HOMEWORK FOR 18.

101, FALL 2007

ASSIGNMENT 1 SOLUTIONS

(1) Given a linear map

T : Rm −→ Rn
Define the operator norm of T as follows:
kT (x)k
kT k := sup
x6=0 kxk
Similarly, if A is a matrix, define the operator norm of A by
kAxk
kAk := sup
x6=0 kxk
(a) Show that kT k is finite.
(b) Show that k·k is a norm.
(c) Show that if AB is defined,
kAk kBk ≥ kABk
Solution:
x
(a) By rescaling x to kxk (which has norm equal to 1), and by using the
fact that T is linear, we get an alternative characterization:
kT k = sup kT (x)k
kxk=1

We will now show that this quantity is finite. Let us think of T as an

m × n matrix in the corresponding standard bases of Rn and Rm . We
find a constant M > 0 such that all the entries of T are bounded in
absolute value by M . Let x = (x1 , . . . , xn ) ∈ Rn be such that kxk = 1.
We know that then |xj | ≤ 1∀j = 1, . . . , n. Using the triangle inequality
and the bounds on the entries of T , it follows that every entry
√ of T (x)
is bounded in absolute value by nM. Hence kT (x)k ≤ nM n. We can
take supremums
√ over all x ∈ Rn such that kxk = 1 to obtain that:
kT k ≤ nM n < +∞ Hence, kT k is indeed finite.
(b) In order to show that k·k indeed gives us a norm on the space L(Rn , Rm )
of linear operators from Rn to Rm , we need to check the following
properties:
(i) Positive Definiteness: For all T ∈ L(Rn , Rm ), kT k ≥ 0 and
kT k = 0 if and only if T = 0.
The fact that kT k ≥ 0, ∀T ∈ L(Rn , Rm ) follows from definition
of k·k . It is also an immediate consequence of the definition that
k0k = 0. Suppose T ∈ L(Rn , Rm ) is such that kT k = 0. Then
T (x) = 0, ∀x ∈ Rn such that kxk = 1. Since every element of
Rn is a scalar multiple of an element of unit norm, it follows by
linearity of T that T = 0 on Rn .
1
2 HOMEWORK FOR 18.101, FALL 2007 ASSIGNMENT 1 SOLUTIONS

(ii) Homogeneity: Let T ∈ L(Rn , Rm ) and λ ∈ R be given. We

must show that:kλT k = |λ| kT k .
To see this, suppose that x ∈ Rn and kxk = 1. Then k(λT )(x)k =
|λ| kT (x)k . Taking supremums over x ∈ Rn with kxk = 1, it fol-
lows that kλT k = |λ| kT k .
(iii) Triangle Inequality: We must show that ∀A, B ∈ L(Rn , Rm )
kA + Bk ≤ kAk + kBk . To see this, let A, B be as above. Let
x ∈ Rn be such that kxk = 1. Then, by the triangle inequality
in Rn , we have that:
k(A + B)(x)k ≤ kA(x)k + kB(x)k
Taking supremums over kxk = 1, we indeed get:
kA + Bk ≤ kAk + kBk
(c) We observe that ∀ T ∈ L(Rn , Rm ), ∀x ∈ Rn , we have: kT (x)k ≤
kT k kxk (∗). Namely, this fact evidently holds in the case that x = 0
and the case x 6= 0 follows from the original definition of kT k . Suppose
now that A and B are such that AB is well defined. Let v be an
element of the domain of B and suppose that kvk = 1. We obtain
that: k(AB)(v)k = kA(B(v))k ≤ {by using (*)} ≤ kAk kB(v)k ≤
{by using (*) again } ≤ kAk kBk kvk . Taking supremums over kvk =
1, it follows that indeed kAk kBk ≥ kABk , as was claimed.
(2) Prove that the two norms |·|s and k·k on Rn give the same topology in the
sense that if U is an open set using the metric from one norm, it is open
using the metric from the other norm. pP
(Recall that we defined |x|s := max |xi |, and kxk := x2i .)
Solution:
n
We observe that ∀x = (x1 , . . . , xn ) ∈ Rp we have: |x|s ≤ kxk. This is
P 2
because for j = 1, . . . , n we have: |xj | ≤ xi .
On the other hand, |x j | ≤ |x| s ∀j = 1, . . . , n. We thus get that: kxk =
pP
2
√
xi ≤ n |x|s .
From the preceding discussion, we deduce that:
√
∀x ∈ Rn , |x|s ≤ kxk ≤ n |x|s (∗)

Using (∗), we now prove the claim.

Let us now define the notation that we will use for the rest of the problem.
Let p ∈ Rn , and r > 0 be given. We denote by
Br (p) := {x ∈ Rn , kx − pk < r}

Br,s (p) := {x ∈ Rn , |x − p|s < r}

• Suppose that U ⊆ Rn is open with respect to k·k . If U is empty, then U
is automatically open with respect to |·|s . Hence, it suffices to consider
the case when U is nonempty. Let us choose p ∈ U . Then, since U
is open with respect to k·k, we can find r > 0 such that Br (p) ⊆ U .
From (*), we have that B √rn ,s (p) ⊆ Br (p). Hence B √rn ,s (p) ⊆ U, so U
is open with respect to |·|s .
 HOMEWORK FOR 18.101, FALL 2007 ASSIGNMENT 1 SOLUTIONS 3

• Suppose that U ⊆ Rn is open with respect to |·|s . As before, we

consider the case of nonempty U and let p ∈ U be given. We can find
r > 0 such that Br,s (p) ⊆ U. By using (*), we know that Br (p) ⊆
Br,s (p).Hence Br (p) ⊆ U,so U is open with respect to k·k .
Conclusion: k·k and |·|s define the same topology.
(3) (a) Show that given any m × n matrix A, the transpose of the matrix AT
is the unique n × m matrix with the property that
(Ax) · y = x · (AT )y ∀x ∈ Rn , y ∈ Rm
(b) Show that
kAk = AT
Solution:
(a) Suppose that A is an m × n matrix. Let {ej , j = 1, . . . , n}, {fi , i =
1, . . . , m} denote the standard bases for Rn and Rm respectively. We
know from bilinearity that an n×m matrix B satisfies (Ax)·y = x·(By)
∀x ∈ Rn , y ∈ Rm if and only if we have that ∀i = 1, . . . , m, j =
1, . . . , n (Aej ) · fi = ej · (Bfi ) Now, ∀i = 1, . . . , m, j = 1, . . . , n, we
have that: (Aej ) · fi = ith entry of the j th column of A, which we
denote by Ai j . Also, ej · (Bfi ) = j th entry of the ith column of B,
which we denote by Bj i . Hence, from the above we may conclude
that, given an n × m matrix B we have that:
(Ax) · y = x · (By) ∀x ∈ Rn , y ∈ Rm
if and only if
Ai j = Bj i ∀i = 1, . . . , m, j = 1, . . . , n.
The latter set of equalities is equivalent to the fact that B = AT , i.e.
that B is the transpose of A.
(b) Let us show that AT ≤ kAk. Since (AT )T = A, i.e since (·)T is an
involution, the claim will then follow. We know from Problem 1 that,
given any linear map B : Rp −→, Rq kBk = supkvk=1 kBvk .
Suppose now that v ∈ Rn with kvk = 1 and w ∈ Rm with kwk = 1 are
given.
Then we know that (AT )w·v = w·Av. Combining the Schwarz-Cauchy
Inequality and the fact that kAvk ≤ kAk kvk, it follows that the right
hand side of the above inequality has absolute value less than or equal
to: kwk kAk kvk = kAk .
In particular, given w ∈ Rm with kwk = 1, we can find v ∈ Rn with
(AT )w
kvk = 1 such that (AT )w · v = (AT )w Namely, we let v := k(A T )wk

if (A )w 6= 0 and we let v be an arbitrary element of unit norm in Rm

T

if (AT )w = 0.
It follows from the previous discussion that, ∀w ∈ Rm with kwk = 1
we have (AT )w ≤ kAk Taking supremums over such w, it follows
that:
AT ≤ kAk
As we noted earlier, from this inequality we may deduce that:
AT = kAk
4 HOMEWORK FOR 18.101, FALL 2007 ASSIGNMENT 1 SOLUTIONS

(4) (a) Suppose that f : Rn −→ Rm is differentiable, and suppose that for all
x, kDf (x)k ≤ 1. Prove that
kf (x) − f (y)k ≤ kx − yk
(Hint: Try taking the dot product with f (x) − f (y), and use the chain
rule to convert this into a single variable problem. Then you can use
the mean value theorem.)
(b) Find a counterexample to the following naive generalization of the
mean value theorem: Given f : Rn −→ Rm differentiable and points
x, y ∈ Rn , there exists some point c on the line segment between x and
y so that
Df (c)(x − y) = f (x) − f (y)
Solution:

(a) Let x, y ∈ Rn be given. Let us define:

g : R −→ R by
g(t) := (f (y+t(x−y))−f (y))·(f (x)−f (y)). We observe that then g is a
differentiable function of t. Also g(0) = 0 and g(1) = (kf (x) − f (y)k)2
By applying the Chain Rule, we know that ∀t ∈ R
g ′ (t) = (Df (y + t(x − y))(x − y)) · (f (x) − f (y))
We can now use the Mean Value Theorem to get that there exists some
t0 ∈ (0, 1) so that
g(1) − g(0) = g ′ (t0 )
Therefore,
2
kf (x) − f (y)k = (Df (y + t0 (x − y))(x − y)) · (f (x) − f (y))
Now we can use the Cauchy-Schwartz inequality
2
kf (x) − f (y)k ≤ kDf (y + t0 (x − y))(x − y)k kf (x) − f (y)k
Therefore, as all terms are non negative,
kf (x) − f (y)k ≤ kDf (y + t0 (x − y))(x − y)k ≤ kDf (y + t0 (x − y))k kx − yk ≤ kx − yk
(b) We consider the function f : R −→ R2 be such that f (t) is:
 
cos t
sin t
Then f is differentiable and Df (t) is given by
 
− sin t
cos t
Since (sin t)2 + (cos t)2 = 1, it follows that Df (t) has rank 1∀t ∈ R, so
Df (t) is injective ∀t ∈ R. Observe that then f (0) = f (2π). However,
if there existed a point c on the line segment joining 0 and 2π such that
Df (c)(2π−0) = f (2π)−f (0), then it would follow that Df (c)(2π) = 0.
This is impossible since 2π 6= 0 and since Df (c) is injective. Thus, the
‘naive’ generalization of the Mean Value Theorem doesn’t hold.
 HOMEWORK FOR 18.101, FALL 2007 ASSIGNMENT 1 SOLUTIONS 5

(5) (a) Let f : Rn −→ Rn be of class C 1 . Prove that the set S ⊂ Rn consisting

of points x ∈ Rn where Df (x) has rank n is open.
(Hint: The determinant has a formula which is a polynomial in the
coefficients of the matrix. This tells you that the determinant in a
continuous function of the coefficients of a matrix. Use this.)
(b) Use the inverse function theorem to prove that f (S) ⊂ Rn is also open.
Solution:
(a) We observe that:
S = {x ∈ Rn , Df (x) has rank n} = {x ∈ Rn , det(Df (x)) 6= 0}
We also know that the determinant of a matrix is a polynomial in its
entries. This follows from the definition of a determinant. Now, for f
as above, the quantity det(Df (x)) is a polynomial in the n2 quantities
∂fj 1
∂xi (x), where i, j = 1, . . . , n. Since f is a class C function, it follows
∂f
that ∂xji (x) is a continuous function in x for i, j = 1, . . . , n. From here
we may deduce that det(Df (x)) is a continuous function in x ∈ Rn .
Hence, by continuity, the set S = {x ∈ Rn , det(Df (x)) 6= 0} is open
as it is the inverse image of an open subset of R.
(b) In order to prove the claim, we want to use the Inverse Function The-
orem.
Suppose that y ∈ f (S) is given. By definition, we can find x ∈ S such
that f (x) = y. Then, since x ∈ S, it follows that det(Df (x)) 6= 0 so
by the Inverse Function Theorem, we can find a neighborhood U of x
in Rn and a neighborhood V of y in Rn such that f is a bijection from
U to V which has an inverse that is of class C 1 . Call this inverse g.
In particular, we know by the Chain Rule that:
(Df (w))◦(Dg(f (w))) = In ∀w ∈ U. Hence Df (w) is invertible ∀w ∈ U .
Thus, it follows that U ⊆ S. Hence, V = f (U ) ⊆ f (S). V is also a
neighborhood of y in Rn by construction. Since such a neighborhood
V can be found for all y ∈ f (S), it follows that f (S) is open, as was
claimed.