Concept Math 2dn
Uploaded by
Ignacio CaamañoConcept Math 2dn
Uploaded by
Ignacio CaamañoB is an idempotent map, a splitting of e consists of an object A together with two maps A= B with rs = 1, and sr =e. It will turn out that in many categories, a device very similar to the one we used in sets will give a splitting for any idempotent endomap. In any case, there cannot be two essentially different splittings for e, as the following exercise shows. Exercise 3: : (In any category) Suppose that both ASB and A'=+~>B split the same idempotent B “+ B. Use these maps to construct an isomdrphism A 4’. One can show that this isomorphism f is the only one ‘compatible’ with the maps we started from, and one can even study how to reconstruct maps between retracts from maps between the large objects B; but this should be enough to give you the crucial idea: all the essential information about A, r, and s is really contained in B and e. Here is an example from arithmetic. Let B be the set of all ‘fraction symbols’ nid with 1 and d whole numbers and d # 0. Different fraction symbols, like 8/6 and 4/3 may represent the same rational number. In school you were taught a ‘reduction process’ B + B: cancel the greatest common factor in the numerator and denomi- nator, and then if the denominator is negative change the signs of both numerator and denominator. For example, e(6/—4) =—3/2. This map e is idempotent, because reducing a reduced fraction doesn’t change it. A rational number can now be described either as a reduced fraction (fixed point of e) or as the cluster of all fractions which reduce to that reduced fraction. In this example there is even a way to test whether two fraction symbols are in the same cluster without reducing them: e(n/d) = e(m/c) exactly when nc = md. This is convenient, because it is easier to multiply large numbers than to find their greatest common factor. Curiously, it is casier to find the greatest common factor of two numbers, by a process called the ‘Euclidean algorithm’, than it is to factor either of them into its prime factors! Recent unbreakable(?) codes depend on the apparent difficulty of factoring large numbers. 3. A puzzle If we think of B as a known set, incorporated into our ‘subjective’ category, what we have achieved is that the less-known set A has been captured by a description in our subjective category, namely by B and its idempotent endomap e. This seems puz- Retracts and idempotents 103 zling. Why would we want to describe the smaller set A in terms of the larger set B? Normally we wouldn’t, and it isn’t too often done in the category of finite abstract sets (but see the examples in the section below.) One exception occurs in program- ming computers, where the sort of set that is most easily managed consists of all strings of zeroes and ones of a particular length, say n. There are 2” of these, and you can see that it might be useful to represent any set A in which you are interested as a retract of a set B of n-strings, provided you can then record nicely the idempotent endomap of B. The study of how to do that gets one into ‘Boolean algebra,’ which is a basic topic in computer science. Still, the main use of ‘describing the smaller in terms of the larger’ occurs in other categories. It often happens that even though B is bigger, it is ‘structurally simpler’ than A. 4. Three kinds of retract problems Let’s return to the two general aspects of maps. We have seen that if a map B > A is given and we seek sections for it, a good way to picture the situation is to regard r as a ‘sorting of B into A sorts’: This enables us to picture a section s of r as choosing for each sort an ‘example’ of that sort. This might be called the ‘museum director’s problem’. Suppose you need to assemble an exhibit of mammals, with one mammal of each species. Then you start with the sorting map r from the set B of mammals B= mammals A= species of mammals Your job is to choose a section s of r; that involves selecting one exemplary specimen for each species. ‘The opposite, or dual, problem is the ‘bird-watcher’s problem.’ The bird-watcher starts with a manual giving an example of each species, a ‘sampling’ or ‘exemplifying’ map s: 104 Session 9 B= birds observed ir? Y A= species of birds His job is to assign to each bird he sees (real or pictured) a species, and his manual at least gives him sufficient direction to ensure that rs = ie To the young is given the most difficult problem. The small child sees a variety of animals, and with whatever assistance can be gathered from picture-books and parents, tries to select an idempotent endomap e: B= Animals The map e should assign to each animal the most familiar animal it closely resem- bles. Having selected e, the child is asked (again with some assistance) to split this idempotent: to form the abstract idea of ‘sorts of animals’ (e.g. cat, dog, cow) and to master the maps: B= Animals rs=1, A= Sorts of animals The map s assigns to each ‘sort’ of animal, say ‘cow’, the most familiar example, say Bossie; r assigns to each particular animal, say ‘the big scary barking thing next door’, its sort. These three kinds of problems were described in a way that made it seem that one solution could be preferable to another. In the rarified world of abstract sets, this would not be so; the sets in our examples have additional structure. Perhaps some abstract pictures may not be out of order, to illustrate that all three problems are solved by giving the large set B additional ‘structure’. Museum director’s problem: Given B —+ A, choose A —» B satisfying rs = 14. Mental picture: View r as sorting B into A sorts: Retracts and idempotents 105 Then choose a ‘cross-section’: Bird-watcher’s problem: Given 4A —+B, choose B—» A satisfying rs= 14. Mental picture: View s as a sampling of B by A: Then choose for each unidentified bird the most similar bird which is identified in the manual s: This constructs the idempotent e, clustering the birds around the sample birds, but then r is easy to find. While we're here, we should see whether we can calculate the number of solutions to the bird-watcher’s problem. Suppose there were a thousand birds and only three species, so the sampling map would look like this: How many retractions are there for this map? 106 Session 9 DANILO: The three birds that are sampled have to go back where they came from, but for the rest it is just any map to the three-element set of species; so there are 3-3) or 3%7 retractions for s. Good. You can see that Danilo's method finds the number of retractions for any map of sets that has any, ie. any map s that ‘preserves distinctness’: if x # y then sx # sy. Child’s problem: Given B, choose a map B— B satisfying ce =e. Having watched children for years, I remain as puzzled as ever about the selection of the idempotent endomap e associating to each animal the most familiar animal it resembles. After that’s done, though, the rest of the job (splitting the idempotent) is easy: The ‘sorts’ are all there in the mental picture, and all that is needed is to learn the names ‘cow,’ ‘cat,’ and ‘dog’ for these sorts. As a picture of the actual learning Process, this description is surely oversimplified, because the selection of the idem- potent map and the learning of the sort-names go on concurrently, 5. Comparing infinite sets There is one glaring omission from our account of ‘same size’ (isomorphism) and ‘at most as big as’ (retract). From our experience with finite sets, we would expect that if both 4
2". If C == D satisfy poqep = p, use the given maps p and q to devise a map q’ satisfying both: Peq'ep=p and qcpeq'=q' (and explain how you know that your q’ has these properties.) 1*. Same question as Problem | at top of page, except that both sets A and B are required to be infinite sets. 108 How to solve the quiz problems I have written down the thoughts that might reasonably go through your mind in trying to solve the problems, to show how you might arrive at a solution; and then how a solution might look. You will notice that the thought-process looks long, if you write it all down, and by comparison, a solution, after you finally find it, seems brief. Exactly how you use all this to help you learn how to solve problems is, of course, up to you. I suggest reading just a little at a time of the description of the possible thought process, and then returning to the problem to see if you are able to finish it without reading the rest. Afterwards, you can compare the way you arrived at a solution with the way this imaginary student did, and perhaps learn some new strategies to add to your techniques for thinking about problems. Problem 1 Let’s see ... I'm asked to pick an awful lot of stuff out of the air here. I have to pick the sets A and B, and the map f. My problem is that I know a lot of examples of sets, and I know a lot of examples of maps from one set to another — which should T choose? First, I had better decide how big I need to make these sets — the smaller I can make them, the better! What do I want? A + B is supposed to be chosen so there is a retraction for f — let me give that retraction for f a name — maybe ‘r’ would be a good letter, to remind me that it is supposed to be a retraction for f. Let me add r to the picture A 2+ B. Which way must r go? That's easy ~ any retraction or section for f goes backwards from f. So my ‘external diagram’ is going to look like this: [ase But of course I have to remember the definition of the phrase ‘r is a retraction for f.’ I memorized the definitions: ‘r is a retraction for f” means rf = 14. (I heeded the warning that was repeated so often in class that I had better learn by heart the difference between ‘g is a retraction for f° and ‘g is a section for f”, since they mean different things!) 109 110 How to solve the quiz problems Now I want to pick two sets A and B and two maps r and f, arranged as in the box above; but not just any two maps, they have to satisfy the equation rof=l, If I remember correctly, a retraction for a map tends to go from a bigger set to a smaller one. I ought to choose my sets so that Bis at least as big as A. In fact, I think it would be safer to choose it a bit bigger. Maybe if I try taking A with no members, and B with one member, it would work? f —— a r WOOPS! I can’t possibly have a map r that goes from B to A, because there is no member in A to be r(b)! Try again: maybe A with one member, B with two members 76) 5 I still must make up my maps r and f. There is only one map from B to 4 (I? = 1), there is no choice for r; it looks like: What about picking my map ASB Here there are two choices, I will just pick one, since they seem to look rather alike, anyway. Did this work? I was supposed to choose r and f so that rof = 1, How to solve the quiz problems HI That means I need, for each member of A, that r(f (that member)) = that member But there is only one member in A! What I need is r(f(a)) = it: r(f(a)) = r(b,) = a. Yes, it’s true! (Actually, now I realize that I didn’t even have to check it! There is only one map from A to A(1' = 1); any two maps, like ref and 14, from A to A have to be the same map.) Have I finished? Let me reread the problem .... Yes, I have done everything, except showing that there is no section for f. How do I do that? Well, a section for f would be a map B +> A satisfying f © s = 1g. (Good thing I learned the defini- tions!) Is there a map s that satisfies that equation? Well, there is only one map from Bto A, my s would have to be that one. And I have already named that ‘r’. I need to know: i, that’s all. Let’s check Is for = 1p or not? That would say F(r(b)) = 1 (*) and S(r(b2)) = br (#*) Are these true? f(r(b,)) = f(a) = by, («) is true; and f(r(b2)) = f(a) = by, (+*) is false! Therefore this map r is not a section for f; and it was the only map from B to A. Thus f has no section. Too bad! No, wait ... that’s what I wanted! f has a retraction, but it does not have any section. GOOD! Maybe, just to make it prettier, and to satisfy this fussy professor who asked me to make the sets and maps ‘explicit’, I will give the example with ‘concrete’ sets, but keep the picture I drew with Take A to be the set whose only member is lan; and take B = {Katie, Sheri} and A 2 Bo be given by f(Jan) = Katie. (— Here is how a good solution would look, written out: 112 How to solve the quiz problems I choose First: I claim f has a retraction r, a map B —> A satisfying rf = 1,. What is it? r is the only map from B to A! And since ref and 14 are both maps from A to A, and there is only one map from A to A, these two maps must be the same: ref = 14. Here is a picture of r, if you want: : fn) Second: I claim there is no section for f. A section for f would be a map B+ A satisfying f °s = 1. The Theorem on Uniqueness of Inverses said: If s is a section for f and r is a retraction for f then r = s. So the only possible sec- tion for f is r! And r is not a section for f, since f(r(Sheri)) = f (Ian) = Katie, shows that for # L;Katie shri): QED Note: After finding his original solution, this student found an alternative argument to show that this f has no section. Either argument would have been fine; but this one is maybe slightly better, because it employs a general principle: If you know that a map 4 + B has a retraction r, then the only possible section for f is r itself; if r is not a section for f, then f has no section! Problem 2(a) This one looks easier: I don’t have to invent everything myself. What I know is that p C==pD @ (read ‘p is a map from C to D and q is a map from D to C) and that Peqep=p What do I need to find out? I need to see that pq is idempotent; of course I need to know what it means to say that a map is ‘idempotent.’ Fortunately I learned that a map e is idempotent if it satisfies the equation ece=e How to solve the quiz problems 113 That seemed a bit peculiar to me at first, since usually if you have a map A + B, ‘fof doesn’t make any sense. You can follow one map by another, like f » g, only if the domains and codomains match up properly: Wyo The only case in which f © f makes sense is if the domain and codomain of f are the same set, like That is, f has to be ~ what was the word? — an endomap. Only endomaps have a chance to be idempotent; and even then, most endomaps are not idempotent. Just to be sure they're not trying to trick me, I had better check: is p » q even an endomap? Well, its domain is — let’s see, q was done first, so the domain of pq is the domain of q, which was D. And the codomain of pq is the codomain of p, which was ... yes, D. Since D -2°4, D is an endomap, at least it has a chance to be idempotent. Let’s write down exactly what it is that I want to show about p> q. I need to see that if you follow this complicated map by itself, you get it back again; i.e. I need to show: (peg)e(peg)2 peg What I know is: p> qop = p. My problem boils down to: KNOW: | pegep=p |* WANT TO SHOW: | (pog)o(pog)=pog |** That should be pretty easy — I have done problems like this before. Here is my solution: (p°q)° (peg) =peqgepeg (I can omit the parentheses) =(peqep)eq (I put parentheses back in to use #) =P°q (by *) Therefore, pq is idempotent. QED Now try Problem 2(b) yourself. Composition of opposed maps We should work through some examples of composition of maps of sets. While the algebra of composition is very simple, involving only the associative and identity laws, the understanding of how this algebra is applied is greatly aided by practice with concrete examples, first in the category of sets and later in richer categories. Let’s consider the following maps: mother Men ———= Women father One of them assigns to each man his mother, and the other assigns to each woman her father. What is the composite mothers father, or more briefly gof, where g = mother and f = father? For example, let’s ask Sheri: Who is gef > Sheri? First you have to decide who is f » Sheri. seri: My father is Mike. And who is g > Mike? seri: My father’s mother was Lee. Good, so gf ° Sheri = Lee. Is the map f gf equal to the map f? How do we test whether two maps of sets are equal? CHAD: When the same input gives the same output. So what about these two maps? ALYSIA: They are equal. Really? Let’s calculate both for the input 1“, women Who is f » Alysia? ALYSIA: Rocco. And who is go Rocco? ALYSIA: Dolores. And f ° Dolores? 114 Composition of opposed maps 115 ALYSIA: I don’t remember his first name, but his family name was R. All right, so fogef Alysia = Mr R. Is f° gf =f? Is Mr R. = Rocco? ALYSIa: No, fogof and f are different maps. Right. There is an input for which they give different outputs, so they are different. Notice that our test for equality of maps A == B of sets f=h] if and only if [Foa=hoa for every point 12rd is equivalent to the following: f*h| if'and only if | foratleastone1**A foathea Any such element a for which f is different from h is a counterexample proving the difference of f and h. So, Alysia, what is the counterexample proving that f' and fogof are different? ALYSIA: Me. Right. Because you are the member of the set of women for which we verified Segof> Alysia ¢ f > Alysia In fact for these two maps fogs fox £ fs x for every woman x, since otherwise we would have the biologically impossible situation that the father of x, y =/'ox, would satisfy y = f° gy; he would be his mother’s father. Often a composite map has a special name because of its importance. With g = mother and f = father, the composite gf, mother of father, is called ‘paternal grandmother.’ Notice how often it is possible to read the symbol ‘+’ as ‘of,’ instead of ‘following,’ This is true also of the symbol ‘x’ for multiplication of numbers. We usually read it as ‘times’, but for fractional factors, as in 3 x 6 = 4, we often say, ‘Two-thirds of six is four.’ Summary/quiz on pairs of ‘opposed’ maps = ee 0.) 4 = B Fill in the blanks; when [ys?] occurs, cross out the false alternative 1. Given two maps f, g with domains and codomains as above, we can (sometimes [vs] always) form the composites gf and fg. All we can say about gef and fg as maps in themselves is that they are : 2. If we know that g is a retraction for f, that means gf is actually ; then we can prove that f © g is not only an » but actually an . The latter means that the equation is true. » . If we even know that f is an isomorphism and that gof = 1,, then fg is not only an idempotent, but is . If, moreover, s is a map for which fos = Ip, we can conclude that s = 4. Going back to 0, i.e. assuming no equations, but only the domain and codomain statements about f and g, the composite f ° g°f (could be different from [ys?] must be the same as) f. Likewise f ogo f °g (could be different from [ys?] must be the same as) fg. 116 Summary If maps A 4+ ¥ £4 A satisfy (*) poj In any category The endomap ¥“* x (call it ‘a for short satisfies ao a we say «vis idempotent. Written out in full, this is Gcp)> Gop) =Gen). We will see more consequences later. On the equation po j = 1, 14, several consequences follow: | In the category of finite sets (1) p satisfies: for each member a of 4, there is at least one member x of X for which p(x) = a; (We say p is surjective.) (2) j satisfies: if j(a,) = j(a,), then a) = ay; (We say j is injective.) (3) HA < #X, and if #4 = 0, then #¥ =0 too! Problems involving the equation (*): (four types) Given X 2+ A, find all AL ¥ satisfying (*). Such a jis called a section for p. |r a | In finite sets, j is also called a ‘choice of representatives’ for p. Unless p is surjective, there will be no sections for p. More generally, the number of sections for pis T] #(p™a) (‘Chad's formula’). ipa Given A 2+ X, find all X £5 A satisfying (*). Such a p is called a retraction for j. In finite sets, unless j is injective, there will be no retractions for j. If is injective, the number of retractions for j is (#4)(#*-#4) (Danilo’s formula’). 4 x : | |r 4 X Given only X and A, find all p, J satisfying (*). If there is at 21) |P? least one such pair we say 4 is a retract of X (via p and j) 4 and sometimes write “A < YX’. Unless #4 < #X, there can be no such pairs p, i.e. A cannot be a retract of X. The formula for the number of pairs p, j in terms of #X and #£4 is rather complicated. XQq Given only an endomap X¥ + X find an A and j, p satisfying (*) | p? and j «p= a. Such a pair p, jis called a splitting for a. Unless 24 @is idempotent, there cannot be a splitting for a. In the category of finite sets, for each idempotent endomap a there is a splitting p, j. The number of elements of the desired turns out to be the number of fixed points of a (elements x of X satisfying a(x) = x). 117 Review of ‘I-words’ Identity map: For each object X there is an identity map ¥ + X. It satisfies 1yf =f and g1y = g whenever (the domains and codomains match, so that) the left side is defined. Inverse, isomorphism: ‘Inverse’ is the basic word, and involves two maps AB £ To say that ‘g is an inverse for f” means fg = 1p and gf = 14. If f has an inverse, it has only one, and we call that one f-!. If f has an inverse, we say that f is an isomorphism. Here is an analogy to get the grammar straight: MAPS g is the INVERSE for f. Not all maps have inverses, but a map can’t have two inverses. J is an ISOMORPHISM Meaning: there is some g which is an inverse for f, in fact exactly one. 7" (the inverse of f ) It is forbidden to use this as a name of a map, unless f has an inverse. a More precisely, if f has an inverse, then f~ also has an inverse, namely f. Idempotent, PEOPLE Ginger is the SPOUSE of Fred. Not all people have spouses, but you are not allowed to have two spouses Fred is MARRIED Meaning: there is some person who is a spouse for F., in fact exactly one. Fred's spouse It is forbidden to use this to specify a person, unless Fred has a spouse. The spouse of the spouse of F. is F. More precisely, if F. has a spouse, then that spouse also has a spouse, namely F. involution: (ee Article 111) Both are properties that only an endomap A+ A an have, since they involve f of. If fof =f, we say f is (an) idempotent. Iffef = 14, we say f is an involution. Remarks: The only idempotent which has an inverse is an identity map. Every involution has an inverse, namely itself. 118 Test 1 1. Throughout this problem (a) Find an invertible map A.A, different from the identity map 1. (b) Find an idempotent map A—*—A, different from the identity map J4. (Draw the ‘special internal diagrams’ of your maps f and e — the diagrams that are available only for endomaps.) (©) Find another set B and two maps s BoA for which ros = Ip and ser =e. (Draw the internal diagrams of r and s. In this part, ¢ is still the map you chose in part (b),) 2. Riis the set of all real numbers, and R—4—R is the map given by the explicit formula f(x) =4x—7 for each input x. show that f has an inverse map. To do this, give an explicit formula for the inverse map g, and then show that (a) (gef)(x) =x for each real number x, and that (b) (f°g)(x) =x for each real number x. 119 SESSION 10 Brouwer’s theorems 1. Balls, spheres, fixed points, and retractions The Dutch mathematician L.E.J. Brouwer (1881-1966) proved some remarkable theorems about ‘continuous’ maps between familiar objects: circle, disk, solid ball, etc. The setting for these was the ‘category of topological spaces and continuous maps.’ For our purposes it is unnecessary to have any precise description of this category; we will instead eventually list certain facts which we will call ‘axioms’ and deduce conclusions from these axioms. Naturally, the axioms will not be selected at random, but will reflect our experience with ‘cohesive sets’ (sets in which it makes sense to speak of closeness of points) and ‘continuous maps.’ (Roughly, a map f is continuous if f(p) doesn’t instantaneously jump from one position to a far away position as we gradually move p. We met this concept in discussing Galileo's idea of acontinuous motion of a particle, i.e. a continuous map from an interval of time into space.) There is even an advantage in not specifying our category precisely: our reasoning will apply to any category in which the axioms are true, and there are, in fact, many such categories (‘topological spaces’, ‘smooth spaces’, etc.). We begin by stating Brouwer’s theorems and by trying to see whether our intuition about continuous maps makes them seem plausible. First we describe the Brouwer fixed point theorems. (1) Let Ibe a line segment, including its endpoints (I for Interval) and suppose that f: 1 —+1is.a continuous endomap. Then this map must have a fixed point: a point x in I for which f(x) = x. Example: Suppose that J is an interval of time, and that R is an interval of road, say the highway from Buffalo to Rochester. Suppose that two cars drive on this road. The first car drives at a constant speed from Buffalo to Rochester, so its motion is described by J “+ R (u for ‘uniform’ motion). Meanwhile, the second car starts anywhere along the road and just travels aimlessly along, perhaps occasionally parking for a while, then retracing its path for a while, and ending its Journey at any point along the road. Let’s denote the motion of this second car by 1s R. Now u is an invertible map, so we get RJ, and let I 4+ J be the composite f = u-! om. Brouwer’s theorem tells us that there must be some time ¢ in J at which f(t) = 4; that is, wu"! m=, so mt = ut, which says there is some 120 Brouwer’s theorems 121 time ¢ at which the two cars are at the same point on the road. This seems not very surprising; if the first car drives from Buffalo to Rochester and the second car is always on the road, then of course the first car must at some time meet the second. The next theorem is similar, but about a disk instead of an interval, and I find it much less obvious. (2) Let D be a closed disk (the plane figure consisting of all the points inside or on a circle), and f a continuous endomap of D. Then f has a fixed point. Example: Rotating the disk by a certain angle gives a continuous endomap of the disk; f could be the process ‘turn 90 degrees.’ ALYSIA: What about the center? Exactly! That is a fixed point. For this map it was easy to see that it has a fixed point, but for other maps it may not be so easy; yet the theorem says that as long as f is continuous, it will have at least one fixed point. This theorem seems to me much more surprising than the previous one. Example: Suppose my disk is a portion of the Washington DC area, say the part inside or on the circular beltway. I also bring a map of the region, drawn on a piece of paper P. My map is thus a continuous map D -"+ P. If I am so callous as to crumple up the map and throw it out of the car window, so that it lands inside the beltway, I get an additional continuous map P—+D (p for projection’), assigning to each point on the crumpled paper the point on the ground directly under it. Brouwer’s theorem, applied to the map f =pom: D— D, tells me that some point x inside the beltway is directly under the point m(x) that represents x on the map. Do you find that surprising? I did when I first heard it. You can try the experiment, but please pick up the map afterward. If it occurred to you that a perfect map would show every detail of the area, even including a picture of the discarded map, congratulate yourself. You have discovered the idea behind Banach’s fixed point theorem for ‘contraction’ maps. You only have to goa step further: the discarded map has a small picture of the discarded map, and 122 Session 10 that picture has a smaller picture which has a smaller picture .... These pictures gradually close in on the one and only fixed point for our endomap. This beautifully simple idea only works for an endomap which shrinks distances, though. Brouwer’s theorem applies to every continuous endomap of the disk. Example: Here is a map to which Brouwer’s theorem applies and Banach’s doesn’t. Suppose D is a disk-shaped room in a doll’s house, and F is a larger- than-life floor plan of that room; we crumple F and discard it on D as before. The composite map D + F "+ D this time will not shrink all distances, so the Banach idea doesn’t apply. (In fact pom may have many fixed points, but they are not so easy to locate. It often happens that if a problem has only one solution, it’s easy to find it; but if there are many solutions, it’s hard to find even one of them.) The next theorem is about .... Any guess? FATIMA: A ball? Exactly! A solid ball. It says the following: (3) Any continuous endomap of a solid ball has a fixed point. To imagine an endomap, think of deforming the ball in any arbitrary way, but without tearing it. DANILO: Something like folding dough? Yes, but without breaking it into separate pieces. I find it easier to imagine this endomap of the ball if I first have two ‘objects,’ a wad W of dough and a ball-shaped region B in space. Then I can use two maps from W to B: a ‘uniform’ placement u: W—+B in which the wad W exactly fills the region B, and the new placement after kneading the dough, p : W— B. Now w is invertible, and the endomap we want is pu. It assigns to each point in the region the new location of the point in the dough that was originally there; it’s a sort of ‘change of address’ map. Now we describe the sequence of theorems known as Brouwer retraction theorems. (1) Consider the inclusion map j : E—-I of the two-point set E as boundary of the interval I. There is no continuous map which is a retraction for j. Recall that this means there is no continuous map r : / — E such that roj = 1g. E . eh r? No In other words, it is not possible to map the interval continuously to its two endpoints and leave the endpoints in place. Isn’t this reasonable? Isn’t it pretty Brouwer’s theorems 123 obvious that one cannot put one part of the interval on one of its endpoints and another part of it on the other without tearing it? The next retraction theorem is about the disk and its boundary. (I) Consider the inclusion map j : C— D of the circle C as boundary of the disk D into the disk. There is no continuous map which is a retraction for j. Circle Disk Again, this should seem quite reasonable. Suppose we have a drum made of a very flexible stretchable sheet. To get a retraction for the inclusion of the boundary we might imagine taking the sheet and squeezing it into the rim but without moving its boundary. One would think that this is not possible without puncturing or tearing the sheet. What this retraction theorem says is that this thought is correct. The third retraction theorem is, as you can imagine, about the ball and its bound- ary (the sphere). (III) Consider the inclusion j :S—+B of the sphere S as boundary of the ball B into the ball. There is no continuous map which is a retraction for j. Now, here is the point about all these theorems: (1) and (1) are actually equivalent theorems, and so are the Theorems (2) and (ID), and also Theorems (3) and (III). In other words, after proving the retraction theorems, which seem so reasonable, Brouwer could easily get as a consequence the fixed point theorems (which seem much less intuitive). We shall illustrate this by showing how Brouwer proved that (II) implies (2), and we'll leave the other cases for you to think about. Let's write clearly what Brouwer promised to show: Uf there is no continuous retraction of the disk to its boundary then every con- tinuous map from the disk to itself has a fixed point. However, Brouwer did not prove this directly. Instead of this he proved the follow- ing: Given a continuous endomap of the disk with no fixed points, one can construct a continuous retraction of the disk to its boundary. 124 Session 10 This is an example of the contrapositive form of a logical statement. The contra- positive form of “A implies B’ is ‘not B implies not A,’ which conveys exactly the same information as ‘A implies B,’ just expressed in a different way. Below is an example of how it is used. 2. Digression on the contrapositive rule A friend of mine, Meeghan, has many uncles. All of Meeghan’s uncles are doctors. In Meeghan’s world (implies) uncle doctor (PARTICULAR SITUATION) I went to her wedding and met some of them. There I had an interesting discussion with an intelligent man who I thought was another uncle, but in the course of the conversation he said that he was a mechanic. So I thought mechanic 2, not doctor (GENERAL KNOWLEDGE about our society) not doctor “2, not Meeghan's uncle (CONTRAPOSITIVE of what is known of the particular situation) Therefore this man is not one of Meeghan’s uncles. M his statement | ae not D u fact about ———, — ow ao SSE | Mecghan’s family not U D 3. Brouwer’s proof We return to Brouwer’s theorems. To prove that the non-existence of a retraction implies that every continuous endomap has a fixed point, all we need to do is to assume that there is a continuous endomap of the disk which does not have any fixed point, and to build from it a continuous retraction for the inclusion of the circle into the disk. Brouwer's theorems 125 So, let j : C —+ D be the inclusion map of the circle into the disk as its boundary, and let’s assume that we have an endomap of the disk, f: D — D, which does not have any fixed point. This means that for every point x in the disk D, f(x) 4 x. From this we are going to build a retraction for j, i.e. a map r: D— C such that roj is the identity on the circle. The key to the construction is the assumed property of f, namely that for every point x in the disk, fix) is different from x. Draw an arrow with its tail at f(x) and its head at x. This arrow will ‘point to’ some point r(x) on the boundary, When x was already a point on the boundary, r(x) is x itself, so that r is a retraction for j, ie. rj = tc. Two things are worth noting: first, that sometimes something that looks impos- sible or hard to prove may be easily deduced from something that looks much more reasonable and is, in fact, easier to prove; and second, that to know that a map has no retraction often has very powerful consequences. The reasoning leading to the proof of Brouwer’s fixed point theorem can be summarized in the following diagram: proof JS without fixed point. ——————————} retraction on the boundary contrapositive principle no retraction b _nofwithout fixed point (reasonable and true) (surprising but now known to be true) DANILO: Your conclusion sounds peculiar. Instead of ‘every f has a fixed point,’ you get ‘there is no f without fixed point.’ You are right. We need to use another principle of logic, that nor(not A) implies 4, to reach ‘every f has a fixed point.’ Brouwer himself seriously questioned this rule of logic; and we will later see that there are examples of useful categories in whose ‘internal’ logic this rule does not hold. (This ‘logical’ difficulty turns out to be connected with the difficulty of actually locating a fixed point for f, if f is not a ‘contraction map’.) 126 Session 10 4. Relation between fixed point and retraction theorems Exercise 1: Let j: C—+D be, as before, the inclusion of the circle into the disk. Suppose that we have two continuous maps D ——: D, and that g satisfies goj =j. Use the retraction theorem to show that there must be a point x in the disk at which f(x) = g(x). (Hint: The fixed point theorem is the special case g = Ip, so try to generalize the argument we used in that special case.) I mentioned earlier that each retraction theorem is equivalent to a fixed point theorem. That means that not only can we deduce the fixed point theorem from the retraction theorem, as we did, but we can also deduce the retraction theorem from the fixed point theorem. This is easier, and doesn’t require a clever geometrical construction. Here is how it goes. Exercise 2: Suppose that A is a ‘retract’ of X, ie. there are maps A-—>X with ros = Ly. Suppose also that X has the fixed point property for maps from T, ie. for every endomap X 4+ X, there is a map T —+ X for which fx = x. Show that A also has the fixed point property for maps from 7. (Hint: The proof should work in any category, so it should only use the algebra of composition of maps.) Now you can apply Exercise 2 to the cases: T is 1 (any one-point space), X is the interval, the disk, or the ball, and A is its boundary (two points, circle, or sphere.) Notice that in each of these cases, there is an obvious ‘antipodal’ endomap a of 4, sending each point to the diametrically opposite point; and a has no fixed point. Exercise 3: Use the result of the preceding exercise, and the fact that the antipodal map has no fixed point, to deduce each retraction theorem from the corresponding fixed point theorem. In solving these exercises, you will notice that you have done more than was required. For example, from the fixed point theorem for the disk, you will have concluded not only that the inclusion map C— D has no retraction, but also Brouwer’s theorems 127 that C is not a retract of D (by any pair of maps.) In fact, the argument even shows that none of E, C, S, is a retract of any of 1, D, B. You will probably have noticed that the same reasoning is used in all dimensions; for instance, Exercise | applies to the interval or ball as well as the disk. In the next section we state things for the ‘ball’ case, but draw the pictures for the ‘disk’ case. 5. How to understand a proof: The objectification and ‘mapification’ of concepts You may have felt that none of our reasoning about Brouwer’s theorems was valid, since we still have no precise notion of ‘continuous map.’ What we wish to do next is to extract those properties which are needed for our reasoning, and see that our conclusions are valid in any category in which these properties (which we will call Axiom | and Axiom 2) hold. Brouwer introduces in his proof, besides the sphere S and ball B and the inclusion map S > B, several new concepts: 1. arrows in B: 2. each arrow has a head, in B: this point is the head of the arrow 3. each arrow in B points to a point in S: this is where the arrow points To analyze his proof, then, we must bring these concepts into our category @. This means that we will need: 1. an object A (whose points are the arrows in B); 2. a map A 4. B (assigning to each arrow its head); and 3. a map A. S (telling where each arrow points). (Remember that a map in @ means a ‘continuous’ map, so that any map obtained by composing maps in @ will automatically be continuous.) 128 Session 10 Now we have three objects and three maps: A Ly | Jer neat i and we can begin to ask: what special properties of these (now ‘objectified’) concepts are used in Brouwer’s proof? First, we observe that if an arrow has its head on the boundary, then its head is the place to which it points: this is both paand ha We will bring this into our category, by noting that a map T -*+ 4 is a (smooth) ‘listing’ of arrows: T > A. Axiom 1: If T is any object in @, and T + A and T ++ S are maps satisfying ha=js, then pa=s. The diagram below shows all the maps involved. Te 17 SB 7 (Instead of just one arrow, we imagine a ‘parameterized family’ of arrows, one for each point in a ‘parameter space’ or ‘test object’ T; ie. a map T —+ A. The rest of the translation process leading to Axiom 1 just requires taking care to notice that p of an arrow is in S, while h of an arrow is in B; so to compare them we need to use the inclusion map S > B.) Already from Axiom 1, we can carry out part of Brouwer’s argument: Theorem 1: If B “+ A satisfies hoj = j, then pa is a retraction for j. Proof: Put T = S,s= 1s, and a = qj in Axiom 1. Corollary: If he = 1p, then pe is a retraction for j. Second, we notice that if two points of B are different, there is an arrow from the first to the second; in fact each arrow in A should be thought of as having its head Brouwer’s theorems 129 and tail distinct, otherwise it wouldn’t ‘point to’ a definite place on the boundary S. ‘We use the method of test objects again, with the idea that for each t, at is the arrow from ft to gt. Axiom 2: If T is any object in @, and T B are any maps, then either there is a point 1 “+ T with ft = gt, or there is a map T -“+ A with ha = g. Now we can finish his argument: Theorem 2: Suppose we have maps and gj = j, then either there is a point 1 + B with fo = gh, or there is a retraction for SB. Proof: Take T= in Axiom 2. We get: either there is a point 1—"+ B with fo=gb, or there is a map B—+A with ha=g; but then haj=gj=j, so Theorem 1 says that pa is a retraction for j. If we take g = J in Theorem 2, we get a corollary. Corollary: if B 4+ B, then either there is a fixed point for f or there is a retraction for ss. (We gave, in Theorem 2, the more general version of Brouwer’s theorem; the cor- ollary is the original version.) We will see later that in many categories @, an object T may be large, and still have no ‘points’ 1-4 T. In such a category, we should notice that we really didn’t use the full strength of Axioms 1 and 2 in our proofs. It was enough to have Axiom 1 just for T = S, and Axiom 2 for T = B. The main thing to study, though, is the way in which by objectifying certain concepts as maps in a category, the combining of concepts becomes composition of maps! Then we can condense a complicated argument into simple calculations using the associative law. Several hundred years ago, Hooke, Leibniz, and other great scientists foresaw the possibility of a ‘philosophical algebra’ which would have such features. This section has been quite condensed, and it may take effort to master it. You will need to go back to our previous discussion of Brouwer’s proof, and carefully compare it with this version. Such a study will be helpful because this example is a model for the method of ‘thinking categorically.” 130 Session 10 6. The eye of the storm Imagine a fluid (liquid or gas) moving in a spherical container. (If you want a two- dimensional example, you can imagine water swirling in a teacup and observe the surface current, say by imagining tiny boats drifting.) Right now, each point in our ball is moving, and we draw an arrow with tail at that point to represent its velocity. That is, the length of the arrow is proportional to the speed of the point, and the arrow points in the direction of travel. Could it be that every point is moving with non-zero speed, or must there be at least one instantaneous ‘eye of the storm?” To answer this, we take a slightly different arrow-object A than we imagined before. Its points are to be the possible velocity arrows of particles moving in our ball with non-zero speed. These arrows are less constrained than in our previous arrow-object, since the head of the arrow may be outside the ball; the only restriction is that if a point is on the surface of the ball, its velocity arrow cannot ‘point out- ward’ — at worst it is tangent to the sphere. Here is a picture for dimension 2: a The arrows a; and a, are allowed as points in A, but a3 is forbidden. Now we'll suppose that every point is moving, so we get a map B -*> A, assigning to each point of B the ‘velocity arrow’ at that point. For the map A ~~» B, we take the map assigning to each arrow its ‘home.’ (Remember that an arrow is supposed to repre- sent the velocity of a moving point, so the tail of the arrow is the current home of the point.) Finally, for the map A + S, we assign to each arrow its imaginary ‘place of birth.’ (It is customary to name winds in this way, as if a wind arriving from the north had always blown in one direction, and came from the farthest point that it could.) a P(a) Axiom 1 says that if the moving point is on the sphere, then its ‘place of birth’ is its current location: Brouwer’s theorems 131 ha) = pa) That is, the dot in the picture above is both A(a) (as a point in the ball) and p(a) (as a point on the sphere). Now you can work out for yourself that the corollary to Theorem | tells us that if there were a storm with no instantaneous ‘eye,’ there would be a retraction for the inclusion of the sphere into the ball. 7. Using maps to formulate guesses Let's return to the one-dimensional case, the two cars traveling on the highway. ae Buffalo Rochester Actually, the highway extends beyond these two cities. Suppose I drive along the road, starting in Buffalo and ending in Rochester; you start and finish at the same times, starting and finishing anywhere between Buffalo and Rochester. During our travels, we're allowed to go anywhere along the highway we want, even west of Buffalo or east of Rochester. Are you convinced that at some time we must meet? Why? Notice that there are now three objects involved: /, an interval; E, its endpoints; and R, the long road. (You can imagine R as the whole line if you want.) We also have two ‘inclusion maps’: HL LA Buffalo Rochester R My travel gives an additional map: J + R, and your travel gives another: I 2+ R. The relations among these four maps are investigated in the exercises below. 132 Session 10 Exercise 4: (a) Express the restrictions given above on my travel and yours by equations involving composition of maps, introducing other objects and maps as needed. (b) Formulate the conclusion that at some time we meet, in terms of composition of maps. (You will need to introduce the object 1.) (c) Guess a stronger version of Brouwer’s fixed point theorem in two dimensions, by replacing E, J, and R by the circle, disk, and plane. (You can do it in three dimensions too, if you want.) (d) Try to test your guess in (c); ¢.g. try to invent maps for which your conjec- tured theorem is not true. PART III Categories of structured sets We use maps to express extra ‘structure’ on sets, leading to graphs, dynamical systems, and other examples of ‘types of structure.’ We then investigate ‘structure-preserving’ maps. ARTICLE III Examples of categories Directed graphs and other structures We recall from Session 10: 1. Given an endomap of the ball with no fixed point, we can construct a retraction of the ball to its boundary. 2. Brouwer proved that no such retraction is possible. We deduced by pure logic: 3. Every endomap of a ball has a fixed point. We saw further that: 4, The sphere and the ball cannot be isomorphic (since the sphere does have a fixed point free endomap, for example, its antipodal map.) It is critical that the category which we were discussing is not the category of abstract sets and arbitrary functions; it must rather be some category of ‘cohesive’ objects and ‘continuous’ maps Precisely which category of this type does not matter for our purposes; it only matters that spheres and balls are related by certain maps with certain properties, which we specified in detail in the case of (1). Crucially, (2) does not hold in the category of sets and functions on which most of our earlier discussions centered. Nor does (3) or (4): for example, the sphere, ball, and circle are all isomorphic in the category of sets! There are, in fact, many categories having the needed properties: the category of topological spaces and continuous maps the category of smooth spaces and smooth maps, etc. These categories differ substantially from each other, and to specify these categories precisely and discuss their differences is a task better left to more advanced works. We propose instead to define and study certain simpler categories which are of great interest in their own right, and which exhibit many of the features of cohesive categories. All our examples will illustrate a basic method: to make precise some imprecisely-known category, we can try to model it by structures in the category of abstract sets. These structures are always expressed by some configuration of given maps. When these categories arise later, we will treat them more slowly and in more detail; our main aim here is to give a rapid preview of some possible notions of 135 136 Article IIT structure, and especially to provide a first introduction to the powerful idea of Structure-preserving map. The exercises in this article, like our previous exercises, involve only the application of the associative law to given definitions. 1. The category S© of endomaps of sets An important example already implicitly alluded to is the category in which an object is a set equipped with a specified endomap. Before defining it, let us denote by S the category of sets and maps which we have been discussing up to now. The maps in $ are to be thought of as ‘arbitrary’; that is, any conceivable process or scheme, which has only the property that to every point of a specified domain it gives a unique value in a specified codomain, counts as a map in S. As a consequence S itself cannot really assign any property to distinguish one point of a set A from another point of A, though the number of points of 4 is an isomorphism-invariant in S. Most of the interesting examples we discussed, such as the time-line, are only partly captured by S, since, for example the order of time involves further ‘structure’; however, all the examples have their shadow in S, and already the calculations we can do in S (using composition, the forthcoming products, etc.) shed some light on the real examples. By ascending to the consideration of categories of objects more richly structured than those in S, we can hope to see a much sharper image above the shadow, and to shed much more light on the examples by the same kind of categorical calculations. Schematically, the program is ae Shamperimage gr = J. s where % denotes an imprecise but real category, T denotes a specific chosen notion of structure, and $7 denotes the category of structures of kind T which can be built in S; the arrows denote the appropriate kind of maps between categories, known as functors, which we will discuss later. Now we return to the category in which an object is an endomap of a set. A suggestive notation for it is S°. An object of S° is any set X equipped with an endomap a. But the most important thing about a category is its maps and how they compose ~ what are the maps of S©? They are maps which ‘respect the given structure,” i.e. a map ~* f YP between two objects of $° is an S-map ¥ G ¥ which, moreover, satisfies foa=Bof Examples of categories 137 After doing several exercises you will see that this equation really is the most appro- priate expression of the idea that f preserves the given structure, i.e. that f is a way of mirroring the structure of a in the structure of 3. Exercise 1: Show that if both f as above and also yobs] 2 -» [P7 are maps in $9, then the composite gof in $ actually defines another map in ©. Hint: What should the domain and the codomain (in the sense of S°) of this third map be? Transfer the definition (given for the case f) to the cases g and gef; then calculate that the equations satisfied by g and f imply the desired equation for gf. An object of $© actually has all the structure suggested by our internal picture of an endomap a: This is because an isomorphism in S© has an inverse which is also a map in S$. It can thus be shown that if between two objects of S© there exists an isomorphism of 5°, then not only do the two sets have the same total number of points (as already mere S-isomorphism would imply), but also equal numbers of fixed points, the same number of cycles of length seven, equal numbers of points that move four steps before stopping, equal numbers of points which move two steps before entering a cycle of length three, etc. and, moreover, equal numbers of components, etc. This array of numbers (which we may learn to organize) describes the kind of structure inherent in an object of SO. 2. Typical applications of S° Objects of S© arise frequently as dynamical systems or automata. The idea is that X is the set of possible states, either of a natural system or of a machine, and that the given endomap a represents the evolution of states, either the natural evolution in one unit of time of the system left to itself, or the change of internal state that will 138 Article IIT occur as a result of pressing a button (or other control) a on the outside of the machine once. If the system happens to be in state x now, then after one unit of time or one activation of the control, it will be in state a(x). After two units of time or pressing the button twice, it will be in the state. a@(a(x)) = (@ a)(x) Similarly, a? = a aca effects the three-step evolution, etc. Questions which could be asked about a particular object of S© thus include the question of accessibility: Given a state x, is it possible to get into that state, i.e. does there exist a state x’ for which a(x") = x? as well as the question of convergence to equilibrium: Given a state x, is it possible by activating o enough times (or waiting long enough, in the natural system view) to arrive at a state which no longer changes, i.e. for some ny al™"(x) = a"(x)? 3. Two subcategories of S° By putting restrictions on the kind of endomaps allowed, we obtain subcategories 2 s* 2 s where S* means the category whose objects are all idempotent endomaps of sets and S> means the category whose objects are all invertible endomaps of sets (also known as automorphisms of sets or just as permutations); in both of these categories the definition of map between objects is the same as that stated above for SO. The numerical (or other) description of the detailed structure of a typical object in one of these two subcategories may be regarded as a specialization (somewhat less com- plicated) of the description for $©. But, as categories in their own right, these three are strikingly different, as we will see. 4. Categories of endomaps If @ is any category, we can build @© from @ in the same way we built S© from S. An object is an endomap in @ and a map is a @-map satisfying the same equation as before. There are many full subcategories of @° (the category whose objects are endomaps in @ and whose maps are ‘equivariant’ maps), for example 5@ 5 a. é é 2 e 2 Examples of categories 139 meaning: automorphisms = isomorphic endomaps. > _involutions 2 » endomaps identities > 2 idempotents where 0 is an involution of A if and only if 0° = 14. Note that an involution is automatically an automorphism (i.e. an endomap which is also an isomorphism) for it has an obvious inverse: 0~' is 0 itself if @ is an involution. Every object 4 of @ has only one identity map, but may have many idempotents and many involutions, some automorphisms which are not involutions, and some endomaps which are neither idempotents nor automorphisms. Question: Could an endomap be both an automorphism and idempotent? Yes, 1, is obviously both. Are there any others? Well, suppose we know both aoa=a acB= 1, Boa=y that is, that a is idempotent and also has a (two-sided) inverse 8. Then 14 = 028 = (00) B= 0(aef) =a0l,=0 in other words, the only idempotent automorphism is the identity. From the proof we see in fact that the only idempotent which has even a section is 14. Exercise 2: What can you prove about an idempotent which has a retraction? When @ = S, what do the internal pictures of such special endomaps look like? If 606 = 14, then the internal diagram of @ must look like this 140 Article HT that is, a certain number of ‘2-cycles’ and a certain number of fixed points (x for which 6(x) = x). Exercise 3: A finite set A has an even number of elements iff (i. if and only if) there is an involution on A with no fixed points; A has an odd number of elements iff there is an involution on A with just one fixed point. Here we rely on known ideas about numbers — but these properties can be used as a definition of odd- ness or evenness that can be verified without counting if the structure of a real situation suggests an involution. The map ‘mate of” in a group A of socks is an obvious example, Let us exemplify the above types of endomaps on the set Z={...—3,-2,-1,0,1,2,3,...} of all (positive and negative) whole numbers, considered as an object of S. Exercise 4: If a(x) = —x is considered as an endomap of Z, is @ an involution or an idem- potent? What are its fixed points? Exercise 5: Same questions as above, if instead a(x) = |x|, the absolute value. Exercise 6: If a is the endomap of Z, defined by the formula at morphism? If so, write the formula for its inverse. =x+3, is a an auto- Exercise 7: Same questions for a(x) = 5x. There are many other subcategories of @°, for example, the one whose objects are all the endomaps a in @ which satisfy acaca=a Exercise 8: Show that both @, @” are subcategories of the category above, i.e. that either an idempotent or an involution will satisfy a3 = a. Examples of categories 141 Exercise 9: In S, consider the endomap a of a three-element set defined by the internal picture . 1e Show that it satisfies a’ = a, but that it is not idempotent and that it is not an involution. 5. Irreflexive graphs There is another important category of structures of which S© itself may be con- sidered to be a subcategory. We refer to the category S'! of (irreflexive directed multi-) graphs. An object of this category is any pair of sets equipped with a parallel pair of maps, as in this diagram: a fa where X is called the set of arrows and P the set of dots of the graph. If xis an ‘arrow’ (clement of X), then s(x) is called the source of x, and ¢(x) is called the target of x. The terminology refers to the fact that any graph has an internal picture of the type Here X has five elements (a,b,...) and P has six (k, m, ...) and s(a) =k, «(e) =r, 1(d) =4q, etc. Exercise 10: Complete the specification of the two maps X—++P and XP which express the source and target relations of the graph pictured above. Is there any element of X at which s and ¢ take the same value in P? Is there any element to which 1 assigns the value k? 142 Article TIT The maps in S"! are again defined so as to respect the graph structure. That is, a map in $'! is defined to be any pair of S-maps ¥ 4 y, Ps Q for which both equations oy Wt are valid in S. We say briefly that ‘f preserves the source and target relations’ of the graphs. (The subscripts 4, D are merely to suggest the part of the map f that operates on arrows and the part that operates on dots.) Exercise 11: If f is as above and if Zz | { R is another map of graphs, show that the pair gy°f4, gp °fp of S-composites is also an S“map. Graphs have many important applications — we might consider dots as towns, and arrows as possible roads; or dots as nouns, and arrows as transitive verbs with specified subject and object. Electrical wiring diagrams, information flow diagrams, etc. are often considered explicitly as graphs, i.e. as objects in S!!, and many impor- tant relationships between graphs are expressed in terms of maps in S!!. Among the many numerical properties of graphs which remain unchanged by isomorphism are, not only the total number of arrows and of dots, but also the number of loops and the number of components. Examples of categories 143, 6. Endomaps as special graphs Why did we say that S° ‘may be considered as a subcategory of $!!7 Any such statement involves a specific way / of inserting Ft go which in this case is the following: Given any set X* equipped with an endomap, we may consider as a special kind of graph, in which the number of arrows is the same as the number of dots, and in which more precisely the source of the arrow named x is the dot also named x, but the target of the arrow named x is the dot named a(x). Now you see the method in our madness: the internal picture of an endomap is a special case of the internal picture of a graph! We said that the category S° could be considered as a subcategory of S''. Since a big part of a category is its maps, this means that our insertion idea must apply to maps as well. Indeed, if (=) + x ¥ ‘lf ‘lp Cg a satisfies the two equations required of a map in S!. in S®, then it is easy to see that — —_ a Exercise 12: If we denote the result of the above process by (f), then I(g°f) = I(g)°I(f) so that our insertion J preserves the fundamental operation of categories. 144 Article HI Exercise 13: (Fullness) Show that if we are given any S!-morphism I x—t+y Cat between the special graphs that come via J from endomaps of sets, then it fol- lows that f, = fp, so that the map itself comes via J from a map in S°. Considering J as understood, we see that our examples are related as i Q oo S285 5 7. The simpler category S!: Objects are just maps of sets A different subcategory of S"! is S!, in which an object is an arbitrary single map between two sets, and a map is a ‘commutative square of maps’ in S. Here the intended inclusion involves considering those graphs for which the source and target structures are the same map (i. graphs all of whose arrows are loops). Since an endomap is a special case of a map, there is also an obvious insertion J of S© into s!; but, crucially, it does not satisfy ‘fullness.’ There are maps J(X2") — J(¥9%) in 's! which do not come via J from maps ¥O° — Y°? in S© Exercise 14: Give an example of S of two endomaps and two maps as in ae ‘ 8 x ra ae which satisfy the equation fp °a = 8» f4, but for which f, # fo. Since it is easy to give many examples as in the last exercise, we may say that the structure preserved by an S!-map F(XO*) — (YO?) Examples of categories 145 is much ‘looser’ than the structure preserved by an actual SO-map [P< )——+ [F This remains true even for isomorphisms, so that the rich structure which S© sees in an endomap is degraded by considering it as just a map (that happens to be an endomap) to the much simpler questions: How many points are in the set and how many a-stacks of each possible size are there? 8. Reflexive graphs A final very important example is reflexive graphs: these may be considered as graphs with a third structural map i x i Ip S| | i=l, B of which both source and target are retractions; or equivalently, iis a given common section of both the source map and the target map. The following exercise asks you to prove certain consequences of these equations. Exercise 15: In a reflexive graph, the two endomaps e, = is, eo = it of the set of arrows are not only idempotent, but even satisfy four equations: exe = 6 for k,j=0,1 Of course, maps of reflexive graphs are required to respect not only source and target, but also the extra ingredient i. You should formulate the definition of map of reflexive graphs before beginning Exercise 16 Exercise 16: Show that if f4, fp in S constitute a map of reflexive graphs, then fy is determined by f, and the internal structure of the two graphs. Exercise 17: Consider a structure involving two sets and four maps as in Mr BT) Ip" (no equations required) poe (for example M = males, F = females, p and ¢' are father, and ys and y’ are mother). Devise a rational definition of map between such structures in order to make them into a category. 146 Article IIT 9. Summary of the examples and their general significance In the diagram below, the horseshoe symbols indicate full insertions. Note that J followed by the inclusion to S!! is not the same as the insertion J. The relation U between reflexive and irrefiexive graphs is not a full insertion, but a forgetful functor {it just neglects the structural ingredient 1); similarly for V. a ei A ae Reflexive es OD a ome) ei rae \ s io In all of the examples, the general kind of ‘structure’ involved can be more pre- cisely described. Each example involves a ‘category’ (species or mode) of cohesive or active sets. As opposed to the abstract sets $, which have zero internal cohesion or internal motion, these ‘sets’ have specific ways of internally sticking together and/or internally moving, and the maps in these categories permit comparing and studying these objects without tearing or interrupting them. By applying specified forgetful functors, we can also study how the objects compare if we imagine permitting (partial) tearing or interrupting to specified degrees. 10. Retractions and injectivity When does a map a have a retraction? An important necessary condition is that it should be injective. Recall the definition: Definition: We say that a map X —+ Y is injective iff for any maps T + X and T + X (in the same category) if ax, = axy then x; = x2 (or, in contrapositive form, ‘the map a does not destroy distinctions,’ i.e. if x; # Xp in the diagram below, then ax, # ax as well) Exercise 18: If a has a retraction, then a is injective. (Assume pa = /y and ax, = ax; then try to show by calculation that x; = x2.) Examples of categories 147 In the category S of abstract sets and arbitrary maps, the converse of the above exercise is almost true: If X — ¥ is any injective map in S for which X # 0, then there exist maps ¥ + X for which pa = 1y, as we have seen before. However, it is very important that this converse is not true in most categories. For example, in a category of continuous maps, the inclusion a of a circle X¥ as the boundary of a disk Y does not have a retraction; any of the S-retractions of p of the underlying sets of points would have to ‘tear’ the disk, i.e. would not underlie a continuous retraction of the spaces. We now consider an example of the same phenomenon in xe Pe | 09D Let ax = y and a0 = 0, with ¥, Y, a, and # as pictured above. Exercise 19: Show that ais a map [X22] 4» [>] in 50. Exercise 20: Show that a is injective Exercise 21: Show that, as a map Y —+ Y in S, a has exactly two retractions p. Exercise 22: how that neither of the maps p found in the preceding exercise is a map yo? |—»| yc | in S*. Hence a has no retractions in Exercise 23: How many of the eight S-maps Y —+ X (if any) are actually $°-maps? PF |_»[ oe Exercise 24: Show that our map a does not have any retractions, even when considered (via the insertion J in Section 7 of this article) as being a map in the ‘looser’ category S!. 148 Article HIT Exercise 25: Show that for any two graphs and any S!!-map between them | t Paar — on y the equation fp °s = fy °t can only be true when f, maps every arrow in X to a loop (relative to s‘, t’) in Y. To say that Ox) is an automorphism would be wrong, since Z doesn’t have fractions in it. On the other hand, there would be a germ of truth in the statement, because if Q denotes the set of rational numbers, then: Exercise 26: ; There is an ‘inclusion’ map Z ++ Q in S for which 1, [POL [PO] is a map in $, and 2. QO js an automorphism, and 3. f is injective. Find the f and prove the three statements. Exercise 27: Consider our standard idempotent Pt=| by and let YO? be any automorphism, Show that any SO-map XO* y°* must be non-injective, i.e. must map both elements of X to the same (fixed) point of f in ¥. Exercise 28: If XO° is any object of S© for which there exists an injective S°-map f to some Y“? where @ is in the subcategory of automorphisms, then a itself must be injective. Examples of categories 149 11. Types of structure A type of structure can be specified by giving the following ingredients: 1. a set of names (perhaps more than one or two) for the objects we expect as components of each single structure of the type; . another set of names for the crucial structural maps that must be specified to determine any single structure of the type; and the specification of which structural component object is required to be the domain and codomain of each structural map, but in terms of the abstract names. v » Each concrete structure of the type is required to have its structural maps conform to the abstract specification. For example, discrete dynamical systems have one com- ponent object of ‘states’ and one structural map, the ‘dynamic’, whereas graphs have two component objects ‘arrows’ and ‘dots’ and two structural maps ‘source’ and ‘target.’ Reflexive graphs have three structural maps. Our discussion of kinship systems involves also component sets and structural maps. (Note that an abstract specification of a type of structure can itself be considered a graph — see Session 17.) The pattern for defining the notion of map in any category of concrete structures is now explicitly the same for all abstract types. Namely, suppose X and Y are two structures of a given type, modeled in sets. Then for each component name A in the type, there are given sets X(A) and Y(A), so that a map XY 4 ¥ is required to involve, for each such A, a map of sets X(4) —» Y(A); but these maps are required to preserve all the structure in order to be considered to constitute together a single map of structures. Namely, for cach structural map-name a in the type, ¥ has specified a map ay : X(A) + ¥(B) where A, B are the source and target of « in the type, and also Y has specified a structural map ay: ¥(A)— ¥(B) with the same name a and the same A, B; thus the natural meaning of the statement that ‘f preserves a’ is that x) > 1 @fi-hoty ey a XO 18) in the background category of sets. To be a map of structures, f is required to preserve ail the structural maps as named by the type of structure. Thus a map in a category of structures has as many component maps as there are component-object 150 Article IIT names in the type, and is required to satisfy one preservation-equation for each structural-map name in the type. ‘There are many more categories than just those given by abstract types of struc- ture; however, those can be construed as full subcategories of the latter, so that the notion of map does not change. Such full subcategories are determined by putting restrictive conditions on the diagram that constitutes an object, the simplest sort of such condition just being a composition-equation that the structural maps are required to satisfy. For example, a dynamical system might be required to be an involution, or a ‘preferred loop’ structure in a graph might be required to have source and target both equal to the identity on dots in order to have a reflexive graph, etc. An abstract structure type often arises from a particular example as follows. Suppose # is a small family of objects and maps in a category %, with the domain and codomain of any map in # being in #. Let each object A in # be considered as the name of ‘A-shaped figures’ and each map a in # be considered as a name a* of structural map. The domain of a* is the codomain of a, and the codomain of a” is the domain of a. Then every object X of % gives rise to an #-structure whose A-th component set is the set of all #-maps A — X and wherein for each B-“+ A the structural map on these figures has for all x a(x) = xoa Exercise 29: t Every map X —+ Y in % gives rise to a map in the category of #-structures, by the associative law. For example, the abstract notion of graph structure can be identified with the con- crete diagram of graphs below because for any graph X the arrows in X may be identified with the graph maps A—+X, the dots in X with the graph maps D — X, and the source of x is then si¢(x) = xs; for any graph map X —+ Y, the associativity f(xs) = (/x)s then sub- stantiates, inside the category itself, the fact that f preserves sources: syof =fosk Examples of categories 151 Any instance of a structure ‘opposite’ to a given type (e.g. the type ‘graph’) in any category @ gives rise to an interpretation of @ into the category of ‘sets with struc- ture’ of the given type. For example if @ is some category of cohesive spaces, we might take in place of the objects D and A the objects 1 and S, a one-point space and an object representing the space of a room. In addition, we need two selected points in the room, 1+ § and 1+ S. Once these data are fixed, each object in the category @ gets an ‘interpretation’ as a graph. For example, if T is the temperature line, a dot of the ‘temperature graph’ is a point of T (a map 1 — T), and an arrow of the graph is a ‘temperature field’ in this room (a map ST). The ‘source’ of a temperature field is the temperature at the point s in the room; the ‘target’ is the temperature at 1. Exercise 30: If S, s, 1 is a given bipointed object as above in a category @, then for each object X of @, the graph of ‘X fields’ on S is actually a reflexive graph, and for each map X “ Y in @, the induced maps on sets constitute a map of reflexive graphs. 12. Guide Several useful examples of categories have been constructed by a common method, and we have begun to explore some ways in which these categories do and do not resemble the category of sets. Extended discussion of these and other categories is given in Sessions 11-18, along with a sample test after Session 17.
You might also like
- (Category Theory Homological Al) Marco Grandis - Category Theory and Applications - A Textbook For Beginners-World Scientific Pub Co Inc (2018)100% (1)(Category Theory Homological Al) Marco Grandis - Category Theory and Applications - A Textbook For Beginners-World Scientific Pub Co Inc (2018)304 pages
- Notes Ankjvkfkd Exercises From Cat, Jvhkdegories For The Working Mathematician100% (1)Notes Ankjvkfkd Exercises From Cat, Jvhkdegories For The Working Mathematician42 pages
- Conceptual Mathematics A First Introduction To Category Theory100% (3)Conceptual Mathematics A First Introduction To Category Theory376 pages
- F. William Lawvere, Stephen H. Schanuel - Conceptual Mathematics - A First Introduction To Categories (2009, Cambridge University Press) (1) - TextNo ratings yetF. William Lawvere, Stephen H. Schanuel - Conceptual Mathematics - A First Introduction To Categories (2009, Cambridge University Press) (1) - Text403 pages
- Category Theory George Janelidze 2020 Updated 20220519100% (1)Category Theory George Janelidze 2020 Updated 2022051952 pages
- Introduction To Categories Work in ProgrNo ratings yetIntroduction To Categories Work in Progr118 pages
- Category Theory: Randall R. Holmes October 8, 2019100% (1)Category Theory: Randall R. Holmes October 8, 201999 pages
- Math - Harvard.edu/ Eriehl/survey - PDF Math - Harvard.edu/ Eriehl/266x/basic PDFNo ratings yetMath - Harvard.edu/ Eriehl/survey - PDF Math - Harvard.edu/ Eriehl/266x/basic PDF14 pages
- (Encyclopedia of Mathematics and Its Applications 50) Francis Borceux - Handbook of Categorical Algebra 1 - Basic Category Theory - Cambridge University Press (1994) (Z-Lib - Io)No ratings yet(Encyclopedia of Mathematics and Its Applications 50) Francis Borceux - Handbook of Categorical Algebra 1 - Basic Category Theory - Cambridge University Press (1994) (Z-Lib - Io)361 pages
- Download Complete Introduction to Category Theory and Categorical Logic Thomas Streicher PDF for All ChaptersNo ratings yetDownload Complete Introduction to Category Theory and Categorical Logic Thomas Streicher PDF for All Chapters55 pages
- MATH325-425-Representation-Theory-of-Finite-Groups-NotesNo ratings yetMATH325-425-Representation-Theory-of-Finite-Groups-Notes104 pages
- Foundations: 2.1 Objects, Relations, and OperationsNo ratings yetFoundations: 2.1 Objects, Relations, and Operations56 pages
- V Sankrithi Krishnan An Introduction To Category Theory North Holland 1981100% (2)V Sankrithi Krishnan An Introduction To Category Theory North Holland 1981181 pages
- Francis Borceux Handbook of Categorical Algebra 1 Basic Category TheoryNo ratings yetFrancis Borceux Handbook of Categorical Algebra 1 Basic Category Theory361 pages
- Generic Figures And Their Glueings A Constructive Approach To Functor Categories Marie La Palme Reyes pdf download100% (1)Generic Figures And Their Glueings A Constructive Approach To Functor Categories Marie La Palme Reyes pdf download78 pages
- Introduction To Abstract Algebra - Math 113 - Paulin PDFNo ratings yetIntroduction To Abstract Algebra - Math 113 - Paulin PDF78 pages
- E. Cheng & A. Lauda - Higher-Dimensional Categories. An Illustrated Guide Book100% (3)E. Cheng & A. Lauda - Higher-Dimensional Categories. An Illustrated Guide Book182 pages
- Lecture Notes in Lie Groups: Vladimir G. Ivancevic Tijana T. IvancevicNo ratings yetLecture Notes in Lie Groups: Vladimir G. Ivancevic Tijana T. Ivancevic74 pages
