AXIOMATIC STABLE HOMOTOPY _ A SURVEY N. P. STRICKLAND Abstract.We survey various approaches to axiomatic stable homotopy theor* *y, with examples including derived categories, categories of (possibly equivariant or loc* *alized) spectra, and stable categories of modular representations of finite groups. We focus mainly* * on representability theorems, localisation, Bousfield classes, and nilpotence. 1. Introduction Axiomatic stable homotopy theory is the study of triangulated categories form* *ally similar to the homotopy category of spectra in the sense of Boardman [1, 49]. While variou* *s authors have used different systems of axioms, there is broad agreement about the main examp* *les, of which the following are a sample: (1) Boardman's category itself, which we denote by B. (2) The subcategories of E(n)-local and K(n)-local spectra [66, 36]. (3) The homotopy category BG of G-spectra (indexed on a complete universe), * *where G is a compact Lie group [48, 50]. (4) The homotopy category of A-modules, where A is a commutative ring spectr* *um. (Here and throughout this paper, the phrase "ring spectrum" refers to a strict* *ly associative monoid in a suitable geometric category of spectra, such as that defined* * in [23].) (5) The derived category DR of R-modules, for a commutative ring R. If we le* *t HR denote the associated Eilenberg-MacLane ring spectrum, then DR ' DHR (by [23, Theor* *em IV.2.4]), so this is a special case of the previous example. (6) The stable category StabkGof kG-modules and projective equivalence class* *es of homo- morphisms, where G is a finite group and k is a field [70]. This occurs * *as a quotient of the category DkG, in which the morphism sets are given by group cohomology; * *the morphism sets in StabkGitself are more closely related to Tate cohomology. (7) The derived category (in a suitable sense) of MU*MU-comodules [33]. There are some further examples that satisfy some authors' axioms but not oth* *ers, or where the axioms have not yet been checked (to the best of my knowledge): (1) The derived category of modules over a noncommutative ring. (2) The derived category of quasicoherent sheaves over a nonaffine scheme. (3) The homotopy category of G-spectra indexed by an incomplete G-universe. (4) Various versions of the category of motivic spectra. (Motivic spaces are* * discussed in [54]; at the time of writing, there is no published account of the correspondi* *ng category of spectra.) It seems an important problem to decide which of the usual axioms apply to the * *motivic stable category, and to see what the axiomatic literature teaches us about this exampl* *e. The main topics that have been discussed from an axiomatic point of view are * *as follows. (a) Theorems saying that certain (covariant or contravariant) functors are r* *epresentable, gen- eralising the Brown representability theorems for (co)homology theories * *on (finite or infi- nite) spectra. (b) Phantom maps. ___________ Date: July 10, 2003. 1991 Mathematics Subject Classification. 55U35. Key words and phrases. axiomatic stable homotopy,triangulated category,Bousf* *ield class . 1 2 N. P. STRICKLAND (c)Various kinds of localisation, generalising Bousfield's theory of localis* *ation with respect to a homology theory. Special cases such as finite, cofinite, algebraic o* *r smashing locali- sations. (d) The lattice of Bousfield classes, and various related lattices (some of t* *hem conjecturally identical to the Bousfield lattice). (e)Nilpotence theorems in the spirit of Devinatz, Hopkins and Smith. (f)Picard groups, Grothendieck rings, and Euler characteristics [28, 51, 52,* * 46]. (g)Projective classes, and generalisations of the Adams spectral sequence [1* *5]. (h) Duality theorems generalising those of Verdier and Gross-Hopkins [57, 24]. For (f) to (h) we refer the reader to the cited papers and their bibliographies* *. This survey will concentrate on (a) to (e). The relevant literature consists partly of papers that are explicitly axiomat* *ic, and partly of pa- pers that are nominally restricted to some particular category, but whose metho* *ds allow straight- forward generalisation to other examples. Some authors are as follows: (1) Margolis's book [49] treats B from an axiomatic point of view; this was * *an important inspiration for much of the later work. Earlier still, there were releva* *nt papers by Freyd, Heller and Joel Cohen. (2) Neeman has written extensively, particularly on questions related to rep* *resentability and localisation [13, 17, 59, 56, 62, 55, 61, 58, 63]. Some papers are restr* *icted to the case of DR; often R need not be commutative or noetherian. When working axiomat* *ically, he has generally assumed that his triangulated category C is öc mpactly gen* *erated", but not that C has a symmetric monoidal structure. The class of categories cons* *idered is thus rather large, but unfortunately it is not closed under Bousfield localis* *ation. Recently he has introduced the more complex notion of a "well-generated" triangulate* *d category to repair this problem. (3) Krause has also written extensively on representability, localisation, a* *nd versions of the Bousfield lattice [11, 3, 39, 40, 43, 42, 41, 44, 45, 46, 47]. (4) Beligiannis has written a long paper [2] covering many themes in axiomat* *ic stable homo- topy, considered as an analog of relative homological algebra in the con* *text of triangulated categories. (5) Benson, Carlson, Rickard and Gnacadja (working in various combinations) * *have proved many results about the categories StabkGand DkG, often using methods tha* *t transfer easily to an axiomatic setting [4, 6, 7, 8, 9, 10, 5, 11, 3, 72, 70, 68,* * 71, 69]. Benson and Wheeler have interpreted the Green correspondence in this context [12]. (6) Hovey, Schwede and Shipley have worked in a more rigid context, studying* * Quillen model categories C0 such that the homotopy category Ho(C0) is triangulated [34* *, 73, 75, 76]. (7) May and coauthors have studied the equivariant stable categories BG, oft* *en using methods that transfer easily to an axiomatic setting. This applies particularly* * to their work on duality, traces, and Picard groups [25, 52, 51, 48]. (8) Hovey and Palmieri and Strickland wrote a memoir [35] on axiomatic stabl* *e homotopy theory. We assumed much more than Neeman, and thus could obtain results* * closer to those previously known for B. In particular, we assume that C has a clos* *ed symmetric monoidal structure. 2. Axioms We next discuss the various axioms that have been used. We start with a categ* *ory C. 2.1. Basics. The category C should be triangulated, and should have coproducts * *for all families of objects (indexed by a set). These are core axioms, used by almost all author* *s. Existence of coproducts should be seen as an important test of the correctness of the techni* *cal details of the definition of C. Boardman's category itself came after several attempts to defi* *ne a good category of spectra, and it was the first to be triangulated and coproduct-complete; it * *rapidly became clear that Boardman's version was much more convenient than all the others. Similarly* *, the earliest AXIOMATIC STABLE HOMOTOPY _ A SURVEY 3 versions of DR incorporated various boundedness conditions, and so were not cop* *roduct-complete. Bokstedt and Neeman [13] adjusted the definitions to remove this problem, and t* *his allowed much smoother comparisons between DR and B. We recall the definition of a triangulation: Definition 2.1. A triangulation of an additive category C is an additive (suspe* *nsion) functor : C -!C giving an automorphism of C, together with a collection 4 of diagrams,* * called distin- guished triangles or cofibre sequences, of the form X -!Y -!Z -! X such that 1. Any diagram isomorphic to a cofibre sequence is a cofibre sequence. 2. Any diagram of the following form is a cofibre sequence: 0 -!X 1-!X -!0 3. The first of the following diagrams a cofibre sequence iff the second is* * a cofibre sequence: X f-!Y -g!Z h-! X Y -g!Z h-! X --f--! Y. 4. For any map f :X -!Y , there is a cofibre sequence of the following form: X f-!Y -!Z -! X 5. Suppose we have a diagram as shown below (with h missing), in which the * *rows are cofibre sequences and the rectangles commute. Then there exists a (nonunique) ma* *p h making the whole diagram commutative. U ________V-________W-_______- U | | || | | | | | |f | |h | f | | | | |? |? ||? |? X ________Y-________Z-_______- X 6. Verdier's octahedral axiom holds: Suppose we have maps X -v!Y u-!Z, an* *d cofibre sequences (X, Y, U), (X, Z, V ) and (Y, Z, W ) as shown in the diagram. * *(A circled arrow U -!O X means a map U -! X.) Then there exist maps r and s as shown, * *making (U, V, W ) into a cofibre sequence, such that the following commutativit* *ies hold: au = rd es = ( v)b sa = f br = c _____-V______ J J] J b c J a J r J J s Æ J J X ___________-Zuv J OE J OE J J c c vJ u J f JJ J J J J J^ J^ Æ U ___________oe___________oecYW d e (If u and v are inclusions of CW spectra, this essentially just says tha* *t (Z=X)=(Y=X) = Z=Y . The diagram can be turned into an octahedron by lifting the outer* * vertices and drawing an extra line from W to U.) 4 N. P. STRICKLAND Following the standard topological notation, we write [X, Y ] for the set C(X* *, Y ) (of morphisms in C from X to Y ). We also put [X, Y ]n = [ nX, Y ]. If C0 is a pointed Quillen model category, then the homotopy category C = Ho(* *C0) automatically has structure close to that described above, except that the functor : C -! C * *need not be an equivalence. Following Hovey, we say that C0 is a stable model category if is* * an equivalence; if so, one can show that C is triangulated [34, Chapter 7]. Similar results appear* * in [51]. Schwede and Shipley have shown [75] that most stable model categories are Quillen-equiv* *alent to DA for some ring spectrum A, and that the general case is only a little more general. Various modifications and refinements of triangulated categories have been co* *nsidered by Nee- man [55], May [51], Franke [26] and probably others. It seems likely that these* * all follow from the existence of an underlying model category C0 as above. For the categories C occ* *urring in practice, it seems that there is always an underlying model category, and that any two na* *tural choices are Quillen equivalent. It thus seems reasonable to assume whenever convenient that* * one is given a category C0. 2.2. Smash products. In [35], we assume that our categories C come equipped wit* *h a symmetric monoidal product. We use notation coming from topology, and thus write X ^ Y fo* *r the monoidal product of X and Y , and S for the unit, so S ^ X = X = X ^ S. We also assume t* *hat there are adjoint function objects F (Y, Z), so [X, F (Y, Z)] ' [X ^ Y, Z], naturally in * *all variables. We write Sn = nS, so Sn ^ Sm = Sn+m and nX = Sn ^ X. We also put ßnX = [Sn, X]. This structure certainly exists in all the categories mentioned so far, excep* *t for the category DA when A is not commutative. However, the theory of blocks in group algebras deco* *mposes StabkG as a product of smaller categories, which need not have a symmetric monoidal st* *ructure. There may be similar examples related to BG. It is natural to require that the smash product be compatible with the triang* *ulation. In [35], we wrote down the most obvious compatibility conditions: (1) The smash product commutes with suspension, so (X ^ Y ) ' X ^ Y . (2) The functors X ^ (-) and F (X, -) preserve cofibre triangles. The contra* *variant functors F (-, Y ) preserve cofibre triangles up to a sign change. (3) The twist map S1 ^ S1 -!S1 ^ S1 is multiplication by -1. However, May has given strong evidence that further conditions should be added.* * To explain this, consider a map f :X -!X. Under suitable finiteness conditions, this has a trace* * ø(f): S -!S. If we have a cofibre sequence X0 -!X1 -!X2 -! X0 and compatible maps fi:Xi -!Xi, it is natural to hope that ø(f0) - ø(f1) + ø(f2* *) = 0; this is suggested by the theory of Lefschetz numbers, among other things. It turns * *out that the statement must be adjusted slightly: given f0 and f1, one can choose a compatib* *le f2 such that ø(f0) - ø(f1) + ø(f2) = 0, but this may not be the case for all compatible f2's* *. This (and various extensions) can be proved in Boardman's category, but the proof cannot * *be transferred to the axiomatic setting without adding some more conditions. In outline, consi* *der two cofibre sequences X0 -!X1 -!X2 -! X0 Y0 -!Y1-!Y2 -! Y0. From these we obtain a 4 x 4 diagram with vertices Xi^ Xj, in which all squares* * commute, except that one square anticommutes. By writing in the diagonal composite in ea* *ch square, we get 18 commutative triangles. Each commutative triangle fits in an octahedron, * *as in Verdier's axiom. The 18 resulting octahedra have many vertices and edges in common, and o* *ne can hope to add in some extra vertices and edges making everything fit together more coh* *erently. May [51] has formulated three axioms about this situation, and explained how they can be* * checked when C = Ho(C0) for some Quillen model category C0. There are many interesting cases of noncommutative rings (or ring spectra) R * *and R0for which R 6' R0 but DR ' DR0; this is a natural extension of Morita theory. Examples c* *ome from Koszul duality, the Fourier-Mukai transform for sheaves on abelian varieties, t* *ilting complexes AXIOMATIC STABLE HOMOTOPY _ A SURVEY 5 in representation theory, and so on [68, 69, 14, 20, 74]. Nonetheless, it seems* * that there are no examples of this type where the categories involved have smash products and the* * equivalence respects them. We know of no rigorous results in this direction, however. 2.3. Generation. A key feature of Boardman's category is that every spectrum X * *has a cell structure, and (essentially equivalently) that if [S0, X]* = 0 then X = 0. More* * generally, suppose we have a set G of objects in a triangulated category C. We say that G generate* *s C if there is no proper triangulated subcategory C0 C closed under all coproducts such that G * * C0. We also say that G detects C if every object X 2 C with [A, X]* = 0 for all A 2 G actually * *has X = 0. (By taking C0= {A | [A, X]* = 0}, we see that generation implies detection, and the* * converse holds under suitable finiteness conditions.) Generating sets for the main examples ar* *e as follows. (1) {S0} generates B, and LES0 generates the subcategory of E-local spectra * *[35, Section 3.5]. (2) If X is a finite spectrum of type n (in the usual chromatic sense) then * *LK(n)X generates the category of K(n)-local spectra [36, Theorem 7.3], and this is genera* *lly a better choice of generator than LK(n)S0. (3) The G-spectra G=H+ (as H runs over conjugacy classes of closed subgroups* *) generate BG. (4) If R is a ring, then R generates DR. This also works for (strictly assoc* *iative) ring spectra. (5) The nonprojective simple kG-modules generate StabkG. All authors in axiomatic stable homotopy theory assume that C is generated by* * some set G of objects, and impose some smallness conditions on G. The details vary between au* *thors, however. One popular condition is as follows. We say that an object A 2 C is small (or* * compact) if the natural map M M [A, Xi] -![A, Xi] i2I i is an isomorphism for all families of objects {Xi}. For example: (1) In B, the small objects are those of the form dX where d 2 Z and X is a* * finite CW complex. (2) In the category of E(n)-local spectra, the small objects are those that * *can be written as a retract of dLE(n)X for some d 2 Z and some finite CW complex X. (3) In the category of K(n)-local spectra, the small objects are those that * *can be written as a retract of dLK(n)X for some d 2 Z and some finite CW complex X of typ* *e n. (4) In BG, the small objects are those that can be written as a retract of * *dX for some d 2 Z and some finite G-CW complex X. (5) In DR, the small objects are the finite complexes of finitely generated * *projective modules. (6) In StabkG, the small objects are the kG-modules M that are finite-dimens* *ional over k. (Most of these facts are proved in [35], for example.) We say that C is compactly generated if there is a set G of small objects tha* *t generates C. In the terminology of [35], a stable homotopy category is algebraic iff it is comp* *actly generated. This is a very convenient condition, and is often satisfied; in particular, the gene* *rators listed above for B, BG, DR and StabkGare all small. In the case of the K(n)-local category, the * *obvious generator LK(n)S0 is not small, but LK(n)X is small whenever X is finite of type n, so th* *e category is nonetheless compactly generated. For a simpler example of the same phenomenon, * *let C be the p-completion of the category DZ; then the obvious generator is Zp (which is not* * small), but the object Z=p is also a generator, and is small. (There is a well-understood axio* *matic framework covering both of these examples: see [35, Section 3.3], and Section 7 of the pr* *esent paper.) More seriously, there are many known spectra E for which the category CE of E-local * *spectra has no nontrivial small objects, so in particular, CE is notWcompactly generated [60][* *36, Appendix B]. For example, this applies with E = BP or E = H or E = 0 n<1 K(n). Another useful condition is dualisability. To formulate this, we need to assu* *me that C has a symmetric monoidal smash product, as in the previous section. We write DA = F (* *A, S), where S is the unit object for the smash product. We say that A is dualisable if the* * natural map DA ^ A -!F (A, A) is an isomorphism; this implies that we have DA ^ B = F (A, B* *) for all B. 6 N. P. STRICKLAND The category of dualisable objects is formally very similar to the category of * *finite dimensional vector spaces over a field. In Boardman's category B, or in the derived category DR, it is known that an * *object is dualisable iff it is small. However, LK(n)S0 is dualisable but not small in the K(n)-local* * category. A number of interesting things are known about K(n)-locally dualisable spectra: (1) A K(n)-local spectrum X is dualisable iff dimK(n)*K(n)*X < 1 (proved in * *[36]). (2) If X is a finite complex, then LK(n)Q1 X is easily seen to be dualisable. (3) If X is a connected space with | k>0ßkX| < 1, then it is probably true * *that K(n)*X is finite-dimensional and so LK(n) 1 X is dualisable. The results in the* * literature involve some additional conditions, however; for example, the claim is true if X* * = BG with G finite [65] (in which case LK(n) 1 BG turns out to be self-dual [77]), o* *r if X is a double loop space [31]. In [35], we assume that our generators are dualisable, but not that they are * *small. This theory has the advantage that any localisation of a category satisfying our axioms, ag* *ain satisfies our axioms. The disadvantages are: (i)We need a smash product to formulate the definition of dualisability, an* *d this is absent or unnatural in many examples, such as DR when R is not commutative. (ii)If C = Ho(C0) for some Quillen model category C0, then there are natural* * conditions on objects in C0 guaranteeing that they are small in C. This is not the cas* *e for dualisability: we need special geometric arguments to show that G=H+ is dualisable in B* *G, for example. (iii)There are naturally occurring cases where the generators are small but * *not dualisable, for example the category of G-spectra based on an incomplete universe, a* *nd possibly also derived categories for nonaffine schemes. In [63], Neeman introduces the notion of a well-generated triangulated catego* *ry; the definition is explained and simplified in [42]. To explain the nature of this concept, we * *recall some generali- sations of Quillen's small object argument. Quillen originally considered a cat* *egory E closed under limits and colimits, and looked for objects A that were small in the sense that* * the functor E(A, -) preserves filtered colimits. Later, it was realised that one can fix a large ca* *rdinal ~ and say that A is ~-small if E(A, -) preserves colimits of sequences indexed by ordinals lar* *ger than ~. In many categories, every object is ~-small for some ~, and in many applications relate* *d to localisation, this is an adequate substitute for smallness. Neeman works with triangulated ca* *tegories, which typically do not have colimits for most diagrams. Thus, the above cannot be app* *lied directly, but a somewhat more elaborate argument leads to Neeman's well-generated categories.* * It is shown in [62] that the derived category of any Grothendieck abelian category is well-* *generated. 2.4. Representability. A cohomology functor on C is a contravariant functor fro* *m C to the category Ab of abelian groups, that converts coproducts to products and cofibre* * sequences to exact sequences. For any fixed object Z, it is well-known that the representable func* *tor X 7! [X, Z] is a cohomology functor. We say that the representability theorem holds for C i* *f the converse is true, so that every cohomology theory on C is representable. Brown proved the r* *epresentability theorem for B, and the same proof works for any compactly generated triangulate* *d category. By much more elaborate arguments, Neeman has extended this to all well-generated t* *riangulated categories [63], and similar results have been obtained by Krause [44, 45] and * *Franke [27]. It is also easy to see that if the representability theorem holds for C, then* * it holds for any localisation of C. In [35], we take the representability theorem as an axiom; this gives a conve* *nient way to treat compactly generated categories and their localisations in parallel. All other a* *uthors assume axioms that turn out to imply the representability theorem. 2.5. Extra axioms. We now discuss some possible additional assumptions. No auth* *or takes any these as a standard axiom, but they define special classes of examples with use* *fully simplified behaviour. (a) Let C be an abelian category in which all monomorphisms split (and thus * *all epimorphisms split), and suppose we have an equivalence : C -!C. We can give C a tri* *angulation by AXIOMATIC STABLE HOMOTOPY _ A SURVEY 7 declaring that all all triangles of the form A B f-!B C g-!C A h-! A B (with f(a, b) = (b, 0) and so on) are cofibre sequences. We call this the* * abelian case. Under mild finiteness conditions, one can show that C is the category of graded* * A*-modules, for some graded ring A* that is a finite product of graded division rings. Th* *is situation is discussed in [35, Section 8]. Examples include the category of rational G-spectra for any finite grou* *p G, or the category StabkGwhen |G| is invertible in k. (b) Suppose that C has a symmetric monoidal structure, and that C is generate* *d by the single object S (the unit for the smash product). We then have a graded-commutat* *ive ring ß*S defined by ßnS = [ nS, S]. If this is noetherian, we say that we are in t* *he noetherian case; this is discussed in [35, Section 6]. Examples include DR where R i* *s commutative and noetherian, and DkG. (c)Suppose again that C has a symmetric monoidal structure, and that C is ge* *nerated by the single object S. If ßnS = 0 for n < 0, then we are in the connective case* *; this is discussed in [35, Section 7]. We will make a number of remarks about the noetherian case below. Beyond that* *, we refer the reader to [35] for further discussion. 3.Functors on small objects Let F be the category of small objects in C, and consider the category A = [F* *op, Ab] of additive contravariant functors from F to the category of abelian groups. This * *is a bicomplete abelian category satisfying the AB5 condition (filtered colimits are exact). Th* *e functor : C -!C induces a functor : A -!A. If C has a good symmetric monoidal structure, then * *so does A. If the objects of F are strongly dualisable, then we have F ' Fop and so A ' [F, A* *b]. One can think of F as a "ring with many objects", and regard A as its module category. There is a Yoneda functor h: C -!A sending X to the functor hX (A) = [A, X] (* *for A 2 F). The structure of A and the behaviour of h have proved to be very useful in the * *study of C, at least when C is compactly generated. If C is merely well-generated, then Neeman* * has developed a partially parallel theory based on more complicated functor categories [63, C* *hapter 6]. In the compactly generated case, Beligiannis [2] has considered categories of the form* * [Gop, Ab] where G is an arbitrary triangulated subcategory of F. Let E A be the category of exact functors: those that send cofibre sequence* *s in F to exact sequences in Ab. It is standard that hX 2 E for all X. In good cases, E is the * *category of objects of finite projective (or injective) dimension in A, and h: C -!E is close to be* *ing an equivalence; see Section 9 for more discussion. The functor h: C -! A always preserves coproducts and sends cofibre sequences* * to exact se- quences. In other words, it is an A-valued homology theory on C. A morphism u: * *X -!Y in C is said to be phantom if hu: hX -!hY is zero. 4.Types of subcategories Let R be a commutative noetherian ring, and let zar(R) denote the space of pr* *ime ideals in R, with the Zariski topology. (We do not use the notation spec(R), to avoid con* *flicting uses of the word "spectrum".) It turns out [30, 56] that we can recover zar(R) from the* * category DR, in several slightly different ways. More precisely, one can recover the lattice* * of radical ideals in R, which is well-known to be anti-isomorphic to the lattice of closed subsets o* *f zar(R), and this lattice determines zar(R) itself [37]. The key is to study various lattices of * *subcategories of DR. Using parallel constructions in the category StabkG, we can recover the space z* *ar(H*(G; Fp)). In the case of Boardman's category, this study makes contact with the chromatic ap* *proach to stable homotopy theory, and the nilpotence theorems of Devinatz, Hopkins and Smith. Th* *is has many important applications that are not visible in the purely algebraic examples. F* *or example, suppose we want to prove that all finite spectra X have some property P (X). Suppose we* * can show that 8 N. P. STRICKLAND o Whenever we have a cofibre sequence X -!Y -! Z in which two terms have pr* *operty P , then the third also has property P o Whenever P (X _ Y ) holds, so do P (X) and P (Y ) o There exists a finite spectrum X such that H*(X; Q) 6= 0 and P (X) holds. Then one can show using the subcategory classification theorems that P (X) hold* *s for all X. The basic definitions are as follows. Definition 4.1. Let D be a full subcategory of C. For simplicity, we assume tha* *t any object in C that is isomorphic to an object in D, is itself in D. Let A be an arbitrary c* *ollection (possibly a proper class) of objects in C. (a) D is thick if (i)The zero object lies in D. (ii)Any retract of any object in D, again lies in D. (iii)Whenever X -!Y -! Z is a cofibre sequence with two terms in D, the * *third term is also in D. (b) If C has a symmetric monoidal structure, we say that D is an ideal if X * *^ Y 2 D whenever X 2 D. Dually, we say that D is a coideal if F (Y, Z) 2 D whenever Z 2 D. (c) D is a localising subcategory if it is thick, and closed under (possibly* * infinite) coproducts. Dually, D is a colocalising subcategory if it is thick, and closed under* * (possibly infinite) products. (d) A (co)localising (co)ideal is a (co)localising subcategory that is also * *a (co)ideal. (e) A bilocalising subcategory is a subcategory that is both a localising su* *bcategory and a colocalising subcategory. A biideal is a subcategory that is both an ide* *al and a coideal. If C is monoidal and the unit object S 2 C is small and generates C, then eve* *ry (co)localising subcategory is a (co)ideal. This holds in the following cases: (1) C = B (but not C = BG for general G) (2) C = DR, where R is commutative (3) C = StabkG, where k has characteristic p and G is a p-group. If we have a functor F between triangulated categories that preserves cofibre* * sequences, then ker(F ) := {X | F X = 0} is evidently a thick subcategory. Similarly, if F is * *a functor from a triangulated category to an abelian category, and F converts cofibre sequences * *to exact sequences, then ker(F ) will again be a thick subcategory. Under various auxiliary conditi* *ons, we can conclude that ker(F ) is a (co)localising subcategory or a (co)localising ideal. Little is known about classification of subcategories that are not ideals. Th* *e examples studied in [8, Section 6] suggest that there is no simple and general picture. On the other hand, there are good classification results for many of our cent* *ral examples; the main method of proof will be discussed in Section 8. To state the results, it * *is convenient to introduce one more definition. Given an object A 2 F, we write thickid for * *the smallest thick ideal containing A. We then say that a thick ideal I is finitely generate* *d if it has the form thickid for some A (this makes senseWbecause the thick ideal generated by A1* *, . .,.Ar is also generated by the single object A = iAi). A classification of finitely generate* *d thick ideals (or thick subcategories) in F extends in a fairly obvious way to give a classificat* *ion of all ideals (or thick subcategories) in F. (a) Let R be a commutative noetherian ring, and put C = DR. Then the localis* *ing subcate- gories of C biject with the colocalising subcategories, and with the sub* *sets of zar(R). The finitely generated thick subcategories of F biject with closed subsets o* *f zar(R). On the other hand, Neeman has considered the nonnoetherian ring R = k[x2, x3, x4, . .].=(x22, x33, x44, . .)., where k is a field. This has only one prime ideal, but DR has an enormou* *s collection of localising subcategories [61]. (b) Let k be a field, let G be a finite group, and put C = StabkG. It is pro* *ved in [8] that the finitely generated thick ideals in F biject with the closed subsets of t* *he projective scheme AXIOMATIC STABLE HOMOTOPY _ A SURVEY 9 proj(H*(G; k)). One can also show that the (co)localising (co)ideals in C* * biject with all subsets of proj(H*(G; k)). (c)In the category of E(n)-local spectra, the (co)localising subcategories i* *n C biject with subsets of {0, 1, . .,.n}, and the thick subcategories of F biject with t* *he subsets of the form {m, m + 1, . .,.n} for some m. All the relevant subcategories are (* *co)ideals [36, Theorem 6.14]. (d) Now let C be the category of K(n)-local spectra. Then 0 and C are the onl* *y localising subcategories of C, and also the only colocalising subcategories of C [36* *, Theorem 7.5]. Similarly, 0 and F are the only thick subcategories of F. (e)Finally, let C be the category of p-local spectra. The thick subcategorie* *s of F are then the categories Fn := {X | K(m)*X = 0 for allm < n}, where 0 n 1; this was* * proved in [32]. The theory of Bousfield classes gives many known examples of (c* *o)localising subcategories and inclusions between them. However, almost nothing is kno* *wn about the collection of all localising subcategories (which might even be a proper * *class). The classification results in Examples (a) and (b) have a partial generalisatio* *n that applies in the noetherian case. A strong but technically complex statement is proved in [35, S* *ection 6.3]; given some additional hypotheses (conjecturally always satisfied) this implies the ev* *ident analog of (a) and (b). So far we have only discussed results about ideals in F; we next consider res* *ults about (co)localising subcategories or (co)ideals in C. On the one hand, given D C we can certainly* * consider the thick subcategory D \ F F; if we have a good understanding of F then this will be a* * useful invariant, but rather a coarse one. On the other hand, given a thick subcategory A F we * *can consider the category A? := {X | [A, X]* = 0 for allA 2 A}, which is easily seen to be a bilocalising subcategory. The telescope conjectur* *e for C is closely related to the statement that every bilocalising subcategory is of the form A? * *for some A. This is known to hold in DR when R is noetherian and commutative, and also in StabkG* *when k has characteristic p and G is a p-group. It is believed to be false in Boardman's c* *ategory, although many years of study have still not produced a watertight argument. 5.Quotient categories and Bousfield localisation Let C be a triangulated category. Given a thick subcategory D, we can look fo* *r a triangulated category C0 and an exact functor Q: C -! C0 that sends all objects in D to zero* *. It turns out that there is an initial example of such a functor, whose target we call C=D. T* *o be more precise, we say that a map s: X -! Y in C is a D-equivalence if the cofibre of s lies in* * D. The class of D-equivalences has a number of useful properties: o Any isomorphism is a D-equivalence. o Given morphisms X s-!Y -t!Y , if any two of {s, t, ts} are D-equivalence* *s then so is the third. o Given maps X f-!Y- sZ in which s is a D-equivalence, there is a commutat* *ive square g W ____//_Z t|| |s| fflffl|fflffl| X __f__//Y in which t is a D-equivalence. We then define C=D as follows: the objects are the same as in C, and the morphi* *sms from X to Y are equivalence classes of öf rmal fractions" gt-1, where g and t fit in a di* *agram of the shape X- t W -g!Y , and t is a D-equivalence. The properties listed above allow us to* * compose and manipulate fractions in a natural way. Krause has considered some more delicate notions of quotient categories, whic* *h are important in the study of smashing localisations; but we will not discuss these here. 10 N. P. STRICKLAND As is well-known, there is a potential problem with the above construction, w* *hich is of great importance in some applications. We always assume implicitly that the morphism * *sets C(X, Y ) are genuine sets; but as defined above, (C=D)(X, Y ) might be a proper class. T* *here are a number of techniques that can be used in different circumstances to show that this pro* *blem does not arise. As far as I know, no one has looked systematically for examples where proper cl* *asses do arise; it is possible that (some version of) our standing axioms are enough to prevent th* *is. If C=D has small Hom sets, then for any Y 2 C we have a functor from C to Ab * *given by X 7! (C=D)(X, Y ). The Representability Theorem shows that this is representabl* *e, so we have an object LX 2 C and an isomorphism C(X, LY ) ' (C=D)(X, Y ), naturally in X. A* * standard argument shows that L can be regarded as a functor C=D -! C, right adjoint to t* *he quotient functor C -!C=D. If we put D? = {Y | C(X, Y ) = 0 for allX 2 D}, we find that L actually gives an equivalence C=D ' D? . We also use the letter * *L for the composite functor C -!C=D -!D? C; in this guise, it is left adjoint to the inclusion of* * D? in C. The functors L: C -!C arising in this way can be characterised by certain wel* *l-known properties: they are exact functors, equipped with a natural map iX :X -!LX such that LiX :* *LX -!L2X is an equivalence, and i*X:[LX, LY ] -![X, LY ] is an isomorphism for all Y . W* *e call such a pair (L, i) a Bousfield localisation functor. We can recover D as the category ker(L* *) = {X | LX = 0}. The above discussion shows that quotients are really localisations. Of course* *, the converse is also true: to invert a class of maps E is the same as to quotient out the local* *ising subcategory generated by the cofibres of the maps in E. In [35], it is assumed that C is symmetric monoidal and that D is a localisin* *g ideal. In this case, the quotient category C=D (or equivalently, the category D? ) inherits a * *symmetric monoidal structure. Given any localisation functor L, there is another functor C and natural tran* *sformations CX -qX-!X -iX-!LX -dX-! CX giving a cofibre sequence for all X. The theory can* * be set up in such a way that C and L play precisely dual r^oles. In any case, the pair (L* *, i) determines (C, q) (up to an obvious notion of equivalence) and vice versa. Given two localisation functors (L, i) and (L0, i0), there is at most one mor* *phism u: L -!L0 with ui = i0. We write L L0 if such a morphism exists. This gives a partial* * order on the collection of isomorphism classes of localisation functors. It is not known whe* *ther this collection is a set or a proper class. 6.Versions of the Bousfield lattice In this section, we assume that C has a symmetric monoidal structure. Without* * such a structure, one could set up a formal theory along the same lines, but it seems hard to ana* *lyze any examples explicitly. We can now define various partially ordered sets i; some of them may actuall* *y be proper classes, but we will suppress this from the terminology. An optimistic conjectu* *re would be that they are all the same; this is known to be true in the noetherian case. The gen* *eral case appears to be open (related work of Gutierrez and Casacuberta turns out not to provide a c* *ounterexample). Definition 6.1. o0is the class of all colocalising coideals (ordered by in* *clusion). For any class A of objects in C, we write colocid for the intersection of all* * colocalising ideals containing A. The poset 0 is actually a lattice, with meet operation D * *^ D0= D \ D0, and join D _ D0= colocid. o 1 is the class of all localising ideals, ordered by reverse inclusion. * *This is a lattice by a dual argument. o For any class A of objects in C, we put A? = {X | F (A, X) = 0 for allA 2 A} ?A = {X | F (X, A) = 0 for allA 2 A}. AXIOMATIC STABLE HOMOTOPY _ A SURVEY 11 It is easy to see that A? 2 0 and ?A 2 1. We say that a colocalising co* *ideal D is closed if it has the form A? for some A, or equivalently if D = (?D)?; we write * * 2 for the set of closed colocalising coideals, so 2 0. o Dually, we say that a localising ideal E is closed if it has the form ?A * *for some A, or equivalently if E = ?(E? ). We write 3 for the set of closed localising* * ideals, so that 3 1. o There are order-preserving maps 0 -! 3 and 1 -! 2, given by D 7! ?D and* * E 7! E? . A purely formal argument shows that these give an isomorphism 2 ' 3. o We say that a colocalising coideal D is reflective if the inclusion D -!C* * has a left adjoint; one can show that this implies that D is closed, so the coreflective coid* *eals give a subset 4 2. Dually, we say that a localising ideal E is coreflective if the * *inclusion has a right adjoint, and these ideals give a subset 5 3. The bijection 2 ' 3 re* *stricts to give a bijection 4 ' 5. If D and E correspond under this bijection, then the* *re is a pair of functors (L, C) as in Section 5, with D = image(L) = ker(C) = ker(L)? = image(C)? E = image(C) = ker(L) = ?ker(L) = ?image(C). It follows that 4 and 5 are equivalent to the poset of localisation fun* *ctors L for which ker(L) is an ideal, or to the poset of colocalisation functors C for whic* *h ker(C) is a coideal. o We say that a localising ideal E is principal if E = locid<{E}> for some * *object E. Note that if E = locid<{Ei}i2I>W(where I is a set, not a proper class) then we also* * have E = locid<{E}> where E = i2IEi, so E is principal. In Boardman's category, it is known * *that principal ideals are coreflective, so they form a subset 6 5. It is not clear i* *n what generality this argument works. If E = locid<{E}> then the corresponding localisati* *on functor is called stable E-nullification, and written P *E. (Confusingly, it was cal* *led colocalisation in Bousfield's original papers.) o We say that a localising ideal E is a Bousfield class if it has the form * *E = = {X | E ^ X = 0} for some E 2 C. We write 7 for the collection of all Bousfield c* *lasses. In Boardman's category, this is contained in 6; it is not clear how far thi* *s fact can be generalised. It is also known that 7 is a set rather than a proper class* * [64, 22]. To see this, for any finite spectrum A and any element x 2 E*A we let annA(x) be* * the set of maps f :A -!B in F such that (E*f)(x) = 0. We then write <> = {annA(x) | A 2 F , x 2 E*A}, and call this the Ohkawa class of E. As F has small Hom sets and only a s* *et of isomorphism classes, we see that there is only a set of possible Ohkawa classes. One * *can check that <> determines , so there is only a set of Bousfield classes. 7.Special types of localisation In this section, we assume that C is compactly generated. Definition 7.1. A Bousfield localisation functor L: C -! C is smashing if image* *(L) (which is automatically a colocalising subcategory) is closed under coproducts (and so is* * also a localising subcategory). In the monoidal case, this implies that there is a natural equiva* *lence LS ^ X -! LX, and the corresponding colocalisation functor C also satisfies CX = CS ^ X. * *The category D = image(L) = ker(C) = ker(L)? = image(C)? is then both a localising ideal and* * a colocalising coideal. We put U = ?D and V = D? , and then bCX = F (LS, X) and bLX = F (CS, X* *). It turns out that bLis a localisation functor, and bCis the corresponding colocalisation* *. Moreover, we have U = image(C) = ker(L) = V = image(bL) = ker(Cb) D = image(Cb) = ker(bL) = image(L) = ker(C) = U? = ?V = . 12 N. P. STRICKLAND It follows that CCb = 0 = bLL, so bLC ' bLand CLb' C. This implies that bL:U -* *! V and C :V -!U are mutually inverse equivalences. Apart from the finite localisations discussed below, the most important examp* *les are the local- isations with respect to the Johnson-Wilson spectra E(n). It is a highly nontri* *vial theorem [67, Chapter 8] that these are smashing. Krause [40] has shown that L is determined by the set ann(L|F ) of morphisms * *u: A -!B in F for which Lu = 0. This means in particular that there is only a set of smashing* * localisations. Definition 7.2. A finite localisation is a localisation functor L: C -!C where * *ker(L) = loc for some thick subcategory A F. Functors of this type are always smashing [53][35* *, Section 3.3]. One formulation of the telescope conjecture for C is the statement that every s* *mashing localisation is a finite localisation. This is known to be true in many noetherian cases, bu* *t believed to be false in Boardman's category. Keller [38] has provided a counterexample in DR for a c* *ertain ring R (but his framework of definitions is slightly different from ours, and we have * *not pinned down the precise relationship). An important example of finite localisation is as follows. Let R be a noether* *ian ring, and put C = DR. Fix an ideal I, and let A consist of the objects X 2 C for which ß*X is* * an I-torsion module. Here the category U = ker(L) consists of the I-torsion objects in C, a* *nd V = ker(Cb) consists of I-complete objects in a suitable sense. Thus, the equivalence U ' V* * shows that the torsion category and the complete category are essentially the same. All this i* *s closely related to the theory of local (co)homology [29]. See [21] and [35, Section 3.3] for other* * perspectives. Definition 7.3. Suppose that C is symmetric monoidal, and is generated by the u* *nit object S. Given a set T of homogeneous elements in the graded ring ß*S, we let A deno* *te the thick subcategory of F generated by the cofibres of the maps in T , and then let L be* * the corresponding finite localisation functor. One can show that ß*LX = (ß*X)[T -1]. Functors of * *this type are called algebraic localisations; in the special case where T Z, they are called arith* *metic localisations. 8. Nilpotence Our understanding of Boardman's category relies heavily on the nilpotence the* *orem of Devinatz, Hopkins and Smith [19] and its consequences [32, 67]. We next explain the forma* *l parts of this story that are amenable to axiomatic generalisation [78][35, Section 5]. We wil* *l assume here that C is compactly generated and has a symmetric monoidal structure, and that all o* *bjects of F are strongly dualisable. We say that an object I 2 F equipped with a map i: I -!S is an ideal if the m* *ap i^1: I^S=I -! S=I is null. (Here S=I denotes the cofibre of i.) We write I J if the map I -* *! S -! S=J is zero. It turns out that if I -i!S and J -j!S are ideals, then so is I ^ J -i^j* *-!S; we will just write IJ for this, making the set of isomorphism classes of ideals into a commu* *tative monoid. We say_that I and J are radically equivalent if for large n we have In J and Jn * * I. We write Id(S) for the set of radical equivalence classes of ideals. Given any A 2 F, th* *e fibre of the unit map S -!F (A, A) is an ideal, and we write ann(A) for its equivalence_class. On* *e can show that the rule ann(A) $ thickid gives a well-defined bijection between Id(S) and t* *he set of finitely generated thick ideals in F. Next, we say that a map u: A -!B in F is smash-nilpotent if the m'th smash po* *wer u(m):A(m)-! B(m)is zero for m 0. One checks that Im J for some m iff the composite I -!* *S -!S=J is smash-nilpotent. Now suppose we are given a set N and a collection of objects K(n) 2 C for eac* *h n 2 N. For any object X 2SC we put supp(X) = {n | K(n) ^ X 6= 0}. Similarly, given a thick* * ideal A F we put supp(A) = A2A supp(A). We say that the K(n)'s detect ideals if whenever A, B 2 F and supp(A) supp(* *B), we have thickid thickid. This implies that the map A -! supp(A) gives an embed* *ding of the lattice of thick ideals in the lattice of subsets of N. (Except in the noetheri* *an case, we know of no general method to determine the image of this map.) AXIOMATIC STABLE HOMOTOPY _ A SURVEY 13 We next explain two versions of what it might mean for the K(n)'s to detect n* *ilpotence. It is a key theorem that if the K(n)'s detect nilpotence, then they also detect ideals. For the most algebraically natural version, we need auxiliary hypotheses. Fir* *st, we assume that each K(n) has a commutative ring structure, and that every nonzero homogeneous * *element in the coefficient ring K(n)* is invertible (so K(n)* is a graded field). We also assu* *me that the resulting Künneth maps K(n)*(X) K(n)*K(n)*(Y ) -!K(n)*(X ^ Y ) are isomorphisms for all X and Y (this is not automatic unless C is generated b* *y {S}). Thus, we can regard K(n) as giving a monoidal functor from F to the category of finit* *e-dimensional vector spaces over K(n)*. We say that a map u: A -!B in F is K(*)-null if the i* *nduced map K(n)*A -! K(n)*B is zero for all n. We say that the K(n)'s detect smash-nilpot* *ence if every K(*)-null map is_smash-nilpotent. Assuming this, a straightforward argument (b* *ased on our discussion of Id(S)) shows that the K(n)'s detect ideals. In a general stable homotopy category C, it is very hard to produce ring obje* *cts K(n) such that K(n)* is a graded field. However, one can use another line of argument wit* *h rather different hypotheses. First, we say that the K(n)'s detect rings ifWfor every nonzero rin* *g object R we have K(n) ^ R 6= 0 for some n. (This will obviously hold if n = in the Bo* *usfield lattice.) Suppose in addition that whenever A 2 F and K(n) ^ A 6= 0 we have = * *. We claim that the K(n)'s detect ideals. To see this, consider a thick ideal A F,* * and let L be the finite localisation functor with ker(L) = loc (and thus ker(L) \ F = A). Giv* *en X 2 F with supp(X) supp(A), we must show that X 2 A. It turns out to be equivalent to sa* *y that the ring object R = F (X, X) ^ LS is zero, so it will suffice to show that K(n) ^ R = 0 * *for all n, and this is easy. The main examples are as follows. (a)In the motivating example [19, 32, 67], C is the category of p-local spec* *tra, N is N [ {1}, and K(n) is the n'th Morava K-theory (which is well-known to be a field t* *heory, up to a slight adjustment of the definition at the prime 2). The proof that these* * theories detect nilpotence is a tour de force of stable homotopy theory, using methods ve* *ry far from those surveyed in this paper. It follows that they also detect smash nilpotence* *. It is deduced in [32] that they detect nilpotence in various other senses, and that the* *y detect rings. Part of this argument can be axiomatised (at least in a connective stable homo* *topy category) but we shall not attempt that here. (b) Let G be a finite group, and let C be the category of p-local G-spectra. * *We then let N be the set of pairs (H, n), where H is a (representative of a) conjugacy cla* *ss of subgroups of G, and n 2 N [ {1}. We take K(H, n) to be the representing object for the* * cohomology theory X 7! K(n)* H X, where H is the geometric fixed point functor, and* * K(n) is the usual nonequivariant Morava K-theory. These representing objects can * *be made quite explicit, but we shall not give the details. It follows quite easily from* * the previous example that they detect smash-nilpotence, and thus that they detect ideals [78]. (c)Let R be a noetherian ring, and put C = DR. We then take N to be the set * *of prime idealsWin R, and let K(p) be the field of fractions of R=p. Here it is no* *t hard to show that p = and that is a minimal Bousfield class; it follows t* *hat these objects detect nilpotence, and also that they detect ideals [56][35, Section 6]. (d) Now consider the case C = StabkG, where k is a field of characteristic p * *and G is a finite p-group. Take N to be the set of homogeneous prime ideals in H*(G; k). Ne* *xt, fix an algebraically closed field L of infinite transcendence degree over k. For* * any p 2 N, the theory of "shifted subgroups" gives an algebra A LG isomorphic to L[u]=* *up and an object K(p) 2 C such that K(p) ^ M = 0 iff L kM is free asWa module over* * A. It follows easily from the infinite version of Dade's Lemma [7] that p = .* * Morever, we see from [7, Theorem 10.8] that = whenever K(p) ^ M 6= * *0, so the K(p)'s detect ideals. 14 N. P. STRICKLAND 9. Brown representability For all authors, it is either an axiom or a theorem that cohomology functors * *defined on the whole category C are representable. It follows easily that the Yoneda functor i* *s an equivalence between C and the category of cohomology functors defined on C. This is a very* * satisfactory result, with many applications (existence of infinite products, existence of Bo* *usfield localisations, Brown-Comenetz duality, and so on). It is desirable to extend this result to various subcategories D C. If the* *re is an exact localisation functor L: C -!D (as in Section 5), then this is easy. In some oth* *er cases, it can be proved using Neeman's theory of well-generated categories [63, Chapter 8]. Similarly, it would be helpful to have a dual theorem. This should say that * *any product- preserving exact covariant functor C -!Ab has the form Y 7! [X, Y ] for some re* *presenting object X. This has also been proved by Neeman [59, 63], under some additional hypothes* *es. Next, let F C be the subcategory of small objects, and suppose that F gener* *ates C. Given a cohomology functor H :Fop -!Ab, it is natural to ask whether there is an obje* *ct Z 2 C and a natural isomorphism HX = [X, Z] for X 2 F. It is equivalent to ask whether H ca* *n be extended to a cohomology functor defined on all of C. We first observe that in the case C = B, this reduces to a more familiar ques* *tion. In that context, the Spanier-Whitehead duality functor D :X 7! F (X, S) gives an equiva* *lence Fop ' F, so the covariant functor H0 = H O D :F -! Ab is homological. A natural isomorph* *ism HX = [X, Z] (for all X 2 F) is thus the same as a H0X = ß0(X ^ Z), and Brown's homol* *ogical representability theorem (in the version proved by Adams) says that such an iso* *morphism can always be found. Adams's proof used some countability arguments, and implicitly* * relied on the existence of an underlying model category, so it could not directly be transfer* *red to our axiomatic setting. Margolis [49] and Neeman [58] independently gave reformulations that d* *o not use model categories. Neeman also showed, however, that the countability hypothesis is es* *sential. To explain this and related results (mostly distilled from [58, 17, 2]), it i* *s convenient to use the category A = [Fop, Ab], the subcategory E A of exact functors, and the Yo* *neda functor h: C -! E, as discussed in Section 3. Brown's theorem says that when C = B, th* *e functor h: C -!E is full and essentially surjective. The work of Margolis and Neeman sa* *ys that the same holds whenever (a) C is compactly generated; and (b) F has only countably many isomorphism classes, and F(A, B) is countable * *for all A, B 2 F. Now consider the case C = D(k[x, y]), where k is a field. Neeman has shown th* *at if |k| @2 then h is not full, and if |k| @3 then h is not essentially surjective. These* * examples are obtained from more general and more complicated statements of two different types. Firs* *tly, there are results relating properties of h to homological algebra in A; secondly, there a* *re relations between homological algebra in A and in the category MR of R-modules, in the case where* * C = DR. For the first step, we define pgldim(C) to be the supremum of the projective * *dimensions in A of all the objects in E. Even though h need not be essentially surjective, this* * is known to be the same as the supremum of the projective dimensions of objects in the image of h.* * It is also known that (a) pgldim(C) 1 iff h is full. (b) If pgldim(C) 2, then h is essentially surjective. (c) Thus, if h is full, then it is essentially surjective. For the second step, we recall some additional definitions. An R-module P is * *said to be pure- projective if it is a retract of a (possibly infinite) direct sum of finitely p* *resented modules. The pure-projective dimension of a module M is the minimum possible length for a pu* *re projective resolution of M. The pure global dimension of R (written pgldim(R)) is the supr* *emum of the pure- projective dimensions of all R-modules. The ring R is said to be hereditary if * *every submodule of a projective module is again projective. It is known that pgldim(R) pgldim(DR* *); the inequality AXIOMATIC STABLE HOMOTOPY _ A SURVEY 15 is an equality when R is hereditary, but can be strict in more general cases. U* *sing this and some additional arguments, one proves the following result: Theorem 9.1. Suppose that C = DR, where R is hereditary. Then the functor h: C * *-!A is (a) full iff pgldim(R) 1 (b) essentially surjective iff pgldim(R) 2. Benson and Gnacadja [9, 10] have proved similar results for the case C = Stab* *kG, involving questions of purity for kG-modules. In particular, they show that the following* * are equivalent: (a) h is full and essentially surjective (b) pgldim(kG) 1 (c) Either k is countable, or the Sylow p-subgroup of G is cyclic (where p i* *s the characteristic of k). They also give a number of intricate examples related to these results. Now consider the case where h is full and essentially surjective, as with the* * original case of Boardman's category of spectra. We then say that C is a Brown category. This ha* *s a number of useful consequences [35, 18, 2]. Firstly, for F 2 A, the following are equivale* *nt: (a) F has finite projective dimension in A (b) F has projective dimension at most one (c) F has finite injective dimension (d) F has injective dimension at most one (e) F 2 E (f)F is in the image of h. Next, we say that a map v in C is phantom if h(v) = 0. In a Brown category, the* * composite of any two phantom maps is zero, so the phantoms form a square-zero ideal. (Benson* * [5] has shown that this can fail when C is not a Brown category; in particular, it fails when* * C = StabkG, k is an uncountable field of characteristic p, and the p-rank of G is at least two.) So* *me further properties of phantom maps are studied in [16]. Finally, consider a diagram X :I -! C, where I is a filtered category. A weak* * colimit for the diagram consists of a object U and compatible maps Xi-! U for i 2 I, such that * *the induced map [U, Y ] -!lim -[Xi, Y ] is surjective for all Y 2 C. Such a weak colimit is min* *imal if the induced I map lim-![Z, Xi] -![Z, U] is a bijection for all Z 2 F. In a Brown category, it* * is known that i (a) Every filtered diagram of small objects has a minimal weak colimit, whic* *h is a retract of any other weak colimit. (b) Every object can be expressed as the minimal weak colimit of a filtered * *diagram of small objects. References [1]J. F. Adams. Stable Homotopy and Generalised Homology. University of Chicag* *o Press, Chicago, 1974. [2]A. Beligiannis. Relative homological algebra and purity in triangulated cat* *egories. J. Algebra, 227(1):268-361, 2000. [3]D. Benson and H. Krause. Pure injectives and the spectrum of the cohomology* * ring of a finite group. J. Reine Angew. Math., 542:23-51, 2002. [4]D. J. Benson. Infinite dimensional modules for finite groups. In Infinite l* *ength modules (Bielefeld, 1998), Trends Math., pages 251-272. Birkhäuser, Basel, 2000. [5]D. J. Benson. Phantom maps and purity in modular representation theory. III* *. J. Algebra, 248(2):747-754, 2002. [6]D. J. Benson, J. F. Carlson, and J. Rickard. Complexity and varieties for i* *nfinitely generated modules. Math. Proc. Cambridge Philos. Soc., 118(2):223-243, 1995. [7]D. J. Benson, J. F. Carlson, and J. Rickard. Complexity and varieties for i* *nfinitely generated modules. II. Math. Proc. Cambridge Philos. Soc., 120(4):597-615, 1996. [8]D. J. Benson, J. F. Carlson, and J. Rickard. Thick subcategories of the sta* *ble module category. Fund. Math., 153(1):59-80, 1997. [9]D. J. Benson and G. P. Gnacadja. Phantom maps and purity in modular represe* *ntation theory. I. Fund. Math., 161(1-2):37-91, 1999. Algebraic topology (Kazimierz Dolny, 1997). 16 N. P. STRICKLAND [10]D. J. Benson and G. P. Gnacadja. Phantom maps and purity in modular represe* *ntation theory. II. Algebr. Represent. Theory, 4(4):395-404, 2001. [11]D. J. Benson and H. Krause. Generic idempotent modules for a finite group. * *Algebr. Represent. Theory, 3(4):337-346, 2000. Special issue dedicated to Klaus Roggenkamp on the occas* *ion of his 60th birthday. [12]D. J. Benson and W. W. Wheeler. The Green correspondence for infinitely gen* *erated modules. J. London Math. Soc. (2), 63(1):69-82, 2001. [13]M. Bökstedt and A. Neeman. Homotopy limits in triangulated categories. Comp* *ositio Math., 86(2):209-234, 1993. [14]T. Bridgeland. Equivalences of triangulated categories and Fourier-Mukai tr* *ansforms. Bull. London Math. Soc., 31(1):25-34, 1999. [15]J. D. Christensen. Ideals in triangulated categories: phantoms, ghosts and * *skeleta. Adv. Math., 136(2):284-339, 1998. [16]J. D. Christensen and M. Hovey. Phantom maps and chromatic phantom maps. Am* *er. J. Math., 122(2):275- 293, 2000. [17]J. D. Christensen, B. Keller, and A. Neeman. Failure of Brown representabil* *ity in derived categories. Topology, 40(6):1339-1361, 2001. [18]J. D. Christensen and N. P. Strickland. Phantom maps and homology theories.* * Topology, 37(2):339-364, 1998. [19]E. S. Devinatz, M. J. Hopkins, and J. H. Smith. Nilpotence and stable homot* *opy theory I. Annals of Mathe- matics, 128:207-242, 1988. [20]D. Dugger and B. Shipley. K-theory and derived equivalences, arXiv:math.KT/* *0209084. [21]W. G. Dwyer and J. P. C. Greenlees. Complete modules and torsion modules. A* *mer. J. Math., 124(1):199-220, 2002. [22]W. G. Dwyer and J. H. Palmieri. Ohkawa's theorem: there is a set of Bousfie* *ld classes. Proc. Amer. Math. Soc., 129(3):881-886, 2001. [23]A. D. Elmendorf, I. Kriz, M. A. Mandell, and J. P. May. Rings, Modules and * *Algebras in Stable Homotopy Theory, volume 47 of Amer. Math. Soc. Surveys and Monographs. American Mathe* *matical Society, 1996. [24]H. Fausk, P. Hu, , and J. May. Isomorphisms between left and right adjoints* *, arXiv:math.AT/0206079. [25]H. Fausk, L. G. Lewis, Jr., and J. P. May. The Picard group of equivariant * *stable homotopy theory. Adv. Math., 163(1):17-33, 2001. [26]J. Franke. Uniqueness theorems for certain triangulated categories posessin* *g an Adams spectral sequence. Preprint, 1996. [27]J. Franke. On the Brown representability theorem for triangulated categorie* *s. Topology, 40(4):667-680, 2001. [28]P. Freyd. The Grothendieck group for stable homotopy is free. Bull. Amer. M* *ath. Soc., 73:84-86, 1967. [29]J. P. C. Greenlees and J. P. May. Derived functors of I-adic completion and* * local homology. Journal of Algebra, 149(2):438-453, 1992. [30]M. J. Hopkins. Global methods in homotopy theory. In E. Rees and J. D. Jone* *s, editors, Homotopy Theory _ Proceedings of the Durham Symposium, pages 73-96. Cambridge University Pre* *ss, 1987. LMSLN 117. [31]M. J. Hopkins, D. C. Ravenel, and W. S. Wilson. Morava Hopf algebras and sp* *aces K(n) equivalent to finite Postnikov systems. In S. O. Kochman and P. Selick, editors, Stable and unsta* *ble homotopy, Fields Institute Communications. AMS, 1998. To appear. [32]M. J. Hopkins and J. H. Smith. Nilpotence and stable homotopy theory. II. A* *nn. of Math. (2), 148(1):1-49, 1998. [33]M. Hovey. Homotopy theory of comodules over a Hopf algebroid, arXiv:math.AT* */0301229. [34]M. Hovey. Model category structures on chain complexes of sheaves. Trans. A* *mer. Math. Soc., 353(6):2441- 2457 (electronic), 2001. [35]M. Hovey, J. H. Palmieri, and N. P. Strickland. Axiomatic stable homotopy t* *heory. Mem. Amer. Math. Soc., 128(610):x+114, 1997. [36]M. Hovey and N. P. Strickland. Morava K-theories and localisation. Mem. Ame* *r. Math. Soc., 139(666):104, 1999. [37]P. T. Johnstone. Stone spaces, volume 3 of Cambridge Studies in Advanced Ma* *thematics. Cambridge University Press, 1982. [38]B. Keller. A remark on the generalized smashing conjecture. Manuscripta Mat* *h., 84(2):193-198, 1994. [39]H. Krause. Decomposing thick subcategories of the stable module category. M* *ath. Ann., 313(1):95-108, 1999. [40]H. Krause. Smashing subcategories and the telescope conjecture_an algebraic* * approach. Invent. Math., 139(1):99-133, 2000. [41]H. Krause. Brown representability and flat covers. J. Pure Appl. Algebra, 1* *57(1):81-86, 2001. [42]H. Krause. On Neeman's well generated triangulated categories. Doc. Math., * *6:121-126 (electronic), 2001. [43]H. Krause. The spectrum of a module category. Mem. Amer. Math. Soc., 149(70* *7):x+125, 2001. [44]H. Krause. A Brown representability theorem via coherent functors. Topology* *, 41(4):853-861, 2002. [45]H. Krause. Coherent functors in stable homotopy theory. Fund. Math., 173(1)* *:33-56, 2002. [46]H. Krause and U. Reichenbach. Endofiniteness in stable homotopy theory. Tra* *ns. Amer. Math. Soc., 353(1):157-173 (electronic), 2001. [47]H. Krause and G. Zwara. Stable equivalence and generic modules. Bull. Londo* *n Math. Soc., 32(5):615-618, 2000. AXIOMATIC STABLE HOMOTOPY _ A SURVEY 17 [48]L. G. Lewis, J. P. May, and M. S. (with contributions by Jim E. McClure). E* *quivariant Stable Homotopy Theory, volume 1213 of Lecture Notes in Mathematics. Springer-Verlag, New Yo* *rk, 1986. [49]H. R. Margolis. Spectra and the Steenrod Algebra. North-Holland, 1983. [50]J. P. May. Equivariant homotopy and cohomology theory. Published for the Co* *nference Board of the Mathe- matical Sciences, Washington, DC, 1996. With contributions by M. Cole, G. Co* *meza~na, S. Costenoble, A. D. Elmendorf, J. P. C. Greenlees, L. G. Lewis, Jr., R. J. Piacenza, G. Triantaf* *illou, and S. Waner. [51]J. P. May. The additivity of traces in triangulated categories. Adv. Math.,* * 163(1):34-73, 2001. [52]J. P. May. Picard groups, Grothendieck rings, and Burnside rings of categor* *ies. Adv. Math., 163(1):1-16, 2001. [53]H. R. Miller. Finite localizations. Boletin de la Sociedad Matematica Mexic* *ana, 37:383-390, 1992. This is a special volume in memory of Jos'e Adem, and is really a book. The editor is * *Enrique Ram'``rez de Arellano. [54]F. Morel and V. Voevodsky. A1-homotopy theory of schemes. Inst. Hautes 'Etu* *des Sci. Publ. Math., (90):45-143 (2001), 1999. [55]A. Neeman. Some new axioms for triangulated categories. Journal of Algebra,* * 139:221-255, 1991. [56]A. Neeman. The chromatic tower for D(R). Topology, 31(3):519-532, 1992. Wit* *h an appendix by Marcel Bökstedt. [57]A. Neeman. The Grothendieck duality theorem via Bousfield's techniques and * *Brown representability. Journal of the American Mathematical Society, 9:205-236, 1996. [58]A. Neeman. On a theorem of Brown and Adams. Topology, 36(3):619-645, 1997. [59]A. Neeman. Brown representability for the dual. Invent. Math., 133(1):97-10* *5, 1998. [60]A. Neeman. Non-compactly generated categories. Topology, 37(5):981-987, 199* *8. [61]A. Neeman. Oddball Bousfield classes. Topology, 39(5):931-935, 2000. [62]A. Neeman. On the derived category of sheaves on a manifold. Doc. Math., 6:* *483-488 (electronic), 2001. [63]A. Neeman. Triangulated categories, volume 148 of Annals of Mathematics Stu* *dies. Princeton University Press, Princeton, NJ, 2001. [64]T. Ohkawa. The injective hull of homotopy types with respect to generalized* * homology functors. Hiroshima Mathematical Journal, 19(3):631-639, 1989. [65]D. C. Ravenel. Morava k-theories and finite groups. In S. Gitler, editor, S* *ymposium on Algebraic Topology in Honor of Jos'e Adem, Contemporary Mathematics, pages 289-292. American Mathe* *matical Society, 1982. [66]D. C. Ravenel. Localization with respect to certain periodic homology theor* *ies. American Journal of Mathe- matics, 106:351-414, 1984. [67]D. C. Ravenel. Nilpotence and Periodicity in Stable Homotopy Theory, volume* * 128 of Annals of Mathematics Studies. Princeton University Press, 1992. [68]J. Rickard. Morita theory for derived categories. J. London Math. Soc. (2),* * 39(3):436-456, 1989. [69]J. Rickard. Splendid equivalences: derived categories and permutation modul* *es. Proc. London Math. Soc. (3), 72(2):331-358, 1996. [70]J. Rickard. Idempotent modules in the stable category. J. London Math. Soc.* * (2), 56(1):149-170, 1997. [71]J. Rickard. Some recent advances in modular representation theory. In Algeb* *ras and modules, I (Trondheim, 1996), volume 23 of CMS Conf. Proc., pages 157-178. Amer. Math. Soc., Provid* *ence, RI, 1998. [72]J. Rickard. Bousfield localization for representation theorists. In Infinit* *e length modules (Bielefeld, 1998), Trends Math., pages 273-283. Birkhäuser, Basel, 2000. [73]S. Schwede. The stable homotopy category has a unique model at the prime 2.* * Adv. Math., 164(1):24-40, 2001. [74]S. Schwede and B. Shipley. Classification of stable model categories, arXiv* *:math.AT/0108143. [75]S. Schwede and B. Shipley. Stable model categories are categories of module* *s. Topology, 42(1):103-153, 2003. [76]B. Shipley. Monoidal uniqueness of stable homotopy theory. Adv. Math., 160(* *2):217-240, 2001. [77]N. P. Strickland. K(N)-local duality for finite groups and groupoids. Topol* *ogy, 39(4):733-772, 2000. [78]N. P. Strickland. Equivariant Bousfield classes. In preparation, 2002. Department of Pure Mathematics, University of Sheffield, Sheffield S3 7RH, UK E-mail address: N.P.Strickland@sheffield.ac.uk