Limits on a Stochastic Background of Gravitational Waves from Gravitational Lensing
Extracted source text ยท arXiv astro-ph/9606065
\documentstyle[11pt,amssym]{article}
\textwidth=6.9in
\textheight=8.4in
\topmargin=0in
\oddsidemargin=0in
\evensidemargin=0in
\parindent=1.2em
\parskip=0.5ex
\columnsep=0.5in
\hoffset=-0.2in
\tolerance=600
\begin{document}
\newcommand{\xV}{\vec{x}}
\newcommand{\tO}{\tau_{0}}
\newcommand{\AT}{A_{\rm T}}
\newcommand{\oS}{q_{S}}
\newcommand{\tS}{\tau_{S}}
\newcommand{\tL}{\tau_{L}}
\newcommand{\kV}{\vec{k}}
\newcommand{\Mtil}{\widetilde{M}}
\newcommand{\Hpres}{\mbox{$100h\,{\rm km\,sec^{-1}\,Mpc^{-1}}$}}
\renewcommand{\thesection}{\Roman{section}}
\begin{center}
{\Large \bf LIMITS ON A STOCHASTIC BACKGROUND OF GRAVITATIONAL
WAVES FROM GRAVITATIONAL LENSING}
\end{center}
\begin{center}
Rennan Bar-Kana\footnote{Electronic address: barkana@arcturus.mit.edu}
{\it Department of Physics, MIT, Cambridge, MA 02139}
(Received 10 June 1996; revised manuscript received 6 August 1996)
\end{center}
\vspace{.2in}
\begin{quote}
\begin{center} {\bf Abstract:} \end{center}
We compute the effects of a stochastic background of gravitational
waves on multiply imaged systems or on weak lensing. There are two
possible observable effects, a static relative deflection of images
or shear, and an induced time dependent shift or proper motion. We
evaluate the rms magnitude of these effects for a COBE normalized,
scale-invariant spectrum, which is an upper limit on spectra
produced by inflation. Previous work has shown that large-scale
structure may cause a relative deflection large enough to affect
observations, but we find that the corresponding effect of gravity
waves is smaller by $\sim 10^4$ and so cannot be observed. This
results from the oscillation in time as well as the redshifting of
the amplitude of gravity waves. We estimate the magnitude of the
proper motion induced by deflection of light due to large-scale
structure, and find it to be $\sim 10^{-8}$ arcsec per year. This
corresponds to $\sim 50$ km/s at cosmological distances, which is
quite small compared to typical peculiar velocities. The COBE
normalized gravity wave spectrum produces motions smaller still by
$\sim 10^2$. We conclude that light deflection due to these
cosmological perturbations cannot produce observable proper motions
of lensed images. On the other hand, there are only a few known
observational limits on a stochastic background of gravity waves at
shorter, astrophysical wavelengths. We calculate the expected
magnitudes of the effects of lensing by gravity waves of such wavelengths,
and find that they are too small to yield interesting limits
on the energy density of gravity waves.
\end{quote}
\vspace{.2in}
\section{Introduction}
Events in the early Universe may have left a stochastic
background of gravitational waves (GW's). In particular, a
generic prediction of inflation is a relic spectrum of
GW's \cite{AbbWs}. Detecting these elusive remnants
would not only establish this prediction of general
relativity, but also serve as a critical test for
inflation. While the predicted background may be
too weak for direct detection \cite{me}, it could be
detected indirectly through its effect on light
propagation in the Universe. Even if the effects of GW's cannot
be distinguished observationally from other effects,
observers who assume no GW's might reach incorrect conclusions about
the distribution of matter in the Universe.
Gravitational lensing is one of the most promising methods
of mapping the distribution of matter at cosmological distances.
Detailed observations of multiple images of quasars have been
used to try to reconstruct the lensing mass distribution (e.g.,
\cite{falco}). It has also long been recognized that
measurements of the time delay between images can be used to
determine the Hubble constant \cite{refsdal}.
Gravitational lenses and sources, however, typically lie at
significant redshifts. Light rays are thus deflected by
large-scale structure (LSS) and GW's as they traverse the
cosmological distance to the observer, and these deflections may
change the simple lensing picture.
GW's may be produced by many sources. Astrophysical sources, such
as close binary systems which include a neutron star or black
hole, radiate GW's, and numerous individual sources may
superpose to create a stochastic background. At the Planck time,
quantum fluctuations in the metric are significant and may produce
gravitons. Phase transitions in the universe may lead to topological
defects such as cosmic strings, which generate GW's. A period of
inflation may leave behind a significant amount of GW's. Whatever the
source, any spectrum which extends over wavelengths comparable to
the present horizon
would contribute to the quadrupole anisotropy of the
cosmic microwave background (CMB) \cite{infl}. Such a spectrum
is therefore limited by the anisotropy measured by the
COBE DMR experiment \cite{COBE}. For our calculations
we adopt a scale-invariant primordial spectrum, i.e., one
which has constant energy density per logarithmic frequency,
which we assume produces the entire measured quadrupole
anisotropy. Inflationary models predict slightly
tilted spectra which are responsible only for some fraction
of the anisotropy \cite{infl,me}, and so are generally
weaker than our adopted case.
In inflation, GW's are produced in conjunction
with density fluctuations. The initial nearly scale-invariant power
spectrum of density fluctuations evolves as modes reenter
the horizon after inflation, and as
structure later forms in a universe dominated by dark matter.
The present spectrum is strongly constrained by galaxy and cluster
surveys, and can be used to study the effects of LSS
on lensing. The induced effects are small but potentially
observable. In weak lensing, the effect is a coherent distortion
of background galaxies by an ellipticity of the order of a few per cent
\cite{weak,kaiser}. In strong lensing, the primary effect is an
external shear which may be
significant for observed four-image systems \cite{me2,seljak}.
In general, the influence at a given time of a weak metric
perturbation on
light propagation is simply described by two effects. Their
magnitudes were
estimated for LSS in Ref.\ \cite{seljak}, which we summarize here.
The first effect is a constant deflection, the same for all nearby
light rays.
This deflection simply displaces the ``true'' angular position of
an observed lens
or source, and is not directly observable. In the case of LSS,
deflections from coherent structures of size $\sim 1$ Mpc
add up in a random walk,
giving an overall deflection of order a few arcminutes at redshift
$1$, which scales as the square root of comoving distance $r$. The
second effect
is a {\it relative} deflection between nearby light rays, which
produces a
focusing and shear with observable effects on weak and strong
lensing.
For two rays at initial angle $\theta$, each coherent structure at
a distance
$r$ causes a relative deflection proportional to their separation
of $\approx r \theta$. The additional random walk gives a relative
angular fluctuation
of $\approx 0.07\ \theta$ at redshift $1$, which scales as
$r^{3/2}$.
It was suggested in Ref.\ \cite{allen} that gravity waves could
significantly
affect the time delays in a multiply imaged system. It was later
pointed
out \cite{surpi} that a correct analysis must include the lensing
constraint,
i.e., the fact that image rays in the presence of GW's follow
different
paths than for no GW's, so that all rays go from the source to a
common destination, the observer. These later authors also
showed that both LSS and GW's have no observable effects on lensing,
to lowest order. However, in their lowest order expansion they
assumed that two image rays that are observed at an angular
separation $\theta$
are separated by a distance of exactly $r\theta$ on the lens plane
at a distance $r$. In other words, they neglected the relative
deflection between light rays, and therefore only included an
overall, constant deflection due to LSS or GW's.
We can easily see why this assumption leads to no observable
effects. In the absence of metric perturbations, we can write the
lens equation for a thin lens as (e.g., \cite{schneider})
$\vec{\beta}=\vec{\theta}-\vec{\alpha}_{\rm lens}
(\vec{\theta})$, where $\vec{\theta}$ and $\vec{\beta}$ are the
image and source angles, respectively, and $\vec{\alpha}_{\rm lens}$
is the scaled deflection
angle, which is determined by the mass distribution of the lens.
If we neglect relative deflections, then LSS or GW's can only cause
an angular shift $\vec{\alpha}_{L}$ between the observer and the
lens, and a shift $\vec{\alpha}_{LS}$ between the lens and the
source. Then the
lens equation becomes $\vec{\beta}=\vec{\theta}-\vec{\alpha}_{\rm lens}
(\vec{\theta})+\vec{\alpha}_{S}$, where $\vec{\theta}$ is now
measured relative to the {\it observed} (and shifted) lens position,
and $\vec{\alpha}_S$ involves $\vec{\alpha}_{L}$ and $\vec{\alpha}_
{LS}$ (see Sec.\ III for the full details).
The constant (i.e., $\vec{\theta}$-independent) deflection
$\vec{\alpha}_{S}$ has no effect on any observables of the lens system
(e.g., \cite{schneider}), since
$\vec{\beta}$ is not directly observable . Fermat's principle then
implies that
the lens equation must be equivalent to $\partial \Delta t/\partial
\vec{\theta}
=0$ at fixed $\vec{\beta}$, where $\Delta t$ is the relative time
delay. There is
thus no observable effect on the time delay, either, since it can
be derived from the lens equation, up to (unobservable)
$\vec{\theta}$-independent terms.
This approximation of neglecting the relative deflection may not be
a good one. Indeed, such deflection can have observational
consequences, which may be sufficiently large to detect in the case
of LSS \cite{me2,seljak}. In this paper, we compute the rms total
and relative deflections between light rays induced by a
scale-invariant stochastic background
of GW's. Unlike LSS, GW's oscillate with time, and so the effect of short
wavelength modes does not amplify, as light rays deflect one way
in crests and the
opposite way in troughs. In addition, the energy density and thus
also the amplitude
of sub-horizon GW's redshift away as the universe expands. The lensing
effect is thus
dominated by wavelengths on the scale of the distance to the source.
Each such mode
acts as a single coherent structure, and so both the
total and relative deflections due to GW's scale approximately
linearly with distance. The effect of different modes must be
convolved with a particular power spectrum and include the
above-mentioned
decay of each mode as the universe expands. We find simple
integral expressions for the scale-invariant spectrum. The
total and relative deflections are smaller than those caused by
large-scale structure by factors of the order of $10^{2}$ and $10^{4},$
respectively. We do not need to explicitly set up the lens equation,
since the rms shear in the lens equation is directly related to
the rms relative deflection of light rays, which we calculate.
This fact was demonstrated for LSS in Ref.\ \cite{me2}, and we give
a general proof in Sec.\ III below. Our results imply that the static
effects of the GW spectrum on lensing are negligible compared to
those of LSS, and cannot be detected in practice.
In addition to the static effects of LSS and GW's on lensing, it
is possible that the fluctuation in the induced deflection with time
would be directly manifested as an observed proper motion of images.
In other words, the sources do not really move but the light rays
from the sources are deflected and so the sources appear to move.
We find that even LSS can only produce motions of order $10^{-8}$
arcsec per year from this effect.
This corresponds to $\sim 50$ km/s at a distance of a Gpc,
and the effect of GW's is smaller still by a factor of $\sim 10^{2}$.
Since typical peculiar velocities are much larger, the proper motion
induced by deflection of light due to LSS is unobservable, and the
same is true for the COBE-normalized scale-invariant spectrum of GW's.
However, we may try to use shear or proper motions of imaged sources
to improve
existing limits on stochastic GW's at a range of astrophysical
wavelengths. There are only a few such limits known:
Single-pulsar timing yields $\Omega_{\lambda}<1 \times 10^{-8}$ at
$\lambda
\approx 2$ pc \cite{latest,kaspi}, binary pulsar timing implies
$\Omega_{\lambda}< 0.04$ over $\lambda \approx 2$ pc to 1 kpc and
$\Omega_{\lambda}< 0.5$ up to $10$ kpc \cite{latest,Rees}, and
the observed angular correlation function of galaxies sets a limit
of $\Omega_{\lambda}<
10^{-3}$ over $\lambda \approx
100$ kpc-$100$ Mpc \cite{linder}. These limits apply to any
stochastic background of GW's,
whether cosmological in origin or generated at low redshift as a
superposition of many discrete sources.
For a cosmological spectrum that existed at early times, there are
also big bang
nucleosynthesis constraints of $\Omega_{\lambda}< 10^{-4}$ for
$\lambda < 100$ pc
\cite{Carr} and CMB limits of $\Omega_{\lambda}< 10^{-12}$ at
Horizon wavelengths (from COBE) and $\Omega_{\lambda}< 10^{-8}$
for $\lambda > 1$ Mpc from small-scale anisotropy \cite{linder2}.
In Ref.\ \cite{linder} it was suggested that highly magnified lensed
sources could increase the sensitivity to detecting proper motions
due to GW's. The angular deviations
induced by GW's produced by an individual source were discussed in
Ref.\ \cite{fakir}. Ref.\ \cite{pyne} considered detecting proper
motions (of unlensed sources) due to GW's through VLBI measurements,
but our approach is simpler than theirs.
For an image of a lensed source, only an angular deflection of
the source {\it relative} to the lens is easily observed, and we
find that this relative motion is small when we assume an
isotropic GW background. Thus we do not find an
interesting limit on the energy density.
\section{Formalism}
In this section we review the formalism describing gravity waves,
their cosmological evolution, and their effect on lensing, as
well as the usual formalism of gravitational lensing.
We work in the framework of a flat Robertson-Walker metric
with small-amplitude tensor metric fluctuations. For weak
perturbations, we can consider the effect of GW's without including
LSS, since the cross terms between them would be of higher order.
In comoving coordinates we can write the line element as
\begin{equation}
ds^2=a^2(\tau)[-d\tau^2+(\delta_{ij}+h_{ij}) dx^{i} dx^{j}]\ .
\label{metric}
\end{equation}
Here $\tau$ is the conformal time, $a(\tau)$ the expansion factor,
and we have set $c=1$. We expand the metric perturbation in plane
waves ($k=2\pi / \lambda$),
\begin{equation}
h_{lm}(\xV,\tau)=\int d^{3}k~ h^{n}(\kV,\tau)\epsilon_{lm}^{n}
(\widehat{k})\,e^{-i\kV \cdot \xV},
\end{equation}
where $\epsilon_{lm}^{n}$ is the polarization tensor which depends
on the direction $\widehat{k}$ ($l$ and $m$ are
spatial indices ranging from 1 to 3, while $n$ ranges over the
polarization components $+,\times$). For a wave
propagating in the $z$-direction, the nonvanishing components
are in the $x$-and $y$- rows and columns: $$\epsilon_{lm}^{+}=
\left( \begin{array}
{rrr} 1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & 0 \end{array} \right),\
\epsilon_{lm}^{\times}=\left( \begin{array}{rrr} 0 & 1 & 0 \\ 1 & 0 & 0
\\ 0 & 0 & 0 \end{array} \right)\ .$$ For
other propagation directions $\kV$, we rotate $\epsilon \rightarrow
R \epsilon R^{{\rm T}}$, with $R$ the standard $3 \times 3$ rotation
matrix.
GW's with a given wavevector $\kV$ are produced during inflation and
then stretched outside the horizon. The amplitude is constant
outside the horizon, but once a mode reenters its energy redshifts
as $a^{-4}$. For the inflationary spectrum the effect of very short
wavelength modes is negligible, and so we can assume that all modes enter
during the matter-dominated era, for which the exact time evolution is
given \cite{AbbWs,me} by a spherical Bessel function,
$3 j_{1}(k \tau)/(k \tau)$. This time evolution is also correct for
all modes long after matter-radiation equality. Inflation
produces Gaussian, stochastic perturbations. The Fourier components
have zero ensemble mean and a covariance
\begin{equation}
\label{frpres}
<h^{i}(\kV,\tau_{1}) h^{j}(\vec{q},\tau_{2})>=\AT k^{-3} \left[
\frac{3 j_{1} (k\tau_{1})} {k \tau_{1}} \right] \left[
\frac{3 j_{1}(k\tau_{2})}
{k \tau_{2}} \right] \delta^{3}(\kV+\vec{q})\delta_{ij},
\end{equation}
for the scale-invariant $k^{-3}$ spectrum. Note that we do not
assume the short-wavelength approximation $h^{i}(\kV,\tau) \propto
a^{-1}(\tau)~ e^{i k \tau}$. The contribution to $\Omega$ at the
present (averaged over several periods) is
\begin{equation}
\label{ompres}
\Omega_{\lambda}=\frac{d\Omega_{\rm GW}}{d \ln k}=\frac{3\pi}{2}\AT
(k\tO)^{-2}\ ,
\end{equation}
where $\tau_{0}=2 H_{0}^{-1}$ is the present value of $\tau$, and
throughout we set $H_{0}=75\,{\rm km\,sec^{-1}\,Mpc^{-1}}.$
Normalization to the full CMB quadrupole anisotropy gives $\AT =6
\times 10^{-11}$.
Consider a photon emitted from a source toward
an observer at the origin, with the photon's final direction defined
as (minus) the $z$-axis. We use $r$ to denote values of
the $z$-coordinate (with $z_{S}$ denoting the source redshift, not
its $z$-coordinate).
GW's affect the distance-redshift relation, but this effect is
separate from that
of the angular deflections which we are interested in, and it
introduces only small
additional corrections in these quantities \cite{linder}.
We can thus neglect this effect, and assume that the photon path
obeys $r(\tau)=\tau_{0}-
\tau$. In a flat, matter-dominated universe, $r_{S}=2 H_{0}^{-1}
[1-(1+z_{S})^{-1/2}].$
The components perpendicular to the $z$-axis of the photon
direction obey \cite{linder}
\begin{equation}
\frac{dx^{i}}{d\tau}(\tau)-\frac{dx^{i}}{d\tau}(\tO)=h_{zi}(\tau)-
h_{zi}(\tO)-\frac{1}{2}\int_{\tau}^{\tO}\nabla_{i}h_{zz}(\tau')\
d\tau'\ .
\end{equation}
Integrating this we find, for the perpendicular components of the
position (with respect to $\xV(\tO)=0$),
\begin{equation}
x^{i}(\tau)=\int_{\tau}^{\tO}\left[\frac{1}{2}(\tau'-\tau)\nabla_{i}
h_{zz}(\tau')+h_{zi}(\tO)-h_{zi}(\tau')\right]d\tau'\ .
\label{thetai}
\end{equation}
We define a (two-component) angle $\beta^{i}=x^{i}(\tau)/r(\tau)$.
In gravitational lensing with a primary thin lens at a distance
$r_L$ (but no LSS or GW's) the lens equation is (e.g.,
\cite{schneider})
\begin{equation}
\vec{\beta}=\vec{\theta}-\vec{\alpha}_{\rm lens} (\vec{\theta})\ ,
\label{1lens}
\end{equation}
where $\vec{\theta}$ is the observed image angle, $\vec{\beta}$ is
the source angle (defined as $\vec{x}_S/r_S$, in terms of the
perpendicular position of the source), and $\vec{\alpha}_{\rm lens}$
is the deflection angle scaled by $r_{LS}/r_S$ (we
define $r_{LS}=r_S-r_L$).
In this case, the fiducial $z$-axis is defined to be in the observed
direction of the lens. The distortion of the image of a small
source is given by the inverse of the Jacobian matrix
\begin{equation}
\frac{\partial \beta^{i}}{\partial \theta^{j}}=\delta^{ij}-
\Psi^{ij}\ ,
\end{equation}
where $\Psi^{ij}$ is also termed the shear tensor of the lens.
\section{Shear induced by GW's on lensing}
In this section we follow the approach used for LSS in Ref.\
\cite{seljak},
i.e., we compute some of the same quantities for GW's and compare the
results. As stated in Sec.\ I, we do not need to include a lens
explicitly, as we now justify.
In the presence of a metric perturbation, but without a primary
lens, the lens
equation has the form $\vec{\beta}=\vec{\theta}-\vec{\alpha}_{OS}
(\vec{\theta})$,
where $\vec{\alpha}_{OS}$ results from the accumulated deflection
between the
observer and the source. As defined in Sec.\ II, the shear tensor for
an image at
$\vec{\theta}$ due to the perturbation is $F^{ij}=\partial
\alpha_{OS}^{i}(\vec{\theta})/
\partial \theta^{j}$. On the other hand, the relative deflection
at $\vec{\theta}$ between
two rays separated by a tiny angle $\vec{\gamma}$ is $\vec{\alpha}_
{OS}(\vec{\theta}+\vec{\gamma})-\vec{\alpha}_{OS}(\vec{\theta})$.
We denote the rms of this quantity by $\sigma_{\Delta\beta}$. We
average over directions of $\vec{\gamma}$ (which
for this calculation is equivalent to fixing $\vec{\gamma}$ and
assuming that $F^{ij}$ is
isotropic) and take $\gamma \rightarrow 0$, obtaining the relation
\footnote{Repeated indices are summed over the x and y directions.
There is no distinction between upper and lower indices.}
\begin{equation}
\left(\frac{\sigma_{\Delta\beta}}{\gamma}\right)^2=\frac{1}{2}
\left<F^{ij}F_{ij}\right>\ ,
\label{FM}
\end{equation}
all evaluated at position $\vec{\theta}$. Thus, $\sigma_{\Delta
\beta}/\gamma$ yields an estimate of the magnitude of the shear
tensor. Indeed, it fully characterises rms
values of $F^{ij}$, since for an isotropic field
\begin{equation}
\left<F_{ij}F_{kl}\right>=\frac{1}{8} \left[\delta_{ij}\delta_{kl}+
\delta_{ik}\delta_{jl}+
\delta_{il}\delta_{jk}\right] \left<F^{mn}F_{mn}\right>\ .
\end{equation}
If we also include a primary lens, in the lens equation we simply
add up all deflections
linearly, assuming all deflections are small. For the primary lens
alone, we have
Eq.\ (\ref{1lens}). Figure \ref{fig} shows this setup
schematically. In the
presence of a metric perturbation, we trace a light ray that is
observed at $r=0$ to
come from the direction $\vec{\theta}$, back to the source.
We find a different form for the lens equation:
\begin{equation}
\vec{\beta}=\vec{\theta}-\vec{\alpha}_{OS}^{(2,3)}-\vec{\alpha}_
{\rm lens} (\vec{\theta}-\vec{\alpha}_{OL}^{(2)})\ .
\label{2lens}
\end{equation}
Here $\vec{\alpha}_{OS}^{(2,3)}$ refers to the integrated deflection
caused by LSS or GW's along the paths labeled 2 and 3 in Figure
\ref{fig}, defined so that the total induced
change in $\vec{x}(\tau_S)$ equals $r_S \vec{\alpha}_{OS}^{(2,3)}$.
Similarly, $r_L \vec{\alpha}_{OL}^{(2)}$ is the induced change in
$\vec{x}(\tau_L)$. In integrating
the deflections along the unperturbed paths 2 and 3, we are assuming
that the relative deflections due to LSS or GW's are small compared
to $\theta$ and $\alpha_{\rm lens}$,
which is true for the cases which we consider below.
When the perturbation is included, $\vec{\theta}$ is no longer an
observable, since it is measured
with respect to the unperturbed position of the lens. The observed
position of the lens (whose actual position has not changed) is now
$\vec{\theta}_{\rm lens}=\vec{\alpha}_{OL}^{(1)}$, so the lens
equation in terms of the observable $\vec{\theta}'=\vec{\theta}-
\vec{\theta}_{\rm lens}$ is
\begin{equation}
\vec{\beta}=\vec{\theta}'+\vec{\alpha}_{OL}^{(1)}-
\vec{\alpha}_{OS}^{(2,3)}-\vec{\alpha}_{\rm lens}(\vec{\theta}'+
\vec{\alpha}_{OL}^{(1)}-\vec{\alpha}_{OL}^{(2)})\ .
\label{3lens}
\end{equation}
If we now calculate the shear tensor resulting from equation
(\ref{3lens}), it will contain
the shear of the primary lens, shear terms from the perturbation,
and also cross terms.
For simplicity, in the case of GW's we only estimate one
characteristic
magnitude, that of the shear resulting from $\vec{\alpha}_{OS}^
{(2,3)}$, by evaluating
the corresponding $\sigma_{\Delta\beta}/\gamma$. In Ref.\
\cite{me2} all the different shear
terms were studied for LSS, showing that $\sigma_{\Delta\beta}/\gamma$
indeed estimates the relative magnitude of the various corrections
due to LSS. Since we find that $\sigma_{\Delta
\beta}/\gamma$ is much smaller for GW's, we do not have any
motive to explore equation (\ref{3lens}) further in this case.
Instead of the path $(2,3)$, we may use a straight path
from $r=0$ to $r=r_S$ to evaluate the rms of various quantities
in this section, since $\alpha_{\rm lens} \ll 1$ and so the
components of vectors and tensors as well as the relative distances
of points along the path (both of which enter
into the rms calculations) are unchanged (except for $O(\alpha_{\rm
lens})$ corrections). Thus we only need to consider the effect of
GW's in the absence of a primary lens.
%%%%%%%%Figure figgw.ps
\begin{figure}[t]
\vspace*{6.3 cm}
\caption{Sketch showing positions of the observer, lens and source,
as well as an image ray and several comoving distances.}
\special{psfile=figgw.ps
voffset=255 hoffset=40 hscale=45 vscale=45}
\label{fig}
\end{figure}
%%%%%%%%
Consider first a single light ray with $\vec{\theta}=0$. In the
absence of GW's (or LSS) it
would follow the straight line $x^{i}(\tau)=0$ for all
$\tau$. We now include the effect of GW's, and compute the rms
fluctuation in the
photon's perpendicular displacement at the source, $\sigma_{\beta}=
\left<\vec{\beta}(\tS)\cdot\vec{\beta}(\tS)\right>
^{1/2}$. This is a measure of the common deflection of all image
rays, and is therefore
not observable, but it is useful for the calculations that follow.
We use equation (\ref{thetai}) and convert the expression to Fourier
space. Consider first only the $h_{zz}$ term, whose contribution
to $\sigma_{\beta}^{2}$ we denote $\sigma_{\beta,a}^{2}$. The
polarization gives $(\epsilon_{zz}^+)^2+(\epsilon_{zz}^{\times})^2=
\sin^4\theta_k$, where $\kV=(k,\theta_k,\phi_k)$ in spherical
coordinates. Performing the angular $\kV$ integrations then yields
$$\sigma_{\beta,a}^{2}=
\frac{432 \pi} {r_{S}^{2}} \AT \int_{\tS}^{\tO}d\tau_{1} \int_{\tS}
^{\tO}d\tau_{2}\int_{0}^{\infty} dk\ k(\tau_{1}-\tS)(\tau_{2}-\tS)
\frac{j_{1}(k\tau_{1})}{k\tau_{1}} \frac{j_{1}(k\tau_{2})}
{k\tau_{2}}\frac{j_{3}[k(\tau_{1}-
\tau_{2})]}{[k(\tau_{1}-\tau_{2})]^3}\ .$$
The $j_{3}[k(\tau_{1}-\tau_{2})]/[k(\tau_{1}-\tau_{2})]^3$ term
represents a
further suppression of short wavelength modes due to phase
cancellations among
different waves in the assumed isotropic stochastic background.
Letting $s=k\tau_{1}$
and $q=\tau_{2}/\tau_{1}$, we can simplify this expression to a
double integral,
\begin{equation}
\sigma_{\beta,a}^{2}=\frac{864 \pi} {r_{S}^{2}} \AT
\int_{\oS}^{1} \left\{q\tO\left(\frac{1}{2}\tO-\tS\right)
-\tS r_{S}+\tS^{2} \left[\frac{1}
{2q}+\ln(q/\oS)\right]\right\}\,W(q)dq\ ,
\end{equation}
where $\oS=\tS/\tO$ and with $u \equiv (1-q)s$ we define
\begin{equation}
W(q)=\int_{0}^{\infty}\frac{j_{1}(s)}{s}\frac{j_{1}(q s)}{q s}
\frac{j_{3}(u)}{u^3}\,s\ ds\ .
\end{equation}
Similarly, the contribution of the $h_{zi}$ terms of Eq.\
(\ref{thetai}) is
\begin{equation}
\sigma_{\beta,b}^{2}=\frac{144 \pi}{r_{S}^{2}} \AT \int_{\oS}^{1}
\left[\frac{\tS^{2}}{2 q^{2}}+\tO\left(\frac{1}{2}\tO-\tS\right)
\right]G(q)dq\ ,
\end{equation}
where
\begin{equation}
G(q)=\int_{0}^{\infty}\left\{\frac{1}{5} \left[\frac{j_{1}(s)}
{s}\right]^{2}+\frac{1}{5}\left[\frac{j_{1}(q s)}{q s}\right]
^{2}-2\,\frac{j_{1}(s)}{s}\frac{j_{1}(q s)}{q s}\left[\frac
{j_1(u)}{u}-2\frac{j_2(u)}{u^2}\right]
\right\}\frac{ds}{s}\ .
\end{equation}
Integrating over angles gives a zero cross term, and so $\sigma_
{\beta}^{2}=\sigma_{\beta,a}^{2}+\sigma_{\beta,b}^{2}$.
Numerically, we find that $\sigma_{\beta}=5 \times 10^{-6}\
(z_{S}=1),\ 9 \times 10^{-6}\ (z_{S}=3)$. This is much smaller
than the estimates for LSS \cite{seljak}, $6 \times 10^{-4}\
(z_{S}=1),\ 7 \times 10^{-4}\ (z_{S}=3)$.
To estimate the relative deflection between rays at $\vec{\theta}
=0$, we choose two
directions (labeled A and B) separated at the observer by an
infinitesimal angle $\gamma$,
and find the rms difference between the deflections due to GW's in
these two directions,
$\sigma_{\Delta\beta}=\left<\left[\vec{\beta}_{A}(\tS)-
\vec{\beta}_{B}(\tS)\right]^{2}\right>^{1/2}$.
We cannot evaluate this with the method used for LSS, which assumes
that horizon size modes are negligible \cite{kaiser}. Instead we
must calculate $\sigma_{\Delta\beta}^{2}$ explicitly and keep all
the terms to lowest order in $\gamma$, i.e., quadratic order. These
include terms which come from multiplying polarization components
for the different directions A and B. The final result is
\begin{equation}
(\sigma_{\Delta\beta}/\gamma)^{2}=4 \sigma_{\beta,a}^{2}+
\sigma_{\Delta\beta,a}^{2}+\sigma_{\Delta\beta,b}^{2}
+\sigma_{\Delta\beta,c}^{2}\ ,
\end{equation}
where
\begin{eqnarray}
\sigma_{\Delta\beta,a}^{2}&=&\frac{576 \pi}{r_{S}^{2}} \AT
\int_{\oS}^{1}\left[\frac{\tO\tS}{q}(1+q)^2-\left(\tO^2+\frac{
\tS^2}{q}\right)(1+q)+(\tO+\tS)r_S \ln(q/q_S)\right] I_1(q)dq\ ,
\nonumber \\
\sigma_{\Delta\beta,b}^{2}&=&\frac{1728 \pi}{r_{S}^{2}} \AT
\int_{\oS}^{1}
\biggl\{(\tO+\tS)^2-\tO^2(1+q)^2+2q\tS(\tS-q\tO)+ \\ \nonumber
& &\hspace{1in}+\left[\tS\tO(1+q^2)+q(\tO+\tS)^2\right]
\ln(q/\oS)\biggl\}I_2(q)dq\ , \nonumber \\
\sigma_{\Delta\beta,c}^{2}&=&\frac{288 \pi}{r_{S}^{2}} \AT
\int_{\oS}^{1} \left[\frac{\tS^{2}}{2 q^{2}}+\tO\left(
\frac{1}{2}\tO-\tS\right)\right] I_3(q) dq\ , \nonumber \\
I_1(q)&=&\int_{0}^{\infty}\frac{j_{1}(s)}{s}\frac{j_{1}(q s)}{q s}
\left[\frac{j_{2}(u)}{u^2}-3\frac{j_{3}(u)}
{u^3}\right]s\,ds\ , \quad
I_2(q)=\int_{0}^{\infty}\frac{j_{1}(s)}{s}\frac{j_{1}(q s)}{q s}
\frac{j_{4}(u)}{u^4}s^3\,ds\ , \nonumber \\
I_3(q)&=&\int_{0}^{\infty}\left\{\frac{2}{15}\left[\frac{j_{1}(s)}
{s}\right]^{2}+\frac{2}{15}\left[\frac{j_{1}(q s)}{q s}\right]
^{2}-2\,\frac{j_{1}(s)}{s}\frac{j_{1}(q s)}{q s}\left[\frac
{j_1(u)}{u}-3\frac{j_2(u)}{u^2}\right]
\right\}\frac{ds}{s}\ . \nonumber
\end{eqnarray}
Then $\sigma_{\Delta\beta}/\gamma=7 \times 10^{-6}\ (z_{S}=1),\ 1.3
\times 10^{-5}\
(z_{S}=3)$. By contrast, LSS gives a $\sigma_{\Delta\beta}/\gamma=0.
07\ (z_{S}=1),\ 0.14\ (z_{S}=3)$. For LSS, the relative deflection
is greatly increased by coherent deflections for short wavelength
modes, but for GW's the effect of these modes is cut off
by the redshifting as well as the temporal oscillations. We also
used the relation
$k r_{S}\gamma \ll 1$ in the calculation of $\sigma_{\Delta\beta}$.
The reason we find a $\sigma_{\Delta\beta}$ of order
$\gamma\sigma_{\beta}$ is that long wavelength modes overlap over
the two light rays, and the relative deflection is small compared
to the total deflection. Indeed, a Taylor expansion suggests that
in general $\sigma_{\Delta\beta}/\gamma \sim k
r_{S} \sigma_{\beta}$, and $k r_{S} \sim 1$ is dominant for this
GW spectrum. As shown above, the shear tensor (which is also used
in weak lensing) is closely related to $\sigma_{\Delta\beta}/
\gamma$, and so the mean square ellipticity at a point induced by GW's
is of order $10^{-5}$, again negligible compared to the few per
cent expected from LSS (e.g., \cite{kaiser}).
We can also try to derive general limits on GW's at astrophysical
wavelengths from the induced shear. To obtain a limit on
$\Omega_{\lambda}$, we compute the $\sigma_{\Delta\beta}/\gamma$
produced by an isotropic background of GW's at a single
wavenumber $k$. Note that for
modes at a given $k$, we can use equations (\ref{frpres}) and
(\ref{ompres}) even for short wavelengths (with $\AT$ a
normalization factor, separate for each $k$), for times $\tau$
long after matter-radiation equality. Since GW's at Horizon wavelengths
are already strongly constrained by the CMB as noted above, we
restrict our calculation to the case $k r_S \gg 1$, in which case
the $h_{zi}$ terms in Eq.\ (\ref{thetai}) can be neglected.
We can estimate from Eq.\ (\ref{thetai}) that in order of
magnitude $\sigma^2_{\beta}$ should equal $\AT/(k \tO)^4$, and thus
that $\sigma^2_{\Delta\beta}/\gamma^2 \sim \AT/(k \tO)^2$. However,
we find from the exact calculation that there is no term this
large, only higher order terms in $1/(k \tO)$. We outline in
the Appendix a mathematical argument showing this cancellation
at small wavelengths. This result requires both the phase
cancellations that come in averaging over an isotropic
background, and also the oscillation with time of the GW's.
With different assumptions, e.g., if we analyzed GW's from a
particular source, which are then not isotropic,
stronger limits may be possible.
\section{Proper motions induced by LSS and GW's}
We now consider the fluctuation of the angular deflection of image
rays with time, and the
resulting proper motion. If the deflection of image rays induced by
LSS or GW's changes
significantly during an observation of a lens system, then the slow
shift in alignment
between the lens and the source will change the impact parameter at
the lens of a given ray from the source. The images will therefore
move, and even tiny motions
may be detected since the source motion is magnified if it is
lensed by a primary lens.
We first show that this effect is still expected to be too small to
measure for LSS and for
the GW power spectrum that we have considered above. However, given
the weakness of existing
limits on GW's at astrophysical wavelengths (Sec.\ I), we also consider
possible limits on a general GW spectrum.
Again we consider a single light ray from the observer out to some
distance $r_S$, in the absence of a primary lens (we consider the
effect of a lens below). Given a ray with a fixed direction at the
observer, its position $x^{i}(r_{S})$ at $r_S$ moves with time, and
it is this motion which we evaluate. In practice, we
are interested in a fixed source at $r_S$, in which case its {\it
apparent} position will drift with the same speed but in the
opposite direction. For LSS we have (e.g., \cite{seljak})
\begin{equation}
x^{i}(r_{S})=-2\int_{r=0}^{r_{S}}(r_{S}-r)\nabla_{i}\phi
(\tau=\tO-r)dr,
\end{equation}
in terms of the Newtonian potential (or scalar metric perturbation)
$\phi$. We are now using the parameter $r$ rather than $\tau$,
since as time changes all comoving distances remain fixed. The only
change is the time of evaluation of $\phi$, and so
to find $dx^{i}(r_{S})/d\tO$ from $x^{i}(r_{S})$ we simply replace
$\phi(\tau=\tO-r)$ by $\dot{\phi}(\tau=\tO-r)$, with the partial
time derivative in $\dot{\phi}$ taken at a fixed position. The rms
value of $dx^{i}(r_{S})/d\tO$ depends on the power spectrum of
$\dot{\phi}$, a quantity which has been estimated by various
authors in connection with the Rees-Sciama effect on the CMB
(e.g., \cite{RS}). While the integrated deflection is dominated by
short ($\sim 1$ Mpc) wavelengths, the LSS potential only evolves
on a cosmological time scale. In an Einstein-deSitter universe,
$\phi$ is time independent in the linear regime of small density
perturbations, but in this case too $\dot{\phi}$ becomes nonzero
when nonlinear structure forms. In general, therefore, the
proper motion induced by LSS is of order $\sigma_{\beta}/\tO \simeq
10^{-8}$ arcsec per year. For the gravity wave spectrum considered
above, horizon size modes are dominant, and so here too the induced
proper motion is of order $\sigma_{\beta}/\tO$, with a $\sigma_
{\beta}$ smaller by $\sim 10^{2}$ than for LSS. Any observed proper
motion will thus be dominated by peculiar velocities of hundreds
of km/s generated, e.g., by the velocity dispersion of stars in a
galaxy or galaxies in a galaxy group or cluster.
We now estimate the lensing limit on stochastic GW's in general, at
any wave number $k$. VLBI observations can directly measure or
limit proper motions, and this then implies a limit on GW's.
Again we restrict ourselves to wavelengths with $k r_S \gg 1$, and
consider first the apparent motion of a source that is not lensed by a
primary lens. The apparent motion due to GW's
of a fixed object at distance $r_S$ is $-d\vec{\beta}(r_S)/
d\tau_0$. Up to corrections of order $1/k r_S$, the mean
square of this motion is
\begin{equation}
\left<\left[\frac{d}{d\tau_0}\beta(r_S)\right]^2\right> = \frac{18
\pi \AT} {5 k^2 \tO^4} \left[1+\frac{2}{3}\cos(2 k \tO)\right]\ .
\label{eqcom}
\end{equation}
However, when there are both a lens and a source, a GW background
will produce correlated proper motions in both, and the relative
motion may be small. Limits from VLBI on proper motions in
gravitational lenses were recently considered
in Ref.\ \cite{koch}, and we proceed similarly. We may
hope for strong limits because, in the presence of lensing, a
proper motion of the source relative to the lens is magnified into
a larger proper motion of the images.
Furthermore, only a {\it relative} motion between images needs to
be detected, as opposed to a more difficult measurement of motion
with respect to an external reference frame, since if the source moves
(relative to the lens), the different images do {\it not} all move
together. In general, different values of the magnification matrix
at the different image positions will produce relative motions
between images of the same order of magnitude as the absolute
motions. Moreover, pairs of highly magnified
images generally have antiparallel motions \cite{koch}.
To analyze how proper motion due to GW's may be magnified, we
start from Eq.\ (\ref{3lens}), and consider the same equation a
time $\Delta t$ later, when the deflections from GW's have changed.
E.g., $\vec{\alpha}_{OL}^{(1)}$ has changed to $\vec{\alpha}_{OL}^
{(1)}+\Delta^{(1)}$, and a total change $\vec{\Delta}$ in the
observed $\vec{\theta}'$ has been induced. Expanding the lens
equation to first order in the small changes and solving for
$\vec{\Delta}$, we obtain
\begin{equation}
\Delta_{i}=\Delta^{(2)}_{i}-\Delta^{(1)}_{i}+M_i^j \left[\Delta^
{(2,3)}_j-\Delta^{(2)}_j\right]\ ,
\label{eqdel}
\end{equation}
where the magnification matrix $M_i^j$ equals the inverse of
$\delta_i^j-\partial_i\alpha_{\rm lens}^j$ (and is evaluated at
$\vec{\theta}'+\vec{\alpha}_{OL}^{(1)}-\vec{\alpha}_{OL}^{(2)}$).
Consider first the magnified term, $\Delta^{(2,3)}_j-\Delta^{(2)}_j$.
Averaging over directions of $\vec{\Delta}^{(2,3)}-\vec{\Delta}^{(2)}$
we obtain a result analogous to Eq.\ (\ref{FM}) for the mean
square. Since $M^{ij}$ is symmetric for a thin lens \cite{schneider},
it has two real eigenvalues $m_a$ and $m_b$ (where the magnification
$M=|m_a m_b|$). Letting
\begin{equation}
\Mtil=\left[\frac{1}{2}(m_a^2+m_b^2)\right]^{1/2}\ ,
\label{Mtil}
\end{equation}
we find that
\begin{equation}
{\rm rms\ }\left|M_i^j \left[\Delta^{(2,3)}_j-\Delta^{(2)}_j\right]
\right| = \Mtil\, \times
{\rm rms\ }\left|\vec{\Delta}^{(2,3)}-\vec{\Delta}^{(2)}\right|\ .
\label{eq2mag}
\end{equation}
In Eq.\ (\ref{eq2mag}) we may evaluate the rms on the right-hand
side using a straight path (as in Sec.\ III).
Letting
\begin{equation}
\frac{d}{d\tO}\vec{\beta}_{LS} =-\frac{d}{d\tO}\vec{\beta}(r_S)+
\frac{d}{d\tO}\vec{\beta}(r_L)\ ,
\end{equation}
we find that in $\left<\dot{\beta}_{LS}^2\right>$ there is no
term of order $\AT /(k^2 \tO^4)$ [as in Eq.\ (\ref{eqcom})],
but only higher order terms in $1/(k \tO)$.
Once again this small wavelength cutoff results from combining
the time oscillation of GW's and the phase cancellations in averaging
over an isotropic background (see the Appendix), and as a result
there is only a very weak limit on $\Omega_{\lambda}$.
\section{Conclusions}
Gravitational lensing is affected by perturbations to the
homogeneous and isotropic
background metric. Such perturbations, whether they are caused by
LSS or GW's, may produce a number of effects on light propagation.
One such effect is an overall shift in the angular positions of
nearby objects, which is not observable. Another is
a relative difference between the induced shifts in nearby light
rays. This relative deflection manifests itself as a shear which
may cause weak lensing
and also affect strong lensing. A third effect is a fluctuation of
the angular position of distant objects with time, leading to a
directly observable proper motion.
The actual amplitude of long wavelength modes of LSS and GW's is
limited by the quadrupole anisotropy of the CMB. Even if both make
comparable contributions to the anisotropy, LSS is dominant in its
effects on lensing. This results from cancellations due to the time
oscillation of short wavelength gravity waves, as well as the
redshifting of their amplitude.
For LSS, on the other hand, the effect of small coherent structures
is amplified as the deflection executes a random walk. We find that
the relative deflection due to GW's is four orders of magnitude
smaller than that of LSS, and is therefore not observable.
The induced proper motions expected for LSS or for GW's generated in
inflation are small compared to typical peculiar velocities, and
thus are not observable. The motions are also too small to yield
interesting limits on the energy density of GW's at shorter
wavelengths.
After this paper was submitted for publication, the bending of
light by gravity waves was analyzed differently in Ref.\
\cite{new}, for the case of short (subhorizon) wavelengths,
in a nonexpanding flat space (i.e., neglecting the redshifting
of the amplitude of GW's). That simplified analysis shows that the
relative proper motion between two sources is small not only if
they are at different redshifts along the same line of sight
(in agreement with our calculation of
$\left<\dot{\beta}_{LS}^2\right>$ in Sec.\ IV), but also
if they are separated on the sky by a small angle. The
treatment presented in Ref.\ \cite{new} changes quantitatively
if expansion is included, but not qualitatively for GW's with
a period short compared to the redshifting time scale (i.e.,
a Hubble time).
\vspace{.1in}
\begin{center}
{\bf Acknowledgements}
\end{center}
It is a pleasure to thank Ed Bertschinger for helpful advice and many
valuable comments, Uro\v{s} Seljak for valuable discussions and
comments, and Paul Schechter, Sean Carroll, Alan Guth and Paul
Steinhardt for helpful discussions. I thank Scott
Dodelson for pointing out an error. This work was supported by NASA
grant No.\ NAG5-2816.
\appendix
\section{Appendix}
In trying to set limits on GW's at short wavelengths $k r_S \gg 1$,
we twice encountered a weaker limit than simple dimensional
analysis would suggest: Once in calculating
$\sigma_{\Delta\beta}/\gamma$ or shear in Sec.\ III, and then in
estimating the magnified proper motion in Sec.\ IV. In this
Appendix we outline the first of these calculations and
show how this result emerges. The second calculation can be
done similarly.
From Eq.\ (\ref{thetai}) in the limit of short wavelengths
(compared with the present Horizon), we derive
$$\sigma_{\Delta\beta}^{2}/\gamma^2=
\frac{1728 \pi} {r_{S}^{2}} \AT k^4 \int_{\tS}^{\tO}d\tau_{1}
\int_{\tS}^{\tO}d\tau_{2}(\tau_{1}-\tS)(\tau_{2}-\tS)
(\tO-\tau_{1})(\tO-\tau_{2})
\frac{\cos(k\tau_{1})}{(k\tau_{1})^2}\frac{\cos(k\tau_{2})}
{(k\tau_{2})^2}\frac{j_{4}[k(\tau_{1}-
\tau_{2})]}{[k(\tau_{1}-\tau_{2})]^4}\ .$$
The $j_{4}[k(\tau_{1}-\tau_{2})]/[k(\tau_{1}-\tau_{2})]^4$
term comes from the angular $\kV$ integrations, including
the angular dependence of the polarizations and assuming
an isotropic background.
Letting $x=k \tau_1$ and $u=k (\tau_1-\tau_2)$ (also $x_0
=k \tO$, etc.) leads to
$$\frac{864 \pi} {k^2 r_S^2} \AT \int_{x_S}^{x_0}dx
\frac{(x-x_S)(x_0-x)}{x^2}\int_{x_S-x}^{x_0-x}du\frac{
(x+u-x_S)(x_0-x-u)}{(x+u)^2} \frac{j_4(u)}{u^4}\left[
\cos u(1+\cos 2 x)-\sin u \sin 2 x\right]\ . $$
We evaluate only the first $\cos u$ term here, since the other
terms in the square brackets can be evaluated similarly. Note
that dimensional analysis suggests that the $x$ and $u$ integrals
should give a term of order 1 (not larger, because of the
oscillating integrand).
To do the $u$ integral we separate the smooth and oscillating
parts and then repeatedly integrate by parts: We let $w^{[n]}(u)$
be the $n$-th indefinite integral of $[\cos u]\,j_4(u)/u^4$ with respect to
$u$, and $v^{[n]}(u)$ the $n$-th derivative of $(x+u-x_S)
(x_0-x-u)/(x+u)^2$ with respect to $u$. For each $n$ such that $w^{[n]}(u)$
converges for $u\rightarrow\pm\infty$, we fix the arbitrary constant by
$w^{[n]}(\infty)+w^{[n]}(-\infty)=0$ (Any constant will do
for other $n$). Then the $u$ integral equals a
series of terms evaluated at the two limits of integration,
$$ \sum_{n=0}^{\infty}(-1)^n\left\{ w^{[n+1]}v^{[n]}|_{u=x_0-x}-
w^{[n+1]}v^{[n]}|_{u=x_S-x}\right\}\ . $$
Since the two series of terms can be handled similarly, we
evaluate here only the $u=x_0-x$ terms. We do the $x$ integration
in the same way as the $u$ integration. Thus we continue to
integrate $w^{[n+1]}(x)$ with respect to $x$, and let $v^{[n,m]}(x)$
be the $m$-th derivative of $(x-x_S)(x_0-x)v^{[n]}(x_0-x)/x^2$
with respect to x. The contribution to $\sigma_{\Delta\beta}^{2}/
\gamma^2$ from the terms we have kept is
$$\frac{864 \pi} {x_S^2} \AT \sum_{n,m=0}^{\infty} (-1)^{n+1}
\left\{w^{[n+m+2]}(0)v^{[n,m]}(x_0)-w^{[n+m+2]}(x_0-x_S)
v^{[n,m]}(x_S)\right\}\ . $$
Now, $v^{[n,m]}(x)$ at $x \gg 1$ is of order $x^{-(n+m)}$,
$w^{[n]}(0)$ is $0$ or a constant, and we find that
$w^{[n]}(x)$ at $|x| \gg 1$ is of order $|x|^{n-5}$. This last
fact, that $w^{[n]}(\pm\infty)$ converges for the first few
$n$, depends on the specific function $w^{[0]}(x)$ which
in turn is determined by the two physical assumptions of
time oscillation and angular averaging. The only term
from the final sum that could give a contribution of order
$\AT/x_S^2$ is the $n=m=0$ term. We find that $w^{[2]}(0)$
is a nonzero constant, but since $v^{[0,0]}(x)=0$ identically,
there is no term of this lowest order.
\begin{thebibliography}{gwaves}
\bibitem{AbbWs} A. A. Starobinskii, Pis'ma Zh. Eksp. Teor. Fiz.
{\bf 30}, 719 (1979) [JETP Lett. {\bf 30}, 682 (1979)];
V. A. Rubakov, M. V. Sazhin, and A. V. Veryaskin,
Phys. Lett. {\bf 115B}, 189 (1982); A. A. Starobinskii,
Pis'ma Astron. Zh. {\bf 11} 323 (1985) [Sov. Astron. Lett. {\bf 11},
133 (1985)]; L. F. Abbott and M. B. Wise, Nucl. Phys.
{\bf B244}, 541 (1984).
\bibitem{me} R. Bar-Kana, Phys. Rev. D {\bf 50},
1157 (1994); A. R. Liddle, {\it ibid.\ } {\bf 49}, 3805 (1994);
M. S. Turner, M. White, and J. E. Lidsey, Phys. Rev. D {\bf 48},
4613 (1993); V. Sahni, {\it Phys. Rev. D} {\bf 42}, 453
(1990); V. Sahni and T. Souradeep, Mod. Phys.
Lett. A {\bf 7}, 3541 (1992).
\bibitem{falco}
E. E. Falco, M. V. Gorenstein, and I. I. Shapiro,
Astrophys. J. {\bf 372}, 364 (1991).
\bibitem{refsdal}
S. Refsdal, Mon. Not. Roy. Astr. Soc. {\bf 128}, 307 (1964);
S. Refsdal, S. Mon. Not. Roy. Astr. Soc. {\bf 132}, 101 (1966).
\bibitem{infl}
L. M. Krauss and M. White, Phys. Rev. Lett. {\bf 69}, 869
(1992); R. L. Davis, H. M. Hodges, G. F. Smoot,
P. J. Steinhardt and M. S. Turner, {\it ibid.} {\bf 69},
1856 (1992); M. S. Turner, Phys. Rev. D {\bf 48}, 5539 (1993).
\bibitem{COBE} G. F. Smoot {\it et al., Astrophys. J. Lett.} {\bf
396}, L1 (1992); K. M. Gorski {\it et al., ibid.\ } {\bf 430},
L89 (1994).
\bibitem{weak}
J. Miralda-Escud\'{e}, Astrophys. J. {\bf 380}, 1 (1991);
R. D. Blandford, A. B. Saust, T. G. Brainerd, and
J. V. Villumsen, Mon. Not. Roy. Astr. Soc. {\bf 251}, 600 (1991);
M. Jaroszynski, C. Park, B. Paczynski, and J. R. Gott
III, Astrophys. J. {\bf 365}, 22 (1991);
vJ. Mould, R. Blandford, J. Villumsen, T. Brainerd,
I. Smail, T. Small, and W. Kells, Mon. Not. Roy. Astr. Soc. {\bf
271}, 31 (1994); J. V. Villumsen, preprint astro-ph/9507007 (1995).
\bibitem{kaiser}
N. Kaiser, Astrophys. J. {\bf 388}, 272 (1992).
\bibitem{me2}
R. Bar-Kana, Astrophys. J. {\bf 468}, 17 (1996).
\bibitem{seljak}
U. Seljak, Astrophys. J. {\bf 436}, 509 (1995).
\bibitem{allen}
B. Allen, Phys. Rev. Lett. {\bf 63}, 2017 (1989); B. Allen,
Gen. Relativ. Grav. {\bf 22}, 1447 (1990).
\bibitem{surpi}
G. C. Surpi, D. D. Harari, and J. A. Frieman, Astrophys. J. {\bf
464}, 54 (1996); J. A. Frieman, D. D. Harari, and G. C. Surpi,
Phys. Rev. D {\bf 50}, 4895 (1994).
\bibitem{schneider}
P. Schneider, J. Ehlers, and E. E. Falco, {\it Gravitational
Lenses}, Springer-Verlag, 1992.
\bibitem{latest}
S. E. Thorsett and R. J. Dewey, Phys. Rev. D {\bf 53}, 3468 (1996).
\bibitem{kaspi}
V. M. Kaspi, J. H. Taylor, and M. F. Ryba, Astrophys. J. {\bf 428},
713 (1994).
\bibitem{Rees}
B. Bertotti, B. J. Carr, and M. J. Rees, Mon. Not. R. Astron. Soc.
{\bf 203}, 945 (1983).
\bibitem{linder}
E. V. Linder, Astrophys. J. {\bf 328}, 77 (1988).
\bibitem{Carr}
B. J. Carr, Astron. \& Astrophys. {\bf 89}, 6 (1980).
\bibitem{linder2}
E. V. Linder, Astrophys. J. {\bf 326}, 517 (1988).
\bibitem{fakir}
R. Fakir, Astrophys. J. {\bf 426}, 74 (1994).
\bibitem{pyne}
T. Pyne, C. R. Gwinn, M. Birkinshaw, T. M. Eubanks, and D. N.
Matsakis, Astrophys. J. {\bf 465}, 566 (1996).
\bibitem{RS}
M. J. Rees and D. W. Sciama, Nature {\bf 517}, 611 (1968); E.
Martinez-Gonzalez, J. L. Sanz, and J. Silk, Phys. Rev. D. {\bf 46},
4196 (1992); E. Martinez-Gonzalez,
J. L. Sanz, and J. Silk, Astrophys. J. {\bf 436}, 1 (1994); L.
Kofman and A. Starobinsky, Sov. Astron. Lett., {\bf 11}, 271 (1985);
M. Kamionkowski and D. N.Spergel, Astrophys. J. {\bf 432}, 7 (1994);
U. Seljak, Astrophys. J. {\bf 460}, 549 (1996).
\bibitem{koch}
C. S. Kochanek, T. S. Kolatt, and M. Bartelmann, submitted to the
Astrophys. J., astro-ph/9602037 (1996).
\bibitem{new}
N. Kaiser and A. Jaffe, submitted to the
Astrophys. J., astro-ph/9609043 (1996).
\end{thebibliography}
\end{document}