Search for a stochastic background of 100-MHz gravitational
waves with laser interferometers
Tomotada Akutsu/'Q Seiji Kawamura, 2 Atsushi Nishizawa, 3
Koji Arai, 2 Kazuhiro Yamamoto, 2 Daisuke Tatsumi, 2 Shigeo Nagano,
Erina Nishida, 5 Takeshi Chiba, 6 Ryuichi Takahashi, 7 Naoshi Sugiyama, 7
Mitsuhiro Fukushima, 2 Toshitaka Yamazaki, 2 and Masa-Katsu Fujimoto 2
1 Department of Astronomy, University of Tokyo, Bunkyo, Tokyo 113-0033, Japan
qq 1 Kyoto University, Kyoto 606-8501, Japan
2 National Astronomical Observatory of Japan, Mitaka, Tokyo 181-8588, Japan
3 Graduate School of Human and Environmental Studies,
^National Institute of Information and Communications
Technology, Koganei, Tokyo 184-8795, Japan
5 Graduate School of Humanities and Sciences,
Ochanomizu University, Bunkyo, Tokyo 112-8610, Japan
6 Department of Physics, College of Humanities and Sciences,
q ', Nihon University, Tokyo 156-8550, Japan
7 Graduate School of Science, Nagoya University, Nagoya 464-8602, Japan
(Dated: August 8, 2008)
^ ■ This letter reports the results of a search for a stochastic background of gravitational waves
(GW) at 100 MHz by laser interferometry. We have developed a GW detector, which is a pair
of 75-cm baseline synchronous recycling (resonant recycling) interferometers. Each interferometer
has a strain sensitivity of ~ 10~ 16 Hz -1 / 2 at 100 MHz. By cross-correlating the outputs of the two
interferometers within 1000 seconds, we found /ifog^gw < 6 x 10 25 to be an upper limit on the
energy density spectrum of the GW background in a 2-kHz bandwidth around 100 MHz, where a
flat spectrum is assumed.
PACS numbers: Valid PACS appear here
Recently, Cruise and Ingley reported on a detector for gravitational waves (GW) at 100
MHz l|. Their GW detector is a pair of waveguide loop cavities, each of which has a
strain sensitivity of ~ 10~ 14 Hz -1 / 2 at the frequency. Except for this, no experiments were
attempted to directly detect GWs at very high frequencies (above 100 kHz), while many
theories predict a stochastic gravitational-wave background (GWB) in a broad range of
frequencies, 10~ 18 — 10 10 Hz. At very high frequencies, a relatively large GWB is predicted
by some models of the early universe and compact astronomical objects (references are
summarized in our previous paper ). Although the amount of the cosmic GWB is indirectly
limited by not only the helium-4 abundance due to big-bang nucleosynthesis [3|, but also
measurements of the cosmic microwave background |4| , direct search experiments for a GWB
at very high frequencies should be significant.
We have developed a more sensitive detector for 100-MHz GWs using laser interferome-
ters. The detector is a pair of synchronous recycling interferometers, where the synchronous
recycling (or resonant recycling) technique was proposed by Drever in the 1980s |5|. In our
previous papers 0, E| , we showed that this interferometer is suitable to detect a GWB at
very high frequency with high signal-to-noise ratio (SNR), and that the SNR can be im-
proved by cross-correlating the outputs of the two interferometers. In this letter, we report
the first results of the search for a stochastic GWB at 100 MHz with the GW detector.
Synchronous recycling interferometer. — The interferometer has a resonant response to
GWs at a specific frequency p], IsJ]. GW signals are enhanced in a recycling cavity (see
Fig. ), which is formed by a recycling mirror (RM), a transfer mirror (TM), and two end
mirrors (EMI and EM2).
The size of the recycling cavity determines the resonant frequency, where the signal
enhancement is proportional to the laser power kept in the cavity. At the entrance of the
interferometer, a laser beam is divided into two orthogonal directions by a beamsplitter
(BS). Thus two beams are incident on the RM, which is a beamsplitter but with relatively
high reflectivity. When the laser frequency is stabilized to the recycling cavity, the two
beams passing through the RM are resonant in the cavity by circulating many times along a
common path in opposite directions (clockwise and counterclockwise). At the same time, the
two circulating beams will experience differential phase shifts due to quadrupole components
of GWs. The phase difference is maximized for the GWs at the same frequency as the free-
spectral range z/fsr, the inverse of the round-trip period of the circulating beams. The phase
Recycling cavity^ £^ End mjrror 2
FIG. 1: Schematic view of a synchronous recycling interferometer. GW signals are enhanced in
the recycling cavity, and detected with the photodetector.
difference is enhanced as the laser power builds up depending on the finesse of the recycling
cavity. The beams that left the cavity are recombined at the BS so that the differential
components (GW signals) are detected with the photodetector (PD).
Experimental setup. — We have developed two synchronous recycling interferometers,
hereafter called IFO-1 and IFO-2. For each interferometer (see Fig. ), we use a Nd:YAG
continuous-wave laser with a wavelength of 1064 nm and a laser power of 0.5 W. The laser
beam passes through an electro-optic phase modulator (EOl) and a Faraday isolator (FI),
and then enters the interferometer. The recycling cavity is designed to have a baseline
length (distance from the RM to the EMI or EM2 [15j) of L ~ 75 cm so that the GW
response is maximized at z/fsr = c/(4L) ~ 100 MHz, where c is the speed of light 16].
Because this experiment is the first step in the direct detection of a GWB at 100 MHz, both
interferometers are constructed in the air, and each recycling cavity is designed to have a
finesse of ~ 100; each RM has relatively low reflectivity (nominal 98.5%). For calibration,
we use the E02 to simulate GWs by modulating the phases of the circulating beams in the
cavity. The size of the Sagnac interferometer, which is formed by the BS, the RM, and two
steering mirrors, is relatively small (12.5-cm square optical path), and thus its GW response
is insignificant compared to that of the recycling cavity.
The laser frequency is stabilized to the recycling cavity by the Pound-Drever-Hall tech-
nique . This technique requires phase- modulation sidebands spaced by a radio frequency
(RF) from the laser-source (carrier) frequency z/ in the optical frequency domain. The RF
LO: 14.696840 MHz
FIG. 2: (Color online.) Schematic view of the experimental setup of one of the interferometers.
EO: electro-optic phase modulator; FI: Faraday isolator; DAQ: data acquisition system. The laser
frequency is stabilized to the recycling cavity by the Pound-Drever-Hall technique. GW signal
sidebands are once converted to intermediate-frequency (IF) signals at the PD2. Then the IF
signals are mixed with a local oscillator (LO), and converted to audio- frequency (AF) signals. The
AF signals are recorded with the DAQ. The E02 is used to simulate GW signal sidebands for
sidebands at v ± / RF are induced at the EOl, where the laser light is phase- modulated
& t /rf = 85.4 MHz. The PD1 detects the light reflected from the cavity and produces a
photocurrent, which contains RF signals modulated by the relative deviation between the
laser and the cavity. We correct the relative deviation using signals demodulated from the
The target GW signals are converted to electrical signals at intermediate frequencies (IF)
~ 15 MHz with the PD2, since it is difficult to make a low-noise photodetector that can
respond to signals at very high frequencies (~ 100 MHz). The PD2 produces IF signals
at /if = few — /rf in response to the beat between the RF sidebands (also used for the
laser stabilization) and signal sidebands (representation of the GW signals in the optical
frequency domain) at z/ ± /gw> where /gw is the GW frequency. A small fraction of the
RF sidebands leaks to the PD2, since the splitting ratio of the BS is not exactly balanced in
the realistic case. Otherwise, non-differential components including the RF sidebands are in
FIG. 3: (Color online.) Strain sensitivities of the interferometers estimated from the IF signals.
The solid red and dashed green lines represent the sensitivities of IFO-1 and IFO-2, respectively.
principle completely reflected into the PD1. The PD2 is designed to have a band-path filter
centered at /if ~ 15 MHz with about 1-MHz bandwidth. Thus the conversion coefficient
from GWs to IF signals contains the filter response of the PD2 as well as the frequency
response of the recycling cavity.
We find the strain sensitivity of each interferometer is about 10 -16 Hz -1 ' 2 around 100 MHz
(Fig. ) . We estimate the sensitivity from IF signals. For calibration of the interfer-
ometer outputs, we estimate the conversion coefficient from the IF signals Vjp to the GW
amplitudes h by modulating the phases of the circulating beams with the E02 driven by
calibration signals V ca ,\ [l8|. The coefficient is the product of the response of the recycling
cavity and the band-path filter of the PD2, and is estimated by
A(f GW )C(f GW ). (1)
V lF (fw) M/if)
where A(rad/V) is the measured modulation efficiency of the E02, and C (strain/rad) is
the calculated conversion coefficient from the phase modulation to the simulated GWs. The
term C is a function of the distance of the E02 from the RM, and it is ~ 25 cm in our
The IF signals vary too quickly to be sampled with an inexpensive data acquisition
(DAQ) system. We convert the IF signals to recordable audio-frequency (AF) signals at
/af = few - (/ c - A//2) with a local oscillator (LO) at f LO = f c - Af/2 - f RF , where
we choose Af = 6.32 kHz as a signal bandwidth to be recorded, and f c = 100.1 MHz as a
center frequency of the bandwidth. They yield /lo = 14.696840 MHz. For example, GWs
at 100.1 MHz corresponds to AF signals at 3.16 kHz.
Cross-correlation analysis. — Using the outputs of the two interferometers, we have per-
formed a cross- correlation analysis to reduce uncorrelated noises between them and improve
the SNR, the ratio of the GW signals to the interferometer noises. The analysis method
is similar to the method used in LIGO
11] . We assume that a GWB is isotropic,
unpolarized, stationary, and Gaussian, and it is so small that the interferometer outputs
are dominated by their noises rather than GW signals. The GWB is often character-
ized by a normalized energy density spectrum per unit logarithmic frequency interval
fi gw (/) = p^ 1 dp(f)/d\n f, where p(f) is the cumulative energy density of GWB included
below / Hz, and p c = 3HqC 2 / (8tcG) is the critical energy density of the universe; here G is
the Newton constant, and Hq = h\oo x lOOkm/s/Mpc is the Hubble constant. In this letter,
we also use the form hl 00 Q gw (f), which is independent of the value of hi 00 .
We define a cross-correlation statistic:
1 I™ x* x (f)x 2 (f)Q(f)df, (2)
J — oo
where x\ and x 2 are Fourier components of the signal outputs from IFO-1 and IFO-2,
respectively; T is the observation time period; Q is the optimal filter that optimizes the
SNR of an expectation value (ensemble average) of Z\ 2 estimated from available data (the
exact definition of Q will be given later in Eq.()).
The expectation value of Z\ 2 and its variance are respectively written as
^ s (zi2) = H SZ *^^mqw> (3)
4 = (Z 2 l2 ) - (Z 12 ) 2 ~ — / dfPi(|/|)P 2 (i/|)|Q(/)i 2 , (4)
where Pi and P 2 are the one-sided power spectral densities (PSD) of the noises in IFO-1
and IFO-2, respectively [h]]; and 712 is called the reduced overlap reduction function. As
the usual overlap reduction function in the low-frequency limit Q] , 712 represents the
reduction of the signal correlation caused by the distance between the two interferometer
sites and the alignment of their arms. In our experiment, 712 ~ 0.93 is nearly constant
around 100 MHz, because the two recycling cavities are co-aligned and almost co-located
(the distance is ~ 10cm) j^ .
The SNR of the estimation is defined as Hz / y&z- To optimize the SNR, the filter Q is
n(f\ - y n gw(l/l)7i2(/) (r .s
Qn \f\ 3 Pi(\f\m\f\y [)
where K is a normalization constant [6|. Using this optimal filter, we find that the SNR is
snr - 3H <> s/t \ r .„ -rumuw) i 1/2
SNR -T^ VT [J_JwmlmW)\ ■ (6)
Thus the SNR in principle increases proportional to VT.
The observation time period used for the cross-correlation analysis is 1070.5 seconds. The
data record is divided into N = 439 segments. For each segment, the cross-correlation and
its uncertainty are calculated based on Eqs.© and (j4"l); we will refer to the calculated ones
as Zu and a|, respectively. The ensemble average \iz defined in Eq.Q is estimated by a
HZ = ^'
V 2 (n) ' ^ ' '
where the superscript "(n)" indicates that the quantity is calculated from the n-th segment
n, = 1, 2, ... , N); a is the uncertainty of fiz and is written as
n=l a Z
As the integration domain in Eq.(T2]), we use a range from 2.08 kHz to 4.19 kHz for AF signals,
which corresponds to a 2-kHz bandwidth around 100.1 MHz for GW signals. Because the
optimal filter in Eq.(jSJ) contains fi gw (/) itself, we need to assume its spectrum in advance.
We assume that the spectrum will be flat in such a narrow bandwidth.
The variation of fiz itself with respect to the observation time period is shown in FigJH
The two curves represent ^±1.65(7^, and the area enclosed by the curves is a two-sided 90%
confidence interval of \xz- After the 1070.5-second observation, we obtain (iz = 4.9 x 10~ 12
with = 3.7 x 10~ 12 . The 90% confidence interval [fiz — 1.65<7 M ,/iz + 1.65(5"^] includes
Hz = 0; in other words, there is a possibility of Q gw = at 100 MHz. Instead, we consider
an upper limit on the amount of the stochastic GWB at 100 MHz.
We define the upper limit as a one-sided 90% confidence level; in terms of Hz, the upper
limit corresponds to fiz + 1.28a M . Then we find h^Q^ < 6 x 10 25 as an upper limit on
the stochastic GWB at around 100.1 MHz from the direct search experiment. Note that
1 1 1 1
i i , , , , i i i i i i
i i i i i i i
i : _
, , , ,
, , , . i . . , , i , , , ,
. ... t .... '
200 400 600 800 1000 1200
Observation time [sec]
FIG. 4: (Color online.) Variation of fiz with respect to the observation time period is shown as
filled red circles. The area enclosed by black curves is its two-sided 90% confidence interval.
this is also an upper limit on the correlated noises between the two interferometers at this
Conclusions. — We searched for a stochastic GWB at 100 MHz by laser interferometry.
The GW detector is a pair of synchronous recycling interferometers. Each interferometer has
a strain sensitivity of ~ 10 -16 Hz -1 ' 2 to GWs at 100 MHz. Using the two interferometers,
we directly searched for a stochastic GWB centered at 100.1 MHz with 2-kHz bandwidth in
1070.5 seconds. We performed a cross-correlation analysis to improve the SNR of the search.
We found h^Q^ < 6 x 10 25 to be an upper limit on the energy density of a stochastic
GWB at 100 MHz.
We plan to improve the GW detector by increasing the finesse of each recycling cavity
up to about 4.5 x 10 4 . For this purpose, each cavity will be constructed in a vacuum with
high-reflectivity mirrors in future. Then each interferometer will have a strain sensitivity
of about 4.7 x 10" 21 Hz~ 1 / 2 . For about a one-year observation, we should obtain a tighter
upper limit as hl 00 fl gw ~ 2.8 x 10 14 around 100 MHz by a cross-correlation analysis with
these two interferometers.
This research is supported by Grant-in- Aid for Scientific Research (A) 17204018 from the
Ministry of Education, Culture, Sports, Science and Technology.
* Electronic address: takutsu@g ravity.mtk.nao.ac.jp|
 A. M. Cruise and R. M. J. Ingley, Class. Quantum. Grav. 23, 6185 (2006).
 A. Nishizawa et al., Phys. Rev. D 77, 022002 (2008).
 M. Maggiore, Phys. Rep. 331, 283 (2000).
 T. L. Smith, E. Pierpaoli, and M. Kamionkowski, Phys. Rev. Lett. 97, 021301 (2006).
 R. W. P. Drever, Gravitational radiation, edited by N. Deruelle and T. Piran (North-Holland,
Amsterdam, 1983), pp.321-338.
 A. Nishizawa et al., arXiv:0801.4149, "Optimal Location of Two Laser-interferometric Detec-
tors for Gravitational Wave Backgrounds at 100 MHz".
 J.-Y. Vinet, B. Meers, C. N. Man, and A. Brillet, Phys. Rev. D 38, 433 (1988).
 B. J. Meers, Phys. Rev. D 38, 2317 (1988).
 R. W. P. Drever, J. L. Hall, F. V. Kowalski, J. Hough, G. M. Ford, and H. Ward, Appl. Phys.
B 31, 97 (1983).
 B. Allen and J. D. Romano, Phys. Rev. D 59, 102001 (1999).
 B. Abbott et al., Astrophys. J. 659, 918 (2007).
 N. Christensen, Phys. Rev. D 46, 5250 (1992).
 E. E. Flanagan, Phys. Rev. D 48, 2389 (1993).
 K. X. Sun, M. M. Fejer, E. Gustafson, and R. L. Byer, Phys. Rev. Lett. 76, 3053 (1996).
 The cavity's optical path is crossed so that it encloses nearly zero area so as to be insensitive
to the Sagnac effect [141 ].
 One can construct a narrowband audio-frequency GW detector by adding delay lines or Fabry-
Perot cavities in the recycling cavity Q] .
 There was a concern on environmental electromagnetic noises around 100 MHz. We use a loop
antenna to measure the noise features, and choose a quiet band for the GW search.
 The calibration signal generator induces noises as shown in Fig.  (the common peaks for both
IFOs at about 100.1 MHz in this case), even when the generator is in a stand-by mode. We
power off the generator during the 1000-second data taking described below.
 The square root of each PSD is the strain sensitivity.
 The overlap reduction function, j(f), defined in our previous paper (| is related to 712 in the
following form: -f(f) = 712 (/)
sin I ^ 1J L -
. Both functions reduce to the usual
overlap reduction function in the low-frequency limit: / — > 0.