### 6.5 Stability of Minkowski spacetime

In this section we apply the validity criterion for semiclassical gravity introduced in Section 3.3 to flat spacetime. The Minkowski metric is a particularly simple and interesting solution of semiclassical gravity. In fact, as we have seen in Section 6.1, when the quantum fields are in the Minkowski vacuum state, one may take the renormalized expectation value of the stress tensor as ; this is equivalent to assuming that the cosmological constant is zero. Then the Minkowski metric is a solution of the semiclassical Einstein equation (8). Thus, we can look for the stability of Minkowski spacetime against quantum matter fields. According to the criteria we have established, we have to look for the behavior of the two-point quantum correlations for the metric perturbations over the Minkowski background, which are given by Equations (16) and (17). As we have emphasized before, these metric fluctuations separate in two parts: the first term on the right-hand side of Equation (17), which corresponds to the intrinsic metric fluctuations, and the second term, which corresponds to the induced metric fluctuations.

#### 6.5.1 Intrinsic metric fluctuations

Let us first consider the intrinsic metric fluctuations,

where are the homogeneous solutions of the Einstein–Langevin equation (15), or equivalently the linearly-perturbed semiclassical equation, and where the statistical average is taken with respect to the Wigner distribution that describes the initial quantum state of the metric perturbations. Since these solutions are described by the linearized semiclassical equation around flat spacetime, we can make use of the results derived in [10, 11, 110, 169]. The solutions for the case of a massless scalar field were first discussed in [169] and an exhaustive description can be found in Appendix A of [110]. It is convenient to decompose the perturbation around Minkowski spacetime into scalar, vectorial and tensorial parts, as
where is a transverse vector and is a transverse and traceless symmetric tensor, i.e., , and . A vector field characterizes the gauge freedom due to infinitesimal diffeomorphisms as . We may use this freedom to choose a gauge; a convenient election is the Lorentz or harmonic gauge defined as
When this gauge is imposed we have the following conditions on the metric perturbations and , which implies . A remaining gauge freedom compatible with the Lorentz gauge is still possible provided the vector field satisfies the condition . One can easily see [203] that the vectorial and scalar part can be eliminated, as well as the contribution of the scalar part , which corresponds to Fourier modes with . Thus, we will assume that we impose the Lorentz gauge with additional gauge transformations, which leave only the tensorial component and the modes of the scalar component with in Fourier space.

Using the metric decomposition (148) we may compute the linearized Einstein tensor . It is found that the vectorial part of the metric perturbation gives no contribution to this tensor, and the scalar and tensorial components give rise, respectively, to scalar and tensorial components: and . Thus, let us now write the Fourier transform of the homogeneous Einstein–Langevin equation (121), which is equivalent to the linearized semiclassical Einstein equation,

Using the previous decomposition of the Einstein tensor this equation can be re-written in terms of its scalar and tensorial parts as

where and are given by Equations (123), and and denote, respectively, the Fourier-transformed scalar and tensorial parts of the linearized Einstein tensor. To simplify the problem and to illustrate, in particular, how the runaway solutions arise, we will consider the case of a massless and conformally coupled field (see [110] for the massless case with arbitrary coupling and [10, 259] for the general massive case). Thus substituting and into the functions and , and using Equation (117), the above equations become

where . Let us consider these two equations separately.

For the scalar component when the only solution is . When the solutions for the scalar component exhibit an oscillatory behavior in spacetime coordinates, which corresponds to a massive scalar field with ; for the solutions correspond to a tachyonic field with . In spacetime coordinates they exhibit an exponential behavior in time, growing or decreasing, for wavelengths larger than and an oscillatory behavior for wavelengths smaller than . On the other hand, the solution is completely trivial since any scalar metric perturbation giving rise to a vanishing linearized Einstein tensor can be eliminated by a gauge transformation.

For the tensorial component, when , where is the Planck length (), the first factor in Equation (154) vanishes for four complex values of of the form and , where is some complex value. This means that, in the corresponding propagator, there are two poles on the upper half-plane of the complex plane and two poles in the lower half-plane. We will consider here the case in which ; a detailed description of the situation for can be found in Appendix A of [110]. The two zeros on the upper half of the complex plane correspond to solutions in spacetime coordinates, which exponentially grow in time, whereas the two on the lower half correspond to solutions exponentially decreasing in time. Strictly speaking, these solutions only exist in spacetime coordinates, since their Fourier transform is not well-defined. They are commonly referred to as runaway solutions and for they grow exponentially in time scales comparable to the Planck time.

Consequently, in addition to the solutions with , there are other solutions that in Fourier space take the form for some particular values of , but all of them exhibit exponential instabilities with characteristic Planckian time scales. In order to deal with those unstable solutions, one possibility is to make use of the order-reduction prescription [295], which we will briefly summarize in Section 6.5.3. Note that the terms in Equations (153) and (154) come from two spacetime derivatives of the Einstein tensor, moreover, the term comes from the nonlocal term of the expectation value of the stress tensor. The order-reduction prescription amounts here to neglecting these higher derivative terms. Thus, neglecting the terms proportional to in Equations (153) and (154), we are left with only the solutions, which satisfy . The result for the metric perturbation in the gauge introduced above can be obtained by solving for the Einstein tensor, which in the Lorentz gauge of Equation (149) reads:

These solutions for simply correspond to free linear gravitational waves propagating in Minkowski spacetime expressed in the transverse and traceless (TT) gauge. When substituting back into Equation (147) and averaging over the initial conditions we simply get the symmetrized quantum correlation function for free gravitons in the TT gauge for the state given by the Wigner distribution. As far as the intrinsic fluctuations are concerned, it seems that the order-reduction prescription is too drastic, at least in the case of Minkowski spacetime, since no effects due to the interaction with the quantum matter fields are left.

A second possibility, proposed by Hawking et al. [161, 162], is to impose boundary conditions, which discard the runaway solutions that grow unbounded in time. These boundary conditions correspond to a special prescription for the integration contour when Fourier transforming back to spacetime coordinates. As we will discuss in more detail in Section 6.5.2, this prescription reduces here to integrating along the real axis in the complex plane. Following that procedure we get, for example, that for a massless conformally-coupled matter field with the intrinsic contribution to the symmetrized quantum correlation function coincides with that of free gravitons plus an extra contribution for the scalar part of the metric perturbations. This extra-massive scalar renders Minkowski spacetime stable, but also plays a crucial role in providing a graceful exit in inflationary models driven by the vacuum polarization of a large number of conformal fields. Such a massive scalar field would not be in conflict with present observations because, for the range of parameters considered, the mass would be far too large to have observational consequences [162].

#### 6.5.2 Induced metric fluctuations

Induced metric fluctuations are described by the second term in Equation (17). They are dependent on the noise kernel that describes the stress-tensor fluctuations of the matter fields,

where here we have written the expression in the large limit, so that , where and is the number of independent free scalar fields. The contribution corresponding to the induced quantum fluctuations is equivalent to the stochastic correlation function obtained by considering just the inhomogeneous part of the solution to the Einstein–Langevin equation. We can make use of the results for the metric correlations obtained in Sections 6.3 and 6.4 for solving the Einstein–Langevin equation. In fact, one should simply take to transform our expressions here to those of Sections 6.3 and 6.4 or, more precisely, one should multiply the noise kernel in these expressions by in order to use those expressions here, as follows from the fact that we now have independent matter fields.

As we have seen in Section 6.4, following [259], the Einstein–Langevin equation can be entirely written in terms of the linearized Einstein tensor. The equation involves second spacetime derivatives of that tensor and, in terms of its Fourier components, is given in Equation (121) as

where we have now used the rescaled coupling . The solution for the linearized Einstein tensor is given in Equation (133) in terms of the retarded propagator defined in Equation (129). Now this propagator, which is written in Equation (130), exhibits two poles in the upper half complex plane and two poles in the lower half-plane, as we have seen analyzing the zeros in Equations (153) and (154) for the massless and conformally coupled case. The retarded propagator in spacetime coordinates is obtained, as usual, by taking the appropriate integration contour in the plane. It is convenient in this case to deform the integration path along the real axis so as to leave the two poles of the upper half-plane below that path. In this way, when closing the contour by an upper half-circle, in order to compute the anti-causal part of the propagator, there will be no contribution. The problem now is that when closing the contour on the lower half-plane, in order to compute the causal part, the contribution of the upper half-plane poles gives an unbounded solution, a runaway instability. If we adopt the Hawking et al. [161, 162] criterion of imposing final boundary conditions, which discard solutions growing unboundedly in time, this implies that we just need to take the integral along the real axis, as was done in Section 6.4.2. But now that the propagator is no longer strictly retarded, there are causality violations in time scales on the order of , which should have no observable consequences. This propagator, however, has a well-defined Fourier transform.

Following the steps after Equation (133), the Fourier transform of the two-point correlation for the linearized Einstein tensor can be written in our case as,

where the noise kernel is given by Equation (125). Note that these correlation functions are invariant under gauge transformations of the metric perturbations because the linearized Einstein tensor is invariant under those transformations.

We may also use the order-reduction prescription, which amounts in this case to neglecting terms in the propagator, which are proportional to , corresponding to two spacetime derivatives of the Einstein tensor. The propagator then becomes a constant, and we have

Finally, we may derive the correlations for the metric perturbations from Equations (158) or (159). In the Lorentz or harmonic gauge the linearized Einstein tensor takes the particularly simple form of Equation (155) in terms of the metric perturbation. One may derive the correlation functions for as it was done in Section 6.4.2 to get
There will be one possible expression for the two-point metric correlation, which corresponds to the Einstein-tensor correlation of Equation (158), and another expression corresponding to Equation (158), when the order-reduction prescription is used. We should note that, contrary to the correlation functions for the Einstein tensor, the two-point metric correlation is not gauge invariant (it is given in the Lorentz gauge). Moreover, when taking the Fourier transform to get the correlations in spacetime coordinates, there is an apparent infrared divergence when in the massless case. This can be seen from the expression for the noise kernel defined in Equation (125). For the massive case no such divergence due to the factor exists, but as one takes the limit it will show up. This infrared divergence, however, is a gauge artifact that has been enforced by the use of the Lorentz gauge. A gauge different from the Lorentz gauge should be used in the massless case; see [203] for a more detailed discussion of this point.

Let us now write the two-point metric correlation function in spacetime coordinates for the massless and conformally coupled fields. In order to avoid runaway solutions we use the prescription that the propagator should have a well-defined Fourier transform by integrating along the real axis in the complex plane. This was, in fact, done in Section 6.4.3 and we may now write Equation (146) as

where the projector is defined in Equation (107). This correlation function for the metric perturbations is in agreement with the real part of the graviton propagator obtained by Tomboulis in [348] using a large expansion with Fermion fields. Note that when the order-reduction prescription is used the terms in the denominator of Equation (161) that are proportional to are neglected. Thus, in contrast to the intrinsic metric fluctuations, there is still a nontrivial contribution to the induced metric fluctuations due to the quantum matter fields in this case.

To estimate the above integral let us follow Section 6.4.3 and consider spacelike separated points and introduce the Planck length . For space separations we have that the two-point correlation (161) goes as and for we have that it goes as Since these metric fluctuations are induced by the matter stress fluctuations we infer that the effect of the matter fields is to suppress metric fluctuations at small scales. On the other hand, at large scales the induced metric fluctuations are small compared to the free graviton propagator, which goes like .

We thus conclude that, once the instabilities giving rise to the unphysical runaway solutions have been discarded, the fluctuations of the metric perturbations around the Minkowski spacetime induced by the interaction with quantum scalar fields are indeed stable (instabilities lead to divergent results when Fourier transforming back to spacetime coordinates). We have found that, indeed, both the intrinsic and the induced contributions to the quantum correlation functions of metric perturbations are stable, and consequently Minkowski spacetime is stable.

#### 6.5.3 Order-reduction prescription and large N

Runaway solutions are a typical feature of equations describing backreaction effects, such as in classical electrodynamics, and are due to higher than two time derivatives in the dynamical equations. Here we will give a qualitative analysis of this problem in semiclassical gravity. In a very schematic way the semiclassical Einstein equations have the form

where, say, stands for the linearized Einstein tensor over the Minkowski background and we have simplified the equation as much as possible. The second term of the equation is due to the vacuum polarization of matter fields and contains four time derivatives of the metric perturbation. Some specific examples of such an equation are, in momentum space, Equations (153) and (154). The order-reduction procedure is based on treating perturbatively the terms involving higher-order derivatives, differentiating the equation under consideration, and substituting back the higher derivative terms in the original equation, keeping only terms up to the required order in the perturbative parameter. In the case of the semiclassical Einstein equation, the perturbative parameter is . If we differentiate Equation (162) with respect to time twice, it is clear that the second-order derivatives of the Einstein tensor are of order . Substituting back into the original equation, we get the following equation up to order : Now there are certainly no runaway solutions but also no effect due to the vacuum polarization of matter fields. Note that the result is not so trivial when there is an inhomogeneous term on the right-hand side of Equation (162), this is what happens with the induced fluctuations predicted by the Einstein–Langevin equation.

Semiclassical gravity is expected to provide reliable results as long as the characteristic length scales under consideration, say , satisfy that  [110]. This can be qualitatively argued by estimating the magnitude of the different contributions to the effective action for the gravitational field, considering the relevant Feynman diagrams and using dimensional arguments. Let us write the effective gravitational action, again in a very schematic way, as

where is the Ricci scalar. The first term is the usual classical Einstein–Hilbert term. The second stands for terms quadratic in the curvature (square of Ricci and Weyl tensors). These terms appear as radiative corrections due to vacuum polarization of matter fields. Here is a dimensionless parameter presumably of order 1 and the terms are higher-order corrections, which appear, for instance, when one considers internal graviton propagators inside matter loops. Let us assume that ; then the different terms in the action are on the order of and . Consequently, when , the term due to matter loops is a small correction to the Einstein–Hilbert term and this term can be treated as a perturbation. The justification for the order-reduction prescription is actually based on this fact. Therefore, significant effects from the vacuum polarization of the matter fields are only expected when their small corrections accumulate in time, as would be the case for an evaporating macroscopic black hole all the way before reaching Planckian scales (see Section 8.3).

However, if we have a large number of matter fields, the estimates for the different terms change in a remarkable way. This is interesting because the large expansion seems, as we have argued in Section 3.3.1, the best justification for semiclassical gravity. In fact, now the vacuum-polarization terms involving loops of matter are of order . For this reason, the contribution of the graviton loops, which is just of order as is any other loop of matter, can be neglected in front of the matter loops; this justifies the semiclassical limit. Similarly, higher-order corrections are of order . Now there is a regime, when , where the Einstein–Hilbert term is comparable to the vacuum polarization of matter fields, , and yet the higher correction terms can be neglected because we still have , provided . This is the kind of situation considered in trace anomaly driven inflationary models [162], such as that originally proposed by Starobinsky [339], see also [355], where exponential inflation is driven by a large number of massless conformal fields. The order-reduction prescription would completely discard the effect from the vacuum polarization of the matter fields even though it is comparable to the Einstein–Hilbert term. In contrast, the procedure proposed by Hawking et al. keeps the contribution from the matter fields. Note that here the actual physical Planck length is considered, not the rescaled one, , which is related to by .

#### 6.5.4 Summary

An analysis of the stability of any solution of semiclassical gravity with respect to small quantum perturbations should include not only the evolution of the expectation value of the metric perturbations around that solution, but also their fluctuations encoded in the quantum correlation functions. Making use of the equivalence (to leading order in , where is the number of matter fields) between the stochastic correlation functions obtained in stochastic semiclassical gravity and the quantum correlation functions for metric perturbations around a solution of semiclassical gravity, the symmetrized two-point quantum correlation function for the metric perturbations can be decomposed into two different parts: the intrinsic metric fluctuations due to the fluctuations of the initial state of the metric perturbations itself, and the fluctuations induced by their interaction with the matter fields. From the linearized perturbations of the semiclassical Einstein equation, information on the intrinsic metric fluctuations can be retrieved. On the other hand, the information on the induced metric fluctuations naturally follows from the solutions of the Einstein–Langevin equation.

We have analyzed the symmetrized two-point quantum correlation function for the metric perturbations around the Minkowski spacetime interacting with scalar fields initially in the Minkowski vacuum state. Once the instabilities that arise in semiclassical gravity, which are commonly regarded as unphysical, have been properly dealt with by using the order-reduction prescription or the procedure proposed by Hawking et al. [161, 162], both the intrinsic and the induced contributions to the quantum correlation function for the metric perturbations are found to be stable [203]. Thus, we conclude that Minkowski spacetime is a valid solution of semiclassical gravity.