Anda di halaman 1dari 33

1 Sequences

Sequences are an infinite list (a1 , a2 , a3 , . . . ), where an is some number. Usually


a sequence can be denoted by a general formula for each value of an for example
an = n2 would give a sequence a1 = 12 , a2 = 22 , a3 = 32 etc.

Definition 1.0.1 The objects an in a sequence are defined as the terms of the
sequence

One of the most famous sequences of all time is called the Fibonacci Se-
quence, where each term is the sum of the previous two.
Fibonacci Sequence: 1, 1, 2, 3, 5, 8, 13, 21, 34, . . .

Example 1.0.1 Some more examples of sequences are:


1 1 1 1
• an = has terms 1, , , , . . .
n 2 3 4
n
• an = has terms 12 , 32 , 34 , 45 , . . .
n+1
• an = (−1)n has terms -1, 1, -1, 1, -1, 1, -1, 1, . . .

1.1 Basic Notation


• an is the nth term in a sequence e.g. a1 is the first term, a2 is the second
term and a45 is the forty fifth term.
• (an )n∈N , (an )n , (an ), (an )∞
n=1 are all used to represent a sequence although
we will generally use (an ).
• We often denote limits (see later) by the Greek letter lambda, λ.
• We denote infinity by the symbol ∞

1.2 Limits
Often once we have found or made a sequence we want to find out more infor-
mation about it. One of the most common properties we look for in sequences
is called a limit.

Definition 1.2.1 We say that a sequence has a limit λ if as n gets large (tends
to infinity) an gets closer and closer to λ. This means that as we go further
and further down our sequence the terms get closer and closer to our limit. We
often write this as an −→ λ for n −→ ∞ or limn→∞ an = λ

Example 1.2.1 Limits of some sequences are:


1
• limn→∞ = 0. This is an important sequence as we use it a lot for finding
n
limits of other sequences.

2
n
• limn→∞ =1
n+1
2n + 1
• limn→∞ =2
n
Sometimes a sequence will not have a limit as the terms never get closer and
closer to one number. Such a sequence we say is divergent. Many sequences
are divergent for example the Fibonacci sequence is divergent, as the terms get
bigger and bigger. Also the sequence an = (−1)n with terms -1, 1, -1, 1, -1, 1, -1,
1, . . . is divergent as it never gets closer and closer to any number. Sometimes
when we have a sequence where the terms get bigger and bigger we say its limit
is infinity, ∞.

1.3 Finding limits


Given a sequence or number of sequences finding limits can be difficult. One
way to find a limit is to write out lots of terms of the sequence and see if you can
spot the limit from that. This can be often really hard as some sequences can
take hundreds of terms before you can spot the limit. Generally most sequences
we will see, have limits that can be found by playing around with the general
formula for each term, and comparing it to sequences we already know of. A
few good sequences to compare things against are:
a
• limn→∞ n1 = 0. This shows us that in general nnb has a limit of 0 if b is
bigger than a, a limit of 1 if b is the same as a and a limit of infinity if a
is bigger than b.
1
• limn→∞ n! = 0 since

1 1 1
= ≤ =0
n! 1 · 2 · 3 · 4···n n

• limn→∞ cos(πn) diverges.


• limn→∞ sin(πn) = 0.
We’ll look at an example
3n2 + 7n + 4
Example 1.3.1 Find the limit of the sequence an = .
9n2 + 24n + 9
This is a complicated example and were we to try and find the limit by writing
all the terms down, it would take a lot of working out for each of the terms
(try working out the term when n is 13!). So we use some clever maths to find
the limit. First we look to see what is the biggest power in the formula is, in
our case it is n2 . Note that in some cases the biggest power may not be on
n3 4
both parts of the fraction e.g. in n4 +34n 2 +3 the biggest power is n . So we then

3
divide both the top and bottom parts of the formula by out biggest power like
so:
3n2 7n 4
3n2 + 7n + 4 n2 + n2 + n2
=
9n2 + 24n + 9 9n2 24n
n2 + n2 + n2
9

Note that the last two terms on both the top and bottom of the fraction are
comparable with the limit of 1/n and so are 0. So we have.
3+0+0 3 1
= =
9+0+0 9 3
Hence our limit is a third.

1.4 Limits of more than one sequence


An interesting thing to do is when we have found the limit of several sequences
is to combine them to see what happens to the limit. When we do this we have
a few rules that help us. Given two sequences an and bn with limits a and b we
have that
1. limn→∞ (an + bn ) = a + b
2. limn→∞ (an − bn ) = a − b
3. limn→∞ (an · bn ) = a × b
an
4. limn→∞ ( )=a÷b
bn
5. If we have a number r then limn→∞ (r · an ) = r × a
So lets look at an example.
2n2 + 12n + 5
 3
3n + 2n2 + 1
 
Example 1.4.1 Find the limit of the sequence an = · .
n2 + 3n + 1 2n3
Ok since we have a multiplication of two sequences we will be using rule 3 from
our list. But first we must work out the limit of each individual sequence. As
before we note that the highest power in the first one is n2 and in the second
one it is n3 , so dividing we get.
2n2 12n 5
2n2 + 12n + 5 n2 + n2 + n2 2+0+0
= = =2
n2 + 3n + 1 n2 3n 1
n2 + n2 + n2
1+0+0
3n3 2n2 1
3n3 + 2n2 + 1 n3 + n3 + n3 3+0+0 3
= 3 = =
2n3 2n
n3
2 2

2n2 + 12n + 5
 3
3n + 2n2 + 1
 
So using rule 3 we have that the limit of an = ·
n2 + 3n + 1 2n3
3
is 2 × which is three.
2

4
2 Series
2.1 Introduction
Series’ are made up from sequences. Given any sequence (an ) we can make a
new sequence (Sn ) as follows.

S0 = a0
S1 = a0 + a1
S2 = a 0 + a 1 + a 2
S3 = a 0 + a 1 + a 2 + a 3
..
.
Sn = a 0 + a 1 + a 2 + a 3 + · · · + a n

We have a limit for this, as n tends to infinity (gets bigger and bigger) we have
that the sequence (Sn ) is simply the infinite sum of all the an
X∞
lim Sn = lim (a0 + a1 + a2 + a3 + · · · + an ) = an
n→∞ n→∞ n=0


X
Definition 2.1.1 The limit of the sequence Sn , an , is called an infinite
n=0
series.

Definition 2.1.2 The terms of the sequence (Sn ) are called partial sums of the
series.

2.2 Arithmetic Series


Definition 2.2.1 A series which is the sum of a sequence (an ) where each term
an is the previous term plus some number d is called an arithmetic series i.e
an = an−1 + d.

Example 2.2.1 Let (an ) be a sequence where a1 = 1 and an = a( n − 1) + 2, i.e



X
the sequence (1,3,5,7,9,11,. . . ). The arithmetic series is an = 1 + 3 + 5 +
n=1
7 + ....

By doing some clever rearrangement one can see that each partial sum Sn is
1
in fact n(a1 + an ). Just looking at our previous example with the sequence
2
(1, 3, 5, 7, 9, 11, . . . ) we see that S4 = 1 + 3 + 5 + 7 which is 16, but using
1 1
our formula we get S4 = 4(1 + 7) = 4 · 8 = 16. This is a nice trick for
2 2
finding partial sums of arithmetic series especially ones where n is large. It
can be used to find the sum all the numbers from 1 to 100. He simply took
1/2 · 100 · (1 + 100) = 1/2 · 100 · 101 = 50 · 101 = 5050.

5
2.3 Geometric Series
Definition 2.3.1 A series where the sequence involved is an = xn for some

X
number x, i.e the series xn is called a geometric series
k=0

Now these type of series are very common and there are a couple of important
results to do with them.
Theorem 2.3.1 Take any number x. Then if | x |< 1 then,

X 1
xn =
1−x
k=0

If | x |≥ 1 then,

X
xn = ∞
k=0

i.e the sum just gets bigger and bigger.

Example 2.3.1 A few examples of geometric series,


∞  n
X 1 1 1 4 1
• = = = =1
4 1 − 1/4 3/4 3 3
k=0

X 1
• = 1 + 1/2 + 1/3 + · · · ≤ 1 + 1/2 + 1/2 + . . . which gets bigger and
n
k=0
bigger so is equal to infinity.

6
In a case that limn→∞ Sn does not exist we say that the infinite sum Σ∞
n=1 an
DIVERGES.

OBSERVATION 1
In a case when all elements of the sequence {an }n∈N −{0} are equal 0 starting
from a certain k ≥ 1 the infinite sum becomes a finite sum, hence the infinite
sum is a generalization of the finite one, and this is why we keep the similar
notation.

EXAMPLE 1
The infinite sum of a geometric sequence an = xk for x ≥ 0, i.e. Σ∞
n=1 x
n

converges if and only if | x |< 1 because

x − xn+1
Σnk=1 xk = Sn = , and
x−1
x(1 − xn ) x x
lim = lim (1 − xn ) = iff | x |< 1,
n→∞ x−1 n→∞ x − 1 x−1
hence
x
Σ∞ k
n=1 x = .
x−1

EXAMPLE 2
The series Σ∞
n=1 1 DIVERGES to ∞ as Sn = Σk=1 1 = n and limn→∞ Sn =
n

limn→∞ n = ∞.

EXAMPLE 3
The infinite sum Σ∞
n=1 (−1)
n
DIVERGES.

EXAMPLE 4
The infinite sum Σ∞ 1
n=0 (k+1)(k+2) CONVERGES to 1; i.e.

1
Σ∞
n=0 = 1.
(k + 1)(k + 2)

1
Proof: first we evaluate Sn = Σnk=1 (k+1)(k+2) as follows.

1 1 n+1 1
Sn = Σnk=0 = Σnk=1 k −2 = − | =− + 1 and
(k + 1)(k + 2) x+1 0 n+2
1
lim Sn = lim − + 1 = 1.
n→∞ n→∞ n+2

2
DEFINITION 3
For any infinite sum (series)
Σ∞
n=1 an

a series
rn = Σ∞
m=n+1 am

is called its n-th REMINDER.

FACT 1
If Σ∞ ∞
n=1 an converges, then so does its n-th REMINDER rn = Σm=n+1 am .

Proof : first, observe that if Σ∞


n=1 an converges, then for any value on n so does
rn = Σ∞m=n+1 m a because

rn = lim (an+1 + ... + an+k ) = lim Sn+k − Sn = Σ∞


m=1 am − Sn .
n→∞ n→∞

So we get

lim rn = Σ∞ ∞ ∞
m=1 am − lim Sn = Σm=1 Sm − Σn=1 an = S − S = 0.
n→∞ n→∞

General Properties of Infinite Sums

THEOREM 1
If the infinite sum

Σ∞
n=1 an converges, then lim an = 0.
n→∞

Proof : observe that an = Sn − Sn−1 and hence

lim an = lim Sn T he − lim Sn−1 = 0,


n→∞ n→∞ n→∞

as limn→∞ Sn = limn→∞ Sn−1 .

REMARK 1
The reverse statement to the theorem 1

If lim an = 0. then Σ∞
n=1 an converges
n→∞

is not always true. There are infinite sums with the term converging to zero
that are not convergent.

3
EXAMPLE 5
The infinite HARMONIC sum
1
H = Σ∞
n=1
n

DIVERGES to ∞, i.e. Σ∞
n=1 n = ∞ but limn→∞
1 1
n = 0.

DEFINITION 4
Infinite sum Σ∞ n
n=1 an is BOUNDED if its sequence of partial sums Sn = Σk=1 ak
is BOUNDED; i.e. there is a number M such that |Sn | < M , for all n ≤ 1, n ∈
N.

FACT 2
Every convergent infinite sum is bounded.

THEOREM 2
If the infinite sums Σ∞ ∞
n=1 an , Σn=1 bn CONVERGE, then the following properties
hold.

Σ∞ ∞ ∞
n=1 (an + bn ) = Σn=1 an + Σn=1 bn ,

Σ∞ ∞
n=1 can = cΣn=1 an , c ∈ R.

Alternating Infinite Sums. Abel Theorem


DEFINITION 5
An infinite sum
Σ∞
n=1 (−1)
n+1
an , for an ≥ 0
is called ALTERNATING infinite sum (alternating series).

EXAMPLE 6
Consider
Σ∞
n=1 (−1)
n+1
= 1 − 1 + 1 − 1 + ....
If we group the terms in pairs, we get

(1 − 1) + (1 − 1) + .... = 0

but if we start the pairing one step later, we get

1 − (1 − 1) − (1 − 1) − ..... = 1 − 0 − 0 − 0 − ... = 1.

4
It shows that grouping terms in a case of infinite sum can lead to inconsistencies
(contrary to the finite case). Look also example on page 59. We need to develop
some strict criteria for manipulations and convergence/divergence of alternating
series.

THEOREM 3
The alternating infinite sum Σ∞
n=1 (−1)
n+1
an , (an ≥ 0) such that
a1 ≥ a2 ≥ a3 ≥ .... and lim an = 0
n→∞

always CONVERGES. Moreover, its partial sums Sn = Σnk=1 (−1)n+1 an fulfil


the condition
S2n ≤ Σ∞
n=1 (−1)
n+1
an ≤ S2n+1 .

Proof : observe that the sequence of S2n is increasing as


S2(n+1) = S2n+2 = S2n +(a2n+1 −a2n+2 , and a2n+1 −a2n+2 ≥ 0, i.e.S2n+2 ≥ S2n.

The sequence of S2n is also bounded as


S2n = a1 − ((a2 − a3 ) + (a4 − a5 ) + ...a2n ) ≤ a1 .
We know that any bounded and increasing sequence is is convergent, so we
proved that S2n converges. Let denote limn→∞ S2n = g.
To prove that Σ∞n=1 (−1)
n+1
an = limn→∞ Sn converges we have to show now
that limn→∞ S2n+1 = g.
Observe that S2n+1 = S2n + a2n+2 and we get
lim S2n+1 = lim S2n + lim a2n+2 = g
n→∞ n→∞ n→∞

as we assumed that limn→∞ an = 0.


We proved that the sequence {S2n } is increasing, in a similar way we prove
that the sequence {S2n+1 } is decreasing. Hence S2n ≤ limn→∞ S2n = g =
Σ∞n=1 (−1)
n+1
an and S2n+1 ≥ limn→∞ S2n+1 = g = Σ∞ n=1 (−1)
n+1
an , i.e
S2n ≤ Σ∞
n=1 (−1)
n+1
an ≤ S2n+1 .

EXAMPLE 7
Consider the INHARMONIC series (infinite sum)
1 1 1 1
AH = Σ∞
n=1 (−1)
n+1
= 1 − + − .....
n 2 3 4
Observe that an = n1 , and n1 ≥ n+1
1
i.e. an ≥ an+1 for all n, so the assumptions
of the theorem 3 are fulfilled for AH and hence AH converges. In fact, it is
proved (by analytical methods, not ours) that
1
AH = Σ∞
n=1 (−1)
n+1
= ln 2.
n

5
EXAMPLE 8
A series (infinite sum)
1 1 1 1 1
Σ∞
n=0 (−1)
n
= 1 − + − + ......
2n + 1 3 5 7 9
converges, by Theorem 3 (proof similar to the one in the example 7). It also
is proved that
1 π
Σ∞n=0 (−1)
n
= .
2n + 1 4

THEOREM 4 (ABEL Theorem)


IF a sequence {an } fulfils the assumptions of the theorem 3 i.e.

a1 ≥ a2 ≥ a3 ≥ .... and lim an = 0


n→∞

and an infinite sum (converging or diverging) Σ∞


n=1 bn is bounded,
THEN the infinite sum
Σ∞n=1 an bn

always converges.

Observe that Theorem 3 is a special case of theorem 4 when bn = (−1)n+1 .

Convergence of Infinite Sums


with Positive Terms

We consider now infinite sums with all its terms being positive real numbers,
i.e.
S = Σ∞n=1 an , for an ≥ 0, an ∈ R.

Observe that if all an ≥ 0, then the sequence {Sn } of partial sums Sn = Σnk=1 ak
is increasing; i.e.
S1 ≤ S2 ≤ .... ≤ Sn ...
and hence the limn→∞ Sn exists and is finite or is ∞. This proves the following
theorem.

THEOREM 5
The infinite sum
S = Σ∞
n=1 an , for an ≥ 0, an ∈ R

always converges, or diverges to ∞.

6
THEOREM 6 (Comparing the series)
Let Σ∞
n=1 an be an infinite sum and {bn } be a sequence such that for all n,

0 ≤ bn ≤ an .

If the infinite sum Σ∞ ∞


n=1 an converges, then Σn=1 bn also converges and

Σ∞ ∞
n=1 bn ≤ Σn=1 an .

Proof: Denote
Sn = Σnk=1 ak , Tn = Σnk=1 bk .
As 0 ≤ bn ≤ an we get that also Sn ≤ Tn . But

Sn ≤ lim Sn = Σ∞ ∞
n=1 an so also Tn ≤ Σn=1 an = S.
n→∞

The inequality Tn ≤ Σ∞ n=1 an = S means that the sequence {Tn } is a bounded


sequence with positive terms, hence by theorem 5, it converges.

By the assumption that Σ∞


n=1 an we get that

Σ∞ n
n=1 an = lim Σk=1 ak = lim Sn = S.
n→∞ n→∞

We just proved that Tn = Σnk=1 bk converges, i.e.

lim Tn = T = Σ∞
n=1 bn .
n→∞

But also we proved that Sn ≤ Tn , hence

lim Sn ≤ lim Tn
n→∞ n→∞

what means that


Σ∞ ∞
n=1 bn ≤ Σn=1 an .

EXAMPLE 9
Let’s use Theorem 5 to prove that the series
1
Σ∞
n=1
(n + 1)2
2
converges. We prove by analytical methods that it converges to π6 , here we
prove only that it does converge. First observe that the series below converges
to 1, i.e.
1
Σ∞
n=1 = 1.
n(n + 1)

7
Consider
1 1 1 1 1 1 1 1 1
Sn = + .... + = (1 − ) + ( − ) + ...( − ) = 1−
1·2 2·3 n(n + 1) 2 2 3 n n+1 n+1
so we get
1 1
Σ∞
n=1 = lim Sn = lim (1 − ) = 1.
n(n + 1) n→∞ n→∞ n+1

Now we observe (easy to prove) that


1 1 1 1 1 1
≤ , ≤ , ..... ≤ , ......
22 1 · 2 32 1·3 (n + 1)2 n(n + 1)

i.e. we proved that all assumptions of Theorem 5 hold, hence Σ∞ 1


n=1 (n+1)2
converges and moreover
1 1
Σ∞
n=1 ≤ Σ∞
n=1 .
(n + 1)2 n(n + 1)

THEOREM 7 (D’Alambert’s Criterium)


Any series with all its terms being positive real numbers, i.e.

Σ∞
n=1 an , for an ≥ 0, an ∈ R

converges if the following condition holds:


an+1
lim < 1.
n→∞ an

Proof: let h be any number such that limn→∞ aan+1 n


< h < 1. It means that
there is k such that for any n ≥ k we have aan+1
n
< h, i.e. an+1 < han and

ak+1 < ak h, ak+2 = ak+1 h < ak h2 , ak+3 = ak+2 h < ak h3 , ......

i.e. all terms an of Σ∞n=k an are smaller then the terms of a converging (as
0 < h < 1) geometric series Σ∞ n 2
n=0 ak h = ak + ak h + ak h + .... By Theorem

5 the series Σn=1 an must converge.

THEOREM 8 (Cauchy’s Criterium)


Any series with all its terms being positive real numbers, i.e.

Σ∞
n=1 an , for an ≥ 0, an ∈ R

converges if the following condition holds:



lim n
an < 1.
n→∞

8
Proof: we carry the proof in a similar way as the proof of theorem 6. Let h

be any number such that limn→∞ n an < h < 1. So it means that there is k

such that for any n ≥ k we have n an < h, i.e. an < hn . This means that all
terms an of Σ∞n=k an are smaller then the terms of a converging (as 0 < h < 1)
geometric series Σ∞ n k
n=k h = h + h
k+1
+ hk+2 + .... By Theorem 5 the series

Σn=1 an must converge.

THEOREM 9 (Divergence Criteria


Any series with all its terms being positive real numbers, i.e.

Σ∞
n=1 an , for an ≥ 0, an ∈ R

diverges if
an+1 √
lim > 1 or lim n
an > 1
n→∞ an n→∞

Proof: observe that if limn→∞ aan+1


n
> 1, then for sufficiently large n we have
that
an+1
> 1, and hence an+1 > an .
an
This means that the limit of the sequence {an } can’t be 0. By theorem 1 we
get that Σ∞n=1 an diverges.

Similarly, if limn→∞ n an > 1, then then for sufficiently large n we have that

n
an > 1 and hence an > 1,

what means that the limit of the sequence {an } can’t be 0. By theorem 1 we
get that Σ∞
n=1 an diverges.

REMARK 2
It can happen that for a certain infinite sum Σ∞
n=1 an )

an+1 √
lim = 1 = lim n an .
n→∞ an n→∞

In this case our Divergence Criteria do not decide whether the infinite sum
converges or diverges. In this case we say that the infinite sum DOES NOT
REACT on the criteria.

EXAMPLE 10
The Harmonic series
1
H = Σ∞
n=1
n
does not react on D’Alambert’s Criterium (Theorem 7) because
n 1
lim = lim = 1.
n→∞ n + 1 n→∞ (1 + n1 )

9
EXAMPLE 11
The series from example 9
1
Σ∞
n=1
(n + 1)2
does not react on D’Alambert’s Criterium (Theorem 7) because
2 1
(n + 1)2 n2 + 2n + 1 1+ n + n2
lim 2
= lim 2 = lim 4 4 = 1.
n→∞ (n + 2) n→∞ n + 4n + 1 n→∞ 1 + +
n n2

REMARK 3 Both series Σ∞ 1 ∞ 1


n=1 n and Σn=1 (n+1)2 do not react on D’Alambert’s,
but first in divergent and the second is convergent.

There are more criteria for convergence, most known are Kumer’s criterium and
Raabe criterium.

EXAMPLE 12
The series
cn
Σ∞
n=1
n!
converges for c > 0.

Proof : Use D’Alambert Criterium.


an+1 cn+1 n! c
= n · =
an c (n + 1)! n+1

and
an+1 c
lim = lim = 0 < 1.
n→∞ an n→∞ n + 1

EXAMPLE 13
The series
n!
Σ∞
n=1
nn
converges.

Proof : Use D’Alambert Criterium.


an+1 n!(n + 1) nn nn an+1 (n + 1)nn n n 1
= · = (n+1) = = =( ) =
an (n + 1) n+1 n! (n + 1) n+1 an n
(n + 1) (n + 1) n+1 (1 + n1 )n

10
CHAPTER 9

Sequences and Series

9.1. Convergence: Definitionand Examples


Sequences

The purpose of this chapter is to introduce a particular way of generating algorithms for finding
the values of functions defined by their properties; for example, transcendental functions. This is the
technique of InfiniteSeries. Computer algorithms for determining the value of a function depend upon
the usual arithmetic operations; thus an exact determination can only be achieved for rational functions
(quotients of polynomials). If a function is transcendental, its values can only be approximated. For
example, we know that

(9.1) ex 

lim 1
n ∞
 nx  n

This expression tells us that if for any n we do the calculation described by the expression on the right,
that these numbers will, for n large enough, be close to the “true” value of e x . Now, it turns out that this
is a very inefficientway to calculate ex , and the expression as an infiniteseries (which we will discover
later in this chapter)

(9.2) ex  1 x x2
 x3
 

 xn
 

2! 3! n!

is far better. Equation 9.2 is taken to mean: add up the numbers of the form x n n!, starting with n 0. If
we add up enough terms, we have a good approximation to e x . Of course, it is important to have estimates
on how good this approximation is, and more generally, to have ways of finding these approximating
sums. That is what we study in this chapter, starting with the idea of convergence in the sense of “good
approximation”.


Definition9.1 A sequence is a list of numbers, denoted an , where an is the nth term of the sequence.

A sequence may be definedby a specific formula or an algorithm for determining the members of
the sequence successively, or recursively.

128
9.1 Convergence: Definitionand Examples 129

Example 9.1 The formulae an  n n 1; bn  n 1


n 2; cn  3 2n  n  0 define the
n 1
sequences, respectively:

(9.3) 1 2 3
 
  n  
  ; 3 4 5
  

   n  1


  ; 3 5 7 9

   3  2n 

 
1 2 3 n 1
  
The last sequence can be definedrecursively by: a0
is given by the recursion a1 1  an an  1 1.   
3, and for n  0  cn cn  1 2. Similarly, the first


The symbol n! (read “ n-factorial”) is used to denote the product of the firstn integers. This also has
the recursive definition:a0 1, and for n  0, an nan  1. (Note that we have defined0! 1).  
Of the sequences described in equation 9.2, the first and the third clearly grow without bound, but
the second is bounded; in fact, if we rewrite the general term as

(9.4) bn  n 1  1 1
n

n 1 1 1
n

 
  a  
 
we see that the sequence bn approaches 1. We say that bn converges to 1, as in the following definition.
Definition9.2 A sequence a1  a2  n converges to a limit L, written

(9.5)

lim an
n ∞
L

if, for every ε  0, there is an n0 such that for all n  n0 we have  an  L  ε .
This just says that we can be sure that an is as close to L as we need it to be, just by taking the index
n large enough. We will rarely have to actually use this definition,relying more on understanding what
it says, and known facts about limits. For example:

f and if we know that limx ∞ f x


 
Proposition 9.1 If the general term an of a sequence can be expressed as f n
for a continuous function
L, then we can conclude that limx ∞ an L.  
As an application, using results from the preceding chapter, we have


Proposition 9.2

a) lim n p ∞ for p  0 

n ∞
1

b) lim p 0 for p  0 
n ∞n
 
c) lim A1 n 1 if A  0
n ∞ 
Let p and q be polynomials.
p n
 p n
 
d) lim

n ∞ q n

0 if deg p  deg q 

lim
n ∞ q n

∞ if deg p  deg q
p n
 a
e) If the polynomials p and q have the same degree, then lim

n ∞ q n
b
 where a and b are the
leading coefficientsof p and q.
p n


f) lim n
n ∞ e
0 for any polynomial p.
p n

g) lim

n ∞ ln n
c
∞ for any polynomial of positive degree and any positive c.
Chapter 9 Sequences and Series 130

These can all be derived by replacing n by x, and using limit theorems already discussed (such as
l’Hˆ
opital’s rule).
n2  
Example 9.2 lim

n ∞ n2  
n 1
1  by e
above

 1
n 
Example 9.3 lim
n ∞ n 0  since the numerator oscillates between -1 and 1, and the denominator
goes to infinity. We should not be perturbed by such oscillation, so long as it remains bounded. For
example we also have

sin n

(9.6)

lim
n ∞ n
0

since the term sin n


remains bounded. The following propositions state the general rule for handling
such cases.

Proposition 9.3
 
    
a) (Squeeze theorem) Given three sequences an  bn  cn , if an  bn  cn for all n  andlimn ∞ an


limn ∞ cn L  then also limn ∞ bn L
b) If an bn cn , the sequence bn is bounded, cn  0 and limn ∞ cn 0, then also limn ∞ an     0.
Let’s see why b) is true, using a). Let M be the bound of the  bn  . Then

(9.7) Mcn  bn cn   Mcn

so a) applies and the conclusion follows.


In some cases where none of the above rules apply, we have to return to the definitionof convergence.
an  
Example 9.4 For any a  0, lim
n ∞ n!
0


To see why this is true, we think of the sequence as recursively defined: the firstterm, a 1 is a, and


each an is obtained by multiplying its predecessor by a n. Now, eventually, that is, for n large enough,
a n  1 2. Thus each term after that is less than half its predecessor. This now surely looks like a
sequence converging to zero. To be more precise, let N be the firstinteger for which a N  1 2. Then
for any k  0,

aN k


1 aN 
(9.8)

N k
!

2k N!

Now the sequence on the right is a fixed number (aN N ) times a sequence (1 2k ) which tends to zero.
Thus our sequence converges to zero, also by the squeeze theorem (proposition 9.3a).
Note that in the above argument, we only had to show that the general term of our sequence is
dominated by the general term of a sequence converging to zero from some point on. What happens to
any finitecollection of terms of a sequence is not relevant to the question of convergence. We shall use
the word eventually to mean “from some point on”, or more precisely, “for all n greater than some fixed
integer N”. We restate proposition 9.3, using the word ”e ventually”:
9.1 Convergence: Definitionand Examples 131

Proposition 9.4
a) (Squeeze theorem) Given three sequences an  bn  cn , if eventually
 
(9.9) an  bn  cn for all n  and lim an

n ∞ 
lim cn
n ∞
L

then also
 L
(9.10)
n ∞ 
lim bn

b) Suppose that an  bn cn eventually, that is, for all n larger than some N. If the sequence b n is
bounded and

(9.11) lim cn
n ∞  0

then also
 0
(9.12)

lim an
n ∞

np  
Example 9.5 For any positive integer p, lim
n ∞ n!
0

The idea here is that the numerator is a product of p terms, whereas the denominator is a product of
n terms, so grows faster than the numerator. To make this precise, write
np   
 n
n 1 
1
 
 n  p 
(9.13)
n! n n  1
n  p
!

Now, if n is so large that n n  p
 2  n  2p will do), then the firstfactor is bounded by 2 p . Thus,
for n  2p, that is, eventually,

(9.14)
np
 2p
1 
n! n  p
!

Since 1 n  p
! 0 and n ∞, the result follows from the squeeze theorem.
Finally, we note that the limit of a sum is the sum of the limits:
Proposition 9.5 If an  bn  cn , and the sequences bn and cn converge, then so does the sequence an ,
and
  lim 
(9.15)

lim an
n ∞ 
lim bn
n ∞  c n ∞
n

Series

For many sequences, in fact, the most important ones, the general term is formed by adding something to
its predecessor; that is, the sequence is formed by the recursion s n sn  1 an , where an is from another  
sequence. Such a sequence is called a series. Explicitly, the terms of the series are

(9.16) a 1  a1  a2  a1  a2  a3 
 
  a  a2  a3  


 a 
 
 
1 n
Chapter 9 Sequences and Series 132

It is useful to use the summation symbol:

   
    n 
(9.17) a1 a2 a3 an ∑ ak
k 1

Definition9.3 The series



(9.18) ∑ ak
k 0

is to be considered as the limit of the sequence

 n 
(9.19) sn ∑ ak
k 0


If the limit L of the sequence sn exists, the series is said to converge, and L is called its sum. If the
limit does not exist, the series diverges. The terms of the sequence s n are called the partial sums of the
series.

Example 9.6 ∑ 2k

1  1
k 1
Let’s look at a few partial sums:

(9.20)
1 3 7 15
   



2 4 8 16


We see that each term adds half the distance of its predecessor from 1, from which we guess that the
partial sum: sn 1  2  n. Let’s now show that to be true. As we have seen it is true for the first four
terms. If it is true for the n  1
th term, it is also true for the nth term:

(9.21) sn  sn   1  1
1
 1  1
2 1  1
1 
1
2n 2n  1 2n 2n 2n
Thus, our guess holds for the fifth, and then the sixth, and, by continued application of equation 9.21,
ultimately, for all terms. So the result is easy to conclude:

∑ 2k
1   1  1
(9.22)
k 1
n ∞ 
lim sn
n ∞ 
lim 1 
2n

Now, remember that the index is a way of relating the partial sums of the series to the general term
from which it is defined, so if we change that relation consistently, we don’t change the series. For
example,

∑ ak  ∑ an  ∑ ak 
∞ ∞ ∞ ∞
(9.23) 

1 ∑ am  8
k 1 n 1 k 0 m 9

and so forth. Each representation comes about by replacing the index with a new index. For example, if
we substitute n for k, we get the firstequality; if we substitute k 1 for n we get the second equality, and 
9.1 Convergence: Definitionand Examples 133


if we replace k 1 by m  8, we get the last one. It is often useful to make a change of index as the next
examples show.

Example 9.7 ∑ 2k

1  2
k 0
For

(9.24)

∑ 2k
1  1 ∞
∑ 2k
1  1 1 2 
k 0 k 1


1  1 
Example 9.8 ∑ 2k 2n  1
k n
First, change the index by k  m  n, and then factor out 2  : n


∑  ∑
1  1 
2  2 
∞ ∞ ∞
1
2 ∑ n n
2 n 1


(9.25)
2k 2m n 2m


k n m 0 m 0

Proposition 9.6 (Geometric Series) :

∑ xk 

1
(9.26) for  x   1 
k 0 1 x

∞ 
(9.27) ∑ xk diverges for  x  1
k 0

To show this, we obtain (by a clever little observation) a formula for the partial sums

 
∑ xk   x  x  
 

n
2
(9.28) sn 1 xn
k 0

Note that

(9.29) sn 

1
 1  x  x  
   x
 x  s  x and
2 n n 1


n
n 1


(9.30) sn 

1
 1  x  x 
    x
 1  xs 
2 n 1


Equating these expressions for sn 

1 , we obtain s  x  1  xs . Solving this for s :


n
n 1


n n

 n
 1  xn 1


(9.31) sn ∑x k
1 x

k 0

so

  1  xn 1


(9.32) ∑ xk 
lim sn
n ∞
lim
n  ∞ 1 x

k 0
Chapter 9 Sequences and Series 134

which equals 1  x
 1 if  x   1 and diverges if  x   1. We look at the cases x  


  1 separately. For
x 1  sn n, so the series diverges. For x  1, the sequence sn is the sequence 1  0  1  0  1  0  , so 
cannot converge to any particular number.

Example 9.9

∑ k k 
1  1  We firstuse the fact that
n 1 1

1  1 1 
(9.33)
k k  1


k k 1 
Thus the partial sum sn can be calculated:

    


   

1 1 1 1 1 1 1
(9.34) sn 1
2 

2 3 

3 4 

n n 1  

 1      
   
   

1 1 1 1 1 1 1 1 1
(9.35) 
2 2 

3 3 

4 4 
 
n n 

n  1

 1
(9.36) 1
n  1


which converges to 1 as n goes to infinity. This is an example of a telescoping series.


Finally, we observe that if a series converges, its general term must go to zero. Be careful: there are
many series whose general term goes to zero which do not converge.

Proposition 9.7 If

∑ ak converges, then limn ∞ ak
 0
k 0

To see this, let sn ∑nk 0 ak  tn ∑nk  
Then,
lim  t . Thus lim  s  t
 0. But s  t  a .
1
0 ak . since these are both sequences of the partial sums
of the series, but indexed differently, limn  ∞ sn n ∞ n n ∞ n n n n n
Finally, Proposition 9.5 gives us:
  cn , and the series ∑ bn and ∑ cn converge, then so does the series ∑ an ,

Proposition 9.8 If an bn
and ∑ an ∑ bn ∑ cn 

9.2. Tests for convergence


Throughout this section, unless otherwise specified,we will be considering series, all of whose terms are
positive. For such a series, the sequence of partial sums is increasing. If they remain bounded, then - just
as in the assertion of theorem 8.1 for functions - the sequence of partial sums will converge.
n ∞
Proposition 9.9 If ak  0 for all k, and there is an M  0 such that ∑ ak 
M for all n, then ∑ ak
k 0 k 0
converges.
9.2 Tests for convergence 135

Because of this proposition, for a series with positive terms, the statements ∑ ak converges, ∑ ak
diverges, are usually written simply as

∑ ak 
∞ ∞
(9.37) ∑ ak  ∞ converges
 ∞ diverges

k 0 k 0

Here is an important application of this proposition:


Proposition 9.10 (Comparison Test). Given two sequences ak  bk with 0 ak bk . Then
a) If ∑ bk  ∞  then ∑ ak  ∞  
 


b) If ∑ ak ∞  then ∑ bk ∞ 
As for (a), the sequence of partial sums sn ∑n0 ak is bounded by ∑∞ 
0 bk , so converges by Proposition 9.9.
In the second case, since the sequence of partial sums ∑ ak has no bound, neither does the sequence of
partial sums of ∑ bk .
It is important to observe that it is not necessary that the inequalities in the hypothesis of proposition
9.10 hold for all k, only that they eventually hold. That is because the issue of convergence series is
determined by the end of the series, and not affected by any finitenumber of terms.
1 
Example 9.10 ∑ rk r  1

 ∞ if 0  r  1

Since rk


1  rk r  1
,
1 1
(9.38)

rk r 1


rk 1
 

so the comparison test applies.

k
Example 9.11 ∑ k  ∞ if r  1

r
Now, here the trouble is that the numerator grows without bound - but it doesn’t grow as fast as a


power. So, what we do is borrow something from the denominator to compensate for the numerator. We
note that eventually k r k 2  1; in fact, this is true as soon as k  2 ln k lnr (which eventually happens,

since k ln k ∞). Then for all k larger than this number

(9.39)
k  k 1

1 
rk  r
k  r
k  r
k

Since r  1, we also have r  1, and so the series


1
(9.40) ∑  r
k

converges, and thus, by comparison, our original series converges.



1 
Example 9.12 ∑ n2  ∞
n 0
Now,

(9.41)
1

1  1

1

n2 n n  1
n 1 n
Chapter 9 Sequences and Series 136

so our series is dominated by a telescoping series which converges (see example 9.9. above).
A very useful application of the comparison test is the following.
Proposition 9.11 (The Integral Test). Suppose that f is a nonnegative, nonincreasing function defined
on an interval M  ∞
. Suppose the an is a sequence such that for n  M, an f n
. Then 
∞ ∞
a) If  f x
dx  ∞ then ∑ an  ∞

M

 ∑ an 
n 0
∞ ∞
b) If  f x
dx ∞ then ∞
M n 0
Let

(9.42) bn  n 1


f x
dx




Then, since the function is nonincreasing, f n


 bn  f an  1
; that is an  bn  an 1 Now, use the 

comparison theorem. For example, if f x


dx  ∞, then ∑ bn converges, so by comparison ∑ an 1 also



converges.

∑n

1
Example 9.13 (The harmonic series). ∞
We apply the integral test using the function f x

n 1
 1 x. Since
(9.43) 
∞ dx  ∞
1 x
as we saw in chapter 8, the result follows.
If we apply example 17 of chapter 8 to series via the integral test we have a result which is very
useful for comparisons:
Proposition 9.12 Let p be a positive number.

1
a) ∑ p  ∞ if p  1
n 1n

1
b) ∑ p ∞ if p 1  
n 1n



This follows from the same result for the integral of 1 x p .



1 
Example 9.14 ∑ n ln n
p
The function f x

n 2
 1 x ln x
p is decreasing. We integrate using the substitution u  ln x:

(9.44)
A dx  ln A du 
x ln x
p up
 

2 ln 2

We know (again from example 17, chapter 8) that this converges if p  1, and otherwise diverges. Thus,
by the integral test,

1
(9.45) ∑ n ln n
p  ∞ if p  1 
n 2
9.2 Tests for convergence 137

and otherwise diverges.


Finally, we need a tool to test for convergence when we cannot realize the general term of the series
in the form f n
for some function f . For example, if the expression for a n involves the factorial, we
proceed to the following.
Proposition 9.13 (Ratio Test). Given the series ∑ an , consider

(9.46) lim
an 1

 L
an
if the limit exists. If L  1, the series converges; if L  1, the series diverges. For the case L  1, we can
draw no conclusion.
a

Suppose that L  1. Then there is a number r with L  r  1 such that eventually a n 1 n  r. That is,


there is an integer N such that an 1 an  r for all n  N. We conclude




(9.47) aN 

1  aN r  aN 

2  aN 

1r  aN r 2  aN 

3  aN 

2r  aN r 3 

and so forth. Thus, we have, for all k  1, aN k  aN rk , so by comparison with the geometric series, our


series converges.
If on the other hand, L  1, there is a number r, L  r  1, such that eventually a n 1 an  r. Follow-



ing the same argument but with the inequalities reversed, we conclude that for all k  1, a N k aN rk , so




we have divergence by comparison with the geometric series. We can conclude nothing if L 1. This is
the case for the all the series of the type ∑ 1 n p, and as we have seen, for some p we get convergence,
and divergence for other p.

an 
Example 9.15 ∑ n!
n 1
We try the ratio test.

 an 1 n! 


an 1 a
 


(9.48) 0
an n 1
! an n 1
as n ∞, so the ratio test gives us convergence.

2n n3 
Example 9.16 ∑ n
n 1 3
Try the ratio test:

(9.49)
an 1


 2n


1 n  1
3 3n  2


n  1 3
2
3n 1 2n n3


an 3 n  3
so we have convergence.
∞ 
Example 9.17 ∑ rn
n 1
Here the ratio test gives

(9.50)
an 1 

 r
an
Chapter 9 Sequences and Series 138

so we conclude that the series converges if r  1, and diverges if r  1. This may seem to be a sim-
plification of proposition 9.6, but in fact it is a fraud. The argument is circular, for we have used the
convergence of the geometric series to derive the ratio test.
We observe that we didn’t really need to know that the limit of an 1 an exists, only that eventually 

these ratios are either less than some number less than 1 to conclude convergence, or greater than some
number greater than 1, for divergence.

9.3. Absolute convergence


There are new difficulties when we have to consider series of negative as well as positive terms. For
example, although the harmonic series ∑ 1 n diverges, if we alternately change signs, the series now
converges.

 
  

  1
n 1


1 1 1
Example 9.18 The series 1 
2 3 4
 ∑ n
converges.
n 1
To see this, we look at the sequences of even partial sums and odd partial sums separately. Since

  1 1
(9.51) s2 n 

1 
s2n
2n  1

2n  2
 s2n

the sequence of even partial sums is increasing. Similarly,

 1
 1
2n  2 
(9.52) s2 n 

1


1
s2n 

1   s2n 

1


2n 3
tells us that the sequence of odd partial sums is decreasing. Now

  1
(9.53) s2n 

1 s2n
2n  1
 s2n 

that is, the odd partial sums are all greater than all the even partial sums. So both sequences are bounded,
and thus converge. But, they converge to the same limit, as we see by taking the limit in equation 9.53:

  lim 1 
(9.54)

lim s
n ∞ 2n 1



lim s
n ∞ 2n  2n  1 lim
n ∞ s n ∞ 2n



since 1 2n 1 0
. Since they both converge to the same limit, the full sequence also converges, and
to the same limit.
This argument actually generalizes to any alternating series, a series whose terms alternate in sign.
 ∞
∑ 
Proposition 9.14 If an is a decreasing sequence, and lim an

n ∞
0 then the series
n 1
1
n an converges.

Definition9.4 Given a sequence an , we say the series ∑ an converges absolutely if, for the series formed
of the absolute values  an  , we have convergence: ∑  an   ∞.

Proposition 9.15 If a series converges absolutely, it converges. That is,



(9.55) If ∑  an  ∞ then ∑ an converges
9.4 Power Series 139

To see that, let sn be the nth partial sum of the sequence, pn the sum of all the positive terms making
up sn , and qn the sum of the absolute values of all the negative terms. Then

(9.56) sn  pn  qn

Both sequences pn and qn are increasing, and bounded by ∑  an  , so converge, to, say p, q respectively.
Then

(9.57) ∑ an  
lim sn
n ∞


lim pn  lim qn
n ∞ n ∞ 
 p q

Because of this peculiarity of sequences of terms with alternating signs, we shall be most interested in
absolute convergence. We can use the tests of section 9.3 (applied to the series of absolute values), to
test for absolute convergence.
∞ 
Example 9.19 ∑ xn converges for  1 x 1
n 1
This is because the sum of the absolute values is just the geometric series.
∞ 
Example 9.20 ∑ n2 xn converges for  1 x 1
n 1
Here we use the ratio test for the absolute values;

(9.58)
 an 

1  n  1
2  x  n


1

n  1

2  x  x

 an  n2  x  n n

Thus, we get convergence for x of absolute value less than 1.

9.4. Power Series

∞  The point c is called the center of


Definition9.5 A power series is a series of the form ∑ an x  c
n
n 0
the power series.
A power series definesa function on the set of points for which it converges by

 ∞ 
(9.59) f x
∑ an x  c
n
n 0

The series provides an effective way of approximately evaluating the function f ; our goal in these last
sections is to show that most functions do have a power series representation. We can use the ratio test
to determine the question of convergence. We take the ratio of successive terms of (4):

 an 1  x  c n 1
  an 1


 

(9.60)  x  c L x  c 
 an   x  c n  an 
Chapter 9 Sequences and Series 140

 

if the limit L limn ∞  an 1   an  exists. In this case the series converges absolutely for  x  c   1 L,


and diverges for  x  c   1 L. Thus, the domains of convergence and divergence of the series are
separated by the circle, centered at c of radius 1 L. It can be shown that, in general, there is a circle
separating these domains, even if the limit of the ratio of successive coefficientsdoesn’t exist.

Proposition 9.16 Given the power series representation f x


 ∞
∑ an x  c
n
 there is a number R  0 

n 0
R ∞ such that we get absolute convergence for all x   x  c   R, and divergence for all x   x  c   R.
a  


1
R is called the radius of convergence of the power series. We have this value of R: lim n 1  if



n ∞  an  R
the limit exists.
The firstexample of a power series representation is that of the geometric series:

∑ xn  

1
Example 9.21 for  x   1 has the radius of convergence R 1.
n 0 1 x

∞ 
Example 9.22 ∑ nk xn converges for  x   1 for any number k. We use the ratio test. The ratio of
n 0
successive coefficientsis

(9.61)
n  1
k

n  1

k
1
nk n
as n ∞.

xn  
Example 9.23 ∑ n! has radius of convergence R ∞ Using the ratio test:
n 0

1 1  1
(9.62)
n  1
! n! n  1
0

so R  ∞, and the series converges for all x. On the other hand, the ratio test shows us that the series

(9.63) ∑ n!xn
n 0

has radius of convergence R  0, so converges only for x  0.


Newton thought of power series as “generalized polynomials” - that is, as polynomials, only longer.
This is justified,because we can operate with power series just as we operate with polynomials: we can
add, multiply, and substitute in them by doing so term by term.
x  ∞ 
∑ xn 1
for R  1 For


Example 9.24
1 x n 0

(9.64)
x  x

1  x 1  x x  2
x3 
 
 x  x2  x3  x4  


1 x 1 x

1  ∞
1  ∞ 
Example 9.25
1 x2 ∑ x2n  1  x2 ∑  1
n x2n for x  1
n 0 n 0
9.4 Power Series 141


To see the first,we note that 1 1  x2
is obtained from 1 1  x
by substituting x2 for x. Thus, the
power series representation is obtained in the same way. In the second, we have substituted  x 2 for x.

Example 9.26 Find a power series expansion for 1 5  2x


centered at the origin. What is its radius
of convergence?


To solve a problem like this, we have to relate the function to another function, whose power series


we know. In this case that would be 1 1  x
. Now 5  2x 5 1  2 5
x
, so our function is obtained
from 1 1  x
by firstreplacing x by 2 5
x, and then dividing by 5. We follow the same instructions
with the power series.
Start with
1  ∞ 
(9.65)
1 x ∑ xn
Replace x by 2 5
x:
n 0

 ∞ 

1 2
(9.66)
1  2 5
x ∑ 5 x
n
n 0

Divide by 5 and clean up:


1  1 ∞ 2  ∞
2 n xn
5 n∑0 5 ∑ 5n
n
(9.67) x

5  2x 1


n 0

We can calculate the radius of convergence using proposition 9.16, or we can reason as follows; since
the series we started with converges for  x   1, our finalseries converges for  2 5
x   1, or  x   5 2.
Finally, we can also integrate and differentiate power series term by term:

Proposition 9.17 Suppose that f x


 ∞
∑ anxn has radius of convergence R. Then
n 0

(9.68) 
x
f t
dt  ∑ n

an
xn


1

0 n 0 1

(9.69) f x
 ∞
∑ nanxn  1

n 1

and both have the same radius of convergence, R.

 ∞
 1
n 
∑ 2n  x2n 1


Example 9.27 arctanx



n 0 1
We know that the derivative of the arc tangent is 1 1 x2
. Now, in example 9.25, we have already
found the power series representation of that function, so we obtain the power series representation of
arctanx by integrating term by term.

 ∞
xn  
Example 9.28 ex ∑ n! for all x Let f x
∑∞ n
n 0 x n! Then, differentiating term by term, we
n 0
find

(9.70) f x
 ∑

nxn  1
 ∑

xn  1  ∞
∑ n!
xn

n 1 n! n 1 n  1
! n 0
Chapter 9 Sequences and Series 142

 

where the last equation is obtained by replacing the index n by n 1. Thus f x

differential equation, y y, definingthe exponential function. Since f 0

f x
, so satisfiesthe
1 also, it is the exponential 
function.

Example 9.29 e   ∞
nx
2n 
x2
∑  1
n!
for all x Just replace x in example 9.28 by  x2 .
n 0

9.5. Taylor Series


Finally we tackle the question: how do we find the power series representation of a given function?
Recalling that the purpose of the power series is to have an effective way to approximate the values of
a function by polynomials, we turn to that question: what is the best way to so approximate a function?
We start with a function f that has derivatives of all orders definedin an interval about the origin. To
begin with, we recall the definitionof the derivative in this context:

f x
 f 0
 
(9.71) lim
x 0  x
f 0

If we rewrite this as
f x
 f 0
 f 0
x

(9.72)

lim
x 0 x
0

we see that the linear function y f 0


 
f 0
x approximates f x
to firstorder: f 0
f 0
x is closer 
which approximates f to second order? Let y a bx cx2 be such a polynomial. Then we want  
to f x
than x is to zero, and by an order of magnitude. We now ask, can we finda quadratic polynomial

f x
 a  bx  cx2
 0
(9.73) lim
x 0  x2


We calculate this limit using l’Hˆ
a f 0
. Then
opital’s rule. First of all, for l’Hˆ
opital’s rule to apply, we have to have

f x
 f 0
bx   cx2
 f x
 b  2cx

 
lH
(9.74) lim lim
x 0 x2 x 0 2x
We can apply l’Hˆ
opital’s rule again, if we have b  f 0
:

f x
 f 0
 2cx
 f x
 2c 
 
lH
(9.75) lim lim 0
x 0 2x x 0 2
if c 
f 0
2. We conclude that the polynomial

(9.76) f 0
 f 0
x  f 0
2
2
x

approximates f to second order: this is closer to f x


than x is to 0 by two orders of magnitude. Further-
more, it is the unique quadratic polynomial to do so.
9.5 Taylor Series 143

We can repeat this procedure as many times as we care to, concluding


Proposition 9.18 The polynomial which approximates f near 0 to nth order is

(9.77) f 0
 f 0
x  f 0
2
x  
   f n 

0

2 n!

Of course we can make the same argument at any point, not just the origin. To summarize:
Definition9.6 Suppose that f is a function with derivatives at all orders definedin an interval about
the point c. The Taylor polynomial of degree n of f , centered at c is

 n
f k c




(9.78) Tc n f
x

x  c
k
k 0 k!

Proposition 9.19 The Taylor polynomial Tc n f is the polynomial of degree n which approximates f near


c to nth order.
So, we can compute effective approximations to the values of f x
near c by these Taylor polynomi-
als; but the question is, how effective is this? More precisely, what is the error? We use this estimate:
Proposition 9.20 Suppose that f is differentiable to order n 1 in the interval c  a  c a centered at  
the point c. Then the error in approximating f in this interval by its Taylor polynomial of degree n, Tc n f 

is bounded by
Mn 1

 x  c n 1





(9.79)
n 1
!

where Mn 

1 is a bound of the values of f n 1




over the interval c  a  c  a . To be precise, we have the


inequality
Mn 1 
 f x
 Tcn f x


 x  c n 1


(9.80) 

n 1
!

In the next chapter we will show how the error estimate is obtained, and see how to work with it.
What we want now is to concentrate on the representation by series.
Definition9.7 Let f be a function which is differentiable to all orders in a neighborhood of the point c.
The Taylor series for f centered at c is

(9.81) Tc f x
 ∞

f n 

x  c
n
n 0 n!

If c is the origin, this series is called the Maclaurin series for f .

Proposition 9.21 Suppose that f is a function which has derivatives of all orders in the interval c 

a  c a
, Let Mn be a bound for the nth derivative of f in the interval. If the sequence
Mn
(9.82)  x  c n 0
n!
Chapter 9 Sequences and Series 144

converges to zero for all x in the interval, then f is given by its Taylor series:

(9.83) f x
 ∑

f n 

x  c
n
n 0 n!

in c  1  c  a
.
As an example, ex has the Maclaurin series

 ∞
xn 
(9.84) ex ∑ n!
n 0


We have already shown by other means. We can verify this using proposition 9.21 as well, since the nth
derivative of ex is still ex , and the value at x 0 is 1. By a parallel calculation we obtain the power series
representation of ex centered at any point:
Example 9.30 For c any point, the function ex has the Taylor series representation centered at c:

 ∞
ec 
(9.85) ex ∑ n! x  c
n
n 0

We do have to verify that the remainders converge to zero; that is the terms 9.82 converge to zero. Since
  

ex is an increasing function, its maximum in the interval a  c  a c is at x a c, so we can take
Mn ea c . Then, for the exponential function we have


Mn   x  c n 
  x  c n ea

c


(9.86) lim lim 0


n ∞ n! n ∞ n!
by example 9.24.
It is useful to make the following observation
Proposition 9.22 Suppose that f has a power series representation:

 ∞ 
(9.87) f x
∑ an x  c
n
n 0

Then, this is its Taylor series. More precisely:

(9.88) an  f n 

c

n!

This is easy to see; if we differentiate 9.87 k times we obtain:

 ∞

  n  k
an x  c
n 

(9.89) f k 

x
∑ n n  1
k
n k


Now, let x c and obtain f k c


k!ak , for all terms but the firsthave the factor x  c.
So, if we have found a power series representative of a function, then that is automatically the Taylors
series for the function.
9.5 Taylor Series 145

Example 9.31 Find the Maclaurin series for the function f x


1  x 5x2  x3 . Since a polynomial  
is already expressed as a sum of powers of x, that expression is a power series, and thus the Maclaurin
series for the polynomial.
  
Example 9.32 Find the Taylor series centered at c 1 for the function f x

have to findthe values of the derivatives of f at c 1:  1 x 5x 2  x3 . We

(9.90) f 1
 4

(9.91) f x
  1  10x  3x2  so f 1
 6

(9.92) f x
 10  6x  so f 1
 4

(9.93) f x
  6 so f 1
  6

and all higher derivatives are zero. Thus the Taylor series is

(9.94) f x
 4 6 x  1
 4
x  1
2 
6
x  1
3  4 6 x  1
 2 x  1
2  x  1
3

2! 3!

Now, we can findthe Maclaurin series for many functions, so long as we know how to differentiate
them. Following is a list of the most important Maclaurin series.
Proposition 9.23
1 ∞

1  x n∑0
a) xn   x   1

b) ex ∑ ∞ n
x
n 0 n!

c) cosx ∑


 1
n 2n
x
n 0 2n
!

d) sin x ∑ 

 1
n 2n 1



x
n 0 2n 1
!

e) arctanx ∑


 1
n 2n 1



x
n 0 2n 1

We have already seen how to get (a), (b) and (d). For the trigonometric functions, we proceed as
follows. First, the cosine:

(9.95) f 0
 1

(9.96) f x
  sin x  so f 1
 0

(9.97) f x
  cosx  so f 1
  1
Chapter 9 Sequences and Series 146

(9.98) f x
 sin x  so f 1
 0
(9.99) f iv 

x
 cos x  so f iv 1
  1
Thus, up to four terms we have

(9.100) cosx  1
x2
2!
 x4
4!
 
  
But, now, since we have returned to cos x, the cycle 1  0   1  0 repeats itself again and again. We
conclude that

(9.101) cos x  1
x2
2!
 x4 x6

4! 6!
 x8
8!
 
  
which can be rewritten as iii
of proposition 9.23 above.
One finalTaylor series is worth noting: since the integral of 1 1  x
is  ln 1  x
, we can findthe
Taylor series centered at 1 for ln x as follows:

(9.102)
1  ∞
∑ tn 
1 t n 0

 ∞
tn 1



(9.103)  ln 1  t
∑n 1 
n 0

Now, make the substitution x  1  t, so t  1  x:

lnx 

1  x
n 1


 ∞
x  1
n 1



 ∑  ∑  1
n 1



(9.104)
n 0 n 1 n 0 n 1

Anda mungkin juga menyukai