Math 31B: Sequences and Series
Math 31B: Sequences and Series
Michael Andrews
UCLA Mathematics Department
October 9, 2017
1 Sequences
1.1 What is one?
A sequence is a list which goes on forever. Heres an example.
31, 30, 31, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, . . .
This sequence lists the number of days in each month starting in October
2017. There are some things we can demonstrate with this sequence.
Theres not a particular nice formula for this sequence and that doesnt
matter.
31, 30, 31, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, . . .
1, 2, 3, 4, 5, . . .
1 1 1 1
1, 2, 3, 4, 5, ...
2, 4, 8, 16, 32, . . .
1 1 1 1 1
2 , 4 , 8 , 16 , 32 , ...
1
The above sequences have nice formulas for their n-th term. We have
1 1
an = n, an = , an = 2n , an = n ,
n 2
respectively.
4. Plugging in a big enough positive integer into the formula an = 21n will
force a rubbish calculator to return 0. The sequence converges to 0.
2
with the formal definition. There is not much time, and so I will not expect
you to come to terms with the formal definition, but you might still find it
useful to think about.
Similarly, there is formal definition of what it means for a sequence (an )
to diverge to . When this definition is satisfied we write
lim an = .
n
Sequences via functions. Suppose an = f (n) for some function f (x) and
that limx f (x) = L. Then limn an = L.
The point of this theorem is that a sequence only has values for each posi-
tive integer:
it is a list. A function takes on even more values: it can make
10
sense at 2, e, , 103 . This means that the condition limx f (x) = L is
a stronger one than limn f (n) = L. However, once we have a function,
methods of calculus (e.g. LHopitals rule) might be applicable, whereas,
before they were not.
For example, if you want to calculate limn (1 + n1 )n , then it is enough
to calculate limx (1 + x1 )x . We have seen, using LHopitals rule, that
limx (1 + x1 )x = e, and so limn (1 + n1 )n = e.
3
The same is true for sequences. We remember the limits of our sequence
friends, and most other limits will follow from some rules about convergent
sequences. Here are your two best sequence friends.
1
1. The sequence with n-th term an = n converges to 0. That is,
1
lim = 0.
n n
2. If r is a number with 1 < r < 1, then the sequence with n-th term
an = rn converges to 0. That is,
lim rn = 0.
n
1. limn k = k;
2. limn (kan ) = k (limn an );
3. limn (an + bn ) = (limn an ) + (limn bn );
4. (an + cn ) diverges;
5. limn (an bn ) = (limn an ) (limn bn );
6. limn ( abnn ) = limn an
limn bn , as long as limn bn 6= 0;
7. limn f (an ) = f (limn an ).
4
Next, we calculate
1 1 1 1 1 1
lim 9 + 3 + 227 = lim 9 + lim 3 + lim 227
n n n n n n n n n n
1 1 1
= lim 9 + 3 lim + 227 lim
n n n n n n
1 1 1
= lim 9 + 3 lim + 227 lim lim
n n n n n n n
= 9 + 3 0 + 227 0 0 = 9.
The first equality uses 3; the second uses 2; the third uses 5; the final equality
uses 1 and the fact that limn n1 = 0.
Since x is continuous, 7 tells us that
s
r
1 1 1 1 1 1
lim 9 + 3 + 227 = lim 9 + 3 + 227 = 9 = 3.
n n n n n n n n
q
Similarly, we can verify that limn 4 + 2 n1 + n1 n1 = 2. Finally,
q
2
4n + 2n + 1 4 + 2 n1 + n1 n1
lim = lim q
n 9n2 + 3n + 227 n 9 + 3 1 + 227 1 1
n n n
q
limn 4 + 2 n1 + n1 n1 2
= q = .
lim 9 + 3 1 + 227 1 1 3
n n n n
5
In calculating such a limit, this is the standard technique to show that the
highest degree terms in the numerator and denominator are all that matter.
1 1 1 1 1 1 1
, ,
20, 20, 20, 20, 20, 20, 20, , , , , , ...
8 9 10 11 12 13 14
1 1 1 1 1 1 1 1 1 1 1
, , , , , , , , , , , ...
100 101 102 103 104 105 106 107 108 109 110
also converge to 0.
lim SQUEEZEDn = L.
n
sin n
Example. limn n = 0 since n1 sin n
n n1 , and limn n1 = 0;
we can take
1 sin n 1
LOWERn = , SQUEEZEDn = , and UPPERn =
n n n
in the squeeze theorem.
n n 8
Example. limn 8n! = 0 because 0 8n! 88! ( 98 )n8 ; you could also
n 8
use the inequality 0 8n! 88! n8 .
Well expand on the second example after the discussion.
6
HELP! Students always struggle with finding the lower and upper sequences
in the squeeze theorem. Here are some pointers to help you.
The first thing to note is the purpose of the squeeze theorem. It is used
to formalize intuition you have about why a sequence converges to a limit.
n
In the previous example, ( 8n! ), Id have said, I think that n! grows faster
than exponents, so my guess is 0. This is the first part of using the squeeze
theorem.
As soon as you have made a sensible guess, L, for the limit of the sequence
in question, this imposes conditions on what your lower and upper sequences
can be. If the following bullet points are not fulfilled then you have gone
badly wrong.
(LOWERn ) and (UPPERn ) must have the SAME limit, and that limit
better be your guess L; otherwise, youre not squeezing! For example,
suppose limn LOWERn = 0 and limn UPPERn = 1. If we take
0 to be BROAD2160E and 1 to be the hill, this is like asking can your
friends both be huggging you if one is in my lecture, and the other is
up the hill? Theyd need bloody long arms!
With all of these points in mind, the most difficult part is the following.
7
This sequence should converge to 0 since I think n! grows faster than
exponents.
It is hopefully clear that LOWERn = 0 is a fine choice. (0) has the easiest of
all limits to calculate, it is 0, and it is trivially true that 0 SQUEEZEDn
because the sequence under consideration consists of positive terms.
To figure out (UPPERn ) requires more thought about the sequence under
consideration. Imagine that you bet your mothers house on the fact that
this sequence converges and you start writing down the terms in the sequence
one after another.
8 8 8 8 8 8 8 8 8 8 8 8 8 8 8
, , , , , ...
1 1 2 1 2 3 1 2 3 4 1 2 3 4 5
OH NOOOOOO! We keep multiplying by a number bigger than 1. This is a
disaster. What am I gonna tell her? Why wasnt I better at math? Whyd
a make a stupid bet? Never bet, she told me, unless its selling all your
pounds before the EU referendum. Fingers crossed. The 9-th term. . .
8 8 8 8 8 8 8 8 8
.
1 2 3 4 5 6 7 8 9
HOLY CRAP! We just multiplied by a number less than 1 to get from the
8-th term to the 9-th term! Maybe its gonna be alright? Maybe I should
go to Vegas after class?! Buy her a condo on Wilshire Blvd?
8n
8 8 8 8 8 8 8 8 8 8 8 8 8
=
n! 1 2 3 4 5 6 7 8 9 10 11 n1 n
8
than or equal to 89 . If there are n terms in total, and eight crappy terms,
then there are n 8 lovely terms. We just said:
8n
8 8 8 8 8 8 8 8 8
=
n! 1 2 7 8 9 10 11 n1 n
n8
88 88
8 8 8 8 8 8
=
8! 9 10 11 n1 n 8! 9
88
Can we take UPPERn = 8! ( 98 )n8 ? Well, it is a friend: we can see this by
writing it as
88 n
8! 8
.
( 89 )8 9
It doesnt involved any n!; sure, theres an 8!, but this is not depending on
n, its a FIXED number. Since 89 < 1, the sequence converges to 0. We just
proved the requisite inequality. DONE!
If wed have enjoyed the thrill of betting and not celebrated 89 so much,
8
we might have let UPPERn = 88! n8 . This works since
8n 88
8 8 8 8 8 8 8 8 8 8
= 1 .
n! 1 2 7 8 9 10 11 n1 n 8! n
If one cares about the development of the real numbers, then these results
are actually incredibly important. We wouldnt know e existed without such
theorems. In Math 31B, we just need to know these results are out there.
9
1.9 Recursively defined sequences (non-examinable)
Let a1 = 1 and for n > 1, an = 2an1 . This means
r q
q
a1 = 1, a2 = 2, a3 = 2 2, a4 = 2 2 2, . . .
To get a feel
for this you could press 1, then = on your calculator,
then enter 2 ANS, and press = over and over again. What does your
answer end up at? It should be 2.
To prove this mathematically there are two steps.
1. Pretend there is an answer, and figure out what it must be using limit
laws.
We can see that an 1 for all n and so L cannot be zero. Thus, L = 2, i.e.
limn an = 2.
To verify that such an L actual exists, we use the driving-a-car-at-a-wall
theorem. We will show that an an+1 2 for all n. Driving-a-car-at-a-wall
will then say that (an ) converges to some L 2. It is important that you
know that you can use such a theorem. But I will never ask you to show
that a sequence is increasing and bounded above; I would always tell you
such information.
a1 = 1 2 = a2 and, if n > 1 and an1 an , then
p
an = 2an1 2an = an+1 .
10
Also, a1 2 and, if n > 1 and an1 2, then
p
an = 2an1 2 2 = 2.
From this (and induction) we learn that an an+1 2 for all n. The keen
student would ask how I guessed 2 was an upper bound. The way I have
presented the problem answers that question: I knew what the sequence had
to converge to before I even showed it converges.
11
2 Series
2.1 What is one?
A series is an infinite sum. The difference with a sequence is that the commas
are replaced by addition signs. For example,
1 + 2 + 3 + 4 + 5 + ...
1 1 1 1
1+ + + + + ...
2 3 4 5
2 + 4 + 8 + 16 + 32, . . .
1 1 1 1 1
+ + + + + ...
2 4 8 16 32
2.2 -notation
are integers, p q, and ap , ap+1 , . . . , aq1 , aq are real numbers, we
If p, q P
write qn=p an for
ap + ap+1 + . . . + aq1 + aq .
Here, n isR the indexing number. You can think of this just as you think of
b
the x in a f (x) dx. p and q tell you where to start and end the summation.
You can think of these like the limits in a definite integral. Just like with
integration, and anything in math, we can use different letters. Sometimes
it is useful to index by k instead. I will try to stick with n unless I have to
use another letter.
We can use -notation for series too. For the above series, we write
X X 1 X n X 1
n, , 2 , .
n 2n
n=1 n=1 n=1 n=1
12
Wed like to make sense of
X
an = a1 + a2 + a3 + . . .
n=1
Let
N
X
sN = an = a1 + a2 + a3 + . . . + aN .
n=1
Then (sN )N =1 is called the sequence of partial sums, and we can talk about
its convergence or divergence. This is exactly the same as talking about the
convergence or divergence of the series
P
n=1 an . In one equation:
X N
X
an = lim an = lim sN .
N N
n=1 n=1
Later on, you will see the similarity to improper integrals. In that case, we
will make sense of an improper integral as a limit of proper integrals. Now,
were making
P sense of an infinite sum as a limit of finite sums. Considering
the series n=1 an is exactly the same as considering the sequence
a1 , a1 + a2 , a1 + a2 + a3 , a1 + a2 + a3 + a4 , . . .
As an example of the content of the previous discussion, lets turn to our
old sequence friend, the one with an = n1 . We love this sequence and will
never forget how grateful
P we1 are that it converges to 0.
Its cousin series, n=1 n , has a different personality. We love it too,
but for different reasons. This series is so special that it has a name: the
harmonic series. Lets show it diverges.
2
s1 = 1 =
2
1 2 1 3
s2 = s1 + = + =
2 2 2 2
1 1 3 1 4
s4 = s2 + + + =
3 4 2 2 2
1 1 1 1 4 1 5
s8 = s4 + + + + + =
5 6 7 8 2 2 2
1 1 1 5 1 6
s16 = s8 + + + ... + + =
9 10 16 2 2 2
1 1 1 6 1 7
s32 = s16 + + + ... + + =
17 18 32 2 2 2
13
We find that s2n n+2
2 so that the sequence (sn )n=1 diverges. By definition,
this means the harmonic series diverges.
This is important. Most people do not expect this result. It is tempting
to think that the series will converge since the terms get smaller; far along
in the summation, you are adding on a very small amount. The calculation
above shows, that although the terms start becoming very small, by grouping
them together cleverly, in sums which are a power of 2 long, they are still
significant.
IfRyou are happier with the improper integral stuff, think about the fact
that 1 x1 dx diverges. It is the same deal: the curve gets real close to the
x-axis, the area available under it is less and less, but theres still enough
that there is an infinite area overall, since limS ln S = .
14
2.5 Rules for series
Suppose
P P P
n=1 an and n=1 bn are convergent series, n=1 cn is a divergent
series, and that k is a real number.
P P P
1. n=1 kan converges to k n=1 an ; if k 6= 0, n=1 kcn diverges.
P P P
2. n=1 (an + bn ) converges to n=1 an + n=1 bn .
P
3. n=1 (an + cn ) diverges.
PNotice, that there are far fewer rules. For instance, there is not one for
n=1 an bn . This is because, even in the finite case, it is not always true
that (a1 + a2 )(b1 + b2 ) = a1 b1 + a2 b2 ; only a Freshman dream would allow
such a thing.
An example of these rules (and recognizing geometric series) is as follows.
8 + 2n+5 X 8 2n+5
X
= +
5n 5n 5n
n=1 n=1
2n+5
X X
8
= +
5n 5n
n=1 n=1
n n
8 15 25 ( 25 )
X X
1 5 2
= 8 + 2 = +
n=1
5
n=1
5 1 15 1 52
P 2 P 1
n=1 n2 and n=1 n3 converge (by the p-test) and the rules say that
X
X 2 1 2 X 1
+ = + .
n2 n3 n2 n3
n=1 n=1 n=1
P (n+1)2
We can also use the rules to show n=1 n3
diverges. First, notice
(n + 1)2 n2 + 2n + 1
X X X 1 2 1
= = + + .
n3 n3 n n2 n3
n=1 n=1 n=1
Since
P 1 P 2 1
n=1 n diverges
P and n=1Pn2 + n3 converges part 3. applies.
It is true that n=1 2n1 1
and 1
n=1 2n+1 diverge (later on, you will be
able to
Pprove this using the
limit comparison test). However, it is NOT true
that n=1 2n1
1
1
2n+1 diverges; it converges to 1 (see the homework).
The rules say nothingPabout this situation
Pand they cannot because of the
following silly series: n=1 (1 + 1) and n=1 (1 1).
15
3 A philosophical discussion
This section is completely non-examinable but some of the things mentioned
are important to understand. Without thinking about some of these issues,
I dont think you can really understand any of the questions Im asking you
about series!
So, you ask, arent series just a special case of sequences? And, if so, why
do we bother talking about them?
Well, yes, series are just a special type of series. But, heres the issue. . .
A non-series-type sequence often has a nice and simple formula attached to
it; in this case, it is often possible to calculate the limit of the sequence. On
the other hand, when we use sums to define the terms of a sequence, our
formula often becomes complicated so that it is difficult or even impossible
to calculate what the limit might be.
For example, what do you think
X 1 1 1 1 X 1 1 1 1
= 1 + + + + . . . and =1+ + + + ...,
n2 4 9 16 n3 8 27 64
n=1 n=1
converge to?
A sensible guess for the first was made by Euler using an infinite product
x2
Y
sin x = x 1 2 2 .
n
n=1
I say guess, but, at the time (Euler lived from 1707 to 1783), his argument
2
was accepted as a proof. His answer was 6 and this turned out to be correct.
You can also let x = 2 in his product formula to see that
2 2 4 4 6 6 8 8 10 10 12 12 14 14
= .
1 3 3 5 5 7 7 9 9 11 11 13 13 15 2
16
This is Wallis product for (1655). On wikipedia, there is a proof of this
result using integration by part and the squeeze theorem: youve made it to
17th century mathematics!
The second number is called Aperys constant. In 1978, it was proved
that this number is irrational, meaning not a fraction. It took until then to
know this!
The integral defines the normal distribution with mean 0 and variance 1;
normal distributions describe many things in real life. The integral is im-
possible to evaluate exactly. However, it is important to know it converges.
Approximations to it are very useful and can be found in any statistics book.
[We will see, later, that it converges by comparing with the integrand with
function e1|t| .]
The series defines the cosine function. Mathematicians can prove all the
familiar formulae involving cosine using this definition. The pitfall of the
formula is that it doesnt allow us to know the values of cos exactly. But,
this never phased you before; you were happy with your calculator giving
you a decimal to 10 decimal places. Now, you can always use Taylors Error
Bound to calculate values of cosine as accurately as you need. [We will see,
later, that the series defining cosine converges using the ratio test.]
In lecture, I will note that decimals only make sense because of series!!
My favorite application of infinite series is called Fourier series, which
tells you about the amplitude of the harmonics in a periodic signal: part of
the reason musical instruments sound different to one another.
Having said all this, I think I now feel less bad about telling you all the
tests for convergence/divergence of integrals and series.
17
4 Tests for convergence/divergence of a series
4.1 The n-th term test
In the previous section, we remarked that most people do not expect for the
harmonic series to diverge: the terms get smaller and smaller; were adding
on less and less. It is important to remember the harmonic series.
The terms getting small is not enough for a series to converge.
On the other hand, you would not expect
1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + ...
to converge. The terms do not get small; we are always adding on a sig-
nificant amount; we cannot expect the series to converge. To see this even
more explicitly, lets consider the sequence of partial sums.
s1 = 1
s2 = 1 + 1 = 2
s3 = 1 + 1 + 1 = 3
s4 = 1 + 1 + 1 + 1 = 4
s5 = 1 + 1 + 1 + 1 + 1 = 5
We have sn = n, and so (sn ) is very definitely divergent.
6 0, then
P
The n-th term test. If limn |an | = n=1 an diverges.
18
The second is the geometric series, with r = 12 , with its first 100 terms
replaced by 1000. This does not suddenly make it diverge. It just changes
its value by the number
99
X 1
100, 000 .
2n
n=0
When someone asks you about the convergence of a series, it is useful to
know that you can ignore the first 1, 000, 000, 000 terms if it is convenient.
For this reason, in later theorems, you only have to check hypothesis for
sufficently large n even if it says for all n.
Ignoring the first 100 values in a convergent sequence, will change the
value which it converges to, however. So if youre asked for a value, dont
randomly ignore the first 100 terms; youd be silly to do so.
2. f (x) 0;
Example.
Let f (x) = x1p , where p > 0. Then f (x) satisfies the hypothesis of the
integral test. Thus,
P 1 R 1
1. n=1 np converges if and only if 1 xp dx converges.
P 1 R 1
2. n=1 np diverges if and only if 1 xp dx diverges.
19
This means 1
P
n=1 np converges if p > 1 and diverges if p 1 (this includes
p 0, although we are not using the integral test for this).
Important. The integral test can be used to show a series converges, but
it does not say much about what it converges to. For example,
2
Z
1 X 1
dx = 1 and = .
1 x2 n2 6
n=1
If one P
is careful, and understands thePproof of the integral
R 1 test, all it says
about 1 P 1 1
n=1 n2 is that n=1 n2 = 1 + n=2 n 2 1 + 1 x2 dx = 1 + 1 = 2.
Additional. In light of the remarks made in section 4.2, one can relax
all the assumptions in the integral test, demanding only that they are true
eventually. How does one express this mathematically? We mean that there
is an M 0 such that:
an = f (n) when n M ;
f (x) 0 for x M ;
In this case,
P R
1. n=1 an converges if and only if M f (x) dx converges.
P R
2. n=1 an diverges if and only if M f (x) dx diverges.
Honest thoughts. I like the integral test because it has an intuitive proof
that highlights many of the ideas which are going on in this stuff. However,
this course is not dedicated to proofs. For this reason, I feel the integral test
is verging on useless for us. After using it to say when p-series converge, one
almost never needs it; other tests are more practical. Did I waste your time
by setting questions on it? No! Thankfully, the saving grace is that to use
the integral test one has to practice improper integrals, u-subs with them,
integration by part with them, and LHopitals rule, so it is good for review.
Also, it gets you to grips with checking hypotheses. In the integral test, the
first few are normally very easy, but the decreasing condition might require
a little work. Since I dont care much for the integral test, I wont grill you
on this on the exam.
20
Example. Heres P one1example where I would choose the integral test over
other tests. . . n=2 n ln n diverges. This is because
Z Z S
1 1
dx = lim dx
e x ln x S e x ln x
Z ln S
1
= lim du
S 1 u
ln S
= lim ln |u|
S 1
= lim ln(ln S) ln(1) = lim ln(ln S) = .
S S
1
and, letting f (x) = x ln x , we have
1. f (x) is defined for x e and continuous;
2. f (x) 0 for x e;
3. f 0 (x) = (x1+ln x
ln x)2
0 for x e, so that f (x) is decreasing for x e.
(You could also deduce this from the fact that x and ln x are increasing
and positive for x > 1, but the derivative is a tool that will pretty much
always work for you.)
Example. Heres one P other 1example where I would choose the integral test
over other tests. . .
n=2 n(ln n)2 converges. This is because
Z Z S
1 1
dx = lim dx
e x(ln x)2 S e x(ln x)2
Z ln S
1
= lim du
S 1 u2
ln S
1
= lim
S u 1
1
= lim 1 = 1,
S ln S
1
and, letting f (x) = x(ln x)2
, we have
1. f (x) is defined for x e and continuous;
2. f (x) 0 for x e;
3. f 0 (x) = x2+ln x
2 (ln x)3 0 for x e, so that f (x) is decreasing for x e.
21
4.4 Direct comparison test
The direct comparison test is my favorite test. It is the most theoretically
useful by a long way; it is, arguably, the most practically useful. However,
it is also the one that students find the most difficult.
Example 1.
How do we use this theorem? Well, suppose we want to prove the con-
vergence or divergence of
X 1
.
n=1
n3 + 3n + 2
Where can we start? We go to our friends: the p-series and the geometric
series. Does it look like either? It definitely looks more like a p-series, but it
sort of looks like lots of them and that is confusing. A couple of Freshman
dreams down the line, you might say
1 1 1 1
= + + (NOOOOOO!!!!).
n3 + 3n + 2 n 3 3n 2
22
These are, indeed, Freshman dreams, and you should check that you do
not do this. However, they do highlight what the difficulty is. We have
a sum underneath a square root, at the bottom of a fraction. Square roots
play badly with sums, and sums play badly with fractions if they are on the
bottom. The only thing we can really do is to try and ignore terms or pair
them up together in some way. Lets try ignoring them and see if we get
anywhere.
Do we know about the convergence of any of
X 1 X 1 1 X 1 X 1 1 X
, = , = 1?
n=1 n3 n=1
3n 3 n=1 n n=1
2 2 n=1
YES! They are all our friends; they are all p-series. The first converges since
3 1
2 > 1. The others diverge since 0, 2 1.
Now we need to figure out what effect ignoring the various terms had.
In each case, we forgot about a positive quantity, under the square root of
the bottom of a fraction. In each case, this makes the bottom of the fraction
smaller, and so the whole thing bigger. That is,
1 1 1 1
, , .
n3 + 3n + 2 n3 3n 2
We now see that 1
P
n=1 n3 +3n+2 is smaller than a convergent thing, so we
Example 2.
Suppose we want to prove the convergence or divergence of
X 1
5
.
n=1
n4 + 6n + 3
23
YES! They are all our friends; they are all p-series. They all diverge since
4 1
5 , 5 , 0 1.
This suggests that 1
P
n=1 5 n4 +6n+3 diverges. To use the comparison test
we better have
1
BIGn = 5
.
n4 + 6n + 3
1
We need SMALLn . To make 5 4 smaller we need to make the things
n +6n+3
under the square root bigger. Wed like it if they paired up nicely, so that
no freshman dreams are required. Heres how.
1 1 1 1
5
5
=
5
5
n4 + 6n + 3 n4 + 6n4 + 3n4 10 n4
We let
1 1
SMALLn =
5
5
.
10 n4
P
n=1 SMALLn diverges by the p-test sinceP54 1. The direct comparison
test tells us that n=1 BIGn diverges, i.e.
P
1
n=1 5 n4 +6n+3 diverges.
PNotice 1that the term we compared to was the one closest to making
n=1 5 4 converge.
n +6n+3
24
Example.
1 1
Let an =
5 and bn = 5 4.
Then
n4 +6n+3 n
5
an n4 1
lim = lim
5
= lim q
n bn n 4
n + 6n + 3 n 5 1 + 63 + 3
n n4
1
=
5
= 1 6= 0, .
1+0+0
that
P P
Thus, the limitPcomparison theorem says P n=1 an diverges if n=1 bn
1 1
diverges, i.e. n=1 5 4 diverges if
n=1 5 n4 diverges.
P n +6n+3
1 4
diverges by the p-test, since 5 1, so
n=1 5 4n
X 1
5
diverges.
n=1
n4 + 6n + 3
There are also versions of the limit comparison test which account for
the cases when L = 0 or L = . The first time I taught this class, I found
that students would often misuse these versions, and so, the second time I
taught the class, I didnt encourage their use as much. Since I found that
some students still wanted to use them, Ive compromised. Questions on the
final will not require these versions - another test will always be applicable
- but if you like them, you are free to use them, and they might help you.
Look up the statements in the textbook. Heres an example of when L = 0.
Example.
10 ln n 1
Let an = n2
and bn = 3 . Then
n2
an 10 ln n 10 ln x ( 10
x) 20
lim = lim = lim = lim 1 = lim = 0.
n bn n n x x x (2 x)
x x
This limit calculation says that for large values of n, an is much smaller than
bn . (WolframAlphaP shows thatP an < bn as long as n > 8100.)
The p-test says n=1 bn =
1
n=1 32 converges and the L = 0 version of
n
the limit comparison test allows us to conclude that
X X 10 ln n
an =
n2
n=1 n=1
converges.
25
4.6 Ratio and root test
All of our tests so far have relied on comparison with one of our friends.
Wouldnt it be great if we had a test that didnt require us being so social?
1. If L < 1, then
P
n=1 an converges (absolutely).
The ratio test is particularly good if you see factorials. n! factorial grows
quicker than pretty much anything. The only way to beat it is with things
like, nn , double exponents, or double factorials. So, if you see a factorial on
the top of a fraction in the n-th term, then the series is likely to diverge,
and if you see it on the bottom of a fraction in the n-th term, the series is
likely to converge. The ratio test is then a good check, but, with the ra-
tio test, unlike the comparison tests, you do not even need an educated guess.
Example 1.
1000n
Consider the series
P
n=0 n! . The n! on the bottom of the fraction
making up the n-th term makes us think it will converge, but even if we
n
didnt think this, we can use the ratio test with an = 1000
n! and we get
1000n+1
an+1 1000
lim = lim (n+1)! n = lim = 0 < 1.
n an n 1000 n n + 1
n!
P 1000n
Thus, n=0 n! converges. In fact, it converges to e1000 .
26
Example 2.
en
Consider the series
P
n=1 nn . Using the ratio test becomes a little com-
plicated: we have
en+1
e nn
an+1 n+1 1 e
= (n+1) = =
en n + 1 (1 + 1 )n
n
an
nn
(n + 1) (n + 1) n
Note. If the ratio or root test is inconclusive because the calculation gives
L = 1, the other will not help you. Dont waste time trying both.
2. limn an = 0.
Then
P P n1 a converges.
n=1 bn = n=1 (1) n
Example.
The alternating harmonic series satisfies the hypothesis for the alternat-
1
ing series test, since n+1 n1 and limn n1 = 0. Thus,
X (1)n1
converges.
n
n=1
27
Example.
n1 n2 converges. Let a = n2
P
n=1 (1) n3 +1 n n3 +1
. One can see that a2 a1 .
x2 0 (x) = x(x 3 2)
3
Also, if f (x) = 3
x +1
, then f 3
(x +1) 2 0 when x 2 so that
1
an+1 an for n 2. Finally, 0 an n , so the squeeze theorem tells us
that limn an = 0. We can now apply the alternating series test.
Definition. If
P P P
n=1 an but n=1 |an | diverges, then n=1 an is said to
be conditionally convergent.
Conditionally convergent series always have the bizarre aforementioned
property. One
P could view this as an interesting but bad property. If we de-
mand that n=1 |an | converges, then the order of summation never matters.
Definition. If
P P
n=1 |an | converges, then n=1 an is said to be absolutely
convergent.
Absolutely convergent series are convergent.
Example.
Consider the series cos n
P
n=1 n2 . We stare Pat this and think that, other
than the cos n, it looks a lot like our friend 1
n=1 n2 which converges since
,
2 > 1. cos n varies between positive and negative, but it is always between
1 and 1. By letting
1 | cos n|
BIGn = and SMALLn =
n2 n2
P
we have BIGn SMALLn 0, and
P n=1 BIGn converges. The direct
comparison test tells us that n=1 SMALLn converges,
X | cos n|
i.e. converges.
n2
n=1
P cos n
This tells us n=1 n2 is absolutely convergent, and thus convergent.
28
4.9 The hierachy of tests
I dont generally have an order of tests which I apply to figure out whether
a series converges or diverges. With enough practice and experience, you
almost always knowP straightaway. But. . . if I had to give a suggested order
of things to try on n=1 an , itd be this. . . This is a fail-safe method; not
the most efficient.
2. Look out for friends. If its a p-series or a geometric series you have
been handed a gift. You understand the convergence and divergence
of p-series, and, in the case of geometric series, when they converge,
you even know what they converge to.
Also, look out for things which are just sums
P or 2 multiples of such
100
things, and apply series rules. For example, n=1 [ n2 + n3 ] converges,
6n 2n
and
P
n=0 [ 3n ] diverges.
What if you can kind of see a friend? Hands rub together. . . Compar-
ison tests might apply!!
3. If you have negative terms, first, check to see if the alternating series
test applies. If it doesnt, apply absolute values to every term in the
hopePthat the series is absolutely convergent; then do subsequent tests
on n=1 |an |.
4. Try the direct comparison test. If you get really good at these then this
will pretty much always work, since the proofs of the ratio test, the root
test, and limit comparison test, all depend on the direct comparison
test. In spirit, this is more like number one on my non-existent list.
5. Perhaps, while trying to use the comparison test, you see that it is one
of those ones where the limit comparison test is a little easier. If so,
use the limit comparison test instead. Beware of the L = 0, cases, if
you use them. I avoided them in lecture since I did not want to cause
additional confusion. If youre confident, then feel free to use them.
29
6. If there are factorials or powers of n in sight, the ratio test might work.
If there are n-th powers, the root test might work. These could also
work in other situations too. Sometimes a factorial might make you
head to the ratio test straightaway, one way in which you can see this
list is a little too inflexible.
8. Sometimes you might have to rewrite the series a little before you can
apply a test to it. For instance, how would you prove
1 1 1 1 1 1 1 1 1 1 1
1+ + + + + + + + ...
3 2 5 7 4 9 11 6 13 15 8
converges? Use parentheses cleverly!
Sometimes using more than one test one after another can work too.
4.10 Examples
P n1 3n2 en3
1. n=1 (1)
3
This converges for pretty much every reason! en decays really quickly
so. . . However, some tests are easier than others. Ill make a couple of
observations before beginning.
3 3
Let f (x) = 3x2 ex . Then f 0 (x) = 3xex (2 3x3 ) and so f 0 (x) 0
when x 1. This means f (x) is decreasing when x 1. Also,
3x2 6x 2
lim f (x) = lim 3 = lim 3 = lim = 0,
x x e x 2
x 3x e x x xex3
30
(b) We dont see any friends.
(c) By using the calculations above, we have
P an+1 n1
and an for all
P n. The alternating series test tells us that
n1 3n2 en3 converges.
n=1 (1) an = n=1 (1)
This required some calculus and LHopitals rule: a bit annoying.
We will show
P P n1 a is ab-
n=1 an converges so that n=1 (1) n
solutely convergent.
(d) The direct comparison test is applicable. We can let SMALLn =
an and LARGEn = 3en . Proving SMALLn LARGEn is a bit
annoying.
(e) The limit comparison is possible, though my choice of bn = en
results in limn abnn = 0 and calculating this requires LHopital:
a bit annoying.
(f)
2 (n+1)3
an+1
lim = lim 3(n + 1) e 3
n an n 3n2 en
n+1 2
2
= lim lim e3n 3n1 = 1 0 = 0 < 1
n n n
In summary, the ratio test was the best; in fact, I would have gone for
it straight away since powers play well with the ratio test.
31
n
P e 1000 +n
2. n=1 n2 +2
.
n
Since e 1000 grows faster than everything in sight I would expect this
to diverge. It also diverges for pretty much every reason, but some are
easier to check than others.
n
e 1000 +n
Let an = n2 +2
.
I think the direct comparison was easiest. The n-th term test wasnt
so bad either.
32
5 Direct Comparison for Integrals
Suppose that BIG(x) and SMALL(x) are continuous on a < x < b and that
for these values we have
BIG(x) SMALL(x) 0.
Rb Rb
1. If a BIG(x) dx converges, then
SMALL(x) dx converges.
a
Rb Rb
2. If a SMALL(x) dx diverges, then a BIG(x) dx diverges.
(The first equivalence uses the fact that et is an increasing function of t; the
last equivalence makes use of the fact that x > 0.)
Do we give up? Never! We just avoid our problems. Notice that
Z Z 1 Z
x2 x2 2
e dx = e dx + ex dx.
0 0 1
Were left with the second integral, and the argument that we just tried to
make does work now. We have BIG(x) SMALL(x) 0 when x 1, and
Z Z R
BIG(x) dx = lim ex dx = lim (e1 eR )
1 R 1 R
33
R
converges to 1e . So the comparison theorem says that 1 SMALL(x) dx =
R x2
1 e dx converges. We have now completed the proof.
An alternative proof, avoiding dividing the integral up, would have been
1
to take BIG(x) = e 4 x :
1 2 1
e 4 x ex x x2
4
1 2
2 1
x x + 0 x 0.
4 2
R
Example 2: 2 ln1x dx diverges.
1 1
Let
R BIG(x) = ln x and SMALL(x) = x .
2 SMALL(x) dx diverges by the p-test. To apply the comparison test,
we just need to show that BIG(x) SMALL(x) 0 when x > 2.
If x 1, then
d 1 d
(x) = 1 = (ln x).
dx x dx
Since 1 ln(1), this shows that for x > 2, x ln x, and so
1 1
when x > 2.
ln x x
(By using the prime number theorem (a very difficult theorem) this example
gives a very roundabout proof that there are infinitely many prime numbers.)
34
6 More 8.7 help.
Here are the type of inequalities I use for solving questions 61-75 in section
8.7.
1. (a) If f (x) > 0, g(x) 0, and a 0, then
1 1
a
.
(f (x) + g(x)) (f (x))a
(b) If f (x) > 0, g(x) k > 0, then
1 1
a
.
(f (x)g(x)) (kf (x))a
2. (a) If f (x) > g(x) 0, and a 0, then
1 1
a
.
(f (x) g(x)) (f (x))a
(b) If nf (x) g(x) > 0 and a 0, then
1 1
a
.
(f (x) + g(x)) ((n + 1)f (x))a
(c) If f (x) > 0, k g(x) > 0, and a 0, then
1 1
.
(f (x)g(x))a (kf (x))a
This is used in 75 with f (x) = x, g(x) = ex + x, k = e + 1, a = 1,
on the interval 0 < x 1.
3. (a) ex 1 for x 0.
(b) If f (x) 0, g(x) 0, then
e(f (x)+g(x)) ef (x) .
35