On the duality of F-theory and the CHL string in seven dimensions (Q2149064)

The paper under review exploits how the duality between \(F\)-theory and the CHL string in seven dimensions is reflected in the geometry and moduli of complex \(K3\) surfaces. To this end, the authors work out neat algebraic correspondences between certain 10-dimensional families of lattice polarized \(K3\) surfaces. In fact, everything is made very explicit by using first of all elliptic fibrations and Weierstrass models, but then also other standard models of \(K3\) surfaces such as double sextics and double covers of \(\mathbb P^1\times\mathbb P^1\) branched along a curve of bidegree \((4,4)\) -- or the minimal resolutions of these varieties in case there are rational double point singularities. Most of this can be phrased in terms of lattice theory (thanks to the Torelli theorem for \(K3\) surfaces), but key input also consists in symplectic involutions provided by 2-torsion sections (referred to as van Geemen--Sarti involutions), base change from rational elliptic surfaces, and double covers of \(K3\) surfaces induced by an even set of 8 disjoint smooth rational curves. A special role is also played by twists by order two elements of the Brauer group, following [\textit{B. van Geemen}, Adv. Math. 197, No. 1, 222--247 (2005; Zbl 1082.14040)]. The authors also study special 6-dimensional subfamilies of \(K3\) surfaces where some of the above notions come together. This is related to \(F\)-theory vacua and CHL string backgrounds in the presence of additional structure. The interested reader should be advised that some of the mathematical statements and arguments have to be taken with care, as there are quite a few typos, inaccuracies and even mistakes. To name some issues which the reviewer struggled with: 1. On p.\ 637, it is stated that the Mordell-Weil lattice is isomorphic to the dual \(S^\vee\) of the narrow Mordell-Weil lattice \(S\). While this notion follows Elkies and Shioda without citing either, This is stated in Thm.\ 9.1 in [\textit{T. Shioda}, Comment. Math. Univ. St. Pauli 39, No. 2, 211--240 (1990; Zbl 0725.14017)] only for the case where the Néron-Severi lattice is unimodular. Indeed this becomes a trivial statement for \(K3\) surfaces as unimodularity of the Néron-Severi lattice is easily seen to imply finite Mordell-Weil group MW, but there are obvious non-unimodular counterexamples, such as the \(K3\) surfaces with Néron-Severi lattice \(H\oplus E_8(-2)\) studied in the present paper. (Here \(H\) denote the hyperbolic plane and \(E_8\) the unique positive definite even unimodular lattice in rank 8.) 2. Let \(X\) be a jacobian elliptic surface \(X\) and consider the sublattice \(H\oplus V\subset\mathrm{NS}(X)\), where \(H\) is generated by zero section and general fibre and \(V\) is generated by all fibre components disjoint from the zero section. (Note that the authors denote \(V\) by \(T\) and call this the trivial lattice while others reserve the letter \(T\) for \(H\oplus V\) and define this as the trivial lattice.) In (3.2), it is stated that \[ \mathrm{NS}(X) \cong H \oplus V' \oplus S\tag{1} \] where \(V'\) denotes the saturation of \(V\) inside \(\mathrm{NS}(X)\) obtained by adjoining the torsion sections (considered in the orthogonal projection with respect to \(H\)). However, it follows directly from the standard isomorphism \[ \mathrm{MW}(X) \cong \mathrm{NS}(X)/(H \oplus V) \] (see Thm.\ 1.3 in [\textit{T. Shioda}, Comment. Math. Univ. St. Pauli 39, No. 2, 211--240 (1990; Zbl 0725.14017)]) that (1) may only hold if, up to the torsion, the Mordell-Weil group equals the narrow Mordell-Weil group. This, however, rarely persists -- for instance, on rational elliptic surfaces, it only holds for the extreme cases of Mordell-Weil rank 0 and 8 by the classification in [\textit{K. Oguiso} and \textit{T. Shioda}, Comment. Math. Univ. St. Pauli 40, No. 1, 83--99; Appendix 1: 96--97; Appendix 2: 97--98 (1991; Zbl 0757.14011)]. On \(K3\) surfaces (which maybe Lemma 3.1 tacitly restricts to), (1) is only guaranteed to hold if \(V=0\) (i.e. there are no reducible fibres at all) or if the Mordell-Weil group is finite. [Unfortunately this error corrupts some of the subsequent arguments, e.g.\ those under 6.] 3. On p.\ 638, the two cubics \(p, q\) don't want to have common components, and the base points of the cubic pencil may only yield independent sections if the pencil has no reducible fibres. 4. In Lemma 3.2 and elsewhere, what the authors refer to as `general \(K3\) surface' should be interpreted as `very general', i.e.\ in the complement of countably many closed subsets of the respective moduli space. 5. Further on p.\ 638, the authors decide to focus on the rational elliptic surface with Mordell-Weil lattice \(D_4\), but this does not exist (again by [\textit{K. Oguiso} and \textit{T. Shioda}, Comment. Math. Univ. St. Pauli 40, No. 1, 83--99; Appendix 1: 96--97; Appendix 2: 97--98 (1991; Zbl 0757.14011)]). The formulae at the bottom of the page indicate that \(D_4^\vee\) must be meant. 6. In Cor.\ 3.6 (1), the lattices are wrong as one can check, for instance, by what's often called the discriminant formula. Indeed the Mordell--Weil lattices will very generally be \(D_4^\vee(-2)\) (which is isometric to \(D_4\)), not \(D_4(-2)\) as stated. 7. Prop.\ 3.7 informs that the members of some 6-dimensional subfamily admit two commuting anti-symplectic involutions preserving the Jacobian fibration. This, however, holds true already on the 10-dimensional family (the involution on the fibers and the deck transformation), so here there is yet another involution commuting with the others. 8. In Prop.\ 3.8, the first lattice has to be corrected since the Mordell--Weil lattice should be \(D_4^\vee(-2)\), again. 9. In Cor.\ 3.11 (1) and its proof, it is not obvious to the reviewer why a second van Geemen-Sarti involution commuting with the first automatically respects the same elliptic fibration, and even if so, why it acts as translation by another 2-torsion section. In (2), it remains unclear why the antisymplectic involution can be assumed to be a deck transformation. For instance, Enriques involutions aren't. 10. In 3.4 on p.\ 644, the lattice \(L=H(2)\oplus E_8(-2)\) is embedded diagonally in \(H^2\oplus E_8(-2)^2\), but the perp \(L^\perp\) seems to be computed in the \(K3\) lattice \(H^3\oplus E_8(-2)^2\). 11. On pp.\ 645 ff., several elliptic fibrations are stated to admit no sections, but this holds only for (the very general \(K3\) surfaces in) some cases (see 13.). If it holds, a proof can be given a posteriori by computation/inspection of the Néron-Severi lattices. 12. Specifically, for \(\mathcal G\), it is not explained why the induced elliptic fibration \(\pi_{\mathcal G}\) has the same singular fibres as \(\mathcal X'\), or why in fact \(\mathcal X'\) appears as Jacobian of this fibration, but this seems to follow by direct computation using the general formulae (here displayed as (4.6)) applied to (4.7). However, this latter fibration has plenty of bisections, in contradiction to Prop.\ 4.2 (assuming that the multi-section index is given by the positive generator of the ideal \[ \{D.F; D\in \mathrm{Div}(\mathcal G)\}\subset\mathbb Z \] where \(F\) is a general fibre; note, however, that the reviewer was not able to find the notion of multi-section index in [\textit{F. R. Cossec} and \textit{I. V. Dolgachev}, Enriques surfaces. I. Boston, MA etc.: Birkhäuser Verlag (1989; Zbl 0665.14017)] to which the authors refer without further specification, but only the above notion of index of a genus one fibration (on p.\ 306 in loc.\ cit.)). 13. In Prop.\ 4.2, judging from the subsequent arguments, the \(K3\) surfaces \(\mathcal G\) should be polarized by the Enriques lattice \(H(2)\oplus E_8(-2)\) (which also appears in the introduction, but it comes out of nowhere in the proof of Prop.\ 4.2). However, this also turns out wrong; instead (2) holds true very generally as we shall briefly argue. In the proof, the statement about \(\mathrm{NS}(\mathcal G)\) being simultaneously a sublattice of \(\mathrm{NS}(\mathcal X')\) and of \(\mathrm{NS}(\mathcal Y)\) probably relies on [\textit{J. Keum}, Trans. Am. Math. Soc. 352, No. 5, 2077--2086 (2000; Zbl 0941.14012)] (cited at the end of the proof). Then this seems to imply that the indices in the present situation can only be 1 (in the presence of a section) or 2 (otherwise). The only solution to this issue seems to be that \(\pi_{\mathcal G'}\) already has a section (contrary to what is stated on p.\ 646) and \[ \mathrm{NS}(\mathcal G)\cong \mathrm{NS}(\mathcal Y)\cong H\oplus E_8(-2) \cong H(2)\oplus N\tag{2} \] (the last isometry being stated as (1.1) in the introduction). Indeed, it is obvious that one can solve for constant \((s:t)\) such that the RHS of (4.8) is a square polynomial in \(u,v\). to obtain the claimed section. Thus the indices are \(2\) and \(1\) instead of \(4\) and \(2\), and many things, especially Fig.\ 3 have to be adjusted. 14. Cor.\ 4.4 may still be valid, but it does not follow from Prop.\ 4.2 anymore. 15. Lem.\ 4.5 seems to miss some lattices as an index two sublattice of \(H^2\oplus E_8(-2)\) is by no means 2-elementary. Easy counterexamples are provided by \(H^2\oplus D_8(-2)\) and \(H^2\oplus A_1(-2)\oplus E_7(-2)\). The subsequent considerations have to be adjusted, including the counts for \(T_{\langle\theta\rangle}\). 16. On p.\ 649, it is unclear why elements \(\pm \theta\in\mathrm{Br}(\mathcal Y)_2\) differ at all. 17. In the proof of Prop.\ 4.6, it is not explained how every suitable element \(\theta\in\mathrm{Br}(\mathcal Y)_2\) determines an equation for \(\mathcal G_{2,0}\). Moreover, it is not quite obvious why the shape of the equation of \(\mathcal Y\) predicts the equation of \(\mathcal G_{2,0}\) to be even with respect to the homogeneous coordinates \(u, v\). In fact, if it took the shape of (4.8), then it would admit a section (as discussed above), but by construction, it doesn't. That is, the \(K3\) surfaces \(\mathcal G_{2,0}\) and what was previously called \(\mathcal G\) are distinguished by their (very general) Néron--Severi lattices, and equation (4.7) can very generally not apply to \(\mathcal G_{2,0}\), contrary to the statement of Prop.\ 4.6. The subsequent claims and arguments of the paper may also have to be revisited. Meanwhile the authors have informed the reviewer that they are preparing a corrected version of the paper; possibly there will also be an erratum published by the journal. They point out that the main results remain valid, in particular the mathematical framework for the F-theory/CHL heterotic string duality provided by \(K3\) surfaces, and that some arguments will be simplified.