# Black hole pair creation in de Sitter space: a complete one-loop analysis

###### Abstract

We present an exact one-loop calculation of the tunneling process in Euclidean quantum gravity describing creation of black hole pairs in a de Sitter universe. Such processes are mediated by gravitational instantons giving an imaginary contribution to the partition function. The required energy is provided by the expansion of the universe. We utilize the thermal properties of de Sitter space to describe the process as the decay of a metastable thermal state. Within the Euclidean path integral approach to gravity, we explicitly determine the spectra of the fluctuation operators, exactly calculate the one-loop fluctuation determinants in the -function regularization scheme, and check the agreement with the expected scaling behaviour. Our results show a constant volume density of created black holes at late times, and a very strong suppression of the nucleation rate for small values of .

and

^{†}

^{†}thanks: Supported by the DFG

## 1 Introduction

Instantons play an important role in flat space gauge field theory [45]. Being stationary points of the Euclidean action, they give the dominant contribution to the Euclidean path integral thus accounting for a variety of important phenomena in QCD-type theories. In addition, self-dual instantons admit supersymmetric extensions, which makes them an important tool for verifying various duality conjectures like the AdS/CFT correspondence [4]. More generally, the Euclidean approach has become the standard method of quantum field theories in flat space.

Since the theory of gravity and Yang-Mills theory are somewhat similar, it is natural to study also gravitational instantons. An impressive amount of work has been done in this direction, leading to a number of important discoveries. A thorough study of instanton solutions of the vacuum Einstein equations and also those with a -term has been carried out [20, 17, 24]. These solutions dominate the path integral of Euclidean quantum gravity, leading to interesting phenomena like black hole nucleation and quantum creation of universes. Perhaps one of the most spectacular achievements of the Euclidean approach is the derivation of black hole entropy from the action of the Schwarzschild instanton [22]. In addition, gravitational instantons are used in the Kaluza-Klein reductions of string theory.

Along with these very suggestive results, the difficulties of Euclidean quantum gravity have been revealed. Apart from the usual problem of the non-renormalizability of gravity, which can probably be resolved only at the level of a more fundamental theory like string theory, the Euclidean approach presents other challenging problems. In field theories in flat space the correlation functions of field operators are holomorphic functions of the global coordinates in a domain that includes negative imaginary values of the time coordinate, , where is real and positive [48]. This allows one to perform the analysis in the Euclidean section and then analytically continue the functions back to the Lorentzian sector to obtain the physical predictions. In curved space the theorems that would ensure the analyticity of any quantities arising in quantum gravity are not known. As a result, even if Euclidean calculations make sense, it is not in general clear how to relate their result to the Lorentzian physics.

This difficulty is most strikingly illustrated by the famous problem of the conformal sector in Euclidean quantum gravity. If one tries evaluating the path integral over Riemannian metrics, then one discovers that it diverges because the Euclidean gravitational action is not bounded from below and can be made arbitrarily large and negative by conformal rescaling of the metric [25]. Such a result is actually expected, for if the integral did converge (with some regularization), then one could give a well-defined meaning to the canonical ensemble of the quantum gravitational field. However, the possibility of having a black hole causes the canonical ensemble to break down – since the degeneracy of black hole states grows faster than the Boltzmann factor decreases. One can, ‘improve’ the Euclidean gravitational action by analytically continuing the conformal modes, let us call them , via , and this improves the convergence of the integral [25]. This shows that if there is a well-defined Euclidean path integral for the gravitational field, then the relation to the Lorentzian sector is more complicated than just via .

Unfortunately, it is unknown at present whether one can in the general case find a physically well-defined and convergent path integral for the gravitational field. At the same time, the idea of constructing it is conceptually simple [46]: one should start from the Hamiltonian path integral over the physical degrees of freedom of the gravitational field. Such an integral certainly makes sense physically and is well-convergent, since the Hamiltonian is positive – at least in the asymptotically flat case. The Hamiltonian approach is not covariant, but one can covariantize it by changing the integration variables, which leads to a manifestly covariant and convergent path integral for gravity. The main problem with this program is that in the general case it is unclear how to isolate the physical degrees of freedom of the gravitational field. For this reason, so far the program has been carried out only for weak fields in the asymptotically flat case [46]. Remarkably, the result has been shown to exactly correspond to the the standard Euclidean path integral with the conformal modes being complex-rotated via . This lends support to the Euclidean approach in gravity and allows one to hope that the difficulties of the method can be consistently resolved; (see, for example, [7, 6] for the recent new developments within the lattice approach).

One can adopt the viewpoint that Euclidean quantum gravity is a meaningful theory within its range of applicability, at least at one-loop level, by assuming that a consistent resolution of its difficulties exists. Then already in its present status the theory can be used for calculating certain processes, most notably for describing tunneling phenomena, in which case the Euclidean amplitude directly determines the probability. The analytic continuation to the Lorentzian sector in this case is not necessary, apart from when the issue of the interpretation of the corresponding gravitational instanton is considered. The important example of a tunneling process in quantum gravity is the creation of black holes in external fields. Black holes are created whenever the energy pumped into the system is enough in order to make a pair of virtual black holes real [33]. The energy can be provided by the heat bath [30, 38, 5], by the background magnetic field [21, 19, 16, 15], by the expansion of the universe [28, 10, 41], by cosmic strings [37], domain walls [11], etc; (see also [43, 35, 36]). Besides, one can consider pair creation of extended multidimensional objects like -branes due to interaction with the background supergravity fields [14]. In all these examples the process is mediated by the corresponding gravitational instanton, and the semiclassical nucleation rate for a pair of objects on a given background is given by

(1.1) |

Here is the classical action of the gravitational instanton mediating creation of the objects, is the action of the background fields alone, and the prefactor includes quantum corrections.

In most cases the existing calculations of black hole pair creation processes consider only the classical term in (1.1). This is easily understood, since loop calculations in quantum gravity for non-trivial backgrounds are extremely complicated. To our knowledge, there is only one example of a next-to-leading-order computation, which was undertaken in [30] by Gross, Perry, and Yaffe for the Schwarzschild instanton background. The aim of the present paper is to consider one more example of a complete one-loop computation in quantum gravity.

The problem we are interested in is the quantum creation of black holes in de Sitter space. This problem was considered by Ginsparg and Perry [28], who identified the instanton responsible for this process, which is the solution of the Euclidean Einstein equations for . Ginsparg and Perry noticed that this solution has one negative mode in the physical sector, which renders the partition function complex, thus indicating the quasi-classical instability of the system. This instability leads to spontaneous nucleation of black holes in the rapidly inflating universe. This is the dominant instability of de Sitter space, since classically the space is stable [28]. The energy necessary for the nucleation is provided by the -term, which drives different parts of the universe apart thereby drugging the members of a virtual black hole pair away from each other. The typical radius of the created black holes is , while the the nucleation rate is of the order of , where is Newton’s constant. As a result, for when inflation is fast, the black holes are produced in abundance but they are small and presumably almost immediately evaporate. Large black holes emerge for when inflation slows down, and these can probably exist for a long time, but the probability of their creation is exponentially small. This scenario was further studied in Refs.[10, 9, 18] (see also references in [9]), where the generalization to the charged case was considered and also the subsequent evolution of the created black holes was analyzed. However, the one-loop contribution so far has not been computed.

A remarkable feature of the instanton is its high symmetry. In what follows, we shall utilize this symmetry in order to explicitly determine spectra of all relevant fluctuation operators in the problem. We shall use the -function regularization scheme in order to compute the one-loop determinants, which will give us the partition function for the small fluctuations around the instanton. We shall then need to normalize this result. The normalization coefficients is , the partition function for small fluctuations around the instanton, which is the Euclidean version of the de Sitter space. The one-loop quantization around the instanton was considered by Gibbons and Perry [27], and by Christensen and Duff [13], but unfortunately in none of these papers the analysis was completed. We shall therefore reconsider the problem by rederiving the spectra of fluctuations around and computing the determinants within the -function scheme, thereby obtaining a closed one-loop expression for .

In our treatment of the path integral we follow the approach of Gibbons and Perry [27]; (see also [42]). In order to have control over the results, we work in a one-parameter family of covariant gauges and perform the Hodge decomposition of the fluctuations. These are then expanded with respect to the complete sets of basis harmonics, and the perturbative path integration measure is defined as the square root of the determinant of the metric on the function space of fluctuations. To insure the convergence of the integral over the conformal modes, which enter the action with the wrong sign, we essentially follow the standard recipe [25]; (see also Ref.[42], where a slightly disguised form of the same prescription was advocated). The subtle issue is that the conformal operator has a finite number, , of negative modes, and these enter the action with the correct sign from the very beginning. Our treatment of these special modes is different from that by Hawking [32], who suggests that such modes should be complex-rotated twice, the partition function then acquiring the overall factor of . However, the presence of this factor in the partition function would lead to unsatisfactory results, and on these grounds we are led to not rotating the special conformal modes at all.

The path integral is computed by integrating over the Fourier expansion coefficients, which leads to infinite products over the eigenvalues. The only conformal modes giving contribution to the result are the special negative modes discussed above. We carefully analyze the resulting products to make sure that all modes are taken into account and that the dependence of the gauge-fixing parameter cancels thereby indicating the correctness of the procedure. We give a detailed consideration to the zero modes of the Faddeev-Popov operator, which arise due to the background isometries. The integration over these modes requires a non-perturbative extension of the path-integration measure, and we find such a non-perturbative measure in the zero mode sector to be proportional to the Haar measure of the isometry group. Collecting all terms yields the partition function for small fluctuations around a background instanton configuration in terms of infinite products over eigenvalues of the gauge-invariant operators. We then use the explicitly known spectra of fluctuations around the and backgrounds in order to calculate the partition functions.

The rest of the paper is organized as follows. In Sec.2 we present our derivation of the black hole nucleation rate within the finite temperature approach. In Sec.3 the path integration procedure is considered. The spectra of small fluctuations around the instanton are computed in Sec.4 via a direct solving of the differential equations in the eigenvalue problems. The spectra of the fluctuations around the instanton are rederived in Sec.5 with the use of group theoretic arguments. The partition functions are computed in Sec.6, and Sec.7 contains the final expression for the black hole nucleation rate together with some remarks. We present a detailed analysis of the -functions in the Appendix. We use units where .

## 2 Black hole nucleation rate

In this section we shall derive the basic formula for the black hole nucleation rate in de Sitter space, whose different parts will be evaluated in the next sections. The existing derivations of the nucleation rate [28, 10] recover only the classical factor in (1.1). In addition, it is not always clear to which volume the rate refers. We argue that our formula (2.15) gives the nucleation probability per Hubble volume and unit time as measured by a freely falling observer. The basic idea of our approach is to utilize the relation between the inflation and thermal properties of de Sitter space. This will allow us to use the standard theory of decay of metastable thermal states [39, 40, 3].

Let us consider the partition function for the gravitational field

(2.2) |

where the integral is taken over Riemannian metrics, and is the Euclidean action for gravity with a positive terms; see Eq.(3.1) below. The path integration procedure will be considered in detail in the next section. At present let us only recall that in the semiclassical approximation the integral is approximated by the sum over the classical extrema of the action , that is

(2.3) |

Here is the partition function for the small gravitational fluctuations around a background manifold with a metric subject to the Euclidean Einstein equations . Schematically one has

(2.4) |

where is the classical action for the -th extremum, and is the operator for the small fluctuations around this background.

The dominant contribution to the sum in (2.4) is given by the instanton, which is the four-dimensional sphere with the radius and the standard metric. Since this is a maximally symmetry space, its action is less than that of any other instanton. Hence,

(2.5) |

On the other hand, the instanton describes the thermal properties of de Sitter space [22, 23], since it can be obtained by an analytic continuation via of the region of the de Sitter solution

(2.6) |

contained inside the event horizon, . Let us call this region a Hubble region. Its boundary, the horizon, has the area . The temperature associated with this horizon is and the free energy . The same values can be obtained by writing the partition function for the instanton as , the entropy

(2.7) |

with . Indeed, since is periodic in all four coordinates, any of them can be chosen to be the ‘imaginary time’. The corresponding period, , can be identified with the proper length of a geodesic on , all of which are circles with the same length. This gives the correct de Sitter temperature. Comparing (2.7) and (2.5) one obtains , the dots denoting the quantum corrections, and this again agrees with the result for the de Sitter space. To recapitulate, the partition function of quantum gravity with is approximately

(2.8) |

where is the de Sitter temperature and is the free energy in the Hubble region.

Let us now consider the contribution of the other instantons. One has

(2.9) |

where the prime indicates that . Now, for all terms in the sum are exponentially small and can safely be neglected as compared to the unity, if only they are real. If there are complex terms, then they will give an exponentially small imaginary contribution. The instanton is distinguished by the fact that its partition function is purely imaginary due to the negative mode in the physical sector [28].

This is the only solution for which is not a local minimum of the action in the class of metrics with constant scalar curvature [20]. Hence (see Fig.1),

(2.10) |

where is purely imaginary. As a result, the partition function can still be represented as , where the real part of is the free energy of the Hubble region, and the exponentially small imaginary part is given by

(2.11) |

It is natural to relate this imaginary quantity also to the free energy. We are therefore led to the conclusion that the free energy of the Hubble region has a small imaginary part, thus indicating that the system is metastable. The decay of this metastable state will lead to a spontaneous nucleation of a black hole in the Hubble region, which can be inferred from the geometrical properties of the instanton.

The instanton can be obtained via the analytic continuation of the Schwarzschild-de Sitter solution [26, 28, 10]

(2.12) |

Here this function has roots at (black hole horizon) and at (cosmological horizon). One has for , and only this portion of the solution can be analytically continued to the Euclidean sector via . The conical singularity at either of the horizons can be removed by a suitable identification of the imaginary time. However, since the two horizons have different surface gravities, the second conical singularity will survive. The situation improves in the extreme limit, , since the surface gravities are then the same and both conical singularities can be removed at the same time. Although one might think that the Euclidean region shrinks to zero when the two horizons merge, this is not so. The limit implies that with . One can introduce new coordinates and via and . Passing to the new coordinates and taking the limit , the result is , and for

(2.13) |

and this metric fulfills the Einstein equations. Since the instanton field determines the initial value for the created real time configuration, one concludes that the instanton is responsible for the creation of a black hole in the Hubble region. This black hole fills the whole region, since its size is equal to the radius of the cosmological horizon.

It is well known that the region of the static coordinate system in (2.6) covers only a small portion of the de Sitter hyperboloid [47]; (see Fig.2). In order to cover the whole space, one can introduce an infinite number of freely falling observers and associate the interior of the static coordinate system with each of them. Hence, the spacetime contains infinitely many Hubble regions. It is also instructive to use global coordinates covering the whole de Sitter space,

(2.14) |

where and . The trajectory of a freely falling observer is (and also , ), and the domain of the associated static coordinate system, the Hubble region, is the intersection of the interiors of the observer’s past and future horizons [34]. Let be a spacelike hypersurface, say . If then is completely contained inside the Hubble region of a single observer with (see Fig.2). However, for late moments of time, , one needs more and more independent observers in order to completely cover by the union of their Hubble regions. One can say that the Hubble regions proliferate with the expansion of the universe.

Since de Sitter space consists of infinitely many Hubble regions, the black hole nucleation will lead to some of the regions being completely filled by a black hole, but most of the regions will be empty. The number of the filled regions divided by the number of those without a black hole is the probability for a black hole nucleation in one region. This is proportional to in (2.11).

One can argue that the black holes are actually created in pairs [33, 36], where the two members of the pair are located at the antipodal points of the de Sitter hyperboloid. This can be inferred from the conformal diagram of the Schwarzschild-de Sitter solution, which contains an infinite sequence of black hole singularities and spacelike infinities; see Fig.3. One can identify the asymptotically de Sitter regions in the diagram related by a horizontal shift, and the spacetime will then consist of two black holes at antipodal points of the closed universe. This agrees with the standard picture of particles in external fields being created in pairs.

The surface gravity of the extreme Schwarzschild-de Sitter solution is finite when defined with respect to the suitably normalized Killing vector [10]. This gives a non-zero value for the temperature of the nucleated black holes, which can be read off also from the metric: it is the inverse proper length of the equator of any of the two spheres, . How can it be that this is different from the temperature of the heat bath, which is the de Sitter space with 30]. However, the global structure of de Sitter space is different from that of Minkowski space. The fluctuations cannot absorb energy from and emit energy into the whole of de Sitter space, but can only exchange energy with the Hubble region. Thus the energy exchange is restricted. As a result, the local temperature in the vicinity of a created defect may be different from that of the heat bath, but reduces to the latter in the asymptotic region far beyond the cosmological horizon. ? For example, in the hot Minkowski space the nucleated black holes have the same temperature as the heat bath [

The relation of the imaginary part of the free energy to the
rate of decay of a metastable thermal state
was considered in [39, 40, 3].
If the decay is only due to tunneling then .
Suppose that
there is an additional possibility to classically jump over
the potential barrier. In this case on top of the barrier
there is a classical saddle point configuration
whose real
time decay rate is determined by the saddle negative mode
.
At low temperatures the tunneling formula is then still correct,
while for one has
.
In our problem the saddle point configuration also exists, the
instanton, but its real time analog,
the Schwarzschild-de Sitter black hole, is stable.
It seems therefore that there is no classical contribution to
the process and the black hole nucleation is a purely quantum
phenomenon.^{1}^{1}1We do not understand the
classical interpretation of the Euclidean saddle point solution
suggested in [30].
The argument uses a family of non-normalizable deformations
of the instanton, and the action is finite as long as they are ‘static’.
However, if one considers a time evolution along such a family then
the action will be infinite, which shows that the classical picture
does not apply. Even if one uses the classical formula for in this case,
one arrives at the quantum result, since =const..
[One can imagine that the effective potential barrier
is infinitely high, such that a classical
transition is forbidden,
but at the same time so narrow that the tunneling rate is non-zero.]
As a result, the rate
of quasiclassical decay of the de Sitter space is given by
. Using Eq.(2.11),

(2.15) |

Here . Since is the proper time of the geodesic observer resting at the origin of the static coordinate system (2.6), we conclude that the formula gives the probability of a black hole nucleation per Hubble volume and unit time of a freely falling observer. is the temperature of the de Sitter heat bath, which was originally defined with respect to the analytically continued Killing vector

In order to use the formula (2.15), we should be able to compute the one-loop partition functions and . Now we shall calculate them within the path integral approach.

## 3 The path integration procedure

In this section we shall consider the path integral for fluctuations around an instanton solution of the Einstein equations in the stationary phase approximation. We shall largely follow the approach of Gibbons and Perry [27].

### 3.1 The second variation of the action

Our starting point is the action for the gravitational field on a compact Riemannian manifold ,

(3.1) |

whose extrema, , are determined by the equations

(3.2) |

Let be an arbitrary solution, and consider small fluctuations around it, . The action expands as

(3.3) |

where is quadratic in and dots denote the higher order terms. One can express directly in terms if . However, it is convenient to use first the standard decomposition of ,

(3.4) |

Here is the transverse tracefree part, , is the trace, and the piece due to is the longitudinal tracefree part. Under the gauge transformations (general diffeomorphisms) generated by one has The TT-tensor is gauge-invariant, while the trace changes as . It follows that

(3.5) |

is gauge-invariant. For further references we note that can in turn be decomposed into its coexact part , for which , the exact part , and the harmonic piece ,

(3.6) |

The number of square-integrable harmonic vectors is a topological invariant, which is equal to the first Betti number of the manifold . Since the latter is zero if is simply-connected, which is the case for , we may safely ignore the harmonic contribution in what follows.

With the decomposition (3.4) the second variation of the action in (3.3) is expressed in terms of the gauge-invariant quantities and alone,

(3.7) |

Here and below we consider the following second order differential operators: the operator for the TT-tensor fluctuations

(3.8) |

the vector operator acting on coexact vectors

(3.9) |

and the scalar operators for , , and

(3.10) |

with being a real parameter. Since for the manifold is compact, these operators are (formally) self-adjoint with respect to the scalar product

(3.11) |

similarly for vectors and scalars .

The action in (3.7) contains only the gauge-invariant amplitudes and , while the dependence on the gauge degrees of freedom cancels. Pure gauge modes are thus zero modes of the action. Fixing of the gauge is therefore necessary in order to carry out the path integration. To fix the gauge we pass from the action to the gauge-fixed action

(3.12) |

where, following [27], we choose the gauge-fixing terms as

(3.13) |

with and being real parameters. We shall shortly see that it is convenient to choose [27]

(3.14) |

This choice, however, implies that does not vanish for . It is often convenient to set , in which case . However, we shall not fix the value of , since this will provide us with a check of the gauge-invariance of our results.

Using the decompositions (3.4), (3.6) the gauge-fixing term reads

(3.15) |

Adding this up with in (3.7) one obtains the gauge-fixed action . It is now convenient to pass from the gauge-invariant variable defined in (3.5) back to the trace , since with the choice in (3.14) the resulting action then becomes diagonal:

This action generically has no zero modes, but it depends on the arbitrary parameter , which reflects the freedom of choice of gauge-fixing. In order to cancel this dependency, the compensating ghost term is needed.

### 3.2 The mode decomposition of the action

We wish to calculate the path integral

(3.17) |

where is the classical action, and the Faddeev-Popov factor is obtained from

(3.18) |

In order to perform the path integration, we introduce the eigenmodes associated with the operators , and :

(3.19) |

Throughout this paper we shall denote the eigenvalues and eigenfunctions of the tensor operator by and , and those for the vector operator by and , respectively. [Later we shall use the symbol also for the argument of the -functions, and this will not lead to any confusion]. Eigenvalues of the scalar operator will be denoted by , and it will be convenient to split the set into three subsets, , where , , and ; see Eqs.(3.25)–(3.27) below. Accordingly, the set of the scalar eigenfunctions will be split as .

Since the manifold is compact, we choose the modes to be orthonormal. This allows us to expand all fields in the problem as

(3.20) |

and

(3.21) |

As a result, the action decomposes into the sum over modes, and the path integral reduces to integrals over the Fourier coefficients.

a) Vector and tensor modes.– Let us consider the mode decomposition for the gauge-fixed action in (3.1). This action is the sum of four terms. For the first two terms we obtain

(3.22) | |||||

(3.23) |

These quadratic forms should be positive definite, since otherwise the integrals over the coefficients would be ill-defined. We can see that the quadratic form in (3.23) for the vector modes is indeed non-negative definite. Next, the expression in (3.22) for the gauge-invariant tensor modes is positive-definite if all eigenvalues are positive. If there is a negative eigenvalue, , as in the case of the instanton background, then it is physically significant. The integration over is performed with the complex contour rotation, which renders the partition function imaginary thus indicating the quasiclassical instability of the system.

Let us consider now the contribution of the longitudinal vector piece to the action (3.1). We obtain

(3.24) |

where and are the eigenvalues of and , respectively. We note that while , the and can be negative and should therefore be treated carefully. Let us split the scalar modes into three groups according to the sign of :

(3.25) | |||

(3.26) | |||

(3.27) |

First we consider the constant mode in (3.25). This exists for any background, and for compact manifolds without boundary this is the only normalizable zero mode of . Since this mode is annihilated by , it does not contribute to the sum in (3.24).

Consider now the scalar modes with the eigenvalue in (3.26). In view of the Lichnerowicz-Obata theorem [49], the lowest non-trivial eigenvalue of for is bounded from below by , and the equality is attained if only the background is . Hence the modes in (3.26) exist only for the instanton, and there can be no modes ‘in between’ (3.25) and (3.26). In the case there are five scalar modes with the eigenvalue , and their gradients are the five conformal Killing vectors that do not correspond to infinitesimal isometries. If , a theorem of Yano an Nagano [49] states that such vectors exist only in the case. Let us call these five scalar modes ‘conformal Killing modes’. Notice that these also do not contribute to the sum in (3.24).

To recapitulate, the lowest lying modes in the scalar spectrum are the constant conformal mode in (3.25), which exists for any background, and also 5 ‘conformal Killing modes’ in (3.26) which exist only for the instanton and generate the conformal isometries. As we shall see, these 1+5 lowest lying modes are physically distinguished, since they are the only scalar modes contributing to the partition function. However, they do not enter the sum in (3.24).

For the remaining infinite number of scalar modes in (3.27) (these are labeled by ) the eigenvalues and are positive, and it is not difficult to see that all the ’s are also positive, provided that the gauge parameter is positive and large enough. To recapitulate, the contribution of the longitudinal vector modes to the action is given by

(3.28) |

which is positive definite. We shall see that all modes contributing to this sum are unphysical and cancel from the path integral.

b) Conformal modes.– We now turn to the last term in the gauge-fixed action (3.1). Using (3.25)–(3.27) we obtain

(3.29) |

The expression on the right has a finite number of positive terms, corresponding to the distinguished lowest lying modes, and infinitely many negative ones. As a result, an increase in the coefficients makes it arbitrarily large and negative, thus rendering the path integral divergent. This represents the well-known problem of conformal modes in Euclidean quantum gravity [25]. A complete solution of this problem is lacking at present, but the origin of the trouble seems to be understood [46]. In brief, the problem is not related to any defects of the theory itself, but arises as a result of the bad choice of the path integral. If one starts from the fundamental Hamiltonian path integral over the physical degrees of freedom of the gravitational field, then one does not encounter this problem. The Hamiltonian path integral, however, is non-covariant and difficult to work with. One can ‘covariantize’ it by adding gauge degrees of freedom, and this leads to the Euclidean path integral described above, up to the important replacement [25]

(3.30) |

The effect of this is to change the overall sign in (3.29), such that the infinite number of negative modes become positive. Unfortunately, such a consistent derivation of the path integral has only been carried out for weak gravitational fields [46] (and for ), since otherwise it is unclear how to choose the physical degrees of freedom. Nevertheless, the rule (3.30) is often used also in the general case [25], and it leads to the cancellation of the unphysical conformal modes. However, some subtle issues can arise.

For the expression in (3.29) contains, apart from infinitely many negative terms, also a finite number of positive ones, which are due to the distinguished lowest lying scalar modes. If we apply the rule (3.30) and change the overall sign of the scalar mode action, then the negative modes will become positive, but the positive ones will become negative. As a result, the path integral will still be divergent. It was therefore suggested by Hawking that the contour for these extra negative modes should be rotated back, the partition function then acquiring the factor , where is the number of such modes [32]. As we know, for the instanton, and for any other solution of with .

Unfortunately, this prescription to rotate the contour twice leads in some cases to physically meaningless results; the examples will be given in a moment. We suggest therefore a slightly different scheme: not to touch the positive modes in (3.29) at all and to change the sign only for the negative modes. The whole expression then becomes

(3.31) |

We make no attempt to rigorously justify such a rule. We note, however, that it is essentially equivalent to the standard recipe (3.30) – up to a finite number of modes which we handle differently as compared to Hawking’s prescription. We shall now comment on this difference.

When compared to Hawking’s recipe [32], the ultimate effect of our prescription is to remove the factor from the partition function. We are unaware of any examples where it would be necessary to insist on this factor being present in the final result. On the contrary, the examples are in favour of the factor being absent. For the instanton one has , such that , and this would render the partition function for hot gravitons in a de Sitter universe negative, which would be physically meaningless. Next, for the instanton, which already has one negative mode in the spin-2 sector, one has . As a result, the factor would make the partition function real instead of being imaginary, and there would be no black hole pair creation !

These arguments suggest that Hawking’s rule should be somehow modified, and we therefore put forward the prescription (3.31). Let us also note that our rule leads to gauge invariant results – the dependence on the gauge parameter cancels after the integration. Finally we note that the lowest lying scalar modes are physically distinguished, and since they are positive, they should be treated similarly to the physical tensor modes.

To recapitulate, the mode expansion of the gauge-fixed action is given by the sum of (3.22), (3.23), (3.28), and (3.31):

(3.32) |

In a similar way we obtain the following mode expansion for the gauge-fixing term in (3.15):

(3.33) | |||||

This expression is non-negative definite.

### 3.3 The path integration measure

In order to compute the path integrals in (3.17),(3.18) we still need to define the path-integration measure. The perturbative measure is defined as the square root of the determinant of the metric on the function space of fluctuations:

(3.34) |

Here it is assumed that the fluctuations are Fourier-expanded and the differentials refer to the Fourier coefficients, while the meaning of the proportionality sign will become clear shortly. Let us first consider . It follows from (3.4),(3.6) that

(3.35) |

Expanding the fields on the right according to (3.20),(3.21) and differentiating with respect to the Fourier coefficients we obtain the metric for the vector fluctuations

(3.36) |

which yields

(3.37) |

Here the prime indicates that terms with do not contribute to the sum in (3.36), and should therefore be omitted in the product in (3.37). To obtain the measure we endow each term in the products in (3.37) with the weight factor :

(3.38) |

Such a normalization implies that

(3.39) |

Here is a parameter with the dimension of an inverse length. In a similar way we obtain the measure , which is normalized as

(3.40) |

we shall shortly comment on the relative normalization of and . The result is

(3.41) | |||||

Here the prime indicates that the zero modes of the vector fluctuation operator do not contribute to the product. Notice, however, that these modes do contribute to the measure .

The following remarks are in order. We use units where all fields and parameters have dimensions of different powers of a length scale . One has . Eigenvalues of all fluctuation operators have the dimension . The coordinates are dimensionless, while . For the vectors, , and for the scalars and We assume that the scalar, vector and tensor eigenfunctions in (3.19) are orthonormal with respect to the scalar product in (3.11). As a result, the dimensions of the eigenfunctions are , , , which gives for the Fourier coefficients in (3.20),(3.21) , , and .

The normalization of can be arbitrary, which is reflected in the presence of the arbitrary parameter in the above formulas. However, the relative normalization of and , which is defined by Eqs.(3.39) and (3.40) is fixed by gauge invariance. Had we chosen instead a different relative normalization, say dividing each mode in (3.38) by 2, then the path integral would acquire a factor of , where is the ‘number of eigenvalues’ of the non-gauge-invariant operator . [The issue of relative normalization of the fluctuation and Faddeev-Popov determinants seldom arises, since in most cases the absolute value of the path integral is irrelevant].

### 3.4 Computation of the path integral

Now we are ready to compute the path integrals in (3.17),(3.18). Let us illustrate the procedure on the example of Eq.(3.18), which determines the Faddeev-Popov factor . Using from Eq.(3.33) and the measure from (3.38) we obtain

(3.42) | |||

which gives

(3.43) |

#### 3.4.1 Zero modes

The factor in (3.43) arises due to the gauge zero modes, for which and the integral is non-Gaussian:

(3.44) |

with the product taken over all such modes. The existence of zero modes of the Faddeev-Popov operator indicates that the gauge is not completely fixed. This can be related to the global aspects of gauge fixing procedure known as the Gribov ambiguity. However, Gribov’s problem is usually not the issue in the perturbative calculations, where zero modes arise rather due to background symmetries. This will be the case in our analysis below. Specifically, the isometries of the background manifold form a subgroup of the full diffeomorphism group. Sometimes is called the stability group; for the and backgrounds is SO(5) and SO(3)SO(3), respectively. Since the isometries do not change (in the linearized approximation), their generators, which are the Killing vectors , are zero modes of the Faddeev-Popov operator.

We therefore conclude that the integration in (3.44) is actually performed over the stability group . Since the latter is compact in the cases under consideration, the integral is finite. In order to actually compute the integral, some further analysis is necessary, in which we shall adopt the approach of Osborn [44]. First of all, let us recall that all eigenmodes in our analysis have unit norm. If we now rescale the zero modes such that the Killing vectors become dimensionless (remember that the coordinates are also dimensionless), then the expression in Eq.(3.44) reads

(3.45) |

where now and . For small values of the parameters they can be regarded as coordinates on the group manifold in the vicinity of the unit element. Since acts on via , one has

In general it is a difficult issue to construct the non-perturbative path integration measure. However, in the zero mode sector this can be done. We note that the measure should be invariant under the group multiplications, , and this uniquely requires that should be the Haar measure for . The normalization is fixed by the requirement that for the perturbative result (3.45) is reproduced. This unambiguously gives

(3.46) |

where is computed at and is the Haar measure of the isometry group normalized such that as