This action might not be possible to undo. Are you sure you want to continue?
WMAP7 Bounds
Levon Pogosian
1
, Amit P.S. Yadav
2
, YiFung Ng
3
and Tanmay Vachaspati
4
1
Department of Physics, Simon Fraser University,
Burnaby, BC, V5A 1S6, Canada
2
Institute for Advanced Study,
Princeton, NJ 08540, USA
3
CERCA, Physics Department,
Case Western Reserve University,
Cleveland, OH 441067079, USA
4
Physics Department, Arizona State University,
Tempe, AZ 85287, USA
Faraday rotation induced Bmodes can provide a distinctive signature of primordial magnetic ﬁelds
because of their characteristic frequency dependence and because they are only weakly damped
on small scales, allowing them to dominate Bmodes from other sources. By numerically solving
the full CMB radiative transport equations, we study the Bmode power spectrum induced by
stochastic magnetic ﬁelds that have signiﬁcant power on scales smaller than the thickness of the last
scattering surface. Constraints on the magnetic ﬁeld energy density and inertial scale are derived
from WMAP 7year data, and are stronger than the big bang nucleosynthesis (BBN) bound for a
range of parameters. Observations of the CMB polarization at smaller angular scales are crucial to
provide tighter constraints or a detection.
I. INTRODUCTION
Many early universe scenarios predict the existence of cosmological magnetic ﬁelds and several observational tech
niques are currently being employed to detect such ﬁelds [1, 2]. There are also recent claims for the detection of an
intergalactic magnetic ﬁeld [3]. With further conﬁrmation and reﬁnement, these observations can be turned into a
probe of the fundamental interactions that are necessary to generate a primordial magnetic ﬁeld and to the physics of
the early universe. A primordial magnetic ﬁeld can also have important implications for the formation of ﬁrst stars,
growth of cosmic structure and the present universe.
Of particular interest to us are magnetogenesis scenarios based on phase transitions such as the electroweak phase
transition, when Maxwellian electromagnetism ﬁrst emerged [4–11]. In this scenario particle physics uncertainties are
minimal and magnetogenesis is tightly related to the creation of matter, or “baryogenesis” – the magnitude of the
magnetic helicity density is approximately equal to the baryon number density. Since the cosmic number density of
baryons is known, the scenario enables a prediction for the magnetic helicity density that is largely independent of the
details of the electroweak model. Furthermore, the lefthanded magnetic helicity is a direct outcome of parity violation
present in the electroweak model and, like baryon number, is a remarkable largescale manifestation of a microscopic
symmetry violation.
Nonvanishing helicity has important consequences for the evolution of a magnetic ﬁeld. Although the ﬁeld is
generated on small scales, magnetic helicity allows for an “inverse cascade” where power is transferred from small
to large scales, resulting in magnetic coherence on larger scales. In the case of magnetic ﬁelds generated at the
electroweak scale, the ﬁnal coherence scale can be on the parsec to kiloparsec scales [6, 12]. While the coherence
scale is large, it is not as large as for ﬁelds that might be generated during an inﬂationary epoch. For purposes of
calculating observational signatures, magnetic ﬁelds generated at a phase transition are stochastic. An important aim
of the present work is to ﬁnd distinctive observational signatures of magnetic ﬁelds that are stochastic on subMpc
coherence scales.
Once magnetic ﬁelds have been injected into the cosmological plasma, the subsequent evolution is described by
magnetohydrodynamical (MHD) equations in an expanding spacetime. Power on very small length scales is expected
to be erased by dissipational mechanisms. Power on very large scales is also suppressed because the magnetic ﬁeld
is injected on microscopic scales and cannot extend to arbitrarily large scales. Hence the spectral distribution of the
magnetic ﬁeld is expected to decay fast on small and large scales and be peaked on some intermediate scale, which is
presumably at the parsec to kiloparsec scale. The spectral form of the magnetic ﬁeld has been investigated recently
in some detail in Refs. [13, 14] and the results are schematically depicted in Fig. 1.
On the observational front, magnetic ﬁelds within galaxies and clusters of galaxies have been studied for many
decades and their origin – cosmological versus astrophysical – remains unsettled, though a hybrid explanation is also
a
r
X
i
v
:
1
1
0
6
.
1
4
3
8
v
3
[
a
s
t
r
o

p
h
.
C
O
]
2
8
S
e
p
2
0
1
1
2
I
0
k
1
k
b
k k
diss
k
k
FIG. 1: The Fourier amplitude b(k) – as given by the square root of the power spectrum – of a stochastic magnetic ﬁeld
generated during a cosmic phase transition is expected to grow for k < kI where kI is an inertial scale, and obey some other
power law for kI < k < k
diss
, where k
diss
is a dissipative scale. The work of [13] suggests k
1
growth at small k, then k
0
behavior
until k
diss
, beyond which the amplitude falls oﬀ very quickly.
conceivable. More recently, observations of TeV gamma ray sources have been used to place lower bounds ∼ 10
−16
G
[3, 15–18], and perhaps a measurement ∼ 10
−15
G [19], on a magnetic ﬁeld in intergalactic space, ∼ 10 Mpc away
from the TeV gamma ray source. It seems likely that such an intergalactic magnetic ﬁeld, unassociated with cosmic
structure, is primordial, but an astrophysical origin, say based on the expulsion of magnetic ﬁelds from active sources,
may also be viable.
A detection of magnetic ﬁelds in the cosmic microwave background (CMB), for example due to Faraday rotation
(FR) of the CMB polarization, would unambiguously point to a cosmological origin because there are no confounding
magnetized structures at last scattering. However, an observed FR of the CMB could also be due to magnetic ﬁelds
along the line of sight, especially within the Milky Way. Hence, it is necessary to ﬁnd distinctive signatures of FR that
occurred at recombination versus that which happened more recently. As we will see, in addition to its characteristic
frequency dependence, FR induced Bmodes are only weakly damped on small angular scales (high ), which means
they are likely to dominate Bmodes from other sources.
Earlier work on FR of the CMB in cosmic magnetic ﬁelds has largely focused on the eﬀect of a uniform magnetic
ﬁeld [20, 21] and, when a stochastic magnetic ﬁeld has been considered, a “thin” LSS was often assumed [22–24]. This
approximation results in a tremendous technical simpliﬁcation but it is not clear if it is suited to study the eﬀects
of a magnetic ﬁeld with coherence scale smaller than the thickness of the LSS. In [25], the ﬁnite thickness eﬀects
of the LSS were modelled by approximating the visibility function with a Gaussian proﬁle. As we show, the thin
LSS approximation is suﬃcient for order of magnitude estimates, but can be wrong by factors of a few and in an
dependent way.
In the present paper, we study the eﬀect of a primordial magnetic ﬁeld on the CMB polarization, focussing on the
eﬀect of a stochastic ﬁeld with power on small (subMpc) scales. We calculate the Bmode correlator, C
BB
l
, induced
by a primordial magnetic ﬁeld. In the thin LSS approximation of Sec. IV, we show that C
BB
l
is directly related to
C
EE
l
multiplied by the correlation function for the FR, C
αα
l
, which we calculate in Sec. III. In Sec. V we move on
to our main calculation of C
BB
l
with a thick LSS. Here we ﬁnd that the physics of FR during recombination can
be described in terms of “window functions” through which the magnetic ﬁeld spectrum (see Eq. (65)) appears in
C
BB
l
. We have to resort to extensive numerical eﬀorts to compute the window functions. Our results are described in
Sec. VI. The window functions themselves are independent of the magnetic ﬁeld power spectrum and are shown in
Fig. 2. When the window functions are convolved with the magnetic ﬁeld power spectrum, we obtain C
BB
l
. In Figs. 3,
4 we show the results for C
BB
l
for a scaleinvariant and a blue magnetic power spectrum. We conclude in Sec. VII.
We have made our window functions publicly available at http://www.sfu.ca/∼levon/faraday.html along with a
short Fortran code for calculating C
BB
l
.
II. STOCHASTIC MAGNETIC FIELDS
A statistically homogeneous and isotropic stochastic magnetic ﬁeld is described by the twopoint correlator in
Fourier space as
¸b
i
(k)b
j
(k
)¸ = (2π)
3
δ
(3)
(k +k
)[(δ
ij
−
ˆ
k
i
ˆ
k
j
)S(k) +iε
ijl
ˆ
k
l
A(k)] , (1)
3
where S(k) and A(k), the symmetric and antisymmetric magnetic power spectra, are real functions of k = [k[.
Throughout the paper, we use Gaussian CGS units. The energy density in modes up to some value of k is given by
M
(k) =
1
8π
_
k
<k
d
3
k
(2π)
3
2S(k
) =
1
(2π)
3
_
k
0
dkk
2
S(k). (2)
We take the form of S(k) to correspond to Fig. 1
S(k) =
_
¸
¸
¸
_
¸
¸
¸
_
S
∗
_
k
k
I
_
2n−3
, 0 < k < k
I
S
∗
_
k
k
I
_
2n
−3
, k
I
< k < k
diss
0, k
diss
< k
(3)
The results in [13, 26] suggest the exponents n = 5/2 and n
= 3/2. In our analysis, along with these values, we
will also consider the case of a nearly scale invariant spectrum, with 2n = 2n
= 0.1, motivated by an inﬂationary
mechanism of generation of magnetic ﬁelds [27, 28].
We will only consider the eﬀect of magnetic ﬁeld on the CMB for modes with l ≤ l
max
= 10
4
, as computations
at higher l are very expensive. This corresponds to a minimum comoving scale of 1 Mpc, or k
max
∼ 1 Mpc
−1
. For
magnetic ﬁelds generated at the electroweak phase transition, the coherence scale is estimated at kpc scales or less
[6, 12]. Hence k
I
may be expected to be 10
3
Mpc
−1
. This is much larger than k
max
and thus the l ≤ 10
4
modes of the
CMB are likely to be aﬀected only by the magnetic ﬁeld modes in the inertial range 0 < k < k
I
, and the form of the
power spectrum for k > k
I
plays no direct role. However, even then, the large k part of the power spectrum would
still enter when we derive a constraint on the magnetic ﬁeld, since the amplitude, S
∗
, is ﬁxed by the total energy
density in the magnetic ﬁeld. We now make this point clearer.
First deﬁne an “eﬀective magnetic ﬁeld”, B
eﬀ
, in terms of the total energy density in the magnetic ﬁeld,
0
,
0
≡
B
2
eﬀ
8π
. (4)
In other words, B
eﬀ
is the ﬁeld strength of a uniform magnetic ﬁeld that has the same total energy density as our
stochastic magnetic ﬁeld. Our constraints will be written in terms of B
eﬀ
. To connect to the amplitude of the power
spectrum, we ﬁrst evaluate the energy density in the magnetic ﬁeld. From Eq. (2) we get
M
(k) =
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
S∗k
3
I
16π
3
n
_
k
k
I
_
2n
, k ≤ k
I
S∗k
3
I
16π
3
n
_
1 +
n
n
_
_
k
k
I
_
2n
−1
__
, k
I
< k ≤ k
diss
S∗k
3
I
16π
3
n
_
1 +
n
n
_
_
k
diss
k
I
_
2n
−1
__
, k
diss
< k
(5)
The total energy density in the magnetic ﬁeld,
0
, is found by setting k →∞, which is the same as
M
(k) for k > k
diss
.
Thus, we can write
0
=
S
∗
k
3
I
κ
16π
3
n
, (6)
where
κ ≡ 1 +
n
n
_
_
k
diss
k
I
_
2n
−1
_
, (7)
and hence
B
eﬀ
=
1
π
_
κS
∗
k
3
I
2n
. (8)
For a ﬁxed exponent n, CMB observations at l ≤ 10
4
will only constrain the combination S
∗
k
3
I
. To then convert the
constraint to a bound on the energy density in magnetic ﬁelds requires knowledge of the exponent n
and the inertial
and dissipation scales. In other words, the CMB signature for l ≤ 10
4
probes the long wavelength tail of the magnetic
4
spectrum and not the modes where the bulk of the energy density resides. This suggests that it may be favorable to
investigate the CMB at yet higher l; indeed, our results do show stronger signatures with growing l.
Big bang nucleosynthesis (BBN) is sensitive to the total energy density in the magnetic ﬁeld since this is what enters
the expansion rate of the universe. The constraint from BBN [29–32] is best expressed in terms of the magnetic ﬁeld
energy density relative to the photon energy density
Ω
Bγ
≡
0
ρ
γ
10
−1
(9)
The relative density, Ω
Bγ
, is independent of cosmological epoch since both magnetic ﬁeld and photon energy density
scale as a
−4
where a(t) is the cosmological scale factor. Note that ρ
γ
is the photon density which is diﬀerent from the
radiation density at BBN epoch since neutrinos also contribute to radiation.
Finally, all the scaling due to the expansion of the universe can be pulled out by converting to rescaled quantities
Ba
2
→B , ρ
γ
a
4
→ρ
γ
, ka →k , (10)
and, in what follows, we will use these comoving quantities unless explicitly stated. The ﬁnal form of the magnetic
ﬁeld power spectrum in the inertial range is
S(k) = Ω
Bγ
ρ
γ
16π
3
n
κk
3
I
_
k
k
I
_
2n−3
, k < k
I
(11)
The power spectrum S(k) will enter the calculation of C
BB
l
in combination with powers of k. So it is convenient to
introduce the dimensionless “power spectrum” using powers of k and also the wavelength of observed radiation, λ
0
,
∆
2
M
(k) ≡ k
3
S(k)
_
3λ
2
0
16π
2
e
_
2
=
_
¸
¸
¸
_
¸
¸
¸
_
∆
2
0
_
k
k
I
_
2n
0 < k < k
I
∆
2
0
_
k
k
I
_
2n
k
I
< k < k
diss
0 k > k
diss
(12)
where
∆
2
0
≡
9n
16πe
2
κ
ρ
γ
λ
4
0
Ω
Bγ
(13)
At the present epoch ρ
γ
(t
0
) = 4.64 10
−34
gm/cm
3
= 2 10
−15
(eV)
4
[33], and so
∆
2
0
= 1.1 10
4
Ω
Bγ
κ
_
2n
5
__
90 GHz
ν
0
_
4
(14)
where we denote the observed CMB frequency by ν
0
. Note that ∆
2
0
is independent of k
I
.
In principle, k
diss
is not an independent parameter. One can estimate its value for a given amplitude and shape
of the magnetic ﬁelds spectrum. According to [34, 35], k
diss
is determined by damping into Alfven waves and can be
related to B
eﬀ
as
k
diss
1Mpc
−1
≈ 1.4 h
1/2
_
10
−7
Gauss
B
eﬀ
_
, (15)
Converting this to Ω
Bγ
, we obtain
k
diss
≈ 0.43
¸
10
−2
h
Ω
Bγ
Mpc
−1
, (16)
where it was useful to know that 1 Gauss = 6.9 10
−20
GeV
2
when converting between natural units and CGS units.
We also note that Eq. (4) implies:
B
eﬀ
= 3.25 10
−6
_
Ω
Bγ
Gauss . (17)
The relation (15) is based on the analysis in Ref. [34] where small perturbations on top of a homogeneous magnetic
ﬁeld were treated. To extend this analysis to a stochastic magnetic ﬁeld with little power on long wavelengths, Ref. [35]
introduced a smoothing procedure and split the spectrum into a “homogeneous” part and a “perturbations” part.
It is not clear to us if this procedure is valid for an arbitrary spectrum, S(k), but we will still use Eq. (15) as an
approximate expression for the dissipation scale.
5
III. FARADAY ROTATION CORRELATORS
The CMB is linearly polarized and an intervening magnetic ﬁeld will rotate the polarization vector at a rate given
by:
dα = λ
2
e
3
2πm
2
e
a n
e
B dl , (18)
where λ is the wavelength of light, a is the scale factor normalized so that a
today
= 1, n
e
is the number density of free
electrons, dl is the comoving length element along the photon trajectory from the source to the observer and we are
using Gaussian natural units with = c = 1. Using the known expression for Thomson scattering crosssection,
σ
T
=
8πe
4
3m
2
e
, (19)
and integrating along the line of sight, we obtain the Faraday rotation of the polarization angle,
α =
3
16π
2
e
λ
2
0
_
˙ τ(x)
˜
B dl (20)
where ˙ τ(x) ≡ n
e
σ
T
a is the diﬀerential optical depth along the line of sight, λ
0
is the observed wavelength of the
radiation and
˜
B ≡ Ba
2
is the “comoving” magnetic ﬁeld. The limits of the integral are from the initial position of
the photon to the ﬁnal position.
FR depends on the free electron density, which becomes negligible towards the end of recombination. Therefore, the
bulk of the rotation is produced during a relatively brief period of time when the electron density is suﬃciently low
for polarization to be produced and yet suﬃciently high for the FR to occur. The average FR (in radians) between
Thomson scatterings due to a tangled magnetic ﬁeld was calculated in Ref. [36] and is given by
F =
3
8π
2
e
B
0
ν
2
0
≈ 0.08
_
B
0
10
−9
G
__
30GHz
ν
0
_
2
, (21)
where B
0
is the current amplitude of the ﬁeld and ν
0
is the radiation frequency observed today.
In this section we will calculate the twopoint correlation functions of the FR angle, α, and this will be related to the
twopoint correlation function of the magnetic ﬁeld as given in Eq. (1). The FR correlator will enter the calculation
of C
BB
l
in the thin LSS approximation of Sec. IV, in which one assumes that all of the polarization was generated
at once independently from the FR. In the general case, which we present in Sec. V, the generation of the CMB
polarization and its FR are entangled.
FR is sensitive only to the line of sight component of the magnetic ﬁeld, whereas magnetic helicity, described by
the helical power spectrum A(k) in Eq. (1), depends on all 3 components of the magnetic ﬁeld. So a correlator of FR
cannot be sensitive to the helical properties [22, 37, 38]
1
. Following Ref. [39] without the helical term, we get
¸α(ˆ n)α(ˆ n
)¸ =
_
3λ
2
0
16π
2
e
_
2
_
d
3
k
(2π)
3
S(k)
_
dη
_
dη
˙ τ(η) ˙ τ(η
)e
−ik·ˆ nη
e
+ik·ˆ n
η
[ˆ n ˆ n
−(
ˆ
k ˆ n)(
ˆ
k ˆ n
)] (22)
where ˆ n, ˆ n
are two directions on the sky. Note that, as is conventional, we have written ˙ τ(x) in Eq. (20) as ˙ τ(η)
in the integrand of Eq. (22) even though x = ˆ n(η
0
− η). We have also ignored inhomogeneities in the free electron
density along diﬀerent directions on the sky since this will only give a higher order correction to FR. The limits of the
η, η
integrations are from the time that the photon last scattered to the present time. In general, the last scattering
time will be diﬀerent for diﬀerent photons but, in the thin LSS approximation, the initial time will be taken to be η
∗
,
the epoch at which the “visibility function”, g(η) ≡ ˙ τe
−τ
, is maximum.
Statistical isotropy implies that ¸α(ˆ n)α(ˆ n
)¸ must be a function of ˆ n ˆ n
. This can also be seen directly by writing
Eq. (22) as
¸α(ˆ n)α(ˆ n
)¸ =
_
3λ
2
0
16π
2
e
_
2
_
k
2
dk
(2π)
3
S(k)
_
dη
_
dη
˙ τ(η) ˙ τ(η
) [ˆ n ˆ n
−∂
kη
∂
kη
]
_
d
2
ˆ
ke
−ik·ˆ nη
e
+ik·ˆ n
η
, (23)
1
Indirectly though, helicity does enter the FR signature because magnetic helicity plays a crucial role in the evolution of magnetic ﬁelds
and the exponent n in S(k) (see Eq. (11)).
6
and using
_
d
2
ˆ
ke
−ik·ˆ nη
e
+ik·ˆ n
η
= 4π
l
(2L + 1)j
L
(kη)j
L
(kη
)P
L
(ˆ n ˆ n
) . (24)
where j
L
(x) are Bessel functions and P
L
(x) are Legendre polynomials. Hence, ¸α(ˆ n)α(ˆ n
)¸ depends only on ˆ n ˆ n
as
expected. As a consequence, the correlator of α’s can be expanded into Legendre series
¸α(ˆ n)α(ˆ n
)¸ =
L
(2L + 1)
4π
C
αα
L
P
L
(ˆ n ˆ n
) (25)
and the FR correlators can also be written as ¸α
∗
LM
α
L
M
¸ = C
αα
L
δ
LL
δ
MM
, where α
LM
are the coeﬃcients in the
spherical harmonic decomposition of the FR angle,
α(ˆ n) =
L,M
α
LM
Y
LM
(ˆ n) . (26)
Now combining (23) and (24), and introducing j
L
(x) ≡ ∂
x
j
L
(x), we can write
¸α(ˆ n)α(ˆ n
)¸ =
_
3λ
2
0
16π
2
e
_
2
2
π
_
k
2
dkS(k)
_
dη
_
dη
˙ τ(η) ˙ τ(η
)
L
(2L + 1)
4π
[(ˆ n ˆ n
)P
L
(ˆ n ˆ n
)j
L
(kη)j
L
(kη
) −j
L
(kη)j
L
(kη
)P
L
(ˆ n ˆ n
)] . (27)
Applying the recursion relation
(L + 1)P
L+1
(x) = (2L + 1)xP
L
(x) −LP
L−1
(x) (28)
to the (ˆ n ˆ n
)P
L
(ˆ n ˆ n
) term in (27) results in
¸α(ˆ n)α(ˆ n
)¸ =
_
3λ
2
0
16π
2
e
_
2
2
π
_
k
2
dkS(k)
_
dη
_
dη
˙ τ(η) ˙ τ(η
)
L
(2L + 1)
4π
P
L
(ˆ n ˆ n
)
_
L
2L + 1
j
L−1
(kη)j
L−1
(kη
) +
L + 1
2L + 1
j
L+1
(kη)j
L+1
(kη
) −j
L
(kη)j
L
(kη
)
_
. (29)
In analogy with the way C
L
’s are evaluated for CMB, it is convenient to introduce “transfer” functions T
L
(k) and
T
(1)
L
(k) deﬁned as
T
L
(k) ≡
_
η0
η∗
dη ˙ τ(η)j
L
(k(η
0
−η))
T
(1)
L
(k) ≡
_
η0
η∗
dη ˙ τ(η)j
L
(k(η
0
−η)) (30)
where, as deﬁned above, η
∗
is the epoch at which the visibility function is maximum and η
0
is the present epoch.
Using these transfer functions in (29) and comparing to (25) allows us to write
C
αα
L
=
2
π
_
dk
k
∆
2
M
(k)
_
L
2L + 1
(T
L−1
(k))
2
+
L + 1
2L + 1
(T
L+1
(k))
2
−(T
(1)
L
(k))
2
_
. (31)
The function T
(1)
L
can be expressed in terms of T functions by using the relation
j
L
(x) =
1
2L + 1
[Lj
L−1
(x) −(L + 1)j
L+1
(x)] . (32)
This gives
T
(1)
L
=
1
2L + 1
[LT
L−1
−(L + 1)T
L+1
] . (33)
The functions T
L
(k) are independent of the magnetic ﬁeld, and can be easily evaluated numerically using the ionization
history from CMBFAST.
Eq. (31) is our ﬁnal result for the FR correlation function. It will be useful in Sec. IV where we ﬁnd C
BB
l
in the
thin LSS approximation. However, the result is not useful in the general case of a thick LSS because then C
BB
l
is not
simply related to C
αα
L
.
7
IV. FARADAY ROTATION OF CMB IN THIN LSS APPROXIMATION
In the limit of instant last scattering one assumes that all of the polarization was generated at the peak of the
visibility function. Since we are interested speciﬁcally in the FR eﬀects, we will neglect primordial tensor modes and
any actively sourced vector and tensor modes (including those sourced by magnetic ﬁelds) so that only E mode is
produced at the instant of last scattering. At subsequent times, because of the residual presence of charged particles,
some of this E mode will be Faraday rotated into B mode. To estimate this eﬀect, we can start with Eq. (6) of [40]
(same as Eq. (20) of [41]) which gives the Bmode coeﬃcients
B
lm
= 2(−1)
m
LM
l2m2
α
LM
E
l2m2
ξ
LM
lml2m2
H
L
ll2
, (34)
where ξ
LM
lml2m2
and H
L
ll2
are deﬁned in terms of Wigner 3 −j symbols as [41]
ξ
LM
lml2m2
≡ (−1)
m
_
(2l + 1)(2L + 1)(2l
2
+ 1)
4π
_
l L l
2
−m M m
2
_
(35)
H
L
ll2
≡
_
l L l
2
2 0 −2
_
, (36)
and the summation is restricted to be only over even L+l
2
+l. From the above, we can derive the expression relating
C
BB
l
to C
EE
l
:
¸B
∗
l
m
B
lm
¸ = 4
LM
L
M
l2m2
l
2
m
2
ξ
LM
lml2m2
H
L
ll2
ξ
L
M
l
m
l
2
m
2
H
L
l
l
2
¸α
∗
LM
E
∗
l2m2
α
L
M
E
l
2
m
2
¸
= δ
ll
δ
mm
4
L
(2L + 1)
4π
C
αα
L
l2
(2l
2
+ 1)C
EE
l2
(H
L
ll2
)
2
(37)
which assumes a statistically isotropic stochastic magnetic ﬁelds with a FR angular spectrum C
αα
L
given by Eq. (31).
Therefore, the Bmode angular spectrum in the thin last scattering approximation is
C
BB
l
=
1
π
L
(2L + 1)C
αα
L
l2
(2l
2
+ 1)C
EE
l2
(H
L
ll2
)
2
. (38)
It is instructive to put the expressions for the CMB observables in a form that separates the wellestablished physics
of FR of CMB polarization from the particular form of the magnetic ﬁeld spectrum. For example, substituting (31)
into (38), we can rewrite the latter as
C
BB
l
=
2
π
_
dk
k
∆
2
M
(k)W
l
(k) , (39)
where W
l
(k) are “window functions” deﬁned as
W
l
(k) = 4
l1L
(2l
1
+ 1)(2L + 1)
4π
(H
L
ll1
)
2
C
EE
l1
_
L
2L + 1
[T
L−1
(k)]
2
+
L + 1
2L + 1
[T
L+1
(k)]
2
−[T
(1)
L
(k)]
2
_
. (40)
They describe the amount of power a given wavelength k of the magnetic ﬁeld spectrum contributes to a given angular
scale l of the Bmode polarization spectrum. We note that Eq. (39) relating C
BB
l
to the magnetic spectrum is formally
independent of the thin LSS approximation – the approximation is used in the calculation of the window functions.
Transfer functions T
L
(k) can be found numerically using the diﬀerential optical depth calculated in CMBFAST [42].
Having evaluated the window functions once, one can store them and use (39) to calculate C
BB
l
for diﬀerent choices
of the magnetic spectrum.
V. EXACT (THICK LSS) TREATMENT OF FARADAY ROTATION
The thin LSS approximation decouples the process of generation of the CMB polarization by Thomson scattering
from its subsequent FR by magnetic ﬁelds. It eﬀectively assumes that the background Emode polarization on all
8
angular scales was created at a single instant in time corresponding to the peak of the visibility function. While this
may suﬃce for order of magnitude estimates, polarization in diﬀerent parts of the sky was created at diﬀerent times
and any choice of a single time is essentially arbitrary. Furthermore, the amount of the FR strongly depends on the
choice of the initial instant, since the amplitude of the rotation is directly proportional on the rapidly decreasing free
electron density. In this Section we derive the exact FR window functions, denoted by
˜
W
l
(k), by solving the radiative
transport equations for the generation and propagation of CMB polarization in the presence of FR by stochastic
magnetic ﬁelds. The form of Eq. (39) relating C
BB
l
to the magnetic spectrum will be the same.
The evolution of CMB Stokes parameters is described by Chandrasekhar’s radiative transport equations [43]. In
the absence of FR, equations for the q Fourier mode of linear polarization parameters Q and U are [44]
˙
P
±
+iqµP
±
= −˙ τP
±
+S
±
, (41)
where P
±
(q, ˆ n, η) = Q±iU, ˆ n is the direction of the line of sight, µ = ˆ q ˆ n, and
S
±
= ˙ τ
√
6
2
a=−2
P
(a)
(q, η)
±2
Y
2a
(ˆ n)
_
4π
5
. (42)
In the above,
±2
Y
lm
(ˆ n) are spin2 spherical harmonics, and P
(a)
(q, η) = [Θ
(a)
2
−
√
6E
(a)
2
]/10, where Θ
(a)
2
and E
(a)
2
are
the quadrupole moments of the CMB temperature and Emode polarization for scalar (a = 0), vector (a = ±1) and
tensor (a=±2) modes. Assuming that polarization generated by vector and tensor sources is negligible, we have
S
±
= ˙ τ
√
6P
(0)
(q, η)
±2
Y
20
(ˆ n)
_
4π
5
. (43)
FR rotates Q into U, and U into Q, leading to a new term on the right hand of (41) [45]:
˙
P
±
+iqµP
±
= −˙ τP
±
∓2iω
B
P
±
+S
±
. (44)
where ω
B
(ˆ n, η) = f ˙ τB(r) ˆ n, r = (η
0
−η)ˆ n, f = 3λ
2
0
/(2πe), and λ
0
and B are the comoving wavelength and magnetic
ﬁeld strength. The ordinary diﬀerential equation (44) has the inhomogeneous solution
P
±
=
_
η0
0
dη ˜ s
±
(q, ˆ n, η)e
∓2i
η
0
η
ω
B
dη
, (45)
with
˜ s
±
= S
±
e
−τ
e
−iqµ(η0−η)
= −˙ τe
−τ
√
6P
(0)
(q, η)
l
(−i)
l
_
4π(2l + 1)[
(0)
l
(q(η
0
−η)) ±iβ
(0)
l
(q(η
0
−η))]
±2
Y
l0
(ˆ n) , (46)
and τ ≡
_
η0
η
dη
˙ τ. In the above, we used the identity (Eq. (16) of [44])
−
_
4π
5
±2
Y
20
(ˆ n)e
i q·ˆ nr
=
l
(−i)
l
_
4π(2l + 1)[
(0)
l
(qr) ±iβ
(0)
l
(qr)]
±2
Y
l0
(ˆ n) , (47)
with (in what follows we will not need β
(0)
l
)
(0)
l
(x) ≡
¸
3
8
(l + 2)!
(l −2)!
j
l
(x)
x
2
. (48)
For small ω
B
, we can write (45) as
P
±
(q, ˆ n) =
_
η0
0
dη ˜ s
±
(q, ˆ n, η)
_
1 ∓2i
_
η0
η
ω
B
dη
_
. (49)
Next, we can use the total angular momentum formalism of [44] to derive an expression for C
BB
l
in terms of the
magnetic spectrum S(k). From Eq. (55) of [44] we have
P
±
(q, ˆ n) =
l
(−i)
l
_
4π
2l + 1
l
m=−l
_
E
(m)
l
(q) ±iB
(m)
l
(q)
_
±2
Y
lm
(ˆ n) , (50)
9
where ˆ q = ˆ z. Note that in Eq. (55) of [44] the sum over m runs only from −2 to 2 because these are the only modes
that can be sourced by scalar, vector and tensor ﬂuctuations in the metric. However, the FR eﬀect on the propagation
of photons is not via perturbations of the metric tensor. Hence, to stay general, we keep the sum to be over all m
modes. Inverting (50) and using (49) we obtain
B
(m)
l
(q) =
1
(−i)
l
_
2l + 1
4π
_
dˆ n
_
η0
0
dη ˙ τe
−τ
√
6P
(0)
(q, η)
l1
(−i)
l1
_
4π(2l
1
+ 1)[
(0)
l1
(
+2
Y
l10
[
+2
Y
lm
]
∗
+
−2
Y
l10
[
−2
Y
lm
]
∗
)
_
η0
η
ω
B
dη
, (51)
where we have assumed that FR is the only source of Bmode. The angular spectrum C
BB
l
can be written in terms
of B
(m)
l
(q) as [44]
(2l + 1)
2
C
BB
l
= 4π
_
d
3
q
(2π)
3
l
m=−l
¸B
(m)∗
l
(q)B
(m)
l
(q)¸ . (52)
Introducing g(η) = ˙ τ exp(−τ) and X
m
ll1
≡
+2
Y
l10
[
+2
Y
lm
]
∗
+
−2
Y
l10
[
−2
Y
lm
]
∗
, and substituting (51) into (52) we obtain
(2l + 1)C
BB
l
= 4π
_
d
3
q
(2π)
3
6
_
η0
0
dη g(η)
_
η0
0
dη
g(η
)¸P
(0)∗
(q, η)P
(0)
(q, η
)¸
l
m=−l
l1
l2
(−i)
l1
i
l2
_
(2l
1
+ 1)(2l
2
+ 1)
(0)
l1
(q(η
0
−η))
(0)
l2
(q(η
0
−η
))
_
dˆ n
_
dˆ n
X
m∗
ll1
(ˆ n)X
m
ll2
(ˆ n
)¸
_
η0
η
dη
_
η0
η
dη
ω
B
(η
, ˆ n)ω
B
(η
, ˆ n
)¸ . (53)
Eq. (27) for the equal time twopoint correlation of rotation measure is easily generalized to the unequal time FR
correlation case above. Namely, we have
¸
_
η0
η
dη
_
η0
η
dη
ω
B
(η
, ˆ n)ω
B
(η
, ˆ n
)¸ =
2
π
_
dk
k
∆
2
M
(k)
L
_
2L + 1
4π
_
P
L
(ˆ n ˆ n
)
L
(k, η, η
) , (54)
where

L
(k, η, η
) =
L
2L + 1
T
L−1
(k, η)T
L−1
(k, η
) +
L + 1
2L + 1
T
L+1
(k, η)T
L+1
(k, η
) −T
(1)
L
(k, η)T
(1)
L
(k, η
) (55)
and the transfer functions are the same as before, except for the range of the time integration:
T
L
(k, η) ≡
_
η0
η
dη
˙ τ(η
)j
L
(k[η
0
−η
])
T
(1)
L
(k, η) ≡
_
η0
η
dη
˙ τ(η
)j
L
(k[η
0
−η
]) . (56)
The 
L
in (55) can be written as a sum of terms with separated η and η
dependencies:

L
(k, η, η
) =
3
c=1
u
(c)
L
(k, η)u
(c)
L
(k, η
) , (57)
where
u
(1)
L
=
_
L
2L + 1
T
L−1
, u
(2)
L
=
_
L + 1
2L + 1
T
L+1
, u
(3)
L
= iT
(1)
L
(58)
We can also relate ¸P
(0)∗
(q, η)P
(0)
(q, η
)¸ to the primordial curvature power spectrum ∆
2
(q) via
¸P
(0)∗
(q, η)P
(0)
(q, η
)¸ = q
−3
∆
2
(q)P
(0)∗
(q, η)P
(0)
(q, η
) . (59)
10
Putting it all in (53) and integrating over the angular dependence of q, we obtain
(2l + 1)C
BB
l
=
2
π
_
dk
k
∆
2
M
(k)
l1,l2,L
mM
:
LM∗
l10lm
:
LM
l20lm
2
π
_
dq
q
∆
2
(q)
3
c=1
d
(c)∗
l1L
(q, k)d
(c)
l2L
(q, k) , (60)
where we have deﬁned
:
LM
l10lm
= (−i)
l1
_
(2l
1
+ 1)
_
dˆ n X
m
ll1
Y
∗
LM
, (61)
and
d
(c)
l1L
(q, k) =
_
η0
0
dη g(η)
√
6P
(0)
(q, η)
(0)
l1
(q(η
0
−η))u
(c)
L
(k, η) . (62)
Using the expression for the integral of a product of three spinweighted spherical harmonics
_
dˆ n
a
Y
l1m1
(ˆ n)
b
Y
LM
(ˆ n)
c
Y
l2m2
(ˆ n) =
_
(2l
1
+ 1)(2L + 1)(2l
2
+ 1)
4π
_
l
1
L l
2
m
1
M m
2
__
l
1
L l
2
−a −b −c
_
(63)
and the orthogonality property of Wigner 3j symbols [46] , we have
mM
:
LM∗
l10lm
:
LM
l20lm
= 4δ
l1l2
(2l + 1)(2l
1
+ 1)(2L + 1)
4π
(1
L
ll1
)
2
if l +l
1
+L = even, and 0 otherwise . (64)
Substituting this into (60) we can write
C
BB
l
=
2
π
_
dk
k
∆
2
M
(k)
˜
W
l
(k) , (65)
with
˜
W
l
(k) = 4
l1L
(2l
1
+ 1)(2L + 1)
4π
(1
L
ll1
)
2
S
l1L
(k) , (66)
where l +l
1
+L = even, and
S
l1L
(k) =
2
π
_
dq
q
∆
2
(q)
3
c=1
[d
(c)
l1L
(q, k)[
2
. (67)
The thick LSS window function in Eq. (66) is very similar to the thin LSS window function given by Eq. (40).
Noting that
C
EE
l1
=
2
π
_
dq
q
∆
2
(q)
__
η0
0
dη g(η)
√
6P
(0)
(q, η)
(0)
l1
(q(η
0
−η))
_
2
, (68)
we can see that
2
π
_
dq
q
∆
2
(q)
3
c=1
[d
(c)
l1L
(q, k)[
2
(69)
of the thick case becomes
C
EE
l1
_
L
2L + 1
[T
L−1
(k)]
2
+
L + 1
2L + 1
[T
L+1
(k)]
2
−[T
(1)
L
(k)]
2
_
(70)
of the thin case if the function d
(c)
l1L
(q, k) deﬁned in Eq. (62) is “factorizable”, that is if
d
(c)
l1L
→u
(c)
L
(k, η
∗
)
_
η0
0
dη g(η)
√
6P
(0)
(q, η)
(0)
l1
(q(η
0
−η)) . (71)
11
1 10 100 1000
0.0001
0.001
l
FIG. 2: Window functions W
l
(k) for diﬀerent values of k plotted vs l as evaluated using the full LSS treatment (solid lines)
and using the thin LSS approximation(dotted lines). These window functions prescribe the way in which a given Fourier mode
k of the stochastic magnetic ﬁeld contributes to the multipole l of C
BB
l
. For example, C
BB
l
at l ∼ 8000 is sensitive to ∆
2
M
at
k = 0.5, 1 Mpc
−1
but not to smaller k. The left and right panels show the same functions plotted on logarithmic and linear
axis respectively.
Thus the thick LSS case reduces to the thin LSS case if we disregard the convolution in Eq. (62).
The convolution in Eq. (62) will in general not be factorizable, i.e. the diﬀerent q modes of the source Emode
polarization are not all created at a single time before FR took place. Instead, functions d
(c)
l1L
(q, k) determine the
relative amount by which a given qmode, projecting into multipole l
1
of C
EE
l1
, is distorted by the kmode of the
magnetic ﬁeld projecting onto multipole L of the FR distortion spectrum.
To evaluate
˜
W
l
(k) numerically, we modiﬁed CMBFAST to calculate sources S
l1L
(k) on a grid in L and k. In order
to accurately account for magnetic ﬁelds on scales up to a given k
max
in Mpc
−1
, one needs to evaluate the source up
to L
max
∼ (10
4
Mpc)k
max
. We included the l
1
modes up to 6000, and conﬁrmed that it is more than suﬃcient for all
k and L because of the exponential suppression of the source Emodes by the Silk damping. Time required for the
evaluation of sources needed for
˜
W
l
(k) was ∼ 4000 CPUhours, where the sources were sampled for 450k bins. Once
the sources are calculated and stored, the sums over l
1
and L in (66) are quick to perform.
Even though computing the exact window functions
˜
W
l
(k) takes a nontrivial amount of CPU time, it only needs to
be done once for a given cosmological model. We have made our window functions, evaluated using the ΛCDM model
with WMAP7 best ﬁt parameters [47], publicly available at http://www.sfu.ca/∼levon/faraday.html, along with
a Fortran code that calculates C
BB
l
for a given ∆
2
M
(k).
VI. RESULTS
In Fig. 2 we show window functions for several values of k as a function of multipole l for the thin (dotted lines)
and thick (solid lines) LSS treatments discussed in previous sections. These results are independent of the spectral
features of the magnetic ﬁeld. Instead, the role of the window is to specify the extent to which a given Fourier mode
12
FIG. 3: The CMB Bmode spectrum from Faraday rotation evaluated in the case of a nearly scaleinvariant magnetic spectrum
with 2n = 2n
= 0.1, kI = 1 Mpc
−1
, and f
cmb
= 61 GHz, using the thin (blue dot) and exact (solid red) treatment of LSS. The
value of the magnetic ﬁeld energy fraction ΩBγ = 5 ×10
−4
corresponds to B
eﬀ
≈ 0.73 ×10
−7
Gauss. The black shortdash line
is the input Emode spectrum, the black dashdot line is the contribution from inﬂationary gravitational waves with r = 0.1,
while the black longdash line is the expected contribution from gravitational lensing by large scale structure.
k of the stochastic magnetic ﬁeld contributes to the multipole l of C
BB
l
. One can see that each window has a peak at
l approximately given by l = 10
4
k, where 10
4
is roughly the distance to LSS in Mpc. One can also see the oscillations
which come from the acoustic oscillations in the Emode spectrum.
Comparing the exact (thick LSS) windows with the ones obtained in the thin LSS approximation, we note that
they have comparable shapes and amplitudes near their peaks, but diﬀer signiﬁcantly at l away from the peaks. This
diﬀerence comes because of the assumption made in the thin LSS approximation that all of the Emode was produced
at the same time, so that the FR occurs at the same rate on all scales. In reality, Emode is rotated as it is being
produced and, since the rate of FR depends on the rapidly decreasing free electron density, Emode scales produced
at diﬀerent times are rotated at diﬀerent rates.
Let us now focus on the Bmode spectra for speciﬁc choices of the magnetic spectrum. In Fig. 3 we consider a nearly
scale invariant magnetic spectrum with 2n = 0.1. The black dash line shows the input Emode, while the blue dot and
the red solid lines show the Bmode spectra obtained using the thin and exact treatment of LSS. One can see that the
exactly calculated spectrum favors the power near the peak at the cost of the power around it. Note that in this case
the shape of the Bmode is essentially a copy of the Emode spectrum, except for the lack of exponential damping
on small scales. While the EE correlations are suppressed by the Silk damping, there is no exponential suppression of
the FR generated small scale Bmode spectrum because the magnetic ﬁeld is correlated on small scales. For 2n = 0.1,
both thin LSS and exact spectra have the asymptotic form of l
−1
at high l. More generally, the asymptotic exponent
is 2n −1, which can be relatively large for stochastic ﬁelds e.g. for n = 5/2, l
2
C
BB
l
∝ l
4
at large l.
The Silk damping dissipates the perturbations in the CMB temperature and Emode on scales smaller than 9 Mpc
or so. In our formalism, it comes through the exponential suppression of the source function P
(0)
(k, η) in Eq. (43).
13
FIG. 4: The CMB Bmode spectrum from Faraday rotation in the 2n = 5, 2n
= 3 case motivated by causally generated ﬁelds,
with kI = 1 Mpc
−1
and f
cmb
= 61 GHz. The thin blue dot line shows the thin LSS calculation, while the exact (thick) LSS
calculation is shown with solid red. The value of the magnetic ﬁeld energy fraction, ΩBγ = 10
−3
, corresponds to B
eﬀ
≈ 10
−7
Gauss. The other lines are the same as in Fig. 3.
However, the dissipation scale of (weak) magnetic ﬁelds at decoupling time is typically much smaller [13, 34] and
for k 1Mpc
−1
the magnetic ﬁeld can be treated as a stiﬀ source, i.e. we can safely assume that its evolution is
independent of the perturbations in the photonbaryon ﬂuid. Then, the only damping of the FR induced Bmode
power is due to averaging over many random rotations along the line of sight. The functional form of this suppression
is a power law and the exponent can be estimated by observing that a random superposition of N perturbations along
the line of sight leads to a statistical reduction in the amplitude of the observed anisotropy by a factor 1/
√
N. For
wavenumber k we have N ∝ ∆ηk, where ∆η is the period of time, comparable to the thickness of LSS, during which
FR is eﬃcient. Thus, the power spectrum on small scales, which is the square of the FR amplitude, is suppressed by
1/k, which translates into the 1/l suppression of the angular spectrum.
In Fig. 4 we show the plot of the Bmode spectrum for the magnetic spectrum index 2n = 5 expected for causally
generated magnetic ﬁelds [13]. In the ﬁgure we used k
I
= 1 Mpc
−1
. For comparison, we show the Bmode from weak
lensing and from gravity waves, as well as the Emode spectrum. Note that at small angular scales (high l), the FR
produced Bmode can dominate the signal as it keeps growing as l
2n−1
.
At present, Bmode have not been detected; there are only upper bounds. Still, even these weak bounds can produce
constraints on the magnetic ﬁeld fraction Ω
Bγ
, and furthermore, the bounds will improve rapidly as CMB observations
are made on smaller angular scales. We derive current constraints on the magnetic ﬁeld energy density, Ω
Bγ
, from the
WMAP 7year polarization data by comparing magnetic ﬁeld induced theoretical CMB Bmode power spectrum C
BB
as given by Eq. (65), with the WMAP observed Bmode power spectrum using the χ
2
statistics. We consider three
WMAP frequency bands Q, V, and W corresponding to frequencies 41 GHz, 61 GHz, and 94 GHz respectively. The
lower frequency bands, K and Ka, are foreground dominated and we do not include those in our analysis. We combine
14
FIG. 5: Constraints from WMAP7 year data on the magnetic ﬁeld density ΩBγ, or analogously magnetic ﬁeld eﬀective
amplitude B
eﬀ
as deﬁned by Eqs. (4) and (17), as a function of the inertial scale wavevector kI. We consider two choices of
magnetic spectrum, the case of a nearly scaleinvariant spectrum 2n = 2n
= 0.1 (dashdot red line) and causal magnetic ﬁelds
with 2n = 5, 2n
= 3 (dashed red line). The BBN constraints on the magnetic ﬁeld density ΩBγ < 0.1 is shown by the dotted
blue line. At large kI the bound is set by the theoretical relation between ΩBγ and k
diss
≥ kI given by Eq. (15) (black solid
line).
diﬀerent frequency channels directly when evaluating the χ
2
and, when evaluating the likelihood, we restrict the
maximum value of Ω
Bγ
to the one given by Eq (16) for k
I
= k
diss
. We used WMAP data for > 32, above which the
errors for individual ’s can be treated as uncorrelated. The maximum multipole considered in the analysis is = 700.
For our analysis we considered two choices of theoretical magnetic spectral indices (n, n
), one corresponding to nearly
scaleinvariant spectrum 2n = 2n
= 0.1, and the other corresponding to causal magnetic ﬁelds 2n = 5, 2n
= 3. Our
analysis assumes that the magnetic ﬁeld is the only source of the Bmode signal, and ignores the possibility of other
sources of Bmodes such as inﬂationary gravitational waves [48], weak gravitational lensing of the CMB [49], and
several other distortions of primary CMB along the line of sight (see Refs. [50, 51] for examples of such distortions).
To obtain bounds on the magnetic ﬁeld, we note that the CMB FR signature constrains ∆
2
0
deﬁned in Eq. (14),
which is proportional to Ω
Bγ
/κ. Now κ is deﬁned in Eq. (7) and depends on the dissipation scale, k
diss
, which can
also be related to Ω
Bγ
by using Eq. (16). Therefore for given values of n and n
, κ is a function of the inertial scale,
k
I
, and the energy fraction in the magnetic ﬁeld, Ω
Bγ
. In Fig. 5 we present the 95% conﬁdence level (C.L.). bounds
on Ω
Bγ
for two choices of (n, n
): the nearly scaleinvariant case, 2n = 2n
= 0.1, and the causal case, 2n = 5, 2n
= 3.
The bound in each case is presented as a function of k
I
, which we treat as an unknown parameter in the model.
The dependence of the constraint on Ω
Bγ
on k
I
in each case can be readily understood. In the 2n = 5, 2n
= 3
case, C
BB
l
is independent of k
I
at very small k
I
because for positive n the integral in Eq. (65) is dominated by the
contribution from k ¸k
I
. The dependence of C
BB
l
on k
I
becomes stronger when k
I
gets closer to the range of scales
with signiﬁcant Emode power (k ∼ 0.1 Mpc
−1
), which is encoded in the shapes of the window functions. However,
15
this happens to be just under the maximum wavenumber constrained by WMAP’s measurement of BB. Thus, WMAP
is unable to probe the dependence on k
I
, and the constraint line in Fig. 5 stays almost horizontal even at k
I
∼ 0.1.
The condition k
I
≤ k
diss
∝ Ω
−1/2
Bγ
imposes its own upper bound on Ω
Bγ
at large k
I
. Namely, at suﬃciently large k
I
,
the constraint curve becomes independent of the CMB constraints, and is a consequence of the constraint arising from
dissipation. The curve k
I
= k
diss
is also shown in Fig. 5.
In the scaleinvariant limit, which is the 2n = 2n
= 0.1 case in Fig. 5, the CMB constraint is independent of k
I
because all k dependence eﬀectively disappears. At large k
I
, the bound is eventually dominated by the k
I
= k
diss
curve, which is independent of the CMB data.
We note that a CMB experiment which can measure Bmodes at ∼ 1000 would be sensitive to changes in k
I
in
the case of a causal spectrum. We will present forecasted bounds from future CMB data in an upcoming paper [52].
VII. CONCLUSIONS
Primordial stochastic magnetic ﬁelds may be produced in the early universe during baryogenesis. Characterizing
such primordial magnetic ﬁeld is extremely valuable for probing early universe physics. In this paper we have calculated
the eﬀect of primordial magnetic ﬁelds on the CMB polarization. This study is especially timely in view of upcoming
and next generation of CMB observations that are focused on measuring the polarization of the CMB.
FR of existing Emodes of the CMB can generate parity odd Bmodes whose amplitude will depend on the obser
vation frequency bands. We generalized the signatures previously obtained assuming “thin” last scattering surface by
solving the full CMB radiative transport equation with FR taken into account. Comparing our full treatment with the
thin last scattering approximation, we ﬁnd the respective Bmode spectra have similar shapes but their magnitudes
at a given multipole may diﬀer by as much as a factor of a few depending on the primordial magnetic ﬁeld spectrum
and the angular scale of observation. The wellestablished physics of FR of CMB can be encoded in window functions,
W
l
(k), which determine the relative amount that a given Fourier mode k of the magnetic ﬁeld contributes to the
multipole l of C
BB
l
. The window functions are independent of the details of the magnetic ﬁeld spectrum and only
need to be computed once for a given cosmological model. We have evaluated them in the best ﬁt ΛCDM model and
made them available at http://www.sfu.ca/∼levon/faraday.html, along with a Fortran code that calculates C
BB
l
for a given magnetic spectrum.
FR can take a smooth Emode ﬁeld and distort it to create Bmodes on arbitrarily small scales. For instance, on
scales smaller than the Silk damping scale, the Emode map is essentially homogeneous. However, FR creates E and
Bmode inhomogeneities on subSilk scales by rotating diﬀerent parts of the homogeneous Emode patch in diﬀerent
random ways. As explained in the previous section, the power of these inhomogeneities is suppressed as 1/l due to
random superposition of multiple small scale rotations along the same line of sight.
FR of the CMB due to magnetic ﬁelds has a distinct frequency dependence (∼ ν
2
) which will allow it to be
distinguished from other sources of CMB Bmodes such as lensing, inﬂationary gravitational waves, topological defects
[53, 54] and FR due to pseudo scalar ﬁelds [55–58]. The growth l
2
C
BB
l
∝ l
2n−1
at large l is also characteristic of a
magnetic ﬁeld and will help discriminate primordial FR from other foreground contamination. We have summarized
current constraints on magnetic enegy Ω
Bγ
from WMAP7 based on the FR induced Bmode spectrum in Fig. 5. For
comparison we have also shown the constraints on magnetic energy obtained by BBN. For a scale invariant magnetic
ﬁeld, the constraints from WMAP 7year data are 2orders of magnitude better than the BBN constraints. For a
causal magnetic ﬁeld with (2n = 5, 2n
= 3), BBN constraints are weaker only by a factor of few. However, note that
the constraints from BBN are not expected to improve much in future, but future observations of CMB polarization
at smaller angular scales will signiﬁcantly improve the constraints on magnetic ﬁeld. We also refer the reader to [59]
for the most recent CMB bounds on magnetic ﬁelds based signatures other than FR.
Although in this paper we have focused on Bmode power spectrum of CMB generated due to FR of the CMB
polarization, recently it has been shown that the stochastic FR of polarization of the CMB couples diﬀerent CMB
angular modes, thus generating nonGaussianity which can be seen in the CMB trispectrum [40, 41, 60]. In a followup
paper [52] we discuss the detectability of primordial magnetic ﬁeld using such nonGaussian features in the CMB.
Acknowledgments
We are grateful to the WMAP collaboration, especially Eiichiro Komatsu and Michael Nolta, for providing the
WMAP Bmode power spectra for individual frequency bands. We thank Karsten Jedamzik, Tina Kahniashvili, and
other participants of the Primordial Magnetism Workshop, 2011, at ASU for input, and Richard Battye and Yun
Li for pointing out several inconsistencies in the earlier version of the paper. LP is supported by a Discovery Grant
from the Natural Sciences and Engineering Research Council of Canada, and acknowledges hospitality at Perimeter
16
Institute for Theoretical Physics where part of this work was completed. A.P.S.Y. gratefully acknowledges funding
support from NASA award number NNX08AG40G and NSF grant number AST0807444. TV is supported by the
Department of Energy at ASU, and is grateful to the Institute for Advanced Study, Princeton (IAS) for hospitality.
We also acknowledge the use of cluster computing at the IAS.
[1] L. M. Widrow, Rev. Mod. Phys. 74, 775 (2002).
[2] A. Kandus, K. E. Kunze, and C. G. Tsagas, Phys. Rep. 505, 1 (2011), 1007.3891.
[3] A. Neronov and I. Vovk, Science 328, 73 (2010), 1006.3504.
[4] T. Vachaspati, Phys.Lett. B265, 258 (1991).
[5] J. M. Cornwall, Phys. Rev. D56, 6146 (1997), hepth/9704022.
[6] T. Vachaspati, Phys. Rev. Lett. 87, 251302 (2001), astroph/0101261.
[7] J. GarciaBellido, M. GarciaPerez, and A. GonzalezArroyo, Phys. Rev. D69, 023504 (2004), hepph/0304285.
[8] C. J. Copi, F. Ferrer, T. Vachaspati, and A. Achucarro, Phys. Rev. Lett. 101, 171302 (2008), 0801.3653.
[9] T. Vachaspati, Phil. Trans. Roy. Soc. Lond. A366, 2915 (2008), 0802.1533.
[10] Y. Ng and T. Vachaspati, Phys. Rev. D82, 023008 (2010), 1001.4817.
[11] Y. Chu, J. Dent, and T. Vachaspati, ArXiv eprints (2011), 2011.
[12] G. Sigl, Phys.Rev. D66, 123002 (2002), astroph/0202424.
[13] K. Jedamzik and G. Sigl, ArXiv eprints (2010), 1012.4794.
[14] R. Banerjee and K. Jedamzik, Phys. Rev. D 70, 123003 (2004), arXiv:astroph/0410032.
[15] F. Tavecchio, G. Ghisellini, G. Bonnoli, and L. Foschini (2010), 1009.1048.
[16] K. Dolag, M. Kachelriess, S. Ostapchenko, and R. Tomas, Astrophys.J. 727, L4 (2011), 1009.1782.
[17] A. Taylor, I. Vovk, and A. Neronov (2011), 1101.0932.
[18] K. Takahashi, M. Mori, K. Ichiki, and S. Inoue (2011), 1103.3835.
[19] S. Ando and A. Kusenko, Astrophys.J. 722, L39 (2010), 1005.1924.
[20] A. Kosowsky and A. Loeb, Astrophys.J. 469, 1 (1996), astroph/9601055.
[21] D. D. Harari, J. D. Hayward, and M. Zaldarriaga, Phys.Rev. D55, 1841 (1997), astroph/9608098.
[22] A. Kosowsky, T. Kahniashvili, G. Lavrelashvili, and B. Ratra, Phys. Rev. D71, 043006 (2005), astroph/0409767.
[23] M. Giovannini and K. E. Kunze, Phys. Rev. D78, 023010 (2008), 0804.3380.
[24] T. Kahniashvili, Y. Maravin, and A. Kosowsky, Phys.Rev. D80, 023009 (2009), 0806.1876.
[25] M. Giovannini and K. E. Kunze, Phys. Rev. D79, 063007 (2009), 0812.2207.
[26] R. Durrer and C. Caprini, JCAP 0311, 010 (2003), astroph/0305059.
[27] M. S. Turner and L. M. Widrow, Phys. Rev. D37, 2743 (1988).
[28] B. Ratra, Astrophys. J. 391, L1 (1992).
[29] J. Matese and R. O’Connell, Nature 222, 649 (1969).
[30] P. J. Kernan, G. D. Starkman, and T. Vachaspati, Phys. Rev. D 54, 7207 (1996), arXiv:astroph/9509126.
[31] D. Grasso and H. R. Rubinstein, Phys. Lett. B379, 73 (1996), astroph/9602055.
[32] B.l. Cheng, A. V. Olinto, D. N. Schramm, and J. W. Truran, Phys. Rev. D54, 4714 (1996), astroph/9606163.
[33] S. Dodelson, Modern Cosmology (Academic Press, San Diego, 2003).
[34] K. Jedamzik, V. Katalinic, and A. V. Olinto, Phys. Rev. D57, 3264 (1998), astroph/9606080.
[35] T. Kahniashvili, A. G. Tevzadze, S. K. Sethi, K. Pandey, and B. Ratra, Phys.Rev. D82, 083005 (2010), 1009.2094.
[36] D. D. Harari, J. D. Hayward, and M. Zaldarriaga, Phys. Rev. D 55, 1841 (1997), arXiv:astroph/9608098.
[37] T. A. Ensslin and C. Vogt, Astron. Astrophys. 401, 835 (2003), astroph/0302426.
[38] L. Campanelli, A. D. Dolgov, M. Giannotti, and F. L. Villante, Astrophys. J. 616, 1 (2004), astroph/0405420.
[39] L. Pogosian, T. Vachaspati, and S. Winitzki, Phys. Rev. D 65, 083502 (2002), arXiv:astroph/0112536.
[40] M. Kamionkowski, ArXiv eprints (2008), 0810.1286.
[41] V. Gluscevic, M. Kamionkowski, and A. Cooray, Phys. Rev. D 80, 023510 (2009), 0905.1687.
[42] U. Seljak and M. Zaldarriaga, Astrophys. J. 469, 437 (1996), astroph/9603033.
[43] S. Chandrasekhar, Radiative Transfer (Dover, New York, 1960).
[44] W. Hu and M. J. White, Phys. Rev. D56, 596 (1997), astroph/9702170.
[45] A. Kosowsky and A. Loeb, Astrophys. J. 469, 1 (1996), astroph/9601055.
[46] D. A. Varshalovich, A. N. Moskalev, and V. K. Kherso´ nskii, Quantum theory of angular momentum (World Scientiﬁc,
Singapore, 1988).
[47] E. Komatsu et al. (WMAP) (2008), 0803.0547.
[48] D. Baumann, M. G. Jackson, P. Adshead, A. Amblard, A. Ashoorioon, N. Bartolo, R. Bean, M. Beltr´ an, F. de Bernardis,
S. Bird, et al., in American Institute of Physics Conference Series, edited by S. Dodelson, D. Baumann, A. Cooray,
J. Dunkley, A. Fraisse, M. G. Jackson, A. Kogut, L. Krauss, M. Zaldarriaga, & K. Smith (2009), vol. 1141 of American
Institute of Physics Conference Series, pp. 10–120, 0811.3919.
[49] K. M. Smith, A. Cooray, S. Das, O. Dor´e, D. Hanson, C. Hirata, M. Kaplinghat, B. Keating, M. LoVerde, N. Miller, et al.,
ArXiv eprints (2008), 0811.3916.
[50] A. P. S. Yadav, M. Su, and M. Zaldarriaga, Phys. Rev. D 81, 063512 (2010), 0912.3532.
17
[51] W. Hu, M. M. Hedman, and M. Zaldarriaga, Phys. Rev. D 67, 043004 (2003), arXiv:astroph/0210096.
[52] A. Yadav, L. Pogosian, and T. Vachaspati (in preparation).
[53] U. Seljak, U.L. Pen, and N. Turok, Phys. Rev. Lett. 79, 1615 (1997), astroph/9704231.
[54] L. Pogosian and M. Wyman, Phys. Rev. D77, 083509 (2008), 0711.0747.
[55] S. M. Carroll, Phys. Rev. Lett. 81, 3067 (1998), astroph/9806099.
[56] G.C. Liu, S. Lee, and K.W. Ng, Phys. Rev. Lett. 97, 161303 (2006), astroph/0606248.
[57] M. Pospelov, A. Ritz, and C. Skordis, Phys. Rev. Lett. 103, 051302 (2009), 0808.0673.
[58] M. Giovannini and K. E. Kunze, Phys. Rev. D79, 087301 (2009), 0812.2804.
[59] D. Paoletti and F. Finelli, Phys. Rev. D 83, 123533 (2011), 1005.0148.
[60] A. P. S. Yadav, R. Biswas, M. Su, and M. Zaldarriaga, Phys. Rev. D 79, 123009 (2009), 0902.4466.
sfu. we study the eﬀect of a primordial magnetic ﬁeld on the CMB polarization. We have to resort to extensive numerical eﬀorts to compute the window functions. VII. Clαα . on a magnetic ﬁeld in intergalactic space. As we will see. As we show. 15–18].html along with a short Fortran code for calculating ClBB . especially within the Milky Way. and perhaps a measurement ∼ 10−15 G [19].ca/∼levon/faraday. III. V we move on to our main calculation of ClBB with a thick LSS. and obey some other power law for kI < k < kdiss . The work of [13] suggests k1 growth at small k. In the present paper. In Sec. In the thin LSS approximation of Sec. II. Here we ﬁnd that the physics of FR during recombination can be described in terms of “window functions” through which the magnetic ﬁeld spectrum (see Eq. Earlier work on FR of the CMB in cosmic magnetic ﬁelds has largely focused on the eﬀect of a uniform magnetic ﬁeld [20. is primordial. A detection of magnetic ﬁelds in the cosmic microwave background (CMB). when a stochastic magnetic ﬁeld has been considered. focussing on the eﬀect of a stochastic ﬁeld with power on small (subMpc) scales. which we calculate in Sec. IV. Our results are described in Sec. the thin LSS approximation is suﬃcient for order of magnitude estimates. it is necessary to ﬁnd distinctive signatures of FR that occurred at recombination versus that which happened more recently. beyond which the amplitude falls oﬀ very quickly. an observed FR of the CMB could also be due to magnetic ﬁelds along the line of sight. conceivable. we show that ClBB is directly related to ClEE multiplied by the correlation function for the FR. the ﬁnite thickness eﬀects of the LSS were modelled by approximating the visibility function with a Gaussian proﬁle. then k0 behavior until kdiss .2 bk k0 k1 kI kdiss k FIG. (65)) appears in ClBB . In [25]. 2. but can be wrong by factors of a few and in an dependent way. may also be viable. unassociated with cosmic structure. (1) . 1: The Fourier amplitude b(k) – as given by the square root of the power spectrum – of a stochastic magnetic ﬁeld generated during a cosmic phase transition is expected to grow for k < kI where kI is an inertial scale. We have made our window functions publicly available at http://www. which means they are likely to dominate Bmodes from other sources. Hence. The window functions themselves are independent of the magnetic ﬁeld power spectrum and are shown in Fig. 3. induced by a primordial magnetic ﬁeld. When the window functions are convolved with the magnetic ﬁeld power spectrum. STOCHASTIC MAGNETIC FIELDS A statistically homogeneous and isotropic stochastic magnetic ﬁeld is described by the twopoint correlator in Fourier space as ˆˆ ˆ bi (k)bj (k ) = (2π)3 δ (3) (k + k )[(δij − ki kj )S(k) + iεijl kl A(k)] . say based on the expulsion of magnetic ﬁelds from active sources. ∼ 10 Mpc away from the TeV gamma ray source. for example due to Faraday rotation (FR) of the CMB polarization. would unambiguously point to a cosmological origin because there are no confounding magnetized structures at last scattering. a “thin” LSS was often assumed [22–24]. 4 we show the results for ClBB for a scaleinvariant and a blue magnetic power spectrum. In Figs. However. observations of TeV gamma ray sources have been used to place lower bounds ∼ 10−16 G [3. We calculate the Bmode correlator. ClBB . 21] and. we obtain ClBB . but an astrophysical origin. It seems likely that such an intergalactic magnetic ﬁeld. This approximation results in a tremendous technical simpliﬁcation but it is not clear if it is suited to study the eﬀects of a magnetic ﬁeld with coherence scale smaller than the thickness of the LSS. VI. where kdiss is a dissipative scale. in addition to its characteristic frequency dependence. More recently. FR induced Bmodes are only weakly damped on small angular scales (high ). We conclude in Sec.
are real functions of k = k. 0 . even then. S∗ . Beﬀ . kdiss < k n kI The total energy density in the magnetic ﬁeld. (7) (8) 3 For a ﬁxed exponent n. From Eq. the symmetric and antisymmetric magnetic power spectra. with 2n = 2n = 0. Throughout the paper. 0 ≡ 2 Beﬀ . we ﬁrst evaluate the energy density in the magnetic ﬁeld. along with these values. This corresponds to a minimum comoving scale of 1 Mpc. CMB observations at l ≤ 104 will only constrain the combination S∗ kI . 1 S k ∗ kI S(k) = S k ∗ kI 0. the coherence scale is estimated at kpc scales or less [6. 0 (2) We take the form of S(k) to correspond to Fig. we can write 0. as computations at higher l are very expensive. Hence kI may be expected to be 103 Mpc−1 . We now make this point clearer. To then convert the constraint to a bound on the energy density in magnetic ﬁelds requires knowledge of the exponent n and the inertial and dissipation scales. In other words. we use Gaussian CGS units. the large k part of the power spectrum would still enter when we derive a constraint on the magnetic ﬁeld. is ﬁxed by the total energy density in the magnetic ﬁeld. To connect to the amplitude of the power spectrum. 2n−3 . 2n −3 0 < k < kI kI < k < kdiss kdiss < k (3) . This is much larger than kmax and thus the l ≤ 104 modes of the CMB are likely to be aﬀected only by the magnetic ﬁeld modes in the inertial range 0 < k < kI . or kmax ∼ 1 Mpc−1 . Thus. Beﬀ is the ﬁeld strength of a uniform magnetic ﬁeld that has the same total energy density as our stochastic magnetic ﬁeld.1. Our constraints will be written in terms of Beﬀ . and the form of the power spectrum for k > kI plays no direct role. is found by setting k → ∞. We will only consider the eﬀect of magnetic ﬁeld on the CMB for modes with l ≤ lmax = 104 .3 where S(k) and A(k). kI < k ≤ kdiss (5) 16π 3 n 1 + n kI M (k) = S k3 2n kdiss ∗ I 1+ n 16π3 n − 1 . 2n n n kdiss kI 2n −1 . First deﬁne an “eﬀective magnetic ﬁeld”. For magnetic ﬁelds generated at the electroweak phase transition. the CMB signature for l ≤ 104 probes the long wavelength tail of the magnetic . (2) we get 2n 3 k S∗ kI 16π3 n kI . motivated by an inﬂationary mechanism of generation of magnetic ﬁelds [27. The energy density in modes up to some value of k is given by M (k) = 1 8π k <k 1 d3 k 2S(k ) = 3 (2π) (2π)3 k dkk 2 S(k). The results in [13. 28]. 0 = (6) where κ≡1+ and hence Beﬀ = 1 π 3 κS∗ kI . k ≤ kI S k3 2n ∗ I n k −1 . However. in terms of the total energy density in the magnetic ﬁeld. 26] suggest the exponents n = 5/2 and n = 3/2. we will also consider the case of a nearly scale invariant spectrum. 8π (4) In other words. In our analysis. 12]. which is the same as 3 S∗ kI κ . 16π 3 n M (k) for k > kdiss . since the amplitude.
λ0 . ΩBγ (16) 10−7 Gauss Beﬀ . all the scaling due to the expansion of the universe can be pulled out by converting to rescaled quantities Ba2 → B . According to [34. (17) The relation (15) is based on the analysis in Ref. S(k).64 × 10−34 gm/cm3 = 2 × 10−15 (eV) [33]. One can estimate its value for a given amplitude and shape of the magnetic ﬁelds spectrum. 35].25 × 10−6 ΩBγ Gauss . we obtain kdiss ≈ 0. (15) as an approximate expression for the dissipation scale. Note that ργ is the photon density which is diﬀerent from the radiation density at BBN epoch since neutrinos also contribute to radiation. The constraint from BBN [29–32] is best expressed in terms of the magnetic ﬁeld energy density relative to the photon energy density ΩBγ ≡ 0 ργ 10−1 (9) The relative density. (15) where it was useful to know that 1 Gauss = 6. but we will still use Eq. k < kI (11) The power spectrum S(k) will enter the calculation of ClBB in combination with powers of k. [35] introduced a smoothing procedure and split the spectrum into a “homogeneous” part and a “perturbations” part. Note that ∆2 is independent of kI . So it is convenient to introduce the dimensionless “power spectrum” using powers of k and also the wavelength of observed radiation. kdiss is determined by damping into Alfven waves and can be related to Beﬀ as kdiss ≈ 1.43 10−2 h Mpc−1 . Big bang nucleosynthesis (BBN) is sensitive to the total energy density in the magnetic ﬁeld since this is what enters the expansion rate of the universe. indeed. ΩBγ . (4) implies: Beﬀ = 3. [34] where small perturbations on top of a homogeneous magnetic ﬁeld were treated.4 spectrum and not the modes where the bulk of the energy density resides. ργ a4 → ργ . we will use these comoving quantities unless explicitly stated. is independent of cosmological epoch since both magnetic ﬁeld and photon energy density scale as a−4 where a(t) is the cosmological scale factor. our results do show stronger signatures with growing l. ka → k . To extend this analysis to a stochastic magnetic ﬁeld with little power on long wavelengths. kdiss is not an independent parameter. This suggests that it may be favorable to investigate the CMB at yet higher l. (10) and. 0 In principle. 2n ∆2 k 0 k 0 < k < kI 2 I 3λ2 0 2n 2 3 = ∆2 k (12) ∆M (k) ≡ k S(k) kI < k < kdiss 0 kI 16π 2 e 0 k > kdiss where ∆2 ≡ 0 9n ργ λ4 ΩBγ 0 16πe2 κ 4 (13) At the present epoch ργ (t0 ) = 4. Ref. and so ∆2 = 1. We also note that Eq. .1 × 104 0 ΩBγ × κ 2n 5 90 GHz ν0 4 (14) where we denote the observed CMB frequency by ν0 . Finally. It is not clear to us if this procedure is valid for an arbitrary spectrum.4 h1/2 1Mpc−1 Converting this to ΩBγ .9 × 10−20 GeV2 when converting between natural units and CGS units. in what follows. The ﬁnal form of the magnetic ﬁeld power spectrum in the inertial range is S(k) = ΩBγ ργ 16π 3 n 3 κkI k kI 2n−3 .
(22) even though x = n(η0 − η). IV. α= 3 λ2 16π 2 e 0 ˜ τ (x) B · dl ˙ (20) where τ (x) ≡ ne σT a is the diﬀerential optical depth along the line of sight. ne is the number density of free electrons. . the initial time will be taken to be η∗ . FR depends on the free electron density. dl is the comoving length element along the photon trajectory from the source to the observer and we are using Gaussian natural units with = c = 1. (22) as α(ˆ )α(ˆ ) = n n 3λ2 0 16π 2 e 2 k 2 dk S(k) (2π)3 dη dη τ (η)τ (η ) [ˆ · n − ∂kη ∂kη ] ˙ ˙ n ˆ n n ˆ d2 ke−ik·ˆ η e+ik·ˆ η .5 III. a is the scale factor normalized so that atoday = 1. as is conventional.08 8π 2 e ν0 B0 10−9 G 30GHz ν0 2 . (1). the bulk of the rotation is produced during a relatively brief period of time when the electron density is suﬃciently low for polarization to be produced and yet suﬃciently high for the FR to occur. (20) as τ (η) ˆ ˆ ˙ ˙ in the integrand of Eq. (1). 2πm2 e (18) where λ is the wavelength of light. η integrations are from the time that the photon last scattered to the present time. 38] 1 . which we present in Sec. [36] and is given by F = 3 B0 2 ≈ 0. This can also be seen directly by writing n n ˆ ˆ Eq. and this will be related to the twopoint correlation function of the magnetic ﬁeld as given in Eq. 37. The FR correlator will enter the calculation of ClBB in the thin LSS approximation of Sec. In general. the last scattering time will be diﬀerent for diﬀerent photons but. the generation of the CMB polarization and its FR are entangled. So a correlator of FR cannot be sensitive to the helical properties [22. g(η) ≡ τ e−τ . 3m2 e (19) and integrating along the line of sight. The average FR (in radians) between Thomson scatterings due to a tangled magnetic ﬁeld was calculated in Ref. (23) 1 Indirectly though. FARADAY ROTATION CORRELATORS The CMB is linearly polarized and an intervening magnetic ﬁeld will rotate the polarization vector at a rate given by: dα = λ2 e3 a ne B · dl . V. depends on all 3 components of the magnetic ﬁeld. we get α(ˆ )α(ˆ ) = n n 3λ2 0 16π 2 e 2 d3 k S(k) (2π)3 dη n n ˆ ˆ ˆ ˆ dη τ (η)τ (η )e−ik·ˆ η e+ik·ˆ η [ˆ · n − (k · n)(k · n )] ˙ ˙ n ˆ (22) where n. λ0 is the observed wavelength of the ˙ ˜ radiation and B ≡ Ba2 is the “comoving” magnetic ﬁeld. σT = 8πe4 . In this section we will calculate the twopoint correlation functions of the FR angle. In the general case. (11)). α. Therefore. we obtain the Faraday rotation of the polarization angle. Following Ref. We have also ignored inhomogeneities in the free electron ˆ density along diﬀerent directions on the sky since this will only give a higher order correction to FR. (21) where B0 is the current amplitude of the ﬁeld and ν0 is the radiation frequency observed today. which becomes negligible towards the end of recombination. The limits of the integral are from the initial position of the photon to the ﬁnal position. described by the helical power spectrum A(k) in Eq. we have written τ (x) in Eq. the epoch at which the “visibility function”. is maximum. helicity does enter the FR signature because magnetic helicity plays a crucial role in the evolution of magnetic ﬁelds and the exponent n in S(k) (see Eq. FR is sensitive only to the line of sight component of the magnetic ﬁeld. Using the known expression for Thomson scattering crosssection. in which one assumes that all of the polarization was generated at once independently from the FR. [39] without the helical term. ˙ Statistical isotropy implies that α(ˆ )α(ˆ ) must be a function of n · n . whereas magnetic helicity. The limits of the η. n are two directions on the sky. Note that. in the thin LSS approximation.
M αLM YLM (ˆ ) .6 and using n n ˆ d2 ke−ik·ˆ η e+ik·ˆ η = 4π l (2L + 1)jL (kη)jL (kη )PL (ˆ · n ) . and can be easily evaluated numerically using the ionization history from CMBFAST. (31) The function TL (1) can be expressed in terms of T functions by using the relation jL (x) = 1 [LjL−1 (x) − (L + 1)jL+1 (x)] . where αLM are the coeﬃcients in the α(ˆ ) = n L. the correlator of α’s can be expanded into Legendre series α(ˆ )α(ˆ ) = n n L (2L + 1) αα CL PL (ˆ · n ) n ˆ 4π (25) ∗ and the FR correlators can also be written as αLM αL M spherical harmonic decomposition of the FR angle. 2L + 1 (32) This gives TL (1) = (33) The functions TL (k) are independent of the magnetic ﬁeld. As a consequence. IV where we ﬁnd ClBB in the thin LSS approximation. (29) L+1 L jL−1 (kη)jL−1 (kη ) + jL+1 (kη)jL+1 (kη ) − jL (kη)jL (kη ) 2L + 1 2L + 1 In analogy with the way CL ’s are evaluated for CMB. η∗ is the epoch at which the visibility function is maximum and η0 is the present epoch. as deﬁned above. It will be useful in Sec. it is convenient to introduce “transfer” functions TL (k) and (1) TL (k) deﬁned as η0 TL (k) ≡ TL (k) ≡ (1) η∗ η0 dη τ (η)jL (k(η0 − η)) ˙ dη τ (η)jL (k(η0 − η)) ˙ η∗ (30) where. n ˆ (24) where jL (x) are Bessel functions and PL (x) are Legendre polynomials. αα = CL δLL δM M . we can write α(ˆ )α(ˆ ) = n n × Applying the recursion relation (L + 1)PL+1 (x) = (2L + 1)xPL (x) − LPL−1 (x) to the (ˆ · n )PL (ˆ · n ) term in (27) results in n ˆ n ˆ α(ˆ )α(ˆ ) = n n × 3λ2 0 16π 2 e 2 L 3λ2 2 0 k 2 dkS(k) dη dη τ (η)τ (η ) ˙ ˙ 16π 2 e π (2L + 1) [(ˆ · n )PL (ˆ · n )jL (kη)jL (kη ) − jL (kη)jL (kη )PL (ˆ · n )] . (31) is our ﬁnal result for the FR correlation function. However. α(ˆ )α(ˆ ) depends only on n · n as n n ˆ ˆ expected. . and introducing jL (x) ≡ ∂x jL (x). 2L + 1 1 [LTL−1 − (L + 1)TL+1 ] . Hence. n ˆ n ˆ n ˆ 4π 2 (27) (28) 2 π k 2 dkS(k) dη dη τ (η)τ (η ) ˙ ˙ L (2L + 1) PL (ˆ · n ) n ˆ 4π . n (26) Now combining (23) and (24). Eq. the result is not useful in the general case of a thick LSS because then ClBB is not αα simply related to CL . Using these transfer functions in (29) and comparing to (25) allows us to write αα CL = 2 π dk 2 L L+1 (1) ∆ (k) (TL−1 (k))2 + (TL+1 (k))2 − (TL (k))2 k M 2L + 1 2L + 1 .
Therefore. For example. we can rewrite the latter as ClBB = where Wl (k) are “window functions” deﬁned as Wl (k) = 4 l1 L 2 π dk 2 ∆ (k)Wl (k) . we can start with Eq. the Bmode angular spectrum in the thin last scattering approximation is ClBB = 1 π αα (2L + 1)CL L l2 L (2l2 + 1)ClEE (Hll2 )2 . (20) of [41]) which gives the Bmode coeﬃcients Blm = 2(−1)m LM l2 m2 LM L where ξlml2 m2 and Hll2 are deﬁned in terms of Wigner 3 − j symbols as [41] LM L αLM El2 m2 ξlml2 m2 Hll2 . k M (39) (2l1 + 1)(2L + 1) L 2 EE (Hll1 ) Cl1 4π L L+1 (1) [TL−1 (k)]2 + [TL+1 (k)]2 − [TL (k)]2 2L + 1 2L + 1 . we will neglect primordial tensor modes and any actively sourced vector and tensor modes (including those sourced by magnetic ﬁelds) so that only E mode is produced at the instant of last scattering. It eﬀectively assumes that the background Emode polarization on all . some of this E mode will be Faraday rotated into B mode. l L l2 −m M m2 (35) (36) l L l2 2 0 −2 and the summation is restricted to be only over even L + l2 + l. To estimate this eﬀect. 2 (38) It is instructive to put the expressions for the CMB observables in a form that separates the wellestablished physics of FR of CMB polarization from the particular form of the magnetic ﬁeld spectrum. we can derive the expression relating ClBB to ClEE : ∗ Bl m Blm = 4 LM L M l2 m2 l2 m2 LM L LM ∗ ξlml2 m2 Hll2 ξl m l2 m2 HlLl2 αLM El∗ m2 αL M El2 m2 2 = δll δmm 4 L (2L + 1) αα CL 4π L (2l2 + 1)ClEE (Hll2 )2 2 l2 (37) αα which assumes a statistically isotropic stochastic magnetic ﬁelds with a FR angular spectrum CL given by Eq. because of the residual presence of charged particles. (40) They describe the amount of power a given wavelength k of the magnetic ﬁeld spectrum contributes to a given angular scale l of the Bmode polarization spectrum. Transfer functions TL (k) can be found numerically using the diﬀerential optical depth calculated in CMBFAST [42]. substituting (31) into (38). FARADAY ROTATION OF CMB IN THIN LSS APPROXIMATION In the limit of instant last scattering one assumes that all of the polarization was generated at the peak of the visibility function. (34) LM ξlml2 m2 ≡ (−1)m L Hll2 ≡ (2l + 1)(2L + 1)(2l2 + 1) 4π .7 IV. At subsequent times. Having evaluated the window functions once. one can store them and use (39) to calculate ClBB for diﬀerent choices of the magnetic spectrum. (39) relating ClBB to the magnetic spectrum is formally independent of the thin LSS approximation – the approximation is used in the calculation of the window functions. (6) of [40] (same as Eq. EXACT (THICK LSS) TREATMENT OF FARADAY ROTATION The thin LSS approximation decouples the process of generation of the CMB polarization by Thomson scattering from its subsequent FR by magnetic ﬁelds. V. (31). From the above. We note that Eq. Since we are interested speciﬁcally in the FR eﬀects.
(47) with (in what follows we will not need βl ) (0) l (x) (0) ≡ 3 (l + 2)! jl (x) . we can write (45) as η0 η0 P± (q. While this may suﬃce for order of magnitude estimates. n. since the amplitude of the rotation is directly proportional on the rapidly decreasing free ˜ electron density. f = 3λ2 /(2πe). The form of Eq. ˙ where P± (q. n. η) = [Θ2 − 6E2 ]/10. we have √ S± = τ 6P (0) (q. denoted by Wl (k). ±2 Ylm (ˆ ) are spin2 spherical harmonics. η) = Q ± iU . In this Section we derive the exact FR window functions. we can use the total angular momentum formalism of [44] to derive an expression for ClBB in terms of the magnetic spectrum S(k). r = (η0 − η)ˆ . (55) of [44] we have P± (q. polarization in diﬀerent parts of the sky was created at diﬀerent times and any choice of a single time is essentially arbitrary. (44) where ωB (ˆ . 8 (l − 2)! x2 (48) For small ωB . and P (a) (q. η) 1 ˜ ˆ 2i η ωB dη . Furthermore. η) ˙ n ±2 Y20 (ˆ ) 4π . (39) relating ClBB to the magnetic spectrum will be the same. η) = f τ B(r) · n. we used the identity (Eq. vector (a = ±1) and tensor (a=±2) modes. and ˆ ˆ ˆ ˆ √ S± = τ 6 ˙ 2 (41) P (a) (q. n) = ˆ 0 dη s± (q. leading to a new term on the right hand of (41) [45]: ˙ P± + iqµP± = −τ P± ˙ 2iωB P± + S± . In the absence of FR. and λ0 and B are the comoving wavelength and magnetic n ˙ ˆ n 0 ﬁeld strength. Assuming that polarization generated by vector and tensor sources is negligible. (45) with s± = S± e−τ e−iqµ(η0 −η) ˜ √ = −τ e−τ 6P (0) (q. n. (46) and τ ≡ η0 η dη τ . and U into Q. by solving the radiative transport equations for the generation and propagation of CMB polarization in the presence of FR by stochastic magnetic ﬁelds.8 angular scales was created at a single instant in time corresponding to the peak of the visibility function. where Θ2 and E2 are n the quadrupole moments of the CMB temperature and Emode polarization for scalar (a = 0). (49) Next. the amount of the FR strongly depends on the choice of the initial instant. (16) of [44]) ˙ − 4π 5 n ±2 Y20 (ˆ )e iq·ˆ r n = l (−i)l 4π(2l + 1)[ (0) l (qr) ± iβl (qr)] (0) n ±2 Yl0 (ˆ ) . η)e ˜ ˆ 2i η0 η ωB dη . η) a=−2 n ±2 Y2a (ˆ ) 4π . (50) . The ordinary diﬀerential equation (44) has the inhomogeneous solution η0 P± = 0 dη s± (q. In the above. equations for the q Fourier mode of linear polarization parameters Q and U are [44] ˙ P± + iqµP± = −τ P± + S± . µ = q · n. 5 (42) √ (a) (a) (a) (a) In the above. η) ˙ l (0) l (q(η0 (0) (−i)l 4π(2l + 1)[ − η)) ± iβl (q(η0 − η))] n ±2 Yl0 (ˆ ) . The evolution of CMB Stokes parameters is described by Chandrasekhar’s radiative transport equations [43]. 5 (43) FR rotates Q into U . n) = ˆ l (−i)l 4π 2l + 1 l El m=−l (m) (q) ± iBl (m) (q) n ±2 Ylm (ˆ ) . n is the direction of the line of sight. From Eq.
η)uL (k. n)ωB (η . η. (51) where we have assumed that FR is the only source of Bmode. uL = iTL 2L + 1 (58) We can also relate P (0)∗ (q. η ) + TL+1 (k. to stay general. n ˆ (54) where UL (k. η ) . except for the range of the time integration: η0 TL (k. (c) (c) (57) where uL = (1) L (2) TL−1 . we keep the sum to be over all m modes. ˆ ˆ Eq. η ) 2L + 1 2L + 1 (55) and the transfer functions are the same as before. (59) . η ) .9 where q = z . (52) m Introducing g(η) = τ exp(−τ ) and Xll1 ≡+2 Yl1 0 [+2 Ylm ]∗ +−2 Yl1 0 [−2 Ylm ]∗ . η) ˙ +−2 Yl1 0 [−2 Ylm ]∗ ) η η0 (−i)l1 l1 4π(2l1 + 1)[ (0) ∗ l1 (+2 Yl1 0 [+2 Ylm ] ωB dη . (27) for the equal time twopoint correlation of rotation measure is easily generalized to the unequal time FR correlation case above. η)TL+1 (k. η ) − TL (k. n ) . η) ≡ η dη τ (η )jL (k[η0 − η ]) ˙ η0 TL (k. The angular spectrum ClBB can be written in terms (m) of Bl (q) as [44] (2l + 1)2 ClBB = 4π d3 q (2π)3 l Bl m=−l (m)∗ (q)Bl (m) (q) . η)P (0) (q. η ) = L L+1 (1) (1) TL−1 (k. η)TL (k. η. η. n)ωB (η . Hence. η ) to the primordial curvature power spectrum ∆2 (q) via P (0)∗ (q. we have η0 η0 dη η η dη ωB (η . Inverting (50) and using (49) we obtain Bl (m) (q) = × 1 (−i)l 2l + 1 4π η0 dˆ n 0 √ dη τ e−τ 6P (0) (q. ˙ η (56) The UL in (55) can be written as a sum of terms with separated η and η dependencies: 3 UL (k. η)P (0) (q. η ) (0) l1 (q(η0 × m=−l l1 l2 (−i)l1 il2 dˆ n (2l1 + 1)(2l2 + 1) η0 η0 − η)) (0) l2 (q(η0 − η )) (53) × dˆ Xll1 (ˆ )Xll2 (ˆ ) n m∗ n m n dη η η dη ωB (η . However. and substituting (51) into (52) we obtain ˙ (2l + 1)ClBB = 4π l d3 q 6 (2π)3 η0 η0 dη g(η) 0 0 dη g(η ) P (0)∗ (q. η) ≡ (1) dη τ (η )jL (k[η0 − η ]) . vector and tensor ﬂuctuations in the metric. Namely. η)P (0) (q. η ) . (55) of [44] the sum over m runs only from −2 to 2 because these are the only modes ˆ ˆ that can be sourced by scalar. η ) = c=1 uL (k. η)TL−1 (k. Note that in Eq. η ) = q −3 ∆2 (q)P (0)∗ (q. n ) = ˆ ˆ 2 π dk 2 ∆ (k) k M L 2L + 1 4π PL (ˆ · n )UL (k. η)P (0) (q. uL = 2L + 1 L+1 (3) (1) TL+1 . the FR eﬀect on the propagation of photons is not via perturbations of the metric tensor.
(66) is very similar to the thin LSS window function given by Eq. η∗ ) (c) (c) η0 0 √ dη g(η) 6P (0) (q.L mM 2 π dq 2 (c)∗ (c) ∆ (q) dl1 L (q. k)dl2 L (q. k) = (c) η0 0 2 π dk 2 ∆ (k) k M ZlLM ∗ ZlLM 1 0lm 2 0lm l1 . n m ∗ (61) √ dη g(η) 6P (0) (q. (71) . (68) dq 2 (c) ∆ (q) dl1 L (q. we have ZlLM ∗ ZlLM = 4δl1 l2 1 0lm 2 0lm mM (2l + 1)(2l1 + 1)(2L + 1) L 2 (Hll1 ) if l + l1 + L = even. η) 2 (0) l1 (q(η0 − η)) . q c=1 3 (60) (2l1 + 1) dˆ Xll1 YLM . (62) is “factorizable”. we obtain (2l + 1)ClBB = where we have deﬁned ZlLM = (−i)l1 1 0lm and dl1 L (q. ∆ (q) q c=1 3 (67) The thick LSS window function in Eq. k) deﬁned in Eq. and Sl1 L (k) = 2 π dq 2 (c) dl1 L (q. η) (0) l1 (q(η0 − η))uL (k. k)2 q c=1 3 (69) L L+1 (1) [TL−1 (k)]2 + [TL+1 (k)]2 − [TL (k)]2 2L + 1 2L + 1 (70) of the thin case if the function dl1 L (q. η) . k M (65) (2l1 + 1)(2L + 1) L 2 (Hll1 ) Sl1 L (k) . Noting that ClEE = 1 we can see that 2 π of the thick case becomes ClEE 1 (c) 2 π dq 2 ∆ (q) q η0 0 √ dη g(η) 6P (0) (q.10 Putting it all in (53) and integrating over the angular dependence of q. k) . 4π (64) Substituting this into (60) we can write ClBB = with ˜ Wl (k) = 4 l1 L 2 π dk 2 ˜ ∆ (k)Wl (k) . 4π (66) where l + l1 + L = even.l2 . (c) (62) Using the expression for the integral of a product of three spinweighted spherical harmonics dˆ a Yl1 m1 (ˆ )b YLM (ˆ )c Yl2 m2 (ˆ ) = n n n n (2l1 + 1)(2L + 1)(2l2 + 1) 4π l1 L l2 m 1 M m2 l1 L l2 −a −b −c (63) and the orthogonality property of Wigner 3j symbols [46] . η) (0) l1 (q(η0 − η)) . that is if dl1 L → uL (k. k)2 . and 0 otherwise . (40).
0001 1 10 100 l 1000 FIG. M VI. The left and right panels show the same functions plotted on logarithmic and linear axis respectively. it only needs to be done once for a given cosmological model. We have made our window functions. projecting into multipole l1 of Cl1 .5. (62) will in general not be factorizable. is distorted by the kmode of the magnetic ﬁeld projecting onto multipole L of the FR distortion spectrum. ClBB at l ∼ 8000 is sensitive to ∆2 at M k = 0. These results are independent of the spectral features of the magnetic ﬁeld. we modiﬁed CMBFAST to calculate sources Sl1 L (k) on a grid in L and k. 1 Mpc−1 but not to smaller k. publicly available at http://www.sfu. In order to accurately account for magnetic ﬁelds on scales up to a given kmax in Mpc−1 . the sums over l1 and L in (66) are quick to perform. Thus the thick LSS case reduces to the thin LSS case if we disregard the convolution in Eq. the role of the window is to specify the extent to which a given Fourier mode . where the sources were sampled for 450k bins. Instead. (62). For example. functions dl1 L (q. We included the l1 modes up to 6000. evaluated using the ΛCDM model with WMAP7 best ﬁt parameters [47].ca/∼levon/faraday. 2 we show window functions for several values of k as a function of multipole l for the thin (dotted lines) and thick (solid lines) LSS treatments discussed in previous sections. ˜ To evaluate Wl (k) numerically. k) determine the EE relative amount by which a given qmode. The convolution in Eq. and conﬁrmed that it is more than suﬃcient for all k and L because of the exponential suppression of the source Emodes by the Silk damping. Time required for the ˜ evaluation of sources needed for Wl (k) was ∼ 4000 CPUhours. along with a Fortran code that calculates ClBB for a given ∆2 (k). Once the sources are calculated and stored. ˜ Even though computing the exact window functions Wl (k) takes a nontrivial amount of CPU time. one needs to evaluate the source up to Lmax ∼ (104 Mpc)kmax . the diﬀerent q modes of the source Emode (c) polarization are not all created at a single time before FR took place. i.html. These window functions prescribe the way in which a given Fourier mode k of the stochastic magnetic ﬁeld contributes to the multipole l of ClBB .e. 2: Window functions Wl (k) for diﬀerent values of k plotted vs l as evaluated using the full LSS treatment (solid lines) and using the thin LSS approximation(dotted lines). RESULTS In Fig. Instead.001 0.11 0.
73 × 10−7 Gauss. and fcmb = 61 GHz. it comes through the exponential suppression of the source function P (0) (k. except for the lack of exponential damping on small scales. The value of the magnetic ﬁeld energy fraction ΩBγ = 5 × 10−4 corresponds to Beﬀ ≈ 0.1. using the thin (blue dot) and exact (solid red) treatment of LSS. In Fig. Let us now focus on the Bmode spectra for speciﬁc choices of the magnetic spectrum. l2 ClBB ∝ l4 at large l. One can also see the oscillations which come from the acoustic oscillations in the Emode spectrum. More generally. Emode scales produced at diﬀerent times are rotated at diﬀerent rates. For 2n = 0. since the rate of FR depends on the rapidly decreasing free electron density. Comparing the exact (thick LSS) windows with the ones obtained in the thin LSS approximation. This diﬀerence comes because of the assumption made in the thin LSS approximation that all of the Emode was produced at the same time. for n = 5/2.1. In reality. where 104 is roughly the distance to LSS in Mpc. The black dash line shows the input Emode. Note that in this case the shape of the Bmode is essentially a copy of the Emode spectrum.g. there is no exponential suppression of the FR generated small scale Bmode spectrum because the magnetic ﬁeld is correlated on small scales. . The black shortdash line is the input Emode spectrum. Emode is rotated as it is being produced and.12 FIG. kI = 1 Mpc−1 . k of the stochastic magnetic ﬁeld contributes to the multipole l of ClBB . the black dashdot line is the contribution from inﬂationary gravitational waves with r = 0. η) in Eq. both thin LSS and exact spectra have the asymptotic form of l−1 at high l. but diﬀer signiﬁcantly at l away from the peaks. we note that they have comparable shapes and amplitudes near their peaks. the asymptotic exponent is 2n − 1. 3 we consider a nearly scale invariant magnetic spectrum with 2n = 0. One can see that each window has a peak at l approximately given by l = 104 k. The Silk damping dissipates the perturbations in the CMB temperature and Emode on scales smaller than 9 Mpc or so. so that the FR occurs at the same rate on all scales. while the blue dot and the red solid lines show the Bmode spectra obtained using the thin and exact treatment of LSS. 3: The CMB Bmode spectrum from Faraday rotation evaluated in the case of a nearly scaleinvariant magnetic spectrum with 2n = 2n = 0. (43). In our formalism.1. while the black longdash line is the expected contribution from gravitational lensing by large scale structure. One can see that the exactly calculated spectrum favors the power near the peak at the cost of the power around it. which can be relatively large for stochastic ﬁelds e. While the EE correlations are suppressed by the Silk damping.1.
2n = 3 case motivated by causally generated ﬁelds. K and Ka.e. However. ΩBγ = 10−3 . the only damping of the FR induced Bmode power is due to averaging over many random rotations along the line of sight. there are only upper bounds. and 94 GHz respectively. where ∆η is the period of time. the FR produced Bmode can dominate the signal as it keeps growing as l2n−1 . are foreground dominated and we do not include those in our analysis. Still. 4 we show the plot of the Bmode spectrum for the magnetic spectrum index 2n = 5 expected for causally generated magnetic ﬁelds [13]. the power spectrum on small scales. In the ﬁgure we used kI = 1 Mpc−1 . and furthermore. even these weak bounds can produce constraints on the magnetic ﬁeld fraction ΩBγ . comparable to the thickness of LSS. 4: The CMB Bmode spectrum from Faraday rotation in the 2n = 5. we show the Bmode from weak lensing and from gravity waves. Thus. with kI = 1 Mpc−1 and fcmb = 61 GHz. 61 GHz. with the WMAP observed Bmode power spectrum using the χ2 statistics. we can safely assume that its evolution is independent of the perturbations in the photonbaryon ﬂuid. 34] and for k 1Mpc−1 the magnetic ﬁeld can be treated as a stiﬀ source. the bounds will improve rapidly as CMB observations are made on smaller angular scales. 3. (65). We consider three WMAP frequency bands Q. ΩBγ . Note that at small angular scales (high l). while the exact (thick) LSS calculation is shown with solid red. The thin blue dot line shows the thin LSS calculation.13 FIG. and W corresponding to frequencies 41 GHz. V. during which FR is eﬃcient. which is the square of the FR amplitude. In Fig. We derive current constraints on the magnetic ﬁeld energy density. For comparison. The value of the magnetic ﬁeld energy fraction. The other lines are the same as in Fig. the dissipation scale of (weak) magnetic ﬁelds at decoupling time is typically much smaller [13. i. from the WMAP 7year polarization data by comparing magnetic ﬁeld induced theoretical CMB Bmode power spectrum C BB as given by Eq. At present. Then. The functional form of this suppression is a power law and the exponent can be estimated by observing that a random superposition of N perturbations along √ the line of sight leads to a statistical reduction in the amplitude of the observed anisotropy by a factor 1/ N . corresponds to Beﬀ ≈ 10−7 Gauss. Bmode have not been detected. We combine . For wavenumber k we have N ∝ ∆ηk. The lower frequency bands. is suppressed by 1/k. which translates into the 1/l suppression of the angular spectrum. as well as the Emode spectrum.
(14). (7) and depends on the dissipation scale.L. 5: Constraints from WMAP7 year data on the magnetic ﬁeld density ΩBγ . when evaluating the likelihood. The maximum multipole considered in the analysis is = 700. We consider two choices of magnetic spectrum.1 Mpc−1 ). (15) (black solid line).1. ClBB is independent of kI at very small kI because for positive n the integral in Eq. and ignores the possibility of other sources of Bmodes such as inﬂationary gravitational waves [48]. [50. the case of a nearly scaleinvariant spectrum 2n = 2n = 0.1. 0 which is proportional to ΩBγ /κ. we restrict the maximum value of ΩBγ to the one given by Eq (16) for kI = kdiss . The BBN constraints on the magnetic ﬁeld density ΩBγ < 0. 2n = 3 (dashed red line). However. 2n = 3 case. ΩBγ . weak gravitational lensing of the CMB [49]. or analogously magnetic ﬁeld eﬀective amplitude Beﬀ as deﬁned by Eqs. and the other corresponding to causal magnetic ﬁelds 2n = 5. (65) is dominated by the contribution from k kI . 5 we present the 95% conﬁdence level (C. and the energy fraction in the magnetic ﬁeld.). we note that the CMB FR signature constrains ∆2 deﬁned in Eq. 51] for examples of such distortions).14 FIG. (16). For our analysis we considered two choices of theoretical magnetic spectral indices (n. diﬀerent frequency channels directly when evaluating the χ2 and. In the 2n = 5.1 is shown by the dotted blue line. κ is a function of the inertial scale. We used WMAP data for > 32. The bound in each case is presented as a function of kI . To obtain bounds on the magnetic ﬁeld. and the causal case. above which the errors for individual ’s can be treated as uncorrelated. Therefore for given values of n and n . kI . . n ). which can also be related to ΩBγ by using Eq. 2n = 2n = 0. 2n = 5. as a function of the inertial scale wavevector kI . Now κ is deﬁned in Eq. one corresponding to nearly scaleinvariant spectrum 2n = 2n = 0. bounds on ΩBγ for two choices of (n. (4) and (17). Our analysis assumes that the magnetic ﬁeld is the only source of the Bmode signal.1 (dashdot red line) and causal magnetic ﬁelds with 2n = 5. In Fig. n ): the nearly scaleinvariant case. The dependence of ClBB on kI becomes stronger when kI gets closer to the range of scales with signiﬁcant Emode power (k ∼ 0. At large kI the bound is set by the theoretical relation between ΩBγ and kdiss ≥ kI given by Eq. The dependence of the constraint on ΩBγ on kI in each case can be readily understood. kdiss . 2n = 3. 2n = 3. which is encoded in the shapes of the window functions. and several other distortions of primary CMB along the line of sight (see Refs. which we treat as an unknown parameter in the model.
Namely. However. which is independent of the CMB data. on scales smaller than the Silk damping scale. thus generating nonGaussianity which can be seen in the CMB trispectrum [40. topological defects [53. the constraints from WMAP 7year data are 2orders of magnitude better than the BBN constraints. and other participants of the Primordial Magnetism Workshop. LP is supported by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada. For a causal magnetic ﬁeld with (2n = 5. This study is especially timely in view of upcoming and next generation of CMB observations that are focused on measuring the polarization of the CMB. the power of these inhomogeneities is suppressed as 1/l due to random superposition of multiple small scale rotations along the same line of sight. The window functions are independent of the details of the magnetic ﬁeld spectrum and only need to be computed once for a given cosmological model. CONCLUSIONS Primordial stochastic magnetic ﬁelds may be produced in the early universe during baryogenesis. We also refer the reader to [59] for the most recent CMB bounds on magnetic ﬁelds based signatures other than FR. inﬂationary gravitational waves. WMAP is unable to probe the dependence on kI .ca/∼levon/faraday. In the scaleinvariant limit. The growth l2 ClBB ∝ l2n−1 at large l is also characteristic of a magnetic ﬁeld and will help discriminate primordial FR from other foreground contamination. 5 stays almost horizontal even at kI ∼ 0. along with a Fortran code that calculates ClBB for a given magnetic spectrum. 5. which determine the relative amount that a given Fourier mode k of the magnetic ﬁeld contributes to the multipole l of ClBB . 5. We thank Karsten Jedamzik. 2n = 3). For instance. We will present forecasted bounds from future CMB data in an upcoming paper [52]. However. FR of the CMB due to magnetic ﬁelds has a distinct frequency dependence (∼ ν 2 ) which will allow it to be distinguished from other sources of CMB Bmodes such as lensing. −1/2 The condition kI ≤ kdiss ∝ ΩBγ imposes its own upper bound on ΩBγ at large kI . Tina Kahniashvili. and acknowledges hospitality at Perimeter . and is a consequence of the constraint arising from dissipation. In this paper we have calculated the eﬀect of primordial magnetic ﬁelds on the CMB polarization. FR of existing Emodes of the CMB can generate parity odd Bmodes whose amplitude will depend on the observation frequency bands. but future observations of CMB polarization at smaller angular scales will signiﬁcantly improve the constraints on magnetic ﬁeld.1. FR creates E and Bmode inhomogeneities on subSilk scales by rotating diﬀerent parts of the homogeneous Emode patch in diﬀerent random ways. the constraint curve becomes independent of the CMB constraints. For comparison we have also shown the constraints on magnetic energy obtained by BBN. As explained in the previous section. and Richard Battye and Yun Li for pointing out several inconsistencies in the earlier version of the paper. We have summarized current constraints on magnetic enegy ΩBγ from WMAP7 based on the FR induced Bmode spectrum in Fig. at ASU for input. The curve kI = kdiss is also shown in Fig. the Emode map is essentially homogeneous. and the constraint line in Fig. In a followup paper [52] we discuss the detectability of primordial magnetic ﬁeld using such nonGaussian features in the CMB. We note that a CMB experiment which can measure Bmodes at ∼ 1000 would be sensitive to changes in kI in the case of a causal spectrum. We generalized the signatures previously obtained assuming “thin” last scattering surface by solving the full CMB radiative transport equation with FR taken into account. Although in this paper we have focused on Bmode power spectrum of CMB generated due to FR of the CMB polarization. Comparing our full treatment with the thin last scattering approximation.html. 54] and FR due to pseudo scalar ﬁelds [55–58]. Thus. 5. we ﬁnd the respective Bmode spectra have similar shapes but their magnitudes at a given multipole may diﬀer by as much as a factor of a few depending on the primordial magnetic ﬁeld spectrum and the angular scale of observation. We have evaluated them in the best ﬁt ΛCDM model and made them available at http://www. VII.15 this happens to be just under the maximum wavenumber constrained by WMAP’s measurement of BB. especially Eiichiro Komatsu and Michael Nolta. BBN constraints are weaker only by a factor of few. at suﬃciently large kI . The wellestablished physics of FR of CMB can be encoded in window functions. 41. 2011. For a scale invariant magnetic ﬁeld. the bound is eventually dominated by the kI = kdiss curve. for providing the WMAP Bmode power spectra for individual frequency bands. 60]. Characterizing such primordial magnetic ﬁeld is extremely valuable for probing early universe physics. FR can take a smooth Emode ﬁeld and distort it to create Bmodes on arbitrarily small scales. which is the 2n = 2n = 0. Acknowledgments We are grateful to the WMAP collaboration. the CMB constraint is independent of kI because all k dependence eﬀectively disappears. At large kI .1 case in Fig.sfu. Wl (k). recently it has been shown that the stochastic FR of polarization of the CMB couples diﬀerent CMB angular modes. note that the constraints from BBN are not expected to improve much in future.
White. Zaldarriaga. Rev. D54.3653. Turner and L. Fraisse.J. 1841 (1997). Vovk. 391. M. Astrophys. Phys. Lond. A. Komatsu et al. Kandus. E. Astrophys. Phys.Rev. 73 (2010). Rev.Rev. and A. Rubinstein. J. GarciaPerez.1286. A. M. Ghisellini. R. D 55.1687. 835 (2003). O. G. 023009 (2009). astroph/9601055. J. Kahniashvili. V. Trans. A. Dent. 727. Y. 649 (1969). Kahniashvili. Kunze. Baumann. D78. Cooray. Maravin. Phys. New York. Vachaspati. Starkman. Ashoorioon. Vovk. Lett. D. edited by S. 87. D57. M. A. Quantum theory of angular momentum (World Scientiﬁc. D82. A. 0905. Cornwall. Bartolo. Rev. 4714 (1996). 775 (2002). Neronov (2011). Phys. and A. Rev. Grasso and H. Pogosian. (WMAP) (2008). in American Institute of Physics Conference Series. . Astrophys. V. 1012. Hu and M. K. Phys. Rev. Harari. D37. Rev. 469. Rev. Phys. 73 (1996). and T. ArXiv eprints (2008). M. J. D. P. de Bernardis. Tevzadze. arXiv:astroph/9509126. Katalinic. and A. A. and J. Jackson. and B. We also acknowledge the use of cluster computing at the IAS. Smith. 2003). Ando and A. 083005 (2010). W. Vachaspati. Seljak and M. Phys. and B. A. S. Rev. S.l. D. Kosowsky. Nature 222. Phys. TV is supported by the Department of Energy at ASU. E.P. 1 (1996). N. M.3916. Cooray. 7207 (1996). astroph/0409767. Hirata. Rev. Ensslin and C. 1101. Caprini. 063512 (2010).1782. J. Soc. and L. 596 (1997). Neronov and I. Phys. 1103. M. J.Lett. Kaplinghat. 251302 (2001). Kunze. Winitzki. O’Connell. D.. Krauss. A. Lett. T. J. J. Loeb. Dodelson.S. 1009. E. 123002 (2002). arXiv:astroph/9608098. N. Campanelli. D55. R. Gluscevic. 0811. Dolag. 1007. F. Baumann. Jedamzik and G.4817. Amblard. L. 505. and M. M. W. Das. D. Zaldarriaga. Astrophys. Phys. D 81. Lett. M. Ng and T. D 70. 023008 (2010). Chandrasekhar. 101. K. 0912. Mod. A. M. Jedamzik. 1006. 1960). 1009. 023510 (2009). 1009. M. K. Vachaspati. et al. K.3919. D66. 401. Vachaspati. B265. P. Roy. S. Rev.2094. LoVerde. GonzalezArroyo. 1001. A. B379. Astrophys. Dunkley. A366. Rep. Radiative Transfer (Dover.3891. astroph/9608098. Durrer and C. 0806. Ferrer. J. 1005. n Singapore. 1841 (1997).16 Institute for Theoretical Physics where part of this work was completed.3835. 1 (2004). R. K.J. E. S.Rev. Rev. Vachaspati. ArXiv eprints (2010). T. T. Phys. F. F. Vachaspati. Pandey. D. Bonnoli. Keating. T. 0811. 023504 (2004). Cooray. Sigl. M. V. D71. Astron. 043006 (2005). Jedamzik. 063007 (2009). 2743 (1988). and S. and F. Harari. 083502 (2002). Ratra.1924. G. 0803. Astrophys. L. D. D82. Olinto. G. K. G. S. Rev. Hayward. Vachaspati. 469. 0802. 6146 (1997). Achucarro. M. Ratra. e ArXiv eprints (2008). D 80.3380. Phys. T.1876. M. Ichiki.2207. Banerjee and K. D56. J. et al. Moskalev. Zaldarriaga. Hayward. astroph/9606163. arXiv:astroph/0410032.3532. R. D. and S. astroph/0101261. Astrophys. Copi. Jackson. Rev. 1 (1996). Phys. Phys. 3264 (1998). M. astroph/0302426. ArXiv eprints (2011). G. Y. Kusenko. Dolgov. Rev. Phys. A. astroph/0405420. A. Smith (2009). Kamionkowski. 616. S. U. astroph/9603033. 258 (1991). Foschini (2010). C. arXiv:astroph/0112536. 469. 010 (2003). Rev. Modern Cosmology (Academic Press. A.. Matese and R. D69. Kogut.J. L. G. Science 328. K. T. 1141 of American Institute of Physics Conference Series. Princeton (IAS) for hospitality. 0812. D 54. Kunze. gratefully acknowledges funding support from NASA award number NNX08AG40G and NSF grant number AST0807444. T. astroph/0305059. Tsagas. Chu. A. S. J. Kosowsky. Rev. and A. D79. hepth/9704022. Dodelson. hepph/0304285. 722. 123003 (2004). 0804.4794. Kosowsky and A. L. A. astroph/9601055. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] L. L1 (1992). and R. Bird. Kherso´skii. Vogt. I. S. San Diego. and V. D. astroph/9602055. Widrow. G. L4 (2011). D. L39 (2010). Dor´. A. Takahashi. Schramm. Rev. 1 (2011). Hanson. M. Astrophys. J. Phys. A. and A. Kernan. V. M. Phys. 74. Bean. Ostapchenko. and M. B. JCAP 0311. Giovannini and K. Phys. Inoue (2011). pp. Miller. Olinto. Tavecchio. Phys. Adshead. J. Kamionkowski. Taylor. Phys.Y. Villante. and A. Phys. astroph/0202424. Giannotti.1533. Yadav. 0810. K. 171302 (2008). K. astroph/9606080.1048. Cheng. Varshalovich. T. a S. vol. Phys. and is grateful to the Institute for Advanced Study.0932. B. D80. N. M. Zaldarriaga. D. Rev. 437 (1996). J. 2011. GarciaBellido. Kosowsky and A. A.0547. 2915 (2008). and C. 0801. Sigl. M. Zaldarriaga. D. Lavrelashvili. Kachelriess. astroph/9702170. Y. D 65. P.Rev. Mori. Giovannini and K. M. A. M. J. C. D. 023010 (2008). Phys. Kahniashvili. Phys. A. 1988). Su. Rev. Tomas. D56. Beltr´n. 10–120. Widrow. Truran. Phys. D. Loeb. K. and M.3504. Phys. T. Ratra. Vachaspati. N. G. G. & K. B. Sethi. Phil. and T.
Phys. S. 087301 (2009). Phys. P. 103. Pogosian and M. 0902. D 79. and T. Skordis. Zaldarriaga. Carroll. and K.4466. Finelli. and N.C. 3067 (1998). Lett. Rev. astroph/9806099. 123533 (2011). 1615 (1997). Pospelov. Rev. 123009 (2009). 79. R. 0711. Seljak. Phys. . arXiv:astroph/0210096. Liu. Ng. Rev. Hu. 0812. M. 043004 (2003). Kunze. Rev. U. 1005. Lett. A. M. E. astroph/9704231. Ritz.0747. 051302 (2009). D 67. S. Biswas. Pen. and C.L. Yadav. U. M. Lett. 161303 (2006). Lee. 81. Phys.2804. D77. D79. M. 0808. Rev. A. Wyman. A. G. 083509 (2008). Phys.0148. Rev. Rev. M. Phys. and M. Rev. Lett. Rev. Pogosian. S. D. Phys. Vachaspati (in preparation). astroph/0606248. D 83. and M. Yadav. Hedman. Zaldarriaga.W. Giovannini and K. Su. L. Paoletti and F. L. Phys. Turok.17 [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] W. 97. M. Phys.0673.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.