1.

Limits and Continuity

It is often the case that a non-linear function of n-variables x = (x1 , . . . , xn ) is not really
defined on all of Rn . For instance f (x1 , x2 ) = xx21−x
x2
2 is not defined when x1 = ±x2 . However,
1 2
I will adopt a convention from the vector calculus notes of Jones and write F : Rn → Rm
regardless, meaning only that the source of F is some subset of Rn . While a bit imprecise,
this will not cause any big problems and will simplify many statements.
I will often distinguish between functions f : Rn → R that are scalar-valued and functions
F : Rn → Rm , m ≥ 2 that are vector-valued, using lower-case letters to denote the former and
upper case letters to denote the latter. Note that any vector-valued function F : Rn → Rm
may be written F = (F1 , . . . , Fm ) where Fj : Rn → R are scalar-valued functions called the
components of F . For example, F : R2 → R2 given by F (x1 , x2 ) = (x1 x2 , x1 + x2 ) is a vector
valued function with components F1 (x1 , x2 ) = x1 x2 and F2 (x1 , x2 ) = x1 + x2 .
Definition 1.1. Let a ∈ Rn be a point and r > 0 be a positive real number. The open ball
of radius r about a is the set
B(a, r) := {x ∈ Rn : kx − ak < r}.
I will also use B ∗ (a, r) to denote the set of all x ∈ B(a, r) except x = a.
Proposition 1.2. Let a, b ∈ Rn be points and r, s > 0 be real numbers. Then
• B(a, r) ⊂ B(b, s) if and only if ka − bk ≤ s − r.
• B(a, r) ∩ B(b, s) = ∅ if and only if ka − bk ≥ s + r.
Proof. Exercise. Both parts depend on the triangle inequality. 

Extending my above convention, I will say that a function F : Rn → Rm is defined near

a point a ∈ Rn if there exists r > 0 such that F (x) is defined for all points x ∈ B(a, r),
except possibly the center x = a.
Below I will need the following inequality relating the magnitude of a vector to the sizes
of its coordinates.
v1
 

Proposition 1.3. For any vector v =  ...  ∈ Rn we have for each 1 ≤ j ≤ n that
vn
√
|vj | ≤ kvk ≤ n max |vj |.
1≤j≤n

Proof. Suppose that vJ is the coordinate of v with largest absolute value. Then for any
index j, we have
Xn
2
vj ≤ vi2 ≤ nvJ2 .
i=1
Taking square roots throughout gives the inequalities in the statement of the lemma. 

Now we come to the main point. The idea of a ‘limit’ is one of the most important in
all of mathematics. In differential calculus, it is the key to relating non-linear (i.e. hard)
functions to linear (i.e. easier) functions.
1
2

Definition 1.4. Suppose that F : Rn → Rm is a function defined near a point a ∈ Rn . We

say that F (x) has limit b ∈ Rm as x approaches a, i.e.
lim F (x) = b ∈ Rm ,
x→a

if for each  > 0 there exists δ > 0 such that 0 < kx − ak < δ implies kF (x) − bk < .
Notice that the final phrase in this definition can be written in terms of balls instead of
magnitudes: for any  > 0 there exists δ > 0 such that x ∈ B ∗ (a, δ) implies F (x) ∈ B(b, ).
A function might or might not have a limit as x approaches some given point a, but it
never has more than one.
Proposition 1.5 (uniqueness of limits). If F : Rn → Rm is defined near a ∈ Rn , then there
is at most one point b ∈ Rm such that limx→a F (x) = b.
Proof. Suppose, in order to reach a contradiction, that F (x) converges
to two different
1
points b, b̃ ∈ Rm as x approaches a. Then the quantity  := 2 b̃ − b is positive. So by

the definition of limit, there exists a number δ1 > 0 such that 0 < kx − ak < δ1 implies
kF
(x) − bk < . Likewise, there exists a number δ2 > 0 such that 0 < kx − ak < δ2 implies
F (x) − b̃ < . So if I let δ = min{δ1 , δ2 } be the smaller bound, then 0 < kx − ak < δ

implies that

b̃ − b = (F (x) − b) − (F (x) − b̃) ≤ kF (x) − bk + F (x) − b̃ <  +  ≤ b̃ − b .


Note that the ‘≤’ in this estimate follows from the triangle inequality, and the ‘<’ follows
from my choice of . At any rate, no real number is smaller than itself, so I have reached a
contradiction and conclude that F cannot have two different limits at a. 

Definition 1.6. We say that a function F : Rn → Rm is continuous at a ∈ Rn if F is

defined near and at a and
lim F (x) = F (a).
x→a
If F is continuous at all points in its domain, we say simply that F is continuous.
Now let us verify that many familiar scalar-valued functions are continuous.
Proposition 1.7. The following are continuous functions.
(a) The constant function f : Rn → R, given by f (x) = c for some fixed c ∈ R and all
x ∈ Rn .
(b) The magnitude function f : Rn → R given by f (x) = kxk.
(c) The addition function f : R2 → R given by f (x1 , x2 ) = x1 + x2 .
(d) The multiplication function f : R2 → R given by f (x1 , x2 ) = x1 x2 .
(e) The reciprocal function f : R → R given by f (x) = 1/x.
Proof. (a) Fix a point a ∈ Rn . Given  > 0, let δ > 0 be any positive number, e.g. δ = 1
(it won’t matter). Then if kx − ak < δ it follows that
|f (x) − f (a)| = |c − c| = 0 < .
So the constant function f (x) = c is continuous at any point a ∈ Rn .
 3
n
(b) Fix a point a ∈ R . Given  > 0, let δ = . Then if kx − ak < δ, it follows that
|f (x) − f (a)| = | kxk − kak | ≤ kx − ak < δ = .
The ‘<’ here follows from Problem 1.2.17 (on Homework 1) in Shifrin. Hence f (x) =
kxk is continuous at any point a ∈ Rn .

(c) Fix a point a = (a1 , a2 ) ∈ R2 . Given  > 0, let δ = /2. Then if kx − ak < δ, it
follows that
|f (x) − f (a)| = |x1 + x2 − a1 − a2 | ≤ |x1 − a1 | + |x2 − a2 | < δ + δ = .
Hence f (x1 , x2 ) = x1 + x2 is continuous at any point (a1 , a2 ) ∈ R2 .

(d) Fix a point a = (a1 , a2 ) ∈ R2 . Given  > 0, let δ = min{1, (1 + |a1 | + |a2 |)−1 }. Then
if kx − ak < δ, it follows that
|f (x) − f (a)| = |x1 x2 − a1 a2 | = |(x1 x2 − x1 a2 ) + (x1 a2 − a1 a2 )|
≤ |x1 x2 − x1 a2 | + |x1 a2 − a1 a2 | = |x1 ||x2 − a2 | + |a2 ||x1 − a1 |
< δ(|x1 | + |a2 |) < δ(|a1 | + 1 + |a2 |) ≤ .
Notice that the final ‘<’ follows from the fact that |x1 −a1 | < δ ≤ 1. Hence f (x1 , x2 ) =
x1 x2 is continuous at any point a = (a1 , a2 ) ∈ R2 .

(e) Homework exercise.



Theorem 1.8. Linear transformations T : Rn → Rm are continuous.

This one requires a little warm-up. If A is an m × n matrix, let us define the magnitude
of A to be the quantity v
u m X n
uX
kAk := t a2ij .
i=1 j=1

That is, we are measuring the size of A as if it were a vector in Rmn . Be aware that elsewhere
in mathematics (including Shifrin), there are other notions of the magnitude of a matrix.
Lemma 1.9. Given any matrix A ∈ Mm×n and any vector x ∈ Rn , we have kAxk ≤
kAk kxk.
Proof. Let rowi denote the ith row of A. Then on the one hand, we have
Xm X n m
X
2
kAk = 2
aij = krowi k2 .
i=1 j=1 i=1

But on the other hand, the ith entry of Ax is rowi · x. Hence from the Cauchy-Schwartz
inequality, we obtain
m
X m
X m
X
2 2 2 2
kAxk = 2
(rowi · x) ≤ krowi k kxk = kxk krowi k2 = kAk2 kxk2 .
i=1 i=1 i=1


 4

Proof of Theorem 1.8. Let A ∈ Mm×n be the standard matrix of the linear transformation

T : Rn → Rm and a ∈ Rn be any point. Given  > 0, I choose δ = kAk . Then if kx − ak < δ,
it follows that
kT (x) − T (a)k = kT (x − a)k = kA(x − a)k ≤ kAk kx − ak < kAk δ = .
Hence T is continuous at any point a ∈ Rn 

Questions about limits of vector-valued functions can always be reduced to questions about
scalar-valued functions.
Proposition 1.10. Suppose that F : Rn → Rm is a vector-valued function F = (F1 , . . . , Fm )
defined near a ∈ Rn . Then the following are equivalent.
(a) limx→a F (x) = b ∈ Rm .
(b) limx→a kF (x) − bk = 0.
(c) limx→a Fj (x) = bj for 1 ≤ j ≤ m.
Proof of Proposition 1.10.
(a) =⇒ (b) Suppose that limx→a F (x) = b and set f (x) := kF (x) − bk. Given  > 0, the
definition of limit gives me δ > 0 such that 0 < kx − ak < δ implies that kF (x) − bk < .
But this last inequality can be rewritten |f (x) − 0| < . Thus limx→a f (x) = 0, i.e. (b)
holds.
(b) =⇒ (a) Similar.
(a) =⇒ (c) Suppose again that limx→a F (x) = b. Fix an index j between 1 and m and let
 > 0 be given. Since limx→a F (x) = b, there exists δ > 0 such that 0 < kx − ak < δ implies
that kF (x) − bk < . Then by Proposition 1.3, I also have that
|Fj (x) − bj | ≤ kF (x) − bk < .
So limx→a Fj (x) = bj .
(c) =⇒ (a) Suppose that limx→a Fj (x) = bj for each 1 ≤ j ≤ m. Then given any  > 0, there
exist real numbers δj > 0 such that 0 < kx − ak < δj implies that |Fj (x) − bj | < √m . Taking
δ = min{δ1 , . . . , δm }, I infer that 0 < kx − ak < δ implies |Fj (x) − bj | < √m for all indices
j. Thus by the lemma,
√ 
kF (x) − bk < m √ = .
m
So limx→a F (x) = b. 

The following theorem is sometimes paraphrased by saying that limits commute with
continuous functions.
Theorem 1.11 (composite limits). Let F : Rn → Rm and G : Rm → Rp be functions and
a ∈ Rn , b ∈ Rm be points such that limx→a F (x) = b and G is continuous at b. Then
lim G ◦ F (x) = G(lim F (x)) = G(b).
x→a x→a

Proof. Let  > 0 be given. By continuity of G at b, there exists a number 0 > 0 such
that ky − bk < 0 implies kG(y) − G(b)k < . Likewise, since limx→a F (x) = b, there
exists δ > 0 such that 0 < kx − ak < δ implies that kF (x) − bk < 0 . Putting these two
 5

things together, I see that 0 < kx − ak < δ further implies that kG(F (x)) − G(b)k < . So
limx→a G(F (x)) = G(b). 

Corollary 1.12. Let F : Rn → Rm and G : Rm → Rp be continuous functions. Then G ◦ F

is continuous.
Proof. Note that G ◦ F is defined at a ∈ Rn precisely when F is defined at a and G is defined
at F (a). Then Since both functions are continuous wherever they are defined, we have
lim G(F (x)) = G(lim F (x)) = G(F (a)).
x→a x→a
n
Hence G ◦ F is continuous at any point a ∈ R where it is defined. 

Corollary 1.13. Let f, g : Rn → R be functions with limits limx→a f (x) = b and limx→a g(x) =
c at some point a ∈ Rn . Then
(a) limx→a |f (x)| = |b|.
(b) limx→a f (x) + g(x) = b + c;
(c) limx→a f (x)g(x) = bc;
1
(d) limx→a f (x) = 1b , provided b 6= 0.
Hence a sum or product of continuous functions is continuous, as is the reciprocal of a
continuous function.
Actually, the corollary extends to dot products, magnitudes and sums of vector-valued
functions F, G : Rn → Rm , too. I’ll let you write down the statements of these facts.
Proof. I prove (c). The other parts are similar. Let F : R → R2 be given by F (x) :=
(f (x), g(x)) and m : R2 → R be the multiplication function m(y1 , y2 ) := y1 y2 . Recall that
m is continuous (Proposition 1.7). Moreover, Proposition 1.10 and our hypotheses about f
and g imply that limx→a F (x) = (b, c). Hence I infer from Theorem 1.11 that
lim f (x)g(x) = lim m(F (x)) = m(lim F (x)) = m(b, c) = bc.
x→a x→a x→a


When used with the fact that functions can’t have more than one limit at a given point,
Theorem 1.11 leads to a useful criterion for establishing that a limit doesn’t exist.
Definition 1.14. A parametrized curve is a continuous function γ : R → Rn .
Corollary 1.15. Given a function F : Rn → Rm defined near a point a ∈ Rn , suppose that
γ1 , γ2 : R → Rn are parametrized curves such that γ1 (t) = γ2 (t) = a if and only if t = 0.
If the limits limt→0 F ◦ γ1 (t) and limt→0 F ◦ γ2 (t) are not equal, then limx→a F (x) does not
exist.
Proof. I will prove the contrapositive statement: suppose that limx→a F (x) = b exists and
γ1 , γ2 : R → Rn are parametrized curves with initial points γj (0) = a but γj (t) 6= a for
t 6= a. Then the limits limx→a F (x) and limt→0 F ◦ γj (t) do not concern the value of F at a.
So I may assume with no loss of generality that F (a) = b, i.e. that F is actually continuous
at a.
Theorem 1.11 then tells me that
lim F ◦ γ1 (t) = F (lim γ1 (t)) = F (γ1 (0)) = F (a) = b.
t→0 t→0
6

The second equality holds because γ1 is continuous. Likewise, limt→0 F ◦ γ2 (t) = b. In

particular, the two limits are the same. 

I remark that if γ : R → Rn is a continuous curve and F : Rn → Rm is a function, then

the composite function F ◦ γ : R → Rm is sometimes called the restriction of F to γ.
One last fact about limits that will prove useful for us is the following.
Theorem 1.16 (The Squeeze Theorem). Suppose that F : Rn → Rm and g : Rn → R are
functions defined near a ∈ Rn . Suppose there exists r > 0 such that
• kF (x)k ≤ |g(x)| for all x ∈ B ∗ (a, r);
• limx→a g(x) = 0.
Then limx→a F (x) = 0.
Proof. Exercise.