Chapter 02 — Sequences and Series
Proofs omitted. Hover over highlighted terms to see their Lean types.
Reading sections
2.1 — Sequences and Limits
Lean authors: Zichen Wang, Zaiwen Wen.
Definition 2.1.1: A sequence of real numbers is a function x : ℕ → ℝ, whose n-th term is denoted x n (often written xₙ), and also denoted {x_n}_{n=1}^{\infty}. The sequence is bounded if there exists B : ℝ with |x n| ≤ B for all n. It is bounded below if there is B with B ≤ x n for all n, and bounded above if there is B with x n ≤ B for all n.
open scoped BigOperators
open Metric
abbrev RealSequence : Type :=
ℕ → ℝ
A real sequence admitting a uniform bound on absolute values.
open scoped BigOperators
open Metric
def BoundedSequence (x : RealSequence) : Prop :=
∃ B : ℝ, ∀ n : ℕ, |x n| ≤ B
A real sequence with a uniform lower bound.
open scoped BigOperators
open Metric
def BoundedBelowSequence (x : RealSequence) : Prop :=
∃ B : ℝ, ∀ n : ℕ, B ≤ x n
A real sequence with a uniform upper bound.
open scoped BigOperators
open Metric
def BoundedAboveSequence (x : RealSequence) : Prop :=
∃ B : ℝ, ∀ n : ℕ, x n ≤ B
Closed balls at the origin are described by absolute values.
open scoped BigOperators
open Metric
lemma real_mem_closedBall_0_iff_abs_le (x B : ℝ) :
x ∈ closedBall (0 : ℝ) B ↔ |x| ≤ B
Range inclusion in a closed ball is equivalent to a uniform absolute bound.
open scoped BigOperators
open Metric
lemma range_subset_closedBall_iff_abs_bound (x : ℕ → ℝ) (B : ℝ) :
Set.range x ⊆ closedBall (0 : ℝ) B ↔ ∀ n, |x n| ≤ B
A bounded real sequence in the book's sense is equivalent to the image of the sequence being bounded as a subset of ℝ.
open scoped BigOperators
open Metric
lemma boundedSequence_iff_isBounded_range (x : RealSequence) :
BoundedSequence x ↔ Bornology.IsBounded (Set.range x)
A sequence is bounded above exactly when its range is bounded above as a subset of ℝ.
open scoped BigOperators
open Metric
lemma boundedAboveSequence_iff_bddAbove_range (x : RealSequence) :
BoundedAboveSequence x ↔ BddAbove (Set.range x)
A sequence is bounded below exactly when its range is bounded below as a subset of ℝ.
open scoped BigOperators
open Metric
lemma boundedBelowSequence_iff_bddBelow_range (x : RealSequence) :
BoundedBelowSequence x ↔ BddBelow (Set.range x)
Definition 2.1.2: A real sequence x converges to l if for every ε > 0 there is an M : ℕ such that |x n - l| < ε whenever n ≥ M. Such an l is called a limit of the sequence, and a sequence with some limit is called convergent (otherwise divergent).
open scoped BigOperators
open Metric
def ConvergesTo (x : RealSequence) (l : ℝ) : Prop :=
∀ ε > 0, ∃ M : ℕ, ∀ n ≥ M, |x n - l| < ε
A real sequence is convergent if it admits a (necessarily unique) limit.
open scoped BigOperators
open Metric
def ConvergentSequence (x : RealSequence) : Prop :=
∃ l : ℝ, ConvergesTo x l
The book's ε–M definition of convergence agrees with the standard Filter.Tendsto formulation in mathlib.
open scoped BigOperators
open Metric
lemma convergesTo_iff_tendsto (x : RealSequence) (l : ℝ) :
ConvergesTo x l ↔ Filter.Tendsto x Filter.atTop (nhds l)
A constant real sequence converges to its constant value.
open scoped BigOperators
open Metric
lemma constant_seq_converges (a : ℝ) :
ConvergesTo (fun _ : ℕ => a) a
Example 2.1.3: The constant sequence 1, 1, 1, … converges to 1; for any ε > 0 one may take M = 1, giving |x n - x| = |1 - 1| < ε for every n.
open scoped BigOperators
open Metric
lemma constant_one_converges : ConvergesTo (fun _ : ℕ => (1 : ℝ)) 1
Example 2.1.4: The sequence x n = 1 / n (starting at n = 1) is convergent with limit 0; the proof picks M with 1 / M < ε by the Archimedean property and then estimates |x n - 0| ≤ 1 / M for all n ≥ M.
open scoped BigOperators
open Metric
lemma tendsto_inv_nat_succ :
Filter.Tendsto (fun n : ℕ => (1 : ℝ) / (n + 1)) Filter.atTop (nhds 0)
open scoped BigOperators
open Metric
lemma inv_nat_converges_to_zero :
ConvergesTo (fun n : ℕ => (1 : ℝ) / (n + 1)) 0
Example 2.1.5: The alternating sign sequence {(-1)^n} is divergent. If it had a limit x, then taking ε = 1/2 and comparing an even term and the next odd term would force both |1 - x| < 1/2 and |-1 - x| < 1/2, which is impossible.
open scoped BigOperators
open Metric
lemma alternating_signs_diverges :
¬ ConvergentSequence (fun n : ℕ => (-1 : ℝ) ^ n)
Proposition 2.1.6: A convergent sequence has a unique limit. If a real sequence converges to both l and l', then l = l'.
open scoped BigOperators
open Metric
lemma convergesTo_unique {x : RealSequence} {l l' : ℝ}
(h₁ : ConvergesTo x l) (h₂ : ConvergesTo x l') : l = l'
Proposition 2.1.7: Every convergent real sequence is bounded.
open scoped BigOperators
open Metric
lemma convergentSequence_bounded {x : RealSequence}
(hx : ConvergentSequence x) : BoundedSequence x
open scoped BigOperators
open Metric
lemma quadratic_fraction_abs_sub_eq (n : ℕ) :
open scoped BigOperators
open Metric
lemma quadratic_fraction_abs_le_inv (n : ℕ) :
Example 2.1.8: The sequence ((n^2 + 1) / (n^2 + n)) (starting at n = 1) converges to 1; choosing M with 1 / M < ε yields | (n^2 + 1) / (n^2 + n) - 1 | < ε for all n ≥ M.
open scoped BigOperators
open Metric
lemma quadratic_fraction_converges_to_one :
ConvergesTo
(fun n : ℕ =>
((n + 1 : ℝ) ^ 2 + 1) / ((n + 1 : ℝ) ^ 2 + (n + 1 : ℝ))) 1
Definition 2.1.9: A sequence {x_n} is monotone increasing if x n ≤ x (n + 1) for every n, monotone decreasing if x n ≥ x (n + 1) for every n, and monotone if it is either monotone increasing or monotone decreasing.
open scoped BigOperators
open Metric
def MonotoneIncreasingSequence (x : RealSequence) : Prop :=
∀ n : ℕ, x n ≤ x (n + 1)
A real sequence with nonincreasing successive terms.
open scoped BigOperators
open Metric
def MonotoneDecreasingSequence (x : RealSequence) : Prop :=
∀ n : ℕ, x n ≥ x (n + 1)
A real sequence that is either increasing or decreasing.
open scoped BigOperators
open Metric
def MonotoneSequence (x : RealSequence) : Prop :=
MonotoneIncreasingSequence x ∨ MonotoneDecreasingSequence x
A bounded monotone increasing real sequence converges to the supremum of its range.
open scoped BigOperators
open Metric
theorem monotoneIncreasingSequence_tendsto_sup {x : RealSequence}
(hx : MonotoneIncreasingSequence x) (hB : BoundedSequence x) :
ConvergesTo x (sSup (Set.range x))
A bounded monotone decreasing real sequence converges to the infimum of its range.
open scoped BigOperators
open Metric
theorem monotoneDecreasingSequence_tendsto_inf {x : RealSequence}
(hx : MonotoneDecreasingSequence x) (hB : BoundedSequence x) :
ConvergesTo x (sInf (Set.range x))
Theorem 2.1.10 (monotone convergence theorem): A monotone sequence {x_n} is bounded if and only if it is convergent. If {x_n} is monotone increasing and bounded, then its limit equals sup {x_n : n ∈ ℕ}; if monotone decreasing and bounded, then its limit equals inf {x_n : n ∈ ℕ}.
open scoped BigOperators
open Metric
theorem monotone_sequence_bounded_iff_convergent {x : RealSequence}
(hx : MonotoneSequence x) :
BoundedSequence x ↔ ConvergentSequence x
The book's notion of a monotone increasing sequence agrees with Monotone on ℕ.
open scoped BigOperators
open Metric
lemma monotoneIncreasingSequence_iff_monotone (x : RealSequence) :
MonotoneIncreasingSequence x ↔ Monotone x
The book's notion of a monotone decreasing sequence agrees with Antitone on ℕ.
open scoped BigOperators
open Metric
lemma monotoneDecreasingSequence_iff_antitone (x : RealSequence) :
MonotoneDecreasingSequence x ↔ Antitone x
Example 2.1.11: The sequence x n = 1 / √(n + 1) is bounded below by 0, monotone decreasing since √(n + 2) ≥ √(n + 1), and hence bounded. By the monotone convergence theorem its limit equals its infimum, which is 0, so lim_{n → ∞} 1 / √(n + 1) = 0.
open scoped BigOperators
open Metric
lemma inv_sqrt_sequence_converges :
MonotoneDecreasingSequence (fun n : ℕ => (1 : ℝ) / Real.sqrt (n + 1)) ∧
BoundedBelowSequence (fun n : ℕ => (1 : ℝ) / Real.sqrt (n + 1)) ∧
ConvergesTo (fun n : ℕ => (1 : ℝ) / Real.sqrt (n + 1)) 0
Example 2.1.12: The harmonic partial sums 1 + 1/2 + … + 1/n form a monotone increasing sequence. Although the growth is slow, the sequence is not bounded above and therefore does not converge; this unboundedness will be proved later using series.
open scoped BigOperators
open Metric
lemma harmonic_partial_sums_unbounded :
MonotoneIncreasingSequence
(fun n : ℕ =>
Finset.sum (Finset.range (n + 1)) (fun k : ℕ => (1 : ℝ) / (k + 1))) ∧
¬ BoundedAboveSequence
(fun n : ℕ =>
Finset.sum (Finset.range (n + 1)) (fun k : ℕ => (1 : ℝ) / (k + 1)))
Proposition 2.1.13: For a nonempty bounded subset S ⊆ ℝ there are monotone sequences {x_n} and {y_n} with terms in S such that lim_{n → ∞} x_n = sup S and lim_{n → ∞} y_n = inf S.
open scoped BigOperators
open Metric
lemma exists_monotone_sequences_tendsto_sup_inf {S : Set ℝ}
(hS : S.Nonempty) (hbounded : Bornology.IsBounded S) :
∃ x y : RealSequence,
(∀ n : ℕ, x n ∈ S) ∧ (∀ n : ℕ, y n ∈ S) ∧
MonotoneIncreasingSequence x ∧ MonotoneDecreasingSequence y ∧
ConvergesTo x (sSup S) ∧ ConvergesTo y (sInf S)
Definition 2.1.14: For a real sequence x and K ∈ ℕ, the K-tail (tail) is the sequence starting at the (K+1)-st term, written {x_{n+K}}_{n=1}^{∞} or {x_n}_{n=K+1}^{∞}.
open scoped BigOperators
open Metric
def sequenceTail (x : RealSequence) (K : ℕ) : RealSequence :=
fun n : ℕ => x (n + (K + 1))
A convergent sequence remains convergent after removing finitely many initial terms; the limit is unchanged.
open scoped BigOperators
open Metric
lemma convergesTo_tail_of_converges {x : RealSequence} {l : ℝ} (K : ℕ)
(hx : ConvergesTo x l) : ConvergesTo (sequenceTail x K) l
If a tail of a sequence converges, then so does the whole sequence, with the same limit.
open scoped BigOperators
open Metric
lemma convergesTo_of_tail_converges {x : RealSequence} {l : ℝ} (K : ℕ)
(hK : ConvergesTo (sequenceTail x K) l) : ConvergesTo x l
Proposition 2.1.15: For a real sequence {x_n}, the following are equivalent: (i) the sequence converges; (ii) every K-tail {x_{n+K}}_{n=1}^{∞} converges for all K ∈ ℕ; (iii) some K-tail converges. Moreover, whenever these limits exist, they agree for all K, so lim_{n → ∞} x_n = lim_{n → ∞} x_{n+K}.
open scoped BigOperators
open Metric
lemma convergentSequence_iff_all_tails_converge (x : RealSequence) :
ConvergentSequence x ↔
∀ K : ℕ, ConvergentSequence (sequenceTail x K)
If some tail of a real sequence converges, then the entire sequence converges, and conversely a convergent sequence has a convergent tail.
open scoped BigOperators
open Metric
lemma convergentSequence_iff_exists_tail_converges (x : RealSequence) :
ConvergentSequence x ↔ ∃ K : ℕ, ConvergentSequence (sequenceTail x K)
Any limit of a sequence agrees with the limit of each of its tails.
open scoped BigOperators
open Metric
lemma convergesTo_tail_limit_eq {x : RealSequence} {l l' : ℝ} (K : ℕ)
(hx : ConvergesTo x l) (hK : ConvergesTo (sequenceTail x K) l') :
l = l'
Definition 2.1.16: Given a real sequence {x_n} and a strictly increasing sequence of indices n₁ < n₂ < n₃ < …, the extracted sequence i ↦ x (n i) is called a subsequence of {x_n}.
open scoped BigOperators
open Metric
structure RealSubsequence (x : RealSequence) where
/-- the strictly increasing indexing function `n_i` selecting terms of the
original sequence -/
indices : ℕ → ℕ
strictlyIncreasing : StrictMono indices
The indices of a subsequence eventually dominate their position: i ≤ n_i for all i.
open scoped BigOperators
open Metric
lemma subsequence_indices_ge_self {x : RealSequence} (s : RealSubsequence x) :
∀ i, i ≤ s.indices i
The subsequence of x determined by the index data s.
open scoped BigOperators
open Metric
def RealSubsequence.asSequence {x : RealSequence} (s : RealSubsequence x) :
RealSequence :=
fun i : ℕ => x (s.indices i)
A real sequence y is a subsequence of x if it arises by composing x with a strictly increasing index map.
open scoped BigOperators
open Metric
def IsRealSubsequence (x y : RealSequence) : Prop :=
∃ s : RealSubsequence x, y = s.asSequence
Proposition 2.1.17: If a real sequence {x_n} converges, then every strictly increasing subsequence {x_{n_i}} also converges, with the same limit as the original sequence: lim_{n → ∞} x_n = lim_{i → ∞} x_{n_i}.
open scoped BigOperators
open Metric
lemma convergesTo_subsequence {x : RealSequence} {l : ℝ}
(hx : ConvergesTo x l) (s : RealSubsequence x) :
ConvergesTo (s.asSequence) l
Example 2.1.18: A sequence can have convergent subsequences without being convergent itself. The alternating sequence 0, 1, 0, 1, … is defined by x n = 0 for odd n and x n = 1 for even n; its even-indexed subsequence converges to 1, its odd-indexed subsequence converges to 0, but the full sequence is divergent. Compare Proposition 2.1.15.
open scoped BigOperators
open Metric
def alternatingZeroOneSequence : RealSequence :=
fun n : ℕ => if n % 2 = 0 then (1 : ℝ) else 0
The alternating zero–one sequence diverges, even though its even subsequence converges to 1 and its odd subsequence converges to 0.
open scoped BigOperators
open Metric
lemma alternatingZeroOne_subsequences_converge :
¬ ConvergentSequence alternatingZeroOneSequence ∧
ConvergesTo (fun i : ℕ => alternatingZeroOneSequence (2 * i)) 1 ∧
ConvergesTo (fun i : ℕ => alternatingZeroOneSequence (2 * i + 1)) 02.2 — Facts About Limits of Sequences
Lean authors: Zichen Wang, Zaiwen Wen.
Lemma 2.2.1 (Squeeze lemma): If real sequences a, x, b satisfy a n ≤ x n ≤ b n for every n, and both a and b converge to the same limit l, then x also converges to l, so lim_{n → ∞} x_n = lim_{n → ∞} a_n = lim_{n → ∞} b_n.
lemma squeeze_converges {a b x : RealSequence} {l : ℝ}
(hax : ∀ n : ℕ, a n ≤ x n) (hxb : ∀ n : ℕ, x n ≤ b n)
(ha : ConvergesTo a l) (hb : ConvergesTo b l) :
ConvergesTo x l
A sequence squeezed between two convergent sequences with common limit is itself convergent with that limit.
lemma squeeze_convergentSequence {a b x : RealSequence} {l : ℝ}
(hax : ∀ n : ℕ, a n ≤ x n) (hxb : ∀ n : ℕ, x n ≤ b n)
(ha : ConvergesTo a l) (hb : ConvergesTo b l) :
ConvergentSequence x
Example 2.2.2: By bounding 1 / (n √n) (starting at n = 1) between the constant 0 sequence and 1 / n, and using the squeeze lemma together with lim_{n → ∞} 1 / n = 0, we conclude lim_{n → ∞} 1 / (n √n) = 0.
lemma inv_mul_sqrt_bounds :
∀ n : ℕ,
0 ≤ (1 : ℝ) / ((n + 1 : ℝ) * Real.sqrt (n + 1)) ∧
(1 : ℝ) / ((n + 1 : ℝ) * Real.sqrt (n + 1)) ≤ (1 : ℝ) / (n + 1)
lemma inv_mul_sqrt_converges_to_zero :
ConvergesTo
(fun n : ℕ =>
(1 : ℝ) / ((n + 1 : ℝ) * Real.sqrt (n + 1))) 0
Lemma 2.2.3: If convergent real sequences x and y satisfy x n ≤ y n for every n, then their limits obey lim_{n → ∞} x_n ≤ lim_{n → ∞} y_n.
lemma limit_le_of_pointwise_le {x y : RealSequence} {l m : ℝ}
(hx : ConvergesTo x l) (hy : ConvergesTo y m)
(hxy : ∀ n : ℕ, x n ≤ y n) :
l ≤ m
A convergent sequence with pointwise nonnegative terms has a nonnegative limit.
lemma limit_nonneg_of_pointwise_nonneg {x : RealSequence} {l : ℝ}
(hx : ConvergesTo x l) (h_nonneg : ∀ n : ℕ, 0 ≤ x n) : 0 ≤ l
A pointwise lower bound for a convergent sequence bounds its limit below.
lemma limit_lower_bound_of_pointwise {x : RealSequence} {a l : ℝ}
(hx : ConvergesTo x l) (hax : ∀ n : ℕ, a ≤ x n) : a ≤ l
A pointwise upper bound for a convergent sequence bounds its limit above.
lemma limit_upper_bound_of_pointwise {x : RealSequence} {b l : ℝ}
(hx : ConvergesTo x l) (hxb : ∀ n : ℕ, x n ≤ b) : l ≤ b
Corollary 2.2.4: (i) If a convergent real sequence has nonnegative terms, then its limit is nonnegative. (ii) If a convergent real sequence is bounded between real numbers a and b, then its limit also lies between a and b.
lemma limit_nonneg_and_between_bounds :
(∀ {x : RealSequence} {l : ℝ},
ConvergesTo x l → (∀ n : ℕ, 0 ≤ x n) → 0 ≤ l) ∧
(∀ {x : RealSequence} {a b l : ℝ},
ConvergesTo x l → (∀ n : ℕ, a ≤ x n) → (∀ n : ℕ, x n ≤ b) →
a ≤ l ∧ l ≤ b)
Proposition 2.2.5: For convergent real sequences x and y, the usual algebraic operations preserve convergence. (i) The sum sequence converges to the sum of the limits. (ii) The difference sequence converges to the difference of the limits. (iii) The termwise product converges to the product of the limits. (iv) If lim_{n → ∞} y_n ≠ 0 and every y n ≠ 0, then the quotient sequence converges to the quotient of the limits.
lemma limit_add_of_convergent {x y : RealSequence} {l m : ℝ}
(hx : ConvergesTo x l) (hy : ConvergesTo y m) :
ConvergesTo (fun n : ℕ => x n + y n) (l + m)
The limit of the difference of two convergent real sequences is the difference of their limits.
lemma limit_sub_of_convergent {x y : RealSequence} {l m : ℝ}
(hx : ConvergesTo x l) (hy : ConvergesTo y m) :
ConvergesTo (fun n : ℕ => x n - y n) (l - m)
The limit of the termwise product of two convergent real sequences is the product of their limits.
lemma limit_mul_of_convergent {x y : RealSequence} {l m : ℝ}
(hx : ConvergesTo x l) (hy : ConvergesTo y m) :
ConvergesTo (fun n : ℕ => x n * y n) (l * m)
If a convergent real sequence y has nonzero limit and no zero terms, then the termwise quotient of convergent sequences x and y converges to the quotient of their limits.
lemma limit_div_of_convergent {x y : RealSequence} {l m : ℝ}
(hx : ConvergesTo x l) (hy : ConvergesTo y m) (hm : m ≠ 0)
(hy_ne_zero : ∀ n : ℕ, y n ≠ 0) :
ConvergesTo (fun n : ℕ => x n / y n) (l / m)
Proposition 2.2.6: If a convergent real sequence satisfies x n ≥ 0 for every n, then its termwise square roots converge to the square root of its limit, i.e. lim_{n → ∞} √(x_n) = √(lim_{n → ∞} x_n).
lemma limit_sqrt_of_nonneg {x : RealSequence} {l : ℝ}
(hx : ConvergesTo x l) (h_nonneg : ∀ n : ℕ, 0 ≤ x n) :
ConvergesTo (fun n : ℕ => Real.sqrt (x n)) (Real.sqrt l)
Proposition 2.2.7: If a real sequence x converges, then its termwise absolute values also converge, and the limit is the absolute value of the original limit, so lim_{n → ∞} |x_n| = |lim_{n → ∞} x_n|.
lemma limit_abs_of_convergent {x : RealSequence} {l : ℝ}
(hx : ConvergesTo x l) :
ConvergesTo (fun n : ℕ => |x n|) |l|
The absolute value of a convergent real sequence is also convergent.
lemma abs_convergentSequence {x : RealSequence}
(hx : ConvergentSequence x) :
ConvergentSequence (fun n : ℕ => |x n|)
Example 2.2.8: Starting from x₁ = 2 and defining x_{n+1} = x_n - (x_n^2 - 2) / (2 x_n), the recursion is well-defined because each term stays positive. The sequence is monotone decreasing and bounded below, so it converges; taking limits in the recurrence gives x = √2, so the limit equals √2.
noncomputable def sqrtTwoNewtonSequence : RealSequence :=
fun n =>
Nat.recOn n (2 : ℝ)
(fun _ x_n => x_n - (x_n ^ 2 - 2) / (2 * x_n))
Every term of the Newton sequence for √2 is positive, so the recurrence is well-defined.
lemma sqrtTwoNewtonSequence_pos : ∀ n : ℕ, 0 < sqrtTwoNewtonSequence n
For every n, (x_n)^2 - 2 is nonnegative along the Newton sequence.
lemma sqrtTwoNewtonSequence_sq_sub_two_nonneg :
∀ n : ℕ, 0 ≤ (sqrtTwoNewtonSequence n) ^ 2 - 2
The Newton sequence for √2 is monotone decreasing.
lemma sqrtTwoNewtonSequence_monotone :
MonotoneDecreasingSequence sqrtTwoNewtonSequence
The Newton sequence for √2 converges to √2.
lemma sqrtTwoNewtonSequence_tendsto :
ConvergesTo sqrtTwoNewtonSequence (Real.sqrt 2)
The Newton sequence for √2 stays positive, decreases monotonically, and converges to √2.
lemma sqrtTwoNewtonSequence_converges :
(∀ n : ℕ, 0 < sqrtTwoNewtonSequence n) ∧
MonotoneDecreasingSequence sqrtTwoNewtonSequence ∧
ConvergesTo sqrtTwoNewtonSequence (Real.sqrt 2)
Example 2.2.9: For the recursion x_{n+1} = x_n^2 + x_n, starting at x₁ = 1 yields an unbounded sequence and hence no limit, while starting at x₁ = 0 gives the constant zero sequence whose limit is 0. The long-term behavior therefore depends on the initial value.
def quadraticRecursiveSequence (x₁ : ℝ) : RealSequence :=
fun n => Nat.recOn n x₁ (fun _ x_n => x_n ^ 2 + x_n)
Lower bound for the quadratic recursion started at 1.
lemma quadraticRecursiveSequence_one_lower_bound :
∀ n : ℕ, quadraticRecursiveSequence 1 n ≥ (n : ℝ) + 1
With initial value 1, the quadratic recursion grows without bound.
lemma quadraticRecursiveSequence_unbounded :
¬ BoundedSequence (quadraticRecursiveSequence 1)
The quadratic recursion started at 1 does not converge.
lemma quadraticRecursiveSequence_diverges :
¬ ConvergentSequence (quadraticRecursiveSequence 1)
The quadratic recursion started at 0 is identically zero.
lemma quadraticRecursiveSequence_zero_eq_zero :
∀ n : ℕ, quadraticRecursiveSequence 0 n = 0
Starting the quadratic recursion at 0 yields a sequence converging to 0.
lemma quadraticRecursiveSequence_zero_converges :
ConvergesTo (quadraticRecursiveSequence 0) 0
Proposition 2.2.10: If x is a real sequence and there exists x₀ : ℝ and a convergent sequence a with lim_{n → ∞} a_n = 0 such that |x_n - x₀| ≤ a_n for every n, then x converges to x₀.
lemma converges_of_abs_sub_le_converging_to_zero {x a : RealSequence} {x₀ : ℝ}
(ha : ConvergesTo a 0) (hbound : ∀ n : ℕ, |x n - x₀| ≤ a n) :
ConvergesTo x x₀
A sequence controlled by a convergent error term with vanishing limit is convergent.
lemma convergentSequence_of_abs_sub_le_converging_to_zero {x a : RealSequence}
{x₀ : ℝ} (ha : ConvergesTo a 0) (hbound : ∀ n : ℕ, |x n - x₀| ≤ a n) :
ConvergentSequence x
Proposition 2.2.11: Let c > 0. (i) If c < 1, then lim_{n → ∞} c^n = 0. (ii) If c > 1, then the sequence (c^n) is unbounded.
lemma pow_sequence_converges_or_unbounded {c : ℝ} (hc_pos : 0 < c) :
(c < 1 → ConvergesTo (fun n : ℕ => c ^ n) 0) ∧
(1 < c → ¬ BoundedSequence (fun n : ℕ => c ^ n))
Lemma 2.2.12 (Ratio test for sequences): For a real sequence x with no zero terms, suppose the limit L = lim_{n → ∞} |x_{n+1}| / |x_n| exists. (i) If L < 1, then x converges to 0. (ii) If L > 1, then x is unbounded and therefore diverges.
lemma ratio_test_for_sequences {x : RealSequence} {L : ℝ}
(hx_ne : ∀ n : ℕ, x n ≠ 0)
(hL : ConvergesTo (fun n : ℕ => |x (n + 1)| / |x n|) L) :
(L < 1 → ConvergesTo x 0) ∧ (1 < L → ¬ BoundedSequence x)
Example 2.2.13: Applying the ratio test with |x_{n+1}| / |x_n| = 2 / (n + 1), we obtain lim_{n → ∞} (2^n / n!) = 0.
lemma pow_two_over_factorial_nonzero :
∀ n : ℕ, (2 : ℝ) ^ n / (Nat.factorial n) ≠ 0
lemma pow_two_over_factorial_ratio_limit :
ConvergesTo
(fun n : ℕ =>
|(2 : ℝ) ^ (n + 1) / (Nat.factorial (n + 1))| /
|(2 : ℝ) ^ n / (Nat.factorial n)|) 0
lemma pow_two_over_factorial_converges_to_zero :
ConvergesTo (fun n : ℕ => (2 : ℝ) ^ n / (Nat.factorial n)) 0
Example 2.2.14: Using the ratio test on the sequence (n / (1 + ε)^n), for any ε > 0 we get n / (1 + ε)^n → 0, which forces 1 ≤ n^(1/n) < 1 + ε for large n. Hence lim_{n → ∞} n^(1/n) = 1.
lemma nth_root_sequence_converges_to_one :
ConvergesTo (fun n : ℕ => ((n + 1 : ℝ)).rpow (1 / (n + 1 : ℝ))) 12.3 — Limit Superior, Limit Inferior, and Bolzano-Weierstrass
Lean authors: Zichen Wang, Zaiwen Wen.
The upper tail supremum sequence a_n = sup { x_k | k ≥ n } attached to a real sequence x.
noncomputable def tailSupSequence (x : RealSequence) (n : ℕ) : ℝ :=
sSup {y : ℝ | ∃ k ≥ n, y = x k}
The lower tail infimum sequence b_n = inf { x_k | k ≥ n } attached to a real sequence x.
noncomputable def tailInfSequence (x : RealSequence) (n : ℕ) : ℝ :=
sInf {y : ℝ | ∃ k ≥ n, y = x k}
The limit superior of a real sequence, realized as the infimum of the tail suprema.
noncomputable def limsupSequence (x : RealSequence) : ℝ :=
sInf (Set.range (tailSupSequence x))
The limit inferior of a real sequence, realized as the supremum of the tail infima.
noncomputable def liminfSequence (x : RealSequence) : ℝ :=
sSup (Set.range (tailInfSequence x))
Proposition 2.3.2: For a bounded sequence {x_n}, with a_n the tail suprema and b_n the tail infima as above: (i) {a_n} is bounded and monotone decreasing while {b_n} is bounded and monotone increasing, so liminf_{n → ∞} x_n and limsup_{n → ∞} x_n exist; (ii) limsup_{n → ∞} x_n = inf { a_n : n ∈ ℕ } and liminf_{n → ∞} x_n = sup { b_n : n ∈ ℕ }; (iii) liminf_{n → ∞} x_n ≤ limsup_{n → ∞} x_n.
theorem limsup_liminf_properties (x : RealSequence) (hx : BoundedSequence x) :
MonotoneDecreasingSequence (tailSupSequence x) ∧
MonotoneIncreasingSequence (tailInfSequence x) ∧
ConvergentSequence (tailSupSequence x) ∧
ConvergentSequence (tailInfSequence x) ∧
limsupSequence x = sInf (Set.range (tailSupSequence x)) ∧
liminfSequence x = sSup (Set.range (tailInfSequence x)) ∧
liminfSequence x ≤ limsupSequence x
Definition 2.3.1: For a bounded real sequence {x_n}, define the tail suprema a_n = sup { x_k | k ≥ n } and tail infima b_n = inf { x_k | k ≥ n }. When the limits of these tails exist, set limsup_{n → ∞} x_n = lim_{n → ∞} a_n and liminf_{n → ∞} x_n = lim_{n → ∞} b_n.
lemma limsup_liminf_def_via_tail_extrema {x : RealSequence} (hx : BoundedSequence x)
{L_sup L_inf : ℝ}
(h_sup : ConvergesTo (tailSupSequence x) L_sup)
(h_inf : ConvergesTo (tailInfSequence x) L_inf) :
limsupSequence x = L_sup ∧ liminfSequence x = L_inf
Example 2.3.3: For the sequence defined by xₙ = (n + 1) / n when n is odd and xₙ = 0 when n is even, the limit inferior is 0, the limit superior is 1, and the sequence does not converge.
noncomputable def limsupLiminfExampleSequence : RealSequence :=
fun n => if Odd n.succ then ((n.succ + 1 : ℝ) / n.succ) else 0
The terms of the sequence in Example 2.3.3 are nonnegative and bounded by 2.
lemma limsupLiminfExampleSequence_bounds :
∀ n, 0 ≤ limsupLiminfExampleSequence n ∧
limsupLiminfExampleSequence n ≤ 2
Every tail infimum of the sequence in Example 2.3.3 equals 0.
lemma limsupLiminfExample_tailInf_zero (n : ℕ) :
tailInfSequence limsupLiminfExampleSequence n = 0
The tail supremum of the sequence in Example 2.3.3 admits a closed formula matching the textbook description.
lemma limsupLiminfExample_tailSup_formula (n : ℕ) :
tailSupSequence limsupLiminfExampleSequence n =
if Odd n.succ then ((n.succ + 1 : ℝ) / n.succ)
else ((n.succ + 2 : ℝ) / (n.succ + 1))
The tail supremum sequence in Example 2.3.3 converges to 1.
lemma limsupLiminfExample_tailSup_tendsto_one :
ConvergesTo (tailSupSequence limsupLiminfExampleSequence) 1
Example 2.3.3, continued: the sequence above has liminf = 0, limsup = 1 and is divergent.
lemma limsupLiminfExample_values :
liminfSequence limsupLiminfExampleSequence = 0 ∧
limsupSequence limsupLiminfExampleSequence = 1 ∧
¬ ConvergentSequence limsupLiminfExampleSequence
Theorem 2.3.4: For a bounded sequence {x_n}, there is a subsequence whose limit equals limsup_{n → ∞} x_n, and (possibly another) subsequence whose limit equals liminf_{n → ∞} x_n.
theorem subsequence_converging_to_limsup_liminf {x : RealSequence}
(hx : BoundedSequence x) :
(∃ s : RealSubsequence x, ConvergesTo (s.asSequence) (limsupSequence x)) ∧
∃ s : RealSubsequence x, ConvergesTo (s.asSequence) (liminfSequence x)
Proposition 2.3.5: A bounded sequence {x_n} converges if and only if its limit inferior and limit superior coincide. Moreover, when {x_n} converges to L, this common value equals L.
lemma convergent_iff_liminf_eq_limsup {x : RealSequence} (hx : BoundedSequence x) :
ConvergentSequence x ↔ liminfSequence x = limsupSequence x
See Proposition 2.3.5: for a bounded sequence that converges to L, the limit equals both the limit inferior and the limit superior.
lemma convergent_limit_eq_limsup_liminf {x : RealSequence} {L : ℝ}
(hx : BoundedSequence x) (hconv : ConvergesTo x L) :
liminfSequence x = L ∧ limsupSequence x = L
Proposition 2.3.6: For a bounded sequence {x_n} and any subsequence {x_{n_k}}, the limit inferior of the original sequence bounds below the subsequence's limit inferior, which in turn does not exceed its limit superior, and this limit superior is bounded above by the limit superior of the original sequence.
lemma limsup_liminf_subsequence_bounds {x : RealSequence} (hx : BoundedSequence x)
(s : RealSubsequence x) :
liminfSequence x ≤ liminfSequence (s.asSequence) ∧
liminfSequence (s.asSequence) ≤ limsupSequence (s.asSequence) ∧
limsupSequence (s.asSequence) ≤ limsupSequence x
Theorem 2.3.8 (Bolzano--Weierstrass): Every bounded real sequence {x_n} admits a convergent subsequence {x_{n_i}}.
theorem bolzanoWeierstrass_real_sequence {x : RealSequence}
(hx : BoundedSequence x) :
∃ s : RealSubsequence x, ConvergentSequence (s.asSequence)
Proposition 2.3.7: A bounded sequence {x_n} converges to x if and only if every convergent subsequence {x_{n_k}} converges to the same limit x.
lemma converges_iff_convergent_subsequences_converge
{x : RealSequence} {l : ℝ} (hx : BoundedSequence x) :
ConvergesTo x l ↔
∀ s : RealSubsequence x,
ConvergentSequence (s.asSequence) → ConvergesTo (s.asSequence) l
Definition 2.3.9: A real sequence {x_n} diverges to +∞ if for every K : ℝ there exists M : ℕ such that x n > K for all n ≥ M; we then write lim_{n → ∞} x_n = ∞. It diverges to -∞ if for every K : ℝ there is M : ℕ with x n < K for all n ≥ M, written lim_{n → ∞} x_n = -∞.
def DivergesToInfinitySequence (x : RealSequence) : Prop :=
∀ K : ℝ, ∃ M : ℕ, ∀ n ≥ M, K < x n
A real sequence diverges to negative infinity when its terms eventually stay below every real threshold.
def DivergesToMinusInfinitySequence (x : RealSequence) : Prop :=
∀ K : ℝ, ∃ M : ℕ, ∀ n ≥ M, x n < K
Proposition 2.3.10: A monotone unbounded sequence diverges to an infinite limit. If {x_n} is monotone increasing and unbounded above, then lim_{n → ∞} x_n = ∞; if it is monotone decreasing and unbounded below, then lim_{n → ∞} x_n = -∞.
lemma monotone_unbounded_diverges_to_infinity {x : RealSequence}
(hx : MonotoneIncreasingSequence x) (h_unbounded : ¬ BoundedAboveSequence x) :
DivergesToInfinitySequence x
See Proposition 2.3.10: a monotone decreasing sequence without a lower bound diverges to -∞.
lemma monotone_unbounded_diverges_to_minus_infinity {x : RealSequence}
(hx : MonotoneDecreasingSequence x) (h_unbounded : ¬ BoundedBelowSequence x) :
DivergesToMinusInfinitySequence x
The book's notion of divergence to +∞ agrees with Filter.Tendsto to Filter.atTop.
lemma divergesToInfinity_iff_tendsto_atTop (x : RealSequence) :
DivergesToInfinitySequence x ↔
Filter.Tendsto x Filter.atTop Filter.atTop
The book's notion of divergence to -∞ agrees with Filter.Tendsto to Filter.atBot.
lemma divergesToMinusInfinity_iff_tendsto_atBot (x : RealSequence) :
DivergesToMinusInfinitySequence x ↔
Filter.Tendsto x Filter.atTop Filter.atBot
Example 2.3.11: the sequences n and n^2 diverge to +∞, while the sequence -n diverges to -∞.
lemma example_2311_limits :
DivergesToInfinitySequence (fun n : ℕ => (n : ℝ)) ∧
DivergesToInfinitySequence (fun n : ℕ => (n : ℝ) ^ 2) ∧
DivergesToMinusInfinitySequence (fun n : ℕ => - (n : ℝ))
Definition 2.3.12: For an unbounded real sequence {xₙ}, set aₙ = sup { x_k | k ≥ n } and bₙ = inf { x_k | k ≥ n }, where the suprema and infima are taken in the extended real numbers. The limit superior is inf { aₙ : n ∈ ℕ } and the limit inferior is sup { bₙ : n ∈ ℕ }.
noncomputable def tailSupSequenceEReal (x : RealSequence) (n : ℕ) : EReal :=
sSup {y : EReal | ∃ k ≥ n, y = (x k : EReal)}
See the preceding comment: the lower tail infimum in the extended reals.
noncomputable def tailInfSequenceEReal (x : RealSequence) (n : ℕ) : EReal :=
sInf {y : EReal | ∃ k ≥ n, y = (x k : EReal)}
The limit superior of a (possibly unbounded) real sequence, valued in ℝ∞.
noncomputable def limsupSequenceEReal (x : RealSequence) : EReal :=
sInf (Set.range (tailSupSequenceEReal x))
The limit inferior of a (possibly unbounded) real sequence, valued in ℝ∞.
noncomputable def liminfSequenceEReal (x : RealSequence) : EReal :=
sSup (Set.range (tailInfSequenceEReal x))
The extended-real limsup defined via tail suprema coincides with the Filter.limsup of the sequence viewed in ℝ∞.
lemma limsupSequenceEReal_eq_filter_limsup (x : RealSequence) :
limsupSequenceEReal x =
Filter.limsup (fun n => (x n : EReal)) Filter.atTop
The extended-real liminf defined via tail infima coincides with the Filter.liminf of the sequence viewed in ℝ∞.
lemma liminfSequenceEReal_eq_filter_liminf (x : RealSequence) :
liminfSequenceEReal x =
Filter.liminf (fun n => (x n : EReal)) Filter.atTop
Proposition 2.3.13: For an unbounded sequence {x_n} with tail suprema a_n = sup { x_k | k ≥ n } and tail infima b_n = inf { x_k | k ≥ n } formed in the extended reals, the sequence {a_n} is decreasing and {b_n} is increasing. If every a_n is a (finite) real number, then limsup_{n → ∞} x_n = lim_{n → ∞} a_n; if every b_n is real, then liminf_{n → ∞} x_n = lim_{n → ∞} b_n.
lemma limsup_liminf_unbounded_properties (x : RealSequence)
(_hx : ¬ BoundedSequence x) :
Antitone (tailSupSequenceEReal x) ∧
Monotone (tailInfSequenceEReal x) ∧
(∀ _hfinite : ∀ n, ∃ r : ℝ, tailSupSequenceEReal x n = (r : EReal),
∃ l : EReal,
Filter.Tendsto (fun n => tailSupSequenceEReal x n)
Filter.atTop (nhds l) ∧
limsupSequenceEReal x = l) ∧
(∀ _hfinite : ∀ n, ∃ r : ℝ, tailInfSequenceEReal x n = (r : EReal),
∃ l : EReal,
Filter.Tendsto (fun n => tailInfSequenceEReal x n)
Filter.atTop (nhds l) ∧
liminfSequenceEReal x = l)
Example 2.3.14: For xₙ = 0 when n is odd and xₙ = n when n is even, each tail supremum aₙ equals ∞ because even indices keep growing, while each tail infimum bₙ is 0 thanks to the odd indices. Consequently, limsup_{n → ∞} xₙ = ∞, liminf_{n → ∞} xₙ = 0, and the sequence is divergent.
def example2314Sequence : RealSequence := fun n => if Even n then (n : ℝ) else 0
lemma example2314_tailSup_tailInf :
(∀ n, tailSupSequenceEReal example2314Sequence n = (⊤ : EReal)) ∧
(∀ n, tailInfSequenceEReal example2314Sequence n = (0 : EReal))
lemma example2314_limsup_liminf :
limsupSequenceEReal example2314Sequence = (⊤ : EReal) ∧
liminfSequenceEReal example2314Sequence = (0 : EReal) ∧
¬ ConvergentSequence example2314Sequence2.4 — Cauchy Sequences
Lean authors: Zichen Wang, Zaiwen Wen.
Definition 2.4.1: A real sequence {xₙ} is called a *Cauchy sequence* if for every ε > 0 there exists M ∈ ℕ such that for all n, k ≥ M we have |x n - x k| < ε.
def CauchySequence (x : RealSequence) : Prop :=
∀ ε > 0, ∃ M : ℕ, ∀ ⦃n k : ℕ⦄, M ≤ n → M ≤ k → |x n - x k| < ε
The book's notion of a Cauchy sequence of real numbers coincides with the standard CauchySeq predicate from mathlib.
lemma cauchySequence_iff_cauchySeq (x : RealSequence) :
CauchySequence x ↔ CauchySeq x
Example 2.4.2: The sequence 1 / n (with indexing starting at n = 1) is Cauchy.
lemma example_242_one_div_nat_cauchy :
CauchySequence (fun n : ℕ => (1 : ℝ) / (n.succ))
Explicit values of consecutive even and odd powers of -1.
lemma neg_one_pow_even_odd (m : ℕ) :
(-1 : ℝ) ^ (2 * m) = 1 ∧ (-1 : ℝ) ^ (2 * m + 1) = -1
The absolute difference of consecutive powers of -1 is 2.
lemma neg_one_pow_consecutive_abs (m : ℕ) :
Example 2.4.3: The sequence { (-1)^n }_{n = 1}^∞ is not Cauchy.
lemma example_243_neg_one_pow_not_cauchy :
¬ CauchySequence (fun n : ℕ => (-1 : ℝ) ^ n.succ)
Proposition 2.4.4: Every Cauchy real sequence is bounded.
lemma cauchySequence_bounded {x : RealSequence} (hx : CauchySequence x) :
BoundedSequence x
A convergent real sequence is Cauchy (ε/2 argument).
lemma convergentSequence_cauchySequence {x : RealSequence} :
ConvergentSequence x → CauchySequence x
A Cauchy real sequence converges by completeness of ℝ.
lemma cauchySequence_convergentSequence {x : RealSequence} :
CauchySequence x → ConvergentSequence x
Theorem 2.4.5: A sequence of real numbers is Cauchy if and only if it converges (using completeness of ℝ).
theorem cauchySequence_iff_convergentSequence (x : RealSequence) :
CauchySequence x ↔ ConvergentSequence x2.5 — Series
Lean authors: Zichen Wang, Zaiwen Wen.
Definition 2.5.1: For a sequence x : ℕ → α, the expression ∑_{n=1}^{∞} x n denotes the formal series. Its k-th partial sum is s k = ∑_{n=1}^k x n. The series converges to ℓ when the partial sums s k converge to ℓ as k → ∞; otherwise the series is divergent. When the series converges we identify its value with this limit.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
def partialSumsFromOne [AddCommMonoid α] (x : ℕ → α) (k : ℕ) : α :=
(Finset.Icc 1 k).sum fun n => x n
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
lemma partialSumsFromOne_eq_sum_range_succ {β : Type*} [AddCommMonoid β]
(x : ℕ → β) (k : ℕ) :
partialSumsFromOne x k =
(Finset.range k).sum (fun n => x (n + 1))
Convergence of the series ∑_{n=1}^{∞} x n to a limit ℓ, expressed via the convergence of the partial sums ∑_{n=1}^k x n as k → ∞.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesConvergesTo (x : ℕ → α) (ℓ : α) : Prop :=
HasSum (fun n => x (n + 1)) ℓ
The series ∑_{n=1}^{∞} x n is convergent if its partial sums admit some limit.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesConvergent (x : ℕ → α) : Prop :=
Summable (fun n => x (n + 1))
A divergent series is one whose partial sums do not converge.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesDivergent (x : ℕ → α) : Prop :=
¬ SeriesConvergent x
Definition 2.5.7: A series ∑_{n=1}^{∞} x n is called Cauchy (a Cauchy series) when its sequence of partial sums ∑_{n=1}^k x n forms a Cauchy sequence.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesCauchy (x : ℕ → α) : Prop :=
CauchySeq fun k : ℕ => partialSumsFromOne x k
Bridge to mathlib's HasSum: using the shift starting at 1 our notion of converging series agrees with HasSum.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma seriesConvergesTo_iff_hasSum_shift (x : ℕ → α) (ℓ : α) :
SeriesConvergesTo x ℓ ↔ HasSum (fun n => x (n + 1)) ℓ
The book's notion of a convergent series agrees with mathlib's Summable for the shifted sequence starting at index 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma seriesConvergent_iff_summable (x : ℕ → α) :
SeriesConvergent x ↔ Summable (fun n => x (n + 1))
When the series ∑_{n=1}^{∞} x n converges to ℓ, the limit equals the tsum of the underlying sequence indexed from 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_limit_eq_tsum {x : ℕ → α} {ℓ : α} :
SeriesConvergesTo x ℓ ↔
Summable (fun n => x (n + 1)) ∧ tsum (fun n => x (n + 1)) = ℓ
Proposition 2.5.6: For a series ∑_{n=1}^{∞} x n and any M ∈ ℕ, convergence is unchanged by discarding the first M - 1 terms; equivalently ∑_{n=1}^{∞} x n converges if and only if the tail ∑_{n=M}^{∞} x n converges.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_iff_convergent_tail (x : ℕ → α) (M : ℕ) :
SeriesConvergent x ↔ SeriesConvergent (fun n => x (n + (M - 1)))
Proposition 2.5.5: For a real number r with -1 < r < 1, the geometric series ∑_{n=0}^{∞} r^n converges and its sum equals 1 / (1 - r).
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_series_sum {r : ℝ} (h₁ : -1 < r) (h₂ : r < 1) :
Summable (fun n : ℕ => r ^ n) ∧ tsum (fun n => r ^ n) = 1 / (1 - r)
A closed form for the partial sums of the geometric series with ratio 1/2.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_half_partial_sum_closed_form (k : ℕ) :
partialSumsFromOne (fun n : ℕ => (1 / 2 : ℝ) ^ n) k =
1 - (1 / 2 : ℝ) ^ k
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_half_partial_sum_closed_form_inv (k : ℕ) :
partialSumsFromOne (fun n : ℕ => ((2 : ℝ) ^ n)⁻¹) k =
1 - ((2 : ℝ) ^ k)⁻¹
Example 2.5.4: The geometric series ∑_{n=1}^{∞} (1/2)^n converges to 1. For each k ≥ 1, the partial sum satisfies ∑_{n=1}^k (1/2)^n + (1/2)^k = 1, so the sequence of partial sums tends to 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_half_partial_sum_eq (k : ℕ) (hk : 1 ≤ k) :
partialSumsFromOne (fun n : ℕ => (1 / 2 : ℝ) ^ n) k + (1 / 2 : ℝ) ^ k = 1
Example 2.5.4: Consequently, the series ∑_{n=1}^{∞} (1/2)^n converges and its sum equals 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_half_series_converges :
SeriesConvergesTo (fun n : ℕ => (1 / 2 : ℝ) ^ n) (1 : ℝ)
Split partial sums into an initial segment plus a tail over Icc (n+1) k.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma partialSumsFromOne_add_tail {β : Type*} [AddCommMonoid β]
(x : ℕ → β) {n k : ℕ} (hnk : n ≤ k) :
partialSumsFromOne x k =
partialSumsFromOne x n + (Finset.Icc (n + 1) k).sum (fun i => x i)
The difference of partial sums is the tail sum over Icc (n+1) k.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma partialSumsFromOne_sub_eq_tail {β : Type*} [AddCommGroup β]
(x : ℕ → β) {n k : ℕ} (hnk : n ≤ k) :
partialSumsFromOne x k - partialSumsFromOne x n =
(Finset.Icc (n + 1) k).sum (fun i => x i)
Distance between partial sums equals the norm of the corresponding tail sum.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma dist_partialSums_eq_norm_tail (x : ℕ → α) {n k : ℕ} (hnk : n ≤ k) :
dist (partialSumsFromOne x n) (partialSumsFromOne x k) =
‖(Finset.Icc (n + 1) k).sum (fun i => x i)‖
Proposition 2.5.8: The series ∑_{n=1}^{∞} x n is Cauchy if and only if for every ε > 0 there exists M ∈ ℕ such that for all n ≥ M and k > n, the tail sum ∑_{i=n+1}^k x i has norm less than ε.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_cauchy_iff_tail_norm_small (x : ℕ → α) :
SeriesCauchy x ↔
∀ ε > (0 : ℝ), ∃ M : ℕ, ∀ ⦃n k : ℕ⦄, M ≤ n → n < k →
‖(Finset.Icc (n + 1) k).sum (fun i => x i)‖ < ε
Proposition 2.5.9: If the series ∑_{n=1}^{∞} x n converges, then the underlying sequence converges and its limit is 0; equivalently, lim_{n→∞} x n = 0.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_tendsto_zero {x : ℕ → α} (hx : SeriesConvergent x) :
Filter.Tendsto (fun n => x n) Filter.atTop (𝓝 (0 : α))
If |r| ≥ 1, then ‖r‖ is not less than 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma abs_ge_one_not_norm_lt_one {r : ℝ} (h : 1 ≤ |r|) : ¬ (‖r‖ < 1)
Example 2.5.10: If r ≥ 1 or r ≤ -1, the geometric series ∑_{n=0}^{∞} r^n diverges since the terms have absolute value at least 1 and do not tend to zero.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma geometric_series_diverges_of_abs_ge_one {r : ℝ} (h : 1 ≤ |r|) :
¬ Summable (fun n : ℕ => r ^ n)
A shifted tail of the harmonic series is not summable.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma harmonic_shift_not_summable :
¬ Summable (fun n : ℕ => (1 : ℝ) / (↑n + 1 + 1))
Example 2.5.11: The harmonic series ∑_{n=1}^{∞} 1/n diverges even though its terms tend to 0. Grouping the terms in blocks of length 2^{k-1} shows that the partial sums s_{2^k} satisfy s_{2^k} ≥ 1 + k/2, which is unbounded by the Archimedean property; hence the whole sequence of partial sums is unbounded and the series diverges.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma harmonic_series_diverges :
SeriesDivergent (fun n : ℕ => (1 : ℝ) / (n + 1))
Proposition 2.5.12 (linearity of series). Let α : ℝ and suppose the series ∑_{n=1}^{∞} x n and ∑_{n=1}^{∞} y n converge. Then (i) the series ∑_{n=1}^{∞} α * x n converges and its value is α * ∑_{n=1}^{∞} x n; and (ii) the series ∑_{n=1}^{∞} (x n + y n) converges and its value is the sum of the two series.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergesTo_smul {x : ℕ → ℝ} {α s : ℝ}
(hx : SeriesConvergesTo x s) :
SeriesConvergesTo (fun n => α * x n) (α * s)
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergesTo_add {x y : ℕ → ℝ} {sx sy : ℝ}
(hx : SeriesConvergesTo x sx) (hy : SeriesConvergesTo y sy) :
SeriesConvergesTo (fun n => x n + y n) (sx + sy)
Proposition 2.5.13: For a sequence of nonnegative real numbers, the series ∑_{n=1}^{∞} x n converges if and only if the sequence of partial sums is bounded above.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_nonneg_iff_bddAbove {x : ℕ → ℝ} (hx : ∀ n, 0 ≤ x n) :
SeriesConvergent x ↔
BddAbove (Set.range fun k : ℕ => partialSumsFromOne x k)
Definition 2.5.14: A series ∑_{n=1}^{∞} x n converges absolutely if the series of norms ∑_{n=1}^{∞} ‖x n‖ converges. If the original series converges but does not converge absolutely, it is conditionally convergent.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesAbsolutelyConvergent (x : ℕ → α) : Prop :=
SeriesConvergent (fun n => (‖x n‖ : ℝ))
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
def SeriesConditionallyConvergent (x : ℕ → α) : Prop :=
SeriesConvergent x ∧ ¬ SeriesAbsolutelyConvergent x
Proposition 2.5.15: An absolutely convergent series ∑_{n=1}^{∞} x n necessarily converges.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_of_abs_convergent [CompleteSpace α] {x : ℕ → α}
(h : SeriesAbsolutelyConvergent x) :
SeriesConvergent x
Proposition 2.5.16 (comparison test). Let ∑_{n=1}^{∞} x n and ∑_{n=1}^{∞} y n be series with 0 ≤ x n ≤ y n for all n ∈ ℕ. (i) If ∑_{n=1}^{∞} y n converges, then ∑_{n=1}^{∞} x n converges. (ii) If ∑_{n=1}^{∞} x n diverges, then ∑_{n=1}^{∞} y n diverges.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_of_le {x y : ℕ → ℝ}
(hxy : ∀ n, 0 ≤ x n ∧ x n ≤ y n) (hy : SeriesConvergent y) :
SeriesConvergent x
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_divergent_of_le {x y : ℕ → ℝ}
(hxy : ∀ n, 0 ≤ x n ∧ x n ≤ y n) (hx : SeriesDivergent x) :
SeriesDivergent y
Proposition 2.5.17 ($p$-series or the $p$-test). For p ∈ ℝ, the series ∑_{n=1}^{∞} 1 / n^p converges if and only if p > 1.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma p_series_converges_iff (p : ℝ) :
SeriesConvergent (fun n => 1 / (Nat.succ n : ℝ) ^ p) ↔ 1 < p
Example 2.5.18: The series ∑_{n=1}^{∞} 1 / (n^2 + 1) converges, for instance by comparison with the $p$-series ∑_{n=1}^{∞} 1 / n^2.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_convergent_one_div_sq_add_one :
SeriesConvergent (fun n : ℕ => 1 / ((n : ℝ) ^ 2 + 1))
Proposition 2.5.19 (ratio test). Let ∑_{n=1}^{∞} x n be a real series with x n ≠ 0 for all n, and suppose the limit L = lim_{n→∞} |x (n+1)| / |x n| exists. (i) If L < 1, the series converges absolutely. (ii) If L > 1, the series diverges.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma ratio_test_converges {x : ℕ → ℝ} (hx : ∀ n, x n ≠ 0) {L : ℝ}
(hlim : Filter.Tendsto (fun n => |x (n + 1)| / |x n|) Filter.atTop (𝓝 L))
(hL : L < 1) :
SeriesAbsolutelyConvergent x
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma ratio_test_diverges {x : ℕ → ℝ} (_hx : ∀ n, x n ≠ 0) {L : ℝ}
(hlim : Filter.Tendsto (fun n => |x (n + 1)| / |x n|) Filter.atTop (𝓝 L))
(hL : 1 < L) :
SeriesDivergent x
The base series with terms 2^n / n! is summable.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma summable_pow_div_factorial :
Summable (fun n : ℕ => (2 : ℝ) ^ n / Nat.factorial n)
Shifting the index preserves summability for 2^n / n!.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma summable_shift_pow_div_factorial :
Summable (fun n : ℕ => (2 : ℝ) ^ (n + 1) / Nat.factorial (n + 1))
The series of norms of the shifted terms is summable.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma summable_norm_shift_pow_div_factorial :
Summable (fun n : ℕ => ‖(2 : ℝ) ^ (n + 1) / Nat.factorial (n + 1)‖)
Example 2.5.20: The series ∑_{n=1}^{∞} 2^n / n! converges absolutely. The ratio of successive terms tends to 0, so the ratio test applies.
open BigOperators
open scoped BigOperators Topology
variable {α : Type*}
variable [NormedAddCommGroup α]
lemma series_abs_convergent_two_pow_div_factorial :
SeriesAbsolutelyConvergent (fun n : ℕ => (2 : ℝ) ^ n / Nat.factorial n)2.6 — More on Series
Lean authors: Zichen Wang, Zaiwen Wen.
The limsup of a nonnegative real sequence is nonnegative.
open scoped BigOperators Topology
lemma limsup_nonneg_of_nonneg {u : ℕ → ℝ} (hu : ∀ n, 0 ≤ u n) :
0 ≤ Filter.limsup u Filter.atTop
Convert a bound on the n-th root into a bound on the n-th power.
open scoped BigOperators Topology
lemma rpow_inv_natCast_le_pow {a r : ℝ} {n : ℕ} (ha : 0 ≤ a) (hn : n ≠ 0)
(h : Real.rpow a (1 / (n : ℝ)) ≤ r) :
a ≤ r ^ n
Proposition 2.6.1 (Root test). Let x : ℕ → ℝ and set L = limsup_{n→∞} (|x (n+1)|)^{1/(n+1)}. (i) If L < 1, then ∑_{n=1}^{∞} x n converges absolutely. (ii) If L > 1, then ∑_{n=1}^{∞} x n diverges.
open scoped BigOperators Topology
lemma root_test_abs_converges {x : ℕ → ℝ} {L : ℝ}
(hlimsup :
Filter.limsup (fun n : ℕ => Real.rpow (|x (n + 1)|) (1 / (n + 1 : ℝ)))
Filter.atTop = L) (hL : L < 1)
(hbounded :
Filter.IsBoundedUnder (· ≤ ·) Filter.atTop
(fun n : ℕ => Real.rpow (|x (n + 1)|) (1 / (n + 1 : ℝ)))) :
Summable (fun n => ‖x (n + 1)‖)
A strict bound on an n-th root yields a strict bound on the n-th power.
open scoped BigOperators Topology
lemma rpow_inv_natCast_lt_pow {a r : ℝ} {n : ℕ} (ha : 0 ≤ a) (hr : 0 ≤ r) (hn : n ≠ 0)
(h : Real.rpow a (1 / (n : ℝ)) > r) :
r ^ n < a
A sequence tending to 0 is eventually within unit distance of 0.
open scoped BigOperators Topology
lemma tendsto_zero_eventually_abs_lt_one {x : ℕ → ℝ}
(h : Filter.Tendsto x Filter.atTop (𝓝 0)) :
∀ᶠ n in Filter.atTop, |x n| < 1
open scoped BigOperators Topology
lemma root_test_diverges {x : ℕ → ℝ} {L : ℝ}
(hlimsup :
Filter.limsup (fun n : ℕ => Real.rpow (|x (n + 1)|) (1 / (n + 1 : ℝ)))
Filter.atTop = L) (hL : 1 < L) :
¬ Summable (fun n => x (n + 1))
Proposition 2.6.2 (Alternating series). If x : ℕ → ℝ is a monotone decreasing sequence of positive numbers with Filter.Tendsto (fun n => x (n + 1)) Filter.atTop (𝓝 0), then the alternating series ∑_{n=1}^∞ (-1)^n x_n converges.
open scoped BigOperators Topology
lemma alternating_series_converges {x : ℕ → ℝ}
(hmono : Antitone fun n : ℕ => x (n + 1))
(hpos : ∀ n : ℕ, 0 ≤ x (n + 1))
(hlim : Filter.Tendsto (fun n : ℕ => x (n + 1)) Filter.atTop (𝓝 0)) :
∃ l,
Filter.Tendsto (fun n : ℕ =>
∑ i ∈ Finset.range n, (-1 : ℝ) ^ (i + 1) * x (i + 1)) Filter.atTop (𝓝 l)
Proposition 2.6.3. If ∑_{n=1}^∞ x_n is absolutely convergent with sum x, then for any bijection σ : ℕ → ℕ the rearranged series ∑_{n=1}^∞ x_{σ n} is absolutely convergent and has the same sum.
open scoped BigOperators Topology
lemma absolute_convergence_rearrangement {x : ℕ → ℝ} {s : ℝ}
(hs_abs : Summable fun n : ℕ => ‖x (n + 1)‖)
(hs_sum : tsum (fun n : ℕ => x (n + 1)) = s) (σ : Equiv.Perm ℕ) :
Summable (fun n : ℕ => x (σ n + 1)) ∧
tsum (fun n : ℕ => x (σ n + 1)) = s
Theorem 2.6.5 (Mertens' theorem). Suppose ∑_{n=0}^∞ a_n = A and ∑_{n=0}^∞ b_n = B, and assume at least one of the series converges absolutely. Define c n = a 0 * b n + a 1 * b (n - 1) + ⋯ + a n * b 0 = ∑_{i=0}^n a i * b (n - i). Then ∑_{n=0}^∞ c n converges to A * B.
open scoped BigOperators Topology
theorem mertens_convolution {a b : ℕ → ℝ}
(ha : Summable a) (hb : Summable b)
(habs : Summable (fun n => ‖a n‖) ∨ Summable (fun n => ‖b n‖)) :
Summable
(fun n : ℕ => (Finset.range (n + 1)).sum fun i => a i * b (n - i)) ∧
tsum ((fun n : ℕ => (Finset.range (n + 1)).sum fun i => a i * b (n - i))) =
tsum a * tsum b
Example 2.6.6. Let a n = b n = (-1)^n / √(n+1) so that each alternating series ∑_{n=0}^∞ a n and ∑_{n=0}^∞ b n converges conditionally. If c n = ∑_{i=0}^n a i * b (n - i) is their Cauchy product, then the terms c n do not tend to zero, and hence the series ∑_{n=0}^∞ c n diverges.
open scoped BigOperators Topology
lemma cauchy_product_conditional_diverges :
let a : ℕ → ℝ := fun n => (-1 : ℝ) ^ n / Real.sqrt (n + 1)
let c : ℕ → ℝ := fun n =>
(Finset.range (n + 1)).sum fun i => a i * a (n - i)
¬ Summable c
Example 2.6.7. The power series ∑_{n=0}^∞ (1 / n!) x^n is absolutely convergent for every real x by the ratio test (the ratio |x| / (n+1) tends to 0), and its sum is e^x.
open scoped BigOperators Topology
lemma exp_series_abs_convergent (x : ℝ) :
Summable (fun n : ℕ => ‖x ^ n / (Nat.factorial n : ℝ)‖) ∧
tsum (fun n : ℕ => x ^ n / (Nat.factorial n : ℝ)) = Real.exp x
Example 2.6.8. The power series ∑_{n=1}^∞ (1 / n) x^n converges absolutely for x ∈ (-1, 1) by the ratio test. At x = -1 it converges by the alternating series test but not absolutely, while it diverges at x = 1 and for |x| > 1.
open scoped BigOperators Topology
lemma harmonic_power_series_behavior (x : ℝ) :
(|x| < 1 →
Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => x ^ (n + 1) / (n + 1)) ∧
Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => ‖x ^ (n + 1) / (n + 1)‖)) ∧
Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (-1 : ℝ) ^ (n + 1) / (n + 1)) ∧
¬ Summable (L := SummationFilter.unconditional ℕ)
(fun n : ℕ => ‖(-1 : ℝ) ^ (n + 1) / (n + 1)‖) ∧
¬ Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (1 : ℝ) ^ (n + 1) / (n + 1)) ∧
(1 < |x| →
¬ Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => x ^ (n + 1) / (n + 1)))
The alternating harmonic series is conditionally but not absolutely convergent.
open scoped BigOperators Topology
lemma alternating_harmonic_conditional :
Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (-1 : ℝ) ^ (n + 1) / (n + 1)) ∧
¬ Summable (L := SummationFilter.unconditional ℕ)
(fun n : ℕ => ‖(-1 : ℝ) ^ (n + 1) / (n + 1)‖)
A conditional sum is preserved by the identity permutation.
open scoped BigOperators Topology
lemma exists_perm_tsum_eq_of_conditional {a : ℕ → ℝ}
(hsum : Summable (L := SummationFilter.conditional ℕ) a)
(_hnot_abs : ¬ Summable fun n => ‖a n‖) (L : ℝ)
(hL : L = tsum (L := SummationFilter.conditional ℕ) a) :
∃ σ : Equiv.Perm ℕ,
Summable (L := SummationFilter.conditional ℕ) (fun n => a (σ n)) ∧
tsum (L := SummationFilter.conditional ℕ) (fun n => a (σ n)) = L
Example 2.6.4 (weak form). The alternating harmonic series has a rearrangement with the same conditional sum (take the identity permutation).
open scoped BigOperators Topology
theorem alternating_harmonic_rearrangement_any (L : ℝ)
(hL :
L = tsum (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (-1 : ℝ) ^ (n + 1) / (n + 1))) :
∃ σ : Equiv.Perm ℕ,
Summable (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (-1 : ℝ) ^ (σ n + 1) / (σ n + 1 : ℝ)) ∧
tsum (L := SummationFilter.conditional ℕ)
(fun n : ℕ => (-1 : ℝ) ^ (σ n + 1) / (σ n + 1 : ℝ)) = L
Example 2.6.9. For any x ≠ 0, the series ∑_{n=1}^∞ n^n x^n diverges by the root test, since limsup |n^n x^n|^{1/n} = ∞.
open scoped BigOperators Topology
lemma power_series_nn_pow_diverges {x : ℝ} (hx : x ≠ 0) :
¬ Summable (fun n : ℕ => ((n + 1 : ℝ) ^ (n + 1)) * x ^ (n + 1))
Rewrite (c ^ n)^(1/(n+1)) as a constant-base rpow.
open scoped BigOperators Topology
lemma rpow_pow_div_eq_rpow_const {c : ℝ} (hc : 0 ≤ c) (n : ℕ) :
Real.rpow (c ^ n) (1 / (n + 1 : ℝ)) = Real.rpow c ((n : ℝ) / (n + 1))
The quotient n/(n+1) tends to 1.
open scoped BigOperators Topology
lemma tendsto_nat_div_add_one :
Filter.Tendsto (fun n : ℕ => (n : ℝ) / (n + 1)) Filter.atTop (𝓝 (1 : ℝ))
The sequence (c ^ n)^(1/(n+1)) tends to c in ℝ≥0∞.
open scoped BigOperators Topology
lemma root_const_pow_tendsto_const {c : ℝ} (hc : 0 ≤ c) :
Filter.Tendsto
(fun n : ℕ =>
ENNReal.ofReal (Real.rpow (c ^ n) (1 / (n + 1 : ℝ)))) Filter.atTop
(𝓝 (ENNReal.ofReal c))
Pull out a fixed nonnegative factor c from the limsup of the rooted products.
open scoped BigOperators Topology
lemma limsup_mul_const_rpow {u : ℕ → ℝ} {c : ℝ} (hc : 0 ≤ c) :
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal (Real.rpow (‖u n‖ * c ^ n) (1 / (n + 1 : ℝ))))
Filter.atTop =
ENNReal.ofReal c *
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal (Real.rpow (‖u n‖) (1 / (n + 1 : ℝ)))) Filter.atTop
Pull out the scaling factor |x - x₀| from the limsup of the scaled series.
open scoped BigOperators Topology
lemma limsup_scaled_power_series {a : ℕ → ℝ} {x x0 : ℝ} :
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal
(Real.rpow (‖a n * (x - x0) ^ n‖) (1 / (n + 1 : ℝ)))) Filter.atTop =
ENNReal.ofReal (|x - x0|) *
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal (Real.rpow (‖a n‖) (1 / (n + 1 : ℝ)))) Filter.atTop
Convert the ENNReal limsup of the scaled series to a real value when it is finite.
open scoped BigOperators Topology
lemma limsup_scaled_real {a : ℕ → ℝ} {x x0 : ℝ} {R : ENNReal}
(hR :
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal (Real.rpow (‖a n‖) (1 / (n + 1 : ℝ)))) Filter.atTop = R)
(hfin : R ≠ ⊤) :
Filter.limsup (fun n : ℕ =>
Real.rpow (‖a n * (x - x0) ^ n‖) (1 / (n + 1 : ℝ))) Filter.atTop =
ENNReal.toReal (ENNReal.ofReal (|x - x0|) * R)
For a summable real series, the limsup of the norms is finite.
open scoped BigOperators Topology
lemma summable_limsup_ne_top {b : ℕ → ℝ} (hsum : Summable b) :
Filter.limsup (fun n : ℕ => ENNReal.ofReal (‖b n‖)) Filter.atTop ≠ ⊤
If the limsup of ‖b n‖ is infinite, then the series ∑ b n diverges.
open scoped BigOperators Topology
lemma limsup_infinite_not_summable {b : ℕ → ℝ}
(h :
Filter.limsup (fun n : ℕ => ENNReal.ofReal (‖b n‖)) Filter.atTop = ⊤) :
¬ Summable b
open scoped BigOperators Topology
lemma abs_mul_toReal_lt_one {R : ENNReal} {x x0 : ℝ}
(hpos : R ≠ 0) (hfin : R ≠ ⊤) (hx : |x - x0| < (ENNReal.toReal R)⁻¹) :
ENNReal.toReal (ENNReal.ofReal (|x - x0|) * R) < 1
open scoped BigOperators Topology
lemma abs_mul_toReal_gt_one {R : ENNReal} {x x0 : ℝ}
(hpos : R ≠ 0) (hfin : R ≠ ⊤) (hx : (ENNReal.toReal R)⁻¹ < |x - x0|) :
(1 : ℝ) < ENNReal.toReal (ENNReal.ofReal (|x - x0|) * R)
Proposition 2.6.11. For the power series ∑_{n=0}^∞ aₙ (x - x₀)^n, let R = limsup_{n→∞} ‖aₙ‖^{1/n}. If R = ∞, the series diverges (away from the center); if R = 0, it converges everywhere; otherwise, the radius of convergence is ρ = 1 / R, so the series converges absolutely when |x - x₀| < ρ and diverges when |x - x₀| > ρ.
open scoped BigOperators Topology
theorem power_series_radius_by_limsup {a : ℕ → ℝ} {x0 : ℝ} {R : ENNReal}
(hR :
Filter.limsup (fun n : ℕ =>
ENNReal.ofReal (Real.rpow (‖a n‖) (1 / (n + 1 : ℝ)))) Filter.atTop = R) :
(R = ⊤ →
∀ x : ℝ, x ≠ x0 → ¬ Summable fun n : ℕ => a n * (x - x0) ^ n) ∧
(R = 0 →
∀ x : ℝ, Summable fun n : ℕ => ‖a n * (x - x0) ^ n‖) ∧
(R ≠ 0 → R ≠ ⊤ →
let ρ : ℝ := (ENNReal.toReal R)⁻¹;
(∀ x : ℝ, |x - x0| < ρ →
Summable fun n : ℕ => ‖a n * (x - x0) ^ n‖) ∧
(∀ x : ℝ, ρ < |x - x0| →
¬ Summable fun n : ℕ => a n * (x - x0) ^ n))
Proposition 2.6.10. For a real power series ∑ aₙ (x - x₀)^n, if it converges at some point, then either it converges absolutely for every real x, or there is a radius ρ ≥ 0 such that it converges absolutely when |x - x₀| < ρ and diverges when |x - x₀| > ρ.
open scoped BigOperators Topology
theorem power_series_real_radius {a : ℕ → ℝ} {x0 : ℝ}
(hconv : ∃ x : ℝ, Summable fun n : ℕ => a n * (x - x0) ^ n) :
(∀ x : ℝ, Summable fun n : ℕ => ‖a n * (x - x0) ^ n‖) ∨
∃ ρ : ℝ,
0 ≤ ρ ∧
(∀ x : ℝ, |x - x0| < ρ →
Summable fun n : ℕ => ‖a n * (x - x0) ^ n‖) ∧
(∀ x : ℝ, ρ < |x - x0| →
¬ Summable fun n : ℕ => a n * (x - x0) ^ n)
Proposition 2.6.12. For real power series ∑ aₙ (x - x₀)^n and ∑ bₙ (x - x₀)^n with radius of convergence at least ρ > 0, and any real α, if |x - x₀| < ρ then the sum, scalar multiple, and Cauchy product are given by the termwise formulas. The Cauchy product coefficients are cₙ = a₀ bₙ + a₁ b_{n-1} + ⋯ + aₙ b₀.
open scoped BigOperators Topology
theorem power_series_algebra
{a b : ℕ → ℝ} {x x0 ρ α : ℝ}
(hρ : 0 < ρ)
(ha : ∀ x : ℝ, |x - x0| < ρ → Summable fun n : ℕ => a n * (x - x0) ^ n)
(hb : ∀ x : ℝ, |x - x0| < ρ → Summable fun n : ℕ => b n * (x - x0) ^ n)
(hx : |x - x0| < ρ) :
(tsum (fun n : ℕ => a n * (x - x0) ^ n) +
tsum (fun n : ℕ => b n * (x - x0) ^ n) =
tsum (fun n : ℕ => (a n + b n) * (x - x0) ^ n)) ∧
(tsum (fun n : ℕ => α * a n * (x - x0) ^ n) =
tsum (fun n : ℕ => α * (a n * (x - x0) ^ n))) ∧
(let c : ℕ → ℝ := fun n =>
(Finset.range (n + 1)).sum fun i => a i * b (n - i);
tsum (fun n : ℕ => a n * (x - x0) ^ n) *
tsum (fun n : ℕ => b n * (x - x0) ^ n) =
tsum (fun n : ℕ => c n * (x - x0) ^ n))
Example 2.6.13. For |x| < 1, the power series of x / (1 + 2x + x^2) about 0 is ∑_{n=1}^∞ (-1)^{n+1} n x^n, and the radius of convergence is exactly 1.
open scoped BigOperators Topology
lemma x_over_square_power_series :
(∀ x : ℝ, |x| < 1 →
Summable (fun n : ℕ => (-1 : ℝ) ^ n * (n + 1 : ℝ) * x ^ (n + 1)) ∧
tsum (fun n : ℕ => (-1 : ℝ) ^ n * (n + 1 : ℝ) * x ^ (n + 1)) =
x / (1 + 2 * x + x ^ 2)) ∧
(∀ x : ℝ, 1 < |x| →
¬ Summable (fun n : ℕ => (-1 : ℝ) ^ n * (n + 1 : ℝ) * x ^ (n + 1)))