Remark: In this sections, all the subsets of R are assumed to be nonempty.
Let E be a subset of R. We say that E is bounded above if there exists a real number U such that x ≤ U for all x ∈ E. In this case, we say that U is an upper bound for E. We say that E is bounded below if there exists a real number L so that x ≥ L for all x ∈ E. In this case, we say that L is a lower bound for E. A subset E of R is said to be bounded if E is both bounded above and bounded below.
Let E be a subset of R.
(1) Suppose E is bounded above. An upper bound U of set E is the least upper bounded of E if for any upper bound U0 of E, U0 ≥ U. If U is the least upper bound of E, we denote U by sup E. The least upper bound for E is also called supremum of E.
(2) Suppose E is bounded below. A lower bounded L of E is said to be the greatest lower bound of E if for any lower bound L0 of E, L ≥ L0. If L is the greatest lower bound for E, we denote L by inf E. The greatest lower bound for E is also called the infimum of E.
Example 1.1. Let a, b be real numbers. The set [a, b] = {x ∈ R : a ≤ x ≤ b} is bounded with sup[a, b] = b and inf[a, b] = a.
Example 1.2. Let a be a real numbers. We denote (−∞, a) = {x ∈ R : x < a}. Then (−∞, a) is bounded above but not bounded below.
Example 1.3. Given a sequence (an) of real numbers, let {an∈ R : n ≥ 1} be the image of (an), i.e. the set of all values of (an). Then (an) is bounded (bounded above, bounded below) if and only if the set {an∈ R : n ≥ 1} is bounded (bounded above, bounded below).
Theorem 1.1. (Property of R) In R, the following hold:
(1) Least upper bound property: Let S be a nonempty set in R that has an upper bound. Then S has a least upper bound.
(2) Greatest lower bound property: Let P be a nonempty subset in R that has a lower bound. Then P has a greatest lower bound in R.
Example 1.4. Consider the set S = {x ∈ R : x2+ x < 3}. Find sup S and inf S.
Example 1.5. Let S = {x ∈ Q : x2< 2}. Find sup S and inf S.
Example 1.6. Let S = {x ∈ R : x3 < 1}. Find sup S. Is S bounded below?
Proposition 1.1. Let E be a bounded subset of R and U ∈ R is an upper bound of E.
Then U is the least upper bound of E if and only if for any > 0, there exists x ∈ E so that x ≥ U − .
Proof. Suppose U = sup E. Then for any > 0, U − < U. Hence U − is not an upper bound of E. Claim: there exists x ∈ E so that x > U − . If there is no x so that x > U − ,
1
then x ≤ U − for all x ∈ E. This implies that U − is again an upper bound for E. By the definition of the least upper bound, U ≤ U − which is absurd since > 0.
Conversely, let U0 = sup E. Since U is an upper bound of E, U ≥ U0. Claim U0 = U. For any > 0, choose x ∈ E so that x > U − . Thus U0 ≥ x > U − . We know that for any U0 > U − for all > 0. Since is arbitrary U0 ≥ U. We conclude that U0 = U.
Lemma 1.1. Let E be a nonempty subset of R.
(1) If E is bounded above and α < sup E, then there exists x ∈ E so that x > α.
(2) If E is bounded above and β > inf E, then there exists x ∈ E so that x < β.
Proof. The result follows from the definition.
Corollary 1.1. Let E be a bounded subset of R and L ∈ R is a lower bound of E. Then L is the greatest upper bound of E if and only if for any > 0, there exists x ∈ E so that x ≤ L − .
Proof. We leave it to the reader as an exercise.
Let E be a nonempty subset of R. We say that M is a maximum of E if M is an element of E and x ≤ M for all x ∈ E. Using the definition, we immediately know that there is only one maximum of E if the maximum elements of E exists. In this case, we denote M by max E. It also follows from the definition that if the maximum of E exists, it must be bounded above.
We say that m is the minimum of a nonempty subset E of R if m is an element of E and x ≥ m for all x ∈ m. We denote m by min E. It follows from the definition that if a set has minimum, it must be bounded below.
Example 1.7. The following subsets of R are all bounded. Hence their greatest lower bound and their least upper bound exist. Determine whether their maximum or minimum exist.
(1) E1 = (0, 1).
(2) E2 = (0, 1].
(3) E3 = [0, 1).
(4) E4 = [0, 1].
Proposition 1.2. Suppose E is a nonempty subset of R. If E is bounded above, then the maximum of E exists if and only if sup E ∈ E. Similarly, if E is bounded below, then the minimum of E exists if and only if inf E ∈ E.
Proof. The proof follows from the definition.
Proposition 1.3. Let E be a nonempty subset of R. Suppose E is a bounded set. Then inf E ≤ sup E.
Proof. We leave it to the reader as an exercise.
Proposition 1.4. Let E and F be nonempty subsets of R. Suppose that E ⊂ F.
(1) If E and F are both bounded below, then inf F ≤ inf E.
(2) If E and F are both bounded above, then sup E ≤ sup F.
Proof. Let us prove (a). (b) is left to the reader.
For any x ∈ F, x ≥ inf F. Since E is a subset of F, x ≥ inf F holds for all x ∈ E. Therefore inf F is a lower bound for E. Since inf E is the greatest lower bound for E, inf E ≥ inf F.
Theorem 1.2. Every bounded monotone sequence is convergent.
Proof. Without loss of generality, we may assume that (an) is a bounded nondecreasing sequence of real numbers. Let a = sup{an : n ≥ 1}. Given > 0, there exists aN so that a ≥ aN > a − . Since (an) is nondecreasing, an≥ aN for every n ≥ N. Hence an> a − for every n ≥ N. In this case, |an− a| = a − an< for n ≥ N. We prove that lim
n→∞an= a.
2. limsup and liminf
Let (an) be a bounded sequence of real numbers. Define a new sequence (xn) by xn= sup{am: m ≥ n}, n ≥ 1,
Since (an) is bounded, xn is a real number for each n ≥ 1. We assume that |an| ≤ M for all n ≥ 1. Then |xn| ≤ M for all n ≥ 1. This shows that (xn) is also a bounded sequence. By Proposition 1.4, (xn) is nonincreasing. By monotone sequence property, x = lim
n→∞xnexists.
We denote x by lim sup
n→∞
an. Similarly, define a sequence (yn) by yn= inf{am: m ≥ n}, n ≥ 1.
Then (yn) is a nondecreasing sequence by Proposition 1.4. By monotone sequence property, y = lim
n→∞yn exists. We denote y by lim inf
n→∞ an. From now on, we will simply denote by
a∗ = lim sup
n→∞
an, a∗= lim inf
n→∞ an.
Since (xn) is nonincreasing and bounded below, its limit equals to inf{xn : n ≥ 1}.
Similarly, (yn) is nondecreasing and bounded above, its limit equals to sup{yn: n ≥ 1}.
Example 2.1. Find lim sup
n→∞
(−1)n and lim inf
n→∞ (−1)n.
Solution: Let us compute these two numbers via definition. Denote (−1)n by an. For each k ≥ 1, we know {an : n ≥ k} = {−1, 1}. For k ≥ 1, xk = sup{an : n ≥ k} = sup{−1, 1} = 1. Similarly, for k ≥ 1, yk = sup{am : m ≥ k} = inf{−1, 1} = −1. Therefore lim sup
n→∞
an = lim
k→∞xk = 1 while lim inf
n→∞ an = lim
k→∞yk = −1. Notice that the subsequence (a2n) of (an) is convergent to 1 and the subsequence (a2n−1) of (an) is convergent to −1.
Later, we will prove that in general, the limit supremum and the limit infimum of a bounded sequence are always the limits of some subsequences of the given sequence.
Example 2.2. Let (an) be the sequence defined by an= 1 − 1
n, n ≥ 1.
Evaluate lim sup
n→∞
an and lim inf
n→∞ an.
Solution: The sequence (an) is increasing and bounded above by 1. Let us prove that sup{an : n ≥ 1} = 1. We have seen that 1 is an upper bound. Now, for each > 0, choose N = [1/] + 1. For n ≥ N, we see
1 − < 1 − 1 n.
By Proposition 1.1, we find 1 is indeed the least upper bound for {an : n ≥ 1}. Therefore 1 = sup{an: n ≥ 1}. We also know that lim
n→∞an= 1. In other words, we prove
n→∞lim
1 − 1
n
= sup
1 − 1
n : n ≥ 1
.
This gives us an example of Theorem 1.2. Similarly, for each k ≥ 1, we can show that 1 is the least upper bound of the set {1 − 1/n : n ≥ k} and −1 is the greatest lower bound for
{1 − 1/n : n ≥ k}. In other words, we find that xk= sup
1 − 1
n : n ≥ k
= 1, yk= inf
1 −1
n : n ≥ k
= −1.
This shows that lim
k→∞xk= 1 and lim
k→∞yk= −1. In other words, lim sup
n→∞
an= lim inf
n→∞ an= 1.
In this case, (an) is convergent to 1 and at the same time, both limit supremum and limit infimum of (an) are also equal to 1. This is not an accident. Later, we will prove that a bounded sequence is convergent if and only if its limit supremum equals to its limit infimum.
Lemma 2.1. Let (an) be a bounded sequence and a ∈ R.
(1) If a > a∗, there exists k ∈ N such that an< a for all n ≥ k.
(2) If a < a∗, then for all k ∈ N, there exists n ∈ N with n ≥ k such that an> a (3) If a < a∗, there exists k ∈ N such that an> a for all n ≥ k.
(4) If a > a∗, then for all k ∈ N, there exists n ∈ N such that an< a.
Proof. Let (xk) be the sequence defined as above, i.e. for each k ≥ 1, xk = sup{an: n ≥ k}.
By definition, a∗= inf{xk: k ≥ 1}.
(1) If a > a∗, a is not a lower bound for {xk : k ≥ 1}. Then there exists an element xk0 ∈ {xk : k ≥ 1} such that xk0 < a. Since a > xk0 and xk0 is the least upper bound for {an: n ≥ k0}, then a is an upper bound for {an: n ≥ k0}. Hence an< a for all n ≥ k0.
(2) If a < a∗, then a is a lower bound for {xk : k ≥ 1}. Then for all k ≥ 1, xk > a. For each k ≥ 1, xk is the least upper bound for {an: n ≥ k}. Since a < xk, a is not an upper bound for {an: n ≥ k}. Hence we can choose an element an∈ {an: n ≥ k} so that an> a.
In other words, we can find n ∈ N with n ≥ k such that an> a.
(3) and (4) are left to the reader.
Definition 2.1. We say that a real number x is a cluster point of a bounded sequence (an) if there exists a subsequence (ank) of (an) whose limit is x.
Corollary 2.1. Let (an) be a bounded sequence. Then both lim sup
n→∞
an and lim inf
n→∞ an are cluster points.
Proof. Since a∗+ 1 > a∗, there exists n1 ∈ N so that an< a∗+ 1 for all n ≥ n1 by Lemma 2.1. Since a∗ − 1 < a∗, for the given n1, we can find m1 ∈ N with m1 ≥ n1 such that am1 > a∗− 1. Since m1 ≥ n1, am1 < a∗+ 1. We find a∗− 1 < am1 < a∗+ 1. Inductively, we obtain a subsequence (amk) of (an) so that
a∗−1
k < amk < a∗+ 1
k, k ≥ 1.
Since lim
k→∞ a∗−
k = lim
k→∞ a∗+
k = a∗, by the Sandwich principle, lim
k→∞amk = a∗. This shows that a∗ is a cluster point.
Similarly, we can show that a∗ is a cluster point.
Lemma 2.2. If (an) is a convergent to a, the all the subsequences (ank) is also convergent to a.
Proof. Suppose (an) is convergent to a. Then for any > 0, there exists N∈ N so that
|an− a| < , whenever n ≥ N.
Let (ank) be any subsequence of (an). For any k ≥ N, nk ≥ k ≥ N. Then |ank − a| < whenever k ≥ N. This shows that lim
k→∞ank = a.
Corollary 2.2. If (an) is convergent to a, then a∗ = a∗= a.
Proof. Since a∗ and a∗ are both cluster points, Lemma 2.2 implies that a∗ = a = a∗. In fact, the converse is also true:
Theorem 2.1. Let (an) be a bounded sequence of real numbers. Then (an) is convergent to a real number a if and only if a∗= a∗ = a.
Proof. We have proved one direction. Conversely, let us assume a∗ = a∗= a.
For any > 0, a + > a∗. By Lemma 2.1, there exists k ∈ N so that for any n ≥ k, an< a∗+ .
Since a − < a∗ = a, Lemma 2.1 implies that there exists j ∈ N so that an > a − whenever n ≥ j. Denote N = max{k, j}. Then for all n ≥ N, we have
a − < an< a + .
Thus |an− a| < whenever n ≥ N. This shows that lim
n→∞an= a.
Let (an) be a bounded sequence. If E is the set of all cluster points of (an), we know that it is nonempty by Bolzano-Weierstrass theorem (since any bounded sequence has a convergent subsequence). For x ∈ E, choose a subsequence (ank) of (an) so that lim
k→∞ank = x. By Theorem 2.1, we know
lim sup
k→∞
ank = lim inf
k→∞ ank = x.
By definition, for any j ≥ 1, the set {ank : k ≥ j} is a subset of {an: n ≥ j}. This is because for k ≥ j, nk ≥ k ≥ j. For each j ≥ 1,
sup{an: n ≥ j} ≥ sup{ank : k ≥ j}.
Denote the left hand side by xj and the right hand side by αj. Then xj ≥ αj for all j ≥ 1.
Since both (xj) and (αj) are convergent, by Lemma ??, we find a∗= lim
j→∞xj ≥ lim
j→∞αj = x.
This shows x ≤ a∗, i.e. E is bounded above by lim sup
n→∞
an. Using a similar argument, we can show that x ≥ a∗, i.e. E is bounded below by a∗. We obtain that E is a nonempty bounded subset of R. Since both a∗ and a∗ are cluster points, i.e. a∗, a∗ ∈ E, we obtain that:
Theorem 2.2. Let (an) be a bounded sequence and E be the set of all cluster points of (an). Then
lim sup
n→∞
an= max E, lim inf
n→∞ an= min E.
Example 2.3. Construct a sequence with exact three different cluster points.
Solution: We leave it to the reader as an exercise.
Example 2.4. Find the limsup and liminf of the following sequence of real numbers.
(1) an= (−1)n+ 1
n, n ≥ 1.
(2) an= cos(nπ +π
6), n ≥ 1.
(3) an=
1
n if n = 3k
1 +n1 if n = 3k + 1
−2 + n1 if n = 3k + 2.
Theorem 2.3. Let (sn) and (tn) be bounded sequences of real numbers. Suppose that there exists N > 0 so that sn≤ tn for n ≥ N. Then
(1) lim inf
n→∞ sn≤ lim inf
n→∞ tn. (2) lim sup
n→∞
sn≤ lim sup
n→∞
tn.
Proof. We leave it to the reader as an exercise.
It follows immediately from this theorem that
Corollary 2.3. Let (an) be a sequence of real numbers. Then (1) If an≤ M for all n, then lim supn→∞an≤ M.
(2) If an≥ M for all n, then lim infn→∞an≥ M.