Calabi–Yau threefolds fibred by high rank lattice polarized K3 surfaces
 66 Downloads
Abstract
We study threefolds fibred by K3 surfaces admitting a lattice polarization by a certain class of rank 19 lattices. We begin by showing that any family of such K3 surfaces is completely determined by a map from the base of the family to the appropriate K3 moduli space, which we call the generalized functional invariant. Then we show that if the threefold total space is a smooth Calabi–Yau, there are only finitely many possibilities for the polarizing lattice and the form of the generalized functional invariant. Finally, we construct explicit examples of Calabi–Yau threefolds realizing each case and compute their Hodge numbers.
1 Introduction
The primary aim of this paper is to study threefolds fibred by K3 surfaces polarized by a certain class of rank 19 lattice, with a particular focus on Calabi–Yau threefolds.
Our study of threefolds fibred by \(M_n\)polarized K3 surfaces was initiated in [9], where we performed a careful examination of threefolds fibred by \(M_2\)polarized K3 surfaces. In particular, we gave a classification of Calabi–Yau threefolds fibred nonisotrivially by \(M_2\)polarized K3 surfaces. This classification raised a natural question: for which n do there exist Calabi–Yau threefolds fibred nonisotrivially by \(M_n\)polarized K3 surfaces, and can they also be classified? The aim of this paper is to answer that question.
Our first main result (Theorem 2.2) is a generalization of [9, Theorem 2.3]. It shows that, for \(n \ge 2\), an \(M_n\)polarized family of K3 surfaces (in the sense of [8, Definition 2.1]) over a quasiprojective base curve U is completely determined by its generalized functional invariant map\(U \rightarrow \mathcal {M}_{M_n}\), where \(\mathcal {M}_{M_n}\) denotes the moduli space of \(M_n\)polarized K3 surfaces. The case \(n = 1\) is excluded here, as it is complicated by the fact that \(M_1\)polarized K3 surfaces admit an antisymplectic involution that fixes the \(M_1\)polarization. This means that the analogue of Theorem 2.2 does not hold for \(M_1\)polarized families of K3 surfaces (see Remark 2.3); we therefore primarily restrict our attention to the cases where \(n \ge 2\). It should, however, be noted that several Calabi–Yau threefolds fibred by \(M_1\)polarized K3 surfaces are known to exist (see [8, Theorem 5.20]), but no classification for them is currently known; the authors intend to address this in future work.
In the final part of this paper we study the converse statement, and ask which of the cases allowed by the theorems mentioned above can actually be realized by examples. Our main result here (Theorem 4.1) constructs explicit birational models for all such Calabi–Yau threefolds, and we compute Hodge numbers in all cases. This gives a substantial extension of the results of [9].
1.1 Structure of the paper
This paper is structured as follows. In Sect. 2 we begin by proving Theorem 2.2, which shows that any nonisotrivial \(M_n\)polarized family of K3 surfaces, for \(n \ge 2\), is uniquely determined by its generalized functional invariant map. Then we prove a Hodgetheoretic version of the same result: that the transcendental variation of Hodge structure associated to an \(M_n\)polarized family of K3 surfaces is the pullback of a fundamental variation of Hodge structure \(\mathbb {V}_n^+\) from the moduli space \(\mathcal {M}_{M_n}\) of \(M_n\)polarized K3 surfaces. Following on from this, we construct the variations of Hodge structure \(\mathbb {V}_n^+\) explicitly and study some of their properties. This allows us to prove Theorems 2.12 and 2.13, which show that a Calabi–Yau threefold may only admit a nonisotrivial \(M_n\)polarized K3 fibration if n is one of a small number of possibilities. Finally, we use results from [10] to prove Theorem 2.17, which places severe restrictions on the possible generalized functional invariants that may give rise to Calabi–Yau threefolds.
In Sect. 3 we study the converse problem, and attempt to construct explicit examples of Calabi–Yau threefolds realizing the possibilities allowed by our earlier results. By Theorem 2.2, these are all birational to pullbacks of certain fundamental families \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) under the generalized functional invariant map. Section 3 is dedicated to a detailed study of these fundamental families \(\mathcal {X}_n\); in particular, we conduct a careful examination of their singular fibres and resolutions.
Finally, Sect. 4 applies the results of the previous sections to the explicit construction of Calabi–Yau threefolds. The main result here is Theorem 4.1, which shows that all of the generalized functional invariants listed in Theorem 2.17 actually give rise to (possibly mildly singular) Calabi–Yau threefolds fibred by \(M_n\)polarized K3 surfaces. Following on from this, Propositions 4.3 and 4.4 explicitly compute the Hodge numbers of these Calabi–Yau threefolds.
Remark 1.1
Throughout this paper, a Calabi–Yau threefold will always be a projective threefold \(\mathcal {X}\) with trivial canonical sheaf \(\omega _{\mathcal {X}} \cong \mathcal {O}_{\mathcal {X}}\) and \(H^1(\mathcal {X},\mathcal {O}_{\mathcal {X}}) = 0\). Unless otherwise specified, we do not necessarily assume that \(\mathcal {X}\) is smooth, but we do assume that any singularities it has are \(\mathbb {Q}\)factorial and terminal.
2 A bound on possible lattices
The aim of this section is to place restrictions on the possible lattices \(M_n\) that may polarize the fibres in a Calabi–Yau threefold fibred by \(M_n\)polarized K3 surfaces. We will begin by establishing some general results that apply to any threefold fibred by \(M_n\)polarized K3 surfaces, then specialize to the Calabi–Yau case in Sect. 2.5.
We begin by setting up some notation. Let \(\mathcal {X}\) be a projective threefold with at worst terminal singularities, that admits a fibration \(\pi :\mathcal {X}\rightarrow B\) by K3 surfaces over a smooth base curve B. Let \(X_p\) denote the fibre of \(\pi \) over \(p \in B\) and let \({{\,\mathrm{NS}\,}}(X_p)\) denote the Néron–Severi group of \(X_p\). We will assume that, for general choice of p, we have \({{\,\mathrm{NS}\,}}(X_p) \cong M_n\), where \(M_n\) denotes the rank 19 lattice \(M_n := H \oplus E_8 \oplus E_8 \oplus \langle 2n \rangle \) (here H denotes the hyperbolic plane lattice and \(E_8\) denotes the negative definite \(E_8\) root lattice). Let \(M_n^{\perp } = H \oplus \langle 2n \rangle \) denote the orthogonal complement of \(M_n\) in the K3 lattice \(\Lambda _{\mathrm {K3}} := H^{\oplus 3} \oplus E_8^{\oplus 2}\).
As any singularities of \(\mathcal {X}\) are terminal, they must be isolated points by [20, Corollaries 5.38, 5.39]. Let \(U \subset B\) denote the open set over which \(\mathcal {X}\) is smooth and the fibres of \(\pi \) are smooth K3 surfaces, let \(j:U \hookrightarrow B\) denote the natural embedding, and let \(\pi ^U:\mathcal {X}^U \rightarrow U\) denote the restriction of \(\mathcal {X}\) to U. Suppose further that \(\pi ^U:\mathcal {X}^U \rightarrow U\) is an \(M_n\)polarized family of K3 surfaces in the sense of the following definition.
Definition 2.1

there is a trivial local subsystem \(\mathcal {L}\) of \(R^2\pi _*\mathbb {Z}\) so that, for each \(p \in \Delta ^*\), the fibre \(\mathcal {L}_p \subset H^2(X_p,\mathbb {Z})\) of \(\mathcal {L}\) over p is a primitive sublattice of \({{\,\mathrm{NS}\,}}(X_p)\) that is isomorphic to L, and

there is a line bundle \(\mathcal {A}\) on \(\mathcal {X}^U\) whose restriction \(\mathcal {A}_p\) to any fibre \(X_p\) is ample with first Chern class \(c_1(\mathcal {A}_p)\) contained in \(\mathcal {L}_p\) and primitive in \({{\,\mathrm{NS}\,}}(X_p)\).
In general, we will call such \(\mathcal {X}\) a threefold fibred by\(M_n\)polarized K3 surfaces. If, in addition, the family \(\pi ^U:\mathcal {X}^U \rightarrow U\) is not isotrivial, we will call \(\mathcal {X}\) a threefold fibred nonisotrivially by\(M_n\)polarized K3 surfaces.
2.1 Modular families and the generalized functional invariant
Results of Dolgachev [13] show that the moduli space of \(M_n\)polarized K3 surfaces is isomorphic to a dense open set in a certain modular curve. This modular curve admits a standard compactification, obtained by adding finitely many points (called cusps), which in turn gives rise to a compactification of the moduli space of \(M_n\)polarized K3 surfaces. We henceforth denote this compactified moduli space by \(\mathcal {M}_{M_n}\); full details of its construction are given in Sect. 2.3, below.
To any \(M_n\)polarized family of K3 surfaces \(\pi ^U:\mathcal {X}^U \rightarrow U\), we may associate a generalized functional invariant map\(g :U \rightarrow \mathcal {M}_{M_n}\), defined to be the map which takes a point \(p \in U\) to the point in moduli corresponding to the fibre \(X_p\). The generalized functional invariant map is important because of the following theorem.
Theorem 2.2
Suppose \(n \ge 2\). Let \(\pi ^U:\mathcal {X}^U \rightarrow U\) denote a nonisotrivial \(M_n\)polarized family of K3 surfaces over a quasiprojective curve U, such that the Néron–Severi group of a general fibre of \(\mathcal {X}^U\) is isomorphic to \(M_n\). Then \(\pi ^U:\mathcal {X}^U \rightarrow U\) is uniquely determined (up to isomorphism) by its generalized functional invariant map \(g :U \rightarrow \mathcal {M}_{M_n}\).
Proof
This was proved for \(n = 2\) in [9, Theorem 2.3]. The result in the general case follows by the same argument, after noting that the discriminant group \(A_{M_n^{\perp }}\) is isomorphic to \(\mathbb {Z}/2n\mathbb {Z}\).
\(\square \)
Remark 2.3
In the case \(n=1\), the existence of an antisymplectic involution fixing the \(M_1\)polarization on an \(M_1\)polarized K3 surface gives rise to several pathologies, one of which is that this result does not hold. For this reason, in this paper we will always restrict our attention to the cases where \(n \ge 2\), unless otherwise stated. More details about the case \(n=1\) may be found in [8, Sect. 5.5]; it is our intention to address this case in more detail in future work.
A consequence of this theorem is that any \(M_n\)polarized family of K3 surfaces \(\pi ^U:\mathcal {X}^U \rightarrow U\), for \(n \ge 2\), is the pullback of a fundamental modular family \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) under the generalized functional invariant map.
In particular, if X is a K3 surface which is lattice polarized by the lattice \(M_n = M \oplus \langle 2n \rangle \), then X can be expressed as a resolution of one of the surfaces X(a, b, d). One may ask whether the subvarieties of the (a, b, d) domain corresponding to \(M_n\)polarized K3 surfaces are indeed the moduli spaces \(\mathcal {M}_{M_n}\). There is a map from \(\mathcal {M}_{M_n}\) to \({\mathbb {WP}}(2,3,6)\), whose degree is given by the index of the subgroup of \(\mathrm {O}(\Lambda _{\mathrm {K3}})\) which acts as the identity on \(M_n\) inside the subgroup of \(\mathrm {O}(\Lambda _{\mathrm {K3}})\) which acts as the identity on M and sends \(M_n^\perp \) to itself. The following result implies that this index is 1.
Lemma 2.4
Suppose \(n \ge 2\). If \(\gamma \in \mathrm {O}(\Lambda _{\mathrm {K3}})\) acts as the identity on the lattice M and sends \(M_n^\perp \) to itself, then \(\gamma \) acts as the identity on \(M_n\).
Proof
This is a simple exercise in lattice theory using [24, Corollary 1.5.2].\(\square \)
Therefore, the curves in \({\mathbb {WP}}(2,3,6)\) which support \(M_n\)polarized K3 surfaces may be identified with the curves \(\mathcal {M}_{M_n}\) whenever \(n \ge 2\). So for \(n \ge 2\) we may define a family \(\mathcal {X}_n\) of K3 surfaces over \(\mathcal {M}_{M_n}\) to be a maximal crepant resolution of the pullback of the family X(a, b, d) by the map \(\mathcal {M}_{M_n} \rightarrow {\mathbb {WP}}(2,3,6)\) (note that the general fibre of \(\mathcal {X}_n\) is smooth, but the total space may still be singular). By construction, the fibres of \(\mathcal {X}_n\) are smooth K3 surfaces admitting \(M_n\)polarizations and the period map is of degree 1 onto the moduli space of \(M_n\)polarized K3 surfaces. The fact that \(\mathcal {X}_n\) defines an \(M_n\)polarized family, in the sense of Definition 2.1, then follows from Lemma 2.4.
We will compute some of the families \(\mathcal {X}_n\) explicitly, for small values of n, in Sect. 3; but in general this is difficult to do. Therefore, in order to deduce information about \(M_n\)polarized families of K3 surfaces \(\pi ^U:\mathcal {X}^U \rightarrow U\) for arbitrary \(n \ge 2\), we instead turn to Hodge theory.
2.2 Some Hodge theory
The family \(\pi ^U:\mathcal {X}^U \rightarrow U\) determines a weight 2 variation of polarized Hodge structure \(\mathcal {H}_{\mathcal {X}}\) on \(R^2\pi ^U_* \mathbb {Z}\). Following the discussion of [8, Sect. 2.1], define \(\mathcal {T}(\mathcal {X}^U)\) to be the integral local subsystem of \(R^2\pi ^U_* \mathbb {Z}\) underlying the smallest integral subvariation of Hodge structure of \(\mathcal {H}_{\mathcal {X}}\) containing \(\mathcal {H}^{2,0}_{\mathcal {X}} := \mathcal {F}^2(R^2\pi ^U_* \mathbb {Z})\). The local system \(\mathcal {T}(\mathcal {X}^U)\) thus supports a variation of polarized Hodge structure, with polarization induced from \(\mathcal {H}_{\mathcal {X}}\) (this variation of Hodge structure is the “essential part” of \(\mathcal {H}_{\mathcal {X}}\), as defined by Saito and Zucker [29, Sect. 4]).
The direct image \(j_*\mathcal {T}(\mathcal {X}_U)\) of this sheaf on B admits a Hodge structure of weight 3 on its cohomology \(H^1(B,j_*\mathcal {T}(\mathcal {X}_U)\otimes \mathbb {Q})\), defined as follows. Indeed, suppose that \(\mathbb {V}\) is any \(\mathbb {Q}\)local system defined on a Zariski open set \(j:U \hookrightarrow B\), which supports a variation of Hodge structure of weight k. Following Deligne [5], the associated Hodge filtration \(\mathcal {F}^i\) may be extended to coherent sheaves \(\widetilde{\mathcal {F}}^i\) over the entire curve B in a (quasi)canonical manner. Zucker [31] then shows how to use the sheaves \(\widetilde{\mathcal {F}}^i\) to equip the cohomology group \(H^1(B,j_*\mathbb {V})\) with a pure polarizable Hodge structure of weight \(k+1\).
Our main use for this Hodge structure arises from the following result, which allows us to deduce information about the Hodge structure on \(H^3(\mathcal {X},\mathbb {C})\) from that on \(H^1(U,j_*\mathcal {T}(\mathcal {X}^U) \otimes \mathbb {C})\). In particular, it will allow us to compute the geometric genus of \(\mathcal {X}\), which we can use to restrict when \(\mathcal {X}\) may be Calabi–Yau.
Lemma 2.5
Proof
Given this, we wish to find a way to study the local system \(\mathcal {T}(\mathcal {X}^U)\). The aim of the next subsection is to prove a Hodge theoretic version of Theorem 2.2 which will do just that: we will prove that the variation of Hodge structure on \(\mathcal {T}(\mathcal {X}^U)\) is obtained as the pullback of a given variation of Hodge structure \(\mathbb {V}_n^+\) on \(\mathcal {M}_{M_n}\) by the generalized functional invariant. To construct this variation of Hodge structure, we first examine the structure of \(\mathcal {M}_{M_n}\).
2.3 Modular curves and variations of Hodge structure
Now let \(\mathcal {F}^2\) be the tautological vector bundle \(\mathcal {O}_{{\mathbb {P}}^2}(1)\) restricted to \(\mathcal {D}_n\), which is a subbundle of the vector bundle \(\mathcal {F}^0 := (M_n^{\perp }\otimes \mathcal {O}_{{\mathbb {P}}^2})_{{\mathcal {D}}_n}\) on \(\mathcal {D}_n\). Setting \(\mathcal {F}^1\) to be the fibrewise orthogonal complement of \(\mathcal {F}^2\) in \(\mathcal {F}^0\), we obtain a weight 2 variation of Hodge structure \(\mathcal {F}^2 \subseteq \mathcal {F}^1 \subseteq \mathcal {F}^0\), which descends to variations of Hodge structure \(\mathbb {V}_n\) and \(\mathbb {V}_n^+\) on the quotients \(\mathcal {D}_n/{\mathrm {SO}}(M_n^{\perp })^*\) and \(\mathcal {D}_n/\mathrm {O}(M_n^{\perp })^*\) respectively. These two variations of Hodge structure are closely related: there is a double covering \(\mathcal {D}_n/{\mathrm {SO}}(M_n^{\perp })^* \rightarrow \mathcal {D}_n/\mathrm {O}(M_n^{\perp })^*\), and \(\mathbb {V}_n\) is the pullback of \(\mathbb {V}_n^+\) under this covering map.
The following should be thought of as a Hodge theoretic version of Theorem 2.2.
Proposition 2.6
Suppose \(n \ge 2\). Let \(\pi ^U:\mathcal {X}^U \rightarrow U\) be a nonisotrivial \(M_n\)polarized family of K3 surfaces over a quasiprojective curve U, such that the Néron–Severi group of a general fibre of \(\mathcal {X}^U\) is isomorphic to \(M_n\). Then the variation of Hodge structure on \(\mathcal {T}(\mathcal {X}^U)\) is the pullback of \(\mathbb {V}_n^+\) by the generalized functional invariant map \(g:U \rightarrow \mathcal {M}_{M_n}\).
Proof
This proposition is proved in largely the same way as Theorem 2.2 (and, by extension, [9, Theorem 2.3]). Assume for a contradiction that the local systems \(\mathcal {T}(\mathcal {X}^U)\) and \(g^*(\mathbb {V}_n^+)\) differ on U. Let \(\{U_i\}\) denote an open cover of U, with each \(U_i\) simply connected. As the local systems \(\mathcal {T}(\mathcal {X}^U)\) and \(g^*(\mathbb {V}_n^+)\) are trivial on each \(U_i\), they must differ in the way that they glue over the intersections \(U_i \cap U_j\). Let V be a connected component of such an intersection, so that the gluing maps differ over V.
Over V, the two gluing maps must differ by a fibrewise automorphism in the group \(\mathrm {O}(M_n^\perp )^*\). But it follows from a slight modification of the argument in [9, Theorem 2.3] (see Theorem 2.2) that no such automorphism exists for \(n > 1\). Thus \(\mathcal {T}(\mathcal {X}^U)\) and \(g^*(\mathbb {V}_n^+)\) are isomorphic as local systems.
It remains to show that the variations of Hodge structure agree. To do this, it suffices to show that \(g^*\mathcal {F}^2(\mathbb {V}^+) = \mathcal {F}^2(\mathcal {T}(\mathcal {X}^U))\). But this is a standard property of the period map. \(\square \)
In order to use this result to study \(M_n\)polarized families of K3 surfaces, we need to understand the variations of Hodge structure \(\mathbb {V}_n^+\). As we shall see, it will turn out to be simpler to study the closely related variations of Hodge structure \(\mathbb {V}_n\), then use them to deduce the properties of \(\mathbb {V}_n^+\).
With this notation, define \(Y_0(n) := \mathbb {H}/\Gamma _0(n)\) and \(Y_0(n)^+ := \mathbb {H}/\Gamma _0(n)^+\), where \(\mathbb {H}\) denotes the upper halfplane in \(\mathbb {C}\), and let \(i:Y_0(n) \hookrightarrow X_0(n) \) and \(i^+:Y_0(n)^+ \hookrightarrow X_0(n)^+\) denote their compactifications. Dolgachev [13, Theorem 7.1] has proved that there is an isomorphism \(\mathcal {D}_n/\mathrm {O}(M_n^{\perp })^* \cong Y_0(n)^+\) so, in particular, \(X_0(n)^+\) defines a compactification \(\mathcal {M}_{M_n}\) of the moduli space of \(M_n\)polarized K3 surfaces.
Let \(\mathcal {D}_n^+ \cong \mathbb {H}\) denote a connected component of \(\mathcal {D}_n\) and let \({\mathrm {SO}}(M_n^{\perp })^{+}\) denote the index 2 subgroup of \({\mathrm {SO}}(M_n^{\perp })^{*}\) that stabilizes \(\mathcal {D}_n^+\). The map A sends \(\Gamma _0(n)\) to the group \({\mathrm {SO}}(M_n^\perp )^+\). A slight modification of the argument in [13, Sect. 7] then shows that the quotient \(\mathcal {D}_n/{\mathrm {SO}}(M_n^{\perp })^* \cong \mathcal {D}_n^+/{\mathrm {SO}}(M_n^{\perp })^{+}\) is isomorphic to the modular curve \(Y_0(n)\).
We may use this to explicitly construct the variation of Hodge structure \(\mathbb {V}_n\) on \(\mathcal {D}_n/{\mathrm {SO}}(M_n^{\perp })^* \cong Y_0(n)\).
Proposition 2.7
The variation of Hodge structure \(\mathbb {V}_n\) is the symmetric square of a polarizable weight one \(\mathbb {Z}\)variation of Hodge structure \(\mathbb {V}_{\sqrt{n}}\) over \(Y_0(n)\).
Proof
Begin by considering the weight 1 variation of Hodge structure \(\mathcal {F}^1 \subseteq \mathcal {F}^0\) on \(\mathbb {H} \times \mathbb {Z}^2\), defined so that \(\mathcal {F}^1\) is the subbundle of \(\mathcal {F}^0 := \mathbb {H} \times \mathbb {C}^2\) whose fibre over \(\tau \in \mathbb {H}\) is given by the subspace in \( \mathbb {C}^2\) spanned by \((\tau ,1)\).
Remark 2.8
Note that a similar result does not hold for the variation of Hodge structure \(\mathbb {V}_n^+\). This is because the matrix \(\iota \in \mathrm {O}(M_n^\perp )^*\) has no preimage under A in \({\mathrm {SL}}_2(\mathbb {R})\), so the action of \(\mathrm {O}(M_n^\perp )^*\) is not the one induced by applying A to \(\Gamma _0(n)^+\). Indeed, in the basis (e, f, g) the matrix of \(\iota \) has determinant \(1\), so it cannot come from an element of \({\mathrm {SL}}_2(\mathbb {R})\).
Now that we understand \(\mathbb {V}_n\), we may use it to deduce properties of \(\mathbb {V}_n^+\). In order to apply Lemma 2.5 and access Hodge theoretic data about threefolds fibred by \(M_n\)polarized K3 surfaces, we need to understand the Hodge decompositions induced on the cohomology groups \(H^1(X_0(n),i_*\mathbb {V}_n)\) and \(H^1(X_0(n)^+,i^+_*\mathbb {V}^+_n)\). We begin with a technical lemma that will allow us to compare these cohomology groups.
Lemma 2.9
Proof
Using this diagram, we see that to prove injectivity of \(g^*\), it suffices to prove that the map \(H^3(\widetilde{\mathcal {X}}_n,\mathbb {C}) \rightarrow H^3(\mathcal {X}',\mathbb {C})\) is injective. But this map is induced by the pullback along the surjective map \(\mathcal {X}' \rightarrow \widetilde{\mathcal {X}}_n\), obtained as the composition of the resolution and the pullback by g, so its injectivity follows immediately from [30, Lemma 7.28]. \(\square \)
Using this, we can prove an important result about the Hodge decompositions on the cohomology groups \(H^1(X_0(n),i_*\mathbb {V}_n)\) and \(H^1(X_0(n)^+,i^+_*\mathbb {V}^+_n)\)
Proposition 2.10
The Hodge numbers \(h^{p,q}\) of the weight 3 Hodge structures on \(H^1(X_0(n),i_*\mathbb {V}_n)\) and \(H^1(X_0(n)^+,i^+_*\mathbb {V}_n^+)\) are trivial if \((p,q) = (2,1)\) or (1, 2).
Proof
Shimura [28] has shown that there is a Hodge structure of weight 3 on the space \(S_2(\Gamma _0(n)) \oplus \overline{S_2(\Gamma _0(n))}\), where \(S_2(\Gamma _0(n))\) denotes the space of \(\mathbb {C}^2\)valued cusp forms of weight 2 for \(\Gamma _0(n)\), which is concentrated in the (0, 3) and (3, 0) components. Zucker [31, Sect. 12] proved that this Hodge structure agrees with the Hodge structure on \(H^1(X_0(n),i_*\mathbb {V}_n)\) described above; this completes the proof in the \(X_0(n)\) case. Given this, the \(X_0(n)^+\) case follows immediately from Lemma 2.9. \(\square \)
We can use this to compute the Hodge numbers \(h^{3,0}\) and \(h^{0,3}\) of the Hodge structure on \(H^1(X_0(n)^+,i^+_*\mathbb {V}^+_n)\). Indeed, since we know that the Hodge numbers \(h^{1,2}\) and \(h^{2,1}\) are trivial, it suffices to compute the rank of \(H^1(X_0(n)^+,i^+_*\mathbb {V}_n^+)\). To do this, we follow the method of [9, Sect. 3.3], which requires us to determine the local monodromy matrices of the local system \(\mathbb {V}_n^+\).
2.4 Computation of the local monodromy
We begin by describing the local monodromy matrices of \(\mathbb {V}_n\), which we compute using Proposition 2.7 and the wellknown local monodromy matrices of \(\mathbb {V}_{\sqrt{n}}\). For \(n > 4\), we can use these to deduce the local monodromy matrices of \(\mathbb {V}_n^+\). The cases \(n \in \{2,3,4\}\) will then be tackled individually.
Lemma 2.11
 (1)
If \(n \ne 4\), the Fricke involution \(\tau _n\) does not fix any cusp of \(X_0(n)\).
 (2)
If \(n > 3\), the Fricke involution \(\tau _n\) does not fix any elliptic orbifold point of \(X_0(n)\).
Proof
The first statement is precisely [25, Proposition 3] and the second follows easily from the discussion following it. In particular, [25] shows that the fixed points of \(\tau _n\) correspond to specific elliptic curves with level n structure and with complex multiplication by \(\mathbb {Q}(\sqrt{n})\). \(\square \)
 (1)
If \(n=2\), the Fricke involution \(\tau _2\) fixes one smooth point in \(X_0(2)\) and the orbifold point of order 2.
 (2)
If \(n=3\), the Fricke involution \(\tau _3\) fixes one smooth point in \(X_0(3)\) and the orbifold point of order 3.
 (3)
If \(n=4\), the Fricke involution \(\tau _4\) fixes one smooth point in \(X_0(4)\) and a cusp of width 4.
To compute the local monodromy matrices in the local systems \(\mathbb {V}_3^+\) and \(\mathbb {V}_4^+\), we use the fundamental modular families \(\mathcal {X}_3 \rightarrow \mathcal {M}_{M_3}\) and \(\mathcal {X}_4 \rightarrow \mathcal {M}_{M_4}\). These families were computed explicitly in [8, Sect. 5.4] (see also Proposition 3.1, below). Moreover, it follows from Proposition 2.6 that \(\mathbb {V}_3^+\) and \(\mathbb {V}_4^+\) agree with the transcendental variations of Hodge structure associated with these families.
2.5 Application to Calabi–Yau threefolds
We wish to use this theory to restrict which lattices \(M_n\) can arise as polarizations in a Calabi–Yau threefold fibred nonisotrivially by \(M_n\)polarized K3 surfaces.
So let \(\pi :\mathcal {X}\rightarrow B\) be a smooth Calabi–Yau threefold fibred nonisotrivially by \(M_n\)polarized K3 surfaces (note that the cohomological condition in the definition of Calabi–Yau threefold necessitates \(B \cong \mathbb {P}^1\)). Define \(\pi ^U:\mathcal {X}^U \rightarrow U\) as before, and let \(j:U \rightarrow B\) denote the inclusion.
 (1)
The curve \(X_0(n)^+\) must have genus 0 (by Hurwitz’s theorem), and
 (2)
The local system \(\mathbb {V}_n^+\) must satisfy \({{\,\mathrm{rank}\,}}H^1(X_0(n)^+, i^+_*\mathbb {V}_n^+) \le 2\).
Theorem 2.12
Proof
This is easy to check explicitly using Eq. (4). \(\square \)
Recall that a smooth Calabi–Yau threefold \(\mathcal {X}\) is called rigid if \(h^{2,1}(\mathcal {X}) = 0\).
Theorem 2.13
Proof
We begin by showing that any small deformation of \(\mathcal {X}\) must also admit a fibration by \(M_n\)polarized K3 surfaces. The first step in this is to show that any small deformation of \(\mathcal {X}\) must contain a K3 surface. This will follow if we can show that any small deformation of \(\mathcal {X}\) lifts to a small deformation of the pair \((\mathcal {X},S)\), where S is a general fibre of the K3 fibration on \(\mathcal {X}\).
Finally, we show that this K3 fibration is \(M_n\)polarized. Let \(G_S\) denote the subgroup of \(H^2(S,\mathbb {Z})\) obtained as the pullback of \(H^2(\mathcal {X},\mathbb {Z})\) under the inclusion of S into \(\mathcal {X}\), equipped with the usual cupproduct pairing. \(G_S\) is a topological invariant of the pair \((\mathcal {X},S)\). Moreover, \(G_S\) is precisely the lattice polarization of \(\pi :\mathcal {X}\rightarrow \mathbb {P}^1\). This shows that any small deformation of \(\mathcal {X}\) remains fibred by K3 surfaces and, more importantly, this K3 surface fibration is \(M_n\)polarized.
Now, by Eq. (5), we have an injective morphism of Hodge structures \(H^1(X_0(n)^+,i^+_* \mathbb {V}^+_0) \hookrightarrow H^3(\mathcal {X},\mathbb {C})\). Moreover, it follows from Theorem 2.12 that \(H^1(X_0(n)^+,i^+_*\mathbb {V}_n^+) =2\) and, by Proposition 2.10, we see that the Hodge structure on \(H^1(X_0(n)^+,i^+_*\mathbb {V}_n^+)\) is concentrated in the (3, 0) and (0, 3) components. Thus, since \(\mathcal {X}\) is a Calabi–Yau threefold, it follows that there is some subHodge structure \(\mathcal {H}\) of \(H^3(\mathcal {X},\mathbb {C})\) with components only of types (1, 2) and (2, 1), so that \(H^3(\mathcal {X}, \mathbb {C})\) is isomorphic to \(H^1(X_0(n)^+,i^+_* \mathbb {V}^+_0) \oplus \mathcal {H}\).
Such a decomposition of Hodge structures also exists for a generic deformation of \(\mathcal {X}\), as such deformations are also fibred by \(M_n\)polarized K3 surfaces. However, Bardelli has proven [1, Proposition 1.1.2] that this implies that a generic deformation has trivial KodairaSpencer map; in other words, \(\mathcal {X}\) must be rigid by the local Torelli theorem for Calabi–Yau manifolds. \(\square \)
Corollary 2.14
Remark 2.15
Note that several examples of smooth, nonrigid Calabi–Yau threefolds fibred nonisotrivially by \(M_1\)polarized K3 surfaces are given in [8, Theorem 5.20]. The remaining cases will be addressed in the rest of this paper.
Remark 2.16
Using the wellknown classification by Iskovskih [18, 19], one may easily compute the hypersurface degrees for all smooth Fano threefolds of Picard rank 1 (c.f. [16, Theorem 2.1]). The set of integers one obtains are precisely those appearing in Corollary 2.14. We postulate that this is not a coincidence. Indeed, if Y is a Fano threefold of Picard rank 1 and hypersurface degree n, then a generic anticanonical hypersurface in Y is a smooth K3 surface with Picard group of rank 1 and Picard lattice isomorphic to \(\langle 2n \rangle \). According to Dolgachev [13], the mirror of a K3 surface with Picard group isomorphic to \(\langle 2n \rangle \) is an \(M_n\)polarized K3 surface. In other words, the \(M_n\)polarized K3 surfaces which appear as the fibres of smooth, nonrigid Calabi–Yau threefolds fibred nonisotrivially by \(M_n\)polarized K3 surfaces are precisely those which are mirror to anticanonical K3 surfaces in smooth Fano threefolds of Picard rank 1. This mirror correspondence is closely related to the FanoLG correspondence for smooth Fano threefolds of Picard rank 1, it is explored further in [11, 21] (see also Remark 4.7, below).
2.6 A further restriction
In order to achieve our goal of completely classifying nonrigid Calabi–Yau threefolds admitting nonisotrivial fibrations by \(M_n\)polarized K3 surfaces with \(n \ge 2\), we need to place some additional restrictions on the form of the generalized functional invariant maps \(g :\mathbb {P}^1 \rightarrow \mathcal {M}_{M_n}\) that may arise.
We begin by setting up some extra notation to describe these generalized functional invariants. Note first that, for \(n \in \{2,3,4,5,6,7,8,9,11\}\), the moduli space \(\mathcal {M}_{M_n}\) has orbifold points of types \((2,\ldots ,2,a,\infty )\), where \(\infty \) denotes a cusp (which has order \(\infty \)) and \(a \ge 2\) is an integer or \(\infty \) (in the case where \(\mathcal {M}_{M_n}\) contains two cusps). A complete list of these orbifold points is given later, in Table 4. Fix a parameter \(\lambda \) on \(\mathcal {M}_{M_n}\) so that the aorbifold point occurs at \(\lambda = 0\), the cusp occurs at \(\lambda = \infty \), and the 2orbifold points occur at \(\lambda = \lambda _1,\ldots ,\lambda _q\). Note that in the case \(n=4\), where we have two cusps of different widths, the cusp of width 1 occurs at \(\lambda = \infty \) and the cusp of width 2 occurs at \(\lambda = 0\).
Theorem 2.17
Suppose that \(\mathcal {X}\) is a smooth nonrigid Calabi–Yau threefold fibred nonisotrivially by \(M_n\)polarized K3 surfaces, with \(n \ge 2\). Then the ramification profile \([y_1,\ldots ,y_l]\) of the generalized functional invariant map of \(\mathcal {X}\) over the aorbifold point in \(\mathcal {M}_{M_n}\) must be one of the possibilities listed in Table 1.
Allowable partitions \([y_1,\ldots ,y_l]\) of d
n  Partitions 

2  [1, 1], [1, 2], [1, 3], [1, 4], [2, 2], [2, 3] [2, 4], [3, 3], [3, 4] [4, 4],[5], [6], [7], [8] 
3  [1, 1], [1, 2], [1, 3], [2, 2], [2, 3], [3, 3], [4], [5], [6] 
4  [1, 1], [1, 2], [2, 2], [3], [4] 
5  [1, 1], [1, 2], [2, 2], [3], [4] 
6  [1, 1], [2] 
7  [1, 1], [2], [3] 
8  [1, 1], [2] 
9  [1, 1], [2] 
11  [1, 1], [2] 
3 Modular families of lattice polarized K3 surfaces
Next we turn our attention to the converse question: when do smooth nonrigid Calabi–Yau threefolds fibred nonisotrivially by \(M_n\)polarized K3 surfaces satisfying the conditions of the previous section actually exist? To answer this question, we will attempt to find an explicit construction for these threefolds.
The construction of such Calabi–Yau threefolds is a subject to which we have already given some attention: several explicit examples for \(2 \le n \le 4\) are studied in [8, Theorem 5.10] and the case \(n = 2\) is treated in detail in [9].
By Theorem 2.2, if \(\pi :\mathcal {X}\rightarrow B\) is such a Calabi–Yau threefold, the family \(\pi ^U:\mathcal {X}^U \rightarrow U\) is given by the pullback of the modular family \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) (constructed in Sect. 2.1) by the generalized functional invariant map \(g :U \rightarrow \mathcal {M}_{M_n}\). Thus in order to construct \(\mathcal {X}\), we begin by studying the families \(\mathcal {X}_n\).
3.1 The modular families \(\mathcal {X}_n\) for small n
The families \(\bar{\mathcal {X}}_n\)
n  Ambient space  Family 

2  \(\mathbb {P}^3[x,y,z,t]\)  \(t^4 + \lambda xyz(x+y+zt) = 0\) 
3  \(\mathbb {P}^1[r,s] \times \mathbb {P}^2[x,y,z]\)  \(s^2z^3 + \lambda r(rs)xy(zxy) = 0\) 
4  \(\prod _{i=1}^3\mathbb {P}^1[r_i,s_i]\)  \(s_1^2s_2^2s_3^2  \lambda r_1(s_1r_1)r_2(s_2r_2)r_3(s_3r_3) = 0\) 
5  \(\mathbb {P}^3[x,y,z,t]\)  \((x^2+xy + xz + xt + yz + yt + zt)^2  \lambda xyzt = 0\) 
6  \(\mathbb {P}^3[x,y,z,t]\)  \((x+z+t)(x+y+z+t)(z+t)(y+z)  \lambda xyzt = 0\) 
7  \(\mathbb {P}^3[x,y,z,t]\)  \( (x+y+z+t)(y+z+t)(z+t)^2+ (x+y+z+t)^2yz  \lambda xyzt = 0\) 
8  \(\mathbb {P}^3[x,y,z,t]\)  \((x+y+z+t)(x+t)(y+t)(z+t)  \lambda xyzt = 0\) 
9  \(\mathbb {P}^3[x,y,z,t]\)  \((x+y+z)(xt^2 + yt^2 + zt^2 + xyt + xzt + yzt + xyz)  \lambda xyzt = 0\) 
11  \(\mathbb {P}^3[x,y,z,t]\)  \((z+t)(x+y+t)(xy+zt)+x^2y^2+xyz^2+3xyzt  \lambda xyzt = 0\) 
Singularities of a generic fibre of \(\bar{\mathcal {X}}_n\)
n  Singularities 

2  \(6 \times A_3\) 
3  \(3 \times A_1\), \(6 \times A_2\) 
4  \(12 \times A_1\) 
5  \(3 \times A_1\), \(3 \times D_4\) 
6  \(3 \times A_1\), \(2 \times A_2\), \(2 \times A_3\) 
7  \(1\times A_1\), \(3 \times A_2\), \(1 \times A_3\), \(1 \times A_4\) 
8  \(6 \times A_1\), \(3 \times A_2\) 
9  \(3 \times A_1\), \(3 \times A_2\), \(1 \times D_4\) 
11  \(2\times A_1\), \(1 \times A_2\), \(2 \times A_3\) 
Orbifold points of \(\mathcal {M}_{M_n}\)
n  Orbifold type  \(\lambda _i\) 

2  \((2,4,\infty )\)  256 
3  \((2,6,\infty )\)  108 
4  \((2,\infty ,\infty )\)  64 
5  \((2,2,2,\infty )\)  \(22 + 10\sqrt{5}\), \(22  10\sqrt{5}\) 
6  \((2,2,\infty ,\infty )\)  \(17 + 12\sqrt{2}\), \(17  12\sqrt{2}\) 
7  \((2,2,3,\infty )\)  \(1\), 27 
8  \((2,2,\infty ,\infty )\)  \(12 + 8\sqrt{2}\), \(12  8\sqrt{2}\) 
9  \((2,2,\infty ,\infty )\)  \(9 + 6\sqrt{3}\), \(9  6\sqrt{3}\) 
11  \((2,2,2,2,\infty )\)  Roots of the cubic \(\lambda ^320\lambda ^2+56\lambda 44\) 
Let \(U_{M_n}\) denote the open subset of \(\mathcal {M}_{M_n}\) obtained by removing these orbifold points. The restriction \(\hat{\mathcal {X}}_{n}^U\) of \(\hat{\mathcal {X}}_n\) to \(U_{M_n}\) is a family of smooth K3 surfaces. The following proposition shows that these families are the fundamental modular families \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) that we have been seeking. In light of this result, in the following sections we will omit the hat and simply denote these families by \(\mathcal {X}_n\), and their restrictions to \(U_{M_n}\) by \(\mathcal {X}_n^U\).
Proposition 3.1
For \(n \in \{2,3,4,5,6,7,8,9,11\}\), the families \(\hat{\mathcal {X}}_n^U \rightarrow U_{M_n}\) defined as above are \(M_n\)polarized families of K3 surfaces in the sense of Definition 2.1. Moreover, these families are modular, in the sense that their generalized functional invariant maps are isomorphisms.
The rest of this subsection is devoted to a proof of this proposition. For \(n \in \{2,3,4\}\), Proposition 3.1 was proved in [8, Sect. 5.4]. So suppose that \(n \in \{5,6,7,8,9,11\}\). In these cases, Doran et al. [7] have shown that \(\hat{\mathcal {X}}_n^U \rightarrow U_{M_n}\) is an \(M_n\)polarized family of K3 surfaces. It therefore only remains to prove modularity of these families.
To do this, we begin by showing that, if \(p \in \mathcal {M}_{M_n}\) is an orbifold point with order k (as given in Table 4), the local monodromy of the transcendental local system \(\mathcal {T}(\hat{\mathcal {X}}_n^U)\) has order k around p.
To compute the orders of the local monodromies of \(\mathcal {T}(\hat{\mathcal {X}}_n^U)\), we turn to the properties of the families \(\hat{\mathcal {X}}_n^U\). Przyjalkowski [26] constructs these families as weak Landau–Ginzburg models of certain rank 1 Fano threefolds. In particular, he shows that the holomorphic periods of the families \(\hat{\mathcal {X}}_n^U\) are equal to the Iseries of the corresponding Fano threefolds. It follows that the regularized quantum differential operators associated to these Fano threefolds, which annihilate the Iseries, are PicardFuchs differential equations for the families \(\hat{\mathcal {X}}_n^U\). These regularized quantum differential operators are computed for all rank 1 Fano threefolds by Golyshev [16]; the family \(\hat{\mathcal {X}}_n^U\) corresponds to row n of the \(d=1\) table in [16, Sect. 5.8].
Once one has the PicardFuchs differential equations for the families \(\hat{\mathcal {X}}_n^U\), computation of the local monodromies of \(\mathcal {T}(\hat{\mathcal {X}}_n^U)\) is a straightforward application of the Frobenius method. Combining the result of this calculation with the computation of the local monodromies of the local systems \(\mathbb {V}^+_n\) performed in Sect. 2.4, we conclude that \(\mathcal {T}(\hat{\mathcal {X}}_n^U)\) and \(\mathbb {V}_n^+\) have the same monodromy profiles for all \(n \in \{5,6,7,8,9,11\}\).
Definition 3.2
Let \(\mathbb {V}\) be a local system. The monodromy profile of \(\mathbb {V}\) is the unordered list of the orders of the local monodromy matrices of \(\mathbb {V}\) (excluding, of course, points where this monodromy has order 1).
Moreover, we know from Proposition 2.6 that \(\mathcal {T}(\hat{\mathcal {X}}_n^U)\) is the pullback of \(\mathbb {V}_n^+\) under the generalized functional invariant map g of \(\hat{\mathcal {X}}_n^U\). Proposition 3.1 will therefore follow if we can prove that \(g^*\mathbb {V}^+_n\) can only have the same monodromy profile as \(\mathbb {V}_n^+\) (for \(n \in \{5,6,7,8,9,11\}\)) if g is an isomorphism.
We will establish this fact through a pair of lemmas. The first lemma deals with the cases \(n \in \{6,8,9\}\).
Lemma 3.3
Let \(\mathbb {V}\) be a local system on a Zariski open subset of \(\mathbb {P}^1\), which has at least two local monodromy matrices of infinite order and at least one other nontrivial local monodromy matrix. Let \(g :\mathbb {P}^1 \rightarrow \mathbb {P}^1\) be a morphism. Then \({g}^* \mathbb {V}\) has the same monodromy profile as \(\mathbb {V}\) if and only if g is an isomorphism.
Proof
This lemma does not apply when we have only one cusp, a situation which occurs whenever n is prime [14, Sect. 3.8]. We therefore need a second lemma to deal with the remaining cases, where \(n \in \{5,7,11\}\).
Lemma 3.4
Let \(\mathbb {V}\) be a local system on a Zariski open subset of \(\mathbb {P}^1\), such that one of the local monodromy matrices of \(\mathbb {V}\) is of infinite order and there are at least three points in \(\mathbb {P}^1\) at which the local monodromy matrices of \(\mathbb {V}\) have finite order \(\ge 2\). Let \(g :\mathbb {P}^1 \rightarrow \mathbb {P}^1\) be a morphism. Then \({g}^* \mathbb {V}\) has the same monodromy profile as \(\mathbb {V}\) if and only if g is an isomorphism.
Proof
This completes the proof of Proposition 3.1.
3.2 Singular fibres of \(\mathcal {X}_n\)
We wish to see which of the generalized functional invariant maps allowed by the restrictions of Sects. 2.5 and 2.6 can actually underlie nonrigid Calabi–Yau threefolds fibred by \(M_n\)polarized K3 surfaces. As such threefolds are birational to the pullbacks of the fundamental families \(\mathcal {X}_n\) under the generalized functional invariant map (by Theorem 2.2), we begin by performing a detailed study of these fundamental families and their pullbacks under various covers.
Of particular interest is the behaviour of these families, and their covers, in neighbourhoods of the orbifold points in \(\mathcal {M}_{M_n}\). The fibres over such points will, in general, be singular. We will examine each of these singular fibres in turn, beginning with the simplest kind: those lying over the 2orbifold points \(\lambda _i\).
Proposition 3.5
Let \(\Delta \subset \mathcal {M}_{M_n}\) denote a small disc around one of the 2orbifold points \(\lambda = \lambda _i\) listed in Table 4. If \(n \ne 7\), the threefold \(\mathcal {X}_n\) is smooth over \(\Delta \) and its singular fibre over \(\lambda = \lambda _i\) is a singular K3 surface containing an \(A_1\) singularity. If we pullback \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) by a \(\mu \)fold cover \(\Delta ' \rightarrow \Delta \) ramified totally over \(\lambda = \lambda _i\), the total space of the pulledback threefold will contain an isolated \(cA_{\mu 1}\) singularity.
In the case \(n = 7\), the fibre over \(\lambda = 27\) behaves as above. However, the fibre over \(\lambda = 1\) contains an \(A_2\) singularity and the threefold \(\mathcal {X}_n\) contains an isolated node (\(cA_1\)) singularity over this point. If we pullback \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) by a \(\mu \)fold cover \(\Delta ' \rightarrow \Delta \) ramified totally over \(\lambda = 1\), the total space of the pulledback threefold will contain an isolated \(cA_{2\mu 1}\) singularity.
Proof
The only difficult case here is when \(n = 7\) and \(\lambda = 1\). In this case, by Table 3, \(\bar{\mathcal {X}}_7\) contains six curves of \(cA_m\) (\(m=1,2,3,4\)) singularities over \(\Delta \). When \(\lambda = 1\), the singularity along one of the \(cA_2\) curves (where \((x,y,z,t) = (0,1,1,0)\)) jumps to a \(cA_3\). After blowingup these six curves of singularities, we are thus left with an isolated \(cA_1\) singularity over \(\lambda = 1\) in \(\mathcal {X}_7\). The fibre of \(\mathcal {X}_7\) over \(\lambda = 1\) is a singular K3, containing an \(A_2\) singularity. Given this, the statement about the covers is easy. \(\square \)
Remark 3.6
In the case \(n=7\), the threefold \(\mathcal {X}_n\) contains an isolated \(cA_1\) singularity. Working analytically, it is always possible to find a small crepant resolution of this singularity, to obtain a smooth threefold with singular fibre a nodal K3 surface as in the other cases. However, it may not be possible to perform such a resolution algebraically.
The next type of singular fibre that we turn our attention to is the fibres over the cusp \(\lambda = \infty \). As we shall see, these fibres all admit a simple common description.
Proposition 3.7
Let \(\Delta \subset \mathcal {M}_{M_n}\) denote a small disc around the cusp \(\lambda = \infty \). The threefold \(\mathcal {X}_n\) is smooth over \(\Delta \) and its singular fibre over \(\lambda = \infty \) is a semistable K3 surface of Type III, containing \(n+2\) components. If we pullback \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) by a \(\mu \)fold cover \(\Delta ' \rightarrow \Delta \) ramified totally over \(\lambda = \infty \) and crepantly resolve any resulting singularities, the resulting threefold is again smooth and contains a semistable fibre of Type III with \(n\mu ^2 + 2\) components.
Proof
Once we have proved the form of the fibres over \(\Delta \), the form of the fibres over the \(\mu \)fold cover \(\Delta ' \rightarrow \Delta \) follows immediately from [15, Proposition 1.2]. We also note that the number of components appearing in the fibre of \(\mathcal {X}_n\) over \(\lambda = \infty \) has previously been computed by Przyjalkowski in [27, Corollary 35], but he does not precisely describe these fibres; our computations corroborate his result.
 (\({M}_2\))

Follows from the proof of [9, Proposition 2.5].
 (\(M_3\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_3\) contains nine curves of \(cA_m\) (\(m=1,2\)) singularities, which form sections of the fibration. Away from these nine curves, \(\bar{\mathcal {X}}_3\) is nonsingular over \(\Delta \). After performing m crepant blowups on each of these curves, we are left with nine isolated singularities of type \(cA_1\) lying over \(\lambda = \infty \). These may be resolved by a small projective resolution, after which the threefold \(\mathcal {X}_3\) becomes smooth over \(\Delta \), with a semistable singular fibre of Type III consisting of five components arranged as a triangular prism.
 (\(M_4\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_4\) contains twelve curves of \(cA_1\) singularities, which form sections of the fibration. Away from these twelve curves, \(\bar{\mathcal {X}}_4\) is nonsingular over \(\Delta \). After performing a crepant blowup on each of these curves, we are left with twelve isolated \(cA_1\) singularities over \(\lambda = \infty \). These admit a small projective resolution, after which the threefold \(\mathcal {X}_4\) is smooth over \(\Delta \), with a semistable singular fibre of Type III consisting of six components arranged in a cube.
 (\(M_5\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_5\) contains three curves of \(cA_1\) singularities, which we denote \(C^{A_1}_i\) (\(i = 1,2,3\)), and three curves of \(cD_4\) singularities, which we denote \(C^{D_4}_j\) (\(j=1,2,3\)), all of which form sections of the fibration. The fibre of \(\bar{\mathcal {X}}_5\) over \(\lambda = \infty \) has four components, given by the four coordinate hyperplanes in \(\mathbb {P}^3\), arranged as a tetrahedron, and the three curves \(C^{D_4}_j\) pass through three of the vertices. Moreover, the singularities along the curves \(C^{D_4}_j\) jump to canonical threefold singularities when \(\lambda = \infty \).
These singularities may be crepantly resolved as follows. First perform a single embedded blowup along each of the curves \(C^{D_4}_j\). Each of these blowups is crepant and gives rise to two exceptional divisors: one lying over the curve \(C^{D_4}_j\), and one lying over the point of intersection between \(C^{D_4}_j\) and the fibre \(\lambda = \infty \). After these blowups, we are left with twelve curves of generically \(cA_1\) singularities: the three \(C^{A_1}_i\) and three coming from each \(C^{D_4}_j\). These may each be crepantly resolved by a single blowup, leaving twelve isolated \(cA_1\) singularities, which may be resolved by small projective blowups.
After doing this, the threefold \(\mathcal {X}_5\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = \infty \) has seven components: the strict transforms of the original four and three additional components coming from the vertices that lie on the \(C^{D_4}_j\) (\(j=1,2,3\)). This fibre is semistable of Type III and has seven components arranged as a tetrahedron with three truncated vertices.
 (\(M_6\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_6\) contains three curves of \(cA_1\) singularities, denoted \(C^{A_1}_1, C^{A_1}_2, C^{A_1}_3\), two curves of \(cA_2\)’s, denoted \(C^{A_2}_1, C^{A_2}_1\), and two curves of \(cA_3\)’s, denoted \(C^{A_3}_1,C^{A_3}_2\), which form sections of the fibration. There are also two further curves of \(cA_1\) singularities lying in the fibre over \(\lambda = \infty \), given by \(D_1 := \{y = z = 0,\ \lambda = \infty \}\) and \(D_2 := \{z = t = 0,\ \lambda = \infty \}\).
The fibre of \(\bar{\mathcal {X}}_6\) over \(\lambda = \infty \) has four components arranged as a tetrahedron. The curves \(C^{A_3}_1\) and \(D_2\) both pass through the point \(x=z=t=0\) (which is a vertex of the tetrahedron), and the curves \(C^{A_2}_1\), \(D_1\) and \(D_2\) all pass through the vertex \(y=z=t=0\). Both of these vertices are canonical threefold singularities.
These singularities may be crepantly resolved by first performing m blowups of each of the curves \(C^{A_m}_i\), then blowing up the curves \(D_j\) once each, and finally performing small projective resolutions on any isolated terminal singularities that remain. After doing this, the threefold \(\mathcal {X}_6\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = \infty \) is semistable of Type III with eight components: the original four, plus one each from the curves \(D_j\) (\(j=1,2\)) and one each from the vertices \(x=z=t=0\) and \(y=z=t=0\).
 (\(M_7\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_7\) contains one curve of \(cA_1\) singularities, denoted \(C^{A_1}_1\), three curves of \(cA_2\)’s, denoted \(C^{A_2}_1,C^{A_2}_2\), one curve of \(cA_3\)’s, denoted \(C^{A_3}_1\), and one curve of \(cA_4\)’s, denoted \(C^{A_4}_1\), all of which form sections of the fibration.
To simplify the computation, first introduce a new variable \(u := x + y + z + t\), then change coordinates \(u \mapsto \lambda u\); this is an isomorphism on a general fibre and performs a birational modification to the fibre over \(\lambda = \infty \), but this birational modification does not affect the number of components appearing in a crepant resolution. The resulting family is given by$$\begin{aligned} \lambda u^2yz + u(y+z+t)(z+t)^2  (\lambda u  y  z t)yzt = 0. \end{aligned}$$(7)The fibre of this new family over \(\lambda = \infty \) consists of four components arranged as a tetrahedron. In addition to the \(C^{A_m}_i\), there are three curves of \(cA_1\) singularities lying in this fibre, given by \(D_1:=\{u=z=0\}\), \(D_2 := \{u=t=0\}\), and \(D_3 := \{u=y=0\}\). The curves \(C^{A_4}_1\), \(D_1\) and \(D_2\) all pass through the vertex \(u=z=t=0\), which is a canonical threefold singularity.
These singularities may be crepantly resolved as follows. First perform m blowups along each of the curves \(C^{A_m}_j\), then blow up each of the \(D_j\) once. Finally, perform small projective resolutions on the isolated terminal singularities that remain. After this process the threefold \(\mathcal {X}_7\) becomes smooth over \(\Delta \). Its fibre over \(\lambda = \infty \) is semistable of Type III with nine components: the strict transforms of the original four, one component from each of the three \(D_j\), and two additional components from the point \(u=z=t=0\).
 (\(M_8\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_8\) contains six curves of \(cA_1\) singularities, denoted \(C^{A_1}_i\) (\(i \in \{1,\ldots ,6\}\)), and three curves of \(cA_2\)’s, denoted \(C^{A_2}_i\) (\(i \in \{1,2,3\}\)), which form sections of the fibration. There are also three additional curves \(D_j\) (\(j=1,2,3\)) of \(cA_1\) singularities lying in the fibre \(\lambda = \infty \), given by \(\{x = t = 0\}\), \(\{y=t=0\}\) and \(\{z=t=0\}\) respectively.
The fibre of \(\bar{\mathcal {X}}_8\) over \(\lambda = \infty \) has four components, arranged as a tetrahedron, and the three \(C^{A_2}_i\) pass through the three vertices \(x=y=t=0\), \(x=z=t=0\) and \(x=y=t=0\), which are all canonical singularities.
These singularities admit a crepant resolution as follows. First perform m blowups of each of the curves \(C^{A_m}_i\), then blow up each of the curves \(D_j\) once. Finally, perform small projective resolutions on the isolated terminal singularities that remain. After doing this, the threefold \(\mathcal {X}_8\) becomes smooth over \(\Delta \). Its fibre over \(\lambda = \infty \) is semistable of Type III with ten components: the strict transforms of the original four, one component from each of the three \(D_i\), and one component from each of the three vertices \(C^{A_2}_i \cap D_i \cap D_j\).
 (\(M_9\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_9\) contains three curves of \(cA_1\) singularities, denoted \(C^{A_1}_i\) (\(i \in \{1,2,3\}\)), three curves of \(cA_2\) singularities, denoted \(C^{A_2}_i\) (\(i \in \{1,2,3\}\)), and one curve of \(cD_4\) singularities, denoted \(C^{D_4}\), all of which form sections of the fibration. There are also three additional curves \(D_j\) (\(j=1,2,3\)) of \(cA_1\) singularities in the fibre over \(\lambda = \infty \), where \(x = t = 0\), \(y=t=0\) and \(z=t=0\) respectively.
The fibre of \(\bar{\mathcal {X}}_8\) over \(\lambda = \infty \) has four components, arranged as a tetrahedron. The three \(C^{A_2}_i\) pass through the three vertices \(x=y=t=0\), \(x=z=t=0\) and \(x=y=t=0\), and \(C^{D_4}\) passes through the fourth vertex \(x=y=z=0\).
These singularities admit a crepant resolution as follows. First perform m blowups of each of the curves \(C^{A_m}_i\), then blow up the curves \(D_j\) once each. Finally, perform an embedded blowup along the curve \(C^{D_4}\). This gives rise to an additional component over the vertex \(\lambda =\infty \), \(x=y=z=0\) and three disjoint curves of \(cA_1\) singularities. These three curves may then be resolved by one further blowup each. Finally, perform small projective resolutions on the isolated terminal singularities that remain. After this, the threefold \(\mathcal {X}_9\) becomes smooth over \(\Delta \). Its fibre over \(\lambda = \infty \) is semistable of Type III with eleven components: the strict transforms of the original four, one component from each of the three \(D_j\), and one component from each of the four vertices of the tetrahedron.
 (\(M_{11}\))

By Table 3, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_{11}\) contains two curves of \(cA_1\) singularities, denoted \(C^{A_1}_1,C^{A_1}_2\), one curve of \(cA_2\) singularities, denoted \(C^{A_2}_1\), and two curves of \(cA_3\) singularities, denoted \(C^{A_3}_1,C^{A_3}_2\), all of which form sections of the fibration. There are also four additional curves \(D_j\) (\(j=1,2,3,4\)) of \(cA_1\) singularities where \(\lambda = \infty \) and \(x = t = 0\), \(y=t=0\), \(x=z=0\), and \(y=z=0\) respectively.
The fibre of \(\bar{\mathcal {X}}_{11}\) over \(\lambda = \infty \) has four components, arranged as a tetrahedron. The curve \(C^{A_2}_1\) passes through the vertex \(x=y=t=0\), which is a canonical singularity, and the two curves \(C^{A_3}_i\) (\(i = 1,2\)) pass through the vertices \(x=z=t=0\) and \(y=z=t=0\), each of which is a canonical singularity.
These singularities may be crepantly resolved as follows. First perform m blowups along each of the curves \(C^{A_m}_i\). After doing this, resolve the curves \(D_j\) by blowing up once each. Finally, perform small projective resolutions on any isolated terminal singularities that remain. After this, the threefold \(\mathcal {X}_{11}\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = \infty \) is semistable of Type III with thirteen components: the strict transforms of the original four, four additional components coming from the four \(D_i\), one component from the vertex \(x=y=t=0\) (where \(C^{A_2}_1\), \(D_1\) and \(D_2\) meet), and two components from each of the two vertices \(x=z=t = 0\) (where \(C^{A_3}_1\), \(D_1\) and \(D_3\) meet) and \(y = z= t = 0\) (where \(C^{A_3}_2\), \(D_2\) and \(D_4\) meet).
This completes the proof of Proposition 3.7.\(\square \)
The last, and most complex, singular fibres lie over the orbifold point \(\lambda = 0\). In this case, Theorem 2.17 heavily restricts the degrees of the possible covers that we need to consider.
Proposition 3.8
Let \(\Delta \subset \mathcal {M}_{M_n}\) denote a small disc around the point \(\lambda = 0\). The number of components of the singular fibre of \(\mathcal {X}_n\) over \(\lambda = 0\) is given by the \(\mu =1\) column of Table 5.
After pullingback \(\mathcal {X}_n \rightarrow \mathcal {M}_{M_n}\) by a \(\mu \)fold cover \(\Delta ' \rightarrow \Delta \) ramified totally over \(\lambda = 0\), we may perform a birational modification (that only affects the fibre over \(\lambda = 0\)) so that the relative canonical bundle of the resulting threefold over \(\Delta '\) is trivial and, unless marked otherwise (see below), the resulting threefold is smooth. In this case, the number of components of the fibre over \(\lambda = 0\) is given by the column corresponding to \(\mu \) in Table 5.

in cases marked with an asterisk (\(^*\)), the corresponding fibre is semistable of Type III or, if there is only one component, is a smooth K3 surface; and

there are several possibilities for fibres having only one component. A fibre denoted \(1^*\) (resp. 1) is a smooth (resp. nodal) K3 surface in a smooth total space. A fibre denoted \(1^{*\dagger }\) (resp. \(1^{\dagger }\)) is a singular K3 surface in a singular total space, that admits a small analytic resolution to a smooth (resp. nodal) K3 surface in a smooth total space. Note that in all cases marked with \(^{\dagger }\) it may not be possible to perform such resolutions algebraically, see Remark 3.6.
Number of components in the fibre over \(\lambda = 0\) in \(\mathcal {X}_n\) and its ramified covers
n  Degree \(\mu \) of ramified cover  

1  2  3  4  5  6  7  8  
2  31  11  4  \(1^*\)  31  11  4  \(1^*\) 
3  21  6  1  21  6  \(1^{*\dagger }\)  −  − 
4  15  \(3^*\)  18  \(6^*\)  −  −  −  − 
5  11  \(1^*\)  11  \(1^*\)  −  −  −  − 
6  \(8^*\)  \(26^*\)  −  −  −  −  −  − 
7  6  15  \(1^{*\dagger }\)  −  −  −  −  − 
8  \(4^*\)  \(10^*\)  −  −  −  −  −  − 
9  \(3^*\)  \(6^*\)  −  −  −  −  −  − 
11  \(1^{\dagger }\)  \(1^{*\dagger }\)  −  −  −  −  −  − 
Proof
 (\(M_2\))

The cases \(\mu = 1,2,4,8\) were computed in the proof of [9, Proposition 2.5].
In the case \(\mu = 3\), after pullingback \(\bar{\mathcal {X}}_2\) by a triple cover the threefold becomes nonnormal along the locus \(\lambda = w = 0\). Blow this locus up in the fourfold ambient space. After blowing up, the strict transform of our threefold is smooth away from the six curves of \(cA_3\) singularities which form sections of the fibration, and its fibre over \(\lambda = 0\) has five components: the strict transform of the original along with four new components. Of these, the component which is a strict transform of the original fibre is exceptional and may be contracted, leaving a threefold with trivial relative canonical bundle. The singular fibre of this threefold over \(\lambda = 0\) has four components, arranged as a cone over four lines in \(\mathbb {P}^2\).
In the cases \(\mu = 5,6,7\), after making a change of coordinates \(t \mapsto \lambda t\) and normalizing, we obtain a singular fibre with the same configuration of singularities as in the cases \(\mu = 1,2,3\), respectively. These may be resolved in the same way, to give singular fibres with respectively 31, 11 and 4 components.
 (\(M_3\))

In \(\bar{\mathcal {X}}_3\), the fibre \(\lambda = 0\) has two components: a component \(S := \{\lambda = s = 0\} \cong \mathbb {P}^2\), which has multiplicity 2, and a component \(Z:= \{\lambda = z = 0\} \cong \mathbb {P}^1 \times \mathbb {P}^1\), which has multiplicity 3. Moreover, over \(\Delta \), the threefold \(\bar{\mathcal {X}}_3\) has 17 curves of singularities: along with the six curves \(C^{A_1}_i\) (\(i \in \{1,\ldots ,6\}\)) of \(cA_1\) and three curves \(C^{A_2}_i\) (\(i\in \{1,2,3\}\)) of \(cA_2\) singularities coming from Table 3, which form sections of the fibration, there are also three curves \(D_i\) of \(cA_1\) singularities in the component S and five curves \(E_i\) of \(cA_2\) singularities in the component Z. The method of [15, Sect. 1] may be used to simultaneously resolve all curves of \(cA_1\)’s, followed by all curves of \(cA_2\)’s, giving rise to a singular fibre with 21 components: the strict transforms of S and Z, one exceptional component from each of the three \(D_i\), two exceptional components from each of the five \(E_i\), and one exceptional component from each of the six points \(C^{A_2}_i \cap E_j \cap E_k\).
In the case \(\mu =2\), suppose that we pullback \(\bar{\mathcal {X}}_3\) by a double cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is nonnormal along its central fibre. After performing a change of coordinates \(s \mapsto \lambda s\), which has the effect of normalizing the pullback of S and contracting the pullback of Z, we obtain a threefold with irreducible central fibre that contains ten curves of singularities over \(\Delta '\): the pullbacks of the nine \(C^{A_m}_i\) (note that the six \(C^{A_2}_i\)’s are no longer disjoint, but intersect in three pairs) and the curve \(D := \{\lambda = r = z = 0\}\), which is a curve of \(cA_2\) singularities. Once again, the configuration of curves of \(cA_2\) singularities may be simultaneously resolved by the method of [15, Sect. 1], leaving three disjoint curves \(C^{A_1}_i\) of \(cA_1\)’s, which may be resolved by one blowup each. The result is a singular fibre with six components: the strict transform of S, two from the curve D, and one each from the three points \(C^{A_2}_i \cap C^{A_2}_j \cap D\).
In the case \(\mu =3\), suppose that we pullback \(\bar{\mathcal {X}_3}\) by a triple cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is nonnormal along its central fibre. After performing a change of coordinates \(z \mapsto \lambda z\), which has the effect of normalizing the pullback of Z and contracting the pullback of S, we find that the resulting threefold is smooth away from the nine curves \(C^{A_j}_i\) (note that the three \(C^{A_1}_i\)’s are no longer disjoint, and all intersect at the point \(\lambda = s = x =y = 0\)). Blowup the six curves \(C^{A_2}_i\) twice each, then simultaneously resolve the three curves \(C^{A_1}_i\) by the method of [15, Sect. 1] to obtain a smooth threefold whose singular fibre over \(\lambda = 0\) is a nodal K3.
Given this it is easy to see that, in the case \(\mu =6\), the pullback of \(\mathcal {X}_3\) by a 6fold cover \(\Delta ' \rightarrow \Delta \) is a singular threefold, with an isolated node (\(cA_1\)) singularity.
In the case \(\mu =4\), suppose that we pullback \(\mathcal {X}_3\) by a 4fold cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). After making a change of coordinates \(z \mapsto \lambda z\), we obtain a central fibre containing three curves of \(cA_1\) singularities and five curves of \(cA_2\) singularities, arranged in the same configuration as the unramified case. This may be crepantly resolved in the same way as the unramified case, to give a fibre with 21 components.
Finally, in the case \(\mu =5\), suppose that we pullback \(\bar{\mathcal {X}}_3\) by a 5fold cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). After performing a change of coordinates \(s \mapsto \lambda s\), \(z \mapsto \lambda z\), we obtain a threefold with irreducible central fibre that contains ten curves of singularities over \(\Delta '\): the pullbacks of the nine \(C^{A_m}_i\) (as in the double covering case, the six \(C^{A_2}_i\)’s are no longer disjoint, but intersect in three pairs, and the three \(C^{A_1}_i\)’s intersect transversely in a triple point) and the curve \(D := \{\lambda = r = z = 0\}\), which is a curve of \(cA_2\) singularities. As in the double covering case, the configuration of curves of \(cA_2\) singularities may be simultaneously resolved by the method of [15, Sect. 1], leaving the three curves \(C^{A_1}_i\) of \(cA_1\)’s, which may be simultaneously resolved by the same method. The result is a singular fibre with six components: the original component, two from the curve D, and one each from the three points \(C^{A_2}_i \cap C^{A_2}_j \cap D\).
 (\(M_4\))

In \(\bar{\mathcal {X}}_4\) the fibre over \(\lambda = 0\) has three components \(S_1,S_2,S_3\), given by \(S_i := \{\lambda = s_i = 0\} \cong \mathbb {P}^1 \times \mathbb {P}^1\), each of which has multiplicity 2. Moreover, over \(\Delta \) the threefold \(\bar{\mathcal {X}}_4\) has 24 curves of singularities: along with the twelve curves \(C_i\) (\(i = 1,\ldots ,12\)) of \(cA_1\) singularities from Table 3, which form sections of the fibration, there are also four curves \(D_{ij}\) (\(i \in \{1,2,3\}\), \(j \in \{1,\ldots ,4\}\)) of \(cA_1\) singularities in each component \(S_i\), corresponding to the intersections of \(S_i\) with the hyperplanes \(\{r_k = 0\}\) and \(\{r_k = s_k\}\) for \(k \ne i\). This configuration can be resolved by the method of [15, Sect. 1], giving rise to a singular fibre with 15 components: the strict transforms of the three \(S_i\) along with one exceptional component from each of the twelve \(D_{ij}\).
In the case \(\mu =2\), suppose that we pullback \(\bar{\mathcal {X}}_4\) by a double cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is nonnormal. The normalization is singular only along the pullbacks of the curves \(C_i\) and its central fibre consists of three components, each of which is a double cover of one of the \(S_i\) ramified over the four curves \(D_{ij}\) (each component thus obtained contains four \(A_1\) singularities, corresponding to the intersections \(D_{ij} \cap D_{ik}\). These coincide with the intersection of \(S_i\) with the curves \(C_l\), which are blown up in the next step). After blowing up the \(C_i\), we obtain a smooth threefold whose central fibre is semistable of Type III, with three components meeting along three double curves and two triple points.
Given this, the form of the fibre over the pullback by a further double cover (corresponding to \(\mu =4\)) then follows easily from [15, Proposition 1.2].
Finally, in the case \(\mu =3\), suppose that we pullback \(\bar{\mathcal {X}}_4\) by a triple cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is again nonnormal. After normalizing, the resulting fibre has \(cA_1\) singularities along the twelve \(C_i\), the twelve \(D_{ij}\), and along the three intersections \(S_i \cap S_j\). Crepantly resolving these singularities using the method of [15, Sect. 1], we obtain a central fibre with 18 components: the original three, plus one from each of the twelve curves \(D_{ij}\) and one from each of the three curves \(S_i \cap S_j\).
 (\(M_5\))

The fibre of \(\mathcal {X}_5\) over \(\lambda = 0\) has been previously studied by Iliev, Katzarkov and Przyjalkowski [17, Sect. 5.2]. They show that after a number of crepant blowups the threefold \(\mathcal {X}_5\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = 0\) has 11 components.
In the case \(\mu =2\), suppose that we pullback \(\bar{\mathcal {X}}_5\) by a double cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is nonnormal. After normalizing, the only singularities remaining are the six curves of \(cA_1\) and \(cD_4\) singularities coming from Table 3. After blowing up these curves, the resulting threefold is smooth over \(\Delta '\) with central fibre a smooth K3 surface. A further double cover preserves this smoothness, giving the case \(\mu =4\).
Finally, in the case \(\mu =3\), suppose we pullback \(\bar{\mathcal {X}}_5\) by a triple cover \(\Delta ' \rightarrow \Delta \) ramified over \(\lambda = 0\). The resulting threefold is nonnormal along its central fibre \(\lambda = 0\). We may normalize by blowing up this fibre in the fourfold total space. The resulting threefold is normal and its fibre over \(\lambda = 0\) contains seven curves of \(cA_1\) singularities, arranged in the same configuration as the unramified case. This may be crepantly resolved by the method of [17, Sect. 5.2], to give a fibre with 11 components.
 (\(M_6\))
 Note there is a natural involution of the family \(\bar{\mathcal {X}}_6\), given byThis involution exchanges the fibres over \(\lambda = 0\) and \(\lambda = \infty \), which must therefore be isomorphic. Given this, the properties for the fibre over \(\lambda = 0\) follow immediately from the properties of the fibre over \(\lambda = \infty \), as computed in Proposition 3.7.$$\begin{aligned} (x,y,z,t,\lambda ) \longmapsto \left( y+z,zt,x+z+t,xyzt,\tfrac{1}{\lambda }\right) .\end{aligned}$$
 (\(M_7\))
 The fibre of \(\mathcal {X}_7\) over \(\lambda = 0\) is the most difficult. We once again use the simplified form (7), but with an additional variable substitution \(w := z+t\). The equation for the family becomesThe fibre of this family over \(\lambda = 0\) has two components, given by \(\{y+w = 0\}\) and \(\{(uw^2+yz(wz) = 0\}\), which meet along the two curves \(D_1 := \{y=w=0\}\) and \(D_2 := \{y+w = uwz(wz) = 0\}\). Over \(\Delta \) the threefold defined by this equation contains eight curves of singularities: along with the six curves \(C^{A_m}_i\) (\(m \in \{1,\ldots ,4\}\), \(i \in \{1,2,3\}\)) of \(cA_m\) singularities coming from Table 3, which form sections of the fibration, there is an additional curve of \(cA_2\) singularities along \(D_1\) and a curve \(D_3 := \{ w=z=0\}\) of \(cA_1\) singularities.$$\begin{aligned} \lambda u^2yz + uw^2(y + w)  yz(\lambda u  y  w)(wz) = 0. \end{aligned}$$
This fibre admits a crepant resolution as follows. First blow up the curve \(D_1\) twice and the curve \(D_3\) once. After this, perform m blowups of each of the curves \(C^{A_m}_i\). Finally, any remaining isolated terminal singularities may be resolved by small projective resolutions. After this process, \(\mathcal {X}_7\) becomes smooth over \(\Delta \), and its fibre over \(\lambda = 0\) has six components: the strict transforms of the original two, two from \(D_1\), one from \(D_3\), and one from the point \(y=z=w=0\) where \(D_1\), \(D_3\) and \(C^{A_2}_1\) meet.
In the case \(\mu =2\), suppose that we pullback by a double cover. The fibre described above contains six components meeting along nine double curves. Under the cover, these double curves become curves of \(cA_1\) singularities, which may be resolved by blowing up once each. The resulting fibre has 15 components.
Finally, in the case \(\mu =3\), suppose that we pullback by a triple cover ramified over \(\lambda = 0\). Perform another change of coordinates \(u \mapsto \frac{u}{\lambda }\), \(w \mapsto \lambda w\) and \(y \mapsto \lambda y\). The resulting family isThis family has an irreducible fibre over \(\lambda = 0\) and the threefold total space is smooth away from the six curves \(C^{A_m}_i\), which form sections of the fibration. However, there is some isolated singularity behaviour where the curves \(C^{A_m}_i\) meet the fibre \(\lambda = 0\): the curves \(C_1^{A_1}\) and \(C_1^{A_2}\) collide in a \(cA_4\) singularity, the curves \(C_2^{A_2}\) and \(C_1^{A_3}\) collide in a \(cA_6\) singularity, and the singularity along the curve \(C_1^{A_4}\) jumps to a \(cD_5\). After blowing up these three singularities, we are left with three isolated nodes (\(cA_1\)’s) in the threefold total space, corresponding to the three points where the isolated singularity behaviour occurs. The fibre over \(\lambda = 0\) in this threefold is a singular K3 surface with three \(A_1\) singularities, one for each threefold node.$$\begin{aligned} u^2yz + uw^2(y+z)  yz(\lambda u  y  w)(\lambda w  z) = 0. \end{aligned}$$  (\(M_8\))

The fibre of \(\mathcal {X}_8\) over \(\lambda = 0\) has been previously studied by Przyjalkowski [26, Example 29]. He shows that after a number of crepant blowups the threefold \(\mathcal {X}_8\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = 0\) is semistable with 4 components, arranged as a tetrahedron.
Given this, in the case \(\mu =2\) the form of the fibre over the double cover \(\Delta ' \rightarrow \Delta \) follows immediately from [15, Proposition 1.2].
 (\(M_9\))

The fibre of \(\mathcal {X}_9\) over \(\lambda = 0\) has also been previously studied by Przyjalkowski [26, Example 30], who shows that after a number of crepant blowups the threefold \(\mathcal {X}_9\) becomes smooth over \(\Delta \) and its fibre over \(\lambda = 0\) is semistable with 3 components.
As before, in the case \(\mu =2\), the form of the fibre over the double cover \(\Delta ' \rightarrow \Delta \) follows immediately from [15, Proposition 1.2].
 (\(M_{11}\))

The fibre of \(\bar{\mathcal {X}}_{11}\) over \(\lambda = 0\) is irreducible, and the only singularities of \(\bar{\mathcal {X}}_{11}\) over \(\Delta \) are the five curves of \(cA_m\) (\(m=1,2,3\)) singularities arising from Table 3 and one isolated node (\(cA_1\)) at the point \((x,y,z,t,\lambda ) = (1,1,1,0,0)\). After blowing up the five curves of \(cA_m\) singularities, \(\mathcal {X}_{11}\) is left with a single isolated node (\(cA_1\)). Its fibre over \(\lambda = 0\) is a singular K3 surface with an \(A_2\) singularity.
In the case \(\mu =2\), if we pullback \(\mathcal {X}_{11}\) by an double cover ramified over \(\lambda = 0\), the fibre over \(\lambda = 0\) is unchanged, but the resulting threefold will contain an isolated \(cA_{3}\) singularity over \(\lambda = 0\).
4 Constructing Calabi–Yau threefolds
The aim of this final section is to use the results of the previous sections to explicitly classify nonrigid Calabi–Yau threefolds fibred nonisotrivially by \(M_n\)polarized K3 surfaces, for \(n \ne 1\). By Theorem 2.2, we know that any such Calabi–Yau threefold is birational to the pullback \(\bar{\mathcal {X}}_{n,g}\) of one of the families \(\bar{\mathcal {X}}_n \rightarrow \mathcal {M}_{M_n}\) under the generalized functional invariant map \(g :\mathbb {P}^1 \rightarrow \mathcal {M}_{M_n}\), and Corollary 2.14 tells us that \(n \in \{2,3,4,5,6,7,8,9,11\}\).
We are thus reduced to classifying the possible generalized functional invariant maps that give rise to Calabi–Yau threefolds in each case. We may use Theorem 2.17 to significantly reduce the number of cases that we need to consider. We thus obtain the following theorem, which should be thought of as a converse to Theorem 2.17.
The generalized functional invariant maps g appearing in this theorem are described using the notation introduced in Sect. 2.6. Recall that d denotes the degree of g (note that \(d \ge 2\), by Theorem 2.17) and \([x_1,\ldots ,x_k]\), \([y_1,\ldots ,y_l]\), \([z_{1,1},\ldots ,z_{m_1,1}],\ldots ,[z_{1,1},\ldots ,z_{m_q,q}]\) denote the partitions of d that encode the ramification profiles of g over \(\lambda = \infty \) (the cusp), \(\lambda = 0\) (the aorbifold point) and \(\lambda = \lambda _1,\ldots ,\lambda _q\) (the 2orbifold points) respectively. Finally, r denotes the degree of ramification of g away from \(\lambda \in \{0,\infty ,\lambda _1,\ldots ,\lambda _q\}\).
Theorem 4.1

\(n \ne 7, 11\), and

g is unramified over \(\lambda \in \{\lambda _1,\ldots ,\lambda _q\}\), and

we are not in the case \(n = 3\), \(l=1\), \([y_1] = [6]\),
Proof
The singularities in \(\bar{\mathcal {X}}_{n,g}\) away from the fibres over the points in the preimage \(g^{1}(\{0,\infty ,\lambda _1,\ldots ,\lambda _q\})\) are given in Table 3; they all lie along smooth curves which form sections of the fibration, so may be crepantly resolved. In neighbourhoods of the fibres over \(g^{1}(\{0,\infty ,\lambda _1,\ldots ,\lambda _q\})\), crepant birational tranformations resolving the singularities in \(\bar{\mathcal {X}}_{n,g}\) are computed in Propositions 3.5, 3.7 and 3.8. Using these explicit birational transformations, it is a straightforward calculation to show that \(\mathcal {X}_{n,g}\) has trivial canonical sheaf. Finally, vanishing of \(H^1(\mathcal {X}_{n,g},\mathcal {O}_{\mathcal {X}_{n,g}})\) follows by the method used to prove [9, Proposition 2.7]. \(\square \)
Remark 4.2
This theorem should be thought of as a generalization of the results of [9, Sect. 2]. Note that, even in the \(n = 2\) case, the main result [9, Corollary 2.8] of that section is somewhat weaker than the result above, as it only classifies the cases where the threefolds \(\bar{\mathcal {X}}_{2,g}\) have Calabi–Yau resolutions. In contrast, the result above classifies all cases where \(\bar{\mathcal {X}}_{2,g}\) is birational to a Calabi–Yau threefold, which is a weaker condition. In particular, here we acquire additional cases \([y_1,\ldots ,y_l] = [1,3]\), [2, 3], [3, 3], [3, 4], [5], [6], and [7] (also see Remark 4.5).
Using the methods from [9], it is now a simple matter to compute the Hodge numbers of these threefolds.
Proposition 4.3
Proof
This is a straightforward computation using [9, Lemma 3.2].\(\square \)
Proposition 4.4

k and l denotes the number of ramification points of g over the points \(\lambda = \infty \) and \(\lambda = 0\) respectively,

\(m_{\mathrm {odd}}\) denotes the number of ramification points of odd order lying over the 2orbifold points in the set \(\{\lambda _1,\ldots ,\lambda _q\}\),

d is the degree of g,
 \(\delta \) is a correction term which equals 2 if \(n=2\) and \([y_1,\ldots ,y_l]\) is [3, 3] or [7]; equals 1 ifand equals 0 otherwise (see Remark 4.5).

\(n = 2\) and \([y_1,\ldots ,y_l]\) is one of [1, 3], [2, 3], [3, 4], [5], [6], [8], or

\(n = 3\) and \([y_1,\ldots ,y_l] = [5]\), or

\(n=5\) and \([y_1,\ldots ,y_l]\) is [3] or [4];

Proof
The proof of this proposition follows by the same method as was used to prove [9, Corollary 3.9]. We begin by setting up some notation: let \(\pi _{n,g}:\mathcal {X}_{n,g} \rightarrow B \cong \mathbb {P}^1\) denote the K3 fibration on \(\mathcal {X}_{n,g}\), let \(j:U \hookrightarrow B\) denote the open set over which \(\mathcal {X}_{n,g}\) is smooth and the fibres of \(\mathcal {X}_{n,g}\) are smooth K3 surfaces, and let \(\pi _{n,g}^U\) denote the restriction of \(\pi _{n,g}\) to U.

if g ramifies to order y at a preimage p of 0, then \(R(p) = 4  {\mathrm {hcf}}(y,2)\),

if g ramifies to order z at a preimage p of \(\lambda _1 = 64\), then \(R(p) = 2  {\mathrm {hcf}}(z,2)\), and

at any preimage p of \(\infty \), we have \(R(p) = 2\).
Remark 4.5

when \(n=2\), the cases \([y_1,\ldots ,y_l] = [a]\), for \(a \in \{5,6,7,8\}\), are smooth degenerations of the cases \([y_1,\ldots ,y_l] = [4,a4]\),

when \(n=3\), the case \([y_1,\ldots ,y_l] = [5]\) is a smooth degeneration of the case \([y_1,\ldots ,y_l] = [2,3]\), and

when \(n=5\), the cases \([y_1,\ldots ,y_l] = [a]\), for \(a \in \{3,4\}\), are smooth degenerations of the cases \([y_1,\ldots ,y_l] = [2,a2]\).
The values of \(h^{2,1}(\mathcal {X}_{n,g})\) must be invariant under these degenerations, and the correction term \(\delta \) ensures that this holds. In this way, \(\delta \) may be thought of as a correction for “nongenericity”. It corrects for the fact that, for certain choices of n, g as above, the Calabi–Yau threefold \(\mathcal {X}_{n,g}\) is not generic in its deformation class; it may be thought of as measuring the codimension of the family \(\mathcal {X}_{n,g}\) in moduli.
Note that this also means that, whilst [9, Corollary 2.8] did not capture all of the threefolds with \(n=2\) from Theorem 4.1 (see Remark 4.2), it did capture the generic member of every deformation class (i.e. all cases with \(\delta = 0\)).
Remark 4.6
This formula for \(h^{2,1}(\mathcal {X}_{n,g})\) can be simplified significantly if we make the genericity assumptions that \(\delta = 0\) (so \(\mathcal {X}_{n,g}\) is generic in its deformation class) and g is unramified over \(\lambda \in \{\lambda _1,\ldots ,\lambda _q\}\). In this case, we have \(m_{\mathrm {odd}} = m_1+\cdots + m_q = qd\), so it follows from Theorem 4.1 that \(h^{2,1}(\mathcal {X}_{n,g}) = r\), the degree of ramification away from \(\lambda \in \{0,\infty ,\lambda _1,\ldots ,\lambda _q\}\).
In this setting, an argument analogous to that used to prove [9, Proposition 4.1] shows that all small deformations of \(\mathcal {X}_{n,g}\) are induced by deformations of g that preserve the ramification profiles over \(\lambda \in \{0,\infty \}\). One may use this to compare the moduli spaces of the Calabi–Yau threefolds \(\mathcal {X}_{n,g}\) to the Hurwitz spaces of the maps g.
Remark 4.7
It is interesting to look at the mirror symmetric interpretation of these results. The philosophy of [11] states that a smooth Calabi–Yau \(\mathcal {X}_{n,g}\) with \(l=2\) (so there are 2 ramification points over \(\lambda = 0\)) should split into a pair of LGmodels, each corresponding to a quasiFano variety. These quasiFano varieties may then be glued to give a Tyurin degeneration, which smooths to give a Calabi–Yau variety mirror to \(\mathcal {X}_{n,g}\). Further evidence for this conjecture is given in [21].
[11] does not provide a mirror interpretation for the Calabi–Yau varieties \(\mathcal {X}_{n,g}\) with \(l=1\). The authors plan to address this in future work.
Notes
Acknowledgements
The computations of singularity types and explicit resolutions of singularities found in this paper were greatly assisted by the Singular computer algebra system [6].
References
 1.Bardelli, F.: On Grothendieck’s generalized Hodge conjecture for a family of threefolds with trivial canonical bundle. J. Reine Angew. Math. 422, 165–200 (1991)MathSciNetzbMATHGoogle Scholar
 2.Beukers, F., Heckman, G.: Monodromy for the hypergeometric function \(_nF_{n1}\). Invent. Math. 95(2), 325–354 (1989)MathSciNetCrossRefzbMATHGoogle Scholar
 3.Clingher, A., Doran, C.F.: Modular invariants for lattice polarized K3 surfaces. Michigan Math. J. 55(2), 355–393 (2007)MathSciNetCrossRefzbMATHGoogle Scholar
 4.del Angel, P.L., MüllerStach, S., van Straten, D., Zuo, K.: Hodge classes associated to \(1\)parameter families of Calabi–Yau \(3\)folds. Acta Math. Vietnam. 35(1), 7–22 (2010)MathSciNetzbMATHGoogle Scholar
 5.Deligne, P.: Théorie de Hodge I, Actes du Congrès International des Mathématiciens (Nice, 1970) Tome 1, GauthierVillars, Paris pp. 425–430 (1971)Google Scholar
 6.Decker, W., Greuel, G.M., Pfister, G., Schönemann, H.: Singular 402—a computer algebra system for polynomial computations. http://www.singular.unikl.de (2015)
 7.Doran, C.F., Harder, A., Katzarkov, L., Lewis, J., Przyjalkowski, V.: Moduli spaces of K3 surfaces and mirror symmetry for Fano threefolds (2016) (Manuscript in preparation) Google Scholar
 8.Doran, C.F., Harder, A., Novoseltsev, A. Y., Thompson, A.: Families of lattice polarized K3 surfaces with monodromy. Int. Math. Res. Notices 23, 12265–12318 (2015)Google Scholar
 9.Doran, C.F., Harder, A., Novoseltsev, A.Y., Thompson, A.: Calabi–Yau threefolds fibred by mirror quartic K3 surfaces. Adv. Math. 298, 369–392 (2016)MathSciNetCrossRefzbMATHGoogle Scholar
 10.Doran, C.F., Harder, A., Thompson, A.: Hodge numbers from Picard–Fuchs equations. SIGMA 13(045), 23 (2017)Google Scholar
 11.Doran, C.F., Harder, A., Thompson, A.: Mirror symmetry, Tyurin degenerations and fibrations on Calabi–Yau manifolds, StringMath 2015. In: Proc. Symp. Pure Math., vol. 96. American Mathematical Society pp. 93–131 (2017)Google Scholar
 12.Doran, C.F., Malmendier, A.: Calabi–Yau manifolds realizing symplectically rigid monodromy tuples, Preprint, (March 2015). arXiv:1503.07500 (2015)
 13.Dolgachev, I.V.: Mirror symmetry for lattice polarised K3 surfaces. J. Math. Sci. 81(3), 2599–2630 (1996)MathSciNetCrossRefzbMATHGoogle Scholar
 14.Diamond, F., Shurman, J.: A first course in modular forms. Graduate Texts in Mathematics, vol. 228. Springer, Berlin (2005)Google Scholar
 15.Friedman, R.: Base change, automorphisms, and stable reduction for type III K3 surfaces. In: Friedman, R., Morrison, D. (eds.) The Birational Geometry of Degenerations. Progr. Math., vol. 29, Birkhäuser, pp. 277–298 (1983)Google Scholar
 16.Golyshev, V.: Classification problems and mirror duality. In: Young, N. (ed.) Surveys in Geometry and Number Theory: Reports on Contemporary Russian Mathematics. London Math. Soc. Lecture Note Ser., vol. 338, Cambridge Univ. Press, pp. 88–121 (2007)Google Scholar
 17.Iliev, A., Katzarkov, L., Przyjalkowski, V.: Double solids, categories and nonrationality. Proc. Edinb. Math. Soc. (2) 57(1), 145–173 (2014)Google Scholar
 18.Iskovskih, V.A.: Fano 3folds. I, Math. USSR Izvestija 11(3), 485–527 (1977)Google Scholar
 19.Iskovskih, V.A.: Fano 3folds. II, Math. USSR Izvestija 12(3), 469–506 (1978)Google Scholar
 20.Kollár, J., Mori, S.: Birational geometry of algebraic varieties. Cambridge Tracts in Mathematics, vol. 134. Cambridge University Press, Cambridge (1998)Google Scholar
 21.Lee, N.H.: Mirror pairs of Calabi–Yau threefolds from mirror pairs of quasiFano threefolds. Preprint. arXiv:1708.02489 (2017)
 22.Levelt, A.H.M.: Hypergeometric functions, Ph.D. thesis, University of Amsterdam (1961)Google Scholar
 23.Nikulin, V.V.: Finite automorphism groups of Kähler \(K3\) surfaces. Trans. Moscow Math. Soc. 38(2), 71–135 (1980)zbMATHGoogle Scholar
 24.Nikulin, V.V.: Integral symmetric bilinear forms and some of their applications. Math. USSR Izv. 14(1), 103–167 (1980)CrossRefzbMATHGoogle Scholar
 25.Ogg, A.P.: Hyperelliptic modular curves. Bull. Soc. Math. France 102, 449–462 (1974)MathSciNetCrossRefzbMATHGoogle Scholar
 26.Przyjalkowski, V.: Weak Landau–Ginzburg models for smooth Fano threefolds. Izv. Math. 77(4), 772–794 (2013)MathSciNetCrossRefzbMATHGoogle Scholar
 27.Przyjalkowski, V.: Calabi–Yau compactifications of toric Landau–Ginzburg models for smooth Fano threefolds. Math. Sb. 208(7), 84–108 (2017)MathSciNetCrossRefzbMATHGoogle Scholar
 28.Shimura, G.: Sur les intégrales attachées aux formes automorphes. J. Math. Soc. Japan 11, 291–311 (1959)MathSciNetCrossRefzbMATHGoogle Scholar
 29.Saito, M.H., Zucker, S.: Classification of nonrigid families of K3 surfaces and a finiteness theorem of Arakelov type. Math. Ann. 289(1), 1–31 (1991)MathSciNetCrossRefzbMATHGoogle Scholar
 30.Voisin, C.: Hodge theory and complex algebraic geometry. I. Cambridge Studies in Advanced Mathematics, vol. 76. Cambridge University Press, Cambridge (2007)Google Scholar
 31.Zucker, S.: Hodge theory with degenerating coefficients: \(L_{2}\) cohomology in the Poincaré metric. Ann. Math. (2) 109(3), 415–476 (1979)Google Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.