2109 03189

ISOGENIES OF CERTAIN K3 SURFACES OF RANK 18
NOAH BRAEGER, ADRIAN CLINGHER, ANDREAS MALMENDIER,

arXiv:2109.03189v1 [math.AG] 7 Sep 2021
AND SHANTEL SPATIG
Abstract. We construct geometric isogenies between three types of two-parameter

families of K3 surfaces of Picard rank 18. One is the family of Kummer surfaces asso-
ciated with Jacobians of genus-two curves admitting an elliptic involution, another
is the family of Kummer surfaces associated with the product of two non-isogeneous
elliptic curves, and the third is the twisted Legendre pencil. The isogenies imply the
existence of algebraic correspondences between these K3 surfaces and prove that the
associated four-dimensional Galois representations are isomorphic. We also apply
our result to several subfamilies of Picard rank 19. The result generalizes work of
van Geemen and Top in [60].
In [1], Ahlgren, Ono and Penniston computed the L-series and zeta function, over
the finite field Fp , for certain families of K3 surfaces of Picard rank 19. The K3
families in question are associated with a specific one-parameter elliptic curve fam-
ily Et and include the twisted Legendre pencil Xt , as well as the Kummer surfaces
Kum(Et × Et ). The Ahlgren-Ono-Penniston results imply that two three-dimensional
ℓ-adic Galois representations, associated with Xt and Kum(Et × Et ), respectively, are
isomorphic, where ℓ is any prime number different from p. In turn, the above iso-
morphism determines a Galois invariant class in a certain étale cohomology group of
Xt × Kum(Et × Et ). The Tate conjecture predicts then the existence of an algebraic
cycle on Xt × Kum(Et × Et ), with coefficients in Qℓ , that realizes this Galois invariant
class. The explicit algebraic correspondence determining the predicted cycle was sub-
sequently constructed by van Geemen and Top [60]. This correspondence is obtained
as the graph of a dominant rational map from Xt to Kum(Et ×Et ), defined over a finite
extension of Q(t).
The present article indicates a generalization of the above work. We extend the
above results to certain two-parameter families of K3 surfaces of generic Picard rank
18. The K3 families in question include: Kummer surfaces associated with Jacobians
of genus-two curves admitting an elliptic involution, Kummer surfaces associated with
the product of two elliptic curves, and the two-parameter twisted Legendre pencil.
To begin with, let us briefly introduce some notation. Let V be a smooth projective
variety over Q, with separable closure Q and let GQ = Gal (Q/Q) be the absolute
Galois group. Fix a prime number ℓ and consider the ℓ-adic cohomology groups, i.e.,
the cohomology groups of the base extension VQ of V to Q with coefficients in the
2020 Mathematics Subject Classification. 14J27, 14J28, 14G25.
Key words and phrases. Isogenies, Kummer surfaces.
1
2 NOAH BRAEGER, ADRIAN CLINGHER, ANDREAS MALMENDIER, AND SHANTEL SPATIG
ℓ-adic integers Zℓ , and corresponding scalars then extended to the ℓ-adic numbers
Qℓ . The étale cohomology theory determines then ℓ-adic cohomology groups for the
algebraic variety V , which carry many of the same properties that the usual singular
cohomology groups have. In addition, however, the étale cohomology groups are
representations of GQ . They also satisfy a form of Poincaré duality on non-singular
projective varieties, and a Künneth formula holds. In particular, when V is a non-
1
singular algebraic curve of genus g, Hét is a free Zℓ -module of rank 2g, dual to the
1
Tate module of the Jacobian variety of V . In turn, Hét is isomorphic to the singular
cohomology (of the complex algebraic curve V ) with Zℓ -coefficients. Getting back
to the general V case, for r > 0, a codimension-r subvariety of V , defined over Q,
determines an element of the cohomology group Hét 2r
(VQ , Qℓ (r)) fixed by GQ . Here,
Qℓ (r) denotes the rth Tate twist. In this context, the Tate conjecture states that the
subspace of Galois invariant classes is in fact the Qℓ -vector space of algebraic cycles
on V of codimension r. For more details on this aspect, we refer the reader to the
excellent survey in [58].
In this article, we consider several special families of K3 surfaces. First, the two-
parameter twisted Legendre pencil X is, by definition, given by the family of K3
surfaces that are minimal resolutions of double covers of P2 = P(z1 , z2 , z3 ) branched
over a special union of six lines (and hence over a sextic curve). Specifically, consider
the family of double sextics, given by
(0.1) X∶ y 2 = z1 z2 (z1 − z2 )(z1 − z3 )(z3 − Az2 )(z3 − Bz2 ) ,
where the parameters A, B satisfy A =/ B. The branch locus of a general member is
the configuration of six lines {ℓ1 , . . . , ℓ6 } in P2 such that, up to permutation, the triple
intersections {p123 } = ℓ1 ∩ ℓ2 ∩ ℓ3 and {p145 } = ℓ1 ∩ ℓ4 ∩ ℓ5 consist of two distinct points
p123 =/ p145 , and the configuration is generic otherwise. Notice that the configuration
(P2 ; ℓ1 , . . . , ℓ6 ) is not a Kummer plane, i.e., the general member of X is not a Kummer
surface. We assume then:
1 + λ2 2 1 + λ3 2
(0.2) A=( ) , B=( )
1 − λ2 1 − λ3
with λ2 , λ3 ∈ Q with λ2 , λ3 ∈ P1 /{0, 1, ∞} and λ2 =/ λ±1
3 . In the generic case the Picard
rank of X is 18. We also remind the reader that the Picard group is generated by
(curve) classes defined over Q. Hence, one has an isomorphism of GQ -representations:
(XQ , Qℓ ) ≅ Tℓ ⊕ Qℓ (−1)⊕18 ,
2 (1)
(0.3) Hét
(1)
where Tℓ is a certain four-dimensional ℓ-adic representation.
We also consider Kummer surfaces Kum(A), of generic Picard rank 18, associated
with certain principally polarized abelian surfaces A. If we denote the minus identity
involution denoted by −I, then Kum(A) is, by definition, the minimal resolution
of the quotients A/⟨−I⟩. We shall realize the quotients A/⟨−I⟩ as double quadric
or double sextic surfaces, as well as Jacobian elliptic surfaces. We shall refer to the
minimal resolutions of the corresponding quadratic-twist surfaces as twisted Kummer
surfaces Kum(A)(ε) with twist factor ε.
ISOGENIES OF CERTAIN K3 SURFACES OF RANK 18 3
One choice of abelian surface we shall consider is the product A = E1 × E2 , where

the elliptic curves El , for l = 1, 2, are not mutually isogenous. Specifically, we consider
the (families of) elliptic curves given by the equation
(0.4) El ∶ yl2 zl = xl (xl − zl )(xl − Λl zl ) ,
such that
(λ2 + λ3 )2 − 4λ2 λ3 2(λ2 + λ3 )
(0.5) Λ1 Λ2 = , Λ1 + Λ2 = − .
(1 − λ2 )2 (1 − λ3 )2 (1 − λ2 )(1 − λ3 )
The Picard rank of Kum(E1 × E2 ) is generically 18. Hence, one has an isomorphism
of GQ -representations, namely
( Kum(E1 × E2 )Q , Qℓ ) ≅ Tℓ ⊕ Qℓ (−1)⊕18 ,
2 (2)
(0.6) Hét
(2)
where Tℓ is a four-dimensional ℓ-adic representation. Then, at the level of GQ -
representations, one has the isomorphism:
≅ Hét (E1 , Qℓ ) ⊗ Hét (E2 , Qℓ ) .
(2) 1 1
(0.7) Tℓ
Finally, we consider Kummer surfaces Kum(A) with abelian surface A realized as
the Jacobian Jac (C0 ) of a smooth genus-two curve C0 admitting an elliptic involution.
The curves C0 are given explicitly by
(0.8) C0 ∶ Y 2 = XZ(X − Z) (X − λ2 λ3 Z) (X − λ2 Z) (X − λ3 Z) .
Due to the elliptic involution existing on the genus-two curve C0 , the transcendental
lattice of A = Jac (C0 ) generically has rank four. As GQ -representations, one has an
isomorphism:
( Kum(Jac C0 ) , Qℓ ) ≅ Tℓ ⊕ Qℓ (−1)⊕18 ,
2 (ε) (3)
(0.9) Hét
Q
(3)
where Tℓ is a second four-dimensional ℓ-adic representation and ε is a certain
quadratic-twist factor.
The main result of this article can be stated as follows:
Theorem 0.1. Assume that λ2 , λ3 ∈ Q satisfy λ2 , λ3 ∈/ {0, 1}, λ2 =/ λ±1
3 , and are generic
otherwise. Further assume that Λ1 , Λ2 satisfy
(λ2 + λ3 )2 − 4λ2 λ3 2(λ2 + λ3 )
(0.10) Λ1 Λ2 = , Λ1 + Λ2 = − ,
(1 − λ2 )2 (1 − λ3 )2 (1 − λ2 )(1 − λ3 )
and ε = λ2 λ3 . Then, there are explicit algebraic correspondences
Γ(1,2) ⊂ Kum (E1 × E2 ) × X ,
Γ(2,3) ⊂ X × Kum ( Jac C0 )
(ε)
(0.11) ,
Γ(3,1) ⊂ Kum ( Jac C0 ) × Kum (E1 × E2 ) ,
(ε)
defined over Q, which induce ℓ-adic GQ -representation isomorphisms:

[Γ(i,j) ] ∶
≅
Ð→ Tℓ ,
(i) (j)
(0.12) Tℓ
for i, j ∈ {1, 2, 3} and i =/ j.

We note that the result above contains three special subfamilies, associated with
the non-generic situation when the elliptic curves E1 and E2 are isomorphic, up to a
quadratic twist. The GQ -representations involved are then three-dimensional. The
details on this case are included in Theorem 3.31. One of these specializations is
exactly the case considered in [60]. To our knowledge, the other two cases have not
appeared in the literature. In addition, in Theorem 3.33 we prove analogous results
for two other subfamilies of Picard rank 19 and 18 - one where the elliptic curves are
two-isogenous to each other and one where one of the elliptic curves is constant with
complex multiplication and j-invariant 123 .
The article structure is as follows. In Section 1, we derive normal forms for Ja-
cobians of genus-two curves admitting an elliptic involution, their elliptic quotient-
curves and certain (2, 2)-isogenous abelian surfaces. In Section 2, we construct the
Kummer surfaces associated with the Jacobian varieties of genus-two curves admit-
ting an elliptic involution and the product abelian surfaces of two non-isogeneous
elliptic curves. These Kummer surfaces are explicitly described as quartic hypersur-
faces, double quadric surfaces, double sextic surfaces, and Jacobian elliptic surfaces.
As a side observation, we show that Legendre’s glueing construction, i.e., the con-
struction of a genus-two curve with involution from its elliptic-curve quotients over
a small field of definition, has an exact analogue for quartic Kummer surfaces. This
is discussed in Theorem 2.33. Using the results from Section 1, we are then able
to construct explicit isogenies between the various Kummer surfaces involved; see
Theorem 2.40. In Section 3 we prove that the two-parameter twisted Legendre pen-
cil fits into a Kummer sandwich of dominant rational maps, directly relating it to
the Kummer surfaces from Section 2. This geometrical aspect is presented in Theo-
rem 3.19. The rational maps involved are defined only over a finite field extension of
the field of definition for the given varieties. The varieties involved have (projective)
models defined over the rationals, but one has to choose the right quadratic-twists in
order to obtain non-trivial rational correspondences between them. We achieve this
goal by arranging for the pullbacks between natural generators of the corresponding
H 2,0 -parts in cohomology to be rational. Theorem 3.30 is the main result of this ar-
ticle and proves that the aforementioned four-dimensional ℓ-adic representations – as
obtained from the corresponding transcendental lattices of families of K3 surfaces of
Picard rank 18 – are isomorphic. Finally, Theorems 3.31 and 3.33 discuss subfamilies
of generic Picard rank 19, one of which recovers the main result from [60].
Acknowledgments. The authors would like to thank the referees for their thought-
ful comments and effort towards improving the manuscript. The authors also thank
Dr. Muhammad Arjumand Masood for some helpful discussions about Jacobian ellip-
tic functions during an early stage of this project. N.B. and S.S. would like to acknowl-
edge the support from the Office of Research and Graduate Studies at Utah State
University. A.C. acknowledges support from a UMSL Mid-Career Research Grant.
A.M. acknowledges support from the Simons Foundation through grant no. 202367.
1. Abelian surfaces admitting an elliptic involution

Let C be an irreducible, smooth, projective curve of genus two, defined over the
complex numbers C. Let M be the coarse moduli space of smooth curves of genus
two. We denote by [C] the isomorphism class of C, i.e., the corresponding point in
M. For a smooth genus-two curve C given as a sextic Y 2 = f6 (X, Z) in the weighted
complex projective space WP(1,3,1) = P(X, Y, Z), we send three roots (λ4 , λ5 , λ6 ) to
(0, 1, ∞) to obtain an isomorphic curve in Rosenhain normal form, given by
(1.1) C∶ Y 2 = X Z (X − Z) (X − λ1 Z) (X − λ2 Z) (X − λ3 Z) .
We denote the hyperelliptic involution on C by ıC . The Weierstrass points of C are
the fixed points of ıC , and they are Pi ∶ [X ∶ Y ∶ Z] = [λi ∶ 0 ∶ 1] for i = 1, . . . , 5, and
P6 ∶ [X ∶ Y ∶ Z] = [1 ∶ 0 ∶ 0]. The tuple (λ1 , λ2 , λ3 ) where the roots λi are all distinct
and different from {0, 1, ∞} determines a point in the moduli space of genus-two
curves with six marked Weierstrass points, denoted by M(2). The Jacobian variety
Jac (C) is the moduli space of degree-zero line bundles on C. It is the connected
component of the identity in the Picard group of C, hence an abelian surface. We
refer to a genus-two curve as generic if it is smooth and its Jacobian Jac (C) has no
extra automorphisms.
The Siegel three-fold is the quasi-projective variety of dimension three, obtained
from the Siegel upper half-plane H2 of degree two1 divided by the action of the
modular transformations Γ2 ∶= Sp4 (Z), i.e.,
(1.2) A2 = H2 /Γ2 .
Each τ = ( ττ11 τ12
12 τ22 ) ∈ H2 determines a complex abelian surface A = C2 /Λ obtained
from the lattice Λ = Z2 ⊕ τ Z2 with the period matrix (I2 , τ ) ∈ Mat(2, 4; C). We
consider two abelian surfaces A and A ̃ isomorphic if and only if there is an element
M ∈ Γ2 such that ̃ τ = M(τ ). The canonical principal polarization of A is given by
the positive definite hermitian form h on C2 such that α = Im h(Λ, Λ) ⊂ Z where
α is the Riemann form α(x1 + x2 τ, y1 + y2 τ ) = xt1 ⋅ y2 − y1t ⋅ x2 on Z2 ⊕ τ Z2 . Such
a hermitian form determines the class of a line bundle L → A in the Néron-Severi
lattice NS(A). Alternatively, τ determines a line bundle L with first Chern class
Im(h) ∈ NS(A) ⊂ ⋀2 H 1 (A, Z). A general fact from algebra asserts that one can
always choose a basis of Λ such that α is given by the matrix ( −D 0 D ) with D = ( d1 0 )
0 0 d2
where d1 , d2 ∈ N, d1 , d2 ≥ 0, and d1 divides d2 . For a principal polarization one has
(d1 , d2 ) = (1, 1). Since transformations in Γ2 preserve the Riemann form α, it follows
that the Siegel three-fold A2 is also the set of isomorphism classes of principally
polarized abelian surfaces, i.e., abelian surfaces with a polarization of type (1, 1). We
also define the subgroup Γ2 (2) = {M ∈ Γ2 ∣ M ≡ I mod 2} such that Γ2 /Γ2 (2) ≅ S6
where S6 is the permutation group of six elements representing the permutations of
1By definition H is the set of two-by-two symmetric matrices over C whose imaginary part is
2
positive definite
the Rosenhain roots (λ1 , λ2 , λ3 , 0, 1, ∞). Then, A2 (2) is the three-dimensional moduli
space of principally polarized abelian surfaces with marked level-two structure.
For a principally polarized abelian variety A the line bundle L defining its principal
polarization is ample and satisfies h0 (L) = 1. There exists an effective divisor Θ such
that L = OA (Θ), uniquely defined only up to translations. The divisor Θ ∈ NS(A) is
called a theta divisor associated with the polarization. It is known that the abelian
surface A is not the product of two elliptic curves if and only if Θ is an irreducible
divisor. Torelli’s theorem implies that the map sending a curve C to its Jacobian
Jac(C) is injective and defines a morphism M ↪ A2 . The complement of its image in
A2 corresponds to abelian surfaces obtained as product of two complex elliptic curves.
Moreover, M ↪ A2 lifts to an injective morphism M(2) ↪ A2 (2) between the coarse
moduli spaces of smooth curves of genus two with marked Weierstrass points and
principally polarized abelian surfaces with marked level-two structures, respectively.
1.1. Isogenies of Jacobian surfaces. Translations of the Jacobian A = Jac (C)
by order-two points in A[2] are isomorphisms of the Jacobian and map the set of
two-torsion points to itself. Moreover, a Göpel group is a two-dimensional subspace
G ≅ (Z/2Z)2 of A[2] such that A′ = A/G is again a principally polarized abelian
surface [49, Sec. 23]. The corresponding isogeny Ψ ∶ A → A′ between principally
polarized abelian surfaces has as its kernel G ⩽ A[2] and is called a (2, 2)-isogeny.
Remark 1.1. In general one defines an isogeny of type (n1 , . . . , nk ) with k ≤ 4 be-
tween abelian surfaces to be an isogeny with kernel isomorphic to ⊕ki=1 Z/ni Z. We
will also consider compositions of two (2, 2)-isogenies, leading to isogenies of type
(2, 2, 2, 2) (which is multiplication by two up to automorphisms), (4, 2, 2), and (4, 4).
This ‘calculus’ for compositions of isogenies also leads to Hecke operators related
to algebraic subvarieties on A × A parametrizing pairs (A, A/G) for suitable finite
subgroups G such that A/G is still a principally polarized abelian surface; see work
by Andrianov for example [2]. This generalizes the construction of modular curves
TN = X0 (N) ⊂ A1 × A1 ; see also Remark 3.6.
For the Jacobian of a genus-two curve, every nontrivial two-torsion point is a dif-
ference of Weierstrass points Pi ∈ C for 1 ≤ i ≤ 6. In fact, the sixteen order-two points
of A = Jac (C) are obtained using the embedding of the curve into the connected
component of the identity in the Picard group, i.e., C ↪ Jac (C) ≅ Pic0 (C). We obtain
the 15 elements Pij = [Pi + Pj − 2 P6 ] ∈ A[2] with 1 ≤ i < j < 6 and set P0 = P66 = [0].
For {i, j, k, l, m, n} = {1, . . . , 6}, the group law on A[2] is given by the relations
(1.3) P0 + Pij = Pij , Pij + Pij = P0 , Pij + Pkl = Pmn , Pij + Pjk = Pik .
The space A[2] of two-torsion points admits a symplectic bilinear form, called the
Weil pairing such that the two-dimensional, maximal isotropic subspace of A[2] with
respect to the Weil pairing are the Göpel groups. The Weil pairing is induced by the
pairing
(1.4) ⟨[Pi − Pj ], [Pk − Pl ]⟩ = #{Pi , Pj } ∩ {Pk , Pl } mod 2 .
It is easy to check that there are exactly 15 inequivalent Göpel groups, and they are
of the form
(1.5) {P0 , Pij , Pkl , Pmn }
such that {i, j, k, l, m, n} = {1, . . . , 6}.
For a general genus-two curve C, one may ask whether the (2, 2)-isogenous abelian
surface A′ = A/G satisfies A′ = Jac (C ′ ) for some smooth curve C ′ of genus two. The
geometric moduli relationship between the two curves of genus two was found by
Richelot [53]; see also [9]. If we choose for C a sextic equation Y 2 = f6 (X, Z), then
from any factorization f6 = A ⋅ B ⋅ C into three degree-two polynomials A, B, C one
obtains a new genus-two curve C ′ , given by
(1.6) C′ ∶ ∆ABC ⋅ Y 2 = [A, B] [A, C] [B, C] ,
where we have set [A, B] ⋅ Z = B ∂X A − A ∂X B with ∂X denoting the derivative with
respect to X and ∆ABC is the determinant of (A, B, C) with respect to the basis
X 2 , XZ, Z 2 . Notice that the Richelot construction is guaranteed to work only for a
general curve of genus two. Later we will also consider isogenies from Jac(C0 ) for
genus-two curves C0 with extra involution to the decomposable principally polarized
abelian varieties that are products of two elliptic curves E1 × E2 .
1.2. Humbert surfaces. Sets of abelian surfaces with the same endomorphism ring
form subvarieties within A2 . The endomorphism ring of principally polarized abelian
surface tensored with Q is either a quartic CM field, an indefinite quaternion algebra,
a real quadratic field or in the generic case Q. Irreducible components of the corre-
sponding subsets in A2 have dimensions 0, 1, 2 and are known as CM points, Shimura
curves, and Humbert surfaces, respectively. (Here, we are excluding the endomor-
phism algebras of abelian surfaces which are isogenous to E1 × E2 .) The Humbert
surface H∆ , with invariant ∆, is the space of principally polarized abelian surfaces
admitting a symmetric endomorphism with discriminant ∆. It turns out that ∆ al-
ways is a positive integer satisfying ∆ ≡ 0, 1 mod 4 and uniquely determined H∆ . In
fact, H∆ is the image inside A2 under the projection of the rational divisor associated
with the equation
(1.7) a τ11 + b τ12 + c τ22 + d (τ12
2
− τ11 τ22 ) + e = 0 ,
with integers a, b, c, d, e satisfying ∆ = b2 − 4 a c − 4 d e and τ = ( ττ11
12
τ12
τ22 ) ∈ H2 . The
following was proven by Birkenhake and Wilhelm in [7]:
Theorem 1.2. For δ ∈ N the Humbert surface Hδ2 is the locus of principally polarized
abelian surfaces (A, L) ∈ A2 admitting an isogeny of degree δ 2 , given by
(1.8) Φ∶ (E1 × E2 , OE1 (δ) ⊠ OE2 (δ)) Ð→ (A, L) ,
where OEl (δ) is a line bundle of degree δ on an elliptic curve El for l = 1, 2.
The Humbert surface H1 is irreducible. In order to underscore this point, we shall
refer to it as H1 . It is the image, under the period projection, of the rational divisor
associated to τ12 = 0. The Humbert surface H4 has two irreducible components, one
of which is H1 . We shall denote the second component by H4 . This is the image of

the divisor corresponding to τ11 − τ22 = 0. Moreover, the singular locus of A2 is given
by H4 = H1 ∪ H4 . As analytic spaces, H1 and H4 are each isomorphic to the Hilbert
modular surface
(1.9) ((SL2 (Z) × SL2 (Z)) ⋊ Z2 )/(H × H) .
For a detailed introduction to Siegel modular forms relative to Γ2 , Humbert sur-
faces, and the Satake compactification of the Siegel modular threefold, we refer the
reader to Freitag’s book [20]. Igusa proved [33, 34] that the ring of modular forms is
generated by the Siegel modular forms ψ4 , ψ6 , χ10 , χ12 and by one more cusp form
χ35 of odd weight 35 whose square is the following polynomial [33, p. 849] in the even
generators
χ10 (224 315 χ512 − 213 39 ψ43 χ412 − 213 39 ψ62 χ412 + 33 ψ46 χ312
1
χ235 =
212 39
− 2 ⋅ 33 ψ43 ψ62 χ312 − 214 38 ψ42 ψ6 χ10 χ312 − 223 312 52 ψ4 χ210 χ312 + 33 ψ64 χ312
+ 211 36 37 ψ44 χ210 χ212 + 211 36 5 ⋅ 7 ψ4 ψ62 χ210 χ212 − 223 39 53 ψ6 χ310 χ212
(1.10) − 32 ψ47 χ210 χ12 + 2 ⋅ 32 ψ44 ψ62 χ210 χ12 + 211 35 5 ⋅ 19 ψ43 ψ6 χ310 χ12
+ 220 38 53 11 ψ42 χ410 χ12 − 32 ψ4 ψ64 χ210 χ12 + 211 35 52 ψ63 χ310 χ12 − 2 ψ46 ψ6 χ310
− 212 34 ψ45 χ410 + 22 ψ43 ψ63 χ310 + 212 34 52 ψ42 ψ62 χ410 + 221 37 54 ψ4 ψ6 χ510
− 2 ψ65 χ310 + 232 39 55 χ610 ) .
Hence, the expression Q ∶= 212 39 χ235 /χ10 is a polynomial of degree 60 in the even
generators. Igusa also proved that each Siegel modular form (with trivial character)
of odd weight is divisible by the form χ35 . The following fact is well-known [23]:
Proposition 1.3. The vanishing divisor of the cusp form χ10 is the Humbert surface
H1 , i.e., a period point τ is equivalent to a point with τ12 = 0 relative to Γ2 if and only
if χ10 (τ ) = 0. The vanishing divisor of Q is the Humbert surface H4 , i.e., a period
point τ is equivalent to a point with τ11 = τ22 relative to Γ2 if and only if Q = 0.
It is known that one has χ10 (τ ) = 0 if and only if the principally polarized abelian
surface A is a product of two elliptic curves A = Eτ11 × Eτ22 with the transcendental
lattice TA = H ⊕ H; see [39]. Here, H denotes the lattice Z2 with quadratic form
v ) = 2v1 v2 . Moreover, for Q(τ ) = 0 the transcendental lattice of the corresponding
q(⃗
abelian surface A is given by TA = H ⊕ ⟨2⟩ ⊕ ⟨−2⟩. Here, ⟨m⟩ denotes the rank-one
lattice Zv with q(v) = m. Similarly, the locus of A2 corresponding to abelian surfaces
A with transcendental lattice TA = H ⊕ ⟨4⟩ ⊕ ⟨−4⟩ is a surface H16 ⊂ A2 , which is a
special irreducible component of H16 .
1.3. Components of the Humbert surface H4 (2). The ring of invariants of binary
sextics is generated by the so-called Igusa-Clebsch invariants (I2 , I4 , I6 , I10 ) which
were studied in [47, p. 319] and also in [32, p. 17]. Igusa [33, p. 848] proved that
the relations between the Igusa invariants of a binary sextic Y 2 = f6 (X, Z) defining
a smooth genus-two curve C and the even Siegel modular forms for the associated
principally polarized abelian surface A = Jac (C) with period matrix τ are as follows:
χ12 (τ )
I2 (f6 ) = −23 ⋅ 3
χ10 (τ )
,
I4 (f6 ) = 22 ψ4 (τ ) ,
ψ4 (τ ) χ12 (τ )
(1.11)
23
I6 (f6 ) = − ψ6 (τ ) − 25
χ10 (τ )
,
3
I10 (f6 ) = −214 χ10 (τ ) =/ 0 .
Here, the Igusa invariant I10 is the discriminant of the sextic f6 (X, Z) and we are
using the same normalization as in [39, 43].
1.3.1. Sextics with extra involution. Bolza [8] described the possible automorphism
groups of genus-two curves defined by sextics. In particular, he proved that a sextic
curve Y 2 = f6 (X, Z) defining a genus-two curve C0 admits an elliptic involution if
and only if Q(τ ) = 0. Moreover, he proved that one can always represent this extra
involution as [X ∶ Y ∶ Z] ↦ [−X ∶ Y ∶ Z]. It follows that the smooth sextic curve C0
with extra involution can be brought into the normal form given by
(1.12) C0 ∶ Y 2 = X 6 + s1 X 4 Z 2 + s2 X 2 Z 4 + Z 6 .
One uses Equations (1.11) and Equation (1.10) to check that Q = 212 39 χ235 /χ10 always
vanishes for the genus-two curve C0 .
Given the elliptic involution [X ∶ Y ∶ Z] ↦ [−X ∶ Y ∶ Z] on C0 , its composition with
the hyperelliptic involution defines a second elliptic involution. Thus, the elliptic
involutions on C0 come naturally in pairs. The two involutions define two elliptic
subfields of degree two for the function field of C0 . We introduce the elliptic curves El
in P2 = P(xl , yl , zl ) for l = 1, 2, given by
(1.13) E1 ∶ y12 z1 = x31 + s2 x21 z1 + s1 x1 z12 + z13 , E2 ∶ y22 z2 = x32 + s1 x22 z2 + s2 x2 z22 + z23 ,
where E1 and E2 have the j-invariants j1 = j(E1 ) and j2 = j(E2 ) with
28 (3s1 − s22 ) 28 (3s2 − s21 )
3 3
(1.14) j1 = j2 =
4(s31 + s32 ) − (s1 s2 )2 − 18s1 s2 + 27 4(s31 + s32 ) − (s1 s2 )2 − 18s1 s2 + 27
, ,
respectively. Here, we use the standard normalization of the j-invariant where the
square torus with the complex structure i satisfies j = 1728 = 123 . The degree-two
quotient maps πEl ∶ C0 → El associated with the involutions are given by
(1.15) πE1 ∶ C0 Ð→ E1 , [X ∶ Y ∶ Z] ↦ [x1 ∶ y1 ∶ z1 ] = [XZ 2 ∶ Y ∶ X 3 ] ,
and
(1.16) πE2 ∶ C0 Ð→ E2 , [X ∶ Y ∶ Z] ↦ [x2 ∶ y2 ∶ z2 ] = [X 2 Z ∶ Y ∶ Z 3 ] ,
respectively, for XZ =/ 0.
At this point a diagram of the (Z/2Z)2 -covering might be helpful. Assume that
the genus-two curve C0 is given by Y 2 = (X 2 − aZ 2 )(X 2 − bZ 2 )(X 2 − cZ 2 ) so that the
C0 ◆◆
πE1 ♣♣♣♣♣ ◆◆◆ π
◆◆◆E2
♣♣♣ ◆◆◆
♣♣ ◆'
w ♣
E1 = C0 /⟨1 ⟩ E = C /⟨ ⟩
♣
P(X, Z) 2 0 2
♣♣ ◆◆◆
♣♣♣ ◆◆◆
♣♣♣ ◆◆◆
x ♣♣ ◆◆&
P(x1 , z1 ) m 1 P(x2 , z2 )
♣
≅
[x1 ∶z1 ]=[z2 ∶x2 ]
Figure 1. (Z/2Z)2 -covering for the genus-two curve C0
Weierstrass points are again paired by a change of sign. On C0 we have the involu-
tions 1 (X, Y, Z) = (X, Y, −Z) and 2 (X, Y, Z) = (X, −Y, −Z), and the corresponding
quotients are E1 ∶ y12 z1 = (a2 x1 − z1 )(b2 x1 − z1 )(c2 x1 − z1 ) and E2 ∶ y22 z2 = (x2 − a2 z2 )(x2 −
b2 z2 )(x2 −c2 z2 ), respectively. We then obtain the diagram in Figure 1. The five points
on P(x1 , z1 ) that determine the cover maps are {[0 ∶ 1], [a2 ∶ 1], [b2 ∶ 1], [c2 ∶ 1], [1 ∶ 0]}
(and similar on P(x2 , z2 )).
Let us state a closing proposition summarizing the main facts in this section:
Proposition 1.4. A point M2 ⊂ A2 lies in H4 if and only if the associated genus-
two curve can be described as in Equation (1.12). In this case Figure 1 gives the
corresponding isogeny.
1.3.2. Pringsheim decomposition. Next we look at the irreducible components of the
inverse image of H4 in A2 (2) which we denote by H4 (2). We start with a smooth
genus-two curve C in Rosenhain form given by Equation (1.1). We use the aforemen-
tioned relations between the Igusa invariants and the Siegel modular forms to expand
the generators of the ring of modular forms in terms of the Rosenhain roots. We
obtain
(1.17)
222 χ235 (τ )
= (λ1 − λ2 λ3 ) (λ2 − λ1 λ3 ) (λ3 − λ1 λ2 )
2 2 2
χ10 (τ )
− 4
× (λ1 − λ2 − λ3 + λ2 λ3 ) ( − λ1 + λ2 − λ3 + λ1 λ3 ) ( − λ1 − λ2 + λ3 + λ1 λ2 )
2 2 2
× (λ1 λ2 + λ1 λ3 − λ2 λ3 − λ1 ) (λ1 λ2 + λ2 λ3 − λ1 λ3 − λ2 ) (λ1 λ3 + λ2 λ3 − λ1 λ2 − λ3 )

2 2 2
⎧
⎪
⎪ (λ1 λ2 − λ1 λ3 − λ1 + λ3 ) (λ1 λ3 − λ2 λ3 − λ1 + λ3 ) (λ1 λ2 − λ2 λ3 − λ1 + λ2 )
2 2 2
× ⎨
⎪
⎪
⎩ (λ1 λ2 − λ1 λ3 + λ1 − λ2 ) (λ1 λ3 − λ2 λ3 + λ2 − λ3 ) (λ1 λ2 − λ2 λ3 − λ2 + λ3 ) .
2 2 2
The vanishing divisor of Equation (1.17) defines 15 components in A2 (2) = H2 /Γ2 (2)
of discriminant ∆ = 4. Notice that each line in Equation (1.17) is arranged to be
invariant under permutations of the three roots λ1 , λ2 , λ3 . Pringsheim proved the
following statement2 in [52]:
2We corrected two minor typos in the statement of the main theorem.
Proposition 1.5 (Pringsheim). There are exactly 15 components of H4 (2) in A2 (2),

given in Table 1. Each of the component is equivalent to τ11 = τ22 relative to the mod-
ular group Γ2 . Moreover, there is a transposition of the six roots (λ1 , λ2 , λ3 , 0, 1, ∞)
in Γ2 /Γ2 (2) ≅ S6 that permutes each pair of components.
Remark 1.6. In Table 1 for each entry the equation of a rational divisor in H2 is
given whose projection to H2 /Γ2 (2) realizes a component in Proposition 1.5. The
corresponding vanishing divisor in terms of the Rosenhain roots is given as well. The
group names (I, II, III, IV) were introduced by Pringsheim, and they are colored to
match the corresponding terms in Equation (1.17).
I
2
2τ12 + τ11 τ22 − τ12 =0 λ1 = λ2 λ3
II
τ11 + 2τ12 = 0 λ1 − λ2 = λ3 (1 − λ2 )
τ11 + 2τ12 − (τ11 τ22 − τ12
2
)=0 λ1 (1 − λ3 ) = λ2 (λ1 − λ3 )
III
2τ12 − τ22 = 0 λ3 = λ1 λ2
2τ12 − τ22 + (τ11 τ22 − τ12
2
)=0 λ2 = λ1 λ3
τ11 − τ22 = 0 λ1 − λ2 = λ2 (λ1 − λ3 )
τ11 − τ22 + (τ11 τ22 − τ12
2
)=0 λ1 − λ3 = λ3 (λ1 − λ2 )
IV
2τ12 = 1 λ2 − λ3 = −λ1 (1 − λ2 )
2τ12 + τ22 = 1 λ3 (1 − λ2 ) = −λ1 (λ2 − λ3 )
τ11 + 2τ12 = 1 λ2 − λ3 = λ1 (1 − λ3 )
τ11 + 2τ12 + τ22 − (τ11 τ22 − τ12
2 )=1 λ1 (λ2 − λ3 ) = λ2 (1 − λ3 )
2
4τ12 + 3(τ11 τ22 − τ12 )=1 λ2 − λ3 = λ2 (λ1 − λ3 )
2
4τ12 + τ22 + 3(τ11 τ22 − τ12 )=1 λ1 − λ3 = λ1 (λ2 − λ3 )
τ11 + 4τ12 + 3(τ11 τ22 − τ12 ) = 1
2
λ2 − λ3 = −λ3 (λ1 − λ2 )
2
τ11 + 4τ12 + τ22 + 2(τ11 τ22 − τ12 )=1 λ1 − λ2 = −λ1 (λ2 − λ3 )
Table 1. Pringsheim components of H4
1.3.3. Normal forms with level-two structure. We will use component (I) in Proposi-
tion 1.5 to construct a smooth genus-two curve in Rosenhain normal form admitting
an elliptic involution. We consider the smooth genus-two curve C0 , given by
(1.18) C0 ∶ Y 2 = XZ (X − Z) (X − λ2 λ3 Z) (X − λ2 Z) (X − λ3 Z) ,
with a discriminant given by
(1.19) λ62 λ63 (λ2 − 1)4 (λ3 − 1)4 (λ2 λ3 − 1)4 (λ2 − λ3 )2 .
Equation (1.18) makes all Weierstrass points rational and moves them into ‘standard’
position, which is equivalent to choosing one of the 15 components in Proposition 1.5.
In Section 1.3.4 we will also consider another normal form for C0 that is based on the
sextic with elliptic involution in Equation (1.12).
Let us denote by Λl ∈ P1 /{0, 1, ∞} the modular parameter for the elliptic curves El
in P2 = P(xl , yl , zl ) for l = 1, 2 defined by the Legendre normal form
(1.20) El ∶ yl2 zl = xl (xl − zl )(xl − Λl zl ) ,
with the hyperelliptic involution given by ıEl ∶ [xl ∶ yl ∶ zl ] ↦ [xl ∶ −yl ∶ zl ]. We have
the following:
Lemma 1.7. The function field of the smooth genus-two curve C0 contains the sub-
fields given by the function fields of the elliptic curves El for l = 1, 2 if
(λ2 + λ3 )2 − 4λ2 λ3 2(λ2 + λ3 )
Λ1 Λ2 = Λ1 + Λ2 = −
(1 − λ2 )2 (1 − λ3 )2 (1 − λ2 )(1 − λ3 )
(1.21) , .
Proof. The genus-two curve in Bolza normal form in Equation (1.12) is isomorphic
to Rosenhain curve in Equation (1.18) if we set
(µ2 ν + 1)(µ − ν) (µ2 ν − 1)(µ − ν)
λ2 = λ3 =
(µ2 ν − 1)(µ + ν) (µ2 ν + 1)(µ + ν)
(1.22) , ,
and
1 1 1
(1.23) s 1 = µ2 + ν 2 + , s2 = + 2 + µ2 ν 2 ,
µ2 ν 2 µ 2 ν
for µ, ν ∈ C× such that (µ2 − ν 2 )(µ4 ν 2 − 1) =/ 0. This statement is proved by computing
the Igusa-Clebsch invariants, using the normalization in [39, 43]. Denoting the Igusa-
Clebsch invariants of the genus-two curve in Equation (1.18) and Equation (1.12) by
[I2 ∶ I4 ∶ I6 ∶ I10 ] ∈ WP(2,4,6,10) and [I2′ ∶ I4′ ∶ I6′ ∶ I10
′
], respectively, one checks that
(1.24) [I2 ∶ I4 ∶ I6 ∶ I10 ] = [s2 I2′ ∶ s4 I4′ ∶ s6 I6′ ∶ s10 I10
′
] = [I2′ ∶ I4′ ∶ I6′ ∶ I10
′
]
with s ∈ C× . The remainder of the proof follows from computing the j-invariants
for the curves in Equation (1.20) and comparing them with Equations (1.14). The
solution of Equations (1.21), up to interchanging Λ1 ↔ Λ2 , is then given by
−µ2 + ν 2 −µ4 ν 2 + µ2 ν 4
(1.25) Λ1 = , Λ2 = ,
µ4 ν 4 − µ2 µ2 ν 4 − 1
and j(E1 ) = j1 and j(E2 ) = j2 in Equations (1.14).
We have the following:
Proposition 1.8. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 , and the mod-
uli of the curves C0 and El in Equations (1.18) and (1.20) satisfy Equations (1.21).
Quotient maps πEl ∶ C0 → El for l = 1, 2 are given by
(1.26) πEl ∶ C0 Ð→ El , [X ∶ Y ∶ Z] ↦ [xl ∶ yl ∶ zl ] for l = 1, 2,
with
(1.27) [xl ∶ yl ∶ zl ] = [r(X − λ2 Z)(X − λ3 Z) XZ ∶ (X − (−1)l qZ) Y ∶ r 3 X 2 Z 2 ] ,
for XZ =/ 0, and [xl ∶ yl ∶ zl ] = [1 ∶ 0 ∶ 0] otherwise. Here, q, r are square roots of

q 2 = λ2 λ3 and r 2 = (1 − λ2 )(1 − λ3 ), respectively. The elliptic involutions l on C0 ,
given by
(1.28) l ∶ [X ∶ Y ∶ Z] ↦ [λ2 λ3 Z ∶ (−1)l+1 λ2 λ3 qY ∶ X] ,
satisfy πEl ○ l = πEl for l = 1, 2.
Proof. The genus-two curve C0 in Equation (1.18) is assumed to be smooth. Thus,
one must have λ2 , λ3 ∈ P1 /{0, 1, ∞} and λ2 =/ λ±1
3 . The remainder of the proof follows
by explicit computation. We note that a choice of square root for r is changed by
composing πEl with the elliptic involution on El . The choice of square root for q is
changed by interchanging πE1 ↔ πE2 .
In the context of hyperelliptic period integrals, the construction of the rational
double covers is known as the (classical) Jacobi reduction of a genus-two curve with
elliptic involution [45]. We note that we are using component (I) in Proposition 1.5
to construct smooth normal forms for the genus-two curve in Rosenhain normal form
and its elliptic subfields representing a specific divisor of H4 (2) ⊂ A2 (2):
Remark 1.9. Using a covering space of component (I) in Proposition 1.5, given by
the set of tuples (k1 , k2 ) with λl = kl2 for l = 2, 3 and λ1 = (k2 k3 )2 , we obtain for
the modular parameters of the elliptic-curve quotients El in Lemma 1.7 the following
algebraic solutions of Equation (1.21):
(k2 − k3 )2 (k2 + k3 )2
Λ1 = − Λ2 = −
(1 − k22 )(1 − k32 ) (1 − k22 )(1 − k32 )
(1.29) , .
A change of square roots (k2 , k3 ) ↦ (±k2 , ±k3 ) leaves Equations (1.29) invariant,
whereas the change (k2 , k3 ) ↦ (±k2 , ∓k3 ) interchanges Λ1 and Λ2 . This means that
the elliptic curves El with level-two structure can be constructed over Q(k2 , k3 ) as a
finite field extension of Q(Λ1 , Λ2 ).
1.3.4. Legendre’s gluing construction. In this section we construct an explicit model
for C0 over Q(Λ1 , Λ2 ) for elliptic curves in Legendre normal form (1.20) with modular
parameters Λ1 , Λ2 . This is often referred to as Legendre’s gluing method.
The Weierstrass points Pi of the curve C0 in Equation (1.18), given by
P1 = [λ2 λ3 ∶ 0 ∶ 1] , P2 = [λ2 ∶ 0 ∶ 1] , P3 = [λ3 ∶ 0 ∶ 1] ,
P4 = [0 ∶ 0 ∶ 1] , P5 = [1 ∶ 0 ∶ 1] , P6 = [1 ∶ 0 ∶ 0] ,
(1.30)
are the fixed points of the hyperelliptic involution ıC0 . The two elliptic involutions l
in Equation (1.28) for l = 1, 2 (each) pairwise interchange the Weierstrass points, i.e.,
(1.31) l ∶ P2 ↔ P3 , P1 ↔ P5 , P4 ↔ P6 .
The points Ql,± , given by
Q1,± = [−q ∶ ±iq 2 (k2 k3 + 1)(k2 + k3 ) ∶ 1] ,
3
(1.32)
Q2,± = [ q ∶ ± q 2 (k2 k3 − 1)(k2 − k3 ) ∶ 1] ,
3
are the ramification points of πEl in Equation (1.26) for l = 1 and l = 2, respec-
tively. The involution 1 fixes the points Q1,+ and Q1,− , and interchanges the points
Q2,+ ↔ Q2,− . An analogous statements holds for the involution 2 . The hyperelliptic
involution of the curve C0 interchanges the elements of the two pairs simultaneously,
i.e., ıC0 ∶ Ql,+ ↔ Ql,− for l = 1, 2.
The pairs {P2 , P3 }, {P1 , P5 }, {Q2,+ , Q2,− }, and {P5 , P6 } are mapped by πE1 to the
two-torsion points p0 ∶ [x1 ∶ y1 ∶ z1 ] = [0 ∶ 0 ∶ 1], p1 ∶ [1 ∶ 0 ∶ 1], pΛ1 ∶ [Λ1 ∶ 0 ∶ 1], and the
identity p∞ ∶ [1 ∶ 0 ∶ 0] in E1 [2]. Similarly, the pairs {P2 , P3 }, {P1 , P5 }, {Q1,+ , Q1,− },
and {P5 , P6 } are mapped by πE2 to the two-torsion points p0 ∶ [x2 ∶ y2 ∶ z2 ] = [0 ∶ 0 ∶ 1],
p1 ∶ [1 ∶ 0 ∶ 1], pΛ2 ∶ [Λ2 ∶ 0 ∶ 1], and the identity p∞ ∶ [1 ∶ 0 ∶ 0] in E2 [2].
We associate with the branch locus of πE1 and πE2 the effective divisor of degree two
B1 = [πE1 (Q1,+ )+πE1 (Q1,− )] on E1 and B2 = [πE2 (Q2,+ )+πE2 (Q2,− )] on E2 , respectively.
One checks that for πE1 (Q1,± ) one has [x1 ∶ z1 ] = [Λ2 ∶ 1], and for πE2 (Q2,± ) one has
[x2 ∶ z2 ] = [Λ1 ∶ 1]. Because of Pic0 (El ) ≅ El the line bundle OEl (Bl ) associated with
the branch locus Bl is equivalent to the line bundle OEl (2pα ) if and only if
(1.33) πEl (Ql,+ ) ⊕ πEl (Ql,− ) = 2pα = pα ⊕ pα ,
where ⊕ refers to the addition with respect to the elliptic curve group law on El .
Using the properties πEl in Equation (1.26) one finds that
(1.34) πEl (Ql,+ ) ⊕ πEl (Ql,− ) = 0 ,
and pα ∈ El [2] is a two-torsion point. Thus, there are four line bundles Ll → El such
that L⊗2
l ≅ OEl (Bl ), namely Ll ≅ OEl (pα ). It follows h (El , Ll ) = 1 by the Riemann-
0
Roch theorem. Thus, the preimage of the vanishing locus of a section of Ll → El

determines a unique double cover pl ∶ C0 → El . The composition on C0 of the elliptic
involution with the hyperelliptic involution defines a second elliptic involution, and a
second elliptic-curve quotient is obtained. Thus, the data (El , Ll , Bl ) for either l = 1 or
l = 2 determines the curve C0 uniquely. As we have seen, this data is equivalent to an
(unordered) pair {Λ1 , Λ2 } of modular parameters Λ1 , Λ2 ∈ P1 /{0, 1, ∞} with Λ1 =/ Λ2 .
Following Serre’s explanation [54, Sec. 27] of Legendre’s gluing method we obtain:
Proposition 1.10. Assume that Λ1 , Λ2 ∈ P1 /{0, 1, ∞} satisfy Λ1 =/ Λ2 , and the moduli
of C0 and El in Equations (1.18) and (1.20) satisfy Equations (1.21). The smooth
genus-two curve, given by
C̃0 ∶ Y 2 = (X 2 − Z 2 ) (X 2 − Z ) (X 2 − Z ),
Λ1 2 1 − Λ1 2
(1.35)
Λ2 1 − Λ2
is isomorphic to C0 over a finite field extension of Q(λ2 , λ3 ). The curve is also iso-
morphic to each of the curves obtained by replacing {Λ1 , Λ2 } in Equation (1.35) by
one of the following pairs:
(1.36) { }, { }, { , }, { } , {1 − Λ1 , 1 − Λ2 }.
1 1 Λ1 − 1 Λ2 − 1 1 1 Λ1 Λ2
, , ,
1 − Λ1 1 − Λ2 Λ1 Λ2 Λ1 Λ2 Λ1 − 1 Λ2 − 1
Proof. Using Remark 1.9 and denoting the Igusa-Clebsch invariants of any two pairs
genus-two curves in the statement of Proposition 1.10 by [I2 ∶ I4 ∶ I6 ∶ I10 ] ∈ WP(2,4,6,10)
and [I2′ ∶ I4′ ∶ I6′ ∶ I10

′
], respectively, one checks that
(1.37) [I2 ∶ I4 ∶ I6 ∶ I10 ] = [s2 I2′ ∶ s4 I4′ ∶ s6 I6′ ∶ s10 I10
′
] = [I2′ ∶ I4′ ∶ I6′ ∶ I10
′
]
√
with s ∈ Q(λ2 , λ3 , λ2 λ3 ).
The anharmonic group is the group acting on the Λ-line of an elliptic curve in
Legendre form (which is A1 (2)) that is generated by the transformations Λ ↦ 1 −
Λ, Λ/(Λ − 1). The six pairs, given by {Λ1 , Λ2 } and Equation (1.36), form an orbit
under the diagonal action of the anharmonic group. We have the following:
Corollary 1.11. Assume that Λ1 , Λ2 ∈ P1 /{0, 1, ∞} satisfy
Λ1 /∈ {Λ2 , , 1 − Λ2 } .
1 Λ2 − 1 1 Λ2
(1.38) , , ,
1 − Λ2 Λ2 Λ2 Λ2 − 1
Using {Λ1 , Λ2 } and the pairs
{Λ1 , } , {Λ1 , } , {Λ1 , } , {Λ1 , } , {Λ1 , 1 − Λ2 },
1 Λ2 − 1 1 Λ2
(1.39)
1 − Λ2 Λ2 Λ2 Λ2 − 1
one obtains six smooth genus-two curves C̃0 admitting an elliptic involution whose
quotients (E1 , E2 ) are pairwise isomorphic and are given by Equations (1.20). Gener-
ically, the six curves are non-isomorphic and satisfy Jac (C̃0 ) ∈ H4 .
Proof. The proof follows from an explicit computation similar to the one in the proof
of the previous proposition.
1.3.5. Relation to Göpel groups. Since the Picard group Pic0 (C0 ) ≅ Jac (C0 ) consists
of elements of the form [P + Q − 2P6 ] for P, Q ∈ C0 , an isogeny Ψ of abelian surfaces
is defined by setting
(1.40) Ψ ∶ Jac (C0 ) Ð→ E1 × E2 , [P + Q − 2P6 ] ↦ (πE1 (P ) ⊕ πE1 (Q), πE2 (P ) ⊕ πE2 (Q)) .
It follows from the construction of the line bundles Ll → El in Section 1.3.4 that
the line bundle L → A = Jac (C0 ) associated with the principal polarization of A
satisfies L ≅ Ψ∗ (L1 ⊠ L2 ). The elliptic involutions l on C0 extend to involutions on
the Jacobian A = Jac (C0 ) that coincide on A[2]. We denote this induced involution
on A[2] by  ∶ A[2] → A[2].
Given the marking (P1 , . . . , P6 ) = (λ1 = λ2 λ3 , λ2 , λ3 , 0, 1, ∞) of the Weierstrass
points for the curve C0 as before, we have the following:
Proposition 1.12. In the situation above, the kernel of Ψ ∶ A = Jac (C0 ) → E1 × E2 is
given by the Göpel group
(1.41) ker Ψ = {P0 , P15 , P23 , P46 } ≅ (Z/2Z)2 ⊂ A[2] .
In particular, Ψ is a (2, 2)-isogeny and A′ = A/ ker Ψ ≅ E1 × E2 with TA′ = H ⊕ H.
Proof. The induced actions of the two elliptic involutions l on Jac (C0 ) fix the divisors
P15 , P23 , P46 , and [Ql,+ + Ql,− − 2P6 ]. It follows from the explicit formulas for πEl in
Proposition 1.8 that [P + Q − 2P6 ] ∈ ker Ψ if and only if P, Q ∈ A[2] and l (P ) = Q
for l = 1, 2. For A′ we have χ10 = 0 whence A′ is a product of two elliptic curves with
the transcendental lattice TA′ = H ⊕ H; see [39].
The remaining Göpel groups fall within two cases: the first case consists of six Göpel
groups such that the involution  fixes one non-trivial element and interchanges the
other two. These groups are given by
{P0 , P12 , P35 , P46 } , {P0 , P13 , P25 , P46 } , {P0 , P14 , P23 , P56 } ,
(1.42)
{P0 , P16 , P23 , P45 } , {P0 , P15 , P24 , P36 } , {P0 , P15 , P26 , P34 } .
By construction, each of them contains one non-trivial element from the special Göpel
group {P0 , P15 , P23 , P46 } that yields the quotient E1 × E2 . We have the following:
Proposition 1.13. For each Göpel group G in Equation (1.42) the abelian surface
A′ = A/G satisfies A′ ≅ Jac (C0′ ) for some smooth genus-two curve C0′ admitting an
elliptic involution. In particular, we have Jac (C0′ ) ∈ H4 and TA′ = H ⊕ ⟨2⟩ ⊕ ⟨−2⟩. The
six genus-two curves correspond to the curves in Corollary 1.11 under (2, 2)-isogeny.
Proof. The proof follows by constructing the Richelot curve in Equation (1.6) and
checking that Q = 0 in Proposition 1.3 for each Göpel subgroup Gn with n = 1, . . . , 6,
in Equation (1.42). For A′ we have Q = 0 whence the transcendental lattice of A′ is
TA′ = H ⊕ ⟨2⟩ ⊕ ⟨−2⟩.
For the elliptic curve El in Equation (1.20), there are three order-two points that
can be used to construct a two-isogenous elliptic curve. Using the order-two point
[xl ∶ yl ∶ zl ] = [0 ∶ Λl ∶ 1], one easily obtains an elliptic curves El′ in P2 = P(Xl , Yl , Zl )
for l = 1, 2, given by
(1.43) El′ ∶ Yl2 Zl = Xl (Xl2 + 2(1 − 2Λl )Xl Zl + Zl2 ) ,
together with an isogenies χEl ∶ El → El′ , and its dual two-isogenies χ′El . Assuming
Equations (1.21), the two-isogenous elliptic curves can be brought into Legendre
normal form, but only over the finite field extension Q(k1 , k2 ). In fact, the elliptic
curve El′ is isomorphic over Q(k1 , k2 ) to the quadratic-twist curve
(1.44) El′ ∶ Yl2 Zl = δ′ Xl (X1 − Z1 ) (X1 − Λ′1 Z1 ) ,
where we have set
(k2 − 1)2 (k3 + 1)2 (k2 + 1)2 (k3 + 1)2
Λ′1 = Λ′2 = δ′ = −(1 − k22 )(1 − k32 ) .
(k2 + 1)2 (k3 − 1)2 (k2 − 1)2 (k3 − 1)2
(1.45) , ,
A change of square roots (k2 , k3 ) ↦ (±k2 , ±k3 ) leaves Equations (1.45) invariant or acts
as the modular transformation Λ′l ↦ 1/Λ′l , whereas the change (k2 , k3 ) ↦ (∓k2 , ±k3 ) in
Equations (1.45) interchanges Λ′1 and Λ′2 (up to a modular transformation). Applying
Proposition 1.10 one obtains the common double cover
C̃0′ ∶
Λ′1 2 1 − Λ′1 2
(1.46) Y 2 = (X 2 − Z 2 ) (X 2 − Z ) (X 2
− Z ).
Λ′2 1 − Λ′2
Proposition 1.14. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1 / −1,
3 and λ2 =
and C0 is given by Equation (1.18) with A = Jac (C0 ). The genus-two curve
(1.47) C0′ ∶ Y 2 = XZ(X − Z)(X − µ2 µ3 Z)(X 2 − (µ2 + µ3 )XZ + µ2 µ3 Z 2 ) ,
is smooth, admits an elliptic involution, and satisfies A′ ≅ A/G for A′ = Jac (C0′ ) and
G = {P0 , P12 , P35 , P46 }. Here, µ2 , µ3 are given by
λ2 λ3 + λ2 − λ3 − 1 + 2d
µ2 = ,
λ2 λ3 + λ2 − λ3 − 1 − 2d
(1.48) (λ2 λ3 + λ2 − λ3 − 1 + 2d)((1 + λ2 )2 k3 + (1 − λ2 )d − 2(1 + λ3 )λ2 )
µ3 =
(λ2 λ3 + λ2 − λ3 − 1 − 2d)((1 + λ2 )2 k3 − (1 − λ2 )d − 2(1 + λ3 )λ2 )
,
with k32 = λ3 and d2 = (λ2 − λ3 )(λ2 λ3 − 1). In particular, the curve C0′ is isomorphic to
C̃0′ in Equation (1.46) over a finite field extension of Q(λ2 , λ3 ).
Remark 1.15. In Equations (1.48) the sign of k3 and d can be chosen indepen-
dently. A change (k3 , d) ↦ (∓k3 , ∓d) acts on the Rosenhain roots as (µ2 , µ3 ) ↦
(1/µ2 , 1/µ3 ). The sign change (k3 , d) ↦ (±k3 , ∓d) acts on the Rosenhain roots as
(µ2 , µ3 ) ↦ (µ2 , µ22 /µ3) or (µ2 , µ3 ) ↦ (1/µ2, µ3 /µ22 ). Isomorphic curves are also ob-
tained by interchanging λ2 ↔ λ3 , k2 ↔ k3 , and changing d ↦ id in Equations (1.48)
(while assuming λ3 =/ −1 instead of λ2 =/ −1).
Proof. The result is obtained by applying the results of [16, 17] in the case λ1 = λ2 λ3 .
In fact, [16, Prop. 2.1] yields the curve
(1.49) Y 2 = (X − 2(1 + λ22 )λ3 Z)(c2 X 2 + c1 XZ + c0 Z 2 )((X + (1 + λ22 )λ3 Z) − 36λ22 λ23 Z 2 )Z ,
2
with
c2 = 1 + λ3 − 2k3 ,
c1 = 2λ3 (λ22 − 6λ2 + 1)(λ3 + 1) + 8(1 + λ22 )λ3 k3 ,
c0 = λ23 (λ42 + 24λ32 − 34λ22 + 24λ2 + 1)(λ3 + 1) + 2λ23 (λ22 − 5)(5λ22 − 1)k3 .
If in Equation (1.49) one substitutes
X = 3(1 + λ2 )2 λ3 X ′ − (1 + 6λ2 + λ22 )λ3 Z ′ , Y = 3 2 (1 + λ2 )5 (1 − k3 )Y ′ , Z = Z′ ,
5
one obtains the curve

(1.50) (Y ′ )2 = X ′ Z ′ (X ′ − Z ′ )(X ′ − λ′2 λ′3 Z ′ )((X ′ )2 − (λ′2 + λ′3 )X ′ Z ′ + λ′2 λ′3 (Z ′ )2 ) ,
where λ′2 , λ′3 satisfy
(1 − λ2 )2 (1 + λ3 )
λ′2 + λ′3 = 2 (1 − ),
4λ2
λ′2 λ′3 =
(1 + λ2 )2 (1 + λ2 )2 (1 + λ3 − 2k3 )
(1.51) ,
with λ3 = k32 . Introducing new moduli µ2 , µ3 , given by

1 − µ2 (1 − µ2 )µ3 1 − λ′2 (1 − λ′2 )λ′3
(λ′2 , λ′3 ) = ( ) ⇔ (µ2 , µ3 ) = ( ),
1 − µ3 (1 − µ3 )µ2 1 − λ′3 (1 − λ′3 )λ′2
(1.52) , ,
one obtains an isomorphic curve over Q(µ2 , µ3 ) given by Equation (1.47). Because of
λ2 =/ λ±1
3 we have d =/ 0. Since we also assumed λ2 =/ −1, we have µ2 , µ3 ∈ P1 /{0, 1, ∞}
and µ2 =/ µ±1
3 and the curve is smooth.
Denoting the Igusa-Clebsch invariants of the genus-two curves in Equation (1.46)
and Equation (1.50) by [I2 ∶ I4 ∶ I6 ∶ I10 ] ∈ WP(2,4,6,10) and [I2′ ∶ I4′ ∶ I6′ ∶ I10
′
], respec-
tively, one checks that
(1.53) [I2 ∶ I4 ∶ I6 ∶ I10 ] = [s2 I2′ ∶ s4 I4′ ∶ s6 I6′ ∶ s10 I10
′
] = [I2′ ∶ I4′ ∶ I6′ ∶ I10
′
]
with s ∈ Q(k2 , k3 ).
The last case of Göpel groups to consider consists of four pairs that are pairwise
interchanged by the action of the involution . These groups are given by
{P0 , P12 , P34 , P56 }, {P0 , P12 , P36 , P45 }, {P0 , P13 , P24 , P56 }, {P0 , P13 , P26 , P45 },
↔
↔
(1.54) ∶
{P0 , P14 , P26 , P35 }, {P0 , P16 , P24 , P35 }, {P0 , P14 , P25 , P36 }, {P0 , P16 , P25 , P34 }.
Proposition 1.16. For each pair of Göpel subgroups G± in Equation (1.54) which are
pairwise interchanged by the action of the involution , the abelian surfaces A′± = A/G±
satisfy A′± ≅ Jac (C0,±′
) for some smooth genus-two curves C0,± ′ ′
with C0,+ ′
≅ C0,− and
Jac (C0,± ) ∈ H16 . In particular, we have TA′± = H ⊕ ⟨4⟩ ⊕ ⟨−4⟩.
′
Proof. The proof follows by constructing the Richelot curves C±′ in Equation (1.6) for
each pair of Göpel groups in Equation (1.54). Denoting the Igusa-Clebsch invariants
′
of the genus-two curves C0,± by [I2± ∶ I4± ∶ I6± ∶ I10
±
] ∈ WP(2,4,6,10) , one checks that
(1.55) [I2+ ∶ I4+ ∶ I6+ ∶ I10
+
] = [I2− ∶ I4− ∶ I6− ∶ I10
−
].
One then checks by a direct computation that Q =/ 0 for each C0,± ′
whence it follows
Jac (C0,± ) /∈ H4 . Using the (2, 2)-isogeny Φ from Theorem 1.2, we obtain the following
′
compositions of isogenies
E1 × E2 Ð→ A = Jac (C0 ) Ð→ A′n,± = Jac (C0,± ) = A/G± .
Φ Ψ ′
(1.56)
Since Ψ is obtained from the projection onto the quotient by a Göpel group, it is a
(2, 2)-isogeny as well. Thus, Ψ ○ Φ is a (4, 4)-isogeny. Using Theorem 1.2 once more
it follows that Jac (C0,±
′
) ∈ H16 .
Corollary 1.17. For a genus-two curve C0 admitting an elliptic involution, there is
exactly one Göpel group G ⩽ A[2] of A = Jac (C0 ) such that A/G ≅ E1 × E2 for two
elliptic curves E1 , E2 . In particular, G is the unique Göpel group of A = Jac (C0 ) which
is fixed under the action of the involution  on A[2].
Remark 1.18. The special Göpel group from Corollary 1.17 has one non-trivial ele-
ment in common with each Göpel group in Proposition 1.13 whose associated (2, 2)-
isogenous Richelot curve has a Jacobian in H4 . In contrast, the intersection with each
Göpel group in Proposition 1.16 is trivial.
Remark 1.19. The results above determine the (2, 2)-isogenies for a point Jac (C0 )
in H4 . There is a similar result for a general point E1 × E2 in H1 . Obviously, 3 ⋅ 3 = 9
such isogenies have images that are products of elliptic curves, so again in H1 . The
remaining six are the ones in Corollary 1.11 and give points in H4 .
2. Kummer surfaces and isogenies
On a principally polarized abelian surface (A, L) with the minus identity involution
denoted by −I, one can always choose a theta divisor L = OA (Θ) to satisfy (−I)∗ Θ = Θ,
that is, to be a symmetric theta divisor. For an irreducible principal polarization the
abelian surface A then maps to the complete linear system ∣2Θ∣, and the rational map
ϕL2 ∶ A → P3 associated with the line bundle L2 factors via an embedding through
the projection A → A/⟨−I⟩; see [6]. In this way, one identifies A/⟨−I⟩ with its image
ϕL2 (A) in P3 , which is a singular quartic projective surface KA with sixteen ordinary
double points, called a nodal surface. The map from A onto its image KA ⊂ P3 is
two-to-one, except on sixteen points of order two where it is injective. Its minimum
resolution is the Kummer surface Kum(A) associated with the principally polarized
abelian surface (A, L). Due to the 166 configuration on KA , there are six hyperplanes
containing the double point ϕL2 (0) and touching KA along a double conic. The linear
projection with center ϕL2 (0) maps these six hyperplanes onto the six lines ℓ1 , . . . , ℓ6
in a projective plane P2 . The configuration (P2 ; ℓ1 , . . . , ℓ6 ) is called the Kummer plane
associated with (A, L). The rich geometry of six-line configurations in the Kummer
plane, as well as their strong connection with theta functions have been the subject
of multiple studies [12, 18, 22, 30, 39, 41, 55] over the last century and a half.
In this section we will construct the Kummer surfaces associated with the product
of two elliptic curves and the Jacobian of a genus-two curve as a double quadric, a
double sextic, and a quartic projective surface. We will also construct several Jacobian
elliptic fibrations on these surfaces. As a reminder, an elliptic surface is a (relatively)
minimal complex surface X , together with a Jacobian elliptic fibration, that is, a
holomorphic map πX ∶ X → P1 to P1 such that the general fiber is a smooth curve of
genus one together with a distinguished section Σ ∶ P1 → X that marks a smooth point
in each fiber. To each Jacobian elliptic fibration πX ∶ X → P1 there is an associated
Weierstrass model obtained by contracting all components of reducible fibers not
meeting Σ. By a slight abuse of notation we will denote the elliptic surface and
its associated Weierstrass model by the same symbol. The complete list of possible
singular fibers has been given by Kodaira [36]. It encompasses two infinite families
(In , In∗ , n ≥ 0) and six exceptional cases (II, III, IV, II ∗ , III ∗ , IV ∗ ). The Weierstrass
model of a smooth K3 surface can always be written in the form

(2.1) y 2 z = 4x3 − g2 (v) xz 2 − g3 (v)z 3 ,
where v is a suitable coordinate on the base curve P1v , and g2 and g3 are polynomials
of degree 8 and 12 in homogeneous coordinates on P1v , respectively. The section is
given by the point at infinity in each smooth fiber. We denote the Mordell-Weil group
of sections on the Jacobian elliptic surface πX ∶ X → P1 by MW(X , πX ). If a Jacobian
elliptic fibration admits in addition a two-torsion section T ∈ MW(X , πX ), we use a
change of coordinates to write Equation (2.1) in the form
(2.2) X∶ y 2 z = x3 + p1 (v) x2 z + p2 (v) xz 2 ,
where p1 and p2 are polynomials of degree 4 and 8 in homogeneous coordinates on
the base curve P1v , respectively. A natural holomorphic two-form on the K3 surface,
obtained as the minimal resolution of the Weierstrass model, is the pullback of the
holomorphic two-form dv∧dx/y in the (affine coordinate) chart z = 1 in Equation (2.2).
We make the following:
Remark 2.1. Oguiso classified the Jacobian elliptic fibrations on the Kummer sur-
faces associated with E1 × E2 where the elliptic curves Ei for i = 1, 2 are not mutually
isogenous in [51]. Kuwata and Shioda furthered Oguiso’s work in [38] where they com-
puted elliptic parameters and Weierstrass equations for all fibrations, and analyzed the
reducible fibers and Mordell-Weil lattices. Similarly, all inequivalent Jacobian ellip-
tic fibrations on the Kummer surface associated with Jac (C) for a general genus-two
curve C were determined explicitly by Kumar in [37]. In particular, Kumar computed
elliptic parameters and Weierstrass equations for all twenty five different fibrations
that appear, and analyzed the reducible fibers and Mordell-Weil lattices.
2.1. Kummer surfaces associated with two elliptic curves. In this section we
will concerned with the Kummer surface Kum(E1 × E2 ), i.e., the minimal resolution of
quotient surface of the product abelian surface E1 ×E2 by the inversion automorphism.
Unless stated otherwise we will assume that the two elliptic curves are not mutually
isogenous.
Let us define a double quadric surface closely related to E1 × E2 for the elliptic
curves given by Equation (1.20). In general, a double quadric surface is obtained as
the double cover branched along a locus of bi-degree (4, 4) in P1 × P1 ; see [5]. It is
well known that the minimal resolution of any double quadric is a K3 surface. In our
situation, we identify P1 = P(xl , zl ) for l = 1, 2 and define the special double quadric
surface Z using the equation
(2.3) Z∶ 2
y1,2 = x1 z1 (x1 − z1 )(x1 − Λ1 z1 ) x2 z2 (x2 − z2 )(x2 − Λ2 z2 ) .
There is a natural projection map π ∶ E1 × E2 ⇢ Z given by y1,2 = z1 z2 y1 y2 , invariant
under the action of ıE1 ×ıE2 . It follows that the minimal resolution of Z is the Kummer
surface Kum(E1 × E2 ).
A holomorphic two-form on Z is given by ωZ = dx1 ∧ dx2 /y1,2 in the coordinate
chart z1 = z2 = 1 in Equation (2.3). The regular two-from on E1 × E2 , given by the
outer tensor product of the elliptic-curve holomorphic one-forms, i.e.,

dx1 dx2 dx1 dx2
(2.4) ∧ ∶= ⊠ ,
y1 y2 y1 y2
descends to the quotient (E1 × E2 )/⟨−I⟩ and coincides with ωZ . Here, we use the
notation η1 ⊠ η2 = (π1∗ η1 ) ⊗ (π2∗ η2 ) where πl is the canonical projection onto El for
l = 1, 2. Let ǫ ∶ Ẑ → Z be the minimal resolution. We obtain a holomorphic two-form
ωKum(E1 ×E2 ) on Kum(E1 × E2 ) by pullback, i.e.,
(2.5) ωKum(E1 ×E2 ) = ǫ∗ ωZ , with ǫ∶ Kum(E1 × E2 ) = Ẑ → Z .
We also introduce the quadratic twist of Z, given by
(2.6) Z (ε) ∶ 2
ŷ1,2 = εx1 z1 (x1 − z1 )(x1 − Λ1 z1 ) x2 z2 (x2 − z2 )(x2 − Λ2 z2 ) ,
√
such that for general ε ∈ Q the surfaces Z and Z (ε) are isomorphic only over Q( ε).
We refer to the minimal resolution of Z (ε) as twisted Kummer surface and denote it
by Kum(E1 × E2 )(ε) . Holomorphic two-forms on the twisted double quadric surface
and the minimal resolution are then obtained analogously to Equation (2.5). We
make the following:
Remark 2.2. Projecting to either P1 in Equation (2.3) defines a Jacobian elliptic
fibration on the Kummer surface Kum(E1 × E2 ) denoted J4 in [38], i.e., a fibration
with the singular fibers 4I0∗ and a Mordell-Weil group of sections (Z/2Z)2 .
Remark 2.3. If the elliptic curves El for l = 1, 2 are not mutually isogenous, then
the Kummer surface Kum(E1 × E2 ) has Picard rank 18. It follows from [48] that the
transcendental lattice is given by
(2.7) TKum(E1 ×E2 ) ≅ TE1 ×E2 (2) = H(2) ⊕ H(2) .
Moreover, we will consider Kum(E1′ ×E2′ ), i.e., the minimal resolution of the quotient
surface of the product abelian surface E1′ × E2′ by the inversion automorphism – asso-
ciated with the two-isogenous elliptic curves El′ in Equation (1.43). The associated
double quadric surface is given by
(2.8) Z ′ ∶ 2
Y1,2 = X1 Z1 (X12 + 2(1 − 2Λ1 )X1 Z1 + Z12 ) X2 Z2 (X22 + 2(1 − 2Λ2 )X2 Z2 + Z22 ) .
On Kum(E1′ × E2′ ) a holomorphic two-form ωKum(E1′ ×E2′ ) is obtained as the pullback of
the two-form ωZ ′ = dX1 ∧ dX2 /Y1,2 in the chart Z1 = Z2 = 1 on Z ′ in Equation (2.8).
The Cartesian product of the dual two-isogenies χ′E1 × χ′E2 of elliptic curves induces a
rational cover Z ′ → Z. We have the following:
4
Lemma 2.4. The map χ ∶ Z ′ ⇢ Z (2 ) , given by
(2.9) χ∶ (X1 , Z1 , X2 , Z2 , Y1,2 ) ↦ (x1 , z1 , x2 , z2 , ŷ1,2 )
with
x1 = 4Λ1 X1 Z1 , z1 = (X1 + Z1 )2 ,
(2.10) x2 = 4Λ2 X2 Z2 , z2 = (X2 + Z2 )2 ,
ŷ1,2 = 16 Λ1 Λ2 (X12 − Z12 )(X22 − Z22 )Y1,2 ,
is a rational covering map defined over Q(Λ1 , Λ2 ). For the holomorphic two-forms
4
ωZ (24 ) = dx1 ∧ dx2 /ŷ1,2 in the chart z1 = z2 = 1 on Z (2 ) and ωZ ′ = dX1 ∧ dX2 /Y1,2 in
the chart Z1 = Z2 = 1 on Z ′ , it follows ωZ ′ = χ∗ ωZ (24 ) .
Proof. The proof follows by an explicit computation using Equations (2.10).
2.1.1. Certain elliptic fibrations. On the double quadric surface Z, a non-isotrivial
elliptic fibration, with fibers over P1 = P(u1 , u2 ) embedded into P2 = P(X, Y, Z), is
given by the Weierstrass model
Y 2 Z = X (X + u1 u2 (u1 − u2 )((ρ1 − ρ2 )u1 − (ρ1 + ρ2 )u2 )Z)
1
2
(2.11)
× (X + u1 u2 ((ρ1 − ρ2 )u21 − 2(ρ1 + 1)u1 u2 + (ρ1 + ρ2 )u22 )Z) .
1
2
The discriminant function of the elliptic fibration is
2−4 u81 u82 (u1 − u2 )2 ((ρ1 − ρ)u1 − (ρ1 + ρ2 )u2 ) ((ρ1 − ρ)u21 − 2(ρ1 + 1)u1 u2 + (ρ1 + ρ)u22 ) .
2 2
We have the immediate:

Lemma 2.5. Equation (2.11) defines a Jacobian elliptic fibration with the singular
fibers 2I2∗ + 4I2 and the Mordell-Weil group of sections (Z/2Z)2 .
In fact, Equation (2.11) defines the Jacobian elliptic fibration on Kum(E1 × E2 )
denoted J6 in [38]. We have the following:
Proposition 2.6. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} and parameters ρ1 , ρ2 given by
ρ1 = Λ1 + Λ2 − 2Λ1 Λ2 − 1 , ρ2 = 1 − Λ1 − Λ2 ,
the double quadric surface Z (2.3) and the Jacobian elliptic surface (2.11) are bira-
tional equivalent over Q(Λ1 , Λ2 ).
Proof. In the chart Z = u2 = 1 in Equation (2.11) and z1 = z2 = 1 in Equation (2.3), a
birational map is given by
x1 x2 x21 x2 (x1 + x2 − 1)(x1 x2 − Λ2 (x1 + x2 ) + Λ2 )y1,2
u1 = Y =
(x1 − 1)(x2 − 1) (x1 − 1)4 (x2 − 1)5
, ,
x1 x2 (x1 + x2 − 1)(x1 − Λ1 )(x1 x2 − Λ2 (x1 + x2 ) + Λ2 )

(2.12)
X=
(x1 − 1)3 (x2 − 1)3
.

The holomorphic two-forms are related as follows:
Corollary 2.7. The birational equivalence of Proposition 2.6 identifies the holo-
morphic two-form du1 ∧ dX/Y in the chart Z = u2 = 1 in Equation (2.11) with
ωZ = dx1 ∧ dx2 /y1,2 in the chart z1 = z2 = 1 in Equation (2.3).
Proof. The proof follows by an explicit computation using the transformation pro-
vided in the proof of Proposition 2.6.
On the double quadric surface Z ′ , another elliptic fibration, with fibers over P1 =
P(v1 , v2 ) embedded into P2 = P(X, Y, Z), is given by the Weierstrass model
(2.13) Y 2 Z = X 3 − 2v2 (2v1 − v2 )(v12 + 2ρ1 v1 v2 + ρ22 v22 )X 2 Z + v24 (v12 + 2ρ1 v1 v2 + ρ22 v22 ) XZ 2 .
2
The discriminant function of the fibration is 16v1 v210 (v1 −v2 )(v12 +2ρ1 v1 v2 +ρ22 v22 )6 . We
have the immediate:
fibers I4∗ + 2I1 + 2I0∗ and the Mordell-Weil group of sections Z/2Z.
Equation (2.13) defines the Jacobian elliptic fibration on Kum(E1′ × E2′ ) denoted J7
in [38]. However, to construct a birational equivalence between the Jacobian elliptic
fibration and Z ′ , one has to use a finite field extension Q(κ1 , κ2 ) with Λ1 = 1/(1 − κ21 )
and Λ2 = 1/(1 − κ22 ). We have the following:
Proposition 2.9. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} let parameters ρ1 , ρ2 be given by
ρ1 = Λ1 + Λ2 − 2Λ1 Λ2 − 1 , ρ2 = 1 − Λ1 − Λ2 .
The double quadric surface Z ′ (2.8) and the Jacobian elliptic surface (2.13) are bira-
tional equivalent over Q(κ1 , κ2 ) with Λl = 1/(1 − κ2l ) for l = 1, 2.
Proof. In the charts v2 = 1, Z = 1 in Equation (3.35) and Z1 = Z2 = 1 in Equation (2.8),
a birational equivalence is given by
(κ21 − 1)(κ1 κ22 − 1)X1 X2 − κ1 (κ1 + 1)(κ22 − 1)X1 + κ1 (κ1 − 1)(κ22 − 1)X22 + (κ21 − 1)(κ1 κ22 + 1)X2
v1 =
(1 − κ21 )(1 − κ22 )X2 ((κ1 + 1)X1 + κ1 − 1)
,
((κ1 + 1)(κ2 − 1)X1 − (κ1 − 1)(κ2 + 1)X2 )((κ1 + 1)(κ2 + 1)X1 − (κ1 − 1)(κ2 − 1)X2 )
X=
(1 − κ21 )2 (1 − κ22 )2
((κ2 + 1)X2 + (κ2 − 1))((κ2 − 1)X2 + (κ2 + 1))κ21 X1
((κ1 + 1)X1 + κ1 − 1)2 X23
(2.14) × ,
((κ1 + 1)(κ2 − 1)X1 − (κ1 − 1)(κ2 + 1)X2 )2 ((κ1 + 1)(κ2 + 1)X1 − (κ1 − 1)(κ2 − 1)X2 )2
Y =−
(1 − κ21 )3 (1 − κ22 )3
((κ2 + 1)X2 + (κ2 − 1))((κ2 − 1)X2 + (κ2 + 1))κ31 Y1,2
((κ1 + 1)X1 + κ1 − 1)4 X25
× .

morphic two-form dv1 ∧ dX/Y in the chart Z = v2 = 1 in Equation (2.13) with
ωZ ′ = dX1 ∧ dX2 /Y1,2 in the chart Z1 = Z2 = 1 in Equation (2.8).
2.2. Jacobian Kummer surfaces. The Kummer surface Kum(Jac C) is the mini-
mal resolution of the quotient surface of the Jacobian Jac (C) by the inversion auto-
morphism where C is a generic, smooth genus-two curve.
Since the Jacobian Jac (C) is birational to the symmetric product Sym2 (C), let
us consider the function field of Sym2 (C) = (C × C)/⟨σ⟩ where σ interchanges two
copies of a generic genus-two curve C. In particular, for a smooth genus-two curve C
in Rosenhain form (1.1), the function field of the variety Sym2 (C)/⟨ıC × ıC ⟩ – where
ıC ×ıC is the involution on Sym2 (C) induced by the hyperelliptic involution ıC on C – is
generated by z1 = Z (1) Z (2) , z2 = X (1) Z (2) +X (2) Z (1) , z3 = X (1) X (2) , and z̃4 = Y (1) Y (2) ,
subject to the relation
3
(2.15) W∶ z̃42 = z1 z3 (z1 − z2 + z3 ) ∏ (λ2i z1 − λi z2 + z3 ) .
i=1
Equation (2.15) is a special case of a double sextic surface, i.e., the double cover of
P2 = P(z1 , z2 , z3 ) branched on the union of six lines (and hence over a sextic curve).
For a configuration of six lines in general position, the minimal resolution of a double
sextic surface is always a K3 surface; see [5]. Equation (2.15) is known as Shioda
sextic associated with Jac (C); see [57]. It follows that the minimal resolution of W
is the Kummer surface Kum(Jac C).
In Equation (2.15) the double cover is branched along the six lines ℓi with 1 ≤ i ≤ 6,
given by
(2.16) λ2i z1 − λi z2 + z3 = 0 with 1 ≤ i ≤ 3, z1 = 0, z1 − z2 + z3 = 0, z3 = 0 .
The six lines are tangent to the common conic K2 = − 4 z1 z3 = 0. Conversely, it z22
is easy to see that any six lines tangent to a common conic can always be brought
into the form of Equations (2.16) using a projective transformation. Humbert proved
that the Kummer plane (P2 ; ℓ1 , . . . , ℓ6 ) inherits essential information of the principally
polarized abelian surface A = Jac (C) itself; see [31]. A picture is provided in Figure 2.
✵✵
✡✡ ✵✵✵
✡ ✵
✶✶ ✡✡✡ ℓ4 ✵✵
✵✵ ✗
✶✶ ✡✡ ✵✵ ✗✗
✶✶ ℓ ✡✡ ✵ℓ✵ 5 ✗ ✗
✶✶ 3 ✡✡ ✵✵ ✗✗
✶✶ ✡✡ ✵✗
✡✶✡✶ K2 = 0 ✗✗✗ ✵✵✵
✡✡ ✶
✡ ✶ ✵
✡✡ ℓ2 ✶✶✶ ✗✗ ✵✵
✶✶ ✗✗ ✗ ℓ6 ✵✵
✵✵
✶
ℓ ✗✗
1
Figure 2. Double cover branched along reducible sextic

Remark 2.11. In [7] the geometry of the Kummer plane was determined over several
Humbert surfaces H∆ . In particular, the following statement were proven [7, Cor. 7.2]
and [7, Cor. 7.3]:
∆ = 4: if (A, L) ∈ H4 if and only if (numbering the six lines on its Kummer plane
(P2 ; ℓ1 , . . . , ℓ6 ) suitably) the three points ℓ1 ∩ ℓ2 , ℓ3 ∩ ℓ4 , ℓ5 ∩ ℓ6 are collinear.
∆ = 16: if (A, L) ∈ H16 then its Kummer plane (P2 ; ℓ1 , . . . , ℓ6 ) admits a cubic pass-
ing smoothly through three of the 15 points ℓm ∩ ℓn and touching the sin-
gular lines ℓn in the remaining intersection points with even multiplicity.
Conversely, if (P2 ; ℓ1 , . . . , ℓ6 ) admits such a curve, then (A, L) ∈ H∆ with
∆ ∈ {4, 8, 12, 16, 20}.
For the transcendental lattice it follows TKum(A) ≅ TA (2) [48, Thm. 10]. In particular
we have
TKum(A) = H(2) ⊕ ⟨4⟩ ⊕ ⟨−4⟩ for (A, L) ∈ H4 ,
TKum(A) = H(2) ⊕ ⟨8⟩ ⊕ ⟨−8⟩ for (A, L) ∈ H16 .
(2.17)
Generally, for A ∈ Hδ2 it follows from [48] that Kum(A) has Picard rank 18, and the
transcendental lattice is given by
(2.18) TKum(A) ≅ TA (2) = H(2) ⊕ ⟨2δ⟩ ⊕ ⟨−2δ⟩ .
2.2.1. Closely related normal forms. The Shioda sextic is closely related to two quartic
equations, known as the Baker quartic and Cassels-Flynn quartic. The Baker quartic
is obtained directly from the Shioda sextic W in Equation (2.15). Using parameters
L4 = 1 + λ1 + λ2 + λ3 , L3 = λ1 + λ2 + λ3 + λ1 λ2 + λ2 λ3 + λ1 λ3 ,
(2.19)
L 1 = λ1 λ2 λ3 , L2 = λ1 λ2 + λ2 λ3 + λ1 λ3 + λ1 λ2 λ3 ,
and a variable transformation given by
W = K2 z1 , X = K2 z2 , Y = K2 z3 ,
2z̃4 − (L1 z12 z2 + L3 z1 z2 z3 − 2L4 z1 z32 + z2 z32 ) .
(2.20)
Z= − 2L2 z12 z3
with K2 given in Equation (2.22). Equation (2.15) is equivalent to the quartic surface
U in P3 = P(W, X, Y, Z) given by the vanishing locus of the Baker determinant, i.e.,
RR 0 Y RRRR
RRR L1 W
R
RRR L1 W 2L2 W + 2Z L3 W − Y X RRRR
−Z
RR
L3 W − Y 2L4 W − 2X W RRRR
(2.21) U∶ RRR −Z = 0.
RRR RRR
RR Y X W 0 RR
The Baker determinant was first derived in [4].
In addition to K2 introduced above, we define homogeneous polynomials Kl =
Kl (z1 , z2 , z3 ) of degree 4 − l for l = 0, 1 with
K2 = z22 − 4z1 z3 , K1 = −2z2 z32 − 2L1 z12 z2 + 4L2 z12 z3 − 2L3 z1 z2 z3 + 4L4 z1 z32 ,
(2.22) K0 = L21 z14 − 2L1 L3 z13 z3 + (2L1 − 4L2 L4 + L23 )z12 z32 + 4L1 L4 z12 z2 z3
+ 3(−2L1 z22 z3 + 2L2 z2 z32 − L3 z33 )z1 + z34 .
A variable transformation in Equation (2.15), given by

(K2 z4 + 2K1 ) ,
1
(2.23) z̃4 =
4
or, equivalently, the variable transformation in Equation (2.21), given by
(2.24) W = K2 z1 , X = K2 z2 , Y = K2 z3 , Z = K2 z4 + K1 ,
yields the quartic projective surface KA in P3 = P(z1 , z2 , z3 , z4 ) given by
(2.25) KA ∶ K2 (z1 , z2 , z3 ) z42 + K1 (z1 , z2 , z3 ) z4 + K0 (z1 , z2 , z3 ) = 0 .
The quartic appeared in Cassels and Flynn [11, Sec. 3] and is called the Cassels-Flynn
quartic. We have the following:
Proposition 2.12. Assume that λ1 , λ2 , λ3 are the Rosenhain roots of a smooth genus-
two curve C. The surfaces in Equations (2.15), (2.21), and (2.25) are birational
equivalent over Q(λ1 , λ2 , λ3 ). The minimal resolution is isomorphic to the Kummer
surface Kum(A) associated with the Jacobian A = Jac (C).
Proof. It is known that under birational equivalence the involution minus −I on
Jac (C) restricts to the product involution ıC × ıC on Sym2 (C); see [15]. Hence, the
Shioda sextic in Equation (2.15) is birational to Jac (C)/⟨−I⟩ associated with the
Jacobian A = Jac (C) of a genus-two curve C in Rosenhain normal form (1.1). One
substitutes Equations (2.20) into Equation (2.21) and checks that the strict transform
is given by Equation (2.3). Moreover, the variable transformation in Equation (2.21)
can be inverted to obtain a rational inverse map.
Remark 2.13. The sixteen nodes pij of the Kummer surface KA in Equation (2.25)
are given in Table 2. They are the images of the order-two points Pij ∈ A[2].
A holomorphic two-form on W is given by the two-form ωW = dz2 ∧ dz3 /z̃4 in the
coordinate chart z1 = 1 in Equation (2.15). A regular two-form on C × C, given as the
antisymmetric linear combination of the outer tensor products of the two holomorphic
one-forms on each copy of C, i.e.,
dX (1) X (2) dX (2) X (1) dX (1) dX (2)
(2.26) ⊠ − ⊠ (2) ,
Y (1) Y (2) Y (1) Y
descends to Sym2 (C)/⟨ıC × ıC ⟩ and coincides with ωW . If ǫ is the minimal resolution,
we obtain a holomorphic two-form ωKum(Jac C) on Kum(Jac C) by pullback, i.e.,
(2.27) ωKum(Jac C) = ǫ∗ ωW , ̂ ⇢ W.
with ǫ∶ Kum(Jac C) = W
We also introduce the twisted Shioda sextic W (ε) , given by
3
(2.28) W (ε) ∶ ẑ42 = εz1 z3 (z1 − z2 + z3 ) ∏ (λ2i z1 − λi z2 + z3 ) ,
i=1
√
such that for general ε ∈ Q the surfaces W (ε) and W are isomorphic only over Q( ε).
Further, we introduce another genus-two curve, namely
(2.29) C̃ ∶ Y 2 = XZ(X − λ0 Z) (X − λ0 λ1 Z) (X − λ0 λ2 Z) (X − λ0 λ3 Z) ,
nodes on KA [z1 ∶ z2 ∶ z3 ∶ z4 ]
p0 [0 ∶ 0 ∶ 0 ∶ 1]
p16 [0 ∶ 1 ∶ λ1 ∶ λ21 ]
p26 [0 ∶ 1 ∶ λ2 ∶ λ22 ]
p36 [0 ∶ 1 ∶ λ3 ∶ λ23 ]
p46 [0 ∶ 1 ∶ 0 ∶ 0]
p56 [0 ∶ 1 ∶ 1 ∶ 1]
p14 [1 ∶ λ1 ∶ 0 ∶ λ2 λ3 ]
p24 [1 ∶ λ2 ∶ 0 ∶ λ1 λ3 ]
p34 [1 ∶ λ3 ∶ 0 ∶ λ1 λ2 ]
p45 [1 ∶ 1 ∶ 0 ∶ λ1 λ2 λ3 ]
p15 [1 ∶ λ1 + 1 ∶ λ1 ∶ λ1 (λ2 + λ3 )]
p25 [1 ∶ λ2 + 1 ∶ λ2 ∶ λ2 (λ1 + λ3 )]
p35 [1 ∶ λ3 + 1 ∶ λ3 ∶ λ3 (λ1 + λ2 )]
p13 [1 ∶ λ1 + λ3 ∶ λ1 λ3 ∶ (λ2 + 1) λ1 λ3 ]
p23 [1 ∶ λ2 + λ3 ∶ λ2 λ3 ∶ (λ1 + 1) λ1 λ2 ]
p12 [1 ∶ λ1 + λ2 ∶ λ1 λ2 ∶ (λ3 + 1) λ1 λ2 ]
Table 2. Nodes on the generic Jacobian Kummer surface KA
and its associated (twisted) Shioda sextic given by

̃(ε) ∶
3
(2.30) W ŷ42 = εy1 y3 (λ20 y1 − λ0 y2 + y3 ) ∏ ((λ0 λi )2 y1 − λ0 λi y2 + y3 ) .
i=1
We refer to the minimal resolutions of W (ε) and W ̃ (ε) as twisted Kummer surfaces
(ε)
and denote them as Kum(Jac C) and Kum(Jac C) ̃ (ε) , respectively. Holomorphic
two-forms on the various twisted Shioda sextic surfaces and their minimal resolutions
are then obtained analogously to Equation (2.27). We have the following:
Proposition 2.14. Assume λ1 , λ2 , λ3 are the Rosenhain roots of a smooth genus-
two curve C and λ0 =/ 0. The surfaces W (ε) and W ̃ (ε/λ40 ) are birational equivalent
over Q(λ0 ) such that the holomorphic two-form dy2 ∧ dy3 /ŷ4 in the chart y1 = 1
in Equation (2.30) equals dz2 ∧ dz3 /ẑ4 in the chart z1 = 1 in Equation (2.15). In
particular, there is an isomorphism
(2.31) ı∶ ̃ (ε/λ0 ) ,
Kum(Jac C)(ε) Ð→ Kum(Jac C)
≅ 4
such that ωKum(Jac C)(ε) = ı∗ ωKum(Jac C)

̃ (ε/λ40 ) .
Proof. For the isomorphism ı ∶ (z1 , z2 , z3 , ẑ4 ) ↦ (y1 , y2 , y3 , ŷ4 ) = (z1 , λ0 z2 , λ20 z3 , λ30 ẑ4 )
the pullback ı∗ (dy2 ∧ dy3/ŷ4 ) equals dz2 ∧ dz3 /ẑ4 .
2.2.2. Certain elliptic fibrations. The pencil of lines, given by tz1 − z3 = 0 in P2 =
P(z1 , z2 , z3 ), induces an elliptic fibration on the Shioda sextic W. The elliptic fibra-
tion, with fibers over P1 = P(u1 , u2 ) embedded into P2 = P(X, Y, Z), is given by the
Weierstrass model
Y 2 Z = X(X + (λ1 − λ3 )(λ2 − 1)u1 u2 (u1 − λ2 u2 )(u1 − λ1 λ3 u2 )Z)
(2.32)
× (X + (λ1 − λ2 )(λ3 − 1)u1 u2 (u1 − λ3 u2 )(u1 − λ1 λ2 u2 )Z) .
The discriminant function of the elliptic fibration is given by
(λ1 − 1)2 (λ2 − 1)2 (λ3 − 1)2 (λ1 − λ2 )2 (λ1 − λ3 )2 (λ2 − λ3 )2 u61 u62 (u1 − λ1 u2 )2
× (u1 − λ2 u2 )2 (u1 − λ3 u2 )2 (u1 − λ1 λ2 u2 )2 (u1 − λ1 λ3 u2 )2 (u1 − λ2 λ3 u2 )2 .
The fibration in Equation (2.32) was denoted fibration (1) in [37]. In [37] the following
lemma was proven:
fibers 6I2 + 2I0∗ and a Mordell-Weil group of sections Z/2Z ⊕ ⟨1⟩.
In the statement above the symbol ⟨m⟩ stands for a rank 1 lattice Zx satisfying
⟨x, x⟩ = m with respect to the height pairing. We also have the following:
two curve. The surfaces in Equation (2.15) and (2.32) are birational equivalent over
Q(λ1 , λ2 , λ3 ).
Proof. In the charts u2 = 1, Z = 1 in Equation (2.32) and z1 = 1 in Equation (2.15), a
birational equivalence is given
(λ1 − 1)(λ2 − 1)(λ3 − 1)u1 (u1 − λ1 )(u1 − λ2 )(u1 − λ3 )
z2 =
X − λ1 (λ2 − 1)(λ3 − 1)u1 (u1 − λ2 )(u1 − λ3 )
+ u1 + 1 ,
(2.33) (λ1 − 1)(λ2 − 1)(λ3 − 1)u1 (u1 − λ1 )(u1 − λ2 )(u1 − λ3 )Y
z̃4 = , z3 = u1 .
(X − λ1 (λ2 − 1)(λ3 − 1)u1 (u1 − λ2 )(u1 − λ3 ))
2

morphic two-form dz2 ∧ dz3 /z̃4 in the chart z1 = 1 in Equation (2.15) with du1 ∧ dX/Y
in the chart Z = u2 = 1 in Equation (2.32).
On the Shioda sextic W, another elliptic fibration, with fibers over P1 = P(w1 , w2 )
embedded into P2 = P(x, y, z), is given by the Weierstrass model
y 2 z = x3 + w1 (w1 + (λ1 + λ2 λ3 )w2 )(w1 + (λ2 + λ1 λ3 )w2 )(w1 + (λ3 + λ1 λ2 )w2 )x2 z
(2.34)
+λ1 λ2 λ3 w22 (w1 + (λ1 + λ2 λ3 )w2 ) (w1 + (λ2 + λ1 λ3 )w2 ) (w1 + (λ3 + λ1 λ2 )w2 ) xz 2 .
2 2 2
The discriminant function of the elliptic fibration is given by

λ21 λ22 λ23 w24 (w12 − 4λ1 λ2 λ3 w22 )(w1 + (λ1 + λ2 λ3 )w2 )
6
(2.35)
× (w1 + (λ2 + λ1 λ3 )w2 ) (w1 + (λ3 + λ1 λ2 )w2 ) .
6 6
We have the immediate:

fibers I4 + 2I1 + 3I0∗ and a Mordell-Weil group of sections Z/2Z.
The fibration in Equation (2.34) was denoted fibration (4) in [37]. We have the
following:
two curve. The surfaces in Equation (2.15) and (2.34) are birational equivalent over
Q(λ1 , λ2 , λ3 ).
Proof. In the charts w2 = 1, z = 1 in Equation (2.34) and z1 = 1 in Equation (2.15), a
birational equivalence is given by
(λ1 (λ2 − λ3 ) − λ2 λ3 (λ1 + 1))z2 z3 + λ1 λ2 λ3 (1 + λ1 + λ2 − λ3 )z2 + λ3 (1 + λ1 + λ2 − λ3 )z32
w1 =
λ3 z2 z3 − λ1 λ2 z2 − z32 + ((λ2 − λ3 )(λ1 + 1) + λ1 − λ2 λ3 )z3 + λ1 λ2 λ3
(λ21 λ2 + λ1 (1 + λ2 )(λ2 − λ23 ) − λ2 λ23 )z3 + λ1 λ2 (λ2 λ3 + λ1 (λ2 λ3 − λ2 + λ3 )
λ3 z2 z3 − λ1 λ2 z2 − z32 + ((λ2 − λ3 )(λ1 + 1) + λ1 − λ2 λ3 )z3 + λ1 λ2 λ3
− ,
(λ1 − λ3 )2 (λ2 − λ3 )2 (1 − λ3 )2 z3 (z2 − z3 − 1)(z3 − λ1 )(z3 − λ2 )(z3 − λ1 λ2 )(λ1 z2 − z3 − λ21 )(λ2 z2 − z3 − λ22 )
x= ,
(λ3 z2 z3 − λ1 λ2 z2 − z32 + ((λ2 − λ3 )(λ1 + 1) + λ1 − λ2 λ3 )z3 + λ1 λ2 λ3 )
3
(λ3 − λ1 )3 (λ2 − λ3 )3 (λ3 − 1)3 (z2 − z3 − 1)(z3 − λ1 )2 (z3 − λ2 )2 (z3 − λ1 λ2 )2 (λ1 z2 − z3 − λ21 )(λ2 z2 − z3 − λ22 )z̃4
y= .
(λ3 z2 z3 − λ1 λ2 z2 − z32 + ((λ2 − λ3 )(λ1 + 1) + λ1 − λ2 λ3 )z3 + λ1 λ2 λ3 )
5

morphic two-form dw1 ∧ dx/y in the chart z = w2 = 1 in Equation (2.34) with
dz2 ∧ dz3 /z̃4 in the chart z1 = 1 in Equation (2.15).
2.2.3. Specialization to Picard rank 18. We consider the special case of the Kummer
surface associated with the Jacobian of the genus-two curve C0 in Equation (1.18),
i.e., a smooth genus-two curve admitting an elliptic involution. We will denote the
Shioda sextic in Equation (2.15) for λ1 = λ2 λ3 by W0 , and the corresponding vanishing
locus for the Baker determinant by U0 . Similarly, we use the symbols C̃0 , W0 , W
(ε) ̃ (ε)
0
to refer to the corresponding specialization for λ1 = λ2 λ3 .
Remark 2.21. The induced involution  on the Shioda sextic W0 in Equation (2.15)
for λ1 = λ2 λ3 is given by
(2.36) ∶ (z1 , z2 , z3 , z̃4 ) ↦ (z3 , λ2 λ3 z2 , (λ2 λ3 )2 z3 , (λ2 λ3 )3 z̃4 ) .
In terms of the elliptic fibration already discussed above, the restriction to the
curve C0 in Equation (1.18) is characterized as follows:
Lemma 2.22. Assume that λ1 , λ2 , λ3 are the Rosenhain roots of a smooth genus-two
curve C. We have the following:
(1) the singular fibers in the fibration of Lemma 2.15 are I4 + 4I2 + 2I0∗ if and only
if λ1 = λ2 λ3 , λ2 = λ1 λ3 , or λ3 = λ1 λ2 ,
(2) the singular fibers in the fibration of Lemma 2.18 are I4 + I1 + I1∗ + 2I0∗ if and
only if λ1 = λ2 λ3 , λ2 = λ1 λ3 , or λ3 = λ1 λ2 .
Proof. We prove the second statement, the first one is analogous. We apply the
resultant to pairs of factors in the discriminant function (2.35) to obtain criteria for
coalescing fibers. The resultant (with respect to [w1 ∶ w2 ]) of the factors in the
reduced discriminant corresponding to fibers of Kodaira-type I0∗ , i.e.,
(2.37) (w1 + (λ1 + λ2 λ3 )w2 )(w1 + (λ2 + λ1 λ3 )w2 )(w1 + (λ3 + λ1 λ2 )w2 ) ,
and the factors in the discriminant corresponding to the fibers of Kodaira-type I1 ,
i.e., w12 − 4λ1 λ2 λ3 w22 , yields
(λ1 − λ2 λ3 ) (λ2 − λ1 λ3 ) (λ3 − λ1 λ2 ) .
2 2 2
(2.38)
The elliptic fibrations in Lemma 2.22 and the quadratic twist are related as follows:
Corollary 2.23. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 . The Weierstrass
model, given by
Y 2 Z = εX(X + λ2 (λ3 − 1)2 u1 u2 (u1 − λ3 u2 )(u1 − λ22 λ3 u2 )Z)
(2.39)
× (X + λ3 (λ2 − 1)2 u1 u2 (u1 − λ2 u2 )(u1 − λ2 λ23 u2 )Z) ,
defines a Jacobian elliptic fibration with the singular fibers I4 +4I2 +2I0∗ and a Mordell-
(ε)
Weil group of sections Z/2Z + ⟨1⟩. The elliptic surface is birational equivalent to W0
over Q(λ2 λ3 , λ2 + λ3 ) and Corollary 2.17 holds.
Corollary 2.24. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 . The Weierstrass
model, given by
y 2 z = x3 + εw1 (w1 + 2λ2 λ3 w2 )(w1 + λ2 (1 + λ23 )w2 )(w1 + λ3 (1 + λ22 )w2 )x2 z
(2.40)
+ε2 (λ2 λ3 )2 w22 (w1 + 2λ2 λ3 w2 ) (w1 + λ2 (1 + λ23 )w2 ) (w1 + λ3 (1 + λ22 )w2 ) xz 2 ,
2 2 2
defines a Jacobian elliptic fibration with the singular fibers I4 + I1 + I1∗ + 2I0∗ and a
Mordell-Weil group of sections Z/2Z. The elliptic surface is birational equivalent to
(ε)
W0 over Q(λ2 , λ3 ) and Corollary 2.20 holds.
2.3. Quartic surfaces. Let us briefly recall the construction of the Göpel-Hudson
quartic for a Kummer surface associated with a principally polarized abelian surface
(A, L) based on results in [6]. Considering the rational map ϕL2 ∶ A → P3 , its image
ϕL2 (A) is a quartic surface in P3 which, using the projective coordinates [w ∶ x ∶ y ∶ z],
can be written as
(2.41) 0 = ξ0 (w4 + x4 + y4 + z4 ) + ξ4 wxyz
+ξ1 (w2 z2 + x2 y2 ) + ξ2 (w2 x2 + y2 z2 ) + ξ3 (w2 y2 + x2 z2 ) ,
for some parameter set [ξ0 ∶ ξ1 ∶ ξ2 ∶ ξ3 ∶ ξ4 ] ∈ P4 . A general member of the family (2.41)
is smooth. As soon as the surface is singular at a general point, it must have sixteen
singular nodal points because of its symmetry. The discriminant turns out to be a
homogeneous polynomial of degree eighteen in the parameters [ξ0 ∶ ξ1 ∶ ξ2 ∶ ξ3 ∶ ξ4 ] ∈ P4
and was determined in [6, Sec. 7.7 (3)]. Thus, the Kummer surfaces form an open set
among these surfaces with parameters [ξ0 ∶ ξ1 ∶ ξ2 ∶ ξ3 ∶ ξ4 ] ∈ P4 , namely the ones that
make the irreducible factor of degree three in the discriminant vanish, i.e.,
(2.42) ξ0 (16ξ02 − 4ξ12 − 4ξ22 − 4ξ33 + ξ42 ) + 4 ξ1 ξ2 ξ3 = 0 .
Setting ξ0 = 1 and using the affine moduli ξ1 = −A, ξ2 = −B, ξ3 = −C, ξ4 = 2D, we ob-
tain a normal form for a nodal quartic surface, known as Göpel-Hudson quartic (GH-
quartic). The Göpel-Hudson quartic is the projective surface in P3 = P(w, x, y, z)
given by
(2.43) 0 = w4 + x4 + y4 + z4 + 2Dwxyz
−A(w2 z2 + x2 y2 ) − B(w2 x2 + y2 z2 ) − C(w2 y2 + x2 z2 ) ,
where A, B, C, D ∈ C satisfy
(2.44) D 2 = A2 + B 2 + C 2 + ABC − 4 .
By construction, we have the following:
Lemma 2.25. The minimal resolution of the quartic surface in Equation (2.43),
for generic parameters (A, B, C, D) satisfying (2.44), is isomorphic to the Kummer
surface Kum(A) associated with a principally polarized abelian surface A.
The abelian surface in Lemma 2.25 for generic parameters (A, B, C, D) does not
admit any additional automorphism and is in fact isomorphic to Jac (C) for a gen-
eral genus-two curve C. In [15, Thm. 4.46] two of the authors determined explicitly
the connection between the parameters (A, B, C, D) and the moduli of the abelian
surface A = Jac (C) for a smooth genus-two curve C. It was also proved in [15] that
(A, B, C, D) are modular functions relative to Γ2 (2). We have the following:
Proposition 2.26. If the minimal resolution of the Hudson quartic (2.43), for param-
eters (A, B, C, D) satisfying Equation (2.44), is isomorphic to the Kummer surface
Kum(A) associated with a principally polarized abelian surface A = Jac (C0 ) for a
smooth genus-two curve C0 admitting an elliptic involution, then one of the following
additional relations holds:

case I ∶ D = 0,
case II ∶ B = ±C ,
(2.45) case III ∶ A = ±B , or A = ±C ,
A + B + C ± D + 6 = 0, − A − B + C ± D + 6 = 0,
{
or or
case IV ∶
A − B − C ± D + 6 = 0, or − A + B − C ± D + 6 = 0.
Remark 2.27. Table 3 shows the one-to-one correspondence between the 15 compo-
nents in Equation (2.45) and the components in H2 /Γ2 (2) covering H4 in the Pring-
sheim decomposition in Proposition 1.5, using the same marking as before.
Remark 2.28. Equation (2.42) defines the Segre cubic threefold, i.e., the projective
dual of the Igusa quartic in P4 that is the Satake compactification of A2 (2). Under
this projective duality the decomposable abelian surfaces in A2 (2), that is H1 (2), are
contracted to 10 nodes of the Segre cubic. The Segre cubic has the simpler equation
∑ xi = ∑ x3i = 0 in P5 such that the permutation of the coordinates corresponds to the
action of S6 on A2 (2) and induces an action on the parameters A, B, C, D [19, p. 182-
183 and p. 156-159] and [59, p. 317–350 and p. 348 for Hδ (2) for δ = 1, 4]. One has to
omit the 15 hyperplanes in the Segre cubic whose divisors correspond to the boundary
components of the Satake compactification A2 (2) [19, Prop. 6]. The 10 nodes of the
Segre cubic correspond to the double quadrics given by (w2 ± x2 ± y2 ± z2 )2 with an
even number of minus signs, and (ab ± cd)2 with {a, b, c, d} = {w, x, y, z}. They are
attained in Equation (2.43) for images of E1 ×E2 . Once all these boundary components
are excluded, the statement in Proposition 2.26 becomes ‘if and only if ’.
Proof. A linear transformation of variables in Equation (2.21) over a suitable finite
extension field yields a Göpel quartic in Equation (2.43). In [15, Thm. 4.46] a corre-
sponding solution for the parameters (A, B, C, D) in terms of the Rosenhain roots of
a genus-two curve C0 in Equation (1.1) was determined explicitly. It is given by
λ1 + 1 λ1 λ2 + λ1 λ3 − 2λ2 λ3 − 2λ1 + λ2 + λ3 λ3 + λ2
A=2 B=2 C =2
(λ2 − λ3 )(λ1 − 1)
, , ,
λ1 − 1 λ3 − λ2
λ1 − λ2 λ3
D=4
(λ2 − λ3 )(λ1 − 1)
(2.46) ,
such that Equation (2.44) is satisfied. All other possible solutions are obtained from
this one by the 15 transpositions acting on the roots (λ1 , λ2 , λ3 , 0, 1, ∞). This is
precisely the relation between the Pringsheim components for H4 given in Proposi-
tion 1.5. We can then rewrite any relation between λ1 , λ2 , λ3 as an additional relation
for A, B, C, D. In this way, we obtain a one-to-one correspondence between the 15
Pringsheim components in Proposition 1.5 and the 15 additional relations for the
Göpel-Hudson quartic in Equation (2.45).
Proposition 2.26 provides a characterization of the 15 components in H2 /Γ2 (2)
covering H4 . Because of Remark 2.11 each of the components also corresponds to a
configuration in the Kummer plane where three nodes on KA in Table 2 are collinear.
I
2
2τ12 + τ11 τ22 − τ12 =0 λ1 = λ2 λ3 D = 0 {p15 , p23 , p46 }
II
τ11 + 2τ12 = 0 λ1 − λ2 = λ3 (1 − λ2 ) B + C = 0 {p14 , p23 , p56 }
τ11 + 2τ12 − (τ11 τ22 − τ12
2
)=0 λ1 (1 − λ3 ) = λ2 (λ1 − λ3 ) B − C = 0 {p16 , p23 , p45 }
III
2τ12 − τ22 = 0 λ3 = λ1 λ2 A−C =0 {p12 , p35 , p46 }
2τ12 − τ22 + (τ11 τ22 − τ12
2
)=0 λ2 = λ1 λ3 A+C =0 {p13 , p25 , p46 }
τ11 − τ22 = 0 λ1 − λ2 = λ2 (λ1 − λ3 ) A+B =0 {p15 , p26 , p34 }
τ11 − τ22 + (τ11 τ22 − τ12
2
)=0 λ1 − λ3 = λ3 (λ1 − λ2 ) A−B =0 {p15 , p24 , p36 }
IV
2τ12 = 1 λ2 − λ3 = −λ1 (1 − λ2 ) −A + B − C − D + 6 = 0 {p12 , p34 , p56 }
2τ12 + τ22 = 1 λ3 (1 − λ2 ) = −λ1 (λ2 − λ3 ) A+B+C +D+6=0 {p12 , p36 , p45 }
τ11 + 2τ12 = 1 λ2 − λ3 = λ1 (1 − λ3 ) −A − B + C + D + 6 = 0 {p13 , p24 , p56 }
τ11 + 2τ12 + τ22 − (τ11 τ22 − τ12
2
) = 1 λ1 (λ2 − λ3 ) = λ2 (1 − λ3 ) A−B−C −D+6=0 {p13 , p26 , p45 }
2
4τ12 + 3(τ11 τ22 − τ12 )=1 λ2 − λ3 = λ2 (λ1 − λ3 ) −A + B − C + D + 6 = 0 {p14 , p35 , p26 }
4τ12 + τ22 + 3(τ11 τ22 − τ122
)=1 λ1 − λ3 = λ1 (λ2 − λ3 ) A+B+C −D+6=0 {p16 , p24 , p35 }
τ11 + 4τ12 + 3(τ11 τ22 − τ122
)=1 λ2 − λ3 = −λ3 (λ1 − λ2 ) −A − B + C − D + 6 = 0 {p14 , p25 , p36 }
2
τ11 + 4τ12 + τ22 + 2(τ11 τ22 − τ12 ) = 1 λ1 − λ2 = −λ1 (λ2 − λ3 ) A−B−C +D+6=0 {p16 , p25 , p34 }
Table 3. Components of H4 with matching constraints for the GH-

quartic
On the other hand, for each genus-two curve C0 with an elliptic involution there is
precisely one Göpel group G ⩽ A[2] with A = Jac (C0 ) such that A/G ≅ E1 × E2 ; see
Corollary 1.17. We then have the following:
Corollary 2.29. For a smooth genus-two curve C0 admitting an elliptic involution
and A = Jac (C0 ), the group G = {P0 , Pij , Pkl , Pmn } ⩽ A[2] is the unique Göpel group
such that A/G ≅ E1 × E2 if and only if for the corresponding Kummer configuration
(P2 ; ℓ1 , . . . , ℓ6 ) on KA the three nodes {pij , pmn , pkl } are collinear.
Remark 2.30. Table 3 shows the 15 components in H2 /Γ2 (2) covering H4 and the
corresponding collinear points in the Kummer plane, using the marking of Weierstrass
points (P1 , . . . , P6 ) = (λ1 = λ2 λ3 , λ2 , λ3 , 0, 1, ∞) for the curve C0 .
Proof. The one-to-one correspondence between the possible configurations of collinear
points in the Kummer plane and the 15 components in H2 /Γ2 (2) covering H4 is
computed explicitly using their respective characterizations in terms of Rosenhain
roots in Table 2 and Table 1, respectively.
Remark 2.31. For A2 = B 2 = C 2 = 1 and D = 0 in Equation (2.43) one finds
the special solutions in Table 4. The minimal resolution of Equation (2.43) is then
isomorphic to the Kummer surface Kum(Jac C0 ) of Picard rank ρ = 20 where C0 is
given by Equation (1.18) with λ2 , λ3 as specified in the table. In this case, the two
elliptic-curve quotients are isomorphic to the elliptic curve E, given by
E∶ y 2 z = x3 + z 3 ,
admitting a Z3 -symmetry x ↦ ω3 x for ω33 = 1.
x λ2 λ3 A B C D Λ1 Λ2 Λ1 + Λ2 j(E1 ) = j(E2 ) ρ
±1 −1 3 1 1 1 0 1 1 0 20
±1 3 −1 1 −1 −1 0 1 1 0 20
±1 −1 31 −1 1 −1 0 1 1 0 20
±1 31 −1 −1 −1 1 0 1 1 0 20
Table 4. Solutions for A2 = B 2 = C 2 = 1 and D = 0
2.3.1. Related normal form. The Shioda sextic W0 in Equation (2.28) has a presenta-
tion as a simple quartic hypersurface, isomorphic to W0 over Q(λ2 + λ3 , λ2 λ3 ). Notice
that this is not the case for the Göpel-Hudson quartic in Equation (2.43) – there, a
finite field extension is needed to construct such an isomorphism.
Each elliptic curve El for l = 1, 2, as given in Legendre form in Equation (1.20), can
also be represented as a complete quadric intersection in P3 . Let I1 be the complete
intersection of the two quadric surfaces in P3 = P(X00 , X01 , X10 , X11 ), given by
⎧
⎪
⎪ X01 = X10 + X11 ,
2 2 2
I1 ∶ ⎨ 2
⎪
⎩ X00 = X10 + (1 − Λ1 )X11 ,
(2.47)
⎪ 2 2
and I2 be the complete intersection in P3 = P(Y00 , Y01 , Y10 , Y11 ), given by

⎧
⎪
⎪ Y01 = Y10 + Y11 ,
2 2 2
I2 ∶ ⎨ 2
⎪
⎩ Y00 = Y10 + (1 − Λ2 )Y11 .
(2.48)
⎪ 2 2

Lemma 2.32. El is birational equivalent to Il for l = 1, 2 over Q(Λl ).
Proof. A rational map E1 ↪ P3 , [x1 ∶ y1 ∶ z1 ] ↦ [X00 , X01 , X10 , X11 ] is given by
(2.49) [X00 ∶ X01 ∶ X10 ∶ X11 ] = [x21 −2Λ1 x1 z1 +Λ1 z12 ∶ x21 −Λ1 z12 ∶ x21 −2x1 z1 +Λ1 z12 ∶ ±2y1 z1 ] .
It has a rational inverse I1 ⇢ E1 , [X00 , X01 , X10 , X11 ] ↦ [x1 ∶ y1 ∶ z1 ], given by
(2.50) [x1 ∶ y1 ∶ z1 ] = [Λ1 (X1,0 − X0,0 ) ∶ ±Λ1 (Λ1 − 1)X1,1 ∶ (1 − Λ1 )X0,1 + Λ1 X1,0 − X0,0 ] .
Thus, we obtain a birational equivalence between E1 and I1 . An analogous argument
holds for E2 and I2 .
There is a well defined map π̃ ∶ I1 × I2 ⇢ P3 with P3 = P(Z00 , Z01 , Z10 , Z11 ) where
one sets
(2.51) [Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] = [X00 Y00 ∶ X01 Y01 ∶ X10 Y10 ∶ X11 Y11 ] ,
for
(2.52) [X00 ∶ X01 ∶ X10 ∶ X11 ] ∈ I1 , [Y00 ∶ Y01 ∶ Y10 ∶ Y11 ] ∈ I1 .
Theorem 2.33. Assume that Λ1 , Λ2 ∈ P1 /{0, 1, ∞} and Λ1 =/ Λ2 . The image π̃(I1 ×I2 )
in P3 = P(Z00 , Z01 , Z10 , Z11 ) is the quartic projective surface V0 , given by
(2.53)
⎧
⎪ Z400 + (1 − Λ1 )(1 − Λ2 )Z401 + Λ1 Λ2 Z410 + Λ1 Λ2 (1 − Λ1 )(1 − Λ2 )Z411
⎪
⎪
⎪
⎪
⎪ −(2 − Λ − Λ )(Z2 Z2 + Λ Λ Z2 Z2 ) − (2Λ Λ − Λ − Λ )(Z2 Z2 + Z2 Z2 )
V0 ∶ ⎨
⎪
⎪
1 2 1 2 10 11 1 2 1 2
⎪
00 01 00 11 01 10
⎪
⎪
⎪ −(Λ1 + Λ2 )(Z00 Z10 + (1 − Λ1 )(1 − Λ2 )Z01 Z11 ) = 0 .
⎩
2 2 2 2
The minimal resolution is isomorphic to a Kummer surface of Picard rank 18. In

particular, the surfaces V0 and W0 in Equation (2.28) for parameters satisfying (1.21)
are birational equivalent over Q(λ2 λ3 , λ2 + λ3 ).
Proof. We multiply (pairwise) Equations (2.47) and (2.47) and then use the vari-
ables in Equation (2.51). Upon eliminating the remaining variables we obtain Equa-
tion (2.53). For elliptic moduli Kl and complementary elliptic moduli Kl′ with
Λl = Kl2 = √1 − (Kl′ )2 ∈ P1 /{0, 1, ∞} for l = 1, 2, the surface V0 is isomorphic over
√
Q( K1 K2 , K1′ K2′ ) to the equation in P3 = P(w, x, y, z) given by
(K1′ )2 + (K2′ )2 2 2
0 = w4 + x4 + y4 + z4 − (w z + x2 y2 )
K1′ K2′
(2.54)
K12 + K22 2 2 (K1 K2′ )2 + (K1′ K2 )2 2 2
− (w y + x2 z2 ) + (w x + y2 z2 ) .
K1 K2 K1 K2 K1′ K2′
In fact, in Equation (2.53) we can rescale
Z01 = √ ′ ′ , Z10 = √ Z11 = √ √
y z w
(2.55) Z00 = x , , ,
K1 K2 K1 K2 K1 K2 K1′ K2′
and obtain Equation (2.54). We check that the latter is a Göpel-Hudson quartic in
Equation (2.43) with D = 0. According to Proposition 2.26, its minimal resolution is
a Jacobian Kummer surface of Picard rank 18.
For parameters satisfying Equations (1.21), a map V0 → U0 , given by
[Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] ↦ [W ∶ X ∶ Y ∶ Z] ,
with W = Z0,1 − Z1,0 − Z1,1 ,
(2.56) X = −(1 − λ2 )(1 − λ3 )Z0,0 + (1 + λ2 λ3 )Z0,1 − (λ2 + λ3 )Z1,0 ,
Y = λ2 λ3 (Z0,1 − Z1,0 + Z1,1 ) ,
Z = −λ2 λ3 ((1 − λ2 )(1 − λ3 )Z0,0 + (1 + λ2 λ3 )Z0,1 − (λ2 + λ3 )Z1,0 ) ,
is an isomorphism, defined over Q(λ2 λ3 , λ2 + λ3 ), between the quartic surface V0
in Equation (2.53) and U0 in Equation (2.21) with λ1 = λ2 λ3 . Substituting Equa-
tions (2.56) into Equation (2.21), we obtain Equation (2.53) up to a non-vanishing
scale factor. Inverting Equations (2.56) yields

Z0,0 = λ2 λ3 X + Z ,
Z0,1 = λ2 λ3 (λ2 + λ3 )W − λ2 λ3 X + (λ2 + λ3 )Y + Z ,
Z1,0 = λ2 λ3 (λ2 λ3 + 1)W − λ2 λ3 X + (λ2 λ3 + 1)Y + Z ,
(2.57)
Z1,1 = −λ2 λ3 (1 − λ2 )(1 − λ3 )W + (1 − λ2 )(1 − λ3 )Y .

Moreover, Equations (2.23) and (2.24) provide a birational equivalence between U0
and W0 . It is easy to see that for λ1 = λ2 λ3 in Equations (2.19) this equivalence is
well defined over Q(λ2 λ3 , λ2 + λ3 )
Remark 2.34. The surface V0 in Equation (2.53) is the Kummer-surface analogue
of the genus-two curve in Equation (1.35) obtained by Legendre’s gluing method.
2.4. Geometric isogeny. We proved in Proposition 1.8 that the smooth curve C0
in Equation (1.18) admits an elliptic involution with the elliptic-curve quotients El
for l = 1, 2 and rational quotient maps πEl ∶ C0 ⇢ El . In this situation, we consider the
rational map, given by
(2.58) ψ0 ∶ C0 × C0 Ð→ E1 × E2 , (P, Q) ↦ (πE1 (P ) ⊕ πE1 (Q), πE2 (P ) ⊕ πE2 (Q)) ,
where the symbol ⊕ refers to the addition of two points on the elliptic curve E1 and
E2 , respectively. One can easily show that the argument from [60, Sec. 4.6] extends,
and the map ψ0 induces a geometric isogeny ψ between the Shioda sextic W0 in
Equation (2.15) with λ1 = λ2 λ3 and the double quadric surface Z in Equation (2.3). If
we additionally use the fact that the Jacobian Jac (C0 ) is birational to the symmetric
product Sym2 (C0 ), it follows that the map ψ is related to the (2, 2)-isogeny Ψ in
Equation (1.40) by the following commutative diagram:
C0 × C0 Ð→ W0 ←Ð Kum(Jac C0 ) ←Ð Jac (C0 )
(2.59) ψ0 ↓ ↓ψ ↓ ↓Ψ
E1 × E2 Ð→ Z ←Ð Kum(E1 × E2 ) ←Ð E1 × E2
Here, the middle horizontal arrows represent minimal resolutions of singularities. We
will now show that, relative to the quartic surface V0 in Equation (2.53), the geometric
isogeny ψ will take a simple form and can be constructed explicitly.
By construction of the√ map ψ0 and Proposition 1.8, the isogeny ψ is defined over
the field Q(q, r 2 ) ≅ Q( λ2 λ3 , λ1 + λ2 ) for q 2 = λ2 λ3 and r 2 = (1 − λ2 )(1 − λ3 ). We then
have the following:
Lemma 2.35. For the holomorphic two-forms ωW0 = dz2 ∧ dz3 /z̃4 in the chart z1 = 1
in Equation (2.15) and ωZ = dx1 ∧ dx2 /y1,2 in the chart z1 = z2 = 1 in Equation (2.3),
it follows
ψ∗ ( )=δ⋅
dx1 ∧ dx2 dz2 ∧ dz3
(2.60)
y12 z̃4
√
with δ = 2 λ2 λ3 (1 − λ2 )(1 − λ3 ).
Proof. The smooth genus-two curve C0 given by Equation (1.18) is defined by a poly-
nomial of degree five. Thus, in the chart Z = 1, the regular differentials from a vector
space with a basis dX/Y and XdX/Y . Similarly, in the charts zl = 1 on the elliptic
curves El holomorphic one-forms are given by dxl /yl for l = 1, 2. One checks by a
direct computation that their pullbacks are give by
p∗l ( ) = (pl1 X + pl0 ) = (rX − (−1)l rq)
dxl dX dX
(2.61) .
yl Y Y
The independence of the pullbacks is equivalent to the non-vanishing of the quantity
(2.62) δ = p21 p10 − p20 p11 = 2r 2 q =/ 0 .
The regular two-from dx1 ∧ dx2 /(y1 y2 ) in Equation (2.4) on E1 × E2 is the pullback
(along the horizontal arrow in Equation (2.59)) of the holomorphic two-form ωZ . It
was shown in [60, Sec. 4.6] that one has
dX (1) ∧ dX (2) d (X (1)
+ X (2) ) ∧ d (X (1) X (2) )
ψ0∗ ( ) = δ(X (2) − X (1) )
dx1 dx2
∧ = δ ,
y1 y2 Y (1) Y (2) Y (1) Y (2)
where [X (l) ∶ Y (l) ∶ Z (l) ] for l = 1, 2 are the coordinates of the two copies of C0 .
One avoids the pre-factor δ in Equation (2.60) by using quadratic twists:
(ε)
Corollary 2.36. The map ψ extends to a rational map between the surfaces W0 in
4
Equation (2.28) and Z (2 ) in Equation (2.6), i.e.,
4
W0 ⇢ Z (2 ) , ε=
(ε) 4
λ2 λ3 (1 − λ2 )2 (1 − λ3 )2
(2.63) ψ∶ with ,
such that the holomorphic two-forms ωW (ε) = dz2 ∧ dz3 /ẑ4 in the chart z1 = 1 and
0
ωZ (24 ) = dx1 ∧ dx2 /ŷ1,2 in the chart z1 = z2 = 1 satisfy ωW (ε) = ψ ∗ ωZ (24 ) .
0
(ε)
Proof. An isomorphism √ between W0 in Equation (2.15) and W0 in Equation (2.28)
is given by z̃4 = ẑ4 / ǫ. Similarly, an isomorphism between Z in Equation (2.15)
4
and Z (2 ) in Equation (2.28) is given by y1,2 = ŷ1,2 /4. Equation (2.60) then becomes
ωW (ε) = ψ ∗ ωZ (24 ) .
0
On the other hand, the double quadric Z in Equation (2.3) is related to the surface
V0 in Equation (2.53) as follows:
Lemma 2.37. In the commutative diagram
E1 × E2 Ð→ I1 × I2
≅
(2.64) π↓ ↓ π̃
Z Ð→ V0
φ±
the maps φ± ∶ Z ⇢ V0 are rational maps of degree two given by

φ± ∶ (x1 , z1 , x2 , z2 , y1,2 ) ↦ [Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] ,
with Z00 = (x21 − 2Λ1 x1 z1 + Λ1 z12 )(x22 − 2Λ2 x2 z2 + Λ2 z22 ) ,

(2.65) Z01 = (x21 − Λ1 z12 )(x22 − Λ2 z22 ) ,
Z10 = (x21 − 2x1 z1 + Λ1 z12 )(x22 − 2x2 z2 + Λ2 z22 ) ,
Z11 = ±4y1,2 ,
such that φ± ○ π ○ ıEl = φ∓ ○ π for l = 1, 2 where ıEl is the hyperelliptic involution on El .
Proof. Equation (2.65) follows immediately from the equivalence E1 ×E2 ≅ I1 ×I2 given
by Lemma 2.32 (with a choice ±1 for the relative sign of y1 and y2 ) and the fact that
the projection map π ∶ E1 × E2 → Z is given by y1,2 = z1 z2 y1 y2 . The fact that the map
is a rational map of degree two follows directly from the fact that Equation (2.3) is a
double quadric surface.
For the elliptic curve El with Λl ∈ P1 /{0, 1, ∞} for l = 1, 2, multiplication by two is
given by the composition of two-isogenies χ′El ○ χEl , or, equivalently, by the map
xl ↦ 2(x2l − Λl zl ) yl zl ,
2
(2.66) yl ↦ (x2l − 2Λl xl zl + Λl zl2 )(x2l − 2xl zl + Λl zl2 )(x2l − Λl zl2 ) ,

zl ↦ 8yl3 zl3 .
The two-isogenies χEl can then be expressed as rational maps originating from the
complete intersection of quadrics I1 in Equation (2.47) (resp. I2 in Equation (2.48)).
It follows that
(2.67) χ′E1 ○ χE1 ∶ [X00 , X01 , X10 , X11 ] ↦ [x1 ∶ y1 ∶ z1 ] = [X20,1 X1,1 ∶ X0,0 X0,1 X1,0 ∶ X31,1 ] ,
and a similar formula holds for χ′E2 ○χE2 . The composition of their Cartesian products,
given by
χ′E ×χ′E
I1 × I2
χE ×χE
E1′ × E2′ E1 × E2 ,
1 2 1 2
(2.68) Ð→ Ð→
induces a rational map from V0 to Z. We have the following:
Proposition 2.38. The family of rational maps ψ± ∶ V0 ⇢ Z of degree two given by
ψ± ∶ [Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] ↦ (x1 , z1 , x2 , z2 , y1,2 ) ,
with x1 = Q , z1 = (Λ1 − Λ2 )Z21,1 ,
(2.69)
x2 = (Λ1 − Λ2 )Z20,1 , z2 = Q ,
y1,2 = ±(Λ1 − Λ2 ) Q Z0,0 Z0,1 Z1,0 Z1,1 ,
2 2
with
(2.70) Q = Z20,0 − (1 − Λ1 )Z20,1 − Λ1 Z21,0 + Λ1 (1 − Λ2 )Z21,1 ,
satisfies that the composition ψ± ○ φ± ∶ Z ⇢ Z is the diagonal action of multiplication
by two on E1 × E2 in Equation (2.66).
Proof. The composition of elliptic-curve isogenies in Equation (2.68) is given by

[X00 ∶ X01 ∶ X10 ∶ X11 ] [x1 ∶ z1 ∶ y1 ] = [X20,1 X1,1 ∶ X31,1 ∶ X0,0 X0,1 X1,0 ]
( )↦( ).
[Y00 ∶ Y01 ∶ Y10 ∶ Y11 ] [x2 ∶ z2 ∶ y2 ] = [Y0,1 ∶ Y0,0 Y0,1 Y1,0 ]
(2.71) 2 3
Y1,1 ∶ Y1,1
The relation, given by
(2.72) [Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] = [X00 Y00 ∶ X01 Y01 ∶ X10 Y10 ∶ X11 Y11 ] ,
induces a correspondence between Z and V0 , given by
x1 x2 = Z20,1 , z1 z2 = Z21,1 , y1,2 = Z0,0 Z0,1 Z1,0 Z1,1 ,
(2.73)
(Λ1 − Λ2 )x1 z2 = Q , (Λ2 − Λ1 )x2 z1 = R , QR = −(Λ1 − Λ2 )2 Z20,1 Z21,1 ,
with Q = (Λ1 − Λ2 )X201 Y11
2
and R = (Λ2 − Λ1 )X211 Y01
2
. One then obtains
Q = Z20,0 − (1 − Λ1 )Z20,1 − Λ1 Z21,0 + Λ1 (1 − Λ2 )Z21,1 ,
(2.74)
R = Z20,0 − (1 − Λ2 )Z20,1 − Λ2 Z21,0 + Λ2 (1 − Λ1 )Z21,1 .
When solving for an equivalence class, given by
(2.75) (x1 , z1 , x2 , z2 , y1,2 ) ∼ (µx1 , µz1 , νx2 , νz2 , µ2 ν 2 y1,2 ) with µ, ν ∈ C× ,
we obtain Equations (2.69). Using the relation QR = −(Λ1 − Λ2 )2 Z20,1 Z21,1 the solution
can also be written as
x1 = (Λ2 − Λ1 )Z20,1 , z1 = R , x2 = R , z2 = (Λ2 − Λ1 )Z21,1 ,
(2.76)
y1,2 = (Λ1 − Λ2 )2 R2 Z0,0 Z0,1 Z1,0 Z1,1 .
Equation (2.66) provides an explicit formula for the (diagonal) action of multiplication
by two on E1 × E2 . We apply the natural projection map π ∶ E1 × E2 → Z given by y1,2 =
z1 z2 y1 y2 to obtain the induced action on the the double quadric Z in Equation (2.3).
We check that this agrees with the composition of maps ψ± ○ φ± ∶ Z ⇢ Z, i.e., we have
(2.77) ψ± ○ φ± = π ((χ′E1 × χ′E2 ) ○ (χE1 × χE2 )) .
The map ψ− in Equation (2.69) can be related to the geometric isogeny ψ in Equa-
tion (2.59). We have the following:
Proposition 2.39. The map ψ in Equation (2.59) with the normalization determined
by Equation (2.60) coincides with ψ− in Proposition 2.38.
Proof. Considering the finite field extension Q(k1 , k2 ) with
(k2 − k3 )2 (k2 + k3 )2
(1 − k22 )(1 − k32 ) (1 − k22 )(1 − k32 )
(2.78) Λ1 = − , Λ2 = − ,
the maps ψ± ∶ V0 ⇢ Z in Proposition 2.38 are equivalently given by

(2.79) ψ± ∶ [Z00 ∶ Z01 ∶ Z10 ∶ Z11 ] ↦ (x1 , z1 , x2 , z2 , y1,2 ) ,
with
x1 = S + , z1 = 4k2 k3 (1 − k22 )(1 − k32 )Z21,1 ,
(2.80) x2 = S − , z2 = 4k2 k3 (1 − k22 )(1 − k32 )Z21,1 ,
y1,2 = ±256k24 k34 (1 − k22 )2 (1 − k32 )2 Z0,0 Z0,1 Z1,0 Z51,1 ,
where S± are the quadrics given by
S± = ±(1 − k22 )2 (1 − k32 )Z20,0 ∓ (1 ∓ k2 k3 )2 (1 − k22 )(1 − k32 )Z20,1
(2.81)
± (k2 ∓ k3 )2 (1 − k22 )(1 − k32 )Z21,0 ∓ (k2 ∓ k3 )2 (1 ± k2 k3 )2 Z21,1 ,
such that S+ S− = 16k22 k32 (1 − k22 )2 (1 − k32 )2 Z20,1 Z21,1 . The proof follows from a tedious
computation using Maple: one first constructs the rational map ψ in Equation (2.59)
as a map from W0 in the chart z1 = 1 in Equation (2.28) to Z in the chart z1 = z2 = 1
in Equation (2.3) over Q(k2 , k3 ). One then uses the identification W0 ≅ V0 from the
proof of Proposition 2.33. The map ψ then simplifies considerably and coincides with
ψ− in Equation (2.80).
Legendre’s gluing method in Section 1.3.4 determines the smooth genus-two curve
in Equation (1.35), admitting an elliptic involution with elliptic-curve quotients E1
and E2 . The quotient maps determine the (2, 2)-isogeny Ψ ∶ Jac (C0 ) → E1 × E2 in
Equation (1.40) and its dual isogeny Φ. The construction descends and yields the
following Kummer sandwich theorem:
Theorem 2.40. The surface V0 in Equation (2.53) and rational maps φ± , ψ± in Equa-
tions (2.65) and (2.69) fit into the following Kummer sandwich:
Kum (E1 × E2 ) Ð→ V0 Ð→ Kum (E1 × E2 ) .
φ± ψ±
The maps φ± , ψ± are defined on families over Q(Λ1 , Λ2 ) and induced (up to the action
of the minus identity involution) by the (2, 2)-isogeny in Equation (1.8) and its dual
(2, 2)-isogeny in Equation (1.40) of principally polarized abelian surfaces, respectively,
i.e., the isogenies in the following diagram:
E1 × E2 Ð→ Jac (C0 ) Ð→ E1 × E2 .
Φ Ψ
(2.82)
C0 is the smooth genus-two curve admitting an elliptic involution, and El are the
elliptic-curve quotients determined in Proposition 1.8.
Proof. General results in [6, Sec. 1] and Theorem 1.2 show that there is a (2, 2)-isogeny
Φ in Equation (1.8), given by
(2.83) Φ∶ E1 × E2 Ð→ Jac (C0 ) .
The dual isogeny Ψ ∶ Jac (C0 ) → E1 × E2 was constructed in Equation (1.40). Up to
isomorphisms, the composition is given by multiplication by two on each factor of
E1 × E2 . Proposition 2.39 shows that Ψ induces the map ψ± . The minimal resolution
of the double quadric surface Z is isomorphic to the Kummer surface Kum(E1 × E2 ).
Moreover, the induced action of Ψ ○ Φ on the associated Kummer surfaces coincides
with the one of ψ± ○ φ± due to Proposition 2.38. On the other hand, the map ψ± ○ φ±
factors though V0 by construction. Proposition 2.33 proves that V0 is isomorphic to

W0 whose minimal resolution is Kum(C0 ) due to Proposition 2.12.
3. K3 surfaces related to the Legendre pencil
The family of projective surfaces, given by the equation
(3.1) y 2 = z1 z2 z3 (z1 + z2 )(z2 + z3 )(z1 + tz3 )
with t ∈ Q is a family of double sextic surfaces considered by van Geemen and Top
in [60]. The minimal resolutions yield K3 surfaces of Picard rank 19. A natural
generalization, referred to as the two-parameter twisted Legendre pencil in [25–29], is
the family X , given by
(3.2) X∶ y 2 = z1 z2 (z1 − z2 )(z1 − z3 ) (z32 + 2ρ1 z2 z3 + ρ22 z22 )
with ρ1 , ρ2 ∈ Q and ρ1 =/ ±ρ2 . If fact, changing variables according to z1 ↦ −z1 ,
z3 ↦ 2ρ1 z3 , and y ↦ 2iρ1 y, the family X is isomorphic to the family in Equation (3.1)
for (ρ1 , ρ2 ) = (t/2, 0) over Q(i). A natural holomorphic two-form ωX on X is given
by dz1 ∧ dz3 /y in the chart z2 = 1 in Equation (3.2). Notice that a general member of
X is not the Kummer associated with an abelian surface.
3.1. Configurations of six lines. The general member of the family in Equa-
tion (3.2) yields a K3 surface of Picard rank 18, and thus its transcendental lattice
TX has rank 4. Hoyt showed in [26, 29] that
(3.3) TX = ⟨2⟩ ⊕ ⟨2⟩ ⊕ ⟨−2⟩ ⊕ ⟨−2⟩ .
There is also a further generalization, known as three-parameter twisted Legendre
pencil, given by
(3.4) y 2 = z1 (z1 − z2 )(z1 − z3 )(Az2 − z3 )(Bz2 − z3 )(Cz2 − z3 ) .
The branch locus consists of six lines, three of which are coincident in a point; see
[13, 26, 40]. The minimal resolutions are K3 surfaces of Picard rank greater than or
equal to 17. In [29] it was shown that the transcendental lattice of the K3 surface
associated with a general member in Equation (3.4) is ⟨2⟩⊕2 ⊕ ⟨−2⟩⊕3 . In particular,
a general member of X is not the Kummer surface.
The case C = ∞ in Equation (3.4), after rescaling, is given by
(3.5) X∶ y 2 = z1 z2 (z1 − z2 )(z1 − z3 )(Az2 − z3 )(Bz2 − z3 ) .
√
This pencil is isomorphic to Equation (3.2) over Q(ρ1 , ρ2 , ρ21 − ρ22 ) with AB = ρ22 and
A + B = −2ρ1 . We have the following:
Proposition 3.1. For A =/ B the general member in Equation (3.5) is branched on
the union of six lines, one of which is coincident with two pairs of lines in two distinct
points and generic otherwise. The configuration is shown in Figure 3. The minimal
resolution has Picard rank 18. For the following specializations the minimal resolution
has Picard rank 19:
(1) AB = 0, (2) (A + B − 4)2 − 4AB = 0, (3) (A − B + 1)(A − B − 1) = 0.
ℓ2 ℓ3 t
✍ tt
✍✍tt
ℓ4 ❖❖❖ t✍t✍t
❖❖❖ tt✍
❖❖❖ tt ✍
❖❖❖ ttt ✍✍✍
t
ℓ5 ❩❩❩❩❩❩❩❩❩❩❩❖❩❖❖❖❖ tttt ✍
❩❩❩❩❖❩t❖❩t❩ ✍✍
tt ❖❖❖ ❩❩❩❩❩❩✍✍❩
t ❖❖❖ ✍ ❩❩❩❩❩❩❩❩❩
tt ❖✍❖ ❩❩❩
ttt ✍✍ ❖❖❖❖
ℓ6 tt ✍ ❖ ❖❖❖
tt ✍✍
tt ❖❖❖
ttt ✍✍ ✍ ❖❖❖
tt ✍ ❖
ℓ1
Figure 3. Line configuration for general member in Proposition 3.1
Remark 3.2. In terms of Equation (3.2) the subsets of the parameter space in Propo-
sition 3.1 are given by
(1) ρ2 = 0, (2) (ρ1 − ρ2 + 2)(ρ1 + ρ2 + 2) = 0, (3) 4ρ21 − 4ρ22 − 1 = 0.
Remark 3.3. In case (1) of Proposition 3.1 the branch locus is the configuration
of six lines {ℓ1 , . . . , ℓ6 } in P2 such that, up to permutation, the triple intersections
{p123 } = ℓ1 ∩ ℓ2 ∩ ℓ3 , {p145 } = ℓ1 ∩ ℓ4 ∩ ℓ5 , and {p246 } = ℓ2 ∩ ℓ4 ∩ ℓ6 consist of three
pairwise distinct points p123 , p145 , p246 , and the configuration is generic otherwise.
Proof. The branch locus of Equation (3.5) is given by the six lines
(3.6) ℓ1 ∶ z2 , ℓ2 ∶ z1 , ℓ3 ∶ z1 − z2 , ℓ4 ∶ Az2 − z3 , ℓ5 ∶ Bz2 − z3 , ℓ6 ∶ z1 − z3 ,
whose intersection pattern looks as follows:
ℓ1 ℓ2 ℓ3 ℓ4 ℓ5 ℓ6
ℓ1 − [0 ∶ 0 ∶ 1] [0 ∶ 0 ∶ 1] [1 ∶ 0 ∶ 0] [1 ∶ 0 ∶ 0] [1 ∶ 0 ∶ 1]
ℓ2 [0 ∶ 0 ∶ 1] − [0 ∶ 0 ∶ 1] [0 ∶ 1 ∶ A] [0 ∶ 1 ∶ B] [0 ∶ 1 ∶ 0]
(3.7) ℓ3 [0 ∶ 0 ∶ 1] [0 ∶ 0 ∶ 1] − [1 ∶ 1 ∶ A] [1 ∶ 1 ∶ B] [1 ∶ 1 ∶ 1]
ℓ4 [1 ∶ 0 ∶ 0] [0 ∶ 1 ∶ A] [1 ∶ 1 ∶ A] − [1 ∶ 0 ∶ 0] [A ∶ 1 ∶ A]
ℓ5 [1 ∶ 0 ∶ 0] [0 ∶ 1 ∶ B] [1 ∶ 1 ∶ B] [1 ∶ 0 ∶ 0] − [B ∶ 1 ∶ B]
ℓ6 [1 ∶ 0 ∶ 1] [0 ∶ 1 ∶ 0] [1 ∶ 1 ∶ 1] [A ∶ 1 ∶ A] [B ∶ 1 ∶ B] −
Statement (1) is immediate. For (2) one can check that the double sextic in Equa-
tion (3.5) admits a Jacobian elliptic fibration with a non-torsion section: the elliptic
fibration is obtained by setting z1 = x, z2 = t, z3 = 1. Then, for x = − 4(At−1)(Bt−1)
(A−B)2 t we
have
4(At − 1)2 (Bt − 1)2 ((A + B)t − 2)2 (4ABt2 + (A2 + B 2 − 2AB − 4A − 4B)t + 4)
y2 = −
(A − B)6 t2
.
The discriminant for the quadratic term in the numerator of y 2 factors into the
product of (A − B)2 times A2 + B 2 − 2AB − 8A − 8B − 16. Since we assumed A =/ B we
obtain a non-torsion section if A2 + B 2 − 2AB − 8A − 8B − 16 = 0. Thus, we set A = α2

and B = (2 ± α)2 and obtain
i(α2 t − 1)((α − 2)2 t − 1)((α2 − 2α + 2)t − 1)(α(α − 2)t + 1)
y=± .
8(α − 1)3 t
2 2 2
A(At−1) (Bt−1) t
For (3) one checks that x = At−1
A−B and y =
2
(A−B)3 is a section if and only if
A − B + 1 = 0. Switching the roles of A and B the statement follows.
In Equation (3.5) the pencil of lines through the point [1 ∶ 0 ∶ 0] (or, equiva-
lently, the point [0 ∶ 0 ∶ 1]) induces an elliptic fibration on X . A simple coordinate
transformation, defined over Q(ρ1 , ρ22 ), yields a Weierstrass model with fibers over
P1 = P(v1 , v2 ) embedded into P2 = P(x, y, z), given by
(3.8) X∶ y 2 z = x(x − v1 v2 (v12 + 2ρ1 v1 v2 + ρ22 v22 )z)(x − v22 (v12 + 2ρ1 v1 v2 + ρ22 v22 )z) .
This change of coordinates also maps via pullback the holomorphic two-form ωX =
dv1 ∧ dx/y in the chart v2 = z = 1 in Equation (3.8) to the holomorphic two-form
dz1 ∧ dz3 /y in the chart z2 = 1 in Equation (3.2). The discriminant function of the
fibration is given by v28 v12 (v1 − v2 )2 (v12 + 2ρ1 v1 v2 + ρ22 v22 )6 . We have the immediate:
fibers I2∗ + 2I2 + 2I0∗ and a Mordell-Weil group of sections (Z/2Z)2 .
Remark 3.5. The pencil of lines through the point [z1 ∶ z2 ∶ z3 ] = [1 ∶ 0 ∶ 1] induces
a second Jacobian elliptic fibration on X with the singular fibers I4∗ + 4I2 + I0∗ and a
Mordell-Weil group of sections (Z/2Z)2 .
Remark 3.6. Hoyt proved in [28] that there are countably many algebraic curves
WN ⊂ C2 in the parameter space {(A, B) ∣ A =/ B and A, B =/ 0, 1} for Equation (3.5)
such that each WN corresponds to a modular correspondence between two elliptic
curves. Moreover, it was shown that the rank of the Mordell-Weil group in Lemma 3.4
is greater than zero if and only if (A, B) ∈ ⋃N WN . We will provide the modular cor-
respondences for the subsets of the parameter space in Proposition 3.1 in Section 3.5.
Remark 3.7. The transcendental lattice in case (1) of Proposition 3.1 is isomorphic
to ⟨2⟩⊕⟨2⟩⊕⟨−2⟩. For the remaining subfamilies in Proposition 3.1 the transcendental
lattices can be determined using the results of [29, Sec. 3].
3.2. Double coverings by Kummer surfaces. We recall that an even eight on a
K3 surface X is a set of eight disjoint (−2)-rational smooth curves E1 , . . . , E8 such
that there is a divisor D ∈ NS(X ) with
(3.9) E1 + ⋅ ⋅ ⋅ + E8 ∼ 2D ,
where the symbol ∼ denotes linear equivalence. Choosing an even eight as the branch
locus on a K3 surface for a two-fold ramified covering, Nikulin’s construction yields a
new K3 surface Y together with a double covering f ∶ Y → X ; see [50]. Equivalently,
there is a Nikulin involution on Y such that the minimal resolution of the quotient
surface of Y by the involution yields X . A Nikulin involution is an analytic involution

of Y which acts trivially on H 2,0(Y). It is well known that such an involution always
has eight fixed points and that by blowing them up and taking the quotient we recover
the K3 surface X [48].
In our situation, reducible fibers of type D6 or D4 arise in Lemma 3.4 as minimal
resolutions of the fibers located over v2 = 0 and v12 + 2ρ1 v1 v2 + ρ22 v22 = 0, respectively.
Recall that a D4 fiber has one (central) component of multiplicity two and four
reduced component, taking the sum of two such fibers one finds that the eight reduced
components are twice the class of a fiber minus the two double components and this
gives the divisibility by two. Thus, configurations of eight non-central components of
these fibers then form even eights on the minimal resolution of X . Since the Mordell-
Weil group of sections in Lemma 3.4 is (Z/2Z)2 , we can always choose the four marked
components in the reducible fibers in such a way that each of them is met by the zero
section or a two-torsion section. In our situation, there are then the following even
eights to consider:
√
(1) non-central components of D6 and D4 over v2 = 0 and v1 = (− ρ21 − ρ22 + ρ1 )v2
meeting sections, √
(2) non-central components of D6 and D4 over v2 = 0 and v1 = (+ ρ21 − ρ22 + ρ1 )v2
meeting sections,
(3) non-central components of D4 ’s over v12 + 2ρ1 v1 v2 + ρ22 v22 = 0.
Choosing any of these even eights as the branch locus of a double cover, Nikulin’s
construction yields a new K3 surface obtained as the minimal resolution of a Jacobian
elliptic surface Yn , together with a rational double cover fn ∶ Yn ⇢ X for n = 1, . . . , 3.
In the case of two D4 fibers, the double cover is induced by the double cover of the
base curve P1 branched over the two corresponding base points. The pull-back of the
elliptic fibration, after a normalization, gives the elliptic K3 surface Yn with smooth
fibers over the ramification points. Moreover, the corresponding Nikulin involution
on Yn is the lift of the covering involution on the base composed with fiberwise
multiplication by −1. This involution has eight fixed points, four in each fiber over
the fixed points on the base, and with a quotient that is the K3 surface X one started
with. Similar results hold in the other cases. Garbagnati considered Jacobian elliptic
K3 surfaces with abelian and dihedral groups of symplectic automorphisms, which
includes an isogeny fn ∶ Yn ⇢ X where X contains two reducible D4 fibers as a special
case [21, Sec. 5-6]. Closely related is also work by Mehran [46] who classified all the
K3 surfaces that admit a rational map of degree two into a given Kummer surface.
3.2.1. Double coverings by Kummer surfaces associated with two elliptic curves. Let
us first construct the Jacobian elliptic surface Y3 and the double cover f3 ∶ Y3 → X .
A double cover between two rational curves, defined over Q(ρ1 , ρ2 ), is given by
(3.10) P1 ⇢ P1 , [u1 ∶ u2 ] ↦ [v1 ∶ v2 ] = [(ρ1 − ρ2 )u21 − 2ρ1 u1 u2 + (ρ1 + ρ2 )u22 ∶ 2u1 u2 ] .
The pullback of Equation (3.8) along the double cover in Equation (3.10) yields an
elliptic fibration, with fibers over P1 = P(u1 , u2 ) embedded into P2 = P(X, Y, Z), given
by the Weierstrass model
Y 2 Z = X (X + u1 u2 (u1 − u2 )((ρ1 − ρ2 )u1 − (ρ1 + ρ2 )u2 )Z)
1
Y3 ∶
2
(3.11)
× (X + u1 u2 ((ρ1 − ρ2 )u21 − 2(ρ1 + 1)u1 u2 + (ρ1 + ρ2 )u22 )Z) .
1
2
In fact, Equation (3.11) is obtained by setting
x = ((ρ1 − ρ2 )u21 − (ρ1 + ρ2 )u22 ) (4X + 2u1 u2 (u1 − u2 )((ρ1 − ρ2 )u1 − (ρ1 + ρ2 )u2 )Z) ,
2
(3.12)
y = 8((ρ1 − ρ2 )u21 − (ρ1 + ρ2 )u22 ) Y ,
3
z=Z,
in Equation (3.8). Together, Equations (3.12) and (3.10) determine the double cover
(3.13) f3 ∶ Y3 ⇢ X , ([u1 ∶ u2 ], [X ∶ Y ∶ Z]) ↦ ([v1 ∶ v2 ], [x ∶ y ∶ z]) ,
which is branched along the even eight on X that consists of the non-central compo-
nents of the two reducible fibers of type D4 obtained as the minimal resolution of the
fibers of Kodaira-type I0∗ located over v12 + 2ρ1 v1 v2 + ρ22 v22 = 0. Conversely, on Y3 in
the chart Z = 1 and u2 = 1 a compatible Nikulin involution is given by
(ρ1 + ρ2 )2 X (ρ1 + ρ2 )3 Y
(u1 , X, Y ) ↦ (u′1 = ),
ρ1 + ρ2
, X′ = ′
(ρ1 − ρ2 )u1 (ρ1 − ρ2 )2 u41 (ρ1 − ρ2 )3 u61
(3.14) , Y = −
and it leaves the holomorphic two-forms ωY3 = du1 ∧ dX/Y invariant. We have the
following:
Lemma 3.8. For the holomorphic two-form ωX = dw1 ∧ dx/y in the chart z = w2 = 1
on X in Equation (3.8) and the holomorphic two-form ωY3 = du1 ∧ dX/Y in the chart
Z = u2 = 1 on Y3 in Equation (3.11), we have ωY3 = f3∗ ωX .
Proof. The proof follows by an explicit computation using the transformation in Equa-
tions (3.12) and (3.10).
It turns out that the minimal resolution of Y3 is isomorphic to the Kummer surface
Kum(E1 ×E2 ) associated with the product surface of two non-isogenous elliptic curves.
Proposition 3.9. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} let parameters ρ1 , ρ2 be given by
(3.15) ρ1 = Λ1 + Λ2 − 2Λ1 Λ2 − 1 , ρ2 = 1 − Λ1 − Λ2 .
The surfaces Y3 in Equation (3.11) and Z in Equation (2.3) are birational equivalent
over Q(Λ1 , Λ2 ). The birational equivalence identifies the holomorphic two-form ωY3 =
du1 ∧ dX/Y in the chart Z = u2 = 1 with ωZ = dx1 ∧ dx2 /y1,2 in the chart z1 = z2 = 1.
Proof. Equation (3.11) is then precisely Equation (2.11), describing a Jacobian elliptic
fibration on the double quadric surface Z whose minimal resolution is Kum(E1 × E2 ).
The proof then follows from Proposition 2.6 and Corollary 2.10.
It turns out that the identification of the elliptic fibration in Equation (3.8) and
the isogeny f3 ∶ Y3 → X are key in generalizing the results of [60]; see Section 3.5.
3.2.2. Double coverings by Jacobian Kummer surfaces. Next we construct the K3
surface Y1 and the double cover f1 ∶ Y1 → X . We introduce new parameters µ2 , µ3
such that in Equation (3.8) we have
(µ3 − 1)2 (µ22 + 6µ2 + 1) (µ3 − 1)4 (µ2 + 1)2 µ2
ρ22 =
8(µ2 µ3 − 1)(µ2 − µ3 ) 4(µ2 µ3 − 1)2 (µ2 − µ3 )2
(3.16) ρ1 = , .
Equations (3.16) are sufficient to obtain a family of Weierstrass models from Equa-
tion (3.8) over Q(µ2 , µ3 ) since the latter family only depends on ρ22 .
A double cover P1 → P1 between two rational curves, defined over Q(µ2 , µ3 ), is
given by [u1 ∶ u2 ] ↦ [v1 ∶ v2 ] with
(3.17) [v1 ∶ v2 ] = [µ2 (µ3 − 1)2 (u1 − µ3 u2 )(u1 − µ22 µ3 u2 ) ∶ (µ2 − µ3 )(1 − µ2 µ3 )(u1 − µ2 µ3 u2 )2 ] .
The pullback of Equation (3.8) along the double cover in Equation (3.17) then yields
an elliptic fibration, with fibers over P1 = P(u1 , u2 ) embedded into P2 = P(X, Y, Z),
given by the Weierstrass model
Y1 ∶ Y 2 Z = ǫX (X + µ2 (µ3 − 1)2 u1 u2 (u1 − µ3 u2 )(u1 − µ22 µ3 u2 )Z)
(3.18)
× (X + µ3 (µ2 − 1)2 u1 u2 (u1 − µ2 u2 )(u1 − µ2 µ23 u2 )Z) ,
with 1/ǫ = 4µ2 µ3 (1 − µ2 µ3 )(µ2 − µ3 ). In fact, Equation (3.18) is obtained by setting
µ2 µ3 (1 − µ2 µ3 )(µ2 − µ3 )(µ2 − 1)4 (µ3 − 1)4 (u1 − µ2 µ3 u2 )2 (u2 + µ2 µ3 u2 )2
1
x=
4
(3.19) × (X + µ2 (µ3 − 1)2 u1 u2 (u1 − µ3 u2 )(u1 − µ22 µ3 u2 )Z) ,
µ µ (µ2 − µ3 )2 (1 − µ2 µ3 )2 (µ2 − 1)6 (µ3 − 1)6 (u1 − µ2 µ3 u2 )3 (u1 + µ2 µ3 u2 )3 Y ,

1 2 2
y=
4 2 3
and z = Z in Equation (3.8). Together, Equations (3.19) and (3.17) then determine a
double cover
(3.20) f1 ∶ Y1 ⇢ X , ([u1 ∶ u2 ], [X ∶ Y ∶ Z]) ↦ ([v1 ∶ v2 ], [x ∶ y ∶ z]) ,
which is branched along the even eight on X that consists of non-central components
of the reducible fibers of type D6 and D4 meeting the sections. In terms of the
parameters µ2 , µ3 , the reducible fibers are the minimal resolution of the fibers of
Kodaira-type I2∗ and I0∗ located over
(µ2 + 1)(1 − µ3 )2
4(1 − µ2 µ3 )(µ2 − µ3 )
(3.21) v2 = 0 , and v1 = v2 ,
whereas the remaining fiber of Kodaira-type I0∗ is located over

µ2 (1 − µ3 )2
(1 − µ2 µ3 )(µ2 − µ3 )
(3.22) v1 = v2 .
A compatible Nikulin involution on Y1 can be constructed in the same way as already

demonstrated for the case of Y3 . We have the following:
Lemma 3.10. For the holomorphic two-form ωX = dw1 ∧ dx/y in the chart z = w2 = 1
on X in Equation (3.8) and the holomorphic two-form ωY1 = du1 ∧ dX/Y in the chart
Z = u2 = 1 on Y1 in Equation (3.18), we have ωY1 = f1∗ ωX .
Proof. The proof follows by an explicit computation using the transformation in Equa-
tions (3.19) and (3.17).
Remark 3.11. For the construction of the double cover √ f2 ∶ Y2 → X the induced map
on the base curves is branched over v2 = 0 and v1 = ( ρ21 − ρ22 + ρ1 )v2 . We introduce
new Rosenhain roots µ ̃2 = ̃ ̃3 = ̃
k22 and µ k32 such that
(µ2 + 1)(1 − µ3 )2 ̃2 (1 − µ
µ ̃3 )2
4(1 − µ2 µ3 )(µ2 − µ3 ) (1 − µ
̃2 µ
̃3 )(̃ ̃3 )
= ,
µ2 − µ
(3.23)
µ2 (1 − µ3 )2 (̃
µ2 + 1)(1 − µ ̃3 )2
(1 − µ2 µ3 )(µ2 − µ3 ) 4(1 − µ
̃2 µ
̃3 )(̃ ̃3 )
= .
µ2 − µ
With respect to the new Rosenhain roots the positions of the two fibers of Kodaira-
type I0∗ in Equations (3.21) and (3.22) are interchanged, whereas the equations for
the parameters ρ1 , ρ22 remain unchanged, i.e.,
(̃
µ3 − 1)2 (̃
µ22 + 6̃
µ2 + 1) (̃
µ3 − 1)4 (̃ ̃2
µ2 + 1)2 µ
ρ22 =
̃3 − 1)(̃ ̃3 ) ̃3 − 1) (̃ ̃3 )2
(3.24) ρ1 = , 2
.
8(̃
µ2 µ µ2 − µ 4(̃
µ2 µ µ2 − µ
The solutions of Equation (3.23) are given by
(i ± ̃
k2 )2 (i ± ̃
k3 )2
(i ∓ ̃ (i ∓ ̃
(3.25) µ2 = − , µ3 = − ,
k2 )2 k3 )2
where the sign choices in the formulas for µ2 and µ3 are independent. Thus, all
results related to f2 ∶ Y2 → X can be derived from those for f1 by replacing the moduli
according to Equation (3.25).
Let us also consider the double sextic surface given by
3
(3.26) W0 ẑ42 = ε′ z1 z3 (z1 − z2 + z3 )((µ2 µ3 )2 z1 − µ2 µ3 z2 + z3 ) ∏ (µ2i z1 − µi z2 + z3 ) .
′ (ε′ )
∶
i=2
′ (ε′ )
It follows from Proposition 2.12 that W0
is birational equivalent to the twisted
′ (ε′ )
Kummer surface Kum(Jac C0 ) associated with the Jacobian of the smooth genus-
two curve given by
(3.27) C0′ ∶ Y 2 = XZ(X − Z)(X − µ2 µ3 Z)(X 2 − (µ2 + µ3 )XZ + µ2 µ3 Z 2 ) .
It turns out that the minimal resolution of Y1 is isomorphic to the Kummer surface
Kum(Jac C0′ ). We have the following:
Proposition 3.12. Assume that µ2 , µ3 ∈ P1 /{0, 1, ∞} satisfy µ2 =/ µ±1
3 , and a quadric-
twist factor is given by
1
ε′ =
4µ2 µ3 (1 − µ2 µ3 )(µ2 − µ3 )
(3.28) .
′ (ε′ )
The surfaces Y1 in Equation (3.18) and W0 in Equation (3.26) are birational equiv-
alent over Q(µ2 µ3 , µ2 + µ3 ). The birational equivalence identifies the holomorphic
two-form ωY1 = du1 ∧ dX/Y in the chart Z = u2 = 1 with ωW ′ (ε′ ) = dz2 ∧ dz3 /ẑ4 in the
0
chart z1 = 1.
Proof. Equation (3.18) is precisely Equation (2.39), with λl ’s replaced by µl ’s for
l = 2, 3. The proof then follows from Corollary 2.23. The constraint that the birational
equivalence identifies ωY1 and ωW ′ (ε′ ) determines ε′ .
0
3.3. Kummer surfaces as quotients. For the affine Weierstrass equation for X of
the form y 2 = x3 +p1 x2 +p2 x, the translation by the two-torsion section T ∶ (x, y) = (0, 0)
in each fiber constitutes a Nikulin involution ı, which is given by addition ⊕ of T with
respect to the group law in each smooth elliptic fiber, i.e.,
(x, y) ↦ (x, y) ⊕ T = ( ,− 2 ) .
p2 p2 y
(3.29) ı∶
x x
By resolving the eight nodes of X /⟨ı⟩ we obtain a new K3 surface as a Jacobian elliptic
surface X ∨ , given by the affine Weierstrass model
(3.30) Y 2 = X 3 − 2 p1 X 2 + (p21 − 4p2 ) X .
One constructs a dual van Geemen-Sarti involution ı∨ analogously. The explicit
formulas for the isogeny and the dual isogeny are well known3 and given by
⎛ y 2 y (x2 − p2 ) ⎞
X ⇢ X∨ , (x, y) ↦
⎝ x2 ⎠
(3.31) ϕ∶ , ,
x2
and
⎛ Y 2 Y (X 2 − p21 + 4 p2 )) ⎞
ϕ∨ ∶ X∨ ⇢ X, (X, Y ) ↦
⎝ 4X 2 ⎠
(3.32) , .
8 X2
Proposition 3.13. The K3 surfaces X and X ∨ admit dual van Geemen-Sarti in-
volutions ı and ı∨ associated with fiberwise translations by the two-torsion section
T ∶ (x, y) = (0, 0) and T ∨ ∶ (X, Y ) = (0, 0), respectively, and the following pair of dual
geometric two-isogenies:
ϕ
, s ❞ ❜ ❴ ❭ ❩ ❲ v
(3.33) ı∨ X∨ ❩ ❭ ❴ ❜ ❞ ❣3 X ı
ϕ∨
Moreover, it follows ωX = ϕ∗ ωX ∨ and 2 ωX ∨ = (ϕ∨ )∗ ωX for the holomorphic two-forms

ωX = dv ∧ dx/y and ωX ∨ = dv ∧ dX/Y on X and X ∨ , respectively.
3They were given explicitly in [14]
Applying Proposition 3.13 to Equation (3.8) using the Nikulin involution ı3 asso-
ciated with translations by the two-torsion section
(3.34) T3 ∶ [X ∶ Y ∶ Z] = [v1 v2 (v12 + 2ρ1 v1 v2 + ρ22 v22 ) ∶ 0 ∶ −1] ,
we obtain a new K3 surface X3∨ as the minimal resolution of X /⟨ı3 ⟩ together with a
quotient map ϕ3 ∶ X ⇢ X3∨ defined over Q(ρ1 , ρ22 ). Here, a Weierstrass model for X3∨
is given by
X3∨ ∶ Y 2 Z = X 3 − 2v2 (2v1 − v2 )(v12 + 2ρ1 v1 v2 + ρ22 v22 )X 2 Z
(3.35)
+ v24 (v12 + 2ρ1 v1 v2 + ρ22 v22 ) XZ 2 .
2
It turns out that the minimal resolution of X3∨ is isomorphic to the Kummer surface
Kum(E1′ × E2′ ) associated with the product surface of the two-isogenous elliptic curves
in Equation (1.43). We have the following:
Proposition 3.14. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} let parameters ρ1 , ρ2 satisfy
(3.36) ρ1 = Λ1 + Λ2 − 2Λ1 Λ2 − 1 , ρ22 = (1 − Λ1 − Λ2 )2 .
The surfaces X3∨ in Equation (3.35) and Z ′ in Equation (2.8) are birationally equiv-
alent over Q(κ1 , κ2 ) with Λl = 1/(1 − κ2l ) for l = 1, 2. The birational equivalence
identifies the holomorphic two-form ωX3∨ = dv1 ∧ dX/Y in the chart Z = v2 = 1 with
ωZ ′ = dX1 ∧ dX2 /Y1,2 in the chart Z1 = Z2 = 1.
Proof. Equation (3.35) is precisely Equation (2.13), describing a Jacobian elliptic
fibration on the Kummer surface Z ′ whose minimal resolution is Kum(E1′ × E2′ ). The
proof then follows from Proposition 2.6 and Corollary 2.10.
Applying Proposition 3.13 to Equation (3.8) using the Nikulin involution ı1 as-
sociated with translations by the two-torsion section T1 ∶ [X ∶ Y ∶ Z] = [0 ∶ 0 ∶ 1],
we obtain a new K3 surface X1∨ as the minimal resolution of X /⟨ı1 ⟩ together with a
rational quotient map ϕ1 ∶ X ⇢ X1∨ defined over Q(ρ1 , ρ22 ). Here, a Weierstrass model
for X1∨ , with fibers over P1 = P(v1 , v2 ) embedded into P2 = P(X, Y, Z), is
X1∨ ∶ Y 2 Z = X 3 + 2v2 (v1 + v2 )(v12 + 2ρ1 v1 v2 + ρ22 v22 )X 2 Z
(3.37)
+ v22 (v1 − v2 ) (v12 + 2ρ1 v1 v2 + ρ22 v23 ) XZ 2 .
2 2
The discriminant function of the fibration is 16v1 v27 (v1 − v2 )4 (v12 + 2ρ1 v1 v2 + ρ22 v22 )6 .
Remark 3.15. Applying Proposition 3.13 using the (different) two-torsion section of
the elliptic fibration (3.8) of X , namely
(3.38) T2 ∶ [X ∶ Y ∶ Z] = [v22 (v12 + 2ρ1 v1 v2 + ρ22 v22 ) ∶ 0 ∶ −1] ,
yields a K3 surface with the same singular fibers as X2∨ , but for different moduli. In
fact, the Weierstrass equation coincides with Equation (3.37) if we replace
(3.39) (ρ1 , ρ22 ) ↦ ( − ρ1 − 1, ρ22 + 2ρ1 + 1) .
In terms of Lemma 3.4 the change of parameters is equivalent to the interchange of

the two fibers of Kodaira-type I2 while keeping the fiber of Kodaira-type I2∗ fixed.
It turns out that the minimal resolution of X1∨ in Equation (3.37) is isomorphic to
the Kummer surface Kum(Jac C0 ) associated with the Jacobian of a smooth genus-two
curve admitting an elliptic involution. We have the following:
Proposition 3.16. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 . Further as-
sume that parameters ρ1 , ρ2 satisfy
λ22 λ23 + λ22 + λ23 − 4λ2 λ3 + 1 (λ2 + 1)2 (λ3 + 1)2
ρ22 =
(λ2 − 1)2 (λ3 − 2)2 (λ2 − 1)2 (λ3 − 1)2
(3.40) ρ1 = − , ,
and a quadric-twist factor is given by

4
λ2 λ3 (1 − λ2 )2 (1 − λ3 )2
(3.41) ε= .
(ε)
The surfaces X1∨ in Equation (3.37) and W0 in Equation (2.28) are birational
equivalent over Q(λ2 + λ3 , λ2 λ3 ). The rational equivalence identifies the two-forms
ωX1∨ = dv1 ∧ dX/Y in the chart Z = v2 = 1 and ωW (ε) = dz2 ∧ dz3 /ẑ4 in the chart z1 = 1.
0
Proof. Since the genus-two curve C0 is smooth, we have λ2 , λ3 ∈ P1 /{0, 1, ∞} and

λ2 =/ λ±1
3 . A birational equivalence relating Equation (2.40) and Equation (3.37) is
given by
w1 = −λ2 λ3 (v1 + v2 ) , w2 = (v1 − v2 ) ,
1
2
3
x = (λ2 λ3 )3 (λ2 λ3 − λ2 − λ3 + 1)2 X , y = (λ2 λ2 ) 2 (λ2 λ3 − λ2 − λ3 + 1)3 Y , z = Z .

ε ε2 9
4 8
We observe that for ε = ε21 /(λ2 λ3 ) the above transformation is well defined over
Q(λ2 + λ3 , λ2 λ3 ). The remainder of the proof follows from Corollary 2.24 and by an
explicit computation using the transformations provided above. The constraint that
the birational equivalence identifies ωX1∨ and ωW (ε) determines ε.
0
3.4. Kummer sandwich theorems. In this section we combine the results of the
previous sections to obtain several Kummer sandwich theorems, a term introduced by
Shioda [56], that relate the Legendre pencil to various Kummer surfaces. In [10, 44]
it was demonstrated that the number of points rational over a finite field on the K3
surfaces in a sandwich are the same.
The first Kummer sandwich relates the twisted Legendre pencil X to two Kum-
mer surfaces associated with the product of two elliptic curves. The first one is the
Kummer surface Kum (E1 × E2 ) associated with the product abelian surface E1 × E2 ,
where the elliptic curves El with modular parameter Λl is given by Equations (1.20).
The second is the Kummer surface Kum (E1′ × E2′ ) where El′ for l = 1, 2 are the two-
isogenous elliptic curves in Equation (1.43). Models for the Kummer surfaces are the
double quadric surfaces in Equation (2.3) and Equation (2.8), respectively. We have
the following:
Proposition 3.17. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} let parameters ρ1 , ρ2 satisfy

(3.42) ρ1 = Λ1 + Λ2 − 2Λ1 Λ2 − 1 , ρ22 = (1 − Λ1 − Λ2 )2 .
The twisted Legendre pencil X in Equation (3.2) fits into the following Kummer sand-
wich of rational maps:
Kum(E1 × E2 ) Ð→ X Ð→ Kum(E1′ × E2′ ) Ð→ Kum(E1 × E2 )(2 ) .
4
(3.43)
The maps are defined on families over Q(κ1 , κ2 ) with Λl = 1/(1 − κ2l ) for l = 1, 2. The
holomorphic two-form ωX equals the pullback of ωKum(E1′ ×E2′ ) , resp. its pullback equals
ωKum(E1 ×E2 ) and the pullback of ωKum(E1 ×E2 )24 equals ωKum(E1′ ×E2′ ) .
Proof. We relate the twisted Legendre pencil X to the double quadric surfaces Z and
Z ′ in Equation (2.3) and (2.8), respectively, by relating it to the Jacobian elliptic
K3 surfaces Y3 and X3∨ in Equations (3.11) and (3.35) first. As a reminder, the
minimal resolution of Z is the Kummer surface Kum (E1 × E2 ). Similarly, the minimal
resolution of Z ′ is the Kummer surface Kum (E1′ × E2′ ).
First we assume ρ2 = 1 − Λ1 − Λ2 . We combine Lemma 3.8, Proposition 3.9, Propo-
sition 3.13, Proposition 3.14, Lemma 2.4 to obtain the following diagram of rational
maps:
Z ≅ Y3 Ð→ X X3∨ ≅ Z ′ Ð→ Z (2
f3 ϕ3 χ 4
)
Ð→
(3.44)
≃
≃
Kum(E1 × E2 ) Ð→ X Ð→ Kum(E1′ × E2′ ) Ð→ Kum(E1 × E2 )(2
4
)
By construction, all maps and equivalences are defined over Q(Λ1 , Λ2 ), except the
equivalence X3∨ ≅ Z ′ which is defined only over the field extension Q(κ1 , κ2 ). The
holomorphic two-forms are related by pullback as follows:
(3.45) ω Z = ω Y3 , ωY3 = f3∗ ωX , ωX = ϕ∗3 ωX3∨ , ωX3∨ = ωZ ′ , ωZ ∨ = χ∗ ωZ (24 ) .
Using Equation (2.5) the statement regarding the two-forms follows.
Finally, the modular transformation (Λ1 , Λ2 ) ↦ (1−Λ1 , 1−Λ2 ) leaves the j-functions
of E1 and E2 invariant and induces a sign change (ρ1 , ρ2 ) ↦ (ρ1 , −ρ2 ). Thus, the same
statements hold for (ρ1 , −ρ2 ).
The second Kummer sandwich relates the twisted Legendre pencil X to certain
Jacobian Kummer surfaces. The first is the Kummer surface Kum(Jac C0 ) associated
with the Jacobian of the smooth genus-two curve C0 admitting an elliptic involution
in Equation (1.18) (with Rosenhain roots λ2 , λ3 ). The second is the Kummer surface
Kum(Jac C0′ ) where C0′ is the (2, 2)-isogenous Richelot-curve in Equation (1.47) (with
Rosenhain roots µ2 , µ3 ). That is, we have Jac (C0 ) ≅ Jac (C0′ )/G for a certain Göpel
group determined in Proposition 1.14. Models for the Kummer surfaces are the double
sextic surfaces in Equation (2.28) and (3.26), respectively. We have the following:
Proposition 3.18. The twisted Legendre pencil X in Equation (3.2) fits into the
following Kummer sandwich of rational maps:
Kum(Jac C0′ )(ε ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum(E1 × E2 )(2 ) .
′ 4
(3.46)
The maps are defined on families over a finite field extension of Q(λ2 , λ3 ). The
holomorphic two-form ωX equals the pullback of ωKum(Jac C0 )(ε) , resp. its pullback equals
ωKum(Jac C0′ )(ε′ ) and the pullback of ωKum(E1 ×E2 )(24 ) equals ωKum(Jac C0 )(ε) .
Here, λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 and λ2 = / −1. The parameters µ2 , µ3 are
given by
λ2 λ3 + λ2 − λ3 − 1 + 2d
µ2 = ,
λ2 λ3 + λ2 − λ3 − 1 − 2d
(λ2 λ3 + λ2 − λ3 − 1 + 2d)((1 + λ2 )2 k3 + (1 − λ2 )d − 2(1 + λ3 )λ2 )
(3.47)
(λ2 λ3 + λ2 − λ3 − 1 − 2d)((1 + λ2 )2 k3 − (1 − λ2 )d − 2(1 + λ3 )λ2 )

µ3 = ,
with k32 = λ3 and d2 = (λ2 − λ3 )(λ2 λ3 − 1), and parameters (ρ1 , ρ2 ) satisfy
λ22 λ23 + λ22 + λ23 − 4λ2 λ3 + 1 (λ2 + 1)2 (λ3 + 1)2
ρ22 =
(λ2 − 1)2 (λ3 − 2)2 (λ2 − 1)2 (λ3 − 1)2
(3.48) ρ1 = − , ,
and quadratic-twist factors are given by

4 1
ε′ =
λ2 λ3 (1 − λ2 )2 (1 − λ3 )2 4µ2 µ3 (1 − µ2 µ3 )(µ2 − µ3 )
(3.49) ε= , .
Proof. We relate the twisted Legendre pencil X to the (twisted) double sextic sur-
(ε) ′ (ε′ )
faces W0 and W0 in Equation (2.28) and (3.26), respectively, by relating it to
the Jacobian elliptic K3 surface Y1 and X1∨ in Equations (3.18) and (3.37) first.
(ε)
As a reminder, the minimal resolution of W0 is isomorphic to the Kummer sur-
′ (ε′ )
face Kum(Jac C0 )(ε) . Similarly, the minimal resolution of W0 is isomorphic to the
′ (ε′ )
Kummer surface Kum(Jac C0 ) .
We combine Lemma 3.10, Proposition 3.12, Proposition 3.13, Proposition 3.16,
Corollary 2.36 to obtain the following diagram of rational maps:
W0 ≅ Y1 Ð→ X X1∨ ≅ W0 Z (2
′ (ε′ ) f1 ϕ1 (ε) ψ 4
)
Ð→ Ð→
(3.50)
≃
Kum(Jac C0′ )(ε ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum(E1 × E2 )(2

′ 4
)
Comparing Equations (3.16) and (3.40), one obtains a relation between the Rosenhain
roots (λ2 , λ3 ) and (µ2 , µ3 ). There are eight solutions for (µ2 , µ3 ); see Remark 1.15.
Since we only assumed λ2 =/ −1 the four solutions in Proposition 1.14 – with the sign
for k3 and d chosen independently – determine a smooth genus-two curve C0′ .
By construction, the maps and equivalences in Equation (3.50) to the left of X are
defined over Q(µ2 , µ3 ). The map ψ is defined over the field extension Q(λ2 + λ3 , q)
with q 2 = λ2 λ3 . The remaining maps and equivalences in Equation (3.50) are families
defined over Q(λ2 , λ3 ). The holomorphic two-forms are related by pullback as follows:
(3.51) ωW ′ (ε′ ) = ωY1 , ωY1 = f1∗ ωX , ωX = ϕ∗1 ωX1∨ , ωX1∨ = ωW (ε) , ωW (ε) = ψ ∗ ωZ (24 ) .
0 0 0
Using Equation (2.27) the statement regarding the two-forms follows.

We combine the results of Proposition 3.17 and Proposition 3.18 to obtain one of
the main results of this article:
Theorem 3.19. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1

3 . Further assume
that Λ1 , Λ2 satisfy
(λ2 + λ3 )2 − 4λ2 λ3 2(λ2 + λ3 )
(1 − λ2 )2 (1 − λ3 )2 (1 − λ2 )(1 − λ3 )
(3.52) Λ1 Λ2 = , Λ1 + Λ2 = − .
and the quadratic-twist factor ε is given by

4
λ2 λ3 (1 − λ2 )2 (1 − λ3 )2
(3.53) ε= .
The twisted Legendre pencil, given by

1 + λ2 2 1 + λ3 2
(3.54) X∶ y 2 = z1 z2 (z1 − z2 )(z1 − z3 ) (z3 − ( ) z2 ) (z3 − ( ) z2 ) ,
1 − λ2 1 − λ3
fits into the following Kummer sandwich of rational maps:
Kum(E1 × E2 ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum(E1 × E2 )(2 ) .
4
(3.55)
The maps are defined on families over Q(k2 , k3 ) with kl2 = λl for l = 2, 3. The holo-
morphic two-form ωX equals the pullback of ωKum(Jac C0 )(ε) , resp. its pullback equals
ωKum(E1 ×E2 ) and the pullback of ωKum(E1 ×E2 )(24 ) equals ωKum(Jac C0 )(ε) .
Proof. For the given relations in Equations (3.48) the quantity ρ21 − ρ22 is a perfect
square of a rational function in Q(λ2 , λ3 ). The twisted Legendre pencil X in Equa-
tion (3.2) then is equivalent to Equation (3.54).
First assume that ρ2 = 1 − Λ1 − Λ2 . Comparing Equations (3.15) and (3.40) one
obtains a relation between the Rosenhain roots (λ2 , λ3 ) and the modular parameters
(Λ1 , Λ2 ). This relation is precisely Equation (1.21), relating a genus-two curve C0 with
an elliptic involution and its elliptic-curve quotients El for l = 1, 2 in Proposition 1.8.
The second solution, obtained by changing ρ2 ↦ −ρ2 , is related to the first one by
a modular transformation, e.g., by applying (Λ1 , Λ2 ) ↦ (1 − Λ1 , 1 − Λ2 ). Following
Remark 1.9 we set λl = kl2 for l = 2, 3 and obtain the modular parameters of the elliptic-
curve quotients El in Lemma 1.7 as the algebraic solutions in Equations (3.52). This
also implies that Λ1 , Λ2 ∈ P1 /{0, 1, ∞} so that the elliptic curves E1 and E2 are smooth.
Combining Proposition 3.17 and Proposition 3.18 we obtain the following diagram
of rational maps:
Z ≅ Y3 Ð→ X X1∨ ≅ W0 Z (2
f3 ϕ1 (ε) ψ 4)
Ð→ Ð→
(3.56)
≃
Kum (E1 × E2 ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum (E1 × E2 )(2

4)
By construction, the maps and equivalences in Equation (3.56) to the left of X are
defined over Q(Λ1 , Λ2 ). The map ψ is defined over Q(λ2 + λ3 , q) with q 2 = λ2 λ3 . The
(ε)
map ϕ1 is defined over Q(ρ1 , ρ2 ), and the equivalence X1∨ ≅ W0 over Q(λ2 +λ3 , λ2 λ3 ).
Thus, the smallest common finite extension field is Q(k1 , k2 ). Thus, all maps and
equivalences are families defined over Q(k2 , k3 ), and the holomorphic two-forms are
related by pullback as follows:
(3.57) ω Z = ω Y3 , ωY = f3∗ ωX , ωX = ϕ∗1 ωX1∨ , ωX1∨ = ωW (ε) , ωW (ε) = ψ ∗ ωZ (24 ) .
0 0
The remainder of the statement follows easily.

Remark 3.20. For generic parameters λ2 , λ3 in Theorem 3.19 the transcendental
lattices of the corresponding K3 surfaces are given by
TKum(E1 ×E2 ) = H(2) ⊕ H(2), TX = ⟨2⟩⊕2 ⊕ ⟨−2⟩⊕2 , TKum(Jac C0 ) = H(2) ⊕ ⟨4⟩ ⊕ ⟨−4⟩.
This follows from Remarks 2.3, 2.11, and Equation (3.3).
3.5. One-parameter subfamilies. In this section we will construct five subfami-
lies associated with certain modular correspondences for the two elliptic curves in
Theorem 3.19. The first four families are associated with the specializations of the
six-line configurations of Picard rank 19 found in Proposition 3.1. Since two alge-
braic K3 surfaces S1 and S2 with a rational map of finite degree f ∶ S1 ⇢ S2 have
the same Picard number [35, Cor. 1.2], all K3 surfaces occurring in the corresponding
Kummer sandwiches have Picard rank 19. The last family is obtained by making one
elliptic-curve factor constant with complex multiplication and j-invariant 123 .
3.5.1. Family with isomorphic elliptic curves. We will consider the three specializa-
tions of the families of K3 and abelian surfaces in Theorem 3.19 obtained by requiring
the two elliptic curves to be isomorphic. The two elliptic curves E1 and E2 are iso-
morphic if and only if their j-invariants coincide. One easily checks that this is the
case if and only if the following expression vanishes:
(3.58) (Λ1 − Λ2 )(1 − Λ1 − Λ2 )(Λ1 Λ2 − Λ1 − Λ2 )(1 − Λ1 Λ2 )(Λ1 Λ2 − Λ1 + 1)(Λ1 Λ2 − Λ2 + 1).
However, it follows from Equation (3.52) that
(Λ1 − Λ2 )2 = =/ 0 ,
16λ2 λ3
(1 − λ2 )2 (1 − λ3 )2
whence we must have Λ1 =/ Λ2 if C0 is a smooth genus-two curve. However, the two
elliptic curves in Theorem 3.19 can still be isomorphic. We have the following:
Lemma 3.21. The elliptic curves E1 and E2 in Theorem 3.19 are isomorphic if the
following polynomial vanishes:
0 = (λ2 + 1)(λ3 + 1)(2 − λ2 − λ3 )(λ2 λ3 − 2λ2 + 1)(2λ2 λ3 − λ2 − λ3 )(λ2 λ3 − 2λ3 + 1)
×( (λ2 + λ3 )4 − 2(λ2 λ3 )(λ2 + λ3 )3 + 3(λ2 λ3 )2 (λ2 + λ3 )2 − 2(λ2 λ3 )3 (λ2 + λ3 )
(3.59)
+ (λ2 λ3 )4 − 2(λ2 + λ3 )3 − 6(λ2 λ3 )(λ2 + λ3 )2 + 10(λ2 λ3 )2 (λ2 + λ3 ) − 8(λ2 λ3 )3
+ 3(λ2 λ3 )2 + 10(λ2 λ3 )(λ2 + λ3 ) − 2(λ2 λ3 )2 − 2(λ2 + λ3 ) − 8(λ2 λ3 ) + 1) .
Proof. Equation (3.59) is obtained by setting j1 − j2 = 0. That is, we start with

256(Λ21 − Λ1 + 1)3 256(Λ22 − Λ2 + 1)3
j1 − j2 = j(E1 ) − j(E2 ) =
Λ21 (Λ1 − 1)2 Λ22 (Λ2 − 1)2
(3.60) − ,
use Equations (3.52) to express the right hand side as a rational function in λ2 , λ3 ,
and then select the numerator.
Thus, we consider the one-parameter subfamilies in Theorem 3.19 with a linear
(more precisely, linear or fractional linear) relation between the Rosenhain roots:
(1) λ3 = −1 (or λ2 = −1) such that 1 − Λ1 − Λ2 = 0,
(2) λ3 = 2 − λ2 such that Λ1 Λ2 − Λ1 − Λ2 = 0, or,
λ−1
3 = 2 − λ2 such that 1 − Λ1 Λ2 = 0,
(3) λ−1 −1
3 = 2 − λ2 such that Λ1 Λ2 − Λ1 − Λ2 = 0, or,
λ3 = 2 − λ−1
2 such that 1 − Λ1 Λ2 = 0.
The case (Λ1 Λ2 −Λ1 +1)(Λ1 Λ2 −Λ2 +1) = 0 occurs exactly if the irreducible polynomial
of degree (4, 4) in Equation (3.59) vanishes. We will not consider this case further.
For each remaining case we introduce the following one-parameter families of elliptic
curves, genus-two curves, double sextic and double quadric surfaces:
(1) for λ3 = −1 or λ3 = 2 − λ2 we set
Ẽ1 ∶ y 2 z = (x − z) (x2 − 1 2
t+1 z
) ,
C̃1 ∶ Y2 = XZ(X 2 + 4XZ + 4(t + 1)Z 2 )(X 2 − 4XZ + 4(t + 1)Z 2 ) ,
(3.61) X̃1 ∶ y2 = z1 z2 z3 (z1 − z2 )(z1 − z3 )(tz2 + z3 ) ,
Z̃1 1 ∶ = ε1 ∏ (xk − zk ) (x2k − )
(ε ) 2 1 2
ŷ12 t+1 zk zk ,
k=1,2
(2) for λ3 = 2 − λ2 or λ−1

3 = 2 − λ2 we set
Ẽ2 ∶ y 2 z = (x − z) (x2 − (t+1)(t−3) z 2 ) ,
2
(t−1)
C̃2 ∶ Y2 = XZ(X + (t2 − 1)Z)(X + (t − 1)2 Z)

×(X + (t + 1)(t − 3)Z)(X + (t − 1)(t − 3)Z) ,
(3.62)
X̃2 ∶ y2 = z1 z2 (z1 − z2 )(z1 − z3 )(t2 z2 − z3 )((t − 2)2 z2 − z3 ) ,
Z̃2 2 ∶ = ε2 ∏ (xk − zk ) (x2k − (t+1)(t−3) zk ) zk ,
(t−1)2
(ε ) 2 2
ŷ12
k=1,2
(3) for λ3 = 2 − λ−1

2 or λ−1
3 = 2 − λ−1
2 we set
Ẽ3 ∶ y 2 z = x(x − z) (x + (t+1)

2
(t−1)2
z) ,
C̃3 ∶ Y2 = XZ(X + 4t2 Z)(X + 2(t2 + 1)Z)
×(X + 2t2 (t2 + 1)Z)(X + (t2 + 1)2 Z) ,
(3.63)
X̃3 ∶ = z1 z2 (z1 − z2 )(z1 − z3 ) ( (t z − z3 ) ( (3t z − z3 ) ,
2 2 2 2
+3) +1)
y2 (t2 −1)2 2 (t2 −1)2 2
Z̃3 3 ∶ = ε3 ∏ xk zk (xk − zk ) (xk + (t−1)2 zk ) .

(ε ) 2 (t+1)2
ŷ12
k=1,2
By construction, the minimal resolution of Z̃n is Kum(Ẽn × Ẽn ) for n = 1, 2, 3. We

have the following:
Corollary 3.22. For t ∈ P1 /Sn with
S1 = {0, −1, ∞} , 1
ε1 = t+1 , δ1,1 = t+1
2 , δ2,1 = 1
8 ,
S2 = {±1, 3, ∞} , 1 (t+1)(t−3) 1
ε2 = (t+1)(t−3) , δ1,2 = 2 , δ2,2 = 2(t−1) ,
S3 = {0, ±1, ±i, ∞} , ε3 = 1 , δ1,3 = t−1
t+1 , δ2,3 = 1
t(t2 +1)(t2 −1)2
,
and δ3,n = 4 δ1,n , the twisted Legendre pencil X̃n for n = 1, 2, 3 fits into the following
smooth Kummer sandwich of rational maps:
(3.64) Kum(Ẽn × Ẽn )(−εn δ1,n ) Ð→ X̃n Ð→ Kum(Jac C̃n )(εn δ2,n ) Ð→ Kum(Ẽn × Ẽn )(−εn δ3,n ) .
2 2 2
The maps are defined on families over a finite field extension of Q(t), and the holo-
morphic two-form ωX̃n equals the pullback of ω 2
̃ (εn δ2,n )
, resp. its pullback equals
Kum(Jac Cn )
ω (−εn δ 2 ) and the pullback of ω (−εn δ 2 ) equals ω (εn δ 2 ) .
Kum(Ẽn ×Ẽn ) 1,n Kum(Ẽn ×Ẽn ) 3,n Kum(Jac C̃n ) 2,n
Proof. In the three cases we introduce a parameter t in the following way:

(1+λ2 ) 2
(1) for λ3 = −1 set t = − (1−λ 2 = 4Λ1 (Λ1 − 1),
2)
t+1 Λ21 (t−1)2
(2) for λ3 = 2 − λ2 set λ2 = t−1 and (Λ1 −2)2 = (t+1)(t+3) , or,
t+1 (Λ1 −1)2 (t−1)2
for λ−1
3 = 2 − λ2 set λ2 = t−1 and (Λ1 +1)2 = (t+1)(t+3) ,
1+t2 (t+1)2
(3) for λ3 = 2 − λ−1
2 set λ2 = 2 and Λ1 = − (t−1)2 , or,
1+t2 2(t2 +1)
for λ−1 −1
3 = 2 − λ2 set λ2 = 2 and Λ1 = (t+1)2 .
The field of definition Q(λ2 , λ3 ) is related to Q(t) at worst by a quadratic field
extension. In fact, in the three cases above, we have Q(t) ⊆ Q(λ2 , λ3 ), Q(λ2 , λ3 ) ≅
Q(t), and Q(λ2 , λ3 ) ⊆ Q(t), respectively.
(−ǫn δ2 )
One checks by an explicit computation that the double quadric surface Z̃n 1,n
for n = 1, 2, 3 is isomorphic to Z in Equation (2.3) over Q(Λ1 , Λ2 ) which is (at worst)
a finite field extension of Q(t). Moreover, the pullback by this isomorphism leaves the
holomorphic two-form invariant. Taking the minimal resolution we obtain the desired
isomorphism between Kum(Ẽn × Ẽn )(−εn δ1,n ) and Kum(E1 × E2 ). One also checks that
2
the curve C̃n for n = 1, 2, 3 is isomorphic to C0 for the given (fractional) linear relation
between the Rosenhain roots. We use Equation (2.29) and Proposition 2.14 with a
suitable parameter λ0 , namely,
4 4
(λ2 − 1)2
(3.65) n = 1 ∶ λ0 = , n = 2 ∶ λ0 = − , n = 3 ∶ λ0 = −4(2λ2 − 1) .
λ2 − 1
(ε)
It then follows that the (twisted) double sextic surfaces W0 in Equation (2.28) is
isomorphic to the quadratic twist of the double sextic defined by C̃n with twist fac-
2
tor εn δ2,n . By construction, the isomorphism is defined over a finite field extension
of Q(t). Proposition 2.14 guarantees that the pullback by the isomorphism leaves
the holomorphic two-form invariant. The proof then follows from applying the con-
structed isomorphisms to the situation in Theorem 3.19. The set Sn is obtained from
the vanishing locus of the discriminant for C̃n .
We make the following two remarks:
Remark 3.23. For the surfaces in Corollary 3.22 the transcendental lattices in Re-
mark 3.20 specialize to
TKum(Ẽn ×Ẽn ) = H(2) ⊕ ⟨4⟩, TX̃n = ⟨2⟩⊕2 ⊕ ⟨−2⟩, TKum(Jac C̃n ) = ⟨4⟩⊕2 ⊕ ⟨−4⟩.
This follows from Corollary 3.22 and results in [27, 60].
Remark 3.24. In the two-dimensional parameter space {(A, B) ∣ A =/ B} of Equa-
tion (3.5) we considered the conic (A + B − 4)2 − 4AB = 0 in Proposition 3.1. One
checks that for
1 + λ2 2 1 + λ3 2
(3.66) A=( ) , B=( ) ,
1 − λ2 1 − λ3
the vanishing locus of the conic becomes
(3.67) ∏ (2 − λσ2 2 − λσ3 3 ) = 0 .
σ2 ,σ3 ∈{±1}
Thus, the double sextic surfaces associated with these four components realize Equa-
tions (3.62) and (3.63).
3.5.2. Family with isogenous elliptic curves. As another example, we consider the
classical modular curve X0 (2), i.e., the irreducible plane algebraic curve of genus
zero, given by
0 = − j12 j22 + j13 + j23 + 1488j1 j2 (j1 + j2 ) − 24 34 53 (j12 + j22 )
(3.68)
+ 34 53 4027j1 j2 + 27 37 55 (j1 + j2 ) − 212 39 59 .
This is the relation between the j-invariants j1 = j(τ1 ) = j(E1 ) and j2 = j(τ2 ) = j(E2 )
of two elliptic curves E1 and E2 , respectively, such that τ2 = 2 ⋅ τ1 ; see [24]. We have
the following:
Lemma 3.25. Assume t ∈ P1 /{0, ±1, ∞}. For the two smooth elliptic curves
(3.69) E1 ∶ y12 z1 = x1 (x1 − z1 )(x1 − t2 z1 ) , E2 ∶ y22 z2 = (x2 − (t2 + 1)z2 )(x22 − 4t2 z22 ) ,
with j-invariants j1 = j(τ1 ) = j(E1 ) and j2 = j(τ2 ) = j(E2 ), respectively, one has τ2 =
2 ⋅ τ1 . Moreover, for the two-isogenous elliptic curves – as defined in Equation (1.43)
– one has E1′ ≅ E2 and E2′ ≅ E1 .
Proof. For two elliptic curves in Legendre normal form as in Equation (1.20) one
irreducible component of Equation (3.68) is given by
(3.70) 0 = 16Λ21 Λ22 − 16Λ1 Λ2 (Λ1 + Λ2 ) + 16Λ1 Λ2 − 1 .
X0 (2) has a rational parametrization given by j1 = (h + 256)3 /h2 and j2 = (h + 16)3 /h
for h ∈ C× ; see [24]. We may take a suitable covering by setting h = 16(t2 − 1)2 /t2
to obtain rational modular parameters for elliptic curves in Legendre form; see [42].
Then, the elliptic curves in Legendre form with Λ1 = −1/(t2 −1) and Λ2 = (t+1)2 /(4t),
i.e., the curves
(t + 1)2
y 2 z = x(x − z) (x + y 2 z = x(x − z) (x −
1
(3.71) E1 ∶ 2
z) , E2 ∶ z) ,
t −1 4t
satisfy j1 = j(E1 ) and j2 = j(E2 ). In particular, the given parameters Λ1 , Λ2 satisfy
Equation (3.70). The two-isogenous elliptic curves in Equation (1.43) are then given
by
2(t2 + 1) t2 + 1
E1′ ∶ Y12 Z1 = X1 (X12 + X1 Z1 + Z12 ) , E2′ ∶ Y22 Z2 = X2 (X22 − X2 Z2 + Z22 ) .
(t − 1)
2 t
The remainder of the statement is checked by an explicit computation.
Remark 3.26. The proof of Lemma 3.25 shows that the double sextic surfaces for case
(4) in Proposition 3.1 are covered by Kummer surfaces associated with two elliptic
curves E1 and E2 such that τ2 = 2 ⋅ τ1 . In particular, the conic (A−B +1)(A−B −1) = 0
is equivalent to the relation
(3.72) 0 = 16Λ21 Λ22 − 16Λ1 Λ2 (Λ1 + Λ2 ) + 16Λ1 Λ2 − 1
between the elliptic modular parameters of the elliptic curves. For the two-isogenous
elliptic curves – as defined by Equation (1.43) – we then have j(E1 ) = j(E2′ ) and
j(E2 ) = j(E1′ ). Using Equations (3.52) this is also equivalent to the relation
0 =(λ22 λ23 − 6λ22 λ3 + 2λ2 λ23 + λ22 + λ23 + 4λ2 λ3 + 2λ2 − 6λ3 + 1)
(3.73)
×(λ22 λ23 + 2λ22 λ3 − 6λ2 λ23 + λ22 + λ23 + 4λ2 λ3 − 6λ2 + 2λ3 + 1) ,
between the Rosenhain roots of a genus-two curve C0 with an elliptic involution.
We define the following one-parameter families of elliptic curves, genus-two curves,
double sextic and double quadric surfaces:
Ẽ4 ∶ y2 z = x(x − z)(x − t2 z) ,
Ẽ′ ∶
4 Y 2 Z = (X − (t2 + 1)Z)(X 2 − 4t2 Z 2 ) ,
C̃4 ∶
2(3t4 +8t3 +6t2 −8t+3) 2 2
Y2 = XZ(X 4 − 4X 3 Z + (t2 +1)2
X Z
(3.74)
Z 4) ,
2 2 2 4
+2t−1)
− 4(t
(t2 −2t−1)2
XZ 3 + (t +2t−1)
(t2 −2t−1)4
X̃4 ∶ z1 z2 (z1 − z2 )(z1 − z3 ) (z3 + 4t(t2 −1) z2 ) (z3 + 4t(t2 −1) z2 ) ,
(t2 +1)2 (t2 −2t−1)2
y2 =
Z̃4 4 ε4 x1 z1 (x1 − z1 )(x1 − t2 z1 )(x2 − (t2 + 1)z2 )(x22 − 4t2 z22 )z2 .
(ε ) 2
∶ ŷ12 =
By construction, the minimal resolution of Z̃4 4 is Kum(Ẽ4 × Ẽ4′ )(ε4 ) . Moreover, the
(ε )
j-invariants of Ẽ4 and Ẽ4′ satisfy Equation (3.68). We have the following:
√
Corollary 3.27. For t ∈ P1 /S4 with S4 = {0, ±( 2 ± 1), ±1, ±i, ∞} and
1 1 (t2 − 2t − 1)2 (t2 + 1)
(3.75) ε4 = , δ1,4 = , δ2,4 = , δ3,4 = 2 ,
t(t2 − 1) 2 8t(t2 − 1)(t2 + 2t − 1)
the twisted Legendre pencil X̃4 fits into the following smooth Kummer sandwich of
rational maps:
Kum(Ẽ4 × Ẽ4′ )(ε4 δ1,4 ) Ð→ X̃4 Ð→ Kum(Jac C̃4 )(δ2,4 ) Ð→ Kum(Ẽ4 × Ẽ4′ )(ε4 δ3,4 ) .
2 2 2
(3.76)
morphic two-form ωX̃4 equals the pullback of ω 2
̃ (δ2,4 )
Kum(Jac C4 )
ω (ε δ 2 ) and the pullback of ω (ε δ 2 ) equals ω (δ2 ) .
Kum(Ẽ4 ×Ẽ4′ ) 4 1,4 Kum(Ẽ4 ×Ẽ4′ ) 4 3,4 Kum(Jac C̃4 ) 2,4
Proof. We assume that two elliptic curves E1 and E2 in Theorem 3.19 are given by
Λ1 = −1/(t2 −1) and Λ2 = (t+1)2 /(4t). It follows from Lemma 3.25 that τ1 = 2τ2 , and for
the two-isogenous elliptic curves in Equation (1.43) we have j(E1 ) = j(E2′ ) = j(Ẽ4 ) and
j(E2 ) = j(E1′ ) = j(Ẽ4′ ). By an explicit computation one checks that the isomorphisms
E1 ≅ (Ẽ4 ) t2 −1 , E2 ≅ (Ẽ4′ ) 4t , E1′ ≅ (Ẽ4′ ) t , E2′ ≅ (Ẽ4 ) t2 −1 ,
1 1 1 1
( ) ( ) ( ) ( )
(3.77)
are defined over Q(t) and identify the corresponding holomorphic one-forms via pull-
back. Furthermore, we use Equation (2.29) and Proposition 2.14 with the parameter
(T 4 + 1)(T 4 + 2T 2 − 1)2
(T 4 − 2T 2 − 1)(T 8 + 4DT 5 − 6T 6 − 4DT 3 + 2T 2 − 1)
(3.78) λ0 = ,
where t = T 2 and D 2 = 1 − T 4 . The remainder of the proof is analogous to the proof

of Corollary 3.22.
3.5.3. Family with trivial factor. The last family is not obtained by a fixed modular
correspondence between two elliptic curves. Instead we make one elliptic-curve factor
constant with complex multiplication and j-invariant 123 . We introduce the following
one-parameter families of elliptic curves, genus-two curves, double sextic and double
quadric surfaces:
Ẽ5 ∶ y2z = (x − (t2 + 1)z) (x2 − 4t2 z 2 ) ,
̃5 ∶
F Y 2 Z = X(X 2 − Z 2 ) ,
C̃5 ∶ Y2 = XZ(X 2 − 2(t2 − 4t + 1)XZ + (t2 + 1)2 Z 2 )
×(X 2 − 2(t2 + 4t + 1)XZ + (t2 + 1)2 Z 2 ) ,
(3.79)
X̃5 ∶ = z1 z2 (z1 − z2 )(z1 − z3 )(z32 − z2 z3 + )

(t2 +1)2 2
y2 16t2 z2 ,
Z̃5 5 ∶ ε5 (x1 − (t2 + 1)z1 ) (x21 − 4t2 z12 ) z1 x2 z2 (x22 − z22 ) .

(ε ) 2
ŷ12 =
By construction, the minimal resolution of Z̃5 5 is Kum(Ẽ5 × F̃5 )(ε5 ) . Moreover, one
(ε )
checks that F̃5 is constant with complex multiplication and j(F̃5 ) = 123 . We have the
following:
Corollary 3.28. For t ∈ P1 /S5 with S5 = {0, ±1, ±i, ∞} and
1 1 1
(3.80) ε5 = , δ1,5 = , δ2,5 = , δ3,5 = 2 ,
4t 2 8t(t2 + 1)
the twisted Legendre pencil X̃5 fits into the following smooth Kummer sandwich of
rational maps:
Kum(Ẽ5 × F
̃5 )(ε5 δ1,5 Ð→ X̃5 Ð→ Kum(Jac C̃5 )(δ2,5 ) Ð→ Kum(Ẽ5 × F
̃5 )(ε4 δ3,5
2 2 2
) )
(3.81) .
morphic two-form ωX̃5 equals the pullback of ω 2
̃ (δ2,5 )
Kum(Jac C5 )
ω (ε5 δ 2 ) and the pullback of ω 2
̃5 )(ε4 δ3,5 )
equals ω (δ 2 ) .
Kum(Ẽ5 ×F
̃5 ) 1,5 Kum(Ẽ5 ×F Kum(Jac C̃5 ) 2,5
Proof. We assume that the two elliptic curves E1 and E2 in Theorem 3.19 are given
by Λ1 = (t + 1)2 /(4t) and Λ2 = 1/2. Furthermore, we use Equation (2.29) and Propo-
sition 2.14 with the parameter
(3.82) λ0 = (u + iω)2 (u − ω)2 ,
where t = u2 and ω = exp (−πi/4). The remainder of the proof is analogous to the
proof of Corollary 3.22.
3.6. Construction of correspondences. In Theorem 3.19 we constructed a Kum-
mer sandwich for the Legendre pencil in Equation 3.54. The surfaces involved are
defined over Q(λ2 , λ3 ), the maps are defined on families over the finite field extension
Q(k2 , k3 ) with kl2 = λl for l = 2, 3.
More generally, let us consider two families of Jacobian elliptic K3 surfaces Vλ and
Wλ defined over Q(λ2 , λ3 ) whose generic members have Picard rank 18. Here, we
denote the tuple (λ2 , λ3 ) by λ. Since we assume that the K3 surfaces are Jacobian
elliptic K3 surfaces, they have Weierstrass models over Q(λ2 , λ3 ). For example, for
Vλ we assume this Weierstrass model to be of the form
(3.83) Vλ ∶ y 2 z = 4x3 − g2 (v) xz 2 − g3 (v)z 3 ,
where v is a suitable coordinate on the base curve P1 , and g2 and g3 are polynomials
of degree 8 and 12, respectively, with coefficients in Q(λ2 , λ3 ). There is a natural
(ε)
notion of twisted K3 surfaces Vλ for ε ∈ Q(λ2 , λ3 ) by taking the minimal resolution
of the twisted Weierstrass model
ŷ 2 z = ε (4x3 − g2 (v) xz 2 − g3 (v)z 3 ) .
(ε)
(3.84) Vλ ∶
Let F ∶ Vλ ⇢ Wλ be a rational map of finite degree. An algebraic correspondence
from Vλ to Wλ is a subset Γ ⊂ Vλ × Wλ such that Γ is finite and surjective over each
component of Vλ . Then, the graph of F ∶ Vλ ⇢ Wλ is clearly such a correspondence
Γ = ΓF . We have the following:
Lemma 3.29. Assume ε = δ 2 =/ 0 with δ ∈ Q(λ2 , λ3 ). Then, there is an alge-
(ε)
braic correspondence from Vλ to Vλ over Q(λ2 , λ3 ) that induces a nonzero map
(ε)
H 2,0 (Vλ ) → H 2,0 (Vλ ).
Proof. Since ε = δ 2 the map from Equation (3.83) to Equation (3.84) is given by
(ε)
F ∶ y ↦ ŷ = δy. Generators of H 2,0 (Vλ ) and H 2,0 (Vλ ) are given in the charts z = 1
by the regular two-forms ω = dv ∧ dx/y and ω ′ = dv ∧ dx/ŷ, respectively. We have

F ∗ ω ′ = ω/δ =/ 0.
For a map F ∶ Vλ ⇢ Wλ which is defined over Q(k2 , k3 ), the graph ΓF determines
a correspondence, but only over Q(k2 , k3 ). However, in our situation F has an addi-
tional property: for generators of H 2,0 (Vλ ) and H 2,0 (Wλ ) given in local coordinates by
the regular two-forms ωVλ and ωWλ , respectively, we have F ∗ ωWλ = ωVλ . We consider
the Gal (Q(k2 , k3 )/Q(λ2 , λ3 ))-conjugate maps F(±,±) ∶ Vλ ⇢ Wλ obtained by replacing
(k1 , k2 ) ↦ (±k1 , ±k2 ) such that F(+,+) = F . Since the surfaces and the two-forms are
defined over Q(λ2 , λ3 ), we obtain
(3.85) ∑
∗
F(σ ω = 4 ωVλ =/ 0 .
2 ,σ3 ) Wλ
σ2 ,σ3 ∈{±}
The sum of the graphs

(3.86) Γ = ΓF(+,+) + ΓF(+,−) + ΓF(−,+) + ΓF(−,−)
then defines a correspondence on Vλ × Wλ over Q(λ2 , λ3 ) that induces a nonzero map
H 2,0 (Wλ ) → H 2,0 (Vλ ). Thus, it must induce an isomorphism on the transcendental
lattices of Wλ and Vλ when tensored with Q. The Artin comparison theorem for
complex and ℓ-adic cohomology implies that the same is true for the corresponding
Galois representations [3]. Therefore, the correspondence Γ induces an isomorphism
of GQ(λ2 ,λ3 ) -representations, denoted by [Γ], between the part of Hét2
(Vλ,Q , Qℓ ) that
is orthogonal to the eighteen algebraically independent étale cycle classes and the
corresponding part of Hét 2
(Wλ,Q , Qℓ ). This isomorphism produces, via the Künneth
formula and Poincaré duality, a Galois invariant class in Hét 4
(Vλ,Q × Wλ,Q , Qℓ ). Of
course, the existence of such a class is already expected due to the Tate conjecture.
In the generic situation of Theorem 3.19 the K3 surfaces have Picard rank 18.
For the continuous Galois representations of GQ = Gal (Q/Q) on the transcendental
lattices, there are isomorphisms of GQ -representations, namely
(XQ , Qℓ ) ≅ Tℓ ⊕ Qℓ (−1)⊕18 ,
2 (1)
Hét
( Kum(E1 × E2 )Q , Qℓ ) ≅ Tℓ ⊕ Qℓ (−1)⊕18 ,
2 (2)
(3.87) Hét
( Kum(Jac C0 ) , Qℓ ) ⊕ Qℓ (−1)⊕18 ,
2 (ε) (3)
Hét ≅ Tℓ
Q
(m)
where Tℓ are the four-dimensional ℓ-adic representations obtained from the corre-
sponding transcendental lattices using the comparison theorem for m = 1, 2, 3. Here,
we used the fact that NSQ is generated by curves defined over Q. This explains the
summand Qℓ (−1)⊕18 in Equation (3.87). We have the following:
Theorem 3.30. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1
3 and are generic
otherwise. Further assume that Λ1 , Λ2 satisfy
(λ2 + λ3 )2 − 4λ2 λ3 2(λ2 + λ3 )
(1 − λ2 )2 (1 − λ3 )2 (1 − λ2 )(1 − λ3 )
(3.88) Λ1 Λ2 = , Λ1 + Λ2 = − ,
and ε = λ2 λ3 . Then, there are explicit algebraic correspondences

Γ(1,2) ⊂ Kum(E1 × E2 ) × X ,
(3.89) Γ(2,3) ⊂ X × Kum(Jac C0 )(ε) ,
Γ(3,1) ⊂ Kum(Jac C0 )(ε) × Kum(E1 × E2 ) ,
defined over Q(λ2 , λ3 ), which induce isomorphisms of the ℓ-adic representations
[Γ(i,j) ] ∶
(i) ≅ (j)
(3.90) Tℓ Ð→ Tℓ ,
for i, j ∈ {1, 2, 3} with i =/ j.
Proof. It follows from Remark 3.23 that for generic parameters λ2 , λ3 ∈ Q, the K3
surfaces in Theorem 3.19 have Picard rank 18. Using the families of maps in Equa-
tion (3.55) defined over Q(k2 , k3 ), we construct correspondences between the different
Kummer surfaces as explained above. Lemma 3.29 allows us to replace the twist fac-
tor ε = λ2 λ3 (1−λ24)2 (1−λ3 )2 by ε′ = λ2 λ3 as ε′ /ε = (λ2 λ3 (1 − λ2 )(1 − λ3 )/2)2 . Similarly for
Γ(3,1) we eliminate the twist factor 24 .
In the situation of Corollary 3.22 the K3 surfaces have Picard rank 19. Then, there
are isomorphisms of GQ -representations, namely
2 ̃
Hét ̃ (1) ⊕ Qℓ (−1)⊕19 ,
(Xn,Q , Qℓ ) ≅ T n,ℓ
̃ (2) ⊕ Qℓ (−1)⊕19 ,
( Kum(Ẽn × Ẽn ) n , Qℓ ) ≅ T
2 (−ε )
Hét n,ℓ
Q
̃ (3) ⊕ Qℓ (−1)⊕19 ,
( Kum(Jac C̃n ) n , Qℓ ) ≅ T
2 (ε )
Hét n,ℓ
Q
where ̃ (m)
T are the three-dimensional ℓ-adic representations obtained from the cor-
n,ℓ
responding transcendental lattices using the comparison theorem for m, n = 1, 2, 3.
In the cases above, the representation T ̃ (2) are naturally isomorphic to a symmet-
n,ℓ
2 1 ̃
ric square, namely Sym (Hét (En,Q , Qℓ ))(χ) where χ is the Dirichlet character of the
quadratic twist given by εn ; see [60]. We have the following:
Theorem 3.31. In Equations (3.61), (3.62), (3.63) let t ∈ P1 /Sn with
S1 = {0, −1, ∞} , S2 = {±1, 3, ∞} S3 = {0, ±1, ±i, ∞} ,
1 1
and ε1 = t+1 , ε2 = (t+1)(t−3) , ε3 = 1. Then, there are explicit algebraic correspondences
̃(1,2) ⊂ Kum(Ẽn × Ẽn )(−εn ) × X̃n ,
Γ n
(3.91) ̃(2,3)
Γ n ⊂ X̃n × Kum(Jac C̃n )(εn ) ,
̃(3,1)
Γ n ⊂ Kum(Jac C̃n )(εn ) × Kum(Ẽn × Ẽn )(−εn ) ,
defined over Q(t), which induce isomorphisms of the ℓ-adic representations
(3.92) ̃(i,j) ] ∶
[Γ ̃ (i) Ð→
T
≅
̃ (j) ,
T
n n,ℓ n,ℓ
for i, j, n ∈ {1, 2, 3} with i =/ j.

Proof. It follows from Remark 3.23 that the K3 surfaces in Corollary 3.22 have Picard
rank 19. Using the families of maps in Equation (3.64) defined over a finite field ex-
tension of Q(t), we construct correspondences between the different Kummer surfaces
as explained above. Lemma 3.29 then allows us to replace the quadratic-twist factors
2
±εn δm,n by ±εn .
Remark 3.32. For n = 1 the statement of Theorem 3.31 is precisely the main result
of [60]. In fact, Equations (3.61) recover the families that appeared in [60]4. To our
knowledge, the other two cases have not appeared in the literature.
In the situation of Corollary 3.27 and 3.28 the K3 surfaces have Picard rank 19 and
18, respectively, and there are isomorphisms of GQ -representations, namely
2 ̃ ̃ (1) ⊕ Qℓ (−1)⊕µn ,
Hét (Xn,Q , Qℓ ) ≅ T n,ℓ
2
Hét ( Kum(Jac C̃n )Q , Qℓ ) ≅ T
̃ (3) ⊕ Qℓ (−1)⊕µn ,
n,ℓ
for n = 4, 5 and µ4 = 19, µ5 = 18. Moreover, we have the isomorphisms of GQ -

representations
( Kum(Ẽn × F
̃n ) , Qℓ ) ̃ (2) ⊕ Qℓ (−1)⊕µn .
2 (εn )
Hét ≅ T n,ℓ
Q
Here, F̃4 = Ẽ4′ is the two-isogenous elliptic curve such that the j-invariants of Ẽ4 and
Ẽ4′ satisfy Equation (3.68), i.e., the equation of the classical modular curve X0 (2).
And F̃5 is the constant elliptic curve with complex multiplication and j(F̃5 ) = 123 .
Theorem 3.33. In Equations (3.74), (3.79) let t ∈ P1 /Sn with
√
S4 = {0, ±( 2 ± 1), ±1, ±i, ∞} , S5 = {0, ±1, ±i, ∞} ,
1 1
and ε4 = t(t2 −1) , ε5 = 4t . Then, there are explicit algebraic correspondences
̃
Γ(1,2)
n ⊂ Kum(Ẽn × F
̃n )(εn ) × X̃n ,
(3.93) ̃
Γ(2,3)
n ⊂ X̃n × Kum(Jac C̃n ) ,
̃
Γ(3,1) ⊂ Kum(Jac C̃n ) × Kum(Ẽn × F
̃n )(εn ) ,
n
defined over Q(t), which induce isomorphisms of the ℓ-adic representations

(3.94) ̃(i,j) ] ∶
[Γ ̃ (i) Ð→
T
≅
̃ (j) ,
T
n n,ℓ n,ℓ
for i, j ∈ {1, 2, 3} with i =/ j and n = 4, 5.

Proof. The proof is analogous to the proof of Theorem 3.31.
4All surfaces in Theorem 3.31 have been twisted by (−1) compared to [60]. This is due to a
different sign choice in the definition of the twisted Legendre pencil in Equation (3.2).
References
[1] Scott Ahlgren, Ken Ono, and David Penniston, Zeta functions of an infinite family of K3
surfaces, Amer. J. Math. 124 (2002), no. 2, 353–368. MR1890996 ↑1
[2] Anatolij N. Andrianov, Quadratic forms and Hecke operators, Grundlehren der Mathematischen
Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 286, Springer-Verlag,
Berlin, 1987. MR884891 ↑6
[3] M. Artin, The etale topology of schemes, Proc. Internat. Congr. Math. (Moscow, 1966), 1968,
pp. 44–56. MR0232775 ↑61
[4] H. F. Baker, On a system of differential equations leading to periodic functions, Acta Math. 27
(1903), no. 1, 135–156. MR1554977 ↑25
[5] Wolf Barth, Even sets of eight rational curves on a K3-surface, Complex geometry (Göttingen,
2000), 2002, pp. 1–25. MR1922094 ↑20, 24
[6] Christina Birkenhake and Herbert Lange, Complex abelian varieties, Second, Grundlehren der
Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 302,
Springer-Verlag, Berlin, 2004. MR2062673 ↑19, 31, 40
[7] Christina Birkenhake and Hannes Wilhelm, Humbert surfaces and the Kummer plane, Trans.
Amer. Math. Soc. 355 (2003), no. 5, 1819–1841. MR1953527 ↑7, 25
[8] Oskar Bolza, On binary sextics with linear transformations into themselves, Amer. J. Math. 10
(1887), no. 1, 47–70. MR1505464 ↑9
[9] Jean-Benoı̂t Bost and Jean-François Mestre, Moyenne arithmético-géométrique et périodes des
courbes de genre 1 et 2, Gaz. Math. 38 (1988), 36–64. MR970659 ↑7
[10] Noah Braeger, Andreas Malmendier, and Yih Sung, Kummer sandwiches and Greene-Plesser
construction, J. Geom. Phys. 154 (2020), 103718, 18. MR4099481 ↑50
[11] J. W. S. Cassels and E. V. Flynn, Prolegomena to a middlebrow arithmetic of curves of genus
2, London Mathematical Society Lecture Note Series, vol. 230, Cambridge University Press,
Cambridge, 1996. MR1406090 ↑26
[12] Adrian Clingher and Charles F. Doran, Modular invariants for lattice polarized K3 surfaces,
Michigan Math. J. 55 (2007), no. 2, 355–393. MR2369941 ↑19
[13] Adrian Clingher and Andreas Malmendier, On the geometry of (1,2)-polarized Kummer surfaces,
arXiv:1704.04884 [math.AG] (201704), available at 1704.04884. ↑41
[14] , Nikulin involutions and the CHL string, Comm. Math. Phys. 370 (2019), no. 3, 959–
994. MR3995925 ↑48
[15] , Normal forms for Kummer surfaces, London Mathematical Society Lecture Note Series
2 (2019), no. 459, 107–162. ↑26, 31, 32
[16] Adrian Clingher, Andreas Malmendier, and Tony Shaska, Geometry of Prym varieties for special
bielliptic curves of genus three and five (2019, arXiv:1901.09846). ↑17
[17] , On isogenies among certain abelian surfaces, Michigan Math. J. (2019,
arXiv:1901.09846). ↑17
[18] Pierre Deligne, La conjecture de Weil pour les surfaces K3, Invent. Math. 15 (1972), 206–226.
MR0296076 ↑19
[19] Igor Dolgachev and David Ortland, Point sets in projective spaces and theta functions,
Astérisque 165 (1988), 210 pp. (1989). MR1007155 ↑32
[20] E. Freitag, Siegelsche Modulfunktionen, Grundlehren der Mathematischen Wissenschaften
[Fundamental Principles of Mathematical Sciences], vol. 254, Springer-Verlag, Berlin, 1983.
MR871067 ↑8
[21] Alice Garbagnati, Elliptic K3 surfaces with abelian and dihedral groups of symplectic automor-
phisms, Comm. Algebra 41 (2013), no. 2, 583–616. MR3011784 ↑44
[22] Maria R. Gonzalez-Dorrego, (16, 6) configurations and geometry of Kummer surfaces in P3 ,

Mem. Amer. Math. Soc. 107 (1994), no. 512, vi+101. MR1182682 ↑19
[23] V. A. Gritsenko and V. V. Nikulin, Igusa modular forms and “the simplest” Lorentzian Kac-
Moody algebras, Mat. Sb. 187 (1996), no. 11, 27–66. MR1438983 ↑8
[24] E. Hecke, Die eindeutige Bestimmung der Modulfunktionen q-ter Stufe durch algebraische Eigen-
schaften, Math. Ann. 111 (1935), no. 1, 293–301. MR1512996 ↑57
[25] William L. Hoyt, On surfaces associated with an indefinite ternary lattice, Number theory (New
York, 1983–84), 1985, pp. 197–210. MR803354 ↑41
[26] , Notes on elliptic K3 surfaces, Number theory (New York, 1984–1985), 1987, pp. 196–
213. MR894512 ↑41
[27] , Elliptic fiberings of Kummer surfaces, Number theory (New York, 1985/1988), 1989,
pp. 89–110. MR1023921 ↑41, 57
[28] , On twisted Legendre equations and Kummer surfaces, Theta functions—Bowdoin 1987,
Part 1 (Brunswick, ME, 1987), 1989, pp. 695–707. MR1013162 ↑41, 43
[29] William L. Hoyt and Charles F. Schwartz, Yoshida surfaces with Picard number ρ ≥ 17, Pro-
ceedings on Moonshine and related topics (Montréal, QC, 1999), 2001, pp. 71–78. MR1877757
↑41, 43
[30] R. W. H. T. Hudson, Kummer’s quartic surface, Cambridge Mathematical Library, Cambridge
University Press, Cambridge, 1990. With a foreword by W. Barth, Revised reprint of the 1905
original. MR1097176 ↑19
[31] Georges Humbert, Sur les fonctionnes abéliennes singulières. I, II, III, J. Math. Pures Appl.
serie 5, V, 233–350 (1899); VI, 279–386 (1900); VII, 97–123 (1901). ↑24
[32] Jun-ichi Igusa, On Siegel modular forms of genus two, Amer. J. Math. 84 (1962), 175–200.
MR0141643 ↑8
[33] , Modular forms and projective invariants, Amer. J. Math. 89 (1967), 817–855.
MR0229643 ↑8
[34] , On the ring of modular forms of degree two over Z, Amer. J. Math. 101 (1979), no. 1,
149–183. MR527830 ↑8
[35] Hiroshi Inose, On certain Kummer surfaces which can be realized as non-singular quartic sur-
faces in P 3 , J. Fac. Sci. Univ. Tokyo Sect. IA Math. 23 (1976), no. 3, 545–560. MR429915
↑54
[36] K. Kodaira, On the structure of compact complex analytic surfaces. I, II, Proc. Nat. Acad. Sci.
U.S.A. 50 (1963), 218–221; ibid. 51 (1963), 1100–1104. MR165541 ↑19
[37] Abhinav Kumar, Elliptic fibrations on a generic Jacobian Kummer surface, J. Algebraic Geom.
23 (2014), no. 4, 599–667. MR3263663 ↑20, 28, 29
[38] Masato Kuwata and Tetsuji Shioda, Elliptic parameters and defining equations for elliptic fibra-
tions on a Kummer surface, Algebraic geometry in East Asia—Hanoi 2005, 2008, pp. 177–215.
MR2409557 ↑20, 21, 22, 23
[39] A. Malmendier and T. Shaska, The Satake sextic in F-theory, J. Geom. Phys. 120 (2017), 290–
305. MR3712162 ↑8, 9, 12, 16, 19
[40] Andreas Malmendier, Kummer surfaces associated with Seiberg-Witten curves, J. Geom. Phys.
62 (2012), no. 1, 107–123. MR2854198 ↑41
[41] Andreas Malmendier and David R. Morrison, K3 surfaces, modular forms, and non-geometric
heterotic compactifications, Lett. Math. Phys. 105 (2015), no. 8, 1085–1118. MR3366121 ↑19
[42] Andreas Malmendier and Ken Ono, Moonshine for M24 and Donaldson invariants of CP2 ,
Commun. Number Theory Phys. 6 (2012), no. 4, 759–770. MR3068406 ↑58
[43] Andreas Malmendier and Tony Shaska, A universal genus-two curve from Siegel modular
forms, SIGMA Symmetry Integrability Geom. Methods Appl. 13 (2017), Paper No. 089, 17.
MR3731039 ↑9, 12
[44] Andreas Malmendier and Yih Sung, Counting rational points on Kummer surfaces, Res. Number
Theory 5 (2019), no. 3, Paper No. 27, 23. MR3992148 ↑50
[45] John Hector McDonald, A problem in the reduction of hyperelliptic integrals, Trans. Amer.
Math. Soc. 7 (1906), no. 4, 578–587. MR1500768 ↑13
[46] Afsaneh Mehran, Double covers of Kummer surfaces, Manuscripta Math. 123 (2007), no. 2,
205–235. MR2306633 ↑44
[47] Jean-François Mestre, Construction de courbes de genre 2 à partir de leurs modules, Effective
methods in algebraic geometry (Castiglioncello, 1990), 1991, pp. 313–334. MR1106431 ↑8
[48] D. R. Morrison, On K3 surfaces with large Picard number, Invent. Math. 75 (1984), no. 1,
105–121. MR728142 ↑21, 25, 44
[49] David Mumford, Abelian varieties, Tata Institute of Fundamental Research Studies in Mathe-
matics, vol. 5, Published for the Tata Institute of Fundamental Research, Bombay; by Hindus-
tan Book Agency, New Delhi, 2008. With appendices by C. P. Ramanujam and Yuri Manin,
Corrected reprint of the second (1974) edition. MR2514037 ↑6
[50] V. V. Nikulin, Kummer surfaces, Izv. Akad. Nauk SSSR Ser. Mat. 39 (1975), no. 2, 278–293,
471. MR0429917 ↑43
[51] Keiji Oguiso, On Jacobian fibrations on the Kummer surfaces of the product of nonisogenous
elliptic curves, J. Math. Soc. Japan 41 (1989), no. 4, 651–680. MR1013073 ↑20
[52] Alfred Pringsheim, Zur Transformation zweiten Grades der hyperelliptischen Functionen erster
Ordnung, Math. Ann. 9 (1876), no. 4, 445–475. MR1509868 ↑10
[53] Fried. Jul. Richelot, De transformatione integralium Abelianorum primi ordinis commentatio.
Caput secundum. De computatione integralium Abelianorum primi ordinis, J. Reine Angew.
Math. 16 (1837), 285–341. MR1578135 ↑7
[54] Jean-Pierre Serre, Rational points on curves over finite fields, notes by F. Gouvea of lectures
at Harvard University (1985). ↑14
[55] E. R. Shafarevitch, Le théorème de Torelli pour les surfaces algébriques de type K3, Actes du
Congrès International des Mathématiciens (Nice, 1970), Tome 1, 1971, pp. 413–417. MR0419459
↑19
[56] Tetsuji Shioda, Kummer sandwich theorem of certain elliptic K3 surfaces, Proc. Japan Acad.
Ser. A Math. Sci. 82 (2006), no. 8, 137–140. MR2279280 ↑50
[57] , Classical Kummer surfaces and Mordell-Weil lattices, Algebraic geometry, 2007,
pp. 213–221. MR2296439 ↑24
[58] Burt Totaro, Recent progress on the Tate conjecture, Bull. Amer. Math. Soc. (N.S.) 54 (2017),
no. 4, 575–590. MR3683625 ↑2
[59] G. van der Geer, On the geometry of a Siegel modular threefold, Math. Ann. 260 (1982), no. 3,
317–350. MR669299 ↑32
[60] Bert van Geemen and Jaap Top, An isogeny of K3 surfaces, Bull. London Math. Soc. 38 (2006),
no. 2, 209–223. MR2214473 ↑1, 4, 36, 37, 41, 46, 57, 62, 63
Dept. of Mathematics & Statistics, Utah State University, Logan, UT 84322

Email address: noah.braeger@usu.edu
Dept. of Mathematics & Statistics, University of Missouri - St. Louis, MO 63121

Email address: clinghera@umsl.edu
Dept. of Mathematics, University of Connecticut, Storrs, Connecticut 06269,

Email address: andreas.malmendier@uconn.edu

Email address: shantel.black@aggiemail.usu.edu

2109 03189

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2109 03189

Uploaded by

Copyright:

Available Formats

ISOGENIES OF CERTAIN K3 SURFACES OF RANK 18

NOAH BRAEGER, ADRIAN CLINGHER, ANDREAS MALMENDIER,

AND SHANTEL SPATIG

Abstract. We construct geometric isogenies between three types of two-parameter

One choice of abelian surface we shall consider is the product A = E1 × E2 , where

defined over Q, which induce ℓ-adic GQ -representation isomorphisms:

for i, j ∈ {1, 2, 3} and i =/ j.

1. Abelian surfaces admitting an elliptic involution

of which is H1 . We shall denote the second component by H4 . This is the image of

Figure 1. (Z/2Z)2 -covering for the genus-two curve C0

× (λ1 λ2 + λ1 λ3 − λ2 λ3 − λ1 ) (λ1 λ2 + λ2 λ3 − λ1 λ3 − λ2 ) (λ1 λ3 + λ2 λ3 − λ1 λ2 − λ3 )

Proposition 1.5 (Pringsheim). There are exactly 15 components of H4 (2) in A2 (2),

Table 1. Pringsheim components of H4

for XZ =/ 0, and [xl ∶ yl ∶ zl ] = [1 ∶ 0 ∶ 0] otherwise. Here, q, r are square roots of

Roch theorem. Thus, the preimage of the vanishing locus of a section of Ll → El

and [I2′ ∶ I4′ ∶ I6′ ∶ I10

one obtains the curve

with λ3 = k32 . Introducing new moduli µ2 , µ3 , given by

model of a smooth K3 surface can always be written in the form

outer tensor product of the elliptic-curve holomorphic one-forms, i.e.,

We have the immediate:

x1 x2 (x1 + x2 − 1)(x1 − Λ1 )(x1 x2 − Λ2 (x1 + x2 ) + Λ2 )

Figure 2. Double cover branched along reducible sextic

A variable transformation in Equation (2.15), given by

Table 2. Nodes on the generic Jacobian Kummer surface KA

and its associated (twisted) Shioda sextic given by

such that ωKum(Jac C)(ε) = ı∗ ωKum(Jac C)

The discriminant function of the elliptic fibration is given by

We have the immediate:

additional relations holds:

Table 3. Components of H4 with matching constraints for the GH-

Table 4. Solutions for A2 = B 2 = C 2 = 1 and D = 0

and I2 be the complete intersection in P3 = P(Y00 , Y01 , Y10 , Y11 ), given by

We have the following:

The minimal resolution is isomorphic to a Kummer surface of Picard rank 18. In

scale factor. Inverting Equations (2.56) yields

Z1,1 = −λ2 λ3 (1 − λ2 )(1 − λ3 )W + (1 − λ2 )(1 − λ3 )Y .

the maps φ± ∶ Z ⇢ V0 are rational maps of degree two given by

with Z00 = (x21 − 2Λ1 x1 z1 + Λ1 z12 )(x22 − 2Λ2 x2 z2 + Λ2 z22 ) ,

(2.66) yl ↦ (x2l − 2Λl xl zl + Λl zl2 )(x2l − 2xl zl + Λl zl2 )(x2l − Λl zl2 ) ,

Proof. The composition of elliptic-curve isogenies in Equation (2.68) is given by

the maps ψ± ∶ V0 ⇢ Z in Proposition 2.38 are equivalently given by

factors though V0 by construction. Proposition 2.33 proves that V0 is isomorphic to

Figure 3. Line configuration for general member in Proposition 3.1

obtain a non-torsion section if A2 + B 2 − 2AB − 8A − 8B − 16 = 0. Thus, we set A = α2

surface of Y by the involution yields X . A Nikulin involution is an analytic involution

µ µ (µ2 − µ3 )2 (1 − µ2 µ3 )2 (µ2 − 1)6 (µ3 − 1)6 (u1 − µ2 µ3 u2 )3 (u1 + µ2 µ3 u2 )3 Y ,

whereas the remaining fiber of Kodaira-type I0∗ is located over

A compatible Nikulin involution on Y1 can be constructed in the same way as already

Moreover, it follows ωX = ϕ∗ ωX ∨ and 2 ωX ∨ = (ϕ∨ )∗ ωX for the holomorphic two-forms

In terms of Lemma 3.4 the change of parameters is equivalent to the interchange of

and a quadric-twist factor is given by

Proof. Since the genus-two curve C0 is smooth, we have λ2 , λ3 ∈ P1 /{0, 1, ∞} and

x = (λ2 λ3 )3 (λ2 λ3 − λ2 − λ3 + 1)2 X , y = (λ2 λ2 ) 2 (λ2 λ3 − λ2 − λ3 + 1)3 Y , z = Z .

Proposition 3.17. For Λ1 , Λ2 ∈ P1 /{0, 1, ∞} let parameters ρ1 , ρ2 satisfy

(λ2 λ3 + λ2 − λ3 − 1 − 2d)((1 + λ2 )2 k3 − (1 − λ2 )d − 2(1 + λ3 )λ2 )

and quadratic-twist factors are given by

Kum(Jac C0′ )(ε ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum(E1 × E2 )(2

Using Equation (2.27) the statement regarding the two-forms follows.

Theorem 3.19. Assume that λ2 , λ3 ∈ P1 /{0, 1, ∞} satisfy λ2 =/ λ±1

and the quadratic-twist factor ε is given by

The twisted Legendre pencil, given by

Kum (E1 × E2 ) Ð→ X Ð→ Kum(Jac C0 )(ε) Ð→ Kum (E1 × E2 )(2