Jump to ContentJump to Main Navigation
The Ambient Metric (AM-178)$

Charles Fefferman and C. Robin Graham

Print publication date: 2011

Print ISBN-13: 9780691153131

Published to Princeton Scholarship Online: October 2017

DOI: 10.23943/princeton/9780691153131.001.0001

Show Summary Details
Page of

PRINTED FROM PRINCETON SCHOLARSHIP ONLINE (www.princeton.universitypressscholarship.com). (c) Copyright Princeton University Press, 2021. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in PRSO for personal use. Subscriber: null; date: 23 January 2022

Jet Isomorphism

Jet Isomorphism

Chapter:
(p.82) Chapter Eight Jet Isomorphism
Source:
The Ambient Metric (AM-178)
Author(s):

Charles Fefferman

C. Robin Graham

Publisher:
Princeton University Press
DOI:10.23943/princeton/9780691153131.003.0008

Abstract and Keywords

A fundamental result in Riemannian geometry is the jet isomorphism theorem which asserts that at the origin in geodesic normal coordinates, the full Taylor expansion of the metric may be recovered from the iterated covariant derivatives of curvature. As a consequence, one deduces that any local invariant of Riemannian metrics has a universal expression in terms of the curvature tensor and its covariant derivatives. Geodesic normal coordinates are determined up to the orthogonal group, so problems involving local invariants are reduced to purely algebraic questions concerning invariants of the orthogonal group on tensors. This chapter proves an analogous jet isomorphism theorem for conformal geometry. By making conformal changes, the Taylor expansion of a metric in geodesic normal coordinates can be further simplified, resulting in a “conformal normal form” for metrics about a point.

Keywords:   jet isomorphism theorem, conformal geometry, ambient metric, Riemannian geometry, geodesic normal coordinates, Taylor expansion

A fundamental result in Riemannian geometry is the jet isomorphism theorem which asserts that at the origin in geodesic normal coordinates, the full Taylor expansion of the metric may be recovered from the iterated covariant derivatives of curvature. As a consequence, one deduces that any local invariant of Riemannian metrics has a universal expression in terms of the curvature tensor and its covariant derivatives. Geodesic normal coordinates are determined up to the orthogonal group, so problems involving local invariants are reduced to purely algebraic questions concerning invariants of the orthogonal group on tensors.

Our goal in this chapter is to prove an analogous jet isomorphism theorem for conformal geometry. By making conformal changes, the Taylor expansion of a metric in geodesic normal coordinates can be further simplified, resulting in a "conformal normal form" for metrics about a point. The jet isomorphism theorem states that the map from the Taylor coefficients of metrics in conformal normal form to the space of all conformal curvature tensors, realized in terms of covariant derivatives of ambient curvature, is an isomorphism. If n is even, the theorem holds only up to a finite order. In the conformal case, the role of the orthogonal group in Riemannian geometry is played by a parabolic subgroup of the conformal group. We assume throughout this chapter and the next that n ≥ 3.

We begin by reviewing the Riemannian theorem in the form we will use it. Fix a reference quadratic form hij of signature (p, q) on n . . In the positive definite case one typically chooses hij = δ‎ij.The background coordinates are geodesic normal coordinates for a metric gij defined near the origin in n if and only if gij (0)= hij and the radial vector field r satisfies r r =0, which is equivalent to Γ jk i x j x k =0 or

2 kgij x j x k = igjk x j x k .

It is easily seen that the coordinates are normal if and only if

g i j x j = h i j x j .

In fact, set F i = g i j x j h i j x j and observe that

kgij x j x k = x i j F i F i , igjk x j x k = i ( x j F j )2 F i .

(p.83) Clearly Fi = 0 implies that (8.1) holds so that the coordinates are normal. Conversely, if the coordinates are normal, then the fact that g and r are parallel along radial lines implies x j F j =0. Substituting this and (8.3) into (8.1) gives x j j F j =0, , which together with the fact that Fi is smooth and vanishes at the origin implies Fi = 0.

We are interested in the space of infinite order jets of metrics at the origin. Taylor expanding shows that (8.2) holds to infinite order if and only if gij (0)= hij and the derivatives of gij satisfy ( k 1 ... k r gi)j (0)=0. Since a 3-tensor symmetric in 2 indices and skew in 2 indices must vanish, this implies in particular that all first derivatives of g vanish at the origin. So we define the space N of jets of metrics in geodesic normal coordinates as follows.

Definition 8.1.

The space N is the set of lists (g(2),g(3), …), where for each r, g (r) 2 n r n satisfies g i ( j , k 1 k r ) ( r ) = 0 Here the comma just serves to separate the first two indices. For N2, N N will denote the set of truncated lists (g(2),g(3),…, g(N) ) with the same conditions on the g(r).

To a metric in geodesic normal coordinates near the origin, we associate the element of N given by g ij, k 1 ... k r (r) = k 1 ... k r g ij (0) for r ≥ 2. Conversely, to an element of N we associate the metric determined to infinite order by these relations together with gij (0)= hij and k gij (0)= 0. In the following, we typically identify the element of N and the jet of the metric g.

Definition 8.2.

The space R is the set of lists (R(0), R(1),…), where for each r, R (r) 2 R n 2 R n r R n , and the usual identities satisfied by covariant derivatives of curvature hold:

  1. (1) R i [ j k l ] , m 1 m r = 0 ;

  2. (2) R i j [ k l , m 1 ] m 2 m r = 0 ;

  3. (3) R i j k l , m 1 [ m s 1 m s ] m r = Q i j k l m 1 m r ( s ) .

    Here Q i j k l m 1 m r ( s ) denotes the quadratic polynomial in the R ( r ) with r'r – 2 which one obtains by covariantly differentiating the usual Ricci identity for commuting covariant derivatives, expanding the differentiations using the Leibnitz rule, and then setting equal to h the metric which contracts the two factors in each term.

For N0, R N will denote the set of truncated lists (R(0), R(1),…,R(N)) with the same conditions on the R(r).

There is a natural map NR induced by evaluation of the covariant derivatives of curvature of a metric, which is polynomial in the sense that (p.84) the corresponding truncated maps N N+2 R N are polynomial. We will say that a map on a subset of a finite-dimensional vector space is polynomial if it is the restriction of a polynomial map defined on the whole space, and a map between subsets of finite-dimensional vector spaces is a polynomial equivalence if it is a bijective polynomial map whose inverse is also polynomial. The jet isomorphism theorem for pseudo-Riemannian geometry is the following.

Theorem 8.3.

The map NR is bijective and for each Ν‎ ≥0 the truncated map N N+2 R N is a polynomial equivalence.

There are two parts to the proof. One is a linearization argument showing that it suffices to show that the linearized map is an isomorphism. The other is the observation that the linearized map is the direct sum over r of isomorphisms between two realizations according to different Young projectors of specific irreducible representations of GL(n,). The reduction to the linearization can be carried out in different ways; one is to argue as we do below in the conformai case. The analysis of the linearized map is contained in [E2]. See also [ABP] for a different construction of a left inverse of the map NR. .

We now consider the further freedom allowed by the possibility of making conformal changes to g.

Proposition 8.4.

Let g be a metric on a manifold M and let p ∈ M. Given Ω 0 C (M) with Ω 0 (p)>0, there is Ω C (M), uniquely determined to infinite order at p, such that Ω‎ − Ω‎0 vanishes to second order at p and Sym( r Ric)( Ω 2 g)(p)=0 for all r ≥ 0. Here Sym( r Ric)( Ω 2 g) denotes the full symmetrization of the rank r + 2 tensor ( r Ric)( Ω 2 g).

Proof.

Write Ω= e ϒ and set g ^ = e 2ϒ g. We are given ϒ(p) and dϒ(p). . Recall the transformation law for conformal change of Ricci curvature:

R ^ ij = R ij (n2) ϒ ij ϒ k k g ij +(n2)( ϒ i ϒ j ϒ k ϒ k g ij ).

Differentiating and conformally transforming the covariant derivative results in

( r Ric ^ ) ij, m 1 ... m r = ( r Ric) ij, m 1 ... m r (n2) ϒ ij m 1 ... m r ϒ k k m 1 ... m r g ij +lots,

where lots denotes terms involving at most r + 1 derivatives of ϒ. . We may replace covariant derivatives of ϒ by coordinate derivatives on the right-hand side of this formula. Then symmetrizing gives

Sym( r Ric ^ )=Sym( r Ric)(n2) r+2 ϒSym(tr( r+2 ϒ)g)+lots.

(p.85) If c > 0, the map ss+cSym(tr(s)g) is invertible on symmetric (r + 2)- tensors. Therefore, for any r ≥ 0, one can uniquely determine r+2 ϒ(p) to make Sym( r Ric ^ )(p)=0. The result follows upon iterating and applying Borel's Lemma.

We remark that there are other natural choices for normalizations of the conformal factor. For example, one such is that the symmetrized covariant derivatives of the tensor Pij vanish at p, where Pij is given by (3.7). Another is that in normal coordinates, det gij − 1 vanishes to infinite order at p (see [LP]).

Definition 8.5.

The space N c N of jets of metrics in conformal normal form is the subset of N consisting of jets of metrics in geodesic normal coordinates for which Sym( r Ric)(g)(0)=0 for all r ≥ 0. For Ν‎ ≥ 2, N c N will denote the subset of N N obtained by requiring that this relation hold for 0 ≤ rN − 2.

Later we will need the following consequence of the proof of Proposition 8.4.

Lemma 8.6.

For each Ν‎ ≥ 3, there is a polynomial map η N : N c N1 2 n N n with zero constant and linear terms, such that the map g(g, η N (g)) maps N c N1 N c N . Here g=( g (2) , g (3) ,..., g (N1) ) N c N1 .

Proof.

View g as the metric on a neighborhood of 0 n given by the prescribed finite Taylor expansion of order Ν‎ – 1. Then the components of Sym( N2 Ric)(g)(0) are polynomials in the g(r) with no constant or linear terms. According to the proof of Proposition 8.4, there is a function Ω‎ of the form Ω‎ = 1 + pΝ‎ with pΝ‎ a homogeneous polynomial of degree Ν‎ whose coefficients are polynomials in the g(r) with no constant or linear terms, such that Sym( N2 Ric)( Ω 2 g)(0)=0. Now Ω‎2g is in geodesic normal coordinates to order Ν‎ – 1 but not necessarily N. However by the construction of geodesic normal coordinates, there is a diffeomorphism ψ‎ = I + qN+1, where qN+1 is a vector-valued homogeneous polynomial of order N+l whose coefficients are linear in the order Ν‎ Taylor coefficient of Ω‎2g, such that ψ ( Ω 2 g) is in geodesic normal coordinates to order N. Since the condition Sym( r Ric)( Ω 2 g)(0)=0 is invariant under diffeomorphisms, η‎Ν‎ defined to be the order Ν‎ Taylor coefficient of ψ ( Ω 2 g) has the required properties.

If g1 and g2 are jets of metrics of signature (p, q) at the origin in n , , we say that g1 and g2 are equivalent if there is a local diffeomorphism ψ‎ defined near 0 satisfying ψ‎(0) = 0,and a positive smooth function Ω‎ defined near 0, so that g 2 = ψ ( Ω 2 g 1 ) to infinite order. It is clear from Proposition 8.4 (p.86) and the existence of geodesic normal coordinates that any jet of a metric is equivalent to one in Nc. In choosing Ω‎ we have the freedom of + × n , and in choosing ψ‎ a freedom of O(p,q). We next describe how these freedoms can be realized as an action on Nc of a subgroup of the conformai group.

Recall that hij is our fixed background quadratic form of signature (p,q) on n . . Define a quadratic form h ˜ IJ on ℝn+2 by

h ˜ IJ =( 0 0 1 0 h ij 0 1 0 0 )

and the quadric Q={[ x I ]: h ˜ IJ x I x J =0} n+1 . The metric h ˜ IJ d x I d x J on ℝn+2 induces a conformal structure of signature (p,q) on Q. The standard action of the orthogonal group O( h ˜ ) on ℝn+2 induces an action on Q by conformal transformations and the adjoint group O( h ˜ )/{±I} can be identified with the conformal group of all conformal transformations of Q.

Let e 0 =( 1 0 0 ) n+2 and let Ρ‎ be the image in O( h ˜ )/{±I} of the isotropy group {AO( h ˜ ):A e 0 =a e 0 ,a}of[ e 0 ]. It is clear that each element of Ρ‎ is represented by exactly one A for which a > 0, so we make the identification P={pO( h ˜ ):p e 0 =a e 0 ,a>0}. The first column of pP is ( a 0 0 ); combining this with the fact that pO( h ˜ ), one finds that

P={ ( a b j c 0 m i j d i 0 0 a 1 ):a>0, m i j O(h),c= 1 2a b j b j , d i = 1 a m ij b j }

where hij is used to raise and lower lowercase indices. It is evident that Ρ‎ = ℝ+ · ℝn · Ο‎(h), where the subgroups ℝ+, ℝn, Ο‎ (h) arise by varying a, bj, mij, resp.

The intersection of Q with the cell {[ x I ]: x 0 0} can be identified with ℝn via n x i [ 1 x i 1 2 |x | 2 ]Q, where |x | 2 = h ij x i x j . In this identification, the conformal structure is represented by the metric h i j d x i d x j on ℝn. If pΡ‎ is as above, the conformal transformation determined by p is realized as

( φ p ( x ) ) i = m i j x j 1 2 | x | 2 d i a + b j x j 1 2 c | x | 2

and one has φ p h = Ω p 2 h for

Ω p = ( a + b j x j 1 2 c | x | 2 ) 1 .

This motivates the following definition of an action of Ρ‎ on N c . Given (p.87) p P as above and g N c , by Proposition 8.4 there is a positive smooth function Ω‎ uniquely determined to infinite order at 0 so that Ω‎ agrees with Ω‎p to second order and such that Sym( r Ric)( Ω 2 g)(0)=0 for all r > 0. Now ( Ω 2 g)(0)= a 2 h, so by the construction of geodesic normal coordinates, there is a diffeomorphism φ‎, uniquely determined to infinite order at 0, so that φ(0)=0,φ'(0)= a 1 m, and such that ( φ 1 ) ( Ω 2 g) is in geodesic normal coordinates to infinite order. We define p.g= ( φ 1 ) ( Ω 2 g). . It is clear by construction of φ‎ that p.gN, and since the condition of vanishing of the symmetrized covariant derivatives of Ricci curvature is diffeomorphism-invariant, it follows by construction of Ω‎ that p.g N c . It is straightforward to check that this defines a left action of Ρ‎ on N c . Note that if g = h, then Ω= Ω p and φ = φ p so that ft is a fixed point of the action. A moment's thought shows that (p.g)(r) depends only on g(s) for s < r. Therefore for each N2, there is an induced action on N c N .

It is clear from the construction of the action that p.g is equivalent to g for all g N c and p P In fact, the P-orbits are exactly the equivalence classes.

Proposition 8.7.

The orbits of the Ρ‎-action on N c are precisely the equivalence classes of jets of metrics in N c under diffeomorphism and conformai change.

Proof. It remains to show that equivalent jets of metrics in N c in the same P-orbit. Suppose that g 1 , g 2 N c are equivalent. Then we can write g 2 = ( φ 1 ) ( Ω 2 g 1 ) to infinite order for a diffeomorphism φ‎ with φ ( 0 ) = 0 and a positive smooth Ω‎. We can uniquely choose the parameters a and b of p P so that Ω Ω p vanishes to second order. Since g1 and g2 both equal h at 0, it follows that a φ (0)O(h), so that we can write φ (0)= a 1 m with mO(h). Together with the already determined parameters a and b, this choice of m uniquely determines a p P Since g 2 N c , all symmetrized covariant derivatives of Ricci curvature of Ω 2 g 1 vanish at the origin, so Ω‎ must be the conformai factor determined when constructing the action of ρ‎ on g1. And since g1 is in geodesic normal coordinates, φ‎ must be the correct diffeomorphism, so that g2 =p.g1.

It is straightforward to calculate from the definition the action on N c of the ℝ+ and O(h) subgroups of P. If we denote by pa the element of P obtained by taking b = 0 and m = I and by pm the element given by a =1 and b = 0, then one finds that paacts by multiplying g(r) by ar and pmacts by transforming each g(r) as an element of r+2 n , where ℝn denotes the standard defining representation of Ο‎(h).

(p.88) The problem of understanding local invariants of metrics under diffeomorphism and conformal change reduces to understanding this action of Ρ‎ on NC. However, it is very difficult to analyze or even concretely exhibit the action of the ℝn part of Ρ‎ directly from the definition. The ambient curvature tensors enable the reformulation of the action in terms of standard tensor representations of P.

Propositions 6.1 and 6.4 show that the covariant derivatives of curvature of an ambient metric g ˜ satisfy relations arising from the homogeneity and Ricci-flatness of the metric. These conditions suggest the following definition.

Definition 8.8.The space R ˜ is the set of lists ( R ˜ (0) , R ˜ (1) ,...), such that R ˜ (r) 2 n+2 2 n+2 r n+2 , and such that the following relations hold:

  1. (1) R ˜ I[JKL], M 1 M r =0;

  2. (2) R ˜ IJ[KL, M 1 ] M 2 M r =0;

  3. (3) h ˜ IK R ˜ IJKL, M 1 M r =0;

  4. (4) R ˜ IJKL, M 1 [ M s1 M s ] M r = Q ˜ IJKL M 1 M r (s) ;

  5. (5) R ˜ IJK0, M 1 M r = s=1 r R ˜ IJK M s , M 1 M s ^ ... M r ;

  6. (6) R ˜ IJKL, M 1 M s 0 M s+1 M r = (s+2) R ˜ IJKL, M 1 M r t=s+1 r R ˜ IJKL, M 1 M s M t M s+1 M t ^ M r .

Here, as in Definition 8.2, Q ˜ IJKL M 1 M r (s) denotes the quadratic polynomial in the components of R ˜ ( r ) for r r2 which one obtains by differentiating the Ricci identity for commuting covariant derivatives, expanding the differentiations using the Leibnitz rule, and then setting equal to h ˜ the metric which contracts the two factors in each term. Condition (5) in case r = 0 is interpreted as R ˜ IJK0 =0.

We remark, as we did in the proof of Proposition 6.1, that condition (6) is superfluous: it is a consequence of (2), (4), (5). But we

will not use this fact.

It will be convenient also to introduce the vector space τ ˜ consisting of the set of lists ( R ˜ (0) , R ˜ (1) ,) with R ˜ (r) 2 n+2 2 n+2 r n+2 such that (l)–(3), (5), (6) hold.

We prepare to define truncated spaces R ˜ N for R ˜ . Recall the notion of

strength from Définition 6.3. Note that it is clear that for each of the relations (l)–(6) except (4) in Definition 8.8, all components R ˜ IJKL, M 1 M r which occur in the relation have the same value for the strength of the index list (p.89) of the component. For Ν‎ > 0, define the following vector spaces of lists of components of tensors. We denote by a list of indices of length |M|=r, and the conditions (l)–(6) refer to Definition 8.8.

τ ˜ N ={ ( R ˜ IJKL,M ) |M|0,IJKLMN+4 : (1)(3), (5), (6) hold} τ ˜ N ={ ( R ˜ IJKL,M ) |M|0,IJKLM=N+4 : (1)(3), (5), (6) hold}

If is a component appearing in an element of τ ˜ N and r > N, then at least one of the indices IJKLM1 ... Mr must be 0. Therefore by (5) and (6), R ˜ IJKL, M 1 M r can be written as a linear combination of

components with r replaced by r – 1. It follows that τ ˜ N and τ ˜ N

are finite-dimensional. Since (l)–(3), (5) and (6) imply that R ˜ IJKL, M 1 M r =0 if IJKL M 1 M r 3, we have T ˜ N = M=0 N T ˜ M and T ˜ = M=0 T ˜ M . As for (4), a typical term in Q ˜ IJKL M 1 M r (s) is

h ˜ AB R ˜ AI M s1 M s , M R ˜ BJKL,M L s2 M ,

where M4' and M4" are lists of indices such that M M is a rearrangement of M s+1 M r . In order that h ˜ AB 0, it must be that AB=2. There-fore AI M s1 M s M +BJKL M 1 M s2 M =IJKL M 1 M r +2.

This implies that if either AI M s1 M s M orBJKL M 1 M s2 M is greater than or equal to IJKL M 1 M r , then the other is less than

or equal to 2. The same reasoning applies to all terms in Q ˜ IJKL M 1 M r (s) . Since (l)–(3), (5) and (6) imply that a component of an R ˜ (t) vanishes if its

strength is at most 3, it follows that any component of an R ˜ (t) which occurs in Q ˜ IJKL M 1 M r (s) with a nonzero coefficient must have strength strictly less IJKL M 1 M r . Hence we will regard (4) as a relation involving components with indices of strength at most IJKL M 1 M r , and the quadratic terms in (4) only involve components of strength less than that of the linear terms. With this understanding, we can now define

R ˜ N ={( R ˜ IJKL, M 1 M r ) τ ˜ N : (4) of Definition 8.8 holds}.

We will also need the corresponding linearized spaces, in which the term Q ˜ IJKL M 1 M r (s) in (4) is replaced by 0. Define vector spaces

T R ˜ ={( R ˜ (0) , R ˜ (1) ,...) τ ˜ : R ˜ IJKL, M 1 ... M r = R ˜ IJKL,( M 1 ... M r ) }

T R ˜ N ={( R ˜ IJKL, M 1 ... M r ) τ ˜ N : R ˜ IJKL, M 1 ... M r = R ˜ IJKL,( M 1 ... M r ) }

σ ˜ N ={( R ˜ IJKL, M 1 ... M r ) τ ˜ N : R ˜ IJKL, M 1 ... M r = R ˜ IJKL,( M 1 ... M r ) }.

Then T R ˜ N = M=0 N σ ˜ M andT R ˜ = M=0 σ ˜ M . Note that in the presence of the condition R ˜ IJKL, M 1 M r = R ˜ IJKL,( M 1 M r ) , condition (6) of Definition 8.8 becomes

R ˜ IJKL, M 1 M r 0 =(r+2) R ˜ IJKL, M 1 M r

(p.90)

and is a consequence of (2) and (5). Note also that if ( R ˜ IJKL, M 1 M r ) σ ˜ N ,

then h ˜ L M s R ˜ IJKL, M 1 M r =0 and h ˜ M s M t R ˜ IJKL, M 1 M r =0.

For w, Let σ w :P denote the character Since Ρ‎ C Ο‎ (h) and Ρ‎ preserves e0 up to scale, it is easily seen that and T ˜ , R ˜ andT R ˜ are invariant subsets of the P-space r=0 ( 4+r n+2 σ 2r ) , where

Rn+22 denotes the standard representation of PGL(n+2,). These inclusions therefore define actions of Ρ‎ on these spaces. These actions of Ρ‎ do not preserve strength, but because Ρ‎ consists of block upper-triangular ma-trices,a component of p.(eR) depends only on components of R of greater trices, a component of p. depends only on strength. So for there are also actions of Ρ‎ on T ˜ N , R ˜ N and '

An easy computation shows that the element acts by multiplying a

component of strength S by a S2 .

We next define our main object of interest. If g is a metric defined in a neighborhood of 0 n , we construct a straight ambient metric in normal form for g as in Chapter 3. We then evaluate the covariant derivatives of curvature of the ambient metric at ρ=0,

t = 1 as described in Chapter 6. If n is odd, the values of all components of these covariant derivatives at the origin

depend only on the derivatives of g at the origin, while if η‎ is even, this is true for components of strength at most n+ 1 by Proposition 6.2. If gij(0)=hij,then g ˜ IJ = h ˜ IJ atρ=0,t=1,x=0. In this case, Propositions 6.1 and 6.4 show that the resulting lists of components satisfy the relations of Definition 8.8. This procedure therefore defines a map c : for n odd, and c: N c R ˜ n3

for n even. Since the conformai curvature tensors are natural polynomial invariants of the metric g, c is a polynomial map.

Proposition 8.9.

The map c: N c R ˜ (or R ˜ n3 if n is even)is equivariant with respect to the P-actions.

Proof.

Recall that the action of Ρ‎ on Nc is given by p.g= ( φ 1 ) ( Ω 2 g), where φ‎ and Ω‎ are determined to map g back to conformai normal form given the initial normalizations defined by p; see the discussion preceding Proposition 8.7 above. By naturality of the conformai curvature tensors,

c( ( φ 1 ) ( Ω 2 g))= ( φ (0) 1 ) (c( Ω 2 g)), where ( φ (0) 1 ) on the right-hand side is interpreted as the pullback in the indices between 1 and n of each of the tensors in the list, leaving the 0 and ∞ indices alone. And c( Ω 2 g) is given by Proposition 6.5. We use these observations to check for each of the generating subgroups O(h), ℝ++ and ℝn of Ρ‎ that where

the P-action on the right-hand side is that on r=0 ( 4+r n+2 σ 2r ).

For p = pm, we have Ω = 1 and φ(x)=mx,soc( p m .g)= ( m 1 ) c(g) is obtained from c(g) by transforming covariantly

underΟ‎(h) the indices

between 1 and n. But this is precisely how pm acts on r=0 4+r n+2 .

(p.91) For P = pa, we have φ(x)= a 1 x and Ω= a 2 By Proposition 6.5, the component R ˜ IJKL, M 1 M r

for a-2g is that for g multiplied by a 2 s 2 Since ( φ (0) 1 ) * acts by multiplying this component by a s M it follows that the

components of c(pa.g) are those of c(g) multiplied by a s M +2 s 2 = a S2 But we noted above that this is precisely how pa acts on r=0 ( 4+r n+2 * σ 2r ) . Finally, for P = pb, we have φ‎'(0) = I and Ω=1 b i x i +O(|x | 2 ), so that the components of c( ( φ 1 ) * ( Ω 2 g)) are given by Proposition 6.5 with ϒ i = b i . But this is precisely how pb, acts on r=0 4+r n+2 * .

Let us examine more carefully the equivariance of c with respect to the subgroup + P The component R ˜ IJKL, M 1 M r of c(g) is a polynomial in the components of the g(s) and this equivariance says that when the g(s) are replaced by as(s) then R ˜ IJKL, M 1 M r

is multiplied by a S2 with S= IJKL M 1 M r In particular, R ˜ IJKL, M 1 M r can only involve for g(s) s >≤ S – 2. This implies that for each N ≥ 0 (satisying also N ≤ n-3 for n even), c induces a P-equivariant polynomial map c N : N c N+2 R ˜ N . Clearly these induced maps satisfy the compatibility conditions c N1 π N+2 = π ˜ N c N , where π N : N c N   N c N1 and π ˜ N : R ˜ N   R ˜ N1 are the natural projections. The main result of this chapter is the following jet isomorphism theorem.

Theorem 8.10. Let N ≤ 0 and assume that if N ≤ n-3is even. Then R ˜ N is a smooth submanifold of T ˜ N whose tangent space at 0 is T R ˜ N and the map C N : N c N+2 R ˜ N is a P-equivariant polynomial equivalence.

For n odd, it follows that c: N c   R ˜ is a P-equivariant bijection since c is the projective limit of the cN.

It will be convenient in the proof of Theorem 8.10 to use Theorem 8.3 to realize Nc in terms of curvature tensors on ℝn rather than Taylor coefficients of metrics. So we make the following definition. Recall the space R introduced in Definition 8.2.

Definition 8.11.

Define the space R c  R to be the subset consisting of lists of tensors (R(0) R(1),…) satisfying in addition to the conditions in Definition 8.2 the following: for each r0

Sym( h ik R ijkl, m 1 m r (r) )=0.

Here Sym refers to the symmetrization over the free indices jl m 1 m r For N0 by R c N we will denote the corresponding set of truncated lists ( R (0) , R (1) ,, R (N) )

For r,N0 we define also the following finite-dimensional vector spaces:

τ r ={ R (x) Λ 2 n* Λ 2 n* r n* :

(p.92)

(8.5) and (1), (2) of Definition 8.2 hold}

σ r ={ R (r) τ τ : R ijkl, m 1 ,..., m r = R ijkl,( m 1 ,..., m r ) }

T N= r=0 N T τ

T R c N = r=0 N σ τ T N .

The bijection N R asserted by Theorem 8.3 clearly restricts to a bijection N c R   c whose truncated maps N c N+2   R c N are polynomial equivalences. By composition, we can regard c and the cN as defined on the corresponding Rcand RNc In the following, we will not have occasion to refer to N c and N c N+2 so no confusion should arise from henceforth using thesame symbols c and cN for the maps defined on Rc and R c N We can transfer the action of P on N c to Rc. The elementpaacts on Rc by multiplying R (r) by a r+2 ; this same prescription gives an ℝ+ action on TN for N ≥ 0. The ℝ+-equivariance of c N implies that a component R ˜ IJKL, M 1 M r of CN(R) with IJKL M 1 M r =N+4 can be written as a linear combination of R(N) plus quadratic and higher terms in the components of the R(r) with rN2

Our starting point for the proof of Theorem 8.10 is the following lemma.

Lemma 8.12.

For each N ≥ 0, the subset R c N T N is a smooth submanifold whose tangent space at 0 is T R c N .

Proof.

We will show that for each N ≥ 1, there is a polynomial equivalence Φ N : T N T N satisfying d Φ N (0)=I and Φ N ( R c N1 × σ N )=  R c N Upon iterating this statement and using σ 0 =  R c 0 =T R c 0 we conclude the existence of a polynomial equivalence : T N   T N whose differential at 0 is the identity, and which maps T R c N   R c N The desired conclusion follows immediately.

When reformulated in terms of the spaces R c N , Lemma 8.6 asserts the existence for each N ≥ 1 of a polynomial map η N : T N1 N+4 n * with zero constant and linear term, such that the map Λ N :R(R, η N (R)) sends R c N1   R c N . Here R denotes the list constituting an element of TN-l. There is no loss of generality in assuming that η N ( T N1 ) τ N so that ɅN : T N1   T N. Define Φ N : T N   T N by Φ N (R, R (N) )=(R, R (N) + η N (R)) It is evident from the form of the relations defining RNc that Φ N ( R c N1 × σ N )=  R c N , and clearly ( Φ N ) 1 (R, R (N) )=(R, R (N) η N (R))

We remark that the same proof could have been carried out in terms of the spaces of normal form coefficients, and shows that the subset. N c N   N N is a smooth submanifold whose tangent space at 0 is obtained by linearizing the (p.93) equation obtained by writing (8.5) in terms of the normal form coefficients

g ij, k 1 k r . At this point we do not know that R ˜ N is a submanifold of T ˜ N with tangent space T R ˜ N but it is clear that the tangent vector at 0 to a smooth curve in R N must lie in T R ˜ N So we conclude for the differential of cN at the origin that d c N :T R c N T R ˜ N The differentiation of the action of Ρ‎ on R c N gives a linear action of Ρ‎ on T R c N and d c N :T R c N T R ˜ N is P-equivariant. By R+-equivariance and linearity of d c N … , it follows that dcN decomposes as a direct sum of maps σ M   σ ˜ M for 0≤M≤NBy the compatibility of the cN as Ν‎ varies, the map σ M σ ˜   M is independent of the choice of Ν‎ > M, so we may as well denote it as d c N : σ N   σ ˜ N The main algebraic fact on which rests the proof of Theorem 8.10 is the following.

Proposition 8.13.

For Ν‎ > 0 (and Ν‎ < η‎ –3 if η‎ is even), d c N : σ N   σ ˜ N is an isomorphism.

Proof of Theorem 8.10 using Proposition 8.13.

We prove by induction on N that there is a polynomial equivalence Φ ˜ N : T ˜ N   T ˜ N satisfying d Φ ˜ N (0)=I and Φ ˜ N ( R ˜ N1 ×  σ ˜ N )=  R ˜ N and that c N : R c N   R ˜ N is a polynomial equivalence. Just as in the proof of Lemma 8.12, iterating the first statement provides a polynomial equivalence : : T ˜ N   T ˜ N whose differential at 0 is the identity and which maps T R ˜ N   R ˜ N from which follows the first statement of Theorem 8.10.

For Ν‎ = 0, we can take Φ ˜ N to be the identity. Since R c 0 =T R c 0 = σ 0 , R 0 =T R 0 =  σ ˜ 0 , and c° is linear and can be identified with d c 0 the second statement is immediate from Proposition 8.13.

Suppose for some Ν‎ > 1 that we have the polynomial equivalence Φ ˜ N1

and we know that cN-1 is a polynomial equivalence. Recall the polynomial maps η N : T N1 τ N , Λ N : T N1 T N and Φ N : T N   T N constructed in Lemma 8.6 and Lemma 8.12. By the induction hypothesis that cN-1 is a polynomial equivalence, we conclude that there is a polynomial map Λ ˜ N : T ˜ N1   T ˜ N such that Λ ˜ N ( R ˜ N1 )  R ˜ N and such that the diagram

Jet Isomorphism

commutes. Using the compatibility of c N1 and cN and the form of the map Λ N , one sees that Λ ˜ N can be taken to have the form Λ ˜ N ( R ˜ )=( R ˜ , η ˜ N ( R ˜ ))

where η ˜ N : T ˜ N1   τ ˜ N has no constant or linear terms. Now define the map Φ ˜ N : T ˜ N   T ˜ N by Φ ˜ N ( R ˜ , R ˜ (N) )=( R ˜ , R ˜ (N) + η ˜ N ( R ˜ )) Clearly d Φ ˜ N (0)=I (p.94) and Φ ˜ N ( R ˜ N1 ×  σ ˜ N )=  R ˜ N by the form of the relations defining R ˜ N It is a straightforward matter to check that the diagram

Jet Isomorphism

commutes. By the induction hypothesis and Proposition 8.13, the vertical map on the left is a polynomial equivalence. We conclude that cN is also a polynomial equivalence, completing the induction step.

Proof of Proposition 8.13.

The proof has two parts. We will first construct an injective map L: σ ˜ N σ N . We will then show that d c N : σ N   σ ˜ N is injective. These statements together imply that dim( σ N )=dim( σ ˜ N ). from which it then follows that d c N is an isomorphsim.

Let ( R ˜ IJKL, M 1 M r )  σ ˜ N . We can consider the components R ˜ ijkl, m 1 m N in which all the indices lie between 1 and n. This defines a map L: σ ˜ N Λ 2 n * Λ 2 n * N n * and clearly everything in the range of L satisfies conditions (1) and (2) of Definition 8.2. We claim that everything in the range of L also satisfies (8.5), so that L: σ ˜ N σ N Condition (3) of Definition 8.8 implies h ˜ IK R ˜ IjKl, m 1 m N =0 This can be written as

R ˜ jl0, m 1 m N + R ˜ lj0, m 1 m N + h ik R ˜ ijkl, m 1 m N =0.

If we apply condition (5) of Definition 8.8 to R ˜ jl0, m 1 m N and then symmetrize over jlm1...mN, the result is 0 by the skew symmetry of R ˜ (N) in the second pair of indices. Similarly for R ˜ lj0, m 1 m N It follows that the symmetrization of vanishes. This proves that (8.5) holds.

Next we show that L is injective. We claim that for ( R ˜ IJKL, M 1 M r )  σ ˜ N any component R ˜ IJKL, M 1 M r can be written as a linear combination of components in which none of the indices IJKLM1 ... M r is ∞. We first show that any component can be written as a linear combination of components in which none of IJKL is ∞. To see this, note that (8.4) and

h ˜ AB R ˜ IJKA,B M 1 M r =0 imply

(r+2) R ˜ IJK, M 1 M r = R ˜ IJK,0 M 1 M r = R ˜ IJK0, M 1 M r h ab R ˜ IJKa,b M 1 M r .

Thus a component R ˜ IJKL, M 1 M r in which L= can be rewritten as a linear combination of components in which L and UK remain unchanged. Repeating this procedure allows the removal of any ∞'s in IJKL. (p.95) The same method allows the removal of ∞'s in M\ ... M h ˜ AB R ˜ IJKL,AB M 2 M r =0 one has

2(r+2) R ˜ IJKL, M 2 M r =2 R ˜ IJKL,0 M 2 M r = h ab R ˜ IJKL,ab M 2 M r .

Thus all oo's can be removed as indices. Now (8.4) and (5) of Definition 8.8 can be used to remove any O's as indices at the expense of permuting the remaining indices between 1 and n. It follows that any component R ˜ IJKL, M 1 M r can be written as a linear combination of components in which all indices are between 1 and n. Thus L is injective.

It remains to prove that d c N is injective. If R ijkl, m 1 m N σ N . we set R jl, m 1 m N = h ik R ijkl, m 1 m N , R m 1 m N = h jl R jl, m 1 m N and

(n2) P jk, m 1 m N = R jk, m 1 m N R m 1 m N h jk /2(n1).

We also denote by W ijkl, m 1 m N Λ 2 n * Λ 2 n * N n * the tensor obtained by taking the trace-free part in ijkl while ignoring m1 ... mn:

W ijkl, m 1 m N = R ijkl, m 1 m N 2 h l[j P i]k, m 1 m N h jl +2 h k[j P i]l, m 1 m N

Then Wjjfc;imimjv satisfies (1) of Definition 8.2 but not necessarily (2). We also define

C jkl, m 1 m N1 =2 P j[k,l] m 1 m N1 .

Contracting the second Bianchi identity (2) of Definition 8.2 in the usual way shows that W i jkl,i m 1 m N1 =(3n) C jkl, m 1 m N1 and P i j,i m 1 m N1 = P i i,j m 1 m N1 .

Lemma 8.14. Let R ijkl, m 1 m N σ N If n ≥ 4 and W ijkl, m 1 m N =0 then R ijkl, m 1 m N =0 If n = 3 and C jkl, m 1 m N1 =0 then R ijkl, m 1 m N =0

Proof. If Ν‎ = 0, this follows from the decomposition of the curvature tensor into its Weyl piece and its Ricci piece. Suppose N ≥ 1. If n ≥ 4, the contracted Bianchi identity above shows that C jkl, m 1 m N1 =0 which is ourhypothesis if n = 3. (The hypothesis Wijkl,m1...mn = 0 for n ≥ 4 is of course automatic for η‎ = 3.) Thus we conclude for any η‎ that Pj[k,l]m1...mn-1 = 0, so P ij, m 1 m N = P (ij, m 1 m N ) Since R ijkl, m 1 m N σ N we also have

R(ij,m1...mN) = 0. Therefore

(n2) P ij, m 1 m N =(n2) P (ij, m 1 m N ) = R ( m 1 m N h ij) /2(n1).

Now h ij P (ij, m 1 m N ) = h ij P ij, m 1 m N = R m 1 m N /2(n1) Hence the symmetric tensor P= P (ij, m 1 m N ) is in the kernel of the operator Ρ‎ —» (η‎ – 2)Ρ‎ + Sym(tr(P)/i). This operator is injective on symmetric tensors, so we conclude that Pijtmi---mN = 0· The conclusion now follows from (8.6). D

(p.96) We will prove that d c N is injective by showing that if R ijkl, m 1 m N

ker(dcN), then the hypotheses of Lemma 8.14 hold. To get the flavor of the argument, consider first the cases Ν‎ = 0,1. Now σ 0 is the space of trace-free curvature tensors. In Chapter 6, we found that is the Weyl piece of such a curvature tensor. So co linear and is obviously injective. In Chapter 6, we also calculated the curvature components R ˜ jkl = C jkl and (see (6.3)). We see that c1 is also linear, so can be identified with dc11. If R ijkl,m ker(d c 1 ) we first conclude by considering R ˜ jkl that C jkl =0 and then by considering R ˜ ijkl,m that W ijkl,m =0 (for n > 4), as desired.

For the general case we need to understand the relation between covariant derivatives with respect to the ambient metric and covariant derivatives with respect to a representative g on M. Recall that the conformai curvature tensors are tensors on M defined by evaluating components of R ˜ IJKL, M 1 M r at ρ‎ = 0 and t = 1. We can take further covariant derivatives of such a tensor with respect to g. We will denote by the tensor on M obtained by such further covariant differentiations. For example, R ˜ ijkl,|m = W ijkl,m whereas R ˜ ijkl,m is the tensor V ijklm given by (6.3). An inspection of (3.16) shows the relation between R ˜ IJKL, M 1 M r p and R ˜ IJKL, M 1 M r |p Recalling that g ij =2 P ij ρ=0 at one sees that R ˜ IJKL, M 1 M r p R ˜ IJKL, M 1 M r |p is a linear combination of components of

Vr R, possibly multiplied by components of g, plus quadratic terms in curvature. Iterating, it follows that R ˜ IJKL, M 1 M r p 1 p s R ˜ IJKL, M 1 M r | p 1 p s is a linear combination of terms of the form R ˜ ABCD, F 1 F u | q 1 ... q t with t < s, possibly multiplied by components of g, plus nonlinear terms in curvature.

Consider now the map d c N : σ N   σ ˜ N The symbol R ˜ IJKL, M 1 M r with IJKL M 1 M r =N+4 is now to be interpreted as the linear function of the R ijkl, m 1 m N obtained by applying d c N Similarly, we now interpret the symbol R ˜ IJKL, M 1 M r | p 1 p s for IJKL M 1 M r +s=N+4 as a linear function of the R ijkl, m 1 m N Suppose that R ijkl, m 1 m N ker(d c N ) We claim that R ˜ IJKL, M 1 M r | p 1 p s =0 for IJKL M 1 M r +s=N+4 The proof is by induction on s. For s = 0, this is just the hypothesis that R ijkl, m 1 m N ker(d c N ) Since we are considering the linearization dcN, the quadratic terms may be ignored in the relation derived in the previous paragraph between ambient covariant derivatives and covariant derivatives on M. A moment's thought shows that this relation provides the induction step to increase s by 1.

Taking r = 0, we conclude that R ˜ IJKL,| m 1 m s =0 For IJKL+s=N+4 For IJKL = ijkl we obtain W ijkl, m 1 m N =0 and for IJKL = ∞jkl we obtain C jkl, m 1 m N1 =0 Lemma 8.14 then shows that . R ijkl, m 1 m N =0 as desired.