Conformal Einstein spaces and Bach tensor generalizations in n dimensions

Transcription

1 Linköping Studies in Science and Technology. Theses No Conformal Einstein spaces and Bach tensor generalizations in n dimensions Jonas Bergman Matematiska institutionen Linköpings universitet, SE Linköping, Sweden Linköping 004

2 ii Conformal Einstein spaces and Bach tensor generalizations in n dimensions c 004 Jonas Bergman Matematiska institutionen Linköpings universitet SE Linköping, Sweden jober@mai.liu.se LiU-TEK-LIC-004:4 ISBN ISSN Printed by UniTryck, Linköping 004

3 iii Abstract In this thesis we investigate necessary and sufficient conditions for an n- dimensional space, n 4, to be locally conformal to an Einstein space. After reviewing the classical results derived in tensors we consider the four-dimensional spinor result of Kozameh, Newman and Tod. The involvement of the four-dimensional Bach tensor which is divergence-free and conformally well-behaved in their result motivates a search for an n-dimensional generalization of the Bach tensor B ab with the same properties. We strengthen a theorem due to Belfagón and Jaén and give a basis U ab, V ab and W ab for all n-dimensional symmetric, divergence-free - index tensors quadratic in the Riemann curvature tensor. We discover the simple relationship B ab = 1 U ab V ab and show that the Bach tensor is the unique tensor with these properties in four dimensions. Unfortunately we have to conclude, in general that there is no direct analogue in higher dimension with all these properties. Nevertheless, we are able to generalize the four-dimensional results due to Kozameh, Newman and Tod to n dimensions. We show that a generic space is conformal to an Einstein space if and only if there exists a vector field satisfying two conditions. The explicit use of dimensionally dependent identities some of which are newly derived in this thesis is also exploited in order to make the two conditions as simple as possible; explicit examples are given in five and six dimensions using these tensor identities. For n dimensions, we define the tensors b abc and B ab, and we show that their vanishing is a conformal invariant property which guarantees that the space with non-degenerate Weyl tensor is a conformal Einstein space. Acknowledgments First and foremost, I would like to thank my supervisors Brian Edgar and Magnus Herberthson for their constant support, encouragement, and generous knowledge sharing, and for all the interesting discussions, and for giving me this chance to work on such an interesting topic. Thanks also goes to all my friends and colleagues at the Department of Mathematics, and especially Arne Enqvist for his support. I would also like to mention Anders Höglund, who let me use his fantastic program Tensign, and Ingemar and Göran, who read the manuscript and gave me valuable comments. Thanks guys! Finally, but not least, I would like to thank Pauline and my family for their support, encouragement, and understanding and for putting up with me, especially during the last month. Jonas Bergman, Linköping, 0 September 004

4 iv

5 Contents Abstract and Acknowledgments Contents iii v 1 Introduction and outline of the thesis 1 Preliminaries 5.1 Conventions and notation Conformal transformations Conformally flat spaces Conformal Einstein equations and classical results Einstein spaces Conformal Einstein spaces The classical results The Bach tensor in four dimensions and possible generalizations The Bach tensor in four dimensions Attempts to find an n-dimensional Bach tensor The tensors U ab, V ab and W ab Four-dimensional Bach tensor expressed in U ab, V ab and W ab An n-dimensional tensor expressed in U ab, V ab and W ab The Kozameh-Newman-Tod four-dimensional result and the Bach tensor Two useful lemmas C-spaces and conformal C-spaces Conformal Einstein spaces J = Listing s result in four dimensions Non-degenerate Weyl tensor Conformal C-spaces Conformal Einstein spaces

6 vi 7 Listing s result in n dimensions Non-degenerate Weyl tensor Conformal C-spaces Conformal Einstein spaces Edgar s result in n dimensions Using the Cayley-Hamilton Theorem Four dimensions Higher dimensions Using dimensionally dependent identities A five-dimensional example Six-dimensional examples Generalizing the Bach tensor in n dimensions A generic Weyl tensor The generalization of the KNT result n dimensions using generic results Five-dimensional spaces using dimensional dependent identities Six-dimensional spaces using dimensional dependent identities Conformal properties of different tensors The tensors b abc and B ac and their conformal properties in generic spaces The tensor L ab and its conformal properties Concluding remarks and future work 67 A The Cayley-Hamilton Theorem and the translation of the Weyl tensor/spinor to a matrix 69 A.1 The Cayley-Hamilton Theorem The case where n = 3 and the matrix is trace-free The case where n = 6 and the matrix is trace-free A. Translation of C ab cd to a matrix C A B A.3 Translation of Ψ AB CD to a matrix Ψ B Dimensionally dependent tensor identities 74 B.1 Four-dimensional identities B. Five-dimensional identities B.3 Six-dimensional identities B.4 Lovelock s quartic six-dimensional identity C Weyl scalar invariants 79 C.1 Weyl scalar invariants in 4 dimensions

7 vii D Computer tools 84 D.1 GRTensor II D. Tensign References 87

8 viii

9 Chapter 1 Introduction and outline of the thesis Within semi-riemannian geometry there are classes of spaces which have special significance from geometrical and/or physical viewpoints; e.g., flat spaces with zero Riemann curvature tensor, conformally flat spaces i.e., spaces conformal to flat spaces with Weyl tensor equal to zero, Einstein spaces with trace-free Ricci tensor equal to zero. There are a number of both physical and geometrical reasons to study conformally Einstein spaces i.e., spaces conformal to Einstein spaces, and it has been a long-standing classical problem to find simple characterizations of these spaces in terms of the Riemann curvature tensor. Therefore, in this thesis, we will investigate necessary and sufficient conditions for an n-dimensional space, n 4, to be locally conformal to an Einstein space, a subject studied since the 190s. Global properties will not be considered here. The first results in this field are due to Brinkmann [6], [7], but also Schouten [36] has contributed to the subject; they both considered the general n- dimensional case. Nevertheless, the set of conditions they found is large, and not useful in practice. Later, in 1964, Szekeres [39] introduced spinor tools into the problem and proposed a partial solution in four dimensions using spinors, restricting the space to be Lorentzian, i.e., to have signature. Nevertheless, the spinor conditions he found are hard to analyse and complicated to translate into tensors. Wünsch [43] pointed out a mistake in Szekeres s paper which means that his conditions are only necessary. However, in 1985, Kozameh, Newman and Tod [7] continued with the spinor approach and found a much simpler set consisting of only two independent necessary and sufficient conditions for four-dimensional spaces; however, the price they paid for this simplicity was that the result was restricted to a subspace of the most general class of spaces those for which

10 one of the scalar invariants of the Weyl tensor is non-zero, i.e., J = 1 C ab cdc cd ef C ef ab i C ab cd C cd ef C ef ab One of their conditions is the vanishing of the Bach tensor B ab ; in four dimensions this tensor has a number of nice properties. The condition J 0 in the result of Kozameh et al. [7] has been relaxed, also using spinor methods, by Wünsch [43], [44], by adding a third condition to the set found by Kozameh et al. This still leaves some spaces excluded; in particular the case when the space is of Petrov type N, although Czapor, McLenaghan and Wünsch [1] have some results in the right direction. The spinor formalism is the natural tool for general relativity in four dimensions in a Lorentzian space [3], [33] since it has built in both four dimensions and signature ; on the other hand, it gives little guidance on how to generalize to n-dimensional semi-riemannian spaces. However, using a more differential geometry point of view, Listing [8] recently generalized the result of Kozameh et al. [7] to n-dimensional semi-riemannian spaces having non-degenerate Weyl tensor. Listing s results have been extended by Edgar [13] using the Cayley-Hamilton Theorem and dimensionally dependent identities [14], [9]. There have been other approaches to this problem. For example, Kozameh, Newman and Nurowski [6] have interpreted and studied the necessary and sufficient condition for a space to be conformal to an Einstein space in terms of curvature restrictions for the corresponding Cartan conformal connection. Also Baston and Mason [3], [4], working with a twistorial formulation of the Einstein equations, found a different set of necessary and sufficient conditions. However, we shall restrict ourselves to a classical semi-riemannian geometry approach. In this thesis we are going to try and find n-dimensional tensors, n 4, generalizing the Bach tensor in such a way that as many good properties of the four-dimensional Bach tensor as possible are carried over to the n-dimensional generalization. We shall also investigate how these generalizations of the Bach tensor link up with conformal Einstein spaces. The first part of this thesis will review the classical tensor results; Chapters 5 to 8 will review and extend a number of the results in both spinors and tensors during the last 0 years. In the remaining chapters we will present some new results and also discuss the directions where this work can develop in the future. We have also included four appendices in which we have collected some old and developed some new results needed in the thesis, but to keep the presentation as clear as possible we have chosen to summarize these at the end. The outline of the thesis is as follows: We begin in Chapter by fixing the conventions and notation used in the thesis and giving some useful relations and identities. The chapter ends by

11 3 reviewing and proving the classical result that a space is conformally flat if and only if the Weyl tensor is identically zero. In Chapter 3 Einstein spaces and conformal Einstein spaces are introduced and the conformal Einstein equations are derived. Some of the earlier results in the field are also mentioned. In Chapter 4 the Bach tensor B ab in four dimensions is defined and various attempts to find an n-dimensional counterpart are investigated. We strengthen a theorem due to Belfagón and Jeán and give a basis U ab, V ab and W ab for all n-dimensional symmetric, divergence-free -index tensors quadratic in the Riemann curvature tensor. We discover the simple relationship B ab = 1 U ab V ab between the four-dimensional Bach tensor and these tensors, and show that this is the only -index tensor up to constant rescaling which in four dimensions is symmetric, divergence-free and quadratic in the Riemann curvature tensor. We also demonstrate that there is no useful analogue in higher dimensions. Chapter 5 deals with the four-dimensional result for spaces in which J 0 due to Kozameh, Newman and Tod, and both explicit and implicit results in their paper are proven and discussed. We also explore a little further the relationship between spinor and tensor results. In Chapter 6 and Chapter 7 the recent work of Listing in spaces with non-degenerate Weyl tensors is reviewed; Chapter 6 deals with the fourdimensional case and Chapter 7 with the n-dimensional case. In Chapter 8 we look at the extension of Listing s result due to Edgar using the Cayley-Hamilton Theorem and dimensionally dependent identities. In Chapter 9 the concept of a generic Weyl tensor and a generic space is defined. The results of Kozameh, Newman and Tod are generalized and generic results presented. We show that an n-dimensional generic space is conformal to an Einstein space if and only if there exists a vector field satisfying two conditions. The explicit use of dimensionally dependent identities is also exploited in order to make these two conditions as simple as possible; explicit examples are given in five and six dimensions. In Chapter 10, for n dimensions, we define the tensors b abc and B ab, whose vanishing guarantees a space with non-degenerate Weyl tensor being a conformal Einstein space. We show that b abc is conformally invariant in all spaces with non-degenerate Weyl tensor, and that B ab is conformally weighted with weight, but only in spaces with non-degenerate Weyl tensor where b abc = 0. We also show that the Listing tensor L ab is conformally invariant in all n-dimensional spaces with non-degenerate Wely tensor. In the final chapter we briefly summarize the thesis and discuss different ways of continuing this work and possible applications of the results given in the earlier chapters. Appendix A deals with the representations of the Weyl spinor/tensor as matrices and discusses the Cayley-Hamilton Theorem for matrices and tensors. In Appendix B dimensionally dependent identities are discussed. A number

12 4 of new tensor identities in five and six dimensions suitable for our purpose are derived; these identities are exploited in Chapter 9. In Appendix C, in four dimensions, we look at the Weyl scalar invariants and derive relations between the two complex invariants naturally arising from spinors and the standard four real tensor invariants. The last appendix briefly comments on the computer tools used for some of the calculations in this thesis. At an early stage of this investigation we became aware of the work of Listing [8], who had also been motivated to generalize the work of Kozameh et al. [7]. So, although we had already anticipated some of the Listing s generalizations independently, we have reviewed these generalizations as part of his work in Chapters 6 and 7. When we were writing up this thesis May 004 a preprint by Gover and Nurowski [16] appeared on the The first part of this preprint obtains some of the results which we have obtained in Chapter 9 in essentially the same manner; however, they do not make the link with dimensionally dependent identities, which we believe makes these results more useful. The second part of this preprint deals with conformally Einstein spaces in a different manner based on the tractor calculus associated with the normal Cartan bundle. Out of this treatment emerges the results on the conformal behavior of b abc and B ac, which we obtained in a more direct manner in Chapter 10. Due to the very recent appearance of [16] we have not referred to this preprint in our thesis, since all of our work was done completely independently of it.

13 Chapter Preliminaries In this chapter we will briefly describe the notation and conventions used in this thesis, but for a more detailed description we refer to [3] and [33]. We also review and prove the classical result that a space is conformally flat if and only if the Weyl tensor of the space is identically zero..1 Conventions and notation All manifolds we consider are assumed to be differentiable and equipped with a symmetric non-degenerate bilinear form g ab = g ba, i.e. a metric. No assumption is imposed on the signature of the metric unless explicitly stated, and we will be considering semi-riemannian or pseudo- Riemannian spaces in general; we will on occasions specialize to proper Riemannian spaces metrics with positive definite signature and Lorentzian spaces metrics with signature All connections,, are assumed to be Levi-Civita, i.e. metric compatible, and torsion-free, e.g. a g bc = 0, and a b b a f = 0 for all scalar fields f, respectively. Whenever tensors are used we will use the abstract index notation, see [3], and when spinors are used we again follow the conventions in [3]. The Riemann curvature tensor is constructed from second order derivatives of the metric but can equivalently be defined as the four-index tensor field R abcd satisfying [a b] ω c = a b ω c b a ω c = R abc d ω d.1 for all covector fields ω a, and it has the following algebraic properties, R abcd = R [ab][cd] = R cdab ; it satisfies the first Bianchi identity, R [abc]d = 0, and it also satisfies the second Bianchi identity, [a R bc]de = 0.

14 6 From the Riemann curvature tensor.1 we define the Ricci curvature tensor, R ab, by the contraction R ab = R acb c. and the Ricci scalar, R, from the contracted Ricci tensor R = R a a = R ab ab..3 For dimensions n 3 the Weyl curvature tensor or the Weyl conformal tensor, C abcd, is defined as the trace-free part of the Riemann curvature tensor, C abcd = R abcd g a[c R d]b g b[c R d]a + n n 1n Rg a[cg d]b.4 and as is obvious from above when R ab = 0 the Riemann curvature tensor reduces to the Weyl tensor. The Weyl tensor has all the algebraic properties of the Riemann curvature tensor, i.e. C abcd = C [ab][cd] = C cdab, C a[bcd] = 0, and in addition is trace-free, i.e. C abc a = 0. It is also well known that the Weyl tensor is identically zero in three dimensions. For clarity we note that due to our convention in defining the Riemann curvature tensor.1 we have for an arbitrary tensor field H b...d f...h that i j j i H b...d f...h = R ijb0 b H b0...d f...h R ijd0 d H b...d0 f...h R ijf f 0 H b...d f 0...h... R ijh h 0 H b...d f...h 0.5 and.5 is sometimes referred to as the Ricci identity. We will use both the conventions in the literature for denoting covariant derivatives, e.g., both the nabla and the semicolon, a v v ;a. Note however the difference in order of the indices in each case, a b v v ;ba. For future reference we write out the twice contracted second Bianchi identity, a R ab 1 br = 0,.6 the second Bianchi identity in terms of the Weyl tensor 0 = R ab[cd;e] = C ab[cd;e] + 1 n 3 g f 1 a[cc de]b ;f + n 3 g f b[cc ed]a ;f, the second contracted Bianchi identity for the Weyl tensor, d C abc d = n 3 1 [a R b]c + n n 1 g c[b a] R.7..8

15 7 Note that.8 also can be written d C d n 3 abc = n C cba.9 where C abc = [a R b]c + 1 n 1 g c[b a] R is the Cotton tensor. The Cotton tensor plays an important role in the study of thee-dimensional spaces [5],[15], [1]. We also have the divergence of.8 C db n 3 abcd; = n R ac; b n 3 b n 1 R n 3n ;ac + n R abr b c n 3 n 3 n Rbd C abcd n g acr bd R bd nn 3 n 1n RR n 3 ac n 1n g b acr ; b n 3 + n 1n g acr..10 Using the second Bianchi identity for the Weyl tensor, the Ricci identity and finally decomposing the Riemann tensor into the Weyl tensor, the Ricci tensor and the Ricci scalar we get the following identity [e d C ab]cd = n 3 n R [ab c f n 3 R e]f = n C [ab c f R e]f..11 In an n-dimensional space, letting H {Ω} a 1...a p = H {Ω} [a 1...a p] denote any tensor with an arbitrary number of indices schematically denoted by {Ω}, plus a set of p n completely antisymmetric indices a 1... a p, we define the Hodge dual H {Ω} a p+1...a n with respect to a 1... a p by H {Ω} a p+1...a n = 1 p! η a 1...a n H {Ω}a 1...a p.1 where η is the totally antisymmetric normalized tensor. Sometimes the is placed over the indices onto which the operation acts, e.g. H {Ω} a p+1...a n. In the special case of taking the dual of a double two-form H abcd = H [ab][cd] in four dimensions there are two ways to perform the dual operation; either acting on the first pair of indices, or the second pair. To separate the two we define the left dual and the the right dual as H ij cd = 1 η abijh abcd.13

16 8 and H ab ij = 1 η cdijh abcd.14 respectively. We are going to encounter highly structured products of Weyl tensors and to get a neater notation we follow [4] and make the following definition: Definition.1.1. For a trace-free, -form T abcd, i.e. for a tensor such that T a bad = 0, T abcd = T [ab]cd = T ab[cd].15 an expression of the form T ab c 1 d 1 T c1d1 c d... T cm dm c m 1 d m 1 T cm 1dm 1 ef }{{} m.16 where the indices a, b, e and f are free, is called a chain of the zeroth kind of length m and is written 1 T [m] ab ef. Hence, we have for instance that C[3] ab cd = C ab ijc ij klc kl cd. On occasions we will use matrices and these will always be written in bold capital letters, e.g. A. In this context O and I will denote the zero- and the identity matrix respectively and we will use square brackets to represent the operation of taking the trace of a matrix, e.g. [A] means the trace of the matrix A. Throughout this thesis we will only be considering spaces of dimension n 4. This is because in two dimensions all spaces are Einstein spaces and in three dimensions all spaces are conformally flat.. Conformal transformations Definition..1. Two metrics g ab and ĝ ab are said to be conformally related if there exists a smooth scalar field Ω > 0 such that holds. ĝ ab = Ω g ab.17 The metric ĝ ab is said to arise from a conformal transformation of g ab. Clearly we have ĝ ab = Ω g ab, since then ĝ ab ĝ bc = g ab g bc = δ a c. 1 In [4] this is denoted T 0 [m] ab ef.

17 9 To the rescaled metric ĝ ab there is a unique symmetric connection,, compatible with ĝ ab, i.e. c ĝ ab = 0. The relation between the two connections can be found in [3] and acting on an arbitrary tensor field H b...d f...h it is a H b...d f...h = a H b...d f...h + Q ab0 b H b 0...d f...h Q ad0 d H b...d 0 f...h Q af f 0 H b...d f 0...h... Q ah h 0 H b...d f...h 0.18 with Q ab c = Υ a δ c b g abυ c.19 where Υ a = Ω 1 a Ω = a ln Ω and Υ a = g ab Υ b. The relations between the tensors defined in and their hatted counterparts, i.e. the ones constructed from ĝ ab, are R abc d =R abc d δ d [a b]υ c + g c[a b] Υ d δ d [b Υ a]υ c + Υ [a g b]c Υ d + g c[a δ d b] Υ eυ e,.0 and R ab = R ab + n a Υ b + g ab c Υ c n Υ a Υ b + n g ab Υ c Υ c,.1 R = Ω R + n 1 c Υ c + n 1n Υ c Υ c,. Ĉ abc d = C abc d..3 Note that the positions of the indices in all equations are crucial since we raise and lower indices with different metrics, e.g. Ĉ abcd = ĝ de Ĉ abc e = Ω g de C abc e = Ω C abcd. Definition... A tensor field H b...d f...h is said to be conformally wellbehaved or conformally weighted with weight ω if under the conformal transformation.17, ĝ ab = Ω g ab, there is a real number w such that H b...d f...h Ĥb...d f...h = Ω ω H b...d f...h..4 If ω = 0 then H b...d f...h is said to be conformally invariant. Following [3] we introduce the tensor P ab = 1 n R 1 ab + n 1n Rg ab.5 These can be found, for instance, in [40], but note that Wald is using a different definition than.1 to define the Riemann curvature tensor.

18 10 and with the notation.5 we can express. -.4 as C ab cd = R ab cd + 4P [a [c g b] d],.6 and R ab = n P ab g ab P c c,.7 R = n 1P c c = n 1P..8 The contracted second Bianchi identity for the Weyl tensor.8 can now be written d C abc d = n 3 [a P b]c..9 Note that P ab is essentially R ab with a different trace term added and that P ab simply replaces R ab to make equations such as.6 and.9 simpler than.4 and.8, and hence to make calculations simpler. Furthermore, under a conformal transformation, P ab = P ab + Υ a Υ b a Υ b 1 g abυ c Υ c,.30 and P = Ω P c Υ c n Υ c Υ c,.31 which are simpler than the corresponding equations.1 and...3 Conformally flat spaces Definition.3.1. A space is called flat if its Riemann curvature tensor vanishes, R abcd = 0. Since flat spaces are well understood we would like a simple condition on the geometry telling us when there exists a conformal transformation making the space flat. Hence we define, Definition.3.. A space is called conformally flat if there exists a conformal transformation ĝ ab = Ω g ab such that the Riemann curvature tensor in the space with metric ĝ ab vanishes, i.e., R abcd = 0. From.0 we know that a space is conformally flat if and only if 0 =R abc d δ d [a b]υ c + g c[a b] Υ d δ d [bυ a] Υ c + Υ [a g b]c Υ d + g c[a δ d b] Υ eυ e.3 for some gradient vector field Υ a. Raising one index in equation.3 we can rewrite this as 0 = R ab cd + 4δ [c [a b]υ d] + 4Υ [a δ [c b] Υd] + δ [c [a δd] b] Υ eυ e = R ab cd + 4P [a [c δ d] b].33

19 11 where from.30 with P ab = 0 we have P ab = a Υ b Υ a Υ b + 1 g abυ c Υ c.34 and P [ab] = We will now prove the important classical theorem 3 Theorem.3.1. A space is conformally flat if and only if its Weyl tensor is zero. Proof. The necessary part follows immediately from the conformal properties of the Weyl tensor.3. To prove the sufficient part we break up the proof into two steps. First we shall show that if there exists some symmetric -tensor P ab in a space such that 0 = R ab cd + 4P [a [c δ d] b].36 P ab = a K b K a K b + 1 g abk c K c ; P [ab] = 0.37 for some vector field K a, then the space is conformally flat. From.37 it follows that [a K b] = 0 which means that locally K a is a gradient vector field, K a = a Φ.38 for some scalar field Φ, and substitution of P ab in.37 with this gradient expression for K a into.36 gives 0 =R abc d δ d [a b] c Φ + g c[a b] d Φ δ d [b a] Φ c Φ + [a Φ g b]c d Φ + g c[a δ d b] e Φ e Φ, i.e.,.0 with Υ a = a Φ, which implies that R abcd = 0 and so the space is conformally flat, ĝ = e Φ g ab. Secondly we shall show that if C abcd = 0 then.36 and.37 are satisfied. From.6 it follows immediately that if C abcd = 0, [37]. 0 = R ab cd + 4P [a [c δ d] b].39 3 The necessary part is originally due to Weyl [41], and the sufficient part to Schouten

20 1 i.e..36 is satisfied. Also, if C abcd = 0, we know from the second Bianchi identity.9 that [a P b]c = To check if.37 can be satisfied we calculate the integrability condition of.37 which is 0 = [c P a]b [c a] Υ b + Υ b [c Υ a] + Υ [a c] Υ b Υ e g b[a c] Υ e 1 = [c P a]b + R ca be + P [b [c δ e] a] Υ e.41 But from.39 and.40 it follows that this condition is identically satisfied. Hence we conclude that.37 is a consequence of.36. To summarize, we have shown that C abcd = 0 implies.36, which in turn implies.36 and.37 with the help of the Bianchi identities, meaning that the space is conformally flat.

21 Chapter 3 Conformal Einstein equations and classical results In this chapter we will define conformal Einstein spaces and derive the conformal Einstein equations in n dimensions. We will also give a short summary of the classical results due to Brinkmann [6] and Schouten [36]. 3.1 Einstein spaces Definition An n-dimensional space is said to be an Einstein space if the trace-free part of the Ricci tensor is identically zero, i.e. R ab 1 n g abr = Expressing this condition using the P ab tensor we get an expression having the same algebraic structure n P ab 1 n g abp = 0, 3. and we also note that in an Einstein space.5 becomes 1 P ab = nn 1 g abr. 3.3 From the contracted Bianchi identity.6 we find for an Einstein space that 0 = a n R ab b R = b R n, 3.4 i.e. the Ricci scalar must be constant.

22 14 3. Conformal Einstein spaces Definition An n-dimensional space with metric g ab is a conformal Einstein space or conformally Einstein if there exists a conformal transformation ĝ ab = Ω g ab such that in the conformal space with metric ĝ ab Rab 1 = 0 nĝab, 3.5 or equivalently P ab 1 P = 0 nĝab. 3.6 Note that from. and 3.4 we have that R = Ω R + n 1 c Υ c + n 1n Υ c Υ c = constant, 3.7 where we used the notation introduced in Chapter.. Further, equation 3.5 is equivalent to and 3.6 to R ab 1 n g n abr + n a Υ b g ab c Υ c n n n Υ a Υ b + g ab Υ c Υ c = 0, 3.8 n P ab 1 n g abp a Υ b + 1 n g ab c Υ c + Υ a Υ b 1 n g abυ c Υ c = respectively. 3.8 or 3.9 is often referred to as the n-dimensional conformal Einstein equations. Taking a derivative of 3.7 gives the relations 0 = a R RΥ a 4n 1Υ a c Υ c n 1n Υ a Υ c Υ c + n 1 a c Υ c + n 1n Υ c a Υ c = a P P Υ a + Υ a c Υ c + n Υ a Υ c Υ c a c Υ c n Υ c a Υ c 3.10 and using this, the first integrability condition of 3.9 is calculated to be or, using.9, [a P b]c + 1 C abcdυ d = 0, 3.11 d C abcd + n 3Υ d C abcd = Taking another derivative and using.9 again we have b [a P b]c + 1 P bd C abcd + n 4Υ b [a P b]c =

23 15 and clearly both 3.11 and 3.13 are necessary conditions for a space to be conformally Einstein. Obviously we could get additional necessary conditions by taking higher derivatives. The last equation 3.13 can also be written b d C abcd n 3 n Rbd C abcd n 3n 4Υ b Υ d C abcd = and in dimension n = 4 this condition 3.14 reduces to a condition only on the geometry, and is independent of Υ a. This condition, B ac b d C abcd 1 Rbd C abcd = 0, 3.15 is a necessary, but not a sufficient, condition for a four-dimensional space to be conformally Einstein. Note that if 3.8 holds for any vector field K a, R ab 1 n g n abr + n K a K b g ab c K c n n n K a K b + g ab K c K c = 0, 3.16 n and remembering that R ab is symmetric, then by antisymmetrising we get [a K b] = 0, i.e. that K a is locally a gradient. Hence we have that a space is locally a conformal Einstein space if and only if 3.16 holds for some vector field K a. Given that P ab is defined by.5 this same statement also holds for P ab 1 n g abp a K b + 1 n g ab c K c + K a K b 1 n g abk c K c = The classical results In 194 Brinkmann [6] found necessary and sufficient conditions for a space to be conformally Einstein. In his approach he derived a large set of differential equations involving Υ a and by exploiting both existence and compatibility of this derived set he was able to formulate necessary and sufficient conditions. However, from his results it is hard to get a constructive set of necessary and sufficient conditions, and his results are not very useful in practice. Brinkmann later also studied in detail some special cases of conformal Einstein spaces [7].

24 16 Schouten [36] used a slightly different approach and looked directly at the explicit form of integrability condition for the conformal Einstein equations, but he did not go beyond Brinkmann s results as regards sufficient conditions. Schouten found the necessary condition 3.13 which we will return to in the following chapters.

25 Chapter 4 The Bach tensor in four dimensions and possible generalizations In this chapter we will take a closer look at the four-dimensional version of the conformal Einstein equations introduced in the previous chapter. We will consider the Bach tensor B ab and derive and discuss its properties. A theorem stating that in n dimensions there only exists three independent symmetric, divergence-free -index tensors U ab, V ab and W ab quadratic in the Riemann curvature tensor is proven, extending a result due to Balfagón and Jaén []. The properties of these tensors are investigated, and we obtain the new result that B ab = 1 U ab V ab. We also seek possible generalizations of the Bach tensor in n dimensions. 4.1 The Bach tensor in four dimensions From 3.8, 3.9 in the previous chapter we know that in four dimensions the conformal Einstein equations are or R ab 1 4 g abr + a Υ b g ab c Υ c Υ a Υ b + g ab Υ c Υ c = P ab 1 4 g abp a Υ b g ab c Υ c + Υ a Υ b 1 4 g abυ c Υ c = 0. 4.

26 18 The necessary conditions 3.11 and 3.13 in four dimensions become [a P b]c + 1 C abcdυ d = and b [a P b]c + 1 P bd C abcd = respectively. The left hand side of this last equation 4.4 defines, as in 3.15, the tensor B ac, which can also be written as B ac = b [a P b]c + 1 P bd C abcd, 4.5 B ac = b d C abcd 1 Rbd C abcd, 4.6 and we see that the necessary condition 4.4 then can be formulated as B ac = 0. The tensor B ab is called the Bach tensor and was first discussed by Bach [1]. As seen above, the origin of the Bach tensor is in an integrability condition for a four-dimensional space to be conformal to an Einstein space. The Bach tensor is a tensor built up from pure geometry, and thereby captures necessary features of a space being conformally Einstein in an intrinsic way. It is obvious from the definition of B ab 4.6 that the Bach tensor is symmetric, trace-free and quadratic in the Riemann curvature tensor. Definition A tensor is said to be quadratic in the Riemann curvature tensor if it is a linear combination of products of two Riemann curvature tensors and/or a linear combination of second derivatives of the Riemann curvature tensor [11]. Calculating the divergence of B ab from 4.6 we get, after twice using.5 to switch the order of the derivatives, c B ac = 1 3 R a c c R + R bc c R ba 1 Rbc a R bc + 1 b c c R ba R aebd b R de 1 1 a c c R 1 6 c a c R =0 4.7 i.e. B ab is divergence-free 1. 1 This was first noted by Hesselbach [].

27 19 Under a conformal transformation, ĝ ab = Ω g ab, we see after some calculation that B ac = b d Ĉ abc d 1 R b dĉabc d = b d C abc d + Υ d C abc d 1 Ω R b d + b Υ d + δ b d c Υ c Υ b Υ d + δ b dυ c Υ c C abc d = Ω b d C abcd 1 Rbd C abcd = Ω B ac 4.8 so that B ac is conformally weighted with weight. We can also express the Bach tensor 4.6 in an alternative form in terms of the Weyl tensor, the Ricci tensor and the Ricci scalar, using the fourdimensional version of.10, b d C abcd = 1 R ac; b b 1 6 R ;ac 1 1 g acr ; b b + R ab R b c 1 Rbd C abcd 1 3 RR ac 1 4 g acr bd R bd g acr, 4.9 and we find B ac = 1 R ac;b b 1 1 g acr ;b b 1 6 R ;ac R bd C abcd + R ab R b c 1 4 g acr db R bd 1 3 RR ac g acr For completeness we also give the Bach tensor expressed in spinor language B ab = B AA BB = C A D B + ΦCD A B ΨABCD To summarize, the Bach tensor B ab in four dimensions given by 4.5, 4.6, 4.10 or 4.11 is symmetric, trace-free, quadratic in the Riemann curvature tensor, divergence-free, and is conformally weighted with weight. Although it is only in four dimensions that the Bach tensor has been defined and has these nice properties, it is natural to ask if there is an n-dimensional counterpart to the Bach tensor. Unfortunately as we shall see in the next section, it is easy to show that if we simply carry over the form of the Bach tensor given in 4.6 or 4.10 into n > 4 dimensions, it does not retain all these useful properties. So, in the subsequent sections we look to see if there is a generalization which retains as many as possible of the useful properties that the Bach tensor has in four dimensions. This result is originally due to Haantjes and Schouten [0].

28 0 4. Attempts to find an n-dimensional Bach tensor Before we begin looking for an n-dimensional Bach tensor we note that, using the notation from Chapter., we can derive the two useful relations b d Ĉ abcd = Ω b d C abcd + n 3C abcd b Υ d + n 4Υ d b C abcd +n 4Υ b d C abcd + n 3n 5Υ b Υ d C abcd and 4.1 R bd Ĉ abcd = R bd C abcd + n C abcd b Υ d n C abcd Υ b Υ d, 4.13 which are used extensively in this chapter. If we simply carry over the tensor in 4.6 to arbitrary n dimensions, and label this tensor B 1 ac, we find that Bac; a n 4 = 1 n B 1 ac = b d C abcd 1 Rbd C abcd, 4.14 C abcd R bd ; a n 1 Ra cr ;a + n 1 RR ;c n 3 R ab R bc;a R bd R bd;c n 4.15 and B 1 ac = Ω Bac + n 4 Υ b d C abcd + Υ d b C abcd C abcd b Υ d + n 7C abcd Υ b Υ d, 4.16 i.e., that its divergence-free and conformally well behaved properties do in general not carry over into n > 4 dimensions. Similarly the alternative form from 4.10, B ac = 1 R ac;b b 1 1 g acr ;b b 1 6 R ;ac R bd C abcd + R ab R b c 1 4 g acr db R bd 1 3 RR ac g acr, 4.17

29 1 also fails to have these properties in general in n > 4 dimensions since Bac; a = R ab R bc;a R bd R bd;c and n 4 n B bc =Ω B bc + n 4Ω n 3n C abcd Υ a Υ d + n 3 + 3n 5 Υ b c R + Υ c b R 4n 1 n + 1 6n 1 Ra cr ;a + n Υ a d C abcd 3n 10 ΘR bc 6 g bc R ad Θ ad Θ ab R a c R ab Θ a c + 3n 10n + 5 RΘ bc + 6n 1 n 3n + 3n 16n n 5 ΘΘ bc 3 3n 5 g bc a a Θ n 1 RR ;c 4.18 n Υ d a C abcd 8n n 1 RΥ bυ c n 73n 5 g bc Υ a a R 1n 1 3n 5n 7 g bc RΥ a Υ a 1n 1 Θ ab Θ a c 1 g bcθ ad Θ ad 3n 10n g bc RΘ 6 Υ c b Θ + Υ b c Θ 1 3 b c Θ n 73n 5 g bc ΘΥ a Υ a Υ a a Θ 6 + 3n 5ΘΥ b Υ c 6n 41n + 53 g bc Θ where Θ ab = a Υ b Υ a Υ b + 1 g abυ c Υ c and Θ = Θ a a = a Υ a + n Υ a Υ a. Going back to the origins of the Bach tensor as an integrability condition for conformal Einstein spaces 3.14, b d C abcd suggests considering the tensor n 3 n Rbd C abcd n 3n 4Υ b Υ d C abcd = Bac = b d n 3 C abcd 3 n Rbd C abcd = B 1 ac n 4 n Rbd C abcd. 4.1 But once again, for dimensions n > 4, we see that a n 4n 3 B c;a = 3 n R ab R bc;a R bd R bd;c 1 1 n 1 Ra cr ;a + n 1 RR ;c 4.

30 and [ B 3 ac = Ω Bac + n 4 Υ b d C abcd + Υ d b C abcd 3 + n 3C abcd Υ b Υ d] 4.3 so this tensor is neither divergence-free nor conformally well-behaved in general. So we need to look elsewhere for possible generalization of the Bach tensor to n > 4 dimensions. 4.3 The tensors U ab, V ab and W ab Many years ago Gregory [17] discovered two symmetric divergence-free tensors in four dimensions, and later Collinson [11] added a third. Recently Robinson [34] and Balfagón and Jaén [] have shown that these three tensors have direct counterparts in n > 4 dimension with the same properties. We shall first of all show that these three tensors, U ab, V ab and W ab, are the only three tensors with these properties and then also examine in more detail their structure and properties. Balfagón and Jaén [] have proven the following theorem 3 : Theorem In an arbitrary n-dimensional semi-riemannian manifold: a There exist 14 independent and quadratic in Riemann, four-index divergence-free tensors. b There are no totally symmetric, quadratic in Riemann, and divergencefree four-index tensors c The complete family of quadratic in Riemann, and divergence-free 3 Note that this is a quotation from their paper [], but here expressed using our conventions.

31 3 four-index tensors T abcd totally symmetric in bcd is T abcd = a S T abcd S + a R T abcd R ; 4.4 T abcd S = Q abcd ; 4.5 Q abcd = 1 3 gac R d ir ib R ab;dc Rbd;ac 4 3 gac R bd;i i + g ac R b i ;di Rb ir acid + R aibj R c i d j 1 gac R ij d k R ijbk 4.6 T abcd R = X ab g cd ; 4.7 X ab = KU ab + LV ab 1 4 W ab 4.8 U ab = G ab;s s G sb;a s + G a pr pb 1 gab G pq R pg 4.9 V ab = R ;ab + g ab R ;s s RS ab 4.30 W ab = G apqr R b pqr 1 4 gab G mpqr R mpqr 4.31 G a c = G ab cb 4.3 G ab cd = R ab cd 4δ [a [c Sb] d] 4.33 S ab = 1 4 gab R 4.34 where a S, a R, K and L are four independent constants 4. First note that there are some differences in signs in 4.6, 4.9 and 4.30 compared to the original definitions in []. This is due to our definition of the Riemann curvature tensor.1 which differs from the one in []. A change of convention makes the change R abcd R abcd, meaning that there is only going to be a difference in sign for the terms created from a odd number of Riemann tensors e.g. here exactly only one. However, our definition agrees with the one in [34] up to an overall sign. Secondly, by a divergence-free four index tensor Balfagón and Jaén mean a tensor T abcd such that a T abcd = 0, i.e. a tensor divergence-free on the first index. Hence, in Theorem a states that there exist only 14 independent such tensors, b states that none of these are totally symmetric, T abcd = T abcd and c gives all tensors T abcd such that T abcd = T abcd and a T abcd = 0. 4 T R is the tensor found by Robinson [34] and T S the tensor found by Sachs [35].

32 4 Note that Theorem implies that the tensors U ab, V ab and W ab are all divergence-free. This follows from the divergence-free property and the construction of the tensor T abcd R via 4.7 and 4.8. We are specially interested in the tensors U ab, V ab and W ab, and to see their inner structure and their properties we write them out in terms of the Riemann curvature tensor, the Ricci tensor and the Ricci scalar, U bc =n 3R abcd R ad n 3R a n 3 bc;a + g bc R ad R ad n 3 + n 3RR cb + g bc R a n 3 ;a g bc R, V bc = R ;bc + g bc R ;a a RR bc g bcr, 4.36 W bc =R b ade R cade 1 4 g bcr fade R fade + R ad R abcd + RR bc R ab R a c + g bc R ad R ad 1 4 g bcrr When we later study the conformal behavior of U ab, V ab and W ab it is useful to have them expressed in terms of the Weyl tensor, the Ricci tensor and the Ricci scalar, U bc =n 3R ad C abcd + n 3R a n 3 a bc;a g bc R ;a n 6n 3 + g bc R de R de 4n 3 + n n R abr a c + n 3n 5n + n 1n RR bc n 3n 3n 6 g bc R, 4n 1n 4.38 V bc = R ;bc + g bc R ;d d RR bc g bcr, 4.39 W bc = C ade b C cade 1 4 g bcc fade n 4 C fade + n C abcdr ad n 3n 4 n R ab R a c + n 3n 4 n g bc R ad R ad nn 3n 4 + n 1n RR n + n 3n 4 bc 4n 1n g bc R From or it is obvious that U ab, V ab and W ab are symmetric and quadratic in the Riemann curvature tensor in all

33 5 dimensions. We note from 4.40 that W bc = 0 in four dimensions because in four dimensions C b ade C cade 1 4 g bcc fade C fade = 0, see Appendix B. This fact was not noticed by Collinson [11] but was subsequently pointed out in [34] and []. By taking the trace of we have U a n 3n a = R a ;a + W a a = n 4n 3 R ad R ad n 4n 3 R, V a a = n 1R ;a a + n 4 C fade C fade + 4 n 4 R, n 3n 4 R ad R ad n nn 3n 4 4n 1n R A simple direct calculation would confirm that U ab, V ab and W ab are all divergence-free, but we have already noted that this can be deduced from Theorem It is easily checked that the three tensors U ab, V ab and W ab are independent and an obvious question is whether there are any more such tensors; we shall now show that there are not. Given any symmetric and divergence-free tensor, Y ab, quadratic in the Riemann curvature tensor we see that the tensor Y ab g cd is a four-index tensor which is totally symmetric over bcd, quadratic in the Riemann curvature tensor and divergence-free on the first index. Hence we know from Theorem c, that there exist constants a S, and a R such that Y ab g cd = a S T abcd S + a R T abcd R 4.44 holds. Taking the trace over c and d of 4.44 using the facts that g cd TR abcd = g cd X abcd = g cd KU ab + LV ab W ab g cd = n + KU ab + LV ab W ab, 4.45 where K and L are constants fixed by 4.44, and g cd T abcd S =g cd Q abcd = 4 3 Ra cr bc R ; ab 14 9 Rab ; c c 1 9 gab R ;d d R cdr adbc Radef R b def 1 9 gab R cd R cd 1 6 gab R cdef R cdef, 4.46

34 6 noting that 4.46 actually can be written as a linear combination of U ab, V ab and W ab, g cd T abcd S = 14 9n 3 U ab 8 9 V ab + 3 W ab, 4.47 we find that Y ab 14a S = a R K 9n n 3 ar a S 3n U ab + a R L 8a S V ab 9n W ab, 4.48 i.e., that Y ab is a linear combination of U ab, V ab and W ab. We summarize this result in the following theorem Theorem In an n-dimensional space there are only three independent symmetric and divergence-free -index tensors quadratic in the Riemann curvature tensor, e.g., U ab, V ab and W ab. Before we investigate the relations between the four-dimensional Bach tensor and U ab, V ab and W ab we note that under the conformal transformation ĝ ab = Ω g ab the tensors U ab, V ab and W ab transform according to [ Û bc =Ω U bc + Ω n 3 n C abcd Θ ad + n 6Υ a a R bc + Υ a c R ab + Υ a b R ac + R bc n 4Θ n 4Υ a Υ a + R ab 4Θ a c nυ a Υ c + R ac 4Θ a b nυ a Υ b + g bc R ef n 6Θ ef + Υ e Υ f Υ b c R Υ c b R n 6 + g bc R g bc Υ a a R + R n 5n + Θ + n 1 Υ b Υ c + n 5n + n 1 n 6 Υ a Υ a + g bc n Υ e Υ f Θ ef n a a Θ Θ bc + n n 6Υ a Υ a n n 11 Θ Θ n n 6 + Θ ad Θ ad n n 6Υ a a Θ + n a a Θ bc + n 6n Υ a a Θ bc n Υ b a Θ ac n Υ c a Θ ab + n Υ a b Θ ac + n Υ a c Θ ab + Θ ab 4n Θ a c nn Υ c Υ a nn Θ ac Υ a Υ b

35 7 + Θ bc n n 4Θ n n 4Υ a Υ a + n ΘΥ b Υ c ], 4.49 V bc =Ω V bc + Ω [ n 1ΘR bc 6RΥ c Υ b n 4RΘ bc n 7g bc RΥ a Υ a + n 4g bc RΘ + 3Υ c b R + 3Υ b c R + n 7g bc Υ a a R n 1 b c Θ + n 1g bc a a Θ + 6n 1Υ c b Θ + 6n 1Υ b c Θ + n 1n 7g bc Υ a a Θ + n 1n 7g bc Θ n 1n 7Θg bc Υ a Υ a 1n 1ΘΥ b Υ c n 1n 4ΘΘ bc ], 4.50 [ Ŵ bc =Ω W bc + Ω n 4 C abcd Θ ad n 3 n R abθ a c n 3 n R acθ a b n 3Θ ab Θ a c n 3 + n g bcθ ad R ad + n 3g bc Θ ad Θ ad nn 3 + n 1n RΘ n 3 bc + n R bcθ + n 3Θ bc Θ n 3g bc Θ nn 3 ] n 1n g bcrθ, 4.51 where Θ ab = a Υ b Υ a Υ b + 1 g abυ c Υ c and Θ = Θ a a = a Υ a + n Υ a Υ a. It is easily seen from that U bc and V bc are not conformally well-behaved in general in four dimensions where W bc = Four-dimensional Bach tensor expressed in U ab, V ab and W ab Since U ab, V ab and W ab constitute a basis for all -index symmetric divergence-free tensors quadratic in the Riemann curvature tensor, and the Bach tensor B ab has these properties, we must be able to express the Bach tensor 4.6 in four dimensions in terms of U ab and V ab, remembering W ab = 0 in four dimensions.

36 8 In four dimensions we know from U bc =R bc;a a 1 g bcr ;a a + C abcd R ad + R ab R c a 1 g bcr ad R ad 1 3 RR bc g bcr 4.5 V bc = R ;bc + g bc R ;a a RR bc g bcrr 4.53 Comparing these equations with 4.10 we conclude that we have the relation B bc = 1 U bc V bc The numerical relationship between the tensors V bc and U bc could also be found using the trace-free property of the Bach tensor. Making the ansatz B bc = αu bc + βv bc 4.55 we see from 4.41 and 4.4 in four dimensions that B b b = αr ;b b + 3βR ;b b = α 3βR ;b b = and hence in general we must have 3α = β. This link between the Bach tensor B ab and U ab and V ab in four dimensions does not seem to have been noted before. 4.5 An n-dimensional tensor expressed in U ab, V ab and W ab If we consider the tensor B bc = 1 U bc V bc 4.57 in n > 4 dimensions, it is clearly divergence-free due to the properties of U ab and V ab, but when we examine its conformal properties we find, after a lot of work and rearranging, Ûbc 6 V bc = Ω 1 U bc V bc + n 4Ω [ 1 n n 3C abcdυ a Υ d + n 3Υ a n 3 a R bc Υ a n 3 c R ab Υ a b R ac n 3 R ab Θ a n 3 c R ac Θ a n 3 b + g bc R ad Θ ad

37 9 3n 7 + R bc Θ 1 3 Υ b c R 1 Υ 3n 17 c b R g bc Υ a a R 1 3n 10 n 73n 5 g bc RΘ + RΥ b Υ c 6 n 1 n 73n 5 + g bc RΥ a Υ a n n 3 + g bc Θ ad Θ ad 1n 1 4 n n 3 n Θ ab Θ a c + 3n ΘΘ bc 3 3n 5 3n 5 + b c Θ g bc a 3n 5 a Θ 3 Υ b c Θ 3n 5 3n 5 3 Υ c b Θ n 7 g bc Υ a a Θ 6 3n 5 3n 5 + ΘΥ b Υ c + n 7 g bc ΘΥ a Υ a 6 6n 41n + 53 g bc Θ ], again getting a tensor that is not conformally well-behaved except in four dimensions. The n-dimensional analogous integrability condition that gave rise to the four-dimensional Bach tensor is 3.14, b d C abcd n 3 n Rbd C abcd n 3n 4Υ b Υ d C abcd = and taking only the terms built up from pure geometry and quadratic in the Riemann curvature tensor, e.g., the first and second terms, suggests that we study B 3 ac = b d C abcd n 3 n Rbd C abcd Although we have already shown that this tensor is neither conformally well-behaved nor divergence-free in n > 4 dimensions, it will be instructive to investigate its relationship to the tensors U ab, V ab and W ab. Using.10 we can equivalently express B 3 ac in the decomposed form n 3 Bac = 3 n R ac; b b n 3 n 1 R ;ac n 3 n 1n g acr ; b b n 3n + n R abr b n 3 c n Rbd C abcd n 3 n g acr bd R bd nn 3 n 1n RR ac n 3 + n 1n g acr. 4.61

38 30 This clearly reduces to the ordinary Bach tensor in four dimensions, but to investigate its other properties we first try to express B 3 ac in terms of U ac, V ac and W ac. Doing this we find Bac = 1 3 n U n 3 ac + n 1 V ac 1 W ac + 1 bde Ca C cbde 1 4 g acc fbde C fbde n 4 n Rbd C abcd, 4.6 and as we have already noted and is easily confirmed directly this tensor is only divergence-free in four dimensions, and not for n > 4 dimensions. However, defining the tensor B 4 ac, Bac = Bac 1 bde Ca C cbde g acc fbde n 4 C fbde + n Rbd C abcd = Bac 1 bde Ca C cbde g acc fbde n 4 C fbde + n Rbd C abcd 1 = n U n 3 ac + n 1 V ac 1 W ac, 4.63 we indeed get a tensor quadratic in the Riemann curvature tensor which is symmetric and divergence-free in all dimensions, and it collapses to the original Bach tensor in four dimensions. To investigate the conformal properties of B 4 ac and thereby also the fourdimensional B ac we use and find [ B ac =Ω B ac + n 4 Υ b d C abcd + Υ d b C abcd C abcd b Υ d + n 4C abcd Υ b Υ d] From 4.64 we see that, in general, it is only in four dimensions that Bac is conformally well-behaved. Hence, in general, there is no obvious 4 n-dimensional symmetric and divergence-free tensor which is quadratic in the Riemann curvature tensor and generalizes the Bach tensor in four dimensions, which also is of good conformal weight. We can now ask more generally if it is possible to construct any n-dimensional -index tensor of good conformal weight from U ab, V ab and W ab, i.e. a tensor which is symmetric, divergence-free, quadratic in the Riemann curvature tensor and of good conformal weight. To investigate this we look at

39 31 αu ab + βv ab + γw ab, where α, β and γ are arbitrary constants, αûbc+β V bc + γŵbc = Ω αu bc + βv bc + γw bc [ + Ω n 4 γc abcd Θ ad αn n 3C abcd Υ a Υ d + αn 3 Υ a a R bc Υ a c R ab Υ a b R ac 1 + n 3 α β n 3 + γ n RΘ bc n 1n n 1 + n 3 αn β n 3 + γ ΘΘ bc n 3 αn + γ Θ ab Θ a c n + n 3 α + γ g bc Θ ad Θ ad 1 + γ n 3 α + γ αn 3 β + γ n R bcθ R ab Θ a c + R ac Θ a b g bc R ad Θ ad n ] g bc RΘ + γn 3g bc Θ nn 3 n 1n + Ω [ 3 αn n 3 βn 1 n 1 RΥ bυ c n 7 + n 1 g bcrυ a Υ a + b c Θ + 6ΘΥ b Υ c n 7g bc Υ a a Θ ] 3 Υ b c Θ + Υ c b Θ g bc a a Θ + n 7g bc ΘΥ a Υ a + Ω [n n 1 3 αn 3 β R bc Θ n 3 n n 3n 11 α βn 1n 7 g bc Θ αn 3 3β Υ b c R αn 3 3β Υ c b R ] n 3n 6 α βn 7 g bc Υ a a R. 4.65