Evaluating Azumaya algebras on cubic surfaces (Q625499)

Let \(K\) be a number field and let \(X\subset\mathbb P^{\,3}_{K}\) be a smooth cubic surface over \(K\). It is known that such \(X\) need not satisfy the Hasse principle for the existence of \(K\)-rational points. Conjecturally, all counterexamples to this principle on geometrically rational varieties such as \(X\) are accounted for by the Brauer-Manin obstruction. Thus it is important to have a good description of the Brauer-Manin pairing which defines this obstruction. Recall that this is the pairing \[ \prod_{\text{all \(v\)}}X(K_{v})\times\text{Br}(X)\to \mathbb Q/\mathbb Z, ((x_{v}),\mathcal A)\mapsto\sum_{v}\text{inv}_{v}\,\mathcal A(x_{v}). \] The object of the paper under review is to give a thorough description of both \(\text{Br}(X)\) and the local evaluation maps \(X(K_{v})\to\mathbb Q/\mathbb Z, x\mapsto \text{inv}_{v}\,\mathcal A(x)\), intervening above for certain types of cubic surfaces \(X\), thereby extending work of Colliot-Thélène, D. Kanevsky and Sansuc who discussed the case of diagonal cubic surfaces in 1985. As in the earlier work just mentioned, it is still straightforward to see that the local evaluation map is constant at places of good reduction for \(X\). At places where \(X\) reduces to a cone, the behaviour described by Colliot-Thélène, Kanevsky and Sansuc still happens as long as the singularity of the model of \(X\) is ``not too severe'', in the following sense: assume that \(\mathcal X\subset\mathbb P^{\,3}_{\mathcal O_{v}}\) is a model of \(X_{v}=X\times_{\text{Spec}\, K}\text{Spec}\, K_{v}\) whose special fiber is a cone over a smooth cubic curve. By choosing coordinates appropriately, the equation of \(\mathcal X\) has the form \[ f(T_0,T_1,T_2)+\pi^{s}g(T_0,T_1,T_2,T_3)=0, \] where \(\pi\) is a uniformizer at \(v\), \(s=1\) or \(2\), \(f\) and \(g\) are cubic homogeneous polynomials with coefficients in \(\mathcal O_{v}\), the reduction of \(f\) defines a smooth curve and \(g(0,0,0,1)\) is a unit in \(\mathcal O_v\), i.e., the singularity of \(\mathcal X\) is ``not too bad'' (the case of residual characteristic 3 must be excluded). For such \(X_v\), the author obtains the following results: 1) The only divisor classes on \(X_v\) defined over \(K_{v}\) are multiples of the class of a plane section; 2) \(\text{Br}(X_v)/\text{Br}(K_v)=(\mathbb Z/3\mathbb Z)^{d}\), where \(d=0,1\) or \(2\); 3) Every element of \(\text{Br}(X_v)\) splits over \(K_{v}^{\text{unr}}(\pi^{1/3})\); 4) if \(\mathcal A\) is any Azumaya algebra on \(X_v\) not equivalent to a constant algebra, then the associated evaluation map \(X(K_v)\to \mathbb Q/\mathbb Z\) takes three distinct values. 5) If \(Y\) is a smooth cubic surface defined over a number field \(K\) such that the base change of \(Y\) to one completion \(K_{v}\) is of the type described above, then there is no Brauer-Manin obstruction to the existence of rational points on \(Y\).