TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY JACK MORAVA Abstract.This very speculative talk suggests that a theory of fundamental groupoids for tensor triangulated categories could be used to describe t* *he ring of integers as the singular fiber in a family of ring-spectra parametriz* *ed by a structure space for the stable homotopy category, and that Bousfield loc* *aliza- tion might be part of a theory of `nearby' cycles for stacks or orbifold* *s. x0, Introduction One of the motivations for this talk comes from John Rognes' Galois theory for structured ring spectra. His paper [36] ends with some very interesting remarks about analogies between classical primes in algebraic number fields and the non- Euclidean primes of the stable homotopy category, and I try here to develop a language in which these analogies can be restated as the assertion that Waldhau* *sen's unfolding specZ ! specS of the integers in the category of brave new rings (or E1 ring-spectra, or com* *mu- tative S-algebras) leads to the existence of commutative diagrams of the form speco__Qp___//specLMUK(n)MU |Gal(__Qp=Qp)| |Dx| fflffl| fflffl| specZp ______//specLK(n)S . The vertical arrow on the left is the Galois cover defined by the ring of integ* *ers in an algebraic closure of the p-adic rationals, but to make sense of the right- hand side would require, among other things, a good theory of structure objects (analogous to the prime ideal spectra of commutative algebra) for some general class of tensor triangulated categories. In this direction I have mostly hopes * *and analogies, summarized in x3. The main result of the first section below, however, is that local classfield t* *heory implies the existence of an interesting system of group homomorphisms __ x ae : W(Q p=Qp) ! D which can plausibly be interpreted as the maps induced on the fundamental groups of these hypothetical structure objects. [The group on the left is Weil's techn* *ical ____________ Date: 6 August 2005. 1991 Mathematics Subject Classification. 11G, 19F, 57R, 81T. The author was supported in part by the NSF. 1 2 JACK MORAVA variant of the Galois group; its topology [30] is slightly subtler than the usu* *al one.] This example is local: it depends on a choice of the prime p. But Rognes [36 x1* *2.2.1] has global results as well; in particular, he identifies the ringspectrum S[BU]* * as the Hopf algebra of functions on an analog Galhotof a Galois group for MU, regarded as an (inseparable) algebraic closure of S. This seems to be in striking agreem* *ent with work of Connes and Marcolli [8] and others [5, 12] on a remarkable `cosmic' generalization of Galois theory, involving a certain motivic Galois group Galmo* *t: there is a very natural morphism S[BU] ! HZ QSymm * of Hopf algebra objects, QSymm * being a certain graded Hopf algebra of qua- sisymmetric functions [19], defined by the inclusion of the symmetric in the qu* *a- sisymmetric functions, and in x2 I suggest that a quotient of this map defines a representation Galmot! Galhot (at least, over Q) which conjecturally plays the role of the homomorphism induc* *ed on fundamental groups by a `geometric realization' construction [33], sending t* *he derived category of mixed Tate motives to some category of complex-oriented spe* *c- tra. The paper ends with a very impressionistic (fauvist?) discussion of a possible * *theory of fundamental groupoids for (sufficiently small) tensor triangulated categorie* *s, which might be flexible enough to encompass both these examples. I am deeply indebted to several mathematicians for conversations about this and related material, eg Bill Dwyer and John Rognes, and in particular Matilde Mar- colli; but none of them bear any responsibility for the many errors it undoubte* *dly contains. x1, Some local Galois representations This section summarizes some_classical local number theory. The first two subse* *c- tions define a system W(Q p=Qp) ! Dx of representations of certain Galois-like groups in the units of suitable p-adic division algebras; then_x1.3_sketches a * *con- jecture about the structure of the absolute inertia group Gal(Q p=Qnrp), which * *says, roughly, that these systems are nicely compatible. 1.0 A local field is a commutative field, with a nontrivial topology in which it is locally compact. The reals or complexes are examples, but I will be concerned here mostly with totally disconnected cases, in particular the fields of charac* *teristic zero obtained as non-Archimedean completions of algebraic number fields. These are finite extensions L of Qp, for some Euclidean prime p; the topology defines a natural equivalence class of valuations, with the elements algebraic over Zp * *as (local) valuation ring. __ The Galois group Gal(Q =Q) of an algebraic closure of the rationals acts on the* * set of prime ideals in the ring of algebraic integers in Q, with orbits correspondi* *ng to the classical primes; the corresponding isotropy groups can be identified with * *the TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 3 __ Galois groups Gal(Q p=Qp) of the algebraic closures of the p-adic rationals. Th* *ese isotropy groups preserve the valuation rings, and hence act on their residue fi* *elds, defining (split) exact sequences __ __ __ 1 ! I(Q p=Qp) ! Gal(Q p=Qp) ! Gal(Fp=Fp) ~=^Z! 0 of profinite topological groups. The cokernel is the closure_of a dense subgrou* *p Z generated by the Frobenius automorphism oe : x 7! xp of Fp, and Weil observed that in some contexts it is more natural to work with the pullback extension __ __ 1 ! I(Q p=Qp) ! W(Q p=Qp) ! Z ! 0 , the so-called Weil group, which is now only locally compact. [These groups are defined much more generally [40] but we won't need that here.] 1.1.1 The work of Lubin and Tate defines very interesting representations of th* *ese groups as automorphisms of certain one-dimensional formal groups. Honda's loga- rithm X i logq(T ) = p-iT q i 0 (with q = pn) defines a formal group law Fq(X, Y ) = log-1q(logq(X) + logq(Y )) 2 Zp[[X, Y ]] (it's not obvious that its coefficients are integral!) whose endomorphism ring * *con- tains, besides the elements T 7! [a](T ) = log-1q(a logq(T )) defined by multiplication in the formal group by a p-adic integer a, the endomo* *r- phism T 7! [!](T ) = log-1q(! logq(T )) = !T defined by multiplication by a (q - 1)st root of unity !. In fact the full rin* *g of endomorphisms of Fq can be identified with the Witt ring W (Fq), which can also be described as the algebraic integers in the unramified extension field Qq obt* *ained from Qp by adjoining !. * *__ 1.1.2 By reducing the coefficients of Fq modulo p we obtain a formal group law * *Fq over Fp, which admits T 7! F (T ) = T p as a further endomorphism. Honda's group law is a Lubin-Tate group [37] for the field Qq, and it can be shown that [p](T ) T qmod p ; __ in other words, F n= p in the ring of endomorphisms of F q. From this it is not hard to see that __ End _Fp(F q) = W (Fq)=(F n- p) (the pointed brackets indicating a ring of noncommutative indeterminates, subje* *ct to the relation aoeF = F a for a 2 W (Fq), with oe 2 Gal(Fq=Fp) ~=Z=nZ being the (other!) Frobenius endo- morphism). This is the ring oD of integers in the division algebra D = Qq=(F n- p) 4 JACK MORAVA with_center Qp; its group oxDof strict units is the full group of automorphisms* * of Fq, and it is convenient to think of 1 ! oxD! Dx ! Z ! 0 as a semidirect product, with quotient generated by F . Thus conjugation by F a* *cts on W (Fq)x oxDas oe 2 Gal(Fq=Fp): the Galois action is encoded in the division algebra structure. 1.2.1 For each n 1, a deep theorem of Weil and Shafarevich defines a homomor- phism wsq : W(Qabq=Qp) ~=W (Fq)x n Z ! oxDn Z = Dx of locally compact topological groups. It extends the local version of Artin's * *reci- procity law, which defines an isomorphism Lx ~=W(Lab=L) for any totally disconnected local field L (with Lab its maximal abelian extens* *ion). When L = Qq is unramified, the sequence 1 ! W(Qabq=Qq) ! W(Qabq=Qp) ! W(Qq=Qp) ! 0 is just the product 0 ! W (Fq)x x Z ! W (Fq)x n Z ! Z=nZ ! 0 of the elementary exact sequence 0 ! Z ! Z ! Z=nZ ! 0 with a copy of the units in W (Fq): in the semidirect product extension above, * *the generator 1 2 Z acts on W (Fq) by oe, so its nth power acts trivially. 1.2.2 More generally, any Galois extension L of degree n over Qp can be embedded in a division algebra of rank n with center Qp, as a maximal commutative subfie* *ld. When the division algebra D has invariant 1=n in the Brauer group Q=Z of Qp (as it does in our case: this invariant equals the class, modulo Z, of the p-order * *of an element (e.g. F ) generating the maximal ideal of oD ), the theorem [44, append* *ix] of Weil and Shafarevich defines an isomorphism of the normalizer of Lx in Dx wi* *th the Weil group of Lab over Qp. [The maximal abelian extension Labis obtained by adjoining the p-torsion points* * of the Lubin-Tate group of L to the maximal unramified extension Lnr; the resulting field acquires an action of the group oxLof units of L (by `complex multiplicat* *ion' on the Lubin-Tate group) together with an action of the automorphisms ^Zof the algebraic closure of the residue field. The point is that the Lubin-Tate group of L is natural in the etale topology: t* *o be precise, any two Lubin-Tate groups for L become isomorphic_over the completion Lnr of the maximal unramified extension Lnr = L W(k) W (k) of L [38 x3.7]. Since an automorphism of L over Qp takes one Lubin-Tate group to another, the resulting group of automorphisms of `the' Lubin-Tate group of L (as a completed Hopf algebra over Zp) is an extension of Gal(L=Qp) by oxLx ^Z, i.e. the profini* *te completion of Lx . This extension is classified by an element of H2(Gal(L=Qp), Lx ) ~=Z=nZ , TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 5 which also classifies those algebras, simple with center Qp, which split after * *ten- soring with L. The extension in question generates this group, by a fundamental result of local classfield theory; but the division algebra with invariant 1=n * *also generates this group, and the associated group extension is the normalizer of L* *x in Dx , QED.] 1.2.3 The remarks in this subsession are a digression, but they will be useful * *in x3.3: Since the normalizer of Lx acts as generalized automorphisms of a Lubin-T* *ate group for L, then for every g 2 W(Lab=Qp) there is a power series [g](T ) 2 oLnr[[T ]] satisfying [g0]([g1](T )) = [g0g1](T ), compatible with the natural action of t* *he Weil group W(Lab=Qp) on the ring of integers oLnr in Lnr. It follows that [g](t) = ff(g)T + . .d.efines a crossed homomorphism ff : W(Lab=Qp) ! (Lnr)x , ie a map satisfying ff(g0g1) = ff(g0) . ff(g1)g0, the superscript denoting the * *action of the Weil group on Lnr through its quotient W(Lnr=Qp). When L = Qq this implies the existence of an extension of the identity homomorphism from Qxqto itself, to a crossed homomorphism from Qxqn Z to (Lnr)x . A corollary is the existence of a representation Qnrp(1) of W(Qabq=Qp) on the completion Qnrpextending the action* * of W (Fq)x by multiplication. [The completions in this construction are cumbersome, and might be unnecessary. Experts may know how to do without them, but I don't.] 1.3 The representations promised in the introduction are the compositions __ ab x aeq : W(Q p=Qp) ! W(Qq =Qp) ! D with the second arrow coming from the Weil-Shafarevich theorem. In the remainder of this section I will sketch a conjecture about the relations* * be- tween these representations. The argument goes back to Serre's Cohomologie Ga- loisienne, and I believe that many people have thought along the lines below, b* *ut I don't know of any place in the literature where this is spelled out. It is b* *ased on a p-adic analog of a conjecture of Deligne, related to an older conjecture of Shafarevich [33; cf. also 14]. 1.3.1 We need some basic facts: i) in order that a pro-p-group be free, it is necessary and sufficient that its* * coho- mological dimension be 1 [37 I x4.2, corollary 2], and ii) the maximal unramified extension of a local field with perfect residue fiel* *d has cohomological dimension 1 [II x3.3 ex c]. I will follow Serre's notation, which is very similar to that used above. K is* * a field complete with respect to a discrete valuation, with residue field k, e.g.* * a finite extension of Qp. Knr will denote its maximal unramified extension, Ks its (separable) algebraic closure, and Ktr will be the union of the tamely ramified Galois extensions (i.e. with Galois group of order prime to p) of K in Ks; thus Ks Ktr Knr K . 6 JACK MORAVA Eventually K will be the quotient field Qq of the ring W (k) of Witt vectors for some k = Fq with q = pf elements, but for the moment we can be more general. We have Y Gal(Ktr=Knr) ~=lim-Fxpn~= Zi= ^Z(1)(:p) l6=p __ [II x4 ex 2a]; this is a module over Gal(k=k) ~=^Z, in which 1 2 Z acts as mult* *ipli- cation by q = #(k) [II x5.6 ex 1]. The kernel in the extension 1 ! Gal(Ks=Ktr) = P ! Gal(Ks=Knr) ! Gal(Ktr=Knr) ! 1 is a pro-p-group (ibid 2b), closed in a group of cohomological dimension one (by assertion (ii)), hence itself of cohomological dimension one [I x3.3 propositio* *n 14], hence free by assertion (i). __ The extension in question is the inertia group of K; it has a natural Gal(k=k)- action. 1.3.2 From now on I will assume that K = Qq is unramified over Qp. Conjecture I, This extension splits: Gal(Ks=Knr) ~=P o ^Z(1)(:p) is a semidirect product. [Since the kernel and quotient have relatively prime order, this would be obvio* *us if either were finite. This may be known to the experts.] Let W (k)x0= (1 + pW (k))x be the group of those units of W (k) congruent to 1 mod p: the logarithm defines an exact sequence 1 ! kx ! W (k)x ! W (k) ! k ! 0 , so taking W (k)x0isomorphically to pW (k). Let __x 0 x 0 __ W(k)0 = lim-{W (k )0 | k finite k} __x be the limit under norms; it is a compactification of W (k)0 . Conjecture_II: The topological abelianization of P is naturally isomorphic to W(k)x0. Alternately: Lazard's group ring __ Zp[[P ]] ~=lim-{T^(W (k0)x0) | k0finite k} of P of is isomorphic to the limit (under norms) of the system of completed ten* *sor algebras of pW (k0) ~=W (k0)x0. It is thus (hypothetically) a kind of noncommut* *ative Iwasawa algebra [I x1.5 proposition 7 (p. 8)]. If both these conjectures are true, then I can think of no natural way for ^Z(1* *)(:p) to act on P , so I will go the rest of the way and conjecture as well that this* * action is trivial. I will also abbreviate the sum of these conjectures as the assertio* *n that the absolute inertia group I_of_Qp is the_product of ^Z(1)(:p) with the pro-free pro-p-group generated by W (k)x0~=pW (k). TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 7 1.3.3 These conjectures seem to be compatible with other known facts of classfi* *eld theory. In particular, they would imply that __ Gal(Q p=Qq) ~=I n ^Z with 1 2 Z acting as oe(x) = xq on k. This abelianizes to W (k)x x ^Z, agreeing* * with Artin's reciprocity law, and if q1 = qm0this would yield an exact sequence __ __ 1 ! Gal(Q p=Qq1) ~=I n ^Z! Gal(Q p=Qq0) ~=I n ^Z! Gal(Fq1=Fq0) ~=Z=mZ ! 0 . Finally, the composition __x x I ! Iab ~=W(k)0 x ^Z(1)(:p) ! W (Fq) defines a compatible system of candidates for the quotient maps __ x ab aeq : Gal(Q p=Qp) ~=I n ^Z! W (Fq) n ^Z~=Gal(Qq =Qp) . x2, Some more global representations This section is concerned with two very global objects, each of which has some unfamiliar features. The first subsection is concerned with the suspension spec* *trum S[BU] = 1 (BU+ ) of the classifying space for the infinite unitary group, and its interpretation* * (follow- ing [36]) as a Hopf algebra object in the category of spectra. The second subse* *ction reviews some properties of the Hopf algebra of quasisymmetric functions, follow* *ing [5, 23]. A remarkable number of the properties of the Hopf algebra of symmetric functions generalize to this context [20], and I will try to keep this fact in * *focus. This Hopf algebra is conjectured to be closely related to a certain motivic gro* *up of interest in arithmetic geometry, and I have tried to say a little about that* *, in particular because it seems to overlap with recent work of Connes and Marcolli, discussed in x2.3. 2.1.1 Since BU is an infinite loopspace, S[BU] becomes an E1 ringspectrum (or, in an alternate language, a commutative S-algebra); but BU is also a space, with a diagonal map, and this structure can be used to make this ringspectrum into a Hopf algebra in the category of S-modules. In complex cobordism, the complete Chern class [1] X ct= cI tI = OE-1stOE(1) 2 MU*(BU+ ) S* (where S* = Z[ti| i 1] is the Landweber-Novikov algebra, with coaction st) re* *p- resents a morphism S[BU] ! MU ^MU of ringspectra. Indeed, algebra morphisms from S[BU] to the Thom spectrum MU correspond to maps CP+1 = MU(1) ! MU of spectra, and hence to elements of MU*(CP+1). The coaction MU ~=S ^ MU ! MU ^ MU is also ring map, so the resulting composition S[BU] ! MU = S ^ MU ! MU ^ MU 8 JACK MORAVA is a morphism of ring spectra. On the other hand the Thom isomorphism OE : MU*(BU+ ) ~=MU*(MU) satisfies Y OE-1stOE(1) = ct= e-1it(ei) P i (where t(e) = tkek+1 2 MU*(CP+1) S*, with e being the Euler, or first Chern, class). In fact this map is also a morphism of Hopf algebra objects: st0( ct) = ct0Ot ct0Ot2 MU*(BU+ ^ BU+ ) (S* S*) , so the diagram S[BU] _______//S[BU] ^ S[BU]_____//_(MU ^ MU) ^ (MU ^ MU) | | | | fflffl| |fflffl MU ^ MU ________________________//(MU ^ MU) ^MU (MU ^ MU) commutes. We can thus think of S[BU] as a kind of Galois group for the category of MU- algebras over S, or alternately for the category of complex-oriented multiplica* *tive cohomology theories. In particular, if E is an MU-algebra, algebra maps from S[BU] to E define elements of E0(CP+1); thus the group Aut(E) of multiplicative automorphisms of E maps to algebra homomorphisms from S[BU] to E. The ad- joint construction thus sends S[BU] to the spectrum of maps from Aut(E) (regard* *ed naively, as a set) to E; but these maps can be regarded as E*-valued functions * *on Aut(E), and hence as elements of the coalgebra E*E. Note that S[BU] is large, so the existence of an honest dual object in the cate- gory of spectra may be problematic. The remark above implies that this perhaps nonexistent group object admits the etale groupschemes oxDof x1 as subgroups. 2.1.2 The Landweber-Novikov algebra represents the group of invertible power series under composition, and it may be useful below to know that (since all on* *e- dimensional formal groups over the rationals are equivalent) (H*(MU, Q), H*(MU ^ MU, Q)) represents the transformation groupoid defined by this group acting on itself by translation. On the other hand H*(BU, Q) represents the group of for- mal power series with leading coefficient 1, under multiplication, and the indu* *ced morphism spec(H*(MU, Q), H*(MU ^ MU, Q)) ! spec(H*(S, Q), H*(BU, Q)) of groupoidschemes sends the pair (g, h) of invertible series, viewed as a morp* *hism from h to gOh, to the translated derivative g0(h(t)). This is essentially the c* *onstruc- tion which assigns to a formal group, its canonical invariant differential. The* * chain rule ensures that this is a homomorphism: we have (g, h) O (k, g O h) = (k O g,* * h), while g0(h(t)) . k0((g O h)(t)) = (k O g)0(h(t)) . 2.2.1 The Z-algebra of symmetric functions manifests itself in topology as the integral homology of BU; it is a commutative and cocommutative Hopf algebra TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 9 (with a canonical nondegenerate inner product [20] which, from the topological point of view, looks quite mysterious). Over Q it is the universal enveloping a* *lgebra of an abelian graded Lie algebra with one generator in each even degree. The graded ring QSymm *of quasi symmetric functions is the commutative Hopf algebra dual to the free associative algebra on (noncommutative!) generators Zk* * of degree 2k, with coproduct X Zk = Zi Zk ; i+j=k it is thus the universal enveloping algebra for a free graded Lie algebra fZ, w* *ith one generator in each (even) degree [19]. There is a natural monomorphism embedding the symmetric functions in the quasisymmetric functions [5 x2.4], dual to the m* *ap on enveloping algebras defined by the homomorphism from the free Lie algebra to its abelianization. This then defines a morphism S[BU] ! HZ QSymm * of (Hopf) ringspectra. The group-valued functors represented by such Hopf algebras are very interestin* *g, and have a large literature. In what follows I will simplify by tensoring every* *thing with Q, which will be general enough for anything I have to say. The formal (Magnus) completion of the Q-algebra of noncommuting power series is dual to the algebra of functions on a free prounipotent [10 x9] groupscheme F. In this cont* *ext, a grading on a Lie algebra can be reinterpreted as an action of the multiplicat* *ive groupscheme Gm , which sends an element x of degree d to ~dx, where ~ is a unit in whatever ring we're working with, so we can describe the map above as defini* *ng a morphism F n Gm ! Galhot of group objects of some sort, over Q. 2.2.2 In arithmetic geometry there is currently great interest in a groupscheme Galmot= Foddn Gm which is conjectured to be (isomorphic to) the motivic Galois group of a certain Tannakian category, that of mixed Tate motives over Z [10, 12]. Usually Fodd is taken to be the prounipotent group defined by the free graded Q-Lie algebra fod* *don generators of degree 4k + 2 (hence `odd' according to the algebraists' conventi* *ons), with k 1; but there are reasons to allow k = 0 as well. We can regard Fodd as* * a subgroupscheme of F, by regarding fodd as a subalgebra of f. A Tannakian category is, roughly, a suitably small k-linear abelian category wi* *th tensor product and duality - such as the category of finite-dimensional linear * *rep- resentations of a proalgebraic group over a field k. Indeed, the main theorem o* *f the subject [11] asserts that (when k has characteristic zero) any Tannakian catego* *ry is of this form; then the relevant group is called the motivic group of the cat* *egory. Present technology extracts mixed Tate motives from a certain triangulated cate- gory of (pieces of) algebraic varieties, constructed as a subcategory of the te* *nsor triangulated category of more general motives [12, 43]. 10 JACK MORAVA There are more details in x3 below, but one of the points of this paper is that* * the language of such tensor categories can be quite useful in more general circumst* *ances. For example, the category of complex-oriented multiplicative cohomology theories behaves very much like (a derived category of) representations, with S[BU] as i* *ts motivic group. Away from the prime two, the fibration SO=SU+ ! BSU+ ! BSO+ (defined by the forgetful map from C to R) splits (even as maps of infinite loo* *pspaces). Over the rationals, this is almost trivial: it corresponds to the splitting of * *the graded abelian Lie algebra with one generator in each even degree, into a sum of two s* *uch Lie algebras, with generators concentrated in degrees congruent to 0 and 2 mod 4 respectively. The composition (the first arrow is the projection of the H-spa* *ce splitting, and the second corresponds to the abelianization of a graded free Lie algebra). S[BU] ! S[SO=SU] ^ S[CP 1] ! HQ QSymm odd is the candidate, promised in the introduction, for a natural representation Galmot! Galhot (over Q, of course!). 2.2.3 The degree two (= 2(2 . 0 + 1)) generator in Fodd is closely related to t* *he S[CP 1] factor in the ring decomposition above; both correspond to exceptional cases that deserve some explanation. The symmetric functions are defined formally as an inverse limit of rings of sy* *m- metric polynomials in finitely many variables xn, n 1. There are thus in- teresting maps from the symmetric functions to other rings, defined by assigning interesting values to the xn; but because infinitely many variables are involve* *d, issues of convergence can arise. For example: if we send xn to 1=n, the nth pow* *er sum pn maps to i(n) 2 R . . . as long as n > 1, for s = 1 is a pole of i(s). This is a delicate matter [5 x2.7, 23], and it turns out to be very natural to * *send p1 to Euler's constant fl. In fact this homomorphism extends, to define a homomorphism from QSymm to R, whose image is sometimes called the ring MZN * of `multizeta numbers'; it has a natural grading. It is classical that for any positive integ* *er n, i(2n) = -1_2B2n fl2n(2ssi) 2 Q(ss) , (where flk denotes the kth divided power) and in some contexts it is natural to* * work with the even-odd graded subring of C obtained by adjoining an invertible eleme* *nt (2ssi) to MZN *. It is known [18 x4] that the multizeta numbers are periods of algebraic integrals, and that the ring generated by all such periods is a Ho* *pf algebra, closely related to the algebra of functions on the (strictly speaking,* * still hypothetical) motivic group of all motives over Q [28]; but it is thought that * *Euler's constant is probably not a period. Nevertheless, from the homotopy-theoretic po* *int of view presented here, it appears quite naturally. 2.3.1 These multizeta numbers may play some universal role in the general theory of asymptotic expansions; in any case, they appear systematically in renormaliz* *a- tion theory. Connes and Marcolli, building on earlier work of Connes, Kreimer TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 11 [7], Broadhurst, and others, have put this in a Galois-theoretic framework. Th* *is subsection summarizes some of their work. Classical techniques [of Bogoliubov, Parasiuk, Hepp, and Zimmerman] in physics have achieved an impressive level of internal consistency, but to mathematicians they lack conceptual coherence. Starting with a suitable Lagrangian density, Co* *nnes and Kreimer define a graded Lie algebra g* generated by a class of Feynman grap* *hs naturally associated to the interactions encoded by the Lagrangian. They interp* *ret the BPHZ dimensional regularization procedure as a Birkhoff decomposition for loops in the associated prounipotent Lie group G, and construct a universal rep- resentation of this group in the formal automorphisms of the line at the origin, yielding a formula for a reparametrized coupling constant which eliminates the divergences in the (perturbative) theory defined by the original Lagrangian. Dimensional regularization involves regarding the number d of space-time dimen- sions as a special value of a complex parameter; divergences are interpreted as poles at its physically significant value. Renormalization is thus expressed in* * terms much like the extraction of a residue, involving a simple closed curve encircli* *ng the relevant value of d (the source of the loop in G mentioned above). Connes and Marcolli [8] reformulate such data in geometric terms, involving flat connectio* *ns on a Gm -equivariant G-bundle over a certain `metaphysical' (not their terminology* *!) base space B. From this they define a Tannakian category (of flat, equisingular vector bundles with connection over B), and identify its motivic group as F n G* *m . Their constructions define a representation of the motivic group in G, and henc* *e in the group of reparametrizations of the coupling constant. 2.3.2 Besides the complex deformation of the space-time dimension, the base spa* *ce B encodes information about the mass scale. It is a (trivial, but not naturally trivialized, cf end of x2.13) principle bundle Gm ! B ! over a complex disk centered around the physical dimension d 2 C. [In this disk is treated as infinitesimal but there may be some use in thinking of it as* * the complement of infinity in CP1.] The renormalization group equations [x2.9] are reformulated in terms of the (mass-rescaling) Gm -action on the principal bundle G x B (the grading on g* endows G with a natural Gm -action), leading to the existence of a unique gauge-equivalence class of flat Gm -equivariant connection forms ~ 2 1(B, g), which are equisingular in the sense that their restrictions* * to sections of B (regarded as a principal bundle over ) which agree at 0 2 are mutually (gauge) equivalent. A key result [8, Theorem 2.25, x2.13] characterizes such forms ~ in terms of a * *graded element fi* 2 g* corresponding to the beta-function of renormalization group th* *eory. In local coordinates (z 2 near the basepoint, u 2 Gm ) we can write ~(z, u) = ~0(z, u) . dz + ~1(z, u) . u-1du with coefficients ~i 2 C{u, z}[z-1] allowed singularities at z = 0; but flatness and equivariance imply that these coefficients determine each other. The former condition [2.166] can be stated as @z~1 = H~0 - [~0, ~1] 12 JACK MORAVA where the grading operator H = (u@u)|u=1 is the infinitesimal generator of the Gm -action; it sends uk to kuk. Regularit* *y of ~0 at u = 0 implies that ~0 = H-1[@z~1 + [H-1@z~1, ~1] + . .]. is determined, at least formally, by knowledge of ~1 in the fiber direction. Co* *nnes and Marcolli's solution [Theorems 2.15, 2.18, eq. 2.173] of the renormalization group equations imply that ~1(u, z) = -z-1u*(fi) for some unique fi 2 g. This characterizes a universal (formal) flat equisingu* *lar connection ~(fi) on B, related to the universal singular frame of [x2.14]. 2.3.3 The Lie algebra V of the group of formal diffeomorphisms of the line at t* *he ori- gin has canonical generators vk = uk+1@u satisfying [vk, vl] = (l - k)vk+1, k, * *l 1, so a graded module with an action of such operators defines a flat equisingular* * vec- tor bundle over B with fi* = v*. A commutative ringspectrum E with S[BU]-action defines a multiplicative complex-oriented cohomology theory, and in particular * *pos- sesses an MU-module structure. On rationalized homotopy groups the morphism E ! E ^ S ! E ^ MU defines an S*-comodule structure map (cf. x2.1.2 above) E*Q! E*Q MU*QMU*MUQ = E*Q S* and thus an action of the group of formal diffeomorphisms; differentiating this* * ac- tion assigns E*Qan action of V. The smash product of two complex-oriented spect* *ra is another such thing, and the functor from S[BU]-representations to V represen* *ta- tions takes this product to the usual tensor product of Lie algebra representat* *ions. This, together with the universal connection constructed above, defines a monoi* *dal functor from S[BU]-representations to the Tannakian category of flat equisingul* *ar vector bundles over B of [x2.16], and thus a (rational) representation of its m* *otivic group in S[BU]. Connes and Marcolli note that their motivic group is isomorphic to the motivic group for mixed Tate motives over the Gaussian integers (which has generators in all degrees, unlike that for mixed Tate motives over the rationals), but they d* *o not try to make this isomorphism canonical. It is striking to me that their formul* *as (cf eq. 2.137) actually take values in the group of odd formal diffeomorphisms. Presumably Gal(Q(i)=Q) acts naturally on the motivic group of mixed Tate mo- tives over the Gaussian integers, and it seems conceivable that these construct* *ions might actually yield a representation of Foddn Gm in the group of odd formal diffeomorphisms [33]; but I have no hard evidence for this. x3, Towards fundamental groupoids of tensor triangulated categories 3.0 A Tannakian category A is a k-linear abelian category, where k is a field, possessing a coherently associative and commutative tensor product [11 x2.5]; moreover, A should be small enough: its objects should be of finite length, its Hom-objects should be finite-dimensional over k, the endomorphism ring of the TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 13 identity object for the tensor product should be k, and it should admit a good internal duality [x2.12]. The specifically Tannakian data, however, consists o* *f a nontrivial exact k-linear functor ! : A ! (k - Vect) which is monoidal in the sense that !(X Y ) ~=!(X) k !(Y ) . Let Aut!(k) be the group of multiplicative automorphisms of !, ie of natural tr* *ans- formations of k-module-valued functors from ! to itself, which commute with the* *se multiplicativity isomorphisms. More generally, the multiplicative automorphisms A 7! Aut!(A) = Aut(! k A) of ! kA define a group-valued functor on the category of commutative k-algebra* *s. It sometimes happens (eg when k is of characteristic zero) that this functor is representable by a suitable Hopf algebra H, ie Aut (! k A) ~=Hom k-alg(H, A) ; then ! lifts to a functor from A to the category of finite-dimensional represen* *tations of the affine (pro)algebraic group represented by H, and it may be that we can * *use this lift to identify A with such a category of representations [x7]. In any c* *ase, when the k-groupscheme specH = ss1(specA, !) exists, it is natural to think of it as a (`motivic') kind of fundamental group* * for the Tannakian category, with the `fiber functor' ! playing the role of basepoint. 3.1 Here are some illustrative examples, and variations on this theme: 3.1.1 The category of local systems (ie of finite-dimensional flat k-vector spa* *ces) over a connected, suitably locally connected topological space X is Tannakian: a basepoint x 2 X defines an exact functor, which sends the system to its fiber o* *ver x, and we recover the fundamental group ss1(X, x) (or, more precisely, its `env* *elope' or best approximation by a proalgebraic group over k) as its automorphism object [41]. 3.1.2 The functor which assigns to a finite-dimensional Q-vector space V_,_the * *p-adic vector space V Q Qp, defines a Tannakian structure on Q-Vect, with Gal(Q p=Qp), regarded as a profinite groupscheme over Qp, as its motivic fundamental group. 3.1.3 One possible variation involves fiber functors taking values in categorie* *s more general than vector spaces over a field. The monoidal category of even-odd grad* *ed (or `super') vector spaces is an important example. When this works, one gets a Hopf algebra object (corresponding to a `super' groupscheme) in the enriched category; the motivic group in this case is the multiplicative group ~2 of squa* *re roots of unity (suitably interpreted, [11 x8.19]). Perhaps in homotopy theory t* *his has something to do with the fact that the first stable homotopy group of the s* *phere spectrum is Z=2Z. 14 JACK MORAVA 3.1.4 The automorphisms of cohomology with Fp coefficients, which is defined not on an abelian category but on a tensor triangulated one, eg finite spectra, is a more exotic and perhaps more compelling example: the even-degree cohomology is a representation of the groupscheme defined by the functor X k A 7! { akT p = a(T ) 2 A[[T ]] | a0 2 Ax } ; k 0 its Hopf algebra is dual to the algebra defined by Steenrod's reduced pth power* *s. The group operation here is composition of power series. The Hopf algebra dual to the full Steenrod algebra can be recovered by working with ~2-graded (super) vector spaces. Note that the groupscheme defined above contains the `torus' A 7! Ax = Hom k-alg(k[t01], A) as the subgroup of series of the form a0T, a0 2 Ax ; this action by the multipl* *ica- tive groupscheme Gm allows us to recover the (even part of) the grading on the cohomology in `intrinsic' terms. However, we're not as lucky here as in the pre* *ced- ing case: the stable homotopy category (at p) is not the same as the category of modules over the Steenrod algebra. The Adams spectral sequence tells us that we have to take higher extensions into consideration. 3.1.5 These motivic fundamental groups are in a natural sense functorial [11 x8* *.15]: Given a commutative diagram j A0 ___________//_A1 !0|| |!1| fflffl|"j fflffl| (k0 - Vect)____//(k1 - Vect) of Tannakian categories Ai, fiber functors !i, and fields k0 ! k1, with j exact k0-linear multiplicative and "j= - k0k1, there is a natural homomorphism j* : ss1(specA1, !1) ! ss1(specA0, !0) xk0k1 of groupschemes over k1 constructed as follows: if ff : !1 ~=!1 is a multiplica* *tive automorphism of !1 (perhaps after some base extension which I won't record), th* *en ff O j : "jO !0 = !1 O j ~=!1 O j = "jO !0 is an element of ss1(specA0, !0), pulled back by "jto become a groupscheme over k1. As x3.1.4 suggests, it is tempting to push these constructions in various direc* *tions; in particular, if T is a suitably small tensor triangulated category (ie a tria* *ngulated category with tensor product satisfying reasonable axioms (for example that ten- soring with a suitable object represents suspension)), and ! is a multiplicat* *ive homological functor (taking distinguished triangles to long exact sequences) wi* *th values in k-vector spaces, we can consider its multiplicative automorphisms, as* * in the Tannakian case; and we might hope that if T is some kind of derived category of A, then we could try to reconstruct T in terms of [13, 34] a derived category of representations of the automorphism group of the homological fiber functor !. [If we ask that the automorphisms behave reasonably under suspension, then the automorphism group will contain a torus encoding gradings, as in x3.1.4.] TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 15 3.2 What seems to be missing from this picture is a compatible understanding of ss0. If p is a prime ideal in a commutative noetherian ring A then A=p is a dom* *ain with quotient field Q(A=p), and the composite !p : A ! A=p ! Q(A=p) = k(p) determines p, so the prime ideal spectrum specA is a set of equivalence classes* * of homomorphisms from A to (varying) fields k. Such a homomorphism lifts to define* * a triangulated functor LAk(p) from the derived category of A-modules to the deri* *ved category of vector spaces over k(p), taking (derived) tensor products to (deriv* *ed) tensor products, and (as in the case of ordinary cohomology above) we can consi* *der the functor of automorphisms associated to such a generalized point. G-equivari* *ant stable homotopy theory is another example of a category with a plentiful supply* * of natural `points', corresponding to conjugacy classes of closed subgoups of G. It would be very useful if we could construct, for suitable T, a fundamental gr* *oupoid ss(specT) with objects corresponding to equivalence classes of multiplicative h* *omo- logical fiber functors !, and morphisms coming from ss1(specT, !). The lattice * *of Bousfield localizations [24] of a triangulated category is in some ways analogo* *us to the Boolean algebra of subsets of the spectrum of an abelian category [3, 21, 3* *5], but we do not seem to understand, in any generality, how to identify in it a su* *blat- tice corresponding to the open sets of a reasonable topology. Moreover, there s* *eem to be subtle finiteness issues surrounding this question: constructing a good s* *s1 at ! may require identifying a suitable subcategory of !-finite or coherent object* *s [26 x8.6], and we might hope that a reasonable topology on specT would suggest a natural theory of adeles [27] associated to chains of specializations. 3.3 This is all quite vague; perhaps some examples will be helpful. The structure map S ! HZ defines a monoidal pullback functor - ^ HZ : (Spectra)! D(Z - Mod ) ; simply regarding a Z-algebra as an S-algebra is not monoidal: H(A B) is not t* *he same as HA ^ HB. 3.3.1 Let ^K*p(-) = K*(-) Zp denote classical complex K-theory, regarded as a functor on finite spectra and p-adically completed. The Adams operations, suita* *bly normalized, define an actiion of Zxpby stable multiplicative automorphisms [39], and the Chern character isomorphism ^K*p(-) Q ! H*(-, Qp) identifies its eigenspaces with the graded components defined by ordinary cohom* *ol- ogy. This can be_expressed in the language developed above, as follows: the fiber fu* *nctor !0 = K^p Qp on finite spectra_is ordinary cohomology in disguise, and so has automorphism group Gm n Gal(Q p=Qp), as does __ __ !1 : M 7! H+ (M Qp) : D(Z - Mod ) ! (Q p- Vect) . The induced morphism __ __ Gm n Gal(Q p=Qp) ! Gm n Gal(Q p=Qp) 16 JACK MORAVA __ of motivic fundamental groups sends (u, g) 2 Gm n Gal(Q p=Qp) to (gabu, g), whe* *re __ __ x g 7! gab : Gal(Q p=Qp) ! Gal(Q p=Qp)ab ! Zp is the p-adic cyclotomic character; in Hopf algebra terms, this corresponds to * *the homomorphism u 7! 1 - t : Qp[u 1] ! Zp[[t]][~p-1] Q ~=Zp[[Zxp]] Q of Iwasawa theory. 3.3.2 This has a conjectural generalization to the spectra obtained by speciali* *zing En, with E*n(S) = Zp[[v1, . .,.vn-1]][u 1] at vi7! 0, 1 i n - 1. [These spectra correspond to the Honda formal group of x1.1.1; they have some kind [15] of multiplicative structure. It is convenient * *to call the associated cohomology functors K(n)*(-; Zp).] It is natural [32] to expect that the group of multiplicative automorphisms of * *this functor maps, under reduction modulo p, to the normalizer of the maximal torus * *Qxq in Dx , as in x1.2.1; in other words, precisely those automorphisms of K(n)*(-;* * Fp) which lift to automorphisms of the formal group law of K(n)*(-; Zp) lift to mul* *ti- plicative automorphisms of the whole functor. It would follow from this, that K(n)*(-; Zp) ZpQnrp~= k2ZH*(-; Qnrp(1) k) , __ where Qnrp(1) is the representation of Gal(Qabq=Qp) (and thus of Gal(Q p=Qp)) c* *on- structed in x1.2.3. The results of the preceding section would then generalize,* * with the role of the cyclotomic character now being played by the crossed homomorphi* *sm ff. 3.4 This suggests that the remarks above about ss0 might be slightly naive: ss0 might make better sense as a small diagram category than as a topological space. Equivariant stable homotopy theory provides further evidence for this, as do de* *rived categories of representations of quivers in groups. The conjecture above asserts that the functor A 7! Aut(K(n)*(-; Zp) ZpA) : (flatA - Mod ) ! (Groups) is i) representable ii) by Gal(Qabq=Qp). Enlarging the useable class of fiber f* *unctors in this way, allowing values in (say) modules over discrete valuation rings, wo* *uld lead to the existence of specialization homomorphisms such as ss1(specZp) ! ss1(specS, K(n)*(-; Zp)) ! ss1(specS, K(n)*(-; Fp)) . This paper is so much a wish list, that while we're at it we might as well ask * *for a way to associate automorphism groups to suitable open subsets or subdiagrams of our hypothetical ss0, together with homomorphisms between them defined by inclusion* *s. One of the beauties of Grothendieck's original account of the fundamental group is his theory of specialization [16 X x2.3] which, as he notes, has no immediate analog in pure algebraic topology. There is evidence [2, 4 x6, 22, 42 x4] that * *some generalization of his theory to a derived category context could accomodate a v* *ery general form of the theory of vanishing cycles [17, 9]. TOWARD A FUNDAMENTAL GROUPOID FOR THE STABLE HOMOTOPY CATEGORY 17 References 1. J.F. Adams, Stable homotopy and generalised homology Chicago Lectures in Mat* *he- matics (1995) 2. M. Ando, J. Morava, H. Sadofsky, Completions of Z=(p)-Tate cohomology of per* *iodic spectra, Geom. Topol. 2 (1998), 145-174 3. Paul Balmer, The spectrum of prime ideals in tensor triangulated categories,* * available at math.CT/0409360 4. M. B"okstedt, A. Neeman, Homotopy limits in triangulated categories, Composi* *tio Math. 86 (1993), 209-234. 5. P. Cartier, Functions polylogarithms, nombres polyz'etas et groupes prounipo* *tents, Sem. Bour- baki 885, in Asterisque 282 (2002) 6. D. Christensen, M. Hovey, Phantom maps and chromatic phantom maps, Amer. J. * *Math. 122 (2000) 275-293. 7. A. Connes, D. Kreimer, Renormalization in quantum field theory and the Riema* *nn-Hilbert problem I: the Hopf algebra structure of graphs and the main theorem; II, : * *the fi-function, diffeomorphisms and the renormalization group; available at hep-th/9912092, * *0003188 8. __-, M. Marcolli, From physics to number theory via noncommutative geometry,* * II: renor- malization, the Riemann-Hilbert correspondence, and motivic Galois theory, a* *vailable at hep-th/0411114 9. K. Consani, Double complexes and Euler L-factors, Compositio Math. 111 (1998* *), 323-358. 10.P. Deligne, Le groupe fondamental de la droite projective moins trois points* *, in Galois groups over Q, MSRI Publ. 16 (79 - 313) 1989 11.__-, Categories Tannakiennes, in the Grothendieck festschrift Vol. II, 111-1* *95, Progr. Math., 87, Birkh"auser (1990) 12.__-, A. Goncharov, Groupes fondementaux motiviques de Tate mixte, available * *at math.NT/0302267 13.J. Franke, Uniqueness theorems for certain triangulated categories with an A* *dams spectral sequence, available at www.maths.abdn.ac.uk/~bensondj/html/archive/franke.ht* *ml 14.H. Furusho, p-adic multiple zeta values. I. p-adic multiple polylogarithms a* *nd the p-adic KZ equation, Invent. Math. 155 (2004) 253 - 286 15.P. Goerss, M. Hopkins, Moduli spaces for structured ring spectra, * *available at hopf.math.purdue.edu 16.A. Grothendieck, Rev`etements 'etales et groupe fondamental (SGA 1), Lecture* * Notes in Math., 224, Springer 17.__-, Groupes de monodromie en g'eom'etrie alg'ebrique (SGA 7), Lecture Notes* * in Mathematics: I, 288; II, 340 18.R. Hain, Lectures on the Hodge-de Rham theory of the fundamental group of P1* *- {0, 1, 1}, available at www.math.duke.edu/~hain/aws/ 19.M. Hazewinkel, The algebra of quasi-symmetric functions is free over the int* *egers, Adv. Math. 164 (2001) 283 - 300 20.__-, Symmetric functions, noncommutative symmetric functions and quasisymmet* *ric func- tions, Acta Appl. Math. 85 (2005) 319-340 21.M. Hopkins, Global methods in homotopy theory, in Homotopy theory (Durham, 1* *985), 73-96, London Math. Soc. Lecture Note Ser., 117, Cambridge Univ. Press (1987) 22.__-, B.H. Gross, The rigid analytic period mapping, Lubin-Tate space, and st* *able homotopy theory, available at math.AT/9401220 23.M. Hoffman, The algebra of multiple harmonic series, J. Alg. 194 (1997) 477 * *- 495 24.M. Hovey, J. Palmieri, The structure of the Bousfield lattice, in Homotopy i* *nvariant al- gebraic structures, 175-196, Contemp. Math., 239, AMS (1999) 25.__-, __-, N. Strickland, Axiomatic stable homotopy theory Mem. Amer. Math. S* *oc. 128 (1997) 26.__-, N. Strickland, Morava K-theories and localisation Mem. Amer. Math. Soc.* * 139, no. 666 (1999) 27.M. Kapranov, Analogies between the Langlands correspondence and topological * *quantum field theory, in Functional analysis on the eve of the 21st century Vol. 1 119-151* *, Progr. Math., 131, Birkh"auser (1995) 18 JACK MORAVA 28.M. Kontsevich, Operads and motives in deformation quantization, avai* *lable at math.QA/990405 29.H. Krause, Cohomological quotients and smashing localizations, available at * *math.RA/0308291 30.S. Lichtenbaum, The Weil-etale topology on schemes over finite fields, Compo* *s. Math. 141 (2005) 689-702 32.J. Morava, The Weil group as automorphisms of the Lubin-Tate group, in Journ* *'ees de g'eom'etrie alg'ebrique de Rennes Vol. I, pp. 169-177, Ast'erisque 63, Soc. * *Math. France (1979) 32.__-, Some Weil group representations motivated by algebraic topology, in Ell* *iptic curves and modular forms in algebraic topology 94-106, Lecture Notes in Math., 1326* *, Springer (1988) 33.__-, The motivic Thom isomorphism, available at math.AT/0306151 34.A. Neeman, Stable homotopy as a triangulated functor, Invent. Math. 109 (199* *2) 17-40. 35.__-, The chromatic tower for D(R). With an appendix by Marcel B"okstedt. Top* *ology 31 (1992) 519-532. 36.J. Rognes, Galois extensions of structured ring spectra, available at math.A* *T/0502183 37.J.P. Serre, Cohomologie galoisienne, Lecture Notes in Mathematics, 5. Spring* *er-Verlag, Berlin, 1994. 38.__-, Local class field theory, in Algebraic number theory (Brighton) 128-161* * Thompson (1967) 39.D. Sullivan, Geometric Topology, Part I. Localization, periodicity and Galoi* *s sym- metry, available at www.maths.ed.ac.uk/~aar/surgery 40.J. Tate, Number theoretic background, in Automorphic forms, representations * *and L-functions 3-26, Proc. Sympos. Pure Math., XXXIII, AMS (1979) 41.B. Toen, Vers une interpr'etation galoisienne de la th'eorie de l'homotopie,* * Cah. Topol. G'eom. Diff'er. Cat'eg. 43 (2002) 257-312 42.T. Torii, On degeneration of one-dimensional formal group laws and applicati* *ons to stable homotopy theory, Amer. J. Math. 125 (2003) 1037-1077. 43.V. Voevodsky, Triangulated categories of motives over a field, in E. Friedla* *nder et al, Cycles, transfers, and motivic homology theories, Annals of Maths. Studies 143, Prin* *ceton (2000) 44.A. Weil, Basic number theory Third edition. Die Grundlehren der Mathematisch* *en Wis- senschaften, Band 144. Springer-Verlag, New York-Berlin, 1974. Department of Mathematics, Johns Hopkins University, Baltimore, Maryland 21218 E-mail address: jack@math.jhu.edu