁿ
• Non-positive n: ζ(n) = (−1)^(−n) · B(1−n) / (1−n) — the "trivial" values, all rational
• Positive odd n: no known closed form exists (this is genuinely open in general — ζ(3) is Apéry's constant); zeta.py prints a warning and returns an Euler-number-based expression rather than a proven identity

piprime.py uses the Euler product ζ(n) = ∏ₚ (1 − p⁻ⁿ)⁻¹ over the first iters primes, rearranged to solve for π:

π = ( ∏ₚ(1 − p⁻ⁿ) · ζ(n) )^(−1/n)

Since ζ(n) is known exactly for even n, this converges to π as more primes are included — a nice way to derive π purely from the distribution of primes.

prime.py and nextPrime.py provide the prime stream piprime.py walks over: isPrime uses pycryptodome's Miller-Rabin test by default (fast, probabilistic but effectively certain for practical sizes), with an optional 6k±1 trial-division mode; NP finds the next prime by stepping along the same 6k±1 wheel.

Examples

>>> from zeta import zeta
>>> zeta(2)
1/6π²
>>> zeta(4)
1/90π⁴
>>> zeta(-1)
-1/12
>>> zeta(-3)
1/120

>>> from piprime import piprime
>>> piprime(2, 1000)
3.1415727030005223
>>> piprime(2, 100000)
3.14159260...

More iterations (more primes in the product) converge more closely to π, though slowly — this is a demonstration of the identity, not an efficient way to compute π.

Limitations

• Odd zeta values (ζ(3), ζ(5), ...) have no known closed form in terms of π and rationals; zeta.py's odd-n branch returns an Euler-number expression as a placeholder, not a mathematically proven identity. Treat this case as exploratory only.
• piprime.py converges slowly — it's a demonstration of the Euler product identity, not a practical π-computation algorithm.
• prime.py's trial-division fallback (r=True, c=False) is only efficient for small n; the default Miller-Rabin path via pycryptodome should be used for anything large.

License

MIT — see LICENSE for details.