Sequences
Definition A sequence {pn} in X is a function f : N → X such that f(n) = pn ∈ X.
Definition The sequence {pn} converges in X if there exists p ∈ X such that for each > 0, there exists N ∈ N such that if n > N then d(p, pn) < . We say {pn} converges to p. Or equivalently, p is the limit of {pn}.
Notation pn→ p ⇐⇒ lim
n→∞{pn} = p.
Definitions
(a) If {pn} does not converge to any point in X then it diverges.
(b) The range of {pn} is {x ∈ X | x = pn for some n}.
(c) We say {pn} isbounded in X if the range of {pn} is bounded in X.
True/False Questions Let {pn} be a sequence in a metric space X.
(a) If pn → p and pn → p0, then p = p0. True.
(b) If {pn} is bounded, then {pn} converges. False.
(c) If pn converges, then {pn} is bounded. True.
(d) If pn → p, then p is a limit point of the range of {pn}. False.
(e) If p is a limit point of E ⊆ X, then there exists some sequence {pn} such that pn → p. True.
(f) pn→ p if and only if every neighborhood of p contains all but finitely many terms in {pn}.
True.
Remark “All but finitely many” is equivalent to “Almost all.”
Solution
a) Let > 0. Then there exists N1 ∈ N such that n > N1, d(p, pn) <
2. Similarly there exists N2 ∈ N such that n > N2, d(p0, pn) <
2. Then for n > max (N1, N2):
d(p, p0) ≤ d(p, pn) + d(p0, pn) < 2
2 = .
Thus d(p, p0) < for all > 0. Thus d(p, p0) = 0. Hence p = p0. b) Consider Ex. (4) where there is oscillation between two points.
c) Suppose pn → p. Since 1 > 0 (our choice of ), there exists N ∈ N such that for n > N implies d(pn, p) < 1. Let r = max (1, d(p, p1), . . . , d(p, pN)) + 1. Then pn∈ B(p, r) for all n ∈ N.
d) Consider Ex. (3) where the sequence is simply one point.
e) p ∈ E0. For all n ∈ N such that pn ∈ E such that d(pn, p) < n1 and pn 6= p. There for a sequence {pn}. Let > 0. Choose N > 1
. Then for each n > N , we have d(p, pn) < 1 n < 1
N < .
Thus this sequences {pn} converges to p as desired.
f) (⇒) Suppose pn → p. Given B(p, ), there exists N ∈ N such that when n > N, it fol- lows that d(p, pn) < , i.e. pn∈ B(p, ), leaving only finitely many points, p1 through pnpossible.
(⇐) For all > 0, B(p, ) contains almost all {pn}. For > 0, let m = max{n ∈ N | pn6∈ B(p, )}.
Then n > m implies pn∈ B(p, ), i.e. d(pn, p) < .
Theorem (Limit Laws) Suppose {sn}, {tn} are sequences in C, and lim
n→∞sn = s, lim
n→∞tn = t.
Then (a) lim
n→∞(sn+ tn) = s + t.
(b) lim
n→∞(csn) = cs for all c ∈ C.
(c) lim
n→∞(c + sn) = c + s for all c ∈ C.
(d) lim
n→∞(sntn) = st.
(e) lim
n→∞
1 sn
= 1
s provided sn6= 0 for all n ∈ N, and s 6= 0.
Outline of the Proof (a) Let > 0. Since lim
n→∞sn = s and lim
n→∞tn= t, there exists N1, N2 ∈ N such that if n ≥ N1, and if n ≥ N2 then |sn− s| <
2 and |tn− s| <
2, respectively. This imples that if n ≥ max (N1, N2), then
|(sn+ tn) − (s + t)| = |(sn+ tn) − (s + t)|
= |sn− s| + |tn− t|
< .
(b) Let > 0. Since lim
n→∞sn= s, there exists N ∈ N such that if n ≥ N, then |sn− s| < 1 + |c|. This imples that if n ≥ N, then |csn− cs| ≤ |c| |sn− s| < |c|
1 + |c| < .
(d) Let > 0. Choose k = max(|s|, |t|, 1, ). Since lim
n→∞sn = s and lim
n→∞tn = t, there exists N1, N2 ∈ N such that if n ≥ N1, and if n ≥ N2 then |sn− s| <
3k and |tn− s| <
3k, respectively.
This imples that if n ≥ max (N1, N2), then
|sntn− st| = |(sn− s)(tn− t) + s(tn− t) + t(sn− s)|
≤ 2
9k2 + |s|
3k +|t|
3k
≤ 9k +
3+
3 since k ≥ |s|, |t|, =⇒ 1 ≥ |s|
k , |t|
k , k
≤ 9+
3+
3 since k ≥ 1
< .
(e) Let > 0. Since lim
n→∞sn = s 6= 0, there exists N1, N2 ∈ N such that if n ≥ N1, then |sn− s| < 1
2|s| =⇒ |s| − |sn| ≤ |sn− s| < 1
2|s| =⇒ 1
2|s| ≤ |sn| if n ≥ N2, then |sn− s| < 1
2|s|2, respectively.
This imples that if n ≥ max (N1, N2), then
1 sn − 1
s
=
sn− s sns
< 2
|s|2 |sn− s|
< .
Subsequences
Definition Let {pn} be a sequence in X and let {ni} be a sequence of natural numbers such that n1 < n2 < n3 < . . . . Then {pni} is a subsequence of {pn}.
Examples
(a) Let {pn} = {1, π,12, π, 13, π, . . . }. Notice this does not converge. But a subsequence does converge. For example, only the π terms. As do the other terms.
(b) Notice that {12,23,34,45, . . . } converges for 1. Notice that any subsequence converges. By property (f ) on Page 1, almost all terms of the subsequence are contained in any neighbor- hood of the limit.
Remarks
(a) Note that if pn → p and {pnk} is any subsequence of {pn}, then pnk → p.
(b) Question: Must every sequence contain a convergent subsequence?
No. Consider the sequence {1, 2, 3, . . . }, and note that there is no convergent subsequence of this sequence.
Recall that an ordered field F is said to have the least-upper-bound property if every nonempty, bounded above subset E of F , the sup E ∈ E0 exists in F. For general metric spaces without the least-upper-bound property, the following generalizations hold.
Theorem
(a) In a compact metric space X, every sequence contains a subsequence converging to a point in X, i.e. if {pn} is a sequence in a compact metric space, then there exists a subsequence {pnj} of {pn} converges to a point p ∈ X.
Proof The proof is simmilar to the proof of Theorem 2.41 (b) =⇒ (c) on p40.
(b) (Bolzano-Weierstrass) Every bounded sequence in Rkcontains a convergent subsequence.
Proof Let R = {pn | n ∈ N} = the range of {pn}.
Suppose R is a finite subset of Rk, there exists a point p ∈ R and a subsequence {nj | j ∈ N} of {n | n ∈ N} such that
pnj = p ∀ j ∈ N by the pigeon-hole principle. Since
j→∞lim pnj = p, pnj is a convergent subsequence of {pn}.
Suppose R is a infinite subset of Rk, since R is bounded, there exists a k-cell I such that R = {pn | n ∈ N} is an infinite subset of the compact set I.
Bisecting sides of I into 2k k-cells {Qi}2i=1k with equal sizes such that
I =
2k
[
i=1
Qi.
Let I1 ∈ {Qi}2i=1k such that
I1∩ R contains infinitely many elements and diam (I1) = 1
2diam (I).
Next subdivide I1 and continue the process to obtain a sequence of k-cells such that I ⊃ I1 ⊃ I2 ⊃ · · ·
Ij∩ R contains infinitely many elements for each j ∈ N diam (Ij) = 1
2diam (Ij−1) = · · · = 1
2j diam (I) → 0 as j → ∞
For each j ∈ N, since Ij contains infinitely many elements of {pn}, there is a subsequence {nj} such that pnj ∈ Ij∩ R. Next, since lim
j→∞diam (Ij) = 0,
∞
\
j=1
Ij contains exactly one point, say p ∈ Rk. This implies that
{p} =
∞
\
j=1
Ij and 0 ≤ lim
j→∞d(pnj, p) ≤ lim
j→∞diam (Ij) = 0 =⇒ lim
j→∞pnj = p.
Theorem Let X be a metric space and let {pn} be a sequence in X. Suppose that E = {pn | n ∈ N} is the set of points determined by {pn} and E∗ is the set of all subsequential limits of {pn} given by
E∗ = {p ∈ X | ∃ a subsequence {pni} ⊂ {pn} such that lim
i→∞pni = p}.
Then E∗ is closed, i.e. E∗0
⊆ E∗.
Proof
Suppose E = {q1, . . . , qm} contains finitely many points of X.
=⇒ E∗ contains finite number of points
=⇒ E∗0
= ∅ ⊆ E∗
=⇒ E∗ is closed.
Suppose E contains infinitely many points of X and assume that E∗0
6= ∅.
=⇒ Given q ∈ E∗0
, let pn1 ∈ {pn} such that pn1 6= q
=⇒ δ = d(q, pn1) > 0.
Since q ∈ E∗0
=⇒ ∃ x1 ∈ Nδ/2(q) ∩ E∗\ {q} and since x1 ∈ E∗
=⇒ ∃ n2 > n1 such that pn2 ∈ Nδ/2(x1) ∩ E.
=⇒ d(q, pn2) ≤ d(q, x1) + d(x1, pn2) < δ.
Similarly, ∃ x2 ∈ Nδ/22(q) ∩ E∗\ {q} and since x2 ∈ E∗
=⇒ ∃ n3 > n2 > n1 such that pn3 ∈ Nδ/22(x2) ∩ E.
=⇒ d(q, pn3) ≤ d(q, x2) + d(x2, pn3) < δ/2.
q pn1
x1 pn2 δ
δ/2
δ/2
x2 pn3
δ/4 δ/4
For each k ≥ 2, suppose pn1, . . . , pnk are chosen such that d(pnj, q) < δ/2j−2 for each 2 ≤ j ≤ k, then we choose
xk ∈ Nδ/2k(q) ∩ E∗\ {q} and pnk+1 ∈ Nδ/2k(xk) ∩ E
such that
d(q, pnk+1) ≤ d(q, xk) + d(xk, pnk+1) < δ/2k−1.
pnk
xqk
pnk+1 δ/2k−2
δ/2k−1
δ/2k
δ/2k
Since lim
k→∞d(q, pnk) ≤ lim
k→∞δ/2k−2 = 0, {pnk} is a subsequence of {pn} converging to q and hence q ∈ E∗. Since q is an arbitrary point from q ∈ E∗0
, we have shown E∗0
⊆ E∗ and E∗ is closed.
Cauchy Sequences
Question How to tell if {pn} converges if you don’t know the limit already?
Definition The sequence {pn} is a Cauchy sequence if for each > 0, there exists N ∈ N such that if m, n > N then it is implied that d(pm, pn) < .
Definition Let E be a subset of a metric space X = (X, d). The diameter of E is defined by diam E = sup{d(p, q) | p, q ∈ E}.
Remarks
(a) Note that 0 ≤ diam E ≤ ∞.
(b) Let {pn} be a sequence in X and let EN = {pn | n ≥ N }. Then {pn} is a Cauchy sequence if and only if
lim
N →∞diam EN = 0.
(c) Let E be a subset of a metric space X = (X, d). Then diam ¯E = diam E.
Proof Since E ⊆ ¯E, we have diam E ≤ diam ¯E.
Given > 0 and for any p, q ∈ ¯E, since ¯E = E ∪ E0, there exist p0, q0 ∈ E such that d(p, p0) <
2 and d(q, q0) < 2.
This implies that
d(p, q) ≤ d(p, p0) + d(p0, q0) + d(q, q0) ≤
2 + diam E +
2 = diam E + ∀ p, q ∈ ¯E, and
diam ¯E ≤ diam E + =⇒ diam ¯E ≤ diam E by letting → 0+. Hence we have diam ¯E = diam E.
(d) If Kn is a sequence of compact sets in X such that Kn ⊃ Kn+1 for all n ∈ N and if
n→∞lim diam Kn= 0,
then
∞
\
n=1
Kn consists of exactly one point.
Theorem
(a) In any metric space X, every convergent sequence is a Cauchy sequence, i.e. if {pn} ⊂ X converges, then it is Cauchy.
Proof Suppose that lim
n→∞pn= p. Then ∀ > 0 there exists N such that n > N d(pn, p) < 2. So for m, n > N we know
d(pn, pm) ≤ d(pn, p) + d(pm, p)
< 2 +
2
= .
Remark Not every Cauchy sequence converges. To see this, let X = Q. Let pn be the smallest such that m
n > π. Then {pn} is Cauchy, but does not converge in Q (in R converge to π).
(b) If X is a compact metric space and if {pn} is a Cauchy sequence in X, then {pn} converges to some point of X.
Proof Consider the set S defined by
S = {pn| n ∈ N}, i.e. the range set of the map p : N → X defined by p(n) = pn. Case 1 S contains finitely many points, say S = {pn1, . . . , pnm}.
Let
δ = min{d(pni, pnj) | 1 ≤ i < j ≤ m}.
Given > 0 such that < δ, since {pk} is Cauchy, there exists N ∈ N such that if k, l ≥ N then d(pk, pl) < =⇒ if k, l ≥ N then d(pk, pl) < δ.
Suppose ∃ k, l ≥ N such that pk 6= pl, then we must have
δ ≤ d(pk, pl) < δ =⇒ δ < δ which is a contradiction.
Hence we have
∀ k, l ≥ N, pk = pl =⇒ if k ≥ N, then pk = pnj for some 1 ≤ j ≤ m.
Therefore we have
lim
k→∞pk = pnj for some 1 ≤ j ≤ m, i.e. {pk} converges to pnj ∈ X.
Case 2 S contains infinitely many points.
Since X is compact and S is an infinite subset of X,
X is closed, S0 6= ∅ and S0 ⊂ X0 = X =⇒ ∃ p ∈ S0 ⊂ X.
This implies that for each k ∈ N, there exists pnk ∈ B1/k(p) ∩ S \ {p}. Thus we obtain a subsequence {pnk} of {pn} converging to p. This imples that
∀ > 0, ∃ N1 ∈ N such that if k ≥ N1, then d(pnk, p) < .
Also since {pn} is a Cauchy sequence,
with the given > 0, ∃ N2 ∈ N such that if n, m ≥ N2, then d(pn, pm) < .
Let N = max{N1, N2}. Since
d(pn, p) ≤ d(pn, pnn) + d(pnn, p) < 2 ∀ n ≥ N =⇒ lim
n→∞pn = p, i.e. {pn} converges to p ∈ X.
(c) In Rk, every Cauchy sequence converges.
Definition A metric space X is complete if every Cauchy sequence converges to a point in X.
Remarks
(a) Compact metric spaces are complete.
Proof Let {xi} be Cauchy in X. Then {xi} has a convergent subsequence. So there exist {xnk} converging to x ∈ X. Fix > 0. Cauchy implies there exists N ∈ N such that i, j > N then d(xi, xj) < 2. By convergence of {xnk}, there exists N0 ∈ N such that whenever nk> N0, then d(x, xnk) < 2. Let N00 = max (N, N0). Then i, nk> N00 implies
d(x, xi) ≤ d(x, xnk) + d(xnk, xi)
< 2+
2
= .
Hence the sequence itself converges to x and therefore X is complete.
(b) Rk is complete but Q is not.
(c) Let X be the ordered field R. Then X has the leat-upper-bound property if and only if every Cauchy sequence converges to a point in X. Thusan ordered field X is complete if it has the least-upper-bound property.
Proof
(=⇒) Let S be a nonempty, bounded from above subset of X = R. Choose a sufficiently large M > 0 such that
M is an upper bound of S and [−M, M ] ∩ S 6= ∅.
Let I1 = [−M, M ]. Divide I1 into 2 equal length subintervals and let I2 be the closed subinterval such that
the right endpoint of I2 is an upper bound of S and I2∩ S 6= ∅.
Continuing this process, we obtain a sequence of closed intervals {In} such that the right endpoint of In is an upper bound of S, In∩ S 6= ∅ ∀ n ≥ 1.
and
I1 ⊃ I2 ⊃ · · · ⊃ In⊃ · · · ; lim
n→∞|In| = lim
n→∞
M 2n−2 = 0, For each n ∈ N, since In∩ S 6= ∅, let xn be a point in In∩ S.
Since lim
n→∞|In| = lim
n→∞
M
2n−2 = 0, {xn} is a Cauchy sequence and hence it converges to a point, say x, in X = R.
(⇐=) Let {xn} be a Cauchy sequence in X = R. Since {xn} is bounded, there exists x ∈ X = R and a closed interval I = [−M, M] such that
x = sup{xn | n ∈ N} and {xn} ⊂ I.
Since I is compact and {xn} is a Cauchy sequence, we have
n→∞lim xn = x.
Theorem Every metric space (X, d) has a completion (X?, ∆). In other words, (X?, ∆) is a complete metric space containing in X.
Let X? = {Cauchy sequences in X}/ ∼. We will say:
{pn} ∼ {p0n} ⇐⇒ lim
n→∞d(pn, p0n) = 0.
Let P, Q ∈ X?. Then P = [{pn}], and Q = [{p0n}]. We define:
∆(P, Q) = lim
n→∞d(pn, qn).
We claim (X?, ∆) is a complete metric space and (X, d) is isometrically embedded in (X?, ∆).
In other words, there is an injection i : X ,→ X? such that d(p, q) = ∆(i(p), i(q)).
Remarks
(a) This is the other construction of R. But Dedekind cuts are more hardcore.
(b) Example If X = Q, then X? = R. In particular, X? is isometrically isomorphic to R. In other there is a distance preserving bijection.
Definition A sequence {sn} in R is monotonically increasing if sn ≤ sn+1 for all n. Similarly {sn} ismonotonically decreasing if sn ≥ sn+1.
Theorem Suppose {sn} is a monotonic sequence of real numbers. Then {sn} converges if and only if it is bounded.
Proof
(⇐=) Suppose sn ≤ sn+1 (the proof is analogous in the other case).
Let
E = {sn| n ∈ N} = the range of {sn}.
If {sn} is bounded, let
s = sup E = sup{sn| n ∈ N}.
Then
sn ≤ s ∀ n ∈ N.
For every > 0, since s − < s and there is an integer N such that s − < sN ≤ s,
for otherwise s − would be an least-upper-bound of E. Since sn+1≥ sn for all n ∈ N, therefore implies
if n ≥ N, then s − < sN ≤ sn ≤, which shows that {sn} converges to s.
(=⇒) Suppose that s = lim
n→∞sn. By taking = 1, there exists K ∈ N such that if n ≥ K, then |sn| − |s| ≤ |sn− n| < = 1 =⇒ |sn| < 1 + |s| ∀n ≥ K.
Thus
if L = max{|s1|, . . . , |sK−1|, 1 + |s|} =⇒ |sn| ≤ L ∀ n ∈ N.
This proves that {sn} is bounded.
Remarks
(a) Monotone Sequence Property Let F be an ordered field. We say F has the monotone sequence property if every bounded above monotonically increasing sequence converges.
(b) Let X be the ordered field R. Then X has the least-upper-bound property if and only if X has the monotone sequence property. Thus an ordered field X is complete if it has the monotone sequence property.
Proof
(⇐=) Let S be a nonempty, bounded from above subset of X = R. Choose a sufficiently large M > 0 such that
M is an upper bound of S and [−M, M ] ∩ S 6= ∅.
Let I1 = [−M, M ]. Divide I1 into 2 equal length subintervals and let I2 be the closed subinterval such that
the right endpoint of I2 is an upper bound of S and I2∩ S 6= ∅.
Continuing this process, we obtain a sequence of closed intervals {In} such that the right endpoint of In is an upper bound of S, In∩ S 6= ∅ ∀ n ≥ 1.
and
I1 ⊃ I2 ⊃ · · · ⊃ In⊃ · · · ; lim
n→∞|In| = lim
n→∞
M 2n−2 = 0,
For each n ∈ N, by letting xn be the left endpoint of In, we obtain a bounded above monotonically increasing sequence {xn}. Hence there exists x ∈ X = R such that
n→∞lim xn = x.
Since lim
n→∞|In| = lim
n→∞
M
2n−2 = 0 and, for each n ≥ 1, the right endpoint of In is an upper bound of S, x = sup{xn| n ∈ N} ∈ X = R.
(=⇒) Let {xn} be a bounded above monotonically increasing sequence in X = R. Since {xn} is bounded above monotonically increasing, there exists x ∈ X = R such that
n→∞lim xn = x.
Upper and Lower Limits
Definition Let {sn} be a sequence of real numbers.
(a) lim
n→∞sn = ∞ iff ∀ M ∈ R, ∃ N ∈ N such that if n ≥ N then sn ≥ M.
(b) lim
n→∞sn = −∞ iff ∀ M ∈ R, ∃ N ∈ N such that if n ≥ N then sn ≤ M.
Definition 3.16 Let {sn} be a sequence of real numbers and let E be the set of subsequential limits of {sn}defined by
E =n
x ∈ (−∞, ∞) ∪ {±∞} | ∃ {snk} ⊂ {sn} such that lim
k→∞snk = xo . Let
lim sup
n→∞
sn = s∗ = sup E, lim inf
n→∞ sn = s∗ = inf E.
The numbers s∗, s∗ are called the upper and lower limits of {sn}; we use the notation lim sup
n→∞
sn = s∗, lim inf
n→∞ sn= s∗.
Theorem 3.17 Let {sn} be a sequence of real numbers and let E be the set of subsequential limits of {sn}. Then s∗ = sup E and s∗ = inf E have the following properties:
(a) s∗, s∗ ∈ E, i.e. there exist subsequences {snk}, {smj} of {sn} such that lim
k→∞snk = s∗, lim
j→∞smj = s∗. Proof
If s∗ = ∞, then E is not bounded above; hence {sn} is not bounded above, and there is a subsequence {snk} such that lim
k→∞snk = ∞.
If s∗ ∈ R = (−∞, ∞), then E is bounded above; and at least one subsequential limit exists, so that (a) follows from the Weierstrass Theorem.
If s∗ = −∞, then E contains only one element, namely −∞, and there is no subsequential limit. Hence for any M ∈ R, there exist at most a finite number of values of n such that sn > M which implies that lim
n→∞= −∞.
(b) If x > s∗ ≥ s∗ > y, there exists an N ∈ N such that n ≥ N implies x > sn > y, i.e. for each
> 0, there exists an N ∈ N such that
if n ≥ N then s∗− < sn< s∗+ .
s∗+ s∗ −
Proof For each > 0, if there are infinitely many sn’s lying in (−∞, s∗− ] ∪ [s∗+ , ∞), then there exists a subsequence {snk} lying completely in (−∞, s∗− ] or [s∗+ , ∞).
This implies that there exists a subsequential limit x ∈ (−∞, s∗− ] ∪ [s∗+ , ∞) ∪ {±∞}
which is a contradiction to the definition of s∗ = inf E or that of s∗ = sup E.
Hence there are only finite number of xn’s lying in (−∞, s∗− ] ∪ [s∗+ , ∞).
Remarks
(a) If k ≥ l ∈ N, then {sm | m ≥ k} ⊆ {sm | m ≥ l} and
inf{sm | m ≥ l} ≤ inf{sm | m ≥ k} ≤ sup{sm | m ≥ k} ≤ sup{sm | m ≥ l},
i.e. sup{sm | m ≥ k} is monotone decreasing in k, and inf{sm | m ≥ k} is monotone increasing in k.
(b) If s∗, s∗ ∈ E, and {snk}, {smj} are subsequences of {sn} such that
k→∞lim snk = s∗, lim
j→∞smj = s∗ =⇒ lim inf
k→∞ snk = s∗ = lim sup
k→∞
snk, lim inf
j→∞ smj = s∗ = lim sup
j→∞
smj. For each > 0, since there only exist finite number of n ∈ N such that sn > s∗ + or sn < s∗− , hence we have
s∗− ≤ inf{sm | m ≥ n} ≤ sup{sm | m ≥ n} ≤ s∗+ for sufficiently large n
=⇒ s∗− ≤ lim
n→∞inf{sm | m ≥ n} ≤ lim
n→∞sup{sm | m ≥ n} ≤ s∗+ ∀ > 0
=⇒ lim
→0+(s∗− ) ≤ lim
n→∞inf{sm | m ≥ n} ≤ lim
n→∞sup{sm | m ≥ n} ≤ lim
→0+(s∗+ )
=⇒ s∗ ≤ lim
n→∞inf{sm | m ≥ n} ≤ lim
n→∞sup{sm | m ≥ n} ≤ s∗.
On the other hand for each l ∈ N, since {smj | j ≥ l} and {snk | k ≥ l} are subsequences of {sn | n ≥ ml} and {sm | m ≥ nl} respectively, we have
inf{sn| n ≥ ml} ≤ smj and snk ≤ sup{sm | m ≥ nl} ∀ j, k ≥ l
=⇒ inf{sn| n ≥ ml} ≤ lim
j→∞smj = s∗ and s∗ = lim
k→∞snk ≤ sup{sm | m ≥ nl} ∀ l ∈ N
=⇒ inf{sn| n ≥ ml} ≤ s∗ ≤ s∗ ≤ sup{sm | m ≥ nl} ∀ l ∈ N
=⇒ lim
l→∞inf{sn | n ≥ ml} ≤ s∗ ≤ s∗ ≤ lim
l→∞sup{sm | m ≥ nl}
=⇒ lim
n→∞inf{sm | m ≥ n} ≤ s∗ ≤ s∗ ≤ lim
n→∞sup{sm | m ≥ n}.
Hence
lim sup
n→∞
sn= s∗ = lim
n→∞sup{sm | m ≥ n} = inf{sup{sm | m ≥ n} | n ∈ N}, lim inf
n→∞ sn= s∗ = lim
n→∞inf{sm | m ≥ n} = sup{inf{sm| m ≥ n} | n ∈ N}.
(c) If x > s∗ = lim
n→∞sup{sm | m ≥ n} and lim
n→∞inf{sm | m ≥ n} = s∗ > y, there exists an N ∈ N depending on x − s∗ > 0 and s∗ − y > 0 such that if n ≥ N then
max{| sup{sm | m ≥ n} − s∗|, |s∗− inf{sm | m ≥ n}|} < min{x − s∗, s∗− y} ∀ n ≥ N
=⇒ sup{sm | m ≥ n} − s∗ < x − s∗ and s∗− inf{sm | m ≥ n} < s∗− y ∀ n ≥ N
=⇒ sn≤ sup{sm | m ≥ n} < x and sn≥ inf{sm | m ≥ n} > y ∀ n ≥ N
=⇒ y < inf{sm | m ≥ n} ≤ sn≤ sup{sm | m ≥ n} < x ∀ n ≥ N.
Examples
(a) Let {sn} = Q, be a sequence containing all rationals. Then every real number is a subse- quential limit, and
lim sup
n→∞
sn= ∞, lim inf
n→∞ sn = −∞, i.e. the set of subsequential limits of {sn} is E = R ∪ {±∞}.
(b) Let sn= (−1)n
1 + (1/n). Then
lim sup
n→∞
sn = 1, lim inf
n→∞ sn = −1.
(c) For a real-valued sequence {sn},
n→∞lim sn = s ⇐⇒ lim sup
n→∞
sn = lim inf
n→∞ sn = s.
Theorem If sn ≤ tn for n ≥ N, where N is fixed, then lim inf
n→∞ sn≤ lim inf
n→∞ tn, lim sup
n→∞
sn ≤ lim sup
n→∞
tn.
Some Special Sequences
Theorem
(a) If p > 0, then lim
n→∞
1 np = 0.
Proof For each > 0, there exists n ∈ N such that n > 1
1/p ⇐⇒ np > 1
⇐⇒ > 1 np = | 1
np − 0|.
(b) p > 0, then lim
n→∞
√n
p = 1.
Proof If p > 1, by letting xn = √n
p − 1 > 0, then 1 + n xn≤ (1 + xn)n= p =⇒ xn ≤ p − 1
n =⇒ lim
n→∞xn = 0 ⇐⇒ lim
n→∞
√n
p = 1.
If p < 1, by letting p = 1
q for some q > 1, we have
n→∞lim
√n
q = 1 =⇒ lim
n→∞
1
√n
q = 1 =⇒ lim
n→∞
√n
p = lim
n→∞
1
√n
q = 1.
(c) lim
n→∞
√n
n = 1.
Proof Let xn = √n
n − 1. Then xn≥ 0 for each n ∈ N and n = (1 + xn)n≥ n(n − 1)
2 x2n =⇒ 0 ≤ xn≤
r 2
n − 1 for n ≥ 2 =⇒ lim
n→∞xn = 0.
(d) If p > 0 and α ∈ R, then lim
n→∞
nα
(1 + p)n = 0.
Proof Let k ∈ N such that k > α. If n ∈ N and n
2 > k, then (1 + p)n>n
k
pk = n(n − 1) · · · (n − k + 1)
k! pk> nkpk
2kk! =⇒ 0 < nα
(1 + p)n < 2kk!
pk nα−k. (e) If |x| < 1, then lim
n→∞xn= 0.
Proof Take α = 0 and p > 0 such that |x| = 1
1 + p as in (d).
Series
Question What does this mean?
1 + 1 4+ 1
9+ 1 16+ 1
25+ · · · = π2 6 .
Remark Sum of natural numbers to negative even powers always has a nice form.
Consider also:
1 − 1 + 1 − 1 + 1 − 1 + . . .
You can associate these differently and get different limits. This tells us we cannot assume associativity in infinite sums.
Notation Let {an} be a real sequence. Then:
j
X
n=i
an = ai+ ai+1+ · · · + aj, when i < j ∈ N.
Definition The nth partial sum of {ak} is sn=
n
X
k=1
ak.
Remark {sn} is a real sequence. Sometimes {sn} is called an infinite series.
Definition If sn → s we write
∞
X
n=1
an = lim
n→∞
n
X
k=1
ak = s
Question When does an infinite series converge? When its sequence of partial sums converge.
Example Does
∞
X
n=1
1
n converge?
Consider partial sums where sn=
n
X
k=1
1
k. We use the Cauchy criterion. For m < n, then d(sm, sn) = sn− sm
= sm+1+ sm+2+ · · · + sn
= 1
m + 1 + 1
m + 2 + · · · + 1
n ≥ n − m n .
The inequality comes from observing that all the terms in the sum are less than or equal to 1 n. Thus s2n − sn ≥ 2n − n
2n = 1
2. Therefore this sequence is not Cauchy. Hence {sn} does not converge.
Cauchy Criterion for series X
an converges if and only if for all > 0 there exists N ∈ N such that m ≥ n > N implies
m
X
k=n
ak
< .
Corollary (Divergence Test) If X
an converges, then lim
n→∞an = 0.
Remarks (a) X 1
n is called the harmonic series.
(b) The corollary’s converse is not true (counter-example is harmonic series).
Theorem If an ≥ 0, then X
an converges if and only if partial sums {sn} form a bounded sequence.
Proof Since ak ≥ 0 for all k ∈ N, {sn} is monotonically increasing. Thus {sn} converges, i.e.
Xan converges, if and only if {sn} is bounded.
Comparison Test 1. If X
cn converges and |an| ≤ cn for almost all n, then X
an converges.
2. If X
dn diverges to +∞ and an ≥ dn for almost all n, then X
an diverges as well.
Proof
1. Let > 0. Since X
cn converges, it satisfies Cauchy Criterion. Thus there exists N ∈ N such that m ≥ n ≥ N implies:
m
X
k=n
ck ≤
m
X
k=n
ck
< .
Thus
m
X
k=n
ak
≤
m
X
k=n
|ak| ≤
m
X
k=n
ck.
This follows from the assumption that |an| ≤ cn for almost all n so let N be at least larger than the last n for which cn < |an|. The resulting inequality satisfies the Cauchy Criterion and thusX
an converges.
2. Follows from (a) via contrapositive. (Also, partial sums form a bounded sequence.)
Geometric Series If |x| < 1, then
∞
X
n=0
xn= 1 1 − x. If |x| ≥ 1, then X
xn diverges.
Proof If x 6= 1, let sn =
n
X
k=0
xk = 1 + x + · · · + xn. Then sn = 1 − xn+1
1 − x by multiplying both sides by 1 − x. Thus it follows:
n→∞lim sn = 1
1 − x if |x| < 1
If |x| > 1, then {sn} does not converge. Similarly if x = ±1, use the divergence test to verify {sn} does not converge.
Example
∞
X
n=0
1
n! converges. Notice
sn= 1 + 1 + 1 2! + 1
3! + 1
4! + · · · + 1 n!
≤ 1 + 1 + 1 2+ 1
22 + 1
23 + · · · + 1 2n−1
< 3.
Thus it is bounded and since each term is nonnegative, it is monotonically increasing. Thus it converges.
Definition e =
∞
X
n=0
1 n!. Theorem e = lim
n→∞(1 + 1 n)n. Proof
Let sn=
n
X
k=0
1
k!, tn = 1 + 1 n
n
. By the binomial theorem,
tn= 1 + 1 + 1
2! 1 − 1 n + 1
3! 1 − 1 n
1 − 2
n + · · · + 1
n! 1 − 1 n
1 − 2
n · · · 1 − n − 1 n .
Hence tn≤ sn ≤ e for all n ∈ N, so that
lim sup
n→∞
tn≤ e.
Next, if n ≥ m, then
tn≥ 1 + 1 + 1
2! 1 − 1
n + · · · + 1
m! 1 − 1 n
1 − 2
n · · · 1 − m − 1 n .
By keeping m fixed and letting n → ∞ on both sides, we get lim inf
n→∞ tn ≥ 1 + 1 + 1
2! + · · · + 1
m! =⇒ lim inf
n→∞ tn ≥ sm ∀ m ∈ N.
Letting m → ∞, we finally get e ≤ lim inf
n→∞ tn ≤ lim sup
n→∞
tn≤ e =⇒ e = lim
n→∞tn = lim
n→∞(1 + 1 n)n.
Remark
e − sn = 1
(n + 1)! + 1
(n + 2)! + . . .
< 1
(n + 1)!(1 + 1
n + 1 + 1
(n + 1)2 + . . . )
= 1
(n + 1)!
1
1 − 1/(n + 1)
= 1 n! n
Theorem e 6∈ Q.
Proof Suppose e = m
n for m, n > 0. Then
0 < n! (e − sn)
| {z }
∈N
< 1 n < 1
Recall A series converges if its partial sums converges.
Example 1 + 1 2+ 1
4+ 1 8+ 1
16+ . . . converges because partial sums converge.
Cauchy’s Theorem If a1 ≥ a2 ≥ a2· · · ≥ 0, i.e. monotonically decreasing, then
∞
X
n=1
an con-
verges if and only if
∞
X
k=0
2ka2k = a1+ 2a2+ 4a4 + 8a8+ . . . converges.
Proof Compare sn = a1 + · · · + an and tk = a1 + 2a2 + · · · + 2ka2k. Consider the following grouping of the finite sum:
sn = (a1) + (a2+ a3) + (a4 + · · · + a7) + · · · + an
tk = (a1) + (a2+ a2) + (a4 + · · · + a4) + · · · + (a2k + · · · + a2k)
If tk converges, since sn < tk ∀ n < 2k and ∀ k ∈ N,
=⇒ sn≤ lim
k→∞tk ∀ n ∈ N
=⇒ sn converges since sn is monotonically increasing and bounded above.
On the other hand,
if sn converges, since 2sn> tk− a1 ∀ n ≥ 3 and ∀ 2k < n,
=⇒ 2 lim
n→∞sn ≥ tk− a1 ∀ k ∈ N
=⇒ tk converges since tk− a1 is monotonically increasing and bounded above.
This proves that both series diverge or converge simultaneously.
Theorem Consider P 1
np. Claim is that this converges if p > 1 and diverges if p ≤ 1.
Proof If p ≤ 0, terms do not go to zero, so the series diverges. If p > 0, look at:
X
k
2k 1
(2k)p =X
k
2(1−p)k,
which is geometric. Thus it converges if and only if 21−p< 1. This only happens when 1 − p < 0 and hence p > 1.
Remark We were able to turn a harmonic-like series into a geometric-like series.
Theorem (Root Test) Given P an, let α = lim
n→∞
p|an n|. Then (a) if α < 1, then P an converges;
(b) if α > 1, then P an diverges;
(c) if α = 1, then the test is inconclusive.
Proof If α < 1, we can choose β so that α < β < 1, and an integer N such that p|an n| < β for n ≥ N, by Theorem 3.17(b).
That is, if n ≥ N, then |an| < βn. Since 0 < β < 1, P βn converges. Therefore, P an converges (absolutely) by the comparison test.
If α > 1, then, again by Theorem 3.17(b), there exists a sequence {nk} such that
nk
q
|ank| → α > 1.
Hence |an| > 1 for infinitely many values of n, and
n→∞lim an6= 0 =⇒ X
an diverges.
Since P 1
n diverges, P 1
n2 converges and α = 1 in both cases, the test is inconclusive if α = 1.
Theorem (Ratio Test) The series P an
(a) converges if lim sup
an+1
an < 1, (b) diverges if
an+1
an
≥ 1 for n ≥ n0, where n0 is some fixed integer.
Proof If lim sup
an+1 an
< 1, there exist β < 1 and an integer N such that
an+1
an
< β < 1 for n ≥ N.
This implies that
|aN +k| < β|aN +k−1| < β2|aN +k−2| < · · · < βk|an| for each k ∈ N.
Therefore,P an converges (absolutely) by the comparison test.
If |an+1| ≥ |an| for n ≥ n0, then
n→∞lim an6= 0 =⇒ X
an diverges.
Definition A power series is of the form
∞
X
k=0
cnzn where cn ∈ C.
Theorem Let α = lim supp|cn n|. Let r = α1. Then the power series converges if |z| < R and diverges |z| > R. We call R the radius of convergence.
Proof Use the root test so consider lim sup
n→∞
p|cn nzn| = lim sup
n→∞
|z|p|cn n| = |z| lim sup
n→∞
p|cn n|.
Notice that this less than 1, and thus converges, when |z| is less than 1 over the limsup.
Definition A series converges absolutely if P |an| converges.
Theorem P an converges absolutely implies P an converges.
Proof |
m
X
k=n
ak| ≤
m
X
k=n
|ak| < , by the Cauchy Criterion since P |ak| converges.
Example If a series converges, it does not necessarily converge absolutely. Consider the series 1 − 12+13 −14 + . . . , which converges by alternating series test but does not converge absolutely.
Question If the terms in a convergent series are rearranged, must it converge to same sum? Not all the time, but it does if the series converges absolutely.
Riemann’s Theorem If a series P an converges but not absolutely, then we can form a rear- rangement that has any limsup and liminf you’d like.
Example Rearrange 1 − 12 + 13 − 14 + . . . to converge to 4. We want to the partial sums to converge to 4. Consider the positive terms in sequence that sum to at least 4 (notice the positive terms diverge). Then use a s many negative terms you need to go back, as many positive terms you need to go forward, et cetera. These partial sums converge to 4.