overcomplete_ica_sample_size_hellinger_comparison
For all real numbers M_q and H_q_sq with 0 < M_q, if M_q / 4 ≤ H_q_sq and H_q_sq ≤ 2 * M_q, then 32 / M_q ≤ 64 / H_q_sq and 64 / H_q_sq ≤ 256 / M_q.
Statement
theorem overcomplete_ica_sample_size_hellinger_comparison (M_q H_q_sq : ℝ) (h : M_q / 4 ≤ H_q_sq) (h2 : H_q_sq ≤ 2 * M_q) (hM : 0 < M_q) : 32 / M_q ≤ 64 / H_q_sq ∧ 64 / H_q_sq ≤ 256 / M_q- Formalizes
- arXiv:2611.11852 Sec. 2.2, eq. (3)
- Verified
- 11 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext
Read back from the Lean
For all real numbers $M_q$ and $H_{q\_sq}$, if $M_q / 4 \le H_{q\_sq}$, $H_{q\_sq} \le 2 \cdot M_q$, and $0 < M_q$, then $32 / M_q \le 64 / H_{q\_sq}$ and $64 / H_{q\_sq} \le 256 / M_q$.
Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.
The proof, explained
The theorem shows that sample size bounds in terms of M_q and H_q^2 are within constant factors of one another when M_q / 4 ≤ H_q^2 ≤ 2 * M_q and M_q > 0. The proof first derives that H_q^2 > 0 from M_q > 0 and M_q / 4 ≤ H_q^2. It then splits the conjunction into two independent fraction inequalities and clears denominators using the positivity of both quantities, reducing each branch to linear inequalities that follow directly from the hypotheses.
0 < H_q_sq
From the assumptions that M_q > 0 and M_q / 4 ≤ H_q_sq, linear arithmetic immediately implies that H_q_sq is strictly positive.
have hH : 0 < H_q_sq := by linarith
It suffices to prove 32 / M_q ≤ 64 / H_q_sq and 64 / H_q_sq ≤ 256 / M_q separately.
The conjunction is split into two independent inequality goals using `constructor`.
constructor32 / M_q ≤ 64 / H_q_sq
Using the positivity of M_q and H_q_sq, clearing denominators with `div_le_div_iff₀` converts the fraction inequality into the equivalent cross-multiplication 32 * H_q_sq ≤ 64 * M_q (or H_q_sq ≤ 2 * M_q). This inequality is an immediate consequence of the hypothesis h2 (H_q_sq ≤ 2 * M_q) and is resolved by linear arithmetic.
uses
div_le_div_iff₀· rw [div_le_div_iff₀ hM hH] linarith
64 / H_q_sq ≤ 256 / M_q
Similarly, because H_q_sq and M_q are both positive, clearing denominators with `div_le_div_iff₀` turns the inequality into 64 * M_q ≤ 256 * H_q_sq (or M_q / 4 ≤ H_q_sq). This matches the hypothesis h (M_q / 4 ≤ H_q_sq) and is resolved by linear arithmetic.
uses
div_le_div_iff₀· rw [div_le_div_iff₀ hH hM] linarith
div_le_div_iff₀- For strictly positive denominators b and d in an ordered group with zero (such as the real numbers), a / b ≤ c / d holds if and only if a * d ≤ c * b.
Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
/-- Sample size bounds in terms of the squared Hellinger distance H_q^2 and the amplitude second moment M_q are comparable. -/theorem overcomplete_ica_sample_size_hellinger_comparison (M_q H_q_sq : ℝ) (h : M_q / 4 ≤ H_q_sq) (h2 : H_q_sq ≤ 2 * M_q) (hM : 0 < M_q) : 32 / M_q ≤ 64 / H_q_sq ∧ 64 / H_q_sq ≤ 256 / M_q := by have hH : 0 < H_q_sq := by linarith constructor · rw [div_le_div_iff₀ hM hH] linarith · rw [div_le_div_iff₀ hH hM] linarith