Subsequence
Definition 2.3.1. (Subsequence) Suppone that {an} is any sequence and let
n1 < n2 < n3 < ... < nk < ...
be infinite subset of ℕ i.e. a strictly increasing sequence of natural numbers. Then {ank} is called the subsequence of the squence {an}. For every intege k ∈ ℕ, by induction on k, nk is at least as large as k □
If {an} is a sequence, if for k ∈ ℕ we set nk = 2k, we get the subsequence {ank} of even-indexed term. Similarly, if by letting nk = 2k − 1, we obtain the subsequence {a2k−1} of odd-indexed terms.
For example consider the sequence {1/n}. Then, the sequence {1/2, 1/4,1/6, ..., 1/2n, ...} is the subsequence of even-indexed terms of {1/n}.
Theorem 2.3.2. If a sequence {an} of real numbers converges to a limit L, then every subsequence {ank} of {an} converges to the same limit.
Proof. Since {an} converges to L, for each ε > 0, there exists a positive integer n0 such that
|an − L| < ε ∀n > n0
Then taking n ≥ nk ≥ n0 we have
|ank − L| < ε ∀n > n0
Hence {ank} converges to L. □
The converge of the last theorem is not generally true. For example consider {(−1)n}. This sequence is not convergent. But the subsequence {ank} = {1,1,1,..} whose all terms are 1 converges to 1 as n ⟶ ∞
Theorem (Nested intervals property) Consider a decreasing sequnce of nested closed intervals in ℝ:
[a1, b1] ⊃ [a2, b2] ⊃ [a3, b3] ⊃ ...
If the lenghts of these intervals vanish, i.e,
limn ⟶ ∞ bn − an = 0
then there is exactly one real number ℓ that is common to all of these intervals, i.e.
limn ⟶ ∞ [bn − an] = {ℓ}
Proof. The nondecreasing sequence of the left end points a1, a2, a3, ... is bounded by any one of the right endpoints bn; so by the Complete Axiom, it has a least upper bound, say α. Similarly, the nonincreasing sequence b1, b2, b3,.. of right end points has a greatest lower bound β. Therefore
a1 ≤ a2 ≤ a3 ≤ ... ≤ α ≤ β ≤ ... ≤ b1 ≤ b2 ≤ b3
In particular, 0 ≤ β − α bn − an for every index n. Since limn ⟶ ∞ (bn − an) = 0, the squeeze theorem implies that β − α = 0. So α = β, and this common value is the number r. □
The nested intervla property holds because the completeness of real numbers enusres that ther are no gaps in ℝ.
We have proved that a convergent sequence is bounded and that the converse of this theorem is false take {(−1)n} for example. The Bolzano–Weierstrass theorem states that every bounded sequence has a convergent subsequence. Considering an = {(−1)n}, note that the subsequence a2n is the constant sequence 1 and so converges to 1. On the ohter hand a2n + 1 is the constant sequence −1 and so congerges to −1.
Consider the following sequences:
{an} = (1,0,1,0,1,0,...); {bn} = (1,2,3,4,5,6,7,...}
Neither {an} nor {bn} converges. Nervertheless, these two instances of nonconvergent sequences don't feel quite the same: the sequence {an} has a convergent sequence "inside it" while {bn} does not have. Indeed {an} is a bounded sequence.
Theorem (The Bolzano–Weierstrass theorem) Every bounded sequence of real numbers has a convergent subsequence (hence, an accumulation point).
Proof. Let {an} be a bounded sequence in ℝ so that there exists b > 0 satisfying |an| ≤ b and let S = {an : n ∈ ℕ} be the range of {an}. If S is a finite set, then at least one point ℓ in it must repeat infinitely ofter; i.e., there is an increasing sequence of indicides n1 < n2 < ... such that anj for all j = 1, 2, 3,.. . It follows that the subsequence anj converges (trivially) to a, making ℓ an accumulation point.
If S is infinite, then given the boundedness of {an}, there are real numbers a,b such that S ⊂ [a,b]. If we partition the closed interval [a,b] into halves
[a,b] = [a, a + b/2] ∪ [a + b/2, b]
then at least one on the half interval must contain infinitely many disctinct elements of S. Denote that half interval by [a1,b1] where either a1 or b1 is (a+b)/2 and note that the lenght of [a1,b1] is half the lenght of [a,b], i.e. (b − a)/2.
If we further partition [a1,b1] into half intervals, then one of the new hald intervals, call it [a2,b2], contains infinitely many (distinct) points of S, and its lenght is (b − a)/4. Repeating this partitioning process generates a nested sequence of hald intervals [ak, bk] such that
[a1, b1] ⊃ [a2, b2] ⊃ [a3, b3] ⊃ ...
the lenght of [ak, bk] is (b − a)/2k for every k = 1, 2, 3, ...
[ak, bk] contains infinitely many ponts of S.
By the nested interval property
This number r is in [ak, bk] for every k, which by (iii) above must also contain at least one term an, say ank, where ank ≠ ℓ. It follows that the distance between ank and ℓ cannot exceed the lenght of the interval [ak, bk], i.e.
|ank − r| < ε, k ≥ N
We conclude that the subsequence ank converges to ℓ and therefore, ℓ is an accumulation point. ■