pell_family_asymptotic_base
The sequence (2 * pell_number (n - 2))^(1/n) converges to 1 + Real.sqrt 2 as n tends to infinity.
Statement
theorem pell_family_asymptotic_base : Filter.Tendsto (fun n : ℕ => ((2 * (pell_number (n - 2) : ℝ)) ^ ((1 : ℝ) / (n : ℝ)))) Filter.atTop (nhds (1 + Real.sqrt 2))
- Formalizes
- arXiv:2610.08525 Sec. 5, Theorem 14
- Verified
- 20 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 1
- pell_numberThe Pell numbers are defined by P(0) = 0, P(1) = 1, and P(m+2) = P(m) + 2 P(m+1).
Read back from the Lean
As $n \to \infty$ through the natural numbers, the sequence given by $(2 \cdot P(n - 2))^{1/n}$ converges to $1 + \sqrt{2}$ (with respect to the standard topology on $\mathbb{R}$), where $P(k)$ is the $k$-th Pell number (with $n - 2$ denoting truncated natural number subtraction).
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 proof establishes the asymptotic limit of (2 P_{n-2})^{1/n} using Binet's closed form P_m = (aᵐ - bᵐ)/(2√2), where a = 1 + √2 and b = 1 - √2. Because |b| ≤ 1/2 < 1, the term bᵐ is bounded by (1/2) aᵐ for m ≥ 1, which sandwiches 2 P_{n-2} between constant multiples of aⁿ: aⁿ c₁ ≤ 2 P_{n-2} ≤ aⁿ c₂. Taking 1/n-th powers sandwiches (2 P_{n-2})^{1/n} between a c₁^{1/n} and a c₂^{1/n}. Since c^{1/n} → 1 as n → ∞ for any positive constant c, both bounds converge to a = 1 + √2, and the squeeze theorem yields the result.
The roots a = 1 + √2 and b = 1 - √2 satisfy a > 0, 1 ≤ a, a² = 2a + 1, b² = 2b + 1, |b| ≤ 1/2, and 2√2 ≠ 0.
We establish rational bounds on √2, namely 1 < √2 < 3/2, by squaring and comparing with rational numbers (1 < 2 < 9/4). The characteristic roots a = 1 + √2 and b = 1 - √2 satisfy the recurrence relation x² = 2x + 1. Using 1 < √2 < 3/2, we get a ≥ 1, a > 0, and |b| = √2 - 1 ≤ 1/2, while 2√2 is strictly positive and therefore non-zero.
uses
Real.lt_sqrt,Real.sqrt_lt',Real.sq_sqrthave hsqrt2_pos : 0 < Real.sqrt 2 := Real.sqrt_pos.2 (by norm_num) have hsqrt2_gt : (1:ℝ) < Real.sqrt 2 := by rw [Real.lt_sqrt (by norm_num)]; norm_num have hsqrt2_lt : Real.sqrt 2 < 3/2 := by rw [Real.sqrt_lt' (by norm_num : (0:ℝ) < 3/2)]; norm_num have hsq2 : Real.sqrt 2 ^ 2 = 2 := Real.sq_sqrt (by norm_num) set b : ℝ := 1 - Real.sqrt 2 with hb_def set a : ℝ := 1 + Real.sqrt 2 with ha_def have a_pos : 0 < a := by rw [ha_def]; linarith have a_ge_one : 1 ≤ a := by rw [ha_def]; linarith have a_sq : a ^ 2 = 2 * a + 1 := by rw [ha_def]; nlinarith have b_sq : b ^ 2 = 2 * b + 1 := by rw [hb_def]; nlinarith have hb : |b| ≤ 1/2 := abs_le.2 ⟨by rw [hb_def]; linarith, by rw [hb_def]; linarith⟩ have hsqrt2_ne : (2 * Real.sqrt 2 : ℝ) ≠ 0 := by positivity
For all m ∈ ℕ, the Pell numbers satisfy the closed-form Binet formula P_m = (aᵐ - bᵐ) / (2√2).
The formula is proved by two-step induction on m. The base cases m = 0 and m = 1 follow directly from P₀ = 0 and P₁ = 1, since a⁰ - b⁰ = 0 and a¹ - b¹ = (1 + √2) - (1 - √2) = 2√2. For the inductive step, substituting the recurrence P_{n+2} = 2 P_{n+1} + P_n and the induction hypotheses reduces the claim to a^{n+2} - b^{n+2} = 2(a^{n+1} - b^{n+1}) + (aⁿ - bⁿ), which holds because a² = 2a + 1 and b² = 2b + 1.
uses
Nat.twoStepInduction,pell_number.eq_1,pell_number.eq_2,pell_number.eq_3,pow_one,div_self,pow_addhave closed_form : ∀ m : ℕ, (pell_number m : ℝ) = (a ^ m - b ^ m) / (2 * Real.sqrt 2) := by intro m induction m using Nat.twoStepInduction with | zero => simp [pell_number.eq_1] | one => have h1 : a ^ 1 - b ^ 1 = 2 * Real.sqrt 2 := by rw [pow_one, pow_one, ha_def, hb_def]; ring rw [pell_number.eq_2, h1, div_self hsqrt2_ne] norm_num | more n ih1 ih2 => have hstep : a ^ (n + 2) - b ^ (n + 2) = 2 * (a ^ (n + 1) - b ^ (n + 1)) + (a ^ n - b ^ n) := by rw [pow_add, pow_add, pow_add, pow_add, a_sq, b_sq] ring change ((pell_number (n + 2) : ℕ) : ℝ) = (a ^ (n + 2) - b ^ (n + 2)) / (2 * Real.sqrt 2) rw [show n + 2 = n.succ.succ from rfl, pell_number.eq_3, hstep] push_cast rw [ih1, ih2] ring
For every m ≥ 1, the quantity 2 P_m satisfies the bounds aᵐ / (2√2) ≤ 2 P_m ≤ 3 aᵐ / (2√2).
Since |b| ≤ 1/2, we have |b|ᵐ ≤ (1/2)ᵐ ≤ 1/2. Because a ≥ 1, we have 1 ≤ aᵐ, so |bᵐ| = |b|ᵐ ≤ (1/2) aᵐ. This gives -(1/2) aᵐ ≤ bᵐ ≤ (1/2) aᵐ, which bounds the numerator: (1/2) aᵐ ≤ aᵐ - bᵐ ≤ (3/2) aᵐ. Since 2 P_m = (aᵐ - bᵐ) / √2, dividing these bounds by √2 establishes the claimed lower and upper bounds on 2 P_m.
uses
pow_le_pow_left₀,abs_nonneg,pow_le_pow_of_le_one,pow_one,one_le_pow₀,mul_le_mul_of_nonneg_left,abs_pow,le_trans,le_abs_self,neg_abs_le,div_le_div_iff₀have bounds : ∀ m : ℕ, 1 ≤ m → 2 * (pell_number m : ℝ) ≤ 3 * a ^ m / (2 * Real.sqrt 2) ∧ a ^ m / (2 * Real.sqrt 2) ≤ 2 * (pell_number m : ℝ) := by intro m hm have hkey : |b| ^ m ≤ (1/2) * a ^ m := by have h1 : |b| ^ m ≤ (1/2) ^ m := pow_le_pow_left₀ (abs_nonneg b) hb m have h2 : (1/2 : ℝ) ^ m ≤ (1/2) ^ 1 := pow_le_pow_of_le_one (by norm_num) (by norm_num) hm rw [pow_one] at h2 have h3 : (1 : ℝ) ≤ a ^ m := one_le_pow₀ a_ge_one have h4 : (1/2 : ℝ) * 1 ≤ (1/2) * a ^ m := by apply mul_le_mul_of_nonneg_left h3 (by norm_num) linarith have hbm : |b ^ m| ≤ (1/2) * a ^ m := by rw [abs_pow]; exact hkey have h1 : b ^ m ≤ (1/2) * a ^ m := le_trans (le_abs_self _) hbm have h2 : -(1/2) * a ^ m ≤ b ^ m := by have := neg_abs_le (b ^ m); linarith have hdu : a ^ m - b ^ m ≤ 3 * a ^ m / 2 := by linarith have hdl : a ^ m / 2 ≤ a ^ m - b ^ m := by linarith have e : (2:ℝ) * ((a ^ m - b ^ m) / (2 * Real.sqrt 2)) = (a ^ m - b ^ m) / Real.sqrt 2 := by field_simp constructor · rw [closed_form m, e, div_le_div_iff₀ hsqrt2_pos (by positivity)] nlinarith · rw [closed_form m, e, div_le_div_iff₀ (by positivity) hsqrt2_pos] nlinarith
For any positive constant c, c^{1/n} converges to 1 as n → ∞.
We express c^{1/n} as exp((ln c) * (1/n)). Because 1/n → 0 as n → ∞, the product (ln c) * (1/n) tends to 0. By continuity of the exponential function, exp((ln c) * (1/n)) converges to exp(0) = 1.
uses
Filter.Tendsto.const_mul,tendsto_one_div_atTop_nhds_zero_nat,mul_zero,Function.comp_def,Real.rpow_def_of_poshave hlim : ∀ c : ℝ, 0 < c → Filter.Tendsto (fun n : ℕ => c ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds 1) := by intro c hc have h1 : Filter.Tendsto (fun n : ℕ => Real.log c * (1/(n:ℝ))) Filter.atTop (nhds (Real.log c * 0)) := Filter.Tendsto.const_mul (Real.log c) tendsto_one_div_atTop_nhds_zero_nat rw [mul_zero] at h1 have h2 : Filter.Tendsto (fun n : ℕ => Real.exp (Real.log c * (1/(n:ℝ)))) Filter.atTop (nhds 1) := by have := Real.continuous_exp.continuousAt.tendsto.comp h1 simpa [Function.comp_def] using this simpa only [Real.rpow_def_of_pos hc] using h2
For positive constants c₁ = 1 / (2√2 a²) and c₂ = 3 / (2√2 a²), every n ≥ 3 satisfies a * c₁^{1/n} ≤ (2 P_{n-2})^{1/n} ≤ a * c₂^{1/n}.
For n ≥ 3, we set m = n - 2 ≥ 1 and rewrite a^{n-2} as aⁿ / a². The bounds on 2 P_{n-2} become aⁿ c₁ ≤ 2 P_{n-2} ≤ aⁿ c₂. Since taking 1/n-th powers is monotone on non-negative reals, (aⁿ c₁)^{1/n} ≤ (2 P_{n-2})^{1/n} ≤ (aⁿ c₂)^{1/n}. Factoring the powers yields (aⁿ cᵢ)^{1/n} = (aⁿ)^{1/n} cᵢ^{1/n} = a cᵢ^{1/n}, giving the desired sandwich bounds.
uses
Real.rpow_natCast,Real.rpow_mul,Real.rpow_one,pow_add,Real.rpow_le_rpow,Real.mul_rpow,pow_nonnegset c₁ : ℝ := 1 / (a ^ 2 * (2 * Real.sqrt 2)) with hc1_def set c₂ : ℝ := 3 / (a ^ 2 * (2 * Real.sqrt 2)) with hc2_def have hc1_pos : 0 < c₁ := by rw [hc1_def]; positivity have hc2_pos : 0 < c₂ := by rw [hc2_def]; positivity have step : ∀ n : ℕ, 3 ≤ n → a * c₁ ^ ((1:ℝ)/(n:ℝ)) ≤ ((2 * (pell_number (n-2) : ℝ)) ^ ((1:ℝ)/(n:ℝ))) ∧ ((2 * (pell_number (n-2) : ℝ)) ^ ((1:ℝ)/(n:ℝ))) ≤ a * c₂ ^ ((1:ℝ)/(n:ℝ)) := by intro n hn have hn3 : 1 ≤ n - 2 := by omega have hnn : n ≠ 0 := by omega have hn0 : (n : ℝ) ≠ 0 := by exact_mod_cast hnn have ha_rpow : (a ^ n : ℝ) ^ ((1:ℝ) / (n:ℝ)) = a := by rw [← Real.rpow_natCast a n, ← Real.rpow_mul a_pos.le] have he : (n:ℝ) * (1/(n:ℝ)) = 1 := by field_simp rw [he, Real.rpow_one] have hpow2 : a ^ (n - 2) = a ^ n / a ^ 2 := by have hpow : a ^ n = a ^ (n - 2) * a ^ 2 := by rw [← pow_add]; congr 1; omega rw [hpow] have : (a^2 : ℝ) ≠ 0 := by positivity field_simp have hbnd := bounds (n-2) hn3 have hlo : a ^ n * c₁ ≤ 2 * (pell_number (n-2) : ℝ) := by have h := hbnd.2 rw [hpow2] at h have he : a ^ n / a ^ 2 / (2 * Real.sqrt 2) = a ^ n * c₁ := by rw [hc1_def]; field_simp rwa [he] at h have hhi : 2 * (pell_number (n-2) : ℝ) ≤ a ^ n * c₂ := by have h := hbnd.1 rw [hpow2] at h have he : 3 * (a ^ n / a ^ 2) / (2 * Real.sqrt 2) = a ^ n * c₂ := by rw [hc2_def]; field_simp rwa [he] at h have hA_nn : (0:ℝ) ≤ 2 * (pell_number (n-2) : ℝ) := by positivity have hexp_nn : (0:ℝ) ≤ 1/(n:ℝ) := by positivity constructor · have h1 := Real.rpow_le_rpow (by positivity : (0:ℝ) ≤ a ^ n * c₁) hlo hexp_nn rwa [Real.mul_rpow (pow_nonneg a_pos.le n) hc1_pos.le, ha_rpow] at h1 · have h1 := Real.rpow_le_rpow hA_nn hhi hexp_nn rwa [Real.mul_rpow (pow_nonneg a_pos.le n) hc2_pos.le, ha_rpow] at h1
Both bounding sequences a * c₁^{1/n} and a * c₂^{1/n} converge to a = 1 + √2 as n → ∞.
Since c₁ > 0 and c₂ > 0, the limit lemma implies that c₁^{1/n} → 1 and c₂^{1/n} → 1. Multiplying each by the constant a shows that both bounding sequences converge to a * 1 = a = 1 + √2.
have hlimlo : Filter.Tendsto (fun n : ℕ => a * c₁ ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds a) := by have h := (hlim c₁ hc1_pos).const_mul a simpa using h have hlimhi : Filter.Tendsto (fun n : ℕ => a * c₂ ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds a) := by have h := (hlim c₂ hc2_pos).const_mul a simpa using h
The sequence (2 P_{n-2})^{1/n} converges to 1 + √2 as n → ∞.
The sandwich bounds a * c₁^{1/n} ≤ (2 P_{n-2})^{1/n} ≤ a * c₂^{1/n} hold for all n ≥ 3, which is an event that holds eventually as n → ∞. Because both bounds converge to 1 + √2, the squeeze theorem (tendsto_of_tendsto_of_tendsto_of_le_of_le') implies that (2 P_{n-2})^{1/n} also converges to 1 + √2.
uses
tendsto_of_tendsto_of_tendsto_of_le_of_le',Filter.eventually_ge_atToprefine tendsto_of_tendsto_of_tendsto_of_le_of_le' hlimlo hlimhi ?_ ?_ · filter_upwards [Filter.eventually_ge_atTop 3] with n hn exact (step n hn).1 · filter_upwards [Filter.eventually_ge_atTop 3] with n hn exact (step n hn).2
Real.lt_sqrt- For non-negative x, x < √y if and only if x² < y.
Real.sqrt_lt'- For positive y, √x < y if and only if x < y².
Real.sq_sqrt- For non-negative x, (√x)² = x.
Nat.twoStepInduction- Two-step induction principle on natural numbers with base cases 0 and 1.
pell_number.eq_1- The 0-th Pell number is 0.
pell_number.eq_2- The 1-st Pell number is 1.
pell_number.eq_3- The recurrence relation P_{m+2} = 2 P_{m+1} + P_m for Pell numbers.
pow_one- Any element raised to the power 1 equals itself.
div_self- Any non-zero element divided by itself equals 1.
pow_add- Exponents add under multiplication: a^{m+n} = a^m a^n.
pow_le_pow_left₀- If 0 ≤ a ≤ b, then aⁿ ≤ bⁿ for all n ∈ ℕ.
abs_nonneg- The absolute value of any element is non-negative.
pow_le_pow_of_le_one- For 0 ≤ a ≤ 1, raising a to a larger power yields a smaller or equal result: if m ≤ n, then aⁿ ≤ aᵐ.
one_le_pow₀- If 1 ≤ a, then 1 ≤ aⁿ for all n ∈ ℕ.
mul_le_mul_of_nonneg_left- Multiplying an inequality b ≤ c on the left by a non-negative element preserves the inequality: a * b ≤ a * c.
abs_pow- The absolute value of a power is the power of the absolute value: |aⁿ| = |a|ⁿ.
le_trans- Transitivity of inequality: if a ≤ b and b ≤ c, then a ≤ c.
le_abs_self- Every number is bounded above by its absolute value: a ≤ |a|.
neg_abs_le- The negative absolute value of a is less than or equal to a: -|a| ≤ a.
div_le_div_iff₀- For positive denominators b and d, a / b ≤ c / d if and only if a * d ≤ c * b.
Filter.Tendsto.const_mul- If f(n) → L as n → ∞, then c * f(n) → c * L.
tendsto_one_div_atTop_nhds_zero_nat- The sequence 1/n converges to 0 as n → ∞.
mul_zero- Multiplying any number by zero yields zero: a * 0 = 0.
Function.comp_def- Definition of function composition: (f ∘ g)(x) = f(g(x)).
Real.rpow_def_of_pos- For x > 0 and real y, xʸ = exp(y * ln x).
Real.rpow_natCast- The real power xⁿ with n cast to ℝ equals the natural power xⁿ.
Real.rpow_mul- For non-negative x, powers multiply: x^{y * z} = (xʸ)ᶻ.
Real.rpow_one- Any real number raised to the real power 1 equals itself.
Real.rpow_le_rpow- For 0 ≤ x ≤ y and non-negative z, xᶻ ≤ yᶻ.
Real.mul_rpow- For non-negative x and y, (x * y)ᶻ = xᶻ * yᶻ.
pow_nonneg- If a ≥ 0, then aⁿ ≥ 0 for all natural numbers n.
tendsto_of_tendsto_of_tendsto_of_le_of_le'- The squeeze theorem: if g(n) ≤ f(n) ≤ h(n) eventually, and both g(n) → L and h(n) → L, then f(n) → L.
Filter.eventually_ge_atTop- For any threshold N, eventually n ≥ N as n → ∞.
- The proofs of algebraic identities for the characteristic roots a and b, as well as several inequalities, rely on automated decision procedures (linarith, nlinarith, ring).
- The casts between natural numbers and real numbers (push_cast, exact_mod_cast) are formal bookkeeping with no paper counterpart.
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.
/-- The sequence (2 * pell_number (n - 2))^(1/n) converges to 1 + √2 as n → ∞. -/theorem pell_family_asymptotic_base : Filter.Tendsto (fun n : ℕ => ((2 * (pell_number (n - 2) : ℝ)) ^ ((1 : ℝ) / (n : ℝ)))) Filter.atTop (nhds (1 + Real.sqrt 2)) := by have hsqrt2_pos : 0 < Real.sqrt 2 := Real.sqrt_pos.2 (by norm_num) have hsqrt2_gt : (1:ℝ) < Real.sqrt 2 := by rw [Real.lt_sqrt (by norm_num)]; norm_num have hsqrt2_lt : Real.sqrt 2 < 3/2 := by rw [Real.sqrt_lt' (by norm_num : (0:ℝ) < 3/2)]; norm_num have hsq2 : Real.sqrt 2 ^ 2 = 2 := Real.sq_sqrt (by norm_num) set b : ℝ := 1 - Real.sqrt 2 with hb_def set a : ℝ := 1 + Real.sqrt 2 with ha_def have a_pos : 0 < a := by rw [ha_def]; linarith have a_ge_one : 1 ≤ a := by rw [ha_def]; linarith have a_sq : a ^ 2 = 2 * a + 1 := by rw [ha_def]; nlinarith have b_sq : b ^ 2 = 2 * b + 1 := by rw [hb_def]; nlinarith have hb : |b| ≤ 1/2 := abs_le.2 ⟨by rw [hb_def]; linarith, by rw [hb_def]; linarith⟩ have hsqrt2_ne : (2 * Real.sqrt 2 : ℝ) ≠ 0 := by positivity have closed_form : ∀ m : ℕ, (pell_number m : ℝ) = (a ^ m - b ^ m) / (2 * Real.sqrt 2) := by intro m induction m using Nat.twoStepInduction with | zero => simp [pell_number.eq_1] | one => have h1 : a ^ 1 - b ^ 1 = 2 * Real.sqrt 2 := by rw [pow_one, pow_one, ha_def, hb_def]; ring rw [pell_number.eq_2, h1, div_self hsqrt2_ne] norm_num | more n ih1 ih2 => have hstep : a ^ (n + 2) - b ^ (n + 2) = 2 * (a ^ (n + 1) - b ^ (n + 1)) + (a ^ n - b ^ n) := by rw [pow_add, pow_add, pow_add, pow_add, a_sq, b_sq] ring change ((pell_number (n + 2) : ℕ) : ℝ) = (a ^ (n + 2) - b ^ (n + 2)) / (2 * Real.sqrt 2) rw [show n + 2 = n.succ.succ from rfl, pell_number.eq_3, hstep] push_cast rw [ih1, ih2] ring have bounds : ∀ m : ℕ, 1 ≤ m → 2 * (pell_number m : ℝ) ≤ 3 * a ^ m / (2 * Real.sqrt 2) ∧ a ^ m / (2 * Real.sqrt 2) ≤ 2 * (pell_number m : ℝ) := by intro m hm have hkey : |b| ^ m ≤ (1/2) * a ^ m := by have h1 : |b| ^ m ≤ (1/2) ^ m := pow_le_pow_left₀ (abs_nonneg b) hb m have h2 : (1/2 : ℝ) ^ m ≤ (1/2) ^ 1 := pow_le_pow_of_le_one (by norm_num) (by norm_num) hm rw [pow_one] at h2 have h3 : (1 : ℝ) ≤ a ^ m := one_le_pow₀ a_ge_one have h4 : (1/2 : ℝ) * 1 ≤ (1/2) * a ^ m := by apply mul_le_mul_of_nonneg_left h3 (by norm_num) linarith have hbm : |b ^ m| ≤ (1/2) * a ^ m := by rw [abs_pow]; exact hkey have h1 : b ^ m ≤ (1/2) * a ^ m := le_trans (le_abs_self _) hbm have h2 : -(1/2) * a ^ m ≤ b ^ m := by have := neg_abs_le (b ^ m); linarith have hdu : a ^ m - b ^ m ≤ 3 * a ^ m / 2 := by linarith have hdl : a ^ m / 2 ≤ a ^ m - b ^ m := by linarith have e : (2:ℝ) * ((a ^ m - b ^ m) / (2 * Real.sqrt 2)) = (a ^ m - b ^ m) / Real.sqrt 2 := by field_simp constructor · rw [closed_form m, e, div_le_div_iff₀ hsqrt2_pos (by positivity)] nlinarith · rw [closed_form m, e, div_le_div_iff₀ (by positivity) hsqrt2_pos] nlinarith have hlim : ∀ c : ℝ, 0 < c → Filter.Tendsto (fun n : ℕ => c ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds 1) := by intro c hc have h1 : Filter.Tendsto (fun n : ℕ => Real.log c * (1/(n:ℝ))) Filter.atTop (nhds (Real.log c * 0)) := Filter.Tendsto.const_mul (Real.log c) tendsto_one_div_atTop_nhds_zero_nat rw [mul_zero] at h1 have h2 : Filter.Tendsto (fun n : ℕ => Real.exp (Real.log c * (1/(n:ℝ)))) Filter.atTop (nhds 1) := by have := Real.continuous_exp.continuousAt.tendsto.comp h1 simpa [Function.comp_def] using this simpa only [Real.rpow_def_of_pos hc] using h2 set c₁ : ℝ := 1 / (a ^ 2 * (2 * Real.sqrt 2)) with hc1_def set c₂ : ℝ := 3 / (a ^ 2 * (2 * Real.sqrt 2)) with hc2_def have hc1_pos : 0 < c₁ := by rw [hc1_def]; positivity have hc2_pos : 0 < c₂ := by rw [hc2_def]; positivity have step : ∀ n : ℕ, 3 ≤ n → a * c₁ ^ ((1:ℝ)/(n:ℝ)) ≤ ((2 * (pell_number (n-2) : ℝ)) ^ ((1:ℝ)/(n:ℝ))) ∧ ((2 * (pell_number (n-2) : ℝ)) ^ ((1:ℝ)/(n:ℝ))) ≤ a * c₂ ^ ((1:ℝ)/(n:ℝ)) := by intro n hn have hn3 : 1 ≤ n - 2 := by omega have hnn : n ≠ 0 := by omega have hn0 : (n : ℝ) ≠ 0 := by exact_mod_cast hnn have ha_rpow : (a ^ n : ℝ) ^ ((1:ℝ) / (n:ℝ)) = a := by rw [← Real.rpow_natCast a n, ← Real.rpow_mul a_pos.le] have he : (n:ℝ) * (1/(n:ℝ)) = 1 := by field_simp rw [he, Real.rpow_one] have hpow2 : a ^ (n - 2) = a ^ n / a ^ 2 := by have hpow : a ^ n = a ^ (n - 2) * a ^ 2 := by rw [← pow_add]; congr 1; omega rw [hpow] have : (a^2 : ℝ) ≠ 0 := by positivity field_simp have hbnd := bounds (n-2) hn3 have hlo : a ^ n * c₁ ≤ 2 * (pell_number (n-2) : ℝ) := by have h := hbnd.2 rw [hpow2] at h have he : a ^ n / a ^ 2 / (2 * Real.sqrt 2) = a ^ n * c₁ := by rw [hc1_def]; field_simp rwa [he] at h have hhi : 2 * (pell_number (n-2) : ℝ) ≤ a ^ n * c₂ := by have h := hbnd.1 rw [hpow2] at h have he : 3 * (a ^ n / a ^ 2) / (2 * Real.sqrt 2) = a ^ n * c₂ := by rw [hc2_def]; field_simp rwa [he] at h have hA_nn : (0:ℝ) ≤ 2 * (pell_number (n-2) : ℝ) := by positivity have hexp_nn : (0:ℝ) ≤ 1/(n:ℝ) := by positivity constructor · have h1 := Real.rpow_le_rpow (by positivity : (0:ℝ) ≤ a ^ n * c₁) hlo hexp_nn rwa [Real.mul_rpow (pow_nonneg a_pos.le n) hc1_pos.le, ha_rpow] at h1 · have h1 := Real.rpow_le_rpow hA_nn hhi hexp_nn rwa [Real.mul_rpow (pow_nonneg a_pos.le n) hc2_pos.le, ha_rpow] at h1 have hlimlo : Filter.Tendsto (fun n : ℕ => a * c₁ ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds a) := by have h := (hlim c₁ hc1_pos).const_mul a simpa using h have hlimhi : Filter.Tendsto (fun n : ℕ => a * c₂ ^ ((1:ℝ)/(n:ℝ))) Filter.atTop (nhds a) := by have h := (hlim c₂ hc2_pos).const_mul a simpa using h refine tendsto_of_tendsto_of_tendsto_of_le_of_le' hlimlo hlimhi ?_ ?_ · filter_upwards [Filter.eventually_ge_atTop 3] with n hn exact (step n hn).1 · filter_upwards [Filter.eventually_ge_atTop 3] with n hn exact (step n hn).2