On the conjectures of Birch and Swinnerton-Dyer in characteristic \(p>0\) (Q1411973)

Let \(K\) be a function field over a finite field and \(A/K\) an abelian variety defined over \(K\). In this breakthrough paper, the authors prove that if there exists a prime number \(\ell\) such that the \(\ell\)-primary part of the Tate-Shafarevich group of \(A\) is finite, then the whole Birch and Swinnerton-Dyer conjectures hold for \(A/K\). We will be more specific in the sequel. This theorem was already known in the following particular situations: in the case where \(A\) is the Jacobian variety of a smooth proper curve over \(K\) (under a mild hypothesis), this is due to \textit{J. Tate} [On the conjectures of Birch and Swinnerton-Dyer and a geometric analog, Dix Exposés Cohomologie Schémas, Advanced Studies Pure Math. 3, 189--214 (1968); Sem. Bourbaki 1965/66, No. 306, 415--440 (1966; Zbl 0199.55604)] and \textit{J. S. Milne} [J. Fac. Sci., Univ. Tokyo, Sect. I A 28, 735--743 (1981; Zbl 0503.14010)]; for constant abelian varieties, \textit{J. S. Milne} [Invent. Math. 6, 91--105 (1968; Zbl 0159.22402)]; for some elliptic curves, and more generally for those of rank at most 1 if \(p>3\), \textit{D. Ulmer} [Ann. Math. (2) 155, No. 1, 295--315 (2002; Zbl 1109.11314) and Geometric non-vanishing, http://arxiv.org/math.NT/0305321]; for elliptic curves when \(p=\text{char}(K)>2\), \textit{J. S. Milne} [Ann. Math. (2) 102, 517--533 (1975; Zbl 0343.14005)]; for abelian varieties modulo \(p\)-primary part, \textit{P. Schneider} [Math. Ann. 260, 495--510 (1982; Zbl 0509.14022)]; and in the case where \(A\) has everywhere good reduction, \textit{W. Bauer} [Invent. Math. 108, 263--287 (1992; Zbl 0807.14014)]. The strategy of the proof is to reduce the problem to the comparison between \(\ell\)-adic étale cohomology and the \(\ell\)-primary part of flat cohomology, when \(\ell\neq p\), this was done in [\textit{P. Schneider}, loc.cit.]; and crystalline cohomology and the \(p\)-primary part of flat cohomology. This is the harder part due to the following reasons. The classical crystalline cohomology does not work well for degenerating coefficients nor at the places where \(A/K\) has bad reduction. Replacing it by rigid cohomology, although it works well for degenerating coefficients, it only does so modulo torsion, so it is not easily related to torsion flat cohomology groups. The authors overcome this difficulty by taking a finite extension \(K'/K\) such that \(A':=A\times_KK'\) has everywhere semi-stable reduction. In this situation there is crystalline cohomology with log poles (which respects torsion and works well with semi-stable degenerations) and syntomic cohomology with log poles relates crystalline cohomology with log poles with flat cohomology (syntomic cohomology without log poles was already used by \textit{W. Bauer} [loc.cit.]. The result is then obtained by comparing rigid cohomology groups associated to \(A\), the crystalline cohomology groups with log poles associated to \(A'\), the flat cohomology groups associated to \(A'\) and the flat cohomology groups associated to \(A\). Let us now actually describe the conjectures mentioned above. Let \(K\) be the function field of a smooth proper connected curve \(X\) defined over the finite field \(\mathbb{F}_p\) of \(p\) elements. Let \(U\subset X\) be an open dense subset where \(A\) has good reduction. Let \[ L(U,A,s):=\prod_{v\notin U}P_v(N(v)^{-s})^{-1}, \] where \(P_v(u)=\det(1-u\varphi_v\,| \,T_{\ell}(A)(-1)\otimes\mathbb{Q}_{\ell})\), \(T_{\ell}(A)\) is the \(\ell\)-adic Tate module of \(A\) for a prime number \(\ell\) different from the characteristic of the residue field \(k(v)\) of \(v\), \(\varphi_v\) is the geometric Frobenius of \(v\) and \(N(v)=\#k(v)\). The first statement is that \(\text{ord}_{s=1}(L(U,A,s))=r=\text{rank} (A(K))\). The second statement is the finiteness of the Tate-Shafarevich group \({\text Ш}(A)=\ker(H^1(K,A)\to \bigoplus_vH^1(A,K_v))\). Let \(A^*/K\) be the dual abelian variety and let \(h:A(K)\times A^*(K)\to\mathbb{R}\) be the height pairing. Let \(e_1,\cdots,e_r\) be a basis of \(A(K)/A(K)_{\text{tor}}\) and \(e_1^*,\cdots,e_r^*\) a basis of \(A^*(K)/A^*(K) _{\text{tor}}\), then the discriminant of the height pairing is defined by \(\text{Disc}(h(\,,\,)):= | \det(h(e_i,e_j^*)_{i,j})| \in\mathbb{R}^*\). The last statement is \[ \lim_{s\to 1}(s-1)^rL(U,A,s)=\frac{\#{\text Ш}(A)\#\text{Disc}(h(\,,\,))}{\#A(K)_{\text{tor}} \#A^*(K)_{\text{tor}}} {\text{vol}}\left(\prod_{v\notin U}A(K_v)\right). \] The authors' theorem says that the first statement is equivalent to the finiteness of the \(\ell\)-primary part \({\text Ш}(A)\{\ell\}\) of \({\text Ш}(A)\) for some prime number \(\ell\) and in this case all three statements are true for \(A/K\).