MATH 255: Lecture 13

Sequences of Functions: Uniform Convergence and Differentiation

If fn (x) = xn /n, the sequence (fn ) converges uniformly to the function f = 0 on [0, 1]. However,
fn0 (x)= xn−1 so that the sequence of derivatives (fn0 ) converges pointwise to the function g, where
g(x) = 0 if x 6= 1 and g(1) = 1. Since f 0 6= g, this shows that one cannot, in general, interchange limits
and derivatives. Nevertheless, one has the following result:
Theorem. Let (fn ) be a sequence of differentiable functions on a bounded interval I such that (fn (x0 ))
converges for some point x0 ∈ I. If the sequence (fn0 ) converges uniformly on I to a function g, then
(fn ) converges uniformly on I to a function f which is differentiable on I and f 0 = g.
Proof. Let a < b be the endpoints of I and let ² > 0 be given. Choose N1 so that, for m, n ≥ N1
0 ² ² ²
|fm (x) − fn0 (x)| < min( , ) and |fn (x0 ) − fm (x0 )| < .
3 3(b − a) 3

For any x ∈ I, apply the Mean Value Theorem for Derivatives to the interval with endpoints x0 , x to
get
0
fm (x) − fn (x) = fm (x0 ) − fn (x0 ) + (x − x0 )(fm (y) − fn0 (y))
for some y between x0 and x. Hence

0 2²
|fm (x) − fn (x)| ≤ |fm (x0 ) − fn (x0 )| + (b − a))|fm (y) − fn0 (y)| < < ².
3
This shows that (fn ) is uniformly convergent on I. If f = lim fn , the f is continuous on I since each
fn is continuous on I.
To show that f is differentiable at a point c ∈ I, we apply the Mean Value Theorem for Derivatives
to fm − fn on the interval with endpoints x, c ∈ I with x 6= c. We get
0
(fm (x) − fn (x)) − (fm (c) − fn (c)) = (x − c)(fm (z) − fn0 (z))

with z between x and c. Dividing by x − c and taking absolute values, we get

¯ ¯
¯ fm (x) − fm (c) fn (x) − fn (c) ¯ ²
¯ − ¯ ≤ |fm
0
(z) − fn0 (z)| < .
¯ x−c x−c ¯ 3

Passing to the limit with respect to m, we get

¯ ¯
¯ f (x) − f (c) fn (x) − fn (c) ¯
¯ − ¯ ≤ ².
¯ x−c x−c ¯ 3

²
Now choose N2 so that |fn0 (c) − g(c)| < for n ≥ N2 . Let N = max(N1 , N2 ) and choose δ > 0 so that
3
¯
¯ fN (x) − fN (c) ²
0 < |x − c| < δ =⇒ ¯¯ 0
− fN (c)| < .
x−c 3

Then, combining these inequalities, we get

¯ ¯
¯ f (x) − f (c) ¯
¯ − g(c)¯ < ².
¯ x−c ¯

Since ² > 0 was arbitrary, this shows that f 0 exists on I and equals g.

1
 P∞
Corollary. If n=1 fn is a series of differentiable functions on a bounded interval I such that
∞
X
(a) fn (x0 ) converges for some point x0 ∈ I,
n=1

∞
X
(b) g = fn0 is uniformly convergent on J,
n=1

∞
X
then f = fn is uniformly convergent and f 0 = g.
n=1

While the uniform convergence of a sequence of continuous functions is sufficient for the limit function
to be continuous, it is not necessary. For example, if


nx if 0 ≤ x ≤ 1/n,
fn (x) = −n(x − 2/n) if 1/n ≤ x ≤ 2/n,


0 if 2/n ≤ x ≤ 2,

the sequence (fn ) converges pointwise to the zero function on [0, 2] but the convergence is not uniform
as fn (1/n) = 1. However, for monotone sequences of continuous functions on a closed interval, uniform
convergence is necessary if the pointwise limit exists and is continuous.

Theorem. (Dini) Let (fn ) be a monotone sequence of continuous functions on a closed interval [a, b].
If (fn ) converges pointwise to a continuous function f , the the convergence is uniform.
Proof. Possibly replacing (fn ) by (−fn ), we can assume that fn+1 (x) ≤ fn (x). Also, after replacing
fn by fn − f , we can assume f = 0. Let ² > 0 be given. For each t ∈ [a, b], there is an N (t) > 0 such
that fN (t) (t) < ². By continuity, there is a δ(t) > 0 such that fN (t) (x) < ² if |x − t| < δ(t). Now let
(P, t) be a δ-fine tagged partition for the gauge δ constructed above and let N = min(N (t1 ), . . . , tn ),
where t = (t1 , . . . , tn ). Then, for any x ∈ [a, b], we have fN (x) < ² since |x − tk | < δ(tk ) for some k.
Hence fn (x) < ² for n ≥ N and all x ∈ [a, b]. QED
Note that the the example fn (x) = xn on (0, 1) shows that the result is false on an interval which
is not closed.

(Last updated 2:00 pm April 14, 2003)