A LAGRANGIAN SPHERE WHICH IS NOT A VANISHING

CYCLE

FRANÇOIS GREER
arXiv:1812.00489v3 [math.SG] 13 Jun 2019

Abstract. We give examples of Calabi-Yau threefolds containing Lagrangian

spheres which are not vanishing cycles of nodal degenerations, answering a
question of Donaldson in the negative.

1. Introduction
The n-dimensional nodal singularity has a 1-parameter versal deformation
n
!
X
zi = t ⊂ Cn+1 × Ct .
2

i=0
The nearby fiber over t =  > 0 retracts onto an n-sphere, the vanishing cycle:
n
!
X
Sn ' x2i =  ,
i=0
P
which is Lagrangian
√ with respect to the standard symplectic form ω = dxi ∧ dyi
(zi = xi + yi −1). A natural question, first raised by Donaldson [4], is whether all
Lagrangian spheres arise in this way.
Question 1. Let Z be a complex projective manifold, and L ⊂ Z embedded sphere,
Lagrangian with respect to a Kähler form on Z. Is L always the vanishing cycle of
a nodal degeneration of Z?
The answer is positive for curves, and unknown for surfaces. For certain K3 surfaces,
the answer is positive [8], modulo the issue of whether Fukaya isomorphism implies
Hamiltonian isotopy. For Horikawa surfaces, a positive answer would distinguish
two particular deformation types as smooth manifolds [1]. We show that the answer
to Question 1 is negative in general:
Theorem 2. There exists a rigid projective Calabi-Yau threefold X̂ with a La-
grangian sphere L ⊂ X̂ which is homologically non-trivial (essential).
Rigidity implies that any degeneration of X̂ is isotrivial. We prove further that
such a degeneration must have monodromy of order ≤ 6 on H3 (X̂). In particu-
lar, this rules out nodal degenerations with vanishing cycle L; their monodromy
would be a Dehn twist by [L] ∈ H3 (X̂), which has infinite order. It was known [9]
that if an essential Lagrangian sphere existed on a rigid CY3, then it could not
be the vanishing cycle of a nodal degeneration. Our construction produces four
non-isomorphic rigid Calabi-Yau threefolds with Euler characteristics 66, 72, 80,
and 96, each containing essential Lagrangian spheres.

1
2 FRANÇOIS GREER

Acknowledgments. The author is grateful to Mark McLean for suggesting the

problem, and to the anonymous referees for valuable comments. He has also ben-
efitted from communications with Denis Auroux, Jonathan Evans, Ivan Smith,
Richard Thomas, and Abigail Ward.

2. The Construction
The counterexamples are among the Calabi-Yau threefolds considered by Schoen [7].
Consider the following pencil of cubics in P2 :
(x + y)(y + z)(z + x) + txyz = 0.
Viewed as a family of curves over P1 , the relatively minimal smooth model
ν : S → P1
has four singular Kodaira fibers of types I6 , I3 , I2 , and I1 over t = ∞, 0, 1, and −8,
respectively. This is one of six semistable elliptic families over P1 (all extremal)
with the minimum number of singular fibers, as constructed by Beauville [2], and
it is isomorphic to the universal family over the compactified modular curve X1 (6).
See [5] for a comprehensive reference on elliptic surfaces with four singular fibers.

Let φ be a non-trivial automorphism of P1 which permutes {∞, 0, 1}, and note

that φ(−8) 6= −8. We form the Cartesian product
X /S
π
φ◦ν
 
S
ν / P1 .
The result is a singular projective threefold with KX = 0, fibered over P1 by abelian
surfaces which are products of non-isogenous elliptic curves. There are five critical
values: ∞, 0, 1, −8, and φ(−8). The total space X has n conifold singularities in
the fibers over {∞, 0, 1}, located at the product of two nodes in the elliptic fibers.
For the different choices of φ, we get n = 33, 36, 40, 48. The value n = 36 arises
when φ acts as a 3-cycle; using either 3-cycle produces the same singular threefold.
The remaining values of n arise from transpositions.

There exists a projective small resolution  : X̂ → X obtained by successive blow

ups of the n Weil divisors which are irreducible components of the singular fibers.
The resolution is crepant, so X̂ is a smooth Calabi-Yau threefold. We will use π
interchangeably with π ◦  when no confusion can arise.
Proposition 3. The Picard group of X̂ has rank n.
Proof. The specialization of π : X̂ → P1 to the generic point η ∈ P1 gives a split
short exact sequence
0 → A → Pic(X̂) → Pic(Xη ) → 0,
where A is the span of the n divisor classes supported over {∞, 0, 1}. They satisfy
2 relations using the rational equivalence over P1 . The generic fibers of ν and φ ◦ ν
are non-isogenous elliptic curves, so
Pic(Xη ) ' Pic(Sη ) ⊕ Pic(Sη ).
 A LAGRANGIAN SPHERE WHICH IS NOT A VANISHING CYCLE 3

The specialization of ν : S → P1 to η gives a split short exact sequence

0 → B → Pic(S) → Pic(Sη ) → 0,
where B is the span of the 12 curves classes supported over {∞, 0, 1, −8}. They
satisfy 3 relations using the rational equivalence over P1 . Since ρ(S) = 10, we find
that Pic(Sη ) has rank 1. In fact, the torsion Mordell-Weil group of Sη is known [6]:
Pic(Sη ) ' Z ⊕ Z/6Z.

Proposition 4. The threefold X̂ is rigid in the sense that H 1 (TX̂ ) = 0.

Proof. The cup product gives an isomorphism H 1 (TX̂ ) ' H 1,2 (X̂)∨ since X̂ is
Calabi-Yau. As X is fibered by tori, its topological Euler characteristic is deter-
mined by the singular fibers Ib × Ib0 . This gives
χtop (X) = n
χtop (X̂) = 2n,
since the small resolution replaces each conifold point with a P1 . On the other
hand, the Calabi-Yau property gives
 
χtop (X̂) = 2 h1,1 (X̂) − h1,2 (X̂) ,

and H 1,1 (X̂) ' Pic(X̂), so we are done by Prop. 3. 

Proposition 5. X̂ contains an embedded 3-sphere L that is Lagrangian with respect

to a Kähler form.
Proof. We adapt a construction from [9]. Let γ : [0, 1] → P1 be an embedded path
missing {∞, 0, 1} with γ(0) = −8 and γ(1) = φ(−8). Choose a Kähler form ω on X̂
which agrees over a neighborhood of γ with the sum of pullbacks of Kähler forms
from each elliptic surface factor S (this is explained below). Using the horizontal
distribution ω-orthogonal to the π-vertical tangent spaces, there is a symplectic
parallel transport along γ. By our choice of ω, the transport
preserves the product
structure E × E 0 on the fibers over γ. Let `0
⊂ π −1 γ( 21 ) be a vanishing loop in
E for the flow toward 0, and `1 ⊂ π −1 γ( 12 ) a vanishing loop in E 0 for the flow
toward 1. The parallel transport along γ sweeps out a Lagrangian:
[
L= (`0 )γ(s) × (`1 )γ(s) ⊂ X̂,
s∈[0,1]

diffeomorphic to S 3 fibered by 2-tori, with S 1 caps on either side. It is smooth at

the caps because it is locally the product of a Lefschetz thimble with S 1 . 
To show the existence of suitable Kähler forms on X̂, we use the main idea of [10].
Replacing φ in the construction of X with a generic nearby automorphism φ0 of P1 ,
we obtain a deformation smoothing Y of X. Schematically,

X̂


Y X.
4 FRANÇOIS GREER

This Y is commonly referred to as the Schoen manifold. All the fibers of Y → P1

have a smooth elliptic factor (ν and φ0 ◦ ν share no critical values), so χtop (Y ) = 0.
By an argument similar to the proof of Prop. 3, we find that h1,1 (Y ) = 19, so
h1,2 (Y ) = 19 as well. Since Y is a smooth fibered product of projective surfaces,
it admits a Kähler form ωY which is the sum of pullbacks of Kähler forms from
each factor. This form induces a closed form on X̂ which is Kähler away from the
exceptional curves. In a slight abuse of notation, we will refer to this form as ωY too.

There are n Lagrangian vanishing spheres Li ⊂ Y for the degeneration to X,

constructed as in the proof of Prop. 5, using the short paths from φ(t) to φ0 (t) for
t ∈ {∞, 0, 1}. The passage from Y to X̂ is called a conifold transition. Topologi-
cally, it is one of the 2n possible surgeries replacing each Lagrangian neighborhood
T ∗ Li ' T ∗ S 3 with
Tot(OP1 (−1) ⊕ OP1 (−1)).
1
In our case, each such P is an exceptional curve Ci of the projective small resolution
 : X̂ → X. For a general conifold transition between 6-manifolds, we have:
Theorem 6. [10] Fix a symplectic 6-manifold Y with a collection of n disjoint
Lagrangian spheres Li . There exists a symplectic structure on one of the 2n conifold
transitions in the Li such that the resulting Ci are symplectic if and only if there is
a relation in homology:
Xn
λi [Li ] = 0 ∈ H3 (Y, Z),
i=1
with λi 6= 0 for all i.
Our conifold transition satisfies the conditions of Thm. 6 because X̂ is projective.
The proof of the ‘if’ part of the theorem proceeds by finding a 4-chain η on Y with
boundary equal to ∪i Li , realizing the relation above. After passing through the
conifold transition surgery, η becomes a 4-cycle whose Poincaré dual is represented
by a closed 2-form σ. Using cutoff functions and the fact that H 2 (Ci , R) ' R, one
can arrange that σ restricts to a J-tame form on a tubular neighborhood Ui of each
exceptional curve Ci , where J is the integrable almost complex structure on X̂.
Since ωY is J-tame away from the Ci and X̂ r ∪i Ui is compact, there exists N  0
such that N ωY + σ is J-tame everywhere. Now,
H 0,2 (X̂) = H 2,0 (X̂) = 0
implies that every closed 2-form has type (1,1), so J-tame forms are Kähler. The
key point is that the modification of ωY is local around η. We will produce a 4-chain
η on Y disjoint from the fibers over the path γ in Prop. 5. This means that the
Kähler form on X̂ can be chosen to agree with ωY in a neighborhood of π −1 (γ).

Let g : [0, 1] → P1 be a path from φ(t) to φ0 (t) for t ∈ {∞, 0, 1}. The fiber of
Y → P1 over g( 12 ) is a product of elliptic curves: E × E 0 . Now, E (resp. E 0 )
contains b (resp. b0 ) disjoint homologous loops which get pinched to nodes in the
fibers by the flow s → 0 (resp. s → 1). If `0 is such a loop in E, `1 and `01 are
adjacent such loops in E 0 , and K ⊂ E 0 is a cylinder with boundary `1 ∪ `01 , then
the parallel transport along g sweeps out a 4-chain:
[
(`0 )g(s) × (K)g(s) ⊂ Y.
s∈[0,1]
 A LAGRANGIAN SPHERE WHICH IS NOT A VANISHING CYCLE 5

Its boundary is the union of two vanishing 3-spheres and an S 1 × S 2 given by

(`0 )g(1) cross a P1 component of (φ0 ◦ ν)−1 (g(1)). Repeating this construction
for all other choices of adjacent loops in either factor, weighted appropriately, we
obtain the desired η, which is supported over the short paths from φ(t) to φ0 (t) for
t ∈ {∞, 0, 1}, disjoint from π −1 (γ).
Remark 7. The construction above only works because b, b0 > 1 (in our case,
they are 2, 3, or 6). Otherwise the cylinder K would not have a boundary. This
observation fits with the description of the projective small resolution  : X̂ → X as
a blow up. The irreducible components of a fiber Ib × Ib0 are non-Cartier divisors
in the threefold if and only if b, b0 > 1; see Lemma 3.1 of [7].
Lastly, we point out that the conifold transition lets us compute the Betti numbers
of X using Thm. 2.11 of [10]. If r is the rank of the span of [Li ] in H3 (Y, Z), then
b2 (X̂) = b2 (X) + n − r = b2 (Y ) + n − r;
b3 (X̂) = b3 (X) − r = b3 (Y ) − 2r;
b4 (X̂) = b4 (X) = b4 (Y ) + n − r.
Since b3 (X̂) = 2 and b3 (Y ) = 40, we deduce that r = 19.

3. Elliptic Modular Surfaces

The classical modular curves X1 (N ) are constructed by compactifying quotients of
the upper half plane H by the congruence subgroups Γ1 (N ) ⊂ SL2 (Z):
  
a b
Γ1 (N ) := ∈ SL2 (Z) : a ≡ d ≡ 1 (6), c ≡ 0 (6) ;
c d

H∗ := H ∪ P1 (Q);
X1 (N ) := H∗ /Γ1 (N ).
The action of Γ1 (N ) on P1 (Q) has finitely many orbits, which become cusps in the
modular curve. The stabilizer of a point in P1 (Q) is a parabolic subgroup of Γ1 (N )
generated by a conjugate of  
1 b
,
0 1
where b ∈ N is called the width of the corresponding cusp. There is a universal
family over H/Γ1 (N ) whose fiber at τ is the elliptic curve C/(Z + Zτ ). This family
admits a compactification over X1 (N ) by adding a fiber of Kodaira type Ib over
each cusp. For 1 ≤ N ≤ 10 and N = 12, the curve X1 (N ) has genus 0, and
the Hauptmodul jN : X1 (N ) → P1 is an isomorphism defined over Q. For our
application, we specialize to the case N = 6 where the elliptic modular surface is
isomorphic to our example ν : S → P1 by [2].

Toward our ultimate goal of describing the homology of the Calabi-Yau threefold
X̂, we record the monodromy representation associated to ν : S → P1 .
Proposition 8. For a chosen base point ∗ ∈ P1 r {−8, ∞, 1, 0}, and a star-shaped,
cyclically ordered collection of paths from ∗ to −8, ∞, 1, and 0, the monodromy
π1 (P1 r {−8, ∞, 1, 0}, ∗) → SL(H1 (ν −1 (∗), Z)) ' SL2 (Z)
6 FRANÇOIS GREER

maps the corresponding positively oriented loops γ−8 , γ∞ , γ1 , γ0 respectively to

       
1 1 1 0 −5 2 −5 3
M1 = , M6 = , M2 = , M3 = ∈ Γ1 (6).
0 1 −6 1 −18 7 −12 7
Proof. The action of Γ1 (6) on H has a fundamental domain G, which can be built
by translating the strip fundamental domain F for P SL2 (Z) by 12 representatives
gi for the left cosets of Γ1 (6) ⊂ P SL2 (Z):
12
[
G= gi−1 F.
i=1

The coset space is identified with the P SL2 (Z) orbit of (1, 0) in (Z/6Z)2 / ± 1. If
we choose the coset representatives
           
1 0 0 1 −1 1 −2 1 −3 1 −4 1
, , , , , ,
0 1 −1 0 −1 0 −1 0 −1 0 −1 0
           
−5 1 −5 3 −7 4 −9 5 −11 4 −14 5
, , , , , ,
−1 0 −2 1 −2 1 −2 1 −3 1 −3 1
then G is contractible and has cusps at τ = i∞, 0, 31 , 12 . The universal family
of elliptic curves over G is topologically trivial, so for any base point τ0 in the
interior of G, we can take the standard basis for H1 (ν −1 (τ0 ), Z) ' Z2 . Choose
non-intersecting paths in G from τ0 limiting to each of the four cusps. The quotient
G/Γ1 (6) is homeomorphic to a sphere with four punctures, and π1 (G/Γ1 (6), τ0 ) is
generated by four oriented loops around the punctures, conjugated by the paths
above, with the relation that their cyclically ordered product is 1. The Γ1 (6)-
stabilizer of each cusp of G is generated by the respective matrix in the statement
of the proposition, and this gives the monodromy action on Z2 . The isomorphism
j6 : X1 (6) → P1 must identify the cusps with the critical values of the pencil ν
matching cusp widths with Kodaira types, so j6 (i∞) = −8, j6 (0) = ∞, j6 ( 31 ) = 1,
and j6 ( 12 ) = 0. We set ∗ = j(τ0 ). 

4. Homology class of L
To show that the Lagrangian sphere L ⊂ X̂ is essential (for all choices of φ and γ),
we construct a second Lagrangian sphere that intersects it non-trivially. Recall that
γ is an embedded path from −8 to φ(−8) missing the other critical values, and let
U ⊂ P1 be a small neighborhood of γ. Let V ⊂ P1 be an open set containing another
critical value c with fiber Ib × Ib0 such that b 6= b0 , overlapping with U as pictured
below in the case where b = 2 and b0 = 6. Such a value exists because φ permutes
{∞, 0, 1} non-trivially. Let γ 0 be another path from −8 to φ(−8) embedded in
U ∪ V r {c}, not homotopic to γ in that complement, and meeting γ only at −8
and φ(−8). The construction of Prop. 5 applied to γ 0 gives a second Lagrangian
sphere L0 . Note that as a set,
L ∩ L0 ⊂ π −1 (−8) ∪ π −1 (φ(−8)).
Proposition 9. The intersection product [L] · [L0 ] is a nonzero multiple of 6.
Proof. Choose a basis for homology of the smooth fiber H1 (E × E 0 , Z) ' Z2 ⊕ Z2
0
  point of γ such that the vanishing loops `0 ⊂ E and `1 ⊂ E each
over an interior
1
have class in their respective copy of Z2 . The intersection L ∩ π −1 (φ(−8)) is
0
 A LAGRANGIAN SPHERE WHICH IS NOT A VANISHING CYCLE 7

the parallel transported loop `0 in the smooth elliptic factor, cross the node in the
I1 factor. The intersection L0 ∩ π −1 (φ(−8)) is a different loop in the smooth elliptic
factor, cross the node in the I1 factor. Hence, the contribution to [L] · [L0 ] from
points in π −1 (φ(−8)) can be computed on the smooth elliptic factor E. Likewise,
the contribution to [L] · [L0 ] from π −1 (−8) can be computed on the smooth elliptic
factor E 0 . In terms of the (conjugated) monodromy matrices from Prop. 8,

       
1 1 1 1
[L] · [L0 ] = · gMb g −1 + · h−1 Mb−1
0 h .
0 0 0 0

Here g, h ∈ Γ1 (6) come from the ambiguity in choosing the paths γ and γ 0 in
the punctured sphere. Each summand above is equal to the lower left entry in a
monodromy matrix. Observe that the matrices Mb (b = 2, 3, 6) in Prop. 8 have
lower left entries −36/b. Conjugating Mb by an element of Γ1 (6) does not change
the lower left entry modulo 36. Therefore, the difference of any two such entries
for b 6= b0 is a nonzero multiple of 6. 
8 FRANÇOIS GREER

5. Degenerations of X̂
Suppose that X̂ admits a Kähler degeneration. That is, X̂ is isomorphic to a fiber
of a proper holomorphic family over a complex disk
f : X → ∆,
−1
and f (0) is singular. By Prop. 4, X̂ has no moduli so f is holomorphically locally
trivial away from 0 ∈ ∆, by the Fischer-Grauert theorem. The fiber bundle
f −1 (∆∗ ) → ∆∗
has monodromy valued in Aut(X̂). In other words,
Z ' π1 (∆∗ ) → Aut(X̂) → Sp(H3 (X̂, Z)) ' SL2 (Z).
To control potential isotrivial degenerations, we prove that Aut(X̂) is finite.
Proposition 10. Every automorphism preserves the fibration π : X̂ → P1 .
Proof. Let ϕ : X̂ → X̂ be an automorphism, and let A be a general fiber of π. If
ϕ does not preserve the fibration, then ϕ(A) surjects onto P1 . The image of ϕ(A)
in S cannot be all of S because complex tori only surject onto projective spaces
and complex tori [3]. Thus, φ(A) maps onto a curve C in S, and the generic fiber
is an elliptic curve. There must be singular fibers because C surjects to P1 , which
contradicts the fact that χtop (A) = 0. 
Proposition 11. The group Aut(X̂) is isomorphic to (Z/6Z)2 .
Proof. Recall that π has three non-isomorphic singular fibers, so any ϕ ∈ Aut(X̂)
satisfies π = π ◦ ϕ. This implies that Aut(X̂) = Aut(Xη ) ' Aut(Sη ) × Aut(Sη ),
since the generic fiber is a product of non-isogenous elliptic curves. As an elliptic
curve over K ' C(t) with non-constant j-invariant, Sη has automorphism group
isomorphic to its group of K-points (Mordell-Weil group), which is Z/6Z by [6]. 
Therefore, the image of monodromy is a finite subgroup of SL2 (Z), so it is abelian
of order ≤ 6. In particular, L is not the vanishing cycle of a nodal degeneration.

References
[1] D. Auroux. The canonical pencils on Horikawa surfaces. Geom. Topol., 10(4):2173–2217, 2006.
[2] A. Beauville. Les familles stables de courbes elliptiques sur P1 admettant quatre fibres sin-
gulières. C.R. Acad. Sc. Paris, 294(19):657–660, 1982.
[3] J.-P. Demailly, J.-M. Hwang, and T. Peternell. Compact manifolds covered by a torus. J.
Geom. Anal., 18(2):324–340, Apr 2008.
[4] S. Donaldson. Polynomials, vanishing cycles and Floer homology. In V. Arnold, M. Atiyah,
P. Lax, and B. Mazur, editors, Mathematics: Frontiers and Perspectives, The American
Mathematical Monthly, pages 55–64. American Mathematical Society, 2000.
[5] S. Herfurtner. Elliptic surfaces with four singular fibres. Math. Ann., 291(1):319–342, 1991.
[6] R. Miranda. The basic theory of elliptic surfaces. Dottorato di Ricerca in Matematica. ETS
Editrice, Pisa, 1989.
[7] C. Schoen. On fiber products of rational elliptic surfaces with section. Math. Z., 197(2):177–
199, Jun 1988.
[8] N. Sheridan and I. Smith. Symplectic topology of K3 surfaces via mirror symmetry. math.SG,
arXiv:1709.09439.
[9] I. Smith and R. Thomas. Symplectic surgeries from singularities. Turkish J. Math., 27(1):231–
250, 2003.
[10] I. Smith, R.P. Thomas, and S.-T. Yau. Symplectic conifold transitions. J. Differential Geom.,
62(2):209–242, 2002.