ANALYSIS I

9 The Cauchy Criterion

9.1 Cauchy’s insight
Our difficulty in proving “an → `” is this: What is `? Cauchy saw that it was enough to
show that if the terms of the sequence got sufficiently close to each other. then completeness
will guarantee convergence.

Remark. In fact Cauchy’s insight would let us construct R out of Q if we had time.

9.2 Definition
Let (an ) be a sequence [R or C]. We say that (an ) is a Cauchy sequence if, for all ε > 0
there exists N ∈ N such that

m, n > N =⇒ |am − an | < ε.

[Is that all? Yes, it is!]

9.3 Cauchy =⇒ Bounded

Theorem. Every Cauchy sequence is bounded [R or C].

Proof. 1 > 0 so there exists N such that m, n > N =⇒ |am − an | < 1. So for m > N ,
|am | 6 1 + |aN | by the ∆ law. So for all m

|am | 6 1 + |a1 | + |a2 | + · · · + |aN |.

9.4 Convergent =⇒ Cauchy [R or C]

Theorem. Every convergent sequence is Cauchy.

Proof. Let an → l and let ε > 0. Then there exists N such that

k > N =⇒ |ak − l| < ε/2

For m, n > N we have

|am − l| < ε/2

|an − l| < ε/2

So
|am − an | 6 |am − l| + |an − l| by the ∆ law
< ε/2 + ε/2 =ε

1
9.5 Cauchy =⇒ Convergent [R]
Theorem. Every real Cauchy sequence is convergent.

Proof. Let the sequence be (an ). By the above, (an ) is bounded. By Bolzano-Weierstrass
(an ) has a convergent subsequence (ank ) → l, say. So let ε > 0. Then

∃N1 such that r > N1 =⇒ |anr − l| < ε/2

∃N2 such that m, n > N2 =⇒ |am − an | < ε/2

Put s := min{r|nr > N2 } and put N = ns . Then

m, n > N =⇒ |am − an |
6 |am − ans | + |ans − l|
< ε/2 + ε/2 =ε

9.6 Cauchy =⇒ Convergent [C]

Theorem. Every complex Cauchy sequence is convergent.

Proof. Put zn = x + iy. Then xn is Cauchy: |xx − xm | 6 |zn − zm | (as |<w| 6 |w|). So
xn → x, yn → y and so zn → x + iy.

9.7 Example
Let
1 1 (−1)n+1
an = 1 −+ + ··· +
2 3 n
Then with m > n, and m − n odd we have
z }| {z }| { z }| {

 1 1 1 1 1 1
|am − an | =  − + − +...+ − 
n + 1 n + 2 n + 3 n + 4 m − 1 m
1 1 1 1 1 1
= − + −...− + −
n + 1
| {z } | n + 2{z n + 3
} | m − 2{z m − 1 m
} |{z}
1 1
6 n+1 6 n

If m − n is even, we write
1 1 2
|am − an | = |am − am−1 + am | 6 |am − am−1 | + |am | 6 + 6
N +1 m n
1
Let ε > 0. Chose N > 2ε and convergence follows.
Question: What is limn→∞ an ? Later we’ll see it is log 2.

2
10 Series
10.1 Definition
Note. So far we have always used sequences defined by functions a : N → R. It is convenient
from now on to start off at a0 , that is to work with functions a : N ∪ {0} → R.
Pm
So let (an )∞ n=0 be a sequence of real (complex) numbers. Define sm := n=0 an , a well
defined real (complex) number.
Consider the P sequence (sn )∞ n=0 . We call this the series defined by the sequence (an ), and we
∞
denote it by n=0 an .

Note. This is a very odd name indeed! Don’t let it mislead you: ∞
P
n=0 an is just the sequence
a0 , a0 + a1 , a0 + a1 + a2 , . . . .

10.2 Examples

(i) The Geometric Series Let an := xn Then xn is

P

(1, 1 + x, 1 + x + x2 , . . . , 1 + x + x2 + · · · + xn , . . . )

1 P 1
(ii) The Harmonic Series Let an := n+1 . Then n+1 is
 
1 1 1
1, 1 + , 1 + + , . . .
2 2 3

(iii) The Exponential Series Let an := xn /n!. Then

P n
x /n! is

x2
 
1, 1 + x, 1 + x, 1 + x + , . . .
2!

(iv) The Cosine Series Let

x2m
 m
an = 2m! (−1) if n = 2m
0 otherwise
P
Then an is
x2 x2 x2 x4
 
1, 1, 1 − , 1− , 1− + , ...
2! 2! 2! 4!

10.3 Convergence and Divergence

P∞
With the setup above if (sn ) converges, then we say “the series n=0 is convergent”.
Otherwise we say that the series is divergent.

3
10.4 More examples

(i) Let an = xn .

xn is divergent.
P
(a) If x = 1 then

Proof. s0 = 1, s1 = 2, . . . , sn = n + 1 by induction.
1−xn+1
(b) If x 6= 1, then sn = 1−x .

Proof. Pure algebra.

P n
(c) If |x| < 1 then x is convergent.
P n
(d) If |x| > 1 then x is divergent.

Proof. If |x| < 0 then xn → 0 as n → ∞. So by AOL sn → 1/(1 − x). If sn → s

then xn+1 → 11 − (1 − x)s by AOL. But if |x| > 1, xn is not convergent.

(ii) Let an = n1 . Then

P1
n is divergent.

1 1
Proof. s2n −1 = 1 + 2 + ··· + 2n =
1 1 1 1 1 1 1




= 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + ...
1 2 4 2n−1
> 1 + 2 + 4 + 8 + ... + 2n
n
> 2

so (sn ) has a subsequence which is not convergent.

1 P 1
(iii) Let an = (n+1)2
. Then n2
is convergent.

Proof.
1 1 1 1
sn = 1 + 4 + 9 + ... + n2
+ (n+1)2
1 1 1
6 1 + 1·2 + 2·3 + ... +
 n(n+1) 
1 1 1 1 1 1



= 1 + 1 − 2 + 2 − 3 + ... + n − n+1
1
= 2− n+1
6 2
1
So (sn ) is monotone increasing [sn+1 = sn + (n+1)2
> sn ] and bounded by 2. So (sn ) is
convergent by Bolzano-Weierstrass (??).

4
10.5 Notation and its abuse
P∞ P∞
More notation: if the series n=0 an is convergent then we often denote the limit by n=0 an ,
and call it the sum.

Note. We must take great care, but Pthis double use is traditional.
P I will try to distinguish
the two uses, and say “The series an ” and “The sum an ”. I suggest that you do the
same for a bit.

Note. Just attaching the label “sum” to the symbol ∞

P
n=0 an does not turn it into a proper
mathematical sum. Look at our axioms for R. They only speak of adding pairs of real
numbers, which we can extend using axiom A2 to finite sets of real numbers. But given an
infinite set of real numbers we can’t simply “add” them and get a “sum”. Instead we have to
talk about sequences . . .

10.6 Tails
Let (an ) be a sequence and (sn ) be the corresponding series.
Sometimes we want to look at (ak , ak+1 , ak+2 , . . . ). We write this series ∞
P
n=k an . We put
Sn = ak + ak+1 + · · · + an and note that what we said about Tails of sequences, and adding
constants, ensures that (sn ) is convergent if and only if Sn is convergent.

10.7 Cauchy’s criterion

We rewrite Cauchy Criterion for series.

Theorem. The series ∞

P
n=0 an is convergent if and only if for all ε > 0 there exists N ∈ N
such that
 
X l 
l > k > N =⇒ an  <ε
 

 
n=k
| {z }
A genuine
sum
Note. Clearly in practice when we estimate the sum we’ll use the ∆ law when we can.

10.8 Absolute Convergence

P P
Let an be a sequence. Then we say that an is absolutely convergent if the series |an |
is convergent.

10.9 Absolute Convergence =⇒ Convergence

P
Theorem. If an is absolutely convergent then it is convergent.

Proof. Let ε > 0 Then there exists N ∈ N such that

 l 
X 
l > k > N =⇒  |an | < ε
 
 
k

5
So  l   l 
X  X l X 
l > k > N =⇒  an  6 |an | =  |an | < ε
   
   
| k {z k } k
By the ∆ law

Note. Why is absolute convergence a good thing to have? Because it makes use of Cauchy
criterion easy!

10.10 Examples

xn absolutely convergent if |x| < 1

P
(i)
P (−1)n
(ii) n2
is absolutely convergent
P sin n
(iii) n3
is absolutely convergent
P (−1)n
(iv) n+1 is convergent, but not absolutely convergent.

10.11 Re-arrangements
P
Let p : N −→ N one-to-one and onto. WePcan then put bn = ap(n) and consider bn , which
we call a rearrangement of the series an .
Funny this can happen! Later on we will be able to prove that

Example.
1 1 1
1− 2 + 3 − 4 + ... → log 2
1 1 1
1+ 3 + 5 − 2 + 17 + 1
9 + 1
11 − 1
4 + ... → log(2/3)
P
Theorem. If (an ) is absolutely convergent and (bn ) is a rearrangement of (an ) then bn is
absolutely convergent too.

10.12 Multiplication of series

P P
Theorem. Suppose an , bn are absolutely convergent. Suppose that
n
X
cn := ar bn−r
r=0

Then
P
(i) cn is absolutely convergent
P P P
(ii) cn = an bn

Proof.

6
 (i) By pure algebra
N
X N
X 2
X 2N
X 2N
X
|an | |bn | 6 N |cr | 6 |an | |bn |
0 0 0 0 0
P P P
So by Sandwich Rule, |cn | is convergent to |an | |bn |.
P
(ii) Hence cn is convergent.

(iii) Given ε > 0 there exist N such that

l
X
l > k > N =⇒ |cn | < ε
k

Then for k > N

 2k k k
   2k
X X X   X  X X
cn − an bn  =  ar bs  6 |ar bs | 6 |cn | < ε
   

   r+s62k, r>k or 
n=0 n=0 n=0 s>k ditto k

Note. Draw a diagram in the (r, s) plane marking out which ar bs are included in the various
sums.

10.13 Two Applications

(i)
X 1
(n + 1)xn =
(1 − x)2

Proof. an := xn , bn := xn , therefore
X
cn = xr xs = (n + 1)xn
r+s=n

(ii)
X xn X y n X (x + y)n
=
n! n! n!
xn yn
Proof. an := n! , bn := n! therefore
X xr y s (x + y)n
cn = =
r+s=n
r! s! n!