ALGEBRAIC COLIMIT CALCULATIONS IN HOMOTOPY THEORY USING FIBRED AND COFIBRED CATEGORIES RONALD BROWN AND RAFAEL SIVERA Bangor University Maths Preprint 06.08 ABSTRACT. Higher Homotopy van Kampen Theorems allow the computation as * *col- imits of certain homotopical invariants of glued spaces. One corollary * *is to describe homotopical excision in critical dimensions in terms of induced modules * *and crossed modules over groupoids. This paper shows how fibred and cofibred categor* *ies give an overall context for discussing and computing such constructions, allowin* *g one result to cover many cases. A useful general result is that the inclusion of a fi* *bre of a fibred category preserves connected colimits. The main homotopical application * *are to pairs of spaces with several base points, but we also describe briefly the sit* *uation for triads. 1 Contents 1 Introduction * * 1 2 Fibrations of categories * * 4 3 Cofibrations of categories * * 8 4 Pushouts and cocartesian morphisms * *11 5 Groupoids bifibred over sets * * 13 6 Groupoid modules bifibred over groupoids * * 14 7 Crossed modules bifibred over groupoids * * 16 8 Crossed complexes and an HHvKT 19 9 Homotopical excision and induced constructions * * 23 10 Crossed squares and triad homotopy groups * *28 1. Introduction One of our aims is to give the framework of fibred and cofibred categories to c* *ertain colimit calculations of algebraic homotopical invariants for spaces with parts glued to* *gether: the data here is information on the invariants of the parts, and of the gluing proc* *ess. _____________ This work was partially supported by a Leverhulme Emeritus Fellowship 2002-4* * for the first author. 2000 Mathematics Subject Classification: 55Q99, 18D30, 18A40. Key words and phrases: higher homotopy van Kampen theorems, homotopical exci* *sion, colimits, fibred and cofibred categories, groupoids, modules, crossed modules . Oc Ronald Brown and Rafael Sivera, . Permission to copy for private use gran* *ted. 129/09/08 1 2 The second aim is to advertise the possibility of such calculations, which a* *re based on various Higher Homotopy van Kampen Theorems (HHvKTs)2, proved in 1978-87. These are of the form that a homotopically defined functor : (topological data) ! (algebraic data) preserves certain colimits [BH78 , BH81b , BL87a ], and where the algebraic dat* *a contains non-Abelian information. The antecedent for dimension 1 of such functors as is the fundamental grou* *p ss1(X, a) of based spaces: the van Kampen theorem in its form due to Crowell, [Cro59 ], g* *ives that this ss1 preserves certain pushouts. The next advance was the fundamental groupoid functor ss1(X, A) from spaces * *X with a set A of base points to groupoids: groupoids in effect carry information in d* *imensions 0 and 1. For example, the groupoid van Kampen theorem [Bro67 , Bro06 ] gives t* *he fundamental group of the circle as the infinite cyclic group C obtained in the * *category of groupoids from the (finite) groupoid I ~=ss1([0, 1], {0, 1}) by identifying 0, * *1. That is, we have the analogous pushouts {0, 1}___//_{0} {0, 1}____//_{0} | | ss1 | | | | 7! | | (* *1) fflffl| |fflffl fflffl| fflffl| [0, 1]____//_S1 I _______//_C the first in the category of spaces, the second in the category of groupoids, a* *nd ss1 takes the first with the appropriate sets of base points, to the second. The aim is * *to have a similar argument at the module level for ssn, n > 1. One of the problems of obtaining such results in higher homotopy theory is t* *hat low dimensional identifications of spaces usually affect high dimensional homotopy * *invariants. To cope with this fact, the algebraic data for the values of the functor must* * have structure interacting from low to high dimensions, in order to model how the sp* *aces are glued together. An example of such dimensional interaction is the operation of the fundament* *al group on the higher homotopy groups, which fascinated the early workers in homotopy t* *heory (private communication, J.H.C. Whitehead, 1957). This operation can be seen as * *necess- itated by the dependence of these groups on the base point, but is somewhat neg* *lected in algebraic topology, perhaps for perceived deficiencies in modes of calculati* *ons. A recognised reason for the difficulty of traditional invariants such as hom* *otopy groups in dealing with gluing spaces, and so obtaining colimit calculations, is the fa* *ilure of excision. If X is the union of open sets U, V with intersection W then the in* *clusion of pairs " : (V, W ) ! (X, U) _____________ 2This name for certain generalisations of van Kampen's Theorem was recently * *suggested by Jim Stasheff. 3 does not induce an isomorphism of relative homotopy groups as it does for singu* *lar ho- mology. However there is a residue of homotopical excision which can be deduced* * from a HHvKT, as in [BH81b ]; it requires connectivity conditions and gives informatio* *n in only the critical dimension, but this information involves the additional structure * *of operations. We state it at this stage for a single base point, as follows: Homotopical Excision Theorem (HET) If U, V and W = U \ V are path connected with base point a 2 W , and (V, W ) is (n - 1)-connected, n > 3, then (X, U) is* * (n - 1)- connected and the excision map "* : ssn(V, W, a) ! ssn(X, U, a) presents the ss1(U, a)-module ssn(X, U, a) as induced by the morphism of fundam* *ental groups ss1(W, a) ! ss1(U, a) from the ss1(W, a)-module ssn(V, W, a). The same * *holds for n = 2 with `module' replaced by `crossed module'. We show that this theorem implies the classical Relative Hurewicz Theorem, (* *Corollary 9.11) for which in more usual terms see for example [Whi78 ]. The case n = 2 im* *plies a deep theorem of Whitehead on free crossed modules (Corollary 9.8) (of which [Bro80 ]* * explains the original proof), and indeed allows the more general calculation of ss2(X [ * *CA, X, a) as a crossed ss1(X, a)-module (Corollary 9.13). That result is applied in [BW95 ,* * BW03 ] to calculate crossed modules representing some homotopy 2-types of mapping cones o* *f maps of classifying spaces of groups. The notion of inducing used here is both an indication of the existence of s* *ome uni- versal properties in higher homotopy theory, and also of the tools needed to ex* *ploit such properties as there are. In the case n > 2 the inducing construction is well k* *nown for modules over groups: if M is a module over the group G, and f : G ! H is a morp* *hism of groups, then the induced module f*(M) is isomorphic to M ZG ZH. For n = 2, * *the analogous construction is non-Abelian. We give the approach of fibred and cofibred categories to this inducing cons* *truction. Also the following simple example of algebraic modelling of a gluing process* * suggests a further generalisation is needed. Consider the following maps involving the * *n-sphere Sn, the first an inclusion and the second an identification: p n 1 Sn -i! Sn _ [0, 1] -! S _ S . (* *2) Here n > 2, the base point 0 of Sn is in the second space identified with 0, an* *d the map p identifies 0, 1 to give the circle S1. This clearly requires an algebraic theory dealing with many base points, and* * so a generalisation from groups to groupoids. The geometry of the base points and t* *heir interconnections can then play a role in the theory and applications. This programme has proved successful for the algebraic data of crossed compl* *exes over groupoids, or equivalent structures, and for catn-groups, or, equivalently, cro* *ssed n-cubes of groups. In both categories there is a HHvKT asserting that a homotopically d* *efined functor preserves certain colimits, and this leads to new calculations of hom* *otopical invariants. 4 The calculation of colimits in these algebraic categories usually requires w* *orking up in dimensions. So it is useful to consider for these hierarchical structures t* *he forgetful functors from higher complex structures to lower structures. The prototype again is groupoids, with the object functor from groupoids to * *sets. The fibre of this functor over a set I is written Gpd I: the objects of this c* *ategory are groupoids with object set I and morphisms are functors which are the identity o* *n I. This fibre has nice properties, and is what is called protomodular; in rough te* *rms it has properties analogous to those of the category of groups. In particular, there i* *s a notion of normal subgroupoid, and of coproduct, so that colimits can be calculated in * *Gpd Ias quotients of a coproduct by a normal subobject. However the whole category Gpd is of interest for homotopical modelling. The* * functor Ob : Gpd ! Sethas the properties of being a fibration and a cofibration in the* * sense of Grothendieck, [Gro68 ]. So a function u : I ! J of sets gives rise to a pair o* *f functors u* : GpdJ ! GpdI, known often as pullback, and u* : GpdI ! GpdJ, which could be* * called `pushin' or `push forward', such that u* is left adjoint to u*. So for G 2 Gpd* * Iwe have a morphism u0: G ! u*G over u with a universal property which is traditionally * *called `cocartesian', and it is also said that u*G is the object induced from X by u. It is interesting to see that the main parts of the pattern for groupoids go* *es over to the general case. Our categorical results are for a fibration of categories :* * X ! B. The first and a main result (Theorem 2.7) is that the inclusion XI ! X of a fibre X* *I of preserves colimits of connected diagrams. Our second set of results relates pus* *houts in X to the construction of the functors u* : XI ! XJ for u a morphism in B (Proposi* *tion 4.2). Finally, we show how the combination of these results uses the computation of c* *olimits in B and in each XI to give the computation of colimits in X (Theorem 4.4). These general results are developed in the spirit of `categories for the wor* *king math- ematician' in sections 2 to 4. We illustrate the use of these results for homo* *topical calculations not only in groupoids (section 5), but also for crossed modules (s* *ection 7) and for modules, in both cases over groupoids. Finally we give a brief account * *of some relevance to crossed squares (section 10), as an indication of a more extensive* * theory, and in which these ideas need development. We are grateful to Thomas Streicher for his Lecture Notes [Str99], on which * *the foll- owing account of fibred categories is based, and for helpful comments leading t* *o im- provements in the proofs. For further accounts of fibred and cofibred categori* *es, see [Gra66 , Bor94, Vis04] and the references there. The first paper gives analogie* *s between fibrations of categories and Hurewicz fibrations of spaces. 2. Fibrations of categories We recall the definition of fibration of categories. 2.1. Definition. Let : X ! B be a functor. A morphism ' : Y ! X in X over u := (') is called cartesian if and only if for all v : K ! J in B and ` : Z !* * X with (`) = uv there is a unique morphism _ : Z ! Y with (_) = v and ` = '_. 5 This is illustrated by the following diagram: __`__________________________________________* *_____________ _______________________________________________* *_____________________________________________________________ ______________%%__________________________________* *_________________________________________________________ Z ______//___________Y//_X| _ ' | __uv______________||__________________________* *_______________________________________________________ _____________________|___________________________* *_____________________________________________________________________________* *_________________________ ______//______$$______fflffl|___________________//_ K v J u I 2 It is straightforward to check that cartesian morphisms are closed under com* *position, and that ' is an isomorphism if and only if ' is a cartesian morphism over an i* *somorphism. A morphism ff : Z ! Y is called vertical (with respect to ) if and only if * * (ff) is an identity morphism in B. In particular, for I 2 B we write XI, called the fibre * *over I, for the subcategory of X consisting of those morphisms ff with (ff) = idI. 2.2. Definition. The functor : X ! B is a fibration or category fibred over * *B if and only if for all u : J ! I in B and X 2 XI there is a cartesian morphism ' : Y !* * X over u: such a ' is called a cartesian lifting of X along u. * * 2 Notice that cartesian liftings of X 2 XI along u : J ! I are unique up to ve* *rtical isomorphism: if ' : Y ! X and _ : Z ! X are cartesian over u, then there exist vertical arrows ff : Z ! Y and fi : Y ! Z with 'ff = _ and _fi = ' respective* *ly, from which it follows by cartesianness of ' and _ that fffi = idY and fiff = i* *dZ as _fiff = 'ff = _ = _ idYand similarly 'fiff = ' idY. 2.3. Example. The forgetful functor, Ob : Gpd ! Set, from the category of grou* *poids to the category of sets is a fibration. We can for a groupoid G over I and fun* *ction u : J ! I define the cartesian lifting ' : H ! G as follows: for j, j02 J set H(j, j0) = {(j, g, j0) | g 2 G(uj, uj0)} with composition (j1, g1, j01)(j, g, j0) = (j1, g1g, j0), with ' given by '(j, g, j0) = g. The universal property is easily verified. The* * groupoid H is usually called the pullback of G by u. This is a well known construction * *(see for example [Mac05 , x2.3], where pullback by u is written u##). * * 2 2.4. Definition. If : X ! B is a fibration, then using the axiom of choice f* *or classes we may select for every u : J ! I in B and X 2 XI a cartesian lifting of X alon* *g u uX : u*X ! X. Such a choice of cartesian liftings is called a cleavage or splitting of . 6 If we fix the morphism u : J ! I in B, the splitting gives a so-called reind* *exing functor u* : XI ! XJ defined on objects by X 7! u*X and the image of a morphism ff : X ! Y is u*ff * *the unique vertical arrow commuting the diagram: _____uX______// u*XO X O || u*ffO ff|| fflfflO fflffl| u*Y _____________//Y uY 2 We can use this reindexing functor to get an adjoint situation for each u : * *J ! I in B. 2.5. Proposition. Suppose : X ! B is a fibration of categories, u : J ! I in* * B, and a reindexing functor u* : XI ! XJ is chosen. Then there is a bijection XJ(Y, u*X) ~=Xu(Y, X) natural in Y 2 XJ, X 2 XI where Xu(Y, X) consists of those morphisms ff 2 X(Y, * *X) with (ff) = u. Proof This is just a restatement of the universal properties concerned. In general for composable maps u : J ! I and v : K ! J in B it does not hold* * that v*u* = (uv)* as may be seen with the fibration of Example 2.3. Nevertheless there is a natur* *al equiv- alence cu,v: v*u* ' (uv)* as shown in the following diagram in which the full a* *rrows are cartesian and where (cu,v)X is the unique vertical arrow making the diagram com* *mute: *X _____vu________//_* v*u*XO u X O || (cu,v)X~=O |uX fflfflO |fflffl| (uv)*X ________________//X (uv)X Let us consider this phenomenon for our main examples: 7 2.6. Example. 1.- Typically, for B = @1 : B2 ! B, the fundamental fibration * *for a category with pullbacks, we do not know how to choose pullbacks in a functorial* * way. 2.- In considering groupoids as a fibration over sets, if u : J ! I is a map* *, we have a reindexing functor, also called pullback, u* : Gpd I! Gpd J. We notice that * *v*u*Q is naturally isomorphic to, but not identical to (uv)*Q. * * 2 A result which aids understanding of our calculation of pushouts and some ot* *her colimits of groupoids, modules, crossed complexes and higher categories is the * *following. Recall that a category C is connected if for any c, c02 C there is a sequence * *of objects c0 = c, c1, . .,.cn-1, cn = c0such that for each i = 0, . .,.n-1 there is a mor* *phism ci! ci+1 or ci+1! ci in C. The sequence of morphisms arising in this way is called a zig* *-zag from c to c0of length n. 2.7. Theorem. Let : X ! B be a fibration, and let J 2 B. Then the inclusi* *on iJ : XJ ! X preserves colimits of connected diagrams. Proof We will need the following diagrams: T_(c)__ T (c) __________________________________FF______________________* *EE __________________________________|`(c)FF_________________* *_________________|`(c)EE __________________________________FFF______________________* *_____________EEE ___________________________________(c)FF___________________* *_________________(c)EE __________________________________FF________________________* *__________EE __________________________fflffl|""F________________________* *__________________fflffl|""E T (f)____________________________________________________________* *Y'_//_Xfl(c)__________________________________Y'_//_X;; __________________________________OO<_____//Y__'__//______________X?? >_ """ > """ `0 >>OO""""~ X0 By the given property of _ there is a unique morphism _0: Y ! X0 in XJ such * *that _0_ = `0. By the cartesian property of ~, there is a unique morphism '0in XJ su* *ch that ~'0= '. Then ~_0_ = ~`0= ` = '_ = ~'0_. 11 By the cartesian property of ~, and since _0_, '0_ are over uv, we have _0_ = '* *0_. By the given property of _, and since '0, _0 are in XJ, we have '0= _0. So ' = ~_0* *, and this proves uniqueness. But we have already checked that ~_0_ = `, so we are done. The following Proposition allows us to prove that a fibration is also a cofi* *bration by constructing the adjoints u* of u* for every u. 3.8. Proposition. Let : X ! B be a fibration of categories. Let u : J ! I * *have reindexing functor u* : XI ! XJ. Then the following are equivalent: (i)For all Y 2 XJ, there is a morphism uY : Y ! u*Y which is cocartesian over* * u; (ii)there is a functor u* : XJ ! XI which is left adjoint to u*. Proof That (ii) implies (i) is clear, using Proposition 3.7, since the adjoin* *tness gives the bijection required for the cocartesian property. To prove that (i) implies (ii) we have to check that the allocation Y 7! u*(* *Y ) gives a functor that is adjoint to u*. As before the adjointness comes from the coca* *rtesian property. We leave to the reader the check the details of the functoriality of u*. To end this section, we give a useful result on compositions. 3.9. Proposition. The composition of fibrations, (cofibrations), is also a f* *ibration (cofibration). Proof We leave this as an exercise. 4. Pushouts and cocartesian morphisms Here is a small result which we use in this section and section 9, as it applie* *s to many examples, such as the fibration Ob : Gpd ! Set. 4.1. Proposition. Let : X ! B be a functor that has a left adjoint D. Then f* *or each K 2 Ob B, D(K) is initial in XK . In fact if u : K ! J in B, then for any X 2 X* *J there is a unique morphism "K : DK ! X over u. Proof This follows immediately from the adjoint relation Xu(DK, X) ~=B(K, X)* * for all X 2 Ob XJ. Special cases of cocartesian morphisms are used in [Bro06 , BH78 , BH81b ], * *and we review these in section 9. A construction which arises naturally from the vario* *us Higher Homotopy van Kampen theorems is given a general setting as follows: 12 4.2. Theorem. Let : X ! B be a fibration of categories which has a left adj* *oint D. Suppose that X admits pushouts. Let v : K ! J be a morphism in B, and let Z 2 X* *K . Then a cocartesian lifting _ : Z ! Y of v is given precisely by the pushout in * *X: D(v) D(K) _______//D(J) | | "K || |"J| (* **) | | fflffl| |fflffl Z _________//_Y _ Proof Suppose first that diagram (*) is a pushout in X. Let u : J ! I in B an* *d let ` : Z ! X satisfy (`) = uv, so that (X) = I. Let f : D(J) ! X be the adjoint * *of u : J ! (X). D(v) D(K) ____//_D(J) 33 "K || "J||333 fflffl|_ fflffl|f333 Z SSSSS___//Y___33_ SSSSS _'_33_ (** **) SSSSS__33____ ` SSSS3ssss))S""_____ X K ___v___//_J_u__//_I Then (fD(v)) = uv = (`"K ) and so by Proposition 4.1, fD(v) = `"K . The pu* *shout property implies there is a unique ' : Y ! X such that '_ = ` and '"J = f. This* * last condition gives (') = u since u = (f) = ('"J) = (') idJ= ('). For the converse, we suppose given f : D(J) ! X and ` : Z ! X such that `"K = fD(v). Then (`) = uv and so there is a cocartesian lifting ' : Y ! X of u. * *The additional condition '"J = f is immediate by Proposition 4.1. 4.3. Corollary. Let : X ! B be a fibration which has a left adjoint and sup* *pose that X admits pushouts. Then is also a cofibration. In view of the construction of hierarchical homotopical invariants as colimi* *ts from the HHvKTs in [BH81b ] and [BL87a ], the following is worth recording, as a consequ* *ence of Theorem 2.7. 4.4. Theorem. Let : X ! B be fibred and cofibred. Assume B and all fibres X* *I are cocomplete. Let T : C ! X be a functor from a small connected category. Then co* *limT exists and may be calculated as follows: (i)First calculate I = colim( T ), with cocone fl : T ) I; (ii)for each c 2 C choose cocartesian morphisms fl0(c) : T (c) ! X(c), over f* *l(c) where X(c) 2 XI; 13 (iii)make c 7! X(c) into a functor X : C ! XI, so that fl0 becomes a natural t* *ransfor- mation fl0: T ) X; (iv)form Y = colimX 2 XI with cocone ~ : X ) Y . Then Y with ~fl0: T ) Y is colimT . Proof We first explain how to make X into a functor. We will in stages build up the following diagram: j ____________________________________________* *_____________________________________________________________________________* *_____________________________________________________________________________* *_____________________________________________________________________________* *__________________________________________@ __________________________________________________* *_____________________________________________________________________________* *_____________________________________________________________________________* *_____________________________ ______________________________________________________* *_____________________________________________________________________________* *___________________________________~(c) T (c)K_______0_________//X(c)O_____//Y__o___))//__________* *_____Z KKK fl (c) O || KKK OX(f) |1 T (f)KK%% fflffl fflffl| (* *3) T (c0)__0__0//_X(c0)_0__//Y___0__//____________* *___Z | fl (c ) ~(c ) o || fflffl| _________// _________//_________//_______// K J 0 I I H T (f) fl(c ) 1 w Let f : c ! c0 be a morphism in C, K = T (c), J = T (c0). By cocartesianness* * of fl0(c), there is a unique vertical morphism X(f) : X(c) ! X(c0) such that X(f)f* *l0(c) = fl0(c0)T (f). It is easy to check, again using cocartesianness, that if furthe* *r g : c0 ! c00, then X(gf) = X(g)X(f), and X(1) = 1. So X is a functor and the above diagram sh* *ows that fl0 becomes a natural transformation T ) X. Let j : T ) Z be a natural transformation to a constant functor Z, and let * *(Z) = H. Since I = colim( T ), there is a unique morphism w : I ! H such that wfl = (j). By the cocartesian property of fl0, there is a natural transformation j0: X * *) Z such that j0fl0= j. Since Y is also a colimit in X of X, we obtain a morphism o : Y ! Z in X suc* *h that o~ = j0. Then o~fl0= j0fl0= j. Let o0 : Y ! Z be another morphism such that o0~fl0 = j. Then (o) = (o0) =* * w, since I is a colimit. Again by cocartesianness, o0~ = o~. By the colimit proper* *ty of Y , o = o0. This with Theorem 4.4 shows how to compute colimits of connected diagrams in* * the examples we discuss in sections 5 to 10, and in all of which a van Kampen type * *theorem is available giving colimits of algebraic data for some glued topological data. 4.5. Corollary. Let : X ! B be a functor satisfying the assumptions of theo* *rem 4.4. Then X is connected complete, i.e. admits colimits of all connected diagra* *ms. 5. Groupoids bifibred over sets We have already seen in Example 2.3 that the functor Ob : Gpd ! Set is a fibr* *ation. It also has a left adjoint D assigning to a set I the discrete groupoid on I, a* *nd a right 14 adjoint assigning to a set I the codiscrete groupoid on I. It follows from general theorems on algebraic theories that the category Gpd* * is co- complete, and in particular admits pushouts, and so it follows from previous re* *sults that Ob : Gpd ! Setis also a cofibration. A construction of the cocartesian lifting* *s of u : I ! J for G a groupoid over I is given in terms of words, generalising the constructi* *on of free groups and free products of groups, in [Hig71, Bro06]. In these references the * *cocartesian lifting of u to G is called a universal morphism, and is written u* : G ! Uu(G)* *. This construction is of interest as it yields a normal form for the elements of Uu(G* *), and hence u* is injective on the set of non-identity elements of G. A homotopical application of this cocartesian lifting is the following theor* *em on the fundamental groupoid. It shows how identification of points of a discrete subse* *t of a space can lead to `identifications of the objects' of the fundamental groupoid: 5.1. Theorem. Let (X, A) be a pair of spaces such that A is discrete and the i* *nclusion A ! X is a closed cofibration. Let f : A ! B be a function to a discrete space * *B. Then the induced morphism ss1(X, A) ! ss1(B [f X, B) is the cocartesian lifting of f. This theorem immediately gives the fundamental group of the circle S1 as the* * infinite cyclic group C, since S1 is obtained from the unit interval [0, 1] by identifyi* *ng 0 and 1, as shown in the Introduction in diagram (1). The theorem is a translation of [Bro0* *6 , 9.2.1], where the words `universal morphism' are used instead of `cocartesian lifting'.* * Section 8.2 of [Bro06 ] shows how free groupoids on directed graphs are obtained by a gener* *alisation of this example. The calculation of colimits in a fibre GpdI is similar to that in the catego* *ry of groups, since both categories are protomodular, [BB04 ]. Thus a colimit is calculated * *as a quo- tient of a coproduct, where quotients are themselves obtained by factoring by a* * normal subgroupoid. Quotients are discussed in [Hig71, Bro06]. Theorem 4.4 now shows how to compute general colimits of groupoids. We refer again to [Hig71, Bro06 ] for further developments and applications * *of the algebra of groupoids; we generalise some aspects to modules, crossed modules an* *d crossed complexes in the next sections. 6. Groupoid modules bifibred over groupoids Modules over groupoids are a useful generalisation of modules over groups, and * *also form part of the basic structure of crossed complexes. Homotopy groups ssn(X; X0), n* * > 2, of a space X with a set X0 of base points form a module over the fundamental group* *oid ss1(X, X0), as do the homotopy groups ssn(Y, X : X0), n > 3, of a pair (Y, X). 6.1. Definition. A module over a groupoid is a pair (M, G), where G is a grou* *poid with set of objects G0, M is a totally disconnected abelian groupoid with the s* *ame set of 15 objects as G, and with a given action of G on M. Thus M comes with a target fun* *ction t : M ! G0, and each M(x) = t-1(x), x 2 G0, has the structure of Abelian group.* * The G-action is given by a family of maps M(x) x G(x, y) ! M(y) for all x, y 2 G0. These maps are denoted by (m, p) 7! mp and satisfy the usual* * properties, 0 (pp0) 0 p p 0p i.e. m1 = m, (mp)p = m and (m + m ) = m + m , whenever these are defined.* * In particular, any p 2 G(x, y) induces an isomorphism m 7! mp from M(x) to M(y). A morphism of modules is a pair (`, f) : (M, G) ! (N, H), where f : G ! H and ` : M ! N are morphisms of groupoids and preserve the action. That is, ` is giv* *en by a family of group morphisms `(x) : M(x) ! N(f(x)) for all x 2 G0 satisfying `(y)(mp) = (`(x)(m))f(p), for all p 2 G(x, y), m 2 M(* *x). This defines the category Mod having modules over groupoids as objects and * *the morphisms of modules as morphisms. If (M, G) is a module, then (M, G)0 is defin* *ed to be G0. We have a forgetful functor M : Mod ! Gpd in which (M, G) 7! G. 6.2. Proposition. The forgetful functor M : Mod ! Gpd has a left adjoint an* *d is fibred and cofibred. Proof The left adjoint of M assigns to a groupoid G the module written 0 ! G* * which has only the trivial group over each x 2 G0. Next, we give the pullback construction to prove that M is fibred. This is* * entirely analogous to the group case, but taking account of the geometry of the groupoid. So let v : G ! H be a morphism of groupoids, and let (N, H) be a module. We * *define (M, G) = v*(N, H) as follows. For x 2 G0 we set M(x) = {x} x N(vx) with additi* *on given by that in N(vx). The operation is given by (x, n)p = (y, nvp) for p 2 G(* *x, y). The construction of N = v*(M, G) for (M, G) a G-module is as follows. For y 2 H0 we let N(y) be the abelian group generated by pairs (m, q) with m* * 2 M, q 2 H, and t(q) = y, s(q) = v(t(m)), so that N(y) = 0 if no such pairs exist. The o* *peration 0 0 0 0 of H on N is given by (m, q)q = (m, qq ), addition is (m, q) + (m , q) = (m + m* * , q) and the relations imposed are (mp, q) = (m, v(p)q) when these make sense. The cocar* *tesian morphism over v is given by _ : m 7! (m, 1vt(m)). 6.3. Remark. We will discuss the relation between a module over a groupoid and* * the restriction to the vertex groups in section 8 in the general context of crossed* * complexes. However it is useful to give the general situation of many base points to descr* *ibe the relative homotopy group ssn(X, A, a0) when X is obtained from A by adding n-cel* *ls at various base points. The natural invariant to consider is then ssn(X, A, A0) wh* *ere A0 is an appropriate set of base points. We now describe free modules over groupoids in terms of the inducing constru* *ction. The interest of this is two fold. Firstly, induced modules arise in homotopy th* *eory from a 16 HHvKT, and we get new proofs of results on free modules in homotopy theory. Sec* *ondly, this indicates the power of the HHvKT since it gives new results. 6.4. Definition. Let Q be a groupoid. A free basis for a module (N, Q) consis* *ts of a pair of functions tB : B ! Q0, i : B ! N such that tN i = tB and with the uni* *versal property that if (L, Q) is a module and f : B ! L is a function such that tLf =* * tN then there is a unique Q-module morphism ' : N ! L such that 'i = f. 6.5. Proposition. Let B be an indexing set, and Q a groupoid. The free Q-modu* *le (F M(t), Q) on t : B ! Q0 may be described as the Q-module induced by t : B ! Q from the discrete free module ZB = (Z x B, B) on B, where B denotes also the di* *screte groupoid on B. Proof This is a direct verification of the universal property. 6.6. Remark. Proposition 3.7 shows that the universal property for a free modu* *le can also be expressed in terms morphisms of modules (F M(t), Q) ! (L, R). 7. Crossed modules bifibred over groupoids Out homotopical example here is the family of second relative homotopy groups o* *f a pair of spaces with many base points. A crossed module over a groupoid, [BH81a ], consists first of a morphism of * *groupoids ~ : M ! P of groupoids with the same set P0 of objects such that ~ is the ident* *ity on objects, and M is a family of groups M(x), x 2 P0; second, there is an action o* *f P on the family of groups M so that if m 2 M(x) and p 2 P (x, y) then mp 2 M(y). This ac* *tion must satisfy the usual axioms for an action with the additional properties: CM1) ~(mp) = p-1~(m)p, and CM2) m-1nm = n~m for all p 2 P , m, n 2 M such that the equations make sense. These form the obj* *ects of the category XMod in which a morphism is a commutative square of morphisms of * *groups g M ____//_N ~ || || fflffl|fflffl| P _____//Q f which preserve the action in the sense that g(mp) = (gm)fp whenever this makes * *sense. The category XMod is equivalent to the well known category 2-Gpd of 2-group* *oids, [BH81c ]. However the advantages of XMod over 2-groupoids are: o crossed modules are closer to the classical invariants of relative homotop* *y groups; o the notion of freeness is clearer in XMod and models a standard topologic* *al situation, that of attaching 1- and 2-cells; 17 o the category XMod has a monoidal closed structure which helps to define a* * notion of homotopy; these constructions are simpler to describe in detail than th* *ose for 2-groupoids, and they extend to all dimensions. We have a forgetful functor 1 : XMod ! Gpd which sends (M ! P ) 7! P . Our* * first main result is: 7.1. Proposition. The forgetful functor 1 : XMod ! Gpd is fibred and has a* * left adjoint. Proof The left adjoint of 1 assigns to a groupoid P the crossed module 0 ! P* * which has only the trivial group over each x 2 P0. Next, we give the pullback construction to prove that 1 is fibred. So let f* * : P ! Q be a morphism of groupoids, and let : N ! Q be a crossed module. We define M = * **(N) as follows. For x 2 P0 we set M(x) to be the subgroup of P (x) x N(fx) of elements (p, n* *) such that fp = n. If p1 2 P (x, x0), n 2 N(fx) we set (p, n)p1 = (p-11pp1, nf(p1))* *, and let ~ : (p, n) 7! p. We leave the reader to verify that this gives a crossed module* *, and that the morphism (p, n) 7! n is cartesian. The following result in the case of crossed modules of groups appeared in [B* *H78 ], described in terms of the crossed module @ : u*(M) ! Q induced from the crossed module ~ : M ! P by a morphism u : P ! Q. 7.2. Proposition. The forgetful functor 1 : XMod ! Gpd is cofibred. Proof We prove this by a direct construction. Let ~ : M ! P be a crossed module, and let f : P ! Q be a morphism of groupo* *ids. The construction of N = f*(M) and of @ : N ! Q requires just care to the geomet* *ry of the partial action in addition to the construction for the group case initiated* * in [BH78 ] and pursued in [BW03 ] and the papers referred to there. Let y 2 Q0. If there is no q 2 Q from a point of f(P0) to y, then we set N(y* *) to be the trivial group. Otherwise, we define F (y) to be the free group on the set of pairs (m, q) s* *uch that 0 * * 0 m 2 M(x) for some x 2 P0 and q 2 Q(fx, y). If q02 Q(y, y0) we set (m, q)q = (m,* * qq ). We define @0 : F (y) ! Q(y) to be (m, q) 7! q-1(fm)q. This gives a precrossed m* *odule over @ : F ! Q, with function i : M ! F given by m 7! (m, 1) where if m 2 M(x) * *then 1 here is the identity in Q(fx). We now wish to change the function i : M ! F to make it an operator morphism. For this, factor F out by the relations (m, q)(m0, q)= (mm0, q), (mp, q)= (m, (fp)q), whenever these are defined, to give a projection F ! F 0and i0 : M ! F 0. As* * in the group case, we have to check that @0 : F ! Q induces @00: F 0! H making thi* *s a 18 precrossed module. To make this a crossed module involves factoring out Peiffer* * elements, whose theory is as for the group case in [BH82 ]. This gives a crossed module m* *orphism (', f) : (M, P ) ! (N, Q) which is cocartesian. We recall the algebraic origin of free crossed modules, but in the groupoid * *context. Let P be a groupoid, with source and target functions written s, t : P ! P0* *. A subgroupoid N of P is said to be normal in P , written N . P , if N is wide in * *P , i.e. N0 = P0, and for all x, y 2 P0 and a 2 P (x, y), a-1N(x)a = N(y). If N is also * *totally intransitive, i.e. N(x, y) = ; when x 6= y, as we now assume, then the quotient* * groupoid P=N is easy to define. (It may also be defined in general but we will need only* * this case.) Now suppose given a family R(x), x 2 P0 of subsets of P (x). Then the norma* *liser NP(R) of R is well defined as the smallest normal subgroupoid of P containing a* *ll the sets R(x). Note that the elements of NP(R) are all consequences of R in P , i.e* *. all well defined products of the form c = (r"11)a1. .(.r"nn)an, "i= 1, ai2 P, n > 0 (* *4) and where ba denotes a-1ba. The quotient P=NP(R) is also written P=R, and calle* *d the quotient of P by the relations r = 1, r 2 R. As in group theory, we need also to allow for repeated relations. So we supp* *ose given a set R and a function ! : R ! P such that s! = t! = fi, say. This `base point * *function', saying where the relations are placed, is a useful part of the information. We now wish to obtain `syzygies' by replacing the normal subgroupoid by a `f* *ree object' on the relations ! : R ! P . As in the group case, this is done using f* *ree crossed modules. 7.3. Remark. There is a subtle reason for this use of crossed modules. A no* *rmal subgroupoid N of P (as defined above) gives a quotient object P=N in the catego* *ry GpdX of groupoids with object set X = P0. Alternatively, N defines a congruence on P* * , which is a particular kind of equivalence relation. Now an equivalence relation is in* * general a particular kind of subobject of a product, but in this case, we must take the p* *roduct in the category Gpd X. As a generalisation of this, one should take a groupoid* * object in the category Gpd X. Since these totally disconnected normal subgroupoids det* *ermine equivalence relations on each P (x, y) which are congruences, it seems clear th* *at a groupoid object internal to GpdX is equivalent to a 2-groupoid with object set X. 7.4. Definition. A free basis for a crossed module @ : C ! P over a groupoid * *P is a set R, function fi : R ! P0 and function i : R ! C such that i(r) 2 C(fir), r 2* * R, with the universal property that if ~ : M ! P is a crossed module and ` : R ! M a fu* *nction over the identity on P0 such that ~` = @i, then there is a unique morphism of c* *rossed P -modules ' : C ! M such that 'i = `. 7.5. Example. Let R be a set and fi : R ! P0 a function. Let id : F1(R) ! F2(* *R) be the identity crossed module on two copies of F (R), the disjoint union of co* *pies C(r) of the infinite cyclic group C with generator cr 2 C(r). Thus F2(R) is a totally i* *ntransitive 19 groupoid with object set R. Let i : R ! F1(R) be the function r 7! cr. Let fi :* * R ! R be the identity function. Then id: F1(R) ! F2(R) is the free crossed module on * *i. The verification of this is simple from the diagram __`_______________________________________* *_____________________________________________________________________________* *_____ _____________________________________________* *_____________________________________________________________________________* *_________________________________________________________________________ ________________&&______________________________* *______________________________________________________________________ R ___i//_F1(R)````//M | f | id| |~ fflffl| fflffl| F2(R) ____//_F2(R) id The morphism f simply maps the generator cr to `r. 7.6. Proposition. Let R be a set, and ~ : M ! P a crossed module over the grou* *poid P . Let fi : R ! P0 be a function. Then the functions i : R ! M such that s~ = * *t~ = fi are bijective with the crossed module morphisms (f, g) f F1(R) _____//M | || id| |~ fflffl| fflffl| F2(R) __g__//P such that sg = fi. Further, the free crossed module @ : C(!) ! P on a function ! : R ! P such t* *hat s! = t! = fi is determined as the crossed module induced from id : F1(R) ! F2(R* *) by the extension of ! to the groupoid morphism F2(R) ! P . Proof The first part is clear since g = ~f and f and i are related by f(cr) =* * i(r), r 2 R. The second part follows from the first part and the universal property of in* *duced crossed modules as shown in the following diagram: __`______________________________________* *_____________________________________________________________________________* *________________________________ _____________________________________________* *_____________________________________________________________________________* *_____________________________________________________________________________* *_________________________________ F1(R)_____//C(!)'```))//M | f | || id| |@ |~ fflffl| fflffl| fflffl| F2(R)___g__//P__=___//_P 8. Crossed complexes and an HHvKT Crossed complexes are analogous to chain complexes but also generalise groupoid* *s to all dimensions and with their base points and operations relate dimensions 0,1 and * *n. The structure and axioms for a crossed complex are those universally satisfied by t* *he main 20 topological example, the fundamental crossed complex X* of a filtered space X** *, where ( X*)1 is the fundamental groupoid ss1(X1, X0) and for n > 2 ( X*)n is the fami* *ly of relative homotopy groups ssn(Xn, Xn-1, x0) for all x0 2 X0, with associated ope* *rations of the fundamental groupoid and boundaries. Crossed complexes fit into our scheme of algebraic structures over a range o* *f dimensions satisfying a HHvKT in that the fundamental crossed complex functor : (filtered spaces) ! (crossed complexes) preserves certain colimits. We state a precise version below. A crossed complex C is in part a sequence of the form ffin ffin-1 ffi3 ffi2 . ._.___//Cn___//_Cn-1___//_......__//C2___//_C1 where all the Cn, n > 1, are groupoids over C0. Here ffi2 : C2 ! C1 is a crosse* *d module and for n > 3 (Cn, C1) is a module. The further axioms are that ffin is an operator* * morphism for n > 2 and that ffi2c operates trivially on Cn for c 2 C2 and n > 3. We ass* *ume the basic facts on crossed complexes as surveyed in for example [Bro99 , Bro04]. Th* *e category of crossed complexes is written Crs. A full exposition of the theory of crossed* * complexes will be given in [BHS08 ]. To state the Higher Homotopy van Kampen Theorem for relative homotopy groups, namely Theorem C of [BH81b , Section 5], we need the following definition: 8.1. Definition. A filtered space X* is said to be connected if the following * *conditions hold for each n > 0 : - '(X*, 0) : If r > 0, the map ss0X0 ! ss0Xr, induced by inclusion, is surjecti* *ve; i.e. X0 meets all path connected components of all stages of the filtration X*. -'(X*, n)(for n > 1): If r > n and x 2 X0, then ssn(Xr, Xn, x) = 0. * * 2 8.2. Theorem. [Higher Homotopy van Kampen Theorem] Let X* be a filtered space and suppose: (i)X is the union of the interiors of subspaces U, V ; (ii)the filtrations U*, V* and W*, formed by intersection with X*, and where * *W = U \V , are connected filtrations. Then (Conn) the filtration X* is connected, and (Pushout) the following diagram of morphisms of crossed complexes induced by in* *clusions W* _____// U* | | | | fflffl| fflffl| V* ____//_ X* is a pushout of crossed complexes. 21 8.3. Remark. The connectivity conclusion is significant, but not as important* * as the algebraic conclusion. This theorem is proved without recourse to traditional me* *thods of algebraic topology such as homology and simplicial approximation. Indeed, the * *impli- cations in dimension 2 are in general nonabelian and so unreachable by the trad* *itional abelian methods. Instead the theorem is proved using the construction of aeX*,* * the cu- bical homotopy !-groupoid of the filtered space X*, defined in dimension n to b* *e the set of filter homotopy classes of maps In*! X*. The properties of this construc* *tion en- able the proof of the 1-dimensional theorem van Kampen theorem to be generalise* *d to higher dimensions, and the theorem on crossed complexes is deduced using a non * *trivial equivalence between the two constructions. The paper [BH81b ] also proves a more general theorem, in which arbitrary un* *ions lead to a coequaliser rather than a pushout. The paper also assumes a J 0 cond* *ition on the filtered spaces; but this can be relaxed by the refined definition of ma* *king filter homotopies of maps In*! X* to be rel vertices, as has been advertised in [BH91 * *]. 2 8.4. Remark. Colimits of crossed complexes may be computed from the colimits * *of the groupoids, crossed modules and modules from which they are constituted. 8.5. Remark. A warning has to be given that some of the algebra is not as str* *aight- forward as that in traditional homological and homotopical algebra. For example* * in an abelian category, a pushout of the form A _____//0 | | i| | fflffl|fflffl| B __p__//C is equivalent to an exact sequence p A -i! B -! C ! 0. However in the category Mod *of modules over groups a pushout of the form (M, G)_____//(0, 1) | | i | | fflffl| fflffl| (N, H)__p__//(P, K) is equivalent to a pair of an exact sequence of groups p G -i! H -! K ! 1, and an exact sequence of induced modules over K p M ZG ZK -i! N ZH ZK -! P ! 0. This shows that pushouts give much more information in our case, but also shows* * that handling the information may be more difficult, and that misinterpretations cou* *ld lead to false conjectures or proofs. * * 2 22 8.6. Proposition. The truncation functor tr1 : Crs! Gpd, C 7! C1, is a bifibra* *tion. Proof The previous results give the constructions on modules and crossed modu* *les. The functoriality of these constructions give the construction of the boundary * *maps, and the axioms for all these follow. We will also need for later applications (Proposition 9.14) the relation of * *a connected crossed complex to the full, reduced (single vertex) crossed complex it contain* *s, analogous to the well known relation of a connected groupoid to any of its vertex groups. Recall that a codiscrete groupoid T is one on which T (x, y) is a singleton * *for all objects x, y 2 T0. This is called a tree groupoid in [Bro06 ]. Similarly, a codiscrete * *crossed complex T is one in which the groupoid T1 is codiscrete and which is trivial in higher * *dimensions. We follow similar conventions for crossed complexes as for groupoids in [Bro* *06 ]. Thus if D and E are crossed complexes, and S = D0 \ E0 then by the free product D * * *E we mean the crossed complex given by the pushout of crossed complexes S ______//_E | | | j| (* *5) fflffl| fflffl| D __i__//D * E where the set S is identified with the discrete crossed complex which is S in d* *imension 0 and trivial in higher dimensions. The following result is analogous to and inde* *ed includes standard facts for groupoids (cf. [Bro06 , 6.7.3, 8.1.5]). 8.7. Proposition. Let C be a connected crossed complex, let x0 2 C0 and let T* * be codiscrete wide subcrossed complex of C. Let C(x0) be the subcrossed complex of* * C at the base point x0. Then the natural morphism ' : C(x0) * T ! C determined by the inclusions is an isomorphism, and T determines a strong defor* *mation retraction r : C ! C(x0). Further, if f : C ! D is a morphism of crossed complexes which is the identi* *ty on C0 ! D0 then we can find a retraction s : D ! D(x0) giving rise to a pushout sq* *uare _r__//_ C C(x0) | | f|| |f0 (* *6) fflffl| fflffl| D _s__//_D(x0) in which f0 is the restriction of f. Proof Let i : C(x0) ! C, j : T ! C be the inclusions. We verify the universal p* *roperty of the free product. Let ff : C(x0) ! E, fi : T ! E be morphisms of crossed com* *plexes 23 agreeing on x0. Suppose g : C ! E satisfies gi = ff, gj = fi. Then g is determi* *ned on C0. Let c 2 C1(x, y). Then c= (ox)((ox)-1c(oy))(oy)-1 (* **) and so gc = g(ox)g((ox)-1c(oy))g(oy)-1 = fi(ox)ff((ox)-1c(oy))fi(oy)-1. If c 2 Cn(x), n > 2, then -1 c= (cox)(ox) (** **) and so -1 g(c)= ff(cox)fi(ox). This proves uniqueness of g, and conversely one checks that this formula define* *s a mor- phism g as required. In effect, equations (*) and (**) give for the elements of C normal forms in* * terms of elements of C(x0) and of T . This isomorphism and the constant map T ! {x0} determine the strong deformat* *ion retraction r : C ! C(x0). The retraction s is defined by the elements fo(x), x 2 C0, and then the diag* *ram (6) is a pushout since it is a retract of the pushout square __1_//_ C C f || f|| 2 fflffl|fflffl| D __1_//_D 9. Homotopical excision and induced constructions We now interpret the HHvKT (Theorem 8.2) when the filtrations have essentially * *just two stages. 9.1. Definition. By a based pair Xo = (X, X1; X0) of spaces we mean a pair (X,* * X1) of spaces together with a set X0 X1 of base points. For such a based pair and* * n > 2 we have an associated filtered space X[n]owhich is X0 in dimension 0, X1 in dim* *ensions 1 6 i < n and X for dimensions i > n. We write nXo for the crossed complex X[* *n]o. This crossed complex is trivial in all dimensions 6= 1, n, and in dimension n i* *s the family of relative homotopy groups ssn(X, X1, x0) for x0 2 X0, considered as a module * *(crossed module if n = 2) over the fundamental groupoid ss1(X1, X0). Colimits of such c* *rossed complexes are equivalent to colimits of the corresponding module or crossed mod* *ule. 2 The following is clear. 24 9.2. Proposition. If Xo = (X, X1; X0) is a based pair, and n > 2 then the asso* *ciated filtered space X[n]ois connected if and only if the based pair Xo is (n - 1)-co* *nnected, i.e. if: o X0 meets each path component of X1 and of X; o each path in X joining points of X0 is deformable in X rel end points to a* * path in X1; and o ssr(X, X1, x0) = 0 for all x0 2 X0 and 1 < r < n. The last part of the condition is of course vacuous if n = 2. * * 2 With this in mind, Theorem 8.2 may be restated as: 9.3. Theorem. Let (X, X1; X0) be a based pair of spaces and suppose: (i)X is the union of the interiors of subspaces U, V ; (ii)the based pairs (U, U1; U0), (V, V1; V0) and (W, W1; W0) formed by inters* *ection with (X, X1; X0), and where W = U \ V , are (n - 1)-connected. Then (Conn) (X, X1; X0) is (n - 1)-connected, and (Pushout) the following diagram of morphisms n(W, W1; W0) _____// n(U, U1; U0) | | | | fflffl| fflffl| n(V, V1; V0)_____// n(X, X1; X0) is a pushout of modules if n > 3 and of crossed modules if n = 2. Proof The important point is the equivalence between colimits of crossed comp* *lexes which for a given n > 2 have Ci= 0 for i 6= 0, 1, n or which are trivial in dim* *ensions > 2, and colimits of the corresponding modules or crossed modules. 9.4. Remark. It is not easy to see for Theorem 9.3 a direct proof in terms of * *modules or crossed modules, since one needs the intermediate structure between 0 and n * *to use the connectivity conditions. The cubical homotopy !-groupoid with connections aeX* * *is found convenient for this inductive process in [BH81b ]. There are two reasons for th* *is: cubical methods are convenient for constructing homotopies, and also for algebraic inve* *rses to subdivision. * * 2 We now concentrate on excision, since this gives rise to cocartesian morphis* *ms and so induced modules and crossed modules. 25 9.5. Theorem. [Homotopy Excision Theorem] Let the topological space X be the u* *nion of the interiors of sets U, V , and let W = U \ V . Let n > 2. Let W0 U0 U * *be such that the based pair (V, W ; W0) is (n - 1)-connected and U0 meets each path com* *ponent of U. Then (X, U; U0) is (n - 1)-connected, and the morphism of modules (crossed i* *f n = 2) ssn(V, W ; W0) ! ssn(X, U; U0) induced by inclusions is cocartesian over the morphism of fundamental groupoids ss1(W, W0) ! ss1(U, U0) induced by inclusion. Proof We deduce this excision theorem from the pushout theorem 9.3, applied t* *o the based pair (X, U; U0) and the following diagram of morphisms : n(W, W ; W0) ____//_ n(U, U; U0) (* *7) | | | | fflffl| fflffl| n(V, W ; W0)____// n(X, U; U0) This is a pushout of modules if n > 3 and of crossed modules if n = 2, by Theor* *em 9.3. However n(W, W ; W0) = (0, ss1(W, W0)), n(U, U; U0) = (0, ss1(U, U0)). So the theorem follows from Theorem 4.2 and our discussion of the examples of m* *odules and crossed modules. 9.6. Corollary. [Homotopical excision for an adjunction] Let i : W ! V be a cl* *osed cofibration and f : W ! U a map. Let W0 be a subset of W meeting each path comp* *onent of W and V , and let U0 be a subset of U meeting reach path component of U and * *such that f(W0) U0. Suppose that the based pair (V, W ) is (n - 1)-connected. Let X = U* * [f V . Then the based pair (X, U) is (n - 1)-connected and the induced morphism of mod* *ules (crossed if n = 2) ssn(V, W ; W0) ! ssn(X, U; U0) is cocartesian over the induced morphism of fundamental groupoids ss1(W, W0) ! ss1(U, U0). Proof This follows from Theorem 9.5 using mapping cylinders in a similar mann* *er to the proof of a corresponding result for the fundamental groupoid [Bro06 , 9.1.2* *]. That is, we form the mapping cylinder Y = M(f)[W . The closed cofibration assumption ens* *ures that the projection from Y to X = U [f V is a homotopy equivalence. 26 9.7. Remark. These methods were used in [BH78 ]. 2 9.8. Corollary. [Attaching n-cells] Let the space Y be obtained from the space* * X by attaching n-cells, n > 2, at a set of base points A of X, so that Y = X [f~en~* *, ~ 2 , where f~ : (Sn-1, 0) ! (X, A). Then ssn(Y, X; A) is isomorphic to the free ss1* *(X, A)- module (crossed if n = 2) on the characteristic maps of the n-cells. 9.9. Remark. The previous corollary for n = 2 was a theorem of J.H.C. Whitehe* *ad. An account of Whitehead's proof is given in [Bro80 ]. There are several other * *proofs in the literature but none give the more general homotopical excision result, theo* *rem 9.5. 2 9.10. Example. We can now explain the example in the Introduction. That Sn is (n - 1)-connected and ssn(Sn, 0) ~= Z follows by induction in the usual way fro* *m the homotopical excision theorem and the calculation ss1(S1, 0) ~= Z by the groupoi* *d van Kampen theorem. Applying the HET to writing Sn _ [0, 1] as a union of Sn and [0* *, 1] we get that ssn(Sn _ [0, 1], [0, 1]; {0, 1}) is the free ss1([0, 1], {0, 1})-modul* *e on one generator. Again applying the HET but now identifying 0, 1 we get that ss1(Sn _ S1, 0) is * *the free ss1(S1, 0)-module in one generator. * * 2 9.11. Corollary. [Relative Hurewicz Theorem] Let A ! X be a closed cofibration and suppose A is path connected and (X, A) is (n - 1)-connected. Then X [ CA is (n - 1)-connected and ssn(X [ CA, x) is isomorphic to ssn(X, A, x) factored by * *the action of ss1(A, x). We now point out that a generalisation of a famous result of Hopf, [Hop42 ],* * is a corollary of the relative Hurewicz theorem. The following for n = 2 is part of* * Hopf's result. The algebraic description of H2(G) which he gives for G a group is sho* *wn in [BH78 ] to follow from the HHvKT. 9.12. Proposition. [Hopf's theorem] Let (V, A) be a pair of pointed spaces suc* *h that: (i)ssi(A) = 0 for 1 < i < n; (ii)ssi(V ) = 0 for 1 < i 6 n; (iii)the inclusion A ! V induces an isomorphism on fundamental groups. Then the pair (V, A) is n-connected, and the inclusion A ! V induces an epimor* *phism HnA ! HnV whose kernel consists of spherical elements, i.e. of the image of ssn* *A under the Hurewicz morphism !n : ssn(A) ! Hn(A). Proof That (V, A) is n-connected follows immediately from the homotopy exact * *se- quence of the pair (V, A) up to ssn(V ). We now consider the next part of the * *exact homotopy sequence and its relation to the homology exact sequence as shown in t* *he 27 commutative diagram: ssn+1(V, A)@__//_ssn(A)___//ssn(V_)__//ssn(V, A) !n+1|| !n || || || fflffl| fflffl| fflffl| fflffl| Hn+1(V, A) ____//_Hn(A)___//_Hn(V_)__//_Hn(V, A) @0 i* The Relative Hurewicz Theorem implies that Hn(V, A) = 0, and that !n+1 is surje* *ctive. Also @ in the top row is surjective, since ssn(V ) = 0. It follows easily that * *the sequence ssn(A) ! Hn(A) ! Hn(V ) ! 0 is exact. There is a nice generalisation of Corollary 9.8, which so far has been prove* *d only as a deduction from a HHvKT. 9.13. Corollary. [Attaching cones] Let A be a space and let S be a set consis* *ting of one point in each path component of A. By CA, the cone on A, we mean the un* *ion of cones on each path component of A. Let f : A ! X be a map, and let S0 be the image of S by f. Then ss2(X [ CA, X; S0) is isomorphic to the ss1(X, S0)-crosse* *d module induced from the identity crossed module ss1(A, S) ! ss1(A, S) by the induced m* *orphism f* : ss1(A, S) ! ss1(X, S0). The paper [BW03 ] uses this result to give explicit calculations for the cr* *ossed modules representing the homotopy 2-types of certain mapping cones. We now explain the relevance to free crossed modules of Proposition 8.7, lea* *ving the module and other cases to the reader. 9.14. Proposition. Let X be a path connected space with base point a, and let* * Y = X [f~{e2~} be obtained by attaching cells by means of pointed maps f~ : (S1, 0)* * ! (X, a~), determining elements x~ 2 ss1(X, a~), ~ 2 . Let A = {a} [ {a~ | ~ 2 }. Let T * *be a tree groupoid in ss1(X, A) determining a retraction r : ss1(X, A) ! ss1(X, a). Then * *ss2(Y, X, a) is isomorphic to the free crossed ss1(X, a)-module on the elements r(x~). Proof We consider the following diagram: F r (0 ! ~Z) _______//_(0 ! P_)____//_(0 ! P (a)) | | | | | | (* *8) F fflffl|F fflffl| fflffl| ( ~Z ! ~Z) _____//(C( ) ! P )s__//_(F ! P (a)) The left hand square is the pushout defining the free crossed module C( ) ! P a* *s an induced crossed module. The right hand square is the special case of crossed mo* *dules of the retraction of Proposition 8.7, and so is also a pushout. Hence the composit* *e square is a pushout. Hence the crossed module F ! P (a) is the free crossed module as des* *cribed. 28 9.15. Remark. An examination of Whitehead's paper [Whi41 ], and the exposition* * of part of it in [Bro80 ], shows that the geometrical side of the last proposition* * is intrinsic to his approach. Of course a good proportion of Whitehead's work was devoted to ex* *tending ideas of combinatorial group theory to higher dimensions in combinatorial homot* *opy theory. The argument here is that this extension requires combinatorial groupoi* *d theory for good modelling of the geometry. * * 2 10. Crossed squares and triad homotopy groups In this section we give a brief sketch of the theory of triad homotopy groups, * *including the exact sequence relating them to homotopical excision, and show that the thi* *rd triad group forms part of a crossed square which, as an algebraic structure with link* *s over several dimensions, in this case dimensions 1,2,3, fits our criteria for a HHvK* *T. Finally we indicate a bifibration from crossed squares, so leading to the notion of ind* *uced crossed square, which is relevant to a triadic Hurewicz theorem in dimension 3. A triad of spaces (X : A, B; x) consists of a pointed space (X, x) and two p* *ointed subspaces (A, x), (B, x). Then ssn(X : A, B; x) is defined for n > 2 as the set* * of homotopy classes of maps (In : @-1In, @-2In; Jn-11,2) ! (X : A, B; x) where Jn-11,2denotes the union of the faces of In other than @-1In, @-2In. For * *n > 3 this set obtains a group structure, using the direction 3, say, which is Abelian for n >* * 4. Further there is an exact sequence ! ssn+1(X : A, B; x) ! ssn(A, C, x) -"! ssn(X, B, x) ! ssn(X : A, B; x)* *(!9) where C = A \ B, and " is the excision map. It was the fact that these groups m* *easure the failure of excision that was their main interest. However they do not shed* * light on the above Homotopical Excision Theorem 9.5. The third triad homotopy group fits into a diagram of possibly non-Abelian g* *roups ss3(X; A, B, x)___//_ss2(B, C, x) (X; A, B, x) := || || (1* *0) fflffl| fflffl| ss2(A, C, x)_____//ss1(C, x) in which ss1(C, x) operates on the other groups and there is also a function ss2(A, C, x) x ss2(B, C, x) ! ss3(X : A, B; x) known as the generalised Whitehead product. This diagram has structure and properties which are known as those of a cros* *sed square, [GWL81 , BL87a ], explained below, and so this gives a homotopical fun* *ctor : (based triads) ! (crossed squares). (1* *1) 29 A crossed square is a commutative diagram of morphisms of groups __~_//_ L M ~0|| |~| (1* *2) fflffl|fflffl| N _____//P together with left actions of P on L, M, N and a function h : M x N ! L satisfy* *ing a number of axioms which we do not give in full here. Suffice it to say that the * *morphisms in the square preserve the action of P , which acts on itself by conjugation; M* *, N act on each other and on L via P ; ~, ~0, ~, and ~~ are crossed modules; and h satis* *fies axioms reminiscent of commutator rules, summarised by saying it is a biderivation. Mor* *phisms of crossed squares are defined in the obvious way, giving a category XSq of crosse* *d squares. Let XMod 2 be the category of pairs of crossed modules ~ : M ! P, : N ! P * *(with P and ~, variable), and with the obvious notion of morphism. There is a forg* *etful functor : XSq ! XMod 2. This functor has a right adjoint D which completes th* *e pair ~ : M ! P, : N ! P with L = M xP N and ~, ~0 given by the projections and h : M x N ! L given by h(m, n) = (nmm-1, n mn-1), m 2 M, n 2 N. More interestin* *gly, it has a left adjoint which to the above pair of crossed P -modules yields the * *`universal crossed square' M N _____//N | | | | (1* *3) fflffl| fflffl| M ___~___//P where M N, as defined in [BL87a ], is the nonabelian tensor product of groups* * which act on each other. Then is a fibration of categories and also a cofibration. Thus we have a n* *otion of induced crossed square, which according to Proposition 4.2 is given by a pushou* *t of the form 0 1 ff fifi ` ' @ A` ' M N N ___ff____fl//_R S S M P R Q 0 1 | |0 1 @u 1A || |@v| 1A 1 1 || ||1 1 ` fflffl|' ` fflffl|' L N _______________//T S M P 0 ffifi1 R Q @ A ff fl in the category of crossed squares, given morphisms (ff, fl) : (M ! P ) ! (R ! * *Q), (fi, fl) : (N ! P ) ! (S ! Q) of crossed modules. 30 The functor is exploited in [BL87a ] for an HHvKT implying some calculatio* *ns of the non-Abelian group ss3(X :A, B; x)3. The applications are developed in [BL8* *7b ] for a triadic Hurewicz Theorem, and for the notion of free crossed square, both bas* *ed on `induced crossed squares'. Free crossed squares are exploited in [Ell93] for ho* *motopy type calculations. In fact the HHvKT works in all dimensions and in the more general setting of* * n-cubes of spaces, although not in a many base point situation. For a recent applicati* *on, see [EM08 ]. References [Bor94] Borceux, F. Handbook of categorical algebra. 2, Encyclopedia of Math* *emat- ics and its Applications, Volume 51. Cambridge University Press, Camb* *ridge (1994). Categories and structures. [BB04] Borceux, F. and Bourn, D. Mal'cev, protomodular, homological and sem* *i- abelian categories, Mathematics and its Applications, Volume 566. Klu* *wer Aca- demic Publishers, Dordrecht (2004). [Bro67] Brown, R. `Groupoids and van Kampen's theorem'. Proc. London Math. So* *c. (3) 17 (1967) 385-401. [Bro80] Brown, R. `On the second relative homotopy group of an adjunction sp* *ace: an exposition of a theorem of J. H. C. Whitehead'. J. London Math. So* *c. (2) 22 (1) (1980) 146-152. [Bro99] Brown, R. `Groupoids and crossed objects in algebraic topology'. Ho* *mology Homotopy Appl. 1 (1999) 1-78 (electronic). [Bro04] Brown, R. `Crossed complexes and homotopy groupoids as non commutative tools for higher dimensional local-to-global problems'. In `Galois th* *eory, Hopf algebras, and semiabelian categories', Fields Inst. Commun., Volume 4* *3. Amer. Math. Soc., Providence, RI (2004), 101-130. [Bro06] Brown, R. Topology and Groupoids. Printed by Booksurge LLC, Charlesto* *n, S. Carolina, third edition (2006). [BG89] Brown, R. and Golasi'nski, M. `A model structure for the homotopy th* *eory of crossed complexes'. Cahiers Topologie G'eom. Diff'erentielle Cat'* *eg. 30 (1) (1989) 61-82. [BH78] Brown, R. and Higgins, P. J. `On the connection between the second re* *lative homotopy groups of some related spaces'. Proc. London Math. Soc. (3) * *36 (2) (1978) 193-212. _____________ 3Earlier results had used homological methods to obtain some Abelian values. 31 [BH81a] Brown, R. and Higgins, P. J. `On the algebra of cubes'. J. Pure Appl.* * Algebra 21 (3) (1981) 233-260. [BH81b] Brown, R. and Higgins, P. J. `Colimit theorems for relative homotopy * *groups'. J. Pure Appl. Algebra 22 (1) (1981) 11-41. [BH81c] Brown, R. and Higgins, P. J. `The equivalence of 1-groupoids and cro* *ssed complexes'. Cahiers Topologie G'eom. Diff'erentielle 22 (4) (1981) 37* *1-386. [BH91] Brown, R. and Higgins, P. J. `The classifying space of a crossed comp* *lex'. Math. Proc. Cambridge Philos. Soc. 110 (1) (1991) 95-120. [BHS08] Brown, R., Higgins, P. J. and Sivera, R. Nonabelian algebraic topolo* *gy (in preparation). www.bangor.ac.uk/r.brown/nonab-a-t.html (2008). [BH82] Brown, R. and Huebschmann, J. `Identities among relations'. In * *`Low- dimensional topology (Bangor, 1979)', London Math. Soc. Lecture Note * *Ser., Volume 48. Cambridge Univ. Press, Cambridge (1982), 153-202. [BL87a] Brown, R. and Loday, J.-L. `Van Kampen theorems for diagrams of spac* *es'. Topology 26 (3) (1987) 311-335. With an appendix by M. Zisman. [BL87b] Brown, R. and Loday, J.-L. `Homotopical excision, and Hurewicz theore* *ms for n-cubes of spaces'. Proc. London Math. Soc. (3) 54 (1) (1987) 176-192. [BW95] Brown, R. and Wensley, C. D. `On finite induced crossed modules, and* * the homotopy 2-type of mapping cones'. Theory Appl. Categ. 1 (1995) No. 3* *, 54- 70 (electronic). [BW03] Brown, R. and Wensley, C. D. `Computation and homotopical application* *s of induced crossed modules'. J. Symbolic Comput. 35 (1) (2003) 59-72. [Cro59] Crowell, R. H. `On the van Kampen theorem'. Pacific J. Math. 9 (1959)* * 43-50. [Ell93] Ellis, G. `Crossed squares and combinatorial homotopy'. Math. Z. 21* *4 (1) (1993) 93-110. [EM08] Ellis, G. J. and Mikhailov, R. `A colimit of classifying spaces'. arX* *iv: [math.GR] (0804.3581v1) (2008) 1-16. [Gra66] Gray, J. W. `Fibred and cofibred categories'. In Proc. Conf. Categori* *cal Algebra (La Jolla, Calif., 1965), Springer, New York (1966), 21-83. [Gro68] Grothendieck, A. Cat'egories cofibr'ees additives et complexe cotange* *nt relatif, Lecture Notes in Mathematics, Volume 79. Springer-Verlag, Berlin (196* *8). 32 [GWL81] Guin-Wal'ery, D. and Loday, J.-L. `Obstruction `a l'excision en K* *-th'eorie alg'ebrique'. In `Algebraic K-theory, Evanston 1980 (Proc. Conf., Nor* *thwestern Univ., Evanston, Ill., 1980)', Lecture Notes in Math., Volume 854. Sp* *ringer, Berlin (1981), 179-216. [Hig71] Higgins, P. J. Notes on categories and groupoids, Mathematical Studi* *es, Vol- ume 32. Van Nostrand Reinhold Co., London (1971); Reprints in Theory * *and Applications of Categories, No. 7 (2005) pp 1-195. [Hop42] Hopf, H. `Fundamentalgruppe und zweite Bettische Gruppe'. Comment. Ma* *th. Helv. 14 (1942) 257-309. [Mac05] Mackenzie, K. C. H. General theory of Lie groupoids and Lie algebroid* *s, London Mathematical Society Lecture Note Series, Volume 213. Cambridge Unive* *rsity Press, Cambridge (2005). [Str99] Streicher, T. `Fibred categories `a la B'enabou'. http://www.mathematik.tu-darmstadt.de/~streicher/ (1999) 1-85. [Vis04] Vistoli, A. `Notes on Grothendieck topologies, fibered categories and* * descent theory'. arXiv:math/0412512v4 (2004) 114 p. [Whi78] Whitehead, G. W. Elements of homotopy theory, Graduate Texts in Mathe* *mat- ics, Volume 61. Springer-Verlag, New York (1978). [Whi41] Whitehead, J. H. C. `On adding relations to homotopy groups'. Ann. of* * Math. (2) 42 (1941) 409-428. School of Computer Science Bangor University, Gwynedd LL57 1UT, U.K. Departamento de Geometr'ia y Topolog'ia Universitat de Val`encia, 46100 Burjassot, Valencia Spain Email: r.brown@bangor.ac.uk, Rafael.Sivera@uv.es