CONVERGENCE TESTS FOR SERIES: COMMENTS AND PROOFS

PART I: THE COMPARISON AND INTEGRAL TESTS

Math 112

The convergence tests for series have nice intuitive reasons why they work, and these
are fairly easy to turn into rigorous proofs. In these notes we will prove the standard
convergence tests and give two tests that aren't in our text.

It's important to remember that the convergence or divergence of a series depends only
on what happens to some tail of the series|the inclusion or omission of a ¯nite number of
terms cannot change a convergent series into a divergent one or vice versa. (If the series
converges, the sum is a®ected, of course.) Because of this, we can let a little convenient
sloppiness into our notation. When it doesn't matter what the starting point is for a series,
P P P1 P1
we can write ak instead of 1 k=1 ak (or k=5 ak or k=k0 ak ). More speci¯cally, the
P
notation a k will always mean an in¯nite sum, and will only be used when the starting
point for the sum is not important.

Slight Generalizations of the Comparison and Integral Tests

The observation above allows the use of the comparison test in cases where the terms
of the two series satisfy the appropriate inequality after some point, but perhaps not for
the ¯rst several terms. Here is the slightly more general comparison test.
P P
Theorem (Comparison Test). Let a k and bk be series. Suppose there is an index
P P
K such that 0 · a k · bk for all k ¸ K. If bk converges, so does ak . Similarly, if
P P
a k diverges, so does bk, and both sums are in¯nite.

The integral test has a similar modi¯cation.

1
 P
Theorem (Integral Test). Let a k be a series. Suppose that K is a positive integer
and that a: [K; 1) ! R is a continuous, decreasing function such that ak = a(k) for each
P R
integer k ¸ K. Then either the series ak and the improper integral K1 a(x) dx both
converge, or they both diverge.

The point here is that the terms of the series and the function don't have to be related for
x and k less than K. In fact the function doesn't have to be continuous, decreasing, or even
R
de¯ned. Note that for K > 1, if the integral K1 a(x) dx converges, we can conclude that
P
the series 1 k=1 ak converges (assuming the index for the series begins with 1). However,
R1
if the series converges, we cannot conclude that the integral 1 a(x) dx converges without
additional information about the function a.

The Comparison Test. The comparison test is pretty intuitive. If ak · bk for all k ¸ K,
P P1 P1 P1
then it should be the case that 1 k=K ak · k=K b k . If k=K b k is ¯nite, so is k=K a k .
P1 P1
Similarly, if k=K a k is in¯nite, so is k=K bk . It's not hard to make this rigorous.
P P
Proof of the Comparison Test. For n ¸ K, we have the inequality nk=K ak · nk=K bk be-
Pn Pn
tween the partial sums. Taking the limit as n ! 1 we have lim k=K ak · lim k=K bk
n!1 n!1
P1 P1
(which is the same as k=K a k · k=K bk ) provided that the limits exist (the limit 1 is
allowed). These limits exist due to the following lemma about increasing sequences.

Lemma 1. Suppose cn , n = 1; 2; 3; : : : , is an increasing sequence. If there is some number

L such that cn · L for all n, then lim cn exists as a ¯nite number. Otherwise lim cn = 1.
n!1 n!1

There are two ways to think about this informally.

(1) If an increasing sequence is bounded above, it has to converge (meaning that the
limit is ¯nite).
(2) An increasing sequence has to have a limit (which may be in¯nite).
P
In the proof of the comparison this lemma is applied to the partial sums An = nk=K ak
P
and Bn = nk=K bk , which are increasing since the terms of these sums are non-negative.
P P1
By the lemma, the limits lim An = 1 k=K a k and lim Bn = k=K bk exist (they may
n!1 n!1

2
be in¯nite), and then the reasoning preceding the lemma may be applied.

The lemma is pretty intuitive also, but its proof uses one of the fundamental properties
of the real numbers, and so it's worth looking at. (You can skip the rest of this section if
you're not interested.)

Proof of the lemma. Let's take care of the easy part ¯rst. Suppose that the sequence is
not bounded above. This means that for every number L, no matter how large, some term
of the sequence is larger than L, that is, there is some index N such that cN > L. Since
the sequence is increasing, we have cn ¸ cN > L for all n ¸ N . Thus, for every number
L, there is an index N such that cn > L for all n ¸ N . This is exactly what lim cn = 1
n!1
means.

Now suppose that cn · L for all n, that is L is an upper bound for the sequence. Now
L is not the only upper bound|any number greater than L will also be an upper bound,
and some numbers less than L may also be upper bounds. Of these upper bounds, one
of them will be the smallest. Let ` be the least upper bound. I claim that ` is the limit
of the sequence. To see this, suppose that a < `. Since ` is the least upper bound of the
sequence, then a is not an upper bound. This means that some term of the sequence is
bigger than a, that is cN > a for some index N . Since the sequence is increasing we have
that a < cN · cn · ` for every n ¸ N. Thus for any a < ` there is an integer N such that
a < cn · ` for all n ¸ N. This is (essentially) what lim cn = ` means.
n!1

In general, lim cn = ` means that for any open interval containing `, the sequence
n!1

eventually gets inside the interval, and stays inside. More precisely, if a and b are any
numbers such that a < ` < b, then there is an index N such that a < cn < b for all
integers n ¸ N. (You may have learned this as \for any ² > 0 there is an index N such
that ` ¡ ² < cn < ` + ², or jcn ¡ `j < ², for all n ¸ N ." This just makes the interval around
` symmetric, but that's not important.) In the lemma we only need to deal with numbers
less than `, since ` is an upper bound for the sequence.

The fundamental property of the real numbers used in the lemma is the least upper

3
bound property.

Least Upper Bound Property. A non-empty set of real numbers that is bounded above
has a least upper bound.

This property distinguishes the real numbers from the rational numbers. It is crucial to
many theorems in calculus and its generalization, analysis. A study of the least upper
bound property and its sometimes subtle uses are part of the content of Math 33 and 34.

Extra Credit Challenge Problems.

(1) Give an example of a non-empty set of rational numbers that is bounded above
but does not have a rational least upper bound.
(2) Find some calculus theorems that aren't true if you restrict yourself to rational
numbers.

The Integral Test. The integral test is also pretty clear once you see the right pictures
(see page 578 of OZ). These suggest that
X1 Z 1 1
X
ak · a(x) dx · a k:
k=K+1 K k=K
P1 R1 R1
As with the comparison test, if k=K a k is ¯nite, so is K a(x) dx. Similarly, if K a(x) dx
P
is ¯nite, so is 1
k=K a k. A rigorous proof involves partial sums and integrals over ¯nite

intervals.

Proof of the Integral Test. For K · k · x · k + 1 we have ak+1 · a(x) · a k, since both
the function and the sequence are decreasing. Integrating from k to k + 1 yields ak+1 ·
R k+1 R k+1
k a(x) dx · a k . If we let a 0k = ak+1 and bk = k a(x) dx, we have a 0k · bk · a k .
P 0 P
Note that ak and a k are really the same series, their indices are just o®set by one.
P P
It then follows from the regular comparison test that a k and bk either both converge
or diverge.
P R1
Now what is 1 k=K bk ? It should be K a(x) dx. Indeed,
Z 1 Zn X Z k+1
n¡1 n¡1
X 1
X
a(x) dx = lim a(x) dx = lim a(x) dx = lim bk = bk :
K n!1 K n!1 k n!1
k=K k=K k=K

4
 R1
There is one subtlety, however: a(x) dx isn't de¯ned as lim Bn , where B n =
K n!1
Rn RX
K a(x) dx; it's de¯ned as lim B(X ), where B(X ) = K a(x) dx. The di®erence is that
X !1
n is an integer whereas X is a real number. These are slightly di®erent types of limits.
They are obviously related, however, and in this case they are equal. This follows from
two lemmas, the second of which is the analogue for functions of the lemma in the previous
section.

Lemma 2. Suppose that f : [0; 1) ! R and that cn = f (n) for all integers n ¸ 0. If
lim f(x) exists, then so does lim cn , and they are equal (they may both be in¯nite).
x!1 n!1

Lemma 3. Suppose that f : [0; 1) ! R is increasing. If there is some number L such

that f (x) · L for all x, then lim f (x) exists as a ¯nite number. Otherwise lim f (x) = 1.
x!1 n!1

Extra Credit Challenge Problems. In problems 5, 6, and 7 explicit formulas are not neces-
sary. Good descriptions are adequate.
R1 P
(3) The integral test says that K a(x) dx converges if and only if 1k=K ak converges.

As with any \if and only if " statement, there are really two implications. Carefully
explain how the two lemmas are used in each of these implications. In particular,
both lemmas are needed for one direction, but only one is needed for the other.
(4) Give a careful proof of the two lemmas.
(5) Give an example of a sequence ak , k = 0; 1; 2; : : : , and a continuous function
a: [0; 1) ! R with ak = a(k) for each integer k ¸ 0 such that lim a k exists, but
k!1
lim a(x) does not.
x!1
P1
(6) Give an example of a series k=0 ak of non-negative terms and a continuous func-
tion a: [0; 1) ! R with ak = a(k) for each integer k ¸ 0 such that the series
R1
converges but the improper integral 0 a(x) dx does not.
P
(7) Give an example of a series 1k=0 ak of non-negative terms and a continuous func-

tion a: [0; 1) ! R with ak = a(k) for each integer k ¸ 0 such that the improper
R1
integral 0 a(x) dx converges but the series does not.

Robert Foote, October 1999