I noticed that a lot of students are very "limit-happy," i.e., many of you like to move limits around without justification. Starting with Homework 3, I will require that you justify any instances where you move a limit around. This is especially important if you want to swap lim with max, min, sup, or inf. (In fact, in many instances, that won't be legal.)
For example, here is a model solution:
Example 1.
Assume that limn→∞xn=x and limn→∞yn=y. Show that
n→∞lim(xn+yn)2=(x+y)2.
Solution.
n→∞lim(xn+yn)2=(n→∞lim(xn+yn))2=(x+y)2(t2 is continuous)(limit law)
Problem 2
While grading this one, I noticed that a lot of you guys wrote something like this:
n→∞lim∣∣xi(n)∣∣=0 for all i⟹n→∞lim1≤i≤kmax∣∣xi(n)∣∣=0
While it turns out this is true for this problem, it's a non-trivial fact, meaning you have to prove it. What's special in this problem is that i ranges over a finite set, i∈{1,…,k}. When you take a maximum or supremum over an infinite set, you can no longer switch lim and max in general:
Example 2.
Let
ai(n)={10if i=n,otherwise.
In other words, a(n) is a sequence which is all 0's except at the n-th term. For example,
n→∞lim∣∣xi(n)∣∣=0 for all i⟺imaxn→∞lim∣∣xi(n)∣∣=0,
but for each n, the maximum coordinate of x(n) is 1, so you also have
n→∞limimax∣∣xi(n)∣∣=n→∞lim1=1.
Moral: Pointwise convergence (i.e., convergence in each coordinate) does not imply uniform convergence (i.e., convergence of the maximum or supremum) in general.
We'll see more examples of this phenomenon later.
Cauchy Sequences and Completeness
If you remember the definition of a Cauchy sequence in R, then this definition should be no surprise:
Definition
Let (X,d) be a metric space. We say a sequence {xn}n is Cauchy if for all ε>0, there exists N such that d(xn,xm)<ε whenever n,m≥N.
Morally, Cauchy sequences "should" converge. In (R,∣⋅∣), a sequence was convergent if and only if it is Cauchy (this is the Bolzano-Weierstrass theorem from MATH 131A). However, this is not true for general metric spaces, which is why we have an additional definition:
Definition
Let (X,d) be a metric space. If every Cauchy sequence converges in X, then we say that X is complete.
Example 3.
R is complete with respect to the Euclidean metric..
(0,1) is not complete with respect to the Euclidean metric.
[0,1] is complete with respect to the Euclidean metric..
Example 4.
Show that (Rk,dℓ2) is complete.
Solution.
To show this, we need to show that any Cauchy sequence {x(n)}n in Rn converges. Each element in our sequence is a vector:
x(n)=(x1(n),…,xk(n))
As you can imagine, we're going to need to use completeness of R to do this.
Let ε>0. Since {x(n)}n is Cauchy, there exists an N such that if n,m≥N, then
if n,m≥N, i.e., the components {xj(n)}n is Cauchy in R. Thus, by completeness, there exists xj∈R such that limn→∞xj(n)=xj for each 1≤j≤k. The vector x=(x1,…,xk) will be our candidate for the limit of the sequence. We just need to prove that {x(n)}n actually converges to x with respect to dℓ2.
This is easy, though. The sum in dℓ2 is a finite sum, so we can apply our limit laws:
n→∞limdℓ2(x(n),x)=n→∞lim(i=1∑k∣∣xi(n)−xi∣∣2)1/2=(n→∞limi=1∑k∣∣xi(n)−xi∣∣2)1/2=(i=1∑kn→∞lim∣∣xi(n)−xi∣∣2)1/2=0(t1/2 is continuous)(limit law; sum is finite)(t2 is continuous; n→∞limxi(n)=xi)
Thus, {x(n)}n converges to x in dℓ2, so (Rk,dℓ2) is complete.
Example 5.
Show that (C([0,1]),d) is complete, where d is the sup-norm.
Solution.
As a reminder:
C([0,1])={f:[0,1]→R∣f is continuous}d(f,g)=x∈[0,1]sup∣f(x)−g(x)∣
Let {fn}n be a Cauchy sequence. Like in the previous example, we'll first want to find a candidate for the limit of this sequence. Notice that for a fixed x∈[0,1],
Like before, this estimate implies that {fn(x)}n is Cauchy for each x, so by completeness of R, there exists a number f(x)∈R such that limn→∞fn(x)=f(x).
There are two more things to show: (i) that {fn}n converges to f with respect to d, (ii) and that f is actually in C([0,1]), i.e., that f is continuous.
To show (i), let ε>0. Because {fn}n is Cauchy, there exists N such that for any x∈[0,1],
and this inequality is true for any n≥N. In other words, {fn}n converges to f with respect to d.
Finally, we need to show (ii). But this is immediate: convergence with respect to d is exactly uniform convergence. Thus, {fn}n is a sequence of continuous functions which converges uniformly to f, so f itself must be continuous (this is a fact from MATH 131A).
The second equality is not true in general: Consider fn(x)=xn on [0,1). Then
n→∞limx∈[0,1)supxn=1butx∈[0,1)supn→∞limxn=0.
This means that you cannot swap sup and lim without additional justification.
Homework Hints
Here's a useful proposition (that's listed as an exercise in Terry's book) for Problem 2, 3(a), and 5:
Proposition (1.2.5(h))
If E is any subset of X, then IntE is open, and given any other open set V⊆E, we have V⊆IntE. Similarly, E is closed, and given any other closed set K⊇E, we have K⊇E.
Proof.
There are a number of things to prove:
IntE is open:
Let x∈IntE. By definition, this means there exists r>0 such that B(x,r)⊆E. We'll show that B(x,r)⊆IntE also, which shows by definition that IntE is open.
B(x,r) is an open set, so given any y∈B(x,r), there exists r′>0 such that B(y,r′)⊆B(x,r)⊆E. In other words, y∈IntE, so B(x,r)⊆IntE.
IntE is the largest open set in E:
Assume V⊆E and that V is open. Given x∈V, this means there exists r>0 such that B(x,r)⊆V⊆E. Thus, x∈IntE by definition, so V⊆IntE.
E is closed:
This was proven in lecture.
E is the smallest closed set containing E:
Assume K⊇E and K is closed. Given x∈E and r>0,
∅=B(x,r)∩E⊆B(x,r)∩K.
Moreover, K is closed, so x∈K=K, which shows that E⊆K.
□
You can use this proposition freely as long as you cite it properly (i.e., write down the proposition number every time you use it).