A Generalized Approach to the Fundamental Group Author(s): Daniel K. Biss Source: The American Mathematical Monthly, Vol. 107, No. 8 (Oct., 2000), pp. 711-720 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/2695468 Accessed: 11/02/2009 15:20 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=maa. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org. Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to The American Mathematical Monthly. http://www.jstor.org A Generalized Approach to the Fundamental Group Daniel K. Biss 1. INTRODUCTION.The classical theory of fundamentalgroups and covering spaces is among the most elegant stories in the mathematicalcanon. The subject, which bears a striking relationship to the study of Galois groups and field extensions, provides historical motivation for a significant portion of modern algebraictopology.The basic idea is to associate a group,the fundamentalgroup, to each topological space, and then to extract topological informationfrom the lattice of subgroupsof the fundamentalgroup. More specifically,elements of the fundamentalgroup of a space X correspond to loops in X modulo the relationthat if one loop can be continuouslydeformed into another,then the two are consideredequivalent.A "cover"of a space X is a map from anotherspace to X such that X can be coveredby open sets Ui whose inverse images are simply disjoint copies of Ui. It then turns out that each subgroupof the fundamentalgroupof a space correspondsto a cover.Intuitively,a subgroupof the fundamentalgroup describessome loops or holes in X, and the correspondingcover gives a procedurefor unwrappingthe loops. This beautiful relationshipis useful in manyways, both because coveringspaces can be used to computethe fundamentalgroupand because knowledgeof the fundamentalgroup tells us when to search for potentiallyinterestingcovers of a space. However, the theory works only for certain relatively nice spaces, namely semi-locallysimplyconnected spaces. Worse yet, it can be shown that the correspondencefails for all spaces that are not semi-locallysimplyconnected!The aim of this paper is to present a slightly modified version of the definitions that generalizesthe standardcorrespondencebetween connectedcoversand subgroups of the fundamentalgroup to a much broader setting. In doing this, we need to introducesome additionalstructureto the fundamentalgroup that is able to see subtlerdifferencesbetween some spaces.This extrastructureallowsus to state the usual theoremsof the subjectin a languagethat generalizesappropriatelyto more exotic spaces. This paper is meant to be a non-technicalcompanionto the more precise [2], in whichthe details not includedhere are filled out. A similarprogram is being carriedout from a slightlydifferentpoint of view by Bogley and Sieradski; see [3], [4], and [7]. We begin, in Section 2, by briefly reviewingthe classical theory. No previous knowledgebesides general point-set topology and basic group theory is assumed. In Section 3, we present the new constructionsand ideas that are necessary to develop the generalizedtheory. Finally, in Section 4, we sketch (mostly without proofs) the new resultsthat are extensionsof the theoremsdiscussedin Section 2. 2. THE CLASSICALTHEORY.The fundamentalgroup was first introducedby Poincarein the groundbreakingpaper [6] in which he essentiallyfounded the field of topology.Poincareoutlined the theory from its very inception, albeit in a style that is somewhatless precise than that of the typicalpiece of modernmathematics. October 2000]GENERALIZEDAPPROACH TO THE FUNDAMENTAL GROUP 711 In this paper,he beganby definingthe objectsof study,which are what would now be called submanifoldsof Rn, modulo some imprecisionin Poincare'sdefinition. Later in the article, he defined homology and the fundamentalgroup, with the object of determiningwhen certain manifolds are not homeomorphicto one another. The motivatingconcept is a simple one: to find a systematicway of measuringthe "holes" in a manifold. For example, how can we prove that the spaces R2 and R2 - {(0, 0)} are not homeomorphic? Let I denote the closed unit interval[0, 1]. Poincare'selegant idea was to fix a point x in a manifold X, and considerthe set of all continuouspaths ry: I -> X such that -y(O)= -y(l) = x. We identify two such paths if they are homotopic equivalent,that is, if one can be continuouslydeformedinto the other. Formally, two paths ryo and -yl are said to be homotopic equivalent if there exists a continuous map F: I x I -* X with F(O, s) = yo(s), F(1, s) = y1(s), and F(t, 0) = F(t, 1) = x. Now, given two paths ryo and ryl, we can form a new path ryo* -yl by concatenatingthe two; explicitly, (y0*^l)(s)= Yo(2s) - 1) l^i(2s for 0 < s < for 2< S < 1. Although concatenationis not an associativeoperation (that is, the paths Pi = (YO* Y) * Y2 and P2 = YO*((Y * Y2) need not be equal), it is associativeup to homotopy (that is, Pi and P2 are homotopic). Thus, we can make the set of equivalenceclasses of paths into a group,called the fundamentalgroup-rT(X,x) of the manifoldX. With some work,one can in fact see that the fundamentalgroupof R2 is trivial and the fundamentalgroup of R2 _ {(0, 0)} is isomorphic to the integers, Z, with the path -y(s) = e 2lTiS acting as a generator. Thus, in this case, the fundamentalgroup does manage to distinguishbetween the two spaces in question. In general, the fundamentalgroup of a space is a tool that encodes which loops in the space can be deformedinto one another.Thus, it providesa technique ideallysuited to the task of differentiatingbetween R2, in which everyloop can be contractedto a point, and [R2 - {(0,0)}, in which loops that go aroundthe origin cannot be contracted. Of course, the fundamentalgroupis nowherenear a subtle enough invariantto distinguishbetween all pairs of spaces (for example, every sphere of dimension greater than one has trivialfundamentalgroup),but it remains a cornerstonein the foundation of algebraic topology. Unfortunately, in spite of (or perhaps because of) its enticinglyintuitivedefinition,the fundamentalgroupwas relatively difficultto computefor manyyears. It was the introductionof coveringspaces into the picturethat made the fundamentalgroup a well-understoodinvariant. The historyof the theory of coveringspaces is a convolutedone, not simplified by the fact that Riemann, perhaps the earliest focal player in the development, publishedalmostnothingon the subject.His ideas were only graduallydiseminated by other mathematicians,such as Betti [1], primarilyin correspondence.The genesis of the idea comes from the followingphenomenon.Considerthe function f: C -> C defined by f(z) = Z2. For every z, we have f(z) = f( -z), so that every nonzerocomplexnumberhas two pointsin its inverseimage;however,f- 1(0) = {0}. Suppose that we would like to constructsome new space X that locally "looks like"C and a function Sq : C -? X that "behaveslike" the function f but is also one-to-one.This would be useful because an inverseto the function Sq would play the role of a square root function. We can constructsuch an X by taking two copies of the complexplane, C1 and C2, cuttingthem both open along the positive 712 ? THE MATHEMATICALASSOCIATIONOF AMERICA [Monthly107 real axis, and then gluing them together along the cuts in such a way that for a positivereal numbera, and small E, a + iE in the first plane is close to a - i E of the second, and vice versa (see Figure 1). Then, we may define the function Sq by Sq(z) {Z2 E C21 for lm(z) ? 0 Now, since X was constructedout of two copies of C, it retains a projection mapping p: X -> C. This map is called a branched covering of C-this means that away from a finite collection of branchingpoints (in this case, 0 is the only branchingpoint), p is a local homeomorphism.More precisely,for every complex number z # 0, there is an open set U c C containing z such that p-1(U) is homeomorphicto two disjointcopies of U, and the restrictionof p to either copy is a homoeomorphism.The general definitionof branchedcoveringdiffersonly in that the "two"in the previoussentence can be replacedby any positiveinteger (or infinity). Figure 1. The construction of the space X. So far, the promisedconnectionwith the fundamentalgroup has not emerged. To see the relationshipbetween these two concepts,let p: X -? C be a branched cover with branching points Pi, P2 ... ., Pn Let C = C - {Pi,* .. , Pnl} X = p (0) and =p I Then p3is a branched covering map with no branching points, or a coveringmap. Now, let us examine more carefully the example of f(z) = z2. Ignoring the point z = 0 (which we need to do anywayin order to make the branchedcover we constructedinto a cover),we have a two-to-onefunction, and so we have troubleconstructinga squareroot functionthat is to be its inverse.Of course, one can try to solve this problemas it is solved in the real case, namelyby simplydeclaringthe squareroot of a positive real numberto be positive,but this brings up a new difficultythat did not present itself in the real case. Indeed, consider a circle of radius one going counterclockwise about the origin: y(t) = e2Tit* Since we declared that the square root of a positive real numberis positive, we find that -y(O)= 1. Now, of course we must alwayshave y(tE) {e-ie, -eTit}, but since we have already declared y (0) = +e0, in order to ensure that our square root function is continuous, as t increases, it must remain the case that y(t) = +ewit.However,takingthe limit as t approaches1, we find a contradiction: 1 = - 1! Thus, the existenceof this branchedcoverX is closelyrelatedto the existenceof a nontrivialelement of the fundamentalgroupof C - {0},represented by y. There is also informationto be uncovered by looking more carefully at the relationshipbetween the fundamentalgroupsof X and C - {0}.Notice first of all October 2000]GENERALIZEDAPPROACHTO THE FUNDAMENTAL GROUP 713 that if we have two pointed spaces (Y1,yl) and (Y2, Y2) and a continuousfunction f : Y1 -> Y2 such that f(Y1) = Y2' we get an induced map f*: ir1(Y1,yl) -71(Y2, Y2). This is done simplyby taking a loop y: I -* Y1 and composingwith f to get a loop f o y : I -> Y2. We next try to understandthe fundamentalgroup of X. Recall that X is made up of two copies of C - {0}, so we start at the point 1 in one of the copies and begin following the path y that goes around the origin counterclockwise.Because of the way the two planes in X are glued together,after going aroundthe originonce, we find ourselveson the other plane, and need to go aroundagainto get back to the point where we started.Denote by jo this path that goes aroundboth planesbefore comingbackto its initialpoint. One can check that the fundamental group of X is Z, generated by the path y. Thus, the projection Z that is multiplicationby 2, since po f :X -> C - {0} inducesthe map p*: Z goes aroundthe originin C twice. Thus, the image of the map fip*is the subgroup 2Z c Z. As we will see, this fits into a very general picture:the coveringspaces lying over a space can be understoodin terms of the subgroupsof the fundmental group of the space. Now, a space X is said to be simplyconnectedif 71(X, x) is the trivialgroupfor all x c X. Furthermore,for any open set U c X, we have the obvious inclusion map i: U " X, which, for any x c U, induces a map i * : ir1(U, x) -s 1(X, x). We say that X is semi-locallysimplyconnectedif every x E X has a neighborhoodU such that the induced map i* is the trivial map. Intuitively speaking, X is semi-locallysimplyconnectedif it does not have a collection of loops that "bunch up" (for a more precise descriptionof such phenomena, see Section 4). We are now ready to list the theorems that make up the relationship between the fundamentalgroup and coveringspaces. Throughoutthe rest of this section, all spaces are assumedto be semi-locallysimplyconnected. Proposition 1. Let p: X -* Y be a covering map. Then for all x map p* : ir1(X, x) -- 'nr1(Y,p(x))is injective. E X, the induced Thus, to understand what a covering map does to fundamental groups, it suffices to know the image of the induced map on 7r1.It turns out that the image of the induced map tells us everythingthere is to know about the coveringmap. That is, two covers P1 : X, -> Y and P2 : X2 -* Y are said to be isomorphic if there is a homeomorphism h: X1 -* X2 such that P2 o h = Pi, and using the fundamental group, it can be determinedpreciselywhen two covers are isomorphic. Theorem 1. For everysubgroupH c ir1(Y, y), there is a coveringmap p : X Yand a point x E X such that p. (r1(X, x)) = H. Furthermore,two coversp1: X1 Y and P2 : X2 -* Y are isomorphic if and only if (P1):* ( r(X,, x)) and (P2)* (7r1(X1I X2)) are conjugate subgroupsof 7r1(Y,y). In particular,there is a unique cover p : X -* Y that correspondsto the trivial subgroup-equivalently, this is the unique cover with X simply connected. Furthermore,if p': X' -> Y is another cover, then the cover of X' correspondingto the trivial subgroup of 7r1(X')will be a covering map q : X -> X' such that p = p' oq. For this reason, the unique simply connected cover X is called the universalcover of Y. For example, the map p : C ?-> - {(0, 0)1 given by p(z) = eZ is the universalcover of C - (0, 0)1.To see the "universality"propertyin action, consider the covering map p': : - ((0, 0)1 -> ? - (0, 0)1 defined by p'(z) = z2. Then our theory predicts the existence of a covering map q : C -? C - (O,0)} satisfyingp = p' o q. The map q(z) = ez/2 turns out to do the trick. 714 ?? THE MATHEMATICALASSOCIATION OF AMERICA [Monthly107 Now, a homeomorphism h: X -> X such that p o h = p is called a deck transfor- mationof X. The set of all deck transformationscan easily be seen to be a group under composition.Magically,it turns out that this group is completely understood. Theorem 2. Let p: X -> Y be a universal cover and y transformationsof X is isomorphic to i1(Y, y). E Y. Then the group of deck We now illustrate all of these phenomena with an example. Although the motivation behind this theory comes from complex analysis, the most easily visualizedcase comes from a more basic space. So, let Y = S1, the circle given by the equation {z E C: lzI = 11. It can be checked that iT1(Y,1) = Z (generatedby the loop y(s) = e2wis). Now, the subgroupsof Z are all of the form nZ for some nonnegativeinteger n. First, suppose that n > 0. Theorem 1 predictsthat there is a unique cover p: X -* Y with p*(X1(X, x)) = nZ. To find such a cover, we first need to understandthe elements of the subgroupnZ c r-1(Y,1). This subgroupis generatedby the path yn(s) = e2n,is. Thus our coveringmap must somehow"wrap around"Y exactly n times. Considerthe map p: Y -- Y given by p(z) = zn. Since p oy, = yn, we find that this map has preciselythe desired property. We still have not dealt with the case n = 0, that is, the case in which X is simplyconnected.To find this universalcover of Y, we need to somehowunwrap the circle Y to get a simply connected space. Indeed, define p R -: Y by p(x) = e2 ix. This does precisely what we hoped for-it is a local homeomorphism in which the preimageof each point is countablyinfinite. We now try to compute the group of deck transformations.Suppose h: R 1Ris a homeomorphismsuch that p o h = p. Then for all x, we know that p(h(x)) = p(x), or e2Tih(x) = e2wix, or = 1, SO that x - h(x) must be an integer. But then the function e2wi(x-h(x)) x - h(x) is a continuousinteger-valuedfunction,so it must be constant.Thus, we have h(x) = x + h(O). Furthermore, since 0 - h(O) is an integer, h(O) must be an integer, so h(x) = x + k for an arbitraryinteger k. Therefore, the group of all such h is preciselythe group Z, which is also the group ,T1(Y,1), as predictedby Theorem2. 3. SOME EXTENDED DEFINITIONS. In the rest of this article, we set up a frameworkthat gives us naturalgeneralizationsof the resultspresentedin Section 2. The basic observationis that the fundamentalgroupis not only a group,but also a topologicalspace. A topologicalgroupis a group G that is a topologicalspace in which the multiplication map G x G -* G and the inversion map G -* G (sending x to x-1) are continuous.It turnsout that the fundamentalgroupcan be naturally given the structureof a topologicalgroup.Then, the conditionof semi-localsimple connectednessof a topologicalspace will correspondto a topologicalconditionon its fundamentalgroup, and this motivatesall our new definitions. We begin by introducingthe compact-opentopology. Let X and Y be two topologicalspaces. The set Hom(X, Y) of continuousmaps from X to Y can be endowed with a topology,called the compact-opentopology,in the followingway. Let K c X be a compactsubset of X, and U c Y an open subset of Y. Denote by KK, U) the set of all continuous maps f: X -* Y such that f(K) c U. Then the compact-opentopologyon Hom(X, Y) is the topologywith sub-basis{(K, U) IK c X compact,U c Y openi, that is, the minimaltopologycontainingall <K, U) as open sets. In particular,we can furnishthe space Hom(S1,X) with the compactopen topology.Now, for any x E X, the space Hom((S1,1),(X, x)) is a subspaceof Hom(S1,X), and can accordinglybe giventhe subspacetopology.Lastly,the group October 2000]GENERALIZEDAPPROACH TO THE FUNDAMENTAL GROUP 715 mT(X,x) is simplya quotient of the space Hom((S1,1), (X, x)) (by the homotopy relation),and hence also comes with a naturaltopology.It is not too hard to check that in this topology,the multiplicationand inversionmaps are continuous,so that m1(X,x) is actuallya topologicalgroup. When we want to stress the topological structureof the fundamentalgroup,we denote it by r-t0P(X,x). Viewing the fundamentalgroup as a topologicalgroup gives us a slightlyfiner invariant that is more naturally suited to the study of spaces that are not semi-locallysimply connected. However, to have any hope of generalizing the results of Section 2, we need also to modify somehow the definition of the other majorplayerin that story,namelythe coveringspace. Recall that a coveringmap is a map p: X -- Y such that every point y E Y has a neighborhoodU such that p- (U) is just a collectionof disjointcopies of U. Anotherway of sayingthis is that p1'(U) is homeomorphicto the space U x F for some discrete space F. Of course, we can make an analogous definition for an arbitrary(not necessarily discrete) space F. A map p: X -- Y is said to be a fiber bundle with fiber F if everypoint y E Y has a neighborhoodU such that there exists a homeomorphism h:p-1(U) -> UxF such that iToh =p Ip-1(U), where iT:-:UxF -* U is the projection map. Now, a covering map is simply a fiber bundle with discrete fiber-to complete our new picture, it suffices to replace the notion of "discrete fiber"with an appropriatelyrelaxedconstraint. In order to decide how preciselyto weaken the definitionof coveringspaces, let us take stock of our situation.First of all, we would like to produce a theory that "works"for some spaces that are not semi-locallysimplyconnected.In orderto do this, we must allow fiber bundleswith non-discretefibers.Secondly,we must insist that the results of Section 2 hold in some guise in our new setting. Thus, it is essentialto isolate preciselywhat propertiesof coveringmaps are needed to make the machine turn. It turns out that most of these proofs (see [5], or, for an especially systematic and general exposition, [8]), hinge on one fundamental property:the fact that coveringmaps satisfy the uniquepath-liftingproperty.This property says that if p: X -* Y is a covering map, y: I -* Y is a path in Y, and x E X is a point such that y(O) = p(x), then there is a unique path : I -* X such that j(O) = x and p o y = y. Furthermore,it can be shown (again, see [8]) that as long as F is a totally disconnectedspace, then any fiber bundle p: X -- Y with fiber F satisfiesthe unique path-liftingproperty.Thus,we choose as our objectsof study rigid covering bundles p: X -* Y, or fiber bundles with totally disconnected fiber. As we will soon see, shiftingour attentionfrom the classificationof coversof a given space Y as in Section 2 to a study of rigid coveringbundles enables us to obtain some resultsfor spaces that are not semi-locallysimplyconnected. 4. THE EXTENDEDTHEORY.In this section, we finally demonstratehow to extend the results of Section 2 to some spaces that are not semi-locallysimply connected. In order to give the results some conceptual meaning, it is useful to have a comparativelysimple example in the back of one's head at all times. One such exampleis the Hawaiianearringspace HE (see Figure 2). The space HE is a subset of the plane consistingof infinitelymany circles of decreasingradii, all of which are tangent to one another. That is, HE is the union over all positive integers n of the circle of radius 1 and center (, 0). This space is not semi-locally simplyconnected,for any open neighborhoodof the origincontainsinfinitelymany of the circles, each of which contributesnon-trivialelements of the fundamental group. It is precisely this type of phenomenonthat prevents a space from being semi-locallysimplyconnected. 716 ?BTHE MATHEMATICALASSOCIATIONOF AMERICA [Monthly107 Figure 2. The Hawaiian Earring HE. The idea that motivatesthe results of this section is very simple. First, notice that a universalcoverof a space Y is a rigidcoveringbundlewhose fiber is discrete and correspondsbijectivelyto r1(Y). Second, such a gadget exists if and only if Y is semi-locallysimply connected. In other words, in the semi-locallysimply connected case, we know that there exists a universalcover, and its fiber has two properties:it is discrete,and there exists a bijectionbetween it and 'n-1(Y).Hence, if only we could view the elements of the fiber of the universalcover as corresponding naturally not just to those of T1(Y),but also to those of T?oP(Y), then a natural syllogismwould suggest itself: Y has a universal cover if and only if 1 toP(Y) is discrete.This indeed turns out to be the case. Theorem 3. Let Y be a topological space. Then Y is semi-locally simply connected if and only if iTOP(Y) is discrete. Thus, if Y has a universal cover, then its fiber is homeomorphic to r1OP(Y). This observationgives us the vantagepoint fromwhichwe can shift our point of view to the more general one we now pursue. Indeed, instead of viewing the universalcover as the objectof study,we can seek out a rigidcoveringbundlewith fiber Tt0P(Y)-clearly,this objectwould be the universalcover in the semi-locally simplyconnected case. Unfortunately,this type of rigid coveringbundle need not alwaysexist, but the conditionunder which it does exist is far weaker than that of semi-localsimple connectedness. Proposition 2. Let Y be a topological space. Then there exists a rigid coveringbundle p : X -> Ywith fiber <TtP(Y) if and only if the path-components of s?oP(Y) arepoints. Although the proof of existence is relativelyinvolved and dry, note that the condition that iTtP(Y) be totally path-disconnected is clearly necessary, simply by the definitionof rigidcoveringbundles.Thus, this result is the best that one could hope for. In fact, we can go slightlyfurtherand obtaina completegeneralizationof Theorem 1. First, we need to point out that an extension of Proposition1 also holds in the new setting. Proposition 3. Let p: X induced map p*: -* Y be a rigid covering bundle. Then for all x E X, the iT1(X, x) -- iT1(Y,p(x)) is injective. Now we can state the classificationof rigid coveringbundles. Of course, we need a notion of isomorphismof rigid coveringbundles;this definitioncan also be October2000]GENERALIZEDAPPROACHTO THE-FUNDAMENTAL GROUP 717 made in analogy with concepts already introduced. That is, an isomorphism -' Y and P2: X2 -* Y is a homeomor- between two rigid covering bundles Pi: X1 phism h: X1 -* X2 such that Pi = P2 o h. Theorem 4. Let H c Ttp(Y, y) be a subgroup such that the left coset space s?oP(Y, y)/H is totally path-disconnected. Then there exists a rigid covering bundle p: X -> Y and a point x E X such that p*(Xr1(X, x)) = H. Furthermore, two rigid covering bundles Pi: X1 -* Y and P2: X2 -* Y are isomorphic if and only if of 'i1(Y,y). (p1):O *T1(X1, x1)) and (P2)* (T1(X1, X2)) are conjugatesubgroups Incidentally,lest it seem aestheticallydissatisfyingfor a criterionto depend on topologicalpropertiesof the left coset space of a subgroupof i1tP, notice that the left and right coset spaces are homeomorphicvia the inversionmap. Finally,we providea generalizationof Theorem2 to the land of rigid covering bundles.As with the other concepts introducedin Section 2, the definitionof the groupof deck transformationsmust be modifiedslightlyin orderto state the result in its strongestform. So, suppose p: X -* Y is a rigid coveringbundle with fiber s-?oP(Y) (or, equivalently,with X simply connected). Then much as before, the group of deck transformationsof the rigid coveringbundle is defined to be the group of homeomorphisms h: X -> X satisfying p o h = p. However, once again, we notice that this groupis actuallya subset of the set of all continuousmaps from X to itself, and as such, inherits a compact-opentopology.Thus, we can view the groupof deck transformationsas a topologicalgroup. Theorem 5. Let p: X -* Y be a rigid covering bundle with X simply connected and locallypath-connected. Then iTOP(Y) is isomorphicas a topologicalgroup to the group of deck transformationsof the bundle. Once again,let us look at some examplesto see how this theoryworks,and how the criteria that need to be satisfied to guarantee the existence of certain rigid coveringbundlescan fail. First,recall the exampleof the HawaiianearringHE. In order to see whether our machinerygives us interestingresults for this space, we first need to understand the fundamentalgroup of HE. For each n > 1, let ,yn: I -> HE be the loop that circumnavigates the nth circle, that is, -yn(t)= sin (2'nTt)). Then each yngives an element of iT1(HE,(0, 0)). It (1 - cos (2-t)), 0 is not too hard to see that there are no relations among the yn that is, that T1(HE,(0,0)) containsa free groupwith generators{y1, 2 ... }. However, it also contains more elements: for example, consider the path that for 1 - 1/2n-1 < t < 1 - 1/2n traces the loop yn at 2n times the ordinaryspeed. This provides a new element of iT1(HE,(0, 0)) that is not an element of the free 0)) can be describedas the set of group {yl, Y29 ... }. It turns out that iT1(HE,(0, infinite (formal) products of the yn and their inverses, subject to the condition that in any particularproduct, each yn appears only finitely many times. Thus, (x2x4x6 ...)( ... x4x3x2x1) is an element of iT1(HE,(0,0)), but x1x2xlx3xlx4xl .1 0)) as having a set of "formal"generators is not. This descriptionof iT1(HE,(0, requiressome care:for example,althoughthe words x1x2x3 ... and ... x-1x-lxj1 are inverses, it is not immediatelyobvious what cancellation laws lead to the = e. A modified approachto combinatorial formula (x1x2x3 ...)( *... x1xj1xj1) grouptheorythat irons out these difficultiesin a way that is applicableto topology is developedin [7]. 718 ? THE MATHEMATICALASSOCIATIONOF AMERICA [Monthly107 In any case, regardless of the technical difficulties, it is pretty clear that (0,0)) is a (non-discrete)topological space with no nonconstantpaths. (Non-discretebecause any open neighborhoodof the origin contains infinitely many loops, and hence each KK,U) contains infinitely many xi.) Thus, by Proposition 2, there exists a rigid covering bundle p: X -> HE with fiber i4tP(HE,(0, 0)). It turns out that not all spaces enjoy this property.For example, considerthe exampleof the HarmonicarchipelagoI1 (see Figure3). This example is due to Bogley and Sieradski;see also [3]. To constructIA, begin with the disc of unit radius centered at the point (1, 0) in the plane; notice that HE is a subspace of this disc. Now (thinking of the x,y-plane a subset of 3-space), we stretch the surface of the disc to form very narrowspires of unit height at each i4tP(HE, point of the form (n + n 9 1,0), that is, along the x-axis between every pair of circlesbelongingto HE. Figure 3. The Harmonic Archipelago H. Now, althoughintuition might suggest that w,(IL4, (0, 0, 0)) ought to be trivial (in fact that I4 ought to be contractible),this turns out to not be the case. For example,consider the loop yl. If it can be contractedto the constant loop, then certainlyas it is being deformed,it must pass throughthe top of each spire, that is, through each point of the form (n + n 0, 1). If F: I x I -* A is a homotopy between yi and the constantpath, then I x I must contain a sequence of points a1, a2,... such that F(an) = ( + n 1 0,1). But by compactnessof I x I, the has a convergent subseqence, whereas no subsequence of sequence a,, a2,... ( + n 1, 0, 1) can converge to a point in A. This provides us with a contradic- tion: thus, T1(1L4, (0, 0, 0)) is not trivial,since Yl representsa non-trivialelement. (0, 0, 0)) is isomorphicto 17T(HE,(0, 0)) modulo Indeed, it turnsout that ,1(1L4, the relation xn xm for all n, m ? 1. Equivalently,a word a is identified with anyword that can be obtainedby choosingfinitelymanyletters in a and changing them to any other letters. In particular,notice that any open set U in 14 containingthe origin contains the images of all but finitely many of the yn, say and any word a contains only finitely many letters of the form -yn 1,., )nk xn1,... ,xnk. Therefore, a is equivalentto some word none of whose letters are of the form xn1, . .,. xnk, and so for any K, we see thata E (K, U>. Since a was an arbitrary word, it follows that for any compact K, (K, U> = v,1(1L4,(0,0,0)). Therefore,ITtoP(IL4,(0, 0, 0)) has the trivialtopology-that is, its only open subsets October2000]GENERALIZEDAPPROACHTO THE FUNDAMENTAL GROUP 719 are itself and the empty set. Thus, for this disastrousexample,Theorem 4 tells us that no rigidcoveringbundlesexist! So, our new theory,while more powerfulthan the old, still has considerablelimitationsin sufficientlyexotic settings. The appeal of the new theory lies in the fact that much of its reasoningfollows in quite the same vein as the classicalsetting, once we observethat the fundamental group can be viewed as a topologicalspace. Thus, without doing a significant amountof extrawork,we are able to removethose old hypothesesthat turn out to be unnecessary,and at the same time gain a clearer understandingof what obstructionslie in the way of the constructionof coveringspaces. ACKNOWLEDGMENTS. I thank Bill Bogley and Al Sieradski for sharing with me their work on this subject, and especially for showing me the striking example of the Harmonic Archipelago. Also, many thanks to Jim Davis and Kent Orr for their insight and advice. REFERENCES 1. E. Betti, Sopra gli spazi di unnumero qualunquedi dimensioni, Ann. Mat. Pura Appl. 4 (1871) 140-158. 2. D. K. Biss, The topological fundamental group and generalized covering spaces, in preparation. 3. W. A. Bogley and A. J. Sieradski, Weightedcombinatorial group theory and wild metric complexes, Groups-Korea 98, (A. C. Kim, ed.), de Gruyter, Pusan, 2000, pp. 53-80. 4. W. A. Bogley and A. J. Sieradski, Universalpath spaces, http: // osu. orst. edu/- bogleyw. 5. J. R. Munkres, Topology: a first course, Prentice-Hall, Englewood Cliffs, NJ, 1975. 6. H. Poincare, Sur l'analysis situs, C. R. Acad. Sci. Paris. 115 (1892) 663-636. edu/-bogleyw. 7. A. J. Sieradski, Omega-groups,http: // osu.orst. 8. E. H. Spanier, Algebraic Topology, McGraw-Hill, New York, 1966. DANIEL BISS was born in 1977 in Akron, OH, and grew up in Bloomington, IN. He received an A.B. summa cum laude in mathematics from Harvard University in 1998, and is now a second-year topology graduate student at MIT. Among his awards are the Thomas Wendel Hoopes prize for an outstanding undergraduate thesis at Harvard, and the 1998 AMS-MAA-SIAM Morgan prize for undergraduate research. He has for several years taken a great interest in the teaching and exposition of mathematics, during which period he has several times been a research advisor at the summer Research Experience for Undergraduates in Duluth, MN, a course assistant at Harvard, and a frustrated member of Harvard's undergraduate mathematics curriculum committee. Daniel also avidly pursues interests in photography, music, writing, and travel. Room 2-251, MassachusettsInstitute of Technology, Cambridge,MA 02139 daniel@math.mit.edu 720 ? THE MATHEMATICALASSOCIATIONOF AMERICA [Monthly107