Returning to our PE manifold, we may replace
${s}_{g}=m(m1)$
, and this lemma gives us a formula for the Pfaffian in terms of the successive contractions of the Weyl tensor. Although the Weyl tensor itself has trivial contraction, this is not true of its higher powers. For instance, from [
17]
,
${\mathcal{C}}^{4}{\mathfrak{W}}^{2}=6W{}^{2}$
.
Theorem 4.5.
Let
$x$
be a special bdf on an even dimensional asymptotically hyperbolic manifold
$(M,g)$
, and assume that
${x}^{2}g=d{x}^{2}+{G}_{x}$
where
$${G}_{x}={\overline{g}}^{\left(0\right)}+...\left(\text{even powers}\right)...+{\overline{g}}^{(m1)}{x}^{m1}+O\left({x}^{m}\right),$$
and
$trac{e}_{{\overline{g}}^{\left(0\right)}}{\overline{g}}^{(m1)}=0$
. Then
$${}^{R}\int Pff=\chi \left(M\right).$$

Proof.
This follows from Chern's GaussBonnet Theorem for the manifold with boundary
${M}_{\varepsilon}:=\{x\ge \varepsilon \}$
,
$$\begin{array}{c}{\int}_{{M}_{\varepsilon}}\text{Pff}+{\int}_{x=\varepsilon}\text{}\text{I}\text{I}\text{}\u203f\text{}\u2040\text{}=\chi \left({M}_{\varepsilon}\right).\end{array}$$ 
(4.7)

Here
$\text{}\text{I}\text{I}\text{}\u203f\text{}\u2040\text{}$
is a polynomial in the curvature and the second fundamental form of
$\{x=\varepsilon \}$
in
${M}_{\varepsilon}$
. In terms of the curvature and connection forms it is given in [10] by
$$\begin{array}{c}\text{}\text{I}\text{I}\text{}\u203f\text{}\u2040\text{}={\sum}_{q=0}^{m/2}{C}_{m,q}{\sum}_{\sigma \in {\Sigma}_{m1}}{\Omega}_{{\sigma}_{1}{\sigma}_{2}}\wedge \dots \wedge {\Omega}_{{\sigma}_{2q1}{\sigma}_{2q}}\wedge {\omega}_{{\sigma}_{2q+1}m}\wedge \dots \wedge {\omega}_{{\sigma}_{m1}m},\end{array}$$ 
(4.8)

for some constants
${C}_{m,q}$
whose precise value we will not need. Note that, for
$\varepsilon $
small,
$\chi \left({M}_{\epsilon}\right)=\chi \left(M\right)$
.
Since the right hand side of
4.7 does not depend on
$\varepsilon $
, neither does the left, and thus
${}^{R}\int Pff+{FP}_{\varepsilon =0}{\int}_{x=\varepsilon}\text{}\text{I}\text{I}\text{}\u203f\text{}\u2040\text{}=\chi \left(M\right).$
So we need only show that the second term vanishes.
If we denote the second fundamental form of
$\{x=\varepsilon \}$
as a double form by
$\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{}\in {\mathcal{D}}^{1,1}$
, 4.8 shows that
$\text{}\text{I}\text{I}\text{}\u203f\text{}\u2040\text{}$
is a linear combination of terms of the form
$$\begin{array}{c}{\mathcal{C}}^{m1}({\Re}^{k}\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}{\text{}}^{m12k})dvol.\end{array}$$ 
(4.9)

It is easy to see that, for any
$\omega $
,
$${\mathcal{C}}^{m1}\left(\omega \right){dvol}_{g}{\downharpoonright}_{\partial M}={x}^{m1}{\overline{\mathcal{C}}}^{m1}\left(\omega \right){dvol}_{\overline{g}}{\downharpoonright}_{\partial M},$$
where
$\overline{\mathcal{C}}$
denotes contracting via a local orthonormal frame of
$\overline{g}$
instead of
$g$
. Recall that
${r}_{ijk\ell}$
and
${\gamma}_{ij}^{m}$
have even expansions mod
${x}^{m1}$
, hence so do
${x}^{4}\Re $
and
${x}^{2}\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{}$
, and the constant term in 4.9 comes from the
${x}^{m1}$
terms in the expansion of these coefficients. Now, since
$$\begin{array}{cccc}{\left({\gamma}_{ij}^{m}\right)}^{\left(0\right)}={\overline{g}}_{ij}^{\left(0\right)}& ,{\left({\gamma}_{ij}^{m}\right)}^{(m1)}=\frac{3m}{2}{\overline{g}}_{ij}^{(m1)},& {\left({r}_{ijk\ell}\right)}^{\left(0\right)}={\left[{\gamma}_{ik}^{m}{\gamma}_{j\ell}^{m}{\gamma}_{jk}^{m}{\gamma}_{i\ell}^{m}\right]}^{\left(0\right)}& ,\text{and}{\left({r}_{ijk\ell}\right)}^{(m1)}={\left[{\gamma}_{ik}^{m}{\gamma}_{j\ell}^{m}{\gamma}_{jk}^{m}{\gamma}_{i\ell}^{m}\right]}^{(m1)},\end{array}$$  
we may conclude that the constant term in the expansion of 4.9 is a multiple of
${\overline{\mathcal{C}}}^{m1}\left({\overline{\mathfrak{g}}}_{\left(0\right)}^{m2}{\overline{\mathfrak{g}}}_{(m1)}\right)\overline{dvol}$
, which vanishes as
$trac{e}_{{\overline{g}}^{\left(0\right)}}{\overline{g}}^{(m1)}=0$
. □
It is instructive to use Lemma 4.4 and write out the renormalized GaussBonnet theorem 4.5 in four dimensions:
$$\begin{array}{c}{\frac{1}{(2\pi {)}^{2}}}^{R}\int {\sum}_{k=0}^{2}\frac{\left(2k\right)!}{k!}{\left(\frac{1}{2}\right)}^{k}\frac{{\mathcal{C}}^{42k}{\mathfrak{W}}^{2k}}{(42k)!(2k)!}=\frac{1}{8(2\pi {)}^{2}}\int W{}^{2}+\frac{3}{(2\pi {)}^{2}}\hat{V}=\chi (M),\end{array}$$ 
(4.10)

and in six dimensions:
$$\begin{array}{c}\frac{1}{(2\pi {)}^{3}}\int \frac{{\mathcal{C}}^{6}{\mathfrak{W}}^{3}}{6!3!}{\frac{1}{8(2\pi {)}^{3}}}^{R}\int W{}^{2}\frac{15}{(2\pi {)}^{3}}\hat{V}=\chi (M).\end{array}$$ 
(4.11)

Equation 4.10 is due to Anderson [
2]
(note that he uses a different convention for
$W{}^{2}$
, see the comment after equation (1.15) in [
2]
). It is only in dimension four that the integrand requires no further renormalization than the volume. Generally, from Lemma 4.4 we know that
$$\begin{array}{c}{}^{R}\int Pff=\int \text{Pff}\left(W\right){+}^{R}\int P\left(W\right)+\frac{(1{)}^{m/2}}{{2}^{m/2}(2\pi {)}^{m/2}}\frac{m!}{(m/2)!}\hat{V},\end{array}$$ 
(4.12)

where
$P\left(W\right)$
is a polynomial in the Weyl curvature, and
$\text{Pff}\left(W\right)$
is the Pfaffian evaluated in the Weyl curvature instead of the full curvature. Equation 4.11 shows that
$P\left(W\right)$
is not zero in general.
As we mentioned in the introduction, Chang, Qing, and Yang [
8]
have recently established a very similar formula 1.5 . It would be interesting to consolidate these formulas.
5 Varying the PoincareEinstein metric
In this section we will consider the variation of the characteristic forms of a PoincaréEinstein manifold when we allow the metric to vary along a family
${g}_{s}$
of PE metrics with the same scalar curvature. After describing the expansion of
$\dot{g}={\partial}_{s}{g}_{s}{}_{s=0}$
, we recover results of Anderson [
2]
and GrahamHirachi [
15]
on the variation of the renormalized volume. Finally, we verify directly that the variation of the renormalized integral of the Pfaffian vanishes.
5.1 Variation of the Pontrjagin Forms
As before, Pontrjagin forms and numbers on PoincaréEinstein manifolds are easily dealt with due to conformal invariance. Indeed, we know (Theorem 4.3 ) that
$$\text{tr}\left({\mathbf{\Omega}}^{\ell}\right)=\text{tr}\left({\overline{\mathbf{\Omega}}}^{\ell}\right),$$
so we need only consider variations of
$\overline{g}$
. Given a family of metrics
${\overline{g}}_{s}$
on
$\overline{M}$
with connection and curvature forms
${\overline{\omega}}_{s}$
,
${\overline{\mathbf{\Omega}}}_{s}$
and a homogeneous invariant polynomial in the curvature,
$P\left({\overline{\mathbf{\Omega}}}_{s}\right)$
, of degree
$q$
with complete polarization
$p({\overline{\mathbf{\Omega}}}_{s},\dots ,{\overline{\mathbf{\Omega}}}_{s})$
, and denoting the derivative at
$s=0$
by an overdot, we have
$$\dot{P}\left(\overline{\mathbf{\Omega}}\right)=q\text{d}p(\dot{\overline{\omega}},\overline{\mathbf{\Omega}},\dots ,\overline{\mathbf{\Omega}}).$$
The derivative of a product of Pontrjagin forms is exact, being the product of closed and exact forms. Note that the boundary of
$\overline{M}$
is totally geodesic, so restricting the LeviCivita connections of
${\overline{g}}_{s}$
yields the LeviCivita connections of the induced metrics. Hence
$\dot{\overline{\omega}}$
when restricted to the boundary is the derivative of the connection 1forms of the boundary metrics. We conclude with the following wellknown result.
Proposition 5.1.
Let
${\overline{g}}_{s}$
a smooth family of metrics on
$\overline{M}$
such that, for every
$s$
,
$(\overline{M},{\overline{g}}_{s})$
is a compact manifold with totally geodesic boundary. For any polynomial
$Q$
, the variation of the characteristic number of
$\overline{M}$
corresponding to
$Q(\mathbf{\Omega})$
is the integral over the boundary of the ChernSimons number corresponding to
$Q(\mathbf{\Omega})$
on
$\partial M$
.
5.2 Variation of the Renormalized Volume and the Pfaffian
Naturally, the renormalized GaussBonnet theorem discussed previously shows that the variation of the renormalized integral of the Pfaffian vanishes. We verify this directly by studying the variation of the renormalized volume and the renormalized Weyl volume of tubes invariants.
Let
${g}_{s}$
be a family of PoincaréEinstein metrics on
$M$
with the same scalar curvature,
$m(m1)$
.
We use
$x$
, a special boundary defining function for
$g:={g}_{0}$
, to define a corresponding family of metrics on the boundary
${g}_{s}^{\left(0\right)}:={x}^{2}{g}_{s}{\downharpoonright}_{\partial M}$
. The metrics
${g}_{s}^{\left(0\right)}$
in turn determine bdfs
${x}_{s}=x{e}^{\omega \left(x\right)}$
, special with respect to
${g}_{s}$
(as in Lemma 2.1 of [
13]
). We shall compute in a frame
${X}_{u}=x{\overline{X}}_{u}$
as in previous sections, and denote derivatives with respect to
$s$
at
$s=0$
by an overdot. Set
$h=\dot{g}$
.
Just as one needs to break conformal invariance by choosing a bdf in order to study the renormalized volume, the study of the Einstein equation requires breaking gauge invariance. The approach followed in [
2]
(see also [
15]
) is to use the FeffermanGraham expansion of the metrics
${g}_{s}$
. A second approach is to impose the so called Bianchi gauge. One defines the operator
$${B}_{g}\left(k\right):={\delta}_{g}k+\frac{1}{2}d\left({\text{tr}}_{g}h\right)$$
and demands that
${g}_{s}$
satisfy the Einstein equation and
${B}_{g}\left({g}_{s}\right)=0$
. This approach was first espoused in [
7]
, though we shall follow [
27]
(see also [
3]
and [
22]
).
The following result appears as Lemma 3.4 in [
3]
.
Lemma 5.2.
$x{\partial}_{x}\left(\dot{\omega}\right)=\frac{1}{2}{h}_{mm}$
.

Proof.
We differentiate the relation
$d{x}_{s}{}_{{\overline{g}}_{s}}^{2}=1$
:
$$\begin{array}{cc}{x}_{s}^{2}{g}_{s}\left(d{x}_{s},d{x}_{s}\right)& ={x}^{2}\left[{g}_{s}\left(dx,dx\right)+2x{g}_{s}\left(d{\omega}_{s},dx\right)+{x}^{2}{g}_{s}\left(d{\omega}_{s},d{\omega}_{s}\right)\right]=1\end{array}$$  
to get
$${x}^{2}\left[\frac{\partial}{\partial s}{}_{s=0}{g}_{s}^{mm}+2xg\left(dx,d\dot{\omega}\right)\right]=0.$$
As
$\partial {g}_{s}^{1}={g}^{1}\left(\partial {g}_{s}\right){g}^{1}$
, we conclude that
$$x{\partial}_{x}\left(\dot{\omega}\right)=\frac{1}{2}{h}_{mm}.$$
□
The linearized Einstein equation,
$Ri{c}^{\prime}\left(h\right)=(m1)h$
, is given by (1.179 in [6] )
$$\begin{array}{c}{E}_{g}^{\prime}\left(h\right):={g}^{st}({C}_{g}h{)}_{uvs,t}\frac{1}{2}Dd{\text{tr}}_{g}h+(m1)h=0,\end{array}$$ 
(5.1)

wherein
$$({C}_{g}h{)}_{uvs}=\frac{1}{2}\left({h}_{sv,u}+{h}_{us,v}{h}_{uv,s}\right).$$
As the scalar curvature is the same for all
$s$
, we also have
$Sca{l}_{g}^{\prime}\left(h\right)=0$
, which by Theorem 1.174(e) of [6] , means
$$\begin{array}{c}\begin{array}{cccc}Sca{l}_{g}^{\prime}\left(h\right)& ={\Delta}_{g}\left(t{r}_{g}h\right)+{\delta}_{g}\left({\delta}_{g}h\right)g\left(Ri{c}_{g},g\right)& & ={\Delta}_{g}\left(t{r}_{g}h\right)+{\delta}_{g}\left({\delta}_{g}h\right)+(m1)t{r}_{g}h=0.\end{array}\end{array}$$ 
(5.2)

It is easy to compute the
$0$
th order term in 5.1 , for instance:
$${\left({E}_{g}^{\prime}\left(h\right)\right)}_{mm}^{\left(0\right)}=(m1){h}_{mm}^{\left(0\right)},$$
so
${E}_{g}^{\prime}\left(h\right)=0$
implies
${h}_{mm}^{\left(0\right)}=0$
, and we gain no more information from the other
$0$
th order terms.
Computing higher order terms will allow us to show that any perturbation of the metric through PE metrics has an expansion like that of the metric.
Proposition 5.3.
Assume that
${h}_{mm}^{\left(0\right)}=0$
, and that for some natural number
$\alpha $
both:
$$\begin{array}{c}\begin{array}{cccc}1)& {\left({X}_{u}\left({h}_{st}\right)\right)}^{\left(\alpha \right)}={\delta}_{um}\left(\alpha {h}_{st}^{\left(\alpha \right)}\right)& 2)& {\left({\Gamma}_{vw}^{u}{h}_{st}\right)}^{\left(\alpha \right)}={\left({\Gamma}_{vw}^{u}\right)}^{\left(\alpha \right)}{\left({h}_{st}\right)}^{\left(0\right)}+{\left({\Gamma}_{vw}^{u}\right)}^{\left(0\right)}{\left({h}_{st}\right)}^{\left(\alpha \right)}\end{array}\end{array}$$ 
(5.3)

hold, then
$$\begin{array}{c}{\left({E}_{g}^{\prime}\left(h\right)\right)}_{uv}^{\left(\alpha \right)}=\{\begin{array}{cc}\frac{\alpha}{2}\left(m1\alpha \right){h}_{uv}^{\left(\alpha \right)}+{\delta}_{uv}\left[\left(m1\alpha \right){h}_{mm}^{\left(\alpha \right)}+\frac{\alpha}{2}({\text{tr}}_{g}h{)}^{\left(\alpha \right)}\right]& \text{if}m\text{/}\text{\u2208}\{u,v\}\text{}\\ \left(1\frac{\alpha}{2}\right)\left[\left(m1\alpha \right){h}_{mm}^{\left(\alpha \right)}+\alpha ({\text{tr}}_{g}h{)}^{\left(\alpha \right)}\right]& \text{if}m=u=v\text{}\end{array}.\end{array}$$ 
(5.4)

Consequently, once gauge invariance is broken, either through the FeffermanGraham expansion or by imposing the Bianchi gauge, we have
$$\begin{array}{c}\mathcal{\mathcal{F}}\left({h}_{uv}\right)=(1{)}^{{\delta}_{um}+{\delta}_{vm}},\end{array}$$ 
(5.5)

(
$\mathcal{\mathcal{F}}$
as in Corollary 3.3 above), though in the odddimensional case log terms can occur with
${x}^{m1}$
.

Proof.
The proof of 5.4 is a straightforward computation using 5.3 in 5.1 .
We first prove
5.5 in the FeffermanGraham expansion approach. We know that
${\overline{g}}_{s}\left({\overline{\theta}}^{m},{\overline{\theta}}^{i}\right)=0\text{for any}i\ne m\text{.}$
A computation like that of Lemma 5.2 gives
${h}_{mi}=xg\left(d\dot{\omega},{\overline{\theta}}^{i}\right)$
. This implies that
${h}_{mi}^{\left(0\right)}=0$
and, together with Lemma 5.2 , that for
$k<m1$
$$\begin{array}{c}\text{if}{h}_{mm}\text{is even up to}{x}^{k}\text{, then}{h}_{mi}\text{is odd up to}{x}^{k+1}\text{.}\end{array}$$ 
(5.6)

This sets up an iterative scheme. In the first step, we use
${h}_{mi}^{\left(0\right)}=0$
to see that the conditions 5.3 are satisfied with
$\alpha =1$
. Then 5.4 shows that
${h}_{uv}^{\left(1\right)}={h}_{mm}^{\left(1\right)}=0$
, and this in turn, by 5.6 , that
${h}_{mi}^{\left(2\right)}=0$
.
Plugging this back into
5.3 with
$\alpha =3$
gives us the next iterative step. This works until
$\alpha =m1$
, which establishes 5.5 .
For the Bianchi gauge approach, consider the operator
$${\Phi}_{g}\left(k\right)=Ri{c}_{k}+(m1)k+{\delta}_{k}^{*}{B}_{g}\left(k\right).$$
Clearly,
${g}_{s}$
is in the kernel of
${\Phi}_{g}$
and
$h$
in that of
${\Phi}_{g}^{\prime}$
. We recall from Proposition 3 of [
27]
that
${\Phi}_{g}^{\prime}$
has no indicial roots between
$0$
and
$m1$
. This will yield 5.5 through the following two observations.
First, we split
$h$
into two pieces,
$h={h}_{e}+{h}_{o}$
, by requiring that
$({h}_{e}{)}_{uv}$
have an even expansion in
$x$
when there are an even number of
$m$
's in
$\{u,v\}$
and have an odd expansion in
$x$
otherwise. It is not hard to see that
${\Phi}_{g}^{\prime}({h}_{e}{)}_{uv}\text{mod}{x}^{m1}$
is even or odd in
$x$
when
$({h}_{e}{)}_{uv}$
is even or odd, and similarly
${\Phi}_{g}^{\prime}({h}_{o}{)}_{uv}$
. Secondly,
${B}_{g}\left({g}_{s}\right)=0$
and
$Sca{l}^{\prime}\left(h\right)=0$
(see 5.2 above) together imply that
${\delta}_{g}\left(h\right)=0$
and
$t{r}_{g}h=0$
.
An easy computation shows that
$({\delta}_{g}h{)}^{\left(0\right)}=0$
implies
${h}_{im}^{\left(0\right)}=0$
.
These observations together show that
${h}_{o}$
vanishes to first order at the boundary, and that
${\Phi}_{g}^{\prime}\left({h}_{o}\right)$
vanishes to order
$m1$
. We conclude that
${h}_{o}$
vanishes to order
$m1$
at the boundary. Thus
$h={h}_{e}\text{mod}{x}^{m1}$
and this is 5.5 . □
We observe that 5.5 shows that the conditions 5.3 are satisfied with
$\alpha =m1$
; this implies by 5.4 that
$(t{r}_{g}h{)}^{(m1)}=0$
.
Once we know the structure of
$h$
, it is a simple matter to compute the variation of the renormalized volume in the evendimensional case (Theorem 2.2 in [
2]
) and the variation of the residue in the odddimensional case (Theorem 1.1 in [
15]
).
Theorem 5.4.
On an evendimensional PE manifold, if
${g}_{s}=g+sh$
to first order with
$h$
as in Proposition 5.3 , the variation of the renormalized volume is
$$\begin{array}{c}{\hat{V}}^{\prime}\left(h\right)=\frac{1}{4}{\int}_{\partial M}\langle {g}^{(m1)},{h}^{\left(0\right)}\rangle ,\end{array}$$ 
(5.7)

the inner product and the integral being taken with respect to
${g}^{\left(0\right)}$
. Similarly, on an odddimensional PE manifold, the variation of the residue of the volume is given by
$$\begin{array}{c}{L}^{\prime}\left(h\right)=\frac{1}{4}{\int}_{\partial M}\langle {g}^{(m1,1)},{h}^{\left(0\right)}\rangle .\end{array}$$ 
(5.8)

Remark In [
15]
, 5.8 is expressed in terms of Branson's
$Q$
curvature and the FeffermanGraham obstruction tensor
$\mathcal{O}$
as
$${\left({\int}_{\partial M}Q\right)}^{\prime}\left(h\right)=(1{)}^{\frac{m1}{2}}\frac{m3}{2}{\int}_{\partial M}\langle \mathcal{O},h\rangle .$$

Proof.
Using Riesz renormalization, we have
$$\begin{array}{c}\begin{array}{cccccccc}\frac{\partial}{\partial s}{}_{s=0}{FP}_{z=0}{\int}_{M}{x}_{s}^{z}{dvol}_{s}& ={FP}_{z=0}{\int}_{M}z{x}^{z1}\dot{x}dvol+{FP}_{z=0}{\int}_{M}{x}^{z}\left(\frac{1}{2}t{r}_{g}h\right)dvol& & ={\int}_{\partial M}{\dot{\omega}}^{(m1)}{dvol}_{0}+{FP}_{z=0}\frac{1}{2(m1)}{\int}_{M}{x}^{z}\left[\Delta \left(t{r}_{g}h\right)+{\delta}_{g}{\delta}_{g}h\right]dvol& & ={\int}_{\partial M}\left[{\dot{\omega}}^{(m1)}+\frac{1}{2}{\left(t{r}_{g}h\right)}^{(m1)}+\frac{1}{2(m1)}{\left({\delta}_{g}h\right)}_{m}^{(m1)}\right]{dvol}_{0}& & =\frac{1}{2(m1)}{\int}_{\partial M}\left[{h}_{mm}^{(m1)}+{\left({\delta}_{g}h\right)}_{m}^{(m1)}\right]{dvol}_{0}.\end{array}\end{array}$$ 
(5.9)

The second equality is 5.2 , the third follows from integrating by parts, and the last uses Lemma 5.2 and the observation made before the statement of the theorem.
Now as in Proposition
5.3 , we can use the structure of
$h$
to compute
$$\begin{array}{c}{\left({\delta}_{g}h\right)}_{m}^{(m1)}={h}_{mm}^{(m1)}\frac{m1}{2}\langle {g}^{(m1)},{h}^{\left(0\right)}\rangle ,\end{array}$$ 
(5.10)

which gives the theorem in the evendimensional case.
In odd dimensions the same proof works. The reason is that the
${x}^{m1}logx$
, being the first
$log$
term, behaves like the first odd term in the evendimensional case. Specifically, it satisfiesconditions 5.3 with
$\alpha =(m1,1)$
which allows us to compute as in 5.9 and 5.10 replacing
$(m1)$
by
$(m1,1)$
. □
To deal with the other terms in the expression for the Pfaffian in section
$\S $
4.3 we use the formalism of double forms, e.g., we interpret
$h$
as an element of
${\mathcal{D}}^{1,1}$
and denote it by
$\mathfrak{h}$
. We benefit greatly from a recent description of the variation of Weyl's volume of tubes invariants by Labbi [
21]
, [
20]
. We denote the integrand in the definition of
${K}_{2\ell}$
by
$${k}_{2\ell}(\Re ):=\frac{{\mathcal{C}}^{2\ell}{\Re}^{\ell}}{(2\ell )!(\ell )!}.$$
Labbi defines the tensors
$$\begin{array}{c}{\mathfrak{E}}_{2\ell}(\Re ):={k}_{2\ell}(\Re )\cdot \mathfrak{g}\frac{{\mathcal{C}}^{2\ell 1}{\Re}^{\ell}}{(2\ell 1)!\ell !},\end{array}$$ 
(5.11)

and establishes that, on any closed Riemannian manifold,
$$\begin{array}{c}{\left({K}_{2\ell}(\Re )\right)}^{\prime}(\mathfrak{h})=\int \langle {\mathfrak{E}}_{2\ell}(\Re ),\mathfrak{h}\rangle .\end{array}$$ 
(5.12)

We will show that a very similar formula holds, on Einstein manifolds, for the variation of the Pfaffian. Our definitions consistently differ from those of Labbi by a factor of
$\ell !$
but agree with those of [
17]
.
The tensors
5.11 were first introduced by Lovelock in [
23]
. His interest was to find all divergencefree symmetric (0,2)tensors built from the metric and its first two covariant derivatives. His theorem is that they are given by arbitrary linear combinations of the
${\mathfrak{E}}_{2\ell}(\Re )$
and the metric. These properties are sometimes used to motivate the Einstein field equation in four dimensions (e.g., 3.7(i) in [
6]
).
Lovelock's generalization plays an analogous role in higher dimensional GB gravity theories (see [
30]
and references therein). In the physics literature, a linear combination of the
${k}_{2\ell}(\Re )$
is known as a Lovelock Lagrangian; an arbitrary linear combination of the
${\mathfrak{E}}_{2\ell}(\Re )$
is known as a Lovelock tensor. Note that
${\mathfrak{E}}_{2}(\Re )$
is the usual stress energy tensor,
$\frac{Scal}{2}gRic$
. Recall that any surface is automatically Einstein and that any higher dimensional Einstein manifold has constant scalar curvature. The following results appear in [
31]
.
Lemma 5.5.
On any
$n$
dimensional manifold, a) (BachLanczos identity) If
$n$
is even,
$${\mathfrak{E}}_{n}(\Re )={\mathfrak{E}}_{n}(\mathfrak{W})=0.$$
b) If
$2\ell <n$
and
${\mathfrak{E}}_{2\ell}(\Re )=\lambda \mathfrak{g}$
for some constant
$\lambda $
, then
${k}_{2\ell}(\Re )$
is constant.

Proof.
a) Any double form in
${\mathcal{D}}^{n,n}$
is of the form
$f{\mathfrak{g}}^{n}$
for some function
$f$
, and it is easy to see that
$$\frac{{\mathcal{C}}^{n}\left(f{\mathfrak{g}}^{n}\right)}{n!\left(\frac{n}{2}\right)!}\mathfrak{g}\frac{{\mathcal{C}}^{n1}\left(f{\mathfrak{g}}^{n}\right)}{(n1)!\left(\frac{n}{2}\right)!}=0.$$
b) We always have
$$\mathcal{C}\left({\mathfrak{E}}_{2\ell}(\Re )\right)=\frac{\left(n2\ell \right)}{\left(2\ell \right)!\ell !}{\mathcal{C}}^{2\ell}{\Re}^{\ell},$$
so if in addition we know that
$\mathcal{C}\left({\mathfrak{E}}_{2\ell}(\Re )\right)=n\lambda $
and
$n\ne 2\ell $
, we find that
${\mathcal{C}}^{2\ell}{\Re}^{\ell}$
is constant. □
Labbi's proof of 5.12 follows from this formula, Lemma 4.2 of [21] (
$\ell \ge 1$
):
$$\begin{array}{c}{\left({k}_{2\ell}(\Re )\right)}^{\prime}(\mathfrak{h})=\frac{1}{2}\langle \frac{{\mathcal{C}}^{2\ell 1}{\Re}^{\ell}}{(2\ell 1)!(\ell !)},\mathfrak{h}\rangle +\frac{(1{)}^{m}}{4}\left(\delta \stackrel{~}{\delta}+\stackrel{~}{\delta}\delta \right)\left(*\left(\frac{{\mathfrak{g}}^{m2\ell}\cdot {\Re}^{\ell 1}}{(m2\ell )!(\ell 1)!}\cdot \mathfrak{h}\right)\right).\end{array}$$ 
(5.13)

where
$*$
extends to double forms via
$$*\left(\omega \otimes \eta \right)=*\omega \otimes *\eta ,$$
and
$\left(\delta \stackrel{~}{\delta}+\stackrel{~}{\delta}\delta \right):{\mathcal{D}}^{1,1}\to {\mathcal{D}}^{0,0}$
is a second order differential operator; all we will need to know here is that its adjoint is twice the Hessian (see comments after equation (10) in [21] ), so that
${\left(\delta \stackrel{~}{\delta}+\stackrel{~}{\delta}\delta \right)}^{*}\left({x}^{z}\right)=2z(1{)}^{m}{x}^{z}\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{},$
with
$\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{}\in {\mathcal{D}}^{1,1}$
the second fundamental form of
$g$
.
For a variation of Einstein metrics preserving the scalar curvature, a very similar formula holds.
The following proposition is proved by adapting Labbi's approach from
$\Re $
to
$\mathfrak{W}$
.
Proposition 5.6.
Consider a family
${g}_{s}$
of PE metrics on an evendimensional manifold, with
${g}_{s}=g+sh$
to first order with
$h$
as in Proposition 5.3 .
a) For
$\ell >2$
$$\frac{\partial}{\partial s}{}_{s=0}\left({}^{R}\int {k}_{2\ell}(\mathfrak{W})\right){=}^{R}\int \left[\frac{\langle {\mathfrak{E}}_{2\ell}(\mathfrak{W}),\mathfrak{h}\rangle}{2}\left(\frac{{s}_{g}}{2m(m1)}\right)\left(m2\ell +1\right)\langle {\mathfrak{E}}_{2\ell 2}(\mathfrak{W}),\mathfrak{h}\rangle \right],$$
and for
$\ell =2$
,
$$\frac{\partial}{\partial s}{}_{s=0}\left({}^{R}\int {k}_{4}(\mathfrak{W})\right)={\frac{1}{2}}^{R}\int \langle {\mathfrak{E}}_{4}(\mathfrak{W}),\mathfrak{h}\rangle +\frac{(m1)(m3)}{4}{\int}_{\partial M}{\langle {g}^{(m1)},{h}^{\left(0\right)}\rangle}_{{g}^{\left(0\right)}}.$$
b) For these variations, the variation of the renormalized integral of the Pfaffian vanishes.
Remark. Notice that
${\mathfrak{E}}_{2}(\mathfrak{W})=0$
because
$\mathcal{C}\mathfrak{W}=0$
. In four dimensions, the functional
$$g\mapsto {K}_{4}(\mathfrak{W})=\int W{}^{2}$$
is well understood. Its gradient is the Bach tensor which vanishes for metrics conformal to Einstein metrics. This is reflected in the vanishing of the interior integral in the formula above for
$\ell =2$
, since by the BachLanczos formula
${\mathfrak{E}}_{4}(\mathfrak{W})=0$
in four dimensions. It would be interesting to understand the behavior of the
${K}_{2\ell}(\mathfrak{W})$
under an arbitrary variation of the metric.

Proof.
Define
$$\begin{array}{cccc}{F}_{\mathfrak{h}}(\Re )& \left((X,Y)(Z,W)\right)& & =h\left(R(X,Y)Z,W\right)h\left(R(X,Y)W,Z\right)+h\left(R(Z,W)X,Y\right)h\left(R(Z,W),Y,X\right).\end{array}$$  
Lemma 4.1 of [21] establishes
$${\Re}^{\prime}(\mathfrak{h})=\frac{1}{4}\left(D\stackrel{~}{D}+\stackrel{~}{D}D\right)(\mathfrak{h})+\frac{1}{4}{F}_{\mathfrak{h}}(\Re ),$$
for some differential operators
$D,\stackrel{~}{D}$
. Simple manipulations using that the metrics
${g}_{s}$
all have the same scalar curvature yield
$$\begin{array}{c}\begin{array}{cccccc}{\mathfrak{W}}^{\prime}\left(h\right)& ={\Re}^{\prime}(\mathfrak{h})\frac{s}{2m(m1)}{\left({\mathfrak{g}}^{2}\right)}^{\prime}(\mathfrak{h})& & =\frac{1}{4}\left(D\stackrel{~}{D}+\stackrel{~}{D}D\right)(\mathfrak{h})+\frac{1}{4}{F}_{\mathfrak{h}}(\Re )\frac{2s}{2m(m1)}\left(\mathfrak{g}\cdot \mathfrak{h}\right)& & =\frac{1}{4}\left(D\stackrel{~}{D}+\stackrel{~}{D}D\right)(\mathfrak{h})+\frac{1}{4}{F}_{\mathfrak{h}}(\mathfrak{W})\frac{s}{2m(m1)}\left(\mathfrak{g}\cdot \mathfrak{h}\right)\end{array}\end{array}$$ 
(5.14)

Labbi's computations in the proof of Lemma 4.2 of [21] go through with
$\Re $
replaced with
$\mathfrak{W}$
and
${\Re}^{\prime}(\mathfrak{h})$
replaced by 5.14 because, on Einstein manifolds, Weyl curvature satisfies the second Bianchi identity. This gives
$$\begin{array}{c}\begin{array}{cccc}{\left({k}_{2\ell}(\mathfrak{W})\right)}^{\prime}(\mathfrak{h})& =\frac{1}{2}\langle \frac{{\mathcal{C}}^{2\ell 1}{\mathfrak{W}}^{\ell}}{(2\ell 1)!(\ell !)},\mathfrak{h}\rangle \left(\frac{{s}_{g}}{2m(m1)}\right)\left(m2\ell +1\right)\langle {\mathfrak{E}}_{2(\ell 1)}(\mathfrak{W}),\mathfrak{h}\rangle & & +\frac{(1{)}^{m}}{4}\left(\delta \stackrel{~}{\delta}+\stackrel{~}{\delta}\delta \right)\left(*\left(\frac{{\mathfrak{g}}^{m2\ell}\cdot {\mathfrak{W}}^{\ell 1}}{(m2\ell )!(\ell 1)!}\cdot \mathfrak{h}\right)\right).\end{array}\end{array}$$ 
(5.15)

Consider the renormalized integral of the last term in equation 5.15 ,
$$\begin{array}{cccc}{FP}_{z=0}{\int}_{M}{x}^{z}(1{)}^{m}\left(\delta \stackrel{~}{\delta}+\stackrel{~}{\delta}\delta \right)\left(*\left({\mathfrak{g}}^{m2\ell}\cdot {\mathfrak{W}}^{\ell 1}\cdot \mathfrak{h}\right)\right)& =2{FP}_{z=0}{\int}_{M}z{x}^{z}\langle \text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{},*\left({\mathfrak{g}}^{m2\ell}\cdot {\mathfrak{W}}^{\ell 1}\cdot \mathfrak{h}\right)\rangle & & =2Res\left[\langle \text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{},*\left({\mathfrak{g}}^{m2\ell}\cdot {\mathfrak{W}}^{\ell 1}\cdot \mathfrak{h}\right)\rangle \right].\end{array}$$  
So this term produces residues, and we need to find the
${x}^{m1}$
term in the expansions. We revisit Lemma 3.1 and ascertain that
$${\langle R({X}_{i},{X}_{j}){X}_{k},{X}_{\ell}\rangle}_{g}^{\left(0\right)}={\left[{\gamma}_{ik}^{m}{\gamma}_{j\ell}^{m}{\gamma}_{jk}^{m}{\gamma}_{i\ell}^{m}\right]}^{\left(0\right)},$$
and
$${\langle R({X}_{i},{X}_{j}){X}_{k},{X}_{\ell}\rangle}_{g}^{(m1)}={\left[{\gamma}_{ik}^{m}{\gamma}_{j\ell}^{m}{\gamma}_{jk}^{m}{\gamma}_{i\ell}^{m}\right]}^{(m1)}.$$
Thus the coefficients of
$\mathfrak{W}$
computed in this frame have no constant term, and the residue vanishes if
$\ell >2$
. To compute the residue for
$\ell =2$
, denote the double form corresponding to
${g}^{\left(0\right)}$
by
${\mathfrak{g}}_{0}$
and similarly
${\mathfrak{g}}_{m1}$
. Note that replacing the coefficients of
$\text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{}$
by their
$0$
th order part produces the double form
${\mathfrak{g}}_{0}$
and replacing the coefficients of
$\mathfrak{W}$
by their
$(m1)$
th order part produces the double form
${\mathfrak{g}}_{0}\cdot \left(\frac{m1}{2}{\mathfrak{g}}_{m1}\right).$
We can use the following formulas from [20]
$$\langle \omega ,*\eta \rangle =\langle *\omega ,\eta \rangle \text{,}\langle \mathfrak{g}\cdot \omega ,\eta \rangle =\langle \omega ,\mathcal{C}\eta \rangle \text{,}*\frac{{\mathfrak{g}}^{k}}{k!}=\frac{{\mathfrak{g}}^{mk}}{(mk)!}\text{, and}{\mathcal{C}}^{k}\left(\frac{{\mathfrak{g}}^{\ell}}{\ell !}\right)=\frac{(m\ell +k)!}{(m\ell )!}\frac{{\mathfrak{g}}^{(\ell k)}}{(\ell k)!},$$
to see that
$$\begin{array}{c}\begin{array}{cccccccc}\frac{1}{2}Res& \left[\langle \text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{},*\left(\frac{{\mathfrak{g}}^{m4}\cdot \mathfrak{W}}{(m4)!}\cdot \mathfrak{h}\right)\rangle \right]=\frac{m1}{4(m4)!}{\int}_{\partial M}\langle {\mathfrak{g}}_{0},*\left({\mathfrak{g}}_{0}^{m4}\cdot {\mathfrak{g}}_{0}\cdot {\mathfrak{g}}_{m1}\cdot {\mathfrak{h}}_{0}\right)\rangle & & =\frac{m1}{4(m4)!}{\int}_{\partial M}\langle (m3)!\frac{{\mathfrak{g}}^{2}}{2},{\mathfrak{g}}_{m1}\cdot {\mathfrak{h}}_{0}\rangle & & =\frac{(m1)(m3)}{4}{\int}_{\partial M}\left[\left({\text{tr}}_{{g}^{\left(0\right)}}{g}^{(m1)}\right)\left({\text{tr}}_{{g}^{\left(0\right)}}{h}^{\left(0\right)}\right){\langle {g}^{(m1)},{h}^{\left(0\right)}\rangle}_{{g}^{\left(0\right)}}\right]& & =\frac{(m1)(m3)}{4}{\int}_{\partial M}{\langle {g}^{(m1)},{h}^{\left(0\right)}\rangle}_{{g}^{\left(0\right)}}.\end{array}\end{array}$$ 
(5.16)

Putting these observations together, we see that the variation of the
$\ell $
th renormalized volume of tube invariant evaluated in the Weyl curvature
$${K}_{2\ell}(\mathfrak{W}){=}^{R}\int \frac{{\mathcal{C}}^{2\ell}\left({\mathfrak{W}}^{\ell}\right)}{(2\ell )!\ell !}dvol$$
is given by
$$\begin{array}{cccccc}{\left({K}_{2\ell}(\mathfrak{W})\right)}^{\prime}(\mathfrak{h})& ={FP}_{z=0}{\int}_{M}z{x}^{z}\dot{\omega}{k}_{2\ell}dvol+{FP}_{z=0}{\int}_{M}{x}^{z}{\dot{k}}_{2\ell}dvol+{FP}_{z=0}{\int}_{M}{x}^{z}{k}_{2\ell}\left(\frac{1}{2}t{r}_{g}h\right)dvol& & ={FP}_{z=0}{\int}_{M}{x}^{z}\left[\frac{\langle {\mathfrak{E}}_{2\ell}(\mathfrak{W}),\mathfrak{h}\rangle}{2}\left(\frac{{s}_{g}}{2m(m1)}\right)\left(m2\ell +1\right)\langle {\mathfrak{E}}_{2\ell 2}(\mathfrak{W}),\mathfrak{h}\rangle \right]dvol& & +Res\left(\dot{\omega}{k}_{2\ell}\right)+\frac{1}{2}Res\left[\langle \text{}\mathcal{\mathcal{I}}\mathcal{\mathcal{I}}\text{},*\left(\frac{{\mathfrak{g}}^{m2\ell}\cdot {\mathfrak{W}}^{\ell 1}}{(m2\ell )!(\ell 1)!}\cdot \mathfrak{h}\right)\rangle \right].\end{array}$$  
The second residue has been computed above for
$\ell =2$
and vanishes otherwise. For the first residue, we know that
$\mathfrak{W}={x}^{2}\overline{\mathfrak{W}}$
(e.g., [19] ) hence neither
$\dot{\omega}$
nor
${k}_{2\ell}(\mathfrak{W})$
have a constant term, and their product does not have an
${x}^{m1}$
term in its expansion, so that residue also vanishes. This establishes part (a) of the proposition.
Using the expression in Lemma
4.4 for the Pfaffian of an Einstein metric, its variation can be read off from part (a) of the proposition and the variation of the volume in Theorem 5.4 . The interior integrals telescope
$$\begin{array}{cccc}\frac{1}{(2\pi {)}^{m/2}}& {}^{R}\int {\sum}_{k=0}^{m/22}\frac{\left(2k\right)!}{k!}{\stackrel{~}{s}}^{k}\left(\frac{\langle {\mathfrak{E}}_{m2k}(\mathfrak{W}),\mathfrak{h}\rangle}{2}\stackrel{~}{s}\left(2k+1\right)\langle {\mathfrak{E}}_{m2k2}(\mathfrak{W}),\mathfrak{h}\rangle \right)& =\frac{1}{(2\pi {)}^{m/2}}& {}^{R}\int {\sum}_{k=0}^{m/22}\frac{\left(2k\right)!}{k!}{\stackrel{~}{s}}^{k}\left(\frac{\langle {\mathfrak{E}}_{m2k}(\mathfrak{W}),\mathfrak{h}\rangle}{2}\stackrel{~}{s}\frac{(2k+1)(2k+2)}{(k+1)}\frac{\langle {\mathfrak{E}}_{m2k2}(\mathfrak{W}),\mathfrak{h}\rangle}{2}\right),\end{array}$$  
where we have abbreviated the scalar curvature factor to
$\stackrel{~}{s}$
. We are left only the
$\langle {\mathfrak{E}}_{m}(\mathfrak{W}),\mathfrak{h}\rangle $
term, which is zero by Lemma 5.5 . There are two terms with residues, one coming from the volume and the other from
$\ell =2$
. The corresponding summands in the formula for the Pfaffian (in Lemma 4.4 ) are
$$\frac{m!}{\left(\frac{m}{2}\right)!}\frac{(1{)}^{\frac{m}{2}}}{{2}^{\frac{m}{2}}}\left[1+\frac{{k}_{4}}{(m1)(m3)}\right],$$
so from 5.16 and Theorem 5.4 , we see that these residues cancel each other out. □
References

Albin, P. The GaussBonnet theorem and Index theory on conformally compact manifolds. Ph.D. Dissertation, Stanford University, in preparation.

Anderson, M. T.
${L}^{2}$
curvature and volume renormalization of AHE metrics on 4manifolds. Math. Res. Lett. 8 (2001), no. 12, 171–188.

Anderson, M. T. Some results on the structure of conformally compact Einstein metrics. preprint, 2005. Available online at arXiv:math.DG/0402198.

Anderson, M. T. Topics in conformally compact Einstein metrics. preprint, 2005. Available online at arXiv:math.DG/0503243.

Avez, A. Characteristic classes and Weyl tensor: Applications to general relativity. Proc. Nat. Acad. Sci. U.S.A. 66 1970 265–268.

Besse, A. L. Einstein manifolds. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), 10. SpringerVerlag, Berlin, 1987. xii+510 pp. ISBN: 3540152792

Biquard, O. Métriques d'Einstein asymptotiquement symétriques. Astérisque No. 265 (2000), vi+109 pp.

Chang, S. Y. A.; Qing, J.; Yang, P. On the Renormalized Volumes for Conformally Compact Einstein Manifolds. Vienna, preprint ESI 1513 (2004), available at www.esi.ac.at/preprints/ESIPreprints.html.

Cheeger, J.; Muller, W.; Schrader, R. Kinematic and tube formulas for piecewise linear spaces. Indiana Univ. Math. J. 35 (1986), no. 4, 737–754.

Chern, S. S. On the curvatura integra in a Riemannian manifold. Ann. of Math. (2) 46, (1945). 674–684.

Donnelly, H. Heat equation and the volume of tubes. Invent. Math. 29 (1975), no. 3, 239–243.

Fefferman, C.; Graham, C. R.
$Q$
curvature and Poincaré metrics. Math. Res. Lett. 9 (2002), no. 23, 139–151.

Graham, C. R. Volume and area renormalizations for conformally compact Einstein metrics. The Proceedings of the 19th Winter School ”Geometry and Physics” (Srni, 1999). Rend. Circ. Mat. Palermo (2) Suppl. No. 63 (2000), 31–42.

Graham, C. R.; Lee, J. M. Einstein metrics with prescribed conformal infinity on the ball. Adv. Math. 87 (1991), no. 2, 186–225.

Graham, C. R.; Hirachi, K. The Ambient Obstruction Tensor and QCurvature. to appear in a proceeding volume, available online at arXiv:math.DG/0405068.

Graham, C. R.; Witten, E. Conformal anomaly of submanifold observables in AdS/CFT correspondence. Nuclear Phys. B 546 (1999), no. 12, 52–64.

Gray, A. Tubes. Second edition. Progress in Mathematics, 221. Birkhuser Verlag, Basel, 2004. xiv+280 pp. ISBN: 3764369078

Guillarmou, C. Meromorphic properties of the resolvent on asymptotically hyperbolic manifolds. to appear in Duke Math. J., available online at arXiv:math.SP/0311424.

Kulkarni, R. S. On the Bianchi Identities. Math. Ann. 199 (1972), 175–204.

Labbi, M. L. Double forms, curvature structures and the
$(p,q)$
curvatures. preprint, 2004, available online at arXiv:math.DG/0404081.

Labbi, M. L. On a variational formula for the H. Weyl curvature invariants. preprint, 2004, available online at arXiv:math.DG/0406548.

Lee, J. M. Fredholm operators and Einstein metrics on conformally compact manifolds. preprint, 2001, available online at arXiv:math.DG/0105046.

Lovelock, D. The Einstein tensor and its generalizations. J. Mathematical Phys. 12 1971 498–501.

Mazzeo, R. The Hodge cohomology of a conformally compact metric. J. Differential Geom. 28 (1988), no. 2, 309–339.

Mazzeo, R. Elliptic theory of differential edge operators. I. Comm. Partial Differential Equations 16 (1991), no. 10, 1615–1664.

Mazzeo, R. R.; Melrose, R. B. Meromorphic extension of the resolvent on complete spaces with asymptotically constant negative curvature. J. Funct. Anal. 75 (1987), no. 2, 260–310.

Mazzeo, R. R.; Pacard, F. Maskit combinations of PoincaréEinstein metrics. preprint, 2002, available online at arXiv:math.DG/0211099.

Melrose, R. B. The AtiyahPatodiSinger index theorem. Research Notes in Mathematics, 4. A K Peters, Ltd., Wellesley, MA, 1993. xiv+377 pp. ISBN 1568810024

Melrose, R. B.; Nistor, V. Homology of pseudodifferential operators I. Manifolds with boundary. preprint, 1996, available online at arXiv:functan/9606005.

Padilla, A. Surface terms and the GaussBonnet Hamiltonian. Classical Quantum Gravity 20 (2003), no. 14, 3129–3149.

Patterson, E. M. A class of critical Riemannian metrics. J. London Math. Soc. (2) 23 (1981), no. 2, 349–358.

Patterson, S. J.; Perry, P. A. The divisor of Selberg's zeta function for Kleinian groups. Appendix A by Charles Epstein. Duke Math. J. 106 (2001), no. 2, 321–390.

Paycha, S. From heatoperators to anomalies; a walk through various regularization techniques in mathematics and physics Emmy Noether Lectures delivered at the University of Goettingen, AprilMay 2003 (available online at http://wwwlma.univbpclermont.fr/ paycha/publications.html)

Witten, E. Anti de Sitter space and holography. Adv. Theor. Math. Phys. 2 (1998), 253291, arXiv:hepth/9802150.
Department of Mathematics, Stanford University Email address : pierre@math.stanford.edu