Math 104: Homework 10

Solutions

The following problems are taken from the textbook, and listed according to the same numbering:

27.1 Prove Theorem 27.5 from Theorem 27.4. Hint: Let φ(x) = (b − a)x + a so that φ maps [0, 1]
onto [a, b]. If f is continuous on [a, b] then f ◦ φ is continuous on [0, 1].
Solution: Let f be a continuous function on [a, b]. For convenience, set d = b − a > 0. As
in the hint, define φ(x) = dx + a (with its domain restricted to [0, 1]). Then as in the hint it
is clear that φ([0, 1]) = [a, b]. This means that f ◦ φ is a function on [0, 1], and since φ is also
clearly continuous we see f ◦ φ is continuous on [0, 1] by Theorem 17.5. So by Theorem 27.4
we see that gn → f ◦ φ uniformly on [0, 1], where gn = Bn [f ◦ φ]. Note that φ has an inverse
function φ−1 : [a, b] → [0, 1] defined by
x−a
φ−1 (x) =
d
which is also linear and hence continuous. Now define pn = gn ◦ φ−1 . Then we indeed have
pn is a function on [a, b]. In fact, we claim pn is a polynomial for each n. To see this, note
that for any x ∈ [a, b] we have

pn (x) = (Bn [f ◦ φ])(φ−1 (x))

n  
X k
k
n−k
= (f ◦ φ) φ−1 (x) 1 − φ−1 (x)
n
k=0
n  
X k
k
n−k
= f d · + a d−1 x − d−1 a 1 − d−1 x + d−1 a .
n
k=0

In the expression on the final line above, it is clear that pn (x) is polynomial, as the factors
(d−1 x − d−1 a) and (1 − d−1 x + d−1 a) are each linear polynomials, so the products in each
summand are polynomial and hence pn itself is as well.
Now, to see that pn → f uniformly on [a, b], let ε > 0. Since we already know gn → f ◦ φ
uniformly on [0, 1] we can take N such that for n > N we have |gn (x) − f ◦ φ(x)| < ε for
n > N and x ∈ [0, 1]. We claim that for this same choice of N we have |pn (x) − f (x)| < ε
for x ∈ [a, b]. To see this, let n > N and x ∈ [a, b]. Then we see that φ−1 (x) ∈ [0, 1], so by
assumption we have
|gn (φ−1 (x)) − (f ◦ φ)(φ−1 (x))| < ε,
but then note that by definition we have gn (φ−1 (x)) = (gn ◦ φ−1 )(x) = pn (x), and also
(f ◦ φ)(φ−1 (x)) = f (φ(φ−1 (x))) = f (x), so indeed

|pn (x) − f (x)| < ε

for x ∈ [a, b] and n > N , as claimed.

27.3 Show there does not exist a sequence of polynomials converging uniformly on R to f if

(a) f (x) = sin x,

Solution: As a lemma, we will show that if p(x) is any polynomial function on R, then
limx→±∞ |p(x)| = ∞. The argument for x → −∞ is almost identical to that for x → ∞,
1
 so we focus on the latter. Let p(x) = a0 + a1 x + . . . + an xn be a polynomial (where of
course an 6= 0). Note that we can write
an−1 an−2 a0
|p(x)| = |x|n an + + 2 + ... + n .
x x x
Now let (xk ) be a sequence with xk → ∞. We will show that |p(xk )| → ∞, which then
means limx→∞ |p(x)| = ∞ by definition. To see this, note that in the above expression
the factor
an−1 an−2 a0
an + + 2 + ... + n − −→|an |,
xk xk xk k→∞
and we have 0 < |an | < ∞ by assumption. On the other hand clearly we have |xk |n → ∞,
so by Theorem 12.1 we see that indeed lim |p(xk )| = ∞, so limx→∞ |p(x)| = ∞.
Now, suppose (pn ) is any sequence of polynomials converging to sin x on R (in fact,
strictly speaking, the assumption of convergence isn’t necessary for our purposes here
but we’ll make it anyway because it feels natural). To see that pn → sin x is not uniform,
consider ε = 1. Then for any n, we have by the preceding argument that there is some
x for which |pn (x)| > 2. But then this means

|pn (x) − sin x| ≥ |pn (x)| − | sin x| > 2 − 1 = ε,

since |pn (x)| > 2 and | sin x| ≤ 1. So, for this ε, no N can exist for which |pn (x)−sin x| <
ε for all n > N and all x ∈ R, showing that indeed pn (x) does not converge uniformly
to sin x.
(b) f (x) = ex .
Solution: Note that for x ≤ 0 we have |f (x)| < 1. So in similar fashion to the previous
part, let (pn ) be any sequence of polynomials converging to ex on R. Then consider ε = 1.
Since for any polynomial p we have by the previous part that limx→−∞ |p(x)| = ∞, for
any n we can find x < 0 such that |pn (x)| > 2. But then by an argument parallel to
that in the previous part, we see that |pn (x) − f (x)| > ε for this choice of x, showing
that the convergence cannot be uniform.
27.6 The Bernstein polynomials were defined for any function f on [0, 1]. Show that if Bn f → f
uniformly on [0, 1], then f is continuous on [0, 1].
Solution: Note that since the Bernstein polynomials are polynomials, they are automatically
continuous on [0, 1] (if you like, this follows inductively by finitely many applications of the
sum and product rules for continuous functions, starting from the base cases that f (x) = x
and constant functions are continuous). So if Bn f → f uniformly on [0, 1] we must have that
f is continuous by Theorem 24.3.
28.4 Let f (x) = x2 sin x1 for x 6= 0 and f (0) = 0.
(a) Use Theorems 28.3 and 28.4 to show f is differentiable at each a 6= 0 and calculate f 0 (a).
Use, without proof, the fact that sin x is differentiable and cos x is its derivative.
Solution: Note that since 1 and x are differentiable functions on R and if a 6= 0 then
x 6= 0 at x = a we have that g(x) = 1/x is differentiable at a by Theorem 28.3(iv). Using
the formula, we see that g 0 (a) = −1/a2 . We also know sine is differentiable everywhere,
so by Theorem 28.4 we have that sin ◦g is differentiable at a, and
1
(sin ◦g)0 (a) = cos(g(a))g 0 (a) = − cos( )/a2 .
a
2
 Lastly, since x is differentiable everywhere, by two applications of (iii) we have that f is
differentiable for x 6= 0 and the value at x = a is
1 1
f 0 (a) = 2a sin − cos .
a a
(b) Use the definition to show f is differentiable at x = 0 and f 0 (0) = 0.
Solution: To show f is differentiable at x = 0 and f 0 (0) = 0, by definition, we must
show that
f (x) − f (0)
lim = 0,
x→0 x−0
or equivalently (since f (0) = 0)
f (x)
lim = 0.
x→0 x

Now note that for x 6= 0 we have f (x)/x = x sin(1/x), and since the existence and value
of the above limit depend only on the values of f (x)/x away from x = 0, we have reduced
the problem to showing  
1
lim x sin = 0.
x→0 x
To see this, let ε > 0. Then take δ < ε. Then for 0 < |x| < δ we clearly have
 
1
x sin ≤ |x| < δ < ε,
x
so indeed we have limx→0 f (x)/x = 0, meaning f 0 (0) exists and f 0 (0) = 0.
(c) Show f 0 is not continuous at x = 0. [Max’s note: you should really study this fact
closely. I think it should defy almost anyone’s intuition that f 0 can exist at a point and
yet fail to be continuous there.]
Solution: Consider the sequence xn = 1/(πn), which has lim xn = 0. Using the formula
from part (a), we see that

f 0 (xn ) = − cos(πn) = (−1)n+1 ,

so lim f 0 (xn ) does not exist, showing f 0 is not continuous at 0.

28.5 Let f (x) = x2 sin x1 for x 6= 0, f (0) = 0, and g(x) = x for x ∈ R.

(a) Observe f and g are differentiable on R.

Solution: We showed in the previous problem that f is differentiable everywhere, and
g is trivially differentiable.
1
(b) Calculate f (x) for x = πn , n = ±1, ±2, . . ..
Solution: We have  
1 1
f = sin(πn) = 0.
πn π 2 n2
g(f (x))−g(f (0))
(c) Explain why limx→0 f (x)−f (0) is meaningless.
Solution: The expression is meaningless because the denominator f (x) − f (0) (equal
to f (x) because f (0) = 0) is zero at all the points x = 1/(nπ) where n ∈ Z \ {0}, so
there is no open interval (−ε, ε) containing zero on which the expression in the limit is
defined (i.e. we always have 1/(nπ) ∈ (−ε, ε) for sufficiently large n).
3
28.8 Let f (x) = x2 for x rational and f (x) = 0 for x irrational.

(a) Prove f is continuous at x = 0.

√
Solution: Let ε > 0. Take δ < ε. Now let |x − 0| = |x| < δ. We claim |f (x) − f (0)| =
|f (x)| < ε. To see this, note that for any x with |x| < δ either f (x) = x2 or f (x) = 0,
and in both cases we have |f (x)| ≤ |x|2 < δ 2 < ε, so indeed |f (x)| < ε as claimed. This
shows f is continuous at 0.
(b) Prove f is discontinuous at all x 6= 0.
Solution: Let x 6= 0. There are two cases: either x is rational or it is irrational. If x is
rational, then we have f (x) = x2 6= 0. However, in this case we can consider a sequence
of points (xn ) which are all irrational such that xn → x (this is easily constructed since
the irrational numbers are dense). For this sequence, we have f (xn ) → 0 6= f (x), so f
cannot be continuous at x. On the other hand, if x is irrational, f (x) = 0, but we can
similarly take a sequence of points (xn ) which are all rational with xn → x, and then
f (xn ) → x2 6= 0 = f (x), so again f is not continuous at x = 0.
(c) Prove f is differentiable at x = 0. Warning: You cannot simply claim f 0 (x) = 2x. [Max’s
note: again, you should take special notice of the pathology exhibited in this problem.
Recall part 1 of the T/F section on midterm 2 – the conclusion was that a function can be
continuous at a single point while failing to be continuous at any other points nearby (or
indeed anywhere else at all). In this case we something even stronger – hence even less
intuitive – that f can be not only continuous but differentiable at a point, while failing
EVEN to be continuous at a single other point. In fact, this example can be generalized
quite broadly to show that a function can have any finite number of derivatives at a
point while being continuous nowhere else (with only slightly more difficulty, one can
even conjure an example of a function that is smooth – having infinitely many derivatives
– at a single point while being continuous nowhere else). This is a powerful testament
to the local nature of the properties of continuity and differentiability. CORRECTION
8/14: This is wrong. The way we have defined the derivative, if f (n) exists at a point,
then in fact all lower derivatives must exist in a neighborhood of that point, meaning
continuity certainly holds in a neighborhood if n > 1. See the posted announcement.]
Solution: We seek to show limx→0 f (x)/x = 0, which suffices to show that f is differen-
tiable at 0 and moreover f 0 (0) = 0. Note that when x 6= 0 we have f (x)/x = x when x is
rational and f (x)/x = 0 when x is irrational, and in either case we have |f (x)/x| ≤ |x|.
So let ε > 0 and take δ < ε. Then for 0 < |x| < δ we have |f (x)/x| ≤ |x| < δ < ε,
thus verifying definitionally that limx→0 f (x)/x = 0, and therefore that f 0 exists at 0
and f 0 (0) = 0.

29.5 Let f be defined on R, and suppose |f (x) − f (y)| ≤ (x − y)2 for all x, y ∈ R. Prove f is a
constant function.
Solution: First we claim that f is differentiable at all a ∈ R, and f 0 (a) = 0. To see this, let
a ∈ R and we will show
f (x) − f (a)
lim = 0.
x→a x−a
To see this, let ε > 0 and take δ < ε. Then for 0 < |x − a| < δ we have by the assumption in
the problem that
|f (x) − f (a)| ≤ |x − a|2 ,

4
 so after dividing both sides by |x − a| we see that

f (x) − f (a)
≤ |x − a| < δ < ε,
x−a

so indeed f is differentiable at a and f 0 (a) = 0. Now since f 0 (x) = 0 for all x, we have by
Theorem 29.4 that f is constant on any open interval (a, b) with a, b finite. But this directly
implies that f is constant on R, since if f (x1 ) 6= f (x2 ) for some x1 < x2 then we would have
that f is not constant on the open interval (x1 − 1, x2 + 1), contradicting what we just proved.

29.7 (a) Suppose f is twice differentiable on an open interval I and f 00 (x) = 0 for all x ∈ I. Show
f has the form f (x) = ax + b for suitable constants a and b. [Max’s note: Your solution
should make NO REFERENCE to the notion of an integral whatsoever. Consider trying
to show that f 0 (x) = a for some a ∈ R.]
Solution: Note that the argument at the end of exercise 29.5 above shows that if a
function f is differentiable on any open interval I (in particular where one or both
of the “endpoints” of I may be infinite) with f 0 = 0 on all of I then we still have
that f is constant. So, in this case, since (f 0 )0 = 0 on I we must have that f 0 is
constant, say f 0 (x) = a for some a ∈ R. Now consider the function g(x) = ax on I. We
straightforwardly have that g 0 (x) = a for x ∈ I, so that g 0 = f 0 on I. Then by Corollary
29.5 (extended to possibly unbounded intervals by the same technique as above) we have
that there is a constant, call it b in this case, so that f (x) = g(x) + b for x ∈ I. But this
means f (x) = ax + b for x ∈ I, as claimed.
(b) Suppose f is three times differentiable on an open interval I and f 000 = 0 on I. What
form does f have? Prove your claim.
Solution: We claim f (x) = ax2 + bx + c for some constants a, b, c ∈ R. To see this, first
note that since (f 0 )00 = 0 we have by the previous part that f 0 (x) = 2ax + b for some
a, b ∈ R. Now consider the function g(x) = ax2 + bx. We see that g 0 (x) = 2ax + b on I,
so we have g 0 = f 0 on I meaning there is some constant, call it c, so that f (x) = g(x) + c
for x ∈ I. This means f (x) = ax2 + bx + c, as desired.