A note on the primitive cohomology lattice of a projective surface
Editorial remark: The following snippets are taken more or less directly from the text. The intersection form of a compact connected orientable \(4n\)-dimensional manifold \(X\) is the bilinear, symmetric form on \(\mathsf H_X= H^{2n}(X,\mathbb{Z})/\text{(torsion)}\) given by cup product. By Poincaré duality this form is \textit{unimodular}, that is, its Gram matrix has determinant \(\pm1\). The pair consisting of \(\mathsf H_X\) and the intersection pairing is called the \textit{intersection lattice} of \(X\). If \(X\subset\mathbb{P}^N \) is a smooth compact complex manifold with hyperplane section \(H\), the orthogonal complement of the class of \(H^n\) in \(H^{2n}(X,\mathbb{Z})\) is called the (middle) \textit{primitive cohomology}, denoted \(\mathsf P_X \). Precise knowledge of this lattice and its group of isometries turns out to be useful, especially for arithmetic questions. This motivates interest in the main result of this note which deals with the case of surfaces: Theorem. Let \(X\) be a complex projective surface with \(p_g(X)\not=0\) and let \(c\in \mathsf H_X\) be a primitive representative of an ample divisor. Then the isometry class of the lattice \(c^\perp\) is uniquely determined by the following data: \begin{itemize} \item[1.] the triple \(( b_1(X), c_1^2(X), c_2(X))\) of topological invariants, \item[2.] whether or not \(c\) is characteristic \item[3.] the self-intersection of \(c\). \end{itemize} The above result implies in particular that for a given surface \(X\) the primitive lattice does not not depend on the particular choice of the projective embedding of \(X\), but only on the degree of \(X\). The proof of the theorem uses firstly Nikulin's reformulation of the classical classification results on integral quadratic forms in terms of the discriminant quadratic form and, secondly, on a fine analysis of the type of intersection lattices occurring for projective surfaces based on the Enriques classification. This result is effective as illustrated for surfaces with small \(c_1^2\), e.g. for some Horikawa surfaces. See Examples 1. Primitive cohomology plays a central role in Hodge theory since the Hodge decomposition together with the intersection pairing gives \(\mathsf P_X\) the structure of a polarized pure Hodge structure of weight \(2n\). To explain why this is the case, consider an embedding \(X\subset\mathbb{P}^N\). The Hodge structure on the middle primitive cohomology in smooth families \({X_s}_{s\in S}\) of smooth varieties embedded in the same \(\mathbb{P}^N\) gives rise to a period map \(S\to \Gamma\backslash D\) where \(D\) is a suitable period domain and where \(\Gamma\), the (maximal) monodromy group, is the isometry group of the primitive lattice of a fibre \(X_s\) (all such groups are isomorphic). More precisely, since monodromy preserves the polarization, \(\Gamma\) is the subgroup of the isometry group of \(P=\mathsf P_{X_s}\) inducing the identity on the discriminant group \(P^*/P\) \begin{itemize} \item[1.] For a complex projective surface \(X\) with \(c_1^2(X)=1\) and \(K_X\) ample and \(X\) embedded by a suitable multiple of \( K_X\), one has \(\mathsf P_X\simeq^s U^t E_8(-1)\) since the index is negative by the index formula. The Noether inequality [\textit{W. P. Barth} et al., Compact complex surfaces. 2nd enlarged ed. Berlin: Springer (2004; Zbl 1036.14016), Theorem VII.3.1] stating that \(p_g\le{\frac 12} c_1^2+2\) implies that \(p_g\le 2\). Furthermore, in case \(q>0\), \textit{O. Debarre} [Bull. Soc. Math. Fr. 110, 319--346 (1982; Zbl 0543.14026); Bull. Soc. Math. Fr. 111, 301--302 (1983; Zbl 0585.14027)] has show that \(2p_g\le c_1^2\) so that \(p_g=0\) in the present situation. From this and the Noether formula, one arrives at the following sets of possible invariants: \begin{center} \begin{tabular}{@{} |c| c|c| c | @{}} \hline \(\chi\) & \(1\) & \(2\) & \(3\) \\ \hline \(c_2\) & \(11\) & \(23\) & 35\\ \hline \((p_g,q)\) & \((0,0) \) & \( (1,0) \) &\( (2,0)\) \\ \hline \((s,t)\) & \((0,1) \) & \( (2,2) \) &\( (4, 3)\) \\ \hline \end{tabular} \end{center} Here are some surfaces within this range of invariants (the list is far from complete!): \begin{itemize} \item The so-called Godeaux-type surfaces, i.e. those with \(p_g=q=0\) and \(c_1^2=1\). For concrete examples, see e.g.\ [\textit{W. P. Barth} et al., Compact complex surfaces. 2nd enlarged ed. Berlin: Springer (2004; Zbl 1036.14016), \S VII.10]. Here \(\mathsf P_X\) is unimodular and negative definite of rank \(8\). It is known that then \(\mathsf P_X\simeq E_8(-1)\) (cf.\ Table 1). \item The Kynev surface from [\textit{V. I. Kynev}, C. R. Acad. Bulg. Sci. 30, 323--325 (1977; Zbl 0363.14005); \textit{A. N. Todorov}, Ann. Sci. Éc. Norm. Supér. (4) 13, 1--21 (1980; Zbl 0478.14030)] with \(c_1^2=1\), \(p_g=1\) and \(q=0\) (so that \(c_2=23\)). \item E.\ Horikawa's (simply connected) surface from [\textit{E. Horikawa}, Ann. Math. (2) 104, 357--387 (1976; Zbl 0339.14024)] with \(c_1^2=1\), \(c_2=35\) (so that \(b_2(X)=33\)). \end{itemize} \item The simplest non-unimodular \(\mathsf P_X\) are obtained for surfaces \(X\) with \(c_1^2(X)=2\) and \( K_X\) ample and \(X\) embedded by a suitable multiple of \( K_X\). Here \({\mathsf P_X}=\pm 2\). As before, using Noether's inequality, Debarre's inequality and the Noether formula, one arrives at the following sets of possible invariants: \begin{center} \begin{tabular}{@{} |c| c|c| c |c| @{}} \hline \(\chi\) & \(1\) & \(2\) & \(3\)& \(4\) \\ \hline \(c_2\) & \(10\) & \(22\) & \(34\) & \(46\) \\ \hline \((p_g,q)\) & \((0,0), (1,1) \) & \( (1,0) \) &\( (2,0)\) &\((3,0)\)\\ \hline %\((s,t)\) & \((0,1) \) & \( (2,2) \) &\( (4, 3)\) \\ % \hline \end{tabular} \end{center} These surfaces are known to exist. I give some examples: \begin{itemize} \item The numerical Campedelli surfaces, i.e., those with \(p_g=q=0\) and \(c_1^2=2\). Again, for examples see e.g.\ [\textit{W. P. Barth} et al., Compact complex surfaces. 2nd enlarged ed. Berlin: Springer (2004; Zbl 1036.14016), \S VII.10] For these, \(\mathsf P_X\) is negative definite of rank \(2\) and with \({\mathsf P_X}=-2\). It is known that \(\mathsf P_X\simeq E_7(-1)\). See Table 1 below. \item The surfaces with \(p_g=q=1\) and \(c_1^2=2\) have been completely classified. See [\textit{F. Catanese}, ``On a class of surfaces of general type, in: Algebraic surfaces. Heidelberg: Springer. 67--284 (2010)]. Here \(\mathsf P_X\) has signature \((2,9)\) and discriminant \(-2\). Such a lattice is isometric to \( \langle 2\rangle U E_8(-1)\). \item Horikawa's surface with \(c_1^2=2, p_g=3, q=0\) (and \(c_2=46\)) from [\textit{E. Horikawa}, Invent. Math. 37, 121--155 (1976; Zbl 0339.14025)]. Here \(\mathsf P_X\) has signature \((6,37)\) and discriminant \(-2\). Such a lattice is isometric to \( \langle 2\rangle^{ 5} U^4 E_8(-1)\). \end{itemize} \item Let \(X\) be an Enriques surface. Then \(\mathsf H_X\simeq UE_8(-1)\). Let \(c\) be a primitive vector in the \(U\) component, say \(c= e + f\) where \({e,f}\) is the standard basis of \(U\). Then \(c^\perp \simeq \langle -2 \rangle E_8(-1)\). By the main theorem in [\textit{G. L. Watson}, Mathematika 25, 57--67 (1978; Zbl 0368.10017)] this lattice has class number \(1\). By loc. cit. for vectors of the form \(c'=d.e+f\), \(d\not=\pm 1\), the class number of the lattice \((c')^\perp\) is larger than \(1\). In fact, to interpret Watson's results, one has to be careful since his terminology differs form what is nowadays usual. First of all, Watson only considers quadratic forms and so the associated bilinear forms (the polar forms) are always even. His notation compares to the one used in this note as follows: \(P=U\), \(Q=\langle 2\rangle\), \(B=A_2\). \(E=E_8\) so that the two forms of rank \(9\) having class number \(1\) are \(F_9= E_8\langle 2\rangle\) and \(G_9\), an indecomposable form of discriminant \(8\) (in loc. cit. the discriminant of forms of odd rank have been divided by \(2\)). The last form is not isometric to \(E_8\langle 8\rangle\) since \((G_9)_2=^3 U A_2 \langle- 3. 2^3\rangle\). \end{itemize} \begin{table}[htp] \caption{List of lattices \(\mathsf P_X\) for \(X\) canonically polarized with \(\chi(X)=12\).} \begin{center} \begin{tabular}{|c|c|c|} \hline \(c_1^2,\text{rank} {P_X}\) & lattice & discrim. form\\ \hline \((1,8)\)& \(E_8(-1)\) & \(0\)\\ \((2,7)\)& \(E_7(-1)\)& \(\langle -1/2\rangle\) \\ \((3,6)\)& \(E_6(-1) \) & \(\langle 1/3\rangle\) \\ \((4,5)\)& \(D_5(-1) \) & \(\langle -1/4\rangle\)\\ \((5,4)\) & \(A_4(-1) \) & \(\langle -4/5\rangle\)\\ \((6,3\))& \(A_2(-1)\langle -2\rangle\) & \(\langle 1/3\rangle\langle -1/2\rangle\) \\ \((7,2)\)&\(\begin{pmatrix} -4&1\\1&-2\end{pmatrix}\) & \(\langle 1/7\rangle \)\\ \((8,1)\)&\(\langle -8\rangle\) & \(\langle-1 / 8\rangle\)\\ \hline \end{tabular} \end{center} \end{table}
This page was built for publication: A note on the primitive cohomology lattice of a projective surface
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6188618)
