Вы находитесь на странице: 1из 89

Sequences in Rm

Distance in Rk

The Euclidean distance (or simply distance)


between x = (x1, x2, . . . , xk ) and
y = (y1, y2, . . . , yk ) in Rk is defined to be
v
u
u k
kx − y k = (xi − yi)2.
u X
u
t
i=1
1
(The case in which k = 1 corresponds to the
distance |x − y| between two real numbers.)
There is a certain mathematical attraction
in defining distances on rather more
abstract or unusual spaces, and this leads to
the notion of a metric space. This is
something we will touch on later in this
course. For the moment, we are primarily
interested in Euclidean space Rk , for some
integer k ≥ 1, and we shall always use the
Euclidean distance.
Note: The Euclidean distance between two
vectors x, y in Rk is simply the norm or
length of x − y, where the norm is the one
arising from the usual inner product on Rk .
(See Linear Algebra.) Consequently, the
distance measure has some nice properties.
For example, by the triangle inequality for
norms, we have that for any x, y, z ∈ Rm,

kx − zk = k(x − y) + (y − z)k ≤ kx − yk + ky − zk.


How would we prove the triangle inequality?

Squaring both sides, it is equivalent to


showing that, for all x, y, z ∈ Rk ,

kx − zk2 ≤ kx − yk2 + ky − zk2 + 2kx − ykky − zk,

that is,

k(x − y) + (y − z)k2 ≤ kx − yk2 + ky − zk2


+ 2kx − ykky − zk.
In terms of inner products, for each vector
a ∈ Rk ,

kak2 = ha, ai.

So

k(x − y) + (y − z)k2
= h(x − y) + (y − z), (x − y) + (y − z)i.
Using the properties of inner products,

h(x − y) + (y − z), (x − y) + (y − z)i


= hx − y, x − yi + hy − z, y − zi
+2hx − y, y − zi
= kx − yk2 + ky − zk2 + 2hx − y, y − zi.

To prove the triangle inequality, we need to


show this is at most

kx − yk2 + ky − zk2 + 2kx − ykky − zk.


In other words, we need, for any x, y, z ∈ Rk ,

hx − y, y − zi ≤ kx − ykky − zk,

which will true, provided, for any a, b ∈ Rk ,

ha, bi ≤ kakkbk.

So let us prove this fact.


For any real z, and any vectors a, b ∈ Rk ,

0 ≤ kz a − bk2 = hz a − b, z a − bi
= z 2ha, ai + hb, bi − 2zha, bi
= z 2kak2 + kbk2 − 2zha, bi.

The right hand side is a quadratic


expression in z, and, since the inequality is
true for all real z, the discriminant must be
non-positive:

[2ha, bi]2 − 4kak2kbk2 ≤ 0.


But

[2ha, bi]2 − 4kak2kbk2 ≤ 0

is equivalent to

ha, bi ≤ kakkbk,

so we have proven the inequality required


and hence the triangle inequality.
Definition [Bounded subset of Rm] B ⊆ Rm
is bounded if there is K > 0 such that for all
x, y ∈ B, kx − yk ≤ K.

Note that K is fixed: it does not depend on


x, y. (The definition would be meaningless if
that were the case.)
Example of a bounded set. Take m = 1,
and let B = [a, b], where 0 ≤ a ≤ b.

Then for all x, y ∈ B, kx − yk = |x − y| ≤ b − a,


so we can take K = b − a.
Example of an unbounded set. Take
m = 1, and let B = [0, ∞) = R+.

Then for all K, we can find x, y ∈ B, such


that kx − yk = |x − y| > K.

For example, take x = 0 and y = K + 1.


Note that, here, for all x, y ∈ B,
kx − yk = |x − y| is finite, and so |x − y| ≤ K,
for any number K ≥ |x − y|. But the point is
we cannot find a number K that would
work for all x, y simultaneously.
Theorem. A subset B of Rm is bounded if
and only if there is some M > 0 such that
kxk ≤ M for all x ∈ B.

This is an “if and only if” statement so


there are two things to prove here.
Suppose first there exists some M > 0 such
that kxk ≤ M for all x ∈ B.

We need to show that B is bounded, that


is, by definition, that there exists some
K > 0 such that, for all x, y ∈ B, kx − yk ≤ K.
Let 0 denote the vector with all components
equal to 0.

Then, by the triangle inequality,

kx − yk = k(x − 0) + (0 − y)k
≤ kx − 0k + k0 − yk
= kxk + kyk ≤ M + M
= 2M,

so we can take K = 2M .
Conversely, suppose that B is bounded. We
need to show that there is some M > 0 such
that kxk ≤ M for all x ∈ B.

We may just as well assume that B is


non-empty, otherwise the statement is
trivial. Let x0 ∈ B. Then kx0k ≥ 0.
Then for all x ∈ B, using the triangle
inequality and the definition of
boundedness,

kxk = k(x − x0) + x0k


≤ kx − x0k + kx0k
≤ K + kx0k,

so we may take M = K + kx0k > 0.


Convergence in Rm

A sequence (xn) in Rm is simply an ordered


list x1, x2, . . . of elements of Rm.
Definition [Convergence of sequences in
Rm] Suppose that (xn) is a sequence of
points of Rm. Then we say that the
sequence has limit x ∈ Rm if for every  > 0
there is N such that if n > N then
kxn − xk < .
Such a sequence is said to be convergent
and to converge towards x.

Equivalently, xn → x as n → ∞ if

kxn − xk → 0 as n → ∞.
Example. In R2, let xn = ( 2n+1 , 1 ). Let
n+1 n+1
x = (2, 0). Then
v
2n + 1 1
u
2
)2
u
kxn − xk = ( − 2) + (
u
t

v
n+1 n+1
(2n + 1 − 2(n + 1))2
u
u
u 1
= u
2
+
(n + 1)2
t
(n + 1)
v
u
u 1 1
= t
u
2
+
(n + 1) (n + 1)2

2
= →0 as n → ∞.
n+1
Theorem Suppose (xn) is a sequence in Rm
and let xn = (x1n, x2n, . . . , xmn). Then
xn → x = (x1, . . . , xm) if and only if, for i
between 1 and m, xin → xi as n → ∞.
Proof Suppose xn → x and let  > 0 be
given. Then there is N such that for n > N ,
kxn − xk < . But
v
u m
(xin − xi)2 ≥ |xin − xi|,
u X
kxn − xk = u
t
i=1

for any i between 1 and m, so for n > N ,


|xin − xi| <  and hence xin → xi.
On the other hand, if, for each i,
|xin − xi| < α, then
v
u m
(xin − xi)2
u X
kxn − xk = u
t

r
i=1
< mα2

= mα,

So if we let α = / m, we have:

|xin − xi| < / m, (i = 1, 2, . . . , m)

=⇒ kxn − xk < .

If xin → xi for each i, we may take / m in
the definition of limit (in place of ) to see
that there is some Ni such that for n > Ni,

|xin − xi| < / m.

Let N be the largest of N1, . . . , Nm. Then



for n > N , |xin − xi| < / m for all i and
hence kxn − xk < . This shows that xn → x.
n  
4n+2 
Example Suppose xn = 

n2
. Then, as

  n2−1
1/4 
n → ∞, xn → x = 
  . To see this, we
1
n 1
can simply observe that → and
4n + 2 4
n2
2
→ 1. Alternatively (though this is
n −1
more difficult), we could calculate kxn − xk
and check that kxn − xk → 0 as n → ∞.
Bolzano-Weierstrass theorem

Definition A sequence in Rm is bounded if


the set {xn : n ∈ N} is a bounded subset of
Rm .

Thus, a sequence (xn) is bounded if there is


some number M such that kxnk ≤ M for all
n.
Many of the results for sequences of real
numbers extend to sequences in Rm for
m > 1. For example, the
Bolzano-Weierstrass theorem has the
following generalisation.

Theorem [Bolzano-Weierstrass] Any


bounded sequence in Rm has a convergent
subsequence.
To prove the Bolzano-Weierstrass theorem,
we assume first m = 1, that is we are in R.

Let (xn) be a sequence of real numbers.


We will now show how to construct a
monotone subsequence (xnk ) of (xn).

Since (xn) is bounded, also (xnk ) is


bounded, above and below. Since (xnk ) is
monotone and bounded, it must converge.
Let us call a positive integer N a record if
n > N implies that xN ≥ xn, that is
xN = max{xn : n ≥ N }.

Suppose first that the sequence has


infinitely many records, n1 < n2 < n3 < . . ..
Note that, by definition of record,
xn1 ≥ xn2 ≥ xn3 ≥ . . .. In other words, the
sequence of records is decreasing, so in
particular, is monotone.
Now suppose that the sequence has only
finitely many records, and let N be the last
one. Take some n1 > N . Since n1 is not a
record, there exists n2 > n1, such that
xn1 < xn2 .

Repeating this process leads to an infinite


subsequence xn1 ≤ xn2 ≤ . . ., i.e. a
monotone increasing subsequence.
For general m, let (xn) be a sequence in Rm,
where xn = (x1n, . . . , xmn), and assume that
(xn) is bounded.

Then there exists K > 0 such that kxnk ≤ K


for all n.

Then also |x1n| ≤ K for all n, so (x1n) is a


bounded sequence of real numbers.
By the Bolzano-Weierstrass theorem in R,
there exists a subset A1 ⊆ N such that
(x1n : n ∈ A1) is a convergent subsequence
of (x1n), with limit L1, say.

But also |x2n| ≤ K for all n ∈ A1, so


(x2n : n ∈ A1) is a bounded sequence of real
numbers.
By the Bolzano-Weierstrass theorem in R,
there exists a subset A2 ⊆ A1 such that
(x2n : n ∈ A2) is a convergent subsequence
of (x2n), with limit L2, say.

Also, (x1n : n ∈ A2) is a subsequence of a


convergent subsequence (x1n : n ∈ A1), so
converges to limit L1.
We can now repeat this argument to
construct a sequence of subsets
Am ⊆ Am−1 ⊆ . . . ⊆ A2 ⊆ A1 such that, for
each i ∈ {1, 2, . . . , m}, (xin : n ∈ Ai)
converges to a limit Li, say.

It then follows that for each


i ∈ {1, 2, . . . , m}, (xin : n ∈ Am) converges to
a limit Li. Hence (xn : n ∈ Am) must
converge to (L1, L2, . . . , Lm).
Limits and continuity of functions
f :R→R

Definition[Limit of a function at a point]


Let f : R → R be a function. We say that L
is the limit of f (x) as x tends to a, denoted
by limx→a f (x) = L, if for each  > 0, there
exists δ > 0 such that

0 < |x − a| < δ =⇒ |f (x) − L| < .


The definition states that if someone gives
us any arbitrarily small , then there is some
neighbourhood of a, (a − δ, a + δ), such that
any x in this neighbourhood — other than a
itself — will have f (x) in the
-neighbourhood (L − , L + ) of L. (Note
that what happens to the function at a is
irrelevant.)
Theorem Let f, g : R → R be two functions
and c any real number. Suppose that
limx→a f (x) = L and limx→a g(x) = M . Then

• limx→a(cf )(x) = cL

• limx→a(|f |)(x) = |L|

• limx→a(f + g)(x) = L + M
Let us prove the third one of these. By
definition of limit, for all  > 0, there exists
δ1 such that

0 < |x − a| < δ1 =⇒ |f (x) − L| < /2.

And there exists δ2 > 0 such that

0 < |x − a| < δ2 =⇒ |g(x) − M | < /2.


Let δ = min{δ1, δ2}. Then, provided that
0 < |x − a| < δ, we have

|(f (x) + g(x)) − (L + M )| ≤ |f (x) − L|


+ |g(x) − M |
< /2 + /2 = .
• limx→a(f − g)(x) = L − M

• limx→a(f g)(x) = LM

• limx→a(f /g)(x) = L/M provided M 6= 0.


Let us prove the third one of these.

Since limx→a g(x) 6= 0, g(x) 6= 0 for each x


in some neighbourhood of a. That is, there
exists some δ0 > 0 such that, as long as
0 < |x − a| < δ0, g(x) 6= 0.
Now, for x such that 0 < |x − a| < δ0,
f (x) L |f (x)M − g(x)L|
| − | = | |
g(x) M |g(x)||M |
|f (x)M − LM |
≤ |
|g(x)||M |
|LM − g(x)L|
+
|g(x)||M |
|f (x) − L|

|g(x)|
|L||M − g(x)|
+ .
|g(x)||M |
We will be done if we can show that, given
 > 0, there exists δ > 0 such that, provided
0 < |x − a| < δ,
|f (x) − L|
< /2
|g(x)|
and
|L||M − g(x)|
< /2.
|g(x)||M |
To do this, note that we can choose δ0
such that inf x:0<|x−a|<δ0 |g(x)| ≥ |M |/2 > 0,
since limx→a g(x) = M .

Since f (x) → L as x → a, there exists δ1 > 0


such that

0 < |x − a| < δ1 =⇒ |f (x) − L| < |M |/4.


Since g(x) → M as x → a, there exists δ2 > 0
such that

0 < |x − a| < δ2 =⇒ |g(x) − M | < M 2/4|L|.

It is enough to take δ = min{δ0, δ1, δ2}.


(CHECK!)
Definition [One-sided limits] Let f : R → R
be a function. We say that L is the limit of
f (x) as x approaches a from the left,
denoted by limx→a− f (x) = L if for each
 > 0, there exists δ > 0 such that

0 < a − x < δ =⇒ |f (x) − L| < .

A similar definition applies to limits from


the right, denoted limx→a+ f (x) = L.
Continuity of functions f : R → R

Definition[Continuity at a point] A function


f : R → R is continuous at the point a if
limx→a f (x) exists and equals f (a).
Definition[Continuity on a set] Suppose
X ⊆ R. A function f : R → R is continuous
on X if for each a ∈ X, the limit of f (x), as
x → a and x ∈ X, exists and equals f (a).
Definition[Continuity on interval] A
function is continuous on the interval [a, b] if
it is continuous at each point in (a, b) and
(i) f (a) = limx→a+ f (x) and (ii)
f (b) = limx→b− f (x).
So, to say that f is continuous on [a, b]
means that f is continuous at each point in
(a, b), and that it is continuous on the left
at b and continuous on the right at a.

Definition[Continuity] A function is
continuous if it is continuous at each point
a.

A function is discontinuous at a if it is not


continuous there.
Theorem[Heredity results for continuity]
Let f, g : R → R be functions that are
continuous at a ∈ R and c be any real
number. Then |f |, (cf ), (f − g), (f + g), (f g)
are all continuous at a, and (f /g) is
continuous provided g(a) 6= 0.
As a corollary, any polynomial
p(x) = i=0 aixi is continuous.
Pk

Recall that if f, g are functions, then we


may define the composite function f (g(x)).
It turns out that if g is continuous at a, and
f is continuous at g(a), then the composite
function f (g(x)) is continuous at a.
To see this, take  > 0. Then, by continuity
of f at g(a) there exists δ1 > 0 such that

|y − g(a)| < δ1 =⇒ |f (y) − f (g(a))| < .

By continuity of g at a, there exists δ2 > 0


such that

|x − a| < δ2 =⇒ |g(x) − g(a)| < δ1.


Take δ = δ2. Then

|x − a| < δ =⇒ |g(x) − g(a)| < δ1


=⇒ |f (g(x)) − f (g(a))| < .
Limits of functions f : Rn → Rm

Suppose that f : R → R. As mentioned


above, we say that f (x) tends to L as x → a
if and only if given any  > 0, there is δ > 0
such that

0 < |x − a| < δ =⇒ |f (x) − L| < .


This appears to use the fact that one can
form the difference between any two real
numbers, However, as above, we can
interpret |x − a| as the distance between the
real numbers x and a, in which case the
above condition can be restated as

0 < distance(x, a) < δ

=⇒ distance(f (x), L) < .


It should be clear from this that the
condition doesn’t really use any algebraic
properties of R, only ‘distance’ properties.
This definition and many of its
consequences will remain if we have as
domain and codomain Rn and Rm for any
m, n ≥ 1.
Definition Suppose f : Rn → Rm, and that
a ∈ Rn and L ∈ Rm. We say that L is the
limit of f (x) as x tends to a if for each
 > 0, there exists δ > 0 such that

0 < kx − ak < δ =⇒ kf (x) − Lk < .

and we write limx→a f (x) = L.

(Note that we use the same notation, k.k,


for the lengths of vectors in both Rn and
Rm.)
The definition can be modified in the
obvious way if the function f maps from
some subset A of Rn: we simply add the
qualification that x ∈ A.
Example Suppose f : R2 \ {(0, 0)T } → R is
given by
 
x1  x1 x2
f 
  = r .
x2 x2
1 + x 2
2

Then f (x) → 0 as x → 0 = (0, 0)T .


To see this, we note that

|f (x) − 0| = |f (x)|
|x1||x2|
= r
x21 + x 2
2
(1/2)(x2 1 + x 2)
2
≤ r
x2
1 + x 2
2
1r 2
= x 1 + x2 2 → 0,
2
where we have used the fact that for any
real numbers a and b, 2ab ≤ a2 + b2. (This
follows from (a − b)2 ≥ 0.)
When looking at limits for functions from R
to R, we noticed that one can define left
and right limits and that these might be
different. A counterpart to the idea of left
and right limits for functions f : Rn → Rm
when n > 1 is the idea of the limit along a
path.
Example Suppose f : R2 \ {(0, 0)T } → R is
given by
x2 2
 
x
 1 − x
f   = 22 1
x2 x1 + x 22
Let’s consider what happens to f (x) as x
tends to 0 = (0, 0)T along the line x2 = αx1;
that is, through x of the form (t, αt)T . We
have
2t2 − t2 2−1
 
t α α
f
  =

2 2 2
= 2 .
αt t +α t α +1
So, f (x) approaches different values as
x → 0 along different lines. In particular,
f (x) does not have a limit as x → 0.
Example Let g : R2 \ {(0, 0)T } → R be

x1 x2
 
x
 1
g   = 2 2 4.
x2 x1 + x 2
If we again investigate what happens as
x → 0 along the lines x2 = αx1, we note
that, for all α,
2t3
 
t α
g
  =

2 4 4
→ 0 as t → 0.
αt t +α t
So, here, the limit as x tends to 0 along all
lines is the same.
However, we cannot deduce from this alone
that g(x) has a limit as x → 0. For, consider
what happens to g(x) as x → 0 along the
parabola given by x1 = αx2 2 (that is,
through points (αt2, t)). We have

αt 2 
4
αt α
g 
  =

2 4 4
= 2 ,
t α t +t α +1
so the limit depends on α, and so g(x) has
no limit as x → 0.
Continuity of functions f : Rn → Rm

Definition[Continuity of f : Rn → Rm]
Suppose that f : Rn → Rm and that a ∈ Rn.
Then, we say that f is continuous at a if
limx→a f (x) exists and equals f (a).
Equivalently, f is continuous at a if given
any  > 0 there exists δ > 0 such that if
kx − ak < δ then kf (x) − f (a)k < .
For a subset X of Rm, we say that f is
continuous on X if for all a ∈ X, the limit of
f (x), as x → a, with x ∈ X, exists and equals
f (a).

If f is continuous on the whole of Rm then


we simply say that f is continuous.
The following result is often useful.

Theorem Suppose that f : Rn → Rm and


that f1, f2, . . . , fm : Rn → R are such that for
all x ∈ Rn,

f (x) = (f1(x), f2(x), . . . , fm(x))T .

Then f is continuous at a ∈ Rn if and only if


f1, f2, . . . , fm are continuous at a.
To prove this, note that, for i = 1, . . . , m,

kf (x) − f (a)k ≥ |fi(x) − fi(a)|,

and, if |fi(x) − fi(a)| < α for each i, then



v
u m
kf (x) − f (a)k =
u X
u
t
2
(fi(x) − fi(a)) < α m.
i=1
Note that if e1, e2, . . . , em are the standard
basis vectors of Rm (so that ei has a 1 in
position i and all other entries equal to 0),
then the functions fi referred to are given by

fi(x) = hf (x), eii = eT


i f (x),

where ha, bi denotes the usual inner product


(scalar product) on Rm. We shall call the
functions fi the component functions of f .
All linear functions are continuous. Recall
that a linear function f : Rn → Rm is one
with the property that for all x, y ∈ Rn and
all α, β ∈ R,

f (αx + β y) = αf (x) + βf (y).


Any linear function can be represented in
matrix form: that is, there is some m × n
matrix M such that f (x) = M x for all x. In
this case, for 1 ≤ i ≤ m,

fi(x) = eT
i f ( x) = e T M x.
i

If we let mi denote M T ei, then we see that


T
fi(x) = eT Te x = mT

i M x = M i i x.
So, for any a ∈ Rn,

kfi(x) − fi(a)k = kmT


i x − mT ak ≤ km kkx − ak.
i i

As x → a, the right hand side tends to 0,


because kmTi k is just a fixed number. This
shows that fi is continuous, for each i, and
it follows that f is continuous.
Sequences and continuity

Theorem Suppose that f : Rn → Rm and


that a ∈ Rn. Then f is continuous at a if
and only if for every sequence (xn)
converging to a we have f (xn) → f (a).
Therefore f is continuous (on the whole of
Rn) if for every convergent sequence (xn) in
Rn, we have lim f (xn) = f (lim xn).
To prove this, assume first that the
function is continuous at a.

Take a sequence (xn) converging to a. We


want to prove that f (xn) → f (a) as n → ∞.

Take  > 0. We need to show that there


exists N such that, for all n > N ,

kf (xn) − f (a)k < .


By continuity of f at a, for all  > 0, there
exists δ > 0 such that, provided kx − ak < δ,

kf (x) − f (a)k < .

By convergence of xn to a, there exists N


such that, for all n > N ,

kxn − ak < δ.
Hence for all n > N ,

kf (xn) − f (a)k < ,

as required.
Now assume that, for every sequence
xn → a, we have f (xn) → f (a). We need to
prove that f is continuous at a.

We argue by contradiction: assume that f


is not continuous at a. Then there exists
 > 0 such that, for all δ > 0, there exists xδ
with kxδ − ak < δ but kf (xδ ) − f (a)k ≥ .
In particular, we can take δ = 1/n for each
n, and obtain xn such that kxn − ak < 1/n
but kf (xn) − f (a)k ≥ .

Then xn → a as n → ∞, but f (xn) 6→ f (a), a


contradiction.
Another worked example. The function
f : R2 \ {(0, 0)T } → R is given, for
x = (x, y)T , by

x2 y
f (x) = 6 2
.
x +y
Prove that limx→0 f (x) does not exist.
(Here 0 = (0, 0)T .)
Let xn = (1/n, α/n)T . Then xn → 0, and

αn −3
f (xn) = f ((1/n, α/n)T ) = −6
n + α2n−2
αn−1
= −4 2
→ 0, as n → ∞.
n +α

But, if xn = (1/n, 1/n3)T , then xn → 0, yet

n−5 n
3 T
f (xn) = f ((1/n, 1/n ) ) = −6 = → ∞,
n +n −6 2
as n → ∞.
So using different sequences tending to the
given point a can be useful in situations
where the limit of f (x) as x → a does not
exist.

This is a similar idea to the path approach


we took to a similar example earlier.
We have
αt 3
f ((t, αt)T ) = 6
t + α2t2
αt
= 4 → 0, as t → 0.
t + α2
But
t5 1
3 T
f ((t, t ) ) = 6 = → ∞, as t → 0.
t +t 6 2t
In particular, we see again it is not enough
to work along straight lines.
If f (x) → l as x → a, then f must tend to
the same limit as x tends to a along any
possible sequence (path).

And conversely, if f (x) tends to l as x → a


along all sequences (paths), then
limx→a f (x) = l.

To repeat, this observation is most useful in


detecting when the limit does not exist.

Вам также может понравиться