Observations on Fermat motives of \(K3\)-type (Q1273705)

Let \(k={\mathbb{F}}_q\), \(q=p^e\), and write \(G(k)=\text{Gal}(\overline{k}/k)\) with \(\varphi_k\in\text{Gal}(\overline{k}/k)\) for the geometric Frobenius. Then a polynomial \(P(t)\in 1+t{\mathbb{Z}}[1/q][t]\) is said to be \(q\)-admissible if all complex reciprocal roots of \(P(t)\) have absolute value \(1\). Such a polynomial is said to be of \(K3\)-type if there is a non-zero rational number \(c_p\) such that the Newton polygon of \(P(t)\) has (normalised) slopes \(\{-c_p,c_p\}\) or \(\{-c_p,0,c_p\}\) with respect to a chosen \(p\)-adic valuation \(\nu\) of \(\overline{\mathbb{Q}}_p\) such that \(\nu(q)=1\), and in both cases one should have \(\text{length}_p(-c_p)=\text{length}_p(c_p)=1\). For a smooth projective variety \(Y\), defined over \(k\), of even dimension \(2d\), and for a prime \(\ell\neq p\), one has the (twisted) \(\ell\)-adic representation \(\rho_{Y,2d}[d]:G(k)\rightarrow\Aut (H^{2d}(Y\times\overline{k},{\mathbb{Q}}_l)(d))\) with characteristic polynomial \(P_{Y,[d]}(t)=\det(1-t\rho_{Y,2d}[d](\varphi_k);H^{2d}(Y\times\overline{k},{\mathbb{Q}}_l)(d))\). This polynomial is \(q\)-admissible. If it is of \(K3\)-type, then \(Y\) is said to be of \(K3\)-type. This notion is due to Yu. Zarkhin. Examples of varieties of \(K3\)-type are provided by ordinary abelian surfaces, ordinary \(K3\)-surfaces, and ordinary cubic 4-folds. The notion of \(K3\)-type can be extended to motives. In particular, consider the Fermat variety \({\mathcal X}:X_0^m+X_1^m+\cdots X_{n+1}^m=0\subset{\mathbb{P}}^n\) of degree \(m\) and dimension \(n=2d\). Then \(({\mathcal X},\Delta_{\mathcal X})\) admits a ``motivic decomposition'' \(({\mathcal X},\Delta_{\mathcal X})=\bigoplus_A{\mathcal M}_A\). Here the motives \(\mathcal M_A\) are determined by the choice of a multiplicative character \(\chi:k^{\times}\rightarrow\mu_m\). Via the diagonal embedding \(\mu_m\hookrightarrow\mu_m^{n+2}\) with image \(\Delta\) one gets a subgroup \(\mathfrak G=\mu^{n+2}_m/\Delta\) of \(\Aut (\mathcal X)\), with character group \(\widehat{\mathfrak G}\) which may be identified with the set \(\{(a_0,a_1,\ldots,a_{n+1})\}\), \(a_i\in{\mathbb{Z}}/m{\mathbb{Z}}\) such that \(\sum a_i\equiv 0\pmod m\). An element \(\mathbf{a}=(a_0,\ldots,a_{n+1})\in\widehat{\mathfrak G}\) gives rise to a Jacobi sum \(j(\mathbf{a})=(-1)^n\sum\chi(v_1)^{a_1}\chi(v_2)^{a_2}\cdots\chi(v_{n+1})^{a_{n+1}}\), with \((v_1,\ldots,v_{n+1})\in(k^{\times})^{n+1}\) such that \(1+v_1+\cdots v_{n+1}=0\). One has a Galois action on the Jacobi sums, and this action is related to the action of \({\mathbb{Z}}/m{\mathbb{Z}}\) on \(\widehat{\mathfrak G}\) by multiplication with \(t\) on each component of an \(\mathbf{a}\in\widehat{\mathfrak G}\) modulo \(m\). One is led to consider the \(({\mathbb{Z}}/m{\mathbb{Z}})^{\times}\)-orbits \(A\) of \(\mathbf{a}\). These define algebraic cycles on \((\mathcal X\times\mathcal X)_k\) with coefficients in \({\mathbb{Z}}[1/m]\) which can be shown to be projectors \(p_A\), thus one has motives \((\mathcal X,p_A)=\mathcal M_A\). The Betti numbers, Hodge numbers and slopes of the \(\mathcal M_A\) can be defined in the usual way. Also, one may define the Hodge and Newton polygons of the motive \(\mathcal M_A\). In particular, one has Mazur's result that the Newton polygon of \(\mathcal M_A\) lies above or on its Hodge polygon. Assume that \(\mathbf{a}\in\widehat{\mathfrak G}\) such that \(a_i\not\equiv 0\pmod m\) for all \(i=0,1,\ldots,n+1\). These form a subset \(\mathfrak U^m_n\) of \(\widehat{\mathfrak G}\). Such \(\mathbf{a}\) will play a major role in the sequel. They determine the zeta-function of the Fermat variety: \[ Z({\mathcal X},t)={\mathcal Q}({\mathcal X},t)^{-1}\prod_{i=0}^{2d}(1-q^it)^{-1}, \] where \(\mathcal Q(\mathcal X,t)\in 1+t{\mathbb{Z}}[t]\), and over \({\mathbb{Q}}(\zeta_m)\) it factors as \(\mathcal Q(\mathcal X,t)=\prod_{\mathbf{a}\in\mathfrak U^m_n}(1-j(\mathbf{a})t)\). One has coincidence between \((1-t)\mathcal Q(\mathcal X,t/q^d)\) and the characteristic polynomial of \(\rho_{\mathcal X,2d}[d]\), thus \(\mathcal Q(\mathcal X,t/q^d)\) is a \(q\)-admissible polynomial. Back to the motives \(\mathcal M_A\). \(\mathcal M_A\) is said to be of \textit{\(K3\)-type} if the polynomial \(\mathcal Q(\mathcal M_A,t/q^d)\) is of \(K3\)-type. Various examples of Fermat motives of \(K3\)-type are given. In their book: ``Arithmetic of diagonal hypersurfaces over finite fields'', Lond. Math. Soc. Lect. Note Ser. 209 (1995; Zbl 0833.14015), \textit{F. Q. Gouvêa} and \textit{N. Yui} defined type I and type II inductive structures in Fermat motives. One has the result: Let \(\mathcal M_A\) be a Fermat motive of \(K3\)-type corresponding to a character \(\mathbf{a}\in\mathfrak U^m_{2d}\). Then any Fermat motive \(\mathcal M_{\widetilde{A}}\) corresponding to a character \(\widetilde{\mathbf{a}}\in\mathfrak U^m_{2d}\) obtained from \(\mathbf{a}\) by type I induction is again of \(K3\)-type. This is not true for type II induction. The theory of \(K3\)-type Fermat motives sheds some light on the Tate conjecture on algebraic cycles. Fermat motives of \(K3\)-type give rise to transcendental cycles. However, powers of \(K3\)-type motives are no longer of \(K3\)-type, so one may ask about Tate's conjecture for powers of Fermat motives of \(K3\)-type. It is shown that Tate's conjecture holds true for \(\mathcal M_A^r\), \(r\geq 2\). For \(\mathcal M_A\) as above, for any natural number \(r\), \(0\leq r\leq d\), let \(w(r)=\sum_{i=0}^{r-1}(r-i)h^{i,n-i}\), where the \(h^{i,j}\) are the Hodge numbers of \(\mathcal M_A\). Then, if \(q=p\equiv 1\pmod m\), there is a positive integer \(\widetilde{B}(r)\), called the Brauer number of \(\mathcal M_A\), with \(\widetilde{B}(r)=\widetilde{B}(n-r)\), such that \(\text{Norm} (1-{j(\mathbf{a})\over q^r})= {\widetilde{B}(r)\over q^{w(r)}}\). Several questions on divisibility properties of the \(\widetilde{B}(r)\) remain unanswered, though numerical results point to some answers. The paper closes with some interesting remarks on the relation between varieties of \(K3\)-type and the corresponding motives, in particular, it is shown that a smooth projective variety \(Y\) of even dimension \(> 2\), defined over \({\mathbb{F}}_q\), which admits a motivic decomposition \(Y=\bigoplus\mathcal M\), and is of \(K3\)-type, has at most one motive \(\mathcal M\) of \(K3\)-type, all the other motives being supersingular, i.e. having Newton polygon pure of slope \(n/2=d\).