ARE211, Fall2013

ANALYSIS2: TUE, SEP 3, 2013 PRINTED: SEPTEMBER 3, 2013 (LEC# 2)

Contents 1. Analysis (cont) 1.4. 1.4.1. 1.4.2. 1.5. Convergence of Sequences Boundedness of Sequences Boundedness of Sequences (cont) Least Upper bounds and Greatest Lower Bounds, Suprema and Inma 1 1 3 5 6

1. Analysis (cont) 1.4. Convergence of Sequences

Last lecture I introduced a number of metrics, the d1 , d2 and d metrics. An important property of these metrics is that they are all the same, in a sense that will be clearer in a minute. In this lecture, Ill refer to the Euclidean metric. In fact there are lots of Euclidean metrics, but they are all equivalent to each other. I emphasized in the previous lecture that the mathematical topic of analysis is all about what it means for things to be close to each other. For this reason, a core concept of analysis is the notion of convergence. Denition: We say that a sequence {xn } in X converges to an element x X in the metric d if > 0 N N such that n > N , d(xn , x) < .
1

ANALYSIS2: TUE, SEP 3, 2013

PRINTED: SEPTEMBER 3, 2013

(LEC# 2)

Denition: Two metrics d and on X are said to be equivalent if a sequence {xn } in X converges to an element x X in the metric d i {xn } in X converges to an element x X in the metric . As an exercise you should try to prove that for any r, q N, the two metrics on Rn dened by dr (x, y) = (
n i=1 (xi

yi )r )1/r and dq (x, y) = (

n i=1 (xi

yi )q )1/q are equivalent.

A sequence that converges to a point is called a convergent sequence. The point that it converges to is called a limit of the sequence. (Its also the limit of the sequence, i.e., theres only one, but thats a result not a denition.) Example: The sequence {xn } in R dened by xn = 1/n converges to zero in the Euclidean metric, but not in the discrete metric. Proof: We need to show that for all , there exists N such that for all n > N , d(xn , 0) < . Now in the Euclidean metric, d(xn , 0) = (xn 0)2 = |xn |, i.e., we just need to show that for n suciently

large, |xn | < . Pick1 N = 1/. Then 1/N < and for n > N , 1/n < 1/N . On the other hand, consider the discrete metric. To show that xn doesnt converge to zero, we need to show that there exists > 0 such that for all N , there exists n > N such that ddiscrete (xn , 0) > . Pick = 1/2. Let N be arbitrary and observe that ddiscrete (xN +1 , 0) = 1 > . Done. Note that what we did here was an example of a general proof technique: to prove that an assertion involving s and s is false, go through and change the s to and the s to . Then for all inequalities not linked to a or , ip the inequality. So for example, in the denition of convergence ( > 0 N N such that n > N , d(xn , x) < ), the only inequality that gets ipped is d(xn , x) < , which gets ipped to d(xn , x) .
1 Given R, is the smallest integer not less than . Similarly, is the largest integer not larger than .

ARE211, Fall2013

Theorem: A sequence {xn } converges to R in the discrete metric if and only if there exists N such that for all n > N , xn = .

1.4.1. Boundedness of Sequences. Denition: A sequence {xn } is bounded above by b R if xn b, for all n N. A sequence is bounded above if it is bounded above by some b R. Defn of bounded below is parallel. Similarly, a set X is bounded above by b R if x b, for all x X . Example: The sequence {xn } dened by xn = (1)n is bounded above by 1 and below by -1. (Its bounded above by lots of other things too, e.g, 2, , etc, and similarly bounded below by lots of other things. The point is that in order to be bounded above, a sequence just has to be bounded above by something). The Axiom of Completeness: Every non-increasing sequence {xn } in R that is bounded above below converges to a point x R in the Euclidean metric. (Not a theorem! Cant prove it. Have to assume it.) Intuition for these is straightforward: take the latter statement; you have an innite number of points that keep declining, or at least doesnt increase; but theres a oor to the decline: the points have to accumulate, because theres nowhere else for them to go. The Axiom is a very special property that depends both on the set R and the Euclidean metric. To see that it is violated for general sets, consider the set of rationals, denoted Q, which consists of all ratios of integers. If you replaced R by Q in the above statement, it would obviously be false: consider a non-increasing sequence of rational numbers which, when viewed as real numbers, converges to the non-rational number . (An example of such a sequence is given below.) Since its non-increasing, its bounded below by . But if you replaced the symbol R everywhere by Q
non-decreasing

ANALYSIS2: TUE, SEP 3, 2013

PRINTED: SEPTEMBER 3, 2013

(LEC# 2)

in the statement of the Axiom above, it would clearly be nonsense. Similarly, the Axiom would be equally nonsensical if the words Euclidean metric were replaced by discrete metric. The above highlights a special property of the real numbers, which is that they are complete, i.e., have no holes in them. (Again, this is an assumption, it is not a result that can be proved by invoking other properties of the real line.)
1) 1 1 1 1 Example: Consider the following sequence {xn }, where xn = (2 n1 , so that {xn } = {1, 3 , 5 , 7 , 9 , ...}.
n

Now dene the new sequence {Xn } by: Xn = 4

n k =1 xn .

The subsequence {Yn } obtained (I.e., Y1 = X1 = 4;

by taking the odd-numbered elements of {Xn }, is strictly decreasing.

Y2 = X3 = 4(1 1/3 + 1/5), Y3 = Y2 + 4(1/7 + 1/9). Also, note that the entire sequence lives in the set of rationals, denoted by Q, which consists of all ratios of integers. Now viewed as a sequence living in R, the sequence {Yn } converges (believe it or not) to . But viewed as a sequence living in Q, it is a non-convergent sequence, since its limitif it had onecan only be , but / Q. The three sequences are illustrated in Fig. 1. To see that the Axiom of Completeness is in fact a highly non-trivial assumption, consider the set of continuous functions C 0 = {f : [0, 1] R}. Intuitively, its fairly obvious what a non-decreasing sequence in C 0 would look like. For example, heres one: the sequence of continuous functions nx if 0 x < 1/n {f1 , f2 , ...fn ...}, where fn (x) = . 1 if 1/n x 1 While Im not going to dene metrics for functions (you need one for the domain and one for the codomain), its intuitive that for most such metrics (not all, though), the sequence {fn } will not converge to any continuous function. So the Axiom of Completeness would be violated if R in the specication above were replaced by C 0 .

ARE211, Fall2013

A nonincreasing NPR sequence of rationals that converges to an irrational

pi

x(n) = 1/(2*n1)*(1)n X(n) = sum(x(1:n)) x,X,Y Y(n)=X(2*n1)

10 n

12

14

16

18

20

Figure 1. Math courtesy of NPR 1.4.2. Boundedness of Sequences (cont). The following questions came up in class in past years. Question: Given a sequence {xn }, suppose that |xn+1 xn | converges to zero. Can we conclude that the sequence is bounded? Answer: No. Consider the sequence {xn } dened by xn =
n k =1 1/k .

Observe that dierence ben k =1 1/k }

tween xn+1 and xn is 1/(n + 1). However, the sequence {1/xn }, i.e., {

doesnt converge.

(This fact is isnt at all obvious (see Berck and Sydsaeter, 7.10 for verication that it is indeed true).) Question: Can a sequence converge to dierent things in dierent metrics?

ANALYSIS2: TUE, SEP 3, 2013

PRINTED: SEPTEMBER 3, 2013

(LEC# 2)

Answer: Not if the sequence belongs to a nite dimensional space, i.e., Rn , and if the metrics you are comparing are both Euclidean but more generally yes. For sequences in Rn , all Euclidean metrics are equivalent in the sense that if a sequence in Rn converges in two distinct metrics, then it converges to the same point in both metrics. In innite dimensional spaces, however, sequences can converge to dierent things in dierent metrics. (Well learn the precise meaning of innite dimensional in Linear algebra.) Consider the example of a sequence of continuous functions that 1 if x 1/n we examined in the last lecture, {f1 , f2 , ...fn ...}, where fn = nx if 1/n < x < 1/n . Depend 1 if x 1/n if x < 0 if x = 0 . if x > 0

ing on your metric, this sequence can converge to either 1 1 if x 0 1 if x < 0 = = f or g = or h 0 1 if x > 0 1 if x 0 1

1.5. Least Upper bounds and Greatest Lower Bounds, Suprema and Inma

Every nite set has a largest and a smallest element. However, this is not necessarily true of innite sets. E.g., the set {1 1 2 , ..., 1 1/n, ...} has no highest element. It nearly gets to 1 however. So, for bounded sets, we dene something called its least upper bound which is almost the maximum, the only possible candidate for maximum, if the set had one. In the above example the least upper bound of the set is 1. More generally, the least upper bound of the interval (x , x ) R is x . Denition: Let S R. A number bR is an upper bound for S if sb, for all sS . A number b is the least upper bound for S if b is an upper bound for S and if b b , for any upper bound b of S .

ARE211, Fall2013

Analogously, we have greatest lower bound of a set, which is almost the minimum, the point that would be the minimum if the set had a minimum. In the above example, {1 1 2 , ..., 1 1/n, ...}, the greatest lower bound of the set is 1/2, which is also the minimum of the set. More generally, the greatest lower bound of the interval (x , x ) R is x . Naturally if a set does indeed have a maximum then the least upper bound of the set is that maximum, e.g., the greatest lower bound and the least upper bound of the set [x , x ] are x and x . Denition: Let S R. A number b R is a lower bound for S if s b, for all s S . A number b is a greatest lower bound for S if b is a lower bound for S and if b b , for any lower bound b of S . If a set has a least upper bound, it is also called the supremum of the set. Similarly, if a set has a greatest lower bound, it is also called the inmum of the set. Some nonempty sets, however, dont have a least upper bound (e.g., R, N). In this case, we say that the supremum of the set is . Similarly, if a nonempty set doesnt have a greatest lower bound, we say that its inmum is . The empty set is special, an exception to many rules: its supremum dened to be and its inmum is dened to be . Notation: sup(S ) and inf(S ) denote, respectively, the supremum and inmum of the set S . A critical property of the real line R is: Theorem: Every nonempty set S R that has an upper bound has a least upper bound. Every nonempty set S R that has an lower bound has a greatest lower bound in R. Again, the above statement would not be true if R were replaced by Q. Consider the subsequence {Yn } dened above on 4. This set belongs to Q. As we saw, it converges to , but / Q. It has lots of upper bounds, indeed, any rational greater than is an upper bound for this set. But it

ANALYSIS2: TUE, SEP 3, 2013

PRINTED: SEPTEMBER 3, 2013

(LEC# 2)

doesnt have a least upper bound in Q: cant be an upper bound because its not in Q; and for any rational q greater than , theres another one thats between and q . Note the dierence between this theorem and the Axiom of Completeness. The latter was about sequences, this is about sets. It is, however, another way of saying that the real line doesnt have any holes in it. It is, clearly, very closely related to the Axiom of Completeness. In fact, we could have made this the axiom and then turned the Axiom of Completeness into a theorem.