Benchmarks¶

Computing pi digits with numbertoolkit versus mpmath, measured on an Apple Silicon laptop:

The gap widens with digit count because numbertoolkit's binary-splitting Chudnovsky implementation scales quasi-linearly, and the Rust extension avoids Python-level overhead in the hot loop.

The same holds for e_digits (factorial series with binary splitting):

Running the benchmarks yourself¶

The benchmark suite compares numbertoolkit against mpmath and stdlib baselines (Machin's formula via decimal for pi, an integer Taylor sum for e), and cross-checks that all implementations agree on the digits they produce:

make bench
# or directly:
uv run --group bench benchmarks/bench_pi.py
uv run --group bench benchmarks/bench_e.py
uv run --group bench benchmarks/bench_sqrt.py

The quadratic stdlib baselines are only run up to 5,000 digits.