Documentation

LeanPool.Chudnovsky.Numerics

Certified numeric facts for the Chudnovsky formula (Phase D1) #

Sorried stubs for the certified numeric bounds consumed by Phase D (Coefficients.lean, ch. 10 of Milla, arXiv:1809.00533v6): high-precision bounds for π, √163, and the nome magnitude e^{-π√163} = ‖q τ₁₆₃‖, together with the exact evaluation q τ₁₆₃ = -e^{-π√163} (a negative real number, since e^{2πi·(1/2)} = -1).

Per PLAN.md (D1), these will eventually be proved by a small ℚ-interval-arithmetic layer (directed-rounding add/mul/div, Real.exp_bound Taylor remainders, rational square-root approximation + nlinarith), extending the machinery of Mathlib.Analysis.Real.Pi.Bounds for the π bounds.

Every digit string below has been verified by exact rational arithmetic (80-digit Decimal computation of π via Machin's formula, √163, and exp (-π·√163), then interval checks with fractions.Fraction):

Bounds for π #

π = 3.14159265358979323846264338327950288419716939937510…

theorem Chudnovsky.pi_gt_high_precision :
3.14159265358979323846 < Real.pi

High-precision lower bound for π (21 significant digits). To be proved by the Mathlib.Analysis.Real.Pi.Bounds machinery (PLAN D1).

theorem Chudnovsky.pi_lt_high_precision :
Real.pi < 3.14159265358979323847

High-precision upper bound for π (21 significant digits). To be proved by the Mathlib.Analysis.Real.Pi.Bounds machinery (PLAN D1).

Bounds for √163 #

√163 = 12.7671453348037046617109520097808923473823637803013…; the endpoint rationals below satisfy lo² < 163 < hi² (verified exactly).

theorem Chudnovsky.sqrt_163_gt :
12.7671453348037046617 < 163

Lower bound for √163 (21 significant digits). To be proved from lo² < 163 via Real.lt_sqrt / nlinarith (PLAN D1).

theorem Chudnovsky.sqrt_163_lt :
163 < 12.7671453348037046618

Upper bound for √163 (21 significant digits). To be proved from 163 < hi² via Real.sqrt_lt' / nlinarith (PLAN D1).

High-precision internal bounds (Phase D1 certificates) #

The two theorems on e^{-π√163} need π and √163 to far more than the 21 digits of pi_gt_high_precision. We certify π and √163 to 38 decimal places internally, using Mathlib's pi_lower_bound/pi_upper_bound witness machinery (the certificate rationals were generated by exact-arithmetic directed rounding) and rational squaring. All arithmetic is checked by the kernel via norm_num.

Bounds for e^{-π√163} #

e^{-π√163} = 3.808980937007652338226231516478005437619629319381…e-18 (30 significant digits certified below). This is ‖q τ₁₆₃‖, the quantity all the truncation-error estimates of ch. 5/10 are evaluated at.

theorem Chudnovsky.exp_neg_pi_sqrt_163_gt :
380898093700765233822623151647e-47 < Real.exp (-(Real.pi * 163))

Lower bound for e^{-π√163} (30 significant digits), i.e. 380898093700765233822623151647 / 10^47 < e^{-π√163}. To be proved by ℚ-interval arithmetic + Real.exp_bound (PLAN D1).

theorem Chudnovsky.exp_neg_pi_sqrt_163_lt :
Real.exp (-(Real.pi * 163)) < 380898093700765233822623151648e-47

Upper bound for e^{-π√163} (30 significant digits), i.e. e^{-π√163} < 380898093700765233822623151648 / 10^47. To be proved by ℚ-interval arithmetic + Real.exp_bound (PLAN D1).

The nome at τ₁₆₃ #

The magnitude of the nome at the CM point: ‖q τ₁₆₃‖ = e^{-π√163}.

The nome at the CM point is real and negative: q τ₁₆₃ = e^{2πi(1/2 + i√163/2)} = e^{πi}·e^{-π√163} = -e^{-π√163}.

Restatement of q_τ₁₆₃_eq via the real exponential: q τ₁₆₃ is the coercion of the negative real number -e^{-π√163}.