2209.09922v1

Probing cosmological singularities with quantum fields:
Open and closed FLRW universes

Abhay Ashtekar1, ∗ and Adrián del Rı́o1, †
1 Institute
for Gravitation and the Cosmos & Physics Department,
The Pennsylvania State University, University Park, PA 16802 U.S.A
(Dated: September 22, 2022)
It was recently pointed out that linear quantum fields φ̂(x) can be meaningfully
arXiv:2209.09922v1 [gr-qc] 20 Sep 2022
propagated across the big bang (and the big crunch) singularities of spatially flat
Friedmann, Lemaı̂tre, Robertson, Walker (FLRW)universes [1]. Recall that φ̂(x), as
well as renormalized observables hφ̂2 (x)iren and hT̂ab (x)iren , are distribution-valued
already in Minkowskian quantum field theories. It was shown that they can be
extended as well-defined distributions even when these space-times are enlarged to
include the big-bang (or the big crunch). We generalize these results to spatially
closed and open FLRW models, showing that this ‘tameness’ of cosmological singu-
larities is not an artifact of the technical simplifications due to spatial flatness. Our
analysis also provides explicit expressions of hφ̂(x) φ̂(x ′ )i, hφ̂2 (x)iren and hT̂ab (x)iren
in closed and open universes for minimally coupled massless scalar fields and discuss
the ambiguities in the definition of hT̂ab (x)iren at the big-bang. While the techni-
cal expressions are more complicated than in the spatially flat case, there is also
an unexpected conceptual simplification: the infrared divergence [2] is now absent
because, in effect, the spatial curvature provides a natural cutoff. Finally, we further
clarify the sense in which quantum field theory can continue to be well defined even
though the extended space-time is not globally hyperbolic because of the singularity,
and suggest directions for further work.
I. INTRODUCTION
Singularity theorems of Penrose, Hawking, Geroch and others are often interpreted to
mean that the big bang (and the big crunch) represent the absolute beginning (and end)
of the universe: since these singularities are space-like, space-time is taken to terminate
there and it is assumed that physics cannot be extended beyond. However, the theorems
only establish geodesic incompleteness. While trajectories of classical test particles do come
to an abrupt end, they are not the appropriate tools to probe the space-time structure
once the curvature enters the Planck regime. It has been long argued that singularities of
general relativity may be quite tame when probed with more realistic tools. For example,
quantum particles were used as probes in certain static, dilatonic black holes in Ref. [3], and
classical test fields were used as probes in superextremal Reissner-Nordström solutions in
Ref. [4]. In these cases, the background space-times are non-dynamical and the singularities
are eternal. The conclusion was that the evolution of these probes remains well-defined
in a precise mathematical sense in spite of the singularity. In this paper we will focus on
cosmological space-times which are dynamical and in which the singularity is physically
∗
[email protected]
†
[email protected]
2
more interesting. Because of the dynamical nature of space-time, there is ‘particle creation’
and quantum test particles are no longer suitable as probes. One has to use quantum fields
which in turn require appropriate mathematical tools to handle the presence of an infinite
number of degrees of freedom. The simplest setting is provided by the spatially flat (or
K = 0) Friedmann, Lemaı̂tre, Robertson, Walker (FLRW) models. Linear, test quantum
fields were recently analyzed on this background [1]. A key point in this analysis was to
note that, already in Minkowski space-time, a quantum field φ̂(x) is an operator-valued
distribution (OVD) on the Fock space, rather than an operator. Indeed, it is because of this
distributional character that one needs a regularization and renormalization procedure to
define their products such as hφ̂2(x)iren . One cannot expect φ̂(x) to do better at the big-
bang singularity! Now, in FLRW models we can extend the space-time across the big-bang
(and/or big-crunch) singularity in an obvious way so that the space-time metric gab is only
continuous (and degenerate) at the singularity. The questions then are:
(i) Can φ̂(x) be extended as a well-defined OVD in the larger space-time?
(ii) Can the associated hφ̂(x) φ̂(x ′ )i be extended as a well-defined bi-distribution? and,
(iii) Can observables hφ̂2 (x)iren and hT̂ab (x)iren be extended as well-defined distributions?
In the K = 0 case, if we consider scale factors of the form a(η) = aβ η β (where η is the
conformal time), we have β = 1 for radiation-filled universes and β = 2 for dust-filled
universes. In the main text of [1] the focus was on these two cases and the situation for
β > 2 was summarized in an Appendix. Somewhat surprisingly, the answers to all three
questions turned out to be in the affirmative. 1
Findings in [1] were facilitated by the technical simplifications associated with the absence
of spatial curvature in the K = 0 case. The question is whether the final results depend
critically on those simplifications. If so, regularity of quantum fields across the big-bang
would be accidental. To gain insight into this issue, in this paper we will consider open
and closed FLRW universes which are technically more complicated due to the presence of
spatial curvature. We will focus on radiation and dust filled universes and show that the
main results of [1] remain unaltered, although the expressions of hφ̂(x) φ̂(x ′ )i, hφ̂2 (x)iren and
hT̂ab (x)iren are now more complicated. (More general equations of state for matter will not
be discussed because the analysis becomes technically even more complicated and findings of
[1] suggest that the main conclusions will not change qualitatively.) Somewhat surprisingly
–at least at first– the presence of spatial curvature leads to a conceptual simplification. In
the K = 0 case, there is an infrared (IR) divergence for dust-filled universes that requires
the introduction of an IR regulator [2]. We will find that this is no longer necessary in closed
and open universes because spatial curvature introduces an effective IR cut-off. We will also
take this opportunity to add some clarifications about the effect of the singularity. Specif-
ically, because the extended space-time that includes the singularity is no longer globally
hyperbolic, some of the familiar results fail to go through. Nonetheless, there is sufficient
structure to enable one to introduce the OVD φ̂(x) and investigate its properties on the
extended space-time that includes the singularity.
The material is organized as follows. In Sec. II we briefly review the theory of linear
quantum fields in FLRW space-times. This discussion will serve to fix the notation and also
enable us to separate the structure that fails to go through when the space-time is extended
1
Another example of direct physical interest is provided by the Schwarzschild space-time in which the
singularity is also space-like and in its vicinity space-time is again dynamical. This has been analyzed
using a formal Schrödinger representation in [5], and at a mathematical level that pays due attention to
the presence of an infinite number of degrees of freedom in [6]. (Bianchi I model has also been investigated
using the formal Schrödinger representation in [7]). The conclusion is again that, when probed with linear,
test quantum fields, the singularity is tame.
3
to include the big-bang, from that which does go through. Relative to the K = 0 case, the
main technical complication is that, whereas the eigenfunctions of the spatial Laplacian are
~
simply the plane waves eik·~x in the K = 0 case, they are much more involved in presence
of spatial curvature. In Sec. III we extend the space-time across the big-bang/big-crunch
of the radiation-filled universes, and show that sufficient structure continues to be available
to establish that the OVD φ̂(x) is well-defined on the extension. As in the K = 0 case,
the time-dependence of mode functions that define the explicit form of φ̂(x) in the standard
Fock representation is quite simple because the space-time scalar curvature R vanishes also
for closed and open universes. We will find that the bi-distribution hφ̂(x) φ̂(x ′ )i continues
to have the standard Hadamard structure away from the big-bang/big-crunch surface, and
if the points x, x′ have a ‘purely time-like’ separation, also across this surface. Using this
bi-distribution we then calculate the observables hφ̂2 (x)iren and hT̂ab (x)iren using Hadamard
renormalization. In Sec. IV we turn to dust-dominated universes. Now the space-time
scalar curvature R no longer vanishes, whence the time dependence of mode functions is
more complicated. Still, one can carry out all the required calculations and again obtain
the explicit expressions of hφ̂(x) φ̂(x ′ )i, hφ̂2(x)iren and hT̂ab (x)iren as in Sec. III. It is manifest
that the IR cut-off that is necessary in the K = 0, dust-filled FLRW universe is no longer
necessary in closed and open universes. For both radiation and dust filled universes, the
final expressions of hφ̂2 (x)iren and hT̂ab (x)iren make it manifest that these observables are
well defined distributions on the full, extended space-time in spite of the singularity. This
can be quite surprising to researchers who focus on cosmological perturbations and relation
between theory and observations. However, we recently learned that this fact was expected
by some in the algebraic quantum field theory and microlocal analysis community, based
on results of [8, 9]. In sections III and IV we also provide explicit expressions based on the
Hadamard renormalization scheme that would be of direct interest to semi-classical gravity.
Our expressions of hT̂ab (x)iren satisfy the standard properties [10], including the non-trivial
conservation requirement. But still the answer is as usual scheme dependent and there is a
further 4-parameter ambiguity in the definitions of these distributions at the big-bang/big-
crunch. Since our primary goal is to show existence rather than uniqueness, we choose
a commonly used prescription motivated by certain mathematical properties to fix them.
Whether they can be fixed using physical criteria is an interesting open issue.
The presence of the singularity in the extended space-time does give rise to certain sub-
tleties. In particular, since the extended space-time is not globally hyperbolic, the standard
notion of retarded and advanced Green’s functions are not available, and the standard no-
tion of Hadamard regularity cannot be applied if the points lie on the big-bang/big-crunch
surface. This is not surprising. Rather, what is surprising is that these issues do not pose
unsurmountable obstacles. At appropriate points in sections III and IV, we explain how
apparent difficulties are circumvented. In Sec. V we summarize the main results, provide a
general perspective on the quantum tameness of space-like singularities of classical general
relativity, and suggest directions for further work.
One of the primary goal of this paper is to try to bridge the gap between the community
that investigates quantum field theory using mathematically rigorous techniques, generally
emphasizing the algebraic approach, and the community that is primarily interested in ap-
plications of the theory to the early universe, emphasizing explicit expressions of φ̂(x) and
hφ̂(x) φ̂(x ′ )i using mode expansions and of hφ̂2 (x)iren and hT̂ab (x)iren using specific renormal-
ization schemes, in the hope that combined ideas from both communities will shed further
light on the nature of the big bang when probed using quantum tools. To make the material
4
accessible to both communities, we have tried to strike a balance between mathematical

rigor and physical and conceptual issues.
Our conventions are as follows. We work in the geometrized units G = c = 1. The metric
signature has signature (−, +, +, +), the Riemann tensor is defined by 2∇[a ∇b] vc =: Rabc d vd
for any 1-form vd ; the Ricci tensor is defined as Rab := Racb c ; and the scalar curvature as
R := g ab Rab . In section II we work with the standard FLRW models that exclude the
big-bang and big-crunch singularities, and all classical fields are assumed to be smooth. In
sections III and IV, the space-time manifold is enlarged to include the big-bang/big-crunch
by extending the range of conformal time η in the obvious manner. Then, the inverse scale
factor, curvature tensor and classical solutions to the Klein-Gordon equation are treated as
distributions which are smooth fields away from the singularity.
II. LINEAR QUANTUM FIELDS IN FLRW SPACE-TIMES
In this section we restrict ourselves to the FLRW space-times that are smooth, i.e.,
exclude the big-bang and big-crunch singularities. Discussion is divided into three parts. In
the first, we recall structures underlying quantum field theory in curved space-times [10–17]
that are needed for our analysis, and in the second, the basics of mode decomposition of
scalar fields in spatially closed and open FLRW models [18]. This summary will help make
the paper self-contained. To keep the summary reasonably short, and to reach the audience
outside the ‘communications in mathematical physics community’, we will have to gloss
over issues that are not essential for our main results, such as those related to topology
on infinite dimensional spaces. Therefore, we will not use notions from C ⋆ -algebras, nor
discuss functional analytic issues such as the distinction between weakly non-degenerate
and strongly non-degenerate symplectic structures. In the third part we use the results from
the first two parts to arrive at a strategy that will facilitate the extension of quantum fields
across the big-bang/big-crunch.
A. Operator algebras and their quasi-free representations
Consider a scalar field φ(x) satisfying the Klein-Gordon equation ( − m2 − ξR) φ(x) = 0
on a globally hyperbolic space-time (M, gab ) where m, ξ are constants and R denotes the
scalar curvature of the space-time metric gab . Since the Klein-Gordon operator is nor-
mally hyperbolic, we have well-defined retarded, advanced and the causal/commutator
Green’s functions, GRet (x, x′ ), GAd (x, x′ ), ∆(x, x′ ), respectively, with ∆(x, x′ ) = (GAd −
GRet )(x, x′ ) [19]. Then, given any C ∞ function f (x) with compact support on M, i.e., an
element of C0∞ (M),
Z
F (x) := d4 V ′ ∆(x, x ′ )f (x ′ ) (2.1)
M
is a smooth solution to the Klein-Gordon equation, whose initial data is of compact support
on any Cauchy slice. (Here d4 V ′ is the volume element of the metric gab on M.) In fact,
any smooth solution F of the Klein-Gordon equation that has spatially compact support
on every Cauchy slice is of this form for some f ∈ C0∞ (M). Note, however, that the map
f (x) → F (x) has a large kernel: If f (x) = ( − m2 − ξR)h(x) for any h(x) ∈ C0∞ (M),
then the corresponding solution is identically zero. Now, the space of these solutions F (x)
5
is naturally equipped with a symplectic structure Ω that defines Poisson brackets in the
classical theory and leads to the canonical commutation relations in the quantum theory:
Z
Ω(F1 (x), F2 (x)) := d3 V F1 (x)(na ∇a F2 (x) − (na ∇a F1 (x))F2 (x))

(2.2)
Σ
where Σ is any Cauchy surface in M and na the unit normal to Σ. Satisfaction of the
Klein-Gordon equation guarantees that the integral on the right side is independent of the
choice of Σ. This fact will play an important role in our extension of the theory across
the big-bang/big-crunch. We can re-express the symplectic structure using any of the test
functions f1 (x) and f2 (x) that give rise to F1 (x) and F2 (x) via (2.1):
Z Z
4
Ω(F1 (x), F2 (x)) = d V d4 V ′ ∆(x, x ′ ) f1 (x) f2 (x ′ ) . (2.3)
M M
Now, in the algebraic approach to quantum field theory in curved space-times, one gen-
erally begins by introducing an (abstractly defined) OVD φ̂(x) that satisfies:
(i) ⋆-relations: φ̂(f ) = φ̂⋆ (f ), where φ̂(f ) = M d4 V φ̂(x)f (x);
R
(ii) The Klein Gordon equation in a distributional sense:
Z
2
φ̂(( − m − ξ R)f ) := d4 V φ̂(x) ( − m2 − ξ R)f (x) = 0 ; (2.4)
M
(iii) The commutation relations:

Z Z
[φ̂(f1 ), φ̂(f2 )] = i~ 4
d V f1 (x) ˆ
d4 V ′ ∆(x, x ′ ) f2 (x ′ ) I, (2.5)
M M
for any f, f1 , f2 in C0∞ (M). One then constructs the free ⋆-algebra A(f ) generated by the
operators φ̂(f ) subject to these three relations. Note that map f → φ̂(f ) has, again, a
large kernel because of (2.4). Therefore, it is often convenient to pass from f (x) ∈ C0∞ (M)
to F (x) and associate field operators Φ̂(F ) with suitably regular classical solutions F (x),
defined formally as follows:
Z
Φ̂(F ) = Ω(φ̂(x), F (x)) ≡ d3 V φ̂(x)(na ∇a F (x) − (na ∇a φ̂(x))F (x)) .

(2.6)
Σ
We can now construct a ⋆-algebra A(F ) generated by Φ̂(F ). Since F (x) is a smooth solution
to the KG equation, we no longer need the condition (ii) above: Φ̂(F ) are subject only to
(i′ ) ⋆-relations: Φ̂(F ) = Φ̂⋆ (F ); and,
(iii′ ) Commutation relations
[Φ̂(F1 ), Φ̂(F2 )] = i~ Ω(F1 , F2 ) Iˆ . (2.7)
Then, in place of A(f ) , we can use the ⋆-algebra A(F ) generated by the operators Φ̂(F )
(defined, again, abstractly), subject to (i′ ) and (iii′ ) [11]. The map φ̂(f ) → Φ̂(F ) –where f
determines F via (2.1)– is an isomorphism from the ⋆-algebra A(f ) to the ⋆-algebra A(F ) .
We will find that A(F ) is particularly useful in extending the quantum field theory beyond
the big-bang because, e.g., this algebra is not tied to the use of Green’s functions, which
refer to global hyperbolicity.
6
Recall that the Gel’fand, Naimark, Segal (GNS) construction [16, 17] provides a direct
and elegant avenue to obtain representations of ⋆-algebras by operators on a Hilbert space.
Given a normalized, positive linear function (PLF) on a ⋆-algebra –in physical terms, an
expectation value function E– the construction provides an explicit, step by step procedure
to build a Hilbert space and represent elements of the algebra by concrete operators on it
such that all algebraic relations are preserved. To specify PLFs E, it is much more convenient
to work with the Weyl operators Ŵ (F ) := exp ~i Φ̂(F ) because the vector space of their linear
combinations is closed under the product:
i
Ŵ (F1 )Ŵ (F2 ) = e− 2~ Ω(F1 ,F2 ) Ŵ (F1 + F2 ) . (2.8)
The Weyl algebra W, generated by Ŵ (F ) is thus spanned just by linear combinations of

Ŵ (F ).2 Because of this property (2.8), the task of specifying a PLF E on the Weyl algebra
W simplifies considerably, since one only has to specify E(W (F )). (By contrast, on A(F ) ,
for example, one has to specify E(Φ̂(F1 ) . . . Φ̂(Fn )) for all n.)
Of particular interest are the quasi-free (or, Gaussian) states that lead to Fock represen-
tations of these ⋆-algebras, in which Φ̂(F ) (and φ̂(f )) are represented as (unbounded) sums
of concretely defined creation and annihilation operators, and Ŵ (F ) by the corresponding
unitary (and therefore bounded) operators. Let us denote by ΓCov the space of suitably
regular solutions F (x) of the Klein-Gordon equation. Then, choice of a quasi-free state on
W is in 1-1 correspondence with the choice of a positive definite metric g on ΓCov which is
compatible with the symplectic structure Ω on ΓCov . Given such a g,
1
E(Ŵ (F )) := e− 4~ g(F,F ) (2.9)
naturally extends to a positive-linear function on all of W. In the resulting GNS repre-
sentation, g is (essentially) the ‘covariance matrix’ of the vacuum state. The compatibility
condition between g and Ω is that g is of the form g(F1 , F2 ) := Ω(F1 , JF2 ), where J is a
complex structure, i.e., a linear operator on ΓCov satisfying J 2 = −1. The triplet (Ω, g, J)
endows ΓCov with a Kähler structure for which the Hermitian inner product is given by
1
hF1 , F2 i = g(F1 , F2 ) + i Ω(F1 , F2 ) (2.10)
2~
1
(so that the PLF has the familiar form E(W (F )) := e− 2 hF | F i ). The Cauchy completion of
ΓCov with respect to this inner product is the 1-particle Hilbert space H in the quasi-free
representation defined by (2.9). The full Hilbert space determined by the GNS construction
is the symmetric Fock space F based on H. The representation map π –determined by the
choice of the metric g or the complex structure J– sends the abstractly defined field operator
Φ̂(F ) to the concrete operator on F ,
π(Φ̂(F )) = ~[Â(F ) + Â† (F )] , (2.11)

where the concrete creation and annihilation operators A† (F ) and A(F ) satisfy the usual
commutation relations:
[Â(F1 ), Â† (F2 )] = hF1 , F2 i Iˆ . (2.12)
2
W can be endowed with the structure of a C ⋆ algebra in a natural manner. It is more appropriate to use
it in the GNS construction, in particular in the discussion of self-adjointness of field operators Φ̂(F ).
7
For simplicity, from now on we will drop the explicit representation map π; the context will
make it clear whether we are referring to abstract operators of their concrete representations
on F .
We will see in sections III and IV that one can naturally extend the familiar phase
spaces ΓCov associated with the post big-bang FLRW space-times –together with the Kähler
structure (J, g, h . | . i) they carry– across the big-bang. This will provide an extension of
standard quantum field theories to larger space-times in spite of the singularity.
B. Klein-Gordon fields in FLRW cosmologies
We will now turn to FLRW cosmologies and summarize certain structures underlying
classical Klein Gordon fields on these space-times that will be used in Sec. III and IV to
construct ⋆-algebras and quasi-free representations.
For models of interest to this paper, the space-time metric gab is given by
gab dxa dxb =: a2 (η)g̊ab dxa dxb ≡ a2 (η) − dη 2 + hij dxi dxj ,

(2.13)
where η is the conformal time and the spatial metrics hij are of constant curvature K. Since
K is the scalar curvature of the 3-manifold, it has dimensions of 1/(length)2 . Following the
usual conventions, we will set
√ √ √ √
K = | K| if K > 0 and K = i| K| if K < 0 (2.14)
√
and, of course, K = 0 if K = 0. Then, hij has the explicit form
√
i j 2 2 2 2 2 sin Kχ
hij dx dx = dχ + Σ(K) (χ) (dθ + sin θdϕ ) where Σ(K) (χ) = √ (2.15)
K
The domain of the radial coordinate χ is (0, ∞) for both K = 0, K < 0, and χ ∈ (0, π) for
K > 0, while (θ, φ) are the usual coordinates on the 2-sphere. The scalar curvature R of gab
′′
is given by R(η) = a26(η) aa +K) where the prime denotes derivative w.r.t. η. Note that that
the 4-metric g̊ab defined in (2.13) is ultra-static –it admits ∂/∂η as a hypersurface-orthogonal,
time-like Killing vector with constant norm– but it is not flat unless K = 0.
Because of the form of the metric gab , without loss of generality we can separate the
spatial and time dependence of solutions to the Klein-Gordon equation ( − m2 − ξR)φ = 0
and consider solutions of the form
√
~ (K)
φ~k (η, ~x) = ψk (η) Y~k (~x) (2.16)
a(η)
where ~x ≡ (χ, θ, φ) and ~k labels the eigenfunctions of the spatial Laplacian ∆(K) defined by
hij . As in the spatially flat case, the solution diverges when the scale factor a(η) vanishes.
With this ansatz, the Klein-Gordon equation separates into two parts, the first dictating the
spatial dependence and the second governing the time evolution:
(K) (K)
△(K) Y~k (~x) + (k 2 − K) Y~k (~x) = 0 , and (2.17)
1
ψk′′ + k 2 + m2 a2 + ξ − a2 R ψk = 0 , (2.18)
6
8
where the prime denotes derivative w.r.t. η in (2.18), and k is the separation constant.
Eq. (2.17) is just the eigenvalue equation for the spatial Laplacian ∆(K) . For K = 0, the
~
solutions are just plane waves eik·~x . But for non-zero K, the solutions are more complicated.
Nonetheless, one can analyze their structure in detail because the spatial metric hij is of
constant curvature both for K > 0 and K < 0 (see Appendix B). However, the time
evolution equation (2.18) does not share this ‘universality’: the form a(η) of the scale factor
varies from one model to another depending on matter content, and the equation also involves
the mass m and the conformal coupling constant ξ. Therefore, in the detailed analysis of
sections III and IV we will set m = 0 and ξ = 0 –thus focusing on the massless, minimally-
coupled Klein-Gordon equation– and restrict ourselves to scale factors a(η) corresponding to
radiation and dust-filled universes. In the remaining part of this sub-section, we will focus
on the spatial equation (2.17).
Main properties of eigenvalues and eigenfunctions of ∆(K) can be summarized as follows.
First, because the 3-metric hij is spherically symmetric, we can separate the eigenfunctions
(K)
Y~k (~x) into a radial and a spherical part:
(K) (K)
Y~k (~x) = Πkℓ (χ) Yℓm (θ, φ) , (2.19)
where Yℓm (θ, φ) are the usual spherical harmonics, with ℓ = 0, 1, 2, . . . , and m = −ℓ, −ℓ +
1, . . . , ℓ. Thus, the label ~k of the eigenfunctions of the Laplacian ∆(K) stands for the triplet
(K)
k, ℓ, m. The structure of the radial modes Πkℓ (χ) is rather complicated and is discussed in
detail in the Appendix B. Here we will simply note the properties that are important to our
analysis. The general solution to the radial equation is given by
ℓ+1 ℓ+1
√ √

(K) ℓ d ℓ d
Πkℓ (χ) = Akℓ sin ( Kχ) √ cos(kχ) + Bkℓ sin ( Kχ) √ sin(kχ)
d cos( Kχ) d cos( Kχ)
(2.20)
where Akℓ , Bkℓ are arbitrary coefficients. In the K → 0 limit, the two linearly independent
solutions reduce to the usual spherical Bessel functions jℓ (kχ), and the Neumann functions
yℓ (kχ), respectively (see, for instance, Eqs. (10.1.25) and (10.1.26) of [20]). As one would
expect from this limit, regularity conditions at the origin χ = 0 require that we set Bkℓ = 0
for all k, ℓ and K. The resulting functions are even in k, so without loss of generality we will
restrict ourselves to k > 0 (note that solutions vanish for k = 0). Furthermore, for K > 0,
a detailed examination shows that regularity on the entire spatial manifold S3 , including
points χ = π, imposes the √ following additional conditions:
(i) k is quantized: k = √K n where n is a positive integer, and
(ii) For any given k ≥ √ K, the quantum number ℓ is bounded above: it can only take
values: ℓ = 0, 1, ..., (k/ K) − 1 .
For the case when the spatial manifold is a hyperboloid H, we have K < 0 and there is no
restriction either on k –which takes values in (0, ∞)– or, on permissible values of ℓ.
Finally, let us choose the coefficients Akl so as to normalize the eigenfunctions appropri-
ately. We will set
√ ℓ+1
√

(K) K ℓ d
Πkℓ (χ) = q sin ( Kχ) √ cos(kχ) (2.21)
π ℓ
Π [k 2 /K − n2 ] d cos( Kχ)
2 n=0
9
Note that this expression is well-defined for k = 0 for any ℓ. This choice ensures (1) the
orthonormality condition
Z
(K) (K)
d3 V̊ Y~k (~x) Ȳ~k′ (~x) = δ(k − k ′ ) δℓ,ℓ′ δm,m′ (2.22)
Σ
for K ≤ 0, where d3 V̊ is the volume element defined by the spatial metric hij ; (2) the
orthonormality condition
1
Z
(K) (K)
d3 V̊ Y~k (~x) Ȳ~k′ (~x) = √ δk,k′ δℓ,ℓ′ δm,m′ (2.23)
Σ K
for K > 0; and, (3) the completeness relation

X (K) (K)
X (K) (K) k sin[k(χ − χ′ )]
Y~k (~x) Ȳ~k (~x′ ) = Πkℓ (χ) Π̄kℓ (χ′ ) |Yℓm (θ, φ)|2 = . (2.24)
ℓm ℓm
2π 2 ΣK (χ − χ′ )
C. Quantization Strategy
Recall from Sec. II A that one can construct the algebra A(F ) of operators and its quasi-
free representations if one has the following ingredients: (i) A space ΓCov of suitably regular
solutions F (x) to the field equation; and, (ii) a complex structure J thereon that is com-
patible with the symplectic structure Ω so that (2.10) is an Hermitian inner product on
ΓCov . This formulation was shown in [1] to be well-suited for extending the theory across
the big-bang/big-crunch in the K = 0 case. In this section we will explain why the same
strategy can be successfully used in closed and open FLRW models.
Let us begin with FLRW space-times that are smooth and exclude singularities. Then,
the first step in our construction would be to pick a complete set of candidate ‘positive
frequency solutions’. A standard practice accomplishes this by specifying a basis ek (η) in
the space of solutions to the time evolution equation (2.18) with following two properties:
(i) For each k, the ek (η), together with their complex conjugates, span the solution space;
and,
(ii)They are normalized such that
ek (η) ∂η ēk (η) − ēk (η) ∂η ek (η) = i . (2.25)
Then the required ΓCov can be constructed by taking suitable linear combinations of
(K)
(ek (η)/a(η)) Y~k (~x) (see Eq. (2.16)). Coefficients in these linear combinations –denoted
by z(~k) below– have to be specified with due care to ensure that the operators Φ̂(F ) are
well-defined on the resulting Fock representation, and continue to be well-defined even when
the FLRW metric is extended across the big-bang. We will choose them as follows. Con-
sider C ∞ functions z(~x) of compact support on the spatial manifolds of constant curvature
(namely, the η = const surfaces) and denote by z(~k) their expansion coefficients in the
normalized basis Y~k (~x). In the K = 0 case, these z(~k) become the Fourier transforms of
(K)
C0∞ functions z(~x) on the Euclidean space. As in that case, the functions z(~k) fall-off in k
faster than any polynomial as k → ∞ ensuring convergence of various k integrals (or sums).
10
Let us now consider the K < 0 case. Then, our ΓCov will consist of solutions F to the
Klein-Gordon equation of the form
∞ ∞ X
ℓ
1
Z h i
(K) (K)
z(~k) ek (η) Y~k (~x) + z̄(~k) ēk (η) Ȳ~k (~x)
X
F (η, ~x) = dk (2.26)
a(η) 0 ℓ=0 m=−ℓ
+ −
=: F (η, ~x) + F (η, ~x) (2.27)
(see Eq. (2.16)). F + (η, ~x) –the first term on the right side of (2.26)– is to be thought
of as the ‘positive frequency part’ of F (η, ~x), and its complex conjugate, the ‘negative fre-
(K)
quency part’. The normalization conditions on ek (η) and Y~k (~x) then ensure that the linear
operator J defined on ΓCov by
(JF )(η, ~x) := i F + (η, ~x) − i F − (η, ~x) (2.28)
is a complex structure on ΓCov that is compatible with the symplectic structure (2.2) thereon.
The resulting positive definite Kähler metric g they define is given simply by
Z ∞ ∞ X
ℓ
|z(~k)|2 .
X
g(F, F ) := 2 dk (2.29)
0 ℓ=0 m=−ℓ
As noted in Sec. II A, g naturally provides a PLF on the Weyl algebra W and carries the
interpretation of the ‘covariance matrix’ of the vacuum in the resulting Fock representation.
On this Fock space, the OVD φ̂(x) admits an expansion that mimics (2.26):
∞ ∞ X
ℓ
1
Z h i
(K) (K)
ek (η) Y~k (~x) Â(~k) + ēk (η) Ȳ~k (~x) Â† (~k)
X
φ̂(η, ~x) = dk (2.30)
a(η) 0 ℓ=0 m=−ℓ
where the creation and annihilation operators satisfy [Â(~k), Â† (~k ′ )] = δ(k − k ′ )δℓ,ℓ′ δm,m′ .
Now a key point is that the time dependent part ea(η) k (η)
that features in the expansion
(2.26) typically diverges at the big bang, η = 0, first because the scale factor vanishes there,
and second because in dust-filled FLRW universes ek (η) also diverges there. Therefore, the
modes –and hence F (η, ~x)– are ill-defined as functions at η = 0. However, as we will see in
sections III and IV, the divergence is only polynomial, i.e., of the form η −n . Now, we know
from the standard distribution theory [21–24] that although η −n are singular as functions
at η = 0, they are well-defined as tempered distributions η −n on the entire real line R that,
furthermore, satisfy the familiar rules of calculus:
d −n
η = −n η −n−1 and, if n > 1 then η η −n = η−n+1 . (2.31)
dη
(For a summary, see Appendix A of [1].) Therefore if we extend the FLRW space-time across
the big-bang by allowing η to take values on the entire real line (−∞, ∞), then ([ek (η)/a(η)]
and) F (η, ~x) are well-defined as distributions satisfying the Klein Gordon equation on the
extended space-time.
But, since the fields F (x) diverge as functions at η = 0, does the symplectic product
Ω(F1 , F2 ) not diverge there? It does not simply because it is conserved. To obtain its explicit
expression let us substitute the expansion (2.26) in the expression (2.2) of the symplectic
11
structure and evaluate it on a Cauchy surface Σ given by η = η0 6= 0. Then, using the

orthonormality relations (2.22) and (2.25) one finds
Z ∞ ∞ X
ℓ
z1 (~k)z̄2 (~k) − z2 (~k)z̄1 (~k) .
X
Ω(F1 , F2 ) = i dk (2.32)
0 ℓ=0 m=−ℓ
Since this expression is independent of the choice of η0 we can extend it by continuity

also to η = 0 surface: the divergence of fields F1 , F2 as we approach the big-bang is exactly
compensated by the shrinking of the volume element [1]. As a consequence, the commutation
relations (2.7) remain well-defined for any pair of solutions F1 , F2 in ΓCov even when the
space-time is extended beyond the big-bang. Therefore, we can construct the ⋆-algebra
A(F ) generated by the operators Φ̂(F ) also on the extended space-time. We can also define
a complex structure J as in Eq. (2.28), arrive at a Fock representation of A(F ) . In the
quantum theory, each of these solutions F (η, ~x) defines a 1-particle state. Again, while
the F (η, ~x) diverge as ‘wave functions’, they are well-defined as elements of the 1-particle
Hilbert space H because their norm remains (2.29) finite. (This is somewhat analogous
to the fact that in quantum mechanics on R3 , the wave function ψ(~r) = e−r /r diverges at
r = 0 but provides a well-defined element of the Hilbert space L2 (R3 ).) Since φ̂(x) of (2.30)
is constructed from the mode functions, as in the K = 0 case discussed in [1], we will find
that it continues to be a well-defined OVD across the big-bang.
Finally, for K > 0 the situation is completely analogous. Minor technical differences arise
because k is now quantized, and the range of ℓ is restricted: (2.30) is now replaced by:
√
√ ∞ (k/ K) −1 ℓ
K X X X h
(K) ~ (K) † ~
i
φ̂(η, ~x) = ek (η) Y~k (~x) A(k) + ēk (η) Ȳ~k (~x) A (k) , (2.33)
a(η) √ ℓ=0 m=−ℓ
k= K
and the creation and annihilation operators satisfy now [Â(~k), Â† (~k ′ )] = √1
K
δk,k′ δℓ,ℓ′ δm,m′ .
Remark: A succinct discussion of the definition and properties of distributions η−n can
be found in Appendix A of [1]. As noted there, the action of η −1 on a test function f is
given by the Cauchy principal value: R∞
η−1 : f (η) → limǫ→0+ R\[−ǫ,ǫ] dη η −1 f (η) = 0 dη η −1 f (η) − f (−η)
R
which has the intuitively expected property of vanishing for test functions f (η) that are even
in η. Distributions η −n are defined by a natural generalization of this procedure and they
satisfy (2.31). Since our primary goal is to show that there exists a consistent extension of
QFT across the big-bang, it suffices to make one concrete choice. For concreteness, we will
make this choice.
However, as pointed out in the Appendix A of [1], if one wishes to regard the distribution
η −n
as the algebraic inverse of the function η n , thenPthere is an n-parameter family of
n−1
ambiguities in the definition [21, 23, 24]: η −n = η −n + i=0 ci δ i (η), where ci are constants
and δ i denotes the ith derivative of the Dirac delta distribution. In any case, the Hilbert
space structure of 1-particle states defined by F is insensitive to this ambiguity (because
the volume element vanishes sufficiently rapidly at η = 0). Similarly hφ̂2 (x)iren is also free
of this ambiguity. On the other hand the ambiguity persists in the definition of hT̂ab (x)iren
because, as we shall see in Sec. III and IV, the expression involves 1/a8 (η) for the radiation
filled universe and 1/a6 (η) for the dust-filled case, while the volume element goes to zero
12
only as a4 (η). Use of η−n corresponds to fixing this ambiguity by setting ci = 0. From a
mathematical consideration, this choice results in intuitively expected properties; for n = 1,
for example, it is only when c0 = 0 that the distribution sends all even test functions f (η)
to zero. But it would be much more satisfactory to remove it using physical requirements.
This issue is open.
III. RADIATION-FILLED, CLOSED AND OPEN FLRW UNIVERSES
This section is divided in three parts. In the first, we show that there is a quasi-free
representation in which the OVD φ̂(x) of Eq. (2.30) remains well-defined even when the
FLRW space-time is extended across the bing-bang. In the second we focus on the K < 0
radiation-filled universes and calculate the observables hφ̂2 (x)iren and hT̂ab (x)iren using the
Hadamard renormalization procedure (summarized in Appendix A), and in the third we
carry out these calculations for K > 0 universes. As expected these results reduce to those
in the K = 0 case reported in [1].
A. Extension across the big-bang
We will now apply this general strategy to radiation-filled universes. It is convenient to

introduce a variable L with dimensions of length, given by
K = 1/L2 if K > 0 and K = −1/L2 if K < 0 ; (3.1)
the limit K → 0 corresponds to L → ∞ in both cases. For radiation-filled universes, the

scale factor is given by
a(η) = (a1 L) sin(η/L), for K > 0 and a(η) = (a1 L) sinh(η/L), for K < 0 (3.2)
where a1 is a constant (with dimensions (length)−1 ). The scale factor vanishes at η = 0 and
the curvature diverges there. This is the big-bang singularity (and can also be the big-crunch
in the K > 0 case). In the standard analysis one restricts the range of η to lie in (0, πL) for
closed universes and in (0, ∞) for open universes, so that η = 0 corresponds to the big-bang
in both cases. (In the limit K → 0 we have a(η) = a1 η, as in the radiation-filled case
discussed in [1].)
Since the stress-energy tensor is trace-free, the space-time scalar curvature R vanishes by
Einstein’s equations. Hence (2.18) becomes ψk′′ + k 2 ψk = 0, which can be solved trivially to
obtain the general solution
ψk (η) = Ck e−ikη + Dk eikη (3.3)
for both closed and open universes, where Ck and Dk are arbitrary constants. This form
suggests that ek (η) := √12k e−ikη would serve as the positive frequency basis satisfying the
normalization condition (2.25). This expectation is correct: One can systematically arrive
at this choice using the following considerations.
Recall first that the FLRW metric (2.13) is of the form gab = a2 (η)g̊ab , where g̊ab is ultra-
static, g̊ab dxa dxb = −dη 2 + hij dxi dxj , and the spatial metric hij is of constant curvature.
This ultra-static metric is obviously well-defined if the underlying manifold M is extended
across the big-bang, so that η ∈ (−∞, ∞) in the K < 0 case and η ∈ (−πL, πL) in the
13
K > 0 case. Let us denote the extended space-time by (M̊ ,g̊ab ). The next step is to note
the relation between Klein Gordon equations w.r.t. the physical metric gab and w.r.t. the
ultra-static metric g̊ab . For any two conformally related metrics gab = a2 (η)g̊ab , we have the
identity
1 ˚ − 1 R̊)φ̊
( − R)φ(x) = a−3 (η) ( where φ̊ = a(η)φ . (3.4)
6 6
Since R = 0 for gab , and R̊ = 6K, we have
˚ − K)φ̊ .
φ(x) = a−3 (η) ( (3.5)
Thus, we are led to solve ( ˚ − K)φ̊ = 0. Using separation of variables, this equation
reduces to (2.17) and (2.18), (with a(η) = 1, ξ = 0, m2 = 61 R = K). Therefore we are
led to construct the phase space Γ̊cov using solutions of the form (2.26) (with a(η) = 1).
Thanks to the ultra-static property of g̊ab , Γ̊cov admits a unique complex structure J. ˚ The
corresponding quasi-free vacuum is a Hadamard state. In the resulting Fock representation,
φ̂◦ (x) is a well-defined OVD on full (M̊,g̊ab ), given by (2.30) (again with a(η) = 1).
Let us return to the the physical FLRW space-time (M, gab ) and extend the metric gab to
M̊ simply by letting η assume negative values in a(η) and, for notational simplicity, continue
to denote the extended metric again by gab . (Since a(η) is C ∞ on entire M̊ , so is the tensor
field gab .) It is immediate from the expression (3.2) of a(η) that η/a(η) is also a smooth
function on M̊. Therefore, it follows that, for every F̊ (x) ∈ Γ̊cov ,
1 1 η
F (x) := F̊ (x) ≡ F̊ (x) (3.6)
a(η) η a(η)
is a well-defined distribution on the extended FLRW space-time (M̊, gab ), satisfying F (x) =
0. Our phase space ΓCov will consist of these solutions (with F̊ (x) ∈ Γ̊cov ). The complex
structure on Γ̊cov induces a natural complex structure J on ΓCov , with a positive frequency
basis ek (η) := √12k e−ikη , as anticipated. For reasons discussed in in section II C, although the
solutions F (x) diverge at η = 0 as functions, the norm (2.29) of the state each F (x) defines
in the 1-particle Hilbert space is finite (and insensitive to the 1-parameter ambiguity in the
definition of the distribution η −1 ). Finally, φ̂(x) = φ̂◦ (x)/a(η) provides us a well-defined
OVD on the resulting Fock space satisfying the Klein-Gordon equation in a distributional
sense on the extended FLRW space-time (M̊, gab ).
Remarks:
1. While we are considering massless scalar fields and the scalar curvature of the radiation-
filled FLRW vanishes, as we see from (4.2), F (x) 6= F̊ (x); there is conformal covariance but
not invariance. Had we been considering the Maxwell field, we would have had conformal
invariance –Fab (x) = F̊ab (x)– and then Fock vacuum in the ultra-static space-time would be
the Fock-vacuum in the FLRW space-time. The presence of 1/a(η) in the relation F (x) =
(1/a(η)) F̊ (x) makes the quantum theory of φ̂(x) different from that of φ̂◦ (x).
2. In particular, since (M̊ ,g̊ab ) is a smooth, globally hyperbolic space-time, the retarded,
advanced and the commutator Green’s function ∆(x, ˚ x′ ) := (G̊Ad − G̊Ret )(x, x′ ) are all well-
defined. On the other hand (M̊ , gab ) is not globally hyperbolic because of the singularity
at η = 0. Therefore, a priori, the retarded and advanced Green’s functions are not well-
defined unless both the points x, x′ lie on the same side of the singularity. Nonetheless, since
14
φ̂(x) = φ̂◦ (x)/a(η), the commutator is given by

Z Z
[φ̂(f ), φ̂(g)] =: i~ 4
d V d4 V ′ f (x)g(x′ ) ∆(x, x′ ) Iˆ
ZM o Z Mo
= i~ d4 x ˚
d4 x′ f (x)a3 (η) g(x′ )a3 (η ′ ) ∆(x, x′ ) Iˆ (3.7)
Mo Mo
and the right hand side is a well-defined for all test functions f and g on the extended
space-time (M̊ , gab ). The physical correctness of the commutator is assured by following
considerations. First, ∆(x, x′ ) satisfies the Klein-Gordon equation w.r.t. gab in both argu-
ments. Second, given any C0∞ test-function f on M̊ , we obtain a solution F (x) by smearing
∆(x, x′ ) with f (x′ ). These solutions have smooth Cauchy data away from the η = 0 surface
and are well-defined as distributions on all of M̊ . Finally, the commutator (3.7) is equivalent
to the commutator [Φ̂(F1 ), Φ̂(F2 )] = i~ Ω(F1 , F2 ) Iˆ on the Fock space that correctly captures
the Poisson bracket relations on ΓCov .
B. Observables: K < 0
While φ̂(x) is a dimension 1 operator, hφ̂2 (x)iren has dimension 2 and hT̂ab (x)iren , dimen-
sion 4. Therefore, it is not a priori clear whether hφ̂2 (x)iren and hT̂ab (x)iren also remain
well-defined as distributions in the extended space-time (M̊ , gab ). In this sub-section we will
compute these quantities and show that they are. To our knowledge these expressions have
not appeared in the literature.
Recall that on the fiducial, ultra-static space-time (M̊ ,g̊ab ) the vacuum is a Hadamard
state: the bi-distribution hφ̂◦ (x) φ◦ (x ′ )i it defines has the Hadamard singularity struc-
ture as the points are brought together. Therefore one might expect that hφ̂(x) φ̂(x ′ )i =
(1/a(η)a(η ′)) hφ̂◦ (x) φ◦ (x ′ )i, would also have the Hadamard ultraviolet behavior away from
η = 0 [25]. This expectation is borne out in the following sense. The FLRW metric is
spatially homogeneous and isotropic and our choice of mode functions –and therefore of
the Fock vacuum they define– respects these symmetries. Consequently, as far as spatial
directions are concerned, it suffices to verify whether the hφ̂(x) φ̂(x ′ )i has the Hadamard
structure if the points are separated on an η = η0 surface along any one direction. We
verified that this is the case if the separation is in the radial direction and η0 6= 0. We also
verified the Hadamard behavior for points with ‘purely time-like’ separation, i.e., for points
of the type (~x, η) and (~x, η + ǫ) assuming that neither of these two points lie on the η = 0
surface. Note that the result holds also if the two points lie on the opposite sides of the
η = 0 surface, i.e., are separated by the big-bang singularity. (Since the metric gab vanishes
at η = 0, the notion of Hadamard behavior is not meaningful if they lies on this surface.)
As explained in Appendix A, in the calculation of hT̂ab (x)iren it is most convenient to
bring together points that have a ‘purely time-like’ separation. Therefore, we will calculate
hφ̂(x) φ̂(x ′ )i for two points x, x′ with ~x = ~x′ . Using Eq. (2.24), one obtains:
Z ∞
′ ~ ~
hφ̂(x) φ̂(x )i = 2 dk k e−ikǫ = lim − − 2 2 (3.8)
4π a(η)a(η − ǫ) 0 Im ǫ→0 4π ǫ a(η)a(η − ǫ)
where ǫ = η−η ′ . When expanding in ǫ << 1, it reproduces the Hadamard singular structure.
15
The physical observables, hφ̂(x) φ̂(x ′ )i and hT̂ab (x)iren can be computed from this bi-
distribution. The procedure using Hadamard renormalization (summarized in Appendix A)
is straightforward, but calculations are rather tedious. Using Mathematica and xAct we
obtain:
~
hφ̂2 (x)iren = − (3.9)
48π a2 (η)L2
2
and
24 cosh (2η/L) − 9 − 11 cosh (4η/L) 4
hρ̂iren = ~ a1 (3.10)
3840π 2 a8 (η)
−16 cosh (2η/L) + 47 + 11 cosh(4η/L) 4
hp̂iren =~ a1 (3.11)
3840π 2a8 (η)
As a non-trivial check, one can verify that the stress-energy tensor is conserved, i.e., the
only non-trivial component hρi′ren + 3 aa (hρiren + hpiren ) = 0 of ∇a hT̂ab (x)iren = 0 is satisfied.
′
Eqs. (3.9) and (3.10) have two interesting properties:

(i) All three expressions are manifestly IR finite; we did not have to introduce an IR regulator;
(ii) Away from the singularity, the expressions are manifestly smooth. But they diverge as
η −n for various values of n. Therefore, all these observables are well-defined as distributions
on the entire extended space-time (M̊, gab ). Note also that the 4-dimensional volume element
goes to zero as a(η)4R, lowering the effective value of n in the computation of the smeared
observables, such as d4 V f (x)hφ̂2(x)iren .
The full hT̂ab (x)iren can be written in a covariant form using curvature tensors:
1 cd 1 1 c d
11 1
hT̂ab (x)iren = R Racbd − Rab + 2Racbd u u − (gab + 4ua ub )
2880π 2 288π 2 (aL)2 1440π 2 (aL)4
(3.12)
where ua is the unit normal to the η = const surfaces. In the spatially-flat limit L → ∞ we
recover the standard result in the literature for a massless minimally coupled scalar field in
a spatially flat, radiated-dominated universe [26].
C. Observables: K > 0
The situation in the K > 0 case is analogous. The main differences are just
(i) Now η ∈ (−πL, πL) on the extended space-time (M̊ , gab ); and,
(ii) the integral over k is now a discrete sum and the sum over ℓ is bounded above as in
(2.33).
These differences arise because the global structure of space-time is quite different in the
K > 0 case from that in the K < 0 case. The vacuum state and the associated modes sense
this difference. Therefore, while in our discussion of the two scale factors (3.2) one could
pass from the K < 0 to the K > 0 by replacing the hyperbolic trigonometric functions
with trigonometric functions, this simple strategy is no longer valid for the renormalized
observables. Therefore, the form of the main equations is quite different from those in
Sec. III B.
As in Sec. III B, one can compute hφ̂(x) φ̂(x ′ )i using the completeness relation (2.24) and
it has Hadamard structure in the same sense. For reasons explained there, we can again
16
restrict ourselves to points that are ‘purely time-like separated’. Then the bi-distribution
reads:
∞
~ X
′ −ik ′ ǫ/L
hφ̂(x) φ̂(x ′ )i = k e
4π 2 L2 a(η)a(η − ǫ) k′ =1
~ 1 1
= lim − (3.13)
Im ǫ→0 8π L a(η)a(η − ǫ) (cos(ǫ/L) − 1)2
2 2
where we have set k ′ = k L and ǫ = η − η ′ . Using Mathematica and xAct we find (see
Appendix A for details):
hφ̂2 (x)iren = 0 , (3.14)
and
~ a41 ~ a41
hρ̂iren = and hp̂iren = (3.15)
960π 2 a(η)8 576π 2a(η)8
As a non-trivial check, one can verify that the stress-energy tensor is conserved, ∇a hT̂ab (x)iren =
0, i.e., hρi′ren + 3 aa (hρiren + hpiren ) = 0. Eq. (3.15) can be written in a manifestly covariant
′
way as
1
hT̂ab (x)iren = Rcd Racbd (3.16)
2880π 2
The three expressions are considerably simpler than those in the K < 0 case: as noted
above one cannot read them off from the K < 0 expressions because the field modes are
sensitive to the global structure of the spacetime –in particular the topology.
Remark: At first it may seem surprising that hφ̂2 (x)iren = 0 in the K = 0 [26], and K > 0
cases but not in the K < 0 case of Sec. III B. This difference can be understood in terms
of differences in the corresponding classical geometries as follows. Note first that by dimen-
sional considerations, hφ̂2 (x)iren has the form hφ̂2 (x)iren = ~S where S is a geometric scalar
with dimensions (length)−2 . In radiation-filled universes, there are only three independent
candidates: the spatial curvature 3R; square of the trace of extrinsic curvature, (a′′ /a2 )2 ;
and the time-time component of the 4-dimensional Ricci tensor, Rab ua ub where ua is the
unit normal to the η = const slices. It is convenient to write the right side of (3.9) as a
linear combination of these three scalars: hφ̂2 (x)iren = (1/576) Rab ua ub + 6(a′ /a2 )2 − 3R for
K < 0. In the case when K = 0 and K = +1/L2 cases, this combination happens to vanish
and we recover the result hφ̂2 (x)iren = 0 in these two cases.
IV. DUST-FILLED, CLOSED AND OPEN FLRW UNIVERSES
In dust-filled FLRW models, the spatial eigenvalue equation (2.17) is the same as in the
radiation-filled case but the time-evolution equation (2.18) is more complicated because the
space-time scalar curvature R is now non-zero. As in Sec. III we will divide our discussion
in three parts. In the first we will discuss the extension of the theory across the big-bang;
in the second we will discuss observables in the K < 0 models and, in the third, the K > 0
models.
17
A. Extension across the big-Bang
As in Sec. III A, it is convenient to introduce a variable L with dimensions of length,

given by K = 1/L2 if K > 0 and K = −1/L2 if K < 0; again, the limit K → 0 corresponds
to L → ∞ in both cases. For dust-filled universes, the scale factor is then given by
a(η) = (−2a2 L2 ) (cos(η/L) − 1), for K > 0 and

a(η) = (2a2 L2 ) (cosh(η/L) − 1), for K < 0 (4.1)
where a2 is a constant with dimensions of (length)−2 . The scale factor again vanishes at
η = 0 and the curvature diverges there. In standard treatments, the range of η is restricted
to lie in (0, π) for closed universes and to lie in (0, ∞) for open universes, so that η = 0
corresponds to the big-bang in both cases. (In the limit K → 0 we have a(η) = a2 η 2 , as
in the dust-filled case discussed in [1].) The space-time metric has the form gab = a2 (η)g̊ab ,
where g̊ab is the same ultra-static metric that we had in Sec. III and hence it can be extended
to the manifold M̊ on which η runs over the full real line, (−∞, ∞). However, with a(η) of
(4.1), the scalar curvature R of gab does not vanish. Therefore the relation between and
˚ is more complicated than in Sec. III. Using the conformal covariance property

1 ˚ − 1 R̊ F̊ (x)
− R F (x) = a−3 (η)

with F̊ (x) = a(η) F (x) (4.2)
6 6
one finds ′′
˚ + a (η) F̊ (x) = 0 .

F (x) = 0 iff (4.3)
a
Thus, now F̊ satisfies the Klein-Gordon equation with a time dependent potential V (η) =
a′′
a
(η) on the ultra-static space-time (M̊,g̊ab ). This makes the analysis more complicated
than that for radiation-filled universes. However, as in the K = 0 case, properties of this
potential enable one to select a preferred complex structure on the covariant phase space
Γ̊cov of solutions F̊ (x) which can then be naturally transferred to the phase space ΓCov of
solutions F (x) to the Klein-Gordon equation on the FLRW space-time (M̊, gab ), using the
relation F (x) = F̊ (x)/a(η) (see Sec. IV of [1]). The final results can be summarized as
follows.
Let us first consider the K < 0 case. In this case, the ‘positive-frequency’ basis functions
which make the action of the complex structure explicit –as in (2.28)– are given by
η −ikη

2kL coth 2L e
ek (η) = √ 1−i √ . (4.4)
2
1 + 4k L2 2kL 2k
They satisfy the evolution equation
h 1 i
e′′k (η) + k 2 − ek = 0, (4.5)
L2 cosh(η/L) − 1
the normalization condition (2.25), and tend to the basis

√ found in [1] in the limit K → 0.
−ikη
Note that, in contrast to the mode functions e / 2k in the radiation-filled case, each
ek (η) now diverges at η = 0 as a function, but η ek (η) is smooth on entire M̊ . Therefore,
each ek (η) is well-defined as a distribution on the extended manifold M̊ . On the extended
18
(K)
FLRW space-time (M̊ , gab ), these ek (η) and the spatial basis functions Y~k (~x) can be used
to define elements F (x) of ΓCov as in (2.26), and the OVD φ̂(x) as in (2.30). They satisfy
the Klein-Gordon equation in a distributional sense.
Finally, let us consider the case K > 0. Now the ‘positive-frequency’ modes satisfying
the normalization condition (2.25) are given by
η −ikη

2kL cot 2L e
ek (η) = √ 1−i √ (4.6)
1 − 4k 2 L2 2kL 2k
and they satisfy the evolution equation
h 1 i
e′′k (η) + k2 + ek = 0 . (4.7)
L2 cos(η/L) − 1
(Note that these modes and the equation they satisfy can be obtained from (4.4) and (4.5)
by the change L → iL. Also, because of the term kL in the numerator –which is absent in
the K = 0 case– there will be no infrared divergences in the present case.) On the original
FLRW space-time (M, gab ), η takes values in the open interval (0, πL) with big-bang at
η = 0 and big-crunch at η = π. For the extension M̊ of this space-time, we are led to
allow η to take values in the open interval (−πL, πL) that includes the big-bang of the
original space-time (which is now also the big-crunch of the portion of M̊ representing the
past big-bang branch). With these changes, the structure is completely analogous to that in
the K < 0 model. Therefore, in the K < 0 equations, we only have to replace the integral
over k by a sum (divided by L) and restrict the sum over ℓ as in (2.33). F (x) and φ̂(x) are
again well-defined and satisfy the Klein-Gordon equation as distributions over the extended
FLRW space-time (M̊ , gab ).
B. Observables: the K < 0 case
Using the expression (2.30) of the OVD φ̂(x) in terms of creation and annihilation oper-
ators, one can calculate the bi-distribution hφ̂(x) φ̂(x ′ )i. It is Hadamard in the same sense
as in Sec. III. More precisely, the mode functions –and hence the Fock vacuum– are again
invariant under the space-time isometries implementing homogeneity and isotropy and we
verified that hφ̂(x) φ̂(x ′ )i has the Hadamard structure for nearby points x, x′ that have ei-
ther ‘purely radial’ or ‘purely time-like’ separation (assuming neither of them lies on the
η = 0 surface). Again, for reasons explained in Appendix A, let us focus on points that have
‘purely time-like’ separation. Using Mathematica one obtains:
~ h ǫ ǫ − η η
hφ̂(x) φ̂(x ′ )i = lim − 3 − cosh + cosh + cosh ×
Im ǫ→0 64L2 π 2 a(η)a(−ǫ + η) L L L
iǫ ǫ − η η 1 ǫ2
Ci csch csch + √ 2G3,1 1,3 − | −1
1
2L 2L 2L π 16L2 −1, 0, 2
1 ǫ ǫ − η η h ǫ i i
+ 2 sinh Iπǫ − 4L csch csch + 2ǫ Shi (4.8)
ǫ 2L 2L 2L 2L
19
R∞ Rz
where Ci(z) := − z cos(t)/tdt is the cosine integral function, Shi(z) := 0 sinh(t)/tdt is
the hyperbolic sine integral function, and G3,1
1,3 is Meijer’s G function. Using this expression
and Hadamard renormalization implemented using Mathematica, we find
√

2 R 5 1
hφ̂ (x)iren = 2
− γ + log( 2µ− La(η)) − (4.9)
48π 6 48π L2 a(η)2
2
where as usual γ is the Euler-Mascheroni constant, and µ− is the renormalization scale for
the K < 0 dust-filled FLRW universe. Note that in the limit R → 0, we recover the result
of Sec. III B for the K < 0, radiation-free universe. For stress-energy tensor we obtain
~a22 h √
hρ̂ren i = 2 6
5 54 log 2a(η)L µ− − 54γ + 11
240π a (η)
a(η) √ a2 (η) i
+ 90 log 2a(η)L µ− − 90γ + 17 − 11 2 4 (4.10)
a2 L2 2a2 L
and
~a22 h √
hp̂ren i = 2 6
15 − 54 log 2a(η)L µ− + 54γ + 7
720π a (η)
2a(η) √ a2 (η) i
− (−90 log 2a(η)L µ − + 90γ + 28 − 11 . (4.11)
a2 L2 2a22 L4
One can verify that the stress-energy tensor is conserved, ∇a T ab = 0, i.e., hρi′ren +3 aa (hρiren +
′
hpiren ) = 0. In contrast to the radiation-filled universe, these expressions exhibit logarithmic

runnings with the renormalization scale µ− . This is a consequence of the fact that now the
equation governing time evolution has a potential, V (η) = a′′ /a. The spatially flat limit
is also more subtle than it was in Sec. III because, while there is no infrared divergence
for K < 0, there is one for K = 0 [2]. When these subtleties are taken into account, one
recovers the K = 0 results in the literature.
Finally, we can collect the results (4.10) and (4.11) to express hT̂ab (x)iren in terms of
space-time geometry:
1 9 2 1 √
hT̂ab (x)iren = − R g ab + RR ab + log( 2L µ − a(η)) − γRR ab
1728π 2 4 64π 2 √
1 1 107 log( 2L µ− a(η)) − γ 1 1
− Rg ab + R ab + Rg ab − 5R ab
1728π 2 (aL)2 10 96π 2 (aL)2 2
11 1
− 2 4
gab + 4ua ub . (4.12)
1440π (aL)
C. Observables: K > 0
As in Sec. III, to begin with there are only two main differences from the K < 0 universes
of Sec. IV B
(i) Now η ∈ (−πL, πL) on the extended space-time (M̊ , gab ); and,
(ii) the integral over k is now a discrete sum and the sum over ℓ is bounded above as in
(2.33).
However, the global structure of space-time is quite different in the K > 0 case from that
20
in the K < 0 case and, as in Sec. III, vacua associated with these modes sense this differ-
ence. Therefore, there are again notable differences in the expressions of the renormalized
observables.
As explained in Appendix A, it suffices to calculate hφ̂(x) φ̂(x ′ )i for points with ‘purely’
time-like separation. The result of this computation using Mathematica is
5iǫ η
e− 2L csc( 2L ) csc2 ( 2Lǫ ) csc( ǫ−η
2L )
3
hφ̂(x) φ̂(x ′ )i = lim − ~
iǫ iǫ
192π 2 L2 a(η)a(η−ǫ)
× 4 −1 + e 3 F2 12 , 2, 2; 1, 25 ; e−
L L
Im ǫ→0
5iǫ
h iǫ
i
+ 3 e 2L − 2 cos ǫ−2η ǫ 3ǫ 2 ǫ −1 − 2L

2L
+ 3 cos 2L
− cos 2L
− 2 sin L
tanh e (4.13)
where 3 F2 is the hypergeometric function. Using Hadamard renormalization we obtain

R √
hφ̂(x)2 i = −1 + log( 8 µ + La(η)) (4.14)
48π 2
Again, in the limit R → 0 we recover the result of Sec. III C for radiation-filled universe
with K > 0. For the renormalized density and pressure we obtain
~a22 η √
hρ̂iren = − 26 + 9 sin( ) 10 log[ 8a(η)µ + L] − 7 (4.15)
240π 2 a6 (η) 2L
~a22 √ 2 η √
hp̂iren = 5 54 log( 8a(η)µ + L) − 61 − 24 sin 10 log( 8a(η)µ + L) − 12
240π 2 a6 (η) 2L
(4.16)
These expressions again satisfy the conservation equation hρi′ren + 3 aa (hρiren + hpiren ) = 0,
′
and we find logarithmic runnings with the renormalization scale µ+ . Finally one can again
combine (4.15) and (4.16) to express hT̂ab (x)iren in terms of space-time geometry:
√
1 2 cd
log( 8L µ+ a(η))
hT̂ab (x)iren = R gab − 39R Racbd + (12Rcd Racbd − R2 gab )
1728π 2 √512π 2
1 1 log( 8L µ+ a(η)) 1
+ (Rgab + 30Rab ) + (Rgab − 10Rab )
640π 2 (aL)2 192π 2 (aL)2
(4.17)
Thus, once again, the observables hφ̂2 (x)iren and hT̂ab (x)iren are well-defined distributions
on the extended FLRW space-time (M̊ , gab ), and realized as smooth functions away from
the η = 0 surface.
V. OUTLOOK
In this work we extended the results of [1] to spatially closed and open universes and
showed that, in sharp contrast to classical test particles, the big-bang and the big-crunch
of the FLRW models are quite harmless to linear, test quantum fields. In particular, these
singularities do not represent an absolute beginning or absolute end of the universe for
them. The analysis of [1] was restricted to the K = 0 FLRW space-times and exploited
some simplifications that occur because of spatial flatness. Results of this paper show that
the tameness of the big-bang (and the big-crunch) experienced by quantum fields was not
an artifact of these simplifications; it is robust vis a vis inclusion of spatial curvature. It
21
turns out that a similar analysis can be carried out also for the Schwarzschild singularity
[6] which is more complicated for two reasons: (i) unlike FLRW models, space-time is not
spatially isotropic; and, (ii) while the Weyl curvature vanishes identically in the FLRW
models, it diverges at the Schwarzschild singularity. Nonetheless, the analysis shows that
this singularity is again tame when probed with test quantum fields. These results hint
at the possibility that all physically interesting space-like singularities of classical general
relativity are harmless when one uses physically appropriate tools to investigate their nature
[27].
A key point in all these investigations is to recognize that, since quantum fields φ̂(x)
are operator-valued distributions (OVDs) already in Minkowski space, one cannot expect
them to be better behaved in the vicinity of a singularity. In the more physically oriented
treatments of quantum field theory on cosmological space-times, one expands out φ̂(x) using
basis functions and it is implicitly assumed that they should be regular functions, as in
Minkowski space. But this is not necessary: For φ̂(x) to be a well-defined OVD, it suffices
that they are well-defined as distributions and satisfy the field equation in the distributional
sense. We found that this is the case for the basis functions normally used for minimally
coupled, massless scalar fields in closed and open FLRW models, even when the space-time
is extended across singularity. A priori there is an n-parameter ambiguity in defining the
distribution η −n that corresponds to inverse powers of conformal time η. However, φ̂(x) as
well as hφ̂2 (x)iren are insensitive to this ambiguity because the scale factor vanishes at η = 0
sufficiently rapidly.
It is also often assumed that the expectation values such as hT̂ab (x)iren should be smooth
tensor fields because the quantum corrected metric is obtained by solving the semi-classical
Einstein’s equation with the right side given by hT̂ab (x)iren , and the semi-classical metric is
expected to be a smooth tensor field. However, smoothness of hT̂ab (x)iren is not essential
from the quantum field theory perspective. Since φ̂(x) is an OVD and hφ̂(x) φ̂(x ′ )i is a bi-
distribution already in Minkowski space, there is no a priori reason to insist that observables
such as hφ̂2 (x)iren and hT̂ab (x)iren must be well-behaved functions. In particular, semi-
classical corrections in tame regions of space-time could be smooth but if one pushes the
theory all the way to the singularity of the classical background, the semi-classical solutions
may well become distributional. This is what happens, for example, when one includes the
back-reaction to the dynamics of evaporating Callen, Giddings, Harvey, Strominger black
holes [28], where quantum corrections soften the singularity and make the semi-classical
metric C 0 but not C 1 there [29, 30]. Indeed, already in classical general relativity, it has
been shown that the metric can be extended across physically interesting Cauchy horizons as
a field that is only C 0 [31, 32], so the curvature diverges there and field equations can hold
only in a distributional sense. Returning to semi-classical gravity, one would expect this
approximation to cease to be physically accurate already in the Planck regime, even before
the singularity is reached. But it could well capture some ‘faithful shadows’ of predictions
of a full-fledged quantum gravity theory. For example, there are strong indications from
Loop quantum gravity that space-time geometry becomes distributional in a specific sense
at Planck scale and the smooth Riemannian geometry of classical general relativity arises
only on coarse graining [33–37]. Using the form of the volume element, we found that
in both Radiation-filled and Dust-filled universes, d4 V hT̂ab (x)iren goes as 1/η 4 . Therefore,
although hT̂ab (x)iren can be made into a well-defined distribution on the extended space-time,
a priori there is a 4-parameter ambiguity in its definition. Since our emphasis was on the
existence of an extension of the theory beyond the big-bang, we chose to fix this ambiguity
22
using the simplest prescription that has intuitively expected mathematical properties. An
interesting open issue is whether this ambiguity can be removed by imposing compelling
physical requirements. Perhaps these will descend from quantum gravity considerations.
More generally, it is of considerable interest to better understand possible relations between
the appearance of fields that satisfy the desired equations only in a distributional sense
at these three different levels –classical general relativity, quantum field theory in curved
space-time and loop quantum gravity [27].
This investigation also suggests some directions for future work in quantum field theory
in curved space-times by itself. Chronologically, the theory was developed through the
introduction of concrete Fock representations of the canonical commutation relations, by
expressing φ̂(x) as a sum of creation and annihilation operators associated with suitable basis
functions (or modes) in specific space-times. Through a series of sustained mathematical
advances, the theory has been made significantly more general and formulated in generic
globally hyperbolic space-times. The emphasis has shifted to an algebraic approach in which
one introduces abstractly defined operator algebras (such as A(f ) and W of Sec. II), and
works with regular states (generally taken to be Hadamard) without necessarily constructing
the associated GNS representation (see, e.g., [12–15]). Powerful techniques from micro-
local analysis and wavefront sets have added much rigor and generality to the framework.
However, at their core these developments appear to be deeply intertwined with notion of
global hyperbolicity, and therefore not applicable –at least directly– once the space-time is
extended to include singularities, such as the (M̊, gab )’s considered in sections II - IV. On
the other hand, we saw that one can extend quantum field theory to such space-times. The
strategy outlined in Sec. II C shifts the emphasis from the abstract algebra A(f ) generated by
φ̂(f ) to the algebra A(F ) generated by Φ̂(F ) associated with suitable classical solutions F (x)
to the field equations, and the corresponding Weyl operators Ŵ (F ). In globally hyperbolic
space-times, we have well-defined retarded, advanced and commutator distributions, and one
can readily go back and forth between φ̂(f ) and Φ̂(F ) (and hence between the corresponding
Weyl operators). But once the FLRW space-time (M, gab ) is extended to (M̊, gab ) that
includes the big-bang/big-crunch singularity, we no longer have global hyperbolicity and it
is not obvious that the powerful techniques that have been developed for globally hyperbolic
space-times can be used. Nonetheless, we could associate operators Φ̂(F ) with certain
judiciously chosen distributional classical solutions F (x) for which, in particular, the 1-
particle norm (2.29) is well-defined in spite of the distributional character of the solutions.
Of course, space-times considered so far –FLRW models and the Schwarzschild ‘interior’–
are very special, like the cosmological space-times and Rindler wedges that were the focus
of attention in the early days of quantum field theory in curved space-times. But as far
as singularities are concerned, the FLRW and the Schwarzschild space-times are among
the most interesting examples from a physical perspective. Therefore, it would well-worth
investigating whether the powerful mathematical methods that have been introduced over
the last 2-3 decades can be extended to a judiciously chosen class of singular space-times.
Understanding the structure of quantum fields from a general perspective would provide
us with significant new insights on way to full quantum gravity which, we believe will
be singularity free because of the ultraviolet regularity of quantum geometry. Perhaps
distributional geometries provided by an appropriate generalization of the current semi-
classical gravity will provide a bridge between the low energy continuum geometries and
the full quantum Riemannian geometry at Planck scale, such as the one provided by loop
quantum gravity. We hope that results presented here and other related works will open a
23
door to extend the well-developed, mathematically rigorous quantum field theory in curved
space-times in these directions which have remained unexplored so far.
ACKNOWLEDGEMENTS
We would like to thank Marc Schneider for numerous discussions as well as comments
on the manuscript, and the York Mathematical Physics group of comments and questions
during a seminar on this subject. This work was supported by the NSF grant PHY-1806356,
and the Eberly Chair funds of Penn State. We acknowledge extensive use of some packages
of xAct for Mathematica.
Appendix A: Hadamard renormalization
For convenience of the reader, in this Appendix we call the renormalization procedure
that underlies our calculations of hφ̂2 (x)i and hTab (x)i used in sections III and IV. Given the
bi-distribution hφ(x)φ(x′ )i it is possible to compute these observables in a straightforward
manner using Hadamard renormalization. We review here the basic formalism, following
Ref. [38].
As is well-known, the product of two operator-valued distributions evaluated at the
same point is not well-defined. The first step towards renormalization typically involves
the regularization of divergences by splitting the points. Let U denote a convex normal
neighborhood of the spacetime manifold M. For any two points, x, x′ ∈ U, there exists
a unique geodesic that connects x and x′ and which lies entirely in U [39]. Let τ (x, x′ )
denote the geodesic distance between x and x′ , and ta (x) the geodesic’s tangent vector
at x, normalized as gab ta tb = ǫ (where ǫ = ±1 for space-like/time-like geodesics, respec-
tively). It is convenient to work instead with σ a (x, x′ ) := τ (x, x′ ) ta (x), which is a rescaled
tangent vector of the geodesic at x, with norm equal to the geodesic distance. This is a
vector at x and a scalar at x′ . This bi-tensor naturally leads to the Synge’s world function
σ(x, x′ ) := 21 σa (x, x′ )σ a (x, x′ ) = 2ǫ τ 2 (x, x′ ).
For sufficiently close points x 6= x′ , the bi-distribution hφ(x)φ(x′ )i computed in any
Hadamard state has the following singular structure: 3
U (x, x′ )

′ 1 ′ ′ ′
hφ̂(x) φ̂(x )i = 2 + V (x, x ) log σ (x, x ) + W (x, x ) (A1)
8π σ (x, x′ )
where U (x, x′ ) and V (x, x′ ) are smooth, real-valued bi-functions that depend only on the
local geometry of the spacetime, while W (x, x′ ) is a smooth, real bi-function that encodes
the information about the quantum state chosen.
It is customary to fix x, interpret U (x, x′ ) , V (x, x′ ) , W (x, x′ ) as functions of x′ , and then
expand them in covariant Taylor expansions around the point x [40]:
∞
′
X (−1)p
U (x, x ) = u(x) + ua1 ...ap (x)σ a1 . . . σ ap (A2)
p=1
p!
3
The precise sense in which the quasi-free states used in the main text satisfy Hadamard conditions is
spelled out in Secs. III and IV.
24
∞ ∞
X X (−1)p
V (x, x′ ) = Vn (x, x′ ) σ n , Vn (x, x′ ) = vn (x) + vna1 ...ap (x)σ a1 . . . σ ap (A3)
n=0 p=1
p!
∞ p
X (−1)
W (x, x′ ) = ω(x) + ωa1 ...ap (x)σ a1 . . . σ ap . (A4)
p=1
p!
The renormalized quantities of interest are given then by the following formulas, which
involve some of the Taylor coefficients in the previous expansions:
1
hφ̂2 (x)iren = ω(x) − v0 (x) log µ2

2
(A5)
8π
1 1 1
hT̂ab (x)iren = 2 −ωab (x) + − ξ ∇a ∇b ω(x) + ξ − gab ω(x) + ξRab ω(x)
8π 2 4
log µ2

1 1
+ v0ab (x) + gab v1 (x) − − ξ ∇a ∇b v0 (x) − ξ − gab v0 (x) − ξRab v0 (x)
8π 2 2 4
1
− 2 gab v1 (x) . (A6)
8π
where µ > 0 is an arbitrary renormalization scale. Notice that the first line in (A6) contains
all the information about the quantum state, while the third line contains the term that
gives the trace anomaly. Using identities from the Hadamard formalism, it can be proven
that ∇a hT̂ab (x)iren = 0.
The geometric contributions, v0 (x), v1 (x), v0ab (x), are universal (in the sense that do not
depend on the choice of the Hadamard state), and are determined by the background metric
according to expressions (A16), (A17) below. In particular, it is not difficult to see that
conformally coupled massless fields do not run with the renormalization scale. In contrast,
the quantities ω(x) and ω0ab (x), which depend on the choice of quantum state, must be
computed in detail for each problem. This is the non-trivial part of the calculation. If we
define the singular part of the distribution as
U (x, x′ )

′ 1 ′ ′
hφ̂(x)φ̂ (x )ising := 2 + V (x, x ) log σ (x, x ) (A7)
8π σ (x, x′ )
then, ω(x) and ω0ab (x) can be obtained from the bi-distribution by directly applying the
formulas:
h i
W (x, x′ ) = 8π 2 hφ(x)φ (x′ )i − hφ(x)φ (x′ )ising (A8)
ω(x) = lim
′
W (x, x′ ) (A9)
x →x
ωab (x) = lim ∇a ∇b W (x, x′ ) (A10)
′ x →x
The strategy to obtain the renormalized observables (A5), (A6) is the following. We
fix two points x and x′ of the spacetime manifold, and compute hφ(x)φ (x′ )i (the strategy
for choosing these two points in the most convenient manner is discussed below). Then,
we compute hφ̂(x)φ̂ (x′ )ising using (A7). To get this quantity we need to evaluate U(x, x′ ),
V (x, x′ ) and σ(x, x′ ) to the required order in point-splitting. Up to the required order to
25
compute (A9) - (A10) in 4 dimensions, the covariant Taylor expansion of the functions
U(x, x′ ) and V (x, x′ ) are
1 1 1
U(x, x′ ) = u(x) − ua (x)σ a + uab (x)σ a σ b − uabc (x)σ a σ b σ c + uabcd (x)σ a σ b σ c σ d + O σ 5/2

2! 3! 4!
V (x, x′ ) = V0 (x, x′ ) + V1 (x, x′ )σ(x, x′ ) + O σ 3/2 (A11)
with
1
V0 (x, x′ ) = v0 (x) − v0a (x)σ a + v0ab (x)σ a σ b + O σ 3/2

(A12)
2!
V1 (x, x′ ) = v1 (x) + O σ 1/2 (A13)
The coefficients in this expansion are completely determined by the background metric,
according to
1 1
u0 = 1, u0a = 0,u0ab = Rab , u0abc = R(ab;c) , (A14)
6 4
3 1 1 pq
u0abcd = R(ab;cd) + R(ab Rcd) + Rp(a|q|b Rcd) (A15)
10 12 15
and

1 2 1 1 1
v0 = m + ξ− R, v0a = ξ− R;a , (A16)
2 6 4 6

1 2 1 3 1 1 1
v0ab = m Rab + ξ− R;ab − Rab + ξ− RRab
12 6 20 120 12 6
1 1 pq 1 pqr
+ Rap Rpb − R Rpaqb − R Rpqrb
90 180 180 a
and
2
m4 m2 R R2 (Rpq Rpq − Rpqrs Rpqrs )

1 R 1 1
v1 = + ξ− − ξ− + ξ− − (A17)
8 4 6 24 5 8 8 720
On the other hand, to obtain the bi-tensor σ a (x, x′ ) we follow Appendix B of [41], and from
this we can readily obtain the Synge world’s function by σ(x, x′ ) := 21 σa (x, x′ )σ a (x, x′ ).
While the procedure just described can be implemented for any two points x, x′ in a
normal neighborhood, the specific calculation can be greatly simplified if x, x′ are cho-
sen conveniently. For a quantum state that respects the homogeneity and isotropy of
the FLRW spacetime background, hT̂ab (x)iren has the perfect fluid form, hT̂ab (x)iren =
(hρ̂iren + hp̂i)ua ub + hpiren gab , with ua , the unit time-like normal to the η = const slices,
and
hρ̂iren = ua ubhT̂ab (x)iren (A18)
1 1
hp̂iren = hab hT̂ab (x)iren = hρ̂iren + hT̂aa iren (A19)
3 3
It follows from Eqs. A6, A9 and A10 that, to evaluate hρ̂iren we need ω(x), ωab (x)ua ub and
to evaluate hT̂aa iren we need ωaa (x). Finally, because of the identity ωaa (x) = (m2 + ξR)ω(x) −
6v1 (x) from the Hadamard formalism, one only needs to calculate ω(x) and ωab (x)ua ub .
To compute ω(x), one can take the point-splitting in any direction. However, to calculate
26
1
ωab (x)ua ub = a(η) 2 ωηη (x) it is crucial to do the splitting in the time-like direction. Therefore,
it suffices to calculate the bi-distribution for points that have a ‘purely time-like’ separation:
x = (η, χ, θ, φ), x′ = (η ′, χ, θ, φ). Then, the quantity of interest ωηη (x) emerges in the
covariant Taylor expansion:
1 2
W (x, x′ ) = w(x) − wη (x) (η − η ′ ) + . . . + wηη (x) (η − η ′ ) + . . . (A20)
2
To perform the calculation we express both hφ(x)φ (x′ )i and hφ(x)φ (x′ )ising in terms of
ǫ = η − η ′ , compute Ŵ (x, ǫ) ≡ W (x, x − ǫ) using (A8), and then we evaluate
ω(x) = Ŵ (x, ǫ) (A21)

ǫ=0
ωηη (x) = a2 ua ub ∇a ∇b Ŵ (x, ǫ) = ∂η2 − H∂η Ŵ (x, ǫ)

. (A22)
ǫ=0 ǫ=0
We then calculate the rights sides of (A18)-(A19) using these results and equation (A6).
The result of doing this calculation for the 4 spacetimes considered in the main text yields:
(i) equations (3.10)-(3.11) for the radiation filled, K = −1/L2 universe; (ii) equations (3.15)
for the radiation-filled K = +1/L2 universe; (iii) equations (4.10)-(4.11) for the dust-filled,
K = −1/L2 universe; and, (iv) equations (4.15)-(4.16) for the dust-filled K = +1/L2
universe.
A useful identity that can be used as a check during intermediate calculations is ωa (x) =
1
∇ ω(x).
2 a
Appendix B: Radial component of the Klein-Gordon equation
For completeness, in this appendix we show how to obtain solutions (2.20) to the radial
equation (2.17) for the field modes, as well as their main properties and the orthonormality
(2.22) and the addition formula (2.24). Several of these results are stated without proof in
the literature (e.g., [18]) but we were not able to find a complete treatment. This Appendix
provides an essentially self-contained derivation for the convenience of the reader.
1. Main solutions and properties
With the ansatz (2.19) the spatial equation (2.20) reduces to

√ √

2 Kℓ(ℓ + 1) 2
∂χ + 2 K cot( Kχ)∂χ − √ + k − K ΠK,kℓ = 0 (B1)
sin2 ( Kχ)
First we note that the solution to this equation is closely related to the associated Legendre
polynomials. These are solutions to
2 µ2

2 d u du
1−z − 2z + ν(ν + 1) − u = 0. (B2)
dz 2 dz 1 − z2
√
By performing a change of variables u = v(z 2 − 1)1/4 , and z = cos( Kx), this equation
becomes
√ √ µ2 − 41

2 3
∂x + 2 K cot( Kx)∂x + K ν(ν + 1) − − √ v = 0. (B3)
4 sin( Kx)
27
This equation can be identified with (B1) provided that µ2 − 14 = ℓ(ℓ+1) and K(ν(ν+1)− 43 ) =
k 2 − K. We can solve these equations to get µ = ±(ℓ + 12 ) and ν = − 21 ± √kK . Therefore,
1 ±(ℓ+ 12 ) √
q √ P 1
−2+√ k (cos( Kx)) (B4)
K
sin( Kx)
are two linearly independent solutions to (B1). 4 We have to show now its connection to
(2.20). Let us focus first on the + solution. We will use the identity (see eqn 8.733(1) in
[43])
dPνµ (x) √
1 − x2 = − 1 − x2 Pνµ+1 (x) − µxPνµ (x) (B5)
dx
From this identity one can easily infer
√
√ √ d √ cos( Kx) a √
Pνa+1 (cos( Kx)) = − sin( Kx) √ Pνa (cos( Kx)) − a √ Pν (cos( Kx))
d(cos( Kx)) sin( Kx)
√
1 d a √ cos( Kx) a √
=√ Pν (cos( Kx)) − a √ Pν (cos( Kx))
K dx sin( Kx)
√ √
sina ( Kx) d Pνa (cos( Kx))
= √ √
K dx sina ( Kx)
√
√ d P a
ν (cos( Kx))
= − sina+1 ( Kx) √ a
√ (B6)
d cos( Kx) sin ( Kx)
Proceeding recursively, we obtain:
ℓ a √
√ √

d Pν (cos( Kx))
Pνa+ℓ (cos(
Kx)) = (−1) sin ( Kx) ℓ a+ℓ
√ √ (B7)
d cos( Kx) sina ( Kx)
1/2 √ q
For a = 1/2 we have (see eqn 8.754(1) in [43]), Pν (cos( Kx)) = π sin(2√Kx) cos(kx), so,
up to an irrelevant constant, one concludes
ℓ+1
√ √

1 (ℓ+ 21 ) ℓ d
q √ P− 1 + √k (cos( Kx)) ∼ sin ( Kx) √ sin(kx) (B8)
sin( Kx)
2 K d cos( Kx)
which is one of the two linearly independent solutions of (2.20). Let us focus now on the −
solution above. For this we will use the following identity (see eqn 8.731, 1(2) in [43]):
dPνµ(z) √
z2 − 1 = (ν + µ)(ν − µ + 1) z 2 − 1Pνµ−1 (z) − µzPνµ (z) (B9)
dz
This identities yields
√
(ν + a)(ν − a + 1) Pνa−1 (cos( Kx))
4 µ −µ
For K = 1/L2 , µ+ + ν is an integer but µ+ isn’t, so by statement 8.707 (4) in [43] Pν + (z) and Pν + (z)
µ
are linearly independent. On the other hand, from the identity P−ν−1 (x) = Pνµ (x) (see eqn 8.733(5) in
µ µ
[43]) we get P− 1 + √k (x) = P− 1 − √k (x) so the one of the two signs of ν± is redundant.
2 K 2 K
28
√
√ d a
√ a cos( Kx) a √
= +i sin( Kx) √ Pν (cos( Kx)) + √ Pν (cos( Kx))
d(cos( Kx)) i sin( Kx)
√
√ d Pνa (cos( Kx))
= −i sin−a+1 ( Kx) √ √ (B10)
d cos( Kx) sin−a ( Kx)
We have skipped most of the steps, as the calculation is very similar to (B6). Proceeding
recursively:
ℓ a √
√ √

Γ(ν − a + 1)Γ(ν + a − ℓ + 1) d P ν (cos( Kx))
Pνa−ℓ (cos( Kx)) = −i sin−a+ℓ ( Kx) √ −a
√
Γ(ν − a + ℓ + 1)Γ(ν + a + 1) d cos( Kx) sin ( Kx)
−1/2
√ q √
For a = −1/2 we have (see eqn 8.754(3) in [43]), Pν (cos( Kx)) = π sin(2√Kx) kK sin(kx).
As a result, we obtain
−(ℓ+ 1 ) √
P− 1 + √2k (cos( Kx)) √ √ ℓ+1
√

2 K 2K −i Γ(k/ K − ℓ) ℓ d
q √ = √ √ sin ( Kx) √ cos(kx) (B11)
πk Γ(k/ K + ℓ + 1) d cos( Kx)
sin( Kx)
which is the remaining linearly indepenent solution of (2.20).
Let us analyze the behavior of these two √ linearly independent solutions (B8) and (B11)
ℓ
when x = 0. Because of the prefactor sin ( Kx) in both of them, √ the result vanishes at
x = 0 unless we can extract an equal number of powers of sin( Kx) from the derivative
d√ 1 √ d
d cos( Kx)
= −√K sin( Kx) dx
. For (B11) we obtain
ℓ+1
√ dℓ d cos(kx)

ℓ d 1
sin ( Kx) √ cos(kx) = √ √
d cos( Kx) x=0
(− K)ℓ dxℓ d cos( Kx) x=0
ℓ
−k d sin(kx)
= √ √ (B12)
(− K) dxℓ sin( Kx)
ℓ+1
x=0
sin(kx)
The function sin( √
Kx)
is smooth around x = 0, so its derivatives are all well-defined for any
ℓ at x = 0. Let us do a similar analysis for (B8)
ℓ+1
√ dℓ d sin(kx)

ℓ d 1
sin ( Kx) √ sin(kx) = √ √
d cos( Kx) (− K) ℓ dxℓ d cos( Kx)
x=0 x=0
k dℓ cos(kx)
= √ √ (B13)
(− K)ℓ+1 dxℓ sin( Kx) x=0
cos(kx)
Kx)
and all its derivatives, in contrast to the previous case, are singular at
x = 0. If the general solution of (B1) is a linear combination of (B8) and (B11), demanding
regularity at the origin x = 0 requires the coefficient of (B8) to be zero.
For K = +1/L2 , the coordinate x is bounded above by √ πL, and ℓone can do a similar
ℓ
analysis of regularity around this point. The prefactor sin ( Kx) = sin (x/L) again vanishes
when x = πL:
ℓ+1 ℓ
√

ℓ d −k d sin(kx)
sin ( Kx) √ cos(kx) = √ √ (B14)
d cos( Kx) (− K) ℓ+1 dx sin( Kx)
x=πL x=πL
29
sin(kx)
Kx)
is not smooth at x = πL unless we demand kL ∈ Z. The property
kL ∈ Z imposes in turn another constraint. Indeed, from the trigonometric identity
n/2 i
X X
i−j n i
cos(nx) = (−1) cosn−2(i−j) x (B15)
2i j
i=0 j=0
√
√ when n is integer, one infers that cos(kx) is a polynomial of2 cos( Kx) of degree
valid
k/ K = kL. As a result, (B11) is zero when ℓ ≥ kL. For K = +1/L we therefore restrict
ℓ < kL.
Notice that, for K = −1/L2 , kL is not necessarily an integer, so the reasoning above
cannot be applied in this case and ℓ is in general unbounded.
2. Orthogonality and normalization
We discuss now the normalization factor for the field modes. We follow the logic of [42]
(in particular, see its section II.D). Before we argued that the radial dependence of the field
modes, subject to regularity conditions at the origin, satisfies
ℓ+1
√

ℓ d
ΠK,kℓ(χ) = Akℓ sin ( Kχ) √ cos(kχ) . (B16)
d cos( Kχ)
The full space-like modes YK,~k (~x) = ΠK,kℓ (χ)Yℓm (θ, φ) are required to be normalized as
(2.22). Using the familiar orthogonality condition for the spherical harmonics, we can par-
tially evaluate this integral:
Z √ 3 |Akℓ |2 xK
Z
YK,~k (~x)ȲK,~k′ (~x) h d x = δℓ,ℓ′ δm,m′ qℓ+1 (k1 , x) qℓ+1 (k2 , x) dx (B17)
Σ K 0
where xK ≡ πL for K = 1/L2 and xK ≡ +∞ for K = −1/L2 . To alleviate the notation we

introduced the function
ℓ
√

ℓ d
qℓ (k, x) = sin ( Kx) √ cos(kx) (B18)
d cos( Kx)
This function satisfies the following identities,
1 dqℓ (k, x) √
qℓ+1 (k, x) = − √ + ℓ cot( Kx)qℓ (k, x) (B19)
dx
K
2 2 Kℓ(ℓ + 1)
0 = ∂x + N − √ qℓ (k, x) (B20)
sin2 ( Kx)
√ dqℓ (k, x) √
(k 2 − K(ℓ − 1)2 )qℓ−1 (k, x) = K + K(ℓ − 1) cot( Kx)qℓ (k, x) (B21)
dx
The first identity is easy to check from (B18). The second identity is a consequence of the
differential equation that satisfies ΠK,kℓ(χ):
√ √

2 Kℓ(ℓ + 1) 2 qℓ+1 (k, χ)
∂χ + 2 K cot( Kχ)∂χ − 2
√ +k −K √ =0 (B22)
sin ( Kχ) sin( Kχ)
30
The third identity can be obtained easily from the first two. Using these functional relations
we can now compute the full integral of interest:
Z xK Z xK √

1 ′
qℓ+1 (k1 , x)qℓ+1 (k2 , x)dx = qℓ+1 (k1 , x) − √ qℓ (k2 , x) + ℓ cot( Kx)qℓ (k2 , x) dx
0 0 K
Z xK √

1 ′
= qℓ (k2 , x) √ qℓ+1 (k1 , x) + ℓ cot( Kx)qℓ+1 (k1 , x) dx
0 K
1
− √ qℓ qℓ+1 |x0 K (B23)
K
To proceed with the calculation we will distinguish the two cases.
1) Let K = 1/L2 . In this case, it easy to find from the definition that qℓ (k, 0) =
qℓ (k, πL) = 0 for odd ℓ. Therefore the boundary term above vanishes for all ℓ. Using
one of the identities introduced above, we get
Z xK 2 Z xK
k1 2
qℓ+1 (k1 , x)qℓ+1 (k2 , x)dx = −ℓ qℓ (k1 , x)qℓ (k2 , x)dx (B24)
0 K 0
Proceeding recursively:
Z xK Z xK
k12 k12
2
2 k1 2
qℓ+1 (k1 , x)qℓ+1 (k2 , x)dx = − 1 ... −ℓ q0 (k1 , x)q0 (k2 , x)dx (B25)
0 K K K 0
The last integral is easy to evaluate:

Z xK xK
π
Z
q0 (k1 , x)q0 (k2 , x)dx = cos(k1 x) cos(k2 x)dx = Lδk1 ,k2 (B26)
0 0 2
R 3
√
Collecting all these results, and demanding Σ
d x hYK,~k (~x)ȲK,~k′ (~x) = Lδk,k′ δℓ,ℓ′ δm,m′ , we
end up with
√
eiφkℓ K
Akℓ = q (B27)
π ℓ
Π [k 2 /K
2 i=0
− i2 ]
where φkℓ is an arbitrary phase. Note that the denominator never vanishes because ℓ < kL.
Note also that (B16) is well-defined for √ k = 0 with √ this normalization factor for any ℓ.
Taking into account that k 2 /K − i2 = (k/ K + i)(k/ K − i), this expression can be further
written as
q √
2K 3/2 Γ(k/ K − ℓ)
Akℓ = eiφkℓ q √ (B28)
π k Γ(k/ K + ℓ + 1)
√ √ √
where we used Γ(k/ K + 1)/Γ(k/ K) = k/ K.
2) Let K = −1/L2 . In this case qℓ (k, 0) = 0 for odd ℓ but qℓ (k, ∞) is oscillatory. To deal
with this case, we work with an integrated version of (B23):
Z +∞ Z xK Z +∞ Z xK √
′
qℓ+1 (k1 , x)

dk2 qℓ+1 (k1 , x)qℓ+1 (k2 , x)dx = dk2 qℓ (k2 , x) √ + ℓ cot( Kx)qℓ+1 (k1 , x) dx
−∞ 0 −∞ 0 K
31
+∞
1
Z
− lim dk2 √ qℓ (k2 , x) qℓ+1 (k1 , x)
x→∞ −∞ K
It is not difficult to see that, when x → ∞, the function qℓ (k, x) asymptotically equals
pℓ (k, x) = f1 (k, ℓ) cos(kx)+f2 (k, ℓ) sin(kx) for some functions f1 , f2 . The product pℓ (k2 , x)pℓ+1 (k1 , x)
will be a linear combination of cos((k1 ± k2 )x), sin((k1 − k2 )x). Subtracting and summing
this function inside the second integral above, and applying the Riemann-Lebesgue lemma,
this integral is zero. Doing now exactly the same calculation as in the K = +1/L2 case,
using the following integral representation of the Dirac delta 5
Z ∞ Z ∞
1 ∞
Z
q0 (k1 , x)q0 (k2 , x)dx = cos(k1 x) cos(k2 x)dx = cos(k1 x) cos(k2 x)dx
0 0 2 −∞
1 ∞
Z
= [cos((k1 − k2 )x) + cos((k1 + k2 )x)]dx
4 −∞
2π π
= (δ(k1 − k2 ) + δ(k1 + k2 )) = δ(k1 − k2 ) (B29)
4 2
√
and demanding Σ d3 x h YK,~k (~x)ȲK,~k′ (~x) = δ(k1 − k2 )δℓ,ℓ′ δm,m′ , we end up with the same
R
coefficient (B28).
3. Addition Theorem
Using (B11) with (B16) and (B28) we can write

−(ℓ+ 1 ) √
s √ P− 1 + √2k (cos( Kx))
√ Γ(k/ K + ℓ + 1) 2 K
ΠK,kℓ (χ) = ieiφkℓ kK 1/4 √ q √ (B30)
Γ(k/ K − ℓ) sin( Kx)
From the relation with the Gegenbauer function [see eqn 8.936(1) in [43]]
−(ℓ+ 12 ) √
P (cos( Kx)) √
ℓ+1
√ − 12 + √k
K (−1)−1/4−ℓ/2 Γ(ℓ + 1 + k/ K)Γ(ℓ + 3/2)
Ck−(ℓ+1) (cos( Kx)) = √ √
sin1/2+ℓ ( Kx) 2−ℓ−1/2 Γ(2ℓ + 2)Γ(k/ K − ℓ)
we can infer
s √
iφ′kℓ
√ √ √ Γ(k/ K − ℓ) 2ℓ+1/2
ΠK,kℓ (χ) = e kK 1/4 sinℓ ( Kx)Ck−(ℓ+1)
ℓ+1
(cos( Kx)) √ √ Γ(ℓ + 1)
Γ(k/ K + ℓ + 1) π
22(ℓ+1)−1
where we used the identity Γ(2ℓ + 2)/Γ(ℓ + 3/2) = √
π
Γ(ℓ + 1) [see eqn 8.335(1) in [43]],
and introduced a new phase φ′kℓ := φkℓ + 5π
4
+ πℓ
2
.
Let us compute the addition formula:
∞
X X
2
X 2ℓ + 1
YK,~k (~x)ȲK,~k (~y ) = ΠK,kℓ(x)Π̄K,kℓ (y)|Yℓm(θ, φ)| = ΠK,kℓ (x)Π̄K,kℓ(y)
ℓm ℓm ℓ=0
4π
5
δ(k1 + k2 ) = 0 because k1 , k2 ≥ 0, and in the case k1 = k2 = 0 the prefactor k12 in (B25) makes it vanish
32
∞
kK 1/2 X ℓ √ ℓ+1
√ ℓ
√ ℓ+1
√
= sin ( Kx)C k−(ℓ+1) (cos( Kx)) sin ( Ky)C k−(ℓ+1) (cos( Ky))
2π 2
ℓ=0
√
22ℓ Γ(k/ K − ℓ)Γ(ℓ + 1)2
× √ (2ℓ + 1)
Γ(k/ K + ℓ + 1)
where in the second equality we used the addition theorem for the spherical harmonics.
Note that, when K = 1/L2 , the sum is finite as the summand is non-zero only when
ℓ = 0, . . . , kL − 1. Using now the addition theorem for Gegenbauer polynomials [see eqn
1/2
8.934(3) in [43]; use the identity Cℓ (cos 0) = Pℓ (1) = 1 from eqn 8.936(3) in [43]] we obtain
X kK 1/2 1 √
YK,~k (~x)ȲK,~k (~y ) = 2
Ck−1(cos( K(x − y))) (B31)
ℓm
2π
Using again the relation between Gegenbauer and Legendre polynomials [eqn 8.936(1) in [43]]
√ −1 √
1
(cos( K(x − y))) = √kK π2 (−4)1−1/4 √ √1
p
we find Ck−1 P− 12+ √k (cos( K(x − y))). We
sin( K(x−y)) 2 K
−1/2
√ q √
also have (see eqn 8.754(3) in [43]), Pν (cos( Kx)) = π sin( Kx) kK sin(kx), so we finally
2√
get
X kK 1/2 sin(k(x − y))
YK,~k (~x)ȲK,~k (~y ) = √ (B32)
ℓm
2π 2 sin( K(x − y))
[1] A. Ashtekar, T. De Lorenzo and M. Schneider, Probing the big bang with quantum fields,
Adv. Theo. Math. Phys. 25, 1651-1702 (2021).
[2] L. H. Ford and L. Parker, Infrared divergences in a class of Robertson-Walker universes, Phys.
Rev. D. 16, 245-250 (1977)
[3] G. T. Horowitz, and D. Marolf, Quantum probes of space-time singularities, Phys. Rev. D.
52, 670-5675 (1995).
[4] Stalker, J. D. and Tahvildar-Zadeh, A. S. Scalar waves on a naked-singularity background,
Class. Quant. Grav. 21, 2831-2848 (2004).
[5] S. Hofmann and M. Schneider, Classical versus quantum completeness, Phys. Rev. D 91,
125028 (2015).
[6] A. Ashtekar and M. Schneider, Probing the Schwarzschild singularity with quantum fields (in
preparation).
[7] S. Hofmann, M. Schneider and M. Urban, Quantum complete prelude to inflation, Phys. Rev.
D 99 065012 (2019).
[8] O. Steinmann, Perturbation Expansions in Axiomatic Field Theory, volume 11 of Lecture
Notes in Physics, (Springer, Berlin, (1971)).
[9] R. Brunetti and C. Fredenhagen, Microlocal analysis and interacting quantum field theories:
renormalization on physical backgrounds, Commun. Math. Phys. 208 623-661 (2000); Section
5.1.
[10] R. M. Wald, Quantum field theory in curved space-times and black hole thermodynamics (Uni-
versity of Chicago Press, Chicago, 1994).
33
[11] A. Ashtekar and A. Magnon, Quantum fields in curved space-times, Proc. R. Soc.(London),
A 346 375-394 (1975).
[12] R. Brunetti, K. Fredenhagen and R. Verch, The generally covariant locality principle: A new
paradigm for local quantum field theory, Commun. Math. Phys. 237, 31-68 (2003).
[13] M. Benini, C. Dappiaggi, and T.P. Hack, Quantum Field Theory on Curved Backgrounds –
A Primer, Int. J. Mod. Phys. A, 28, 1330023 (2013).
[14] S. Hollands and R. M. Wald, Quantum fields in Curved space-times, in General Relativity and
Gravitation: A Centennial Perspective, A. Ashtekar, B. Berger, J. Isenberg and M. A. H. Mc-
Callum (Eds).; Cambridge University Press: Cambridge (2015).
[15] C. J. Fewster and K. Rejzner, Algebraic Quantum Field Theory – an introduction, in: Progress
and Visions in Quantum Theory in View of Gravity - Bridging Foundations of Physics and
Mathematics, eds. F. Finster, D. Giulini, J. Kleiner, J. Tolksdorf, (Birkhäuser, Chem, 2020),
p 1-62; arXiv:1904.04051v2.
[16] I. M. Gel’fand and M. A. Naimark, On the embedding of normed rings into the ring of
operators in Hilbert space, Mat. Sobrn. 12, 197-217 (1943).
[17] I. E. Segal, Postulates of general quantum mechanics, Ann. Math.48, 930-948 (1947).
[18] L. Parker and S. A. Fulling, Phys. Rev. D 9, 341 (1974); T. S. Bunch, Phys. Rev. D 18, 1844
(1978); P.R. Anderson and L. Parker, Phys. Rev. D 36, 2963 (1987).
[19] A. Lichnerowicz, Propagateurs, commutateurs et anticommutateurs en relativité générale,
1964 Les Houches Lectures, English translation reprinted in General Relativit. and Grav. 50,
145 (2018).
[20] M. Abramowitz; I. A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs,
and Mathematical Tables.. Applied Mathematics Series. Vol. 55 (Dover Publications, N.Y.
(1995)) .
[21] L. Schwartz, Some Applications of the Theory of Distributions, in Lectures on Modern Math-
ematics, vol I (Wiley, New York, (1963)), p23-58
[22] I. Gel’fand and G. Shilov, Generalized Functions, Vol. 1 Gosudrstv. Izdat. Fiz-Mat. Lib.:
Moscow 1958; English translation: (AMS Chelea Publication: Providence, 2016).
[23] L. Hörmander, On the division of distributions by polynomials, Arkiv für Math. 3, 555-569
(1958).
[24] S. Lojasiewicz, Sur le problèm de la division, Stud. Math. 19, 87-136 (1958).
[25] S. Tadaki, Hadamard regularization and conformal transformation, Prog. Theo. Phys. 81,
891-903 (1988).
[26] T. S. Bunch, P. C. W. Davies, Non-conformal renormalised stress tensors in Robertson-Walker
space-times, J. Phys. A: Math. Gen. 11, 1315-1328 (1978).
[27] A. Ashtekar, A. del Rı́o and M. Schneider, Space-like Singularities of General Relativity: A
Phantom menace? Gen. Relativ.Grav. 54, 45 (2022).
[28] C. G. Callan, S. B. Giddings, J. A. Harvey and A. Strominger, Evanescent black holes, Phys.
Rev. D 45, R1005-R1009 (1992).
[29] A. Ori, Approximate solution to the CGHS field equations for two-dimensional evaporating
black holes, Phys. Rev.D 82, 104009 (2010).
[30] A. Ashtekar, F. Pretorius and F. Ramazanoğlu, Surprises in the Evaporation of 2-Dimensional
Black Holes, Phys. Rev. Lett. 106 161303 (2011);
Evaporation of 2-Dimensional Black Holes, Phys.Rev.D 83, 044040 (2011).
[31] M. Dafermos, The interior of charged black holes and the problem of uniqueness in general
relativity, Comm. Pure Appl. Math. 58, 0445–0504 (2005).
34
[32] M. Dafermos and J. Luk, The interior of dynamical vacuum black holes I: The C 0 -stability of
the Kerr Cauchy horizon, arXiv:1710.01722.
[33] Ashtekar, A and Lewandowski, J. 2004 Background independent quantum gravity: A status
report, Class. Quant. Grav. 21, R53-R152 (2004).
[34] C. Rovelli, Quantum Gravity, (Cambridge UP, Cambridge (2004)).
[35] Thiemann, T. 2007 Introduction to Modern Canonical Quantum General Relativity. (Cam-
bridge UP, Cambridge (2007)).
[36] K. Giesel, Quantum Geometry, in Loop Quantum Gravity: The First 30 Years, A. Ashtekar
and J. Pullin Eds, (World Scientific, Singapore 2017).
[37] A. Ashtekar and E. Bianchi, A short review of loop quantum gravity, Rep. Prog. Phys. 84,
042001 (2021)
[38] D. Bernard and A. Folacci, Phys. Rev. D 34, 2286 (1986); Y. Decanini and A. Folacci Phys.
Rev. D 78 044025, (2008).
[39] B. O’Neill, Semi-Riemannian Geometry with Applications to Relativity (Academic Press, San
Diego, California (1983)).
[40] S. M. Christensen, Vacuum expectation value of the stress tensor in an arbitrary curved
background: The covariant point-separation method, Phys. Rev. D 14, 2490-2501 (1976);
Phys. Rev. D, 17, 946-963 (1978).
[41] P. R. Anderson, W. A. Hiscock, and D. A. Samuel, Phys. Rev. D 51, 4337-4358 (1995).
[42] M. Bander and C. Itzykson, Rev. Mod. Phys. 38, 346-358 (1966).
[43] I. S. Gradshteyn, I. M. Ryzhik, Table of Integrals, Series, and Products. Translated by Scripta
Technica, Inc. (3 ed.). Academic Press.

2209.09922v1

Uploaded by

Copyright:

Available Formats

2209.09922v1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2209.09922v1

Uploaded by

Copyright:

Available Formats

Probing cosmological singularities with quantum fields:

Open and closed FLRW universes

accessible to both communities, we have tried to strike a balance between mathematical

II. LINEAR QUANTUM FIELDS IN FLRW SPACE-TIMES

A. Operator algebras and their quasi-free representations

(iii) The commutation relations:

[Φ̂(F1 ), Φ̂(F2 )] = i~ Ω(F1 , F2 ) Iˆ . (2.7)

The Weyl algebra W, generated by Ŵ (F ) is thus spanned just by linear combinations of

π(Φ̂(F )) = ~[Â(F ) + Â† (F )] , (2.11)

B. Klein-Gordon fields in FLRW cosmologies

for K > 0; and, (3) the completeness relation

ek (η) ∂η ēk (η) − ēk (η) ∂η ek (η) = i . (2.25)

(JF )(η, ~x) := i F + (η, ~x) − i F − (η, ~x) (2.28)

structure and evaluate it on a Cauchy surface Σ given by η = η0 6= 0. Then, using the

Since this expression is independent of the choice of η0 we can extend it by continuity

III. RADIATION-FILLED, CLOSED AND OPEN FLRW UNIVERSES

A. Extension across the big-bang

We will now apply this general strategy to radiation-filled universes. It is convenient to

K = 1/L2 if K > 0 and K = −1/L2 if K < 0 ; (3.1)

the limit K → 0 corresponds to L → ∞ in both cases. For radiation-filled universes, the

φ̂(x) = φ̂◦ (x)/a(η), the commutator is given by

Eqs. (3.9) and (3.10) have two interesting properties:

IV. DUST-FILLED, CLOSED AND OPEN FLRW UNIVERSES

A. Extension across the big-Bang

As in Sec. III A, it is convenient to introduce a variable L with dimensions of length,

a(η) = (−2a2 L2 ) (cos(η/L) − 1), for K > 0 and

the normalization condition (2.25), and tend to the basis

B. Observables: the K < 0 case

hpiren ) = 0. In contrast to the radiation-filled universe, these expressions exhibit logarithmic

where 3 F2 is the hypergeometric function. Using Hadamard renormalization we obtain

Appendix A: Hadamard renormalization

ω(x) = Ŵ (x, ǫ) (A21)

ωηη (x) = a2 ua ub ∇a ∇b Ŵ (x, ǫ) = ∂η2 − H∂η Ŵ (x, ǫ)

Appendix B: Radial component of the Klein-Gordon equation

1. Main solutions and properties

With the ansatz (2.19) the spatial equation (2.20) reduces to

2. Orthogonality and normalization

where xK ≡ πL for K = 1/L2 and xK ≡ +∞ for K = −1/L2 . To alleviate the notation we

The last integral is easy to evaluate:

Using (B11) with (B16) and (B28) we can write

You might also like