Uniform Continuity: Ryan Acosta Babb

Uniform Continuity
Ryan Acosta Babb

r.acosta-babb@warwick.ac.uk
MA260 Support Class – Week 3
Summary
A brief motivation and review of uniform continuity in R. We discuss
the importance of quantifier order and variable dependence through an
example that helps motivate the need for uniform continuity. This review
is intended as a “warm up” before tackling compactness.
1
Let f : (0, 1) → R be a continuous function and 0 < an < n for all n ∈ N.
Question Does (f (an ))n∈N converge?

A rough attempt follows:
“Proof 1”. Note that an → 0 and f is continuous, so f (an ) → f (0).
This clearly will not do, as f is not necessarily defined at x = 0! Indeed,
Example 1. Take f (x) := x1 . Then f (an ) > n for all n ∈ N, so it clearly does
not converge.
What if we strengthen our assumptions to prevent the blow-up at the end-
points?
Question 2 Suppose f : (0, 1) → R is continuous and bounded, with an as

above. Does (f (an ))n∈N converge?
Our first “proof” failed because we didn’t know the behaviour of f at the
“limit point” f (0). If we want to show that f (an ) converges without knowing a
priori what the limit might be, we can check whether it is a Cauchy sequence!
Proof 2. Let ε > 0. Then,
(i) By continuity of f , there exists a δ > 0 such that
|x − y| < δ =⇒ |f (x) − f (y)| < ε.
(ii) By convergence of (an )n∈N , it is a Cauchy sequence, so there is an N ∈ N

such that, for all n, m > N ,
|an − am | < δ.
1
(iii) Hence, by (i) and (ii),
|f (an ) − f (am )| < ε for all n, m > N.
This proves that (f (an ))n∈N is Cauchy in R and, therefore, convergent.

Does it? Well, no. Our proof only used the assumption that f is continuous
(we never invoked boundedness!), and we saw in Example 1 that (f (an ))n∈N
does not converge in general under this assumption. So let us take a closer look
at our “proof” to see what is going on.
In (i) we applied continuity to get a δ > 0, which we then fed into the
definition of Cauchy sequence to produce the index N ∈ N. But we run into
trouble when we try to apply the continuity of f to the condition |an − am | < δ,
since this is expected to hold for all an , am with n, m > N . However, the δ only
tells us about points y within δ of a specific x!
Let us run through the argument again, noting carefully the dependence of
the different constants:
(i) For any x ∈ (0, 1) there is a δx > 0 depending on x (and ε) such that
|x − y| < δx =⇒ |f (x) − f (y)| < ε.
(ii) For each δx there is an Nδx ≡ Nx ∈ N depending on x (actually δx , which

depends on x and ε, but we care about the x dependence) such that
|an − am | < δx for all n, m > Nx .
We therefore have possibly uncountably many Nx to choose from! How are we

to pick one N to rule them all?
The issue is that regular continuity is a local property: it tells us how f
changes near a single point, say x. So while we may know how to bound f (y)
to within ε of f (x) for y close to x (say within δ of x), if we now move to a
different point x0 we have to start all over again, since getting δ-close to x0 may
not be enough to bound |f (x0 ) − f (y)| < ε (same ε and δ as before).
A better behaved function might not care about the specific choice of x, but
only the closeness δ.
Example 2. Let f (x) := |x|. Then, by the reverse triangle inequality, we have
|f (x) − f (y)| 6 |x − y| for all x, y ∈ R.
Thus, choosing δ = ε for given ε > 0, we have
|f (x) − f (y)| < ε (1)
at any two points with |x − y| < ε. It doesn’t matter whether we pick x = 0

and y = ε/2 or if we take x = −400 + ε/67 and y = −400: as long as they span
an interval no greater than ε, the bound (1) holds. (See Figure 1.)
2
5
f (x) = |x|
x
−4 −2 2 4
Figure 1: The red marks identify intervals of width ε. No matter where we take
them on the x-axis, the range of the function f stays within a vertical interval
of height ε.
Example 3. Consider now g(x) := x1 defined on (0, 1). Take ε = 1/2. Then,
for any δ ∈ (0, 1) we can choose x = δ/4 and y = 3δ/4 so clearly

1 1 8
|x − y| < δ and = =
>ε
x y 3δ
(as 0 < δ < 1). Provided we pick x and y close to 0, where the function g blows
up, we can make their difference as small as we like, and still |g(x) − g(y)| > ε.
2
If, however, we fix x ∈ (0, 1), for any ε > 0 we can take δ = x2 ε (depending
on x!). This “pinning of x in place” gives us more control over g(y) for y close
to x. Compare this to the case above, were we needed to squish both x and y
close to 0 to get the blow up. (See Figure 2.)
This you should recognise as uniform continuity from Analysis III. (If you
don’t, go back and revise your notes!) Compare:
(i) A function f : (a, b) → R is continuous if
∀ε > 0∀x ∈ (a, b)∃δx,ε > 0∀y ∈ (a, b) (|x − y| < δx,ε =⇒ |f (x) − f (y)| < ε)
(ii) A function f : (a, b) → R is uniformly continuous if
∀ε > 0∃δε > 0∀x ∈ (a, b)∀y ∈ (a, b) (|x − y| < δε =⇒ |f (x) − f (y)| < ε)
Note the quantifier flip! Thus, the δ no longer depends on x.

We can now state and answer our question correctly and affirmatively:
3
1
g(x) = x
20
15
10
x
0.1 0.2 0.8 0.9 1
Figure 2: The interval of width 0.1 from 0.1 to 0.2 yields a vertical jump of
5 units, while an interval of the same length measured at 0.8 yields a jump of
about 0.14.
Proposition 4. Suppose f : (0, 1) → R is uniformly continuous, with 0 < an <

1
n . Then (f (an ))n∈N converges in R.
Proof. We use the uniform continuity of f to free our δ from the restriction of
x.
(i) Let ε > 0. By uniform continuity of f , find δ > 0 such that
|x − y| < δ =⇒ |f (x) − f (y)| < ε for all x, y ∈ (a, b).
(ii) With this single δ > 0, apply the Cauchy criterion to (an )n∈N to get an
N ∈ N (depending on δ, but that’s fine!) such that
|an − am | < δ for all n, m > N.
(iii) Taking x = an and y = am in (i), (ii) implies that

|f (an ) − f (am )| < ε for all n, m > N.
This proves that (f (an ))n∈N is Cauchy in R, and therefore convergent.

The key step in the proof is finding a δ that works uniformly on the an , am ,
without being tied down to a particular choice of x, since the an are changing
(what if we pick n = N + 1 or n = N + 2 or n = N + 3 or . . . ?) and we
need to control all of them in one go. This sort of idea is of crucial importance
in analysis, as you will have ample opportunity to discover in later modules.
For now, keep it in mind, as it will be immensely helpful in understanding the
notion of compactness.

Uniform Continuity: Ryan Acosta Babb

Uploaded by

Copyright:

Available Formats

Uniform Continuity: Ryan Acosta Babb

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Uniform Continuity: Ryan Acosta Babb

Uploaded by

Copyright:

Available Formats

Uniform Continuity

Ryan Acosta Babb

Question Does (f (an ))n∈N converge?

Question 2 Suppose f : (0, 1) → R is continuous and bounded, with an as

|x − y| < δ =⇒ |f (x) − f (y)| < ε.

(ii) By convergence of (an )n∈N , it is a Cauchy sequence, so there is an N ∈ N

|f (an ) − f (am )| < ε for all n, m > N.

This proves that (f (an ))n∈N is Cauchy in R and, therefore, convergent.

|x − y| < δx =⇒ |f (x) − f (y)| < ε.

(ii) For each δx there is an Nδx ≡ Nx ∈ N depending on x (actually δx , which

|an − am | < δx for all n, m > Nx .

We therefore have possibly uncountably many Nx to choose from! How are we

|f (x) − f (y)| 6 |x − y| for all x, y ∈ R.

Thus, choosing δ = ε for given ε > 0, we have

|f (x) − f (y)| < ε (1)

at any two points with |x − y| < ε. It doesn’t matter whether we pick x = 0

(i) A function f : (a, b) → R is continuous if

(ii) A function f : (a, b) → R is uniformly continuous if

Note the quantifier flip! Thus, the δ no longer depends on x.

Proposition 4. Suppose f : (0, 1) → R is uniformly continuous, with 0 < an <

(iii) Taking x = an and y = am in (i), (ii) implies that

This proves that (f (an ))n∈N is Cauchy in R, and therefore convergent.

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.