energy_filter_gaussian_lower_step
For n>0, t≥0, and t+1≤n, exp(−((t+1)²/(n−t)−t²/(n−t+1))) is at most (n−t)/(n+t+1).
Statement
theorem energy_filter_gaussian_lower_step (n t : ℝ)
(hn : 0 < n) (ht : 0 ≤ t) (htn : t + 1 ≤ n) :
Real.exp (-(((t+1)^2/(n-t))-(t^2/(n-t+1)))) ≤ (n-t)/(n+t+1)- Source
- OP-892: centered binomial fourth-mass asymptotic via rescaled step functions and dominated convergence.
- Verified
- 14 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext- Used by 1
- energy_filter_centered_binomial_gaussian_sandwichFor natural k≥1 and 0≤j≤k, choose(2k,k+j)/choose(2k,k) lies between exp(−j²/(k−j+1)) and exp(−j²/(k+j)).
The proof, explained
The proof compares both sides via the intermediate quantity x = (2t + 1) / (n - t). Expanding (t + 1)² = t² + 2t + 1 and using that the subtracted denominator n - t + 1 is larger than n - t bounds the exponent difference below by x. Negating and exponentiating gives exp(-Δ) ≤ exp(-x) = 1 / exp(x). Finally, the standard inequality 1 + x ≤ exp(x) inverts to 1 / exp(x) ≤ 1 / (1 + x), which algebraically simplifies to (n - t) / (n + t + 1).
0 < n - t and 0 < n - t + 1, so t² / (n - t + 1) ≤ t² / (n - t)
From the hypothesis t + 1 ≤ n, linear arithmetic shows that n - t ≥ 1 > 0, so both denominators n - t and n - t + 1 are strictly positive. Since t² ≥ 0 and n - t ≤ n - t + 1, dividing t² by the larger positive denominator yields a smaller or equal quotient.
uses
div_le_div_of_nonneg_left,sq_nonneghave hp : 0 < n-t := by linarith have hp' : 0 < n-t+1 := by linarith have hdiv : t^2/(n-t+1) ≤ t^2/(n-t) := div_le_div_of_nonneg_left (sq_nonneg t) hp (by linarith)
(2t + 1) / (n - t) ≤ (t + 1)² / (n - t) - t² / (n - t + 1)
Expanding the numerator (t + 1)² = t² + (2t + 1) and splitting the fraction into t² / (n - t) + (2t + 1) / (n - t), subtracting t² / (n - t + 1) leaves at least (2t + 1) / (n - t) because t² / (n - t + 1) ≤ t² / (n - t).
uses
add_divhave hdelta : (2*t+1)/(n-t) ≤ (t+1)^2/(n-t) - t^2/(n-t+1) := by have hpoly : (t+1)^2 = t^2 + (2*t+1) := by ring rw [hpoly] simp only [add_div] linarith
1 + (2t + 1) / (n - t) > 0 and (n - t) / (n + t + 1) = 1 / (1 + (2t + 1) / (n - t))
Because t ≥ 0 and n - t > 0, the fraction (2t + 1) / (n - t) is positive, making 1 + (2t + 1) / (n - t) strictly positive. Multiplying the numerator and denominator of 1 / (1 + (2t + 1) / (n - t)) by n - t simplifies it to (n - t) / ((n - t) + (2t + 1)) = (n - t) / (n + t + 1).
have hz : 0 < 1+(2*t+1)/(n-t) := by positivity have hid : (n-t)/(n+t+1) = 1/(1+(2*t+1)/(n-t)) := by field_simp; ring
1 + (2t + 1) / (n - t) ≤ exp((2t + 1) / (n - t))
This applies the standard inequality 1 + x ≤ exp(x) to x = (2t + 1) / (n - t).
uses
Real.add_one_le_exphave hexp : 1+(2*t+1)/(n-t) ≤ Real.exp ((2*t+1)/(n-t)) := by linarith [Real.add_one_le_exp ((2*t+1)/(n-t))]
exp(-((t + 1)² / (n - t) - t² / (n - t + 1))) ≤ (n - t) / (n + t + 1)
A calculation chains the inequalities together: 1. By monotonicity of the exponential function, the lower bound on the exponent difference gives exp(-Δ) ≤ exp(-x) for x = (2t + 1) / (n - t). 2. Rewriting exp(-x) as 1 / exp(x) turns this into 1 / exp(x). 3. Taking reciprocals of 1 + x ≤ exp(x) yields 1 / exp(x) ≤ 1 / (1 + x). 4. Replacing 1 / (1 + x) with (n - t) / (n + t + 1) completes the proof.
uses
Real.exp_le_exp.mpr,Real.exp_neg,one_div,one_div_le_one_div_of_lecalc _ ≤ Real.exp (-((2*t+1)/(n-t))) := by apply Real.exp_le_exp.mpr; linarith _ = 1 / Real.exp ((2*t+1)/(n-t)) := by rw [Real.exp_neg, one_div] _ ≤ 1 / (1+(2*t+1)/(n-t)) := one_div_le_one_div_of_le hz hexp _ = _ := hid.symm
div_le_div_of_nonneg_left- For non-negative a and positive c, if c ≤ b, then a / b ≤ a / c.
sq_nonneg- The square of any real number is non-negative: 0 ≤ a².
add_div- Division distributes over addition from the right: (a + b) / c = a / c + b / c.
Real.add_one_le_exp- For any real number x, 1 + x ≤ exp(x).
Real.exp_le_exp.mpr- The real exponential function is monotone: if x ≤ y, then exp(x) ≤ exp(y).
Real.exp_neg- The exponential of a negated number is its multiplicative inverse: exp(-x) = (exp(x))⁻¹.
one_div- The fraction 1 / a equals the inverse a⁻¹.
one_div_le_one_div_of_le- Taking reciprocals reverses inequality between positive numbers: if 0 < a ≤ b, then 1 / b ≤ 1 / a.
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.
theorem energy_filter_gaussian_lower_step (n t : ℝ) (hn : 0 < n) (ht : 0 ≤ t) (htn : t + 1 ≤ n) : Real.exp (-(((t+1)^2/(n-t))-(t^2/(n-t+1)))) ≤ (n-t)/(n+t+1) := by have hp : 0 < n-t := by linarith have hp' : 0 < n-t+1 := by linarith have hdiv : t^2/(n-t+1) ≤ t^2/(n-t) := div_le_div_of_nonneg_left (sq_nonneg t) hp (by linarith) have hdelta : (2*t+1)/(n-t) ≤ (t+1)^2/(n-t) - t^2/(n-t+1) := by have hpoly : (t+1)^2 = t^2 + (2*t+1) := by ring rw [hpoly] simp only [add_div] linarith have hz : 0 < 1+(2*t+1)/(n-t) := by positivity have hid : (n-t)/(n+t+1) = 1/(1+(2*t+1)/(n-t)) := by field_simp; ring have hexp : 1+(2*t+1)/(n-t) ≤ Real.exp ((2*t+1)/(n-t)) := by linarith [Real.add_one_le_exp ((2*t+1)/(n-t))] calc _ ≤ Real.exp (-((2*t+1)/(n-t))) := by apply Real.exp_le_exp.mpr; linarith _ = 1 / Real.exp ((2*t+1)/(n-t)) := by rw [Real.exp_neg, one_div] _ ≤ 1 / (1+(2*t+1)/(n-t)) := one_div_le_one_div_of_le hz hexp _ = _ := hid.symm