Physics Letters B 545 (2002) 23–29

www.elsevier.com/locate/npe

A phantom menace?
Cosmological consequences of a dark energy component with
super-negative equation of state
R.R. Caldwell
Department of Physics & Astronomy, Dartmouth College, Hanover, NH 03755, USA
Received 29 April 2002; received in revised form 24 August 2002; accepted 24 August 2002
Editor: J. Frieman

Abstract
It is extraordinary that a number of observations indicate that we live in a spatially flat, low matter density Universe, which
is currently undergoing a period of accelerating expansion. The effort to explain this current state has focused attention on
cosmological models in which the dominant component of the cosmic energy density has negative pressure, with an equation of
state w
−1. Remarking that most observations are consistent with models right up to the w = −1 or cosmological constant (Λ)
limit, it is natural to ask what lies on the other side, at w < −1. In this regard, we construct a toy model of a “phantom” energy
component which possesses an equation of state w < −1. Such a component is found to be compatible with most classical
tests of cosmology based on current data, including the recent type 1a SNe data as well as the cosmic microwave background
anisotropy and mass power spectrum. If the future observations continue to allow w < −1, then barring unanticipated systematic
effects, the dominant component of the cosmic energy density may be stranger than anything expected.
 2002 Published by Elsevier Science B.V.

Arguments have been put forward that we live in vincing theory has yet been constructed to explain this
a spatially flat, low matter density Universe which is state of affairs, although cosmological models based
currently undergoing a period of accelerating expan- on a dark energy component, such as the cosmological
sion. If the observational evidence upon which these constant (Λ) or quintessence (Q), are leading candi-
claims are based are reinforced and strengthened by dates. At this stage we are lead to notice that the pa-
future experiments, the implications for cosmology rameterization of the dominant energy component as
will be incredible. It would then appear that the cos- a fluid with an equation of state (defined as the ratio
mological fluid is dominated by some sort of fantastic of pressure to energy density) w(≡ p/ρ)
−1 leads
energy density, which has negative pressure, and has to the curious situation that most observational con-
just begun to play an important role today. No con- straints are consistent with models that go right up to
the w = −1 border [1]. It is natural to ask what lies on
the other side of this boundary. The focus of this Let-
ter is the investigation of cosmological models with a
E-mail address: [email protected]
(R.R. Caldwell). dominant component for which w < −1.

0370-2693/02/$ – see front matter  2002 Published by Elsevier Science B.V.

PII: S 0 3 7 0 - 2 6 9 3 ( 0 2 ) 0 2 5 8 9 - 3
24 R.R. Caldwell / Physics Letters B 545 (2002) 23–29

We begin by constructing a classical cosmology:

a spatially flat FRW space–time filled by a cold
dark matter (CDM) component and a “phantom” (P )
energy component. (A phantom is something which
is apparent to the sight or other senses but has no
corporeal existence—an appropriate description for a
form of energy necessarily described by unorthodox
physics, as will be seen later.) The phantom energy has
positive energy density, ρp > 0, but negative pressure,
such that ρp + pp < 0. (It is no fatal flaw for a
component to violate the dominant energy condition
for a finite time, as can arise from a bulk viscous stress Fig. 1. The age in units of the Hubble time is plotted vs. Ωm for a
due to particle production [2].) It immediately follows series of cosmological models containing dark energy with different
values of w.
from the equation of state w < −1 that the phantom
energy density grows with time. If this energy has
begun to dominate today, it must have come on like a
bolt from the blue. Taking the constant value w = −3,
as an example, then ρp ∝ a 6 where a is the expansion
scale factor. So if Ωm ∼ 0.3 today, then the Universe
contained 90% CDM at z ∼ 0.4 as opposed to z ∼ 1.8
in a Λ-dominated cosmology. Hence, the phantom
energy exerts its influence very late. We now turn to
quantify the remarkable effect on the cosmology of
a component with super-negative equation of state,
w < −1.
The expansion scale factor grows rapidly when the Fig. 2. The volume–red shift relationship is shown for phantom
phantom energy comes to dominate the Universe, as energy models with w = −3, −3/2, ΛCDM with w = −1, QCDM
ȧ/a actually grows and the deceleration parameter with w = −1/2, and CDM. All the dark energy models have
q0 = (1 + 3wΩp )/2 becomes very negative. In fact, Ωm = 0.3.
the scale factor diverges in finite time: if the expansion
is matter dominated until the time tm , then we can fact, for Ωm = 0.3, the bound on the age in units of
write the phantom energy-driven scale factor as the Hubble time is H0 t0  1.2.
The volume–red shift relationship is demonstrated

2/3(1+w)
a(t) = a(tm ) −w + (1 + w)t/tm at t > tm . in Fig. 2. For the same matter density, the phantom
energy model gives the largest differential number of
For w = −3 then a diverges when t = 3tm /2. Unless objects per red shift interval. Although evolutionary
Ωm  1, however, t0 < tm w/(1 + w) and so the lim- effects are important in cosmological tests based
iting cosmological time occurs well after the present on this relationship, if all other features are held
day. fixed, the phantom model will predict more strong
The expansion age, and similarly the horizon dis- gravitationally lensed quasars than Q or Λ models.
tance in a phantom energy cosmology are larger than The magnitude–red shift relationship is demon-
in the analogous Λ model. In Fig. 1 we show the age strated in Fig. 3. The predictions for several cosmolog-
for a sequence of w  −1 models. In the range of in- ical models have been shown along with a summary
terest, near Ωm ∼ 0.3–0.4, we see that the phantom of the observational results. All other parameters be-
energy can increase the age by up to ∼ 30% over the ing equal, one expects high red shift supernovae to be
w = −1 age. For a given value of Ωm , as w becomes dimmer in a phantom energy Universe.
very negative the phantom energy becomes important Considering the constraint to the allowed range of
later and later with diminishing effect on the age. In w and Ωm due to the supernovae, we find that phantom
 R.R. Caldwell / Physics Letters B 545 (2002) 23–29 25

Fig. 3. The magnitude–red shift relationship is shown for the

red shift-binned type 1a SNe data (HZT—[17]; SCP—[18]; bin-
ning—[19]), alongside the predictions of various cosmological
models. Both the phantoms (w = −3/2, Ωm = 0.4; −3/2, 0.3) and
the Λ model (−1, 0.3) provide good fits to the data (low χ 2 /d.o.f.).
The magnitudes are calculated relative to an empty, open Universe.
The light, dashed lines are for pure phantom, de Sitter, Milne, and
Einstein–de Sitter, from top to bottom.

energy models even with very negative w are in accord

with the observations. We see in Fig. 4 that the contour
region is extensive for w < −1, preferring a slightly
higher matter density than in Quintessence models. An
alternative way to visualize this constraint on phantom Fig. 4. The constraint on the allowed values of Ωm in phantom and
models is to construct a parameter space in terms of quintessence dark energy models is shown as a function of w, or
the red shift at which the energy density drops to alternatively the red shift at which matter-domination ends, when
Ωm = 0.9. We have traversed the w = −1 boundary, finding that
90% matter, signaling the end of the matter-dominated
there are phantom energy models in accord with the observations.
epoch, rather than the constant equation of state w.
The bottom panel above makes clear that an important
difference with other dark energy models is the very
late onset of the phantom energy. (with metric signature + − − −). The important point
The next challenge is to construct a microphysical is that we have switched the sign of the kinetic term
model of the phantom energy in order to consistently in the scalar field Lagrangian Lp . In this way, ρp =
determine the impact on the cold dark matter, baryon, −φ̇ 2 /2 + V and pp = −φ̇ 2 /2 − V so that w  −1
and photon fluctuation spectra. is attained. The equations of motion are φ̈ + 3H φ̇ =
We describe the system of gravitation, normal +V,φ , so that the field will tend to run up, not down,
matter, and phantom energy with the Lagrangian a potential towards larger energy. The key ingredient
for this crude toy model, and the more detailed model
L = −R/16πG + Lm + Lp , we introduce later, is the non-canonical kinetic energy
term, examples of which occur in supergravities [3]
where Lm represents the Lagrangian for the “normal”
and in higher derivative theories of gravity [4].
matter, CDM plus the standard model particles and
The cosmological spectrum of fluctuations in the
fields. Since it is not possible to achieve w < −1 with
phantom field develop in a fashion similar to the case
a free scalar field, and we wish to avoid interactions
of Quintessence [5]. The Fourier transform of the
with other fields in order to keep the energy “dark”,
linearized perturbation equation is
we start with the unorthodox Lagrangian
 
Lp = −∂µ φ∂ µ φ/2 − V (φ) ¨ + 3H δφ
δφ ˙ + k 2 − V,φφ δφ = −ḣφ̇/2,
26 R.R. Caldwell / Physics Letters B 545 (2002) 23–29

where h is the synchronous gauge metric perturbation. began to dominate and the present. For wp < −1, this
In this equation, the sign of V,φφ is different than in the is a narrow range in red shift, leaving little opportu-
standard case. Just as for Quintessence, perturbations nity for the effects of time-variation to be distinct. As
on small scales, for k 2 |V φφ |, are suppressed. On we have modified CMB fast [11] to include a phantom
larger scales, however, the phantom energy develops dark energy component based on both approaches,
inhomogeneities in response to the surrounding matter we now proceed to discuss the resulting perturbation
and density perturbations. If the effective mass, (k 2 − spectra.
V,φφ )1/2 , should become imaginary then δφ would The main feature that distinguishes the phantom
develop a growing (tachyon) instability. However, if energy case from Λ or Q is that the onset of phantom
we implement a constant equation of state, then energy dominance happens at the very last moment—
  so late that most evolutionary effects which occur in Λ
3 3 2
V,φφ = (1 − w) Ḣ − H (1 + w) , and Q models are suppressed. In the cases illustrated
2 2 in Fig. 5 the strength of the late-time ISW effect
which is negative for a limited range of values of w diminishes as w becomes more and more negative,
and Ωm . Consequently, the effective mass is real, and because the expansion is CDM-dominated until later
there is no instability for constant w. and later. On small angular scales the locations of the
A more versatile phantom model can be con- acoustic peaks shift to higher multipole moments as w
structed using a Lagrangian with a non-standard de- becomes more negative, due to the increased distance
pendence on the field gradients, to the last scattering surface. This may be an important
factor in attempts to develop a phantom dark energy
Lp = a(φ)(∇φ)2 + b(φ)(∇φ)4 + · · · ,
cosmological model to accommodate the newer CMB
motivated by the appearance of higher derivatives in results, also represented in Fig. 5. A more detailed
string theory. This approach has been employed in analysis will be forthcoming.
the k-inflation and k-essence scenarios [6–8], and it The phantom-dominated background evolution has
has been demonstrated [6,9] that such models can an important effect on the fluctuation spectra, as well.
also be used to describe an energy component with a Making the assumptions that the matter component
super-negative equation of state, wp < −1, which is of the cosmological fluid carries a spectrum of scale-
also stable to perturbations (no tachyonic instability). invariant perturbations generated by inflation, and
Details of the treatment of cosmological perturbations that, similar to Quintessence, the phantom energy
in the k-essence scenario are given in [10]; the itself does not fluctuate on scales well below the
approach is sufficiently general that the case of a Hubble horizon, then we may follow the growth of
phantom is simply accommodated. Hence, starting linear perturbations in the CDM and baryons. The
from a theory giving the functional dependence of growth suppression factor g = D(a)/a is shown in
Lp on φ and (∇φ)2 , one can readily determine Fig. 6, where we see that perturbations grow as
the stress-energy ρ(z), p(z) and the sound speed of D ∝ a until very late, owing to the very late time at
perturbations cp (z) as they evolve in time or red shift. which the phantom energy begins to dominate. Hence,
Turning this around, one may instead dictate the the evolution is very similar to the standard CDM
histories wp (z) and cp (z) in order to specify the phan- scenario, but with a lower matter density.
tom dark energy model. In the present investigation we Next, we consider the behavior of the mass power
focus on constant-wp and cp = 1 models. Although spectrum for w < −1. Because the phantom field
this is not a generic prediction of the dynamical mod- does not fluctuate on scales well below the hori-
els described above, the observational differences are zon, the overall shape of the mass power spectrum
not expected to be very large (as some exploratory is well-described by the BBKS parameterization [12].
analyses have confirmed). Since the observational ef- (See [13] for instructions on how to adapt the fitting
fects rely on a substantial fractional phantom energy functions to quintessence.) The normalization is ob-
density, and since ρ(z) is actually growing with time, tained by comparison with the CMB. Using the am-
then the relevant values of wp (z) and cp (z) are the plitude of the first acoustic peak, then the normal-
mean values between the red shift when the phantom ization is similar to a ΛCDM model with the same
 R.R. Caldwell / Physics Letters B 545 (2002) 23–29 27

Fig. 5. The CMB anisotropy spectra are shown for various cosmological models with equal input power, to demonstrate the effect of w < −1.
On large angular scales, the strength of the late-time ISW effect diminishes as w becomes more and more negative, because the expansion
is CDM-dominated until later and later. The locations of the acoustic peaks shift to smaller angular scales as w becomes more negative, due
to the increased distance to the last scattering surface. The models shown all have Ωm = 0.35, Ωb h2 = 0.02, and h = 0.70. The horizontal
axis is logarithmic for 2 < l < 200 in order to give enough space to both large and small angular scale features. Comparing the curves with
a compilation of current CMB data [20] suggest a more negative dark energy equation-of-state may allow for a better fit to the small angular
scale data.

tuations in the phantom field must also be taken into

account. However, owing to the late onset of the phan-
tom, there is very little late ISW and the direct fluc-
tuations are also negligible. As mentioned earlier, the
low-' spectrum is very flat, as in SCDM, and up to
∼ 5% lower than in the comparable ΛCDM model.
Hence, the COBE normalized spectrum will result in
a power spectrum with a σ8 up to ∼ 20 + 5% higher
than for Λ.
In Fig. 7 we show the prediction and observational
constraint on σ8 for a sequence of models varying
Fig. 6. The amplitude of the growth suppression factor g = D(a)/a in w. While the amplitude of the CMB-normalized
vs. the scale factor is shown for various cosmological models.
mass power spectrum grows slightly as w becomes
We see that the perturbation growth is slowed later, with weaker
effect, in the phantom energy models. The models shown all have more negative, the implied σ8 based on the observed
Ωm = 0.3. For SCDM, D(a)/a = 1. abundance of clusters grows as well. (The cluster
abundance calculation is based on an extension of [14]
to the case of w < −1.) We see that the intersection
cosmological parameters, but with a σ8 higher by a of the 2σ regions of σ8 grows with decreasing w.
ratio of the growth suppression factors gP /gΛ . Since Further constraints based on large scale structure one
gΛ (Ωm = 0.35, a = 1) = 0.8, and gP is at most unity may consider, but which are beyond the scope of this
in the limit of w  −1, then σ8 is higher by up to Letter, include the evolution of the abundance of rich
20%. Using COBE for the normalization, on the other clusters and the growth rate of the linear fluctuation
hand, the effect of the late ISW and the direct fluc- spectrum. Our lesson here is that phantom dark energy
28 R.R. Caldwell / Physics Letters B 545 (2002) 23–29

such a strongly negative equation of state is simply

a phantom—that the apparent accelerating expansion
is due to more conventional, though unanticipated
causes (e.g., dust or evolution for the SNe). If these
systematic effects can be eliminated, and the data
continue to support w < −1 then the implications for
fundamental physics would be astounding, since w <
−1 cannot be achieved with Einstein gravity and a
canonical Lagrangian. It is premature to shift attention
towards building w < −1 models, but it is important to
be aware of the properties and implications of models
in each direction of the cosmological parameter space.
Fig. 7. The cluster abundance constraint is shown for a sequence As has been discussed elsewhere, current observa-
of Q (w > −1), Λ (w = −1), and phantom (w < −1) models
with Ωm = 0.3, h = 0.65, Ωb h2 = 0.02, n = 1. The upper shaded
tions suggest the presence of a dark energy component
region shows the 2σ range of σ8 based on the observed abundance with w  −0.5 [1], with many constraints pushing to-
of x-ray clusters, as interpreted for a range of w (based on an wards w = −1: the observations are teetering at the
extension of the work in Ref. [14]). The lower shaded region shows edge of a previously unfamiliar boundary.
the predicted σ8 for COBE-normalized models. The intersection of
the two regions grows as w becomes more negative. The letters P
and Q indicate the Phantom and Quintessence regions, to the left
and right of w = −1. Acknowledgements

models do exist which, upon first inspection, satisfy The SNe data used in Fig. 3 were obtained from
Adam Riess [16]. The cluster abundance constraint in
constraints based on the CMB and the mass power
Fig. 7 was evaluated by Limin Wang. This work was
spectrum.
In summary, we have investigated the properties supported by the US DoE grant DE-FG02-91ER40671
(at Princeton) and the NSF grant PHY-0099543 (at
of cosmological models in which the dominant en-
ergy density component today has an equation of state Dartmouth).
w < −1. We have demonstrated the impact on the cos-
mological age, the volume–red shift and magnitude–
red shift relations, the CMB, and the mass power References
spectrum, finding broad agreement with current ob-
servational constraints. Current data appears to be [1] L. Wang, R.R. Caldwell, J. Ostriker, P.J. Steinhardt, Astrophys.
consistent with a phantom, although a more careful J. 530 (2000) 17.
analysis will be necessary to quantify the observa- [2] J.D. Barrow, Nucl. Phys. B 310 (1988) 743.
[3] H.P. Nilles, Phys. Rep. 110 (1984) 1.
tional status of w < −1. [4] M.D. Pollock, Phys. Lett. B 215 (1988) 635.
While there is no single, best method for determin- [5] R.R. Caldwell, R. Dave, P.J. Steinhardt, Phys. Rev. Lett. 80
ing if the equation of state is strongly negative, combi- (1998) 1586.
nations of measurements of the type described in this [6] C. Armendariz-Picon, T. Damour, V. Mukhanov, Phys. Lett.
Letter will substantially improve our understanding of B 458 (1999) 209.
[7] C. Armendariz-Picon, V. Mukhanov, P.J. Steinhardt, Phys. Rev.
the dark energy. One of our central points is the im- Lett. 85 (2000) 4438.
portance that such analyses are open to the possibil- [8] C. Armendariz-Picon, V. Mukhanov, P.J. Steinhardt, Phys. Rev.
ity of a phantom dark energy—unjustified biases and D 63 (2001) 103510.
priors can lead to a gross misinterpretation of the ob- [9] T. Chiba, T. Okabe, M. Yamaguchi, Phys. Rev. D 62 (2000)
servational evidence. (See [15] for an analysis of the 023511.
[10] J.K. Erickson, et al., Phys. Rev. Lett. 88 (2002) 121301.
pitfalls of assuming a constant w or w
−1.) [11] U. Seljak, M. Zaldarriaga, Astrophys. J. 469 (1996) 437.
Of course, we do not want to overlook the distinct [12] J.M. Bardeen, J.R. Bond, N. Kaiser, A.S. Szalay, Astrophys.
possibility that the observational evidence in favor of J. 304 (1986) 15.
 R.R. Caldwell / Physics Letters B 545 (2002) 23–29 29

[13] C.-P. Ma, R.R. Caldwell, P. Bode, L. Wang, Astrophys. J. 521 [16] A.G. Riess, private communication (2002).
(1999) L1. [17] A.G. Riess, et al., Astronom. J. 116 (1998) 1009.
[14] L. Wang, P.J. Steinhardt, Astrophys. J. 508 (1998) 483. [18] S. Perlmutter, et al., Astrophys. J. 517 (1999) 565.
[15] I. Maor, R. Brustein, J. McMahon, P.J. Steinhardt, astro- [19] A.G. Riess, et al., Astrophys. J. 560 (2001) 49.
ph/0112526. [20] X. Wang, M. Tegmark, M. Zaldarriaga, astro-ph/0105091.