February 7, 2014

Math 361: Homework 2 Solutions

1. Let U be an open subset in Rn , f, g : U → Rm be two differentiable functions and a, b be any two real
numbers. Show that af + bg is again differentiable and

D(af + bg) = aDf + bDg

Since f is differentiable, we know that for all x ∈ U , we can write:

f (x + h) − f (x) = f 0 (x)h + rf (h) (1)

|rf (h)|
where f 0 (x) is Df evaluated at x and where limh→0 |h| = 0.
Similarly, since g is differentiable, we can do the same for a sublinear remainder rg (h):

g(x + h) − g(x) = g 0 (x)h + rg (h) (2)

Next, we try to find the derivative of af + bg by considering (af + bg)(x + h) − (af + bg)(x):

(af + bg)(x + h) − (af + bg)(x) = af (x + h) + bg(x + h) − (af (x) + bg(x)) (3)

= a(f (x + h) − f (x)) + b(g(x + h) − g(x)) (4)
= a(f 0 (x)h + rf (h)) + b(g 0 (x)h + rg (h)) (5)
= (af 0 (x) + bg 0 (x))h + arf (h) + brg (h) (6)

Now, we argue that (arf (h) + brg (h)) is sublinear:

|arf (h) + brg (h)| |arf (h)| + |brg (h)|

lim ≤ lim (triangle inequality) (7)
h→0 |h| h→0 |h|
|arf (h)| |brg (h)|
= lim + lim (limits exist individually) (8)
h→0 |h| h→0 |h|
|rf (h)| |brg (h)|
= |a| lim + |b| lim (a, b ∈ R) (9)
h→0 |h| h→0 |h|
= |a| · 0 + |b| · 0 = 0 (differentiability of f, g) (10)

|ar (h)+br (h)|

We have shown that limh→0 f |h| g ≤ 0, but because norms are always non-negative, this
indicates that the limit is actually equal to 0.
We let arf (h) + brg (h) = raf +bg (h). We have shown that:

(af + bg)(x + h) − (af + bg)(x) = (af 0 (x) + bg 0 (x))h + raf +bg (h) (11)
|r (h)|
where af +bg
|h| → 0 as h → 0. Therefore, by the Landau definition of differentiability, we have shown
that af + bg is differentiable at every point x ∈ U and that its derivative is equal to af 0 (x) + bg 0 (x) =
aDf + bDg. Note that this derivative is unique by Theorem 9.12 in Rudin.

2. Let U be an open subset in Rn , and f, g : U → Rm be two differentiable real-valued functions. Show

that the product function f g is again differentiable and

D(f g) = gDf + f Dg

1
Note that here g times Df as a 1 × n matrix is just the entrywise multiplication.

We set up the problem in the same way that we did problem 1: Since f, g are both differentiable for
all x ∈ U :
f (x + h) − f (x) = f 0 (x)h + rf (h) (1)
g(x + h) − g(x) = g 0 (x)h + rg (h) (2)
where rf (h) and rg (h) are sublinear.
Next, we try to find the derivative of f g by considering (f g)(x + h) − (f g)(x):

(f g)(x + h) − (f g)(x) = f (x + h)g(x + h) − f (x)g(x) (3)

= f (x + h)g(x + h) − f (x + h)g(x) + f (x + h)g(x) − f (x)g(x) (4)
= f (x + h)(g(x + h) − g(x)) + g(x)(f (x + h) − f (x)) (5)
= f (x + h)(g 0 (x)h + rg (h)) + g(x)(f 0 (x)h + rf (h)) (6)
= f (x + h)g 0 (x)h + f (x + h)rg (h) + g(x)f 0 (x)h + g(x)rf (h) (7)
0 0 0
= [f (x) + f (x)h + rf (h)]g (x)h + f (x + h)rg (h) + g(x)f (x)h + g(x)rf (h) (8)
= [f (x)g 0 (x) + g(x)f 0 (x)]h + [f (x + h)rg (h) + g(x)rf (h) + f 0 (x)g 0 (x)h2 (9)
0
+ rf (h)g (x)h]

Note that in the above calculation we subbed in for g(x + h) − g(x), f (x + h) − f (x) and then again
for just f (x + h) all from the definition of differentiability in (1) and (2).
We let rf g = f (x + h)rg (h) + g(x)rf (h) + f 0 (x)g 0 (x)h2 + rf (h)g 0 (x)h. Now, we argue that each term
in rf g is actually sublinear:

|f (x + h)rg (h)| |rg (h)|

lim = lim |f (x + h)| · lim = |f (x)| · 0 = 0 (10)
h→0 |h| h→0
| {z }
h→0 |h|
| {z }
exists because f continuous exists because g differentiable

(11)
|g(x)rf (h)| |rf (h)|
lim = |g(x)| · lim = |g(x)| · 0 = 0 (12)
h→0 |h| | {z } h→0 |h|
doesn’t depend on h

(13)
0 0 2 2
|f (x)g (x)h | |h |
lim = |f 0 (x)g 0 (x)| · lim = |f 0 (x)g 0 (x)| · lim |h| = |f 0 (x)g 0 (x)| · 0 = 0 (14)
h→0 |h| | {z } h→0 |h| h→0
doesn’t depend on h

(15)
0
|rf (h)g (x)h| |rf (h)|
lim = |g 0 (x)| · lim · lim |h| = |g 0 (x)| · 0 · 0 = 0 (16)
h→0 |h| | {z } h→0 |h| h→0
doesn’t depend on h | {z }
exists because f differentiable

Now, since each term is sublinear, it must mean that rf g is sublinear:

|rf g | |f (x + h)rg (h)| |g(x)rf (h)| |f 0 (x)g 0 (x)h2 | |rf (h)g 0 (x)h|
lim ≤ lim + lim + lim + lim (17)
h→0 |h| h→0 |h| h→0 |h| h→0 |h| h→0 |h|
=0+0+0+0=0 (18)
|rf g |
This implies that limh→0 |h| actually equals zero because the norms can never be negative.

2
 Therefore, we have shown that:
(f g)(x + h) − (f g)(x) = (f (x)g 0 (x) + g(x)f 0 (x))h + rf g (h) (19)
|r (h)|
where f|h| g
→ 0 as h → 0. Therefore, by the Landau definition of differentiability, we have shown
that f g is differentiable at every point x ∈ U and that its derivative is equal to f (x)g 0 (x) + g(x)f 0 (x) =
f Dg + gDf . Note that this derivative is unique by Theorem 9.12 in Rudin.

3. Let T be a linear transformation from Rn to Rm . Show that T : Rn → Rm is differentiable as a map

and compute its total derivative.

In this question, we will use the limit definition of differentiability (just to mix it up a little). A function
f : Rn → Rm is differentiable at x ∈ Rm if there exists some linear transformation A of Rn into Rm
such that:
|f (x + h) − f (x) − A(h)|
lim =0 (1)
h→0 |h|
If this is the case, we call A the total derivative of f .
Now, let us plug in our function T into this definition. We claim that T is differentiable and that its
derivative is just T itself (i.e., A = T ).
|T (x + h) − T (x) − A(h)| |T (x) + T (h) − T (x) − A(h)|
lim = lim (T is linear) (2)
h→0 |h| h→0 |h|
|T (h) − A(h)|
= lim (T (x) cancels) (3)
h→0 |h|
|T (h) − T (h)|
= lim (we get to choose A) (4)
h→0 |h|
0
= lim =0 (5)
h→0 |h|

Therefore, we have found that there does exist a transformation that makes the limit in (1) go to 0,
which means that T is indeed differentiable and that its total derivative is itself. In addition, we know
that this is the unique total derivative by the same Theorem 9.12 in Rudin as before.

4. Pugh Chapter 5 #20: Assume that U is a connected open subset of Rn and f : U → Rm is differentiable
everywhere on U . If (Df )p = 0 for all p ∈ U , show that f is constant.

Lemma: f is locally constant, i.e., for any x ∈ U , there always exists an open neighborhood Nδ (x) of
x inside U such that f is constant over Nδ (x).
Since f is continuous and defined on an open domain, we can take δ small enough to form an open
ball Nδ (x) for any x ∈ U such that Nδ (x) ⊂ U . Now, we want to show that f must be constant under
all elements in Nδ (x).
Let p be an arbitrary point inside Nδ (x). We know that the segment [x, p] is contained within U
because x, p ∈ Nδ (x) ⊂ U . Since f is differentiable, by the multivariable Mean Value Theorem, we
know that:
|f (p) − f (x)| ≤ M |p − x| where M = sup k(Df )x k (1)
x∈U

However, we know that (Df )x = 0 for all x ∈ U . This means that M = supx∈U k0k = 0. Hence, we
have that:
|f (p) − f (x)| ≤ 0 · |p − x| =⇒ |f (p) − f (x)| ≤ 0 =⇒ f (p) − f (x) = 0 =⇒ f (p) = f (x) (2)

3
 In other words, we have just shown that for any p ∈ Nδ (x), it must be the case that f (p) = f (x),
i.e., the function f is locally constant. Therefore, we have proven the lemma and we now move on to
proving that f must be constant over all of U .

We argue by contradiction. Suppose f is not constant. Then we know that the cardinality of Im(f )
must be strictly greater than 1. Let y ∈ Im(f ), and consider the following two sets:
A = f −1 ({y}) B = f −1 (Im(f ) \ {y}) (3)

We argue that the sets A and B have the following characteristics:

• A and B are both nonempty.
We know that A is nonempty because y ∈ Im(f ), which means there must exist at least one
element in U that maps to {y} =⇒ f −1 ({y}) is nonempty.
Since the cardinality of Im(f ) is greater than 1, it must be the case that Im(f ) \ {y} is nonempty.
Therefore, similarly to above, there must be at least one element in U that maps to Im(f ) \ {y}
which implies that f −1 (Im(f ) \ {y}) is nonempty.

• A∪B =U
It’s clear that {y} ∪ Im(f ) \ {y} = Im(f ), and by definition we know that f −1 (Im(f )) = U .
Therefore, it follows that f −1 ({y} ∪ Im(f ) \ {y}) = U . Finally, since {y} and Im(f ) \ {y} are
disjoint, the previous statement must imply that the union of their inverse images must make up
all of U , i.e., f −1 ({y}) ∪ f −1 (Im(f ) \ {y}) = U . That is, A ∪ B = U .

• A∩B =∅
Suppose not. This must mean that there is an element x ∈ U such that f (x) = y and f (x) = z
where z ∈ Im(f ) \ {y}. This means that f (x) equals two distinct elements in Rn , which would
mean that f is not a function. Therefore, we have a contradiction and we have shown that
A ∩ B = ∅.

• A and B are both open.

Let x be any arbitrary element in A, i.e., f (x) = y. Since A ⊂ U , by our lemma we know that
we can form a δ−neighborhood around x (call it Nδ (x)) such that the value of f under Nδ (x) is
constant. That is, for any t ∈ Nδ (x), we know that f (t) = y. In other words, f (Nδ (x)) = {y}
which implies that Nδ (x) ⊂ f −1 ({y}). Therefore, we have shown that for any x in A, Nδ (x) ⊂ A,
which means that A must be open.
The exact same argument works for B, since our lemma holds for any element in U : Let x be any
arbitrary element in B. We form a δ−neighborhood around x (call it Nδ (x)) such that the value
of f under Nδ (x) is constant. That is, for any t ∈ Nδ (x), we know that f (t) = f (x) ∈ Im(f ) \ {y}.
In other words, f (Nδ (x)) = Im(f ) \ {y} which implies that Nδ (x) ⊂ f −1 (Im(f ) \ {y}). Therefore,
we have shown that for any x in B, Nδ (x) ⊂ B, which means that B must be open.
Hence, we have shown that there exist two open, nonempty, disjoint sets A and B whose union is
U . This must mean that U is disconnected, which contradicts our assumption of U being connected.
Therefore, we know that the cardinality of Im(f ) must equal 1, i.e., f is constant.

7. Pugh Chapter 5 #24: Show that all second partial derivatives of the function f : R2 → R defined by
2 2
 xy(x − y ) if (x, y) 6= (0, 0)


f (x, y) = 2
x +y 2

0 if (x, y) = (0, 0)


4
exist everywhere, but the mixed second partials are unequal at the origin, ∂ 2 f (0, 0)/∂x∂y 6= ∂ 2 f (0, 0)/∂y∂x.

To show that the partial derivatives exist everywhere, we take the derivatives with respect to x and y
by holding the other variable constant and using the product rule and chain rule. For (x, y) 6= (0, 0):
∂f −2x 1
= xy(x2 − y 2 ) · 2 + (xy · 2x + y(x2 − y 2 )) 2 (1)
∂x (x + y 2 )2 x + y2
2 2 2 2 2 2
 −2x y(x − y ) + 2x y + y(x − y ) if (x, y) 6= (0, 0)

∂f
=⇒ = (x2 + y 2 )2 x2 + y 2 (2)
∂x 
0 if (x, y) = (0, 0)
We calculate the partial with respect to y similarly:
∂f −2y 1
= xy(x2 − y 2 ) · 2 + (xy · (−2y) + x(x2 − y 2 )) 2 (3)
∂y (x + y 2 )2 x + y2
2 2 2 2 2 2
 −2xy (x − y ) + −2xy + x(x − y ) if (x, y) 6= (0, 0)

∂f 2 2 2 2 2
=⇒ = (x + y ) x +y (4)
∂x 
0 if (x, y) = (0, 0)
In order to show that all second order partial derivatives exist, we would have to take the derivative
of ∂f ∂f
∂x twice more (once with respect to x and again with respect to y) and the derivative of ∂y twice
more as well. However, notice that when we take the derivatives, we always have some polynomial
of x and y in the numerator and a power of (x2 + y 2 ) in the denominator. This indicates that the
partials will exist whenever the denominator is nonzero (because polynomials are continuous), and the
denominator is zero only at (0, 0), which we define to equal zero. This means that all of the partial
derivatives will exist everywhere.

To show that that the mixed partials do not agree at the origin, we calculate the mixed second partial
derivatives of f using the limit definition of partial derivatives at a point:

∂2f
   
∂ ∂f ∂ f (0 + h, 0) − f (0, 0)
(0, 0) = (0, 0) = lim = (5)
∂y∂x ∂y ∂x ∂y h→0 h
 f (0 + h, 0 + k) − f (0, 0 + k) f (0 + h, 0) − f (0, 0) 
lim − lim
= lim  h→0 h h→0 h
= (6)
 
k→0 k

hk(h2 − k 2 ) 0 · k(0 − k 2 ) h · 0(h2 − 0)

 
2 2
− − 0
 lim h + k
 0 + k2 − lim h2 + 0 

 h→0
= lim  h h→0 h
= (7)

k→0  k 
 

hk(h2 − k 2 ) 0−0 k(0 − k 2 )

   
lim
 h→0 h(h2 + k 2 ) − lim 2
− 0
h→0 h  = lim  (0 + k )
 
= lim  = (8)
k→0  k  k→0  k 

−k 3
 
= lim = lim −1 = −1 (9)
k→0 k3 k→0

5
 ∂2f
   
∂ ∂f ∂ f (0, 0 + h) − f (0, 0)
(0, 0) = (0, 0) = lim = (10)
∂x∂y ∂x ∂y ∂x h→0 h
 f (0 + k, 0 + h) − f (0 + k, 0) f (0, 0 + h) − f (0, 0) 
lim − lim
= lim  h→0 h h→0 h
= (11)
 
k→0 k

kh(k 2 − h2 ) k · 0(k 2 − 0) 0 · h(0 − h2 )

 
2 2
− − 0
 lim k + h
 k2 + 0 − lim 0 + h2 

 h→0
= lim  h h→0 h
= (12)

k→0  k 
 

kh(k 2 − h2 ) 0−0 k(k 2 − 0)

   
lim
 h→0 h(k 2 + h2 ) − lim  2 − 0
h→0 h  = lim  (k + 0)

= lim  = (13)
k→0  k  k→0  k 

k3
 
= lim = lim 1 = 1 (14)
k→0 k3 k→0

Therefore, we have shown that:

∂2f ∂2f
(0, 0) = 1 6= −1 = (0, 0) (15)
∂x∂y ∂y∂x

8. Rudin Chapter 9 #28: For t ≥ 0, put

 √
x (0 ≤ x ≤ t)
√ √ √

ϕ(x, t) = −x + 2 t ( t ≤ x ≤ 2 t)

0 (otherwise),


and put ϕ(x, t) = −ϕ(x, |t|) if t < 0. Show that ϕ is continuous on R2 , and
(D2 ϕ)(x, 0) = 0
for all x. Define Z 1
f (t) = ϕ(x, t)dx
−1

Show that f (t) = t if |t| < 14 . Hence

Z 1
f 0 (0) 6= (D2 ϕ)(x, 0)dx.
−1
√
To show that ϕ is continuous on R2 , we first note that the three functions that make up ϕ (x, −x+2 t,
and 0) are all continuous.
√ Therefore, we √ just have to show that the pieces of ϕ agree on the boundaries,
i.e., at x = 0, at x = t, and at x = 2 t.
lim ϕ(x, t) = lim 0 = 0 (1)
x→0− x→0

lim ϕ(x, t) = lim x = 0 (2)

x→0+ x→0

6
Therefore, the left limit and right limit agree and ϕ is continuous at x = 0.

√
lim
√ − ϕ(x, t) = lim
√ x= t (3)
x→ t x→ t
√ √ √ √
lim
+
√ (−x + 2 t) = − t + 2 t =
√ ϕ(x, t) = lim t (4)
x→ t x→ t

√
Therefore, ϕ is continuous at x = t.

√ √ √
lim
√ ϕ(x, t) = lim√ (−x + 2 t) = −2 t + 2 t = 0 (5)
−
x→2 t x→2 t

lim
√ + ϕ(x, t) = lim√ 0 = 0 (6)
x→2 t x→2 t

√
Therefore, ϕ is continuous at x = 2 t. That is, ϕ is continuous on every boundary and also everywhere
else (because the functions that make it up are continuous), which means that ϕ is continuous on R2 .
Next, we want to show that (D2 ϕ)(x, 0) = 0. We do this by using the definition of partial derivatives:

ϕ(x, 0 + h) − ϕ(x, 0) x−x

(D2 ϕ)(x, 0) = lim = lim = lim 0 = 0 (7)
h→0 h h→0 h h→0

√
Note that we used the definition of ϕ close to 0 because as h → 0 so does h.
1
√
As for the last part of the problem, first consider 0 < t < Note that in this case, 0 < t < 14 =⇒
4.
√
0 < 2 t < 1. This means that when we compute f (t) for 0 < t < 41 , we only need to worry about
integrating between 0 and 1:
√ √
Z 1 Z t Z 2 2 √
f (t) = ϕ(x, t)dx = xdx + √ (−x + 2 t)dx = (8)
0 0 t
√t   2√t
x2  √ x2 
 
t 4t t
= + 2 tx − = − 0 + 4t − − 2t − = (9)
2 0 2  √t 2 2 2
t t
= + 2t − 2t + = t (10)
2 2

When t = 0, ϕ(x, t) = 0 for any x. Therefore, we know that f (0) = 0. When − 14 < t < 0, we know
that ϕ(x, t) = −ϕ(x, |t|), which is to say that for these values of t:

f (x, t) = −f (x, −t) = −(−t) = t (11)

1
Therefore, it follows that for any |t| < 4, f (t) = t. By normal single-variable differentiation, this
implies that f 0 (t) = 1 for these t values.
We have shown that (D2 ϕ)(x, 0) = 0, therefore:
Z 1
0
f (0) = 1 6= 0 = (D2 ϕ)(x, 0)dx (12)
−1