THE L2 GEOMETRY OF THE SYMPLECTOMORPHISM GROUP
A Dissertation
Submitted to the Graduate School of the University of Notre Dame in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
by
James Benn,
Gerard Misio lek,Director
Graduate Program in Mathematics Notre Dame, Indiana April 2015 c Copyright by James Benn 2015 All Rights Reserved THE L2 GEOMETRY OF THE SYMPLECTOMORPHISM GROUP
Abstract by James Benn
In this thesis we study the geometry of the group of Symplectic diffeomor- phisms of a closed Symplectic manifold M, equipped with the L2 weak Riemannian metric. It is known that the group of Symplectic diffeomorphisms is geodesically complete with respect to this L2 metric and admits an exponential mapping which is defined on the whole tangent space. Our primary objective is to describe the structure of the set of singularities of associated weak Riemannian exponential mapping, which are known as conjugate points. We construct examples of con- jugate points on the Symplectomorphism group and solve the Jacobi equation explicitly along geodesics consisting of isometries of M. Using the functional cal- culus and spectral theory, we show that every such geodesic contains conjugate points, all of which have even multiplicity. A macroscopic view of conjugate points is then given by showing that the exponential mapping of the L2 metric is a non- linear Fredholm map of index zero, from which we deduce that conjugate points constitute a set of first Baire category in the Symplectic diffeomorphism group. Finally, using the Fredholm properties of the exponential mapping, we give a new characterization of conjugate points along stationary geodesics in terms of the linearized geodesic equation and coadjoint orbits. CONTENTS
ACKNOWLEDGMENTS ...... iv
CHAPTER 1: INTRODUCTION AND OVERVIEW ...... 1
CHAPTER 2: PRELIMINARIES ...... 8 2.1 Diffeomorphism Groups ...... 8 2.1.1 Manifolds of Mappings ...... 8 2.1.2 Groups of Diffeomorphisms ...... 11 2.1.3 Exponential Mappings ...... 12 2.2 Weak Riemannian Structure on the Diffeomorphism Group . . . . 14 s 2.3 The Symplectic Diffeomorphism Group Dω ...... 19 s 2.3.1 The Manifold Structure of Dω ...... 19 2.3.2 Hodge Theory for Manifolds ...... 24 s 2.3.3 Weak Riemannian Structure for Dω(M)...... 33 2.4 L2 Geodesic Completeness ...... 36 2.5 Curvature ...... 42
CHAPTER 3: CONJUGATE POINTS IN THE SYMPLECTOMORPHISM GROUP ...... 49 3.1 Introduction ...... 49 3.2 Adjoint and Coadjoint Operators ...... 51 3.3 The Second Fundamental Tensor and Curvature ...... 62 s n 3.4 Conjugate Points on Dω(CP )...... 71 3.5 Geodesics in the Isometry Subgroup ...... 77 3.6 The Jacobi Equation along Geodesics of Isometries ...... 82
CHAPTER 4: FREDHOLM PROPERTIES OF THE L2 EXPONENTIAL s MAP ON Dω ...... 92 4.1 Introduction ...... 92 4.2 The Jacobi Equation ...... 94 4.3 Proof of Fredholmness in Hσ ...... 98 4.4 Proof of Fredholmness in Hs ...... 108
ii s n CHAPTER 5: CONJUGATE POINTS ON DHam(M ) IN DIMENSIONS n =2AND4 ...... 114 5.1 Introduction ...... 114 5.2 A Conservation Law ...... 118 5.3 A Characterization of Conjugate Points along Stationary Geodesics 120 5.4 Proof of Proposition 5.3.1 ...... 122
APPENDIX A: SOBOLEV SPACES ...... 130 A.1 Sobolev Spaces of Bounded Domains in Rn ...... 130 A.2 Sobolev Spaces on Compact Manifolds ...... 131 A.3 Sobolev Spaces on Vector Bundles ...... 132
APPENDIX B: SYMPLECTIC VECTOR SPCAES AND MANIFOLDS . 135 B.1 Symplectic Manifolds ...... 135 B.2 Symplectomorphisms ...... 137
APPENDIX C: FREDHOLM OPERATORS ...... 140 C.1 Compact Operators ...... 140 C.2 Fredholm Operators ...... 142
APPENDIX D: SPECTRAL INTEGRALS ...... 146 D.1 Spectral Measures ...... 146 D.2 Stone’s Theorem ...... 148
BIBLIOGRAPHY ...... 151
iii ACKNOWLEDGMENTS
A very special thanks goes out to my friend and advisor, Gerard Misio lek,for introducing me to the subject and for his endless ideas. Thankyou to Alex Hi- monas for his ongoing support and to Steve Preston for his interest and useful suggestions. My family, my friends; I thank all of you deeply.
iv CHAPTER 1
INTRODUCTION AND OVERVIEW
Let M be a closed Symplectic manifold with Symplectic form ω and Rie- mannian metric g. We assume that ω and g are compatible, in the sense that there exists an almost complex structure J : TM → TM satisfying J 2 = −I, g(Jv, Jw) = g(v, w), and g(v, Jw) = ω(v, w), for any vector fields v, w on M (see
s appendix A). Let Dω (M) denote the group of all diffeomorphisms of Sobolev class
s dim M s H preserving the Symplectic form ω on M. If s > 2 + 1 then Dω becomes an infinite dimensional Hilbert manifold whose tangent spaces at a point η consists of Hs sections X of the pull-back bundle η∗TM for which the corresponding vector
−1 field v = X ◦ η on M satisfies Lvω = 0, where L is the usual Lie derivative. Using right-translations, the L2 inner product on vector fields,
Z s (u, v)L2 = g (u, v) dµ, u, v ∈ TeDω, (1.1) M defines a right-invariant metric on the group. This thesis is concerned with the
2 s s L geometry of the group Dω, and it’s finite codimensional subgroup DHam - the group of Hamiltonian diffeomorphisms. Diffeomorphism groups can be realized as the configuration spaces of a number of equations in mathematical physics, which provides a strong motivation to study their geometry. Perhaps the most famous example is the Euler equations of hy-
1 drodynamics, where Arnold, [A], noticed that a curve η(t) in the group of smooth
s 2 volume preserving diffeomorphisms (Dµ) is a geodesic of the L metric (1.1) if and only if the vector field v, defined by ∂tη = v ◦ η, solves the Euler equations of hydrodynamics. The L2 metric is simply the kinetic energy of the fluid, and the geodesic equation is a manifestation of Newton’s second law F = ma.
s 2 Analogously, a curve η(t) in Dω(M) is a geodesic of the L metric starting from the identity in the direction vo if and only if the time dependent vector field v =η ˙ ◦ η−1 on M solves the Symplectic Euler equations
ω ∂tv + Pe (∇vv) = 0 (1.2)
Lvω = 0
v(0) = vo,
ω where Pe is the orthogonal projection onto the space of Symplectic vector fields. The subgroup of Hamiltonian diffeomorphisms plays a role in plasma dynamics analogous to the role played by the volume preserving diffeomorphism group in incompressible hydrodynamics, see Arnold and Khesin [A-Kh], Holm and Tronci [H-T], Morrison [Mo], Marsden and Weinstein, [M-W], for details. Chapter 2 contains a review of the manifold structure of mapping spaces and diffeomorphism groups. We briefly describe the shortcomings of the group expo- nential map on diffeomorphism groups and motivate the endowment of a weak Riemannian structure (the L2 metric (1.1)) on these manifolds. Section 2.3 fo- cuses our attention on the Symplectomorphism and Hamiltonian subgroups (and submanifolds!) of the diffeomorphism group. Here we recall the Hodge decompo- sition of forms and the fundamental results of Ebin and Marsden, [E-M], which
2 prove the existence of a smooth right-invariant connection and exponential map- ping associated to the weak L2 metric. We define the weak Riemannian curvature tensors on the Symplectomorphism group and indicate that they are trilinear op- erators bounded in the strong Sobolev Hs topology, as shown in Misiolek [M1]. Finally, in section 2.4 we review Ebin’s [Eb] and Khesin’s [Kh] proof that solutions to the Symplectic Euler equations (1.2) exist globally in time for any Symplectic
s 2 manifold, so that the group Dω(M) is L geodesically complete. Arnold ([A]) computed sectional curvatures of diffeomorphism groups and found that they were mostly negative, although in some small regions they were
s positive. He asked if there are conjugate points on Dµ and called for a description
s of them. Much progress in understanding conjugate points on Dµ has been made since the work of Misiolek. In [M1], some simple examples of conjugate points in the Volumorphism group were constructed, answering Arnold’s first question in the affirmative. More examples were later provided by Misiolek [M2], Preston [P2], and Shnirelman [Sh2]. In contrast with finite dimensional geometry, two types of conjugate points can occur in infinite dimensions. Grossman [Gro] gave the first examples of the two types of conjugacies that may occur: on a sphere in Hilbert space the differential of the exponential map may have infinite dimensional kernel corresponding to an infinite-dimensional family of geodesics joining two antipodal points. In addition, on an infinite dimensional ellipsoid the exponential map differential fails to be surjective, even though it is injective, in certain directions.
s A natural question to ask is whether conjugate points exist on Dω. This is the contents of chapter 3: Some simple examples of conjugate points are constructed on the Symplectomorphism group of the complex projective plane. In particular,
3 s n n Theorem. (3.4.3) Conjugate points exist on Dω (CP ), for s > 2 + 1 and n ≥ 2.
s We then show that geodesics which lie in the isometry subgroup of Dω always carry conjugate points, all of which have even multiplicity.
2 Theorem. (3.6.2, 3.6.4) Let η(t) = exp(tvo) be a geodesic of the L metric (1.1) generated by a Killing vector field vo. Let J(t) be a Jacobi field along η(t), with
0 initial conditions J(0) = 0, J (0) = wo. Then
tKv0 −tKv e − I J(t) = Dη(t) · e 0 w0, Kv0
s where Kvo (·) is a compact, skew self-adjoint operator on TeDω, and we have the
−tK R itλ etKv0 −I R eitλ−1 spectral representations e vo = e dE(λ) and = dE(λ) which R Kv0 R iλ s are linear operators on TeDω. Consequently, the multiplicity of each conjugate point along η(t) is even.
s Conjugate points on Dµ(M), M a closed surface, were studied extensively by Ebin, Misiolek and Preston in [E-M-P], where it was shown that the exponential map of the L2 metric is a non-linear Fredholm map of index zero. A corollary of this is that the two types of conjugacies mentioned above coincide. Moreover, conjugate points are isolated and of finite multiplicity. The conclusion is that the exponential map behaves like that of a finite dimensional manifold. When M is
s three dimensional, the singularities of the exponential mapping on Dµ typically behave pathologically and the exponential map is no longer a Fredholm map, cf. [E-M-P], Misiolek-Preston [M-P], [P2], [Sh2]. In chapter 4 we will show
Theorem. (4.1.1) Let M be a closed Symplectic manifold of dimension n = 2m
2 s and s > m + 1. Then the exponential map of the L metric on Dω(M) is a nonlinear Fredholm map of index zero.
4 This result provides a distinction between Symplectic diffeomorphisms and Volume preserving diffeomorphisms, when equipped with the L2 metric. It holds
s for any closed Symplectic manifold of dimension 2n, but fails for the group Dµ of manifolds of dimension 3 and higher, cf. [E-M-P], [P2]. The relationship between Fredholmness of the L2 exponential map and known classifications (e.g. C0 closure, Gromov’s non-squeezing Theorem) of Symplectic diffeomorphisms is, at this point, unclear. In chapter 5 we give a new geometric characterization of conjugate points along a stationary geodesic, and relate their existence to the right-invariance of the L2 metric.
Theorem. (5.3.2) Let M be a two or four dimensional Symplectic manifold and
2 s dim M η(t) a stationary geodesic of the L metric on Dω, s > 2 + 1, with initial ∗ velocity vo. Then η(t ) is conjugate to the identity if and only if there exists a
⊥ s v ∈ (ker Kvo ) ⊂ TeDω such that
∗ ∗ S(t )v = Adη−1(t∗)v,
∗ 2 where Adη(t) is the formal L adjoint of the push forward of vector fields operation,
s Kvo (·) a compact skew self-adjoint operator on TeDω, and S(t) is the solution operator of the linearized Symplectic Euler equations.
That is, conjugate points occur when a solution to the linearized Symplectic
s Euler equations (with initial value v), which can be thought of as a curve in TeDω, intersects the coadjoint orbit of its initial value v which can also be thought of as
s a curve in TeDω. Moreover, we are able to express solutions of the Jacobi equation in terms of solutions to the linearized Symplectic Euler equation and coadjoint
5 orbits. Namely,
Theorem. (5.3.3) Let M be a two or four dimensional Symplectic manifold and
2 s dim M η(t) be a stationary geodesic of the L metric on Dω, s > 2 +1. Then the Jacobi 0 ⊥ field J(t) = u(t) ◦ η(t) with initial conditions J(0) = 0, J (0) = wo ∈ (ker Kvo ) is given by X 1 u(t) = (g (t) − a (t)) v , λ i i i i i
0 where {vi}i∈N is a complete orthonormal set of eigenvectors of Kvo spanning TeDω, P Kvo vi = λivi, S(t)wo = i gi(t)vi solves the linearized Symplectic Euler equations ∗ P and Adη−1(t)wo = ai(t)vi.
In particular, the growth of Jacobi fields (measured in some norm) is deter- mined by how much solutions of the linearized Symplectic Euler equations differ from coadjoint orbits (measured in the same norm). It is interesting to understand what information conjugate points carry about the qualitative behavior of flows. For example, the inviscid Burgers’ equation
∂tv + ∇vv = 0
v(0) = vo describes the motion of a collection of particles moving without any internal forces. It is also the geodesic equation of the L2 metric on the full diffeomorphism group. It has been shown by Khesin and Misiolek, [Kh-M], that the Burgers’ equation has solutions in which particles begin colliding with one-another, forming shock- waves, in finite time. That is, geodesics (i.e. particle trajectories) cease to be diffeomorphisms (or reach the boundary of the group) in finite time. The first conjugate point along a geodesic generated by such a solution signals the onset of
6 shock-waves in the material space. On the other hand, solutions of the 2D Euler- equations of hydrodynamics (and the Symplectic Euler equations (1.2)) exist for all time and yet some of the corresponding geodesics contain conjugate points.
7 CHAPTER 2
PRELIMINARIES
2.1 Diffeomorphism Groups
2.1.1 Manifolds of Mappings
The basic idea of giving a manifold structure to mapping spaces was first laid down by Eells [E1] in 1958 where he constructed a smooth manifold out of the set of continuous maps between two manifolds. Constructing a manifold from Ck- diffeomorphisms of a compact manifold without boundary was done independently around 1966 by Abraham, Eells and Leslie, [E2], [L]. The Sobolev Hs case was done by Ebin and Marsden [E-M] where they gave a manifold structure to the Hs diffeomorphism group, the Volume-preserving diffeomorphism subgroup and the Symplectic diffeomorphism subgroup of a compact manifold with, or without, boundary. The construction is as follows. Let M and N be two compact manifolds each endowed with a Riemannian metric, g and h, and suppose N is without boundary. For an integer s, a map f : M → N is of Sobolev class Hs (write f ∈ Hs(M,N)) if for any point p ∈ M and chart around p,(Up, ϕ), and any chart (V, ψ) around f(p) the composite map ψ◦f◦ϕ−1 : ϕ(U) → Rn is in Hs(ϕ(U), Rn). If the sobolev index
m s 0 s satisfies s > 2 then by the Sobolev embedding Lemma, H (M,N) ⊂ C (M,N), and the above notion is well-defined and independent of the charts chosen. We
8 refer the reader to appendix A for a review of Sobolev spaces. In order to define charts on Hs(M,N) we need to determine the correct mod- eling space for Hs(M,N). Just as in finite dimensions, we use the tangent space as the model space for a manifold and we shall proceed similarly here. With this in mind, we shall look for a good description of the tangent space at a point f ∈ Hs(M,N). Consider a curve c :(−, ) → Hs(M,N) such that c(0) = f. For a point p ∈ M, the map t 7→ c(t)(p) is a curve in N. Now c(0)(p) = f(p) and so the
d derivative of this curve at 0 is dt c(t)(p)|t=0 and is an element of Tf(p)N. Therefore, d the map p 7→ dt c(t)(p)|t=0 is a map from M to TN and such that the canonical projection πN : TN → N covers f. Making the identification
d d c(t)| (p) = c(t)(p)| , dt t=0 dt t=0 the tangent space at a point f ∈ Hs(M,N) is
s s Tf H (M,N) = {X : M → TN : X ∈ H (M,TN), πN ◦ X = f} .
Here, Hs(M,TN) is the space of all sections from M to TN which have L2 deriva- tives up to order s. Define the inner product
Z X k k (V,W )s = g ∇ V, ∇ W dµ |k|≤s M where ∇k means the k − th order covariant derivative given by the Riemannian metric on N. With this inner product, Hs(M,TN) is a Hilbert space and all
9 m sections are continuous by the Sobolev Lemma (since s > 2 ) and the topology of Hs(M,TN) is stronger than that of uniform convergence. In order to construct an f-centered chart for Hs(M,N) we use the Riemannian exponential map of N. Since N is closed, it is geodesically complete and for each
N x ∈ N, the exponential map exp x : TxN → N is defined on the whole of TxN.
N N Consequently, exp x can be extended to a map exp : TN → N, where for
N N vx ∈ TxN, exp (vx) = exp x(vx). Since f(M) is compact, there is a number
λf > 0 such that any point of N whose distance from f(x) is less than λf can be joined by a unique geodesic arc of length less than λf . That is, for any point p of
N whose distance from f(x) is less than λf there is an X(x) ∈ Tf(x)N which lies
N in the disk of radius λf centered at 0, such that expf(x) X(x) = p and the map
N x 7→ expf(x) X(x) is a map from M to N. Consequently, the map
s s Ψ: Tf H (M,N) → H (M,N)
X 7→ expN X
s gives a bijective correspondence between the disc of radius λf in Tf H (M,N)
s centered at 0 and the disc of radius λf in H (M,N) centered at f. Since exp : TN → N is a local diffeomorphism, the transition functions are compositions of smooth maps and hence Hs(M,N) has a smooth manifold struc- ture. Moreover, compactness of M and N is used to show that the topology defined on Hs(M,N) is independent of the metric.
10 2.1.2 Groups of Diffeomorphisms
The diffeomorphism groups have a very rich and complicated structure which is still not very well understood. Let M be a compact manifold without boundary
n s 1 1 and assume that s > 2 + 1 so that the H topology is stronger than C . Let C D be the group of C1 diffeomorphisms of M, to itself, and let Ds(M) = Hs(M,M)∩ C1D. According to Theorem 1.7 of Hirsch’s Differential Topology, [Hi], C1D is open in C1(M,M), so that Ds(M) is open in Hs(M,M) and hence inherits its manifold structure. Ds(M) is also a topological group with composition as the group operation. Right multiplication is smooth:
s s Rη : D → D
ξ 7→ ξ ◦ η.
s ˙ Indeed, let t 7→ ξ(t) be a curve in D with ξ(0) = ξ, ξ(0) = X, then dξRη(X) =
d d dt |t=0(Rη(ξ(t))) = dt |t=0(ξ(t) ◦ η) = X ◦ η which is another right translation. However, Left multiplication
s s Lη : D → D
ξ 7→ η ◦ ξ
is only continuous and its tangent map is given by dξLη(X) = Dηξ · X. The manifold Ds(M) is not precisely a Lie group but has some similarities. If we were to work with D∞(M) instead of Ds(M) then we would have a genuine Lie group; however, we would no longer have a Banach manifold in which important theorems like the inverse function theorem hold, see [H].
11 If G is a Lie group and e ∈ G the identity element then the Lie algebra of G is
s s s identified with TeG. Similarly, TeD (M) = H (TM), which are H vector fields on M, serves as the Lie Algebra of Ds(M). Since right multiplication is smooth we are able to talk about right-invariant vector fields on Ds. Given any vector
s field v on M, we define a right-invariant vector field vη on D (M) by the formula
vη = v ◦ η.
Since these are fields of class C1 we are able to define the Lie bracket. The Lie bracket [uη, vη] is calculated as
[uη, vη](η) = dvη(uη) − duη(vη) = (dv ◦ η)(u ◦ η) − (du ◦ η)(v ◦ η)
= (dv · u − du · v) ◦ η = dRη [u, v] .
s s−1 s However, since u and v are both in H , the bracket is only H and hence TeD is not closed under the bracket operation.
2.1.3 Exponential Mappings
Let M be a closed, orientable manifold of dimension n and let D∞ be the
∞ group of smooth diffeomorphisms of M with Lie Algebra TeD (M) of smooth
∞ vector fields on M. For a vector field v ∈ TeD (M), it’s flow η(t): M → M is defined and is called a one-parameter subgroup of D∞(M). Therefore, the group exponential mapping
G ∞ ∞ exp : TeD → D
v 7→ η1
12 is defined. Here, η1 is the value of the one-parameter subgroup ηt = exp (tv) corresponding to t = 1. However, the group exponential mapping has a number of shortcomings. It is not even a homeomorphism in a neighborhood of the identity. There exist diffeomorphisms arbitrarily close to the identity which are not embeddable in a flow ([Ko]). This was shown for even the simplest case of M = S1. The group exponential map is in fact much worse! In the work of Grabowski ([G1]) it was shown that for the group of compactly supported diffeomorphisms of a Ck manifold (k = 0, 1, 2, ...) there exist arcwise-connected, non-trivial free subgroups of diffeomorphisms which embed in no flow. It is shown in [E-M] that, just like a Lie group, the group of Sobolev Hs dif- feomorphisms of a closed, orientable manifold M admits an exponential mapping which associates to every tangent vector at the identity a one parameter subgroup of Ds(M). Recall that such a tangent vector is an Hs vector field on M and the one parameter subgroup is the flow generated by the vector field.
n Theorem 2.1.1. Let M be a compact manifold without boundary, s > 2 + 2 and Ds(M) the group of Hs diffeomorphisms.
s 1. If V is an H vector field on M, its flow ηt is a one-parameter subgroup of Ds.
1 2. The curve t 7→ ηt is of class C .
s s 1 3. The mapping exp : TeD → D , V 7→ η1 is continuous (but not C ).
We refer the reader to [E-M] for the proof of this Theorem. Note that the exponen- tial mapping is not C1 because it does not cover any neighborhood of the identity. Since smooth diffeomorphisms embed densely in the set of Hs diffeomorphisms the
13 above results regarding local injectivity and surjectivity of the exponential map- ping apply to the exponential mapping of the Hs diffeomorphisms: there exist diffeomorphisms arbitrarily close to the identity which embed in no flow. More- over, there exist arcwise-connected, non-trivial free subgroup of diffeomorphisms which embed in no flow. The solution to this problem is to use an exponential map associated to a weaker metric defined on Ds(M). It is not automatic that a weak metric admit an exponential mapping but this turns out to be the case on Ds(M).
2.2 Weak Riemannian Structure on the Diffeomorphism Group
Let M be a closed, orientable manifold of dimension n, endowed with a Rie- mannian metric g and volume form dµ defined by g. The Riemannian metric
s s on M defines a weak Riemannian structure on D (M): Let η ∈ D (M) and uη,
s vη ∈ TηD (M), then,
Z −1 ∗ (uη, vη)L2 = g(u, v)(η ) dµ. (2.1) M
s This is a symmetric bilinear form defined on each tangent space TηD (M) whose norm is an L2-norm. The Riemannian structure is said to be weak because it generates the L2 topology on Ds(M) rather than the stronger Hs topology. Geodesics on a Hilbert manifold N locally minimize the L2-energy functional R t 2 0 kη˙(s)kN ds. The equations defining geodesics are then a second order system of differential equations η¨(t) = F (η, η˙) with F a vector field on TTN ∼= TN. According to the fundamental existence
14 and uniqueness theorem of ODE’s in Banach space, if F is smooth in both it’s arguments η andη ˙ then there exists a unique solution η(t) defined on an open interval around 0 ∈ R. It would then follow that around every point x ∈ N there exists a neighbourhood U of x and a number ε > 0 so that for each p ∈ U and each tangent vector v ∈ TpN with length less than ε there is a unique geodesic
η :(δ1, δ2) → N satisfying the initial conditions
η(0) = p η˙(0) = v.
Let v ∈ TpN be a tangent vector at a point p ∈ N and suppose there exists a geodesic η(t) : [0, 1] → N satisfying the initial conditions η(0) = p, η˙(0) = v. The point η(1) ∈ N will be denoted by
N expp (v) and called the exponential of the tangent vector v. The geodesic η(t) can then be described by
N η(t) = expp (tv)
N and expp is called the exponential map.
Theorem 2.2.1. ([E-M] Theorem 9.1) Let M be compact, without boundary,
s and endowed with a Riemannian metric g and in every tangent space TηD , s >
dim M 2 + 1, let the inner product be defined by (2.1). Then
1. (2.1) defines a weak Riemannian structure on Ds(M)
15 2. (2.1) has a unique torsion-free affine connection ∇¯ associated to it; that is, for smooth vector fields u, v and w on Ds(M), we have