You are on page 1of 3

Notes on Continuous Functions

Alex Nelson
Email: pqnelson@gmail.com
November 5, 2011

Abstract
Starting from Cauchys definition x x0 implies f (x) f (x0 ), we make these
notions rigorous using sequences. Then we get rid of the sequences by generalizing to
the familiar notion of continuity in real analysis.

Contents
1 Why Bother With Continuity? 1
1.1 Enter Cauchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1 Why Bother With Continuity?


We learn calculus by symbolic rules for differentiation, e.g.
d n
x = nxn1
dx
d ix
e = ieix
dx
and so on. But is this always defined? Consider some pathological functions:

X
f (x) = an cos(bn x), (1.1)
n=0

where 0 < a < 1, b N is odd, and


3
ab > 1 + . (1.2)
2
What is its derivative?
For some pictures of the type of functions we would prefer to avoid, consider the
following:

x0 sin(1/x)

x0
This
is a page from http://code.google.com/p/notebk/
Compiled: November 5, 2011 at 4:38pm (PST)

1
1 Why Bother With Continuity? 2
These cause problems at the points x0 . If you dont believe me, try taking the derivative
of the function, then take the limit x x0 . Wouldnt it be nice if we could just weed out
these pathological types beforehand? Thats one of the roles continuity plays.
But we cannot simply say Ignore the pathological cases, whatever that means. We
need to give some criteria to see if a function is pathological or not. Let us kill each case.

1.1 Enter Cauchy


Cauchy defined continuity. If we say x is Infinitesimally Close to x0 , then we
denote it by
x x0 . (1.3)
What does it mean? Well, we dont know (this was the flaw in Cauchys approach).
We will say that f is Continuous at x0 if and only if

x x0 = f (x) f (x0 ). (1.4)

Is this a good definition?


Well, yes and no. Yes, because no matter how one defines infinitesimally close, it
avoids jump discontinuities:

x0

But this also has the problem that we dont have any rigorous notion of infinitesimals.
  If we insisted that we may use hyperreal numbers and define things this way, we still have
problems. Consider f (x) = 1 Q (x) which vanishes if x Q and is 1 otherwise. Then
using hyperreal numbers to define infinitesimally close and make Cauchys notion of continuity
rigorous, then f is continuous on R except at countably many points (namely, when x Q). This
example should tell us that a literal translation of Cauchys definition is not good, since the only
discontinuity is a jump discontinuity.

1.2 Sequences
We have a sequence {xn }nN converge to x0 if and only if for each > 0 there exists
an N N such that |xm xN | < for all m > N .
Intuitively, if xn is an arbitrary sequence which converges to x0 , then eventually xn gets
infinitesimally close to x0 . This is a better way to rigorously describe getting infinitesimally
close, because we avoid all discontinuities (not just jumps). We thus translate Cauchys
intuitive notion of continuity thus:
Definition 1.1. Let f be a function, x0 dom(f ). We say f is Continuous at x0 if for
any sequence {xk }kN which converges to x0 we have

f (xk ) f (x0 ). (1.5)

Remark 1.2. This is a better definition, because we can generalize this notion to more liberal
settings (e.g., the space of square matrices). Also, we avoid using infinitesimals (which only
cause messes).
What have we done? We basically switched the condition of being infinitesimally close
to convergent sequences. This is quite clever, it preserves the intuition. Can we define a
notion of continuity that (1) doesnt use sequences, and (2) is logically equivalent to the
definition we have produced? Consider first the Cauchy condition for convergence:
References 3
Theorem 1.3 (Cauchy). A sequence of real numbers {xn }nN is convergent if and only if
they satisfy the Cauchy condition
for each > 0 there is a corresponding N N such that |xm xn | < for
m, n > N .
We wont produce a proof (its not hard, hint: think). But lets consider this: discon-
tinuities occur because not all sequences f (xn ) can converge to f (x0 ). So convergence of
f (xn ) is critical. We reserve > 0 to measure how close f (xn ) is to f (x).
But we want to consider getting infinitesimally close to x0 in the domain. This
amounts to considering the identity function

(x) = x (1.6)

and its continuity at x0 . We reserved , so lets use another variable, say, > 0. For each
> 0 there is a corresponding N N such that

|(xn ) (x0 )| < (1.7)

for all n > N . Observe we plug in the definition of to find this implies

|xn x0 | < (1.8)

which intuitively corresponds to xn x0 . Cauchys definition said this should imply

|f (xn ) f (x0 )| < . (1.9)

This seems like a good way to define continuity.


Question. Are the variables and independent of each other?
Should they be? Intuitively, no they should depend on each other. But again, we are
concerned about the convergence of the sequence f (xn ) f (x0 ). So > 0 can be arbitrary,
but to each one there is a corresponding (but not necessarily unique) > 0 such that

|x x0 | < = |f (x) f (x0 )| < . (1.10)

This includes all sequences contained within this scope. So is this a good definition?
Amazingly, yes! We could alternatively formulate it as for each interval centered at
f (x0 ) there is a corresponding interval centered at x0 such that

|x x0 | < = |f (x) f (x0 )| < (1.11)

What does it tell us? Well, continuous functions preserve open intervals (duh). But there is
more, if we extend it to a closed interval [x0 , x0 + ], then the image is necessarily a closed
interval. This is unexciting to the neophyte, but to the topologist it means that compactness
is preserved under continuous mappings. For more, see my notes on topology [1].

References
[1] Alex Nelson,
A Rapid Introduction to Topology.
Eprint: http://notebk.googlecode.com/files/rapidIntroTopology.pdf
Posted: October 25, 2011. 10 pages.

You might also like