Lagrangian geometry of algebraic varieties

Nikolay Tyurin
BLTPh JINR (Dubna) and Volga Region Mat Center (Kazan)
arXiv:2109.00209v1 [math.AG] 1 Sep 2021

Abstract
Every algebraic variety can be regarded as a symplectic manifold being
equipped with a Kahler form. Therefore it is natural to study lagrangian
geometry of any algebraic variety. We present two basic constructions
which can be applied to a sufficiently wide set of algebraic varieties.
Lagrangian geometry. Every compact algebraic variety X by the very
definition (see [1]) admits principal polarization so a very ample line bundle
L which defines an embedding X ֒→ CPN to a projective space. Therefore if
a standard Kahler form on the last one is fixed it induces the corresponding
Kahler form on X; the form is not unique but the corresponding lagrangian
geometry essentially depends on the cohomology class c1 (L) ∈ H 2 (X, Z) only.
Thus the choice c1 (L) on X discovers a new direction in the studies of X:
taking the corresponding Kahler form ω as a symplectic form one can pose the
following questions. First, one says that a real submanifold S ⊂ X is lagrangian
if ω|S identically vanishes and dimC X = dimR S. And the first classification
question arises as follows: which homology classes νi ∈ Hn (X, Z) are realizable
by smooth lagrangian submanifolds? And the next basic question is what are
possible topological types of smooth lagrangian submanifolds?
Example: the complex projective plane CP2 admits essentially unique prin-
cipal polarization given by L = O(1) which corresponds to the standard Kahler
form for the Fubini - Study metric. Just recently the answer to the basic ques-
tion of the lagrangian geometry has been completed for this case: one proves
that the Klein bottle can not appear as a smooth lagrangian submanifold, and
the complete list of possible lagrangian submanifolds contains real projective
plane RP2 and torus T 2 both representing the trivial homology class.
More specific questions arise when one starts the classification programme
for lagrangian submanifolds: if two lagrangian submanifolds S1 and S2 can be
joined by a family of lagrangian submanifolds St then one says that S1 and S2
are equivalent up to lagrangian deformation; if it exists a Hamiltonian function
H(x, t) such that the corresponding flow φtXH moves S1 to S2 then these are
called Hamiltonian isotopic. The last equivalence is much more delicate then the
first one. Then a high level classification question for a given algebraic variety
with a fixed principal polarization asks how many equivalence classes (up to
lagrangian deformation or up to Hamiltonian isotopy) exists for a given class
νi ∈ Hn (X, Z) and a given topological type topS. It is not hard to see that the
answer does not depend on the particular choice of the Kahler form but on the
class c1 (L) only.
Example. Consider the case of lagrangian tori in CP2 . Since the projective
plane is toric then it exists the family of Liouville tori which are lagrangian by

1
the Liouville theorem; at the same time one of them is the famous Clifford torus.
Untill 1996 one thought that no other lagrangian tori of the same periods but
Hamiltonian non isotopic do not exist, and Yu. Chekanov construction (see [2])
presented an example of lagrangian torus which has the same periods but which
is not Hamiltonian isotopic to the Clifford torus. This torus was called exotic.
Generalization of Chekanov construction. More general, every toric
variety by the very definition admits Liouville type tori: such a variety is pre-
sented by the phase space of completely integrable system, and the common level
sets for the first integrals are lagrangian tori. Thus toric geometry is a subject
where algebraic geometry meets symplectic geometry: T. Delzant proved that
all the data are encoded by a simple combinatoric object, a convex polytop PX ,
and both the symplectic geometry and algebraic geometry can be recovered
from PX (see [3]). The Chekanov construction can be translated to the alge-
braic language as follows: the components of the boundary divisor (= preimage
of the boundary ∂PX ) can be combined in the elements of a pencil of algebraic
divisors such that each element is invariant under the Hamiltonian action of
one first integral, say, f1 . Therefore for CP2 instead of a pair of first integrals
(f1 , f2 ) one considers the pair (f1 , {Q}), where the second element is a pencil
of conics on CP2 : the integrability requires that the Hamiltonian action of f1
preserves the fibers of the pencil. Then the base of the pencil CP1 admits two
points p± which underly singular fibers, and every smooth loop γ ⊂ CP1 \{p± }
gives a lagrangian torus in whole CP2 ; if γ is non contractible then the resulting
torus is of the Clifford type, if γ is contractible then the torus is exotic (or of
the Chekanov type).
Therefore we can generalize the Chekanov construction for any toric alge-
braic variety X. First of all we need to extract certain information from the
convex polytop PX ⊂ Rn . It is formed by r linear inequalities bounded by facets
∆i of PX such that the Picard group PicX subjects the exact sequence

0 → Zn → Zr → PicX → 0,

where the second Zr is spanned on the components Di of the boundary divisor

Db . Then each point of Rn gives a relation on Di of the form
X
r
ai [Di ] = 0 in PicX.
i=1

Since each Di is effective, realized by a subvariety, all the coefficients ai can not
be of the same signs, and we can rearrange the last equality as follows
X X
ai [Di ] = |ai |[Di ],
ai ≥0 ai <0

denoting the corresponding divisors as D+ and D− ; by the very construction

they represent the same class [D+ ] = [D− ] in the Picard group. Therefore if
we take the corresponding line bundle L → X such that c1 (L) = P.D.[D± ] then
it exists a pencil < D+ , D− >⊂ |L| in the complete linear system, defined by
holomorphic sections of L. Excluding the base set B = D+ ∩ D− from X one
gets a map ψ : X\B → CP1 ; it is not hard to see that the points p± = ψ(D± 0
)
exhaust the set of points underly singular fibers. At the same time the choice
of the pencil < D+ , D− > derives a linear condition on the complete set of first

2
integrals (f1 , ..., fn ) such that it exists a set (f˜1 , ..., f˜n−1 ) of integrals such that
each element of the pencil is invariant with respect to the Hamiltonian action
of each f˜i .
Therefore for a toric algebraic variety X one can fix data (f˜1 , ..., f˜n−1 , ψ)
choosing a direction in Rn ; for a fixed data every smooth loop γ on the comple-
ment < D+ , D− > \{p± } and a generic value set c1 , ..., cn−1 for functions f˜i we
have the union
Tγ,ci = ∪p∈γ (ψ −1 (p) ∩ N (c1 , ..., cn−1 )),
where N (c1 , ..., cn−1 ) is the common level set for the functions f˜i . The Main
theorem says that Tγ,ci is a smooth lagrangian torus: if γ is non contractible
in < D+ , D− > \{p± } then the torus is of the standard type, if γ is contractible
then the resulting torus is of exotic type, see [4]. Moreover, in [5] one proves
that if X admits a standard monotone lagrangian torus then it exists an exotic
lagrangian torus which is monotone as well.
At the same time one can apply the construction to a non toric algebraic
variety: in [4] one presents the structure (f1 , ..., fn−1 , ψ) on flag variety F 3 ,
which is the full flag in C3 . This variety is non toric, however it admits natural
non complete set of first integrals (f1 , f2 ) and it is possible to find a pencil
ψ : F 3 \B → CP1 such that the fibers are invariant under the Hamiltonian
action of fi . In this non toric situation the base CP1 contains three points
which underly singular fibers therefore we have more complicated picture since
a smooth loop on the complement ”base minus three points” can represent
different classes being non contractible. In this setup one can reconstruct the
famous Gelfand - Zeytlin lagrangian sphere, see [4]; different types of smooth
lagrangian tori can be derived from the picture, see [5].
Generalization of Mironov construction. In [6] A. Mironov presented
new example of (Hamiltonian) minimal lagrangian submanifolds in Cn and CPn .
The geometric essence of the Mironov construction is the following: one uses
together the standard toric structures and the natural real structures which
exist on both the varieties. From this point of view it is not hard to find a
generalization of the Mironov construction: it can be applied to an algebraic
variety which admits a real structure and a (possible non complete) toric action
which is transversal to the real structure. If X is our algebraic variety and
ω is compatible with the real structure then the real part XR ⊂ X must be
isotropical with respect to ω. If moreover the real part has maximal possible
dimension dimR XR = dimC X then XR is lagrangian.
Suppose we have a Hamiltonian toric action spanned by moment maps (or
first integrals) f1 , ..., fk . Then for a generic value set c1 , ..., ck one takes the
common level set N (c1 , ..., ck ) = {fi = ci } and the intersection SR (c1 , ..., ck ) =
XR ∩N (c1 , ..., ck ). Since the toric action is transversal to the real structure by the
assumption the last submanifold SR (c1 , ..., ck ) is transversal to the Hamiltonian
vector fields Xfi , and if we apply the toric action, generated by Xfi , to the last
submanifold we get T k (SR (c1 , ..., cn )) ⊂ X which is a subcycle (since possibly it
has self intersections). In [7] one proves the Main theorem which states that
T k (SR (c1 , ..., ck )) is a lagrangian immersion (in the smooth case – lagrangian
submanifold) in X. Using different sets of fi ’s one presents different types of
lagrangian submanifolds in the Grassmannian Gr(2, 4). At the end in [7] one
proposes that the Grassmannian Gr(2, n) admits at least n + 1 topologically
distinct lagrangian submanifolds.

3
 Indeed, the Grassmann variety Gr(k, n) gives a natural example of algebraic
variety which admits natural real structure and toric structure of the desired
types. As the space of k − 1 - dimensional projective subspaces of CPn−1 it
inherits the toric action of T n−1 coming from the standard toric action on
CPn−1 ; on the other hand it admits the natural real structure such that the
real part GrR (k, n) has right dimension. For the construction one can take
any subtorus in T n−1 , which leads to the natural grading by the rank of this
subtorus, starting with XR itself, which is claimed to be of the zero level. In
[8] one realizes the programme for the level 1 case. Main theorem here reads
as follows: Mironov cycle is smooth Lagrangian submanifold, isomorphic to
topologically non trivial fiber bundle over real Grassmannian GrR (k, n − 1),
where the fiber is either S 1 × S k−1 for even k or generalized Klein bottle for
for odd k. In particular for Gr(2, n) for any n it gives certain topologically non
trivial T 2 - bundle over GrR (2, n − 1).
References:
[1] P. Griffits, J. Harris, ”Principles of algebraic geometry”, NY, Wiley, 1978;
[2] Yu. Chekanov, ”Lagrangian tori in a symplectic vector space and global
symplectomorphisms”, Math. Z., 223: 4 (1996), 547–559;
[3] T. Delzant, ”Hamiltoniens périodiques et images convexes de l’application
moment”, Bull. Soc. Math. France, 116: 3 (1988), 315–339;
[4] N. Tyurin, ”Pseudotoric structures: Lagrangian submanifolds and La-
grangian fibrations”, Russian Mathematical Surveys (2017), 72: 3, 513 - 546;
[5] N. Tyurin, ”Monotonic Lagrangian Tori of Standard and Nonstandard
Types in Toric and Pseudotoric Fano Varieties”, Proc. Steklov Inst. Math.,
307 (2019), 267–280;
[6] A. Mironov, ”New examples of Hamilton-minimal and minimal Lagrangian
manifolds in Cn and CPn ”, Sb. Math., 195: 1 (2004), 85–96;
[7] N. Tyurin, ”Mironov Lagrangian cycles in algebraic varieties”, Sb. Math.,
212:3 (2021), 389–398;
[8] N. Tyurin, ”Examples of Mironov cycles in Grassmann varieties”, Siberian
Math. J., 62: 2 (2021), 457–465.