int-llm — the math foundation, drawn
These are the seed visualizations behind fp_math.h — the pure-integer
Q16.48 math library at the core of int-llm.
They were built first, to convince myself that transcendental constants and functions
can be approximated deterministically with nothing but integers, shifts and adds. The library is what
grew out of them; the LLM is what the library turned out to be good enough to run.
Exploratory aids, not proofs — but they show what the math actually does.
Constants from integers
- √2 — dyadic refinement ·
fp_sqrt,isqrt128 - e — dyadic refinement ·
fp_exp - π — dyadic refinement ·
fp_compute_pi(Machin)
Functions & the payoff
- Transcendentals ·
fp_exp,fp_log,fp_sincos - eiπ + 1 = 0 — the complete journey · the most beautiful equation, in pure integers
The full library reference is in
FP_MATH.md. Euler's identity here is the
same one checked as a passing unit test in fp_test.c — computed end-to-end
in Q16.48 fixed point, no floating point anywhere.