DISPLACEABILITY IN SYMPLECTIC GEOMETRY
A Dissertation Presented to the Faculty of the Graduate School
of Cornell University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
by Frederik De Keersmaeker August 2020 c 2020 Frederik De Keersmaeker ALL RIGHTS RESERVED DISPLACEABILITY IN SYMPLECTIC GEOMETRY Frederik De Keersmaeker, Ph.D. Cornell University 2020
We use a topological condition by Albers that is sufficient for nondisplaceabil- ity to describe a large class of nondisplaceable Lagrangians, namely the anti- diagonals of closed monotone symplectic manifolds. We provide some exam- ples of closed monotone symplectic manifolds of Euler characteristic zero.
Furthermore, we study the displaceability of fibers of the bending flow sys- tem on equilateral pentagon space. Besides torus fibers over points in the in- terior of moment map image, this completely integrable system has two La- grangian sphere fibers over boundary points of the moment polytope. They are nondisplaceable for topological reasons. Most of the regular torus fibers are displaceable using McDuff’s probes technique. We prove that the central torus fiber is nondisplaceable by showing that we can lift a Hamiltonian diffeo- morphism displacing it to a Hamiltonian diffeomorphism displacing the cen- tral torus fiber of the Gelfand-Cetlin system on the complex Grassmannian of
2-planes. This fiber is known to be nondisplaceable. BIOGRAPHICAL SKETCH
Frederik De Keersmaeker was born in Belgium in 1989. He started studying at
Ghent University in 2007, where he got an M.S. in Engineering Physics in 2012 and an M.S. in Mathematics in 2013. After working as a teaching assistant for a year at KULeuven KULAK, he enrolled in the Ph.D. program in Mathematics at
Cornell University.
iii This document is dedicated to all Cornell graduate students.
iv ACKNOWLEDGEMENTS
I cannot imagine what my experience in graduate school would have been like without my adviser, Tara Holm. She is the most supportive and caring per- son. Thank you for guiding me through my research over the past six years, for encouraging me to be a better teacher and to do outreach, for suggesting I go to as many interesting workshops and conferences as possible, and for providing a safe space to talk about the ups and downs. The Cornell Math Department is a wonderful place, and I have a lot of peo- ple to thank there. Let me start with my minor committee members: Reyer Sjamaar and Allen Knutson. Thank you for reading this dissertation and for the many useful comments. I have loved teaching at Cornell; Kelly Delp and Ja- son Manning played a big role is making it a positive experience. Crossing the border to get into the U.S. would have been much harder without those letter Melissa Totman wrote for me, saying I was a valued member of the department.
The heart of the department is its graduate students. There are a lot of names to list here, and I hope I have conveyed my appreciation for you all during our time at Cornell. Thank you Ian, My, Thomas, Balazs, Daoji, Emily, David,
Lila, Smaranda, Trevor, Ryan, Ellie, Nicki, Fiona, Portia, Joe, Amin, Sasha, Jeff, Kelsey, Valente, and Voula. My symplectic buddy, Benjamin Hoffman, deserves a special shout-out. So does Hannah Keese for teaching in prison with me for two semesters. I learned a lot during the research visits I went on. I want to thank Nguyen Tien Zung at Universite´ Paul Sabatier in Toulouse and Katrin Wehrheim at UC
Berkeley for talking to me about my research. Ithaca has been a wonderful place to live with my friends. Thank you Lau- ren, Michelle, Naomi, Drew, Gennie, Jaron, Emily, Smaranda, Patrick, Margaux,
v Trevor, Benjamin, Phoebe, Jan, Patsy, and Valeria! Whenever I was back in Belgium, Emilie, Annabelle, Margo, Emma, and Margot made me feel like I had never left! Thank you for all the support!
Finally, I want to thank my parents and my family for cheering me on and believing in me.
vi TABLE OF CONTENTS
Biographical Sketch ...... iii Dedication ...... iv Acknowledgements ...... v Table of Contents ...... vii List of Tables ...... ix List of Figures ...... x
1 Introduction 1
2 Preliminary Material 3 2.1 A Brief Overview of Symplectic Geometry ...... 3 2.1.1 Symplectic Manifolds: Definition and Examples ...... 3 2.1.2 Local Normal Forms ...... 6 2.1.3 Hamiltonian Vector Fields ...... 7 2.1.4 Hamiltonian Diffeomorphisms ...... 9 2.1.5 Hamiltonian Group Actions and Symplectic Reduction . . 13 2.1.6 Compatible Almost Complex Structures ...... 15 2.2 Nondisplaceable Lagrangians ...... 16 2.2.1 Definition and Properties ...... 16 2.2.2 Examples ...... 17 2.3 Monotone and Semipositive Symplectic Manifolds ...... 18 2.4 The Maslov Class and Monotone Lagrangians ...... 21
3 Floer Homology and the PSS Isomorphism 29 3.1 Morse Theory ...... 29 3.1.1 Morse Functions ...... 29 3.1.2 The Morse Complex ...... 30 3.1.3 Continuation Maps ...... 33 3.1.4 Functoriality and Induced Maps ...... 36 3.1.5 Poincare´ Duality ...... 37 3.2 Periodic Orbits of Hamiltonian Vector Fields ...... 39 3.2.1 Definition and Properties ...... 40 3.2.2 The Arnold Conjecture ...... 41 3.3 Floer Homology ...... 42 3.3.1 The Novikov Ring ...... 43 3.3.2 The Loop Space and the Action Functional ...... 45 3.3.3 The Conley-Zehnder Index ...... 47 3.3.4 The Floer Chain Groups ...... 51 3.3.5 The Boundary Operator ...... 52 3.4 Moduli Spaces of Floer Cylinders ...... 59 3.4.1 Fredholm Operators and the Implicit Function Theorem . 59 3.4.2 Smoothness, Linearization, and Transversality ...... 61
vii 3.4.3 Compactness ...... 62 3.5 The Piunikhin-Salamon-Schwarz (PSS) Isomorphism ...... 69 3.6 Representing Submanifolds Using PSS ...... 75 3.7 An Application to Displaceability ...... 82 3.8 Nondisplaceable Anti-Diagonals ...... 84
4 Polygon Spaces 89 4.1 Definition and Elementary Properties ...... 89 4.2 The Bending Flow System ...... 90 4.3 McDuff’s Probes Technique ...... 94 4.4 Displacing Fibers and Symplectic Reduction ...... 98 4.5 Polygon Spaces and Grassmannians ...... 100 4.6 The Gelfand-Cetlin System ...... 104 4.7 Displacing the Central Fiber ...... 107
Bibliography 110
viii LIST OF TABLES
2.1 Semipositivity of products of two complex projective spaces in low dimension...... 21 2.2 Semipositivity of products of three complex projective spaces in low dimension...... 21
3.1 Different moduli spaces of Floer cylinders ...... 62
ix LIST OF FIGURES
3.1 Floer cylinder with disk cappings connecting two contractible periodic solutions (in blue)...... 55 3.2 Broken configuration with two Floer cylinders, with periodic or- bits in blue...... 57 3.3 A plumber’s helper solution...... 70 3.4 The two possible cases for a broken plumber’s helper solution with one break: either the Morse flow half-line breaks (top) or the cylindrical part breaks (bottom)...... 73 3.5 A broken configuration: periodic orbits in red connected by Floer cylinders, and Lagrangian boundary condition for the rightmost cylinder, with sphere bubbles in green and disk bub- bles in blue...... 78
4.1 A pentagon with its two diagonals as red dashed line segments. 91 4.2 The image of the bending flow system for r = (2, 3, 3, 2, 3)..... 92 4.3 The image of the bending flow system for r = (1, 1, 1, 1, 1). The polytope is not smooth at the two blue vertices...... 93 4.4 A equilateral pentagon with a vanishing diagonal consists of an equilateral triangle and a line segment...... 93 4.5 Bending flow fibers displaced by vertical (left) and horizontal (right) probes in equilateral pentagon space...... 96 4.6 Bending flow fibers displaced by probes in equilateral pentagon space...... 97
x CHAPTER 1 INTRODUCTION
The main subject of this dissertation is nondispaceable Lagrangians. They are symplectic analogues of n-dimensional submanifolds of an 2n-dimensional smooth manifold with nonzero self-intersection number.
We start Chapter 2 with a brief overview of symplectic geometry. We define displaceable and nondisplaceable Lagrangians and introduce monotonicity of symplectic manifolds and their Lagrangians.
In Chapter 3 we present a topological condition (Theorem 15) by Peter Al- bers that guarantees nondisplaceability. We describe a large class of nondis- placeable Lagrangians, namely the anti-diagonals of closed monotone symplec- tic manifolds (Theorem 16). If their Euler characteristic is nonzero, they are not displaceable by maps isotopic to the identity (a class that contains Hamiltonian diffeomorphisms). We provide some examples of closed monotone symplectic manifolds of Euler characteristic zero (Examples 9 and 10).
In Chapter 4 we study the displaceability of fibers of the bending flow sys- tem on equilateral pentagon space. Besides torus fibers over points in the in- terior of moment map image, this completely integrable system has two La- grangian sphere fibers over boundary points of the moment polytope. They are nondisplaceable for topological reasons (Proposition 25). Most of the regular torus fibers are displaceable using McDuff’s probes technique (Theorem 17). We prove that the central torus fiber is nondisplaceable (Theorem 23) by showing that we can lift a Hamiltonian diffeomorphism displacing it to a Hamiltonian diffeomorphism displacing the central torus fiber of the Gelfand-Cetlin system
1 on the complex Grassmannian of 2-planes. This fiber is known to be nondis- placeable.
2 CHAPTER 2 PRELIMINARY MATERIAL
After a brief overview of symplectic geometry, we define what it means for a Lagrangian submanifold in a symplectic manifold to be displaceable (by Hamil- tonian diffeomorphisms). We introduce monotonicity for symplectic manifolds and Lagrangians.
2.1 A Brief Overview of Symplectic Geometry
We review the basic concepts from symplectic geometry that we will need this dissertation. We use this opportunity to introduce notation and conventions.
Unless stated otherwise, the material presented in this section can be found in [MS17] or [CdS01].
2.1.1 Symplectic Manifolds: Definition and Examples
A symplectic form on a smooth manifold M is a closed nondegenerate 2-form,
2 i.e. an element ω of Ω (M) satisfying dω = 0, such that ωp : TpM × TpM R is a nondegenerate skew-symmetric bilinear form for all p ∈ M. Every nonzero → tangent vector v ∈ TpM has a nonzero symplectic buddy w ∈ TpM, unique up to multiplication by a scalar, such that ωp(v, w) 6= 0. We call the pair (M, ω) a symplectic manifold.
The existence of a symplectic form severely restricts the topology of M:
• Since nondegenerate skew-symmetric bilinear forms only exist on even-
3 dimensional vector spaces, the dimension dim(M) = 2n must be even. We will often write M2n to mean that M is a manifold of dimension 2n.
• Nondegeneracy is equivalent to the top form ωn = ω ∧ ··· ∧ ω ∈ Ω2n(M) being a volume form. Therefore M must be orientable; it carries an orien- tation induced by the symplectic form.
2 • Since ω is closed, it defines a de Rham cohomology class [ω] ∈ HdR(M). If M is closed, then [ω] cannot be zero: if ω = dα for some α ∈ Ω1(M), then
ωn = d(α∧ωn−1). The integral of a closed top form is zero by Stokes’ The- orem, contradicting the fact that the integral of a volume form is nonzero.
2i In fact, this argument implies that HdR(M) 6= 0 for all i ∈ {0, 1, . . . , n}.
Example 1. We give some elementary examples of symplectic manifolds.
1. Any area form on a closed orientable surface Σg is symplectic. The
closedness condition is automatically satisfied. In particular, the two- dimensional sphere S2 is symplectic.
2. If the k-dimensional sphere Sk admits a symplectic form for k ≥ 4, then
2 k 0 2 HdR(S ) 6= 0 since it is closed. Therefore S and S are the only spheres that are symplectic.
3. The complex projective spaces CPn admit a symplectic form, called the
Fubini-Study form ωFS. Its de Rham cohomology class is determined by the fact that it assign symplectic area π to any complex line. More gener- ally, any complex submanifold of CPn equipped with the restriction of the
Fubini-Study form is symplectic.
4. As follows from the previous example, all complex Grassmannians and
flag manifolds admit symplectic forms.
4 5. Consider Cn =∼ R2n with coordinates zj = xj + iyj for j ∈ {1, . . . , n}. In light of the Darboux Theorem (see Theorem 1), we call
n i n ω := dxj ∧ dyj = dzj ∧ dzj st 2 j=1 j=1 X X the standard symplectic form.
6. Let Xn be a smooth manifold and T ∗X its cotangent bundle. Let (x1, . . . , xn) be a coordinate system defined on an open subset U ⊂ X and (ξ1, . . . , ξn) the associated fiber coordinates on T ∗U. It is straightforward to check that
the forms
n n i i i i αU := ξ dx and ωU := dx ∧ dξ = −dαU i=1 i=1 X X on T ∗U do not depend on the chosen coordinates. They define global
∗ 1 ∗ forms on T X, called tautological form αtaut ∈ Ω (T X) and the canoni-
2 ∗ cal symplectic form ωcan ∈ Ω (T X).
A diffeomorphism f : M1 M2 between two symplectic manifolds (M1, ω1)
∗ and (M2, ω2) is a symplectomorphism if f ω2 = ω1. The symplectomorphisms → of (M, ω) form a subgroup of the diffeomorphism group Diff(M), which we denote by Sympl(M, ω).
Let (M2n, ω) be a symplectic manifold and i : S M the inclusion of a submanifold. We say S is isotropic if the restriction i∗ω of the symplectic form to → S vanishes. We say S is Lagrangian if it is maximally isotropic, i.e. isotropic and not strictly contained another isotropic submanifold. Symplectic linear algebra
1 guarantees that dim S ≤ n = 2 dim M if S is isotropic. Lagrangian submanifolds are exactly the n-dimensional isotropic submanifolds.
5 2.1.2 Local Normal Forms
We consider two local normal forms in this subsection, the first describing what a neighborhood of a point in a symplectic manifold looks like, and the second describing what a neighborhood of a Lagrangian looks like.
Theorem 1 (Darboux, [Dar82]). Let (M2n, ω) be a symplectic manifold. Then every point in M is contained in a coordinate chart ϕ : U ⊂ M R2n such that
n ! ∗ ∗ j j ω|U = ϕ ωst = ϕ dx ∧ dy . → j=1 X
The Darboux Theorem states that any point in a symplectic manifold of di- mension 2n has an open neighborhood that is symplectomorphic to an open
2n ball in the standard symplectic manifold (R , ωst). We use this opportunity to list some key differences between symplectic and Riemannian geometry.
• Since all symplectic manifolds of the same dimension are locally symplec- tomorphic, symplectic geometry has no local invariants other than dimen-
sion, in contrast to Riemannian geometry, which has local invariants like curvature. An example of a global symplectic invariant is the symplectic volume 1 vol(M2n, ω) := ωn. n! ZM • While any smooth manifold admits a Riemannian metric, we have already discussed that the existence of a symplectic form restricts the topology of
the underlying manifold.
• The isometry group of a compact Riemannian manifold is a compact Lie group [MS39]. Symplectomorphism groups tend to be infinite-
dimensional. The natural map on the cotangent bundle T ∗X induced by a
6 ∗ diffeomorphism f : X X is a symplectomorphism, so Sympl(T X, ωcan) contains a subgroup isomorphic to the diffeomorphism group of X. Cotan- → gent bundles are not compact, but even in the compact case, the symplec-
tomorphism group contains the infinite-dimensional subgroup of Hamil- tonian diffeomorphisms, which we introduce in the next subsection.
Weinstein proved in [Wei71] that the way a Lagrangian is embedded in a symplectic manifold only depends on the diffeomorphism type of the La- grangian.
Theorem 2 (Weinstein’s Lagrangian Neighborhood Theorem, [CdS01, Thm. 9.3]). Let (M, ω) be a symplectic manifold and L ⊂ M a Lagrangian. Con- sider the cotangent bundle T ∗L equipped with its canonical symplectic form.
Then the zero section L T ∗L can be extended to a symplectomorphism from a tubular neighborhood of L in M to a tubular neighborhood of the zero section → in T ∗L.
2.1.3 Hamiltonian Vector Fields
The proof of the Darboux Theorem relies on the fact that dω = 0. We dis- cuss another consequence of closedness here: the Jacobi identity for the Poisson bracket.
A symplectic vector field is a vector field X on M for which LXω = 0. Equiv-
t t ∗ alently, its time-t flow φX : M M satisfies (φX) ω = ω for all t ∈ R. Using Cartan’s magic formula and the closedness of the symplectic form, we write →
LXω = dιXω + ιXdω = dιXω.
7 A vector field X is therefore symplectic if and only if ιXω is closed.
t Remark 1. To ensure that the time-t flow φX of a vector field X is well-defined for all t ∈ R, we assume that X is compactly supported or that the manifold M is closed.
A Hamiltonian vector field is a vector field X on M for which ιXω is exact.
All Hamiltonian vector fields are symplectic. The Hamiltonian vector field Xf
1 corresponding to f ∈ C (M) is defined by the condition ω(Xf, ·) = −df. It ∞ uniquely defines Xf because ω is nondegenerate. We say that f is a Hamiltonian for Xf. Two functions define the same Hamiltonian vector field if their difference is locally constant. Since
LXf (f) = df(Xf) = ω(Xf,Xf) = 0,
the flow of Xf preserves the level sets of its Hamiltonian f.
If X and Y are two symplectic vector fields, then
ι[X,Y]ω = LXιYω − ιYLXω
= dιXιYω + ιXdιYω − ιYdιXω − ιYιXdω (2.1)
= −d ω(X, Y),
so their Lie bracket [X, Y] is the Hamiltonian vector field Xω(X,Y). In the Lie al- gebra X(M) of vector fields on M, the space Xsympl(M, ω) of symplectic vector
fields and the space XHam(M, ω) of Hamiltonian vector fields are Lie subalge- bras. In fact, XHam(M, ω) is an ideal of Xsympl(M, ω). The Poisson bracket of two functions f, g ∈ C (M) is the function {f, g} := ω(Xf,Xg) = dg(Xf). It satis-
1 ∞ Some authors [CdS01, MS17] use the convention ω(Xf, ·) = df. In that case the map in (2.2) is a Lie algebra anti-homomorphism.
8 fies the Jacobi identity since
1 {{f, g}, h} + {{g, h}, f} + {{h, f}, g} = (dω)(X ,X ,X ) = 0, 2 f g h so it is a Lie bracket on C (M). The definition of the Poisson bracket only relied ∞ on the nondegeneracy of the symplectic form. We observe that the closedness of ω is equivalent to the requirement that the Poisson bracket satisfy the Jacobi identity. The computation in (2.1) shows that the map
C (M) X(M): f 7 Xf (2.2) ∞ is a Lie algebra homomorphism. → →
We say that two functions on M are in involution if their Poisson bracket vanishes. An integrable system is a collection (f1, . . . , fk) of functions on M that are pairwise in involution and linearly independent, i.e. there is an open dense subset Ω ⊆ M such that their differentials dpf1, . . . , dpfk are linearly in-
dependent in TpM for all p ∈ Ω. Then the tangent vectors Xf1 (p),...,Xfk (p) are linearly independent and span an isotropic subspace of (TpM, ωp) for all p ∈ Ω, so k ≤ n. An integrable system is complete if k = n.
Remark 2. Some authors require the functions in an integrable system to be smooth. We will adopt a more generous requirement: we only require that the functions are continuous everywhere, and smooth on an open dense subset.
2.1.4 Hamiltonian Diffeomorphisms
We introduce Hamiltonian diffeomorphisms in this subsection as the flows of Hamiltonian vector fields. To get a sufficiently rich class of diffeomorphisms, we
9 allow the Hamiltonians to be time-dependent. The material in this subsection is from [HZ11].
Let (M, ω) be a closed symplectic manifold and H ∈ C (R × M). We can ∞ interpret H as a collection (Ht)t∈R of smooth functions on M that is smoothly indexed by time. Let (Xt)t∈R be the corresponding collection of Hamiltonian vector fields on M. It is a time-dependent vector field XH on M, which we call the time-dependent Hamiltonian vector field associated to H.
Remark 3. With a slight abuse of notation, we will often attribute objects and concepts to a Hamiltonian H that are traditionally attributed to its Hamiltonian X (φt ) vector field H. As an example, we will sometimes refer to the flow XH t∈R of t XH as the flow of H and denote it by (φH)t∈R. More abuse of notation in this vein is to follow.
t Recall that the flow of XH is a collection (φH)t∈R of diffeomorphisms of M
0 determined by φH = idM and
dφt H = X ◦ φt dt H H for all t ∈ R. In the autonomous (i.e. time-independent) case, the flows are
s t s+t symplectomorphisms that satisfy the group law φH ◦ φH = φH for all s, t ∈ R. While the group law no longer holds in the time-dependent case, the flows are still symplectomorphisms:
d (φt )∗ω = (φt )∗L ω dt H H Xt t ∗ = (φH) (dιXt ω + ιXt dω)
= 0,
t ∗ 0 ∗ and therefore (φH) ω = (φH) ω = ω.
10 We say a diffeomorphism on M is a Hamiltonian diffeomorphism if it is the
1 time-1 flow φH : M M of a time-dependent Hamiltonian vector field. We denote the set of Hamiltonian diffeomorphisms by Ham(M, ω). Since they play → an important role in this dissertation, we explore their properties in the rest of this section.
The next result shows that we can restrict our attention to 1-periodic Hamil- tonians H : S1 × M R, where we identified the circle S1 with R/Z. In the rest of this dissertation, we will use the term periodic to mean 1-periodic. →
Proposition 1. Let H : R×M R be a time-dependent Hamiltonian. Then there exists a periodic Hamiltonian K : S1 × M R such that φ1 = φ1 . → H K
→ Proof. Let α :[0, 1] [0, 1] be a smooth function with the property that there exists > 0 such that α(t) = 0 for all t < and α(t) = 1 for all t > 1 − . Define → K : R × M R by K(t, p) = α0(t)H(α(t), p) for t ∈ [0, 1] and then periodically
α(0) 0 extending it. Note that K is smooth. Since φ = φ idM and → H H d φα(t)(p) = α0(t)X φα(t)(p) dt H Ht H α(t) 0 = Xα (t)Ht φH (p)
α(t) = XKt φH (p)
α(t) t for all t ∈ [0, 1] and p ∈ M, we conclude that φH = φK for all t ∈ [0, 1]. The result follows since α(1) = 1.
To show that compositions and inverses of Hamiltonian diffeomorphisms are Hamiltonian diffeomorphisms, we need the following operations: given two periodic Hamiltonians H and K, let H # K : S1 × M R and H : S1 × M R be
→ →
11 the functions defined by