redistribution_harmonic_scalar_bound
For every natural m and real y with 0<=y<1, harmonic(m) times y^m is at most -log(1-y). This is the scalar multiplier estimate in the published harmonic trade-error proof.
Statement
theorem redistribution_harmonic_scalar_bound
(m : ℕ) (y : ℝ) (hy0 : 0 ≤ y) (hy1 : y < 1) :
(harmonic m : ℝ) * y^m ≤ -Real.log (1-y)- Source
- Uniform helper for IN-251, the proof of arXiv:2610.10995 Lemmas4.2/4.3. The disk logarithmic mean theorem and logarithm series already exist in the installed Mathlib; the upper-half-plane boundary passage, singular cosine integral and main harmonic bound remain unresolved. No research advance.
- Verified
- 27 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext
The proof, explained
The proof expands the harmonic number $H_m$ into the finite sum $\sum_{i=0}^{m-1} \frac{1}{i+1}$ and distributes $y^m$ across the terms. Because $0 \le y < 1$ and $i+1 \le m$, the powers satisfy $y^m \le y^{i+1}$, which bounds $H_m y^m$ term-by-term by the partial sum $\sum_{i=0}^{m-1} \frac{y^{i+1}}{i+1}$. Since every term is non-negative, this finite sum is bounded by the full infinite series $\sum_{n=0}^\infty \frac{y^{n+1}}{n+1}$, which sums to $-\log(1-y)$.
The power series $\sum_{n=0}^\infty \frac{y^{n+1}}{n+1}$ converges to $-\log(1-y)$.
Since $y \ge 0$, the absolute value $|y|$ equals $y$ by `abs_of_nonneg`. Thus $|y| < 1$ follows directly from the assumption $y < 1$, which allows applying `Real.hasSum_pow_div_log_of_abs_lt_one` to obtain the Taylor series representation of $-\log(1-y)$.
uses
Real.hasSum_pow_div_log_of_abs_lt_one,abs_of_nonneghave hs := Real.hasSum_pow_div_log_of_abs_lt_one (x := y) (by simpa only [abs_of_nonneg hy0] using hy1)
The product $(H_m : \mathbb{R}) \cdot y^m$ can be rewritten as the finite sum $\sum_{i=0}^{m-1} \frac{y^m}{i+1}$.
By definition of the harmonic number, $H_m = \sum_{i=0}^{m-1} \frac{1}{i+1}$. Multiplying by $y^m$ and distributing the multiplication across the finite sum via `sum_mul` expresses the left-hand side as $\sum_{i=0}^{m-1} \frac{1}{i+1} \cdot y^m$.
uses
sum_mulhave hf : (harmonic m : ℝ) = ∑ i ∈ range m, (1 / ((i+1 : ℕ) : ℝ)) := by simp [harmonic] rw [hf, sum_mul]
The sum $\sum_{i=0}^{m-1} \frac{y^m}{i+1}$ is bounded above by $\sum_{i=0}^{m-1} \frac{y^{i+1}}{i+1}$.
By `sum_le_sum`, it suffices to establish the inequality term by term for each index $i \in \{0, \dots, m-1\}$. Since $i < m$, we have $i+1 \le m$. For a base $0 \le y \le 1$, raising $y$ to a higher power decreases or preserves its value, so $y^m \le y^{i+1}$ by `pow_le_pow_of_le_one`. Multiplying both sides by the non-negative factor $(i+1)^{-1}$ via `mul_le_mul_of_nonneg_right` (using `one_div`, `div_eq_mul_inv`, `mul_comm`, and `mul_one` to rearrange division into multiplication by the inverse) gives $\frac{y^m}{i+1} \le \frac{y^{i+1}}{i+1}$.
uses
sum_le_sum,pow_le_pow_of_le_one,one_div,div_eq_mul_inv,mul_comm,mul_one,mul_le_mul_of_nonneg_rightcalc _ ≤ ∑ i ∈ range m, y^(i+1) / ((i+1 : ℕ) : ℝ) := by apply sum_le_sum intro i hi have hp := pow_le_pow_of_le_one hy0 hy1.le (by simpa using hi : i+1 ≤ m) simpa only [one_div, div_eq_mul_inv, mul_comm, mul_one] using mul_le_mul_of_nonneg_right hp (by positivity : 0 ≤ (((i+1 : ℕ) : ℝ))⁻¹)
The partial sum $\sum_{i=0}^{m-1} \frac{y^{i+1}}{i+1}$ is bounded above by $-\log(1-y)$.
Each term $y^{i+1}/(i+1)$ is non-negative because $y \ge 0$ (verified by `div_nonneg` and `pow_nonneg`). Any finite partial sum of a convergent series with non-negative terms is bounded by the total sum of the series, which is $-\log(1-y)$ by the power series evaluation from the first step.
uses
div_nonneg,pow_nonneg,Nat.cast_add,Nat.cast_one_ ≤ -Real.log (1-y) := by have ht := hs.summable.sum_le_tsum (range m) (fun i _ => div_nonneg (pow_nonneg hy0 _) (by positivity)) simpa only [hs.tsum_eq, Nat.cast_add, Nat.cast_one] using ht
Real.hasSum_pow_div_log_of_abs_lt_one- For real $x$ with $|x| < 1$, the power series $\sum_{n=0}^\infty \frac{x^{n+1}}{n+1}$ converges to $-\log(1 - x)$.
abs_of_nonneg- For any non-negative element $a \ge 0$, its absolute value $|a|$ equals $a$.
sum_mul- A factor distributes over a finite sum: $(\sum_{i \in s} f(i)) \cdot a = \sum_{i \in s} (f(i) \cdot a)$.
sum_le_sum- If $f(i) \le g(i)$ for all indices $i \in s$, then $\sum_{i \in s} f(i) \le \sum_{i \in s} g(i)$.
pow_le_pow_of_le_one- For a base $a$ satisfying $0 \le a \le 1$, if $m \le n$ then $a^n \le a^m$.
one_div- The quotient $1 / a$ equals the multiplicative inverse $a^{-1}$.
div_eq_mul_inv- Division equals multiplication by the inverse: $a / b = a \cdot b^{-1}$.
mul_comm- Multiplication is commutative: $a \cdot b = b \cdot a$.
mul_one- Multiplying by $1$ leaves an element unchanged: $a \cdot 1 = a$.
mul_le_mul_of_nonneg_right- If $b \le c$ and $0 \le a$, then $b \cdot a \le c \cdot a$.
div_nonneg- If $a \ge 0$ and $b \ge 0$, then $a / b \ge 0$.
pow_nonneg- If $a \ge 0$, then $a^n \ge 0$ for any natural number $n$.
Nat.cast_add- Casting a sum of natural numbers preserves addition: $\uparrow(m + n) = \uparrow m + \uparrow n$.
Nat.cast_one- Casting the natural number $1$ yields the identity $1$.
- In Lean, `harmonic m` and the power series for $-\log(1-y)$ are indexed starting at $0$ with denominator $i+1$, rather than the conventional $1$-based indexing $\sum_{k=1}^m \frac{1}{k}$.
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.
open Finset Real MeasureTheory Metric MeromorphicOntheorem redistribution_harmonic_scalar_bound (m : ℕ) (y : ℝ) (hy0 : 0 ≤ y) (hy1 : y < 1) : (harmonic m : ℝ) * y^m ≤ -Real.log (1-y) := by have hs := Real.hasSum_pow_div_log_of_abs_lt_one (x := y) (by simpa only [abs_of_nonneg hy0] using hy1) have hf : (harmonic m : ℝ) = ∑ i ∈ range m, (1 / ((i+1 : ℕ) : ℝ)) := by simp [harmonic] rw [hf, sum_mul] calc _ ≤ ∑ i ∈ range m, y^(i+1) / ((i+1 : ℕ) : ℝ) := by apply sum_le_sum intro i hi have hp := pow_le_pow_of_le_one hy0 hy1.le (by simpa using hi : i+1 ≤ m) simpa only [one_div, div_eq_mul_inv, mul_comm, mul_one] using mul_le_mul_of_nonneg_right hp (by positivity : 0 ≤ (((i+1 : ℕ) : ℝ))⁻¹) _ ≤ -Real.log (1-y) := by have ht := hs.summable.sum_le_tsum (range m) (fun i _ => div_nonneg (pow_nonneg hy0 _) (by positivity)) simpa only [hs.tsum_eq, Nat.cast_add, Nat.cast_one] using ht