MATH 1023/24 Lecture Notes

Chapter 1 - Sequences

Hong Kong University of Science and Technology

July 29, 2023

1 - Limit of convergent sequences

A sequence {𝑥𝑛 } converges if there exists a real number 𝐿 ∈ (−∞, +∞) such
that |𝑥𝑛 − 𝐿| can be arbitrarily small for sufficiently large 𝑛. In other words,
for any 𝜖 > 0, there exists 𝑁 > 0 such that |𝑥𝑛 − 𝐿| < 𝜖 whenever 𝑛 > 𝑁 . The
real number 𝐿 is called the limit of {𝑥𝑛 }.
Example 1.1. Suppose 𝑥𝑛 → 𝐴. Then since |𝑥𝑛 |−|𝑥𝑛 − 𝐴| ≤ |𝐴| ≤ |𝐴 − 𝑥𝑛 |+
|𝑥𝑛 | ⟹ |𝐴 − 𝑥𝑛 | ≥ ||𝐴| − |𝑥𝑛 ||, we have |𝑥𝑛 | → |𝐴|.
The argument used in the proof |𝑎−𝑏| ≥ ||𝑎| − |𝑏|| is a corollary of the triangular
inequality and is often used to bound expressions from below.
Example 1.2. Suppose 𝑥𝑛 → 𝐴. Show that

𝑥1 + 𝑥 2 + … + 𝑥 𝑛
→ 𝐴.
𝑛

Proof. Suppose 𝑥𝑛 ∈ (𝐴 − 𝜖, 𝐴 + 𝜖) for 𝑛 > 𝑁 . Then, we have

𝑥1 + … + 𝑥 𝑛 𝑥 + … + 𝑥𝑁 𝑥𝑁+1 + … + 𝑥𝑛 𝑆 𝑛−𝑁 𝑆 𝑛−𝑁

= 1 + ∈( 𝑁 + ⋅ (𝐴 − 𝜖), 𝑁 + ⋅ (𝐴 + 𝜖)) .
𝑛 𝑛 𝑛 𝑛 𝑛 𝑛 𝑛

Choose 𝑛 sufficiently large (how large?), then we will have

𝑥1 + … + 𝑥 𝑛 𝜖 𝜖 𝜖2 𝜖2
∈ (𝜖 + (1 − ) (𝐴 − 𝜖), 𝜖 + (1 − ) (𝐴 + 𝜖)) = (𝐴 − 𝜖 + , 𝐴 + 𝜖 − ) .
𝑛 𝐴 𝐴 𝐴 𝐴

This is a subset of (𝐴 − 𝜖, 𝐴 + 𝜖) for small enough 𝜖, concluding our proof.

Using the definition of limit, it is not hard to verify that

1
Proposition 1.3. The limit of a sequence is unique, and limits preserves the
four arithmetic operations (in particular, addition and multiplication).
Proof. For the readers to complete.
Example 1.3.1. We first present a simple but powerful result in proving results
related to polynomials or in which binomial theorem is involved:

𝑛𝑘 (𝑘 + 1)!𝑛𝑘 (𝑘 + 1)! 𝑛 𝑛 𝑛
𝑛 = = ⋅ ⋅ ⋅⋯⋅ → 0.
(𝑘+1 ) 𝑛 ⋅ (𝑛 − 1) ⋅ ⋯ ⋅ (𝑛 − 𝑘) 𝑛 𝑛 𝑛−1 𝑛−𝑘

Example 1.3.2. Suppose {𝑥𝑛 } is such that

𝑥1 + 𝑥 2 + … + 𝑥 𝑛
→ 𝐿 ∈ (−∞, +∞).
𝑛

Let 𝑆𝑛 = 𝑥1 + … + 𝑥𝑛 . Note that

𝑥𝑛+1 𝑆 𝑛 𝑆
= 𝑛+1 − ⋅ 𝑛 → 𝐿 − 1 ⋅ 𝐿 = 0.
𝑛+1 𝑛+1 𝑛+1 𝑛

Example 1.3.3. Suppose |𝑥| < 1, then

∞ ∞
𝑘−1 𝑛 𝑛−1 1
(1 − 𝑥) ∏ (1 + 𝑥2 ) = lim (1 − 𝑥2 ) = 1 ⟹ ∏ (1 + 𝑥2 )= .
𝑘=1
𝑛→∞
𝑛=1
1−𝑥

Example 1.3.4. Suppose lim𝑛→∞ (𝑥𝑛 − 𝑥𝑛−1 ) = 𝐴 ≠ 𝑘𝜋. Suppose that sin 𝑥𝑛 →
𝐿. Using the rotation matrix by the angle of 𝑥𝑛 −𝑥𝑛−1 (or equivalently trigonom-
etry identities), we get the following identities

′
sin 𝑥𝑛+1 = sin (𝑥𝑛+1 − 𝑥𝑛 ) cos 𝑥𝑛 +cos (𝑥𝑛+1 − 𝑥𝑛 ) sin 𝑥𝑛 . → 𝐿 = sin 𝐴⋅𝐿 +cos 𝐴⋅𝐿.

′
sin 𝑥𝑛−1 = sin 𝑥𝑛 cos (𝑥𝑛 − 𝑥𝑛−1 )−cos 𝑥𝑛 sin (𝑥𝑛 − 𝑥𝑛−1 ) . → 𝐿 = 𝐿⋅cos 𝐴−𝐿 ⋅sin 𝐴.

′ ′2
Since sin 𝑥𝑛 → 𝐿, we have cos 𝑥𝑛 → 𝐿 such that 𝐿2 + 𝐿 = 1. Substituting
the limits into the above identities, we get (by assuming the continuity of sine
and cosine functions)

′
𝐿 (1 − cos 𝐴) = 0; 𝐿 ⋅ sin 𝐴 = 0.

2
 ′
Since sin 𝐴 ≠ 0, then 1 − cos 𝐴 ≠ 0 and hence 𝐿 = 𝐿 = 0, which is impossible.
Therefore, sin 𝑥𝑛 must diverge whenever 𝐴 ≠ 𝑘𝜋.
Example 1.3.5. Suppose

1
𝑥𝑛+1 = 1 + , 𝑥 = 1.
1 + 𝑥𝑛 1

If the sequence {𝑥𝑛 } converges to a real number 𝐿, then 𝐿 > 0, and hence

1 √
𝐿=1+ ⟹ 𝐿 = 2.
1+𝐿
√
Let 𝜆𝑛 = ∣𝑥𝑛 − 2∣. Then

𝜆𝑛+1 √ 1
= ( 2 − 1) ⋅ .
𝜆𝑛 𝑥𝑛 + 1
√
We can then use induction to show that 𝑥2𝑛−1 < 𝑥2𝑛+1 < 2 < 𝑥2𝑛 < 𝑥2𝑛−2 .
This step is a standard but crucial step. This result implies 1 ≤ 𝑥𝑛 ≤ 32 , which
gives

√ 𝑛
2−1
𝜆𝑛 ≤ 2 ⋅ ( ) → 0.
2
√
Therefore, 𝑥𝑛 → 2, and we also estimated the speed of its convergence.
Proposition 1.4. Suppose 𝑎𝑛 ≤ 𝑏𝑛 ≤ 𝑐𝑛 for sufficiently large 𝑛, and 𝑎𝑛 , 𝑐𝑛 → 𝐴.
Then for sufficiently large 𝑛,

𝐴 − 𝜖 ≤ 𝑎𝑛 ≤ 𝑏𝑛 ≤ 𝑐𝑛 ≤ 𝐴 + 𝜖.

Therefore 𝑏𝑛 → 𝐴. The key is to use various techniques and tricks to find and
construct the binding convergent sequences {𝑎𝑛 } , {𝑐𝑛 }.
Example 1.4.1. Note that

1! + 2! + … + 𝑛! < (𝑛 − 2)(𝑛 − 2)! + (𝑛 − 1)! + 𝑛! < 2(𝑛 − 1)! + 𝑛!.

Therefore, we have

1! + 2! + … + 𝑛! 2(𝑛 − 1)! + 𝑛! 2
1< < = + 1.
𝑛! 𝑛! 𝑛
(1!+2!+…+𝑛!)
By squeeze theorem, we get 𝑛! → 1.

3
Example 1.4.2. Suppose 𝑝(𝑛) is a polynomial of 𝑛 and 0 < 𝑎 < 1. Using the
result 1.3.1, we can show that 𝑝(𝑛)𝑎𝑛 → 0. Suppose 𝑎−1 = 1 + 𝐴, 𝐴 > 0, and
|𝑝(𝑛)| < 𝑛𝑘 for some large enough integer 𝑘. Note that

𝑛 𝑛𝑘
(1 + 𝐴)𝑛 > ( )𝐴𝑘+1 ⟹ 0 ≤ |𝑝(𝑛)| 𝑎𝑛 < 𝐴−(𝑘+1) ⋅ 𝑛 → 0.
𝑘+1 (𝑘+1)

Therefore, we get |𝑝(𝑛)| 𝑎𝑛 → 0.

Example 1.4.3. Suppose further that the leading coefficient is positive. Using a
similar argument and 1.3.1, we can show that √𝑝(𝑛)
𝑛
→ 1.
Proposition 1.5. Suppose 𝑥𝑛 → 𝐴. If 𝑏 < 𝐴 < 𝐵, then for sufficiently large 𝑛,
𝑏 < 𝑥𝑛 < 𝐵. Conversely, if for sufficiently large 𝑛, 𝑏 < 𝑥𝑛 < 𝐵, then 𝑏 ≤ 𝐴 ≤ 𝐵.
Example 1.5.1. It follows naturally that a convergent sequence is bounded.
√
Example 1.5.2. Suppose 𝑞 > 1. For sufficiently large 𝑛, 𝑛 𝑛 < 𝑞 𝜖 . Then
log𝑞 𝑛 log𝑞 𝑛
𝑛 < 𝜖. Therefore, 𝑛 → 0.
∞ ∞
A subsequence {𝑥𝑛𝑘 } of {𝑥𝑛 }𝑛=1 is a sequence, whose terms are from
𝑘=1
∞
{𝑥𝑛 }𝑛=1 and are placed in the original order.
∞
The set of all the limits of convergent subsequences of a sequence {𝑥𝑛 }𝑛=1 is
denoted as 𝐿𝐼𝑀 {𝑥𝑛 }.
Proposition 1.6. If a sequence {𝑥𝑛 } converges to 𝐿, then every subsequence
converges, and 𝐿𝐼𝑀 {𝑥𝑛 } = {𝐿}. The converse is obviously true. In fact, if
{𝑥2𝑛 } and {𝑥2𝑛−1 } converges to the same 𝐿, then 𝑥𝑛 → 𝐿.
Suppose 𝑋 is an open interval, and 𝐷 ⊂ 𝑋. 𝐷 is called dense in 𝑋 if for any
(𝑎, 𝑏) ⊂ 𝑋, there exists 𝑟 ∈ 𝐷 such that 𝑟 ∈ (𝑎, 𝑏).
Proposition 1.7. 𝐷 is dense in 𝑋 if and only if for any 𝑥 ∈ 𝑋, there is a
sequence {𝑟𝑛 } in 𝐷 such that 𝑟𝑛 → 𝑥.
Proof. Suppose 𝐷 is dense in 𝑋. Then for any 𝑥 ∈ 𝑋, we have (𝑥 − 𝜖, 𝑥 + 𝜖) ⊂ 𝑋
for some 𝜖 > 0. Choose 𝜖𝑛 → 0. By density, there exists 𝑟𝑛 ∈ (𝑥 − 𝜖𝑛 , 𝑥 + 𝜖𝑛 ).
Hence 𝑟𝑛 → 𝑥.
Suppose for any 𝑥 ∈ 𝑋, there is a sequence {𝑟𝑛 } in 𝐷 such that 𝑟𝑛 → 𝑥. Then
for any (𝑎, 𝑏) ⊂ 𝑋, choose 𝑟𝑛 → 21 (𝑎 + 𝑏). Then by proposition 1.4, 𝑟𝑛 ∈ (𝑎, 𝑏)
for sufficiently large 𝑛.
Example 1.8. Suppose 𝑎 < 𝑏. Then find a sufficiently large 𝑁 such that
|𝑎−𝑏|𝑁 > 1. Then, there exists an integer 𝑀 ∈ (𝑎𝑁 , 𝑏𝑁 ). Therefore, 𝑀
𝑁 ∈ (𝑎, 𝑏).
This is the density of rational numbers.
Example 1.9. Suppose 𝛼 > 0 is an irrational number. Consider the set

4
 𝑍𝛼 = {𝑛𝛼 − [𝑛𝛼] ∶ 𝑛 ∈ ℤ} .

Here, we also try to find a maximum distance as in 1.8, using pigeonhole prin-
ciple. Let 𝑚 be a positive integer. Then, there must exist distinct 𝑖 < 𝑗 ∈
{1, 2, … , 𝑚 + 1} such that, for some 𝑘 ∈ {1, 2, … , 𝑚}, we have

𝑘−1 𝑘
≤ 𝑖𝛼 − [𝑖𝛼], 𝑗𝛼 − [𝑗𝛼] < .
𝑚 𝑚

This implies that any two elements from the subset of 𝑍𝛼 , 𝑍𝛼𝑚 =
{𝑛 ((𝑗 − 𝑖)𝛼 − [(𝑗 − 𝑖)𝛼]) ∶ 𝑛 ∈ ℤ}, are less than 𝑚−1 apart in distance.
Suppose 0 ≤ 𝑐 < 𝑑 < 1, then we build a corresponding 𝑍𝛼𝑚 using a suitable,
large enough 𝑚 such that 𝑚−1 < |𝑑 − 𝑐|. Then, there exists 𝛽 ∈ 𝑍𝛼𝑚 ⊂ 𝑍𝛼 such
that 𝛽 ∈ (𝑐, 𝑑). This shows the density of 𝑍𝛼 in [0, 1).
Example 1.9.1. Consider the sequence {sin(𝑛)}. We claim that 𝐿𝐼𝑀 {sin(𝑛)} =
[−1, 1].
This directly follows from the above example. To see why, draw a unit circle
𝜔. Observe that it is sufficient (by using the continuity of sine function) to
show that for any radius vector v pointing from the origin 𝑂 to a point on 𝜔,
there is a sequence of radius vectors corresponding to the sequence {sin(𝑛)}
𝑛
that approaches it. In rigorous language, the set { 2𝜋 𝑚𝑜𝑑 1} is dense in [0, 1)
(number of cycles), which is given by 1.8.

2 - Monotone sequences, the number 𝑒

Theorem 2.1. (*) Bounded, (ultimately) monotone sequences converge.
Example 2.1.1. Consider the harmonic series 𝐻𝑛 . Suppose 𝑎𝑛 = 𝐻2𝑛 − 𝐻𝑛 .
𝑛
Then 𝑎𝑛+1 > 𝑎𝑛 , while 𝑎𝑛 < 𝑛+1 < 1. Therefore, the sequence {𝑎𝑛 } converges.
Example 2.1.2. Suppose

2
(2𝑛)!! 1
𝑎𝑛 = [ ] ⋅ .
(2𝑛 − 1)!! 2𝑛 + 1

Then easy to verify 𝑎𝑛+1 > 𝑎𝑛 . Also,

2⋅4 4⋅6 (2𝑛 − 2) ⋅ 2𝑛 2𝑛

𝑎𝑛 = 2 ⋅ ⋅ 2 ⋅⋯⋅ ⋅ < 2.
32 5 (2𝑛 − 1)2 2𝑛 + 1

Therefore, the sequence converges. This is the sequence in the Wallis formula.
Example 2.1.3. Suppose

5
 𝑚
1
𝑆𝑚 = ∑ 𝑝
, 𝑝 > 1.
𝑛=1
𝑛

Clearly, 𝑆𝑚+1 > 𝑆𝑚 . Also, note that

1 1 1 1
𝑆2𝑚 −1 ≤ 1 + 𝑝
⋅ 2 + 𝑝 ⋅ 4 + ⋯ + (𝑚−1)𝑝 ⋅ 2𝑚−1 = 1 + 𝑟 + 𝑟2 + … + 𝑟𝑚−1 < .
2 4 2 1−𝑟

Since 𝑆𝑚 ≤ 𝑆2𝑚 −1 , we have 𝑆𝑚 must converge.

Example 2.1.4. Suppose

1 𝑛
𝑒𝑛 = (1 + ) .
𝑛

First, we apply the AM-GM inequality to the sequence and get

𝑛+1
1 𝑛 𝑛 (1 + 𝑛1 ) + 1 1 𝑛+1
𝑒𝑛 = 1 ⋅ (1 + ) < [ ] = (1 + ) = 𝑒𝑛+1 .
𝑛 𝑛+1 𝑛+1

Also, by using the binomial theorem,

𝑛 𝑛 𝑛
𝑛 1 𝑛! 𝑛(𝑛 − 1) ⋯ (𝑛 − 𝑗 + 1)
𝑒𝑛 = ∑ ( ) 𝑗 = ∑ 𝑗
= 2 + ∑ .
𝑗=0
𝑗 𝑛 𝑗=0
(𝑛 − 𝑗)!𝑗!𝑛 𝑗=2
𝑗!𝑛𝑗

Note that

𝑛 𝑛 𝑛 𝑛
𝑛(𝑛 − 1) ⋯ (𝑛 − 𝑗 + 1) 1 1 𝑗−1 1 1
∑ = ∑ (1 − ) ⋯ (1 − ) < ∑ < ∑ < 1.
𝑗=2
𝑗!𝑛𝑗 𝑗=2
𝑗! 𝑛 𝑛 𝑗=2
𝑗! 𝑗=2
𝑗(𝑗 − 1)

Therefore, since 𝑒𝑛 is monotone increasing and bounded from above, it converges.

We denote

1 𝑛
𝑒 ≔ lim 𝑒𝑛 = lim (1 + ) .
𝑛→∞ 𝑛→∞ 𝑛

The above limit is also called the natural constant.

Example 2.1.5. Both {𝑥2𝑛 } and {𝑥2𝑛−1 } in 1.3.4 are monotone sequences, and
hence converges. If they converge to the same limit, then by proposition 1.6,
the sequence {𝑥𝑛 } converges.

6
To see this, we only need to see that both {𝑥2𝑛 } and {𝑥2𝑛−1 } follow the same
difference equation,√and hence they have the same limit. Therefore, the sequence
{𝑥𝑛 } converges to 2.
Example 2.1.6. Suppose 𝑎1 > 𝑏1 > 0, and

2𝑎𝑛 𝑏𝑛
𝑎𝑛+1 = ,𝑏 = √𝑎𝑛+1 𝑏𝑛 .
𝑎𝑛 + 𝑏𝑛 𝑛+1

Using induction, it is easy to see that 𝑏1 < 𝑏2 < 𝑎2 < 𝑎1 ⟹ 𝑏𝑛 < 𝑏𝑛+1 <
𝑎𝑛+1 < 𝑎𝑛 . Suppose 𝑎𝑛 → 𝐴, 𝑏𝑛 → 𝐵, then

2𝐴𝐵
𝐴= , 𝐴𝐵 ≠ 0 ⟹ 𝐴 = 𝐵.
𝐴+𝐵

Example 2.1.4 introduces us to the constant 𝑒. It is of crucial importance in

analysis, and many limits can be explicitly computed using this result.
Now we give some first properties of 𝑒. We first show that
Proposition 2.2.

∞
1 1 1
𝑒=∑ = 1 + 1 + + + ….
𝑛=0
𝑛! 2! 3!

First, using theorem 2.1, the series on RHS converges. Now we prove the equal-
ity.
We notice that the terms are kind of familiar to us, as they appear in our
derivation of the upper bound for 𝑒 in 2.1.4. To show the above proposition,
use that

1 𝑛 𝑚
1 1 𝑘−1 1 𝑚 𝑚
1 1 𝑘−1
(1 + ) > 2+∑ (1 − ) ⋯ (1 − ), (1 + ) < 2+∑ (1 − ) ⋯ (1 − ).
𝑛 𝑘=2
𝑘! 𝑛 𝑛 𝑚 𝑘=2
𝑘! 𝑛 𝑛

Using proposition 1.5, by pushing 𝑛 → ∞, the first inequality shows that for
𝑚 1
any given 𝑚, 𝑒 ≥ ∑𝑘=0 𝑘! . By letting 𝑚 → ∞ and apply 1.5 again, we get
∞ 1
𝑒 ≥ ∑𝑘=0 𝑘! .
1𝑚
The second inequality shows, by pushing 𝑛 → ∞, that 𝑒𝑚 ≤ ∑𝑘=0 𝑘! , which is
∞ 1
now independent of 𝑛. By letting 𝑚 → ∞, we get 𝑒 ≤ ∑𝑘=0 𝑘! . Therefore, we
∞ 1
get 𝑒 = ∑𝑘=0 𝑘! .
We may also want to bound the error term. In fact, we have
Proposition 2.3. For any 𝑚,

7
 𝑚
1 1 1
<𝑒−∑ < .
(𝑚 + 1)! 𝑘=0
𝑘! 𝑚!𝑚

𝑚 1 ∞ 1
Proof. Just directly compute 𝑒 − ∑𝑘=0 𝑘! = ∑𝑘=𝑚+1 𝑘! and see

𝑛 𝑛−𝑚−1
1 1 1 𝑚+1 1
∑ < ∑ 𝑗
< × = .
𝑘=𝑚+1
𝑘! 𝑗=0
(𝑚 + 1)!(𝑚 + 1) (𝑚 + 1)! 𝑚 𝑚!𝑚

Then let 𝑛 → ∞. The above result means

1 1 𝜃 𝑚
𝑒 − (1 + 1 + +…+ )= 𝑚 , < 𝜃𝑚 < 1.
2! 𝑚! 𝑚!𝑚 𝑚 + 1
1
We can get a much better estimation by adding the term 𝑚!𝑚 to our estimation.
To show this, we would need the following identity

1 1 1 1 1
1+1+ +…+ + =3− −⋯− .
2! 𝑚! 𝑚!𝑚 2! ⋅ 1 ⋅ 2 𝑚!(𝑚 − 1)𝑚

Note the use of the telescoping series trick by noticing

1 1 1 1 1 1
− = , − = .
𝑚!(𝑚 − 1) 𝑚!𝑚 𝑚!(𝑚 − 1)𝑚 𝑚!(𝑚 − 1) (𝑚 − 1)!(𝑚 − 1) 𝑚!

This shows

∞
1
𝑒=3−∑ .
𝑘=2
𝑘!(𝑘 − 1)𝑘

Hence

∞
1 1 1 1
∣𝑒 − (1 + 1 + +…+ + )∣ = ∑ .
2! 𝑚! 𝑚!𝑚 𝑘=𝑚+1
𝑘!(𝑘 − 1)𝑘

Derive the upper bound for the error

𝑛 𝑛
1 1 1 1 1 2
∑ < − ∑ < − = .
𝑘=𝑚+1
𝑘!(𝑘 − 1)𝑘 (𝑚 + 1)!𝑚 𝑘=𝑚+1 (𝑘 + 1)! (𝑚 + 1)!𝑚 (𝑚 + 2)! (𝑚 + 2)!𝑚

Using 2.2 and a standard technique, we can show

8
Proposition 2.4. 𝑒 is irrational.
𝑝
Proof. If not, then 𝑒 = 𝑞 for some coprime (positive) integers 𝑝, 𝑞 where 𝑞 ≠ 1.
Then

𝑝 1 1
= (1 + 1 + + … + ) + 𝜖𝑞 .
𝑞 2! 𝑞!

Therefore, by multiplying on both sides 𝑞!, we get 𝑞!𝜖𝑞 must be an integer (since
it is the difference between two integers). However,

1 1
0 < 𝜖𝑞 < ⟹ 0 < 𝑞!𝜖𝑞 < < 1.
𝑞!𝑞 𝑞

This means 𝑞!𝜖𝑞 can never be an integer, contradictory to our assumption that
𝑒 is rational. The above technique is often used in proving similar arguments in
similar situations.
Selected exercises for Section 1 & 2
Exercise 1. Show that

2𝑛 1 𝑥∈ℚ
lim { lim [cos(𝜋𝑚!𝑥)] } = { .
𝑚→∞ 𝑛→∞ 0 𝑥 ∈ ℚ𝑐

2𝑛
We call the function 𝐷(𝑥) = lim𝑚→∞ {lim𝑛→∞ [cos(𝜋𝑚!𝑥)] } Dirichlet’s func-
tion.
𝑥𝑛
Exercise 2. (a) Suppose lim𝑛→∞ (𝑥𝑛 − 𝑥𝑛−1 ) = 𝑑. Show that 𝑛 → 𝑑.
𝑥 √
(b) Suppose 𝑥𝑛 > 0 and lim𝑛→∞ 𝑥𝑛+1 = 𝐿. Show that lim𝑛→∞ 𝑛 𝑥
𝑛 = 𝐿.
𝑛

𝑐𝑛
(c) Suppose 𝑎𝑛 → 𝛼, 𝑏𝑛 → 𝛽, 𝑐𝑛 = 𝑎0 𝑏𝑛 + … + 𝑎𝑛 𝑏0 . Show that 𝑛 → 𝛼𝛽.
Exercise 3. (a) Prove the following generalization of 1.2. Suppose 𝑎𝑛 → +∞
and is strictly increasing, and

𝑏𝑛 − 𝑏𝑛−1
lim = 𝐿 ∈ [−∞, +∞].
𝑛→∞ 𝑎𝑛 − 𝑎𝑛−1

Then

𝑏𝑛
lim = 𝐿.
𝑛→∞ 𝑎𝑛

(b) Compute

9
 1 𝑛 𝑛
lim ∑ ln ( ).
𝑛→∞ 𝑛2 𝑘
𝑘=0

(c) Suppose 0 < 𝑎1 < 𝑞 −1 , 0 < 𝑞 ≤ 1, 𝑎𝑛+1 = 𝑎𝑛 (1 − 𝑞𝑎𝑛 ). Show that

lim𝑛→∞ 𝑛𝑎𝑛 = 𝑞 −1 .
Exercise 4. Let 𝑝(𝑛) be the number of prime numbers that divide 𝑛. Show that
𝑝(𝑛)
𝑛 → 0.

Exercise 5. Let

1 𝑛+1
𝑏𝑛 = (1 + ) .
𝑛

(a) Show that 𝑏𝑛 > 𝑏𝑛+1 .

(b) Hence, show that 𝑏𝑛 → 𝑒 from above, and

1 1 1
< ln (1 + ) < .
𝑛+1 𝑛 𝑛

(c) By using 𝑏𝑛 and 𝑒𝑛 and showing that they have the same limit, one can
give an alternative proof to the existence of 𝑒. Use a similar technique to show
𝑐𝑛 = 𝐻𝑛 − ln 𝑛 converges.
(d) Show that

𝑘 𝑘 𝑘
< ln (1 + ) < .
𝑛+𝑘 𝑛 𝑛

Hint: Use the technique in (b).

Exercise 6. (a) Show that 𝑐𝑛 = 𝐻𝑛 − ln 𝑛 converges. The limit of {𝑐𝑛 } is the
so-called Euler’s constant 𝛾.
Hint: use 4(b).
(b) We have shown previously that 𝑎𝑛 = 𝐻2𝑛 − 𝐻𝑛 converges. Use (a) to find
the limit of {𝑎𝑛 }.
Exercise 7. (a) Write the first few terms and use theorem 2.1 to show the
convergence of

1 1 1 (−1)𝑛−1
𝑢𝑛 = 1 − + − +…+ .
2 3 4 𝑛

(b) Show that 𝑢2𝑛 = 𝑎𝑛 , where 𝑎𝑛 = 𝐻2𝑛 − 𝐻𝑛 . Hence find the limit of {𝑢𝑛 }.

10
Exercise 8. (*) (a) Show that

1 1 𝑛 𝑒
< 𝑒 − (1 + ) < .
2𝑛 𝑛 𝑛

(b) Show that there exists 𝐵 > 0 independent of 𝑛 such that, for sufficiently
large 𝑛,

1 1
∣𝑛2 (ln (1 + ) − )∣ ≤ 𝐵.
𝑛 𝑛

In other words,

1 1 1
ln (1 + ) = + 𝑂( 2).
𝑛 𝑛 𝑛

Then use this to conclude that 𝑐𝑛 = 𝐻𝑛 − ln 𝑛 converges.

∞ 𝑘
Exercise 9. It is known that 1𝑒 = ∑𝑘=0 (−1)
𝑘! . Suppose 𝑎, 𝑏, 𝑐 are integers, and
𝑏 > 0. Show that there does not exist such 𝑎, 𝑏, 𝑐 that

𝑏
𝑎𝑒 + = 𝑐.
𝑒
1 1 𝑛
Exercise 10. (a) Compute lim𝑛→∞ (1 + 𝑛 + 𝑛2 ) .
𝑛
(b) Show that lim𝑛→∞ √
𝑛
𝑛!
= 𝑒 using exercise 1(a). Then show again using
exercise 1(b).
1
(c) Use (b) or exercise 2(a), compute lim𝑛→∞ (𝑛!) 𝑛2 .
(d) Compute

1 2 𝑛
lim (1 + 2
) (1 + 2 ) … (1 + 2 ).
𝑛→∞ 𝑛 𝑛 𝑛

(e) Suppose 𝑎1 = 1, 𝑎𝑛 = 𝑛(𝑎𝑛−1 + 1). Compute

𝑛
1
lim ∏ (1 + ).
𝑛→∞
𝑘=1
𝑎𝑘

(f) Let 𝐾𝑛 be the least 𝑘 such that 𝐻𝑘 ≥ 𝑛. Show that

𝐾𝑛+1
lim = 𝑒.
𝑛→∞ 𝐾𝑛

11
 𝑒𝐻 𝑛
Hint: Use 𝑛 → 𝑒𝛾 .
(g) It is known that lim𝑛→∞ 𝑛 sin 𝑛𝜋 = 𝜋. Use this and 2.3 to show that

lim 𝑛 sin(2𝜋𝑒𝑛!) = 2𝜋.

𝑛→∞

Exercise 11. Let 𝑃 (𝑥) and 𝑄(𝑥) be two non-constant polynomials of the same
degree and assume that 𝑃 (0)𝑄(0) ≠ 0. Suppose {𝑎𝑛 } , {𝑏𝑛 } are sequences of
positive numbers such that

𝑎𝑛+1 𝑃 ( 𝑎𝑏𝑛 ) 𝑏
lim = 𝐿, 𝑎𝑛+1 = 𝑛
, 𝑏𝑛+1 = 𝑛 .
𝑛→∞ 𝑎𝑛 𝑄 ( 𝑎𝑏𝑛 ) 𝑎𝑛
𝑛

Show that 𝐿 = 1.
Exercise 12. (*) To compute the exact limit of is often much more complicated
as it may involve finding the closed form for the sequences using various tricks
and techniques.
Consider the sequences {𝑎𝑛 } , {𝑏𝑛 } in 2.1.6. In this exercise, we find a closed
form of the two sequences using trigonometric substitution.
𝑏𝑛
Let 𝑐𝑛 = 𝑏𝑛+1 .
2
(a) Show that 𝑐𝑛 = 2𝑐𝑛+1 − 1.

(b) Let 𝜃 = arccos ( 𝑎𝑏1 ). Note that 𝜃 = 2 arccos (√ 𝑎12𝑎+𝑏1 ). Show that 𝑐𝑛 =
1 1

cos ( 2𝜃𝑛 ).
√
(c) Use lim𝑛→∞ 𝑛 sin 𝑛𝜋 = 𝜋 to show that 𝑎𝑛 , 𝑏𝑛 → 𝜋 when 𝑎1 = 2 3, 𝑏1 = 3.

3 - Iteration
In this section, we consider sequences of the form 𝑎𝑛+1 = 𝑓(𝑎𝑛 ), where 𝑓 ∶ 𝐼 → ℝ
is a function that is independent of 𝑛.
The following proposition summarizes the relation between monotonicity of 𝑓
and the sequence.
Proposition 3.1. Suppose {𝑥𝑛 } ∶ 𝑥𝑛+1 = 𝑓 (𝑥𝑛 ), and 𝑓 is monotone on 𝐼. If
𝑥𝑛 ∈ 𝐼, ∀𝑛, then
(1) when 𝑓 is monotone increasing, {𝑥𝑛 } is monotone, and
(2) when 𝑓 is monotone decreasing, {𝑥2𝑛 } , {𝑥2𝑛−1 } are monotone sequences with
opposite monotonicity.

12
Proof. We only show (1) since 𝑓 ∘ 𝑓 in (2) is monotone increasing. While (1) is
trivial using mathematical induction: if 𝑥1 ≤ 𝑓 (𝑥1 ) = 𝑥2 , then 𝑥3 = 𝑓 (𝑥2 ) ≤
𝑓 (𝑥1 ) = 𝑥2 .
Example 3.2. Let 𝐴 > 0, 𝑥1 > 0, and

1 𝐴
𝑥𝑛+1 = (𝑥𝑛 + ).
2 𝑥𝑛
√
By induction, 𝑥𝑛 > 𝐴. Then

√ 2
√ 1 𝐴 √ 1 (𝑥𝑛 − 𝐴) 1 √
𝑥𝑛+1 − 𝐴 = (𝑥𝑛 + − 2 𝐴) = ⋅ ≤ (𝑥𝑛 − 𝐴) .
2 𝑥𝑛 2 𝑥𝑛 2

√
This shows 𝑥𝑛 → 𝐴 and gives us an estimate on its speed. Moreover,

√ 2
√ 1 (𝑥𝑛 − 𝐴) 1 √ 2
𝑥𝑛+1 − 𝐴= ⋅ ≤ √ ⋅ (𝑥𝑛 − 𝐴) .
2 𝑥𝑛 2 𝐴

This is called convergence in order 2. The no. of effective digits is doubled

(why?) after each iteration.
Selected exercises for Section 3
Exercise 1. Another algorithm for computing square roots is to use the iteration

𝑥𝑛 (𝑥2𝑛 + 3𝐴)
𝑥𝑛+1 = .
3𝑥2𝑛 + 𝐴
√
Show that it indeed estimates 𝐴 and estimate the speed of its convergence.
2
Exercise 2. Suppose 𝑦0 ≥ 2, 𝑦𝑛 = 𝑦𝑛−1 − 2. Let

1 1 1
𝑆𝑛 = + +…+ .
𝑦0 𝑦0 𝑦1 𝑦0 𝑦1 ⋯ 𝑦 𝑛

Show that

𝑦0 − √𝑦02 − 4
lim 𝑆𝑛 = .
𝑛→∞ 2

Hint: let 𝑦0 = 𝑎 + 𝑎−1 for some 𝑎, and use repeatedly the following telescoping
trick for product

13
 1 𝑎2𝑁 − 𝑎−2𝑁
𝑎𝑁 + 𝑁
= 𝑁 .
𝑎 𝑎 − 𝑎−𝑁

14