Anisotropic Hubble expansion of large scale structures

Abstract

We investigate the dynamics of an homogenous distribution of galaxies moving under the cosmological expansion through Euler-Poisson equations system. The solutions are interpreted with the aim of understanding the cosmic velocity fields in the Local Super Cluster, and in particular the presence of a bulk flow. Among several solutions, we shows a planar kinematics with constant (eternal) and rotational distortion, the velocity field is not potential.

Full text

Page 1

arXiv:gr-qc/0505095v4 14 Oct 2005
Anisotropic Hubble expansion of large scale
structures
H.H. Fliche†, J.M. Souriau‡ and R. Triay‡§
† LMMT (UPRES EA 2596) Fac. des Sciences et Techniques de St J´ erˆ ome, av.
Normandie-Niemen, 13397 Marseille Cedex 20, France
‡ Centre de Physique Th´ eorique?, CNRS Luminy Case 907, 13288 Marseille Cedex 9,
France
Abstract.
investigate the dynamics of a pressureless distribution of gravitational sources moving
under an anisotropic generalization of Hubble expansion and constraint by Euler-
Poisson equations system. As a result, it turns out that such a behavior requires the
distribution to be homogenous, similarly to Hubble law. Among several solutions, we
show a planar kinematics with constant (eternal) and rotational distortion, where the
velocity field is not potential. Within this class, the one with no rotational distortion
identifies to a bulk flow. To apply this model within cosmic structures as the Local
Super Cluster, the solutions are interpreted as approximations providing us with an
hint on the behavior of the cosmic flow just after decoupling era up to present date.
Such a result suggests that the observed bulk flow may not be due exclusively to tidal
forces but has a primordial origin.
With the aim of understanding the cosmic velocity fields at large scale, we
PACS numbers: 98.62.Py, 98.80., 98.80.Es
Submitted to: Class. Quantum Grav.
§ triay@cpt.univ-mrs.fr
? Unit´ e Mixte de Recherche (UMR 6207) du CNRS, et des universit´ es Aix-Marseille I, Aix-Marseille
II et du Sud Toulon-Var. Laboratoire affili´ e ` a la FRUMAM (FR 2291).

Page 2

H.H. Fliche et al.2
1. Introduction
In the past, the investigations of cosmic velocity fields from redshifts surveys were not as
successful as that for providing us with powerful tools for the estimation of cosmological
parameters and the local density fields, see e.g. [1]. Later on, least action methods
[2] were also used in reconstruction procedures with more reliable estimates, see e.g.
[3, 4, 5, 6, 7]. Both approaches assume that the peculiar velocity field decomposes
into a “divergent” component due to density fluctuations inside the surveyed volume,
and a tidal (shear) component, consisting of a bulk velocity and higher moments, due
to the matter distribution outside the surveyed volume [8].
justified by the properties that the irrotational linear perturbations dominate with time
[9, 10] together with Kelvin theorem which ensures the irrotational characteristics of
motions. However, it turns out that theses investigations might provide us with biased
results because of their inherent dependence of sampling characteristics. For example,
the presence of an high region (Great Attractor) was proposed for accounting of an
unexpected feature in the cosmic velocities [11, 12, 13, 14], when from recent surveys
Shapley Concentration seems to be a better candidate, although it is not responsible
for all of the SMAC flow [15, 16, 17]. Since such an interpretation is not as clear
cut as that, one might ask whether the presence of a bulk flow originates necessarily
from a density excess in the spatial distribution of gravitational structures or another
alternative could be envisaged. For answering this question, we investigate Euler Poisson
equations system describing an anisotropic Hubble flow of pressureless distribution of
gravitational sources in a Newtonian schema.
These hypotheses are
2. Dynamics of the cosmic expansion
The dynamics of a pressureless distribution of gravitational sources (dust) is investigated
by assuming that the motion of sources satisfy the following kinematics
? r = A? r◦,A(t◦) = 1 l(1)
where A = A(t) stands for a 3×3 matrix depending on cosmic time t which has to be
determined by an observer at rest with respect to Cosmological Background Radiation
(CMB). The present investigation limits on collisionless motions, which is ensured by a
non vanishing determinant of matrix A, and because it reads as a unit matrix at given
t = t◦one has detA > 0. Hence, the velocity field ? v = ? v(? r,t) is given by
? v =d? r
dt=˙AA−1? r,
˙A =d
dtA (2)
where A−1stands for the inverse matrix (A−1A = AA−1= 1 l). Moreover, accordingly
to Hubble law, we assume a radial acceleration field
? g =d? v
dt=¨AA−1? r,? g ∝ ? r(3)

Page 3

H.H. Fliche et al.3
We assume now that these motions are constraint by Euler equations system
∂ρ
∂ t= − div(ρ? v)
∂? v
∂ t= −∂? v
where ρ = ρ(? r,t) stands for the density. By using the trace, the determinant and the
equality
detA, eq.(4) transforms
(4)
∂? r? v +? g(5)
d
dtdetA = Tr
˙ ρ =∂ρ
?˙AA−1?
∂ t+? v·
−→
grad ρ
= − ρ div
?˙AA−1? r
?
= −ρ tr
?˙AA−1?
= −3ρ˙ a
a
(6)
where the doted variables stand for time derivatives, and
3√detA,
a(t) =
a◦= a(t◦) = 1(7)
for the (generalized) expansion factor. Hence, with eq.(1) one obtains
ρ(? r,t) =ρ(A−1? r,t◦)
a3
which simply accounts for the mass conservation ρ◦dr3
One has a unique matrix decomposition
˙AA−1= H1 l +H◦
a2B,
where H = H(t) acts as the usual Hubble factor and B = B(t) stands for a traceless
matrix herein called distortion matrix. It characterizes a deviation from isotropy of the
(dimensionless) velocity field, its amplitude is defined by the matrix norm
=ρ◦
a3,ρ◦= ρ◦(? r◦)(8)
◦= ρdr3= ρdetAdr3
◦.
H =˙ a
a,
H◦= H(t◦),tr B = 0(9)
?B? =
?
tr(BtB)(10)
where the sign ”t” stands for the matrix transposition. For convenience, let us write
βn(t) = tr(Bn),n = 1,2,3(11)
According to eq.(2,5,9), one has
? g =
??˙H + H2?
1 l +H◦
a2
?
˙B +H◦
a2B2
??
? r(12)
We assume that the gravitational field satisfies Poisson-Newton equations
div? g = − 4πGρ + Λ
−→
rot ? g =?0
(13)
(14)
where G is Newton constant of gravitation and Λ the cosmological constant. According
to eq.(12), the left hand term of eq.(13) reads
div? g = tr
?∂? g
∂? r
?
= 3
?˙H + H2?
+H2
◦
a4β2
(15)
which does not depend on spatial coordinates. Hence, eq.(13) tells us that the space
distribution of sources is homogenous, i.e.
ρ = ρ(t)(16)

Page 4

H.H. Fliche et al.4
With eq.(13,15) one has
1
H2
◦
?˙H + H2?
=
1
H2
◦
¨ a
a= −Ω◦
2a3+ λ◦−
1
3a4β2
(17)
where
Ω◦=8πG
3H2
◦
ρ◦,λ◦=
Λ
3H2
◦
,H◦= H(t◦)(18)
are motion parameters. By multiplying each term of eq.(17) by 2˙ aa one easily identifies
the following constant of motion
˙ a2
H2
◦
3
1
κ◦=Ω◦
a
+ λ◦a2−−2
?a
β2
a3da = Ω◦+ λ◦− 1(19)
Hence, the chronology is given by
dt =
1
H◦
ada
?
P(a)
(20)
where H◦> 0 accounts for an expansion (according to observations, the case H◦< 0
which accounts for a collapse is not envisaged), and
P(a) = λ◦a4− κ◦a2+ Ω◦a −2
The constraint given by eq.(14) can be written in matrix form as follows
3a2
?a
1
β2
a3da ≥ 0,P(1) = 1(21)
∂? g
∂? r−
?∂? g
∂? r
?t
= 0(22)
Such a symmetric property with eq.(12,17) shows that the matrix
ˆB = B2−1
3β21 l +a2
H◦
˙B (23)
is traceless and symmetric
trˆB = 0,
ˆBt=ˆB(24)
According to eq.(3,12,23,24), since the field ? g, which reads
? g
H2
◦
=
?
λ◦−Ω◦
2a3+1
a4ˆB
?? r
a2
(25)
is radial, the matrixˆB must be scalar and because it is traceless, one has necessarily
ˆB = 0(26)
2.1. Reference map
Instead of (t,? r), it is more convenient to analyze the dynamics of the cosmic flow in the
(τ,? q) coordinates defined by
dτ = H◦dt
a2
(27)
? q=? r
a
(28)

Page 5

H.H. Fliche et al.5
herein called reference map. According to eq.(2,20,23,26), the equations of motion read
d? q
dτ= B? q,
d2? q
dτ2=
a2
3H◦β2? q(29)
dτ =
da
a
?
P(a)
(30)
where the distortion matrix B satisfies
dB
dτ
The resolution of these equations can be performed by mean of numerical techniques;
having solved eq.(31), which gives the evolution with time of distortion matrix B, the
particles trajectories τ ?→ ? q(τ) are obtained by integrating eq.(29) and the evolution of
the generalized expansion factor a from eq.(30).
=1
3β21 l − B2
(31)
2.2. Discussion
As a result, it is interesting to mention that (as it is the case for Hubble law) this
anisotropic generalization accounts for homogeneous space distributions of matter.
Hence, one is forced to ask whether it describes correctly the dynamics of cosmic
structures because of the presence of strong density inhomogeneities in the space
distribution of galaxies catalogs. In principle, such a remark should be also sensible
to question Hubble law when, regardless the isotropy, it is a fact that perturbations are
not so dominant otherwise it would never have been highlighted. Actually, homogeneity
is implicitly assumed for the interpretation of CMB isotropy and the redshift of distant
sources, which provides us with an expanding background.
space of FL world model onto which the gravitational instability theory is applied for
understanding the formation of cosmic structures. It is with such a schema in mind that
this anisotropic Hubble law may provides us with an hint on the behavior of the cosmic
flow from decoupling era up to present date in order to answer whether the observed
bulk flow is due exclusively to tidal forces.
Namely the comoving
3. Analysis of analytic solutions
In this section, we investigate some analytic solutions of eq.(29,30,31) that are obtained
thanks to particular properties of distortion matrix B. The parameters λ◦, Ω◦ and
H◦given in eq.(18) correspond to cosmological Friedmann-Lemaˆ ıtre (FL) world model
parameters. Moreover, the constraint β2 = 0 in eq.(20,21) provides us with the FL
chronology, where κ◦ given in eq.(19) represents the curvature parameter in the FL
model (i.e. the dimensionless scalar curvature Ωkof the comoving space, see [18]), while
the flatness of (simultaneous events) Newton space. It must be noted that the particle
position ? q as defined in eq.(28) does not identify to the usual FL comoving coordinate
because the (generalized) expansion factor a depends on the anisotropy unless β2= 0.

Page 6

H.H. Fliche et al.6
3.1. Evolution of functions βn=2,3
Because B is a traceless matrix, its characteristic polynomial reads
Q(s) = det(s1 l − B) = s3−1
according to Leverrier-Souriau’s algorithm [20]. With Cayley-Hamilton’s theorem (i.e.
Q(B) = 0) and eq.(31) we obtain the following differential equations system
2β2s −1
3β3
(32)
d
dτβ2= − 2β3
d
dτβ3= −1
(33)
2β2
2
(34)
and we note that the discriminant of third order polynomial Q, it is proportional to
α = 3β2
3−1
2β3
2
(35)
is a constant of motion (i.e., dα/dτ = 0). The integration of eq.(33,34) is performed by
defining β2by a quadrature
√6
2
β2(τ◦)
τ = τ◦+ ǫ
?β2(τ)
dx
√2α + x3,ǫ = ±1(36)
and hence β3from eq.(33); in addition of the singular solution
β2= β3= 0,(i.e.,B3= 0) (37)
defined equivalently either by β2= 0 or β3= 0, according to eq.(33,34).
The related dynamics depends on roots ηi=1,2,3of characteristic polynomial Q given
in eq.(32), i.e. the eigenvalues of distortion matrix B. Their real values identify to
dilatation rates at time τ toward the corresponding (time dependent) eigenvectors (not
necessarily orthogonal). Their sum is null (β1 = 0) and their product (β3 = 3detB)
is either decreasing with time or is null, according to eq.(34). The sign of α given in
eq.(35) is used to classify the solutions as follows :
• if α = 0 then Q has a real double root η1= η2and a simple one η3. The related
instantaneous kinematic shows a planar-axial symmetry (either a contraction within
a plane with an expansion toward a transverse direction or vice versa), see sec. 3.3.
If η1= η3then both vanish and the related solution identifies to the singular one
defined in eq.(37), see sec.3.2;
• if α > 0 then Q has a single real root η1;
• if α < 0 then Q has three distinct real roots ηi=1,2,3. Their order is conserved during
the evolution (since a coincidence of eigenvalues makes α = 0), the largest one must
be positive while the smallest one must be negative (because β1= 0).

Page 7

H.H. Fliche et al.7
3.2. Planar kinematics
The singular solution B3= 0 shows a FL chronology and the distortion matrix
B = −B2
◦τ + B◦,B3
◦= 0(38)
is solely defined by its initial value B◦, according to eq.(31). It is neither symmetric
nor asymmetric (otherwise it vanishes), see eq.(10). Hence, eq.(29) transforms
d? q
dτ=
?
−B2
◦τ + B◦
?
? q(39)
which accounts for eternal motions
? q = exp
?
−B2
◦
τ2
2+ B◦τ
?
? q◦= (1 + B◦τ)? q◦
(40)
The trajectory of a particle located at initial position ? q◦ identifies to a straight line
toward the direction B◦? q◦. The analysis of B◦range (i.e., its image) provides us with
characteristics of trajectories flow. The nilpotent property of B◦shows that its kernel is
not empty Ker(B◦) ?= ∅. Its dimension dim(Ker(B◦)) = m characterizes the kinematics,
which is either planar (m = 1) or directional (m = 2), i.e. a bulk flow. Conversely, if
the kernel of distorsion matrix B is not empty then β3= 3det(B) = 0, and thus β2= 0,
see to eq.(33,34). Therefore, all planar kinematics can be described by such a model.
3.3. Planar-Axial kinematics
If β2?= 0 then the chronology differentiates from FL one. Let us focus on the α = 0 with
two distinct eigenvalues η1?= η3class of solutions. With eq.(36), eq.(33,34) integrate
6
(τ − τ⋆)n,
which shows a singularity at date τ = τ⋆ > 0 that splits the motion in two regimes
τ < τ⋆and τ > τ⋆. The complete investigation of this singularity problem demands to
solve an integro-differential equation, see eq.(20,21). The roots of Q read
1
(τ⋆− τ),
where η1stands for the double root. Among others, two class of solutions are defined
by mean of a constant (time independent) matrix P, the projector associated to η1,
see [20],
βn=
(n = 2,3),τ⋆= τ◦+
√6β−1/2
2
(τ◦)(41)
η1= η3= −2η1
(42)
P2= P,trP = 2(43)
They describe distinct kinematics depending on whether matrix B is diagonalizable.

Page 8

H.H. Fliche et al.8
3.3.1. (Irrotational) motionsIf B is diagonalizable then
B = η1(3P − 21 l)(44)
From eq.(29,44), one has
d? q
dτ= η1(3P − 21 l)? q(45)
and the solution reads
? q = −η1P?ξ +1
η2
1
(1 l − P)?ξ
(46)
where?ξ is constant. If the eigenvectors are orthogonal then the kinematics accounts for
irrotational motions.
3.3.2. Rotational motionsIf B is not diagonalizable then
B = η1(3P − 21 l) +1
η2
d? q
dτ=
1
N(47)
?
η1(3P − 21 l) +1
η2
1
N
?
? q(48)
where N is a constant nilpotent matrix, which accounts for rotational motions on the
eigenplane of P. The solution reads
? q = −η1P?ξ +1
η2
1
η2
where?ξ is constant.
(1 l − P)?ξ +1
1
N?ξ
(49)
4. Application to flat large scale structures (B3= 0)
The B3= 0 class of solutions has interesting properties with regard to the stability of
large scale structures that show a flat spatial distribution. To answer the question of
whether observations define unambiguously the kinematics, the distortion matrix B is
decomposed as follows
B = S + j(? ω), trS = 0(50)
where S and j(? ω) stand for its symmetric and its asymmetric¶ component, and
? ω =a2
H◦? σ,
2
accounts for the motion rotation, ? σ being the swirl vector. Hence, eq.(38) gives
? σ =1
−→
rot ? v(51)
B? ω = S? ω(52)
The evolution of the anisotropy with time is defined by
S= −
?
S2
◦+ j(? ω◦)j(? ω◦)
?
τ + S◦
(53)
j(? ω) = − (S◦j(? ω◦) + j(? ω◦)S◦)τ + j(? ω◦)
¶ The operator j stands for the vector product, ? u × ? ω = j(? u)(? ω).
(54)

Page 9

H.H. Fliche et al.9
which couples the symmetric and the antisymmetric parts of the distortion matrix. The
swirl magnitude reads
ω =
?
?? ω,? ω? =
?
1
2trS2
(55)
according to eq.(51), since β2= 0. Its orientation cannot be determined from the data
because the above equations describe two distinct kinematics corresponding to ±? ω◦that
cannot be disentangle. According to eq.(50), if (and only if) the rotation ω = 0 then
the distortion vanishes S = 0 since B is either a symmetric or antisymmetric. In other
words, a planar distortion has necessarily to account for a rotation.
4.1. Constant distortion
Among above solutions which show planar kinematics, let us investigate the (simplest)
one defined by B2
linear calculus shows that the distortion is constant
◦= 0. In such a case,?k◦∝ ? ω and S? ω =?0. According to eq.(53,54,55),
S = S◦,ω = ω◦
(56)
Such a distortion in the Hubble flows produces a rotating planar velocities field with
magnitude ∝ H◦a−2. In the present case, the model parameters can be easily evaluated
from data. The observed cosmic velocity fields are partially determined by their radial
component
vr= ?? v,? r
where m, z, ? u, t stand respectively for the apparent magnitude, the redshift, the line of
sight, the photon emission date of the galaxy and c the speed of the light. According to
eq.(2,9,50), the radial velocity of a galaxy located at position ? r is given by
r? = cz,? r = ? r(m) = r? u = a? q, r = ct (57)
vr=
?
H +˜H? u
?
r,
˜H? u=H◦
a2? u · S? u (58)
Because trS = 0, it is clear that the sum of three radial velocities vrcorresponding to
galaxies located in the sky toward orthogonal directions and at same distance r provides
us with the quantity H. Hence, simple algebra shows that the sample average of radial
velocities within a sphere a radius r is equal to
?vr? = H?? r?
(59)
Therefore, for motions described by eq.(1), the statistics given in eq.(59) provides us
with a genuine interpretation of Hubble parameter H = H(t). Hence, according to
eq.(9,20,21), one obtains the (generalized) expansion factor
a(t) = exp
?t
t◦H(t)dt(60)
Hence, the cosmological parameters can be estimated by fitting the data to the function
ψ(t)λ◦,κ◦,Ω◦=
?
P(a) = a2H/H◦,λ◦+ κ◦+ Ω◦= 1(61)

Page 10

H.H. Fliche et al.10
The component of matrix S can be estimated by substituting H in eq.(58), and ω is
obtained from eq.(56).
It is clear that the above model is derived in the Galilean reference frame, where
the Euler-Poisson equations system can be applied. Hence, a non vanishing velocity of
the observer with respect to this frame an produces a bipolar harmonic signal in the˜H? u
distribution of data in the sky, which can be (identified and then) subtracted.
4.2. Discussion
The dynamics of a homogenous medium and anisotropic moving under Newton gravity
was already studied by describing the evolution of an ellipsoid [21]. The current approach
enables us to identify characteristics of the dynamics of the deformation from isotropic
Hubble law in a more systematic way by mean of the distorsion matrix.
At first glance, if the planar anisotropic of the space distribution of galaxies within
the Local Super Cluster (LSC) is stable then the above solution can be used for
understanding its cosmic velocity fields,?k◦being orthogonal to LSC plane. It is well
known however that the distribution of galaxies is not so homogenous as that, whereas
this model describes motions of an homogenous distribution of gravitational sources.
However, such an approximation level is similar to the one which provides us with the
observed Hubble law, that is included in this model (S = 0).
5. Conclusion
The present anisotropic solution of Euler Poisson equations system generalizes the
Hubble law and provides us with a better understanding of cosmic velocity fields
within large scale structures as long as Newton approximation is valid. As a result,
this generalization of Hubble motion implies necessarily an homogenous distribution of
gravitational sources, as similarly to Hubble law. Because the chronology identifies to FL
chronology for a vanishing distortion, this model interprets as a Newton approximation
of anisotropic cosmological solutions. The motions are characterized by means of a
constant of motion α.Among them, particular solutions can be easily derived for
α = 0. They describe all planar distortions, in addition of two classes of planar-axial
distortions with or without rotation. Among these solutions, the one which ensures a
planar kinematics is of particular interest because it describes constant (eternal) and
rotational distortions. This solution can be fully determined from observational data
except for the orientation of the rotation. The sensible result is that the velocity field
is not potential. It is interesting to note that this model accounts for motions which
might be interpreted as due to tidal forces whereas the density is homogeneous. It is an
alternative to models which assume the presence of gravitational structures similar to
Great Attractor as origin of a bulk flow.

Page 11

H.H. Fliche et al. 11
References
[1] M.A. Strauss in Cosmic Flows Workshop, S. Courteau & J. Willick ed., ASP Conference Series,
201, 3 (2000)
[2] P.J.E. Peebles, Astrophys. J. 344, L53 (1989)
[3] M. Susperregi, J. Binney, Monthly Not. Roy. Astron. Soc. 271, 719 (1994)
[4] I.M. Schmoldt, P. Saha, Astrophys. J. 115, 2231 (1998)
[5] M. Susperregi, Astrophys. J. 563, 473 (2001)
[6] D.M. Goldberg, Astrophys. J. 522, 413 (2001)
[7] D.M. Goldberg, D.N. Spergel, Astrophys. J. 544, 21 (2000)
[8] S. Courteau, A. Dekel, in Astrophysical Ages and Times Scales, T. von Hippel, C. Simpson, and
N. Manset ed., ASP Conference Series Vol. 245, 584 (2001)
[9] E.M. Lifshitz, I.M. Khalatnikov, Adv. Phys. 12,185 (Usp. Fiz. Nauk.,80,391) (1963)
[10] H.H. Fliche, in´Evaluation des paramtres cosmologiques ` a l’aide des proprit´ es optiques des quasars.
Fluctuations des mod` eles de Friedmann-Lemaˆ ıtre, th` ese de doctorat d’´ etat, Universit´ e de
Provence, C.N.R.S. - C.P.T. 81/ P. 1282, Marseille (1981)
[11] D. Lynden-Bell, S.M. Faber, D. Burstein, R.L. Davies, A. Dressler, R.J. Terlevich, G. Wegner,
Astrophys. J. 326, 19 (1988)
[12] E. Bertschinger, A. Dekel, Astrophys. J. 336, L5 (1989)
[13] A. Dekel, E. Bertschinger, S.M. Faber, Astrophys. J. 364, 349 (1990)
[14] A. Dekel, A. Eldar, T. Kolatt, A. Yahil, J.A. Willick, S.M. Faber, S. Courteau, D. Burstein,
Astrophys. J. 522, 1 (1999)
[15] E. Branchini, L. Teodoro, C. S. Frenk, I. Schmoldt, G. Efstathiou, S. D. M. White, W. Saunders,
W. Sutherland, M. Rowan-Robinson6 O. Keeble, H. Tadros, S. Maddox and S. Oliver, Monthly
Not. Roy. Astron. Soc. 308, 1 (1999)
[16] W. Saunders, K. DMellow, H. Valentine, B. Tully, E. Carrasco, B. Mobasher, S. Maddox, G. Hau,
W. Sutherland, D. Clements, L. Staveley-Smith, in Mapping the Hidden Universe: The Universe
behind the Milky Way - The Universe in HI. ASP Conference Proceedings R.C. Kraan-Korteweg,
P.A. Henning, and H. Andernach ed. Astronomical Society of the Pacific, 218,141 (2000)
[17] M.J. Hudson, R.J. Smith, J.R. Lucey, E. Branchini, Monthly Not. Roy. Astron. Soc. 352, 61 (2004)
[18] R. Triay, L. Spinelli, R. Lafaye, Monthly Not. Roy. Astron. Soc. 279, 564 (1996)
[19] R. Triay, V. Villalba, Gen. Relat. & Gravit. 31, 1913 (1999)
[20] J.M. Souriau, in Calcul lin´ eaire (edit. Jacques Gabay) (1992)
[21] Ya. Zel’dovich, I.D. Novikov, in The Structure and Evolution of the Universe vol. 2 (edit. G.
Steigman, The University of Chicago Press) p480 (1983)