CONNECTING HOMOMORPHISMS ASSOCIATED TO TATE SEQUENCES PAUL RICHARD BUCKINGHAM 1 Abstract. Tatesequencesareanimportanttoolfortacklingproblems 1 related to the (ill-understood) Galois structure of groups of S-units. 0 The relatively recent Tate sequence “for small S” of Ritter and Weiss 2 allowsonetousethe sequencewithoutassumingthevanishingofthe S- n class-group,asignificantadvanceinthetheory.AssociatedtoRitterand a Weiss’s version of the sequence are connecting homomorphisms in Tate J cohomology,involvingthe S-class-group,that do not exist inthe earlier 0 theory. In the present article, we give explicit descriptions of certain of 1 theseconnectinghomomorphismsundersomeassumptionsonthesetS. ] T N . h 1. Introduction t a m In [19, 21], Tate constructs, for a Galois extension L/K of number fields [ with Galois group G, a Tate sequence 0 → O× → A → B → X → 0 L,S 1 v of Z[G]-modules whose extension class in Ext2Z[G](X,OL×,S) is the so-called 0 Tate canonical class, where O× is the group of S-units in L and X is a 5 L,S 8 finitely generated, Z-torsion-free module which will be defined in Section 2. 1 . This is done under the assumptions that S contains the ramified places and 1 0 that the S-class-group of L is trivial. 1 1 Tate’s construction shows that A and B may in fact be chosen finitely : v generated and cohomologically trivial. This has the consequence that the i X Galois cohomology of O× can be identified with that of X, after a dimen- L,S r sion shift of 2. This is reminiscent of the Artin–Tate formulation of class a field theory [1] in which the cohomology of the idele class-group is identi- fied with the cohomology of Z after a dimension shift of 2, with a similar statement in the local case. In fact, this is no coincidence, since it is via this interpretation of class field theory that the Tate sequence is constructed. Thesequenceisaprimaryfeatureinnumerousapplicationstomultiplica- tive Galois module structure. Tate uses it in [21, Ch. II] to prove Stark’s Conjecture for rationalcharacters, andChinburg employs it in theconstruc- tion of his third Ω-invariant, which is central to his root number conjecture – see [4, 5]. In fact, the Tate sequence also appears in the definition of 2010 Mathematics Subject Classification. Primary 11R29; Secondary 11R34. Key words and phrases. Tate sequence, Class-group,Galois cohomology. 1 2 P. R.BUCKINGHAM the lifted Ω-invariant in the Lifted Root Number Conjecture of Gruenberg– Ritter–Weiss [9]. Further, a variant of the sequence is used to construct the equivariant Tamagawa number in the Equivariant Tamagawa Number Conjecture (ETNC) for the motive h0(Spec(L))(0), as in Burns–Flach [2] for example. A significant step forward in the theory of Tate sequences was Ritter and Weiss’s Tate sequence “for small S”, which allows the sequence to be constructed for an arbitrary set of places S. More precisely, they found in [15] a sequence (1.1) 0 → O× → A → B → ∇ → 0 L,S for a Galois extension L/K of number fields with Galois group G, with no restriction on the set S other than that it contains the infinite places. Whereasinthepreceding theoryofTatesequences, themoduleinplaceof∇ was the Z-torsion-free Z[G]-module X whose Galois structure is described in terms of very basic arithmetic information, ∇ itself fits into an exact sequence (1.2) 0 → Cl (L) → ∇ → X → 0. S An early use of this refined Tate sequence “for small S” was in the proof by Ritter and Weiss [16] of the Strong Stark Conjecture for abelian extensions of Q. More recent applications have been the consideration of the minus part of the ETNC for tame extensions, as in Nickel [13], and the study of Fitting ideals of (duals of) minus parts of class-groups, as in Greither [8]. Despitethesequence’s growinguseinmultiplicative Galoismodulestruc- ture, one aspect of it that has not yet been determined is the precise ef- fect of the connecting homomorphisms that are naturally associated to the sequence. Our aim is to make certain such maps explicit, under some as- sumptions on the set S, namely that S will contain the infinite and ramified places and at least one place with full decomposition group. We expect to be able to remove this last assumption in the future. However, we empha- size that the S-class-group will not be assumed trivial. The maps we make explicit are the connecting homomorphism H−2(G,X) → H−1(G,Cl (L)) S associated to (1.2), and the map H−1(G,Cl (L)) → H1(G,O× ) that re- S L,S sults from H−1(G,Cl (L)) → H−1(G,∇) together with the appropriate S connecting homomorphisms (i.e. starting in dimension −1 then 0) obtained by splitting (1.1) into two short exact sequences. After building up preliminaries in Sections 2, 3, 4 and 5, we compute the maps in Sections 6 and 7, and give some corollaries in Section 8. CONNECTING HOMOMORPHISMS ASSOCIATED TO TATE SEQUENCES 3 2. Notation, assumptions and conventions We will assume throughout that L/K is Galois with Galois group G. S will denote a finite set of places of K containing the infinite and ramified places. By a minor abuse of notation, O will denote the S -integers in L,S L L, where S consists of the places of L above those in S. Thus O× is the L L,S group of S -units in L. The S-class-group of K will be written Cl (K), and L S the S -class-group of L will be just Cl (L), the redundant subscript L again L S being dropped. By L , we mean the Hilbert S -class field of L, that is, the maximal S L unramified abelian extension of L in which all places in S split completely. L Gal(L /L) is isomorphic, via the Artin map, to Cl (L). S S We will denote the element σ ∈ Z[G] by N. Multiplication by σ∈G N induces an endomorphism of Cl (L), but we stress that this is different PS from the map Nm : Cl (L) → Cl (K), induced by the norm of ideals, that S S is featured in Section 8. All number fields will lie in a fixed algebraic closure Q¯ of Q. If p is a place of K, we will fix once and for all a place of Q¯ above p. Then given any number field F containing K, p(F) will denote the place of F below the chosen place of Q¯ above p. For shorthand, we will denote the completion F by F , and if F/K is Galois, the decomposition group Gal(F/K) p(F) p p(F) will bedenoted simply Gal(F/K) . Notethat if p ∈ S, then since p(L) splits p completely in L , restriction Gal(L /K) → G defines an isomorphism, S S p p and we denote the composition ≃ G → Gal(L /K) → Gal(L /K) p S p S by ι . p For a finite place p of K, let ϕ˜ denote a lift in Gal(Lur/K ) of the Frobe- p p p nius of Kur/K , and ϕ¯ its image in Gal(L /K). Further, if p is unramified p p p S in L/K, ϕ will be the associated Frobenius element in G . p p Animportant object appearing throughout the article is theGaloismod- ule X defined in Definition 2.1: Definition 2.1. Let Y be the free abelian group on S , and X the kernel L of the augmentation map Y → Z sending every place P ∈ S to 1. L 4 P. R.BUCKINGHAM To give some context, we remark that X appears in the Dirichlet regu- lator map, i.e. the isomorphism R⊗ O× → R⊗ X Z L,S Z 1⊗u 7→ logkuk w, w wX∈SL where the absolute values k·k are normalized in the particular canonical w way which makes the product formula hold, as in [12, Ch. III, Section 1]. For any group G, ∆G will denote its augmentation ideal, that is, the kernel of the augmentation map Z[G] → Z which sends each group element to 1. 2.1. Key assumptions. InSection 4andSection 5.1, weassume that L/K is Galois and S contains the ramified places. In Section 5.2 and Sections 6, 7 and 8, we further assume that there is a place p ∈ S such that G = G. 0 p0 We observe that this last assumption forces G to be solvable, since then G is the Galois group of a Galois extension of local fields. 3. Group cohomology CohomologywillbeTatecohomologythroughout.Wewillusethefollow- ing model for Tate cohomology groups in negative degrees, which we shall think of as homology groups, that is Hi(G,A) = H (G,A) for i < −1: −i−1 For n ≥ 0, let P be the free right Z[G]-module on Gn, and for n ≥ 1 define n a boundary map d : P → P by sending [σ ,...,σ ] to n n n−1 1 n n−1 [σ ,...,σ ]·σ + (−1)n−i[σ ,...,σ σ ,...,σ ]+(−1)n[σ ,...,σ ]. 1 n−1 n 1 i i+1 n 2 n i=1 X aug Then P → Z → 0 is a free resolution of Z as a right Z[G]-module, and • for a left Z[G]-module A, we view H (G,A) as the ith homology group of i the chain complex P ⊗ A when i > 0. The Tate cohomology group • Z[G] H−1(G,A) will be taken to mean {a ∈ A | σa = 0} (3.1) σ∈G . Z{(σ −1)a | σ ∈ G,a ∈ A} P Notethat the denominator in (3.1) is indeed aZ[G]-submodule, i.e. is closed under the action of G. Lemma 3.1. Let G be a finite group and H a subgroup. There is a well- defined group isomorphism Hab → H−2(G,Z[G]⊗ Z) Z[H] sending σ[H,H] to [σ]⊗1⊗1. CONNECTING HOMOMORPHISMS ASSOCIATED TO TATE SEQUENCES 5 Proof. That Hab is isomorphic to H−2(G,Z[G]⊗ Z) is simply Shapiro’s Z[H] Lemma – as in [22, Lemma 6.3.2] for example – together with the fact that H−2(H,Z) ≃ Hab. The explicit description of the isomorphism is left to the (cid:3) reader. 3.1. Extension classes. Supposethat AandC areZ[G]-modulesandthat C is Z-free. Lemma 3.2. There is a canonical isomorphism Ext1 (C,A) ≃ H1(G,Hom (C,A)). Z[G] Z Proof. Since thisiswell known, we onlysketch theproof.Supposing wehave an exact sequence 0 → A → B → C → 0 of Z[G]-modules, choose a section s : C → B oftheZ-modulehomomorphismB → C,sothats ∈ Hom (C,B) Z maps to 1 in Hom (C,C). This is possible by the assumption on C. Then C Z for each σ ∈ G, σs−s is the image of a unique f ∈ Hom (C,A). The map σ Z σ 7→ f is a 1-cocycle G → Hom (C,A). σ Z Conversely, given a 1-cocycle f : G → Hom (C,A), we can endow the Z direct sum B = A⊕C of Z-modules with a G-action by (3.2) σ(a,c) = (σa+f(σ)(σc),σc). One checks that these constructions pass to mutually inverse maps between Ext1 (C,A) and H1(G,Hom (C,A)). (cid:3) Z[G] Z Lemma 3.3. Suppose 0 → A → B → C → 0 is an exact sequence of Z[G]-modules such that C is Z-free. Viewing its extension class ξ as an element of H1(G,Hom (C,A)), the connecting homomorphism Hi(G,C) → Z Hi+1(G,A) is given by following cup-product with ξ by the evaluation map Hi+1(G,C ⊗ Hom (C,A)) → Hi+1(G,A). Z Z (cid:3) Proof. See [3, Ch. XII, Prop. 6.1]. By Lemma 3.2, Ext1 (∆G,A) ≃ H1(G,Hom (∆G,A)) for each Z[G]- Z[G] Z module A. On the other hand, the exact sequence 0 → A → Hom (Z[G],A) → Hom (∆G,A) → 0 Z Z together with the cohomological triviality of Hom (Z[G],A) gives an iso- Z morphism H1(G,Hom (∆G,A)) → H2(G,A). Thus we have: Z Lemma 3.4. For any Z[G]-module A, Ext1 (∆G,A) ≃ H2(G,A). Z[G] 6 P. R.BUCKINGHAM 4. The modules WS′, RS′ and BS′ 4.1. The modules WS′ and RS′. We let S′ denote a finite set of places of K containing S, and further satisfying: (i) G = G, p∈S′ p and (ii) ClS′(L) = 0. S Such an S′ can always be chosen, by the Chebotarev Density Theorem. Following [15, Section 1], we define WS′ = ∆pG⊕ Z[G], p∈S q∈S′rS M M where ∆ G is the left ideal in Z[G] generated by ∆G . We observe that p p this description of WS′ relies on S containing the ramified primes. For a treatment of the general case, see [15] itself. As in [15, Section 4], we now define RS′ to be the kernel of the map WS′ → ∆G that is inclusion on ∆pG and left multiplication by ϕq −1 on the copy of Z[G] corresponding to q ∈ S′rS. 4.2. The module BS′. After [15, Section 4], we let BS′ be the kernel of the map Z[G] → Z[G] p∈S′ M that is the identity on the copy of Z[G] corresponding to p ∈ S, and mul- tiplication by ϕ −1 on the copy of Z[G] corresponding to q ∈ S′rS. We q notethatBS′ is projective, andthereforecohomologicallytrivial. To provide more context, we note that BS′ is the module B appearing in (1.1). The map RS′ → BS′ in (4.5) is induced by the inclusion map WS′ → Z[G]. p∈S′ M The map BS′ → X is induced by the map Z[G] → Y p∈S′ M that is zero on Z[G] corresponding to q ∈ S′rS, and sends α ∈ Z[G] corresponding to p ∈ S to αp(L). 4.3. Class field theory and diagrams. Denote the idele class-group of L by C . We let L 0 → C → V → ∆G → 0 L be the extension corresponding under the isomorphism of Lemma 3.4 to the globalfundamental classinH2(G,C ).Infact,forconcreteness, wewill take L thefollowingdescriptionofV:SupposetheelementofH1(G,Hom (∆G,C )) Z L CONNECTING HOMOMORPHISMS ASSOCIATED TO TATE SEQUENCES 7 corresponding to the global fundamental class is represented by the 1- cocycle f. Then we view V as C ⊕ ∆G (direct sum as Z-modules) with L G-action given as in (3.2). Ritter and Weiss also define local versions V of V for each p ∈ S′. For p p ∈ S, V is simply the analogous construction to V with C replaced by L× p L p and the global fundamental class replaced by the local one. For p ∈ S′rS, the definition is more subtle, but still uses local class field-theoretic data. We will only need the definitions in a very simple situation, which we will turn to in Section 7.2. The reader wishing to see the complete definition may refer to [15, Sections 1,3]. Following Ritter and Weiss, we set (4.1) VS′ = Z[G]⊗Z[Gp] Vp ⊕ UP, Mp∈S′ ! PM6∈SL′   where U is the group of units in L . Note that the first direct sum runs P P through primes p of K (in S′), whereas the second direct sum runs through primes P of L (not above S′). The important diagrams involving the modules WS′, RS′ and BS′ are (4.2) 0 0 0 (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) 0 // OL×,S // A // RS′ s ED (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) 0 // JS // VS′ // WS′ // 0 BC (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) GF 0 // C // V // ∆G // 0 L (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) @A // Cl (L) // 0 0 S (cid:15)(cid:15) 0 and (4.3) 0 // RS′ // BS′ // X // 0 s t 1 (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) 0 // Cl (L) // ∇ // X // 0. S 8 P. R.BUCKINGHAM In (4.2), the module A is defined to be the kernel of VS′ → V, and (4.2) is simply the snake diagramarising fromthemiddle two rows. Thesurjectivity of VS′ → V in (4.2) and the origin of (4.3) are treated in [15, Section 4]. RS′ fits into an exact sequence (4.4) 0 → OL×,S → A → RS′ →s ClS(L) → 0, where A is the (cohomologically trivial) Z[G]-module appearing in the Tate sequence (1.1). Ritter and Weiss term the map s the ‘snake map’, be- cause it is the snake map of diagram (4.2) – see the discussion following [15, Theorem 1]. The significance of the snake map is as follows: Applying Hom (−,Cl (L)) to the short exact sequence Z S (4.5) 0 → RS′ → BS′ → X → 0 occurring in (4.3), and remembering that X is Z-free, we obtain the exact sequence 0 → HomZ(X,ClS(L)) → HomZ(BS′,ClS(L)) → HomZ(RS′,ClS(L)) → 0. Identifying Ext1 (X,Cl (L))withH1(G,Hom (X,Cl (L))),theextension Z[G] S Z S class of (1.2) is the image of the class of −s under the connecting homo- morphism H0(G,HomZ(RS′,ClS(L))) → H1(G,HomZ(X,ClS(L))). Since (1.2) is central to our purpose, getting a strong hold on s will be one of our main goals, and will be carried out in Section 5. 5. The snake map 5.1. Explicit description of s. In this section, L/K is a Galois extension of number fields and S a finite set of places of K containing the infinite and ramified ones. We let GS = Gal(L /K). S We now come to the description of the snake map s : RS′ → ClS(L) given in [15, Section 5]. It is defined first of all as a prime-by-prime map WS′ → H where ∆GS H = . Ker(∆GS → ∆G)∆GS (We note that an element σ of G acts on H by left multiplication by any preimage of σ in GS.) For p ∈ S, the map ∆ G → H sends βα to β(ι (α)) for α ∈ ∆G and p p p β ∈ Z[G], where the bar denotes class in H. CONNECTING HOMOMORPHISMS ASSOCIATED TO TATE SEQUENCES 9 Recall the element ϕ¯ ∈ GS = Gal(L /K) defined in Section 2. For q S q ∈ S′rS, the image of an element β inside the copy of Z[G] in WS′ corresponding to q is mapped to β(ϕ¯ ) in H. q There is an embedding Gal(L /L) → H which sends σ to the class of S σ − 1, and the restriction of the map WS′ → H to RS′ has its image in Im(Gal(L /L) → H). Identifying Gal(L /L) with Cl (L), we have thus S S S described the snake map s : RS′ → ClS(L). 5.2. The extension class of ∇. We now assume that there exists a finite place p ∈ S such that G = G. For each p ∈ S, fix a set D of represen- 0 p0 p tatives for (G/G ) containing 1, and given σ ∈ G let ρ (σ) be the chosen p left p representative of the coset σG . It is possible to choose S′ as in Section 4 p with the further condition that every place q ∈ S′rS splits completely in L/K. For this, we are already using that G = G. We are assuming p∈S p more, of course: G = G. With S′ chosen in this way, we have p0 S (5.1) RS′ = Ker ∆pG → ∆G ⊕ Z[G]. p∈S ! q∈S′rS M M Definition 5.1. Given p ∈ Sr{p }, σ ∈ G and τ ∈ D , let r(p) be 0 p σ,τ the element of RS′ = Ker( p∈S ∆pG → ∆G) ⊕ q∈S′rSZ[G] which has σρ (σ−1τ)−τ in the p-component, τ −σρ (σ−1τ) in the p -component, and p L p L 0 zero elsewhere. Lemma 5.2. Define g : G → Hom (X,Cl (L)) by sending σ ∈ G to the Z S map τp(L)−p (L) 7→ s(r(p)) 0 σ,τ for τ ∈ Dp, p ∈ Sr{p0}. Then the image of −s ∈ H0(G,HomZ(RS′,ClS(L))) in H1(G,Hom (X,Cl (L))) is g. Z S Proof. We follow the diagram (5.2) Map(G,Hom (X,Cl (L))) Z S (cid:15)(cid:15) HomZ(BS′,ClS(L)) // Map(G,HomZ(BS′,ClS(L))) (cid:15)(cid:15) HomZ(RS′,ClS(L)) from bottom-left to top-right. Note that BS′ = Ker Z[G] → Z[G] ⊕ Z[G]. p∈S ! q∈S′rS M M 10 P. R.BUCKINGHAM We first look for a Z-splitting of 0 → RS′ → BS′ → X → 0. We use the Z-basis {τp(L)−p (L) | τ ∈ D } 0 p p∈S[r{p0} of X. For p ∈ S, let e be the element of Z[G] which has 1 in the p p∈S p-component and zero everywhere else. ThenLa lift of τp(L)−p0(L) to BS′ is τe −τe . Therefore if Z is the Z-span of these lifts, p p0 BS′ = RS′ ⊕Z. Define µ : BS′ → ClS(L) by µ|RS′ = −s and µ|Z = 0. Let f be the image of µ under the horizontal map in (5.2), i.e. if σ ∈ G, f(σ) = σµ−µ. Now take τ ∈ D for some p ∈ S. p (σµ)(τe −τe ) = σµ(σ−1τe −σ−1τe ), p p0 p p0 and one sees that σ−1τe −σ−1τe decomposes as p p0 [(σ−1τ −ρ (σ−1τ))e +(ρ (σ−1τ)−σ−1τ)e ] p p p p0 (5.3) + [ρ (σ−1τ)e −ρ (σ−1τ)e ] p p p p0 in RS′ ⊕ Z. Also, we recognize the first bracketed element in (5.3) as −σ−1r(p). Thus σ,τ (σµ)(τe −τe ) = −σµ(σ−1r(p)) p p0 σ,τ = σs(σ−1r(p)) σ,τ = s(r(p)). σ,τ Since µ(τe −τe ) = 0, we therefore obtain that p p0 f(σ)(τe −τe ) = s(r(p)). p p0 σ,τ Further, the map g : G → Hom (X,Cl (L)) appearing in the statement Z S of the lemma has image f in Map(G,HomZ(BS′,ClS(L))). The lemma is (cid:3) proven. Definition 5.3. If p ∈ Sr{p } and τ ∈ G , then ι (τ)−1ι (τ) ∈ Gal(L /L), 0 p p0 p S and we let its image in Cl (L) under the Artin map be c (τ). S p Lemma 5.4. Ifp ∈ Sr{p }, τ ∈ D and σ ∈ G,then s(r(p)) = τc (τ−1σρ (σ−1τ)). 0 p σ,τ p p Proof. SincesisaZ[G]-modulehomomorphism,thestatement ofthelemma is equivalent to s(τ−1r(p)) = c (τ−1σρ (σ−1τ)). Now, recalling the notation σ,τ p p e introduced in the proof of Lemma 5.2, p τ−1r(p) = (τ−1σρ (σ−1τ)−1)e −(τ−1σρ (σ−1τ)−1)e , σ,τ p p p p0