Congruences between derivatives of abelian \(L\)-functions at \(s =0\) (Q2385046)

This paper proves a beautiful and very strong result on Rubin-Stark units attached to abelian extensions \(K/k\) of global fields and the relations that hold between these units when the top field varies. The result is conditional on the validity of the Equivariant Tamagawa Number Conjecture for a certain standard motive; we will comment on this below. Let us first try to describe the main result, and then its consequences. Let \(G\) be the Galois group of \(K/k\). One always needs a finite set \(S\) of \(k\)-places containing at least the infinite places and those that ramify in \(K\). To make the unit groups torsion-free, one is also obliged to work with another finite set \(T\) of places, disjoint with \(S\); the subscript \(T\) always means that one takes only \(S\)-units congruent to 1 at all places in \(T\). The equivariant \(L\)-function attached to \(K/k\) has an \(S\)-truncated, \(T\)-modified variant \(\theta(K/k,S,T; s)\). Let \(r=r_K\) be the number of places in \(S\) that split completely in \(K/k\), if this is less than \(|S|\), and \(r=|S|-1\) otherwise. Rubin's conjecture predicts the existence of an element \(\eta_{K,S,T}\) in the \(r\)-th exterior power of \(O_{K,S,T}^\times\) tensored with \(\mathbb{Q}\), but one does expect certain integrality properties. We will call this element the Rubin-Stark ``unit'', by an abuse of language. It is linked to the \(r\)-fold derivative of \(\theta(K/k,S,T; s)\) at \(s=0\) by the regulator map. The problem is to make precise the integrality condition on \( \eta_{K,S,T}\), and still more importantly, the relations that hold between \(\eta_{K,S,T}\) and \(\eta_{L,S,T}\) for intermediate fields \(k\subset L \subset K\). Since the number \(r_L\) is usually bigger than \(r_K\), this cannot just be a compatibility under a canonical map, since the involved exterior powers do not match. What makes matters a little tough at first sight is that one always needs ``testing homomorphisms'' \(\Phi \in \wedge^r \text{Hom}_G(O_{K,S,T}^\times,\mathbb{Z}[G])\). Let \(r'=r_L\) (so, let us repeat this, we often have \(r'>r\)). Then applying \(\Phi\) to \(\eta_{K,S,T}\) gives an element of \(\mathbb{Z}[G]\), assuming Rubin's conjecture. Applying the same \(\Phi\) to \(\eta_{L,S,T}\) however still leaves us with a wedge product of units (exponent \(r'-r\)). Now one can construct a homomorphism \(f_{L,S}\) from \(\wedge^{r'-r} O_{L,S,T}^\times\) to a certain factor ring of \(\mathbb{Z}[G]\), using local Artin symbols. This is a kind of algebraic counterpart to the Dirichlet regulator. The main result (Theorem 3.1) then says (we are neglecting quite a few technical points here, so the reader should beware!) that ETNC for the pair \((h^0(\text{Spec}(K)), \mathbb{Z}[G])\) implies: \(\Phi(\eta_{K,S,T})\) is equal, in this factor ring, to \(f_{L,S}(\Phi^H(\eta_{L,S,T}))\). Here \(H\) is the group fixing \(L\), and \(\Phi^H\) is a straightforward modification of \(\Phi\), which it is appropriate to apply to elements known to be fixed by \(H\). Since the right hand side can be shown to lie in a certain power of the image of the augmentation ideal of \(\mathbb{Z}[G]\), this gives ``vanishing results'', in other words: a statement concerning the position of the Rubin-Stark unit in the wedge product of the unit group. It seems worthwile to discuss an example: we essentially take this from \textit{A. Hayward}\,'s paper [Compos. Math. 140, No. 1, 99--129 (2004; Zbl 1060.11075)]. Let \(k\) be the rationals and \(K=k(\zeta_m)^+\) where \(m\) is the product of the distinct odd prime numbers \(p_1,\ldots, p_s\). We also impose the very mild restriction that \(m\) is the exact conductor of \(K\). Take \(S\) to consist of the \(p_i\) plus the place at infinity. Since that place splits completely and no other place does, we have \(r=1\). Now we take \(L=k\) (smallest possible choice); this gives \(r'=s\). The Rubin-Stark unit of \(K\) is an actual unit (\(r=1\)) and is given essentially by a cyclotomic unit. The Rubin-Stark ``unit'' of \(k\) is essentially the wedge \(p_1\wedge \ldots \wedge p_s\); and the result of \(f_{k,S}\) on a wedge \(p_1\wedge \ldots \wedge p_{s-1}\) can be given as the determinant of a matrix whose off-diagonal entries have the form \(1-f_{p_i}(p_j)\), with \(f_{p_i}(p_j)\) the Frobenius of \(p_j\) in the \(p_i\)-adic completion of \(K/k\). (This instance of the main result of the article under review had already been proved by Hayward [loc. cit.]) Another illuminating example, which shows that ETNC implies the so-called refined \(p\)-adic abelian Stark conjecture, is discussed in the paper, see Corollary 4.3. The proof of this main result combines a great many ideas and techniques. It is not possible to go into detail here. Of course the ETNC comes in crucially, so its statement must be unravelled. This involves a complex with known cohomology and the regulator map. One major advance of the paper is to provide a very explicit variant of this complex, consisting of just two terms: \(F \overset{\phi}{\to}F\) in degree 0 and 1. Here \(F\) is a free \(\mathbb{Z}[G]\)-module of sufficiently high rank. The kernel of \(\phi\) is closely related to \(O_{K,S,T}^\times\), and the cokernel is closely related to the standard lattice \(X_{K,S}\). This leads to a description of the Rubin-Stark units in terms of the Laplace expansion of a certain matrix. It should be mentioned that it is only possible to construct this simple ``approximation'' \(F \overset{\phi}{\to} F\) after enlarging the set \(S\) suitably so as to make Pic\((O_{K,S})\) trivial and ensure another technical condition involving \(T\); it is shown in \S5 that such a procedure is possible and involves no loss of generality in the proof of the main theorem. It is also well explained in the article under review that the main result has a considerable number of important consequences: If \(K\) is an abelian extension of \(\mathbb{Q}\), or belongs to an explicit list of other cases, then the author obtains the validity of conjectures made by Rubin, Popescu, Gross, and many others. (For full details see the paper, Thm.~A on p. 453f.) This is achieved via the main result: in all of the listed cases the relevant instance of ETNC is proved, so the main theorem 3.1 applies, and it is explained how all the conjectures are derived from it. Remark: The 2-adic case of ETNC was left out in the paper of Burns and the reviewer [Invent. Math. 153, No. 2, 303--359 (2003; Zbl 1142.11076)] and subsequently dealt with in the survey paper of \textit{M. Flach} [Contemp. Math. 358, 79--125 (2004; Zbl 1070.11025)]. The reviewer recently learnt that the argument for the 2-adic case given in that survey contains a subtle flaw, which has been repaired in a forthcoming work of Flach. The article under review is remarkable by its insight and generality. It is, unavoidably, rather complex, but well organised; we refer in particular to the introduction and the outline of proof given in 6.4.