0% found this document useful (0 votes)
48 views32 pages

Weinstein Manifolds Revisited: Yakov Eliashberg

This document provides an introduction to Weinstein manifolds and related concepts: 1) It defines Liouville domains and manifolds, and discusses how they are related. Liouville domains can be completed to Liouville manifolds by attaching cylindrical ends. 2) Weinstein structures impose additional conditions on Liouville structures, including requiring a Lyapunov function for the Liouville vector field. This ensures the core is stratified by isotropic submanifolds. 3) Weinstein cobordisms generalize Weinstein structures to cobordisms between contact manifolds, with additional conditions on the Liouville vector field and Lyapunov function along the boundaries.

Uploaded by

muler
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
48 views32 pages

Weinstein Manifolds Revisited: Yakov Eliashberg

This document provides an introduction to Weinstein manifolds and related concepts: 1) It defines Liouville domains and manifolds, and discusses how they are related. Liouville domains can be completed to Liouville manifolds by attaching cylindrical ends. 2) Weinstein structures impose additional conditions on Liouville structures, including requiring a Lyapunov function for the Liouville vector field. This ensures the core is stratified by isotropic submanifolds. 3) Weinstein cobordisms generalize Weinstein structures to cobordisms between contact manifolds, with additional conditions on the Liouville vector field and Lyapunov function along the boundaries.

Uploaded by

muler
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 32

Weinstein manifolds revisited

Yakov Eliashberg∗
arXiv:1707.03442v2 [math.SG] 29 Aug 2017

Department of Mathematics, Stanford University

Abstract

This is a very biased and incomplete survey of some basic notions, old and new results, as
well as open problems concerning Weinstein symplectic manifolds.

1 Weinstein manifolds, domains, cobordisms

We begin with a notion of a Liouville domain. Let (X, ω) be a 2n-dimensional com-


pact symplectic manifold with boundary equipped with an exact symplectic form ω.
A Liouville structure on (X, ω) is a choice of a primitive λ, dλ = ω, called Liou-
ville form such that λ|∂X is a contact form and the orientation of ∂X by the form
λ ∧ dλn−1 |∂X coincides with its orientation as the boundary of symplectic manifold
(X, ω). The vector field Z, that is ω-dual to λ, i.e. ι(Z)ω = λ, is also called Liou-
ville. It satisfies the condition LZ ω = ω which means that its flow is conformally
symplectically expanding. The contact boundary condition is equivalent to the out-
ward transversality of Z to ∂X. A Liouville domain X can always be completed to
a Liouville manifold X b by attaching a cylindrical end:

b := X ∪ (∂X × [0, ∞))


X
b as equal to es (λ|∂X ) on the attached end. We will be constantly
and extending λ to X
going back and forth between these two tightly related notions of Liouville domains
and Liouville manifolds.
Given a Liouville structure L = (X, ω, Z) we say that a Liouville structure L0 =
(X 0 , ω, Z) is obtained by a radial deformation from L if there exists a function h :

Partially supported by NSF grant DMS-1505910

1
1 Weinstein manifolds, domains, cobordisms 2

Xb → R such that X 0 ⊂ X b →X
b is the image of X under the time 1 map ψ : X b of
the flow of the vector field hZ on the completion X.
b The completions of the radially
0
equivalent Liouville domains L and L are canonically isomorphic.
The space of Liouville structures for (X, ω) is convex, and hence any two Liouville
structures are canonically homotopic. Given a homotopy of completed Liouville
structures (X, b → X
b ωt , λt ) there exists an isotopy φt : X b such that φ∗ ωt = ω0 .
t
Moreover, one can always arrange that φ∗t λt = λ0 + dHt , see [11], Sections 11.1 and
11.2. In particular on completed Liouville manifolds it is always sufficient to con-
sider homotopies fixing the symplectic form, and, moreover, changing the Liouville
form by adding an exact form. Homotopic non-completed Liouville domains are
symplectomorphic up to radial deformation.
Given a Liouville domain L = (X, ω, λ) consider a compact set
\
Core(L) = Z−t (X),
t>0

the attractor of the negative flow of the Liouville vector field Z. We will call Core(L)
the core, or the skeleton of the Liouville structure L. While Core(L) has obviously its
2n-dimensional Lebesgue measure equal to 0, it still can be pretty large if no extra
conditions are imposed on the Liouville structure. For instance, McDuff constructed
in [31] a Liouville structure on T ∗ Sg \ Sg for a closed surface Sg of genus g > 1, whose
core has codimension 1.
However, the situation changes if one requires existence of a Lyapunov function for
the Liouville vector field Z. A Weinstein structure on a domain X is a Liouville
structure L together with a function φ : X → R which is Lyapunov for the Liouville
vector field Z, i.e.

(L1) dφ(Z) > c||Z||2 for a positive constant c and some Riemann metric on X.

Note that condition (L1) implies that Core(X, λ) is the union of Z-stable manifolds
of critical points of φ (i.e. points converging to the critical locus in forward time).
In [11] it was required in addition that φ is either Morse or generalized Morse (i.e.
may have death-birth critical points). Under these assumptions it was shown in [11],
see also [20, 15], that

(L2) the core is stratified by isotropic for λ, and hence for ω submanifolds.

F. Laudenbach proved, see [30], that if the flow of Z is Morse-Smale (i.e. stable and
unstable manifolds of critical points intersect transversely) and near critical points
1 Weinstein manifolds, domains, cobordisms 3

Core(Σ)

Fig. 1.1: Skeleton of a Weinstein domain

the vector field Z is gradient with respect to an Euclidean metric, then the skeleton
can be further Whitney substratified. It is likely that the Whitney condition also
holds if near its zeroes the vector field Z is gradient with respect to any Riemannian
metric. However, as far as know, this was never verified in the literature. The
Whitney condition need not hold if eigenvalues of the linearization of Z at critical
points have non-vanishing imaginary parts, as a spiraling phenomenon of trajectories
may occur.1
Condition (L2) holds for a much more general class of taming functions (e.g. when
φ is Morse-Bott), and hence for the the purposes of this paper we will take the
following working definition of a Weinstein structure, extending the class considered
in [11]: W = (X, λ, Z, φ) is Weinstein if it satisfies conditions (L1) and (L2) with
the Whitney condition and also condition

(L3) there exists a smooth family of Weinstein structures Wt = (X, λt , φt ), t ≥ 0


such that (λ, φ) = (λ0 , φ0 ) and φt is Morse for t > 0.
Problem 1.1. Which conditions (or maybe none?) on φ and Z are needed to deduce
(L2) and (L3) from (L1)?

A. Oancea suggested to me that a good sufficiently general condition on a Weinstein


structure could be to require that near critical points it is generated by a J-convex
1
I thank Francois Laudenbach for the discussion of the involved issues.
1 Weinstein manifolds, domains, cobordisms 4

function with respect to some (not necessarily integrable) almost complex structure
J, see [11], Chapter 1, for the details.
Remark 1.2. Note that not every closed subset C of a symplectic manifold which is
stratified by isotropic strata may serve as the skeleton for an appropriately chosen
Weinstein structure on a neighborhood of C (compatible with the given ambient
symplectic form). Examples of this kind exist already in R2 . For instance, let

C := {x = 0, y ≥ 0} ∪ {x = y 2 , y ≥ 0} ∪ {y = 0, x ≥ 0} ∪ {y = x2 , x ≥ 0}

be the union of 4 arcs emanating from the origin. Then there is no Liouville structure
on a neighborhood U 3 0 which has C ∩ U as a part of its skeleton. Indeed it is
straightforward to check that for any 1-form λ vanishing on C ∩ U one has (dλ)0 = 0.
Problem 1.3. Find a necessary and sufficient condition on a compact subset C of
a symplectic manifold to serve as the skeleton of some

a) Liouville, or

b) Weinstein

structure on its neighborhood. In particular, is it true that a Whitney stratified subset


C which is the skeleton of a Liouville structure on its neighborhood also serves as the
skeleton of a Weinstein structure?

+W

2 2
W W

−W
Fig. 1.2: Sutured Weinstein cobordism W with corners.

It is also useful to consider a notion of a Weinstein cobordism. This is a cobordism


(W, ∂− W = Y− , ∂+ W = Y+ ) endowed with a Liouville form λ, whose Liouville vector
2 Weinstein hypersurfaces and Weinstein pairs 5

field Z is outward transverse to ∂+ W and inward transverse to ∂− W , and a Lyapunov


(i.e. satisfying condition (L1)) function φ : W → R for the field Z. We also postulate
(L3) and an analog of condition (L2) for the core of the Weinstein cobordism, which
we define in that case as the stable manifold of the critical locus of φ. We will also
be considering Weinstein cobordisms between manifolds with boundary ∂± W . We
will view these cobordisms as sutured manifolds with a corner along the suture, see
Fig. 1.2. More precisely, we assume that the boundary ∂W is presented as the
union of two manifolds ∂− W and ∂+ W with common boundary ∂ 2 W = ∂+ W ∩ ∂− W
along which it has a corner. The vector field Z transversely enters W through ∂− W
and exits through ∂+ W , but of course, in this case the function φ cannot be chosen
constant on ∂− W and ∂+ W .
While any two Weinstein structures on the same symplectic manifold are (canon-
ically) homotopic as Liouville structures, the problem of existence of a Weinstein
homotopy is widely open.
Problem 1.4. Let (X,b λ0 , φ0 ) and (X,
b λ1 , φ1 ) be two completed Weinstein structures
on the same symplectic manifold (X, b ω). Are they homotopic as Weinstein struc-
tures?
In particular,
Problem 1.5. Let W = (X, b ω, λ, φ) be a completed Weinstein structure, and f :
Xb →X b a symplectomorphism. Is the pull-back Weinstein structure f ∗ W is Wein-
stein homotopic to W?
The Weinstein structure notion was introduced in [20] as a symplectic counterpart
of the notion of Stein complex structure, and inspired by the work of A. Weinstein
[45], see also [16, 15, 11].
I discussed the notions and problems considered in this paper with many people. I
am especially grateful to Daniel Alvarez-Gavela, Oleg Lazarev, David Nadler, Sheel
Ganatra, Vivek Shende, Laura Starkston and Kyler Siegel for contributing many
ideas and suggestions for improvement of the current text. I am very grateful to the
anonymous referee for critical remarks and many useful suggestions. Special thanks
to Nikolai Mishachev for making the pictures.

2 Weinstein hypersurfaces and Weinstein pairs

Weinstein hypersurfaces are special cases of Liouville hypersurfaces introduced by


Avdek in [3]. This and other related notions discussed in this paper are also similar to
2 Weinstein hypersurfaces and Weinstein pairs 6

“stops” of Sylvan, [42] and Liouville sectors of Ganatra-Pardon-Shende, [26]. Related


constructions are also considered in Ekholm-Lekili’s paper [14].

Weinstein hypersurfaces in a contact manifold


Let (Y, ξ) be a contact manifold. A codimension 1 submanifold Σ ⊂ Y with boundary
is called Weinstein hypersurface if there exists a contact form λ for ξ such that
(Σ, λ|Σ ) is compatible with a Weinstein structure on Σ, i.e. dλ|Σ is symplectic and
the Liouville vector field ZΣ on Σ dual to the Liouville form λ|Σ is outward transverse
to ∂Σ and admits a Lyapunov function φ : Σ → R. The Reeb vector field for λ is
transverse to Σ and the boundary ∂Σ of a Weinstein hypersurface Σ is a codimension
two contact submanifold of (Y, ξ).
Though the induced Weinstein structure on Σ depends on the choice of a contact
form, its skeleton is independent of this choice. Indeed, the Liouville fields for the
Liouville structures λ and f λ for a positive f > 0 are proportional. In fact, as
it is computed in Lemma 12.1 in [11] the form f λ is Liouville if and only if k :=
inf(f + df (Z)) > 0, where Z is the Liouville form for λ, and in that case the Liouville
vector field for f λ is equal to k1 Z. Moreover, the space of functions f for which f λ
is Liouville (and hence in the considered case Weinstein) is contractible.
It follows that the skeleton Core(Σ, λ|Σ ) is a stratified subset of Y which consists of
strata which are isotropic, and in the top dimension n − 1 are Legendrian for the
contact structure ξ.
Example 2.1. (i) Weinstein thickening of a Legendrian submanifold. Let Λ ⊂
(Y, ξ) be a Legendrian submanifold. Then it admits a Darboux neighbor-
hood U (Λ) isomorphic to (J 1 (Λ), dz − pdq), q ∈ Λ, ||p||2 + z 2 ≤ ε2 . Then
Σ(Λ) := U (Λ) ∩ {z = 0} is a Weinstein hypersurface symplectomorphic to the
cotangent ball bundle of Λ. Up to Weinstein isotopy the Weinstein thickening
Σ(Λ) is independent of all the choices. 2

(ii) Pages of open books. According to Giroux’s theorem [25], any contact manifold
admits an open book decomposition whose pages are Weinstein hypersurfaces.

(iii) Halves of convex hypersurfaces. Recall that a hypersurface Σ in a contact


manifold is called convex if it admits a transverse contact vector field, see
[20, 24]. The set D of points where the contact vector field is tangent to the
2
Warning: unlike the case of a Legendrian isotopy, an isotopy of Weinstein hypersurfaces does
not extend in general to an ambient contact diffeotopy.
2 Weinstein hypersurfaces and Weinstein pairs 7

contact plane field, called a dividing set, is generically a smooth hypersurface


which divides Σ into two Liouville manifolds. In many interesting examples
these Liouville manifolds are, in fact, Weinstein, and hence serve a rich source
of Weinstein hypersurfaces.
Given two Legendrian isotopic submanifolds Λ0 , Λ1 ⊂ (Y, ξ) their Weinstein thicken-
ings Σ(Λ0 ) and Σ(Λ1 ) are isotopic as Weinstein hypersurfaces.

Problem 2.2. Is the converse true?

Here by isotopy we mean an isotopy of unparameterized submanifolds.


Note that an isotopy of Weinstein hypersurfaces carries Λ0 to an exact Lagrangian
submanifold Λe 1 ⊂ Σ(Λ1 ). Moreover, there is a symplectomorphism ψ : Σ(Λ1 ) →
Σ(Λ1 ) such that ψ(Λe 1 ) = Λ1 . Hence, the positive answer to Problem 2.2 would
follow from the positive resolution of the following special case of the nearby La-
grangian conjecture: Lagrangians which are images of the 0-section under a global
symplectomorphism are Hamiltonian isotopic to the 0-section.
If the contact manifold (Y, ξ) is symplectically fillable then one can prove that the
Legendrian algebras LHA(Λ0 ) and LHA(Λ1 ) are isomorphic3 . It is likely that this
claim could be generalized to the case of a general contact manifold (Y, ξ).

Problem 2.3. Is there an analog of the Legendrian algebra LHA(Λ) for a general
Weinstein hypersurface?

Let us return to the case of the Legendrian homology algebra of a Legendrian sub-
manifold Λ and pick a contact form λ such that its Reeb vector field is tangent to
the contact submanifold ∆ := ∂Σ(Λ). We also choose an almost complex structure
J on ξ such that ξ ∩ T (∆) are J-invariant. This allows us to define a deformation
(A[t], D) of the Legendrian differentialP
algebra (A, ∂) := LHA(Λ) as follows. For a
generating chord c ∈ A define D(c) = (∂k c)tk , where ∂0 = ∂ and ∂k c counts holo-
k≥0
morphic curves with the intersection index k with the symplectization of ∆. This
symplectization is a complex hypersurface in the symplectization of Y , and hence
k ≥ 0. The sum defining differential D is finite due to the Gromov compactness.

Problem 2.4. Explore whether the above construction yields a genuinely new in-
variant of a Legendrian submanifold.
3
I thank Sheel Ganatra and Tobias Ekholm for the discussion of this problem.
2 Weinstein hypersurfaces and Weinstein pairs 8

Given a Weinstein hypersurface Σ ⊂ Y we slightly extend it to a larger Weinstein


hypersurface Σ e ⊃ Σ such that on Σ e \ Σ the Liouville form λ can be written as tλ|∂Σ ,
t ∈ [1, 1 + ε]. The extended hypersurface Σ e has a neighborhood U e diffeomorphic to
e × (−ε, ε) such that λ| e = π ∗ (λ| e ) + du where u is the coordinate corresponding to
Σ U Σ
the second factor and π the projection U e → Σ.
e Note that the level sets {u = const}
are translates of Σe under the Reeb flow of the contact form λ. Pick a non-negative
function h : Σe → R which is equal to 0 on Σ and to t − 1 near ∂ Σ e and set U (Σ) =
2 2 2
Uε (Σ) := {h + u ≤ ε } ⊂ U . The neighborhood U (Σ) will be called the contact
e
surrounding of a Weinstein hypersurface Σ.
Proposition 2.5. Contact manifolds Y \ U (Σ), Y \ Σ and Y \ Core(Σ, λ|Σ ) are
contactomorphic.

Let us first recall a few basic facts about convex hypersurfaces in contact manifolds.
If a germ ξ of a contact structure along a closed hypersurface V in a (2n − 1)-
dimensional manifold admits a transverse contact vector field v then we canonically
can construct a contact structure ξb on V × R which is invariant with respect to
translations along the second factor and whose germ along any slice V × t, t ∈ R, is
isomorphic to ξ. We will call ξb the invariant extension of the convex germ ξ.
Lemma 2.6. Let V be a closed (2n − 2)-dimensional manifold and ξ a contact
structure on Y = V ×[0, ∞) which admits a contact vector field v inward transverse to
V ×0 and such that its trajectories intersecting V ×0 fill the whole manifold Y (we do
not require v to be complete). Then (Y, ξ) is contactomorphic to (V ×[0, ∞), ξ),
b where
ξb is the invariant extension of the germ of ξ along V × 0. Moreover, for any compact
set C ⊂ Y , Int C ⊃ V ×0, there exists a contactomorphism h : (Y, ξ) → (V ×[0, ∞), ξ)
b
which is equal to the identity on V × 0 and which sends the contact vector field v|C

to the vector field ∂t .

Proof. It is sufficient to construct a complete contact vector field ve on V × [0, 1)


which coincides with v on C and whose trajectories intersecting V × 0 fill the whole
manifold V × [0, 1). We will construct it using the following inductive process. Take

S
a sequence of compact sets C0 = C, C1 , . . . , which exhausts Y , i.e. Cj = Y and
0
Cj ⊂ Int Cj , j = 0, 1, . . . . Let v1 be a contact vector field obtained by cutting off
v outside C0 but inside C1 . Let h1 be the time T1 > 1 flow map of v1 , where T1
is chosen sufficiently large to ensure that h1 (V × 0) ⊂ C1 \ C0 . Denote by C e0 the
domain bounded by V × 0 and h1 (V × 0) and by ve1 the contact vector field equal to
v1 on Ce0 and to the push-forward vector field (h1 )∗ v on Y \ C e0 . Let v2 be a contact
2 Weinstein hypersurfaces and Weinstein pairs 9

vector field obtained by cutting off ve1 outside C1 but inside C2 and denote by h2
the time T2 > T1 + 1 flow of v2 , where T2 is chosen such that h2 (V × 0) ⊂ C2 \ C1 .
Denote by C e1 the domain bounded by V × 0 and h2 (V × 0) and by ve2 the contact
e1 and to the push-forward vector field (h1 )∗ ve1 on Y \ C
vector field equal to v2 on C e1 .
Continuing this process we construct a sequence of contact vector fields ve1 , ve2 , . . . ,
which stabilize on compact sets C1 , C2 , . . . and converge to the contact vector field
ve on Y with the required properties.

Proof of Proposition 2.5. The contact vector field v = −ZΣ − u ∂u is transverse to
∂U (σ) and retracts U (Σ) to Core(Σ, λΣ ), and hence the contact structure on U (Σ) \
Core(Σ, λΣ ) is canonically isomorphic to ∂U (Σ) × [0, ∞) endowed with the invariant
extension ξb of the germ of contact structure
Sξ along ∂U (Σ). On the other hand, v is
transverse to ∂Uδ (σ) for each δ ≤ ε and ∂Uδ (Σ) = U (Σ) \ Σ. Hence, applying
δ∈(0,ε]

Lemma 2.6 we conclude that (U (Σ) \ Σ, ξ) is contactomorphic to (∂U (Σ) × [0, ∞), ξ),
b
and the claim follows.

Remark 2.7. One of the corollaries of Lemma 2.6 is that any open domain in the
2n+1
P
standard contact (R , dz + Pxi dyi − yi dxi ) which is star-shaped with respect to

the contact vector field 2z ∂z + xi ∂x∂ i + yi ∂y∂ i is contactomorphic to R2n+1 . On the
other hand, in the standard contact R3 any open domain diffeomorphic to R3 is
contactomorphic to R3 , see [17].

Problem 2.8. Is there a domain in the standard contact R2n+1 , n > 2, which is
diffeomorphic to the closed ball, has convex in contact sense boundary, but whose
interior is not contactomorphic to the standard R2n+1 ? Or even are there any open
domains in the standard contact R2n+1 , n > 2, which are diffeomorphic but not
contactomorphic to R2n+1 ?

Weinstein pairs
A Weinstein pair (W, Σ) consists of a Weinstein domain W = (X, λ, φ) together with
a Weinstein hypersurface (Σ, λ|Σ ) in its boundary ∂X. Equivalently, a Weinstein pair
can be viewed as a Weinstein manifold with cylindrical end, together with a Weinstein
hypersurface in its ideal contact boundary.
2 Weinstein hypersurfaces and Weinstein pairs 10

Let Λ = Core(Σ) be the skeleton of Σ and


[
Λ
b := Z −t (Λ)
t≥0

be its saturation by the trajectories of the Liouville vector field Z. The union

Core(X, Σ) := Core(X) ∪ Λ
b

is called the core, or the skeleton of the Weinstein pair.


It turns out that it is possible to modify the Liouville form λ on X in a neighborhood
of Σ in X to make the attractor of the modified Liouville vector field equal to the
skeleton Core(X, Σ).
Given a Weinstein pair (W, Σ), W = (X, ω, λ, Z, φ), let U = U (Σ) ⊂ ∂X be its
contact surrounding. Denote by ZΣ the Liouville field dual to λ|Σ and by φΣ its
Lyapunov function. A Liouville form λ0 , the corresponding Liouville vector field Z0
for ω on X and a smooth function φ0 : X → R are called adjusted to the structure
of the pair if (see Fig. 2.1)

• Z0 is tangent to ∂X on U (Σ) and transverse to ∂X elsewhere;



• Z0 |U (Σ) = ZΣ + u ∂u ;
T −t
• the attractor Z0 (X) of the Liouville vector field −Z0 coincides with the
t≥0
core Core(X, Σ) of the Weinstein pair;

• the function φ0 : X → R is Lyapunov for Z0 and such that φ0 |U (Σ) = φΣ + u2


and φ0 has no critical values ≥ ε2 = φ0 |∂U (Σ) .

Proposition 2.9. Given a Weinstein pair (W, Σ), W = (X, λ, φ), there exist a
Liouville form λ0 for ω and a function φ0 : X → R such that

• λ0 , φ0 are adjusted to (W, Σ);

• λ0 coincides with λ outside a neighborhood of Σ;

Moreover, there exists an extension λ, e ⊃X


e φe of (λ0 , φ0 ) to a slightly bigger domain X
such that the W
f := (X,e λ,
e φ)
e is a Weinstein domain and Core(W) f = Core(W, Σ).
2 Weinstein hypersurfaces and Weinstein pairs 11

To construct the adjusted Liouville field Z0 let us write the form λ near ∂X as
s(du + λΣ ) near U (Σ). Note that the Hamiltonian vector field Y for a function
∂ ∂
su near U (Σ) coincides with −s ∂s + u ∂u + ZΣ , and hence by appropriately cutting
off the function su outside a neighborhood of U (Σ) and subtracting the differential
dg of the resulting function g to the Liouville form λ we get the Liouville form λ0
with the required properties. Note that the form λ0 |U is no more contact. Instead,
λ0 |U = π ∗ (λ|Σ ).
Suppose that λ0 , φ0 are adjusted to the Weinstein pair (W, Σ). Recall that φ0 |∂U (Σ) =
ε2 . Denote X0 = {φ0 ≤ ε2 }. We note that φ0 has no critical points in X \ Int X0 ,
and hence X0 is a manifold with boundary with a corner along ∂U (Σ) which is
homeomorphic to X. We will sometime refer to (X0 , λ0 , φ0 ) as the cornered version
of the Weinstein pair (W, Σ).4 For instance, the cornered version of the standard
Weinstein ball B 2n is the cotangent ball bundle of Dn . Thus, it is always possible
to go back and forth between the original and adjusted (cornered) versions of a
Weinstein pair, and we will be using the term “Weinstein pair” for both versions.

U(Σ) U(Σ) U(Σ)

Ζ Σ Ζ0 Σ Ζ− Σ

Fig. 2.1: Modifications of a Weinstein pair structure.

Remark 2.10. There are several other useful adjustments of a Weinstein pair struc-
ture. Ekholm and Lekili in [14], Section B.3, are doing a similar to the cornered
version construction by deforming the boundary ∂X near U (Σ) without changing Z,
as on Fig. 2.2. Without defining here Sylvan’s stop structure we just say that for a
given Weinstein pair there is a contractible space of choices of stop structures on the
completion.
One can also transform a Weinstein pair into a Weinstein cobordism whose negative
boundary is U (Σ), see Fig. 2.1:
4
The completion of the cornered version of a Weinstein pair is a special case of a Liouville sector
in the sense of [26].
3 Operations on Weinstein pairs 12

Let (X0 , λ0 , Z0 , φ0 ) be the cornered adjusted version of a Liouville pair structure


(W, Σ), as in Proposition 2.9. There exists a Liouville form λ− on X0 such that
(X0 , λ− , φ0 ) is a sutured Weinstein cobordism structure with ∂− X0 = U (Σ), and
Core(X0 , λ− , φ0 ) = Core(W, Σ).
To obtain such a form λ− one subtracts from λ the differential of the appropriately
cut off function 2su instead of the function su used to modify λ into λ0 .

Fig. 2.2: Ekholm-Lekili deformation of ∂X.

3 Operations on Weinstein pairs

3.1 Splitting and gluing of Weinstein pairs


Let W = (X, λ, Z, φ) be a Weinstein domain. A hypersurface (P, ∂P ) ⊂ (X, ∂X) is
called splitting for W if it satisfies the following conditions:

- ∂P splits the boundary ∂X into two parts, ∂X = Y− ∪ Y+ with ∂Y− = ∂Y+ =


Y+ ∩ Y− = ∂P (and respectively, P divides X into two parts X+ and X− with
∂X− = P ∪ Y− , ∂X+ = P ∪ Y+ and X+ ∩ X− = P ;

- the Liouville vector field Z is tangent to P ;

- there exists a hypersurface (S, ∂S) ⊂ (P, ∂P ) which is Weinstein for the restricted
Liouville form λ|S , tangent to the vector field Z and intersects all leaves of
the characteristic foliation F of the hypersurface P ; we will refer to S as the
Weinstein soul of the splitting hypersurface P and denote it by Soul(P ).
3 Operations on Weinstein pairs 13

Note that the latter condition together with Lemma 2.6 imply that P is contacto-
morphic to the contact surrounding of its Weinstein soul.
It follows that (S, λ|S , φ|S ; ∂S) is a codimension two Weinstein subdomain of X and
Core(S, λ|S , φ|S ) = Core(W) ∩ P. Moreover, (W± ; S), where W± := (X± , λ|X± , φ|X± )
are cornered Weinstein pairs and Core(W± ; S) = Core(W) ∩ X± .
The gluing construction reverses the splitting. This operation was considered by
Avdek in [3] in the context of Liouville hypersurfaces. Let (W, Σ) and (W0 , Σ0 )
be two Weinstein pairs and (X0 , λ0 , φ0 ), (X00 , λ00 , φ00 ) their cornered forms. Let F :
(Σ, λ|Σ , φ|Σ ) → (Σ0 , λ0 |Σ0 , φ0 |Σ0 ) be a Weinstein isomorphism. We extend F to a
contactomorphism U (Σ) → U (Σ0 ), still denoted by F , and use it to define a domain

X t X 0 := X0 t X00 /{(x ∈ U (Σ)) ∼ (F (x) ∈ U (Σ0 )).


F

Then the Liouville forms λ0 and λ00 , as well as Lyapunov functions φ0 : X0 → R and
φ00 : X00 → R, can be glued together to define a Weinstein structure (W, Σ) ∪(W0 , Σ0 ) :=
F
(XF , λF , φF ), see Fig. 3.1.

F
X Σ Σ X U( Σ ) =F(U( Σ))

XU X
F

Fig. 3.1: Gluing of Weinstein pairs.

Note that
Core(XF , λF , φF ) = Core(X, Σ) ∪ Core(X0 , Σ0 ).
F|Core(Σ)

Note that the constructed Weinstein domain XF contains U (Σ) as its splitting hy-
persurface. Applying the above described splitting construction we get back the
Weinstein pairs (W, Σ) and (W0 , Σ0 ).
3 Operations on Weinstein pairs 14

The gluing of Weinstein pairs is a generalization of the Legendrian surgery con-


struction (or rather Weinstein handle attachment). When Σ = Σ(Λ) for a Legen-
drian Λ ⊂ ∂X, X 0 = B 2n and Σ0 = Σ(Λ0 ), where Λ0 is the Legendrian unknot in
S 2n−1 = ∂B, then (XF , λF , φF ) is the Weinstein n-handle attachment to X along Λ.
Conversely, the general gluing operation (W, Σ) ∪(W0 , Σ0 ) can be decomposed into
F
a sequence of subcritical and critical handle attachments. To do that, one fixes first
a Weinstein handle decomposition of Σ, and then for each handle of index k of this
decomposition one needs to attach a handle of index k + 1 to the glued domains. For
instance, for a handle of index 0 centered at a point p ∈ Σ one attaches a handle of
index 1 along an arc connecting the point p ∈ Σ with its image p0 = F (p) ∈ Σ0 under
the gluing map.
Both, splitting and gluing constructions can be naturally generalized to the rela-
tive setting. Let (W, Σ) and (W0 , Σ0 ) be two Weinstein pairs. Suppose that Σ
and Σ0 are split by splitting hypersurfaces T ⊂ Σ and T 0 ⊂ Σ0 as Σ = Σ− ∪ Σ+
and Σ0 = Σ0− ∪ Σ0+ and we are given a Weinstein isomorphism F : Σ+ → Σ0− .
Then the result of the partial gluing is the pair (W, Σ) ∪ 0 (W0 , Σ0 ) which con-
F,T,T
sists of the Weinstein domain (W, Σ− ) ∪(W0 , Σ0+ ) together with the Weinstein hy-
F
persurface Σ− ∪ Σ0+ ⊂ X ∪ X 0 which is the result of gluing the Weinstein pairs
F |Soul(T ) F
(Σ+ , Soul(T )) and (Σ− , Soul(T 0 ))
0
using the Weinstein isomorphism F |Soul(T ) .
The reverse operation to the partial gluing of Weinstein pairs is a splitting of a
Weinstein pair (W, Σ), W = (X, λ, φ), along a splitting hypersurface (P, Q := ∂P ) ⊂
(X, ∂X) for the Weinstein domain X where in addition P satisfies the following
condition:

- Q intersects Σ transversely, Q ∩ Σ = Soul(Q) and Q ∩ Σ is a splitting hypersurface


for Σ, which splits it into Σ+ and Σ− ;

The result of this splitting are two Weinstein pairs (X− , Σ


e − ) and (X+ , Σ
e + ), where
the Weinstein hypersurface Σ± ⊂ ∂X± = Y± ∪ P is the result of gluing of Weinstein
e
pairs (Σ± , Soul(Q ∩ Σ)) and (Soul(P ), Soul(Q ∩ Σ)).
As in the absolute case, the gluing operation of Weinstein pairs glues their skeleta
along the skeleta of glued hypersurfaces. Conversely, a splitting of the skeleton of a
Weinstein domain lifts to a splitting of a Weinstein domain into two Weinstein pairs.
3 Operations on Weinstein pairs 15

3.2 Product and Stabilization of Weinstein pairs


Given two Weinstein pairs (W, Σ) and (W0 , Σ0 ), where W = (X, λ, φ), W0 = (X 0 , λ0 , φ0 )
we define their product as the Weinstein pair
(W, Σ) × (W0 , Σ0 ) := (X × X 0 , λ ⊕ λ0 , (Σ × X 0 ; Σ × Σ0 ) t(X × Σ0 , Σ × Σ0 )).
Id

Here (Σ × X 0 ; Σ × Σ0 ) t(X × Σ0 , Σ × Σ0 )) is the result of gluing of two Weinstein pairs


Id
by the identity map between the Weinstein hypersurfaces Σ × Σ0 ⊂ ∂(X × Σ0 ) and
Σ × Σ0 ⊂ ∂(Σ × X 0 ). We note that
Core ((W, Σ) × (W0 , Σ0 )) = Core(W, Σ) × Core(W0 , Σ0 ).

In the case when (X 0 , Σ0 ) is the Weinstein pair (T ∗ Dk , T ∗ S k−1 ) the product operation
is called the stabilization (or k-stabilization). It was first proposed in a slightly
different form by M. Kontsevich, [28]. The core of the k-stabilized pair (W, Σ) is
equal to Core(W, Σ) × Dk .
It is important to stress the point that the result of the stabilization is always a
Weinstein pair with a non-empty hypersurface in the boundary, even if we begin
with the absolute case of a Weinstein domain.

3.3 Weinstein homotopy as a Weinstein pair


Consider a Weinstein structure W0 := (X, ω, λ0 , φ0 ) and its 1-stabilization Wst :=
W × T ∗ I, viewed as a Weinstein pair (X × T ∗ I, λ0 + udt, X × 0 ∪ X × 1). Consider a
Weinstein homotopy Wt := (X, λt = λ0 +dht , φt ), t ∈ [0, 1]. We assume, in addition,
that ḣ1 = ḣ0 = 0, where we denoted ḣt := dh dt
t
(t). This condition can always be
arranged by a re-parameterization of the homotopy. Consider the product X × T ∗ I
with the symplectic form Ω := ω ⊕ du ∧ dt , where (u, t) are canonical coordinates on
T ∗ I (so that u = 0 defines the 0-section). Note that the 1-form λ e := λt + (u + ḣt )dt is
a Liouville form for Ω. Indeed, dλ e := dλt + dt ∧ dḣt + dḣt ∧ dt + du ∧ dt = ω + du ∧ dt.
We have λ| e X = λ0 and λ|
0
e X = λ1 .
1

Proposition 3.1. There exists a function φe = X × T ∗ I → R such that


 
(X × T ∗ I, λ e X0 ∪ X1
e := λt + (u + ḣt )dt, φ;

where X0 := X × {t = u = 0}, ; X1 := X × {t = 1, u = 0},


is a Weinstein pair.
4 Looseness and Flexibility 16

We call this pair the concordance generated by the homotopy Wt .

Proof. Note that the corresponding to λ e Liouville vector field is given by the formula

Ze = Zt + (u + ḣt ) ∂u , where Zt is the Liouville vector field corresponding to λt . Define
the function φe by the formula φe = φt + k2 (u + ḣt )2 , where a positive constant k will
be chosen later. Then we have

dφ( e = dφt (Zt ) + k(u + ḣt )2 + k(u + ḣt )dḣt (Zt ).


e Z)

Not that |dφt (Zt ) ≥ a||Zt ||2 and |dḣt (Zt )| ≤ b||Zt || for some constants a, b > 0.
Denoting X := ||Zt ||, Y := u + ḣt we can write

|dφ( e ≥ a||Zt ||2 + k(u + ḣt )2 − bk|u + ḣt |||Zt ||


e Z)|
aX 2 + kY 2 − bkXY.

The quadratic form aX 2 + kY 2 − bkXY is positive definite if b2 k 2 − 4ak < 0 or


k < 4ab2
. Under this condition, which can be arranged by choosing the constant k
sufficiently small, we get |dφ( e ≥ c(X 2 + Y 2 ) ≥ e
e Z)| e 2 for positive constants c, e
c||Z|| c.
This concludes the proof.
Remark 3.2. The critical point locus of φe (= the zero locus of Z)
e is equal to

e = {(x, t, u); x is a critical point of φt , u = ḣt (x), t ∈ [0, 1]}.


C

The stable manifold of a critical point (x0 , t0 , u0 ) projects to the stable manifold of the
critical point x0 of φt0 . Its u-coordinate can be found by solving the inhomogeneous
linear ODE
du(γ(s))
= u(γ(s)) + ḣt0 (γ(s))
ds
with the asymptotic boundary condition lim u(γ(s)) = u0 , where γ(s) is a trajectory
s→∞
of Xt0 converging to the critical point x0 .

4 Looseness and Flexibility

Let us recall that in contact manifolds of dimension 2n − 1 ≥ 5 there is a local


modification construction for Legendrian submanifolds, called stabilization5 , see
[16, 35, 11]. This operation can be performed in an arbitrarily small neighborhood
5
The term “stabilization” is used here in a completely different sense than in Section 3.2.
4 Looseness and Flexibility 17

of any point of a Legendrian. Moreover, it can also be performed without changing


the formal Legendrian isotopy class of the Legendrian submanifold. In her 2012
paper [35] Emmy Murphy called a Legendrian submanifold loose if it is isotopic to a
stabilization of another Legendrian submanifold, and showed that loose Legendrians
satisfy an h-principle: any two loose formally isotopic Legendrians can be connected
by a Legendrian isotopy.
The notion of flexibility, see [11], for Weinstein cobordisms is tightly related to
the looseness property of Legendrian knots. One first defines flexibility for ele-
mentary Weinstein cobordisms, i.e. Weinstein cobordisms (W, ω, Z, φ) without any
Z-trajectories connecting critical points of the Lyapunov function φ. An elementary
2n-dimensional, n > 2, Weinstein cobordism (W, ω, X, φ) is called flexible if the at-
taching spheres of all index n handles form in ∂− W a loose Legendrian link (i.e. each
sphere is loose in the complement of the others). A Weinstein structure is called
flexible if it is homotopic to one which can be decomposed into elementary flexible
cobordisms.
As it was shown by E. Murphy and K. Siegel in [37] existence of a decomposition into
flexible elementary cobordisms really depends on the choice of a particular Weinstein
structure in the given homotopy class. Moreover, there exist non-flexible Weinstein
domains which become flexible after attaching an n-handle.
Flexible Weinstein structures are indeed flexible: they abide a number of h-principles.
Theorem 4.1. (i) ([11]) Any two flexible Weinstein structures on a given smooth
cobordism are homotopic as Weinstein structures provided that the correspond-
ing symplectic forms are in the same homotopy class of non-degenerate (but
not necessarily closed) 2-forms.
(ii) ([11]) Let (X, ω, Z, φ) be any flexible Weinstein structure and φt , t ∈ [0, 1], be
a family of generalized Morse functions such that φ0 = φ. Then there exists a
homotopy (X, ωt , Zt , φt ) of Weinstein structures.
(iii) ([22]) Let (X± , ω± , Z± , φ± ) be two Weinstein structures. Suppose that the struc-
ture (X− , ω− , Z− , φ− ) is flexible and that there exists an embedding f : X− →
X+ such that the forms ω− and f ∗ ω+ are homotopic as non-degenerate (but
not necessarily closed) 2-forms. Then there exists a homotopy of Weinstein
t
structures (X− , ω− , Z−t , φt− ), t ∈ [0, 1], beginning with (X− , ω−0
= ω− , Z−0 =
0 t 0
Z− , φ− = φ− ) and an isotopy f : X− → X+ beginning with f = f such that
(f 1 )∗ ω+ = ω−
1
.

At first glance Theorem 4.1 implies that symplectic topology of flexible Weinstein
4 Looseness and Flexibility 18

manifolds is quite boring. This is also confirmed by the fact that symplectic homology
in all its flavors of a flexible Weinstein manifold is trivial. However, as we will see
below in Section 7 the contact boundaries of flexible Weinstein domains have a rich
contact topology.
The looseness property of a Legendrian submanifold can be naturally extended to
Weinstein hypersurfaces of contact manifolds. A Weinstein hypersurface Σ of a
contact manifold Y of dimension 2n + 1 ≥ 5 is called loose if for each n-dimensional
strata S of the skeleton Core(Σ) there is a ball BS ⊂ Y \ (Core(Σ) \ S) such that
BS ∩ S is loose in BS relative ∂(BS ∩ S). A canonical Weinstein thickening of a loose
Legendrian knot is loose. However, it is unclear whether looseness is preserved under
Weinstein isotopy.

Problem 4.2. Is looseness property preserved under a Weinstein isotopy of Σ. In


particular, suppose that a Weinstein thickening Σ(Λ) of a Legendrian knot Λ is iso-
topic in the class of Weinstein hypersurfaces to a loose Weinstein hypersurface. Does
this imply that Λ itself is loose?

Proposition 4.3. Let (W, Σ) and (W0 , Σ0 ) where W = (X, λ, φ), W0 = (X 0 , λ0 , φ0 ),


be two Weinstein pairs and

(W, Σ) × (W0 , Σ0 ) := (X × X 0 , λ ⊕ λ0 , Σ
e := (Σ × X 0 ; Σ × Σ0 ) t(X × Σ0 , Σ × Σ0 ))
Id

e is loose in ∂(X × X 0 ).
be their product. Suppose that Σ is loose in ∂X. Then Σ

Indeed, this is straightforward from the following fact: given any contact manifold
(Y, {α = 0}), a Liouville manifold (U, µ), a loose Legendrian Λ ⊂ Y and a Lagrangian
L ⊂ U with µ|L = 0, then the Legendrian Λ × L ⊂ (Y × U, {α ⊕ µ = 0}) is loose as
well.
Let us stress the point that while flexibility of a Weinstein manifold is its intrinsic
property, the looseness of a Weinstein hypersurface depends on its embedding in the
contact manifold. However, the above fact about the looseness of a product shows
that flexibility always implies looseness (I thank the referee for this argiment).

Proposition 4.4. Let (Y, ξ) be a contact manifold of dimension ≥ 7, and Σ ⊂ Y a


flexible Weinstein hypersurface. Then Σ is loose.

Indeed, let α be a contact form for ξ which restricts to a Liouville form µ on Σ.


Consider a Weinstein subdomain Σ0 ⊂ Σ and let a Lagrangian disc ∆ ⊂ Σ \ Σ0 be
attached to Σ0 along a loose Legendrian sphere Λ := ∂∆ ⊂ ∂Σ0 . In a neighborhood
4 Looseness and Flexibility 19

U ⊃ ∂Σ0 in Σ the Liouville form µ can be written as sβ, s ∈ (1 − ε, 1 + ε) for a


contact form on ∂Σ0 , and on a neighborhood U e of U in Y the contact form α can
be written as dt + sβ = s(udt + β), |t| < ε, u = 1s . Hence, U
e can be viewed as the
product of the contact manifold (∂Σ0 , β) and a Liouville subdomain
 
1 1
Q := {(u, t) ∈ − , × (−ε, ε)} ⊂ (R2 , udt),
1+ε 1−ε

while ∆ ∩ Ue = Λ × {t = 0, 1 < u ≤ 1]} ⊂ Σ0 × Q. Hence looseness of attaching


1−ε
spheres of top index Weinstein handles of Σ implies looseness of their Lagrangian
cores viewed as Legendrian submanifolds of Y .
The notion of flexibility naturally extends to Weinstein pairs. A Weinstein pair
(W = (X, λ, φ), Σ) is called flexible if it is flexible viewed as a cobordism between
∂X− = U (Σ) and ∂+ X = X \ Int U (Σ), see Remark 2.10. It is straightforward to
see that flexibility is preserved under the stabilization construction. However, the
converse is not clear.

Problem 4.5. Suppose that the stabilization of a Weinstein pair is flexible. Does this
imply that the Weinstein pair itself is flexible? More generally, does existence of a
homotopy between stabilizations of two Weinstein (pair) structures implies existence
of a homotopy between the structures themselves?

Attaching a critical handle along a loose Legendrian knot to a flexible Weinstein


domain by definition preserves its flexibility. This generalizes to the following

Proposition 4.6. Let (W = (X, λ, φ), Σ) and (W0 = (X 0 , λ0 , φ0 ), Σ0 ) be two Wein-


stein pairs. Let Σ and Σ0 be decomposed as Σ = Σ− ∪ Σ+ , Σ0 = Σ0− ∪ Σ0+ by splitting
hypersurfaces T ⊂ Σ and T 0 ⊂ Σ0 , see Section 3.1. Suppose that

- there exists a Weinstein isomorphism F : Σ+ → Σ0− ,

- Σ− is loose in ∂X and

- pairs (W, Σ− ) and (W0 , Σ0+ ) are flexible.

Then the glued pair (W, Σ) ∪ 0 (W0 , Σ0 ) is flexible. In particular, the result of gluing
F,T,T
of two flexible Weinstein domains along Weinstein hypersurfaces one of which is
loose is flexible.
5 Lagrangian submanifolds of Weinstein domains 20

This follows from the fact that the gluing operations of two Weinstein pairs can be
decomposed into a sequence of handle attachments, and the looseness assumption
for the Weinstein hypersurface in one of the glued parts implies that all the critical
handles are attached along loose knots.
As a corollary Proposition 4.6 implies the following generalization of the following
result of E. Murphy and K. Siegel, [37]:
Proposition 4.7. The product of two Weinstein pairs, one of which is flexible, is
flexible.

Indeed, the product of two Weinstein pairs can always be built by a sequence of
gluing of various stabilizations of the first pair.

5 Lagrangian submanifolds of Weinstein domains

In this section we discuss exact Lagrangian submanifolds in a Weinstein domain


(X, λ, φ). The Lagrangians will always be assumed either closed or with Legendrian
boundary in ∂L ⊂ ∂X.
Let Σ(∂L) be the Weinstein thickening of the (possibly empty) Legendrian boundary
∂L. A Lagrangian L is called regular, see [18], if the Weinstein pair (X, Σ(∂L)) admits
a skeleton which contains L.
Problem 5.1. Are there non-regular exact Lagrangians?

The problem is widely open. While no examples of non-regular Lagrangians are


known, in the opposite direction in the case of a closed exact Lagrangian L in a gen-
eral Weinstein domain X it is even unknown whether L realizes a non-zero homology
class in Hn (X) (which is a necessary condition for its regularity).
If L ⊂ X is regular then by removing its tubular neighborhood N (L) one gets a
Weinstein cobordism XL := (W \N (L), ∂− XL := ∂N (L) \ ∂X, ∂+ XL := ∂X \ N (L))
(between manifolds with boundary if ∂L 6= ∅) whose negative boundary is the unit
cotangent bundle of L. The Lagrangian L is called flexible, see [18]), if the cobordism
XL is flexible.
It was shown in [18] that any flexible (X, λ) admits a surprising abundance of flexible
Lagrangians with non-empty Legendrian boundary. In particular,
Theorem 5.2. Let L be an n-manifold with non-empty boundary, equipped with a
fixed trivialization η of its complexified tangent bundle T L ⊗ C. Then there exists
6 Symplectic topology of Weinstein manifolds 21

a flexible Lagrangian embedding with Legendrian boundary (L, ∂L) → (B 2n , ∂B 2n )


where B 2n is the standard symplectic 2n-ball, realizing the trivialization η. In partic-
ular, any 3-manifold with boundary can be realized as a flexible Lagrangian subma-
nifold of B 6 with Legendrian boundary in ∂B 6 .

6 Symplectic topology of Weinstein manifolds

While flexible Weinstein structures enjoy a full parametric h-principle, there is plenty
of symplectic rigidity and fine symplectic invariants of non-flexible ones. I will not
discuss in this survey any such invariants and just mention that until recently most
examples of formally homotopic but not symplectomorphic Weinstein manifolds were
distinguished by their (possibly appropriately deformed) symplectic cohomology. For
instance, there are infinitely many non-symplectomorphic Weinstein structures on
R2n for any n > 2 ([41, 33]) and by taking connected sums of these examples with
flexible Weinstein manifolds one gets infinitely many non-symplectomorphic Wein-
stein structures on any given “almost Weinstein” (i.e. an almost complex manifold
of homotopy type of a half-dimensional CW-complex) manifolds, see [2].
Note that Theorem 5.2 can also be used for constructing exotic Weinstein structures.
In particular,
Theorem 6.1 ([18]). Let L be a closed 3-manifold. Then there exists a unique
up so symplectomorphism Weinstein structure W(L) = (ωL , ZL , φL ) on T ∗ S 3 which
contains L as its flexible Lagrangian submanifold in the homology class of the 0-
section (with Z/2-coefficients in the non-orientable case). Moreover, infinitely many
of these W(L) are pairwise non-symplectomorphic.

Note that there exists only 1 homotopy class of almost complex structures on T ∗ S 3 .
While the symplectic structure of W(L) carries a lot of information about the topol-
ogy of L, the following problem is open:
Problem 6.2. Suppose W(L) is symplectomorphic to W(L0 )? Does it imply that L
is diffeomorphic to L0 ?

The famous ”nearby Lagrangian problem” asks whether there is a unique up to


Hamiltonian isotopy exact closed Lagrangian submanifold in the standard T ∗ M for
a closed M . Though in this form the answer is unknown except for M = S 2 and T 2 ,
see [27, 12], the answer is positive up to simple homotopy equivalence, [1], and hence
according to Smale, Freedman and Perelman for M = S n up to homeomorphism,
7 Topology of Weinstein fillings 22

and for some dimensions, e.g. n = 3, 5, 6, 12, even up to diffeomorphism, [34]. As


it was pointed out to me by O. Lazarev, one can show using methods of [9] that
certain exotic T ∗ S n may contain several not homotopy equivalent regular closed
exact Lagrangian submanifolds.

Problem 6.3. Can the uniquenes results from [1] be extended to a more general
class of Weinstein structures on T ∗ S n ?

The proof of Theorem 5.2 yields also the following slightly stronger result.

Theorem 6.4. Let (X, ω, λ, φ) be a 6-dimensional Weinstein domain such that φ


has exactly 1 critical point of index 3 (and any number of critical points of smaller
indices). Suppose also that the symplectic vector bundle (T X, dλ) is trivial. Then
there exists a Weinstein structure (ωX , λX , φX ) on T ∗ S 3 which admits an embedding

(X, ω, λ, φ) → (T ∗ S 3 , ωX , λX , φX )

onto a Weinstein subdomain with a flexible complement.

7 Topology of Weinstein fillings

Contact manifolds appeared as boundaries of Weinstein domains are called Weinstein


fillable. The fact that a Weinstein filling has a homotopy type of a half-dimensional
CW-complex imposes constraints on the topology of its contact boundary and the
stable almost complex class which can be realized by Weinstein fillable contact struc-
tures on a given smooth manifold. This question was studied in detail by Bowden-
Crowley-Stipsicz in [5, 6]. In particular, they showed that there are classes of homo-
topy spheres which do not admit any Weinstein fillable contact structure.
Given a contact manifold (Y, ξ) one can try to describe (symplectic) topology of its
Weinstein fillings. In this section we discuss this problem for contact manifolds of
dimension 2n − 1 > 3, see [40] for a survey of results for 3-dimensional manifolds.
First of all notice that the fact that X retracts to its n-dimensional skeleton implies
that the inclusion Y = ∂X ,→ X is (n − 1)-connected, and in particular, if Y is
a homotopy sphere then X is (n − 1)-connected. It turns out that some contact
structures know much more about the topology of their fillings.

Theorem 7.1 ([31]). Any Weinstein filling of the standard contact sphere (S 2n−1 , ξstd )
is diffeomorphic to the ball B 2n .
7 Topology of Weinstein fillings 23

Generalizing Theorem 7.1 K. Barth, H. Geiges and K. Zehmisch proved in [7]:


Theorem 7.2. All Weinstein fillings of a simply connected contact manifold admit-
ting a subcritical filling are diffeomorphic.

In fact, both Theorems 7.1 and 7.2 hold in a stronger form for a more general class
of symplectic, and not necessarily Weinstein fillings. We also note that while it fol-
lows from Theorem 4.1 that all completed subcritical Weinstein fillings of a given
contact manifold are symplectomorphic (we note that the (n − 1)-connectedness of
the inclusion map ∂X ,→ X implies that the homotopy class of an almost complex
structure on a subcritical manifold is determined by the homotopy class of its re-
striction to the boundary), it is unknown for n > 2 whether all completed fillings of
a contact manifold admitting a subcritical filling (e.g. the standard contact sphere)
are symplectomorphic.
The following theorem of Oleg Lazarev constrains topology of flexible Weinstein
manifolds.
Theorem 7.3 ([29]). All flexible fillings of of a contact manifold (Y, ξ) with c1 (Y, ξ) =
0 have canonically isomorphic integral homology.

In particular, as Lazarev observed, Theorem 7.3 together Smale’s classification of


2-connected 6-manifolds from [43] and the fact that π3 (O/U ) = 0 yield a complete
classification of flexibly fillable contact structures on S 5 .
Corollary 7.4 ([29]). There exists a sequence ξn , n = 0, 1, . . . , of pairwise non-
contactomorphic contact structures on S 5 such that

• any flexibly fillable contact structure on S 5 is contactomorphic to one of the


structures from this sequence;

• the contact structure ξ0 is standard;

• for n ≥ 1 the contact sphere (S 5 , ξn ) admits a unique up to symplectomorphism


n
flexible Weinstein filling diffeomorphic to (# S 3 × S 3 ) \ B 6 .
1

There are further constraints on the topology of flexible Weinstein fillings. In par-
ticular,
Theorem 7.5 ([19]). Let (S 4n−1 , ξ) be a flexibly fillable contact structure. Then the
signature of its flexible filling is uniquely determined by the contact structure ξ.
8 Nadler’s program of arborealization 24

Problem 7.6. Does a contact structure (Y, ξ) remember


a) the diffeomorphism type of its flexible Weinstein filling (X, ω, Z, φ)?
b) the almost symplectic homotopy class [ω] of the symplectic structure ω?

We note that the diffeomorphism type of X together with the homotopy class [ω]
determine a flexible Weinstein structure up to Weinstein homotopy, and hence the
positive answer to a) and b) would imply that the contact structure (Y, ξ) remember
the symplectomorphism type of the completion of its flexible filling.

8 Nadler’s program of arborealization

A priori, a skeleton of a Weinstein domain can have very complicated singularities.


However, David Nadler conjectured that up to Weinstein homotopy the singularities
of the skeleton can be reduced to a finite list in any dimension, see [38]. For 2n-
dimensional symplectic Weinstein manifolds the list of Nadler’s singularities, which
he calls arboreal, are enumerated by decorated rooted trees with ≤ n + 1 vertices. It
is remarkable that the singularity of each given type has a unique symplectic realiza-
tion. Nadler also proposed in [39] a procedure for arborealization of the skeleton of a
Weinstein structure. His procedure replaces a given Weinstein structure by another
one whih an arboreal skeleton. Nadler proved in [39] that the constructed Weinstein
manifold has microlocal sheaf-theoretic invariants equivalent to those of the Wein-
stein manifold. Conjecturally this implies that the wrapped Fukaya categories are
also the same for the original and modified Weinstein manifold. However, it is un-
clear whether Nadler’s modification yields a Weinstein structure which is homotopic,
or even symplectomorphic to the original one.
In an ongoing joint project [23] with David Nadler and Laura Starkston we are
exploring a somewhat different strategy for arborealization of the Weinstein skeleton
via a Weinstein homotopy using simplification of singularities type technique in the
spirit of a recent paper of D. Álvarez-Gavela, [4]. In some special cases this program
was already carried out by Starkston in [44].
In this section we discuss the arboreal singularities with more detail and give precise
statements of some of the results from [23].
8 Nadler’s program of arborealization 25

8.1 Definition of an arboreal singularity


While we define below arboreal models as closed properly embedded subsets of the
standard symplectic vector space, we are interested only in germs of these models at
the origin.
Consider a tree T with ≤ n + 1 vertices and a fixed vertex R, the root. Suppose in
addition that all edges, except the terminal ones are decorated with ±1. We will
denote by ε the decoration, and by |T | the total number of vertices. With each
decorated rooted tree (T, ε) we associate a unique up to symplectomorphism model
A(T, ε, m) ⊂ R2m = T ∗ Rm in each dimension m ≥ n of the skeleton. The models will
be stratified by strata which are isotropic for the Liouville form pdq. In dimension
m > n we have A(T, ε, m) = A(T, ε, n) × Rm−n ⊂ T ∗ Rn × T ∗ Rm−n = T ∗ Rm .
The model A(T, ε, n) will be defined inductively in n. For a tree T which consists
of one vertex we define A(T, 0) to be a point (in the 0-dimensional symplectic space
T ∗ R0 ), and respectively A(T, n) = Rn ⊂ T ∗ Rn .
As it was already stated above, the Liouville form pdq vanishes on each stratum of
the model A(T, ε, n) ⊂ T ∗ Rn . Hence, if we view T ∗ R2n as a (Weinstein) hypersur-
face {z = 0} in the contact space (R2n+1 = T ∗ Rn × R, pdq + dz), then all strata
of A(T, ε, n) ⊂ T ∗ Rn are also isotropic for the contact form pdq + dz. However,
unless A(T, ε, n) is a Lagrangian plane, the front projection (p, q, z) 7→ (q, z) is very
degenerate, because it collapses the image to the hyperplane {z = 0}. We want to
deform the model A(T, ε, n) in R2n+1 to make the front projection more generic. To
do that, consider a contactomorphism S : R2n+1 → R2n+1 given by the formula
p21
S(p1 , . . . , pn , q1 , q2 , . . . , qn , z) = (p1 , . . . , pn , q1 + p1 , q2 , . . . , qn , z − ).
2
Then
p21
S −1 (p1 , . . . , pn , q1 , q2 , . . . , qn , z) = (p1 , . . . , pn , q1 − p1 , q2 , . . . , qn , z + ).
2
Denote
b− (T, ε, n) := S −1 (A(T, ε, n)).
b+ (T, ε, n) := S(A(T, ε, n)), A
A
The sets A b± (T, ε, n) are stratified by isotropic for the contact form dz + pdq strata.
If |T | = 1 we have A b+ (T, n) = A(T, n).
Suppose that we already defined models for all decorated rooted trees (T, ε) with
|T | ≤ n. Consider a rooted tree (T, ε) with |T | = n + 1. By removing the root
R and all edges adjacent to R we get k decorated trees (T1 , ε), . . . , (Tk , εk ) with
8 Nadler’s program of arborealization 26

+ −

Fig. 8.1: Arboreal singularities labeled by rooted decorated trees. The picture repre-
sents Lagrangian skeleta themselves, and not their front projections. Free
boundaries of vertical strata form Legendrian trees, while their traces at
the horizontal plane are fronts of these trees.

|T1 | = n1 , . . . , |Tk | = nk , n1 + · · · + nk = n. For each of them we choose as its root


the vertex which was connected in T to R. Let σj = ±1 be the decoration of the
edge which was connecting the root R with the root of the tree Tj , j = 1, . . . , k.
Consider already defined models A(T1 , ε1 , n − 1), . . . , A(Tk , εk , n − 1) ⊂ T ∗ Rn−1 × R.
j
P
Denote N0 := 0, Nj := ni , j = 1, . . . , k −1. For each j = 0, . . . , k −1 consider the
i=1
n
hyperplane Πj = {pNj +1 = 1} in R2n = T ∗ Rn with the Liouville form λ =
P
pj dqj .
1
n
pj ∂p∂ j , or equivalently
P
Note that Πj is transverse to the Liouville vector field Z =
1
8 Nadler’s program of arborealization 27

P
λ|Πj = dqNj +1 + pi dqi is a contact form. Cyclically ordering coordinates
i∈{1,...,n},i6=Nj +1
qNj +2 , . . . , qn , q1 , . . . , qNj and taking the coordinate qNj +1 as z we identify Πj with
T ∗ Rn−1 × R. Consider Asign(σj ) (Tj , εj , n − 1) ⊂ Πj .
Denote
k
[
∗ n bsign(σj ) (Tj , εj , n − 1))}.
B(T, ε, n) := {(tp, q) ∈ T R ; t ∈ [0, ∞), (p, q) ∈ A
j=1

Note that B(T, ε, n) ∩ {p = 0} is the union of front projections of Legendrian com-


plexes Absign(σj ) (Tj , εj , n−1)), and B(T, ε, n) is the positive conormal of this stratified
set co-oriented by the vector field ∂qN∂ +1 . Finally, we define
j

A(T, ε, n) := {p = 0} ∪ B(T, ε, n).


Singularities of the form A(T, ε, n) where (T, ε) is a decorated rooted tree are called
primary arboreal.
Note that up to linear symplectomorphism the result of the above construction is in-
dependent of the ordering of the trees T1 , . . . , Tk . Indeed, the corresponding symplec-
tomorphism is the symplectization of the linear automorphism of Rn appropriately
permuting the coordinates q1 , . . . , qn .
As an example, let us explicitly construct the models shown on Fig, 8.1. For a tree
with 2 vertices we take the standard symplectic R2 with coordinates (p, q). Then
Π = {p = 1}. For the 1-vertex tree T1 the model A(T1 , 0) coincides is the point
{p = 1, q = 0} ∈ Π and A(T b 1 , 0) = A(T1 , 0). Hence B(T, 1) = {(t, 0), t ≥ 0} is the
positive p-semi-axis, and A(T, 1) = {p = 0} ∪ B(T, 2), is the union of the coordinate
line q with this semi-axis, as it is shown on the left side of Fig. 8.1.
For the rooted tree with three vertices and the central root, as on the lower picture in
Fig. 8.1, each of the trees T1 , T2 has 1 vertex. Hence, Π1 = {p1 = 1}, Π2 = {p2 = 1},
and identifying this hyperplanes with the standard contact R3 we get A(T1 , 1) =
{p2 = q1 = 0} ⊂ Π1 and A(T2 , 1) = {p1 = q2 = 0} ⊂ Π2 . Therefore,
A(T, 2) = {p = 0} ∪ {p2 = q1 = 0, p1 ≥ 0} ∪ {p1 = q2 = 0, p2 ≥ 0}.

Finally, consider the right models on Fig. 8.1. The models are contained in the
standard symplectic R4 with canonical coordinates (p1 , q1 , p2 , q2 ), and we have Π =
Π1 = {p1 = 1} The tree T1 in this case consists of two vertices, and identifying Π
with the standard symplectic R2 , we find that
b± (T1 , 1) = {q1 = p2 = 0} ∪ {q1 = ∓p22 , p2 = ±q2 , p2 ≥ 0}.
A
8 Nadler’s program of arborealization 28

Note that the second stratum in the union can also be written as {q1 = ∓q22 , p2 =
±q2 , p2 ≥ 0} Thus we have
q22
B(T, +1, 2) = {p2 = 0, q1 = 0, p1 ≥ 0} ∪ {q1 = − , p2 = p1 q2 , p1 , p2 ≥ 0},
2
q22
B(T, −1, 2) = {p2 = 0, q1 = 0, p1 ≥ 0} ∪ {q1 = , p2 = −p1 q2 , p1 , p2 ≥ 0}
2
q2
Note that B(T, ±1, 2) ∩ {p = 0} = {q2 = 0} ∪ {q1 = ∓ 22 } is the front of the
Legendrian tree Ab± (T1 , 1), while B(T, ±1, 2) is the positive conormal of this front
co-oriented by the vector field ∂q∂1 .
A general arboreal singularity is associated to a double decorated rooted tree with
an additional decoration β which assigns 0 or 1 to all terminal vertices of the tree T .
We extend β to all vertices by setting β(v) = 0 for all non-terminal vertices. Primary
arboreal singularities correspond to the case when the decoration β is identically 0.
P
We denote |β| := β(v), where the sum is taken over all terminal vertices v of
the tree T . With each double decorated tree (T, ε, β) we associate a unique up to
symplectomorphism model A(T, ε, β, m) ⊂ T ∗ Rm for each m ≥ |T | + |β| − 1. In
dimension m ≥ n := |T | + |β| − 1 we have
A(T, ε, β, m) = A(T, ε, β, n) × Rm−n ⊂ T ∗ Rn × T ∗ Rm−n = T ∗ Rm .

The model A(T, ε, β, m) ⊂ T ∗ Rm with m = |T | + |β| − 1 is defined by a similar


inductive procedure as for primary arboreal singularities, beginning with
A(T, ε, β, 1) = {p = 0, q ≥ 0} ⊂ T ∗ R for |T | = 1 and |β| = 1.

Every model A(T, ε, β, m) ⊂ T ∗ Rm can be presented as a union of Lagrangian sheets


Lv enumerated by vertices of the graph T . Denote by d(v) the distance between v
and the root. Then Lv is diffeomorphic to the quadrant
{(x1 , . . . , xn ) ∈ Rn ; x1 , . . . xk ≥ 0}, k = d(v) + β(v).

Note that the model A(T, ε, β, m) inherits a smooth structure (i.e. the algebra of
smooth functions) from the ambient space R2n . By an n-dimensional arboreal com-
plex we mean a set covered by charts diffeomorphic to one of the models A(T, ε, β, n).
Hence, every arboreal complex can be canonically stratified by strata ST,ε,β of dimen-
sion n − |T | − |β| + 1. A diffeomorphism f : C → C 0 between two arboreal complexes
induces a diffeomorphism between the corresponding strata, but not every contin-
uous map f : C → C 0 which is a diffeomorphism on the corresponding strata is a
diffeomorphism of arboreal complexes C and C 0 .
8 Nadler’s program of arborealization 29

8.2 Main results


Proposition 8.1 ([23]). For each arboreal complex C there exists a unique up to
symplectomorphism Weinstein domain W(C) = (X, ω, Z, φ), “the cotangent bundle”
of C such that C = Core(X, ω, Z). Any two such Weinstein structures (X, ω, Z, φ)
and (X, ω, Z 0 , φ0 ) are homotopic through a family of Weinstein structures with a fixed
core.

Theorem 8.2 ([23]). (i) Any Weinstein structure is homotopic to a Weinstein


structure with an arboreal skeleton.

(ii) Let Wt , t ∈ [0, 1] be a Weinstein homotopy such W0 and W1 have arboreal


skeleta. Then there exists a Weinstein pair structure (W; W0 ∪W1 ) on X ×T ∗ I
with an arboreal skeleton which is homotopic to the Weinstein pair associated
to the homotopy Wt (see Section 3.3).

Under some topological constraints on the manifold X one can further restrict the
list of necessary singularities.

Theorem 8.3 ([23]). Let W = (X, ω, Z, φ) be a Weinstein structure. Suppose that

a) the manifold X is (n − 2)-connected;

b) there exists a field of Lagrangian planes τ ⊂ T X; in other words, T X with its


homotopically canonical almost complex structure is isomorphic to the complex-
ification of a real n-dimensional vector bundle.

Then the Weinstein structure W is homotopic to a Weinstein structure Wf = (X, ω, Z,


e φ)
e
whose skeleton is an arboreal complex with singularities of type (T, ε, β) where the
distance from the root of the tree T to any other vertex is no more than 2 and the
decoration ε takes only positive values.

References

[1] M. Abouzaid and T. Kragh, Simple homotopy equivalence of nearby Lagrangians,


arXiv:1603.05431.

[2] M. Abouzaid and P. Seidel, Altering symplectic manifolds by homologous recom-


bination, arXiv:1007.3281.
8 Nadler’s program of arborealization 30

[3] R. Avdek, Liouville hypersurfaces and connect sum cobordisms, arXiv:1204.3145.

[4] D. Álvarez-Gavela, The simplification of singularities of Lagrangian and Legen-


drian fronts, arXiv:1605.07259.

[5] J. Bowden, D. Crowley and A. Stipsicz, The topology of Stein fillable manifolds
in high dimensions I, arXiv:arXiv:1306.2746.

[6] J. Bowden, D. Crowley, A. Stipsicz, The topology of Stein fillable manifolds in


high dimensions II, Geom. Topol. 19(2015), 2995–3030.

[7] K. Barth, H. Geiges and K. Zehmisch, The diffeomorphism type of symplectic


fillings, arXiv:1607.03310.

[8] F. Bourgeois, T. Ekholm and Y. Eliashberg, Effect of Legendrian Surgery, Geom.


Topol., 16(2012), 301–389.

[9] C. Cao, N. Gallup, K. Hayden, J. Sabloff, Topologically Distinct Lagrangian and


Symplectic Fillings, arXiv:1307.7998.

[10] K. Cieliebak, Handle attaching in symplectic homology and the Chord Conjec-
ture, J. Eur. Math. Soc.. 4(2002), 115–142.

[11] K. Cieliebak and Y. Eliashberg, From Stein to Weinstein and Back – Symplectic
Geometry of Affine Complex Manifolds, Colloquium Publications Vol. 59, Amer.
Math. Soc. (2012).

[12] G. Dimitroglou Rizell, E. Goodman and A. Ivrii, Lagrangian isotopy of tori in


S 2 × S 2 and CP 2 , arXiv:1602.08821.

[13] T. Ekholm, Rational SFT, linearized Legendrian contact homology, and La-
grangian Floer cohomology, Perspectives in analysis, geometry, and topology,
109–145, Progr. Math., 296, Birkhäuser/Springer, New York, 2012.

[14] T. Ekholm and Y. Lekili, Duality between Lagrangian and Legendrian invariants,
arXiv:1701.01284.

[15] Y. Eliashberg, Symplectic geometry of plurisubharmonic functions, in “Gauge


Theory and Symplectic Geometry”, Vol. 488 of the series NATO ASI Series,
49–67.

[16] Y. Eliashberg, Topological characterization of Stein manifolds of dimension > 2,


Internat. J. Math. 1(1990), no. 1, 29-46.
8 Nadler’s program of arborealization 31

[17] Y. Eliashberg, Classification of contact structures on R3 , Int. Math. Res. No-


tices, 1993, no. 3, 87–91.

[18] Y. Eliashberg, S. Ganatra and O. Lazarev, Flexible Lagrangians,


arXiv:1510.01287.

[19] Y. Eliashberg, S. Ganatra and O. Lazarev, Topology of flexible fillings, in prepa-


ration.

[20] Y. Eliashberg, M Gromov, Convex symplectic manifolds, Proc. Symp. Pure


Math., 52(1991), Amer. Math. Soc., Providence, RI, 135–162.

[21] Y. Eliashberg, M Gromov, Lagrangian intersection theory. Finite-dimensional


approach, AMS Transl., 186(1998), N2, 27–116.

[22] Y. Eliashberg and E. Murphy, Lagrangian caps, Geom. and Funct. Anal.,
23(2013), Volume 23, 1483–1514.

[23] Y. Eliashberg, D. Nadler and L. Starkston, in preparation.

[24] E. Giroux, Convexité en topologie de contact, Comment. Math. Helv., 66 (1991),


637–677.

[25] E. Giroux, Géométrie de contact: de la dimension trois vers les dimesions


supérieures, Proc. of the ICM, Vol. II (Beijing, 2002), Higher Ed. Press, Beijing,
2002, 405–414.

[26] S. Ganatra, J. Pardon and V. Shende, Covariantly functorial Floer theory on


Liouville sectors, arXiv:1706.03152.

[27] R. Hind, Lagrangian spheres in S 2 ×S 2 , Geom. Funct. Anal., 14(2004), 303–318.

[28] M. Kontsevich, Symplectic geometry of homological algebra,


http://www.ihes.fr/∼maxim/TEXTS/Symplectic− AT2009.pdf.

[29] O. Lazarev, Contact manifolds with flexible fillings, arXiv:1610.04837.

[30] F. Laudenbach, On the Thom–Smale complex, an Appendix to Bismut-Zhang,


An extension of a Theorem by Cheeger and Müller, Astérisque 205(1992).

[31] D. McDuff, Symplectic manifolds with contact type boundaries, Invent. Math.
103(1991), 651–671.
8 Nadler’s program of arborealization 32

[32] M. Maydanskiy and P. Seidel, Lefschetz fibrations and exotic symplectic struc-
tures on cotangent bundles of spheres, J. Topol. 3 (2010), 157–180.

[33] M. McLean, Lefschetz fibrations and symplectic homology, Geom. Topol.


13(2009), 1877–1944.

[34] M. Kervaire, and J.W. Milnor, Groups of homotopy spheres: I. Ann. of Math.,
77(3), 504–537.

[35] E. Murphy, Loose Legendrian embeddings in high dimensional contact manifolds,


arXiv:1201.2245.

[36] E. Murphy, Closed exact Lagrangians in the symplectization of contact mani-


folds, arXiv:1304.6620.

[37] E. Murphy and K. Siegel, Subflexible symplectic manifolds, arXiv:1510.01867.

[38] D. Nadler, Arboreal Singularities, arXiv:1309.4122

[39] D. Nadler, Non-characteristic expansion of Legendrian singularities,


arXiv:1507.01513.

[40] B. Ozbagci, On the topology of fillings of contact 3-manifolds,


http://home.ku.edu.tr/∼bozbagci/SurveyFillings.pdf.

[41] P. Seidel and I. Smith, The symplectic topology of Ramanujams surface, Com-
ment. Math. Helv. 80(2005), 859–881.

[42] Z. Sylvan, On partially wrapped Fukaya categories, arXiv: 1604:02540v2.

[43] S. Smale, On the structure of manifolds, Amer. J. Math., 84(1962) pp. 387–399.

[44] L. Starkston, Arboreal Singularities in Weinstein Skeleta, arXiv:1707.03446.

[45] A. Weinstein, Contact surgery and symplectic handlebodies, Hokkaido


Math. J. 20(1991), 241–251.

You might also like