This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Hyperelliptically fibred surfaces with nodes

E. Ballico, E. Gasparim, and B. Suzuki
Abstract.

Using elementary methods of algebraic geometry, we present constructions of hyperelliptically fibred surfaces containing nodal fibres.

1. Motivation

Hyperelliptically fibred surfaces appear often in string theory and when they contain singular fibres these provoke the existence of DD-branes. Elliptic fibrations are quite popular and well understood, while the precise role hyperelliptic fibrations might play in string theory remains to be understood. Recall that a curve CC of genus 2\geq 2 is said to be hyperelliptic if there exists a morphism C1C\rightarrow\mathbb{P}^{1} of degree 2. Examples of string theoretical work considering hyperelliptic fibrations are [30, 42]. We were asked by physicists to provide a summary of results on the existence of hyperelliptic fibrations with certain types of singularities, and this was the motivation to produce this note. Most results here were collected from the standard algebraic geometric literature, however we do provide original constructions in sections 2.3, 2.4, and 2.5 showing explicit examples of genus 2 families acquiring nodes. Nodes may occur in any of the following ways, as illustrated in Figure 1: irreducible of genus 2 with a single node, irreducible of genus 2 with more than 1 nodes, reducible.

Refer to caption
Figure 1. Types of nodes on genus 2 curves

In the review parts of sections 2 and 3 our goal was to collect properties of numerical invariants of fibred surfaces, with focus on inequalities involving bounds on Chern numbers. Physics properties dictate that positivity of Chern numbers will imply the existence of D-branes, but this theme will remain for another paper dedicated to applications to F-theory.

Reversing a construction of Eisenbud, in subsection 2.3 we find a nice family of all genus gg hyperelliptic curves together with also nodal ones in the same parameter space, for example such a family includes the irreducible curve with a prescribed g\leq g number of nodes and the union of an elliptic curve and a g1g-1 hyperelliptic curve.

Given that the motivation coming from string theory emphasises applications of genus 22 curves, in subsection 2.4 we present a second construction specifically tailored for the case of genus 22. Our second construction produces a large family such that a general element is a smooth curve of genus 22, it contains a 22 parameter family of integral nodal curves with arithmetic genus 22 and exactly one node, while it also contains a 1 parameter family of curves which are a nodal union of 22 elliptic curves with a unique common point. In subsection 2.5, we present some further details of the toric case. We observe also that an entire chapter about hyperelliptic curves will appear in the upcoming book by Eisenbud and Harris [14].

In section 3 we summarise some properties of surfaces fibred by smooth fibres. The most interesting case being that of the very interesting Kodaira surfaces which are not locally trivial. Therefore a Kodaira surface is not a fibre bundle, hence not all fibres being isomorphic. Furthermore, each fibre of a Kodaira surface is isomorphic to only finitely many fibres. Such wild local behaviour made it almost impossible for us to illustrate section 3 with figures in the same way as we did in section 2. A reader with better suggestions of how to draw a Kodaira surface is welcome to share their idea with us.

2. Fibrations containing singular fibres

Let XX be a smooth compact complex surface. The holomorphic Euler characteristic of XX is

χ(𝒪𝒳):=1𝒽1(𝒪𝒳)+𝓅(𝒳)=𝒽0(𝒪𝒳)𝒽1(𝒪𝒳)+𝒽2(𝒪𝒳),\chi(\cal{O}_{X})\mathrel{\mathop{:}}=1-h^{1}(\cal{O}_{X})+p_{g}(X)=h^{0}(\cal{O}_{X})-h^{1}(\cal{O}_{X})+h^{2}(\cal{O}_{X}),

while the topological Euler characteristic of XX is

e(X)=h0(TX)h1(TX)+h2(TX)h3(TX)+h4(TX)e(X)=h^{0}(TX)-h^{1}(TX)+h^{2}(TX)-h^{3}(TX)+h^{4}(TX)

for any smooth compact manifold of real dimension 4.

For a smooth compact complex surface XX (i.e. dimX=2\dim_{\mathbb{C}}X=2), Noether’s formula gives

12χ(𝒪𝒳)=𝒸12(𝒳)+𝒸2(𝒳)=(𝒦𝒦)+(𝒳)\boxed{12\chi(\cal{O}_{X})=c_{1}^{2}(X)+c_{2}(X)=(K\cdot K)+e(X)}

where KK is the canonical divisor class [6, Eq.(4)p.26][16, p.472]. In contrast, we also observe that if CC is a smooth compact curve (i.e. a Riemann surface) and hence the two numbers h0(ΩC1)h^{0}(\Omega^{1}_{C}) and h1(𝒪𝒞)h^{1}(\cal{O}_{C}) are the same, we have that the topological and holomorphic Euler characteristics satisfy 2χ(𝒪𝒞)=(𝒞).\boxed{2\chi(\cal{O}_{C})=e(C)}.

Remark 2.1.

Let XX^{\prime} be the surface obtained from XX blowing up one point. We have that the irregularities satisfy h1(𝒪𝒳)=𝒽1(𝒪𝒳)h^{1}(\cal{O}_{X^{\prime}})=h^{1}(\cal{O}_{X}), the plurigenera satisfy pg(X)=pg(X)p_{g}(X^{\prime})=p_{g}(X) and hence the holomorphic Euler characteristics satisfy χ(𝒪𝒳)=χ(𝒪𝒳)\chi(\cal{O}_{X^{\prime}})=\chi(\cal{O}_{X}). We have c12(X)=c12(X)1c_{1}^{2}(X^{\prime})=c_{1}^{2}(X)-1 and consequently c2(X)=c2(X)+1c_{2}(X^{\prime})=c_{2}(X)+1, see [6].

We recall the fundamental local-triviality theorem of Grauert–Fischer.

Theorem 2.2.

[6, p. 36] Let f:XYf\colon X\to Y be a smooth holomorphic family of compact complex manifolds. The holomorphic map ff is locally trivial (in the Euclidean topology) over YY if and only if all fibres of ff are biholomorphic.

Note that in Theorem 2.2 we do not assume that YY is compact, we only assume that ff is a smooth and proper holomorphic map. Let f:XDf\colon X\to D be a proper holomorphic map with XX a smooth and connected complex surface (often called a fibration of curves with DD as a base) and DD a (not necessarily compact) Riemann surface (even not compact algebraic, e.g. a disc of \mathbb{C}). The sheaf f(𝒪𝒳)f_{\ast}(\cal{O}_{X}) is a locally free 𝒪𝒟\cal{O}_{D}-sheaf, say of rank r1r\geq 1 with 𝒪𝒟\cal{O}_{D} as a direct factor and f=h1f1f=h_{1}\circ f_{1}, where D1D_{1} is a Riemann surface, h1:DDh_{1}\colon D^{\prime}\to D is a finite holomorphic map of degree rr, f1:XD1f_{1}\colon X\to D_{1} is proper and f1(𝒪𝒳)=𝒪𝒟f_{1\ast}(\cal{O}_{X})=\cal{O}_{D^{\prime}}. The map h1h_{1} is the identity if r=1r=1. All fibres of f1f_{1} are connected, while a general fibre of ff has exactly rr connected components. Most books say that ff is a fibration only if f(𝒪𝒳)=𝒪𝒟f_{\ast}(\cal{O}_{X})=\cal{O}_{D}, because the general case is reduced to a fibration by taking a finite map of Riemann surfaces [6, Ch. III, §8]. Assume f(𝒪𝒳)=𝒪𝒟f_{\ast}(\cal{O}_{X})=\cal{O}_{D}, i.e. assume that a general fibre is connected (and so all fibres of ff are connected). Let AA be any smooth fibre of ff and As:=f1(s)A_{s}\mathrel{\mathop{:}}=f^{-1}(s) any fibre of ff. Then the Euler numbers satisfy e(As)e(A)e(A_{s})\geq e(A). If XX is compact, i.e. if DD is compact, then by [6, Prop. III.11.4] we have

e(X)=e(A)e(D)+sD(e(A)e(As)).e(X)=e(A)e(D)+\sum_{s\in D}(e(A)-e(A_{s})). (1)

Moreover, there is an easy criterion which gives e(As)>e(A)e(A_{s})>e(A) if AsA_{s} is singular and not a multiple of a smooth elliptic curve [6, III.11.5]. Hence if DD is compact of genus g2g_{2} and a general fibre of ff is smooth of genus g1g_{1}, then

e(X)4(g11)(g21)\boxed{e(X)\geq 4(g_{1}-1)(g_{2}-1)}

(recall here that the Euler number equals the top Chern number for compact XX) with strict inequality if some of the fibres are singular and either g11g_{1}\neq 1 or g1=1g_{1}=1, but there is at least one fibre with is not a multiple of an elliptic curve [6, III. 11.6]. See [6, §III. 9] for conditions that prevent the existence of multiple fibres, while their for existence see [31].

Remark 2.3.

Let XX be a smooth and connected complex surface. We are interested in the case in which there is a surjective holomorphic map f:XDf\colon X\to D with DD a smooth projective curve and as general fibre of ff a smooth curve of genus g2g\geq 2. A surface XX which has such a fibration is always a projective surface for the following reasons, all due to Kodaira. Let (X)\mathbb{C}(X) be the field of all meromorphic functions on XX. The field (X)\mathbb{C}(X) is a finitely generated extension of the field \mathbb{C} and it has transcendental degree at most 22, see [6, Thm. I.7.1]. The transcendence degree a(X)a(X) of the field (X)\mathbb{C}(X) over \mathbb{C} is called the algebraic dimension of XX. If a(X)=0a(X)=0, then XX has only finitely many curves and, in particular, there is no surjective map f:XDf\colon X\to D with DD a curve [6, Thm. IV.8.2], [22, Thm. 5.1]. Any surface with a(X)=1a(X)=1 admits a fibration u:XDu\colon X\to D with DD a smooth curve and such that a general fibre of uu is an elliptic curve, see [6, VI.5.1],[22, Thm. 4.1], and any irreducible curve TXT\subset X is contained in a fibre of uu [22, Thm. 4.3]. Surfaces with a(X)=2a(X)=2 are algebraic and projective, see[6, Cor. IV.6.5], [22, Thm. 3.1].

2.1. Elliptic fibrations and elliptic surfaces

Let DD be a compact complex curve of genus g20g_{2}\geq 0 and f:XDf\colon X\to D be a proper holomorphic map with XX a compact complex surface and such that a general fibre of ff is an elliptic curve.

Let SS be the set of all sDs\in D such that f1(s)f^{-1}(s) is singular (it may even be a multiple fibre). For any smooth genus 11 curve, let j(E){0}j(E)\in\mathbb{C}\setminus\{0\} be its jj-invariant. Recall that the jj-invariant of an elliptic curve

y2=x3+vex+gy^{2}=x^{3}+vex+g

is

j(τ)=4(24f)3Δ,Δ=4f3+27g2,j(\tau)=\frac{4(24f)^{3}}{\Delta},\quad\Delta=4f^{3}+27g^{2},

and two smooth elliptic curves are isomorphic if and only if they have the same jj-invariant.

For each sDs\in D set Es:=f1(s)E_{s}\mathrel{\mathop{:}}=f^{-1}(s) and note that it is connected [6, p. 200–216]. We will analyse two cases separately, depending on whether there are singular fibres or not.

2.1.1. Case S=S=\emptyset

In the case where there are no singular fibres (as illustrated in Figure 2), i.e. assuming that ff is a submersion, by (1) we have e(X)=0e(X)=0. No way to change that.

Refer to caption
Figure 2. Family without singular fibres

Next, we will use the following result.

Theorem 2.4.

[6, III. 15.4] If XX is compact, i.e. if DD is compact, ff has no singular fibres and either D1D\cong\mathbb{P}^{1} or DD is an elliptic curve, then ff is locally trivial over the base and in particular all fibres of ff are biholomorphic.

Remark 2.5.

Assuming g21g_{2}\leq 1 for the base of an elliptic fibration, then by Theorem 2.4 we obtain that all fibres of ff are isomorphic to the same elliptic curve, EE, and there is a finite open covering {Ui}\{U_{i}\} of DD such that f1(Ui)Ui×Ef^{-1}(U_{i})\cong U_{i}\times E for all ii.

Kodaira gave a classification of the possible fibres occurring over SS [15, Ch. 7].

2.1.2. Case SS\neq\emptyset

For a description of all possible singular fibres (these are of 8 types) see [23, Thm. 6.2]. One case is illustrated in Figure 3.

Refer to caption
Figure 3. Elliptic family with 1 node
Definition 2.6.

A fibration f:XDf\colon X\to D is said to be minimal (or relatively minimal) if no fibre of ff contains a (1)(-1)-curve.

We assume that the elliptic fibration f:XDf\colon X\to D is relatively minimal. The general case is obtained from the relatively minimal one, by making a finite sequence of blowing ups of points. For a relatively minimal elliptic surface XX we have KX2=0K_{X}^{2}=0 and hence the second Chern class becomes expressed in terms of the Euler characteristics and the degree dd of the dual of the line bundle of R1f(𝒪X)R^{1}f_{\ast}(\mathcal{O}_{X}) [15, Corollary 16]:

c2(X)=12χ(𝒪X)=12dfor an elliptic fibration.\boxed{c_{2}(X)=12\chi(\mathcal{O}_{X})=12d\quad\textnormal{for an elliptic fibration}.}

Furthermore, d0d\geq 0 and d=0d=0 if and only if the only singular fibres are multiple fibres whose reduction is smooth.

Therefore, we have noticed here that the presence of singular fibres in the case of elliptic fibrations immediately imply that the second Chern number is positive.

2.2. Fibrations by curves of genus 22

In this section we take a surface XX fibred by curves of genus g1=2g_{1}=2 over a curve of genus g2g_{2}, and use the results of Xiao [40]. Recall that XX is projective for free. We set q(X):=h1(𝒪𝒳)=𝒽0(Ω𝒳1)q(X)\mathrel{\mathop{:}}=h^{1}(\cal{O}_{X})=h^{0}(\Omega_{X}^{1}), and pg(X):=h0(ωX)=h2(𝒪𝒳)p_{g}(X)\mathrel{\mathop{:}}=h^{0}(\omega_{X})=h^{2}(\cal{O}_{X}). Since χ(𝒪𝒳):=1𝓆(𝒳)+𝓅(𝒳)\chi(\cal{O}_{X})\mathrel{\mathop{:}}=1-q(X)+p_{g}(X), we have

χ(𝒪𝒳)21for1=2\boxed{\chi(\cal{O}_{X})\geq g_{2}-1\quad\text{for}\quad g_{1}=2}

and equality holds if and only ff has no singular fibre and all fibres are isomorphic, see [7] or [40, p. 7].

Recall that g2qg2+2g_{2}\leq q\leq g_{2}+2 and that XX is known if q=g2+2q=g_{2}+2. Recall that E:=f(ωX/D)E:=f_{\ast}(\omega_{X/D}) is a rank 22 vector bundle on DD with nonnegative degree and with all rank 11 quotients of nonnegative degree. Let E1E_{1} a rank 11 subsheaf of EE with maximal degree. The maximality of the integer deg(E1)\deg(E_{1}) means that the coherent sheaf E/E1E/E_{1} has no torsion. Since DD is a smooth curve and E/E1E/E_{1} is a rank 11 torsion free sheaf, E/E1E/E_{1} is a line bundle. Set ϵ:=deg(E1)deg(E/E1)\epsilon:=\deg(E_{1})-\deg(E/E_{1}). The vector bundle EE is said to be unstable (resp. properly semistable, resp. stable) if and only if ϵ>0\epsilon>0 (resp. ϵ=0\epsilon=0 (resp. ϵ<0\epsilon<0) (in [21, Ex. V.2.8]) the integer ee is our integer ϵ-\epsilon). Since deg(E)=deg(E1)+deg(E/E1)\deg(E)=\deg(E_{1})+\deg(E/E_{1}) and ϵ=deg(E1)deg(E2)\epsilon=\deg(E_{1})-\deg(E_{2}), we have ϵdeg(E)(mod2)\epsilon\equiv\deg(E)\pmod{2}. If ϵ>0\epsilon>0, i.e. if EE is stable, then the maximal degree rank 11 subsheaf E1E_{1} of EE is unique, because it corresponds to a unique section of the associated ruled surface over DD with negative self-intersection [21, Prop. V.2.21]. The 1\mathbb{P}^{1}-bundle PP is associated to a rank 22 vector bundle FF on DD which is strongly related to EE. If ϵ>0\epsilon>0 this is the Hirzebruch surface FϵF_{\epsilon} and the uniqueness of E1E_{1} corresponds to the fact that the ruled surface Fϵ1F_{\epsilon}\to\mathbb{P}^{1}, ϵ>0\epsilon>0, has a unique section D0D_{0} with negative self-intersection, i.e. with D02=ϵD_{0}^{2}=-\epsilon [21, Th. V.2.17].

The following result is a summary of [40, p. 16], see eq. (9), Thm. 2.1 and the Rmq. that follows it.

Theorem 2.7.

We have ϵpg+1\epsilon\leq p_{g}+1, ϵχ+g21(mod2)\epsilon\equiv\chi+g_{2}-1\pmod{2}, g2ϵχg2+1-g_{2}\leq\epsilon\leq\chi-g_{2}+1. If q>g2q>g_{2}, then ϵ=χg2+1\epsilon=\chi-g_{2}+1; q=g2+1q=g_{2}+1 if and only if ϵ=pg+12g2\epsilon=p_{g}+1-2g_{2}.

With this notation, ϵ>0\epsilon>0 is equivalent to the instability of the vector bundle FF and in this case there is a unique section D0D_{0} of the 1\mathbb{P}^{1}-bundle with negative self-intersection. There is also an effective divisor RPR\subset P [40, p. 12].

Theorem 2.8.

[40, Thm. 2.2]

(i) Assume ϵ>0\epsilon>0 and RD0R\supseteq D_{0}. Then

2χ+6(g21)KX23χ+5(g21)2ϵ2\chi+6(g_{2}-1)\leq K_{X}^{2}\leq 3\chi+5(g_{2}-1)-2\epsilon

and hence

ϵ(χg2+1)/2.\epsilon\leq(\chi-g_{2}+1)/2.

(ii) If either ϵ0\epsilon\leq 0 or else both ϵ>0\epsilon>0 and RD0R\not\supseteq D_{0}, then

max{2χ+6(g21),χ+7(g21)+3ϵ}KX2min{6pg5q+3g2+2,7χ+g21}.\max\{2\chi+6(g_{2}-1),\chi+7(g_{2}-1)+3\epsilon\}\leq K_{X}^{2}\leq\min\{6p_{g}-5q+3g_{2}+2,7\chi+g_{2}-1\}.
Corollary 2.9.

[40, p.18] KX28χ(𝒪𝒳)K_{X}^{2}\leq 8\chi(\cal{O}_{X}).

Corollary 2.9 shows that the invariants of XX are far from extremal: for surfaces of general type the bound is KX29χ(𝒪𝒳)K_{X}^{2}\leq 9\chi(\cal{O}_{X}) [6, Ch. VII, §4].

There is a very long description of surfaces with q=g2+1q=g_{2}+1 in [40, §3] and almost all genus 22 fibrations have q=g2q=g_{2}. For such fibrations the upper bounds in Thm. 2.8 give

KX2min{6χ+4(g21),7χ+g21}.K_{X}^{2}\leq\min\{6\chi+4(g_{2}-1),7\chi+g_{2}-1\}.

Moreover, [40, p. 19] gives a picture showing some forbidden parts on the plane (KX2,χ)(K_{X}^{2},\chi). In a small range of integers g2g_{2}, χ:=χ(𝒪𝒳)\chi:=\chi(\cal{O}_{X}) and ϵ>0\epsilon>0 all numerical invariant are the numerical invariants of some genus 22 fibration.

Theorem 2.10.

[40, Thm. 2.9] Fix integers xx, g20g_{2}\geq 0, ϵ0\epsilon\geq 0, χg21\chi\geq g_{2}-1 such that

ϵχg2+1andϵχ+g21(mod2).\epsilon\leq\chi-g_{2}+1\quad\text{and}\quad\epsilon\equiv\chi+g_{2}-1\pmod{2}.

If ϵ0\epsilon\neq 0 assume that xx satisfies the inequalities of case (ii) of Theorem 2.8 (with xx in the place of KX2K_{X}^{2}). Then there is a genus 22 fibration XDX\to D with χ(𝒪𝒳)=χ\chi(\cal{O}_{X})=\chi, KX2=xK^{2}_{X}=x, g(D)=g2g(D)=g_{2} and with ϵ\epsilon as the degree of stability.

2.2.1. c2(X)c_{2}(X) and c1(X)2c_{1}(X)^{2}

Let XX be a smooth connected complex surface and π:X~X\pi\colon\widetilde{X}\to X be the blow up of XX at one point [16, p. 473]. Then [16, p. 576] shows that

c2(X~)=c2(X)+1,c1(X~)2=c1(X)21,andc_{2}(\widetilde{X})=c_{2}(X)+1,\quad\quad c_{1}(\widetilde{X})^{2}=c_{1}(X)^{2}-1,\text{and}
c1(X)2+c2(X)=12χ(𝒪𝒳)=12χ(𝒪𝒳~)=𝒸1(𝒳~)2+𝒸2(𝒳~).c_{1}(X)^{2}+c_{2}(X)=12\chi(\cal{O}_{X})=12\chi(\cal{O}_{\widetilde{X}})=c_{1}(\widetilde{X})^{2}+c_{2}(\widetilde{X}).

Thus, one often assumes that the surface XX has no negative curve of the first kind, i.e. no curve TXT\subset X such that T1T\cong\mathbb{P}^{1} and T2=1T^{2}=-1. Indeed, if any such TT exists we may blow it down to get another smooth compact surface. Every smooth compact surface has a minimal model, i.e. there is a finite sequence u:XX1u\colon X\to X_{1} of blowing downs of curves of the first kind with X1X_{1} a minimal model and one may then only study c2(X1)c_{2}(X_{1}) and c1(X1)2c_{1}(X_{1})^{2}. The situation is a bit different if we wish to study fibrations.

Remark 2.11.

In [40, § 6] there is a complete list of all surfaces of general type with more than one fibration of genus 22 curves. Fix an integer g1g\geq 1. A surface XX of general type has at most finitely many fibrations whose general fibre is a genus gg curve [40, Prop. 6.1]. There is a surface of general type XX with fibrations by genus gg curves for infinitely many genera gg [40, Ex. 6.3]. For each g1g\geq 1 there is an example of a surface XX with κ(X)=1\kappa(X)=1 and infinitely many (countably many) fibrations whose general fibre is a smooth curve of genus gg [40, Ex. 6.2].

Remark 2.12.

There are many compact complex surfaces which admit no non-constant holomorphic maps f:XDf\colon X\to D with DD any smooth compact curve. For instance, no surface with Pic(X)\mbox{Pic}(X)\cong\mathbb{Z} has such an ff for any compact DD. Accordingly, 2\mathbb{P}^{2} has no such map ff, neither do even some K3 surfaces, many surfaces of general type, and also all complete intersection surfaces. Lefschetz pencils show that, for any XX, there exists a blowing up at finitely many points u:X~Xu\colon\widetilde{X}\to X such that X~\widetilde{X} has a surjection f:X~1f\colon\widetilde{X}\to\mathbb{P}^{1} with each fibre either smooth or irreducible with only one node. However, uu is usually a blow up of XX at many points.

For a surface of general type c1(X)23c2(X)\boxed{c_{1}(X)^{2}\leq 3c_{2}(X)}, see [6, Ch. VII, Th. 4.1], and for minimal surfaces of general type c1(X)2>0\boxed{c_{1}(X)^{2}>0}.

Now we consider surfaces (minimal or not) with a fibration.

Remark 2.13.

Let f:XDf\colon X\to D be a minimal fibration as in Def. 2.6, with DD a smooth curve of genus g2>0g_{2}>0. Since every holomorphic map 1D\mathbb{P}^{1}\to D is constant, XX contains no rational curve not even singular ones. Indeed, suppose there is a, possibly singular, rational curve TXT\subset X and call w:1Tw\colon\mathbb{P}^{1}\to T the normalisation map. Since fw:1Df\circ w\colon\mathbb{P}^{1}\to D is constant, TT is contained in a fibre of ff. Thus if ff is a minimal, then XX is minimal. In particular XX has no exceptional curve of the first kind, i.e. it is a minimal surface as an abstract surface.

For general properties of surfaces, we recommend [6, Ch. IV], in particular the first 6 sections. If we allow the non-algebraic case and call a(X){0,1,2}a(X)\in\{0,1,2\} the transcendental degree of the field (X)\mathbb{C}(X) over \mathbb{C}, then the existence of a morphism f:XDf\colon X\to D implies that XX contains a one-parameter family of curves and hence a(X)>0a(X)>0. The case a(X)=1a(X)=1 may arise only if DD has genus 0 and the fibres of ff have genus 11, so a case not of interest in this subsection. We have a(X)=2a(X)=2 if and only XX is projective [6, Cor. IV.6.5].

We always have q(X)g1+g2q(X)\leq g_{1}+g_{2}, where g1g_{1} is the genus of the general fibre of ff and g2g_{2} is the genus of the base, and

χ(𝒪𝒳)2(11)(21),\boxed{\chi(\cal{O}_{X})\geq 2(g_{1}-1)(g_{2}-1)},

see [6, Ch. III, Cor. 11.6], and for further details see [6, Ch. III, Prop. 11.4(ii)] where the exact contribution of the singular fibres is computed.

If both g12g_{1}\geq 2 and g22g_{2}\geq 2, then XX is of general type, because in such cases any rational or elliptic curves are contained in fibres of ff and there are only finitely many of those. In these cases we also have q(X)g12q(X)\geq g_{1}\geq 2. Now assume (any g1g_{1}) that XX is of general type. If we call XX^{\prime} the minimal reduction of XX, we have not only c1(X)23c2(X)c_{1}(X^{\prime})^{2}\leq 3c_{2}(X^{\prime}) (which is better than just using the same inequality for XX), but also the Noether inequality

pg(X)12c1(X)2+2p_{g}(X^{\prime})\leq\frac{1}{2}c_{1}(X^{\prime})^{2}+2

[6, Ch. VII, Thm. 3.1], with pg(X)=pg(X)p_{g}(X)=p_{g}(X^{\prime}). Surfaces XX^{\prime} for which equality holds are called surfaces on the Noether’s line. Recall also that q(X)=q(X)q(X^{\prime})=q(X). As corollaries one obtains other inequalities and a sufficient condition for having q(X)=0q(X^{\prime})=0 [6, IV. Cor. 3.2&3.3].

We also recall the Albanese mapping αX:XAlb(X)\alpha_{X}\colon X\to\mathrm{Alb}(X) with Alb(X)=(Pic0X)\mathrm{Alb}(X)=(\mathrm{Pic}_{0}X)^{\vee} a complex compact torus (it is algebraic because XX is algebraic and hence Kähler and in such a case dimAlb(X)=q(X)\dim\mathrm{Alb}(X)=q(X) [6, pp. 44–47]. If αX(X)\alpha_{X}(X) is a curve, then the curve αX(X)\alpha_{X}(X) is smooth, connected and of genus q(X)q(X) [6, Cor. I.13.9(iii)].

Since any morphism from 1\mathbb{P}^{1} to a compact complex torus is constant, αX(T)\alpha_{X}(T) is a point for every rational curve TXT\subset X (even singular TT, but with 1\mathbb{P}^{1} as its normalisation), if u:XXu\colon X\to X^{\prime} is the minimal reduction of XX, then αX=αXu\alpha_{X}=\alpha_{X^{\prime}}\circ u.

Remark 2.14.

Assuming q(X)=1q(X)=1, then we have that Alb(X)\mathrm{Alb}(X) is an elliptic curve and αX:XAlb(X)\alpha_{X}\colon X\to\mathrm{Alb}(X) is a fibration (i.e. it has connected fibres and only finitely many singular fibres). A general fibre may be rational, but in such a case XX is birational to D×1D\times\mathbb{P}^{1}. All possible genera may occur as the genus of a general fibre of αX\alpha_{X}.

We always have q(X)g2q(X)\geq g_{2}, since f:Pic(D)Pic(X)f^{\ast}\colon\mathrm{Pic}(D)\to\mathrm{Pic}(X) is injective, given that the fibres of ff are connected. Thus, a necessary condition to have a fibration with target of positive genus is that q(X)>0q(X)>0. By Remark 2.14 all XX with q(X)=1q(X)=1 have a fibration with g2=1g_{2}=1 and the fibration is unique, up to isomorphisms of XX and the base.

For fibrations by curves of genus 3 or higher see [5].

2.3. Examples of hyperelliptic fibrations with nodes

By definition, a smooth curve CC of genus g2g\geq 2 is hyperelliptic if there is a degree 22 morphism u:C1u:C\to\mathbb{P}^{1}. Equivalently, by the universal property of the projective line, CC is hyperelliptic if and only if there exist a degree 22 line bundle LL on CC and a 22-dimensional linear space VH0(L)V\subseteq H^{0}(L) with no base points (hence inducing the map uu).

Refer to caption
Figure 4. Hyperelliptic family with irreducible singular fibre

Around 1817, N. Abels showed that such genus gg curves may be understood as follows. Fix a degree 2g+22g+2 polynomial f(x)[x]f(x)\in\mathbb{C}[x] without multiple roots. Consider the affine curve

y2=f(x).y^{2}=f(x).

Its unique smooth projective completion is a hyperelliptic curve of genus gg, the morphism uu being induced by the projection (x,y)x(x,y)\to x. If we view 2\mathbb{C}^{2} as an affine chart of the plane 2\mathbb{P}^{2}, then the corresponding closure of the affine curve is singular.

A natural ambient for hyperelliptic curves is the weighted projective plane (1,1,g+1)\mathbb{P}(1,1,g+1). We take coordinates x0,x1,yx_{0},x_{1},y and call h(x0,x1)h(x_{0},x_{1}) the homogeneous degree g+1g+1 polynomial such that h(1,x1)=f(x)h(1,x_{1})=f(x). Giving weight g+1g+1 to the variable yy, the equation y2=h(x0,x1)y^{2}=h(x_{0},x_{1}) defines our curve in (1,1,g+1)\mathbb{P}(1,1,g+1). Furthermore, fixing any integer tt such that 1tg1\leq t\leq g, we may take as f(x)f(x) a degree 2g+22g+2 polynomial with 2g+22t2g+2-2t simple roots and tt roots of multiplicity 22. Then the curve y2=h(x0,x1)y^{2}=h(x_{0},x_{1}) is an irreducible and nodal curve of arithmetic genus gg with exactly tt nodes.

Now we explain two other constructions containing all smooth genus gg hyperelliptic curves within an ambient surface (here a smooth surface) and, containing in the same linear system (so the same arithmetic genus gg) irreducible curves with exactly tt nodes for all t=1,,gt=1,\dots,g.

First construction: Set F0:=1×1F_{0}:=\mathbb{P}^{1}\times\mathbb{P}^{1}. We have Pic(S)2\mathrm{Pic}(S)\cong\mathbb{Z}^{2} and we may take as a generators of Pic(F0)\mathrm{Pic}(F_{0}) the isomorphism classes 𝒪0(1,0)\cal{O}_{F_{0}}(1,0) and 𝒪0(0,1)\cal{O}_{F_{0}}(0,1) of the fibres of the 22 projections F01F_{0}\to\mathbb{P}^{1}. For all (a,b)2(a,b)\in\mathbb{N}^{2} we have h0(𝒪0(𝒶,𝒷))=(𝒶+1)(𝒷+1)h^{0}(\cal{O}_{F_{0}}(a,b))=(a+1)(b+1).

Now assume a>0a>0 and b>0b>0. In this case 𝒪0(𝒶,𝒷)\cal{O}_{F_{0}}(a,b) is very ample. Hence, a general element |𝒪0(𝒶,𝒷)||\cal{O}_{F_{0}}(a,b)| is smooth and connected. We have h0(𝒪𝒞)=1h^{0}(\cal{O}_{C})=1 for any C|𝒪0(𝒶,𝒷)|C\in|\cal{O}_{F_{0}}(a,b)|, even for the ones with multiple components.

Since ωF0𝒪0(2,2)\omega_{F_{0}}\cong\cal{O}_{F_{0}}(-2,-2), ωC𝒪𝒞(𝒶2,𝒷2)\omega_{C}\cong\cal{O}_{C}(a-2,b-2) for any C|𝒪0(𝒶,𝒷)|C\in|\cal{O}_{F_{0}}(a,b)|. Thus, all C|𝒪0(𝒶,𝒷)|C\in|\cal{O}_{F_{0}}(a,b)| have arithmetic genus 1+abab1+ab-a-b.

(a) The parameter space: Fix an integer tt such that 0t1+abab0\leq t\leq 1+ab-a-b. Let V(a,b,t)V(a,b,t) denote the set of all irreducible and nodal C|𝒪0(𝒶,𝒷)|C\in|\cal{O}_{F_{0}}(a,b)| with exactly tt nodes. We have

V(a,b,t),V(a,b,t)\neq\emptyset,

and V(a,b,t)V(a,b,t) is irreducible with dimV(a,b,t)=dim|𝒪0(𝒶,𝒷)|𝓉=𝒶𝒷+𝒶+𝒷𝓉\dim V(a,b,t)=\dim|\cal{O}_{F_{0}}(a,b)|-t=ab+a+b-t, see [37, 39]. Take a=2a=2 and b=g+1b=g+1. Each C|𝒪0(2,+1)|C\in|\cal{O}_{F_{0}}(2,g+1)| has arithmetic genus gg. Each smooth CC is hyperelliptic (use either of the 22 projections F01F_{0}\to\mathbb{P}^{1}).

Let DD be a smooth hyperelliptic curve of genus gg. Fix a general LPicg+1(D)L\in\mathrm{Pic}^{g+1}(D). Since g2g\geq 2 and LL is general, h0(L)=2h^{0}(L)=2 and LL is base point free. Thus |L||L| induces a degree g+1g+1 morphism v:D1v:D\to\mathbb{P}^{1}. Let u:D1u:D\to\mathbb{P}^{1} be the degree 22 map given by the hyperellipticity of DD and RPic2(D)R\in\mathrm{Pic}^{2}(D) the associated line bundle. Let w=(u,v)1×1w=(u,v)\to\mathbb{P}^{1}\times\mathbb{P}^{1} be the morphism induced by uu and vv. Since LL is not a multiple of RR and 4deg(u)=24\deg(u)=2, ww is birational onto its image. The curve Im(w)|𝒪0(2,+1)|\mathrm{Im}(w)\in|\cal{O}_{F_{0}}(2,g+1)| has arithmetic genus gg and the smooth genus DD curve as its normalization. Thus Im(w)D\mathrm{Im}(w)\cong D and w(D)|𝒪0(2,+1)|w(D)\in|\cal{O}_{F_{0}}(2,g+1)|.

In general, we may take any Hirzebruch surface FeF_{e}, e0e\geq 0 [21, Ch. V,§2], and as in step (b) below we may use explicit equations for the linear systems in FeF_{e}, many of which contain hyperelliptic curves. For any 0tpa0\leq t\leq p_{a}, where pap_{a} is the arithmetic genus of any element of |𝒪(𝒶𝒽+𝒷𝒻)||\cal{O}_{F_{e}}(ah+bf)|, the existence, irreducibility, and the dimension of the family all irreducible nodal curves in a prescribed linear system |𝒪(𝒶𝒽+𝒷𝒻)||\cal{O}_{F_{e}}(ah+bf)| with exactly tt nodes are given in [37, 39].

(b) Hyperelliptic families with irreducible fibres having 1 or 2 nodes: Fix an integer g2g\geq 2. Let ()\cal{H}(g) be the stack of all smooth hyperelliptic curves of genus gg. We recall that any smooth genus 22 curve is hyperelliptic. For any X()X\in\cal{H}(g) call hX:X1h_{X}\colon X\to\mathbb{P}^{1} the degree 2 morphism, usually denoted g21g^{1}_{2}, corresponding (by definition) to the hyperelliptic curve XX.

We recall that such degree 2 morphism is unique for each hyperelliptic curve of genus g>1g>1 and for g=2g=2 it is the canonical map. Denote by RR the only spanned degree 22 line bundle on XX, so that hXh_{X} is the morphism associated to the complete linear system |R||R|. Since deg(L)=2g+22g+1\deg(L)=2g+2\geq 2g+1, the line bundle L:=R(g+1)L:=R^{\otimes(g+1)} is very ample and non-special. Thus h0(L)=deg(L)+1gh^{0}(L)=\deg(L)+1-g and |L||L| induces an embedding f:Xg+1f\colon X\to\mathbb{P}^{g+1}. Fix D1,D2,D3|R|D_{1},D_{2},D_{3}\in|R| such that DiDjD_{i}\neq D_{j} for all iji\neq j. Since ff is an embedding and deg(Di)=2\deg(D_{i})=2, f(Di)f(D_{i}) spans a line Di\langle D_{i}\rangle. The line bundle R(g1)R^{\otimes(g-1)} induces a degree 22 map with as image 1\mathbb{P}^{1} (case g=2g=2) or a rational normal curve of g1\mathbb{P}^{g-1} (case g2g\geq 2). We have h0(Rg2)=g1h^{0}(R^{\otimes g-2})=g-1. We see that the lines Di\langle D_{i}\rangle and Dj\langle D_{j}\rangle span a plane and hence they meet. The 33 lines D1D2D3\langle D_{1}\rangle\cup\langle D_{2}\rangle\cup\langle D_{3}\rangle span a 3\mathbb{P}^{3}, because h0(Rg2)=g1=h0(R(g+1))4h^{0}(R^{\otimes g-2})=g-1=h^{0}(R^{\otimes(g+1)})-4. We obtain that f(X)f(X) is contained in a cone TT with vertex of(X)o\notin f(X) and as a base a rational normal curve of g+1\mathbb{P}^{g+1}, see [12]. Let u:YTu\colon Y\to T be the minimal resolution of TT. The surface YY is isomorphic to the Hirzebruch surface Fg+1F_{g+1} and h:=u1(o)1h:=u^{-1}(o)\cong\mathbb{P}^{1}. We have Pic(Y)2\mathrm{Pic}(Y)\cong\mathbb{Z}^{2} with hh and a fibre ff of its ruling (i.e. the strict transform of a line of TT passing through oo) as a basis over \mathbb{Z}, with intersection numbers

f2=0,hf=1,andh2=g1.f^{2}=0,\quad h\cdot f=1,\quad\text{and}\quad h^{2}=-g-1.

For any curve DFD\subset F (even reducible) which is not a member of |cf||cf| for any c>0c>0, there are integers a>0a>0 and ba(g+1)b\geq a(g+1) such that D|𝒪𝒴(𝒶𝒽+𝒷𝒻)|D\in|\cal{O}_{Y}(ah+bf)|. The linear system |𝒪𝒴(𝒶𝒽+𝒷𝒻)||\cal{O}_{Y}(ah+bf)| contains a curve DD such that Dh=D\cap h=\emptyset, i.e. such that ou(D)o\notin u(D) if and only if b=a(g+1)b=a(g+1).

The case a=2a=2 and b=2g+2b=2g+2 is the linear system corresponding to our genus gg hyperelliptic curves. We take an arbitrary integer a>0a>0 and give an explicit parameter space for the part of the linear system |𝒪𝒴(𝒶𝒽+𝒶(+1)𝒻)||\cal{O}_{Y}(ah+a(g+1)f)| not intersecting hh.

Fix variables x0,x1,yx_{0},x_{1},y and give weight 11 to the variables x0x_{0} and x1x_{1} and weight g+1g+1 to the variable yy. Let V(g+1,a)V(g+1,a) denote the set of all f[x0,x1,y]f\in\mathbb{C}[x_{0},x_{1},y] which are weighted homogeneous with total degree a(g+1)a(g+1). The zero-locus of any vV(a+1,g)v\in V(a+1,g) is an element D|𝒪𝒴(𝒶𝒽+(+1)𝒻|D\in|\cal{O}_{Y}(ah+(g+1)f| such that Dh=D\cap h=\emptyset. Those belonging to the linear system |𝒪𝒴(𝒶𝒽+𝒶(+1)𝒻)||\cal{O}_{Y}(ah+a(g+1)f)| such that DhD\cap h\neq\emptyset have hh as a component, because the intersection number of h+(g+1)fh+(g+1)f and hh is 0.

Take again a=2a=2. In this case we get a linear system whose smooth elements are smooth genus gg hyperelliptic curves. Their nodal degenerations may contain tt nodes, for any 0tg0\leq t\leq g by [37, 39] and also the reducible nodal curves of the form D1D2D_{1}\cup D_{2} with D1D21D_{1}\cong D_{2}\cong\mathbb{P}^{1} and #(D1D2)=g+1\#(D_{1}\cap D_{2})=g+1. It is sufficient to take as D1D_{1} and D2D_{2} two general elements of |h+(g+1)f||h+(g+1)f|.

Refer to caption
Figure 5. Family with 1 irreducible fibre having 2 nodes

(c) Hyperelliptic families with reducible nodal fibres: For an integer g2g\geq 2, let ¯g\overline{\cal{M}}_{g} denote the coarse moduli space of stable genus gg curves. In the boundary ¯g\overline{\cal{M}}_{g}\setminus\cal{M}_{g} we get all nodal reducible curves CEC\cup E with CC a smooth curve of genus gg, EE an elliptic curve and CEC\cap E a unique point. If g=2g=2 we get a nodal union of 22 elliptic curves which occurs as a limit of a family of smooth genus 22 curves. Now assume g3g\geq 3 and assume that CC is hyperelliptic. Both the theory of generalised coverings [20] and that of limit linear series [14] give that CEC\cup E is the flat limit of a family of hyperelliptic curves. See also [19, Ch. 6 C]. In ¯g\overline{\cal{M}}_{g} there exists a point representing the nodal reducible curve D1D2D_{1}\cup D_{2} described at the end of part (b).

2.4. Examples of genus 2 fibrations with nodes

We use another construction to build more examples of hyperelliptic fibrations with nodes.

2.4.1. Case SS\neq\emptyset

Second construction: Let XX be a smooth Del Pezzo surface of degree 11 [11, Ch. 8]. This smooth surface XX may be realised as a blowing up of 2\mathbb{P}^{2} at a set S2S\subset\mathbb{P}^{2} of 88 points, with the conditions that no 33 of the points of SS are colinear, no 66 of the points of SS are contained in a conic and no cubic surface passing through each point in S that is singular at one of them.

It is most useful to see XX as a sextic hypersurface in the weighted projective space (1,1,2,3)\mathbb{P}(1,1,2,3) (We recall here that this notation of the 44 coordinates weights means that λ[x0,x1,x2,x3]=[λx0,λx1,λ2x2,λ3x3]\lambda[x_{0},x_{1},x_{2},x_{3}]=[\lambda x_{0},\lambda x_{1},\lambda^{2}x_{2},\lambda^{3}x_{3}].

For any integer r1r\geq 1 we have

dim|rKX|=r(r+1)/2,\dim|-rK_{X}|=r(r+1)/2,

this is the case d=1d=1 of [11, Lemma 8.3.1]. Hence dim|KX|=1\dim|-K_{X}|=1 and dim|2KX|=3\dim|-2K_{X}|=3. The pencil |KX||-K_{X}| had a unique base point pp, and a general element of |KX||-K_{X}| is a smooth elliptic curve. Every element of |2KX||-2K_{X}| has arithmetic genus 33, a general element of |2KX||-2K_{X}| is smooth and among the elements of |2KX||-2K_{X}| there are two curves E1E2E_{1}\cup E_{2} with (E1,E2)(E_{1},E_{2}) general in |KX||-K_{X}|.

(a) Existence of reducible curves with one node: So we have a 22-dimensional family of nodal unions of 22 elliptic curves with a unique node at pp, inside the 33-dimensional projective space |2KX||-2K_{X}|. This is illustrated in Figure 6.

Refer to caption
Figure 6. Family with node on a reducible fibre

A general line in |2KX||-2K_{X}| gives a fibration with 1\mathbb{P}^{1} as its base. This type of construction, which was earlier considered by Halphen, is described further details on the section of Halphen’s pencils in [11, Ex. 7.20].

(b) Existence of irreducible curves with nodes on SS: This paragraph is due to the referee. The linear system |2KX||-2K_{X}| gives a double cover π:X(1,1,2)\pi\colon X\rightarrow\mathbb{P}(1,1,2) branched over the vertex of the quadric cone (1,1,2)3\mathbb{P}(1,1,2)\subset\mathbb{P}^{3} and a smooth curve RR of degree 66. Take general point PP in X that is mapped to R, and let 𝒫\mathcal{P} be the pencil in|2KX||-2K_{X}| that consists of pre-images of hyperplane sections of the cone (1,1,2)\mathbb{P}(1,1,2) that are tangent to RR at π(P)\pi(P). Then a general member of the pencil 𝒫\mathcal{P} is irreducible and has a node at PP.

Refer to caption
Figure 7. Family with 2 nodes on different fibres

2.5. Families of hyperelliptic curves in 1×1\mathbb{P}^{1}\times\mathbb{P}^{1}

For completeness we present some further details of the toric case. However, we observe that our constructions provide genus 2 unions of elliptic curves inside toric varieties only in the case just described above, which is a very particular construction on a blowing up of 2\mathbb{P}^{2}.

Fix integers g2g\geq 2 and 0cg0\leq c\leq g and a general S1×1S\subset\mathbb{P}^{1}\times\mathbb{P}^{1} with #S=c\#S=c. Now we consider the zero-dimensional scheme ZZ defined by

Z:=pS2p.Z:=\cup_{p\in S}2p.

Elements of |𝒪1×1(2,+1)||\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1)| containing ZZ are exactly the elements of |𝒪1×1(2,+1)||\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1)| singular at all points of SS. We have h0(𝒪1×1(2,+1))=3(+2)h^{0}(\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1))=3(g+2). It is well-known that h0(𝒵(2,+1))=3(+2)3𝒸h^{0}(\cal{I}_{Z}(2,g+1))=3(g+2)-3c ([27]), i.e.

dim|𝒵(2,𝒸)|=3+53𝒸\dim|\cal{I}_{Z}(2,c)|=3g+5-3c

and that a general D|𝒵(2,𝒸)|D\in|\cal{I}_{Z}(2,c)| is an integral curve of arithmetic genus gg, geometric genus gcg-c and exactly cc ordinary nodes as singularities. If SS is defined over \mathbb{R} we may even find DD defined over \mathbb{R}.

A general C|𝒪1×1(2,+1)|C\in|\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1)| is smooth. Since ω1×1𝒪1×1(2,2)\omega_{\mathbb{P}^{1}\times\mathbb{P}^{1}}\cong\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(-2,-2), the adjunction formula gives ωC𝒪𝒞(0,1)\omega_{C}\cong\cal{O}_{C}(0,g-1). So, deg(ωC)=2g2\deg(\omega_{C})=2g-2 and CC has genus gg.

Therefore, we get a 3g+53g+5-dimensional family of smooth genus gg curves. All of them are hyperelliptic. Indeed, the 22-to-11 morphism C1C\to\mathbb{P}^{1} is induced by the projection 1×1\mathbb{P}^{1}\times\mathbb{P}^{1} onto one of its factor induced by the linear system |𝒪1×1(0,1)||\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(0,1)| (we have h0(𝒪1×1(0,1))=2h^{0}(\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(0,1))=2 and the intersection number (0,1)C(0,1)\cdot C is 22, because (0,1)(2,g+1)=2(0,1)\cdot(2,g+1)=2.

Remark 2.15.

The linear system |𝒪1×1(2,+1)||\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1)|, g2g\geq 2, contains all smooth genus gg hyperelliptic curves. The linear system |𝒪1×1(1,1)||\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(1,1)| embeds 1×1\mathbb{P}^{1}\times\mathbb{P}^{1} as a smooth quadric surface QQ. In these examples each smooth C|𝒪1×1(2,+1)|C\in|\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(2,g+1)| is embedded as a degree g+3g+3 non-special curve in 3\mathbb{P}^{3}. Conversely, take any smooth CQC\subset Q of degree g+3g+3 and genus g2g\geq 2, say C|𝒪1×1(𝒶,𝒷)|C\in|\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(a,b)|. Because g>0g>0, we have a2a\geq 2 and b2b\geq 2. The isomorphism 𝒪3(1,1)|𝒬𝒪1×1(1,1)\cal{O}_{\mathbb{P}^{3}}(1,1)_{|Q}\cong\cal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(1,1) gives g+3=deg(C)=a+bg+3=\deg(C)=a+b. Since ω1×1=𝒪1×1(2,2)\omega_{\mathbb{P}^{1}\times\mathbb{P}^{1}}=\mathcal{O}_{\mathbb{P}^{1}\times\mathbb{P}^{1}}(-2,-2), the adjunction formula gives ωC𝒪𝒞(𝒶2,𝒷2)\omega_{C}\cong\cal{O}_{C}(a-2,b-2) and consequently g=abab+1g=ab-a-b+1. Up to a change of the two factors of 1×1\mathbb{P}^{1}\times\mathbb{P}^{1} we may assume aba\leq b. We get a=2a=2 and b=g+1b=g+1. Thus, curves of degree g+3g+3 and genus gg contained in QQ are hyperelliptic.

3. Fibrations with smooth fibres

We describe the case of surfaces. Let f:XDf\colon X\to D be a holomorphic submersion with XX a smooth surface and DD a smooth (even open) Riemann surface. They are described in [6, Ch V §4–6]. Let g1g_{1} be the genus of the fibre, call YY the isomorphism class of a fibre.

The easy case of ruled surfaces is described in [6, p. 189-192], followed by the case of elliptic fibre bundles [6, p. 193–198], with classification when D1D\cong\mathbb{P}^{1} and many results (even classifications of some subcases) for DD of genus 11.

As for the case of higher genus fibre bundles, when YY has genus g2g\geq 2, Aut(Y)\mathrm{Aut}(Y) is finite (indeed #Aut(Y)84(g1)\#\mathrm{Aut}(Y)\leq 84(g-1) by Hurwitz automorphism theorem). Every fibre bundle over the smooth curve DD with YY as a fibre is given by a representation π1(D)Aut(Y)\pi_{1}(D)\to\mathrm{Aut}(Y). There is a finite unramified covering u:D~Du\colon\tilde{D}\to D such that making the fibre product with uu we get a fibre bundle f~:X~D~\tilde{f}\colon\tilde{X}\to\tilde{D} with X~=Y×D~\tilde{X}=Y\times\tilde{D} and f~\tilde{f} the projection onto the second factor [6, pp. 199–200].

3.1. Smooth semistable fibrations

In this section we consider certain fibrations whose general fibre is a smooth curve of genus g2g\geq 2 following [6, §III.10]. We allow singular fibres, but only with ordinary double points.

Definition 3.1.

A semistable fibration f:XDf\colon X\to D is a fibration with connected fibres, (i.e. f(𝒪𝒳)=𝒪𝒟f_{\ast}(\cal{O}_{X})=\cal{O}_{D}) with XX a smooth and connected complex surface (even not compact) DD a Riemann surface (not necessarily compact or algebraic), a general fibre of ff is a smooth curve of genus g2g\geq 2, all fibres have at most ordinary double points as singularity and no fibre contains a (1)(-1)-curve.

Take f:XDf\colon X\to D semistable as in Definition 3.1; XX may have (1)(-1)-curves, say JJ (not contained in a fibre by assumption), but in this case DD is compact, DD has genus 0 and f|J:JDf_{|J}:J\to D is a finite map. In all other cases XX is a minimal surface.

The reason that Definition 3.1 requires that XX has no (1)(-1)-curve contained in a fibre is that if it has one you contract it and get f1:X1Df_{1}\colon X_{1}\to D with all the other properties and XX1X\to X_{1} a blowing up of one point.

Remark 3.2.

A statement analogous to Theorem 2.4 is no longer true when the base is allowed to have genus higher than 1. Indeed, the famous Kodaira surfaces, which we discuss in the following section, are counterexamples to such a statement.

3.2. Fibrations with general fibre hyperelliptic

Let f:XDf\colon X\to D be a fibration whose fibres are smooth hyperelliptic curves. In this case there is a monograph [40] proving certain ranges of Chern numbers are allowed/not allowed for XX.

There are similar results for fibrations f:XDf\colon X\to D whose general fibre is a smooth hyperelliptic curve; the most important modern result was proved by Gujar, Paul and Purnaprajna in [17], showing that ff has at most multiple fibres of multiplicity 22 and that if gg is even no multiple fibre at all [17, Thm. 2].

Remark 3.3.

If f:XDf\colon X\to D is a hyperelliptic fibration, then there is a rational map Φ:XP\Phi\colon X\dasharrow P, where u:PDu\colon P\to D is a 1\mathbb{P}^{1}-bundle on DD and vΦ=fv\circ\Phi=f, hence the diagram

X\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}f\scriptstyle{f}φ\scriptstyle{\varphi}P\textstyle{P\ignorespaces\ignorespaces\ignorespaces\ignorespaces}u\scriptstyle{u}D\textstyle{D}

commutes, but Φ\Phi is only a rational map generically 22 to 11, which on a smooth fibre FF of ff is the degree 22 morphism coming from the definition of hyperelliptic curve.

An observation for the case of smooth compact complex surfaces; if the genus of the fibre satisfies g12g_{1}\geq 2 we may drop the projectivity of XX, because it is a consequence of the existence of the fibration in curves of genus at least 22 [6, Ch.VI §5]. The relative canonical sheaf of ff (usually denoted with ωX/D\omega_{X/D}) is the line bundle ωX/D:=ωXf(ωD)\omega_{X/D}\mathrel{\mathop{:}}=\omega_{X}\otimes f^{\ast}(\omega_{D}). Often you see KXK_{X} instead of ωX\omega_{X} and KDK_{D} instead of ωD\omega_{D}; they are the same line bundles, just a different notation. The sheaf E:=f(ωX/D)E\mathrel{\mathop{:}}=f_{\ast}(\omega_{X/D}) is a rank gg vector bundle on DD. It has the following type of nonnegativity.

Theorem 3.4.

[40, Thm. 1.1] Every vector bundle FF on DD such that there is a surjection EFE\to F has nonnegative degree.

Let g2g_{2} the genus of DD. A. Beauville proved the second inequality in [7, p. 345], while the first inequality in the following lemma is obvious.

Lemma 3.5.

We have g2qg1+g2g_{2}\leq q\leq g_{1}+g_{2}. Moreover, q=g1+g2q=g_{1}+g_{2} if and only if XX is birational to D×CD\times C with CC a curve of genus g1g_{1}.

Fibrations by smooth hyperelliptic fibres are not very varied, more precisely by [40, Proposition 2.10]:

Refer to caption
Figure 8. Smooth hyperelliptic fibration
Proposition 3.6.

Let DD be a smooth projective curve, SS a smooth compact surface and f:SDf\colon S\to D be a morphism such that all its fibres are smooth hyperelliptic curves of genus at least 22. Then ff is isotrivial.

Recall that isotrivial means that the smooth fibres are all isomorphic, as in Figure 8. For more details on isotrivially fibred surfaces see [35, 36].

3.3. Kodaira fibrations

Following [6, Ch. V§14] we discuss Kodaira fibrations as examples of families of curves that behave very differently from families of hyperelliptic curves.

Definition 3.7.

A Kodaira fibration is a smooth compact complex surface XX such that there is a submersion f:XDf\colon X\to D with DD a smooth compact curve, all fibres of ff are smooth of genus gg, but they do not form a locally trivial fibre bundle in the holomorphic category (of course it is a differentially locally trivial fibre bundle).

Refer to caption
Figure 9. Kodaira fibration
Remark 3.8.

In view of the Grauert–Fischer Theorem 2.2 this means that, though all fibres are smooth curves, their complex structure varies. It follows immediately from the uniqueness of 1\mathbb{P}^{1} as a curve of genus 0 and the existence of the JJ-fibration [6, Sec. 9] that the fibre genus of a Kodaira fibration is at least 22. Such an inequality also holds for the base genus. Hence, for a Kodaira surfaces we have g12g_{1}\geq 2 and g22g_{2}\geq 2, see [6, p. 220].

Remark 3.9.

There does not exist any Kodaira surface whose fibres are smooth curves of genus 22 [40, Prop. 2.10]. In further generality, if f:XDf\colon X\to D is any submersion whose fibres are smooth hyperelliptic curves, then all fibres of ff are isomorphic [40, Prop. 2.10]. Since by Theorem 2.2 not being a fibre bundle is equivalent to not all fibres being isomorphic, it follows that fibres of a Kodaira fibration are not hyperelliptic curves.

Nowadays, the preferred approach to Kodaira surfaces uses moduli spaces and moduli stacks for both curves and surfaces. The reader may find this approach at its best in [8] by F. Catanese, in references therein, and certainly also in papers quoting [8]. We summarise a few very interesting facts. Let \cal{M}_{g}, g2g\geq 2, be the coarse moduli scheme of genus gg curves. Any genus gg Kodaira fibration f:XDf\colon X\to D induces a non-constant morphism w:Dw\colon D\to\cal{M}_{g}. Thus, the known fact that there is no genus 22 Kodaira fibration follows from the fact that 2\cal{M}_{2} is affine, while the existence of genus g3g\geq 3 Kodaira fibrations follows from the well known result that for genus g3g\geq 3 the coarse moduli scheme \cal{M}_{g} contains projective curves. Furthermore, if g3g\geq 3 then for a general [C][C]\in\cal{M}_{g} there exists a projective curve TT\subset\cal{M}_{g} such that [C]T[C]\in T, see [19, Thm. 2.33]. Consequently, for every g3g\geq 3 there exists a Kodaira fibration containing a general genus gg curve.

Lemma 3.10.

Fix an integer g3g\geq 3. In any genus gg Kodaira fibration each fibre is isomorphic to only finitely many fibres.

Proof.

Let g\mathcal{M}_{g} be the coarse moduli scheme of genus gg curves, which is a quasi-projective variety. The morphism f:XDf\colon X\to D is a flat family of genus gg curves. By the universal property of coarse moduli spaces the family ff induces a morphism w:Dgw\colon D\to\mathcal{M}_{g}. If the fibres F1F_{1} and F2F_{2} are isomorphic as abstract curves, then w(F1)=w(F2)w(F_{1})=w(F_{2}). Since not all fibres of ff are isomorphic, dimw(D)=1\dim w(D)=1. Since ww is an algebraic map, if a fibre has dimension 0, its reduction is a finite set. Hence each fibre of ww is finite. Thus each fibre of ff is isomorphic to only finitely many fibres of ff. ∎

All Kodaira fibrations are projective surfaces [6, p. 220], and for any Kodaira fibration there are bounds

2<c12(X)c2(X)<3.\boxed{2<\frac{c_{1}^{2}(X)}{c_{2}(X)}<3}.

The following paragraph is shamelessly copied from [10]: The number ν=c12(X)c2(X)\nu=\frac{c_{1}^{2}(X)}{c_{2}(X)} is an important invariant of Kodaira fibred surfaces, called the slope, that can be seen as a quantitative measure of the non-multiplicativity of the signature. In fact, every product Kodaira surface satisfies ν=2\nu=2; on the other hand, if SS is a Kodaira fibred surface, then Arakelov inequality (see [7]) implies ν(S)>2\nu(S)>2, while Liu inequality (see [28]) yields ν(S)<3\nu(S)<3, so that for such a surface the slope lies in the open interval (2,3)(2,3). The original examples by Atiyah, Hirzebruch and Kodaira have slope lying in (2,2+1/3](2,2+1/3], (see [6, p. 221]), and the first examples with higher slope were given by Catanese and Rollenske in [9] using double Kodaira that satisfy ν(S)=2+2/3\nu(S)=2+2/3. More examples of double Kodaira fibrations where given by Causin and Polizzi in [10] and further explored by Polizzi in [34]. At present it is unknown whether the slope of a Kodaira fibred surface can be arbitrarily close to 33.

J. Jost and S.-T. Yau proved that every deformation of a Kodaira fibration is a Kodaira fibration, see [6, p. 223] and [25].

4. Acknowledgements

We greatly appreciate thorough and comprehensive referee report which contributed to improve the quality of our text. We are grateful to Francesco Polizzi for pointing out necessary corrections to text. We thank Maria Pilar Garcia del Moral Zabala and Camilo las Heras for asking us to write this note about existence and numerical invariants of hyperelliptic fibrations. Ballico is a member of MUR and GNSAGA of INdAM (Italy). Gasparim and Suzuki thank the University of Trento for the support and excellent hospitality during their visit under the research in pairs program of CIRM. Suzuki was supported by Grant 2021/11750-7 São Paulo Research Foundation - FAPESP. Gasparim is a senior associate the Abdus Salam International Centre for Theoretical Physics, Italy.

References

  • [1] J. Alexander, A. Hirschowitz, Un lemme d’Horace différentiel: application aux singularité hyperquartiques de 𝐏5\mathbf{P}^{5}, J. Alg. Geom. 1 (1992) 411–426.
  • [2] J. Alexander, A. Hirschowitz, La méthode d’Horace éclaté: application à l’interpolation en degré quatre, Invent. Math. 107 (1992) 585–602.
  • [3] J. Alexander, A. Hirschowitz, Polynomial interpolation in several variables, J. Alg. Geom. 4 (1995) 201–222.
  • [4] E. Arbarello, M. Cornalba, P. Griffiths, Geometry of Algebraic Curves II with a contribution by J. Harris, Springer, Berlin–Heidelberg (2011).
  • [5] M. A. Barja, L. Stoppino, Slopes of trigonal fibred surfaces and of higher dimensional fibrations, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (5) VIII (2009) 647–658.
  • [6] W. P. Barth, K. Hulek, C. A. Peters, A. Van de Ven, Compact complex surfaces, Second edition, Ergebnisse der Mathematik und ihrer Grenzgebiete, 4 Springer–Verlag, Berlin, 2004.
  • [7] A. Beauville, L’inégalité pg2q4p_{g}\geq 2q-4 pour les surfaces de type général, Appendice à O. Debarre: Inégalités numériques pour les surfaces de type général, Bull. Soc. Math. France 110 n.3 (1982) 319–344.
  • [8] F. Catanese, Kodaira fibrations and beyond: methods for moduli theory, Japan J. Math. 12 (2017) 91–174.
  • [9] F. Catanese, S. Rollenske, Double Kodaira fibrations, J. Reine Angew. Math. 628 (2009) 205–233.
  • [10] A. Causin, F. Polizzi, Surface braid groups, finite Heisenberg covers and double Kodaira fibrations, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) XXII (2021) 1309–1352.
  • [11] I. V. Dolgachev, Classical algebraic geometry: A modern view, Cambridge University Press, Cambridge (2012)
  • [12] D. Eisenbud, Transcanonical embeddings of hyperelliptic curves, J. Pure Appl. Algebra 19 (1980) 77–85.
  • [13] D. Eisenbud, J. Harris, Limit linear series: basic theory, Invent. Math. 85 (1986) 337–371.
  • [14] D. Eisenbud, J. Harris, The Practice of Algebraic Curves, book in preparation.
  • [15] R. Friedman, Algebraic surfaces and holomorphic vector bundles, Springer, New York, 2012.
  • [16] J. P. Griffiths, J. Harris, Principles of Algebraic Geometry, John Wiley & Sons, New York, 1978
  • [17] R. V. Gujar, S. Paul, B. P. Purnaprajna, On the fundamental group of hyperelliptic fibrations and some applications, Invent. Math. 186 (2011) 237–254
  • [18] J. Harris, On the Severi problem, Invent. Math. 84 n.3 (1986) 445–461.
  • [19] J. Harris, I. Morrison, Moduli of curves, Springer, 1998.
  • [20] J. Harris, D. Mumford, On the Kodaira dimension of the moduli space of curves, with an appendix by William Fulton, Invent. Math. 67 (1982) 23–88.
  • [21] R. Hartshorne, Algebraic Geometry, Springer, Berlin, 1977.
  • [22] K. Kodaira, On Compact Complex Analytic Surfaces: I, Ann. Math. (2) 71 n.1 (1960) 1–152.
  • [23] K. Kodaira, On Compact Analytic Surfaces: II, Ann. Math. 77 n.3 (1963) 563–626.
  • [24] K. Kodaira, A certain type of irregular algebraic surfaces, J. Anal. Math. 19 (1967) 207–215.
  • [25] J. Jost, S.-T. Yau, Harmonic mappings and Kähler manifolds, Math. Ann. 262 n.2 (1983) 145–166.
  • [26] A. Kas, On deformations of a certain type of irregular algebraic surface, Amer. J. Math. 90 (1968) 1008–1042.
  • [27] A. Laface, On linear systems of curves on rational scrolls, Geom. Dedicata 90 (2002) 127–144.
  • [28] K. Liu, Geometric height inequalities, Math.Res.Lett. 3 (1996) 693–702.
  • [29] X. Lu, K. Zuo, On the slope of hyperelliptic fibrations with positive relative irregularity, arXiv:1311.7271.
  • [30] L. Martucci, J. F. Morales, D. R. Pacifici, Branes, U-folds and hyperelliptic fibrations, J. High Energ. Phys. 145 (2013).
  • [31] K. Mitsui, Multiple fibers of elliptic fibrations, Kyoto University (2011) 44–54.
  • [32] M. Murakami, Notes on hyperelliptic fibrations of genus 3, I, arXiv:1209.6278.
  • [33] M. Murakami, Notes on hyperelliptic fibrations of genus 3, II, arXiv:1303.5151.
  • [34] F. Polizzi, Diagonal Double Kodaira Structures on Finite Groups, In: Current Trends in Analysis, its Applications and Computation, Cerejeiras, Reissig, Sabadini, Toft, (eds), Trends in Mathematics. Birkhäuser, Cham. (2022) 111–128.
  • [35] J. Sawon, Isotrivial elliptic K3 surfaces and Lagrangian fibrations, arXiv:1406.1233.
  • [36] F. Serrano, Isotrivial Fibred Surfaces, Ann. di Mat. Pura ed App. 171 (1996) 63–81.
  • [37] A. Tannenbaum, Families of algebraic curves with nodes, Compositio Math. 41 (1980), 107–119.
  • [38] R. Treger, Plane curves with nodes, Canad. J. Math. 41 n.2 (1989) 193–212.
  • [39] I. Tyomkin, On Severi varieties on Hirzebruch surfaces Int. Math. Res. Not. 23 (2007) rnm109.
  • [40] G. Xiao, Surfaces fibrèes en courbes de genre deux, LNM 1137 Springer-Verlag, Berlin, 1985.
  • [41] G. Xiao, Irregular families of hyperelliptic curves, Algebraic geometry and algebraic number theory (Tianjin, 1989–1990), Nankai Ser. Pure Appl. Math. Theoret. Phys. 3 World Sci. Publ., River Edge, NJ (1992) 152–156.
  • [42] D. Xie, Z Yu, Hyperelliptic families and 4d4d 𝒩=2\mathcal{N}=2 SCFT, arXiv:2310.02793.

Edoardo Ballico, [email protected]
Department of Mathematics, University of Trento
Trento, Italy

Elizabeth Gasparim, [email protected]
Departamento de Matemáticas, Universidade Católica del Norte
Antofagasta, Chile

Bruno Suzuki, [email protected]
Instituto de Matemática e Estatística, Universidade de São Paulo
São Paulo, Brazil