ARE211, Fall2013

ANALYSIS1: THU, AUG 29, 2013 PRINTED: AUGUST 29, 2013 (LEC# 1)

Contents 1. Analysis 1.1. 1.2. 1.3. 1.4. References Countable vs Uncountable innity Sequences Distance/Metrics 1 2 2 4 5

1. Analysis

Heavy emphasis on proofs in this section. Many students think that the proofs are the hardest part of Econ 201. Only way to master the art of proofs is to do a lot of them. The best topic in which to learn how to do proofs is analysis. A secondary goal in this topic is to get you to be comfortable jumping between dierent notations. When you read more formal journal articles, every author has his/her own notation system; need to learn how to jump back and forth between dierent notational systems. Youve just spent a few weeks in Math Camp on analysis. Why do more of it? Consensus is that a few weeks is too short a time to master the topic adequately: people talk about the re-hose approach to teaching math; too much too quick; one can get a mechanical understanding of what the material means, but its very hard in this short a time to develop intuitions for the concepts.

ANALYSIS1: THU, AUG 29, 2013

PRINTED: AUGUST 29, 2013

(LEC# 1)

The goal of the few lectures Ill spend on analysis is to consolidate what youve learned, help you understand the material more intuitively. Having said that, students in dierent years have dierent responses to math camp. Some classes nd it excruciating and totally inadequate preparation for graduate school; others nd that it gives them all that they need. I dont know a priori which category this classes fall into, so Ill want feedback on this after a couple of lectures.

1.1. References

Chapter 12 in Simon-Blume Chapter 1-2 in De La Fuente Appendix F in MasCollel-Whinston-Green Chapter 1 and 2: Elementary Classical Analysis, by J. Marsden

1.2. Countable vs Uncountable innity

This is a distinction thats fundamental in math but tricky to grasp at rst. Examples are easy to understand, but its a very dicult distinction formally. Its not much more than 100 years since it was formally proved that there really is a dierence between them. Three part distinction between sets: (1) nite sets: Example: {1, 2, ...N }. (2) countably innite sets. Example: the natural numbers, denoted N, are 1,2,3,4 ..., going on for ever.

ARE211, Fall2013

(3) uncountably innite sets. Example: the closed unit interval, denoted [0, 1]. Distinction between nite and innite sets: Denition: A set is innite if it can be placed in 1-1 correspondence with a strict subset of itself.

Example: {1, 2} is a strict subset of {1, 2, 3}. You cant map each element of the rst set to each element of the second. The even natural numbers, {2, 4, 6, ...} are a strict subset of the natural numbers N. You can map each element of the rst set to each element of the second. Distinction between countably and uncountably innite sets: Informal Denition: A set is countably innite if you can count its elements, i.e., you can identify the rst element, the second element, etc Example: Its easy to count the natural numbers, but you cant count the unit interval; for the closed unit interval, theres a rst element, but there isnt a second one. Mapping terminology for nite, countably and uncountably innite sets: Its useful to compare the following kinds of mappings. The only distinction between them is the domain of the mapping. The language Im going to use here is conventional but by no means universal.

(1) v : {1, ..., N } {1}, i.e., (1, ..., 1). This is more commonly refered to as an N -vector of ones, Distinguishing feature of a vector is that the domain of the mapping is nite. (2) x : N {1}. This is a sequence of ones, i.e., xn = 1, for all n. Distinguishing feature of a sequence is that the domain of the mapping is countably innite.

ANALYSIS1: THU, AUG 29, 2013

PRINTED: AUGUST 29, 2013

(LEC# 1)

(3) f : R+ {1}. This is a continuous function, mapping the non-negative real numbers to 1, i.e., f () = 1. Distinguishing feature of a function is that the domain of the mapping is uncountably innite.

Until youre taught to think otherwise, youd probably think of only the latter as a real function. Actually, all three mappings satisfy the true denition of a function, i.e., each of them assigns a unique point in the codomain to each point in its respective domain.

1.3. Sequences

A sequence is a mapping from the natural numbers to a set S , i.e., f : N S ; f (n) is the nth element of the sequence. Typically, we suppress the functional notation: instead of writing the image of n under f as f (n) we denote it by xn and write the sequence as {x1 , x2 , ..., xn , ...}, i.e., f (n) = xn . Ill emphasize repeatedly that what distinguishes a sequence from any other kind of mapping is the nature of its domain: it is a countably innite set. A collection {y1 , y2 , ..., yn , ...} is a subsequence of another sequence {x1 , x2 , ..., xn , ...} if there exists a strictly increasing mapping : N N such that for all n N, yn = x (n) . Note that maps the domain of the subsequence into the domain of the original sequence. For example, consider the two sequences {xn } = {3, 6, ..., 3n, ... } and {yn } = {6, 12, ..., 6n, ... }. Obviously {yn } is a subsequence of {xn }. To prove this formally, we need to come up with an appropriate function. The one we need is (n) = 2n, i.e., for all n, yn = x2n : e.g., y1 = x2 = 6, y2 = x4 = 12. In general, you construct a subsequence by discarding some elements of the original sequence, but keeping an innite number of the original elements and preserving their order. Some examples of sequences:

ARE211, Fall2013

(1) {1, 2, 3, 4... } (2) {1, 1/2, 4, 1/8... } (3) {1, 1, 1, 1... } (4) {1, 1/2, 1/3, 1/4... } (5) there are no restrictions on what can be the range of a sequence. In particular, sequences arent necessarily maps from N into scalars. For example, consider we could map N into 1 if x 1/n the set of continuous functions: {f1 , f2 , ...fn ...}, where fn = nx if 1/n < x < 1/n 1 if x 1/n

(6) sequences dont necessarily have a closed form representation, i.e., you cant necessarily

write down a formula that expresses the sequence. For example, if you started generating random numbers and continued forever, you would have a sequence.

Note that the dierence between a sequence and a set of comparable size is that the order of the elements in a sequence matters, while the order of the elements of a set does not. Thus, the sequences {1, 2, 3, 4... } and {2, 1, 3, 4... } are dierent, while the sets {1, 2, 3, 4... } and {2, 1, 3, 4... } are the same. Moreover, the set {1, 1, 1, 1... } is formally equivalent to the singleton set {1}, while the sequence {1, 1, 1, 1... } is a dierent object than the scalar 1.

1.4. Distance/Metrics

Analysis is all about how close things are to each other. Does a sequence converge to a point? In words, given a point x and a collection of points x1 , x2 , ..., what does it mean to say that this collection is close to the point? There are lots of notions of closeness in mathematics, some of them more intuitive than others. So long as we are considering closeness in the context of Euclidean space, most notions of closeness are essentially equivalent. However, once we get to consider closeness in

ANALYSIS1: THU, AUG 29, 2013

PRINTED: AUGUST 29, 2013

(LEC# 1)

the context of functions, there is a vast variety of quite dierent notions. Mathematicians have an abstract notion of what is a legitimate measure of closeness. Denition: a metric or distance function on a set S is a function d : S S R satisfying, for all x, y S :

(1) d(x, y ) = d(y, x) (symmetry) (2) d(x, y ) 0 (nonnegativity) (3) d(x, y ) = 0 i x = y (two elements are a positive distance apart i they are dierent from each other) (4) d(x, y ) d(x, z ) + d(z, y ), for all z S (the triangle inequality)

The last property of a metric is the one that has the most bite, and the one that really captures the spirit of distance: it states that the shortest distance between two points is a straight line. Examples of metrics

(1) on R: d1 (x, y ) = |x y |. (2) on Rn : d2 (x, y) =

yi )2

(3) on Rn : d (x, y) = max {|xi yi | : i = 1, ..., n}. 1 if x = y n discrete (4) on R : d (x, y) = (well call this the discrete metric). 0 if x = y Lets check that the function ddiscrete (x, y) = 1 0 if x = y is indeed a metric. It clearly satises if x = y

the rst three properties. What about the triangle inequality. First observe that if x = y, then ddiscrete (x, y) = 0. Since ddiscrete (x, z) + ddiscrete (z, y) is necessarily nonnegative, then the triangle inequality holds. Now suppose that x = y so that ddiscrete (x, y) = 1. In this case, for all z, either

ARE211, Fall2013

z = x or z = y in which case either ddiscrete (x, z) or ddiscrete (z, y) is 1 so the triangle inequality is again satised. An example of a function that is not a metric is e(x, y) = min{|xi yi | : i = 1, ..., n}. To see this, rst note that it fails the third condition, since e((1, 1), (1, 2)) = 0, but (1, 1) = (1, 2). Moreover, e also fails the last condition: set x = (1, 1), y = (2, 2), z = (1, 2), e((1, 1), (2, 2)) = 1 but e((1, 1), (1, 2)) = e((1, 2), (2, 2)) = 0 so that e(x, y) > e(x, z) + e(z, y). When S is a space of functions, condition (3) in the above denition of a metric is too restrictive. In economics, for example, we often encounter functions that arent equal to each other, but are said to be of distance zero from each other. In particular, it is often natural to say that the distance, d, between two functions is the integral of the absolute value of the dierence between them, i.e., d(f, g) = |f (x) g(x)|dx. But if f and g dier at only a nite (indeed countable) number of

points, then in this sense, the dierence between them will be zero. Condition (3) is inconsistent with this usage. To deal with this problem, we dene a function to be a pseudo-metric if it satises all of the conditions above except condition (3). E.g., if S is the set of integrable functions mapping R to R, then the function : S S R dened by (f, g) = | f (x)dx g(x)dx| is a pseudo-metric but not a metric. (Note that this distance notion is quite dierent from the d mentioned in the preceding paragraph!)

(1) To see that is not a metric, consider the function f1 dened above as the rst element of the sequence {fn } on p. 5 as example (5). Because the function is so symmetric, clearly f1 dx = other. f1 dx = 0, so that (f1 , f1 ) = 0, but these functions are not equal to each

ANALYSIS1: THU, AUG 29, 2013

PRINTED: AUGUST 29, 2013

(LEC# 1)

(2) On the other hand, to see that is a pseudo-metric, observe that its obviously symmetric and non-negative. The only thing remaining to check is that it satises the triangle inequality. To prove this, we could use the following Lemma, but wont go thru it in class

Lemma: for any x, y R, |x| + |y | |x + y |. Proof of the Lemma: Its obvious that if x and y both have the same sign then |x| + |y | = |x + y |. Now suppose without loss of generality (w.l.o.g.) that x 0 > y . In this case, |x| + |y | = x + (y ) > x > |x (y )| = |x + y |

We can now check that satises the triangle inequality. For any functions f, g, h S , (f, h) + (h, g) which from the lemma is = = f dx f dx (f, g) hdx + gdx hdx gdx = f dx hdx + hdx gdx